]>
Commit | Line | Data |
---|---|---|
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 | |
10 | * Authors: Ross Biro | |
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> | |
78 | #include <linux/capability.h> | |
79 | #include <linux/cpu.h> | |
80 | #include <linux/types.h> | |
81 | #include <linux/kernel.h> | |
82 | #include <linux/sched.h> | |
83 | #include <linux/mutex.h> | |
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> | |
93 | #include <linux/ethtool.h> | |
94 | #include <linux/notifier.h> | |
95 | #include <linux/skbuff.h> | |
96 | #include <net/net_namespace.h> | |
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> | |
103 | #include <linux/if_macvlan.h> | |
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> | |
111 | #include <linux/kallsyms.h> | |
112 | #include <linux/netpoll.h> | |
113 | #include <linux/rcupdate.h> | |
114 | #include <linux/delay.h> | |
115 | #include <net/wext.h> | |
116 | #include <net/iw_handler.h> | |
117 | #include <asm/current.h> | |
118 | #include <linux/audit.h> | |
119 | #include <linux/dmaengine.h> | |
120 | #include <linux/err.h> | |
121 | #include <linux/ctype.h> | |
122 | #include <linux/if_arp.h> | |
123 | #include <linux/if_vlan.h> | |
124 | ||
125 | #include "net-sysfs.h" | |
126 | ||
127 | /* | |
128 | * The list of packet types we will receive (as opposed to discard) | |
129 | * and the routines to invoke. | |
130 | * | |
131 | * Why 16. Because with 16 the only overlap we get on a hash of the | |
132 | * low nibble of the protocol value is RARP/SNAP/X.25. | |
133 | * | |
134 | * NOTE: That is no longer true with the addition of VLAN tags. Not | |
135 | * sure which should go first, but I bet it won't make much | |
136 | * difference if we are running VLANs. The good news is that | |
137 | * this protocol won't be in the list unless compiled in, so | |
138 | * the average user (w/out VLANs) will not be adversely affected. | |
139 | * --BLG | |
140 | * | |
141 | * 0800 IP | |
142 | * 8100 802.1Q VLAN | |
143 | * 0001 802.3 | |
144 | * 0002 AX.25 | |
145 | * 0004 802.2 | |
146 | * 8035 RARP | |
147 | * 0005 SNAP | |
148 | * 0805 X.25 | |
149 | * 0806 ARP | |
150 | * 8137 IPX | |
151 | * 0009 Localtalk | |
152 | * 86DD IPv6 | |
153 | */ | |
154 | ||
155 | #define PTYPE_HASH_SIZE (16) | |
156 | #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1) | |
157 | ||
158 | static DEFINE_SPINLOCK(ptype_lock); | |
159 | static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; | |
160 | static struct list_head ptype_all __read_mostly; /* Taps */ | |
161 | ||
162 | #ifdef CONFIG_NET_DMA | |
163 | struct net_dma { | |
164 | struct dma_client client; | |
165 | spinlock_t lock; | |
166 | cpumask_t channel_mask; | |
167 | struct dma_chan **channels; | |
168 | }; | |
169 | ||
170 | static enum dma_state_client | |
171 | netdev_dma_event(struct dma_client *client, struct dma_chan *chan, | |
172 | enum dma_state state); | |
173 | ||
174 | static struct net_dma net_dma = { | |
175 | .client = { | |
176 | .event_callback = netdev_dma_event, | |
177 | }, | |
178 | }; | |
179 | #endif | |
180 | ||
181 | /* | |
182 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl | |
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 | |
188 | * dev_base_head list, and hold dev_base_lock for writing when they do the | |
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 | */ | |
200 | DEFINE_RWLOCK(dev_base_lock); | |
201 | ||
202 | EXPORT_SYMBOL(dev_base_lock); | |
203 | ||
204 | #define NETDEV_HASHBITS 8 | |
205 | #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS) | |
206 | ||
207 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) | |
208 | { | |
209 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); | |
210 | return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)]; | |
211 | } | |
212 | ||
213 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) | |
214 | { | |
215 | return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)]; | |
216 | } | |
217 | ||
218 | /* Device list insertion */ | |
219 | static int list_netdevice(struct net_device *dev) | |
220 | { | |
221 | struct net *net = dev_net(dev); | |
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 | ||
246 | /* | |
247 | * Our notifier list | |
248 | */ | |
249 | ||
250 | static RAW_NOTIFIER_HEAD(netdev_chain); | |
251 | ||
252 | /* | |
253 | * Device drivers call our routines to queue packets here. We empty the | |
254 | * queue in the local softnet handler. | |
255 | */ | |
256 | ||
257 | DEFINE_PER_CPU(struct softnet_data, softnet_data); | |
258 | ||
259 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
260 | /* | |
261 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class | |
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, | |
278 | ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID, | |
279 | ARPHRD_NONE}; | |
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", | |
295 | "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID", | |
296 | "_xmit_NONE"}; | |
297 | ||
298 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
299 | ||
300 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
301 | { | |
302 | int i; | |
303 | ||
304 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
305 | if (netdev_lock_type[i] == dev_type) | |
306 | return i; | |
307 | /* the last key is used by default */ | |
308 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
309 | } | |
310 | ||
311 | static inline void netdev_set_lockdep_class(spinlock_t *lock, | |
312 | unsigned short dev_type) | |
313 | { | |
314 | int i; | |
315 | ||
316 | i = netdev_lock_pos(dev_type); | |
317 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
318 | netdev_lock_name[i]); | |
319 | } | |
320 | #else | |
321 | static inline void netdev_set_lockdep_class(spinlock_t *lock, | |
322 | unsigned short dev_type) | |
323 | { | |
324 | } | |
325 | #endif | |
326 | ||
327 | /******************************************************************************* | |
328 | ||
329 | Protocol management and registration routines | |
330 | ||
331 | *******************************************************************************/ | |
332 | ||
333 | /* | |
334 | * Add a protocol ID to the list. Now that the input handler is | |
335 | * smarter we can dispense with all the messy stuff that used to be | |
336 | * here. | |
337 | * | |
338 | * BEWARE!!! Protocol handlers, mangling input packets, | |
339 | * MUST BE last in hash buckets and checking protocol handlers | |
340 | * MUST start from promiscuous ptype_all chain in net_bh. | |
341 | * It is true now, do not change it. | |
342 | * Explanation follows: if protocol handler, mangling packet, will | |
343 | * be the first on list, it is not able to sense, that packet | |
344 | * is cloned and should be copied-on-write, so that it will | |
345 | * change it and subsequent readers will get broken packet. | |
346 | * --ANK (980803) | |
347 | */ | |
348 | ||
349 | /** | |
350 | * dev_add_pack - add packet handler | |
351 | * @pt: packet type declaration | |
352 | * | |
353 | * Add a protocol handler to the networking stack. The passed &packet_type | |
354 | * is linked into kernel lists and may not be freed until it has been | |
355 | * removed from the kernel lists. | |
356 | * | |
357 | * This call does not sleep therefore it can not | |
358 | * guarantee all CPU's that are in middle of receiving packets | |
359 | * will see the new packet type (until the next received packet). | |
360 | */ | |
361 | ||
362 | void dev_add_pack(struct packet_type *pt) | |
363 | { | |
364 | int hash; | |
365 | ||
366 | spin_lock_bh(&ptype_lock); | |
367 | if (pt->type == htons(ETH_P_ALL)) | |
368 | list_add_rcu(&pt->list, &ptype_all); | |
369 | else { | |
370 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; | |
371 | list_add_rcu(&pt->list, &ptype_base[hash]); | |
372 | } | |
373 | spin_unlock_bh(&ptype_lock); | |
374 | } | |
375 | ||
376 | /** | |
377 | * __dev_remove_pack - remove packet handler | |
378 | * @pt: packet type declaration | |
379 | * | |
380 | * Remove a protocol handler that was previously added to the kernel | |
381 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
382 | * from the kernel lists and can be freed or reused once this function | |
383 | * returns. | |
384 | * | |
385 | * The packet type might still be in use by receivers | |
386 | * and must not be freed until after all the CPU's have gone | |
387 | * through a quiescent state. | |
388 | */ | |
389 | void __dev_remove_pack(struct packet_type *pt) | |
390 | { | |
391 | struct list_head *head; | |
392 | struct packet_type *pt1; | |
393 | ||
394 | spin_lock_bh(&ptype_lock); | |
395 | ||
396 | if (pt->type == htons(ETH_P_ALL)) | |
397 | head = &ptype_all; | |
398 | else | |
399 | head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
400 | ||
401 | list_for_each_entry(pt1, head, list) { | |
402 | if (pt == pt1) { | |
403 | list_del_rcu(&pt->list); | |
404 | goto out; | |
405 | } | |
406 | } | |
407 | ||
408 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); | |
409 | out: | |
410 | spin_unlock_bh(&ptype_lock); | |
411 | } | |
412 | /** | |
413 | * dev_remove_pack - remove packet handler | |
414 | * @pt: packet type declaration | |
415 | * | |
416 | * Remove a protocol handler that was previously added to the kernel | |
417 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
418 | * from the kernel lists and can be freed or reused once this function | |
419 | * returns. | |
420 | * | |
421 | * This call sleeps to guarantee that no CPU is looking at the packet | |
422 | * type after return. | |
423 | */ | |
424 | void dev_remove_pack(struct packet_type *pt) | |
425 | { | |
426 | __dev_remove_pack(pt); | |
427 | ||
428 | synchronize_net(); | |
429 | } | |
430 | ||
431 | /****************************************************************************** | |
432 | ||
433 | Device Boot-time Settings Routines | |
434 | ||
435 | *******************************************************************************/ | |
436 | ||
437 | /* Boot time configuration table */ | |
438 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
439 | ||
440 | /** | |
441 | * netdev_boot_setup_add - add new setup entry | |
442 | * @name: name of the device | |
443 | * @map: configured settings for the device | |
444 | * | |
445 | * Adds new setup entry to the dev_boot_setup list. The function | |
446 | * returns 0 on error and 1 on success. This is a generic routine to | |
447 | * all netdevices. | |
448 | */ | |
449 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
450 | { | |
451 | struct netdev_boot_setup *s; | |
452 | int i; | |
453 | ||
454 | s = dev_boot_setup; | |
455 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
456 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
457 | memset(s[i].name, 0, sizeof(s[i].name)); | |
458 | strlcpy(s[i].name, name, IFNAMSIZ); | |
459 | memcpy(&s[i].map, map, sizeof(s[i].map)); | |
460 | break; | |
461 | } | |
462 | } | |
463 | ||
464 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
465 | } | |
466 | ||
467 | /** | |
468 | * netdev_boot_setup_check - check boot time settings | |
469 | * @dev: the netdevice | |
470 | * | |
471 | * Check boot time settings for the device. | |
472 | * The found settings are set for the device to be used | |
473 | * later in the device probing. | |
474 | * Returns 0 if no settings found, 1 if they are. | |
475 | */ | |
476 | int netdev_boot_setup_check(struct net_device *dev) | |
477 | { | |
478 | struct netdev_boot_setup *s = dev_boot_setup; | |
479 | int i; | |
480 | ||
481 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
482 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
483 | !strcmp(dev->name, s[i].name)) { | |
484 | dev->irq = s[i].map.irq; | |
485 | dev->base_addr = s[i].map.base_addr; | |
486 | dev->mem_start = s[i].map.mem_start; | |
487 | dev->mem_end = s[i].map.mem_end; | |
488 | return 1; | |
489 | } | |
490 | } | |
491 | return 0; | |
492 | } | |
493 | ||
494 | ||
495 | /** | |
496 | * netdev_boot_base - get address from boot time settings | |
497 | * @prefix: prefix for network device | |
498 | * @unit: id for network device | |
499 | * | |
500 | * Check boot time settings for the base address of device. | |
501 | * The found settings are set for the device to be used | |
502 | * later in the device probing. | |
503 | * Returns 0 if no settings found. | |
504 | */ | |
505 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
506 | { | |
507 | const struct netdev_boot_setup *s = dev_boot_setup; | |
508 | char name[IFNAMSIZ]; | |
509 | int i; | |
510 | ||
511 | sprintf(name, "%s%d", prefix, unit); | |
512 | ||
513 | /* | |
514 | * If device already registered then return base of 1 | |
515 | * to indicate not to probe for this interface | |
516 | */ | |
517 | if (__dev_get_by_name(&init_net, name)) | |
518 | return 1; | |
519 | ||
520 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
521 | if (!strcmp(name, s[i].name)) | |
522 | return s[i].map.base_addr; | |
523 | return 0; | |
524 | } | |
525 | ||
526 | /* | |
527 | * Saves at boot time configured settings for any netdevice. | |
528 | */ | |
529 | int __init netdev_boot_setup(char *str) | |
530 | { | |
531 | int ints[5]; | |
532 | struct ifmap map; | |
533 | ||
534 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
535 | if (!str || !*str) | |
536 | return 0; | |
537 | ||
538 | /* Save settings */ | |
539 | memset(&map, 0, sizeof(map)); | |
540 | if (ints[0] > 0) | |
541 | map.irq = ints[1]; | |
542 | if (ints[0] > 1) | |
543 | map.base_addr = ints[2]; | |
544 | if (ints[0] > 2) | |
545 | map.mem_start = ints[3]; | |
546 | if (ints[0] > 3) | |
547 | map.mem_end = ints[4]; | |
548 | ||
549 | /* Add new entry to the list */ | |
550 | return netdev_boot_setup_add(str, &map); | |
551 | } | |
552 | ||
553 | __setup("netdev=", netdev_boot_setup); | |
554 | ||
555 | /******************************************************************************* | |
556 | ||
557 | Device Interface Subroutines | |
558 | ||
559 | *******************************************************************************/ | |
560 | ||
561 | /** | |
562 | * __dev_get_by_name - find a device by its name | |
563 | * @net: the applicable net namespace | |
564 | * @name: name to find | |
565 | * | |
566 | * Find an interface by name. Must be called under RTNL semaphore | |
567 | * or @dev_base_lock. If the name is found a pointer to the device | |
568 | * is returned. If the name is not found then %NULL is returned. The | |
569 | * reference counters are not incremented so the caller must be | |
570 | * careful with locks. | |
571 | */ | |
572 | ||
573 | struct net_device *__dev_get_by_name(struct net *net, const char *name) | |
574 | { | |
575 | struct hlist_node *p; | |
576 | ||
577 | hlist_for_each(p, dev_name_hash(net, name)) { | |
578 | struct net_device *dev | |
579 | = hlist_entry(p, struct net_device, name_hlist); | |
580 | if (!strncmp(dev->name, name, IFNAMSIZ)) | |
581 | return dev; | |
582 | } | |
583 | return NULL; | |
584 | } | |
585 | ||
586 | /** | |
587 | * dev_get_by_name - find a device by its name | |
588 | * @net: the applicable net namespace | |
589 | * @name: name to find | |
590 | * | |
591 | * Find an interface by name. This can be called from any | |
592 | * context and does its own locking. The returned handle has | |
593 | * the usage count incremented and the caller must use dev_put() to | |
594 | * release it when it is no longer needed. %NULL is returned if no | |
595 | * matching device is found. | |
596 | */ | |
597 | ||
598 | struct net_device *dev_get_by_name(struct net *net, const char *name) | |
599 | { | |
600 | struct net_device *dev; | |
601 | ||
602 | read_lock(&dev_base_lock); | |
603 | dev = __dev_get_by_name(net, name); | |
604 | if (dev) | |
605 | dev_hold(dev); | |
606 | read_unlock(&dev_base_lock); | |
607 | return dev; | |
608 | } | |
609 | ||
610 | /** | |
611 | * __dev_get_by_index - find a device by its ifindex | |
612 | * @net: the applicable net namespace | |
613 | * @ifindex: index of device | |
614 | * | |
615 | * Search for an interface by index. Returns %NULL if the device | |
616 | * is not found or a pointer to the device. The device has not | |
617 | * had its reference counter increased so the caller must be careful | |
618 | * about locking. The caller must hold either the RTNL semaphore | |
619 | * or @dev_base_lock. | |
620 | */ | |
621 | ||
622 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) | |
623 | { | |
624 | struct hlist_node *p; | |
625 | ||
626 | hlist_for_each(p, dev_index_hash(net, ifindex)) { | |
627 | struct net_device *dev | |
628 | = hlist_entry(p, struct net_device, index_hlist); | |
629 | if (dev->ifindex == ifindex) | |
630 | return dev; | |
631 | } | |
632 | return NULL; | |
633 | } | |
634 | ||
635 | ||
636 | /** | |
637 | * dev_get_by_index - find a device by its ifindex | |
638 | * @net: the applicable net namespace | |
639 | * @ifindex: index of device | |
640 | * | |
641 | * Search for an interface by index. Returns NULL if the device | |
642 | * is not found or a pointer to the device. The device returned has | |
643 | * had a reference added and the pointer is safe until the user calls | |
644 | * dev_put to indicate they have finished with it. | |
645 | */ | |
646 | ||
647 | struct net_device *dev_get_by_index(struct net *net, int ifindex) | |
648 | { | |
649 | struct net_device *dev; | |
650 | ||
651 | read_lock(&dev_base_lock); | |
652 | dev = __dev_get_by_index(net, ifindex); | |
653 | if (dev) | |
654 | dev_hold(dev); | |
655 | read_unlock(&dev_base_lock); | |
656 | return dev; | |
657 | } | |
658 | ||
659 | /** | |
660 | * dev_getbyhwaddr - find a device by its hardware address | |
661 | * @net: the applicable net namespace | |
662 | * @type: media type of device | |
663 | * @ha: hardware address | |
664 | * | |
665 | * Search for an interface by MAC address. Returns NULL if the device | |
666 | * is not found or a pointer to the device. The caller must hold the | |
667 | * rtnl semaphore. The returned device has not had its ref count increased | |
668 | * and the caller must therefore be careful about locking | |
669 | * | |
670 | * BUGS: | |
671 | * If the API was consistent this would be __dev_get_by_hwaddr | |
672 | */ | |
673 | ||
674 | struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha) | |
675 | { | |
676 | struct net_device *dev; | |
677 | ||
678 | ASSERT_RTNL(); | |
679 | ||
680 | for_each_netdev(net, dev) | |
681 | if (dev->type == type && | |
682 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
683 | return dev; | |
684 | ||
685 | return NULL; | |
686 | } | |
687 | ||
688 | EXPORT_SYMBOL(dev_getbyhwaddr); | |
689 | ||
690 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) | |
691 | { | |
692 | struct net_device *dev; | |
693 | ||
694 | ASSERT_RTNL(); | |
695 | for_each_netdev(net, dev) | |
696 | if (dev->type == type) | |
697 | return dev; | |
698 | ||
699 | return NULL; | |
700 | } | |
701 | ||
702 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); | |
703 | ||
704 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) | |
705 | { | |
706 | struct net_device *dev; | |
707 | ||
708 | rtnl_lock(); | |
709 | dev = __dev_getfirstbyhwtype(net, type); | |
710 | if (dev) | |
711 | dev_hold(dev); | |
712 | rtnl_unlock(); | |
713 | return dev; | |
714 | } | |
715 | ||
716 | EXPORT_SYMBOL(dev_getfirstbyhwtype); | |
717 | ||
718 | /** | |
719 | * dev_get_by_flags - find any device with given flags | |
720 | * @net: the applicable net namespace | |
721 | * @if_flags: IFF_* values | |
722 | * @mask: bitmask of bits in if_flags to check | |
723 | * | |
724 | * Search for any interface with the given flags. Returns NULL if a device | |
725 | * is not found or a pointer to the device. The device returned has | |
726 | * had a reference added and the pointer is safe until the user calls | |
727 | * dev_put to indicate they have finished with it. | |
728 | */ | |
729 | ||
730 | struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask) | |
731 | { | |
732 | struct net_device *dev, *ret; | |
733 | ||
734 | ret = NULL; | |
735 | read_lock(&dev_base_lock); | |
736 | for_each_netdev(net, dev) { | |
737 | if (((dev->flags ^ if_flags) & mask) == 0) { | |
738 | dev_hold(dev); | |
739 | ret = dev; | |
740 | break; | |
741 | } | |
742 | } | |
743 | read_unlock(&dev_base_lock); | |
744 | return ret; | |
745 | } | |
746 | ||
747 | /** | |
748 | * dev_valid_name - check if name is okay for network device | |
749 | * @name: name string | |
750 | * | |
751 | * Network device names need to be valid file names to | |
752 | * to allow sysfs to work. We also disallow any kind of | |
753 | * whitespace. | |
754 | */ | |
755 | int dev_valid_name(const char *name) | |
756 | { | |
757 | if (*name == '\0') | |
758 | return 0; | |
759 | if (strlen(name) >= IFNAMSIZ) | |
760 | return 0; | |
761 | if (!strcmp(name, ".") || !strcmp(name, "..")) | |
762 | return 0; | |
763 | ||
764 | while (*name) { | |
765 | if (*name == '/' || isspace(*name)) | |
766 | return 0; | |
767 | name++; | |
768 | } | |
769 | return 1; | |
770 | } | |
771 | ||
772 | /** | |
773 | * __dev_alloc_name - allocate a name for a device | |
774 | * @net: network namespace to allocate the device name in | |
775 | * @name: name format string | |
776 | * @buf: scratch buffer and result name string | |
777 | * | |
778 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
779 | * id. It scans list of devices to build up a free map, then chooses | |
780 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
781 | * while allocating the name and adding the device in order to avoid | |
782 | * duplicates. | |
783 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
784 | * Returns the number of the unit assigned or a negative errno code. | |
785 | */ | |
786 | ||
787 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) | |
788 | { | |
789 | int i = 0; | |
790 | const char *p; | |
791 | const int max_netdevices = 8*PAGE_SIZE; | |
792 | unsigned long *inuse; | |
793 | struct net_device *d; | |
794 | ||
795 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
796 | if (p) { | |
797 | /* | |
798 | * Verify the string as this thing may have come from | |
799 | * the user. There must be either one "%d" and no other "%" | |
800 | * characters. | |
801 | */ | |
802 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
803 | return -EINVAL; | |
804 | ||
805 | /* Use one page as a bit array of possible slots */ | |
806 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); | |
807 | if (!inuse) | |
808 | return -ENOMEM; | |
809 | ||
810 | for_each_netdev(net, d) { | |
811 | if (!sscanf(d->name, name, &i)) | |
812 | continue; | |
813 | if (i < 0 || i >= max_netdevices) | |
814 | continue; | |
815 | ||
816 | /* avoid cases where sscanf is not exact inverse of printf */ | |
817 | snprintf(buf, IFNAMSIZ, name, i); | |
818 | if (!strncmp(buf, d->name, IFNAMSIZ)) | |
819 | set_bit(i, inuse); | |
820 | } | |
821 | ||
822 | i = find_first_zero_bit(inuse, max_netdevices); | |
823 | free_page((unsigned long) inuse); | |
824 | } | |
825 | ||
826 | snprintf(buf, IFNAMSIZ, name, i); | |
827 | if (!__dev_get_by_name(net, buf)) | |
828 | return i; | |
829 | ||
830 | /* It is possible to run out of possible slots | |
831 | * when the name is long and there isn't enough space left | |
832 | * for the digits, or if all bits are used. | |
833 | */ | |
834 | return -ENFILE; | |
835 | } | |
836 | ||
837 | /** | |
838 | * dev_alloc_name - allocate a name for a device | |
839 | * @dev: device | |
840 | * @name: name format string | |
841 | * | |
842 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
843 | * id. It scans list of devices to build up a free map, then chooses | |
844 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
845 | * while allocating the name and adding the device in order to avoid | |
846 | * duplicates. | |
847 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
848 | * Returns the number of the unit assigned or a negative errno code. | |
849 | */ | |
850 | ||
851 | int dev_alloc_name(struct net_device *dev, const char *name) | |
852 | { | |
853 | char buf[IFNAMSIZ]; | |
854 | struct net *net; | |
855 | int ret; | |
856 | ||
857 | BUG_ON(!dev_net(dev)); | |
858 | net = dev_net(dev); | |
859 | ret = __dev_alloc_name(net, name, buf); | |
860 | if (ret >= 0) | |
861 | strlcpy(dev->name, buf, IFNAMSIZ); | |
862 | return ret; | |
863 | } | |
864 | ||
865 | ||
866 | /** | |
867 | * dev_change_name - change name of a device | |
868 | * @dev: device | |
869 | * @newname: name (or format string) must be at least IFNAMSIZ | |
870 | * | |
871 | * Change name of a device, can pass format strings "eth%d". | |
872 | * for wildcarding. | |
873 | */ | |
874 | int dev_change_name(struct net_device *dev, char *newname) | |
875 | { | |
876 | char oldname[IFNAMSIZ]; | |
877 | int err = 0; | |
878 | int ret; | |
879 | struct net *net; | |
880 | ||
881 | ASSERT_RTNL(); | |
882 | BUG_ON(!dev_net(dev)); | |
883 | ||
884 | net = dev_net(dev); | |
885 | if (dev->flags & IFF_UP) | |
886 | return -EBUSY; | |
887 | ||
888 | if (!dev_valid_name(newname)) | |
889 | return -EINVAL; | |
890 | ||
891 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) | |
892 | return 0; | |
893 | ||
894 | memcpy(oldname, dev->name, IFNAMSIZ); | |
895 | ||
896 | if (strchr(newname, '%')) { | |
897 | err = dev_alloc_name(dev, newname); | |
898 | if (err < 0) | |
899 | return err; | |
900 | strcpy(newname, dev->name); | |
901 | } | |
902 | else if (__dev_get_by_name(net, newname)) | |
903 | return -EEXIST; | |
904 | else | |
905 | strlcpy(dev->name, newname, IFNAMSIZ); | |
906 | ||
907 | rollback: | |
908 | err = device_rename(&dev->dev, dev->name); | |
909 | if (err) { | |
910 | memcpy(dev->name, oldname, IFNAMSIZ); | |
911 | return err; | |
912 | } | |
913 | ||
914 | write_lock_bh(&dev_base_lock); | |
915 | hlist_del(&dev->name_hlist); | |
916 | hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
917 | write_unlock_bh(&dev_base_lock); | |
918 | ||
919 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); | |
920 | ret = notifier_to_errno(ret); | |
921 | ||
922 | if (ret) { | |
923 | if (err) { | |
924 | printk(KERN_ERR | |
925 | "%s: name change rollback failed: %d.\n", | |
926 | dev->name, ret); | |
927 | } else { | |
928 | err = ret; | |
929 | memcpy(dev->name, oldname, IFNAMSIZ); | |
930 | goto rollback; | |
931 | } | |
932 | } | |
933 | ||
934 | return err; | |
935 | } | |
936 | ||
937 | /** | |
938 | * netdev_features_change - device changes features | |
939 | * @dev: device to cause notification | |
940 | * | |
941 | * Called to indicate a device has changed features. | |
942 | */ | |
943 | void netdev_features_change(struct net_device *dev) | |
944 | { | |
945 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); | |
946 | } | |
947 | EXPORT_SYMBOL(netdev_features_change); | |
948 | ||
949 | /** | |
950 | * netdev_state_change - device changes state | |
951 | * @dev: device to cause notification | |
952 | * | |
953 | * Called to indicate a device has changed state. This function calls | |
954 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
955 | * to the routing socket. | |
956 | */ | |
957 | void netdev_state_change(struct net_device *dev) | |
958 | { | |
959 | if (dev->flags & IFF_UP) { | |
960 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
961 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); | |
962 | } | |
963 | } | |
964 | ||
965 | void netdev_bonding_change(struct net_device *dev) | |
966 | { | |
967 | call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev); | |
968 | } | |
969 | EXPORT_SYMBOL(netdev_bonding_change); | |
970 | ||
971 | /** | |
972 | * dev_load - load a network module | |
973 | * @net: the applicable net namespace | |
974 | * @name: name of interface | |
975 | * | |
976 | * If a network interface is not present and the process has suitable | |
977 | * privileges this function loads the module. If module loading is not | |
978 | * available in this kernel then it becomes a nop. | |
979 | */ | |
980 | ||
981 | void dev_load(struct net *net, const char *name) | |
982 | { | |
983 | struct net_device *dev; | |
984 | ||
985 | read_lock(&dev_base_lock); | |
986 | dev = __dev_get_by_name(net, name); | |
987 | read_unlock(&dev_base_lock); | |
988 | ||
989 | if (!dev && capable(CAP_SYS_MODULE)) | |
990 | request_module("%s", name); | |
991 | } | |
992 | ||
993 | /** | |
994 | * dev_open - prepare an interface for use. | |
995 | * @dev: device to open | |
996 | * | |
997 | * Takes a device from down to up state. The device's private open | |
998 | * function is invoked and then the multicast lists are loaded. Finally | |
999 | * the device is moved into the up state and a %NETDEV_UP message is | |
1000 | * sent to the netdev notifier chain. | |
1001 | * | |
1002 | * Calling this function on an active interface is a nop. On a failure | |
1003 | * a negative errno code is returned. | |
1004 | */ | |
1005 | int dev_open(struct net_device *dev) | |
1006 | { | |
1007 | int ret = 0; | |
1008 | ||
1009 | ASSERT_RTNL(); | |
1010 | ||
1011 | /* | |
1012 | * Is it already up? | |
1013 | */ | |
1014 | ||
1015 | if (dev->flags & IFF_UP) | |
1016 | return 0; | |
1017 | ||
1018 | /* | |
1019 | * Is it even present? | |
1020 | */ | |
1021 | if (!netif_device_present(dev)) | |
1022 | return -ENODEV; | |
1023 | ||
1024 | /* | |
1025 | * Call device private open method | |
1026 | */ | |
1027 | set_bit(__LINK_STATE_START, &dev->state); | |
1028 | ||
1029 | if (dev->validate_addr) | |
1030 | ret = dev->validate_addr(dev); | |
1031 | ||
1032 | if (!ret && dev->open) | |
1033 | ret = dev->open(dev); | |
1034 | ||
1035 | /* | |
1036 | * If it went open OK then: | |
1037 | */ | |
1038 | ||
1039 | if (ret) | |
1040 | clear_bit(__LINK_STATE_START, &dev->state); | |
1041 | else { | |
1042 | /* | |
1043 | * Set the flags. | |
1044 | */ | |
1045 | dev->flags |= IFF_UP; | |
1046 | ||
1047 | /* | |
1048 | * Initialize multicasting status | |
1049 | */ | |
1050 | dev_set_rx_mode(dev); | |
1051 | ||
1052 | /* | |
1053 | * Wakeup transmit queue engine | |
1054 | */ | |
1055 | dev_activate(dev); | |
1056 | ||
1057 | /* | |
1058 | * ... and announce new interface. | |
1059 | */ | |
1060 | call_netdevice_notifiers(NETDEV_UP, dev); | |
1061 | } | |
1062 | ||
1063 | return ret; | |
1064 | } | |
1065 | ||
1066 | /** | |
1067 | * dev_close - shutdown an interface. | |
1068 | * @dev: device to shutdown | |
1069 | * | |
1070 | * This function moves an active device into down state. A | |
1071 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1072 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1073 | * chain. | |
1074 | */ | |
1075 | int dev_close(struct net_device *dev) | |
1076 | { | |
1077 | ASSERT_RTNL(); | |
1078 | ||
1079 | might_sleep(); | |
1080 | ||
1081 | if (!(dev->flags & IFF_UP)) | |
1082 | return 0; | |
1083 | ||
1084 | /* | |
1085 | * Tell people we are going down, so that they can | |
1086 | * prepare to death, when device is still operating. | |
1087 | */ | |
1088 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); | |
1089 | ||
1090 | clear_bit(__LINK_STATE_START, &dev->state); | |
1091 | ||
1092 | /* Synchronize to scheduled poll. We cannot touch poll list, | |
1093 | * it can be even on different cpu. So just clear netif_running(). | |
1094 | * | |
1095 | * dev->stop() will invoke napi_disable() on all of it's | |
1096 | * napi_struct instances on this device. | |
1097 | */ | |
1098 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ | |
1099 | ||
1100 | dev_deactivate(dev); | |
1101 | ||
1102 | /* | |
1103 | * Call the device specific close. This cannot fail. | |
1104 | * Only if device is UP | |
1105 | * | |
1106 | * We allow it to be called even after a DETACH hot-plug | |
1107 | * event. | |
1108 | */ | |
1109 | if (dev->stop) | |
1110 | dev->stop(dev); | |
1111 | ||
1112 | /* | |
1113 | * Device is now down. | |
1114 | */ | |
1115 | ||
1116 | dev->flags &= ~IFF_UP; | |
1117 | ||
1118 | /* | |
1119 | * Tell people we are down | |
1120 | */ | |
1121 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
1122 | ||
1123 | return 0; | |
1124 | } | |
1125 | ||
1126 | ||
1127 | /** | |
1128 | * dev_disable_lro - disable Large Receive Offload on a device | |
1129 | * @dev: device | |
1130 | * | |
1131 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1132 | * called under RTNL. This is needed if received packets may be | |
1133 | * forwarded to another interface. | |
1134 | */ | |
1135 | void dev_disable_lro(struct net_device *dev) | |
1136 | { | |
1137 | if (dev->ethtool_ops && dev->ethtool_ops->get_flags && | |
1138 | dev->ethtool_ops->set_flags) { | |
1139 | u32 flags = dev->ethtool_ops->get_flags(dev); | |
1140 | if (flags & ETH_FLAG_LRO) { | |
1141 | flags &= ~ETH_FLAG_LRO; | |
1142 | dev->ethtool_ops->set_flags(dev, flags); | |
1143 | } | |
1144 | } | |
1145 | WARN_ON(dev->features & NETIF_F_LRO); | |
1146 | } | |
1147 | EXPORT_SYMBOL(dev_disable_lro); | |
1148 | ||
1149 | ||
1150 | static int dev_boot_phase = 1; | |
1151 | ||
1152 | /* | |
1153 | * Device change register/unregister. These are not inline or static | |
1154 | * as we export them to the world. | |
1155 | */ | |
1156 | ||
1157 | /** | |
1158 | * register_netdevice_notifier - register a network notifier block | |
1159 | * @nb: notifier | |
1160 | * | |
1161 | * Register a notifier to be called when network device events occur. | |
1162 | * The notifier passed is linked into the kernel structures and must | |
1163 | * not be reused until it has been unregistered. A negative errno code | |
1164 | * is returned on a failure. | |
1165 | * | |
1166 | * When registered all registration and up events are replayed | |
1167 | * to the new notifier to allow device to have a race free | |
1168 | * view of the network device list. | |
1169 | */ | |
1170 | ||
1171 | int register_netdevice_notifier(struct notifier_block *nb) | |
1172 | { | |
1173 | struct net_device *dev; | |
1174 | struct net_device *last; | |
1175 | struct net *net; | |
1176 | int err; | |
1177 | ||
1178 | rtnl_lock(); | |
1179 | err = raw_notifier_chain_register(&netdev_chain, nb); | |
1180 | if (err) | |
1181 | goto unlock; | |
1182 | if (dev_boot_phase) | |
1183 | goto unlock; | |
1184 | for_each_net(net) { | |
1185 | for_each_netdev(net, dev) { | |
1186 | err = nb->notifier_call(nb, NETDEV_REGISTER, dev); | |
1187 | err = notifier_to_errno(err); | |
1188 | if (err) | |
1189 | goto rollback; | |
1190 | ||
1191 | if (!(dev->flags & IFF_UP)) | |
1192 | continue; | |
1193 | ||
1194 | nb->notifier_call(nb, NETDEV_UP, dev); | |
1195 | } | |
1196 | } | |
1197 | ||
1198 | unlock: | |
1199 | rtnl_unlock(); | |
1200 | return err; | |
1201 | ||
1202 | rollback: | |
1203 | last = dev; | |
1204 | for_each_net(net) { | |
1205 | for_each_netdev(net, dev) { | |
1206 | if (dev == last) | |
1207 | break; | |
1208 | ||
1209 | if (dev->flags & IFF_UP) { | |
1210 | nb->notifier_call(nb, NETDEV_GOING_DOWN, dev); | |
1211 | nb->notifier_call(nb, NETDEV_DOWN, dev); | |
1212 | } | |
1213 | nb->notifier_call(nb, NETDEV_UNREGISTER, dev); | |
1214 | } | |
1215 | } | |
1216 | ||
1217 | raw_notifier_chain_unregister(&netdev_chain, nb); | |
1218 | goto unlock; | |
1219 | } | |
1220 | ||
1221 | /** | |
1222 | * unregister_netdevice_notifier - unregister a network notifier block | |
1223 | * @nb: notifier | |
1224 | * | |
1225 | * Unregister a notifier previously registered by | |
1226 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1227 | * kernel structures and may then be reused. A negative errno code | |
1228 | * is returned on a failure. | |
1229 | */ | |
1230 | ||
1231 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1232 | { | |
1233 | int err; | |
1234 | ||
1235 | rtnl_lock(); | |
1236 | err = raw_notifier_chain_unregister(&netdev_chain, nb); | |
1237 | rtnl_unlock(); | |
1238 | return err; | |
1239 | } | |
1240 | ||
1241 | /** | |
1242 | * call_netdevice_notifiers - call all network notifier blocks | |
1243 | * @val: value passed unmodified to notifier function | |
1244 | * @dev: net_device pointer passed unmodified to notifier function | |
1245 | * | |
1246 | * Call all network notifier blocks. Parameters and return value | |
1247 | * are as for raw_notifier_call_chain(). | |
1248 | */ | |
1249 | ||
1250 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) | |
1251 | { | |
1252 | return raw_notifier_call_chain(&netdev_chain, val, dev); | |
1253 | } | |
1254 | ||
1255 | /* When > 0 there are consumers of rx skb time stamps */ | |
1256 | static atomic_t netstamp_needed = ATOMIC_INIT(0); | |
1257 | ||
1258 | void net_enable_timestamp(void) | |
1259 | { | |
1260 | atomic_inc(&netstamp_needed); | |
1261 | } | |
1262 | ||
1263 | void net_disable_timestamp(void) | |
1264 | { | |
1265 | atomic_dec(&netstamp_needed); | |
1266 | } | |
1267 | ||
1268 | static inline void net_timestamp(struct sk_buff *skb) | |
1269 | { | |
1270 | if (atomic_read(&netstamp_needed)) | |
1271 | __net_timestamp(skb); | |
1272 | else | |
1273 | skb->tstamp.tv64 = 0; | |
1274 | } | |
1275 | ||
1276 | /* | |
1277 | * Support routine. Sends outgoing frames to any network | |
1278 | * taps currently in use. | |
1279 | */ | |
1280 | ||
1281 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) | |
1282 | { | |
1283 | struct packet_type *ptype; | |
1284 | ||
1285 | net_timestamp(skb); | |
1286 | ||
1287 | rcu_read_lock(); | |
1288 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
1289 | /* Never send packets back to the socket | |
1290 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1291 | */ | |
1292 | if ((ptype->dev == dev || !ptype->dev) && | |
1293 | (ptype->af_packet_priv == NULL || | |
1294 | (struct sock *)ptype->af_packet_priv != skb->sk)) { | |
1295 | struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC); | |
1296 | if (!skb2) | |
1297 | break; | |
1298 | ||
1299 | /* skb->nh should be correctly | |
1300 | set by sender, so that the second statement is | |
1301 | just protection against buggy protocols. | |
1302 | */ | |
1303 | skb_reset_mac_header(skb2); | |
1304 | ||
1305 | if (skb_network_header(skb2) < skb2->data || | |
1306 | skb2->network_header > skb2->tail) { | |
1307 | if (net_ratelimit()) | |
1308 | printk(KERN_CRIT "protocol %04x is " | |
1309 | "buggy, dev %s\n", | |
1310 | skb2->protocol, dev->name); | |
1311 | skb_reset_network_header(skb2); | |
1312 | } | |
1313 | ||
1314 | skb2->transport_header = skb2->network_header; | |
1315 | skb2->pkt_type = PACKET_OUTGOING; | |
1316 | ptype->func(skb2, skb->dev, ptype, skb->dev); | |
1317 | } | |
1318 | } | |
1319 | rcu_read_unlock(); | |
1320 | } | |
1321 | ||
1322 | ||
1323 | void __netif_schedule(struct netdev_queue *txq) | |
1324 | { | |
1325 | struct net_device *dev = txq->dev; | |
1326 | ||
1327 | if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) { | |
1328 | struct softnet_data *sd; | |
1329 | unsigned long flags; | |
1330 | ||
1331 | local_irq_save(flags); | |
1332 | sd = &__get_cpu_var(softnet_data); | |
1333 | txq->next_sched = sd->output_queue; | |
1334 | sd->output_queue = txq; | |
1335 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1336 | local_irq_restore(flags); | |
1337 | } | |
1338 | } | |
1339 | EXPORT_SYMBOL(__netif_schedule); | |
1340 | ||
1341 | void dev_kfree_skb_irq(struct sk_buff *skb) | |
1342 | { | |
1343 | if (atomic_dec_and_test(&skb->users)) { | |
1344 | struct softnet_data *sd; | |
1345 | unsigned long flags; | |
1346 | ||
1347 | local_irq_save(flags); | |
1348 | sd = &__get_cpu_var(softnet_data); | |
1349 | skb->next = sd->completion_queue; | |
1350 | sd->completion_queue = skb; | |
1351 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
1352 | local_irq_restore(flags); | |
1353 | } | |
1354 | } | |
1355 | EXPORT_SYMBOL(dev_kfree_skb_irq); | |
1356 | ||
1357 | void dev_kfree_skb_any(struct sk_buff *skb) | |
1358 | { | |
1359 | if (in_irq() || irqs_disabled()) | |
1360 | dev_kfree_skb_irq(skb); | |
1361 | else | |
1362 | dev_kfree_skb(skb); | |
1363 | } | |
1364 | EXPORT_SYMBOL(dev_kfree_skb_any); | |
1365 | ||
1366 | ||
1367 | /** | |
1368 | * netif_device_detach - mark device as removed | |
1369 | * @dev: network device | |
1370 | * | |
1371 | * Mark device as removed from system and therefore no longer available. | |
1372 | */ | |
1373 | void netif_device_detach(struct net_device *dev) | |
1374 | { | |
1375 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1376 | netif_running(dev)) { | |
1377 | netif_stop_queue(dev); | |
1378 | } | |
1379 | } | |
1380 | EXPORT_SYMBOL(netif_device_detach); | |
1381 | ||
1382 | /** | |
1383 | * netif_device_attach - mark device as attached | |
1384 | * @dev: network device | |
1385 | * | |
1386 | * Mark device as attached from system and restart if needed. | |
1387 | */ | |
1388 | void netif_device_attach(struct net_device *dev) | |
1389 | { | |
1390 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
1391 | netif_running(dev)) { | |
1392 | netif_wake_queue(dev); | |
1393 | __netdev_watchdog_up(dev); | |
1394 | } | |
1395 | } | |
1396 | EXPORT_SYMBOL(netif_device_attach); | |
1397 | ||
1398 | static bool can_checksum_protocol(unsigned long features, __be16 protocol) | |
1399 | { | |
1400 | return ((features & NETIF_F_GEN_CSUM) || | |
1401 | ((features & NETIF_F_IP_CSUM) && | |
1402 | protocol == htons(ETH_P_IP)) || | |
1403 | ((features & NETIF_F_IPV6_CSUM) && | |
1404 | protocol == htons(ETH_P_IPV6))); | |
1405 | } | |
1406 | ||
1407 | static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb) | |
1408 | { | |
1409 | if (can_checksum_protocol(dev->features, skb->protocol)) | |
1410 | return true; | |
1411 | ||
1412 | if (skb->protocol == htons(ETH_P_8021Q)) { | |
1413 | struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data; | |
1414 | if (can_checksum_protocol(dev->features & dev->vlan_features, | |
1415 | veh->h_vlan_encapsulated_proto)) | |
1416 | return true; | |
1417 | } | |
1418 | ||
1419 | return false; | |
1420 | } | |
1421 | ||
1422 | /* | |
1423 | * Invalidate hardware checksum when packet is to be mangled, and | |
1424 | * complete checksum manually on outgoing path. | |
1425 | */ | |
1426 | int skb_checksum_help(struct sk_buff *skb) | |
1427 | { | |
1428 | __wsum csum; | |
1429 | int ret = 0, offset; | |
1430 | ||
1431 | if (skb->ip_summed == CHECKSUM_COMPLETE) | |
1432 | goto out_set_summed; | |
1433 | ||
1434 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
1435 | /* Let GSO fix up the checksum. */ | |
1436 | goto out_set_summed; | |
1437 | } | |
1438 | ||
1439 | offset = skb->csum_start - skb_headroom(skb); | |
1440 | BUG_ON(offset >= skb_headlen(skb)); | |
1441 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1442 | ||
1443 | offset += skb->csum_offset; | |
1444 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
1445 | ||
1446 | if (skb_cloned(skb) && | |
1447 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1448 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
1449 | if (ret) | |
1450 | goto out; | |
1451 | } | |
1452 | ||
1453 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); | |
1454 | out_set_summed: | |
1455 | skb->ip_summed = CHECKSUM_NONE; | |
1456 | out: | |
1457 | return ret; | |
1458 | } | |
1459 | ||
1460 | /** | |
1461 | * skb_gso_segment - Perform segmentation on skb. | |
1462 | * @skb: buffer to segment | |
1463 | * @features: features for the output path (see dev->features) | |
1464 | * | |
1465 | * This function segments the given skb and returns a list of segments. | |
1466 | * | |
1467 | * It may return NULL if the skb requires no segmentation. This is | |
1468 | * only possible when GSO is used for verifying header integrity. | |
1469 | */ | |
1470 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features) | |
1471 | { | |
1472 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
1473 | struct packet_type *ptype; | |
1474 | __be16 type = skb->protocol; | |
1475 | int err; | |
1476 | ||
1477 | BUG_ON(skb_shinfo(skb)->frag_list); | |
1478 | ||
1479 | skb_reset_mac_header(skb); | |
1480 | skb->mac_len = skb->network_header - skb->mac_header; | |
1481 | __skb_pull(skb, skb->mac_len); | |
1482 | ||
1483 | if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) { | |
1484 | if (skb_header_cloned(skb) && | |
1485 | (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) | |
1486 | return ERR_PTR(err); | |
1487 | } | |
1488 | ||
1489 | rcu_read_lock(); | |
1490 | list_for_each_entry_rcu(ptype, | |
1491 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
1492 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { | |
1493 | if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { | |
1494 | err = ptype->gso_send_check(skb); | |
1495 | segs = ERR_PTR(err); | |
1496 | if (err || skb_gso_ok(skb, features)) | |
1497 | break; | |
1498 | __skb_push(skb, (skb->data - | |
1499 | skb_network_header(skb))); | |
1500 | } | |
1501 | segs = ptype->gso_segment(skb, features); | |
1502 | break; | |
1503 | } | |
1504 | } | |
1505 | rcu_read_unlock(); | |
1506 | ||
1507 | __skb_push(skb, skb->data - skb_mac_header(skb)); | |
1508 | ||
1509 | return segs; | |
1510 | } | |
1511 | ||
1512 | EXPORT_SYMBOL(skb_gso_segment); | |
1513 | ||
1514 | /* Take action when hardware reception checksum errors are detected. */ | |
1515 | #ifdef CONFIG_BUG | |
1516 | void netdev_rx_csum_fault(struct net_device *dev) | |
1517 | { | |
1518 | if (net_ratelimit()) { | |
1519 | printk(KERN_ERR "%s: hw csum failure.\n", | |
1520 | dev ? dev->name : "<unknown>"); | |
1521 | dump_stack(); | |
1522 | } | |
1523 | } | |
1524 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
1525 | #endif | |
1526 | ||
1527 | /* Actually, we should eliminate this check as soon as we know, that: | |
1528 | * 1. IOMMU is present and allows to map all the memory. | |
1529 | * 2. No high memory really exists on this machine. | |
1530 | */ | |
1531 | ||
1532 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) | |
1533 | { | |
1534 | #ifdef CONFIG_HIGHMEM | |
1535 | int i; | |
1536 | ||
1537 | if (dev->features & NETIF_F_HIGHDMA) | |
1538 | return 0; | |
1539 | ||
1540 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1541 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) | |
1542 | return 1; | |
1543 | ||
1544 | #endif | |
1545 | return 0; | |
1546 | } | |
1547 | ||
1548 | struct dev_gso_cb { | |
1549 | void (*destructor)(struct sk_buff *skb); | |
1550 | }; | |
1551 | ||
1552 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) | |
1553 | ||
1554 | static void dev_gso_skb_destructor(struct sk_buff *skb) | |
1555 | { | |
1556 | struct dev_gso_cb *cb; | |
1557 | ||
1558 | do { | |
1559 | struct sk_buff *nskb = skb->next; | |
1560 | ||
1561 | skb->next = nskb->next; | |
1562 | nskb->next = NULL; | |
1563 | kfree_skb(nskb); | |
1564 | } while (skb->next); | |
1565 | ||
1566 | cb = DEV_GSO_CB(skb); | |
1567 | if (cb->destructor) | |
1568 | cb->destructor(skb); | |
1569 | } | |
1570 | ||
1571 | /** | |
1572 | * dev_gso_segment - Perform emulated hardware segmentation on skb. | |
1573 | * @skb: buffer to segment | |
1574 | * | |
1575 | * This function segments the given skb and stores the list of segments | |
1576 | * in skb->next. | |
1577 | */ | |
1578 | static int dev_gso_segment(struct sk_buff *skb) | |
1579 | { | |
1580 | struct net_device *dev = skb->dev; | |
1581 | struct sk_buff *segs; | |
1582 | int features = dev->features & ~(illegal_highdma(dev, skb) ? | |
1583 | NETIF_F_SG : 0); | |
1584 | ||
1585 | segs = skb_gso_segment(skb, features); | |
1586 | ||
1587 | /* Verifying header integrity only. */ | |
1588 | if (!segs) | |
1589 | return 0; | |
1590 | ||
1591 | if (IS_ERR(segs)) | |
1592 | return PTR_ERR(segs); | |
1593 | ||
1594 | skb->next = segs; | |
1595 | DEV_GSO_CB(skb)->destructor = skb->destructor; | |
1596 | skb->destructor = dev_gso_skb_destructor; | |
1597 | ||
1598 | return 0; | |
1599 | } | |
1600 | ||
1601 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev, | |
1602 | struct netdev_queue *txq) | |
1603 | { | |
1604 | if (likely(!skb->next)) { | |
1605 | if (!list_empty(&ptype_all)) | |
1606 | dev_queue_xmit_nit(skb, dev); | |
1607 | ||
1608 | if (netif_needs_gso(dev, skb)) { | |
1609 | if (unlikely(dev_gso_segment(skb))) | |
1610 | goto out_kfree_skb; | |
1611 | if (skb->next) | |
1612 | goto gso; | |
1613 | } | |
1614 | ||
1615 | return dev->hard_start_xmit(skb, dev); | |
1616 | } | |
1617 | ||
1618 | gso: | |
1619 | do { | |
1620 | struct sk_buff *nskb = skb->next; | |
1621 | int rc; | |
1622 | ||
1623 | skb->next = nskb->next; | |
1624 | nskb->next = NULL; | |
1625 | rc = dev->hard_start_xmit(nskb, dev); | |
1626 | if (unlikely(rc)) { | |
1627 | nskb->next = skb->next; | |
1628 | skb->next = nskb; | |
1629 | return rc; | |
1630 | } | |
1631 | if (unlikely(netif_tx_queue_stopped(txq) && skb->next)) | |
1632 | return NETDEV_TX_BUSY; | |
1633 | } while (skb->next); | |
1634 | ||
1635 | skb->destructor = DEV_GSO_CB(skb)->destructor; | |
1636 | ||
1637 | out_kfree_skb: | |
1638 | kfree_skb(skb); | |
1639 | return 0; | |
1640 | } | |
1641 | ||
1642 | /** | |
1643 | * dev_queue_xmit - transmit a buffer | |
1644 | * @skb: buffer to transmit | |
1645 | * | |
1646 | * Queue a buffer for transmission to a network device. The caller must | |
1647 | * have set the device and priority and built the buffer before calling | |
1648 | * this function. The function can be called from an interrupt. | |
1649 | * | |
1650 | * A negative errno code is returned on a failure. A success does not | |
1651 | * guarantee the frame will be transmitted as it may be dropped due | |
1652 | * to congestion or traffic shaping. | |
1653 | * | |
1654 | * ----------------------------------------------------------------------------------- | |
1655 | * I notice this method can also return errors from the queue disciplines, | |
1656 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
1657 | * be positive. | |
1658 | * | |
1659 | * Regardless of the return value, the skb is consumed, so it is currently | |
1660 | * difficult to retry a send to this method. (You can bump the ref count | |
1661 | * before sending to hold a reference for retry if you are careful.) | |
1662 | * | |
1663 | * When calling this method, interrupts MUST be enabled. This is because | |
1664 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
1665 | * --BLG | |
1666 | */ | |
1667 | ||
1668 | static struct netdev_queue *dev_pick_tx(struct net_device *dev, | |
1669 | struct sk_buff *skb) | |
1670 | { | |
1671 | u16 queue_index = 0; | |
1672 | ||
1673 | skb_set_queue_mapping(skb, queue_index); | |
1674 | return netdev_get_tx_queue(dev, queue_index); | |
1675 | } | |
1676 | ||
1677 | int dev_queue_xmit(struct sk_buff *skb) | |
1678 | { | |
1679 | struct net_device *dev = skb->dev; | |
1680 | struct netdev_queue *txq; | |
1681 | struct Qdisc *q; | |
1682 | int rc = -ENOMEM; | |
1683 | ||
1684 | /* GSO will handle the following emulations directly. */ | |
1685 | if (netif_needs_gso(dev, skb)) | |
1686 | goto gso; | |
1687 | ||
1688 | if (skb_shinfo(skb)->frag_list && | |
1689 | !(dev->features & NETIF_F_FRAGLIST) && | |
1690 | __skb_linearize(skb)) | |
1691 | goto out_kfree_skb; | |
1692 | ||
1693 | /* Fragmented skb is linearized if device does not support SG, | |
1694 | * or if at least one of fragments is in highmem and device | |
1695 | * does not support DMA from it. | |
1696 | */ | |
1697 | if (skb_shinfo(skb)->nr_frags && | |
1698 | (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) && | |
1699 | __skb_linearize(skb)) | |
1700 | goto out_kfree_skb; | |
1701 | ||
1702 | /* If packet is not checksummed and device does not support | |
1703 | * checksumming for this protocol, complete checksumming here. | |
1704 | */ | |
1705 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
1706 | skb_set_transport_header(skb, skb->csum_start - | |
1707 | skb_headroom(skb)); | |
1708 | if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb)) | |
1709 | goto out_kfree_skb; | |
1710 | } | |
1711 | ||
1712 | gso: | |
1713 | txq = dev_pick_tx(dev, skb); | |
1714 | spin_lock_prefetch(&txq->lock); | |
1715 | ||
1716 | /* Disable soft irqs for various locks below. Also | |
1717 | * stops preemption for RCU. | |
1718 | */ | |
1719 | rcu_read_lock_bh(); | |
1720 | ||
1721 | /* Updates of qdisc are serialized by queue->lock. | |
1722 | * The struct Qdisc which is pointed to by qdisc is now a | |
1723 | * rcu structure - it may be accessed without acquiring | |
1724 | * a lock (but the structure may be stale.) The freeing of the | |
1725 | * qdisc will be deferred until it's known that there are no | |
1726 | * more references to it. | |
1727 | * | |
1728 | * If the qdisc has an enqueue function, we still need to | |
1729 | * hold the queue->lock before calling it, since queue->lock | |
1730 | * also serializes access to the device queue. | |
1731 | */ | |
1732 | ||
1733 | q = rcu_dereference(txq->qdisc); | |
1734 | #ifdef CONFIG_NET_CLS_ACT | |
1735 | skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS); | |
1736 | #endif | |
1737 | if (q->enqueue) { | |
1738 | /* Grab device queue */ | |
1739 | spin_lock(&txq->lock); | |
1740 | q = txq->qdisc; | |
1741 | if (q->enqueue) { | |
1742 | rc = q->enqueue(skb, q); | |
1743 | qdisc_run(txq); | |
1744 | spin_unlock(&txq->lock); | |
1745 | ||
1746 | rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc; | |
1747 | goto out; | |
1748 | } | |
1749 | spin_unlock(&txq->lock); | |
1750 | } | |
1751 | ||
1752 | /* The device has no queue. Common case for software devices: | |
1753 | loopback, all the sorts of tunnels... | |
1754 | ||
1755 | Really, it is unlikely that netif_tx_lock protection is necessary | |
1756 | here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
1757 | counters.) | |
1758 | However, it is possible, that they rely on protection | |
1759 | made by us here. | |
1760 | ||
1761 | Check this and shot the lock. It is not prone from deadlocks. | |
1762 | Either shot noqueue qdisc, it is even simpler 8) | |
1763 | */ | |
1764 | if (dev->flags & IFF_UP) { | |
1765 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
1766 | ||
1767 | if (txq->xmit_lock_owner != cpu) { | |
1768 | ||
1769 | HARD_TX_LOCK(dev, txq, cpu); | |
1770 | ||
1771 | if (!netif_tx_queue_stopped(txq)) { | |
1772 | rc = 0; | |
1773 | if (!dev_hard_start_xmit(skb, dev, txq)) { | |
1774 | HARD_TX_UNLOCK(dev, txq); | |
1775 | goto out; | |
1776 | } | |
1777 | } | |
1778 | HARD_TX_UNLOCK(dev, txq); | |
1779 | if (net_ratelimit()) | |
1780 | printk(KERN_CRIT "Virtual device %s asks to " | |
1781 | "queue packet!\n", dev->name); | |
1782 | } else { | |
1783 | /* Recursion is detected! It is possible, | |
1784 | * unfortunately */ | |
1785 | if (net_ratelimit()) | |
1786 | printk(KERN_CRIT "Dead loop on virtual device " | |
1787 | "%s, fix it urgently!\n", dev->name); | |
1788 | } | |
1789 | } | |
1790 | ||
1791 | rc = -ENETDOWN; | |
1792 | rcu_read_unlock_bh(); | |
1793 | ||
1794 | out_kfree_skb: | |
1795 | kfree_skb(skb); | |
1796 | return rc; | |
1797 | out: | |
1798 | rcu_read_unlock_bh(); | |
1799 | return rc; | |
1800 | } | |
1801 | ||
1802 | ||
1803 | /*======================================================================= | |
1804 | Receiver routines | |
1805 | =======================================================================*/ | |
1806 | ||
1807 | int netdev_max_backlog __read_mostly = 1000; | |
1808 | int netdev_budget __read_mostly = 300; | |
1809 | int weight_p __read_mostly = 64; /* old backlog weight */ | |
1810 | ||
1811 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; | |
1812 | ||
1813 | ||
1814 | /** | |
1815 | * netif_rx - post buffer to the network code | |
1816 | * @skb: buffer to post | |
1817 | * | |
1818 | * This function receives a packet from a device driver and queues it for | |
1819 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
1820 | * may be dropped during processing for congestion control or by the | |
1821 | * protocol layers. | |
1822 | * | |
1823 | * return values: | |
1824 | * NET_RX_SUCCESS (no congestion) | |
1825 | * NET_RX_DROP (packet was dropped) | |
1826 | * | |
1827 | */ | |
1828 | ||
1829 | int netif_rx(struct sk_buff *skb) | |
1830 | { | |
1831 | struct softnet_data *queue; | |
1832 | unsigned long flags; | |
1833 | ||
1834 | /* if netpoll wants it, pretend we never saw it */ | |
1835 | if (netpoll_rx(skb)) | |
1836 | return NET_RX_DROP; | |
1837 | ||
1838 | if (!skb->tstamp.tv64) | |
1839 | net_timestamp(skb); | |
1840 | ||
1841 | /* | |
1842 | * The code is rearranged so that the path is the most | |
1843 | * short when CPU is congested, but is still operating. | |
1844 | */ | |
1845 | local_irq_save(flags); | |
1846 | queue = &__get_cpu_var(softnet_data); | |
1847 | ||
1848 | __get_cpu_var(netdev_rx_stat).total++; | |
1849 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { | |
1850 | if (queue->input_pkt_queue.qlen) { | |
1851 | enqueue: | |
1852 | dev_hold(skb->dev); | |
1853 | __skb_queue_tail(&queue->input_pkt_queue, skb); | |
1854 | local_irq_restore(flags); | |
1855 | return NET_RX_SUCCESS; | |
1856 | } | |
1857 | ||
1858 | napi_schedule(&queue->backlog); | |
1859 | goto enqueue; | |
1860 | } | |
1861 | ||
1862 | __get_cpu_var(netdev_rx_stat).dropped++; | |
1863 | local_irq_restore(flags); | |
1864 | ||
1865 | kfree_skb(skb); | |
1866 | return NET_RX_DROP; | |
1867 | } | |
1868 | ||
1869 | int netif_rx_ni(struct sk_buff *skb) | |
1870 | { | |
1871 | int err; | |
1872 | ||
1873 | preempt_disable(); | |
1874 | err = netif_rx(skb); | |
1875 | if (local_softirq_pending()) | |
1876 | do_softirq(); | |
1877 | preempt_enable(); | |
1878 | ||
1879 | return err; | |
1880 | } | |
1881 | ||
1882 | EXPORT_SYMBOL(netif_rx_ni); | |
1883 | ||
1884 | static inline struct net_device *skb_bond(struct sk_buff *skb) | |
1885 | { | |
1886 | struct net_device *dev = skb->dev; | |
1887 | ||
1888 | if (dev->master) { | |
1889 | if (skb_bond_should_drop(skb)) { | |
1890 | kfree_skb(skb); | |
1891 | return NULL; | |
1892 | } | |
1893 | skb->dev = dev->master; | |
1894 | } | |
1895 | ||
1896 | return dev; | |
1897 | } | |
1898 | ||
1899 | ||
1900 | static void net_tx_action(struct softirq_action *h) | |
1901 | { | |
1902 | struct softnet_data *sd = &__get_cpu_var(softnet_data); | |
1903 | ||
1904 | if (sd->completion_queue) { | |
1905 | struct sk_buff *clist; | |
1906 | ||
1907 | local_irq_disable(); | |
1908 | clist = sd->completion_queue; | |
1909 | sd->completion_queue = NULL; | |
1910 | local_irq_enable(); | |
1911 | ||
1912 | while (clist) { | |
1913 | struct sk_buff *skb = clist; | |
1914 | clist = clist->next; | |
1915 | ||
1916 | BUG_TRAP(!atomic_read(&skb->users)); | |
1917 | __kfree_skb(skb); | |
1918 | } | |
1919 | } | |
1920 | ||
1921 | if (sd->output_queue) { | |
1922 | struct netdev_queue *head; | |
1923 | ||
1924 | local_irq_disable(); | |
1925 | head = sd->output_queue; | |
1926 | sd->output_queue = NULL; | |
1927 | local_irq_enable(); | |
1928 | ||
1929 | while (head) { | |
1930 | struct netdev_queue *txq = head; | |
1931 | struct net_device *dev = txq->dev; | |
1932 | head = head->next_sched; | |
1933 | ||
1934 | smp_mb__before_clear_bit(); | |
1935 | clear_bit(__LINK_STATE_SCHED, &dev->state); | |
1936 | ||
1937 | if (spin_trylock(&txq->lock)) { | |
1938 | qdisc_run(txq); | |
1939 | spin_unlock(&txq->lock); | |
1940 | } else { | |
1941 | netif_schedule_queue(txq); | |
1942 | } | |
1943 | } | |
1944 | } | |
1945 | } | |
1946 | ||
1947 | static inline int deliver_skb(struct sk_buff *skb, | |
1948 | struct packet_type *pt_prev, | |
1949 | struct net_device *orig_dev) | |
1950 | { | |
1951 | atomic_inc(&skb->users); | |
1952 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1953 | } | |
1954 | ||
1955 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) | |
1956 | /* These hooks defined here for ATM */ | |
1957 | struct net_bridge; | |
1958 | struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br, | |
1959 | unsigned char *addr); | |
1960 | void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly; | |
1961 | ||
1962 | /* | |
1963 | * If bridge module is loaded call bridging hook. | |
1964 | * returns NULL if packet was consumed. | |
1965 | */ | |
1966 | struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p, | |
1967 | struct sk_buff *skb) __read_mostly; | |
1968 | static inline struct sk_buff *handle_bridge(struct sk_buff *skb, | |
1969 | struct packet_type **pt_prev, int *ret, | |
1970 | struct net_device *orig_dev) | |
1971 | { | |
1972 | struct net_bridge_port *port; | |
1973 | ||
1974 | if (skb->pkt_type == PACKET_LOOPBACK || | |
1975 | (port = rcu_dereference(skb->dev->br_port)) == NULL) | |
1976 | return skb; | |
1977 | ||
1978 | if (*pt_prev) { | |
1979 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
1980 | *pt_prev = NULL; | |
1981 | } | |
1982 | ||
1983 | return br_handle_frame_hook(port, skb); | |
1984 | } | |
1985 | #else | |
1986 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb) | |
1987 | #endif | |
1988 | ||
1989 | #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE) | |
1990 | struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly; | |
1991 | EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook); | |
1992 | ||
1993 | static inline struct sk_buff *handle_macvlan(struct sk_buff *skb, | |
1994 | struct packet_type **pt_prev, | |
1995 | int *ret, | |
1996 | struct net_device *orig_dev) | |
1997 | { | |
1998 | if (skb->dev->macvlan_port == NULL) | |
1999 | return skb; | |
2000 | ||
2001 | if (*pt_prev) { | |
2002 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2003 | *pt_prev = NULL; | |
2004 | } | |
2005 | return macvlan_handle_frame_hook(skb); | |
2006 | } | |
2007 | #else | |
2008 | #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb) | |
2009 | #endif | |
2010 | ||
2011 | #ifdef CONFIG_NET_CLS_ACT | |
2012 | /* TODO: Maybe we should just force sch_ingress to be compiled in | |
2013 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions | |
2014 | * a compare and 2 stores extra right now if we dont have it on | |
2015 | * but have CONFIG_NET_CLS_ACT | |
2016 | * NOTE: This doesnt stop any functionality; if you dont have | |
2017 | * the ingress scheduler, you just cant add policies on ingress. | |
2018 | * | |
2019 | */ | |
2020 | static int ing_filter(struct sk_buff *skb) | |
2021 | { | |
2022 | struct net_device *dev = skb->dev; | |
2023 | u32 ttl = G_TC_RTTL(skb->tc_verd); | |
2024 | struct netdev_queue *rxq; | |
2025 | int result = TC_ACT_OK; | |
2026 | struct Qdisc *q; | |
2027 | ||
2028 | if (MAX_RED_LOOP < ttl++) { | |
2029 | printk(KERN_WARNING | |
2030 | "Redir loop detected Dropping packet (%d->%d)\n", | |
2031 | skb->iif, dev->ifindex); | |
2032 | return TC_ACT_SHOT; | |
2033 | } | |
2034 | ||
2035 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl); | |
2036 | skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS); | |
2037 | ||
2038 | rxq = &dev->rx_queue; | |
2039 | ||
2040 | spin_lock(&rxq->lock); | |
2041 | if ((q = rxq->qdisc) != NULL) | |
2042 | result = q->enqueue(skb, q); | |
2043 | spin_unlock(&rxq->lock); | |
2044 | ||
2045 | return result; | |
2046 | } | |
2047 | ||
2048 | static inline struct sk_buff *handle_ing(struct sk_buff *skb, | |
2049 | struct packet_type **pt_prev, | |
2050 | int *ret, struct net_device *orig_dev) | |
2051 | { | |
2052 | if (!skb->dev->rx_queue.qdisc) | |
2053 | goto out; | |
2054 | ||
2055 | if (*pt_prev) { | |
2056 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
2057 | *pt_prev = NULL; | |
2058 | } else { | |
2059 | /* Huh? Why does turning on AF_PACKET affect this? */ | |
2060 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); | |
2061 | } | |
2062 | ||
2063 | switch (ing_filter(skb)) { | |
2064 | case TC_ACT_SHOT: | |
2065 | case TC_ACT_STOLEN: | |
2066 | kfree_skb(skb); | |
2067 | return NULL; | |
2068 | } | |
2069 | ||
2070 | out: | |
2071 | skb->tc_verd = 0; | |
2072 | return skb; | |
2073 | } | |
2074 | #endif | |
2075 | ||
2076 | /* | |
2077 | * netif_nit_deliver - deliver received packets to network taps | |
2078 | * @skb: buffer | |
2079 | * | |
2080 | * This function is used to deliver incoming packets to network | |
2081 | * taps. It should be used when the normal netif_receive_skb path | |
2082 | * is bypassed, for example because of VLAN acceleration. | |
2083 | */ | |
2084 | void netif_nit_deliver(struct sk_buff *skb) | |
2085 | { | |
2086 | struct packet_type *ptype; | |
2087 | ||
2088 | if (list_empty(&ptype_all)) | |
2089 | return; | |
2090 | ||
2091 | skb_reset_network_header(skb); | |
2092 | skb_reset_transport_header(skb); | |
2093 | skb->mac_len = skb->network_header - skb->mac_header; | |
2094 | ||
2095 | rcu_read_lock(); | |
2096 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2097 | if (!ptype->dev || ptype->dev == skb->dev) | |
2098 | deliver_skb(skb, ptype, skb->dev); | |
2099 | } | |
2100 | rcu_read_unlock(); | |
2101 | } | |
2102 | ||
2103 | /** | |
2104 | * netif_receive_skb - process receive buffer from network | |
2105 | * @skb: buffer to process | |
2106 | * | |
2107 | * netif_receive_skb() is the main receive data processing function. | |
2108 | * It always succeeds. The buffer may be dropped during processing | |
2109 | * for congestion control or by the protocol layers. | |
2110 | * | |
2111 | * This function may only be called from softirq context and interrupts | |
2112 | * should be enabled. | |
2113 | * | |
2114 | * Return values (usually ignored): | |
2115 | * NET_RX_SUCCESS: no congestion | |
2116 | * NET_RX_DROP: packet was dropped | |
2117 | */ | |
2118 | int netif_receive_skb(struct sk_buff *skb) | |
2119 | { | |
2120 | struct packet_type *ptype, *pt_prev; | |
2121 | struct net_device *orig_dev; | |
2122 | int ret = NET_RX_DROP; | |
2123 | __be16 type; | |
2124 | ||
2125 | /* if we've gotten here through NAPI, check netpoll */ | |
2126 | if (netpoll_receive_skb(skb)) | |
2127 | return NET_RX_DROP; | |
2128 | ||
2129 | if (!skb->tstamp.tv64) | |
2130 | net_timestamp(skb); | |
2131 | ||
2132 | if (!skb->iif) | |
2133 | skb->iif = skb->dev->ifindex; | |
2134 | ||
2135 | orig_dev = skb_bond(skb); | |
2136 | ||
2137 | if (!orig_dev) | |
2138 | return NET_RX_DROP; | |
2139 | ||
2140 | __get_cpu_var(netdev_rx_stat).total++; | |
2141 | ||
2142 | skb_reset_network_header(skb); | |
2143 | skb_reset_transport_header(skb); | |
2144 | skb->mac_len = skb->network_header - skb->mac_header; | |
2145 | ||
2146 | pt_prev = NULL; | |
2147 | ||
2148 | rcu_read_lock(); | |
2149 | ||
2150 | /* Don't receive packets in an exiting network namespace */ | |
2151 | if (!net_alive(dev_net(skb->dev))) | |
2152 | goto out; | |
2153 | ||
2154 | #ifdef CONFIG_NET_CLS_ACT | |
2155 | if (skb->tc_verd & TC_NCLS) { | |
2156 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); | |
2157 | goto ncls; | |
2158 | } | |
2159 | #endif | |
2160 | ||
2161 | list_for_each_entry_rcu(ptype, &ptype_all, list) { | |
2162 | if (!ptype->dev || ptype->dev == skb->dev) { | |
2163 | if (pt_prev) | |
2164 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
2165 | pt_prev = ptype; | |
2166 | } | |
2167 | } | |
2168 | ||
2169 | #ifdef CONFIG_NET_CLS_ACT | |
2170 | skb = handle_ing(skb, &pt_prev, &ret, orig_dev); | |
2171 | if (!skb) | |
2172 | goto out; | |
2173 | ncls: | |
2174 | #endif | |
2175 | ||
2176 | skb = handle_bridge(skb, &pt_prev, &ret, orig_dev); | |
2177 | if (!skb) | |
2178 | goto out; | |
2179 | skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev); | |
2180 | if (!skb) | |
2181 | goto out; | |
2182 | ||
2183 | type = skb->protocol; | |
2184 | list_for_each_entry_rcu(ptype, | |
2185 | &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) { | |
2186 | if (ptype->type == type && | |
2187 | (!ptype->dev || ptype->dev == skb->dev)) { | |
2188 | if (pt_prev) | |
2189 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
2190 | pt_prev = ptype; | |
2191 | } | |
2192 | } | |
2193 | ||
2194 | if (pt_prev) { | |
2195 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
2196 | } else { | |
2197 | kfree_skb(skb); | |
2198 | /* Jamal, now you will not able to escape explaining | |
2199 | * me how you were going to use this. :-) | |
2200 | */ | |
2201 | ret = NET_RX_DROP; | |
2202 | } | |
2203 | ||
2204 | out: | |
2205 | rcu_read_unlock(); | |
2206 | return ret; | |
2207 | } | |
2208 | ||
2209 | static int process_backlog(struct napi_struct *napi, int quota) | |
2210 | { | |
2211 | int work = 0; | |
2212 | struct softnet_data *queue = &__get_cpu_var(softnet_data); | |
2213 | unsigned long start_time = jiffies; | |
2214 | ||
2215 | napi->weight = weight_p; | |
2216 | do { | |
2217 | struct sk_buff *skb; | |
2218 | struct net_device *dev; | |
2219 | ||
2220 | local_irq_disable(); | |
2221 | skb = __skb_dequeue(&queue->input_pkt_queue); | |
2222 | if (!skb) { | |
2223 | __napi_complete(napi); | |
2224 | local_irq_enable(); | |
2225 | break; | |
2226 | } | |
2227 | ||
2228 | local_irq_enable(); | |
2229 | ||
2230 | dev = skb->dev; | |
2231 | ||
2232 | netif_receive_skb(skb); | |
2233 | ||
2234 | dev_put(dev); | |
2235 | } while (++work < quota && jiffies == start_time); | |
2236 | ||
2237 | return work; | |
2238 | } | |
2239 | ||
2240 | /** | |
2241 | * __napi_schedule - schedule for receive | |
2242 | * @n: entry to schedule | |
2243 | * | |
2244 | * The entry's receive function will be scheduled to run | |
2245 | */ | |
2246 | void __napi_schedule(struct napi_struct *n) | |
2247 | { | |
2248 | unsigned long flags; | |
2249 | ||
2250 | local_irq_save(flags); | |
2251 | list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); | |
2252 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2253 | local_irq_restore(flags); | |
2254 | } | |
2255 | EXPORT_SYMBOL(__napi_schedule); | |
2256 | ||
2257 | ||
2258 | static void net_rx_action(struct softirq_action *h) | |
2259 | { | |
2260 | struct list_head *list = &__get_cpu_var(softnet_data).poll_list; | |
2261 | unsigned long start_time = jiffies; | |
2262 | int budget = netdev_budget; | |
2263 | void *have; | |
2264 | ||
2265 | local_irq_disable(); | |
2266 | ||
2267 | while (!list_empty(list)) { | |
2268 | struct napi_struct *n; | |
2269 | int work, weight; | |
2270 | ||
2271 | /* If softirq window is exhuasted then punt. | |
2272 | * | |
2273 | * Note that this is a slight policy change from the | |
2274 | * previous NAPI code, which would allow up to 2 | |
2275 | * jiffies to pass before breaking out. The test | |
2276 | * used to be "jiffies - start_time > 1". | |
2277 | */ | |
2278 | if (unlikely(budget <= 0 || jiffies != start_time)) | |
2279 | goto softnet_break; | |
2280 | ||
2281 | local_irq_enable(); | |
2282 | ||
2283 | /* Even though interrupts have been re-enabled, this | |
2284 | * access is safe because interrupts can only add new | |
2285 | * entries to the tail of this list, and only ->poll() | |
2286 | * calls can remove this head entry from the list. | |
2287 | */ | |
2288 | n = list_entry(list->next, struct napi_struct, poll_list); | |
2289 | ||
2290 | have = netpoll_poll_lock(n); | |
2291 | ||
2292 | weight = n->weight; | |
2293 | ||
2294 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
2295 | * with netpoll's poll_napi(). Only the entity which | |
2296 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
2297 | * actually make the ->poll() call. Therefore we avoid | |
2298 | * accidently calling ->poll() when NAPI is not scheduled. | |
2299 | */ | |
2300 | work = 0; | |
2301 | if (test_bit(NAPI_STATE_SCHED, &n->state)) | |
2302 | work = n->poll(n, weight); | |
2303 | ||
2304 | WARN_ON_ONCE(work > weight); | |
2305 | ||
2306 | budget -= work; | |
2307 | ||
2308 | local_irq_disable(); | |
2309 | ||
2310 | /* Drivers must not modify the NAPI state if they | |
2311 | * consume the entire weight. In such cases this code | |
2312 | * still "owns" the NAPI instance and therefore can | |
2313 | * move the instance around on the list at-will. | |
2314 | */ | |
2315 | if (unlikely(work == weight)) { | |
2316 | if (unlikely(napi_disable_pending(n))) | |
2317 | __napi_complete(n); | |
2318 | else | |
2319 | list_move_tail(&n->poll_list, list); | |
2320 | } | |
2321 | ||
2322 | netpoll_poll_unlock(have); | |
2323 | } | |
2324 | out: | |
2325 | local_irq_enable(); | |
2326 | ||
2327 | #ifdef CONFIG_NET_DMA | |
2328 | /* | |
2329 | * There may not be any more sk_buffs coming right now, so push | |
2330 | * any pending DMA copies to hardware | |
2331 | */ | |
2332 | if (!cpus_empty(net_dma.channel_mask)) { | |
2333 | int chan_idx; | |
2334 | for_each_cpu_mask(chan_idx, net_dma.channel_mask) { | |
2335 | struct dma_chan *chan = net_dma.channels[chan_idx]; | |
2336 | if (chan) | |
2337 | dma_async_memcpy_issue_pending(chan); | |
2338 | } | |
2339 | } | |
2340 | #endif | |
2341 | ||
2342 | return; | |
2343 | ||
2344 | softnet_break: | |
2345 | __get_cpu_var(netdev_rx_stat).time_squeeze++; | |
2346 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
2347 | goto out; | |
2348 | } | |
2349 | ||
2350 | static gifconf_func_t * gifconf_list [NPROTO]; | |
2351 | ||
2352 | /** | |
2353 | * register_gifconf - register a SIOCGIF handler | |
2354 | * @family: Address family | |
2355 | * @gifconf: Function handler | |
2356 | * | |
2357 | * Register protocol dependent address dumping routines. The handler | |
2358 | * that is passed must not be freed or reused until it has been replaced | |
2359 | * by another handler. | |
2360 | */ | |
2361 | int register_gifconf(unsigned int family, gifconf_func_t * gifconf) | |
2362 | { | |
2363 | if (family >= NPROTO) | |
2364 | return -EINVAL; | |
2365 | gifconf_list[family] = gifconf; | |
2366 | return 0; | |
2367 | } | |
2368 | ||
2369 | ||
2370 | /* | |
2371 | * Map an interface index to its name (SIOCGIFNAME) | |
2372 | */ | |
2373 | ||
2374 | /* | |
2375 | * We need this ioctl for efficient implementation of the | |
2376 | * if_indextoname() function required by the IPv6 API. Without | |
2377 | * it, we would have to search all the interfaces to find a | |
2378 | * match. --pb | |
2379 | */ | |
2380 | ||
2381 | static int dev_ifname(struct net *net, struct ifreq __user *arg) | |
2382 | { | |
2383 | struct net_device *dev; | |
2384 | struct ifreq ifr; | |
2385 | ||
2386 | /* | |
2387 | * Fetch the caller's info block. | |
2388 | */ | |
2389 | ||
2390 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
2391 | return -EFAULT; | |
2392 | ||
2393 | read_lock(&dev_base_lock); | |
2394 | dev = __dev_get_by_index(net, ifr.ifr_ifindex); | |
2395 | if (!dev) { | |
2396 | read_unlock(&dev_base_lock); | |
2397 | return -ENODEV; | |
2398 | } | |
2399 | ||
2400 | strcpy(ifr.ifr_name, dev->name); | |
2401 | read_unlock(&dev_base_lock); | |
2402 | ||
2403 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) | |
2404 | return -EFAULT; | |
2405 | return 0; | |
2406 | } | |
2407 | ||
2408 | /* | |
2409 | * Perform a SIOCGIFCONF call. This structure will change | |
2410 | * size eventually, and there is nothing I can do about it. | |
2411 | * Thus we will need a 'compatibility mode'. | |
2412 | */ | |
2413 | ||
2414 | static int dev_ifconf(struct net *net, char __user *arg) | |
2415 | { | |
2416 | struct ifconf ifc; | |
2417 | struct net_device *dev; | |
2418 | char __user *pos; | |
2419 | int len; | |
2420 | int total; | |
2421 | int i; | |
2422 | ||
2423 | /* | |
2424 | * Fetch the caller's info block. | |
2425 | */ | |
2426 | ||
2427 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) | |
2428 | return -EFAULT; | |
2429 | ||
2430 | pos = ifc.ifc_buf; | |
2431 | len = ifc.ifc_len; | |
2432 | ||
2433 | /* | |
2434 | * Loop over the interfaces, and write an info block for each. | |
2435 | */ | |
2436 | ||
2437 | total = 0; | |
2438 | for_each_netdev(net, dev) { | |
2439 | for (i = 0; i < NPROTO; i++) { | |
2440 | if (gifconf_list[i]) { | |
2441 | int done; | |
2442 | if (!pos) | |
2443 | done = gifconf_list[i](dev, NULL, 0); | |
2444 | else | |
2445 | done = gifconf_list[i](dev, pos + total, | |
2446 | len - total); | |
2447 | if (done < 0) | |
2448 | return -EFAULT; | |
2449 | total += done; | |
2450 | } | |
2451 | } | |
2452 | } | |
2453 | ||
2454 | /* | |
2455 | * All done. Write the updated control block back to the caller. | |
2456 | */ | |
2457 | ifc.ifc_len = total; | |
2458 | ||
2459 | /* | |
2460 | * Both BSD and Solaris return 0 here, so we do too. | |
2461 | */ | |
2462 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; | |
2463 | } | |
2464 | ||
2465 | #ifdef CONFIG_PROC_FS | |
2466 | /* | |
2467 | * This is invoked by the /proc filesystem handler to display a device | |
2468 | * in detail. | |
2469 | */ | |
2470 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) | |
2471 | __acquires(dev_base_lock) | |
2472 | { | |
2473 | struct net *net = seq_file_net(seq); | |
2474 | loff_t off; | |
2475 | struct net_device *dev; | |
2476 | ||
2477 | read_lock(&dev_base_lock); | |
2478 | if (!*pos) | |
2479 | return SEQ_START_TOKEN; | |
2480 | ||
2481 | off = 1; | |
2482 | for_each_netdev(net, dev) | |
2483 | if (off++ == *pos) | |
2484 | return dev; | |
2485 | ||
2486 | return NULL; | |
2487 | } | |
2488 | ||
2489 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2490 | { | |
2491 | struct net *net = seq_file_net(seq); | |
2492 | ++*pos; | |
2493 | return v == SEQ_START_TOKEN ? | |
2494 | first_net_device(net) : next_net_device((struct net_device *)v); | |
2495 | } | |
2496 | ||
2497 | void dev_seq_stop(struct seq_file *seq, void *v) | |
2498 | __releases(dev_base_lock) | |
2499 | { | |
2500 | read_unlock(&dev_base_lock); | |
2501 | } | |
2502 | ||
2503 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) | |
2504 | { | |
2505 | struct net_device_stats *stats = dev->get_stats(dev); | |
2506 | ||
2507 | seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " | |
2508 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", | |
2509 | dev->name, stats->rx_bytes, stats->rx_packets, | |
2510 | stats->rx_errors, | |
2511 | stats->rx_dropped + stats->rx_missed_errors, | |
2512 | stats->rx_fifo_errors, | |
2513 | stats->rx_length_errors + stats->rx_over_errors + | |
2514 | stats->rx_crc_errors + stats->rx_frame_errors, | |
2515 | stats->rx_compressed, stats->multicast, | |
2516 | stats->tx_bytes, stats->tx_packets, | |
2517 | stats->tx_errors, stats->tx_dropped, | |
2518 | stats->tx_fifo_errors, stats->collisions, | |
2519 | stats->tx_carrier_errors + | |
2520 | stats->tx_aborted_errors + | |
2521 | stats->tx_window_errors + | |
2522 | stats->tx_heartbeat_errors, | |
2523 | stats->tx_compressed); | |
2524 | } | |
2525 | ||
2526 | /* | |
2527 | * Called from the PROCfs module. This now uses the new arbitrary sized | |
2528 | * /proc/net interface to create /proc/net/dev | |
2529 | */ | |
2530 | static int dev_seq_show(struct seq_file *seq, void *v) | |
2531 | { | |
2532 | if (v == SEQ_START_TOKEN) | |
2533 | seq_puts(seq, "Inter-| Receive " | |
2534 | " | Transmit\n" | |
2535 | " face |bytes packets errs drop fifo frame " | |
2536 | "compressed multicast|bytes packets errs " | |
2537 | "drop fifo colls carrier compressed\n"); | |
2538 | else | |
2539 | dev_seq_printf_stats(seq, v); | |
2540 | return 0; | |
2541 | } | |
2542 | ||
2543 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) | |
2544 | { | |
2545 | struct netif_rx_stats *rc = NULL; | |
2546 | ||
2547 | while (*pos < nr_cpu_ids) | |
2548 | if (cpu_online(*pos)) { | |
2549 | rc = &per_cpu(netdev_rx_stat, *pos); | |
2550 | break; | |
2551 | } else | |
2552 | ++*pos; | |
2553 | return rc; | |
2554 | } | |
2555 | ||
2556 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) | |
2557 | { | |
2558 | return softnet_get_online(pos); | |
2559 | } | |
2560 | ||
2561 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2562 | { | |
2563 | ++*pos; | |
2564 | return softnet_get_online(pos); | |
2565 | } | |
2566 | ||
2567 | static void softnet_seq_stop(struct seq_file *seq, void *v) | |
2568 | { | |
2569 | } | |
2570 | ||
2571 | static int softnet_seq_show(struct seq_file *seq, void *v) | |
2572 | { | |
2573 | struct netif_rx_stats *s = v; | |
2574 | ||
2575 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n", | |
2576 | s->total, s->dropped, s->time_squeeze, 0, | |
2577 | 0, 0, 0, 0, /* was fastroute */ | |
2578 | s->cpu_collision ); | |
2579 | return 0; | |
2580 | } | |
2581 | ||
2582 | static const struct seq_operations dev_seq_ops = { | |
2583 | .start = dev_seq_start, | |
2584 | .next = dev_seq_next, | |
2585 | .stop = dev_seq_stop, | |
2586 | .show = dev_seq_show, | |
2587 | }; | |
2588 | ||
2589 | static int dev_seq_open(struct inode *inode, struct file *file) | |
2590 | { | |
2591 | return seq_open_net(inode, file, &dev_seq_ops, | |
2592 | sizeof(struct seq_net_private)); | |
2593 | } | |
2594 | ||
2595 | static const struct file_operations dev_seq_fops = { | |
2596 | .owner = THIS_MODULE, | |
2597 | .open = dev_seq_open, | |
2598 | .read = seq_read, | |
2599 | .llseek = seq_lseek, | |
2600 | .release = seq_release_net, | |
2601 | }; | |
2602 | ||
2603 | static const struct seq_operations softnet_seq_ops = { | |
2604 | .start = softnet_seq_start, | |
2605 | .next = softnet_seq_next, | |
2606 | .stop = softnet_seq_stop, | |
2607 | .show = softnet_seq_show, | |
2608 | }; | |
2609 | ||
2610 | static int softnet_seq_open(struct inode *inode, struct file *file) | |
2611 | { | |
2612 | return seq_open(file, &softnet_seq_ops); | |
2613 | } | |
2614 | ||
2615 | static const struct file_operations softnet_seq_fops = { | |
2616 | .owner = THIS_MODULE, | |
2617 | .open = softnet_seq_open, | |
2618 | .read = seq_read, | |
2619 | .llseek = seq_lseek, | |
2620 | .release = seq_release, | |
2621 | }; | |
2622 | ||
2623 | static void *ptype_get_idx(loff_t pos) | |
2624 | { | |
2625 | struct packet_type *pt = NULL; | |
2626 | loff_t i = 0; | |
2627 | int t; | |
2628 | ||
2629 | list_for_each_entry_rcu(pt, &ptype_all, list) { | |
2630 | if (i == pos) | |
2631 | return pt; | |
2632 | ++i; | |
2633 | } | |
2634 | ||
2635 | for (t = 0; t < PTYPE_HASH_SIZE; t++) { | |
2636 | list_for_each_entry_rcu(pt, &ptype_base[t], list) { | |
2637 | if (i == pos) | |
2638 | return pt; | |
2639 | ++i; | |
2640 | } | |
2641 | } | |
2642 | return NULL; | |
2643 | } | |
2644 | ||
2645 | static void *ptype_seq_start(struct seq_file *seq, loff_t *pos) | |
2646 | __acquires(RCU) | |
2647 | { | |
2648 | rcu_read_lock(); | |
2649 | return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN; | |
2650 | } | |
2651 | ||
2652 | static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2653 | { | |
2654 | struct packet_type *pt; | |
2655 | struct list_head *nxt; | |
2656 | int hash; | |
2657 | ||
2658 | ++*pos; | |
2659 | if (v == SEQ_START_TOKEN) | |
2660 | return ptype_get_idx(0); | |
2661 | ||
2662 | pt = v; | |
2663 | nxt = pt->list.next; | |
2664 | if (pt->type == htons(ETH_P_ALL)) { | |
2665 | if (nxt != &ptype_all) | |
2666 | goto found; | |
2667 | hash = 0; | |
2668 | nxt = ptype_base[0].next; | |
2669 | } else | |
2670 | hash = ntohs(pt->type) & PTYPE_HASH_MASK; | |
2671 | ||
2672 | while (nxt == &ptype_base[hash]) { | |
2673 | if (++hash >= PTYPE_HASH_SIZE) | |
2674 | return NULL; | |
2675 | nxt = ptype_base[hash].next; | |
2676 | } | |
2677 | found: | |
2678 | return list_entry(nxt, struct packet_type, list); | |
2679 | } | |
2680 | ||
2681 | static void ptype_seq_stop(struct seq_file *seq, void *v) | |
2682 | __releases(RCU) | |
2683 | { | |
2684 | rcu_read_unlock(); | |
2685 | } | |
2686 | ||
2687 | static void ptype_seq_decode(struct seq_file *seq, void *sym) | |
2688 | { | |
2689 | #ifdef CONFIG_KALLSYMS | |
2690 | unsigned long offset = 0, symsize; | |
2691 | const char *symname; | |
2692 | char *modname; | |
2693 | char namebuf[128]; | |
2694 | ||
2695 | symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset, | |
2696 | &modname, namebuf); | |
2697 | ||
2698 | if (symname) { | |
2699 | char *delim = ":"; | |
2700 | ||
2701 | if (!modname) | |
2702 | modname = delim = ""; | |
2703 | seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim, | |
2704 | symname, offset); | |
2705 | return; | |
2706 | } | |
2707 | #endif | |
2708 | ||
2709 | seq_printf(seq, "[%p]", sym); | |
2710 | } | |
2711 | ||
2712 | static int ptype_seq_show(struct seq_file *seq, void *v) | |
2713 | { | |
2714 | struct packet_type *pt = v; | |
2715 | ||
2716 | if (v == SEQ_START_TOKEN) | |
2717 | seq_puts(seq, "Type Device Function\n"); | |
2718 | else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) { | |
2719 | if (pt->type == htons(ETH_P_ALL)) | |
2720 | seq_puts(seq, "ALL "); | |
2721 | else | |
2722 | seq_printf(seq, "%04x", ntohs(pt->type)); | |
2723 | ||
2724 | seq_printf(seq, " %-8s ", | |
2725 | pt->dev ? pt->dev->name : ""); | |
2726 | ptype_seq_decode(seq, pt->func); | |
2727 | seq_putc(seq, '\n'); | |
2728 | } | |
2729 | ||
2730 | return 0; | |
2731 | } | |
2732 | ||
2733 | static const struct seq_operations ptype_seq_ops = { | |
2734 | .start = ptype_seq_start, | |
2735 | .next = ptype_seq_next, | |
2736 | .stop = ptype_seq_stop, | |
2737 | .show = ptype_seq_show, | |
2738 | }; | |
2739 | ||
2740 | static int ptype_seq_open(struct inode *inode, struct file *file) | |
2741 | { | |
2742 | return seq_open_net(inode, file, &ptype_seq_ops, | |
2743 | sizeof(struct seq_net_private)); | |
2744 | } | |
2745 | ||
2746 | static const struct file_operations ptype_seq_fops = { | |
2747 | .owner = THIS_MODULE, | |
2748 | .open = ptype_seq_open, | |
2749 | .read = seq_read, | |
2750 | .llseek = seq_lseek, | |
2751 | .release = seq_release_net, | |
2752 | }; | |
2753 | ||
2754 | ||
2755 | static int __net_init dev_proc_net_init(struct net *net) | |
2756 | { | |
2757 | int rc = -ENOMEM; | |
2758 | ||
2759 | if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops)) | |
2760 | goto out; | |
2761 | if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops)) | |
2762 | goto out_dev; | |
2763 | if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops)) | |
2764 | goto out_softnet; | |
2765 | ||
2766 | if (wext_proc_init(net)) | |
2767 | goto out_ptype; | |
2768 | rc = 0; | |
2769 | out: | |
2770 | return rc; | |
2771 | out_ptype: | |
2772 | proc_net_remove(net, "ptype"); | |
2773 | out_softnet: | |
2774 | proc_net_remove(net, "softnet_stat"); | |
2775 | out_dev: | |
2776 | proc_net_remove(net, "dev"); | |
2777 | goto out; | |
2778 | } | |
2779 | ||
2780 | static void __net_exit dev_proc_net_exit(struct net *net) | |
2781 | { | |
2782 | wext_proc_exit(net); | |
2783 | ||
2784 | proc_net_remove(net, "ptype"); | |
2785 | proc_net_remove(net, "softnet_stat"); | |
2786 | proc_net_remove(net, "dev"); | |
2787 | } | |
2788 | ||
2789 | static struct pernet_operations __net_initdata dev_proc_ops = { | |
2790 | .init = dev_proc_net_init, | |
2791 | .exit = dev_proc_net_exit, | |
2792 | }; | |
2793 | ||
2794 | static int __init dev_proc_init(void) | |
2795 | { | |
2796 | return register_pernet_subsys(&dev_proc_ops); | |
2797 | } | |
2798 | #else | |
2799 | #define dev_proc_init() 0 | |
2800 | #endif /* CONFIG_PROC_FS */ | |
2801 | ||
2802 | ||
2803 | /** | |
2804 | * netdev_set_master - set up master/slave pair | |
2805 | * @slave: slave device | |
2806 | * @master: new master device | |
2807 | * | |
2808 | * Changes the master device of the slave. Pass %NULL to break the | |
2809 | * bonding. The caller must hold the RTNL semaphore. On a failure | |
2810 | * a negative errno code is returned. On success the reference counts | |
2811 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the | |
2812 | * function returns zero. | |
2813 | */ | |
2814 | int netdev_set_master(struct net_device *slave, struct net_device *master) | |
2815 | { | |
2816 | struct net_device *old = slave->master; | |
2817 | ||
2818 | ASSERT_RTNL(); | |
2819 | ||
2820 | if (master) { | |
2821 | if (old) | |
2822 | return -EBUSY; | |
2823 | dev_hold(master); | |
2824 | } | |
2825 | ||
2826 | slave->master = master; | |
2827 | ||
2828 | synchronize_net(); | |
2829 | ||
2830 | if (old) | |
2831 | dev_put(old); | |
2832 | ||
2833 | if (master) | |
2834 | slave->flags |= IFF_SLAVE; | |
2835 | else | |
2836 | slave->flags &= ~IFF_SLAVE; | |
2837 | ||
2838 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); | |
2839 | return 0; | |
2840 | } | |
2841 | ||
2842 | static int __dev_set_promiscuity(struct net_device *dev, int inc) | |
2843 | { | |
2844 | unsigned short old_flags = dev->flags; | |
2845 | ||
2846 | ASSERT_RTNL(); | |
2847 | ||
2848 | dev->flags |= IFF_PROMISC; | |
2849 | dev->promiscuity += inc; | |
2850 | if (dev->promiscuity == 0) { | |
2851 | /* | |
2852 | * Avoid overflow. | |
2853 | * If inc causes overflow, untouch promisc and return error. | |
2854 | */ | |
2855 | if (inc < 0) | |
2856 | dev->flags &= ~IFF_PROMISC; | |
2857 | else { | |
2858 | dev->promiscuity -= inc; | |
2859 | printk(KERN_WARNING "%s: promiscuity touches roof, " | |
2860 | "set promiscuity failed, promiscuity feature " | |
2861 | "of device might be broken.\n", dev->name); | |
2862 | return -EOVERFLOW; | |
2863 | } | |
2864 | } | |
2865 | if (dev->flags != old_flags) { | |
2866 | printk(KERN_INFO "device %s %s promiscuous mode\n", | |
2867 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : | |
2868 | "left"); | |
2869 | if (audit_enabled) | |
2870 | audit_log(current->audit_context, GFP_ATOMIC, | |
2871 | AUDIT_ANOM_PROMISCUOUS, | |
2872 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
2873 | dev->name, (dev->flags & IFF_PROMISC), | |
2874 | (old_flags & IFF_PROMISC), | |
2875 | audit_get_loginuid(current), | |
2876 | current->uid, current->gid, | |
2877 | audit_get_sessionid(current)); | |
2878 | ||
2879 | if (dev->change_rx_flags) | |
2880 | dev->change_rx_flags(dev, IFF_PROMISC); | |
2881 | } | |
2882 | return 0; | |
2883 | } | |
2884 | ||
2885 | /** | |
2886 | * dev_set_promiscuity - update promiscuity count on a device | |
2887 | * @dev: device | |
2888 | * @inc: modifier | |
2889 | * | |
2890 | * Add or remove promiscuity from a device. While the count in the device | |
2891 | * remains above zero the interface remains promiscuous. Once it hits zero | |
2892 | * the device reverts back to normal filtering operation. A negative inc | |
2893 | * value is used to drop promiscuity on the device. | |
2894 | * Return 0 if successful or a negative errno code on error. | |
2895 | */ | |
2896 | int dev_set_promiscuity(struct net_device *dev, int inc) | |
2897 | { | |
2898 | unsigned short old_flags = dev->flags; | |
2899 | int err; | |
2900 | ||
2901 | err = __dev_set_promiscuity(dev, inc); | |
2902 | if (err < 0) | |
2903 | return err; | |
2904 | if (dev->flags != old_flags) | |
2905 | dev_set_rx_mode(dev); | |
2906 | return err; | |
2907 | } | |
2908 | ||
2909 | /** | |
2910 | * dev_set_allmulti - update allmulti count on a device | |
2911 | * @dev: device | |
2912 | * @inc: modifier | |
2913 | * | |
2914 | * Add or remove reception of all multicast frames to a device. While the | |
2915 | * count in the device remains above zero the interface remains listening | |
2916 | * to all interfaces. Once it hits zero the device reverts back to normal | |
2917 | * filtering operation. A negative @inc value is used to drop the counter | |
2918 | * when releasing a resource needing all multicasts. | |
2919 | * Return 0 if successful or a negative errno code on error. | |
2920 | */ | |
2921 | ||
2922 | int dev_set_allmulti(struct net_device *dev, int inc) | |
2923 | { | |
2924 | unsigned short old_flags = dev->flags; | |
2925 | ||
2926 | ASSERT_RTNL(); | |
2927 | ||
2928 | dev->flags |= IFF_ALLMULTI; | |
2929 | dev->allmulti += inc; | |
2930 | if (dev->allmulti == 0) { | |
2931 | /* | |
2932 | * Avoid overflow. | |
2933 | * If inc causes overflow, untouch allmulti and return error. | |
2934 | */ | |
2935 | if (inc < 0) | |
2936 | dev->flags &= ~IFF_ALLMULTI; | |
2937 | else { | |
2938 | dev->allmulti -= inc; | |
2939 | printk(KERN_WARNING "%s: allmulti touches roof, " | |
2940 | "set allmulti failed, allmulti feature of " | |
2941 | "device might be broken.\n", dev->name); | |
2942 | return -EOVERFLOW; | |
2943 | } | |
2944 | } | |
2945 | if (dev->flags ^ old_flags) { | |
2946 | if (dev->change_rx_flags) | |
2947 | dev->change_rx_flags(dev, IFF_ALLMULTI); | |
2948 | dev_set_rx_mode(dev); | |
2949 | } | |
2950 | return 0; | |
2951 | } | |
2952 | ||
2953 | /* | |
2954 | * Upload unicast and multicast address lists to device and | |
2955 | * configure RX filtering. When the device doesn't support unicast | |
2956 | * filtering it is put in promiscuous mode while unicast addresses | |
2957 | * are present. | |
2958 | */ | |
2959 | void __dev_set_rx_mode(struct net_device *dev) | |
2960 | { | |
2961 | /* dev_open will call this function so the list will stay sane. */ | |
2962 | if (!(dev->flags&IFF_UP)) | |
2963 | return; | |
2964 | ||
2965 | if (!netif_device_present(dev)) | |
2966 | return; | |
2967 | ||
2968 | if (dev->set_rx_mode) | |
2969 | dev->set_rx_mode(dev); | |
2970 | else { | |
2971 | /* Unicast addresses changes may only happen under the rtnl, | |
2972 | * therefore calling __dev_set_promiscuity here is safe. | |
2973 | */ | |
2974 | if (dev->uc_count > 0 && !dev->uc_promisc) { | |
2975 | __dev_set_promiscuity(dev, 1); | |
2976 | dev->uc_promisc = 1; | |
2977 | } else if (dev->uc_count == 0 && dev->uc_promisc) { | |
2978 | __dev_set_promiscuity(dev, -1); | |
2979 | dev->uc_promisc = 0; | |
2980 | } | |
2981 | ||
2982 | if (dev->set_multicast_list) | |
2983 | dev->set_multicast_list(dev); | |
2984 | } | |
2985 | } | |
2986 | ||
2987 | void dev_set_rx_mode(struct net_device *dev) | |
2988 | { | |
2989 | netif_addr_lock_bh(dev); | |
2990 | __dev_set_rx_mode(dev); | |
2991 | netif_addr_unlock_bh(dev); | |
2992 | } | |
2993 | ||
2994 | int __dev_addr_delete(struct dev_addr_list **list, int *count, | |
2995 | void *addr, int alen, int glbl) | |
2996 | { | |
2997 | struct dev_addr_list *da; | |
2998 | ||
2999 | for (; (da = *list) != NULL; list = &da->next) { | |
3000 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
3001 | alen == da->da_addrlen) { | |
3002 | if (glbl) { | |
3003 | int old_glbl = da->da_gusers; | |
3004 | da->da_gusers = 0; | |
3005 | if (old_glbl == 0) | |
3006 | break; | |
3007 | } | |
3008 | if (--da->da_users) | |
3009 | return 0; | |
3010 | ||
3011 | *list = da->next; | |
3012 | kfree(da); | |
3013 | (*count)--; | |
3014 | return 0; | |
3015 | } | |
3016 | } | |
3017 | return -ENOENT; | |
3018 | } | |
3019 | ||
3020 | int __dev_addr_add(struct dev_addr_list **list, int *count, | |
3021 | void *addr, int alen, int glbl) | |
3022 | { | |
3023 | struct dev_addr_list *da; | |
3024 | ||
3025 | for (da = *list; da != NULL; da = da->next) { | |
3026 | if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 && | |
3027 | da->da_addrlen == alen) { | |
3028 | if (glbl) { | |
3029 | int old_glbl = da->da_gusers; | |
3030 | da->da_gusers = 1; | |
3031 | if (old_glbl) | |
3032 | return 0; | |
3033 | } | |
3034 | da->da_users++; | |
3035 | return 0; | |
3036 | } | |
3037 | } | |
3038 | ||
3039 | da = kzalloc(sizeof(*da), GFP_ATOMIC); | |
3040 | if (da == NULL) | |
3041 | return -ENOMEM; | |
3042 | memcpy(da->da_addr, addr, alen); | |
3043 | da->da_addrlen = alen; | |
3044 | da->da_users = 1; | |
3045 | da->da_gusers = glbl ? 1 : 0; | |
3046 | da->next = *list; | |
3047 | *list = da; | |
3048 | (*count)++; | |
3049 | return 0; | |
3050 | } | |
3051 | ||
3052 | /** | |
3053 | * dev_unicast_delete - Release secondary unicast address. | |
3054 | * @dev: device | |
3055 | * @addr: address to delete | |
3056 | * @alen: length of @addr | |
3057 | * | |
3058 | * Release reference to a secondary unicast address and remove it | |
3059 | * from the device if the reference count drops to zero. | |
3060 | * | |
3061 | * The caller must hold the rtnl_mutex. | |
3062 | */ | |
3063 | int dev_unicast_delete(struct net_device *dev, void *addr, int alen) | |
3064 | { | |
3065 | int err; | |
3066 | ||
3067 | ASSERT_RTNL(); | |
3068 | ||
3069 | netif_addr_lock_bh(dev); | |
3070 | err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0); | |
3071 | if (!err) | |
3072 | __dev_set_rx_mode(dev); | |
3073 | netif_addr_unlock_bh(dev); | |
3074 | return err; | |
3075 | } | |
3076 | EXPORT_SYMBOL(dev_unicast_delete); | |
3077 | ||
3078 | /** | |
3079 | * dev_unicast_add - add a secondary unicast address | |
3080 | * @dev: device | |
3081 | * @addr: address to add | |
3082 | * @alen: length of @addr | |
3083 | * | |
3084 | * Add a secondary unicast address to the device or increase | |
3085 | * the reference count if it already exists. | |
3086 | * | |
3087 | * The caller must hold the rtnl_mutex. | |
3088 | */ | |
3089 | int dev_unicast_add(struct net_device *dev, void *addr, int alen) | |
3090 | { | |
3091 | int err; | |
3092 | ||
3093 | ASSERT_RTNL(); | |
3094 | ||
3095 | netif_addr_lock_bh(dev); | |
3096 | err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0); | |
3097 | if (!err) | |
3098 | __dev_set_rx_mode(dev); | |
3099 | netif_addr_unlock_bh(dev); | |
3100 | return err; | |
3101 | } | |
3102 | EXPORT_SYMBOL(dev_unicast_add); | |
3103 | ||
3104 | int __dev_addr_sync(struct dev_addr_list **to, int *to_count, | |
3105 | struct dev_addr_list **from, int *from_count) | |
3106 | { | |
3107 | struct dev_addr_list *da, *next; | |
3108 | int err = 0; | |
3109 | ||
3110 | da = *from; | |
3111 | while (da != NULL) { | |
3112 | next = da->next; | |
3113 | if (!da->da_synced) { | |
3114 | err = __dev_addr_add(to, to_count, | |
3115 | da->da_addr, da->da_addrlen, 0); | |
3116 | if (err < 0) | |
3117 | break; | |
3118 | da->da_synced = 1; | |
3119 | da->da_users++; | |
3120 | } else if (da->da_users == 1) { | |
3121 | __dev_addr_delete(to, to_count, | |
3122 | da->da_addr, da->da_addrlen, 0); | |
3123 | __dev_addr_delete(from, from_count, | |
3124 | da->da_addr, da->da_addrlen, 0); | |
3125 | } | |
3126 | da = next; | |
3127 | } | |
3128 | return err; | |
3129 | } | |
3130 | ||
3131 | void __dev_addr_unsync(struct dev_addr_list **to, int *to_count, | |
3132 | struct dev_addr_list **from, int *from_count) | |
3133 | { | |
3134 | struct dev_addr_list *da, *next; | |
3135 | ||
3136 | da = *from; | |
3137 | while (da != NULL) { | |
3138 | next = da->next; | |
3139 | if (da->da_synced) { | |
3140 | __dev_addr_delete(to, to_count, | |
3141 | da->da_addr, da->da_addrlen, 0); | |
3142 | da->da_synced = 0; | |
3143 | __dev_addr_delete(from, from_count, | |
3144 | da->da_addr, da->da_addrlen, 0); | |
3145 | } | |
3146 | da = next; | |
3147 | } | |
3148 | } | |
3149 | ||
3150 | /** | |
3151 | * dev_unicast_sync - Synchronize device's unicast list to another device | |
3152 | * @to: destination device | |
3153 | * @from: source device | |
3154 | * | |
3155 | * Add newly added addresses to the destination device and release | |
3156 | * addresses that have no users left. The source device must be | |
3157 | * locked by netif_tx_lock_bh. | |
3158 | * | |
3159 | * This function is intended to be called from the dev->set_rx_mode | |
3160 | * function of layered software devices. | |
3161 | */ | |
3162 | int dev_unicast_sync(struct net_device *to, struct net_device *from) | |
3163 | { | |
3164 | int err = 0; | |
3165 | ||
3166 | netif_addr_lock_bh(to); | |
3167 | err = __dev_addr_sync(&to->uc_list, &to->uc_count, | |
3168 | &from->uc_list, &from->uc_count); | |
3169 | if (!err) | |
3170 | __dev_set_rx_mode(to); | |
3171 | netif_addr_unlock_bh(to); | |
3172 | return err; | |
3173 | } | |
3174 | EXPORT_SYMBOL(dev_unicast_sync); | |
3175 | ||
3176 | /** | |
3177 | * dev_unicast_unsync - Remove synchronized addresses from the destination device | |
3178 | * @to: destination device | |
3179 | * @from: source device | |
3180 | * | |
3181 | * Remove all addresses that were added to the destination device by | |
3182 | * dev_unicast_sync(). This function is intended to be called from the | |
3183 | * dev->stop function of layered software devices. | |
3184 | */ | |
3185 | void dev_unicast_unsync(struct net_device *to, struct net_device *from) | |
3186 | { | |
3187 | netif_addr_lock_bh(from); | |
3188 | netif_addr_lock(to); | |
3189 | ||
3190 | __dev_addr_unsync(&to->uc_list, &to->uc_count, | |
3191 | &from->uc_list, &from->uc_count); | |
3192 | __dev_set_rx_mode(to); | |
3193 | ||
3194 | netif_addr_unlock(to); | |
3195 | netif_addr_unlock_bh(from); | |
3196 | } | |
3197 | EXPORT_SYMBOL(dev_unicast_unsync); | |
3198 | ||
3199 | static void __dev_addr_discard(struct dev_addr_list **list) | |
3200 | { | |
3201 | struct dev_addr_list *tmp; | |
3202 | ||
3203 | while (*list != NULL) { | |
3204 | tmp = *list; | |
3205 | *list = tmp->next; | |
3206 | if (tmp->da_users > tmp->da_gusers) | |
3207 | printk("__dev_addr_discard: address leakage! " | |
3208 | "da_users=%d\n", tmp->da_users); | |
3209 | kfree(tmp); | |
3210 | } | |
3211 | } | |
3212 | ||
3213 | static void dev_addr_discard(struct net_device *dev) | |
3214 | { | |
3215 | netif_addr_lock_bh(dev); | |
3216 | ||
3217 | __dev_addr_discard(&dev->uc_list); | |
3218 | dev->uc_count = 0; | |
3219 | ||
3220 | __dev_addr_discard(&dev->mc_list); | |
3221 | dev->mc_count = 0; | |
3222 | ||
3223 | netif_addr_unlock_bh(dev); | |
3224 | } | |
3225 | ||
3226 | unsigned dev_get_flags(const struct net_device *dev) | |
3227 | { | |
3228 | unsigned flags; | |
3229 | ||
3230 | flags = (dev->flags & ~(IFF_PROMISC | | |
3231 | IFF_ALLMULTI | | |
3232 | IFF_RUNNING | | |
3233 | IFF_LOWER_UP | | |
3234 | IFF_DORMANT)) | | |
3235 | (dev->gflags & (IFF_PROMISC | | |
3236 | IFF_ALLMULTI)); | |
3237 | ||
3238 | if (netif_running(dev)) { | |
3239 | if (netif_oper_up(dev)) | |
3240 | flags |= IFF_RUNNING; | |
3241 | if (netif_carrier_ok(dev)) | |
3242 | flags |= IFF_LOWER_UP; | |
3243 | if (netif_dormant(dev)) | |
3244 | flags |= IFF_DORMANT; | |
3245 | } | |
3246 | ||
3247 | return flags; | |
3248 | } | |
3249 | ||
3250 | int dev_change_flags(struct net_device *dev, unsigned flags) | |
3251 | { | |
3252 | int ret, changes; | |
3253 | int old_flags = dev->flags; | |
3254 | ||
3255 | ASSERT_RTNL(); | |
3256 | ||
3257 | /* | |
3258 | * Set the flags on our device. | |
3259 | */ | |
3260 | ||
3261 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
3262 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
3263 | IFF_AUTOMEDIA)) | | |
3264 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
3265 | IFF_ALLMULTI)); | |
3266 | ||
3267 | /* | |
3268 | * Load in the correct multicast list now the flags have changed. | |
3269 | */ | |
3270 | ||
3271 | if (dev->change_rx_flags && (old_flags ^ flags) & IFF_MULTICAST) | |
3272 | dev->change_rx_flags(dev, IFF_MULTICAST); | |
3273 | ||
3274 | dev_set_rx_mode(dev); | |
3275 | ||
3276 | /* | |
3277 | * Have we downed the interface. We handle IFF_UP ourselves | |
3278 | * according to user attempts to set it, rather than blindly | |
3279 | * setting it. | |
3280 | */ | |
3281 | ||
3282 | ret = 0; | |
3283 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ | |
3284 | ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev); | |
3285 | ||
3286 | if (!ret) | |
3287 | dev_set_rx_mode(dev); | |
3288 | } | |
3289 | ||
3290 | if (dev->flags & IFF_UP && | |
3291 | ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI | | |
3292 | IFF_VOLATILE))) | |
3293 | call_netdevice_notifiers(NETDEV_CHANGE, dev); | |
3294 | ||
3295 | if ((flags ^ dev->gflags) & IFF_PROMISC) { | |
3296 | int inc = (flags & IFF_PROMISC) ? +1 : -1; | |
3297 | dev->gflags ^= IFF_PROMISC; | |
3298 | dev_set_promiscuity(dev, inc); | |
3299 | } | |
3300 | ||
3301 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
3302 | is important. Some (broken) drivers set IFF_PROMISC, when | |
3303 | IFF_ALLMULTI is requested not asking us and not reporting. | |
3304 | */ | |
3305 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
3306 | int inc = (flags & IFF_ALLMULTI) ? +1 : -1; | |
3307 | dev->gflags ^= IFF_ALLMULTI; | |
3308 | dev_set_allmulti(dev, inc); | |
3309 | } | |
3310 | ||
3311 | /* Exclude state transition flags, already notified */ | |
3312 | changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING); | |
3313 | if (changes) | |
3314 | rtmsg_ifinfo(RTM_NEWLINK, dev, changes); | |
3315 | ||
3316 | return ret; | |
3317 | } | |
3318 | ||
3319 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
3320 | { | |
3321 | int err; | |
3322 | ||
3323 | if (new_mtu == dev->mtu) | |
3324 | return 0; | |
3325 | ||
3326 | /* MTU must be positive. */ | |
3327 | if (new_mtu < 0) | |
3328 | return -EINVAL; | |
3329 | ||
3330 | if (!netif_device_present(dev)) | |
3331 | return -ENODEV; | |
3332 | ||
3333 | err = 0; | |
3334 | if (dev->change_mtu) | |
3335 | err = dev->change_mtu(dev, new_mtu); | |
3336 | else | |
3337 | dev->mtu = new_mtu; | |
3338 | if (!err && dev->flags & IFF_UP) | |
3339 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
3340 | return err; | |
3341 | } | |
3342 | ||
3343 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) | |
3344 | { | |
3345 | int err; | |
3346 | ||
3347 | if (!dev->set_mac_address) | |
3348 | return -EOPNOTSUPP; | |
3349 | if (sa->sa_family != dev->type) | |
3350 | return -EINVAL; | |
3351 | if (!netif_device_present(dev)) | |
3352 | return -ENODEV; | |
3353 | err = dev->set_mac_address(dev, sa); | |
3354 | if (!err) | |
3355 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
3356 | return err; | |
3357 | } | |
3358 | ||
3359 | /* | |
3360 | * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock) | |
3361 | */ | |
3362 | static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
3363 | { | |
3364 | int err; | |
3365 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
3366 | ||
3367 | if (!dev) | |
3368 | return -ENODEV; | |
3369 | ||
3370 | switch (cmd) { | |
3371 | case SIOCGIFFLAGS: /* Get interface flags */ | |
3372 | ifr->ifr_flags = dev_get_flags(dev); | |
3373 | return 0; | |
3374 | ||
3375 | case SIOCGIFMETRIC: /* Get the metric on the interface | |
3376 | (currently unused) */ | |
3377 | ifr->ifr_metric = 0; | |
3378 | return 0; | |
3379 | ||
3380 | case SIOCGIFMTU: /* Get the MTU of a device */ | |
3381 | ifr->ifr_mtu = dev->mtu; | |
3382 | return 0; | |
3383 | ||
3384 | case SIOCGIFHWADDR: | |
3385 | if (!dev->addr_len) | |
3386 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); | |
3387 | else | |
3388 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, | |
3389 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
3390 | ifr->ifr_hwaddr.sa_family = dev->type; | |
3391 | return 0; | |
3392 | ||
3393 | case SIOCGIFSLAVE: | |
3394 | err = -EINVAL; | |
3395 | break; | |
3396 | ||
3397 | case SIOCGIFMAP: | |
3398 | ifr->ifr_map.mem_start = dev->mem_start; | |
3399 | ifr->ifr_map.mem_end = dev->mem_end; | |
3400 | ifr->ifr_map.base_addr = dev->base_addr; | |
3401 | ifr->ifr_map.irq = dev->irq; | |
3402 | ifr->ifr_map.dma = dev->dma; | |
3403 | ifr->ifr_map.port = dev->if_port; | |
3404 | return 0; | |
3405 | ||
3406 | case SIOCGIFINDEX: | |
3407 | ifr->ifr_ifindex = dev->ifindex; | |
3408 | return 0; | |
3409 | ||
3410 | case SIOCGIFTXQLEN: | |
3411 | ifr->ifr_qlen = dev->tx_queue_len; | |
3412 | return 0; | |
3413 | ||
3414 | default: | |
3415 | /* dev_ioctl() should ensure this case | |
3416 | * is never reached | |
3417 | */ | |
3418 | WARN_ON(1); | |
3419 | err = -EINVAL; | |
3420 | break; | |
3421 | ||
3422 | } | |
3423 | return err; | |
3424 | } | |
3425 | ||
3426 | /* | |
3427 | * Perform the SIOCxIFxxx calls, inside rtnl_lock() | |
3428 | */ | |
3429 | static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd) | |
3430 | { | |
3431 | int err; | |
3432 | struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name); | |
3433 | ||
3434 | if (!dev) | |
3435 | return -ENODEV; | |
3436 | ||
3437 | switch (cmd) { | |
3438 | case SIOCSIFFLAGS: /* Set interface flags */ | |
3439 | return dev_change_flags(dev, ifr->ifr_flags); | |
3440 | ||
3441 | case SIOCSIFMETRIC: /* Set the metric on the interface | |
3442 | (currently unused) */ | |
3443 | return -EOPNOTSUPP; | |
3444 | ||
3445 | case SIOCSIFMTU: /* Set the MTU of a device */ | |
3446 | return dev_set_mtu(dev, ifr->ifr_mtu); | |
3447 | ||
3448 | case SIOCSIFHWADDR: | |
3449 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); | |
3450 | ||
3451 | case SIOCSIFHWBROADCAST: | |
3452 | if (ifr->ifr_hwaddr.sa_family != dev->type) | |
3453 | return -EINVAL; | |
3454 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, | |
3455 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); | |
3456 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); | |
3457 | return 0; | |
3458 | ||
3459 | case SIOCSIFMAP: | |
3460 | if (dev->set_config) { | |
3461 | if (!netif_device_present(dev)) | |
3462 | return -ENODEV; | |
3463 | return dev->set_config(dev, &ifr->ifr_map); | |
3464 | } | |
3465 | return -EOPNOTSUPP; | |
3466 | ||
3467 | case SIOCADDMULTI: | |
3468 | if ((!dev->set_multicast_list && !dev->set_rx_mode) || | |
3469 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
3470 | return -EINVAL; | |
3471 | if (!netif_device_present(dev)) | |
3472 | return -ENODEV; | |
3473 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, | |
3474 | dev->addr_len, 1); | |
3475 | ||
3476 | case SIOCDELMULTI: | |
3477 | if ((!dev->set_multicast_list && !dev->set_rx_mode) || | |
3478 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) | |
3479 | return -EINVAL; | |
3480 | if (!netif_device_present(dev)) | |
3481 | return -ENODEV; | |
3482 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, | |
3483 | dev->addr_len, 1); | |
3484 | ||
3485 | case SIOCSIFTXQLEN: | |
3486 | if (ifr->ifr_qlen < 0) | |
3487 | return -EINVAL; | |
3488 | dev->tx_queue_len = ifr->ifr_qlen; | |
3489 | return 0; | |
3490 | ||
3491 | case SIOCSIFNAME: | |
3492 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; | |
3493 | return dev_change_name(dev, ifr->ifr_newname); | |
3494 | ||
3495 | /* | |
3496 | * Unknown or private ioctl | |
3497 | */ | |
3498 | ||
3499 | default: | |
3500 | if ((cmd >= SIOCDEVPRIVATE && | |
3501 | cmd <= SIOCDEVPRIVATE + 15) || | |
3502 | cmd == SIOCBONDENSLAVE || | |
3503 | cmd == SIOCBONDRELEASE || | |
3504 | cmd == SIOCBONDSETHWADDR || | |
3505 | cmd == SIOCBONDSLAVEINFOQUERY || | |
3506 | cmd == SIOCBONDINFOQUERY || | |
3507 | cmd == SIOCBONDCHANGEACTIVE || | |
3508 | cmd == SIOCGMIIPHY || | |
3509 | cmd == SIOCGMIIREG || | |
3510 | cmd == SIOCSMIIREG || | |
3511 | cmd == SIOCBRADDIF || | |
3512 | cmd == SIOCBRDELIF || | |
3513 | cmd == SIOCWANDEV) { | |
3514 | err = -EOPNOTSUPP; | |
3515 | if (dev->do_ioctl) { | |
3516 | if (netif_device_present(dev)) | |
3517 | err = dev->do_ioctl(dev, ifr, | |
3518 | cmd); | |
3519 | else | |
3520 | err = -ENODEV; | |
3521 | } | |
3522 | } else | |
3523 | err = -EINVAL; | |
3524 | ||
3525 | } | |
3526 | return err; | |
3527 | } | |
3528 | ||
3529 | /* | |
3530 | * This function handles all "interface"-type I/O control requests. The actual | |
3531 | * 'doing' part of this is dev_ifsioc above. | |
3532 | */ | |
3533 | ||
3534 | /** | |
3535 | * dev_ioctl - network device ioctl | |
3536 | * @net: the applicable net namespace | |
3537 | * @cmd: command to issue | |
3538 | * @arg: pointer to a struct ifreq in user space | |
3539 | * | |
3540 | * Issue ioctl functions to devices. This is normally called by the | |
3541 | * user space syscall interfaces but can sometimes be useful for | |
3542 | * other purposes. The return value is the return from the syscall if | |
3543 | * positive or a negative errno code on error. | |
3544 | */ | |
3545 | ||
3546 | int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg) | |
3547 | { | |
3548 | struct ifreq ifr; | |
3549 | int ret; | |
3550 | char *colon; | |
3551 | ||
3552 | /* One special case: SIOCGIFCONF takes ifconf argument | |
3553 | and requires shared lock, because it sleeps writing | |
3554 | to user space. | |
3555 | */ | |
3556 | ||
3557 | if (cmd == SIOCGIFCONF) { | |
3558 | rtnl_lock(); | |
3559 | ret = dev_ifconf(net, (char __user *) arg); | |
3560 | rtnl_unlock(); | |
3561 | return ret; | |
3562 | } | |
3563 | if (cmd == SIOCGIFNAME) | |
3564 | return dev_ifname(net, (struct ifreq __user *)arg); | |
3565 | ||
3566 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) | |
3567 | return -EFAULT; | |
3568 | ||
3569 | ifr.ifr_name[IFNAMSIZ-1] = 0; | |
3570 | ||
3571 | colon = strchr(ifr.ifr_name, ':'); | |
3572 | if (colon) | |
3573 | *colon = 0; | |
3574 | ||
3575 | /* | |
3576 | * See which interface the caller is talking about. | |
3577 | */ | |
3578 | ||
3579 | switch (cmd) { | |
3580 | /* | |
3581 | * These ioctl calls: | |
3582 | * - can be done by all. | |
3583 | * - atomic and do not require locking. | |
3584 | * - return a value | |
3585 | */ | |
3586 | case SIOCGIFFLAGS: | |
3587 | case SIOCGIFMETRIC: | |
3588 | case SIOCGIFMTU: | |
3589 | case SIOCGIFHWADDR: | |
3590 | case SIOCGIFSLAVE: | |
3591 | case SIOCGIFMAP: | |
3592 | case SIOCGIFINDEX: | |
3593 | case SIOCGIFTXQLEN: | |
3594 | dev_load(net, ifr.ifr_name); | |
3595 | read_lock(&dev_base_lock); | |
3596 | ret = dev_ifsioc_locked(net, &ifr, cmd); | |
3597 | read_unlock(&dev_base_lock); | |
3598 | if (!ret) { | |
3599 | if (colon) | |
3600 | *colon = ':'; | |
3601 | if (copy_to_user(arg, &ifr, | |
3602 | sizeof(struct ifreq))) | |
3603 | ret = -EFAULT; | |
3604 | } | |
3605 | return ret; | |
3606 | ||
3607 | case SIOCETHTOOL: | |
3608 | dev_load(net, ifr.ifr_name); | |
3609 | rtnl_lock(); | |
3610 | ret = dev_ethtool(net, &ifr); | |
3611 | rtnl_unlock(); | |
3612 | if (!ret) { | |
3613 | if (colon) | |
3614 | *colon = ':'; | |
3615 | if (copy_to_user(arg, &ifr, | |
3616 | sizeof(struct ifreq))) | |
3617 | ret = -EFAULT; | |
3618 | } | |
3619 | return ret; | |
3620 | ||
3621 | /* | |
3622 | * These ioctl calls: | |
3623 | * - require superuser power. | |
3624 | * - require strict serialization. | |
3625 | * - return a value | |
3626 | */ | |
3627 | case SIOCGMIIPHY: | |
3628 | case SIOCGMIIREG: | |
3629 | case SIOCSIFNAME: | |
3630 | if (!capable(CAP_NET_ADMIN)) | |
3631 | return -EPERM; | |
3632 | dev_load(net, ifr.ifr_name); | |
3633 | rtnl_lock(); | |
3634 | ret = dev_ifsioc(net, &ifr, cmd); | |
3635 | rtnl_unlock(); | |
3636 | if (!ret) { | |
3637 | if (colon) | |
3638 | *colon = ':'; | |
3639 | if (copy_to_user(arg, &ifr, | |
3640 | sizeof(struct ifreq))) | |
3641 | ret = -EFAULT; | |
3642 | } | |
3643 | return ret; | |
3644 | ||
3645 | /* | |
3646 | * These ioctl calls: | |
3647 | * - require superuser power. | |
3648 | * - require strict serialization. | |
3649 | * - do not return a value | |
3650 | */ | |
3651 | case SIOCSIFFLAGS: | |
3652 | case SIOCSIFMETRIC: | |
3653 | case SIOCSIFMTU: | |
3654 | case SIOCSIFMAP: | |
3655 | case SIOCSIFHWADDR: | |
3656 | case SIOCSIFSLAVE: | |
3657 | case SIOCADDMULTI: | |
3658 | case SIOCDELMULTI: | |
3659 | case SIOCSIFHWBROADCAST: | |
3660 | case SIOCSIFTXQLEN: | |
3661 | case SIOCSMIIREG: | |
3662 | case SIOCBONDENSLAVE: | |
3663 | case SIOCBONDRELEASE: | |
3664 | case SIOCBONDSETHWADDR: | |
3665 | case SIOCBONDCHANGEACTIVE: | |
3666 | case SIOCBRADDIF: | |
3667 | case SIOCBRDELIF: | |
3668 | if (!capable(CAP_NET_ADMIN)) | |
3669 | return -EPERM; | |
3670 | /* fall through */ | |
3671 | case SIOCBONDSLAVEINFOQUERY: | |
3672 | case SIOCBONDINFOQUERY: | |
3673 | dev_load(net, ifr.ifr_name); | |
3674 | rtnl_lock(); | |
3675 | ret = dev_ifsioc(net, &ifr, cmd); | |
3676 | rtnl_unlock(); | |
3677 | return ret; | |
3678 | ||
3679 | case SIOCGIFMEM: | |
3680 | /* Get the per device memory space. We can add this but | |
3681 | * currently do not support it */ | |
3682 | case SIOCSIFMEM: | |
3683 | /* Set the per device memory buffer space. | |
3684 | * Not applicable in our case */ | |
3685 | case SIOCSIFLINK: | |
3686 | return -EINVAL; | |
3687 | ||
3688 | /* | |
3689 | * Unknown or private ioctl. | |
3690 | */ | |
3691 | default: | |
3692 | if (cmd == SIOCWANDEV || | |
3693 | (cmd >= SIOCDEVPRIVATE && | |
3694 | cmd <= SIOCDEVPRIVATE + 15)) { | |
3695 | dev_load(net, ifr.ifr_name); | |
3696 | rtnl_lock(); | |
3697 | ret = dev_ifsioc(net, &ifr, cmd); | |
3698 | rtnl_unlock(); | |
3699 | if (!ret && copy_to_user(arg, &ifr, | |
3700 | sizeof(struct ifreq))) | |
3701 | ret = -EFAULT; | |
3702 | return ret; | |
3703 | } | |
3704 | /* Take care of Wireless Extensions */ | |
3705 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) | |
3706 | return wext_handle_ioctl(net, &ifr, cmd, arg); | |
3707 | return -EINVAL; | |
3708 | } | |
3709 | } | |
3710 | ||
3711 | ||
3712 | /** | |
3713 | * dev_new_index - allocate an ifindex | |
3714 | * @net: the applicable net namespace | |
3715 | * | |
3716 | * Returns a suitable unique value for a new device interface | |
3717 | * number. The caller must hold the rtnl semaphore or the | |
3718 | * dev_base_lock to be sure it remains unique. | |
3719 | */ | |
3720 | static int dev_new_index(struct net *net) | |
3721 | { | |
3722 | static int ifindex; | |
3723 | for (;;) { | |
3724 | if (++ifindex <= 0) | |
3725 | ifindex = 1; | |
3726 | if (!__dev_get_by_index(net, ifindex)) | |
3727 | return ifindex; | |
3728 | } | |
3729 | } | |
3730 | ||
3731 | /* Delayed registration/unregisteration */ | |
3732 | static DEFINE_SPINLOCK(net_todo_list_lock); | |
3733 | static LIST_HEAD(net_todo_list); | |
3734 | ||
3735 | static void net_set_todo(struct net_device *dev) | |
3736 | { | |
3737 | spin_lock(&net_todo_list_lock); | |
3738 | list_add_tail(&dev->todo_list, &net_todo_list); | |
3739 | spin_unlock(&net_todo_list_lock); | |
3740 | } | |
3741 | ||
3742 | static void rollback_registered(struct net_device *dev) | |
3743 | { | |
3744 | BUG_ON(dev_boot_phase); | |
3745 | ASSERT_RTNL(); | |
3746 | ||
3747 | /* Some devices call without registering for initialization unwind. */ | |
3748 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
3749 | printk(KERN_DEBUG "unregister_netdevice: device %s/%p never " | |
3750 | "was registered\n", dev->name, dev); | |
3751 | ||
3752 | WARN_ON(1); | |
3753 | return; | |
3754 | } | |
3755 | ||
3756 | BUG_ON(dev->reg_state != NETREG_REGISTERED); | |
3757 | ||
3758 | /* If device is running, close it first. */ | |
3759 | dev_close(dev); | |
3760 | ||
3761 | /* And unlink it from device chain. */ | |
3762 | unlist_netdevice(dev); | |
3763 | ||
3764 | dev->reg_state = NETREG_UNREGISTERING; | |
3765 | ||
3766 | synchronize_net(); | |
3767 | ||
3768 | /* Shutdown queueing discipline. */ | |
3769 | dev_shutdown(dev); | |
3770 | ||
3771 | ||
3772 | /* Notify protocols, that we are about to destroy | |
3773 | this device. They should clean all the things. | |
3774 | */ | |
3775 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
3776 | ||
3777 | /* | |
3778 | * Flush the unicast and multicast chains | |
3779 | */ | |
3780 | dev_addr_discard(dev); | |
3781 | ||
3782 | if (dev->uninit) | |
3783 | dev->uninit(dev); | |
3784 | ||
3785 | /* Notifier chain MUST detach us from master device. */ | |
3786 | BUG_TRAP(!dev->master); | |
3787 | ||
3788 | /* Remove entries from kobject tree */ | |
3789 | netdev_unregister_kobject(dev); | |
3790 | ||
3791 | synchronize_net(); | |
3792 | ||
3793 | dev_put(dev); | |
3794 | } | |
3795 | ||
3796 | static void __netdev_init_queue_locks_one(struct net_device *dev, | |
3797 | struct netdev_queue *dev_queue, | |
3798 | void *_unused) | |
3799 | { | |
3800 | spin_lock_init(&dev_queue->_xmit_lock); | |
3801 | netdev_set_lockdep_class(&dev_queue->_xmit_lock, dev->type); | |
3802 | dev_queue->xmit_lock_owner = -1; | |
3803 | } | |
3804 | ||
3805 | static void netdev_init_queue_locks(struct net_device *dev) | |
3806 | { | |
3807 | netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL); | |
3808 | __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL); | |
3809 | } | |
3810 | ||
3811 | /** | |
3812 | * register_netdevice - register a network device | |
3813 | * @dev: device to register | |
3814 | * | |
3815 | * Take a completed network device structure and add it to the kernel | |
3816 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
3817 | * chain. 0 is returned on success. A negative errno code is returned | |
3818 | * on a failure to set up the device, or if the name is a duplicate. | |
3819 | * | |
3820 | * Callers must hold the rtnl semaphore. You may want | |
3821 | * register_netdev() instead of this. | |
3822 | * | |
3823 | * BUGS: | |
3824 | * The locking appears insufficient to guarantee two parallel registers | |
3825 | * will not get the same name. | |
3826 | */ | |
3827 | ||
3828 | int register_netdevice(struct net_device *dev) | |
3829 | { | |
3830 | struct hlist_head *head; | |
3831 | struct hlist_node *p; | |
3832 | int ret; | |
3833 | struct net *net; | |
3834 | ||
3835 | BUG_ON(dev_boot_phase); | |
3836 | ASSERT_RTNL(); | |
3837 | ||
3838 | might_sleep(); | |
3839 | ||
3840 | /* When net_device's are persistent, this will be fatal. */ | |
3841 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
3842 | BUG_ON(!dev_net(dev)); | |
3843 | net = dev_net(dev); | |
3844 | ||
3845 | spin_lock_init(&dev->addr_list_lock); | |
3846 | netdev_init_queue_locks(dev); | |
3847 | ||
3848 | dev->iflink = -1; | |
3849 | ||
3850 | /* Init, if this function is available */ | |
3851 | if (dev->init) { | |
3852 | ret = dev->init(dev); | |
3853 | if (ret) { | |
3854 | if (ret > 0) | |
3855 | ret = -EIO; | |
3856 | goto out; | |
3857 | } | |
3858 | } | |
3859 | ||
3860 | if (!dev_valid_name(dev->name)) { | |
3861 | ret = -EINVAL; | |
3862 | goto err_uninit; | |
3863 | } | |
3864 | ||
3865 | dev->ifindex = dev_new_index(net); | |
3866 | if (dev->iflink == -1) | |
3867 | dev->iflink = dev->ifindex; | |
3868 | ||
3869 | /* Check for existence of name */ | |
3870 | head = dev_name_hash(net, dev->name); | |
3871 | hlist_for_each(p, head) { | |
3872 | struct net_device *d | |
3873 | = hlist_entry(p, struct net_device, name_hlist); | |
3874 | if (!strncmp(d->name, dev->name, IFNAMSIZ)) { | |
3875 | ret = -EEXIST; | |
3876 | goto err_uninit; | |
3877 | } | |
3878 | } | |
3879 | ||
3880 | /* Fix illegal checksum combinations */ | |
3881 | if ((dev->features & NETIF_F_HW_CSUM) && | |
3882 | (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
3883 | printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n", | |
3884 | dev->name); | |
3885 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); | |
3886 | } | |
3887 | ||
3888 | if ((dev->features & NETIF_F_NO_CSUM) && | |
3889 | (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
3890 | printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n", | |
3891 | dev->name); | |
3892 | dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM); | |
3893 | } | |
3894 | ||
3895 | ||
3896 | /* Fix illegal SG+CSUM combinations. */ | |
3897 | if ((dev->features & NETIF_F_SG) && | |
3898 | !(dev->features & NETIF_F_ALL_CSUM)) { | |
3899 | printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n", | |
3900 | dev->name); | |
3901 | dev->features &= ~NETIF_F_SG; | |
3902 | } | |
3903 | ||
3904 | /* TSO requires that SG is present as well. */ | |
3905 | if ((dev->features & NETIF_F_TSO) && | |
3906 | !(dev->features & NETIF_F_SG)) { | |
3907 | printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n", | |
3908 | dev->name); | |
3909 | dev->features &= ~NETIF_F_TSO; | |
3910 | } | |
3911 | if (dev->features & NETIF_F_UFO) { | |
3912 | if (!(dev->features & NETIF_F_HW_CSUM)) { | |
3913 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " | |
3914 | "NETIF_F_HW_CSUM feature.\n", | |
3915 | dev->name); | |
3916 | dev->features &= ~NETIF_F_UFO; | |
3917 | } | |
3918 | if (!(dev->features & NETIF_F_SG)) { | |
3919 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " | |
3920 | "NETIF_F_SG feature.\n", | |
3921 | dev->name); | |
3922 | dev->features &= ~NETIF_F_UFO; | |
3923 | } | |
3924 | } | |
3925 | ||
3926 | netdev_initialize_kobject(dev); | |
3927 | ret = netdev_register_kobject(dev); | |
3928 | if (ret) | |
3929 | goto err_uninit; | |
3930 | dev->reg_state = NETREG_REGISTERED; | |
3931 | ||
3932 | /* | |
3933 | * Default initial state at registry is that the | |
3934 | * device is present. | |
3935 | */ | |
3936 | ||
3937 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
3938 | ||
3939 | dev_init_scheduler(dev); | |
3940 | dev_hold(dev); | |
3941 | list_netdevice(dev); | |
3942 | ||
3943 | /* Notify protocols, that a new device appeared. */ | |
3944 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
3945 | ret = notifier_to_errno(ret); | |
3946 | if (ret) { | |
3947 | rollback_registered(dev); | |
3948 | dev->reg_state = NETREG_UNREGISTERED; | |
3949 | } | |
3950 | ||
3951 | out: | |
3952 | return ret; | |
3953 | ||
3954 | err_uninit: | |
3955 | if (dev->uninit) | |
3956 | dev->uninit(dev); | |
3957 | goto out; | |
3958 | } | |
3959 | ||
3960 | /** | |
3961 | * register_netdev - register a network device | |
3962 | * @dev: device to register | |
3963 | * | |
3964 | * Take a completed network device structure and add it to the kernel | |
3965 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
3966 | * chain. 0 is returned on success. A negative errno code is returned | |
3967 | * on a failure to set up the device, or if the name is a duplicate. | |
3968 | * | |
3969 | * This is a wrapper around register_netdevice that takes the rtnl semaphore | |
3970 | * and expands the device name if you passed a format string to | |
3971 | * alloc_netdev. | |
3972 | */ | |
3973 | int register_netdev(struct net_device *dev) | |
3974 | { | |
3975 | int err; | |
3976 | ||
3977 | rtnl_lock(); | |
3978 | ||
3979 | /* | |
3980 | * If the name is a format string the caller wants us to do a | |
3981 | * name allocation. | |
3982 | */ | |
3983 | if (strchr(dev->name, '%')) { | |
3984 | err = dev_alloc_name(dev, dev->name); | |
3985 | if (err < 0) | |
3986 | goto out; | |
3987 | } | |
3988 | ||
3989 | err = register_netdevice(dev); | |
3990 | out: | |
3991 | rtnl_unlock(); | |
3992 | return err; | |
3993 | } | |
3994 | EXPORT_SYMBOL(register_netdev); | |
3995 | ||
3996 | /* | |
3997 | * netdev_wait_allrefs - wait until all references are gone. | |
3998 | * | |
3999 | * This is called when unregistering network devices. | |
4000 | * | |
4001 | * Any protocol or device that holds a reference should register | |
4002 | * for netdevice notification, and cleanup and put back the | |
4003 | * reference if they receive an UNREGISTER event. | |
4004 | * We can get stuck here if buggy protocols don't correctly | |
4005 | * call dev_put. | |
4006 | */ | |
4007 | static void netdev_wait_allrefs(struct net_device *dev) | |
4008 | { | |
4009 | unsigned long rebroadcast_time, warning_time; | |
4010 | ||
4011 | rebroadcast_time = warning_time = jiffies; | |
4012 | while (atomic_read(&dev->refcnt) != 0) { | |
4013 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { | |
4014 | rtnl_lock(); | |
4015 | ||
4016 | /* Rebroadcast unregister notification */ | |
4017 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
4018 | ||
4019 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, | |
4020 | &dev->state)) { | |
4021 | /* We must not have linkwatch events | |
4022 | * pending on unregister. If this | |
4023 | * happens, we simply run the queue | |
4024 | * unscheduled, resulting in a noop | |
4025 | * for this device. | |
4026 | */ | |
4027 | linkwatch_run_queue(); | |
4028 | } | |
4029 | ||
4030 | __rtnl_unlock(); | |
4031 | ||
4032 | rebroadcast_time = jiffies; | |
4033 | } | |
4034 | ||
4035 | msleep(250); | |
4036 | ||
4037 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
4038 | printk(KERN_EMERG "unregister_netdevice: " | |
4039 | "waiting for %s to become free. Usage " | |
4040 | "count = %d\n", | |
4041 | dev->name, atomic_read(&dev->refcnt)); | |
4042 | warning_time = jiffies; | |
4043 | } | |
4044 | } | |
4045 | } | |
4046 | ||
4047 | /* The sequence is: | |
4048 | * | |
4049 | * rtnl_lock(); | |
4050 | * ... | |
4051 | * register_netdevice(x1); | |
4052 | * register_netdevice(x2); | |
4053 | * ... | |
4054 | * unregister_netdevice(y1); | |
4055 | * unregister_netdevice(y2); | |
4056 | * ... | |
4057 | * rtnl_unlock(); | |
4058 | * free_netdev(y1); | |
4059 | * free_netdev(y2); | |
4060 | * | |
4061 | * We are invoked by rtnl_unlock() after it drops the semaphore. | |
4062 | * This allows us to deal with problems: | |
4063 | * 1) We can delete sysfs objects which invoke hotplug | |
4064 | * without deadlocking with linkwatch via keventd. | |
4065 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
4066 | * safely in order to wait for the netdev refcnt to drop to zero. | |
4067 | */ | |
4068 | static DEFINE_MUTEX(net_todo_run_mutex); | |
4069 | void netdev_run_todo(void) | |
4070 | { | |
4071 | struct list_head list; | |
4072 | ||
4073 | /* Need to guard against multiple cpu's getting out of order. */ | |
4074 | mutex_lock(&net_todo_run_mutex); | |
4075 | ||
4076 | /* Not safe to do outside the semaphore. We must not return | |
4077 | * until all unregister events invoked by the local processor | |
4078 | * have been completed (either by this todo run, or one on | |
4079 | * another cpu). | |
4080 | */ | |
4081 | if (list_empty(&net_todo_list)) | |
4082 | goto out; | |
4083 | ||
4084 | /* Snapshot list, allow later requests */ | |
4085 | spin_lock(&net_todo_list_lock); | |
4086 | list_replace_init(&net_todo_list, &list); | |
4087 | spin_unlock(&net_todo_list_lock); | |
4088 | ||
4089 | while (!list_empty(&list)) { | |
4090 | struct net_device *dev | |
4091 | = list_entry(list.next, struct net_device, todo_list); | |
4092 | list_del(&dev->todo_list); | |
4093 | ||
4094 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { | |
4095 | printk(KERN_ERR "network todo '%s' but state %d\n", | |
4096 | dev->name, dev->reg_state); | |
4097 | dump_stack(); | |
4098 | continue; | |
4099 | } | |
4100 | ||
4101 | dev->reg_state = NETREG_UNREGISTERED; | |
4102 | ||
4103 | netdev_wait_allrefs(dev); | |
4104 | ||
4105 | /* paranoia */ | |
4106 | BUG_ON(atomic_read(&dev->refcnt)); | |
4107 | BUG_TRAP(!dev->ip_ptr); | |
4108 | BUG_TRAP(!dev->ip6_ptr); | |
4109 | BUG_TRAP(!dev->dn_ptr); | |
4110 | ||
4111 | if (dev->destructor) | |
4112 | dev->destructor(dev); | |
4113 | ||
4114 | /* Free network device */ | |
4115 | kobject_put(&dev->dev.kobj); | |
4116 | } | |
4117 | ||
4118 | out: | |
4119 | mutex_unlock(&net_todo_run_mutex); | |
4120 | } | |
4121 | ||
4122 | static struct net_device_stats *internal_stats(struct net_device *dev) | |
4123 | { | |
4124 | return &dev->stats; | |
4125 | } | |
4126 | ||
4127 | static void netdev_init_one_queue(struct net_device *dev, | |
4128 | struct netdev_queue *queue, | |
4129 | void *_unused) | |
4130 | { | |
4131 | spin_lock_init(&queue->lock); | |
4132 | queue->dev = dev; | |
4133 | } | |
4134 | ||
4135 | static void netdev_init_queues(struct net_device *dev) | |
4136 | { | |
4137 | netdev_init_one_queue(dev, &dev->rx_queue, NULL); | |
4138 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); | |
4139 | } | |
4140 | ||
4141 | /** | |
4142 | * alloc_netdev_mq - allocate network device | |
4143 | * @sizeof_priv: size of private data to allocate space for | |
4144 | * @name: device name format string | |
4145 | * @setup: callback to initialize device | |
4146 | * @queue_count: the number of subqueues to allocate | |
4147 | * | |
4148 | * Allocates a struct net_device with private data area for driver use | |
4149 | * and performs basic initialization. Also allocates subquue structs | |
4150 | * for each queue on the device at the end of the netdevice. | |
4151 | */ | |
4152 | struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name, | |
4153 | void (*setup)(struct net_device *), unsigned int queue_count) | |
4154 | { | |
4155 | struct netdev_queue *tx; | |
4156 | struct net_device *dev; | |
4157 | int alloc_size; | |
4158 | void *p; | |
4159 | ||
4160 | BUG_ON(strlen(name) >= sizeof(dev->name)); | |
4161 | ||
4162 | alloc_size = sizeof(struct net_device); | |
4163 | if (sizeof_priv) { | |
4164 | /* ensure 32-byte alignment of private area */ | |
4165 | alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST; | |
4166 | alloc_size += sizeof_priv; | |
4167 | } | |
4168 | /* ensure 32-byte alignment of whole construct */ | |
4169 | alloc_size += NETDEV_ALIGN_CONST; | |
4170 | ||
4171 | p = kzalloc(alloc_size, GFP_KERNEL); | |
4172 | if (!p) { | |
4173 | printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n"); | |
4174 | return NULL; | |
4175 | } | |
4176 | ||
4177 | tx = kzalloc(sizeof(struct netdev_queue) * queue_count, GFP_KERNEL); | |
4178 | if (!tx) { | |
4179 | printk(KERN_ERR "alloc_netdev: Unable to allocate " | |
4180 | "tx qdiscs.\n"); | |
4181 | kfree(p); | |
4182 | return NULL; | |
4183 | } | |
4184 | ||
4185 | dev = (struct net_device *) | |
4186 | (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST); | |
4187 | dev->padded = (char *)dev - (char *)p; | |
4188 | dev_net_set(dev, &init_net); | |
4189 | ||
4190 | dev->_tx = tx; | |
4191 | dev->num_tx_queues = queue_count; | |
4192 | dev->real_num_tx_queues = queue_count; | |
4193 | ||
4194 | if (sizeof_priv) { | |
4195 | dev->priv = ((char *)dev + | |
4196 | ((sizeof(struct net_device) + NETDEV_ALIGN_CONST) | |
4197 | & ~NETDEV_ALIGN_CONST)); | |
4198 | } | |
4199 | ||
4200 | dev->gso_max_size = GSO_MAX_SIZE; | |
4201 | ||
4202 | netdev_init_queues(dev); | |
4203 | ||
4204 | dev->get_stats = internal_stats; | |
4205 | netpoll_netdev_init(dev); | |
4206 | setup(dev); | |
4207 | strcpy(dev->name, name); | |
4208 | return dev; | |
4209 | } | |
4210 | EXPORT_SYMBOL(alloc_netdev_mq); | |
4211 | ||
4212 | /** | |
4213 | * free_netdev - free network device | |
4214 | * @dev: device | |
4215 | * | |
4216 | * This function does the last stage of destroying an allocated device | |
4217 | * interface. The reference to the device object is released. | |
4218 | * If this is the last reference then it will be freed. | |
4219 | */ | |
4220 | void free_netdev(struct net_device *dev) | |
4221 | { | |
4222 | release_net(dev_net(dev)); | |
4223 | ||
4224 | kfree(dev->_tx); | |
4225 | ||
4226 | /* Compatibility with error handling in drivers */ | |
4227 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
4228 | kfree((char *)dev - dev->padded); | |
4229 | return; | |
4230 | } | |
4231 | ||
4232 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
4233 | dev->reg_state = NETREG_RELEASED; | |
4234 | ||
4235 | /* will free via device release */ | |
4236 | put_device(&dev->dev); | |
4237 | } | |
4238 | ||
4239 | /* Synchronize with packet receive processing. */ | |
4240 | void synchronize_net(void) | |
4241 | { | |
4242 | might_sleep(); | |
4243 | synchronize_rcu(); | |
4244 | } | |
4245 | ||
4246 | /** | |
4247 | * unregister_netdevice - remove device from the kernel | |
4248 | * @dev: device | |
4249 | * | |
4250 | * This function shuts down a device interface and removes it | |
4251 | * from the kernel tables. | |
4252 | * | |
4253 | * Callers must hold the rtnl semaphore. You may want | |
4254 | * unregister_netdev() instead of this. | |
4255 | */ | |
4256 | ||
4257 | void unregister_netdevice(struct net_device *dev) | |
4258 | { | |
4259 | ASSERT_RTNL(); | |
4260 | ||
4261 | rollback_registered(dev); | |
4262 | /* Finish processing unregister after unlock */ | |
4263 | net_set_todo(dev); | |
4264 | } | |
4265 | ||
4266 | /** | |
4267 | * unregister_netdev - remove device from the kernel | |
4268 | * @dev: device | |
4269 | * | |
4270 | * This function shuts down a device interface and removes it | |
4271 | * from the kernel tables. | |
4272 | * | |
4273 | * This is just a wrapper for unregister_netdevice that takes | |
4274 | * the rtnl semaphore. In general you want to use this and not | |
4275 | * unregister_netdevice. | |
4276 | */ | |
4277 | void unregister_netdev(struct net_device *dev) | |
4278 | { | |
4279 | rtnl_lock(); | |
4280 | unregister_netdevice(dev); | |
4281 | rtnl_unlock(); | |
4282 | } | |
4283 | ||
4284 | EXPORT_SYMBOL(unregister_netdev); | |
4285 | ||
4286 | /** | |
4287 | * dev_change_net_namespace - move device to different nethost namespace | |
4288 | * @dev: device | |
4289 | * @net: network namespace | |
4290 | * @pat: If not NULL name pattern to try if the current device name | |
4291 | * is already taken in the destination network namespace. | |
4292 | * | |
4293 | * This function shuts down a device interface and moves it | |
4294 | * to a new network namespace. On success 0 is returned, on | |
4295 | * a failure a netagive errno code is returned. | |
4296 | * | |
4297 | * Callers must hold the rtnl semaphore. | |
4298 | */ | |
4299 | ||
4300 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
4301 | { | |
4302 | char buf[IFNAMSIZ]; | |
4303 | const char *destname; | |
4304 | int err; | |
4305 | ||
4306 | ASSERT_RTNL(); | |
4307 | ||
4308 | /* Don't allow namespace local devices to be moved. */ | |
4309 | err = -EINVAL; | |
4310 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
4311 | goto out; | |
4312 | ||
4313 | /* Ensure the device has been registrered */ | |
4314 | err = -EINVAL; | |
4315 | if (dev->reg_state != NETREG_REGISTERED) | |
4316 | goto out; | |
4317 | ||
4318 | /* Get out if there is nothing todo */ | |
4319 | err = 0; | |
4320 | if (net_eq(dev_net(dev), net)) | |
4321 | goto out; | |
4322 | ||
4323 | /* Pick the destination device name, and ensure | |
4324 | * we can use it in the destination network namespace. | |
4325 | */ | |
4326 | err = -EEXIST; | |
4327 | destname = dev->name; | |
4328 | if (__dev_get_by_name(net, destname)) { | |
4329 | /* We get here if we can't use the current device name */ | |
4330 | if (!pat) | |
4331 | goto out; | |
4332 | if (!dev_valid_name(pat)) | |
4333 | goto out; | |
4334 | if (strchr(pat, '%')) { | |
4335 | if (__dev_alloc_name(net, pat, buf) < 0) | |
4336 | goto out; | |
4337 | destname = buf; | |
4338 | } else | |
4339 | destname = pat; | |
4340 | if (__dev_get_by_name(net, destname)) | |
4341 | goto out; | |
4342 | } | |
4343 | ||
4344 | /* | |
4345 | * And now a mini version of register_netdevice unregister_netdevice. | |
4346 | */ | |
4347 | ||
4348 | /* If device is running close it first. */ | |
4349 | dev_close(dev); | |
4350 | ||
4351 | /* And unlink it from device chain */ | |
4352 | err = -ENODEV; | |
4353 | unlist_netdevice(dev); | |
4354 | ||
4355 | synchronize_net(); | |
4356 | ||
4357 | /* Shutdown queueing discipline. */ | |
4358 | dev_shutdown(dev); | |
4359 | ||
4360 | /* Notify protocols, that we are about to destroy | |
4361 | this device. They should clean all the things. | |
4362 | */ | |
4363 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); | |
4364 | ||
4365 | /* | |
4366 | * Flush the unicast and multicast chains | |
4367 | */ | |
4368 | dev_addr_discard(dev); | |
4369 | ||
4370 | /* Actually switch the network namespace */ | |
4371 | dev_net_set(dev, net); | |
4372 | ||
4373 | /* Assign the new device name */ | |
4374 | if (destname != dev->name) | |
4375 | strcpy(dev->name, destname); | |
4376 | ||
4377 | /* If there is an ifindex conflict assign a new one */ | |
4378 | if (__dev_get_by_index(net, dev->ifindex)) { | |
4379 | int iflink = (dev->iflink == dev->ifindex); | |
4380 | dev->ifindex = dev_new_index(net); | |
4381 | if (iflink) | |
4382 | dev->iflink = dev->ifindex; | |
4383 | } | |
4384 | ||
4385 | /* Fixup kobjects */ | |
4386 | netdev_unregister_kobject(dev); | |
4387 | err = netdev_register_kobject(dev); | |
4388 | WARN_ON(err); | |
4389 | ||
4390 | /* Add the device back in the hashes */ | |
4391 | list_netdevice(dev); | |
4392 | ||
4393 | /* Notify protocols, that a new device appeared. */ | |
4394 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
4395 | ||
4396 | synchronize_net(); | |
4397 | err = 0; | |
4398 | out: | |
4399 | return err; | |
4400 | } | |
4401 | ||
4402 | static int dev_cpu_callback(struct notifier_block *nfb, | |
4403 | unsigned long action, | |
4404 | void *ocpu) | |
4405 | { | |
4406 | struct sk_buff **list_skb; | |
4407 | struct netdev_queue **list_net; | |
4408 | struct sk_buff *skb; | |
4409 | unsigned int cpu, oldcpu = (unsigned long)ocpu; | |
4410 | struct softnet_data *sd, *oldsd; | |
4411 | ||
4412 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) | |
4413 | return NOTIFY_OK; | |
4414 | ||
4415 | local_irq_disable(); | |
4416 | cpu = smp_processor_id(); | |
4417 | sd = &per_cpu(softnet_data, cpu); | |
4418 | oldsd = &per_cpu(softnet_data, oldcpu); | |
4419 | ||
4420 | /* Find end of our completion_queue. */ | |
4421 | list_skb = &sd->completion_queue; | |
4422 | while (*list_skb) | |
4423 | list_skb = &(*list_skb)->next; | |
4424 | /* Append completion queue from offline CPU. */ | |
4425 | *list_skb = oldsd->completion_queue; | |
4426 | oldsd->completion_queue = NULL; | |
4427 | ||
4428 | /* Find end of our output_queue. */ | |
4429 | list_net = &sd->output_queue; | |
4430 | while (*list_net) | |
4431 | list_net = &(*list_net)->next_sched; | |
4432 | /* Append output queue from offline CPU. */ | |
4433 | *list_net = oldsd->output_queue; | |
4434 | oldsd->output_queue = NULL; | |
4435 | ||
4436 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
4437 | local_irq_enable(); | |
4438 | ||
4439 | /* Process offline CPU's input_pkt_queue */ | |
4440 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) | |
4441 | netif_rx(skb); | |
4442 | ||
4443 | return NOTIFY_OK; | |
4444 | } | |
4445 | ||
4446 | #ifdef CONFIG_NET_DMA | |
4447 | /** | |
4448 | * net_dma_rebalance - try to maintain one DMA channel per CPU | |
4449 | * @net_dma: DMA client and associated data (lock, channels, channel_mask) | |
4450 | * | |
4451 | * This is called when the number of channels allocated to the net_dma client | |
4452 | * changes. The net_dma client tries to have one DMA channel per CPU. | |
4453 | */ | |
4454 | ||
4455 | static void net_dma_rebalance(struct net_dma *net_dma) | |
4456 | { | |
4457 | unsigned int cpu, i, n, chan_idx; | |
4458 | struct dma_chan *chan; | |
4459 | ||
4460 | if (cpus_empty(net_dma->channel_mask)) { | |
4461 | for_each_online_cpu(cpu) | |
4462 | rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL); | |
4463 | return; | |
4464 | } | |
4465 | ||
4466 | i = 0; | |
4467 | cpu = first_cpu(cpu_online_map); | |
4468 | ||
4469 | for_each_cpu_mask(chan_idx, net_dma->channel_mask) { | |
4470 | chan = net_dma->channels[chan_idx]; | |
4471 | ||
4472 | n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask)) | |
4473 | + (i < (num_online_cpus() % | |
4474 | cpus_weight(net_dma->channel_mask)) ? 1 : 0)); | |
4475 | ||
4476 | while(n) { | |
4477 | per_cpu(softnet_data, cpu).net_dma = chan; | |
4478 | cpu = next_cpu(cpu, cpu_online_map); | |
4479 | n--; | |
4480 | } | |
4481 | i++; | |
4482 | } | |
4483 | } | |
4484 | ||
4485 | /** | |
4486 | * netdev_dma_event - event callback for the net_dma_client | |
4487 | * @client: should always be net_dma_client | |
4488 | * @chan: DMA channel for the event | |
4489 | * @state: DMA state to be handled | |
4490 | */ | |
4491 | static enum dma_state_client | |
4492 | netdev_dma_event(struct dma_client *client, struct dma_chan *chan, | |
4493 | enum dma_state state) | |
4494 | { | |
4495 | int i, found = 0, pos = -1; | |
4496 | struct net_dma *net_dma = | |
4497 | container_of(client, struct net_dma, client); | |
4498 | enum dma_state_client ack = DMA_DUP; /* default: take no action */ | |
4499 | ||
4500 | spin_lock(&net_dma->lock); | |
4501 | switch (state) { | |
4502 | case DMA_RESOURCE_AVAILABLE: | |
4503 | for (i = 0; i < nr_cpu_ids; i++) | |
4504 | if (net_dma->channels[i] == chan) { | |
4505 | found = 1; | |
4506 | break; | |
4507 | } else if (net_dma->channels[i] == NULL && pos < 0) | |
4508 | pos = i; | |
4509 | ||
4510 | if (!found && pos >= 0) { | |
4511 | ack = DMA_ACK; | |
4512 | net_dma->channels[pos] = chan; | |
4513 | cpu_set(pos, net_dma->channel_mask); | |
4514 | net_dma_rebalance(net_dma); | |
4515 | } | |
4516 | break; | |
4517 | case DMA_RESOURCE_REMOVED: | |
4518 | for (i = 0; i < nr_cpu_ids; i++) | |
4519 | if (net_dma->channels[i] == chan) { | |
4520 | found = 1; | |
4521 | pos = i; | |
4522 | break; | |
4523 | } | |
4524 | ||
4525 | if (found) { | |
4526 | ack = DMA_ACK; | |
4527 | cpu_clear(pos, net_dma->channel_mask); | |
4528 | net_dma->channels[i] = NULL; | |
4529 | net_dma_rebalance(net_dma); | |
4530 | } | |
4531 | break; | |
4532 | default: | |
4533 | break; | |
4534 | } | |
4535 | spin_unlock(&net_dma->lock); | |
4536 | ||
4537 | return ack; | |
4538 | } | |
4539 | ||
4540 | /** | |
4541 | * netdev_dma_regiser - register the networking subsystem as a DMA client | |
4542 | */ | |
4543 | static int __init netdev_dma_register(void) | |
4544 | { | |
4545 | net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma), | |
4546 | GFP_KERNEL); | |
4547 | if (unlikely(!net_dma.channels)) { | |
4548 | printk(KERN_NOTICE | |
4549 | "netdev_dma: no memory for net_dma.channels\n"); | |
4550 | return -ENOMEM; | |
4551 | } | |
4552 | spin_lock_init(&net_dma.lock); | |
4553 | dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask); | |
4554 | dma_async_client_register(&net_dma.client); | |
4555 | dma_async_client_chan_request(&net_dma.client); | |
4556 | return 0; | |
4557 | } | |
4558 | ||
4559 | #else | |
4560 | static int __init netdev_dma_register(void) { return -ENODEV; } | |
4561 | #endif /* CONFIG_NET_DMA */ | |
4562 | ||
4563 | /** | |
4564 | * netdev_compute_feature - compute conjunction of two feature sets | |
4565 | * @all: first feature set | |
4566 | * @one: second feature set | |
4567 | * | |
4568 | * Computes a new feature set after adding a device with feature set | |
4569 | * @one to the master device with current feature set @all. Returns | |
4570 | * the new feature set. | |
4571 | */ | |
4572 | int netdev_compute_features(unsigned long all, unsigned long one) | |
4573 | { | |
4574 | /* if device needs checksumming, downgrade to hw checksumming */ | |
4575 | if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM)) | |
4576 | all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM; | |
4577 | ||
4578 | /* if device can't do all checksum, downgrade to ipv4/ipv6 */ | |
4579 | if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM)) | |
4580 | all ^= NETIF_F_HW_CSUM | |
4581 | | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; | |
4582 | ||
4583 | if (one & NETIF_F_GSO) | |
4584 | one |= NETIF_F_GSO_SOFTWARE; | |
4585 | one |= NETIF_F_GSO; | |
4586 | ||
4587 | /* If even one device supports robust GSO, enable it for all. */ | |
4588 | if (one & NETIF_F_GSO_ROBUST) | |
4589 | all |= NETIF_F_GSO_ROBUST; | |
4590 | ||
4591 | all &= one | NETIF_F_LLTX; | |
4592 | ||
4593 | if (!(all & NETIF_F_ALL_CSUM)) | |
4594 | all &= ~NETIF_F_SG; | |
4595 | if (!(all & NETIF_F_SG)) | |
4596 | all &= ~NETIF_F_GSO_MASK; | |
4597 | ||
4598 | return all; | |
4599 | } | |
4600 | EXPORT_SYMBOL(netdev_compute_features); | |
4601 | ||
4602 | static struct hlist_head *netdev_create_hash(void) | |
4603 | { | |
4604 | int i; | |
4605 | struct hlist_head *hash; | |
4606 | ||
4607 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
4608 | if (hash != NULL) | |
4609 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
4610 | INIT_HLIST_HEAD(&hash[i]); | |
4611 | ||
4612 | return hash; | |
4613 | } | |
4614 | ||
4615 | /* Initialize per network namespace state */ | |
4616 | static int __net_init netdev_init(struct net *net) | |
4617 | { | |
4618 | INIT_LIST_HEAD(&net->dev_base_head); | |
4619 | ||
4620 | net->dev_name_head = netdev_create_hash(); | |
4621 | if (net->dev_name_head == NULL) | |
4622 | goto err_name; | |
4623 | ||
4624 | net->dev_index_head = netdev_create_hash(); | |
4625 | if (net->dev_index_head == NULL) | |
4626 | goto err_idx; | |
4627 | ||
4628 | return 0; | |
4629 | ||
4630 | err_idx: | |
4631 | kfree(net->dev_name_head); | |
4632 | err_name: | |
4633 | return -ENOMEM; | |
4634 | } | |
4635 | ||
4636 | static void __net_exit netdev_exit(struct net *net) | |
4637 | { | |
4638 | kfree(net->dev_name_head); | |
4639 | kfree(net->dev_index_head); | |
4640 | } | |
4641 | ||
4642 | static struct pernet_operations __net_initdata netdev_net_ops = { | |
4643 | .init = netdev_init, | |
4644 | .exit = netdev_exit, | |
4645 | }; | |
4646 | ||
4647 | static void __net_exit default_device_exit(struct net *net) | |
4648 | { | |
4649 | struct net_device *dev, *next; | |
4650 | /* | |
4651 | * Push all migratable of the network devices back to the | |
4652 | * initial network namespace | |
4653 | */ | |
4654 | rtnl_lock(); | |
4655 | for_each_netdev_safe(net, dev, next) { | |
4656 | int err; | |
4657 | char fb_name[IFNAMSIZ]; | |
4658 | ||
4659 | /* Ignore unmoveable devices (i.e. loopback) */ | |
4660 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
4661 | continue; | |
4662 | ||
4663 | /* Push remaing network devices to init_net */ | |
4664 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); | |
4665 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
4666 | if (err) { | |
4667 | printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n", | |
4668 | __func__, dev->name, err); | |
4669 | BUG(); | |
4670 | } | |
4671 | } | |
4672 | rtnl_unlock(); | |
4673 | } | |
4674 | ||
4675 | static struct pernet_operations __net_initdata default_device_ops = { | |
4676 | .exit = default_device_exit, | |
4677 | }; | |
4678 | ||
4679 | /* | |
4680 | * Initialize the DEV module. At boot time this walks the device list and | |
4681 | * unhooks any devices that fail to initialise (normally hardware not | |
4682 | * present) and leaves us with a valid list of present and active devices. | |
4683 | * | |
4684 | */ | |
4685 | ||
4686 | /* | |
4687 | * This is called single threaded during boot, so no need | |
4688 | * to take the rtnl semaphore. | |
4689 | */ | |
4690 | static int __init net_dev_init(void) | |
4691 | { | |
4692 | int i, rc = -ENOMEM; | |
4693 | ||
4694 | BUG_ON(!dev_boot_phase); | |
4695 | ||
4696 | if (dev_proc_init()) | |
4697 | goto out; | |
4698 | ||
4699 | if (netdev_kobject_init()) | |
4700 | goto out; | |
4701 | ||
4702 | INIT_LIST_HEAD(&ptype_all); | |
4703 | for (i = 0; i < PTYPE_HASH_SIZE; i++) | |
4704 | INIT_LIST_HEAD(&ptype_base[i]); | |
4705 | ||
4706 | if (register_pernet_subsys(&netdev_net_ops)) | |
4707 | goto out; | |
4708 | ||
4709 | if (register_pernet_device(&default_device_ops)) | |
4710 | goto out; | |
4711 | ||
4712 | /* | |
4713 | * Initialise the packet receive queues. | |
4714 | */ | |
4715 | ||
4716 | for_each_possible_cpu(i) { | |
4717 | struct softnet_data *queue; | |
4718 | ||
4719 | queue = &per_cpu(softnet_data, i); | |
4720 | skb_queue_head_init(&queue->input_pkt_queue); | |
4721 | queue->completion_queue = NULL; | |
4722 | INIT_LIST_HEAD(&queue->poll_list); | |
4723 | ||
4724 | queue->backlog.poll = process_backlog; | |
4725 | queue->backlog.weight = weight_p; | |
4726 | } | |
4727 | ||
4728 | netdev_dma_register(); | |
4729 | ||
4730 | dev_boot_phase = 0; | |
4731 | ||
4732 | open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL); | |
4733 | open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL); | |
4734 | ||
4735 | hotcpu_notifier(dev_cpu_callback, 0); | |
4736 | dst_init(); | |
4737 | dev_mcast_init(); | |
4738 | rc = 0; | |
4739 | out: | |
4740 | return rc; | |
4741 | } | |
4742 | ||
4743 | subsys_initcall(net_dev_init); | |
4744 | ||
4745 | EXPORT_SYMBOL(__dev_get_by_index); | |
4746 | EXPORT_SYMBOL(__dev_get_by_name); | |
4747 | EXPORT_SYMBOL(__dev_remove_pack); | |
4748 | EXPORT_SYMBOL(dev_valid_name); | |
4749 | EXPORT_SYMBOL(dev_add_pack); | |
4750 | EXPORT_SYMBOL(dev_alloc_name); | |
4751 | EXPORT_SYMBOL(dev_close); | |
4752 | EXPORT_SYMBOL(dev_get_by_flags); | |
4753 | EXPORT_SYMBOL(dev_get_by_index); | |
4754 | EXPORT_SYMBOL(dev_get_by_name); | |
4755 | EXPORT_SYMBOL(dev_open); | |
4756 | EXPORT_SYMBOL(dev_queue_xmit); | |
4757 | EXPORT_SYMBOL(dev_remove_pack); | |
4758 | EXPORT_SYMBOL(dev_set_allmulti); | |
4759 | EXPORT_SYMBOL(dev_set_promiscuity); | |
4760 | EXPORT_SYMBOL(dev_change_flags); | |
4761 | EXPORT_SYMBOL(dev_set_mtu); | |
4762 | EXPORT_SYMBOL(dev_set_mac_address); | |
4763 | EXPORT_SYMBOL(free_netdev); | |
4764 | EXPORT_SYMBOL(netdev_boot_setup_check); | |
4765 | EXPORT_SYMBOL(netdev_set_master); | |
4766 | EXPORT_SYMBOL(netdev_state_change); | |
4767 | EXPORT_SYMBOL(netif_receive_skb); | |
4768 | EXPORT_SYMBOL(netif_rx); | |
4769 | EXPORT_SYMBOL(register_gifconf); | |
4770 | EXPORT_SYMBOL(register_netdevice); | |
4771 | EXPORT_SYMBOL(register_netdevice_notifier); | |
4772 | EXPORT_SYMBOL(skb_checksum_help); | |
4773 | EXPORT_SYMBOL(synchronize_net); | |
4774 | EXPORT_SYMBOL(unregister_netdevice); | |
4775 | EXPORT_SYMBOL(unregister_netdevice_notifier); | |
4776 | EXPORT_SYMBOL(net_enable_timestamp); | |
4777 | EXPORT_SYMBOL(net_disable_timestamp); | |
4778 | EXPORT_SYMBOL(dev_get_flags); | |
4779 | ||
4780 | #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE) | |
4781 | EXPORT_SYMBOL(br_handle_frame_hook); | |
4782 | EXPORT_SYMBOL(br_fdb_get_hook); | |
4783 | EXPORT_SYMBOL(br_fdb_put_hook); | |
4784 | #endif | |
4785 | ||
4786 | #ifdef CONFIG_KMOD | |
4787 | EXPORT_SYMBOL(dev_load); | |
4788 | #endif | |
4789 | ||
4790 | EXPORT_PER_CPU_SYMBOL(softnet_data); |