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 |
0a9627f2 TH |
2206 | /* |
2207 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
2208 | * CPU from the RPS map of the receiving queue for a given skb. | |
2209 | */ | |
2210 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb) | |
2211 | { | |
2212 | struct ipv6hdr *ip6; | |
2213 | struct iphdr *ip; | |
2214 | struct netdev_rx_queue *rxqueue; | |
2215 | struct rps_map *map; | |
2216 | int cpu = -1; | |
2217 | u8 ip_proto; | |
2218 | u32 addr1, addr2, ports, ihl; | |
2219 | ||
2220 | rcu_read_lock(); | |
2221 | ||
2222 | if (skb_rx_queue_recorded(skb)) { | |
2223 | u16 index = skb_get_rx_queue(skb); | |
2224 | if (unlikely(index >= dev->num_rx_queues)) { | |
2225 | if (net_ratelimit()) { | |
7a161ea9 ED |
2226 | pr_warning("%s received packet on queue " |
2227 | "%u, but number of RX queues is %u\n", | |
2228 | dev->name, index, dev->num_rx_queues); | |
0a9627f2 TH |
2229 | } |
2230 | goto done; | |
2231 | } | |
2232 | rxqueue = dev->_rx + index; | |
2233 | } else | |
2234 | rxqueue = dev->_rx; | |
2235 | ||
2236 | if (!rxqueue->rps_map) | |
2237 | goto done; | |
2238 | ||
2239 | if (skb->rxhash) | |
2240 | goto got_hash; /* Skip hash computation on packet header */ | |
2241 | ||
2242 | switch (skb->protocol) { | |
2243 | case __constant_htons(ETH_P_IP): | |
2244 | if (!pskb_may_pull(skb, sizeof(*ip))) | |
2245 | goto done; | |
2246 | ||
2247 | ip = (struct iphdr *) skb->data; | |
2248 | ip_proto = ip->protocol; | |
2249 | addr1 = ip->saddr; | |
2250 | addr2 = ip->daddr; | |
2251 | ihl = ip->ihl; | |
2252 | break; | |
2253 | case __constant_htons(ETH_P_IPV6): | |
2254 | if (!pskb_may_pull(skb, sizeof(*ip6))) | |
2255 | goto done; | |
2256 | ||
2257 | ip6 = (struct ipv6hdr *) skb->data; | |
2258 | ip_proto = ip6->nexthdr; | |
2259 | addr1 = ip6->saddr.s6_addr32[3]; | |
2260 | addr2 = ip6->daddr.s6_addr32[3]; | |
2261 | ihl = (40 >> 2); | |
2262 | break; | |
2263 | default: | |
2264 | goto done; | |
2265 | } | |
2266 | ports = 0; | |
2267 | switch (ip_proto) { | |
2268 | case IPPROTO_TCP: | |
2269 | case IPPROTO_UDP: | |
2270 | case IPPROTO_DCCP: | |
2271 | case IPPROTO_ESP: | |
2272 | case IPPROTO_AH: | |
2273 | case IPPROTO_SCTP: | |
2274 | case IPPROTO_UDPLITE: | |
2275 | if (pskb_may_pull(skb, (ihl * 4) + 4)) | |
2276 | ports = *((u32 *) (skb->data + (ihl * 4))); | |
2277 | break; | |
2278 | ||
2279 | default: | |
2280 | break; | |
2281 | } | |
2282 | ||
2283 | skb->rxhash = jhash_3words(addr1, addr2, ports, hashrnd); | |
2284 | if (!skb->rxhash) | |
2285 | skb->rxhash = 1; | |
2286 | ||
2287 | got_hash: | |
2288 | map = rcu_dereference(rxqueue->rps_map); | |
2289 | if (map) { | |
2290 | u16 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32]; | |
2291 | ||
2292 | if (cpu_online(tcpu)) { | |
2293 | cpu = tcpu; | |
2294 | goto done; | |
2295 | } | |
2296 | } | |
2297 | ||
2298 | done: | |
2299 | rcu_read_unlock(); | |
2300 | return cpu; | |
2301 | } | |
2302 | ||
2303 | /* | |
2304 | * This structure holds the per-CPU mask of CPUs for which IPIs are scheduled | |
2305 | * to be sent to kick remote softirq processing. There are two masks since | |
2306 | * the sending of IPIs must be done with interrupts enabled. The select field | |
2307 | * indicates the current mask that enqueue_backlog uses to schedule IPIs. | |
2308 | * select is flipped before net_rps_action is called while still under lock, | |
2309 | * net_rps_action then uses the non-selected mask to send the IPIs and clears | |
2310 | * it without conflicting with enqueue_backlog operation. | |
2311 | */ | |
2312 | struct rps_remote_softirq_cpus { | |
2313 | cpumask_t mask[2]; | |
2314 | int select; | |
2315 | }; | |
2316 | static DEFINE_PER_CPU(struct rps_remote_softirq_cpus, rps_remote_softirq_cpus); | |
2317 | ||
2318 | /* Called from hardirq (IPI) context */ | |
2319 | static void trigger_softirq(void *data) | |
2320 | { | |
2321 | struct softnet_data *queue = data; | |
2322 | __napi_schedule(&queue->backlog); | |
2323 | __get_cpu_var(netdev_rx_stat).received_rps++; | |
2324 | } | |
1e94d72f | 2325 | #endif /* CONFIG_SMP */ |
0a9627f2 TH |
2326 | |
2327 | /* | |
2328 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
2329 | * queue (may be a remote CPU queue). | |
2330 | */ | |
2331 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu) | |
2332 | { | |
2333 | struct softnet_data *queue; | |
2334 | unsigned long flags; | |
2335 | ||
2336 | queue = &per_cpu(softnet_data, cpu); | |
2337 | ||
2338 | local_irq_save(flags); | |
2339 | __get_cpu_var(netdev_rx_stat).total++; | |
2340 | ||
152102c7 | 2341 | rps_lock(queue); |
0a9627f2 TH |
2342 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { |
2343 | if (queue->input_pkt_queue.qlen) { | |
2344 | enqueue: | |
2345 | __skb_queue_tail(&queue->input_pkt_queue, skb); | |
152102c7 CG |
2346 | rps_unlock(queue); |
2347 | local_irq_restore(flags); | |
0a9627f2 TH |
2348 | return NET_RX_SUCCESS; |
2349 | } | |
2350 | ||
2351 | /* Schedule NAPI for backlog device */ | |
2352 | if (napi_schedule_prep(&queue->backlog)) { | |
df334545 | 2353 | #ifdef CONFIG_RPS |
0a9627f2 TH |
2354 | if (cpu != smp_processor_id()) { |
2355 | struct rps_remote_softirq_cpus *rcpus = | |
2356 | &__get_cpu_var(rps_remote_softirq_cpus); | |
2357 | ||
2358 | cpu_set(cpu, rcpus->mask[rcpus->select]); | |
2359 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2360 | } else | |
2361 | __napi_schedule(&queue->backlog); | |
1e94d72f TH |
2362 | #else |
2363 | __napi_schedule(&queue->backlog); | |
2364 | #endif | |
0a9627f2 TH |
2365 | } |
2366 | goto enqueue; | |
2367 | } | |
2368 | ||
152102c7 | 2369 | rps_unlock(queue); |
0a9627f2 TH |
2370 | |
2371 | __get_cpu_var(netdev_rx_stat).dropped++; | |
2372 | local_irq_restore(flags); | |
2373 | ||
2374 | kfree_skb(skb); | |
2375 | return NET_RX_DROP; | |
2376 | } | |
1da177e4 | 2377 | |
1da177e4 LT |
2378 | /** |
2379 | * netif_rx - post buffer to the network code | |
2380 | * @skb: buffer to post | |
2381 | * | |
2382 | * This function receives a packet from a device driver and queues it for | |
2383 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
2384 | * may be dropped during processing for congestion control or by the | |
2385 | * protocol layers. | |
2386 | * | |
2387 | * return values: | |
2388 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
2389 | * NET_RX_DROP (packet was dropped) |
2390 | * | |
2391 | */ | |
2392 | ||
2393 | int netif_rx(struct sk_buff *skb) | |
2394 | { | |
0a9627f2 | 2395 | int cpu; |
1da177e4 LT |
2396 | |
2397 | /* if netpoll wants it, pretend we never saw it */ | |
2398 | if (netpoll_rx(skb)) | |
2399 | return NET_RX_DROP; | |
2400 | ||
b7aa0bf7 | 2401 | if (!skb->tstamp.tv64) |
a61bbcf2 | 2402 | net_timestamp(skb); |
1da177e4 | 2403 | |
df334545 | 2404 | #ifdef CONFIG_RPS |
0a9627f2 TH |
2405 | cpu = get_rps_cpu(skb->dev, skb); |
2406 | if (cpu < 0) | |
2407 | cpu = smp_processor_id(); | |
1e94d72f TH |
2408 | #else |
2409 | cpu = smp_processor_id(); | |
2410 | #endif | |
1da177e4 | 2411 | |
0a9627f2 | 2412 | return enqueue_to_backlog(skb, cpu); |
1da177e4 | 2413 | } |
d1b19dff | 2414 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
2415 | |
2416 | int netif_rx_ni(struct sk_buff *skb) | |
2417 | { | |
2418 | int err; | |
2419 | ||
2420 | preempt_disable(); | |
2421 | err = netif_rx(skb); | |
2422 | if (local_softirq_pending()) | |
2423 | do_softirq(); | |
2424 | preempt_enable(); | |
2425 | ||
2426 | return err; | |
2427 | } | |
1da177e4 LT |
2428 | EXPORT_SYMBOL(netif_rx_ni); |
2429 | ||
1da177e4 LT |
2430 | static void net_tx_action(struct softirq_action *h) |
2431 | { | |
2432 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2433 | ||
2434 | if (sd->completion_queue) { | |
2435 | struct sk_buff *clist; | |
2436 | ||
2437 | local_irq_disable(); | |
2438 | clist = sd->completion_queue; | |
2439 | sd->completion_queue = NULL; | |
2440 | local_irq_enable(); | |
2441 | ||
2442 | while (clist) { | |
2443 | struct sk_buff *skb = clist; | |
2444 | clist = clist->next; | |
2445 | ||
547b792c | 2446 | WARN_ON(atomic_read(&skb->users)); |
1da177e4 LT |
2447 | __kfree_skb(skb); |
2448 | } | |
2449 | } | |
2450 | ||
2451 | if (sd->output_queue) { | |
37437bb2 | 2452 | struct Qdisc *head; |
1da177e4 LT |
2453 | |
2454 | local_irq_disable(); | |
2455 | head = sd->output_queue; | |
2456 | sd->output_queue = NULL; | |
2457 | local_irq_enable(); | |
2458 | ||
2459 | while (head) { | |
37437bb2 DM |
2460 | struct Qdisc *q = head; |
2461 | spinlock_t *root_lock; | |
2462 | ||
1da177e4 LT |
2463 | head = head->next_sched; |
2464 | ||
5fb66229 | 2465 | root_lock = qdisc_lock(q); |
37437bb2 | 2466 | if (spin_trylock(root_lock)) { |
def82a1d JP |
2467 | smp_mb__before_clear_bit(); |
2468 | clear_bit(__QDISC_STATE_SCHED, | |
2469 | &q->state); | |
37437bb2 DM |
2470 | qdisc_run(q); |
2471 | spin_unlock(root_lock); | |
1da177e4 | 2472 | } else { |
195648bb | 2473 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 2474 | &q->state)) { |
195648bb | 2475 | __netif_reschedule(q); |
e8a83e10 JP |
2476 | } else { |
2477 | smp_mb__before_clear_bit(); | |
2478 | clear_bit(__QDISC_STATE_SCHED, | |
2479 | &q->state); | |
2480 | } | |
1da177e4 LT |
2481 | } |
2482 | } | |
2483 | } | |
2484 | } | |
2485 | ||
6f05f629 SH |
2486 | static inline int deliver_skb(struct sk_buff *skb, |
2487 | struct packet_type *pt_prev, | |
2488 | struct net_device *orig_dev) | |
1da177e4 LT |
2489 | { |
2490 | atomic_inc(&skb->users); | |
f2ccd8fa | 2491 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2492 | } |
2493 | ||
2494 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) | |
da678292 MM |
2495 | |
2496 | #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE) | |
2497 | /* This hook is defined here for ATM LANE */ | |
2498 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
2499 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 2500 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 2501 | #endif |
1da177e4 | 2502 | |
6229e362 SH |
2503 | /* |
2504 | * If bridge module is loaded call bridging hook. | |
2505 | * returns NULL if packet was consumed. | |
2506 | */ | |
2507 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, | |
2508 | struct sk_buff *skb) __read_mostly; | |
4fb019a0 | 2509 | EXPORT_SYMBOL_GPL(br_handle_frame_hook); |
da678292 | 2510 | |
6229e362 SH |
2511 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb, |
2512 | struct packet_type **pt_prev, int *ret, | |
2513 | struct net_device *orig_dev) | |
1da177e4 LT |
2514 | { |
2515 | struct net_bridge_port *port; | |
2516 | ||
6229e362 SH |
2517 | if (skb->pkt_type == PACKET_LOOPBACK || |
2518 | (port = rcu_dereference(skb->dev->br_port)) == NULL) | |
2519 | return skb; | |
1da177e4 LT |
2520 | |
2521 | if (*pt_prev) { | |
6229e362 | 2522 | *ret = deliver_skb(skb, *pt_prev, orig_dev); |
1da177e4 | 2523 | *pt_prev = NULL; |
4ec93edb YH |
2524 | } |
2525 | ||
6229e362 | 2526 | return br_handle_frame_hook(port, skb); |
1da177e4 LT |
2527 | } |
2528 | #else | |
6229e362 | 2529 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) |
1da177e4 LT |
2530 | #endif |
2531 | ||
b863ceb7 PM |
2532 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) |
2533 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; | |
2534 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); | |
2535 | ||
2536 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, | |
2537 | struct packet_type **pt_prev, | |
2538 | int *ret, | |
2539 | struct net_device *orig_dev) | |
2540 | { | |
2541 | if (skb->dev->macvlan_port == NULL) | |
2542 | return skb; | |
2543 | ||
2544 | if (*pt_prev) { | |
2545 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2546 | *pt_prev = NULL; | |
2547 | } | |
2548 | return macvlan_handle_frame_hook(skb); | |
2549 | } | |
2550 | #else | |
2551 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) | |
2552 | #endif | |
2553 | ||
1da177e4 LT |
2554 | #ifdef CONFIG_NET_CLS_ACT |
2555 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2556 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2557 | * a compare and 2 stores extra right now if we dont have it on | |
2558 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 2559 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
2560 | * the ingress scheduler, you just cant add policies on ingress. |
2561 | * | |
2562 | */ | |
4ec93edb | 2563 | static int ing_filter(struct sk_buff *skb) |
1da177e4 | 2564 | { |
1da177e4 | 2565 | struct net_device *dev = skb->dev; |
f697c3e8 | 2566 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
2567 | struct netdev_queue *rxq; |
2568 | int result = TC_ACT_OK; | |
2569 | struct Qdisc *q; | |
4ec93edb | 2570 | |
f697c3e8 HX |
2571 | if (MAX_RED_LOOP < ttl++) { |
2572 | printk(KERN_WARNING | |
2573 | "Redir loop detected Dropping packet (%d->%d)\n", | |
8964be4a | 2574 | skb->skb_iif, dev->ifindex); |
f697c3e8 HX |
2575 | return TC_ACT_SHOT; |
2576 | } | |
1da177e4 | 2577 | |
f697c3e8 HX |
2578 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
2579 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 2580 | |
555353cf DM |
2581 | rxq = &dev->rx_queue; |
2582 | ||
83874000 | 2583 | q = rxq->qdisc; |
8d50b53d | 2584 | if (q != &noop_qdisc) { |
83874000 | 2585 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
2586 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
2587 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
2588 | spin_unlock(qdisc_lock(q)); |
2589 | } | |
f697c3e8 HX |
2590 | |
2591 | return result; | |
2592 | } | |
86e65da9 | 2593 | |
f697c3e8 HX |
2594 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
2595 | struct packet_type **pt_prev, | |
2596 | int *ret, struct net_device *orig_dev) | |
2597 | { | |
8d50b53d | 2598 | if (skb->dev->rx_queue.qdisc == &noop_qdisc) |
f697c3e8 | 2599 | goto out; |
1da177e4 | 2600 | |
f697c3e8 HX |
2601 | if (*pt_prev) { |
2602 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2603 | *pt_prev = NULL; | |
2604 | } else { | |
2605 | /* Huh? Why does turning on AF_PACKET affect this? */ | |
2606 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); | |
1da177e4 LT |
2607 | } |
2608 | ||
f697c3e8 HX |
2609 | switch (ing_filter(skb)) { |
2610 | case TC_ACT_SHOT: | |
2611 | case TC_ACT_STOLEN: | |
2612 | kfree_skb(skb); | |
2613 | return NULL; | |
2614 | } | |
2615 | ||
2616 | out: | |
2617 | skb->tc_verd = 0; | |
2618 | return skb; | |
1da177e4 LT |
2619 | } |
2620 | #endif | |
2621 | ||
bc1d0411 PM |
2622 | /* |
2623 | * netif_nit_deliver - deliver received packets to network taps | |
2624 | * @skb: buffer | |
2625 | * | |
2626 | * This function is used to deliver incoming packets to network | |
2627 | * taps. It should be used when the normal netif_receive_skb path | |
2628 | * is bypassed, for example because of VLAN acceleration. | |
2629 | */ | |
2630 | void netif_nit_deliver(struct sk_buff *skb) | |
2631 | { | |
2632 | struct packet_type *ptype; | |
2633 | ||
2634 | if (list_empty(&ptype_all)) | |
2635 | return; | |
2636 | ||
2637 | skb_reset_network_header(skb); | |
2638 | skb_reset_transport_header(skb); | |
2639 | skb->mac_len = skb->network_header - skb->mac_header; | |
2640 | ||
2641 | rcu_read_lock(); | |
2642 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2643 | if (!ptype->dev || ptype->dev == skb->dev) | |
2644 | deliver_skb(skb, ptype, skb->dev); | |
2645 | } | |
2646 | rcu_read_unlock(); | |
2647 | } | |
2648 | ||
acbbc071 ED |
2649 | static inline void skb_bond_set_mac_by_master(struct sk_buff *skb, |
2650 | struct net_device *master) | |
2651 | { | |
2652 | if (skb->pkt_type == PACKET_HOST) { | |
2653 | u16 *dest = (u16 *) eth_hdr(skb)->h_dest; | |
2654 | ||
2655 | memcpy(dest, master->dev_addr, ETH_ALEN); | |
2656 | } | |
2657 | } | |
2658 | ||
2659 | /* On bonding slaves other than the currently active slave, suppress | |
2660 | * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and | |
2661 | * ARP on active-backup slaves with arp_validate enabled. | |
2662 | */ | |
2663 | int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master) | |
2664 | { | |
2665 | struct net_device *dev = skb->dev; | |
2666 | ||
2667 | if (master->priv_flags & IFF_MASTER_ARPMON) | |
2668 | dev->last_rx = jiffies; | |
2669 | ||
2670 | if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) { | |
2671 | /* Do address unmangle. The local destination address | |
2672 | * will be always the one master has. Provides the right | |
2673 | * functionality in a bridge. | |
2674 | */ | |
2675 | skb_bond_set_mac_by_master(skb, master); | |
2676 | } | |
2677 | ||
2678 | if (dev->priv_flags & IFF_SLAVE_INACTIVE) { | |
2679 | if ((dev->priv_flags & IFF_SLAVE_NEEDARP) && | |
2680 | skb->protocol == __cpu_to_be16(ETH_P_ARP)) | |
2681 | return 0; | |
2682 | ||
2683 | if (master->priv_flags & IFF_MASTER_ALB) { | |
2684 | if (skb->pkt_type != PACKET_BROADCAST && | |
2685 | skb->pkt_type != PACKET_MULTICAST) | |
2686 | return 0; | |
2687 | } | |
2688 | if (master->priv_flags & IFF_MASTER_8023AD && | |
2689 | skb->protocol == __cpu_to_be16(ETH_P_SLOW)) | |
2690 | return 0; | |
2691 | ||
2692 | return 1; | |
2693 | } | |
2694 | return 0; | |
2695 | } | |
2696 | EXPORT_SYMBOL(__skb_bond_should_drop); | |
2697 | ||
10f744d2 | 2698 | static int __netif_receive_skb(struct sk_buff *skb) |
1da177e4 LT |
2699 | { |
2700 | struct packet_type *ptype, *pt_prev; | |
f2ccd8fa | 2701 | struct net_device *orig_dev; |
0641e4fb | 2702 | struct net_device *master; |
0d7a3681 | 2703 | struct net_device *null_or_orig; |
ca8d9ea3 | 2704 | struct net_device *null_or_bond; |
1da177e4 | 2705 | int ret = NET_RX_DROP; |
252e3346 | 2706 | __be16 type; |
1da177e4 | 2707 | |
81bbb3d4 ED |
2708 | if (!skb->tstamp.tv64) |
2709 | net_timestamp(skb); | |
2710 | ||
05423b24 | 2711 | if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb)) |
9b22ea56 PM |
2712 | return NET_RX_SUCCESS; |
2713 | ||
1da177e4 | 2714 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 2715 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
2716 | return NET_RX_DROP; |
2717 | ||
8964be4a ED |
2718 | if (!skb->skb_iif) |
2719 | skb->skb_iif = skb->dev->ifindex; | |
86e65da9 | 2720 | |
0d7a3681 | 2721 | null_or_orig = NULL; |
cc9bd5ce | 2722 | orig_dev = skb->dev; |
0641e4fb ED |
2723 | master = ACCESS_ONCE(orig_dev->master); |
2724 | if (master) { | |
2725 | if (skb_bond_should_drop(skb, master)) | |
0d7a3681 JE |
2726 | null_or_orig = orig_dev; /* deliver only exact match */ |
2727 | else | |
0641e4fb | 2728 | skb->dev = master; |
cc9bd5ce | 2729 | } |
8f903c70 | 2730 | |
1da177e4 LT |
2731 | __get_cpu_var(netdev_rx_stat).total++; |
2732 | ||
c1d2bbe1 | 2733 | skb_reset_network_header(skb); |
badff6d0 | 2734 | skb_reset_transport_header(skb); |
b0e380b1 | 2735 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
2736 | |
2737 | pt_prev = NULL; | |
2738 | ||
2739 | rcu_read_lock(); | |
2740 | ||
2741 | #ifdef CONFIG_NET_CLS_ACT | |
2742 | if (skb->tc_verd & TC_NCLS) { | |
2743 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2744 | goto ncls; | |
2745 | } | |
2746 | #endif | |
2747 | ||
2748 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
f982307f JE |
2749 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
2750 | ptype->dev == orig_dev) { | |
4ec93edb | 2751 | if (pt_prev) |
f2ccd8fa | 2752 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2753 | pt_prev = ptype; |
2754 | } | |
2755 | } | |
2756 | ||
2757 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
2758 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
2759 | if (!skb) | |
1da177e4 | 2760 | goto out; |
1da177e4 LT |
2761 | ncls: |
2762 | #endif | |
2763 | ||
6229e362 | 2764 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); |
b863ceb7 PM |
2765 | if (!skb) |
2766 | goto out; | |
2767 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); | |
6229e362 | 2768 | if (!skb) |
1da177e4 LT |
2769 | goto out; |
2770 | ||
1f3c8804 AG |
2771 | /* |
2772 | * Make sure frames received on VLAN interfaces stacked on | |
2773 | * bonding interfaces still make their way to any base bonding | |
2774 | * device that may have registered for a specific ptype. The | |
2775 | * handler may have to adjust skb->dev and orig_dev. | |
1f3c8804 | 2776 | */ |
ca8d9ea3 | 2777 | null_or_bond = NULL; |
1f3c8804 AG |
2778 | if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) && |
2779 | (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) { | |
ca8d9ea3 | 2780 | null_or_bond = vlan_dev_real_dev(skb->dev); |
1f3c8804 AG |
2781 | } |
2782 | ||
1da177e4 | 2783 | type = skb->protocol; |
82d8a867 PE |
2784 | list_for_each_entry_rcu(ptype, |
2785 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1f3c8804 | 2786 | if (ptype->type == type && (ptype->dev == null_or_orig || |
ca8d9ea3 AG |
2787 | ptype->dev == skb->dev || ptype->dev == orig_dev || |
2788 | ptype->dev == null_or_bond)) { | |
4ec93edb | 2789 | if (pt_prev) |
f2ccd8fa | 2790 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2791 | pt_prev = ptype; |
2792 | } | |
2793 | } | |
2794 | ||
2795 | if (pt_prev) { | |
f2ccd8fa | 2796 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2797 | } else { |
2798 | kfree_skb(skb); | |
2799 | /* Jamal, now you will not able to escape explaining | |
2800 | * me how you were going to use this. :-) | |
2801 | */ | |
2802 | ret = NET_RX_DROP; | |
2803 | } | |
2804 | ||
2805 | out: | |
2806 | rcu_read_unlock(); | |
2807 | return ret; | |
2808 | } | |
0a9627f2 TH |
2809 | |
2810 | /** | |
2811 | * netif_receive_skb - process receive buffer from network | |
2812 | * @skb: buffer to process | |
2813 | * | |
2814 | * netif_receive_skb() is the main receive data processing function. | |
2815 | * It always succeeds. The buffer may be dropped during processing | |
2816 | * for congestion control or by the protocol layers. | |
2817 | * | |
2818 | * This function may only be called from softirq context and interrupts | |
2819 | * should be enabled. | |
2820 | * | |
2821 | * Return values (usually ignored): | |
2822 | * NET_RX_SUCCESS: no congestion | |
2823 | * NET_RX_DROP: packet was dropped | |
2824 | */ | |
2825 | int netif_receive_skb(struct sk_buff *skb) | |
2826 | { | |
df334545 | 2827 | #ifdef CONFIG_RPS |
0a9627f2 TH |
2828 | int cpu; |
2829 | ||
2830 | cpu = get_rps_cpu(skb->dev, skb); | |
2831 | ||
2832 | if (cpu < 0) | |
2833 | return __netif_receive_skb(skb); | |
2834 | else | |
2835 | return enqueue_to_backlog(skb, cpu); | |
1e94d72f TH |
2836 | #else |
2837 | return __netif_receive_skb(skb); | |
2838 | #endif | |
0a9627f2 | 2839 | } |
d1b19dff | 2840 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 2841 | |
6e583ce5 | 2842 | /* Network device is going away, flush any packets still pending */ |
152102c7 | 2843 | static void flush_backlog(void *arg) |
6e583ce5 | 2844 | { |
152102c7 CG |
2845 | struct net_device *dev = arg; |
2846 | struct softnet_data *queue = &__get_cpu_var(softnet_data); | |
6e583ce5 SH |
2847 | struct sk_buff *skb, *tmp; |
2848 | ||
152102c7 | 2849 | rps_lock(queue); |
6e583ce5 SH |
2850 | skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp) |
2851 | if (skb->dev == dev) { | |
2852 | __skb_unlink(skb, &queue->input_pkt_queue); | |
2853 | kfree_skb(skb); | |
2854 | } | |
152102c7 | 2855 | rps_unlock(queue); |
6e583ce5 SH |
2856 | } |
2857 | ||
d565b0a1 HX |
2858 | static int napi_gro_complete(struct sk_buff *skb) |
2859 | { | |
2860 | struct packet_type *ptype; | |
2861 | __be16 type = skb->protocol; | |
2862 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
2863 | int err = -ENOENT; | |
2864 | ||
fc59f9a3 HX |
2865 | if (NAPI_GRO_CB(skb)->count == 1) { |
2866 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 2867 | goto out; |
fc59f9a3 | 2868 | } |
d565b0a1 HX |
2869 | |
2870 | rcu_read_lock(); | |
2871 | list_for_each_entry_rcu(ptype, head, list) { | |
2872 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
2873 | continue; | |
2874 | ||
2875 | err = ptype->gro_complete(skb); | |
2876 | break; | |
2877 | } | |
2878 | rcu_read_unlock(); | |
2879 | ||
2880 | if (err) { | |
2881 | WARN_ON(&ptype->list == head); | |
2882 | kfree_skb(skb); | |
2883 | return NET_RX_SUCCESS; | |
2884 | } | |
2885 | ||
2886 | out: | |
d565b0a1 HX |
2887 | return netif_receive_skb(skb); |
2888 | } | |
2889 | ||
11380a4b | 2890 | static void napi_gro_flush(struct napi_struct *napi) |
d565b0a1 HX |
2891 | { |
2892 | struct sk_buff *skb, *next; | |
2893 | ||
2894 | for (skb = napi->gro_list; skb; skb = next) { | |
2895 | next = skb->next; | |
2896 | skb->next = NULL; | |
2897 | napi_gro_complete(skb); | |
2898 | } | |
2899 | ||
4ae5544f | 2900 | napi->gro_count = 0; |
d565b0a1 HX |
2901 | napi->gro_list = NULL; |
2902 | } | |
d565b0a1 | 2903 | |
5b252f0c | 2904 | enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
2905 | { |
2906 | struct sk_buff **pp = NULL; | |
2907 | struct packet_type *ptype; | |
2908 | __be16 type = skb->protocol; | |
2909 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 2910 | int same_flow; |
d565b0a1 | 2911 | int mac_len; |
5b252f0c | 2912 | enum gro_result ret; |
d565b0a1 HX |
2913 | |
2914 | if (!(skb->dev->features & NETIF_F_GRO)) | |
2915 | goto normal; | |
2916 | ||
4cf704fb | 2917 | if (skb_is_gso(skb) || skb_has_frags(skb)) |
f17f5c91 HX |
2918 | goto normal; |
2919 | ||
d565b0a1 HX |
2920 | rcu_read_lock(); |
2921 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
2922 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
2923 | continue; | |
2924 | ||
86911732 | 2925 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
2926 | mac_len = skb->network_header - skb->mac_header; |
2927 | skb->mac_len = mac_len; | |
2928 | NAPI_GRO_CB(skb)->same_flow = 0; | |
2929 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 2930 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 2931 | |
d565b0a1 HX |
2932 | pp = ptype->gro_receive(&napi->gro_list, skb); |
2933 | break; | |
2934 | } | |
2935 | rcu_read_unlock(); | |
2936 | ||
2937 | if (&ptype->list == head) | |
2938 | goto normal; | |
2939 | ||
0da2afd5 | 2940 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 2941 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 2942 | |
d565b0a1 HX |
2943 | if (pp) { |
2944 | struct sk_buff *nskb = *pp; | |
2945 | ||
2946 | *pp = nskb->next; | |
2947 | nskb->next = NULL; | |
2948 | napi_gro_complete(nskb); | |
4ae5544f | 2949 | napi->gro_count--; |
d565b0a1 HX |
2950 | } |
2951 | ||
0da2afd5 | 2952 | if (same_flow) |
d565b0a1 HX |
2953 | goto ok; |
2954 | ||
4ae5544f | 2955 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 2956 | goto normal; |
d565b0a1 | 2957 | |
4ae5544f | 2958 | napi->gro_count++; |
d565b0a1 | 2959 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 2960 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
2961 | skb->next = napi->gro_list; |
2962 | napi->gro_list = skb; | |
5d0d9be8 | 2963 | ret = GRO_HELD; |
d565b0a1 | 2964 | |
ad0f9904 | 2965 | pull: |
cb18978c HX |
2966 | if (skb_headlen(skb) < skb_gro_offset(skb)) { |
2967 | int grow = skb_gro_offset(skb) - skb_headlen(skb); | |
2968 | ||
2969 | BUG_ON(skb->end - skb->tail < grow); | |
2970 | ||
2971 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
2972 | ||
2973 | skb->tail += grow; | |
2974 | skb->data_len -= grow; | |
2975 | ||
2976 | skb_shinfo(skb)->frags[0].page_offset += grow; | |
2977 | skb_shinfo(skb)->frags[0].size -= grow; | |
2978 | ||
2979 | if (unlikely(!skb_shinfo(skb)->frags[0].size)) { | |
2980 | put_page(skb_shinfo(skb)->frags[0].page); | |
2981 | memmove(skb_shinfo(skb)->frags, | |
2982 | skb_shinfo(skb)->frags + 1, | |
2983 | --skb_shinfo(skb)->nr_frags); | |
2984 | } | |
ad0f9904 HX |
2985 | } |
2986 | ||
d565b0a1 | 2987 | ok: |
5d0d9be8 | 2988 | return ret; |
d565b0a1 HX |
2989 | |
2990 | normal: | |
ad0f9904 HX |
2991 | ret = GRO_NORMAL; |
2992 | goto pull; | |
5d38a079 | 2993 | } |
96e93eab HX |
2994 | EXPORT_SYMBOL(dev_gro_receive); |
2995 | ||
5b252f0c BH |
2996 | static gro_result_t |
2997 | __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
96e93eab HX |
2998 | { |
2999 | struct sk_buff *p; | |
3000 | ||
d1c76af9 HX |
3001 | if (netpoll_rx_on(skb)) |
3002 | return GRO_NORMAL; | |
3003 | ||
96e93eab | 3004 | for (p = napi->gro_list; p; p = p->next) { |
f64f9e71 JP |
3005 | NAPI_GRO_CB(p)->same_flow = |
3006 | (p->dev == skb->dev) && | |
3007 | !compare_ether_header(skb_mac_header(p), | |
3008 | skb_gro_mac_header(skb)); | |
96e93eab HX |
3009 | NAPI_GRO_CB(p)->flush = 0; |
3010 | } | |
3011 | ||
3012 | return dev_gro_receive(napi, skb); | |
3013 | } | |
5d38a079 | 3014 | |
c7c4b3b6 | 3015 | gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 3016 | { |
5d0d9be8 HX |
3017 | switch (ret) { |
3018 | case GRO_NORMAL: | |
c7c4b3b6 BH |
3019 | if (netif_receive_skb(skb)) |
3020 | ret = GRO_DROP; | |
3021 | break; | |
5d38a079 | 3022 | |
5d0d9be8 | 3023 | case GRO_DROP: |
5d0d9be8 | 3024 | case GRO_MERGED_FREE: |
5d38a079 HX |
3025 | kfree_skb(skb); |
3026 | break; | |
5b252f0c BH |
3027 | |
3028 | case GRO_HELD: | |
3029 | case GRO_MERGED: | |
3030 | break; | |
5d38a079 HX |
3031 | } |
3032 | ||
c7c4b3b6 | 3033 | return ret; |
5d0d9be8 HX |
3034 | } |
3035 | EXPORT_SYMBOL(napi_skb_finish); | |
3036 | ||
78a478d0 HX |
3037 | void skb_gro_reset_offset(struct sk_buff *skb) |
3038 | { | |
3039 | NAPI_GRO_CB(skb)->data_offset = 0; | |
3040 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
7489594c | 3041 | NAPI_GRO_CB(skb)->frag0_len = 0; |
78a478d0 | 3042 | |
78d3fd0b | 3043 | if (skb->mac_header == skb->tail && |
7489594c | 3044 | !PageHighMem(skb_shinfo(skb)->frags[0].page)) { |
78a478d0 HX |
3045 | NAPI_GRO_CB(skb)->frag0 = |
3046 | page_address(skb_shinfo(skb)->frags[0].page) + | |
3047 | skb_shinfo(skb)->frags[0].page_offset; | |
7489594c HX |
3048 | NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size; |
3049 | } | |
78a478d0 HX |
3050 | } |
3051 | EXPORT_SYMBOL(skb_gro_reset_offset); | |
3052 | ||
c7c4b3b6 | 3053 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 3054 | { |
86911732 HX |
3055 | skb_gro_reset_offset(skb); |
3056 | ||
5d0d9be8 | 3057 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
3058 | } |
3059 | EXPORT_SYMBOL(napi_gro_receive); | |
3060 | ||
96e93eab HX |
3061 | void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
3062 | { | |
96e93eab HX |
3063 | __skb_pull(skb, skb_headlen(skb)); |
3064 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
3065 | ||
3066 | napi->skb = skb; | |
3067 | } | |
3068 | EXPORT_SYMBOL(napi_reuse_skb); | |
3069 | ||
76620aaf | 3070 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 3071 | { |
5d38a079 | 3072 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
3073 | |
3074 | if (!skb) { | |
89d71a66 ED |
3075 | skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD); |
3076 | if (skb) | |
3077 | napi->skb = skb; | |
80595d59 | 3078 | } |
96e93eab HX |
3079 | return skb; |
3080 | } | |
76620aaf | 3081 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 3082 | |
c7c4b3b6 BH |
3083 | gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, |
3084 | gro_result_t ret) | |
96e93eab | 3085 | { |
5d0d9be8 HX |
3086 | switch (ret) { |
3087 | case GRO_NORMAL: | |
86911732 | 3088 | case GRO_HELD: |
e76b69cc | 3089 | skb->protocol = eth_type_trans(skb, skb->dev); |
86911732 | 3090 | |
c7c4b3b6 BH |
3091 | if (ret == GRO_HELD) |
3092 | skb_gro_pull(skb, -ETH_HLEN); | |
3093 | else if (netif_receive_skb(skb)) | |
3094 | ret = GRO_DROP; | |
86911732 | 3095 | break; |
5d38a079 | 3096 | |
5d0d9be8 | 3097 | case GRO_DROP: |
5d0d9be8 HX |
3098 | case GRO_MERGED_FREE: |
3099 | napi_reuse_skb(napi, skb); | |
3100 | break; | |
5b252f0c BH |
3101 | |
3102 | case GRO_MERGED: | |
3103 | break; | |
5d0d9be8 | 3104 | } |
5d38a079 | 3105 | |
c7c4b3b6 | 3106 | return ret; |
5d38a079 | 3107 | } |
5d0d9be8 HX |
3108 | EXPORT_SYMBOL(napi_frags_finish); |
3109 | ||
76620aaf HX |
3110 | struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
3111 | { | |
3112 | struct sk_buff *skb = napi->skb; | |
3113 | struct ethhdr *eth; | |
a5b1cf28 HX |
3114 | unsigned int hlen; |
3115 | unsigned int off; | |
76620aaf HX |
3116 | |
3117 | napi->skb = NULL; | |
3118 | ||
3119 | skb_reset_mac_header(skb); | |
3120 | skb_gro_reset_offset(skb); | |
3121 | ||
a5b1cf28 HX |
3122 | off = skb_gro_offset(skb); |
3123 | hlen = off + sizeof(*eth); | |
3124 | eth = skb_gro_header_fast(skb, off); | |
3125 | if (skb_gro_header_hard(skb, hlen)) { | |
3126 | eth = skb_gro_header_slow(skb, hlen, off); | |
3127 | if (unlikely(!eth)) { | |
3128 | napi_reuse_skb(napi, skb); | |
3129 | skb = NULL; | |
3130 | goto out; | |
3131 | } | |
76620aaf HX |
3132 | } |
3133 | ||
3134 | skb_gro_pull(skb, sizeof(*eth)); | |
3135 | ||
3136 | /* | |
3137 | * This works because the only protocols we care about don't require | |
3138 | * special handling. We'll fix it up properly at the end. | |
3139 | */ | |
3140 | skb->protocol = eth->h_proto; | |
3141 | ||
3142 | out: | |
3143 | return skb; | |
3144 | } | |
3145 | EXPORT_SYMBOL(napi_frags_skb); | |
3146 | ||
c7c4b3b6 | 3147 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 3148 | { |
76620aaf | 3149 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
3150 | |
3151 | if (!skb) | |
c7c4b3b6 | 3152 | return GRO_DROP; |
5d0d9be8 HX |
3153 | |
3154 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
3155 | } | |
5d38a079 HX |
3156 | EXPORT_SYMBOL(napi_gro_frags); |
3157 | ||
bea3348e | 3158 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
3159 | { |
3160 | int work = 0; | |
1da177e4 LT |
3161 | struct softnet_data *queue = &__get_cpu_var(softnet_data); |
3162 | unsigned long start_time = jiffies; | |
3163 | ||
bea3348e SH |
3164 | napi->weight = weight_p; |
3165 | do { | |
1da177e4 | 3166 | struct sk_buff *skb; |
1da177e4 | 3167 | |
152102c7 CG |
3168 | local_irq_disable(); |
3169 | rps_lock(queue); | |
1da177e4 | 3170 | skb = __skb_dequeue(&queue->input_pkt_queue); |
bea3348e | 3171 | if (!skb) { |
8f1ead2d | 3172 | __napi_complete(napi); |
5a6d234e | 3173 | rps_unlock(queue); |
e4008276 | 3174 | local_irq_enable(); |
8f1ead2d | 3175 | break; |
bea3348e | 3176 | } |
152102c7 CG |
3177 | rps_unlock(queue); |
3178 | local_irq_enable(); | |
1da177e4 | 3179 | |
0a9627f2 | 3180 | __netif_receive_skb(skb); |
bea3348e | 3181 | } while (++work < quota && jiffies == start_time); |
1da177e4 | 3182 | |
bea3348e SH |
3183 | return work; |
3184 | } | |
1da177e4 | 3185 | |
bea3348e SH |
3186 | /** |
3187 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 3188 | * @n: entry to schedule |
bea3348e SH |
3189 | * |
3190 | * The entry's receive function will be scheduled to run | |
3191 | */ | |
b5606c2d | 3192 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
3193 | { |
3194 | unsigned long flags; | |
1da177e4 | 3195 | |
bea3348e SH |
3196 | local_irq_save(flags); |
3197 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); | |
3198 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3199 | local_irq_restore(flags); | |
1da177e4 | 3200 | } |
bea3348e SH |
3201 | EXPORT_SYMBOL(__napi_schedule); |
3202 | ||
d565b0a1 HX |
3203 | void __napi_complete(struct napi_struct *n) |
3204 | { | |
3205 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
3206 | BUG_ON(n->gro_list); | |
3207 | ||
3208 | list_del(&n->poll_list); | |
3209 | smp_mb__before_clear_bit(); | |
3210 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
3211 | } | |
3212 | EXPORT_SYMBOL(__napi_complete); | |
3213 | ||
3214 | void napi_complete(struct napi_struct *n) | |
3215 | { | |
3216 | unsigned long flags; | |
3217 | ||
3218 | /* | |
3219 | * don't let napi dequeue from the cpu poll list | |
3220 | * just in case its running on a different cpu | |
3221 | */ | |
3222 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
3223 | return; | |
3224 | ||
3225 | napi_gro_flush(n); | |
3226 | local_irq_save(flags); | |
3227 | __napi_complete(n); | |
3228 | local_irq_restore(flags); | |
3229 | } | |
3230 | EXPORT_SYMBOL(napi_complete); | |
3231 | ||
3232 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
3233 | int (*poll)(struct napi_struct *, int), int weight) | |
3234 | { | |
3235 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 3236 | napi->gro_count = 0; |
d565b0a1 | 3237 | napi->gro_list = NULL; |
5d38a079 | 3238 | napi->skb = NULL; |
d565b0a1 HX |
3239 | napi->poll = poll; |
3240 | napi->weight = weight; | |
3241 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 3242 | napi->dev = dev; |
5d38a079 | 3243 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
3244 | spin_lock_init(&napi->poll_lock); |
3245 | napi->poll_owner = -1; | |
3246 | #endif | |
3247 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
3248 | } | |
3249 | EXPORT_SYMBOL(netif_napi_add); | |
3250 | ||
3251 | void netif_napi_del(struct napi_struct *napi) | |
3252 | { | |
3253 | struct sk_buff *skb, *next; | |
3254 | ||
d7b06636 | 3255 | list_del_init(&napi->dev_list); |
76620aaf | 3256 | napi_free_frags(napi); |
d565b0a1 HX |
3257 | |
3258 | for (skb = napi->gro_list; skb; skb = next) { | |
3259 | next = skb->next; | |
3260 | skb->next = NULL; | |
3261 | kfree_skb(skb); | |
3262 | } | |
3263 | ||
3264 | napi->gro_list = NULL; | |
4ae5544f | 3265 | napi->gro_count = 0; |
d565b0a1 HX |
3266 | } |
3267 | EXPORT_SYMBOL(netif_napi_del); | |
3268 | ||
df334545 | 3269 | #ifdef CONFIG_RPS |
0a9627f2 TH |
3270 | /* |
3271 | * net_rps_action sends any pending IPI's for rps. This is only called from | |
3272 | * softirq and interrupts must be enabled. | |
3273 | */ | |
3274 | static void net_rps_action(cpumask_t *mask) | |
3275 | { | |
3276 | int cpu; | |
3277 | ||
3278 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
3279 | for_each_cpu_mask_nr(cpu, *mask) { | |
3280 | struct softnet_data *queue = &per_cpu(softnet_data, cpu); | |
3281 | if (cpu_online(cpu)) | |
3282 | __smp_call_function_single(cpu, &queue->csd, 0); | |
3283 | } | |
3284 | cpus_clear(*mask); | |
3285 | } | |
1e94d72f | 3286 | #endif |
1da177e4 LT |
3287 | |
3288 | static void net_rx_action(struct softirq_action *h) | |
3289 | { | |
bea3348e | 3290 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list; |
24f8b238 | 3291 | unsigned long time_limit = jiffies + 2; |
51b0bded | 3292 | int budget = netdev_budget; |
53fb95d3 | 3293 | void *have; |
df334545 | 3294 | #ifdef CONFIG_RPS |
0a9627f2 TH |
3295 | int select; |
3296 | struct rps_remote_softirq_cpus *rcpus; | |
1e94d72f | 3297 | #endif |
53fb95d3 | 3298 | |
1da177e4 LT |
3299 | local_irq_disable(); |
3300 | ||
bea3348e SH |
3301 | while (!list_empty(list)) { |
3302 | struct napi_struct *n; | |
3303 | int work, weight; | |
1da177e4 | 3304 | |
bea3348e | 3305 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
3306 | * Allow this to run for 2 jiffies since which will allow |
3307 | * an average latency of 1.5/HZ. | |
bea3348e | 3308 | */ |
24f8b238 | 3309 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
3310 | goto softnet_break; |
3311 | ||
3312 | local_irq_enable(); | |
3313 | ||
bea3348e SH |
3314 | /* Even though interrupts have been re-enabled, this |
3315 | * access is safe because interrupts can only add new | |
3316 | * entries to the tail of this list, and only ->poll() | |
3317 | * calls can remove this head entry from the list. | |
3318 | */ | |
e5e26d75 | 3319 | n = list_first_entry(list, struct napi_struct, poll_list); |
1da177e4 | 3320 | |
bea3348e SH |
3321 | have = netpoll_poll_lock(n); |
3322 | ||
3323 | weight = n->weight; | |
3324 | ||
0a7606c1 DM |
3325 | /* This NAPI_STATE_SCHED test is for avoiding a race |
3326 | * with netpoll's poll_napi(). Only the entity which | |
3327 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
3328 | * actually make the ->poll() call. Therefore we avoid | |
3329 | * accidently calling ->poll() when NAPI is not scheduled. | |
3330 | */ | |
3331 | work = 0; | |
4ea7e386 | 3332 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { |
0a7606c1 | 3333 | work = n->poll(n, weight); |
4ea7e386 NH |
3334 | trace_napi_poll(n); |
3335 | } | |
bea3348e SH |
3336 | |
3337 | WARN_ON_ONCE(work > weight); | |
3338 | ||
3339 | budget -= work; | |
3340 | ||
3341 | local_irq_disable(); | |
3342 | ||
3343 | /* Drivers must not modify the NAPI state if they | |
3344 | * consume the entire weight. In such cases this code | |
3345 | * still "owns" the NAPI instance and therefore can | |
3346 | * move the instance around on the list at-will. | |
3347 | */ | |
fed17f30 | 3348 | if (unlikely(work == weight)) { |
ff780cd8 HX |
3349 | if (unlikely(napi_disable_pending(n))) { |
3350 | local_irq_enable(); | |
3351 | napi_complete(n); | |
3352 | local_irq_disable(); | |
3353 | } else | |
fed17f30 DM |
3354 | list_move_tail(&n->poll_list, list); |
3355 | } | |
bea3348e SH |
3356 | |
3357 | netpoll_poll_unlock(have); | |
1da177e4 LT |
3358 | } |
3359 | out: | |
df334545 | 3360 | #ifdef CONFIG_RPS |
0a9627f2 TH |
3361 | rcpus = &__get_cpu_var(rps_remote_softirq_cpus); |
3362 | select = rcpus->select; | |
3363 | rcpus->select ^= 1; | |
3364 | ||
515e06c4 | 3365 | local_irq_enable(); |
bea3348e | 3366 | |
0a9627f2 | 3367 | net_rps_action(&rcpus->mask[select]); |
1e94d72f TH |
3368 | #else |
3369 | local_irq_enable(); | |
3370 | #endif | |
0a9627f2 | 3371 | |
db217334 CL |
3372 | #ifdef CONFIG_NET_DMA |
3373 | /* | |
3374 | * There may not be any more sk_buffs coming right now, so push | |
3375 | * any pending DMA copies to hardware | |
3376 | */ | |
2ba05622 | 3377 | dma_issue_pending_all(); |
db217334 | 3378 | #endif |
bea3348e | 3379 | |
1da177e4 LT |
3380 | return; |
3381 | ||
3382 | softnet_break: | |
3383 | __get_cpu_var(netdev_rx_stat).time_squeeze++; | |
3384 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3385 | goto out; | |
3386 | } | |
3387 | ||
d1b19dff | 3388 | static gifconf_func_t *gifconf_list[NPROTO]; |
1da177e4 LT |
3389 | |
3390 | /** | |
3391 | * register_gifconf - register a SIOCGIF handler | |
3392 | * @family: Address family | |
3393 | * @gifconf: Function handler | |
3394 | * | |
3395 | * Register protocol dependent address dumping routines. The handler | |
3396 | * that is passed must not be freed or reused until it has been replaced | |
3397 | * by another handler. | |
3398 | */ | |
d1b19dff | 3399 | int register_gifconf(unsigned int family, gifconf_func_t *gifconf) |
1da177e4 LT |
3400 | { |
3401 | if (family >= NPROTO) | |
3402 | return -EINVAL; | |
3403 | gifconf_list[family] = gifconf; | |
3404 | return 0; | |
3405 | } | |
d1b19dff | 3406 | EXPORT_SYMBOL(register_gifconf); |
1da177e4 LT |
3407 | |
3408 | ||
3409 | /* | |
3410 | * Map an interface index to its name (SIOCGIFNAME) | |
3411 | */ | |
3412 | ||
3413 | /* | |
3414 | * We need this ioctl for efficient implementation of the | |
3415 | * if_indextoname() function required by the IPv6 API. Without | |
3416 | * it, we would have to search all the interfaces to find a | |
3417 | * match. --pb | |
3418 | */ | |
3419 | ||
881d966b | 3420 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
3421 | { |
3422 | struct net_device *dev; | |
3423 | struct ifreq ifr; | |
3424 | ||
3425 | /* | |
3426 | * Fetch the caller's info block. | |
3427 | */ | |
3428 | ||
3429 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3430 | return -EFAULT; | |
3431 | ||
fb699dfd ED |
3432 | rcu_read_lock(); |
3433 | dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex); | |
1da177e4 | 3434 | if (!dev) { |
fb699dfd | 3435 | rcu_read_unlock(); |
1da177e4 LT |
3436 | return -ENODEV; |
3437 | } | |
3438 | ||
3439 | strcpy(ifr.ifr_name, dev->name); | |
fb699dfd | 3440 | rcu_read_unlock(); |
1da177e4 LT |
3441 | |
3442 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
3443 | return -EFAULT; | |
3444 | return 0; | |
3445 | } | |
3446 | ||
3447 | /* | |
3448 | * Perform a SIOCGIFCONF call. This structure will change | |
3449 | * size eventually, and there is nothing I can do about it. | |
3450 | * Thus we will need a 'compatibility mode'. | |
3451 | */ | |
3452 | ||
881d966b | 3453 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
3454 | { |
3455 | struct ifconf ifc; | |
3456 | struct net_device *dev; | |
3457 | char __user *pos; | |
3458 | int len; | |
3459 | int total; | |
3460 | int i; | |
3461 | ||
3462 | /* | |
3463 | * Fetch the caller's info block. | |
3464 | */ | |
3465 | ||
3466 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
3467 | return -EFAULT; | |
3468 | ||
3469 | pos = ifc.ifc_buf; | |
3470 | len = ifc.ifc_len; | |
3471 | ||
3472 | /* | |
3473 | * Loop over the interfaces, and write an info block for each. | |
3474 | */ | |
3475 | ||
3476 | total = 0; | |
881d966b | 3477 | for_each_netdev(net, dev) { |
1da177e4 LT |
3478 | for (i = 0; i < NPROTO; i++) { |
3479 | if (gifconf_list[i]) { | |
3480 | int done; | |
3481 | if (!pos) | |
3482 | done = gifconf_list[i](dev, NULL, 0); | |
3483 | else | |
3484 | done = gifconf_list[i](dev, pos + total, | |
3485 | len - total); | |
3486 | if (done < 0) | |
3487 | return -EFAULT; | |
3488 | total += done; | |
3489 | } | |
3490 | } | |
4ec93edb | 3491 | } |
1da177e4 LT |
3492 | |
3493 | /* | |
3494 | * All done. Write the updated control block back to the caller. | |
3495 | */ | |
3496 | ifc.ifc_len = total; | |
3497 | ||
3498 | /* | |
3499 | * Both BSD and Solaris return 0 here, so we do too. | |
3500 | */ | |
3501 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
3502 | } | |
3503 | ||
3504 | #ifdef CONFIG_PROC_FS | |
3505 | /* | |
3506 | * This is invoked by the /proc filesystem handler to display a device | |
3507 | * in detail. | |
3508 | */ | |
7562f876 | 3509 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
c6d14c84 | 3510 | __acquires(RCU) |
1da177e4 | 3511 | { |
e372c414 | 3512 | struct net *net = seq_file_net(seq); |
7562f876 | 3513 | loff_t off; |
1da177e4 | 3514 | struct net_device *dev; |
1da177e4 | 3515 | |
c6d14c84 | 3516 | rcu_read_lock(); |
7562f876 PE |
3517 | if (!*pos) |
3518 | return SEQ_START_TOKEN; | |
1da177e4 | 3519 | |
7562f876 | 3520 | off = 1; |
c6d14c84 | 3521 | for_each_netdev_rcu(net, dev) |
7562f876 PE |
3522 | if (off++ == *pos) |
3523 | return dev; | |
1da177e4 | 3524 | |
7562f876 | 3525 | return NULL; |
1da177e4 LT |
3526 | } |
3527 | ||
3528 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3529 | { | |
c6d14c84 ED |
3530 | struct net_device *dev = (v == SEQ_START_TOKEN) ? |
3531 | first_net_device(seq_file_net(seq)) : | |
3532 | next_net_device((struct net_device *)v); | |
3533 | ||
1da177e4 | 3534 | ++*pos; |
c6d14c84 | 3535 | return rcu_dereference(dev); |
1da177e4 LT |
3536 | } |
3537 | ||
3538 | void dev_seq_stop(struct seq_file *seq, void *v) | |
c6d14c84 | 3539 | __releases(RCU) |
1da177e4 | 3540 | { |
c6d14c84 | 3541 | rcu_read_unlock(); |
1da177e4 LT |
3542 | } |
3543 | ||
3544 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
3545 | { | |
eeda3fd6 | 3546 | const struct net_device_stats *stats = dev_get_stats(dev); |
1da177e4 | 3547 | |
2d13bafe | 3548 | seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " |
5a1b5898 RR |
3549 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", |
3550 | dev->name, stats->rx_bytes, stats->rx_packets, | |
3551 | stats->rx_errors, | |
3552 | stats->rx_dropped + stats->rx_missed_errors, | |
3553 | stats->rx_fifo_errors, | |
3554 | stats->rx_length_errors + stats->rx_over_errors + | |
3555 | stats->rx_crc_errors + stats->rx_frame_errors, | |
3556 | stats->rx_compressed, stats->multicast, | |
3557 | stats->tx_bytes, stats->tx_packets, | |
3558 | stats->tx_errors, stats->tx_dropped, | |
3559 | stats->tx_fifo_errors, stats->collisions, | |
3560 | stats->tx_carrier_errors + | |
3561 | stats->tx_aborted_errors + | |
3562 | stats->tx_window_errors + | |
3563 | stats->tx_heartbeat_errors, | |
3564 | stats->tx_compressed); | |
1da177e4 LT |
3565 | } |
3566 | ||
3567 | /* | |
3568 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
3569 | * /proc/net interface to create /proc/net/dev | |
3570 | */ | |
3571 | static int dev_seq_show(struct seq_file *seq, void *v) | |
3572 | { | |
3573 | if (v == SEQ_START_TOKEN) | |
3574 | seq_puts(seq, "Inter-| Receive " | |
3575 | " | Transmit\n" | |
3576 | " face |bytes packets errs drop fifo frame " | |
3577 | "compressed multicast|bytes packets errs " | |
3578 | "drop fifo colls carrier compressed\n"); | |
3579 | else | |
3580 | dev_seq_printf_stats(seq, v); | |
3581 | return 0; | |
3582 | } | |
3583 | ||
3584 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) | |
3585 | { | |
3586 | struct netif_rx_stats *rc = NULL; | |
3587 | ||
0c0b0aca | 3588 | while (*pos < nr_cpu_ids) |
4ec93edb | 3589 | if (cpu_online(*pos)) { |
1da177e4 LT |
3590 | rc = &per_cpu(netdev_rx_stat, *pos); |
3591 | break; | |
3592 | } else | |
3593 | ++*pos; | |
3594 | return rc; | |
3595 | } | |
3596 | ||
3597 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
3598 | { | |
3599 | return softnet_get_online(pos); | |
3600 | } | |
3601 | ||
3602 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3603 | { | |
3604 | ++*pos; | |
3605 | return softnet_get_online(pos); | |
3606 | } | |
3607 | ||
3608 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
3609 | { | |
3610 | } | |
3611 | ||
3612 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
3613 | { | |
3614 | struct netif_rx_stats *s = v; | |
3615 | ||
0a9627f2 | 3616 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
31aa02c5 | 3617 | s->total, s->dropped, s->time_squeeze, 0, |
c1ebcdb8 | 3618 | 0, 0, 0, 0, /* was fastroute */ |
0a9627f2 | 3619 | s->cpu_collision, s->received_rps); |
1da177e4 LT |
3620 | return 0; |
3621 | } | |
3622 | ||
f690808e | 3623 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
3624 | .start = dev_seq_start, |
3625 | .next = dev_seq_next, | |
3626 | .stop = dev_seq_stop, | |
3627 | .show = dev_seq_show, | |
3628 | }; | |
3629 | ||
3630 | static int dev_seq_open(struct inode *inode, struct file *file) | |
3631 | { | |
e372c414 DL |
3632 | return seq_open_net(inode, file, &dev_seq_ops, |
3633 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
3634 | } |
3635 | ||
9a32144e | 3636 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
3637 | .owner = THIS_MODULE, |
3638 | .open = dev_seq_open, | |
3639 | .read = seq_read, | |
3640 | .llseek = seq_lseek, | |
e372c414 | 3641 | .release = seq_release_net, |
1da177e4 LT |
3642 | }; |
3643 | ||
f690808e | 3644 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
3645 | .start = softnet_seq_start, |
3646 | .next = softnet_seq_next, | |
3647 | .stop = softnet_seq_stop, | |
3648 | .show = softnet_seq_show, | |
3649 | }; | |
3650 | ||
3651 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
3652 | { | |
3653 | return seq_open(file, &softnet_seq_ops); | |
3654 | } | |
3655 | ||
9a32144e | 3656 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
3657 | .owner = THIS_MODULE, |
3658 | .open = softnet_seq_open, | |
3659 | .read = seq_read, | |
3660 | .llseek = seq_lseek, | |
3661 | .release = seq_release, | |
3662 | }; | |
3663 | ||
0e1256ff SH |
3664 | static void *ptype_get_idx(loff_t pos) |
3665 | { | |
3666 | struct packet_type *pt = NULL; | |
3667 | loff_t i = 0; | |
3668 | int t; | |
3669 | ||
3670 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
3671 | if (i == pos) | |
3672 | return pt; | |
3673 | ++i; | |
3674 | } | |
3675 | ||
82d8a867 | 3676 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
3677 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
3678 | if (i == pos) | |
3679 | return pt; | |
3680 | ++i; | |
3681 | } | |
3682 | } | |
3683 | return NULL; | |
3684 | } | |
3685 | ||
3686 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 3687 | __acquires(RCU) |
0e1256ff SH |
3688 | { |
3689 | rcu_read_lock(); | |
3690 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
3691 | } | |
3692 | ||
3693 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3694 | { | |
3695 | struct packet_type *pt; | |
3696 | struct list_head *nxt; | |
3697 | int hash; | |
3698 | ||
3699 | ++*pos; | |
3700 | if (v == SEQ_START_TOKEN) | |
3701 | return ptype_get_idx(0); | |
3702 | ||
3703 | pt = v; | |
3704 | nxt = pt->list.next; | |
3705 | if (pt->type == htons(ETH_P_ALL)) { | |
3706 | if (nxt != &ptype_all) | |
3707 | goto found; | |
3708 | hash = 0; | |
3709 | nxt = ptype_base[0].next; | |
3710 | } else | |
82d8a867 | 3711 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
3712 | |
3713 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 3714 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
3715 | return NULL; |
3716 | nxt = ptype_base[hash].next; | |
3717 | } | |
3718 | found: | |
3719 | return list_entry(nxt, struct packet_type, list); | |
3720 | } | |
3721 | ||
3722 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 3723 | __releases(RCU) |
0e1256ff SH |
3724 | { |
3725 | rcu_read_unlock(); | |
3726 | } | |
3727 | ||
0e1256ff SH |
3728 | static int ptype_seq_show(struct seq_file *seq, void *v) |
3729 | { | |
3730 | struct packet_type *pt = v; | |
3731 | ||
3732 | if (v == SEQ_START_TOKEN) | |
3733 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 3734 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
3735 | if (pt->type == htons(ETH_P_ALL)) |
3736 | seq_puts(seq, "ALL "); | |
3737 | else | |
3738 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
3739 | ||
908cd2da AD |
3740 | seq_printf(seq, " %-8s %pF\n", |
3741 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
3742 | } |
3743 | ||
3744 | return 0; | |
3745 | } | |
3746 | ||
3747 | static const struct seq_operations ptype_seq_ops = { | |
3748 | .start = ptype_seq_start, | |
3749 | .next = ptype_seq_next, | |
3750 | .stop = ptype_seq_stop, | |
3751 | .show = ptype_seq_show, | |
3752 | }; | |
3753 | ||
3754 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
3755 | { | |
2feb27db PE |
3756 | return seq_open_net(inode, file, &ptype_seq_ops, |
3757 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
3758 | } |
3759 | ||
3760 | static const struct file_operations ptype_seq_fops = { | |
3761 | .owner = THIS_MODULE, | |
3762 | .open = ptype_seq_open, | |
3763 | .read = seq_read, | |
3764 | .llseek = seq_lseek, | |
2feb27db | 3765 | .release = seq_release_net, |
0e1256ff SH |
3766 | }; |
3767 | ||
3768 | ||
4665079c | 3769 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
3770 | { |
3771 | int rc = -ENOMEM; | |
3772 | ||
881d966b | 3773 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 3774 | goto out; |
881d966b | 3775 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 3776 | goto out_dev; |
881d966b | 3777 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 3778 | goto out_softnet; |
0e1256ff | 3779 | |
881d966b | 3780 | if (wext_proc_init(net)) |
457c4cbc | 3781 | goto out_ptype; |
1da177e4 LT |
3782 | rc = 0; |
3783 | out: | |
3784 | return rc; | |
457c4cbc | 3785 | out_ptype: |
881d966b | 3786 | proc_net_remove(net, "ptype"); |
1da177e4 | 3787 | out_softnet: |
881d966b | 3788 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 3789 | out_dev: |
881d966b | 3790 | proc_net_remove(net, "dev"); |
1da177e4 LT |
3791 | goto out; |
3792 | } | |
881d966b | 3793 | |
4665079c | 3794 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
3795 | { |
3796 | wext_proc_exit(net); | |
3797 | ||
3798 | proc_net_remove(net, "ptype"); | |
3799 | proc_net_remove(net, "softnet_stat"); | |
3800 | proc_net_remove(net, "dev"); | |
3801 | } | |
3802 | ||
022cbae6 | 3803 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
3804 | .init = dev_proc_net_init, |
3805 | .exit = dev_proc_net_exit, | |
3806 | }; | |
3807 | ||
3808 | static int __init dev_proc_init(void) | |
3809 | { | |
3810 | return register_pernet_subsys(&dev_proc_ops); | |
3811 | } | |
1da177e4 LT |
3812 | #else |
3813 | #define dev_proc_init() 0 | |
3814 | #endif /* CONFIG_PROC_FS */ | |
3815 | ||
3816 | ||
3817 | /** | |
3818 | * netdev_set_master - set up master/slave pair | |
3819 | * @slave: slave device | |
3820 | * @master: new master device | |
3821 | * | |
3822 | * Changes the master device of the slave. Pass %NULL to break the | |
3823 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
3824 | * a negative errno code is returned. On success the reference counts | |
3825 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
3826 | * function returns zero. | |
3827 | */ | |
3828 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
3829 | { | |
3830 | struct net_device *old = slave->master; | |
3831 | ||
3832 | ASSERT_RTNL(); | |
3833 | ||
3834 | if (master) { | |
3835 | if (old) | |
3836 | return -EBUSY; | |
3837 | dev_hold(master); | |
3838 | } | |
3839 | ||
3840 | slave->master = master; | |
4ec93edb | 3841 | |
283f2fe8 ED |
3842 | if (old) { |
3843 | synchronize_net(); | |
1da177e4 | 3844 | dev_put(old); |
283f2fe8 | 3845 | } |
1da177e4 LT |
3846 | if (master) |
3847 | slave->flags |= IFF_SLAVE; | |
3848 | else | |
3849 | slave->flags &= ~IFF_SLAVE; | |
3850 | ||
3851 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
3852 | return 0; | |
3853 | } | |
d1b19dff | 3854 | EXPORT_SYMBOL(netdev_set_master); |
1da177e4 | 3855 | |
b6c40d68 PM |
3856 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
3857 | { | |
d314774c SH |
3858 | const struct net_device_ops *ops = dev->netdev_ops; |
3859 | ||
3860 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
3861 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
3862 | } |
3863 | ||
dad9b335 | 3864 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
3865 | { |
3866 | unsigned short old_flags = dev->flags; | |
8192b0c4 DH |
3867 | uid_t uid; |
3868 | gid_t gid; | |
1da177e4 | 3869 | |
24023451 PM |
3870 | ASSERT_RTNL(); |
3871 | ||
dad9b335 WC |
3872 | dev->flags |= IFF_PROMISC; |
3873 | dev->promiscuity += inc; | |
3874 | if (dev->promiscuity == 0) { | |
3875 | /* | |
3876 | * Avoid overflow. | |
3877 | * If inc causes overflow, untouch promisc and return error. | |
3878 | */ | |
3879 | if (inc < 0) | |
3880 | dev->flags &= ~IFF_PROMISC; | |
3881 | else { | |
3882 | dev->promiscuity -= inc; | |
3883 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
3884 | "set promiscuity failed, promiscuity feature " | |
3885 | "of device might be broken.\n", dev->name); | |
3886 | return -EOVERFLOW; | |
3887 | } | |
3888 | } | |
52609c0b | 3889 | if (dev->flags != old_flags) { |
1da177e4 LT |
3890 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
3891 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 3892 | "left"); |
8192b0c4 DH |
3893 | if (audit_enabled) { |
3894 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
3895 | audit_log(current->audit_context, GFP_ATOMIC, |
3896 | AUDIT_ANOM_PROMISCUOUS, | |
3897 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
3898 | dev->name, (dev->flags & IFF_PROMISC), | |
3899 | (old_flags & IFF_PROMISC), | |
3900 | audit_get_loginuid(current), | |
8192b0c4 | 3901 | uid, gid, |
7759db82 | 3902 | audit_get_sessionid(current)); |
8192b0c4 | 3903 | } |
24023451 | 3904 | |
b6c40d68 | 3905 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 3906 | } |
dad9b335 | 3907 | return 0; |
1da177e4 LT |
3908 | } |
3909 | ||
4417da66 PM |
3910 | /** |
3911 | * dev_set_promiscuity - update promiscuity count on a device | |
3912 | * @dev: device | |
3913 | * @inc: modifier | |
3914 | * | |
3915 | * Add or remove promiscuity from a device. While the count in the device | |
3916 | * remains above zero the interface remains promiscuous. Once it hits zero | |
3917 | * the device reverts back to normal filtering operation. A negative inc | |
3918 | * value is used to drop promiscuity on the device. | |
dad9b335 | 3919 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 3920 | */ |
dad9b335 | 3921 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
3922 | { |
3923 | unsigned short old_flags = dev->flags; | |
dad9b335 | 3924 | int err; |
4417da66 | 3925 | |
dad9b335 | 3926 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 3927 | if (err < 0) |
dad9b335 | 3928 | return err; |
4417da66 PM |
3929 | if (dev->flags != old_flags) |
3930 | dev_set_rx_mode(dev); | |
dad9b335 | 3931 | return err; |
4417da66 | 3932 | } |
d1b19dff | 3933 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 3934 | |
1da177e4 LT |
3935 | /** |
3936 | * dev_set_allmulti - update allmulti count on a device | |
3937 | * @dev: device | |
3938 | * @inc: modifier | |
3939 | * | |
3940 | * Add or remove reception of all multicast frames to a device. While the | |
3941 | * count in the device remains above zero the interface remains listening | |
3942 | * to all interfaces. Once it hits zero the device reverts back to normal | |
3943 | * filtering operation. A negative @inc value is used to drop the counter | |
3944 | * when releasing a resource needing all multicasts. | |
dad9b335 | 3945 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
3946 | */ |
3947 | ||
dad9b335 | 3948 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
3949 | { |
3950 | unsigned short old_flags = dev->flags; | |
3951 | ||
24023451 PM |
3952 | ASSERT_RTNL(); |
3953 | ||
1da177e4 | 3954 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
3955 | dev->allmulti += inc; |
3956 | if (dev->allmulti == 0) { | |
3957 | /* | |
3958 | * Avoid overflow. | |
3959 | * If inc causes overflow, untouch allmulti and return error. | |
3960 | */ | |
3961 | if (inc < 0) | |
3962 | dev->flags &= ~IFF_ALLMULTI; | |
3963 | else { | |
3964 | dev->allmulti -= inc; | |
3965 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
3966 | "set allmulti failed, allmulti feature of " | |
3967 | "device might be broken.\n", dev->name); | |
3968 | return -EOVERFLOW; | |
3969 | } | |
3970 | } | |
24023451 | 3971 | if (dev->flags ^ old_flags) { |
b6c40d68 | 3972 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 3973 | dev_set_rx_mode(dev); |
24023451 | 3974 | } |
dad9b335 | 3975 | return 0; |
4417da66 | 3976 | } |
d1b19dff | 3977 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
3978 | |
3979 | /* | |
3980 | * Upload unicast and multicast address lists to device and | |
3981 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 3982 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
3983 | * are present. |
3984 | */ | |
3985 | void __dev_set_rx_mode(struct net_device *dev) | |
3986 | { | |
d314774c SH |
3987 | const struct net_device_ops *ops = dev->netdev_ops; |
3988 | ||
4417da66 PM |
3989 | /* dev_open will call this function so the list will stay sane. */ |
3990 | if (!(dev->flags&IFF_UP)) | |
3991 | return; | |
3992 | ||
3993 | if (!netif_device_present(dev)) | |
40b77c94 | 3994 | return; |
4417da66 | 3995 | |
d314774c SH |
3996 | if (ops->ndo_set_rx_mode) |
3997 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
3998 | else { |
3999 | /* Unicast addresses changes may only happen under the rtnl, | |
4000 | * therefore calling __dev_set_promiscuity here is safe. | |
4001 | */ | |
32e7bfc4 | 4002 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
4417da66 PM |
4003 | __dev_set_promiscuity(dev, 1); |
4004 | dev->uc_promisc = 1; | |
32e7bfc4 | 4005 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
4417da66 PM |
4006 | __dev_set_promiscuity(dev, -1); |
4007 | dev->uc_promisc = 0; | |
4008 | } | |
4009 | ||
d314774c SH |
4010 | if (ops->ndo_set_multicast_list) |
4011 | ops->ndo_set_multicast_list(dev); | |
4417da66 PM |
4012 | } |
4013 | } | |
4014 | ||
4015 | void dev_set_rx_mode(struct net_device *dev) | |
4016 | { | |
b9e40857 | 4017 | netif_addr_lock_bh(dev); |
4417da66 | 4018 | __dev_set_rx_mode(dev); |
b9e40857 | 4019 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
4020 | } |
4021 | ||
f0db275a SH |
4022 | /** |
4023 | * dev_get_flags - get flags reported to userspace | |
4024 | * @dev: device | |
4025 | * | |
4026 | * Get the combination of flag bits exported through APIs to userspace. | |
4027 | */ | |
1da177e4 LT |
4028 | unsigned dev_get_flags(const struct net_device *dev) |
4029 | { | |
4030 | unsigned flags; | |
4031 | ||
4032 | flags = (dev->flags & ~(IFF_PROMISC | | |
4033 | IFF_ALLMULTI | | |
b00055aa SR |
4034 | IFF_RUNNING | |
4035 | IFF_LOWER_UP | | |
4036 | IFF_DORMANT)) | | |
1da177e4 LT |
4037 | (dev->gflags & (IFF_PROMISC | |
4038 | IFF_ALLMULTI)); | |
4039 | ||
b00055aa SR |
4040 | if (netif_running(dev)) { |
4041 | if (netif_oper_up(dev)) | |
4042 | flags |= IFF_RUNNING; | |
4043 | if (netif_carrier_ok(dev)) | |
4044 | flags |= IFF_LOWER_UP; | |
4045 | if (netif_dormant(dev)) | |
4046 | flags |= IFF_DORMANT; | |
4047 | } | |
1da177e4 LT |
4048 | |
4049 | return flags; | |
4050 | } | |
d1b19dff | 4051 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 4052 | |
bd380811 | 4053 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 4054 | { |
1da177e4 | 4055 | int old_flags = dev->flags; |
bd380811 | 4056 | int ret; |
1da177e4 | 4057 | |
24023451 PM |
4058 | ASSERT_RTNL(); |
4059 | ||
1da177e4 LT |
4060 | /* |
4061 | * Set the flags on our device. | |
4062 | */ | |
4063 | ||
4064 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
4065 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
4066 | IFF_AUTOMEDIA)) | | |
4067 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
4068 | IFF_ALLMULTI)); | |
4069 | ||
4070 | /* | |
4071 | * Load in the correct multicast list now the flags have changed. | |
4072 | */ | |
4073 | ||
b6c40d68 PM |
4074 | if ((old_flags ^ flags) & IFF_MULTICAST) |
4075 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 4076 | |
4417da66 | 4077 | dev_set_rx_mode(dev); |
1da177e4 LT |
4078 | |
4079 | /* | |
4080 | * Have we downed the interface. We handle IFF_UP ourselves | |
4081 | * according to user attempts to set it, rather than blindly | |
4082 | * setting it. | |
4083 | */ | |
4084 | ||
4085 | ret = 0; | |
4086 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
bd380811 | 4087 | ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev); |
1da177e4 LT |
4088 | |
4089 | if (!ret) | |
4417da66 | 4090 | dev_set_rx_mode(dev); |
1da177e4 LT |
4091 | } |
4092 | ||
1da177e4 | 4093 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff ED |
4094 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
4095 | ||
1da177e4 LT |
4096 | dev->gflags ^= IFF_PROMISC; |
4097 | dev_set_promiscuity(dev, inc); | |
4098 | } | |
4099 | ||
4100 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
4101 | is important. Some (broken) drivers set IFF_PROMISC, when | |
4102 | IFF_ALLMULTI is requested not asking us and not reporting. | |
4103 | */ | |
4104 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
4105 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
4106 | ||
1da177e4 LT |
4107 | dev->gflags ^= IFF_ALLMULTI; |
4108 | dev_set_allmulti(dev, inc); | |
4109 | } | |
4110 | ||
bd380811 PM |
4111 | return ret; |
4112 | } | |
4113 | ||
4114 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags) | |
4115 | { | |
4116 | unsigned int changes = dev->flags ^ old_flags; | |
4117 | ||
4118 | if (changes & IFF_UP) { | |
4119 | if (dev->flags & IFF_UP) | |
4120 | call_netdevice_notifiers(NETDEV_UP, dev); | |
4121 | else | |
4122 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
4123 | } | |
4124 | ||
4125 | if (dev->flags & IFF_UP && | |
4126 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) | |
4127 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
4128 | } | |
4129 | ||
4130 | /** | |
4131 | * dev_change_flags - change device settings | |
4132 | * @dev: device | |
4133 | * @flags: device state flags | |
4134 | * | |
4135 | * Change settings on device based state flags. The flags are | |
4136 | * in the userspace exported format. | |
4137 | */ | |
4138 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
4139 | { | |
4140 | int ret, changes; | |
4141 | int old_flags = dev->flags; | |
4142 | ||
4143 | ret = __dev_change_flags(dev, flags); | |
4144 | if (ret < 0) | |
4145 | return ret; | |
4146 | ||
4147 | changes = old_flags ^ dev->flags; | |
7c355f53 TG |
4148 | if (changes) |
4149 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 | 4150 | |
bd380811 | 4151 | __dev_notify_flags(dev, old_flags); |
1da177e4 LT |
4152 | return ret; |
4153 | } | |
d1b19dff | 4154 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 4155 | |
f0db275a SH |
4156 | /** |
4157 | * dev_set_mtu - Change maximum transfer unit | |
4158 | * @dev: device | |
4159 | * @new_mtu: new transfer unit | |
4160 | * | |
4161 | * Change the maximum transfer size of the network device. | |
4162 | */ | |
1da177e4 LT |
4163 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
4164 | { | |
d314774c | 4165 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4166 | int err; |
4167 | ||
4168 | if (new_mtu == dev->mtu) | |
4169 | return 0; | |
4170 | ||
4171 | /* MTU must be positive. */ | |
4172 | if (new_mtu < 0) | |
4173 | return -EINVAL; | |
4174 | ||
4175 | if (!netif_device_present(dev)) | |
4176 | return -ENODEV; | |
4177 | ||
4178 | err = 0; | |
d314774c SH |
4179 | if (ops->ndo_change_mtu) |
4180 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
4181 | else |
4182 | dev->mtu = new_mtu; | |
d314774c | 4183 | |
1da177e4 | 4184 | if (!err && dev->flags & IFF_UP) |
056925ab | 4185 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
4186 | return err; |
4187 | } | |
d1b19dff | 4188 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 4189 | |
f0db275a SH |
4190 | /** |
4191 | * dev_set_mac_address - Change Media Access Control Address | |
4192 | * @dev: device | |
4193 | * @sa: new address | |
4194 | * | |
4195 | * Change the hardware (MAC) address of the device | |
4196 | */ | |
1da177e4 LT |
4197 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
4198 | { | |
d314774c | 4199 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
4200 | int err; |
4201 | ||
d314774c | 4202 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
4203 | return -EOPNOTSUPP; |
4204 | if (sa->sa_family != dev->type) | |
4205 | return -EINVAL; | |
4206 | if (!netif_device_present(dev)) | |
4207 | return -ENODEV; | |
d314774c | 4208 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 4209 | if (!err) |
056925ab | 4210 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
4211 | return err; |
4212 | } | |
d1b19dff | 4213 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 LT |
4214 | |
4215 | /* | |
3710becf | 4216 | * Perform the SIOCxIFxxx calls, inside rcu_read_lock() |
1da177e4 | 4217 | */ |
14e3e079 | 4218 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
4219 | { |
4220 | int err; | |
3710becf | 4221 | struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name); |
1da177e4 LT |
4222 | |
4223 | if (!dev) | |
4224 | return -ENODEV; | |
4225 | ||
4226 | switch (cmd) { | |
d1b19dff ED |
4227 | case SIOCGIFFLAGS: /* Get interface flags */ |
4228 | ifr->ifr_flags = (short) dev_get_flags(dev); | |
4229 | return 0; | |
1da177e4 | 4230 | |
d1b19dff ED |
4231 | case SIOCGIFMETRIC: /* Get the metric on the interface |
4232 | (currently unused) */ | |
4233 | ifr->ifr_metric = 0; | |
4234 | return 0; | |
1da177e4 | 4235 | |
d1b19dff ED |
4236 | case SIOCGIFMTU: /* Get the MTU of a device */ |
4237 | ifr->ifr_mtu = dev->mtu; | |
4238 | return 0; | |
1da177e4 | 4239 | |
d1b19dff ED |
4240 | case SIOCGIFHWADDR: |
4241 | if (!dev->addr_len) | |
4242 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
4243 | else | |
4244 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
4245 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4246 | ifr->ifr_hwaddr.sa_family = dev->type; | |
4247 | return 0; | |
1da177e4 | 4248 | |
d1b19dff ED |
4249 | case SIOCGIFSLAVE: |
4250 | err = -EINVAL; | |
4251 | break; | |
14e3e079 | 4252 | |
d1b19dff ED |
4253 | case SIOCGIFMAP: |
4254 | ifr->ifr_map.mem_start = dev->mem_start; | |
4255 | ifr->ifr_map.mem_end = dev->mem_end; | |
4256 | ifr->ifr_map.base_addr = dev->base_addr; | |
4257 | ifr->ifr_map.irq = dev->irq; | |
4258 | ifr->ifr_map.dma = dev->dma; | |
4259 | ifr->ifr_map.port = dev->if_port; | |
4260 | return 0; | |
14e3e079 | 4261 | |
d1b19dff ED |
4262 | case SIOCGIFINDEX: |
4263 | ifr->ifr_ifindex = dev->ifindex; | |
4264 | return 0; | |
14e3e079 | 4265 | |
d1b19dff ED |
4266 | case SIOCGIFTXQLEN: |
4267 | ifr->ifr_qlen = dev->tx_queue_len; | |
4268 | return 0; | |
14e3e079 | 4269 | |
d1b19dff ED |
4270 | default: |
4271 | /* dev_ioctl() should ensure this case | |
4272 | * is never reached | |
4273 | */ | |
4274 | WARN_ON(1); | |
4275 | err = -EINVAL; | |
4276 | break; | |
14e3e079 JG |
4277 | |
4278 | } | |
4279 | return err; | |
4280 | } | |
4281 | ||
4282 | /* | |
4283 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
4284 | */ | |
4285 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
4286 | { | |
4287 | int err; | |
4288 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 4289 | const struct net_device_ops *ops; |
14e3e079 JG |
4290 | |
4291 | if (!dev) | |
4292 | return -ENODEV; | |
4293 | ||
5f2f6da7 JP |
4294 | ops = dev->netdev_ops; |
4295 | ||
14e3e079 | 4296 | switch (cmd) { |
d1b19dff ED |
4297 | case SIOCSIFFLAGS: /* Set interface flags */ |
4298 | return dev_change_flags(dev, ifr->ifr_flags); | |
14e3e079 | 4299 | |
d1b19dff ED |
4300 | case SIOCSIFMETRIC: /* Set the metric on the interface |
4301 | (currently unused) */ | |
4302 | return -EOPNOTSUPP; | |
14e3e079 | 4303 | |
d1b19dff ED |
4304 | case SIOCSIFMTU: /* Set the MTU of a device */ |
4305 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
1da177e4 | 4306 | |
d1b19dff ED |
4307 | case SIOCSIFHWADDR: |
4308 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
1da177e4 | 4309 | |
d1b19dff ED |
4310 | case SIOCSIFHWBROADCAST: |
4311 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
4312 | return -EINVAL; | |
4313 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
4314 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
4315 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
4316 | return 0; | |
1da177e4 | 4317 | |
d1b19dff ED |
4318 | case SIOCSIFMAP: |
4319 | if (ops->ndo_set_config) { | |
1da177e4 LT |
4320 | if (!netif_device_present(dev)) |
4321 | return -ENODEV; | |
d1b19dff ED |
4322 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
4323 | } | |
4324 | return -EOPNOTSUPP; | |
1da177e4 | 4325 | |
d1b19dff ED |
4326 | case SIOCADDMULTI: |
4327 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4328 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4329 | return -EINVAL; | |
4330 | if (!netif_device_present(dev)) | |
4331 | return -ENODEV; | |
22bedad3 | 4332 | return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data); |
d1b19dff ED |
4333 | |
4334 | case SIOCDELMULTI: | |
4335 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || | |
4336 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
4337 | return -EINVAL; | |
4338 | if (!netif_device_present(dev)) | |
4339 | return -ENODEV; | |
22bedad3 | 4340 | return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data); |
1da177e4 | 4341 | |
d1b19dff ED |
4342 | case SIOCSIFTXQLEN: |
4343 | if (ifr->ifr_qlen < 0) | |
4344 | return -EINVAL; | |
4345 | dev->tx_queue_len = ifr->ifr_qlen; | |
4346 | return 0; | |
1da177e4 | 4347 | |
d1b19dff ED |
4348 | case SIOCSIFNAME: |
4349 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4350 | return dev_change_name(dev, ifr->ifr_newname); | |
1da177e4 | 4351 | |
d1b19dff ED |
4352 | /* |
4353 | * Unknown or private ioctl | |
4354 | */ | |
4355 | default: | |
4356 | if ((cmd >= SIOCDEVPRIVATE && | |
4357 | cmd <= SIOCDEVPRIVATE + 15) || | |
4358 | cmd == SIOCBONDENSLAVE || | |
4359 | cmd == SIOCBONDRELEASE || | |
4360 | cmd == SIOCBONDSETHWADDR || | |
4361 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4362 | cmd == SIOCBONDINFOQUERY || | |
4363 | cmd == SIOCBONDCHANGEACTIVE || | |
4364 | cmd == SIOCGMIIPHY || | |
4365 | cmd == SIOCGMIIREG || | |
4366 | cmd == SIOCSMIIREG || | |
4367 | cmd == SIOCBRADDIF || | |
4368 | cmd == SIOCBRDELIF || | |
4369 | cmd == SIOCSHWTSTAMP || | |
4370 | cmd == SIOCWANDEV) { | |
4371 | err = -EOPNOTSUPP; | |
4372 | if (ops->ndo_do_ioctl) { | |
4373 | if (netif_device_present(dev)) | |
4374 | err = ops->ndo_do_ioctl(dev, ifr, cmd); | |
4375 | else | |
4376 | err = -ENODEV; | |
4377 | } | |
4378 | } else | |
4379 | err = -EINVAL; | |
1da177e4 LT |
4380 | |
4381 | } | |
4382 | return err; | |
4383 | } | |
4384 | ||
4385 | /* | |
4386 | * This function handles all "interface"-type I/O control requests. The actual | |
4387 | * 'doing' part of this is dev_ifsioc above. | |
4388 | */ | |
4389 | ||
4390 | /** | |
4391 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4392 | * @net: the applicable net namespace |
1da177e4 LT |
4393 | * @cmd: command to issue |
4394 | * @arg: pointer to a struct ifreq in user space | |
4395 | * | |
4396 | * Issue ioctl functions to devices. This is normally called by the | |
4397 | * user space syscall interfaces but can sometimes be useful for | |
4398 | * other purposes. The return value is the return from the syscall if | |
4399 | * positive or a negative errno code on error. | |
4400 | */ | |
4401 | ||
881d966b | 4402 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4403 | { |
4404 | struct ifreq ifr; | |
4405 | int ret; | |
4406 | char *colon; | |
4407 | ||
4408 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4409 | and requires shared lock, because it sleeps writing | |
4410 | to user space. | |
4411 | */ | |
4412 | ||
4413 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 4414 | rtnl_lock(); |
881d966b | 4415 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 4416 | rtnl_unlock(); |
1da177e4 LT |
4417 | return ret; |
4418 | } | |
4419 | if (cmd == SIOCGIFNAME) | |
881d966b | 4420 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
4421 | |
4422 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4423 | return -EFAULT; | |
4424 | ||
4425 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4426 | ||
4427 | colon = strchr(ifr.ifr_name, ':'); | |
4428 | if (colon) | |
4429 | *colon = 0; | |
4430 | ||
4431 | /* | |
4432 | * See which interface the caller is talking about. | |
4433 | */ | |
4434 | ||
4435 | switch (cmd) { | |
d1b19dff ED |
4436 | /* |
4437 | * These ioctl calls: | |
4438 | * - can be done by all. | |
4439 | * - atomic and do not require locking. | |
4440 | * - return a value | |
4441 | */ | |
4442 | case SIOCGIFFLAGS: | |
4443 | case SIOCGIFMETRIC: | |
4444 | case SIOCGIFMTU: | |
4445 | case SIOCGIFHWADDR: | |
4446 | case SIOCGIFSLAVE: | |
4447 | case SIOCGIFMAP: | |
4448 | case SIOCGIFINDEX: | |
4449 | case SIOCGIFTXQLEN: | |
4450 | dev_load(net, ifr.ifr_name); | |
3710becf | 4451 | rcu_read_lock(); |
d1b19dff | 4452 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
3710becf | 4453 | rcu_read_unlock(); |
d1b19dff ED |
4454 | if (!ret) { |
4455 | if (colon) | |
4456 | *colon = ':'; | |
4457 | if (copy_to_user(arg, &ifr, | |
4458 | sizeof(struct ifreq))) | |
4459 | ret = -EFAULT; | |
4460 | } | |
4461 | return ret; | |
1da177e4 | 4462 | |
d1b19dff ED |
4463 | case SIOCETHTOOL: |
4464 | dev_load(net, ifr.ifr_name); | |
4465 | rtnl_lock(); | |
4466 | ret = dev_ethtool(net, &ifr); | |
4467 | rtnl_unlock(); | |
4468 | if (!ret) { | |
4469 | if (colon) | |
4470 | *colon = ':'; | |
4471 | if (copy_to_user(arg, &ifr, | |
4472 | sizeof(struct ifreq))) | |
4473 | ret = -EFAULT; | |
4474 | } | |
4475 | return ret; | |
1da177e4 | 4476 | |
d1b19dff ED |
4477 | /* |
4478 | * These ioctl calls: | |
4479 | * - require superuser power. | |
4480 | * - require strict serialization. | |
4481 | * - return a value | |
4482 | */ | |
4483 | case SIOCGMIIPHY: | |
4484 | case SIOCGMIIREG: | |
4485 | case SIOCSIFNAME: | |
4486 | if (!capable(CAP_NET_ADMIN)) | |
4487 | return -EPERM; | |
4488 | dev_load(net, ifr.ifr_name); | |
4489 | rtnl_lock(); | |
4490 | ret = dev_ifsioc(net, &ifr, cmd); | |
4491 | rtnl_unlock(); | |
4492 | if (!ret) { | |
4493 | if (colon) | |
4494 | *colon = ':'; | |
4495 | if (copy_to_user(arg, &ifr, | |
4496 | sizeof(struct ifreq))) | |
4497 | ret = -EFAULT; | |
4498 | } | |
4499 | return ret; | |
1da177e4 | 4500 | |
d1b19dff ED |
4501 | /* |
4502 | * These ioctl calls: | |
4503 | * - require superuser power. | |
4504 | * - require strict serialization. | |
4505 | * - do not return a value | |
4506 | */ | |
4507 | case SIOCSIFFLAGS: | |
4508 | case SIOCSIFMETRIC: | |
4509 | case SIOCSIFMTU: | |
4510 | case SIOCSIFMAP: | |
4511 | case SIOCSIFHWADDR: | |
4512 | case SIOCSIFSLAVE: | |
4513 | case SIOCADDMULTI: | |
4514 | case SIOCDELMULTI: | |
4515 | case SIOCSIFHWBROADCAST: | |
4516 | case SIOCSIFTXQLEN: | |
4517 | case SIOCSMIIREG: | |
4518 | case SIOCBONDENSLAVE: | |
4519 | case SIOCBONDRELEASE: | |
4520 | case SIOCBONDSETHWADDR: | |
4521 | case SIOCBONDCHANGEACTIVE: | |
4522 | case SIOCBRADDIF: | |
4523 | case SIOCBRDELIF: | |
4524 | case SIOCSHWTSTAMP: | |
4525 | if (!capable(CAP_NET_ADMIN)) | |
4526 | return -EPERM; | |
4527 | /* fall through */ | |
4528 | case SIOCBONDSLAVEINFOQUERY: | |
4529 | case SIOCBONDINFOQUERY: | |
4530 | dev_load(net, ifr.ifr_name); | |
4531 | rtnl_lock(); | |
4532 | ret = dev_ifsioc(net, &ifr, cmd); | |
4533 | rtnl_unlock(); | |
4534 | return ret; | |
4535 | ||
4536 | case SIOCGIFMEM: | |
4537 | /* Get the per device memory space. We can add this but | |
4538 | * currently do not support it */ | |
4539 | case SIOCSIFMEM: | |
4540 | /* Set the per device memory buffer space. | |
4541 | * Not applicable in our case */ | |
4542 | case SIOCSIFLINK: | |
4543 | return -EINVAL; | |
4544 | ||
4545 | /* | |
4546 | * Unknown or private ioctl. | |
4547 | */ | |
4548 | default: | |
4549 | if (cmd == SIOCWANDEV || | |
4550 | (cmd >= SIOCDEVPRIVATE && | |
4551 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 4552 | dev_load(net, ifr.ifr_name); |
1da177e4 | 4553 | rtnl_lock(); |
881d966b | 4554 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 | 4555 | rtnl_unlock(); |
d1b19dff ED |
4556 | if (!ret && copy_to_user(arg, &ifr, |
4557 | sizeof(struct ifreq))) | |
4558 | ret = -EFAULT; | |
1da177e4 | 4559 | return ret; |
d1b19dff ED |
4560 | } |
4561 | /* Take care of Wireless Extensions */ | |
4562 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
4563 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
4564 | return -EINVAL; | |
1da177e4 LT |
4565 | } |
4566 | } | |
4567 | ||
4568 | ||
4569 | /** | |
4570 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 4571 | * @net: the applicable net namespace |
1da177e4 LT |
4572 | * |
4573 | * Returns a suitable unique value for a new device interface | |
4574 | * number. The caller must hold the rtnl semaphore or the | |
4575 | * dev_base_lock to be sure it remains unique. | |
4576 | */ | |
881d966b | 4577 | static int dev_new_index(struct net *net) |
1da177e4 LT |
4578 | { |
4579 | static int ifindex; | |
4580 | for (;;) { | |
4581 | if (++ifindex <= 0) | |
4582 | ifindex = 1; | |
881d966b | 4583 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
4584 | return ifindex; |
4585 | } | |
4586 | } | |
4587 | ||
1da177e4 | 4588 | /* Delayed registration/unregisteration */ |
3b5b34fd | 4589 | static LIST_HEAD(net_todo_list); |
1da177e4 | 4590 | |
6f05f629 | 4591 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 4592 | { |
1da177e4 | 4593 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
4594 | } |
4595 | ||
9b5e383c | 4596 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 4597 | { |
e93737b0 | 4598 | struct net_device *dev, *tmp; |
9b5e383c | 4599 | |
93ee31f1 DL |
4600 | BUG_ON(dev_boot_phase); |
4601 | ASSERT_RTNL(); | |
4602 | ||
e93737b0 | 4603 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 4604 | /* Some devices call without registering |
e93737b0 KK |
4605 | * for initialization unwind. Remove those |
4606 | * devices and proceed with the remaining. | |
9b5e383c ED |
4607 | */ |
4608 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
4609 | pr_debug("unregister_netdevice: device %s/%p never " | |
4610 | "was registered\n", dev->name, dev); | |
93ee31f1 | 4611 | |
9b5e383c | 4612 | WARN_ON(1); |
e93737b0 KK |
4613 | list_del(&dev->unreg_list); |
4614 | continue; | |
9b5e383c | 4615 | } |
93ee31f1 | 4616 | |
9b5e383c | 4617 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
93ee31f1 | 4618 | |
9b5e383c ED |
4619 | /* If device is running, close it first. */ |
4620 | dev_close(dev); | |
93ee31f1 | 4621 | |
9b5e383c ED |
4622 | /* And unlink it from device chain. */ |
4623 | unlist_netdevice(dev); | |
93ee31f1 | 4624 | |
9b5e383c ED |
4625 | dev->reg_state = NETREG_UNREGISTERING; |
4626 | } | |
93ee31f1 DL |
4627 | |
4628 | synchronize_net(); | |
4629 | ||
9b5e383c ED |
4630 | list_for_each_entry(dev, head, unreg_list) { |
4631 | /* Shutdown queueing discipline. */ | |
4632 | dev_shutdown(dev); | |
93ee31f1 DL |
4633 | |
4634 | ||
9b5e383c ED |
4635 | /* Notify protocols, that we are about to destroy |
4636 | this device. They should clean all the things. | |
4637 | */ | |
4638 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
93ee31f1 | 4639 | |
a2835763 PM |
4640 | if (!dev->rtnl_link_ops || |
4641 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
4642 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U); | |
4643 | ||
9b5e383c ED |
4644 | /* |
4645 | * Flush the unicast and multicast chains | |
4646 | */ | |
a748ee24 | 4647 | dev_uc_flush(dev); |
22bedad3 | 4648 | dev_mc_flush(dev); |
93ee31f1 | 4649 | |
9b5e383c ED |
4650 | if (dev->netdev_ops->ndo_uninit) |
4651 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 4652 | |
9b5e383c ED |
4653 | /* Notifier chain MUST detach us from master device. */ |
4654 | WARN_ON(dev->master); | |
93ee31f1 | 4655 | |
9b5e383c ED |
4656 | /* Remove entries from kobject tree */ |
4657 | netdev_unregister_kobject(dev); | |
4658 | } | |
93ee31f1 | 4659 | |
a5ee1551 | 4660 | /* Process any work delayed until the end of the batch */ |
e5e26d75 | 4661 | dev = list_first_entry(head, struct net_device, unreg_list); |
a5ee1551 | 4662 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
93ee31f1 | 4663 | |
a5ee1551 | 4664 | synchronize_net(); |
395264d5 | 4665 | |
a5ee1551 | 4666 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
4667 | dev_put(dev); |
4668 | } | |
4669 | ||
4670 | static void rollback_registered(struct net_device *dev) | |
4671 | { | |
4672 | LIST_HEAD(single); | |
4673 | ||
4674 | list_add(&dev->unreg_list, &single); | |
4675 | rollback_registered_many(&single); | |
93ee31f1 DL |
4676 | } |
4677 | ||
e8a0464c DM |
4678 | static void __netdev_init_queue_locks_one(struct net_device *dev, |
4679 | struct netdev_queue *dev_queue, | |
4680 | void *_unused) | |
c773e847 DM |
4681 | { |
4682 | spin_lock_init(&dev_queue->_xmit_lock); | |
cf508b12 | 4683 | netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type); |
c773e847 DM |
4684 | dev_queue->xmit_lock_owner = -1; |
4685 | } | |
4686 | ||
4687 | static void netdev_init_queue_locks(struct net_device *dev) | |
4688 | { | |
e8a0464c DM |
4689 | netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL); |
4690 | __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL); | |
c773e847 DM |
4691 | } |
4692 | ||
b63365a2 HX |
4693 | unsigned long netdev_fix_features(unsigned long features, const char *name) |
4694 | { | |
4695 | /* Fix illegal SG+CSUM combinations. */ | |
4696 | if ((features & NETIF_F_SG) && | |
4697 | !(features & NETIF_F_ALL_CSUM)) { | |
4698 | if (name) | |
4699 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " | |
4700 | "checksum feature.\n", name); | |
4701 | features &= ~NETIF_F_SG; | |
4702 | } | |
4703 | ||
4704 | /* TSO requires that SG is present as well. */ | |
4705 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { | |
4706 | if (name) | |
4707 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " | |
4708 | "SG feature.\n", name); | |
4709 | features &= ~NETIF_F_TSO; | |
4710 | } | |
4711 | ||
4712 | if (features & NETIF_F_UFO) { | |
4713 | if (!(features & NETIF_F_GEN_CSUM)) { | |
4714 | if (name) | |
4715 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4716 | "since no NETIF_F_HW_CSUM feature.\n", | |
4717 | name); | |
4718 | features &= ~NETIF_F_UFO; | |
4719 | } | |
4720 | ||
4721 | if (!(features & NETIF_F_SG)) { | |
4722 | if (name) | |
4723 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4724 | "since no NETIF_F_SG feature.\n", name); | |
4725 | features &= ~NETIF_F_UFO; | |
4726 | } | |
4727 | } | |
4728 | ||
4729 | return features; | |
4730 | } | |
4731 | EXPORT_SYMBOL(netdev_fix_features); | |
4732 | ||
fc4a7489 PM |
4733 | /** |
4734 | * netif_stacked_transfer_operstate - transfer operstate | |
4735 | * @rootdev: the root or lower level device to transfer state from | |
4736 | * @dev: the device to transfer operstate to | |
4737 | * | |
4738 | * Transfer operational state from root to device. This is normally | |
4739 | * called when a stacking relationship exists between the root | |
4740 | * device and the device(a leaf device). | |
4741 | */ | |
4742 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
4743 | struct net_device *dev) | |
4744 | { | |
4745 | if (rootdev->operstate == IF_OPER_DORMANT) | |
4746 | netif_dormant_on(dev); | |
4747 | else | |
4748 | netif_dormant_off(dev); | |
4749 | ||
4750 | if (netif_carrier_ok(rootdev)) { | |
4751 | if (!netif_carrier_ok(dev)) | |
4752 | netif_carrier_on(dev); | |
4753 | } else { | |
4754 | if (netif_carrier_ok(dev)) | |
4755 | netif_carrier_off(dev); | |
4756 | } | |
4757 | } | |
4758 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
4759 | ||
1da177e4 LT |
4760 | /** |
4761 | * register_netdevice - register a network device | |
4762 | * @dev: device to register | |
4763 | * | |
4764 | * Take a completed network device structure and add it to the kernel | |
4765 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
4766 | * chain. 0 is returned on success. A negative errno code is returned | |
4767 | * on a failure to set up the device, or if the name is a duplicate. | |
4768 | * | |
4769 | * Callers must hold the rtnl semaphore. You may want | |
4770 | * register_netdev() instead of this. | |
4771 | * | |
4772 | * BUGS: | |
4773 | * The locking appears insufficient to guarantee two parallel registers | |
4774 | * will not get the same name. | |
4775 | */ | |
4776 | ||
4777 | int register_netdevice(struct net_device *dev) | |
4778 | { | |
1da177e4 | 4779 | int ret; |
d314774c | 4780 | struct net *net = dev_net(dev); |
1da177e4 LT |
4781 | |
4782 | BUG_ON(dev_boot_phase); | |
4783 | ASSERT_RTNL(); | |
4784 | ||
b17a7c17 SH |
4785 | might_sleep(); |
4786 | ||
1da177e4 LT |
4787 | /* When net_device's are persistent, this will be fatal. */ |
4788 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 4789 | BUG_ON(!net); |
1da177e4 | 4790 | |
f1f28aa3 | 4791 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 4792 | netdev_set_addr_lockdep_class(dev); |
c773e847 | 4793 | netdev_init_queue_locks(dev); |
1da177e4 | 4794 | |
1da177e4 LT |
4795 | dev->iflink = -1; |
4796 | ||
df334545 | 4797 | #ifdef CONFIG_RPS |
0a9627f2 TH |
4798 | if (!dev->num_rx_queues) { |
4799 | /* | |
4800 | * Allocate a single RX queue if driver never called | |
4801 | * alloc_netdev_mq | |
4802 | */ | |
4803 | ||
4804 | dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL); | |
4805 | if (!dev->_rx) { | |
4806 | ret = -ENOMEM; | |
4807 | goto out; | |
4808 | } | |
4809 | ||
4810 | dev->_rx->first = dev->_rx; | |
4811 | atomic_set(&dev->_rx->count, 1); | |
4812 | dev->num_rx_queues = 1; | |
4813 | } | |
df334545 | 4814 | #endif |
1da177e4 | 4815 | /* Init, if this function is available */ |
d314774c SH |
4816 | if (dev->netdev_ops->ndo_init) { |
4817 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
4818 | if (ret) { |
4819 | if (ret > 0) | |
4820 | ret = -EIO; | |
90833aa4 | 4821 | goto out; |
1da177e4 LT |
4822 | } |
4823 | } | |
4ec93edb | 4824 | |
d9031024 OP |
4825 | ret = dev_get_valid_name(net, dev->name, dev->name, 0); |
4826 | if (ret) | |
7ce1b0ed | 4827 | goto err_uninit; |
1da177e4 | 4828 | |
881d966b | 4829 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
4830 | if (dev->iflink == -1) |
4831 | dev->iflink = dev->ifindex; | |
4832 | ||
d212f87b SH |
4833 | /* Fix illegal checksum combinations */ |
4834 | if ((dev->features & NETIF_F_HW_CSUM) && | |
4835 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
4836 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
4837 | dev->name); | |
4838 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
4839 | } | |
4840 | ||
4841 | if ((dev->features & NETIF_F_NO_CSUM) && | |
4842 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
4843 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
4844 | dev->name); | |
4845 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
4846 | } | |
4847 | ||
b63365a2 | 4848 | dev->features = netdev_fix_features(dev->features, dev->name); |
1da177e4 | 4849 | |
e5a4a72d LB |
4850 | /* Enable software GSO if SG is supported. */ |
4851 | if (dev->features & NETIF_F_SG) | |
4852 | dev->features |= NETIF_F_GSO; | |
4853 | ||
aaf8cdc3 | 4854 | netdev_initialize_kobject(dev); |
7ffbe3fd JB |
4855 | |
4856 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); | |
4857 | ret = notifier_to_errno(ret); | |
4858 | if (ret) | |
4859 | goto err_uninit; | |
4860 | ||
8b41d188 | 4861 | ret = netdev_register_kobject(dev); |
b17a7c17 | 4862 | if (ret) |
7ce1b0ed | 4863 | goto err_uninit; |
b17a7c17 SH |
4864 | dev->reg_state = NETREG_REGISTERED; |
4865 | ||
1da177e4 LT |
4866 | /* |
4867 | * Default initial state at registry is that the | |
4868 | * device is present. | |
4869 | */ | |
4870 | ||
4871 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
4872 | ||
1da177e4 | 4873 | dev_init_scheduler(dev); |
1da177e4 | 4874 | dev_hold(dev); |
ce286d32 | 4875 | list_netdevice(dev); |
1da177e4 LT |
4876 | |
4877 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 4878 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 4879 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
4880 | if (ret) { |
4881 | rollback_registered(dev); | |
4882 | dev->reg_state = NETREG_UNREGISTERED; | |
4883 | } | |
d90a909e EB |
4884 | /* |
4885 | * Prevent userspace races by waiting until the network | |
4886 | * device is fully setup before sending notifications. | |
4887 | */ | |
a2835763 PM |
4888 | if (!dev->rtnl_link_ops || |
4889 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
4890 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
1da177e4 LT |
4891 | |
4892 | out: | |
4893 | return ret; | |
7ce1b0ed HX |
4894 | |
4895 | err_uninit: | |
d314774c SH |
4896 | if (dev->netdev_ops->ndo_uninit) |
4897 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 4898 | goto out; |
1da177e4 | 4899 | } |
d1b19dff | 4900 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 4901 | |
937f1ba5 BH |
4902 | /** |
4903 | * init_dummy_netdev - init a dummy network device for NAPI | |
4904 | * @dev: device to init | |
4905 | * | |
4906 | * This takes a network device structure and initialize the minimum | |
4907 | * amount of fields so it can be used to schedule NAPI polls without | |
4908 | * registering a full blown interface. This is to be used by drivers | |
4909 | * that need to tie several hardware interfaces to a single NAPI | |
4910 | * poll scheduler due to HW limitations. | |
4911 | */ | |
4912 | int init_dummy_netdev(struct net_device *dev) | |
4913 | { | |
4914 | /* Clear everything. Note we don't initialize spinlocks | |
4915 | * are they aren't supposed to be taken by any of the | |
4916 | * NAPI code and this dummy netdev is supposed to be | |
4917 | * only ever used for NAPI polls | |
4918 | */ | |
4919 | memset(dev, 0, sizeof(struct net_device)); | |
4920 | ||
4921 | /* make sure we BUG if trying to hit standard | |
4922 | * register/unregister code path | |
4923 | */ | |
4924 | dev->reg_state = NETREG_DUMMY; | |
4925 | ||
4926 | /* initialize the ref count */ | |
4927 | atomic_set(&dev->refcnt, 1); | |
4928 | ||
4929 | /* NAPI wants this */ | |
4930 | INIT_LIST_HEAD(&dev->napi_list); | |
4931 | ||
4932 | /* a dummy interface is started by default */ | |
4933 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
4934 | set_bit(__LINK_STATE_START, &dev->state); | |
4935 | ||
4936 | return 0; | |
4937 | } | |
4938 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
4939 | ||
4940 | ||
1da177e4 LT |
4941 | /** |
4942 | * register_netdev - register a network device | |
4943 | * @dev: device to register | |
4944 | * | |
4945 | * Take a completed network device structure and add it to the kernel | |
4946 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
4947 | * chain. 0 is returned on success. A negative errno code is returned | |
4948 | * on a failure to set up the device, or if the name is a duplicate. | |
4949 | * | |
38b4da38 | 4950 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
4951 | * and expands the device name if you passed a format string to |
4952 | * alloc_netdev. | |
4953 | */ | |
4954 | int register_netdev(struct net_device *dev) | |
4955 | { | |
4956 | int err; | |
4957 | ||
4958 | rtnl_lock(); | |
4959 | ||
4960 | /* | |
4961 | * If the name is a format string the caller wants us to do a | |
4962 | * name allocation. | |
4963 | */ | |
4964 | if (strchr(dev->name, '%')) { | |
4965 | err = dev_alloc_name(dev, dev->name); | |
4966 | if (err < 0) | |
4967 | goto out; | |
4968 | } | |
4ec93edb | 4969 | |
1da177e4 LT |
4970 | err = register_netdevice(dev); |
4971 | out: | |
4972 | rtnl_unlock(); | |
4973 | return err; | |
4974 | } | |
4975 | EXPORT_SYMBOL(register_netdev); | |
4976 | ||
4977 | /* | |
4978 | * netdev_wait_allrefs - wait until all references are gone. | |
4979 | * | |
4980 | * This is called when unregistering network devices. | |
4981 | * | |
4982 | * Any protocol or device that holds a reference should register | |
4983 | * for netdevice notification, and cleanup and put back the | |
4984 | * reference if they receive an UNREGISTER event. | |
4985 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 4986 | * call dev_put. |
1da177e4 LT |
4987 | */ |
4988 | static void netdev_wait_allrefs(struct net_device *dev) | |
4989 | { | |
4990 | unsigned long rebroadcast_time, warning_time; | |
4991 | ||
e014debe ED |
4992 | linkwatch_forget_dev(dev); |
4993 | ||
1da177e4 LT |
4994 | rebroadcast_time = warning_time = jiffies; |
4995 | while (atomic_read(&dev->refcnt) != 0) { | |
4996 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
6756ae4b | 4997 | rtnl_lock(); |
1da177e4 LT |
4998 | |
4999 | /* Rebroadcast unregister notification */ | |
056925ab | 5000 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
a5ee1551 | 5001 | /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users |
395264d5 | 5002 | * should have already handle it the first time */ |
1da177e4 LT |
5003 | |
5004 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
5005 | &dev->state)) { | |
5006 | /* We must not have linkwatch events | |
5007 | * pending on unregister. If this | |
5008 | * happens, we simply run the queue | |
5009 | * unscheduled, resulting in a noop | |
5010 | * for this device. | |
5011 | */ | |
5012 | linkwatch_run_queue(); | |
5013 | } | |
5014 | ||
6756ae4b | 5015 | __rtnl_unlock(); |
1da177e4 LT |
5016 | |
5017 | rebroadcast_time = jiffies; | |
5018 | } | |
5019 | ||
5020 | msleep(250); | |
5021 | ||
5022 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
5023 | printk(KERN_EMERG "unregister_netdevice: " | |
5024 | "waiting for %s to become free. Usage " | |
5025 | "count = %d\n", | |
5026 | dev->name, atomic_read(&dev->refcnt)); | |
5027 | warning_time = jiffies; | |
5028 | } | |
5029 | } | |
5030 | } | |
5031 | ||
5032 | /* The sequence is: | |
5033 | * | |
5034 | * rtnl_lock(); | |
5035 | * ... | |
5036 | * register_netdevice(x1); | |
5037 | * register_netdevice(x2); | |
5038 | * ... | |
5039 | * unregister_netdevice(y1); | |
5040 | * unregister_netdevice(y2); | |
5041 | * ... | |
5042 | * rtnl_unlock(); | |
5043 | * free_netdev(y1); | |
5044 | * free_netdev(y2); | |
5045 | * | |
58ec3b4d | 5046 | * We are invoked by rtnl_unlock(). |
1da177e4 | 5047 | * This allows us to deal with problems: |
b17a7c17 | 5048 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
5049 | * without deadlocking with linkwatch via keventd. |
5050 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
5051 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
5052 | * |
5053 | * We must not return until all unregister events added during | |
5054 | * the interval the lock was held have been completed. | |
1da177e4 | 5055 | */ |
1da177e4 LT |
5056 | void netdev_run_todo(void) |
5057 | { | |
626ab0e6 | 5058 | struct list_head list; |
1da177e4 | 5059 | |
1da177e4 | 5060 | /* Snapshot list, allow later requests */ |
626ab0e6 | 5061 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
5062 | |
5063 | __rtnl_unlock(); | |
626ab0e6 | 5064 | |
1da177e4 LT |
5065 | while (!list_empty(&list)) { |
5066 | struct net_device *dev | |
e5e26d75 | 5067 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
5068 | list_del(&dev->todo_list); |
5069 | ||
b17a7c17 SH |
5070 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
5071 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
5072 | dev->name, dev->reg_state); | |
5073 | dump_stack(); | |
5074 | continue; | |
5075 | } | |
1da177e4 | 5076 | |
b17a7c17 | 5077 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 5078 | |
152102c7 | 5079 | on_each_cpu(flush_backlog, dev, 1); |
6e583ce5 | 5080 | |
b17a7c17 | 5081 | netdev_wait_allrefs(dev); |
1da177e4 | 5082 | |
b17a7c17 SH |
5083 | /* paranoia */ |
5084 | BUG_ON(atomic_read(&dev->refcnt)); | |
547b792c IJ |
5085 | WARN_ON(dev->ip_ptr); |
5086 | WARN_ON(dev->ip6_ptr); | |
5087 | WARN_ON(dev->dn_ptr); | |
1da177e4 | 5088 | |
b17a7c17 SH |
5089 | if (dev->destructor) |
5090 | dev->destructor(dev); | |
9093bbb2 SH |
5091 | |
5092 | /* Free network device */ | |
5093 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 5094 | } |
1da177e4 LT |
5095 | } |
5096 | ||
d83345ad ED |
5097 | /** |
5098 | * dev_txq_stats_fold - fold tx_queues stats | |
5099 | * @dev: device to get statistics from | |
5100 | * @stats: struct net_device_stats to hold results | |
5101 | */ | |
5102 | void dev_txq_stats_fold(const struct net_device *dev, | |
5103 | struct net_device_stats *stats) | |
5104 | { | |
5105 | unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0; | |
5106 | unsigned int i; | |
5107 | struct netdev_queue *txq; | |
5108 | ||
5109 | for (i = 0; i < dev->num_tx_queues; i++) { | |
5110 | txq = netdev_get_tx_queue(dev, i); | |
5111 | tx_bytes += txq->tx_bytes; | |
5112 | tx_packets += txq->tx_packets; | |
5113 | tx_dropped += txq->tx_dropped; | |
5114 | } | |
5115 | if (tx_bytes || tx_packets || tx_dropped) { | |
5116 | stats->tx_bytes = tx_bytes; | |
5117 | stats->tx_packets = tx_packets; | |
5118 | stats->tx_dropped = tx_dropped; | |
5119 | } | |
5120 | } | |
5121 | EXPORT_SYMBOL(dev_txq_stats_fold); | |
5122 | ||
eeda3fd6 SH |
5123 | /** |
5124 | * dev_get_stats - get network device statistics | |
5125 | * @dev: device to get statistics from | |
5126 | * | |
5127 | * Get network statistics from device. The device driver may provide | |
5128 | * its own method by setting dev->netdev_ops->get_stats; otherwise | |
5129 | * the internal statistics structure is used. | |
5130 | */ | |
5131 | const struct net_device_stats *dev_get_stats(struct net_device *dev) | |
7004bf25 | 5132 | { |
eeda3fd6 SH |
5133 | const struct net_device_ops *ops = dev->netdev_ops; |
5134 | ||
5135 | if (ops->ndo_get_stats) | |
5136 | return ops->ndo_get_stats(dev); | |
d83345ad ED |
5137 | |
5138 | dev_txq_stats_fold(dev, &dev->stats); | |
5139 | return &dev->stats; | |
c45d286e | 5140 | } |
eeda3fd6 | 5141 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 5142 | |
dc2b4847 | 5143 | static void netdev_init_one_queue(struct net_device *dev, |
e8a0464c DM |
5144 | struct netdev_queue *queue, |
5145 | void *_unused) | |
dc2b4847 | 5146 | { |
dc2b4847 DM |
5147 | queue->dev = dev; |
5148 | } | |
5149 | ||
bb949fbd DM |
5150 | static void netdev_init_queues(struct net_device *dev) |
5151 | { | |
e8a0464c DM |
5152 | netdev_init_one_queue(dev, &dev->rx_queue, NULL); |
5153 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); | |
c3f26a26 | 5154 | spin_lock_init(&dev->tx_global_lock); |
bb949fbd DM |
5155 | } |
5156 | ||
1da177e4 | 5157 | /** |
f25f4e44 | 5158 | * alloc_netdev_mq - allocate network device |
1da177e4 LT |
5159 | * @sizeof_priv: size of private data to allocate space for |
5160 | * @name: device name format string | |
5161 | * @setup: callback to initialize device | |
f25f4e44 | 5162 | * @queue_count: the number of subqueues to allocate |
1da177e4 LT |
5163 | * |
5164 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 PWJ |
5165 | * and performs basic initialization. Also allocates subquue structs |
5166 | * for each queue on the device at the end of the netdevice. | |
1da177e4 | 5167 | */ |
f25f4e44 PWJ |
5168 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, |
5169 | void (*setup)(struct net_device *), unsigned int queue_count) | |
1da177e4 | 5170 | { |
e8a0464c | 5171 | struct netdev_queue *tx; |
1da177e4 | 5172 | struct net_device *dev; |
7943986c | 5173 | size_t alloc_size; |
1ce8e7b5 | 5174 | struct net_device *p; |
df334545 ED |
5175 | #ifdef CONFIG_RPS |
5176 | struct netdev_rx_queue *rx; | |
0a9627f2 | 5177 | int i; |
df334545 | 5178 | #endif |
1da177e4 | 5179 | |
b6fe17d6 SH |
5180 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
5181 | ||
fd2ea0a7 | 5182 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
5183 | if (sizeof_priv) { |
5184 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 5185 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
5186 | alloc_size += sizeof_priv; |
5187 | } | |
5188 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 5189 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 5190 | |
31380de9 | 5191 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 5192 | if (!p) { |
b6fe17d6 | 5193 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
5194 | return NULL; |
5195 | } | |
1da177e4 | 5196 | |
7943986c | 5197 | tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL); |
e8a0464c DM |
5198 | if (!tx) { |
5199 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5200 | "tx qdiscs.\n"); | |
ab9c73cc | 5201 | goto free_p; |
e8a0464c DM |
5202 | } |
5203 | ||
df334545 | 5204 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5205 | rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL); |
5206 | if (!rx) { | |
5207 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
5208 | "rx queues.\n"); | |
5209 | goto free_tx; | |
5210 | } | |
5211 | ||
5212 | atomic_set(&rx->count, queue_count); | |
5213 | ||
5214 | /* | |
5215 | * Set a pointer to first element in the array which holds the | |
5216 | * reference count. | |
5217 | */ | |
5218 | for (i = 0; i < queue_count; i++) | |
5219 | rx[i].first = rx; | |
df334545 | 5220 | #endif |
0a9627f2 | 5221 | |
1ce8e7b5 | 5222 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 5223 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc JP |
5224 | |
5225 | if (dev_addr_init(dev)) | |
0a9627f2 | 5226 | goto free_rx; |
ab9c73cc | 5227 | |
22bedad3 | 5228 | dev_mc_init(dev); |
a748ee24 | 5229 | dev_uc_init(dev); |
ccffad25 | 5230 | |
c346dca1 | 5231 | dev_net_set(dev, &init_net); |
1da177e4 | 5232 | |
e8a0464c DM |
5233 | dev->_tx = tx; |
5234 | dev->num_tx_queues = queue_count; | |
fd2ea0a7 | 5235 | dev->real_num_tx_queues = queue_count; |
e8a0464c | 5236 | |
df334545 | 5237 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5238 | dev->_rx = rx; |
5239 | dev->num_rx_queues = queue_count; | |
df334545 | 5240 | #endif |
0a9627f2 | 5241 | |
82cc1a7a | 5242 | dev->gso_max_size = GSO_MAX_SIZE; |
1da177e4 | 5243 | |
bb949fbd DM |
5244 | netdev_init_queues(dev); |
5245 | ||
15682bc4 PWJ |
5246 | INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list); |
5247 | dev->ethtool_ntuple_list.count = 0; | |
d565b0a1 | 5248 | INIT_LIST_HEAD(&dev->napi_list); |
9fdce099 | 5249 | INIT_LIST_HEAD(&dev->unreg_list); |
e014debe | 5250 | INIT_LIST_HEAD(&dev->link_watch_list); |
93f154b5 | 5251 | dev->priv_flags = IFF_XMIT_DST_RELEASE; |
1da177e4 LT |
5252 | setup(dev); |
5253 | strcpy(dev->name, name); | |
5254 | return dev; | |
ab9c73cc | 5255 | |
0a9627f2 | 5256 | free_rx: |
df334545 | 5257 | #ifdef CONFIG_RPS |
0a9627f2 | 5258 | kfree(rx); |
ab9c73cc | 5259 | free_tx: |
df334545 | 5260 | #endif |
ab9c73cc | 5261 | kfree(tx); |
ab9c73cc JP |
5262 | free_p: |
5263 | kfree(p); | |
5264 | return NULL; | |
1da177e4 | 5265 | } |
f25f4e44 | 5266 | EXPORT_SYMBOL(alloc_netdev_mq); |
1da177e4 LT |
5267 | |
5268 | /** | |
5269 | * free_netdev - free network device | |
5270 | * @dev: device | |
5271 | * | |
4ec93edb YH |
5272 | * This function does the last stage of destroying an allocated device |
5273 | * interface. The reference to the device object is released. | |
1da177e4 LT |
5274 | * If this is the last reference then it will be freed. |
5275 | */ | |
5276 | void free_netdev(struct net_device *dev) | |
5277 | { | |
d565b0a1 HX |
5278 | struct napi_struct *p, *n; |
5279 | ||
f3005d7f DL |
5280 | release_net(dev_net(dev)); |
5281 | ||
e8a0464c DM |
5282 | kfree(dev->_tx); |
5283 | ||
f001fde5 JP |
5284 | /* Flush device addresses */ |
5285 | dev_addr_flush(dev); | |
5286 | ||
15682bc4 PWJ |
5287 | /* Clear ethtool n-tuple list */ |
5288 | ethtool_ntuple_flush(dev); | |
5289 | ||
d565b0a1 HX |
5290 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
5291 | netif_napi_del(p); | |
5292 | ||
3041a069 | 5293 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
5294 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
5295 | kfree((char *)dev - dev->padded); | |
5296 | return; | |
5297 | } | |
5298 | ||
5299 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
5300 | dev->reg_state = NETREG_RELEASED; | |
5301 | ||
43cb76d9 GKH |
5302 | /* will free via device release */ |
5303 | put_device(&dev->dev); | |
1da177e4 | 5304 | } |
d1b19dff | 5305 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 5306 | |
f0db275a SH |
5307 | /** |
5308 | * synchronize_net - Synchronize with packet receive processing | |
5309 | * | |
5310 | * Wait for packets currently being received to be done. | |
5311 | * Does not block later packets from starting. | |
5312 | */ | |
4ec93edb | 5313 | void synchronize_net(void) |
1da177e4 LT |
5314 | { |
5315 | might_sleep(); | |
fbd568a3 | 5316 | synchronize_rcu(); |
1da177e4 | 5317 | } |
d1b19dff | 5318 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
5319 | |
5320 | /** | |
44a0873d | 5321 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 5322 | * @dev: device |
44a0873d | 5323 | * @head: list |
6ebfbc06 | 5324 | * |
1da177e4 | 5325 | * This function shuts down a device interface and removes it |
d59b54b1 | 5326 | * from the kernel tables. |
44a0873d | 5327 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
5328 | * |
5329 | * Callers must hold the rtnl semaphore. You may want | |
5330 | * unregister_netdev() instead of this. | |
5331 | */ | |
5332 | ||
44a0873d | 5333 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 5334 | { |
a6620712 HX |
5335 | ASSERT_RTNL(); |
5336 | ||
44a0873d | 5337 | if (head) { |
9fdce099 | 5338 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
5339 | } else { |
5340 | rollback_registered(dev); | |
5341 | /* Finish processing unregister after unlock */ | |
5342 | net_set_todo(dev); | |
5343 | } | |
1da177e4 | 5344 | } |
44a0873d | 5345 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 5346 | |
9b5e383c ED |
5347 | /** |
5348 | * unregister_netdevice_many - unregister many devices | |
5349 | * @head: list of devices | |
9b5e383c ED |
5350 | */ |
5351 | void unregister_netdevice_many(struct list_head *head) | |
5352 | { | |
5353 | struct net_device *dev; | |
5354 | ||
5355 | if (!list_empty(head)) { | |
5356 | rollback_registered_many(head); | |
5357 | list_for_each_entry(dev, head, unreg_list) | |
5358 | net_set_todo(dev); | |
5359 | } | |
5360 | } | |
63c8099d | 5361 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 5362 | |
1da177e4 LT |
5363 | /** |
5364 | * unregister_netdev - remove device from the kernel | |
5365 | * @dev: device | |
5366 | * | |
5367 | * This function shuts down a device interface and removes it | |
d59b54b1 | 5368 | * from the kernel tables. |
1da177e4 LT |
5369 | * |
5370 | * This is just a wrapper for unregister_netdevice that takes | |
5371 | * the rtnl semaphore. In general you want to use this and not | |
5372 | * unregister_netdevice. | |
5373 | */ | |
5374 | void unregister_netdev(struct net_device *dev) | |
5375 | { | |
5376 | rtnl_lock(); | |
5377 | unregister_netdevice(dev); | |
5378 | rtnl_unlock(); | |
5379 | } | |
1da177e4 LT |
5380 | EXPORT_SYMBOL(unregister_netdev); |
5381 | ||
ce286d32 EB |
5382 | /** |
5383 | * dev_change_net_namespace - move device to different nethost namespace | |
5384 | * @dev: device | |
5385 | * @net: network namespace | |
5386 | * @pat: If not NULL name pattern to try if the current device name | |
5387 | * is already taken in the destination network namespace. | |
5388 | * | |
5389 | * This function shuts down a device interface and moves it | |
5390 | * to a new network namespace. On success 0 is returned, on | |
5391 | * a failure a netagive errno code is returned. | |
5392 | * | |
5393 | * Callers must hold the rtnl semaphore. | |
5394 | */ | |
5395 | ||
5396 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
5397 | { | |
ce286d32 EB |
5398 | int err; |
5399 | ||
5400 | ASSERT_RTNL(); | |
5401 | ||
5402 | /* Don't allow namespace local devices to be moved. */ | |
5403 | err = -EINVAL; | |
5404 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5405 | goto out; | |
5406 | ||
3891845e EB |
5407 | #ifdef CONFIG_SYSFS |
5408 | /* Don't allow real devices to be moved when sysfs | |
5409 | * is enabled. | |
5410 | */ | |
5411 | err = -EINVAL; | |
5412 | if (dev->dev.parent) | |
5413 | goto out; | |
5414 | #endif | |
5415 | ||
ce286d32 EB |
5416 | /* Ensure the device has been registrered */ |
5417 | err = -EINVAL; | |
5418 | if (dev->reg_state != NETREG_REGISTERED) | |
5419 | goto out; | |
5420 | ||
5421 | /* Get out if there is nothing todo */ | |
5422 | err = 0; | |
878628fb | 5423 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
5424 | goto out; |
5425 | ||
5426 | /* Pick the destination device name, and ensure | |
5427 | * we can use it in the destination network namespace. | |
5428 | */ | |
5429 | err = -EEXIST; | |
d9031024 | 5430 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
5431 | /* We get here if we can't use the current device name */ |
5432 | if (!pat) | |
5433 | goto out; | |
d9031024 | 5434 | if (dev_get_valid_name(net, pat, dev->name, 1)) |
ce286d32 EB |
5435 | goto out; |
5436 | } | |
5437 | ||
5438 | /* | |
5439 | * And now a mini version of register_netdevice unregister_netdevice. | |
5440 | */ | |
5441 | ||
5442 | /* If device is running close it first. */ | |
9b772652 | 5443 | dev_close(dev); |
ce286d32 EB |
5444 | |
5445 | /* And unlink it from device chain */ | |
5446 | err = -ENODEV; | |
5447 | unlist_netdevice(dev); | |
5448 | ||
5449 | synchronize_net(); | |
5450 | ||
5451 | /* Shutdown queueing discipline. */ | |
5452 | dev_shutdown(dev); | |
5453 | ||
5454 | /* Notify protocols, that we are about to destroy | |
5455 | this device. They should clean all the things. | |
5456 | */ | |
5457 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
a5ee1551 | 5458 | call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev); |
ce286d32 EB |
5459 | |
5460 | /* | |
5461 | * Flush the unicast and multicast chains | |
5462 | */ | |
a748ee24 | 5463 | dev_uc_flush(dev); |
22bedad3 | 5464 | dev_mc_flush(dev); |
ce286d32 | 5465 | |
3891845e EB |
5466 | netdev_unregister_kobject(dev); |
5467 | ||
ce286d32 | 5468 | /* Actually switch the network namespace */ |
c346dca1 | 5469 | dev_net_set(dev, net); |
ce286d32 | 5470 | |
ce286d32 EB |
5471 | /* If there is an ifindex conflict assign a new one */ |
5472 | if (__dev_get_by_index(net, dev->ifindex)) { | |
5473 | int iflink = (dev->iflink == dev->ifindex); | |
5474 | dev->ifindex = dev_new_index(net); | |
5475 | if (iflink) | |
5476 | dev->iflink = dev->ifindex; | |
5477 | } | |
5478 | ||
8b41d188 | 5479 | /* Fixup kobjects */ |
aaf8cdc3 | 5480 | err = netdev_register_kobject(dev); |
8b41d188 | 5481 | WARN_ON(err); |
ce286d32 EB |
5482 | |
5483 | /* Add the device back in the hashes */ | |
5484 | list_netdevice(dev); | |
5485 | ||
5486 | /* Notify protocols, that a new device appeared. */ | |
5487 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5488 | ||
d90a909e EB |
5489 | /* |
5490 | * Prevent userspace races by waiting until the network | |
5491 | * device is fully setup before sending notifications. | |
5492 | */ | |
5493 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U); | |
5494 | ||
ce286d32 EB |
5495 | synchronize_net(); |
5496 | err = 0; | |
5497 | out: | |
5498 | return err; | |
5499 | } | |
463d0183 | 5500 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 5501 | |
1da177e4 LT |
5502 | static int dev_cpu_callback(struct notifier_block *nfb, |
5503 | unsigned long action, | |
5504 | void *ocpu) | |
5505 | { | |
5506 | struct sk_buff **list_skb; | |
37437bb2 | 5507 | struct Qdisc **list_net; |
1da177e4 LT |
5508 | struct sk_buff *skb; |
5509 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
5510 | struct softnet_data *sd, *oldsd; | |
5511 | ||
8bb78442 | 5512 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
5513 | return NOTIFY_OK; |
5514 | ||
5515 | local_irq_disable(); | |
5516 | cpu = smp_processor_id(); | |
5517 | sd = &per_cpu(softnet_data, cpu); | |
5518 | oldsd = &per_cpu(softnet_data, oldcpu); | |
5519 | ||
5520 | /* Find end of our completion_queue. */ | |
5521 | list_skb = &sd->completion_queue; | |
5522 | while (*list_skb) | |
5523 | list_skb = &(*list_skb)->next; | |
5524 | /* Append completion queue from offline CPU. */ | |
5525 | *list_skb = oldsd->completion_queue; | |
5526 | oldsd->completion_queue = NULL; | |
5527 | ||
5528 | /* Find end of our output_queue. */ | |
5529 | list_net = &sd->output_queue; | |
5530 | while (*list_net) | |
5531 | list_net = &(*list_net)->next_sched; | |
5532 | /* Append output queue from offline CPU. */ | |
5533 | *list_net = oldsd->output_queue; | |
5534 | oldsd->output_queue = NULL; | |
5535 | ||
5536 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
5537 | local_irq_enable(); | |
5538 | ||
5539 | /* Process offline CPU's input_pkt_queue */ | |
5540 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) | |
5541 | netif_rx(skb); | |
5542 | ||
5543 | return NOTIFY_OK; | |
5544 | } | |
1da177e4 LT |
5545 | |
5546 | ||
7f353bf2 | 5547 | /** |
b63365a2 HX |
5548 | * netdev_increment_features - increment feature set by one |
5549 | * @all: current feature set | |
5550 | * @one: new feature set | |
5551 | * @mask: mask feature set | |
7f353bf2 HX |
5552 | * |
5553 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
5554 | * @one to the master device with current feature set @all. Will not |
5555 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 5556 | */ |
b63365a2 HX |
5557 | unsigned long netdev_increment_features(unsigned long all, unsigned long one, |
5558 | unsigned long mask) | |
5559 | { | |
5560 | /* If device needs checksumming, downgrade to it. */ | |
d1b19dff | 5561 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) |
b63365a2 HX |
5562 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); |
5563 | else if (mask & NETIF_F_ALL_CSUM) { | |
5564 | /* If one device supports v4/v6 checksumming, set for all. */ | |
5565 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && | |
5566 | !(all & NETIF_F_GEN_CSUM)) { | |
5567 | all &= ~NETIF_F_ALL_CSUM; | |
5568 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
5569 | } | |
e2a6b852 | 5570 | |
b63365a2 HX |
5571 | /* If one device supports hw checksumming, set for all. */ |
5572 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { | |
5573 | all &= ~NETIF_F_ALL_CSUM; | |
5574 | all |= NETIF_F_HW_CSUM; | |
5575 | } | |
5576 | } | |
7f353bf2 | 5577 | |
b63365a2 | 5578 | one |= NETIF_F_ALL_CSUM; |
7f353bf2 | 5579 | |
b63365a2 | 5580 | one |= all & NETIF_F_ONE_FOR_ALL; |
d9f5950f | 5581 | all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO; |
b63365a2 | 5582 | all |= one & mask & NETIF_F_ONE_FOR_ALL; |
7f353bf2 HX |
5583 | |
5584 | return all; | |
5585 | } | |
b63365a2 | 5586 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 5587 | |
30d97d35 PE |
5588 | static struct hlist_head *netdev_create_hash(void) |
5589 | { | |
5590 | int i; | |
5591 | struct hlist_head *hash; | |
5592 | ||
5593 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
5594 | if (hash != NULL) | |
5595 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
5596 | INIT_HLIST_HEAD(&hash[i]); | |
5597 | ||
5598 | return hash; | |
5599 | } | |
5600 | ||
881d966b | 5601 | /* Initialize per network namespace state */ |
4665079c | 5602 | static int __net_init netdev_init(struct net *net) |
881d966b | 5603 | { |
881d966b | 5604 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 5605 | |
30d97d35 PE |
5606 | net->dev_name_head = netdev_create_hash(); |
5607 | if (net->dev_name_head == NULL) | |
5608 | goto err_name; | |
881d966b | 5609 | |
30d97d35 PE |
5610 | net->dev_index_head = netdev_create_hash(); |
5611 | if (net->dev_index_head == NULL) | |
5612 | goto err_idx; | |
881d966b EB |
5613 | |
5614 | return 0; | |
30d97d35 PE |
5615 | |
5616 | err_idx: | |
5617 | kfree(net->dev_name_head); | |
5618 | err_name: | |
5619 | return -ENOMEM; | |
881d966b EB |
5620 | } |
5621 | ||
f0db275a SH |
5622 | /** |
5623 | * netdev_drivername - network driver for the device | |
5624 | * @dev: network device | |
5625 | * @buffer: buffer for resulting name | |
5626 | * @len: size of buffer | |
5627 | * | |
5628 | * Determine network driver for device. | |
5629 | */ | |
cf04a4c7 | 5630 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) |
6579e57b | 5631 | { |
cf04a4c7 SH |
5632 | const struct device_driver *driver; |
5633 | const struct device *parent; | |
6579e57b AV |
5634 | |
5635 | if (len <= 0 || !buffer) | |
5636 | return buffer; | |
5637 | buffer[0] = 0; | |
5638 | ||
5639 | parent = dev->dev.parent; | |
5640 | ||
5641 | if (!parent) | |
5642 | return buffer; | |
5643 | ||
5644 | driver = parent->driver; | |
5645 | if (driver && driver->name) | |
5646 | strlcpy(buffer, driver->name, len); | |
5647 | return buffer; | |
5648 | } | |
5649 | ||
4665079c | 5650 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
5651 | { |
5652 | kfree(net->dev_name_head); | |
5653 | kfree(net->dev_index_head); | |
5654 | } | |
5655 | ||
022cbae6 | 5656 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
5657 | .init = netdev_init, |
5658 | .exit = netdev_exit, | |
5659 | }; | |
5660 | ||
4665079c | 5661 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 5662 | { |
e008b5fc | 5663 | struct net_device *dev, *aux; |
ce286d32 | 5664 | /* |
e008b5fc | 5665 | * Push all migratable network devices back to the |
ce286d32 EB |
5666 | * initial network namespace |
5667 | */ | |
5668 | rtnl_lock(); | |
e008b5fc | 5669 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 5670 | int err; |
aca51397 | 5671 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
5672 | |
5673 | /* Ignore unmoveable devices (i.e. loopback) */ | |
5674 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5675 | continue; | |
5676 | ||
e008b5fc EB |
5677 | /* Leave virtual devices for the generic cleanup */ |
5678 | if (dev->rtnl_link_ops) | |
5679 | continue; | |
d0c082ce | 5680 | |
ce286d32 | 5681 | /* Push remaing network devices to init_net */ |
aca51397 PE |
5682 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
5683 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 5684 | if (err) { |
aca51397 | 5685 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 5686 | __func__, dev->name, err); |
aca51397 | 5687 | BUG(); |
ce286d32 EB |
5688 | } |
5689 | } | |
5690 | rtnl_unlock(); | |
5691 | } | |
5692 | ||
04dc7f6b EB |
5693 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
5694 | { | |
5695 | /* At exit all network devices most be removed from a network | |
5696 | * namespace. Do this in the reverse order of registeration. | |
5697 | * Do this across as many network namespaces as possible to | |
5698 | * improve batching efficiency. | |
5699 | */ | |
5700 | struct net_device *dev; | |
5701 | struct net *net; | |
5702 | LIST_HEAD(dev_kill_list); | |
5703 | ||
5704 | rtnl_lock(); | |
5705 | list_for_each_entry(net, net_list, exit_list) { | |
5706 | for_each_netdev_reverse(net, dev) { | |
5707 | if (dev->rtnl_link_ops) | |
5708 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); | |
5709 | else | |
5710 | unregister_netdevice_queue(dev, &dev_kill_list); | |
5711 | } | |
5712 | } | |
5713 | unregister_netdevice_many(&dev_kill_list); | |
5714 | rtnl_unlock(); | |
5715 | } | |
5716 | ||
022cbae6 | 5717 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 5718 | .exit = default_device_exit, |
04dc7f6b | 5719 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
5720 | }; |
5721 | ||
1da177e4 LT |
5722 | /* |
5723 | * Initialize the DEV module. At boot time this walks the device list and | |
5724 | * unhooks any devices that fail to initialise (normally hardware not | |
5725 | * present) and leaves us with a valid list of present and active devices. | |
5726 | * | |
5727 | */ | |
5728 | ||
5729 | /* | |
5730 | * This is called single threaded during boot, so no need | |
5731 | * to take the rtnl semaphore. | |
5732 | */ | |
5733 | static int __init net_dev_init(void) | |
5734 | { | |
5735 | int i, rc = -ENOMEM; | |
5736 | ||
5737 | BUG_ON(!dev_boot_phase); | |
5738 | ||
1da177e4 LT |
5739 | if (dev_proc_init()) |
5740 | goto out; | |
5741 | ||
8b41d188 | 5742 | if (netdev_kobject_init()) |
1da177e4 LT |
5743 | goto out; |
5744 | ||
5745 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 5746 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
5747 | INIT_LIST_HEAD(&ptype_base[i]); |
5748 | ||
881d966b EB |
5749 | if (register_pernet_subsys(&netdev_net_ops)) |
5750 | goto out; | |
1da177e4 LT |
5751 | |
5752 | /* | |
5753 | * Initialise the packet receive queues. | |
5754 | */ | |
5755 | ||
6f912042 | 5756 | for_each_possible_cpu(i) { |
1da177e4 LT |
5757 | struct softnet_data *queue; |
5758 | ||
5759 | queue = &per_cpu(softnet_data, i); | |
5760 | skb_queue_head_init(&queue->input_pkt_queue); | |
1da177e4 LT |
5761 | queue->completion_queue = NULL; |
5762 | INIT_LIST_HEAD(&queue->poll_list); | |
bea3348e | 5763 | |
df334545 | 5764 | #ifdef CONFIG_RPS |
0a9627f2 TH |
5765 | queue->csd.func = trigger_softirq; |
5766 | queue->csd.info = queue; | |
5767 | queue->csd.flags = 0; | |
1e94d72f | 5768 | #endif |
0a9627f2 | 5769 | |
bea3348e SH |
5770 | queue->backlog.poll = process_backlog; |
5771 | queue->backlog.weight = weight_p; | |
d565b0a1 | 5772 | queue->backlog.gro_list = NULL; |
4ae5544f | 5773 | queue->backlog.gro_count = 0; |
1da177e4 LT |
5774 | } |
5775 | ||
1da177e4 LT |
5776 | dev_boot_phase = 0; |
5777 | ||
505d4f73 EB |
5778 | /* The loopback device is special if any other network devices |
5779 | * is present in a network namespace the loopback device must | |
5780 | * be present. Since we now dynamically allocate and free the | |
5781 | * loopback device ensure this invariant is maintained by | |
5782 | * keeping the loopback device as the first device on the | |
5783 | * list of network devices. Ensuring the loopback devices | |
5784 | * is the first device that appears and the last network device | |
5785 | * that disappears. | |
5786 | */ | |
5787 | if (register_pernet_device(&loopback_net_ops)) | |
5788 | goto out; | |
5789 | ||
5790 | if (register_pernet_device(&default_device_ops)) | |
5791 | goto out; | |
5792 | ||
962cf36c CM |
5793 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
5794 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
5795 | |
5796 | hotcpu_notifier(dev_cpu_callback, 0); | |
5797 | dst_init(); | |
5798 | dev_mcast_init(); | |
5799 | rc = 0; | |
5800 | out: | |
5801 | return rc; | |
5802 | } | |
5803 | ||
5804 | subsys_initcall(net_dev_init); | |
5805 | ||
e88721f8 KK |
5806 | static int __init initialize_hashrnd(void) |
5807 | { | |
0a9627f2 | 5808 | get_random_bytes(&hashrnd, sizeof(hashrnd)); |
e88721f8 KK |
5809 | return 0; |
5810 | } | |
5811 | ||
5812 | late_initcall_sync(initialize_hashrnd); | |
5813 |