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> | |
76 | #include <asm/system.h> | |
77 | #include <linux/bitops.h> | |
4fc268d2 | 78 | #include <linux/capability.h> |
1da177e4 LT |
79 | #include <linux/cpu.h> |
80 | #include <linux/types.h> | |
81 | #include <linux/kernel.h> | |
08e9897d | 82 | #include <linux/hash.h> |
5a0e3ad6 | 83 | #include <linux/slab.h> |
1da177e4 | 84 | #include <linux/sched.h> |
4a3e2f71 | 85 | #include <linux/mutex.h> |
1da177e4 LT |
86 | #include <linux/string.h> |
87 | #include <linux/mm.h> | |
88 | #include <linux/socket.h> | |
89 | #include <linux/sockios.h> | |
90 | #include <linux/errno.h> | |
91 | #include <linux/interrupt.h> | |
92 | #include <linux/if_ether.h> | |
93 | #include <linux/netdevice.h> | |
94 | #include <linux/etherdevice.h> | |
0187bdfb | 95 | #include <linux/ethtool.h> |
1da177e4 LT |
96 | #include <linux/notifier.h> |
97 | #include <linux/skbuff.h> | |
457c4cbc | 98 | #include <net/net_namespace.h> |
1da177e4 LT |
99 | #include <net/sock.h> |
100 | #include <linux/rtnetlink.h> | |
101 | #include <linux/proc_fs.h> | |
102 | #include <linux/seq_file.h> | |
103 | #include <linux/stat.h> | |
104 | #include <linux/if_bridge.h> | |
b863ceb7 | 105 | #include <linux/if_macvlan.h> |
1da177e4 LT |
106 | #include <net/dst.h> |
107 | #include <net/pkt_sched.h> | |
108 | #include <net/checksum.h> | |
44540960 | 109 | #include <net/xfrm.h> |
1da177e4 LT |
110 | #include <linux/highmem.h> |
111 | #include <linux/init.h> | |
112 | #include <linux/kmod.h> | |
113 | #include <linux/module.h> | |
1da177e4 LT |
114 | #include <linux/netpoll.h> |
115 | #include <linux/rcupdate.h> | |
116 | #include <linux/delay.h> | |
295f4a1f | 117 | #include <net/wext.h> |
1da177e4 | 118 | #include <net/iw_handler.h> |
1da177e4 | 119 | #include <asm/current.h> |
5bdb9886 | 120 | #include <linux/audit.h> |
db217334 | 121 | #include <linux/dmaengine.h> |
f6a78bfc | 122 | #include <linux/err.h> |
c7fa9d18 | 123 | #include <linux/ctype.h> |
723e98b7 | 124 | #include <linux/if_arp.h> |
6de329e2 | 125 | #include <linux/if_vlan.h> |
8f0f2223 | 126 | #include <linux/ip.h> |
ad55dcaf | 127 | #include <net/ip.h> |
8f0f2223 DM |
128 | #include <linux/ipv6.h> |
129 | #include <linux/in.h> | |
b6b2fed1 DM |
130 | #include <linux/jhash.h> |
131 | #include <linux/random.h> | |
9cbc1cb8 | 132 | #include <trace/events/napi.h> |
5acbbd42 | 133 | #include <linux/pci.h> |
1da177e4 | 134 | |
342709ef PE |
135 | #include "net-sysfs.h" |
136 | ||
d565b0a1 HX |
137 | /* Instead of increasing this, you should create a hash table. */ |
138 | #define MAX_GRO_SKBS 8 | |
139 | ||
5d38a079 HX |
140 | /* This should be increased if a protocol with a bigger head is added. */ |
141 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
142 | ||
1da177e4 LT |
143 | /* |
144 | * The list of packet types we will receive (as opposed to discard) | |
145 | * and the routines to invoke. | |
146 | * | |
147 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
148 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
149 | * | |
150 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
151 | * sure which should go first, but I bet it won't make much | |
152 | * difference if we are running VLANs. The good news is that | |
153 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 154 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
155 | * --BLG |
156 | * | |
157 | * 0800 IP | |
158 | * 8100 802.1Q VLAN | |
159 | * 0001 802.3 | |
160 | * 0002 AX.25 | |
161 | * 0004 802.2 | |
162 | * 8035 RARP | |
163 | * 0005 SNAP | |
164 | * 0805 X.25 | |
165 | * 0806 ARP | |
166 | * 8137 IPX | |
167 | * 0009 Localtalk | |
168 | * 86DD IPv6 | |
169 | */ | |
170 | ||
82d8a867 PE |
171 | #define PTYPE_HASH_SIZE (16) |
172 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
173 | ||
1da177e4 | 174 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 175 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 176 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 177 | |
1da177e4 | 178 | /* |
7562f876 | 179 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
180 | * semaphore. |
181 | * | |
c6d14c84 | 182 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
183 | * |
184 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 185 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
186 | * actual updates. This allows pure readers to access the list even |
187 | * while a writer is preparing to update it. | |
188 | * | |
189 | * To put it another way, dev_base_lock is held for writing only to | |
190 | * protect against pure readers; the rtnl semaphore provides the | |
191 | * protection against other writers. | |
192 | * | |
193 | * See, for example usages, register_netdevice() and | |
194 | * unregister_netdevice(), which must be called with the rtnl | |
195 | * semaphore held. | |
196 | */ | |
1da177e4 | 197 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
198 | EXPORT_SYMBOL(dev_base_lock); |
199 | ||
881d966b | 200 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 LT |
201 | { |
202 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
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 EB |
225 | /* Device list insertion */ |
226 | static int list_netdevice(struct net_device *dev) | |
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 EB |
237 | write_unlock_bh(&dev_base_lock); |
238 | return 0; | |
239 | } | |
240 | ||
fb699dfd ED |
241 | /* Device list removal |
242 | * caller must respect a RCU grace period before freeing/reusing dev | |
243 | */ | |
ce286d32 EB |
244 | static void unlist_netdevice(struct net_device *dev) |
245 | { | |
246 | ASSERT_RTNL(); | |
247 | ||
248 | /* Unlink dev from the device chain */ | |
249 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 250 | list_del_rcu(&dev->dev_list); |
72c9528b | 251 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 252 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 EB |
253 | write_unlock_bh(&dev_base_lock); |
254 | } | |
255 | ||
1da177e4 LT |
256 | /* |
257 | * Our notifier list | |
258 | */ | |
259 | ||
f07d5b94 | 260 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
261 | |
262 | /* | |
263 | * Device drivers call our routines to queue packets here. We empty the | |
264 | * queue in the local softnet handler. | |
265 | */ | |
bea3348e | 266 | |
9958da05 | 267 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 268 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 269 | |
cf508b12 | 270 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 271 | /* |
c773e847 | 272 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
273 | * according to dev->type |
274 | */ | |
275 | static const unsigned short netdev_lock_type[] = | |
276 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
277 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
278 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
279 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
280 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
281 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
282 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
283 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
284 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
285 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
286 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
287 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
288 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
2d91d78b | 289 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, |
929122cd | 290 | ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154, |
fcb94e42 | 291 | ARPHRD_VOID, ARPHRD_NONE}; |
723e98b7 | 292 | |
36cbd3dc | 293 | static const char *const netdev_lock_name[] = |
723e98b7 JP |
294 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", |
295 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
296 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
297 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
298 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
299 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
300 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
301 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
302 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
303 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
304 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
305 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
306 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
2d91d78b | 307 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", |
929122cd | 308 | "_xmit_PHONET_PIPE", "_xmit_IEEE802154", |
fcb94e42 | 309 | "_xmit_VOID", "_xmit_NONE"}; |
723e98b7 JP |
310 | |
311 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 312 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
313 | |
314 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
315 | { | |
316 | int i; | |
317 | ||
318 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
319 | if (netdev_lock_type[i] == dev_type) | |
320 | return i; | |
321 | /* the last key is used by default */ | |
322 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
323 | } | |
324 | ||
cf508b12 DM |
325 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
326 | unsigned short dev_type) | |
723e98b7 JP |
327 | { |
328 | int i; | |
329 | ||
330 | i = netdev_lock_pos(dev_type); | |
331 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
332 | netdev_lock_name[i]); | |
333 | } | |
cf508b12 DM |
334 | |
335 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
336 | { | |
337 | int i; | |
338 | ||
339 | i = netdev_lock_pos(dev->type); | |
340 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
341 | &netdev_addr_lock_key[i], | |
342 | netdev_lock_name[i]); | |
343 | } | |
723e98b7 | 344 | #else |
cf508b12 DM |
345 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
346 | unsigned short dev_type) | |
347 | { | |
348 | } | |
349 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
350 | { |
351 | } | |
352 | #endif | |
1da177e4 LT |
353 | |
354 | /******************************************************************************* | |
355 | ||
356 | Protocol management and registration routines | |
357 | ||
358 | *******************************************************************************/ | |
359 | ||
1da177e4 LT |
360 | /* |
361 | * Add a protocol ID to the list. Now that the input handler is | |
362 | * smarter we can dispense with all the messy stuff that used to be | |
363 | * here. | |
364 | * | |
365 | * BEWARE!!! Protocol handlers, mangling input packets, | |
366 | * MUST BE last in hash buckets and checking protocol handlers | |
367 | * MUST start from promiscuous ptype_all chain in net_bh. | |
368 | * It is true now, do not change it. | |
369 | * Explanation follows: if protocol handler, mangling packet, will | |
370 | * be the first on list, it is not able to sense, that packet | |
371 | * is cloned and should be copied-on-write, so that it will | |
372 | * change it and subsequent readers will get broken packet. | |
373 | * --ANK (980803) | |
374 | */ | |
375 | ||
376 | /** | |
377 | * dev_add_pack - add packet handler | |
378 | * @pt: packet type declaration | |
379 | * | |
380 | * Add a protocol handler to the networking stack. The passed &packet_type | |
381 | * is linked into kernel lists and may not be freed until it has been | |
382 | * removed from the kernel lists. | |
383 | * | |
4ec93edb | 384 | * This call does not sleep therefore it can not |
1da177e4 LT |
385 | * guarantee all CPU's that are in middle of receiving packets |
386 | * will see the new packet type (until the next received packet). | |
387 | */ | |
388 | ||
389 | void dev_add_pack(struct packet_type *pt) | |
390 | { | |
391 | int hash; | |
392 | ||
393 | spin_lock_bh(&ptype_lock); | |
9be9a6b9 | 394 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 395 | list_add_rcu(&pt->list, &ptype_all); |
9be9a6b9 | 396 | else { |
82d8a867 | 397 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
1da177e4 LT |
398 | list_add_rcu(&pt->list, &ptype_base[hash]); |
399 | } | |
400 | spin_unlock_bh(&ptype_lock); | |
401 | } | |
d1b19dff | 402 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 403 | |
1da177e4 LT |
404 | /** |
405 | * __dev_remove_pack - remove packet handler | |
406 | * @pt: packet type declaration | |
407 | * | |
408 | * Remove a protocol handler that was previously added to the kernel | |
409 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
410 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 411 | * returns. |
1da177e4 LT |
412 | * |
413 | * The packet type might still be in use by receivers | |
414 | * and must not be freed until after all the CPU's have gone | |
415 | * through a quiescent state. | |
416 | */ | |
417 | void __dev_remove_pack(struct packet_type *pt) | |
418 | { | |
419 | struct list_head *head; | |
420 | struct packet_type *pt1; | |
421 | ||
422 | spin_lock_bh(&ptype_lock); | |
423 | ||
9be9a6b9 | 424 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 425 | head = &ptype_all; |
9be9a6b9 | 426 | else |
82d8a867 | 427 | head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; |
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 | ||
436 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
437 | out: | |
438 | spin_unlock_bh(&ptype_lock); | |
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 LT |
461 | |
462 | /****************************************************************************** | |
463 | ||
464 | Device Boot-time Settings Routines | |
465 | ||
466 | *******************************************************************************/ | |
467 | ||
468 | /* Boot time configuration table */ | |
469 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
470 | ||
471 | /** | |
472 | * netdev_boot_setup_add - add new setup entry | |
473 | * @name: name of the device | |
474 | * @map: configured settings for the device | |
475 | * | |
476 | * Adds new setup entry to the dev_boot_setup list. The function | |
477 | * returns 0 on error and 1 on success. This is a generic routine to | |
478 | * all netdevices. | |
479 | */ | |
480 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
481 | { | |
482 | struct netdev_boot_setup *s; | |
483 | int i; | |
484 | ||
485 | s = dev_boot_setup; | |
486 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
487 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
488 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 489 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
490 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
491 | break; | |
492 | } | |
493 | } | |
494 | ||
495 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
496 | } | |
497 | ||
498 | /** | |
499 | * netdev_boot_setup_check - check boot time settings | |
500 | * @dev: the netdevice | |
501 | * | |
502 | * Check boot time settings for the device. | |
503 | * The found settings are set for the device to be used | |
504 | * later in the device probing. | |
505 | * Returns 0 if no settings found, 1 if they are. | |
506 | */ | |
507 | int netdev_boot_setup_check(struct net_device *dev) | |
508 | { | |
509 | struct netdev_boot_setup *s = dev_boot_setup; | |
510 | int i; | |
511 | ||
512 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
513 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 514 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
515 | dev->irq = s[i].map.irq; |
516 | dev->base_addr = s[i].map.base_addr; | |
517 | dev->mem_start = s[i].map.mem_start; | |
518 | dev->mem_end = s[i].map.mem_end; | |
519 | return 1; | |
520 | } | |
521 | } | |
522 | return 0; | |
523 | } | |
d1b19dff | 524 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
525 | |
526 | ||
527 | /** | |
528 | * netdev_boot_base - get address from boot time settings | |
529 | * @prefix: prefix for network device | |
530 | * @unit: id for network device | |
531 | * | |
532 | * Check boot time settings for the base address of device. | |
533 | * The found settings are set for the device to be used | |
534 | * later in the device probing. | |
535 | * Returns 0 if no settings found. | |
536 | */ | |
537 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
538 | { | |
539 | const struct netdev_boot_setup *s = dev_boot_setup; | |
540 | char name[IFNAMSIZ]; | |
541 | int i; | |
542 | ||
543 | sprintf(name, "%s%d", prefix, unit); | |
544 | ||
545 | /* | |
546 | * If device already registered then return base of 1 | |
547 | * to indicate not to probe for this interface | |
548 | */ | |
881d966b | 549 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
550 | return 1; |
551 | ||
552 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
553 | if (!strcmp(name, s[i].name)) | |
554 | return s[i].map.base_addr; | |
555 | return 0; | |
556 | } | |
557 | ||
558 | /* | |
559 | * Saves at boot time configured settings for any netdevice. | |
560 | */ | |
561 | int __init netdev_boot_setup(char *str) | |
562 | { | |
563 | int ints[5]; | |
564 | struct ifmap map; | |
565 | ||
566 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
567 | if (!str || !*str) | |
568 | return 0; | |
569 | ||
570 | /* Save settings */ | |
571 | memset(&map, 0, sizeof(map)); | |
572 | if (ints[0] > 0) | |
573 | map.irq = ints[1]; | |
574 | if (ints[0] > 1) | |
575 | map.base_addr = ints[2]; | |
576 | if (ints[0] > 2) | |
577 | map.mem_start = ints[3]; | |
578 | if (ints[0] > 3) | |
579 | map.mem_end = ints[4]; | |
580 | ||
581 | /* Add new entry to the list */ | |
582 | return netdev_boot_setup_add(str, &map); | |
583 | } | |
584 | ||
585 | __setup("netdev=", netdev_boot_setup); | |
586 | ||
587 | /******************************************************************************* | |
588 | ||
589 | Device Interface Subroutines | |
590 | ||
591 | *******************************************************************************/ | |
592 | ||
593 | /** | |
594 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 595 | * @net: the applicable net namespace |
1da177e4 LT |
596 | * @name: name to find |
597 | * | |
598 | * Find an interface by name. Must be called under RTNL semaphore | |
599 | * or @dev_base_lock. If the name is found a pointer to the device | |
600 | * is returned. If the name is not found then %NULL is returned. The | |
601 | * reference counters are not incremented so the caller must be | |
602 | * careful with locks. | |
603 | */ | |
604 | ||
881d966b | 605 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
606 | { |
607 | struct hlist_node *p; | |
0bd8d536 ED |
608 | struct net_device *dev; |
609 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 610 | |
0bd8d536 | 611 | hlist_for_each_entry(dev, p, head, name_hlist) |
1da177e4 LT |
612 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
613 | return dev; | |
0bd8d536 | 614 | |
1da177e4 LT |
615 | return NULL; |
616 | } | |
d1b19dff | 617 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 618 | |
72c9528b ED |
619 | /** |
620 | * dev_get_by_name_rcu - find a device by its name | |
621 | * @net: the applicable net namespace | |
622 | * @name: name to find | |
623 | * | |
624 | * Find an interface by name. | |
625 | * If the name is found a pointer to the device is returned. | |
626 | * If the name is not found then %NULL is returned. | |
627 | * The reference counters are not incremented so the caller must be | |
628 | * careful with locks. The caller must hold RCU lock. | |
629 | */ | |
630 | ||
631 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
632 | { | |
633 | struct hlist_node *p; | |
634 | struct net_device *dev; | |
635 | struct hlist_head *head = dev_name_hash(net, name); | |
636 | ||
637 | hlist_for_each_entry_rcu(dev, p, head, name_hlist) | |
638 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
639 | return dev; | |
640 | ||
641 | return NULL; | |
642 | } | |
643 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
644 | ||
1da177e4 LT |
645 | /** |
646 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 647 | * @net: the applicable net namespace |
1da177e4 LT |
648 | * @name: name to find |
649 | * | |
650 | * Find an interface by name. This can be called from any | |
651 | * context and does its own locking. The returned handle has | |
652 | * the usage count incremented and the caller must use dev_put() to | |
653 | * release it when it is no longer needed. %NULL is returned if no | |
654 | * matching device is found. | |
655 | */ | |
656 | ||
881d966b | 657 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
658 | { |
659 | struct net_device *dev; | |
660 | ||
72c9528b ED |
661 | rcu_read_lock(); |
662 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
663 | if (dev) |
664 | dev_hold(dev); | |
72c9528b | 665 | rcu_read_unlock(); |
1da177e4 LT |
666 | return dev; |
667 | } | |
d1b19dff | 668 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
669 | |
670 | /** | |
671 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 672 | * @net: the applicable net namespace |
1da177e4 LT |
673 | * @ifindex: index of device |
674 | * | |
675 | * Search for an interface by index. Returns %NULL if the device | |
676 | * is not found or a pointer to the device. The device has not | |
677 | * had its reference counter increased so the caller must be careful | |
678 | * about locking. The caller must hold either the RTNL semaphore | |
679 | * or @dev_base_lock. | |
680 | */ | |
681 | ||
881d966b | 682 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
683 | { |
684 | struct hlist_node *p; | |
0bd8d536 ED |
685 | struct net_device *dev; |
686 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 687 | |
0bd8d536 | 688 | hlist_for_each_entry(dev, p, head, index_hlist) |
1da177e4 LT |
689 | if (dev->ifindex == ifindex) |
690 | return dev; | |
0bd8d536 | 691 | |
1da177e4 LT |
692 | return NULL; |
693 | } | |
d1b19dff | 694 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 695 | |
fb699dfd ED |
696 | /** |
697 | * dev_get_by_index_rcu - find a device by its ifindex | |
698 | * @net: the applicable net namespace | |
699 | * @ifindex: index of device | |
700 | * | |
701 | * Search for an interface by index. Returns %NULL if the device | |
702 | * is not found or a pointer to the device. The device has not | |
703 | * had its reference counter increased so the caller must be careful | |
704 | * about locking. The caller must hold RCU lock. | |
705 | */ | |
706 | ||
707 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
708 | { | |
709 | struct hlist_node *p; | |
710 | struct net_device *dev; | |
711 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
712 | ||
713 | hlist_for_each_entry_rcu(dev, p, head, index_hlist) | |
714 | if (dev->ifindex == ifindex) | |
715 | return dev; | |
716 | ||
717 | return NULL; | |
718 | } | |
719 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
720 | ||
1da177e4 LT |
721 | |
722 | /** | |
723 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 724 | * @net: the applicable net namespace |
1da177e4 LT |
725 | * @ifindex: index of device |
726 | * | |
727 | * Search for an interface by index. Returns NULL if the device | |
728 | * is not found or a pointer to the device. The device returned has | |
729 | * had a reference added and the pointer is safe until the user calls | |
730 | * dev_put to indicate they have finished with it. | |
731 | */ | |
732 | ||
881d966b | 733 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
734 | { |
735 | struct net_device *dev; | |
736 | ||
fb699dfd ED |
737 | rcu_read_lock(); |
738 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
739 | if (dev) |
740 | dev_hold(dev); | |
fb699dfd | 741 | rcu_read_unlock(); |
1da177e4 LT |
742 | return dev; |
743 | } | |
d1b19dff | 744 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 LT |
745 | |
746 | /** | |
747 | * dev_getbyhwaddr - find a device by its hardware address | |
c4ea43c5 | 748 | * @net: the applicable net namespace |
1da177e4 LT |
749 | * @type: media type of device |
750 | * @ha: hardware address | |
751 | * | |
752 | * Search for an interface by MAC address. Returns NULL if the device | |
753 | * is not found or a pointer to the device. The caller must hold the | |
754 | * rtnl semaphore. The returned device has not had its ref count increased | |
755 | * and the caller must therefore be careful about locking | |
756 | * | |
757 | * BUGS: | |
758 | * If the API was consistent this would be __dev_get_by_hwaddr | |
759 | */ | |
760 | ||
881d966b | 761 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha) |
1da177e4 LT |
762 | { |
763 | struct net_device *dev; | |
764 | ||
765 | ASSERT_RTNL(); | |
766 | ||
81103a52 | 767 | for_each_netdev(net, dev) |
1da177e4 LT |
768 | if (dev->type == type && |
769 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
770 | return dev; |
771 | ||
772 | return NULL; | |
1da177e4 | 773 | } |
cf309e3f JF |
774 | EXPORT_SYMBOL(dev_getbyhwaddr); |
775 | ||
881d966b | 776 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
777 | { |
778 | struct net_device *dev; | |
779 | ||
4e9cac2b | 780 | ASSERT_RTNL(); |
881d966b | 781 | for_each_netdev(net, dev) |
4e9cac2b | 782 | if (dev->type == type) |
7562f876 PE |
783 | return dev; |
784 | ||
785 | return NULL; | |
4e9cac2b | 786 | } |
4e9cac2b PM |
787 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
788 | ||
881d966b | 789 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 790 | { |
99fe3c39 | 791 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 792 | |
99fe3c39 ED |
793 | rcu_read_lock(); |
794 | for_each_netdev_rcu(net, dev) | |
795 | if (dev->type == type) { | |
796 | dev_hold(dev); | |
797 | ret = dev; | |
798 | break; | |
799 | } | |
800 | rcu_read_unlock(); | |
801 | return ret; | |
1da177e4 | 802 | } |
1da177e4 LT |
803 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
804 | ||
805 | /** | |
bb69ae04 | 806 | * dev_get_by_flags_rcu - find any device with given flags |
c4ea43c5 | 807 | * @net: the applicable net namespace |
1da177e4 LT |
808 | * @if_flags: IFF_* values |
809 | * @mask: bitmask of bits in if_flags to check | |
810 | * | |
811 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 ED |
812 | * is not found or a pointer to the device. Must be called inside |
813 | * rcu_read_lock(), and result refcount is unchanged. | |
1da177e4 LT |
814 | */ |
815 | ||
bb69ae04 | 816 | struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags, |
d1b19dff | 817 | unsigned short mask) |
1da177e4 | 818 | { |
7562f876 | 819 | struct net_device *dev, *ret; |
1da177e4 | 820 | |
7562f876 | 821 | ret = NULL; |
c6d14c84 | 822 | for_each_netdev_rcu(net, dev) { |
1da177e4 | 823 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 824 | ret = dev; |
1da177e4 LT |
825 | break; |
826 | } | |
827 | } | |
7562f876 | 828 | return ret; |
1da177e4 | 829 | } |
bb69ae04 | 830 | EXPORT_SYMBOL(dev_get_by_flags_rcu); |
1da177e4 LT |
831 | |
832 | /** | |
833 | * dev_valid_name - check if name is okay for network device | |
834 | * @name: name string | |
835 | * | |
836 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
837 | * to allow sysfs to work. We also disallow any kind of |
838 | * whitespace. | |
1da177e4 | 839 | */ |
c2373ee9 | 840 | int dev_valid_name(const char *name) |
1da177e4 | 841 | { |
c7fa9d18 DM |
842 | if (*name == '\0') |
843 | return 0; | |
b6fe17d6 SH |
844 | if (strlen(name) >= IFNAMSIZ) |
845 | return 0; | |
c7fa9d18 DM |
846 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
847 | return 0; | |
848 | ||
849 | while (*name) { | |
850 | if (*name == '/' || isspace(*name)) | |
851 | return 0; | |
852 | name++; | |
853 | } | |
854 | return 1; | |
1da177e4 | 855 | } |
d1b19dff | 856 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
857 | |
858 | /** | |
b267b179 EB |
859 | * __dev_alloc_name - allocate a name for a device |
860 | * @net: network namespace to allocate the device name in | |
1da177e4 | 861 | * @name: name format string |
b267b179 | 862 | * @buf: scratch buffer and result name string |
1da177e4 LT |
863 | * |
864 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
865 | * id. It scans list of devices to build up a free map, then chooses |
866 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
867 | * while allocating the name and adding the device in order to avoid | |
868 | * duplicates. | |
869 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
870 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
871 | */ |
872 | ||
b267b179 | 873 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
874 | { |
875 | int i = 0; | |
1da177e4 LT |
876 | const char *p; |
877 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 878 | unsigned long *inuse; |
1da177e4 LT |
879 | struct net_device *d; |
880 | ||
881 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
882 | if (p) { | |
883 | /* | |
884 | * Verify the string as this thing may have come from | |
885 | * the user. There must be either one "%d" and no other "%" | |
886 | * characters. | |
887 | */ | |
888 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
889 | return -EINVAL; | |
890 | ||
891 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 892 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
893 | if (!inuse) |
894 | return -ENOMEM; | |
895 | ||
881d966b | 896 | for_each_netdev(net, d) { |
1da177e4 LT |
897 | if (!sscanf(d->name, name, &i)) |
898 | continue; | |
899 | if (i < 0 || i >= max_netdevices) | |
900 | continue; | |
901 | ||
902 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 903 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
904 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
905 | set_bit(i, inuse); | |
906 | } | |
907 | ||
908 | i = find_first_zero_bit(inuse, max_netdevices); | |
909 | free_page((unsigned long) inuse); | |
910 | } | |
911 | ||
d9031024 OP |
912 | if (buf != name) |
913 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 914 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 915 | return i; |
1da177e4 LT |
916 | |
917 | /* It is possible to run out of possible slots | |
918 | * when the name is long and there isn't enough space left | |
919 | * for the digits, or if all bits are used. | |
920 | */ | |
921 | return -ENFILE; | |
922 | } | |
923 | ||
b267b179 EB |
924 | /** |
925 | * dev_alloc_name - allocate a name for a device | |
926 | * @dev: device | |
927 | * @name: name format string | |
928 | * | |
929 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
930 | * id. It scans list of devices to build up a free map, then chooses | |
931 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
932 | * while allocating the name and adding the device in order to avoid | |
933 | * duplicates. | |
934 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
935 | * Returns the number of the unit assigned or a negative errno code. | |
936 | */ | |
937 | ||
938 | int dev_alloc_name(struct net_device *dev, const char *name) | |
939 | { | |
940 | char buf[IFNAMSIZ]; | |
941 | struct net *net; | |
942 | int ret; | |
943 | ||
c346dca1 YH |
944 | BUG_ON(!dev_net(dev)); |
945 | net = dev_net(dev); | |
b267b179 EB |
946 | ret = __dev_alloc_name(net, name, buf); |
947 | if (ret >= 0) | |
948 | strlcpy(dev->name, buf, IFNAMSIZ); | |
949 | return ret; | |
950 | } | |
d1b19dff | 951 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 952 | |
8ce6cebc | 953 | static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt) |
d9031024 | 954 | { |
8ce6cebc DL |
955 | struct net *net; |
956 | ||
957 | BUG_ON(!dev_net(dev)); | |
958 | net = dev_net(dev); | |
959 | ||
d9031024 OP |
960 | if (!dev_valid_name(name)) |
961 | return -EINVAL; | |
962 | ||
963 | if (fmt && strchr(name, '%')) | |
8ce6cebc | 964 | return dev_alloc_name(dev, name); |
d9031024 OP |
965 | else if (__dev_get_by_name(net, name)) |
966 | return -EEXIST; | |
8ce6cebc DL |
967 | else if (dev->name != name) |
968 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
969 | |
970 | return 0; | |
971 | } | |
1da177e4 LT |
972 | |
973 | /** | |
974 | * dev_change_name - change name of a device | |
975 | * @dev: device | |
976 | * @newname: name (or format string) must be at least IFNAMSIZ | |
977 | * | |
978 | * Change name of a device, can pass format strings "eth%d". | |
979 | * for wildcarding. | |
980 | */ | |
cf04a4c7 | 981 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 982 | { |
fcc5a03a | 983 | char oldname[IFNAMSIZ]; |
1da177e4 | 984 | int err = 0; |
fcc5a03a | 985 | int ret; |
881d966b | 986 | struct net *net; |
1da177e4 LT |
987 | |
988 | ASSERT_RTNL(); | |
c346dca1 | 989 | BUG_ON(!dev_net(dev)); |
1da177e4 | 990 | |
c346dca1 | 991 | net = dev_net(dev); |
1da177e4 LT |
992 | if (dev->flags & IFF_UP) |
993 | return -EBUSY; | |
994 | ||
c8d90dca SH |
995 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
996 | return 0; | |
997 | ||
fcc5a03a HX |
998 | memcpy(oldname, dev->name, IFNAMSIZ); |
999 | ||
8ce6cebc | 1000 | err = dev_get_valid_name(dev, newname, 1); |
d9031024 OP |
1001 | if (err < 0) |
1002 | return err; | |
1da177e4 | 1003 | |
fcc5a03a | 1004 | rollback: |
a1b3f594 EB |
1005 | ret = device_rename(&dev->dev, dev->name); |
1006 | if (ret) { | |
1007 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1008 | return ret; | |
dcc99773 | 1009 | } |
7f988eab HX |
1010 | |
1011 | write_lock_bh(&dev_base_lock); | |
92749821 | 1012 | hlist_del(&dev->name_hlist); |
72c9528b ED |
1013 | write_unlock_bh(&dev_base_lock); |
1014 | ||
1015 | synchronize_rcu(); | |
1016 | ||
1017 | write_lock_bh(&dev_base_lock); | |
1018 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1019 | write_unlock_bh(&dev_base_lock); |
1020 | ||
056925ab | 1021 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1022 | ret = notifier_to_errno(ret); |
1023 | ||
1024 | if (ret) { | |
91e9c07b ED |
1025 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1026 | if (err >= 0) { | |
fcc5a03a HX |
1027 | err = ret; |
1028 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1029 | goto rollback; | |
91e9c07b ED |
1030 | } else { |
1031 | printk(KERN_ERR | |
1032 | "%s: name change rollback failed: %d.\n", | |
1033 | dev->name, ret); | |
fcc5a03a HX |
1034 | } |
1035 | } | |
1da177e4 LT |
1036 | |
1037 | return err; | |
1038 | } | |
1039 | ||
0b815a1a SH |
1040 | /** |
1041 | * dev_set_alias - change ifalias of a device | |
1042 | * @dev: device | |
1043 | * @alias: name up to IFALIASZ | |
f0db275a | 1044 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1045 | * |
1046 | * Set ifalias for a device, | |
1047 | */ | |
1048 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1049 | { | |
1050 | ASSERT_RTNL(); | |
1051 | ||
1052 | if (len >= IFALIASZ) | |
1053 | return -EINVAL; | |
1054 | ||
96ca4a2c OH |
1055 | if (!len) { |
1056 | if (dev->ifalias) { | |
1057 | kfree(dev->ifalias); | |
1058 | dev->ifalias = NULL; | |
1059 | } | |
1060 | return 0; | |
1061 | } | |
1062 | ||
d1b19dff | 1063 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
0b815a1a SH |
1064 | if (!dev->ifalias) |
1065 | return -ENOMEM; | |
1066 | ||
1067 | strlcpy(dev->ifalias, alias, len+1); | |
1068 | return len; | |
1069 | } | |
1070 | ||
1071 | ||
d8a33ac4 | 1072 | /** |
3041a069 | 1073 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1074 | * @dev: device to cause notification |
1075 | * | |
1076 | * Called to indicate a device has changed features. | |
1077 | */ | |
1078 | void netdev_features_change(struct net_device *dev) | |
1079 | { | |
056925ab | 1080 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1081 | } |
1082 | EXPORT_SYMBOL(netdev_features_change); | |
1083 | ||
1da177e4 LT |
1084 | /** |
1085 | * netdev_state_change - device changes state | |
1086 | * @dev: device to cause notification | |
1087 | * | |
1088 | * Called to indicate a device has changed state. This function calls | |
1089 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1090 | * to the routing socket. | |
1091 | */ | |
1092 | void netdev_state_change(struct net_device *dev) | |
1093 | { | |
1094 | if (dev->flags & IFF_UP) { | |
056925ab | 1095 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1096 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1097 | } | |
1098 | } | |
d1b19dff | 1099 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1100 | |
3ca5b404 | 1101 | int netdev_bonding_change(struct net_device *dev, unsigned long event) |
c1da4ac7 | 1102 | { |
3ca5b404 | 1103 | return call_netdevice_notifiers(event, dev); |
c1da4ac7 OG |
1104 | } |
1105 | EXPORT_SYMBOL(netdev_bonding_change); | |
1106 | ||
1da177e4 LT |
1107 | /** |
1108 | * dev_load - load a network module | |
c4ea43c5 | 1109 | * @net: the applicable net namespace |
1da177e4 LT |
1110 | * @name: name of interface |
1111 | * | |
1112 | * If a network interface is not present and the process has suitable | |
1113 | * privileges this function loads the module. If module loading is not | |
1114 | * available in this kernel then it becomes a nop. | |
1115 | */ | |
1116 | ||
881d966b | 1117 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1118 | { |
4ec93edb | 1119 | struct net_device *dev; |
1da177e4 | 1120 | |
72c9528b ED |
1121 | rcu_read_lock(); |
1122 | dev = dev_get_by_name_rcu(net, name); | |
1123 | rcu_read_unlock(); | |
1da177e4 | 1124 | |
a8f80e8f | 1125 | if (!dev && capable(CAP_NET_ADMIN)) |
1da177e4 LT |
1126 | request_module("%s", name); |
1127 | } | |
d1b19dff | 1128 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1129 | |
bd380811 | 1130 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1131 | { |
d314774c | 1132 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1133 | int ret; |
1da177e4 | 1134 | |
e46b66bc BH |
1135 | ASSERT_RTNL(); |
1136 | ||
1da177e4 LT |
1137 | /* |
1138 | * Is it even present? | |
1139 | */ | |
1140 | if (!netif_device_present(dev)) | |
1141 | return -ENODEV; | |
1142 | ||
3b8bcfd5 JB |
1143 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1144 | ret = notifier_to_errno(ret); | |
1145 | if (ret) | |
1146 | return ret; | |
1147 | ||
1da177e4 LT |
1148 | /* |
1149 | * Call device private open method | |
1150 | */ | |
1151 | set_bit(__LINK_STATE_START, &dev->state); | |
bada339b | 1152 | |
d314774c SH |
1153 | if (ops->ndo_validate_addr) |
1154 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1155 | |
d314774c SH |
1156 | if (!ret && ops->ndo_open) |
1157 | ret = ops->ndo_open(dev); | |
1da177e4 | 1158 | |
4ec93edb | 1159 | /* |
1da177e4 LT |
1160 | * If it went open OK then: |
1161 | */ | |
1162 | ||
bada339b JG |
1163 | if (ret) |
1164 | clear_bit(__LINK_STATE_START, &dev->state); | |
1165 | else { | |
1da177e4 LT |
1166 | /* |
1167 | * Set the flags. | |
1168 | */ | |
1169 | dev->flags |= IFF_UP; | |
1170 | ||
649274d9 DW |
1171 | /* |
1172 | * Enable NET_DMA | |
1173 | */ | |
b4bd07c2 | 1174 | net_dmaengine_get(); |
649274d9 | 1175 | |
1da177e4 LT |
1176 | /* |
1177 | * Initialize multicasting status | |
1178 | */ | |
4417da66 | 1179 | dev_set_rx_mode(dev); |
1da177e4 LT |
1180 | |
1181 | /* | |
1182 | * Wakeup transmit queue engine | |
1183 | */ | |
1184 | dev_activate(dev); | |
1da177e4 | 1185 | } |
bada339b | 1186 | |
1da177e4 LT |
1187 | return ret; |
1188 | } | |
1189 | ||
1190 | /** | |
bd380811 PM |
1191 | * dev_open - prepare an interface for use. |
1192 | * @dev: device to open | |
1da177e4 | 1193 | * |
bd380811 PM |
1194 | * Takes a device from down to up state. The device's private open |
1195 | * function is invoked and then the multicast lists are loaded. Finally | |
1196 | * the device is moved into the up state and a %NETDEV_UP message is | |
1197 | * sent to the netdev notifier chain. | |
1198 | * | |
1199 | * Calling this function on an active interface is a nop. On a failure | |
1200 | * a negative errno code is returned. | |
1da177e4 | 1201 | */ |
bd380811 PM |
1202 | int dev_open(struct net_device *dev) |
1203 | { | |
1204 | int ret; | |
1205 | ||
1206 | /* | |
1207 | * Is it already up? | |
1208 | */ | |
1209 | if (dev->flags & IFF_UP) | |
1210 | return 0; | |
1211 | ||
1212 | /* | |
1213 | * Open device | |
1214 | */ | |
1215 | ret = __dev_open(dev); | |
1216 | if (ret < 0) | |
1217 | return ret; | |
1218 | ||
1219 | /* | |
1220 | * ... and announce new interface. | |
1221 | */ | |
1222 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1223 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1224 | ||
1225 | return ret; | |
1226 | } | |
1227 | EXPORT_SYMBOL(dev_open); | |
1228 | ||
1229 | static int __dev_close(struct net_device *dev) | |
1da177e4 | 1230 | { |
d314774c | 1231 | const struct net_device_ops *ops = dev->netdev_ops; |
e46b66bc | 1232 | |
bd380811 | 1233 | ASSERT_RTNL(); |
9d5010db DM |
1234 | might_sleep(); |
1235 | ||
1da177e4 LT |
1236 | /* |
1237 | * Tell people we are going down, so that they can | |
1238 | * prepare to death, when device is still operating. | |
1239 | */ | |
056925ab | 1240 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1241 | |
1da177e4 LT |
1242 | clear_bit(__LINK_STATE_START, &dev->state); |
1243 | ||
1244 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
bea3348e SH |
1245 | * it can be even on different cpu. So just clear netif_running(). |
1246 | * | |
1247 | * dev->stop() will invoke napi_disable() on all of it's | |
1248 | * napi_struct instances on this device. | |
1249 | */ | |
1da177e4 | 1250 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ |
1da177e4 | 1251 | |
d8b2a4d2 ML |
1252 | dev_deactivate(dev); |
1253 | ||
1da177e4 LT |
1254 | /* |
1255 | * Call the device specific close. This cannot fail. | |
1256 | * Only if device is UP | |
1257 | * | |
1258 | * We allow it to be called even after a DETACH hot-plug | |
1259 | * event. | |
1260 | */ | |
d314774c SH |
1261 | if (ops->ndo_stop) |
1262 | ops->ndo_stop(dev); | |
1da177e4 LT |
1263 | |
1264 | /* | |
1265 | * Device is now down. | |
1266 | */ | |
1267 | ||
1268 | dev->flags &= ~IFF_UP; | |
1269 | ||
1270 | /* | |
bd380811 | 1271 | * Shutdown NET_DMA |
1da177e4 | 1272 | */ |
bd380811 PM |
1273 | net_dmaengine_put(); |
1274 | ||
1275 | return 0; | |
1276 | } | |
1277 | ||
1278 | /** | |
1279 | * dev_close - shutdown an interface. | |
1280 | * @dev: device to shutdown | |
1281 | * | |
1282 | * This function moves an active device into down state. A | |
1283 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1284 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1285 | * chain. | |
1286 | */ | |
1287 | int dev_close(struct net_device *dev) | |
1288 | { | |
1289 | if (!(dev->flags & IFF_UP)) | |
1290 | return 0; | |
1291 | ||
1292 | __dev_close(dev); | |
1da177e4 | 1293 | |
649274d9 | 1294 | /* |
bd380811 | 1295 | * Tell people we are down |
649274d9 | 1296 | */ |
bd380811 PM |
1297 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1298 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
649274d9 | 1299 | |
1da177e4 LT |
1300 | return 0; |
1301 | } | |
d1b19dff | 1302 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1303 | |
1304 | ||
0187bdfb BH |
1305 | /** |
1306 | * dev_disable_lro - disable Large Receive Offload on a device | |
1307 | * @dev: device | |
1308 | * | |
1309 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1310 | * called under RTNL. This is needed if received packets may be | |
1311 | * forwarded to another interface. | |
1312 | */ | |
1313 | void dev_disable_lro(struct net_device *dev) | |
1314 | { | |
1315 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1316 | dev->ethtool_ops->set_flags) { | |
1317 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1318 | if (flags & ETH_FLAG_LRO) { | |
1319 | flags &= ~ETH_FLAG_LRO; | |
1320 | dev->ethtool_ops->set_flags(dev, flags); | |
1321 | } | |
1322 | } | |
1323 | WARN_ON(dev->features & NETIF_F_LRO); | |
1324 | } | |
1325 | EXPORT_SYMBOL(dev_disable_lro); | |
1326 | ||
1327 | ||
881d966b EB |
1328 | static int dev_boot_phase = 1; |
1329 | ||
1da177e4 LT |
1330 | /* |
1331 | * Device change register/unregister. These are not inline or static | |
1332 | * as we export them to the world. | |
1333 | */ | |
1334 | ||
1335 | /** | |
1336 | * register_netdevice_notifier - register a network notifier block | |
1337 | * @nb: notifier | |
1338 | * | |
1339 | * Register a notifier to be called when network device events occur. | |
1340 | * The notifier passed is linked into the kernel structures and must | |
1341 | * not be reused until it has been unregistered. A negative errno code | |
1342 | * is returned on a failure. | |
1343 | * | |
1344 | * When registered all registration and up events are replayed | |
4ec93edb | 1345 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1346 | * view of the network device list. |
1347 | */ | |
1348 | ||
1349 | int register_netdevice_notifier(struct notifier_block *nb) | |
1350 | { | |
1351 | struct net_device *dev; | |
fcc5a03a | 1352 | struct net_device *last; |
881d966b | 1353 | struct net *net; |
1da177e4 LT |
1354 | int err; |
1355 | ||
1356 | rtnl_lock(); | |
f07d5b94 | 1357 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1358 | if (err) |
1359 | goto unlock; | |
881d966b EB |
1360 | if (dev_boot_phase) |
1361 | goto unlock; | |
1362 | for_each_net(net) { | |
1363 | for_each_netdev(net, dev) { | |
1364 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1365 | err = notifier_to_errno(err); | |
1366 | if (err) | |
1367 | goto rollback; | |
1368 | ||
1369 | if (!(dev->flags & IFF_UP)) | |
1370 | continue; | |
1da177e4 | 1371 | |
881d966b EB |
1372 | nb->notifier_call(nb, NETDEV_UP, dev); |
1373 | } | |
1da177e4 | 1374 | } |
fcc5a03a HX |
1375 | |
1376 | unlock: | |
1da177e4 LT |
1377 | rtnl_unlock(); |
1378 | return err; | |
fcc5a03a HX |
1379 | |
1380 | rollback: | |
1381 | last = dev; | |
881d966b EB |
1382 | for_each_net(net) { |
1383 | for_each_netdev(net, dev) { | |
1384 | if (dev == last) | |
1385 | break; | |
fcc5a03a | 1386 | |
881d966b EB |
1387 | if (dev->flags & IFF_UP) { |
1388 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1389 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1390 | } | |
1391 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
a5ee1551 | 1392 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); |
fcc5a03a | 1393 | } |
fcc5a03a | 1394 | } |
c67625a1 PE |
1395 | |
1396 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1397 | goto unlock; |
1da177e4 | 1398 | } |
d1b19dff | 1399 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1400 | |
1401 | /** | |
1402 | * unregister_netdevice_notifier - unregister a network notifier block | |
1403 | * @nb: notifier | |
1404 | * | |
1405 | * Unregister a notifier previously registered by | |
1406 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1407 | * kernel structures and may then be reused. A negative errno code | |
1408 | * is returned on a failure. | |
1409 | */ | |
1410 | ||
1411 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1412 | { | |
9f514950 HX |
1413 | int err; |
1414 | ||
1415 | rtnl_lock(); | |
f07d5b94 | 1416 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1417 | rtnl_unlock(); |
1418 | return err; | |
1da177e4 | 1419 | } |
d1b19dff | 1420 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1421 | |
1422 | /** | |
1423 | * call_netdevice_notifiers - call all network notifier blocks | |
1424 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1425 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1426 | * |
1427 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1428 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1429 | */ |
1430 | ||
ad7379d4 | 1431 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1432 | { |
ab930471 | 1433 | ASSERT_RTNL(); |
ad7379d4 | 1434 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 LT |
1435 | } |
1436 | ||
1437 | /* When > 0 there are consumers of rx skb time stamps */ | |
1438 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1439 | ||
1440 | void net_enable_timestamp(void) | |
1441 | { | |
1442 | atomic_inc(&netstamp_needed); | |
1443 | } | |
d1b19dff | 1444 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1445 | |
1446 | void net_disable_timestamp(void) | |
1447 | { | |
1448 | atomic_dec(&netstamp_needed); | |
1449 | } | |
d1b19dff | 1450 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1451 | |
3b098e2d | 1452 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 LT |
1453 | { |
1454 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1455 | __net_timestamp(skb); |
b7aa0bf7 ED |
1456 | else |
1457 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1458 | } |
1459 | ||
3b098e2d ED |
1460 | static inline void net_timestamp_check(struct sk_buff *skb) |
1461 | { | |
1462 | if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed)) | |
1463 | __net_timestamp(skb); | |
1464 | } | |
1465 | ||
44540960 AB |
1466 | /** |
1467 | * dev_forward_skb - loopback an skb to another netif | |
1468 | * | |
1469 | * @dev: destination network device | |
1470 | * @skb: buffer to forward | |
1471 | * | |
1472 | * return values: | |
1473 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1474 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1475 | * |
1476 | * dev_forward_skb can be used for injecting an skb from the | |
1477 | * start_xmit function of one device into the receive queue | |
1478 | * of another device. | |
1479 | * | |
1480 | * The receiving device may be in another namespace, so | |
1481 | * we have to clear all information in the skb that could | |
1482 | * impact namespace isolation. | |
1483 | */ | |
1484 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1485 | { | |
1486 | skb_orphan(skb); | |
1487 | ||
6ec82562 ED |
1488 | if (!(dev->flags & IFF_UP) || |
1489 | (skb->len > (dev->mtu + dev->hard_header_len))) { | |
1490 | kfree_skb(skb); | |
44540960 | 1491 | return NET_RX_DROP; |
6ec82562 | 1492 | } |
8a83a00b | 1493 | skb_set_dev(skb, dev); |
44540960 AB |
1494 | skb->tstamp.tv64 = 0; |
1495 | skb->pkt_type = PACKET_HOST; | |
1496 | skb->protocol = eth_type_trans(skb, dev); | |
44540960 AB |
1497 | return netif_rx(skb); |
1498 | } | |
1499 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1500 | ||
1da177e4 LT |
1501 | /* |
1502 | * Support routine. Sends outgoing frames to any network | |
1503 | * taps currently in use. | |
1504 | */ | |
1505 | ||
f6a78bfc | 1506 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1507 | { |
1508 | struct packet_type *ptype; | |
a61bbcf2 | 1509 | |
8caf1539 JP |
1510 | #ifdef CONFIG_NET_CLS_ACT |
1511 | if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS))) | |
3b098e2d | 1512 | net_timestamp_set(skb); |
8caf1539 | 1513 | #else |
3b098e2d | 1514 | net_timestamp_set(skb); |
8caf1539 | 1515 | #endif |
1da177e4 LT |
1516 | |
1517 | rcu_read_lock(); | |
1518 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1519 | /* Never send packets back to the socket | |
1520 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1521 | */ | |
1522 | if ((ptype->dev == dev || !ptype->dev) && | |
1523 | (ptype->af_packet_priv == NULL || | |
1524 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
d1b19dff | 1525 | struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); |
1da177e4 LT |
1526 | if (!skb2) |
1527 | break; | |
1528 | ||
1529 | /* skb->nh should be correctly | |
1530 | set by sender, so that the second statement is | |
1531 | just protection against buggy protocols. | |
1532 | */ | |
459a98ed | 1533 | skb_reset_mac_header(skb2); |
1da177e4 | 1534 | |
d56f90a7 | 1535 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1536 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1537 | if (net_ratelimit()) |
1538 | printk(KERN_CRIT "protocol %04x is " | |
1539 | "buggy, dev %s\n", | |
1540 | skb2->protocol, dev->name); | |
c1d2bbe1 | 1541 | skb_reset_network_header(skb2); |
1da177e4 LT |
1542 | } |
1543 | ||
b0e380b1 | 1544 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1545 | skb2->pkt_type = PACKET_OUTGOING; |
f2ccd8fa | 1546 | ptype->func(skb2, skb->dev, ptype, skb->dev); |
1da177e4 LT |
1547 | } |
1548 | } | |
1549 | rcu_read_unlock(); | |
1550 | } | |
1551 | ||
56079431 | 1552 | |
def82a1d | 1553 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1554 | { |
def82a1d JP |
1555 | struct softnet_data *sd; |
1556 | unsigned long flags; | |
56079431 | 1557 | |
def82a1d JP |
1558 | local_irq_save(flags); |
1559 | sd = &__get_cpu_var(softnet_data); | |
a9cbd588 CG |
1560 | q->next_sched = NULL; |
1561 | *sd->output_queue_tailp = q; | |
1562 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
1563 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
1564 | local_irq_restore(flags); | |
1565 | } | |
1566 | ||
1567 | void __netif_schedule(struct Qdisc *q) | |
1568 | { | |
1569 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1570 | __netif_reschedule(q); | |
56079431 DV |
1571 | } |
1572 | EXPORT_SYMBOL(__netif_schedule); | |
1573 | ||
bea3348e | 1574 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1575 | { |
15e83ed7 ED |
1576 | if (!skb->destructor) |
1577 | dev_kfree_skb(skb); | |
1578 | else if (atomic_dec_and_test(&skb->users)) { | |
bea3348e SH |
1579 | struct softnet_data *sd; |
1580 | unsigned long flags; | |
56079431 | 1581 | |
bea3348e SH |
1582 | local_irq_save(flags); |
1583 | sd = &__get_cpu_var(softnet_data); | |
1584 | skb->next = sd->completion_queue; | |
1585 | sd->completion_queue = skb; | |
1586 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1587 | local_irq_restore(flags); | |
1588 | } | |
56079431 | 1589 | } |
bea3348e | 1590 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1591 | |
1592 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1593 | { | |
1594 | if (in_irq() || irqs_disabled()) | |
1595 | dev_kfree_skb_irq(skb); | |
1596 | else | |
1597 | dev_kfree_skb(skb); | |
1598 | } | |
1599 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1600 | ||
1601 | ||
bea3348e SH |
1602 | /** |
1603 | * netif_device_detach - mark device as removed | |
1604 | * @dev: network device | |
1605 | * | |
1606 | * Mark device as removed from system and therefore no longer available. | |
1607 | */ | |
56079431 DV |
1608 | void netif_device_detach(struct net_device *dev) |
1609 | { | |
1610 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1611 | netif_running(dev)) { | |
d543103a | 1612 | netif_tx_stop_all_queues(dev); |
56079431 DV |
1613 | } |
1614 | } | |
1615 | EXPORT_SYMBOL(netif_device_detach); | |
1616 | ||
bea3348e SH |
1617 | /** |
1618 | * netif_device_attach - mark device as attached | |
1619 | * @dev: network device | |
1620 | * | |
1621 | * Mark device as attached from system and restart if needed. | |
1622 | */ | |
56079431 DV |
1623 | void netif_device_attach(struct net_device *dev) |
1624 | { | |
1625 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1626 | netif_running(dev)) { | |
d543103a | 1627 | netif_tx_wake_all_queues(dev); |
4ec93edb | 1628 | __netdev_watchdog_up(dev); |
56079431 DV |
1629 | } |
1630 | } | |
1631 | EXPORT_SYMBOL(netif_device_attach); | |
1632 | ||
6de329e2 BH |
1633 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) |
1634 | { | |
1635 | return ((features & NETIF_F_GEN_CSUM) || | |
1636 | ((features & NETIF_F_IP_CSUM) && | |
1637 | protocol == htons(ETH_P_IP)) || | |
1638 | ((features & NETIF_F_IPV6_CSUM) && | |
1c8dbcf6 YZ |
1639 | protocol == htons(ETH_P_IPV6)) || |
1640 | ((features & NETIF_F_FCOE_CRC) && | |
1641 | protocol == htons(ETH_P_FCOE))); | |
6de329e2 BH |
1642 | } |
1643 | ||
1644 | static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb) | |
1645 | { | |
1646 | if (can_checksum_protocol(dev->features, skb->protocol)) | |
1647 | return true; | |
1648 | ||
1649 | if (skb->protocol == htons(ETH_P_8021Q)) { | |
1650 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
1651 | if (can_checksum_protocol(dev->features & dev->vlan_features, | |
1652 | veh->h_vlan_encapsulated_proto)) | |
1653 | return true; | |
1654 | } | |
1655 | ||
1656 | return false; | |
1657 | } | |
56079431 | 1658 | |
8a83a00b AB |
1659 | /** |
1660 | * skb_dev_set -- assign a new device to a buffer | |
1661 | * @skb: buffer for the new device | |
1662 | * @dev: network device | |
1663 | * | |
1664 | * If an skb is owned by a device already, we have to reset | |
1665 | * all data private to the namespace a device belongs to | |
1666 | * before assigning it a new device. | |
1667 | */ | |
1668 | #ifdef CONFIG_NET_NS | |
1669 | void skb_set_dev(struct sk_buff *skb, struct net_device *dev) | |
1670 | { | |
1671 | skb_dst_drop(skb); | |
1672 | if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) { | |
1673 | secpath_reset(skb); | |
1674 | nf_reset(skb); | |
1675 | skb_init_secmark(skb); | |
1676 | skb->mark = 0; | |
1677 | skb->priority = 0; | |
1678 | skb->nf_trace = 0; | |
1679 | skb->ipvs_property = 0; | |
1680 | #ifdef CONFIG_NET_SCHED | |
1681 | skb->tc_index = 0; | |
1682 | #endif | |
1683 | } | |
1684 | skb->dev = dev; | |
1685 | } | |
1686 | EXPORT_SYMBOL(skb_set_dev); | |
1687 | #endif /* CONFIG_NET_NS */ | |
1688 | ||
1da177e4 LT |
1689 | /* |
1690 | * Invalidate hardware checksum when packet is to be mangled, and | |
1691 | * complete checksum manually on outgoing path. | |
1692 | */ | |
84fa7933 | 1693 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1694 | { |
d3bc23e7 | 1695 | __wsum csum; |
663ead3b | 1696 | int ret = 0, offset; |
1da177e4 | 1697 | |
84fa7933 | 1698 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1699 | goto out_set_summed; |
1700 | ||
1701 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1702 | /* Let GSO fix up the checksum. */ |
1703 | goto out_set_summed; | |
1da177e4 LT |
1704 | } |
1705 | ||
a030847e HX |
1706 | offset = skb->csum_start - skb_headroom(skb); |
1707 | BUG_ON(offset >= skb_headlen(skb)); | |
1708 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1709 | ||
1710 | offset += skb->csum_offset; | |
1711 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1712 | ||
1713 | if (skb_cloned(skb) && | |
1714 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1715 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1716 | if (ret) | |
1717 | goto out; | |
1718 | } | |
1719 | ||
a030847e | 1720 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1721 | out_set_summed: |
1da177e4 | 1722 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1723 | out: |
1da177e4 LT |
1724 | return ret; |
1725 | } | |
d1b19dff | 1726 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 1727 | |
f6a78bfc HX |
1728 | /** |
1729 | * skb_gso_segment - Perform segmentation on skb. | |
1730 | * @skb: buffer to segment | |
576a30eb | 1731 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1732 | * |
1733 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1734 | * |
1735 | * It may return NULL if the skb requires no segmentation. This is | |
1736 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1737 | */ |
576a30eb | 1738 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc HX |
1739 | { |
1740 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1741 | struct packet_type *ptype; | |
252e3346 | 1742 | __be16 type = skb->protocol; |
a430a43d | 1743 | int err; |
f6a78bfc | 1744 | |
459a98ed | 1745 | skb_reset_mac_header(skb); |
b0e380b1 | 1746 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1747 | __skb_pull(skb, skb->mac_len); |
1748 | ||
67fd1a73 HX |
1749 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
1750 | struct net_device *dev = skb->dev; | |
1751 | struct ethtool_drvinfo info = {}; | |
1752 | ||
1753 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1754 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1755 | ||
1756 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d " | |
1757 | "ip_summed=%d", | |
1758 | info.driver, dev ? dev->features : 0L, | |
1759 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1760 | skb->len, skb->data_len, skb->ip_summed); | |
1761 | ||
a430a43d HX |
1762 | if (skb_header_cloned(skb) && |
1763 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1764 | return ERR_PTR(err); | |
1765 | } | |
1766 | ||
f6a78bfc | 1767 | rcu_read_lock(); |
82d8a867 PE |
1768 | list_for_each_entry_rcu(ptype, |
1769 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1770 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1771 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1772 | err = ptype->gso_send_check(skb); |
1773 | segs = ERR_PTR(err); | |
1774 | if (err || skb_gso_ok(skb, features)) | |
1775 | break; | |
d56f90a7 ACM |
1776 | __skb_push(skb, (skb->data - |
1777 | skb_network_header(skb))); | |
a430a43d | 1778 | } |
576a30eb | 1779 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1780 | break; |
1781 | } | |
1782 | } | |
1783 | rcu_read_unlock(); | |
1784 | ||
98e399f8 | 1785 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1786 | |
f6a78bfc HX |
1787 | return segs; |
1788 | } | |
f6a78bfc HX |
1789 | EXPORT_SYMBOL(skb_gso_segment); |
1790 | ||
fb286bb2 HX |
1791 | /* Take action when hardware reception checksum errors are detected. */ |
1792 | #ifdef CONFIG_BUG | |
1793 | void netdev_rx_csum_fault(struct net_device *dev) | |
1794 | { | |
1795 | if (net_ratelimit()) { | |
4ec93edb | 1796 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1797 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1798 | dump_stack(); |
1799 | } | |
1800 | } | |
1801 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1802 | #endif | |
1803 | ||
1da177e4 LT |
1804 | /* Actually, we should eliminate this check as soon as we know, that: |
1805 | * 1. IOMMU is present and allows to map all the memory. | |
1806 | * 2. No high memory really exists on this machine. | |
1807 | */ | |
1808 | ||
9092c658 | 1809 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 1810 | { |
3d3a8533 | 1811 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1812 | int i; |
5acbbd42 FT |
1813 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
1814 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1815 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1816 | return 1; | |
1817 | } | |
1da177e4 | 1818 | |
5acbbd42 FT |
1819 | if (PCI_DMA_BUS_IS_PHYS) { |
1820 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 1821 | |
9092c658 ED |
1822 | if (!pdev) |
1823 | return 0; | |
5acbbd42 FT |
1824 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
1825 | dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page); | |
1826 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) | |
1827 | return 1; | |
1828 | } | |
1829 | } | |
3d3a8533 | 1830 | #endif |
1da177e4 LT |
1831 | return 0; |
1832 | } | |
1da177e4 | 1833 | |
f6a78bfc HX |
1834 | struct dev_gso_cb { |
1835 | void (*destructor)(struct sk_buff *skb); | |
1836 | }; | |
1837 | ||
1838 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1839 | ||
1840 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1841 | { | |
1842 | struct dev_gso_cb *cb; | |
1843 | ||
1844 | do { | |
1845 | struct sk_buff *nskb = skb->next; | |
1846 | ||
1847 | skb->next = nskb->next; | |
1848 | nskb->next = NULL; | |
1849 | kfree_skb(nskb); | |
1850 | } while (skb->next); | |
1851 | ||
1852 | cb = DEV_GSO_CB(skb); | |
1853 | if (cb->destructor) | |
1854 | cb->destructor(skb); | |
1855 | } | |
1856 | ||
1857 | /** | |
1858 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1859 | * @skb: buffer to segment | |
1860 | * | |
1861 | * This function segments the given skb and stores the list of segments | |
1862 | * in skb->next. | |
1863 | */ | |
1864 | static int dev_gso_segment(struct sk_buff *skb) | |
1865 | { | |
1866 | struct net_device *dev = skb->dev; | |
1867 | struct sk_buff *segs; | |
576a30eb HX |
1868 | int features = dev->features & ~(illegal_highdma(dev, skb) ? |
1869 | NETIF_F_SG : 0); | |
1870 | ||
1871 | segs = skb_gso_segment(skb, features); | |
1872 | ||
1873 | /* Verifying header integrity only. */ | |
1874 | if (!segs) | |
1875 | return 0; | |
f6a78bfc | 1876 | |
801678c5 | 1877 | if (IS_ERR(segs)) |
f6a78bfc HX |
1878 | return PTR_ERR(segs); |
1879 | ||
1880 | skb->next = segs; | |
1881 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1882 | skb->destructor = dev_gso_skb_destructor; | |
1883 | ||
1884 | return 0; | |
1885 | } | |
1886 | ||
fc6055a5 ED |
1887 | /* |
1888 | * Try to orphan skb early, right before transmission by the device. | |
1889 | * We cannot orphan skb if tx timestamp is requested, since | |
1890 | * drivers need to call skb_tstamp_tx() to send the timestamp. | |
1891 | */ | |
1892 | static inline void skb_orphan_try(struct sk_buff *skb) | |
1893 | { | |
1894 | if (!skb_tx(skb)->flags) | |
1895 | skb_orphan(skb); | |
1896 | } | |
1897 | ||
6afff0ca JF |
1898 | /* |
1899 | * Returns true if either: | |
1900 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
1901 | * 2. skb is fragmented and the device does not support SG, or if | |
1902 | * at least one of fragments is in highmem and device does not | |
1903 | * support DMA from it. | |
1904 | */ | |
1905 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
1906 | struct net_device *dev) | |
1907 | { | |
1908 | return skb_is_nonlinear(skb) && | |
1909 | ((skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) || | |
1910 | (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) || | |
1911 | illegal_highdma(dev, skb)))); | |
1912 | } | |
1913 | ||
fd2ea0a7 DM |
1914 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
1915 | struct netdev_queue *txq) | |
f6a78bfc | 1916 | { |
00829823 | 1917 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 1918 | int rc = NETDEV_TX_OK; |
00829823 | 1919 | |
f6a78bfc | 1920 | if (likely(!skb->next)) { |
9be9a6b9 | 1921 | if (!list_empty(&ptype_all)) |
f6a78bfc HX |
1922 | dev_queue_xmit_nit(skb, dev); |
1923 | ||
93f154b5 ED |
1924 | /* |
1925 | * If device doesnt need skb->dst, release it right now while | |
1926 | * its hot in this cpu cache | |
1927 | */ | |
adf30907 ED |
1928 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
1929 | skb_dst_drop(skb); | |
1930 | ||
fc6055a5 | 1931 | skb_orphan_try(skb); |
9ccb8975 DM |
1932 | |
1933 | if (netif_needs_gso(dev, skb)) { | |
1934 | if (unlikely(dev_gso_segment(skb))) | |
1935 | goto out_kfree_skb; | |
1936 | if (skb->next) | |
1937 | goto gso; | |
6afff0ca JF |
1938 | } else { |
1939 | if (skb_needs_linearize(skb, dev) && | |
1940 | __skb_linearize(skb)) | |
1941 | goto out_kfree_skb; | |
1942 | ||
1943 | /* If packet is not checksummed and device does not | |
1944 | * support checksumming for this protocol, complete | |
1945 | * checksumming here. | |
1946 | */ | |
1947 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
1948 | skb_set_transport_header(skb, skb->csum_start - | |
1949 | skb_headroom(skb)); | |
1950 | if (!dev_can_checksum(dev, skb) && | |
1951 | skb_checksum_help(skb)) | |
1952 | goto out_kfree_skb; | |
1953 | } | |
9ccb8975 DM |
1954 | } |
1955 | ||
ac45f602 | 1956 | rc = ops->ndo_start_xmit(skb, dev); |
ec634fe3 | 1957 | if (rc == NETDEV_TX_OK) |
08baf561 | 1958 | txq_trans_update(txq); |
ac45f602 | 1959 | return rc; |
f6a78bfc HX |
1960 | } |
1961 | ||
576a30eb | 1962 | gso: |
f6a78bfc HX |
1963 | do { |
1964 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
1965 | |
1966 | skb->next = nskb->next; | |
1967 | nskb->next = NULL; | |
068a2de5 KK |
1968 | |
1969 | /* | |
1970 | * If device doesnt need nskb->dst, release it right now while | |
1971 | * its hot in this cpu cache | |
1972 | */ | |
1973 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
1974 | skb_dst_drop(nskb); | |
1975 | ||
00829823 | 1976 | rc = ops->ndo_start_xmit(nskb, dev); |
ec634fe3 | 1977 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
1978 | if (rc & ~NETDEV_TX_MASK) |
1979 | goto out_kfree_gso_skb; | |
f54d9e8d | 1980 | nskb->next = skb->next; |
f6a78bfc HX |
1981 | skb->next = nskb; |
1982 | return rc; | |
1983 | } | |
08baf561 | 1984 | txq_trans_update(txq); |
fd2ea0a7 | 1985 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) |
f54d9e8d | 1986 | return NETDEV_TX_BUSY; |
f6a78bfc | 1987 | } while (skb->next); |
4ec93edb | 1988 | |
572a9d7b PM |
1989 | out_kfree_gso_skb: |
1990 | if (likely(skb->next == NULL)) | |
1991 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
1992 | out_kfree_skb: |
1993 | kfree_skb(skb); | |
572a9d7b | 1994 | return rc; |
f6a78bfc HX |
1995 | } |
1996 | ||
0a9627f2 | 1997 | static u32 hashrnd __read_mostly; |
b6b2fed1 | 1998 | |
9247744e | 1999 | u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb) |
8f0f2223 | 2000 | { |
7019298a | 2001 | u32 hash; |
b6b2fed1 | 2002 | |
513de11b DM |
2003 | if (skb_rx_queue_recorded(skb)) { |
2004 | hash = skb_get_rx_queue(skb); | |
d1b19dff | 2005 | while (unlikely(hash >= dev->real_num_tx_queues)) |
513de11b DM |
2006 | hash -= dev->real_num_tx_queues; |
2007 | return hash; | |
2008 | } | |
ec581f6a ED |
2009 | |
2010 | if (skb->sk && skb->sk->sk_hash) | |
7019298a | 2011 | hash = skb->sk->sk_hash; |
ec581f6a | 2012 | else |
b249dcb8 | 2013 | hash = (__force u16) skb->protocol; |
d5a9e24a | 2014 | |
0a9627f2 | 2015 | hash = jhash_1word(hash, hashrnd); |
b6b2fed1 DM |
2016 | |
2017 | return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32); | |
8f0f2223 | 2018 | } |
9247744e | 2019 | EXPORT_SYMBOL(skb_tx_hash); |
8f0f2223 | 2020 | |
ed04642f ED |
2021 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
2022 | { | |
2023 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
2024 | if (net_ratelimit()) { | |
7a161ea9 ED |
2025 | pr_warning("%s selects TX queue %d, but " |
2026 | "real number of TX queues is %d\n", | |
2027 | dev->name, queue_index, dev->real_num_tx_queues); | |
ed04642f ED |
2028 | } |
2029 | return 0; | |
2030 | } | |
2031 | return queue_index; | |
2032 | } | |
2033 | ||
e8a0464c DM |
2034 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
2035 | struct sk_buff *skb) | |
2036 | { | |
a4ee3ce3 KK |
2037 | u16 queue_index; |
2038 | struct sock *sk = skb->sk; | |
2039 | ||
2040 | if (sk_tx_queue_recorded(sk)) { | |
2041 | queue_index = sk_tx_queue_get(sk); | |
2042 | } else { | |
2043 | const struct net_device_ops *ops = dev->netdev_ops; | |
2044 | ||
2045 | if (ops->ndo_select_queue) { | |
2046 | queue_index = ops->ndo_select_queue(dev, skb); | |
ed04642f | 2047 | queue_index = dev_cap_txqueue(dev, queue_index); |
a4ee3ce3 KK |
2048 | } else { |
2049 | queue_index = 0; | |
2050 | if (dev->real_num_tx_queues > 1) | |
2051 | queue_index = skb_tx_hash(dev, skb); | |
fd2ea0a7 | 2052 | |
8728c544 | 2053 | if (sk) { |
87eb3670 | 2054 | struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1); |
8728c544 ED |
2055 | |
2056 | if (dst && skb_dst(skb) == dst) | |
2057 | sk_tx_queue_set(sk, queue_index); | |
2058 | } | |
a4ee3ce3 KK |
2059 | } |
2060 | } | |
eae792b7 | 2061 | |
fd2ea0a7 DM |
2062 | skb_set_queue_mapping(skb, queue_index); |
2063 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
2064 | } |
2065 | ||
bbd8a0d3 KK |
2066 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2067 | struct net_device *dev, | |
2068 | struct netdev_queue *txq) | |
2069 | { | |
2070 | spinlock_t *root_lock = qdisc_lock(q); | |
79640a4c | 2071 | bool contended = qdisc_is_running(q); |
bbd8a0d3 KK |
2072 | int rc; |
2073 | ||
79640a4c ED |
2074 | /* |
2075 | * Heuristic to force contended enqueues to serialize on a | |
2076 | * separate lock before trying to get qdisc main lock. | |
2077 | * This permits __QDISC_STATE_RUNNING owner to get the lock more often | |
2078 | * and dequeue packets faster. | |
2079 | */ | |
2080 | if (unlikely(contended)) | |
2081 | spin_lock(&q->busylock); | |
2082 | ||
bbd8a0d3 KK |
2083 | spin_lock(root_lock); |
2084 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2085 | kfree_skb(skb); | |
2086 | rc = NET_XMIT_DROP; | |
2087 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 2088 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
2089 | /* |
2090 | * This is a work-conserving queue; there are no old skbs | |
2091 | * waiting to be sent out; and the qdisc is not running - | |
2092 | * xmit the skb directly. | |
2093 | */ | |
7fee226a ED |
2094 | if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) |
2095 | skb_dst_force(skb); | |
bbd8a0d3 | 2096 | __qdisc_update_bstats(q, skb->len); |
79640a4c ED |
2097 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { |
2098 | if (unlikely(contended)) { | |
2099 | spin_unlock(&q->busylock); | |
2100 | contended = false; | |
2101 | } | |
bbd8a0d3 | 2102 | __qdisc_run(q); |
79640a4c | 2103 | } else |
bc135b23 | 2104 | qdisc_run_end(q); |
bbd8a0d3 KK |
2105 | |
2106 | rc = NET_XMIT_SUCCESS; | |
2107 | } else { | |
7fee226a | 2108 | skb_dst_force(skb); |
bbd8a0d3 | 2109 | rc = qdisc_enqueue_root(skb, q); |
79640a4c ED |
2110 | if (qdisc_run_begin(q)) { |
2111 | if (unlikely(contended)) { | |
2112 | spin_unlock(&q->busylock); | |
2113 | contended = false; | |
2114 | } | |
2115 | __qdisc_run(q); | |
2116 | } | |
bbd8a0d3 KK |
2117 | } |
2118 | spin_unlock(root_lock); | |
79640a4c ED |
2119 | if (unlikely(contended)) |
2120 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
2121 | return rc; |
2122 | } | |
2123 | ||
d29f749e DJ |
2124 | /** |
2125 | * dev_queue_xmit - transmit a buffer | |
2126 | * @skb: buffer to transmit | |
2127 | * | |
2128 | * Queue a buffer for transmission to a network device. The caller must | |
2129 | * have set the device and priority and built the buffer before calling | |
2130 | * this function. The function can be called from an interrupt. | |
2131 | * | |
2132 | * A negative errno code is returned on a failure. A success does not | |
2133 | * guarantee the frame will be transmitted as it may be dropped due | |
2134 | * to congestion or traffic shaping. | |
2135 | * | |
2136 | * ----------------------------------------------------------------------------------- | |
2137 | * I notice this method can also return errors from the queue disciplines, | |
2138 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2139 | * be positive. | |
2140 | * | |
2141 | * Regardless of the return value, the skb is consumed, so it is currently | |
2142 | * difficult to retry a send to this method. (You can bump the ref count | |
2143 | * before sending to hold a reference for retry if you are careful.) | |
2144 | * | |
2145 | * When calling this method, interrupts MUST be enabled. This is because | |
2146 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2147 | * --BLG | |
2148 | */ | |
1da177e4 LT |
2149 | int dev_queue_xmit(struct sk_buff *skb) |
2150 | { | |
2151 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2152 | struct netdev_queue *txq; |
1da177e4 LT |
2153 | struct Qdisc *q; |
2154 | int rc = -ENOMEM; | |
2155 | ||
4ec93edb YH |
2156 | /* Disable soft irqs for various locks below. Also |
2157 | * stops preemption for RCU. | |
1da177e4 | 2158 | */ |
4ec93edb | 2159 | rcu_read_lock_bh(); |
1da177e4 | 2160 | |
eae792b7 | 2161 | txq = dev_pick_tx(dev, skb); |
a898def2 | 2162 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2163 | |
1da177e4 | 2164 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2165 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 LT |
2166 | #endif |
2167 | if (q->enqueue) { | |
bbd8a0d3 | 2168 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2169 | goto out; |
1da177e4 LT |
2170 | } |
2171 | ||
2172 | /* The device has no queue. Common case for software devices: | |
2173 | loopback, all the sorts of tunnels... | |
2174 | ||
932ff279 HX |
2175 | Really, it is unlikely that netif_tx_lock protection is necessary |
2176 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2177 | counters.) |
2178 | However, it is possible, that they rely on protection | |
2179 | made by us here. | |
2180 | ||
2181 | Check this and shot the lock. It is not prone from deadlocks. | |
2182 | Either shot noqueue qdisc, it is even simpler 8) | |
2183 | */ | |
2184 | if (dev->flags & IFF_UP) { | |
2185 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2186 | ||
c773e847 | 2187 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2188 | |
c773e847 | 2189 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2190 | |
fd2ea0a7 | 2191 | if (!netif_tx_queue_stopped(txq)) { |
572a9d7b PM |
2192 | rc = dev_hard_start_xmit(skb, dev, txq); |
2193 | if (dev_xmit_complete(rc)) { | |
c773e847 | 2194 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2195 | goto out; |
2196 | } | |
2197 | } | |
c773e847 | 2198 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2199 | if (net_ratelimit()) |
2200 | printk(KERN_CRIT "Virtual device %s asks to " | |
2201 | "queue packet!\n", dev->name); | |
2202 | } else { | |
2203 | /* Recursion is detected! It is possible, | |
2204 | * unfortunately */ | |
2205 | if (net_ratelimit()) | |
2206 | printk(KERN_CRIT "Dead loop on virtual device " | |
2207 | "%s, fix it urgently!\n", dev->name); | |
2208 | } | |
2209 | } | |
2210 | ||
2211 | rc = -ENETDOWN; | |
d4828d85 | 2212 | rcu_read_unlock_bh(); |
1da177e4 | 2213 | |
1da177e4 LT |
2214 | kfree_skb(skb); |
2215 | return rc; | |
2216 | out: | |
d4828d85 | 2217 | rcu_read_unlock_bh(); |
1da177e4 LT |
2218 | return rc; |
2219 | } | |
d1b19dff | 2220 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2221 | |
2222 | ||
2223 | /*======================================================================= | |
2224 | Receiver routines | |
2225 | =======================================================================*/ | |
2226 | ||
6b2bedc3 | 2227 | int netdev_max_backlog __read_mostly = 1000; |
3b098e2d | 2228 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 SH |
2229 | int netdev_budget __read_mostly = 300; |
2230 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 | 2231 | |
eecfd7c4 ED |
2232 | /* Called with irq disabled */ |
2233 | static inline void ____napi_schedule(struct softnet_data *sd, | |
2234 | struct napi_struct *napi) | |
2235 | { | |
2236 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
2237 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2238 | } | |
2239 | ||
df334545 | 2240 | #ifdef CONFIG_RPS |
fec5e652 TH |
2241 | |
2242 | /* One global table that all flow-based protocols share. */ | |
8770acf0 | 2243 | struct rps_sock_flow_table *rps_sock_flow_table __read_mostly; |
fec5e652 TH |
2244 | EXPORT_SYMBOL(rps_sock_flow_table); |
2245 | ||
0a9627f2 TH |
2246 | /* |
2247 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2248 | * CPU from the RPS map of the receiving queue for a given skb. | |
b0e28f1e | 2249 | * rcu_read_lock must be held on entry. |
0a9627f2 | 2250 | */ |
fec5e652 TH |
2251 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, |
2252 | struct rps_dev_flow **rflowp) | |
0a9627f2 TH |
2253 | { |
2254 | struct ipv6hdr *ip6; | |
2255 | struct iphdr *ip; | |
2256 | struct netdev_rx_queue *rxqueue; | |
2257 | struct rps_map *map; | |
fec5e652 TH |
2258 | struct rps_dev_flow_table *flow_table; |
2259 | struct rps_sock_flow_table *sock_flow_table; | |
0a9627f2 TH |
2260 | int cpu = -1; |
2261 | u8 ip_proto; | |
fec5e652 | 2262 | u16 tcpu; |
8c52d509 CG |
2263 | u32 addr1, addr2, ihl; |
2264 | union { | |
2265 | u32 v32; | |
2266 | u16 v16[2]; | |
2267 | } ports; | |
0a9627f2 | 2268 | |
0a9627f2 TH |
2269 | if (skb_rx_queue_recorded(skb)) { |
2270 | u16 index = skb_get_rx_queue(skb); | |
2271 | if (unlikely(index >= dev->num_rx_queues)) { | |
08c801f8 TG |
2272 | WARN_ONCE(dev->num_rx_queues > 1, "%s received packet " |
2273 | "on queue %u, but number of RX queues is %u\n", | |
2274 | dev->name, index, dev->num_rx_queues); | |
0a9627f2 TH |
2275 | goto done; |
2276 | } | |
2277 | rxqueue = dev->_rx + index; | |
2278 | } else | |
2279 | rxqueue = dev->_rx; | |
2280 | ||
fec5e652 | 2281 | if (!rxqueue->rps_map && !rxqueue->rps_flow_table) |
0a9627f2 TH |
2282 | goto done; |
2283 | ||
2284 | if (skb->rxhash) | |
2285 | goto got_hash; /* Skip hash computation on packet header */ | |
2286 | ||
2287 | switch (skb->protocol) { | |
2288 | case __constant_htons(ETH_P_IP): | |
2289 | if (!pskb_may_pull(skb, sizeof(*ip))) | |
2290 | goto done; | |
2291 | ||
2292 | ip = (struct iphdr *) skb->data; | |
2293 | ip_proto = ip->protocol; | |
b249dcb8 ED |
2294 | addr1 = (__force u32) ip->saddr; |
2295 | addr2 = (__force u32) ip->daddr; | |
0a9627f2 TH |
2296 | ihl = ip->ihl; |
2297 | break; | |
2298 | case __constant_htons(ETH_P_IPV6): | |
2299 | if (!pskb_may_pull(skb, sizeof(*ip6))) | |
2300 | goto done; | |
2301 | ||
2302 | ip6 = (struct ipv6hdr *) skb->data; | |
2303 | ip_proto = ip6->nexthdr; | |
b249dcb8 ED |
2304 | addr1 = (__force u32) ip6->saddr.s6_addr32[3]; |
2305 | addr2 = (__force u32) ip6->daddr.s6_addr32[3]; | |
0a9627f2 TH |
2306 | ihl = (40 >> 2); |
2307 | break; | |
2308 | default: | |
2309 | goto done; | |
2310 | } | |
0a9627f2 TH |
2311 | switch (ip_proto) { |
2312 | case IPPROTO_TCP: | |
2313 | case IPPROTO_UDP: | |
2314 | case IPPROTO_DCCP: | |
2315 | case IPPROTO_ESP: | |
2316 | case IPPROTO_AH: | |
2317 | case IPPROTO_SCTP: | |
2318 | case IPPROTO_UDPLITE: | |
b249dcb8 | 2319 | if (pskb_may_pull(skb, (ihl * 4) + 4)) { |
8c52d509 CG |
2320 | ports.v32 = * (__force u32 *) (skb->data + (ihl * 4)); |
2321 | if (ports.v16[1] < ports.v16[0]) | |
2322 | swap(ports.v16[0], ports.v16[1]); | |
2323 | break; | |
b249dcb8 | 2324 | } |
0a9627f2 | 2325 | default: |
8c52d509 | 2326 | ports.v32 = 0; |
0a9627f2 TH |
2327 | break; |
2328 | } | |
2329 | ||
b249dcb8 ED |
2330 | /* get a consistent hash (same value on both flow directions) */ |
2331 | if (addr2 < addr1) | |
2332 | swap(addr1, addr2); | |
8c52d509 | 2333 | skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd); |
0a9627f2 TH |
2334 | if (!skb->rxhash) |
2335 | skb->rxhash = 1; | |
2336 | ||
2337 | got_hash: | |
fec5e652 TH |
2338 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
2339 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2340 | if (flow_table && sock_flow_table) { | |
2341 | u16 next_cpu; | |
2342 | struct rps_dev_flow *rflow; | |
2343 | ||
2344 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2345 | tcpu = rflow->cpu; | |
2346 | ||
2347 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2348 | sock_flow_table->mask]; | |
2349 | ||
2350 | /* | |
2351 | * If the desired CPU (where last recvmsg was done) is | |
2352 | * different from current CPU (one in the rx-queue flow | |
2353 | * table entry), switch if one of the following holds: | |
2354 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2355 | * - Current CPU is offline. | |
2356 | * - The current CPU's queue tail has advanced beyond the | |
2357 | * last packet that was enqueued using this table entry. | |
2358 | * This guarantees that all previous packets for the flow | |
2359 | * have been dequeued, thus preserving in order delivery. | |
2360 | */ | |
2361 | if (unlikely(tcpu != next_cpu) && | |
2362 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2363 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
2364 | rflow->last_qtail)) >= 0)) { | |
2365 | tcpu = rflow->cpu = next_cpu; | |
2366 | if (tcpu != RPS_NO_CPU) | |
2367 | rflow->last_qtail = per_cpu(softnet_data, | |
2368 | tcpu).input_queue_head; | |
2369 | } | |
2370 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { | |
2371 | *rflowp = rflow; | |
2372 | cpu = tcpu; | |
2373 | goto done; | |
2374 | } | |
2375 | } | |
2376 | ||
0a9627f2 TH |
2377 | map = rcu_dereference(rxqueue->rps_map); |
2378 | if (map) { | |
fec5e652 | 2379 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
2380 | |
2381 | if (cpu_online(tcpu)) { | |
2382 | cpu = tcpu; | |
2383 | goto done; | |
2384 | } | |
2385 | } | |
2386 | ||
2387 | done: | |
0a9627f2 TH |
2388 | return cpu; |
2389 | } | |
2390 | ||
0a9627f2 | 2391 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 2392 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 2393 | { |
e36fa2f7 ED |
2394 | struct softnet_data *sd = data; |
2395 | ||
eecfd7c4 | 2396 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 2397 | sd->received_rps++; |
0a9627f2 | 2398 | } |
e36fa2f7 | 2399 | |
fec5e652 | 2400 | #endif /* CONFIG_RPS */ |
0a9627f2 | 2401 | |
e36fa2f7 ED |
2402 | /* |
2403 | * Check if this softnet_data structure is another cpu one | |
2404 | * If yes, queue it to our IPI list and return 1 | |
2405 | * If no, return 0 | |
2406 | */ | |
2407 | static int rps_ipi_queued(struct softnet_data *sd) | |
2408 | { | |
2409 | #ifdef CONFIG_RPS | |
2410 | struct softnet_data *mysd = &__get_cpu_var(softnet_data); | |
2411 | ||
2412 | if (sd != mysd) { | |
2413 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
2414 | mysd->rps_ipi_list = sd; | |
2415 | ||
2416 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2417 | return 1; | |
2418 | } | |
2419 | #endif /* CONFIG_RPS */ | |
2420 | return 0; | |
2421 | } | |
2422 | ||
0a9627f2 TH |
2423 | /* |
2424 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2425 | * queue (may be a remote CPU queue). | |
2426 | */ | |
fec5e652 TH |
2427 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
2428 | unsigned int *qtail) | |
0a9627f2 | 2429 | { |
e36fa2f7 | 2430 | struct softnet_data *sd; |
0a9627f2 TH |
2431 | unsigned long flags; |
2432 | ||
e36fa2f7 | 2433 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
2434 | |
2435 | local_irq_save(flags); | |
0a9627f2 | 2436 | |
e36fa2f7 | 2437 | rps_lock(sd); |
6e7676c1 CG |
2438 | if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) { |
2439 | if (skb_queue_len(&sd->input_pkt_queue)) { | |
0a9627f2 | 2440 | enqueue: |
e36fa2f7 | 2441 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 2442 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 2443 | rps_unlock(sd); |
152102c7 | 2444 | local_irq_restore(flags); |
0a9627f2 TH |
2445 | return NET_RX_SUCCESS; |
2446 | } | |
2447 | ||
ebda37c2 ED |
2448 | /* Schedule NAPI for backlog device |
2449 | * We can use non atomic operation since we own the queue lock | |
2450 | */ | |
2451 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 2452 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 2453 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
2454 | } |
2455 | goto enqueue; | |
2456 | } | |
2457 | ||
dee42870 | 2458 | sd->dropped++; |
e36fa2f7 | 2459 | rps_unlock(sd); |
0a9627f2 | 2460 | |
0a9627f2 TH |
2461 | local_irq_restore(flags); |
2462 | ||
2463 | kfree_skb(skb); | |
2464 | return NET_RX_DROP; | |
2465 | } | |
1da177e4 | 2466 | |
1da177e4 LT |
2467 | /** |
2468 | * netif_rx - post buffer to the network code | |
2469 | * @skb: buffer to post | |
2470 | * | |
2471 | * This function receives a packet from a device driver and queues it for | |
2472 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2473 | * may be dropped during processing for congestion control or by the | |
2474 | * protocol layers. | |
2475 | * | |
2476 | * return values: | |
2477 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2478 | * NET_RX_DROP (packet was dropped) |
2479 | * | |
2480 | */ | |
2481 | ||
2482 | int netif_rx(struct sk_buff *skb) | |
2483 | { | |
b0e28f1e | 2484 | int ret; |
1da177e4 LT |
2485 | |
2486 | /* if netpoll wants it, pretend we never saw it */ | |
2487 | if (netpoll_rx(skb)) | |
2488 | return NET_RX_DROP; | |
2489 | ||
3b098e2d ED |
2490 | if (netdev_tstamp_prequeue) |
2491 | net_timestamp_check(skb); | |
1da177e4 | 2492 | |
df334545 | 2493 | #ifdef CONFIG_RPS |
b0e28f1e | 2494 | { |
fec5e652 | 2495 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
2496 | int cpu; |
2497 | ||
2498 | rcu_read_lock(); | |
fec5e652 TH |
2499 | |
2500 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
2501 | if (cpu < 0) |
2502 | cpu = smp_processor_id(); | |
fec5e652 TH |
2503 | |
2504 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
2505 | ||
b0e28f1e ED |
2506 | rcu_read_unlock(); |
2507 | } | |
1e94d72f | 2508 | #else |
fec5e652 TH |
2509 | { |
2510 | unsigned int qtail; | |
2511 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
2512 | put_cpu(); | |
2513 | } | |
1e94d72f | 2514 | #endif |
b0e28f1e | 2515 | return ret; |
1da177e4 | 2516 | } |
d1b19dff | 2517 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
2518 | |
2519 | int netif_rx_ni(struct sk_buff *skb) | |
2520 | { | |
2521 | int err; | |
2522 | ||
2523 | preempt_disable(); | |
2524 | err = netif_rx(skb); | |
2525 | if (local_softirq_pending()) | |
2526 | do_softirq(); | |
2527 | preempt_enable(); | |
2528 | ||
2529 | return err; | |
2530 | } | |
1da177e4 LT |
2531 | EXPORT_SYMBOL(netif_rx_ni); |
2532 | ||
1da177e4 LT |
2533 | static void net_tx_action(struct softirq_action *h) |
2534 | { | |
2535 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2536 | ||
2537 | if (sd->completion_queue) { | |
2538 | struct sk_buff *clist; | |
2539 | ||
2540 | local_irq_disable(); | |
2541 | clist = sd->completion_queue; | |
2542 | sd->completion_queue = NULL; | |
2543 | local_irq_enable(); | |
2544 | ||
2545 | while (clist) { | |
2546 | struct sk_buff *skb = clist; | |
2547 | clist = clist->next; | |
2548 | ||
547b792c | 2549 | WARN_ON(atomic_read(&skb->users)); |
1da177e4 LT |
2550 | __kfree_skb(skb); |
2551 | } | |
2552 | } | |
2553 | ||
2554 | if (sd->output_queue) { | |
37437bb2 | 2555 | struct Qdisc *head; |
1da177e4 LT |
2556 | |
2557 | local_irq_disable(); | |
2558 | head = sd->output_queue; | |
2559 | sd->output_queue = NULL; | |
a9cbd588 | 2560 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
2561 | local_irq_enable(); |
2562 | ||
2563 | while (head) { | |
37437bb2 DM |
2564 | struct Qdisc *q = head; |
2565 | spinlock_t *root_lock; | |
2566 | ||
1da177e4 LT |
2567 | head = head->next_sched; |
2568 | ||
5fb66229 | 2569 | root_lock = qdisc_lock(q); |
37437bb2 | 2570 | if (spin_trylock(root_lock)) { |
def82a1d JP |
2571 | smp_mb__before_clear_bit(); |
2572 | clear_bit(__QDISC_STATE_SCHED, | |
2573 | &q->state); | |
37437bb2 DM |
2574 | qdisc_run(q); |
2575 | spin_unlock(root_lock); | |
1da177e4 | 2576 | } else { |
195648bb | 2577 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 2578 | &q->state)) { |
195648bb | 2579 | __netif_reschedule(q); |
e8a83e10 JP |
2580 | } else { |
2581 | smp_mb__before_clear_bit(); | |
2582 | clear_bit(__QDISC_STATE_SCHED, | |
2583 | &q->state); | |
2584 | } | |
1da177e4 LT |
2585 | } |
2586 | } | |
2587 | } | |
2588 | } | |
2589 | ||
6f05f629 SH |
2590 | static inline int deliver_skb(struct sk_buff *skb, |
2591 | struct packet_type *pt_prev, | |
2592 | struct net_device *orig_dev) | |
1da177e4 LT |
2593 | { |
2594 | atomic_inc(&skb->users); | |
f2ccd8fa | 2595 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2596 | } |
2597 | ||
ab95bfe0 JP |
2598 | #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \ |
2599 | (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)) | |
da678292 MM |
2600 | /* This hook is defined here for ATM LANE */ |
2601 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
2602 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 2603 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 2604 | #endif |
1da177e4 | 2605 | |
1da177e4 LT |
2606 | #ifdef CONFIG_NET_CLS_ACT |
2607 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2608 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2609 | * a compare and 2 stores extra right now if we dont have it on | |
2610 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 2611 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
2612 | * the ingress scheduler, you just cant add policies on ingress. |
2613 | * | |
2614 | */ | |
4ec93edb | 2615 | static int ing_filter(struct sk_buff *skb) |
1da177e4 | 2616 | { |
1da177e4 | 2617 | struct net_device *dev = skb->dev; |
f697c3e8 | 2618 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
2619 | struct netdev_queue *rxq; |
2620 | int result = TC_ACT_OK; | |
2621 | struct Qdisc *q; | |
4ec93edb | 2622 | |
f697c3e8 HX |
2623 | if (MAX_RED_LOOP < ttl++) { |
2624 | printk(KERN_WARNING | |
2625 | "Redir loop detected Dropping packet (%d->%d)\n", | |
8964be4a | 2626 | skb->skb_iif, dev->ifindex); |
f697c3e8 HX |
2627 | return TC_ACT_SHOT; |
2628 | } | |
1da177e4 | 2629 | |
f697c3e8 HX |
2630 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
2631 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 2632 | |
555353cf DM |
2633 | rxq = &dev->rx_queue; |
2634 | ||
83874000 | 2635 | q = rxq->qdisc; |
8d50b53d | 2636 | if (q != &noop_qdisc) { |
83874000 | 2637 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
2638 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
2639 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
2640 | spin_unlock(qdisc_lock(q)); |
2641 | } | |
f697c3e8 HX |
2642 | |
2643 | return result; | |
2644 | } | |
86e65da9 | 2645 | |
f697c3e8 HX |
2646 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
2647 | struct packet_type **pt_prev, | |
2648 | int *ret, struct net_device *orig_dev) | |
2649 | { | |
8d50b53d | 2650 | if (skb->dev->rx_queue.qdisc == &noop_qdisc) |
f697c3e8 | 2651 | goto out; |
1da177e4 | 2652 | |
f697c3e8 HX |
2653 | if (*pt_prev) { |
2654 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2655 | *pt_prev = NULL; | |
1da177e4 LT |
2656 | } |
2657 | ||
f697c3e8 HX |
2658 | switch (ing_filter(skb)) { |
2659 | case TC_ACT_SHOT: | |
2660 | case TC_ACT_STOLEN: | |
2661 | kfree_skb(skb); | |
2662 | return NULL; | |
2663 | } | |
2664 | ||
2665 | out: | |
2666 | skb->tc_verd = 0; | |
2667 | return skb; | |
1da177e4 LT |
2668 | } |
2669 | #endif | |
2670 | ||
bc1d0411 PM |
2671 | /* |
2672 | * netif_nit_deliver - deliver received packets to network taps | |
2673 | * @skb: buffer | |
2674 | * | |
2675 | * This function is used to deliver incoming packets to network | |
2676 | * taps. It should be used when the normal netif_receive_skb path | |
2677 | * is bypassed, for example because of VLAN acceleration. | |
2678 | */ | |
2679 | void netif_nit_deliver(struct sk_buff *skb) | |
2680 | { | |
2681 | struct packet_type *ptype; | |
2682 | ||
2683 | if (list_empty(&ptype_all)) | |
2684 | return; | |
2685 | ||
2686 | skb_reset_network_header(skb); | |
2687 | skb_reset_transport_header(skb); | |
2688 | skb->mac_len = skb->network_header - skb->mac_header; | |
2689 | ||
2690 | rcu_read_lock(); | |
2691 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2692 | if (!ptype->dev || ptype->dev == skb->dev) | |
2693 | deliver_skb(skb, ptype, skb->dev); | |
2694 | } | |
2695 | rcu_read_unlock(); | |
2696 | } | |
2697 | ||
ab95bfe0 JP |
2698 | /** |
2699 | * netdev_rx_handler_register - register receive handler | |
2700 | * @dev: device to register a handler for | |
2701 | * @rx_handler: receive handler to register | |
93e2c32b | 2702 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 JP |
2703 | * |
2704 | * Register a receive hander for a device. This handler will then be | |
2705 | * called from __netif_receive_skb. A negative errno code is returned | |
2706 | * on a failure. | |
2707 | * | |
2708 | * The caller must hold the rtnl_mutex. | |
2709 | */ | |
2710 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
2711 | rx_handler_func_t *rx_handler, |
2712 | void *rx_handler_data) | |
ab95bfe0 JP |
2713 | { |
2714 | ASSERT_RTNL(); | |
2715 | ||
2716 | if (dev->rx_handler) | |
2717 | return -EBUSY; | |
2718 | ||
93e2c32b | 2719 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
2720 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
2721 | ||
2722 | return 0; | |
2723 | } | |
2724 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
2725 | ||
2726 | /** | |
2727 | * netdev_rx_handler_unregister - unregister receive handler | |
2728 | * @dev: device to unregister a handler from | |
2729 | * | |
2730 | * Unregister a receive hander from a device. | |
2731 | * | |
2732 | * The caller must hold the rtnl_mutex. | |
2733 | */ | |
2734 | void netdev_rx_handler_unregister(struct net_device *dev) | |
2735 | { | |
2736 | ||
2737 | ASSERT_RTNL(); | |
2738 | rcu_assign_pointer(dev->rx_handler, NULL); | |
93e2c32b | 2739 | rcu_assign_pointer(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
2740 | } |
2741 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
2742 | ||
acbbc071 ED |
2743 | static inline void skb_bond_set_mac_by_master(struct sk_buff *skb, |
2744 | struct net_device *master) | |
2745 | { | |
2746 | if (skb->pkt_type == PACKET_HOST) { | |
2747 | u16 *dest = (u16 *) eth_hdr(skb)->h_dest; | |
2748 | ||
2749 | memcpy(dest, master->dev_addr, ETH_ALEN); | |
2750 | } | |
2751 | } | |
2752 | ||
2753 | /* On bonding slaves other than the currently active slave, suppress | |
2754 | * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and | |
2755 | * ARP on active-backup slaves with arp_validate enabled. | |
2756 | */ | |
2757 | int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master) | |
2758 | { | |
2759 | struct net_device *dev = skb->dev; | |
2760 | ||
2761 | if (master->priv_flags & IFF_MASTER_ARPMON) | |
2762 | dev->last_rx = jiffies; | |
2763 | ||
f350a0a8 JP |
2764 | if ((master->priv_flags & IFF_MASTER_ALB) && |
2765 | (master->priv_flags & IFF_BRIDGE_PORT)) { | |
acbbc071 ED |
2766 | /* Do address unmangle. The local destination address |
2767 | * will be always the one master has. Provides the right | |
2768 | * functionality in a bridge. | |
2769 | */ | |
2770 | skb_bond_set_mac_by_master(skb, master); | |
2771 | } | |
2772 | ||
2773 | if (dev->priv_flags & IFF_SLAVE_INACTIVE) { | |
2774 | if ((dev->priv_flags & IFF_SLAVE_NEEDARP) && | |
2775 | skb->protocol == __cpu_to_be16(ETH_P_ARP)) | |
2776 | return 0; | |
2777 | ||
2778 | if (master->priv_flags & IFF_MASTER_ALB) { | |
2779 | if (skb->pkt_type != PACKET_BROADCAST && | |
2780 | skb->pkt_type != PACKET_MULTICAST) | |
2781 | return 0; | |
2782 | } | |
2783 | if (master->priv_flags & IFF_MASTER_8023AD && | |
2784 | skb->protocol == __cpu_to_be16(ETH_P_SLOW)) | |
2785 | return 0; | |
2786 | ||
2787 | return 1; | |
2788 | } | |
2789 | return 0; | |
2790 | } | |
2791 | EXPORT_SYMBOL(__skb_bond_should_drop); | |
2792 | ||
10f744d2 | 2793 | static int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
2794 | { |
2795 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 2796 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 2797 | struct net_device *orig_dev; |
0641e4fb | 2798 | struct net_device *master; |
0d7a3681 | 2799 | struct net_device *null_or_orig; |
2df4a0fa | 2800 | struct net_device *orig_or_bond; |
1da177e4 | 2801 | int ret = NET_RX_DROP; |
252e3346 | 2802 | __be16 type; |
1da177e4 | 2803 | |
3b098e2d ED |
2804 | if (!netdev_tstamp_prequeue) |
2805 | net_timestamp_check(skb); | |
81bbb3d4 | 2806 | |
05423b24 | 2807 | if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb)) |
9b22ea56 PM |
2808 | return NET_RX_SUCCESS; |
2809 | ||
1da177e4 | 2810 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 2811 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
2812 | return NET_RX_DROP; |
2813 | ||
8964be4a ED |
2814 | if (!skb->skb_iif) |
2815 | skb->skb_iif = skb->dev->ifindex; | |
86e65da9 | 2816 | |
597a264b JF |
2817 | /* |
2818 | * bonding note: skbs received on inactive slaves should only | |
2819 | * be delivered to pkt handlers that are exact matches. Also | |
2820 | * the deliver_no_wcard flag will be set. If packet handlers | |
2821 | * are sensitive to duplicate packets these skbs will need to | |
2822 | * be dropped at the handler. The vlan accel path may have | |
2823 | * already set the deliver_no_wcard flag. | |
2824 | */ | |
0d7a3681 | 2825 | null_or_orig = NULL; |
cc9bd5ce | 2826 | orig_dev = skb->dev; |
0641e4fb | 2827 | master = ACCESS_ONCE(orig_dev->master); |
597a264b JF |
2828 | if (skb->deliver_no_wcard) |
2829 | null_or_orig = orig_dev; | |
2830 | else if (master) { | |
2831 | if (skb_bond_should_drop(skb, master)) { | |
2832 | skb->deliver_no_wcard = 1; | |
0d7a3681 | 2833 | null_or_orig = orig_dev; /* deliver only exact match */ |
597a264b | 2834 | } else |
0641e4fb | 2835 | skb->dev = master; |
cc9bd5ce | 2836 | } |
8f903c70 | 2837 | |
27f39c73 | 2838 | __this_cpu_inc(softnet_data.processed); |
c1d2bbe1 | 2839 | skb_reset_network_header(skb); |
badff6d0 | 2840 | skb_reset_transport_header(skb); |
b0e380b1 | 2841 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
2842 | |
2843 | pt_prev = NULL; | |
2844 | ||
2845 | rcu_read_lock(); | |
2846 | ||
2847 | #ifdef CONFIG_NET_CLS_ACT | |
2848 | if (skb->tc_verd & TC_NCLS) { | |
2849 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2850 | goto ncls; | |
2851 | } | |
2852 | #endif | |
2853 | ||
2854 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
f982307f JE |
2855 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
2856 | ptype->dev == orig_dev) { | |
4ec93edb | 2857 | if (pt_prev) |
f2ccd8fa | 2858 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2859 | pt_prev = ptype; |
2860 | } | |
2861 | } | |
2862 | ||
2863 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
2864 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
2865 | if (!skb) | |
1da177e4 | 2866 | goto out; |
1da177e4 LT |
2867 | ncls: |
2868 | #endif | |
2869 | ||
ab95bfe0 JP |
2870 | /* Handle special case of bridge or macvlan */ |
2871 | rx_handler = rcu_dereference(skb->dev->rx_handler); | |
2872 | if (rx_handler) { | |
2873 | if (pt_prev) { | |
2874 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
2875 | pt_prev = NULL; | |
2876 | } | |
2877 | skb = rx_handler(skb); | |
2878 | if (!skb) | |
2879 | goto out; | |
2880 | } | |
1da177e4 | 2881 | |
1f3c8804 AG |
2882 | /* |
2883 | * Make sure frames received on VLAN interfaces stacked on | |
2884 | * bonding interfaces still make their way to any base bonding | |
2885 | * device that may have registered for a specific ptype. The | |
2886 | * handler may have to adjust skb->dev and orig_dev. | |
1f3c8804 | 2887 | */ |
2df4a0fa | 2888 | orig_or_bond = orig_dev; |
1f3c8804 AG |
2889 | if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) && |
2890 | (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) { | |
2df4a0fa | 2891 | orig_or_bond = vlan_dev_real_dev(skb->dev); |
1f3c8804 AG |
2892 | } |
2893 | ||
1da177e4 | 2894 | type = skb->protocol; |
82d8a867 PE |
2895 | list_for_each_entry_rcu(ptype, |
2896 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1f3c8804 | 2897 | if (ptype->type == type && (ptype->dev == null_or_orig || |
ca8d9ea3 | 2898 | ptype->dev == skb->dev || ptype->dev == orig_dev || |
2df4a0fa | 2899 | ptype->dev == orig_or_bond)) { |
4ec93edb | 2900 | if (pt_prev) |
f2ccd8fa | 2901 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2902 | pt_prev = ptype; |
2903 | } | |
2904 | } | |
2905 | ||
2906 | if (pt_prev) { | |
f2ccd8fa | 2907 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2908 | } else { |
2909 | kfree_skb(skb); | |
2910 | /* Jamal, now you will not able to escape explaining | |
2911 | * me how you were going to use this. :-) | |
2912 | */ | |
2913 | ret = NET_RX_DROP; | |
2914 | } | |
2915 | ||
2916 | out: | |
2917 | rcu_read_unlock(); | |
2918 | return ret; | |
2919 | } | |
0a9627f2 TH |
2920 | |
2921 | /** | |
2922 | * netif_receive_skb - process receive buffer from network | |
2923 | * @skb: buffer to process | |
2924 | * | |
2925 | * netif_receive_skb() is the main receive data processing function. | |
2926 | * It always succeeds. The buffer may be dropped during processing | |
2927 | * for congestion control or by the protocol layers. | |
2928 | * | |
2929 | * This function may only be called from softirq context and interrupts | |
2930 | * should be enabled. | |
2931 | * | |
2932 | * Return values (usually ignored): | |
2933 | * NET_RX_SUCCESS: no congestion | |
2934 | * NET_RX_DROP: packet was dropped | |
2935 | */ | |
2936 | int netif_receive_skb(struct sk_buff *skb) | |
2937 | { | |
3b098e2d ED |
2938 | if (netdev_tstamp_prequeue) |
2939 | net_timestamp_check(skb); | |
2940 | ||
df334545 | 2941 | #ifdef CONFIG_RPS |
3b098e2d ED |
2942 | { |
2943 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
2944 | int cpu, ret; | |
fec5e652 | 2945 | |
3b098e2d ED |
2946 | rcu_read_lock(); |
2947 | ||
2948 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
0a9627f2 | 2949 | |
3b098e2d ED |
2950 | if (cpu >= 0) { |
2951 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
2952 | rcu_read_unlock(); | |
2953 | } else { | |
2954 | rcu_read_unlock(); | |
2955 | ret = __netif_receive_skb(skb); | |
2956 | } | |
0a9627f2 | 2957 | |
3b098e2d | 2958 | return ret; |
fec5e652 | 2959 | } |
1e94d72f TH |
2960 | #else |
2961 | return __netif_receive_skb(skb); | |
2962 | #endif | |
0a9627f2 | 2963 | } |
d1b19dff | 2964 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 2965 | |
88751275 ED |
2966 | /* Network device is going away, flush any packets still pending |
2967 | * Called with irqs disabled. | |
2968 | */ | |
152102c7 | 2969 | static void flush_backlog(void *arg) |
6e583ce5 | 2970 | { |
152102c7 | 2971 | struct net_device *dev = arg; |
e36fa2f7 | 2972 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
6e583ce5 SH |
2973 | struct sk_buff *skb, *tmp; |
2974 | ||
e36fa2f7 | 2975 | rps_lock(sd); |
6e7676c1 | 2976 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
6e583ce5 | 2977 | if (skb->dev == dev) { |
e36fa2f7 | 2978 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 2979 | kfree_skb(skb); |
76cc8b13 | 2980 | input_queue_head_incr(sd); |
6e583ce5 | 2981 | } |
6e7676c1 | 2982 | } |
e36fa2f7 | 2983 | rps_unlock(sd); |
6e7676c1 CG |
2984 | |
2985 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
2986 | if (skb->dev == dev) { | |
2987 | __skb_unlink(skb, &sd->process_queue); | |
2988 | kfree_skb(skb); | |
76cc8b13 | 2989 | input_queue_head_incr(sd); |
6e7676c1 CG |
2990 | } |
2991 | } | |
6e583ce5 SH |
2992 | } |
2993 | ||
d565b0a1 HX |
2994 | static int napi_gro_complete(struct sk_buff *skb) |
2995 | { | |
2996 | struct packet_type *ptype; | |
2997 | __be16 type = skb->protocol; | |
2998 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
2999 | int err = -ENOENT; | |
3000 | ||
fc59f9a3 HX |
3001 | if (NAPI_GRO_CB(skb)->count == 1) { |
3002 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 3003 | goto out; |
fc59f9a3 | 3004 | } |
d565b0a1 HX |
3005 | |
3006 | rcu_read_lock(); | |
3007 | list_for_each_entry_rcu(ptype, head, list) { | |
3008 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
3009 | continue; | |
3010 | ||
3011 | err = ptype->gro_complete(skb); | |
3012 | break; | |
3013 | } | |
3014 | rcu_read_unlock(); | |
3015 | ||
3016 | if (err) { | |
3017 | WARN_ON(&ptype->list == head); | |
3018 | kfree_skb(skb); | |
3019 | return NET_RX_SUCCESS; | |
3020 | } | |
3021 | ||
3022 | out: | |
d565b0a1 HX |
3023 | return netif_receive_skb(skb); |
3024 | } | |
3025 | ||
11380a4b | 3026 | static void napi_gro_flush(struct napi_struct *napi) |
d565b0a1 HX |
3027 | { |
3028 | struct sk_buff *skb, *next; | |
3029 | ||
3030 | for (skb = napi->gro_list; skb; skb = next) { | |
3031 | next = skb->next; | |
3032 | skb->next = NULL; | |
3033 | napi_gro_complete(skb); | |
3034 | } | |
3035 | ||
4ae5544f | 3036 | napi->gro_count = 0; |
d565b0a1 HX |
3037 | napi->gro_list = NULL; |
3038 | } | |
d565b0a1 | 3039 | |
5b252f0c | 3040 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
3041 | { |
3042 | struct sk_buff **pp = NULL; | |
3043 | struct packet_type *ptype; | |
3044 | __be16 type = skb->protocol; | |
3045 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 3046 | int same_flow; |
d565b0a1 | 3047 | int mac_len; |
5b252f0c | 3048 | enum gro_result ret; |
d565b0a1 HX |
3049 | |
3050 | if (!(skb->dev->features & NETIF_F_GRO)) | |
3051 | goto normal; | |
3052 | ||
4cf704fb | 3053 | if (skb_is_gso(skb) || skb_has_frags(skb)) |
f17f5c91 HX |
3054 | goto normal; |
3055 | ||
d565b0a1 HX |
3056 | rcu_read_lock(); |
3057 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
3058 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
3059 | continue; | |
3060 | ||
86911732 | 3061 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
3062 | mac_len = skb->network_header - skb->mac_header; |
3063 | skb->mac_len = mac_len; | |
3064 | NAPI_GRO_CB(skb)->same_flow = 0; | |
3065 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3066 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3067 | |
d565b0a1 HX |
3068 | pp = ptype->gro_receive(&napi->gro_list, skb); |
3069 | break; | |
3070 | } | |
3071 | rcu_read_unlock(); | |
3072 | ||
3073 | if (&ptype->list == head) | |
3074 | goto normal; | |
3075 | ||
0da2afd5 | 3076 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3077 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3078 | |
d565b0a1 HX |
3079 | if (pp) { |
3080 | struct sk_buff *nskb = *pp; | |
3081 | ||
3082 | *pp = nskb->next; | |
3083 | nskb->next = NULL; | |
3084 | napi_gro_complete(nskb); | |
4ae5544f | 3085 | napi->gro_count--; |
d565b0a1 HX |
3086 | } |
3087 | ||
0da2afd5 | 3088 | if (same_flow) |
d565b0a1 HX |
3089 | goto ok; |
3090 | ||
4ae5544f | 3091 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3092 | goto normal; |
d565b0a1 | 3093 | |
4ae5544f | 3094 | napi->gro_count++; |
d565b0a1 | 3095 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 3096 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3097 | skb->next = napi->gro_list; |
3098 | napi->gro_list = skb; | |
5d0d9be8 | 3099 | ret = GRO_HELD; |
d565b0a1 | 3100 | |
ad0f9904 | 3101 | pull: |
cb18978c HX |
3102 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3103 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3104 | ||
3105 | BUG_ON(skb->end - skb->tail < grow); | |
3106 | ||
3107 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3108 | ||
3109 | skb->tail += grow; | |
3110 | skb->data_len -= grow; | |
3111 | ||
3112 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
3113 | skb_shinfo(skb)->frags[0].size -= grow; | |
3114 | ||
3115 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) { | |
3116 | put_page(skb_shinfo(skb)->frags[0].page); | |
3117 | memmove(skb_shinfo(skb)->frags, | |
3118 | skb_shinfo(skb)->frags + 1, | |
3119 | --skb_shinfo(skb)->nr_frags); | |
3120 | } | |
ad0f9904 HX |
3121 | } |
3122 | ||
d565b0a1 | 3123 | ok: |
5d0d9be8 | 3124 | return ret; |
d565b0a1 HX |
3125 | |
3126 | normal: | |
ad0f9904 HX |
3127 | ret = GRO_NORMAL; |
3128 | goto pull; | |
5d38a079 | 3129 | } |
96e93eab HX |
3130 | EXPORT_SYMBOL(dev_gro_receive); |
3131 | ||
5b252f0c BH |
3132 | static gro_result_t |
3133 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
96e93eab HX |
3134 | { |
3135 | struct sk_buff *p; | |
3136 | ||
d1c76af9 HX |
3137 | if (netpoll_rx_on(skb)) |
3138 | return GRO_NORMAL; | |
3139 | ||
96e93eab | 3140 | for (p = napi->gro_list; p; p = p->next) { |
f64f9e71 JP |
3141 | NAPI_GRO_CB(p)->same_flow = |
3142 | (p->dev == skb->dev) && | |
3143 | !compare_ether_header(skb_mac_header(p), | |
3144 | skb_gro_mac_header(skb)); | |
96e93eab HX |
3145 | NAPI_GRO_CB(p)->flush = 0; |
3146 | } | |
3147 | ||
3148 | return dev_gro_receive(napi, skb); | |
3149 | } | |
5d38a079 | 3150 | |
c7c4b3b6 | 3151 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3152 | { |
5d0d9be8 HX |
3153 | switch (ret) { |
3154 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3155 | if (netif_receive_skb(skb)) |
3156 | ret = GRO_DROP; | |
3157 | break; | |
5d38a079 | 3158 | |
5d0d9be8 | 3159 | case GRO_DROP: |
5d0d9be8 | 3160 | case GRO_MERGED_FREE: |
5d38a079 HX |
3161 | kfree_skb(skb); |
3162 | break; | |
5b252f0c BH |
3163 | |
3164 | case GRO_HELD: | |
3165 | case GRO_MERGED: | |
3166 | break; | |
5d38a079 HX |
3167 | } |
3168 | ||
c7c4b3b6 | 3169 | return ret; |
5d0d9be8 HX |
3170 | } |
3171 | EXPORT_SYMBOL(napi_skb_finish); | |
3172 | ||
78a478d0 HX |
3173 | void skb_gro_reset_offset(struct sk_buff *skb) |
3174 | { | |
3175 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3176 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3177 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3178 | |
78d3fd0b | 3179 | if (skb->mac_header == skb->tail && |
7489594c | 3180 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) { |
78a478d0 HX |
3181 | NAPI_GRO_CB(skb)->frag0 = |
3182 | page_address(skb_shinfo(skb)->frags[0].page) + | |
3183 | skb_shinfo(skb)->frags[0].page_offset; | |
7489594c HX |
3184 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size; |
3185 | } | |
78a478d0 HX |
3186 | } |
3187 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
3188 | ||
c7c4b3b6 | 3189 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3190 | { |
86911732 HX |
3191 | skb_gro_reset_offset(skb); |
3192 | ||
5d0d9be8 | 3193 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3194 | } |
3195 | EXPORT_SYMBOL(napi_gro_receive); | |
3196 | ||
96e93eab HX |
3197 | void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
3198 | { | |
96e93eab HX |
3199 | __skb_pull(skb, skb_headlen(skb)); |
3200 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
3201 | ||
3202 | napi->skb = skb; | |
3203 | } | |
3204 | EXPORT_SYMBOL(napi_reuse_skb); | |
3205 | ||
76620aaf | 3206 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3207 | { |
5d38a079 | 3208 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3209 | |
3210 | if (!skb) { | |
89d71a66 ED |
3211 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3212 | if (skb) | |
3213 | napi->skb = skb; | |
80595d59 | 3214 | } |
96e93eab HX |
3215 | return skb; |
3216 | } | |
76620aaf | 3217 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3218 | |
c7c4b3b6 BH |
3219 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
3220 | gro_result_t ret) | |
96e93eab | 3221 | { |
5d0d9be8 HX |
3222 | switch (ret) { |
3223 | case GRO_NORMAL: | |
86911732 | 3224 | case GRO_HELD: |
e76b69cc | 3225 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3226 | |
c7c4b3b6 BH |
3227 | if (ret == GRO_HELD) |
3228 | skb_gro_pull(skb, -ETH_HLEN); | |
3229 | else if (netif_receive_skb(skb)) | |
3230 | ret = GRO_DROP; | |
86911732 | 3231 | break; |
5d38a079 | 3232 | |
5d0d9be8 | 3233 | case GRO_DROP: |
5d0d9be8 HX |
3234 | case GRO_MERGED_FREE: |
3235 | napi_reuse_skb(napi, skb); | |
3236 | break; | |
5b252f0c BH |
3237 | |
3238 | case GRO_MERGED: | |
3239 | break; | |
5d0d9be8 | 3240 | } |
5d38a079 | 3241 | |
c7c4b3b6 | 3242 | return ret; |
5d38a079 | 3243 | } |
5d0d9be8 HX |
3244 | EXPORT_SYMBOL(napi_frags_finish); |
3245 | ||
76620aaf HX |
3246 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
3247 | { | |
3248 | struct sk_buff *skb = napi->skb; | |
3249 | struct ethhdr *eth; | |
a5b1cf28 HX |
3250 | unsigned int hlen; |
3251 | unsigned int off; | |
76620aaf HX |
3252 | |
3253 | napi->skb = NULL; | |
3254 | ||
3255 | skb_reset_mac_header(skb); | |
3256 | skb_gro_reset_offset(skb); | |
3257 | ||
a5b1cf28 HX |
3258 | off = skb_gro_offset(skb); |
3259 | hlen = off + sizeof(*eth); | |
3260 | eth = skb_gro_header_fast(skb, off); | |
3261 | if (skb_gro_header_hard(skb, hlen)) { | |
3262 | eth = skb_gro_header_slow(skb, hlen, off); | |
3263 | if (unlikely(!eth)) { | |
3264 | napi_reuse_skb(napi, skb); | |
3265 | skb = NULL; | |
3266 | goto out; | |
3267 | } | |
76620aaf HX |
3268 | } |
3269 | ||
3270 | skb_gro_pull(skb, sizeof(*eth)); | |
3271 | ||
3272 | /* | |
3273 | * This works because the only protocols we care about don't require | |
3274 | * special handling. We'll fix it up properly at the end. | |
3275 | */ | |
3276 | skb->protocol = eth->h_proto; | |
3277 | ||
3278 | out: | |
3279 | return skb; | |
3280 | } | |
3281 | EXPORT_SYMBOL(napi_frags_skb); | |
3282 | ||
c7c4b3b6 | 3283 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3284 | { |
76620aaf | 3285 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3286 | |
3287 | if (!skb) | |
c7c4b3b6 | 3288 | return GRO_DROP; |
5d0d9be8 HX |
3289 | |
3290 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3291 | } | |
5d38a079 HX |
3292 | EXPORT_SYMBOL(napi_gro_frags); |
3293 | ||
e326bed2 ED |
3294 | /* |
3295 | * net_rps_action sends any pending IPI's for rps. | |
3296 | * Note: called with local irq disabled, but exits with local irq enabled. | |
3297 | */ | |
3298 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
3299 | { | |
3300 | #ifdef CONFIG_RPS | |
3301 | struct softnet_data *remsd = sd->rps_ipi_list; | |
3302 | ||
3303 | if (remsd) { | |
3304 | sd->rps_ipi_list = NULL; | |
3305 | ||
3306 | local_irq_enable(); | |
3307 | ||
3308 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3309 | while (remsd) { | |
3310 | struct softnet_data *next = remsd->rps_ipi_next; | |
3311 | ||
3312 | if (cpu_online(remsd->cpu)) | |
3313 | __smp_call_function_single(remsd->cpu, | |
3314 | &remsd->csd, 0); | |
3315 | remsd = next; | |
3316 | } | |
3317 | } else | |
3318 | #endif | |
3319 | local_irq_enable(); | |
3320 | } | |
3321 | ||
bea3348e | 3322 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3323 | { |
3324 | int work = 0; | |
eecfd7c4 | 3325 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
1da177e4 | 3326 | |
e326bed2 ED |
3327 | #ifdef CONFIG_RPS |
3328 | /* Check if we have pending ipi, its better to send them now, | |
3329 | * not waiting net_rx_action() end. | |
3330 | */ | |
3331 | if (sd->rps_ipi_list) { | |
3332 | local_irq_disable(); | |
3333 | net_rps_action_and_irq_enable(sd); | |
3334 | } | |
3335 | #endif | |
bea3348e | 3336 | napi->weight = weight_p; |
6e7676c1 CG |
3337 | local_irq_disable(); |
3338 | while (work < quota) { | |
1da177e4 | 3339 | struct sk_buff *skb; |
6e7676c1 CG |
3340 | unsigned int qlen; |
3341 | ||
3342 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
3343 | local_irq_enable(); | |
3344 | __netif_receive_skb(skb); | |
6e7676c1 | 3345 | local_irq_disable(); |
76cc8b13 TH |
3346 | input_queue_head_incr(sd); |
3347 | if (++work >= quota) { | |
3348 | local_irq_enable(); | |
3349 | return work; | |
3350 | } | |
6e7676c1 | 3351 | } |
1da177e4 | 3352 | |
e36fa2f7 | 3353 | rps_lock(sd); |
6e7676c1 | 3354 | qlen = skb_queue_len(&sd->input_pkt_queue); |
76cc8b13 | 3355 | if (qlen) |
6e7676c1 CG |
3356 | skb_queue_splice_tail_init(&sd->input_pkt_queue, |
3357 | &sd->process_queue); | |
76cc8b13 | 3358 | |
6e7676c1 | 3359 | if (qlen < quota - work) { |
eecfd7c4 ED |
3360 | /* |
3361 | * Inline a custom version of __napi_complete(). | |
3362 | * only current cpu owns and manipulates this napi, | |
3363 | * and NAPI_STATE_SCHED is the only possible flag set on backlog. | |
3364 | * we can use a plain write instead of clear_bit(), | |
3365 | * and we dont need an smp_mb() memory barrier. | |
3366 | */ | |
3367 | list_del(&napi->poll_list); | |
3368 | napi->state = 0; | |
3369 | ||
6e7676c1 | 3370 | quota = work + qlen; |
bea3348e | 3371 | } |
e36fa2f7 | 3372 | rps_unlock(sd); |
6e7676c1 CG |
3373 | } |
3374 | local_irq_enable(); | |
1da177e4 | 3375 | |
bea3348e SH |
3376 | return work; |
3377 | } | |
1da177e4 | 3378 | |
bea3348e SH |
3379 | /** |
3380 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 3381 | * @n: entry to schedule |
bea3348e SH |
3382 | * |
3383 | * The entry's receive function will be scheduled to run | |
3384 | */ | |
b5606c2d | 3385 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
3386 | { |
3387 | unsigned long flags; | |
1da177e4 | 3388 | |
bea3348e | 3389 | local_irq_save(flags); |
eecfd7c4 | 3390 | ____napi_schedule(&__get_cpu_var(softnet_data), n); |
bea3348e | 3391 | local_irq_restore(flags); |
1da177e4 | 3392 | } |
bea3348e SH |
3393 | EXPORT_SYMBOL(__napi_schedule); |
3394 | ||
d565b0a1 HX |
3395 | void __napi_complete(struct napi_struct *n) |
3396 | { | |
3397 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3398 | BUG_ON(n->gro_list); | |
3399 | ||
3400 | list_del(&n->poll_list); | |
3401 | smp_mb__before_clear_bit(); | |
3402 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3403 | } | |
3404 | EXPORT_SYMBOL(__napi_complete); | |
3405 | ||
3406 | void napi_complete(struct napi_struct *n) | |
3407 | { | |
3408 | unsigned long flags; | |
3409 | ||
3410 | /* | |
3411 | * don't let napi dequeue from the cpu poll list | |
3412 | * just in case its running on a different cpu | |
3413 | */ | |
3414 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3415 | return; | |
3416 | ||
3417 | napi_gro_flush(n); | |
3418 | local_irq_save(flags); | |
3419 | __napi_complete(n); | |
3420 | local_irq_restore(flags); | |
3421 | } | |
3422 | EXPORT_SYMBOL(napi_complete); | |
3423 | ||
3424 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3425 | int (*poll)(struct napi_struct *, int), int weight) | |
3426 | { | |
3427 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 3428 | napi->gro_count = 0; |
d565b0a1 | 3429 | napi->gro_list = NULL; |
5d38a079 | 3430 | napi->skb = NULL; |
d565b0a1 HX |
3431 | napi->poll = poll; |
3432 | napi->weight = weight; | |
3433 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 3434 | napi->dev = dev; |
5d38a079 | 3435 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
3436 | spin_lock_init(&napi->poll_lock); |
3437 | napi->poll_owner = -1; | |
3438 | #endif | |
3439 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3440 | } | |
3441 | EXPORT_SYMBOL(netif_napi_add); | |
3442 | ||
3443 | void netif_napi_del(struct napi_struct *napi) | |
3444 | { | |
3445 | struct sk_buff *skb, *next; | |
3446 | ||
d7b06636 | 3447 | list_del_init(&napi->dev_list); |
76620aaf | 3448 | napi_free_frags(napi); |
d565b0a1 HX |
3449 | |
3450 | for (skb = napi->gro_list; skb; skb = next) { | |
3451 | next = skb->next; | |
3452 | skb->next = NULL; | |
3453 | kfree_skb(skb); | |
3454 | } | |
3455 | ||
3456 | napi->gro_list = NULL; | |
4ae5544f | 3457 | napi->gro_count = 0; |
d565b0a1 HX |
3458 | } |
3459 | EXPORT_SYMBOL(netif_napi_del); | |
3460 | ||
1da177e4 LT |
3461 | static void net_rx_action(struct softirq_action *h) |
3462 | { | |
e326bed2 | 3463 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
24f8b238 | 3464 | unsigned long time_limit = jiffies + 2; |
51b0bded | 3465 | int budget = netdev_budget; |
53fb95d3 MM |
3466 | void *have; |
3467 | ||
1da177e4 LT |
3468 | local_irq_disable(); |
3469 | ||
e326bed2 | 3470 | while (!list_empty(&sd->poll_list)) { |
bea3348e SH |
3471 | struct napi_struct *n; |
3472 | int work, weight; | |
1da177e4 | 3473 | |
bea3348e | 3474 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
3475 | * Allow this to run for 2 jiffies since which will allow |
3476 | * an average latency of 1.5/HZ. | |
bea3348e | 3477 | */ |
24f8b238 | 3478 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
3479 | goto softnet_break; |
3480 | ||
3481 | local_irq_enable(); | |
3482 | ||
bea3348e SH |
3483 | /* Even though interrupts have been re-enabled, this |
3484 | * access is safe because interrupts can only add new | |
3485 | * entries to the tail of this list, and only ->poll() | |
3486 | * calls can remove this head entry from the list. | |
3487 | */ | |
e326bed2 | 3488 | n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list); |
1da177e4 | 3489 | |
bea3348e SH |
3490 | have = netpoll_poll_lock(n); |
3491 | ||
3492 | weight = n->weight; | |
3493 | ||
0a7606c1 DM |
3494 | /* This NAPI_STATE_SCHED test is for avoiding a race |
3495 | * with netpoll's poll_napi(). Only the entity which | |
3496 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3497 | * actually make the ->poll() call. Therefore we avoid | |
3498 | * accidently calling ->poll() when NAPI is not scheduled. | |
3499 | */ | |
3500 | work = 0; | |
4ea7e386 | 3501 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3502 | work = n->poll(n, weight); |
4ea7e386 NH |
3503 | trace_napi_poll(n); |
3504 | } | |
bea3348e SH |
3505 | |
3506 | WARN_ON_ONCE(work > weight); | |
3507 | ||
3508 | budget -= work; | |
3509 | ||
3510 | local_irq_disable(); | |
3511 | ||
3512 | /* Drivers must not modify the NAPI state if they | |
3513 | * consume the entire weight. In such cases this code | |
3514 | * still "owns" the NAPI instance and therefore can | |
3515 | * move the instance around on the list at-will. | |
3516 | */ | |
fed17f30 | 3517 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3518 | if (unlikely(napi_disable_pending(n))) { |
3519 | local_irq_enable(); | |
3520 | napi_complete(n); | |
3521 | local_irq_disable(); | |
3522 | } else | |
e326bed2 | 3523 | list_move_tail(&n->poll_list, &sd->poll_list); |
fed17f30 | 3524 | } |
bea3348e SH |
3525 | |
3526 | netpoll_poll_unlock(have); | |
1da177e4 LT |
3527 | } |
3528 | out: | |
e326bed2 | 3529 | net_rps_action_and_irq_enable(sd); |
0a9627f2 | 3530 | |
db217334 CL |
3531 | #ifdef CONFIG_NET_DMA |
3532 | /* | |
3533 | * There may not be any more sk_buffs coming right now, so push | |
3534 | * any pending DMA copies to hardware | |
3535 | */ | |
2ba05622 | 3536 | dma_issue_pending_all(); |
db217334 | 3537 | #endif |
bea3348e | 3538 | |
1da177e4 LT |
3539 | return; |
3540 | ||
3541 | softnet_break: | |
dee42870 | 3542 | sd->time_squeeze++; |
1da177e4 LT |
3543 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
3544 | goto out; | |
3545 | } | |
3546 | ||
d1b19dff | 3547 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
3548 | |
3549 | /** | |
3550 | * register_gifconf - register a SIOCGIF handler | |
3551 | * @family: Address family | |
3552 | * @gifconf: Function handler | |
3553 | * | |
3554 | * Register protocol dependent address dumping routines. The handler | |
3555 | * that is passed must not be freed or reused until it has been replaced | |
3556 | * by another handler. | |
3557 | */ | |
d1b19dff | 3558 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
3559 | { |
3560 | if (family >= NPROTO) | |
3561 | return -EINVAL; | |
3562 | gifconf_list[family] = gifconf; | |
3563 | return 0; | |
3564 | } | |
d1b19dff | 3565 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
3566 | |
3567 | ||
3568 | /* | |
3569 | * Map an interface index to its name (SIOCGIFNAME) | |
3570 | */ | |
3571 | ||
3572 | /* | |
3573 | * We need this ioctl for efficient implementation of the | |
3574 | * if_indextoname() function required by the IPv6 API. Without | |
3575 | * it, we would have to search all the interfaces to find a | |
3576 | * match. --pb | |
3577 | */ | |
3578 | ||
881d966b | 3579 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
3580 | { |
3581 | struct net_device *dev; | |
3582 | struct ifreq ifr; | |
3583 | ||
3584 | /* | |
3585 | * Fetch the caller's info block. | |
3586 | */ | |
3587 | ||
3588 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3589 | return -EFAULT; | |
3590 | ||
fb699dfd ED |
3591 | rcu_read_lock(); |
3592 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 3593 | if (!dev) { |
fb699dfd | 3594 | rcu_read_unlock(); |
1da177e4 LT |
3595 | return -ENODEV; |
3596 | } | |
3597 | ||
3598 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 3599 | rcu_read_unlock(); |
1da177e4 LT |
3600 | |
3601 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3602 | return -EFAULT; | |
3603 | return 0; | |
3604 | } | |
3605 | ||
3606 | /* | |
3607 | * Perform a SIOCGIFCONF call. This structure will change | |
3608 | * size eventually, and there is nothing I can do about it. | |
3609 | * Thus we will need a 'compatibility mode'. | |
3610 | */ | |
3611 | ||
881d966b | 3612 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
3613 | { |
3614 | struct ifconf ifc; | |
3615 | struct net_device *dev; | |
3616 | char __user *pos; | |
3617 | int len; | |
3618 | int total; | |
3619 | int i; | |
3620 | ||
3621 | /* | |
3622 | * Fetch the caller's info block. | |
3623 | */ | |
3624 | ||
3625 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3626 | return -EFAULT; | |
3627 | ||
3628 | pos = ifc.ifc_buf; | |
3629 | len = ifc.ifc_len; | |
3630 | ||
3631 | /* | |
3632 | * Loop over the interfaces, and write an info block for each. | |
3633 | */ | |
3634 | ||
3635 | total = 0; | |
881d966b | 3636 | for_each_netdev(net, dev) { |
1da177e4 LT |
3637 | for (i = 0; i < NPROTO; i++) { |
3638 | if (gifconf_list[i]) { | |
3639 | int done; | |
3640 | if (!pos) | |
3641 | done = gifconf_list[i](dev, NULL, 0); | |
3642 | else | |
3643 | done = gifconf_list[i](dev, pos + total, | |
3644 | len - total); | |
3645 | if (done < 0) | |
3646 | return -EFAULT; | |
3647 | total += done; | |
3648 | } | |
3649 | } | |
4ec93edb | 3650 | } |
1da177e4 LT |
3651 | |
3652 | /* | |
3653 | * All done. Write the updated control block back to the caller. | |
3654 | */ | |
3655 | ifc.ifc_len = total; | |
3656 | ||
3657 | /* | |
3658 | * Both BSD and Solaris return 0 here, so we do too. | |
3659 | */ | |
3660 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
3661 | } | |
3662 | ||
3663 | #ifdef CONFIG_PROC_FS | |
3664 | /* | |
3665 | * This is invoked by the /proc filesystem handler to display a device | |
3666 | * in detail. | |
3667 | */ | |
7562f876 | 3668 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 3669 | __acquires(RCU) |
1da177e4 | 3670 | { |
e372c414 | 3671 | struct net *net = seq_file_net(seq); |
7562f876 | 3672 | loff_t off; |
1da177e4 | 3673 | struct net_device *dev; |
1da177e4 | 3674 | |
c6d14c84 | 3675 | rcu_read_lock(); |
7562f876 PE |
3676 | if (!*pos) |
3677 | return SEQ_START_TOKEN; | |
1da177e4 | 3678 | |
7562f876 | 3679 | off = 1; |
c6d14c84 | 3680 | for_each_netdev_rcu(net, dev) |
7562f876 PE |
3681 | if (off++ == *pos) |
3682 | return dev; | |
1da177e4 | 3683 | |
7562f876 | 3684 | return NULL; |
1da177e4 LT |
3685 | } |
3686 | ||
3687 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3688 | { | |
c6d14c84 ED |
3689 | struct net_device *dev = (v == SEQ_START_TOKEN) ? |
3690 | first_net_device(seq_file_net(seq)) : | |
3691 | next_net_device((struct net_device *)v); | |
3692 | ||
1da177e4 | 3693 | ++*pos; |
c6d14c84 | 3694 | return rcu_dereference(dev); |
1da177e4 LT |
3695 | } |
3696 | ||
3697 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 3698 | __releases(RCU) |
1da177e4 | 3699 | { |
c6d14c84 | 3700 | rcu_read_unlock(); |
1da177e4 LT |
3701 | } |
3702 | ||
3703 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
3704 | { | |
be1f3c2c | 3705 | const struct rtnl_link_stats64 *stats = dev_get_stats(dev); |
1da177e4 | 3706 | |
be1f3c2c BH |
3707 | seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu " |
3708 | "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n", | |
5a1b5898 RR |
3709 | dev->name, stats->rx_bytes, stats->rx_packets, |
3710 | stats->rx_errors, | |
3711 | stats->rx_dropped + stats->rx_missed_errors, | |
3712 | stats->rx_fifo_errors, | |
3713 | stats->rx_length_errors + stats->rx_over_errors + | |
3714 | stats->rx_crc_errors + stats->rx_frame_errors, | |
3715 | stats->rx_compressed, stats->multicast, | |
3716 | stats->tx_bytes, stats->tx_packets, | |
3717 | stats->tx_errors, stats->tx_dropped, | |
3718 | stats->tx_fifo_errors, stats->collisions, | |
3719 | stats->tx_carrier_errors + | |
3720 | stats->tx_aborted_errors + | |
3721 | stats->tx_window_errors + | |
3722 | stats->tx_heartbeat_errors, | |
3723 | stats->tx_compressed); | |
1da177e4 LT |
3724 | } |
3725 | ||
3726 | /* | |
3727 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
3728 | * /proc/net interface to create /proc/net/dev | |
3729 | */ | |
3730 | static int dev_seq_show(struct seq_file *seq, void *v) | |
3731 | { | |
3732 | if (v == SEQ_START_TOKEN) | |
3733 | seq_puts(seq, "Inter-| Receive " | |
3734 | " | Transmit\n" | |
3735 | " face |bytes packets errs drop fifo frame " | |
3736 | "compressed multicast|bytes packets errs " | |
3737 | "drop fifo colls carrier compressed\n"); | |
3738 | else | |
3739 | dev_seq_printf_stats(seq, v); | |
3740 | return 0; | |
3741 | } | |
3742 | ||
dee42870 | 3743 | static struct softnet_data *softnet_get_online(loff_t *pos) |
1da177e4 | 3744 | { |
dee42870 | 3745 | struct softnet_data *sd = NULL; |
1da177e4 | 3746 | |
0c0b0aca | 3747 | while (*pos < nr_cpu_ids) |
4ec93edb | 3748 | if (cpu_online(*pos)) { |
dee42870 | 3749 | sd = &per_cpu(softnet_data, *pos); |
1da177e4 LT |
3750 | break; |
3751 | } else | |
3752 | ++*pos; | |
dee42870 | 3753 | return sd; |
1da177e4 LT |
3754 | } |
3755 | ||
3756 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
3757 | { | |
3758 | return softnet_get_online(pos); | |
3759 | } | |
3760 | ||
3761 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3762 | { | |
3763 | ++*pos; | |
3764 | return softnet_get_online(pos); | |
3765 | } | |
3766 | ||
3767 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
3768 | { | |
3769 | } | |
3770 | ||
3771 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
3772 | { | |
dee42870 | 3773 | struct softnet_data *sd = v; |
1da177e4 | 3774 | |
0a9627f2 | 3775 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
dee42870 | 3776 | sd->processed, sd->dropped, sd->time_squeeze, 0, |
c1ebcdb8 | 3777 | 0, 0, 0, 0, /* was fastroute */ |
dee42870 | 3778 | sd->cpu_collision, sd->received_rps); |
1da177e4 LT |
3779 | return 0; |
3780 | } | |
3781 | ||
f690808e | 3782 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
3783 | .start = dev_seq_start, |
3784 | .next = dev_seq_next, | |
3785 | .stop = dev_seq_stop, | |
3786 | .show = dev_seq_show, | |
3787 | }; | |
3788 | ||
3789 | static int dev_seq_open(struct inode *inode, struct file *file) | |
3790 | { | |
e372c414 DL |
3791 | return seq_open_net(inode, file, &dev_seq_ops, |
3792 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
3793 | } |
3794 | ||
9a32144e | 3795 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
3796 | .owner = THIS_MODULE, |
3797 | .open = dev_seq_open, | |
3798 | .read = seq_read, | |
3799 | .llseek = seq_lseek, | |
e372c414 | 3800 | .release = seq_release_net, |
1da177e4 LT |
3801 | }; |
3802 | ||
f690808e | 3803 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
3804 | .start = softnet_seq_start, |
3805 | .next = softnet_seq_next, | |
3806 | .stop = softnet_seq_stop, | |
3807 | .show = softnet_seq_show, | |
3808 | }; | |
3809 | ||
3810 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
3811 | { | |
3812 | return seq_open(file, &softnet_seq_ops); | |
3813 | } | |
3814 | ||
9a32144e | 3815 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
3816 | .owner = THIS_MODULE, |
3817 | .open = softnet_seq_open, | |
3818 | .read = seq_read, | |
3819 | .llseek = seq_lseek, | |
3820 | .release = seq_release, | |
3821 | }; | |
3822 | ||
0e1256ff SH |
3823 | static void *ptype_get_idx(loff_t pos) |
3824 | { | |
3825 | struct packet_type *pt = NULL; | |
3826 | loff_t i = 0; | |
3827 | int t; | |
3828 | ||
3829 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
3830 | if (i == pos) | |
3831 | return pt; | |
3832 | ++i; | |
3833 | } | |
3834 | ||
82d8a867 | 3835 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
3836 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
3837 | if (i == pos) | |
3838 | return pt; | |
3839 | ++i; | |
3840 | } | |
3841 | } | |
3842 | return NULL; | |
3843 | } | |
3844 | ||
3845 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 3846 | __acquires(RCU) |
0e1256ff SH |
3847 | { |
3848 | rcu_read_lock(); | |
3849 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
3850 | } | |
3851 | ||
3852 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3853 | { | |
3854 | struct packet_type *pt; | |
3855 | struct list_head *nxt; | |
3856 | int hash; | |
3857 | ||
3858 | ++*pos; | |
3859 | if (v == SEQ_START_TOKEN) | |
3860 | return ptype_get_idx(0); | |
3861 | ||
3862 | pt = v; | |
3863 | nxt = pt->list.next; | |
3864 | if (pt->type == htons(ETH_P_ALL)) { | |
3865 | if (nxt != &ptype_all) | |
3866 | goto found; | |
3867 | hash = 0; | |
3868 | nxt = ptype_base[0].next; | |
3869 | } else | |
82d8a867 | 3870 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
3871 | |
3872 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 3873 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
3874 | return NULL; |
3875 | nxt = ptype_base[hash].next; | |
3876 | } | |
3877 | found: | |
3878 | return list_entry(nxt, struct packet_type, list); | |
3879 | } | |
3880 | ||
3881 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 3882 | __releases(RCU) |
0e1256ff SH |
3883 | { |
3884 | rcu_read_unlock(); | |
3885 | } | |
3886 | ||
0e1256ff SH |
3887 | static int ptype_seq_show(struct seq_file *seq, void *v) |
3888 | { | |
3889 | struct packet_type *pt = v; | |
3890 | ||
3891 | if (v == SEQ_START_TOKEN) | |
3892 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 3893 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
3894 | if (pt->type == htons(ETH_P_ALL)) |
3895 | seq_puts(seq, "ALL "); | |
3896 | else | |
3897 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
3898 | ||
908cd2da AD |
3899 | seq_printf(seq, " %-8s %pF\n", |
3900 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
3901 | } |
3902 | ||
3903 | return 0; | |
3904 | } | |
3905 | ||
3906 | static const struct seq_operations ptype_seq_ops = { | |
3907 | .start = ptype_seq_start, | |
3908 | .next = ptype_seq_next, | |
3909 | .stop = ptype_seq_stop, | |
3910 | .show = ptype_seq_show, | |
3911 | }; | |
3912 | ||
3913 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
3914 | { | |
2feb27db PE |
3915 | return seq_open_net(inode, file, &ptype_seq_ops, |
3916 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
3917 | } |
3918 | ||
3919 | static const struct file_operations ptype_seq_fops = { | |
3920 | .owner = THIS_MODULE, | |
3921 | .open = ptype_seq_open, | |
3922 | .read = seq_read, | |
3923 | .llseek = seq_lseek, | |
2feb27db | 3924 | .release = seq_release_net, |
0e1256ff SH |
3925 | }; |
3926 | ||
3927 | ||
4665079c | 3928 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
3929 | { |
3930 | int rc = -ENOMEM; | |
3931 | ||
881d966b | 3932 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 3933 | goto out; |
881d966b | 3934 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 3935 | goto out_dev; |
881d966b | 3936 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 3937 | goto out_softnet; |
0e1256ff | 3938 | |
881d966b | 3939 | if (wext_proc_init(net)) |
457c4cbc | 3940 | goto out_ptype; |
1da177e4 LT |
3941 | rc = 0; |
3942 | out: | |
3943 | return rc; | |
457c4cbc | 3944 | out_ptype: |
881d966b | 3945 | proc_net_remove(net, "ptype"); |
1da177e4 | 3946 | out_softnet: |
881d966b | 3947 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 3948 | out_dev: |
881d966b | 3949 | proc_net_remove(net, "dev"); |
1da177e4 LT |
3950 | goto out; |
3951 | } | |
881d966b | 3952 | |
4665079c | 3953 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
3954 | { |
3955 | wext_proc_exit(net); | |
3956 | ||
3957 | proc_net_remove(net, "ptype"); | |
3958 | proc_net_remove(net, "softnet_stat"); | |
3959 | proc_net_remove(net, "dev"); | |
3960 | } | |
3961 | ||
022cbae6 | 3962 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
3963 | .init = dev_proc_net_init, |
3964 | .exit = dev_proc_net_exit, | |
3965 | }; | |
3966 | ||
3967 | static int __init dev_proc_init(void) | |
3968 | { | |
3969 | return register_pernet_subsys(&dev_proc_ops); | |
3970 | } | |
1da177e4 LT |
3971 | #else |
3972 | #define dev_proc_init() 0 | |
3973 | #endif /* CONFIG_PROC_FS */ | |
3974 | ||
3975 | ||
3976 | /** | |
3977 | * netdev_set_master - set up master/slave pair | |
3978 | * @slave: slave device | |
3979 | * @master: new master device | |
3980 | * | |
3981 | * Changes the master device of the slave. Pass %NULL to break the | |
3982 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
3983 | * a negative errno code is returned. On success the reference counts | |
3984 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
3985 | * function returns zero. | |
3986 | */ | |
3987 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
3988 | { | |
3989 | struct net_device *old = slave->master; | |
3990 | ||
3991 | ASSERT_RTNL(); | |
3992 | ||
3993 | if (master) { | |
3994 | if (old) | |
3995 | return -EBUSY; | |
3996 | dev_hold(master); | |
3997 | } | |
3998 | ||
3999 | slave->master = master; | |
4ec93edb | 4000 | |
283f2fe8 ED |
4001 | if (old) { |
4002 | synchronize_net(); | |
1da177e4 | 4003 | dev_put(old); |
283f2fe8 | 4004 | } |
1da177e4 LT |
4005 | if (master) |
4006 | slave->flags |= IFF_SLAVE; | |
4007 | else | |
4008 | slave->flags &= ~IFF_SLAVE; | |
4009 | ||
4010 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
4011 | return 0; | |
4012 | } | |
d1b19dff | 4013 | EXPORT_SYMBOL(netdev_set_master); |
1da177e4 | 4014 | |
b6c40d68 PM |
4015 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
4016 | { | |
d314774c SH |
4017 | const struct net_device_ops *ops = dev->netdev_ops; |
4018 | ||
4019 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
4020 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
4021 | } |
4022 | ||
dad9b335 | 4023 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
4024 | { |
4025 | unsigned short old_flags = dev->flags; | |
8192b0c4 DH |
4026 | uid_t uid; |
4027 | gid_t gid; | |
1da177e4 | 4028 | |
24023451 PM |
4029 | ASSERT_RTNL(); |
4030 | ||
dad9b335 WC |
4031 | dev->flags |= IFF_PROMISC; |
4032 | dev->promiscuity += inc; | |
4033 | if (dev->promiscuity == 0) { | |
4034 | /* | |
4035 | * Avoid overflow. | |
4036 | * If inc causes overflow, untouch promisc and return error. | |
4037 | */ | |
4038 | if (inc < 0) | |
4039 | dev->flags &= ~IFF_PROMISC; | |
4040 | else { | |
4041 | dev->promiscuity -= inc; | |
4042 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
4043 | "set promiscuity failed, promiscuity feature " | |
4044 | "of device might be broken.\n", dev->name); | |
4045 | return -EOVERFLOW; | |
4046 | } | |
4047 | } | |
52609c0b | 4048 | if (dev->flags != old_flags) { |
1da177e4 LT |
4049 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
4050 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 4051 | "left"); |
8192b0c4 DH |
4052 | if (audit_enabled) { |
4053 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
4054 | audit_log(current->audit_context, GFP_ATOMIC, |
4055 | AUDIT_ANOM_PROMISCUOUS, | |
4056 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
4057 | dev->name, (dev->flags & IFF_PROMISC), | |
4058 | (old_flags & IFF_PROMISC), | |
4059 | audit_get_loginuid(current), | |
8192b0c4 | 4060 | uid, gid, |
7759db82 | 4061 | audit_get_sessionid(current)); |
8192b0c4 | 4062 | } |
24023451 | 4063 | |
b6c40d68 | 4064 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 4065 | } |
dad9b335 | 4066 | return 0; |
1da177e4 LT |
4067 | } |
4068 | ||
4417da66 PM |
4069 | /** |
4070 | * dev_set_promiscuity - update promiscuity count on a device | |
4071 | * @dev: device | |
4072 | * @inc: modifier | |
4073 | * | |
4074 | * Add or remove promiscuity from a device. While the count in the device | |
4075 | * remains above zero the interface remains promiscuous. Once it hits zero | |
4076 | * the device reverts back to normal filtering operation. A negative inc | |
4077 | * value is used to drop promiscuity on the device. | |
dad9b335 | 4078 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 4079 | */ |
dad9b335 | 4080 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
4081 | { |
4082 | unsigned short old_flags = dev->flags; | |
dad9b335 | 4083 | int err; |
4417da66 | 4084 | |
dad9b335 | 4085 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4086 | if (err < 0) |
dad9b335 | 4087 | return err; |
4417da66 PM |
4088 | if (dev->flags != old_flags) |
4089 | dev_set_rx_mode(dev); | |
dad9b335 | 4090 | return err; |
4417da66 | 4091 | } |
d1b19dff | 4092 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 4093 | |
1da177e4 LT |
4094 | /** |
4095 | * dev_set_allmulti - update allmulti count on a device | |
4096 | * @dev: device | |
4097 | * @inc: modifier | |
4098 | * | |
4099 | * Add or remove reception of all multicast frames to a device. While the | |
4100 | * count in the device remains above zero the interface remains listening | |
4101 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4102 | * filtering operation. A negative @inc value is used to drop the counter | |
4103 | * when releasing a resource needing all multicasts. | |
dad9b335 | 4104 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
4105 | */ |
4106 | ||
dad9b335 | 4107 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
4108 | { |
4109 | unsigned short old_flags = dev->flags; | |
4110 | ||
24023451 PM |
4111 | ASSERT_RTNL(); |
4112 | ||
1da177e4 | 4113 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
4114 | dev->allmulti += inc; |
4115 | if (dev->allmulti == 0) { | |
4116 | /* | |
4117 | * Avoid overflow. | |
4118 | * If inc causes overflow, untouch allmulti and return error. | |
4119 | */ | |
4120 | if (inc < 0) | |
4121 | dev->flags &= ~IFF_ALLMULTI; | |
4122 | else { | |
4123 | dev->allmulti -= inc; | |
4124 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
4125 | "set allmulti failed, allmulti feature of " | |
4126 | "device might be broken.\n", dev->name); | |
4127 | return -EOVERFLOW; | |
4128 | } | |
4129 | } | |
24023451 | 4130 | if (dev->flags ^ old_flags) { |
b6c40d68 | 4131 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 4132 | dev_set_rx_mode(dev); |
24023451 | 4133 | } |
dad9b335 | 4134 | return 0; |
4417da66 | 4135 | } |
d1b19dff | 4136 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
4137 | |
4138 | /* | |
4139 | * Upload unicast and multicast address lists to device and | |
4140 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 4141 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
4142 | * are present. |
4143 | */ | |
4144 | void __dev_set_rx_mode(struct net_device *dev) | |
4145 | { | |
d314774c SH |
4146 | const struct net_device_ops *ops = dev->netdev_ops; |
4147 | ||
4417da66 PM |
4148 | /* dev_open will call this function so the list will stay sane. */ |
4149 | if (!(dev->flags&IFF_UP)) | |
4150 | return; | |
4151 | ||
4152 | if (!netif_device_present(dev)) | |
40b77c94 | 4153 | return; |
4417da66 | 4154 | |
d314774c SH |
4155 | if (ops->ndo_set_rx_mode) |
4156 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
4157 | else { |
4158 | /* Unicast addresses changes may only happen under the rtnl, | |
4159 | * therefore calling __dev_set_promiscuity here is safe. | |
4160 | */ | |
32e7bfc4 | 4161 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 PM |
4162 | __dev_set_promiscuity(dev, 1); |
4163 | dev->uc_promisc = 1; | |
32e7bfc4 | 4164 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 PM |
4165 | __dev_set_promiscuity(dev, -1); |
4166 | dev->uc_promisc = 0; | |
4167 | } | |
4168 | ||
d314774c SH |
4169 | if (ops->ndo_set_multicast_list) |
4170 | ops->ndo_set_multicast_list(dev); | |
4417da66 PM |
4171 | } |
4172 | } | |
4173 | ||
4174 | void dev_set_rx_mode(struct net_device *dev) | |
4175 | { | |
b9e40857 | 4176 | netif_addr_lock_bh(dev); |
4417da66 | 4177 | __dev_set_rx_mode(dev); |
b9e40857 | 4178 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4179 | } |
4180 | ||
f0db275a SH |
4181 | /** |
4182 | * dev_get_flags - get flags reported to userspace | |
4183 | * @dev: device | |
4184 | * | |
4185 | * Get the combination of flag bits exported through APIs to userspace. | |
4186 | */ | |
1da177e4 LT |
4187 | unsigned dev_get_flags(const struct net_device *dev) |
4188 | { | |
4189 | unsigned flags; | |
4190 | ||
4191 | flags = (dev->flags & ~(IFF_PROMISC | | |
4192 | IFF_ALLMULTI | | |
b00055aa SR |
4193 | IFF_RUNNING | |
4194 | IFF_LOWER_UP | | |
4195 | IFF_DORMANT)) | | |
1da177e4 LT |
4196 | (dev->gflags & (IFF_PROMISC | |
4197 | IFF_ALLMULTI)); | |
4198 | ||
b00055aa SR |
4199 | if (netif_running(dev)) { |
4200 | if (netif_oper_up(dev)) | |
4201 | flags |= IFF_RUNNING; | |
4202 | if (netif_carrier_ok(dev)) | |
4203 | flags |= IFF_LOWER_UP; | |
4204 | if (netif_dormant(dev)) | |
4205 | flags |= IFF_DORMANT; | |
4206 | } | |
1da177e4 LT |
4207 | |
4208 | return flags; | |
4209 | } | |
d1b19dff | 4210 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4211 | |
bd380811 | 4212 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4213 | { |
1da177e4 | 4214 | int old_flags = dev->flags; |
bd380811 | 4215 | int ret; |
1da177e4 | 4216 | |
24023451 PM |
4217 | ASSERT_RTNL(); |
4218 | ||
1da177e4 LT |
4219 | /* |
4220 | * Set the flags on our device. | |
4221 | */ | |
4222 | ||
4223 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4224 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4225 | IFF_AUTOMEDIA)) | | |
4226 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4227 | IFF_ALLMULTI)); | |
4228 | ||
4229 | /* | |
4230 | * Load in the correct multicast list now the flags have changed. | |
4231 | */ | |
4232 | ||
b6c40d68 PM |
4233 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4234 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4235 | |
4417da66 | 4236 | dev_set_rx_mode(dev); |
1da177e4 LT |
4237 | |
4238 | /* | |
4239 | * Have we downed the interface. We handle IFF_UP ourselves | |
4240 | * according to user attempts to set it, rather than blindly | |
4241 | * setting it. | |
4242 | */ | |
4243 | ||
4244 | ret = 0; | |
4245 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4246 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4247 | |
4248 | if (!ret) | |
4417da66 | 4249 | dev_set_rx_mode(dev); |
1da177e4 LT |
4250 | } |
4251 | ||
1da177e4 | 4252 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4253 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4254 | ||
1da177e4 LT |
4255 | dev->gflags ^= IFF_PROMISC; |
4256 | dev_set_promiscuity(dev, inc); | |
4257 | } | |
4258 | ||
4259 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4260 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4261 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4262 | */ | |
4263 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4264 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4265 | ||
1da177e4 LT |
4266 | dev->gflags ^= IFF_ALLMULTI; |
4267 | dev_set_allmulti(dev, inc); | |
4268 | } | |
4269 | ||
bd380811 PM |
4270 | return ret; |
4271 | } | |
4272 | ||
4273 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4274 | { | |
4275 | unsigned int changes = dev->flags ^ old_flags; | |
4276 | ||
4277 | if (changes & IFF_UP) { | |
4278 | if (dev->flags & IFF_UP) | |
4279 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4280 | else | |
4281 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4282 | } | |
4283 | ||
4284 | if (dev->flags & IFF_UP && | |
4285 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4286 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4287 | } | |
4288 | ||
4289 | /** | |
4290 | * dev_change_flags - change device settings | |
4291 | * @dev: device | |
4292 | * @flags: device state flags | |
4293 | * | |
4294 | * Change settings on device based state flags. The flags are | |
4295 | * in the userspace exported format. | |
4296 | */ | |
4297 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
4298 | { | |
4299 | int ret, changes; | |
4300 | int old_flags = dev->flags; | |
4301 | ||
4302 | ret = __dev_change_flags(dev, flags); | |
4303 | if (ret < 0) | |
4304 | return ret; | |
4305 | ||
4306 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4307 | if (changes) |
4308 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4309 | |
bd380811 | 4310 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4311 | return ret; |
4312 | } | |
d1b19dff | 4313 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4314 | |
f0db275a SH |
4315 | /** |
4316 | * dev_set_mtu - Change maximum transfer unit | |
4317 | * @dev: device | |
4318 | * @new_mtu: new transfer unit | |
4319 | * | |
4320 | * Change the maximum transfer size of the network device. | |
4321 | */ | |
1da177e4 LT |
4322 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4323 | { | |
d314774c | 4324 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4325 | int err; |
4326 | ||
4327 | if (new_mtu == dev->mtu) | |
4328 | return 0; | |
4329 | ||
4330 | /* MTU must be positive. */ | |
4331 | if (new_mtu < 0) | |
4332 | return -EINVAL; | |
4333 | ||
4334 | if (!netif_device_present(dev)) | |
4335 | return -ENODEV; | |
4336 | ||
4337 | err = 0; | |
d314774c SH |
4338 | if (ops->ndo_change_mtu) |
4339 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4340 | else |
4341 | dev->mtu = new_mtu; | |
d314774c | 4342 | |
1da177e4 | 4343 | if (!err && dev->flags & IFF_UP) |
056925ab | 4344 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4345 | return err; |
4346 | } | |
d1b19dff | 4347 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4348 | |
f0db275a SH |
4349 | /** |
4350 | * dev_set_mac_address - Change Media Access Control Address | |
4351 | * @dev: device | |
4352 | * @sa: new address | |
4353 | * | |
4354 | * Change the hardware (MAC) address of the device | |
4355 | */ | |
1da177e4 LT |
4356 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4357 | { | |
d314774c | 4358 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4359 | int err; |
4360 | ||
d314774c | 4361 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4362 | return -EOPNOTSUPP; |
4363 | if (sa->sa_family != dev->type) | |
4364 | return -EINVAL; | |
4365 | if (!netif_device_present(dev)) | |
4366 | return -ENODEV; | |
d314774c | 4367 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4368 | if (!err) |
056925ab | 4369 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
4370 | return err; |
4371 | } | |
d1b19dff | 4372 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4373 | |
4374 | /* | |
3710becf | 4375 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4376 | */ |
14e3e079 | 4377 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4378 | { |
4379 | int err; | |
3710becf | 4380 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4381 | |
4382 | if (!dev) | |
4383 | return -ENODEV; | |
4384 | ||
4385 | switch (cmd) { | |
d1b19dff ED |
4386 | case SIOCGIFFLAGS: /* Get interface flags */ |
4387 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4388 | return 0; | |
1da177e4 | 4389 | |
d1b19dff ED |
4390 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4391 | (currently unused) */ | |
4392 | ifr->ifr_metric = 0; | |
4393 | return 0; | |
1da177e4 | 4394 | |
d1b19dff ED |
4395 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4396 | ifr->ifr_mtu = dev->mtu; | |
4397 | return 0; | |
1da177e4 | 4398 | |
d1b19dff ED |
4399 | case SIOCGIFHWADDR: |
4400 | if (!dev->addr_len) | |
4401 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4402 | else | |
4403 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4404 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4405 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4406 | return 0; | |
1da177e4 | 4407 | |
d1b19dff ED |
4408 | case SIOCGIFSLAVE: |
4409 | err = -EINVAL; | |
4410 | break; | |
14e3e079 | 4411 | |
d1b19dff ED |
4412 | case SIOCGIFMAP: |
4413 | ifr->ifr_map.mem_start = dev->mem_start; | |
4414 | ifr->ifr_map.mem_end = dev->mem_end; | |
4415 | ifr->ifr_map.base_addr = dev->base_addr; | |
4416 | ifr->ifr_map.irq = dev->irq; | |
4417 | ifr->ifr_map.dma = dev->dma; | |
4418 | ifr->ifr_map.port = dev->if_port; | |
4419 | return 0; | |
14e3e079 | 4420 | |
d1b19dff ED |
4421 | case SIOCGIFINDEX: |
4422 | ifr->ifr_ifindex = dev->ifindex; | |
4423 | return 0; | |
14e3e079 | 4424 | |
d1b19dff ED |
4425 | case SIOCGIFTXQLEN: |
4426 | ifr->ifr_qlen = dev->tx_queue_len; | |
4427 | return 0; | |
14e3e079 | 4428 | |
d1b19dff ED |
4429 | default: |
4430 | /* dev_ioctl() should ensure this case | |
4431 | * is never reached | |
4432 | */ | |
4433 | WARN_ON(1); | |
4434 | err = -EINVAL; | |
4435 | break; | |
14e3e079 JG |
4436 | |
4437 | } | |
4438 | return err; | |
4439 | } | |
4440 | ||
4441 | /* | |
4442 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4443 | */ | |
4444 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4445 | { | |
4446 | int err; | |
4447 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4448 | const struct net_device_ops *ops; |
14e3e079 JG |
4449 | |
4450 | if (!dev) | |
4451 | return -ENODEV; | |
4452 | ||
5f2f6da7 JP |
4453 | ops = dev->netdev_ops; |
4454 | ||
14e3e079 | 4455 | switch (cmd) { |
d1b19dff ED |
4456 | case SIOCSIFFLAGS: /* Set interface flags */ |
4457 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 4458 | |
d1b19dff ED |
4459 | case SIOCSIFMETRIC: /* Set the metric on the interface |
4460 | (currently unused) */ | |
4461 | return -EOPNOTSUPP; | |
14e3e079 | 4462 | |
d1b19dff ED |
4463 | case SIOCSIFMTU: /* Set the MTU of a device */ |
4464 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 4465 | |
d1b19dff ED |
4466 | case SIOCSIFHWADDR: |
4467 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 4468 | |
d1b19dff ED |
4469 | case SIOCSIFHWBROADCAST: |
4470 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4471 | return -EINVAL; | |
4472 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4473 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4474 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4475 | return 0; | |
1da177e4 | 4476 | |
d1b19dff ED |
4477 | case SIOCSIFMAP: |
4478 | if (ops->ndo_set_config) { | |
1da177e4 LT |
4479 | if (!netif_device_present(dev)) |
4480 | return -ENODEV; | |
d1b19dff ED |
4481 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
4482 | } | |
4483 | return -EOPNOTSUPP; | |
1da177e4 | 4484 | |
d1b19dff ED |
4485 | case SIOCADDMULTI: |
4486 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4487 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4488 | return -EINVAL; | |
4489 | if (!netif_device_present(dev)) | |
4490 | return -ENODEV; | |
22bedad3 | 4491 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); |
d1b19dff ED |
4492 | |
4493 | case SIOCDELMULTI: | |
4494 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4495 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4496 | return -EINVAL; | |
4497 | if (!netif_device_present(dev)) | |
4498 | return -ENODEV; | |
22bedad3 | 4499 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); |
1da177e4 | 4500 | |
d1b19dff ED |
4501 | case SIOCSIFTXQLEN: |
4502 | if (ifr->ifr_qlen < 0) | |
4503 | return -EINVAL; | |
4504 | dev->tx_queue_len = ifr->ifr_qlen; | |
4505 | return 0; | |
1da177e4 | 4506 | |
d1b19dff ED |
4507 | case SIOCSIFNAME: |
4508 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4509 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 4510 | |
d1b19dff ED |
4511 | /* |
4512 | * Unknown or private ioctl | |
4513 | */ | |
4514 | default: | |
4515 | if ((cmd >= SIOCDEVPRIVATE && | |
4516 | cmd <= SIOCDEVPRIVATE + 15) || | |
4517 | cmd == SIOCBONDENSLAVE || | |
4518 | cmd == SIOCBONDRELEASE || | |
4519 | cmd == SIOCBONDSETHWADDR || | |
4520 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4521 | cmd == SIOCBONDINFOQUERY || | |
4522 | cmd == SIOCBONDCHANGEACTIVE || | |
4523 | cmd == SIOCGMIIPHY || | |
4524 | cmd == SIOCGMIIREG || | |
4525 | cmd == SIOCSMIIREG || | |
4526 | cmd == SIOCBRADDIF || | |
4527 | cmd == SIOCBRDELIF || | |
4528 | cmd == SIOCSHWTSTAMP || | |
4529 | cmd == SIOCWANDEV) { | |
4530 | err = -EOPNOTSUPP; | |
4531 | if (ops->ndo_do_ioctl) { | |
4532 | if (netif_device_present(dev)) | |
4533 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4534 | else | |
4535 | err = -ENODEV; | |
4536 | } | |
4537 | } else | |
4538 | err = -EINVAL; | |
1da177e4 LT |
4539 | |
4540 | } | |
4541 | return err; | |
4542 | } | |
4543 | ||
4544 | /* | |
4545 | * This function handles all "interface"-type I/O control requests. The actual | |
4546 | * 'doing' part of this is dev_ifsioc above. | |
4547 | */ | |
4548 | ||
4549 | /** | |
4550 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4551 | * @net: the applicable net namespace |
1da177e4 LT |
4552 | * @cmd: command to issue |
4553 | * @arg: pointer to a struct ifreq in user space | |
4554 | * | |
4555 | * Issue ioctl functions to devices. This is normally called by the | |
4556 | * user space syscall interfaces but can sometimes be useful for | |
4557 | * other purposes. The return value is the return from the syscall if | |
4558 | * positive or a negative errno code on error. | |
4559 | */ | |
4560 | ||
881d966b | 4561 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4562 | { |
4563 | struct ifreq ifr; | |
4564 | int ret; | |
4565 | char *colon; | |
4566 | ||
4567 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4568 | and requires shared lock, because it sleeps writing | |
4569 | to user space. | |
4570 | */ | |
4571 | ||
4572 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 4573 | rtnl_lock(); |
881d966b | 4574 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 4575 | rtnl_unlock(); |
1da177e4 LT |
4576 | return ret; |
4577 | } | |
4578 | if (cmd == SIOCGIFNAME) | |
881d966b | 4579 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
4580 | |
4581 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4582 | return -EFAULT; | |
4583 | ||
4584 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4585 | ||
4586 | colon = strchr(ifr.ifr_name, ':'); | |
4587 | if (colon) | |
4588 | *colon = 0; | |
4589 | ||
4590 | /* | |
4591 | * See which interface the caller is talking about. | |
4592 | */ | |
4593 | ||
4594 | switch (cmd) { | |
d1b19dff ED |
4595 | /* |
4596 | * These ioctl calls: | |
4597 | * - can be done by all. | |
4598 | * - atomic and do not require locking. | |
4599 | * - return a value | |
4600 | */ | |
4601 | case SIOCGIFFLAGS: | |
4602 | case SIOCGIFMETRIC: | |
4603 | case SIOCGIFMTU: | |
4604 | case SIOCGIFHWADDR: | |
4605 | case SIOCGIFSLAVE: | |
4606 | case SIOCGIFMAP: | |
4607 | case SIOCGIFINDEX: | |
4608 | case SIOCGIFTXQLEN: | |
4609 | dev_load(net, ifr.ifr_name); | |
3710becf | 4610 | rcu_read_lock(); |
d1b19dff | 4611 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 4612 | rcu_read_unlock(); |
d1b19dff ED |
4613 | if (!ret) { |
4614 | if (colon) | |
4615 | *colon = ':'; | |
4616 | if (copy_to_user(arg, &ifr, | |
4617 | sizeof(struct ifreq))) | |
4618 | ret = -EFAULT; | |
4619 | } | |
4620 | return ret; | |
1da177e4 | 4621 | |
d1b19dff ED |
4622 | case SIOCETHTOOL: |
4623 | dev_load(net, ifr.ifr_name); | |
4624 | rtnl_lock(); | |
4625 | ret = dev_ethtool(net, &ifr); | |
4626 | rtnl_unlock(); | |
4627 | if (!ret) { | |
4628 | if (colon) | |
4629 | *colon = ':'; | |
4630 | if (copy_to_user(arg, &ifr, | |
4631 | sizeof(struct ifreq))) | |
4632 | ret = -EFAULT; | |
4633 | } | |
4634 | return ret; | |
1da177e4 | 4635 | |
d1b19dff ED |
4636 | /* |
4637 | * These ioctl calls: | |
4638 | * - require superuser power. | |
4639 | * - require strict serialization. | |
4640 | * - return a value | |
4641 | */ | |
4642 | case SIOCGMIIPHY: | |
4643 | case SIOCGMIIREG: | |
4644 | case SIOCSIFNAME: | |
4645 | if (!capable(CAP_NET_ADMIN)) | |
4646 | return -EPERM; | |
4647 | dev_load(net, ifr.ifr_name); | |
4648 | rtnl_lock(); | |
4649 | ret = dev_ifsioc(net, &ifr, cmd); | |
4650 | rtnl_unlock(); | |
4651 | if (!ret) { | |
4652 | if (colon) | |
4653 | *colon = ':'; | |
4654 | if (copy_to_user(arg, &ifr, | |
4655 | sizeof(struct ifreq))) | |
4656 | ret = -EFAULT; | |
4657 | } | |
4658 | return ret; | |
1da177e4 | 4659 | |
d1b19dff ED |
4660 | /* |
4661 | * These ioctl calls: | |
4662 | * - require superuser power. | |
4663 | * - require strict serialization. | |
4664 | * - do not return a value | |
4665 | */ | |
4666 | case SIOCSIFFLAGS: | |
4667 | case SIOCSIFMETRIC: | |
4668 | case SIOCSIFMTU: | |
4669 | case SIOCSIFMAP: | |
4670 | case SIOCSIFHWADDR: | |
4671 | case SIOCSIFSLAVE: | |
4672 | case SIOCADDMULTI: | |
4673 | case SIOCDELMULTI: | |
4674 | case SIOCSIFHWBROADCAST: | |
4675 | case SIOCSIFTXQLEN: | |
4676 | case SIOCSMIIREG: | |
4677 | case SIOCBONDENSLAVE: | |
4678 | case SIOCBONDRELEASE: | |
4679 | case SIOCBONDSETHWADDR: | |
4680 | case SIOCBONDCHANGEACTIVE: | |
4681 | case SIOCBRADDIF: | |
4682 | case SIOCBRDELIF: | |
4683 | case SIOCSHWTSTAMP: | |
4684 | if (!capable(CAP_NET_ADMIN)) | |
4685 | return -EPERM; | |
4686 | /* fall through */ | |
4687 | case SIOCBONDSLAVEINFOQUERY: | |
4688 | case SIOCBONDINFOQUERY: | |
4689 | dev_load(net, ifr.ifr_name); | |
4690 | rtnl_lock(); | |
4691 | ret = dev_ifsioc(net, &ifr, cmd); | |
4692 | rtnl_unlock(); | |
4693 | return ret; | |
4694 | ||
4695 | case SIOCGIFMEM: | |
4696 | /* Get the per device memory space. We can add this but | |
4697 | * currently do not support it */ | |
4698 | case SIOCSIFMEM: | |
4699 | /* Set the per device memory buffer space. | |
4700 | * Not applicable in our case */ | |
4701 | case SIOCSIFLINK: | |
4702 | return -EINVAL; | |
4703 | ||
4704 | /* | |
4705 | * Unknown or private ioctl. | |
4706 | */ | |
4707 | default: | |
4708 | if (cmd == SIOCWANDEV || | |
4709 | (cmd >= SIOCDEVPRIVATE && | |
4710 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 4711 | dev_load(net, ifr.ifr_name); |
1da177e4 | 4712 | rtnl_lock(); |
881d966b | 4713 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 4714 | rtnl_unlock(); |
d1b19dff ED |
4715 | if (!ret && copy_to_user(arg, &ifr, |
4716 | sizeof(struct ifreq))) | |
4717 | ret = -EFAULT; | |
1da177e4 | 4718 | return ret; |
d1b19dff ED |
4719 | } |
4720 | /* Take care of Wireless Extensions */ | |
4721 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
4722 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
4723 | return -EINVAL; | |
1da177e4 LT |
4724 | } |
4725 | } | |
4726 | ||
4727 | ||
4728 | /** | |
4729 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 4730 | * @net: the applicable net namespace |
1da177e4 LT |
4731 | * |
4732 | * Returns a suitable unique value for a new device interface | |
4733 | * number. The caller must hold the rtnl semaphore or the | |
4734 | * dev_base_lock to be sure it remains unique. | |
4735 | */ | |
881d966b | 4736 | static int dev_new_index(struct net *net) |
1da177e4 LT |
4737 | { |
4738 | static int ifindex; | |
4739 | for (;;) { | |
4740 | if (++ifindex <= 0) | |
4741 | ifindex = 1; | |
881d966b | 4742 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
4743 | return ifindex; |
4744 | } | |
4745 | } | |
4746 | ||
1da177e4 | 4747 | /* Delayed registration/unregisteration */ |
3b5b34fd | 4748 | static LIST_HEAD(net_todo_list); |
1da177e4 | 4749 | |
6f05f629 | 4750 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 4751 | { |
1da177e4 | 4752 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
4753 | } |
4754 | ||
9b5e383c | 4755 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 4756 | { |
e93737b0 | 4757 | struct net_device *dev, *tmp; |
9b5e383c | 4758 | |
93ee31f1 DL |
4759 | BUG_ON(dev_boot_phase); |
4760 | ASSERT_RTNL(); | |
4761 | ||
e93737b0 | 4762 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 4763 | /* Some devices call without registering |
e93737b0 KK |
4764 | * for initialization unwind. Remove those |
4765 | * devices and proceed with the remaining. | |
9b5e383c ED |
4766 | */ |
4767 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
4768 | pr_debug("unregister_netdevice: device %s/%p never " | |
4769 | "was registered\n", dev->name, dev); | |
93ee31f1 | 4770 | |
9b5e383c | 4771 | WARN_ON(1); |
e93737b0 KK |
4772 | list_del(&dev->unreg_list); |
4773 | continue; | |
9b5e383c | 4774 | } |
93ee31f1 | 4775 | |
9b5e383c | 4776 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
93ee31f1 | 4777 | |
9b5e383c ED |
4778 | /* If device is running, close it first. */ |
4779 | dev_close(dev); | |
93ee31f1 | 4780 | |
9b5e383c ED |
4781 | /* And unlink it from device chain. */ |
4782 | unlist_netdevice(dev); | |
93ee31f1 | 4783 | |
9b5e383c ED |
4784 | dev->reg_state = NETREG_UNREGISTERING; |
4785 | } | |
93ee31f1 DL |
4786 | |
4787 | synchronize_net(); | |
4788 | ||
9b5e383c ED |
4789 | list_for_each_entry(dev, head, unreg_list) { |
4790 | /* Shutdown queueing discipline. */ | |
4791 | dev_shutdown(dev); | |
93ee31f1 DL |
4792 | |
4793 | ||
9b5e383c ED |
4794 | /* Notify protocols, that we are about to destroy |
4795 | this device. They should clean all the things. | |
4796 | */ | |
4797 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 4798 | |
a2835763 PM |
4799 | if (!dev->rtnl_link_ops || |
4800 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
4801 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
4802 | ||
9b5e383c ED |
4803 | /* |
4804 | * Flush the unicast and multicast chains | |
4805 | */ | |
a748ee24 | 4806 | dev_uc_flush(dev); |
22bedad3 | 4807 | dev_mc_flush(dev); |
93ee31f1 | 4808 | |
9b5e383c ED |
4809 | if (dev->netdev_ops->ndo_uninit) |
4810 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 4811 | |
9b5e383c ED |
4812 | /* Notifier chain MUST detach us from master device. */ |
4813 | WARN_ON(dev->master); | |
93ee31f1 | 4814 | |
9b5e383c ED |
4815 | /* Remove entries from kobject tree */ |
4816 | netdev_unregister_kobject(dev); | |
4817 | } | |
93ee31f1 | 4818 | |
a5ee1551 | 4819 | /* Process any work delayed until the end of the batch */ |
e5e26d75 | 4820 | dev = list_first_entry(head, struct net_device, unreg_list); |
a5ee1551 | 4821 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
93ee31f1 | 4822 | |
a5ee1551 | 4823 | synchronize_net(); |
395264d5 | 4824 | |
a5ee1551 | 4825 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
4826 | dev_put(dev); |
4827 | } | |
4828 | ||
4829 | static void rollback_registered(struct net_device *dev) | |
4830 | { | |
4831 | LIST_HEAD(single); | |
4832 | ||
4833 | list_add(&dev->unreg_list, &single); | |
4834 | rollback_registered_many(&single); | |
93ee31f1 DL |
4835 | } |
4836 | ||
e8a0464c DM |
4837 | static void __netdev_init_queue_locks_one(struct net_device *dev, |
4838 | struct netdev_queue *dev_queue, | |
4839 | void *_unused) | |
c773e847 DM |
4840 | { |
4841 | spin_lock_init(&dev_queue->_xmit_lock); | |
cf508b12 | 4842 | netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type); |
c773e847 DM |
4843 | dev_queue->xmit_lock_owner = -1; |
4844 | } | |
4845 | ||
4846 | static void netdev_init_queue_locks(struct net_device *dev) | |
4847 | { | |
e8a0464c DM |
4848 | netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL); |
4849 | __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL); | |
c773e847 DM |
4850 | } |
4851 | ||
b63365a2 HX |
4852 | unsigned long netdev_fix_features(unsigned long features, const char *name) |
4853 | { | |
4854 | /* Fix illegal SG+CSUM combinations. */ | |
4855 | if ((features & NETIF_F_SG) && | |
4856 | !(features & NETIF_F_ALL_CSUM)) { | |
4857 | if (name) | |
4858 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " | |
4859 | "checksum feature.\n", name); | |
4860 | features &= ~NETIF_F_SG; | |
4861 | } | |
4862 | ||
4863 | /* TSO requires that SG is present as well. */ | |
4864 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { | |
4865 | if (name) | |
4866 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " | |
4867 | "SG feature.\n", name); | |
4868 | features &= ~NETIF_F_TSO; | |
4869 | } | |
4870 | ||
4871 | if (features & NETIF_F_UFO) { | |
4872 | if (!(features & NETIF_F_GEN_CSUM)) { | |
4873 | if (name) | |
4874 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4875 | "since no NETIF_F_HW_CSUM feature.\n", | |
4876 | name); | |
4877 | features &= ~NETIF_F_UFO; | |
4878 | } | |
4879 | ||
4880 | if (!(features & NETIF_F_SG)) { | |
4881 | if (name) | |
4882 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4883 | "since no NETIF_F_SG feature.\n", name); | |
4884 | features &= ~NETIF_F_UFO; | |
4885 | } | |
4886 | } | |
4887 | ||
4888 | return features; | |
4889 | } | |
4890 | EXPORT_SYMBOL(netdev_fix_features); | |
4891 | ||
fc4a7489 PM |
4892 | /** |
4893 | * netif_stacked_transfer_operstate - transfer operstate | |
4894 | * @rootdev: the root or lower level device to transfer state from | |
4895 | * @dev: the device to transfer operstate to | |
4896 | * | |
4897 | * Transfer operational state from root to device. This is normally | |
4898 | * called when a stacking relationship exists between the root | |
4899 | * device and the device(a leaf device). | |
4900 | */ | |
4901 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
4902 | struct net_device *dev) | |
4903 | { | |
4904 | if (rootdev->operstate == IF_OPER_DORMANT) | |
4905 | netif_dormant_on(dev); | |
4906 | else | |
4907 | netif_dormant_off(dev); | |
4908 | ||
4909 | if (netif_carrier_ok(rootdev)) { | |
4910 | if (!netif_carrier_ok(dev)) | |
4911 | netif_carrier_on(dev); | |
4912 | } else { | |
4913 | if (netif_carrier_ok(dev)) | |
4914 | netif_carrier_off(dev); | |
4915 | } | |
4916 | } | |
4917 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
4918 | ||
1da177e4 LT |
4919 | /** |
4920 | * register_netdevice - register a network device | |
4921 | * @dev: device to register | |
4922 | * | |
4923 | * Take a completed network device structure and add it to the kernel | |
4924 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
4925 | * chain. 0 is returned on success. A negative errno code is returned | |
4926 | * on a failure to set up the device, or if the name is a duplicate. | |
4927 | * | |
4928 | * Callers must hold the rtnl semaphore. You may want | |
4929 | * register_netdev() instead of this. | |
4930 | * | |
4931 | * BUGS: | |
4932 | * The locking appears insufficient to guarantee two parallel registers | |
4933 | * will not get the same name. | |
4934 | */ | |
4935 | ||
4936 | int register_netdevice(struct net_device *dev) | |
4937 | { | |
1da177e4 | 4938 | int ret; |
d314774c | 4939 | struct net *net = dev_net(dev); |
1da177e4 LT |
4940 | |
4941 | BUG_ON(dev_boot_phase); | |
4942 | ASSERT_RTNL(); | |
4943 | ||
b17a7c17 SH |
4944 | might_sleep(); |
4945 | ||
1da177e4 LT |
4946 | /* When net_device's are persistent, this will be fatal. */ |
4947 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 4948 | BUG_ON(!net); |
1da177e4 | 4949 | |
f1f28aa3 | 4950 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 4951 | netdev_set_addr_lockdep_class(dev); |
c773e847 | 4952 | netdev_init_queue_locks(dev); |
1da177e4 | 4953 | |
1da177e4 LT |
4954 | dev->iflink = -1; |
4955 | ||
df334545 | 4956 | #ifdef CONFIG_RPS |
0a9627f2 TH |
4957 | if (!dev->num_rx_queues) { |
4958 | /* | |
4959 | * Allocate a single RX queue if driver never called | |
4960 | * alloc_netdev_mq | |
4961 | */ | |
4962 | ||
4963 | dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL); | |
4964 | if (!dev->_rx) { | |
4965 | ret = -ENOMEM; | |
4966 | goto out; | |
4967 | } | |
4968 | ||
4969 | dev->_rx->first = dev->_rx; | |
4970 | atomic_set(&dev->_rx->count, 1); | |
4971 | dev->num_rx_queues = 1; | |
4972 | } | |
df334545 | 4973 | #endif |
1da177e4 | 4974 | /* Init, if this function is available */ |
d314774c SH |
4975 | if (dev->netdev_ops->ndo_init) { |
4976 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
4977 | if (ret) { |
4978 | if (ret > 0) | |
4979 | ret = -EIO; | |
90833aa4 | 4980 | goto out; |
1da177e4 LT |
4981 | } |
4982 | } | |
4ec93edb | 4983 | |
8ce6cebc | 4984 | ret = dev_get_valid_name(dev, dev->name, 0); |
d9031024 | 4985 | if (ret) |
7ce1b0ed | 4986 | goto err_uninit; |
1da177e4 | 4987 | |
881d966b | 4988 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
4989 | if (dev->iflink == -1) |
4990 | dev->iflink = dev->ifindex; | |
4991 | ||
d212f87b SH |
4992 | /* Fix illegal checksum combinations */ |
4993 | if ((dev->features & NETIF_F_HW_CSUM) && | |
4994 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
4995 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
4996 | dev->name); | |
4997 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
4998 | } | |
4999 | ||
5000 | if ((dev->features & NETIF_F_NO_CSUM) && | |
5001 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
5002 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
5003 | dev->name); | |
5004 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
5005 | } | |
5006 | ||
b63365a2 | 5007 | dev->features = netdev_fix_features(dev->features, dev->name); |
1da177e4 | 5008 | |
e5a4a72d LB |
5009 | /* Enable software GSO if SG is supported. */ |
5010 | if (dev->features & NETIF_F_SG) | |
5011 | dev->features |= NETIF_F_GSO; | |
5012 | ||
7ffbe3fd JB |
5013 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
5014 | ret = notifier_to_errno(ret); | |
5015 | if (ret) | |
5016 | goto err_uninit; | |
5017 | ||
8b41d188 | 5018 | ret = netdev_register_kobject(dev); |
b17a7c17 | 5019 | if (ret) |
7ce1b0ed | 5020 | goto err_uninit; |
b17a7c17 SH |
5021 | dev->reg_state = NETREG_REGISTERED; |
5022 | ||
1da177e4 LT |
5023 | /* |
5024 | * Default initial state at registry is that the | |
5025 | * device is present. | |
5026 | */ | |
5027 | ||
5028 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5029 | ||
1da177e4 | 5030 | dev_init_scheduler(dev); |
1da177e4 | 5031 | dev_hold(dev); |
ce286d32 | 5032 | list_netdevice(dev); |
1da177e4 LT |
5033 | |
5034 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 5035 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 5036 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
5037 | if (ret) { |
5038 | rollback_registered(dev); | |
5039 | dev->reg_state = NETREG_UNREGISTERED; | |
5040 | } | |
d90a909e EB |
5041 | /* |
5042 | * Prevent userspace races by waiting until the network | |
5043 | * device is fully setup before sending notifications. | |
5044 | */ | |
a2835763 PM |
5045 | if (!dev->rtnl_link_ops || |
5046 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
5047 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
5048 | |
5049 | out: | |
5050 | return ret; | |
7ce1b0ed HX |
5051 | |
5052 | err_uninit: | |
d314774c SH |
5053 | if (dev->netdev_ops->ndo_uninit) |
5054 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 5055 | goto out; |
1da177e4 | 5056 | } |
d1b19dff | 5057 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 5058 | |
937f1ba5 BH |
5059 | /** |
5060 | * init_dummy_netdev - init a dummy network device for NAPI | |
5061 | * @dev: device to init | |
5062 | * | |
5063 | * This takes a network device structure and initialize the minimum | |
5064 | * amount of fields so it can be used to schedule NAPI polls without | |
5065 | * registering a full blown interface. This is to be used by drivers | |
5066 | * that need to tie several hardware interfaces to a single NAPI | |
5067 | * poll scheduler due to HW limitations. | |
5068 | */ | |
5069 | int init_dummy_netdev(struct net_device *dev) | |
5070 | { | |
5071 | /* Clear everything. Note we don't initialize spinlocks | |
5072 | * are they aren't supposed to be taken by any of the | |
5073 | * NAPI code and this dummy netdev is supposed to be | |
5074 | * only ever used for NAPI polls | |
5075 | */ | |
5076 | memset(dev, 0, sizeof(struct net_device)); | |
5077 | ||
5078 | /* make sure we BUG if trying to hit standard | |
5079 | * register/unregister code path | |
5080 | */ | |
5081 | dev->reg_state = NETREG_DUMMY; | |
5082 | ||
5083 | /* initialize the ref count */ | |
5084 | atomic_set(&dev->refcnt, 1); | |
5085 | ||
5086 | /* NAPI wants this */ | |
5087 | INIT_LIST_HEAD(&dev->napi_list); | |
5088 | ||
5089 | /* a dummy interface is started by default */ | |
5090 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5091 | set_bit(__LINK_STATE_START, &dev->state); | |
5092 | ||
5093 | return 0; | |
5094 | } | |
5095 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5096 | ||
5097 | ||
1da177e4 LT |
5098 | /** |
5099 | * register_netdev - register a network device | |
5100 | * @dev: device to register | |
5101 | * | |
5102 | * Take a completed network device structure and add it to the kernel | |
5103 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5104 | * chain. 0 is returned on success. A negative errno code is returned | |
5105 | * on a failure to set up the device, or if the name is a duplicate. | |
5106 | * | |
38b4da38 | 5107 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5108 | * and expands the device name if you passed a format string to |
5109 | * alloc_netdev. | |
5110 | */ | |
5111 | int register_netdev(struct net_device *dev) | |
5112 | { | |
5113 | int err; | |
5114 | ||
5115 | rtnl_lock(); | |
5116 | ||
5117 | /* | |
5118 | * If the name is a format string the caller wants us to do a | |
5119 | * name allocation. | |
5120 | */ | |
5121 | if (strchr(dev->name, '%')) { | |
5122 | err = dev_alloc_name(dev, dev->name); | |
5123 | if (err < 0) | |
5124 | goto out; | |
5125 | } | |
4ec93edb | 5126 | |
1da177e4 LT |
5127 | err = register_netdevice(dev); |
5128 | out: | |
5129 | rtnl_unlock(); | |
5130 | return err; | |
5131 | } | |
5132 | EXPORT_SYMBOL(register_netdev); | |
5133 | ||
5134 | /* | |
5135 | * netdev_wait_allrefs - wait until all references are gone. | |
5136 | * | |
5137 | * This is called when unregistering network devices. | |
5138 | * | |
5139 | * Any protocol or device that holds a reference should register | |
5140 | * for netdevice notification, and cleanup and put back the | |
5141 | * reference if they receive an UNREGISTER event. | |
5142 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5143 | * call dev_put. |
1da177e4 LT |
5144 | */ |
5145 | static void netdev_wait_allrefs(struct net_device *dev) | |
5146 | { | |
5147 | unsigned long rebroadcast_time, warning_time; | |
5148 | ||
e014debe ED |
5149 | linkwatch_forget_dev(dev); |
5150 | ||
1da177e4 LT |
5151 | rebroadcast_time = warning_time = jiffies; |
5152 | while (atomic_read(&dev->refcnt) != 0) { | |
5153 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
6756ae4b | 5154 | rtnl_lock(); |
1da177e4 LT |
5155 | |
5156 | /* Rebroadcast unregister notification */ | |
056925ab | 5157 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
a5ee1551 | 5158 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users |
395264d5 | 5159 | * should have already handle it the first time */ |
1da177e4 LT |
5160 | |
5161 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5162 | &dev->state)) { | |
5163 | /* We must not have linkwatch events | |
5164 | * pending on unregister. If this | |
5165 | * happens, we simply run the queue | |
5166 | * unscheduled, resulting in a noop | |
5167 | * for this device. | |
5168 | */ | |
5169 | linkwatch_run_queue(); | |
5170 | } | |
5171 | ||
6756ae4b | 5172 | __rtnl_unlock(); |
1da177e4 LT |
5173 | |
5174 | rebroadcast_time = jiffies; | |
5175 | } | |
5176 | ||
5177 | msleep(250); | |
5178 | ||
5179 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
5180 | printk(KERN_EMERG "unregister_netdevice: " | |
5181 | "waiting for %s to become free. Usage " | |
5182 | "count = %d\n", | |
5183 | dev->name, atomic_read(&dev->refcnt)); | |
5184 | warning_time = jiffies; | |
5185 | } | |
5186 | } | |
5187 | } | |
5188 | ||
5189 | /* The sequence is: | |
5190 | * | |
5191 | * rtnl_lock(); | |
5192 | * ... | |
5193 | * register_netdevice(x1); | |
5194 | * register_netdevice(x2); | |
5195 | * ... | |
5196 | * unregister_netdevice(y1); | |
5197 | * unregister_netdevice(y2); | |
5198 | * ... | |
5199 | * rtnl_unlock(); | |
5200 | * free_netdev(y1); | |
5201 | * free_netdev(y2); | |
5202 | * | |
58ec3b4d | 5203 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5204 | * This allows us to deal with problems: |
b17a7c17 | 5205 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5206 | * without deadlocking with linkwatch via keventd. |
5207 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5208 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5209 | * |
5210 | * We must not return until all unregister events added during | |
5211 | * the interval the lock was held have been completed. | |
1da177e4 | 5212 | */ |
1da177e4 LT |
5213 | void netdev_run_todo(void) |
5214 | { | |
626ab0e6 | 5215 | struct list_head list; |
1da177e4 | 5216 | |
1da177e4 | 5217 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5218 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5219 | |
5220 | __rtnl_unlock(); | |
626ab0e6 | 5221 | |
1da177e4 LT |
5222 | while (!list_empty(&list)) { |
5223 | struct net_device *dev | |
e5e26d75 | 5224 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5225 | list_del(&dev->todo_list); |
5226 | ||
b17a7c17 SH |
5227 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
5228 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5229 | dev->name, dev->reg_state); | |
5230 | dump_stack(); | |
5231 | continue; | |
5232 | } | |
1da177e4 | 5233 | |
b17a7c17 | 5234 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5235 | |
152102c7 | 5236 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5237 | |
b17a7c17 | 5238 | netdev_wait_allrefs(dev); |
1da177e4 | 5239 | |
b17a7c17 SH |
5240 | /* paranoia */ |
5241 | BUG_ON(atomic_read(&dev->refcnt)); | |
547b792c IJ |
5242 | WARN_ON(dev->ip_ptr); |
5243 | WARN_ON(dev->ip6_ptr); | |
5244 | WARN_ON(dev->dn_ptr); | |
1da177e4 | 5245 | |
b17a7c17 SH |
5246 | if (dev->destructor) |
5247 | dev->destructor(dev); | |
9093bbb2 SH |
5248 | |
5249 | /* Free network device */ | |
5250 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5251 | } |
1da177e4 LT |
5252 | } |
5253 | ||
d83345ad ED |
5254 | /** |
5255 | * dev_txq_stats_fold - fold tx_queues stats | |
5256 | * @dev: device to get statistics from | |
5257 | * @stats: struct net_device_stats to hold results | |
5258 | */ | |
5259 | void dev_txq_stats_fold(const struct net_device *dev, | |
5260 | struct net_device_stats *stats) | |
5261 | { | |
5262 | unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0; | |
5263 | unsigned int i; | |
5264 | struct netdev_queue *txq; | |
5265 | ||
5266 | for (i = 0; i < dev->num_tx_queues; i++) { | |
5267 | txq = netdev_get_tx_queue(dev, i); | |
5268 | tx_bytes += txq->tx_bytes; | |
5269 | tx_packets += txq->tx_packets; | |
5270 | tx_dropped += txq->tx_dropped; | |
5271 | } | |
5272 | if (tx_bytes || tx_packets || tx_dropped) { | |
5273 | stats->tx_bytes = tx_bytes; | |
5274 | stats->tx_packets = tx_packets; | |
5275 | stats->tx_dropped = tx_dropped; | |
5276 | } | |
5277 | } | |
5278 | EXPORT_SYMBOL(dev_txq_stats_fold); | |
5279 | ||
eeda3fd6 SH |
5280 | /** |
5281 | * dev_get_stats - get network device statistics | |
5282 | * @dev: device to get statistics from | |
5283 | * | |
5284 | * Get network statistics from device. The device driver may provide | |
be1f3c2c BH |
5285 | * its own method by setting dev->netdev_ops->get_stats64 or |
5286 | * dev->netdev_ops->get_stats; otherwise the internal statistics | |
5287 | * structure is used. | |
eeda3fd6 | 5288 | */ |
be1f3c2c | 5289 | const struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev) |
7004bf25 | 5290 | { |
eeda3fd6 SH |
5291 | const struct net_device_ops *ops = dev->netdev_ops; |
5292 | ||
be1f3c2c BH |
5293 | if (ops->ndo_get_stats64) |
5294 | return ops->ndo_get_stats64(dev); | |
eeda3fd6 | 5295 | if (ops->ndo_get_stats) |
be1f3c2c | 5296 | return (struct rtnl_link_stats64 *)ops->ndo_get_stats(dev); |
d83345ad ED |
5297 | |
5298 | dev_txq_stats_fold(dev, &dev->stats); | |
be1f3c2c | 5299 | return &dev->stats64; |
c45d286e | 5300 | } |
eeda3fd6 | 5301 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5302 | |
dc2b4847 | 5303 | static void netdev_init_one_queue(struct net_device *dev, |
e8a0464c DM |
5304 | struct netdev_queue *queue, |
5305 | void *_unused) | |
dc2b4847 | 5306 | { |
dc2b4847 DM |
5307 | queue->dev = dev; |
5308 | } | |
5309 | ||
bb949fbd DM |
5310 | static void netdev_init_queues(struct net_device *dev) |
5311 | { | |
e8a0464c DM |
5312 | netdev_init_one_queue(dev, &dev->rx_queue, NULL); |
5313 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); | |
c3f26a26 | 5314 | spin_lock_init(&dev->tx_global_lock); |
bb949fbd DM |
5315 | } |
5316 | ||
1da177e4 | 5317 | /** |
f25f4e44 | 5318 | * alloc_netdev_mq - allocate network device |
1da177e4 LT |
5319 | * @sizeof_priv: size of private data to allocate space for |
5320 | * @name: device name format string | |
5321 | * @setup: callback to initialize device | |
f25f4e44 | 5322 | * @queue_count: the number of subqueues to allocate |
1da177e4 LT |
5323 | * |
5324 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 PWJ |
5325 | * and performs basic initialization. Also allocates subquue structs |
5326 | * for each queue on the device at the end of the netdevice. | |
1da177e4 | 5327 | */ |
f25f4e44 PWJ |
5328 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, |
5329 | void (*setup)(struct net_device *), unsigned int queue_count) | |
1da177e4 | 5330 | { |
e8a0464c | 5331 | struct netdev_queue *tx; |
1da177e4 | 5332 | struct net_device *dev; |
7943986c | 5333 | size_t alloc_size; |
1ce8e7b5 | 5334 | struct net_device *p; |
df334545 ED |
5335 | #ifdef CONFIG_RPS |
5336 | struct netdev_rx_queue *rx; | |
0a9627f2 | 5337 | int i; |
df334545 | 5338 | #endif |
1da177e4 | 5339 | |
b6fe17d6 SH |
5340 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5341 | ||
fd2ea0a7 | 5342 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5343 | if (sizeof_priv) { |
5344 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5345 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5346 | alloc_size += sizeof_priv; |
5347 | } | |
5348 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5349 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5350 | |
31380de9 | 5351 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 5352 | if (!p) { |
b6fe17d6 | 5353 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
5354 | return NULL; |
5355 | } | |
1da177e4 | 5356 | |
7943986c | 5357 | tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL); |
e8a0464c DM |
5358 | if (!tx) { |
5359 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5360 | "tx qdiscs.\n"); | |
ab9c73cc | 5361 | goto free_p; |
e8a0464c DM |
5362 | } |
5363 | ||
df334545 | 5364 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5365 | rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
5366 | if (!rx) { | |
5367 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5368 | "rx queues.\n"); | |
5369 | goto free_tx; | |
5370 | } | |
5371 | ||
5372 | atomic_set(&rx->count, queue_count); | |
5373 | ||
5374 | /* | |
5375 | * Set a pointer to first element in the array which holds the | |
5376 | * reference count. | |
5377 | */ | |
5378 | for (i = 0; i < queue_count; i++) | |
5379 | rx[i].first = rx; | |
df334545 | 5380 | #endif |
0a9627f2 | 5381 | |
1ce8e7b5 | 5382 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5383 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc JP |
5384 | |
5385 | if (dev_addr_init(dev)) | |
0a9627f2 | 5386 | goto free_rx; |
ab9c73cc | 5387 | |
22bedad3 | 5388 | dev_mc_init(dev); |
a748ee24 | 5389 | dev_uc_init(dev); |
ccffad25 | 5390 | |
c346dca1 | 5391 | dev_net_set(dev, &init_net); |
1da177e4 | 5392 | |
e8a0464c DM |
5393 | dev->_tx = tx; |
5394 | dev->num_tx_queues = queue_count; | |
fd2ea0a7 | 5395 | dev->real_num_tx_queues = queue_count; |
e8a0464c | 5396 | |
df334545 | 5397 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5398 | dev->_rx = rx; |
5399 | dev->num_rx_queues = queue_count; | |
df334545 | 5400 | #endif |
0a9627f2 | 5401 | |
82cc1a7a | 5402 | dev->gso_max_size = GSO_MAX_SIZE; |
1da177e4 | 5403 | |
bb949fbd DM |
5404 | netdev_init_queues(dev); |
5405 | ||
15682bc4 PWJ |
5406 | INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list); |
5407 | dev->ethtool_ntuple_list.count = 0; | |
d565b0a1 | 5408 | INIT_LIST_HEAD(&dev->napi_list); |
9fdce099 | 5409 | INIT_LIST_HEAD(&dev->unreg_list); |
e014debe | 5410 | INIT_LIST_HEAD(&dev->link_watch_list); |
93f154b5 | 5411 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
1da177e4 LT |
5412 | setup(dev); |
5413 | strcpy(dev->name, name); | |
5414 | return dev; | |
ab9c73cc | 5415 | |
0a9627f2 | 5416 | free_rx: |
df334545 | 5417 | #ifdef CONFIG_RPS |
0a9627f2 | 5418 | kfree(rx); |
ab9c73cc | 5419 | free_tx: |
df334545 | 5420 | #endif |
ab9c73cc | 5421 | kfree(tx); |
ab9c73cc JP |
5422 | free_p: |
5423 | kfree(p); | |
5424 | return NULL; | |
1da177e4 | 5425 | } |
f25f4e44 | 5426 | EXPORT_SYMBOL(alloc_netdev_mq); |
1da177e4 LT |
5427 | |
5428 | /** | |
5429 | * free_netdev - free network device | |
5430 | * @dev: device | |
5431 | * | |
4ec93edb YH |
5432 | * This function does the last stage of destroying an allocated device |
5433 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5434 | * If this is the last reference then it will be freed. |
5435 | */ | |
5436 | void free_netdev(struct net_device *dev) | |
5437 | { | |
d565b0a1 HX |
5438 | struct napi_struct *p, *n; |
5439 | ||
f3005d7f DL |
5440 | release_net(dev_net(dev)); |
5441 | ||
e8a0464c DM |
5442 | kfree(dev->_tx); |
5443 | ||
f001fde5 JP |
5444 | /* Flush device addresses */ |
5445 | dev_addr_flush(dev); | |
5446 | ||
15682bc4 PWJ |
5447 | /* Clear ethtool n-tuple list */ |
5448 | ethtool_ntuple_flush(dev); | |
5449 | ||
d565b0a1 HX |
5450 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5451 | netif_napi_del(p); | |
5452 | ||
3041a069 | 5453 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5454 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5455 | kfree((char *)dev - dev->padded); | |
5456 | return; | |
5457 | } | |
5458 | ||
5459 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5460 | dev->reg_state = NETREG_RELEASED; | |
5461 | ||
43cb76d9 GKH |
5462 | /* will free via device release */ |
5463 | put_device(&dev->dev); | |
1da177e4 | 5464 | } |
d1b19dff | 5465 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 5466 | |
f0db275a SH |
5467 | /** |
5468 | * synchronize_net - Synchronize with packet receive processing | |
5469 | * | |
5470 | * Wait for packets currently being received to be done. | |
5471 | * Does not block later packets from starting. | |
5472 | */ | |
4ec93edb | 5473 | void synchronize_net(void) |
1da177e4 LT |
5474 | { |
5475 | might_sleep(); | |
fbd568a3 | 5476 | synchronize_rcu(); |
1da177e4 | 5477 | } |
d1b19dff | 5478 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
5479 | |
5480 | /** | |
44a0873d | 5481 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 5482 | * @dev: device |
44a0873d | 5483 | * @head: list |
6ebfbc06 | 5484 | * |
1da177e4 | 5485 | * This function shuts down a device interface and removes it |
d59b54b1 | 5486 | * from the kernel tables. |
44a0873d | 5487 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
5488 | * |
5489 | * Callers must hold the rtnl semaphore. You may want | |
5490 | * unregister_netdev() instead of this. | |
5491 | */ | |
5492 | ||
44a0873d | 5493 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 5494 | { |
a6620712 HX |
5495 | ASSERT_RTNL(); |
5496 | ||
44a0873d | 5497 | if (head) { |
9fdce099 | 5498 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
5499 | } else { |
5500 | rollback_registered(dev); | |
5501 | /* Finish processing unregister after unlock */ | |
5502 | net_set_todo(dev); | |
5503 | } | |
1da177e4 | 5504 | } |
44a0873d | 5505 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 5506 | |
9b5e383c ED |
5507 | /** |
5508 | * unregister_netdevice_many - unregister many devices | |
5509 | * @head: list of devices | |
9b5e383c ED |
5510 | */ |
5511 | void unregister_netdevice_many(struct list_head *head) | |
5512 | { | |
5513 | struct net_device *dev; | |
5514 | ||
5515 | if (!list_empty(head)) { | |
5516 | rollback_registered_many(head); | |
5517 | list_for_each_entry(dev, head, unreg_list) | |
5518 | net_set_todo(dev); | |
5519 | } | |
5520 | } | |
63c8099d | 5521 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 5522 | |
1da177e4 LT |
5523 | /** |
5524 | * unregister_netdev - remove device from the kernel | |
5525 | * @dev: device | |
5526 | * | |
5527 | * This function shuts down a device interface and removes it | |
d59b54b1 | 5528 | * from the kernel tables. |
1da177e4 LT |
5529 | * |
5530 | * This is just a wrapper for unregister_netdevice that takes | |
5531 | * the rtnl semaphore. In general you want to use this and not | |
5532 | * unregister_netdevice. | |
5533 | */ | |
5534 | void unregister_netdev(struct net_device *dev) | |
5535 | { | |
5536 | rtnl_lock(); | |
5537 | unregister_netdevice(dev); | |
5538 | rtnl_unlock(); | |
5539 | } | |
1da177e4 LT |
5540 | EXPORT_SYMBOL(unregister_netdev); |
5541 | ||
ce286d32 EB |
5542 | /** |
5543 | * dev_change_net_namespace - move device to different nethost namespace | |
5544 | * @dev: device | |
5545 | * @net: network namespace | |
5546 | * @pat: If not NULL name pattern to try if the current device name | |
5547 | * is already taken in the destination network namespace. | |
5548 | * | |
5549 | * This function shuts down a device interface and moves it | |
5550 | * to a new network namespace. On success 0 is returned, on | |
5551 | * a failure a netagive errno code is returned. | |
5552 | * | |
5553 | * Callers must hold the rtnl semaphore. | |
5554 | */ | |
5555 | ||
5556 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5557 | { | |
ce286d32 EB |
5558 | int err; |
5559 | ||
5560 | ASSERT_RTNL(); | |
5561 | ||
5562 | /* Don't allow namespace local devices to be moved. */ | |
5563 | err = -EINVAL; | |
5564 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5565 | goto out; | |
5566 | ||
5567 | /* Ensure the device has been registrered */ | |
5568 | err = -EINVAL; | |
5569 | if (dev->reg_state != NETREG_REGISTERED) | |
5570 | goto out; | |
5571 | ||
5572 | /* Get out if there is nothing todo */ | |
5573 | err = 0; | |
878628fb | 5574 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
5575 | goto out; |
5576 | ||
5577 | /* Pick the destination device name, and ensure | |
5578 | * we can use it in the destination network namespace. | |
5579 | */ | |
5580 | err = -EEXIST; | |
d9031024 | 5581 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
5582 | /* We get here if we can't use the current device name */ |
5583 | if (!pat) | |
5584 | goto out; | |
8ce6cebc | 5585 | if (dev_get_valid_name(dev, pat, 1)) |
ce286d32 EB |
5586 | goto out; |
5587 | } | |
5588 | ||
5589 | /* | |
5590 | * And now a mini version of register_netdevice unregister_netdevice. | |
5591 | */ | |
5592 | ||
5593 | /* If device is running close it first. */ | |
9b772652 | 5594 | dev_close(dev); |
ce286d32 EB |
5595 | |
5596 | /* And unlink it from device chain */ | |
5597 | err = -ENODEV; | |
5598 | unlist_netdevice(dev); | |
5599 | ||
5600 | synchronize_net(); | |
5601 | ||
5602 | /* Shutdown queueing discipline. */ | |
5603 | dev_shutdown(dev); | |
5604 | ||
5605 | /* Notify protocols, that we are about to destroy | |
5606 | this device. They should clean all the things. | |
5607 | */ | |
5608 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
a5ee1551 | 5609 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
ce286d32 EB |
5610 | |
5611 | /* | |
5612 | * Flush the unicast and multicast chains | |
5613 | */ | |
a748ee24 | 5614 | dev_uc_flush(dev); |
22bedad3 | 5615 | dev_mc_flush(dev); |
ce286d32 EB |
5616 | |
5617 | /* Actually switch the network namespace */ | |
c346dca1 | 5618 | dev_net_set(dev, net); |
ce286d32 | 5619 | |
ce286d32 EB |
5620 | /* If there is an ifindex conflict assign a new one */ |
5621 | if (__dev_get_by_index(net, dev->ifindex)) { | |
5622 | int iflink = (dev->iflink == dev->ifindex); | |
5623 | dev->ifindex = dev_new_index(net); | |
5624 | if (iflink) | |
5625 | dev->iflink = dev->ifindex; | |
5626 | } | |
5627 | ||
8b41d188 | 5628 | /* Fixup kobjects */ |
a1b3f594 | 5629 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 5630 | WARN_ON(err); |
ce286d32 EB |
5631 | |
5632 | /* Add the device back in the hashes */ | |
5633 | list_netdevice(dev); | |
5634 | ||
5635 | /* Notify protocols, that a new device appeared. */ | |
5636 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5637 | ||
d90a909e EB |
5638 | /* |
5639 | * Prevent userspace races by waiting until the network | |
5640 | * device is fully setup before sending notifications. | |
5641 | */ | |
5642 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5643 | ||
ce286d32 EB |
5644 | synchronize_net(); |
5645 | err = 0; | |
5646 | out: | |
5647 | return err; | |
5648 | } | |
463d0183 | 5649 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 5650 | |
1da177e4 LT |
5651 | static int dev_cpu_callback(struct notifier_block *nfb, |
5652 | unsigned long action, | |
5653 | void *ocpu) | |
5654 | { | |
5655 | struct sk_buff **list_skb; | |
1da177e4 LT |
5656 | struct sk_buff *skb; |
5657 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
5658 | struct softnet_data *sd, *oldsd; | |
5659 | ||
8bb78442 | 5660 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
5661 | return NOTIFY_OK; |
5662 | ||
5663 | local_irq_disable(); | |
5664 | cpu = smp_processor_id(); | |
5665 | sd = &per_cpu(softnet_data, cpu); | |
5666 | oldsd = &per_cpu(softnet_data, oldcpu); | |
5667 | ||
5668 | /* Find end of our completion_queue. */ | |
5669 | list_skb = &sd->completion_queue; | |
5670 | while (*list_skb) | |
5671 | list_skb = &(*list_skb)->next; | |
5672 | /* Append completion queue from offline CPU. */ | |
5673 | *list_skb = oldsd->completion_queue; | |
5674 | oldsd->completion_queue = NULL; | |
5675 | ||
1da177e4 | 5676 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
5677 | if (oldsd->output_queue) { |
5678 | *sd->output_queue_tailp = oldsd->output_queue; | |
5679 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
5680 | oldsd->output_queue = NULL; | |
5681 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
5682 | } | |
1da177e4 LT |
5683 | |
5684 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
5685 | local_irq_enable(); | |
5686 | ||
5687 | /* Process offline CPU's input_pkt_queue */ | |
76cc8b13 | 5688 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
1da177e4 | 5689 | netif_rx(skb); |
76cc8b13 | 5690 | input_queue_head_incr(oldsd); |
fec5e652 | 5691 | } |
76cc8b13 | 5692 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
6e7676c1 | 5693 | netif_rx(skb); |
76cc8b13 TH |
5694 | input_queue_head_incr(oldsd); |
5695 | } | |
1da177e4 LT |
5696 | |
5697 | return NOTIFY_OK; | |
5698 | } | |
1da177e4 LT |
5699 | |
5700 | ||
7f353bf2 | 5701 | /** |
b63365a2 HX |
5702 | * netdev_increment_features - increment feature set by one |
5703 | * @all: current feature set | |
5704 | * @one: new feature set | |
5705 | * @mask: mask feature set | |
7f353bf2 HX |
5706 | * |
5707 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
5708 | * @one to the master device with current feature set @all. Will not |
5709 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 5710 | */ |
b63365a2 HX |
5711 | unsigned long netdev_increment_features(unsigned long all, unsigned long one, |
5712 | unsigned long mask) | |
5713 | { | |
5714 | /* If device needs checksumming, downgrade to it. */ | |
d1b19dff | 5715 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) |
b63365a2 HX |
5716 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); |
5717 | else if (mask & NETIF_F_ALL_CSUM) { | |
5718 | /* If one device supports v4/v6 checksumming, set for all. */ | |
5719 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && | |
5720 | !(all & NETIF_F_GEN_CSUM)) { | |
5721 | all &= ~NETIF_F_ALL_CSUM; | |
5722 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
5723 | } | |
e2a6b852 | 5724 | |
b63365a2 HX |
5725 | /* If one device supports hw checksumming, set for all. */ |
5726 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { | |
5727 | all &= ~NETIF_F_ALL_CSUM; | |
5728 | all |= NETIF_F_HW_CSUM; | |
5729 | } | |
5730 | } | |
7f353bf2 | 5731 | |
b63365a2 | 5732 | one |= NETIF_F_ALL_CSUM; |
7f353bf2 | 5733 | |
b63365a2 | 5734 | one |= all & NETIF_F_ONE_FOR_ALL; |
d9f5950f | 5735 | all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO; |
b63365a2 | 5736 | all |= one & mask & NETIF_F_ONE_FOR_ALL; |
7f353bf2 HX |
5737 | |
5738 | return all; | |
5739 | } | |
b63365a2 | 5740 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 5741 | |
30d97d35 PE |
5742 | static struct hlist_head *netdev_create_hash(void) |
5743 | { | |
5744 | int i; | |
5745 | struct hlist_head *hash; | |
5746 | ||
5747 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
5748 | if (hash != NULL) | |
5749 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
5750 | INIT_HLIST_HEAD(&hash[i]); | |
5751 | ||
5752 | return hash; | |
5753 | } | |
5754 | ||
881d966b | 5755 | /* Initialize per network namespace state */ |
4665079c | 5756 | static int __net_init netdev_init(struct net *net) |
881d966b | 5757 | { |
881d966b | 5758 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 5759 | |
30d97d35 PE |
5760 | net->dev_name_head = netdev_create_hash(); |
5761 | if (net->dev_name_head == NULL) | |
5762 | goto err_name; | |
881d966b | 5763 | |
30d97d35 PE |
5764 | net->dev_index_head = netdev_create_hash(); |
5765 | if (net->dev_index_head == NULL) | |
5766 | goto err_idx; | |
881d966b EB |
5767 | |
5768 | return 0; | |
30d97d35 PE |
5769 | |
5770 | err_idx: | |
5771 | kfree(net->dev_name_head); | |
5772 | err_name: | |
5773 | return -ENOMEM; | |
881d966b EB |
5774 | } |
5775 | ||
f0db275a SH |
5776 | /** |
5777 | * netdev_drivername - network driver for the device | |
5778 | * @dev: network device | |
5779 | * @buffer: buffer for resulting name | |
5780 | * @len: size of buffer | |
5781 | * | |
5782 | * Determine network driver for device. | |
5783 | */ | |
cf04a4c7 | 5784 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) |
6579e57b | 5785 | { |
cf04a4c7 SH |
5786 | const struct device_driver *driver; |
5787 | const struct device *parent; | |
6579e57b AV |
5788 | |
5789 | if (len <= 0 || !buffer) | |
5790 | return buffer; | |
5791 | buffer[0] = 0; | |
5792 | ||
5793 | parent = dev->dev.parent; | |
5794 | ||
5795 | if (!parent) | |
5796 | return buffer; | |
5797 | ||
5798 | driver = parent->driver; | |
5799 | if (driver && driver->name) | |
5800 | strlcpy(buffer, driver->name, len); | |
5801 | return buffer; | |
5802 | } | |
5803 | ||
4665079c | 5804 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
5805 | { |
5806 | kfree(net->dev_name_head); | |
5807 | kfree(net->dev_index_head); | |
5808 | } | |
5809 | ||
022cbae6 | 5810 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
5811 | .init = netdev_init, |
5812 | .exit = netdev_exit, | |
5813 | }; | |
5814 | ||
4665079c | 5815 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 5816 | { |
e008b5fc | 5817 | struct net_device *dev, *aux; |
ce286d32 | 5818 | /* |
e008b5fc | 5819 | * Push all migratable network devices back to the |
ce286d32 EB |
5820 | * initial network namespace |
5821 | */ | |
5822 | rtnl_lock(); | |
e008b5fc | 5823 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 5824 | int err; |
aca51397 | 5825 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
5826 | |
5827 | /* Ignore unmoveable devices (i.e. loopback) */ | |
5828 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5829 | continue; | |
5830 | ||
e008b5fc EB |
5831 | /* Leave virtual devices for the generic cleanup */ |
5832 | if (dev->rtnl_link_ops) | |
5833 | continue; | |
d0c082ce | 5834 | |
ce286d32 | 5835 | /* Push remaing network devices to init_net */ |
aca51397 PE |
5836 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
5837 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 5838 | if (err) { |
aca51397 | 5839 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 5840 | __func__, dev->name, err); |
aca51397 | 5841 | BUG(); |
ce286d32 EB |
5842 | } |
5843 | } | |
5844 | rtnl_unlock(); | |
5845 | } | |
5846 | ||
04dc7f6b EB |
5847 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
5848 | { | |
5849 | /* At exit all network devices most be removed from a network | |
5850 | * namespace. Do this in the reverse order of registeration. | |
5851 | * Do this across as many network namespaces as possible to | |
5852 | * improve batching efficiency. | |
5853 | */ | |
5854 | struct net_device *dev; | |
5855 | struct net *net; | |
5856 | LIST_HEAD(dev_kill_list); | |
5857 | ||
5858 | rtnl_lock(); | |
5859 | list_for_each_entry(net, net_list, exit_list) { | |
5860 | for_each_netdev_reverse(net, dev) { | |
5861 | if (dev->rtnl_link_ops) | |
5862 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
5863 | else | |
5864 | unregister_netdevice_queue(dev, &dev_kill_list); | |
5865 | } | |
5866 | } | |
5867 | unregister_netdevice_many(&dev_kill_list); | |
5868 | rtnl_unlock(); | |
5869 | } | |
5870 | ||
022cbae6 | 5871 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 5872 | .exit = default_device_exit, |
04dc7f6b | 5873 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
5874 | }; |
5875 | ||
1da177e4 LT |
5876 | /* |
5877 | * Initialize the DEV module. At boot time this walks the device list and | |
5878 | * unhooks any devices that fail to initialise (normally hardware not | |
5879 | * present) and leaves us with a valid list of present and active devices. | |
5880 | * | |
5881 | */ | |
5882 | ||
5883 | /* | |
5884 | * This is called single threaded during boot, so no need | |
5885 | * to take the rtnl semaphore. | |
5886 | */ | |
5887 | static int __init net_dev_init(void) | |
5888 | { | |
5889 | int i, rc = -ENOMEM; | |
5890 | ||
5891 | BUG_ON(!dev_boot_phase); | |
5892 | ||
1da177e4 LT |
5893 | if (dev_proc_init()) |
5894 | goto out; | |
5895 | ||
8b41d188 | 5896 | if (netdev_kobject_init()) |
1da177e4 LT |
5897 | goto out; |
5898 | ||
5899 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 5900 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
5901 | INIT_LIST_HEAD(&ptype_base[i]); |
5902 | ||
881d966b EB |
5903 | if (register_pernet_subsys(&netdev_net_ops)) |
5904 | goto out; | |
1da177e4 LT |
5905 | |
5906 | /* | |
5907 | * Initialise the packet receive queues. | |
5908 | */ | |
5909 | ||
6f912042 | 5910 | for_each_possible_cpu(i) { |
e36fa2f7 | 5911 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 5912 | |
dee42870 | 5913 | memset(sd, 0, sizeof(*sd)); |
e36fa2f7 | 5914 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 5915 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 ED |
5916 | sd->completion_queue = NULL; |
5917 | INIT_LIST_HEAD(&sd->poll_list); | |
a9cbd588 CG |
5918 | sd->output_queue = NULL; |
5919 | sd->output_queue_tailp = &sd->output_queue; | |
df334545 | 5920 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
5921 | sd->csd.func = rps_trigger_softirq; |
5922 | sd->csd.info = sd; | |
5923 | sd->csd.flags = 0; | |
5924 | sd->cpu = i; | |
1e94d72f | 5925 | #endif |
0a9627f2 | 5926 | |
e36fa2f7 ED |
5927 | sd->backlog.poll = process_backlog; |
5928 | sd->backlog.weight = weight_p; | |
5929 | sd->backlog.gro_list = NULL; | |
5930 | sd->backlog.gro_count = 0; | |
1da177e4 LT |
5931 | } |
5932 | ||
1da177e4 LT |
5933 | dev_boot_phase = 0; |
5934 | ||
505d4f73 EB |
5935 | /* The loopback device is special if any other network devices |
5936 | * is present in a network namespace the loopback device must | |
5937 | * be present. Since we now dynamically allocate and free the | |
5938 | * loopback device ensure this invariant is maintained by | |
5939 | * keeping the loopback device as the first device on the | |
5940 | * list of network devices. Ensuring the loopback devices | |
5941 | * is the first device that appears and the last network device | |
5942 | * that disappears. | |
5943 | */ | |
5944 | if (register_pernet_device(&loopback_net_ops)) | |
5945 | goto out; | |
5946 | ||
5947 | if (register_pernet_device(&default_device_ops)) | |
5948 | goto out; | |
5949 | ||
962cf36c CM |
5950 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
5951 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
5952 | |
5953 | hotcpu_notifier(dev_cpu_callback, 0); | |
5954 | dst_init(); | |
5955 | dev_mcast_init(); | |
5956 | rc = 0; | |
5957 | out: | |
5958 | return rc; | |
5959 | } | |
5960 | ||
5961 | subsys_initcall(net_dev_init); | |
5962 | ||
e88721f8 KK |
5963 | static int __init initialize_hashrnd(void) |
5964 | { | |
0a9627f2 | 5965 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
5966 | return 0; |
5967 | } | |
5968 | ||
5969 | late_initcall_sync(initialize_hashrnd); | |
5970 |