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 | ||
152102c7 CG |
211 | static inline void rps_lock(struct softnet_data *queue) |
212 | { | |
213 | #ifdef CONFIG_RPS | |
214 | spin_lock(&queue->input_pkt_queue.lock); | |
215 | #endif | |
216 | } | |
217 | ||
218 | static inline void rps_unlock(struct softnet_data *queue) | |
219 | { | |
220 | #ifdef CONFIG_RPS | |
221 | spin_unlock(&queue->input_pkt_queue.lock); | |
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 SH |
266 | |
267 | DEFINE_PER_CPU(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 | /** | |
806 | * dev_get_by_flags - 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 | |
4ec93edb | 812 | * is not found or a pointer to the device. The device returned has |
1da177e4 LT |
813 | * had a reference added and the pointer is safe until the user calls |
814 | * dev_put to indicate they have finished with it. | |
815 | */ | |
816 | ||
d1b19dff ED |
817 | struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags, |
818 | unsigned short mask) | |
1da177e4 | 819 | { |
7562f876 | 820 | struct net_device *dev, *ret; |
1da177e4 | 821 | |
7562f876 | 822 | ret = NULL; |
c6d14c84 ED |
823 | rcu_read_lock(); |
824 | for_each_netdev_rcu(net, dev) { | |
1da177e4 LT |
825 | if (((dev->flags ^ if_flags) & mask) == 0) { |
826 | dev_hold(dev); | |
7562f876 | 827 | ret = dev; |
1da177e4 LT |
828 | break; |
829 | } | |
830 | } | |
c6d14c84 | 831 | rcu_read_unlock(); |
7562f876 | 832 | return ret; |
1da177e4 | 833 | } |
d1b19dff | 834 | EXPORT_SYMBOL(dev_get_by_flags); |
1da177e4 LT |
835 | |
836 | /** | |
837 | * dev_valid_name - check if name is okay for network device | |
838 | * @name: name string | |
839 | * | |
840 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
841 | * to allow sysfs to work. We also disallow any kind of |
842 | * whitespace. | |
1da177e4 | 843 | */ |
c2373ee9 | 844 | int dev_valid_name(const char *name) |
1da177e4 | 845 | { |
c7fa9d18 DM |
846 | if (*name == '\0') |
847 | return 0; | |
b6fe17d6 SH |
848 | if (strlen(name) >= IFNAMSIZ) |
849 | return 0; | |
c7fa9d18 DM |
850 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
851 | return 0; | |
852 | ||
853 | while (*name) { | |
854 | if (*name == '/' || isspace(*name)) | |
855 | return 0; | |
856 | name++; | |
857 | } | |
858 | return 1; | |
1da177e4 | 859 | } |
d1b19dff | 860 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
861 | |
862 | /** | |
b267b179 EB |
863 | * __dev_alloc_name - allocate a name for a device |
864 | * @net: network namespace to allocate the device name in | |
1da177e4 | 865 | * @name: name format string |
b267b179 | 866 | * @buf: scratch buffer and result name string |
1da177e4 LT |
867 | * |
868 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
869 | * id. It scans list of devices to build up a free map, then chooses |
870 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
871 | * while allocating the name and adding the device in order to avoid | |
872 | * duplicates. | |
873 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
874 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
875 | */ |
876 | ||
b267b179 | 877 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
878 | { |
879 | int i = 0; | |
1da177e4 LT |
880 | const char *p; |
881 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 882 | unsigned long *inuse; |
1da177e4 LT |
883 | struct net_device *d; |
884 | ||
885 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
886 | if (p) { | |
887 | /* | |
888 | * Verify the string as this thing may have come from | |
889 | * the user. There must be either one "%d" and no other "%" | |
890 | * characters. | |
891 | */ | |
892 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
893 | return -EINVAL; | |
894 | ||
895 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 896 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
897 | if (!inuse) |
898 | return -ENOMEM; | |
899 | ||
881d966b | 900 | for_each_netdev(net, d) { |
1da177e4 LT |
901 | if (!sscanf(d->name, name, &i)) |
902 | continue; | |
903 | if (i < 0 || i >= max_netdevices) | |
904 | continue; | |
905 | ||
906 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 907 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
908 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
909 | set_bit(i, inuse); | |
910 | } | |
911 | ||
912 | i = find_first_zero_bit(inuse, max_netdevices); | |
913 | free_page((unsigned long) inuse); | |
914 | } | |
915 | ||
d9031024 OP |
916 | if (buf != name) |
917 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 918 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 919 | return i; |
1da177e4 LT |
920 | |
921 | /* It is possible to run out of possible slots | |
922 | * when the name is long and there isn't enough space left | |
923 | * for the digits, or if all bits are used. | |
924 | */ | |
925 | return -ENFILE; | |
926 | } | |
927 | ||
b267b179 EB |
928 | /** |
929 | * dev_alloc_name - allocate a name for a device | |
930 | * @dev: device | |
931 | * @name: name format string | |
932 | * | |
933 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
934 | * id. It scans list of devices to build up a free map, then chooses | |
935 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
936 | * while allocating the name and adding the device in order to avoid | |
937 | * duplicates. | |
938 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
939 | * Returns the number of the unit assigned or a negative errno code. | |
940 | */ | |
941 | ||
942 | int dev_alloc_name(struct net_device *dev, const char *name) | |
943 | { | |
944 | char buf[IFNAMSIZ]; | |
945 | struct net *net; | |
946 | int ret; | |
947 | ||
c346dca1 YH |
948 | BUG_ON(!dev_net(dev)); |
949 | net = dev_net(dev); | |
b267b179 EB |
950 | ret = __dev_alloc_name(net, name, buf); |
951 | if (ret >= 0) | |
952 | strlcpy(dev->name, buf, IFNAMSIZ); | |
953 | return ret; | |
954 | } | |
d1b19dff | 955 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 956 | |
d9031024 OP |
957 | static int dev_get_valid_name(struct net *net, const char *name, char *buf, |
958 | bool fmt) | |
959 | { | |
960 | if (!dev_valid_name(name)) | |
961 | return -EINVAL; | |
962 | ||
963 | if (fmt && strchr(name, '%')) | |
964 | return __dev_alloc_name(net, name, buf); | |
965 | else if (__dev_get_by_name(net, name)) | |
966 | return -EEXIST; | |
967 | else if (buf != name) | |
968 | strlcpy(buf, name, IFNAMSIZ); | |
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 | ||
d9031024 OP |
1000 | err = dev_get_valid_name(net, newname, dev->name, 1); |
1001 | if (err < 0) | |
1002 | return err; | |
1da177e4 | 1003 | |
fcc5a03a | 1004 | rollback: |
3891845e EB |
1005 | /* For now only devices in the initial network namespace |
1006 | * are in sysfs. | |
1007 | */ | |
09ad9bc7 | 1008 | if (net_eq(net, &init_net)) { |
3891845e EB |
1009 | ret = device_rename(&dev->dev, dev->name); |
1010 | if (ret) { | |
1011 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1012 | return ret; | |
1013 | } | |
dcc99773 | 1014 | } |
7f988eab HX |
1015 | |
1016 | write_lock_bh(&dev_base_lock); | |
92749821 | 1017 | hlist_del(&dev->name_hlist); |
72c9528b ED |
1018 | write_unlock_bh(&dev_base_lock); |
1019 | ||
1020 | synchronize_rcu(); | |
1021 | ||
1022 | write_lock_bh(&dev_base_lock); | |
1023 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1024 | write_unlock_bh(&dev_base_lock); |
1025 | ||
056925ab | 1026 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1027 | ret = notifier_to_errno(ret); |
1028 | ||
1029 | if (ret) { | |
91e9c07b ED |
1030 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1031 | if (err >= 0) { | |
fcc5a03a HX |
1032 | err = ret; |
1033 | memcpy(dev->name, oldname, IFNAMSIZ); | |
1034 | goto rollback; | |
91e9c07b ED |
1035 | } else { |
1036 | printk(KERN_ERR | |
1037 | "%s: name change rollback failed: %d.\n", | |
1038 | dev->name, ret); | |
fcc5a03a HX |
1039 | } |
1040 | } | |
1da177e4 LT |
1041 | |
1042 | return err; | |
1043 | } | |
1044 | ||
0b815a1a SH |
1045 | /** |
1046 | * dev_set_alias - change ifalias of a device | |
1047 | * @dev: device | |
1048 | * @alias: name up to IFALIASZ | |
f0db275a | 1049 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1050 | * |
1051 | * Set ifalias for a device, | |
1052 | */ | |
1053 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1054 | { | |
1055 | ASSERT_RTNL(); | |
1056 | ||
1057 | if (len >= IFALIASZ) | |
1058 | return -EINVAL; | |
1059 | ||
96ca4a2c OH |
1060 | if (!len) { |
1061 | if (dev->ifalias) { | |
1062 | kfree(dev->ifalias); | |
1063 | dev->ifalias = NULL; | |
1064 | } | |
1065 | return 0; | |
1066 | } | |
1067 | ||
d1b19dff | 1068 | dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL); |
0b815a1a SH |
1069 | if (!dev->ifalias) |
1070 | return -ENOMEM; | |
1071 | ||
1072 | strlcpy(dev->ifalias, alias, len+1); | |
1073 | return len; | |
1074 | } | |
1075 | ||
1076 | ||
d8a33ac4 | 1077 | /** |
3041a069 | 1078 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1079 | * @dev: device to cause notification |
1080 | * | |
1081 | * Called to indicate a device has changed features. | |
1082 | */ | |
1083 | void netdev_features_change(struct net_device *dev) | |
1084 | { | |
056925ab | 1085 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1086 | } |
1087 | EXPORT_SYMBOL(netdev_features_change); | |
1088 | ||
1da177e4 LT |
1089 | /** |
1090 | * netdev_state_change - device changes state | |
1091 | * @dev: device to cause notification | |
1092 | * | |
1093 | * Called to indicate a device has changed state. This function calls | |
1094 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1095 | * to the routing socket. | |
1096 | */ | |
1097 | void netdev_state_change(struct net_device *dev) | |
1098 | { | |
1099 | if (dev->flags & IFF_UP) { | |
056925ab | 1100 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1101 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1102 | } | |
1103 | } | |
d1b19dff | 1104 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1105 | |
3ca5b404 | 1106 | int netdev_bonding_change(struct net_device *dev, unsigned long event) |
c1da4ac7 | 1107 | { |
3ca5b404 | 1108 | return call_netdevice_notifiers(event, dev); |
c1da4ac7 OG |
1109 | } |
1110 | EXPORT_SYMBOL(netdev_bonding_change); | |
1111 | ||
1da177e4 LT |
1112 | /** |
1113 | * dev_load - load a network module | |
c4ea43c5 | 1114 | * @net: the applicable net namespace |
1da177e4 LT |
1115 | * @name: name of interface |
1116 | * | |
1117 | * If a network interface is not present and the process has suitable | |
1118 | * privileges this function loads the module. If module loading is not | |
1119 | * available in this kernel then it becomes a nop. | |
1120 | */ | |
1121 | ||
881d966b | 1122 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1123 | { |
4ec93edb | 1124 | struct net_device *dev; |
1da177e4 | 1125 | |
72c9528b ED |
1126 | rcu_read_lock(); |
1127 | dev = dev_get_by_name_rcu(net, name); | |
1128 | rcu_read_unlock(); | |
1da177e4 | 1129 | |
a8f80e8f | 1130 | if (!dev && capable(CAP_NET_ADMIN)) |
1da177e4 LT |
1131 | request_module("%s", name); |
1132 | } | |
d1b19dff | 1133 | EXPORT_SYMBOL(dev_load); |
1da177e4 | 1134 | |
bd380811 | 1135 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1136 | { |
d314774c | 1137 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1138 | int ret; |
1da177e4 | 1139 | |
e46b66bc BH |
1140 | ASSERT_RTNL(); |
1141 | ||
1da177e4 LT |
1142 | /* |
1143 | * Is it even present? | |
1144 | */ | |
1145 | if (!netif_device_present(dev)) | |
1146 | return -ENODEV; | |
1147 | ||
3b8bcfd5 JB |
1148 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1149 | ret = notifier_to_errno(ret); | |
1150 | if (ret) | |
1151 | return ret; | |
1152 | ||
1da177e4 LT |
1153 | /* |
1154 | * Call device private open method | |
1155 | */ | |
1156 | set_bit(__LINK_STATE_START, &dev->state); | |
bada339b | 1157 | |
d314774c SH |
1158 | if (ops->ndo_validate_addr) |
1159 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1160 | |
d314774c SH |
1161 | if (!ret && ops->ndo_open) |
1162 | ret = ops->ndo_open(dev); | |
1da177e4 | 1163 | |
4ec93edb | 1164 | /* |
1da177e4 LT |
1165 | * If it went open OK then: |
1166 | */ | |
1167 | ||
bada339b JG |
1168 | if (ret) |
1169 | clear_bit(__LINK_STATE_START, &dev->state); | |
1170 | else { | |
1da177e4 LT |
1171 | /* |
1172 | * Set the flags. | |
1173 | */ | |
1174 | dev->flags |= IFF_UP; | |
1175 | ||
649274d9 DW |
1176 | /* |
1177 | * Enable NET_DMA | |
1178 | */ | |
b4bd07c2 | 1179 | net_dmaengine_get(); |
649274d9 | 1180 | |
1da177e4 LT |
1181 | /* |
1182 | * Initialize multicasting status | |
1183 | */ | |
4417da66 | 1184 | dev_set_rx_mode(dev); |
1da177e4 LT |
1185 | |
1186 | /* | |
1187 | * Wakeup transmit queue engine | |
1188 | */ | |
1189 | dev_activate(dev); | |
1da177e4 | 1190 | } |
bada339b | 1191 | |
1da177e4 LT |
1192 | return ret; |
1193 | } | |
1194 | ||
1195 | /** | |
bd380811 PM |
1196 | * dev_open - prepare an interface for use. |
1197 | * @dev: device to open | |
1da177e4 | 1198 | * |
bd380811 PM |
1199 | * Takes a device from down to up state. The device's private open |
1200 | * function is invoked and then the multicast lists are loaded. Finally | |
1201 | * the device is moved into the up state and a %NETDEV_UP message is | |
1202 | * sent to the netdev notifier chain. | |
1203 | * | |
1204 | * Calling this function on an active interface is a nop. On a failure | |
1205 | * a negative errno code is returned. | |
1da177e4 | 1206 | */ |
bd380811 PM |
1207 | int dev_open(struct net_device *dev) |
1208 | { | |
1209 | int ret; | |
1210 | ||
1211 | /* | |
1212 | * Is it already up? | |
1213 | */ | |
1214 | if (dev->flags & IFF_UP) | |
1215 | return 0; | |
1216 | ||
1217 | /* | |
1218 | * Open device | |
1219 | */ | |
1220 | ret = __dev_open(dev); | |
1221 | if (ret < 0) | |
1222 | return ret; | |
1223 | ||
1224 | /* | |
1225 | * ... and announce new interface. | |
1226 | */ | |
1227 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); | |
1228 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1229 | ||
1230 | return ret; | |
1231 | } | |
1232 | EXPORT_SYMBOL(dev_open); | |
1233 | ||
1234 | static int __dev_close(struct net_device *dev) | |
1da177e4 | 1235 | { |
d314774c | 1236 | const struct net_device_ops *ops = dev->netdev_ops; |
e46b66bc | 1237 | |
bd380811 | 1238 | ASSERT_RTNL(); |
9d5010db DM |
1239 | might_sleep(); |
1240 | ||
1da177e4 LT |
1241 | /* |
1242 | * Tell people we are going down, so that they can | |
1243 | * prepare to death, when device is still operating. | |
1244 | */ | |
056925ab | 1245 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1246 | |
1da177e4 LT |
1247 | clear_bit(__LINK_STATE_START, &dev->state); |
1248 | ||
1249 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
bea3348e SH |
1250 | * it can be even on different cpu. So just clear netif_running(). |
1251 | * | |
1252 | * dev->stop() will invoke napi_disable() on all of it's | |
1253 | * napi_struct instances on this device. | |
1254 | */ | |
1da177e4 | 1255 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ |
1da177e4 | 1256 | |
d8b2a4d2 ML |
1257 | dev_deactivate(dev); |
1258 | ||
1da177e4 LT |
1259 | /* |
1260 | * Call the device specific close. This cannot fail. | |
1261 | * Only if device is UP | |
1262 | * | |
1263 | * We allow it to be called even after a DETACH hot-plug | |
1264 | * event. | |
1265 | */ | |
d314774c SH |
1266 | if (ops->ndo_stop) |
1267 | ops->ndo_stop(dev); | |
1da177e4 LT |
1268 | |
1269 | /* | |
1270 | * Device is now down. | |
1271 | */ | |
1272 | ||
1273 | dev->flags &= ~IFF_UP; | |
1274 | ||
1275 | /* | |
bd380811 | 1276 | * Shutdown NET_DMA |
1da177e4 | 1277 | */ |
bd380811 PM |
1278 | net_dmaengine_put(); |
1279 | ||
1280 | return 0; | |
1281 | } | |
1282 | ||
1283 | /** | |
1284 | * dev_close - shutdown an interface. | |
1285 | * @dev: device to shutdown | |
1286 | * | |
1287 | * This function moves an active device into down state. A | |
1288 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1289 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1290 | * chain. | |
1291 | */ | |
1292 | int dev_close(struct net_device *dev) | |
1293 | { | |
1294 | if (!(dev->flags & IFF_UP)) | |
1295 | return 0; | |
1296 | ||
1297 | __dev_close(dev); | |
1da177e4 | 1298 | |
649274d9 | 1299 | /* |
bd380811 | 1300 | * Tell people we are down |
649274d9 | 1301 | */ |
bd380811 PM |
1302 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING); |
1303 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
649274d9 | 1304 | |
1da177e4 LT |
1305 | return 0; |
1306 | } | |
d1b19dff | 1307 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1308 | |
1309 | ||
0187bdfb BH |
1310 | /** |
1311 | * dev_disable_lro - disable Large Receive Offload on a device | |
1312 | * @dev: device | |
1313 | * | |
1314 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1315 | * called under RTNL. This is needed if received packets may be | |
1316 | * forwarded to another interface. | |
1317 | */ | |
1318 | void dev_disable_lro(struct net_device *dev) | |
1319 | { | |
1320 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1321 | dev->ethtool_ops->set_flags) { | |
1322 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1323 | if (flags & ETH_FLAG_LRO) { | |
1324 | flags &= ~ETH_FLAG_LRO; | |
1325 | dev->ethtool_ops->set_flags(dev, flags); | |
1326 | } | |
1327 | } | |
1328 | WARN_ON(dev->features & NETIF_F_LRO); | |
1329 | } | |
1330 | EXPORT_SYMBOL(dev_disable_lro); | |
1331 | ||
1332 | ||
881d966b EB |
1333 | static int dev_boot_phase = 1; |
1334 | ||
1da177e4 LT |
1335 | /* |
1336 | * Device change register/unregister. These are not inline or static | |
1337 | * as we export them to the world. | |
1338 | */ | |
1339 | ||
1340 | /** | |
1341 | * register_netdevice_notifier - register a network notifier block | |
1342 | * @nb: notifier | |
1343 | * | |
1344 | * Register a notifier to be called when network device events occur. | |
1345 | * The notifier passed is linked into the kernel structures and must | |
1346 | * not be reused until it has been unregistered. A negative errno code | |
1347 | * is returned on a failure. | |
1348 | * | |
1349 | * When registered all registration and up events are replayed | |
4ec93edb | 1350 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1351 | * view of the network device list. |
1352 | */ | |
1353 | ||
1354 | int register_netdevice_notifier(struct notifier_block *nb) | |
1355 | { | |
1356 | struct net_device *dev; | |
fcc5a03a | 1357 | struct net_device *last; |
881d966b | 1358 | struct net *net; |
1da177e4 LT |
1359 | int err; |
1360 | ||
1361 | rtnl_lock(); | |
f07d5b94 | 1362 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1363 | if (err) |
1364 | goto unlock; | |
881d966b EB |
1365 | if (dev_boot_phase) |
1366 | goto unlock; | |
1367 | for_each_net(net) { | |
1368 | for_each_netdev(net, dev) { | |
1369 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1370 | err = notifier_to_errno(err); | |
1371 | if (err) | |
1372 | goto rollback; | |
1373 | ||
1374 | if (!(dev->flags & IFF_UP)) | |
1375 | continue; | |
1da177e4 | 1376 | |
881d966b EB |
1377 | nb->notifier_call(nb, NETDEV_UP, dev); |
1378 | } | |
1da177e4 | 1379 | } |
fcc5a03a HX |
1380 | |
1381 | unlock: | |
1da177e4 LT |
1382 | rtnl_unlock(); |
1383 | return err; | |
fcc5a03a HX |
1384 | |
1385 | rollback: | |
1386 | last = dev; | |
881d966b EB |
1387 | for_each_net(net) { |
1388 | for_each_netdev(net, dev) { | |
1389 | if (dev == last) | |
1390 | break; | |
fcc5a03a | 1391 | |
881d966b EB |
1392 | if (dev->flags & IFF_UP) { |
1393 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1394 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1395 | } | |
1396 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
a5ee1551 | 1397 | nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev); |
fcc5a03a | 1398 | } |
fcc5a03a | 1399 | } |
c67625a1 PE |
1400 | |
1401 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1402 | goto unlock; |
1da177e4 | 1403 | } |
d1b19dff | 1404 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1405 | |
1406 | /** | |
1407 | * unregister_netdevice_notifier - unregister a network notifier block | |
1408 | * @nb: notifier | |
1409 | * | |
1410 | * Unregister a notifier previously registered by | |
1411 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1412 | * kernel structures and may then be reused. A negative errno code | |
1413 | * is returned on a failure. | |
1414 | */ | |
1415 | ||
1416 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1417 | { | |
9f514950 HX |
1418 | int err; |
1419 | ||
1420 | rtnl_lock(); | |
f07d5b94 | 1421 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1422 | rtnl_unlock(); |
1423 | return err; | |
1da177e4 | 1424 | } |
d1b19dff | 1425 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 LT |
1426 | |
1427 | /** | |
1428 | * call_netdevice_notifiers - call all network notifier blocks | |
1429 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1430 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1431 | * |
1432 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1433 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1434 | */ |
1435 | ||
ad7379d4 | 1436 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1437 | { |
ad7379d4 | 1438 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 LT |
1439 | } |
1440 | ||
1441 | /* When > 0 there are consumers of rx skb time stamps */ | |
1442 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1443 | ||
1444 | void net_enable_timestamp(void) | |
1445 | { | |
1446 | atomic_inc(&netstamp_needed); | |
1447 | } | |
d1b19dff | 1448 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1449 | |
1450 | void net_disable_timestamp(void) | |
1451 | { | |
1452 | atomic_dec(&netstamp_needed); | |
1453 | } | |
d1b19dff | 1454 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1455 | |
a61bbcf2 | 1456 | static inline void net_timestamp(struct sk_buff *skb) |
1da177e4 LT |
1457 | { |
1458 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1459 | __net_timestamp(skb); |
b7aa0bf7 ED |
1460 | else |
1461 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1462 | } |
1463 | ||
44540960 AB |
1464 | /** |
1465 | * dev_forward_skb - loopback an skb to another netif | |
1466 | * | |
1467 | * @dev: destination network device | |
1468 | * @skb: buffer to forward | |
1469 | * | |
1470 | * return values: | |
1471 | * NET_RX_SUCCESS (no congestion) | |
1472 | * NET_RX_DROP (packet was dropped) | |
1473 | * | |
1474 | * dev_forward_skb can be used for injecting an skb from the | |
1475 | * start_xmit function of one device into the receive queue | |
1476 | * of another device. | |
1477 | * | |
1478 | * The receiving device may be in another namespace, so | |
1479 | * we have to clear all information in the skb that could | |
1480 | * impact namespace isolation. | |
1481 | */ | |
1482 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1483 | { | |
1484 | skb_orphan(skb); | |
1485 | ||
1486 | if (!(dev->flags & IFF_UP)) | |
1487 | return NET_RX_DROP; | |
1488 | ||
1489 | if (skb->len > (dev->mtu + dev->hard_header_len)) | |
1490 | return NET_RX_DROP; | |
1491 | ||
8a83a00b | 1492 | skb_set_dev(skb, dev); |
44540960 AB |
1493 | skb->tstamp.tv64 = 0; |
1494 | skb->pkt_type = PACKET_HOST; | |
1495 | skb->protocol = eth_type_trans(skb, dev); | |
44540960 AB |
1496 | return netif_rx(skb); |
1497 | } | |
1498 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1499 | ||
1da177e4 LT |
1500 | /* |
1501 | * Support routine. Sends outgoing frames to any network | |
1502 | * taps currently in use. | |
1503 | */ | |
1504 | ||
f6a78bfc | 1505 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1506 | { |
1507 | struct packet_type *ptype; | |
a61bbcf2 | 1508 | |
8caf1539 JP |
1509 | #ifdef CONFIG_NET_CLS_ACT |
1510 | if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS))) | |
1511 | net_timestamp(skb); | |
1512 | #else | |
a61bbcf2 | 1513 | net_timestamp(skb); |
8caf1539 | 1514 | #endif |
1da177e4 LT |
1515 | |
1516 | rcu_read_lock(); | |
1517 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1518 | /* Never send packets back to the socket | |
1519 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1520 | */ | |
1521 | if ((ptype->dev == dev || !ptype->dev) && | |
1522 | (ptype->af_packet_priv == NULL || | |
1523 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
d1b19dff | 1524 | struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); |
1da177e4 LT |
1525 | if (!skb2) |
1526 | break; | |
1527 | ||
1528 | /* skb->nh should be correctly | |
1529 | set by sender, so that the second statement is | |
1530 | just protection against buggy protocols. | |
1531 | */ | |
459a98ed | 1532 | skb_reset_mac_header(skb2); |
1da177e4 | 1533 | |
d56f90a7 | 1534 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1535 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1536 | if (net_ratelimit()) |
1537 | printk(KERN_CRIT "protocol %04x is " | |
1538 | "buggy, dev %s\n", | |
1539 | skb2->protocol, dev->name); | |
c1d2bbe1 | 1540 | skb_reset_network_header(skb2); |
1da177e4 LT |
1541 | } |
1542 | ||
b0e380b1 | 1543 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1544 | skb2->pkt_type = PACKET_OUTGOING; |
f2ccd8fa | 1545 | ptype->func(skb2, skb->dev, ptype, skb->dev); |
1da177e4 LT |
1546 | } |
1547 | } | |
1548 | rcu_read_unlock(); | |
1549 | } | |
1550 | ||
56079431 | 1551 | |
def82a1d | 1552 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1553 | { |
def82a1d JP |
1554 | struct softnet_data *sd; |
1555 | unsigned long flags; | |
56079431 | 1556 | |
def82a1d JP |
1557 | local_irq_save(flags); |
1558 | sd = &__get_cpu_var(softnet_data); | |
1559 | q->next_sched = sd->output_queue; | |
1560 | sd->output_queue = q; | |
1561 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1562 | local_irq_restore(flags); | |
1563 | } | |
1564 | ||
1565 | void __netif_schedule(struct Qdisc *q) | |
1566 | { | |
1567 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1568 | __netif_reschedule(q); | |
56079431 DV |
1569 | } |
1570 | EXPORT_SYMBOL(__netif_schedule); | |
1571 | ||
bea3348e | 1572 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1573 | { |
bea3348e SH |
1574 | if (atomic_dec_and_test(&skb->users)) { |
1575 | struct softnet_data *sd; | |
1576 | unsigned long flags; | |
56079431 | 1577 | |
bea3348e SH |
1578 | local_irq_save(flags); |
1579 | sd = &__get_cpu_var(softnet_data); | |
1580 | skb->next = sd->completion_queue; | |
1581 | sd->completion_queue = skb; | |
1582 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1583 | local_irq_restore(flags); | |
1584 | } | |
56079431 | 1585 | } |
bea3348e | 1586 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1587 | |
1588 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1589 | { | |
1590 | if (in_irq() || irqs_disabled()) | |
1591 | dev_kfree_skb_irq(skb); | |
1592 | else | |
1593 | dev_kfree_skb(skb); | |
1594 | } | |
1595 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1596 | ||
1597 | ||
bea3348e SH |
1598 | /** |
1599 | * netif_device_detach - mark device as removed | |
1600 | * @dev: network device | |
1601 | * | |
1602 | * Mark device as removed from system and therefore no longer available. | |
1603 | */ | |
56079431 DV |
1604 | void netif_device_detach(struct net_device *dev) |
1605 | { | |
1606 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1607 | netif_running(dev)) { | |
d543103a | 1608 | netif_tx_stop_all_queues(dev); |
56079431 DV |
1609 | } |
1610 | } | |
1611 | EXPORT_SYMBOL(netif_device_detach); | |
1612 | ||
bea3348e SH |
1613 | /** |
1614 | * netif_device_attach - mark device as attached | |
1615 | * @dev: network device | |
1616 | * | |
1617 | * Mark device as attached from system and restart if needed. | |
1618 | */ | |
56079431 DV |
1619 | void netif_device_attach(struct net_device *dev) |
1620 | { | |
1621 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1622 | netif_running(dev)) { | |
d543103a | 1623 | netif_tx_wake_all_queues(dev); |
4ec93edb | 1624 | __netdev_watchdog_up(dev); |
56079431 DV |
1625 | } |
1626 | } | |
1627 | EXPORT_SYMBOL(netif_device_attach); | |
1628 | ||
6de329e2 BH |
1629 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) |
1630 | { | |
1631 | return ((features & NETIF_F_GEN_CSUM) || | |
1632 | ((features & NETIF_F_IP_CSUM) && | |
1633 | protocol == htons(ETH_P_IP)) || | |
1634 | ((features & NETIF_F_IPV6_CSUM) && | |
1c8dbcf6 YZ |
1635 | protocol == htons(ETH_P_IPV6)) || |
1636 | ((features & NETIF_F_FCOE_CRC) && | |
1637 | protocol == htons(ETH_P_FCOE))); | |
6de329e2 BH |
1638 | } |
1639 | ||
1640 | static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb) | |
1641 | { | |
1642 | if (can_checksum_protocol(dev->features, skb->protocol)) | |
1643 | return true; | |
1644 | ||
1645 | if (skb->protocol == htons(ETH_P_8021Q)) { | |
1646 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
1647 | if (can_checksum_protocol(dev->features & dev->vlan_features, | |
1648 | veh->h_vlan_encapsulated_proto)) | |
1649 | return true; | |
1650 | } | |
1651 | ||
1652 | return false; | |
1653 | } | |
56079431 | 1654 | |
8a83a00b AB |
1655 | /** |
1656 | * skb_dev_set -- assign a new device to a buffer | |
1657 | * @skb: buffer for the new device | |
1658 | * @dev: network device | |
1659 | * | |
1660 | * If an skb is owned by a device already, we have to reset | |
1661 | * all data private to the namespace a device belongs to | |
1662 | * before assigning it a new device. | |
1663 | */ | |
1664 | #ifdef CONFIG_NET_NS | |
1665 | void skb_set_dev(struct sk_buff *skb, struct net_device *dev) | |
1666 | { | |
1667 | skb_dst_drop(skb); | |
1668 | if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) { | |
1669 | secpath_reset(skb); | |
1670 | nf_reset(skb); | |
1671 | skb_init_secmark(skb); | |
1672 | skb->mark = 0; | |
1673 | skb->priority = 0; | |
1674 | skb->nf_trace = 0; | |
1675 | skb->ipvs_property = 0; | |
1676 | #ifdef CONFIG_NET_SCHED | |
1677 | skb->tc_index = 0; | |
1678 | #endif | |
1679 | } | |
1680 | skb->dev = dev; | |
1681 | } | |
1682 | EXPORT_SYMBOL(skb_set_dev); | |
1683 | #endif /* CONFIG_NET_NS */ | |
1684 | ||
1da177e4 LT |
1685 | /* |
1686 | * Invalidate hardware checksum when packet is to be mangled, and | |
1687 | * complete checksum manually on outgoing path. | |
1688 | */ | |
84fa7933 | 1689 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1690 | { |
d3bc23e7 | 1691 | __wsum csum; |
663ead3b | 1692 | int ret = 0, offset; |
1da177e4 | 1693 | |
84fa7933 | 1694 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1695 | goto out_set_summed; |
1696 | ||
1697 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1698 | /* Let GSO fix up the checksum. */ |
1699 | goto out_set_summed; | |
1da177e4 LT |
1700 | } |
1701 | ||
a030847e HX |
1702 | offset = skb->csum_start - skb_headroom(skb); |
1703 | BUG_ON(offset >= skb_headlen(skb)); | |
1704 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1705 | ||
1706 | offset += skb->csum_offset; | |
1707 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1708 | ||
1709 | if (skb_cloned(skb) && | |
1710 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1711 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1712 | if (ret) | |
1713 | goto out; | |
1714 | } | |
1715 | ||
a030847e | 1716 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1717 | out_set_summed: |
1da177e4 | 1718 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1719 | out: |
1da177e4 LT |
1720 | return ret; |
1721 | } | |
d1b19dff | 1722 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 1723 | |
f6a78bfc HX |
1724 | /** |
1725 | * skb_gso_segment - Perform segmentation on skb. | |
1726 | * @skb: buffer to segment | |
576a30eb | 1727 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1728 | * |
1729 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1730 | * |
1731 | * It may return NULL if the skb requires no segmentation. This is | |
1732 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1733 | */ |
576a30eb | 1734 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc HX |
1735 | { |
1736 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1737 | struct packet_type *ptype; | |
252e3346 | 1738 | __be16 type = skb->protocol; |
a430a43d | 1739 | int err; |
f6a78bfc | 1740 | |
459a98ed | 1741 | skb_reset_mac_header(skb); |
b0e380b1 | 1742 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1743 | __skb_pull(skb, skb->mac_len); |
1744 | ||
67fd1a73 HX |
1745 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
1746 | struct net_device *dev = skb->dev; | |
1747 | struct ethtool_drvinfo info = {}; | |
1748 | ||
1749 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1750 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1751 | ||
1752 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d " | |
1753 | "ip_summed=%d", | |
1754 | info.driver, dev ? dev->features : 0L, | |
1755 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1756 | skb->len, skb->data_len, skb->ip_summed); | |
1757 | ||
a430a43d HX |
1758 | if (skb_header_cloned(skb) && |
1759 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1760 | return ERR_PTR(err); | |
1761 | } | |
1762 | ||
f6a78bfc | 1763 | rcu_read_lock(); |
82d8a867 PE |
1764 | list_for_each_entry_rcu(ptype, |
1765 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1766 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1767 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1768 | err = ptype->gso_send_check(skb); |
1769 | segs = ERR_PTR(err); | |
1770 | if (err || skb_gso_ok(skb, features)) | |
1771 | break; | |
d56f90a7 ACM |
1772 | __skb_push(skb, (skb->data - |
1773 | skb_network_header(skb))); | |
a430a43d | 1774 | } |
576a30eb | 1775 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1776 | break; |
1777 | } | |
1778 | } | |
1779 | rcu_read_unlock(); | |
1780 | ||
98e399f8 | 1781 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1782 | |
f6a78bfc HX |
1783 | return segs; |
1784 | } | |
f6a78bfc HX |
1785 | EXPORT_SYMBOL(skb_gso_segment); |
1786 | ||
fb286bb2 HX |
1787 | /* Take action when hardware reception checksum errors are detected. */ |
1788 | #ifdef CONFIG_BUG | |
1789 | void netdev_rx_csum_fault(struct net_device *dev) | |
1790 | { | |
1791 | if (net_ratelimit()) { | |
4ec93edb | 1792 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1793 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1794 | dump_stack(); |
1795 | } | |
1796 | } | |
1797 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1798 | #endif | |
1799 | ||
1da177e4 LT |
1800 | /* Actually, we should eliminate this check as soon as we know, that: |
1801 | * 1. IOMMU is present and allows to map all the memory. | |
1802 | * 2. No high memory really exists on this machine. | |
1803 | */ | |
1804 | ||
9092c658 | 1805 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 1806 | { |
3d3a8533 | 1807 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1808 | int i; |
5acbbd42 FT |
1809 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
1810 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1811 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1812 | return 1; | |
1813 | } | |
1da177e4 | 1814 | |
5acbbd42 FT |
1815 | if (PCI_DMA_BUS_IS_PHYS) { |
1816 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 1817 | |
9092c658 ED |
1818 | if (!pdev) |
1819 | return 0; | |
5acbbd42 FT |
1820 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
1821 | dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page); | |
1822 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) | |
1823 | return 1; | |
1824 | } | |
1825 | } | |
3d3a8533 | 1826 | #endif |
1da177e4 LT |
1827 | return 0; |
1828 | } | |
1da177e4 | 1829 | |
f6a78bfc HX |
1830 | struct dev_gso_cb { |
1831 | void (*destructor)(struct sk_buff *skb); | |
1832 | }; | |
1833 | ||
1834 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1835 | ||
1836 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1837 | { | |
1838 | struct dev_gso_cb *cb; | |
1839 | ||
1840 | do { | |
1841 | struct sk_buff *nskb = skb->next; | |
1842 | ||
1843 | skb->next = nskb->next; | |
1844 | nskb->next = NULL; | |
1845 | kfree_skb(nskb); | |
1846 | } while (skb->next); | |
1847 | ||
1848 | cb = DEV_GSO_CB(skb); | |
1849 | if (cb->destructor) | |
1850 | cb->destructor(skb); | |
1851 | } | |
1852 | ||
1853 | /** | |
1854 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1855 | * @skb: buffer to segment | |
1856 | * | |
1857 | * This function segments the given skb and stores the list of segments | |
1858 | * in skb->next. | |
1859 | */ | |
1860 | static int dev_gso_segment(struct sk_buff *skb) | |
1861 | { | |
1862 | struct net_device *dev = skb->dev; | |
1863 | struct sk_buff *segs; | |
576a30eb HX |
1864 | int features = dev->features & ~(illegal_highdma(dev, skb) ? |
1865 | NETIF_F_SG : 0); | |
1866 | ||
1867 | segs = skb_gso_segment(skb, features); | |
1868 | ||
1869 | /* Verifying header integrity only. */ | |
1870 | if (!segs) | |
1871 | return 0; | |
f6a78bfc | 1872 | |
801678c5 | 1873 | if (IS_ERR(segs)) |
f6a78bfc HX |
1874 | return PTR_ERR(segs); |
1875 | ||
1876 | skb->next = segs; | |
1877 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1878 | skb->destructor = dev_gso_skb_destructor; | |
1879 | ||
1880 | return 0; | |
1881 | } | |
1882 | ||
fd2ea0a7 DM |
1883 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
1884 | struct netdev_queue *txq) | |
f6a78bfc | 1885 | { |
00829823 | 1886 | const struct net_device_ops *ops = dev->netdev_ops; |
572a9d7b | 1887 | int rc = NETDEV_TX_OK; |
00829823 | 1888 | |
f6a78bfc | 1889 | if (likely(!skb->next)) { |
9be9a6b9 | 1890 | if (!list_empty(&ptype_all)) |
f6a78bfc HX |
1891 | dev_queue_xmit_nit(skb, dev); |
1892 | ||
576a30eb HX |
1893 | if (netif_needs_gso(dev, skb)) { |
1894 | if (unlikely(dev_gso_segment(skb))) | |
1895 | goto out_kfree_skb; | |
1896 | if (skb->next) | |
1897 | goto gso; | |
1898 | } | |
f6a78bfc | 1899 | |
93f154b5 ED |
1900 | /* |
1901 | * If device doesnt need skb->dst, release it right now while | |
1902 | * its hot in this cpu cache | |
1903 | */ | |
adf30907 ED |
1904 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) |
1905 | skb_dst_drop(skb); | |
1906 | ||
ac45f602 | 1907 | rc = ops->ndo_start_xmit(skb, dev); |
ec634fe3 | 1908 | if (rc == NETDEV_TX_OK) |
08baf561 | 1909 | txq_trans_update(txq); |
ac45f602 PO |
1910 | /* |
1911 | * TODO: if skb_orphan() was called by | |
1912 | * dev->hard_start_xmit() (for example, the unmodified | |
1913 | * igb driver does that; bnx2 doesn't), then | |
1914 | * skb_tx_software_timestamp() will be unable to send | |
1915 | * back the time stamp. | |
1916 | * | |
1917 | * How can this be prevented? Always create another | |
1918 | * reference to the socket before calling | |
1919 | * dev->hard_start_xmit()? Prevent that skb_orphan() | |
1920 | * does anything in dev->hard_start_xmit() by clearing | |
1921 | * the skb destructor before the call and restoring it | |
1922 | * afterwards, then doing the skb_orphan() ourselves? | |
1923 | */ | |
ac45f602 | 1924 | return rc; |
f6a78bfc HX |
1925 | } |
1926 | ||
576a30eb | 1927 | gso: |
f6a78bfc HX |
1928 | do { |
1929 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
1930 | |
1931 | skb->next = nskb->next; | |
1932 | nskb->next = NULL; | |
068a2de5 KK |
1933 | |
1934 | /* | |
1935 | * If device doesnt need nskb->dst, release it right now while | |
1936 | * its hot in this cpu cache | |
1937 | */ | |
1938 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
1939 | skb_dst_drop(nskb); | |
1940 | ||
00829823 | 1941 | rc = ops->ndo_start_xmit(nskb, dev); |
ec634fe3 | 1942 | if (unlikely(rc != NETDEV_TX_OK)) { |
572a9d7b PM |
1943 | if (rc & ~NETDEV_TX_MASK) |
1944 | goto out_kfree_gso_skb; | |
f54d9e8d | 1945 | nskb->next = skb->next; |
f6a78bfc HX |
1946 | skb->next = nskb; |
1947 | return rc; | |
1948 | } | |
08baf561 | 1949 | txq_trans_update(txq); |
fd2ea0a7 | 1950 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) |
f54d9e8d | 1951 | return NETDEV_TX_BUSY; |
f6a78bfc | 1952 | } while (skb->next); |
4ec93edb | 1953 | |
572a9d7b PM |
1954 | out_kfree_gso_skb: |
1955 | if (likely(skb->next == NULL)) | |
1956 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
f6a78bfc HX |
1957 | out_kfree_skb: |
1958 | kfree_skb(skb); | |
572a9d7b | 1959 | return rc; |
f6a78bfc HX |
1960 | } |
1961 | ||
0a9627f2 | 1962 | static u32 hashrnd __read_mostly; |
b6b2fed1 | 1963 | |
9247744e | 1964 | u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb) |
8f0f2223 | 1965 | { |
7019298a | 1966 | u32 hash; |
b6b2fed1 | 1967 | |
513de11b DM |
1968 | if (skb_rx_queue_recorded(skb)) { |
1969 | hash = skb_get_rx_queue(skb); | |
d1b19dff | 1970 | while (unlikely(hash >= dev->real_num_tx_queues)) |
513de11b DM |
1971 | hash -= dev->real_num_tx_queues; |
1972 | return hash; | |
1973 | } | |
ec581f6a ED |
1974 | |
1975 | if (skb->sk && skb->sk->sk_hash) | |
7019298a | 1976 | hash = skb->sk->sk_hash; |
ec581f6a | 1977 | else |
7019298a | 1978 | hash = skb->protocol; |
d5a9e24a | 1979 | |
0a9627f2 | 1980 | hash = jhash_1word(hash, hashrnd); |
b6b2fed1 DM |
1981 | |
1982 | return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32); | |
8f0f2223 | 1983 | } |
9247744e | 1984 | EXPORT_SYMBOL(skb_tx_hash); |
8f0f2223 | 1985 | |
ed04642f ED |
1986 | static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index) |
1987 | { | |
1988 | if (unlikely(queue_index >= dev->real_num_tx_queues)) { | |
1989 | if (net_ratelimit()) { | |
7a161ea9 ED |
1990 | pr_warning("%s selects TX queue %d, but " |
1991 | "real number of TX queues is %d\n", | |
1992 | dev->name, queue_index, dev->real_num_tx_queues); | |
ed04642f ED |
1993 | } |
1994 | return 0; | |
1995 | } | |
1996 | return queue_index; | |
1997 | } | |
1998 | ||
e8a0464c DM |
1999 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
2000 | struct sk_buff *skb) | |
2001 | { | |
a4ee3ce3 KK |
2002 | u16 queue_index; |
2003 | struct sock *sk = skb->sk; | |
2004 | ||
2005 | if (sk_tx_queue_recorded(sk)) { | |
2006 | queue_index = sk_tx_queue_get(sk); | |
2007 | } else { | |
2008 | const struct net_device_ops *ops = dev->netdev_ops; | |
2009 | ||
2010 | if (ops->ndo_select_queue) { | |
2011 | queue_index = ops->ndo_select_queue(dev, skb); | |
ed04642f | 2012 | queue_index = dev_cap_txqueue(dev, queue_index); |
a4ee3ce3 KK |
2013 | } else { |
2014 | queue_index = 0; | |
2015 | if (dev->real_num_tx_queues > 1) | |
2016 | queue_index = skb_tx_hash(dev, skb); | |
fd2ea0a7 | 2017 | |
b6c6712a | 2018 | if (sk && rcu_dereference_check(sk->sk_dst_cache, 1)) |
a4ee3ce3 KK |
2019 | sk_tx_queue_set(sk, queue_index); |
2020 | } | |
2021 | } | |
eae792b7 | 2022 | |
fd2ea0a7 DM |
2023 | skb_set_queue_mapping(skb, queue_index); |
2024 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
2025 | } |
2026 | ||
bbd8a0d3 KK |
2027 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
2028 | struct net_device *dev, | |
2029 | struct netdev_queue *txq) | |
2030 | { | |
2031 | spinlock_t *root_lock = qdisc_lock(q); | |
2032 | int rc; | |
2033 | ||
2034 | spin_lock(root_lock); | |
2035 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
2036 | kfree_skb(skb); | |
2037 | rc = NET_XMIT_DROP; | |
2038 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
2039 | !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) { | |
2040 | /* | |
2041 | * This is a work-conserving queue; there are no old skbs | |
2042 | * waiting to be sent out; and the qdisc is not running - | |
2043 | * xmit the skb directly. | |
2044 | */ | |
2045 | __qdisc_update_bstats(q, skb->len); | |
2046 | if (sch_direct_xmit(skb, q, dev, txq, root_lock)) | |
2047 | __qdisc_run(q); | |
2048 | else | |
2049 | clear_bit(__QDISC_STATE_RUNNING, &q->state); | |
2050 | ||
2051 | rc = NET_XMIT_SUCCESS; | |
2052 | } else { | |
2053 | rc = qdisc_enqueue_root(skb, q); | |
2054 | qdisc_run(q); | |
2055 | } | |
2056 | spin_unlock(root_lock); | |
2057 | ||
2058 | return rc; | |
2059 | } | |
2060 | ||
4b258461 KK |
2061 | /* |
2062 | * Returns true if either: | |
2063 | * 1. skb has frag_list and the device doesn't support FRAGLIST, or | |
2064 | * 2. skb is fragmented and the device does not support SG, or if | |
2065 | * at least one of fragments is in highmem and device does not | |
2066 | * support DMA from it. | |
2067 | */ | |
2068 | static inline int skb_needs_linearize(struct sk_buff *skb, | |
2069 | struct net_device *dev) | |
2070 | { | |
2071 | return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) || | |
2072 | (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) || | |
2073 | illegal_highdma(dev, skb))); | |
2074 | } | |
2075 | ||
d29f749e DJ |
2076 | /** |
2077 | * dev_queue_xmit - transmit a buffer | |
2078 | * @skb: buffer to transmit | |
2079 | * | |
2080 | * Queue a buffer for transmission to a network device. The caller must | |
2081 | * have set the device and priority and built the buffer before calling | |
2082 | * this function. The function can be called from an interrupt. | |
2083 | * | |
2084 | * A negative errno code is returned on a failure. A success does not | |
2085 | * guarantee the frame will be transmitted as it may be dropped due | |
2086 | * to congestion or traffic shaping. | |
2087 | * | |
2088 | * ----------------------------------------------------------------------------------- | |
2089 | * I notice this method can also return errors from the queue disciplines, | |
2090 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
2091 | * be positive. | |
2092 | * | |
2093 | * Regardless of the return value, the skb is consumed, so it is currently | |
2094 | * difficult to retry a send to this method. (You can bump the ref count | |
2095 | * before sending to hold a reference for retry if you are careful.) | |
2096 | * | |
2097 | * When calling this method, interrupts MUST be enabled. This is because | |
2098 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
2099 | * --BLG | |
2100 | */ | |
1da177e4 LT |
2101 | int dev_queue_xmit(struct sk_buff *skb) |
2102 | { | |
2103 | struct net_device *dev = skb->dev; | |
dc2b4847 | 2104 | struct netdev_queue *txq; |
1da177e4 LT |
2105 | struct Qdisc *q; |
2106 | int rc = -ENOMEM; | |
2107 | ||
f6a78bfc HX |
2108 | /* GSO will handle the following emulations directly. */ |
2109 | if (netif_needs_gso(dev, skb)) | |
2110 | goto gso; | |
2111 | ||
4b258461 KK |
2112 | /* Convert a paged skb to linear, if required */ |
2113 | if (skb_needs_linearize(skb, dev) && __skb_linearize(skb)) | |
1da177e4 LT |
2114 | goto out_kfree_skb; |
2115 | ||
2116 | /* If packet is not checksummed and device does not support | |
2117 | * checksumming for this protocol, complete checksumming here. | |
2118 | */ | |
663ead3b HX |
2119 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
2120 | skb_set_transport_header(skb, skb->csum_start - | |
2121 | skb_headroom(skb)); | |
6de329e2 BH |
2122 | if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb)) |
2123 | goto out_kfree_skb; | |
663ead3b | 2124 | } |
1da177e4 | 2125 | |
f6a78bfc | 2126 | gso: |
4ec93edb YH |
2127 | /* Disable soft irqs for various locks below. Also |
2128 | * stops preemption for RCU. | |
1da177e4 | 2129 | */ |
4ec93edb | 2130 | rcu_read_lock_bh(); |
1da177e4 | 2131 | |
eae792b7 | 2132 | txq = dev_pick_tx(dev, skb); |
a898def2 | 2133 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 2134 | |
1da177e4 | 2135 | #ifdef CONFIG_NET_CLS_ACT |
d1b19dff | 2136 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS); |
1da177e4 LT |
2137 | #endif |
2138 | if (q->enqueue) { | |
bbd8a0d3 | 2139 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 2140 | goto out; |
1da177e4 LT |
2141 | } |
2142 | ||
2143 | /* The device has no queue. Common case for software devices: | |
2144 | loopback, all the sorts of tunnels... | |
2145 | ||
932ff279 HX |
2146 | Really, it is unlikely that netif_tx_lock protection is necessary |
2147 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
2148 | counters.) |
2149 | However, it is possible, that they rely on protection | |
2150 | made by us here. | |
2151 | ||
2152 | Check this and shot the lock. It is not prone from deadlocks. | |
2153 | Either shot noqueue qdisc, it is even simpler 8) | |
2154 | */ | |
2155 | if (dev->flags & IFF_UP) { | |
2156 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
2157 | ||
c773e847 | 2158 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 2159 | |
c773e847 | 2160 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 2161 | |
fd2ea0a7 | 2162 | if (!netif_tx_queue_stopped(txq)) { |
572a9d7b PM |
2163 | rc = dev_hard_start_xmit(skb, dev, txq); |
2164 | if (dev_xmit_complete(rc)) { | |
c773e847 | 2165 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2166 | goto out; |
2167 | } | |
2168 | } | |
c773e847 | 2169 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
2170 | if (net_ratelimit()) |
2171 | printk(KERN_CRIT "Virtual device %s asks to " | |
2172 | "queue packet!\n", dev->name); | |
2173 | } else { | |
2174 | /* Recursion is detected! It is possible, | |
2175 | * unfortunately */ | |
2176 | if (net_ratelimit()) | |
2177 | printk(KERN_CRIT "Dead loop on virtual device " | |
2178 | "%s, fix it urgently!\n", dev->name); | |
2179 | } | |
2180 | } | |
2181 | ||
2182 | rc = -ENETDOWN; | |
d4828d85 | 2183 | rcu_read_unlock_bh(); |
1da177e4 LT |
2184 | |
2185 | out_kfree_skb: | |
2186 | kfree_skb(skb); | |
2187 | return rc; | |
2188 | out: | |
d4828d85 | 2189 | rcu_read_unlock_bh(); |
1da177e4 LT |
2190 | return rc; |
2191 | } | |
d1b19dff | 2192 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 LT |
2193 | |
2194 | ||
2195 | /*======================================================================= | |
2196 | Receiver routines | |
2197 | =======================================================================*/ | |
2198 | ||
6b2bedc3 SH |
2199 | int netdev_max_backlog __read_mostly = 1000; |
2200 | int netdev_budget __read_mostly = 300; | |
2201 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 LT |
2202 | |
2203 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; | |
2204 | ||
df334545 | 2205 | #ifdef CONFIG_RPS |
fec5e652 TH |
2206 | |
2207 | /* One global table that all flow-based protocols share. */ | |
2208 | struct rps_sock_flow_table *rps_sock_flow_table; | |
2209 | EXPORT_SYMBOL(rps_sock_flow_table); | |
2210 | ||
0a9627f2 TH |
2211 | /* |
2212 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2213 | * CPU from the RPS map of the receiving queue for a given skb. | |
b0e28f1e | 2214 | * rcu_read_lock must be held on entry. |
0a9627f2 | 2215 | */ |
fec5e652 TH |
2216 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, |
2217 | struct rps_dev_flow **rflowp) | |
0a9627f2 TH |
2218 | { |
2219 | struct ipv6hdr *ip6; | |
2220 | struct iphdr *ip; | |
2221 | struct netdev_rx_queue *rxqueue; | |
2222 | struct rps_map *map; | |
fec5e652 TH |
2223 | struct rps_dev_flow_table *flow_table; |
2224 | struct rps_sock_flow_table *sock_flow_table; | |
0a9627f2 TH |
2225 | int cpu = -1; |
2226 | u8 ip_proto; | |
fec5e652 | 2227 | u16 tcpu; |
0a9627f2 TH |
2228 | u32 addr1, addr2, ports, ihl; |
2229 | ||
0a9627f2 TH |
2230 | if (skb_rx_queue_recorded(skb)) { |
2231 | u16 index = skb_get_rx_queue(skb); | |
2232 | if (unlikely(index >= dev->num_rx_queues)) { | |
2233 | if (net_ratelimit()) { | |
7a161ea9 ED |
2234 | pr_warning("%s received packet on queue " |
2235 | "%u, but number of RX queues is %u\n", | |
2236 | dev->name, index, dev->num_rx_queues); | |
0a9627f2 TH |
2237 | } |
2238 | goto done; | |
2239 | } | |
2240 | rxqueue = dev->_rx + index; | |
2241 | } else | |
2242 | rxqueue = dev->_rx; | |
2243 | ||
fec5e652 | 2244 | if (!rxqueue->rps_map && !rxqueue->rps_flow_table) |
0a9627f2 TH |
2245 | goto done; |
2246 | ||
2247 | if (skb->rxhash) | |
2248 | goto got_hash; /* Skip hash computation on packet header */ | |
2249 | ||
2250 | switch (skb->protocol) { | |
2251 | case __constant_htons(ETH_P_IP): | |
2252 | if (!pskb_may_pull(skb, sizeof(*ip))) | |
2253 | goto done; | |
2254 | ||
2255 | ip = (struct iphdr *) skb->data; | |
2256 | ip_proto = ip->protocol; | |
2257 | addr1 = ip->saddr; | |
2258 | addr2 = ip->daddr; | |
2259 | ihl = ip->ihl; | |
2260 | break; | |
2261 | case __constant_htons(ETH_P_IPV6): | |
2262 | if (!pskb_may_pull(skb, sizeof(*ip6))) | |
2263 | goto done; | |
2264 | ||
2265 | ip6 = (struct ipv6hdr *) skb->data; | |
2266 | ip_proto = ip6->nexthdr; | |
2267 | addr1 = ip6->saddr.s6_addr32[3]; | |
2268 | addr2 = ip6->daddr.s6_addr32[3]; | |
2269 | ihl = (40 >> 2); | |
2270 | break; | |
2271 | default: | |
2272 | goto done; | |
2273 | } | |
2274 | ports = 0; | |
2275 | switch (ip_proto) { | |
2276 | case IPPROTO_TCP: | |
2277 | case IPPROTO_UDP: | |
2278 | case IPPROTO_DCCP: | |
2279 | case IPPROTO_ESP: | |
2280 | case IPPROTO_AH: | |
2281 | case IPPROTO_SCTP: | |
2282 | case IPPROTO_UDPLITE: | |
2283 | if (pskb_may_pull(skb, (ihl * 4) + 4)) | |
2284 | ports = *((u32 *) (skb->data + (ihl * 4))); | |
2285 | break; | |
2286 | ||
2287 | default: | |
2288 | break; | |
2289 | } | |
2290 | ||
2291 | skb->rxhash = jhash_3words(addr1, addr2, ports, hashrnd); | |
2292 | if (!skb->rxhash) | |
2293 | skb->rxhash = 1; | |
2294 | ||
2295 | got_hash: | |
fec5e652 TH |
2296 | flow_table = rcu_dereference(rxqueue->rps_flow_table); |
2297 | sock_flow_table = rcu_dereference(rps_sock_flow_table); | |
2298 | if (flow_table && sock_flow_table) { | |
2299 | u16 next_cpu; | |
2300 | struct rps_dev_flow *rflow; | |
2301 | ||
2302 | rflow = &flow_table->flows[skb->rxhash & flow_table->mask]; | |
2303 | tcpu = rflow->cpu; | |
2304 | ||
2305 | next_cpu = sock_flow_table->ents[skb->rxhash & | |
2306 | sock_flow_table->mask]; | |
2307 | ||
2308 | /* | |
2309 | * If the desired CPU (where last recvmsg was done) is | |
2310 | * different from current CPU (one in the rx-queue flow | |
2311 | * table entry), switch if one of the following holds: | |
2312 | * - Current CPU is unset (equal to RPS_NO_CPU). | |
2313 | * - Current CPU is offline. | |
2314 | * - The current CPU's queue tail has advanced beyond the | |
2315 | * last packet that was enqueued using this table entry. | |
2316 | * This guarantees that all previous packets for the flow | |
2317 | * have been dequeued, thus preserving in order delivery. | |
2318 | */ | |
2319 | if (unlikely(tcpu != next_cpu) && | |
2320 | (tcpu == RPS_NO_CPU || !cpu_online(tcpu) || | |
2321 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - | |
2322 | rflow->last_qtail)) >= 0)) { | |
2323 | tcpu = rflow->cpu = next_cpu; | |
2324 | if (tcpu != RPS_NO_CPU) | |
2325 | rflow->last_qtail = per_cpu(softnet_data, | |
2326 | tcpu).input_queue_head; | |
2327 | } | |
2328 | if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) { | |
2329 | *rflowp = rflow; | |
2330 | cpu = tcpu; | |
2331 | goto done; | |
2332 | } | |
2333 | } | |
2334 | ||
0a9627f2 TH |
2335 | map = rcu_dereference(rxqueue->rps_map); |
2336 | if (map) { | |
fec5e652 | 2337 | tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; |
0a9627f2 TH |
2338 | |
2339 | if (cpu_online(tcpu)) { | |
2340 | cpu = tcpu; | |
2341 | goto done; | |
2342 | } | |
2343 | } | |
2344 | ||
2345 | done: | |
0a9627f2 TH |
2346 | return cpu; |
2347 | } | |
2348 | ||
2349 | /* | |
2350 | * This structure holds the per-CPU mask of CPUs for which IPIs are scheduled | |
2351 | * to be sent to kick remote softirq processing. There are two masks since | |
2352 | * the sending of IPIs must be done with interrupts enabled. The select field | |
2353 | * indicates the current mask that enqueue_backlog uses to schedule IPIs. | |
2354 | * select is flipped before net_rps_action is called while still under lock, | |
2355 | * net_rps_action then uses the non-selected mask to send the IPIs and clears | |
2356 | * it without conflicting with enqueue_backlog operation. | |
2357 | */ | |
2358 | struct rps_remote_softirq_cpus { | |
2359 | cpumask_t mask[2]; | |
2360 | int select; | |
2361 | }; | |
2362 | static DEFINE_PER_CPU(struct rps_remote_softirq_cpus, rps_remote_softirq_cpus); | |
2363 | ||
2364 | /* Called from hardirq (IPI) context */ | |
2365 | static void trigger_softirq(void *data) | |
2366 | { | |
2367 | struct softnet_data *queue = data; | |
2368 | __napi_schedule(&queue->backlog); | |
2369 | __get_cpu_var(netdev_rx_stat).received_rps++; | |
2370 | } | |
fec5e652 | 2371 | #endif /* CONFIG_RPS */ |
0a9627f2 TH |
2372 | |
2373 | /* | |
2374 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2375 | * queue (may be a remote CPU queue). | |
2376 | */ | |
fec5e652 TH |
2377 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
2378 | unsigned int *qtail) | |
0a9627f2 TH |
2379 | { |
2380 | struct softnet_data *queue; | |
2381 | unsigned long flags; | |
2382 | ||
2383 | queue = &per_cpu(softnet_data, cpu); | |
2384 | ||
2385 | local_irq_save(flags); | |
2386 | __get_cpu_var(netdev_rx_stat).total++; | |
2387 | ||
152102c7 | 2388 | rps_lock(queue); |
0a9627f2 TH |
2389 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { |
2390 | if (queue->input_pkt_queue.qlen) { | |
2391 | enqueue: | |
2392 | __skb_queue_tail(&queue->input_pkt_queue, skb); | |
fec5e652 TH |
2393 | #ifdef CONFIG_RPS |
2394 | *qtail = queue->input_queue_head + | |
2395 | queue->input_pkt_queue.qlen; | |
2396 | #endif | |
152102c7 CG |
2397 | rps_unlock(queue); |
2398 | local_irq_restore(flags); | |
0a9627f2 TH |
2399 | return NET_RX_SUCCESS; |
2400 | } | |
2401 | ||
2402 | /* Schedule NAPI for backlog device */ | |
2403 | if (napi_schedule_prep(&queue->backlog)) { | |
df334545 | 2404 | #ifdef CONFIG_RPS |
0a9627f2 TH |
2405 | if (cpu != smp_processor_id()) { |
2406 | struct rps_remote_softirq_cpus *rcpus = | |
2407 | &__get_cpu_var(rps_remote_softirq_cpus); | |
2408 | ||
2409 | cpu_set(cpu, rcpus->mask[rcpus->select]); | |
2410 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
fec5e652 TH |
2411 | goto enqueue; |
2412 | } | |
1e94d72f | 2413 | #endif |
fec5e652 | 2414 | __napi_schedule(&queue->backlog); |
0a9627f2 TH |
2415 | } |
2416 | goto enqueue; | |
2417 | } | |
2418 | ||
152102c7 | 2419 | rps_unlock(queue); |
0a9627f2 TH |
2420 | |
2421 | __get_cpu_var(netdev_rx_stat).dropped++; | |
2422 | local_irq_restore(flags); | |
2423 | ||
2424 | kfree_skb(skb); | |
2425 | return NET_RX_DROP; | |
2426 | } | |
1da177e4 | 2427 | |
1da177e4 LT |
2428 | /** |
2429 | * netif_rx - post buffer to the network code | |
2430 | * @skb: buffer to post | |
2431 | * | |
2432 | * This function receives a packet from a device driver and queues it for | |
2433 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2434 | * may be dropped during processing for congestion control or by the | |
2435 | * protocol layers. | |
2436 | * | |
2437 | * return values: | |
2438 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2439 | * NET_RX_DROP (packet was dropped) |
2440 | * | |
2441 | */ | |
2442 | ||
2443 | int netif_rx(struct sk_buff *skb) | |
2444 | { | |
b0e28f1e | 2445 | int ret; |
1da177e4 LT |
2446 | |
2447 | /* if netpoll wants it, pretend we never saw it */ | |
2448 | if (netpoll_rx(skb)) | |
2449 | return NET_RX_DROP; | |
2450 | ||
b7aa0bf7 | 2451 | if (!skb->tstamp.tv64) |
a61bbcf2 | 2452 | net_timestamp(skb); |
1da177e4 | 2453 | |
df334545 | 2454 | #ifdef CONFIG_RPS |
b0e28f1e | 2455 | { |
fec5e652 | 2456 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
2457 | int cpu; |
2458 | ||
2459 | rcu_read_lock(); | |
fec5e652 TH |
2460 | |
2461 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
2462 | if (cpu < 0) |
2463 | cpu = smp_processor_id(); | |
fec5e652 TH |
2464 | |
2465 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
2466 | ||
b0e28f1e ED |
2467 | rcu_read_unlock(); |
2468 | } | |
1e94d72f | 2469 | #else |
fec5e652 TH |
2470 | { |
2471 | unsigned int qtail; | |
2472 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); | |
2473 | put_cpu(); | |
2474 | } | |
1e94d72f | 2475 | #endif |
b0e28f1e | 2476 | return ret; |
1da177e4 | 2477 | } |
d1b19dff | 2478 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
2479 | |
2480 | int netif_rx_ni(struct sk_buff *skb) | |
2481 | { | |
2482 | int err; | |
2483 | ||
2484 | preempt_disable(); | |
2485 | err = netif_rx(skb); | |
2486 | if (local_softirq_pending()) | |
2487 | do_softirq(); | |
2488 | preempt_enable(); | |
2489 | ||
2490 | return err; | |
2491 | } | |
1da177e4 LT |
2492 | EXPORT_SYMBOL(netif_rx_ni); |
2493 | ||
1da177e4 LT |
2494 | static void net_tx_action(struct softirq_action *h) |
2495 | { | |
2496 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2497 | ||
2498 | if (sd->completion_queue) { | |
2499 | struct sk_buff *clist; | |
2500 | ||
2501 | local_irq_disable(); | |
2502 | clist = sd->completion_queue; | |
2503 | sd->completion_queue = NULL; | |
2504 | local_irq_enable(); | |
2505 | ||
2506 | while (clist) { | |
2507 | struct sk_buff *skb = clist; | |
2508 | clist = clist->next; | |
2509 | ||
547b792c | 2510 | WARN_ON(atomic_read(&skb->users)); |
1da177e4 LT |
2511 | __kfree_skb(skb); |
2512 | } | |
2513 | } | |
2514 | ||
2515 | if (sd->output_queue) { | |
37437bb2 | 2516 | struct Qdisc *head; |
1da177e4 LT |
2517 | |
2518 | local_irq_disable(); | |
2519 | head = sd->output_queue; | |
2520 | sd->output_queue = NULL; | |
2521 | local_irq_enable(); | |
2522 | ||
2523 | while (head) { | |
37437bb2 DM |
2524 | struct Qdisc *q = head; |
2525 | spinlock_t *root_lock; | |
2526 | ||
1da177e4 LT |
2527 | head = head->next_sched; |
2528 | ||
5fb66229 | 2529 | root_lock = qdisc_lock(q); |
37437bb2 | 2530 | if (spin_trylock(root_lock)) { |
def82a1d JP |
2531 | smp_mb__before_clear_bit(); |
2532 | clear_bit(__QDISC_STATE_SCHED, | |
2533 | &q->state); | |
37437bb2 DM |
2534 | qdisc_run(q); |
2535 | spin_unlock(root_lock); | |
1da177e4 | 2536 | } else { |
195648bb | 2537 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 2538 | &q->state)) { |
195648bb | 2539 | __netif_reschedule(q); |
e8a83e10 JP |
2540 | } else { |
2541 | smp_mb__before_clear_bit(); | |
2542 | clear_bit(__QDISC_STATE_SCHED, | |
2543 | &q->state); | |
2544 | } | |
1da177e4 LT |
2545 | } |
2546 | } | |
2547 | } | |
2548 | } | |
2549 | ||
6f05f629 SH |
2550 | static inline int deliver_skb(struct sk_buff *skb, |
2551 | struct packet_type *pt_prev, | |
2552 | struct net_device *orig_dev) | |
1da177e4 LT |
2553 | { |
2554 | atomic_inc(&skb->users); | |
f2ccd8fa | 2555 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2556 | } |
2557 | ||
2558 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) | |
da678292 MM |
2559 | |
2560 | #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE) | |
2561 | /* This hook is defined here for ATM LANE */ | |
2562 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
2563 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 2564 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 2565 | #endif |
1da177e4 | 2566 | |
6229e362 SH |
2567 | /* |
2568 | * If bridge module is loaded call bridging hook. | |
2569 | * returns NULL if packet was consumed. | |
2570 | */ | |
2571 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, | |
2572 | struct sk_buff *skb) __read_mostly; | |
4fb019a0 | 2573 | EXPORT_SYMBOL_GPL(br_handle_frame_hook); |
da678292 | 2574 | |
6229e362 SH |
2575 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb, |
2576 | struct packet_type **pt_prev, int *ret, | |
2577 | struct net_device *orig_dev) | |
1da177e4 LT |
2578 | { |
2579 | struct net_bridge_port *port; | |
2580 | ||
6229e362 SH |
2581 | if (skb->pkt_type == PACKET_LOOPBACK || |
2582 | (port = rcu_dereference(skb->dev->br_port)) == NULL) | |
2583 | return skb; | |
1da177e4 LT |
2584 | |
2585 | if (*pt_prev) { | |
6229e362 | 2586 | *ret = deliver_skb(skb, *pt_prev, orig_dev); |
1da177e4 | 2587 | *pt_prev = NULL; |
4ec93edb YH |
2588 | } |
2589 | ||
6229e362 | 2590 | return br_handle_frame_hook(port, skb); |
1da177e4 LT |
2591 | } |
2592 | #else | |
6229e362 | 2593 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) |
1da177e4 LT |
2594 | #endif |
2595 | ||
b863ceb7 PM |
2596 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) |
2597 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; | |
2598 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); | |
2599 | ||
2600 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, | |
2601 | struct packet_type **pt_prev, | |
2602 | int *ret, | |
2603 | struct net_device *orig_dev) | |
2604 | { | |
2605 | if (skb->dev->macvlan_port == NULL) | |
2606 | return skb; | |
2607 | ||
2608 | if (*pt_prev) { | |
2609 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2610 | *pt_prev = NULL; | |
2611 | } | |
2612 | return macvlan_handle_frame_hook(skb); | |
2613 | } | |
2614 | #else | |
2615 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) | |
2616 | #endif | |
2617 | ||
1da177e4 LT |
2618 | #ifdef CONFIG_NET_CLS_ACT |
2619 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2620 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2621 | * a compare and 2 stores extra right now if we dont have it on | |
2622 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 2623 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
2624 | * the ingress scheduler, you just cant add policies on ingress. |
2625 | * | |
2626 | */ | |
4ec93edb | 2627 | static int ing_filter(struct sk_buff *skb) |
1da177e4 | 2628 | { |
1da177e4 | 2629 | struct net_device *dev = skb->dev; |
f697c3e8 | 2630 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
2631 | struct netdev_queue *rxq; |
2632 | int result = TC_ACT_OK; | |
2633 | struct Qdisc *q; | |
4ec93edb | 2634 | |
f697c3e8 HX |
2635 | if (MAX_RED_LOOP < ttl++) { |
2636 | printk(KERN_WARNING | |
2637 | "Redir loop detected Dropping packet (%d->%d)\n", | |
8964be4a | 2638 | skb->skb_iif, dev->ifindex); |
f697c3e8 HX |
2639 | return TC_ACT_SHOT; |
2640 | } | |
1da177e4 | 2641 | |
f697c3e8 HX |
2642 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
2643 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 2644 | |
555353cf DM |
2645 | rxq = &dev->rx_queue; |
2646 | ||
83874000 | 2647 | q = rxq->qdisc; |
8d50b53d | 2648 | if (q != &noop_qdisc) { |
83874000 | 2649 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
2650 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
2651 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
2652 | spin_unlock(qdisc_lock(q)); |
2653 | } | |
f697c3e8 HX |
2654 | |
2655 | return result; | |
2656 | } | |
86e65da9 | 2657 | |
f697c3e8 HX |
2658 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
2659 | struct packet_type **pt_prev, | |
2660 | int *ret, struct net_device *orig_dev) | |
2661 | { | |
8d50b53d | 2662 | if (skb->dev->rx_queue.qdisc == &noop_qdisc) |
f697c3e8 | 2663 | goto out; |
1da177e4 | 2664 | |
f697c3e8 HX |
2665 | if (*pt_prev) { |
2666 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2667 | *pt_prev = NULL; | |
2668 | } else { | |
2669 | /* Huh? Why does turning on AF_PACKET affect this? */ | |
2670 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); | |
1da177e4 LT |
2671 | } |
2672 | ||
f697c3e8 HX |
2673 | switch (ing_filter(skb)) { |
2674 | case TC_ACT_SHOT: | |
2675 | case TC_ACT_STOLEN: | |
2676 | kfree_skb(skb); | |
2677 | return NULL; | |
2678 | } | |
2679 | ||
2680 | out: | |
2681 | skb->tc_verd = 0; | |
2682 | return skb; | |
1da177e4 LT |
2683 | } |
2684 | #endif | |
2685 | ||
bc1d0411 PM |
2686 | /* |
2687 | * netif_nit_deliver - deliver received packets to network taps | |
2688 | * @skb: buffer | |
2689 | * | |
2690 | * This function is used to deliver incoming packets to network | |
2691 | * taps. It should be used when the normal netif_receive_skb path | |
2692 | * is bypassed, for example because of VLAN acceleration. | |
2693 | */ | |
2694 | void netif_nit_deliver(struct sk_buff *skb) | |
2695 | { | |
2696 | struct packet_type *ptype; | |
2697 | ||
2698 | if (list_empty(&ptype_all)) | |
2699 | return; | |
2700 | ||
2701 | skb_reset_network_header(skb); | |
2702 | skb_reset_transport_header(skb); | |
2703 | skb->mac_len = skb->network_header - skb->mac_header; | |
2704 | ||
2705 | rcu_read_lock(); | |
2706 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2707 | if (!ptype->dev || ptype->dev == skb->dev) | |
2708 | deliver_skb(skb, ptype, skb->dev); | |
2709 | } | |
2710 | rcu_read_unlock(); | |
2711 | } | |
2712 | ||
acbbc071 ED |
2713 | static inline void skb_bond_set_mac_by_master(struct sk_buff *skb, |
2714 | struct net_device *master) | |
2715 | { | |
2716 | if (skb->pkt_type == PACKET_HOST) { | |
2717 | u16 *dest = (u16 *) eth_hdr(skb)->h_dest; | |
2718 | ||
2719 | memcpy(dest, master->dev_addr, ETH_ALEN); | |
2720 | } | |
2721 | } | |
2722 | ||
2723 | /* On bonding slaves other than the currently active slave, suppress | |
2724 | * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and | |
2725 | * ARP on active-backup slaves with arp_validate enabled. | |
2726 | */ | |
2727 | int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master) | |
2728 | { | |
2729 | struct net_device *dev = skb->dev; | |
2730 | ||
2731 | if (master->priv_flags & IFF_MASTER_ARPMON) | |
2732 | dev->last_rx = jiffies; | |
2733 | ||
2734 | if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) { | |
2735 | /* Do address unmangle. The local destination address | |
2736 | * will be always the one master has. Provides the right | |
2737 | * functionality in a bridge. | |
2738 | */ | |
2739 | skb_bond_set_mac_by_master(skb, master); | |
2740 | } | |
2741 | ||
2742 | if (dev->priv_flags & IFF_SLAVE_INACTIVE) { | |
2743 | if ((dev->priv_flags & IFF_SLAVE_NEEDARP) && | |
2744 | skb->protocol == __cpu_to_be16(ETH_P_ARP)) | |
2745 | return 0; | |
2746 | ||
2747 | if (master->priv_flags & IFF_MASTER_ALB) { | |
2748 | if (skb->pkt_type != PACKET_BROADCAST && | |
2749 | skb->pkt_type != PACKET_MULTICAST) | |
2750 | return 0; | |
2751 | } | |
2752 | if (master->priv_flags & IFF_MASTER_8023AD && | |
2753 | skb->protocol == __cpu_to_be16(ETH_P_SLOW)) | |
2754 | return 0; | |
2755 | ||
2756 | return 1; | |
2757 | } | |
2758 | return 0; | |
2759 | } | |
2760 | EXPORT_SYMBOL(__skb_bond_should_drop); | |
2761 | ||
10f744d2 | 2762 | static int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
2763 | { |
2764 | struct packet_type *ptype, *pt_prev; | |
f2ccd8fa | 2765 | struct net_device *orig_dev; |
0641e4fb | 2766 | struct net_device *master; |
0d7a3681 | 2767 | struct net_device *null_or_orig; |
ca8d9ea3 | 2768 | struct net_device *null_or_bond; |
1da177e4 | 2769 | int ret = NET_RX_DROP; |
252e3346 | 2770 | __be16 type; |
1da177e4 | 2771 | |
81bbb3d4 ED |
2772 | if (!skb->tstamp.tv64) |
2773 | net_timestamp(skb); | |
2774 | ||
05423b24 | 2775 | if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb)) |
9b22ea56 PM |
2776 | return NET_RX_SUCCESS; |
2777 | ||
1da177e4 | 2778 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 2779 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
2780 | return NET_RX_DROP; |
2781 | ||
8964be4a ED |
2782 | if (!skb->skb_iif) |
2783 | skb->skb_iif = skb->dev->ifindex; | |
86e65da9 | 2784 | |
0d7a3681 | 2785 | null_or_orig = NULL; |
cc9bd5ce | 2786 | orig_dev = skb->dev; |
0641e4fb ED |
2787 | master = ACCESS_ONCE(orig_dev->master); |
2788 | if (master) { | |
2789 | if (skb_bond_should_drop(skb, master)) | |
0d7a3681 JE |
2790 | null_or_orig = orig_dev; /* deliver only exact match */ |
2791 | else | |
0641e4fb | 2792 | skb->dev = master; |
cc9bd5ce | 2793 | } |
8f903c70 | 2794 | |
1da177e4 LT |
2795 | __get_cpu_var(netdev_rx_stat).total++; |
2796 | ||
c1d2bbe1 | 2797 | skb_reset_network_header(skb); |
badff6d0 | 2798 | skb_reset_transport_header(skb); |
b0e380b1 | 2799 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
2800 | |
2801 | pt_prev = NULL; | |
2802 | ||
2803 | rcu_read_lock(); | |
2804 | ||
2805 | #ifdef CONFIG_NET_CLS_ACT | |
2806 | if (skb->tc_verd & TC_NCLS) { | |
2807 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2808 | goto ncls; | |
2809 | } | |
2810 | #endif | |
2811 | ||
2812 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
f982307f JE |
2813 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
2814 | ptype->dev == orig_dev) { | |
4ec93edb | 2815 | if (pt_prev) |
f2ccd8fa | 2816 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2817 | pt_prev = ptype; |
2818 | } | |
2819 | } | |
2820 | ||
2821 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
2822 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
2823 | if (!skb) | |
1da177e4 | 2824 | goto out; |
1da177e4 LT |
2825 | ncls: |
2826 | #endif | |
2827 | ||
6229e362 | 2828 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); |
b863ceb7 PM |
2829 | if (!skb) |
2830 | goto out; | |
2831 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); | |
6229e362 | 2832 | if (!skb) |
1da177e4 LT |
2833 | goto out; |
2834 | ||
1f3c8804 AG |
2835 | /* |
2836 | * Make sure frames received on VLAN interfaces stacked on | |
2837 | * bonding interfaces still make their way to any base bonding | |
2838 | * device that may have registered for a specific ptype. The | |
2839 | * handler may have to adjust skb->dev and orig_dev. | |
1f3c8804 | 2840 | */ |
ca8d9ea3 | 2841 | null_or_bond = NULL; |
1f3c8804 AG |
2842 | if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) && |
2843 | (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) { | |
ca8d9ea3 | 2844 | null_or_bond = vlan_dev_real_dev(skb->dev); |
1f3c8804 AG |
2845 | } |
2846 | ||
1da177e4 | 2847 | type = skb->protocol; |
82d8a867 PE |
2848 | list_for_each_entry_rcu(ptype, |
2849 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1f3c8804 | 2850 | if (ptype->type == type && (ptype->dev == null_or_orig || |
ca8d9ea3 AG |
2851 | ptype->dev == skb->dev || ptype->dev == orig_dev || |
2852 | ptype->dev == null_or_bond)) { | |
4ec93edb | 2853 | if (pt_prev) |
f2ccd8fa | 2854 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2855 | pt_prev = ptype; |
2856 | } | |
2857 | } | |
2858 | ||
2859 | if (pt_prev) { | |
f2ccd8fa | 2860 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2861 | } else { |
2862 | kfree_skb(skb); | |
2863 | /* Jamal, now you will not able to escape explaining | |
2864 | * me how you were going to use this. :-) | |
2865 | */ | |
2866 | ret = NET_RX_DROP; | |
2867 | } | |
2868 | ||
2869 | out: | |
2870 | rcu_read_unlock(); | |
2871 | return ret; | |
2872 | } | |
0a9627f2 TH |
2873 | |
2874 | /** | |
2875 | * netif_receive_skb - process receive buffer from network | |
2876 | * @skb: buffer to process | |
2877 | * | |
2878 | * netif_receive_skb() is the main receive data processing function. | |
2879 | * It always succeeds. The buffer may be dropped during processing | |
2880 | * for congestion control or by the protocol layers. | |
2881 | * | |
2882 | * This function may only be called from softirq context and interrupts | |
2883 | * should be enabled. | |
2884 | * | |
2885 | * Return values (usually ignored): | |
2886 | * NET_RX_SUCCESS: no congestion | |
2887 | * NET_RX_DROP: packet was dropped | |
2888 | */ | |
2889 | int netif_receive_skb(struct sk_buff *skb) | |
2890 | { | |
df334545 | 2891 | #ifdef CONFIG_RPS |
fec5e652 TH |
2892 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2893 | int cpu, ret; | |
2894 | ||
2895 | rcu_read_lock(); | |
0a9627f2 | 2896 | |
fec5e652 | 2897 | cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 2898 | |
fec5e652 TH |
2899 | if (cpu >= 0) { |
2900 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
2901 | rcu_read_unlock(); | |
2902 | } else { | |
2903 | rcu_read_unlock(); | |
2904 | ret = __netif_receive_skb(skb); | |
2905 | } | |
2906 | ||
2907 | return ret; | |
1e94d72f TH |
2908 | #else |
2909 | return __netif_receive_skb(skb); | |
2910 | #endif | |
0a9627f2 | 2911 | } |
d1b19dff | 2912 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 2913 | |
6e583ce5 | 2914 | /* Network device is going away, flush any packets still pending */ |
152102c7 | 2915 | static void flush_backlog(void *arg) |
6e583ce5 | 2916 | { |
152102c7 CG |
2917 | struct net_device *dev = arg; |
2918 | struct softnet_data *queue = &__get_cpu_var(softnet_data); | |
6e583ce5 SH |
2919 | struct sk_buff *skb, *tmp; |
2920 | ||
152102c7 | 2921 | rps_lock(queue); |
6e583ce5 SH |
2922 | skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp) |
2923 | if (skb->dev == dev) { | |
2924 | __skb_unlink(skb, &queue->input_pkt_queue); | |
2925 | kfree_skb(skb); | |
fec5e652 | 2926 | incr_input_queue_head(queue); |
6e583ce5 | 2927 | } |
152102c7 | 2928 | rps_unlock(queue); |
6e583ce5 SH |
2929 | } |
2930 | ||
d565b0a1 HX |
2931 | static int napi_gro_complete(struct sk_buff *skb) |
2932 | { | |
2933 | struct packet_type *ptype; | |
2934 | __be16 type = skb->protocol; | |
2935 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
2936 | int err = -ENOENT; | |
2937 | ||
fc59f9a3 HX |
2938 | if (NAPI_GRO_CB(skb)->count == 1) { |
2939 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 2940 | goto out; |
fc59f9a3 | 2941 | } |
d565b0a1 HX |
2942 | |
2943 | rcu_read_lock(); | |
2944 | list_for_each_entry_rcu(ptype, head, list) { | |
2945 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
2946 | continue; | |
2947 | ||
2948 | err = ptype->gro_complete(skb); | |
2949 | break; | |
2950 | } | |
2951 | rcu_read_unlock(); | |
2952 | ||
2953 | if (err) { | |
2954 | WARN_ON(&ptype->list == head); | |
2955 | kfree_skb(skb); | |
2956 | return NET_RX_SUCCESS; | |
2957 | } | |
2958 | ||
2959 | out: | |
d565b0a1 HX |
2960 | return netif_receive_skb(skb); |
2961 | } | |
2962 | ||
11380a4b | 2963 | static void napi_gro_flush(struct napi_struct *napi) |
d565b0a1 HX |
2964 | { |
2965 | struct sk_buff *skb, *next; | |
2966 | ||
2967 | for (skb = napi->gro_list; skb; skb = next) { | |
2968 | next = skb->next; | |
2969 | skb->next = NULL; | |
2970 | napi_gro_complete(skb); | |
2971 | } | |
2972 | ||
4ae5544f | 2973 | napi->gro_count = 0; |
d565b0a1 HX |
2974 | napi->gro_list = NULL; |
2975 | } | |
d565b0a1 | 2976 | |
5b252f0c | 2977 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
2978 | { |
2979 | struct sk_buff **pp = NULL; | |
2980 | struct packet_type *ptype; | |
2981 | __be16 type = skb->protocol; | |
2982 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 2983 | int same_flow; |
d565b0a1 | 2984 | int mac_len; |
5b252f0c | 2985 | enum gro_result ret; |
d565b0a1 HX |
2986 | |
2987 | if (!(skb->dev->features & NETIF_F_GRO)) | |
2988 | goto normal; | |
2989 | ||
4cf704fb | 2990 | if (skb_is_gso(skb) || skb_has_frags(skb)) |
f17f5c91 HX |
2991 | goto normal; |
2992 | ||
d565b0a1 HX |
2993 | rcu_read_lock(); |
2994 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
2995 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
2996 | continue; | |
2997 | ||
86911732 | 2998 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
2999 | mac_len = skb->network_header - skb->mac_header; |
3000 | skb->mac_len = mac_len; | |
3001 | NAPI_GRO_CB(skb)->same_flow = 0; | |
3002 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 3003 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 3004 | |
d565b0a1 HX |
3005 | pp = ptype->gro_receive(&napi->gro_list, skb); |
3006 | break; | |
3007 | } | |
3008 | rcu_read_unlock(); | |
3009 | ||
3010 | if (&ptype->list == head) | |
3011 | goto normal; | |
3012 | ||
0da2afd5 | 3013 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 3014 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 3015 | |
d565b0a1 HX |
3016 | if (pp) { |
3017 | struct sk_buff *nskb = *pp; | |
3018 | ||
3019 | *pp = nskb->next; | |
3020 | nskb->next = NULL; | |
3021 | napi_gro_complete(nskb); | |
4ae5544f | 3022 | napi->gro_count--; |
d565b0a1 HX |
3023 | } |
3024 | ||
0da2afd5 | 3025 | if (same_flow) |
d565b0a1 HX |
3026 | goto ok; |
3027 | ||
4ae5544f | 3028 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 3029 | goto normal; |
d565b0a1 | 3030 | |
4ae5544f | 3031 | napi->gro_count++; |
d565b0a1 | 3032 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 3033 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
3034 | skb->next = napi->gro_list; |
3035 | napi->gro_list = skb; | |
5d0d9be8 | 3036 | ret = GRO_HELD; |
d565b0a1 | 3037 | |
ad0f9904 | 3038 | pull: |
cb18978c HX |
3039 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
3040 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
3041 | ||
3042 | BUG_ON(skb->end - skb->tail < grow); | |
3043 | ||
3044 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
3045 | ||
3046 | skb->tail += grow; | |
3047 | skb->data_len -= grow; | |
3048 | ||
3049 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
3050 | skb_shinfo(skb)->frags[0].size -= grow; | |
3051 | ||
3052 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) { | |
3053 | put_page(skb_shinfo(skb)->frags[0].page); | |
3054 | memmove(skb_shinfo(skb)->frags, | |
3055 | skb_shinfo(skb)->frags + 1, | |
3056 | --skb_shinfo(skb)->nr_frags); | |
3057 | } | |
ad0f9904 HX |
3058 | } |
3059 | ||
d565b0a1 | 3060 | ok: |
5d0d9be8 | 3061 | return ret; |
d565b0a1 HX |
3062 | |
3063 | normal: | |
ad0f9904 HX |
3064 | ret = GRO_NORMAL; |
3065 | goto pull; | |
5d38a079 | 3066 | } |
96e93eab HX |
3067 | EXPORT_SYMBOL(dev_gro_receive); |
3068 | ||
5b252f0c BH |
3069 | static gro_result_t |
3070 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
96e93eab HX |
3071 | { |
3072 | struct sk_buff *p; | |
3073 | ||
d1c76af9 HX |
3074 | if (netpoll_rx_on(skb)) |
3075 | return GRO_NORMAL; | |
3076 | ||
96e93eab | 3077 | for (p = napi->gro_list; p; p = p->next) { |
f64f9e71 JP |
3078 | NAPI_GRO_CB(p)->same_flow = |
3079 | (p->dev == skb->dev) && | |
3080 | !compare_ether_header(skb_mac_header(p), | |
3081 | skb_gro_mac_header(skb)); | |
96e93eab HX |
3082 | NAPI_GRO_CB(p)->flush = 0; |
3083 | } | |
3084 | ||
3085 | return dev_gro_receive(napi, skb); | |
3086 | } | |
5d38a079 | 3087 | |
c7c4b3b6 | 3088 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3089 | { |
5d0d9be8 HX |
3090 | switch (ret) { |
3091 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3092 | if (netif_receive_skb(skb)) |
3093 | ret = GRO_DROP; | |
3094 | break; | |
5d38a079 | 3095 | |
5d0d9be8 | 3096 | case GRO_DROP: |
5d0d9be8 | 3097 | case GRO_MERGED_FREE: |
5d38a079 HX |
3098 | kfree_skb(skb); |
3099 | break; | |
5b252f0c BH |
3100 | |
3101 | case GRO_HELD: | |
3102 | case GRO_MERGED: | |
3103 | break; | |
5d38a079 HX |
3104 | } |
3105 | ||
c7c4b3b6 | 3106 | return ret; |
5d0d9be8 HX |
3107 | } |
3108 | EXPORT_SYMBOL(napi_skb_finish); | |
3109 | ||
78a478d0 HX |
3110 | void skb_gro_reset_offset(struct sk_buff *skb) |
3111 | { | |
3112 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3113 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3114 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3115 | |
78d3fd0b | 3116 | if (skb->mac_header == skb->tail && |
7489594c | 3117 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) { |
78a478d0 HX |
3118 | NAPI_GRO_CB(skb)->frag0 = |
3119 | page_address(skb_shinfo(skb)->frags[0].page) + | |
3120 | skb_shinfo(skb)->frags[0].page_offset; | |
7489594c HX |
3121 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size; |
3122 | } | |
78a478d0 HX |
3123 | } |
3124 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
3125 | ||
c7c4b3b6 | 3126 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3127 | { |
86911732 HX |
3128 | skb_gro_reset_offset(skb); |
3129 | ||
5d0d9be8 | 3130 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3131 | } |
3132 | EXPORT_SYMBOL(napi_gro_receive); | |
3133 | ||
96e93eab HX |
3134 | void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
3135 | { | |
96e93eab HX |
3136 | __skb_pull(skb, skb_headlen(skb)); |
3137 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
3138 | ||
3139 | napi->skb = skb; | |
3140 | } | |
3141 | EXPORT_SYMBOL(napi_reuse_skb); | |
3142 | ||
76620aaf | 3143 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3144 | { |
5d38a079 | 3145 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3146 | |
3147 | if (!skb) { | |
89d71a66 ED |
3148 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3149 | if (skb) | |
3150 | napi->skb = skb; | |
80595d59 | 3151 | } |
96e93eab HX |
3152 | return skb; |
3153 | } | |
76620aaf | 3154 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3155 | |
c7c4b3b6 BH |
3156 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
3157 | gro_result_t ret) | |
96e93eab | 3158 | { |
5d0d9be8 HX |
3159 | switch (ret) { |
3160 | case GRO_NORMAL: | |
86911732 | 3161 | case GRO_HELD: |
e76b69cc | 3162 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3163 | |
c7c4b3b6 BH |
3164 | if (ret == GRO_HELD) |
3165 | skb_gro_pull(skb, -ETH_HLEN); | |
3166 | else if (netif_receive_skb(skb)) | |
3167 | ret = GRO_DROP; | |
86911732 | 3168 | break; |
5d38a079 | 3169 | |
5d0d9be8 | 3170 | case GRO_DROP: |
5d0d9be8 HX |
3171 | case GRO_MERGED_FREE: |
3172 | napi_reuse_skb(napi, skb); | |
3173 | break; | |
5b252f0c BH |
3174 | |
3175 | case GRO_MERGED: | |
3176 | break; | |
5d0d9be8 | 3177 | } |
5d38a079 | 3178 | |
c7c4b3b6 | 3179 | return ret; |
5d38a079 | 3180 | } |
5d0d9be8 HX |
3181 | EXPORT_SYMBOL(napi_frags_finish); |
3182 | ||
76620aaf HX |
3183 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
3184 | { | |
3185 | struct sk_buff *skb = napi->skb; | |
3186 | struct ethhdr *eth; | |
a5b1cf28 HX |
3187 | unsigned int hlen; |
3188 | unsigned int off; | |
76620aaf HX |
3189 | |
3190 | napi->skb = NULL; | |
3191 | ||
3192 | skb_reset_mac_header(skb); | |
3193 | skb_gro_reset_offset(skb); | |
3194 | ||
a5b1cf28 HX |
3195 | off = skb_gro_offset(skb); |
3196 | hlen = off + sizeof(*eth); | |
3197 | eth = skb_gro_header_fast(skb, off); | |
3198 | if (skb_gro_header_hard(skb, hlen)) { | |
3199 | eth = skb_gro_header_slow(skb, hlen, off); | |
3200 | if (unlikely(!eth)) { | |
3201 | napi_reuse_skb(napi, skb); | |
3202 | skb = NULL; | |
3203 | goto out; | |
3204 | } | |
76620aaf HX |
3205 | } |
3206 | ||
3207 | skb_gro_pull(skb, sizeof(*eth)); | |
3208 | ||
3209 | /* | |
3210 | * This works because the only protocols we care about don't require | |
3211 | * special handling. We'll fix it up properly at the end. | |
3212 | */ | |
3213 | skb->protocol = eth->h_proto; | |
3214 | ||
3215 | out: | |
3216 | return skb; | |
3217 | } | |
3218 | EXPORT_SYMBOL(napi_frags_skb); | |
3219 | ||
c7c4b3b6 | 3220 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3221 | { |
76620aaf | 3222 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3223 | |
3224 | if (!skb) | |
c7c4b3b6 | 3225 | return GRO_DROP; |
5d0d9be8 HX |
3226 | |
3227 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3228 | } | |
5d38a079 HX |
3229 | EXPORT_SYMBOL(napi_gro_frags); |
3230 | ||
bea3348e | 3231 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3232 | { |
3233 | int work = 0; | |
1da177e4 LT |
3234 | struct softnet_data *queue = &__get_cpu_var(softnet_data); |
3235 | unsigned long start_time = jiffies; | |
3236 | ||
bea3348e SH |
3237 | napi->weight = weight_p; |
3238 | do { | |
1da177e4 | 3239 | struct sk_buff *skb; |
1da177e4 | 3240 | |
152102c7 CG |
3241 | local_irq_disable(); |
3242 | rps_lock(queue); | |
1da177e4 | 3243 | skb = __skb_dequeue(&queue->input_pkt_queue); |
bea3348e | 3244 | if (!skb) { |
8f1ead2d | 3245 | __napi_complete(napi); |
5a6d234e | 3246 | rps_unlock(queue); |
e4008276 | 3247 | local_irq_enable(); |
8f1ead2d | 3248 | break; |
bea3348e | 3249 | } |
fec5e652 | 3250 | incr_input_queue_head(queue); |
152102c7 CG |
3251 | rps_unlock(queue); |
3252 | local_irq_enable(); | |
1da177e4 | 3253 | |
0a9627f2 | 3254 | __netif_receive_skb(skb); |
bea3348e | 3255 | } while (++work < quota && jiffies == start_time); |
1da177e4 | 3256 | |
bea3348e SH |
3257 | return work; |
3258 | } | |
1da177e4 | 3259 | |
bea3348e SH |
3260 | /** |
3261 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 3262 | * @n: entry to schedule |
bea3348e SH |
3263 | * |
3264 | * The entry's receive function will be scheduled to run | |
3265 | */ | |
b5606c2d | 3266 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
3267 | { |
3268 | unsigned long flags; | |
1da177e4 | 3269 | |
bea3348e SH |
3270 | local_irq_save(flags); |
3271 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); | |
3272 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3273 | local_irq_restore(flags); | |
1da177e4 | 3274 | } |
bea3348e SH |
3275 | EXPORT_SYMBOL(__napi_schedule); |
3276 | ||
d565b0a1 HX |
3277 | void __napi_complete(struct napi_struct *n) |
3278 | { | |
3279 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3280 | BUG_ON(n->gro_list); | |
3281 | ||
3282 | list_del(&n->poll_list); | |
3283 | smp_mb__before_clear_bit(); | |
3284 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3285 | } | |
3286 | EXPORT_SYMBOL(__napi_complete); | |
3287 | ||
3288 | void napi_complete(struct napi_struct *n) | |
3289 | { | |
3290 | unsigned long flags; | |
3291 | ||
3292 | /* | |
3293 | * don't let napi dequeue from the cpu poll list | |
3294 | * just in case its running on a different cpu | |
3295 | */ | |
3296 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3297 | return; | |
3298 | ||
3299 | napi_gro_flush(n); | |
3300 | local_irq_save(flags); | |
3301 | __napi_complete(n); | |
3302 | local_irq_restore(flags); | |
3303 | } | |
3304 | EXPORT_SYMBOL(napi_complete); | |
3305 | ||
3306 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3307 | int (*poll)(struct napi_struct *, int), int weight) | |
3308 | { | |
3309 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 3310 | napi->gro_count = 0; |
d565b0a1 | 3311 | napi->gro_list = NULL; |
5d38a079 | 3312 | napi->skb = NULL; |
d565b0a1 HX |
3313 | napi->poll = poll; |
3314 | napi->weight = weight; | |
3315 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 3316 | napi->dev = dev; |
5d38a079 | 3317 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
3318 | spin_lock_init(&napi->poll_lock); |
3319 | napi->poll_owner = -1; | |
3320 | #endif | |
3321 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3322 | } | |
3323 | EXPORT_SYMBOL(netif_napi_add); | |
3324 | ||
3325 | void netif_napi_del(struct napi_struct *napi) | |
3326 | { | |
3327 | struct sk_buff *skb, *next; | |
3328 | ||
d7b06636 | 3329 | list_del_init(&napi->dev_list); |
76620aaf | 3330 | napi_free_frags(napi); |
d565b0a1 HX |
3331 | |
3332 | for (skb = napi->gro_list; skb; skb = next) { | |
3333 | next = skb->next; | |
3334 | skb->next = NULL; | |
3335 | kfree_skb(skb); | |
3336 | } | |
3337 | ||
3338 | napi->gro_list = NULL; | |
4ae5544f | 3339 | napi->gro_count = 0; |
d565b0a1 HX |
3340 | } |
3341 | EXPORT_SYMBOL(netif_napi_del); | |
3342 | ||
df334545 | 3343 | #ifdef CONFIG_RPS |
0a9627f2 TH |
3344 | /* |
3345 | * net_rps_action sends any pending IPI's for rps. This is only called from | |
3346 | * softirq and interrupts must be enabled. | |
3347 | */ | |
3348 | static void net_rps_action(cpumask_t *mask) | |
3349 | { | |
3350 | int cpu; | |
3351 | ||
3352 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3353 | for_each_cpu_mask_nr(cpu, *mask) { | |
3354 | struct softnet_data *queue = &per_cpu(softnet_data, cpu); | |
3355 | if (cpu_online(cpu)) | |
3356 | __smp_call_function_single(cpu, &queue->csd, 0); | |
3357 | } | |
3358 | cpus_clear(*mask); | |
3359 | } | |
1e94d72f | 3360 | #endif |
1da177e4 LT |
3361 | |
3362 | static void net_rx_action(struct softirq_action *h) | |
3363 | { | |
bea3348e | 3364 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list; |
24f8b238 | 3365 | unsigned long time_limit = jiffies + 2; |
51b0bded | 3366 | int budget = netdev_budget; |
53fb95d3 | 3367 | void *have; |
df334545 | 3368 | #ifdef CONFIG_RPS |
0a9627f2 TH |
3369 | int select; |
3370 | struct rps_remote_softirq_cpus *rcpus; | |
1e94d72f | 3371 | #endif |
53fb95d3 | 3372 | |
1da177e4 LT |
3373 | local_irq_disable(); |
3374 | ||
bea3348e SH |
3375 | while (!list_empty(list)) { |
3376 | struct napi_struct *n; | |
3377 | int work, weight; | |
1da177e4 | 3378 | |
bea3348e | 3379 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
3380 | * Allow this to run for 2 jiffies since which will allow |
3381 | * an average latency of 1.5/HZ. | |
bea3348e | 3382 | */ |
24f8b238 | 3383 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
3384 | goto softnet_break; |
3385 | ||
3386 | local_irq_enable(); | |
3387 | ||
bea3348e SH |
3388 | /* Even though interrupts have been re-enabled, this |
3389 | * access is safe because interrupts can only add new | |
3390 | * entries to the tail of this list, and only ->poll() | |
3391 | * calls can remove this head entry from the list. | |
3392 | */ | |
e5e26d75 | 3393 | n = list_first_entry(list, struct napi_struct, poll_list); |
1da177e4 | 3394 | |
bea3348e SH |
3395 | have = netpoll_poll_lock(n); |
3396 | ||
3397 | weight = n->weight; | |
3398 | ||
0a7606c1 DM |
3399 | /* This NAPI_STATE_SCHED test is for avoiding a race |
3400 | * with netpoll's poll_napi(). Only the entity which | |
3401 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3402 | * actually make the ->poll() call. Therefore we avoid | |
3403 | * accidently calling ->poll() when NAPI is not scheduled. | |
3404 | */ | |
3405 | work = 0; | |
4ea7e386 | 3406 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3407 | work = n->poll(n, weight); |
4ea7e386 NH |
3408 | trace_napi_poll(n); |
3409 | } | |
bea3348e SH |
3410 | |
3411 | WARN_ON_ONCE(work > weight); | |
3412 | ||
3413 | budget -= work; | |
3414 | ||
3415 | local_irq_disable(); | |
3416 | ||
3417 | /* Drivers must not modify the NAPI state if they | |
3418 | * consume the entire weight. In such cases this code | |
3419 | * still "owns" the NAPI instance and therefore can | |
3420 | * move the instance around on the list at-will. | |
3421 | */ | |
fed17f30 | 3422 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3423 | if (unlikely(napi_disable_pending(n))) { |
3424 | local_irq_enable(); | |
3425 | napi_complete(n); | |
3426 | local_irq_disable(); | |
3427 | } else | |
fed17f30 DM |
3428 | list_move_tail(&n->poll_list, list); |
3429 | } | |
bea3348e SH |
3430 | |
3431 | netpoll_poll_unlock(have); | |
1da177e4 LT |
3432 | } |
3433 | out: | |
df334545 | 3434 | #ifdef CONFIG_RPS |
0a9627f2 TH |
3435 | rcpus = &__get_cpu_var(rps_remote_softirq_cpus); |
3436 | select = rcpus->select; | |
3437 | rcpus->select ^= 1; | |
3438 | ||
515e06c4 | 3439 | local_irq_enable(); |
bea3348e | 3440 | |
0a9627f2 | 3441 | net_rps_action(&rcpus->mask[select]); |
1e94d72f TH |
3442 | #else |
3443 | local_irq_enable(); | |
3444 | #endif | |
0a9627f2 | 3445 | |
db217334 CL |
3446 | #ifdef CONFIG_NET_DMA |
3447 | /* | |
3448 | * There may not be any more sk_buffs coming right now, so push | |
3449 | * any pending DMA copies to hardware | |
3450 | */ | |
2ba05622 | 3451 | dma_issue_pending_all(); |
db217334 | 3452 | #endif |
bea3348e | 3453 | |
1da177e4 LT |
3454 | return; |
3455 | ||
3456 | softnet_break: | |
3457 | __get_cpu_var(netdev_rx_stat).time_squeeze++; | |
3458 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3459 | goto out; | |
3460 | } | |
3461 | ||
d1b19dff | 3462 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
3463 | |
3464 | /** | |
3465 | * register_gifconf - register a SIOCGIF handler | |
3466 | * @family: Address family | |
3467 | * @gifconf: Function handler | |
3468 | * | |
3469 | * Register protocol dependent address dumping routines. The handler | |
3470 | * that is passed must not be freed or reused until it has been replaced | |
3471 | * by another handler. | |
3472 | */ | |
d1b19dff | 3473 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
3474 | { |
3475 | if (family >= NPROTO) | |
3476 | return -EINVAL; | |
3477 | gifconf_list[family] = gifconf; | |
3478 | return 0; | |
3479 | } | |
d1b19dff | 3480 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
3481 | |
3482 | ||
3483 | /* | |
3484 | * Map an interface index to its name (SIOCGIFNAME) | |
3485 | */ | |
3486 | ||
3487 | /* | |
3488 | * We need this ioctl for efficient implementation of the | |
3489 | * if_indextoname() function required by the IPv6 API. Without | |
3490 | * it, we would have to search all the interfaces to find a | |
3491 | * match. --pb | |
3492 | */ | |
3493 | ||
881d966b | 3494 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
3495 | { |
3496 | struct net_device *dev; | |
3497 | struct ifreq ifr; | |
3498 | ||
3499 | /* | |
3500 | * Fetch the caller's info block. | |
3501 | */ | |
3502 | ||
3503 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3504 | return -EFAULT; | |
3505 | ||
fb699dfd ED |
3506 | rcu_read_lock(); |
3507 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 3508 | if (!dev) { |
fb699dfd | 3509 | rcu_read_unlock(); |
1da177e4 LT |
3510 | return -ENODEV; |
3511 | } | |
3512 | ||
3513 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 3514 | rcu_read_unlock(); |
1da177e4 LT |
3515 | |
3516 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3517 | return -EFAULT; | |
3518 | return 0; | |
3519 | } | |
3520 | ||
3521 | /* | |
3522 | * Perform a SIOCGIFCONF call. This structure will change | |
3523 | * size eventually, and there is nothing I can do about it. | |
3524 | * Thus we will need a 'compatibility mode'. | |
3525 | */ | |
3526 | ||
881d966b | 3527 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
3528 | { |
3529 | struct ifconf ifc; | |
3530 | struct net_device *dev; | |
3531 | char __user *pos; | |
3532 | int len; | |
3533 | int total; | |
3534 | int i; | |
3535 | ||
3536 | /* | |
3537 | * Fetch the caller's info block. | |
3538 | */ | |
3539 | ||
3540 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3541 | return -EFAULT; | |
3542 | ||
3543 | pos = ifc.ifc_buf; | |
3544 | len = ifc.ifc_len; | |
3545 | ||
3546 | /* | |
3547 | * Loop over the interfaces, and write an info block for each. | |
3548 | */ | |
3549 | ||
3550 | total = 0; | |
881d966b | 3551 | for_each_netdev(net, dev) { |
1da177e4 LT |
3552 | for (i = 0; i < NPROTO; i++) { |
3553 | if (gifconf_list[i]) { | |
3554 | int done; | |
3555 | if (!pos) | |
3556 | done = gifconf_list[i](dev, NULL, 0); | |
3557 | else | |
3558 | done = gifconf_list[i](dev, pos + total, | |
3559 | len - total); | |
3560 | if (done < 0) | |
3561 | return -EFAULT; | |
3562 | total += done; | |
3563 | } | |
3564 | } | |
4ec93edb | 3565 | } |
1da177e4 LT |
3566 | |
3567 | /* | |
3568 | * All done. Write the updated control block back to the caller. | |
3569 | */ | |
3570 | ifc.ifc_len = total; | |
3571 | ||
3572 | /* | |
3573 | * Both BSD and Solaris return 0 here, so we do too. | |
3574 | */ | |
3575 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
3576 | } | |
3577 | ||
3578 | #ifdef CONFIG_PROC_FS | |
3579 | /* | |
3580 | * This is invoked by the /proc filesystem handler to display a device | |
3581 | * in detail. | |
3582 | */ | |
7562f876 | 3583 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 3584 | __acquires(RCU) |
1da177e4 | 3585 | { |
e372c414 | 3586 | struct net *net = seq_file_net(seq); |
7562f876 | 3587 | loff_t off; |
1da177e4 | 3588 | struct net_device *dev; |
1da177e4 | 3589 | |
c6d14c84 | 3590 | rcu_read_lock(); |
7562f876 PE |
3591 | if (!*pos) |
3592 | return SEQ_START_TOKEN; | |
1da177e4 | 3593 | |
7562f876 | 3594 | off = 1; |
c6d14c84 | 3595 | for_each_netdev_rcu(net, dev) |
7562f876 PE |
3596 | if (off++ == *pos) |
3597 | return dev; | |
1da177e4 | 3598 | |
7562f876 | 3599 | return NULL; |
1da177e4 LT |
3600 | } |
3601 | ||
3602 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3603 | { | |
c6d14c84 ED |
3604 | struct net_device *dev = (v == SEQ_START_TOKEN) ? |
3605 | first_net_device(seq_file_net(seq)) : | |
3606 | next_net_device((struct net_device *)v); | |
3607 | ||
1da177e4 | 3608 | ++*pos; |
c6d14c84 | 3609 | return rcu_dereference(dev); |
1da177e4 LT |
3610 | } |
3611 | ||
3612 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 3613 | __releases(RCU) |
1da177e4 | 3614 | { |
c6d14c84 | 3615 | rcu_read_unlock(); |
1da177e4 LT |
3616 | } |
3617 | ||
3618 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
3619 | { | |
eeda3fd6 | 3620 | const struct net_device_stats *stats = dev_get_stats(dev); |
1da177e4 | 3621 | |
2d13bafe | 3622 | seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " |
5a1b5898 RR |
3623 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", |
3624 | dev->name, stats->rx_bytes, stats->rx_packets, | |
3625 | stats->rx_errors, | |
3626 | stats->rx_dropped + stats->rx_missed_errors, | |
3627 | stats->rx_fifo_errors, | |
3628 | stats->rx_length_errors + stats->rx_over_errors + | |
3629 | stats->rx_crc_errors + stats->rx_frame_errors, | |
3630 | stats->rx_compressed, stats->multicast, | |
3631 | stats->tx_bytes, stats->tx_packets, | |
3632 | stats->tx_errors, stats->tx_dropped, | |
3633 | stats->tx_fifo_errors, stats->collisions, | |
3634 | stats->tx_carrier_errors + | |
3635 | stats->tx_aborted_errors + | |
3636 | stats->tx_window_errors + | |
3637 | stats->tx_heartbeat_errors, | |
3638 | stats->tx_compressed); | |
1da177e4 LT |
3639 | } |
3640 | ||
3641 | /* | |
3642 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
3643 | * /proc/net interface to create /proc/net/dev | |
3644 | */ | |
3645 | static int dev_seq_show(struct seq_file *seq, void *v) | |
3646 | { | |
3647 | if (v == SEQ_START_TOKEN) | |
3648 | seq_puts(seq, "Inter-| Receive " | |
3649 | " | Transmit\n" | |
3650 | " face |bytes packets errs drop fifo frame " | |
3651 | "compressed multicast|bytes packets errs " | |
3652 | "drop fifo colls carrier compressed\n"); | |
3653 | else | |
3654 | dev_seq_printf_stats(seq, v); | |
3655 | return 0; | |
3656 | } | |
3657 | ||
3658 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) | |
3659 | { | |
3660 | struct netif_rx_stats *rc = NULL; | |
3661 | ||
0c0b0aca | 3662 | while (*pos < nr_cpu_ids) |
4ec93edb | 3663 | if (cpu_online(*pos)) { |
1da177e4 LT |
3664 | rc = &per_cpu(netdev_rx_stat, *pos); |
3665 | break; | |
3666 | } else | |
3667 | ++*pos; | |
3668 | return rc; | |
3669 | } | |
3670 | ||
3671 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
3672 | { | |
3673 | return softnet_get_online(pos); | |
3674 | } | |
3675 | ||
3676 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3677 | { | |
3678 | ++*pos; | |
3679 | return softnet_get_online(pos); | |
3680 | } | |
3681 | ||
3682 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
3683 | { | |
3684 | } | |
3685 | ||
3686 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
3687 | { | |
3688 | struct netif_rx_stats *s = v; | |
3689 | ||
0a9627f2 | 3690 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
31aa02c5 | 3691 | s->total, s->dropped, s->time_squeeze, 0, |
c1ebcdb8 | 3692 | 0, 0, 0, 0, /* was fastroute */ |
0a9627f2 | 3693 | s->cpu_collision, s->received_rps); |
1da177e4 LT |
3694 | return 0; |
3695 | } | |
3696 | ||
f690808e | 3697 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
3698 | .start = dev_seq_start, |
3699 | .next = dev_seq_next, | |
3700 | .stop = dev_seq_stop, | |
3701 | .show = dev_seq_show, | |
3702 | }; | |
3703 | ||
3704 | static int dev_seq_open(struct inode *inode, struct file *file) | |
3705 | { | |
e372c414 DL |
3706 | return seq_open_net(inode, file, &dev_seq_ops, |
3707 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
3708 | } |
3709 | ||
9a32144e | 3710 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
3711 | .owner = THIS_MODULE, |
3712 | .open = dev_seq_open, | |
3713 | .read = seq_read, | |
3714 | .llseek = seq_lseek, | |
e372c414 | 3715 | .release = seq_release_net, |
1da177e4 LT |
3716 | }; |
3717 | ||
f690808e | 3718 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
3719 | .start = softnet_seq_start, |
3720 | .next = softnet_seq_next, | |
3721 | .stop = softnet_seq_stop, | |
3722 | .show = softnet_seq_show, | |
3723 | }; | |
3724 | ||
3725 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
3726 | { | |
3727 | return seq_open(file, &softnet_seq_ops); | |
3728 | } | |
3729 | ||
9a32144e | 3730 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
3731 | .owner = THIS_MODULE, |
3732 | .open = softnet_seq_open, | |
3733 | .read = seq_read, | |
3734 | .llseek = seq_lseek, | |
3735 | .release = seq_release, | |
3736 | }; | |
3737 | ||
0e1256ff SH |
3738 | static void *ptype_get_idx(loff_t pos) |
3739 | { | |
3740 | struct packet_type *pt = NULL; | |
3741 | loff_t i = 0; | |
3742 | int t; | |
3743 | ||
3744 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
3745 | if (i == pos) | |
3746 | return pt; | |
3747 | ++i; | |
3748 | } | |
3749 | ||
82d8a867 | 3750 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
3751 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
3752 | if (i == pos) | |
3753 | return pt; | |
3754 | ++i; | |
3755 | } | |
3756 | } | |
3757 | return NULL; | |
3758 | } | |
3759 | ||
3760 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 3761 | __acquires(RCU) |
0e1256ff SH |
3762 | { |
3763 | rcu_read_lock(); | |
3764 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
3765 | } | |
3766 | ||
3767 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3768 | { | |
3769 | struct packet_type *pt; | |
3770 | struct list_head *nxt; | |
3771 | int hash; | |
3772 | ||
3773 | ++*pos; | |
3774 | if (v == SEQ_START_TOKEN) | |
3775 | return ptype_get_idx(0); | |
3776 | ||
3777 | pt = v; | |
3778 | nxt = pt->list.next; | |
3779 | if (pt->type == htons(ETH_P_ALL)) { | |
3780 | if (nxt != &ptype_all) | |
3781 | goto found; | |
3782 | hash = 0; | |
3783 | nxt = ptype_base[0].next; | |
3784 | } else | |
82d8a867 | 3785 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
3786 | |
3787 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 3788 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
3789 | return NULL; |
3790 | nxt = ptype_base[hash].next; | |
3791 | } | |
3792 | found: | |
3793 | return list_entry(nxt, struct packet_type, list); | |
3794 | } | |
3795 | ||
3796 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 3797 | __releases(RCU) |
0e1256ff SH |
3798 | { |
3799 | rcu_read_unlock(); | |
3800 | } | |
3801 | ||
0e1256ff SH |
3802 | static int ptype_seq_show(struct seq_file *seq, void *v) |
3803 | { | |
3804 | struct packet_type *pt = v; | |
3805 | ||
3806 | if (v == SEQ_START_TOKEN) | |
3807 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 3808 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
3809 | if (pt->type == htons(ETH_P_ALL)) |
3810 | seq_puts(seq, "ALL "); | |
3811 | else | |
3812 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
3813 | ||
908cd2da AD |
3814 | seq_printf(seq, " %-8s %pF\n", |
3815 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
3816 | } |
3817 | ||
3818 | return 0; | |
3819 | } | |
3820 | ||
3821 | static const struct seq_operations ptype_seq_ops = { | |
3822 | .start = ptype_seq_start, | |
3823 | .next = ptype_seq_next, | |
3824 | .stop = ptype_seq_stop, | |
3825 | .show = ptype_seq_show, | |
3826 | }; | |
3827 | ||
3828 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
3829 | { | |
2feb27db PE |
3830 | return seq_open_net(inode, file, &ptype_seq_ops, |
3831 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
3832 | } |
3833 | ||
3834 | static const struct file_operations ptype_seq_fops = { | |
3835 | .owner = THIS_MODULE, | |
3836 | .open = ptype_seq_open, | |
3837 | .read = seq_read, | |
3838 | .llseek = seq_lseek, | |
2feb27db | 3839 | .release = seq_release_net, |
0e1256ff SH |
3840 | }; |
3841 | ||
3842 | ||
4665079c | 3843 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
3844 | { |
3845 | int rc = -ENOMEM; | |
3846 | ||
881d966b | 3847 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 3848 | goto out; |
881d966b | 3849 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 3850 | goto out_dev; |
881d966b | 3851 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 3852 | goto out_softnet; |
0e1256ff | 3853 | |
881d966b | 3854 | if (wext_proc_init(net)) |
457c4cbc | 3855 | goto out_ptype; |
1da177e4 LT |
3856 | rc = 0; |
3857 | out: | |
3858 | return rc; | |
457c4cbc | 3859 | out_ptype: |
881d966b | 3860 | proc_net_remove(net, "ptype"); |
1da177e4 | 3861 | out_softnet: |
881d966b | 3862 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 3863 | out_dev: |
881d966b | 3864 | proc_net_remove(net, "dev"); |
1da177e4 LT |
3865 | goto out; |
3866 | } | |
881d966b | 3867 | |
4665079c | 3868 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
3869 | { |
3870 | wext_proc_exit(net); | |
3871 | ||
3872 | proc_net_remove(net, "ptype"); | |
3873 | proc_net_remove(net, "softnet_stat"); | |
3874 | proc_net_remove(net, "dev"); | |
3875 | } | |
3876 | ||
022cbae6 | 3877 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
3878 | .init = dev_proc_net_init, |
3879 | .exit = dev_proc_net_exit, | |
3880 | }; | |
3881 | ||
3882 | static int __init dev_proc_init(void) | |
3883 | { | |
3884 | return register_pernet_subsys(&dev_proc_ops); | |
3885 | } | |
1da177e4 LT |
3886 | #else |
3887 | #define dev_proc_init() 0 | |
3888 | #endif /* CONFIG_PROC_FS */ | |
3889 | ||
3890 | ||
3891 | /** | |
3892 | * netdev_set_master - set up master/slave pair | |
3893 | * @slave: slave device | |
3894 | * @master: new master device | |
3895 | * | |
3896 | * Changes the master device of the slave. Pass %NULL to break the | |
3897 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
3898 | * a negative errno code is returned. On success the reference counts | |
3899 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
3900 | * function returns zero. | |
3901 | */ | |
3902 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
3903 | { | |
3904 | struct net_device *old = slave->master; | |
3905 | ||
3906 | ASSERT_RTNL(); | |
3907 | ||
3908 | if (master) { | |
3909 | if (old) | |
3910 | return -EBUSY; | |
3911 | dev_hold(master); | |
3912 | } | |
3913 | ||
3914 | slave->master = master; | |
4ec93edb | 3915 | |
283f2fe8 ED |
3916 | if (old) { |
3917 | synchronize_net(); | |
1da177e4 | 3918 | dev_put(old); |
283f2fe8 | 3919 | } |
1da177e4 LT |
3920 | if (master) |
3921 | slave->flags |= IFF_SLAVE; | |
3922 | else | |
3923 | slave->flags &= ~IFF_SLAVE; | |
3924 | ||
3925 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
3926 | return 0; | |
3927 | } | |
d1b19dff | 3928 | EXPORT_SYMBOL(netdev_set_master); |
1da177e4 | 3929 | |
b6c40d68 PM |
3930 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
3931 | { | |
d314774c SH |
3932 | const struct net_device_ops *ops = dev->netdev_ops; |
3933 | ||
3934 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
3935 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
3936 | } |
3937 | ||
dad9b335 | 3938 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
3939 | { |
3940 | unsigned short old_flags = dev->flags; | |
8192b0c4 DH |
3941 | uid_t uid; |
3942 | gid_t gid; | |
1da177e4 | 3943 | |
24023451 PM |
3944 | ASSERT_RTNL(); |
3945 | ||
dad9b335 WC |
3946 | dev->flags |= IFF_PROMISC; |
3947 | dev->promiscuity += inc; | |
3948 | if (dev->promiscuity == 0) { | |
3949 | /* | |
3950 | * Avoid overflow. | |
3951 | * If inc causes overflow, untouch promisc and return error. | |
3952 | */ | |
3953 | if (inc < 0) | |
3954 | dev->flags &= ~IFF_PROMISC; | |
3955 | else { | |
3956 | dev->promiscuity -= inc; | |
3957 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
3958 | "set promiscuity failed, promiscuity feature " | |
3959 | "of device might be broken.\n", dev->name); | |
3960 | return -EOVERFLOW; | |
3961 | } | |
3962 | } | |
52609c0b | 3963 | if (dev->flags != old_flags) { |
1da177e4 LT |
3964 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
3965 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 3966 | "left"); |
8192b0c4 DH |
3967 | if (audit_enabled) { |
3968 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
3969 | audit_log(current->audit_context, GFP_ATOMIC, |
3970 | AUDIT_ANOM_PROMISCUOUS, | |
3971 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
3972 | dev->name, (dev->flags & IFF_PROMISC), | |
3973 | (old_flags & IFF_PROMISC), | |
3974 | audit_get_loginuid(current), | |
8192b0c4 | 3975 | uid, gid, |
7759db82 | 3976 | audit_get_sessionid(current)); |
8192b0c4 | 3977 | } |
24023451 | 3978 | |
b6c40d68 | 3979 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 3980 | } |
dad9b335 | 3981 | return 0; |
1da177e4 LT |
3982 | } |
3983 | ||
4417da66 PM |
3984 | /** |
3985 | * dev_set_promiscuity - update promiscuity count on a device | |
3986 | * @dev: device | |
3987 | * @inc: modifier | |
3988 | * | |
3989 | * Add or remove promiscuity from a device. While the count in the device | |
3990 | * remains above zero the interface remains promiscuous. Once it hits zero | |
3991 | * the device reverts back to normal filtering operation. A negative inc | |
3992 | * value is used to drop promiscuity on the device. | |
dad9b335 | 3993 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 3994 | */ |
dad9b335 | 3995 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
3996 | { |
3997 | unsigned short old_flags = dev->flags; | |
dad9b335 | 3998 | int err; |
4417da66 | 3999 | |
dad9b335 | 4000 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 4001 | if (err < 0) |
dad9b335 | 4002 | return err; |
4417da66 PM |
4003 | if (dev->flags != old_flags) |
4004 | dev_set_rx_mode(dev); | |
dad9b335 | 4005 | return err; |
4417da66 | 4006 | } |
d1b19dff | 4007 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 4008 | |
1da177e4 LT |
4009 | /** |
4010 | * dev_set_allmulti - update allmulti count on a device | |
4011 | * @dev: device | |
4012 | * @inc: modifier | |
4013 | * | |
4014 | * Add or remove reception of all multicast frames to a device. While the | |
4015 | * count in the device remains above zero the interface remains listening | |
4016 | * to all interfaces. Once it hits zero the device reverts back to normal | |
4017 | * filtering operation. A negative @inc value is used to drop the counter | |
4018 | * when releasing a resource needing all multicasts. | |
dad9b335 | 4019 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
4020 | */ |
4021 | ||
dad9b335 | 4022 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
4023 | { |
4024 | unsigned short old_flags = dev->flags; | |
4025 | ||
24023451 PM |
4026 | ASSERT_RTNL(); |
4027 | ||
1da177e4 | 4028 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
4029 | dev->allmulti += inc; |
4030 | if (dev->allmulti == 0) { | |
4031 | /* | |
4032 | * Avoid overflow. | |
4033 | * If inc causes overflow, untouch allmulti and return error. | |
4034 | */ | |
4035 | if (inc < 0) | |
4036 | dev->flags &= ~IFF_ALLMULTI; | |
4037 | else { | |
4038 | dev->allmulti -= inc; | |
4039 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
4040 | "set allmulti failed, allmulti feature of " | |
4041 | "device might be broken.\n", dev->name); | |
4042 | return -EOVERFLOW; | |
4043 | } | |
4044 | } | |
24023451 | 4045 | if (dev->flags ^ old_flags) { |
b6c40d68 | 4046 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 4047 | dev_set_rx_mode(dev); |
24023451 | 4048 | } |
dad9b335 | 4049 | return 0; |
4417da66 | 4050 | } |
d1b19dff | 4051 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
4052 | |
4053 | /* | |
4054 | * Upload unicast and multicast address lists to device and | |
4055 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 4056 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
4057 | * are present. |
4058 | */ | |
4059 | void __dev_set_rx_mode(struct net_device *dev) | |
4060 | { | |
d314774c SH |
4061 | const struct net_device_ops *ops = dev->netdev_ops; |
4062 | ||
4417da66 PM |
4063 | /* dev_open will call this function so the list will stay sane. */ |
4064 | if (!(dev->flags&IFF_UP)) | |
4065 | return; | |
4066 | ||
4067 | if (!netif_device_present(dev)) | |
40b77c94 | 4068 | return; |
4417da66 | 4069 | |
d314774c SH |
4070 | if (ops->ndo_set_rx_mode) |
4071 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
4072 | else { |
4073 | /* Unicast addresses changes may only happen under the rtnl, | |
4074 | * therefore calling __dev_set_promiscuity here is safe. | |
4075 | */ | |
32e7bfc4 | 4076 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 PM |
4077 | __dev_set_promiscuity(dev, 1); |
4078 | dev->uc_promisc = 1; | |
32e7bfc4 | 4079 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 PM |
4080 | __dev_set_promiscuity(dev, -1); |
4081 | dev->uc_promisc = 0; | |
4082 | } | |
4083 | ||
d314774c SH |
4084 | if (ops->ndo_set_multicast_list) |
4085 | ops->ndo_set_multicast_list(dev); | |
4417da66 PM |
4086 | } |
4087 | } | |
4088 | ||
4089 | void dev_set_rx_mode(struct net_device *dev) | |
4090 | { | |
b9e40857 | 4091 | netif_addr_lock_bh(dev); |
4417da66 | 4092 | __dev_set_rx_mode(dev); |
b9e40857 | 4093 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4094 | } |
4095 | ||
f0db275a SH |
4096 | /** |
4097 | * dev_get_flags - get flags reported to userspace | |
4098 | * @dev: device | |
4099 | * | |
4100 | * Get the combination of flag bits exported through APIs to userspace. | |
4101 | */ | |
1da177e4 LT |
4102 | unsigned dev_get_flags(const struct net_device *dev) |
4103 | { | |
4104 | unsigned flags; | |
4105 | ||
4106 | flags = (dev->flags & ~(IFF_PROMISC | | |
4107 | IFF_ALLMULTI | | |
b00055aa SR |
4108 | IFF_RUNNING | |
4109 | IFF_LOWER_UP | | |
4110 | IFF_DORMANT)) | | |
1da177e4 LT |
4111 | (dev->gflags & (IFF_PROMISC | |
4112 | IFF_ALLMULTI)); | |
4113 | ||
b00055aa SR |
4114 | if (netif_running(dev)) { |
4115 | if (netif_oper_up(dev)) | |
4116 | flags |= IFF_RUNNING; | |
4117 | if (netif_carrier_ok(dev)) | |
4118 | flags |= IFF_LOWER_UP; | |
4119 | if (netif_dormant(dev)) | |
4120 | flags |= IFF_DORMANT; | |
4121 | } | |
1da177e4 LT |
4122 | |
4123 | return flags; | |
4124 | } | |
d1b19dff | 4125 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4126 | |
bd380811 | 4127 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4128 | { |
1da177e4 | 4129 | int old_flags = dev->flags; |
bd380811 | 4130 | int ret; |
1da177e4 | 4131 | |
24023451 PM |
4132 | ASSERT_RTNL(); |
4133 | ||
1da177e4 LT |
4134 | /* |
4135 | * Set the flags on our device. | |
4136 | */ | |
4137 | ||
4138 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4139 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4140 | IFF_AUTOMEDIA)) | | |
4141 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4142 | IFF_ALLMULTI)); | |
4143 | ||
4144 | /* | |
4145 | * Load in the correct multicast list now the flags have changed. | |
4146 | */ | |
4147 | ||
b6c40d68 PM |
4148 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4149 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4150 | |
4417da66 | 4151 | dev_set_rx_mode(dev); |
1da177e4 LT |
4152 | |
4153 | /* | |
4154 | * Have we downed the interface. We handle IFF_UP ourselves | |
4155 | * according to user attempts to set it, rather than blindly | |
4156 | * setting it. | |
4157 | */ | |
4158 | ||
4159 | ret = 0; | |
4160 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4161 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4162 | |
4163 | if (!ret) | |
4417da66 | 4164 | dev_set_rx_mode(dev); |
1da177e4 LT |
4165 | } |
4166 | ||
1da177e4 | 4167 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4168 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4169 | ||
1da177e4 LT |
4170 | dev->gflags ^= IFF_PROMISC; |
4171 | dev_set_promiscuity(dev, inc); | |
4172 | } | |
4173 | ||
4174 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4175 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4176 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4177 | */ | |
4178 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4179 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4180 | ||
1da177e4 LT |
4181 | dev->gflags ^= IFF_ALLMULTI; |
4182 | dev_set_allmulti(dev, inc); | |
4183 | } | |
4184 | ||
bd380811 PM |
4185 | return ret; |
4186 | } | |
4187 | ||
4188 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4189 | { | |
4190 | unsigned int changes = dev->flags ^ old_flags; | |
4191 | ||
4192 | if (changes & IFF_UP) { | |
4193 | if (dev->flags & IFF_UP) | |
4194 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4195 | else | |
4196 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4197 | } | |
4198 | ||
4199 | if (dev->flags & IFF_UP && | |
4200 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4201 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4202 | } | |
4203 | ||
4204 | /** | |
4205 | * dev_change_flags - change device settings | |
4206 | * @dev: device | |
4207 | * @flags: device state flags | |
4208 | * | |
4209 | * Change settings on device based state flags. The flags are | |
4210 | * in the userspace exported format. | |
4211 | */ | |
4212 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
4213 | { | |
4214 | int ret, changes; | |
4215 | int old_flags = dev->flags; | |
4216 | ||
4217 | ret = __dev_change_flags(dev, flags); | |
4218 | if (ret < 0) | |
4219 | return ret; | |
4220 | ||
4221 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4222 | if (changes) |
4223 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4224 | |
bd380811 | 4225 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4226 | return ret; |
4227 | } | |
d1b19dff | 4228 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4229 | |
f0db275a SH |
4230 | /** |
4231 | * dev_set_mtu - Change maximum transfer unit | |
4232 | * @dev: device | |
4233 | * @new_mtu: new transfer unit | |
4234 | * | |
4235 | * Change the maximum transfer size of the network device. | |
4236 | */ | |
1da177e4 LT |
4237 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4238 | { | |
d314774c | 4239 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4240 | int err; |
4241 | ||
4242 | if (new_mtu == dev->mtu) | |
4243 | return 0; | |
4244 | ||
4245 | /* MTU must be positive. */ | |
4246 | if (new_mtu < 0) | |
4247 | return -EINVAL; | |
4248 | ||
4249 | if (!netif_device_present(dev)) | |
4250 | return -ENODEV; | |
4251 | ||
4252 | err = 0; | |
d314774c SH |
4253 | if (ops->ndo_change_mtu) |
4254 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4255 | else |
4256 | dev->mtu = new_mtu; | |
d314774c | 4257 | |
1da177e4 | 4258 | if (!err && dev->flags & IFF_UP) |
056925ab | 4259 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4260 | return err; |
4261 | } | |
d1b19dff | 4262 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4263 | |
f0db275a SH |
4264 | /** |
4265 | * dev_set_mac_address - Change Media Access Control Address | |
4266 | * @dev: device | |
4267 | * @sa: new address | |
4268 | * | |
4269 | * Change the hardware (MAC) address of the device | |
4270 | */ | |
1da177e4 LT |
4271 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4272 | { | |
d314774c | 4273 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4274 | int err; |
4275 | ||
d314774c | 4276 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4277 | return -EOPNOTSUPP; |
4278 | if (sa->sa_family != dev->type) | |
4279 | return -EINVAL; | |
4280 | if (!netif_device_present(dev)) | |
4281 | return -ENODEV; | |
d314774c | 4282 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4283 | if (!err) |
056925ab | 4284 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
4285 | return err; |
4286 | } | |
d1b19dff | 4287 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4288 | |
4289 | /* | |
3710becf | 4290 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4291 | */ |
14e3e079 | 4292 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4293 | { |
4294 | int err; | |
3710becf | 4295 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4296 | |
4297 | if (!dev) | |
4298 | return -ENODEV; | |
4299 | ||
4300 | switch (cmd) { | |
d1b19dff ED |
4301 | case SIOCGIFFLAGS: /* Get interface flags */ |
4302 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4303 | return 0; | |
1da177e4 | 4304 | |
d1b19dff ED |
4305 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4306 | (currently unused) */ | |
4307 | ifr->ifr_metric = 0; | |
4308 | return 0; | |
1da177e4 | 4309 | |
d1b19dff ED |
4310 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4311 | ifr->ifr_mtu = dev->mtu; | |
4312 | return 0; | |
1da177e4 | 4313 | |
d1b19dff ED |
4314 | case SIOCGIFHWADDR: |
4315 | if (!dev->addr_len) | |
4316 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4317 | else | |
4318 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4319 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4320 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4321 | return 0; | |
1da177e4 | 4322 | |
d1b19dff ED |
4323 | case SIOCGIFSLAVE: |
4324 | err = -EINVAL; | |
4325 | break; | |
14e3e079 | 4326 | |
d1b19dff ED |
4327 | case SIOCGIFMAP: |
4328 | ifr->ifr_map.mem_start = dev->mem_start; | |
4329 | ifr->ifr_map.mem_end = dev->mem_end; | |
4330 | ifr->ifr_map.base_addr = dev->base_addr; | |
4331 | ifr->ifr_map.irq = dev->irq; | |
4332 | ifr->ifr_map.dma = dev->dma; | |
4333 | ifr->ifr_map.port = dev->if_port; | |
4334 | return 0; | |
14e3e079 | 4335 | |
d1b19dff ED |
4336 | case SIOCGIFINDEX: |
4337 | ifr->ifr_ifindex = dev->ifindex; | |
4338 | return 0; | |
14e3e079 | 4339 | |
d1b19dff ED |
4340 | case SIOCGIFTXQLEN: |
4341 | ifr->ifr_qlen = dev->tx_queue_len; | |
4342 | return 0; | |
14e3e079 | 4343 | |
d1b19dff ED |
4344 | default: |
4345 | /* dev_ioctl() should ensure this case | |
4346 | * is never reached | |
4347 | */ | |
4348 | WARN_ON(1); | |
4349 | err = -EINVAL; | |
4350 | break; | |
14e3e079 JG |
4351 | |
4352 | } | |
4353 | return err; | |
4354 | } | |
4355 | ||
4356 | /* | |
4357 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4358 | */ | |
4359 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4360 | { | |
4361 | int err; | |
4362 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4363 | const struct net_device_ops *ops; |
14e3e079 JG |
4364 | |
4365 | if (!dev) | |
4366 | return -ENODEV; | |
4367 | ||
5f2f6da7 JP |
4368 | ops = dev->netdev_ops; |
4369 | ||
14e3e079 | 4370 | switch (cmd) { |
d1b19dff ED |
4371 | case SIOCSIFFLAGS: /* Set interface flags */ |
4372 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 4373 | |
d1b19dff ED |
4374 | case SIOCSIFMETRIC: /* Set the metric on the interface |
4375 | (currently unused) */ | |
4376 | return -EOPNOTSUPP; | |
14e3e079 | 4377 | |
d1b19dff ED |
4378 | case SIOCSIFMTU: /* Set the MTU of a device */ |
4379 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 4380 | |
d1b19dff ED |
4381 | case SIOCSIFHWADDR: |
4382 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 4383 | |
d1b19dff ED |
4384 | case SIOCSIFHWBROADCAST: |
4385 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4386 | return -EINVAL; | |
4387 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4388 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4389 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4390 | return 0; | |
1da177e4 | 4391 | |
d1b19dff ED |
4392 | case SIOCSIFMAP: |
4393 | if (ops->ndo_set_config) { | |
1da177e4 LT |
4394 | if (!netif_device_present(dev)) |
4395 | return -ENODEV; | |
d1b19dff ED |
4396 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
4397 | } | |
4398 | return -EOPNOTSUPP; | |
1da177e4 | 4399 | |
d1b19dff ED |
4400 | case SIOCADDMULTI: |
4401 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4402 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4403 | return -EINVAL; | |
4404 | if (!netif_device_present(dev)) | |
4405 | return -ENODEV; | |
22bedad3 | 4406 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); |
d1b19dff ED |
4407 | |
4408 | case SIOCDELMULTI: | |
4409 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4410 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4411 | return -EINVAL; | |
4412 | if (!netif_device_present(dev)) | |
4413 | return -ENODEV; | |
22bedad3 | 4414 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); |
1da177e4 | 4415 | |
d1b19dff ED |
4416 | case SIOCSIFTXQLEN: |
4417 | if (ifr->ifr_qlen < 0) | |
4418 | return -EINVAL; | |
4419 | dev->tx_queue_len = ifr->ifr_qlen; | |
4420 | return 0; | |
1da177e4 | 4421 | |
d1b19dff ED |
4422 | case SIOCSIFNAME: |
4423 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4424 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 4425 | |
d1b19dff ED |
4426 | /* |
4427 | * Unknown or private ioctl | |
4428 | */ | |
4429 | default: | |
4430 | if ((cmd >= SIOCDEVPRIVATE && | |
4431 | cmd <= SIOCDEVPRIVATE + 15) || | |
4432 | cmd == SIOCBONDENSLAVE || | |
4433 | cmd == SIOCBONDRELEASE || | |
4434 | cmd == SIOCBONDSETHWADDR || | |
4435 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4436 | cmd == SIOCBONDINFOQUERY || | |
4437 | cmd == SIOCBONDCHANGEACTIVE || | |
4438 | cmd == SIOCGMIIPHY || | |
4439 | cmd == SIOCGMIIREG || | |
4440 | cmd == SIOCSMIIREG || | |
4441 | cmd == SIOCBRADDIF || | |
4442 | cmd == SIOCBRDELIF || | |
4443 | cmd == SIOCSHWTSTAMP || | |
4444 | cmd == SIOCWANDEV) { | |
4445 | err = -EOPNOTSUPP; | |
4446 | if (ops->ndo_do_ioctl) { | |
4447 | if (netif_device_present(dev)) | |
4448 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4449 | else | |
4450 | err = -ENODEV; | |
4451 | } | |
4452 | } else | |
4453 | err = -EINVAL; | |
1da177e4 LT |
4454 | |
4455 | } | |
4456 | return err; | |
4457 | } | |
4458 | ||
4459 | /* | |
4460 | * This function handles all "interface"-type I/O control requests. The actual | |
4461 | * 'doing' part of this is dev_ifsioc above. | |
4462 | */ | |
4463 | ||
4464 | /** | |
4465 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4466 | * @net: the applicable net namespace |
1da177e4 LT |
4467 | * @cmd: command to issue |
4468 | * @arg: pointer to a struct ifreq in user space | |
4469 | * | |
4470 | * Issue ioctl functions to devices. This is normally called by the | |
4471 | * user space syscall interfaces but can sometimes be useful for | |
4472 | * other purposes. The return value is the return from the syscall if | |
4473 | * positive or a negative errno code on error. | |
4474 | */ | |
4475 | ||
881d966b | 4476 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4477 | { |
4478 | struct ifreq ifr; | |
4479 | int ret; | |
4480 | char *colon; | |
4481 | ||
4482 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4483 | and requires shared lock, because it sleeps writing | |
4484 | to user space. | |
4485 | */ | |
4486 | ||
4487 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 4488 | rtnl_lock(); |
881d966b | 4489 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 4490 | rtnl_unlock(); |
1da177e4 LT |
4491 | return ret; |
4492 | } | |
4493 | if (cmd == SIOCGIFNAME) | |
881d966b | 4494 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
4495 | |
4496 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4497 | return -EFAULT; | |
4498 | ||
4499 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4500 | ||
4501 | colon = strchr(ifr.ifr_name, ':'); | |
4502 | if (colon) | |
4503 | *colon = 0; | |
4504 | ||
4505 | /* | |
4506 | * See which interface the caller is talking about. | |
4507 | */ | |
4508 | ||
4509 | switch (cmd) { | |
d1b19dff ED |
4510 | /* |
4511 | * These ioctl calls: | |
4512 | * - can be done by all. | |
4513 | * - atomic and do not require locking. | |
4514 | * - return a value | |
4515 | */ | |
4516 | case SIOCGIFFLAGS: | |
4517 | case SIOCGIFMETRIC: | |
4518 | case SIOCGIFMTU: | |
4519 | case SIOCGIFHWADDR: | |
4520 | case SIOCGIFSLAVE: | |
4521 | case SIOCGIFMAP: | |
4522 | case SIOCGIFINDEX: | |
4523 | case SIOCGIFTXQLEN: | |
4524 | dev_load(net, ifr.ifr_name); | |
3710becf | 4525 | rcu_read_lock(); |
d1b19dff | 4526 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 4527 | rcu_read_unlock(); |
d1b19dff ED |
4528 | if (!ret) { |
4529 | if (colon) | |
4530 | *colon = ':'; | |
4531 | if (copy_to_user(arg, &ifr, | |
4532 | sizeof(struct ifreq))) | |
4533 | ret = -EFAULT; | |
4534 | } | |
4535 | return ret; | |
1da177e4 | 4536 | |
d1b19dff ED |
4537 | case SIOCETHTOOL: |
4538 | dev_load(net, ifr.ifr_name); | |
4539 | rtnl_lock(); | |
4540 | ret = dev_ethtool(net, &ifr); | |
4541 | rtnl_unlock(); | |
4542 | if (!ret) { | |
4543 | if (colon) | |
4544 | *colon = ':'; | |
4545 | if (copy_to_user(arg, &ifr, | |
4546 | sizeof(struct ifreq))) | |
4547 | ret = -EFAULT; | |
4548 | } | |
4549 | return ret; | |
1da177e4 | 4550 | |
d1b19dff ED |
4551 | /* |
4552 | * These ioctl calls: | |
4553 | * - require superuser power. | |
4554 | * - require strict serialization. | |
4555 | * - return a value | |
4556 | */ | |
4557 | case SIOCGMIIPHY: | |
4558 | case SIOCGMIIREG: | |
4559 | case SIOCSIFNAME: | |
4560 | if (!capable(CAP_NET_ADMIN)) | |
4561 | return -EPERM; | |
4562 | dev_load(net, ifr.ifr_name); | |
4563 | rtnl_lock(); | |
4564 | ret = dev_ifsioc(net, &ifr, cmd); | |
4565 | rtnl_unlock(); | |
4566 | if (!ret) { | |
4567 | if (colon) | |
4568 | *colon = ':'; | |
4569 | if (copy_to_user(arg, &ifr, | |
4570 | sizeof(struct ifreq))) | |
4571 | ret = -EFAULT; | |
4572 | } | |
4573 | return ret; | |
1da177e4 | 4574 | |
d1b19dff ED |
4575 | /* |
4576 | * These ioctl calls: | |
4577 | * - require superuser power. | |
4578 | * - require strict serialization. | |
4579 | * - do not return a value | |
4580 | */ | |
4581 | case SIOCSIFFLAGS: | |
4582 | case SIOCSIFMETRIC: | |
4583 | case SIOCSIFMTU: | |
4584 | case SIOCSIFMAP: | |
4585 | case SIOCSIFHWADDR: | |
4586 | case SIOCSIFSLAVE: | |
4587 | case SIOCADDMULTI: | |
4588 | case SIOCDELMULTI: | |
4589 | case SIOCSIFHWBROADCAST: | |
4590 | case SIOCSIFTXQLEN: | |
4591 | case SIOCSMIIREG: | |
4592 | case SIOCBONDENSLAVE: | |
4593 | case SIOCBONDRELEASE: | |
4594 | case SIOCBONDSETHWADDR: | |
4595 | case SIOCBONDCHANGEACTIVE: | |
4596 | case SIOCBRADDIF: | |
4597 | case SIOCBRDELIF: | |
4598 | case SIOCSHWTSTAMP: | |
4599 | if (!capable(CAP_NET_ADMIN)) | |
4600 | return -EPERM; | |
4601 | /* fall through */ | |
4602 | case SIOCBONDSLAVEINFOQUERY: | |
4603 | case SIOCBONDINFOQUERY: | |
4604 | dev_load(net, ifr.ifr_name); | |
4605 | rtnl_lock(); | |
4606 | ret = dev_ifsioc(net, &ifr, cmd); | |
4607 | rtnl_unlock(); | |
4608 | return ret; | |
4609 | ||
4610 | case SIOCGIFMEM: | |
4611 | /* Get the per device memory space. We can add this but | |
4612 | * currently do not support it */ | |
4613 | case SIOCSIFMEM: | |
4614 | /* Set the per device memory buffer space. | |
4615 | * Not applicable in our case */ | |
4616 | case SIOCSIFLINK: | |
4617 | return -EINVAL; | |
4618 | ||
4619 | /* | |
4620 | * Unknown or private ioctl. | |
4621 | */ | |
4622 | default: | |
4623 | if (cmd == SIOCWANDEV || | |
4624 | (cmd >= SIOCDEVPRIVATE && | |
4625 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 4626 | dev_load(net, ifr.ifr_name); |
1da177e4 | 4627 | rtnl_lock(); |
881d966b | 4628 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 4629 | rtnl_unlock(); |
d1b19dff ED |
4630 | if (!ret && copy_to_user(arg, &ifr, |
4631 | sizeof(struct ifreq))) | |
4632 | ret = -EFAULT; | |
1da177e4 | 4633 | return ret; |
d1b19dff ED |
4634 | } |
4635 | /* Take care of Wireless Extensions */ | |
4636 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
4637 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
4638 | return -EINVAL; | |
1da177e4 LT |
4639 | } |
4640 | } | |
4641 | ||
4642 | ||
4643 | /** | |
4644 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 4645 | * @net: the applicable net namespace |
1da177e4 LT |
4646 | * |
4647 | * Returns a suitable unique value for a new device interface | |
4648 | * number. The caller must hold the rtnl semaphore or the | |
4649 | * dev_base_lock to be sure it remains unique. | |
4650 | */ | |
881d966b | 4651 | static int dev_new_index(struct net *net) |
1da177e4 LT |
4652 | { |
4653 | static int ifindex; | |
4654 | for (;;) { | |
4655 | if (++ifindex <= 0) | |
4656 | ifindex = 1; | |
881d966b | 4657 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
4658 | return ifindex; |
4659 | } | |
4660 | } | |
4661 | ||
1da177e4 | 4662 | /* Delayed registration/unregisteration */ |
3b5b34fd | 4663 | static LIST_HEAD(net_todo_list); |
1da177e4 | 4664 | |
6f05f629 | 4665 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 4666 | { |
1da177e4 | 4667 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
4668 | } |
4669 | ||
9b5e383c | 4670 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 4671 | { |
e93737b0 | 4672 | struct net_device *dev, *tmp; |
9b5e383c | 4673 | |
93ee31f1 DL |
4674 | BUG_ON(dev_boot_phase); |
4675 | ASSERT_RTNL(); | |
4676 | ||
e93737b0 | 4677 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 4678 | /* Some devices call without registering |
e93737b0 KK |
4679 | * for initialization unwind. Remove those |
4680 | * devices and proceed with the remaining. | |
9b5e383c ED |
4681 | */ |
4682 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
4683 | pr_debug("unregister_netdevice: device %s/%p never " | |
4684 | "was registered\n", dev->name, dev); | |
93ee31f1 | 4685 | |
9b5e383c | 4686 | WARN_ON(1); |
e93737b0 KK |
4687 | list_del(&dev->unreg_list); |
4688 | continue; | |
9b5e383c | 4689 | } |
93ee31f1 | 4690 | |
9b5e383c | 4691 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
93ee31f1 | 4692 | |
9b5e383c ED |
4693 | /* If device is running, close it first. */ |
4694 | dev_close(dev); | |
93ee31f1 | 4695 | |
9b5e383c ED |
4696 | /* And unlink it from device chain. */ |
4697 | unlist_netdevice(dev); | |
93ee31f1 | 4698 | |
9b5e383c ED |
4699 | dev->reg_state = NETREG_UNREGISTERING; |
4700 | } | |
93ee31f1 DL |
4701 | |
4702 | synchronize_net(); | |
4703 | ||
9b5e383c ED |
4704 | list_for_each_entry(dev, head, unreg_list) { |
4705 | /* Shutdown queueing discipline. */ | |
4706 | dev_shutdown(dev); | |
93ee31f1 DL |
4707 | |
4708 | ||
9b5e383c ED |
4709 | /* Notify protocols, that we are about to destroy |
4710 | this device. They should clean all the things. | |
4711 | */ | |
4712 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 4713 | |
a2835763 PM |
4714 | if (!dev->rtnl_link_ops || |
4715 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
4716 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
4717 | ||
9b5e383c ED |
4718 | /* |
4719 | * Flush the unicast and multicast chains | |
4720 | */ | |
a748ee24 | 4721 | dev_uc_flush(dev); |
22bedad3 | 4722 | dev_mc_flush(dev); |
93ee31f1 | 4723 | |
9b5e383c ED |
4724 | if (dev->netdev_ops->ndo_uninit) |
4725 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 4726 | |
9b5e383c ED |
4727 | /* Notifier chain MUST detach us from master device. */ |
4728 | WARN_ON(dev->master); | |
93ee31f1 | 4729 | |
9b5e383c ED |
4730 | /* Remove entries from kobject tree */ |
4731 | netdev_unregister_kobject(dev); | |
4732 | } | |
93ee31f1 | 4733 | |
a5ee1551 | 4734 | /* Process any work delayed until the end of the batch */ |
e5e26d75 | 4735 | dev = list_first_entry(head, struct net_device, unreg_list); |
a5ee1551 | 4736 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
93ee31f1 | 4737 | |
a5ee1551 | 4738 | synchronize_net(); |
395264d5 | 4739 | |
a5ee1551 | 4740 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
4741 | dev_put(dev); |
4742 | } | |
4743 | ||
4744 | static void rollback_registered(struct net_device *dev) | |
4745 | { | |
4746 | LIST_HEAD(single); | |
4747 | ||
4748 | list_add(&dev->unreg_list, &single); | |
4749 | rollback_registered_many(&single); | |
93ee31f1 DL |
4750 | } |
4751 | ||
e8a0464c DM |
4752 | static void __netdev_init_queue_locks_one(struct net_device *dev, |
4753 | struct netdev_queue *dev_queue, | |
4754 | void *_unused) | |
c773e847 DM |
4755 | { |
4756 | spin_lock_init(&dev_queue->_xmit_lock); | |
cf508b12 | 4757 | netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type); |
c773e847 DM |
4758 | dev_queue->xmit_lock_owner = -1; |
4759 | } | |
4760 | ||
4761 | static void netdev_init_queue_locks(struct net_device *dev) | |
4762 | { | |
e8a0464c DM |
4763 | netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL); |
4764 | __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL); | |
c773e847 DM |
4765 | } |
4766 | ||
b63365a2 HX |
4767 | unsigned long netdev_fix_features(unsigned long features, const char *name) |
4768 | { | |
4769 | /* Fix illegal SG+CSUM combinations. */ | |
4770 | if ((features & NETIF_F_SG) && | |
4771 | !(features & NETIF_F_ALL_CSUM)) { | |
4772 | if (name) | |
4773 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " | |
4774 | "checksum feature.\n", name); | |
4775 | features &= ~NETIF_F_SG; | |
4776 | } | |
4777 | ||
4778 | /* TSO requires that SG is present as well. */ | |
4779 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { | |
4780 | if (name) | |
4781 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " | |
4782 | "SG feature.\n", name); | |
4783 | features &= ~NETIF_F_TSO; | |
4784 | } | |
4785 | ||
4786 | if (features & NETIF_F_UFO) { | |
4787 | if (!(features & NETIF_F_GEN_CSUM)) { | |
4788 | if (name) | |
4789 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4790 | "since no NETIF_F_HW_CSUM feature.\n", | |
4791 | name); | |
4792 | features &= ~NETIF_F_UFO; | |
4793 | } | |
4794 | ||
4795 | if (!(features & NETIF_F_SG)) { | |
4796 | if (name) | |
4797 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4798 | "since no NETIF_F_SG feature.\n", name); | |
4799 | features &= ~NETIF_F_UFO; | |
4800 | } | |
4801 | } | |
4802 | ||
4803 | return features; | |
4804 | } | |
4805 | EXPORT_SYMBOL(netdev_fix_features); | |
4806 | ||
fc4a7489 PM |
4807 | /** |
4808 | * netif_stacked_transfer_operstate - transfer operstate | |
4809 | * @rootdev: the root or lower level device to transfer state from | |
4810 | * @dev: the device to transfer operstate to | |
4811 | * | |
4812 | * Transfer operational state from root to device. This is normally | |
4813 | * called when a stacking relationship exists between the root | |
4814 | * device and the device(a leaf device). | |
4815 | */ | |
4816 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
4817 | struct net_device *dev) | |
4818 | { | |
4819 | if (rootdev->operstate == IF_OPER_DORMANT) | |
4820 | netif_dormant_on(dev); | |
4821 | else | |
4822 | netif_dormant_off(dev); | |
4823 | ||
4824 | if (netif_carrier_ok(rootdev)) { | |
4825 | if (!netif_carrier_ok(dev)) | |
4826 | netif_carrier_on(dev); | |
4827 | } else { | |
4828 | if (netif_carrier_ok(dev)) | |
4829 | netif_carrier_off(dev); | |
4830 | } | |
4831 | } | |
4832 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
4833 | ||
1da177e4 LT |
4834 | /** |
4835 | * register_netdevice - register a network device | |
4836 | * @dev: device to register | |
4837 | * | |
4838 | * Take a completed network device structure and add it to the kernel | |
4839 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
4840 | * chain. 0 is returned on success. A negative errno code is returned | |
4841 | * on a failure to set up the device, or if the name is a duplicate. | |
4842 | * | |
4843 | * Callers must hold the rtnl semaphore. You may want | |
4844 | * register_netdev() instead of this. | |
4845 | * | |
4846 | * BUGS: | |
4847 | * The locking appears insufficient to guarantee two parallel registers | |
4848 | * will not get the same name. | |
4849 | */ | |
4850 | ||
4851 | int register_netdevice(struct net_device *dev) | |
4852 | { | |
1da177e4 | 4853 | int ret; |
d314774c | 4854 | struct net *net = dev_net(dev); |
1da177e4 LT |
4855 | |
4856 | BUG_ON(dev_boot_phase); | |
4857 | ASSERT_RTNL(); | |
4858 | ||
b17a7c17 SH |
4859 | might_sleep(); |
4860 | ||
1da177e4 LT |
4861 | /* When net_device's are persistent, this will be fatal. */ |
4862 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 4863 | BUG_ON(!net); |
1da177e4 | 4864 | |
f1f28aa3 | 4865 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 4866 | netdev_set_addr_lockdep_class(dev); |
c773e847 | 4867 | netdev_init_queue_locks(dev); |
1da177e4 | 4868 | |
1da177e4 LT |
4869 | dev->iflink = -1; |
4870 | ||
df334545 | 4871 | #ifdef CONFIG_RPS |
0a9627f2 TH |
4872 | if (!dev->num_rx_queues) { |
4873 | /* | |
4874 | * Allocate a single RX queue if driver never called | |
4875 | * alloc_netdev_mq | |
4876 | */ | |
4877 | ||
4878 | dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL); | |
4879 | if (!dev->_rx) { | |
4880 | ret = -ENOMEM; | |
4881 | goto out; | |
4882 | } | |
4883 | ||
4884 | dev->_rx->first = dev->_rx; | |
4885 | atomic_set(&dev->_rx->count, 1); | |
4886 | dev->num_rx_queues = 1; | |
4887 | } | |
df334545 | 4888 | #endif |
1da177e4 | 4889 | /* Init, if this function is available */ |
d314774c SH |
4890 | if (dev->netdev_ops->ndo_init) { |
4891 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
4892 | if (ret) { |
4893 | if (ret > 0) | |
4894 | ret = -EIO; | |
90833aa4 | 4895 | goto out; |
1da177e4 LT |
4896 | } |
4897 | } | |
4ec93edb | 4898 | |
d9031024 OP |
4899 | ret = dev_get_valid_name(net, dev->name, dev->name, 0); |
4900 | if (ret) | |
7ce1b0ed | 4901 | goto err_uninit; |
1da177e4 | 4902 | |
881d966b | 4903 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
4904 | if (dev->iflink == -1) |
4905 | dev->iflink = dev->ifindex; | |
4906 | ||
d212f87b SH |
4907 | /* Fix illegal checksum combinations */ |
4908 | if ((dev->features & NETIF_F_HW_CSUM) && | |
4909 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
4910 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
4911 | dev->name); | |
4912 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
4913 | } | |
4914 | ||
4915 | if ((dev->features & NETIF_F_NO_CSUM) && | |
4916 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
4917 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
4918 | dev->name); | |
4919 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
4920 | } | |
4921 | ||
b63365a2 | 4922 | dev->features = netdev_fix_features(dev->features, dev->name); |
1da177e4 | 4923 | |
e5a4a72d LB |
4924 | /* Enable software GSO if SG is supported. */ |
4925 | if (dev->features & NETIF_F_SG) | |
4926 | dev->features |= NETIF_F_GSO; | |
4927 | ||
aaf8cdc3 | 4928 | netdev_initialize_kobject(dev); |
7ffbe3fd JB |
4929 | |
4930 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); | |
4931 | ret = notifier_to_errno(ret); | |
4932 | if (ret) | |
4933 | goto err_uninit; | |
4934 | ||
8b41d188 | 4935 | ret = netdev_register_kobject(dev); |
b17a7c17 | 4936 | if (ret) |
7ce1b0ed | 4937 | goto err_uninit; |
b17a7c17 SH |
4938 | dev->reg_state = NETREG_REGISTERED; |
4939 | ||
1da177e4 LT |
4940 | /* |
4941 | * Default initial state at registry is that the | |
4942 | * device is present. | |
4943 | */ | |
4944 | ||
4945 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
4946 | ||
1da177e4 | 4947 | dev_init_scheduler(dev); |
1da177e4 | 4948 | dev_hold(dev); |
ce286d32 | 4949 | list_netdevice(dev); |
1da177e4 LT |
4950 | |
4951 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 4952 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 4953 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
4954 | if (ret) { |
4955 | rollback_registered(dev); | |
4956 | dev->reg_state = NETREG_UNREGISTERED; | |
4957 | } | |
d90a909e EB |
4958 | /* |
4959 | * Prevent userspace races by waiting until the network | |
4960 | * device is fully setup before sending notifications. | |
4961 | */ | |
a2835763 PM |
4962 | if (!dev->rtnl_link_ops || |
4963 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
4964 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
4965 | |
4966 | out: | |
4967 | return ret; | |
7ce1b0ed HX |
4968 | |
4969 | err_uninit: | |
d314774c SH |
4970 | if (dev->netdev_ops->ndo_uninit) |
4971 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 4972 | goto out; |
1da177e4 | 4973 | } |
d1b19dff | 4974 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 4975 | |
937f1ba5 BH |
4976 | /** |
4977 | * init_dummy_netdev - init a dummy network device for NAPI | |
4978 | * @dev: device to init | |
4979 | * | |
4980 | * This takes a network device structure and initialize the minimum | |
4981 | * amount of fields so it can be used to schedule NAPI polls without | |
4982 | * registering a full blown interface. This is to be used by drivers | |
4983 | * that need to tie several hardware interfaces to a single NAPI | |
4984 | * poll scheduler due to HW limitations. | |
4985 | */ | |
4986 | int init_dummy_netdev(struct net_device *dev) | |
4987 | { | |
4988 | /* Clear everything. Note we don't initialize spinlocks | |
4989 | * are they aren't supposed to be taken by any of the | |
4990 | * NAPI code and this dummy netdev is supposed to be | |
4991 | * only ever used for NAPI polls | |
4992 | */ | |
4993 | memset(dev, 0, sizeof(struct net_device)); | |
4994 | ||
4995 | /* make sure we BUG if trying to hit standard | |
4996 | * register/unregister code path | |
4997 | */ | |
4998 | dev->reg_state = NETREG_DUMMY; | |
4999 | ||
5000 | /* initialize the ref count */ | |
5001 | atomic_set(&dev->refcnt, 1); | |
5002 | ||
5003 | /* NAPI wants this */ | |
5004 | INIT_LIST_HEAD(&dev->napi_list); | |
5005 | ||
5006 | /* a dummy interface is started by default */ | |
5007 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
5008 | set_bit(__LINK_STATE_START, &dev->state); | |
5009 | ||
5010 | return 0; | |
5011 | } | |
5012 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
5013 | ||
5014 | ||
1da177e4 LT |
5015 | /** |
5016 | * register_netdev - register a network device | |
5017 | * @dev: device to register | |
5018 | * | |
5019 | * Take a completed network device structure and add it to the kernel | |
5020 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
5021 | * chain. 0 is returned on success. A negative errno code is returned | |
5022 | * on a failure to set up the device, or if the name is a duplicate. | |
5023 | * | |
38b4da38 | 5024 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
5025 | * and expands the device name if you passed a format string to |
5026 | * alloc_netdev. | |
5027 | */ | |
5028 | int register_netdev(struct net_device *dev) | |
5029 | { | |
5030 | int err; | |
5031 | ||
5032 | rtnl_lock(); | |
5033 | ||
5034 | /* | |
5035 | * If the name is a format string the caller wants us to do a | |
5036 | * name allocation. | |
5037 | */ | |
5038 | if (strchr(dev->name, '%')) { | |
5039 | err = dev_alloc_name(dev, dev->name); | |
5040 | if (err < 0) | |
5041 | goto out; | |
5042 | } | |
4ec93edb | 5043 | |
1da177e4 LT |
5044 | err = register_netdevice(dev); |
5045 | out: | |
5046 | rtnl_unlock(); | |
5047 | return err; | |
5048 | } | |
5049 | EXPORT_SYMBOL(register_netdev); | |
5050 | ||
5051 | /* | |
5052 | * netdev_wait_allrefs - wait until all references are gone. | |
5053 | * | |
5054 | * This is called when unregistering network devices. | |
5055 | * | |
5056 | * Any protocol or device that holds a reference should register | |
5057 | * for netdevice notification, and cleanup and put back the | |
5058 | * reference if they receive an UNREGISTER event. | |
5059 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 5060 | * call dev_put. |
1da177e4 LT |
5061 | */ |
5062 | static void netdev_wait_allrefs(struct net_device *dev) | |
5063 | { | |
5064 | unsigned long rebroadcast_time, warning_time; | |
5065 | ||
e014debe ED |
5066 | linkwatch_forget_dev(dev); |
5067 | ||
1da177e4 LT |
5068 | rebroadcast_time = warning_time = jiffies; |
5069 | while (atomic_read(&dev->refcnt) != 0) { | |
5070 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
6756ae4b | 5071 | rtnl_lock(); |
1da177e4 LT |
5072 | |
5073 | /* Rebroadcast unregister notification */ | |
056925ab | 5074 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
a5ee1551 | 5075 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users |
395264d5 | 5076 | * should have already handle it the first time */ |
1da177e4 LT |
5077 | |
5078 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5079 | &dev->state)) { | |
5080 | /* We must not have linkwatch events | |
5081 | * pending on unregister. If this | |
5082 | * happens, we simply run the queue | |
5083 | * unscheduled, resulting in a noop | |
5084 | * for this device. | |
5085 | */ | |
5086 | linkwatch_run_queue(); | |
5087 | } | |
5088 | ||
6756ae4b | 5089 | __rtnl_unlock(); |
1da177e4 LT |
5090 | |
5091 | rebroadcast_time = jiffies; | |
5092 | } | |
5093 | ||
5094 | msleep(250); | |
5095 | ||
5096 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
5097 | printk(KERN_EMERG "unregister_netdevice: " | |
5098 | "waiting for %s to become free. Usage " | |
5099 | "count = %d\n", | |
5100 | dev->name, atomic_read(&dev->refcnt)); | |
5101 | warning_time = jiffies; | |
5102 | } | |
5103 | } | |
5104 | } | |
5105 | ||
5106 | /* The sequence is: | |
5107 | * | |
5108 | * rtnl_lock(); | |
5109 | * ... | |
5110 | * register_netdevice(x1); | |
5111 | * register_netdevice(x2); | |
5112 | * ... | |
5113 | * unregister_netdevice(y1); | |
5114 | * unregister_netdevice(y2); | |
5115 | * ... | |
5116 | * rtnl_unlock(); | |
5117 | * free_netdev(y1); | |
5118 | * free_netdev(y2); | |
5119 | * | |
58ec3b4d | 5120 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5121 | * This allows us to deal with problems: |
b17a7c17 | 5122 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5123 | * without deadlocking with linkwatch via keventd. |
5124 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5125 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5126 | * |
5127 | * We must not return until all unregister events added during | |
5128 | * the interval the lock was held have been completed. | |
1da177e4 | 5129 | */ |
1da177e4 LT |
5130 | void netdev_run_todo(void) |
5131 | { | |
626ab0e6 | 5132 | struct list_head list; |
1da177e4 | 5133 | |
1da177e4 | 5134 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5135 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5136 | |
5137 | __rtnl_unlock(); | |
626ab0e6 | 5138 | |
1da177e4 LT |
5139 | while (!list_empty(&list)) { |
5140 | struct net_device *dev | |
e5e26d75 | 5141 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5142 | list_del(&dev->todo_list); |
5143 | ||
b17a7c17 SH |
5144 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
5145 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5146 | dev->name, dev->reg_state); | |
5147 | dump_stack(); | |
5148 | continue; | |
5149 | } | |
1da177e4 | 5150 | |
b17a7c17 | 5151 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5152 | |
152102c7 | 5153 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5154 | |
b17a7c17 | 5155 | netdev_wait_allrefs(dev); |
1da177e4 | 5156 | |
b17a7c17 SH |
5157 | /* paranoia */ |
5158 | BUG_ON(atomic_read(&dev->refcnt)); | |
547b792c IJ |
5159 | WARN_ON(dev->ip_ptr); |
5160 | WARN_ON(dev->ip6_ptr); | |
5161 | WARN_ON(dev->dn_ptr); | |
1da177e4 | 5162 | |
b17a7c17 SH |
5163 | if (dev->destructor) |
5164 | dev->destructor(dev); | |
9093bbb2 SH |
5165 | |
5166 | /* Free network device */ | |
5167 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5168 | } |
1da177e4 LT |
5169 | } |
5170 | ||
d83345ad ED |
5171 | /** |
5172 | * dev_txq_stats_fold - fold tx_queues stats | |
5173 | * @dev: device to get statistics from | |
5174 | * @stats: struct net_device_stats to hold results | |
5175 | */ | |
5176 | void dev_txq_stats_fold(const struct net_device *dev, | |
5177 | struct net_device_stats *stats) | |
5178 | { | |
5179 | unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0; | |
5180 | unsigned int i; | |
5181 | struct netdev_queue *txq; | |
5182 | ||
5183 | for (i = 0; i < dev->num_tx_queues; i++) { | |
5184 | txq = netdev_get_tx_queue(dev, i); | |
5185 | tx_bytes += txq->tx_bytes; | |
5186 | tx_packets += txq->tx_packets; | |
5187 | tx_dropped += txq->tx_dropped; | |
5188 | } | |
5189 | if (tx_bytes || tx_packets || tx_dropped) { | |
5190 | stats->tx_bytes = tx_bytes; | |
5191 | stats->tx_packets = tx_packets; | |
5192 | stats->tx_dropped = tx_dropped; | |
5193 | } | |
5194 | } | |
5195 | EXPORT_SYMBOL(dev_txq_stats_fold); | |
5196 | ||
eeda3fd6 SH |
5197 | /** |
5198 | * dev_get_stats - get network device statistics | |
5199 | * @dev: device to get statistics from | |
5200 | * | |
5201 | * Get network statistics from device. The device driver may provide | |
5202 | * its own method by setting dev->netdev_ops->get_stats; otherwise | |
5203 | * the internal statistics structure is used. | |
5204 | */ | |
5205 | const struct net_device_stats *dev_get_stats(struct net_device *dev) | |
7004bf25 | 5206 | { |
eeda3fd6 SH |
5207 | const struct net_device_ops *ops = dev->netdev_ops; |
5208 | ||
5209 | if (ops->ndo_get_stats) | |
5210 | return ops->ndo_get_stats(dev); | |
d83345ad ED |
5211 | |
5212 | dev_txq_stats_fold(dev, &dev->stats); | |
5213 | return &dev->stats; | |
c45d286e | 5214 | } |
eeda3fd6 | 5215 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5216 | |
dc2b4847 | 5217 | static void netdev_init_one_queue(struct net_device *dev, |
e8a0464c DM |
5218 | struct netdev_queue *queue, |
5219 | void *_unused) | |
dc2b4847 | 5220 | { |
dc2b4847 DM |
5221 | queue->dev = dev; |
5222 | } | |
5223 | ||
bb949fbd DM |
5224 | static void netdev_init_queues(struct net_device *dev) |
5225 | { | |
e8a0464c DM |
5226 | netdev_init_one_queue(dev, &dev->rx_queue, NULL); |
5227 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); | |
c3f26a26 | 5228 | spin_lock_init(&dev->tx_global_lock); |
bb949fbd DM |
5229 | } |
5230 | ||
1da177e4 | 5231 | /** |
f25f4e44 | 5232 | * alloc_netdev_mq - allocate network device |
1da177e4 LT |
5233 | * @sizeof_priv: size of private data to allocate space for |
5234 | * @name: device name format string | |
5235 | * @setup: callback to initialize device | |
f25f4e44 | 5236 | * @queue_count: the number of subqueues to allocate |
1da177e4 LT |
5237 | * |
5238 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 PWJ |
5239 | * and performs basic initialization. Also allocates subquue structs |
5240 | * for each queue on the device at the end of the netdevice. | |
1da177e4 | 5241 | */ |
f25f4e44 PWJ |
5242 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, |
5243 | void (*setup)(struct net_device *), unsigned int queue_count) | |
1da177e4 | 5244 | { |
e8a0464c | 5245 | struct netdev_queue *tx; |
1da177e4 | 5246 | struct net_device *dev; |
7943986c | 5247 | size_t alloc_size; |
1ce8e7b5 | 5248 | struct net_device *p; |
df334545 ED |
5249 | #ifdef CONFIG_RPS |
5250 | struct netdev_rx_queue *rx; | |
0a9627f2 | 5251 | int i; |
df334545 | 5252 | #endif |
1da177e4 | 5253 | |
b6fe17d6 SH |
5254 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5255 | ||
fd2ea0a7 | 5256 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5257 | if (sizeof_priv) { |
5258 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5259 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5260 | alloc_size += sizeof_priv; |
5261 | } | |
5262 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5263 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5264 | |
31380de9 | 5265 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 5266 | if (!p) { |
b6fe17d6 | 5267 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
5268 | return NULL; |
5269 | } | |
1da177e4 | 5270 | |
7943986c | 5271 | tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL); |
e8a0464c DM |
5272 | if (!tx) { |
5273 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5274 | "tx qdiscs.\n"); | |
ab9c73cc | 5275 | goto free_p; |
e8a0464c DM |
5276 | } |
5277 | ||
df334545 | 5278 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5279 | rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
5280 | if (!rx) { | |
5281 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5282 | "rx queues.\n"); | |
5283 | goto free_tx; | |
5284 | } | |
5285 | ||
5286 | atomic_set(&rx->count, queue_count); | |
5287 | ||
5288 | /* | |
5289 | * Set a pointer to first element in the array which holds the | |
5290 | * reference count. | |
5291 | */ | |
5292 | for (i = 0; i < queue_count; i++) | |
5293 | rx[i].first = rx; | |
df334545 | 5294 | #endif |
0a9627f2 | 5295 | |
1ce8e7b5 | 5296 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5297 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc JP |
5298 | |
5299 | if (dev_addr_init(dev)) | |
0a9627f2 | 5300 | goto free_rx; |
ab9c73cc | 5301 | |
22bedad3 | 5302 | dev_mc_init(dev); |
a748ee24 | 5303 | dev_uc_init(dev); |
ccffad25 | 5304 | |
c346dca1 | 5305 | dev_net_set(dev, &init_net); |
1da177e4 | 5306 | |
e8a0464c DM |
5307 | dev->_tx = tx; |
5308 | dev->num_tx_queues = queue_count; | |
fd2ea0a7 | 5309 | dev->real_num_tx_queues = queue_count; |
e8a0464c | 5310 | |
df334545 | 5311 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5312 | dev->_rx = rx; |
5313 | dev->num_rx_queues = queue_count; | |
df334545 | 5314 | #endif |
0a9627f2 | 5315 | |
82cc1a7a | 5316 | dev->gso_max_size = GSO_MAX_SIZE; |
1da177e4 | 5317 | |
bb949fbd DM |
5318 | netdev_init_queues(dev); |
5319 | ||
15682bc4 PWJ |
5320 | INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list); |
5321 | dev->ethtool_ntuple_list.count = 0; | |
d565b0a1 | 5322 | INIT_LIST_HEAD(&dev->napi_list); |
9fdce099 | 5323 | INIT_LIST_HEAD(&dev->unreg_list); |
e014debe | 5324 | INIT_LIST_HEAD(&dev->link_watch_list); |
93f154b5 | 5325 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
1da177e4 LT |
5326 | setup(dev); |
5327 | strcpy(dev->name, name); | |
5328 | return dev; | |
ab9c73cc | 5329 | |
0a9627f2 | 5330 | free_rx: |
df334545 | 5331 | #ifdef CONFIG_RPS |
0a9627f2 | 5332 | kfree(rx); |
ab9c73cc | 5333 | free_tx: |
df334545 | 5334 | #endif |
ab9c73cc | 5335 | kfree(tx); |
ab9c73cc JP |
5336 | free_p: |
5337 | kfree(p); | |
5338 | return NULL; | |
1da177e4 | 5339 | } |
f25f4e44 | 5340 | EXPORT_SYMBOL(alloc_netdev_mq); |
1da177e4 LT |
5341 | |
5342 | /** | |
5343 | * free_netdev - free network device | |
5344 | * @dev: device | |
5345 | * | |
4ec93edb YH |
5346 | * This function does the last stage of destroying an allocated device |
5347 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5348 | * If this is the last reference then it will be freed. |
5349 | */ | |
5350 | void free_netdev(struct net_device *dev) | |
5351 | { | |
d565b0a1 HX |
5352 | struct napi_struct *p, *n; |
5353 | ||
f3005d7f DL |
5354 | release_net(dev_net(dev)); |
5355 | ||
e8a0464c DM |
5356 | kfree(dev->_tx); |
5357 | ||
f001fde5 JP |
5358 | /* Flush device addresses */ |
5359 | dev_addr_flush(dev); | |
5360 | ||
15682bc4 PWJ |
5361 | /* Clear ethtool n-tuple list */ |
5362 | ethtool_ntuple_flush(dev); | |
5363 | ||
d565b0a1 HX |
5364 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5365 | netif_napi_del(p); | |
5366 | ||
3041a069 | 5367 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5368 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5369 | kfree((char *)dev - dev->padded); | |
5370 | return; | |
5371 | } | |
5372 | ||
5373 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5374 | dev->reg_state = NETREG_RELEASED; | |
5375 | ||
43cb76d9 GKH |
5376 | /* will free via device release */ |
5377 | put_device(&dev->dev); | |
1da177e4 | 5378 | } |
d1b19dff | 5379 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 5380 | |
f0db275a SH |
5381 | /** |
5382 | * synchronize_net - Synchronize with packet receive processing | |
5383 | * | |
5384 | * Wait for packets currently being received to be done. | |
5385 | * Does not block later packets from starting. | |
5386 | */ | |
4ec93edb | 5387 | void synchronize_net(void) |
1da177e4 LT |
5388 | { |
5389 | might_sleep(); | |
fbd568a3 | 5390 | synchronize_rcu(); |
1da177e4 | 5391 | } |
d1b19dff | 5392 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
5393 | |
5394 | /** | |
44a0873d | 5395 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 5396 | * @dev: device |
44a0873d | 5397 | * @head: list |
6ebfbc06 | 5398 | * |
1da177e4 | 5399 | * This function shuts down a device interface and removes it |
d59b54b1 | 5400 | * from the kernel tables. |
44a0873d | 5401 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
5402 | * |
5403 | * Callers must hold the rtnl semaphore. You may want | |
5404 | * unregister_netdev() instead of this. | |
5405 | */ | |
5406 | ||
44a0873d | 5407 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 5408 | { |
a6620712 HX |
5409 | ASSERT_RTNL(); |
5410 | ||
44a0873d | 5411 | if (head) { |
9fdce099 | 5412 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
5413 | } else { |
5414 | rollback_registered(dev); | |
5415 | /* Finish processing unregister after unlock */ | |
5416 | net_set_todo(dev); | |
5417 | } | |
1da177e4 | 5418 | } |
44a0873d | 5419 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 5420 | |
9b5e383c ED |
5421 | /** |
5422 | * unregister_netdevice_many - unregister many devices | |
5423 | * @head: list of devices | |
9b5e383c ED |
5424 | */ |
5425 | void unregister_netdevice_many(struct list_head *head) | |
5426 | { | |
5427 | struct net_device *dev; | |
5428 | ||
5429 | if (!list_empty(head)) { | |
5430 | rollback_registered_many(head); | |
5431 | list_for_each_entry(dev, head, unreg_list) | |
5432 | net_set_todo(dev); | |
5433 | } | |
5434 | } | |
63c8099d | 5435 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 5436 | |
1da177e4 LT |
5437 | /** |
5438 | * unregister_netdev - remove device from the kernel | |
5439 | * @dev: device | |
5440 | * | |
5441 | * This function shuts down a device interface and removes it | |
d59b54b1 | 5442 | * from the kernel tables. |
1da177e4 LT |
5443 | * |
5444 | * This is just a wrapper for unregister_netdevice that takes | |
5445 | * the rtnl semaphore. In general you want to use this and not | |
5446 | * unregister_netdevice. | |
5447 | */ | |
5448 | void unregister_netdev(struct net_device *dev) | |
5449 | { | |
5450 | rtnl_lock(); | |
5451 | unregister_netdevice(dev); | |
5452 | rtnl_unlock(); | |
5453 | } | |
1da177e4 LT |
5454 | EXPORT_SYMBOL(unregister_netdev); |
5455 | ||
ce286d32 EB |
5456 | /** |
5457 | * dev_change_net_namespace - move device to different nethost namespace | |
5458 | * @dev: device | |
5459 | * @net: network namespace | |
5460 | * @pat: If not NULL name pattern to try if the current device name | |
5461 | * is already taken in the destination network namespace. | |
5462 | * | |
5463 | * This function shuts down a device interface and moves it | |
5464 | * to a new network namespace. On success 0 is returned, on | |
5465 | * a failure a netagive errno code is returned. | |
5466 | * | |
5467 | * Callers must hold the rtnl semaphore. | |
5468 | */ | |
5469 | ||
5470 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5471 | { | |
ce286d32 EB |
5472 | int err; |
5473 | ||
5474 | ASSERT_RTNL(); | |
5475 | ||
5476 | /* Don't allow namespace local devices to be moved. */ | |
5477 | err = -EINVAL; | |
5478 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5479 | goto out; | |
5480 | ||
3891845e EB |
5481 | #ifdef CONFIG_SYSFS |
5482 | /* Don't allow real devices to be moved when sysfs | |
5483 | * is enabled. | |
5484 | */ | |
5485 | err = -EINVAL; | |
5486 | if (dev->dev.parent) | |
5487 | goto out; | |
5488 | #endif | |
5489 | ||
ce286d32 EB |
5490 | /* Ensure the device has been registrered */ |
5491 | err = -EINVAL; | |
5492 | if (dev->reg_state != NETREG_REGISTERED) | |
5493 | goto out; | |
5494 | ||
5495 | /* Get out if there is nothing todo */ | |
5496 | err = 0; | |
878628fb | 5497 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
5498 | goto out; |
5499 | ||
5500 | /* Pick the destination device name, and ensure | |
5501 | * we can use it in the destination network namespace. | |
5502 | */ | |
5503 | err = -EEXIST; | |
d9031024 | 5504 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
5505 | /* We get here if we can't use the current device name */ |
5506 | if (!pat) | |
5507 | goto out; | |
d9031024 | 5508 | if (dev_get_valid_name(net, pat, dev->name, 1)) |
ce286d32 EB |
5509 | goto out; |
5510 | } | |
5511 | ||
5512 | /* | |
5513 | * And now a mini version of register_netdevice unregister_netdevice. | |
5514 | */ | |
5515 | ||
5516 | /* If device is running close it first. */ | |
9b772652 | 5517 | dev_close(dev); |
ce286d32 EB |
5518 | |
5519 | /* And unlink it from device chain */ | |
5520 | err = -ENODEV; | |
5521 | unlist_netdevice(dev); | |
5522 | ||
5523 | synchronize_net(); | |
5524 | ||
5525 | /* Shutdown queueing discipline. */ | |
5526 | dev_shutdown(dev); | |
5527 | ||
5528 | /* Notify protocols, that we are about to destroy | |
5529 | this device. They should clean all the things. | |
5530 | */ | |
5531 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
a5ee1551 | 5532 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
ce286d32 EB |
5533 | |
5534 | /* | |
5535 | * Flush the unicast and multicast chains | |
5536 | */ | |
a748ee24 | 5537 | dev_uc_flush(dev); |
22bedad3 | 5538 | dev_mc_flush(dev); |
ce286d32 | 5539 | |
3891845e EB |
5540 | netdev_unregister_kobject(dev); |
5541 | ||
ce286d32 | 5542 | /* Actually switch the network namespace */ |
c346dca1 | 5543 | dev_net_set(dev, net); |
ce286d32 | 5544 | |
ce286d32 EB |
5545 | /* If there is an ifindex conflict assign a new one */ |
5546 | if (__dev_get_by_index(net, dev->ifindex)) { | |
5547 | int iflink = (dev->iflink == dev->ifindex); | |
5548 | dev->ifindex = dev_new_index(net); | |
5549 | if (iflink) | |
5550 | dev->iflink = dev->ifindex; | |
5551 | } | |
5552 | ||
8b41d188 | 5553 | /* Fixup kobjects */ |
aaf8cdc3 | 5554 | err = netdev_register_kobject(dev); |
8b41d188 | 5555 | WARN_ON(err); |
ce286d32 EB |
5556 | |
5557 | /* Add the device back in the hashes */ | |
5558 | list_netdevice(dev); | |
5559 | ||
5560 | /* Notify protocols, that a new device appeared. */ | |
5561 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5562 | ||
d90a909e EB |
5563 | /* |
5564 | * Prevent userspace races by waiting until the network | |
5565 | * device is fully setup before sending notifications. | |
5566 | */ | |
5567 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5568 | ||
ce286d32 EB |
5569 | synchronize_net(); |
5570 | err = 0; | |
5571 | out: | |
5572 | return err; | |
5573 | } | |
463d0183 | 5574 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 5575 | |
1da177e4 LT |
5576 | static int dev_cpu_callback(struct notifier_block *nfb, |
5577 | unsigned long action, | |
5578 | void *ocpu) | |
5579 | { | |
5580 | struct sk_buff **list_skb; | |
37437bb2 | 5581 | struct Qdisc **list_net; |
1da177e4 LT |
5582 | struct sk_buff *skb; |
5583 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
5584 | struct softnet_data *sd, *oldsd; | |
5585 | ||
8bb78442 | 5586 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
5587 | return NOTIFY_OK; |
5588 | ||
5589 | local_irq_disable(); | |
5590 | cpu = smp_processor_id(); | |
5591 | sd = &per_cpu(softnet_data, cpu); | |
5592 | oldsd = &per_cpu(softnet_data, oldcpu); | |
5593 | ||
5594 | /* Find end of our completion_queue. */ | |
5595 | list_skb = &sd->completion_queue; | |
5596 | while (*list_skb) | |
5597 | list_skb = &(*list_skb)->next; | |
5598 | /* Append completion queue from offline CPU. */ | |
5599 | *list_skb = oldsd->completion_queue; | |
5600 | oldsd->completion_queue = NULL; | |
5601 | ||
5602 | /* Find end of our output_queue. */ | |
5603 | list_net = &sd->output_queue; | |
5604 | while (*list_net) | |
5605 | list_net = &(*list_net)->next_sched; | |
5606 | /* Append output queue from offline CPU. */ | |
5607 | *list_net = oldsd->output_queue; | |
5608 | oldsd->output_queue = NULL; | |
5609 | ||
5610 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
5611 | local_irq_enable(); | |
5612 | ||
5613 | /* Process offline CPU's input_pkt_queue */ | |
fec5e652 | 5614 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) { |
1da177e4 | 5615 | netif_rx(skb); |
fec5e652 TH |
5616 | incr_input_queue_head(oldsd); |
5617 | } | |
1da177e4 LT |
5618 | |
5619 | return NOTIFY_OK; | |
5620 | } | |
1da177e4 LT |
5621 | |
5622 | ||
7f353bf2 | 5623 | /** |
b63365a2 HX |
5624 | * netdev_increment_features - increment feature set by one |
5625 | * @all: current feature set | |
5626 | * @one: new feature set | |
5627 | * @mask: mask feature set | |
7f353bf2 HX |
5628 | * |
5629 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
5630 | * @one to the master device with current feature set @all. Will not |
5631 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 5632 | */ |
b63365a2 HX |
5633 | unsigned long netdev_increment_features(unsigned long all, unsigned long one, |
5634 | unsigned long mask) | |
5635 | { | |
5636 | /* If device needs checksumming, downgrade to it. */ | |
d1b19dff | 5637 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) |
b63365a2 HX |
5638 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); |
5639 | else if (mask & NETIF_F_ALL_CSUM) { | |
5640 | /* If one device supports v4/v6 checksumming, set for all. */ | |
5641 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && | |
5642 | !(all & NETIF_F_GEN_CSUM)) { | |
5643 | all &= ~NETIF_F_ALL_CSUM; | |
5644 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
5645 | } | |
e2a6b852 | 5646 | |
b63365a2 HX |
5647 | /* If one device supports hw checksumming, set for all. */ |
5648 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { | |
5649 | all &= ~NETIF_F_ALL_CSUM; | |
5650 | all |= NETIF_F_HW_CSUM; | |
5651 | } | |
5652 | } | |
7f353bf2 | 5653 | |
b63365a2 | 5654 | one |= NETIF_F_ALL_CSUM; |
7f353bf2 | 5655 | |
b63365a2 | 5656 | one |= all & NETIF_F_ONE_FOR_ALL; |
d9f5950f | 5657 | all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO; |
b63365a2 | 5658 | all |= one & mask & NETIF_F_ONE_FOR_ALL; |
7f353bf2 HX |
5659 | |
5660 | return all; | |
5661 | } | |
b63365a2 | 5662 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 5663 | |
30d97d35 PE |
5664 | static struct hlist_head *netdev_create_hash(void) |
5665 | { | |
5666 | int i; | |
5667 | struct hlist_head *hash; | |
5668 | ||
5669 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
5670 | if (hash != NULL) | |
5671 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
5672 | INIT_HLIST_HEAD(&hash[i]); | |
5673 | ||
5674 | return hash; | |
5675 | } | |
5676 | ||
881d966b | 5677 | /* Initialize per network namespace state */ |
4665079c | 5678 | static int __net_init netdev_init(struct net *net) |
881d966b | 5679 | { |
881d966b | 5680 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 5681 | |
30d97d35 PE |
5682 | net->dev_name_head = netdev_create_hash(); |
5683 | if (net->dev_name_head == NULL) | |
5684 | goto err_name; | |
881d966b | 5685 | |
30d97d35 PE |
5686 | net->dev_index_head = netdev_create_hash(); |
5687 | if (net->dev_index_head == NULL) | |
5688 | goto err_idx; | |
881d966b EB |
5689 | |
5690 | return 0; | |
30d97d35 PE |
5691 | |
5692 | err_idx: | |
5693 | kfree(net->dev_name_head); | |
5694 | err_name: | |
5695 | return -ENOMEM; | |
881d966b EB |
5696 | } |
5697 | ||
f0db275a SH |
5698 | /** |
5699 | * netdev_drivername - network driver for the device | |
5700 | * @dev: network device | |
5701 | * @buffer: buffer for resulting name | |
5702 | * @len: size of buffer | |
5703 | * | |
5704 | * Determine network driver for device. | |
5705 | */ | |
cf04a4c7 | 5706 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) |
6579e57b | 5707 | { |
cf04a4c7 SH |
5708 | const struct device_driver *driver; |
5709 | const struct device *parent; | |
6579e57b AV |
5710 | |
5711 | if (len <= 0 || !buffer) | |
5712 | return buffer; | |
5713 | buffer[0] = 0; | |
5714 | ||
5715 | parent = dev->dev.parent; | |
5716 | ||
5717 | if (!parent) | |
5718 | return buffer; | |
5719 | ||
5720 | driver = parent->driver; | |
5721 | if (driver && driver->name) | |
5722 | strlcpy(buffer, driver->name, len); | |
5723 | return buffer; | |
5724 | } | |
5725 | ||
4665079c | 5726 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
5727 | { |
5728 | kfree(net->dev_name_head); | |
5729 | kfree(net->dev_index_head); | |
5730 | } | |
5731 | ||
022cbae6 | 5732 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
5733 | .init = netdev_init, |
5734 | .exit = netdev_exit, | |
5735 | }; | |
5736 | ||
4665079c | 5737 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 5738 | { |
e008b5fc | 5739 | struct net_device *dev, *aux; |
ce286d32 | 5740 | /* |
e008b5fc | 5741 | * Push all migratable network devices back to the |
ce286d32 EB |
5742 | * initial network namespace |
5743 | */ | |
5744 | rtnl_lock(); | |
e008b5fc | 5745 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 5746 | int err; |
aca51397 | 5747 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
5748 | |
5749 | /* Ignore unmoveable devices (i.e. loopback) */ | |
5750 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5751 | continue; | |
5752 | ||
e008b5fc EB |
5753 | /* Leave virtual devices for the generic cleanup */ |
5754 | if (dev->rtnl_link_ops) | |
5755 | continue; | |
d0c082ce | 5756 | |
ce286d32 | 5757 | /* Push remaing network devices to init_net */ |
aca51397 PE |
5758 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
5759 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 5760 | if (err) { |
aca51397 | 5761 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 5762 | __func__, dev->name, err); |
aca51397 | 5763 | BUG(); |
ce286d32 EB |
5764 | } |
5765 | } | |
5766 | rtnl_unlock(); | |
5767 | } | |
5768 | ||
04dc7f6b EB |
5769 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
5770 | { | |
5771 | /* At exit all network devices most be removed from a network | |
5772 | * namespace. Do this in the reverse order of registeration. | |
5773 | * Do this across as many network namespaces as possible to | |
5774 | * improve batching efficiency. | |
5775 | */ | |
5776 | struct net_device *dev; | |
5777 | struct net *net; | |
5778 | LIST_HEAD(dev_kill_list); | |
5779 | ||
5780 | rtnl_lock(); | |
5781 | list_for_each_entry(net, net_list, exit_list) { | |
5782 | for_each_netdev_reverse(net, dev) { | |
5783 | if (dev->rtnl_link_ops) | |
5784 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
5785 | else | |
5786 | unregister_netdevice_queue(dev, &dev_kill_list); | |
5787 | } | |
5788 | } | |
5789 | unregister_netdevice_many(&dev_kill_list); | |
5790 | rtnl_unlock(); | |
5791 | } | |
5792 | ||
022cbae6 | 5793 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 5794 | .exit = default_device_exit, |
04dc7f6b | 5795 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
5796 | }; |
5797 | ||
1da177e4 LT |
5798 | /* |
5799 | * Initialize the DEV module. At boot time this walks the device list and | |
5800 | * unhooks any devices that fail to initialise (normally hardware not | |
5801 | * present) and leaves us with a valid list of present and active devices. | |
5802 | * | |
5803 | */ | |
5804 | ||
5805 | /* | |
5806 | * This is called single threaded during boot, so no need | |
5807 | * to take the rtnl semaphore. | |
5808 | */ | |
5809 | static int __init net_dev_init(void) | |
5810 | { | |
5811 | int i, rc = -ENOMEM; | |
5812 | ||
5813 | BUG_ON(!dev_boot_phase); | |
5814 | ||
1da177e4 LT |
5815 | if (dev_proc_init()) |
5816 | goto out; | |
5817 | ||
8b41d188 | 5818 | if (netdev_kobject_init()) |
1da177e4 LT |
5819 | goto out; |
5820 | ||
5821 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 5822 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
5823 | INIT_LIST_HEAD(&ptype_base[i]); |
5824 | ||
881d966b EB |
5825 | if (register_pernet_subsys(&netdev_net_ops)) |
5826 | goto out; | |
1da177e4 LT |
5827 | |
5828 | /* | |
5829 | * Initialise the packet receive queues. | |
5830 | */ | |
5831 | ||
6f912042 | 5832 | for_each_possible_cpu(i) { |
1da177e4 LT |
5833 | struct softnet_data *queue; |
5834 | ||
5835 | queue = &per_cpu(softnet_data, i); | |
5836 | skb_queue_head_init(&queue->input_pkt_queue); | |
1da177e4 LT |
5837 | queue->completion_queue = NULL; |
5838 | INIT_LIST_HEAD(&queue->poll_list); | |
bea3348e | 5839 | |
df334545 | 5840 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5841 | queue->csd.func = trigger_softirq; |
5842 | queue->csd.info = queue; | |
5843 | queue->csd.flags = 0; | |
1e94d72f | 5844 | #endif |
0a9627f2 | 5845 | |
bea3348e SH |
5846 | queue->backlog.poll = process_backlog; |
5847 | queue->backlog.weight = weight_p; | |
d565b0a1 | 5848 | queue->backlog.gro_list = NULL; |
4ae5544f | 5849 | queue->backlog.gro_count = 0; |
1da177e4 LT |
5850 | } |
5851 | ||
1da177e4 LT |
5852 | dev_boot_phase = 0; |
5853 | ||
505d4f73 EB |
5854 | /* The loopback device is special if any other network devices |
5855 | * is present in a network namespace the loopback device must | |
5856 | * be present. Since we now dynamically allocate and free the | |
5857 | * loopback device ensure this invariant is maintained by | |
5858 | * keeping the loopback device as the first device on the | |
5859 | * list of network devices. Ensuring the loopback devices | |
5860 | * is the first device that appears and the last network device | |
5861 | * that disappears. | |
5862 | */ | |
5863 | if (register_pernet_device(&loopback_net_ops)) | |
5864 | goto out; | |
5865 | ||
5866 | if (register_pernet_device(&default_device_ops)) | |
5867 | goto out; | |
5868 | ||
962cf36c CM |
5869 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
5870 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
5871 | |
5872 | hotcpu_notifier(dev_cpu_callback, 0); | |
5873 | dst_init(); | |
5874 | dev_mcast_init(); | |
5875 | rc = 0; | |
5876 | out: | |
5877 | return rc; | |
5878 | } | |
5879 | ||
5880 | subsys_initcall(net_dev_init); | |
5881 | ||
e88721f8 KK |
5882 | static int __init initialize_hashrnd(void) |
5883 | { | |
0a9627f2 | 5884 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
5885 | return 0; |
5886 | } | |
5887 | ||
5888 | late_initcall_sync(initialize_hashrnd); | |
5889 |