blob: 4dc93cc4d5b79dc8fd26ddd21e77cb8e916ac1f8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * 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>
Randy Dunlap4fc268d2006-01-11 12:17:47 -080078#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#include <linux/cpu.h>
80#include <linux/types.h>
81#include <linux/kernel.h>
82#include <linux/sched.h>
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -080083#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070084#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/notifier.h>
94#include <linux/skbuff.h>
95#include <net/sock.h>
96#include <linux/rtnetlink.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/stat.h>
100#include <linux/if_bridge.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101#include <net/dst.h>
102#include <net/pkt_sched.h>
103#include <net/checksum.h>
104#include <linux/highmem.h>
105#include <linux/init.h>
106#include <linux/kmod.h>
107#include <linux/module.h>
108#include <linux/kallsyms.h>
109#include <linux/netpoll.h>
110#include <linux/rcupdate.h>
111#include <linux/delay.h>
Adrian Bunkd86b5e02006-01-21 00:46:55 +0100112#include <linux/wireless.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113#include <net/iw_handler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114#include <asm/current.h>
Steve Grubb5bdb9882005-12-03 08:39:35 -0500115#include <linux/audit.h>
Chris Leechdb217332006-06-17 21:24:58 -0700116#include <linux/dmaengine.h>
Herbert Xuf6a78bf2006-06-22 02:57:17 -0700117#include <linux/err.h>
David S. Millerc7fa9d12006-08-15 16:34:13 -0700118#include <linux/ctype.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120/*
121 * The list of packet types we will receive (as opposed to discard)
122 * and the routines to invoke.
123 *
124 * Why 16. Because with 16 the only overlap we get on a hash of the
125 * low nibble of the protocol value is RARP/SNAP/X.25.
126 *
127 * NOTE: That is no longer true with the addition of VLAN tags. Not
128 * sure which should go first, but I bet it won't make much
129 * difference if we are running VLANs. The good news is that
130 * this protocol won't be in the list unless compiled in, so
Stephen Hemminger3041a062006-05-26 13:25:24 -0700131 * the average user (w/out VLANs) will not be adversely affected.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 * --BLG
133 *
134 * 0800 IP
135 * 8100 802.1Q VLAN
136 * 0001 802.3
137 * 0002 AX.25
138 * 0004 802.2
139 * 8035 RARP
140 * 0005 SNAP
141 * 0805 X.25
142 * 0806 ARP
143 * 8137 IPX
144 * 0009 Localtalk
145 * 86DD IPv6
146 */
147
148static DEFINE_SPINLOCK(ptype_lock);
149static struct list_head ptype_base[16]; /* 16 way hashed list */
150static struct list_head ptype_all; /* Taps */
151
Chris Leechdb217332006-06-17 21:24:58 -0700152#ifdef CONFIG_NET_DMA
153static struct dma_client *net_dma_client;
154static unsigned int net_dma_count;
155static spinlock_t net_dma_event_lock;
156#endif
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158/*
Stephen Hemminger3041a062006-05-26 13:25:24 -0700159 * The @dev_base list is protected by @dev_base_lock and the rtnl
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 * semaphore.
161 *
162 * Pure readers hold dev_base_lock for reading.
163 *
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
168 *
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
172 *
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
175 * semaphore held.
176 */
177struct net_device *dev_base;
178static struct net_device **dev_tail = &dev_base;
179DEFINE_RWLOCK(dev_base_lock);
180
181EXPORT_SYMBOL(dev_base);
182EXPORT_SYMBOL(dev_base_lock);
183
184#define NETDEV_HASHBITS 8
185static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
186static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
187
188static inline struct hlist_head *dev_name_hash(const char *name)
189{
190 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
192}
193
194static inline struct hlist_head *dev_index_hash(int ifindex)
195{
196 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
197}
198
199/*
200 * Our notifier list
201 */
202
Alan Sternf07d5b92006-05-09 15:23:03 -0700203static RAW_NOTIFIER_HEAD(netdev_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
205/*
206 * Device drivers call our routines to queue packets here. We empty the
207 * queue in the local softnet handler.
208 */
Stephen Hemminger31aa02c2005-06-23 20:12:48 -0700209DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211#ifdef CONFIG_SYSFS
212extern int netdev_sysfs_init(void);
213extern int netdev_register_sysfs(struct net_device *);
214extern void netdev_unregister_sysfs(struct net_device *);
215#else
216#define netdev_sysfs_init() (0)
217#define netdev_register_sysfs(dev) (0)
218#define netdev_unregister_sysfs(dev) do { } while(0)
219#endif
220
221
222/*******************************************************************************
223
224 Protocol management and registration routines
225
226*******************************************************************************/
227
228/*
229 * For efficiency
230 */
231
Adrian Bunk60481262006-06-25 23:58:10 -0700232static int netdev_nit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
234/*
235 * Add a protocol ID to the list. Now that the input handler is
236 * smarter we can dispense with all the messy stuff that used to be
237 * here.
238 *
239 * BEWARE!!! Protocol handlers, mangling input packets,
240 * MUST BE last in hash buckets and checking protocol handlers
241 * MUST start from promiscuous ptype_all chain in net_bh.
242 * It is true now, do not change it.
243 * Explanation follows: if protocol handler, mangling packet, will
244 * be the first on list, it is not able to sense, that packet
245 * is cloned and should be copied-on-write, so that it will
246 * change it and subsequent readers will get broken packet.
247 * --ANK (980803)
248 */
249
250/**
251 * dev_add_pack - add packet handler
252 * @pt: packet type declaration
253 *
254 * Add a protocol handler to the networking stack. The passed &packet_type
255 * is linked into kernel lists and may not be freed until it has been
256 * removed from the kernel lists.
257 *
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900258 * This call does not sleep therefore it can not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 * guarantee all CPU's that are in middle of receiving packets
260 * will see the new packet type (until the next received packet).
261 */
262
263void dev_add_pack(struct packet_type *pt)
264{
265 int hash;
266
267 spin_lock_bh(&ptype_lock);
268 if (pt->type == htons(ETH_P_ALL)) {
269 netdev_nit++;
270 list_add_rcu(&pt->list, &ptype_all);
271 } else {
272 hash = ntohs(pt->type) & 15;
273 list_add_rcu(&pt->list, &ptype_base[hash]);
274 }
275 spin_unlock_bh(&ptype_lock);
276}
277
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278/**
279 * __dev_remove_pack - remove packet handler
280 * @pt: packet type declaration
281 *
282 * Remove a protocol handler that was previously added to the kernel
283 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
284 * from the kernel lists and can be freed or reused once this function
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900285 * returns.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 *
287 * The packet type might still be in use by receivers
288 * and must not be freed until after all the CPU's have gone
289 * through a quiescent state.
290 */
291void __dev_remove_pack(struct packet_type *pt)
292{
293 struct list_head *head;
294 struct packet_type *pt1;
295
296 spin_lock_bh(&ptype_lock);
297
298 if (pt->type == htons(ETH_P_ALL)) {
299 netdev_nit--;
300 head = &ptype_all;
301 } else
302 head = &ptype_base[ntohs(pt->type) & 15];
303
304 list_for_each_entry(pt1, head, list) {
305 if (pt == pt1) {
306 list_del_rcu(&pt->list);
307 goto out;
308 }
309 }
310
311 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
312out:
313 spin_unlock_bh(&ptype_lock);
314}
315/**
316 * dev_remove_pack - remove packet handler
317 * @pt: packet type declaration
318 *
319 * Remove a protocol handler that was previously added to the kernel
320 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
321 * from the kernel lists and can be freed or reused once this function
322 * returns.
323 *
324 * This call sleeps to guarantee that no CPU is looking at the packet
325 * type after return.
326 */
327void dev_remove_pack(struct packet_type *pt)
328{
329 __dev_remove_pack(pt);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 synchronize_net();
332}
333
334/******************************************************************************
335
336 Device Boot-time Settings Routines
337
338*******************************************************************************/
339
340/* Boot time configuration table */
341static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
342
343/**
344 * netdev_boot_setup_add - add new setup entry
345 * @name: name of the device
346 * @map: configured settings for the device
347 *
348 * Adds new setup entry to the dev_boot_setup list. The function
349 * returns 0 on error and 1 on success. This is a generic routine to
350 * all netdevices.
351 */
352static int netdev_boot_setup_add(char *name, struct ifmap *map)
353{
354 struct netdev_boot_setup *s;
355 int i;
356
357 s = dev_boot_setup;
358 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
359 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
360 memset(s[i].name, 0, sizeof(s[i].name));
361 strcpy(s[i].name, name);
362 memcpy(&s[i].map, map, sizeof(s[i].map));
363 break;
364 }
365 }
366
367 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
368}
369
370/**
371 * netdev_boot_setup_check - check boot time settings
372 * @dev: the netdevice
373 *
374 * Check boot time settings for the device.
375 * The found settings are set for the device to be used
376 * later in the device probing.
377 * Returns 0 if no settings found, 1 if they are.
378 */
379int netdev_boot_setup_check(struct net_device *dev)
380{
381 struct netdev_boot_setup *s = dev_boot_setup;
382 int i;
383
384 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
385 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
386 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
387 dev->irq = s[i].map.irq;
388 dev->base_addr = s[i].map.base_addr;
389 dev->mem_start = s[i].map.mem_start;
390 dev->mem_end = s[i].map.mem_end;
391 return 1;
392 }
393 }
394 return 0;
395}
396
397
398/**
399 * netdev_boot_base - get address from boot time settings
400 * @prefix: prefix for network device
401 * @unit: id for network device
402 *
403 * Check boot time settings for the base address of device.
404 * The found settings are set for the device to be used
405 * later in the device probing.
406 * Returns 0 if no settings found.
407 */
408unsigned long netdev_boot_base(const char *prefix, int unit)
409{
410 const struct netdev_boot_setup *s = dev_boot_setup;
411 char name[IFNAMSIZ];
412 int i;
413
414 sprintf(name, "%s%d", prefix, unit);
415
416 /*
417 * If device already registered then return base of 1
418 * to indicate not to probe for this interface
419 */
420 if (__dev_get_by_name(name))
421 return 1;
422
423 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
424 if (!strcmp(name, s[i].name))
425 return s[i].map.base_addr;
426 return 0;
427}
428
429/*
430 * Saves at boot time configured settings for any netdevice.
431 */
432int __init netdev_boot_setup(char *str)
433{
434 int ints[5];
435 struct ifmap map;
436
437 str = get_options(str, ARRAY_SIZE(ints), ints);
438 if (!str || !*str)
439 return 0;
440
441 /* Save settings */
442 memset(&map, 0, sizeof(map));
443 if (ints[0] > 0)
444 map.irq = ints[1];
445 if (ints[0] > 1)
446 map.base_addr = ints[2];
447 if (ints[0] > 2)
448 map.mem_start = ints[3];
449 if (ints[0] > 3)
450 map.mem_end = ints[4];
451
452 /* Add new entry to the list */
453 return netdev_boot_setup_add(str, &map);
454}
455
456__setup("netdev=", netdev_boot_setup);
457
458/*******************************************************************************
459
460 Device Interface Subroutines
461
462*******************************************************************************/
463
464/**
465 * __dev_get_by_name - find a device by its name
466 * @name: name to find
467 *
468 * Find an interface by name. Must be called under RTNL semaphore
469 * or @dev_base_lock. If the name is found a pointer to the device
470 * is returned. If the name is not found then %NULL is returned. The
471 * reference counters are not incremented so the caller must be
472 * careful with locks.
473 */
474
475struct net_device *__dev_get_by_name(const char *name)
476{
477 struct hlist_node *p;
478
479 hlist_for_each(p, dev_name_hash(name)) {
480 struct net_device *dev
481 = hlist_entry(p, struct net_device, name_hlist);
482 if (!strncmp(dev->name, name, IFNAMSIZ))
483 return dev;
484 }
485 return NULL;
486}
487
488/**
489 * dev_get_by_name - find a device by its name
490 * @name: name to find
491 *
492 * Find an interface by name. This can be called from any
493 * context and does its own locking. The returned handle has
494 * the usage count incremented and the caller must use dev_put() to
495 * release it when it is no longer needed. %NULL is returned if no
496 * matching device is found.
497 */
498
499struct net_device *dev_get_by_name(const char *name)
500{
501 struct net_device *dev;
502
503 read_lock(&dev_base_lock);
504 dev = __dev_get_by_name(name);
505 if (dev)
506 dev_hold(dev);
507 read_unlock(&dev_base_lock);
508 return dev;
509}
510
511/**
512 * __dev_get_by_index - find a device by its ifindex
513 * @ifindex: index of device
514 *
515 * Search for an interface by index. Returns %NULL if the device
516 * is not found or a pointer to the device. The device has not
517 * had its reference counter increased so the caller must be careful
518 * about locking. The caller must hold either the RTNL semaphore
519 * or @dev_base_lock.
520 */
521
522struct net_device *__dev_get_by_index(int ifindex)
523{
524 struct hlist_node *p;
525
526 hlist_for_each(p, dev_index_hash(ifindex)) {
527 struct net_device *dev
528 = hlist_entry(p, struct net_device, index_hlist);
529 if (dev->ifindex == ifindex)
530 return dev;
531 }
532 return NULL;
533}
534
535
536/**
537 * dev_get_by_index - find a device by its ifindex
538 * @ifindex: index of device
539 *
540 * Search for an interface by index. Returns NULL if the device
541 * is not found or a pointer to the device. The device returned has
542 * had a reference added and the pointer is safe until the user calls
543 * dev_put to indicate they have finished with it.
544 */
545
546struct net_device *dev_get_by_index(int ifindex)
547{
548 struct net_device *dev;
549
550 read_lock(&dev_base_lock);
551 dev = __dev_get_by_index(ifindex);
552 if (dev)
553 dev_hold(dev);
554 read_unlock(&dev_base_lock);
555 return dev;
556}
557
558/**
559 * dev_getbyhwaddr - find a device by its hardware address
560 * @type: media type of device
561 * @ha: hardware address
562 *
563 * Search for an interface by MAC address. Returns NULL if the device
564 * is not found or a pointer to the device. The caller must hold the
565 * rtnl semaphore. The returned device has not had its ref count increased
566 * and the caller must therefore be careful about locking
567 *
568 * BUGS:
569 * If the API was consistent this would be __dev_get_by_hwaddr
570 */
571
572struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
573{
574 struct net_device *dev;
575
576 ASSERT_RTNL();
577
578 for (dev = dev_base; dev; dev = dev->next)
579 if (dev->type == type &&
580 !memcmp(dev->dev_addr, ha, dev->addr_len))
581 break;
582 return dev;
583}
584
Jochen Friedrichcf309e32005-09-22 04:44:55 -0300585EXPORT_SYMBOL(dev_getbyhwaddr);
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587struct net_device *dev_getfirstbyhwtype(unsigned short type)
588{
589 struct net_device *dev;
590
591 rtnl_lock();
592 for (dev = dev_base; dev; dev = dev->next) {
593 if (dev->type == type) {
594 dev_hold(dev);
595 break;
596 }
597 }
598 rtnl_unlock();
599 return dev;
600}
601
602EXPORT_SYMBOL(dev_getfirstbyhwtype);
603
604/**
605 * dev_get_by_flags - find any device with given flags
606 * @if_flags: IFF_* values
607 * @mask: bitmask of bits in if_flags to check
608 *
609 * Search for any interface with the given flags. Returns NULL if a device
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900610 * is not found or a pointer to the device. The device returned has
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 * had a reference added and the pointer is safe until the user calls
612 * dev_put to indicate they have finished with it.
613 */
614
615struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
616{
617 struct net_device *dev;
618
619 read_lock(&dev_base_lock);
620 for (dev = dev_base; dev != NULL; dev = dev->next) {
621 if (((dev->flags ^ if_flags) & mask) == 0) {
622 dev_hold(dev);
623 break;
624 }
625 }
626 read_unlock(&dev_base_lock);
627 return dev;
628}
629
630/**
631 * dev_valid_name - check if name is okay for network device
632 * @name: name string
633 *
634 * Network device names need to be valid file names to
David S. Millerc7fa9d12006-08-15 16:34:13 -0700635 * to allow sysfs to work. We also disallow any kind of
636 * whitespace.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 */
Mitch Williamsc2373ee2005-11-09 10:34:45 -0800638int dev_valid_name(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
David S. Millerc7fa9d12006-08-15 16:34:13 -0700640 if (*name == '\0')
641 return 0;
Stephen Hemmingerb6fe17d2006-08-29 17:06:13 -0700642 if (strlen(name) >= IFNAMSIZ)
643 return 0;
David S. Millerc7fa9d12006-08-15 16:34:13 -0700644 if (!strcmp(name, ".") || !strcmp(name, ".."))
645 return 0;
646
647 while (*name) {
648 if (*name == '/' || isspace(*name))
649 return 0;
650 name++;
651 }
652 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653}
654
655/**
656 * dev_alloc_name - allocate a name for a device
657 * @dev: device
658 * @name: name format string
659 *
660 * Passed a format string - eg "lt%d" it will try and find a suitable
Stephen Hemminger3041a062006-05-26 13:25:24 -0700661 * id. It scans list of devices to build up a free map, then chooses
662 * the first empty slot. The caller must hold the dev_base or rtnl lock
663 * while allocating the name and adding the device in order to avoid
664 * duplicates.
665 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
666 * Returns the number of the unit assigned or a negative errno code.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 */
668
669int dev_alloc_name(struct net_device *dev, const char *name)
670{
671 int i = 0;
672 char buf[IFNAMSIZ];
673 const char *p;
674 const int max_netdevices = 8*PAGE_SIZE;
675 long *inuse;
676 struct net_device *d;
677
678 p = strnchr(name, IFNAMSIZ-1, '%');
679 if (p) {
680 /*
681 * Verify the string as this thing may have come from
682 * the user. There must be either one "%d" and no other "%"
683 * characters.
684 */
685 if (p[1] != 'd' || strchr(p + 2, '%'))
686 return -EINVAL;
687
688 /* Use one page as a bit array of possible slots */
689 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
690 if (!inuse)
691 return -ENOMEM;
692
693 for (d = dev_base; d; d = d->next) {
694 if (!sscanf(d->name, name, &i))
695 continue;
696 if (i < 0 || i >= max_netdevices)
697 continue;
698
699 /* avoid cases where sscanf is not exact inverse of printf */
700 snprintf(buf, sizeof(buf), name, i);
701 if (!strncmp(buf, d->name, IFNAMSIZ))
702 set_bit(i, inuse);
703 }
704
705 i = find_first_zero_bit(inuse, max_netdevices);
706 free_page((unsigned long) inuse);
707 }
708
709 snprintf(buf, sizeof(buf), name, i);
710 if (!__dev_get_by_name(buf)) {
711 strlcpy(dev->name, buf, IFNAMSIZ);
712 return i;
713 }
714
715 /* It is possible to run out of possible slots
716 * when the name is long and there isn't enough space left
717 * for the digits, or if all bits are used.
718 */
719 return -ENFILE;
720}
721
722
723/**
724 * dev_change_name - change name of a device
725 * @dev: device
726 * @newname: name (or format string) must be at least IFNAMSIZ
727 *
728 * Change name of a device, can pass format strings "eth%d".
729 * for wildcarding.
730 */
731int dev_change_name(struct net_device *dev, char *newname)
732{
733 int err = 0;
734
735 ASSERT_RTNL();
736
737 if (dev->flags & IFF_UP)
738 return -EBUSY;
739
740 if (!dev_valid_name(newname))
741 return -EINVAL;
742
743 if (strchr(newname, '%')) {
744 err = dev_alloc_name(dev, newname);
745 if (err < 0)
746 return err;
747 strcpy(newname, dev->name);
748 }
749 else if (__dev_get_by_name(newname))
750 return -EEXIST;
751 else
752 strlcpy(dev->name, newname, IFNAMSIZ);
753
Eric W. Biederman92749822007-04-03 00:07:30 -0600754 device_rename(&dev->dev, dev->name);
755 hlist_del(&dev->name_hlist);
756 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
757 raw_notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
759 return err;
760}
761
762/**
Stephen Hemminger3041a062006-05-26 13:25:24 -0700763 * netdev_features_change - device changes features
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700764 * @dev: device to cause notification
765 *
766 * Called to indicate a device has changed features.
767 */
768void netdev_features_change(struct net_device *dev)
769{
Alan Sternf07d5b92006-05-09 15:23:03 -0700770 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700771}
772EXPORT_SYMBOL(netdev_features_change);
773
774/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 * netdev_state_change - device changes state
776 * @dev: device to cause notification
777 *
778 * Called to indicate a device has changed state. This function calls
779 * the notifier chains for netdev_chain and sends a NEWLINK message
780 * to the routing socket.
781 */
782void netdev_state_change(struct net_device *dev)
783{
784 if (dev->flags & IFF_UP) {
Alan Sternf07d5b92006-05-09 15:23:03 -0700785 raw_notifier_call_chain(&netdev_chain,
Alan Sterne041c682006-03-27 01:16:30 -0800786 NETDEV_CHANGE, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
788 }
789}
790
791/**
792 * dev_load - load a network module
793 * @name: name of interface
794 *
795 * If a network interface is not present and the process has suitable
796 * privileges this function loads the module. If module loading is not
797 * available in this kernel then it becomes a nop.
798 */
799
800void dev_load(const char *name)
801{
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900802 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804 read_lock(&dev_base_lock);
805 dev = __dev_get_by_name(name);
806 read_unlock(&dev_base_lock);
807
808 if (!dev && capable(CAP_SYS_MODULE))
809 request_module("%s", name);
810}
811
812static int default_rebuild_header(struct sk_buff *skb)
813{
814 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
815 skb->dev ? skb->dev->name : "NULL!!!");
816 kfree_skb(skb);
817 return 1;
818}
819
820
821/**
822 * dev_open - prepare an interface for use.
823 * @dev: device to open
824 *
825 * Takes a device from down to up state. The device's private open
826 * function is invoked and then the multicast lists are loaded. Finally
827 * the device is moved into the up state and a %NETDEV_UP message is
828 * sent to the netdev notifier chain.
829 *
830 * Calling this function on an active interface is a nop. On a failure
831 * a negative errno code is returned.
832 */
833int dev_open(struct net_device *dev)
834{
835 int ret = 0;
836
837 /*
838 * Is it already up?
839 */
840
841 if (dev->flags & IFF_UP)
842 return 0;
843
844 /*
845 * Is it even present?
846 */
847 if (!netif_device_present(dev))
848 return -ENODEV;
849
850 /*
851 * Call device private open method
852 */
853 set_bit(__LINK_STATE_START, &dev->state);
854 if (dev->open) {
855 ret = dev->open(dev);
856 if (ret)
857 clear_bit(__LINK_STATE_START, &dev->state);
858 }
859
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900860 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 * If it went open OK then:
862 */
863
864 if (!ret) {
865 /*
866 * Set the flags.
867 */
868 dev->flags |= IFF_UP;
869
870 /*
871 * Initialize multicasting status
872 */
873 dev_mc_upload(dev);
874
875 /*
876 * Wakeup transmit queue engine
877 */
878 dev_activate(dev);
879
880 /*
881 * ... and announce new interface.
882 */
Alan Sternf07d5b92006-05-09 15:23:03 -0700883 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 }
885 return ret;
886}
887
888/**
889 * dev_close - shutdown an interface.
890 * @dev: device to shutdown
891 *
892 * This function moves an active device into down state. A
893 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
894 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
895 * chain.
896 */
897int dev_close(struct net_device *dev)
898{
899 if (!(dev->flags & IFF_UP))
900 return 0;
901
902 /*
903 * Tell people we are going down, so that they can
904 * prepare to death, when device is still operating.
905 */
Alan Sternf07d5b92006-05-09 15:23:03 -0700906 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 dev_deactivate(dev);
909
910 clear_bit(__LINK_STATE_START, &dev->state);
911
912 /* Synchronize to scheduled poll. We cannot touch poll list,
913 * it can be even on different cpu. So just clear netif_running(),
914 * and wait when poll really will happen. Actually, the best place
915 * for this is inside dev->stop() after device stopped its irq
916 * engine, but this requires more changes in devices. */
917
918 smp_mb__after_clear_bit(); /* Commit netif_running(). */
919 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
920 /* No hurry. */
David S. Miller6192b542005-07-28 12:12:58 -0700921 msleep(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 }
923
924 /*
925 * Call the device specific close. This cannot fail.
926 * Only if device is UP
927 *
928 * We allow it to be called even after a DETACH hot-plug
929 * event.
930 */
931 if (dev->stop)
932 dev->stop(dev);
933
934 /*
935 * Device is now down.
936 */
937
938 dev->flags &= ~IFF_UP;
939
940 /*
941 * Tell people we are down
942 */
Alan Sternf07d5b92006-05-09 15:23:03 -0700943 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
945 return 0;
946}
947
948
949/*
950 * Device change register/unregister. These are not inline or static
951 * as we export them to the world.
952 */
953
954/**
955 * register_netdevice_notifier - register a network notifier block
956 * @nb: notifier
957 *
958 * Register a notifier to be called when network device events occur.
959 * The notifier passed is linked into the kernel structures and must
960 * not be reused until it has been unregistered. A negative errno code
961 * is returned on a failure.
962 *
963 * When registered all registration and up events are replayed
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900964 * to the new notifier to allow device to have a race free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 * view of the network device list.
966 */
967
968int register_netdevice_notifier(struct notifier_block *nb)
969{
970 struct net_device *dev;
971 int err;
972
973 rtnl_lock();
Alan Sternf07d5b92006-05-09 15:23:03 -0700974 err = raw_notifier_chain_register(&netdev_chain, nb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 if (!err) {
976 for (dev = dev_base; dev; dev = dev->next) {
977 nb->notifier_call(nb, NETDEV_REGISTER, dev);
978
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900979 if (dev->flags & IFF_UP)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 nb->notifier_call(nb, NETDEV_UP, dev);
981 }
982 }
983 rtnl_unlock();
984 return err;
985}
986
987/**
988 * unregister_netdevice_notifier - unregister a network notifier block
989 * @nb: notifier
990 *
991 * Unregister a notifier previously registered by
992 * register_netdevice_notifier(). The notifier is unlinked into the
993 * kernel structures and may then be reused. A negative errno code
994 * is returned on a failure.
995 */
996
997int unregister_netdevice_notifier(struct notifier_block *nb)
998{
Herbert Xu9f514952006-03-25 01:24:25 -0800999 int err;
1000
1001 rtnl_lock();
Alan Sternf07d5b92006-05-09 15:23:03 -07001002 err = raw_notifier_chain_unregister(&netdev_chain, nb);
Herbert Xu9f514952006-03-25 01:24:25 -08001003 rtnl_unlock();
1004 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005}
1006
1007/**
1008 * call_netdevice_notifiers - call all network notifier blocks
1009 * @val: value passed unmodified to notifier function
1010 * @v: pointer passed unmodified to notifier function
1011 *
1012 * Call all network notifier blocks. Parameters and return value
Alan Sternf07d5b92006-05-09 15:23:03 -07001013 * are as for raw_notifier_call_chain().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 */
1015
1016int call_netdevice_notifiers(unsigned long val, void *v)
1017{
Alan Sternf07d5b92006-05-09 15:23:03 -07001018 return raw_notifier_call_chain(&netdev_chain, val, v);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019}
1020
1021/* When > 0 there are consumers of rx skb time stamps */
1022static atomic_t netstamp_needed = ATOMIC_INIT(0);
1023
1024void net_enable_timestamp(void)
1025{
1026 atomic_inc(&netstamp_needed);
1027}
1028
1029void net_disable_timestamp(void)
1030{
1031 atomic_dec(&netstamp_needed);
1032}
1033
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001034void __net_timestamp(struct sk_buff *skb)
1035{
1036 struct timeval tv;
1037
1038 do_gettimeofday(&tv);
1039 skb_set_timestamp(skb, &tv);
1040}
1041EXPORT_SYMBOL(__net_timestamp);
1042
1043static inline void net_timestamp(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044{
1045 if (atomic_read(&netstamp_needed))
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001046 __net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 else {
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001048 skb->tstamp.off_sec = 0;
1049 skb->tstamp.off_usec = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
1051}
1052
1053/*
1054 * Support routine. Sends outgoing frames to any network
1055 * taps currently in use.
1056 */
1057
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001058static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
1060 struct packet_type *ptype;
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001061
1062 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
1064 rcu_read_lock();
1065 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1066 /* Never send packets back to the socket
1067 * they originated from - MvS (miquels@drinkel.ow.org)
1068 */
1069 if ((ptype->dev == dev || !ptype->dev) &&
1070 (ptype->af_packet_priv == NULL ||
1071 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1072 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1073 if (!skb2)
1074 break;
1075
1076 /* skb->nh should be correctly
1077 set by sender, so that the second statement is
1078 just protection against buggy protocols.
1079 */
1080 skb2->mac.raw = skb2->data;
1081
1082 if (skb2->nh.raw < skb2->data ||
1083 skb2->nh.raw > skb2->tail) {
1084 if (net_ratelimit())
1085 printk(KERN_CRIT "protocol %04x is "
1086 "buggy, dev %s\n",
1087 skb2->protocol, dev->name);
1088 skb2->nh.raw = skb2->data;
1089 }
1090
1091 skb2->h.raw = skb2->nh.raw;
1092 skb2->pkt_type = PACKET_OUTGOING;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001093 ptype->func(skb2, skb->dev, ptype, skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 }
1095 }
1096 rcu_read_unlock();
1097}
1098
Denis Vlasenko56079432006-03-29 15:57:29 -08001099
1100void __netif_schedule(struct net_device *dev)
1101{
1102 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1103 unsigned long flags;
1104 struct softnet_data *sd;
1105
1106 local_irq_save(flags);
1107 sd = &__get_cpu_var(softnet_data);
1108 dev->next_sched = sd->output_queue;
1109 sd->output_queue = dev;
1110 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1111 local_irq_restore(flags);
1112 }
1113}
1114EXPORT_SYMBOL(__netif_schedule);
1115
1116void __netif_rx_schedule(struct net_device *dev)
1117{
1118 unsigned long flags;
1119
1120 local_irq_save(flags);
1121 dev_hold(dev);
1122 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1123 if (dev->quota < 0)
1124 dev->quota += dev->weight;
1125 else
1126 dev->quota = dev->weight;
1127 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1128 local_irq_restore(flags);
1129}
1130EXPORT_SYMBOL(__netif_rx_schedule);
1131
1132void dev_kfree_skb_any(struct sk_buff *skb)
1133{
1134 if (in_irq() || irqs_disabled())
1135 dev_kfree_skb_irq(skb);
1136 else
1137 dev_kfree_skb(skb);
1138}
1139EXPORT_SYMBOL(dev_kfree_skb_any);
1140
1141
1142/* Hot-plugging. */
1143void netif_device_detach(struct net_device *dev)
1144{
1145 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1146 netif_running(dev)) {
1147 netif_stop_queue(dev);
1148 }
1149}
1150EXPORT_SYMBOL(netif_device_detach);
1151
1152void netif_device_attach(struct net_device *dev)
1153{
1154 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1155 netif_running(dev)) {
1156 netif_wake_queue(dev);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001157 __netdev_watchdog_up(dev);
Denis Vlasenko56079432006-03-29 15:57:29 -08001158 }
1159}
1160EXPORT_SYMBOL(netif_device_attach);
1161
1162
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163/*
1164 * Invalidate hardware checksum when packet is to be mangled, and
1165 * complete checksum manually on outgoing path.
1166 */
Patrick McHardy84fa7932006-08-29 16:44:56 -07001167int skb_checksum_help(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168{
Al Virod3bc23e2006-11-14 21:24:49 -08001169 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 int ret = 0, offset = skb->h.raw - skb->data;
1171
Patrick McHardy84fa7932006-08-29 16:44:56 -07001172 if (skb->ip_summed == CHECKSUM_COMPLETE)
Herbert Xua430a432006-07-08 13:34:56 -07001173 goto out_set_summed;
1174
1175 if (unlikely(skb_shinfo(skb)->gso_size)) {
Herbert Xua430a432006-07-08 13:34:56 -07001176 /* Let GSO fix up the checksum. */
1177 goto out_set_summed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 }
1179
1180 if (skb_cloned(skb)) {
1181 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1182 if (ret)
1183 goto out;
1184 }
1185
Kris Katterjohn09a62662006-01-08 22:24:28 -08001186 BUG_ON(offset > (int)skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1188
1189 offset = skb->tail - skb->h.raw;
Kris Katterjohn09a62662006-01-08 22:24:28 -08001190 BUG_ON(offset <= 0);
Al Viroff1dcad2006-11-20 18:07:29 -08001191 BUG_ON(skb->csum_offset + 2 > offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
Al Viroff1dcad2006-11-20 18:07:29 -08001193 *(__sum16*)(skb->h.raw + skb->csum_offset) = csum_fold(csum);
Herbert Xua430a432006-07-08 13:34:56 -07001194
1195out_set_summed:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 skb->ip_summed = CHECKSUM_NONE;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001197out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 return ret;
1199}
1200
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001201/**
1202 * skb_gso_segment - Perform segmentation on skb.
1203 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07001204 * @features: features for the output path (see dev->features)
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001205 *
1206 * This function segments the given skb and returns a list of segments.
Herbert Xu576a30e2006-06-27 13:22:38 -07001207 *
1208 * It may return NULL if the skb requires no segmentation. This is
1209 * only possible when GSO is used for verifying header integrity.
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001210 */
Herbert Xu576a30e2006-06-27 13:22:38 -07001211struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001212{
1213 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1214 struct packet_type *ptype;
Al Viro252e3342006-11-14 20:48:11 -08001215 __be16 type = skb->protocol;
Herbert Xua430a432006-07-08 13:34:56 -07001216 int err;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001217
1218 BUG_ON(skb_shinfo(skb)->frag_list);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001219
1220 skb->mac.raw = skb->data;
1221 skb->mac_len = skb->nh.raw - skb->data;
1222 __skb_pull(skb, skb->mac_len);
1223
Patrick McHardy84fa7932006-08-29 16:44:56 -07001224 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
Herbert Xua430a432006-07-08 13:34:56 -07001225 if (skb_header_cloned(skb) &&
1226 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1227 return ERR_PTR(err);
1228 }
1229
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001230 rcu_read_lock();
1231 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1232 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
Patrick McHardy84fa7932006-08-29 16:44:56 -07001233 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
Herbert Xua430a432006-07-08 13:34:56 -07001234 err = ptype->gso_send_check(skb);
1235 segs = ERR_PTR(err);
1236 if (err || skb_gso_ok(skb, features))
1237 break;
1238 __skb_push(skb, skb->data - skb->nh.raw);
1239 }
Herbert Xu576a30e2006-06-27 13:22:38 -07001240 segs = ptype->gso_segment(skb, features);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001241 break;
1242 }
1243 }
1244 rcu_read_unlock();
1245
Herbert Xu576a30e2006-06-27 13:22:38 -07001246 __skb_push(skb, skb->data - skb->mac.raw);
1247
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001248 return segs;
1249}
1250
1251EXPORT_SYMBOL(skb_gso_segment);
1252
Herbert Xufb286bb2005-11-10 13:01:24 -08001253/* Take action when hardware reception checksum errors are detected. */
1254#ifdef CONFIG_BUG
1255void netdev_rx_csum_fault(struct net_device *dev)
1256{
1257 if (net_ratelimit()) {
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001258 printk(KERN_ERR "%s: hw csum failure.\n",
Stephen Hemminger246a4212005-12-08 15:21:39 -08001259 dev ? dev->name : "<unknown>");
Herbert Xufb286bb2005-11-10 13:01:24 -08001260 dump_stack();
1261 }
1262}
1263EXPORT_SYMBOL(netdev_rx_csum_fault);
1264#endif
1265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266/* Actually, we should eliminate this check as soon as we know, that:
1267 * 1. IOMMU is present and allows to map all the memory.
1268 * 2. No high memory really exists on this machine.
1269 */
1270
1271static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1272{
Herbert Xu3d3a8532006-06-27 13:33:10 -07001273#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 int i;
1275
1276 if (dev->features & NETIF_F_HIGHDMA)
1277 return 0;
1278
1279 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1280 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1281 return 1;
1282
Herbert Xu3d3a8532006-06-27 13:33:10 -07001283#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 return 0;
1285}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001287struct dev_gso_cb {
1288 void (*destructor)(struct sk_buff *skb);
1289};
1290
1291#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1292
1293static void dev_gso_skb_destructor(struct sk_buff *skb)
1294{
1295 struct dev_gso_cb *cb;
1296
1297 do {
1298 struct sk_buff *nskb = skb->next;
1299
1300 skb->next = nskb->next;
1301 nskb->next = NULL;
1302 kfree_skb(nskb);
1303 } while (skb->next);
1304
1305 cb = DEV_GSO_CB(skb);
1306 if (cb->destructor)
1307 cb->destructor(skb);
1308}
1309
1310/**
1311 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1312 * @skb: buffer to segment
1313 *
1314 * This function segments the given skb and stores the list of segments
1315 * in skb->next.
1316 */
1317static int dev_gso_segment(struct sk_buff *skb)
1318{
1319 struct net_device *dev = skb->dev;
1320 struct sk_buff *segs;
Herbert Xu576a30e2006-06-27 13:22:38 -07001321 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1322 NETIF_F_SG : 0);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001323
Herbert Xu576a30e2006-06-27 13:22:38 -07001324 segs = skb_gso_segment(skb, features);
1325
1326 /* Verifying header integrity only. */
1327 if (!segs)
1328 return 0;
1329
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001330 if (unlikely(IS_ERR(segs)))
1331 return PTR_ERR(segs);
1332
1333 skb->next = segs;
1334 DEV_GSO_CB(skb)->destructor = skb->destructor;
1335 skb->destructor = dev_gso_skb_destructor;
1336
1337 return 0;
1338}
1339
1340int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1341{
1342 if (likely(!skb->next)) {
1343 if (netdev_nit)
1344 dev_queue_xmit_nit(skb, dev);
1345
Herbert Xu576a30e2006-06-27 13:22:38 -07001346 if (netif_needs_gso(dev, skb)) {
1347 if (unlikely(dev_gso_segment(skb)))
1348 goto out_kfree_skb;
1349 if (skb->next)
1350 goto gso;
1351 }
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001352
Herbert Xu576a30e2006-06-27 13:22:38 -07001353 return dev->hard_start_xmit(skb, dev);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001354 }
1355
Herbert Xu576a30e2006-06-27 13:22:38 -07001356gso:
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001357 do {
1358 struct sk_buff *nskb = skb->next;
1359 int rc;
1360
1361 skb->next = nskb->next;
1362 nskb->next = NULL;
1363 rc = dev->hard_start_xmit(nskb, dev);
1364 if (unlikely(rc)) {
Michael Chanf54d9e82006-06-25 23:57:04 -07001365 nskb->next = skb->next;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001366 skb->next = nskb;
1367 return rc;
1368 }
Michael Chanf54d9e82006-06-25 23:57:04 -07001369 if (unlikely(netif_queue_stopped(dev) && skb->next))
1370 return NETDEV_TX_BUSY;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001371 } while (skb->next);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001372
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001373 skb->destructor = DEV_GSO_CB(skb)->destructor;
1374
1375out_kfree_skb:
1376 kfree_skb(skb);
1377 return 0;
1378}
1379
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380#define HARD_TX_LOCK(dev, cpu) { \
1381 if ((dev->features & NETIF_F_LLTX) == 0) { \
Herbert Xu932ff272006-06-09 12:20:56 -07001382 netif_tx_lock(dev); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 } \
1384}
1385
1386#define HARD_TX_UNLOCK(dev) { \
1387 if ((dev->features & NETIF_F_LLTX) == 0) { \
Herbert Xu932ff272006-06-09 12:20:56 -07001388 netif_tx_unlock(dev); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 } \
1390}
1391
1392/**
1393 * dev_queue_xmit - transmit a buffer
1394 * @skb: buffer to transmit
1395 *
1396 * Queue a buffer for transmission to a network device. The caller must
1397 * have set the device and priority and built the buffer before calling
1398 * this function. The function can be called from an interrupt.
1399 *
1400 * A negative errno code is returned on a failure. A success does not
1401 * guarantee the frame will be transmitted as it may be dropped due
1402 * to congestion or traffic shaping.
Ben Greearaf191362005-04-24 20:12:36 -07001403 *
1404 * -----------------------------------------------------------------------------------
1405 * I notice this method can also return errors from the queue disciplines,
1406 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1407 * be positive.
1408 *
1409 * Regardless of the return value, the skb is consumed, so it is currently
1410 * difficult to retry a send to this method. (You can bump the ref count
1411 * before sending to hold a reference for retry if you are careful.)
1412 *
1413 * When calling this method, interrupts MUST be enabled. This is because
1414 * the BH enable code must have IRQs enabled so that it will not deadlock.
1415 * --BLG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 */
1417
1418int dev_queue_xmit(struct sk_buff *skb)
1419{
1420 struct net_device *dev = skb->dev;
1421 struct Qdisc *q;
1422 int rc = -ENOMEM;
1423
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001424 /* GSO will handle the following emulations directly. */
1425 if (netif_needs_gso(dev, skb))
1426 goto gso;
1427
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 if (skb_shinfo(skb)->frag_list &&
1429 !(dev->features & NETIF_F_FRAGLIST) &&
Herbert Xu364c6ba2006-06-09 16:10:40 -07001430 __skb_linearize(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 goto out_kfree_skb;
1432
1433 /* Fragmented skb is linearized if device does not support SG,
1434 * or if at least one of fragments is in highmem and device
1435 * does not support DMA from it.
1436 */
1437 if (skb_shinfo(skb)->nr_frags &&
1438 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
Herbert Xu364c6ba2006-06-09 16:10:40 -07001439 __skb_linearize(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 goto out_kfree_skb;
1441
1442 /* If packet is not checksummed and device does not support
1443 * checksumming for this protocol, complete checksumming here.
1444 */
Patrick McHardy84fa7932006-08-29 16:44:56 -07001445 if (skb->ip_summed == CHECKSUM_PARTIAL &&
Herbert Xu8648b302006-06-17 22:06:05 -07001446 (!(dev->features & NETIF_F_GEN_CSUM) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 (!(dev->features & NETIF_F_IP_CSUM) ||
1448 skb->protocol != htons(ETH_P_IP))))
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001449 if (skb_checksum_help(skb))
1450 goto out_kfree_skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001452gso:
Eric Dumazet2d7ceec2005-09-27 15:22:58 -07001453 spin_lock_prefetch(&dev->queue_lock);
1454
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001455 /* Disable soft irqs for various locks below. Also
1456 * stops preemption for RCU.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001458 rcu_read_lock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001460 /* Updates of qdisc are serialized by queue_lock.
1461 * The struct Qdisc which is pointed to by qdisc is now a
1462 * rcu structure - it may be accessed without acquiring
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 * a lock (but the structure may be stale.) The freeing of the
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001464 * qdisc will be deferred until it's known that there are no
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 * more references to it.
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001466 *
1467 * If the qdisc has an enqueue function, we still need to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 * hold the queue_lock before calling it, since queue_lock
1469 * also serializes access to the device queue.
1470 */
1471
1472 q = rcu_dereference(dev->qdisc);
1473#ifdef CONFIG_NET_CLS_ACT
1474 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1475#endif
1476 if (q->enqueue) {
1477 /* Grab device queue */
1478 spin_lock(&dev->queue_lock);
Patrick McHardy85670cc2006-09-27 16:45:45 -07001479 q = dev->qdisc;
1480 if (q->enqueue) {
1481 rc = q->enqueue(skb, q);
1482 qdisc_run(dev);
1483 spin_unlock(&dev->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Patrick McHardy85670cc2006-09-27 16:45:45 -07001485 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1486 goto out;
1487 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 spin_unlock(&dev->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 }
1490
1491 /* The device has no queue. Common case for software devices:
1492 loopback, all the sorts of tunnels...
1493
Herbert Xu932ff272006-06-09 12:20:56 -07001494 Really, it is unlikely that netif_tx_lock protection is necessary
1495 here. (f.e. loopback and IP tunnels are clean ignoring statistics
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 counters.)
1497 However, it is possible, that they rely on protection
1498 made by us here.
1499
1500 Check this and shot the lock. It is not prone from deadlocks.
1501 Either shot noqueue qdisc, it is even simpler 8)
1502 */
1503 if (dev->flags & IFF_UP) {
1504 int cpu = smp_processor_id(); /* ok because BHs are off */
1505
1506 if (dev->xmit_lock_owner != cpu) {
1507
1508 HARD_TX_LOCK(dev, cpu);
1509
1510 if (!netif_queue_stopped(dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 rc = 0;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001512 if (!dev_hard_start_xmit(skb, dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 HARD_TX_UNLOCK(dev);
1514 goto out;
1515 }
1516 }
1517 HARD_TX_UNLOCK(dev);
1518 if (net_ratelimit())
1519 printk(KERN_CRIT "Virtual device %s asks to "
1520 "queue packet!\n", dev->name);
1521 } else {
1522 /* Recursion is detected! It is possible,
1523 * unfortunately */
1524 if (net_ratelimit())
1525 printk(KERN_CRIT "Dead loop on virtual device "
1526 "%s, fix it urgently!\n", dev->name);
1527 }
1528 }
1529
1530 rc = -ENETDOWN;
Herbert Xud4828d82006-06-22 02:28:18 -07001531 rcu_read_unlock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
1533out_kfree_skb:
1534 kfree_skb(skb);
1535 return rc;
1536out:
Herbert Xud4828d82006-06-22 02:28:18 -07001537 rcu_read_unlock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 return rc;
1539}
1540
1541
1542/*=======================================================================
1543 Receiver routines
1544 =======================================================================*/
1545
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001546int netdev_max_backlog = 1000;
1547int netdev_budget = 300;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548int weight_p = 64; /* old backlog weight */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1551
1552
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553/**
1554 * netif_rx - post buffer to the network code
1555 * @skb: buffer to post
1556 *
1557 * This function receives a packet from a device driver and queues it for
1558 * the upper (protocol) levels to process. It always succeeds. The buffer
1559 * may be dropped during processing for congestion control or by the
1560 * protocol layers.
1561 *
1562 * return values:
1563 * NET_RX_SUCCESS (no congestion)
1564 * NET_RX_CN_LOW (low congestion)
1565 * NET_RX_CN_MOD (moderate congestion)
1566 * NET_RX_CN_HIGH (high congestion)
1567 * NET_RX_DROP (packet was dropped)
1568 *
1569 */
1570
1571int netif_rx(struct sk_buff *skb)
1572{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 struct softnet_data *queue;
1574 unsigned long flags;
1575
1576 /* if netpoll wants it, pretend we never saw it */
1577 if (netpoll_rx(skb))
1578 return NET_RX_DROP;
1579
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001580 if (!skb->tstamp.off_sec)
1581 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582
1583 /*
1584 * The code is rearranged so that the path is the most
1585 * short when CPU is congested, but is still operating.
1586 */
1587 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 queue = &__get_cpu_var(softnet_data);
1589
1590 __get_cpu_var(netdev_rx_stat).total++;
1591 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1592 if (queue->input_pkt_queue.qlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593enqueue:
1594 dev_hold(skb->dev);
1595 __skb_queue_tail(&queue->input_pkt_queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 local_irq_restore(flags);
Stephen Hemminger34008d82005-06-23 20:10:00 -07001597 return NET_RX_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 }
1599
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 netif_rx_schedule(&queue->backlog_dev);
1601 goto enqueue;
1602 }
1603
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 __get_cpu_var(netdev_rx_stat).dropped++;
1605 local_irq_restore(flags);
1606
1607 kfree_skb(skb);
1608 return NET_RX_DROP;
1609}
1610
1611int netif_rx_ni(struct sk_buff *skb)
1612{
1613 int err;
1614
1615 preempt_disable();
1616 err = netif_rx(skb);
1617 if (local_softirq_pending())
1618 do_softirq();
1619 preempt_enable();
1620
1621 return err;
1622}
1623
1624EXPORT_SYMBOL(netif_rx_ni);
1625
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001626static inline struct net_device *skb_bond(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627{
1628 struct net_device *dev = skb->dev;
1629
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001630 if (dev->master) {
David S. Miller7ea49ed2006-08-14 17:08:36 -07001631 if (skb_bond_should_drop(skb)) {
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001632 kfree_skb(skb);
1633 return NULL;
1634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 skb->dev = dev->master;
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001636 }
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001637
1638 return dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639}
1640
1641static void net_tx_action(struct softirq_action *h)
1642{
1643 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1644
1645 if (sd->completion_queue) {
1646 struct sk_buff *clist;
1647
1648 local_irq_disable();
1649 clist = sd->completion_queue;
1650 sd->completion_queue = NULL;
1651 local_irq_enable();
1652
1653 while (clist) {
1654 struct sk_buff *skb = clist;
1655 clist = clist->next;
1656
1657 BUG_TRAP(!atomic_read(&skb->users));
1658 __kfree_skb(skb);
1659 }
1660 }
1661
1662 if (sd->output_queue) {
1663 struct net_device *head;
1664
1665 local_irq_disable();
1666 head = sd->output_queue;
1667 sd->output_queue = NULL;
1668 local_irq_enable();
1669
1670 while (head) {
1671 struct net_device *dev = head;
1672 head = head->next_sched;
1673
1674 smp_mb__before_clear_bit();
1675 clear_bit(__LINK_STATE_SCHED, &dev->state);
1676
1677 if (spin_trylock(&dev->queue_lock)) {
1678 qdisc_run(dev);
1679 spin_unlock(&dev->queue_lock);
1680 } else {
1681 netif_schedule(dev);
1682 }
1683 }
1684 }
1685}
1686
1687static __inline__ int deliver_skb(struct sk_buff *skb,
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001688 struct packet_type *pt_prev,
1689 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690{
1691 atomic_inc(&skb->users);
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001692 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693}
1694
1695#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1696int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1697struct net_bridge;
1698struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1699 unsigned char *addr);
1700void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1701
1702static __inline__ int handle_bridge(struct sk_buff **pskb,
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001703 struct packet_type **pt_prev, int *ret,
1704 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705{
1706 struct net_bridge_port *port;
1707
1708 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1709 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1710 return 0;
1711
1712 if (*pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001713 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 *pt_prev = NULL;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001715 }
1716
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 return br_handle_frame_hook(port, pskb);
1718}
1719#else
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001720#define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721#endif
1722
1723#ifdef CONFIG_NET_CLS_ACT
1724/* TODO: Maybe we should just force sch_ingress to be compiled in
1725 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1726 * a compare and 2 stores extra right now if we dont have it on
1727 * but have CONFIG_NET_CLS_ACT
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001728 * NOTE: This doesnt stop any functionality; if you dont have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 * the ingress scheduler, you just cant add policies on ingress.
1730 *
1731 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001732static int ing_filter(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733{
1734 struct Qdisc *q;
1735 struct net_device *dev = skb->dev;
1736 int result = TC_ACT_OK;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001737
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 if (dev->qdisc_ingress) {
1739 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1740 if (MAX_RED_LOOP < ttl++) {
Patrick McHardyc01003c2007-03-29 11:46:52 -07001741 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1742 skb->iif, skb->dev->ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 return TC_ACT_SHOT;
1744 }
1745
1746 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1747
1748 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
David S. Miller86e65da2005-08-09 19:36:29 -07001749
Patrick McHardy035832a2007-03-24 22:13:25 -07001750 spin_lock(&dev->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 if ((q = dev->qdisc_ingress) != NULL)
1752 result = q->enqueue(skb, q);
Patrick McHardy035832a2007-03-24 22:13:25 -07001753 spin_unlock(&dev->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754
1755 }
1756
1757 return result;
1758}
1759#endif
1760
1761int netif_receive_skb(struct sk_buff *skb)
1762{
1763 struct packet_type *ptype, *pt_prev;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001764 struct net_device *orig_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 int ret = NET_RX_DROP;
Al Viro252e3342006-11-14 20:48:11 -08001766 __be16 type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767
1768 /* if we've gotten here through NAPI, check netpoll */
1769 if (skb->dev->poll && netpoll_rx(skb))
1770 return NET_RX_DROP;
1771
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001772 if (!skb->tstamp.off_sec)
1773 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
Patrick McHardyc01003c2007-03-29 11:46:52 -07001775 if (!skb->iif)
1776 skb->iif = skb->dev->ifindex;
David S. Miller86e65da2005-08-09 19:36:29 -07001777
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001778 orig_dev = skb_bond(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001780 if (!orig_dev)
1781 return NET_RX_DROP;
1782
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 __get_cpu_var(netdev_rx_stat).total++;
1784
1785 skb->h.raw = skb->nh.raw = skb->data;
1786 skb->mac_len = skb->nh.raw - skb->mac.raw;
1787
1788 pt_prev = NULL;
1789
1790 rcu_read_lock();
1791
1792#ifdef CONFIG_NET_CLS_ACT
1793 if (skb->tc_verd & TC_NCLS) {
1794 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1795 goto ncls;
1796 }
1797#endif
1798
1799 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1800 if (!ptype->dev || ptype->dev == skb->dev) {
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001801 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001802 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 pt_prev = ptype;
1804 }
1805 }
1806
1807#ifdef CONFIG_NET_CLS_ACT
1808 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001809 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 pt_prev = NULL; /* noone else should process this after*/
1811 } else {
1812 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1813 }
1814
1815 ret = ing_filter(skb);
1816
1817 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1818 kfree_skb(skb);
1819 goto out;
1820 }
1821
1822 skb->tc_verd = 0;
1823ncls:
1824#endif
1825
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001826 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 goto out;
1828
1829 type = skb->protocol;
1830 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1831 if (ptype->type == type &&
1832 (!ptype->dev || ptype->dev == skb->dev)) {
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001833 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001834 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 pt_prev = ptype;
1836 }
1837 }
1838
1839 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001840 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 } else {
1842 kfree_skb(skb);
1843 /* Jamal, now you will not able to escape explaining
1844 * me how you were going to use this. :-)
1845 */
1846 ret = NET_RX_DROP;
1847 }
1848
1849out:
1850 rcu_read_unlock();
1851 return ret;
1852}
1853
1854static int process_backlog(struct net_device *backlog_dev, int *budget)
1855{
1856 int work = 0;
1857 int quota = min(backlog_dev->quota, *budget);
1858 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1859 unsigned long start_time = jiffies;
1860
Stephen Hemmingere3876602005-06-08 14:56:01 -07001861 backlog_dev->weight = weight_p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 for (;;) {
1863 struct sk_buff *skb;
1864 struct net_device *dev;
1865
1866 local_irq_disable();
1867 skb = __skb_dequeue(&queue->input_pkt_queue);
1868 if (!skb)
1869 goto job_done;
1870 local_irq_enable();
1871
1872 dev = skb->dev;
1873
1874 netif_receive_skb(skb);
1875
1876 dev_put(dev);
1877
1878 work++;
1879
1880 if (work >= quota || jiffies - start_time > 1)
1881 break;
1882
1883 }
1884
1885 backlog_dev->quota -= work;
1886 *budget -= work;
1887 return -1;
1888
1889job_done:
1890 backlog_dev->quota -= work;
1891 *budget -= work;
1892
1893 list_del(&backlog_dev->poll_list);
1894 smp_mb__before_clear_bit();
1895 netif_poll_enable(backlog_dev);
1896
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 local_irq_enable();
1898 return 0;
1899}
1900
1901static void net_rx_action(struct softirq_action *h)
1902{
1903 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1904 unsigned long start_time = jiffies;
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001905 int budget = netdev_budget;
Matt Mackall53fb95d2005-08-11 19:27:43 -07001906 void *have;
1907
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 local_irq_disable();
1909
1910 while (!list_empty(&queue->poll_list)) {
1911 struct net_device *dev;
1912
1913 if (budget <= 0 || jiffies - start_time > 1)
1914 goto softnet_break;
1915
1916 local_irq_enable();
1917
1918 dev = list_entry(queue->poll_list.next,
1919 struct net_device, poll_list);
Matt Mackall53fb95d2005-08-11 19:27:43 -07001920 have = netpoll_poll_lock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
1922 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
Matt Mackall53fb95d2005-08-11 19:27:43 -07001923 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 local_irq_disable();
Stephen Hemminger8aca8a22006-03-20 22:26:39 -08001925 list_move_tail(&dev->poll_list, &queue->poll_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 if (dev->quota < 0)
1927 dev->quota += dev->weight;
1928 else
1929 dev->quota = dev->weight;
1930 } else {
Matt Mackall53fb95d2005-08-11 19:27:43 -07001931 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 dev_put(dev);
1933 local_irq_disable();
1934 }
1935 }
1936out:
Chris Leechdb217332006-06-17 21:24:58 -07001937#ifdef CONFIG_NET_DMA
1938 /*
1939 * There may not be any more sk_buffs coming right now, so push
1940 * any pending DMA copies to hardware
1941 */
1942 if (net_dma_client) {
1943 struct dma_chan *chan;
1944 rcu_read_lock();
1945 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1946 dma_async_memcpy_issue_pending(chan);
1947 rcu_read_unlock();
1948 }
1949#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 local_irq_enable();
1951 return;
1952
1953softnet_break:
1954 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1955 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1956 goto out;
1957}
1958
1959static gifconf_func_t * gifconf_list [NPROTO];
1960
1961/**
1962 * register_gifconf - register a SIOCGIF handler
1963 * @family: Address family
1964 * @gifconf: Function handler
1965 *
1966 * Register protocol dependent address dumping routines. The handler
1967 * that is passed must not be freed or reused until it has been replaced
1968 * by another handler.
1969 */
1970int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1971{
1972 if (family >= NPROTO)
1973 return -EINVAL;
1974 gifconf_list[family] = gifconf;
1975 return 0;
1976}
1977
1978
1979/*
1980 * Map an interface index to its name (SIOCGIFNAME)
1981 */
1982
1983/*
1984 * We need this ioctl for efficient implementation of the
1985 * if_indextoname() function required by the IPv6 API. Without
1986 * it, we would have to search all the interfaces to find a
1987 * match. --pb
1988 */
1989
1990static int dev_ifname(struct ifreq __user *arg)
1991{
1992 struct net_device *dev;
1993 struct ifreq ifr;
1994
1995 /*
1996 * Fetch the caller's info block.
1997 */
1998
1999 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2000 return -EFAULT;
2001
2002 read_lock(&dev_base_lock);
2003 dev = __dev_get_by_index(ifr.ifr_ifindex);
2004 if (!dev) {
2005 read_unlock(&dev_base_lock);
2006 return -ENODEV;
2007 }
2008
2009 strcpy(ifr.ifr_name, dev->name);
2010 read_unlock(&dev_base_lock);
2011
2012 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2013 return -EFAULT;
2014 return 0;
2015}
2016
2017/*
2018 * Perform a SIOCGIFCONF call. This structure will change
2019 * size eventually, and there is nothing I can do about it.
2020 * Thus we will need a 'compatibility mode'.
2021 */
2022
2023static int dev_ifconf(char __user *arg)
2024{
2025 struct ifconf ifc;
2026 struct net_device *dev;
2027 char __user *pos;
2028 int len;
2029 int total;
2030 int i;
2031
2032 /*
2033 * Fetch the caller's info block.
2034 */
2035
2036 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2037 return -EFAULT;
2038
2039 pos = ifc.ifc_buf;
2040 len = ifc.ifc_len;
2041
2042 /*
2043 * Loop over the interfaces, and write an info block for each.
2044 */
2045
2046 total = 0;
2047 for (dev = dev_base; dev; dev = dev->next) {
2048 for (i = 0; i < NPROTO; i++) {
2049 if (gifconf_list[i]) {
2050 int done;
2051 if (!pos)
2052 done = gifconf_list[i](dev, NULL, 0);
2053 else
2054 done = gifconf_list[i](dev, pos + total,
2055 len - total);
2056 if (done < 0)
2057 return -EFAULT;
2058 total += done;
2059 }
2060 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002061 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062
2063 /*
2064 * All done. Write the updated control block back to the caller.
2065 */
2066 ifc.ifc_len = total;
2067
2068 /*
2069 * Both BSD and Solaris return 0 here, so we do too.
2070 */
2071 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2072}
2073
2074#ifdef CONFIG_PROC_FS
2075/*
2076 * This is invoked by the /proc filesystem handler to display a device
2077 * in detail.
2078 */
2079static __inline__ struct net_device *dev_get_idx(loff_t pos)
2080{
2081 struct net_device *dev;
2082 loff_t i;
2083
2084 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2085
2086 return i == pos ? dev : NULL;
2087}
2088
2089void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2090{
2091 read_lock(&dev_base_lock);
2092 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2093}
2094
2095void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2096{
2097 ++*pos;
2098 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2099}
2100
2101void dev_seq_stop(struct seq_file *seq, void *v)
2102{
2103 read_unlock(&dev_base_lock);
2104}
2105
2106static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2107{
2108 if (dev->get_stats) {
2109 struct net_device_stats *stats = dev->get_stats(dev);
2110
2111 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2112 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2113 dev->name, stats->rx_bytes, stats->rx_packets,
2114 stats->rx_errors,
2115 stats->rx_dropped + stats->rx_missed_errors,
2116 stats->rx_fifo_errors,
2117 stats->rx_length_errors + stats->rx_over_errors +
2118 stats->rx_crc_errors + stats->rx_frame_errors,
2119 stats->rx_compressed, stats->multicast,
2120 stats->tx_bytes, stats->tx_packets,
2121 stats->tx_errors, stats->tx_dropped,
2122 stats->tx_fifo_errors, stats->collisions,
2123 stats->tx_carrier_errors +
2124 stats->tx_aborted_errors +
2125 stats->tx_window_errors +
2126 stats->tx_heartbeat_errors,
2127 stats->tx_compressed);
2128 } else
2129 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2130}
2131
2132/*
2133 * Called from the PROCfs module. This now uses the new arbitrary sized
2134 * /proc/net interface to create /proc/net/dev
2135 */
2136static int dev_seq_show(struct seq_file *seq, void *v)
2137{
2138 if (v == SEQ_START_TOKEN)
2139 seq_puts(seq, "Inter-| Receive "
2140 " | Transmit\n"
2141 " face |bytes packets errs drop fifo frame "
2142 "compressed multicast|bytes packets errs "
2143 "drop fifo colls carrier compressed\n");
2144 else
2145 dev_seq_printf_stats(seq, v);
2146 return 0;
2147}
2148
2149static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2150{
2151 struct netif_rx_stats *rc = NULL;
2152
2153 while (*pos < NR_CPUS)
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002154 if (cpu_online(*pos)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 rc = &per_cpu(netdev_rx_stat, *pos);
2156 break;
2157 } else
2158 ++*pos;
2159 return rc;
2160}
2161
2162static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2163{
2164 return softnet_get_online(pos);
2165}
2166
2167static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2168{
2169 ++*pos;
2170 return softnet_get_online(pos);
2171}
2172
2173static void softnet_seq_stop(struct seq_file *seq, void *v)
2174{
2175}
2176
2177static int softnet_seq_show(struct seq_file *seq, void *v)
2178{
2179 struct netif_rx_stats *s = v;
2180
2181 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
Stephen Hemminger31aa02c2005-06-23 20:12:48 -07002182 s->total, s->dropped, s->time_squeeze, 0,
Stephen Hemmingerc1ebcdb2005-06-23 20:08:59 -07002183 0, 0, 0, 0, /* was fastroute */
2184 s->cpu_collision );
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 return 0;
2186}
2187
2188static struct seq_operations dev_seq_ops = {
2189 .start = dev_seq_start,
2190 .next = dev_seq_next,
2191 .stop = dev_seq_stop,
2192 .show = dev_seq_show,
2193};
2194
2195static int dev_seq_open(struct inode *inode, struct file *file)
2196{
2197 return seq_open(file, &dev_seq_ops);
2198}
2199
Arjan van de Ven9a321442007-02-12 00:55:35 -08002200static const struct file_operations dev_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 .owner = THIS_MODULE,
2202 .open = dev_seq_open,
2203 .read = seq_read,
2204 .llseek = seq_lseek,
2205 .release = seq_release,
2206};
2207
2208static struct seq_operations softnet_seq_ops = {
2209 .start = softnet_seq_start,
2210 .next = softnet_seq_next,
2211 .stop = softnet_seq_stop,
2212 .show = softnet_seq_show,
2213};
2214
2215static int softnet_seq_open(struct inode *inode, struct file *file)
2216{
2217 return seq_open(file, &softnet_seq_ops);
2218}
2219
Arjan van de Ven9a321442007-02-12 00:55:35 -08002220static const struct file_operations softnet_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 .owner = THIS_MODULE,
2222 .open = softnet_seq_open,
2223 .read = seq_read,
2224 .llseek = seq_lseek,
2225 .release = seq_release,
2226};
2227
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002228#ifdef CONFIG_WIRELESS_EXT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229extern int wireless_proc_init(void);
2230#else
2231#define wireless_proc_init() 0
2232#endif
2233
2234static int __init dev_proc_init(void)
2235{
2236 int rc = -ENOMEM;
2237
2238 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2239 goto out;
2240 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2241 goto out_dev;
2242 if (wireless_proc_init())
2243 goto out_softnet;
2244 rc = 0;
2245out:
2246 return rc;
2247out_softnet:
2248 proc_net_remove("softnet_stat");
2249out_dev:
2250 proc_net_remove("dev");
2251 goto out;
2252}
2253#else
2254#define dev_proc_init() 0
2255#endif /* CONFIG_PROC_FS */
2256
2257
2258/**
2259 * netdev_set_master - set up master/slave pair
2260 * @slave: slave device
2261 * @master: new master device
2262 *
2263 * Changes the master device of the slave. Pass %NULL to break the
2264 * bonding. The caller must hold the RTNL semaphore. On a failure
2265 * a negative errno code is returned. On success the reference counts
2266 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2267 * function returns zero.
2268 */
2269int netdev_set_master(struct net_device *slave, struct net_device *master)
2270{
2271 struct net_device *old = slave->master;
2272
2273 ASSERT_RTNL();
2274
2275 if (master) {
2276 if (old)
2277 return -EBUSY;
2278 dev_hold(master);
2279 }
2280
2281 slave->master = master;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002282
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 synchronize_net();
2284
2285 if (old)
2286 dev_put(old);
2287
2288 if (master)
2289 slave->flags |= IFF_SLAVE;
2290 else
2291 slave->flags &= ~IFF_SLAVE;
2292
2293 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2294 return 0;
2295}
2296
2297/**
2298 * dev_set_promiscuity - update promiscuity count on a device
2299 * @dev: device
2300 * @inc: modifier
2301 *
Stephen Hemminger3041a062006-05-26 13:25:24 -07002302 * Add or remove promiscuity from a device. While the count in the device
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 * remains above zero the interface remains promiscuous. Once it hits zero
2304 * the device reverts back to normal filtering operation. A negative inc
2305 * value is used to drop promiscuity on the device.
2306 */
2307void dev_set_promiscuity(struct net_device *dev, int inc)
2308{
2309 unsigned short old_flags = dev->flags;
2310
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 if ((dev->promiscuity += inc) == 0)
2312 dev->flags &= ~IFF_PROMISC;
David Chau52609c02005-07-05 15:11:06 -07002313 else
2314 dev->flags |= IFF_PROMISC;
2315 if (dev->flags != old_flags) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 dev_mc_upload(dev);
2317 printk(KERN_INFO "device %s %s promiscuous mode\n",
2318 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002319 "left");
Steve Grubb5bdb9882005-12-03 08:39:35 -05002320 audit_log(current->audit_context, GFP_ATOMIC,
2321 AUDIT_ANOM_PROMISCUOUS,
2322 "dev=%s prom=%d old_prom=%d auid=%u",
2323 dev->name, (dev->flags & IFF_PROMISC),
2324 (old_flags & IFF_PROMISC),
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002325 audit_get_loginuid(current->audit_context));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 }
2327}
2328
2329/**
2330 * dev_set_allmulti - update allmulti count on a device
2331 * @dev: device
2332 * @inc: modifier
2333 *
2334 * Add or remove reception of all multicast frames to a device. While the
2335 * count in the device remains above zero the interface remains listening
2336 * to all interfaces. Once it hits zero the device reverts back to normal
2337 * filtering operation. A negative @inc value is used to drop the counter
2338 * when releasing a resource needing all multicasts.
2339 */
2340
2341void dev_set_allmulti(struct net_device *dev, int inc)
2342{
2343 unsigned short old_flags = dev->flags;
2344
2345 dev->flags |= IFF_ALLMULTI;
2346 if ((dev->allmulti += inc) == 0)
2347 dev->flags &= ~IFF_ALLMULTI;
2348 if (dev->flags ^ old_flags)
2349 dev_mc_upload(dev);
2350}
2351
2352unsigned dev_get_flags(const struct net_device *dev)
2353{
2354 unsigned flags;
2355
2356 flags = (dev->flags & ~(IFF_PROMISC |
2357 IFF_ALLMULTI |
Stefan Rompfb00055a2006-03-20 17:09:11 -08002358 IFF_RUNNING |
2359 IFF_LOWER_UP |
2360 IFF_DORMANT)) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 (dev->gflags & (IFF_PROMISC |
2362 IFF_ALLMULTI));
2363
Stefan Rompfb00055a2006-03-20 17:09:11 -08002364 if (netif_running(dev)) {
2365 if (netif_oper_up(dev))
2366 flags |= IFF_RUNNING;
2367 if (netif_carrier_ok(dev))
2368 flags |= IFF_LOWER_UP;
2369 if (netif_dormant(dev))
2370 flags |= IFF_DORMANT;
2371 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
2373 return flags;
2374}
2375
2376int dev_change_flags(struct net_device *dev, unsigned flags)
2377{
2378 int ret;
2379 int old_flags = dev->flags;
2380
2381 /*
2382 * Set the flags on our device.
2383 */
2384
2385 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2386 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2387 IFF_AUTOMEDIA)) |
2388 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2389 IFF_ALLMULTI));
2390
2391 /*
2392 * Load in the correct multicast list now the flags have changed.
2393 */
2394
2395 dev_mc_upload(dev);
2396
2397 /*
2398 * Have we downed the interface. We handle IFF_UP ourselves
2399 * according to user attempts to set it, rather than blindly
2400 * setting it.
2401 */
2402
2403 ret = 0;
2404 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2405 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2406
2407 if (!ret)
2408 dev_mc_upload(dev);
2409 }
2410
2411 if (dev->flags & IFF_UP &&
2412 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2413 IFF_VOLATILE)))
Alan Sternf07d5b92006-05-09 15:23:03 -07002414 raw_notifier_call_chain(&netdev_chain,
Alan Sterne041c682006-03-27 01:16:30 -08002415 NETDEV_CHANGE, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
2417 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2418 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2419 dev->gflags ^= IFF_PROMISC;
2420 dev_set_promiscuity(dev, inc);
2421 }
2422
2423 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2424 is important. Some (broken) drivers set IFF_PROMISC, when
2425 IFF_ALLMULTI is requested not asking us and not reporting.
2426 */
2427 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2428 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2429 dev->gflags ^= IFF_ALLMULTI;
2430 dev_set_allmulti(dev, inc);
2431 }
2432
2433 if (old_flags ^ dev->flags)
2434 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2435
2436 return ret;
2437}
2438
2439int dev_set_mtu(struct net_device *dev, int new_mtu)
2440{
2441 int err;
2442
2443 if (new_mtu == dev->mtu)
2444 return 0;
2445
2446 /* MTU must be positive. */
2447 if (new_mtu < 0)
2448 return -EINVAL;
2449
2450 if (!netif_device_present(dev))
2451 return -ENODEV;
2452
2453 err = 0;
2454 if (dev->change_mtu)
2455 err = dev->change_mtu(dev, new_mtu);
2456 else
2457 dev->mtu = new_mtu;
2458 if (!err && dev->flags & IFF_UP)
Alan Sternf07d5b92006-05-09 15:23:03 -07002459 raw_notifier_call_chain(&netdev_chain,
Alan Sterne041c682006-03-27 01:16:30 -08002460 NETDEV_CHANGEMTU, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 return err;
2462}
2463
2464int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2465{
2466 int err;
2467
2468 if (!dev->set_mac_address)
2469 return -EOPNOTSUPP;
2470 if (sa->sa_family != dev->type)
2471 return -EINVAL;
2472 if (!netif_device_present(dev))
2473 return -ENODEV;
2474 err = dev->set_mac_address(dev, sa);
2475 if (!err)
Alan Sternf07d5b92006-05-09 15:23:03 -07002476 raw_notifier_call_chain(&netdev_chain,
Alan Sterne041c682006-03-27 01:16:30 -08002477 NETDEV_CHANGEADDR, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 return err;
2479}
2480
2481/*
2482 * Perform the SIOCxIFxxx calls.
2483 */
2484static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2485{
2486 int err;
2487 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2488
2489 if (!dev)
2490 return -ENODEV;
2491
2492 switch (cmd) {
2493 case SIOCGIFFLAGS: /* Get interface flags */
2494 ifr->ifr_flags = dev_get_flags(dev);
2495 return 0;
2496
2497 case SIOCSIFFLAGS: /* Set interface flags */
2498 return dev_change_flags(dev, ifr->ifr_flags);
2499
2500 case SIOCGIFMETRIC: /* Get the metric on the interface
2501 (currently unused) */
2502 ifr->ifr_metric = 0;
2503 return 0;
2504
2505 case SIOCSIFMETRIC: /* Set the metric on the interface
2506 (currently unused) */
2507 return -EOPNOTSUPP;
2508
2509 case SIOCGIFMTU: /* Get the MTU of a device */
2510 ifr->ifr_mtu = dev->mtu;
2511 return 0;
2512
2513 case SIOCSIFMTU: /* Set the MTU of a device */
2514 return dev_set_mtu(dev, ifr->ifr_mtu);
2515
2516 case SIOCGIFHWADDR:
2517 if (!dev->addr_len)
2518 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2519 else
2520 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2521 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2522 ifr->ifr_hwaddr.sa_family = dev->type;
2523 return 0;
2524
2525 case SIOCSIFHWADDR:
2526 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2527
2528 case SIOCSIFHWBROADCAST:
2529 if (ifr->ifr_hwaddr.sa_family != dev->type)
2530 return -EINVAL;
2531 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2532 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
Alan Sternf07d5b92006-05-09 15:23:03 -07002533 raw_notifier_call_chain(&netdev_chain,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 NETDEV_CHANGEADDR, dev);
2535 return 0;
2536
2537 case SIOCGIFMAP:
2538 ifr->ifr_map.mem_start = dev->mem_start;
2539 ifr->ifr_map.mem_end = dev->mem_end;
2540 ifr->ifr_map.base_addr = dev->base_addr;
2541 ifr->ifr_map.irq = dev->irq;
2542 ifr->ifr_map.dma = dev->dma;
2543 ifr->ifr_map.port = dev->if_port;
2544 return 0;
2545
2546 case SIOCSIFMAP:
2547 if (dev->set_config) {
2548 if (!netif_device_present(dev))
2549 return -ENODEV;
2550 return dev->set_config(dev, &ifr->ifr_map);
2551 }
2552 return -EOPNOTSUPP;
2553
2554 case SIOCADDMULTI:
2555 if (!dev->set_multicast_list ||
2556 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2557 return -EINVAL;
2558 if (!netif_device_present(dev))
2559 return -ENODEV;
2560 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2561 dev->addr_len, 1);
2562
2563 case SIOCDELMULTI:
2564 if (!dev->set_multicast_list ||
2565 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2566 return -EINVAL;
2567 if (!netif_device_present(dev))
2568 return -ENODEV;
2569 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2570 dev->addr_len, 1);
2571
2572 case SIOCGIFINDEX:
2573 ifr->ifr_ifindex = dev->ifindex;
2574 return 0;
2575
2576 case SIOCGIFTXQLEN:
2577 ifr->ifr_qlen = dev->tx_queue_len;
2578 return 0;
2579
2580 case SIOCSIFTXQLEN:
2581 if (ifr->ifr_qlen < 0)
2582 return -EINVAL;
2583 dev->tx_queue_len = ifr->ifr_qlen;
2584 return 0;
2585
2586 case SIOCSIFNAME:
2587 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2588 return dev_change_name(dev, ifr->ifr_newname);
2589
2590 /*
2591 * Unknown or private ioctl
2592 */
2593
2594 default:
2595 if ((cmd >= SIOCDEVPRIVATE &&
2596 cmd <= SIOCDEVPRIVATE + 15) ||
2597 cmd == SIOCBONDENSLAVE ||
2598 cmd == SIOCBONDRELEASE ||
2599 cmd == SIOCBONDSETHWADDR ||
2600 cmd == SIOCBONDSLAVEINFOQUERY ||
2601 cmd == SIOCBONDINFOQUERY ||
2602 cmd == SIOCBONDCHANGEACTIVE ||
2603 cmd == SIOCGMIIPHY ||
2604 cmd == SIOCGMIIREG ||
2605 cmd == SIOCSMIIREG ||
2606 cmd == SIOCBRADDIF ||
2607 cmd == SIOCBRDELIF ||
2608 cmd == SIOCWANDEV) {
2609 err = -EOPNOTSUPP;
2610 if (dev->do_ioctl) {
2611 if (netif_device_present(dev))
2612 err = dev->do_ioctl(dev, ifr,
2613 cmd);
2614 else
2615 err = -ENODEV;
2616 }
2617 } else
2618 err = -EINVAL;
2619
2620 }
2621 return err;
2622}
2623
2624/*
2625 * This function handles all "interface"-type I/O control requests. The actual
2626 * 'doing' part of this is dev_ifsioc above.
2627 */
2628
2629/**
2630 * dev_ioctl - network device ioctl
2631 * @cmd: command to issue
2632 * @arg: pointer to a struct ifreq in user space
2633 *
2634 * Issue ioctl functions to devices. This is normally called by the
2635 * user space syscall interfaces but can sometimes be useful for
2636 * other purposes. The return value is the return from the syscall if
2637 * positive or a negative errno code on error.
2638 */
2639
2640int dev_ioctl(unsigned int cmd, void __user *arg)
2641{
2642 struct ifreq ifr;
2643 int ret;
2644 char *colon;
2645
2646 /* One special case: SIOCGIFCONF takes ifconf argument
2647 and requires shared lock, because it sleeps writing
2648 to user space.
2649 */
2650
2651 if (cmd == SIOCGIFCONF) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002652 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 ret = dev_ifconf((char __user *) arg);
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002654 rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 return ret;
2656 }
2657 if (cmd == SIOCGIFNAME)
2658 return dev_ifname((struct ifreq __user *)arg);
2659
2660 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2661 return -EFAULT;
2662
2663 ifr.ifr_name[IFNAMSIZ-1] = 0;
2664
2665 colon = strchr(ifr.ifr_name, ':');
2666 if (colon)
2667 *colon = 0;
2668
2669 /*
2670 * See which interface the caller is talking about.
2671 */
2672
2673 switch (cmd) {
2674 /*
2675 * These ioctl calls:
2676 * - can be done by all.
2677 * - atomic and do not require locking.
2678 * - return a value
2679 */
2680 case SIOCGIFFLAGS:
2681 case SIOCGIFMETRIC:
2682 case SIOCGIFMTU:
2683 case SIOCGIFHWADDR:
2684 case SIOCGIFSLAVE:
2685 case SIOCGIFMAP:
2686 case SIOCGIFINDEX:
2687 case SIOCGIFTXQLEN:
2688 dev_load(ifr.ifr_name);
2689 read_lock(&dev_base_lock);
2690 ret = dev_ifsioc(&ifr, cmd);
2691 read_unlock(&dev_base_lock);
2692 if (!ret) {
2693 if (colon)
2694 *colon = ':';
2695 if (copy_to_user(arg, &ifr,
2696 sizeof(struct ifreq)))
2697 ret = -EFAULT;
2698 }
2699 return ret;
2700
2701 case SIOCETHTOOL:
2702 dev_load(ifr.ifr_name);
2703 rtnl_lock();
2704 ret = dev_ethtool(&ifr);
2705 rtnl_unlock();
2706 if (!ret) {
2707 if (colon)
2708 *colon = ':';
2709 if (copy_to_user(arg, &ifr,
2710 sizeof(struct ifreq)))
2711 ret = -EFAULT;
2712 }
2713 return ret;
2714
2715 /*
2716 * These ioctl calls:
2717 * - require superuser power.
2718 * - require strict serialization.
2719 * - return a value
2720 */
2721 case SIOCGMIIPHY:
2722 case SIOCGMIIREG:
2723 case SIOCSIFNAME:
2724 if (!capable(CAP_NET_ADMIN))
2725 return -EPERM;
2726 dev_load(ifr.ifr_name);
2727 rtnl_lock();
2728 ret = dev_ifsioc(&ifr, cmd);
2729 rtnl_unlock();
2730 if (!ret) {
2731 if (colon)
2732 *colon = ':';
2733 if (copy_to_user(arg, &ifr,
2734 sizeof(struct ifreq)))
2735 ret = -EFAULT;
2736 }
2737 return ret;
2738
2739 /*
2740 * These ioctl calls:
2741 * - require superuser power.
2742 * - require strict serialization.
2743 * - do not return a value
2744 */
2745 case SIOCSIFFLAGS:
2746 case SIOCSIFMETRIC:
2747 case SIOCSIFMTU:
2748 case SIOCSIFMAP:
2749 case SIOCSIFHWADDR:
2750 case SIOCSIFSLAVE:
2751 case SIOCADDMULTI:
2752 case SIOCDELMULTI:
2753 case SIOCSIFHWBROADCAST:
2754 case SIOCSIFTXQLEN:
2755 case SIOCSMIIREG:
2756 case SIOCBONDENSLAVE:
2757 case SIOCBONDRELEASE:
2758 case SIOCBONDSETHWADDR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 case SIOCBONDCHANGEACTIVE:
2760 case SIOCBRADDIF:
2761 case SIOCBRDELIF:
2762 if (!capable(CAP_NET_ADMIN))
2763 return -EPERM;
Thomas Grafcabcac02006-01-24 12:46:33 -08002764 /* fall through */
2765 case SIOCBONDSLAVEINFOQUERY:
2766 case SIOCBONDINFOQUERY:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 dev_load(ifr.ifr_name);
2768 rtnl_lock();
2769 ret = dev_ifsioc(&ifr, cmd);
2770 rtnl_unlock();
2771 return ret;
2772
2773 case SIOCGIFMEM:
2774 /* Get the per device memory space. We can add this but
2775 * currently do not support it */
2776 case SIOCSIFMEM:
2777 /* Set the per device memory buffer space.
2778 * Not applicable in our case */
2779 case SIOCSIFLINK:
2780 return -EINVAL;
2781
2782 /*
2783 * Unknown or private ioctl.
2784 */
2785 default:
2786 if (cmd == SIOCWANDEV ||
2787 (cmd >= SIOCDEVPRIVATE &&
2788 cmd <= SIOCDEVPRIVATE + 15)) {
2789 dev_load(ifr.ifr_name);
2790 rtnl_lock();
2791 ret = dev_ifsioc(&ifr, cmd);
2792 rtnl_unlock();
2793 if (!ret && copy_to_user(arg, &ifr,
2794 sizeof(struct ifreq)))
2795 ret = -EFAULT;
2796 return ret;
2797 }
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002798#ifdef CONFIG_WIRELESS_EXT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 /* Take care of Wireless Extensions */
2800 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2801 /* If command is `set a parameter', or
2802 * `get the encoding parameters', check if
2803 * the user has the right to do it */
Jean Tourrilhesa4170162006-04-04 15:53:43 -07002804 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2805 || cmd == SIOCGIWENCODEEXT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 if (!capable(CAP_NET_ADMIN))
2807 return -EPERM;
2808 }
2809 dev_load(ifr.ifr_name);
2810 rtnl_lock();
2811 /* Follow me in net/core/wireless.c */
2812 ret = wireless_process_ioctl(&ifr, cmd);
2813 rtnl_unlock();
2814 if (IW_IS_GET(cmd) &&
2815 copy_to_user(arg, &ifr,
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002816 sizeof(struct ifreq)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 ret = -EFAULT;
2818 return ret;
2819 }
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002820#endif /* CONFIG_WIRELESS_EXT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 return -EINVAL;
2822 }
2823}
2824
2825
2826/**
2827 * dev_new_index - allocate an ifindex
2828 *
2829 * Returns a suitable unique value for a new device interface
2830 * number. The caller must hold the rtnl semaphore or the
2831 * dev_base_lock to be sure it remains unique.
2832 */
2833static int dev_new_index(void)
2834{
2835 static int ifindex;
2836 for (;;) {
2837 if (++ifindex <= 0)
2838 ifindex = 1;
2839 if (!__dev_get_by_index(ifindex))
2840 return ifindex;
2841 }
2842}
2843
2844static int dev_boot_phase = 1;
2845
2846/* Delayed registration/unregisteration */
2847static DEFINE_SPINLOCK(net_todo_list_lock);
2848static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2849
2850static inline void net_set_todo(struct net_device *dev)
2851{
2852 spin_lock(&net_todo_list_lock);
2853 list_add_tail(&dev->todo_list, &net_todo_list);
2854 spin_unlock(&net_todo_list_lock);
2855}
2856
2857/**
2858 * register_netdevice - register a network device
2859 * @dev: device to register
2860 *
2861 * Take a completed network device structure and add it to the kernel
2862 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2863 * chain. 0 is returned on success. A negative errno code is returned
2864 * on a failure to set up the device, or if the name is a duplicate.
2865 *
2866 * Callers must hold the rtnl semaphore. You may want
2867 * register_netdev() instead of this.
2868 *
2869 * BUGS:
2870 * The locking appears insufficient to guarantee two parallel registers
2871 * will not get the same name.
2872 */
2873
2874int register_netdevice(struct net_device *dev)
2875{
2876 struct hlist_head *head;
2877 struct hlist_node *p;
2878 int ret;
2879
2880 BUG_ON(dev_boot_phase);
2881 ASSERT_RTNL();
2882
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07002883 might_sleep();
2884
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 /* When net_device's are persistent, this will be fatal. */
2886 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2887
2888 spin_lock_init(&dev->queue_lock);
Herbert Xu932ff272006-06-09 12:20:56 -07002889 spin_lock_init(&dev->_xmit_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 dev->xmit_lock_owner = -1;
2891#ifdef CONFIG_NET_CLS_ACT
2892 spin_lock_init(&dev->ingress_lock);
2893#endif
2894
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 dev->iflink = -1;
2896
2897 /* Init, if this function is available */
2898 if (dev->init) {
2899 ret = dev->init(dev);
2900 if (ret) {
2901 if (ret > 0)
2902 ret = -EIO;
Adrian Bunk90833aa2006-11-13 16:02:22 -08002903 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 }
2905 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002906
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 if (!dev_valid_name(dev->name)) {
2908 ret = -EINVAL;
Adrian Bunk90833aa2006-11-13 16:02:22 -08002909 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 }
2911
2912 dev->ifindex = dev_new_index();
2913 if (dev->iflink == -1)
2914 dev->iflink = dev->ifindex;
2915
2916 /* Check for existence of name */
2917 head = dev_name_hash(dev->name);
2918 hlist_for_each(p, head) {
2919 struct net_device *d
2920 = hlist_entry(p, struct net_device, name_hlist);
2921 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2922 ret = -EEXIST;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002923 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002925 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926
2927 /* Fix illegal SG+CSUM combinations. */
2928 if ((dev->features & NETIF_F_SG) &&
Herbert Xu8648b302006-06-17 22:06:05 -07002929 !(dev->features & NETIF_F_ALL_CSUM)) {
Stephen Hemminger5a8da022006-07-07 16:54:05 -07002930 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 dev->name);
2932 dev->features &= ~NETIF_F_SG;
2933 }
2934
2935 /* TSO requires that SG is present as well. */
2936 if ((dev->features & NETIF_F_TSO) &&
2937 !(dev->features & NETIF_F_SG)) {
Stephen Hemminger5a8da022006-07-07 16:54:05 -07002938 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 dev->name);
2940 dev->features &= ~NETIF_F_TSO;
2941 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002942 if (dev->features & NETIF_F_UFO) {
2943 if (!(dev->features & NETIF_F_HW_CSUM)) {
2944 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2945 "NETIF_F_HW_CSUM feature.\n",
2946 dev->name);
2947 dev->features &= ~NETIF_F_UFO;
2948 }
2949 if (!(dev->features & NETIF_F_SG)) {
2950 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2951 "NETIF_F_SG feature.\n",
2952 dev->name);
2953 dev->features &= ~NETIF_F_UFO;
2954 }
2955 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956
2957 /*
2958 * nil rebuild_header routine,
2959 * that should be never called and used as just bug trap.
2960 */
2961
2962 if (!dev->rebuild_header)
2963 dev->rebuild_header = default_rebuild_header;
2964
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07002965 ret = netdev_register_sysfs(dev);
2966 if (ret)
Adrian Bunk90833aa2006-11-13 16:02:22 -08002967 goto out;
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07002968 dev->reg_state = NETREG_REGISTERED;
2969
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970 /*
2971 * Default initial state at registry is that the
2972 * device is present.
2973 */
2974
2975 set_bit(__LINK_STATE_PRESENT, &dev->state);
2976
2977 dev->next = NULL;
2978 dev_init_scheduler(dev);
2979 write_lock_bh(&dev_base_lock);
2980 *dev_tail = dev;
2981 dev_tail = &dev->next;
2982 hlist_add_head(&dev->name_hlist, head);
2983 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2984 dev_hold(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 write_unlock_bh(&dev_base_lock);
2986
2987 /* Notify protocols, that a new device appeared. */
Alan Sternf07d5b92006-05-09 15:23:03 -07002988 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 ret = 0;
2991
2992out:
2993 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994}
2995
2996/**
2997 * register_netdev - register a network device
2998 * @dev: device to register
2999 *
3000 * Take a completed network device structure and add it to the kernel
3001 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3002 * chain. 0 is returned on success. A negative errno code is returned
3003 * on a failure to set up the device, or if the name is a duplicate.
3004 *
3005 * This is a wrapper around register_netdev that takes the rtnl semaphore
3006 * and expands the device name if you passed a format string to
3007 * alloc_netdev.
3008 */
3009int register_netdev(struct net_device *dev)
3010{
3011 int err;
3012
3013 rtnl_lock();
3014
3015 /*
3016 * If the name is a format string the caller wants us to do a
3017 * name allocation.
3018 */
3019 if (strchr(dev->name, '%')) {
3020 err = dev_alloc_name(dev, dev->name);
3021 if (err < 0)
3022 goto out;
3023 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003024
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025 err = register_netdevice(dev);
3026out:
3027 rtnl_unlock();
3028 return err;
3029}
3030EXPORT_SYMBOL(register_netdev);
3031
3032/*
3033 * netdev_wait_allrefs - wait until all references are gone.
3034 *
3035 * This is called when unregistering network devices.
3036 *
3037 * Any protocol or device that holds a reference should register
3038 * for netdevice notification, and cleanup and put back the
3039 * reference if they receive an UNREGISTER event.
3040 * We can get stuck here if buggy protocols don't correctly
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003041 * call dev_put.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 */
3043static void netdev_wait_allrefs(struct net_device *dev)
3044{
3045 unsigned long rebroadcast_time, warning_time;
3046
3047 rebroadcast_time = warning_time = jiffies;
3048 while (atomic_read(&dev->refcnt) != 0) {
3049 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08003050 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051
3052 /* Rebroadcast unregister notification */
Alan Sternf07d5b92006-05-09 15:23:03 -07003053 raw_notifier_call_chain(&netdev_chain,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 NETDEV_UNREGISTER, dev);
3055
3056 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3057 &dev->state)) {
3058 /* We must not have linkwatch events
3059 * pending on unregister. If this
3060 * happens, we simply run the queue
3061 * unscheduled, resulting in a noop
3062 * for this device.
3063 */
3064 linkwatch_run_queue();
3065 }
3066
Stephen Hemminger6756ae42006-03-20 22:23:58 -08003067 __rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068
3069 rebroadcast_time = jiffies;
3070 }
3071
3072 msleep(250);
3073
3074 if (time_after(jiffies, warning_time + 10 * HZ)) {
3075 printk(KERN_EMERG "unregister_netdevice: "
3076 "waiting for %s to become free. Usage "
3077 "count = %d\n",
3078 dev->name, atomic_read(&dev->refcnt));
3079 warning_time = jiffies;
3080 }
3081 }
3082}
3083
3084/* The sequence is:
3085 *
3086 * rtnl_lock();
3087 * ...
3088 * register_netdevice(x1);
3089 * register_netdevice(x2);
3090 * ...
3091 * unregister_netdevice(y1);
3092 * unregister_netdevice(y2);
3093 * ...
3094 * rtnl_unlock();
3095 * free_netdev(y1);
3096 * free_netdev(y2);
3097 *
3098 * We are invoked by rtnl_unlock() after it drops the semaphore.
3099 * This allows us to deal with problems:
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003100 * 1) We can delete sysfs objects which invoke hotplug
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 * without deadlocking with linkwatch via keventd.
3102 * 2) Since we run with the RTNL semaphore not held, we can sleep
3103 * safely in order to wait for the netdev refcnt to drop to zero.
3104 */
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08003105static DEFINE_MUTEX(net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106void netdev_run_todo(void)
3107{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003108 struct list_head list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109
3110 /* Need to guard against multiple cpu's getting out of order. */
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08003111 mutex_lock(&net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112
3113 /* Not safe to do outside the semaphore. We must not return
3114 * until all unregister events invoked by the local processor
3115 * have been completed (either by this todo run, or one on
3116 * another cpu).
3117 */
3118 if (list_empty(&net_todo_list))
3119 goto out;
3120
3121 /* Snapshot list, allow later requests */
3122 spin_lock(&net_todo_list_lock);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003123 list_replace_init(&net_todo_list, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124 spin_unlock(&net_todo_list_lock);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003125
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 while (!list_empty(&list)) {
3127 struct net_device *dev
3128 = list_entry(list.next, struct net_device, todo_list);
3129 list_del(&dev->todo_list);
3130
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003131 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132 printk(KERN_ERR "network todo '%s' but state %d\n",
3133 dev->name, dev->reg_state);
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003134 dump_stack();
3135 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 }
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003137
3138 netdev_unregister_sysfs(dev);
3139 dev->reg_state = NETREG_UNREGISTERED;
3140
3141 netdev_wait_allrefs(dev);
3142
3143 /* paranoia */
3144 BUG_ON(atomic_read(&dev->refcnt));
3145 BUG_TRAP(!dev->ip_ptr);
3146 BUG_TRAP(!dev->ip6_ptr);
3147 BUG_TRAP(!dev->dn_ptr);
3148
3149 /* It must be the very last action,
3150 * after this 'dev' may point to freed up memory.
3151 */
3152 if (dev->destructor)
3153 dev->destructor(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 }
3155
3156out:
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08003157 mutex_unlock(&net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158}
3159
3160/**
3161 * alloc_netdev - allocate network device
3162 * @sizeof_priv: size of private data to allocate space for
3163 * @name: device name format string
3164 * @setup: callback to initialize device
3165 *
3166 * Allocates a struct net_device with private data area for driver use
3167 * and performs basic initialization.
3168 */
3169struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3170 void (*setup)(struct net_device *))
3171{
3172 void *p;
3173 struct net_device *dev;
3174 int alloc_size;
3175
Stephen Hemmingerb6fe17d2006-08-29 17:06:13 -07003176 BUG_ON(strlen(name) >= sizeof(dev->name));
3177
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178 /* ensure 32-byte alignment of both the device and private area */
3179 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3180 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3181
Paolo 'Blaisorblade' Giarrusso31380de2006-04-06 22:38:28 -07003182 p = kzalloc(alloc_size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 if (!p) {
Stephen Hemmingerb6fe17d2006-08-29 17:06:13 -07003184 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185 return NULL;
3186 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187
3188 dev = (struct net_device *)
3189 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3190 dev->padded = (char *)dev - (char *)p;
3191
3192 if (sizeof_priv)
3193 dev->priv = netdev_priv(dev);
3194
3195 setup(dev);
3196 strcpy(dev->name, name);
3197 return dev;
3198}
3199EXPORT_SYMBOL(alloc_netdev);
3200
3201/**
3202 * free_netdev - free network device
3203 * @dev: device
3204 *
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003205 * This function does the last stage of destroying an allocated device
3206 * interface. The reference to the device object is released.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 * If this is the last reference then it will be freed.
3208 */
3209void free_netdev(struct net_device *dev)
3210{
3211#ifdef CONFIG_SYSFS
Stephen Hemminger3041a062006-05-26 13:25:24 -07003212 /* Compatibility with error handling in drivers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213 if (dev->reg_state == NETREG_UNINITIALIZED) {
3214 kfree((char *)dev - dev->padded);
3215 return;
3216 }
3217
3218 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3219 dev->reg_state = NETREG_RELEASED;
3220
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07003221 /* will free via device release */
3222 put_device(&dev->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223#else
3224 kfree((char *)dev - dev->padded);
3225#endif
3226}
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003227
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228/* Synchronize with packet receive processing. */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003229void synchronize_net(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230{
3231 might_sleep();
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07003232 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233}
3234
3235/**
3236 * unregister_netdevice - remove device from the kernel
3237 * @dev: device
3238 *
3239 * This function shuts down a device interface and removes it
3240 * from the kernel tables. On success 0 is returned, on a failure
3241 * a negative errno code is returned.
3242 *
3243 * Callers must hold the rtnl semaphore. You may want
3244 * unregister_netdev() instead of this.
3245 */
3246
Stephen Hemminger22f8cde2007-02-07 00:09:58 -08003247void unregister_netdevice(struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248{
3249 struct net_device *d, **dp;
3250
3251 BUG_ON(dev_boot_phase);
3252 ASSERT_RTNL();
3253
3254 /* Some devices call without registering for initialization unwind. */
3255 if (dev->reg_state == NETREG_UNINITIALIZED) {
3256 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3257 "was registered\n", dev->name, dev);
Stephen Hemminger22f8cde2007-02-07 00:09:58 -08003258
3259 WARN_ON(1);
3260 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261 }
3262
3263 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3264
3265 /* If device is running, close it first. */
3266 if (dev->flags & IFF_UP)
3267 dev_close(dev);
3268
3269 /* And unlink it from device chain. */
3270 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3271 if (d == dev) {
3272 write_lock_bh(&dev_base_lock);
3273 hlist_del(&dev->name_hlist);
3274 hlist_del(&dev->index_hlist);
3275 if (dev_tail == &dev->next)
3276 dev_tail = dp;
3277 *dp = d->next;
3278 write_unlock_bh(&dev_base_lock);
3279 break;
3280 }
3281 }
Stephen Hemminger22f8cde2007-02-07 00:09:58 -08003282 BUG_ON(!d);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283
3284 dev->reg_state = NETREG_UNREGISTERING;
3285
3286 synchronize_net();
3287
3288 /* Shutdown queueing discipline. */
3289 dev_shutdown(dev);
3290
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003291
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292 /* Notify protocols, that we are about to destroy
3293 this device. They should clean all the things.
3294 */
Alan Sternf07d5b92006-05-09 15:23:03 -07003295 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003296
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 /*
3298 * Flush the multicast chain
3299 */
3300 dev_mc_discard(dev);
3301
3302 if (dev->uninit)
3303 dev->uninit(dev);
3304
3305 /* Notifier chain MUST detach us from master device. */
3306 BUG_TRAP(!dev->master);
3307
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 /* Finish processing unregister after unlock */
3309 net_set_todo(dev);
3310
3311 synchronize_net();
3312
3313 dev_put(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314}
3315
3316/**
3317 * unregister_netdev - remove device from the kernel
3318 * @dev: device
3319 *
3320 * This function shuts down a device interface and removes it
3321 * from the kernel tables. On success 0 is returned, on a failure
3322 * a negative errno code is returned.
3323 *
3324 * This is just a wrapper for unregister_netdevice that takes
3325 * the rtnl semaphore. In general you want to use this and not
3326 * unregister_netdevice.
3327 */
3328void unregister_netdev(struct net_device *dev)
3329{
3330 rtnl_lock();
3331 unregister_netdevice(dev);
3332 rtnl_unlock();
3333}
3334
3335EXPORT_SYMBOL(unregister_netdev);
3336
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337static int dev_cpu_callback(struct notifier_block *nfb,
3338 unsigned long action,
3339 void *ocpu)
3340{
3341 struct sk_buff **list_skb;
3342 struct net_device **list_net;
3343 struct sk_buff *skb;
3344 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3345 struct softnet_data *sd, *oldsd;
3346
3347 if (action != CPU_DEAD)
3348 return NOTIFY_OK;
3349
3350 local_irq_disable();
3351 cpu = smp_processor_id();
3352 sd = &per_cpu(softnet_data, cpu);
3353 oldsd = &per_cpu(softnet_data, oldcpu);
3354
3355 /* Find end of our completion_queue. */
3356 list_skb = &sd->completion_queue;
3357 while (*list_skb)
3358 list_skb = &(*list_skb)->next;
3359 /* Append completion queue from offline CPU. */
3360 *list_skb = oldsd->completion_queue;
3361 oldsd->completion_queue = NULL;
3362
3363 /* Find end of our output_queue. */
3364 list_net = &sd->output_queue;
3365 while (*list_net)
3366 list_net = &(*list_net)->next_sched;
3367 /* Append output queue from offline CPU. */
3368 *list_net = oldsd->output_queue;
3369 oldsd->output_queue = NULL;
3370
3371 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3372 local_irq_enable();
3373
3374 /* Process offline CPU's input_pkt_queue */
3375 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3376 netif_rx(skb);
3377
3378 return NOTIFY_OK;
3379}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380
Chris Leechdb217332006-06-17 21:24:58 -07003381#ifdef CONFIG_NET_DMA
3382/**
3383 * net_dma_rebalance -
3384 * This is called when the number of channels allocated to the net_dma_client
3385 * changes. The net_dma_client tries to have one DMA channel per CPU.
3386 */
3387static void net_dma_rebalance(void)
3388{
3389 unsigned int cpu, i, n;
3390 struct dma_chan *chan;
3391
Chris Leechdb217332006-06-17 21:24:58 -07003392 if (net_dma_count == 0) {
3393 for_each_online_cpu(cpu)
Alexey Dobriyan29bbd722006-08-02 15:02:31 -07003394 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
Chris Leechdb217332006-06-17 21:24:58 -07003395 return;
3396 }
3397
3398 i = 0;
3399 cpu = first_cpu(cpu_online_map);
3400
3401 rcu_read_lock();
3402 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3403 n = ((num_online_cpus() / net_dma_count)
3404 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3405
3406 while(n) {
Alexey Dobriyan29bbd722006-08-02 15:02:31 -07003407 per_cpu(softnet_data, cpu).net_dma = chan;
Chris Leechdb217332006-06-17 21:24:58 -07003408 cpu = next_cpu(cpu, cpu_online_map);
3409 n--;
3410 }
3411 i++;
3412 }
3413 rcu_read_unlock();
Chris Leechdb217332006-06-17 21:24:58 -07003414}
3415
3416/**
3417 * netdev_dma_event - event callback for the net_dma_client
3418 * @client: should always be net_dma_client
Randy Dunlapf4b8ea72006-06-22 16:00:11 -07003419 * @chan: DMA channel for the event
3420 * @event: event type
Chris Leechdb217332006-06-17 21:24:58 -07003421 */
3422static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3423 enum dma_event event)
3424{
3425 spin_lock(&net_dma_event_lock);
3426 switch (event) {
3427 case DMA_RESOURCE_ADDED:
3428 net_dma_count++;
3429 net_dma_rebalance();
3430 break;
3431 case DMA_RESOURCE_REMOVED:
3432 net_dma_count--;
3433 net_dma_rebalance();
3434 break;
3435 default:
3436 break;
3437 }
3438 spin_unlock(&net_dma_event_lock);
3439}
3440
3441/**
3442 * netdev_dma_regiser - register the networking subsystem as a DMA client
3443 */
3444static int __init netdev_dma_register(void)
3445{
3446 spin_lock_init(&net_dma_event_lock);
3447 net_dma_client = dma_async_client_register(netdev_dma_event);
3448 if (net_dma_client == NULL)
3449 return -ENOMEM;
3450
3451 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3452 return 0;
3453}
3454
3455#else
3456static int __init netdev_dma_register(void) { return -ENODEV; }
3457#endif /* CONFIG_NET_DMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458
3459/*
3460 * Initialize the DEV module. At boot time this walks the device list and
3461 * unhooks any devices that fail to initialise (normally hardware not
3462 * present) and leaves us with a valid list of present and active devices.
3463 *
3464 */
3465
3466/*
3467 * This is called single threaded during boot, so no need
3468 * to take the rtnl semaphore.
3469 */
3470static int __init net_dev_init(void)
3471{
3472 int i, rc = -ENOMEM;
3473
3474 BUG_ON(!dev_boot_phase);
3475
Linus Torvalds1da177e2005-04-16 15:20:36 -07003476 if (dev_proc_init())
3477 goto out;
3478
3479 if (netdev_sysfs_init())
3480 goto out;
3481
3482 INIT_LIST_HEAD(&ptype_all);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003483 for (i = 0; i < 16; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 INIT_LIST_HEAD(&ptype_base[i]);
3485
3486 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3487 INIT_HLIST_HEAD(&dev_name_head[i]);
3488
3489 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3490 INIT_HLIST_HEAD(&dev_index_head[i]);
3491
3492 /*
3493 * Initialise the packet receive queues.
3494 */
3495
KAMEZAWA Hiroyuki6f912042006-04-10 22:52:50 -07003496 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497 struct softnet_data *queue;
3498
3499 queue = &per_cpu(softnet_data, i);
3500 skb_queue_head_init(&queue->input_pkt_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501 queue->completion_queue = NULL;
3502 INIT_LIST_HEAD(&queue->poll_list);
3503 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3504 queue->backlog_dev.weight = weight_p;
3505 queue->backlog_dev.poll = process_backlog;
3506 atomic_set(&queue->backlog_dev.refcnt, 1);
3507 }
3508
Chris Leechdb217332006-06-17 21:24:58 -07003509 netdev_dma_register();
3510
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511 dev_boot_phase = 0;
3512
3513 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3514 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3515
3516 hotcpu_notifier(dev_cpu_callback, 0);
3517 dst_init();
3518 dev_mcast_init();
3519 rc = 0;
3520out:
3521 return rc;
3522}
3523
3524subsys_initcall(net_dev_init);
3525
3526EXPORT_SYMBOL(__dev_get_by_index);
3527EXPORT_SYMBOL(__dev_get_by_name);
3528EXPORT_SYMBOL(__dev_remove_pack);
Mitch Williamsc2373ee2005-11-09 10:34:45 -08003529EXPORT_SYMBOL(dev_valid_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530EXPORT_SYMBOL(dev_add_pack);
3531EXPORT_SYMBOL(dev_alloc_name);
3532EXPORT_SYMBOL(dev_close);
3533EXPORT_SYMBOL(dev_get_by_flags);
3534EXPORT_SYMBOL(dev_get_by_index);
3535EXPORT_SYMBOL(dev_get_by_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536EXPORT_SYMBOL(dev_open);
3537EXPORT_SYMBOL(dev_queue_xmit);
3538EXPORT_SYMBOL(dev_remove_pack);
3539EXPORT_SYMBOL(dev_set_allmulti);
3540EXPORT_SYMBOL(dev_set_promiscuity);
3541EXPORT_SYMBOL(dev_change_flags);
3542EXPORT_SYMBOL(dev_set_mtu);
3543EXPORT_SYMBOL(dev_set_mac_address);
3544EXPORT_SYMBOL(free_netdev);
3545EXPORT_SYMBOL(netdev_boot_setup_check);
3546EXPORT_SYMBOL(netdev_set_master);
3547EXPORT_SYMBOL(netdev_state_change);
3548EXPORT_SYMBOL(netif_receive_skb);
3549EXPORT_SYMBOL(netif_rx);
3550EXPORT_SYMBOL(register_gifconf);
3551EXPORT_SYMBOL(register_netdevice);
3552EXPORT_SYMBOL(register_netdevice_notifier);
3553EXPORT_SYMBOL(skb_checksum_help);
3554EXPORT_SYMBOL(synchronize_net);
3555EXPORT_SYMBOL(unregister_netdevice);
3556EXPORT_SYMBOL(unregister_netdevice_notifier);
3557EXPORT_SYMBOL(net_enable_timestamp);
3558EXPORT_SYMBOL(net_disable_timestamp);
3559EXPORT_SYMBOL(dev_get_flags);
3560
3561#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3562EXPORT_SYMBOL(br_handle_frame_hook);
3563EXPORT_SYMBOL(br_fdb_get_hook);
3564EXPORT_SYMBOL(br_fdb_put_hook);
3565#endif
3566
3567#ifdef CONFIG_KMOD
3568EXPORT_SYMBOL(dev_load);
3569#endif
3570
3571EXPORT_PER_CPU_SYMBOL(softnet_data);