]>
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> | |
82 | #include <linux/sched.h> | |
4a3e2f71 | 83 | #include <linux/mutex.h> |
1da177e4 LT |
84 | #include <linux/string.h> |
85 | #include <linux/mm.h> | |
86 | #include <linux/socket.h> | |
87 | #include <linux/sockios.h> | |
88 | #include <linux/errno.h> | |
89 | #include <linux/interrupt.h> | |
90 | #include <linux/if_ether.h> | |
91 | #include <linux/netdevice.h> | |
92 | #include <linux/etherdevice.h> | |
0187bdfb | 93 | #include <linux/ethtool.h> |
1da177e4 LT |
94 | #include <linux/notifier.h> |
95 | #include <linux/skbuff.h> | |
457c4cbc | 96 | #include <net/net_namespace.h> |
1da177e4 LT |
97 | #include <net/sock.h> |
98 | #include <linux/rtnetlink.h> | |
99 | #include <linux/proc_fs.h> | |
100 | #include <linux/seq_file.h> | |
101 | #include <linux/stat.h> | |
102 | #include <linux/if_bridge.h> | |
b863ceb7 | 103 | #include <linux/if_macvlan.h> |
1da177e4 LT |
104 | #include <net/dst.h> |
105 | #include <net/pkt_sched.h> | |
106 | #include <net/checksum.h> | |
107 | #include <linux/highmem.h> | |
108 | #include <linux/init.h> | |
109 | #include <linux/kmod.h> | |
110 | #include <linux/module.h> | |
1da177e4 LT |
111 | #include <linux/netpoll.h> |
112 | #include <linux/rcupdate.h> | |
113 | #include <linux/delay.h> | |
295f4a1f | 114 | #include <net/wext.h> |
1da177e4 | 115 | #include <net/iw_handler.h> |
1da177e4 | 116 | #include <asm/current.h> |
5bdb9886 | 117 | #include <linux/audit.h> |
db217334 | 118 | #include <linux/dmaengine.h> |
f6a78bfc | 119 | #include <linux/err.h> |
c7fa9d18 | 120 | #include <linux/ctype.h> |
723e98b7 | 121 | #include <linux/if_arp.h> |
6de329e2 | 122 | #include <linux/if_vlan.h> |
8f0f2223 | 123 | #include <linux/ip.h> |
ad55dcaf | 124 | #include <net/ip.h> |
8f0f2223 DM |
125 | #include <linux/ipv6.h> |
126 | #include <linux/in.h> | |
b6b2fed1 DM |
127 | #include <linux/jhash.h> |
128 | #include <linux/random.h> | |
1da177e4 | 129 | |
342709ef PE |
130 | #include "net-sysfs.h" |
131 | ||
d565b0a1 HX |
132 | /* Instead of increasing this, you should create a hash table. */ |
133 | #define MAX_GRO_SKBS 8 | |
134 | ||
5d38a079 HX |
135 | /* This should be increased if a protocol with a bigger head is added. */ |
136 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
137 | ||
5d0d9be8 HX |
138 | enum { |
139 | GRO_MERGED, | |
140 | GRO_MERGED_FREE, | |
141 | GRO_HELD, | |
142 | GRO_NORMAL, | |
143 | GRO_DROP, | |
144 | }; | |
145 | ||
1da177e4 LT |
146 | /* |
147 | * The list of packet types we will receive (as opposed to discard) | |
148 | * and the routines to invoke. | |
149 | * | |
150 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
151 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
152 | * | |
153 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
154 | * sure which should go first, but I bet it won't make much | |
155 | * difference if we are running VLANs. The good news is that | |
156 | * this protocol won't be in the list unless compiled in, so | |
3041a069 | 157 | * the average user (w/out VLANs) will not be adversely affected. |
1da177e4 LT |
158 | * --BLG |
159 | * | |
160 | * 0800 IP | |
161 | * 8100 802.1Q VLAN | |
162 | * 0001 802.3 | |
163 | * 0002 AX.25 | |
164 | * 0004 802.2 | |
165 | * 8035 RARP | |
166 | * 0005 SNAP | |
167 | * 0805 X.25 | |
168 | * 0806 ARP | |
169 | * 8137 IPX | |
170 | * 0009 Localtalk | |
171 | * 86DD IPv6 | |
172 | */ | |
173 | ||
82d8a867 PE |
174 | #define PTYPE_HASH_SIZE (16) |
175 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
176 | ||
1da177e4 | 177 | static DEFINE_SPINLOCK(ptype_lock); |
82d8a867 | 178 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
6b2bedc3 | 179 | static struct list_head ptype_all __read_mostly; /* Taps */ |
1da177e4 | 180 | |
1da177e4 | 181 | /* |
7562f876 | 182 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
183 | * semaphore. |
184 | * | |
185 | * Pure readers hold dev_base_lock for reading. | |
186 | * | |
187 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 188 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
189 | * actual updates. This allows pure readers to access the list even |
190 | * while a writer is preparing to update it. | |
191 | * | |
192 | * To put it another way, dev_base_lock is held for writing only to | |
193 | * protect against pure readers; the rtnl semaphore provides the | |
194 | * protection against other writers. | |
195 | * | |
196 | * See, for example usages, register_netdevice() and | |
197 | * unregister_netdevice(), which must be called with the rtnl | |
198 | * semaphore held. | |
199 | */ | |
1da177e4 LT |
200 | DEFINE_RWLOCK(dev_base_lock); |
201 | ||
1da177e4 LT |
202 | EXPORT_SYMBOL(dev_base_lock); |
203 | ||
204 | #define NETDEV_HASHBITS 8 | |
881d966b | 205 | #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS) |
1da177e4 | 206 | |
881d966b | 207 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 LT |
208 | { |
209 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
881d966b | 210 | return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)]; |
1da177e4 LT |
211 | } |
212 | ||
881d966b | 213 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 214 | { |
881d966b | 215 | return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)]; |
1da177e4 LT |
216 | } |
217 | ||
ce286d32 EB |
218 | /* Device list insertion */ |
219 | static int list_netdevice(struct net_device *dev) | |
220 | { | |
c346dca1 | 221 | struct net *net = dev_net(dev); |
ce286d32 EB |
222 | |
223 | ASSERT_RTNL(); | |
224 | ||
225 | write_lock_bh(&dev_base_lock); | |
226 | list_add_tail(&dev->dev_list, &net->dev_base_head); | |
227 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
228 | hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex)); | |
229 | write_unlock_bh(&dev_base_lock); | |
230 | return 0; | |
231 | } | |
232 | ||
233 | /* Device list removal */ | |
234 | static void unlist_netdevice(struct net_device *dev) | |
235 | { | |
236 | ASSERT_RTNL(); | |
237 | ||
238 | /* Unlink dev from the device chain */ | |
239 | write_lock_bh(&dev_base_lock); | |
240 | list_del(&dev->dev_list); | |
241 | hlist_del(&dev->name_hlist); | |
242 | hlist_del(&dev->index_hlist); | |
243 | write_unlock_bh(&dev_base_lock); | |
244 | } | |
245 | ||
1da177e4 LT |
246 | /* |
247 | * Our notifier list | |
248 | */ | |
249 | ||
f07d5b94 | 250 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
251 | |
252 | /* | |
253 | * Device drivers call our routines to queue packets here. We empty the | |
254 | * queue in the local softnet handler. | |
255 | */ | |
bea3348e SH |
256 | |
257 | DEFINE_PER_CPU(struct softnet_data, softnet_data); | |
1da177e4 | 258 | |
cf508b12 | 259 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 260 | /* |
c773e847 | 261 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
262 | * according to dev->type |
263 | */ | |
264 | static const unsigned short netdev_lock_type[] = | |
265 | {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
266 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, | |
267 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
268 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
269 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
270 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
271 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
272 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
273 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
274 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
275 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
276 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
277 | ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211, | |
2d91d78b | 278 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, |
57c81fff | 279 | ARPHRD_PHONET_PIPE, ARPHRD_VOID, ARPHRD_NONE}; |
723e98b7 JP |
280 | |
281 | static const char *netdev_lock_name[] = | |
282 | {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
283 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
284 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
285 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
286 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
287 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
288 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
289 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
290 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
291 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
292 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
293 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
294 | "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211", | |
2d91d78b | 295 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", |
57c81fff | 296 | "_xmit_PHONET_PIPE", "_xmit_VOID", "_xmit_NONE"}; |
723e98b7 JP |
297 | |
298 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 299 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
300 | |
301 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
302 | { | |
303 | int i; | |
304 | ||
305 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
306 | if (netdev_lock_type[i] == dev_type) | |
307 | return i; | |
308 | /* the last key is used by default */ | |
309 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
310 | } | |
311 | ||
cf508b12 DM |
312 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
313 | unsigned short dev_type) | |
723e98b7 JP |
314 | { |
315 | int i; | |
316 | ||
317 | i = netdev_lock_pos(dev_type); | |
318 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
319 | netdev_lock_name[i]); | |
320 | } | |
cf508b12 DM |
321 | |
322 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
323 | { | |
324 | int i; | |
325 | ||
326 | i = netdev_lock_pos(dev->type); | |
327 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
328 | &netdev_addr_lock_key[i], | |
329 | netdev_lock_name[i]); | |
330 | } | |
723e98b7 | 331 | #else |
cf508b12 DM |
332 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
333 | unsigned short dev_type) | |
334 | { | |
335 | } | |
336 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
337 | { |
338 | } | |
339 | #endif | |
1da177e4 LT |
340 | |
341 | /******************************************************************************* | |
342 | ||
343 | Protocol management and registration routines | |
344 | ||
345 | *******************************************************************************/ | |
346 | ||
1da177e4 LT |
347 | /* |
348 | * Add a protocol ID to the list. Now that the input handler is | |
349 | * smarter we can dispense with all the messy stuff that used to be | |
350 | * here. | |
351 | * | |
352 | * BEWARE!!! Protocol handlers, mangling input packets, | |
353 | * MUST BE last in hash buckets and checking protocol handlers | |
354 | * MUST start from promiscuous ptype_all chain in net_bh. | |
355 | * It is true now, do not change it. | |
356 | * Explanation follows: if protocol handler, mangling packet, will | |
357 | * be the first on list, it is not able to sense, that packet | |
358 | * is cloned and should be copied-on-write, so that it will | |
359 | * change it and subsequent readers will get broken packet. | |
360 | * --ANK (980803) | |
361 | */ | |
362 | ||
363 | /** | |
364 | * dev_add_pack - add packet handler | |
365 | * @pt: packet type declaration | |
366 | * | |
367 | * Add a protocol handler to the networking stack. The passed &packet_type | |
368 | * is linked into kernel lists and may not be freed until it has been | |
369 | * removed from the kernel lists. | |
370 | * | |
4ec93edb | 371 | * This call does not sleep therefore it can not |
1da177e4 LT |
372 | * guarantee all CPU's that are in middle of receiving packets |
373 | * will see the new packet type (until the next received packet). | |
374 | */ | |
375 | ||
376 | void dev_add_pack(struct packet_type *pt) | |
377 | { | |
378 | int hash; | |
379 | ||
380 | spin_lock_bh(&ptype_lock); | |
9be9a6b9 | 381 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 382 | list_add_rcu(&pt->list, &ptype_all); |
9be9a6b9 | 383 | else { |
82d8a867 | 384 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
1da177e4 LT |
385 | list_add_rcu(&pt->list, &ptype_base[hash]); |
386 | } | |
387 | spin_unlock_bh(&ptype_lock); | |
388 | } | |
389 | ||
1da177e4 LT |
390 | /** |
391 | * __dev_remove_pack - remove packet handler | |
392 | * @pt: packet type declaration | |
393 | * | |
394 | * Remove a protocol handler that was previously added to the kernel | |
395 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
396 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 397 | * returns. |
1da177e4 LT |
398 | * |
399 | * The packet type might still be in use by receivers | |
400 | * and must not be freed until after all the CPU's have gone | |
401 | * through a quiescent state. | |
402 | */ | |
403 | void __dev_remove_pack(struct packet_type *pt) | |
404 | { | |
405 | struct list_head *head; | |
406 | struct packet_type *pt1; | |
407 | ||
408 | spin_lock_bh(&ptype_lock); | |
409 | ||
9be9a6b9 | 410 | if (pt->type == htons(ETH_P_ALL)) |
1da177e4 | 411 | head = &ptype_all; |
9be9a6b9 | 412 | else |
82d8a867 | 413 | head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; |
1da177e4 LT |
414 | |
415 | list_for_each_entry(pt1, head, list) { | |
416 | if (pt == pt1) { | |
417 | list_del_rcu(&pt->list); | |
418 | goto out; | |
419 | } | |
420 | } | |
421 | ||
422 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
423 | out: | |
424 | spin_unlock_bh(&ptype_lock); | |
425 | } | |
426 | /** | |
427 | * dev_remove_pack - remove packet handler | |
428 | * @pt: packet type declaration | |
429 | * | |
430 | * Remove a protocol handler that was previously added to the kernel | |
431 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
432 | * from the kernel lists and can be freed or reused once this function | |
433 | * returns. | |
434 | * | |
435 | * This call sleeps to guarantee that no CPU is looking at the packet | |
436 | * type after return. | |
437 | */ | |
438 | void dev_remove_pack(struct packet_type *pt) | |
439 | { | |
440 | __dev_remove_pack(pt); | |
4ec93edb | 441 | |
1da177e4 LT |
442 | synchronize_net(); |
443 | } | |
444 | ||
445 | /****************************************************************************** | |
446 | ||
447 | Device Boot-time Settings Routines | |
448 | ||
449 | *******************************************************************************/ | |
450 | ||
451 | /* Boot time configuration table */ | |
452 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
453 | ||
454 | /** | |
455 | * netdev_boot_setup_add - add new setup entry | |
456 | * @name: name of the device | |
457 | * @map: configured settings for the device | |
458 | * | |
459 | * Adds new setup entry to the dev_boot_setup list. The function | |
460 | * returns 0 on error and 1 on success. This is a generic routine to | |
461 | * all netdevices. | |
462 | */ | |
463 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
464 | { | |
465 | struct netdev_boot_setup *s; | |
466 | int i; | |
467 | ||
468 | s = dev_boot_setup; | |
469 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
470 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
471 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 472 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
473 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
474 | break; | |
475 | } | |
476 | } | |
477 | ||
478 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
479 | } | |
480 | ||
481 | /** | |
482 | * netdev_boot_setup_check - check boot time settings | |
483 | * @dev: the netdevice | |
484 | * | |
485 | * Check boot time settings for the device. | |
486 | * The found settings are set for the device to be used | |
487 | * later in the device probing. | |
488 | * Returns 0 if no settings found, 1 if they are. | |
489 | */ | |
490 | int netdev_boot_setup_check(struct net_device *dev) | |
491 | { | |
492 | struct netdev_boot_setup *s = dev_boot_setup; | |
493 | int i; | |
494 | ||
495 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
496 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 497 | !strcmp(dev->name, s[i].name)) { |
1da177e4 LT |
498 | dev->irq = s[i].map.irq; |
499 | dev->base_addr = s[i].map.base_addr; | |
500 | dev->mem_start = s[i].map.mem_start; | |
501 | dev->mem_end = s[i].map.mem_end; | |
502 | return 1; | |
503 | } | |
504 | } | |
505 | return 0; | |
506 | } | |
507 | ||
508 | ||
509 | /** | |
510 | * netdev_boot_base - get address from boot time settings | |
511 | * @prefix: prefix for network device | |
512 | * @unit: id for network device | |
513 | * | |
514 | * Check boot time settings for the base address of device. | |
515 | * The found settings are set for the device to be used | |
516 | * later in the device probing. | |
517 | * Returns 0 if no settings found. | |
518 | */ | |
519 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
520 | { | |
521 | const struct netdev_boot_setup *s = dev_boot_setup; | |
522 | char name[IFNAMSIZ]; | |
523 | int i; | |
524 | ||
525 | sprintf(name, "%s%d", prefix, unit); | |
526 | ||
527 | /* | |
528 | * If device already registered then return base of 1 | |
529 | * to indicate not to probe for this interface | |
530 | */ | |
881d966b | 531 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
532 | return 1; |
533 | ||
534 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
535 | if (!strcmp(name, s[i].name)) | |
536 | return s[i].map.base_addr; | |
537 | return 0; | |
538 | } | |
539 | ||
540 | /* | |
541 | * Saves at boot time configured settings for any netdevice. | |
542 | */ | |
543 | int __init netdev_boot_setup(char *str) | |
544 | { | |
545 | int ints[5]; | |
546 | struct ifmap map; | |
547 | ||
548 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
549 | if (!str || !*str) | |
550 | return 0; | |
551 | ||
552 | /* Save settings */ | |
553 | memset(&map, 0, sizeof(map)); | |
554 | if (ints[0] > 0) | |
555 | map.irq = ints[1]; | |
556 | if (ints[0] > 1) | |
557 | map.base_addr = ints[2]; | |
558 | if (ints[0] > 2) | |
559 | map.mem_start = ints[3]; | |
560 | if (ints[0] > 3) | |
561 | map.mem_end = ints[4]; | |
562 | ||
563 | /* Add new entry to the list */ | |
564 | return netdev_boot_setup_add(str, &map); | |
565 | } | |
566 | ||
567 | __setup("netdev=", netdev_boot_setup); | |
568 | ||
569 | /******************************************************************************* | |
570 | ||
571 | Device Interface Subroutines | |
572 | ||
573 | *******************************************************************************/ | |
574 | ||
575 | /** | |
576 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 577 | * @net: the applicable net namespace |
1da177e4 LT |
578 | * @name: name to find |
579 | * | |
580 | * Find an interface by name. Must be called under RTNL semaphore | |
581 | * or @dev_base_lock. If the name is found a pointer to the device | |
582 | * is returned. If the name is not found then %NULL is returned. The | |
583 | * reference counters are not incremented so the caller must be | |
584 | * careful with locks. | |
585 | */ | |
586 | ||
881d966b | 587 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
588 | { |
589 | struct hlist_node *p; | |
590 | ||
881d966b | 591 | hlist_for_each(p, dev_name_hash(net, name)) { |
1da177e4 LT |
592 | struct net_device *dev |
593 | = hlist_entry(p, struct net_device, name_hlist); | |
594 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
595 | return dev; | |
596 | } | |
597 | return NULL; | |
598 | } | |
599 | ||
600 | /** | |
601 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 602 | * @net: the applicable net namespace |
1da177e4 LT |
603 | * @name: name to find |
604 | * | |
605 | * Find an interface by name. This can be called from any | |
606 | * context and does its own locking. The returned handle has | |
607 | * the usage count incremented and the caller must use dev_put() to | |
608 | * release it when it is no longer needed. %NULL is returned if no | |
609 | * matching device is found. | |
610 | */ | |
611 | ||
881d966b | 612 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
613 | { |
614 | struct net_device *dev; | |
615 | ||
616 | read_lock(&dev_base_lock); | |
881d966b | 617 | dev = __dev_get_by_name(net, name); |
1da177e4 LT |
618 | if (dev) |
619 | dev_hold(dev); | |
620 | read_unlock(&dev_base_lock); | |
621 | return dev; | |
622 | } | |
623 | ||
624 | /** | |
625 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 626 | * @net: the applicable net namespace |
1da177e4 LT |
627 | * @ifindex: index of device |
628 | * | |
629 | * Search for an interface by index. Returns %NULL if the device | |
630 | * is not found or a pointer to the device. The device has not | |
631 | * had its reference counter increased so the caller must be careful | |
632 | * about locking. The caller must hold either the RTNL semaphore | |
633 | * or @dev_base_lock. | |
634 | */ | |
635 | ||
881d966b | 636 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
637 | { |
638 | struct hlist_node *p; | |
639 | ||
881d966b | 640 | hlist_for_each(p, dev_index_hash(net, ifindex)) { |
1da177e4 LT |
641 | struct net_device *dev |
642 | = hlist_entry(p, struct net_device, index_hlist); | |
643 | if (dev->ifindex == ifindex) | |
644 | return dev; | |
645 | } | |
646 | return NULL; | |
647 | } | |
648 | ||
649 | ||
650 | /** | |
651 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 652 | * @net: the applicable net namespace |
1da177e4 LT |
653 | * @ifindex: index of device |
654 | * | |
655 | * Search for an interface by index. Returns NULL if the device | |
656 | * is not found or a pointer to the device. The device returned has | |
657 | * had a reference added and the pointer is safe until the user calls | |
658 | * dev_put to indicate they have finished with it. | |
659 | */ | |
660 | ||
881d966b | 661 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
662 | { |
663 | struct net_device *dev; | |
664 | ||
665 | read_lock(&dev_base_lock); | |
881d966b | 666 | dev = __dev_get_by_index(net, ifindex); |
1da177e4 LT |
667 | if (dev) |
668 | dev_hold(dev); | |
669 | read_unlock(&dev_base_lock); | |
670 | return dev; | |
671 | } | |
672 | ||
673 | /** | |
674 | * dev_getbyhwaddr - find a device by its hardware address | |
c4ea43c5 | 675 | * @net: the applicable net namespace |
1da177e4 LT |
676 | * @type: media type of device |
677 | * @ha: hardware address | |
678 | * | |
679 | * Search for an interface by MAC address. Returns NULL if the device | |
680 | * is not found or a pointer to the device. The caller must hold the | |
681 | * rtnl semaphore. The returned device has not had its ref count increased | |
682 | * and the caller must therefore be careful about locking | |
683 | * | |
684 | * BUGS: | |
685 | * If the API was consistent this would be __dev_get_by_hwaddr | |
686 | */ | |
687 | ||
881d966b | 688 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha) |
1da177e4 LT |
689 | { |
690 | struct net_device *dev; | |
691 | ||
692 | ASSERT_RTNL(); | |
693 | ||
81103a52 | 694 | for_each_netdev(net, dev) |
1da177e4 LT |
695 | if (dev->type == type && |
696 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
697 | return dev; |
698 | ||
699 | return NULL; | |
1da177e4 LT |
700 | } |
701 | ||
cf309e3f JF |
702 | EXPORT_SYMBOL(dev_getbyhwaddr); |
703 | ||
881d966b | 704 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
705 | { |
706 | struct net_device *dev; | |
707 | ||
4e9cac2b | 708 | ASSERT_RTNL(); |
881d966b | 709 | for_each_netdev(net, dev) |
4e9cac2b | 710 | if (dev->type == type) |
7562f876 PE |
711 | return dev; |
712 | ||
713 | return NULL; | |
4e9cac2b PM |
714 | } |
715 | ||
716 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); | |
717 | ||
881d966b | 718 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b PM |
719 | { |
720 | struct net_device *dev; | |
721 | ||
722 | rtnl_lock(); | |
881d966b | 723 | dev = __dev_getfirstbyhwtype(net, type); |
4e9cac2b PM |
724 | if (dev) |
725 | dev_hold(dev); | |
1da177e4 LT |
726 | rtnl_unlock(); |
727 | return dev; | |
728 | } | |
729 | ||
730 | EXPORT_SYMBOL(dev_getfirstbyhwtype); | |
731 | ||
732 | /** | |
733 | * dev_get_by_flags - find any device with given flags | |
c4ea43c5 | 734 | * @net: the applicable net namespace |
1da177e4 LT |
735 | * @if_flags: IFF_* values |
736 | * @mask: bitmask of bits in if_flags to check | |
737 | * | |
738 | * Search for any interface with the given flags. Returns NULL if a device | |
4ec93edb | 739 | * is not found or a pointer to the device. The device returned has |
1da177e4 LT |
740 | * had a reference added and the pointer is safe until the user calls |
741 | * dev_put to indicate they have finished with it. | |
742 | */ | |
743 | ||
881d966b | 744 | struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask) |
1da177e4 | 745 | { |
7562f876 | 746 | struct net_device *dev, *ret; |
1da177e4 | 747 | |
7562f876 | 748 | ret = NULL; |
1da177e4 | 749 | read_lock(&dev_base_lock); |
881d966b | 750 | for_each_netdev(net, dev) { |
1da177e4 LT |
751 | if (((dev->flags ^ if_flags) & mask) == 0) { |
752 | dev_hold(dev); | |
7562f876 | 753 | ret = dev; |
1da177e4 LT |
754 | break; |
755 | } | |
756 | } | |
757 | read_unlock(&dev_base_lock); | |
7562f876 | 758 | return ret; |
1da177e4 LT |
759 | } |
760 | ||
761 | /** | |
762 | * dev_valid_name - check if name is okay for network device | |
763 | * @name: name string | |
764 | * | |
765 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
766 | * to allow sysfs to work. We also disallow any kind of |
767 | * whitespace. | |
1da177e4 | 768 | */ |
c2373ee9 | 769 | int dev_valid_name(const char *name) |
1da177e4 | 770 | { |
c7fa9d18 DM |
771 | if (*name == '\0') |
772 | return 0; | |
b6fe17d6 SH |
773 | if (strlen(name) >= IFNAMSIZ) |
774 | return 0; | |
c7fa9d18 DM |
775 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
776 | return 0; | |
777 | ||
778 | while (*name) { | |
779 | if (*name == '/' || isspace(*name)) | |
780 | return 0; | |
781 | name++; | |
782 | } | |
783 | return 1; | |
1da177e4 LT |
784 | } |
785 | ||
786 | /** | |
b267b179 EB |
787 | * __dev_alloc_name - allocate a name for a device |
788 | * @net: network namespace to allocate the device name in | |
1da177e4 | 789 | * @name: name format string |
b267b179 | 790 | * @buf: scratch buffer and result name string |
1da177e4 LT |
791 | * |
792 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
793 | * id. It scans list of devices to build up a free map, then chooses |
794 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
795 | * while allocating the name and adding the device in order to avoid | |
796 | * duplicates. | |
797 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
798 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
799 | */ |
800 | ||
b267b179 | 801 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
802 | { |
803 | int i = 0; | |
1da177e4 LT |
804 | const char *p; |
805 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 806 | unsigned long *inuse; |
1da177e4 LT |
807 | struct net_device *d; |
808 | ||
809 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
810 | if (p) { | |
811 | /* | |
812 | * Verify the string as this thing may have come from | |
813 | * the user. There must be either one "%d" and no other "%" | |
814 | * characters. | |
815 | */ | |
816 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
817 | return -EINVAL; | |
818 | ||
819 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 820 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
821 | if (!inuse) |
822 | return -ENOMEM; | |
823 | ||
881d966b | 824 | for_each_netdev(net, d) { |
1da177e4 LT |
825 | if (!sscanf(d->name, name, &i)) |
826 | continue; | |
827 | if (i < 0 || i >= max_netdevices) | |
828 | continue; | |
829 | ||
830 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 831 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
832 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
833 | set_bit(i, inuse); | |
834 | } | |
835 | ||
836 | i = find_first_zero_bit(inuse, max_netdevices); | |
837 | free_page((unsigned long) inuse); | |
838 | } | |
839 | ||
b267b179 EB |
840 | snprintf(buf, IFNAMSIZ, name, i); |
841 | if (!__dev_get_by_name(net, buf)) | |
1da177e4 | 842 | return i; |
1da177e4 LT |
843 | |
844 | /* It is possible to run out of possible slots | |
845 | * when the name is long and there isn't enough space left | |
846 | * for the digits, or if all bits are used. | |
847 | */ | |
848 | return -ENFILE; | |
849 | } | |
850 | ||
b267b179 EB |
851 | /** |
852 | * dev_alloc_name - allocate a name for a device | |
853 | * @dev: device | |
854 | * @name: name format string | |
855 | * | |
856 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
857 | * id. It scans list of devices to build up a free map, then chooses | |
858 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
859 | * while allocating the name and adding the device in order to avoid | |
860 | * duplicates. | |
861 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
862 | * Returns the number of the unit assigned or a negative errno code. | |
863 | */ | |
864 | ||
865 | int dev_alloc_name(struct net_device *dev, const char *name) | |
866 | { | |
867 | char buf[IFNAMSIZ]; | |
868 | struct net *net; | |
869 | int ret; | |
870 | ||
c346dca1 YH |
871 | BUG_ON(!dev_net(dev)); |
872 | net = dev_net(dev); | |
b267b179 EB |
873 | ret = __dev_alloc_name(net, name, buf); |
874 | if (ret >= 0) | |
875 | strlcpy(dev->name, buf, IFNAMSIZ); | |
876 | return ret; | |
877 | } | |
878 | ||
1da177e4 LT |
879 | |
880 | /** | |
881 | * dev_change_name - change name of a device | |
882 | * @dev: device | |
883 | * @newname: name (or format string) must be at least IFNAMSIZ | |
884 | * | |
885 | * Change name of a device, can pass format strings "eth%d". | |
886 | * for wildcarding. | |
887 | */ | |
cf04a4c7 | 888 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 889 | { |
fcc5a03a | 890 | char oldname[IFNAMSIZ]; |
1da177e4 | 891 | int err = 0; |
fcc5a03a | 892 | int ret; |
881d966b | 893 | struct net *net; |
1da177e4 LT |
894 | |
895 | ASSERT_RTNL(); | |
c346dca1 | 896 | BUG_ON(!dev_net(dev)); |
1da177e4 | 897 | |
c346dca1 | 898 | net = dev_net(dev); |
1da177e4 LT |
899 | if (dev->flags & IFF_UP) |
900 | return -EBUSY; | |
901 | ||
902 | if (!dev_valid_name(newname)) | |
903 | return -EINVAL; | |
904 | ||
c8d90dca SH |
905 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) |
906 | return 0; | |
907 | ||
fcc5a03a HX |
908 | memcpy(oldname, dev->name, IFNAMSIZ); |
909 | ||
1da177e4 LT |
910 | if (strchr(newname, '%')) { |
911 | err = dev_alloc_name(dev, newname); | |
912 | if (err < 0) | |
913 | return err; | |
1da177e4 | 914 | } |
881d966b | 915 | else if (__dev_get_by_name(net, newname)) |
1da177e4 LT |
916 | return -EEXIST; |
917 | else | |
918 | strlcpy(dev->name, newname, IFNAMSIZ); | |
919 | ||
fcc5a03a | 920 | rollback: |
3891845e EB |
921 | /* For now only devices in the initial network namespace |
922 | * are in sysfs. | |
923 | */ | |
924 | if (net == &init_net) { | |
925 | ret = device_rename(&dev->dev, dev->name); | |
926 | if (ret) { | |
927 | memcpy(dev->name, oldname, IFNAMSIZ); | |
928 | return ret; | |
929 | } | |
dcc99773 | 930 | } |
7f988eab HX |
931 | |
932 | write_lock_bh(&dev_base_lock); | |
92749821 | 933 | hlist_del(&dev->name_hlist); |
881d966b | 934 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); |
7f988eab HX |
935 | write_unlock_bh(&dev_base_lock); |
936 | ||
056925ab | 937 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
938 | ret = notifier_to_errno(ret); |
939 | ||
940 | if (ret) { | |
941 | if (err) { | |
942 | printk(KERN_ERR | |
943 | "%s: name change rollback failed: %d.\n", | |
944 | dev->name, ret); | |
945 | } else { | |
946 | err = ret; | |
947 | memcpy(dev->name, oldname, IFNAMSIZ); | |
948 | goto rollback; | |
949 | } | |
950 | } | |
1da177e4 LT |
951 | |
952 | return err; | |
953 | } | |
954 | ||
0b815a1a SH |
955 | /** |
956 | * dev_set_alias - change ifalias of a device | |
957 | * @dev: device | |
958 | * @alias: name up to IFALIASZ | |
f0db275a | 959 | * @len: limit of bytes to copy from info |
0b815a1a SH |
960 | * |
961 | * Set ifalias for a device, | |
962 | */ | |
963 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
964 | { | |
965 | ASSERT_RTNL(); | |
966 | ||
967 | if (len >= IFALIASZ) | |
968 | return -EINVAL; | |
969 | ||
96ca4a2c OH |
970 | if (!len) { |
971 | if (dev->ifalias) { | |
972 | kfree(dev->ifalias); | |
973 | dev->ifalias = NULL; | |
974 | } | |
975 | return 0; | |
976 | } | |
977 | ||
0b815a1a SH |
978 | dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL); |
979 | if (!dev->ifalias) | |
980 | return -ENOMEM; | |
981 | ||
982 | strlcpy(dev->ifalias, alias, len+1); | |
983 | return len; | |
984 | } | |
985 | ||
986 | ||
d8a33ac4 | 987 | /** |
3041a069 | 988 | * netdev_features_change - device changes features |
d8a33ac4 SH |
989 | * @dev: device to cause notification |
990 | * | |
991 | * Called to indicate a device has changed features. | |
992 | */ | |
993 | void netdev_features_change(struct net_device *dev) | |
994 | { | |
056925ab | 995 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
996 | } |
997 | EXPORT_SYMBOL(netdev_features_change); | |
998 | ||
1da177e4 LT |
999 | /** |
1000 | * netdev_state_change - device changes state | |
1001 | * @dev: device to cause notification | |
1002 | * | |
1003 | * Called to indicate a device has changed state. This function calls | |
1004 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1005 | * to the routing socket. | |
1006 | */ | |
1007 | void netdev_state_change(struct net_device *dev) | |
1008 | { | |
1009 | if (dev->flags & IFF_UP) { | |
056925ab | 1010 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
1011 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
1012 | } | |
1013 | } | |
1014 | ||
c1da4ac7 OG |
1015 | void netdev_bonding_change(struct net_device *dev) |
1016 | { | |
1017 | call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev); | |
1018 | } | |
1019 | EXPORT_SYMBOL(netdev_bonding_change); | |
1020 | ||
1da177e4 LT |
1021 | /** |
1022 | * dev_load - load a network module | |
c4ea43c5 | 1023 | * @net: the applicable net namespace |
1da177e4 LT |
1024 | * @name: name of interface |
1025 | * | |
1026 | * If a network interface is not present and the process has suitable | |
1027 | * privileges this function loads the module. If module loading is not | |
1028 | * available in this kernel then it becomes a nop. | |
1029 | */ | |
1030 | ||
881d966b | 1031 | void dev_load(struct net *net, const char *name) |
1da177e4 | 1032 | { |
4ec93edb | 1033 | struct net_device *dev; |
1da177e4 LT |
1034 | |
1035 | read_lock(&dev_base_lock); | |
881d966b | 1036 | dev = __dev_get_by_name(net, name); |
1da177e4 LT |
1037 | read_unlock(&dev_base_lock); |
1038 | ||
1039 | if (!dev && capable(CAP_SYS_MODULE)) | |
1040 | request_module("%s", name); | |
1041 | } | |
1042 | ||
1da177e4 LT |
1043 | /** |
1044 | * dev_open - prepare an interface for use. | |
1045 | * @dev: device to open | |
1046 | * | |
1047 | * Takes a device from down to up state. The device's private open | |
1048 | * function is invoked and then the multicast lists are loaded. Finally | |
1049 | * the device is moved into the up state and a %NETDEV_UP message is | |
1050 | * sent to the netdev notifier chain. | |
1051 | * | |
1052 | * Calling this function on an active interface is a nop. On a failure | |
1053 | * a negative errno code is returned. | |
1054 | */ | |
1055 | int dev_open(struct net_device *dev) | |
1056 | { | |
d314774c | 1057 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
1058 | int ret = 0; |
1059 | ||
e46b66bc BH |
1060 | ASSERT_RTNL(); |
1061 | ||
1da177e4 LT |
1062 | /* |
1063 | * Is it already up? | |
1064 | */ | |
1065 | ||
1066 | if (dev->flags & IFF_UP) | |
1067 | return 0; | |
1068 | ||
1069 | /* | |
1070 | * Is it even present? | |
1071 | */ | |
1072 | if (!netif_device_present(dev)) | |
1073 | return -ENODEV; | |
1074 | ||
1075 | /* | |
1076 | * Call device private open method | |
1077 | */ | |
1078 | set_bit(__LINK_STATE_START, &dev->state); | |
bada339b | 1079 | |
d314774c SH |
1080 | if (ops->ndo_validate_addr) |
1081 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1082 | |
d314774c SH |
1083 | if (!ret && ops->ndo_open) |
1084 | ret = ops->ndo_open(dev); | |
1da177e4 | 1085 | |
4ec93edb | 1086 | /* |
1da177e4 LT |
1087 | * If it went open OK then: |
1088 | */ | |
1089 | ||
bada339b JG |
1090 | if (ret) |
1091 | clear_bit(__LINK_STATE_START, &dev->state); | |
1092 | else { | |
1da177e4 LT |
1093 | /* |
1094 | * Set the flags. | |
1095 | */ | |
1096 | dev->flags |= IFF_UP; | |
1097 | ||
649274d9 DW |
1098 | /* |
1099 | * Enable NET_DMA | |
1100 | */ | |
b4bd07c2 | 1101 | net_dmaengine_get(); |
649274d9 | 1102 | |
1da177e4 LT |
1103 | /* |
1104 | * Initialize multicasting status | |
1105 | */ | |
4417da66 | 1106 | dev_set_rx_mode(dev); |
1da177e4 LT |
1107 | |
1108 | /* | |
1109 | * Wakeup transmit queue engine | |
1110 | */ | |
1111 | dev_activate(dev); | |
1112 | ||
1113 | /* | |
1114 | * ... and announce new interface. | |
1115 | */ | |
056925ab | 1116 | call_netdevice_notifiers(NETDEV_UP, dev); |
1da177e4 | 1117 | } |
bada339b | 1118 | |
1da177e4 LT |
1119 | return ret; |
1120 | } | |
1121 | ||
1122 | /** | |
1123 | * dev_close - shutdown an interface. | |
1124 | * @dev: device to shutdown | |
1125 | * | |
1126 | * This function moves an active device into down state. A | |
1127 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1128 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1129 | * chain. | |
1130 | */ | |
1131 | int dev_close(struct net_device *dev) | |
1132 | { | |
d314774c | 1133 | const struct net_device_ops *ops = dev->netdev_ops; |
e46b66bc BH |
1134 | ASSERT_RTNL(); |
1135 | ||
9d5010db DM |
1136 | might_sleep(); |
1137 | ||
1da177e4 LT |
1138 | if (!(dev->flags & IFF_UP)) |
1139 | return 0; | |
1140 | ||
1141 | /* | |
1142 | * Tell people we are going down, so that they can | |
1143 | * prepare to death, when device is still operating. | |
1144 | */ | |
056925ab | 1145 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1146 | |
1da177e4 LT |
1147 | clear_bit(__LINK_STATE_START, &dev->state); |
1148 | ||
1149 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
bea3348e SH |
1150 | * it can be even on different cpu. So just clear netif_running(). |
1151 | * | |
1152 | * dev->stop() will invoke napi_disable() on all of it's | |
1153 | * napi_struct instances on this device. | |
1154 | */ | |
1da177e4 | 1155 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ |
1da177e4 | 1156 | |
d8b2a4d2 ML |
1157 | dev_deactivate(dev); |
1158 | ||
1da177e4 LT |
1159 | /* |
1160 | * Call the device specific close. This cannot fail. | |
1161 | * Only if device is UP | |
1162 | * | |
1163 | * We allow it to be called even after a DETACH hot-plug | |
1164 | * event. | |
1165 | */ | |
d314774c SH |
1166 | if (ops->ndo_stop) |
1167 | ops->ndo_stop(dev); | |
1da177e4 LT |
1168 | |
1169 | /* | |
1170 | * Device is now down. | |
1171 | */ | |
1172 | ||
1173 | dev->flags &= ~IFF_UP; | |
1174 | ||
1175 | /* | |
1176 | * Tell people we are down | |
1177 | */ | |
056925ab | 1178 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
1da177e4 | 1179 | |
649274d9 DW |
1180 | /* |
1181 | * Shutdown NET_DMA | |
1182 | */ | |
b4bd07c2 | 1183 | net_dmaengine_put(); |
649274d9 | 1184 | |
1da177e4 LT |
1185 | return 0; |
1186 | } | |
1187 | ||
1188 | ||
0187bdfb BH |
1189 | /** |
1190 | * dev_disable_lro - disable Large Receive Offload on a device | |
1191 | * @dev: device | |
1192 | * | |
1193 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1194 | * called under RTNL. This is needed if received packets may be | |
1195 | * forwarded to another interface. | |
1196 | */ | |
1197 | void dev_disable_lro(struct net_device *dev) | |
1198 | { | |
1199 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1200 | dev->ethtool_ops->set_flags) { | |
1201 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1202 | if (flags & ETH_FLAG_LRO) { | |
1203 | flags &= ~ETH_FLAG_LRO; | |
1204 | dev->ethtool_ops->set_flags(dev, flags); | |
1205 | } | |
1206 | } | |
1207 | WARN_ON(dev->features & NETIF_F_LRO); | |
1208 | } | |
1209 | EXPORT_SYMBOL(dev_disable_lro); | |
1210 | ||
1211 | ||
881d966b EB |
1212 | static int dev_boot_phase = 1; |
1213 | ||
1da177e4 LT |
1214 | /* |
1215 | * Device change register/unregister. These are not inline or static | |
1216 | * as we export them to the world. | |
1217 | */ | |
1218 | ||
1219 | /** | |
1220 | * register_netdevice_notifier - register a network notifier block | |
1221 | * @nb: notifier | |
1222 | * | |
1223 | * Register a notifier to be called when network device events occur. | |
1224 | * The notifier passed is linked into the kernel structures and must | |
1225 | * not be reused until it has been unregistered. A negative errno code | |
1226 | * is returned on a failure. | |
1227 | * | |
1228 | * When registered all registration and up events are replayed | |
4ec93edb | 1229 | * to the new notifier to allow device to have a race free |
1da177e4 LT |
1230 | * view of the network device list. |
1231 | */ | |
1232 | ||
1233 | int register_netdevice_notifier(struct notifier_block *nb) | |
1234 | { | |
1235 | struct net_device *dev; | |
fcc5a03a | 1236 | struct net_device *last; |
881d966b | 1237 | struct net *net; |
1da177e4 LT |
1238 | int err; |
1239 | ||
1240 | rtnl_lock(); | |
f07d5b94 | 1241 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1242 | if (err) |
1243 | goto unlock; | |
881d966b EB |
1244 | if (dev_boot_phase) |
1245 | goto unlock; | |
1246 | for_each_net(net) { | |
1247 | for_each_netdev(net, dev) { | |
1248 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1249 | err = notifier_to_errno(err); | |
1250 | if (err) | |
1251 | goto rollback; | |
1252 | ||
1253 | if (!(dev->flags & IFF_UP)) | |
1254 | continue; | |
1da177e4 | 1255 | |
881d966b EB |
1256 | nb->notifier_call(nb, NETDEV_UP, dev); |
1257 | } | |
1da177e4 | 1258 | } |
fcc5a03a HX |
1259 | |
1260 | unlock: | |
1da177e4 LT |
1261 | rtnl_unlock(); |
1262 | return err; | |
fcc5a03a HX |
1263 | |
1264 | rollback: | |
1265 | last = dev; | |
881d966b EB |
1266 | for_each_net(net) { |
1267 | for_each_netdev(net, dev) { | |
1268 | if (dev == last) | |
1269 | break; | |
fcc5a03a | 1270 | |
881d966b EB |
1271 | if (dev->flags & IFF_UP) { |
1272 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1273 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1274 | } | |
1275 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
fcc5a03a | 1276 | } |
fcc5a03a | 1277 | } |
c67625a1 PE |
1278 | |
1279 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
fcc5a03a | 1280 | goto unlock; |
1da177e4 LT |
1281 | } |
1282 | ||
1283 | /** | |
1284 | * unregister_netdevice_notifier - unregister a network notifier block | |
1285 | * @nb: notifier | |
1286 | * | |
1287 | * Unregister a notifier previously registered by | |
1288 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1289 | * kernel structures and may then be reused. A negative errno code | |
1290 | * is returned on a failure. | |
1291 | */ | |
1292 | ||
1293 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1294 | { | |
9f514950 HX |
1295 | int err; |
1296 | ||
1297 | rtnl_lock(); | |
f07d5b94 | 1298 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
9f514950 HX |
1299 | rtnl_unlock(); |
1300 | return err; | |
1da177e4 LT |
1301 | } |
1302 | ||
1303 | /** | |
1304 | * call_netdevice_notifiers - call all network notifier blocks | |
1305 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1306 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1307 | * |
1308 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1309 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1310 | */ |
1311 | ||
ad7379d4 | 1312 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1313 | { |
ad7379d4 | 1314 | return raw_notifier_call_chain(&netdev_chain, val, dev); |
1da177e4 LT |
1315 | } |
1316 | ||
1317 | /* When > 0 there are consumers of rx skb time stamps */ | |
1318 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1319 | ||
1320 | void net_enable_timestamp(void) | |
1321 | { | |
1322 | atomic_inc(&netstamp_needed); | |
1323 | } | |
1324 | ||
1325 | void net_disable_timestamp(void) | |
1326 | { | |
1327 | atomic_dec(&netstamp_needed); | |
1328 | } | |
1329 | ||
a61bbcf2 | 1330 | static inline void net_timestamp(struct sk_buff *skb) |
1da177e4 LT |
1331 | { |
1332 | if (atomic_read(&netstamp_needed)) | |
a61bbcf2 | 1333 | __net_timestamp(skb); |
b7aa0bf7 ED |
1334 | else |
1335 | skb->tstamp.tv64 = 0; | |
1da177e4 LT |
1336 | } |
1337 | ||
1338 | /* | |
1339 | * Support routine. Sends outgoing frames to any network | |
1340 | * taps currently in use. | |
1341 | */ | |
1342 | ||
f6a78bfc | 1343 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1344 | { |
1345 | struct packet_type *ptype; | |
a61bbcf2 PM |
1346 | |
1347 | net_timestamp(skb); | |
1da177e4 LT |
1348 | |
1349 | rcu_read_lock(); | |
1350 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1351 | /* Never send packets back to the socket | |
1352 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1353 | */ | |
1354 | if ((ptype->dev == dev || !ptype->dev) && | |
1355 | (ptype->af_packet_priv == NULL || | |
1356 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
1357 | struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC); | |
1358 | if (!skb2) | |
1359 | break; | |
1360 | ||
1361 | /* skb->nh should be correctly | |
1362 | set by sender, so that the second statement is | |
1363 | just protection against buggy protocols. | |
1364 | */ | |
459a98ed | 1365 | skb_reset_mac_header(skb2); |
1da177e4 | 1366 | |
d56f90a7 | 1367 | if (skb_network_header(skb2) < skb2->data || |
27a884dc | 1368 | skb2->network_header > skb2->tail) { |
1da177e4 LT |
1369 | if (net_ratelimit()) |
1370 | printk(KERN_CRIT "protocol %04x is " | |
1371 | "buggy, dev %s\n", | |
1372 | skb2->protocol, dev->name); | |
c1d2bbe1 | 1373 | skb_reset_network_header(skb2); |
1da177e4 LT |
1374 | } |
1375 | ||
b0e380b1 | 1376 | skb2->transport_header = skb2->network_header; |
1da177e4 | 1377 | skb2->pkt_type = PACKET_OUTGOING; |
f2ccd8fa | 1378 | ptype->func(skb2, skb->dev, ptype, skb->dev); |
1da177e4 LT |
1379 | } |
1380 | } | |
1381 | rcu_read_unlock(); | |
1382 | } | |
1383 | ||
56079431 | 1384 | |
def82a1d | 1385 | static inline void __netif_reschedule(struct Qdisc *q) |
56079431 | 1386 | { |
def82a1d JP |
1387 | struct softnet_data *sd; |
1388 | unsigned long flags; | |
56079431 | 1389 | |
def82a1d JP |
1390 | local_irq_save(flags); |
1391 | sd = &__get_cpu_var(softnet_data); | |
1392 | q->next_sched = sd->output_queue; | |
1393 | sd->output_queue = q; | |
1394 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1395 | local_irq_restore(flags); | |
1396 | } | |
1397 | ||
1398 | void __netif_schedule(struct Qdisc *q) | |
1399 | { | |
1400 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
1401 | __netif_reschedule(q); | |
56079431 DV |
1402 | } |
1403 | EXPORT_SYMBOL(__netif_schedule); | |
1404 | ||
bea3348e | 1405 | void dev_kfree_skb_irq(struct sk_buff *skb) |
56079431 | 1406 | { |
bea3348e SH |
1407 | if (atomic_dec_and_test(&skb->users)) { |
1408 | struct softnet_data *sd; | |
1409 | unsigned long flags; | |
56079431 | 1410 | |
bea3348e SH |
1411 | local_irq_save(flags); |
1412 | sd = &__get_cpu_var(softnet_data); | |
1413 | skb->next = sd->completion_queue; | |
1414 | sd->completion_queue = skb; | |
1415 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1416 | local_irq_restore(flags); | |
1417 | } | |
56079431 | 1418 | } |
bea3348e | 1419 | EXPORT_SYMBOL(dev_kfree_skb_irq); |
56079431 DV |
1420 | |
1421 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1422 | { | |
1423 | if (in_irq() || irqs_disabled()) | |
1424 | dev_kfree_skb_irq(skb); | |
1425 | else | |
1426 | dev_kfree_skb(skb); | |
1427 | } | |
1428 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1429 | ||
1430 | ||
bea3348e SH |
1431 | /** |
1432 | * netif_device_detach - mark device as removed | |
1433 | * @dev: network device | |
1434 | * | |
1435 | * Mark device as removed from system and therefore no longer available. | |
1436 | */ | |
56079431 DV |
1437 | void netif_device_detach(struct net_device *dev) |
1438 | { | |
1439 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1440 | netif_running(dev)) { | |
1441 | netif_stop_queue(dev); | |
1442 | } | |
1443 | } | |
1444 | EXPORT_SYMBOL(netif_device_detach); | |
1445 | ||
bea3348e SH |
1446 | /** |
1447 | * netif_device_attach - mark device as attached | |
1448 | * @dev: network device | |
1449 | * | |
1450 | * Mark device as attached from system and restart if needed. | |
1451 | */ | |
56079431 DV |
1452 | void netif_device_attach(struct net_device *dev) |
1453 | { | |
1454 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1455 | netif_running(dev)) { | |
1456 | netif_wake_queue(dev); | |
4ec93edb | 1457 | __netdev_watchdog_up(dev); |
56079431 DV |
1458 | } |
1459 | } | |
1460 | EXPORT_SYMBOL(netif_device_attach); | |
1461 | ||
6de329e2 BH |
1462 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) |
1463 | { | |
1464 | return ((features & NETIF_F_GEN_CSUM) || | |
1465 | ((features & NETIF_F_IP_CSUM) && | |
1466 | protocol == htons(ETH_P_IP)) || | |
1467 | ((features & NETIF_F_IPV6_CSUM) && | |
1468 | protocol == htons(ETH_P_IPV6))); | |
1469 | } | |
1470 | ||
1471 | static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb) | |
1472 | { | |
1473 | if (can_checksum_protocol(dev->features, skb->protocol)) | |
1474 | return true; | |
1475 | ||
1476 | if (skb->protocol == htons(ETH_P_8021Q)) { | |
1477 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
1478 | if (can_checksum_protocol(dev->features & dev->vlan_features, | |
1479 | veh->h_vlan_encapsulated_proto)) | |
1480 | return true; | |
1481 | } | |
1482 | ||
1483 | return false; | |
1484 | } | |
56079431 | 1485 | |
1da177e4 LT |
1486 | /* |
1487 | * Invalidate hardware checksum when packet is to be mangled, and | |
1488 | * complete checksum manually on outgoing path. | |
1489 | */ | |
84fa7933 | 1490 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 1491 | { |
d3bc23e7 | 1492 | __wsum csum; |
663ead3b | 1493 | int ret = 0, offset; |
1da177e4 | 1494 | |
84fa7933 | 1495 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
1496 | goto out_set_summed; |
1497 | ||
1498 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
a430a43d HX |
1499 | /* Let GSO fix up the checksum. */ |
1500 | goto out_set_summed; | |
1da177e4 LT |
1501 | } |
1502 | ||
a030847e HX |
1503 | offset = skb->csum_start - skb_headroom(skb); |
1504 | BUG_ON(offset >= skb_headlen(skb)); | |
1505 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1506 | ||
1507 | offset += skb->csum_offset; | |
1508 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1509 | ||
1510 | if (skb_cloned(skb) && | |
1511 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
1512 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
1513 | if (ret) | |
1514 | goto out; | |
1515 | } | |
1516 | ||
a030847e | 1517 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); |
a430a43d | 1518 | out_set_summed: |
1da177e4 | 1519 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 1520 | out: |
1da177e4 LT |
1521 | return ret; |
1522 | } | |
1523 | ||
f6a78bfc HX |
1524 | /** |
1525 | * skb_gso_segment - Perform segmentation on skb. | |
1526 | * @skb: buffer to segment | |
576a30eb | 1527 | * @features: features for the output path (see dev->features) |
f6a78bfc HX |
1528 | * |
1529 | * This function segments the given skb and returns a list of segments. | |
576a30eb HX |
1530 | * |
1531 | * It may return NULL if the skb requires no segmentation. This is | |
1532 | * only possible when GSO is used for verifying header integrity. | |
f6a78bfc | 1533 | */ |
576a30eb | 1534 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) |
f6a78bfc HX |
1535 | { |
1536 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1537 | struct packet_type *ptype; | |
252e3346 | 1538 | __be16 type = skb->protocol; |
a430a43d | 1539 | int err; |
f6a78bfc | 1540 | |
459a98ed | 1541 | skb_reset_mac_header(skb); |
b0e380b1 | 1542 | skb->mac_len = skb->network_header - skb->mac_header; |
f6a78bfc HX |
1543 | __skb_pull(skb, skb->mac_len); |
1544 | ||
67fd1a73 HX |
1545 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
1546 | struct net_device *dev = skb->dev; | |
1547 | struct ethtool_drvinfo info = {}; | |
1548 | ||
1549 | if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo) | |
1550 | dev->ethtool_ops->get_drvinfo(dev, &info); | |
1551 | ||
1552 | WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d " | |
1553 | "ip_summed=%d", | |
1554 | info.driver, dev ? dev->features : 0L, | |
1555 | skb->sk ? skb->sk->sk_route_caps : 0L, | |
1556 | skb->len, skb->data_len, skb->ip_summed); | |
1557 | ||
a430a43d HX |
1558 | if (skb_header_cloned(skb) && |
1559 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1560 | return ERR_PTR(err); | |
1561 | } | |
1562 | ||
f6a78bfc | 1563 | rcu_read_lock(); |
82d8a867 PE |
1564 | list_for_each_entry_rcu(ptype, |
1565 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
f6a78bfc | 1566 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
84fa7933 | 1567 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { |
a430a43d HX |
1568 | err = ptype->gso_send_check(skb); |
1569 | segs = ERR_PTR(err); | |
1570 | if (err || skb_gso_ok(skb, features)) | |
1571 | break; | |
d56f90a7 ACM |
1572 | __skb_push(skb, (skb->data - |
1573 | skb_network_header(skb))); | |
a430a43d | 1574 | } |
576a30eb | 1575 | segs = ptype->gso_segment(skb, features); |
f6a78bfc HX |
1576 | break; |
1577 | } | |
1578 | } | |
1579 | rcu_read_unlock(); | |
1580 | ||
98e399f8 | 1581 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 1582 | |
f6a78bfc HX |
1583 | return segs; |
1584 | } | |
1585 | ||
1586 | EXPORT_SYMBOL(skb_gso_segment); | |
1587 | ||
fb286bb2 HX |
1588 | /* Take action when hardware reception checksum errors are detected. */ |
1589 | #ifdef CONFIG_BUG | |
1590 | void netdev_rx_csum_fault(struct net_device *dev) | |
1591 | { | |
1592 | if (net_ratelimit()) { | |
4ec93edb | 1593 | printk(KERN_ERR "%s: hw csum failure.\n", |
246a4212 | 1594 | dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
1595 | dump_stack(); |
1596 | } | |
1597 | } | |
1598 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1599 | #endif | |
1600 | ||
1da177e4 LT |
1601 | /* Actually, we should eliminate this check as soon as we know, that: |
1602 | * 1. IOMMU is present and allows to map all the memory. | |
1603 | * 2. No high memory really exists on this machine. | |
1604 | */ | |
1605 | ||
1606 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) | |
1607 | { | |
3d3a8533 | 1608 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
1609 | int i; |
1610 | ||
1611 | if (dev->features & NETIF_F_HIGHDMA) | |
1612 | return 0; | |
1613 | ||
1614 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1615 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1616 | return 1; | |
1617 | ||
3d3a8533 | 1618 | #endif |
1da177e4 LT |
1619 | return 0; |
1620 | } | |
1da177e4 | 1621 | |
f6a78bfc HX |
1622 | struct dev_gso_cb { |
1623 | void (*destructor)(struct sk_buff *skb); | |
1624 | }; | |
1625 | ||
1626 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1627 | ||
1628 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1629 | { | |
1630 | struct dev_gso_cb *cb; | |
1631 | ||
1632 | do { | |
1633 | struct sk_buff *nskb = skb->next; | |
1634 | ||
1635 | skb->next = nskb->next; | |
1636 | nskb->next = NULL; | |
1637 | kfree_skb(nskb); | |
1638 | } while (skb->next); | |
1639 | ||
1640 | cb = DEV_GSO_CB(skb); | |
1641 | if (cb->destructor) | |
1642 | cb->destructor(skb); | |
1643 | } | |
1644 | ||
1645 | /** | |
1646 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1647 | * @skb: buffer to segment | |
1648 | * | |
1649 | * This function segments the given skb and stores the list of segments | |
1650 | * in skb->next. | |
1651 | */ | |
1652 | static int dev_gso_segment(struct sk_buff *skb) | |
1653 | { | |
1654 | struct net_device *dev = skb->dev; | |
1655 | struct sk_buff *segs; | |
576a30eb HX |
1656 | int features = dev->features & ~(illegal_highdma(dev, skb) ? |
1657 | NETIF_F_SG : 0); | |
1658 | ||
1659 | segs = skb_gso_segment(skb, features); | |
1660 | ||
1661 | /* Verifying header integrity only. */ | |
1662 | if (!segs) | |
1663 | return 0; | |
f6a78bfc | 1664 | |
801678c5 | 1665 | if (IS_ERR(segs)) |
f6a78bfc HX |
1666 | return PTR_ERR(segs); |
1667 | ||
1668 | skb->next = segs; | |
1669 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1670 | skb->destructor = dev_gso_skb_destructor; | |
1671 | ||
1672 | return 0; | |
1673 | } | |
1674 | ||
ac45f602 PO |
1675 | static void tstamp_tx(struct sk_buff *skb) |
1676 | { | |
1677 | union skb_shared_tx *shtx = | |
1678 | skb_tx(skb); | |
1679 | if (unlikely(shtx->software && | |
1680 | !shtx->in_progress)) { | |
1681 | skb_tstamp_tx(skb, NULL); | |
1682 | } | |
1683 | } | |
1684 | ||
fd2ea0a7 DM |
1685 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, |
1686 | struct netdev_queue *txq) | |
f6a78bfc | 1687 | { |
00829823 | 1688 | const struct net_device_ops *ops = dev->netdev_ops; |
ac45f602 | 1689 | int rc; |
00829823 SH |
1690 | |
1691 | prefetch(&dev->netdev_ops->ndo_start_xmit); | |
f6a78bfc | 1692 | if (likely(!skb->next)) { |
9be9a6b9 | 1693 | if (!list_empty(&ptype_all)) |
f6a78bfc HX |
1694 | dev_queue_xmit_nit(skb, dev); |
1695 | ||
576a30eb HX |
1696 | if (netif_needs_gso(dev, skb)) { |
1697 | if (unlikely(dev_gso_segment(skb))) | |
1698 | goto out_kfree_skb; | |
1699 | if (skb->next) | |
1700 | goto gso; | |
1701 | } | |
f6a78bfc | 1702 | |
ac45f602 PO |
1703 | rc = ops->ndo_start_xmit(skb, dev); |
1704 | /* | |
1705 | * TODO: if skb_orphan() was called by | |
1706 | * dev->hard_start_xmit() (for example, the unmodified | |
1707 | * igb driver does that; bnx2 doesn't), then | |
1708 | * skb_tx_software_timestamp() will be unable to send | |
1709 | * back the time stamp. | |
1710 | * | |
1711 | * How can this be prevented? Always create another | |
1712 | * reference to the socket before calling | |
1713 | * dev->hard_start_xmit()? Prevent that skb_orphan() | |
1714 | * does anything in dev->hard_start_xmit() by clearing | |
1715 | * the skb destructor before the call and restoring it | |
1716 | * afterwards, then doing the skb_orphan() ourselves? | |
1717 | */ | |
1718 | if (likely(!rc)) | |
1719 | tstamp_tx(skb); | |
1720 | return rc; | |
f6a78bfc HX |
1721 | } |
1722 | ||
576a30eb | 1723 | gso: |
f6a78bfc HX |
1724 | do { |
1725 | struct sk_buff *nskb = skb->next; | |
f6a78bfc HX |
1726 | |
1727 | skb->next = nskb->next; | |
1728 | nskb->next = NULL; | |
00829823 | 1729 | rc = ops->ndo_start_xmit(nskb, dev); |
f6a78bfc | 1730 | if (unlikely(rc)) { |
f54d9e8d | 1731 | nskb->next = skb->next; |
f6a78bfc HX |
1732 | skb->next = nskb; |
1733 | return rc; | |
1734 | } | |
ac45f602 | 1735 | tstamp_tx(skb); |
fd2ea0a7 | 1736 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) |
f54d9e8d | 1737 | return NETDEV_TX_BUSY; |
f6a78bfc | 1738 | } while (skb->next); |
4ec93edb | 1739 | |
f6a78bfc HX |
1740 | skb->destructor = DEV_GSO_CB(skb)->destructor; |
1741 | ||
1742 | out_kfree_skb: | |
1743 | kfree_skb(skb); | |
1744 | return 0; | |
1745 | } | |
1746 | ||
7019298a DM |
1747 | static u32 skb_tx_hashrnd; |
1748 | static int skb_tx_hashrnd_initialized = 0; | |
b6b2fed1 | 1749 | |
7019298a | 1750 | static u16 skb_tx_hash(struct net_device *dev, struct sk_buff *skb) |
8f0f2223 | 1751 | { |
7019298a | 1752 | u32 hash; |
b6b2fed1 | 1753 | |
7019298a DM |
1754 | if (unlikely(!skb_tx_hashrnd_initialized)) { |
1755 | get_random_bytes(&skb_tx_hashrnd, 4); | |
1756 | skb_tx_hashrnd_initialized = 1; | |
b6b2fed1 | 1757 | } |
8f0f2223 | 1758 | |
d5a9e24a | 1759 | if (skb_rx_queue_recorded(skb)) { |
7019298a DM |
1760 | hash = skb_get_rx_queue(skb); |
1761 | } else if (skb->sk && skb->sk->sk_hash) { | |
1762 | hash = skb->sk->sk_hash; | |
1763 | } else | |
1764 | hash = skb->protocol; | |
d5a9e24a | 1765 | |
7019298a | 1766 | hash = jhash_1word(hash, skb_tx_hashrnd); |
b6b2fed1 DM |
1767 | |
1768 | return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32); | |
8f0f2223 DM |
1769 | } |
1770 | ||
e8a0464c DM |
1771 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
1772 | struct sk_buff *skb) | |
1773 | { | |
00829823 | 1774 | const struct net_device_ops *ops = dev->netdev_ops; |
fd2ea0a7 DM |
1775 | u16 queue_index = 0; |
1776 | ||
00829823 SH |
1777 | if (ops->ndo_select_queue) |
1778 | queue_index = ops->ndo_select_queue(dev, skb); | |
8f0f2223 | 1779 | else if (dev->real_num_tx_queues > 1) |
7019298a | 1780 | queue_index = skb_tx_hash(dev, skb); |
eae792b7 | 1781 | |
fd2ea0a7 DM |
1782 | skb_set_queue_mapping(skb, queue_index); |
1783 | return netdev_get_tx_queue(dev, queue_index); | |
e8a0464c DM |
1784 | } |
1785 | ||
d29f749e DJ |
1786 | /** |
1787 | * dev_queue_xmit - transmit a buffer | |
1788 | * @skb: buffer to transmit | |
1789 | * | |
1790 | * Queue a buffer for transmission to a network device. The caller must | |
1791 | * have set the device and priority and built the buffer before calling | |
1792 | * this function. The function can be called from an interrupt. | |
1793 | * | |
1794 | * A negative errno code is returned on a failure. A success does not | |
1795 | * guarantee the frame will be transmitted as it may be dropped due | |
1796 | * to congestion or traffic shaping. | |
1797 | * | |
1798 | * ----------------------------------------------------------------------------------- | |
1799 | * I notice this method can also return errors from the queue disciplines, | |
1800 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
1801 | * be positive. | |
1802 | * | |
1803 | * Regardless of the return value, the skb is consumed, so it is currently | |
1804 | * difficult to retry a send to this method. (You can bump the ref count | |
1805 | * before sending to hold a reference for retry if you are careful.) | |
1806 | * | |
1807 | * When calling this method, interrupts MUST be enabled. This is because | |
1808 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
1809 | * --BLG | |
1810 | */ | |
1da177e4 LT |
1811 | int dev_queue_xmit(struct sk_buff *skb) |
1812 | { | |
1813 | struct net_device *dev = skb->dev; | |
dc2b4847 | 1814 | struct netdev_queue *txq; |
1da177e4 LT |
1815 | struct Qdisc *q; |
1816 | int rc = -ENOMEM; | |
1817 | ||
f6a78bfc HX |
1818 | /* GSO will handle the following emulations directly. */ |
1819 | if (netif_needs_gso(dev, skb)) | |
1820 | goto gso; | |
1821 | ||
1da177e4 LT |
1822 | if (skb_shinfo(skb)->frag_list && |
1823 | !(dev->features & NETIF_F_FRAGLIST) && | |
364c6bad | 1824 | __skb_linearize(skb)) |
1da177e4 LT |
1825 | goto out_kfree_skb; |
1826 | ||
1827 | /* Fragmented skb is linearized if device does not support SG, | |
1828 | * or if at least one of fragments is in highmem and device | |
1829 | * does not support DMA from it. | |
1830 | */ | |
1831 | if (skb_shinfo(skb)->nr_frags && | |
1832 | (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) && | |
364c6bad | 1833 | __skb_linearize(skb)) |
1da177e4 LT |
1834 | goto out_kfree_skb; |
1835 | ||
1836 | /* If packet is not checksummed and device does not support | |
1837 | * checksumming for this protocol, complete checksumming here. | |
1838 | */ | |
663ead3b HX |
1839 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
1840 | skb_set_transport_header(skb, skb->csum_start - | |
1841 | skb_headroom(skb)); | |
6de329e2 BH |
1842 | if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb)) |
1843 | goto out_kfree_skb; | |
663ead3b | 1844 | } |
1da177e4 | 1845 | |
f6a78bfc | 1846 | gso: |
4ec93edb YH |
1847 | /* Disable soft irqs for various locks below. Also |
1848 | * stops preemption for RCU. | |
1da177e4 | 1849 | */ |
4ec93edb | 1850 | rcu_read_lock_bh(); |
1da177e4 | 1851 | |
eae792b7 | 1852 | txq = dev_pick_tx(dev, skb); |
b0e1e646 | 1853 | q = rcu_dereference(txq->qdisc); |
37437bb2 | 1854 | |
1da177e4 LT |
1855 | #ifdef CONFIG_NET_CLS_ACT |
1856 | skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS); | |
1857 | #endif | |
1858 | if (q->enqueue) { | |
5fb66229 | 1859 | spinlock_t *root_lock = qdisc_lock(q); |
37437bb2 DM |
1860 | |
1861 | spin_lock(root_lock); | |
1862 | ||
a9312ae8 | 1863 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { |
96d20316 | 1864 | kfree_skb(skb); |
a9312ae8 | 1865 | rc = NET_XMIT_DROP; |
96d20316 DM |
1866 | } else { |
1867 | rc = qdisc_enqueue_root(skb, q); | |
1868 | qdisc_run(q); | |
a9312ae8 | 1869 | } |
37437bb2 DM |
1870 | spin_unlock(root_lock); |
1871 | ||
37437bb2 | 1872 | goto out; |
1da177e4 LT |
1873 | } |
1874 | ||
1875 | /* The device has no queue. Common case for software devices: | |
1876 | loopback, all the sorts of tunnels... | |
1877 | ||
932ff279 HX |
1878 | Really, it is unlikely that netif_tx_lock protection is necessary |
1879 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1da177e4 LT |
1880 | counters.) |
1881 | However, it is possible, that they rely on protection | |
1882 | made by us here. | |
1883 | ||
1884 | Check this and shot the lock. It is not prone from deadlocks. | |
1885 | Either shot noqueue qdisc, it is even simpler 8) | |
1886 | */ | |
1887 | if (dev->flags & IFF_UP) { | |
1888 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
1889 | ||
c773e847 | 1890 | if (txq->xmit_lock_owner != cpu) { |
1da177e4 | 1891 | |
c773e847 | 1892 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 1893 | |
fd2ea0a7 | 1894 | if (!netif_tx_queue_stopped(txq)) { |
1da177e4 | 1895 | rc = 0; |
fd2ea0a7 | 1896 | if (!dev_hard_start_xmit(skb, dev, txq)) { |
c773e847 | 1897 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
1898 | goto out; |
1899 | } | |
1900 | } | |
c773e847 | 1901 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
1902 | if (net_ratelimit()) |
1903 | printk(KERN_CRIT "Virtual device %s asks to " | |
1904 | "queue packet!\n", dev->name); | |
1905 | } else { | |
1906 | /* Recursion is detected! It is possible, | |
1907 | * unfortunately */ | |
1908 | if (net_ratelimit()) | |
1909 | printk(KERN_CRIT "Dead loop on virtual device " | |
1910 | "%s, fix it urgently!\n", dev->name); | |
1911 | } | |
1912 | } | |
1913 | ||
1914 | rc = -ENETDOWN; | |
d4828d85 | 1915 | rcu_read_unlock_bh(); |
1da177e4 LT |
1916 | |
1917 | out_kfree_skb: | |
1918 | kfree_skb(skb); | |
1919 | return rc; | |
1920 | out: | |
d4828d85 | 1921 | rcu_read_unlock_bh(); |
1da177e4 LT |
1922 | return rc; |
1923 | } | |
1924 | ||
1925 | ||
1926 | /*======================================================================= | |
1927 | Receiver routines | |
1928 | =======================================================================*/ | |
1929 | ||
6b2bedc3 SH |
1930 | int netdev_max_backlog __read_mostly = 1000; |
1931 | int netdev_budget __read_mostly = 300; | |
1932 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1da177e4 LT |
1933 | |
1934 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; | |
1935 | ||
1936 | ||
1da177e4 LT |
1937 | /** |
1938 | * netif_rx - post buffer to the network code | |
1939 | * @skb: buffer to post | |
1940 | * | |
1941 | * This function receives a packet from a device driver and queues it for | |
1942 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
1943 | * may be dropped during processing for congestion control or by the | |
1944 | * protocol layers. | |
1945 | * | |
1946 | * return values: | |
1947 | * NET_RX_SUCCESS (no congestion) | |
1da177e4 LT |
1948 | * NET_RX_DROP (packet was dropped) |
1949 | * | |
1950 | */ | |
1951 | ||
1952 | int netif_rx(struct sk_buff *skb) | |
1953 | { | |
1da177e4 LT |
1954 | struct softnet_data *queue; |
1955 | unsigned long flags; | |
1956 | ||
1957 | /* if netpoll wants it, pretend we never saw it */ | |
1958 | if (netpoll_rx(skb)) | |
1959 | return NET_RX_DROP; | |
1960 | ||
b7aa0bf7 | 1961 | if (!skb->tstamp.tv64) |
a61bbcf2 | 1962 | net_timestamp(skb); |
1da177e4 LT |
1963 | |
1964 | /* | |
1965 | * The code is rearranged so that the path is the most | |
1966 | * short when CPU is congested, but is still operating. | |
1967 | */ | |
1968 | local_irq_save(flags); | |
1da177e4 LT |
1969 | queue = &__get_cpu_var(softnet_data); |
1970 | ||
1971 | __get_cpu_var(netdev_rx_stat).total++; | |
1972 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { | |
1973 | if (queue->input_pkt_queue.qlen) { | |
1da177e4 | 1974 | enqueue: |
1da177e4 | 1975 | __skb_queue_tail(&queue->input_pkt_queue, skb); |
1da177e4 | 1976 | local_irq_restore(flags); |
34008d8c | 1977 | return NET_RX_SUCCESS; |
1da177e4 LT |
1978 | } |
1979 | ||
bea3348e | 1980 | napi_schedule(&queue->backlog); |
1da177e4 LT |
1981 | goto enqueue; |
1982 | } | |
1983 | ||
1da177e4 LT |
1984 | __get_cpu_var(netdev_rx_stat).dropped++; |
1985 | local_irq_restore(flags); | |
1986 | ||
1987 | kfree_skb(skb); | |
1988 | return NET_RX_DROP; | |
1989 | } | |
1990 | ||
1991 | int netif_rx_ni(struct sk_buff *skb) | |
1992 | { | |
1993 | int err; | |
1994 | ||
1995 | preempt_disable(); | |
1996 | err = netif_rx(skb); | |
1997 | if (local_softirq_pending()) | |
1998 | do_softirq(); | |
1999 | preempt_enable(); | |
2000 | ||
2001 | return err; | |
2002 | } | |
2003 | ||
2004 | EXPORT_SYMBOL(netif_rx_ni); | |
2005 | ||
1da177e4 LT |
2006 | static void net_tx_action(struct softirq_action *h) |
2007 | { | |
2008 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
2009 | ||
2010 | if (sd->completion_queue) { | |
2011 | struct sk_buff *clist; | |
2012 | ||
2013 | local_irq_disable(); | |
2014 | clist = sd->completion_queue; | |
2015 | sd->completion_queue = NULL; | |
2016 | local_irq_enable(); | |
2017 | ||
2018 | while (clist) { | |
2019 | struct sk_buff *skb = clist; | |
2020 | clist = clist->next; | |
2021 | ||
547b792c | 2022 | WARN_ON(atomic_read(&skb->users)); |
1da177e4 LT |
2023 | __kfree_skb(skb); |
2024 | } | |
2025 | } | |
2026 | ||
2027 | if (sd->output_queue) { | |
37437bb2 | 2028 | struct Qdisc *head; |
1da177e4 LT |
2029 | |
2030 | local_irq_disable(); | |
2031 | head = sd->output_queue; | |
2032 | sd->output_queue = NULL; | |
2033 | local_irq_enable(); | |
2034 | ||
2035 | while (head) { | |
37437bb2 DM |
2036 | struct Qdisc *q = head; |
2037 | spinlock_t *root_lock; | |
2038 | ||
1da177e4 LT |
2039 | head = head->next_sched; |
2040 | ||
5fb66229 | 2041 | root_lock = qdisc_lock(q); |
37437bb2 | 2042 | if (spin_trylock(root_lock)) { |
def82a1d JP |
2043 | smp_mb__before_clear_bit(); |
2044 | clear_bit(__QDISC_STATE_SCHED, | |
2045 | &q->state); | |
37437bb2 DM |
2046 | qdisc_run(q); |
2047 | spin_unlock(root_lock); | |
1da177e4 | 2048 | } else { |
195648bb | 2049 | if (!test_bit(__QDISC_STATE_DEACTIVATED, |
e8a83e10 | 2050 | &q->state)) { |
195648bb | 2051 | __netif_reschedule(q); |
e8a83e10 JP |
2052 | } else { |
2053 | smp_mb__before_clear_bit(); | |
2054 | clear_bit(__QDISC_STATE_SCHED, | |
2055 | &q->state); | |
2056 | } | |
1da177e4 LT |
2057 | } |
2058 | } | |
2059 | } | |
2060 | } | |
2061 | ||
6f05f629 SH |
2062 | static inline int deliver_skb(struct sk_buff *skb, |
2063 | struct packet_type *pt_prev, | |
2064 | struct net_device *orig_dev) | |
1da177e4 LT |
2065 | { |
2066 | atomic_inc(&skb->users); | |
f2ccd8fa | 2067 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2068 | } |
2069 | ||
2070 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) | |
6229e362 | 2071 | /* These hooks defined here for ATM */ |
1da177e4 LT |
2072 | struct net_bridge; |
2073 | struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br, | |
2074 | unsigned char *addr); | |
6229e362 | 2075 | void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly; |
1da177e4 | 2076 | |
6229e362 SH |
2077 | /* |
2078 | * If bridge module is loaded call bridging hook. | |
2079 | * returns NULL if packet was consumed. | |
2080 | */ | |
2081 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, | |
2082 | struct sk_buff *skb) __read_mostly; | |
2083 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb, | |
2084 | struct packet_type **pt_prev, int *ret, | |
2085 | struct net_device *orig_dev) | |
1da177e4 LT |
2086 | { |
2087 | struct net_bridge_port *port; | |
2088 | ||
6229e362 SH |
2089 | if (skb->pkt_type == PACKET_LOOPBACK || |
2090 | (port = rcu_dereference(skb->dev->br_port)) == NULL) | |
2091 | return skb; | |
1da177e4 LT |
2092 | |
2093 | if (*pt_prev) { | |
6229e362 | 2094 | *ret = deliver_skb(skb, *pt_prev, orig_dev); |
1da177e4 | 2095 | *pt_prev = NULL; |
4ec93edb YH |
2096 | } |
2097 | ||
6229e362 | 2098 | return br_handle_frame_hook(port, skb); |
1da177e4 LT |
2099 | } |
2100 | #else | |
6229e362 | 2101 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) |
1da177e4 LT |
2102 | #endif |
2103 | ||
b863ceb7 PM |
2104 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) |
2105 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; | |
2106 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); | |
2107 | ||
2108 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, | |
2109 | struct packet_type **pt_prev, | |
2110 | int *ret, | |
2111 | struct net_device *orig_dev) | |
2112 | { | |
2113 | if (skb->dev->macvlan_port == NULL) | |
2114 | return skb; | |
2115 | ||
2116 | if (*pt_prev) { | |
2117 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2118 | *pt_prev = NULL; | |
2119 | } | |
2120 | return macvlan_handle_frame_hook(skb); | |
2121 | } | |
2122 | #else | |
2123 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) | |
2124 | #endif | |
2125 | ||
1da177e4 LT |
2126 | #ifdef CONFIG_NET_CLS_ACT |
2127 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2128 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2129 | * a compare and 2 stores extra right now if we dont have it on | |
2130 | * but have CONFIG_NET_CLS_ACT | |
4ec93edb | 2131 | * NOTE: This doesnt stop any functionality; if you dont have |
1da177e4 LT |
2132 | * the ingress scheduler, you just cant add policies on ingress. |
2133 | * | |
2134 | */ | |
4ec93edb | 2135 | static int ing_filter(struct sk_buff *skb) |
1da177e4 | 2136 | { |
1da177e4 | 2137 | struct net_device *dev = skb->dev; |
f697c3e8 | 2138 | u32 ttl = G_TC_RTTL(skb->tc_verd); |
555353cf DM |
2139 | struct netdev_queue *rxq; |
2140 | int result = TC_ACT_OK; | |
2141 | struct Qdisc *q; | |
4ec93edb | 2142 | |
f697c3e8 HX |
2143 | if (MAX_RED_LOOP < ttl++) { |
2144 | printk(KERN_WARNING | |
2145 | "Redir loop detected Dropping packet (%d->%d)\n", | |
2146 | skb->iif, dev->ifindex); | |
2147 | return TC_ACT_SHOT; | |
2148 | } | |
1da177e4 | 2149 | |
f697c3e8 HX |
2150 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); |
2151 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
1da177e4 | 2152 | |
555353cf DM |
2153 | rxq = &dev->rx_queue; |
2154 | ||
83874000 | 2155 | q = rxq->qdisc; |
8d50b53d | 2156 | if (q != &noop_qdisc) { |
83874000 | 2157 | spin_lock(qdisc_lock(q)); |
a9312ae8 DM |
2158 | if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) |
2159 | result = qdisc_enqueue_root(skb, q); | |
83874000 DM |
2160 | spin_unlock(qdisc_lock(q)); |
2161 | } | |
f697c3e8 HX |
2162 | |
2163 | return result; | |
2164 | } | |
86e65da9 | 2165 | |
f697c3e8 HX |
2166 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, |
2167 | struct packet_type **pt_prev, | |
2168 | int *ret, struct net_device *orig_dev) | |
2169 | { | |
8d50b53d | 2170 | if (skb->dev->rx_queue.qdisc == &noop_qdisc) |
f697c3e8 | 2171 | goto out; |
1da177e4 | 2172 | |
f697c3e8 HX |
2173 | if (*pt_prev) { |
2174 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2175 | *pt_prev = NULL; | |
2176 | } else { | |
2177 | /* Huh? Why does turning on AF_PACKET affect this? */ | |
2178 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); | |
1da177e4 LT |
2179 | } |
2180 | ||
f697c3e8 HX |
2181 | switch (ing_filter(skb)) { |
2182 | case TC_ACT_SHOT: | |
2183 | case TC_ACT_STOLEN: | |
2184 | kfree_skb(skb); | |
2185 | return NULL; | |
2186 | } | |
2187 | ||
2188 | out: | |
2189 | skb->tc_verd = 0; | |
2190 | return skb; | |
1da177e4 LT |
2191 | } |
2192 | #endif | |
2193 | ||
bc1d0411 PM |
2194 | /* |
2195 | * netif_nit_deliver - deliver received packets to network taps | |
2196 | * @skb: buffer | |
2197 | * | |
2198 | * This function is used to deliver incoming packets to network | |
2199 | * taps. It should be used when the normal netif_receive_skb path | |
2200 | * is bypassed, for example because of VLAN acceleration. | |
2201 | */ | |
2202 | void netif_nit_deliver(struct sk_buff *skb) | |
2203 | { | |
2204 | struct packet_type *ptype; | |
2205 | ||
2206 | if (list_empty(&ptype_all)) | |
2207 | return; | |
2208 | ||
2209 | skb_reset_network_header(skb); | |
2210 | skb_reset_transport_header(skb); | |
2211 | skb->mac_len = skb->network_header - skb->mac_header; | |
2212 | ||
2213 | rcu_read_lock(); | |
2214 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2215 | if (!ptype->dev || ptype->dev == skb->dev) | |
2216 | deliver_skb(skb, ptype, skb->dev); | |
2217 | } | |
2218 | rcu_read_unlock(); | |
2219 | } | |
2220 | ||
3b582cc1 SH |
2221 | /** |
2222 | * netif_receive_skb - process receive buffer from network | |
2223 | * @skb: buffer to process | |
2224 | * | |
2225 | * netif_receive_skb() is the main receive data processing function. | |
2226 | * It always succeeds. The buffer may be dropped during processing | |
2227 | * for congestion control or by the protocol layers. | |
2228 | * | |
2229 | * This function may only be called from softirq context and interrupts | |
2230 | * should be enabled. | |
2231 | * | |
2232 | * Return values (usually ignored): | |
2233 | * NET_RX_SUCCESS: no congestion | |
2234 | * NET_RX_DROP: packet was dropped | |
2235 | */ | |
1da177e4 LT |
2236 | int netif_receive_skb(struct sk_buff *skb) |
2237 | { | |
2238 | struct packet_type *ptype, *pt_prev; | |
f2ccd8fa | 2239 | struct net_device *orig_dev; |
0d7a3681 | 2240 | struct net_device *null_or_orig; |
1da177e4 | 2241 | int ret = NET_RX_DROP; |
252e3346 | 2242 | __be16 type; |
1da177e4 | 2243 | |
9b22ea56 PM |
2244 | if (skb->vlan_tci && vlan_hwaccel_do_receive(skb)) |
2245 | return NET_RX_SUCCESS; | |
2246 | ||
1da177e4 | 2247 | /* if we've gotten here through NAPI, check netpoll */ |
bea3348e | 2248 | if (netpoll_receive_skb(skb)) |
1da177e4 LT |
2249 | return NET_RX_DROP; |
2250 | ||
b7aa0bf7 | 2251 | if (!skb->tstamp.tv64) |
a61bbcf2 | 2252 | net_timestamp(skb); |
1da177e4 | 2253 | |
c01003c2 PM |
2254 | if (!skb->iif) |
2255 | skb->iif = skb->dev->ifindex; | |
86e65da9 | 2256 | |
0d7a3681 | 2257 | null_or_orig = NULL; |
cc9bd5ce JE |
2258 | orig_dev = skb->dev; |
2259 | if (orig_dev->master) { | |
0d7a3681 JE |
2260 | if (skb_bond_should_drop(skb)) |
2261 | null_or_orig = orig_dev; /* deliver only exact match */ | |
2262 | else | |
2263 | skb->dev = orig_dev->master; | |
cc9bd5ce | 2264 | } |
8f903c70 | 2265 | |
1da177e4 LT |
2266 | __get_cpu_var(netdev_rx_stat).total++; |
2267 | ||
c1d2bbe1 | 2268 | skb_reset_network_header(skb); |
badff6d0 | 2269 | skb_reset_transport_header(skb); |
b0e380b1 | 2270 | skb->mac_len = skb->network_header - skb->mac_header; |
1da177e4 LT |
2271 | |
2272 | pt_prev = NULL; | |
2273 | ||
2274 | rcu_read_lock(); | |
2275 | ||
b9f75f45 | 2276 | /* Don't receive packets in an exiting network namespace */ |
0a36b345 EB |
2277 | if (!net_alive(dev_net(skb->dev))) { |
2278 | kfree_skb(skb); | |
b9f75f45 | 2279 | goto out; |
0a36b345 | 2280 | } |
b9f75f45 | 2281 | |
1da177e4 LT |
2282 | #ifdef CONFIG_NET_CLS_ACT |
2283 | if (skb->tc_verd & TC_NCLS) { | |
2284 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2285 | goto ncls; | |
2286 | } | |
2287 | #endif | |
2288 | ||
2289 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
f982307f JE |
2290 | if (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
2291 | ptype->dev == orig_dev) { | |
4ec93edb | 2292 | if (pt_prev) |
f2ccd8fa | 2293 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2294 | pt_prev = ptype; |
2295 | } | |
2296 | } | |
2297 | ||
2298 | #ifdef CONFIG_NET_CLS_ACT | |
f697c3e8 HX |
2299 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); |
2300 | if (!skb) | |
1da177e4 | 2301 | goto out; |
1da177e4 LT |
2302 | ncls: |
2303 | #endif | |
2304 | ||
6229e362 | 2305 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); |
b863ceb7 PM |
2306 | if (!skb) |
2307 | goto out; | |
2308 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); | |
6229e362 | 2309 | if (!skb) |
1da177e4 LT |
2310 | goto out; |
2311 | ||
9a279bcb HX |
2312 | skb_orphan(skb); |
2313 | ||
1da177e4 | 2314 | type = skb->protocol; |
82d8a867 PE |
2315 | list_for_each_entry_rcu(ptype, |
2316 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1da177e4 | 2317 | if (ptype->type == type && |
f982307f JE |
2318 | (ptype->dev == null_or_orig || ptype->dev == skb->dev || |
2319 | ptype->dev == orig_dev)) { | |
4ec93edb | 2320 | if (pt_prev) |
f2ccd8fa | 2321 | ret = deliver_skb(skb, pt_prev, orig_dev); |
1da177e4 LT |
2322 | pt_prev = ptype; |
2323 | } | |
2324 | } | |
2325 | ||
2326 | if (pt_prev) { | |
f2ccd8fa | 2327 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1da177e4 LT |
2328 | } else { |
2329 | kfree_skb(skb); | |
2330 | /* Jamal, now you will not able to escape explaining | |
2331 | * me how you were going to use this. :-) | |
2332 | */ | |
2333 | ret = NET_RX_DROP; | |
2334 | } | |
2335 | ||
2336 | out: | |
2337 | rcu_read_unlock(); | |
2338 | return ret; | |
2339 | } | |
2340 | ||
6e583ce5 SH |
2341 | /* Network device is going away, flush any packets still pending */ |
2342 | static void flush_backlog(void *arg) | |
2343 | { | |
2344 | struct net_device *dev = arg; | |
2345 | struct softnet_data *queue = &__get_cpu_var(softnet_data); | |
2346 | struct sk_buff *skb, *tmp; | |
2347 | ||
2348 | skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp) | |
2349 | if (skb->dev == dev) { | |
2350 | __skb_unlink(skb, &queue->input_pkt_queue); | |
2351 | kfree_skb(skb); | |
2352 | } | |
2353 | } | |
2354 | ||
d565b0a1 HX |
2355 | static int napi_gro_complete(struct sk_buff *skb) |
2356 | { | |
2357 | struct packet_type *ptype; | |
2358 | __be16 type = skb->protocol; | |
2359 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
2360 | int err = -ENOENT; | |
2361 | ||
5d38a079 | 2362 | if (NAPI_GRO_CB(skb)->count == 1) |
d565b0a1 HX |
2363 | goto out; |
2364 | ||
2365 | rcu_read_lock(); | |
2366 | list_for_each_entry_rcu(ptype, head, list) { | |
2367 | if (ptype->type != type || ptype->dev || !ptype->gro_complete) | |
2368 | continue; | |
2369 | ||
2370 | err = ptype->gro_complete(skb); | |
2371 | break; | |
2372 | } | |
2373 | rcu_read_unlock(); | |
2374 | ||
2375 | if (err) { | |
2376 | WARN_ON(&ptype->list == head); | |
2377 | kfree_skb(skb); | |
2378 | return NET_RX_SUCCESS; | |
2379 | } | |
2380 | ||
2381 | out: | |
b530256d | 2382 | skb_shinfo(skb)->gso_size = 0; |
d565b0a1 HX |
2383 | return netif_receive_skb(skb); |
2384 | } | |
2385 | ||
2386 | void napi_gro_flush(struct napi_struct *napi) | |
2387 | { | |
2388 | struct sk_buff *skb, *next; | |
2389 | ||
2390 | for (skb = napi->gro_list; skb; skb = next) { | |
2391 | next = skb->next; | |
2392 | skb->next = NULL; | |
2393 | napi_gro_complete(skb); | |
2394 | } | |
2395 | ||
4ae5544f | 2396 | napi->gro_count = 0; |
d565b0a1 HX |
2397 | napi->gro_list = NULL; |
2398 | } | |
2399 | EXPORT_SYMBOL(napi_gro_flush); | |
2400 | ||
86911732 HX |
2401 | void *skb_gro_header(struct sk_buff *skb, unsigned int hlen) |
2402 | { | |
2403 | unsigned int offset = skb_gro_offset(skb); | |
2404 | ||
2405 | hlen += offset; | |
2406 | if (hlen <= skb_headlen(skb)) | |
2407 | return skb->data + offset; | |
2408 | ||
2409 | if (unlikely(!skb_shinfo(skb)->nr_frags || | |
2410 | skb_shinfo(skb)->frags[0].size <= | |
2411 | hlen - skb_headlen(skb) || | |
2412 | PageHighMem(skb_shinfo(skb)->frags[0].page))) | |
2413 | return pskb_may_pull(skb, hlen) ? skb->data + offset : NULL; | |
2414 | ||
2415 | return page_address(skb_shinfo(skb)->frags[0].page) + | |
56035022 HX |
2416 | skb_shinfo(skb)->frags[0].page_offset + |
2417 | offset - skb_headlen(skb); | |
86911732 HX |
2418 | } |
2419 | EXPORT_SYMBOL(skb_gro_header); | |
2420 | ||
96e93eab | 2421 | int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
2422 | { |
2423 | struct sk_buff **pp = NULL; | |
2424 | struct packet_type *ptype; | |
2425 | __be16 type = skb->protocol; | |
2426 | struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK]; | |
0da2afd5 | 2427 | int same_flow; |
d565b0a1 | 2428 | int mac_len; |
5d0d9be8 | 2429 | int ret; |
d565b0a1 HX |
2430 | |
2431 | if (!(skb->dev->features & NETIF_F_GRO)) | |
2432 | goto normal; | |
2433 | ||
f17f5c91 HX |
2434 | if (skb_is_gso(skb) || skb_shinfo(skb)->frag_list) |
2435 | goto normal; | |
2436 | ||
d565b0a1 HX |
2437 | rcu_read_lock(); |
2438 | list_for_each_entry_rcu(ptype, head, list) { | |
d565b0a1 HX |
2439 | if (ptype->type != type || ptype->dev || !ptype->gro_receive) |
2440 | continue; | |
2441 | ||
86911732 | 2442 | skb_set_network_header(skb, skb_gro_offset(skb)); |
d565b0a1 HX |
2443 | mac_len = skb->network_header - skb->mac_header; |
2444 | skb->mac_len = mac_len; | |
2445 | NAPI_GRO_CB(skb)->same_flow = 0; | |
2446 | NAPI_GRO_CB(skb)->flush = 0; | |
5d38a079 | 2447 | NAPI_GRO_CB(skb)->free = 0; |
d565b0a1 | 2448 | |
d565b0a1 HX |
2449 | pp = ptype->gro_receive(&napi->gro_list, skb); |
2450 | break; | |
2451 | } | |
2452 | rcu_read_unlock(); | |
2453 | ||
2454 | if (&ptype->list == head) | |
2455 | goto normal; | |
2456 | ||
0da2afd5 | 2457 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 2458 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 2459 | |
d565b0a1 HX |
2460 | if (pp) { |
2461 | struct sk_buff *nskb = *pp; | |
2462 | ||
2463 | *pp = nskb->next; | |
2464 | nskb->next = NULL; | |
2465 | napi_gro_complete(nskb); | |
4ae5544f | 2466 | napi->gro_count--; |
d565b0a1 HX |
2467 | } |
2468 | ||
0da2afd5 | 2469 | if (same_flow) |
d565b0a1 HX |
2470 | goto ok; |
2471 | ||
4ae5544f | 2472 | if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS) |
d565b0a1 | 2473 | goto normal; |
d565b0a1 | 2474 | |
4ae5544f | 2475 | napi->gro_count++; |
d565b0a1 | 2476 | NAPI_GRO_CB(skb)->count = 1; |
86911732 | 2477 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
2478 | skb->next = napi->gro_list; |
2479 | napi->gro_list = skb; | |
5d0d9be8 | 2480 | ret = GRO_HELD; |
d565b0a1 | 2481 | |
ad0f9904 HX |
2482 | pull: |
2483 | if (unlikely(!pskb_may_pull(skb, skb_gro_offset(skb)))) { | |
2484 | if (napi->gro_list == skb) | |
2485 | napi->gro_list = skb->next; | |
2486 | ret = GRO_DROP; | |
2487 | } | |
2488 | ||
d565b0a1 | 2489 | ok: |
5d0d9be8 | 2490 | return ret; |
d565b0a1 HX |
2491 | |
2492 | normal: | |
ad0f9904 HX |
2493 | ret = GRO_NORMAL; |
2494 | goto pull; | |
5d38a079 | 2495 | } |
96e93eab HX |
2496 | EXPORT_SYMBOL(dev_gro_receive); |
2497 | ||
2498 | static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
2499 | { | |
2500 | struct sk_buff *p; | |
2501 | ||
2502 | for (p = napi->gro_list; p; p = p->next) { | |
aa4b9f53 HX |
2503 | NAPI_GRO_CB(p)->same_flow = !compare_ether_header( |
2504 | skb_mac_header(p), skb_gro_mac_header(skb)); | |
96e93eab HX |
2505 | NAPI_GRO_CB(p)->flush = 0; |
2506 | } | |
2507 | ||
2508 | return dev_gro_receive(napi, skb); | |
2509 | } | |
5d38a079 | 2510 | |
5d0d9be8 | 2511 | int napi_skb_finish(int ret, struct sk_buff *skb) |
5d38a079 | 2512 | { |
5d0d9be8 HX |
2513 | int err = NET_RX_SUCCESS; |
2514 | ||
2515 | switch (ret) { | |
2516 | case GRO_NORMAL: | |
5d38a079 HX |
2517 | return netif_receive_skb(skb); |
2518 | ||
5d0d9be8 HX |
2519 | case GRO_DROP: |
2520 | err = NET_RX_DROP; | |
2521 | /* fall through */ | |
2522 | ||
2523 | case GRO_MERGED_FREE: | |
5d38a079 HX |
2524 | kfree_skb(skb); |
2525 | break; | |
2526 | } | |
2527 | ||
5d0d9be8 HX |
2528 | return err; |
2529 | } | |
2530 | EXPORT_SYMBOL(napi_skb_finish); | |
2531 | ||
2532 | int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) | |
2533 | { | |
86911732 HX |
2534 | skb_gro_reset_offset(skb); |
2535 | ||
5d0d9be8 | 2536 | return napi_skb_finish(__napi_gro_receive(napi, skb), skb); |
d565b0a1 HX |
2537 | } |
2538 | EXPORT_SYMBOL(napi_gro_receive); | |
2539 | ||
96e93eab HX |
2540 | void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
2541 | { | |
96e93eab HX |
2542 | __skb_pull(skb, skb_headlen(skb)); |
2543 | skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb)); | |
2544 | ||
2545 | napi->skb = skb; | |
2546 | } | |
2547 | EXPORT_SYMBOL(napi_reuse_skb); | |
2548 | ||
2549 | struct sk_buff *napi_fraginfo_skb(struct napi_struct *napi, | |
2550 | struct napi_gro_fraginfo *info) | |
5d38a079 HX |
2551 | { |
2552 | struct net_device *dev = napi->dev; | |
2553 | struct sk_buff *skb = napi->skb; | |
86911732 | 2554 | struct ethhdr *eth; |
80595d59 HX |
2555 | skb_frag_t *frag; |
2556 | int i; | |
5d38a079 HX |
2557 | |
2558 | napi->skb = NULL; | |
2559 | ||
2560 | if (!skb) { | |
2561 | skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN); | |
2562 | if (!skb) | |
2563 | goto out; | |
2564 | ||
2565 | skb_reserve(skb, NET_IP_ALIGN); | |
2566 | } | |
2567 | ||
2568 | BUG_ON(info->nr_frags > MAX_SKB_FRAGS); | |
80595d59 HX |
2569 | frag = &info->frags[info->nr_frags - 1]; |
2570 | ||
2571 | for (i = skb_shinfo(skb)->nr_frags; i < info->nr_frags; i++) { | |
2572 | skb_fill_page_desc(skb, i, frag->page, frag->page_offset, | |
2573 | frag->size); | |
2574 | frag++; | |
2575 | } | |
5d38a079 | 2576 | skb_shinfo(skb)->nr_frags = info->nr_frags; |
5d38a079 HX |
2577 | |
2578 | skb->data_len = info->len; | |
2579 | skb->len += info->len; | |
2580 | skb->truesize += info->len; | |
2581 | ||
86911732 HX |
2582 | skb_reset_mac_header(skb); |
2583 | skb_gro_reset_offset(skb); | |
2584 | ||
2585 | eth = skb_gro_header(skb, sizeof(*eth)); | |
2586 | if (!eth) { | |
96e93eab | 2587 | napi_reuse_skb(napi, skb); |
9a8e47ff | 2588 | skb = NULL; |
96e93eab HX |
2589 | goto out; |
2590 | } | |
5d38a079 | 2591 | |
86911732 HX |
2592 | skb_gro_pull(skb, sizeof(*eth)); |
2593 | ||
2594 | /* | |
2595 | * This works because the only protocols we care about don't require | |
2596 | * special handling. We'll fix it up properly at the end. | |
2597 | */ | |
2598 | skb->protocol = eth->h_proto; | |
5d38a079 HX |
2599 | |
2600 | skb->ip_summed = info->ip_summed; | |
2601 | skb->csum = info->csum; | |
2602 | ||
96e93eab HX |
2603 | out: |
2604 | return skb; | |
2605 | } | |
2606 | EXPORT_SYMBOL(napi_fraginfo_skb); | |
2607 | ||
5d0d9be8 | 2608 | int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret) |
96e93eab | 2609 | { |
5d0d9be8 | 2610 | int err = NET_RX_SUCCESS; |
96e93eab | 2611 | |
5d0d9be8 HX |
2612 | switch (ret) { |
2613 | case GRO_NORMAL: | |
86911732 | 2614 | case GRO_HELD: |
86911732 HX |
2615 | skb->protocol = eth_type_trans(skb, napi->dev); |
2616 | ||
2617 | if (ret == GRO_NORMAL) | |
2618 | return netif_receive_skb(skb); | |
2619 | ||
2620 | skb_gro_pull(skb, -ETH_HLEN); | |
2621 | break; | |
5d38a079 | 2622 | |
5d0d9be8 HX |
2623 | case GRO_DROP: |
2624 | err = NET_RX_DROP; | |
2625 | /* fall through */ | |
5d38a079 | 2626 | |
5d0d9be8 HX |
2627 | case GRO_MERGED_FREE: |
2628 | napi_reuse_skb(napi, skb); | |
2629 | break; | |
2630 | } | |
5d38a079 | 2631 | |
5d38a079 HX |
2632 | return err; |
2633 | } | |
5d0d9be8 HX |
2634 | EXPORT_SYMBOL(napi_frags_finish); |
2635 | ||
2636 | int napi_gro_frags(struct napi_struct *napi, struct napi_gro_fraginfo *info) | |
2637 | { | |
2638 | struct sk_buff *skb = napi_fraginfo_skb(napi, info); | |
2639 | ||
2640 | if (!skb) | |
2641 | return NET_RX_DROP; | |
2642 | ||
2643 | return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb)); | |
2644 | } | |
5d38a079 HX |
2645 | EXPORT_SYMBOL(napi_gro_frags); |
2646 | ||
bea3348e | 2647 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 LT |
2648 | { |
2649 | int work = 0; | |
1da177e4 LT |
2650 | struct softnet_data *queue = &__get_cpu_var(softnet_data); |
2651 | unsigned long start_time = jiffies; | |
2652 | ||
bea3348e SH |
2653 | napi->weight = weight_p; |
2654 | do { | |
1da177e4 | 2655 | struct sk_buff *skb; |
1da177e4 LT |
2656 | |
2657 | local_irq_disable(); | |
2658 | skb = __skb_dequeue(&queue->input_pkt_queue); | |
bea3348e SH |
2659 | if (!skb) { |
2660 | __napi_complete(napi); | |
2661 | local_irq_enable(); | |
2662 | break; | |
2663 | } | |
1da177e4 LT |
2664 | local_irq_enable(); |
2665 | ||
d565b0a1 | 2666 | napi_gro_receive(napi, skb); |
bea3348e | 2667 | } while (++work < quota && jiffies == start_time); |
1da177e4 | 2668 | |
d565b0a1 HX |
2669 | napi_gro_flush(napi); |
2670 | ||
bea3348e SH |
2671 | return work; |
2672 | } | |
1da177e4 | 2673 | |
bea3348e SH |
2674 | /** |
2675 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 2676 | * @n: entry to schedule |
bea3348e SH |
2677 | * |
2678 | * The entry's receive function will be scheduled to run | |
2679 | */ | |
b5606c2d | 2680 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
2681 | { |
2682 | unsigned long flags; | |
1da177e4 | 2683 | |
bea3348e SH |
2684 | local_irq_save(flags); |
2685 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); | |
2686 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2687 | local_irq_restore(flags); | |
1da177e4 | 2688 | } |
bea3348e SH |
2689 | EXPORT_SYMBOL(__napi_schedule); |
2690 | ||
d565b0a1 HX |
2691 | void __napi_complete(struct napi_struct *n) |
2692 | { | |
2693 | BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); | |
2694 | BUG_ON(n->gro_list); | |
2695 | ||
2696 | list_del(&n->poll_list); | |
2697 | smp_mb__before_clear_bit(); | |
2698 | clear_bit(NAPI_STATE_SCHED, &n->state); | |
2699 | } | |
2700 | EXPORT_SYMBOL(__napi_complete); | |
2701 | ||
2702 | void napi_complete(struct napi_struct *n) | |
2703 | { | |
2704 | unsigned long flags; | |
2705 | ||
2706 | /* | |
2707 | * don't let napi dequeue from the cpu poll list | |
2708 | * just in case its running on a different cpu | |
2709 | */ | |
2710 | if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state))) | |
2711 | return; | |
2712 | ||
2713 | napi_gro_flush(n); | |
2714 | local_irq_save(flags); | |
2715 | __napi_complete(n); | |
2716 | local_irq_restore(flags); | |
2717 | } | |
2718 | EXPORT_SYMBOL(napi_complete); | |
2719 | ||
2720 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
2721 | int (*poll)(struct napi_struct *, int), int weight) | |
2722 | { | |
2723 | INIT_LIST_HEAD(&napi->poll_list); | |
4ae5544f | 2724 | napi->gro_count = 0; |
d565b0a1 | 2725 | napi->gro_list = NULL; |
5d38a079 | 2726 | napi->skb = NULL; |
d565b0a1 HX |
2727 | napi->poll = poll; |
2728 | napi->weight = weight; | |
2729 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 2730 | napi->dev = dev; |
5d38a079 | 2731 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
2732 | spin_lock_init(&napi->poll_lock); |
2733 | napi->poll_owner = -1; | |
2734 | #endif | |
2735 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
2736 | } | |
2737 | EXPORT_SYMBOL(netif_napi_add); | |
2738 | ||
2739 | void netif_napi_del(struct napi_struct *napi) | |
2740 | { | |
2741 | struct sk_buff *skb, *next; | |
2742 | ||
d7b06636 | 2743 | list_del_init(&napi->dev_list); |
5d38a079 | 2744 | kfree(napi->skb); |
d565b0a1 HX |
2745 | |
2746 | for (skb = napi->gro_list; skb; skb = next) { | |
2747 | next = skb->next; | |
2748 | skb->next = NULL; | |
2749 | kfree_skb(skb); | |
2750 | } | |
2751 | ||
2752 | napi->gro_list = NULL; | |
4ae5544f | 2753 | napi->gro_count = 0; |
d565b0a1 HX |
2754 | } |
2755 | EXPORT_SYMBOL(netif_napi_del); | |
2756 | ||
1da177e4 LT |
2757 | |
2758 | static void net_rx_action(struct softirq_action *h) | |
2759 | { | |
bea3348e | 2760 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list; |
24f8b238 | 2761 | unsigned long time_limit = jiffies + 2; |
51b0bded | 2762 | int budget = netdev_budget; |
53fb95d3 MM |
2763 | void *have; |
2764 | ||
1da177e4 LT |
2765 | local_irq_disable(); |
2766 | ||
bea3348e SH |
2767 | while (!list_empty(list)) { |
2768 | struct napi_struct *n; | |
2769 | int work, weight; | |
1da177e4 | 2770 | |
bea3348e | 2771 | /* If softirq window is exhuasted then punt. |
24f8b238 SH |
2772 | * Allow this to run for 2 jiffies since which will allow |
2773 | * an average latency of 1.5/HZ. | |
bea3348e | 2774 | */ |
24f8b238 | 2775 | if (unlikely(budget <= 0 || time_after(jiffies, time_limit))) |
1da177e4 LT |
2776 | goto softnet_break; |
2777 | ||
2778 | local_irq_enable(); | |
2779 | ||
bea3348e SH |
2780 | /* Even though interrupts have been re-enabled, this |
2781 | * access is safe because interrupts can only add new | |
2782 | * entries to the tail of this list, and only ->poll() | |
2783 | * calls can remove this head entry from the list. | |
2784 | */ | |
2785 | n = list_entry(list->next, struct napi_struct, poll_list); | |
1da177e4 | 2786 | |
bea3348e SH |
2787 | have = netpoll_poll_lock(n); |
2788 | ||
2789 | weight = n->weight; | |
2790 | ||
0a7606c1 DM |
2791 | /* This NAPI_STATE_SCHED test is for avoiding a race |
2792 | * with netpoll's poll_napi(). Only the entity which | |
2793 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
2794 | * actually make the ->poll() call. Therefore we avoid | |
2795 | * accidently calling ->poll() when NAPI is not scheduled. | |
2796 | */ | |
2797 | work = 0; | |
2798 | if (test_bit(NAPI_STATE_SCHED, &n->state)) | |
2799 | work = n->poll(n, weight); | |
bea3348e SH |
2800 | |
2801 | WARN_ON_ONCE(work > weight); | |
2802 | ||
2803 | budget -= work; | |
2804 | ||
2805 | local_irq_disable(); | |
2806 | ||
2807 | /* Drivers must not modify the NAPI state if they | |
2808 | * consume the entire weight. In such cases this code | |
2809 | * still "owns" the NAPI instance and therefore can | |
2810 | * move the instance around on the list at-will. | |
2811 | */ | |
fed17f30 DM |
2812 | if (unlikely(work == weight)) { |
2813 | if (unlikely(napi_disable_pending(n))) | |
2814 | __napi_complete(n); | |
2815 | else | |
2816 | list_move_tail(&n->poll_list, list); | |
2817 | } | |
bea3348e SH |
2818 | |
2819 | netpoll_poll_unlock(have); | |
1da177e4 LT |
2820 | } |
2821 | out: | |
515e06c4 | 2822 | local_irq_enable(); |
bea3348e | 2823 | |
db217334 CL |
2824 | #ifdef CONFIG_NET_DMA |
2825 | /* | |
2826 | * There may not be any more sk_buffs coming right now, so push | |
2827 | * any pending DMA copies to hardware | |
2828 | */ | |
2ba05622 | 2829 | dma_issue_pending_all(); |
db217334 | 2830 | #endif |
bea3348e | 2831 | |
1da177e4 LT |
2832 | return; |
2833 | ||
2834 | softnet_break: | |
2835 | __get_cpu_var(netdev_rx_stat).time_squeeze++; | |
2836 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2837 | goto out; | |
2838 | } | |
2839 | ||
2840 | static gifconf_func_t * gifconf_list [NPROTO]; | |
2841 | ||
2842 | /** | |
2843 | * register_gifconf - register a SIOCGIF handler | |
2844 | * @family: Address family | |
2845 | * @gifconf: Function handler | |
2846 | * | |
2847 | * Register protocol dependent address dumping routines. The handler | |
2848 | * that is passed must not be freed or reused until it has been replaced | |
2849 | * by another handler. | |
2850 | */ | |
2851 | int register_gifconf(unsigned int family, gifconf_func_t * gifconf) | |
2852 | { | |
2853 | if (family >= NPROTO) | |
2854 | return -EINVAL; | |
2855 | gifconf_list[family] = gifconf; | |
2856 | return 0; | |
2857 | } | |
2858 | ||
2859 | ||
2860 | /* | |
2861 | * Map an interface index to its name (SIOCGIFNAME) | |
2862 | */ | |
2863 | ||
2864 | /* | |
2865 | * We need this ioctl for efficient implementation of the | |
2866 | * if_indextoname() function required by the IPv6 API. Without | |
2867 | * it, we would have to search all the interfaces to find a | |
2868 | * match. --pb | |
2869 | */ | |
2870 | ||
881d966b | 2871 | static int dev_ifname(struct net *net, struct ifreq __user *arg) |
1da177e4 LT |
2872 | { |
2873 | struct net_device *dev; | |
2874 | struct ifreq ifr; | |
2875 | ||
2876 | /* | |
2877 | * Fetch the caller's info block. | |
2878 | */ | |
2879 | ||
2880 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
2881 | return -EFAULT; | |
2882 | ||
2883 | read_lock(&dev_base_lock); | |
881d966b | 2884 | dev = __dev_get_by_index(net, ifr.ifr_ifindex); |
1da177e4 LT |
2885 | if (!dev) { |
2886 | read_unlock(&dev_base_lock); | |
2887 | return -ENODEV; | |
2888 | } | |
2889 | ||
2890 | strcpy(ifr.ifr_name, dev->name); | |
2891 | read_unlock(&dev_base_lock); | |
2892 | ||
2893 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
2894 | return -EFAULT; | |
2895 | return 0; | |
2896 | } | |
2897 | ||
2898 | /* | |
2899 | * Perform a SIOCGIFCONF call. This structure will change | |
2900 | * size eventually, and there is nothing I can do about it. | |
2901 | * Thus we will need a 'compatibility mode'. | |
2902 | */ | |
2903 | ||
881d966b | 2904 | static int dev_ifconf(struct net *net, char __user *arg) |
1da177e4 LT |
2905 | { |
2906 | struct ifconf ifc; | |
2907 | struct net_device *dev; | |
2908 | char __user *pos; | |
2909 | int len; | |
2910 | int total; | |
2911 | int i; | |
2912 | ||
2913 | /* | |
2914 | * Fetch the caller's info block. | |
2915 | */ | |
2916 | ||
2917 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
2918 | return -EFAULT; | |
2919 | ||
2920 | pos = ifc.ifc_buf; | |
2921 | len = ifc.ifc_len; | |
2922 | ||
2923 | /* | |
2924 | * Loop over the interfaces, and write an info block for each. | |
2925 | */ | |
2926 | ||
2927 | total = 0; | |
881d966b | 2928 | for_each_netdev(net, dev) { |
1da177e4 LT |
2929 | for (i = 0; i < NPROTO; i++) { |
2930 | if (gifconf_list[i]) { | |
2931 | int done; | |
2932 | if (!pos) | |
2933 | done = gifconf_list[i](dev, NULL, 0); | |
2934 | else | |
2935 | done = gifconf_list[i](dev, pos + total, | |
2936 | len - total); | |
2937 | if (done < 0) | |
2938 | return -EFAULT; | |
2939 | total += done; | |
2940 | } | |
2941 | } | |
4ec93edb | 2942 | } |
1da177e4 LT |
2943 | |
2944 | /* | |
2945 | * All done. Write the updated control block back to the caller. | |
2946 | */ | |
2947 | ifc.ifc_len = total; | |
2948 | ||
2949 | /* | |
2950 | * Both BSD and Solaris return 0 here, so we do too. | |
2951 | */ | |
2952 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
2953 | } | |
2954 | ||
2955 | #ifdef CONFIG_PROC_FS | |
2956 | /* | |
2957 | * This is invoked by the /proc filesystem handler to display a device | |
2958 | * in detail. | |
2959 | */ | |
7562f876 | 2960 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
9a429c49 | 2961 | __acquires(dev_base_lock) |
1da177e4 | 2962 | { |
e372c414 | 2963 | struct net *net = seq_file_net(seq); |
7562f876 | 2964 | loff_t off; |
1da177e4 | 2965 | struct net_device *dev; |
1da177e4 | 2966 | |
7562f876 PE |
2967 | read_lock(&dev_base_lock); |
2968 | if (!*pos) | |
2969 | return SEQ_START_TOKEN; | |
1da177e4 | 2970 | |
7562f876 | 2971 | off = 1; |
881d966b | 2972 | for_each_netdev(net, dev) |
7562f876 PE |
2973 | if (off++ == *pos) |
2974 | return dev; | |
1da177e4 | 2975 | |
7562f876 | 2976 | return NULL; |
1da177e4 LT |
2977 | } |
2978 | ||
2979 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2980 | { | |
e372c414 | 2981 | struct net *net = seq_file_net(seq); |
1da177e4 | 2982 | ++*pos; |
7562f876 | 2983 | return v == SEQ_START_TOKEN ? |
881d966b | 2984 | first_net_device(net) : next_net_device((struct net_device *)v); |
1da177e4 LT |
2985 | } |
2986 | ||
2987 | void dev_seq_stop(struct seq_file *seq, void *v) | |
9a429c49 | 2988 | __releases(dev_base_lock) |
1da177e4 LT |
2989 | { |
2990 | read_unlock(&dev_base_lock); | |
2991 | } | |
2992 | ||
2993 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
2994 | { | |
eeda3fd6 | 2995 | const struct net_device_stats *stats = dev_get_stats(dev); |
1da177e4 | 2996 | |
5a1b5898 RR |
2997 | seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " |
2998 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", | |
2999 | dev->name, stats->rx_bytes, stats->rx_packets, | |
3000 | stats->rx_errors, | |
3001 | stats->rx_dropped + stats->rx_missed_errors, | |
3002 | stats->rx_fifo_errors, | |
3003 | stats->rx_length_errors + stats->rx_over_errors + | |
3004 | stats->rx_crc_errors + stats->rx_frame_errors, | |
3005 | stats->rx_compressed, stats->multicast, | |
3006 | stats->tx_bytes, stats->tx_packets, | |
3007 | stats->tx_errors, stats->tx_dropped, | |
3008 | stats->tx_fifo_errors, stats->collisions, | |
3009 | stats->tx_carrier_errors + | |
3010 | stats->tx_aborted_errors + | |
3011 | stats->tx_window_errors + | |
3012 | stats->tx_heartbeat_errors, | |
3013 | stats->tx_compressed); | |
1da177e4 LT |
3014 | } |
3015 | ||
3016 | /* | |
3017 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
3018 | * /proc/net interface to create /proc/net/dev | |
3019 | */ | |
3020 | static int dev_seq_show(struct seq_file *seq, void *v) | |
3021 | { | |
3022 | if (v == SEQ_START_TOKEN) | |
3023 | seq_puts(seq, "Inter-| Receive " | |
3024 | " | Transmit\n" | |
3025 | " face |bytes packets errs drop fifo frame " | |
3026 | "compressed multicast|bytes packets errs " | |
3027 | "drop fifo colls carrier compressed\n"); | |
3028 | else | |
3029 | dev_seq_printf_stats(seq, v); | |
3030 | return 0; | |
3031 | } | |
3032 | ||
3033 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) | |
3034 | { | |
3035 | struct netif_rx_stats *rc = NULL; | |
3036 | ||
0c0b0aca | 3037 | while (*pos < nr_cpu_ids) |
4ec93edb | 3038 | if (cpu_online(*pos)) { |
1da177e4 LT |
3039 | rc = &per_cpu(netdev_rx_stat, *pos); |
3040 | break; | |
3041 | } else | |
3042 | ++*pos; | |
3043 | return rc; | |
3044 | } | |
3045 | ||
3046 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
3047 | { | |
3048 | return softnet_get_online(pos); | |
3049 | } | |
3050 | ||
3051 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3052 | { | |
3053 | ++*pos; | |
3054 | return softnet_get_online(pos); | |
3055 | } | |
3056 | ||
3057 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
3058 | { | |
3059 | } | |
3060 | ||
3061 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
3062 | { | |
3063 | struct netif_rx_stats *s = v; | |
3064 | ||
3065 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n", | |
31aa02c5 | 3066 | s->total, s->dropped, s->time_squeeze, 0, |
c1ebcdb8 SH |
3067 | 0, 0, 0, 0, /* was fastroute */ |
3068 | s->cpu_collision ); | |
1da177e4 LT |
3069 | return 0; |
3070 | } | |
3071 | ||
f690808e | 3072 | static const struct seq_operations dev_seq_ops = { |
1da177e4 LT |
3073 | .start = dev_seq_start, |
3074 | .next = dev_seq_next, | |
3075 | .stop = dev_seq_stop, | |
3076 | .show = dev_seq_show, | |
3077 | }; | |
3078 | ||
3079 | static int dev_seq_open(struct inode *inode, struct file *file) | |
3080 | { | |
e372c414 DL |
3081 | return seq_open_net(inode, file, &dev_seq_ops, |
3082 | sizeof(struct seq_net_private)); | |
1da177e4 LT |
3083 | } |
3084 | ||
9a32144e | 3085 | static const struct file_operations dev_seq_fops = { |
1da177e4 LT |
3086 | .owner = THIS_MODULE, |
3087 | .open = dev_seq_open, | |
3088 | .read = seq_read, | |
3089 | .llseek = seq_lseek, | |
e372c414 | 3090 | .release = seq_release_net, |
1da177e4 LT |
3091 | }; |
3092 | ||
f690808e | 3093 | static const struct seq_operations softnet_seq_ops = { |
1da177e4 LT |
3094 | .start = softnet_seq_start, |
3095 | .next = softnet_seq_next, | |
3096 | .stop = softnet_seq_stop, | |
3097 | .show = softnet_seq_show, | |
3098 | }; | |
3099 | ||
3100 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
3101 | { | |
3102 | return seq_open(file, &softnet_seq_ops); | |
3103 | } | |
3104 | ||
9a32144e | 3105 | static const struct file_operations softnet_seq_fops = { |
1da177e4 LT |
3106 | .owner = THIS_MODULE, |
3107 | .open = softnet_seq_open, | |
3108 | .read = seq_read, | |
3109 | .llseek = seq_lseek, | |
3110 | .release = seq_release, | |
3111 | }; | |
3112 | ||
0e1256ff SH |
3113 | static void *ptype_get_idx(loff_t pos) |
3114 | { | |
3115 | struct packet_type *pt = NULL; | |
3116 | loff_t i = 0; | |
3117 | int t; | |
3118 | ||
3119 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
3120 | if (i == pos) | |
3121 | return pt; | |
3122 | ++i; | |
3123 | } | |
3124 | ||
82d8a867 | 3125 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { |
0e1256ff SH |
3126 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { |
3127 | if (i == pos) | |
3128 | return pt; | |
3129 | ++i; | |
3130 | } | |
3131 | } | |
3132 | return NULL; | |
3133 | } | |
3134 | ||
3135 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
72348a42 | 3136 | __acquires(RCU) |
0e1256ff SH |
3137 | { |
3138 | rcu_read_lock(); | |
3139 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
3140 | } | |
3141 | ||
3142 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
3143 | { | |
3144 | struct packet_type *pt; | |
3145 | struct list_head *nxt; | |
3146 | int hash; | |
3147 | ||
3148 | ++*pos; | |
3149 | if (v == SEQ_START_TOKEN) | |
3150 | return ptype_get_idx(0); | |
3151 | ||
3152 | pt = v; | |
3153 | nxt = pt->list.next; | |
3154 | if (pt->type == htons(ETH_P_ALL)) { | |
3155 | if (nxt != &ptype_all) | |
3156 | goto found; | |
3157 | hash = 0; | |
3158 | nxt = ptype_base[0].next; | |
3159 | } else | |
82d8a867 | 3160 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; |
0e1256ff SH |
3161 | |
3162 | while (nxt == &ptype_base[hash]) { | |
82d8a867 | 3163 | if (++hash >= PTYPE_HASH_SIZE) |
0e1256ff SH |
3164 | return NULL; |
3165 | nxt = ptype_base[hash].next; | |
3166 | } | |
3167 | found: | |
3168 | return list_entry(nxt, struct packet_type, list); | |
3169 | } | |
3170 | ||
3171 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
72348a42 | 3172 | __releases(RCU) |
0e1256ff SH |
3173 | { |
3174 | rcu_read_unlock(); | |
3175 | } | |
3176 | ||
0e1256ff SH |
3177 | static int ptype_seq_show(struct seq_file *seq, void *v) |
3178 | { | |
3179 | struct packet_type *pt = v; | |
3180 | ||
3181 | if (v == SEQ_START_TOKEN) | |
3182 | seq_puts(seq, "Type Device Function\n"); | |
c346dca1 | 3183 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { |
0e1256ff SH |
3184 | if (pt->type == htons(ETH_P_ALL)) |
3185 | seq_puts(seq, "ALL "); | |
3186 | else | |
3187 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
3188 | ||
908cd2da AD |
3189 | seq_printf(seq, " %-8s %pF\n", |
3190 | pt->dev ? pt->dev->name : "", pt->func); | |
0e1256ff SH |
3191 | } |
3192 | ||
3193 | return 0; | |
3194 | } | |
3195 | ||
3196 | static const struct seq_operations ptype_seq_ops = { | |
3197 | .start = ptype_seq_start, | |
3198 | .next = ptype_seq_next, | |
3199 | .stop = ptype_seq_stop, | |
3200 | .show = ptype_seq_show, | |
3201 | }; | |
3202 | ||
3203 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
3204 | { | |
2feb27db PE |
3205 | return seq_open_net(inode, file, &ptype_seq_ops, |
3206 | sizeof(struct seq_net_private)); | |
0e1256ff SH |
3207 | } |
3208 | ||
3209 | static const struct file_operations ptype_seq_fops = { | |
3210 | .owner = THIS_MODULE, | |
3211 | .open = ptype_seq_open, | |
3212 | .read = seq_read, | |
3213 | .llseek = seq_lseek, | |
2feb27db | 3214 | .release = seq_release_net, |
0e1256ff SH |
3215 | }; |
3216 | ||
3217 | ||
4665079c | 3218 | static int __net_init dev_proc_net_init(struct net *net) |
1da177e4 LT |
3219 | { |
3220 | int rc = -ENOMEM; | |
3221 | ||
881d966b | 3222 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) |
1da177e4 | 3223 | goto out; |
881d966b | 3224 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) |
1da177e4 | 3225 | goto out_dev; |
881d966b | 3226 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) |
457c4cbc | 3227 | goto out_softnet; |
0e1256ff | 3228 | |
881d966b | 3229 | if (wext_proc_init(net)) |
457c4cbc | 3230 | goto out_ptype; |
1da177e4 LT |
3231 | rc = 0; |
3232 | out: | |
3233 | return rc; | |
457c4cbc | 3234 | out_ptype: |
881d966b | 3235 | proc_net_remove(net, "ptype"); |
1da177e4 | 3236 | out_softnet: |
881d966b | 3237 | proc_net_remove(net, "softnet_stat"); |
1da177e4 | 3238 | out_dev: |
881d966b | 3239 | proc_net_remove(net, "dev"); |
1da177e4 LT |
3240 | goto out; |
3241 | } | |
881d966b | 3242 | |
4665079c | 3243 | static void __net_exit dev_proc_net_exit(struct net *net) |
881d966b EB |
3244 | { |
3245 | wext_proc_exit(net); | |
3246 | ||
3247 | proc_net_remove(net, "ptype"); | |
3248 | proc_net_remove(net, "softnet_stat"); | |
3249 | proc_net_remove(net, "dev"); | |
3250 | } | |
3251 | ||
022cbae6 | 3252 | static struct pernet_operations __net_initdata dev_proc_ops = { |
881d966b EB |
3253 | .init = dev_proc_net_init, |
3254 | .exit = dev_proc_net_exit, | |
3255 | }; | |
3256 | ||
3257 | static int __init dev_proc_init(void) | |
3258 | { | |
3259 | return register_pernet_subsys(&dev_proc_ops); | |
3260 | } | |
1da177e4 LT |
3261 | #else |
3262 | #define dev_proc_init() 0 | |
3263 | #endif /* CONFIG_PROC_FS */ | |
3264 | ||
3265 | ||
3266 | /** | |
3267 | * netdev_set_master - set up master/slave pair | |
3268 | * @slave: slave device | |
3269 | * @master: new master device | |
3270 | * | |
3271 | * Changes the master device of the slave. Pass %NULL to break the | |
3272 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
3273 | * a negative errno code is returned. On success the reference counts | |
3274 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
3275 | * function returns zero. | |
3276 | */ | |
3277 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
3278 | { | |
3279 | struct net_device *old = slave->master; | |
3280 | ||
3281 | ASSERT_RTNL(); | |
3282 | ||
3283 | if (master) { | |
3284 | if (old) | |
3285 | return -EBUSY; | |
3286 | dev_hold(master); | |
3287 | } | |
3288 | ||
3289 | slave->master = master; | |
4ec93edb | 3290 | |
1da177e4 LT |
3291 | synchronize_net(); |
3292 | ||
3293 | if (old) | |
3294 | dev_put(old); | |
3295 | ||
3296 | if (master) | |
3297 | slave->flags |= IFF_SLAVE; | |
3298 | else | |
3299 | slave->flags &= ~IFF_SLAVE; | |
3300 | ||
3301 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
3302 | return 0; | |
3303 | } | |
3304 | ||
b6c40d68 PM |
3305 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
3306 | { | |
d314774c SH |
3307 | const struct net_device_ops *ops = dev->netdev_ops; |
3308 | ||
3309 | if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags) | |
3310 | ops->ndo_change_rx_flags(dev, flags); | |
b6c40d68 PM |
3311 | } |
3312 | ||
dad9b335 | 3313 | static int __dev_set_promiscuity(struct net_device *dev, int inc) |
1da177e4 LT |
3314 | { |
3315 | unsigned short old_flags = dev->flags; | |
8192b0c4 DH |
3316 | uid_t uid; |
3317 | gid_t gid; | |
1da177e4 | 3318 | |
24023451 PM |
3319 | ASSERT_RTNL(); |
3320 | ||
dad9b335 WC |
3321 | dev->flags |= IFF_PROMISC; |
3322 | dev->promiscuity += inc; | |
3323 | if (dev->promiscuity == 0) { | |
3324 | /* | |
3325 | * Avoid overflow. | |
3326 | * If inc causes overflow, untouch promisc and return error. | |
3327 | */ | |
3328 | if (inc < 0) | |
3329 | dev->flags &= ~IFF_PROMISC; | |
3330 | else { | |
3331 | dev->promiscuity -= inc; | |
3332 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
3333 | "set promiscuity failed, promiscuity feature " | |
3334 | "of device might be broken.\n", dev->name); | |
3335 | return -EOVERFLOW; | |
3336 | } | |
3337 | } | |
52609c0b | 3338 | if (dev->flags != old_flags) { |
1da177e4 LT |
3339 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
3340 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
4ec93edb | 3341 | "left"); |
8192b0c4 DH |
3342 | if (audit_enabled) { |
3343 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
3344 | audit_log(current->audit_context, GFP_ATOMIC, |
3345 | AUDIT_ANOM_PROMISCUOUS, | |
3346 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
3347 | dev->name, (dev->flags & IFF_PROMISC), | |
3348 | (old_flags & IFF_PROMISC), | |
3349 | audit_get_loginuid(current), | |
8192b0c4 | 3350 | uid, gid, |
7759db82 | 3351 | audit_get_sessionid(current)); |
8192b0c4 | 3352 | } |
24023451 | 3353 | |
b6c40d68 | 3354 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 3355 | } |
dad9b335 | 3356 | return 0; |
1da177e4 LT |
3357 | } |
3358 | ||
4417da66 PM |
3359 | /** |
3360 | * dev_set_promiscuity - update promiscuity count on a device | |
3361 | * @dev: device | |
3362 | * @inc: modifier | |
3363 | * | |
3364 | * Add or remove promiscuity from a device. While the count in the device | |
3365 | * remains above zero the interface remains promiscuous. Once it hits zero | |
3366 | * the device reverts back to normal filtering operation. A negative inc | |
3367 | * value is used to drop promiscuity on the device. | |
dad9b335 | 3368 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 3369 | */ |
dad9b335 | 3370 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 PM |
3371 | { |
3372 | unsigned short old_flags = dev->flags; | |
dad9b335 | 3373 | int err; |
4417da66 | 3374 | |
dad9b335 | 3375 | err = __dev_set_promiscuity(dev, inc); |
4b5a698e | 3376 | if (err < 0) |
dad9b335 | 3377 | return err; |
4417da66 PM |
3378 | if (dev->flags != old_flags) |
3379 | dev_set_rx_mode(dev); | |
dad9b335 | 3380 | return err; |
4417da66 PM |
3381 | } |
3382 | ||
1da177e4 LT |
3383 | /** |
3384 | * dev_set_allmulti - update allmulti count on a device | |
3385 | * @dev: device | |
3386 | * @inc: modifier | |
3387 | * | |
3388 | * Add or remove reception of all multicast frames to a device. While the | |
3389 | * count in the device remains above zero the interface remains listening | |
3390 | * to all interfaces. Once it hits zero the device reverts back to normal | |
3391 | * filtering operation. A negative @inc value is used to drop the counter | |
3392 | * when releasing a resource needing all multicasts. | |
dad9b335 | 3393 | * Return 0 if successful or a negative errno code on error. |
1da177e4 LT |
3394 | */ |
3395 | ||
dad9b335 | 3396 | int dev_set_allmulti(struct net_device *dev, int inc) |
1da177e4 LT |
3397 | { |
3398 | unsigned short old_flags = dev->flags; | |
3399 | ||
24023451 PM |
3400 | ASSERT_RTNL(); |
3401 | ||
1da177e4 | 3402 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
3403 | dev->allmulti += inc; |
3404 | if (dev->allmulti == 0) { | |
3405 | /* | |
3406 | * Avoid overflow. | |
3407 | * If inc causes overflow, untouch allmulti and return error. | |
3408 | */ | |
3409 | if (inc < 0) | |
3410 | dev->flags &= ~IFF_ALLMULTI; | |
3411 | else { | |
3412 | dev->allmulti -= inc; | |
3413 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
3414 | "set allmulti failed, allmulti feature of " | |
3415 | "device might be broken.\n", dev->name); | |
3416 | return -EOVERFLOW; | |
3417 | } | |
3418 | } | |
24023451 | 3419 | if (dev->flags ^ old_flags) { |
b6c40d68 | 3420 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 3421 | dev_set_rx_mode(dev); |
24023451 | 3422 | } |
dad9b335 | 3423 | return 0; |
4417da66 PM |
3424 | } |
3425 | ||
3426 | /* | |
3427 | * Upload unicast and multicast address lists to device and | |
3428 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 3429 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
3430 | * are present. |
3431 | */ | |
3432 | void __dev_set_rx_mode(struct net_device *dev) | |
3433 | { | |
d314774c SH |
3434 | const struct net_device_ops *ops = dev->netdev_ops; |
3435 | ||
4417da66 PM |
3436 | /* dev_open will call this function so the list will stay sane. */ |
3437 | if (!(dev->flags&IFF_UP)) | |
3438 | return; | |
3439 | ||
3440 | if (!netif_device_present(dev)) | |
40b77c94 | 3441 | return; |
4417da66 | 3442 | |
d314774c SH |
3443 | if (ops->ndo_set_rx_mode) |
3444 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
3445 | else { |
3446 | /* Unicast addresses changes may only happen under the rtnl, | |
3447 | * therefore calling __dev_set_promiscuity here is safe. | |
3448 | */ | |
3449 | if (dev->uc_count > 0 && !dev->uc_promisc) { | |
3450 | __dev_set_promiscuity(dev, 1); | |
3451 | dev->uc_promisc = 1; | |
3452 | } else if (dev->uc_count == 0 && dev->uc_promisc) { | |
3453 | __dev_set_promiscuity(dev, -1); | |
3454 | dev->uc_promisc = 0; | |
3455 | } | |
3456 | ||
d314774c SH |
3457 | if (ops->ndo_set_multicast_list) |
3458 | ops->ndo_set_multicast_list(dev); | |
4417da66 PM |
3459 | } |
3460 | } | |
3461 | ||
3462 | void dev_set_rx_mode(struct net_device *dev) | |
3463 | { | |
b9e40857 | 3464 | netif_addr_lock_bh(dev); |
4417da66 | 3465 | __dev_set_rx_mode(dev); |
b9e40857 | 3466 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
3467 | } |
3468 | ||
61cbc2fc PM |
3469 | int __dev_addr_delete(struct dev_addr_list **list, int *count, |
3470 | void *addr, int alen, int glbl) | |
bf742482 PM |
3471 | { |
3472 | struct dev_addr_list *da; | |
3473 | ||
3474 | for (; (da = *list) != NULL; list = &da->next) { | |
3475 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
3476 | alen == da->da_addrlen) { | |
3477 | if (glbl) { | |
3478 | int old_glbl = da->da_gusers; | |
3479 | da->da_gusers = 0; | |
3480 | if (old_glbl == 0) | |
3481 | break; | |
3482 | } | |
3483 | if (--da->da_users) | |
3484 | return 0; | |
3485 | ||
3486 | *list = da->next; | |
3487 | kfree(da); | |
61cbc2fc | 3488 | (*count)--; |
bf742482 PM |
3489 | return 0; |
3490 | } | |
3491 | } | |
3492 | return -ENOENT; | |
3493 | } | |
3494 | ||
61cbc2fc PM |
3495 | int __dev_addr_add(struct dev_addr_list **list, int *count, |
3496 | void *addr, int alen, int glbl) | |
bf742482 PM |
3497 | { |
3498 | struct dev_addr_list *da; | |
3499 | ||
3500 | for (da = *list; da != NULL; da = da->next) { | |
3501 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
3502 | da->da_addrlen == alen) { | |
3503 | if (glbl) { | |
3504 | int old_glbl = da->da_gusers; | |
3505 | da->da_gusers = 1; | |
3506 | if (old_glbl) | |
3507 | return 0; | |
3508 | } | |
3509 | da->da_users++; | |
3510 | return 0; | |
3511 | } | |
3512 | } | |
3513 | ||
12aa343a | 3514 | da = kzalloc(sizeof(*da), GFP_ATOMIC); |
bf742482 PM |
3515 | if (da == NULL) |
3516 | return -ENOMEM; | |
3517 | memcpy(da->da_addr, addr, alen); | |
3518 | da->da_addrlen = alen; | |
3519 | da->da_users = 1; | |
3520 | da->da_gusers = glbl ? 1 : 0; | |
3521 | da->next = *list; | |
3522 | *list = da; | |
61cbc2fc | 3523 | (*count)++; |
bf742482 PM |
3524 | return 0; |
3525 | } | |
3526 | ||
4417da66 PM |
3527 | /** |
3528 | * dev_unicast_delete - Release secondary unicast address. | |
3529 | * @dev: device | |
0ed72ec4 RD |
3530 | * @addr: address to delete |
3531 | * @alen: length of @addr | |
4417da66 PM |
3532 | * |
3533 | * Release reference to a secondary unicast address and remove it | |
0ed72ec4 | 3534 | * from the device if the reference count drops to zero. |
4417da66 PM |
3535 | * |
3536 | * The caller must hold the rtnl_mutex. | |
3537 | */ | |
3538 | int dev_unicast_delete(struct net_device *dev, void *addr, int alen) | |
3539 | { | |
3540 | int err; | |
3541 | ||
3542 | ASSERT_RTNL(); | |
3543 | ||
b9e40857 | 3544 | netif_addr_lock_bh(dev); |
61cbc2fc PM |
3545 | err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0); |
3546 | if (!err) | |
4417da66 | 3547 | __dev_set_rx_mode(dev); |
b9e40857 | 3548 | netif_addr_unlock_bh(dev); |
4417da66 PM |
3549 | return err; |
3550 | } | |
3551 | EXPORT_SYMBOL(dev_unicast_delete); | |
3552 | ||
3553 | /** | |
3554 | * dev_unicast_add - add a secondary unicast address | |
3555 | * @dev: device | |
5dbaec5d | 3556 | * @addr: address to add |
0ed72ec4 | 3557 | * @alen: length of @addr |
4417da66 PM |
3558 | * |
3559 | * Add a secondary unicast address to the device or increase | |
3560 | * the reference count if it already exists. | |
3561 | * | |
3562 | * The caller must hold the rtnl_mutex. | |
3563 | */ | |
3564 | int dev_unicast_add(struct net_device *dev, void *addr, int alen) | |
3565 | { | |
3566 | int err; | |
3567 | ||
3568 | ASSERT_RTNL(); | |
3569 | ||
b9e40857 | 3570 | netif_addr_lock_bh(dev); |
61cbc2fc PM |
3571 | err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0); |
3572 | if (!err) | |
4417da66 | 3573 | __dev_set_rx_mode(dev); |
b9e40857 | 3574 | netif_addr_unlock_bh(dev); |
4417da66 PM |
3575 | return err; |
3576 | } | |
3577 | EXPORT_SYMBOL(dev_unicast_add); | |
3578 | ||
e83a2ea8 CL |
3579 | int __dev_addr_sync(struct dev_addr_list **to, int *to_count, |
3580 | struct dev_addr_list **from, int *from_count) | |
3581 | { | |
3582 | struct dev_addr_list *da, *next; | |
3583 | int err = 0; | |
3584 | ||
3585 | da = *from; | |
3586 | while (da != NULL) { | |
3587 | next = da->next; | |
3588 | if (!da->da_synced) { | |
3589 | err = __dev_addr_add(to, to_count, | |
3590 | da->da_addr, da->da_addrlen, 0); | |
3591 | if (err < 0) | |
3592 | break; | |
3593 | da->da_synced = 1; | |
3594 | da->da_users++; | |
3595 | } else if (da->da_users == 1) { | |
3596 | __dev_addr_delete(to, to_count, | |
3597 | da->da_addr, da->da_addrlen, 0); | |
3598 | __dev_addr_delete(from, from_count, | |
3599 | da->da_addr, da->da_addrlen, 0); | |
3600 | } | |
3601 | da = next; | |
3602 | } | |
3603 | return err; | |
3604 | } | |
3605 | ||
3606 | void __dev_addr_unsync(struct dev_addr_list **to, int *to_count, | |
3607 | struct dev_addr_list **from, int *from_count) | |
3608 | { | |
3609 | struct dev_addr_list *da, *next; | |
3610 | ||
3611 | da = *from; | |
3612 | while (da != NULL) { | |
3613 | next = da->next; | |
3614 | if (da->da_synced) { | |
3615 | __dev_addr_delete(to, to_count, | |
3616 | da->da_addr, da->da_addrlen, 0); | |
3617 | da->da_synced = 0; | |
3618 | __dev_addr_delete(from, from_count, | |
3619 | da->da_addr, da->da_addrlen, 0); | |
3620 | } | |
3621 | da = next; | |
3622 | } | |
3623 | } | |
3624 | ||
3625 | /** | |
3626 | * dev_unicast_sync - Synchronize device's unicast list to another device | |
3627 | * @to: destination device | |
3628 | * @from: source device | |
3629 | * | |
3630 | * Add newly added addresses to the destination device and release | |
3631 | * addresses that have no users left. The source device must be | |
3632 | * locked by netif_tx_lock_bh. | |
3633 | * | |
3634 | * This function is intended to be called from the dev->set_rx_mode | |
3635 | * function of layered software devices. | |
3636 | */ | |
3637 | int dev_unicast_sync(struct net_device *to, struct net_device *from) | |
3638 | { | |
3639 | int err = 0; | |
3640 | ||
b9e40857 | 3641 | netif_addr_lock_bh(to); |
e83a2ea8 CL |
3642 | err = __dev_addr_sync(&to->uc_list, &to->uc_count, |
3643 | &from->uc_list, &from->uc_count); | |
3644 | if (!err) | |
3645 | __dev_set_rx_mode(to); | |
b9e40857 | 3646 | netif_addr_unlock_bh(to); |
e83a2ea8 CL |
3647 | return err; |
3648 | } | |
3649 | EXPORT_SYMBOL(dev_unicast_sync); | |
3650 | ||
3651 | /** | |
bc2cda1e | 3652 | * dev_unicast_unsync - Remove synchronized addresses from the destination device |
e83a2ea8 CL |
3653 | * @to: destination device |
3654 | * @from: source device | |
3655 | * | |
3656 | * Remove all addresses that were added to the destination device by | |
3657 | * dev_unicast_sync(). This function is intended to be called from the | |
3658 | * dev->stop function of layered software devices. | |
3659 | */ | |
3660 | void dev_unicast_unsync(struct net_device *to, struct net_device *from) | |
3661 | { | |
b9e40857 | 3662 | netif_addr_lock_bh(from); |
e308a5d8 | 3663 | netif_addr_lock(to); |
e83a2ea8 CL |
3664 | |
3665 | __dev_addr_unsync(&to->uc_list, &to->uc_count, | |
3666 | &from->uc_list, &from->uc_count); | |
3667 | __dev_set_rx_mode(to); | |
3668 | ||
e308a5d8 | 3669 | netif_addr_unlock(to); |
b9e40857 | 3670 | netif_addr_unlock_bh(from); |
e83a2ea8 CL |
3671 | } |
3672 | EXPORT_SYMBOL(dev_unicast_unsync); | |
3673 | ||
12972621 DC |
3674 | static void __dev_addr_discard(struct dev_addr_list **list) |
3675 | { | |
3676 | struct dev_addr_list *tmp; | |
3677 | ||
3678 | while (*list != NULL) { | |
3679 | tmp = *list; | |
3680 | *list = tmp->next; | |
3681 | if (tmp->da_users > tmp->da_gusers) | |
3682 | printk("__dev_addr_discard: address leakage! " | |
3683 | "da_users=%d\n", tmp->da_users); | |
3684 | kfree(tmp); | |
3685 | } | |
3686 | } | |
3687 | ||
26cc2522 | 3688 | static void dev_addr_discard(struct net_device *dev) |
4417da66 | 3689 | { |
b9e40857 | 3690 | netif_addr_lock_bh(dev); |
26cc2522 | 3691 | |
4417da66 PM |
3692 | __dev_addr_discard(&dev->uc_list); |
3693 | dev->uc_count = 0; | |
4417da66 | 3694 | |
456ad75c DC |
3695 | __dev_addr_discard(&dev->mc_list); |
3696 | dev->mc_count = 0; | |
26cc2522 | 3697 | |
b9e40857 | 3698 | netif_addr_unlock_bh(dev); |
456ad75c DC |
3699 | } |
3700 | ||
f0db275a SH |
3701 | /** |
3702 | * dev_get_flags - get flags reported to userspace | |
3703 | * @dev: device | |
3704 | * | |
3705 | * Get the combination of flag bits exported through APIs to userspace. | |
3706 | */ | |
1da177e4 LT |
3707 | unsigned dev_get_flags(const struct net_device *dev) |
3708 | { | |
3709 | unsigned flags; | |
3710 | ||
3711 | flags = (dev->flags & ~(IFF_PROMISC | | |
3712 | IFF_ALLMULTI | | |
b00055aa SR |
3713 | IFF_RUNNING | |
3714 | IFF_LOWER_UP | | |
3715 | IFF_DORMANT)) | | |
1da177e4 LT |
3716 | (dev->gflags & (IFF_PROMISC | |
3717 | IFF_ALLMULTI)); | |
3718 | ||
b00055aa SR |
3719 | if (netif_running(dev)) { |
3720 | if (netif_oper_up(dev)) | |
3721 | flags |= IFF_RUNNING; | |
3722 | if (netif_carrier_ok(dev)) | |
3723 | flags |= IFF_LOWER_UP; | |
3724 | if (netif_dormant(dev)) | |
3725 | flags |= IFF_DORMANT; | |
3726 | } | |
1da177e4 LT |
3727 | |
3728 | return flags; | |
3729 | } | |
3730 | ||
f0db275a SH |
3731 | /** |
3732 | * dev_change_flags - change device settings | |
3733 | * @dev: device | |
3734 | * @flags: device state flags | |
3735 | * | |
3736 | * Change settings on device based state flags. The flags are | |
3737 | * in the userspace exported format. | |
3738 | */ | |
1da177e4 LT |
3739 | int dev_change_flags(struct net_device *dev, unsigned flags) |
3740 | { | |
7c355f53 | 3741 | int ret, changes; |
1da177e4 LT |
3742 | int old_flags = dev->flags; |
3743 | ||
24023451 PM |
3744 | ASSERT_RTNL(); |
3745 | ||
1da177e4 LT |
3746 | /* |
3747 | * Set the flags on our device. | |
3748 | */ | |
3749 | ||
3750 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
3751 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
3752 | IFF_AUTOMEDIA)) | | |
3753 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
3754 | IFF_ALLMULTI)); | |
3755 | ||
3756 | /* | |
3757 | * Load in the correct multicast list now the flags have changed. | |
3758 | */ | |
3759 | ||
b6c40d68 PM |
3760 | if ((old_flags ^ flags) & IFF_MULTICAST) |
3761 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 3762 | |
4417da66 | 3763 | dev_set_rx_mode(dev); |
1da177e4 LT |
3764 | |
3765 | /* | |
3766 | * Have we downed the interface. We handle IFF_UP ourselves | |
3767 | * according to user attempts to set it, rather than blindly | |
3768 | * setting it. | |
3769 | */ | |
3770 | ||
3771 | ret = 0; | |
3772 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
3773 | ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev); | |
3774 | ||
3775 | if (!ret) | |
4417da66 | 3776 | dev_set_rx_mode(dev); |
1da177e4 LT |
3777 | } |
3778 | ||
3779 | if (dev->flags & IFF_UP && | |
3780 | ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI | | |
3781 | IFF_VOLATILE))) | |
056925ab | 3782 | call_netdevice_notifiers(NETDEV_CHANGE, dev); |
1da177e4 LT |
3783 | |
3784 | if ((flags ^ dev->gflags) & IFF_PROMISC) { | |
3785 | int inc = (flags & IFF_PROMISC) ? +1 : -1; | |
3786 | dev->gflags ^= IFF_PROMISC; | |
3787 | dev_set_promiscuity(dev, inc); | |
3788 | } | |
3789 | ||
3790 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
3791 | is important. Some (broken) drivers set IFF_PROMISC, when | |
3792 | IFF_ALLMULTI is requested not asking us and not reporting. | |
3793 | */ | |
3794 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
3795 | int inc = (flags & IFF_ALLMULTI) ? +1 : -1; | |
3796 | dev->gflags ^= IFF_ALLMULTI; | |
3797 | dev_set_allmulti(dev, inc); | |
3798 | } | |
3799 | ||
7c355f53 TG |
3800 | /* Exclude state transition flags, already notified */ |
3801 | changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING); | |
3802 | if (changes) | |
3803 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
1da177e4 LT |
3804 | |
3805 | return ret; | |
3806 | } | |
3807 | ||
f0db275a SH |
3808 | /** |
3809 | * dev_set_mtu - Change maximum transfer unit | |
3810 | * @dev: device | |
3811 | * @new_mtu: new transfer unit | |
3812 | * | |
3813 | * Change the maximum transfer size of the network device. | |
3814 | */ | |
1da177e4 LT |
3815 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
3816 | { | |
d314774c | 3817 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
3818 | int err; |
3819 | ||
3820 | if (new_mtu == dev->mtu) | |
3821 | return 0; | |
3822 | ||
3823 | /* MTU must be positive. */ | |
3824 | if (new_mtu < 0) | |
3825 | return -EINVAL; | |
3826 | ||
3827 | if (!netif_device_present(dev)) | |
3828 | return -ENODEV; | |
3829 | ||
3830 | err = 0; | |
d314774c SH |
3831 | if (ops->ndo_change_mtu) |
3832 | err = ops->ndo_change_mtu(dev, new_mtu); | |
1da177e4 LT |
3833 | else |
3834 | dev->mtu = new_mtu; | |
d314774c | 3835 | |
1da177e4 | 3836 | if (!err && dev->flags & IFF_UP) |
056925ab | 3837 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); |
1da177e4 LT |
3838 | return err; |
3839 | } | |
3840 | ||
f0db275a SH |
3841 | /** |
3842 | * dev_set_mac_address - Change Media Access Control Address | |
3843 | * @dev: device | |
3844 | * @sa: new address | |
3845 | * | |
3846 | * Change the hardware (MAC) address of the device | |
3847 | */ | |
1da177e4 LT |
3848 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
3849 | { | |
d314774c | 3850 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
3851 | int err; |
3852 | ||
d314774c | 3853 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
3854 | return -EOPNOTSUPP; |
3855 | if (sa->sa_family != dev->type) | |
3856 | return -EINVAL; | |
3857 | if (!netif_device_present(dev)) | |
3858 | return -ENODEV; | |
d314774c | 3859 | err = ops->ndo_set_mac_address(dev, sa); |
1da177e4 | 3860 | if (!err) |
056925ab | 3861 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
3862 | return err; |
3863 | } | |
3864 | ||
3865 | /* | |
14e3e079 | 3866 | * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock) |
1da177e4 | 3867 | */ |
14e3e079 | 3868 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) |
1da177e4 LT |
3869 | { |
3870 | int err; | |
881d966b | 3871 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); |
1da177e4 LT |
3872 | |
3873 | if (!dev) | |
3874 | return -ENODEV; | |
3875 | ||
3876 | switch (cmd) { | |
3877 | case SIOCGIFFLAGS: /* Get interface flags */ | |
3878 | ifr->ifr_flags = dev_get_flags(dev); | |
3879 | return 0; | |
3880 | ||
1da177e4 LT |
3881 | case SIOCGIFMETRIC: /* Get the metric on the interface |
3882 | (currently unused) */ | |
3883 | ifr->ifr_metric = 0; | |
3884 | return 0; | |
3885 | ||
1da177e4 LT |
3886 | case SIOCGIFMTU: /* Get the MTU of a device */ |
3887 | ifr->ifr_mtu = dev->mtu; | |
3888 | return 0; | |
3889 | ||
1da177e4 LT |
3890 | case SIOCGIFHWADDR: |
3891 | if (!dev->addr_len) | |
3892 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
3893 | else | |
3894 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
3895 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
3896 | ifr->ifr_hwaddr.sa_family = dev->type; | |
3897 | return 0; | |
3898 | ||
14e3e079 JG |
3899 | case SIOCGIFSLAVE: |
3900 | err = -EINVAL; | |
3901 | break; | |
3902 | ||
3903 | case SIOCGIFMAP: | |
3904 | ifr->ifr_map.mem_start = dev->mem_start; | |
3905 | ifr->ifr_map.mem_end = dev->mem_end; | |
3906 | ifr->ifr_map.base_addr = dev->base_addr; | |
3907 | ifr->ifr_map.irq = dev->irq; | |
3908 | ifr->ifr_map.dma = dev->dma; | |
3909 | ifr->ifr_map.port = dev->if_port; | |
3910 | return 0; | |
3911 | ||
3912 | case SIOCGIFINDEX: | |
3913 | ifr->ifr_ifindex = dev->ifindex; | |
3914 | return 0; | |
3915 | ||
3916 | case SIOCGIFTXQLEN: | |
3917 | ifr->ifr_qlen = dev->tx_queue_len; | |
3918 | return 0; | |
3919 | ||
3920 | default: | |
3921 | /* dev_ioctl() should ensure this case | |
3922 | * is never reached | |
3923 | */ | |
3924 | WARN_ON(1); | |
3925 | err = -EINVAL; | |
3926 | break; | |
3927 | ||
3928 | } | |
3929 | return err; | |
3930 | } | |
3931 | ||
3932 | /* | |
3933 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
3934 | */ | |
3935 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
3936 | { | |
3937 | int err; | |
3938 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
5f2f6da7 | 3939 | const struct net_device_ops *ops; |
14e3e079 JG |
3940 | |
3941 | if (!dev) | |
3942 | return -ENODEV; | |
3943 | ||
5f2f6da7 JP |
3944 | ops = dev->netdev_ops; |
3945 | ||
14e3e079 JG |
3946 | switch (cmd) { |
3947 | case SIOCSIFFLAGS: /* Set interface flags */ | |
3948 | return dev_change_flags(dev, ifr->ifr_flags); | |
3949 | ||
3950 | case SIOCSIFMETRIC: /* Set the metric on the interface | |
3951 | (currently unused) */ | |
3952 | return -EOPNOTSUPP; | |
3953 | ||
3954 | case SIOCSIFMTU: /* Set the MTU of a device */ | |
3955 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
3956 | ||
1da177e4 LT |
3957 | case SIOCSIFHWADDR: |
3958 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
3959 | ||
3960 | case SIOCSIFHWBROADCAST: | |
3961 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
3962 | return -EINVAL; | |
3963 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
3964 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
056925ab | 3965 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
1da177e4 LT |
3966 | return 0; |
3967 | ||
1da177e4 | 3968 | case SIOCSIFMAP: |
d314774c | 3969 | if (ops->ndo_set_config) { |
1da177e4 LT |
3970 | if (!netif_device_present(dev)) |
3971 | return -ENODEV; | |
d314774c | 3972 | return ops->ndo_set_config(dev, &ifr->ifr_map); |
1da177e4 LT |
3973 | } |
3974 | return -EOPNOTSUPP; | |
3975 | ||
3976 | case SIOCADDMULTI: | |
d314774c | 3977 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || |
1da177e4 LT |
3978 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
3979 | return -EINVAL; | |
3980 | if (!netif_device_present(dev)) | |
3981 | return -ENODEV; | |
3982 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, | |
3983 | dev->addr_len, 1); | |
3984 | ||
3985 | case SIOCDELMULTI: | |
d314774c | 3986 | if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) || |
1da177e4 LT |
3987 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
3988 | return -EINVAL; | |
3989 | if (!netif_device_present(dev)) | |
3990 | return -ENODEV; | |
3991 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, | |
3992 | dev->addr_len, 1); | |
3993 | ||
1da177e4 LT |
3994 | case SIOCSIFTXQLEN: |
3995 | if (ifr->ifr_qlen < 0) | |
3996 | return -EINVAL; | |
3997 | dev->tx_queue_len = ifr->ifr_qlen; | |
3998 | return 0; | |
3999 | ||
4000 | case SIOCSIFNAME: | |
4001 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
4002 | return dev_change_name(dev, ifr->ifr_newname); | |
4003 | ||
4004 | /* | |
4005 | * Unknown or private ioctl | |
4006 | */ | |
4007 | ||
4008 | default: | |
4009 | if ((cmd >= SIOCDEVPRIVATE && | |
4010 | cmd <= SIOCDEVPRIVATE + 15) || | |
4011 | cmd == SIOCBONDENSLAVE || | |
4012 | cmd == SIOCBONDRELEASE || | |
4013 | cmd == SIOCBONDSETHWADDR || | |
4014 | cmd == SIOCBONDSLAVEINFOQUERY || | |
4015 | cmd == SIOCBONDINFOQUERY || | |
4016 | cmd == SIOCBONDCHANGEACTIVE || | |
4017 | cmd == SIOCGMIIPHY || | |
4018 | cmd == SIOCGMIIREG || | |
4019 | cmd == SIOCSMIIREG || | |
4020 | cmd == SIOCBRADDIF || | |
4021 | cmd == SIOCBRDELIF || | |
d24fff22 | 4022 | cmd == SIOCSHWTSTAMP || |
1da177e4 LT |
4023 | cmd == SIOCWANDEV) { |
4024 | err = -EOPNOTSUPP; | |
d314774c | 4025 | if (ops->ndo_do_ioctl) { |
1da177e4 | 4026 | if (netif_device_present(dev)) |
d314774c | 4027 | err = ops->ndo_do_ioctl(dev, ifr, cmd); |
1da177e4 LT |
4028 | else |
4029 | err = -ENODEV; | |
4030 | } | |
4031 | } else | |
4032 | err = -EINVAL; | |
4033 | ||
4034 | } | |
4035 | return err; | |
4036 | } | |
4037 | ||
4038 | /* | |
4039 | * This function handles all "interface"-type I/O control requests. The actual | |
4040 | * 'doing' part of this is dev_ifsioc above. | |
4041 | */ | |
4042 | ||
4043 | /** | |
4044 | * dev_ioctl - network device ioctl | |
c4ea43c5 | 4045 | * @net: the applicable net namespace |
1da177e4 LT |
4046 | * @cmd: command to issue |
4047 | * @arg: pointer to a struct ifreq in user space | |
4048 | * | |
4049 | * Issue ioctl functions to devices. This is normally called by the | |
4050 | * user space syscall interfaces but can sometimes be useful for | |
4051 | * other purposes. The return value is the return from the syscall if | |
4052 | * positive or a negative errno code on error. | |
4053 | */ | |
4054 | ||
881d966b | 4055 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 LT |
4056 | { |
4057 | struct ifreq ifr; | |
4058 | int ret; | |
4059 | char *colon; | |
4060 | ||
4061 | /* One special case: SIOCGIFCONF takes ifconf argument | |
4062 | and requires shared lock, because it sleeps writing | |
4063 | to user space. | |
4064 | */ | |
4065 | ||
4066 | if (cmd == SIOCGIFCONF) { | |
6756ae4b | 4067 | rtnl_lock(); |
881d966b | 4068 | ret = dev_ifconf(net, (char __user *) arg); |
6756ae4b | 4069 | rtnl_unlock(); |
1da177e4 LT |
4070 | return ret; |
4071 | } | |
4072 | if (cmd == SIOCGIFNAME) | |
881d966b | 4073 | return dev_ifname(net, (struct ifreq __user *)arg); |
1da177e4 LT |
4074 | |
4075 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
4076 | return -EFAULT; | |
4077 | ||
4078 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
4079 | ||
4080 | colon = strchr(ifr.ifr_name, ':'); | |
4081 | if (colon) | |
4082 | *colon = 0; | |
4083 | ||
4084 | /* | |
4085 | * See which interface the caller is talking about. | |
4086 | */ | |
4087 | ||
4088 | switch (cmd) { | |
4089 | /* | |
4090 | * These ioctl calls: | |
4091 | * - can be done by all. | |
4092 | * - atomic and do not require locking. | |
4093 | * - return a value | |
4094 | */ | |
4095 | case SIOCGIFFLAGS: | |
4096 | case SIOCGIFMETRIC: | |
4097 | case SIOCGIFMTU: | |
4098 | case SIOCGIFHWADDR: | |
4099 | case SIOCGIFSLAVE: | |
4100 | case SIOCGIFMAP: | |
4101 | case SIOCGIFINDEX: | |
4102 | case SIOCGIFTXQLEN: | |
881d966b | 4103 | dev_load(net, ifr.ifr_name); |
1da177e4 | 4104 | read_lock(&dev_base_lock); |
14e3e079 | 4105 | ret = dev_ifsioc_locked(net, &ifr, cmd); |
1da177e4 LT |
4106 | read_unlock(&dev_base_lock); |
4107 | if (!ret) { | |
4108 | if (colon) | |
4109 | *colon = ':'; | |
4110 | if (copy_to_user(arg, &ifr, | |
4111 | sizeof(struct ifreq))) | |
4112 | ret = -EFAULT; | |
4113 | } | |
4114 | return ret; | |
4115 | ||
4116 | case SIOCETHTOOL: | |
881d966b | 4117 | dev_load(net, ifr.ifr_name); |
1da177e4 | 4118 | rtnl_lock(); |
881d966b | 4119 | ret = dev_ethtool(net, &ifr); |
1da177e4 LT |
4120 | rtnl_unlock(); |
4121 | if (!ret) { | |
4122 | if (colon) | |
4123 | *colon = ':'; | |
4124 | if (copy_to_user(arg, &ifr, | |
4125 | sizeof(struct ifreq))) | |
4126 | ret = -EFAULT; | |
4127 | } | |
4128 | return ret; | |
4129 | ||
4130 | /* | |
4131 | * These ioctl calls: | |
4132 | * - require superuser power. | |
4133 | * - require strict serialization. | |
4134 | * - return a value | |
4135 | */ | |
4136 | case SIOCGMIIPHY: | |
4137 | case SIOCGMIIREG: | |
4138 | case SIOCSIFNAME: | |
4139 | if (!capable(CAP_NET_ADMIN)) | |
4140 | return -EPERM; | |
881d966b | 4141 | dev_load(net, ifr.ifr_name); |
1da177e4 | 4142 | rtnl_lock(); |
881d966b | 4143 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
4144 | rtnl_unlock(); |
4145 | if (!ret) { | |
4146 | if (colon) | |
4147 | *colon = ':'; | |
4148 | if (copy_to_user(arg, &ifr, | |
4149 | sizeof(struct ifreq))) | |
4150 | ret = -EFAULT; | |
4151 | } | |
4152 | return ret; | |
4153 | ||
4154 | /* | |
4155 | * These ioctl calls: | |
4156 | * - require superuser power. | |
4157 | * - require strict serialization. | |
4158 | * - do not return a value | |
4159 | */ | |
4160 | case SIOCSIFFLAGS: | |
4161 | case SIOCSIFMETRIC: | |
4162 | case SIOCSIFMTU: | |
4163 | case SIOCSIFMAP: | |
4164 | case SIOCSIFHWADDR: | |
4165 | case SIOCSIFSLAVE: | |
4166 | case SIOCADDMULTI: | |
4167 | case SIOCDELMULTI: | |
4168 | case SIOCSIFHWBROADCAST: | |
4169 | case SIOCSIFTXQLEN: | |
4170 | case SIOCSMIIREG: | |
4171 | case SIOCBONDENSLAVE: | |
4172 | case SIOCBONDRELEASE: | |
4173 | case SIOCBONDSETHWADDR: | |
1da177e4 LT |
4174 | case SIOCBONDCHANGEACTIVE: |
4175 | case SIOCBRADDIF: | |
4176 | case SIOCBRDELIF: | |
d24fff22 | 4177 | case SIOCSHWTSTAMP: |
1da177e4 LT |
4178 | if (!capable(CAP_NET_ADMIN)) |
4179 | return -EPERM; | |
cabcac0b TG |
4180 | /* fall through */ |
4181 | case SIOCBONDSLAVEINFOQUERY: | |
4182 | case SIOCBONDINFOQUERY: | |
881d966b | 4183 | dev_load(net, ifr.ifr_name); |
1da177e4 | 4184 | rtnl_lock(); |
881d966b | 4185 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
4186 | rtnl_unlock(); |
4187 | return ret; | |
4188 | ||
4189 | case SIOCGIFMEM: | |
4190 | /* Get the per device memory space. We can add this but | |
4191 | * currently do not support it */ | |
4192 | case SIOCSIFMEM: | |
4193 | /* Set the per device memory buffer space. | |
4194 | * Not applicable in our case */ | |
4195 | case SIOCSIFLINK: | |
4196 | return -EINVAL; | |
4197 | ||
4198 | /* | |
4199 | * Unknown or private ioctl. | |
4200 | */ | |
4201 | default: | |
4202 | if (cmd == SIOCWANDEV || | |
4203 | (cmd >= SIOCDEVPRIVATE && | |
4204 | cmd <= SIOCDEVPRIVATE + 15)) { | |
881d966b | 4205 | dev_load(net, ifr.ifr_name); |
1da177e4 | 4206 | rtnl_lock(); |
881d966b | 4207 | ret = dev_ifsioc(net, &ifr, cmd); |
1da177e4 LT |
4208 | rtnl_unlock(); |
4209 | if (!ret && copy_to_user(arg, &ifr, | |
4210 | sizeof(struct ifreq))) | |
4211 | ret = -EFAULT; | |
4212 | return ret; | |
4213 | } | |
1da177e4 | 4214 | /* Take care of Wireless Extensions */ |
295f4a1f | 4215 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) |
881d966b | 4216 | return wext_handle_ioctl(net, &ifr, cmd, arg); |
1da177e4 LT |
4217 | return -EINVAL; |
4218 | } | |
4219 | } | |
4220 | ||
4221 | ||
4222 | /** | |
4223 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 4224 | * @net: the applicable net namespace |
1da177e4 LT |
4225 | * |
4226 | * Returns a suitable unique value for a new device interface | |
4227 | * number. The caller must hold the rtnl semaphore or the | |
4228 | * dev_base_lock to be sure it remains unique. | |
4229 | */ | |
881d966b | 4230 | static int dev_new_index(struct net *net) |
1da177e4 LT |
4231 | { |
4232 | static int ifindex; | |
4233 | for (;;) { | |
4234 | if (++ifindex <= 0) | |
4235 | ifindex = 1; | |
881d966b | 4236 | if (!__dev_get_by_index(net, ifindex)) |
1da177e4 LT |
4237 | return ifindex; |
4238 | } | |
4239 | } | |
4240 | ||
1da177e4 | 4241 | /* Delayed registration/unregisteration */ |
3b5b34fd | 4242 | static LIST_HEAD(net_todo_list); |
1da177e4 | 4243 | |
6f05f629 | 4244 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 4245 | { |
1da177e4 | 4246 | list_add_tail(&dev->todo_list, &net_todo_list); |
1da177e4 LT |
4247 | } |
4248 | ||
93ee31f1 DL |
4249 | static void rollback_registered(struct net_device *dev) |
4250 | { | |
4251 | BUG_ON(dev_boot_phase); | |
4252 | ASSERT_RTNL(); | |
4253 | ||
4254 | /* Some devices call without registering for initialization unwind. */ | |
4255 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
4256 | printk(KERN_DEBUG "unregister_netdevice: device %s/%p never " | |
4257 | "was registered\n", dev->name, dev); | |
4258 | ||
4259 | WARN_ON(1); | |
4260 | return; | |
4261 | } | |
4262 | ||
4263 | BUG_ON(dev->reg_state != NETREG_REGISTERED); | |
4264 | ||
4265 | /* If device is running, close it first. */ | |
4266 | dev_close(dev); | |
4267 | ||
4268 | /* And unlink it from device chain. */ | |
4269 | unlist_netdevice(dev); | |
4270 | ||
4271 | dev->reg_state = NETREG_UNREGISTERING; | |
4272 | ||
4273 | synchronize_net(); | |
4274 | ||
4275 | /* Shutdown queueing discipline. */ | |
4276 | dev_shutdown(dev); | |
4277 | ||
4278 | ||
4279 | /* Notify protocols, that we are about to destroy | |
4280 | this device. They should clean all the things. | |
4281 | */ | |
4282 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
4283 | ||
4284 | /* | |
4285 | * Flush the unicast and multicast chains | |
4286 | */ | |
4287 | dev_addr_discard(dev); | |
4288 | ||
d314774c SH |
4289 | if (dev->netdev_ops->ndo_uninit) |
4290 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 DL |
4291 | |
4292 | /* Notifier chain MUST detach us from master device. */ | |
547b792c | 4293 | WARN_ON(dev->master); |
93ee31f1 DL |
4294 | |
4295 | /* Remove entries from kobject tree */ | |
4296 | netdev_unregister_kobject(dev); | |
4297 | ||
4298 | synchronize_net(); | |
4299 | ||
4300 | dev_put(dev); | |
4301 | } | |
4302 | ||
e8a0464c DM |
4303 | static void __netdev_init_queue_locks_one(struct net_device *dev, |
4304 | struct netdev_queue *dev_queue, | |
4305 | void *_unused) | |
c773e847 DM |
4306 | { |
4307 | spin_lock_init(&dev_queue->_xmit_lock); | |
cf508b12 | 4308 | netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type); |
c773e847 DM |
4309 | dev_queue->xmit_lock_owner = -1; |
4310 | } | |
4311 | ||
4312 | static void netdev_init_queue_locks(struct net_device *dev) | |
4313 | { | |
e8a0464c DM |
4314 | netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL); |
4315 | __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL); | |
c773e847 DM |
4316 | } |
4317 | ||
b63365a2 HX |
4318 | unsigned long netdev_fix_features(unsigned long features, const char *name) |
4319 | { | |
4320 | /* Fix illegal SG+CSUM combinations. */ | |
4321 | if ((features & NETIF_F_SG) && | |
4322 | !(features & NETIF_F_ALL_CSUM)) { | |
4323 | if (name) | |
4324 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no " | |
4325 | "checksum feature.\n", name); | |
4326 | features &= ~NETIF_F_SG; | |
4327 | } | |
4328 | ||
4329 | /* TSO requires that SG is present as well. */ | |
4330 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) { | |
4331 | if (name) | |
4332 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no " | |
4333 | "SG feature.\n", name); | |
4334 | features &= ~NETIF_F_TSO; | |
4335 | } | |
4336 | ||
4337 | if (features & NETIF_F_UFO) { | |
4338 | if (!(features & NETIF_F_GEN_CSUM)) { | |
4339 | if (name) | |
4340 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4341 | "since no NETIF_F_HW_CSUM feature.\n", | |
4342 | name); | |
4343 | features &= ~NETIF_F_UFO; | |
4344 | } | |
4345 | ||
4346 | if (!(features & NETIF_F_SG)) { | |
4347 | if (name) | |
4348 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO " | |
4349 | "since no NETIF_F_SG feature.\n", name); | |
4350 | features &= ~NETIF_F_UFO; | |
4351 | } | |
4352 | } | |
4353 | ||
4354 | return features; | |
4355 | } | |
4356 | EXPORT_SYMBOL(netdev_fix_features); | |
4357 | ||
1da177e4 LT |
4358 | /** |
4359 | * register_netdevice - register a network device | |
4360 | * @dev: device to register | |
4361 | * | |
4362 | * Take a completed network device structure and add it to the kernel | |
4363 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
4364 | * chain. 0 is returned on success. A negative errno code is returned | |
4365 | * on a failure to set up the device, or if the name is a duplicate. | |
4366 | * | |
4367 | * Callers must hold the rtnl semaphore. You may want | |
4368 | * register_netdev() instead of this. | |
4369 | * | |
4370 | * BUGS: | |
4371 | * The locking appears insufficient to guarantee two parallel registers | |
4372 | * will not get the same name. | |
4373 | */ | |
4374 | ||
4375 | int register_netdevice(struct net_device *dev) | |
4376 | { | |
4377 | struct hlist_head *head; | |
4378 | struct hlist_node *p; | |
4379 | int ret; | |
d314774c | 4380 | struct net *net = dev_net(dev); |
1da177e4 LT |
4381 | |
4382 | BUG_ON(dev_boot_phase); | |
4383 | ASSERT_RTNL(); | |
4384 | ||
b17a7c17 SH |
4385 | might_sleep(); |
4386 | ||
1da177e4 LT |
4387 | /* When net_device's are persistent, this will be fatal. */ |
4388 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 4389 | BUG_ON(!net); |
1da177e4 | 4390 | |
f1f28aa3 | 4391 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 4392 | netdev_set_addr_lockdep_class(dev); |
c773e847 | 4393 | netdev_init_queue_locks(dev); |
1da177e4 | 4394 | |
1da177e4 LT |
4395 | dev->iflink = -1; |
4396 | ||
d314774c SH |
4397 | #ifdef CONFIG_COMPAT_NET_DEV_OPS |
4398 | /* Netdevice_ops API compatiability support. | |
4399 | * This is temporary until all network devices are converted. | |
4400 | */ | |
4401 | if (dev->netdev_ops) { | |
4402 | const struct net_device_ops *ops = dev->netdev_ops; | |
4403 | ||
4404 | dev->init = ops->ndo_init; | |
4405 | dev->uninit = ops->ndo_uninit; | |
4406 | dev->open = ops->ndo_open; | |
4407 | dev->change_rx_flags = ops->ndo_change_rx_flags; | |
4408 | dev->set_rx_mode = ops->ndo_set_rx_mode; | |
4409 | dev->set_multicast_list = ops->ndo_set_multicast_list; | |
4410 | dev->set_mac_address = ops->ndo_set_mac_address; | |
4411 | dev->validate_addr = ops->ndo_validate_addr; | |
4412 | dev->do_ioctl = ops->ndo_do_ioctl; | |
4413 | dev->set_config = ops->ndo_set_config; | |
4414 | dev->change_mtu = ops->ndo_change_mtu; | |
4415 | dev->tx_timeout = ops->ndo_tx_timeout; | |
4416 | dev->get_stats = ops->ndo_get_stats; | |
4417 | dev->vlan_rx_register = ops->ndo_vlan_rx_register; | |
4418 | dev->vlan_rx_add_vid = ops->ndo_vlan_rx_add_vid; | |
4419 | dev->vlan_rx_kill_vid = ops->ndo_vlan_rx_kill_vid; | |
4420 | #ifdef CONFIG_NET_POLL_CONTROLLER | |
4421 | dev->poll_controller = ops->ndo_poll_controller; | |
4422 | #endif | |
4423 | } else { | |
4424 | char drivername[64]; | |
4425 | pr_info("%s (%s): not using net_device_ops yet\n", | |
4426 | dev->name, netdev_drivername(dev, drivername, 64)); | |
4427 | ||
4428 | /* This works only because net_device_ops and the | |
4429 | compatiablity structure are the same. */ | |
4430 | dev->netdev_ops = (void *) &(dev->init); | |
4431 | } | |
4432 | #endif | |
4433 | ||
1da177e4 | 4434 | /* Init, if this function is available */ |
d314774c SH |
4435 | if (dev->netdev_ops->ndo_init) { |
4436 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
4437 | if (ret) { |
4438 | if (ret > 0) | |
4439 | ret = -EIO; | |
90833aa4 | 4440 | goto out; |
1da177e4 LT |
4441 | } |
4442 | } | |
4ec93edb | 4443 | |
1da177e4 LT |
4444 | if (!dev_valid_name(dev->name)) { |
4445 | ret = -EINVAL; | |
7ce1b0ed | 4446 | goto err_uninit; |
1da177e4 LT |
4447 | } |
4448 | ||
881d966b | 4449 | dev->ifindex = dev_new_index(net); |
1da177e4 LT |
4450 | if (dev->iflink == -1) |
4451 | dev->iflink = dev->ifindex; | |
4452 | ||
4453 | /* Check for existence of name */ | |
881d966b | 4454 | head = dev_name_hash(net, dev->name); |
1da177e4 LT |
4455 | hlist_for_each(p, head) { |
4456 | struct net_device *d | |
4457 | = hlist_entry(p, struct net_device, name_hlist); | |
4458 | if (!strncmp(d->name, dev->name, IFNAMSIZ)) { | |
4459 | ret = -EEXIST; | |
7ce1b0ed | 4460 | goto err_uninit; |
1da177e4 | 4461 | } |
4ec93edb | 4462 | } |
1da177e4 | 4463 | |
d212f87b SH |
4464 | /* Fix illegal checksum combinations */ |
4465 | if ((dev->features & NETIF_F_HW_CSUM) && | |
4466 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
4467 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
4468 | dev->name); | |
4469 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
4470 | } | |
4471 | ||
4472 | if ((dev->features & NETIF_F_NO_CSUM) && | |
4473 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
4474 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
4475 | dev->name); | |
4476 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
4477 | } | |
4478 | ||
b63365a2 | 4479 | dev->features = netdev_fix_features(dev->features, dev->name); |
1da177e4 | 4480 | |
e5a4a72d LB |
4481 | /* Enable software GSO if SG is supported. */ |
4482 | if (dev->features & NETIF_F_SG) | |
4483 | dev->features |= NETIF_F_GSO; | |
4484 | ||
aaf8cdc3 | 4485 | netdev_initialize_kobject(dev); |
8b41d188 | 4486 | ret = netdev_register_kobject(dev); |
b17a7c17 | 4487 | if (ret) |
7ce1b0ed | 4488 | goto err_uninit; |
b17a7c17 SH |
4489 | dev->reg_state = NETREG_REGISTERED; |
4490 | ||
1da177e4 LT |
4491 | /* |
4492 | * Default initial state at registry is that the | |
4493 | * device is present. | |
4494 | */ | |
4495 | ||
4496 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
4497 | ||
1da177e4 | 4498 | dev_init_scheduler(dev); |
1da177e4 | 4499 | dev_hold(dev); |
ce286d32 | 4500 | list_netdevice(dev); |
1da177e4 LT |
4501 | |
4502 | /* Notify protocols, that a new device appeared. */ | |
056925ab | 4503 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 4504 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
4505 | if (ret) { |
4506 | rollback_registered(dev); | |
4507 | dev->reg_state = NETREG_UNREGISTERED; | |
4508 | } | |
1da177e4 LT |
4509 | |
4510 | out: | |
4511 | return ret; | |
7ce1b0ed HX |
4512 | |
4513 | err_uninit: | |
d314774c SH |
4514 | if (dev->netdev_ops->ndo_uninit) |
4515 | dev->netdev_ops->ndo_uninit(dev); | |
7ce1b0ed | 4516 | goto out; |
1da177e4 LT |
4517 | } |
4518 | ||
937f1ba5 BH |
4519 | /** |
4520 | * init_dummy_netdev - init a dummy network device for NAPI | |
4521 | * @dev: device to init | |
4522 | * | |
4523 | * This takes a network device structure and initialize the minimum | |
4524 | * amount of fields so it can be used to schedule NAPI polls without | |
4525 | * registering a full blown interface. This is to be used by drivers | |
4526 | * that need to tie several hardware interfaces to a single NAPI | |
4527 | * poll scheduler due to HW limitations. | |
4528 | */ | |
4529 | int init_dummy_netdev(struct net_device *dev) | |
4530 | { | |
4531 | /* Clear everything. Note we don't initialize spinlocks | |
4532 | * are they aren't supposed to be taken by any of the | |
4533 | * NAPI code and this dummy netdev is supposed to be | |
4534 | * only ever used for NAPI polls | |
4535 | */ | |
4536 | memset(dev, 0, sizeof(struct net_device)); | |
4537 | ||
4538 | /* make sure we BUG if trying to hit standard | |
4539 | * register/unregister code path | |
4540 | */ | |
4541 | dev->reg_state = NETREG_DUMMY; | |
4542 | ||
4543 | /* initialize the ref count */ | |
4544 | atomic_set(&dev->refcnt, 1); | |
4545 | ||
4546 | /* NAPI wants this */ | |
4547 | INIT_LIST_HEAD(&dev->napi_list); | |
4548 | ||
4549 | /* a dummy interface is started by default */ | |
4550 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
4551 | set_bit(__LINK_STATE_START, &dev->state); | |
4552 | ||
4553 | return 0; | |
4554 | } | |
4555 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
4556 | ||
4557 | ||
1da177e4 LT |
4558 | /** |
4559 | * register_netdev - register a network device | |
4560 | * @dev: device to register | |
4561 | * | |
4562 | * Take a completed network device structure and add it to the kernel | |
4563 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
4564 | * chain. 0 is returned on success. A negative errno code is returned | |
4565 | * on a failure to set up the device, or if the name is a duplicate. | |
4566 | * | |
38b4da38 | 4567 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
4568 | * and expands the device name if you passed a format string to |
4569 | * alloc_netdev. | |
4570 | */ | |
4571 | int register_netdev(struct net_device *dev) | |
4572 | { | |
4573 | int err; | |
4574 | ||
4575 | rtnl_lock(); | |
4576 | ||
4577 | /* | |
4578 | * If the name is a format string the caller wants us to do a | |
4579 | * name allocation. | |
4580 | */ | |
4581 | if (strchr(dev->name, '%')) { | |
4582 | err = dev_alloc_name(dev, dev->name); | |
4583 | if (err < 0) | |
4584 | goto out; | |
4585 | } | |
4ec93edb | 4586 | |
1da177e4 LT |
4587 | err = register_netdevice(dev); |
4588 | out: | |
4589 | rtnl_unlock(); | |
4590 | return err; | |
4591 | } | |
4592 | EXPORT_SYMBOL(register_netdev); | |
4593 | ||
4594 | /* | |
4595 | * netdev_wait_allrefs - wait until all references are gone. | |
4596 | * | |
4597 | * This is called when unregistering network devices. | |
4598 | * | |
4599 | * Any protocol or device that holds a reference should register | |
4600 | * for netdevice notification, and cleanup and put back the | |
4601 | * reference if they receive an UNREGISTER event. | |
4602 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 4603 | * call dev_put. |
1da177e4 LT |
4604 | */ |
4605 | static void netdev_wait_allrefs(struct net_device *dev) | |
4606 | { | |
4607 | unsigned long rebroadcast_time, warning_time; | |
4608 | ||
4609 | rebroadcast_time = warning_time = jiffies; | |
4610 | while (atomic_read(&dev->refcnt) != 0) { | |
4611 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
6756ae4b | 4612 | rtnl_lock(); |
1da177e4 LT |
4613 | |
4614 | /* Rebroadcast unregister notification */ | |
056925ab | 4615 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 LT |
4616 | |
4617 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
4618 | &dev->state)) { | |
4619 | /* We must not have linkwatch events | |
4620 | * pending on unregister. If this | |
4621 | * happens, we simply run the queue | |
4622 | * unscheduled, resulting in a noop | |
4623 | * for this device. | |
4624 | */ | |
4625 | linkwatch_run_queue(); | |
4626 | } | |
4627 | ||
6756ae4b | 4628 | __rtnl_unlock(); |
1da177e4 LT |
4629 | |
4630 | rebroadcast_time = jiffies; | |
4631 | } | |
4632 | ||
4633 | msleep(250); | |
4634 | ||
4635 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
4636 | printk(KERN_EMERG "unregister_netdevice: " | |
4637 | "waiting for %s to become free. Usage " | |
4638 | "count = %d\n", | |
4639 | dev->name, atomic_read(&dev->refcnt)); | |
4640 | warning_time = jiffies; | |
4641 | } | |
4642 | } | |
4643 | } | |
4644 | ||
4645 | /* The sequence is: | |
4646 | * | |
4647 | * rtnl_lock(); | |
4648 | * ... | |
4649 | * register_netdevice(x1); | |
4650 | * register_netdevice(x2); | |
4651 | * ... | |
4652 | * unregister_netdevice(y1); | |
4653 | * unregister_netdevice(y2); | |
4654 | * ... | |
4655 | * rtnl_unlock(); | |
4656 | * free_netdev(y1); | |
4657 | * free_netdev(y2); | |
4658 | * | |
58ec3b4d | 4659 | * We are invoked by rtnl_unlock(). |
1da177e4 | 4660 | * This allows us to deal with problems: |
b17a7c17 | 4661 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
4662 | * without deadlocking with linkwatch via keventd. |
4663 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
4664 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
4665 | * |
4666 | * We must not return until all unregister events added during | |
4667 | * the interval the lock was held have been completed. | |
1da177e4 | 4668 | */ |
1da177e4 LT |
4669 | void netdev_run_todo(void) |
4670 | { | |
626ab0e6 | 4671 | struct list_head list; |
1da177e4 | 4672 | |
1da177e4 | 4673 | /* Snapshot list, allow later requests */ |
626ab0e6 | 4674 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
4675 | |
4676 | __rtnl_unlock(); | |
626ab0e6 | 4677 | |
1da177e4 LT |
4678 | while (!list_empty(&list)) { |
4679 | struct net_device *dev | |
4680 | = list_entry(list.next, struct net_device, todo_list); | |
4681 | list_del(&dev->todo_list); | |
4682 | ||
b17a7c17 SH |
4683 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
4684 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
4685 | dev->name, dev->reg_state); | |
4686 | dump_stack(); | |
4687 | continue; | |
4688 | } | |
1da177e4 | 4689 | |
b17a7c17 | 4690 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 4691 | |
6e583ce5 SH |
4692 | on_each_cpu(flush_backlog, dev, 1); |
4693 | ||
b17a7c17 | 4694 | netdev_wait_allrefs(dev); |
1da177e4 | 4695 | |
b17a7c17 SH |
4696 | /* paranoia */ |
4697 | BUG_ON(atomic_read(&dev->refcnt)); | |
547b792c IJ |
4698 | WARN_ON(dev->ip_ptr); |
4699 | WARN_ON(dev->ip6_ptr); | |
4700 | WARN_ON(dev->dn_ptr); | |
1da177e4 | 4701 | |
b17a7c17 SH |
4702 | if (dev->destructor) |
4703 | dev->destructor(dev); | |
9093bbb2 SH |
4704 | |
4705 | /* Free network device */ | |
4706 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 4707 | } |
1da177e4 LT |
4708 | } |
4709 | ||
eeda3fd6 SH |
4710 | /** |
4711 | * dev_get_stats - get network device statistics | |
4712 | * @dev: device to get statistics from | |
4713 | * | |
4714 | * Get network statistics from device. The device driver may provide | |
4715 | * its own method by setting dev->netdev_ops->get_stats; otherwise | |
4716 | * the internal statistics structure is used. | |
4717 | */ | |
4718 | const struct net_device_stats *dev_get_stats(struct net_device *dev) | |
4719 | { | |
4720 | const struct net_device_ops *ops = dev->netdev_ops; | |
4721 | ||
4722 | if (ops->ndo_get_stats) | |
4723 | return ops->ndo_get_stats(dev); | |
4724 | else | |
4725 | return &dev->stats; | |
c45d286e | 4726 | } |
eeda3fd6 | 4727 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 4728 | |
dc2b4847 | 4729 | static void netdev_init_one_queue(struct net_device *dev, |
e8a0464c DM |
4730 | struct netdev_queue *queue, |
4731 | void *_unused) | |
dc2b4847 | 4732 | { |
dc2b4847 DM |
4733 | queue->dev = dev; |
4734 | } | |
4735 | ||
bb949fbd DM |
4736 | static void netdev_init_queues(struct net_device *dev) |
4737 | { | |
e8a0464c DM |
4738 | netdev_init_one_queue(dev, &dev->rx_queue, NULL); |
4739 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); | |
c3f26a26 | 4740 | spin_lock_init(&dev->tx_global_lock); |
bb949fbd DM |
4741 | } |
4742 | ||
1da177e4 | 4743 | /** |
f25f4e44 | 4744 | * alloc_netdev_mq - allocate network device |
1da177e4 LT |
4745 | * @sizeof_priv: size of private data to allocate space for |
4746 | * @name: device name format string | |
4747 | * @setup: callback to initialize device | |
f25f4e44 | 4748 | * @queue_count: the number of subqueues to allocate |
1da177e4 LT |
4749 | * |
4750 | * Allocates a struct net_device with private data area for driver use | |
f25f4e44 PWJ |
4751 | * and performs basic initialization. Also allocates subquue structs |
4752 | * for each queue on the device at the end of the netdevice. | |
1da177e4 | 4753 | */ |
f25f4e44 PWJ |
4754 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, |
4755 | void (*setup)(struct net_device *), unsigned int queue_count) | |
1da177e4 | 4756 | { |
e8a0464c | 4757 | struct netdev_queue *tx; |
1da177e4 | 4758 | struct net_device *dev; |
7943986c | 4759 | size_t alloc_size; |
e8a0464c | 4760 | void *p; |
1da177e4 | 4761 | |
b6fe17d6 SH |
4762 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
4763 | ||
fd2ea0a7 | 4764 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
4765 | if (sizeof_priv) { |
4766 | /* ensure 32-byte alignment of private area */ | |
4767 | alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST; | |
4768 | alloc_size += sizeof_priv; | |
4769 | } | |
4770 | /* ensure 32-byte alignment of whole construct */ | |
4771 | alloc_size += NETDEV_ALIGN_CONST; | |
1da177e4 | 4772 | |
31380de9 | 4773 | p = kzalloc(alloc_size, GFP_KERNEL); |
1da177e4 | 4774 | if (!p) { |
b6fe17d6 | 4775 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); |
1da177e4 LT |
4776 | return NULL; |
4777 | } | |
1da177e4 | 4778 | |
7943986c | 4779 | tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL); |
e8a0464c DM |
4780 | if (!tx) { |
4781 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
4782 | "tx qdiscs.\n"); | |
4783 | kfree(p); | |
4784 | return NULL; | |
4785 | } | |
4786 | ||
1da177e4 LT |
4787 | dev = (struct net_device *) |
4788 | (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST); | |
4789 | dev->padded = (char *)dev - (char *)p; | |
c346dca1 | 4790 | dev_net_set(dev, &init_net); |
1da177e4 | 4791 | |
e8a0464c DM |
4792 | dev->_tx = tx; |
4793 | dev->num_tx_queues = queue_count; | |
fd2ea0a7 | 4794 | dev->real_num_tx_queues = queue_count; |
e8a0464c | 4795 | |
82cc1a7a | 4796 | dev->gso_max_size = GSO_MAX_SIZE; |
1da177e4 | 4797 | |
bb949fbd DM |
4798 | netdev_init_queues(dev); |
4799 | ||
d565b0a1 | 4800 | INIT_LIST_HEAD(&dev->napi_list); |
1da177e4 LT |
4801 | setup(dev); |
4802 | strcpy(dev->name, name); | |
4803 | return dev; | |
4804 | } | |
f25f4e44 | 4805 | EXPORT_SYMBOL(alloc_netdev_mq); |
1da177e4 LT |
4806 | |
4807 | /** | |
4808 | * free_netdev - free network device | |
4809 | * @dev: device | |
4810 | * | |
4ec93edb YH |
4811 | * This function does the last stage of destroying an allocated device |
4812 | * interface. The reference to the device object is released. | |
1da177e4 LT |
4813 | * If this is the last reference then it will be freed. |
4814 | */ | |
4815 | void free_netdev(struct net_device *dev) | |
4816 | { | |
d565b0a1 HX |
4817 | struct napi_struct *p, *n; |
4818 | ||
f3005d7f DL |
4819 | release_net(dev_net(dev)); |
4820 | ||
e8a0464c DM |
4821 | kfree(dev->_tx); |
4822 | ||
d565b0a1 HX |
4823 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
4824 | netif_napi_del(p); | |
4825 | ||
3041a069 | 4826 | /* Compatibility with error handling in drivers */ |
1da177e4 LT |
4827 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
4828 | kfree((char *)dev - dev->padded); | |
4829 | return; | |
4830 | } | |
4831 | ||
4832 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
4833 | dev->reg_state = NETREG_RELEASED; | |
4834 | ||
43cb76d9 GKH |
4835 | /* will free via device release */ |
4836 | put_device(&dev->dev); | |
1da177e4 | 4837 | } |
4ec93edb | 4838 | |
f0db275a SH |
4839 | /** |
4840 | * synchronize_net - Synchronize with packet receive processing | |
4841 | * | |
4842 | * Wait for packets currently being received to be done. | |
4843 | * Does not block later packets from starting. | |
4844 | */ | |
4ec93edb | 4845 | void synchronize_net(void) |
1da177e4 LT |
4846 | { |
4847 | might_sleep(); | |
fbd568a3 | 4848 | synchronize_rcu(); |
1da177e4 LT |
4849 | } |
4850 | ||
4851 | /** | |
4852 | * unregister_netdevice - remove device from the kernel | |
4853 | * @dev: device | |
4854 | * | |
4855 | * This function shuts down a device interface and removes it | |
d59b54b1 | 4856 | * from the kernel tables. |
1da177e4 LT |
4857 | * |
4858 | * Callers must hold the rtnl semaphore. You may want | |
4859 | * unregister_netdev() instead of this. | |
4860 | */ | |
4861 | ||
22f8cde5 | 4862 | void unregister_netdevice(struct net_device *dev) |
1da177e4 | 4863 | { |
a6620712 HX |
4864 | ASSERT_RTNL(); |
4865 | ||
93ee31f1 | 4866 | rollback_registered(dev); |
1da177e4 LT |
4867 | /* Finish processing unregister after unlock */ |
4868 | net_set_todo(dev); | |
1da177e4 LT |
4869 | } |
4870 | ||
4871 | /** | |
4872 | * unregister_netdev - remove device from the kernel | |
4873 | * @dev: device | |
4874 | * | |
4875 | * This function shuts down a device interface and removes it | |
d59b54b1 | 4876 | * from the kernel tables. |
1da177e4 LT |
4877 | * |
4878 | * This is just a wrapper for unregister_netdevice that takes | |
4879 | * the rtnl semaphore. In general you want to use this and not | |
4880 | * unregister_netdevice. | |
4881 | */ | |
4882 | void unregister_netdev(struct net_device *dev) | |
4883 | { | |
4884 | rtnl_lock(); | |
4885 | unregister_netdevice(dev); | |
4886 | rtnl_unlock(); | |
4887 | } | |
4888 | ||
4889 | EXPORT_SYMBOL(unregister_netdev); | |
4890 | ||
ce286d32 EB |
4891 | /** |
4892 | * dev_change_net_namespace - move device to different nethost namespace | |
4893 | * @dev: device | |
4894 | * @net: network namespace | |
4895 | * @pat: If not NULL name pattern to try if the current device name | |
4896 | * is already taken in the destination network namespace. | |
4897 | * | |
4898 | * This function shuts down a device interface and moves it | |
4899 | * to a new network namespace. On success 0 is returned, on | |
4900 | * a failure a netagive errno code is returned. | |
4901 | * | |
4902 | * Callers must hold the rtnl semaphore. | |
4903 | */ | |
4904 | ||
4905 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
4906 | { | |
4907 | char buf[IFNAMSIZ]; | |
4908 | const char *destname; | |
4909 | int err; | |
4910 | ||
4911 | ASSERT_RTNL(); | |
4912 | ||
4913 | /* Don't allow namespace local devices to be moved. */ | |
4914 | err = -EINVAL; | |
4915 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
4916 | goto out; | |
4917 | ||
3891845e EB |
4918 | #ifdef CONFIG_SYSFS |
4919 | /* Don't allow real devices to be moved when sysfs | |
4920 | * is enabled. | |
4921 | */ | |
4922 | err = -EINVAL; | |
4923 | if (dev->dev.parent) | |
4924 | goto out; | |
4925 | #endif | |
4926 | ||
ce286d32 EB |
4927 | /* Ensure the device has been registrered */ |
4928 | err = -EINVAL; | |
4929 | if (dev->reg_state != NETREG_REGISTERED) | |
4930 | goto out; | |
4931 | ||
4932 | /* Get out if there is nothing todo */ | |
4933 | err = 0; | |
878628fb | 4934 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
4935 | goto out; |
4936 | ||
4937 | /* Pick the destination device name, and ensure | |
4938 | * we can use it in the destination network namespace. | |
4939 | */ | |
4940 | err = -EEXIST; | |
4941 | destname = dev->name; | |
4942 | if (__dev_get_by_name(net, destname)) { | |
4943 | /* We get here if we can't use the current device name */ | |
4944 | if (!pat) | |
4945 | goto out; | |
4946 | if (!dev_valid_name(pat)) | |
4947 | goto out; | |
4948 | if (strchr(pat, '%')) { | |
4949 | if (__dev_alloc_name(net, pat, buf) < 0) | |
4950 | goto out; | |
4951 | destname = buf; | |
4952 | } else | |
4953 | destname = pat; | |
4954 | if (__dev_get_by_name(net, destname)) | |
4955 | goto out; | |
4956 | } | |
4957 | ||
4958 | /* | |
4959 | * And now a mini version of register_netdevice unregister_netdevice. | |
4960 | */ | |
4961 | ||
4962 | /* If device is running close it first. */ | |
9b772652 | 4963 | dev_close(dev); |
ce286d32 EB |
4964 | |
4965 | /* And unlink it from device chain */ | |
4966 | err = -ENODEV; | |
4967 | unlist_netdevice(dev); | |
4968 | ||
4969 | synchronize_net(); | |
4970 | ||
4971 | /* Shutdown queueing discipline. */ | |
4972 | dev_shutdown(dev); | |
4973 | ||
4974 | /* Notify protocols, that we are about to destroy | |
4975 | this device. They should clean all the things. | |
4976 | */ | |
4977 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
4978 | ||
4979 | /* | |
4980 | * Flush the unicast and multicast chains | |
4981 | */ | |
4982 | dev_addr_discard(dev); | |
4983 | ||
3891845e EB |
4984 | netdev_unregister_kobject(dev); |
4985 | ||
ce286d32 | 4986 | /* Actually switch the network namespace */ |
c346dca1 | 4987 | dev_net_set(dev, net); |
ce286d32 EB |
4988 | |
4989 | /* Assign the new device name */ | |
4990 | if (destname != dev->name) | |
4991 | strcpy(dev->name, destname); | |
4992 | ||
4993 | /* If there is an ifindex conflict assign a new one */ | |
4994 | if (__dev_get_by_index(net, dev->ifindex)) { | |
4995 | int iflink = (dev->iflink == dev->ifindex); | |
4996 | dev->ifindex = dev_new_index(net); | |
4997 | if (iflink) | |
4998 | dev->iflink = dev->ifindex; | |
4999 | } | |
5000 | ||
8b41d188 | 5001 | /* Fixup kobjects */ |
aaf8cdc3 | 5002 | err = netdev_register_kobject(dev); |
8b41d188 | 5003 | WARN_ON(err); |
ce286d32 EB |
5004 | |
5005 | /* Add the device back in the hashes */ | |
5006 | list_netdevice(dev); | |
5007 | ||
5008 | /* Notify protocols, that a new device appeared. */ | |
5009 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
5010 | ||
5011 | synchronize_net(); | |
5012 | err = 0; | |
5013 | out: | |
5014 | return err; | |
5015 | } | |
5016 | ||
1da177e4 LT |
5017 | static int dev_cpu_callback(struct notifier_block *nfb, |
5018 | unsigned long action, | |
5019 | void *ocpu) | |
5020 | { | |
5021 | struct sk_buff **list_skb; | |
37437bb2 | 5022 | struct Qdisc **list_net; |
1da177e4 LT |
5023 | struct sk_buff *skb; |
5024 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
5025 | struct softnet_data *sd, *oldsd; | |
5026 | ||
8bb78442 | 5027 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
1da177e4 LT |
5028 | return NOTIFY_OK; |
5029 | ||
5030 | local_irq_disable(); | |
5031 | cpu = smp_processor_id(); | |
5032 | sd = &per_cpu(softnet_data, cpu); | |
5033 | oldsd = &per_cpu(softnet_data, oldcpu); | |
5034 | ||
5035 | /* Find end of our completion_queue. */ | |
5036 | list_skb = &sd->completion_queue; | |
5037 | while (*list_skb) | |
5038 | list_skb = &(*list_skb)->next; | |
5039 | /* Append completion queue from offline CPU. */ | |
5040 | *list_skb = oldsd->completion_queue; | |
5041 | oldsd->completion_queue = NULL; | |
5042 | ||
5043 | /* Find end of our output_queue. */ | |
5044 | list_net = &sd->output_queue; | |
5045 | while (*list_net) | |
5046 | list_net = &(*list_net)->next_sched; | |
5047 | /* Append output queue from offline CPU. */ | |
5048 | *list_net = oldsd->output_queue; | |
5049 | oldsd->output_queue = NULL; | |
5050 | ||
5051 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
5052 | local_irq_enable(); | |
5053 | ||
5054 | /* Process offline CPU's input_pkt_queue */ | |
5055 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) | |
5056 | netif_rx(skb); | |
5057 | ||
5058 | return NOTIFY_OK; | |
5059 | } | |
1da177e4 LT |
5060 | |
5061 | ||
7f353bf2 | 5062 | /** |
b63365a2 HX |
5063 | * netdev_increment_features - increment feature set by one |
5064 | * @all: current feature set | |
5065 | * @one: new feature set | |
5066 | * @mask: mask feature set | |
7f353bf2 HX |
5067 | * |
5068 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
5069 | * @one to the master device with current feature set @all. Will not |
5070 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 5071 | */ |
b63365a2 HX |
5072 | unsigned long netdev_increment_features(unsigned long all, unsigned long one, |
5073 | unsigned long mask) | |
5074 | { | |
5075 | /* If device needs checksumming, downgrade to it. */ | |
5076 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) | |
5077 | all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM); | |
5078 | else if (mask & NETIF_F_ALL_CSUM) { | |
5079 | /* If one device supports v4/v6 checksumming, set for all. */ | |
5080 | if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) && | |
5081 | !(all & NETIF_F_GEN_CSUM)) { | |
5082 | all &= ~NETIF_F_ALL_CSUM; | |
5083 | all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); | |
5084 | } | |
e2a6b852 | 5085 | |
b63365a2 HX |
5086 | /* If one device supports hw checksumming, set for all. */ |
5087 | if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) { | |
5088 | all &= ~NETIF_F_ALL_CSUM; | |
5089 | all |= NETIF_F_HW_CSUM; | |
5090 | } | |
5091 | } | |
7f353bf2 | 5092 | |
b63365a2 | 5093 | one |= NETIF_F_ALL_CSUM; |
7f353bf2 | 5094 | |
b63365a2 HX |
5095 | one |= all & NETIF_F_ONE_FOR_ALL; |
5096 | all &= one | NETIF_F_LLTX | NETIF_F_GSO; | |
5097 | all |= one & mask & NETIF_F_ONE_FOR_ALL; | |
7f353bf2 HX |
5098 | |
5099 | return all; | |
5100 | } | |
b63365a2 | 5101 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 5102 | |
30d97d35 PE |
5103 | static struct hlist_head *netdev_create_hash(void) |
5104 | { | |
5105 | int i; | |
5106 | struct hlist_head *hash; | |
5107 | ||
5108 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
5109 | if (hash != NULL) | |
5110 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
5111 | INIT_HLIST_HEAD(&hash[i]); | |
5112 | ||
5113 | return hash; | |
5114 | } | |
5115 | ||
881d966b | 5116 | /* Initialize per network namespace state */ |
4665079c | 5117 | static int __net_init netdev_init(struct net *net) |
881d966b | 5118 | { |
881d966b | 5119 | INIT_LIST_HEAD(&net->dev_base_head); |
881d966b | 5120 | |
30d97d35 PE |
5121 | net->dev_name_head = netdev_create_hash(); |
5122 | if (net->dev_name_head == NULL) | |
5123 | goto err_name; | |
881d966b | 5124 | |
30d97d35 PE |
5125 | net->dev_index_head = netdev_create_hash(); |
5126 | if (net->dev_index_head == NULL) | |
5127 | goto err_idx; | |
881d966b EB |
5128 | |
5129 | return 0; | |
30d97d35 PE |
5130 | |
5131 | err_idx: | |
5132 | kfree(net->dev_name_head); | |
5133 | err_name: | |
5134 | return -ENOMEM; | |
881d966b EB |
5135 | } |
5136 | ||
f0db275a SH |
5137 | /** |
5138 | * netdev_drivername - network driver for the device | |
5139 | * @dev: network device | |
5140 | * @buffer: buffer for resulting name | |
5141 | * @len: size of buffer | |
5142 | * | |
5143 | * Determine network driver for device. | |
5144 | */ | |
cf04a4c7 | 5145 | char *netdev_drivername(const struct net_device *dev, char *buffer, int len) |
6579e57b | 5146 | { |
cf04a4c7 SH |
5147 | const struct device_driver *driver; |
5148 | const struct device *parent; | |
6579e57b AV |
5149 | |
5150 | if (len <= 0 || !buffer) | |
5151 | return buffer; | |
5152 | buffer[0] = 0; | |
5153 | ||
5154 | parent = dev->dev.parent; | |
5155 | ||
5156 | if (!parent) | |
5157 | return buffer; | |
5158 | ||
5159 | driver = parent->driver; | |
5160 | if (driver && driver->name) | |
5161 | strlcpy(buffer, driver->name, len); | |
5162 | return buffer; | |
5163 | } | |
5164 | ||
4665079c | 5165 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
5166 | { |
5167 | kfree(net->dev_name_head); | |
5168 | kfree(net->dev_index_head); | |
5169 | } | |
5170 | ||
022cbae6 | 5171 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
5172 | .init = netdev_init, |
5173 | .exit = netdev_exit, | |
5174 | }; | |
5175 | ||
4665079c | 5176 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 5177 | { |
8eb79863 | 5178 | struct net_device *dev; |
ce286d32 EB |
5179 | /* |
5180 | * Push all migratable of the network devices back to the | |
5181 | * initial network namespace | |
5182 | */ | |
5183 | rtnl_lock(); | |
8eb79863 EB |
5184 | restart: |
5185 | for_each_netdev(net, dev) { | |
ce286d32 | 5186 | int err; |
aca51397 | 5187 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
5188 | |
5189 | /* Ignore unmoveable devices (i.e. loopback) */ | |
5190 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
5191 | continue; | |
5192 | ||
d0c082ce EB |
5193 | /* Delete virtual devices */ |
5194 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) { | |
5195 | dev->rtnl_link_ops->dellink(dev); | |
8eb79863 | 5196 | goto restart; |
d0c082ce EB |
5197 | } |
5198 | ||
ce286d32 | 5199 | /* Push remaing network devices to init_net */ |
aca51397 PE |
5200 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
5201 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 5202 | if (err) { |
aca51397 | 5203 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", |
ce286d32 | 5204 | __func__, dev->name, err); |
aca51397 | 5205 | BUG(); |
ce286d32 | 5206 | } |
8eb79863 | 5207 | goto restart; |
ce286d32 EB |
5208 | } |
5209 | rtnl_unlock(); | |
5210 | } | |
5211 | ||
022cbae6 | 5212 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 EB |
5213 | .exit = default_device_exit, |
5214 | }; | |
5215 | ||
1da177e4 LT |
5216 | /* |
5217 | * Initialize the DEV module. At boot time this walks the device list and | |
5218 | * unhooks any devices that fail to initialise (normally hardware not | |
5219 | * present) and leaves us with a valid list of present and active devices. | |
5220 | * | |
5221 | */ | |
5222 | ||
5223 | /* | |
5224 | * This is called single threaded during boot, so no need | |
5225 | * to take the rtnl semaphore. | |
5226 | */ | |
5227 | static int __init net_dev_init(void) | |
5228 | { | |
5229 | int i, rc = -ENOMEM; | |
5230 | ||
5231 | BUG_ON(!dev_boot_phase); | |
5232 | ||
1da177e4 LT |
5233 | if (dev_proc_init()) |
5234 | goto out; | |
5235 | ||
8b41d188 | 5236 | if (netdev_kobject_init()) |
1da177e4 LT |
5237 | goto out; |
5238 | ||
5239 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 5240 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
5241 | INIT_LIST_HEAD(&ptype_base[i]); |
5242 | ||
881d966b EB |
5243 | if (register_pernet_subsys(&netdev_net_ops)) |
5244 | goto out; | |
1da177e4 LT |
5245 | |
5246 | /* | |
5247 | * Initialise the packet receive queues. | |
5248 | */ | |
5249 | ||
6f912042 | 5250 | for_each_possible_cpu(i) { |
1da177e4 LT |
5251 | struct softnet_data *queue; |
5252 | ||
5253 | queue = &per_cpu(softnet_data, i); | |
5254 | skb_queue_head_init(&queue->input_pkt_queue); | |
1da177e4 LT |
5255 | queue->completion_queue = NULL; |
5256 | INIT_LIST_HEAD(&queue->poll_list); | |
bea3348e SH |
5257 | |
5258 | queue->backlog.poll = process_backlog; | |
5259 | queue->backlog.weight = weight_p; | |
d565b0a1 | 5260 | queue->backlog.gro_list = NULL; |
4ae5544f | 5261 | queue->backlog.gro_count = 0; |
1da177e4 LT |
5262 | } |
5263 | ||
1da177e4 LT |
5264 | dev_boot_phase = 0; |
5265 | ||
505d4f73 EB |
5266 | /* The loopback device is special if any other network devices |
5267 | * is present in a network namespace the loopback device must | |
5268 | * be present. Since we now dynamically allocate and free the | |
5269 | * loopback device ensure this invariant is maintained by | |
5270 | * keeping the loopback device as the first device on the | |
5271 | * list of network devices. Ensuring the loopback devices | |
5272 | * is the first device that appears and the last network device | |
5273 | * that disappears. | |
5274 | */ | |
5275 | if (register_pernet_device(&loopback_net_ops)) | |
5276 | goto out; | |
5277 | ||
5278 | if (register_pernet_device(&default_device_ops)) | |
5279 | goto out; | |
5280 | ||
962cf36c CM |
5281 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
5282 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 LT |
5283 | |
5284 | hotcpu_notifier(dev_cpu_callback, 0); | |
5285 | dst_init(); | |
5286 | dev_mcast_init(); | |
5287 | rc = 0; | |
5288 | out: | |
5289 | return rc; | |
5290 | } | |
5291 | ||
5292 | subsys_initcall(net_dev_init); | |
5293 | ||
5294 | EXPORT_SYMBOL(__dev_get_by_index); | |
5295 | EXPORT_SYMBOL(__dev_get_by_name); | |
5296 | EXPORT_SYMBOL(__dev_remove_pack); | |
c2373ee9 | 5297 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
5298 | EXPORT_SYMBOL(dev_add_pack); |
5299 | EXPORT_SYMBOL(dev_alloc_name); | |
5300 | EXPORT_SYMBOL(dev_close); | |
5301 | EXPORT_SYMBOL(dev_get_by_flags); | |
5302 | EXPORT_SYMBOL(dev_get_by_index); | |
5303 | EXPORT_SYMBOL(dev_get_by_name); | |
1da177e4 LT |
5304 | EXPORT_SYMBOL(dev_open); |
5305 | EXPORT_SYMBOL(dev_queue_xmit); | |
5306 | EXPORT_SYMBOL(dev_remove_pack); | |
5307 | EXPORT_SYMBOL(dev_set_allmulti); | |
5308 | EXPORT_SYMBOL(dev_set_promiscuity); | |
5309 | EXPORT_SYMBOL(dev_change_flags); | |
5310 | EXPORT_SYMBOL(dev_set_mtu); | |
5311 | EXPORT_SYMBOL(dev_set_mac_address); | |
5312 | EXPORT_SYMBOL(free_netdev); | |
5313 | EXPORT_SYMBOL(netdev_boot_setup_check); | |
5314 | EXPORT_SYMBOL(netdev_set_master); | |
5315 | EXPORT_SYMBOL(netdev_state_change); | |
5316 | EXPORT_SYMBOL(netif_receive_skb); | |
5317 | EXPORT_SYMBOL(netif_rx); | |
5318 | EXPORT_SYMBOL(register_gifconf); | |
5319 | EXPORT_SYMBOL(register_netdevice); | |
5320 | EXPORT_SYMBOL(register_netdevice_notifier); | |
5321 | EXPORT_SYMBOL(skb_checksum_help); | |
5322 | EXPORT_SYMBOL(synchronize_net); | |
5323 | EXPORT_SYMBOL(unregister_netdevice); | |
5324 | EXPORT_SYMBOL(unregister_netdevice_notifier); | |
5325 | EXPORT_SYMBOL(net_enable_timestamp); | |
5326 | EXPORT_SYMBOL(net_disable_timestamp); | |
5327 | EXPORT_SYMBOL(dev_get_flags); | |
5328 | ||
5329 | #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE) | |
5330 | EXPORT_SYMBOL(br_handle_frame_hook); | |
5331 | EXPORT_SYMBOL(br_fdb_get_hook); | |
5332 | EXPORT_SYMBOL(br_fdb_put_hook); | |
5333 | #endif | |
5334 | ||
1da177e4 | 5335 | EXPORT_SYMBOL(dev_load); |
1da177e4 LT |
5336 | |
5337 | EXPORT_PER_CPU_SYMBOL(softnet_data); |