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> | |
a7862b45 | 97 | #include <linux/bpf.h> |
457c4cbc | 98 | #include <net/net_namespace.h> |
1da177e4 | 99 | #include <net/sock.h> |
02d62e86 | 100 | #include <net/busy_poll.h> |
1da177e4 | 101 | #include <linux/rtnetlink.h> |
1da177e4 | 102 | #include <linux/stat.h> |
1da177e4 | 103 | #include <net/dst.h> |
fc4099f1 | 104 | #include <net/dst_metadata.h> |
1da177e4 LT |
105 | #include <net/pkt_sched.h> |
106 | #include <net/checksum.h> | |
44540960 | 107 | #include <net/xfrm.h> |
1da177e4 LT |
108 | #include <linux/highmem.h> |
109 | #include <linux/init.h> | |
1da177e4 | 110 | #include <linux/module.h> |
1da177e4 LT |
111 | #include <linux/netpoll.h> |
112 | #include <linux/rcupdate.h> | |
113 | #include <linux/delay.h> | |
1da177e4 | 114 | #include <net/iw_handler.h> |
1da177e4 | 115 | #include <asm/current.h> |
5bdb9886 | 116 | #include <linux/audit.h> |
db217334 | 117 | #include <linux/dmaengine.h> |
f6a78bfc | 118 | #include <linux/err.h> |
c7fa9d18 | 119 | #include <linux/ctype.h> |
723e98b7 | 120 | #include <linux/if_arp.h> |
6de329e2 | 121 | #include <linux/if_vlan.h> |
8f0f2223 | 122 | #include <linux/ip.h> |
ad55dcaf | 123 | #include <net/ip.h> |
25cd9ba0 | 124 | #include <net/mpls.h> |
8f0f2223 DM |
125 | #include <linux/ipv6.h> |
126 | #include <linux/in.h> | |
b6b2fed1 DM |
127 | #include <linux/jhash.h> |
128 | #include <linux/random.h> | |
9cbc1cb8 | 129 | #include <trace/events/napi.h> |
cf66ba58 | 130 | #include <trace/events/net.h> |
07dc22e7 | 131 | #include <trace/events/skb.h> |
5acbbd42 | 132 | #include <linux/pci.h> |
caeda9b9 | 133 | #include <linux/inetdevice.h> |
c445477d | 134 | #include <linux/cpu_rmap.h> |
c5905afb | 135 | #include <linux/static_key.h> |
af12fa6e | 136 | #include <linux/hashtable.h> |
60877a32 | 137 | #include <linux/vmalloc.h> |
529d0489 | 138 | #include <linux/if_macvlan.h> |
e7fd2885 | 139 | #include <linux/errqueue.h> |
3b47d303 | 140 | #include <linux/hrtimer.h> |
e687ad60 | 141 | #include <linux/netfilter_ingress.h> |
6ae23ad3 | 142 | #include <linux/sctp.h> |
40e4e713 | 143 | #include <linux/crash_dump.h> |
1da177e4 | 144 | |
342709ef PE |
145 | #include "net-sysfs.h" |
146 | ||
d565b0a1 HX |
147 | /* Instead of increasing this, you should create a hash table. */ |
148 | #define MAX_GRO_SKBS 8 | |
149 | ||
5d38a079 HX |
150 | /* This should be increased if a protocol with a bigger head is added. */ |
151 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
152 | ||
1da177e4 | 153 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 154 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
155 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
156 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 157 | static struct list_head offload_base __read_mostly; |
1da177e4 | 158 | |
ae78dbfa | 159 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
160 | static int call_netdevice_notifiers_info(unsigned long val, |
161 | struct net_device *dev, | |
162 | struct netdev_notifier_info *info); | |
ae78dbfa | 163 | |
1da177e4 | 164 | /* |
7562f876 | 165 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
166 | * semaphore. |
167 | * | |
c6d14c84 | 168 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
169 | * |
170 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 171 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
172 | * actual updates. This allows pure readers to access the list even |
173 | * while a writer is preparing to update it. | |
174 | * | |
175 | * To put it another way, dev_base_lock is held for writing only to | |
176 | * protect against pure readers; the rtnl semaphore provides the | |
177 | * protection against other writers. | |
178 | * | |
179 | * See, for example usages, register_netdevice() and | |
180 | * unregister_netdevice(), which must be called with the rtnl | |
181 | * semaphore held. | |
182 | */ | |
1da177e4 | 183 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
184 | EXPORT_SYMBOL(dev_base_lock); |
185 | ||
af12fa6e ET |
186 | /* protects napi_hash addition/deletion and napi_gen_id */ |
187 | static DEFINE_SPINLOCK(napi_hash_lock); | |
188 | ||
52bd2d62 | 189 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 190 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 191 | |
18afa4b0 | 192 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 193 | |
4e985ada TG |
194 | static inline void dev_base_seq_inc(struct net *net) |
195 | { | |
196 | while (++net->dev_base_seq == 0); | |
197 | } | |
198 | ||
881d966b | 199 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 200 | { |
95c96174 ED |
201 | unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
202 | ||
08e9897d | 203 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
204 | } |
205 | ||
881d966b | 206 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 207 | { |
7c28bd0b | 208 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
209 | } |
210 | ||
e36fa2f7 | 211 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
212 | { |
213 | #ifdef CONFIG_RPS | |
e36fa2f7 | 214 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
215 | #endif |
216 | } | |
217 | ||
e36fa2f7 | 218 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
219 | { |
220 | #ifdef CONFIG_RPS | |
e36fa2f7 | 221 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
222 | #endif |
223 | } | |
224 | ||
ce286d32 | 225 | /* Device list insertion */ |
53759be9 | 226 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 227 | { |
c346dca1 | 228 | struct net *net = dev_net(dev); |
ce286d32 EB |
229 | |
230 | ASSERT_RTNL(); | |
231 | ||
232 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 233 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 234 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
235 | hlist_add_head_rcu(&dev->index_hlist, |
236 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 237 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
238 | |
239 | dev_base_seq_inc(net); | |
ce286d32 EB |
240 | } |
241 | ||
fb699dfd ED |
242 | /* Device list removal |
243 | * caller must respect a RCU grace period before freeing/reusing dev | |
244 | */ | |
ce286d32 EB |
245 | static void unlist_netdevice(struct net_device *dev) |
246 | { | |
247 | ASSERT_RTNL(); | |
248 | ||
249 | /* Unlink dev from the device chain */ | |
250 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 251 | list_del_rcu(&dev->dev_list); |
72c9528b | 252 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 253 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 254 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
255 | |
256 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
257 | } |
258 | ||
1da177e4 LT |
259 | /* |
260 | * Our notifier list | |
261 | */ | |
262 | ||
f07d5b94 | 263 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
264 | |
265 | /* | |
266 | * Device drivers call our routines to queue packets here. We empty the | |
267 | * queue in the local softnet handler. | |
268 | */ | |
bea3348e | 269 | |
9958da05 | 270 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 271 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 272 | |
cf508b12 | 273 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 274 | /* |
c773e847 | 275 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
276 | * according to dev->type |
277 | */ | |
278 | static const unsigned short netdev_lock_type[] = | |
279 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
280 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
281 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
282 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
283 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
284 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
285 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
286 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
287 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
288 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
289 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
290 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
291 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
292 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
293 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 294 | |
36cbd3dc | 295 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
296 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
297 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
298 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
299 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
300 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
301 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
302 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
303 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
304 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
305 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
306 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
307 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
211ed865 PG |
308 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", |
309 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
310 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
311 | |
312 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 313 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
314 | |
315 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
316 | { | |
317 | int i; | |
318 | ||
319 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
320 | if (netdev_lock_type[i] == dev_type) | |
321 | return i; | |
322 | /* the last key is used by default */ | |
323 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
324 | } | |
325 | ||
cf508b12 DM |
326 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
327 | unsigned short dev_type) | |
723e98b7 JP |
328 | { |
329 | int i; | |
330 | ||
331 | i = netdev_lock_pos(dev_type); | |
332 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
333 | netdev_lock_name[i]); | |
334 | } | |
cf508b12 DM |
335 | |
336 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
337 | { | |
338 | int i; | |
339 | ||
340 | i = netdev_lock_pos(dev->type); | |
341 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
342 | &netdev_addr_lock_key[i], | |
343 | netdev_lock_name[i]); | |
344 | } | |
723e98b7 | 345 | #else |
cf508b12 DM |
346 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
347 | unsigned short dev_type) | |
348 | { | |
349 | } | |
350 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
351 | { |
352 | } | |
353 | #endif | |
1da177e4 LT |
354 | |
355 | /******************************************************************************* | |
356 | ||
357 | Protocol management and registration routines | |
358 | ||
359 | *******************************************************************************/ | |
360 | ||
1da177e4 LT |
361 | /* |
362 | * Add a protocol ID to the list. Now that the input handler is | |
363 | * smarter we can dispense with all the messy stuff that used to be | |
364 | * here. | |
365 | * | |
366 | * BEWARE!!! Protocol handlers, mangling input packets, | |
367 | * MUST BE last in hash buckets and checking protocol handlers | |
368 | * MUST start from promiscuous ptype_all chain in net_bh. | |
369 | * It is true now, do not change it. | |
370 | * Explanation follows: if protocol handler, mangling packet, will | |
371 | * be the first on list, it is not able to sense, that packet | |
372 | * is cloned and should be copied-on-write, so that it will | |
373 | * change it and subsequent readers will get broken packet. | |
374 | * --ANK (980803) | |
375 | */ | |
376 | ||
c07b68e8 ED |
377 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
378 | { | |
379 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 380 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 381 | else |
7866a621 SN |
382 | return pt->dev ? &pt->dev->ptype_specific : |
383 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
384 | } |
385 | ||
1da177e4 LT |
386 | /** |
387 | * dev_add_pack - add packet handler | |
388 | * @pt: packet type declaration | |
389 | * | |
390 | * Add a protocol handler to the networking stack. The passed &packet_type | |
391 | * is linked into kernel lists and may not be freed until it has been | |
392 | * removed from the kernel lists. | |
393 | * | |
4ec93edb | 394 | * This call does not sleep therefore it can not |
1da177e4 LT |
395 | * guarantee all CPU's that are in middle of receiving packets |
396 | * will see the new packet type (until the next received packet). | |
397 | */ | |
398 | ||
399 | void dev_add_pack(struct packet_type *pt) | |
400 | { | |
c07b68e8 | 401 | struct list_head *head = ptype_head(pt); |
1da177e4 | 402 | |
c07b68e8 ED |
403 | spin_lock(&ptype_lock); |
404 | list_add_rcu(&pt->list, head); | |
405 | spin_unlock(&ptype_lock); | |
1da177e4 | 406 | } |
d1b19dff | 407 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 408 | |
1da177e4 LT |
409 | /** |
410 | * __dev_remove_pack - remove packet handler | |
411 | * @pt: packet type declaration | |
412 | * | |
413 | * Remove a protocol handler that was previously added to the kernel | |
414 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
415 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 416 | * returns. |
1da177e4 LT |
417 | * |
418 | * The packet type might still be in use by receivers | |
419 | * and must not be freed until after all the CPU's have gone | |
420 | * through a quiescent state. | |
421 | */ | |
422 | void __dev_remove_pack(struct packet_type *pt) | |
423 | { | |
c07b68e8 | 424 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
425 | struct packet_type *pt1; |
426 | ||
c07b68e8 | 427 | spin_lock(&ptype_lock); |
1da177e4 LT |
428 | |
429 | list_for_each_entry(pt1, head, list) { | |
430 | if (pt == pt1) { | |
431 | list_del_rcu(&pt->list); | |
432 | goto out; | |
433 | } | |
434 | } | |
435 | ||
7b6cd1ce | 436 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 437 | out: |
c07b68e8 | 438 | spin_unlock(&ptype_lock); |
1da177e4 | 439 | } |
d1b19dff ED |
440 | EXPORT_SYMBOL(__dev_remove_pack); |
441 | ||
1da177e4 LT |
442 | /** |
443 | * dev_remove_pack - remove packet handler | |
444 | * @pt: packet type declaration | |
445 | * | |
446 | * Remove a protocol handler that was previously added to the kernel | |
447 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
448 | * from the kernel lists and can be freed or reused once this function | |
449 | * returns. | |
450 | * | |
451 | * This call sleeps to guarantee that no CPU is looking at the packet | |
452 | * type after return. | |
453 | */ | |
454 | void dev_remove_pack(struct packet_type *pt) | |
455 | { | |
456 | __dev_remove_pack(pt); | |
4ec93edb | 457 | |
1da177e4 LT |
458 | synchronize_net(); |
459 | } | |
d1b19dff | 460 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 461 | |
62532da9 VY |
462 | |
463 | /** | |
464 | * dev_add_offload - register offload handlers | |
465 | * @po: protocol offload declaration | |
466 | * | |
467 | * Add protocol offload handlers to the networking stack. The passed | |
468 | * &proto_offload is linked into kernel lists and may not be freed until | |
469 | * it has been removed from the kernel lists. | |
470 | * | |
471 | * This call does not sleep therefore it can not | |
472 | * guarantee all CPU's that are in middle of receiving packets | |
473 | * will see the new offload handlers (until the next received packet). | |
474 | */ | |
475 | void dev_add_offload(struct packet_offload *po) | |
476 | { | |
bdef7de4 | 477 | struct packet_offload *elem; |
62532da9 VY |
478 | |
479 | spin_lock(&offload_lock); | |
bdef7de4 DM |
480 | list_for_each_entry(elem, &offload_base, list) { |
481 | if (po->priority < elem->priority) | |
482 | break; | |
483 | } | |
484 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
485 | spin_unlock(&offload_lock); |
486 | } | |
487 | EXPORT_SYMBOL(dev_add_offload); | |
488 | ||
489 | /** | |
490 | * __dev_remove_offload - remove offload handler | |
491 | * @po: packet offload declaration | |
492 | * | |
493 | * Remove a protocol offload handler that was previously added to the | |
494 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
495 | * is removed from the kernel lists and can be freed or reused once this | |
496 | * function returns. | |
497 | * | |
498 | * The packet type might still be in use by receivers | |
499 | * and must not be freed until after all the CPU's have gone | |
500 | * through a quiescent state. | |
501 | */ | |
1d143d9f | 502 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
503 | { |
504 | struct list_head *head = &offload_base; | |
505 | struct packet_offload *po1; | |
506 | ||
c53aa505 | 507 | spin_lock(&offload_lock); |
62532da9 VY |
508 | |
509 | list_for_each_entry(po1, head, list) { | |
510 | if (po == po1) { | |
511 | list_del_rcu(&po->list); | |
512 | goto out; | |
513 | } | |
514 | } | |
515 | ||
516 | pr_warn("dev_remove_offload: %p not found\n", po); | |
517 | out: | |
c53aa505 | 518 | spin_unlock(&offload_lock); |
62532da9 | 519 | } |
62532da9 VY |
520 | |
521 | /** | |
522 | * dev_remove_offload - remove packet offload handler | |
523 | * @po: packet offload declaration | |
524 | * | |
525 | * Remove a packet offload handler that was previously added to the kernel | |
526 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
527 | * removed from the kernel lists and can be freed or reused once this | |
528 | * function returns. | |
529 | * | |
530 | * This call sleeps to guarantee that no CPU is looking at the packet | |
531 | * type after return. | |
532 | */ | |
533 | void dev_remove_offload(struct packet_offload *po) | |
534 | { | |
535 | __dev_remove_offload(po); | |
536 | ||
537 | synchronize_net(); | |
538 | } | |
539 | EXPORT_SYMBOL(dev_remove_offload); | |
540 | ||
1da177e4 LT |
541 | /****************************************************************************** |
542 | ||
543 | Device Boot-time Settings Routines | |
544 | ||
545 | *******************************************************************************/ | |
546 | ||
547 | /* Boot time configuration table */ | |
548 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
549 | ||
550 | /** | |
551 | * netdev_boot_setup_add - add new setup entry | |
552 | * @name: name of the device | |
553 | * @map: configured settings for the device | |
554 | * | |
555 | * Adds new setup entry to the dev_boot_setup list. The function | |
556 | * returns 0 on error and 1 on success. This is a generic routine to | |
557 | * all netdevices. | |
558 | */ | |
559 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
560 | { | |
561 | struct netdev_boot_setup *s; | |
562 | int i; | |
563 | ||
564 | s = dev_boot_setup; | |
565 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
566 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
567 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 568 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
569 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
570 | break; | |
571 | } | |
572 | } | |
573 | ||
574 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
575 | } | |
576 | ||
577 | /** | |
578 | * netdev_boot_setup_check - check boot time settings | |
579 | * @dev: the netdevice | |
580 | * | |
581 | * Check boot time settings for the device. | |
582 | * The found settings are set for the device to be used | |
583 | * later in the device probing. | |
584 | * Returns 0 if no settings found, 1 if they are. | |
585 | */ | |
586 | int netdev_boot_setup_check(struct net_device *dev) | |
587 | { | |
588 | struct netdev_boot_setup *s = dev_boot_setup; | |
589 | int i; | |
590 | ||
591 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
592 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 593 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
594 | dev->irq = s[i].map.irq; |
595 | dev->base_addr = s[i].map.base_addr; | |
596 | dev->mem_start = s[i].map.mem_start; | |
597 | dev->mem_end = s[i].map.mem_end; | |
598 | return 1; | |
599 | } | |
600 | } | |
601 | return 0; | |
602 | } | |
d1b19dff | 603 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
604 | |
605 | ||
606 | /** | |
607 | * netdev_boot_base - get address from boot time settings | |
608 | * @prefix: prefix for network device | |
609 | * @unit: id for network device | |
610 | * | |
611 | * Check boot time settings for the base address of device. | |
612 | * The found settings are set for the device to be used | |
613 | * later in the device probing. | |
614 | * Returns 0 if no settings found. | |
615 | */ | |
616 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
617 | { | |
618 | const struct netdev_boot_setup *s = dev_boot_setup; | |
619 | char name[IFNAMSIZ]; | |
620 | int i; | |
621 | ||
622 | sprintf(name, "%s%d", prefix, unit); | |
623 | ||
624 | /* | |
625 | * If device already registered then return base of 1 | |
626 | * to indicate not to probe for this interface | |
627 | */ | |
881d966b | 628 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
629 | return 1; |
630 | ||
631 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
632 | if (!strcmp(name, s[i].name)) | |
633 | return s[i].map.base_addr; | |
634 | return 0; | |
635 | } | |
636 | ||
637 | /* | |
638 | * Saves at boot time configured settings for any netdevice. | |
639 | */ | |
640 | int __init netdev_boot_setup(char *str) | |
641 | { | |
642 | int ints[5]; | |
643 | struct ifmap map; | |
644 | ||
645 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
646 | if (!str || !*str) | |
647 | return 0; | |
648 | ||
649 | /* Save settings */ | |
650 | memset(&map, 0, sizeof(map)); | |
651 | if (ints[0] > 0) | |
652 | map.irq = ints[1]; | |
653 | if (ints[0] > 1) | |
654 | map.base_addr = ints[2]; | |
655 | if (ints[0] > 2) | |
656 | map.mem_start = ints[3]; | |
657 | if (ints[0] > 3) | |
658 | map.mem_end = ints[4]; | |
659 | ||
660 | /* Add new entry to the list */ | |
661 | return netdev_boot_setup_add(str, &map); | |
662 | } | |
663 | ||
664 | __setup("netdev=", netdev_boot_setup); | |
665 | ||
666 | /******************************************************************************* | |
667 | ||
668 | Device Interface Subroutines | |
669 | ||
670 | *******************************************************************************/ | |
671 | ||
a54acb3a ND |
672 | /** |
673 | * dev_get_iflink - get 'iflink' value of a interface | |
674 | * @dev: targeted interface | |
675 | * | |
676 | * Indicates the ifindex the interface is linked to. | |
677 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
678 | */ | |
679 | ||
680 | int dev_get_iflink(const struct net_device *dev) | |
681 | { | |
682 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
683 | return dev->netdev_ops->ndo_get_iflink(dev); | |
684 | ||
7a66bbc9 | 685 | return dev->ifindex; |
a54acb3a ND |
686 | } |
687 | EXPORT_SYMBOL(dev_get_iflink); | |
688 | ||
fc4099f1 PS |
689 | /** |
690 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
691 | * @dev: targeted interface | |
692 | * @skb: The packet. | |
693 | * | |
694 | * For better visibility of tunnel traffic OVS needs to retrieve | |
695 | * egress tunnel information for a packet. Following API allows | |
696 | * user to get this info. | |
697 | */ | |
698 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
699 | { | |
700 | struct ip_tunnel_info *info; | |
701 | ||
702 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
703 | return -EINVAL; | |
704 | ||
705 | info = skb_tunnel_info_unclone(skb); | |
706 | if (!info) | |
707 | return -ENOMEM; | |
708 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
709 | return -EINVAL; | |
710 | ||
711 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
712 | } | |
713 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
714 | ||
1da177e4 LT |
715 | /** |
716 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 717 | * @net: the applicable net namespace |
1da177e4 LT |
718 | * @name: name to find |
719 | * | |
720 | * Find an interface by name. Must be called under RTNL semaphore | |
721 | * or @dev_base_lock. If the name is found a pointer to the device | |
722 | * is returned. If the name is not found then %NULL is returned. The | |
723 | * reference counters are not incremented so the caller must be | |
724 | * careful with locks. | |
725 | */ | |
726 | ||
881d966b | 727 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 728 | { |
0bd8d536 ED |
729 | struct net_device *dev; |
730 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 731 | |
b67bfe0d | 732 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
733 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
734 | return dev; | |
0bd8d536 | 735 | |
1da177e4 LT |
736 | return NULL; |
737 | } | |
d1b19dff | 738 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 739 | |
72c9528b ED |
740 | /** |
741 | * dev_get_by_name_rcu - find a device by its name | |
742 | * @net: the applicable net namespace | |
743 | * @name: name to find | |
744 | * | |
745 | * Find an interface by name. | |
746 | * If the name is found a pointer to the device is returned. | |
747 | * If the name is not found then %NULL is returned. | |
748 | * The reference counters are not incremented so the caller must be | |
749 | * careful with locks. The caller must hold RCU lock. | |
750 | */ | |
751 | ||
752 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
753 | { | |
72c9528b ED |
754 | struct net_device *dev; |
755 | struct hlist_head *head = dev_name_hash(net, name); | |
756 | ||
b67bfe0d | 757 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
758 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
759 | return dev; | |
760 | ||
761 | return NULL; | |
762 | } | |
763 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
764 | ||
1da177e4 LT |
765 | /** |
766 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 767 | * @net: the applicable net namespace |
1da177e4 LT |
768 | * @name: name to find |
769 | * | |
770 | * Find an interface by name. This can be called from any | |
771 | * context and does its own locking. The returned handle has | |
772 | * the usage count incremented and the caller must use dev_put() to | |
773 | * release it when it is no longer needed. %NULL is returned if no | |
774 | * matching device is found. | |
775 | */ | |
776 | ||
881d966b | 777 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
778 | { |
779 | struct net_device *dev; | |
780 | ||
72c9528b ED |
781 | rcu_read_lock(); |
782 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
783 | if (dev) |
784 | dev_hold(dev); | |
72c9528b | 785 | rcu_read_unlock(); |
1da177e4 LT |
786 | return dev; |
787 | } | |
d1b19dff | 788 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
789 | |
790 | /** | |
791 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 792 | * @net: the applicable net namespace |
1da177e4 LT |
793 | * @ifindex: index of device |
794 | * | |
795 | * Search for an interface by index. Returns %NULL if the device | |
796 | * is not found or a pointer to the device. The device has not | |
797 | * had its reference counter increased so the caller must be careful | |
798 | * about locking. The caller must hold either the RTNL semaphore | |
799 | * or @dev_base_lock. | |
800 | */ | |
801 | ||
881d966b | 802 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 803 | { |
0bd8d536 ED |
804 | struct net_device *dev; |
805 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 806 | |
b67bfe0d | 807 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
808 | if (dev->ifindex == ifindex) |
809 | return dev; | |
0bd8d536 | 810 | |
1da177e4 LT |
811 | return NULL; |
812 | } | |
d1b19dff | 813 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 814 | |
fb699dfd ED |
815 | /** |
816 | * dev_get_by_index_rcu - find a device by its ifindex | |
817 | * @net: the applicable net namespace | |
818 | * @ifindex: index of device | |
819 | * | |
820 | * Search for an interface by index. Returns %NULL if the device | |
821 | * is not found or a pointer to the device. The device has not | |
822 | * had its reference counter increased so the caller must be careful | |
823 | * about locking. The caller must hold RCU lock. | |
824 | */ | |
825 | ||
826 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
827 | { | |
fb699dfd ED |
828 | struct net_device *dev; |
829 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
830 | ||
b67bfe0d | 831 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
832 | if (dev->ifindex == ifindex) |
833 | return dev; | |
834 | ||
835 | return NULL; | |
836 | } | |
837 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
838 | ||
1da177e4 LT |
839 | |
840 | /** | |
841 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 842 | * @net: the applicable net namespace |
1da177e4 LT |
843 | * @ifindex: index of device |
844 | * | |
845 | * Search for an interface by index. Returns NULL if the device | |
846 | * is not found or a pointer to the device. The device returned has | |
847 | * had a reference added and the pointer is safe until the user calls | |
848 | * dev_put to indicate they have finished with it. | |
849 | */ | |
850 | ||
881d966b | 851 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
852 | { |
853 | struct net_device *dev; | |
854 | ||
fb699dfd ED |
855 | rcu_read_lock(); |
856 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
857 | if (dev) |
858 | dev_hold(dev); | |
fb699dfd | 859 | rcu_read_unlock(); |
1da177e4 LT |
860 | return dev; |
861 | } | |
d1b19dff | 862 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 863 | |
5dbe7c17 NS |
864 | /** |
865 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
866 | * @net: network namespace | |
867 | * @name: a pointer to the buffer where the name will be stored. | |
868 | * @ifindex: the ifindex of the interface to get the name from. | |
869 | * | |
870 | * The use of raw_seqcount_begin() and cond_resched() before | |
871 | * retrying is required as we want to give the writers a chance | |
872 | * to complete when CONFIG_PREEMPT is not set. | |
873 | */ | |
874 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
875 | { | |
876 | struct net_device *dev; | |
877 | unsigned int seq; | |
878 | ||
879 | retry: | |
880 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
881 | rcu_read_lock(); | |
882 | dev = dev_get_by_index_rcu(net, ifindex); | |
883 | if (!dev) { | |
884 | rcu_read_unlock(); | |
885 | return -ENODEV; | |
886 | } | |
887 | ||
888 | strcpy(name, dev->name); | |
889 | rcu_read_unlock(); | |
890 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
891 | cond_resched(); | |
892 | goto retry; | |
893 | } | |
894 | ||
895 | return 0; | |
896 | } | |
897 | ||
1da177e4 | 898 | /** |
941666c2 | 899 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 900 | * @net: the applicable net namespace |
1da177e4 LT |
901 | * @type: media type of device |
902 | * @ha: hardware address | |
903 | * | |
904 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
905 | * is not found or a pointer to the device. |
906 | * The caller must hold RCU or RTNL. | |
941666c2 | 907 | * The returned device has not had its ref count increased |
1da177e4 LT |
908 | * and the caller must therefore be careful about locking |
909 | * | |
1da177e4 LT |
910 | */ |
911 | ||
941666c2 ED |
912 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
913 | const char *ha) | |
1da177e4 LT |
914 | { |
915 | struct net_device *dev; | |
916 | ||
941666c2 | 917 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
918 | if (dev->type == type && |
919 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
920 | return dev; |
921 | ||
922 | return NULL; | |
1da177e4 | 923 | } |
941666c2 | 924 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 925 | |
881d966b | 926 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
927 | { |
928 | struct net_device *dev; | |
929 | ||
4e9cac2b | 930 | ASSERT_RTNL(); |
881d966b | 931 | for_each_netdev(net, dev) |
4e9cac2b | 932 | if (dev->type == type) |
7562f876 PE |
933 | return dev; |
934 | ||
935 | return NULL; | |
4e9cac2b | 936 | } |
4e9cac2b PM |
937 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
938 | ||
881d966b | 939 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 940 | { |
99fe3c39 | 941 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 942 | |
99fe3c39 ED |
943 | rcu_read_lock(); |
944 | for_each_netdev_rcu(net, dev) | |
945 | if (dev->type == type) { | |
946 | dev_hold(dev); | |
947 | ret = dev; | |
948 | break; | |
949 | } | |
950 | rcu_read_unlock(); | |
951 | return ret; | |
1da177e4 | 952 | } |
1da177e4 LT |
953 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
954 | ||
955 | /** | |
6c555490 | 956 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 957 | * @net: the applicable net namespace |
1da177e4 LT |
958 | * @if_flags: IFF_* values |
959 | * @mask: bitmask of bits in if_flags to check | |
960 | * | |
961 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 962 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 963 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
964 | */ |
965 | ||
6c555490 WC |
966 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
967 | unsigned short mask) | |
1da177e4 | 968 | { |
7562f876 | 969 | struct net_device *dev, *ret; |
1da177e4 | 970 | |
6c555490 WC |
971 | ASSERT_RTNL(); |
972 | ||
7562f876 | 973 | ret = NULL; |
6c555490 | 974 | for_each_netdev(net, dev) { |
1da177e4 | 975 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 976 | ret = dev; |
1da177e4 LT |
977 | break; |
978 | } | |
979 | } | |
7562f876 | 980 | return ret; |
1da177e4 | 981 | } |
6c555490 | 982 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
983 | |
984 | /** | |
985 | * dev_valid_name - check if name is okay for network device | |
986 | * @name: name string | |
987 | * | |
988 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
989 | * to allow sysfs to work. We also disallow any kind of |
990 | * whitespace. | |
1da177e4 | 991 | */ |
95f050bf | 992 | bool dev_valid_name(const char *name) |
1da177e4 | 993 | { |
c7fa9d18 | 994 | if (*name == '\0') |
95f050bf | 995 | return false; |
b6fe17d6 | 996 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 997 | return false; |
c7fa9d18 | 998 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 999 | return false; |
c7fa9d18 DM |
1000 | |
1001 | while (*name) { | |
a4176a93 | 1002 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1003 | return false; |
c7fa9d18 DM |
1004 | name++; |
1005 | } | |
95f050bf | 1006 | return true; |
1da177e4 | 1007 | } |
d1b19dff | 1008 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1009 | |
1010 | /** | |
b267b179 EB |
1011 | * __dev_alloc_name - allocate a name for a device |
1012 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1013 | * @name: name format string |
b267b179 | 1014 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1015 | * |
1016 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1017 | * id. It scans list of devices to build up a free map, then chooses |
1018 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1019 | * while allocating the name and adding the device in order to avoid | |
1020 | * duplicates. | |
1021 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1022 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1023 | */ |
1024 | ||
b267b179 | 1025 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1026 | { |
1027 | int i = 0; | |
1da177e4 LT |
1028 | const char *p; |
1029 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1030 | unsigned long *inuse; |
1da177e4 LT |
1031 | struct net_device *d; |
1032 | ||
1033 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
1034 | if (p) { | |
1035 | /* | |
1036 | * Verify the string as this thing may have come from | |
1037 | * the user. There must be either one "%d" and no other "%" | |
1038 | * characters. | |
1039 | */ | |
1040 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1041 | return -EINVAL; | |
1042 | ||
1043 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1044 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1045 | if (!inuse) |
1046 | return -ENOMEM; | |
1047 | ||
881d966b | 1048 | for_each_netdev(net, d) { |
1da177e4 LT |
1049 | if (!sscanf(d->name, name, &i)) |
1050 | continue; | |
1051 | if (i < 0 || i >= max_netdevices) | |
1052 | continue; | |
1053 | ||
1054 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1055 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1056 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1057 | set_bit(i, inuse); | |
1058 | } | |
1059 | ||
1060 | i = find_first_zero_bit(inuse, max_netdevices); | |
1061 | free_page((unsigned long) inuse); | |
1062 | } | |
1063 | ||
d9031024 OP |
1064 | if (buf != name) |
1065 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1066 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1067 | return i; |
1da177e4 LT |
1068 | |
1069 | /* It is possible to run out of possible slots | |
1070 | * when the name is long and there isn't enough space left | |
1071 | * for the digits, or if all bits are used. | |
1072 | */ | |
1073 | return -ENFILE; | |
1074 | } | |
1075 | ||
b267b179 EB |
1076 | /** |
1077 | * dev_alloc_name - allocate a name for a device | |
1078 | * @dev: device | |
1079 | * @name: name format string | |
1080 | * | |
1081 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1082 | * id. It scans list of devices to build up a free map, then chooses | |
1083 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1084 | * while allocating the name and adding the device in order to avoid | |
1085 | * duplicates. | |
1086 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1087 | * Returns the number of the unit assigned or a negative errno code. | |
1088 | */ | |
1089 | ||
1090 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1091 | { | |
1092 | char buf[IFNAMSIZ]; | |
1093 | struct net *net; | |
1094 | int ret; | |
1095 | ||
c346dca1 YH |
1096 | BUG_ON(!dev_net(dev)); |
1097 | net = dev_net(dev); | |
b267b179 EB |
1098 | ret = __dev_alloc_name(net, name, buf); |
1099 | if (ret >= 0) | |
1100 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1101 | return ret; | |
1102 | } | |
d1b19dff | 1103 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1104 | |
828de4f6 G |
1105 | static int dev_alloc_name_ns(struct net *net, |
1106 | struct net_device *dev, | |
1107 | const char *name) | |
d9031024 | 1108 | { |
828de4f6 G |
1109 | char buf[IFNAMSIZ]; |
1110 | int ret; | |
8ce6cebc | 1111 | |
828de4f6 G |
1112 | ret = __dev_alloc_name(net, name, buf); |
1113 | if (ret >= 0) | |
1114 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1115 | return ret; | |
1116 | } | |
1117 | ||
1118 | static int dev_get_valid_name(struct net *net, | |
1119 | struct net_device *dev, | |
1120 | const char *name) | |
1121 | { | |
1122 | BUG_ON(!net); | |
8ce6cebc | 1123 | |
d9031024 OP |
1124 | if (!dev_valid_name(name)) |
1125 | return -EINVAL; | |
1126 | ||
1c5cae81 | 1127 | if (strchr(name, '%')) |
828de4f6 | 1128 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1129 | else if (__dev_get_by_name(net, name)) |
1130 | return -EEXIST; | |
8ce6cebc DL |
1131 | else if (dev->name != name) |
1132 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1133 | |
1134 | return 0; | |
1135 | } | |
1da177e4 LT |
1136 | |
1137 | /** | |
1138 | * dev_change_name - change name of a device | |
1139 | * @dev: device | |
1140 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1141 | * | |
1142 | * Change name of a device, can pass format strings "eth%d". | |
1143 | * for wildcarding. | |
1144 | */ | |
cf04a4c7 | 1145 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1146 | { |
238fa362 | 1147 | unsigned char old_assign_type; |
fcc5a03a | 1148 | char oldname[IFNAMSIZ]; |
1da177e4 | 1149 | int err = 0; |
fcc5a03a | 1150 | int ret; |
881d966b | 1151 | struct net *net; |
1da177e4 LT |
1152 | |
1153 | ASSERT_RTNL(); | |
c346dca1 | 1154 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1155 | |
c346dca1 | 1156 | net = dev_net(dev); |
1da177e4 LT |
1157 | if (dev->flags & IFF_UP) |
1158 | return -EBUSY; | |
1159 | ||
30e6c9fa | 1160 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1161 | |
1162 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1163 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1164 | return 0; |
c91f6df2 | 1165 | } |
c8d90dca | 1166 | |
fcc5a03a HX |
1167 | memcpy(oldname, dev->name, IFNAMSIZ); |
1168 | ||
828de4f6 | 1169 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1170 | if (err < 0) { |
30e6c9fa | 1171 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1172 | return err; |
c91f6df2 | 1173 | } |
1da177e4 | 1174 | |
6fe82a39 VF |
1175 | if (oldname[0] && !strchr(oldname, '%')) |
1176 | netdev_info(dev, "renamed from %s\n", oldname); | |
1177 | ||
238fa362 TG |
1178 | old_assign_type = dev->name_assign_type; |
1179 | dev->name_assign_type = NET_NAME_RENAMED; | |
1180 | ||
fcc5a03a | 1181 | rollback: |
a1b3f594 EB |
1182 | ret = device_rename(&dev->dev, dev->name); |
1183 | if (ret) { | |
1184 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1185 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1186 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1187 | return ret; |
dcc99773 | 1188 | } |
7f988eab | 1189 | |
30e6c9fa | 1190 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1191 | |
5bb025fa VF |
1192 | netdev_adjacent_rename_links(dev, oldname); |
1193 | ||
7f988eab | 1194 | write_lock_bh(&dev_base_lock); |
372b2312 | 1195 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1196 | write_unlock_bh(&dev_base_lock); |
1197 | ||
1198 | synchronize_rcu(); | |
1199 | ||
1200 | write_lock_bh(&dev_base_lock); | |
1201 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1202 | write_unlock_bh(&dev_base_lock); |
1203 | ||
056925ab | 1204 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1205 | ret = notifier_to_errno(ret); |
1206 | ||
1207 | if (ret) { | |
91e9c07b ED |
1208 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1209 | if (err >= 0) { | |
fcc5a03a | 1210 | err = ret; |
30e6c9fa | 1211 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1212 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1213 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1214 | dev->name_assign_type = old_assign_type; |
1215 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1216 | goto rollback; |
91e9c07b | 1217 | } else { |
7b6cd1ce | 1218 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1219 | dev->name, ret); |
fcc5a03a HX |
1220 | } |
1221 | } | |
1da177e4 LT |
1222 | |
1223 | return err; | |
1224 | } | |
1225 | ||
0b815a1a SH |
1226 | /** |
1227 | * dev_set_alias - change ifalias of a device | |
1228 | * @dev: device | |
1229 | * @alias: name up to IFALIASZ | |
f0db275a | 1230 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1231 | * |
1232 | * Set ifalias for a device, | |
1233 | */ | |
1234 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1235 | { | |
7364e445 AK |
1236 | char *new_ifalias; |
1237 | ||
0b815a1a SH |
1238 | ASSERT_RTNL(); |
1239 | ||
1240 | if (len >= IFALIASZ) | |
1241 | return -EINVAL; | |
1242 | ||
96ca4a2c | 1243 | if (!len) { |
388dfc2d SK |
1244 | kfree(dev->ifalias); |
1245 | dev->ifalias = NULL; | |
96ca4a2c OH |
1246 | return 0; |
1247 | } | |
1248 | ||
7364e445 AK |
1249 | new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
1250 | if (!new_ifalias) | |
0b815a1a | 1251 | return -ENOMEM; |
7364e445 | 1252 | dev->ifalias = new_ifalias; |
0b815a1a SH |
1253 | |
1254 | strlcpy(dev->ifalias, alias, len+1); | |
1255 | return len; | |
1256 | } | |
1257 | ||
1258 | ||
d8a33ac4 | 1259 | /** |
3041a069 | 1260 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1261 | * @dev: device to cause notification |
1262 | * | |
1263 | * Called to indicate a device has changed features. | |
1264 | */ | |
1265 | void netdev_features_change(struct net_device *dev) | |
1266 | { | |
056925ab | 1267 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1268 | } |
1269 | EXPORT_SYMBOL(netdev_features_change); | |
1270 | ||
1da177e4 LT |
1271 | /** |
1272 | * netdev_state_change - device changes state | |
1273 | * @dev: device to cause notification | |
1274 | * | |
1275 | * Called to indicate a device has changed state. This function calls | |
1276 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1277 | * to the routing socket. | |
1278 | */ | |
1279 | void netdev_state_change(struct net_device *dev) | |
1280 | { | |
1281 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1282 | struct netdev_notifier_change_info change_info; |
1283 | ||
1284 | change_info.flags_changed = 0; | |
1285 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1286 | &change_info.info); | |
7f294054 | 1287 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1288 | } |
1289 | } | |
d1b19dff | 1290 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1291 | |
ee89bab1 AW |
1292 | /** |
1293 | * netdev_notify_peers - notify network peers about existence of @dev | |
1294 | * @dev: network device | |
1295 | * | |
1296 | * Generate traffic such that interested network peers are aware of | |
1297 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1298 | * a device wants to inform the rest of the network about some sort of | |
1299 | * reconfiguration such as a failover event or virtual machine | |
1300 | * migration. | |
1301 | */ | |
1302 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1303 | { |
ee89bab1 AW |
1304 | rtnl_lock(); |
1305 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
1306 | rtnl_unlock(); | |
c1da4ac7 | 1307 | } |
ee89bab1 | 1308 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1309 | |
bd380811 | 1310 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1311 | { |
d314774c | 1312 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1313 | int ret; |
1da177e4 | 1314 | |
e46b66bc BH |
1315 | ASSERT_RTNL(); |
1316 | ||
1da177e4 LT |
1317 | if (!netif_device_present(dev)) |
1318 | return -ENODEV; | |
1319 | ||
ca99ca14 NH |
1320 | /* Block netpoll from trying to do any rx path servicing. |
1321 | * If we don't do this there is a chance ndo_poll_controller | |
1322 | * or ndo_poll may be running while we open the device | |
1323 | */ | |
66b5552f | 1324 | netpoll_poll_disable(dev); |
ca99ca14 | 1325 | |
3b8bcfd5 JB |
1326 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1327 | ret = notifier_to_errno(ret); | |
1328 | if (ret) | |
1329 | return ret; | |
1330 | ||
1da177e4 | 1331 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1332 | |
d314774c SH |
1333 | if (ops->ndo_validate_addr) |
1334 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1335 | |
d314774c SH |
1336 | if (!ret && ops->ndo_open) |
1337 | ret = ops->ndo_open(dev); | |
1da177e4 | 1338 | |
66b5552f | 1339 | netpoll_poll_enable(dev); |
ca99ca14 | 1340 | |
bada339b JG |
1341 | if (ret) |
1342 | clear_bit(__LINK_STATE_START, &dev->state); | |
1343 | else { | |
1da177e4 | 1344 | dev->flags |= IFF_UP; |
4417da66 | 1345 | dev_set_rx_mode(dev); |
1da177e4 | 1346 | dev_activate(dev); |
7bf23575 | 1347 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1348 | } |
bada339b | 1349 | |
1da177e4 LT |
1350 | return ret; |
1351 | } | |
1352 | ||
1353 | /** | |
bd380811 PM |
1354 | * dev_open - prepare an interface for use. |
1355 | * @dev: device to open | |
1da177e4 | 1356 | * |
bd380811 PM |
1357 | * Takes a device from down to up state. The device's private open |
1358 | * function is invoked and then the multicast lists are loaded. Finally | |
1359 | * the device is moved into the up state and a %NETDEV_UP message is | |
1360 | * sent to the netdev notifier chain. | |
1361 | * | |
1362 | * Calling this function on an active interface is a nop. On a failure | |
1363 | * a negative errno code is returned. | |
1da177e4 | 1364 | */ |
bd380811 PM |
1365 | int dev_open(struct net_device *dev) |
1366 | { | |
1367 | int ret; | |
1368 | ||
bd380811 PM |
1369 | if (dev->flags & IFF_UP) |
1370 | return 0; | |
1371 | ||
bd380811 PM |
1372 | ret = __dev_open(dev); |
1373 | if (ret < 0) | |
1374 | return ret; | |
1375 | ||
7f294054 | 1376 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1377 | call_netdevice_notifiers(NETDEV_UP, dev); |
1378 | ||
1379 | return ret; | |
1380 | } | |
1381 | EXPORT_SYMBOL(dev_open); | |
1382 | ||
44345724 | 1383 | static int __dev_close_many(struct list_head *head) |
1da177e4 | 1384 | { |
44345724 | 1385 | struct net_device *dev; |
e46b66bc | 1386 | |
bd380811 | 1387 | ASSERT_RTNL(); |
9d5010db DM |
1388 | might_sleep(); |
1389 | ||
5cde2829 | 1390 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1391 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1392 | netpoll_poll_disable(dev); |
3f4df206 | 1393 | |
44345724 | 1394 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1395 | |
44345724 | 1396 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1397 | |
44345724 OP |
1398 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1399 | * can be even on different cpu. So just clear netif_running(). | |
1400 | * | |
1401 | * dev->stop() will invoke napi_disable() on all of it's | |
1402 | * napi_struct instances on this device. | |
1403 | */ | |
4e857c58 | 1404 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1405 | } |
1da177e4 | 1406 | |
44345724 | 1407 | dev_deactivate_many(head); |
d8b2a4d2 | 1408 | |
5cde2829 | 1409 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1410 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1411 | |
44345724 OP |
1412 | /* |
1413 | * Call the device specific close. This cannot fail. | |
1414 | * Only if device is UP | |
1415 | * | |
1416 | * We allow it to be called even after a DETACH hot-plug | |
1417 | * event. | |
1418 | */ | |
1419 | if (ops->ndo_stop) | |
1420 | ops->ndo_stop(dev); | |
1421 | ||
44345724 | 1422 | dev->flags &= ~IFF_UP; |
66b5552f | 1423 | netpoll_poll_enable(dev); |
44345724 OP |
1424 | } |
1425 | ||
1426 | return 0; | |
1427 | } | |
1428 | ||
1429 | static int __dev_close(struct net_device *dev) | |
1430 | { | |
f87e6f47 | 1431 | int retval; |
44345724 OP |
1432 | LIST_HEAD(single); |
1433 | ||
5cde2829 | 1434 | list_add(&dev->close_list, &single); |
f87e6f47 LT |
1435 | retval = __dev_close_many(&single); |
1436 | list_del(&single); | |
ca99ca14 | 1437 | |
f87e6f47 | 1438 | return retval; |
44345724 OP |
1439 | } |
1440 | ||
99c4a26a | 1441 | int dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1442 | { |
1443 | struct net_device *dev, *tmp; | |
1da177e4 | 1444 | |
5cde2829 EB |
1445 | /* Remove the devices that don't need to be closed */ |
1446 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1447 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1448 | list_del_init(&dev->close_list); |
44345724 OP |
1449 | |
1450 | __dev_close_many(head); | |
1da177e4 | 1451 | |
5cde2829 | 1452 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1453 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1454 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1455 | if (unlink) |
1456 | list_del_init(&dev->close_list); | |
44345724 | 1457 | } |
bd380811 PM |
1458 | |
1459 | return 0; | |
1460 | } | |
99c4a26a | 1461 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1462 | |
1463 | /** | |
1464 | * dev_close - shutdown an interface. | |
1465 | * @dev: device to shutdown | |
1466 | * | |
1467 | * This function moves an active device into down state. A | |
1468 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1469 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1470 | * chain. | |
1471 | */ | |
1472 | int dev_close(struct net_device *dev) | |
1473 | { | |
e14a5993 ED |
1474 | if (dev->flags & IFF_UP) { |
1475 | LIST_HEAD(single); | |
1da177e4 | 1476 | |
5cde2829 | 1477 | list_add(&dev->close_list, &single); |
99c4a26a | 1478 | dev_close_many(&single, true); |
e14a5993 ED |
1479 | list_del(&single); |
1480 | } | |
da6e378b | 1481 | return 0; |
1da177e4 | 1482 | } |
d1b19dff | 1483 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1484 | |
1485 | ||
0187bdfb BH |
1486 | /** |
1487 | * dev_disable_lro - disable Large Receive Offload on a device | |
1488 | * @dev: device | |
1489 | * | |
1490 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1491 | * called under RTNL. This is needed if received packets may be | |
1492 | * forwarded to another interface. | |
1493 | */ | |
1494 | void dev_disable_lro(struct net_device *dev) | |
1495 | { | |
fbe168ba MK |
1496 | struct net_device *lower_dev; |
1497 | struct list_head *iter; | |
529d0489 | 1498 | |
bc5787c6 MM |
1499 | dev->wanted_features &= ~NETIF_F_LRO; |
1500 | netdev_update_features(dev); | |
27660515 | 1501 | |
22d5969f MM |
1502 | if (unlikely(dev->features & NETIF_F_LRO)) |
1503 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1504 | |
1505 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1506 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1507 | } |
1508 | EXPORT_SYMBOL(dev_disable_lro); | |
1509 | ||
351638e7 JP |
1510 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1511 | struct net_device *dev) | |
1512 | { | |
1513 | struct netdev_notifier_info info; | |
1514 | ||
1515 | netdev_notifier_info_init(&info, dev); | |
1516 | return nb->notifier_call(nb, val, &info); | |
1517 | } | |
0187bdfb | 1518 | |
881d966b EB |
1519 | static int dev_boot_phase = 1; |
1520 | ||
1da177e4 LT |
1521 | /** |
1522 | * register_netdevice_notifier - register a network notifier block | |
1523 | * @nb: notifier | |
1524 | * | |
1525 | * Register a notifier to be called when network device events occur. | |
1526 | * The notifier passed is linked into the kernel structures and must | |
1527 | * not be reused until it has been unregistered. A negative errno code | |
1528 | * is returned on a failure. | |
1529 | * | |
1530 | * When registered all registration and up events are replayed | |
4ec93edb | 1531 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1532 | * view of the network device list. |
1533 | */ | |
1534 | ||
1535 | int register_netdevice_notifier(struct notifier_block *nb) | |
1536 | { | |
1537 | struct net_device *dev; | |
fcc5a03a | 1538 | struct net_device *last; |
881d966b | 1539 | struct net *net; |
1da177e4 LT |
1540 | int err; |
1541 | ||
1542 | rtnl_lock(); | |
f07d5b94 | 1543 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1544 | if (err) |
1545 | goto unlock; | |
881d966b EB |
1546 | if (dev_boot_phase) |
1547 | goto unlock; | |
1548 | for_each_net(net) { | |
1549 | for_each_netdev(net, dev) { | |
351638e7 | 1550 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1551 | err = notifier_to_errno(err); |
1552 | if (err) | |
1553 | goto rollback; | |
1554 | ||
1555 | if (!(dev->flags & IFF_UP)) | |
1556 | continue; | |
1da177e4 | 1557 | |
351638e7 | 1558 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1559 | } |
1da177e4 | 1560 | } |
fcc5a03a HX |
1561 | |
1562 | unlock: | |
1da177e4 LT |
1563 | rtnl_unlock(); |
1564 | return err; | |
fcc5a03a HX |
1565 | |
1566 | rollback: | |
1567 | last = dev; | |
881d966b EB |
1568 | for_each_net(net) { |
1569 | for_each_netdev(net, dev) { | |
1570 | if (dev == last) | |
8f891489 | 1571 | goto outroll; |
fcc5a03a | 1572 | |
881d966b | 1573 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1574 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1575 | dev); | |
1576 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1577 | } |
351638e7 | 1578 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1579 | } |
fcc5a03a | 1580 | } |
c67625a1 | 1581 | |
8f891489 | 1582 | outroll: |
c67625a1 | 1583 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1584 | goto unlock; |
1da177e4 | 1585 | } |
d1b19dff | 1586 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1587 | |
1588 | /** | |
1589 | * unregister_netdevice_notifier - unregister a network notifier block | |
1590 | * @nb: notifier | |
1591 | * | |
1592 | * Unregister a notifier previously registered by | |
1593 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1594 | * kernel structures and may then be reused. A negative errno code | |
1595 | * is returned on a failure. | |
7d3d43da EB |
1596 | * |
1597 | * After unregistering unregister and down device events are synthesized | |
1598 | * for all devices on the device list to the removed notifier to remove | |
1599 | * the need for special case cleanup code. | |
1da177e4 LT |
1600 | */ |
1601 | ||
1602 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1603 | { | |
7d3d43da EB |
1604 | struct net_device *dev; |
1605 | struct net *net; | |
9f514950 HX |
1606 | int err; |
1607 | ||
1608 | rtnl_lock(); | |
f07d5b94 | 1609 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1610 | if (err) |
1611 | goto unlock; | |
1612 | ||
1613 | for_each_net(net) { | |
1614 | for_each_netdev(net, dev) { | |
1615 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1616 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1617 | dev); | |
1618 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1619 | } |
351638e7 | 1620 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1621 | } |
1622 | } | |
1623 | unlock: | |
9f514950 HX |
1624 | rtnl_unlock(); |
1625 | return err; | |
1da177e4 | 1626 | } |
d1b19dff | 1627 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1628 | |
351638e7 JP |
1629 | /** |
1630 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1631 | * @val: value passed unmodified to notifier function | |
1632 | * @dev: net_device pointer passed unmodified to notifier function | |
1633 | * @info: notifier information data | |
1634 | * | |
1635 | * Call all network notifier blocks. Parameters and return value | |
1636 | * are as for raw_notifier_call_chain(). | |
1637 | */ | |
1638 | ||
1d143d9f | 1639 | static int call_netdevice_notifiers_info(unsigned long val, |
1640 | struct net_device *dev, | |
1641 | struct netdev_notifier_info *info) | |
351638e7 JP |
1642 | { |
1643 | ASSERT_RTNL(); | |
1644 | netdev_notifier_info_init(info, dev); | |
1645 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1646 | } | |
351638e7 | 1647 | |
1da177e4 LT |
1648 | /** |
1649 | * call_netdevice_notifiers - call all network notifier blocks | |
1650 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1651 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1652 | * |
1653 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1654 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1655 | */ |
1656 | ||
ad7379d4 | 1657 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1658 | { |
351638e7 JP |
1659 | struct netdev_notifier_info info; |
1660 | ||
1661 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1662 | } |
edf947f1 | 1663 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1664 | |
1cf51900 | 1665 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
1666 | static struct static_key ingress_needed __read_mostly; |
1667 | ||
1668 | void net_inc_ingress_queue(void) | |
1669 | { | |
1670 | static_key_slow_inc(&ingress_needed); | |
1671 | } | |
1672 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1673 | ||
1674 | void net_dec_ingress_queue(void) | |
1675 | { | |
1676 | static_key_slow_dec(&ingress_needed); | |
1677 | } | |
1678 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1679 | #endif | |
1680 | ||
1f211a1b DB |
1681 | #ifdef CONFIG_NET_EGRESS |
1682 | static struct static_key egress_needed __read_mostly; | |
1683 | ||
1684 | void net_inc_egress_queue(void) | |
1685 | { | |
1686 | static_key_slow_inc(&egress_needed); | |
1687 | } | |
1688 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
1689 | ||
1690 | void net_dec_egress_queue(void) | |
1691 | { | |
1692 | static_key_slow_dec(&egress_needed); | |
1693 | } | |
1694 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
1695 | #endif | |
1696 | ||
c5905afb | 1697 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1698 | #ifdef HAVE_JUMP_LABEL |
c5905afb | 1699 | /* We are not allowed to call static_key_slow_dec() from irq context |
b90e5794 | 1700 | * If net_disable_timestamp() is called from irq context, defer the |
c5905afb | 1701 | * static_key_slow_dec() calls. |
b90e5794 ED |
1702 | */ |
1703 | static atomic_t netstamp_needed_deferred; | |
1704 | #endif | |
1da177e4 LT |
1705 | |
1706 | void net_enable_timestamp(void) | |
1707 | { | |
b90e5794 ED |
1708 | #ifdef HAVE_JUMP_LABEL |
1709 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); | |
1710 | ||
1711 | if (deferred) { | |
1712 | while (--deferred) | |
c5905afb | 1713 | static_key_slow_dec(&netstamp_needed); |
b90e5794 ED |
1714 | return; |
1715 | } | |
1716 | #endif | |
c5905afb | 1717 | static_key_slow_inc(&netstamp_needed); |
1da177e4 | 1718 | } |
d1b19dff | 1719 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1720 | |
1721 | void net_disable_timestamp(void) | |
1722 | { | |
b90e5794 ED |
1723 | #ifdef HAVE_JUMP_LABEL |
1724 | if (in_interrupt()) { | |
1725 | atomic_inc(&netstamp_needed_deferred); | |
1726 | return; | |
1727 | } | |
1728 | #endif | |
c5905afb | 1729 | static_key_slow_dec(&netstamp_needed); |
1da177e4 | 1730 | } |
d1b19dff | 1731 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1732 | |
3b098e2d | 1733 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1734 | { |
588f0330 | 1735 | skb->tstamp.tv64 = 0; |
c5905afb | 1736 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1737 | __net_timestamp(skb); |
1da177e4 LT |
1738 | } |
1739 | ||
588f0330 | 1740 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1741 | if (static_key_false(&netstamp_needed)) { \ |
588f0330 ED |
1742 | if ((COND) && !(SKB)->tstamp.tv64) \ |
1743 | __net_timestamp(SKB); \ | |
1744 | } \ | |
3b098e2d | 1745 | |
f4b05d27 | 1746 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
1747 | { |
1748 | unsigned int len; | |
1749 | ||
1750 | if (!(dev->flags & IFF_UP)) | |
1751 | return false; | |
1752 | ||
1753 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1754 | if (skb->len <= len) | |
1755 | return true; | |
1756 | ||
1757 | /* if TSO is enabled, we don't care about the length as the packet | |
1758 | * could be forwarded without being segmented before | |
1759 | */ | |
1760 | if (skb_is_gso(skb)) | |
1761 | return true; | |
1762 | ||
1763 | return false; | |
1764 | } | |
1ee481fb | 1765 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1766 | |
a0265d28 HX |
1767 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1768 | { | |
bbbf2df0 WB |
1769 | if (skb_orphan_frags(skb, GFP_ATOMIC) || |
1770 | unlikely(!is_skb_forwardable(dev, skb))) { | |
a0265d28 HX |
1771 | atomic_long_inc(&dev->rx_dropped); |
1772 | kfree_skb(skb); | |
1773 | return NET_RX_DROP; | |
1774 | } | |
1775 | ||
1776 | skb_scrub_packet(skb, true); | |
08b4b8ea | 1777 | skb->priority = 0; |
a0265d28 | 1778 | skb->protocol = eth_type_trans(skb, dev); |
2c26d34b | 1779 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); |
a0265d28 HX |
1780 | |
1781 | return 0; | |
1782 | } | |
1783 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1784 | ||
44540960 AB |
1785 | /** |
1786 | * dev_forward_skb - loopback an skb to another netif | |
1787 | * | |
1788 | * @dev: destination network device | |
1789 | * @skb: buffer to forward | |
1790 | * | |
1791 | * return values: | |
1792 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1793 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1794 | * |
1795 | * dev_forward_skb can be used for injecting an skb from the | |
1796 | * start_xmit function of one device into the receive queue | |
1797 | * of another device. | |
1798 | * | |
1799 | * The receiving device may be in another namespace, so | |
1800 | * we have to clear all information in the skb that could | |
1801 | * impact namespace isolation. | |
1802 | */ | |
1803 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1804 | { | |
a0265d28 | 1805 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1806 | } |
1807 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1808 | ||
71d9dec2 CG |
1809 | static inline int deliver_skb(struct sk_buff *skb, |
1810 | struct packet_type *pt_prev, | |
1811 | struct net_device *orig_dev) | |
1812 | { | |
1080e512 MT |
1813 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
1814 | return -ENOMEM; | |
71d9dec2 CG |
1815 | atomic_inc(&skb->users); |
1816 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1817 | } | |
1818 | ||
7866a621 SN |
1819 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1820 | struct packet_type **pt, | |
fbcb2170 JP |
1821 | struct net_device *orig_dev, |
1822 | __be16 type, | |
7866a621 SN |
1823 | struct list_head *ptype_list) |
1824 | { | |
1825 | struct packet_type *ptype, *pt_prev = *pt; | |
1826 | ||
1827 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1828 | if (ptype->type != type) | |
1829 | continue; | |
1830 | if (pt_prev) | |
fbcb2170 | 1831 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1832 | pt_prev = ptype; |
1833 | } | |
1834 | *pt = pt_prev; | |
1835 | } | |
1836 | ||
c0de08d0 EL |
1837 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1838 | { | |
a3d744e9 | 1839 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1840 | return false; |
1841 | ||
1842 | if (ptype->id_match) | |
1843 | return ptype->id_match(ptype, skb->sk); | |
1844 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1845 | return true; | |
1846 | ||
1847 | return false; | |
1848 | } | |
1849 | ||
1da177e4 LT |
1850 | /* |
1851 | * Support routine. Sends outgoing frames to any network | |
1852 | * taps currently in use. | |
1853 | */ | |
1854 | ||
74b20582 | 1855 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1856 | { |
1857 | struct packet_type *ptype; | |
71d9dec2 CG |
1858 | struct sk_buff *skb2 = NULL; |
1859 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1860 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1861 | |
1da177e4 | 1862 | rcu_read_lock(); |
7866a621 SN |
1863 | again: |
1864 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1865 | /* Never send packets back to the socket |
1866 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1867 | */ | |
7866a621 SN |
1868 | if (skb_loop_sk(ptype, skb)) |
1869 | continue; | |
71d9dec2 | 1870 | |
7866a621 SN |
1871 | if (pt_prev) { |
1872 | deliver_skb(skb2, pt_prev, skb->dev); | |
1873 | pt_prev = ptype; | |
1874 | continue; | |
1875 | } | |
1da177e4 | 1876 | |
7866a621 SN |
1877 | /* need to clone skb, done only once */ |
1878 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1879 | if (!skb2) | |
1880 | goto out_unlock; | |
70978182 | 1881 | |
7866a621 | 1882 | net_timestamp_set(skb2); |
1da177e4 | 1883 | |
7866a621 SN |
1884 | /* skb->nh should be correctly |
1885 | * set by sender, so that the second statement is | |
1886 | * just protection against buggy protocols. | |
1887 | */ | |
1888 | skb_reset_mac_header(skb2); | |
1889 | ||
1890 | if (skb_network_header(skb2) < skb2->data || | |
1891 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1892 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1893 | ntohs(skb2->protocol), | |
1894 | dev->name); | |
1895 | skb_reset_network_header(skb2); | |
1da177e4 | 1896 | } |
7866a621 SN |
1897 | |
1898 | skb2->transport_header = skb2->network_header; | |
1899 | skb2->pkt_type = PACKET_OUTGOING; | |
1900 | pt_prev = ptype; | |
1901 | } | |
1902 | ||
1903 | if (ptype_list == &ptype_all) { | |
1904 | ptype_list = &dev->ptype_all; | |
1905 | goto again; | |
1da177e4 | 1906 | } |
7866a621 | 1907 | out_unlock: |
71d9dec2 CG |
1908 | if (pt_prev) |
1909 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1da177e4 LT |
1910 | rcu_read_unlock(); |
1911 | } | |
74b20582 | 1912 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 1913 | |
2c53040f BH |
1914 | /** |
1915 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1916 | * @dev: Network device |
1917 | * @txq: number of queues available | |
1918 | * | |
1919 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1920 | * valid. To resolve this verify the tc mapping remains valid and if | |
1921 | * not NULL the mapping. With no priorities mapping to this | |
1922 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1923 | * is invalid nothing can be done so disable priority mappings. If is | |
1924 | * expected that drivers will fix this mapping if they can before | |
1925 | * calling netif_set_real_num_tx_queues. | |
1926 | */ | |
bb134d22 | 1927 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
1928 | { |
1929 | int i; | |
1930 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
1931 | ||
1932 | /* If TC0 is invalidated disable TC mapping */ | |
1933 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 1934 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
1935 | dev->num_tc = 0; |
1936 | return; | |
1937 | } | |
1938 | ||
1939 | /* Invalidated prio to tc mappings set to TC0 */ | |
1940 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
1941 | int q = netdev_get_prio_tc_map(dev, i); | |
1942 | ||
1943 | tc = &dev->tc_to_txq[q]; | |
1944 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
1945 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
1946 | i, q); | |
4f57c087 JF |
1947 | netdev_set_prio_tc_map(dev, i, 0); |
1948 | } | |
1949 | } | |
1950 | } | |
1951 | ||
537c00de AD |
1952 | #ifdef CONFIG_XPS |
1953 | static DEFINE_MUTEX(xps_map_mutex); | |
1954 | #define xmap_dereference(P) \ | |
1955 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
1956 | ||
10cdc3f3 AD |
1957 | static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps, |
1958 | int cpu, u16 index) | |
537c00de | 1959 | { |
10cdc3f3 AD |
1960 | struct xps_map *map = NULL; |
1961 | int pos; | |
537c00de | 1962 | |
10cdc3f3 AD |
1963 | if (dev_maps) |
1964 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
537c00de | 1965 | |
10cdc3f3 AD |
1966 | for (pos = 0; map && pos < map->len; pos++) { |
1967 | if (map->queues[pos] == index) { | |
537c00de AD |
1968 | if (map->len > 1) { |
1969 | map->queues[pos] = map->queues[--map->len]; | |
1970 | } else { | |
10cdc3f3 | 1971 | RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL); |
537c00de AD |
1972 | kfree_rcu(map, rcu); |
1973 | map = NULL; | |
1974 | } | |
10cdc3f3 | 1975 | break; |
537c00de | 1976 | } |
537c00de AD |
1977 | } |
1978 | ||
10cdc3f3 AD |
1979 | return map; |
1980 | } | |
1981 | ||
024e9679 | 1982 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
10cdc3f3 AD |
1983 | { |
1984 | struct xps_dev_maps *dev_maps; | |
024e9679 | 1985 | int cpu, i; |
10cdc3f3 AD |
1986 | bool active = false; |
1987 | ||
1988 | mutex_lock(&xps_map_mutex); | |
1989 | dev_maps = xmap_dereference(dev->xps_maps); | |
1990 | ||
1991 | if (!dev_maps) | |
1992 | goto out_no_maps; | |
1993 | ||
1994 | for_each_possible_cpu(cpu) { | |
024e9679 AD |
1995 | for (i = index; i < dev->num_tx_queues; i++) { |
1996 | if (!remove_xps_queue(dev_maps, cpu, i)) | |
1997 | break; | |
1998 | } | |
1999 | if (i == dev->num_tx_queues) | |
10cdc3f3 AD |
2000 | active = true; |
2001 | } | |
2002 | ||
2003 | if (!active) { | |
537c00de AD |
2004 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
2005 | kfree_rcu(dev_maps, rcu); | |
2006 | } | |
2007 | ||
024e9679 AD |
2008 | for (i = index; i < dev->num_tx_queues; i++) |
2009 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), | |
2010 | NUMA_NO_NODE); | |
2011 | ||
537c00de AD |
2012 | out_no_maps: |
2013 | mutex_unlock(&xps_map_mutex); | |
2014 | } | |
2015 | ||
01c5f864 AD |
2016 | static struct xps_map *expand_xps_map(struct xps_map *map, |
2017 | int cpu, u16 index) | |
2018 | { | |
2019 | struct xps_map *new_map; | |
2020 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2021 | int i, pos; | |
2022 | ||
2023 | for (pos = 0; map && pos < map->len; pos++) { | |
2024 | if (map->queues[pos] != index) | |
2025 | continue; | |
2026 | return map; | |
2027 | } | |
2028 | ||
2029 | /* Need to add queue to this CPU's existing map */ | |
2030 | if (map) { | |
2031 | if (pos < map->alloc_len) | |
2032 | return map; | |
2033 | ||
2034 | alloc_len = map->alloc_len * 2; | |
2035 | } | |
2036 | ||
2037 | /* Need to allocate new map to store queue on this CPU's map */ | |
2038 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2039 | cpu_to_node(cpu)); | |
2040 | if (!new_map) | |
2041 | return NULL; | |
2042 | ||
2043 | for (i = 0; i < pos; i++) | |
2044 | new_map->queues[i] = map->queues[i]; | |
2045 | new_map->alloc_len = alloc_len; | |
2046 | new_map->len = pos; | |
2047 | ||
2048 | return new_map; | |
2049 | } | |
2050 | ||
3573540c MT |
2051 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
2052 | u16 index) | |
537c00de | 2053 | { |
01c5f864 | 2054 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
537c00de | 2055 | struct xps_map *map, *new_map; |
537c00de | 2056 | int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES); |
01c5f864 AD |
2057 | int cpu, numa_node_id = -2; |
2058 | bool active = false; | |
537c00de AD |
2059 | |
2060 | mutex_lock(&xps_map_mutex); | |
2061 | ||
2062 | dev_maps = xmap_dereference(dev->xps_maps); | |
2063 | ||
01c5f864 AD |
2064 | /* allocate memory for queue storage */ |
2065 | for_each_online_cpu(cpu) { | |
2066 | if (!cpumask_test_cpu(cpu, mask)) | |
2067 | continue; | |
2068 | ||
2069 | if (!new_dev_maps) | |
2070 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2071 | if (!new_dev_maps) { |
2072 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2073 | return -ENOMEM; |
2bb60cb9 | 2074 | } |
01c5f864 AD |
2075 | |
2076 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2077 | NULL; | |
2078 | ||
2079 | map = expand_xps_map(map, cpu, index); | |
2080 | if (!map) | |
2081 | goto error; | |
2082 | ||
2083 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
2084 | } | |
2085 | ||
2086 | if (!new_dev_maps) | |
2087 | goto out_no_new_maps; | |
2088 | ||
537c00de | 2089 | for_each_possible_cpu(cpu) { |
01c5f864 AD |
2090 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2091 | /* add queue to CPU maps */ | |
2092 | int pos = 0; | |
2093 | ||
2094 | map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2095 | while ((pos < map->len) && (map->queues[pos] != index)) | |
2096 | pos++; | |
2097 | ||
2098 | if (pos == map->len) | |
2099 | map->queues[map->len++] = index; | |
537c00de | 2100 | #ifdef CONFIG_NUMA |
537c00de AD |
2101 | if (numa_node_id == -2) |
2102 | numa_node_id = cpu_to_node(cpu); | |
2103 | else if (numa_node_id != cpu_to_node(cpu)) | |
2104 | numa_node_id = -1; | |
537c00de | 2105 | #endif |
01c5f864 AD |
2106 | } else if (dev_maps) { |
2107 | /* fill in the new device map from the old device map */ | |
2108 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2109 | RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map); | |
537c00de | 2110 | } |
01c5f864 | 2111 | |
537c00de AD |
2112 | } |
2113 | ||
01c5f864 AD |
2114 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2115 | ||
537c00de | 2116 | /* Cleanup old maps */ |
01c5f864 AD |
2117 | if (dev_maps) { |
2118 | for_each_possible_cpu(cpu) { | |
2119 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2120 | map = xmap_dereference(dev_maps->cpu_map[cpu]); | |
2121 | if (map && map != new_map) | |
2122 | kfree_rcu(map, rcu); | |
2123 | } | |
537c00de | 2124 | |
01c5f864 | 2125 | kfree_rcu(dev_maps, rcu); |
537c00de AD |
2126 | } |
2127 | ||
01c5f864 AD |
2128 | dev_maps = new_dev_maps; |
2129 | active = true; | |
537c00de | 2130 | |
01c5f864 AD |
2131 | out_no_new_maps: |
2132 | /* update Tx queue numa node */ | |
537c00de AD |
2133 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2134 | (numa_node_id >= 0) ? numa_node_id : | |
2135 | NUMA_NO_NODE); | |
2136 | ||
01c5f864 AD |
2137 | if (!dev_maps) |
2138 | goto out_no_maps; | |
2139 | ||
2140 | /* removes queue from unused CPUs */ | |
2141 | for_each_possible_cpu(cpu) { | |
2142 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) | |
2143 | continue; | |
2144 | ||
2145 | if (remove_xps_queue(dev_maps, cpu, index)) | |
2146 | active = true; | |
2147 | } | |
2148 | ||
2149 | /* free map if not active */ | |
2150 | if (!active) { | |
2151 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2152 | kfree_rcu(dev_maps, rcu); | |
2153 | } | |
2154 | ||
2155 | out_no_maps: | |
537c00de AD |
2156 | mutex_unlock(&xps_map_mutex); |
2157 | ||
2158 | return 0; | |
2159 | error: | |
01c5f864 AD |
2160 | /* remove any maps that we added */ |
2161 | for_each_possible_cpu(cpu) { | |
2162 | new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]); | |
2163 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) : | |
2164 | NULL; | |
2165 | if (new_map && new_map != map) | |
2166 | kfree(new_map); | |
2167 | } | |
2168 | ||
537c00de AD |
2169 | mutex_unlock(&xps_map_mutex); |
2170 | ||
537c00de AD |
2171 | kfree(new_dev_maps); |
2172 | return -ENOMEM; | |
2173 | } | |
2174 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2175 | ||
2176 | #endif | |
f0796d5c JF |
2177 | /* |
2178 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2179 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2180 | */ | |
e6484930 | 2181 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2182 | { |
1d24eb48 TH |
2183 | int rc; |
2184 | ||
e6484930 TH |
2185 | if (txq < 1 || txq > dev->num_tx_queues) |
2186 | return -EINVAL; | |
f0796d5c | 2187 | |
5c56580b BH |
2188 | if (dev->reg_state == NETREG_REGISTERED || |
2189 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2190 | ASSERT_RTNL(); |
2191 | ||
1d24eb48 TH |
2192 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2193 | txq); | |
bf264145 TH |
2194 | if (rc) |
2195 | return rc; | |
2196 | ||
4f57c087 JF |
2197 | if (dev->num_tc) |
2198 | netif_setup_tc(dev, txq); | |
2199 | ||
024e9679 | 2200 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2201 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2202 | #ifdef CONFIG_XPS |
2203 | netif_reset_xps_queues_gt(dev, txq); | |
2204 | #endif | |
2205 | } | |
f0796d5c | 2206 | } |
e6484930 TH |
2207 | |
2208 | dev->real_num_tx_queues = txq; | |
2209 | return 0; | |
f0796d5c JF |
2210 | } |
2211 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2212 | |
a953be53 | 2213 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2214 | /** |
2215 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2216 | * @dev: Network device | |
2217 | * @rxq: Actual number of RX queues | |
2218 | * | |
2219 | * This must be called either with the rtnl_lock held or before | |
2220 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2221 | * negative error code. If called before registration, it always |
2222 | * succeeds. | |
62fe0b40 BH |
2223 | */ |
2224 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2225 | { | |
2226 | int rc; | |
2227 | ||
bd25fa7b TH |
2228 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2229 | return -EINVAL; | |
2230 | ||
62fe0b40 BH |
2231 | if (dev->reg_state == NETREG_REGISTERED) { |
2232 | ASSERT_RTNL(); | |
2233 | ||
62fe0b40 BH |
2234 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2235 | rxq); | |
2236 | if (rc) | |
2237 | return rc; | |
62fe0b40 BH |
2238 | } |
2239 | ||
2240 | dev->real_num_rx_queues = rxq; | |
2241 | return 0; | |
2242 | } | |
2243 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2244 | #endif | |
2245 | ||
2c53040f BH |
2246 | /** |
2247 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2248 | * |
2249 | * This routine should set an upper limit on the number of RSS queues | |
2250 | * used by default by multiqueue devices. | |
2251 | */ | |
a55b138b | 2252 | int netif_get_num_default_rss_queues(void) |
16917b87 | 2253 | { |
40e4e713 HS |
2254 | return is_kdump_kernel() ? |
2255 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
2256 | } |
2257 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2258 | ||
3bcb846c | 2259 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 2260 | { |
def82a1d JP |
2261 | struct softnet_data *sd; |
2262 | unsigned long flags; | |
56079431 | 2263 | |
def82a1d | 2264 | local_irq_save(flags); |
903ceff7 | 2265 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2266 | q->next_sched = NULL; |
2267 | *sd->output_queue_tailp = q; | |
2268 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2269 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2270 | local_irq_restore(flags); | |
2271 | } | |
2272 | ||
2273 | void __netif_schedule(struct Qdisc *q) | |
2274 | { | |
2275 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2276 | __netif_reschedule(q); | |
56079431 DV |
2277 | } |
2278 | EXPORT_SYMBOL(__netif_schedule); | |
2279 | ||
e6247027 ED |
2280 | struct dev_kfree_skb_cb { |
2281 | enum skb_free_reason reason; | |
2282 | }; | |
2283 | ||
2284 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2285 | { |
e6247027 ED |
2286 | return (struct dev_kfree_skb_cb *)skb->cb; |
2287 | } | |
2288 | ||
46e5da40 JF |
2289 | void netif_schedule_queue(struct netdev_queue *txq) |
2290 | { | |
2291 | rcu_read_lock(); | |
2292 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2293 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2294 | ||
2295 | __netif_schedule(q); | |
2296 | } | |
2297 | rcu_read_unlock(); | |
2298 | } | |
2299 | EXPORT_SYMBOL(netif_schedule_queue); | |
2300 | ||
2301 | /** | |
2302 | * netif_wake_subqueue - allow sending packets on subqueue | |
2303 | * @dev: network device | |
2304 | * @queue_index: sub queue index | |
2305 | * | |
2306 | * Resume individual transmit queue of a device with multiple transmit queues. | |
2307 | */ | |
2308 | void netif_wake_subqueue(struct net_device *dev, u16 queue_index) | |
2309 | { | |
2310 | struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index); | |
2311 | ||
2312 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) { | |
2313 | struct Qdisc *q; | |
2314 | ||
2315 | rcu_read_lock(); | |
2316 | q = rcu_dereference(txq->qdisc); | |
2317 | __netif_schedule(q); | |
2318 | rcu_read_unlock(); | |
2319 | } | |
2320 | } | |
2321 | EXPORT_SYMBOL(netif_wake_subqueue); | |
2322 | ||
2323 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) | |
2324 | { | |
2325 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2326 | struct Qdisc *q; | |
2327 | ||
2328 | rcu_read_lock(); | |
2329 | q = rcu_dereference(dev_queue->qdisc); | |
2330 | __netif_schedule(q); | |
2331 | rcu_read_unlock(); | |
2332 | } | |
2333 | } | |
2334 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2335 | ||
e6247027 | 2336 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2337 | { |
e6247027 | 2338 | unsigned long flags; |
56079431 | 2339 | |
e6247027 ED |
2340 | if (likely(atomic_read(&skb->users) == 1)) { |
2341 | smp_rmb(); | |
2342 | atomic_set(&skb->users, 0); | |
2343 | } else if (likely(!atomic_dec_and_test(&skb->users))) { | |
2344 | return; | |
bea3348e | 2345 | } |
e6247027 ED |
2346 | get_kfree_skb_cb(skb)->reason = reason; |
2347 | local_irq_save(flags); | |
2348 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2349 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2350 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2351 | local_irq_restore(flags); | |
56079431 | 2352 | } |
e6247027 | 2353 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2354 | |
e6247027 | 2355 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2356 | { |
2357 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2358 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2359 | else |
2360 | dev_kfree_skb(skb); | |
2361 | } | |
e6247027 | 2362 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2363 | |
2364 | ||
bea3348e SH |
2365 | /** |
2366 | * netif_device_detach - mark device as removed | |
2367 | * @dev: network device | |
2368 | * | |
2369 | * Mark device as removed from system and therefore no longer available. | |
2370 | */ | |
56079431 DV |
2371 | void netif_device_detach(struct net_device *dev) |
2372 | { | |
2373 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2374 | netif_running(dev)) { | |
d543103a | 2375 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2376 | } |
2377 | } | |
2378 | EXPORT_SYMBOL(netif_device_detach); | |
2379 | ||
bea3348e SH |
2380 | /** |
2381 | * netif_device_attach - mark device as attached | |
2382 | * @dev: network device | |
2383 | * | |
2384 | * Mark device as attached from system and restart if needed. | |
2385 | */ | |
56079431 DV |
2386 | void netif_device_attach(struct net_device *dev) |
2387 | { | |
2388 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2389 | netif_running(dev)) { | |
d543103a | 2390 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2391 | __netdev_watchdog_up(dev); |
56079431 DV |
2392 | } |
2393 | } | |
2394 | EXPORT_SYMBOL(netif_device_attach); | |
2395 | ||
5605c762 JP |
2396 | /* |
2397 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2398 | * to be used as a distribution range. | |
2399 | */ | |
2400 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, | |
2401 | unsigned int num_tx_queues) | |
2402 | { | |
2403 | u32 hash; | |
2404 | u16 qoffset = 0; | |
2405 | u16 qcount = num_tx_queues; | |
2406 | ||
2407 | if (skb_rx_queue_recorded(skb)) { | |
2408 | hash = skb_get_rx_queue(skb); | |
2409 | while (unlikely(hash >= num_tx_queues)) | |
2410 | hash -= num_tx_queues; | |
2411 | return hash; | |
2412 | } | |
2413 | ||
2414 | if (dev->num_tc) { | |
2415 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2416 | qoffset = dev->tc_to_txq[tc].offset; | |
2417 | qcount = dev->tc_to_txq[tc].count; | |
2418 | } | |
2419 | ||
2420 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2421 | } | |
2422 | EXPORT_SYMBOL(__skb_tx_hash); | |
2423 | ||
36c92474 BH |
2424 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2425 | { | |
84d15ae5 | 2426 | static const netdev_features_t null_features; |
36c92474 | 2427 | struct net_device *dev = skb->dev; |
88ad4175 | 2428 | const char *name = ""; |
36c92474 | 2429 | |
c846ad9b BG |
2430 | if (!net_ratelimit()) |
2431 | return; | |
2432 | ||
88ad4175 BM |
2433 | if (dev) { |
2434 | if (dev->dev.parent) | |
2435 | name = dev_driver_string(dev->dev.parent); | |
2436 | else | |
2437 | name = netdev_name(dev); | |
2438 | } | |
36c92474 BH |
2439 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2440 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2441 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2442 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2443 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2444 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2445 | } | |
2446 | ||
1da177e4 LT |
2447 | /* |
2448 | * Invalidate hardware checksum when packet is to be mangled, and | |
2449 | * complete checksum manually on outgoing path. | |
2450 | */ | |
84fa7933 | 2451 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2452 | { |
d3bc23e7 | 2453 | __wsum csum; |
663ead3b | 2454 | int ret = 0, offset; |
1da177e4 | 2455 | |
84fa7933 | 2456 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2457 | goto out_set_summed; |
2458 | ||
2459 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2460 | skb_warn_bad_offload(skb); |
2461 | return -EINVAL; | |
1da177e4 LT |
2462 | } |
2463 | ||
cef401de ED |
2464 | /* Before computing a checksum, we should make sure no frag could |
2465 | * be modified by an external entity : checksum could be wrong. | |
2466 | */ | |
2467 | if (skb_has_shared_frag(skb)) { | |
2468 | ret = __skb_linearize(skb); | |
2469 | if (ret) | |
2470 | goto out; | |
2471 | } | |
2472 | ||
55508d60 | 2473 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2474 | BUG_ON(offset >= skb_headlen(skb)); |
2475 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2476 | ||
2477 | offset += skb->csum_offset; | |
2478 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2479 | ||
2480 | if (skb_cloned(skb) && | |
2481 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2482 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2483 | if (ret) | |
2484 | goto out; | |
2485 | } | |
2486 | ||
a030847e | 2487 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 2488 | out_set_summed: |
1da177e4 | 2489 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2490 | out: |
1da177e4 LT |
2491 | return ret; |
2492 | } | |
d1b19dff | 2493 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2494 | |
6ae23ad3 TH |
2495 | /* skb_csum_offload_check - Driver helper function to determine if a device |
2496 | * with limited checksum offload capabilities is able to offload the checksum | |
2497 | * for a given packet. | |
2498 | * | |
2499 | * Arguments: | |
2500 | * skb - sk_buff for the packet in question | |
2501 | * spec - contains the description of what device can offload | |
2502 | * csum_encapped - returns true if the checksum being offloaded is | |
2503 | * encpasulated. That is it is checksum for the transport header | |
2504 | * in the inner headers. | |
2505 | * checksum_help - when set indicates that helper function should | |
2506 | * call skb_checksum_help if offload checks fail | |
2507 | * | |
2508 | * Returns: | |
2509 | * true: Packet has passed the checksum checks and should be offloadable to | |
2510 | * the device (a driver may still need to check for additional | |
2511 | * restrictions of its device) | |
2512 | * false: Checksum is not offloadable. If checksum_help was set then | |
2513 | * skb_checksum_help was called to resolve checksum for non-GSO | |
2514 | * packets and when IP protocol is not SCTP | |
2515 | */ | |
2516 | bool __skb_csum_offload_chk(struct sk_buff *skb, | |
2517 | const struct skb_csum_offl_spec *spec, | |
2518 | bool *csum_encapped, | |
2519 | bool csum_help) | |
2520 | { | |
2521 | struct iphdr *iph; | |
2522 | struct ipv6hdr *ipv6; | |
2523 | void *nhdr; | |
2524 | int protocol; | |
2525 | u8 ip_proto; | |
2526 | ||
2527 | if (skb->protocol == htons(ETH_P_8021Q) || | |
2528 | skb->protocol == htons(ETH_P_8021AD)) { | |
2529 | if (!spec->vlan_okay) | |
2530 | goto need_help; | |
2531 | } | |
2532 | ||
2533 | /* We check whether the checksum refers to a transport layer checksum in | |
2534 | * the outermost header or an encapsulated transport layer checksum that | |
2535 | * corresponds to the inner headers of the skb. If the checksum is for | |
2536 | * something else in the packet we need help. | |
2537 | */ | |
2538 | if (skb_checksum_start_offset(skb) == skb_transport_offset(skb)) { | |
2539 | /* Non-encapsulated checksum */ | |
2540 | protocol = eproto_to_ipproto(vlan_get_protocol(skb)); | |
2541 | nhdr = skb_network_header(skb); | |
2542 | *csum_encapped = false; | |
2543 | if (spec->no_not_encapped) | |
2544 | goto need_help; | |
2545 | } else if (skb->encapsulation && spec->encap_okay && | |
2546 | skb_checksum_start_offset(skb) == | |
2547 | skb_inner_transport_offset(skb)) { | |
2548 | /* Encapsulated checksum */ | |
2549 | *csum_encapped = true; | |
2550 | switch (skb->inner_protocol_type) { | |
2551 | case ENCAP_TYPE_ETHER: | |
2552 | protocol = eproto_to_ipproto(skb->inner_protocol); | |
2553 | break; | |
2554 | case ENCAP_TYPE_IPPROTO: | |
2555 | protocol = skb->inner_protocol; | |
2556 | break; | |
2557 | } | |
2558 | nhdr = skb_inner_network_header(skb); | |
2559 | } else { | |
2560 | goto need_help; | |
2561 | } | |
2562 | ||
2563 | switch (protocol) { | |
2564 | case IPPROTO_IP: | |
2565 | if (!spec->ipv4_okay) | |
2566 | goto need_help; | |
2567 | iph = nhdr; | |
2568 | ip_proto = iph->protocol; | |
2569 | if (iph->ihl != 5 && !spec->ip_options_okay) | |
2570 | goto need_help; | |
2571 | break; | |
2572 | case IPPROTO_IPV6: | |
2573 | if (!spec->ipv6_okay) | |
2574 | goto need_help; | |
2575 | if (spec->no_encapped_ipv6 && *csum_encapped) | |
2576 | goto need_help; | |
2577 | ipv6 = nhdr; | |
2578 | nhdr += sizeof(*ipv6); | |
2579 | ip_proto = ipv6->nexthdr; | |
2580 | break; | |
2581 | default: | |
2582 | goto need_help; | |
2583 | } | |
2584 | ||
2585 | ip_proto_again: | |
2586 | switch (ip_proto) { | |
2587 | case IPPROTO_TCP: | |
2588 | if (!spec->tcp_okay || | |
2589 | skb->csum_offset != offsetof(struct tcphdr, check)) | |
2590 | goto need_help; | |
2591 | break; | |
2592 | case IPPROTO_UDP: | |
2593 | if (!spec->udp_okay || | |
2594 | skb->csum_offset != offsetof(struct udphdr, check)) | |
2595 | goto need_help; | |
2596 | break; | |
2597 | case IPPROTO_SCTP: | |
2598 | if (!spec->sctp_okay || | |
2599 | skb->csum_offset != offsetof(struct sctphdr, checksum)) | |
2600 | goto cant_help; | |
2601 | break; | |
2602 | case NEXTHDR_HOP: | |
2603 | case NEXTHDR_ROUTING: | |
2604 | case NEXTHDR_DEST: { | |
2605 | u8 *opthdr = nhdr; | |
2606 | ||
2607 | if (protocol != IPPROTO_IPV6 || !spec->ext_hdrs_okay) | |
2608 | goto need_help; | |
2609 | ||
2610 | ip_proto = opthdr[0]; | |
2611 | nhdr += (opthdr[1] + 1) << 3; | |
2612 | ||
2613 | goto ip_proto_again; | |
2614 | } | |
2615 | default: | |
2616 | goto need_help; | |
2617 | } | |
2618 | ||
2619 | /* Passed the tests for offloading checksum */ | |
2620 | return true; | |
2621 | ||
2622 | need_help: | |
2623 | if (csum_help && !skb_shinfo(skb)->gso_size) | |
2624 | skb_checksum_help(skb); | |
2625 | cant_help: | |
2626 | return false; | |
2627 | } | |
2628 | EXPORT_SYMBOL(__skb_csum_offload_chk); | |
2629 | ||
53d6471c | 2630 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2631 | { |
252e3346 | 2632 | __be16 type = skb->protocol; |
f6a78bfc | 2633 | |
19acc327 PS |
2634 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2635 | if (type == htons(ETH_P_TEB)) { | |
2636 | struct ethhdr *eth; | |
2637 | ||
2638 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2639 | return 0; | |
2640 | ||
2641 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2642 | type = eth->h_proto; | |
2643 | } | |
2644 | ||
d4bcef3f | 2645 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2646 | } |
2647 | ||
2648 | /** | |
2649 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2650 | * @skb: buffer to segment | |
2651 | * @features: features for the output path (see dev->features) | |
2652 | */ | |
2653 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2654 | netdev_features_t features) | |
2655 | { | |
2656 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2657 | struct packet_offload *ptype; | |
53d6471c VY |
2658 | int vlan_depth = skb->mac_len; |
2659 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2660 | |
2661 | if (unlikely(!type)) | |
2662 | return ERR_PTR(-EINVAL); | |
2663 | ||
53d6471c | 2664 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2665 | |
2666 | rcu_read_lock(); | |
22061d80 | 2667 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2668 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2669 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2670 | break; |
2671 | } | |
2672 | } | |
2673 | rcu_read_unlock(); | |
2674 | ||
98e399f8 | 2675 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2676 | |
f6a78bfc HX |
2677 | return segs; |
2678 | } | |
05e8ef4a PS |
2679 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2680 | ||
2681 | ||
2682 | /* openvswitch calls this on rx path, so we need a different check. | |
2683 | */ | |
2684 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2685 | { | |
2686 | if (tx_path) | |
2687 | return skb->ip_summed != CHECKSUM_PARTIAL; | |
2688 | else | |
2689 | return skb->ip_summed == CHECKSUM_NONE; | |
2690 | } | |
2691 | ||
2692 | /** | |
2693 | * __skb_gso_segment - Perform segmentation on skb. | |
2694 | * @skb: buffer to segment | |
2695 | * @features: features for the output path (see dev->features) | |
2696 | * @tx_path: whether it is called in TX path | |
2697 | * | |
2698 | * This function segments the given skb and returns a list of segments. | |
2699 | * | |
2700 | * It may return NULL if the skb requires no segmentation. This is | |
2701 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 KK |
2702 | * |
2703 | * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. | |
05e8ef4a PS |
2704 | */ |
2705 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2706 | netdev_features_t features, bool tx_path) | |
2707 | { | |
2708 | if (unlikely(skb_needs_check(skb, tx_path))) { | |
2709 | int err; | |
2710 | ||
2711 | skb_warn_bad_offload(skb); | |
2712 | ||
a40e0a66 | 2713 | err = skb_cow_head(skb, 0); |
2714 | if (err < 0) | |
05e8ef4a PS |
2715 | return ERR_PTR(err); |
2716 | } | |
2717 | ||
802ab55a AD |
2718 | /* Only report GSO partial support if it will enable us to |
2719 | * support segmentation on this frame without needing additional | |
2720 | * work. | |
2721 | */ | |
2722 | if (features & NETIF_F_GSO_PARTIAL) { | |
2723 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
2724 | struct net_device *dev = skb->dev; | |
2725 | ||
2726 | partial_features |= dev->features & dev->gso_partial_features; | |
2727 | if (!skb_gso_ok(skb, features | partial_features)) | |
2728 | features &= ~NETIF_F_GSO_PARTIAL; | |
2729 | } | |
2730 | ||
9207f9d4 KK |
2731 | BUILD_BUG_ON(SKB_SGO_CB_OFFSET + |
2732 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); | |
2733 | ||
68c33163 | 2734 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2735 | SKB_GSO_CB(skb)->encap_level = 0; |
2736 | ||
05e8ef4a PS |
2737 | skb_reset_mac_header(skb); |
2738 | skb_reset_mac_len(skb); | |
2739 | ||
2740 | return skb_mac_gso_segment(skb, features); | |
2741 | } | |
12b0004d | 2742 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2743 | |
fb286bb2 HX |
2744 | /* Take action when hardware reception checksum errors are detected. */ |
2745 | #ifdef CONFIG_BUG | |
2746 | void netdev_rx_csum_fault(struct net_device *dev) | |
2747 | { | |
2748 | if (net_ratelimit()) { | |
7b6cd1ce | 2749 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2750 | dump_stack(); |
2751 | } | |
2752 | } | |
2753 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2754 | #endif | |
2755 | ||
1da177e4 LT |
2756 | /* Actually, we should eliminate this check as soon as we know, that: |
2757 | * 1. IOMMU is present and allows to map all the memory. | |
2758 | * 2. No high memory really exists on this machine. | |
2759 | */ | |
2760 | ||
c1e756bf | 2761 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2762 | { |
3d3a8533 | 2763 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2764 | int i; |
5acbbd42 | 2765 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2766 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2767 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
2768 | if (PageHighMem(skb_frag_page(frag))) | |
5acbbd42 | 2769 | return 1; |
ea2ab693 | 2770 | } |
5acbbd42 | 2771 | } |
1da177e4 | 2772 | |
5acbbd42 FT |
2773 | if (PCI_DMA_BUS_IS_PHYS) { |
2774 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2775 | |
9092c658 ED |
2776 | if (!pdev) |
2777 | return 0; | |
5acbbd42 | 2778 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2779 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2780 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
5acbbd42 FT |
2781 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2782 | return 1; | |
2783 | } | |
2784 | } | |
3d3a8533 | 2785 | #endif |
1da177e4 LT |
2786 | return 0; |
2787 | } | |
1da177e4 | 2788 | |
3b392ddb SH |
2789 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2790 | * instead of standard features for the netdev. | |
2791 | */ | |
d0edc7bf | 2792 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2793 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2794 | netdev_features_t features, | |
2795 | __be16 type) | |
2796 | { | |
25cd9ba0 | 2797 | if (eth_p_mpls(type)) |
3b392ddb SH |
2798 | features &= skb->dev->mpls_features; |
2799 | ||
2800 | return features; | |
2801 | } | |
2802 | #else | |
2803 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2804 | netdev_features_t features, | |
2805 | __be16 type) | |
2806 | { | |
2807 | return features; | |
2808 | } | |
2809 | #endif | |
2810 | ||
c8f44aff | 2811 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2812 | netdev_features_t features) |
f01a5236 | 2813 | { |
53d6471c | 2814 | int tmp; |
3b392ddb SH |
2815 | __be16 type; |
2816 | ||
2817 | type = skb_network_protocol(skb, &tmp); | |
2818 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2819 | |
c0d680e5 | 2820 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2821 | !can_checksum_protocol(features, type)) { |
996e8021 | 2822 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
c1e756bf | 2823 | } else if (illegal_highdma(skb->dev, skb)) { |
f01a5236 JG |
2824 | features &= ~NETIF_F_SG; |
2825 | } | |
2826 | ||
2827 | return features; | |
2828 | } | |
2829 | ||
e38f3025 TM |
2830 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2831 | struct net_device *dev, | |
2832 | netdev_features_t features) | |
2833 | { | |
2834 | return features; | |
2835 | } | |
2836 | EXPORT_SYMBOL(passthru_features_check); | |
2837 | ||
8cb65d00 TM |
2838 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2839 | struct net_device *dev, | |
2840 | netdev_features_t features) | |
2841 | { | |
2842 | return vlan_features_check(skb, features); | |
2843 | } | |
2844 | ||
cbc53e08 AD |
2845 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
2846 | struct net_device *dev, | |
2847 | netdev_features_t features) | |
2848 | { | |
2849 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
2850 | ||
2851 | if (gso_segs > dev->gso_max_segs) | |
2852 | return features & ~NETIF_F_GSO_MASK; | |
2853 | ||
802ab55a AD |
2854 | /* Support for GSO partial features requires software |
2855 | * intervention before we can actually process the packets | |
2856 | * so we need to strip support for any partial features now | |
2857 | * and we can pull them back in after we have partially | |
2858 | * segmented the frame. | |
2859 | */ | |
2860 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
2861 | features &= ~dev->gso_partial_features; | |
2862 | ||
2863 | /* Make sure to clear the IPv4 ID mangling feature if the | |
2864 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
2865 | */ |
2866 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
2867 | struct iphdr *iph = skb->encapsulation ? | |
2868 | inner_ip_hdr(skb) : ip_hdr(skb); | |
2869 | ||
2870 | if (!(iph->frag_off & htons(IP_DF))) | |
2871 | features &= ~NETIF_F_TSO_MANGLEID; | |
2872 | } | |
2873 | ||
2874 | return features; | |
2875 | } | |
2876 | ||
c1e756bf | 2877 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2878 | { |
5f35227e | 2879 | struct net_device *dev = skb->dev; |
fcbeb976 | 2880 | netdev_features_t features = dev->features; |
58e998c6 | 2881 | |
cbc53e08 AD |
2882 | if (skb_is_gso(skb)) |
2883 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 2884 | |
5f35227e JG |
2885 | /* If encapsulation offload request, verify we are testing |
2886 | * hardware encapsulation features instead of standard | |
2887 | * features for the netdev | |
2888 | */ | |
2889 | if (skb->encapsulation) | |
2890 | features &= dev->hw_enc_features; | |
2891 | ||
f5a7fb88 TM |
2892 | if (skb_vlan_tagged(skb)) |
2893 | features = netdev_intersect_features(features, | |
2894 | dev->vlan_features | | |
2895 | NETIF_F_HW_VLAN_CTAG_TX | | |
2896 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2897 | |
5f35227e JG |
2898 | if (dev->netdev_ops->ndo_features_check) |
2899 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2900 | features); | |
8cb65d00 TM |
2901 | else |
2902 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2903 | |
c1e756bf | 2904 | return harmonize_features(skb, features); |
58e998c6 | 2905 | } |
c1e756bf | 2906 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2907 | |
2ea25513 | 2908 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2909 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2910 | { |
2ea25513 DM |
2911 | unsigned int len; |
2912 | int rc; | |
00829823 | 2913 | |
7866a621 | 2914 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 2915 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 2916 | |
2ea25513 DM |
2917 | len = skb->len; |
2918 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 2919 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 2920 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 2921 | |
2ea25513 DM |
2922 | return rc; |
2923 | } | |
7b9c6090 | 2924 | |
8dcda22a DM |
2925 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
2926 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
2927 | { |
2928 | struct sk_buff *skb = first; | |
2929 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 2930 | |
7f2e870f DM |
2931 | while (skb) { |
2932 | struct sk_buff *next = skb->next; | |
fc70fb64 | 2933 | |
7f2e870f | 2934 | skb->next = NULL; |
95f6b3dd | 2935 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
2936 | if (unlikely(!dev_xmit_complete(rc))) { |
2937 | skb->next = next; | |
2938 | goto out; | |
2939 | } | |
6afff0ca | 2940 | |
7f2e870f DM |
2941 | skb = next; |
2942 | if (netif_xmit_stopped(txq) && skb) { | |
2943 | rc = NETDEV_TX_BUSY; | |
2944 | break; | |
9ccb8975 | 2945 | } |
7f2e870f | 2946 | } |
9ccb8975 | 2947 | |
7f2e870f DM |
2948 | out: |
2949 | *ret = rc; | |
2950 | return skb; | |
2951 | } | |
b40863c6 | 2952 | |
1ff0dc94 ED |
2953 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
2954 | netdev_features_t features) | |
f6a78bfc | 2955 | { |
df8a39de | 2956 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
2957 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
2958 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
2959 | return skb; |
2960 | } | |
f6a78bfc | 2961 | |
55a93b3e | 2962 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
2963 | { |
2964 | netdev_features_t features; | |
f6a78bfc | 2965 | |
eae3f88e DM |
2966 | features = netif_skb_features(skb); |
2967 | skb = validate_xmit_vlan(skb, features); | |
2968 | if (unlikely(!skb)) | |
2969 | goto out_null; | |
7b9c6090 | 2970 | |
8b86a61d | 2971 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
2972 | struct sk_buff *segs; |
2973 | ||
2974 | segs = skb_gso_segment(skb, features); | |
cecda693 | 2975 | if (IS_ERR(segs)) { |
af6dabc9 | 2976 | goto out_kfree_skb; |
cecda693 JW |
2977 | } else if (segs) { |
2978 | consume_skb(skb); | |
2979 | skb = segs; | |
f6a78bfc | 2980 | } |
eae3f88e DM |
2981 | } else { |
2982 | if (skb_needs_linearize(skb, features) && | |
2983 | __skb_linearize(skb)) | |
2984 | goto out_kfree_skb; | |
4ec93edb | 2985 | |
eae3f88e DM |
2986 | /* If packet is not checksummed and device does not |
2987 | * support checksumming for this protocol, complete | |
2988 | * checksumming here. | |
2989 | */ | |
2990 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
2991 | if (skb->encapsulation) | |
2992 | skb_set_inner_transport_header(skb, | |
2993 | skb_checksum_start_offset(skb)); | |
2994 | else | |
2995 | skb_set_transport_header(skb, | |
2996 | skb_checksum_start_offset(skb)); | |
a188222b | 2997 | if (!(features & NETIF_F_CSUM_MASK) && |
eae3f88e DM |
2998 | skb_checksum_help(skb)) |
2999 | goto out_kfree_skb; | |
7b9c6090 | 3000 | } |
0c772159 | 3001 | } |
7b9c6090 | 3002 | |
eae3f88e | 3003 | return skb; |
fc70fb64 | 3004 | |
f6a78bfc HX |
3005 | out_kfree_skb: |
3006 | kfree_skb(skb); | |
eae3f88e | 3007 | out_null: |
d21fd63e | 3008 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3009 | return NULL; |
3010 | } | |
6afff0ca | 3011 | |
55a93b3e ED |
3012 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
3013 | { | |
3014 | struct sk_buff *next, *head = NULL, *tail; | |
3015 | ||
bec3cfdc | 3016 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
3017 | next = skb->next; |
3018 | skb->next = NULL; | |
bec3cfdc ED |
3019 | |
3020 | /* in case skb wont be segmented, point to itself */ | |
3021 | skb->prev = skb; | |
3022 | ||
55a93b3e | 3023 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
3024 | if (!skb) |
3025 | continue; | |
55a93b3e | 3026 | |
bec3cfdc ED |
3027 | if (!head) |
3028 | head = skb; | |
3029 | else | |
3030 | tail->next = skb; | |
3031 | /* If skb was segmented, skb->prev points to | |
3032 | * the last segment. If not, it still contains skb. | |
3033 | */ | |
3034 | tail = skb->prev; | |
55a93b3e ED |
3035 | } |
3036 | return head; | |
f6a78bfc HX |
3037 | } |
3038 | ||
1def9238 ED |
3039 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3040 | { | |
3041 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3042 | ||
3043 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3044 | ||
3045 | /* To get more precise estimation of bytes sent on wire, | |
3046 | * we add to pkt_len the headers size of all segments | |
3047 | */ | |
3048 | if (shinfo->gso_size) { | |
757b8b1d | 3049 | unsigned int hdr_len; |
15e5a030 | 3050 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3051 | |
757b8b1d ED |
3052 | /* mac layer + network layer */ |
3053 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3054 | ||
3055 | /* + transport layer */ | |
1def9238 ED |
3056 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
3057 | hdr_len += tcp_hdrlen(skb); | |
3058 | else | |
3059 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
3060 | |
3061 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3062 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3063 | shinfo->gso_size); | |
3064 | ||
3065 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3066 | } |
3067 | } | |
3068 | ||
bbd8a0d3 KK |
3069 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3070 | struct net_device *dev, | |
3071 | struct netdev_queue *txq) | |
3072 | { | |
3073 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3074 | struct sk_buff *to_free = NULL; |
a2da570d | 3075 | bool contended; |
bbd8a0d3 KK |
3076 | int rc; |
3077 | ||
a2da570d | 3078 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
3079 | /* |
3080 | * Heuristic to force contended enqueues to serialize on a | |
3081 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3082 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3083 | * often and dequeue packets faster. |
79640a4c | 3084 | */ |
a2da570d | 3085 | contended = qdisc_is_running(q); |
79640a4c ED |
3086 | if (unlikely(contended)) |
3087 | spin_lock(&q->busylock); | |
3088 | ||
bbd8a0d3 KK |
3089 | spin_lock(root_lock); |
3090 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3091 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3092 | rc = NET_XMIT_DROP; |
3093 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3094 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3095 | /* |
3096 | * This is a work-conserving queue; there are no old skbs | |
3097 | * waiting to be sent out; and the qdisc is not running - | |
3098 | * xmit the skb directly. | |
3099 | */ | |
bfe0d029 | 3100 | |
bfe0d029 ED |
3101 | qdisc_bstats_update(q, skb); |
3102 | ||
55a93b3e | 3103 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3104 | if (unlikely(contended)) { |
3105 | spin_unlock(&q->busylock); | |
3106 | contended = false; | |
3107 | } | |
bbd8a0d3 | 3108 | __qdisc_run(q); |
79640a4c | 3109 | } else |
bc135b23 | 3110 | qdisc_run_end(q); |
bbd8a0d3 KK |
3111 | |
3112 | rc = NET_XMIT_SUCCESS; | |
3113 | } else { | |
520ac30f | 3114 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3115 | if (qdisc_run_begin(q)) { |
3116 | if (unlikely(contended)) { | |
3117 | spin_unlock(&q->busylock); | |
3118 | contended = false; | |
3119 | } | |
3120 | __qdisc_run(q); | |
3121 | } | |
bbd8a0d3 KK |
3122 | } |
3123 | spin_unlock(root_lock); | |
520ac30f ED |
3124 | if (unlikely(to_free)) |
3125 | kfree_skb_list(to_free); | |
79640a4c ED |
3126 | if (unlikely(contended)) |
3127 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3128 | return rc; |
3129 | } | |
3130 | ||
86f8515f | 3131 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3132 | static void skb_update_prio(struct sk_buff *skb) |
3133 | { | |
6977a79d | 3134 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 3135 | |
91c68ce2 | 3136 | if (!skb->priority && skb->sk && map) { |
2a56a1fe TH |
3137 | unsigned int prioidx = |
3138 | sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); | |
91c68ce2 ED |
3139 | |
3140 | if (prioidx < map->priomap_len) | |
3141 | skb->priority = map->priomap[prioidx]; | |
3142 | } | |
5bc1421e NH |
3143 | } |
3144 | #else | |
3145 | #define skb_update_prio(skb) | |
3146 | #endif | |
3147 | ||
f60e5990 | 3148 | DEFINE_PER_CPU(int, xmit_recursion); |
3149 | EXPORT_SYMBOL(xmit_recursion); | |
3150 | ||
95603e22 MM |
3151 | /** |
3152 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3153 | * @net: network namespace this loopback is happening in |
3154 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3155 | * @skb: buffer to transmit |
3156 | */ | |
0c4b51f0 | 3157 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3158 | { |
3159 | skb_reset_mac_header(skb); | |
3160 | __skb_pull(skb, skb_network_offset(skb)); | |
3161 | skb->pkt_type = PACKET_LOOPBACK; | |
3162 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3163 | WARN_ON(!skb_dst(skb)); | |
3164 | skb_dst_force(skb); | |
3165 | netif_rx_ni(skb); | |
3166 | return 0; | |
3167 | } | |
3168 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3169 | ||
1f211a1b DB |
3170 | #ifdef CONFIG_NET_EGRESS |
3171 | static struct sk_buff * | |
3172 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3173 | { | |
3174 | struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list); | |
3175 | struct tcf_result cl_res; | |
3176 | ||
3177 | if (!cl) | |
3178 | return skb; | |
3179 | ||
3180 | /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set | |
3181 | * earlier by the caller. | |
3182 | */ | |
3183 | qdisc_bstats_cpu_update(cl->q, skb); | |
3184 | ||
3185 | switch (tc_classify(skb, cl, &cl_res, false)) { | |
3186 | case TC_ACT_OK: | |
3187 | case TC_ACT_RECLASSIFY: | |
3188 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3189 | break; | |
3190 | case TC_ACT_SHOT: | |
3191 | qdisc_qstats_cpu_drop(cl->q); | |
3192 | *ret = NET_XMIT_DROP; | |
7e2c3aea DB |
3193 | kfree_skb(skb); |
3194 | return NULL; | |
1f211a1b DB |
3195 | case TC_ACT_STOLEN: |
3196 | case TC_ACT_QUEUED: | |
3197 | *ret = NET_XMIT_SUCCESS; | |
7e2c3aea | 3198 | consume_skb(skb); |
1f211a1b DB |
3199 | return NULL; |
3200 | case TC_ACT_REDIRECT: | |
3201 | /* No need to push/pop skb's mac_header here on egress! */ | |
3202 | skb_do_redirect(skb); | |
3203 | *ret = NET_XMIT_SUCCESS; | |
3204 | return NULL; | |
3205 | default: | |
3206 | break; | |
3207 | } | |
3208 | ||
3209 | return skb; | |
3210 | } | |
3211 | #endif /* CONFIG_NET_EGRESS */ | |
3212 | ||
638b2a69 JP |
3213 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
3214 | { | |
3215 | #ifdef CONFIG_XPS | |
3216 | struct xps_dev_maps *dev_maps; | |
3217 | struct xps_map *map; | |
3218 | int queue_index = -1; | |
3219 | ||
3220 | rcu_read_lock(); | |
3221 | dev_maps = rcu_dereference(dev->xps_maps); | |
3222 | if (dev_maps) { | |
3223 | map = rcu_dereference( | |
3224 | dev_maps->cpu_map[skb->sender_cpu - 1]); | |
3225 | if (map) { | |
3226 | if (map->len == 1) | |
3227 | queue_index = map->queues[0]; | |
3228 | else | |
3229 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), | |
3230 | map->len)]; | |
3231 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3232 | queue_index = -1; | |
3233 | } | |
3234 | } | |
3235 | rcu_read_unlock(); | |
3236 | ||
3237 | return queue_index; | |
3238 | #else | |
3239 | return -1; | |
3240 | #endif | |
3241 | } | |
3242 | ||
3243 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) | |
3244 | { | |
3245 | struct sock *sk = skb->sk; | |
3246 | int queue_index = sk_tx_queue_get(sk); | |
3247 | ||
3248 | if (queue_index < 0 || skb->ooo_okay || | |
3249 | queue_index >= dev->real_num_tx_queues) { | |
3250 | int new_index = get_xps_queue(dev, skb); | |
3251 | if (new_index < 0) | |
3252 | new_index = skb_tx_hash(dev, skb); | |
3253 | ||
3254 | if (queue_index != new_index && sk && | |
004a5d01 | 3255 | sk_fullsock(sk) && |
638b2a69 JP |
3256 | rcu_access_pointer(sk->sk_dst_cache)) |
3257 | sk_tx_queue_set(sk, new_index); | |
3258 | ||
3259 | queue_index = new_index; | |
3260 | } | |
3261 | ||
3262 | return queue_index; | |
3263 | } | |
3264 | ||
3265 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
3266 | struct sk_buff *skb, | |
3267 | void *accel_priv) | |
3268 | { | |
3269 | int queue_index = 0; | |
3270 | ||
3271 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3272 | u32 sender_cpu = skb->sender_cpu - 1; |
3273 | ||
3274 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3275 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3276 | #endif | |
3277 | ||
3278 | if (dev->real_num_tx_queues != 1) { | |
3279 | const struct net_device_ops *ops = dev->netdev_ops; | |
3280 | if (ops->ndo_select_queue) | |
3281 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, | |
3282 | __netdev_pick_tx); | |
3283 | else | |
3284 | queue_index = __netdev_pick_tx(dev, skb); | |
3285 | ||
3286 | if (!accel_priv) | |
3287 | queue_index = netdev_cap_txqueue(dev, queue_index); | |
3288 | } | |
3289 | ||
3290 | skb_set_queue_mapping(skb, queue_index); | |
3291 | return netdev_get_tx_queue(dev, queue_index); | |
3292 | } | |
3293 | ||
d29f749e | 3294 | /** |
9d08dd3d | 3295 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3296 | * @skb: buffer to transmit |
9d08dd3d | 3297 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
3298 | * |
3299 | * Queue a buffer for transmission to a network device. The caller must | |
3300 | * have set the device and priority and built the buffer before calling | |
3301 | * this function. The function can be called from an interrupt. | |
3302 | * | |
3303 | * A negative errno code is returned on a failure. A success does not | |
3304 | * guarantee the frame will be transmitted as it may be dropped due | |
3305 | * to congestion or traffic shaping. | |
3306 | * | |
3307 | * ----------------------------------------------------------------------------------- | |
3308 | * I notice this method can also return errors from the queue disciplines, | |
3309 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3310 | * be positive. | |
3311 | * | |
3312 | * Regardless of the return value, the skb is consumed, so it is currently | |
3313 | * difficult to retry a send to this method. (You can bump the ref count | |
3314 | * before sending to hold a reference for retry if you are careful.) | |
3315 | * | |
3316 | * When calling this method, interrupts MUST be enabled. This is because | |
3317 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3318 | * --BLG | |
3319 | */ | |
0a59f3a9 | 3320 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
3321 | { |
3322 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3323 | struct netdev_queue *txq; |
1da177e4 LT |
3324 | struct Qdisc *q; |
3325 | int rc = -ENOMEM; | |
3326 | ||
6d1ccff6 ED |
3327 | skb_reset_mac_header(skb); |
3328 | ||
e7fd2885 WB |
3329 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3330 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3331 | ||
4ec93edb YH |
3332 | /* Disable soft irqs for various locks below. Also |
3333 | * stops preemption for RCU. | |
1da177e4 | 3334 | */ |
4ec93edb | 3335 | rcu_read_lock_bh(); |
1da177e4 | 3336 | |
5bc1421e NH |
3337 | skb_update_prio(skb); |
3338 | ||
1f211a1b DB |
3339 | qdisc_pkt_len_init(skb); |
3340 | #ifdef CONFIG_NET_CLS_ACT | |
3341 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); | |
3342 | # ifdef CONFIG_NET_EGRESS | |
3343 | if (static_key_false(&egress_needed)) { | |
3344 | skb = sch_handle_egress(skb, &rc, dev); | |
3345 | if (!skb) | |
3346 | goto out; | |
3347 | } | |
3348 | # endif | |
3349 | #endif | |
02875878 ED |
3350 | /* If device/qdisc don't need skb->dst, release it right now while |
3351 | * its hot in this cpu cache. | |
3352 | */ | |
3353 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3354 | skb_dst_drop(skb); | |
3355 | else | |
3356 | skb_dst_force(skb); | |
3357 | ||
0c4f691f SF |
3358 | #ifdef CONFIG_NET_SWITCHDEV |
3359 | /* Don't forward if offload device already forwarded */ | |
3360 | if (skb->offload_fwd_mark && | |
3361 | skb->offload_fwd_mark == dev->offload_fwd_mark) { | |
3362 | consume_skb(skb); | |
3363 | rc = NET_XMIT_SUCCESS; | |
3364 | goto out; | |
3365 | } | |
3366 | #endif | |
3367 | ||
f663dd9a | 3368 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 3369 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3370 | |
cf66ba58 | 3371 | trace_net_dev_queue(skb); |
1da177e4 | 3372 | if (q->enqueue) { |
bbd8a0d3 | 3373 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3374 | goto out; |
1da177e4 LT |
3375 | } |
3376 | ||
3377 | /* The device has no queue. Common case for software devices: | |
3378 | loopback, all the sorts of tunnels... | |
3379 | ||
932ff279 HX |
3380 | Really, it is unlikely that netif_tx_lock protection is necessary |
3381 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
3382 | counters.) |
3383 | However, it is possible, that they rely on protection | |
3384 | made by us here. | |
3385 | ||
3386 | Check this and shot the lock. It is not prone from deadlocks. | |
3387 | Either shot noqueue qdisc, it is even simpler 8) | |
3388 | */ | |
3389 | if (dev->flags & IFF_UP) { | |
3390 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3391 | ||
c773e847 | 3392 | if (txq->xmit_lock_owner != cpu) { |
a70b506e DB |
3393 | if (unlikely(__this_cpu_read(xmit_recursion) > |
3394 | XMIT_RECURSION_LIMIT)) | |
745e20f1 ED |
3395 | goto recursion_alert; |
3396 | ||
1f59533f JDB |
3397 | skb = validate_xmit_skb(skb, dev); |
3398 | if (!skb) | |
d21fd63e | 3399 | goto out; |
1f59533f | 3400 | |
c773e847 | 3401 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3402 | |
73466498 | 3403 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 3404 | __this_cpu_inc(xmit_recursion); |
ce93718f | 3405 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 3406 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 3407 | if (dev_xmit_complete(rc)) { |
c773e847 | 3408 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3409 | goto out; |
3410 | } | |
3411 | } | |
c773e847 | 3412 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3413 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3414 | dev->name); | |
1da177e4 LT |
3415 | } else { |
3416 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3417 | * unfortunately |
3418 | */ | |
3419 | recursion_alert: | |
e87cc472 JP |
3420 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3421 | dev->name); | |
1da177e4 LT |
3422 | } |
3423 | } | |
3424 | ||
3425 | rc = -ENETDOWN; | |
d4828d85 | 3426 | rcu_read_unlock_bh(); |
1da177e4 | 3427 | |
015f0688 | 3428 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3429 | kfree_skb_list(skb); |
1da177e4 LT |
3430 | return rc; |
3431 | out: | |
d4828d85 | 3432 | rcu_read_unlock_bh(); |
1da177e4 LT |
3433 | return rc; |
3434 | } | |
f663dd9a | 3435 | |
2b4aa3ce | 3436 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3437 | { |
3438 | return __dev_queue_xmit(skb, NULL); | |
3439 | } | |
2b4aa3ce | 3440 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3441 | |
f663dd9a JW |
3442 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3443 | { | |
3444 | return __dev_queue_xmit(skb, accel_priv); | |
3445 | } | |
3446 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3447 | ||
1da177e4 LT |
3448 | |
3449 | /*======================================================================= | |
3450 | Receiver routines | |
3451 | =======================================================================*/ | |
3452 | ||
6b2bedc3 | 3453 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3454 | EXPORT_SYMBOL(netdev_max_backlog); |
3455 | ||
3b098e2d | 3456 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
3457 | int netdev_budget __read_mostly = 300; |
3458 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 3459 | |
eecfd7c4 ED |
3460 | /* Called with irq disabled */ |
3461 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3462 | struct napi_struct *napi) | |
3463 | { | |
3464 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3465 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3466 | } | |
3467 | ||
bfb564e7 KK |
3468 | #ifdef CONFIG_RPS |
3469 | ||
3470 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3471 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3472 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3473 | u32 rps_cpu_mask __read_mostly; |
3474 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3475 | |
c5905afb | 3476 | struct static_key rps_needed __read_mostly; |
3df97ba8 | 3477 | EXPORT_SYMBOL(rps_needed); |
adc9300e | 3478 | |
c445477d BH |
3479 | static struct rps_dev_flow * |
3480 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3481 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3482 | { | |
a31196b0 | 3483 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3484 | #ifdef CONFIG_RFS_ACCEL |
3485 | struct netdev_rx_queue *rxqueue; | |
3486 | struct rps_dev_flow_table *flow_table; | |
3487 | struct rps_dev_flow *old_rflow; | |
3488 | u32 flow_id; | |
3489 | u16 rxq_index; | |
3490 | int rc; | |
3491 | ||
3492 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3493 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3494 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3495 | goto out; |
3496 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3497 | if (rxq_index == skb_get_rx_queue(skb)) | |
3498 | goto out; | |
3499 | ||
3500 | rxqueue = dev->_rx + rxq_index; | |
3501 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3502 | if (!flow_table) | |
3503 | goto out; | |
61b905da | 3504 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3505 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3506 | rxq_index, flow_id); | |
3507 | if (rc < 0) | |
3508 | goto out; | |
3509 | old_rflow = rflow; | |
3510 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3511 | rflow->filter = rc; |
3512 | if (old_rflow->filter == rflow->filter) | |
3513 | old_rflow->filter = RPS_NO_FILTER; | |
3514 | out: | |
3515 | #endif | |
3516 | rflow->last_qtail = | |
09994d1b | 3517 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3518 | } |
3519 | ||
09994d1b | 3520 | rflow->cpu = next_cpu; |
c445477d BH |
3521 | return rflow; |
3522 | } | |
3523 | ||
bfb564e7 KK |
3524 | /* |
3525 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3526 | * CPU from the RPS map of the receiving queue for a given skb. | |
3527 | * rcu_read_lock must be held on entry. | |
3528 | */ | |
3529 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3530 | struct rps_dev_flow **rflowp) | |
3531 | { | |
567e4b79 ED |
3532 | const struct rps_sock_flow_table *sock_flow_table; |
3533 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3534 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3535 | struct rps_map *map; |
bfb564e7 | 3536 | int cpu = -1; |
567e4b79 | 3537 | u32 tcpu; |
61b905da | 3538 | u32 hash; |
bfb564e7 KK |
3539 | |
3540 | if (skb_rx_queue_recorded(skb)) { | |
3541 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3542 | |
62fe0b40 BH |
3543 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3544 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3545 | "%s received packet on queue %u, but number " | |
3546 | "of RX queues is %u\n", | |
3547 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3548 | goto done; |
3549 | } | |
567e4b79 ED |
3550 | rxqueue += index; |
3551 | } | |
bfb564e7 | 3552 | |
567e4b79 ED |
3553 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3554 | ||
3555 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3556 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3557 | if (!flow_table && !map) |
bfb564e7 KK |
3558 | goto done; |
3559 | ||
2d47b459 | 3560 | skb_reset_network_header(skb); |
61b905da TH |
3561 | hash = skb_get_hash(skb); |
3562 | if (!hash) | |
bfb564e7 KK |
3563 | goto done; |
3564 | ||
fec5e652 TH |
3565 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3566 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3567 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3568 | u32 next_cpu; |
3569 | u32 ident; | |
3570 | ||
3571 | /* First check into global flow table if there is a match */ | |
3572 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3573 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3574 | goto try_rps; | |
fec5e652 | 3575 | |
567e4b79 ED |
3576 | next_cpu = ident & rps_cpu_mask; |
3577 | ||
3578 | /* OK, now we know there is a match, | |
3579 | * we can look at the local (per receive queue) flow table | |
3580 | */ | |
61b905da | 3581 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3582 | tcpu = rflow->cpu; |
3583 | ||
fec5e652 TH |
3584 | /* |
3585 | * If the desired CPU (where last recvmsg was done) is | |
3586 | * different from current CPU (one in the rx-queue flow | |
3587 | * table entry), switch if one of the following holds: | |
a31196b0 | 3588 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
3589 | * - Current CPU is offline. |
3590 | * - The current CPU's queue tail has advanced beyond the | |
3591 | * last packet that was enqueued using this table entry. | |
3592 | * This guarantees that all previous packets for the flow | |
3593 | * have been dequeued, thus preserving in order delivery. | |
3594 | */ | |
3595 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 3596 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 3597 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
3598 | rflow->last_qtail)) >= 0)) { |
3599 | tcpu = next_cpu; | |
c445477d | 3600 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3601 | } |
c445477d | 3602 | |
a31196b0 | 3603 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
3604 | *rflowp = rflow; |
3605 | cpu = tcpu; | |
3606 | goto done; | |
3607 | } | |
3608 | } | |
3609 | ||
567e4b79 ED |
3610 | try_rps: |
3611 | ||
0a9627f2 | 3612 | if (map) { |
8fc54f68 | 3613 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3614 | if (cpu_online(tcpu)) { |
3615 | cpu = tcpu; | |
3616 | goto done; | |
3617 | } | |
3618 | } | |
3619 | ||
3620 | done: | |
0a9627f2 TH |
3621 | return cpu; |
3622 | } | |
3623 | ||
c445477d BH |
3624 | #ifdef CONFIG_RFS_ACCEL |
3625 | ||
3626 | /** | |
3627 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3628 | * @dev: Device on which the filter was set | |
3629 | * @rxq_index: RX queue index | |
3630 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3631 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3632 | * | |
3633 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3634 | * this function for each installed filter and remove the filters for | |
3635 | * which it returns %true. | |
3636 | */ | |
3637 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3638 | u32 flow_id, u16 filter_id) | |
3639 | { | |
3640 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3641 | struct rps_dev_flow_table *flow_table; | |
3642 | struct rps_dev_flow *rflow; | |
3643 | bool expire = true; | |
a31196b0 | 3644 | unsigned int cpu; |
c445477d BH |
3645 | |
3646 | rcu_read_lock(); | |
3647 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3648 | if (flow_table && flow_id <= flow_table->mask) { | |
3649 | rflow = &flow_table->flows[flow_id]; | |
3650 | cpu = ACCESS_ONCE(rflow->cpu); | |
a31196b0 | 3651 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
3652 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
3653 | rflow->last_qtail) < | |
3654 | (int)(10 * flow_table->mask))) | |
3655 | expire = false; | |
3656 | } | |
3657 | rcu_read_unlock(); | |
3658 | return expire; | |
3659 | } | |
3660 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3661 | ||
3662 | #endif /* CONFIG_RFS_ACCEL */ | |
3663 | ||
0a9627f2 | 3664 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3665 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3666 | { |
e36fa2f7 ED |
3667 | struct softnet_data *sd = data; |
3668 | ||
eecfd7c4 | 3669 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3670 | sd->received_rps++; |
0a9627f2 | 3671 | } |
e36fa2f7 | 3672 | |
fec5e652 | 3673 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3674 | |
e36fa2f7 ED |
3675 | /* |
3676 | * Check if this softnet_data structure is another cpu one | |
3677 | * If yes, queue it to our IPI list and return 1 | |
3678 | * If no, return 0 | |
3679 | */ | |
3680 | static int rps_ipi_queued(struct softnet_data *sd) | |
3681 | { | |
3682 | #ifdef CONFIG_RPS | |
903ceff7 | 3683 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3684 | |
3685 | if (sd != mysd) { | |
3686 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3687 | mysd->rps_ipi_list = sd; | |
3688 | ||
3689 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3690 | return 1; | |
3691 | } | |
3692 | #endif /* CONFIG_RPS */ | |
3693 | return 0; | |
3694 | } | |
3695 | ||
99bbc707 WB |
3696 | #ifdef CONFIG_NET_FLOW_LIMIT |
3697 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3698 | #endif | |
3699 | ||
3700 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3701 | { | |
3702 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3703 | struct sd_flow_limit *fl; | |
3704 | struct softnet_data *sd; | |
3705 | unsigned int old_flow, new_flow; | |
3706 | ||
3707 | if (qlen < (netdev_max_backlog >> 1)) | |
3708 | return false; | |
3709 | ||
903ceff7 | 3710 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3711 | |
3712 | rcu_read_lock(); | |
3713 | fl = rcu_dereference(sd->flow_limit); | |
3714 | if (fl) { | |
3958afa1 | 3715 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3716 | old_flow = fl->history[fl->history_head]; |
3717 | fl->history[fl->history_head] = new_flow; | |
3718 | ||
3719 | fl->history_head++; | |
3720 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3721 | ||
3722 | if (likely(fl->buckets[old_flow])) | |
3723 | fl->buckets[old_flow]--; | |
3724 | ||
3725 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3726 | fl->count++; | |
3727 | rcu_read_unlock(); | |
3728 | return true; | |
3729 | } | |
3730 | } | |
3731 | rcu_read_unlock(); | |
3732 | #endif | |
3733 | return false; | |
3734 | } | |
3735 | ||
0a9627f2 TH |
3736 | /* |
3737 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3738 | * queue (may be a remote CPU queue). | |
3739 | */ | |
fec5e652 TH |
3740 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3741 | unsigned int *qtail) | |
0a9627f2 | 3742 | { |
e36fa2f7 | 3743 | struct softnet_data *sd; |
0a9627f2 | 3744 | unsigned long flags; |
99bbc707 | 3745 | unsigned int qlen; |
0a9627f2 | 3746 | |
e36fa2f7 | 3747 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3748 | |
3749 | local_irq_save(flags); | |
0a9627f2 | 3750 | |
e36fa2f7 | 3751 | rps_lock(sd); |
e9e4dd32 JA |
3752 | if (!netif_running(skb->dev)) |
3753 | goto drop; | |
99bbc707 WB |
3754 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3755 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3756 | if (qlen) { |
0a9627f2 | 3757 | enqueue: |
e36fa2f7 | 3758 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3759 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3760 | rps_unlock(sd); |
152102c7 | 3761 | local_irq_restore(flags); |
0a9627f2 TH |
3762 | return NET_RX_SUCCESS; |
3763 | } | |
3764 | ||
ebda37c2 ED |
3765 | /* Schedule NAPI for backlog device |
3766 | * We can use non atomic operation since we own the queue lock | |
3767 | */ | |
3768 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3769 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3770 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3771 | } |
3772 | goto enqueue; | |
3773 | } | |
3774 | ||
e9e4dd32 | 3775 | drop: |
dee42870 | 3776 | sd->dropped++; |
e36fa2f7 | 3777 | rps_unlock(sd); |
0a9627f2 | 3778 | |
0a9627f2 TH |
3779 | local_irq_restore(flags); |
3780 | ||
caf586e5 | 3781 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3782 | kfree_skb(skb); |
3783 | return NET_RX_DROP; | |
3784 | } | |
1da177e4 | 3785 | |
ae78dbfa | 3786 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 3787 | { |
b0e28f1e | 3788 | int ret; |
1da177e4 | 3789 | |
588f0330 | 3790 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 3791 | |
cf66ba58 | 3792 | trace_netif_rx(skb); |
df334545 | 3793 | #ifdef CONFIG_RPS |
c5905afb | 3794 | if (static_key_false(&rps_needed)) { |
fec5e652 | 3795 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
3796 | int cpu; |
3797 | ||
cece1945 | 3798 | preempt_disable(); |
b0e28f1e | 3799 | rcu_read_lock(); |
fec5e652 TH |
3800 | |
3801 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
3802 | if (cpu < 0) |
3803 | cpu = smp_processor_id(); | |
fec5e652 TH |
3804 | |
3805 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
3806 | ||
b0e28f1e | 3807 | rcu_read_unlock(); |
cece1945 | 3808 | preempt_enable(); |
adc9300e ED |
3809 | } else |
3810 | #endif | |
fec5e652 TH |
3811 | { |
3812 | unsigned int qtail; | |
3813 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
3814 | put_cpu(); | |
3815 | } | |
b0e28f1e | 3816 | return ret; |
1da177e4 | 3817 | } |
ae78dbfa BH |
3818 | |
3819 | /** | |
3820 | * netif_rx - post buffer to the network code | |
3821 | * @skb: buffer to post | |
3822 | * | |
3823 | * This function receives a packet from a device driver and queues it for | |
3824 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
3825 | * may be dropped during processing for congestion control or by the | |
3826 | * protocol layers. | |
3827 | * | |
3828 | * return values: | |
3829 | * NET_RX_SUCCESS (no congestion) | |
3830 | * NET_RX_DROP (packet was dropped) | |
3831 | * | |
3832 | */ | |
3833 | ||
3834 | int netif_rx(struct sk_buff *skb) | |
3835 | { | |
3836 | trace_netif_rx_entry(skb); | |
3837 | ||
3838 | return netif_rx_internal(skb); | |
3839 | } | |
d1b19dff | 3840 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
3841 | |
3842 | int netif_rx_ni(struct sk_buff *skb) | |
3843 | { | |
3844 | int err; | |
3845 | ||
ae78dbfa BH |
3846 | trace_netif_rx_ni_entry(skb); |
3847 | ||
1da177e4 | 3848 | preempt_disable(); |
ae78dbfa | 3849 | err = netif_rx_internal(skb); |
1da177e4 LT |
3850 | if (local_softirq_pending()) |
3851 | do_softirq(); | |
3852 | preempt_enable(); | |
3853 | ||
3854 | return err; | |
3855 | } | |
1da177e4 LT |
3856 | EXPORT_SYMBOL(netif_rx_ni); |
3857 | ||
1da177e4 LT |
3858 | static void net_tx_action(struct softirq_action *h) |
3859 | { | |
903ceff7 | 3860 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
3861 | |
3862 | if (sd->completion_queue) { | |
3863 | struct sk_buff *clist; | |
3864 | ||
3865 | local_irq_disable(); | |
3866 | clist = sd->completion_queue; | |
3867 | sd->completion_queue = NULL; | |
3868 | local_irq_enable(); | |
3869 | ||
3870 | while (clist) { | |
3871 | struct sk_buff *skb = clist; | |
3872 | clist = clist->next; | |
3873 | ||
547b792c | 3874 | WARN_ON(atomic_read(&skb->users)); |
e6247027 ED |
3875 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
3876 | trace_consume_skb(skb); | |
3877 | else | |
3878 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
3879 | |
3880 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
3881 | __kfree_skb(skb); | |
3882 | else | |
3883 | __kfree_skb_defer(skb); | |
1da177e4 | 3884 | } |
15fad714 JDB |
3885 | |
3886 | __kfree_skb_flush(); | |
1da177e4 LT |
3887 | } |
3888 | ||
3889 | if (sd->output_queue) { | |
37437bb2 | 3890 | struct Qdisc *head; |
1da177e4 LT |
3891 | |
3892 | local_irq_disable(); | |
3893 | head = sd->output_queue; | |
3894 | sd->output_queue = NULL; | |
a9cbd588 | 3895 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
3896 | local_irq_enable(); |
3897 | ||
3898 | while (head) { | |
37437bb2 DM |
3899 | struct Qdisc *q = head; |
3900 | spinlock_t *root_lock; | |
3901 | ||
1da177e4 LT |
3902 | head = head->next_sched; |
3903 | ||
5fb66229 | 3904 | root_lock = qdisc_lock(q); |
3bcb846c ED |
3905 | spin_lock(root_lock); |
3906 | /* We need to make sure head->next_sched is read | |
3907 | * before clearing __QDISC_STATE_SCHED | |
3908 | */ | |
3909 | smp_mb__before_atomic(); | |
3910 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
3911 | qdisc_run(q); | |
3912 | spin_unlock(root_lock); | |
1da177e4 LT |
3913 | } |
3914 | } | |
3915 | } | |
3916 | ||
ab95bfe0 JP |
3917 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
3918 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
3919 | /* This hook is defined here for ATM LANE */ |
3920 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
3921 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 3922 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 3923 | #endif |
1da177e4 | 3924 | |
1f211a1b DB |
3925 | static inline struct sk_buff * |
3926 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
3927 | struct net_device *orig_dev) | |
f697c3e8 | 3928 | { |
e7582bab | 3929 | #ifdef CONFIG_NET_CLS_ACT |
d2788d34 DB |
3930 | struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); |
3931 | struct tcf_result cl_res; | |
24824a09 | 3932 | |
c9e99fd0 DB |
3933 | /* If there's at least one ingress present somewhere (so |
3934 | * we get here via enabled static key), remaining devices | |
3935 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 3936 | * out here. |
c9e99fd0 | 3937 | */ |
d2788d34 | 3938 | if (!cl) |
4577139b | 3939 | return skb; |
f697c3e8 HX |
3940 | if (*pt_prev) { |
3941 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
3942 | *pt_prev = NULL; | |
1da177e4 LT |
3943 | } |
3944 | ||
3365495c | 3945 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
c9e99fd0 | 3946 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); |
24ea591d | 3947 | qdisc_bstats_cpu_update(cl->q, skb); |
c9e99fd0 | 3948 | |
3b3ae880 | 3949 | switch (tc_classify(skb, cl, &cl_res, false)) { |
d2788d34 DB |
3950 | case TC_ACT_OK: |
3951 | case TC_ACT_RECLASSIFY: | |
3952 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3953 | break; | |
3954 | case TC_ACT_SHOT: | |
24ea591d | 3955 | qdisc_qstats_cpu_drop(cl->q); |
8a3a4c6e ED |
3956 | kfree_skb(skb); |
3957 | return NULL; | |
d2788d34 DB |
3958 | case TC_ACT_STOLEN: |
3959 | case TC_ACT_QUEUED: | |
8a3a4c6e | 3960 | consume_skb(skb); |
d2788d34 | 3961 | return NULL; |
27b29f63 AS |
3962 | case TC_ACT_REDIRECT: |
3963 | /* skb_mac_header check was done by cls/act_bpf, so | |
3964 | * we can safely push the L2 header back before | |
3965 | * redirecting to another netdev | |
3966 | */ | |
3967 | __skb_push(skb, skb->mac_len); | |
3968 | skb_do_redirect(skb); | |
3969 | return NULL; | |
d2788d34 DB |
3970 | default: |
3971 | break; | |
f697c3e8 | 3972 | } |
e7582bab | 3973 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
3974 | return skb; |
3975 | } | |
1da177e4 | 3976 | |
ab95bfe0 JP |
3977 | /** |
3978 | * netdev_rx_handler_register - register receive handler | |
3979 | * @dev: device to register a handler for | |
3980 | * @rx_handler: receive handler to register | |
93e2c32b | 3981 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 3982 | * |
e227867f | 3983 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
3984 | * called from __netif_receive_skb. A negative errno code is returned |
3985 | * on a failure. | |
3986 | * | |
3987 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
3988 | * |
3989 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
3990 | */ |
3991 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
3992 | rx_handler_func_t *rx_handler, |
3993 | void *rx_handler_data) | |
ab95bfe0 JP |
3994 | { |
3995 | ASSERT_RTNL(); | |
3996 | ||
3997 | if (dev->rx_handler) | |
3998 | return -EBUSY; | |
3999 | ||
00cfec37 | 4000 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 4001 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
4002 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
4003 | ||
4004 | return 0; | |
4005 | } | |
4006 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
4007 | ||
4008 | /** | |
4009 | * netdev_rx_handler_unregister - unregister receive handler | |
4010 | * @dev: device to unregister a handler from | |
4011 | * | |
166ec369 | 4012 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
4013 | * |
4014 | * The caller must hold the rtnl_mutex. | |
4015 | */ | |
4016 | void netdev_rx_handler_unregister(struct net_device *dev) | |
4017 | { | |
4018 | ||
4019 | ASSERT_RTNL(); | |
a9b3cd7f | 4020 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
4021 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
4022 | * section has a guarantee to see a non NULL rx_handler_data | |
4023 | * as well. | |
4024 | */ | |
4025 | synchronize_net(); | |
a9b3cd7f | 4026 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
4027 | } |
4028 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
4029 | ||
b4b9e355 MG |
4030 | /* |
4031 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
4032 | * the special handling of PFMEMALLOC skbs. | |
4033 | */ | |
4034 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
4035 | { | |
4036 | switch (skb->protocol) { | |
2b8837ae JP |
4037 | case htons(ETH_P_ARP): |
4038 | case htons(ETH_P_IP): | |
4039 | case htons(ETH_P_IPV6): | |
4040 | case htons(ETH_P_8021Q): | |
4041 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4042 | return true; |
4043 | default: | |
4044 | return false; | |
4045 | } | |
4046 | } | |
4047 | ||
e687ad60 PN |
4048 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4049 | int *ret, struct net_device *orig_dev) | |
4050 | { | |
e7582bab | 4051 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 PN |
4052 | if (nf_hook_ingress_active(skb)) { |
4053 | if (*pt_prev) { | |
4054 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4055 | *pt_prev = NULL; | |
4056 | } | |
4057 | ||
4058 | return nf_hook_ingress(skb); | |
4059 | } | |
e7582bab | 4060 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
4061 | return 0; |
4062 | } | |
e687ad60 | 4063 | |
9754e293 | 4064 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
4065 | { |
4066 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4067 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4068 | struct net_device *orig_dev; |
8a4eb573 | 4069 | bool deliver_exact = false; |
1da177e4 | 4070 | int ret = NET_RX_DROP; |
252e3346 | 4071 | __be16 type; |
1da177e4 | 4072 | |
588f0330 | 4073 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 4074 | |
cf66ba58 | 4075 | trace_netif_receive_skb(skb); |
9b22ea56 | 4076 | |
cc9bd5ce | 4077 | orig_dev = skb->dev; |
8f903c70 | 4078 | |
c1d2bbe1 | 4079 | skb_reset_network_header(skb); |
fda55eca ED |
4080 | if (!skb_transport_header_was_set(skb)) |
4081 | skb_reset_transport_header(skb); | |
0b5c9db1 | 4082 | skb_reset_mac_len(skb); |
1da177e4 LT |
4083 | |
4084 | pt_prev = NULL; | |
4085 | ||
63d8ea7f | 4086 | another_round: |
b6858177 | 4087 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
4088 | |
4089 | __this_cpu_inc(softnet_data.processed); | |
4090 | ||
8ad227ff PM |
4091 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
4092 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 4093 | skb = skb_vlan_untag(skb); |
bcc6d479 | 4094 | if (unlikely(!skb)) |
2c17d27c | 4095 | goto out; |
bcc6d479 JP |
4096 | } |
4097 | ||
1da177e4 LT |
4098 | #ifdef CONFIG_NET_CLS_ACT |
4099 | if (skb->tc_verd & TC_NCLS) { | |
4100 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
4101 | goto ncls; | |
4102 | } | |
4103 | #endif | |
4104 | ||
9754e293 | 4105 | if (pfmemalloc) |
b4b9e355 MG |
4106 | goto skip_taps; |
4107 | ||
1da177e4 | 4108 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
4109 | if (pt_prev) |
4110 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4111 | pt_prev = ptype; | |
4112 | } | |
4113 | ||
4114 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
4115 | if (pt_prev) | |
4116 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4117 | pt_prev = ptype; | |
1da177e4 LT |
4118 | } |
4119 | ||
b4b9e355 | 4120 | skip_taps: |
1cf51900 | 4121 | #ifdef CONFIG_NET_INGRESS |
4577139b | 4122 | if (static_key_false(&ingress_needed)) { |
1f211a1b | 4123 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 4124 | if (!skb) |
2c17d27c | 4125 | goto out; |
e687ad60 PN |
4126 | |
4127 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 4128 | goto out; |
4577139b | 4129 | } |
1cf51900 PN |
4130 | #endif |
4131 | #ifdef CONFIG_NET_CLS_ACT | |
4577139b | 4132 | skb->tc_verd = 0; |
1da177e4 LT |
4133 | ncls: |
4134 | #endif | |
9754e293 | 4135 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
4136 | goto drop; |
4137 | ||
df8a39de | 4138 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
4139 | if (pt_prev) { |
4140 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4141 | pt_prev = NULL; | |
4142 | } | |
48cc32d3 | 4143 | if (vlan_do_receive(&skb)) |
2425717b JF |
4144 | goto another_round; |
4145 | else if (unlikely(!skb)) | |
2c17d27c | 4146 | goto out; |
2425717b JF |
4147 | } |
4148 | ||
48cc32d3 | 4149 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
4150 | if (rx_handler) { |
4151 | if (pt_prev) { | |
4152 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4153 | pt_prev = NULL; | |
4154 | } | |
8a4eb573 JP |
4155 | switch (rx_handler(&skb)) { |
4156 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 4157 | ret = NET_RX_SUCCESS; |
2c17d27c | 4158 | goto out; |
8a4eb573 | 4159 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 4160 | goto another_round; |
8a4eb573 JP |
4161 | case RX_HANDLER_EXACT: |
4162 | deliver_exact = true; | |
4163 | case RX_HANDLER_PASS: | |
4164 | break; | |
4165 | default: | |
4166 | BUG(); | |
4167 | } | |
ab95bfe0 | 4168 | } |
1da177e4 | 4169 | |
df8a39de JP |
4170 | if (unlikely(skb_vlan_tag_present(skb))) { |
4171 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
4172 | skb->pkt_type = PACKET_OTHERHOST; |
4173 | /* Note: we might in the future use prio bits | |
4174 | * and set skb->priority like in vlan_do_receive() | |
4175 | * For the time being, just ignore Priority Code Point | |
4176 | */ | |
4177 | skb->vlan_tci = 0; | |
4178 | } | |
48cc32d3 | 4179 | |
7866a621 SN |
4180 | type = skb->protocol; |
4181 | ||
63d8ea7f | 4182 | /* deliver only exact match when indicated */ |
7866a621 SN |
4183 | if (likely(!deliver_exact)) { |
4184 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4185 | &ptype_base[ntohs(type) & | |
4186 | PTYPE_HASH_MASK]); | |
4187 | } | |
1f3c8804 | 4188 | |
7866a621 SN |
4189 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
4190 | &orig_dev->ptype_specific); | |
4191 | ||
4192 | if (unlikely(skb->dev != orig_dev)) { | |
4193 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4194 | &skb->dev->ptype_specific); | |
1da177e4 LT |
4195 | } |
4196 | ||
4197 | if (pt_prev) { | |
1080e512 | 4198 | if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) |
0e698bf6 | 4199 | goto drop; |
1080e512 MT |
4200 | else |
4201 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 4202 | } else { |
b4b9e355 | 4203 | drop: |
6e7333d3 JW |
4204 | if (!deliver_exact) |
4205 | atomic_long_inc(&skb->dev->rx_dropped); | |
4206 | else | |
4207 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
4208 | kfree_skb(skb); |
4209 | /* Jamal, now you will not able to escape explaining | |
4210 | * me how you were going to use this. :-) | |
4211 | */ | |
4212 | ret = NET_RX_DROP; | |
4213 | } | |
4214 | ||
2c17d27c | 4215 | out: |
9754e293 DM |
4216 | return ret; |
4217 | } | |
4218 | ||
4219 | static int __netif_receive_skb(struct sk_buff *skb) | |
4220 | { | |
4221 | int ret; | |
4222 | ||
4223 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
4224 | unsigned long pflags = current->flags; | |
4225 | ||
4226 | /* | |
4227 | * PFMEMALLOC skbs are special, they should | |
4228 | * - be delivered to SOCK_MEMALLOC sockets only | |
4229 | * - stay away from userspace | |
4230 | * - have bounded memory usage | |
4231 | * | |
4232 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
4233 | * context down to all allocation sites. | |
4234 | */ | |
4235 | current->flags |= PF_MEMALLOC; | |
4236 | ret = __netif_receive_skb_core(skb, true); | |
4237 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
4238 | } else | |
4239 | ret = __netif_receive_skb_core(skb, false); | |
4240 | ||
1da177e4 LT |
4241 | return ret; |
4242 | } | |
0a9627f2 | 4243 | |
ae78dbfa | 4244 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 4245 | { |
2c17d27c JA |
4246 | int ret; |
4247 | ||
588f0330 | 4248 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 4249 | |
c1f19b51 RC |
4250 | if (skb_defer_rx_timestamp(skb)) |
4251 | return NET_RX_SUCCESS; | |
4252 | ||
2c17d27c JA |
4253 | rcu_read_lock(); |
4254 | ||
df334545 | 4255 | #ifdef CONFIG_RPS |
c5905afb | 4256 | if (static_key_false(&rps_needed)) { |
3b098e2d | 4257 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 4258 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 4259 | |
3b098e2d ED |
4260 | if (cpu >= 0) { |
4261 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4262 | rcu_read_unlock(); | |
adc9300e | 4263 | return ret; |
3b098e2d | 4264 | } |
fec5e652 | 4265 | } |
1e94d72f | 4266 | #endif |
2c17d27c JA |
4267 | ret = __netif_receive_skb(skb); |
4268 | rcu_read_unlock(); | |
4269 | return ret; | |
0a9627f2 | 4270 | } |
ae78dbfa BH |
4271 | |
4272 | /** | |
4273 | * netif_receive_skb - process receive buffer from network | |
4274 | * @skb: buffer to process | |
4275 | * | |
4276 | * netif_receive_skb() is the main receive data processing function. | |
4277 | * It always succeeds. The buffer may be dropped during processing | |
4278 | * for congestion control or by the protocol layers. | |
4279 | * | |
4280 | * This function may only be called from softirq context and interrupts | |
4281 | * should be enabled. | |
4282 | * | |
4283 | * Return values (usually ignored): | |
4284 | * NET_RX_SUCCESS: no congestion | |
4285 | * NET_RX_DROP: packet was dropped | |
4286 | */ | |
04eb4489 | 4287 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa BH |
4288 | { |
4289 | trace_netif_receive_skb_entry(skb); | |
4290 | ||
4291 | return netif_receive_skb_internal(skb); | |
4292 | } | |
04eb4489 | 4293 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 4294 | |
88751275 ED |
4295 | /* Network device is going away, flush any packets still pending |
4296 | * Called with irqs disabled. | |
4297 | */ | |
152102c7 | 4298 | static void flush_backlog(void *arg) |
6e583ce5 | 4299 | { |
152102c7 | 4300 | struct net_device *dev = arg; |
903ceff7 | 4301 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
6e583ce5 SH |
4302 | struct sk_buff *skb, *tmp; |
4303 | ||
e36fa2f7 | 4304 | rps_lock(sd); |
6e7676c1 | 4305 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 4306 | if (skb->dev == dev) { |
e36fa2f7 | 4307 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 4308 | kfree_skb(skb); |
76cc8b13 | 4309 | input_queue_head_incr(sd); |
6e583ce5 | 4310 | } |
6e7676c1 | 4311 | } |
e36fa2f7 | 4312 | rps_unlock(sd); |
6e7676c1 CG |
4313 | |
4314 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
4315 | if (skb->dev == dev) { | |
4316 | __skb_unlink(skb, &sd->process_queue); | |
4317 | kfree_skb(skb); | |
76cc8b13 | 4318 | input_queue_head_incr(sd); |
6e7676c1 CG |
4319 | } |
4320 | } | |
6e583ce5 SH |
4321 | } |
4322 | ||
d565b0a1 HX |
4323 | static int napi_gro_complete(struct sk_buff *skb) |
4324 | { | |
22061d80 | 4325 | struct packet_offload *ptype; |
d565b0a1 | 4326 | __be16 type = skb->protocol; |
22061d80 | 4327 | struct list_head *head = &offload_base; |
d565b0a1 HX |
4328 | int err = -ENOENT; |
4329 | ||
c3c7c254 ED |
4330 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
4331 | ||
fc59f9a3 HX |
4332 | if (NAPI_GRO_CB(skb)->count == 1) { |
4333 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 4334 | goto out; |
fc59f9a3 | 4335 | } |
d565b0a1 HX |
4336 | |
4337 | rcu_read_lock(); | |
4338 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4339 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
4340 | continue; |
4341 | ||
299603e8 | 4342 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
4343 | break; |
4344 | } | |
4345 | rcu_read_unlock(); | |
4346 | ||
4347 | if (err) { | |
4348 | WARN_ON(&ptype->list == head); | |
4349 | kfree_skb(skb); | |
4350 | return NET_RX_SUCCESS; | |
4351 | } | |
4352 | ||
4353 | out: | |
ae78dbfa | 4354 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
4355 | } |
4356 | ||
2e71a6f8 ED |
4357 | /* napi->gro_list contains packets ordered by age. |
4358 | * youngest packets at the head of it. | |
4359 | * Complete skbs in reverse order to reduce latencies. | |
4360 | */ | |
4361 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 4362 | { |
2e71a6f8 | 4363 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 4364 | |
2e71a6f8 ED |
4365 | /* scan list and build reverse chain */ |
4366 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
4367 | skb->prev = prev; | |
4368 | prev = skb; | |
4369 | } | |
4370 | ||
4371 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 4372 | skb->next = NULL; |
2e71a6f8 ED |
4373 | |
4374 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
4375 | return; | |
4376 | ||
4377 | prev = skb->prev; | |
d565b0a1 | 4378 | napi_gro_complete(skb); |
2e71a6f8 | 4379 | napi->gro_count--; |
d565b0a1 HX |
4380 | } |
4381 | ||
4382 | napi->gro_list = NULL; | |
4383 | } | |
86cac58b | 4384 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 4385 | |
89c5fa33 ED |
4386 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
4387 | { | |
4388 | struct sk_buff *p; | |
4389 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 4390 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
4391 | |
4392 | for (p = napi->gro_list; p; p = p->next) { | |
4393 | unsigned long diffs; | |
4394 | ||
0b4cec8c TH |
4395 | NAPI_GRO_CB(p)->flush = 0; |
4396 | ||
4397 | if (hash != skb_get_hash_raw(p)) { | |
4398 | NAPI_GRO_CB(p)->same_flow = 0; | |
4399 | continue; | |
4400 | } | |
4401 | ||
89c5fa33 ED |
4402 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
4403 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
ce87fc6c | 4404 | diffs |= skb_metadata_dst_cmp(p, skb); |
89c5fa33 ED |
4405 | if (maclen == ETH_HLEN) |
4406 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 4407 | skb_mac_header(skb)); |
89c5fa33 ED |
4408 | else if (!diffs) |
4409 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 4410 | skb_mac_header(skb), |
89c5fa33 ED |
4411 | maclen); |
4412 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
4413 | } |
4414 | } | |
4415 | ||
299603e8 JC |
4416 | static void skb_gro_reset_offset(struct sk_buff *skb) |
4417 | { | |
4418 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4419 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4420 | ||
4421 | NAPI_GRO_CB(skb)->data_offset = 0; | |
4422 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
4423 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
4424 | ||
4425 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
4426 | pinfo->nr_frags && | |
4427 | !PageHighMem(skb_frag_page(frag0))) { | |
4428 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
4429 | NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0); | |
89c5fa33 ED |
4430 | } |
4431 | } | |
4432 | ||
a50e233c ED |
4433 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
4434 | { | |
4435 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4436 | ||
4437 | BUG_ON(skb->end - skb->tail < grow); | |
4438 | ||
4439 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4440 | ||
4441 | skb->data_len -= grow; | |
4442 | skb->tail += grow; | |
4443 | ||
4444 | pinfo->frags[0].page_offset += grow; | |
4445 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4446 | ||
4447 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4448 | skb_frag_unref(skb, 0); | |
4449 | memmove(pinfo->frags, pinfo->frags + 1, | |
4450 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4451 | } | |
4452 | } | |
4453 | ||
bb728820 | 4454 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4455 | { |
4456 | struct sk_buff **pp = NULL; | |
22061d80 | 4457 | struct packet_offload *ptype; |
d565b0a1 | 4458 | __be16 type = skb->protocol; |
22061d80 | 4459 | struct list_head *head = &offload_base; |
0da2afd5 | 4460 | int same_flow; |
5b252f0c | 4461 | enum gro_result ret; |
a50e233c | 4462 | int grow; |
d565b0a1 | 4463 | |
9c62a68d | 4464 | if (!(skb->dev->features & NETIF_F_GRO)) |
d565b0a1 HX |
4465 | goto normal; |
4466 | ||
5a212329 | 4467 | if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad) |
f17f5c91 HX |
4468 | goto normal; |
4469 | ||
89c5fa33 ED |
4470 | gro_list_prepare(napi, skb); |
4471 | ||
d565b0a1 HX |
4472 | rcu_read_lock(); |
4473 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4474 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4475 | continue; |
4476 | ||
86911732 | 4477 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4478 | skb_reset_mac_len(skb); |
d565b0a1 HX |
4479 | NAPI_GRO_CB(skb)->same_flow = 0; |
4480 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 4481 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 4482 | NAPI_GRO_CB(skb)->encap_mark = 0; |
a0ca153f | 4483 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 4484 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 4485 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4486 | |
662880f4 TH |
4487 | /* Setup for GRO checksum validation */ |
4488 | switch (skb->ip_summed) { | |
4489 | case CHECKSUM_COMPLETE: | |
4490 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4491 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4492 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4493 | break; | |
4494 | case CHECKSUM_UNNECESSARY: | |
4495 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4496 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4497 | break; | |
4498 | default: | |
4499 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4500 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4501 | } | |
d565b0a1 | 4502 | |
f191a1d1 | 4503 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4504 | break; |
4505 | } | |
4506 | rcu_read_unlock(); | |
4507 | ||
4508 | if (&ptype->list == head) | |
4509 | goto normal; | |
4510 | ||
0da2afd5 | 4511 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4512 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4513 | |
d565b0a1 HX |
4514 | if (pp) { |
4515 | struct sk_buff *nskb = *pp; | |
4516 | ||
4517 | *pp = nskb->next; | |
4518 | nskb->next = NULL; | |
4519 | napi_gro_complete(nskb); | |
4ae5544f | 4520 | napi->gro_count--; |
d565b0a1 HX |
4521 | } |
4522 | ||
0da2afd5 | 4523 | if (same_flow) |
d565b0a1 HX |
4524 | goto ok; |
4525 | ||
600adc18 | 4526 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4527 | goto normal; |
d565b0a1 | 4528 | |
600adc18 ED |
4529 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4530 | struct sk_buff *nskb = napi->gro_list; | |
4531 | ||
4532 | /* locate the end of the list to select the 'oldest' flow */ | |
4533 | while (nskb->next) { | |
4534 | pp = &nskb->next; | |
4535 | nskb = *pp; | |
4536 | } | |
4537 | *pp = NULL; | |
4538 | nskb->next = NULL; | |
4539 | napi_gro_complete(nskb); | |
4540 | } else { | |
4541 | napi->gro_count++; | |
4542 | } | |
d565b0a1 | 4543 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4544 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4545 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4546 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4547 | skb->next = napi->gro_list; |
4548 | napi->gro_list = skb; | |
5d0d9be8 | 4549 | ret = GRO_HELD; |
d565b0a1 | 4550 | |
ad0f9904 | 4551 | pull: |
a50e233c ED |
4552 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4553 | if (grow > 0) | |
4554 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4555 | ok: |
5d0d9be8 | 4556 | return ret; |
d565b0a1 HX |
4557 | |
4558 | normal: | |
ad0f9904 HX |
4559 | ret = GRO_NORMAL; |
4560 | goto pull; | |
5d38a079 | 4561 | } |
96e93eab | 4562 | |
bf5a755f JC |
4563 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4564 | { | |
4565 | struct list_head *offload_head = &offload_base; | |
4566 | struct packet_offload *ptype; | |
4567 | ||
4568 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4569 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4570 | continue; | |
4571 | return ptype; | |
4572 | } | |
4573 | return NULL; | |
4574 | } | |
e27a2f83 | 4575 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4576 | |
4577 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4578 | { | |
4579 | struct list_head *offload_head = &offload_base; | |
4580 | struct packet_offload *ptype; | |
4581 | ||
4582 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4583 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4584 | continue; | |
4585 | return ptype; | |
4586 | } | |
4587 | return NULL; | |
4588 | } | |
e27a2f83 | 4589 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4590 | |
bb728820 | 4591 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4592 | { |
5d0d9be8 HX |
4593 | switch (ret) { |
4594 | case GRO_NORMAL: | |
ae78dbfa | 4595 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4596 | ret = GRO_DROP; |
4597 | break; | |
5d38a079 | 4598 | |
5d0d9be8 | 4599 | case GRO_DROP: |
5d38a079 HX |
4600 | kfree_skb(skb); |
4601 | break; | |
5b252f0c | 4602 | |
daa86548 | 4603 | case GRO_MERGED_FREE: |
ce87fc6c JG |
4604 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) { |
4605 | skb_dst_drop(skb); | |
d7e8883c | 4606 | kmem_cache_free(skbuff_head_cache, skb); |
ce87fc6c | 4607 | } else { |
d7e8883c | 4608 | __kfree_skb(skb); |
ce87fc6c | 4609 | } |
daa86548 ED |
4610 | break; |
4611 | ||
5b252f0c BH |
4612 | case GRO_HELD: |
4613 | case GRO_MERGED: | |
4614 | break; | |
5d38a079 HX |
4615 | } |
4616 | ||
c7c4b3b6 | 4617 | return ret; |
5d0d9be8 | 4618 | } |
5d0d9be8 | 4619 | |
c7c4b3b6 | 4620 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4621 | { |
93f93a44 | 4622 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 4623 | trace_napi_gro_receive_entry(skb); |
86911732 | 4624 | |
a50e233c ED |
4625 | skb_gro_reset_offset(skb); |
4626 | ||
89c5fa33 | 4627 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4628 | } |
4629 | EXPORT_SYMBOL(napi_gro_receive); | |
4630 | ||
d0c2b0d2 | 4631 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4632 | { |
93a35f59 ED |
4633 | if (unlikely(skb->pfmemalloc)) { |
4634 | consume_skb(skb); | |
4635 | return; | |
4636 | } | |
96e93eab | 4637 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4638 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4639 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4640 | skb->vlan_tci = 0; |
66c46d74 | 4641 | skb->dev = napi->dev; |
6d152e23 | 4642 | skb->skb_iif = 0; |
c3caf119 JC |
4643 | skb->encapsulation = 0; |
4644 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4645 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
96e93eab HX |
4646 | |
4647 | napi->skb = skb; | |
4648 | } | |
96e93eab | 4649 | |
76620aaf | 4650 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 4651 | { |
5d38a079 | 4652 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
4653 | |
4654 | if (!skb) { | |
fd11a83d | 4655 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
4656 | if (skb) { |
4657 | napi->skb = skb; | |
4658 | skb_mark_napi_id(skb, napi); | |
4659 | } | |
80595d59 | 4660 | } |
96e93eab HX |
4661 | return skb; |
4662 | } | |
76620aaf | 4663 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 4664 | |
a50e233c ED |
4665 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
4666 | struct sk_buff *skb, | |
4667 | gro_result_t ret) | |
96e93eab | 4668 | { |
5d0d9be8 HX |
4669 | switch (ret) { |
4670 | case GRO_NORMAL: | |
a50e233c ED |
4671 | case GRO_HELD: |
4672 | __skb_push(skb, ETH_HLEN); | |
4673 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4674 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 4675 | ret = GRO_DROP; |
86911732 | 4676 | break; |
5d38a079 | 4677 | |
5d0d9be8 | 4678 | case GRO_DROP: |
5d0d9be8 HX |
4679 | case GRO_MERGED_FREE: |
4680 | napi_reuse_skb(napi, skb); | |
4681 | break; | |
5b252f0c BH |
4682 | |
4683 | case GRO_MERGED: | |
4684 | break; | |
5d0d9be8 | 4685 | } |
5d38a079 | 4686 | |
c7c4b3b6 | 4687 | return ret; |
5d38a079 | 4688 | } |
5d0d9be8 | 4689 | |
a50e233c ED |
4690 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
4691 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
4692 | * We copy ethernet header into skb->data to have a common layout. | |
4693 | */ | |
4adb9c4a | 4694 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
4695 | { |
4696 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
4697 | const struct ethhdr *eth; |
4698 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
4699 | |
4700 | napi->skb = NULL; | |
4701 | ||
a50e233c ED |
4702 | skb_reset_mac_header(skb); |
4703 | skb_gro_reset_offset(skb); | |
4704 | ||
4705 | eth = skb_gro_header_fast(skb, 0); | |
4706 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
4707 | eth = skb_gro_header_slow(skb, hlen, 0); | |
4708 | if (unlikely(!eth)) { | |
4da46ceb AC |
4709 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
4710 | __func__, napi->dev->name); | |
a50e233c ED |
4711 | napi_reuse_skb(napi, skb); |
4712 | return NULL; | |
4713 | } | |
4714 | } else { | |
4715 | gro_pull_from_frag0(skb, hlen); | |
4716 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
4717 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 4718 | } |
a50e233c ED |
4719 | __skb_pull(skb, hlen); |
4720 | ||
4721 | /* | |
4722 | * This works because the only protocols we care about don't require | |
4723 | * special handling. | |
4724 | * We'll fix it up properly in napi_frags_finish() | |
4725 | */ | |
4726 | skb->protocol = eth->h_proto; | |
76620aaf | 4727 | |
76620aaf HX |
4728 | return skb; |
4729 | } | |
76620aaf | 4730 | |
c7c4b3b6 | 4731 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 4732 | { |
76620aaf | 4733 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
4734 | |
4735 | if (!skb) | |
c7c4b3b6 | 4736 | return GRO_DROP; |
5d0d9be8 | 4737 | |
ae78dbfa BH |
4738 | trace_napi_gro_frags_entry(skb); |
4739 | ||
89c5fa33 | 4740 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 4741 | } |
5d38a079 HX |
4742 | EXPORT_SYMBOL(napi_gro_frags); |
4743 | ||
573e8fca TH |
4744 | /* Compute the checksum from gro_offset and return the folded value |
4745 | * after adding in any pseudo checksum. | |
4746 | */ | |
4747 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
4748 | { | |
4749 | __wsum wsum; | |
4750 | __sum16 sum; | |
4751 | ||
4752 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
4753 | ||
4754 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
4755 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
4756 | if (likely(!sum)) { | |
4757 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
4758 | !skb->csum_complete_sw) | |
4759 | netdev_rx_csum_fault(skb->dev); | |
4760 | } | |
4761 | ||
4762 | NAPI_GRO_CB(skb)->csum = wsum; | |
4763 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4764 | ||
4765 | return sum; | |
4766 | } | |
4767 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
4768 | ||
e326bed2 | 4769 | /* |
855abcf0 | 4770 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
4771 | * Note: called with local irq disabled, but exits with local irq enabled. |
4772 | */ | |
4773 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
4774 | { | |
4775 | #ifdef CONFIG_RPS | |
4776 | struct softnet_data *remsd = sd->rps_ipi_list; | |
4777 | ||
4778 | if (remsd) { | |
4779 | sd->rps_ipi_list = NULL; | |
4780 | ||
4781 | local_irq_enable(); | |
4782 | ||
4783 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
4784 | while (remsd) { | |
4785 | struct softnet_data *next = remsd->rps_ipi_next; | |
4786 | ||
4787 | if (cpu_online(remsd->cpu)) | |
c46fff2a | 4788 | smp_call_function_single_async(remsd->cpu, |
fce8ad15 | 4789 | &remsd->csd); |
e326bed2 ED |
4790 | remsd = next; |
4791 | } | |
4792 | } else | |
4793 | #endif | |
4794 | local_irq_enable(); | |
4795 | } | |
4796 | ||
d75b1ade ED |
4797 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
4798 | { | |
4799 | #ifdef CONFIG_RPS | |
4800 | return sd->rps_ipi_list != NULL; | |
4801 | #else | |
4802 | return false; | |
4803 | #endif | |
4804 | } | |
4805 | ||
bea3348e | 4806 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
4807 | { |
4808 | int work = 0; | |
eecfd7c4 | 4809 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 4810 | |
e326bed2 ED |
4811 | /* Check if we have pending ipi, its better to send them now, |
4812 | * not waiting net_rx_action() end. | |
4813 | */ | |
d75b1ade | 4814 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
4815 | local_irq_disable(); |
4816 | net_rps_action_and_irq_enable(sd); | |
4817 | } | |
d75b1ade | 4818 | |
bea3348e | 4819 | napi->weight = weight_p; |
6e7676c1 | 4820 | local_irq_disable(); |
11ef7a89 | 4821 | while (1) { |
1da177e4 | 4822 | struct sk_buff *skb; |
6e7676c1 CG |
4823 | |
4824 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 4825 | rcu_read_lock(); |
6e7676c1 CG |
4826 | local_irq_enable(); |
4827 | __netif_receive_skb(skb); | |
2c17d27c | 4828 | rcu_read_unlock(); |
6e7676c1 | 4829 | local_irq_disable(); |
76cc8b13 TH |
4830 | input_queue_head_incr(sd); |
4831 | if (++work >= quota) { | |
4832 | local_irq_enable(); | |
4833 | return work; | |
4834 | } | |
6e7676c1 | 4835 | } |
1da177e4 | 4836 | |
e36fa2f7 | 4837 | rps_lock(sd); |
11ef7a89 | 4838 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
4839 | /* |
4840 | * Inline a custom version of __napi_complete(). | |
4841 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
4842 | * and NAPI_STATE_SCHED is the only possible flag set |
4843 | * on backlog. | |
4844 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
4845 | * and we dont need an smp_mb() memory barrier. |
4846 | */ | |
eecfd7c4 | 4847 | napi->state = 0; |
11ef7a89 | 4848 | rps_unlock(sd); |
eecfd7c4 | 4849 | |
11ef7a89 | 4850 | break; |
bea3348e | 4851 | } |
11ef7a89 TH |
4852 | |
4853 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
4854 | &sd->process_queue); | |
e36fa2f7 | 4855 | rps_unlock(sd); |
6e7676c1 CG |
4856 | } |
4857 | local_irq_enable(); | |
1da177e4 | 4858 | |
bea3348e SH |
4859 | return work; |
4860 | } | |
1da177e4 | 4861 | |
bea3348e SH |
4862 | /** |
4863 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 4864 | * @n: entry to schedule |
bea3348e | 4865 | * |
bc9ad166 ED |
4866 | * The entry's receive function will be scheduled to run. |
4867 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 4868 | */ |
b5606c2d | 4869 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
4870 | { |
4871 | unsigned long flags; | |
1da177e4 | 4872 | |
bea3348e | 4873 | local_irq_save(flags); |
903ceff7 | 4874 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 4875 | local_irq_restore(flags); |
1da177e4 | 4876 | } |
bea3348e SH |
4877 | EXPORT_SYMBOL(__napi_schedule); |
4878 | ||
bc9ad166 ED |
4879 | /** |
4880 | * __napi_schedule_irqoff - schedule for receive | |
4881 | * @n: entry to schedule | |
4882 | * | |
4883 | * Variant of __napi_schedule() assuming hard irqs are masked | |
4884 | */ | |
4885 | void __napi_schedule_irqoff(struct napi_struct *n) | |
4886 | { | |
4887 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
4888 | } | |
4889 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
4890 | ||
d565b0a1 HX |
4891 | void __napi_complete(struct napi_struct *n) |
4892 | { | |
4893 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
d565b0a1 | 4894 | |
d75b1ade | 4895 | list_del_init(&n->poll_list); |
4e857c58 | 4896 | smp_mb__before_atomic(); |
d565b0a1 HX |
4897 | clear_bit(NAPI_STATE_SCHED, &n->state); |
4898 | } | |
4899 | EXPORT_SYMBOL(__napi_complete); | |
4900 | ||
3b47d303 | 4901 | void napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 HX |
4902 | { |
4903 | unsigned long flags; | |
4904 | ||
4905 | /* | |
4906 | * don't let napi dequeue from the cpu poll list | |
4907 | * just in case its running on a different cpu | |
4908 | */ | |
4909 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
4910 | return; | |
4911 | ||
3b47d303 ED |
4912 | if (n->gro_list) { |
4913 | unsigned long timeout = 0; | |
d75b1ade | 4914 | |
3b47d303 ED |
4915 | if (work_done) |
4916 | timeout = n->dev->gro_flush_timeout; | |
4917 | ||
4918 | if (timeout) | |
4919 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
4920 | HRTIMER_MODE_REL_PINNED); | |
4921 | else | |
4922 | napi_gro_flush(n, false); | |
4923 | } | |
d75b1ade ED |
4924 | if (likely(list_empty(&n->poll_list))) { |
4925 | WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state)); | |
4926 | } else { | |
4927 | /* If n->poll_list is not empty, we need to mask irqs */ | |
4928 | local_irq_save(flags); | |
4929 | __napi_complete(n); | |
4930 | local_irq_restore(flags); | |
4931 | } | |
d565b0a1 | 4932 | } |
3b47d303 | 4933 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 4934 | |
af12fa6e | 4935 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 4936 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
4937 | { |
4938 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
4939 | struct napi_struct *napi; | |
4940 | ||
4941 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
4942 | if (napi->napi_id == napi_id) | |
4943 | return napi; | |
4944 | ||
4945 | return NULL; | |
4946 | } | |
02d62e86 ED |
4947 | |
4948 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
ce6aea93 | 4949 | #define BUSY_POLL_BUDGET 8 |
02d62e86 ED |
4950 | bool sk_busy_loop(struct sock *sk, int nonblock) |
4951 | { | |
4952 | unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0; | |
ce6aea93 | 4953 | int (*busy_poll)(struct napi_struct *dev); |
02d62e86 ED |
4954 | struct napi_struct *napi; |
4955 | int rc = false; | |
4956 | ||
2a028ecb | 4957 | rcu_read_lock(); |
02d62e86 ED |
4958 | |
4959 | napi = napi_by_id(sk->sk_napi_id); | |
4960 | if (!napi) | |
4961 | goto out; | |
4962 | ||
ce6aea93 ED |
4963 | /* Note: ndo_busy_poll method is optional in linux-4.5 */ |
4964 | busy_poll = napi->dev->netdev_ops->ndo_busy_poll; | |
02d62e86 ED |
4965 | |
4966 | do { | |
ce6aea93 | 4967 | rc = 0; |
2a028ecb | 4968 | local_bh_disable(); |
ce6aea93 ED |
4969 | if (busy_poll) { |
4970 | rc = busy_poll(napi); | |
4971 | } else if (napi_schedule_prep(napi)) { | |
4972 | void *have = netpoll_poll_lock(napi); | |
4973 | ||
4974 | if (test_bit(NAPI_STATE_SCHED, &napi->state)) { | |
4975 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
1db19db7 | 4976 | trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); |
ce6aea93 ED |
4977 | if (rc == BUSY_POLL_BUDGET) { |
4978 | napi_complete_done(napi, rc); | |
4979 | napi_schedule(napi); | |
4980 | } | |
4981 | } | |
4982 | netpoll_poll_unlock(have); | |
4983 | } | |
2a028ecb | 4984 | if (rc > 0) |
02a1d6e7 ED |
4985 | __NET_ADD_STATS(sock_net(sk), |
4986 | LINUX_MIB_BUSYPOLLRXPACKETS, rc); | |
2a028ecb | 4987 | local_bh_enable(); |
02d62e86 ED |
4988 | |
4989 | if (rc == LL_FLUSH_FAILED) | |
4990 | break; /* permanent failure */ | |
4991 | ||
02d62e86 | 4992 | cpu_relax(); |
02d62e86 ED |
4993 | } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) && |
4994 | !need_resched() && !busy_loop_timeout(end_time)); | |
4995 | ||
4996 | rc = !skb_queue_empty(&sk->sk_receive_queue); | |
4997 | out: | |
2a028ecb | 4998 | rcu_read_unlock(); |
02d62e86 ED |
4999 | return rc; |
5000 | } | |
5001 | EXPORT_SYMBOL(sk_busy_loop); | |
5002 | ||
5003 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e ET |
5004 | |
5005 | void napi_hash_add(struct napi_struct *napi) | |
5006 | { | |
d64b5e85 ED |
5007 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
5008 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 5009 | return; |
af12fa6e | 5010 | |
52bd2d62 | 5011 | spin_lock(&napi_hash_lock); |
af12fa6e | 5012 | |
52bd2d62 ED |
5013 | /* 0..NR_CPUS+1 range is reserved for sender_cpu use */ |
5014 | do { | |
5015 | if (unlikely(++napi_gen_id < NR_CPUS + 1)) | |
5016 | napi_gen_id = NR_CPUS + 1; | |
5017 | } while (napi_by_id(napi_gen_id)); | |
5018 | napi->napi_id = napi_gen_id; | |
af12fa6e | 5019 | |
52bd2d62 ED |
5020 | hlist_add_head_rcu(&napi->napi_hash_node, |
5021 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 5022 | |
52bd2d62 | 5023 | spin_unlock(&napi_hash_lock); |
af12fa6e ET |
5024 | } |
5025 | EXPORT_SYMBOL_GPL(napi_hash_add); | |
5026 | ||
5027 | /* Warning : caller is responsible to make sure rcu grace period | |
5028 | * is respected before freeing memory containing @napi | |
5029 | */ | |
34cbe27e | 5030 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 5031 | { |
34cbe27e ED |
5032 | bool rcu_sync_needed = false; |
5033 | ||
af12fa6e ET |
5034 | spin_lock(&napi_hash_lock); |
5035 | ||
34cbe27e ED |
5036 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
5037 | rcu_sync_needed = true; | |
af12fa6e | 5038 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 5039 | } |
af12fa6e | 5040 | spin_unlock(&napi_hash_lock); |
34cbe27e | 5041 | return rcu_sync_needed; |
af12fa6e ET |
5042 | } |
5043 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
5044 | ||
3b47d303 ED |
5045 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
5046 | { | |
5047 | struct napi_struct *napi; | |
5048 | ||
5049 | napi = container_of(timer, struct napi_struct, timer); | |
5050 | if (napi->gro_list) | |
5051 | napi_schedule(napi); | |
5052 | ||
5053 | return HRTIMER_NORESTART; | |
5054 | } | |
5055 | ||
d565b0a1 HX |
5056 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
5057 | int (*poll)(struct napi_struct *, int), int weight) | |
5058 | { | |
5059 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
5060 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
5061 | napi->timer.function = napi_watchdog; | |
4ae5544f | 5062 | napi->gro_count = 0; |
d565b0a1 | 5063 | napi->gro_list = NULL; |
5d38a079 | 5064 | napi->skb = NULL; |
d565b0a1 | 5065 | napi->poll = poll; |
82dc3c63 ED |
5066 | if (weight > NAPI_POLL_WEIGHT) |
5067 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
5068 | weight, dev->name); | |
d565b0a1 HX |
5069 | napi->weight = weight; |
5070 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 5071 | napi->dev = dev; |
5d38a079 | 5072 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
5073 | spin_lock_init(&napi->poll_lock); |
5074 | napi->poll_owner = -1; | |
5075 | #endif | |
5076 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 5077 | napi_hash_add(napi); |
d565b0a1 HX |
5078 | } |
5079 | EXPORT_SYMBOL(netif_napi_add); | |
5080 | ||
3b47d303 ED |
5081 | void napi_disable(struct napi_struct *n) |
5082 | { | |
5083 | might_sleep(); | |
5084 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
5085 | ||
5086 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
5087 | msleep(1); | |
2d8bff12 NH |
5088 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
5089 | msleep(1); | |
3b47d303 ED |
5090 | |
5091 | hrtimer_cancel(&n->timer); | |
5092 | ||
5093 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
5094 | } | |
5095 | EXPORT_SYMBOL(napi_disable); | |
5096 | ||
93d05d4a | 5097 | /* Must be called in process context */ |
d565b0a1 HX |
5098 | void netif_napi_del(struct napi_struct *napi) |
5099 | { | |
93d05d4a ED |
5100 | might_sleep(); |
5101 | if (napi_hash_del(napi)) | |
5102 | synchronize_net(); | |
d7b06636 | 5103 | list_del_init(&napi->dev_list); |
76620aaf | 5104 | napi_free_frags(napi); |
d565b0a1 | 5105 | |
289dccbe | 5106 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 5107 | napi->gro_list = NULL; |
4ae5544f | 5108 | napi->gro_count = 0; |
d565b0a1 HX |
5109 | } |
5110 | EXPORT_SYMBOL(netif_napi_del); | |
5111 | ||
726ce70e HX |
5112 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
5113 | { | |
5114 | void *have; | |
5115 | int work, weight; | |
5116 | ||
5117 | list_del_init(&n->poll_list); | |
5118 | ||
5119 | have = netpoll_poll_lock(n); | |
5120 | ||
5121 | weight = n->weight; | |
5122 | ||
5123 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
5124 | * with netpoll's poll_napi(). Only the entity which | |
5125 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
5126 | * actually make the ->poll() call. Therefore we avoid | |
5127 | * accidentally calling ->poll() when NAPI is not scheduled. | |
5128 | */ | |
5129 | work = 0; | |
5130 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
5131 | work = n->poll(n, weight); | |
1db19db7 | 5132 | trace_napi_poll(n, work, weight); |
726ce70e HX |
5133 | } |
5134 | ||
5135 | WARN_ON_ONCE(work > weight); | |
5136 | ||
5137 | if (likely(work < weight)) | |
5138 | goto out_unlock; | |
5139 | ||
5140 | /* Drivers must not modify the NAPI state if they | |
5141 | * consume the entire weight. In such cases this code | |
5142 | * still "owns" the NAPI instance and therefore can | |
5143 | * move the instance around on the list at-will. | |
5144 | */ | |
5145 | if (unlikely(napi_disable_pending(n))) { | |
5146 | napi_complete(n); | |
5147 | goto out_unlock; | |
5148 | } | |
5149 | ||
5150 | if (n->gro_list) { | |
5151 | /* flush too old packets | |
5152 | * If HZ < 1000, flush all packets. | |
5153 | */ | |
5154 | napi_gro_flush(n, HZ >= 1000); | |
5155 | } | |
5156 | ||
001ce546 HX |
5157 | /* Some drivers may have called napi_schedule |
5158 | * prior to exhausting their budget. | |
5159 | */ | |
5160 | if (unlikely(!list_empty(&n->poll_list))) { | |
5161 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
5162 | n->dev ? n->dev->name : "backlog"); | |
5163 | goto out_unlock; | |
5164 | } | |
5165 | ||
726ce70e HX |
5166 | list_add_tail(&n->poll_list, repoll); |
5167 | ||
5168 | out_unlock: | |
5169 | netpoll_poll_unlock(have); | |
5170 | ||
5171 | return work; | |
5172 | } | |
5173 | ||
1da177e4 LT |
5174 | static void net_rx_action(struct softirq_action *h) |
5175 | { | |
903ceff7 | 5176 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
24f8b238 | 5177 | unsigned long time_limit = jiffies + 2; |
51b0bded | 5178 | int budget = netdev_budget; |
d75b1ade ED |
5179 | LIST_HEAD(list); |
5180 | LIST_HEAD(repoll); | |
53fb95d3 | 5181 | |
1da177e4 | 5182 | local_irq_disable(); |
d75b1ade ED |
5183 | list_splice_init(&sd->poll_list, &list); |
5184 | local_irq_enable(); | |
1da177e4 | 5185 | |
ceb8d5bf | 5186 | for (;;) { |
bea3348e | 5187 | struct napi_struct *n; |
1da177e4 | 5188 | |
ceb8d5bf HX |
5189 | if (list_empty(&list)) { |
5190 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
5191 | return; | |
5192 | break; | |
5193 | } | |
5194 | ||
6bd373eb HX |
5195 | n = list_first_entry(&list, struct napi_struct, poll_list); |
5196 | budget -= napi_poll(n, &repoll); | |
5197 | ||
d75b1ade | 5198 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
5199 | * Allow this to run for 2 jiffies since which will allow |
5200 | * an average latency of 1.5/HZ. | |
bea3348e | 5201 | */ |
ceb8d5bf HX |
5202 | if (unlikely(budget <= 0 || |
5203 | time_after_eq(jiffies, time_limit))) { | |
5204 | sd->time_squeeze++; | |
5205 | break; | |
5206 | } | |
1da177e4 | 5207 | } |
d75b1ade | 5208 | |
795bb1c0 | 5209 | __kfree_skb_flush(); |
d75b1ade ED |
5210 | local_irq_disable(); |
5211 | ||
5212 | list_splice_tail_init(&sd->poll_list, &list); | |
5213 | list_splice_tail(&repoll, &list); | |
5214 | list_splice(&list, &sd->poll_list); | |
5215 | if (!list_empty(&sd->poll_list)) | |
5216 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
5217 | ||
e326bed2 | 5218 | net_rps_action_and_irq_enable(sd); |
1da177e4 LT |
5219 | } |
5220 | ||
aa9d8560 | 5221 | struct netdev_adjacent { |
9ff162a8 | 5222 | struct net_device *dev; |
5d261913 VF |
5223 | |
5224 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 5225 | bool master; |
5d261913 | 5226 | |
5d261913 VF |
5227 | /* counter for the number of times this device was added to us */ |
5228 | u16 ref_nr; | |
5229 | ||
402dae96 VF |
5230 | /* private field for the users */ |
5231 | void *private; | |
5232 | ||
9ff162a8 JP |
5233 | struct list_head list; |
5234 | struct rcu_head rcu; | |
9ff162a8 JP |
5235 | }; |
5236 | ||
6ea29da1 | 5237 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 5238 | struct list_head *adj_list) |
9ff162a8 | 5239 | { |
5d261913 | 5240 | struct netdev_adjacent *adj; |
5d261913 | 5241 | |
2f268f12 | 5242 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
5243 | if (adj->dev == adj_dev) |
5244 | return adj; | |
9ff162a8 JP |
5245 | } |
5246 | return NULL; | |
5247 | } | |
5248 | ||
5249 | /** | |
5250 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
5251 | * @dev: device | |
5252 | * @upper_dev: upper device to check | |
5253 | * | |
5254 | * Find out if a device is linked to specified upper device and return true | |
5255 | * in case it is. Note that this checks only immediate upper device, | |
5256 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
5257 | */ | |
5258 | bool netdev_has_upper_dev(struct net_device *dev, | |
5259 | struct net_device *upper_dev) | |
5260 | { | |
5261 | ASSERT_RTNL(); | |
5262 | ||
6ea29da1 | 5263 | return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper); |
9ff162a8 JP |
5264 | } |
5265 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
5266 | ||
5267 | /** | |
5268 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
5269 | * @dev: device | |
5270 | * | |
5271 | * Find out if a device is linked to an upper device and return true in case | |
5272 | * it is. The caller must hold the RTNL lock. | |
5273 | */ | |
1d143d9f | 5274 | static bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
5275 | { |
5276 | ASSERT_RTNL(); | |
5277 | ||
2f268f12 | 5278 | return !list_empty(&dev->all_adj_list.upper); |
9ff162a8 | 5279 | } |
9ff162a8 JP |
5280 | |
5281 | /** | |
5282 | * netdev_master_upper_dev_get - Get master upper device | |
5283 | * @dev: device | |
5284 | * | |
5285 | * Find a master upper device and return pointer to it or NULL in case | |
5286 | * it's not there. The caller must hold the RTNL lock. | |
5287 | */ | |
5288 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
5289 | { | |
aa9d8560 | 5290 | struct netdev_adjacent *upper; |
9ff162a8 JP |
5291 | |
5292 | ASSERT_RTNL(); | |
5293 | ||
2f268f12 | 5294 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
5295 | return NULL; |
5296 | ||
2f268f12 | 5297 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 5298 | struct netdev_adjacent, list); |
9ff162a8 JP |
5299 | if (likely(upper->master)) |
5300 | return upper->dev; | |
5301 | return NULL; | |
5302 | } | |
5303 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
5304 | ||
b6ccba4c VF |
5305 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
5306 | { | |
5307 | struct netdev_adjacent *adj; | |
5308 | ||
5309 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
5310 | ||
5311 | return adj->private; | |
5312 | } | |
5313 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
5314 | ||
44a40855 VY |
5315 | /** |
5316 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
5317 | * @dev: device | |
5318 | * @iter: list_head ** of the current position | |
5319 | * | |
5320 | * Gets the next device from the dev's upper list, starting from iter | |
5321 | * position. The caller must hold RCU read lock. | |
5322 | */ | |
5323 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
5324 | struct list_head **iter) | |
5325 | { | |
5326 | struct netdev_adjacent *upper; | |
5327 | ||
5328 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5329 | ||
5330 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5331 | ||
5332 | if (&upper->list == &dev->adj_list.upper) | |
5333 | return NULL; | |
5334 | ||
5335 | *iter = &upper->list; | |
5336 | ||
5337 | return upper->dev; | |
5338 | } | |
5339 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
5340 | ||
31088a11 VF |
5341 | /** |
5342 | * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list | |
48311f46 VF |
5343 | * @dev: device |
5344 | * @iter: list_head ** of the current position | |
5345 | * | |
5346 | * Gets the next device from the dev's upper list, starting from iter | |
5347 | * position. The caller must hold RCU read lock. | |
5348 | */ | |
2f268f12 VF |
5349 | struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev, |
5350 | struct list_head **iter) | |
48311f46 VF |
5351 | { |
5352 | struct netdev_adjacent *upper; | |
5353 | ||
85328240 | 5354 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); |
48311f46 VF |
5355 | |
5356 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5357 | ||
2f268f12 | 5358 | if (&upper->list == &dev->all_adj_list.upper) |
48311f46 VF |
5359 | return NULL; |
5360 | ||
5361 | *iter = &upper->list; | |
5362 | ||
5363 | return upper->dev; | |
5364 | } | |
2f268f12 | 5365 | EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu); |
48311f46 | 5366 | |
31088a11 VF |
5367 | /** |
5368 | * netdev_lower_get_next_private - Get the next ->private from the | |
5369 | * lower neighbour list | |
5370 | * @dev: device | |
5371 | * @iter: list_head ** of the current position | |
5372 | * | |
5373 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5374 | * list, starting from iter position. The caller must hold either hold the | |
5375 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 5376 | * list will remain unchanged. |
31088a11 VF |
5377 | */ |
5378 | void *netdev_lower_get_next_private(struct net_device *dev, | |
5379 | struct list_head **iter) | |
5380 | { | |
5381 | struct netdev_adjacent *lower; | |
5382 | ||
5383 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5384 | ||
5385 | if (&lower->list == &dev->adj_list.lower) | |
5386 | return NULL; | |
5387 | ||
6859e7df | 5388 | *iter = lower->list.next; |
31088a11 VF |
5389 | |
5390 | return lower->private; | |
5391 | } | |
5392 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
5393 | ||
5394 | /** | |
5395 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
5396 | * lower neighbour list, RCU | |
5397 | * variant | |
5398 | * @dev: device | |
5399 | * @iter: list_head ** of the current position | |
5400 | * | |
5401 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5402 | * list, starting from iter position. The caller must hold RCU read lock. | |
5403 | */ | |
5404 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
5405 | struct list_head **iter) | |
5406 | { | |
5407 | struct netdev_adjacent *lower; | |
5408 | ||
5409 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5410 | ||
5411 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5412 | ||
5413 | if (&lower->list == &dev->adj_list.lower) | |
5414 | return NULL; | |
5415 | ||
6859e7df | 5416 | *iter = &lower->list; |
31088a11 VF |
5417 | |
5418 | return lower->private; | |
5419 | } | |
5420 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
5421 | ||
4085ebe8 VY |
5422 | /** |
5423 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
5424 | * list | |
5425 | * @dev: device | |
5426 | * @iter: list_head ** of the current position | |
5427 | * | |
5428 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
5429 | * list, starting from iter position. The caller must hold RTNL lock or | |
5430 | * its own locking that guarantees that the neighbour lower | |
b469139e | 5431 | * list will remain unchanged. |
4085ebe8 VY |
5432 | */ |
5433 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5434 | { | |
5435 | struct netdev_adjacent *lower; | |
5436 | ||
cfdd28be | 5437 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
5438 | |
5439 | if (&lower->list == &dev->adj_list.lower) | |
5440 | return NULL; | |
5441 | ||
cfdd28be | 5442 | *iter = lower->list.next; |
4085ebe8 VY |
5443 | |
5444 | return lower->dev; | |
5445 | } | |
5446 | EXPORT_SYMBOL(netdev_lower_get_next); | |
5447 | ||
7ce856aa JP |
5448 | /** |
5449 | * netdev_all_lower_get_next - Get the next device from all lower neighbour list | |
5450 | * @dev: device | |
5451 | * @iter: list_head ** of the current position | |
5452 | * | |
5453 | * Gets the next netdev_adjacent from the dev's all lower neighbour | |
5454 | * list, starting from iter position. The caller must hold RTNL lock or | |
5455 | * its own locking that guarantees that the neighbour all lower | |
5456 | * list will remain unchanged. | |
5457 | */ | |
5458 | struct net_device *netdev_all_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5459 | { | |
5460 | struct netdev_adjacent *lower; | |
5461 | ||
5462 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5463 | ||
5464 | if (&lower->list == &dev->all_adj_list.lower) | |
5465 | return NULL; | |
5466 | ||
5467 | *iter = lower->list.next; | |
5468 | ||
5469 | return lower->dev; | |
5470 | } | |
5471 | EXPORT_SYMBOL(netdev_all_lower_get_next); | |
5472 | ||
5473 | /** | |
5474 | * netdev_all_lower_get_next_rcu - Get the next device from all | |
5475 | * lower neighbour list, RCU variant | |
5476 | * @dev: device | |
5477 | * @iter: list_head ** of the current position | |
5478 | * | |
5479 | * Gets the next netdev_adjacent from the dev's all lower neighbour | |
5480 | * list, starting from iter position. The caller must hold RCU read lock. | |
5481 | */ | |
5482 | struct net_device *netdev_all_lower_get_next_rcu(struct net_device *dev, | |
5483 | struct list_head **iter) | |
5484 | { | |
5485 | struct netdev_adjacent *lower; | |
5486 | ||
5487 | lower = list_first_or_null_rcu(&dev->all_adj_list.lower, | |
5488 | struct netdev_adjacent, list); | |
5489 | ||
5490 | return lower ? lower->dev : NULL; | |
5491 | } | |
5492 | EXPORT_SYMBOL(netdev_all_lower_get_next_rcu); | |
5493 | ||
e001bfad | 5494 | /** |
5495 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
5496 | * lower neighbour list, RCU | |
5497 | * variant | |
5498 | * @dev: device | |
5499 | * | |
5500 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
5501 | * list. The caller must hold RCU read lock. | |
5502 | */ | |
5503 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
5504 | { | |
5505 | struct netdev_adjacent *lower; | |
5506 | ||
5507 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
5508 | struct netdev_adjacent, list); | |
5509 | if (lower) | |
5510 | return lower->private; | |
5511 | return NULL; | |
5512 | } | |
5513 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
5514 | ||
9ff162a8 JP |
5515 | /** |
5516 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
5517 | * @dev: device | |
5518 | * | |
5519 | * Find a master upper device and return pointer to it or NULL in case | |
5520 | * it's not there. The caller must hold the RCU read lock. | |
5521 | */ | |
5522 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
5523 | { | |
aa9d8560 | 5524 | struct netdev_adjacent *upper; |
9ff162a8 | 5525 | |
2f268f12 | 5526 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 5527 | struct netdev_adjacent, list); |
9ff162a8 JP |
5528 | if (upper && likely(upper->master)) |
5529 | return upper->dev; | |
5530 | return NULL; | |
5531 | } | |
5532 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
5533 | ||
0a59f3a9 | 5534 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
5535 | struct net_device *adj_dev, |
5536 | struct list_head *dev_list) | |
5537 | { | |
5538 | char linkname[IFNAMSIZ+7]; | |
5539 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
5540 | "upper_%s" : "lower_%s", adj_dev->name); | |
5541 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
5542 | linkname); | |
5543 | } | |
0a59f3a9 | 5544 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
5545 | char *name, |
5546 | struct list_head *dev_list) | |
5547 | { | |
5548 | char linkname[IFNAMSIZ+7]; | |
5549 | sprintf(linkname, dev_list == &dev->adj_list.upper ? | |
5550 | "upper_%s" : "lower_%s", name); | |
5551 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
5552 | } | |
5553 | ||
7ce64c79 AF |
5554 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
5555 | struct net_device *adj_dev, | |
5556 | struct list_head *dev_list) | |
5557 | { | |
5558 | return (dev_list == &dev->adj_list.upper || | |
5559 | dev_list == &dev->adj_list.lower) && | |
5560 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
5561 | } | |
3ee32707 | 5562 | |
5d261913 VF |
5563 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
5564 | struct net_device *adj_dev, | |
7863c054 | 5565 | struct list_head *dev_list, |
402dae96 | 5566 | void *private, bool master) |
5d261913 VF |
5567 | { |
5568 | struct netdev_adjacent *adj; | |
842d67a7 | 5569 | int ret; |
5d261913 | 5570 | |
6ea29da1 | 5571 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
5572 | |
5573 | if (adj) { | |
5d261913 VF |
5574 | adj->ref_nr++; |
5575 | return 0; | |
5576 | } | |
5577 | ||
5578 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
5579 | if (!adj) | |
5580 | return -ENOMEM; | |
5581 | ||
5582 | adj->dev = adj_dev; | |
5583 | adj->master = master; | |
5d261913 | 5584 | adj->ref_nr = 1; |
402dae96 | 5585 | adj->private = private; |
5d261913 | 5586 | dev_hold(adj_dev); |
2f268f12 VF |
5587 | |
5588 | pr_debug("dev_hold for %s, because of link added from %s to %s\n", | |
5589 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 | 5590 | |
7ce64c79 | 5591 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 5592 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
5593 | if (ret) |
5594 | goto free_adj; | |
5595 | } | |
5596 | ||
7863c054 | 5597 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
5598 | if (master) { |
5599 | ret = sysfs_create_link(&(dev->dev.kobj), | |
5600 | &(adj_dev->dev.kobj), "master"); | |
5601 | if (ret) | |
5831d66e | 5602 | goto remove_symlinks; |
842d67a7 | 5603 | |
7863c054 | 5604 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 5605 | } else { |
7863c054 | 5606 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 5607 | } |
5d261913 VF |
5608 | |
5609 | return 0; | |
842d67a7 | 5610 | |
5831d66e | 5611 | remove_symlinks: |
7ce64c79 | 5612 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5613 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
5614 | free_adj: |
5615 | kfree(adj); | |
974daef7 | 5616 | dev_put(adj_dev); |
842d67a7 VF |
5617 | |
5618 | return ret; | |
5d261913 VF |
5619 | } |
5620 | ||
1d143d9f | 5621 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
5622 | struct net_device *adj_dev, | |
5623 | struct list_head *dev_list) | |
5d261913 VF |
5624 | { |
5625 | struct netdev_adjacent *adj; | |
5626 | ||
6ea29da1 | 5627 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 5628 | |
2f268f12 VF |
5629 | if (!adj) { |
5630 | pr_err("tried to remove device %s from %s\n", | |
5631 | dev->name, adj_dev->name); | |
5d261913 | 5632 | BUG(); |
2f268f12 | 5633 | } |
5d261913 VF |
5634 | |
5635 | if (adj->ref_nr > 1) { | |
2f268f12 VF |
5636 | pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name, |
5637 | adj->ref_nr-1); | |
5d261913 VF |
5638 | adj->ref_nr--; |
5639 | return; | |
5640 | } | |
5641 | ||
842d67a7 VF |
5642 | if (adj->master) |
5643 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
5644 | ||
7ce64c79 | 5645 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 5646 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 5647 | |
5d261913 | 5648 | list_del_rcu(&adj->list); |
2f268f12 VF |
5649 | pr_debug("dev_put for %s, because link removed from %s to %s\n", |
5650 | adj_dev->name, dev->name, adj_dev->name); | |
5d261913 VF |
5651 | dev_put(adj_dev); |
5652 | kfree_rcu(adj, rcu); | |
5653 | } | |
5654 | ||
1d143d9f | 5655 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
5656 | struct net_device *upper_dev, | |
5657 | struct list_head *up_list, | |
5658 | struct list_head *down_list, | |
5659 | void *private, bool master) | |
5d261913 VF |
5660 | { |
5661 | int ret; | |
5662 | ||
402dae96 VF |
5663 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private, |
5664 | master); | |
5d261913 VF |
5665 | if (ret) |
5666 | return ret; | |
5667 | ||
402dae96 VF |
5668 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private, |
5669 | false); | |
5d261913 | 5670 | if (ret) { |
2f268f12 | 5671 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5d261913 VF |
5672 | return ret; |
5673 | } | |
5674 | ||
5675 | return 0; | |
5676 | } | |
5677 | ||
1d143d9f | 5678 | static int __netdev_adjacent_dev_link(struct net_device *dev, |
5679 | struct net_device *upper_dev) | |
5d261913 | 5680 | { |
2f268f12 VF |
5681 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
5682 | &dev->all_adj_list.upper, | |
5683 | &upper_dev->all_adj_list.lower, | |
402dae96 | 5684 | NULL, false); |
5d261913 VF |
5685 | } |
5686 | ||
1d143d9f | 5687 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
5688 | struct net_device *upper_dev, | |
5689 | struct list_head *up_list, | |
5690 | struct list_head *down_list) | |
5d261913 | 5691 | { |
2f268f12 VF |
5692 | __netdev_adjacent_dev_remove(dev, upper_dev, up_list); |
5693 | __netdev_adjacent_dev_remove(upper_dev, dev, down_list); | |
5d261913 VF |
5694 | } |
5695 | ||
1d143d9f | 5696 | static void __netdev_adjacent_dev_unlink(struct net_device *dev, |
5697 | struct net_device *upper_dev) | |
5d261913 | 5698 | { |
2f268f12 VF |
5699 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, |
5700 | &dev->all_adj_list.upper, | |
5701 | &upper_dev->all_adj_list.lower); | |
5702 | } | |
5703 | ||
1d143d9f | 5704 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
5705 | struct net_device *upper_dev, | |
5706 | void *private, bool master) | |
2f268f12 VF |
5707 | { |
5708 | int ret = __netdev_adjacent_dev_link(dev, upper_dev); | |
5709 | ||
5710 | if (ret) | |
5711 | return ret; | |
5712 | ||
5713 | ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, | |
5714 | &dev->adj_list.upper, | |
5715 | &upper_dev->adj_list.lower, | |
402dae96 | 5716 | private, master); |
2f268f12 VF |
5717 | if (ret) { |
5718 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5719 | return ret; | |
5720 | } | |
5721 | ||
5722 | return 0; | |
5d261913 VF |
5723 | } |
5724 | ||
1d143d9f | 5725 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
5726 | struct net_device *upper_dev) | |
2f268f12 VF |
5727 | { |
5728 | __netdev_adjacent_dev_unlink(dev, upper_dev); | |
5729 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, | |
5730 | &dev->adj_list.upper, | |
5731 | &upper_dev->adj_list.lower); | |
5732 | } | |
5d261913 | 5733 | |
9ff162a8 | 5734 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 5735 | struct net_device *upper_dev, bool master, |
29bf24af | 5736 | void *upper_priv, void *upper_info) |
9ff162a8 | 5737 | { |
0e4ead9d | 5738 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 VF |
5739 | struct netdev_adjacent *i, *j, *to_i, *to_j; |
5740 | int ret = 0; | |
9ff162a8 JP |
5741 | |
5742 | ASSERT_RTNL(); | |
5743 | ||
5744 | if (dev == upper_dev) | |
5745 | return -EBUSY; | |
5746 | ||
5747 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
6ea29da1 | 5748 | if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper)) |
9ff162a8 JP |
5749 | return -EBUSY; |
5750 | ||
6ea29da1 | 5751 | if (__netdev_find_adj(upper_dev, &dev->adj_list.upper)) |
9ff162a8 JP |
5752 | return -EEXIST; |
5753 | ||
5754 | if (master && netdev_master_upper_dev_get(dev)) | |
5755 | return -EBUSY; | |
5756 | ||
0e4ead9d JP |
5757 | changeupper_info.upper_dev = upper_dev; |
5758 | changeupper_info.master = master; | |
5759 | changeupper_info.linking = true; | |
29bf24af | 5760 | changeupper_info.upper_info = upper_info; |
0e4ead9d | 5761 | |
573c7ba0 JP |
5762 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
5763 | &changeupper_info.info); | |
5764 | ret = notifier_to_errno(ret); | |
5765 | if (ret) | |
5766 | return ret; | |
5767 | ||
6dffb044 | 5768 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 5769 | master); |
5d261913 VF |
5770 | if (ret) |
5771 | return ret; | |
9ff162a8 | 5772 | |
5d261913 | 5773 | /* Now that we linked these devs, make all the upper_dev's |
2f268f12 | 5774 | * all_adj_list.upper visible to every dev's all_adj_list.lower an |
5d261913 VF |
5775 | * versa, and don't forget the devices itself. All of these |
5776 | * links are non-neighbours. | |
5777 | */ | |
2f268f12 VF |
5778 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5779 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5780 | pr_debug("Interlinking %s with %s, non-neighbour\n", | |
5781 | i->dev->name, j->dev->name); | |
5d261913 VF |
5782 | ret = __netdev_adjacent_dev_link(i->dev, j->dev); |
5783 | if (ret) | |
5784 | goto rollback_mesh; | |
5785 | } | |
5786 | } | |
5787 | ||
5788 | /* add dev to every upper_dev's upper device */ | |
2f268f12 VF |
5789 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5790 | pr_debug("linking %s's upper device %s with %s\n", | |
5791 | upper_dev->name, i->dev->name, dev->name); | |
5d261913 VF |
5792 | ret = __netdev_adjacent_dev_link(dev, i->dev); |
5793 | if (ret) | |
5794 | goto rollback_upper_mesh; | |
5795 | } | |
5796 | ||
5797 | /* add upper_dev to every dev's lower device */ | |
2f268f12 VF |
5798 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5799 | pr_debug("linking %s's lower device %s with %s\n", dev->name, | |
5800 | i->dev->name, upper_dev->name); | |
5d261913 VF |
5801 | ret = __netdev_adjacent_dev_link(i->dev, upper_dev); |
5802 | if (ret) | |
5803 | goto rollback_lower_mesh; | |
5804 | } | |
9ff162a8 | 5805 | |
b03804e7 IS |
5806 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
5807 | &changeupper_info.info); | |
5808 | ret = notifier_to_errno(ret); | |
5809 | if (ret) | |
5810 | goto rollback_lower_mesh; | |
5811 | ||
9ff162a8 | 5812 | return 0; |
5d261913 VF |
5813 | |
5814 | rollback_lower_mesh: | |
5815 | to_i = i; | |
2f268f12 | 5816 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5d261913 VF |
5817 | if (i == to_i) |
5818 | break; | |
5819 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); | |
5820 | } | |
5821 | ||
5822 | i = NULL; | |
5823 | ||
5824 | rollback_upper_mesh: | |
5825 | to_i = i; | |
2f268f12 | 5826 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) { |
5d261913 VF |
5827 | if (i == to_i) |
5828 | break; | |
5829 | __netdev_adjacent_dev_unlink(dev, i->dev); | |
5830 | } | |
5831 | ||
5832 | i = j = NULL; | |
5833 | ||
5834 | rollback_mesh: | |
5835 | to_i = i; | |
5836 | to_j = j; | |
2f268f12 VF |
5837 | list_for_each_entry(i, &dev->all_adj_list.lower, list) { |
5838 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) { | |
5d261913 VF |
5839 | if (i == to_i && j == to_j) |
5840 | break; | |
5841 | __netdev_adjacent_dev_unlink(i->dev, j->dev); | |
5842 | } | |
5843 | if (i == to_i) | |
5844 | break; | |
5845 | } | |
5846 | ||
2f268f12 | 5847 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5848 | |
5849 | return ret; | |
9ff162a8 JP |
5850 | } |
5851 | ||
5852 | /** | |
5853 | * netdev_upper_dev_link - Add a link to the upper device | |
5854 | * @dev: device | |
5855 | * @upper_dev: new upper device | |
5856 | * | |
5857 | * Adds a link to device which is upper to this one. The caller must hold | |
5858 | * the RTNL lock. On a failure a negative errno code is returned. | |
5859 | * On success the reference counts are adjusted and the function | |
5860 | * returns zero. | |
5861 | */ | |
5862 | int netdev_upper_dev_link(struct net_device *dev, | |
5863 | struct net_device *upper_dev) | |
5864 | { | |
29bf24af | 5865 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL); |
9ff162a8 JP |
5866 | } |
5867 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
5868 | ||
5869 | /** | |
5870 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
5871 | * @dev: device | |
5872 | * @upper_dev: new upper device | |
6dffb044 | 5873 | * @upper_priv: upper device private |
29bf24af | 5874 | * @upper_info: upper info to be passed down via notifier |
9ff162a8 JP |
5875 | * |
5876 | * Adds a link to device which is upper to this one. In this case, only | |
5877 | * one master upper device can be linked, although other non-master devices | |
5878 | * might be linked as well. The caller must hold the RTNL lock. | |
5879 | * On a failure a negative errno code is returned. On success the reference | |
5880 | * counts are adjusted and the function returns zero. | |
5881 | */ | |
5882 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 5883 | struct net_device *upper_dev, |
29bf24af | 5884 | void *upper_priv, void *upper_info) |
9ff162a8 | 5885 | { |
29bf24af JP |
5886 | return __netdev_upper_dev_link(dev, upper_dev, true, |
5887 | upper_priv, upper_info); | |
9ff162a8 JP |
5888 | } |
5889 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
5890 | ||
5891 | /** | |
5892 | * netdev_upper_dev_unlink - Removes a link to upper device | |
5893 | * @dev: device | |
5894 | * @upper_dev: new upper device | |
5895 | * | |
5896 | * Removes a link to device which is upper to this one. The caller must hold | |
5897 | * the RTNL lock. | |
5898 | */ | |
5899 | void netdev_upper_dev_unlink(struct net_device *dev, | |
5900 | struct net_device *upper_dev) | |
5901 | { | |
0e4ead9d | 5902 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 | 5903 | struct netdev_adjacent *i, *j; |
9ff162a8 JP |
5904 | ASSERT_RTNL(); |
5905 | ||
0e4ead9d JP |
5906 | changeupper_info.upper_dev = upper_dev; |
5907 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; | |
5908 | changeupper_info.linking = false; | |
5909 | ||
573c7ba0 JP |
5910 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
5911 | &changeupper_info.info); | |
5912 | ||
2f268f12 | 5913 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
5914 | |
5915 | /* Here is the tricky part. We must remove all dev's lower | |
5916 | * devices from all upper_dev's upper devices and vice | |
5917 | * versa, to maintain the graph relationship. | |
5918 | */ | |
2f268f12 VF |
5919 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5920 | list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) | |
5d261913 VF |
5921 | __netdev_adjacent_dev_unlink(i->dev, j->dev); |
5922 | ||
5923 | /* remove also the devices itself from lower/upper device | |
5924 | * list | |
5925 | */ | |
2f268f12 | 5926 | list_for_each_entry(i, &dev->all_adj_list.lower, list) |
5d261913 VF |
5927 | __netdev_adjacent_dev_unlink(i->dev, upper_dev); |
5928 | ||
2f268f12 | 5929 | list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) |
5d261913 VF |
5930 | __netdev_adjacent_dev_unlink(dev, i->dev); |
5931 | ||
0e4ead9d JP |
5932 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
5933 | &changeupper_info.info); | |
9ff162a8 JP |
5934 | } |
5935 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
5936 | ||
61bd3857 MS |
5937 | /** |
5938 | * netdev_bonding_info_change - Dispatch event about slave change | |
5939 | * @dev: device | |
4a26e453 | 5940 | * @bonding_info: info to dispatch |
61bd3857 MS |
5941 | * |
5942 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
5943 | * The caller must hold the RTNL lock. | |
5944 | */ | |
5945 | void netdev_bonding_info_change(struct net_device *dev, | |
5946 | struct netdev_bonding_info *bonding_info) | |
5947 | { | |
5948 | struct netdev_notifier_bonding_info info; | |
5949 | ||
5950 | memcpy(&info.bonding_info, bonding_info, | |
5951 | sizeof(struct netdev_bonding_info)); | |
5952 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, | |
5953 | &info.info); | |
5954 | } | |
5955 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
5956 | ||
2ce1ee17 | 5957 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
5958 | { |
5959 | struct netdev_adjacent *iter; | |
5960 | ||
5961 | struct net *net = dev_net(dev); | |
5962 | ||
5963 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 5964 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
5965 | continue; |
5966 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5967 | &iter->dev->adj_list.lower); | |
5968 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5969 | &dev->adj_list.upper); | |
5970 | } | |
5971 | ||
5972 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 5973 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
5974 | continue; |
5975 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
5976 | &iter->dev->adj_list.upper); | |
5977 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
5978 | &dev->adj_list.lower); | |
5979 | } | |
5980 | } | |
5981 | ||
2ce1ee17 | 5982 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
5983 | { |
5984 | struct netdev_adjacent *iter; | |
5985 | ||
5986 | struct net *net = dev_net(dev); | |
5987 | ||
5988 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 5989 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
5990 | continue; |
5991 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
5992 | &iter->dev->adj_list.lower); | |
5993 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
5994 | &dev->adj_list.upper); | |
5995 | } | |
5996 | ||
5997 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 5998 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
5999 | continue; |
6000 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6001 | &iter->dev->adj_list.upper); | |
6002 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6003 | &dev->adj_list.lower); | |
6004 | } | |
6005 | } | |
6006 | ||
5bb025fa | 6007 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 6008 | { |
5bb025fa | 6009 | struct netdev_adjacent *iter; |
402dae96 | 6010 | |
4c75431a AF |
6011 | struct net *net = dev_net(dev); |
6012 | ||
5bb025fa | 6013 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 6014 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6015 | continue; |
5bb025fa VF |
6016 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6017 | &iter->dev->adj_list.lower); | |
6018 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6019 | &iter->dev->adj_list.lower); | |
6020 | } | |
402dae96 | 6021 | |
5bb025fa | 6022 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 6023 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6024 | continue; |
5bb025fa VF |
6025 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6026 | &iter->dev->adj_list.upper); | |
6027 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6028 | &iter->dev->adj_list.upper); | |
6029 | } | |
402dae96 | 6030 | } |
402dae96 VF |
6031 | |
6032 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
6033 | struct net_device *lower_dev) | |
6034 | { | |
6035 | struct netdev_adjacent *lower; | |
6036 | ||
6037 | if (!lower_dev) | |
6038 | return NULL; | |
6ea29da1 | 6039 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
6040 | if (!lower) |
6041 | return NULL; | |
6042 | ||
6043 | return lower->private; | |
6044 | } | |
6045 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
6046 | ||
4085ebe8 VY |
6047 | |
6048 | int dev_get_nest_level(struct net_device *dev, | |
b618aaa9 | 6049 | bool (*type_check)(const struct net_device *dev)) |
4085ebe8 VY |
6050 | { |
6051 | struct net_device *lower = NULL; | |
6052 | struct list_head *iter; | |
6053 | int max_nest = -1; | |
6054 | int nest; | |
6055 | ||
6056 | ASSERT_RTNL(); | |
6057 | ||
6058 | netdev_for_each_lower_dev(dev, lower, iter) { | |
6059 | nest = dev_get_nest_level(lower, type_check); | |
6060 | if (max_nest < nest) | |
6061 | max_nest = nest; | |
6062 | } | |
6063 | ||
6064 | if (type_check(dev)) | |
6065 | max_nest++; | |
6066 | ||
6067 | return max_nest; | |
6068 | } | |
6069 | EXPORT_SYMBOL(dev_get_nest_level); | |
6070 | ||
04d48266 JP |
6071 | /** |
6072 | * netdev_lower_change - Dispatch event about lower device state change | |
6073 | * @lower_dev: device | |
6074 | * @lower_state_info: state to dispatch | |
6075 | * | |
6076 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
6077 | * The caller must hold the RTNL lock. | |
6078 | */ | |
6079 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
6080 | void *lower_state_info) | |
6081 | { | |
6082 | struct netdev_notifier_changelowerstate_info changelowerstate_info; | |
6083 | ||
6084 | ASSERT_RTNL(); | |
6085 | changelowerstate_info.lower_state_info = lower_state_info; | |
6086 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev, | |
6087 | &changelowerstate_info.info); | |
6088 | } | |
6089 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
6090 | ||
18bfb924 JP |
6091 | int netdev_default_l2upper_neigh_construct(struct net_device *dev, |
6092 | struct neighbour *n) | |
6093 | { | |
6094 | struct net_device *lower_dev, *stop_dev; | |
6095 | struct list_head *iter; | |
6096 | int err; | |
6097 | ||
6098 | netdev_for_each_lower_dev(dev, lower_dev, iter) { | |
6099 | if (!lower_dev->netdev_ops->ndo_neigh_construct) | |
6100 | continue; | |
6101 | err = lower_dev->netdev_ops->ndo_neigh_construct(lower_dev, n); | |
6102 | if (err) { | |
6103 | stop_dev = lower_dev; | |
6104 | goto rollback; | |
6105 | } | |
6106 | } | |
6107 | return 0; | |
6108 | ||
6109 | rollback: | |
6110 | netdev_for_each_lower_dev(dev, lower_dev, iter) { | |
6111 | if (lower_dev == stop_dev) | |
6112 | break; | |
6113 | if (!lower_dev->netdev_ops->ndo_neigh_destroy) | |
6114 | continue; | |
6115 | lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n); | |
6116 | } | |
6117 | return err; | |
6118 | } | |
6119 | EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_construct); | |
6120 | ||
6121 | void netdev_default_l2upper_neigh_destroy(struct net_device *dev, | |
6122 | struct neighbour *n) | |
6123 | { | |
6124 | struct net_device *lower_dev; | |
6125 | struct list_head *iter; | |
6126 | ||
6127 | netdev_for_each_lower_dev(dev, lower_dev, iter) { | |
6128 | if (!lower_dev->netdev_ops->ndo_neigh_destroy) | |
6129 | continue; | |
6130 | lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n); | |
6131 | } | |
6132 | } | |
6133 | EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_destroy); | |
6134 | ||
b6c40d68 PM |
6135 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
6136 | { | |
d314774c SH |
6137 | const struct net_device_ops *ops = dev->netdev_ops; |
6138 | ||
d2615bf4 | 6139 | if (ops->ndo_change_rx_flags) |
d314774c | 6140 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
6141 | } |
6142 | ||
991fb3f7 | 6143 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6144 | { |
b536db93 | 6145 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
6146 | kuid_t uid; |
6147 | kgid_t gid; | |
1da177e4 | 6148 | |
24023451 PM |
6149 | ASSERT_RTNL(); |
6150 | ||
dad9b335 WC |
6151 | dev->flags |= IFF_PROMISC; |
6152 | dev->promiscuity += inc; | |
6153 | if (dev->promiscuity == 0) { | |
6154 | /* | |
6155 | * Avoid overflow. | |
6156 | * If inc causes overflow, untouch promisc and return error. | |
6157 | */ | |
6158 | if (inc < 0) | |
6159 | dev->flags &= ~IFF_PROMISC; | |
6160 | else { | |
6161 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
6162 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
6163 | dev->name); | |
dad9b335 WC |
6164 | return -EOVERFLOW; |
6165 | } | |
6166 | } | |
52609c0b | 6167 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
6168 | pr_info("device %s %s promiscuous mode\n", |
6169 | dev->name, | |
6170 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
6171 | if (audit_enabled) { |
6172 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
6173 | audit_log(current->audit_context, GFP_ATOMIC, |
6174 | AUDIT_ANOM_PROMISCUOUS, | |
6175 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
6176 | dev->name, (dev->flags & IFF_PROMISC), | |
6177 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 6178 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
6179 | from_kuid(&init_user_ns, uid), |
6180 | from_kgid(&init_user_ns, gid), | |
7759db82 | 6181 | audit_get_sessionid(current)); |
8192b0c4 | 6182 | } |
24023451 | 6183 | |
b6c40d68 | 6184 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 6185 | } |
991fb3f7 ND |
6186 | if (notify) |
6187 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 6188 | return 0; |
1da177e4 LT |
6189 | } |
6190 | ||
4417da66 PM |
6191 | /** |
6192 | * dev_set_promiscuity - update promiscuity count on a device | |
6193 | * @dev: device | |
6194 | * @inc: modifier | |
6195 | * | |
6196 | * Add or remove promiscuity from a device. While the count in the device | |
6197 | * remains above zero the interface remains promiscuous. Once it hits zero | |
6198 | * the device reverts back to normal filtering operation. A negative inc | |
6199 | * value is used to drop promiscuity on the device. | |
dad9b335 | 6200 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 6201 | */ |
dad9b335 | 6202 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 6203 | { |
b536db93 | 6204 | unsigned int old_flags = dev->flags; |
dad9b335 | 6205 | int err; |
4417da66 | 6206 | |
991fb3f7 | 6207 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 6208 | if (err < 0) |
dad9b335 | 6209 | return err; |
4417da66 PM |
6210 | if (dev->flags != old_flags) |
6211 | dev_set_rx_mode(dev); | |
dad9b335 | 6212 | return err; |
4417da66 | 6213 | } |
d1b19dff | 6214 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 6215 | |
991fb3f7 | 6216 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6217 | { |
991fb3f7 | 6218 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 6219 | |
24023451 PM |
6220 | ASSERT_RTNL(); |
6221 | ||
1da177e4 | 6222 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
6223 | dev->allmulti += inc; |
6224 | if (dev->allmulti == 0) { | |
6225 | /* | |
6226 | * Avoid overflow. | |
6227 | * If inc causes overflow, untouch allmulti and return error. | |
6228 | */ | |
6229 | if (inc < 0) | |
6230 | dev->flags &= ~IFF_ALLMULTI; | |
6231 | else { | |
6232 | dev->allmulti -= inc; | |
7b6cd1ce JP |
6233 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
6234 | dev->name); | |
dad9b335 WC |
6235 | return -EOVERFLOW; |
6236 | } | |
6237 | } | |
24023451 | 6238 | if (dev->flags ^ old_flags) { |
b6c40d68 | 6239 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 6240 | dev_set_rx_mode(dev); |
991fb3f7 ND |
6241 | if (notify) |
6242 | __dev_notify_flags(dev, old_flags, | |
6243 | dev->gflags ^ old_gflags); | |
24023451 | 6244 | } |
dad9b335 | 6245 | return 0; |
4417da66 | 6246 | } |
991fb3f7 ND |
6247 | |
6248 | /** | |
6249 | * dev_set_allmulti - update allmulti count on a device | |
6250 | * @dev: device | |
6251 | * @inc: modifier | |
6252 | * | |
6253 | * Add or remove reception of all multicast frames to a device. While the | |
6254 | * count in the device remains above zero the interface remains listening | |
6255 | * to all interfaces. Once it hits zero the device reverts back to normal | |
6256 | * filtering operation. A negative @inc value is used to drop the counter | |
6257 | * when releasing a resource needing all multicasts. | |
6258 | * Return 0 if successful or a negative errno code on error. | |
6259 | */ | |
6260 | ||
6261 | int dev_set_allmulti(struct net_device *dev, int inc) | |
6262 | { | |
6263 | return __dev_set_allmulti(dev, inc, true); | |
6264 | } | |
d1b19dff | 6265 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
6266 | |
6267 | /* | |
6268 | * Upload unicast and multicast address lists to device and | |
6269 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 6270 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
6271 | * are present. |
6272 | */ | |
6273 | void __dev_set_rx_mode(struct net_device *dev) | |
6274 | { | |
d314774c SH |
6275 | const struct net_device_ops *ops = dev->netdev_ops; |
6276 | ||
4417da66 PM |
6277 | /* dev_open will call this function so the list will stay sane. */ |
6278 | if (!(dev->flags&IFF_UP)) | |
6279 | return; | |
6280 | ||
6281 | if (!netif_device_present(dev)) | |
40b77c94 | 6282 | return; |
4417da66 | 6283 | |
01789349 | 6284 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
6285 | /* Unicast addresses changes may only happen under the rtnl, |
6286 | * therefore calling __dev_set_promiscuity here is safe. | |
6287 | */ | |
32e7bfc4 | 6288 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 6289 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 6290 | dev->uc_promisc = true; |
32e7bfc4 | 6291 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 6292 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 6293 | dev->uc_promisc = false; |
4417da66 | 6294 | } |
4417da66 | 6295 | } |
01789349 JP |
6296 | |
6297 | if (ops->ndo_set_rx_mode) | |
6298 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
6299 | } |
6300 | ||
6301 | void dev_set_rx_mode(struct net_device *dev) | |
6302 | { | |
b9e40857 | 6303 | netif_addr_lock_bh(dev); |
4417da66 | 6304 | __dev_set_rx_mode(dev); |
b9e40857 | 6305 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
6306 | } |
6307 | ||
f0db275a SH |
6308 | /** |
6309 | * dev_get_flags - get flags reported to userspace | |
6310 | * @dev: device | |
6311 | * | |
6312 | * Get the combination of flag bits exported through APIs to userspace. | |
6313 | */ | |
95c96174 | 6314 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 6315 | { |
95c96174 | 6316 | unsigned int flags; |
1da177e4 LT |
6317 | |
6318 | flags = (dev->flags & ~(IFF_PROMISC | | |
6319 | IFF_ALLMULTI | | |
b00055aa SR |
6320 | IFF_RUNNING | |
6321 | IFF_LOWER_UP | | |
6322 | IFF_DORMANT)) | | |
1da177e4 LT |
6323 | (dev->gflags & (IFF_PROMISC | |
6324 | IFF_ALLMULTI)); | |
6325 | ||
b00055aa SR |
6326 | if (netif_running(dev)) { |
6327 | if (netif_oper_up(dev)) | |
6328 | flags |= IFF_RUNNING; | |
6329 | if (netif_carrier_ok(dev)) | |
6330 | flags |= IFF_LOWER_UP; | |
6331 | if (netif_dormant(dev)) | |
6332 | flags |= IFF_DORMANT; | |
6333 | } | |
1da177e4 LT |
6334 | |
6335 | return flags; | |
6336 | } | |
d1b19dff | 6337 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 6338 | |
bd380811 | 6339 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 6340 | { |
b536db93 | 6341 | unsigned int old_flags = dev->flags; |
bd380811 | 6342 | int ret; |
1da177e4 | 6343 | |
24023451 PM |
6344 | ASSERT_RTNL(); |
6345 | ||
1da177e4 LT |
6346 | /* |
6347 | * Set the flags on our device. | |
6348 | */ | |
6349 | ||
6350 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
6351 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
6352 | IFF_AUTOMEDIA)) | | |
6353 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
6354 | IFF_ALLMULTI)); | |
6355 | ||
6356 | /* | |
6357 | * Load in the correct multicast list now the flags have changed. | |
6358 | */ | |
6359 | ||
b6c40d68 PM |
6360 | if ((old_flags ^ flags) & IFF_MULTICAST) |
6361 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 6362 | |
4417da66 | 6363 | dev_set_rx_mode(dev); |
1da177e4 LT |
6364 | |
6365 | /* | |
6366 | * Have we downed the interface. We handle IFF_UP ourselves | |
6367 | * according to user attempts to set it, rather than blindly | |
6368 | * setting it. | |
6369 | */ | |
6370 | ||
6371 | ret = 0; | |
d215d10f | 6372 | if ((old_flags ^ flags) & IFF_UP) |
bd380811 | 6373 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 | 6374 | |
1da177e4 | 6375 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 6376 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 6377 | unsigned int old_flags = dev->flags; |
d1b19dff | 6378 | |
1da177e4 | 6379 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
6380 | |
6381 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
6382 | if (dev->flags != old_flags) | |
6383 | dev_set_rx_mode(dev); | |
1da177e4 LT |
6384 | } |
6385 | ||
6386 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
6387 | is important. Some (broken) drivers set IFF_PROMISC, when | |
6388 | IFF_ALLMULTI is requested not asking us and not reporting. | |
6389 | */ | |
6390 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
6391 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
6392 | ||
1da177e4 | 6393 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 6394 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
6395 | } |
6396 | ||
bd380811 PM |
6397 | return ret; |
6398 | } | |
6399 | ||
a528c219 ND |
6400 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
6401 | unsigned int gchanges) | |
bd380811 PM |
6402 | { |
6403 | unsigned int changes = dev->flags ^ old_flags; | |
6404 | ||
a528c219 | 6405 | if (gchanges) |
7f294054 | 6406 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 6407 | |
bd380811 PM |
6408 | if (changes & IFF_UP) { |
6409 | if (dev->flags & IFF_UP) | |
6410 | call_netdevice_notifiers(NETDEV_UP, dev); | |
6411 | else | |
6412 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
6413 | } | |
6414 | ||
6415 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
6416 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
6417 | struct netdev_notifier_change_info change_info; | |
6418 | ||
6419 | change_info.flags_changed = changes; | |
6420 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
6421 | &change_info.info); | |
6422 | } | |
bd380811 PM |
6423 | } |
6424 | ||
6425 | /** | |
6426 | * dev_change_flags - change device settings | |
6427 | * @dev: device | |
6428 | * @flags: device state flags | |
6429 | * | |
6430 | * Change settings on device based state flags. The flags are | |
6431 | * in the userspace exported format. | |
6432 | */ | |
b536db93 | 6433 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 6434 | { |
b536db93 | 6435 | int ret; |
991fb3f7 | 6436 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
6437 | |
6438 | ret = __dev_change_flags(dev, flags); | |
6439 | if (ret < 0) | |
6440 | return ret; | |
6441 | ||
991fb3f7 | 6442 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 6443 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
6444 | return ret; |
6445 | } | |
d1b19dff | 6446 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 6447 | |
2315dc91 VF |
6448 | static int __dev_set_mtu(struct net_device *dev, int new_mtu) |
6449 | { | |
6450 | const struct net_device_ops *ops = dev->netdev_ops; | |
6451 | ||
6452 | if (ops->ndo_change_mtu) | |
6453 | return ops->ndo_change_mtu(dev, new_mtu); | |
6454 | ||
6455 | dev->mtu = new_mtu; | |
6456 | return 0; | |
6457 | } | |
6458 | ||
f0db275a SH |
6459 | /** |
6460 | * dev_set_mtu - Change maximum transfer unit | |
6461 | * @dev: device | |
6462 | * @new_mtu: new transfer unit | |
6463 | * | |
6464 | * Change the maximum transfer size of the network device. | |
6465 | */ | |
1da177e4 LT |
6466 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
6467 | { | |
2315dc91 | 6468 | int err, orig_mtu; |
1da177e4 LT |
6469 | |
6470 | if (new_mtu == dev->mtu) | |
6471 | return 0; | |
6472 | ||
6473 | /* MTU must be positive. */ | |
6474 | if (new_mtu < 0) | |
6475 | return -EINVAL; | |
6476 | ||
6477 | if (!netif_device_present(dev)) | |
6478 | return -ENODEV; | |
6479 | ||
1d486bfb VF |
6480 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
6481 | err = notifier_to_errno(err); | |
6482 | if (err) | |
6483 | return err; | |
d314774c | 6484 | |
2315dc91 VF |
6485 | orig_mtu = dev->mtu; |
6486 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 6487 | |
2315dc91 VF |
6488 | if (!err) { |
6489 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6490 | err = notifier_to_errno(err); | |
6491 | if (err) { | |
6492 | /* setting mtu back and notifying everyone again, | |
6493 | * so that they have a chance to revert changes. | |
6494 | */ | |
6495 | __dev_set_mtu(dev, orig_mtu); | |
6496 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6497 | } | |
6498 | } | |
1da177e4 LT |
6499 | return err; |
6500 | } | |
d1b19dff | 6501 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 6502 | |
cbda10fa VD |
6503 | /** |
6504 | * dev_set_group - Change group this device belongs to | |
6505 | * @dev: device | |
6506 | * @new_group: group this device should belong to | |
6507 | */ | |
6508 | void dev_set_group(struct net_device *dev, int new_group) | |
6509 | { | |
6510 | dev->group = new_group; | |
6511 | } | |
6512 | EXPORT_SYMBOL(dev_set_group); | |
6513 | ||
f0db275a SH |
6514 | /** |
6515 | * dev_set_mac_address - Change Media Access Control Address | |
6516 | * @dev: device | |
6517 | * @sa: new address | |
6518 | * | |
6519 | * Change the hardware (MAC) address of the device | |
6520 | */ | |
1da177e4 LT |
6521 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
6522 | { | |
d314774c | 6523 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
6524 | int err; |
6525 | ||
d314774c | 6526 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
6527 | return -EOPNOTSUPP; |
6528 | if (sa->sa_family != dev->type) | |
6529 | return -EINVAL; | |
6530 | if (!netif_device_present(dev)) | |
6531 | return -ENODEV; | |
d314774c | 6532 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
6533 | if (err) |
6534 | return err; | |
fbdeca2d | 6535 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 6536 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 6537 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 6538 | return 0; |
1da177e4 | 6539 | } |
d1b19dff | 6540 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 6541 | |
4bf84c35 JP |
6542 | /** |
6543 | * dev_change_carrier - Change device carrier | |
6544 | * @dev: device | |
691b3b7e | 6545 | * @new_carrier: new value |
4bf84c35 JP |
6546 | * |
6547 | * Change device carrier | |
6548 | */ | |
6549 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
6550 | { | |
6551 | const struct net_device_ops *ops = dev->netdev_ops; | |
6552 | ||
6553 | if (!ops->ndo_change_carrier) | |
6554 | return -EOPNOTSUPP; | |
6555 | if (!netif_device_present(dev)) | |
6556 | return -ENODEV; | |
6557 | return ops->ndo_change_carrier(dev, new_carrier); | |
6558 | } | |
6559 | EXPORT_SYMBOL(dev_change_carrier); | |
6560 | ||
66b52b0d JP |
6561 | /** |
6562 | * dev_get_phys_port_id - Get device physical port ID | |
6563 | * @dev: device | |
6564 | * @ppid: port ID | |
6565 | * | |
6566 | * Get device physical port ID | |
6567 | */ | |
6568 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 6569 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
6570 | { |
6571 | const struct net_device_ops *ops = dev->netdev_ops; | |
6572 | ||
6573 | if (!ops->ndo_get_phys_port_id) | |
6574 | return -EOPNOTSUPP; | |
6575 | return ops->ndo_get_phys_port_id(dev, ppid); | |
6576 | } | |
6577 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
6578 | ||
db24a904 DA |
6579 | /** |
6580 | * dev_get_phys_port_name - Get device physical port name | |
6581 | * @dev: device | |
6582 | * @name: port name | |
ed49e650 | 6583 | * @len: limit of bytes to copy to name |
db24a904 DA |
6584 | * |
6585 | * Get device physical port name | |
6586 | */ | |
6587 | int dev_get_phys_port_name(struct net_device *dev, | |
6588 | char *name, size_t len) | |
6589 | { | |
6590 | const struct net_device_ops *ops = dev->netdev_ops; | |
6591 | ||
6592 | if (!ops->ndo_get_phys_port_name) | |
6593 | return -EOPNOTSUPP; | |
6594 | return ops->ndo_get_phys_port_name(dev, name, len); | |
6595 | } | |
6596 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
6597 | ||
d746d707 AK |
6598 | /** |
6599 | * dev_change_proto_down - update protocol port state information | |
6600 | * @dev: device | |
6601 | * @proto_down: new value | |
6602 | * | |
6603 | * This info can be used by switch drivers to set the phys state of the | |
6604 | * port. | |
6605 | */ | |
6606 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
6607 | { | |
6608 | const struct net_device_ops *ops = dev->netdev_ops; | |
6609 | ||
6610 | if (!ops->ndo_change_proto_down) | |
6611 | return -EOPNOTSUPP; | |
6612 | if (!netif_device_present(dev)) | |
6613 | return -ENODEV; | |
6614 | return ops->ndo_change_proto_down(dev, proto_down); | |
6615 | } | |
6616 | EXPORT_SYMBOL(dev_change_proto_down); | |
6617 | ||
a7862b45 BB |
6618 | /** |
6619 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
6620 | * @dev: device | |
6621 | * @fd: new program fd or negative value to clear | |
6622 | * | |
6623 | * Set or clear a bpf program for a device | |
6624 | */ | |
6625 | int dev_change_xdp_fd(struct net_device *dev, int fd) | |
6626 | { | |
6627 | const struct net_device_ops *ops = dev->netdev_ops; | |
6628 | struct bpf_prog *prog = NULL; | |
6629 | struct netdev_xdp xdp = {}; | |
6630 | int err; | |
6631 | ||
6632 | if (!ops->ndo_xdp) | |
6633 | return -EOPNOTSUPP; | |
6634 | if (fd >= 0) { | |
6635 | prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP); | |
6636 | if (IS_ERR(prog)) | |
6637 | return PTR_ERR(prog); | |
6638 | } | |
6639 | ||
6640 | xdp.command = XDP_SETUP_PROG; | |
6641 | xdp.prog = prog; | |
6642 | err = ops->ndo_xdp(dev, &xdp); | |
6643 | if (err < 0 && prog) | |
6644 | bpf_prog_put(prog); | |
6645 | ||
6646 | return err; | |
6647 | } | |
6648 | EXPORT_SYMBOL(dev_change_xdp_fd); | |
6649 | ||
1da177e4 LT |
6650 | /** |
6651 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 6652 | * @net: the applicable net namespace |
1da177e4 LT |
6653 | * |
6654 | * Returns a suitable unique value for a new device interface | |
6655 | * number. The caller must hold the rtnl semaphore or the | |
6656 | * dev_base_lock to be sure it remains unique. | |
6657 | */ | |
881d966b | 6658 | static int dev_new_index(struct net *net) |
1da177e4 | 6659 | { |
aa79e66e | 6660 | int ifindex = net->ifindex; |
1da177e4 LT |
6661 | for (;;) { |
6662 | if (++ifindex <= 0) | |
6663 | ifindex = 1; | |
881d966b | 6664 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 6665 | return net->ifindex = ifindex; |
1da177e4 LT |
6666 | } |
6667 | } | |
6668 | ||
1da177e4 | 6669 | /* Delayed registration/unregisteration */ |
3b5b34fd | 6670 | static LIST_HEAD(net_todo_list); |
200b916f | 6671 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 6672 | |
6f05f629 | 6673 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 6674 | { |
1da177e4 | 6675 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 6676 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
6677 | } |
6678 | ||
9b5e383c | 6679 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 6680 | { |
e93737b0 | 6681 | struct net_device *dev, *tmp; |
5cde2829 | 6682 | LIST_HEAD(close_head); |
9b5e383c | 6683 | |
93ee31f1 DL |
6684 | BUG_ON(dev_boot_phase); |
6685 | ASSERT_RTNL(); | |
6686 | ||
e93737b0 | 6687 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 6688 | /* Some devices call without registering |
e93737b0 KK |
6689 | * for initialization unwind. Remove those |
6690 | * devices and proceed with the remaining. | |
9b5e383c ED |
6691 | */ |
6692 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
6693 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
6694 | dev->name, dev); | |
93ee31f1 | 6695 | |
9b5e383c | 6696 | WARN_ON(1); |
e93737b0 KK |
6697 | list_del(&dev->unreg_list); |
6698 | continue; | |
9b5e383c | 6699 | } |
449f4544 | 6700 | dev->dismantle = true; |
9b5e383c | 6701 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 6702 | } |
93ee31f1 | 6703 | |
44345724 | 6704 | /* If device is running, close it first. */ |
5cde2829 EB |
6705 | list_for_each_entry(dev, head, unreg_list) |
6706 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 6707 | dev_close_many(&close_head, true); |
93ee31f1 | 6708 | |
44345724 | 6709 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
6710 | /* And unlink it from device chain. */ |
6711 | unlist_netdevice(dev); | |
93ee31f1 | 6712 | |
9b5e383c | 6713 | dev->reg_state = NETREG_UNREGISTERING; |
e9e4dd32 | 6714 | on_each_cpu(flush_backlog, dev, 1); |
9b5e383c | 6715 | } |
93ee31f1 DL |
6716 | |
6717 | synchronize_net(); | |
6718 | ||
9b5e383c | 6719 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
6720 | struct sk_buff *skb = NULL; |
6721 | ||
9b5e383c ED |
6722 | /* Shutdown queueing discipline. */ |
6723 | dev_shutdown(dev); | |
93ee31f1 DL |
6724 | |
6725 | ||
9b5e383c ED |
6726 | /* Notify protocols, that we are about to destroy |
6727 | this device. They should clean all the things. | |
6728 | */ | |
6729 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 6730 | |
395eea6c MB |
6731 | if (!dev->rtnl_link_ops || |
6732 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
6733 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, | |
6734 | GFP_KERNEL); | |
6735 | ||
9b5e383c ED |
6736 | /* |
6737 | * Flush the unicast and multicast chains | |
6738 | */ | |
a748ee24 | 6739 | dev_uc_flush(dev); |
22bedad3 | 6740 | dev_mc_flush(dev); |
93ee31f1 | 6741 | |
9b5e383c ED |
6742 | if (dev->netdev_ops->ndo_uninit) |
6743 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 6744 | |
395eea6c MB |
6745 | if (skb) |
6746 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 6747 | |
9ff162a8 JP |
6748 | /* Notifier chain MUST detach us all upper devices. */ |
6749 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
93ee31f1 | 6750 | |
9b5e383c ED |
6751 | /* Remove entries from kobject tree */ |
6752 | netdev_unregister_kobject(dev); | |
024e9679 AD |
6753 | #ifdef CONFIG_XPS |
6754 | /* Remove XPS queueing entries */ | |
6755 | netif_reset_xps_queues_gt(dev, 0); | |
6756 | #endif | |
9b5e383c | 6757 | } |
93ee31f1 | 6758 | |
850a545b | 6759 | synchronize_net(); |
395264d5 | 6760 | |
a5ee1551 | 6761 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
6762 | dev_put(dev); |
6763 | } | |
6764 | ||
6765 | static void rollback_registered(struct net_device *dev) | |
6766 | { | |
6767 | LIST_HEAD(single); | |
6768 | ||
6769 | list_add(&dev->unreg_list, &single); | |
6770 | rollback_registered_many(&single); | |
ceaaec98 | 6771 | list_del(&single); |
93ee31f1 DL |
6772 | } |
6773 | ||
fd867d51 JW |
6774 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
6775 | struct net_device *upper, netdev_features_t features) | |
6776 | { | |
6777 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
6778 | netdev_features_t feature; | |
5ba3f7d6 | 6779 | int feature_bit; |
fd867d51 | 6780 | |
5ba3f7d6 JW |
6781 | for_each_netdev_feature(&upper_disables, feature_bit) { |
6782 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
6783 | if (!(upper->wanted_features & feature) |
6784 | && (features & feature)) { | |
6785 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
6786 | &feature, upper->name); | |
6787 | features &= ~feature; | |
6788 | } | |
6789 | } | |
6790 | ||
6791 | return features; | |
6792 | } | |
6793 | ||
6794 | static void netdev_sync_lower_features(struct net_device *upper, | |
6795 | struct net_device *lower, netdev_features_t features) | |
6796 | { | |
6797 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
6798 | netdev_features_t feature; | |
5ba3f7d6 | 6799 | int feature_bit; |
fd867d51 | 6800 | |
5ba3f7d6 JW |
6801 | for_each_netdev_feature(&upper_disables, feature_bit) { |
6802 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
6803 | if (!(features & feature) && (lower->features & feature)) { |
6804 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
6805 | &feature, lower->name); | |
6806 | lower->wanted_features &= ~feature; | |
6807 | netdev_update_features(lower); | |
6808 | ||
6809 | if (unlikely(lower->features & feature)) | |
6810 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
6811 | &feature, lower->name); | |
6812 | } | |
6813 | } | |
6814 | } | |
6815 | ||
c8f44aff MM |
6816 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
6817 | netdev_features_t features) | |
b63365a2 | 6818 | { |
57422dc5 MM |
6819 | /* Fix illegal checksum combinations */ |
6820 | if ((features & NETIF_F_HW_CSUM) && | |
6821 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6822 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
6823 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
6824 | } | |
6825 | ||
b63365a2 | 6826 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 6827 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 6828 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 6829 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
6830 | } |
6831 | ||
ec5f0615 PS |
6832 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
6833 | !(features & NETIF_F_IP_CSUM)) { | |
6834 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
6835 | features &= ~NETIF_F_TSO; | |
6836 | features &= ~NETIF_F_TSO_ECN; | |
6837 | } | |
6838 | ||
6839 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
6840 | !(features & NETIF_F_IPV6_CSUM)) { | |
6841 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
6842 | features &= ~NETIF_F_TSO6; | |
6843 | } | |
6844 | ||
b1dc497b AD |
6845 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
6846 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
6847 | features &= ~NETIF_F_TSO_MANGLEID; | |
6848 | ||
31d8b9e0 BH |
6849 | /* TSO ECN requires that TSO is present as well. */ |
6850 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
6851 | features &= ~NETIF_F_TSO_ECN; | |
6852 | ||
212b573f MM |
6853 | /* Software GSO depends on SG. */ |
6854 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 6855 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
6856 | features &= ~NETIF_F_GSO; |
6857 | } | |
6858 | ||
acd1130e | 6859 | /* UFO needs SG and checksumming */ |
b63365a2 | 6860 | if (features & NETIF_F_UFO) { |
79032644 | 6861 | /* maybe split UFO into V4 and V6? */ |
c8cd0989 TH |
6862 | if (!(features & NETIF_F_HW_CSUM) && |
6863 | ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) != | |
6864 | (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 6865 | netdev_dbg(dev, |
acd1130e | 6866 | "Dropping NETIF_F_UFO since no checksum offload features.\n"); |
b63365a2 HX |
6867 | features &= ~NETIF_F_UFO; |
6868 | } | |
6869 | ||
6870 | if (!(features & NETIF_F_SG)) { | |
6f404e44 | 6871 | netdev_dbg(dev, |
acd1130e | 6872 | "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n"); |
b63365a2 HX |
6873 | features &= ~NETIF_F_UFO; |
6874 | } | |
6875 | } | |
6876 | ||
802ab55a AD |
6877 | /* GSO partial features require GSO partial be set */ |
6878 | if ((features & dev->gso_partial_features) && | |
6879 | !(features & NETIF_F_GSO_PARTIAL)) { | |
6880 | netdev_dbg(dev, | |
6881 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
6882 | features &= ~dev->gso_partial_features; | |
6883 | } | |
6884 | ||
d0290214 JP |
6885 | #ifdef CONFIG_NET_RX_BUSY_POLL |
6886 | if (dev->netdev_ops->ndo_busy_poll) | |
6887 | features |= NETIF_F_BUSY_POLL; | |
6888 | else | |
6889 | #endif | |
6890 | features &= ~NETIF_F_BUSY_POLL; | |
6891 | ||
b63365a2 HX |
6892 | return features; |
6893 | } | |
b63365a2 | 6894 | |
6cb6a27c | 6895 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 6896 | { |
fd867d51 | 6897 | struct net_device *upper, *lower; |
c8f44aff | 6898 | netdev_features_t features; |
fd867d51 | 6899 | struct list_head *iter; |
e7868a85 | 6900 | int err = -1; |
5455c699 | 6901 | |
87267485 MM |
6902 | ASSERT_RTNL(); |
6903 | ||
5455c699 MM |
6904 | features = netdev_get_wanted_features(dev); |
6905 | ||
6906 | if (dev->netdev_ops->ndo_fix_features) | |
6907 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
6908 | ||
6909 | /* driver might be less strict about feature dependencies */ | |
6910 | features = netdev_fix_features(dev, features); | |
6911 | ||
fd867d51 JW |
6912 | /* some features can't be enabled if they're off an an upper device */ |
6913 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
6914 | features = netdev_sync_upper_features(dev, upper, features); | |
6915 | ||
5455c699 | 6916 | if (dev->features == features) |
e7868a85 | 6917 | goto sync_lower; |
5455c699 | 6918 | |
c8f44aff MM |
6919 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
6920 | &dev->features, &features); | |
5455c699 MM |
6921 | |
6922 | if (dev->netdev_ops->ndo_set_features) | |
6923 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
6924 | else |
6925 | err = 0; | |
5455c699 | 6926 | |
6cb6a27c | 6927 | if (unlikely(err < 0)) { |
5455c699 | 6928 | netdev_err(dev, |
c8f44aff MM |
6929 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
6930 | err, &features, &dev->features); | |
17b85d29 NA |
6931 | /* return non-0 since some features might have changed and |
6932 | * it's better to fire a spurious notification than miss it | |
6933 | */ | |
6934 | return -1; | |
6cb6a27c MM |
6935 | } |
6936 | ||
e7868a85 | 6937 | sync_lower: |
fd867d51 JW |
6938 | /* some features must be disabled on lower devices when disabled |
6939 | * on an upper device (think: bonding master or bridge) | |
6940 | */ | |
6941 | netdev_for_each_lower_dev(dev, lower, iter) | |
6942 | netdev_sync_lower_features(dev, lower, features); | |
6943 | ||
6cb6a27c MM |
6944 | if (!err) |
6945 | dev->features = features; | |
6946 | ||
e7868a85 | 6947 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
6948 | } |
6949 | ||
afe12cc8 MM |
6950 | /** |
6951 | * netdev_update_features - recalculate device features | |
6952 | * @dev: the device to check | |
6953 | * | |
6954 | * Recalculate dev->features set and send notifications if it | |
6955 | * has changed. Should be called after driver or hardware dependent | |
6956 | * conditions might have changed that influence the features. | |
6957 | */ | |
6cb6a27c MM |
6958 | void netdev_update_features(struct net_device *dev) |
6959 | { | |
6960 | if (__netdev_update_features(dev)) | |
6961 | netdev_features_change(dev); | |
5455c699 MM |
6962 | } |
6963 | EXPORT_SYMBOL(netdev_update_features); | |
6964 | ||
afe12cc8 MM |
6965 | /** |
6966 | * netdev_change_features - recalculate device features | |
6967 | * @dev: the device to check | |
6968 | * | |
6969 | * Recalculate dev->features set and send notifications even | |
6970 | * if they have not changed. Should be called instead of | |
6971 | * netdev_update_features() if also dev->vlan_features might | |
6972 | * have changed to allow the changes to be propagated to stacked | |
6973 | * VLAN devices. | |
6974 | */ | |
6975 | void netdev_change_features(struct net_device *dev) | |
6976 | { | |
6977 | __netdev_update_features(dev); | |
6978 | netdev_features_change(dev); | |
6979 | } | |
6980 | EXPORT_SYMBOL(netdev_change_features); | |
6981 | ||
fc4a7489 PM |
6982 | /** |
6983 | * netif_stacked_transfer_operstate - transfer operstate | |
6984 | * @rootdev: the root or lower level device to transfer state from | |
6985 | * @dev: the device to transfer operstate to | |
6986 | * | |
6987 | * Transfer operational state from root to device. This is normally | |
6988 | * called when a stacking relationship exists between the root | |
6989 | * device and the device(a leaf device). | |
6990 | */ | |
6991 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
6992 | struct net_device *dev) | |
6993 | { | |
6994 | if (rootdev->operstate == IF_OPER_DORMANT) | |
6995 | netif_dormant_on(dev); | |
6996 | else | |
6997 | netif_dormant_off(dev); | |
6998 | ||
6999 | if (netif_carrier_ok(rootdev)) { | |
7000 | if (!netif_carrier_ok(dev)) | |
7001 | netif_carrier_on(dev); | |
7002 | } else { | |
7003 | if (netif_carrier_ok(dev)) | |
7004 | netif_carrier_off(dev); | |
7005 | } | |
7006 | } | |
7007 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
7008 | ||
a953be53 | 7009 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
7010 | static int netif_alloc_rx_queues(struct net_device *dev) |
7011 | { | |
1b4bf461 | 7012 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 7013 | struct netdev_rx_queue *rx; |
10595902 | 7014 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 7015 | |
bd25fa7b | 7016 | BUG_ON(count < 1); |
1b4bf461 | 7017 | |
10595902 PG |
7018 | rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
7019 | if (!rx) { | |
7020 | rx = vzalloc(sz); | |
7021 | if (!rx) | |
7022 | return -ENOMEM; | |
7023 | } | |
bd25fa7b TH |
7024 | dev->_rx = rx; |
7025 | ||
bd25fa7b | 7026 | for (i = 0; i < count; i++) |
fe822240 | 7027 | rx[i].dev = dev; |
1b4bf461 ED |
7028 | return 0; |
7029 | } | |
bf264145 | 7030 | #endif |
1b4bf461 | 7031 | |
aa942104 CG |
7032 | static void netdev_init_one_queue(struct net_device *dev, |
7033 | struct netdev_queue *queue, void *_unused) | |
7034 | { | |
7035 | /* Initialize queue lock */ | |
7036 | spin_lock_init(&queue->_xmit_lock); | |
7037 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
7038 | queue->xmit_lock_owner = -1; | |
b236da69 | 7039 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 7040 | queue->dev = dev; |
114cf580 TH |
7041 | #ifdef CONFIG_BQL |
7042 | dql_init(&queue->dql, HZ); | |
7043 | #endif | |
aa942104 CG |
7044 | } |
7045 | ||
60877a32 ED |
7046 | static void netif_free_tx_queues(struct net_device *dev) |
7047 | { | |
4cb28970 | 7048 | kvfree(dev->_tx); |
60877a32 ED |
7049 | } |
7050 | ||
e6484930 TH |
7051 | static int netif_alloc_netdev_queues(struct net_device *dev) |
7052 | { | |
7053 | unsigned int count = dev->num_tx_queues; | |
7054 | struct netdev_queue *tx; | |
60877a32 | 7055 | size_t sz = count * sizeof(*tx); |
e6484930 | 7056 | |
d339727c ED |
7057 | if (count < 1 || count > 0xffff) |
7058 | return -EINVAL; | |
62b5942a | 7059 | |
60877a32 ED |
7060 | tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
7061 | if (!tx) { | |
7062 | tx = vzalloc(sz); | |
7063 | if (!tx) | |
7064 | return -ENOMEM; | |
7065 | } | |
e6484930 | 7066 | dev->_tx = tx; |
1d24eb48 | 7067 | |
e6484930 TH |
7068 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
7069 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
7070 | |
7071 | return 0; | |
e6484930 TH |
7072 | } |
7073 | ||
a2029240 DV |
7074 | void netif_tx_stop_all_queues(struct net_device *dev) |
7075 | { | |
7076 | unsigned int i; | |
7077 | ||
7078 | for (i = 0; i < dev->num_tx_queues; i++) { | |
7079 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
7080 | netif_tx_stop_queue(txq); | |
7081 | } | |
7082 | } | |
7083 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
7084 | ||
1da177e4 LT |
7085 | /** |
7086 | * register_netdevice - register a network device | |
7087 | * @dev: device to register | |
7088 | * | |
7089 | * Take a completed network device structure and add it to the kernel | |
7090 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7091 | * chain. 0 is returned on success. A negative errno code is returned | |
7092 | * on a failure to set up the device, or if the name is a duplicate. | |
7093 | * | |
7094 | * Callers must hold the rtnl semaphore. You may want | |
7095 | * register_netdev() instead of this. | |
7096 | * | |
7097 | * BUGS: | |
7098 | * The locking appears insufficient to guarantee two parallel registers | |
7099 | * will not get the same name. | |
7100 | */ | |
7101 | ||
7102 | int register_netdevice(struct net_device *dev) | |
7103 | { | |
1da177e4 | 7104 | int ret; |
d314774c | 7105 | struct net *net = dev_net(dev); |
1da177e4 LT |
7106 | |
7107 | BUG_ON(dev_boot_phase); | |
7108 | ASSERT_RTNL(); | |
7109 | ||
b17a7c17 SH |
7110 | might_sleep(); |
7111 | ||
1da177e4 LT |
7112 | /* When net_device's are persistent, this will be fatal. */ |
7113 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 7114 | BUG_ON(!net); |
1da177e4 | 7115 | |
f1f28aa3 | 7116 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 7117 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 7118 | |
828de4f6 | 7119 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
7120 | if (ret < 0) |
7121 | goto out; | |
7122 | ||
1da177e4 | 7123 | /* Init, if this function is available */ |
d314774c SH |
7124 | if (dev->netdev_ops->ndo_init) { |
7125 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
7126 | if (ret) { |
7127 | if (ret > 0) | |
7128 | ret = -EIO; | |
90833aa4 | 7129 | goto out; |
1da177e4 LT |
7130 | } |
7131 | } | |
4ec93edb | 7132 | |
f646968f PM |
7133 | if (((dev->hw_features | dev->features) & |
7134 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
7135 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
7136 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
7137 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
7138 | ret = -EINVAL; | |
7139 | goto err_uninit; | |
7140 | } | |
7141 | ||
9c7dafbf PE |
7142 | ret = -EBUSY; |
7143 | if (!dev->ifindex) | |
7144 | dev->ifindex = dev_new_index(net); | |
7145 | else if (__dev_get_by_index(net, dev->ifindex)) | |
7146 | goto err_uninit; | |
7147 | ||
5455c699 MM |
7148 | /* Transfer changeable features to wanted_features and enable |
7149 | * software offloads (GSO and GRO). | |
7150 | */ | |
7151 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f MM |
7152 | dev->features |= NETIF_F_SOFT_FEATURES; |
7153 | dev->wanted_features = dev->features & dev->hw_features; | |
1da177e4 | 7154 | |
cbc53e08 | 7155 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 7156 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 7157 | |
7f348a60 AD |
7158 | /* If IPv4 TCP segmentation offload is supported we should also |
7159 | * allow the device to enable segmenting the frame with the option | |
7160 | * of ignoring a static IP ID value. This doesn't enable the | |
7161 | * feature itself but allows the user to enable it later. | |
7162 | */ | |
cbc53e08 AD |
7163 | if (dev->hw_features & NETIF_F_TSO) |
7164 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
7165 | if (dev->vlan_features & NETIF_F_TSO) |
7166 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
7167 | if (dev->mpls_features & NETIF_F_TSO) | |
7168 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
7169 | if (dev->hw_enc_features & NETIF_F_TSO) | |
7170 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 7171 | |
1180e7d6 | 7172 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 7173 | */ |
1180e7d6 | 7174 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 7175 | |
ee579677 PS |
7176 | /* Make NETIF_F_SG inheritable to tunnel devices. |
7177 | */ | |
802ab55a | 7178 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 7179 | |
0d89d203 SH |
7180 | /* Make NETIF_F_SG inheritable to MPLS. |
7181 | */ | |
7182 | dev->mpls_features |= NETIF_F_SG; | |
7183 | ||
7ffbe3fd JB |
7184 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
7185 | ret = notifier_to_errno(ret); | |
7186 | if (ret) | |
7187 | goto err_uninit; | |
7188 | ||
8b41d188 | 7189 | ret = netdev_register_kobject(dev); |
b17a7c17 | 7190 | if (ret) |
7ce1b0ed | 7191 | goto err_uninit; |
b17a7c17 SH |
7192 | dev->reg_state = NETREG_REGISTERED; |
7193 | ||
6cb6a27c | 7194 | __netdev_update_features(dev); |
8e9b59b2 | 7195 | |
1da177e4 LT |
7196 | /* |
7197 | * Default initial state at registry is that the | |
7198 | * device is present. | |
7199 | */ | |
7200 | ||
7201 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7202 | ||
8f4cccbb BH |
7203 | linkwatch_init_dev(dev); |
7204 | ||
1da177e4 | 7205 | dev_init_scheduler(dev); |
1da177e4 | 7206 | dev_hold(dev); |
ce286d32 | 7207 | list_netdevice(dev); |
7bf23575 | 7208 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 7209 | |
948b337e JP |
7210 | /* If the device has permanent device address, driver should |
7211 | * set dev_addr and also addr_assign_type should be set to | |
7212 | * NET_ADDR_PERM (default value). | |
7213 | */ | |
7214 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
7215 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
7216 | ||
1da177e4 | 7217 | /* Notify protocols, that a new device appeared. */ |
056925ab | 7218 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 7219 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
7220 | if (ret) { |
7221 | rollback_registered(dev); | |
7222 | dev->reg_state = NETREG_UNREGISTERED; | |
7223 | } | |
d90a909e EB |
7224 | /* |
7225 | * Prevent userspace races by waiting until the network | |
7226 | * device is fully setup before sending notifications. | |
7227 | */ | |
a2835763 PM |
7228 | if (!dev->rtnl_link_ops || |
7229 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 7230 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
7231 | |
7232 | out: | |
7233 | return ret; | |
7ce1b0ed HX |
7234 | |
7235 | err_uninit: | |
d314774c SH |
7236 | if (dev->netdev_ops->ndo_uninit) |
7237 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 7238 | goto out; |
1da177e4 | 7239 | } |
d1b19dff | 7240 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 7241 | |
937f1ba5 BH |
7242 | /** |
7243 | * init_dummy_netdev - init a dummy network device for NAPI | |
7244 | * @dev: device to init | |
7245 | * | |
7246 | * This takes a network device structure and initialize the minimum | |
7247 | * amount of fields so it can be used to schedule NAPI polls without | |
7248 | * registering a full blown interface. This is to be used by drivers | |
7249 | * that need to tie several hardware interfaces to a single NAPI | |
7250 | * poll scheduler due to HW limitations. | |
7251 | */ | |
7252 | int init_dummy_netdev(struct net_device *dev) | |
7253 | { | |
7254 | /* Clear everything. Note we don't initialize spinlocks | |
7255 | * are they aren't supposed to be taken by any of the | |
7256 | * NAPI code and this dummy netdev is supposed to be | |
7257 | * only ever used for NAPI polls | |
7258 | */ | |
7259 | memset(dev, 0, sizeof(struct net_device)); | |
7260 | ||
7261 | /* make sure we BUG if trying to hit standard | |
7262 | * register/unregister code path | |
7263 | */ | |
7264 | dev->reg_state = NETREG_DUMMY; | |
7265 | ||
937f1ba5 BH |
7266 | /* NAPI wants this */ |
7267 | INIT_LIST_HEAD(&dev->napi_list); | |
7268 | ||
7269 | /* a dummy interface is started by default */ | |
7270 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7271 | set_bit(__LINK_STATE_START, &dev->state); | |
7272 | ||
29b4433d ED |
7273 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
7274 | * because users of this 'device' dont need to change | |
7275 | * its refcount. | |
7276 | */ | |
7277 | ||
937f1ba5 BH |
7278 | return 0; |
7279 | } | |
7280 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
7281 | ||
7282 | ||
1da177e4 LT |
7283 | /** |
7284 | * register_netdev - register a network device | |
7285 | * @dev: device to register | |
7286 | * | |
7287 | * Take a completed network device structure and add it to the kernel | |
7288 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7289 | * chain. 0 is returned on success. A negative errno code is returned | |
7290 | * on a failure to set up the device, or if the name is a duplicate. | |
7291 | * | |
38b4da38 | 7292 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
7293 | * and expands the device name if you passed a format string to |
7294 | * alloc_netdev. | |
7295 | */ | |
7296 | int register_netdev(struct net_device *dev) | |
7297 | { | |
7298 | int err; | |
7299 | ||
7300 | rtnl_lock(); | |
1da177e4 | 7301 | err = register_netdevice(dev); |
1da177e4 LT |
7302 | rtnl_unlock(); |
7303 | return err; | |
7304 | } | |
7305 | EXPORT_SYMBOL(register_netdev); | |
7306 | ||
29b4433d ED |
7307 | int netdev_refcnt_read(const struct net_device *dev) |
7308 | { | |
7309 | int i, refcnt = 0; | |
7310 | ||
7311 | for_each_possible_cpu(i) | |
7312 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
7313 | return refcnt; | |
7314 | } | |
7315 | EXPORT_SYMBOL(netdev_refcnt_read); | |
7316 | ||
2c53040f | 7317 | /** |
1da177e4 | 7318 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 7319 | * @dev: target net_device |
1da177e4 LT |
7320 | * |
7321 | * This is called when unregistering network devices. | |
7322 | * | |
7323 | * Any protocol or device that holds a reference should register | |
7324 | * for netdevice notification, and cleanup and put back the | |
7325 | * reference if they receive an UNREGISTER event. | |
7326 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 7327 | * call dev_put. |
1da177e4 LT |
7328 | */ |
7329 | static void netdev_wait_allrefs(struct net_device *dev) | |
7330 | { | |
7331 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 7332 | int refcnt; |
1da177e4 | 7333 | |
e014debe ED |
7334 | linkwatch_forget_dev(dev); |
7335 | ||
1da177e4 | 7336 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
7337 | refcnt = netdev_refcnt_read(dev); |
7338 | ||
7339 | while (refcnt != 0) { | |
1da177e4 | 7340 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 7341 | rtnl_lock(); |
1da177e4 LT |
7342 | |
7343 | /* Rebroadcast unregister notification */ | |
056925ab | 7344 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 7345 | |
748e2d93 | 7346 | __rtnl_unlock(); |
0115e8e3 | 7347 | rcu_barrier(); |
748e2d93 ED |
7348 | rtnl_lock(); |
7349 | ||
0115e8e3 | 7350 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
7351 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
7352 | &dev->state)) { | |
7353 | /* We must not have linkwatch events | |
7354 | * pending on unregister. If this | |
7355 | * happens, we simply run the queue | |
7356 | * unscheduled, resulting in a noop | |
7357 | * for this device. | |
7358 | */ | |
7359 | linkwatch_run_queue(); | |
7360 | } | |
7361 | ||
6756ae4b | 7362 | __rtnl_unlock(); |
1da177e4 LT |
7363 | |
7364 | rebroadcast_time = jiffies; | |
7365 | } | |
7366 | ||
7367 | msleep(250); | |
7368 | ||
29b4433d ED |
7369 | refcnt = netdev_refcnt_read(dev); |
7370 | ||
1da177e4 | 7371 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
7372 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
7373 | dev->name, refcnt); | |
1da177e4 LT |
7374 | warning_time = jiffies; |
7375 | } | |
7376 | } | |
7377 | } | |
7378 | ||
7379 | /* The sequence is: | |
7380 | * | |
7381 | * rtnl_lock(); | |
7382 | * ... | |
7383 | * register_netdevice(x1); | |
7384 | * register_netdevice(x2); | |
7385 | * ... | |
7386 | * unregister_netdevice(y1); | |
7387 | * unregister_netdevice(y2); | |
7388 | * ... | |
7389 | * rtnl_unlock(); | |
7390 | * free_netdev(y1); | |
7391 | * free_netdev(y2); | |
7392 | * | |
58ec3b4d | 7393 | * We are invoked by rtnl_unlock(). |
1da177e4 | 7394 | * This allows us to deal with problems: |
b17a7c17 | 7395 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
7396 | * without deadlocking with linkwatch via keventd. |
7397 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
7398 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
7399 | * |
7400 | * We must not return until all unregister events added during | |
7401 | * the interval the lock was held have been completed. | |
1da177e4 | 7402 | */ |
1da177e4 LT |
7403 | void netdev_run_todo(void) |
7404 | { | |
626ab0e6 | 7405 | struct list_head list; |
1da177e4 | 7406 | |
1da177e4 | 7407 | /* Snapshot list, allow later requests */ |
626ab0e6 | 7408 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
7409 | |
7410 | __rtnl_unlock(); | |
626ab0e6 | 7411 | |
0115e8e3 ED |
7412 | |
7413 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
7414 | if (!list_empty(&list)) |
7415 | rcu_barrier(); | |
7416 | ||
1da177e4 LT |
7417 | while (!list_empty(&list)) { |
7418 | struct net_device *dev | |
e5e26d75 | 7419 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
7420 | list_del(&dev->todo_list); |
7421 | ||
748e2d93 | 7422 | rtnl_lock(); |
0115e8e3 | 7423 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 7424 | __rtnl_unlock(); |
0115e8e3 | 7425 | |
b17a7c17 | 7426 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 7427 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
7428 | dev->name, dev->reg_state); |
7429 | dump_stack(); | |
7430 | continue; | |
7431 | } | |
1da177e4 | 7432 | |
b17a7c17 | 7433 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 7434 | |
b17a7c17 | 7435 | netdev_wait_allrefs(dev); |
1da177e4 | 7436 | |
b17a7c17 | 7437 | /* paranoia */ |
29b4433d | 7438 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
7439 | BUG_ON(!list_empty(&dev->ptype_all)); |
7440 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
7441 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
7442 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 7443 | WARN_ON(dev->dn_ptr); |
1da177e4 | 7444 | |
b17a7c17 SH |
7445 | if (dev->destructor) |
7446 | dev->destructor(dev); | |
9093bbb2 | 7447 | |
50624c93 EB |
7448 | /* Report a network device has been unregistered */ |
7449 | rtnl_lock(); | |
7450 | dev_net(dev)->dev_unreg_count--; | |
7451 | __rtnl_unlock(); | |
7452 | wake_up(&netdev_unregistering_wq); | |
7453 | ||
9093bbb2 SH |
7454 | /* Free network device */ |
7455 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 7456 | } |
1da177e4 LT |
7457 | } |
7458 | ||
9256645a JW |
7459 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
7460 | * all the same fields in the same order as net_device_stats, with only | |
7461 | * the type differing, but rtnl_link_stats64 may have additional fields | |
7462 | * at the end for newer counters. | |
3cfde79c | 7463 | */ |
77a1abf5 ED |
7464 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
7465 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
7466 | { |
7467 | #if BITS_PER_LONG == 64 | |
9256645a | 7468 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
77a1abf5 | 7469 | memcpy(stats64, netdev_stats, sizeof(*stats64)); |
9256645a JW |
7470 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7471 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
7472 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 7473 | #else |
9256645a | 7474 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
7475 | const unsigned long *src = (const unsigned long *)netdev_stats; |
7476 | u64 *dst = (u64 *)stats64; | |
7477 | ||
9256645a | 7478 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
7479 | for (i = 0; i < n; i++) |
7480 | dst[i] = src[i]; | |
9256645a JW |
7481 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7482 | memset((char *)stats64 + n * sizeof(u64), 0, | |
7483 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
7484 | #endif |
7485 | } | |
77a1abf5 | 7486 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 7487 | |
eeda3fd6 SH |
7488 | /** |
7489 | * dev_get_stats - get network device statistics | |
7490 | * @dev: device to get statistics from | |
28172739 | 7491 | * @storage: place to store stats |
eeda3fd6 | 7492 | * |
d7753516 BH |
7493 | * Get network statistics from device. Return @storage. |
7494 | * The device driver may provide its own method by setting | |
7495 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
7496 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 7497 | */ |
d7753516 BH |
7498 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
7499 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 7500 | { |
eeda3fd6 SH |
7501 | const struct net_device_ops *ops = dev->netdev_ops; |
7502 | ||
28172739 ED |
7503 | if (ops->ndo_get_stats64) { |
7504 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
7505 | ops->ndo_get_stats64(dev, storage); |
7506 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 7507 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
7508 | } else { |
7509 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 7510 | } |
caf586e5 | 7511 | storage->rx_dropped += atomic_long_read(&dev->rx_dropped); |
015f0688 | 7512 | storage->tx_dropped += atomic_long_read(&dev->tx_dropped); |
6e7333d3 | 7513 | storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler); |
28172739 | 7514 | return storage; |
c45d286e | 7515 | } |
eeda3fd6 | 7516 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 7517 | |
24824a09 | 7518 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 7519 | { |
24824a09 | 7520 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 7521 | |
24824a09 ED |
7522 | #ifdef CONFIG_NET_CLS_ACT |
7523 | if (queue) | |
7524 | return queue; | |
7525 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
7526 | if (!queue) | |
7527 | return NULL; | |
7528 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 7529 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
7530 | queue->qdisc_sleeping = &noop_qdisc; |
7531 | rcu_assign_pointer(dev->ingress_queue, queue); | |
7532 | #endif | |
7533 | return queue; | |
bb949fbd DM |
7534 | } |
7535 | ||
2c60db03 ED |
7536 | static const struct ethtool_ops default_ethtool_ops; |
7537 | ||
d07d7507 SG |
7538 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
7539 | const struct ethtool_ops *ops) | |
7540 | { | |
7541 | if (dev->ethtool_ops == &default_ethtool_ops) | |
7542 | dev->ethtool_ops = ops; | |
7543 | } | |
7544 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
7545 | ||
74d332c1 ED |
7546 | void netdev_freemem(struct net_device *dev) |
7547 | { | |
7548 | char *addr = (char *)dev - dev->padded; | |
7549 | ||
4cb28970 | 7550 | kvfree(addr); |
74d332c1 ED |
7551 | } |
7552 | ||
1da177e4 | 7553 | /** |
36909ea4 | 7554 | * alloc_netdev_mqs - allocate network device |
c835a677 TG |
7555 | * @sizeof_priv: size of private data to allocate space for |
7556 | * @name: device name format string | |
7557 | * @name_assign_type: origin of device name | |
7558 | * @setup: callback to initialize device | |
7559 | * @txqs: the number of TX subqueues to allocate | |
7560 | * @rxqs: the number of RX subqueues to allocate | |
1da177e4 LT |
7561 | * |
7562 | * Allocates a struct net_device with private data area for driver use | |
90e51adf | 7563 | * and performs basic initialization. Also allocates subqueue structs |
36909ea4 | 7564 | * for each queue on the device. |
1da177e4 | 7565 | */ |
36909ea4 | 7566 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 7567 | unsigned char name_assign_type, |
36909ea4 TH |
7568 | void (*setup)(struct net_device *), |
7569 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 7570 | { |
1da177e4 | 7571 | struct net_device *dev; |
7943986c | 7572 | size_t alloc_size; |
1ce8e7b5 | 7573 | struct net_device *p; |
1da177e4 | 7574 | |
b6fe17d6 SH |
7575 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
7576 | ||
36909ea4 | 7577 | if (txqs < 1) { |
7b6cd1ce | 7578 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
7579 | return NULL; |
7580 | } | |
7581 | ||
a953be53 | 7582 | #ifdef CONFIG_SYSFS |
36909ea4 | 7583 | if (rxqs < 1) { |
7b6cd1ce | 7584 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
7585 | return NULL; |
7586 | } | |
7587 | #endif | |
7588 | ||
fd2ea0a7 | 7589 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
7590 | if (sizeof_priv) { |
7591 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 7592 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
7593 | alloc_size += sizeof_priv; |
7594 | } | |
7595 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 7596 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 7597 | |
74d332c1 ED |
7598 | p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
7599 | if (!p) | |
7600 | p = vzalloc(alloc_size); | |
62b5942a | 7601 | if (!p) |
1da177e4 | 7602 | return NULL; |
1da177e4 | 7603 | |
1ce8e7b5 | 7604 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 7605 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 7606 | |
29b4433d ED |
7607 | dev->pcpu_refcnt = alloc_percpu(int); |
7608 | if (!dev->pcpu_refcnt) | |
74d332c1 | 7609 | goto free_dev; |
ab9c73cc | 7610 | |
ab9c73cc | 7611 | if (dev_addr_init(dev)) |
29b4433d | 7612 | goto free_pcpu; |
ab9c73cc | 7613 | |
22bedad3 | 7614 | dev_mc_init(dev); |
a748ee24 | 7615 | dev_uc_init(dev); |
ccffad25 | 7616 | |
c346dca1 | 7617 | dev_net_set(dev, &init_net); |
1da177e4 | 7618 | |
8d3bdbd5 | 7619 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 7620 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 7621 | |
8d3bdbd5 DM |
7622 | INIT_LIST_HEAD(&dev->napi_list); |
7623 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 7624 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 7625 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
7626 | INIT_LIST_HEAD(&dev->adj_list.upper); |
7627 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7628 | INIT_LIST_HEAD(&dev->all_adj_list.upper); | |
7629 | INIT_LIST_HEAD(&dev->all_adj_list.lower); | |
7866a621 SN |
7630 | INIT_LIST_HEAD(&dev->ptype_all); |
7631 | INIT_LIST_HEAD(&dev->ptype_specific); | |
02875878 | 7632 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
7633 | setup(dev); |
7634 | ||
a813104d | 7635 | if (!dev->tx_queue_len) { |
f84bb1ea | 7636 | dev->priv_flags |= IFF_NO_QUEUE; |
a813104d PS |
7637 | dev->tx_queue_len = 1; |
7638 | } | |
906470c1 | 7639 | |
36909ea4 TH |
7640 | dev->num_tx_queues = txqs; |
7641 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 7642 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 7643 | goto free_all; |
e8a0464c | 7644 | |
a953be53 | 7645 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
7646 | dev->num_rx_queues = rxqs; |
7647 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 7648 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 7649 | goto free_all; |
df334545 | 7650 | #endif |
0a9627f2 | 7651 | |
1da177e4 | 7652 | strcpy(dev->name, name); |
c835a677 | 7653 | dev->name_assign_type = name_assign_type; |
cbda10fa | 7654 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
7655 | if (!dev->ethtool_ops) |
7656 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
7657 | |
7658 | nf_hook_ingress_init(dev); | |
7659 | ||
1da177e4 | 7660 | return dev; |
ab9c73cc | 7661 | |
8d3bdbd5 DM |
7662 | free_all: |
7663 | free_netdev(dev); | |
7664 | return NULL; | |
7665 | ||
29b4433d ED |
7666 | free_pcpu: |
7667 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
7668 | free_dev: |
7669 | netdev_freemem(dev); | |
ab9c73cc | 7670 | return NULL; |
1da177e4 | 7671 | } |
36909ea4 | 7672 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
7673 | |
7674 | /** | |
7675 | * free_netdev - free network device | |
7676 | * @dev: device | |
7677 | * | |
4ec93edb YH |
7678 | * This function does the last stage of destroying an allocated device |
7679 | * interface. The reference to the device object is released. | |
1da177e4 | 7680 | * If this is the last reference then it will be freed. |
93d05d4a | 7681 | * Must be called in process context. |
1da177e4 LT |
7682 | */ |
7683 | void free_netdev(struct net_device *dev) | |
7684 | { | |
d565b0a1 HX |
7685 | struct napi_struct *p, *n; |
7686 | ||
93d05d4a | 7687 | might_sleep(); |
60877a32 | 7688 | netif_free_tx_queues(dev); |
a953be53 | 7689 | #ifdef CONFIG_SYSFS |
10595902 | 7690 | kvfree(dev->_rx); |
fe822240 | 7691 | #endif |
e8a0464c | 7692 | |
33d480ce | 7693 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 7694 | |
f001fde5 JP |
7695 | /* Flush device addresses */ |
7696 | dev_addr_flush(dev); | |
7697 | ||
d565b0a1 HX |
7698 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
7699 | netif_napi_del(p); | |
7700 | ||
29b4433d ED |
7701 | free_percpu(dev->pcpu_refcnt); |
7702 | dev->pcpu_refcnt = NULL; | |
7703 | ||
3041a069 | 7704 | /* Compatibility with error handling in drivers */ |
1da177e4 | 7705 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 7706 | netdev_freemem(dev); |
1da177e4 LT |
7707 | return; |
7708 | } | |
7709 | ||
7710 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
7711 | dev->reg_state = NETREG_RELEASED; | |
7712 | ||
43cb76d9 GKH |
7713 | /* will free via device release */ |
7714 | put_device(&dev->dev); | |
1da177e4 | 7715 | } |
d1b19dff | 7716 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 7717 | |
f0db275a SH |
7718 | /** |
7719 | * synchronize_net - Synchronize with packet receive processing | |
7720 | * | |
7721 | * Wait for packets currently being received to be done. | |
7722 | * Does not block later packets from starting. | |
7723 | */ | |
4ec93edb | 7724 | void synchronize_net(void) |
1da177e4 LT |
7725 | { |
7726 | might_sleep(); | |
be3fc413 ED |
7727 | if (rtnl_is_locked()) |
7728 | synchronize_rcu_expedited(); | |
7729 | else | |
7730 | synchronize_rcu(); | |
1da177e4 | 7731 | } |
d1b19dff | 7732 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
7733 | |
7734 | /** | |
44a0873d | 7735 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 7736 | * @dev: device |
44a0873d | 7737 | * @head: list |
6ebfbc06 | 7738 | * |
1da177e4 | 7739 | * This function shuts down a device interface and removes it |
d59b54b1 | 7740 | * from the kernel tables. |
44a0873d | 7741 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
7742 | * |
7743 | * Callers must hold the rtnl semaphore. You may want | |
7744 | * unregister_netdev() instead of this. | |
7745 | */ | |
7746 | ||
44a0873d | 7747 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 7748 | { |
a6620712 HX |
7749 | ASSERT_RTNL(); |
7750 | ||
44a0873d | 7751 | if (head) { |
9fdce099 | 7752 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
7753 | } else { |
7754 | rollback_registered(dev); | |
7755 | /* Finish processing unregister after unlock */ | |
7756 | net_set_todo(dev); | |
7757 | } | |
1da177e4 | 7758 | } |
44a0873d | 7759 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 7760 | |
9b5e383c ED |
7761 | /** |
7762 | * unregister_netdevice_many - unregister many devices | |
7763 | * @head: list of devices | |
87757a91 ED |
7764 | * |
7765 | * Note: As most callers use a stack allocated list_head, | |
7766 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
7767 | */ |
7768 | void unregister_netdevice_many(struct list_head *head) | |
7769 | { | |
7770 | struct net_device *dev; | |
7771 | ||
7772 | if (!list_empty(head)) { | |
7773 | rollback_registered_many(head); | |
7774 | list_for_each_entry(dev, head, unreg_list) | |
7775 | net_set_todo(dev); | |
87757a91 | 7776 | list_del(head); |
9b5e383c ED |
7777 | } |
7778 | } | |
63c8099d | 7779 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 7780 | |
1da177e4 LT |
7781 | /** |
7782 | * unregister_netdev - remove device from the kernel | |
7783 | * @dev: device | |
7784 | * | |
7785 | * This function shuts down a device interface and removes it | |
d59b54b1 | 7786 | * from the kernel tables. |
1da177e4 LT |
7787 | * |
7788 | * This is just a wrapper for unregister_netdevice that takes | |
7789 | * the rtnl semaphore. In general you want to use this and not | |
7790 | * unregister_netdevice. | |
7791 | */ | |
7792 | void unregister_netdev(struct net_device *dev) | |
7793 | { | |
7794 | rtnl_lock(); | |
7795 | unregister_netdevice(dev); | |
7796 | rtnl_unlock(); | |
7797 | } | |
1da177e4 LT |
7798 | EXPORT_SYMBOL(unregister_netdev); |
7799 | ||
ce286d32 EB |
7800 | /** |
7801 | * dev_change_net_namespace - move device to different nethost namespace | |
7802 | * @dev: device | |
7803 | * @net: network namespace | |
7804 | * @pat: If not NULL name pattern to try if the current device name | |
7805 | * is already taken in the destination network namespace. | |
7806 | * | |
7807 | * This function shuts down a device interface and moves it | |
7808 | * to a new network namespace. On success 0 is returned, on | |
7809 | * a failure a netagive errno code is returned. | |
7810 | * | |
7811 | * Callers must hold the rtnl semaphore. | |
7812 | */ | |
7813 | ||
7814 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
7815 | { | |
ce286d32 EB |
7816 | int err; |
7817 | ||
7818 | ASSERT_RTNL(); | |
7819 | ||
7820 | /* Don't allow namespace local devices to be moved. */ | |
7821 | err = -EINVAL; | |
7822 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
7823 | goto out; | |
7824 | ||
7825 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
7826 | if (dev->reg_state != NETREG_REGISTERED) |
7827 | goto out; | |
7828 | ||
7829 | /* Get out if there is nothing todo */ | |
7830 | err = 0; | |
878628fb | 7831 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
7832 | goto out; |
7833 | ||
7834 | /* Pick the destination device name, and ensure | |
7835 | * we can use it in the destination network namespace. | |
7836 | */ | |
7837 | err = -EEXIST; | |
d9031024 | 7838 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
7839 | /* We get here if we can't use the current device name */ |
7840 | if (!pat) | |
7841 | goto out; | |
828de4f6 | 7842 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
7843 | goto out; |
7844 | } | |
7845 | ||
7846 | /* | |
7847 | * And now a mini version of register_netdevice unregister_netdevice. | |
7848 | */ | |
7849 | ||
7850 | /* If device is running close it first. */ | |
9b772652 | 7851 | dev_close(dev); |
ce286d32 EB |
7852 | |
7853 | /* And unlink it from device chain */ | |
7854 | err = -ENODEV; | |
7855 | unlist_netdevice(dev); | |
7856 | ||
7857 | synchronize_net(); | |
7858 | ||
7859 | /* Shutdown queueing discipline. */ | |
7860 | dev_shutdown(dev); | |
7861 | ||
7862 | /* Notify protocols, that we are about to destroy | |
7863 | this device. They should clean all the things. | |
3b27e105 DL |
7864 | |
7865 | Note that dev->reg_state stays at NETREG_REGISTERED. | |
7866 | This is wanted because this way 8021q and macvlan know | |
7867 | the device is just moving and can keep their slaves up. | |
ce286d32 EB |
7868 | */ |
7869 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
6549dd43 G |
7870 | rcu_barrier(); |
7871 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 7872 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
7873 | |
7874 | /* | |
7875 | * Flush the unicast and multicast chains | |
7876 | */ | |
a748ee24 | 7877 | dev_uc_flush(dev); |
22bedad3 | 7878 | dev_mc_flush(dev); |
ce286d32 | 7879 | |
4e66ae2e SH |
7880 | /* Send a netdev-removed uevent to the old namespace */ |
7881 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 7882 | netdev_adjacent_del_links(dev); |
4e66ae2e | 7883 | |
ce286d32 | 7884 | /* Actually switch the network namespace */ |
c346dca1 | 7885 | dev_net_set(dev, net); |
ce286d32 | 7886 | |
ce286d32 | 7887 | /* If there is an ifindex conflict assign a new one */ |
7a66bbc9 | 7888 | if (__dev_get_by_index(net, dev->ifindex)) |
ce286d32 | 7889 | dev->ifindex = dev_new_index(net); |
ce286d32 | 7890 | |
4e66ae2e SH |
7891 | /* Send a netdev-add uevent to the new namespace */ |
7892 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 7893 | netdev_adjacent_add_links(dev); |
4e66ae2e | 7894 | |
8b41d188 | 7895 | /* Fixup kobjects */ |
a1b3f594 | 7896 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 7897 | WARN_ON(err); |
ce286d32 EB |
7898 | |
7899 | /* Add the device back in the hashes */ | |
7900 | list_netdevice(dev); | |
7901 | ||
7902 | /* Notify protocols, that a new device appeared. */ | |
7903 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
7904 | ||
d90a909e EB |
7905 | /* |
7906 | * Prevent userspace races by waiting until the network | |
7907 | * device is fully setup before sending notifications. | |
7908 | */ | |
7f294054 | 7909 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 7910 | |
ce286d32 EB |
7911 | synchronize_net(); |
7912 | err = 0; | |
7913 | out: | |
7914 | return err; | |
7915 | } | |
463d0183 | 7916 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 7917 | |
1da177e4 LT |
7918 | static int dev_cpu_callback(struct notifier_block *nfb, |
7919 | unsigned long action, | |
7920 | void *ocpu) | |
7921 | { | |
7922 | struct sk_buff **list_skb; | |
1da177e4 LT |
7923 | struct sk_buff *skb; |
7924 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
7925 | struct softnet_data *sd, *oldsd; | |
7926 | ||
8bb78442 | 7927 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
7928 | return NOTIFY_OK; |
7929 | ||
7930 | local_irq_disable(); | |
7931 | cpu = smp_processor_id(); | |
7932 | sd = &per_cpu(softnet_data, cpu); | |
7933 | oldsd = &per_cpu(softnet_data, oldcpu); | |
7934 | ||
7935 | /* Find end of our completion_queue. */ | |
7936 | list_skb = &sd->completion_queue; | |
7937 | while (*list_skb) | |
7938 | list_skb = &(*list_skb)->next; | |
7939 | /* Append completion queue from offline CPU. */ | |
7940 | *list_skb = oldsd->completion_queue; | |
7941 | oldsd->completion_queue = NULL; | |
7942 | ||
1da177e4 | 7943 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
7944 | if (oldsd->output_queue) { |
7945 | *sd->output_queue_tailp = oldsd->output_queue; | |
7946 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
7947 | oldsd->output_queue = NULL; | |
7948 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
7949 | } | |
ac64da0b ED |
7950 | /* Append NAPI poll list from offline CPU, with one exception : |
7951 | * process_backlog() must be called by cpu owning percpu backlog. | |
7952 | * We properly handle process_queue & input_pkt_queue later. | |
7953 | */ | |
7954 | while (!list_empty(&oldsd->poll_list)) { | |
7955 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
7956 | struct napi_struct, | |
7957 | poll_list); | |
7958 | ||
7959 | list_del_init(&napi->poll_list); | |
7960 | if (napi->poll == process_backlog) | |
7961 | napi->state = 0; | |
7962 | else | |
7963 | ____napi_schedule(sd, napi); | |
264524d5 | 7964 | } |
1da177e4 LT |
7965 | |
7966 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
7967 | local_irq_enable(); | |
7968 | ||
7969 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 7970 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 7971 | netif_rx_ni(skb); |
76cc8b13 | 7972 | input_queue_head_incr(oldsd); |
fec5e652 | 7973 | } |
ac64da0b | 7974 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 7975 | netif_rx_ni(skb); |
76cc8b13 TH |
7976 | input_queue_head_incr(oldsd); |
7977 | } | |
1da177e4 LT |
7978 | |
7979 | return NOTIFY_OK; | |
7980 | } | |
1da177e4 LT |
7981 | |
7982 | ||
7f353bf2 | 7983 | /** |
b63365a2 HX |
7984 | * netdev_increment_features - increment feature set by one |
7985 | * @all: current feature set | |
7986 | * @one: new feature set | |
7987 | * @mask: mask feature set | |
7f353bf2 HX |
7988 | * |
7989 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
7990 | * @one to the master device with current feature set @all. Will not |
7991 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 7992 | */ |
c8f44aff MM |
7993 | netdev_features_t netdev_increment_features(netdev_features_t all, |
7994 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 7995 | { |
c8cd0989 | 7996 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 7997 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 7998 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 7999 | |
a188222b | 8000 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 8001 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 8002 | |
1742f183 | 8003 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
8004 | if (all & NETIF_F_HW_CSUM) |
8005 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
8006 | |
8007 | return all; | |
8008 | } | |
b63365a2 | 8009 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 8010 | |
430f03cd | 8011 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
8012 | { |
8013 | int i; | |
8014 | struct hlist_head *hash; | |
8015 | ||
8016 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
8017 | if (hash != NULL) | |
8018 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
8019 | INIT_HLIST_HEAD(&hash[i]); | |
8020 | ||
8021 | return hash; | |
8022 | } | |
8023 | ||
881d966b | 8024 | /* Initialize per network namespace state */ |
4665079c | 8025 | static int __net_init netdev_init(struct net *net) |
881d966b | 8026 | { |
734b6541 RM |
8027 | if (net != &init_net) |
8028 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 8029 | |
30d97d35 PE |
8030 | net->dev_name_head = netdev_create_hash(); |
8031 | if (net->dev_name_head == NULL) | |
8032 | goto err_name; | |
881d966b | 8033 | |
30d97d35 PE |
8034 | net->dev_index_head = netdev_create_hash(); |
8035 | if (net->dev_index_head == NULL) | |
8036 | goto err_idx; | |
881d966b EB |
8037 | |
8038 | return 0; | |
30d97d35 PE |
8039 | |
8040 | err_idx: | |
8041 | kfree(net->dev_name_head); | |
8042 | err_name: | |
8043 | return -ENOMEM; | |
881d966b EB |
8044 | } |
8045 | ||
f0db275a SH |
8046 | /** |
8047 | * netdev_drivername - network driver for the device | |
8048 | * @dev: network device | |
f0db275a SH |
8049 | * |
8050 | * Determine network driver for device. | |
8051 | */ | |
3019de12 | 8052 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 8053 | { |
cf04a4c7 SH |
8054 | const struct device_driver *driver; |
8055 | const struct device *parent; | |
3019de12 | 8056 | const char *empty = ""; |
6579e57b AV |
8057 | |
8058 | parent = dev->dev.parent; | |
6579e57b | 8059 | if (!parent) |
3019de12 | 8060 | return empty; |
6579e57b AV |
8061 | |
8062 | driver = parent->driver; | |
8063 | if (driver && driver->name) | |
3019de12 DM |
8064 | return driver->name; |
8065 | return empty; | |
6579e57b AV |
8066 | } |
8067 | ||
6ea754eb JP |
8068 | static void __netdev_printk(const char *level, const struct net_device *dev, |
8069 | struct va_format *vaf) | |
256df2f3 | 8070 | { |
b004ff49 | 8071 | if (dev && dev->dev.parent) { |
6ea754eb JP |
8072 | dev_printk_emit(level[1] - '0', |
8073 | dev->dev.parent, | |
8074 | "%s %s %s%s: %pV", | |
8075 | dev_driver_string(dev->dev.parent), | |
8076 | dev_name(dev->dev.parent), | |
8077 | netdev_name(dev), netdev_reg_state(dev), | |
8078 | vaf); | |
b004ff49 | 8079 | } else if (dev) { |
6ea754eb JP |
8080 | printk("%s%s%s: %pV", |
8081 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 8082 | } else { |
6ea754eb | 8083 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 8084 | } |
256df2f3 JP |
8085 | } |
8086 | ||
6ea754eb JP |
8087 | void netdev_printk(const char *level, const struct net_device *dev, |
8088 | const char *format, ...) | |
256df2f3 JP |
8089 | { |
8090 | struct va_format vaf; | |
8091 | va_list args; | |
256df2f3 JP |
8092 | |
8093 | va_start(args, format); | |
8094 | ||
8095 | vaf.fmt = format; | |
8096 | vaf.va = &args; | |
8097 | ||
6ea754eb | 8098 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 8099 | |
256df2f3 | 8100 | va_end(args); |
256df2f3 JP |
8101 | } |
8102 | EXPORT_SYMBOL(netdev_printk); | |
8103 | ||
8104 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 8105 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 8106 | { \ |
256df2f3 JP |
8107 | struct va_format vaf; \ |
8108 | va_list args; \ | |
8109 | \ | |
8110 | va_start(args, fmt); \ | |
8111 | \ | |
8112 | vaf.fmt = fmt; \ | |
8113 | vaf.va = &args; \ | |
8114 | \ | |
6ea754eb | 8115 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 8116 | \ |
256df2f3 | 8117 | va_end(args); \ |
256df2f3 JP |
8118 | } \ |
8119 | EXPORT_SYMBOL(func); | |
8120 | ||
8121 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
8122 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
8123 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
8124 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
8125 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
8126 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
8127 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
8128 | ||
4665079c | 8129 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
8130 | { |
8131 | kfree(net->dev_name_head); | |
8132 | kfree(net->dev_index_head); | |
8133 | } | |
8134 | ||
022cbae6 | 8135 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
8136 | .init = netdev_init, |
8137 | .exit = netdev_exit, | |
8138 | }; | |
8139 | ||
4665079c | 8140 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 8141 | { |
e008b5fc | 8142 | struct net_device *dev, *aux; |
ce286d32 | 8143 | /* |
e008b5fc | 8144 | * Push all migratable network devices back to the |
ce286d32 EB |
8145 | * initial network namespace |
8146 | */ | |
8147 | rtnl_lock(); | |
e008b5fc | 8148 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 8149 | int err; |
aca51397 | 8150 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
8151 | |
8152 | /* Ignore unmoveable devices (i.e. loopback) */ | |
8153 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8154 | continue; | |
8155 | ||
e008b5fc EB |
8156 | /* Leave virtual devices for the generic cleanup */ |
8157 | if (dev->rtnl_link_ops) | |
8158 | continue; | |
d0c082ce | 8159 | |
25985edc | 8160 | /* Push remaining network devices to init_net */ |
aca51397 PE |
8161 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
8162 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 8163 | if (err) { |
7b6cd1ce JP |
8164 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
8165 | __func__, dev->name, err); | |
aca51397 | 8166 | BUG(); |
ce286d32 EB |
8167 | } |
8168 | } | |
8169 | rtnl_unlock(); | |
8170 | } | |
8171 | ||
50624c93 EB |
8172 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
8173 | { | |
8174 | /* Return with the rtnl_lock held when there are no network | |
8175 | * devices unregistering in any network namespace in net_list. | |
8176 | */ | |
8177 | struct net *net; | |
8178 | bool unregistering; | |
ff960a73 | 8179 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 8180 | |
ff960a73 | 8181 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 8182 | for (;;) { |
50624c93 EB |
8183 | unregistering = false; |
8184 | rtnl_lock(); | |
8185 | list_for_each_entry(net, net_list, exit_list) { | |
8186 | if (net->dev_unreg_count > 0) { | |
8187 | unregistering = true; | |
8188 | break; | |
8189 | } | |
8190 | } | |
8191 | if (!unregistering) | |
8192 | break; | |
8193 | __rtnl_unlock(); | |
ff960a73 PZ |
8194 | |
8195 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 8196 | } |
ff960a73 | 8197 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
8198 | } |
8199 | ||
04dc7f6b EB |
8200 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
8201 | { | |
8202 | /* At exit all network devices most be removed from a network | |
b595076a | 8203 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
8204 | * Do this across as many network namespaces as possible to |
8205 | * improve batching efficiency. | |
8206 | */ | |
8207 | struct net_device *dev; | |
8208 | struct net *net; | |
8209 | LIST_HEAD(dev_kill_list); | |
8210 | ||
50624c93 EB |
8211 | /* To prevent network device cleanup code from dereferencing |
8212 | * loopback devices or network devices that have been freed | |
8213 | * wait here for all pending unregistrations to complete, | |
8214 | * before unregistring the loopback device and allowing the | |
8215 | * network namespace be freed. | |
8216 | * | |
8217 | * The netdev todo list containing all network devices | |
8218 | * unregistrations that happen in default_device_exit_batch | |
8219 | * will run in the rtnl_unlock() at the end of | |
8220 | * default_device_exit_batch. | |
8221 | */ | |
8222 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
8223 | list_for_each_entry(net, net_list, exit_list) { |
8224 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 8225 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
8226 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
8227 | else | |
8228 | unregister_netdevice_queue(dev, &dev_kill_list); | |
8229 | } | |
8230 | } | |
8231 | unregister_netdevice_many(&dev_kill_list); | |
8232 | rtnl_unlock(); | |
8233 | } | |
8234 | ||
022cbae6 | 8235 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 8236 | .exit = default_device_exit, |
04dc7f6b | 8237 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
8238 | }; |
8239 | ||
1da177e4 LT |
8240 | /* |
8241 | * Initialize the DEV module. At boot time this walks the device list and | |
8242 | * unhooks any devices that fail to initialise (normally hardware not | |
8243 | * present) and leaves us with a valid list of present and active devices. | |
8244 | * | |
8245 | */ | |
8246 | ||
8247 | /* | |
8248 | * This is called single threaded during boot, so no need | |
8249 | * to take the rtnl semaphore. | |
8250 | */ | |
8251 | static int __init net_dev_init(void) | |
8252 | { | |
8253 | int i, rc = -ENOMEM; | |
8254 | ||
8255 | BUG_ON(!dev_boot_phase); | |
8256 | ||
1da177e4 LT |
8257 | if (dev_proc_init()) |
8258 | goto out; | |
8259 | ||
8b41d188 | 8260 | if (netdev_kobject_init()) |
1da177e4 LT |
8261 | goto out; |
8262 | ||
8263 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 8264 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
8265 | INIT_LIST_HEAD(&ptype_base[i]); |
8266 | ||
62532da9 VY |
8267 | INIT_LIST_HEAD(&offload_base); |
8268 | ||
881d966b EB |
8269 | if (register_pernet_subsys(&netdev_net_ops)) |
8270 | goto out; | |
1da177e4 LT |
8271 | |
8272 | /* | |
8273 | * Initialise the packet receive queues. | |
8274 | */ | |
8275 | ||
6f912042 | 8276 | for_each_possible_cpu(i) { |
e36fa2f7 | 8277 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 8278 | |
e36fa2f7 | 8279 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 8280 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 8281 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 8282 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 8283 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
8284 | sd->csd.func = rps_trigger_softirq; |
8285 | sd->csd.info = sd; | |
e36fa2f7 | 8286 | sd->cpu = i; |
1e94d72f | 8287 | #endif |
0a9627f2 | 8288 | |
e36fa2f7 ED |
8289 | sd->backlog.poll = process_backlog; |
8290 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
8291 | } |
8292 | ||
1da177e4 LT |
8293 | dev_boot_phase = 0; |
8294 | ||
505d4f73 EB |
8295 | /* The loopback device is special if any other network devices |
8296 | * is present in a network namespace the loopback device must | |
8297 | * be present. Since we now dynamically allocate and free the | |
8298 | * loopback device ensure this invariant is maintained by | |
8299 | * keeping the loopback device as the first device on the | |
8300 | * list of network devices. Ensuring the loopback devices | |
8301 | * is the first device that appears and the last network device | |
8302 | * that disappears. | |
8303 | */ | |
8304 | if (register_pernet_device(&loopback_net_ops)) | |
8305 | goto out; | |
8306 | ||
8307 | if (register_pernet_device(&default_device_ops)) | |
8308 | goto out; | |
8309 | ||
962cf36c CM |
8310 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
8311 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
8312 | |
8313 | hotcpu_notifier(dev_cpu_callback, 0); | |
f38a9eb1 | 8314 | dst_subsys_init(); |
1da177e4 LT |
8315 | rc = 0; |
8316 | out: | |
8317 | return rc; | |
8318 | } | |
8319 | ||
8320 | subsys_initcall(net_dev_init); |