]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET3 Protocol independent device support routines. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Derived from the non IP parts of dev.c 1.0.19 | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * | |
14 | * Additional Authors: | |
15 | * Florian la Roche <rzsfl@rz.uni-sb.de> | |
16 | * Alan Cox <gw4pts@gw4pts.ampr.org> | |
17 | * David Hinds <dahinds@users.sourceforge.net> | |
18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> | |
19 | * Adam Sulmicki <adam@cfar.umd.edu> | |
20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> | |
21 | * | |
22 | * Changes: | |
23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set | |
24 | * to 2 if register_netdev gets called | |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
27 | * Alan Cox : device private ioctl copies fields back. | |
28 | * Alan Cox : Transmit queue code does relevant | |
29 | * stunts to keep the queue safe. | |
30 | * Alan Cox : Fixed double lock. | |
31 | * Alan Cox : Fixed promisc NULL pointer trap | |
32 | * ???????? : Support the full private ioctl range | |
33 | * Alan Cox : Moved ioctl permission check into | |
34 | * drivers | |
35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI | |
36 | * Alan Cox : 100 backlog just doesn't cut it when | |
37 | * you start doing multicast video 8) | |
38 | * Alan Cox : Rewrote net_bh and list manager. | |
39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. | |
40 | * Alan Cox : Took out transmit every packet pass | |
41 | * Saved a few bytes in the ioctl handler | |
42 | * Alan Cox : Network driver sets packet type before | |
43 | * calling netif_rx. Saves a function | |
44 | * call a packet. | |
45 | * Alan Cox : Hashed net_bh() | |
46 | * Richard Kooijman: Timestamp fixes. | |
47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR | |
48 | * Alan Cox : Device lock protection. | |
49 | * Alan Cox : Fixed nasty side effect of device close | |
50 | * changes. | |
51 | * Rudi Cilibrasi : Pass the right thing to | |
52 | * set_mac_address() | |
53 | * Dave Miller : 32bit quantity for the device lock to | |
54 | * make it work out on a Sparc. | |
55 | * Bjorn Ekwall : Added KERNELD hack. | |
56 | * Alan Cox : Cleaned up the backlog initialise. | |
57 | * Craig Metz : SIOCGIFCONF fix if space for under | |
58 | * 1 device. | |
59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there | |
60 | * is no device open function. | |
61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF | |
62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF | |
63 | * Cyrus Durgin : Cleaned for KMOD | |
64 | * Adam Sulmicki : Bug Fix : Network Device Unload | |
65 | * A network device unload needs to purge | |
66 | * the backlog queue. | |
67 | * Paul Rusty Russell : SIOCSIFNAME | |
68 | * Pekka Riikonen : Netdev boot-time settings code | |
69 | * Andrew Morton : Make unregister_netdevice wait | |
70 | * indefinitely on dev->refcnt | |
71 | * J Hadi Salim : - Backlog queue sampling | |
72 | * - netif_rx() feedback | |
73 | */ | |
74 | ||
75 | #include <asm/uaccess.h> | |
1da177e4 | 76 | #include <linux/bitops.h> |
4fc268d2 | 77 | #include <linux/capability.h> |
1da177e4 LT |
78 | #include <linux/cpu.h> |
79 | #include <linux/types.h> | |
80 | #include <linux/kernel.h> | |
08e9897d | 81 | #include <linux/hash.h> |
5a0e3ad6 | 82 | #include <linux/slab.h> |
1da177e4 | 83 | #include <linux/sched.h> |
4a3e2f71 | 84 | #include <linux/mutex.h> |
1da177e4 LT |
85 | #include <linux/string.h> |
86 | #include <linux/mm.h> | |
87 | #include <linux/socket.h> | |
88 | #include <linux/sockios.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/interrupt.h> | |
91 | #include <linux/if_ether.h> | |
92 | #include <linux/netdevice.h> | |
93 | #include <linux/etherdevice.h> | |
0187bdfb | 94 | #include <linux/ethtool.h> |
1da177e4 LT |
95 | #include <linux/notifier.h> |
96 | #include <linux/skbuff.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> |
1da177e4 | 102 | #include <net/dst.h> |
fc4099f1 | 103 | #include <net/dst_metadata.h> |
1da177e4 LT |
104 | #include <net/pkt_sched.h> |
105 | #include <net/checksum.h> | |
44540960 | 106 | #include <net/xfrm.h> |
1da177e4 LT |
107 | #include <linux/highmem.h> |
108 | #include <linux/init.h> | |
1da177e4 | 109 | #include <linux/module.h> |
1da177e4 LT |
110 | #include <linux/netpoll.h> |
111 | #include <linux/rcupdate.h> | |
112 | #include <linux/delay.h> | |
1da177e4 | 113 | #include <net/iw_handler.h> |
1da177e4 | 114 | #include <asm/current.h> |
5bdb9886 | 115 | #include <linux/audit.h> |
db217334 | 116 | #include <linux/dmaengine.h> |
f6a78bfc | 117 | #include <linux/err.h> |
c7fa9d18 | 118 | #include <linux/ctype.h> |
723e98b7 | 119 | #include <linux/if_arp.h> |
6de329e2 | 120 | #include <linux/if_vlan.h> |
8f0f2223 | 121 | #include <linux/ip.h> |
ad55dcaf | 122 | #include <net/ip.h> |
25cd9ba0 | 123 | #include <net/mpls.h> |
8f0f2223 DM |
124 | #include <linux/ipv6.h> |
125 | #include <linux/in.h> | |
b6b2fed1 DM |
126 | #include <linux/jhash.h> |
127 | #include <linux/random.h> | |
9cbc1cb8 | 128 | #include <trace/events/napi.h> |
cf66ba58 | 129 | #include <trace/events/net.h> |
07dc22e7 | 130 | #include <trace/events/skb.h> |
5acbbd42 | 131 | #include <linux/pci.h> |
caeda9b9 | 132 | #include <linux/inetdevice.h> |
c445477d | 133 | #include <linux/cpu_rmap.h> |
c5905afb | 134 | #include <linux/static_key.h> |
af12fa6e | 135 | #include <linux/hashtable.h> |
60877a32 | 136 | #include <linux/vmalloc.h> |
529d0489 | 137 | #include <linux/if_macvlan.h> |
e7fd2885 | 138 | #include <linux/errqueue.h> |
3b47d303 | 139 | #include <linux/hrtimer.h> |
e687ad60 | 140 | #include <linux/netfilter_ingress.h> |
6ae23ad3 | 141 | #include <linux/sctp.h> |
1da177e4 | 142 | |
342709ef PE |
143 | #include "net-sysfs.h" |
144 | ||
d565b0a1 HX |
145 | /* Instead of increasing this, you should create a hash table. */ |
146 | #define MAX_GRO_SKBS 8 | |
147 | ||
5d38a079 HX |
148 | /* This should be increased if a protocol with a bigger head is added. */ |
149 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
150 | ||
1da177e4 | 151 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 152 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
153 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
154 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 155 | static struct list_head offload_base __read_mostly; |
1da177e4 | 156 | |
ae78dbfa | 157 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
158 | static int call_netdevice_notifiers_info(unsigned long val, |
159 | struct net_device *dev, | |
160 | struct netdev_notifier_info *info); | |
ae78dbfa | 161 | |
1da177e4 | 162 | /* |
7562f876 | 163 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
164 | * semaphore. |
165 | * | |
c6d14c84 | 166 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
167 | * |
168 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 169 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
170 | * actual updates. This allows pure readers to access the list even |
171 | * while a writer is preparing to update it. | |
172 | * | |
173 | * To put it another way, dev_base_lock is held for writing only to | |
174 | * protect against pure readers; the rtnl semaphore provides the | |
175 | * protection against other writers. | |
176 | * | |
177 | * See, for example usages, register_netdevice() and | |
178 | * unregister_netdevice(), which must be called with the rtnl | |
179 | * semaphore held. | |
180 | */ | |
1da177e4 | 181 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
182 | EXPORT_SYMBOL(dev_base_lock); |
183 | ||
af12fa6e ET |
184 | /* protects napi_hash addition/deletion and napi_gen_id */ |
185 | static DEFINE_SPINLOCK(napi_hash_lock); | |
186 | ||
52bd2d62 | 187 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 188 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 189 | |
18afa4b0 | 190 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 191 | |
4e985ada TG |
192 | static inline void dev_base_seq_inc(struct net *net) |
193 | { | |
194 | while (++net->dev_base_seq == 0); | |
195 | } | |
196 | ||
881d966b | 197 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 198 | { |
95c96174 ED |
199 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
200 | ||
08e9897d | 201 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
202 | } |
203 | ||
881d966b | 204 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 205 | { |
7c28bd0b | 206 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
207 | } |
208 | ||
e36fa2f7 | 209 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
210 | { |
211 | #ifdef CONFIG_RPS | |
e36fa2f7 | 212 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
213 | #endif |
214 | } | |
215 | ||
e36fa2f7 | 216 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
217 | { |
218 | #ifdef CONFIG_RPS | |
e36fa2f7 | 219 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
220 | #endif |
221 | } | |
222 | ||
ce286d32 | 223 | /* Device list insertion */ |
53759be9 | 224 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 225 | { |
c346dca1 | 226 | struct net *net = dev_net(dev); |
ce286d32 EB |
227 | |
228 | ASSERT_RTNL(); | |
229 | ||
230 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 231 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 232 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
233 | hlist_add_head_rcu(&dev->index_hlist, |
234 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 235 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
236 | |
237 | dev_base_seq_inc(net); | |
ce286d32 EB |
238 | } |
239 | ||
fb699dfd ED |
240 | /* Device list removal |
241 | * caller must respect a RCU grace period before freeing/reusing dev | |
242 | */ | |
ce286d32 EB |
243 | static void unlist_netdevice(struct net_device *dev) |
244 | { | |
245 | ASSERT_RTNL(); | |
246 | ||
247 | /* Unlink dev from the device chain */ | |
248 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 249 | list_del_rcu(&dev->dev_list); |
72c9528b | 250 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 251 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 252 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
253 | |
254 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
255 | } |
256 | ||
1da177e4 LT |
257 | /* |
258 | * Our notifier list | |
259 | */ | |
260 | ||
f07d5b94 | 261 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
262 | |
263 | /* | |
264 | * Device drivers call our routines to queue packets here. We empty the | |
265 | * queue in the local softnet handler. | |
266 | */ | |
bea3348e | 267 | |
9958da05 | 268 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 269 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 270 | |
cf508b12 | 271 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 272 | /* |
c773e847 | 273 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
274 | * according to dev->type |
275 | */ | |
276 | static const unsigned short netdev_lock_type[] = | |
277 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
278 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
279 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
280 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
281 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
282 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
283 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
284 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
285 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
286 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
287 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
288 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
289 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
290 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
291 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 292 | |
36cbd3dc | 293 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
294 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
295 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
296 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
297 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
298 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
299 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
300 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
301 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
302 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
303 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
304 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
305 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
306 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
307 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
308 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
309 | |
310 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 311 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
312 | |
313 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
314 | { | |
315 | int i; | |
316 | ||
317 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
318 | if (netdev_lock_type[i] == dev_type) | |
319 | return i; | |
320 | /* the last key is used by default */ | |
321 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
322 | } | |
323 | ||
cf508b12 DM |
324 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
325 | unsigned short dev_type) | |
723e98b7 JP |
326 | { |
327 | int i; | |
328 | ||
329 | i = netdev_lock_pos(dev_type); | |
330 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
331 | netdev_lock_name[i]); | |
332 | } | |
cf508b12 DM |
333 | |
334 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
335 | { | |
336 | int i; | |
337 | ||
338 | i = netdev_lock_pos(dev->type); | |
339 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
340 | &netdev_addr_lock_key[i], | |
341 | netdev_lock_name[i]); | |
342 | } | |
723e98b7 | 343 | #else |
cf508b12 DM |
344 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
345 | unsigned short dev_type) | |
346 | { | |
347 | } | |
348 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
349 | { |
350 | } | |
351 | #endif | |
1da177e4 LT |
352 | |
353 | /******************************************************************************* | |
354 | ||
355 | Protocol management and registration routines | |
356 | ||
357 | *******************************************************************************/ | |
358 | ||
1da177e4 LT |
359 | /* |
360 | * Add a protocol ID to the list. Now that the input handler is | |
361 | * smarter we can dispense with all the messy stuff that used to be | |
362 | * here. | |
363 | * | |
364 | * BEWARE!!! Protocol handlers, mangling input packets, | |
365 | * MUST BE last in hash buckets and checking protocol handlers | |
366 | * MUST start from promiscuous ptype_all chain in net_bh. | |
367 | * It is true now, do not change it. | |
368 | * Explanation follows: if protocol handler, mangling packet, will | |
369 | * be the first on list, it is not able to sense, that packet | |
370 | * is cloned and should be copied-on-write, so that it will | |
371 | * change it and subsequent readers will get broken packet. | |
372 | * --ANK (980803) | |
373 | */ | |
374 | ||
c07b68e8 ED |
375 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
376 | { | |
377 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 378 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 379 | else |
7866a621 SN |
380 | return pt->dev ? &pt->dev->ptype_specific : |
381 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
382 | } |
383 | ||
1da177e4 LT |
384 | /** |
385 | * dev_add_pack - add packet handler | |
386 | * @pt: packet type declaration | |
387 | * | |
388 | * Add a protocol handler to the networking stack. The passed &packet_type | |
389 | * is linked into kernel lists and may not be freed until it has been | |
390 | * removed from the kernel lists. | |
391 | * | |
4ec93edb | 392 | * This call does not sleep therefore it can not |
1da177e4 LT |
393 | * guarantee all CPU's that are in middle of receiving packets |
394 | * will see the new packet type (until the next received packet). | |
395 | */ | |
396 | ||
397 | void dev_add_pack(struct packet_type *pt) | |
398 | { | |
c07b68e8 | 399 | struct list_head *head = ptype_head(pt); |
1da177e4 | 400 | |
c07b68e8 ED |
401 | spin_lock(&ptype_lock); |
402 | list_add_rcu(&pt->list, head); | |
403 | spin_unlock(&ptype_lock); | |
1da177e4 | 404 | } |
d1b19dff | 405 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 406 | |
1da177e4 LT |
407 | /** |
408 | * __dev_remove_pack - remove packet handler | |
409 | * @pt: packet type declaration | |
410 | * | |
411 | * Remove a protocol handler that was previously added to the kernel | |
412 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
413 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 414 | * returns. |
1da177e4 LT |
415 | * |
416 | * The packet type might still be in use by receivers | |
417 | * and must not be freed until after all the CPU's have gone | |
418 | * through a quiescent state. | |
419 | */ | |
420 | void __dev_remove_pack(struct packet_type *pt) | |
421 | { | |
c07b68e8 | 422 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
423 | struct packet_type *pt1; |
424 | ||
c07b68e8 | 425 | spin_lock(&ptype_lock); |
1da177e4 LT |
426 | |
427 | list_for_each_entry(pt1, head, list) { | |
428 | if (pt == pt1) { | |
429 | list_del_rcu(&pt->list); | |
430 | goto out; | |
431 | } | |
432 | } | |
433 | ||
7b6cd1ce | 434 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 435 | out: |
c07b68e8 | 436 | spin_unlock(&ptype_lock); |
1da177e4 | 437 | } |
d1b19dff ED |
438 | EXPORT_SYMBOL(__dev_remove_pack); |
439 | ||
1da177e4 LT |
440 | /** |
441 | * dev_remove_pack - remove packet handler | |
442 | * @pt: packet type declaration | |
443 | * | |
444 | * Remove a protocol handler that was previously added to the kernel | |
445 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
446 | * from the kernel lists and can be freed or reused once this function | |
447 | * returns. | |
448 | * | |
449 | * This call sleeps to guarantee that no CPU is looking at the packet | |
450 | * type after return. | |
451 | */ | |
452 | void dev_remove_pack(struct packet_type *pt) | |
453 | { | |
454 | __dev_remove_pack(pt); | |
4ec93edb | 455 | |
1da177e4 LT |
456 | synchronize_net(); |
457 | } | |
d1b19dff | 458 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 459 | |
62532da9 VY |
460 | |
461 | /** | |
462 | * dev_add_offload - register offload handlers | |
463 | * @po: protocol offload declaration | |
464 | * | |
465 | * Add protocol offload handlers to the networking stack. The passed | |
466 | * &proto_offload is linked into kernel lists and may not be freed until | |
467 | * it has been removed from the kernel lists. | |
468 | * | |
469 | * This call does not sleep therefore it can not | |
470 | * guarantee all CPU's that are in middle of receiving packets | |
471 | * will see the new offload handlers (until the next received packet). | |
472 | */ | |
473 | void dev_add_offload(struct packet_offload *po) | |
474 | { | |
bdef7de4 | 475 | struct packet_offload *elem; |
62532da9 VY |
476 | |
477 | spin_lock(&offload_lock); | |
bdef7de4 DM |
478 | list_for_each_entry(elem, &offload_base, list) { |
479 | if (po->priority < elem->priority) | |
480 | break; | |
481 | } | |
482 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
483 | spin_unlock(&offload_lock); |
484 | } | |
485 | EXPORT_SYMBOL(dev_add_offload); | |
486 | ||
487 | /** | |
488 | * __dev_remove_offload - remove offload handler | |
489 | * @po: packet offload declaration | |
490 | * | |
491 | * Remove a protocol offload handler that was previously added to the | |
492 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
493 | * is removed from the kernel lists and can be freed or reused once this | |
494 | * function returns. | |
495 | * | |
496 | * The packet type might still be in use by receivers | |
497 | * and must not be freed until after all the CPU's have gone | |
498 | * through a quiescent state. | |
499 | */ | |
1d143d9f | 500 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
501 | { |
502 | struct list_head *head = &offload_base; | |
503 | struct packet_offload *po1; | |
504 | ||
c53aa505 | 505 | spin_lock(&offload_lock); |
62532da9 VY |
506 | |
507 | list_for_each_entry(po1, head, list) { | |
508 | if (po == po1) { | |
509 | list_del_rcu(&po->list); | |
510 | goto out; | |
511 | } | |
512 | } | |
513 | ||
514 | pr_warn("dev_remove_offload: %p not found\n", po); | |
515 | out: | |
c53aa505 | 516 | spin_unlock(&offload_lock); |
62532da9 | 517 | } |
62532da9 VY |
518 | |
519 | /** | |
520 | * dev_remove_offload - remove packet offload handler | |
521 | * @po: packet offload declaration | |
522 | * | |
523 | * Remove a packet offload handler that was previously added to the kernel | |
524 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
525 | * removed from the kernel lists and can be freed or reused once this | |
526 | * function returns. | |
527 | * | |
528 | * This call sleeps to guarantee that no CPU is looking at the packet | |
529 | * type after return. | |
530 | */ | |
531 | void dev_remove_offload(struct packet_offload *po) | |
532 | { | |
533 | __dev_remove_offload(po); | |
534 | ||
535 | synchronize_net(); | |
536 | } | |
537 | EXPORT_SYMBOL(dev_remove_offload); | |
538 | ||
1da177e4 LT |
539 | /****************************************************************************** |
540 | ||
541 | Device Boot-time Settings Routines | |
542 | ||
543 | *******************************************************************************/ | |
544 | ||
545 | /* Boot time configuration table */ | |
546 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
547 | ||
548 | /** | |
549 | * netdev_boot_setup_add - add new setup entry | |
550 | * @name: name of the device | |
551 | * @map: configured settings for the device | |
552 | * | |
553 | * Adds new setup entry to the dev_boot_setup list. The function | |
554 | * returns 0 on error and 1 on success. This is a generic routine to | |
555 | * all netdevices. | |
556 | */ | |
557 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
558 | { | |
559 | struct netdev_boot_setup *s; | |
560 | int i; | |
561 | ||
562 | s = dev_boot_setup; | |
563 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
564 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
565 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 566 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
567 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
568 | break; | |
569 | } | |
570 | } | |
571 | ||
572 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
573 | } | |
574 | ||
575 | /** | |
576 | * netdev_boot_setup_check - check boot time settings | |
577 | * @dev: the netdevice | |
578 | * | |
579 | * Check boot time settings for the device. | |
580 | * The found settings are set for the device to be used | |
581 | * later in the device probing. | |
582 | * Returns 0 if no settings found, 1 if they are. | |
583 | */ | |
584 | int netdev_boot_setup_check(struct net_device *dev) | |
585 | { | |
586 | struct netdev_boot_setup *s = dev_boot_setup; | |
587 | int i; | |
588 | ||
589 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
590 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 591 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
592 | dev->irq = s[i].map.irq; |
593 | dev->base_addr = s[i].map.base_addr; | |
594 | dev->mem_start = s[i].map.mem_start; | |
595 | dev->mem_end = s[i].map.mem_end; | |
596 | return 1; | |
597 | } | |
598 | } | |
599 | return 0; | |
600 | } | |
d1b19dff | 601 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
602 | |
603 | ||
604 | /** | |
605 | * netdev_boot_base - get address from boot time settings | |
606 | * @prefix: prefix for network device | |
607 | * @unit: id for network device | |
608 | * | |
609 | * Check boot time settings for the base address of device. | |
610 | * The found settings are set for the device to be used | |
611 | * later in the device probing. | |
612 | * Returns 0 if no settings found. | |
613 | */ | |
614 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
615 | { | |
616 | const struct netdev_boot_setup *s = dev_boot_setup; | |
617 | char name[IFNAMSIZ]; | |
618 | int i; | |
619 | ||
620 | sprintf(name, "%s%d", prefix, unit); | |
621 | ||
622 | /* | |
623 | * If device already registered then return base of 1 | |
624 | * to indicate not to probe for this interface | |
625 | */ | |
881d966b | 626 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
627 | return 1; |
628 | ||
629 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
630 | if (!strcmp(name, s[i].name)) | |
631 | return s[i].map.base_addr; | |
632 | return 0; | |
633 | } | |
634 | ||
635 | /* | |
636 | * Saves at boot time configured settings for any netdevice. | |
637 | */ | |
638 | int __init netdev_boot_setup(char *str) | |
639 | { | |
640 | int ints[5]; | |
641 | struct ifmap map; | |
642 | ||
643 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
644 | if (!str || !*str) | |
645 | return 0; | |
646 | ||
647 | /* Save settings */ | |
648 | memset(&map, 0, sizeof(map)); | |
649 | if (ints[0] > 0) | |
650 | map.irq = ints[1]; | |
651 | if (ints[0] > 1) | |
652 | map.base_addr = ints[2]; | |
653 | if (ints[0] > 2) | |
654 | map.mem_start = ints[3]; | |
655 | if (ints[0] > 3) | |
656 | map.mem_end = ints[4]; | |
657 | ||
658 | /* Add new entry to the list */ | |
659 | return netdev_boot_setup_add(str, &map); | |
660 | } | |
661 | ||
662 | __setup("netdev=", netdev_boot_setup); | |
663 | ||
664 | /******************************************************************************* | |
665 | ||
666 | Device Interface Subroutines | |
667 | ||
668 | *******************************************************************************/ | |
669 | ||
a54acb3a ND |
670 | /** |
671 | * dev_get_iflink - get 'iflink' value of a interface | |
672 | * @dev: targeted interface | |
673 | * | |
674 | * Indicates the ifindex the interface is linked to. | |
675 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
676 | */ | |
677 | ||
678 | int dev_get_iflink(const struct net_device *dev) | |
679 | { | |
680 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
681 | return dev->netdev_ops->ndo_get_iflink(dev); | |
682 | ||
7a66bbc9 | 683 | return dev->ifindex; |
a54acb3a ND |
684 | } |
685 | EXPORT_SYMBOL(dev_get_iflink); | |
686 | ||
fc4099f1 PS |
687 | /** |
688 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
689 | * @dev: targeted interface | |
690 | * @skb: The packet. | |
691 | * | |
692 | * For better visibility of tunnel traffic OVS needs to retrieve | |
693 | * egress tunnel information for a packet. Following API allows | |
694 | * user to get this info. | |
695 | */ | |
696 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
697 | { | |
698 | struct ip_tunnel_info *info; | |
699 | ||
700 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
701 | return -EINVAL; | |
702 | ||
703 | info = skb_tunnel_info_unclone(skb); | |
704 | if (!info) | |
705 | return -ENOMEM; | |
706 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
707 | return -EINVAL; | |
708 | ||
709 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
710 | } | |
711 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
712 | ||
1da177e4 LT |
713 | /** |
714 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 715 | * @net: the applicable net namespace |
1da177e4 LT |
716 | * @name: name to find |
717 | * | |
718 | * Find an interface by name. Must be called under RTNL semaphore | |
719 | * or @dev_base_lock. If the name is found a pointer to the device | |
720 | * is returned. If the name is not found then %NULL is returned. The | |
721 | * reference counters are not incremented so the caller must be | |
722 | * careful with locks. | |
723 | */ | |
724 | ||
881d966b | 725 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 726 | { |
0bd8d536 ED |
727 | struct net_device *dev; |
728 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 729 | |
b67bfe0d | 730 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
731 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
732 | return dev; | |
0bd8d536 | 733 | |
1da177e4 LT |
734 | return NULL; |
735 | } | |
d1b19dff | 736 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 737 | |
72c9528b ED |
738 | /** |
739 | * dev_get_by_name_rcu - find a device by its name | |
740 | * @net: the applicable net namespace | |
741 | * @name: name to find | |
742 | * | |
743 | * Find an interface by name. | |
744 | * If the name is found a pointer to the device is returned. | |
745 | * If the name is not found then %NULL is returned. | |
746 | * The reference counters are not incremented so the caller must be | |
747 | * careful with locks. The caller must hold RCU lock. | |
748 | */ | |
749 | ||
750 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
751 | { | |
72c9528b ED |
752 | struct net_device *dev; |
753 | struct hlist_head *head = dev_name_hash(net, name); | |
754 | ||
b67bfe0d | 755 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
756 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
757 | return dev; | |
758 | ||
759 | return NULL; | |
760 | } | |
761 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
762 | ||
1da177e4 LT |
763 | /** |
764 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 765 | * @net: the applicable net namespace |
1da177e4 LT |
766 | * @name: name to find |
767 | * | |
768 | * Find an interface by name. This can be called from any | |
769 | * context and does its own locking. The returned handle has | |
770 | * the usage count incremented and the caller must use dev_put() to | |
771 | * release it when it is no longer needed. %NULL is returned if no | |
772 | * matching device is found. | |
773 | */ | |
774 | ||
881d966b | 775 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
776 | { |
777 | struct net_device *dev; | |
778 | ||
72c9528b ED |
779 | rcu_read_lock(); |
780 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
781 | if (dev) |
782 | dev_hold(dev); | |
72c9528b | 783 | rcu_read_unlock(); |
1da177e4 LT |
784 | return dev; |
785 | } | |
d1b19dff | 786 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
787 | |
788 | /** | |
789 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 790 | * @net: the applicable net namespace |
1da177e4 LT |
791 | * @ifindex: index of device |
792 | * | |
793 | * Search for an interface by index. Returns %NULL if the device | |
794 | * is not found or a pointer to the device. The device has not | |
795 | * had its reference counter increased so the caller must be careful | |
796 | * about locking. The caller must hold either the RTNL semaphore | |
797 | * or @dev_base_lock. | |
798 | */ | |
799 | ||
881d966b | 800 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 801 | { |
0bd8d536 ED |
802 | struct net_device *dev; |
803 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 804 | |
b67bfe0d | 805 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
806 | if (dev->ifindex == ifindex) |
807 | return dev; | |
0bd8d536 | 808 | |
1da177e4 LT |
809 | return NULL; |
810 | } | |
d1b19dff | 811 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 812 | |
fb699dfd ED |
813 | /** |
814 | * dev_get_by_index_rcu - find a device by its ifindex | |
815 | * @net: the applicable net namespace | |
816 | * @ifindex: index of device | |
817 | * | |
818 | * Search for an interface by index. Returns %NULL if the device | |
819 | * is not found or a pointer to the device. The device has not | |
820 | * had its reference counter increased so the caller must be careful | |
821 | * about locking. The caller must hold RCU lock. | |
822 | */ | |
823 | ||
824 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
825 | { | |
fb699dfd ED |
826 | struct net_device *dev; |
827 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
828 | ||
b67bfe0d | 829 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
830 | if (dev->ifindex == ifindex) |
831 | return dev; | |
832 | ||
833 | return NULL; | |
834 | } | |
835 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
836 | ||
1da177e4 LT |
837 | |
838 | /** | |
839 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 840 | * @net: the applicable net namespace |
1da177e4 LT |
841 | * @ifindex: index of device |
842 | * | |
843 | * Search for an interface by index. Returns NULL if the device | |
844 | * is not found or a pointer to the device. The device returned has | |
845 | * had a reference added and the pointer is safe until the user calls | |
846 | * dev_put to indicate they have finished with it. | |
847 | */ | |
848 | ||
881d966b | 849 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
850 | { |
851 | struct net_device *dev; | |
852 | ||
fb699dfd ED |
853 | rcu_read_lock(); |
854 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
855 | if (dev) |
856 | dev_hold(dev); | |
fb699dfd | 857 | rcu_read_unlock(); |
1da177e4 LT |
858 | return dev; |
859 | } | |
d1b19dff | 860 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 861 | |
5dbe7c17 NS |
862 | /** |
863 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
864 | * @net: network namespace | |
865 | * @name: a pointer to the buffer where the name will be stored. | |
866 | * @ifindex: the ifindex of the interface to get the name from. | |
867 | * | |
868 | * The use of raw_seqcount_begin() and cond_resched() before | |
869 | * retrying is required as we want to give the writers a chance | |
870 | * to complete when CONFIG_PREEMPT is not set. | |
871 | */ | |
872 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
873 | { | |
874 | struct net_device *dev; | |
875 | unsigned int seq; | |
876 | ||
877 | retry: | |
878 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
879 | rcu_read_lock(); | |
880 | dev = dev_get_by_index_rcu(net, ifindex); | |
881 | if (!dev) { | |
882 | rcu_read_unlock(); | |
883 | return -ENODEV; | |
884 | } | |
885 | ||
886 | strcpy(name, dev->name); | |
887 | rcu_read_unlock(); | |
888 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
889 | cond_resched(); | |
890 | goto retry; | |
891 | } | |
892 | ||
893 | return 0; | |
894 | } | |
895 | ||
1da177e4 | 896 | /** |
941666c2 | 897 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 898 | * @net: the applicable net namespace |
1da177e4 LT |
899 | * @type: media type of device |
900 | * @ha: hardware address | |
901 | * | |
902 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
903 | * is not found or a pointer to the device. |
904 | * The caller must hold RCU or RTNL. | |
941666c2 | 905 | * The returned device has not had its ref count increased |
1da177e4 LT |
906 | * and the caller must therefore be careful about locking |
907 | * | |
1da177e4 LT |
908 | */ |
909 | ||
941666c2 ED |
910 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
911 | const char *ha) | |
1da177e4 LT |
912 | { |
913 | struct net_device *dev; | |
914 | ||
941666c2 | 915 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
916 | if (dev->type == type && |
917 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
918 | return dev; |
919 | ||
920 | return NULL; | |
1da177e4 | 921 | } |
941666c2 | 922 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 923 | |
881d966b | 924 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
925 | { |
926 | struct net_device *dev; | |
927 | ||
4e9cac2b | 928 | ASSERT_RTNL(); |
881d966b | 929 | for_each_netdev(net, dev) |
4e9cac2b | 930 | if (dev->type == type) |
7562f876 PE |
931 | return dev; |
932 | ||
933 | return NULL; | |
4e9cac2b | 934 | } |
4e9cac2b PM |
935 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
936 | ||
881d966b | 937 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 938 | { |
99fe3c39 | 939 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 940 | |
99fe3c39 ED |
941 | rcu_read_lock(); |
942 | for_each_netdev_rcu(net, dev) | |
943 | if (dev->type == type) { | |
944 | dev_hold(dev); | |
945 | ret = dev; | |
946 | break; | |
947 | } | |
948 | rcu_read_unlock(); | |
949 | return ret; | |
1da177e4 | 950 | } |
1da177e4 LT |
951 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
952 | ||
953 | /** | |
6c555490 | 954 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 955 | * @net: the applicable net namespace |
1da177e4 LT |
956 | * @if_flags: IFF_* values |
957 | * @mask: bitmask of bits in if_flags to check | |
958 | * | |
959 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 960 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 961 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
962 | */ |
963 | ||
6c555490 WC |
964 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
965 | unsigned short mask) | |
1da177e4 | 966 | { |
7562f876 | 967 | struct net_device *dev, *ret; |
1da177e4 | 968 | |
6c555490 WC |
969 | ASSERT_RTNL(); |
970 | ||
7562f876 | 971 | ret = NULL; |
6c555490 | 972 | for_each_netdev(net, dev) { |
1da177e4 | 973 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 974 | ret = dev; |
1da177e4 LT |
975 | break; |
976 | } | |
977 | } | |
7562f876 | 978 | return ret; |
1da177e4 | 979 | } |
6c555490 | 980 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
981 | |
982 | /** | |
983 | * dev_valid_name - check if name is okay for network device | |
984 | * @name: name string | |
985 | * | |
986 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
987 | * to allow sysfs to work. We also disallow any kind of |
988 | * whitespace. | |
1da177e4 | 989 | */ |
95f050bf | 990 | bool dev_valid_name(const char *name) |
1da177e4 | 991 | { |
c7fa9d18 | 992 | if (*name == '\0') |
95f050bf | 993 | return false; |
b6fe17d6 | 994 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 995 | return false; |
c7fa9d18 | 996 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 997 | return false; |
c7fa9d18 DM |
998 | |
999 | while (*name) { | |
a4176a93 | 1000 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1001 | return false; |
c7fa9d18 DM |
1002 | name++; |
1003 | } | |
95f050bf | 1004 | return true; |
1da177e4 | 1005 | } |
d1b19dff | 1006 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1007 | |
1008 | /** | |
b267b179 EB |
1009 | * __dev_alloc_name - allocate a name for a device |
1010 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1011 | * @name: name format string |
b267b179 | 1012 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1013 | * |
1014 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1015 | * id. It scans list of devices to build up a free map, then chooses |
1016 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1017 | * while allocating the name and adding the device in order to avoid | |
1018 | * duplicates. | |
1019 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1020 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1021 | */ |
1022 | ||
b267b179 | 1023 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1024 | { |
1025 | int i = 0; | |
1da177e4 LT |
1026 | const char *p; |
1027 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1028 | unsigned long *inuse; |
1da177e4 LT |
1029 | struct net_device *d; |
1030 | ||
1031 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
1032 | if (p) { | |
1033 | /* | |
1034 | * Verify the string as this thing may have come from | |
1035 | * the user. There must be either one "%d" and no other "%" | |
1036 | * characters. | |
1037 | */ | |
1038 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1039 | return -EINVAL; | |
1040 | ||
1041 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1042 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1043 | if (!inuse) |
1044 | return -ENOMEM; | |
1045 | ||
881d966b | 1046 | for_each_netdev(net, d) { |
1da177e4 LT |
1047 | if (!sscanf(d->name, name, &i)) |
1048 | continue; | |
1049 | if (i < 0 || i >= max_netdevices) | |
1050 | continue; | |
1051 | ||
1052 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1053 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1054 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1055 | set_bit(i, inuse); | |
1056 | } | |
1057 | ||
1058 | i = find_first_zero_bit(inuse, max_netdevices); | |
1059 | free_page((unsigned long) inuse); | |
1060 | } | |
1061 | ||
d9031024 OP |
1062 | if (buf != name) |
1063 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1064 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1065 | return i; |
1da177e4 LT |
1066 | |
1067 | /* It is possible to run out of possible slots | |
1068 | * when the name is long and there isn't enough space left | |
1069 | * for the digits, or if all bits are used. | |
1070 | */ | |
1071 | return -ENFILE; | |
1072 | } | |
1073 | ||
b267b179 EB |
1074 | /** |
1075 | * dev_alloc_name - allocate a name for a device | |
1076 | * @dev: device | |
1077 | * @name: name format string | |
1078 | * | |
1079 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1080 | * id. It scans list of devices to build up a free map, then chooses | |
1081 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1082 | * while allocating the name and adding the device in order to avoid | |
1083 | * duplicates. | |
1084 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1085 | * Returns the number of the unit assigned or a negative errno code. | |
1086 | */ | |
1087 | ||
1088 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1089 | { | |
1090 | char buf[IFNAMSIZ]; | |
1091 | struct net *net; | |
1092 | int ret; | |
1093 | ||
c346dca1 YH |
1094 | BUG_ON(!dev_net(dev)); |
1095 | net = dev_net(dev); | |
b267b179 EB |
1096 | ret = __dev_alloc_name(net, name, buf); |
1097 | if (ret >= 0) | |
1098 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1099 | return ret; | |
1100 | } | |
d1b19dff | 1101 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1102 | |
828de4f6 G |
1103 | static int dev_alloc_name_ns(struct net *net, |
1104 | struct net_device *dev, | |
1105 | const char *name) | |
d9031024 | 1106 | { |
828de4f6 G |
1107 | char buf[IFNAMSIZ]; |
1108 | int ret; | |
8ce6cebc | 1109 | |
828de4f6 G |
1110 | ret = __dev_alloc_name(net, name, buf); |
1111 | if (ret >= 0) | |
1112 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1113 | return ret; | |
1114 | } | |
1115 | ||
1116 | static int dev_get_valid_name(struct net *net, | |
1117 | struct net_device *dev, | |
1118 | const char *name) | |
1119 | { | |
1120 | BUG_ON(!net); | |
8ce6cebc | 1121 | |
d9031024 OP |
1122 | if (!dev_valid_name(name)) |
1123 | return -EINVAL; | |
1124 | ||
1c5cae81 | 1125 | if (strchr(name, '%')) |
828de4f6 | 1126 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1127 | else if (__dev_get_by_name(net, name)) |
1128 | return -EEXIST; | |
8ce6cebc DL |
1129 | else if (dev->name != name) |
1130 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1131 | |
1132 | return 0; | |
1133 | } | |
1da177e4 LT |
1134 | |
1135 | /** | |
1136 | * dev_change_name - change name of a device | |
1137 | * @dev: device | |
1138 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1139 | * | |
1140 | * Change name of a device, can pass format strings "eth%d". | |
1141 | * for wildcarding. | |
1142 | */ | |
cf04a4c7 | 1143 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1144 | { |
238fa362 | 1145 | unsigned char old_assign_type; |
fcc5a03a | 1146 | char oldname[IFNAMSIZ]; |
1da177e4 | 1147 | int err = 0; |
fcc5a03a | 1148 | int ret; |
881d966b | 1149 | struct net *net; |
1da177e4 LT |
1150 | |
1151 | ASSERT_RTNL(); | |
c346dca1 | 1152 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1153 | |
c346dca1 | 1154 | net = dev_net(dev); |
1da177e4 LT |
1155 | if (dev->flags & IFF_UP) |
1156 | return -EBUSY; | |
1157 | ||
30e6c9fa | 1158 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1159 | |
1160 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1161 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1162 | return 0; |
c91f6df2 | 1163 | } |
c8d90dca | 1164 | |
fcc5a03a HX |
1165 | memcpy(oldname, dev->name, IFNAMSIZ); |
1166 | ||
828de4f6 | 1167 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1168 | if (err < 0) { |
30e6c9fa | 1169 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1170 | return err; |
c91f6df2 | 1171 | } |
1da177e4 | 1172 | |
6fe82a39 VF |
1173 | if (oldname[0] && !strchr(oldname, '%')) |
1174 | netdev_info(dev, "renamed from %s\n", oldname); | |
1175 | ||
238fa362 TG |
1176 | old_assign_type = dev->name_assign_type; |
1177 | dev->name_assign_type = NET_NAME_RENAMED; | |
1178 | ||
fcc5a03a | 1179 | rollback: |
a1b3f594 EB |
1180 | ret = device_rename(&dev->dev, dev->name); |
1181 | if (ret) { | |
1182 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1183 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1184 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1185 | return ret; |
dcc99773 | 1186 | } |
7f988eab | 1187 | |
30e6c9fa | 1188 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1189 | |
5bb025fa VF |
1190 | netdev_adjacent_rename_links(dev, oldname); |
1191 | ||
7f988eab | 1192 | write_lock_bh(&dev_base_lock); |
372b2312 | 1193 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1194 | write_unlock_bh(&dev_base_lock); |
1195 | ||
1196 | synchronize_rcu(); | |
1197 | ||
1198 | write_lock_bh(&dev_base_lock); | |
1199 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1200 | write_unlock_bh(&dev_base_lock); |
1201 | ||
056925ab | 1202 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1203 | ret = notifier_to_errno(ret); |
1204 | ||
1205 | if (ret) { | |
91e9c07b ED |
1206 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1207 | if (err >= 0) { | |
fcc5a03a | 1208 | err = ret; |
30e6c9fa | 1209 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1210 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1211 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1212 | dev->name_assign_type = old_assign_type; |
1213 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1214 | goto rollback; |
91e9c07b | 1215 | } else { |
7b6cd1ce | 1216 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1217 | dev->name, ret); |
fcc5a03a HX |
1218 | } |
1219 | } | |
1da177e4 LT |
1220 | |
1221 | return err; | |
1222 | } | |
1223 | ||
0b815a1a SH |
1224 | /** |
1225 | * dev_set_alias - change ifalias of a device | |
1226 | * @dev: device | |
1227 | * @alias: name up to IFALIASZ | |
f0db275a | 1228 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1229 | * |
1230 | * Set ifalias for a device, | |
1231 | */ | |
1232 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1233 | { | |
7364e445 AK |
1234 | char *new_ifalias; |
1235 | ||
0b815a1a SH |
1236 | ASSERT_RTNL(); |
1237 | ||
1238 | if (len >= IFALIASZ) | |
1239 | return -EINVAL; | |
1240 | ||
96ca4a2c | 1241 | if (!len) { |
388dfc2d SK |
1242 | kfree(dev->ifalias); |
1243 | dev->ifalias = NULL; | |
96ca4a2c OH |
1244 | return 0; |
1245 | } | |
1246 | ||
7364e445 AK |
1247 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1248 | if (!new_ifalias) | |
0b815a1a | 1249 | return -ENOMEM; |
7364e445 | 1250 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1251 | |
1252 | strlcpy(dev->ifalias, alias, len+1); | |
1253 | return len; | |
1254 | } | |
1255 | ||
1256 | ||
d8a33ac4 | 1257 | /** |
3041a069 | 1258 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1259 | * @dev: device to cause notification |
1260 | * | |
1261 | * Called to indicate a device has changed features. | |
1262 | */ | |
1263 | void netdev_features_change(struct net_device *dev) | |
1264 | { | |
056925ab | 1265 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1266 | } |
1267 | EXPORT_SYMBOL(netdev_features_change); | |
1268 | ||
1da177e4 LT |
1269 | /** |
1270 | * netdev_state_change - device changes state | |
1271 | * @dev: device to cause notification | |
1272 | * | |
1273 | * Called to indicate a device has changed state. This function calls | |
1274 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1275 | * to the routing socket. | |
1276 | */ | |
1277 | void netdev_state_change(struct net_device *dev) | |
1278 | { | |
1279 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1280 | struct netdev_notifier_change_info change_info; |
1281 | ||
1282 | change_info.flags_changed = 0; | |
1283 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1284 | &change_info.info); | |
7f294054 | 1285 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1286 | } |
1287 | } | |
d1b19dff | 1288 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1289 | |
ee89bab1 AW |
1290 | /** |
1291 | * netdev_notify_peers - notify network peers about existence of @dev | |
1292 | * @dev: network device | |
1293 | * | |
1294 | * Generate traffic such that interested network peers are aware of | |
1295 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1296 | * a device wants to inform the rest of the network about some sort of | |
1297 | * reconfiguration such as a failover event or virtual machine | |
1298 | * migration. | |
1299 | */ | |
1300 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1301 | { |
ee89bab1 AW |
1302 | rtnl_lock(); |
1303 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1304 | rtnl_unlock(); | |
c1da4ac7 | 1305 | } |
ee89bab1 | 1306 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1307 | |
bd380811 | 1308 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1309 | { |
d314774c | 1310 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1311 | int ret; |
1da177e4 | 1312 | |
e46b66bc BH |
1313 | ASSERT_RTNL(); |
1314 | ||
1da177e4 LT |
1315 | if (!netif_device_present(dev)) |
1316 | return -ENODEV; | |
1317 | ||
ca99ca14 NH |
1318 | /* Block netpoll from trying to do any rx path servicing. |
1319 | * If we don't do this there is a chance ndo_poll_controller | |
1320 | * or ndo_poll may be running while we open the device | |
1321 | */ | |
66b5552f | 1322 | netpoll_poll_disable(dev); |
ca99ca14 | 1323 | |
3b8bcfd5 JB |
1324 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1325 | ret = notifier_to_errno(ret); | |
1326 | if (ret) | |
1327 | return ret; | |
1328 | ||
1da177e4 | 1329 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1330 | |
d314774c SH |
1331 | if (ops->ndo_validate_addr) |
1332 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1333 | |
d314774c SH |
1334 | if (!ret && ops->ndo_open) |
1335 | ret = ops->ndo_open(dev); | |
1da177e4 | 1336 | |
66b5552f | 1337 | netpoll_poll_enable(dev); |
ca99ca14 | 1338 | |
bada339b JG |
1339 | if (ret) |
1340 | clear_bit(__LINK_STATE_START, &dev->state); | |
1341 | else { | |
1da177e4 | 1342 | dev->flags |= IFF_UP; |
4417da66 | 1343 | dev_set_rx_mode(dev); |
1da177e4 | 1344 | dev_activate(dev); |
7bf23575 | 1345 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1346 | } |
bada339b | 1347 | |
1da177e4 LT |
1348 | return ret; |
1349 | } | |
1350 | ||
1351 | /** | |
bd380811 PM |
1352 | * dev_open - prepare an interface for use. |
1353 | * @dev: device to open | |
1da177e4 | 1354 | * |
bd380811 PM |
1355 | * Takes a device from down to up state. The device's private open |
1356 | * function is invoked and then the multicast lists are loaded. Finally | |
1357 | * the device is moved into the up state and a %NETDEV_UP message is | |
1358 | * sent to the netdev notifier chain. | |
1359 | * | |
1360 | * Calling this function on an active interface is a nop. On a failure | |
1361 | * a negative errno code is returned. | |
1da177e4 | 1362 | */ |
bd380811 PM |
1363 | int dev_open(struct net_device *dev) |
1364 | { | |
1365 | int ret; | |
1366 | ||
bd380811 PM |
1367 | if (dev->flags & IFF_UP) |
1368 | return 0; | |
1369 | ||
bd380811 PM |
1370 | ret = __dev_open(dev); |
1371 | if (ret < 0) | |
1372 | return ret; | |
1373 | ||
7f294054 | 1374 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1375 | call_netdevice_notifiers(NETDEV_UP, dev); |
1376 | ||
1377 | return ret; | |
1378 | } | |
1379 | EXPORT_SYMBOL(dev_open); | |
1380 | ||
44345724 | 1381 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1382 | { |
44345724 | 1383 | struct net_device *dev; |
e46b66bc | 1384 | |
bd380811 | 1385 | ASSERT_RTNL(); |
9d5010db DM |
1386 | might_sleep(); |
1387 | ||
5cde2829 | 1388 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1389 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1390 | netpoll_poll_disable(dev); |
3f4df206 | 1391 | |
44345724 | 1392 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1393 | |
44345724 | 1394 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1395 | |
44345724 OP |
1396 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1397 | * can be even on different cpu. So just clear netif_running(). | |
1398 | * | |
1399 | * dev->stop() will invoke napi_disable() on all of it's | |
1400 | * napi_struct instances on this device. | |
1401 | */ | |
4e857c58 | 1402 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1403 | } |
1da177e4 | 1404 | |
44345724 | 1405 | dev_deactivate_many(head); |
d8b2a4d2 | 1406 | |
5cde2829 | 1407 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1408 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1409 | |
44345724 OP |
1410 | /* |
1411 | * Call the device specific close. This cannot fail. | |
1412 | * Only if device is UP | |
1413 | * | |
1414 | * We allow it to be called even after a DETACH hot-plug | |
1415 | * event. | |
1416 | */ | |
1417 | if (ops->ndo_stop) | |
1418 | ops->ndo_stop(dev); | |
1419 | ||
44345724 | 1420 | dev->flags &= ~IFF_UP; |
66b5552f | 1421 | netpoll_poll_enable(dev); |
44345724 OP |
1422 | } |
1423 | ||
1424 | return 0; | |
1425 | } | |
1426 | ||
1427 | static int __dev_close(struct net_device *dev) | |
1428 | { | |
f87e6f47 | 1429 | int retval; |
44345724 OP |
1430 | LIST_HEAD(single); |
1431 | ||
5cde2829 | 1432 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1433 | retval = __dev_close_many(&single); |
1434 | list_del(&single); | |
ca99ca14 | 1435 | |
f87e6f47 | 1436 | return retval; |
44345724 OP |
1437 | } |
1438 | ||
99c4a26a | 1439 | int dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1440 | { |
1441 | struct net_device *dev, *tmp; | |
1da177e4 | 1442 | |
5cde2829 EB |
1443 | /* Remove the devices that don't need to be closed */ |
1444 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1445 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1446 | list_del_init(&dev->close_list); |
44345724 OP |
1447 | |
1448 | __dev_close_many(head); | |
1da177e4 | 1449 | |
5cde2829 | 1450 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1451 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1452 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1453 | if (unlink) |
1454 | list_del_init(&dev->close_list); | |
44345724 | 1455 | } |
bd380811 PM |
1456 | |
1457 | return 0; | |
1458 | } | |
99c4a26a | 1459 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1460 | |
1461 | /** | |
1462 | * dev_close - shutdown an interface. | |
1463 | * @dev: device to shutdown | |
1464 | * | |
1465 | * This function moves an active device into down state. A | |
1466 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1467 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1468 | * chain. | |
1469 | */ | |
1470 | int dev_close(struct net_device *dev) | |
1471 | { | |
e14a5993 ED |
1472 | if (dev->flags & IFF_UP) { |
1473 | LIST_HEAD(single); | |
1da177e4 | 1474 | |
5cde2829 | 1475 | list_add(&dev->close_list, &single); |
99c4a26a | 1476 | dev_close_many(&single, true); |
e14a5993 ED |
1477 | list_del(&single); |
1478 | } | |
da6e378b | 1479 | return 0; |
1da177e4 | 1480 | } |
d1b19dff | 1481 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1482 | |
1483 | ||
0187bdfb BH |
1484 | /** |
1485 | * dev_disable_lro - disable Large Receive Offload on a device | |
1486 | * @dev: device | |
1487 | * | |
1488 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1489 | * called under RTNL. This is needed if received packets may be | |
1490 | * forwarded to another interface. | |
1491 | */ | |
1492 | void dev_disable_lro(struct net_device *dev) | |
1493 | { | |
fbe168ba MK |
1494 | struct net_device *lower_dev; |
1495 | struct list_head *iter; | |
529d0489 | 1496 | |
bc5787c6 MM |
1497 | dev->wanted_features &= ~NETIF_F_LRO; |
1498 | netdev_update_features(dev); | |
27660515 | 1499 | |
22d5969f MM |
1500 | if (unlikely(dev->features & NETIF_F_LRO)) |
1501 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1502 | |
1503 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1504 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1505 | } |
1506 | EXPORT_SYMBOL(dev_disable_lro); | |
1507 | ||
351638e7 JP |
1508 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1509 | struct net_device *dev) | |
1510 | { | |
1511 | struct netdev_notifier_info info; | |
1512 | ||
1513 | netdev_notifier_info_init(&info, dev); | |
1514 | return nb->notifier_call(nb, val, &info); | |
1515 | } | |
0187bdfb | 1516 | |
881d966b EB |
1517 | static int dev_boot_phase = 1; |
1518 | ||
1da177e4 LT |
1519 | /** |
1520 | * register_netdevice_notifier - register a network notifier block | |
1521 | * @nb: notifier | |
1522 | * | |
1523 | * Register a notifier to be called when network device events occur. | |
1524 | * The notifier passed is linked into the kernel structures and must | |
1525 | * not be reused until it has been unregistered. A negative errno code | |
1526 | * is returned on a failure. | |
1527 | * | |
1528 | * When registered all registration and up events are replayed | |
4ec93edb | 1529 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1530 | * view of the network device list. |
1531 | */ | |
1532 | ||
1533 | int register_netdevice_notifier(struct notifier_block *nb) | |
1534 | { | |
1535 | struct net_device *dev; | |
fcc5a03a | 1536 | struct net_device *last; |
881d966b | 1537 | struct net *net; |
1da177e4 LT |
1538 | int err; |
1539 | ||
1540 | rtnl_lock(); | |
f07d5b94 | 1541 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1542 | if (err) |
1543 | goto unlock; | |
881d966b EB |
1544 | if (dev_boot_phase) |
1545 | goto unlock; | |
1546 | for_each_net(net) { | |
1547 | for_each_netdev(net, dev) { | |
351638e7 | 1548 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1549 | err = notifier_to_errno(err); |
1550 | if (err) | |
1551 | goto rollback; | |
1552 | ||
1553 | if (!(dev->flags & IFF_UP)) | |
1554 | continue; | |
1da177e4 | 1555 | |
351638e7 | 1556 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1557 | } |
1da177e4 | 1558 | } |
fcc5a03a HX |
1559 | |
1560 | unlock: | |
1da177e4 LT |
1561 | rtnl_unlock(); |
1562 | return err; | |
fcc5a03a HX |
1563 | |
1564 | rollback: | |
1565 | last = dev; | |
881d966b EB |
1566 | for_each_net(net) { |
1567 | for_each_netdev(net, dev) { | |
1568 | if (dev == last) | |
8f891489 | 1569 | goto outroll; |
fcc5a03a | 1570 | |
881d966b | 1571 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1572 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1573 | dev); | |
1574 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1575 | } |
351638e7 | 1576 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1577 | } |
fcc5a03a | 1578 | } |
c67625a1 | 1579 | |
8f891489 | 1580 | outroll: |
c67625a1 | 1581 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1582 | goto unlock; |
1da177e4 | 1583 | } |
d1b19dff | 1584 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1585 | |
1586 | /** | |
1587 | * unregister_netdevice_notifier - unregister a network notifier block | |
1588 | * @nb: notifier | |
1589 | * | |
1590 | * Unregister a notifier previously registered by | |
1591 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1592 | * kernel structures and may then be reused. A negative errno code | |
1593 | * is returned on a failure. | |
7d3d43da EB |
1594 | * |
1595 | * After unregistering unregister and down device events are synthesized | |
1596 | * for all devices on the device list to the removed notifier to remove | |
1597 | * the need for special case cleanup code. | |
1da177e4 LT |
1598 | */ |
1599 | ||
1600 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1601 | { | |
7d3d43da EB |
1602 | struct net_device *dev; |
1603 | struct net *net; | |
9f514950 HX |
1604 | int err; |
1605 | ||
1606 | rtnl_lock(); | |
f07d5b94 | 1607 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1608 | if (err) |
1609 | goto unlock; | |
1610 | ||
1611 | for_each_net(net) { | |
1612 | for_each_netdev(net, dev) { | |
1613 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1614 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1615 | dev); | |
1616 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1617 | } |
351638e7 | 1618 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1619 | } |
1620 | } | |
1621 | unlock: | |
9f514950 HX |
1622 | rtnl_unlock(); |
1623 | return err; | |
1da177e4 | 1624 | } |
d1b19dff | 1625 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1626 | |
351638e7 JP |
1627 | /** |
1628 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1629 | * @val: value passed unmodified to notifier function | |
1630 | * @dev: net_device pointer passed unmodified to notifier function | |
1631 | * @info: notifier information data | |
1632 | * | |
1633 | * Call all network notifier blocks. Parameters and return value | |
1634 | * are as for raw_notifier_call_chain(). | |
1635 | */ | |
1636 | ||
1d143d9f | 1637 | static int call_netdevice_notifiers_info(unsigned long val, |
1638 | struct net_device *dev, | |
1639 | struct netdev_notifier_info *info) | |
351638e7 JP |
1640 | { |
1641 | ASSERT_RTNL(); | |
1642 | netdev_notifier_info_init(info, dev); | |
1643 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1644 | } | |
351638e7 | 1645 | |
1da177e4 LT |
1646 | /** |
1647 | * call_netdevice_notifiers - call all network notifier blocks | |
1648 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1649 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1650 | * |
1651 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1652 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1653 | */ |
1654 | ||
ad7379d4 | 1655 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1656 | { |
351638e7 JP |
1657 | struct netdev_notifier_info info; |
1658 | ||
1659 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1660 | } |
edf947f1 | 1661 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1662 | |
1cf51900 | 1663 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
1664 | static struct static_key ingress_needed __read_mostly; |
1665 | ||
1666 | void net_inc_ingress_queue(void) | |
1667 | { | |
1668 | static_key_slow_inc(&ingress_needed); | |
1669 | } | |
1670 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1671 | ||
1672 | void net_dec_ingress_queue(void) | |
1673 | { | |
1674 | static_key_slow_dec(&ingress_needed); | |
1675 | } | |
1676 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1677 | #endif | |
1678 | ||
1f211a1b DB |
1679 | #ifdef CONFIG_NET_EGRESS |
1680 | static struct static_key egress_needed __read_mostly; | |
1681 | ||
1682 | void net_inc_egress_queue(void) | |
1683 | { | |
1684 | static_key_slow_inc(&egress_needed); | |
1685 | } | |
1686 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
1687 | ||
1688 | void net_dec_egress_queue(void) | |
1689 | { | |
1690 | static_key_slow_dec(&egress_needed); | |
1691 | } | |
1692 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
1693 | #endif | |
1694 | ||
c5905afb | 1695 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1696 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1697 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1698 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1699 | * static_key_slow_dec() calls. |
b90e5794 ED |
1700 | */ |
1701 | static atomic_t netstamp_needed_deferred; | |
1702 | #endif | |
1da177e4 LT |
1703 | |
1704 | void net_enable_timestamp(void) | |
1705 | { | |
b90e5794 ED |
1706 | #ifdef HAVE_JUMP_LABEL |
1707 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1708 | ||
1709 | if (deferred) { | |
1710 | while (--deferred) | |
c5905afb | 1711 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1712 | return; |
1713 | } | |
1714 | #endif | |
c5905afb | 1715 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1716 | } |
d1b19dff | 1717 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1718 | |
1719 | void net_disable_timestamp(void) | |
1720 | { | |
b90e5794 ED |
1721 | #ifdef HAVE_JUMP_LABEL |
1722 | if (in_interrupt()) { | |
1723 | atomic_inc(&netstamp_needed_deferred); | |
1724 | return; | |
1725 | } | |
1726 | #endif | |
c5905afb | 1727 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1728 | } |
d1b19dff | 1729 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1730 | |
3b098e2d | 1731 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1732 | { |
588f0330 | 1733 | skb->tstamp.tv64 = 0; |
c5905afb | 1734 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1735 | __net_timestamp(skb); |
1da177e4 LT |
1736 | } |
1737 | ||
588f0330 | 1738 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1739 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1740 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1741 | __net_timestamp(SKB); \ | |
1742 | } \ | |
3b098e2d | 1743 | |
1ee481fb | 1744 | bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb) |
79b569f0 DL |
1745 | { |
1746 | unsigned int len; | |
1747 | ||
1748 | if (!(dev->flags & IFF_UP)) | |
1749 | return false; | |
1750 | ||
1751 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1752 | if (skb->len <= len) | |
1753 | return true; | |
1754 | ||
1755 | /* if TSO is enabled, we don't care about the length as the packet | |
1756 | * could be forwarded without being segmented before | |
1757 | */ | |
1758 | if (skb_is_gso(skb)) | |
1759 | return true; | |
1760 | ||
1761 | return false; | |
1762 | } | |
1ee481fb | 1763 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1764 | |
a0265d28 HX |
1765 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1766 | { | |
bbbf2df0 WB |
1767 | if (skb_orphan_frags(skb, GFP_ATOMIC) || |
1768 | unlikely(!is_skb_forwardable(dev, skb))) { | |
a0265d28 HX |
1769 | atomic_long_inc(&dev->rx_dropped); |
1770 | kfree_skb(skb); | |
1771 | return NET_RX_DROP; | |
1772 | } | |
1773 | ||
1774 | skb_scrub_packet(skb, true); | |
08b4b8ea | 1775 | skb->priority = 0; |
a0265d28 | 1776 | skb->protocol = eth_type_trans(skb, dev); |
2c26d34b | 1777 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); |
a0265d28 HX |
1778 | |
1779 | return 0; | |
1780 | } | |
1781 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1782 | ||
44540960 AB |
1783 | /** |
1784 | * dev_forward_skb - loopback an skb to another netif | |
1785 | * | |
1786 | * @dev: destination network device | |
1787 | * @skb: buffer to forward | |
1788 | * | |
1789 | * return values: | |
1790 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1791 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1792 | * |
1793 | * dev_forward_skb can be used for injecting an skb from the | |
1794 | * start_xmit function of one device into the receive queue | |
1795 | * of another device. | |
1796 | * | |
1797 | * The receiving device may be in another namespace, so | |
1798 | * we have to clear all information in the skb that could | |
1799 | * impact namespace isolation. | |
1800 | */ | |
1801 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1802 | { | |
a0265d28 | 1803 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1804 | } |
1805 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1806 | ||
71d9dec2 CG |
1807 | static inline int deliver_skb(struct sk_buff *skb, |
1808 | struct packet_type *pt_prev, | |
1809 | struct net_device *orig_dev) | |
1810 | { | |
1080e512 MT |
1811 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1812 | return -ENOMEM; | |
71d9dec2 CG |
1813 | atomic_inc(&skb->users); |
1814 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1815 | } | |
1816 | ||
7866a621 SN |
1817 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1818 | struct packet_type **pt, | |
fbcb2170 JP |
1819 | struct net_device *orig_dev, |
1820 | __be16 type, | |
7866a621 SN |
1821 | struct list_head *ptype_list) |
1822 | { | |
1823 | struct packet_type *ptype, *pt_prev = *pt; | |
1824 | ||
1825 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1826 | if (ptype->type != type) | |
1827 | continue; | |
1828 | if (pt_prev) | |
fbcb2170 | 1829 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1830 | pt_prev = ptype; |
1831 | } | |
1832 | *pt = pt_prev; | |
1833 | } | |
1834 | ||
c0de08d0 EL |
1835 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1836 | { | |
a3d744e9 | 1837 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1838 | return false; |
1839 | ||
1840 | if (ptype->id_match) | |
1841 | return ptype->id_match(ptype, skb->sk); | |
1842 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1843 | return true; | |
1844 | ||
1845 | return false; | |
1846 | } | |
1847 | ||
1da177e4 LT |
1848 | /* |
1849 | * Support routine. Sends outgoing frames to any network | |
1850 | * taps currently in use. | |
1851 | */ | |
1852 | ||
f6a78bfc | 1853 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1854 | { |
1855 | struct packet_type *ptype; | |
71d9dec2 CG |
1856 | struct sk_buff *skb2 = NULL; |
1857 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1858 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1859 | |
1da177e4 | 1860 | rcu_read_lock(); |
7866a621 SN |
1861 | again: |
1862 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1863 | /* Never send packets back to the socket |
1864 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1865 | */ | |
7866a621 SN |
1866 | if (skb_loop_sk(ptype, skb)) |
1867 | continue; | |
71d9dec2 | 1868 | |
7866a621 SN |
1869 | if (pt_prev) { |
1870 | deliver_skb(skb2, pt_prev, skb->dev); | |
1871 | pt_prev = ptype; | |
1872 | continue; | |
1873 | } | |
1da177e4 | 1874 | |
7866a621 SN |
1875 | /* need to clone skb, done only once */ |
1876 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1877 | if (!skb2) | |
1878 | goto out_unlock; | |
70978182 | 1879 | |
7866a621 | 1880 | net_timestamp_set(skb2); |
1da177e4 | 1881 | |
7866a621 SN |
1882 | /* skb->nh should be correctly |
1883 | * set by sender, so that the second statement is | |
1884 | * just protection against buggy protocols. | |
1885 | */ | |
1886 | skb_reset_mac_header(skb2); | |
1887 | ||
1888 | if (skb_network_header(skb2) < skb2->data || | |
1889 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1890 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1891 | ntohs(skb2->protocol), | |
1892 | dev->name); | |
1893 | skb_reset_network_header(skb2); | |
1da177e4 | 1894 | } |
7866a621 SN |
1895 | |
1896 | skb2->transport_header = skb2->network_header; | |
1897 | skb2->pkt_type = PACKET_OUTGOING; | |
1898 | pt_prev = ptype; | |
1899 | } | |
1900 | ||
1901 | if (ptype_list == &ptype_all) { | |
1902 | ptype_list = &dev->ptype_all; | |
1903 | goto again; | |
1da177e4 | 1904 | } |
7866a621 | 1905 | out_unlock: |
71d9dec2 CG |
1906 | if (pt_prev) |
1907 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1908 | rcu_read_unlock(); |
1909 | } | |
1910 | ||
2c53040f BH |
1911 | /** |
1912 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1913 | * @dev: Network device |
1914 | * @txq: number of queues available | |
1915 | * | |
1916 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1917 | * valid. To resolve this verify the tc mapping remains valid and if | |
1918 | * not NULL the mapping. With no priorities mapping to this | |
1919 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1920 | * is invalid nothing can be done so disable priority mappings. If is | |
1921 | * expected that drivers will fix this mapping if they can before | |
1922 | * calling netif_set_real_num_tx_queues. | |
1923 | */ | |
bb134d22 | 1924 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1925 | { |
1926 | int i; | |
1927 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1928 | ||
1929 | /* If TC0 is invalidated disable TC mapping */ | |
1930 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1931 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1932 | dev->num_tc = 0; |
1933 | return; | |
1934 | } | |
1935 | ||
1936 | /* Invalidated prio to tc mappings set to TC0 */ | |
1937 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1938 | int q = netdev_get_prio_tc_map(dev, i); | |
1939 | ||
1940 | tc = &dev->tc_to_txq[q]; | |
1941 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1942 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1943 | i, q); | |
4f57c087 JF |
1944 | netdev_set_prio_tc_map(dev, i, 0); |
1945 | } | |
1946 | } | |
1947 | } | |
1948 | ||
537c00de AD |
1949 | #ifdef CONFIG_XPS |
1950 | static DEFINE_MUTEX(xps_map_mutex); | |
1951 | #define xmap_dereference(P) \ | |
1952 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1953 | ||
10cdc3f3 AD |
1954 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1955 | int cpu, u16 index) | |
537c00de | 1956 | { |
10cdc3f3 AD |
1957 | struct xps_map *map = NULL; |
1958 | int pos; | |
537c00de | 1959 | |
10cdc3f3 AD |
1960 | if (dev_maps) |
1961 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1962 | |
10cdc3f3 AD |
1963 | for (pos = 0; map && pos < map->len; pos++) { |
1964 | if (map->queues[pos] == index) { | |
537c00de AD |
1965 | if (map->len > 1) { |
1966 | map->queues[pos] = map->queues[--map->len]; | |
1967 | } else { | |
10cdc3f3 | 1968 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1969 | kfree_rcu(map, rcu); |
1970 | map = NULL; | |
1971 | } | |
10cdc3f3 | 1972 | break; |
537c00de | 1973 | } |
537c00de AD |
1974 | } |
1975 | ||
10cdc3f3 AD |
1976 | return map; |
1977 | } | |
1978 | ||
024e9679 | 1979 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1980 | { |
1981 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1982 | int cpu, i; |
10cdc3f3 AD |
1983 | bool active = false; |
1984 | ||
1985 | mutex_lock(&xps_map_mutex); | |
1986 | dev_maps = xmap_dereference(dev->xps_maps); | |
1987 | ||
1988 | if (!dev_maps) | |
1989 | goto out_no_maps; | |
1990 | ||
1991 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1992 | for (i = index; i < dev->num_tx_queues; i++) { |
1993 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1994 | break; | |
1995 | } | |
1996 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
1997 | active = true; |
1998 | } | |
1999 | ||
2000 | if (!active) { | |
537c00de AD |
2001 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
2002 | kfree_rcu(dev_maps, rcu); | |
2003 | } | |
2004 | ||
024e9679 AD |
2005 | for (i = index; i < dev->num_tx_queues; i++) |
2006 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
2007 | NUMA_NO_NODE); | |
2008 | ||
537c00de AD |
2009 | out_no_maps: |
2010 | mutex_unlock(&xps_map_mutex); | |
2011 | } | |
2012 | ||
01c5f864 AD |
2013 | static struct xps_map *expand_xps_map(struct xps_map *map, |
2014 | int cpu, u16 index) | |
2015 | { | |
2016 | struct xps_map *new_map; | |
2017 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2018 | int i, pos; | |
2019 | ||
2020 | for (pos = 0; map && pos < map->len; pos++) { | |
2021 | if (map->queues[pos] != index) | |
2022 | continue; | |
2023 | return map; | |
2024 | } | |
2025 | ||
2026 | /* Need to add queue to this CPU's existing map */ | |
2027 | if (map) { | |
2028 | if (pos < map->alloc_len) | |
2029 | return map; | |
2030 | ||
2031 | alloc_len = map->alloc_len * 2; | |
2032 | } | |
2033 | ||
2034 | /* Need to allocate new map to store queue on this CPU's map */ | |
2035 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2036 | cpu_to_node(cpu)); | |
2037 | if (!new_map) | |
2038 | return NULL; | |
2039 | ||
2040 | for (i = 0; i < pos; i++) | |
2041 | new_map->queues[i] = map->queues[i]; | |
2042 | new_map->alloc_len = alloc_len; | |
2043 | new_map->len = pos; | |
2044 | ||
2045 | return new_map; | |
2046 | } | |
2047 | ||
3573540c MT |
2048 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
2049 | u16 index) | |
537c00de | 2050 | { |
01c5f864 | 2051 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 2052 | struct xps_map *map, *new_map; |
537c00de | 2053 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
2054 | int cpu, numa_node_id = -2; |
2055 | bool active = false; | |
537c00de AD |
2056 | |
2057 | mutex_lock(&xps_map_mutex); | |
2058 | ||
2059 | dev_maps = xmap_dereference(dev->xps_maps); | |
2060 | ||
01c5f864 AD |
2061 | /* allocate memory for queue storage */ |
2062 | for_each_online_cpu(cpu) { | |
2063 | if (!cpumask_test_cpu(cpu, mask)) | |
2064 | continue; | |
2065 | ||
2066 | if (!new_dev_maps) | |
2067 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2068 | if (!new_dev_maps) { |
2069 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2070 | return -ENOMEM; |
2bb60cb9 | 2071 | } |
01c5f864 AD |
2072 | |
2073 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2074 | NULL; | |
2075 | ||
2076 | map = expand_xps_map(map, cpu, index); | |
2077 | if (!map) | |
2078 | goto error; | |
2079 | ||
2080 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
2081 | } | |
2082 | ||
2083 | if (!new_dev_maps) | |
2084 | goto out_no_new_maps; | |
2085 | ||
537c00de | 2086 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
2087 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2088 | /* add queue to CPU maps */ | |
2089 | int pos = 0; | |
2090 | ||
2091 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2092 | while ((pos < map->len) && (map->queues[pos] != index)) | |
2093 | pos++; | |
2094 | ||
2095 | if (pos == map->len) | |
2096 | map->queues[map->len++] = index; | |
537c00de | 2097 | #ifdef CONFIG_NUMA |
537c00de AD |
2098 | if (numa_node_id == -2) |
2099 | numa_node_id = cpu_to_node(cpu); | |
2100 | else if (numa_node_id != cpu_to_node(cpu)) | |
2101 | numa_node_id = -1; | |
537c00de | 2102 | #endif |
01c5f864 AD |
2103 | } else if (dev_maps) { |
2104 | /* fill in the new device map from the old device map */ | |
2105 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2106 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 2107 | } |
01c5f864 | 2108 | |
537c00de AD |
2109 | } |
2110 | ||
01c5f864 AD |
2111 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2112 | ||
537c00de | 2113 | /* Cleanup old maps */ |
01c5f864 AD |
2114 | if (dev_maps) { |
2115 | for_each_possible_cpu(cpu) { | |
2116 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2117 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2118 | if (map && map != new_map) | |
2119 | kfree_rcu(map, rcu); | |
2120 | } | |
537c00de | 2121 | |
01c5f864 | 2122 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
2123 | } |
2124 | ||
01c5f864 AD |
2125 | dev_maps = new_dev_maps; |
2126 | active = true; | |
537c00de | 2127 | |
01c5f864 AD |
2128 | out_no_new_maps: |
2129 | /* update Tx queue numa node */ | |
537c00de AD |
2130 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2131 | (numa_node_id >= 0) ? numa_node_id : | |
2132 | NUMA_NO_NODE); | |
2133 | ||
01c5f864 AD |
2134 | if (!dev_maps) |
2135 | goto out_no_maps; | |
2136 | ||
2137 | /* removes queue from unused CPUs */ | |
2138 | for_each_possible_cpu(cpu) { | |
2139 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2140 | continue; | |
2141 | ||
2142 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2143 | active = true; | |
2144 | } | |
2145 | ||
2146 | /* free map if not active */ | |
2147 | if (!active) { | |
2148 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2149 | kfree_rcu(dev_maps, rcu); | |
2150 | } | |
2151 | ||
2152 | out_no_maps: | |
537c00de AD |
2153 | mutex_unlock(&xps_map_mutex); |
2154 | ||
2155 | return 0; | |
2156 | error: | |
01c5f864 AD |
2157 | /* remove any maps that we added */ |
2158 | for_each_possible_cpu(cpu) { | |
2159 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2160 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2161 | NULL; | |
2162 | if (new_map && new_map != map) | |
2163 | kfree(new_map); | |
2164 | } | |
2165 | ||
537c00de AD |
2166 | mutex_unlock(&xps_map_mutex); |
2167 | ||
537c00de AD |
2168 | kfree(new_dev_maps); |
2169 | return -ENOMEM; | |
2170 | } | |
2171 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2172 | ||
2173 | #endif | |
f0796d5c JF |
2174 | /* |
2175 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2176 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2177 | */ | |
e6484930 | 2178 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2179 | { |
1d24eb48 TH |
2180 | int rc; |
2181 | ||
e6484930 TH |
2182 | if (txq < 1 || txq > dev->num_tx_queues) |
2183 | return -EINVAL; | |
f0796d5c | 2184 | |
5c56580b BH |
2185 | if (dev->reg_state == NETREG_REGISTERED || |
2186 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2187 | ASSERT_RTNL(); |
2188 | ||
1d24eb48 TH |
2189 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2190 | txq); | |
bf264145 TH |
2191 | if (rc) |
2192 | return rc; | |
2193 | ||
4f57c087 JF |
2194 | if (dev->num_tc) |
2195 | netif_setup_tc(dev, txq); | |
2196 | ||
024e9679 | 2197 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2198 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2199 | #ifdef CONFIG_XPS |
2200 | netif_reset_xps_queues_gt(dev, txq); | |
2201 | #endif | |
2202 | } | |
f0796d5c | 2203 | } |
e6484930 TH |
2204 | |
2205 | dev->real_num_tx_queues = txq; | |
2206 | return 0; | |
f0796d5c JF |
2207 | } |
2208 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2209 | |
a953be53 | 2210 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2211 | /** |
2212 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2213 | * @dev: Network device | |
2214 | * @rxq: Actual number of RX queues | |
2215 | * | |
2216 | * This must be called either with the rtnl_lock held or before | |
2217 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2218 | * negative error code. If called before registration, it always |
2219 | * succeeds. | |
62fe0b40 BH |
2220 | */ |
2221 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2222 | { | |
2223 | int rc; | |
2224 | ||
bd25fa7b TH |
2225 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2226 | return -EINVAL; | |
2227 | ||
62fe0b40 BH |
2228 | if (dev->reg_state == NETREG_REGISTERED) { |
2229 | ASSERT_RTNL(); | |
2230 | ||
62fe0b40 BH |
2231 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2232 | rxq); | |
2233 | if (rc) | |
2234 | return rc; | |
62fe0b40 BH |
2235 | } |
2236 | ||
2237 | dev->real_num_rx_queues = rxq; | |
2238 | return 0; | |
2239 | } | |
2240 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2241 | #endif | |
2242 | ||
2c53040f BH |
2243 | /** |
2244 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2245 | * |
2246 | * This routine should set an upper limit on the number of RSS queues | |
2247 | * used by default by multiqueue devices. | |
2248 | */ | |
a55b138b | 2249 | int netif_get_num_default_rss_queues(void) |
16917b87 YM |
2250 | { |
2251 | return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
2252 | } | |
2253 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2254 | ||
def82a1d | 2255 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 2256 | { |
def82a1d JP |
2257 | struct softnet_data *sd; |
2258 | unsigned long flags; | |
56079431 | 2259 | |
def82a1d | 2260 | local_irq_save(flags); |
903ceff7 | 2261 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2262 | q->next_sched = NULL; |
2263 | *sd->output_queue_tailp = q; | |
2264 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2265 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2266 | local_irq_restore(flags); | |
2267 | } | |
2268 | ||
2269 | void __netif_schedule(struct Qdisc *q) | |
2270 | { | |
2271 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2272 | __netif_reschedule(q); | |
56079431 DV |
2273 | } |
2274 | EXPORT_SYMBOL(__netif_schedule); | |
2275 | ||
e6247027 ED |
2276 | struct dev_kfree_skb_cb { |
2277 | enum skb_free_reason reason; | |
2278 | }; | |
2279 | ||
2280 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2281 | { |
e6247027 ED |
2282 | return (struct dev_kfree_skb_cb *)skb->cb; |
2283 | } | |
2284 | ||
46e5da40 JF |
2285 | void netif_schedule_queue(struct netdev_queue *txq) |
2286 | { | |
2287 | rcu_read_lock(); | |
2288 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2289 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2290 | ||
2291 | __netif_schedule(q); | |
2292 | } | |
2293 | rcu_read_unlock(); | |
2294 | } | |
2295 | EXPORT_SYMBOL(netif_schedule_queue); | |
2296 | ||
2297 | /** | |
2298 | * netif_wake_subqueue - allow sending packets on subqueue | |
2299 | * @dev: network device | |
2300 | * @queue_index: sub queue index | |
2301 | * | |
2302 | * Resume individual transmit queue of a device with multiple transmit queues. | |
2303 | */ | |
2304 | void netif_wake_subqueue(struct net_device *dev, u16 queue_index) | |
2305 | { | |
2306 | struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index); | |
2307 | ||
2308 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) { | |
2309 | struct Qdisc *q; | |
2310 | ||
2311 | rcu_read_lock(); | |
2312 | q = rcu_dereference(txq->qdisc); | |
2313 | __netif_schedule(q); | |
2314 | rcu_read_unlock(); | |
2315 | } | |
2316 | } | |
2317 | EXPORT_SYMBOL(netif_wake_subqueue); | |
2318 | ||
2319 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) | |
2320 | { | |
2321 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2322 | struct Qdisc *q; | |
2323 | ||
2324 | rcu_read_lock(); | |
2325 | q = rcu_dereference(dev_queue->qdisc); | |
2326 | __netif_schedule(q); | |
2327 | rcu_read_unlock(); | |
2328 | } | |
2329 | } | |
2330 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2331 | ||
e6247027 | 2332 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2333 | { |
e6247027 | 2334 | unsigned long flags; |
56079431 | 2335 | |
e6247027 ED |
2336 | if (likely(atomic_read(&skb->users) == 1)) { |
2337 | smp_rmb(); | |
2338 | atomic_set(&skb->users, 0); | |
2339 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2340 | return; | |
bea3348e | 2341 | } |
e6247027 ED |
2342 | get_kfree_skb_cb(skb)->reason = reason; |
2343 | local_irq_save(flags); | |
2344 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2345 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2346 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2347 | local_irq_restore(flags); | |
56079431 | 2348 | } |
e6247027 | 2349 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2350 | |
e6247027 | 2351 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2352 | { |
2353 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2354 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2355 | else |
2356 | dev_kfree_skb(skb); | |
2357 | } | |
e6247027 | 2358 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2359 | |
2360 | ||
bea3348e SH |
2361 | /** |
2362 | * netif_device_detach - mark device as removed | |
2363 | * @dev: network device | |
2364 | * | |
2365 | * Mark device as removed from system and therefore no longer available. | |
2366 | */ | |
56079431 DV |
2367 | void netif_device_detach(struct net_device *dev) |
2368 | { | |
2369 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2370 | netif_running(dev)) { | |
d543103a | 2371 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2372 | } |
2373 | } | |
2374 | EXPORT_SYMBOL(netif_device_detach); | |
2375 | ||
bea3348e SH |
2376 | /** |
2377 | * netif_device_attach - mark device as attached | |
2378 | * @dev: network device | |
2379 | * | |
2380 | * Mark device as attached from system and restart if needed. | |
2381 | */ | |
56079431 DV |
2382 | void netif_device_attach(struct net_device *dev) |
2383 | { | |
2384 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2385 | netif_running(dev)) { | |
d543103a | 2386 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2387 | __netdev_watchdog_up(dev); |
56079431 DV |
2388 | } |
2389 | } | |
2390 | EXPORT_SYMBOL(netif_device_attach); | |
2391 | ||
5605c762 JP |
2392 | /* |
2393 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2394 | * to be used as a distribution range. | |
2395 | */ | |
2396 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, | |
2397 | unsigned int num_tx_queues) | |
2398 | { | |
2399 | u32 hash; | |
2400 | u16 qoffset = 0; | |
2401 | u16 qcount = num_tx_queues; | |
2402 | ||
2403 | if (skb_rx_queue_recorded(skb)) { | |
2404 | hash = skb_get_rx_queue(skb); | |
2405 | while (unlikely(hash >= num_tx_queues)) | |
2406 | hash -= num_tx_queues; | |
2407 | return hash; | |
2408 | } | |
2409 | ||
2410 | if (dev->num_tc) { | |
2411 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2412 | qoffset = dev->tc_to_txq[tc].offset; | |
2413 | qcount = dev->tc_to_txq[tc].count; | |
2414 | } | |
2415 | ||
2416 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2417 | } | |
2418 | EXPORT_SYMBOL(__skb_tx_hash); | |
2419 | ||
36c92474 BH |
2420 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2421 | { | |
65e9d2fa | 2422 | static const netdev_features_t null_features = 0; |
36c92474 | 2423 | struct net_device *dev = skb->dev; |
88ad4175 | 2424 | const char *name = ""; |
36c92474 | 2425 | |
c846ad9b BG |
2426 | if (!net_ratelimit()) |
2427 | return; | |
2428 | ||
88ad4175 BM |
2429 | if (dev) { |
2430 | if (dev->dev.parent) | |
2431 | name = dev_driver_string(dev->dev.parent); | |
2432 | else | |
2433 | name = netdev_name(dev); | |
2434 | } | |
36c92474 BH |
2435 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2436 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2437 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2438 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2439 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2440 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2441 | } | |
2442 | ||
1da177e4 LT |
2443 | /* |
2444 | * Invalidate hardware checksum when packet is to be mangled, and | |
2445 | * complete checksum manually on outgoing path. | |
2446 | */ | |
84fa7933 | 2447 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2448 | { |
d3bc23e7 | 2449 | __wsum csum; |
663ead3b | 2450 | int ret = 0, offset; |
1da177e4 | 2451 | |
84fa7933 | 2452 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2453 | goto out_set_summed; |
2454 | ||
2455 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2456 | skb_warn_bad_offload(skb); |
2457 | return -EINVAL; | |
1da177e4 LT |
2458 | } |
2459 | ||
cef401de ED |
2460 | /* Before computing a checksum, we should make sure no frag could |
2461 | * be modified by an external entity : checksum could be wrong. | |
2462 | */ | |
2463 | if (skb_has_shared_frag(skb)) { | |
2464 | ret = __skb_linearize(skb); | |
2465 | if (ret) | |
2466 | goto out; | |
2467 | } | |
2468 | ||
55508d60 | 2469 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2470 | BUG_ON(offset >= skb_headlen(skb)); |
2471 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2472 | ||
2473 | offset += skb->csum_offset; | |
2474 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2475 | ||
2476 | if (skb_cloned(skb) && | |
2477 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2478 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2479 | if (ret) | |
2480 | goto out; | |
2481 | } | |
2482 | ||
a030847e | 2483 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2484 | out_set_summed: |
1da177e4 | 2485 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2486 | out: |
1da177e4 LT |
2487 | return ret; |
2488 | } | |
d1b19dff | 2489 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2490 | |
6ae23ad3 TH |
2491 | /* skb_csum_offload_check - Driver helper function to determine if a device |
2492 | * with limited checksum offload capabilities is able to offload the checksum | |
2493 | * for a given packet. | |
2494 | * | |
2495 | * Arguments: | |
2496 | * skb - sk_buff for the packet in question | |
2497 | * spec - contains the description of what device can offload | |
2498 | * csum_encapped - returns true if the checksum being offloaded is | |
2499 | * encpasulated. That is it is checksum for the transport header | |
2500 | * in the inner headers. | |
2501 | * checksum_help - when set indicates that helper function should | |
2502 | * call skb_checksum_help if offload checks fail | |
2503 | * | |
2504 | * Returns: | |
2505 | * true: Packet has passed the checksum checks and should be offloadable to | |
2506 | * the device (a driver may still need to check for additional | |
2507 | * restrictions of its device) | |
2508 | * false: Checksum is not offloadable. If checksum_help was set then | |
2509 | * skb_checksum_help was called to resolve checksum for non-GSO | |
2510 | * packets and when IP protocol is not SCTP | |
2511 | */ | |
2512 | bool __skb_csum_offload_chk(struct sk_buff *skb, | |
2513 | const struct skb_csum_offl_spec *spec, | |
2514 | bool *csum_encapped, | |
2515 | bool csum_help) | |
2516 | { | |
2517 | struct iphdr *iph; | |
2518 | struct ipv6hdr *ipv6; | |
2519 | void *nhdr; | |
2520 | int protocol; | |
2521 | u8 ip_proto; | |
2522 | ||
2523 | if (skb->protocol == htons(ETH_P_8021Q) || | |
2524 | skb->protocol == htons(ETH_P_8021AD)) { | |
2525 | if (!spec->vlan_okay) | |
2526 | goto need_help; | |
2527 | } | |
2528 | ||
2529 | /* We check whether the checksum refers to a transport layer checksum in | |
2530 | * the outermost header or an encapsulated transport layer checksum that | |
2531 | * corresponds to the inner headers of the skb. If the checksum is for | |
2532 | * something else in the packet we need help. | |
2533 | */ | |
2534 | if (skb_checksum_start_offset(skb) == skb_transport_offset(skb)) { | |
2535 | /* Non-encapsulated checksum */ | |
2536 | protocol = eproto_to_ipproto(vlan_get_protocol(skb)); | |
2537 | nhdr = skb_network_header(skb); | |
2538 | *csum_encapped = false; | |
2539 | if (spec->no_not_encapped) | |
2540 | goto need_help; | |
2541 | } else if (skb->encapsulation && spec->encap_okay && | |
2542 | skb_checksum_start_offset(skb) == | |
2543 | skb_inner_transport_offset(skb)) { | |
2544 | /* Encapsulated checksum */ | |
2545 | *csum_encapped = true; | |
2546 | switch (skb->inner_protocol_type) { | |
2547 | case ENCAP_TYPE_ETHER: | |
2548 | protocol = eproto_to_ipproto(skb->inner_protocol); | |
2549 | break; | |
2550 | case ENCAP_TYPE_IPPROTO: | |
2551 | protocol = skb->inner_protocol; | |
2552 | break; | |
2553 | } | |
2554 | nhdr = skb_inner_network_header(skb); | |
2555 | } else { | |
2556 | goto need_help; | |
2557 | } | |
2558 | ||
2559 | switch (protocol) { | |
2560 | case IPPROTO_IP: | |
2561 | if (!spec->ipv4_okay) | |
2562 | goto need_help; | |
2563 | iph = nhdr; | |
2564 | ip_proto = iph->protocol; | |
2565 | if (iph->ihl != 5 && !spec->ip_options_okay) | |
2566 | goto need_help; | |
2567 | break; | |
2568 | case IPPROTO_IPV6: | |
2569 | if (!spec->ipv6_okay) | |
2570 | goto need_help; | |
2571 | if (spec->no_encapped_ipv6 && *csum_encapped) | |
2572 | goto need_help; | |
2573 | ipv6 = nhdr; | |
2574 | nhdr += sizeof(*ipv6); | |
2575 | ip_proto = ipv6->nexthdr; | |
2576 | break; | |
2577 | default: | |
2578 | goto need_help; | |
2579 | } | |
2580 | ||
2581 | ip_proto_again: | |
2582 | switch (ip_proto) { | |
2583 | case IPPROTO_TCP: | |
2584 | if (!spec->tcp_okay || | |
2585 | skb->csum_offset != offsetof(struct tcphdr, check)) | |
2586 | goto need_help; | |
2587 | break; | |
2588 | case IPPROTO_UDP: | |
2589 | if (!spec->udp_okay || | |
2590 | skb->csum_offset != offsetof(struct udphdr, check)) | |
2591 | goto need_help; | |
2592 | break; | |
2593 | case IPPROTO_SCTP: | |
2594 | if (!spec->sctp_okay || | |
2595 | skb->csum_offset != offsetof(struct sctphdr, checksum)) | |
2596 | goto cant_help; | |
2597 | break; | |
2598 | case NEXTHDR_HOP: | |
2599 | case NEXTHDR_ROUTING: | |
2600 | case NEXTHDR_DEST: { | |
2601 | u8 *opthdr = nhdr; | |
2602 | ||
2603 | if (protocol != IPPROTO_IPV6 || !spec->ext_hdrs_okay) | |
2604 | goto need_help; | |
2605 | ||
2606 | ip_proto = opthdr[0]; | |
2607 | nhdr += (opthdr[1] + 1) << 3; | |
2608 | ||
2609 | goto ip_proto_again; | |
2610 | } | |
2611 | default: | |
2612 | goto need_help; | |
2613 | } | |
2614 | ||
2615 | /* Passed the tests for offloading checksum */ | |
2616 | return true; | |
2617 | ||
2618 | need_help: | |
2619 | if (csum_help && !skb_shinfo(skb)->gso_size) | |
2620 | skb_checksum_help(skb); | |
2621 | cant_help: | |
2622 | return false; | |
2623 | } | |
2624 | EXPORT_SYMBOL(__skb_csum_offload_chk); | |
2625 | ||
53d6471c | 2626 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2627 | { |
252e3346 | 2628 | __be16 type = skb->protocol; |
f6a78bfc | 2629 | |
19acc327 PS |
2630 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2631 | if (type == htons(ETH_P_TEB)) { | |
2632 | struct ethhdr *eth; | |
2633 | ||
2634 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2635 | return 0; | |
2636 | ||
2637 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2638 | type = eth->h_proto; | |
2639 | } | |
2640 | ||
d4bcef3f | 2641 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2642 | } |
2643 | ||
2644 | /** | |
2645 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2646 | * @skb: buffer to segment | |
2647 | * @features: features for the output path (see dev->features) | |
2648 | */ | |
2649 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2650 | netdev_features_t features) | |
2651 | { | |
2652 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2653 | struct packet_offload *ptype; | |
53d6471c VY |
2654 | int vlan_depth = skb->mac_len; |
2655 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2656 | |
2657 | if (unlikely(!type)) | |
2658 | return ERR_PTR(-EINVAL); | |
2659 | ||
53d6471c | 2660 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2661 | |
2662 | rcu_read_lock(); | |
22061d80 | 2663 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2664 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2665 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2666 | break; |
2667 | } | |
2668 | } | |
2669 | rcu_read_unlock(); | |
2670 | ||
98e399f8 | 2671 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2672 | |
f6a78bfc HX |
2673 | return segs; |
2674 | } | |
05e8ef4a PS |
2675 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2676 | ||
2677 | ||
2678 | /* openvswitch calls this on rx path, so we need a different check. | |
2679 | */ | |
2680 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2681 | { | |
2682 | if (tx_path) | |
2683 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2684 | else | |
2685 | return skb->ip_summed == CHECKSUM_NONE; | |
2686 | } | |
2687 | ||
2688 | /** | |
2689 | * __skb_gso_segment - Perform segmentation on skb. | |
2690 | * @skb: buffer to segment | |
2691 | * @features: features for the output path (see dev->features) | |
2692 | * @tx_path: whether it is called in TX path | |
2693 | * | |
2694 | * This function segments the given skb and returns a list of segments. | |
2695 | * | |
2696 | * It may return NULL if the skb requires no segmentation. This is | |
2697 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 KK |
2698 | * |
2699 | * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. | |
05e8ef4a PS |
2700 | */ |
2701 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2702 | netdev_features_t features, bool tx_path) | |
2703 | { | |
2704 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2705 | int err; | |
2706 | ||
2707 | skb_warn_bad_offload(skb); | |
2708 | ||
a40e0a66 | 2709 | err = skb_cow_head(skb, 0); |
2710 | if (err < 0) | |
05e8ef4a PS |
2711 | return ERR_PTR(err); |
2712 | } | |
2713 | ||
9207f9d4 KK |
2714 | BUILD_BUG_ON(SKB_SGO_CB_OFFSET + |
2715 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); | |
2716 | ||
68c33163 | 2717 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2718 | SKB_GSO_CB(skb)->encap_level = 0; |
2719 | ||
05e8ef4a PS |
2720 | skb_reset_mac_header(skb); |
2721 | skb_reset_mac_len(skb); | |
2722 | ||
2723 | return skb_mac_gso_segment(skb, features); | |
2724 | } | |
12b0004d | 2725 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2726 | |
fb286bb2 HX |
2727 | /* Take action when hardware reception checksum errors are detected. */ |
2728 | #ifdef CONFIG_BUG | |
2729 | void netdev_rx_csum_fault(struct net_device *dev) | |
2730 | { | |
2731 | if (net_ratelimit()) { | |
7b6cd1ce | 2732 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2733 | dump_stack(); |
2734 | } | |
2735 | } | |
2736 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2737 | #endif | |
2738 | ||
1da177e4 LT |
2739 | /* Actually, we should eliminate this check as soon as we know, that: |
2740 | * 1. IOMMU is present and allows to map all the memory. | |
2741 | * 2. No high memory really exists on this machine. | |
2742 | */ | |
2743 | ||
c1e756bf | 2744 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2745 | { |
3d3a8533 | 2746 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2747 | int i; |
5acbbd42 | 2748 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2749 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2750 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2751 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2752 | return 1; |
ea2ab693 | 2753 | } |
5acbbd42 | 2754 | } |
1da177e4 | 2755 | |
5acbbd42 FT |
2756 | if (PCI_DMA_BUS_IS_PHYS) { |
2757 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2758 | |
9092c658 ED |
2759 | if (!pdev) |
2760 | return 0; | |
5acbbd42 | 2761 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2762 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2763 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2764 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2765 | return 1; | |
2766 | } | |
2767 | } | |
3d3a8533 | 2768 | #endif |
1da177e4 LT |
2769 | return 0; |
2770 | } | |
1da177e4 | 2771 | |
3b392ddb SH |
2772 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2773 | * instead of standard features for the netdev. | |
2774 | */ | |
d0edc7bf | 2775 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2776 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2777 | netdev_features_t features, | |
2778 | __be16 type) | |
2779 | { | |
25cd9ba0 | 2780 | if (eth_p_mpls(type)) |
3b392ddb SH |
2781 | features &= skb->dev->mpls_features; |
2782 | ||
2783 | return features; | |
2784 | } | |
2785 | #else | |
2786 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2787 | netdev_features_t features, | |
2788 | __be16 type) | |
2789 | { | |
2790 | return features; | |
2791 | } | |
2792 | #endif | |
2793 | ||
c8f44aff | 2794 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2795 | netdev_features_t features) |
f01a5236 | 2796 | { |
53d6471c | 2797 | int tmp; |
3b392ddb SH |
2798 | __be16 type; |
2799 | ||
2800 | type = skb_network_protocol(skb, &tmp); | |
2801 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2802 | |
c0d680e5 | 2803 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2804 | !can_checksum_protocol(features, type)) { |
a188222b | 2805 | features &= ~NETIF_F_CSUM_MASK; |
c1e756bf | 2806 | } else if (illegal_highdma(skb->dev, skb)) { |
f01a5236 JG |
2807 | features &= ~NETIF_F_SG; |
2808 | } | |
2809 | ||
2810 | return features; | |
2811 | } | |
2812 | ||
e38f3025 TM |
2813 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2814 | struct net_device *dev, | |
2815 | netdev_features_t features) | |
2816 | { | |
2817 | return features; | |
2818 | } | |
2819 | EXPORT_SYMBOL(passthru_features_check); | |
2820 | ||
8cb65d00 TM |
2821 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2822 | struct net_device *dev, | |
2823 | netdev_features_t features) | |
2824 | { | |
2825 | return vlan_features_check(skb, features); | |
2826 | } | |
2827 | ||
c1e756bf | 2828 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2829 | { |
5f35227e | 2830 | struct net_device *dev = skb->dev; |
fcbeb976 ED |
2831 | netdev_features_t features = dev->features; |
2832 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
58e998c6 | 2833 | |
743b03a8 | 2834 | if (gso_segs > dev->gso_max_segs) |
30b678d8 BH |
2835 | features &= ~NETIF_F_GSO_MASK; |
2836 | ||
5f35227e JG |
2837 | /* If encapsulation offload request, verify we are testing |
2838 | * hardware encapsulation features instead of standard | |
2839 | * features for the netdev | |
2840 | */ | |
2841 | if (skb->encapsulation) | |
2842 | features &= dev->hw_enc_features; | |
2843 | ||
f5a7fb88 TM |
2844 | if (skb_vlan_tagged(skb)) |
2845 | features = netdev_intersect_features(features, | |
2846 | dev->vlan_features | | |
2847 | NETIF_F_HW_VLAN_CTAG_TX | | |
2848 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2849 | |
5f35227e JG |
2850 | if (dev->netdev_ops->ndo_features_check) |
2851 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2852 | features); | |
8cb65d00 TM |
2853 | else |
2854 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2855 | |
c1e756bf | 2856 | return harmonize_features(skb, features); |
58e998c6 | 2857 | } |
c1e756bf | 2858 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2859 | |
2ea25513 | 2860 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2861 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2862 | { |
2ea25513 DM |
2863 | unsigned int len; |
2864 | int rc; | |
00829823 | 2865 | |
7866a621 | 2866 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 2867 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 2868 | |
2ea25513 DM |
2869 | len = skb->len; |
2870 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2871 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 2872 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 2873 | |
2ea25513 DM |
2874 | return rc; |
2875 | } | |
7b9c6090 | 2876 | |
8dcda22a DM |
2877 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2878 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2879 | { |
2880 | struct sk_buff *skb = first; | |
2881 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 2882 | |
7f2e870f DM |
2883 | while (skb) { |
2884 | struct sk_buff *next = skb->next; | |
fc70fb64 | 2885 | |
7f2e870f | 2886 | skb->next = NULL; |
95f6b3dd | 2887 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
2888 | if (unlikely(!dev_xmit_complete(rc))) { |
2889 | skb->next = next; | |
2890 | goto out; | |
2891 | } | |
6afff0ca | 2892 | |
7f2e870f DM |
2893 | skb = next; |
2894 | if (netif_xmit_stopped(txq) && skb) { | |
2895 | rc = NETDEV_TX_BUSY; | |
2896 | break; | |
9ccb8975 | 2897 | } |
7f2e870f | 2898 | } |
9ccb8975 | 2899 | |
7f2e870f DM |
2900 | out: |
2901 | *ret = rc; | |
2902 | return skb; | |
2903 | } | |
b40863c6 | 2904 | |
1ff0dc94 ED |
2905 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
2906 | netdev_features_t features) | |
f6a78bfc | 2907 | { |
df8a39de | 2908 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
2909 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
2910 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
2911 | return skb; |
2912 | } | |
f6a78bfc | 2913 | |
55a93b3e | 2914 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
2915 | { |
2916 | netdev_features_t features; | |
f6a78bfc | 2917 | |
eae3f88e DM |
2918 | if (skb->next) |
2919 | return skb; | |
068a2de5 | 2920 | |
eae3f88e DM |
2921 | features = netif_skb_features(skb); |
2922 | skb = validate_xmit_vlan(skb, features); | |
2923 | if (unlikely(!skb)) | |
2924 | goto out_null; | |
7b9c6090 | 2925 | |
8b86a61d | 2926 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
2927 | struct sk_buff *segs; |
2928 | ||
2929 | segs = skb_gso_segment(skb, features); | |
cecda693 | 2930 | if (IS_ERR(segs)) { |
af6dabc9 | 2931 | goto out_kfree_skb; |
cecda693 JW |
2932 | } else if (segs) { |
2933 | consume_skb(skb); | |
2934 | skb = segs; | |
f6a78bfc | 2935 | } |
eae3f88e DM |
2936 | } else { |
2937 | if (skb_needs_linearize(skb, features) && | |
2938 | __skb_linearize(skb)) | |
2939 | goto out_kfree_skb; | |
4ec93edb | 2940 | |
eae3f88e DM |
2941 | /* If packet is not checksummed and device does not |
2942 | * support checksumming for this protocol, complete | |
2943 | * checksumming here. | |
2944 | */ | |
2945 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2946 | if (skb->encapsulation) | |
2947 | skb_set_inner_transport_header(skb, | |
2948 | skb_checksum_start_offset(skb)); | |
2949 | else | |
2950 | skb_set_transport_header(skb, | |
2951 | skb_checksum_start_offset(skb)); | |
a188222b | 2952 | if (!(features & NETIF_F_CSUM_MASK) && |
eae3f88e DM |
2953 | skb_checksum_help(skb)) |
2954 | goto out_kfree_skb; | |
7b9c6090 | 2955 | } |
0c772159 | 2956 | } |
7b9c6090 | 2957 | |
eae3f88e | 2958 | return skb; |
fc70fb64 | 2959 | |
f6a78bfc HX |
2960 | out_kfree_skb: |
2961 | kfree_skb(skb); | |
eae3f88e DM |
2962 | out_null: |
2963 | return NULL; | |
2964 | } | |
6afff0ca | 2965 | |
55a93b3e ED |
2966 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
2967 | { | |
2968 | struct sk_buff *next, *head = NULL, *tail; | |
2969 | ||
bec3cfdc | 2970 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
2971 | next = skb->next; |
2972 | skb->next = NULL; | |
bec3cfdc ED |
2973 | |
2974 | /* in case skb wont be segmented, point to itself */ | |
2975 | skb->prev = skb; | |
2976 | ||
55a93b3e | 2977 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
2978 | if (!skb) |
2979 | continue; | |
55a93b3e | 2980 | |
bec3cfdc ED |
2981 | if (!head) |
2982 | head = skb; | |
2983 | else | |
2984 | tail->next = skb; | |
2985 | /* If skb was segmented, skb->prev points to | |
2986 | * the last segment. If not, it still contains skb. | |
2987 | */ | |
2988 | tail = skb->prev; | |
55a93b3e ED |
2989 | } |
2990 | return head; | |
f6a78bfc HX |
2991 | } |
2992 | ||
1def9238 ED |
2993 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
2994 | { | |
2995 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2996 | ||
2997 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
2998 | ||
2999 | /* To get more precise estimation of bytes sent on wire, | |
3000 | * we add to pkt_len the headers size of all segments | |
3001 | */ | |
3002 | if (shinfo->gso_size) { | |
757b8b1d | 3003 | unsigned int hdr_len; |
15e5a030 | 3004 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3005 | |
757b8b1d ED |
3006 | /* mac layer + network layer */ |
3007 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3008 | ||
3009 | /* + transport layer */ | |
1def9238 ED |
3010 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
3011 | hdr_len += tcp_hdrlen(skb); | |
3012 | else | |
3013 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
3014 | |
3015 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3016 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3017 | shinfo->gso_size); | |
3018 | ||
3019 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3020 | } |
3021 | } | |
3022 | ||
bbd8a0d3 KK |
3023 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3024 | struct net_device *dev, | |
3025 | struct netdev_queue *txq) | |
3026 | { | |
3027 | spinlock_t *root_lock = qdisc_lock(q); | |
a2da570d | 3028 | bool contended; |
bbd8a0d3 KK |
3029 | int rc; |
3030 | ||
a2da570d | 3031 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
3032 | /* |
3033 | * Heuristic to force contended enqueues to serialize on a | |
3034 | * separate lock before trying to get qdisc main lock. | |
9bf2b8c2 YX |
3035 | * This permits __QDISC___STATE_RUNNING owner to get the lock more |
3036 | * often and dequeue packets faster. | |
79640a4c | 3037 | */ |
a2da570d | 3038 | contended = qdisc_is_running(q); |
79640a4c ED |
3039 | if (unlikely(contended)) |
3040 | spin_lock(&q->busylock); | |
3041 | ||
bbd8a0d3 KK |
3042 | spin_lock(root_lock); |
3043 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
3044 | kfree_skb(skb); | |
3045 | rc = NET_XMIT_DROP; | |
3046 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3047 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3048 | /* |
3049 | * This is a work-conserving queue; there are no old skbs | |
3050 | * waiting to be sent out; and the qdisc is not running - | |
3051 | * xmit the skb directly. | |
3052 | */ | |
bfe0d029 | 3053 | |
bfe0d029 ED |
3054 | qdisc_bstats_update(q, skb); |
3055 | ||
55a93b3e | 3056 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3057 | if (unlikely(contended)) { |
3058 | spin_unlock(&q->busylock); | |
3059 | contended = false; | |
3060 | } | |
bbd8a0d3 | 3061 | __qdisc_run(q); |
79640a4c | 3062 | } else |
bc135b23 | 3063 | qdisc_run_end(q); |
bbd8a0d3 KK |
3064 | |
3065 | rc = NET_XMIT_SUCCESS; | |
3066 | } else { | |
a2da570d | 3067 | rc = q->enqueue(skb, q) & NET_XMIT_MASK; |
79640a4c ED |
3068 | if (qdisc_run_begin(q)) { |
3069 | if (unlikely(contended)) { | |
3070 | spin_unlock(&q->busylock); | |
3071 | contended = false; | |
3072 | } | |
3073 | __qdisc_run(q); | |
3074 | } | |
bbd8a0d3 KK |
3075 | } |
3076 | spin_unlock(root_lock); | |
79640a4c ED |
3077 | if (unlikely(contended)) |
3078 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3079 | return rc; |
3080 | } | |
3081 | ||
86f8515f | 3082 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3083 | static void skb_update_prio(struct sk_buff *skb) |
3084 | { | |
6977a79d | 3085 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 3086 | |
91c68ce2 | 3087 | if (!skb->priority && skb->sk && map) { |
2a56a1fe TH |
3088 | unsigned int prioidx = |
3089 | sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); | |
91c68ce2 ED |
3090 | |
3091 | if (prioidx < map->priomap_len) | |
3092 | skb->priority = map->priomap[prioidx]; | |
3093 | } | |
5bc1421e NH |
3094 | } |
3095 | #else | |
3096 | #define skb_update_prio(skb) | |
3097 | #endif | |
3098 | ||
f60e5990 | 3099 | DEFINE_PER_CPU(int, xmit_recursion); |
3100 | EXPORT_SYMBOL(xmit_recursion); | |
3101 | ||
11a766ce | 3102 | #define RECURSION_LIMIT 10 |
745e20f1 | 3103 | |
95603e22 MM |
3104 | /** |
3105 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3106 | * @net: network namespace this loopback is happening in |
3107 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3108 | * @skb: buffer to transmit |
3109 | */ | |
0c4b51f0 | 3110 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3111 | { |
3112 | skb_reset_mac_header(skb); | |
3113 | __skb_pull(skb, skb_network_offset(skb)); | |
3114 | skb->pkt_type = PACKET_LOOPBACK; | |
3115 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3116 | WARN_ON(!skb_dst(skb)); | |
3117 | skb_dst_force(skb); | |
3118 | netif_rx_ni(skb); | |
3119 | return 0; | |
3120 | } | |
3121 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3122 | ||
1f211a1b DB |
3123 | #ifdef CONFIG_NET_EGRESS |
3124 | static struct sk_buff * | |
3125 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3126 | { | |
3127 | struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list); | |
3128 | struct tcf_result cl_res; | |
3129 | ||
3130 | if (!cl) | |
3131 | return skb; | |
3132 | ||
3133 | /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set | |
3134 | * earlier by the caller. | |
3135 | */ | |
3136 | qdisc_bstats_cpu_update(cl->q, skb); | |
3137 | ||
3138 | switch (tc_classify(skb, cl, &cl_res, false)) { | |
3139 | case TC_ACT_OK: | |
3140 | case TC_ACT_RECLASSIFY: | |
3141 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3142 | break; | |
3143 | case TC_ACT_SHOT: | |
3144 | qdisc_qstats_cpu_drop(cl->q); | |
3145 | *ret = NET_XMIT_DROP; | |
3146 | goto drop; | |
3147 | case TC_ACT_STOLEN: | |
3148 | case TC_ACT_QUEUED: | |
3149 | *ret = NET_XMIT_SUCCESS; | |
3150 | drop: | |
3151 | kfree_skb(skb); | |
3152 | return NULL; | |
3153 | case TC_ACT_REDIRECT: | |
3154 | /* No need to push/pop skb's mac_header here on egress! */ | |
3155 | skb_do_redirect(skb); | |
3156 | *ret = NET_XMIT_SUCCESS; | |
3157 | return NULL; | |
3158 | default: | |
3159 | break; | |
3160 | } | |
3161 | ||
3162 | return skb; | |
3163 | } | |
3164 | #endif /* CONFIG_NET_EGRESS */ | |
3165 | ||
638b2a69 JP |
3166 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
3167 | { | |
3168 | #ifdef CONFIG_XPS | |
3169 | struct xps_dev_maps *dev_maps; | |
3170 | struct xps_map *map; | |
3171 | int queue_index = -1; | |
3172 | ||
3173 | rcu_read_lock(); | |
3174 | dev_maps = rcu_dereference(dev->xps_maps); | |
3175 | if (dev_maps) { | |
3176 | map = rcu_dereference( | |
3177 | dev_maps->cpu_map[skb->sender_cpu - 1]); | |
3178 | if (map) { | |
3179 | if (map->len == 1) | |
3180 | queue_index = map->queues[0]; | |
3181 | else | |
3182 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), | |
3183 | map->len)]; | |
3184 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3185 | queue_index = -1; | |
3186 | } | |
3187 | } | |
3188 | rcu_read_unlock(); | |
3189 | ||
3190 | return queue_index; | |
3191 | #else | |
3192 | return -1; | |
3193 | #endif | |
3194 | } | |
3195 | ||
3196 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) | |
3197 | { | |
3198 | struct sock *sk = skb->sk; | |
3199 | int queue_index = sk_tx_queue_get(sk); | |
3200 | ||
3201 | if (queue_index < 0 || skb->ooo_okay || | |
3202 | queue_index >= dev->real_num_tx_queues) { | |
3203 | int new_index = get_xps_queue(dev, skb); | |
3204 | if (new_index < 0) | |
3205 | new_index = skb_tx_hash(dev, skb); | |
3206 | ||
3207 | if (queue_index != new_index && sk && | |
004a5d01 | 3208 | sk_fullsock(sk) && |
638b2a69 JP |
3209 | rcu_access_pointer(sk->sk_dst_cache)) |
3210 | sk_tx_queue_set(sk, new_index); | |
3211 | ||
3212 | queue_index = new_index; | |
3213 | } | |
3214 | ||
3215 | return queue_index; | |
3216 | } | |
3217 | ||
3218 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
3219 | struct sk_buff *skb, | |
3220 | void *accel_priv) | |
3221 | { | |
3222 | int queue_index = 0; | |
3223 | ||
3224 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3225 | u32 sender_cpu = skb->sender_cpu - 1; |
3226 | ||
3227 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3228 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3229 | #endif | |
3230 | ||
3231 | if (dev->real_num_tx_queues != 1) { | |
3232 | const struct net_device_ops *ops = dev->netdev_ops; | |
3233 | if (ops->ndo_select_queue) | |
3234 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, | |
3235 | __netdev_pick_tx); | |
3236 | else | |
3237 | queue_index = __netdev_pick_tx(dev, skb); | |
3238 | ||
3239 | if (!accel_priv) | |
3240 | queue_index = netdev_cap_txqueue(dev, queue_index); | |
3241 | } | |
3242 | ||
3243 | skb_set_queue_mapping(skb, queue_index); | |
3244 | return netdev_get_tx_queue(dev, queue_index); | |
3245 | } | |
3246 | ||
d29f749e | 3247 | /** |
9d08dd3d | 3248 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3249 | * @skb: buffer to transmit |
9d08dd3d | 3250 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
3251 | * |
3252 | * Queue a buffer for transmission to a network device. The caller must | |
3253 | * have set the device and priority and built the buffer before calling | |
3254 | * this function. The function can be called from an interrupt. | |
3255 | * | |
3256 | * A negative errno code is returned on a failure. A success does not | |
3257 | * guarantee the frame will be transmitted as it may be dropped due | |
3258 | * to congestion or traffic shaping. | |
3259 | * | |
3260 | * ----------------------------------------------------------------------------------- | |
3261 | * I notice this method can also return errors from the queue disciplines, | |
3262 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3263 | * be positive. | |
3264 | * | |
3265 | * Regardless of the return value, the skb is consumed, so it is currently | |
3266 | * difficult to retry a send to this method. (You can bump the ref count | |
3267 | * before sending to hold a reference for retry if you are careful.) | |
3268 | * | |
3269 | * When calling this method, interrupts MUST be enabled. This is because | |
3270 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3271 | * --BLG | |
3272 | */ | |
0a59f3a9 | 3273 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
3274 | { |
3275 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3276 | struct netdev_queue *txq; |
1da177e4 LT |
3277 | struct Qdisc *q; |
3278 | int rc = -ENOMEM; | |
3279 | ||
6d1ccff6 ED |
3280 | skb_reset_mac_header(skb); |
3281 | ||
e7fd2885 WB |
3282 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3283 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3284 | ||
4ec93edb YH |
3285 | /* Disable soft irqs for various locks below. Also |
3286 | * stops preemption for RCU. | |
1da177e4 | 3287 | */ |
4ec93edb | 3288 | rcu_read_lock_bh(); |
1da177e4 | 3289 | |
5bc1421e NH |
3290 | skb_update_prio(skb); |
3291 | ||
1f211a1b DB |
3292 | qdisc_pkt_len_init(skb); |
3293 | #ifdef CONFIG_NET_CLS_ACT | |
3294 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); | |
3295 | # ifdef CONFIG_NET_EGRESS | |
3296 | if (static_key_false(&egress_needed)) { | |
3297 | skb = sch_handle_egress(skb, &rc, dev); | |
3298 | if (!skb) | |
3299 | goto out; | |
3300 | } | |
3301 | # endif | |
3302 | #endif | |
02875878 ED |
3303 | /* If device/qdisc don't need skb->dst, release it right now while |
3304 | * its hot in this cpu cache. | |
3305 | */ | |
3306 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3307 | skb_dst_drop(skb); | |
3308 | else | |
3309 | skb_dst_force(skb); | |
3310 | ||
0c4f691f SF |
3311 | #ifdef CONFIG_NET_SWITCHDEV |
3312 | /* Don't forward if offload device already forwarded */ | |
3313 | if (skb->offload_fwd_mark && | |
3314 | skb->offload_fwd_mark == dev->offload_fwd_mark) { | |
3315 | consume_skb(skb); | |
3316 | rc = NET_XMIT_SUCCESS; | |
3317 | goto out; | |
3318 | } | |
3319 | #endif | |
3320 | ||
f663dd9a | 3321 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 3322 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3323 | |
cf66ba58 | 3324 | trace_net_dev_queue(skb); |
1da177e4 | 3325 | if (q->enqueue) { |
bbd8a0d3 | 3326 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3327 | goto out; |
1da177e4 LT |
3328 | } |
3329 | ||
3330 | /* The device has no queue. Common case for software devices: | |
3331 | loopback, all the sorts of tunnels... | |
3332 | ||
932ff279 HX |
3333 | Really, it is unlikely that netif_tx_lock protection is necessary |
3334 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
3335 | counters.) |
3336 | However, it is possible, that they rely on protection | |
3337 | made by us here. | |
3338 | ||
3339 | Check this and shot the lock. It is not prone from deadlocks. | |
3340 | Either shot noqueue qdisc, it is even simpler 8) | |
3341 | */ | |
3342 | if (dev->flags & IFF_UP) { | |
3343 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3344 | ||
c773e847 | 3345 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 3346 | |
745e20f1 ED |
3347 | if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT) |
3348 | goto recursion_alert; | |
3349 | ||
1f59533f JDB |
3350 | skb = validate_xmit_skb(skb, dev); |
3351 | if (!skb) | |
3352 | goto drop; | |
3353 | ||
c773e847 | 3354 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3355 | |
73466498 | 3356 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 3357 | __this_cpu_inc(xmit_recursion); |
ce93718f | 3358 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 3359 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 3360 | if (dev_xmit_complete(rc)) { |
c773e847 | 3361 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3362 | goto out; |
3363 | } | |
3364 | } | |
c773e847 | 3365 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3366 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3367 | dev->name); | |
1da177e4 LT |
3368 | } else { |
3369 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3370 | * unfortunately |
3371 | */ | |
3372 | recursion_alert: | |
e87cc472 JP |
3373 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3374 | dev->name); | |
1da177e4 LT |
3375 | } |
3376 | } | |
3377 | ||
3378 | rc = -ENETDOWN; | |
1f59533f | 3379 | drop: |
d4828d85 | 3380 | rcu_read_unlock_bh(); |
1da177e4 | 3381 | |
015f0688 | 3382 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3383 | kfree_skb_list(skb); |
1da177e4 LT |
3384 | return rc; |
3385 | out: | |
d4828d85 | 3386 | rcu_read_unlock_bh(); |
1da177e4 LT |
3387 | return rc; |
3388 | } | |
f663dd9a | 3389 | |
2b4aa3ce | 3390 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3391 | { |
3392 | return __dev_queue_xmit(skb, NULL); | |
3393 | } | |
2b4aa3ce | 3394 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3395 | |
f663dd9a JW |
3396 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3397 | { | |
3398 | return __dev_queue_xmit(skb, accel_priv); | |
3399 | } | |
3400 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3401 | ||
1da177e4 LT |
3402 | |
3403 | /*======================================================================= | |
3404 | Receiver routines | |
3405 | =======================================================================*/ | |
3406 | ||
6b2bedc3 | 3407 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3408 | EXPORT_SYMBOL(netdev_max_backlog); |
3409 | ||
3b098e2d | 3410 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
3411 | int netdev_budget __read_mostly = 300; |
3412 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 3413 | |
eecfd7c4 ED |
3414 | /* Called with irq disabled */ |
3415 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3416 | struct napi_struct *napi) | |
3417 | { | |
3418 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3419 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3420 | } | |
3421 | ||
bfb564e7 KK |
3422 | #ifdef CONFIG_RPS |
3423 | ||
3424 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3425 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3426 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3427 | u32 rps_cpu_mask __read_mostly; |
3428 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3429 | |
c5905afb | 3430 | struct static_key rps_needed __read_mostly; |
adc9300e | 3431 | |
c445477d BH |
3432 | static struct rps_dev_flow * |
3433 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3434 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3435 | { | |
a31196b0 | 3436 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3437 | #ifdef CONFIG_RFS_ACCEL |
3438 | struct netdev_rx_queue *rxqueue; | |
3439 | struct rps_dev_flow_table *flow_table; | |
3440 | struct rps_dev_flow *old_rflow; | |
3441 | u32 flow_id; | |
3442 | u16 rxq_index; | |
3443 | int rc; | |
3444 | ||
3445 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3446 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3447 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3448 | goto out; |
3449 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3450 | if (rxq_index == skb_get_rx_queue(skb)) | |
3451 | goto out; | |
3452 | ||
3453 | rxqueue = dev->_rx + rxq_index; | |
3454 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3455 | if (!flow_table) | |
3456 | goto out; | |
61b905da | 3457 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3458 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3459 | rxq_index, flow_id); | |
3460 | if (rc < 0) | |
3461 | goto out; | |
3462 | old_rflow = rflow; | |
3463 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3464 | rflow->filter = rc; |
3465 | if (old_rflow->filter == rflow->filter) | |
3466 | old_rflow->filter = RPS_NO_FILTER; | |
3467 | out: | |
3468 | #endif | |
3469 | rflow->last_qtail = | |
09994d1b | 3470 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3471 | } |
3472 | ||
09994d1b | 3473 | rflow->cpu = next_cpu; |
c445477d BH |
3474 | return rflow; |
3475 | } | |
3476 | ||
bfb564e7 KK |
3477 | /* |
3478 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3479 | * CPU from the RPS map of the receiving queue for a given skb. | |
3480 | * rcu_read_lock must be held on entry. | |
3481 | */ | |
3482 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3483 | struct rps_dev_flow **rflowp) | |
3484 | { | |
567e4b79 ED |
3485 | const struct rps_sock_flow_table *sock_flow_table; |
3486 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3487 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3488 | struct rps_map *map; |
bfb564e7 | 3489 | int cpu = -1; |
567e4b79 | 3490 | u32 tcpu; |
61b905da | 3491 | u32 hash; |
bfb564e7 KK |
3492 | |
3493 | if (skb_rx_queue_recorded(skb)) { | |
3494 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3495 | |
62fe0b40 BH |
3496 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3497 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3498 | "%s received packet on queue %u, but number " | |
3499 | "of RX queues is %u\n", | |
3500 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3501 | goto done; |
3502 | } | |
567e4b79 ED |
3503 | rxqueue += index; |
3504 | } | |
bfb564e7 | 3505 | |
567e4b79 ED |
3506 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3507 | ||
3508 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3509 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3510 | if (!flow_table && !map) |
bfb564e7 KK |
3511 | goto done; |
3512 | ||
2d47b459 | 3513 | skb_reset_network_header(skb); |
61b905da TH |
3514 | hash = skb_get_hash(skb); |
3515 | if (!hash) | |
bfb564e7 KK |
3516 | goto done; |
3517 | ||
fec5e652 TH |
3518 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3519 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3520 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3521 | u32 next_cpu; |
3522 | u32 ident; | |
3523 | ||
3524 | /* First check into global flow table if there is a match */ | |
3525 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3526 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3527 | goto try_rps; | |
fec5e652 | 3528 | |
567e4b79 ED |
3529 | next_cpu = ident & rps_cpu_mask; |
3530 | ||
3531 | /* OK, now we know there is a match, | |
3532 | * we can look at the local (per receive queue) flow table | |
3533 | */ | |
61b905da | 3534 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3535 | tcpu = rflow->cpu; |
3536 | ||
fec5e652 TH |
3537 | /* |
3538 | * If the desired CPU (where last recvmsg was done) is | |
3539 | * different from current CPU (one in the rx-queue flow | |
3540 | * table entry), switch if one of the following holds: | |
a31196b0 | 3541 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
3542 | * - Current CPU is offline. |
3543 | * - The current CPU's queue tail has advanced beyond the | |
3544 | * last packet that was enqueued using this table entry. | |
3545 | * This guarantees that all previous packets for the flow | |
3546 | * have been dequeued, thus preserving in order delivery. | |
3547 | */ | |
3548 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 3549 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 3550 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
3551 | rflow->last_qtail)) >= 0)) { |
3552 | tcpu = next_cpu; | |
c445477d | 3553 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3554 | } |
c445477d | 3555 | |
a31196b0 | 3556 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
3557 | *rflowp = rflow; |
3558 | cpu = tcpu; | |
3559 | goto done; | |
3560 | } | |
3561 | } | |
3562 | ||
567e4b79 ED |
3563 | try_rps: |
3564 | ||
0a9627f2 | 3565 | if (map) { |
8fc54f68 | 3566 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3567 | if (cpu_online(tcpu)) { |
3568 | cpu = tcpu; | |
3569 | goto done; | |
3570 | } | |
3571 | } | |
3572 | ||
3573 | done: | |
0a9627f2 TH |
3574 | return cpu; |
3575 | } | |
3576 | ||
c445477d BH |
3577 | #ifdef CONFIG_RFS_ACCEL |
3578 | ||
3579 | /** | |
3580 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3581 | * @dev: Device on which the filter was set | |
3582 | * @rxq_index: RX queue index | |
3583 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3584 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3585 | * | |
3586 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3587 | * this function for each installed filter and remove the filters for | |
3588 | * which it returns %true. | |
3589 | */ | |
3590 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3591 | u32 flow_id, u16 filter_id) | |
3592 | { | |
3593 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3594 | struct rps_dev_flow_table *flow_table; | |
3595 | struct rps_dev_flow *rflow; | |
3596 | bool expire = true; | |
a31196b0 | 3597 | unsigned int cpu; |
c445477d BH |
3598 | |
3599 | rcu_read_lock(); | |
3600 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3601 | if (flow_table && flow_id <= flow_table->mask) { | |
3602 | rflow = &flow_table->flows[flow_id]; | |
3603 | cpu = ACCESS_ONCE(rflow->cpu); | |
a31196b0 | 3604 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
3605 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
3606 | rflow->last_qtail) < | |
3607 | (int)(10 * flow_table->mask))) | |
3608 | expire = false; | |
3609 | } | |
3610 | rcu_read_unlock(); | |
3611 | return expire; | |
3612 | } | |
3613 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3614 | ||
3615 | #endif /* CONFIG_RFS_ACCEL */ | |
3616 | ||
0a9627f2 | 3617 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3618 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3619 | { |
e36fa2f7 ED |
3620 | struct softnet_data *sd = data; |
3621 | ||
eecfd7c4 | 3622 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3623 | sd->received_rps++; |
0a9627f2 | 3624 | } |
e36fa2f7 | 3625 | |
fec5e652 | 3626 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3627 | |
e36fa2f7 ED |
3628 | /* |
3629 | * Check if this softnet_data structure is another cpu one | |
3630 | * If yes, queue it to our IPI list and return 1 | |
3631 | * If no, return 0 | |
3632 | */ | |
3633 | static int rps_ipi_queued(struct softnet_data *sd) | |
3634 | { | |
3635 | #ifdef CONFIG_RPS | |
903ceff7 | 3636 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3637 | |
3638 | if (sd != mysd) { | |
3639 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3640 | mysd->rps_ipi_list = sd; | |
3641 | ||
3642 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3643 | return 1; | |
3644 | } | |
3645 | #endif /* CONFIG_RPS */ | |
3646 | return 0; | |
3647 | } | |
3648 | ||
99bbc707 WB |
3649 | #ifdef CONFIG_NET_FLOW_LIMIT |
3650 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3651 | #endif | |
3652 | ||
3653 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3654 | { | |
3655 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3656 | struct sd_flow_limit *fl; | |
3657 | struct softnet_data *sd; | |
3658 | unsigned int old_flow, new_flow; | |
3659 | ||
3660 | if (qlen < (netdev_max_backlog >> 1)) | |
3661 | return false; | |
3662 | ||
903ceff7 | 3663 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3664 | |
3665 | rcu_read_lock(); | |
3666 | fl = rcu_dereference(sd->flow_limit); | |
3667 | if (fl) { | |
3958afa1 | 3668 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3669 | old_flow = fl->history[fl->history_head]; |
3670 | fl->history[fl->history_head] = new_flow; | |
3671 | ||
3672 | fl->history_head++; | |
3673 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3674 | ||
3675 | if (likely(fl->buckets[old_flow])) | |
3676 | fl->buckets[old_flow]--; | |
3677 | ||
3678 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3679 | fl->count++; | |
3680 | rcu_read_unlock(); | |
3681 | return true; | |
3682 | } | |
3683 | } | |
3684 | rcu_read_unlock(); | |
3685 | #endif | |
3686 | return false; | |
3687 | } | |
3688 | ||
0a9627f2 TH |
3689 | /* |
3690 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3691 | * queue (may be a remote CPU queue). | |
3692 | */ | |
fec5e652 TH |
3693 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3694 | unsigned int *qtail) | |
0a9627f2 | 3695 | { |
e36fa2f7 | 3696 | struct softnet_data *sd; |
0a9627f2 | 3697 | unsigned long flags; |
99bbc707 | 3698 | unsigned int qlen; |
0a9627f2 | 3699 | |
e36fa2f7 | 3700 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3701 | |
3702 | local_irq_save(flags); | |
0a9627f2 | 3703 | |
e36fa2f7 | 3704 | rps_lock(sd); |
e9e4dd32 JA |
3705 | if (!netif_running(skb->dev)) |
3706 | goto drop; | |
99bbc707 WB |
3707 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3708 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3709 | if (qlen) { |
0a9627f2 | 3710 | enqueue: |
e36fa2f7 | 3711 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3712 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3713 | rps_unlock(sd); |
152102c7 | 3714 | local_irq_restore(flags); |
0a9627f2 TH |
3715 | return NET_RX_SUCCESS; |
3716 | } | |
3717 | ||
ebda37c2 ED |
3718 | /* Schedule NAPI for backlog device |
3719 | * We can use non atomic operation since we own the queue lock | |
3720 | */ | |
3721 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3722 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3723 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3724 | } |
3725 | goto enqueue; | |
3726 | } | |
3727 | ||
e9e4dd32 | 3728 | drop: |
dee42870 | 3729 | sd->dropped++; |
e36fa2f7 | 3730 | rps_unlock(sd); |
0a9627f2 | 3731 | |
0a9627f2 TH |
3732 | local_irq_restore(flags); |
3733 | ||
caf586e5 | 3734 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3735 | kfree_skb(skb); |
3736 | return NET_RX_DROP; | |
3737 | } | |
1da177e4 | 3738 | |
ae78dbfa | 3739 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3740 | { |
b0e28f1e | 3741 | int ret; |
1da177e4 | 3742 | |
588f0330 | 3743 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3744 | |
cf66ba58 | 3745 | trace_netif_rx(skb); |
df334545 | 3746 | #ifdef CONFIG_RPS |
c5905afb | 3747 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3748 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3749 | int cpu; |
3750 | ||
cece1945 | 3751 | preempt_disable(); |
b0e28f1e | 3752 | rcu_read_lock(); |
fec5e652 TH |
3753 | |
3754 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3755 | if (cpu < 0) |
3756 | cpu = smp_processor_id(); | |
fec5e652 TH |
3757 | |
3758 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3759 | ||
b0e28f1e | 3760 | rcu_read_unlock(); |
cece1945 | 3761 | preempt_enable(); |
adc9300e ED |
3762 | } else |
3763 | #endif | |
fec5e652 TH |
3764 | { |
3765 | unsigned int qtail; | |
3766 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3767 | put_cpu(); | |
3768 | } | |
b0e28f1e | 3769 | return ret; |
1da177e4 | 3770 | } |
ae78dbfa BH |
3771 | |
3772 | /** | |
3773 | * netif_rx - post buffer to the network code | |
3774 | * @skb: buffer to post | |
3775 | * | |
3776 | * This function receives a packet from a device driver and queues it for | |
3777 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3778 | * may be dropped during processing for congestion control or by the | |
3779 | * protocol layers. | |
3780 | * | |
3781 | * return values: | |
3782 | * NET_RX_SUCCESS (no congestion) | |
3783 | * NET_RX_DROP (packet was dropped) | |
3784 | * | |
3785 | */ | |
3786 | ||
3787 | int netif_rx(struct sk_buff *skb) | |
3788 | { | |
3789 | trace_netif_rx_entry(skb); | |
3790 | ||
3791 | return netif_rx_internal(skb); | |
3792 | } | |
d1b19dff | 3793 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3794 | |
3795 | int netif_rx_ni(struct sk_buff *skb) | |
3796 | { | |
3797 | int err; | |
3798 | ||
ae78dbfa BH |
3799 | trace_netif_rx_ni_entry(skb); |
3800 | ||
1da177e4 | 3801 | preempt_disable(); |
ae78dbfa | 3802 | err = netif_rx_internal(skb); |
1da177e4 LT |
3803 | if (local_softirq_pending()) |
3804 | do_softirq(); | |
3805 | preempt_enable(); | |
3806 | ||
3807 | return err; | |
3808 | } | |
1da177e4 LT |
3809 | EXPORT_SYMBOL(netif_rx_ni); |
3810 | ||
1da177e4 LT |
3811 | static void net_tx_action(struct softirq_action *h) |
3812 | { | |
903ceff7 | 3813 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
3814 | |
3815 | if (sd->completion_queue) { | |
3816 | struct sk_buff *clist; | |
3817 | ||
3818 | local_irq_disable(); | |
3819 | clist = sd->completion_queue; | |
3820 | sd->completion_queue = NULL; | |
3821 | local_irq_enable(); | |
3822 | ||
3823 | while (clist) { | |
3824 | struct sk_buff *skb = clist; | |
3825 | clist = clist->next; | |
3826 | ||
547b792c | 3827 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3828 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3829 | trace_consume_skb(skb); | |
3830 | else | |
3831 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
3832 | |
3833 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
3834 | __kfree_skb(skb); | |
3835 | else | |
3836 | __kfree_skb_defer(skb); | |
1da177e4 | 3837 | } |
15fad714 JDB |
3838 | |
3839 | __kfree_skb_flush(); | |
1da177e4 LT |
3840 | } |
3841 | ||
3842 | if (sd->output_queue) { | |
37437bb2 | 3843 | struct Qdisc *head; |
1da177e4 LT |
3844 | |
3845 | local_irq_disable(); | |
3846 | head = sd->output_queue; | |
3847 | sd->output_queue = NULL; | |
a9cbd588 | 3848 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3849 | local_irq_enable(); |
3850 | ||
3851 | while (head) { | |
37437bb2 DM |
3852 | struct Qdisc *q = head; |
3853 | spinlock_t *root_lock; | |
3854 | ||
1da177e4 LT |
3855 | head = head->next_sched; |
3856 | ||
5fb66229 | 3857 | root_lock = qdisc_lock(q); |
37437bb2 | 3858 | if (spin_trylock(root_lock)) { |
4e857c58 | 3859 | smp_mb__before_atomic(); |
def82a1d JP |
3860 | clear_bit(__QDISC_STATE_SCHED, |
3861 | &q->state); | |
37437bb2 DM |
3862 | qdisc_run(q); |
3863 | spin_unlock(root_lock); | |
1da177e4 | 3864 | } else { |
195648bb | 3865 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 3866 | &q->state)) { |
195648bb | 3867 | __netif_reschedule(q); |
e8a83e10 | 3868 | } else { |
4e857c58 | 3869 | smp_mb__before_atomic(); |
e8a83e10 JP |
3870 | clear_bit(__QDISC_STATE_SCHED, |
3871 | &q->state); | |
3872 | } | |
1da177e4 LT |
3873 | } |
3874 | } | |
3875 | } | |
3876 | } | |
3877 | ||
ab95bfe0 JP |
3878 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3879 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3880 | /* This hook is defined here for ATM LANE */ |
3881 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3882 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3883 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3884 | #endif |
1da177e4 | 3885 | |
1f211a1b DB |
3886 | static inline struct sk_buff * |
3887 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
3888 | struct net_device *orig_dev) | |
f697c3e8 | 3889 | { |
e7582bab | 3890 | #ifdef CONFIG_NET_CLS_ACT |
d2788d34 DB |
3891 | struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); |
3892 | struct tcf_result cl_res; | |
24824a09 | 3893 | |
c9e99fd0 DB |
3894 | /* If there's at least one ingress present somewhere (so |
3895 | * we get here via enabled static key), remaining devices | |
3896 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 3897 | * out here. |
c9e99fd0 | 3898 | */ |
d2788d34 | 3899 | if (!cl) |
4577139b | 3900 | return skb; |
f697c3e8 HX |
3901 | if (*pt_prev) { |
3902 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3903 | *pt_prev = NULL; | |
1da177e4 LT |
3904 | } |
3905 | ||
3365495c | 3906 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
c9e99fd0 | 3907 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); |
24ea591d | 3908 | qdisc_bstats_cpu_update(cl->q, skb); |
c9e99fd0 | 3909 | |
3b3ae880 | 3910 | switch (tc_classify(skb, cl, &cl_res, false)) { |
d2788d34 DB |
3911 | case TC_ACT_OK: |
3912 | case TC_ACT_RECLASSIFY: | |
3913 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3914 | break; | |
3915 | case TC_ACT_SHOT: | |
24ea591d | 3916 | qdisc_qstats_cpu_drop(cl->q); |
d2788d34 DB |
3917 | case TC_ACT_STOLEN: |
3918 | case TC_ACT_QUEUED: | |
3919 | kfree_skb(skb); | |
3920 | return NULL; | |
27b29f63 AS |
3921 | case TC_ACT_REDIRECT: |
3922 | /* skb_mac_header check was done by cls/act_bpf, so | |
3923 | * we can safely push the L2 header back before | |
3924 | * redirecting to another netdev | |
3925 | */ | |
3926 | __skb_push(skb, skb->mac_len); | |
3927 | skb_do_redirect(skb); | |
3928 | return NULL; | |
d2788d34 DB |
3929 | default: |
3930 | break; | |
f697c3e8 | 3931 | } |
e7582bab | 3932 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
3933 | return skb; |
3934 | } | |
1da177e4 | 3935 | |
ab95bfe0 JP |
3936 | /** |
3937 | * netdev_rx_handler_register - register receive handler | |
3938 | * @dev: device to register a handler for | |
3939 | * @rx_handler: receive handler to register | |
93e2c32b | 3940 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 3941 | * |
e227867f | 3942 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
3943 | * called from __netif_receive_skb. A negative errno code is returned |
3944 | * on a failure. | |
3945 | * | |
3946 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3947 | * |
3948 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3949 | */ |
3950 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3951 | rx_handler_func_t *rx_handler, |
3952 | void *rx_handler_data) | |
ab95bfe0 JP |
3953 | { |
3954 | ASSERT_RTNL(); | |
3955 | ||
3956 | if (dev->rx_handler) | |
3957 | return -EBUSY; | |
3958 | ||
00cfec37 | 3959 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 3960 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
3961 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
3962 | ||
3963 | return 0; | |
3964 | } | |
3965 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
3966 | ||
3967 | /** | |
3968 | * netdev_rx_handler_unregister - unregister receive handler | |
3969 | * @dev: device to unregister a handler from | |
3970 | * | |
166ec369 | 3971 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
3972 | * |
3973 | * The caller must hold the rtnl_mutex. | |
3974 | */ | |
3975 | void netdev_rx_handler_unregister(struct net_device *dev) | |
3976 | { | |
3977 | ||
3978 | ASSERT_RTNL(); | |
a9b3cd7f | 3979 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
3980 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
3981 | * section has a guarantee to see a non NULL rx_handler_data | |
3982 | * as well. | |
3983 | */ | |
3984 | synchronize_net(); | |
a9b3cd7f | 3985 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
3986 | } |
3987 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
3988 | ||
b4b9e355 MG |
3989 | /* |
3990 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
3991 | * the special handling of PFMEMALLOC skbs. | |
3992 | */ | |
3993 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
3994 | { | |
3995 | switch (skb->protocol) { | |
2b8837ae JP |
3996 | case htons(ETH_P_ARP): |
3997 | case htons(ETH_P_IP): | |
3998 | case htons(ETH_P_IPV6): | |
3999 | case htons(ETH_P_8021Q): | |
4000 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4001 | return true; |
4002 | default: | |
4003 | return false; | |
4004 | } | |
4005 | } | |
4006 | ||
e687ad60 PN |
4007 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4008 | int *ret, struct net_device *orig_dev) | |
4009 | { | |
e7582bab | 4010 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 PN |
4011 | if (nf_hook_ingress_active(skb)) { |
4012 | if (*pt_prev) { | |
4013 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4014 | *pt_prev = NULL; | |
4015 | } | |
4016 | ||
4017 | return nf_hook_ingress(skb); | |
4018 | } | |
e7582bab | 4019 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
4020 | return 0; |
4021 | } | |
e687ad60 | 4022 | |
9754e293 | 4023 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
4024 | { |
4025 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4026 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4027 | struct net_device *orig_dev; |
8a4eb573 | 4028 | bool deliver_exact = false; |
1da177e4 | 4029 | int ret = NET_RX_DROP; |
252e3346 | 4030 | __be16 type; |
1da177e4 | 4031 | |
588f0330 | 4032 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 4033 | |
cf66ba58 | 4034 | trace_netif_receive_skb(skb); |
9b22ea56 | 4035 | |
cc9bd5ce | 4036 | orig_dev = skb->dev; |
8f903c70 | 4037 | |
c1d2bbe1 | 4038 | skb_reset_network_header(skb); |
fda55eca ED |
4039 | if (!skb_transport_header_was_set(skb)) |
4040 | skb_reset_transport_header(skb); | |
0b5c9db1 | 4041 | skb_reset_mac_len(skb); |
1da177e4 LT |
4042 | |
4043 | pt_prev = NULL; | |
4044 | ||
63d8ea7f | 4045 | another_round: |
b6858177 | 4046 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
4047 | |
4048 | __this_cpu_inc(softnet_data.processed); | |
4049 | ||
8ad227ff PM |
4050 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
4051 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 4052 | skb = skb_vlan_untag(skb); |
bcc6d479 | 4053 | if (unlikely(!skb)) |
2c17d27c | 4054 | goto out; |
bcc6d479 JP |
4055 | } |
4056 | ||
1da177e4 LT |
4057 | #ifdef CONFIG_NET_CLS_ACT |
4058 | if (skb->tc_verd & TC_NCLS) { | |
4059 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
4060 | goto ncls; | |
4061 | } | |
4062 | #endif | |
4063 | ||
9754e293 | 4064 | if (pfmemalloc) |
b4b9e355 MG |
4065 | goto skip_taps; |
4066 | ||
1da177e4 | 4067 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
4068 | if (pt_prev) |
4069 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4070 | pt_prev = ptype; | |
4071 | } | |
4072 | ||
4073 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
4074 | if (pt_prev) | |
4075 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4076 | pt_prev = ptype; | |
1da177e4 LT |
4077 | } |
4078 | ||
b4b9e355 | 4079 | skip_taps: |
1cf51900 | 4080 | #ifdef CONFIG_NET_INGRESS |
4577139b | 4081 | if (static_key_false(&ingress_needed)) { |
1f211a1b | 4082 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 4083 | if (!skb) |
2c17d27c | 4084 | goto out; |
e687ad60 PN |
4085 | |
4086 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 4087 | goto out; |
4577139b | 4088 | } |
1cf51900 PN |
4089 | #endif |
4090 | #ifdef CONFIG_NET_CLS_ACT | |
4577139b | 4091 | skb->tc_verd = 0; |
1da177e4 LT |
4092 | ncls: |
4093 | #endif | |
9754e293 | 4094 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
4095 | goto drop; |
4096 | ||
df8a39de | 4097 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
4098 | if (pt_prev) { |
4099 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4100 | pt_prev = NULL; | |
4101 | } | |
48cc32d3 | 4102 | if (vlan_do_receive(&skb)) |
2425717b JF |
4103 | goto another_round; |
4104 | else if (unlikely(!skb)) | |
2c17d27c | 4105 | goto out; |
2425717b JF |
4106 | } |
4107 | ||
48cc32d3 | 4108 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
4109 | if (rx_handler) { |
4110 | if (pt_prev) { | |
4111 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4112 | pt_prev = NULL; | |
4113 | } | |
8a4eb573 JP |
4114 | switch (rx_handler(&skb)) { |
4115 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 4116 | ret = NET_RX_SUCCESS; |
2c17d27c | 4117 | goto out; |
8a4eb573 | 4118 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 4119 | goto another_round; |
8a4eb573 JP |
4120 | case RX_HANDLER_EXACT: |
4121 | deliver_exact = true; | |
4122 | case RX_HANDLER_PASS: | |
4123 | break; | |
4124 | default: | |
4125 | BUG(); | |
4126 | } | |
ab95bfe0 | 4127 | } |
1da177e4 | 4128 | |
df8a39de JP |
4129 | if (unlikely(skb_vlan_tag_present(skb))) { |
4130 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
4131 | skb->pkt_type = PACKET_OTHERHOST; |
4132 | /* Note: we might in the future use prio bits | |
4133 | * and set skb->priority like in vlan_do_receive() | |
4134 | * For the time being, just ignore Priority Code Point | |
4135 | */ | |
4136 | skb->vlan_tci = 0; | |
4137 | } | |
48cc32d3 | 4138 | |
7866a621 SN |
4139 | type = skb->protocol; |
4140 | ||
63d8ea7f | 4141 | /* deliver only exact match when indicated */ |
7866a621 SN |
4142 | if (likely(!deliver_exact)) { |
4143 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4144 | &ptype_base[ntohs(type) & | |
4145 | PTYPE_HASH_MASK]); | |
4146 | } | |
1f3c8804 | 4147 | |
7866a621 SN |
4148 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
4149 | &orig_dev->ptype_specific); | |
4150 | ||
4151 | if (unlikely(skb->dev != orig_dev)) { | |
4152 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4153 | &skb->dev->ptype_specific); | |
1da177e4 LT |
4154 | } |
4155 | ||
4156 | if (pt_prev) { | |
1080e512 | 4157 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 4158 | goto drop; |
1080e512 MT |
4159 | else |
4160 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 4161 | } else { |
b4b9e355 | 4162 | drop: |
6e7333d3 JW |
4163 | if (!deliver_exact) |
4164 | atomic_long_inc(&skb->dev->rx_dropped); | |
4165 | else | |
4166 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
4167 | kfree_skb(skb); |
4168 | /* Jamal, now you will not able to escape explaining | |
4169 | * me how you were going to use this. :-) | |
4170 | */ | |
4171 | ret = NET_RX_DROP; | |
4172 | } | |
4173 | ||
2c17d27c | 4174 | out: |
9754e293 DM |
4175 | return ret; |
4176 | } | |
4177 | ||
4178 | static int __netif_receive_skb(struct sk_buff *skb) | |
4179 | { | |
4180 | int ret; | |
4181 | ||
4182 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
4183 | unsigned long pflags = current->flags; | |
4184 | ||
4185 | /* | |
4186 | * PFMEMALLOC skbs are special, they should | |
4187 | * - be delivered to SOCK_MEMALLOC sockets only | |
4188 | * - stay away from userspace | |
4189 | * - have bounded memory usage | |
4190 | * | |
4191 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
4192 | * context down to all allocation sites. | |
4193 | */ | |
4194 | current->flags |= PF_MEMALLOC; | |
4195 | ret = __netif_receive_skb_core(skb, true); | |
4196 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
4197 | } else | |
4198 | ret = __netif_receive_skb_core(skb, false); | |
4199 | ||
1da177e4 LT |
4200 | return ret; |
4201 | } | |
0a9627f2 | 4202 | |
ae78dbfa | 4203 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 4204 | { |
2c17d27c JA |
4205 | int ret; |
4206 | ||
588f0330 | 4207 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 4208 | |
c1f19b51 RC |
4209 | if (skb_defer_rx_timestamp(skb)) |
4210 | return NET_RX_SUCCESS; | |
4211 | ||
2c17d27c JA |
4212 | rcu_read_lock(); |
4213 | ||
df334545 | 4214 | #ifdef CONFIG_RPS |
c5905afb | 4215 | if (static_key_false(&rps_needed)) { |
3b098e2d | 4216 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 4217 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 4218 | |
3b098e2d ED |
4219 | if (cpu >= 0) { |
4220 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4221 | rcu_read_unlock(); | |
adc9300e | 4222 | return ret; |
3b098e2d | 4223 | } |
fec5e652 | 4224 | } |
1e94d72f | 4225 | #endif |
2c17d27c JA |
4226 | ret = __netif_receive_skb(skb); |
4227 | rcu_read_unlock(); | |
4228 | return ret; | |
0a9627f2 | 4229 | } |
ae78dbfa BH |
4230 | |
4231 | /** | |
4232 | * netif_receive_skb - process receive buffer from network | |
4233 | * @skb: buffer to process | |
4234 | * | |
4235 | * netif_receive_skb() is the main receive data processing function. | |
4236 | * It always succeeds. The buffer may be dropped during processing | |
4237 | * for congestion control or by the protocol layers. | |
4238 | * | |
4239 | * This function may only be called from softirq context and interrupts | |
4240 | * should be enabled. | |
4241 | * | |
4242 | * Return values (usually ignored): | |
4243 | * NET_RX_SUCCESS: no congestion | |
4244 | * NET_RX_DROP: packet was dropped | |
4245 | */ | |
04eb4489 | 4246 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa BH |
4247 | { |
4248 | trace_netif_receive_skb_entry(skb); | |
4249 | ||
4250 | return netif_receive_skb_internal(skb); | |
4251 | } | |
04eb4489 | 4252 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 4253 | |
88751275 ED |
4254 | /* Network device is going away, flush any packets still pending |
4255 | * Called with irqs disabled. | |
4256 | */ | |
152102c7 | 4257 | static void flush_backlog(void *arg) |
6e583ce5 | 4258 | { |
152102c7 | 4259 | struct net_device *dev = arg; |
903ceff7 | 4260 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
6e583ce5 SH |
4261 | struct sk_buff *skb, *tmp; |
4262 | ||
e36fa2f7 | 4263 | rps_lock(sd); |
6e7676c1 | 4264 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 4265 | if (skb->dev == dev) { |
e36fa2f7 | 4266 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 4267 | kfree_skb(skb); |
76cc8b13 | 4268 | input_queue_head_incr(sd); |
6e583ce5 | 4269 | } |
6e7676c1 | 4270 | } |
e36fa2f7 | 4271 | rps_unlock(sd); |
6e7676c1 CG |
4272 | |
4273 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
4274 | if (skb->dev == dev) { | |
4275 | __skb_unlink(skb, &sd->process_queue); | |
4276 | kfree_skb(skb); | |
76cc8b13 | 4277 | input_queue_head_incr(sd); |
6e7676c1 CG |
4278 | } |
4279 | } | |
6e583ce5 SH |
4280 | } |
4281 | ||
d565b0a1 HX |
4282 | static int napi_gro_complete(struct sk_buff *skb) |
4283 | { | |
22061d80 | 4284 | struct packet_offload *ptype; |
d565b0a1 | 4285 | __be16 type = skb->protocol; |
22061d80 | 4286 | struct list_head *head = &offload_base; |
d565b0a1 HX |
4287 | int err = -ENOENT; |
4288 | ||
c3c7c254 ED |
4289 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
4290 | ||
fc59f9a3 HX |
4291 | if (NAPI_GRO_CB(skb)->count == 1) { |
4292 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 4293 | goto out; |
fc59f9a3 | 4294 | } |
d565b0a1 HX |
4295 | |
4296 | rcu_read_lock(); | |
4297 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4298 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
4299 | continue; |
4300 | ||
299603e8 | 4301 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
4302 | break; |
4303 | } | |
4304 | rcu_read_unlock(); | |
4305 | ||
4306 | if (err) { | |
4307 | WARN_ON(&ptype->list == head); | |
4308 | kfree_skb(skb); | |
4309 | return NET_RX_SUCCESS; | |
4310 | } | |
4311 | ||
4312 | out: | |
ae78dbfa | 4313 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
4314 | } |
4315 | ||
2e71a6f8 ED |
4316 | /* napi->gro_list contains packets ordered by age. |
4317 | * youngest packets at the head of it. | |
4318 | * Complete skbs in reverse order to reduce latencies. | |
4319 | */ | |
4320 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 4321 | { |
2e71a6f8 | 4322 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 4323 | |
2e71a6f8 ED |
4324 | /* scan list and build reverse chain */ |
4325 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
4326 | skb->prev = prev; | |
4327 | prev = skb; | |
4328 | } | |
4329 | ||
4330 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 4331 | skb->next = NULL; |
2e71a6f8 ED |
4332 | |
4333 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
4334 | return; | |
4335 | ||
4336 | prev = skb->prev; | |
d565b0a1 | 4337 | napi_gro_complete(skb); |
2e71a6f8 | 4338 | napi->gro_count--; |
d565b0a1 HX |
4339 | } |
4340 | ||
4341 | napi->gro_list = NULL; | |
4342 | } | |
86cac58b | 4343 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 4344 | |
89c5fa33 ED |
4345 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
4346 | { | |
4347 | struct sk_buff *p; | |
4348 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 4349 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
4350 | |
4351 | for (p = napi->gro_list; p; p = p->next) { | |
4352 | unsigned long diffs; | |
4353 | ||
0b4cec8c TH |
4354 | NAPI_GRO_CB(p)->flush = 0; |
4355 | ||
4356 | if (hash != skb_get_hash_raw(p)) { | |
4357 | NAPI_GRO_CB(p)->same_flow = 0; | |
4358 | continue; | |
4359 | } | |
4360 | ||
89c5fa33 ED |
4361 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
4362 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
ce87fc6c | 4363 | diffs |= skb_metadata_dst_cmp(p, skb); |
89c5fa33 ED |
4364 | if (maclen == ETH_HLEN) |
4365 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 4366 | skb_mac_header(skb)); |
89c5fa33 ED |
4367 | else if (!diffs) |
4368 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 4369 | skb_mac_header(skb), |
89c5fa33 ED |
4370 | maclen); |
4371 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
4372 | } |
4373 | } | |
4374 | ||
299603e8 JC |
4375 | static void skb_gro_reset_offset(struct sk_buff *skb) |
4376 | { | |
4377 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4378 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4379 | ||
4380 | NAPI_GRO_CB(skb)->data_offset = 0; | |
4381 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
4382 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
4383 | ||
4384 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
4385 | pinfo->nr_frags && | |
4386 | !PageHighMem(skb_frag_page(frag0))) { | |
4387 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
4388 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
89c5fa33 ED |
4389 | } |
4390 | } | |
4391 | ||
a50e233c ED |
4392 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
4393 | { | |
4394 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4395 | ||
4396 | BUG_ON(skb->end - skb->tail < grow); | |
4397 | ||
4398 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4399 | ||
4400 | skb->data_len -= grow; | |
4401 | skb->tail += grow; | |
4402 | ||
4403 | pinfo->frags[0].page_offset += grow; | |
4404 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4405 | ||
4406 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4407 | skb_frag_unref(skb, 0); | |
4408 | memmove(pinfo->frags, pinfo->frags + 1, | |
4409 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4410 | } | |
4411 | } | |
4412 | ||
bb728820 | 4413 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4414 | { |
4415 | struct sk_buff **pp = NULL; | |
22061d80 | 4416 | struct packet_offload *ptype; |
d565b0a1 | 4417 | __be16 type = skb->protocol; |
22061d80 | 4418 | struct list_head *head = &offload_base; |
0da2afd5 | 4419 | int same_flow; |
5b252f0c | 4420 | enum gro_result ret; |
a50e233c | 4421 | int grow; |
d565b0a1 | 4422 | |
9c62a68d | 4423 | if (!(skb->dev->features & NETIF_F_GRO)) |
d565b0a1 HX |
4424 | goto normal; |
4425 | ||
5a212329 | 4426 | if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad) |
f17f5c91 HX |
4427 | goto normal; |
4428 | ||
89c5fa33 ED |
4429 | gro_list_prepare(napi, skb); |
4430 | ||
d565b0a1 HX |
4431 | rcu_read_lock(); |
4432 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4433 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4434 | continue; |
4435 | ||
86911732 | 4436 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4437 | skb_reset_mac_len(skb); |
d565b0a1 HX |
4438 | NAPI_GRO_CB(skb)->same_flow = 0; |
4439 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 4440 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 4441 | NAPI_GRO_CB(skb)->encap_mark = 0; |
a0ca153f | 4442 | NAPI_GRO_CB(skb)->is_fou = 0; |
15e2396d | 4443 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4444 | |
662880f4 TH |
4445 | /* Setup for GRO checksum validation */ |
4446 | switch (skb->ip_summed) { | |
4447 | case CHECKSUM_COMPLETE: | |
4448 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4449 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4450 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4451 | break; | |
4452 | case CHECKSUM_UNNECESSARY: | |
4453 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4454 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4455 | break; | |
4456 | default: | |
4457 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4458 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4459 | } | |
d565b0a1 | 4460 | |
f191a1d1 | 4461 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4462 | break; |
4463 | } | |
4464 | rcu_read_unlock(); | |
4465 | ||
4466 | if (&ptype->list == head) | |
4467 | goto normal; | |
4468 | ||
0da2afd5 | 4469 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4470 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4471 | |
d565b0a1 HX |
4472 | if (pp) { |
4473 | struct sk_buff *nskb = *pp; | |
4474 | ||
4475 | *pp = nskb->next; | |
4476 | nskb->next = NULL; | |
4477 | napi_gro_complete(nskb); | |
4ae5544f | 4478 | napi->gro_count--; |
d565b0a1 HX |
4479 | } |
4480 | ||
0da2afd5 | 4481 | if (same_flow) |
d565b0a1 HX |
4482 | goto ok; |
4483 | ||
600adc18 | 4484 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4485 | goto normal; |
d565b0a1 | 4486 | |
600adc18 ED |
4487 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4488 | struct sk_buff *nskb = napi->gro_list; | |
4489 | ||
4490 | /* locate the end of the list to select the 'oldest' flow */ | |
4491 | while (nskb->next) { | |
4492 | pp = &nskb->next; | |
4493 | nskb = *pp; | |
4494 | } | |
4495 | *pp = NULL; | |
4496 | nskb->next = NULL; | |
4497 | napi_gro_complete(nskb); | |
4498 | } else { | |
4499 | napi->gro_count++; | |
4500 | } | |
d565b0a1 | 4501 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4502 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4503 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4504 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4505 | skb->next = napi->gro_list; |
4506 | napi->gro_list = skb; | |
5d0d9be8 | 4507 | ret = GRO_HELD; |
d565b0a1 | 4508 | |
ad0f9904 | 4509 | pull: |
a50e233c ED |
4510 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4511 | if (grow > 0) | |
4512 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4513 | ok: |
5d0d9be8 | 4514 | return ret; |
d565b0a1 HX |
4515 | |
4516 | normal: | |
ad0f9904 HX |
4517 | ret = GRO_NORMAL; |
4518 | goto pull; | |
5d38a079 | 4519 | } |
96e93eab | 4520 | |
bf5a755f JC |
4521 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4522 | { | |
4523 | struct list_head *offload_head = &offload_base; | |
4524 | struct packet_offload *ptype; | |
4525 | ||
4526 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4527 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4528 | continue; | |
4529 | return ptype; | |
4530 | } | |
4531 | return NULL; | |
4532 | } | |
e27a2f83 | 4533 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4534 | |
4535 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4536 | { | |
4537 | struct list_head *offload_head = &offload_base; | |
4538 | struct packet_offload *ptype; | |
4539 | ||
4540 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4541 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4542 | continue; | |
4543 | return ptype; | |
4544 | } | |
4545 | return NULL; | |
4546 | } | |
e27a2f83 | 4547 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4548 | |
bb728820 | 4549 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4550 | { |
5d0d9be8 HX |
4551 | switch (ret) { |
4552 | case GRO_NORMAL: | |
ae78dbfa | 4553 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4554 | ret = GRO_DROP; |
4555 | break; | |
5d38a079 | 4556 | |
5d0d9be8 | 4557 | case GRO_DROP: |
5d38a079 HX |
4558 | kfree_skb(skb); |
4559 | break; | |
5b252f0c | 4560 | |
daa86548 | 4561 | case GRO_MERGED_FREE: |
ce87fc6c JG |
4562 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) { |
4563 | skb_dst_drop(skb); | |
d7e8883c | 4564 | kmem_cache_free(skbuff_head_cache, skb); |
ce87fc6c | 4565 | } else { |
d7e8883c | 4566 | __kfree_skb(skb); |
ce87fc6c | 4567 | } |
daa86548 ED |
4568 | break; |
4569 | ||
5b252f0c BH |
4570 | case GRO_HELD: |
4571 | case GRO_MERGED: | |
4572 | break; | |
5d38a079 HX |
4573 | } |
4574 | ||
c7c4b3b6 | 4575 | return ret; |
5d0d9be8 | 4576 | } |
5d0d9be8 | 4577 | |
c7c4b3b6 | 4578 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4579 | { |
93f93a44 | 4580 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 4581 | trace_napi_gro_receive_entry(skb); |
86911732 | 4582 | |
a50e233c ED |
4583 | skb_gro_reset_offset(skb); |
4584 | ||
89c5fa33 | 4585 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4586 | } |
4587 | EXPORT_SYMBOL(napi_gro_receive); | |
4588 | ||
d0c2b0d2 | 4589 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4590 | { |
93a35f59 ED |
4591 | if (unlikely(skb->pfmemalloc)) { |
4592 | consume_skb(skb); | |
4593 | return; | |
4594 | } | |
96e93eab | 4595 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4596 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4597 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4598 | skb->vlan_tci = 0; |
66c46d74 | 4599 | skb->dev = napi->dev; |
6d152e23 | 4600 | skb->skb_iif = 0; |
c3caf119 JC |
4601 | skb->encapsulation = 0; |
4602 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4603 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
96e93eab HX |
4604 | |
4605 | napi->skb = skb; | |
4606 | } | |
96e93eab | 4607 | |
76620aaf | 4608 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4609 | { |
5d38a079 | 4610 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4611 | |
4612 | if (!skb) { | |
fd11a83d | 4613 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
4614 | if (skb) { |
4615 | napi->skb = skb; | |
4616 | skb_mark_napi_id(skb, napi); | |
4617 | } | |
80595d59 | 4618 | } |
96e93eab HX |
4619 | return skb; |
4620 | } | |
76620aaf | 4621 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4622 | |
a50e233c ED |
4623 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4624 | struct sk_buff *skb, | |
4625 | gro_result_t ret) | |
96e93eab | 4626 | { |
5d0d9be8 HX |
4627 | switch (ret) { |
4628 | case GRO_NORMAL: | |
a50e233c ED |
4629 | case GRO_HELD: |
4630 | __skb_push(skb, ETH_HLEN); | |
4631 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4632 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4633 | ret = GRO_DROP; |
86911732 | 4634 | break; |
5d38a079 | 4635 | |
5d0d9be8 | 4636 | case GRO_DROP: |
5d0d9be8 HX |
4637 | case GRO_MERGED_FREE: |
4638 | napi_reuse_skb(napi, skb); | |
4639 | break; | |
5b252f0c BH |
4640 | |
4641 | case GRO_MERGED: | |
4642 | break; | |
5d0d9be8 | 4643 | } |
5d38a079 | 4644 | |
c7c4b3b6 | 4645 | return ret; |
5d38a079 | 4646 | } |
5d0d9be8 | 4647 | |
a50e233c ED |
4648 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4649 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4650 | * We copy ethernet header into skb->data to have a common layout. | |
4651 | */ | |
4adb9c4a | 4652 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4653 | { |
4654 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4655 | const struct ethhdr *eth; |
4656 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4657 | |
4658 | napi->skb = NULL; | |
4659 | ||
a50e233c ED |
4660 | skb_reset_mac_header(skb); |
4661 | skb_gro_reset_offset(skb); | |
4662 | ||
4663 | eth = skb_gro_header_fast(skb, 0); | |
4664 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4665 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4666 | if (unlikely(!eth)) { | |
4da46ceb AC |
4667 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
4668 | __func__, napi->dev->name); | |
a50e233c ED |
4669 | napi_reuse_skb(napi, skb); |
4670 | return NULL; | |
4671 | } | |
4672 | } else { | |
4673 | gro_pull_from_frag0(skb, hlen); | |
4674 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4675 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4676 | } |
a50e233c ED |
4677 | __skb_pull(skb, hlen); |
4678 | ||
4679 | /* | |
4680 | * This works because the only protocols we care about don't require | |
4681 | * special handling. | |
4682 | * We'll fix it up properly in napi_frags_finish() | |
4683 | */ | |
4684 | skb->protocol = eth->h_proto; | |
76620aaf | 4685 | |
76620aaf HX |
4686 | return skb; |
4687 | } | |
76620aaf | 4688 | |
c7c4b3b6 | 4689 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4690 | { |
76620aaf | 4691 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4692 | |
4693 | if (!skb) | |
c7c4b3b6 | 4694 | return GRO_DROP; |
5d0d9be8 | 4695 | |
ae78dbfa BH |
4696 | trace_napi_gro_frags_entry(skb); |
4697 | ||
89c5fa33 | 4698 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4699 | } |
5d38a079 HX |
4700 | EXPORT_SYMBOL(napi_gro_frags); |
4701 | ||
573e8fca TH |
4702 | /* Compute the checksum from gro_offset and return the folded value |
4703 | * after adding in any pseudo checksum. | |
4704 | */ | |
4705 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
4706 | { | |
4707 | __wsum wsum; | |
4708 | __sum16 sum; | |
4709 | ||
4710 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
4711 | ||
4712 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
4713 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
4714 | if (likely(!sum)) { | |
4715 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
4716 | !skb->csum_complete_sw) | |
4717 | netdev_rx_csum_fault(skb->dev); | |
4718 | } | |
4719 | ||
4720 | NAPI_GRO_CB(skb)->csum = wsum; | |
4721 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4722 | ||
4723 | return sum; | |
4724 | } | |
4725 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
4726 | ||
e326bed2 | 4727 | /* |
855abcf0 | 4728 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4729 | * Note: called with local irq disabled, but exits with local irq enabled. |
4730 | */ | |
4731 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4732 | { | |
4733 | #ifdef CONFIG_RPS | |
4734 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4735 | ||
4736 | if (remsd) { | |
4737 | sd->rps_ipi_list = NULL; | |
4738 | ||
4739 | local_irq_enable(); | |
4740 | ||
4741 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4742 | while (remsd) { | |
4743 | struct softnet_data *next = remsd->rps_ipi_next; | |
4744 | ||
4745 | if (cpu_online(remsd->cpu)) | |
c46fff2a | 4746 | smp_call_function_single_async(remsd->cpu, |
fce8ad15 | 4747 | &remsd->csd); |
e326bed2 ED |
4748 | remsd = next; |
4749 | } | |
4750 | } else | |
4751 | #endif | |
4752 | local_irq_enable(); | |
4753 | } | |
4754 | ||
d75b1ade ED |
4755 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
4756 | { | |
4757 | #ifdef CONFIG_RPS | |
4758 | return sd->rps_ipi_list != NULL; | |
4759 | #else | |
4760 | return false; | |
4761 | #endif | |
4762 | } | |
4763 | ||
bea3348e | 4764 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4765 | { |
4766 | int work = 0; | |
eecfd7c4 | 4767 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4768 | |
e326bed2 ED |
4769 | /* Check if we have pending ipi, its better to send them now, |
4770 | * not waiting net_rx_action() end. | |
4771 | */ | |
d75b1ade | 4772 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
4773 | local_irq_disable(); |
4774 | net_rps_action_and_irq_enable(sd); | |
4775 | } | |
d75b1ade | 4776 | |
bea3348e | 4777 | napi->weight = weight_p; |
6e7676c1 | 4778 | local_irq_disable(); |
11ef7a89 | 4779 | while (1) { |
1da177e4 | 4780 | struct sk_buff *skb; |
6e7676c1 CG |
4781 | |
4782 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 4783 | rcu_read_lock(); |
6e7676c1 CG |
4784 | local_irq_enable(); |
4785 | __netif_receive_skb(skb); | |
2c17d27c | 4786 | rcu_read_unlock(); |
6e7676c1 | 4787 | local_irq_disable(); |
76cc8b13 TH |
4788 | input_queue_head_incr(sd); |
4789 | if (++work >= quota) { | |
4790 | local_irq_enable(); | |
4791 | return work; | |
4792 | } | |
6e7676c1 | 4793 | } |
1da177e4 | 4794 | |
e36fa2f7 | 4795 | rps_lock(sd); |
11ef7a89 | 4796 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
4797 | /* |
4798 | * Inline a custom version of __napi_complete(). | |
4799 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
4800 | * and NAPI_STATE_SCHED is the only possible flag set |
4801 | * on backlog. | |
4802 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
4803 | * and we dont need an smp_mb() memory barrier. |
4804 | */ | |
eecfd7c4 | 4805 | napi->state = 0; |
11ef7a89 | 4806 | rps_unlock(sd); |
eecfd7c4 | 4807 | |
11ef7a89 | 4808 | break; |
bea3348e | 4809 | } |
11ef7a89 TH |
4810 | |
4811 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
4812 | &sd->process_queue); | |
e36fa2f7 | 4813 | rps_unlock(sd); |
6e7676c1 CG |
4814 | } |
4815 | local_irq_enable(); | |
1da177e4 | 4816 | |
bea3348e SH |
4817 | return work; |
4818 | } | |
1da177e4 | 4819 | |
bea3348e SH |
4820 | /** |
4821 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4822 | * @n: entry to schedule |
bea3348e | 4823 | * |
bc9ad166 ED |
4824 | * The entry's receive function will be scheduled to run. |
4825 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 4826 | */ |
b5606c2d | 4827 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4828 | { |
4829 | unsigned long flags; | |
1da177e4 | 4830 | |
bea3348e | 4831 | local_irq_save(flags); |
903ceff7 | 4832 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 4833 | local_irq_restore(flags); |
1da177e4 | 4834 | } |
bea3348e SH |
4835 | EXPORT_SYMBOL(__napi_schedule); |
4836 | ||
bc9ad166 ED |
4837 | /** |
4838 | * __napi_schedule_irqoff - schedule for receive | |
4839 | * @n: entry to schedule | |
4840 | * | |
4841 | * Variant of __napi_schedule() assuming hard irqs are masked | |
4842 | */ | |
4843 | void __napi_schedule_irqoff(struct napi_struct *n) | |
4844 | { | |
4845 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
4846 | } | |
4847 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
4848 | ||
d565b0a1 HX |
4849 | void __napi_complete(struct napi_struct *n) |
4850 | { | |
4851 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
d565b0a1 | 4852 | |
d75b1ade | 4853 | list_del_init(&n->poll_list); |
4e857c58 | 4854 | smp_mb__before_atomic(); |
d565b0a1 HX |
4855 | clear_bit(NAPI_STATE_SCHED, &n->state); |
4856 | } | |
4857 | EXPORT_SYMBOL(__napi_complete); | |
4858 | ||
3b47d303 | 4859 | void napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 HX |
4860 | { |
4861 | unsigned long flags; | |
4862 | ||
4863 | /* | |
4864 | * don't let napi dequeue from the cpu poll list | |
4865 | * just in case its running on a different cpu | |
4866 | */ | |
4867 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4868 | return; | |
4869 | ||
3b47d303 ED |
4870 | if (n->gro_list) { |
4871 | unsigned long timeout = 0; | |
d75b1ade | 4872 | |
3b47d303 ED |
4873 | if (work_done) |
4874 | timeout = n->dev->gro_flush_timeout; | |
4875 | ||
4876 | if (timeout) | |
4877 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
4878 | HRTIMER_MODE_REL_PINNED); | |
4879 | else | |
4880 | napi_gro_flush(n, false); | |
4881 | } | |
d75b1ade ED |
4882 | if (likely(list_empty(&n->poll_list))) { |
4883 | WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state)); | |
4884 | } else { | |
4885 | /* If n->poll_list is not empty, we need to mask irqs */ | |
4886 | local_irq_save(flags); | |
4887 | __napi_complete(n); | |
4888 | local_irq_restore(flags); | |
4889 | } | |
d565b0a1 | 4890 | } |
3b47d303 | 4891 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 4892 | |
af12fa6e | 4893 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 4894 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
4895 | { |
4896 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4897 | struct napi_struct *napi; | |
4898 | ||
4899 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4900 | if (napi->napi_id == napi_id) | |
4901 | return napi; | |
4902 | ||
4903 | return NULL; | |
4904 | } | |
02d62e86 ED |
4905 | |
4906 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
ce6aea93 | 4907 | #define BUSY_POLL_BUDGET 8 |
02d62e86 ED |
4908 | bool sk_busy_loop(struct sock *sk, int nonblock) |
4909 | { | |
4910 | unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0; | |
ce6aea93 | 4911 | int (*busy_poll)(struct napi_struct *dev); |
02d62e86 ED |
4912 | struct napi_struct *napi; |
4913 | int rc = false; | |
4914 | ||
2a028ecb | 4915 | rcu_read_lock(); |
02d62e86 ED |
4916 | |
4917 | napi = napi_by_id(sk->sk_napi_id); | |
4918 | if (!napi) | |
4919 | goto out; | |
4920 | ||
ce6aea93 ED |
4921 | /* Note: ndo_busy_poll method is optional in linux-4.5 */ |
4922 | busy_poll = napi->dev->netdev_ops->ndo_busy_poll; | |
02d62e86 ED |
4923 | |
4924 | do { | |
ce6aea93 | 4925 | rc = 0; |
2a028ecb | 4926 | local_bh_disable(); |
ce6aea93 ED |
4927 | if (busy_poll) { |
4928 | rc = busy_poll(napi); | |
4929 | } else if (napi_schedule_prep(napi)) { | |
4930 | void *have = netpoll_poll_lock(napi); | |
4931 | ||
4932 | if (test_bit(NAPI_STATE_SCHED, &napi->state)) { | |
4933 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
4934 | trace_napi_poll(napi); | |
4935 | if (rc == BUSY_POLL_BUDGET) { | |
4936 | napi_complete_done(napi, rc); | |
4937 | napi_schedule(napi); | |
4938 | } | |
4939 | } | |
4940 | netpoll_poll_unlock(have); | |
4941 | } | |
2a028ecb ED |
4942 | if (rc > 0) |
4943 | NET_ADD_STATS_BH(sock_net(sk), | |
4944 | LINUX_MIB_BUSYPOLLRXPACKETS, rc); | |
4945 | local_bh_enable(); | |
02d62e86 ED |
4946 | |
4947 | if (rc == LL_FLUSH_FAILED) | |
4948 | break; /* permanent failure */ | |
4949 | ||
02d62e86 | 4950 | cpu_relax(); |
02d62e86 ED |
4951 | } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) && |
4952 | !need_resched() && !busy_loop_timeout(end_time)); | |
4953 | ||
4954 | rc = !skb_queue_empty(&sk->sk_receive_queue); | |
4955 | out: | |
2a028ecb | 4956 | rcu_read_unlock(); |
02d62e86 ED |
4957 | return rc; |
4958 | } | |
4959 | EXPORT_SYMBOL(sk_busy_loop); | |
4960 | ||
4961 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e ET |
4962 | |
4963 | void napi_hash_add(struct napi_struct *napi) | |
4964 | { | |
d64b5e85 ED |
4965 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
4966 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 4967 | return; |
af12fa6e | 4968 | |
52bd2d62 | 4969 | spin_lock(&napi_hash_lock); |
af12fa6e | 4970 | |
52bd2d62 ED |
4971 | /* 0..NR_CPUS+1 range is reserved for sender_cpu use */ |
4972 | do { | |
4973 | if (unlikely(++napi_gen_id < NR_CPUS + 1)) | |
4974 | napi_gen_id = NR_CPUS + 1; | |
4975 | } while (napi_by_id(napi_gen_id)); | |
4976 | napi->napi_id = napi_gen_id; | |
af12fa6e | 4977 | |
52bd2d62 ED |
4978 | hlist_add_head_rcu(&napi->napi_hash_node, |
4979 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 4980 | |
52bd2d62 | 4981 | spin_unlock(&napi_hash_lock); |
af12fa6e ET |
4982 | } |
4983 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
4984 | ||
4985 | /* Warning : caller is responsible to make sure rcu grace period | |
4986 | * is respected before freeing memory containing @napi | |
4987 | */ | |
34cbe27e | 4988 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 4989 | { |
34cbe27e ED |
4990 | bool rcu_sync_needed = false; |
4991 | ||
af12fa6e ET |
4992 | spin_lock(&napi_hash_lock); |
4993 | ||
34cbe27e ED |
4994 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
4995 | rcu_sync_needed = true; | |
af12fa6e | 4996 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 4997 | } |
af12fa6e | 4998 | spin_unlock(&napi_hash_lock); |
34cbe27e | 4999 | return rcu_sync_needed; |
af12fa6e ET |
5000 | } |
5001 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
5002 | ||
3b47d303 ED |
5003 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
5004 | { | |
5005 | struct napi_struct *napi; | |
5006 | ||
5007 | napi = container_of(timer, struct napi_struct, timer); | |
5008 | if (napi->gro_list) | |
5009 | napi_schedule(napi); | |
5010 | ||
5011 | return HRTIMER_NORESTART; | |
5012 | } | |
5013 | ||
d565b0a1 HX |
5014 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
5015 | int (*poll)(struct napi_struct *, int), int weight) | |
5016 | { | |
5017 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
5018 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
5019 | napi->timer.function = napi_watchdog; | |
4ae5544f | 5020 | napi->gro_count = 0; |
d565b0a1 | 5021 | napi->gro_list = NULL; |
5d38a079 | 5022 | napi->skb = NULL; |
d565b0a1 | 5023 | napi->poll = poll; |
82dc3c63 ED |
5024 | if (weight > NAPI_POLL_WEIGHT) |
5025 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
5026 | weight, dev->name); | |
d565b0a1 HX |
5027 | napi->weight = weight; |
5028 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 5029 | napi->dev = dev; |
5d38a079 | 5030 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
5031 | spin_lock_init(&napi->poll_lock); |
5032 | napi->poll_owner = -1; | |
5033 | #endif | |
5034 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 5035 | napi_hash_add(napi); |
d565b0a1 HX |
5036 | } |
5037 | EXPORT_SYMBOL(netif_napi_add); | |
5038 | ||
3b47d303 ED |
5039 | void napi_disable(struct napi_struct *n) |
5040 | { | |
5041 | might_sleep(); | |
5042 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
5043 | ||
5044 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
5045 | msleep(1); | |
2d8bff12 NH |
5046 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
5047 | msleep(1); | |
3b47d303 ED |
5048 | |
5049 | hrtimer_cancel(&n->timer); | |
5050 | ||
5051 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
5052 | } | |
5053 | EXPORT_SYMBOL(napi_disable); | |
5054 | ||
93d05d4a | 5055 | /* Must be called in process context */ |
d565b0a1 HX |
5056 | void netif_napi_del(struct napi_struct *napi) |
5057 | { | |
93d05d4a ED |
5058 | might_sleep(); |
5059 | if (napi_hash_del(napi)) | |
5060 | synchronize_net(); | |
d7b06636 | 5061 | list_del_init(&napi->dev_list); |
76620aaf | 5062 | napi_free_frags(napi); |
d565b0a1 | 5063 | |
289dccbe | 5064 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 5065 | napi->gro_list = NULL; |
4ae5544f | 5066 | napi->gro_count = 0; |
d565b0a1 HX |
5067 | } |
5068 | EXPORT_SYMBOL(netif_napi_del); | |
5069 | ||
726ce70e HX |
5070 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
5071 | { | |
5072 | void *have; | |
5073 | int work, weight; | |
5074 | ||
5075 | list_del_init(&n->poll_list); | |
5076 | ||
5077 | have = netpoll_poll_lock(n); | |
5078 | ||
5079 | weight = n->weight; | |
5080 | ||
5081 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
5082 | * with netpoll's poll_napi(). Only the entity which | |
5083 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
5084 | * actually make the ->poll() call. Therefore we avoid | |
5085 | * accidentally calling ->poll() when NAPI is not scheduled. | |
5086 | */ | |
5087 | work = 0; | |
5088 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
5089 | work = n->poll(n, weight); | |
5090 | trace_napi_poll(n); | |
5091 | } | |
5092 | ||
5093 | WARN_ON_ONCE(work > weight); | |
5094 | ||
5095 | if (likely(work < weight)) | |
5096 | goto out_unlock; | |
5097 | ||
5098 | /* Drivers must not modify the NAPI state if they | |
5099 | * consume the entire weight. In such cases this code | |
5100 | * still "owns" the NAPI instance and therefore can | |
5101 | * move the instance around on the list at-will. | |
5102 | */ | |
5103 | if (unlikely(napi_disable_pending(n))) { | |
5104 | napi_complete(n); | |
5105 | goto out_unlock; | |
5106 | } | |
5107 | ||
5108 | if (n->gro_list) { | |
5109 | /* flush too old packets | |
5110 | * If HZ < 1000, flush all packets. | |
5111 | */ | |
5112 | napi_gro_flush(n, HZ >= 1000); | |
5113 | } | |
5114 | ||
001ce546 HX |
5115 | /* Some drivers may have called napi_schedule |
5116 | * prior to exhausting their budget. | |
5117 | */ | |
5118 | if (unlikely(!list_empty(&n->poll_list))) { | |
5119 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
5120 | n->dev ? n->dev->name : "backlog"); | |
5121 | goto out_unlock; | |
5122 | } | |
5123 | ||
726ce70e HX |
5124 | list_add_tail(&n->poll_list, repoll); |
5125 | ||
5126 | out_unlock: | |
5127 | netpoll_poll_unlock(have); | |
5128 | ||
5129 | return work; | |
5130 | } | |
5131 | ||
1da177e4 LT |
5132 | static void net_rx_action(struct softirq_action *h) |
5133 | { | |
903ceff7 | 5134 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
24f8b238 | 5135 | unsigned long time_limit = jiffies + 2; |
51b0bded | 5136 | int budget = netdev_budget; |
d75b1ade ED |
5137 | LIST_HEAD(list); |
5138 | LIST_HEAD(repoll); | |
53fb95d3 | 5139 | |
1da177e4 | 5140 | local_irq_disable(); |
d75b1ade ED |
5141 | list_splice_init(&sd->poll_list, &list); |
5142 | local_irq_enable(); | |
1da177e4 | 5143 | |
ceb8d5bf | 5144 | for (;;) { |
bea3348e | 5145 | struct napi_struct *n; |
1da177e4 | 5146 | |
ceb8d5bf HX |
5147 | if (list_empty(&list)) { |
5148 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
5149 | return; | |
5150 | break; | |
5151 | } | |
5152 | ||
6bd373eb HX |
5153 | n = list_first_entry(&list, struct napi_struct, poll_list); |
5154 | budget -= napi_poll(n, &repoll); | |
5155 | ||
d75b1ade | 5156 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
5157 | * Allow this to run for 2 jiffies since which will allow |
5158 | * an average latency of 1.5/HZ. | |
bea3348e | 5159 | */ |
ceb8d5bf HX |
5160 | if (unlikely(budget <= 0 || |
5161 | time_after_eq(jiffies, time_limit))) { | |
5162 | sd->time_squeeze++; | |
5163 | break; | |
5164 | } | |
1da177e4 | 5165 | } |
d75b1ade | 5166 | |
795bb1c0 | 5167 | __kfree_skb_flush(); |
d75b1ade ED |
5168 | local_irq_disable(); |
5169 | ||
5170 | list_splice_tail_init(&sd->poll_list, &list); | |
5171 | list_splice_tail(&repoll, &list); | |
5172 | list_splice(&list, &sd->poll_list); | |
5173 | if (!list_empty(&sd->poll_list)) | |
5174 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
5175 | ||
e326bed2 | 5176 | net_rps_action_and_irq_enable(sd); |
1da177e4 LT |
5177 | } |
5178 | ||
aa9d8560 | 5179 | struct netdev_adjacent { |
9ff162a8 | 5180 | struct net_device *dev; |
5d261913 VF |
5181 | |
5182 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 5183 | bool master; |
5d261913 | 5184 | |
5d261913 VF |
5185 | /* counter for the number of times this device was added to us */ |
5186 | u16 ref_nr; | |
5187 | ||
402dae96 VF |
5188 | /* private field for the users */ |
5189 | void *private; | |
5190 | ||
9ff162a8 JP |
5191 | struct list_head list; |
5192 | struct rcu_head rcu; | |
9ff162a8 JP |
5193 | }; |
5194 | ||
6ea29da1 | 5195 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 5196 | struct list_head *adj_list) |
9ff162a8 | 5197 | { |
5d261913 | 5198 | struct netdev_adjacent *adj; |
5d261913 | 5199 | |
2f268f12 | 5200 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
5201 | if (adj->dev == adj_dev) |
5202 | return adj; | |
9ff162a8 JP |
5203 | } |
5204 | return NULL; | |
5205 | } | |
5206 | ||
5207 | /** | |
5208 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
5209 | * @dev: device | |
5210 | * @upper_dev: upper device to check | |
5211 | * | |
5212 | * Find out if a device is linked to specified upper device and return true | |
5213 | * in case it is. Note that this checks only immediate upper device, | |
5214 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
5215 | */ | |
5216 | bool netdev_has_upper_dev(struct net_device *dev, | |
5217 | struct net_device *upper_dev) | |
5218 | { | |
5219 | ASSERT_RTNL(); | |
5220 | ||
6ea29da1 | 5221 | return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper); |
9ff162a8 JP |
5222 | } |
5223 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
5224 | ||
5225 | /** | |
5226 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
5227 | * @dev: device | |
5228 | * | |
5229 | * Find out if a device is linked to an upper device and return true in case | |
5230 | * it is. The caller must hold the RTNL lock. | |
5231 | */ | |
1d143d9f | 5232 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
5233 | { |
5234 | ASSERT_RTNL(); | |
5235 | ||
2f268f12 | 5236 | return !list_empty(&dev->all_adj_list.upper); |
9ff162a8 | 5237 | } |
9ff162a8 JP |
5238 | |
5239 | /** | |
5240 | * netdev_master_upper_dev_get - Get master upper device | |
5241 | * @dev: device | |
5242 | * | |
5243 | * Find a master upper device and return pointer to it or NULL in case | |
5244 | * it's not there. The caller must hold the RTNL lock. | |
5245 | */ | |
5246 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
5247 | { | |
aa9d8560 | 5248 | struct netdev_adjacent *upper; |
9ff162a8 JP |
5249 | |
5250 | ASSERT_RTNL(); | |
5251 | ||
2f268f12 | 5252 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
5253 | return NULL; |
5254 | ||
2f268f12 | 5255 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 5256 | struct netdev_adjacent, list); |
9ff162a8 JP |
5257 | if (likely(upper->master)) |
5258 | return upper->dev; | |
5259 | return NULL; | |
5260 | } | |
5261 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
5262 | ||
b6ccba4c VF |
5263 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
5264 | { | |
5265 | struct netdev_adjacent *adj; | |
5266 | ||
5267 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
5268 | ||
5269 | return adj->private; | |
5270 | } | |
5271 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
5272 | ||
44a40855 VY |
5273 | /** |
5274 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
5275 | * @dev: device | |
5276 | * @iter: list_head ** of the current position | |
5277 | * | |
5278 | * Gets the next device from the dev's upper list, starting from iter | |
5279 | * position. The caller must hold RCU read lock. | |
5280 | */ | |
5281 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
5282 | struct list_head **iter) | |
5283 | { | |
5284 | struct netdev_adjacent *upper; | |
5285 | ||
5286 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5287 | ||
5288 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5289 | ||
5290 | if (&upper->list == &dev->adj_list.upper) | |
5291 | return NULL; | |
5292 | ||
5293 | *iter = &upper->list; | |
5294 | ||
5295 | return upper->dev; | |
5296 | } | |
5297 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
5298 | ||
31088a11 VF |
5299 | /** |
5300 | * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list | |
48311f46 VF |
5301 | * @dev: device |
5302 | * @iter: list_head ** of the current position | |
5303 | * | |
5304 | * Gets the next device from the dev's upper list, starting from iter | |
5305 | * position. The caller must hold RCU read lock. | |
5306 | */ | |
2f268f12 VF |
5307 | struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, |
5308 | struct list_head **iter) | |
48311f46 VF |
5309 | { |
5310 | struct netdev_adjacent *upper; | |
5311 | ||
85328240 | 5312 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); |
48311f46 VF |
5313 | |
5314 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5315 | ||
2f268f12 | 5316 | if (&upper->list == &dev->all_adj_list.upper) |
48311f46 VF |
5317 | return NULL; |
5318 | ||
5319 | *iter = &upper->list; | |
5320 | ||
5321 | return upper->dev; | |
5322 | } | |
2f268f12 | 5323 | EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); |
48311f46 | 5324 | |
31088a11 VF |
5325 | /** |
5326 | * netdev_lower_get_next_private - Get the next ->private from the | |
5327 | * lower neighbour list | |
5328 | * @dev: device | |
5329 | * @iter: list_head ** of the current position | |
5330 | * | |
5331 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5332 | * list, starting from iter position. The caller must hold either hold the | |
5333 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 5334 | * list will remain unchanged. |
31088a11 VF |
5335 | */ |
5336 | void *netdev_lower_get_next_private(struct net_device *dev, | |
5337 | struct list_head **iter) | |
5338 | { | |
5339 | struct netdev_adjacent *lower; | |
5340 | ||
5341 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5342 | ||
5343 | if (&lower->list == &dev->adj_list.lower) | |
5344 | return NULL; | |
5345 | ||
6859e7df | 5346 | *iter = lower->list.next; |
31088a11 VF |
5347 | |
5348 | return lower->private; | |
5349 | } | |
5350 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
5351 | ||
5352 | /** | |
5353 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
5354 | * lower neighbour list, RCU | |
5355 | * variant | |
5356 | * @dev: device | |
5357 | * @iter: list_head ** of the current position | |
5358 | * | |
5359 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5360 | * list, starting from iter position. The caller must hold RCU read lock. | |
5361 | */ | |
5362 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
5363 | struct list_head **iter) | |
5364 | { | |
5365 | struct netdev_adjacent *lower; | |
5366 | ||
5367 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5368 | ||
5369 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5370 | ||
5371 | if (&lower->list == &dev->adj_list.lower) | |
5372 | return NULL; | |
5373 | ||
6859e7df | 5374 | *iter = &lower->list; |
31088a11 VF |
5375 | |
5376 | return lower->private; | |
5377 | } | |
5378 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
5379 | ||
4085ebe8 VY |
5380 | /** |
5381 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
5382 | * list | |
5383 | * @dev: device | |
5384 | * @iter: list_head ** of the current position | |
5385 | * | |
5386 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
5387 | * list, starting from iter position. The caller must hold RTNL lock or | |
5388 | * its own locking that guarantees that the neighbour lower | |
b469139e | 5389 | * list will remain unchanged. |
4085ebe8 VY |
5390 | */ |
5391 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5392 | { | |
5393 | struct netdev_adjacent *lower; | |
5394 | ||
cfdd28be | 5395 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
5396 | |
5397 | if (&lower->list == &dev->adj_list.lower) | |
5398 | return NULL; | |
5399 | ||
cfdd28be | 5400 | *iter = lower->list.next; |
4085ebe8 VY |
5401 | |
5402 | return lower->dev; | |
5403 | } | |
5404 | EXPORT_SYMBOL(netdev_lower_get_next); | |
5405 | ||
e001bfad | 5406 | /** |
5407 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
5408 | * lower neighbour list, RCU | |
5409 | * variant | |
5410 | * @dev: device | |
5411 | * | |
5412 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
5413 | * list. The caller must hold RCU read lock. | |
5414 | */ | |
5415 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
5416 | { | |
5417 | struct netdev_adjacent *lower; | |
5418 | ||
5419 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
5420 | struct netdev_adjacent, list); | |
5421 | if (lower) | |
5422 | return lower->private; | |
5423 | return NULL; | |
5424 | } | |
5425 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
5426 | ||
9ff162a8 JP |
5427 | /** |
5428 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
5429 | * @dev: device | |
5430 | * | |
5431 | * Find a master upper device and return pointer to it or NULL in case | |
5432 | * it's not there. The caller must hold the RCU read lock. | |
5433 | */ | |
5434 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
5435 | { | |
aa9d8560 | 5436 | struct netdev_adjacent *upper; |
9ff162a8 | 5437 | |
2f268f12 | 5438 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 5439 | struct netdev_adjacent, list); |
9ff162a8 JP |
5440 | if (upper && likely(upper->master)) |
5441 | return upper->dev; | |
5442 | return NULL; | |
5443 | } | |
5444 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
5445 | ||
0a59f3a9 | 5446 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
5447 | struct net_device *adj_dev, |
5448 | struct list_head *dev_list) | |
5449 | { | |
5450 | char linkname[IFNAMSIZ+7]; | |
5451 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
5452 | "upper_%s" : "lower_%s", adj_dev->name); | |
5453 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
5454 | linkname); | |
5455 | } | |
0a59f3a9 | 5456 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
5457 | char *name, |
5458 | struct list_head *dev_list) | |
5459 | { | |
5460 | char linkname[IFNAMSIZ+7]; | |
5461 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
5462 | "upper_%s" : "lower_%s", name); | |
5463 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
5464 | } | |
5465 | ||
7ce64c79 AF |
5466 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
5467 | struct net_device *adj_dev, | |
5468 | struct list_head *dev_list) | |
5469 | { | |
5470 | return (dev_list == &dev->adj_list.upper || | |
5471 | dev_list == &dev->adj_list.lower) && | |
5472 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
5473 | } | |
3ee32707 | 5474 | |
5d261913 VF |
5475 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
5476 | struct net_device *adj_dev, | |
7863c054 | 5477 | struct list_head *dev_list, |
402dae96 | 5478 | void *private, bool master) |
5d261913 VF |
5479 | { |
5480 | struct netdev_adjacent *adj; | |
842d67a7 | 5481 | int ret; |
5d261913 | 5482 | |
6ea29da1 | 5483 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
5484 | |
5485 | if (adj) { | |
5d261913 VF |
5486 | adj->ref_nr++; |
5487 | return 0; | |
5488 | } | |
5489 | ||
5490 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
5491 | if (!adj) | |
5492 | return -ENOMEM; | |
5493 | ||
5494 | adj->dev = adj_dev; | |
5495 | adj->master = master; | |
5d261913 | 5496 | adj->ref_nr = 1; |
402dae96 | 5497 | adj->private = private; |
5d261913 | 5498 | dev_hold(adj_dev); |
2f268f12 VF |
5499 | |
5500 | pr_debug("dev_hold for %s, because of link added from %s to %s\n", | |
5501 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 | 5502 | |
7ce64c79 | 5503 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 5504 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
5505 | if (ret) |
5506 | goto free_adj; | |
5507 | } | |
5508 | ||
7863c054 | 5509 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
5510 | if (master) { |
5511 | ret = sysfs_create_link(&(dev->dev.kobj), | |
5512 | &(adj_dev->dev.kobj), "master"); | |
5513 | if (ret) | |
5831d66e | 5514 | goto remove_symlinks; |
842d67a7 | 5515 | |
7863c054 | 5516 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 5517 | } else { |
7863c054 | 5518 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 5519 | } |
5d261913 VF |
5520 | |
5521 | return 0; | |
842d67a7 | 5522 | |
5831d66e | 5523 | remove_symlinks: |
7ce64c79 | 5524 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5525 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
5526 | free_adj: |
5527 | kfree(adj); | |
974daef7 | 5528 | dev_put(adj_dev); |
842d67a7 VF |
5529 | |
5530 | return ret; | |
5d261913 VF |
5531 | } |
5532 | ||
1d143d9f | 5533 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
5534 | struct net_device *adj_dev, | |
5535 | struct list_head *dev_list) | |
5d261913 VF |
5536 | { |
5537 | struct netdev_adjacent *adj; | |
5538 | ||
6ea29da1 | 5539 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 5540 | |
2f268f12 VF |
5541 | if (!adj) { |
5542 | pr_err("tried to remove device %s from %s\n", | |
5543 | dev->name, adj_dev->name); | |
5d261913 | 5544 | BUG(); |
2f268f12 | 5545 | } |
5d261913 VF |
5546 | |
5547 | if (adj->ref_nr > 1) { | |
2f268f12 VF |
5548 | pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, |
5549 | adj->ref_nr-1); | |
5d261913 VF |
5550 | adj->ref_nr--; |
5551 | return; | |
5552 | } | |
5553 | ||
842d67a7 VF |
5554 | if (adj->master) |
5555 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
5556 | ||
7ce64c79 | 5557 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5558 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 5559 | |
5d261913 | 5560 | list_del_rcu(&adj->list); |
2f268f12 VF |
5561 | pr_debug("dev_put for %s, because link removed from %s to %s\n", |
5562 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 VF |
5563 | dev_put(adj_dev); |
5564 | kfree_rcu(adj, rcu); | |
5565 | } | |
5566 | ||
1d143d9f | 5567 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
5568 | struct net_device *upper_dev, | |
5569 | struct list_head *up_list, | |
5570 | struct list_head *down_list, | |
5571 | void *private, bool master) | |
5d261913 VF |
5572 | { |
5573 | int ret; | |
5574 | ||
402dae96 VF |
5575 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, |
5576 | master); | |
5d261913 VF |
5577 | if (ret) |
5578 | return ret; | |
5579 | ||
402dae96 VF |
5580 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, |
5581 | false); | |
5d261913 | 5582 | if (ret) { |
2f268f12 | 5583 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5d261913 VF |
5584 | return ret; |
5585 | } | |
5586 | ||
5587 | return 0; | |
5588 | } | |
5589 | ||
1d143d9f | 5590 | static int __netdev_adjacent_dev_link(struct net_device *dev, |
5591 | struct net_device *upper_dev) | |
5d261913 | 5592 | { |
2f268f12 VF |
5593 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
5594 | &dev->all_adj_list.upper, | |
5595 | &upper_dev->all_adj_list.lower, | |
402dae96 | 5596 | NULL, false); |
5d261913 VF |
5597 | } |
5598 | ||
1d143d9f | 5599 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
5600 | struct net_device *upper_dev, | |
5601 | struct list_head *up_list, | |
5602 | struct list_head *down_list) | |
5d261913 | 5603 | { |
2f268f12 VF |
5604 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5605 | __netdev_adjacent_dev_remove(upper_dev, dev, down_list); | |
5d261913 VF |
5606 | } |
5607 | ||
1d143d9f | 5608 | static void __netdev_adjacent_dev_unlink(struct net_device *dev, |
5609 | struct net_device *upper_dev) | |
5d261913 | 5610 | { |
2f268f12 VF |
5611 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, |
5612 | &dev->all_adj_list.upper, | |
5613 | &upper_dev->all_adj_list.lower); | |
5614 | } | |
5615 | ||
1d143d9f | 5616 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
5617 | struct net_device *upper_dev, | |
5618 | void *private, bool master) | |
2f268f12 VF |
5619 | { |
5620 | int ret = __netdev_adjacent_dev_link(dev, upper_dev); | |
5621 | ||
5622 | if (ret) | |
5623 | return ret; | |
5624 | ||
5625 | ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, | |
5626 | &dev->adj_list.upper, | |
5627 | &upper_dev->adj_list.lower, | |
402dae96 | 5628 | private, master); |
2f268f12 VF |
5629 | if (ret) { |
5630 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5631 | return ret; | |
5632 | } | |
5633 | ||
5634 | return 0; | |
5d261913 VF |
5635 | } |
5636 | ||
1d143d9f | 5637 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
5638 | struct net_device *upper_dev) | |
2f268f12 VF |
5639 | { |
5640 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5641 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, | |
5642 | &dev->adj_list.upper, | |
5643 | &upper_dev->adj_list.lower); | |
5644 | } | |
5d261913 | 5645 | |
9ff162a8 | 5646 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 5647 | struct net_device *upper_dev, bool master, |
29bf24af | 5648 | void *upper_priv, void *upper_info) |
9ff162a8 | 5649 | { |
0e4ead9d | 5650 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 VF |
5651 | struct netdev_adjacent *i, *j, *to_i, *to_j; |
5652 | int ret = 0; | |
9ff162a8 JP |
5653 | |
5654 | ASSERT_RTNL(); | |
5655 | ||
5656 | if (dev == upper_dev) | |
5657 | return -EBUSY; | |
5658 | ||
5659 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
6ea29da1 | 5660 | if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper)) |
9ff162a8 JP |
5661 | return -EBUSY; |
5662 | ||
6ea29da1 | 5663 | if (__netdev_find_adj(upper_dev, &dev->adj_list.upper)) |
9ff162a8 JP |
5664 | return -EEXIST; |
5665 | ||
5666 | if (master && netdev_master_upper_dev_get(dev)) | |
5667 | return -EBUSY; | |
5668 | ||
0e4ead9d JP |
5669 | changeupper_info.upper_dev = upper_dev; |
5670 | changeupper_info.master = master; | |
5671 | changeupper_info.linking = true; | |
29bf24af | 5672 | changeupper_info.upper_info = upper_info; |
0e4ead9d | 5673 | |
573c7ba0 JP |
5674 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
5675 | &changeupper_info.info); | |
5676 | ret = notifier_to_errno(ret); | |
5677 | if (ret) | |
5678 | return ret; | |
5679 | ||
6dffb044 | 5680 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 5681 | master); |
5d261913 VF |
5682 | if (ret) |
5683 | return ret; | |
9ff162a8 | 5684 | |
5d261913 | 5685 | /* Now that we linked these devs, make all the upper_dev's |
2f268f12 | 5686 | * all_adj_list.upper visible to every dev's all_adj_list.lower an |
5d261913 VF |
5687 | * versa, and don't forget the devices itself. All of these |
5688 | * links are non-neighbours. | |
5689 | */ | |
2f268f12 VF |
5690 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5691 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5692 | pr_debug("Interlinking %s with %s, non-neighbour\n", | |
5693 | i->dev->name, j->dev->name); | |
5d261913 VF |
5694 | ret = __netdev_adjacent_dev_link(i->dev, j->dev); |
5695 | if (ret) | |
5696 | goto rollback_mesh; | |
5697 | } | |
5698 | } | |
5699 | ||
5700 | /* add dev to every upper_dev's upper device */ | |
2f268f12 VF |
5701 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5702 | pr_debug("linking %s's upper device %s with %s\n", | |
5703 | upper_dev->name, i->dev->name, dev->name); | |
5d261913 VF |
5704 | ret = __netdev_adjacent_dev_link(dev, i->dev); |
5705 | if (ret) | |
5706 | goto rollback_upper_mesh; | |
5707 | } | |
5708 | ||
5709 | /* add upper_dev to every dev's lower device */ | |
2f268f12 VF |
5710 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5711 | pr_debug("linking %s's lower device %s with %s\n", dev->name, | |
5712 | i->dev->name, upper_dev->name); | |
5d261913 VF |
5713 | ret = __netdev_adjacent_dev_link(i->dev, upper_dev); |
5714 | if (ret) | |
5715 | goto rollback_lower_mesh; | |
5716 | } | |
9ff162a8 | 5717 | |
b03804e7 IS |
5718 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
5719 | &changeupper_info.info); | |
5720 | ret = notifier_to_errno(ret); | |
5721 | if (ret) | |
5722 | goto rollback_lower_mesh; | |
5723 | ||
9ff162a8 | 5724 | return 0; |
5d261913 VF |
5725 | |
5726 | rollback_lower_mesh: | |
5727 | to_i = i; | |
2f268f12 | 5728 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5d261913 VF |
5729 | if (i == to_i) |
5730 | break; | |
5731 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); | |
5732 | } | |
5733 | ||
5734 | i = NULL; | |
5735 | ||
5736 | rollback_upper_mesh: | |
5737 | to_i = i; | |
2f268f12 | 5738 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5d261913 VF |
5739 | if (i == to_i) |
5740 | break; | |
5741 | __netdev_adjacent_dev_unlink(dev, i->dev); | |
5742 | } | |
5743 | ||
5744 | i = j = NULL; | |
5745 | ||
5746 | rollback_mesh: | |
5747 | to_i = i; | |
5748 | to_j = j; | |
2f268f12 VF |
5749 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5750 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5d261913 VF |
5751 | if (i == to_i && j == to_j) |
5752 | break; | |
5753 | __netdev_adjacent_dev_unlink(i->dev, j->dev); | |
5754 | } | |
5755 | if (i == to_i) | |
5756 | break; | |
5757 | } | |
5758 | ||
2f268f12 | 5759 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5760 | |
5761 | return ret; | |
9ff162a8 JP |
5762 | } |
5763 | ||
5764 | /** | |
5765 | * netdev_upper_dev_link - Add a link to the upper device | |
5766 | * @dev: device | |
5767 | * @upper_dev: new upper device | |
5768 | * | |
5769 | * Adds a link to device which is upper to this one. The caller must hold | |
5770 | * the RTNL lock. On a failure a negative errno code is returned. | |
5771 | * On success the reference counts are adjusted and the function | |
5772 | * returns zero. | |
5773 | */ | |
5774 | int netdev_upper_dev_link(struct net_device *dev, | |
5775 | struct net_device *upper_dev) | |
5776 | { | |
29bf24af | 5777 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL); |
9ff162a8 JP |
5778 | } |
5779 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
5780 | ||
5781 | /** | |
5782 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
5783 | * @dev: device | |
5784 | * @upper_dev: new upper device | |
6dffb044 | 5785 | * @upper_priv: upper device private |
29bf24af | 5786 | * @upper_info: upper info to be passed down via notifier |
9ff162a8 JP |
5787 | * |
5788 | * Adds a link to device which is upper to this one. In this case, only | |
5789 | * one master upper device can be linked, although other non-master devices | |
5790 | * might be linked as well. The caller must hold the RTNL lock. | |
5791 | * On a failure a negative errno code is returned. On success the reference | |
5792 | * counts are adjusted and the function returns zero. | |
5793 | */ | |
5794 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 5795 | struct net_device *upper_dev, |
29bf24af | 5796 | void *upper_priv, void *upper_info) |
9ff162a8 | 5797 | { |
29bf24af JP |
5798 | return __netdev_upper_dev_link(dev, upper_dev, true, |
5799 | upper_priv, upper_info); | |
9ff162a8 JP |
5800 | } |
5801 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
5802 | ||
5803 | /** | |
5804 | * netdev_upper_dev_unlink - Removes a link to upper device | |
5805 | * @dev: device | |
5806 | * @upper_dev: new upper device | |
5807 | * | |
5808 | * Removes a link to device which is upper to this one. The caller must hold | |
5809 | * the RTNL lock. | |
5810 | */ | |
5811 | void netdev_upper_dev_unlink(struct net_device *dev, | |
5812 | struct net_device *upper_dev) | |
5813 | { | |
0e4ead9d | 5814 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 | 5815 | struct netdev_adjacent *i, *j; |
9ff162a8 JP |
5816 | ASSERT_RTNL(); |
5817 | ||
0e4ead9d JP |
5818 | changeupper_info.upper_dev = upper_dev; |
5819 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; | |
5820 | changeupper_info.linking = false; | |
5821 | ||
573c7ba0 JP |
5822 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
5823 | &changeupper_info.info); | |
5824 | ||
2f268f12 | 5825 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5826 | |
5827 | /* Here is the tricky part. We must remove all dev's lower | |
5828 | * devices from all upper_dev's upper devices and vice | |
5829 | * versa, to maintain the graph relationship. | |
5830 | */ | |
2f268f12 VF |
5831 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5832 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) | |
5d261913 VF |
5833 | __netdev_adjacent_dev_unlink(i->dev, j->dev); |
5834 | ||
5835 | /* remove also the devices itself from lower/upper device | |
5836 | * list | |
5837 | */ | |
2f268f12 | 5838 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5d261913 VF |
5839 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); |
5840 | ||
2f268f12 | 5841 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) |
5d261913 VF |
5842 | __netdev_adjacent_dev_unlink(dev, i->dev); |
5843 | ||
0e4ead9d JP |
5844 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
5845 | &changeupper_info.info); | |
9ff162a8 JP |
5846 | } |
5847 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
5848 | ||
61bd3857 MS |
5849 | /** |
5850 | * netdev_bonding_info_change - Dispatch event about slave change | |
5851 | * @dev: device | |
4a26e453 | 5852 | * @bonding_info: info to dispatch |
61bd3857 MS |
5853 | * |
5854 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
5855 | * The caller must hold the RTNL lock. | |
5856 | */ | |
5857 | void netdev_bonding_info_change(struct net_device *dev, | |
5858 | struct netdev_bonding_info *bonding_info) | |
5859 | { | |
5860 | struct netdev_notifier_bonding_info info; | |
5861 | ||
5862 | memcpy(&info.bonding_info, bonding_info, | |
5863 | sizeof(struct netdev_bonding_info)); | |
5864 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, | |
5865 | &info.info); | |
5866 | } | |
5867 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
5868 | ||
2ce1ee17 | 5869 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
5870 | { |
5871 | struct netdev_adjacent *iter; | |
5872 | ||
5873 | struct net *net = dev_net(dev); | |
5874 | ||
5875 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5876 | if (!net_eq(net,dev_net(iter->dev))) | |
5877 | continue; | |
5878 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5879 | &iter->dev->adj_list.lower); | |
5880 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5881 | &dev->adj_list.upper); | |
5882 | } | |
5883 | ||
5884 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5885 | if (!net_eq(net,dev_net(iter->dev))) | |
5886 | continue; | |
5887 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5888 | &iter->dev->adj_list.upper); | |
5889 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5890 | &dev->adj_list.lower); | |
5891 | } | |
5892 | } | |
5893 | ||
2ce1ee17 | 5894 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
5895 | { |
5896 | struct netdev_adjacent *iter; | |
5897 | ||
5898 | struct net *net = dev_net(dev); | |
5899 | ||
5900 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
5901 | if (!net_eq(net,dev_net(iter->dev))) | |
5902 | continue; | |
5903 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5904 | &iter->dev->adj_list.lower); | |
5905 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5906 | &dev->adj_list.upper); | |
5907 | } | |
5908 | ||
5909 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
5910 | if (!net_eq(net,dev_net(iter->dev))) | |
5911 | continue; | |
5912 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5913 | &iter->dev->adj_list.upper); | |
5914 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5915 | &dev->adj_list.lower); | |
5916 | } | |
5917 | } | |
5918 | ||
5bb025fa | 5919 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 5920 | { |
5bb025fa | 5921 | struct netdev_adjacent *iter; |
402dae96 | 5922 | |
4c75431a AF |
5923 | struct net *net = dev_net(dev); |
5924 | ||
5bb025fa | 5925 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
4c75431a AF |
5926 | if (!net_eq(net,dev_net(iter->dev))) |
5927 | continue; | |
5bb025fa VF |
5928 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5929 | &iter->dev->adj_list.lower); | |
5930 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5931 | &iter->dev->adj_list.lower); | |
5932 | } | |
402dae96 | 5933 | |
5bb025fa | 5934 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
4c75431a AF |
5935 | if (!net_eq(net,dev_net(iter->dev))) |
5936 | continue; | |
5bb025fa VF |
5937 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
5938 | &iter->dev->adj_list.upper); | |
5939 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5940 | &iter->dev->adj_list.upper); | |
5941 | } | |
402dae96 | 5942 | } |
402dae96 VF |
5943 | |
5944 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
5945 | struct net_device *lower_dev) | |
5946 | { | |
5947 | struct netdev_adjacent *lower; | |
5948 | ||
5949 | if (!lower_dev) | |
5950 | return NULL; | |
6ea29da1 | 5951 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
5952 | if (!lower) |
5953 | return NULL; | |
5954 | ||
5955 | return lower->private; | |
5956 | } | |
5957 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
5958 | ||
4085ebe8 VY |
5959 | |
5960 | int dev_get_nest_level(struct net_device *dev, | |
b618aaa9 | 5961 | bool (*type_check)(const struct net_device *dev)) |
4085ebe8 VY |
5962 | { |
5963 | struct net_device *lower = NULL; | |
5964 | struct list_head *iter; | |
5965 | int max_nest = -1; | |
5966 | int nest; | |
5967 | ||
5968 | ASSERT_RTNL(); | |
5969 | ||
5970 | netdev_for_each_lower_dev(dev, lower, iter) { | |
5971 | nest = dev_get_nest_level(lower, type_check); | |
5972 | if (max_nest < nest) | |
5973 | max_nest = nest; | |
5974 | } | |
5975 | ||
5976 | if (type_check(dev)) | |
5977 | max_nest++; | |
5978 | ||
5979 | return max_nest; | |
5980 | } | |
5981 | EXPORT_SYMBOL(dev_get_nest_level); | |
5982 | ||
04d48266 JP |
5983 | /** |
5984 | * netdev_lower_change - Dispatch event about lower device state change | |
5985 | * @lower_dev: device | |
5986 | * @lower_state_info: state to dispatch | |
5987 | * | |
5988 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
5989 | * The caller must hold the RTNL lock. | |
5990 | */ | |
5991 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
5992 | void *lower_state_info) | |
5993 | { | |
5994 | struct netdev_notifier_changelowerstate_info changelowerstate_info; | |
5995 | ||
5996 | ASSERT_RTNL(); | |
5997 | changelowerstate_info.lower_state_info = lower_state_info; | |
5998 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev, | |
5999 | &changelowerstate_info.info); | |
6000 | } | |
6001 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
6002 | ||
b6c40d68 PM |
6003 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
6004 | { | |
d314774c SH |
6005 | const struct net_device_ops *ops = dev->netdev_ops; |
6006 | ||
d2615bf4 | 6007 | if (ops->ndo_change_rx_flags) |
d314774c | 6008 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
6009 | } |
6010 | ||
991fb3f7 | 6011 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6012 | { |
b536db93 | 6013 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
6014 | kuid_t uid; |
6015 | kgid_t gid; | |
1da177e4 | 6016 | |
24023451 PM |
6017 | ASSERT_RTNL(); |
6018 | ||
dad9b335 WC |
6019 | dev->flags |= IFF_PROMISC; |
6020 | dev->promiscuity += inc; | |
6021 | if (dev->promiscuity == 0) { | |
6022 | /* | |
6023 | * Avoid overflow. | |
6024 | * If inc causes overflow, untouch promisc and return error. | |
6025 | */ | |
6026 | if (inc < 0) | |
6027 | dev->flags &= ~IFF_PROMISC; | |
6028 | else { | |
6029 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
6030 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
6031 | dev->name); | |
dad9b335 WC |
6032 | return -EOVERFLOW; |
6033 | } | |
6034 | } | |
52609c0b | 6035 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
6036 | pr_info("device %s %s promiscuous mode\n", |
6037 | dev->name, | |
6038 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
6039 | if (audit_enabled) { |
6040 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
6041 | audit_log(current->audit_context, GFP_ATOMIC, |
6042 | AUDIT_ANOM_PROMISCUOUS, | |
6043 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
6044 | dev->name, (dev->flags & IFF_PROMISC), | |
6045 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 6046 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
6047 | from_kuid(&init_user_ns, uid), |
6048 | from_kgid(&init_user_ns, gid), | |
7759db82 | 6049 | audit_get_sessionid(current)); |
8192b0c4 | 6050 | } |
24023451 | 6051 | |
b6c40d68 | 6052 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 6053 | } |
991fb3f7 ND |
6054 | if (notify) |
6055 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 6056 | return 0; |
1da177e4 LT |
6057 | } |
6058 | ||
4417da66 PM |
6059 | /** |
6060 | * dev_set_promiscuity - update promiscuity count on a device | |
6061 | * @dev: device | |
6062 | * @inc: modifier | |
6063 | * | |
6064 | * Add or remove promiscuity from a device. While the count in the device | |
6065 | * remains above zero the interface remains promiscuous. Once it hits zero | |
6066 | * the device reverts back to normal filtering operation. A negative inc | |
6067 | * value is used to drop promiscuity on the device. | |
dad9b335 | 6068 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 6069 | */ |
dad9b335 | 6070 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 6071 | { |
b536db93 | 6072 | unsigned int old_flags = dev->flags; |
dad9b335 | 6073 | int err; |
4417da66 | 6074 | |
991fb3f7 | 6075 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 6076 | if (err < 0) |
dad9b335 | 6077 | return err; |
4417da66 PM |
6078 | if (dev->flags != old_flags) |
6079 | dev_set_rx_mode(dev); | |
dad9b335 | 6080 | return err; |
4417da66 | 6081 | } |
d1b19dff | 6082 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 6083 | |
991fb3f7 | 6084 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6085 | { |
991fb3f7 | 6086 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 6087 | |
24023451 PM |
6088 | ASSERT_RTNL(); |
6089 | ||
1da177e4 | 6090 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
6091 | dev->allmulti += inc; |
6092 | if (dev->allmulti == 0) { | |
6093 | /* | |
6094 | * Avoid overflow. | |
6095 | * If inc causes overflow, untouch allmulti and return error. | |
6096 | */ | |
6097 | if (inc < 0) | |
6098 | dev->flags &= ~IFF_ALLMULTI; | |
6099 | else { | |
6100 | dev->allmulti -= inc; | |
7b6cd1ce JP |
6101 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
6102 | dev->name); | |
dad9b335 WC |
6103 | return -EOVERFLOW; |
6104 | } | |
6105 | } | |
24023451 | 6106 | if (dev->flags ^ old_flags) { |
b6c40d68 | 6107 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 6108 | dev_set_rx_mode(dev); |
991fb3f7 ND |
6109 | if (notify) |
6110 | __dev_notify_flags(dev, old_flags, | |
6111 | dev->gflags ^ old_gflags); | |
24023451 | 6112 | } |
dad9b335 | 6113 | return 0; |
4417da66 | 6114 | } |
991fb3f7 ND |
6115 | |
6116 | /** | |
6117 | * dev_set_allmulti - update allmulti count on a device | |
6118 | * @dev: device | |
6119 | * @inc: modifier | |
6120 | * | |
6121 | * Add or remove reception of all multicast frames to a device. While the | |
6122 | * count in the device remains above zero the interface remains listening | |
6123 | * to all interfaces. Once it hits zero the device reverts back to normal | |
6124 | * filtering operation. A negative @inc value is used to drop the counter | |
6125 | * when releasing a resource needing all multicasts. | |
6126 | * Return 0 if successful or a negative errno code on error. | |
6127 | */ | |
6128 | ||
6129 | int dev_set_allmulti(struct net_device *dev, int inc) | |
6130 | { | |
6131 | return __dev_set_allmulti(dev, inc, true); | |
6132 | } | |
d1b19dff | 6133 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
6134 | |
6135 | /* | |
6136 | * Upload unicast and multicast address lists to device and | |
6137 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 6138 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
6139 | * are present. |
6140 | */ | |
6141 | void __dev_set_rx_mode(struct net_device *dev) | |
6142 | { | |
d314774c SH |
6143 | const struct net_device_ops *ops = dev->netdev_ops; |
6144 | ||
4417da66 PM |
6145 | /* dev_open will call this function so the list will stay sane. */ |
6146 | if (!(dev->flags&IFF_UP)) | |
6147 | return; | |
6148 | ||
6149 | if (!netif_device_present(dev)) | |
40b77c94 | 6150 | return; |
4417da66 | 6151 | |
01789349 | 6152 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
6153 | /* Unicast addresses changes may only happen under the rtnl, |
6154 | * therefore calling __dev_set_promiscuity here is safe. | |
6155 | */ | |
32e7bfc4 | 6156 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 6157 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 6158 | dev->uc_promisc = true; |
32e7bfc4 | 6159 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 6160 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 6161 | dev->uc_promisc = false; |
4417da66 | 6162 | } |
4417da66 | 6163 | } |
01789349 JP |
6164 | |
6165 | if (ops->ndo_set_rx_mode) | |
6166 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
6167 | } |
6168 | ||
6169 | void dev_set_rx_mode(struct net_device *dev) | |
6170 | { | |
b9e40857 | 6171 | netif_addr_lock_bh(dev); |
4417da66 | 6172 | __dev_set_rx_mode(dev); |
b9e40857 | 6173 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
6174 | } |
6175 | ||
f0db275a SH |
6176 | /** |
6177 | * dev_get_flags - get flags reported to userspace | |
6178 | * @dev: device | |
6179 | * | |
6180 | * Get the combination of flag bits exported through APIs to userspace. | |
6181 | */ | |
95c96174 | 6182 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 6183 | { |
95c96174 | 6184 | unsigned int flags; |
1da177e4 LT |
6185 | |
6186 | flags = (dev->flags & ~(IFF_PROMISC | | |
6187 | IFF_ALLMULTI | | |
b00055aa SR |
6188 | IFF_RUNNING | |
6189 | IFF_LOWER_UP | | |
6190 | IFF_DORMANT)) | | |
1da177e4 LT |
6191 | (dev->gflags & (IFF_PROMISC | |
6192 | IFF_ALLMULTI)); | |
6193 | ||
b00055aa SR |
6194 | if (netif_running(dev)) { |
6195 | if (netif_oper_up(dev)) | |
6196 | flags |= IFF_RUNNING; | |
6197 | if (netif_carrier_ok(dev)) | |
6198 | flags |= IFF_LOWER_UP; | |
6199 | if (netif_dormant(dev)) | |
6200 | flags |= IFF_DORMANT; | |
6201 | } | |
1da177e4 LT |
6202 | |
6203 | return flags; | |
6204 | } | |
d1b19dff | 6205 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 6206 | |
bd380811 | 6207 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 6208 | { |
b536db93 | 6209 | unsigned int old_flags = dev->flags; |
bd380811 | 6210 | int ret; |
1da177e4 | 6211 | |
24023451 PM |
6212 | ASSERT_RTNL(); |
6213 | ||
1da177e4 LT |
6214 | /* |
6215 | * Set the flags on our device. | |
6216 | */ | |
6217 | ||
6218 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
6219 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
6220 | IFF_AUTOMEDIA)) | | |
6221 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
6222 | IFF_ALLMULTI)); | |
6223 | ||
6224 | /* | |
6225 | * Load in the correct multicast list now the flags have changed. | |
6226 | */ | |
6227 | ||
b6c40d68 PM |
6228 | if ((old_flags ^ flags) & IFF_MULTICAST) |
6229 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 6230 | |
4417da66 | 6231 | dev_set_rx_mode(dev); |
1da177e4 LT |
6232 | |
6233 | /* | |
6234 | * Have we downed the interface. We handle IFF_UP ourselves | |
6235 | * according to user attempts to set it, rather than blindly | |
6236 | * setting it. | |
6237 | */ | |
6238 | ||
6239 | ret = 0; | |
d215d10f | 6240 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 6241 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 6242 | |
1da177e4 | 6243 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 6244 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 6245 | unsigned int old_flags = dev->flags; |
d1b19dff | 6246 | |
1da177e4 | 6247 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
6248 | |
6249 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
6250 | if (dev->flags != old_flags) | |
6251 | dev_set_rx_mode(dev); | |
1da177e4 LT |
6252 | } |
6253 | ||
6254 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
6255 | is important. Some (broken) drivers set IFF_PROMISC, when | |
6256 | IFF_ALLMULTI is requested not asking us and not reporting. | |
6257 | */ | |
6258 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
6259 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
6260 | ||
1da177e4 | 6261 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 6262 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
6263 | } |
6264 | ||
bd380811 PM |
6265 | return ret; |
6266 | } | |
6267 | ||
a528c219 ND |
6268 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
6269 | unsigned int gchanges) | |
bd380811 PM |
6270 | { |
6271 | unsigned int changes = dev->flags ^ old_flags; | |
6272 | ||
a528c219 | 6273 | if (gchanges) |
7f294054 | 6274 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 6275 | |
bd380811 PM |
6276 | if (changes & IFF_UP) { |
6277 | if (dev->flags & IFF_UP) | |
6278 | call_netdevice_notifiers(NETDEV_UP, dev); | |
6279 | else | |
6280 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
6281 | } | |
6282 | ||
6283 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
6284 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
6285 | struct netdev_notifier_change_info change_info; | |
6286 | ||
6287 | change_info.flags_changed = changes; | |
6288 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
6289 | &change_info.info); | |
6290 | } | |
bd380811 PM |
6291 | } |
6292 | ||
6293 | /** | |
6294 | * dev_change_flags - change device settings | |
6295 | * @dev: device | |
6296 | * @flags: device state flags | |
6297 | * | |
6298 | * Change settings on device based state flags. The flags are | |
6299 | * in the userspace exported format. | |
6300 | */ | |
b536db93 | 6301 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 6302 | { |
b536db93 | 6303 | int ret; |
991fb3f7 | 6304 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
6305 | |
6306 | ret = __dev_change_flags(dev, flags); | |
6307 | if (ret < 0) | |
6308 | return ret; | |
6309 | ||
991fb3f7 | 6310 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 6311 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
6312 | return ret; |
6313 | } | |
d1b19dff | 6314 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 6315 | |
2315dc91 VF |
6316 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
6317 | { | |
6318 | const struct net_device_ops *ops = dev->netdev_ops; | |
6319 | ||
6320 | if (ops->ndo_change_mtu) | |
6321 | return ops->ndo_change_mtu(dev, new_mtu); | |
6322 | ||
6323 | dev->mtu = new_mtu; | |
6324 | return 0; | |
6325 | } | |
6326 | ||
f0db275a SH |
6327 | /** |
6328 | * dev_set_mtu - Change maximum transfer unit | |
6329 | * @dev: device | |
6330 | * @new_mtu: new transfer unit | |
6331 | * | |
6332 | * Change the maximum transfer size of the network device. | |
6333 | */ | |
1da177e4 LT |
6334 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
6335 | { | |
2315dc91 | 6336 | int err, orig_mtu; |
1da177e4 LT |
6337 | |
6338 | if (new_mtu == dev->mtu) | |
6339 | return 0; | |
6340 | ||
6341 | /* MTU must be positive. */ | |
6342 | if (new_mtu < 0) | |
6343 | return -EINVAL; | |
6344 | ||
6345 | if (!netif_device_present(dev)) | |
6346 | return -ENODEV; | |
6347 | ||
1d486bfb VF |
6348 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
6349 | err = notifier_to_errno(err); | |
6350 | if (err) | |
6351 | return err; | |
d314774c | 6352 | |
2315dc91 VF |
6353 | orig_mtu = dev->mtu; |
6354 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 6355 | |
2315dc91 VF |
6356 | if (!err) { |
6357 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6358 | err = notifier_to_errno(err); | |
6359 | if (err) { | |
6360 | /* setting mtu back and notifying everyone again, | |
6361 | * so that they have a chance to revert changes. | |
6362 | */ | |
6363 | __dev_set_mtu(dev, orig_mtu); | |
6364 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6365 | } | |
6366 | } | |
1da177e4 LT |
6367 | return err; |
6368 | } | |
d1b19dff | 6369 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 6370 | |
cbda10fa VD |
6371 | /** |
6372 | * dev_set_group - Change group this device belongs to | |
6373 | * @dev: device | |
6374 | * @new_group: group this device should belong to | |
6375 | */ | |
6376 | void dev_set_group(struct net_device *dev, int new_group) | |
6377 | { | |
6378 | dev->group = new_group; | |
6379 | } | |
6380 | EXPORT_SYMBOL(dev_set_group); | |
6381 | ||
f0db275a SH |
6382 | /** |
6383 | * dev_set_mac_address - Change Media Access Control Address | |
6384 | * @dev: device | |
6385 | * @sa: new address | |
6386 | * | |
6387 | * Change the hardware (MAC) address of the device | |
6388 | */ | |
1da177e4 LT |
6389 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
6390 | { | |
d314774c | 6391 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
6392 | int err; |
6393 | ||
d314774c | 6394 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
6395 | return -EOPNOTSUPP; |
6396 | if (sa->sa_family != dev->type) | |
6397 | return -EINVAL; | |
6398 | if (!netif_device_present(dev)) | |
6399 | return -ENODEV; | |
d314774c | 6400 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
6401 | if (err) |
6402 | return err; | |
fbdeca2d | 6403 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 6404 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 6405 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 6406 | return 0; |
1da177e4 | 6407 | } |
d1b19dff | 6408 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 6409 | |
4bf84c35 JP |
6410 | /** |
6411 | * dev_change_carrier - Change device carrier | |
6412 | * @dev: device | |
691b3b7e | 6413 | * @new_carrier: new value |
4bf84c35 JP |
6414 | * |
6415 | * Change device carrier | |
6416 | */ | |
6417 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
6418 | { | |
6419 | const struct net_device_ops *ops = dev->netdev_ops; | |
6420 | ||
6421 | if (!ops->ndo_change_carrier) | |
6422 | return -EOPNOTSUPP; | |
6423 | if (!netif_device_present(dev)) | |
6424 | return -ENODEV; | |
6425 | return ops->ndo_change_carrier(dev, new_carrier); | |
6426 | } | |
6427 | EXPORT_SYMBOL(dev_change_carrier); | |
6428 | ||
66b52b0d JP |
6429 | /** |
6430 | * dev_get_phys_port_id - Get device physical port ID | |
6431 | * @dev: device | |
6432 | * @ppid: port ID | |
6433 | * | |
6434 | * Get device physical port ID | |
6435 | */ | |
6436 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 6437 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
6438 | { |
6439 | const struct net_device_ops *ops = dev->netdev_ops; | |
6440 | ||
6441 | if (!ops->ndo_get_phys_port_id) | |
6442 | return -EOPNOTSUPP; | |
6443 | return ops->ndo_get_phys_port_id(dev, ppid); | |
6444 | } | |
6445 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
6446 | ||
db24a904 DA |
6447 | /** |
6448 | * dev_get_phys_port_name - Get device physical port name | |
6449 | * @dev: device | |
6450 | * @name: port name | |
ed49e650 | 6451 | * @len: limit of bytes to copy to name |
db24a904 DA |
6452 | * |
6453 | * Get device physical port name | |
6454 | */ | |
6455 | int dev_get_phys_port_name(struct net_device *dev, | |
6456 | char *name, size_t len) | |
6457 | { | |
6458 | const struct net_device_ops *ops = dev->netdev_ops; | |
6459 | ||
6460 | if (!ops->ndo_get_phys_port_name) | |
6461 | return -EOPNOTSUPP; | |
6462 | return ops->ndo_get_phys_port_name(dev, name, len); | |
6463 | } | |
6464 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
6465 | ||
d746d707 AK |
6466 | /** |
6467 | * dev_change_proto_down - update protocol port state information | |
6468 | * @dev: device | |
6469 | * @proto_down: new value | |
6470 | * | |
6471 | * This info can be used by switch drivers to set the phys state of the | |
6472 | * port. | |
6473 | */ | |
6474 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
6475 | { | |
6476 | const struct net_device_ops *ops = dev->netdev_ops; | |
6477 | ||
6478 | if (!ops->ndo_change_proto_down) | |
6479 | return -EOPNOTSUPP; | |
6480 | if (!netif_device_present(dev)) | |
6481 | return -ENODEV; | |
6482 | return ops->ndo_change_proto_down(dev, proto_down); | |
6483 | } | |
6484 | EXPORT_SYMBOL(dev_change_proto_down); | |
6485 | ||
1da177e4 LT |
6486 | /** |
6487 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 6488 | * @net: the applicable net namespace |
1da177e4 LT |
6489 | * |
6490 | * Returns a suitable unique value for a new device interface | |
6491 | * number. The caller must hold the rtnl semaphore or the | |
6492 | * dev_base_lock to be sure it remains unique. | |
6493 | */ | |
881d966b | 6494 | static int dev_new_index(struct net *net) |
1da177e4 | 6495 | { |
aa79e66e | 6496 | int ifindex = net->ifindex; |
1da177e4 LT |
6497 | for (;;) { |
6498 | if (++ifindex <= 0) | |
6499 | ifindex = 1; | |
881d966b | 6500 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 6501 | return net->ifindex = ifindex; |
1da177e4 LT |
6502 | } |
6503 | } | |
6504 | ||
1da177e4 | 6505 | /* Delayed registration/unregisteration */ |
3b5b34fd | 6506 | static LIST_HEAD(net_todo_list); |
200b916f | 6507 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 6508 | |
6f05f629 | 6509 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 6510 | { |
1da177e4 | 6511 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 6512 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
6513 | } |
6514 | ||
9b5e383c | 6515 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 6516 | { |
e93737b0 | 6517 | struct net_device *dev, *tmp; |
5cde2829 | 6518 | LIST_HEAD(close_head); |
9b5e383c | 6519 | |
93ee31f1 DL |
6520 | BUG_ON(dev_boot_phase); |
6521 | ASSERT_RTNL(); | |
6522 | ||
e93737b0 | 6523 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 6524 | /* Some devices call without registering |
e93737b0 KK |
6525 | * for initialization unwind. Remove those |
6526 | * devices and proceed with the remaining. | |
9b5e383c ED |
6527 | */ |
6528 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
6529 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
6530 | dev->name, dev); | |
93ee31f1 | 6531 | |
9b5e383c | 6532 | WARN_ON(1); |
e93737b0 KK |
6533 | list_del(&dev->unreg_list); |
6534 | continue; | |
9b5e383c | 6535 | } |
449f4544 | 6536 | dev->dismantle = true; |
9b5e383c | 6537 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 6538 | } |
93ee31f1 | 6539 | |
44345724 | 6540 | /* If device is running, close it first. */ |
5cde2829 EB |
6541 | list_for_each_entry(dev, head, unreg_list) |
6542 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 6543 | dev_close_many(&close_head, true); |
93ee31f1 | 6544 | |
44345724 | 6545 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
6546 | /* And unlink it from device chain. */ |
6547 | unlist_netdevice(dev); | |
93ee31f1 | 6548 | |
9b5e383c | 6549 | dev->reg_state = NETREG_UNREGISTERING; |
e9e4dd32 | 6550 | on_each_cpu(flush_backlog, dev, 1); |
9b5e383c | 6551 | } |
93ee31f1 DL |
6552 | |
6553 | synchronize_net(); | |
6554 | ||
9b5e383c | 6555 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
6556 | struct sk_buff *skb = NULL; |
6557 | ||
9b5e383c ED |
6558 | /* Shutdown queueing discipline. */ |
6559 | dev_shutdown(dev); | |
93ee31f1 DL |
6560 | |
6561 | ||
9b5e383c ED |
6562 | /* Notify protocols, that we are about to destroy |
6563 | this device. They should clean all the things. | |
6564 | */ | |
6565 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 6566 | |
395eea6c MB |
6567 | if (!dev->rtnl_link_ops || |
6568 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
6569 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, | |
6570 | GFP_KERNEL); | |
6571 | ||
9b5e383c ED |
6572 | /* |
6573 | * Flush the unicast and multicast chains | |
6574 | */ | |
a748ee24 | 6575 | dev_uc_flush(dev); |
22bedad3 | 6576 | dev_mc_flush(dev); |
93ee31f1 | 6577 | |
9b5e383c ED |
6578 | if (dev->netdev_ops->ndo_uninit) |
6579 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 6580 | |
395eea6c MB |
6581 | if (skb) |
6582 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 6583 | |
9ff162a8 JP |
6584 | /* Notifier chain MUST detach us all upper devices. */ |
6585 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 6586 | |
9b5e383c ED |
6587 | /* Remove entries from kobject tree */ |
6588 | netdev_unregister_kobject(dev); | |
024e9679 AD |
6589 | #ifdef CONFIG_XPS |
6590 | /* Remove XPS queueing entries */ | |
6591 | netif_reset_xps_queues_gt(dev, 0); | |
6592 | #endif | |
9b5e383c | 6593 | } |
93ee31f1 | 6594 | |
850a545b | 6595 | synchronize_net(); |
395264d5 | 6596 | |
a5ee1551 | 6597 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
6598 | dev_put(dev); |
6599 | } | |
6600 | ||
6601 | static void rollback_registered(struct net_device *dev) | |
6602 | { | |
6603 | LIST_HEAD(single); | |
6604 | ||
6605 | list_add(&dev->unreg_list, &single); | |
6606 | rollback_registered_many(&single); | |
ceaaec98 | 6607 | list_del(&single); |
93ee31f1 DL |
6608 | } |
6609 | ||
fd867d51 JW |
6610 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
6611 | struct net_device *upper, netdev_features_t features) | |
6612 | { | |
6613 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
6614 | netdev_features_t feature; | |
5ba3f7d6 | 6615 | int feature_bit; |
fd867d51 | 6616 | |
5ba3f7d6 JW |
6617 | for_each_netdev_feature(&upper_disables, feature_bit) { |
6618 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
6619 | if (!(upper->wanted_features & feature) |
6620 | && (features & feature)) { | |
6621 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
6622 | &feature, upper->name); | |
6623 | features &= ~feature; | |
6624 | } | |
6625 | } | |
6626 | ||
6627 | return features; | |
6628 | } | |
6629 | ||
6630 | static void netdev_sync_lower_features(struct net_device *upper, | |
6631 | struct net_device *lower, netdev_features_t features) | |
6632 | { | |
6633 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
6634 | netdev_features_t feature; | |
5ba3f7d6 | 6635 | int feature_bit; |
fd867d51 | 6636 | |
5ba3f7d6 JW |
6637 | for_each_netdev_feature(&upper_disables, feature_bit) { |
6638 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
6639 | if (!(features & feature) && (lower->features & feature)) { |
6640 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
6641 | &feature, lower->name); | |
6642 | lower->wanted_features &= ~feature; | |
6643 | netdev_update_features(lower); | |
6644 | ||
6645 | if (unlikely(lower->features & feature)) | |
6646 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
6647 | &feature, lower->name); | |
6648 | } | |
6649 | } | |
6650 | } | |
6651 | ||
c8f44aff MM |
6652 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
6653 | netdev_features_t features) | |
b63365a2 | 6654 | { |
57422dc5 MM |
6655 | /* Fix illegal checksum combinations */ |
6656 | if ((features & NETIF_F_HW_CSUM) && | |
6657 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6658 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
6659 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
6660 | } | |
6661 | ||
b63365a2 | 6662 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 6663 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 6664 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 6665 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
6666 | } |
6667 | ||
ec5f0615 PS |
6668 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
6669 | !(features & NETIF_F_IP_CSUM)) { | |
6670 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
6671 | features &= ~NETIF_F_TSO; | |
6672 | features &= ~NETIF_F_TSO_ECN; | |
6673 | } | |
6674 | ||
6675 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
6676 | !(features & NETIF_F_IPV6_CSUM)) { | |
6677 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
6678 | features &= ~NETIF_F_TSO6; | |
6679 | } | |
6680 | ||
31d8b9e0 BH |
6681 | /* TSO ECN requires that TSO is present as well. */ |
6682 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
6683 | features &= ~NETIF_F_TSO_ECN; | |
6684 | ||
212b573f MM |
6685 | /* Software GSO depends on SG. */ |
6686 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 6687 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
6688 | features &= ~NETIF_F_GSO; |
6689 | } | |
6690 | ||
acd1130e | 6691 | /* UFO needs SG and checksumming */ |
b63365a2 | 6692 | if (features & NETIF_F_UFO) { |
79032644 | 6693 | /* maybe split UFO into V4 and V6? */ |
c8cd0989 TH |
6694 | if (!(features & NETIF_F_HW_CSUM) && |
6695 | ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) != | |
6696 | (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6697 | netdev_dbg(dev, |
acd1130e | 6698 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
6699 | features &= ~NETIF_F_UFO; |
6700 | } | |
6701 | ||
6702 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 6703 | netdev_dbg(dev, |
acd1130e | 6704 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
6705 | features &= ~NETIF_F_UFO; |
6706 | } | |
6707 | } | |
6708 | ||
d0290214 JP |
6709 | #ifdef CONFIG_NET_RX_BUSY_POLL |
6710 | if (dev->netdev_ops->ndo_busy_poll) | |
6711 | features |= NETIF_F_BUSY_POLL; | |
6712 | else | |
6713 | #endif | |
6714 | features &= ~NETIF_F_BUSY_POLL; | |
6715 | ||
b63365a2 HX |
6716 | return features; |
6717 | } | |
b63365a2 | 6718 | |
6cb6a27c | 6719 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 6720 | { |
fd867d51 | 6721 | struct net_device *upper, *lower; |
c8f44aff | 6722 | netdev_features_t features; |
fd867d51 | 6723 | struct list_head *iter; |
e7868a85 | 6724 | int err = -1; |
5455c699 | 6725 | |
87267485 MM |
6726 | ASSERT_RTNL(); |
6727 | ||
5455c699 MM |
6728 | features = netdev_get_wanted_features(dev); |
6729 | ||
6730 | if (dev->netdev_ops->ndo_fix_features) | |
6731 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
6732 | ||
6733 | /* driver might be less strict about feature dependencies */ | |
6734 | features = netdev_fix_features(dev, features); | |
6735 | ||
fd867d51 JW |
6736 | /* some features can't be enabled if they're off an an upper device */ |
6737 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
6738 | features = netdev_sync_upper_features(dev, upper, features); | |
6739 | ||
5455c699 | 6740 | if (dev->features == features) |
e7868a85 | 6741 | goto sync_lower; |
5455c699 | 6742 | |
c8f44aff MM |
6743 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
6744 | &dev->features, &features); | |
5455c699 MM |
6745 | |
6746 | if (dev->netdev_ops->ndo_set_features) | |
6747 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
6748 | else |
6749 | err = 0; | |
5455c699 | 6750 | |
6cb6a27c | 6751 | if (unlikely(err < 0)) { |
5455c699 | 6752 | netdev_err(dev, |
c8f44aff MM |
6753 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
6754 | err, &features, &dev->features); | |
17b85d29 NA |
6755 | /* return non-0 since some features might have changed and |
6756 | * it's better to fire a spurious notification than miss it | |
6757 | */ | |
6758 | return -1; | |
6cb6a27c MM |
6759 | } |
6760 | ||
e7868a85 | 6761 | sync_lower: |
fd867d51 JW |
6762 | /* some features must be disabled on lower devices when disabled |
6763 | * on an upper device (think: bonding master or bridge) | |
6764 | */ | |
6765 | netdev_for_each_lower_dev(dev, lower, iter) | |
6766 | netdev_sync_lower_features(dev, lower, features); | |
6767 | ||
6cb6a27c MM |
6768 | if (!err) |
6769 | dev->features = features; | |
6770 | ||
e7868a85 | 6771 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
6772 | } |
6773 | ||
afe12cc8 MM |
6774 | /** |
6775 | * netdev_update_features - recalculate device features | |
6776 | * @dev: the device to check | |
6777 | * | |
6778 | * Recalculate dev->features set and send notifications if it | |
6779 | * has changed. Should be called after driver or hardware dependent | |
6780 | * conditions might have changed that influence the features. | |
6781 | */ | |
6cb6a27c MM |
6782 | void netdev_update_features(struct net_device *dev) |
6783 | { | |
6784 | if (__netdev_update_features(dev)) | |
6785 | netdev_features_change(dev); | |
5455c699 MM |
6786 | } |
6787 | EXPORT_SYMBOL(netdev_update_features); | |
6788 | ||
afe12cc8 MM |
6789 | /** |
6790 | * netdev_change_features - recalculate device features | |
6791 | * @dev: the device to check | |
6792 | * | |
6793 | * Recalculate dev->features set and send notifications even | |
6794 | * if they have not changed. Should be called instead of | |
6795 | * netdev_update_features() if also dev->vlan_features might | |
6796 | * have changed to allow the changes to be propagated to stacked | |
6797 | * VLAN devices. | |
6798 | */ | |
6799 | void netdev_change_features(struct net_device *dev) | |
6800 | { | |
6801 | __netdev_update_features(dev); | |
6802 | netdev_features_change(dev); | |
6803 | } | |
6804 | EXPORT_SYMBOL(netdev_change_features); | |
6805 | ||
fc4a7489 PM |
6806 | /** |
6807 | * netif_stacked_transfer_operstate - transfer operstate | |
6808 | * @rootdev: the root or lower level device to transfer state from | |
6809 | * @dev: the device to transfer operstate to | |
6810 | * | |
6811 | * Transfer operational state from root to device. This is normally | |
6812 | * called when a stacking relationship exists between the root | |
6813 | * device and the device(a leaf device). | |
6814 | */ | |
6815 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
6816 | struct net_device *dev) | |
6817 | { | |
6818 | if (rootdev->operstate == IF_OPER_DORMANT) | |
6819 | netif_dormant_on(dev); | |
6820 | else | |
6821 | netif_dormant_off(dev); | |
6822 | ||
6823 | if (netif_carrier_ok(rootdev)) { | |
6824 | if (!netif_carrier_ok(dev)) | |
6825 | netif_carrier_on(dev); | |
6826 | } else { | |
6827 | if (netif_carrier_ok(dev)) | |
6828 | netif_carrier_off(dev); | |
6829 | } | |
6830 | } | |
6831 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
6832 | ||
a953be53 | 6833 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
6834 | static int netif_alloc_rx_queues(struct net_device *dev) |
6835 | { | |
1b4bf461 | 6836 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 6837 | struct netdev_rx_queue *rx; |
10595902 | 6838 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 6839 | |
bd25fa7b | 6840 | BUG_ON(count < 1); |
1b4bf461 | 6841 | |
10595902 PG |
6842 | rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6843 | if (!rx) { | |
6844 | rx = vzalloc(sz); | |
6845 | if (!rx) | |
6846 | return -ENOMEM; | |
6847 | } | |
bd25fa7b TH |
6848 | dev->_rx = rx; |
6849 | ||
bd25fa7b | 6850 | for (i = 0; i < count; i++) |
fe822240 | 6851 | rx[i].dev = dev; |
1b4bf461 ED |
6852 | return 0; |
6853 | } | |
bf264145 | 6854 | #endif |
1b4bf461 | 6855 | |
aa942104 CG |
6856 | static void netdev_init_one_queue(struct net_device *dev, |
6857 | struct netdev_queue *queue, void *_unused) | |
6858 | { | |
6859 | /* Initialize queue lock */ | |
6860 | spin_lock_init(&queue->_xmit_lock); | |
6861 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
6862 | queue->xmit_lock_owner = -1; | |
b236da69 | 6863 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 6864 | queue->dev = dev; |
114cf580 TH |
6865 | #ifdef CONFIG_BQL |
6866 | dql_init(&queue->dql, HZ); | |
6867 | #endif | |
aa942104 CG |
6868 | } |
6869 | ||
60877a32 ED |
6870 | static void netif_free_tx_queues(struct net_device *dev) |
6871 | { | |
4cb28970 | 6872 | kvfree(dev->_tx); |
60877a32 ED |
6873 | } |
6874 | ||
e6484930 TH |
6875 | static int netif_alloc_netdev_queues(struct net_device *dev) |
6876 | { | |
6877 | unsigned int count = dev->num_tx_queues; | |
6878 | struct netdev_queue *tx; | |
60877a32 | 6879 | size_t sz = count * sizeof(*tx); |
e6484930 | 6880 | |
d339727c ED |
6881 | if (count < 1 || count > 0xffff) |
6882 | return -EINVAL; | |
62b5942a | 6883 | |
60877a32 ED |
6884 | tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
6885 | if (!tx) { | |
6886 | tx = vzalloc(sz); | |
6887 | if (!tx) | |
6888 | return -ENOMEM; | |
6889 | } | |
e6484930 | 6890 | dev->_tx = tx; |
1d24eb48 | 6891 | |
e6484930 TH |
6892 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
6893 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
6894 | |
6895 | return 0; | |
e6484930 TH |
6896 | } |
6897 | ||
a2029240 DV |
6898 | void netif_tx_stop_all_queues(struct net_device *dev) |
6899 | { | |
6900 | unsigned int i; | |
6901 | ||
6902 | for (i = 0; i < dev->num_tx_queues; i++) { | |
6903 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
6904 | netif_tx_stop_queue(txq); | |
6905 | } | |
6906 | } | |
6907 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
6908 | ||
1da177e4 LT |
6909 | /** |
6910 | * register_netdevice - register a network device | |
6911 | * @dev: device to register | |
6912 | * | |
6913 | * Take a completed network device structure and add it to the kernel | |
6914 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
6915 | * chain. 0 is returned on success. A negative errno code is returned | |
6916 | * on a failure to set up the device, or if the name is a duplicate. | |
6917 | * | |
6918 | * Callers must hold the rtnl semaphore. You may want | |
6919 | * register_netdev() instead of this. | |
6920 | * | |
6921 | * BUGS: | |
6922 | * The locking appears insufficient to guarantee two parallel registers | |
6923 | * will not get the same name. | |
6924 | */ | |
6925 | ||
6926 | int register_netdevice(struct net_device *dev) | |
6927 | { | |
1da177e4 | 6928 | int ret; |
d314774c | 6929 | struct net *net = dev_net(dev); |
1da177e4 LT |
6930 | |
6931 | BUG_ON(dev_boot_phase); | |
6932 | ASSERT_RTNL(); | |
6933 | ||
b17a7c17 SH |
6934 | might_sleep(); |
6935 | ||
1da177e4 LT |
6936 | /* When net_device's are persistent, this will be fatal. */ |
6937 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 6938 | BUG_ON(!net); |
1da177e4 | 6939 | |
f1f28aa3 | 6940 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 6941 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 6942 | |
828de4f6 | 6943 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
6944 | if (ret < 0) |
6945 | goto out; | |
6946 | ||
1da177e4 | 6947 | /* Init, if this function is available */ |
d314774c SH |
6948 | if (dev->netdev_ops->ndo_init) { |
6949 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
6950 | if (ret) { |
6951 | if (ret > 0) | |
6952 | ret = -EIO; | |
90833aa4 | 6953 | goto out; |
1da177e4 LT |
6954 | } |
6955 | } | |
4ec93edb | 6956 | |
f646968f PM |
6957 | if (((dev->hw_features | dev->features) & |
6958 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
6959 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
6960 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
6961 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
6962 | ret = -EINVAL; | |
6963 | goto err_uninit; | |
6964 | } | |
6965 | ||
9c7dafbf PE |
6966 | ret = -EBUSY; |
6967 | if (!dev->ifindex) | |
6968 | dev->ifindex = dev_new_index(net); | |
6969 | else if (__dev_get_by_index(net, dev->ifindex)) | |
6970 | goto err_uninit; | |
6971 | ||
5455c699 MM |
6972 | /* Transfer changeable features to wanted_features and enable |
6973 | * software offloads (GSO and GRO). | |
6974 | */ | |
6975 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
6976 | dev->features |= NETIF_F_SOFT_FEATURES; |
6977 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 6978 | |
34324dc2 MM |
6979 | if (!(dev->flags & IFF_LOOPBACK)) { |
6980 | dev->hw_features |= NETIF_F_NOCACHE_COPY; | |
c6e1a0d1 TH |
6981 | } |
6982 | ||
1180e7d6 | 6983 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 6984 | */ |
1180e7d6 | 6985 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 6986 | |
ee579677 PS |
6987 | /* Make NETIF_F_SG inheritable to tunnel devices. |
6988 | */ | |
6989 | dev->hw_enc_features |= NETIF_F_SG; | |
6990 | ||
0d89d203 SH |
6991 | /* Make NETIF_F_SG inheritable to MPLS. |
6992 | */ | |
6993 | dev->mpls_features |= NETIF_F_SG; | |
6994 | ||
7ffbe3fd JB |
6995 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
6996 | ret = notifier_to_errno(ret); | |
6997 | if (ret) | |
6998 | goto err_uninit; | |
6999 | ||
8b41d188 | 7000 | ret = netdev_register_kobject(dev); |
b17a7c17 | 7001 | if (ret) |
7ce1b0ed | 7002 | goto err_uninit; |
b17a7c17 SH |
7003 | dev->reg_state = NETREG_REGISTERED; |
7004 | ||
6cb6a27c | 7005 | __netdev_update_features(dev); |
8e9b59b2 | 7006 | |
1da177e4 LT |
7007 | /* |
7008 | * Default initial state at registry is that the | |
7009 | * device is present. | |
7010 | */ | |
7011 | ||
7012 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7013 | ||
8f4cccbb BH |
7014 | linkwatch_init_dev(dev); |
7015 | ||
1da177e4 | 7016 | dev_init_scheduler(dev); |
1da177e4 | 7017 | dev_hold(dev); |
ce286d32 | 7018 | list_netdevice(dev); |
7bf23575 | 7019 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 7020 | |
948b337e JP |
7021 | /* If the device has permanent device address, driver should |
7022 | * set dev_addr and also addr_assign_type should be set to | |
7023 | * NET_ADDR_PERM (default value). | |
7024 | */ | |
7025 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
7026 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
7027 | ||
1da177e4 | 7028 | /* Notify protocols, that a new device appeared. */ |
056925ab | 7029 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 7030 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
7031 | if (ret) { |
7032 | rollback_registered(dev); | |
7033 | dev->reg_state = NETREG_UNREGISTERED; | |
7034 | } | |
d90a909e EB |
7035 | /* |
7036 | * Prevent userspace races by waiting until the network | |
7037 | * device is fully setup before sending notifications. | |
7038 | */ | |
a2835763 PM |
7039 | if (!dev->rtnl_link_ops || |
7040 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 7041 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
7042 | |
7043 | out: | |
7044 | return ret; | |
7ce1b0ed HX |
7045 | |
7046 | err_uninit: | |
d314774c SH |
7047 | if (dev->netdev_ops->ndo_uninit) |
7048 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 7049 | goto out; |
1da177e4 | 7050 | } |
d1b19dff | 7051 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 7052 | |
937f1ba5 BH |
7053 | /** |
7054 | * init_dummy_netdev - init a dummy network device for NAPI | |
7055 | * @dev: device to init | |
7056 | * | |
7057 | * This takes a network device structure and initialize the minimum | |
7058 | * amount of fields so it can be used to schedule NAPI polls without | |
7059 | * registering a full blown interface. This is to be used by drivers | |
7060 | * that need to tie several hardware interfaces to a single NAPI | |
7061 | * poll scheduler due to HW limitations. | |
7062 | */ | |
7063 | int init_dummy_netdev(struct net_device *dev) | |
7064 | { | |
7065 | /* Clear everything. Note we don't initialize spinlocks | |
7066 | * are they aren't supposed to be taken by any of the | |
7067 | * NAPI code and this dummy netdev is supposed to be | |
7068 | * only ever used for NAPI polls | |
7069 | */ | |
7070 | memset(dev, 0, sizeof(struct net_device)); | |
7071 | ||
7072 | /* make sure we BUG if trying to hit standard | |
7073 | * register/unregister code path | |
7074 | */ | |
7075 | dev->reg_state = NETREG_DUMMY; | |
7076 | ||
937f1ba5 BH |
7077 | /* NAPI wants this */ |
7078 | INIT_LIST_HEAD(&dev->napi_list); | |
7079 | ||
7080 | /* a dummy interface is started by default */ | |
7081 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7082 | set_bit(__LINK_STATE_START, &dev->state); | |
7083 | ||
29b4433d ED |
7084 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
7085 | * because users of this 'device' dont need to change | |
7086 | * its refcount. | |
7087 | */ | |
7088 | ||
937f1ba5 BH |
7089 | return 0; |
7090 | } | |
7091 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
7092 | ||
7093 | ||
1da177e4 LT |
7094 | /** |
7095 | * register_netdev - register a network device | |
7096 | * @dev: device to register | |
7097 | * | |
7098 | * Take a completed network device structure and add it to the kernel | |
7099 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7100 | * chain. 0 is returned on success. A negative errno code is returned | |
7101 | * on a failure to set up the device, or if the name is a duplicate. | |
7102 | * | |
38b4da38 | 7103 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
7104 | * and expands the device name if you passed a format string to |
7105 | * alloc_netdev. | |
7106 | */ | |
7107 | int register_netdev(struct net_device *dev) | |
7108 | { | |
7109 | int err; | |
7110 | ||
7111 | rtnl_lock(); | |
1da177e4 | 7112 | err = register_netdevice(dev); |
1da177e4 LT |
7113 | rtnl_unlock(); |
7114 | return err; | |
7115 | } | |
7116 | EXPORT_SYMBOL(register_netdev); | |
7117 | ||
29b4433d ED |
7118 | int netdev_refcnt_read(const struct net_device *dev) |
7119 | { | |
7120 | int i, refcnt = 0; | |
7121 | ||
7122 | for_each_possible_cpu(i) | |
7123 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
7124 | return refcnt; | |
7125 | } | |
7126 | EXPORT_SYMBOL(netdev_refcnt_read); | |
7127 | ||
2c53040f | 7128 | /** |
1da177e4 | 7129 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 7130 | * @dev: target net_device |
1da177e4 LT |
7131 | * |
7132 | * This is called when unregistering network devices. | |
7133 | * | |
7134 | * Any protocol or device that holds a reference should register | |
7135 | * for netdevice notification, and cleanup and put back the | |
7136 | * reference if they receive an UNREGISTER event. | |
7137 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 7138 | * call dev_put. |
1da177e4 LT |
7139 | */ |
7140 | static void netdev_wait_allrefs(struct net_device *dev) | |
7141 | { | |
7142 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 7143 | int refcnt; |
1da177e4 | 7144 | |
e014debe ED |
7145 | linkwatch_forget_dev(dev); |
7146 | ||
1da177e4 | 7147 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
7148 | refcnt = netdev_refcnt_read(dev); |
7149 | ||
7150 | while (refcnt != 0) { | |
1da177e4 | 7151 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 7152 | rtnl_lock(); |
1da177e4 LT |
7153 | |
7154 | /* Rebroadcast unregister notification */ | |
056925ab | 7155 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 7156 | |
748e2d93 | 7157 | __rtnl_unlock(); |
0115e8e3 | 7158 | rcu_barrier(); |
748e2d93 ED |
7159 | rtnl_lock(); |
7160 | ||
0115e8e3 | 7161 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
7162 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
7163 | &dev->state)) { | |
7164 | /* We must not have linkwatch events | |
7165 | * pending on unregister. If this | |
7166 | * happens, we simply run the queue | |
7167 | * unscheduled, resulting in a noop | |
7168 | * for this device. | |
7169 | */ | |
7170 | linkwatch_run_queue(); | |
7171 | } | |
7172 | ||
6756ae4b | 7173 | __rtnl_unlock(); |
1da177e4 LT |
7174 | |
7175 | rebroadcast_time = jiffies; | |
7176 | } | |
7177 | ||
7178 | msleep(250); | |
7179 | ||
29b4433d ED |
7180 | refcnt = netdev_refcnt_read(dev); |
7181 | ||
1da177e4 | 7182 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
7183 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
7184 | dev->name, refcnt); | |
1da177e4 LT |
7185 | warning_time = jiffies; |
7186 | } | |
7187 | } | |
7188 | } | |
7189 | ||
7190 | /* The sequence is: | |
7191 | * | |
7192 | * rtnl_lock(); | |
7193 | * ... | |
7194 | * register_netdevice(x1); | |
7195 | * register_netdevice(x2); | |
7196 | * ... | |
7197 | * unregister_netdevice(y1); | |
7198 | * unregister_netdevice(y2); | |
7199 | * ... | |
7200 | * rtnl_unlock(); | |
7201 | * free_netdev(y1); | |
7202 | * free_netdev(y2); | |
7203 | * | |
58ec3b4d | 7204 | * We are invoked by rtnl_unlock(). |
1da177e4 | 7205 | * This allows us to deal with problems: |
b17a7c17 | 7206 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
7207 | * without deadlocking with linkwatch via keventd. |
7208 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
7209 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
7210 | * |
7211 | * We must not return until all unregister events added during | |
7212 | * the interval the lock was held have been completed. | |
1da177e4 | 7213 | */ |
1da177e4 LT |
7214 | void netdev_run_todo(void) |
7215 | { | |
626ab0e6 | 7216 | struct list_head list; |
1da177e4 | 7217 | |
1da177e4 | 7218 | /* Snapshot list, allow later requests */ |
626ab0e6 | 7219 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
7220 | |
7221 | __rtnl_unlock(); | |
626ab0e6 | 7222 | |
0115e8e3 ED |
7223 | |
7224 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
7225 | if (!list_empty(&list)) |
7226 | rcu_barrier(); | |
7227 | ||
1da177e4 LT |
7228 | while (!list_empty(&list)) { |
7229 | struct net_device *dev | |
e5e26d75 | 7230 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
7231 | list_del(&dev->todo_list); |
7232 | ||
748e2d93 | 7233 | rtnl_lock(); |
0115e8e3 | 7234 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 7235 | __rtnl_unlock(); |
0115e8e3 | 7236 | |
b17a7c17 | 7237 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 7238 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
7239 | dev->name, dev->reg_state); |
7240 | dump_stack(); | |
7241 | continue; | |
7242 | } | |
1da177e4 | 7243 | |
b17a7c17 | 7244 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 7245 | |
b17a7c17 | 7246 | netdev_wait_allrefs(dev); |
1da177e4 | 7247 | |
b17a7c17 | 7248 | /* paranoia */ |
29b4433d | 7249 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
7250 | BUG_ON(!list_empty(&dev->ptype_all)); |
7251 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
7252 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
7253 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 7254 | WARN_ON(dev->dn_ptr); |
1da177e4 | 7255 | |
b17a7c17 SH |
7256 | if (dev->destructor) |
7257 | dev->destructor(dev); | |
9093bbb2 | 7258 | |
50624c93 EB |
7259 | /* Report a network device has been unregistered */ |
7260 | rtnl_lock(); | |
7261 | dev_net(dev)->dev_unreg_count--; | |
7262 | __rtnl_unlock(); | |
7263 | wake_up(&netdev_unregistering_wq); | |
7264 | ||
9093bbb2 SH |
7265 | /* Free network device */ |
7266 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 7267 | } |
1da177e4 LT |
7268 | } |
7269 | ||
9256645a JW |
7270 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
7271 | * all the same fields in the same order as net_device_stats, with only | |
7272 | * the type differing, but rtnl_link_stats64 may have additional fields | |
7273 | * at the end for newer counters. | |
3cfde79c | 7274 | */ |
77a1abf5 ED |
7275 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
7276 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
7277 | { |
7278 | #if BITS_PER_LONG == 64 | |
9256645a | 7279 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
77a1abf5 | 7280 | memcpy(stats64, netdev_stats, sizeof(*stats64)); |
9256645a JW |
7281 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7282 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
7283 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 7284 | #else |
9256645a | 7285 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
7286 | const unsigned long *src = (const unsigned long *)netdev_stats; |
7287 | u64 *dst = (u64 *)stats64; | |
7288 | ||
9256645a | 7289 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
7290 | for (i = 0; i < n; i++) |
7291 | dst[i] = src[i]; | |
9256645a JW |
7292 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7293 | memset((char *)stats64 + n * sizeof(u64), 0, | |
7294 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
7295 | #endif |
7296 | } | |
77a1abf5 | 7297 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 7298 | |
eeda3fd6 SH |
7299 | /** |
7300 | * dev_get_stats - get network device statistics | |
7301 | * @dev: device to get statistics from | |
28172739 | 7302 | * @storage: place to store stats |
eeda3fd6 | 7303 | * |
d7753516 BH |
7304 | * Get network statistics from device. Return @storage. |
7305 | * The device driver may provide its own method by setting | |
7306 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
7307 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 7308 | */ |
d7753516 BH |
7309 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
7310 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 7311 | { |
eeda3fd6 SH |
7312 | const struct net_device_ops *ops = dev->netdev_ops; |
7313 | ||
28172739 ED |
7314 | if (ops->ndo_get_stats64) { |
7315 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
7316 | ops->ndo_get_stats64(dev, storage); |
7317 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 7318 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
7319 | } else { |
7320 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 7321 | } |
caf586e5 | 7322 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
015f0688 | 7323 | storage->tx_dropped += atomic_long_read(&dev->tx_dropped); |
6e7333d3 | 7324 | storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler); |
28172739 | 7325 | return storage; |
c45d286e | 7326 | } |
eeda3fd6 | 7327 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 7328 | |
24824a09 | 7329 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 7330 | { |
24824a09 | 7331 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 7332 | |
24824a09 ED |
7333 | #ifdef CONFIG_NET_CLS_ACT |
7334 | if (queue) | |
7335 | return queue; | |
7336 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
7337 | if (!queue) | |
7338 | return NULL; | |
7339 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 7340 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
7341 | queue->qdisc_sleeping = &noop_qdisc; |
7342 | rcu_assign_pointer(dev->ingress_queue, queue); | |
7343 | #endif | |
7344 | return queue; | |
bb949fbd DM |
7345 | } |
7346 | ||
2c60db03 ED |
7347 | static const struct ethtool_ops default_ethtool_ops; |
7348 | ||
d07d7507 SG |
7349 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
7350 | const struct ethtool_ops *ops) | |
7351 | { | |
7352 | if (dev->ethtool_ops == &default_ethtool_ops) | |
7353 | dev->ethtool_ops = ops; | |
7354 | } | |
7355 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
7356 | ||
74d332c1 ED |
7357 | void netdev_freemem(struct net_device *dev) |
7358 | { | |
7359 | char *addr = (char *)dev - dev->padded; | |
7360 | ||
4cb28970 | 7361 | kvfree(addr); |
74d332c1 ED |
7362 | } |
7363 | ||
1da177e4 | 7364 | /** |
36909ea4 | 7365 | * alloc_netdev_mqs - allocate network device |
c835a677 TG |
7366 | * @sizeof_priv: size of private data to allocate space for |
7367 | * @name: device name format string | |
7368 | * @name_assign_type: origin of device name | |
7369 | * @setup: callback to initialize device | |
7370 | * @txqs: the number of TX subqueues to allocate | |
7371 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
7372 | * |
7373 | * Allocates a struct net_device with private data area for driver use | |
90e51adf | 7374 | * and performs basic initialization. Also allocates subqueue structs |
36909ea4 | 7375 | * for each queue on the device. |
1da177e4 | 7376 | */ |
36909ea4 | 7377 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 7378 | unsigned char name_assign_type, |
36909ea4 TH |
7379 | void (*setup)(struct net_device *), |
7380 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 7381 | { |
1da177e4 | 7382 | struct net_device *dev; |
7943986c | 7383 | size_t alloc_size; |
1ce8e7b5 | 7384 | struct net_device *p; |
1da177e4 | 7385 | |
b6fe17d6 SH |
7386 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
7387 | ||
36909ea4 | 7388 | if (txqs < 1) { |
7b6cd1ce | 7389 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
7390 | return NULL; |
7391 | } | |
7392 | ||
a953be53 | 7393 | #ifdef CONFIG_SYSFS |
36909ea4 | 7394 | if (rxqs < 1) { |
7b6cd1ce | 7395 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
7396 | return NULL; |
7397 | } | |
7398 | #endif | |
7399 | ||
fd2ea0a7 | 7400 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
7401 | if (sizeof_priv) { |
7402 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 7403 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
7404 | alloc_size += sizeof_priv; |
7405 | } | |
7406 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 7407 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 7408 | |
74d332c1 ED |
7409 | p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
7410 | if (!p) | |
7411 | p = vzalloc(alloc_size); | |
62b5942a | 7412 | if (!p) |
1da177e4 | 7413 | return NULL; |
1da177e4 | 7414 | |
1ce8e7b5 | 7415 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 7416 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 7417 | |
29b4433d ED |
7418 | dev->pcpu_refcnt = alloc_percpu(int); |
7419 | if (!dev->pcpu_refcnt) | |
74d332c1 | 7420 | goto free_dev; |
ab9c73cc | 7421 | |
ab9c73cc | 7422 | if (dev_addr_init(dev)) |
29b4433d | 7423 | goto free_pcpu; |
ab9c73cc | 7424 | |
22bedad3 | 7425 | dev_mc_init(dev); |
a748ee24 | 7426 | dev_uc_init(dev); |
ccffad25 | 7427 | |
c346dca1 | 7428 | dev_net_set(dev, &init_net); |
1da177e4 | 7429 | |
8d3bdbd5 | 7430 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 7431 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 7432 | |
8d3bdbd5 DM |
7433 | INIT_LIST_HEAD(&dev->napi_list); |
7434 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 7435 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 7436 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
7437 | INIT_LIST_HEAD(&dev->adj_list.upper); |
7438 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7439 | INIT_LIST_HEAD(&dev->all_adj_list.upper); | |
7440 | INIT_LIST_HEAD(&dev->all_adj_list.lower); | |
7866a621 SN |
7441 | INIT_LIST_HEAD(&dev->ptype_all); |
7442 | INIT_LIST_HEAD(&dev->ptype_specific); | |
02875878 | 7443 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
7444 | setup(dev); |
7445 | ||
a813104d | 7446 | if (!dev->tx_queue_len) { |
f84bb1ea | 7447 | dev->priv_flags |= IFF_NO_QUEUE; |
a813104d PS |
7448 | dev->tx_queue_len = 1; |
7449 | } | |
906470c1 | 7450 | |
36909ea4 TH |
7451 | dev->num_tx_queues = txqs; |
7452 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 7453 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 7454 | goto free_all; |
e8a0464c | 7455 | |
a953be53 | 7456 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
7457 | dev->num_rx_queues = rxqs; |
7458 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 7459 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 7460 | goto free_all; |
df334545 | 7461 | #endif |
0a9627f2 | 7462 | |
1da177e4 | 7463 | strcpy(dev->name, name); |
c835a677 | 7464 | dev->name_assign_type = name_assign_type; |
cbda10fa | 7465 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
7466 | if (!dev->ethtool_ops) |
7467 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
7468 | |
7469 | nf_hook_ingress_init(dev); | |
7470 | ||
1da177e4 | 7471 | return dev; |
ab9c73cc | 7472 | |
8d3bdbd5 DM |
7473 | free_all: |
7474 | free_netdev(dev); | |
7475 | return NULL; | |
7476 | ||
29b4433d ED |
7477 | free_pcpu: |
7478 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
7479 | free_dev: |
7480 | netdev_freemem(dev); | |
ab9c73cc | 7481 | return NULL; |
1da177e4 | 7482 | } |
36909ea4 | 7483 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
7484 | |
7485 | /** | |
7486 | * free_netdev - free network device | |
7487 | * @dev: device | |
7488 | * | |
4ec93edb YH |
7489 | * This function does the last stage of destroying an allocated device |
7490 | * interface. The reference to the device object is released. | |
1da177e4 | 7491 | * If this is the last reference then it will be freed. |
93d05d4a | 7492 | * Must be called in process context. |
1da177e4 LT |
7493 | */ |
7494 | void free_netdev(struct net_device *dev) | |
7495 | { | |
d565b0a1 HX |
7496 | struct napi_struct *p, *n; |
7497 | ||
93d05d4a | 7498 | might_sleep(); |
60877a32 | 7499 | netif_free_tx_queues(dev); |
a953be53 | 7500 | #ifdef CONFIG_SYSFS |
10595902 | 7501 | kvfree(dev->_rx); |
fe822240 | 7502 | #endif |
e8a0464c | 7503 | |
33d480ce | 7504 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 7505 | |
f001fde5 JP |
7506 | /* Flush device addresses */ |
7507 | dev_addr_flush(dev); | |
7508 | ||
d565b0a1 HX |
7509 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
7510 | netif_napi_del(p); | |
7511 | ||
29b4433d ED |
7512 | free_percpu(dev->pcpu_refcnt); |
7513 | dev->pcpu_refcnt = NULL; | |
7514 | ||
3041a069 | 7515 | /* Compatibility with error handling in drivers */ |
1da177e4 | 7516 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 7517 | netdev_freemem(dev); |
1da177e4 LT |
7518 | return; |
7519 | } | |
7520 | ||
7521 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
7522 | dev->reg_state = NETREG_RELEASED; | |
7523 | ||
43cb76d9 GKH |
7524 | /* will free via device release */ |
7525 | put_device(&dev->dev); | |
1da177e4 | 7526 | } |
d1b19dff | 7527 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 7528 | |
f0db275a SH |
7529 | /** |
7530 | * synchronize_net - Synchronize with packet receive processing | |
7531 | * | |
7532 | * Wait for packets currently being received to be done. | |
7533 | * Does not block later packets from starting. | |
7534 | */ | |
4ec93edb | 7535 | void synchronize_net(void) |
1da177e4 LT |
7536 | { |
7537 | might_sleep(); | |
be3fc413 ED |
7538 | if (rtnl_is_locked()) |
7539 | synchronize_rcu_expedited(); | |
7540 | else | |
7541 | synchronize_rcu(); | |
1da177e4 | 7542 | } |
d1b19dff | 7543 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
7544 | |
7545 | /** | |
44a0873d | 7546 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 7547 | * @dev: device |
44a0873d | 7548 | * @head: list |
6ebfbc06 | 7549 | * |
1da177e4 | 7550 | * This function shuts down a device interface and removes it |
d59b54b1 | 7551 | * from the kernel tables. |
44a0873d | 7552 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
7553 | * |
7554 | * Callers must hold the rtnl semaphore. You may want | |
7555 | * unregister_netdev() instead of this. | |
7556 | */ | |
7557 | ||
44a0873d | 7558 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 7559 | { |
a6620712 HX |
7560 | ASSERT_RTNL(); |
7561 | ||
44a0873d | 7562 | if (head) { |
9fdce099 | 7563 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
7564 | } else { |
7565 | rollback_registered(dev); | |
7566 | /* Finish processing unregister after unlock */ | |
7567 | net_set_todo(dev); | |
7568 | } | |
1da177e4 | 7569 | } |
44a0873d | 7570 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 7571 | |
9b5e383c ED |
7572 | /** |
7573 | * unregister_netdevice_many - unregister many devices | |
7574 | * @head: list of devices | |
87757a91 ED |
7575 | * |
7576 | * Note: As most callers use a stack allocated list_head, | |
7577 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
7578 | */ |
7579 | void unregister_netdevice_many(struct list_head *head) | |
7580 | { | |
7581 | struct net_device *dev; | |
7582 | ||
7583 | if (!list_empty(head)) { | |
7584 | rollback_registered_many(head); | |
7585 | list_for_each_entry(dev, head, unreg_list) | |
7586 | net_set_todo(dev); | |
87757a91 | 7587 | list_del(head); |
9b5e383c ED |
7588 | } |
7589 | } | |
63c8099d | 7590 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 7591 | |
1da177e4 LT |
7592 | /** |
7593 | * unregister_netdev - remove device from the kernel | |
7594 | * @dev: device | |
7595 | * | |
7596 | * This function shuts down a device interface and removes it | |
d59b54b1 | 7597 | * from the kernel tables. |
1da177e4 LT |
7598 | * |
7599 | * This is just a wrapper for unregister_netdevice that takes | |
7600 | * the rtnl semaphore. In general you want to use this and not | |
7601 | * unregister_netdevice. | |
7602 | */ | |
7603 | void unregister_netdev(struct net_device *dev) | |
7604 | { | |
7605 | rtnl_lock(); | |
7606 | unregister_netdevice(dev); | |
7607 | rtnl_unlock(); | |
7608 | } | |
1da177e4 LT |
7609 | EXPORT_SYMBOL(unregister_netdev); |
7610 | ||
ce286d32 EB |
7611 | /** |
7612 | * dev_change_net_namespace - move device to different nethost namespace | |
7613 | * @dev: device | |
7614 | * @net: network namespace | |
7615 | * @pat: If not NULL name pattern to try if the current device name | |
7616 | * is already taken in the destination network namespace. | |
7617 | * | |
7618 | * This function shuts down a device interface and moves it | |
7619 | * to a new network namespace. On success 0 is returned, on | |
7620 | * a failure a netagive errno code is returned. | |
7621 | * | |
7622 | * Callers must hold the rtnl semaphore. | |
7623 | */ | |
7624 | ||
7625 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
7626 | { | |
ce286d32 EB |
7627 | int err; |
7628 | ||
7629 | ASSERT_RTNL(); | |
7630 | ||
7631 | /* Don't allow namespace local devices to be moved. */ | |
7632 | err = -EINVAL; | |
7633 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7634 | goto out; | |
7635 | ||
7636 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
7637 | if (dev->reg_state != NETREG_REGISTERED) |
7638 | goto out; | |
7639 | ||
7640 | /* Get out if there is nothing todo */ | |
7641 | err = 0; | |
878628fb | 7642 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
7643 | goto out; |
7644 | ||
7645 | /* Pick the destination device name, and ensure | |
7646 | * we can use it in the destination network namespace. | |
7647 | */ | |
7648 | err = -EEXIST; | |
d9031024 | 7649 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
7650 | /* We get here if we can't use the current device name */ |
7651 | if (!pat) | |
7652 | goto out; | |
828de4f6 | 7653 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
7654 | goto out; |
7655 | } | |
7656 | ||
7657 | /* | |
7658 | * And now a mini version of register_netdevice unregister_netdevice. | |
7659 | */ | |
7660 | ||
7661 | /* If device is running close it first. */ | |
9b772652 | 7662 | dev_close(dev); |
ce286d32 EB |
7663 | |
7664 | /* And unlink it from device chain */ | |
7665 | err = -ENODEV; | |
7666 | unlist_netdevice(dev); | |
7667 | ||
7668 | synchronize_net(); | |
7669 | ||
7670 | /* Shutdown queueing discipline. */ | |
7671 | dev_shutdown(dev); | |
7672 | ||
7673 | /* Notify protocols, that we are about to destroy | |
7674 | this device. They should clean all the things. | |
3b27e105 DL |
7675 | |
7676 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
7677 | This is wanted because this way 8021q and macvlan know | |
7678 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
7679 | */ |
7680 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
7681 | rcu_barrier(); |
7682 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 7683 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
7684 | |
7685 | /* | |
7686 | * Flush the unicast and multicast chains | |
7687 | */ | |
a748ee24 | 7688 | dev_uc_flush(dev); |
22bedad3 | 7689 | dev_mc_flush(dev); |
ce286d32 | 7690 | |
4e66ae2e SH |
7691 | /* Send a netdev-removed uevent to the old namespace */ |
7692 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 7693 | netdev_adjacent_del_links(dev); |
4e66ae2e | 7694 | |
ce286d32 | 7695 | /* Actually switch the network namespace */ |
c346dca1 | 7696 | dev_net_set(dev, net); |
ce286d32 | 7697 | |
ce286d32 | 7698 | /* If there is an ifindex conflict assign a new one */ |
7a66bbc9 | 7699 | if (__dev_get_by_index(net, dev->ifindex)) |
ce286d32 | 7700 | dev->ifindex = dev_new_index(net); |
ce286d32 | 7701 | |
4e66ae2e SH |
7702 | /* Send a netdev-add uevent to the new namespace */ |
7703 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 7704 | netdev_adjacent_add_links(dev); |
4e66ae2e | 7705 | |
8b41d188 | 7706 | /* Fixup kobjects */ |
a1b3f594 | 7707 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 7708 | WARN_ON(err); |
ce286d32 EB |
7709 | |
7710 | /* Add the device back in the hashes */ | |
7711 | list_netdevice(dev); | |
7712 | ||
7713 | /* Notify protocols, that a new device appeared. */ | |
7714 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
7715 | ||
d90a909e EB |
7716 | /* |
7717 | * Prevent userspace races by waiting until the network | |
7718 | * device is fully setup before sending notifications. | |
7719 | */ | |
7f294054 | 7720 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 7721 | |
ce286d32 EB |
7722 | synchronize_net(); |
7723 | err = 0; | |
7724 | out: | |
7725 | return err; | |
7726 | } | |
463d0183 | 7727 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 7728 | |
1da177e4 LT |
7729 | static int dev_cpu_callback(struct notifier_block *nfb, |
7730 | unsigned long action, | |
7731 | void *ocpu) | |
7732 | { | |
7733 | struct sk_buff **list_skb; | |
1da177e4 LT |
7734 | struct sk_buff *skb; |
7735 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
7736 | struct softnet_data *sd, *oldsd; | |
7737 | ||
8bb78442 | 7738 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
7739 | return NOTIFY_OK; |
7740 | ||
7741 | local_irq_disable(); | |
7742 | cpu = smp_processor_id(); | |
7743 | sd = &per_cpu(softnet_data, cpu); | |
7744 | oldsd = &per_cpu(softnet_data, oldcpu); | |
7745 | ||
7746 | /* Find end of our completion_queue. */ | |
7747 | list_skb = &sd->completion_queue; | |
7748 | while (*list_skb) | |
7749 | list_skb = &(*list_skb)->next; | |
7750 | /* Append completion queue from offline CPU. */ | |
7751 | *list_skb = oldsd->completion_queue; | |
7752 | oldsd->completion_queue = NULL; | |
7753 | ||
1da177e4 | 7754 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
7755 | if (oldsd->output_queue) { |
7756 | *sd->output_queue_tailp = oldsd->output_queue; | |
7757 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
7758 | oldsd->output_queue = NULL; | |
7759 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
7760 | } | |
ac64da0b ED |
7761 | /* Append NAPI poll list from offline CPU, with one exception : |
7762 | * process_backlog() must be called by cpu owning percpu backlog. | |
7763 | * We properly handle process_queue & input_pkt_queue later. | |
7764 | */ | |
7765 | while (!list_empty(&oldsd->poll_list)) { | |
7766 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
7767 | struct napi_struct, | |
7768 | poll_list); | |
7769 | ||
7770 | list_del_init(&napi->poll_list); | |
7771 | if (napi->poll == process_backlog) | |
7772 | napi->state = 0; | |
7773 | else | |
7774 | ____napi_schedule(sd, napi); | |
264524d5 | 7775 | } |
1da177e4 LT |
7776 | |
7777 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
7778 | local_irq_enable(); | |
7779 | ||
7780 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 7781 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 7782 | netif_rx_ni(skb); |
76cc8b13 | 7783 | input_queue_head_incr(oldsd); |
fec5e652 | 7784 | } |
ac64da0b | 7785 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 7786 | netif_rx_ni(skb); |
76cc8b13 TH |
7787 | input_queue_head_incr(oldsd); |
7788 | } | |
1da177e4 LT |
7789 | |
7790 | return NOTIFY_OK; | |
7791 | } | |
1da177e4 LT |
7792 | |
7793 | ||
7f353bf2 | 7794 | /** |
b63365a2 HX |
7795 | * netdev_increment_features - increment feature set by one |
7796 | * @all: current feature set | |
7797 | * @one: new feature set | |
7798 | * @mask: mask feature set | |
7f353bf2 HX |
7799 | * |
7800 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
7801 | * @one to the master device with current feature set @all. Will not |
7802 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 7803 | */ |
c8f44aff MM |
7804 | netdev_features_t netdev_increment_features(netdev_features_t all, |
7805 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 7806 | { |
c8cd0989 | 7807 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 7808 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 7809 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 7810 | |
a188222b | 7811 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 7812 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 7813 | |
1742f183 | 7814 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
7815 | if (all & NETIF_F_HW_CSUM) |
7816 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
7817 | |
7818 | return all; | |
7819 | } | |
b63365a2 | 7820 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 7821 | |
430f03cd | 7822 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
7823 | { |
7824 | int i; | |
7825 | struct hlist_head *hash; | |
7826 | ||
7827 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
7828 | if (hash != NULL) | |
7829 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
7830 | INIT_HLIST_HEAD(&hash[i]); | |
7831 | ||
7832 | return hash; | |
7833 | } | |
7834 | ||
881d966b | 7835 | /* Initialize per network namespace state */ |
4665079c | 7836 | static int __net_init netdev_init(struct net *net) |
881d966b | 7837 | { |
734b6541 RM |
7838 | if (net != &init_net) |
7839 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 7840 | |
30d97d35 PE |
7841 | net->dev_name_head = netdev_create_hash(); |
7842 | if (net->dev_name_head == NULL) | |
7843 | goto err_name; | |
881d966b | 7844 | |
30d97d35 PE |
7845 | net->dev_index_head = netdev_create_hash(); |
7846 | if (net->dev_index_head == NULL) | |
7847 | goto err_idx; | |
881d966b EB |
7848 | |
7849 | return 0; | |
30d97d35 PE |
7850 | |
7851 | err_idx: | |
7852 | kfree(net->dev_name_head); | |
7853 | err_name: | |
7854 | return -ENOMEM; | |
881d966b EB |
7855 | } |
7856 | ||
f0db275a SH |
7857 | /** |
7858 | * netdev_drivername - network driver for the device | |
7859 | * @dev: network device | |
f0db275a SH |
7860 | * |
7861 | * Determine network driver for device. | |
7862 | */ | |
3019de12 | 7863 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 7864 | { |
cf04a4c7 SH |
7865 | const struct device_driver *driver; |
7866 | const struct device *parent; | |
3019de12 | 7867 | const char *empty = ""; |
6579e57b AV |
7868 | |
7869 | parent = dev->dev.parent; | |
6579e57b | 7870 | if (!parent) |
3019de12 | 7871 | return empty; |
6579e57b AV |
7872 | |
7873 | driver = parent->driver; | |
7874 | if (driver && driver->name) | |
3019de12 DM |
7875 | return driver->name; |
7876 | return empty; | |
6579e57b AV |
7877 | } |
7878 | ||
6ea754eb JP |
7879 | static void __netdev_printk(const char *level, const struct net_device *dev, |
7880 | struct va_format *vaf) | |
256df2f3 | 7881 | { |
b004ff49 | 7882 | if (dev && dev->dev.parent) { |
6ea754eb JP |
7883 | dev_printk_emit(level[1] - '0', |
7884 | dev->dev.parent, | |
7885 | "%s %s %s%s: %pV", | |
7886 | dev_driver_string(dev->dev.parent), | |
7887 | dev_name(dev->dev.parent), | |
7888 | netdev_name(dev), netdev_reg_state(dev), | |
7889 | vaf); | |
b004ff49 | 7890 | } else if (dev) { |
6ea754eb JP |
7891 | printk("%s%s%s: %pV", |
7892 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 7893 | } else { |
6ea754eb | 7894 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 7895 | } |
256df2f3 JP |
7896 | } |
7897 | ||
6ea754eb JP |
7898 | void netdev_printk(const char *level, const struct net_device *dev, |
7899 | const char *format, ...) | |
256df2f3 JP |
7900 | { |
7901 | struct va_format vaf; | |
7902 | va_list args; | |
256df2f3 JP |
7903 | |
7904 | va_start(args, format); | |
7905 | ||
7906 | vaf.fmt = format; | |
7907 | vaf.va = &args; | |
7908 | ||
6ea754eb | 7909 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 7910 | |
256df2f3 | 7911 | va_end(args); |
256df2f3 JP |
7912 | } |
7913 | EXPORT_SYMBOL(netdev_printk); | |
7914 | ||
7915 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 7916 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 7917 | { \ |
256df2f3 JP |
7918 | struct va_format vaf; \ |
7919 | va_list args; \ | |
7920 | \ | |
7921 | va_start(args, fmt); \ | |
7922 | \ | |
7923 | vaf.fmt = fmt; \ | |
7924 | vaf.va = &args; \ | |
7925 | \ | |
6ea754eb | 7926 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 7927 | \ |
256df2f3 | 7928 | va_end(args); \ |
256df2f3 JP |
7929 | } \ |
7930 | EXPORT_SYMBOL(func); | |
7931 | ||
7932 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
7933 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
7934 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
7935 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
7936 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
7937 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
7938 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
7939 | ||
4665079c | 7940 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
7941 | { |
7942 | kfree(net->dev_name_head); | |
7943 | kfree(net->dev_index_head); | |
7944 | } | |
7945 | ||
022cbae6 | 7946 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
7947 | .init = netdev_init, |
7948 | .exit = netdev_exit, | |
7949 | }; | |
7950 | ||
4665079c | 7951 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 7952 | { |
e008b5fc | 7953 | struct net_device *dev, *aux; |
ce286d32 | 7954 | /* |
e008b5fc | 7955 | * Push all migratable network devices back to the |
ce286d32 EB |
7956 | * initial network namespace |
7957 | */ | |
7958 | rtnl_lock(); | |
e008b5fc | 7959 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 7960 | int err; |
aca51397 | 7961 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
7962 | |
7963 | /* Ignore unmoveable devices (i.e. loopback) */ | |
7964 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7965 | continue; | |
7966 | ||
e008b5fc EB |
7967 | /* Leave virtual devices for the generic cleanup */ |
7968 | if (dev->rtnl_link_ops) | |
7969 | continue; | |
d0c082ce | 7970 | |
25985edc | 7971 | /* Push remaining network devices to init_net */ |
aca51397 PE |
7972 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
7973 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 7974 | if (err) { |
7b6cd1ce JP |
7975 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
7976 | __func__, dev->name, err); | |
aca51397 | 7977 | BUG(); |
ce286d32 EB |
7978 | } |
7979 | } | |
7980 | rtnl_unlock(); | |
7981 | } | |
7982 | ||
50624c93 EB |
7983 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
7984 | { | |
7985 | /* Return with the rtnl_lock held when there are no network | |
7986 | * devices unregistering in any network namespace in net_list. | |
7987 | */ | |
7988 | struct net *net; | |
7989 | bool unregistering; | |
ff960a73 | 7990 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 7991 | |
ff960a73 | 7992 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 7993 | for (;;) { |
50624c93 EB |
7994 | unregistering = false; |
7995 | rtnl_lock(); | |
7996 | list_for_each_entry(net, net_list, exit_list) { | |
7997 | if (net->dev_unreg_count > 0) { | |
7998 | unregistering = true; | |
7999 | break; | |
8000 | } | |
8001 | } | |
8002 | if (!unregistering) | |
8003 | break; | |
8004 | __rtnl_unlock(); | |
ff960a73 PZ |
8005 | |
8006 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 8007 | } |
ff960a73 | 8008 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
8009 | } |
8010 | ||
04dc7f6b EB |
8011 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
8012 | { | |
8013 | /* At exit all network devices most be removed from a network | |
b595076a | 8014 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
8015 | * Do this across as many network namespaces as possible to |
8016 | * improve batching efficiency. | |
8017 | */ | |
8018 | struct net_device *dev; | |
8019 | struct net *net; | |
8020 | LIST_HEAD(dev_kill_list); | |
8021 | ||
50624c93 EB |
8022 | /* To prevent network device cleanup code from dereferencing |
8023 | * loopback devices or network devices that have been freed | |
8024 | * wait here for all pending unregistrations to complete, | |
8025 | * before unregistring the loopback device and allowing the | |
8026 | * network namespace be freed. | |
8027 | * | |
8028 | * The netdev todo list containing all network devices | |
8029 | * unregistrations that happen in default_device_exit_batch | |
8030 | * will run in the rtnl_unlock() at the end of | |
8031 | * default_device_exit_batch. | |
8032 | */ | |
8033 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
8034 | list_for_each_entry(net, net_list, exit_list) { |
8035 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 8036 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
8037 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
8038 | else | |
8039 | unregister_netdevice_queue(dev, &dev_kill_list); | |
8040 | } | |
8041 | } | |
8042 | unregister_netdevice_many(&dev_kill_list); | |
8043 | rtnl_unlock(); | |
8044 | } | |
8045 | ||
022cbae6 | 8046 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 8047 | .exit = default_device_exit, |
04dc7f6b | 8048 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
8049 | }; |
8050 | ||
1da177e4 LT |
8051 | /* |
8052 | * Initialize the DEV module. At boot time this walks the device list and | |
8053 | * unhooks any devices that fail to initialise (normally hardware not | |
8054 | * present) and leaves us with a valid list of present and active devices. | |
8055 | * | |
8056 | */ | |
8057 | ||
8058 | /* | |
8059 | * This is called single threaded during boot, so no need | |
8060 | * to take the rtnl semaphore. | |
8061 | */ | |
8062 | static int __init net_dev_init(void) | |
8063 | { | |
8064 | int i, rc = -ENOMEM; | |
8065 | ||
8066 | BUG_ON(!dev_boot_phase); | |
8067 | ||
1da177e4 LT |
8068 | if (dev_proc_init()) |
8069 | goto out; | |
8070 | ||
8b41d188 | 8071 | if (netdev_kobject_init()) |
1da177e4 LT |
8072 | goto out; |
8073 | ||
8074 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 8075 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
8076 | INIT_LIST_HEAD(&ptype_base[i]); |
8077 | ||
62532da9 VY |
8078 | INIT_LIST_HEAD(&offload_base); |
8079 | ||
881d966b EB |
8080 | if (register_pernet_subsys(&netdev_net_ops)) |
8081 | goto out; | |
1da177e4 LT |
8082 | |
8083 | /* | |
8084 | * Initialise the packet receive queues. | |
8085 | */ | |
8086 | ||
6f912042 | 8087 | for_each_possible_cpu(i) { |
e36fa2f7 | 8088 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 8089 | |
e36fa2f7 | 8090 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 8091 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 8092 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 8093 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 8094 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
8095 | sd->csd.func = rps_trigger_softirq; |
8096 | sd->csd.info = sd; | |
e36fa2f7 | 8097 | sd->cpu = i; |
1e94d72f | 8098 | #endif |
0a9627f2 | 8099 | |
e36fa2f7 ED |
8100 | sd->backlog.poll = process_backlog; |
8101 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
8102 | } |
8103 | ||
1da177e4 LT |
8104 | dev_boot_phase = 0; |
8105 | ||
505d4f73 EB |
8106 | /* The loopback device is special if any other network devices |
8107 | * is present in a network namespace the loopback device must | |
8108 | * be present. Since we now dynamically allocate and free the | |
8109 | * loopback device ensure this invariant is maintained by | |
8110 | * keeping the loopback device as the first device on the | |
8111 | * list of network devices. Ensuring the loopback devices | |
8112 | * is the first device that appears and the last network device | |
8113 | * that disappears. | |
8114 | */ | |
8115 | if (register_pernet_device(&loopback_net_ops)) | |
8116 | goto out; | |
8117 | ||
8118 | if (register_pernet_device(&default_device_ops)) | |
8119 | goto out; | |
8120 | ||
962cf36c CM |
8121 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
8122 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
8123 | |
8124 | hotcpu_notifier(dev_cpu_callback, 0); | |
f38a9eb1 | 8125 | dst_subsys_init(); |
1da177e4 LT |
8126 | rc = 0; |
8127 | out: | |
8128 | return rc; | |
8129 | } | |
8130 | ||
8131 | subsys_initcall(net_dev_init); |