blob: 8e1dc3051222f23e06b70025f006bb6256ab5194 [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/config.h>
80#include <linux/cpu.h>
81#include <linux/types.h>
82#include <linux/kernel.h>
83#include <linux/sched.h>
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -080084#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070085#include <linux/string.h>
86#include <linux/mm.h>
87#include <linux/socket.h>
88#include <linux/sockios.h>
89#include <linux/errno.h>
90#include <linux/interrupt.h>
91#include <linux/if_ether.h>
92#include <linux/netdevice.h>
93#include <linux/etherdevice.h>
94#include <linux/notifier.h>
95#include <linux/skbuff.h>
96#include <net/sock.h>
97#include <linux/rtnetlink.h>
98#include <linux/proc_fs.h>
99#include <linux/seq_file.h>
100#include <linux/stat.h>
101#include <linux/if_bridge.h>
102#include <linux/divert.h>
103#include <net/dst.h>
104#include <net/pkt_sched.h>
105#include <net/checksum.h>
106#include <linux/highmem.h>
107#include <linux/init.h>
108#include <linux/kmod.h>
109#include <linux/module.h>
110#include <linux/kallsyms.h>
111#include <linux/netpoll.h>
112#include <linux/rcupdate.h>
113#include <linux/delay.h>
Adrian Bunkd86b5e02006-01-21 00:46:55 +0100114#include <linux/wireless.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115#include <net/iw_handler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116#include <asm/current.h>
Steve Grubb5bdb9882005-12-03 08:39:35 -0500117#include <linux/audit.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119/*
120 * The list of packet types we will receive (as opposed to discard)
121 * and the routines to invoke.
122 *
123 * Why 16. Because with 16 the only overlap we get on a hash of the
124 * low nibble of the protocol value is RARP/SNAP/X.25.
125 *
126 * NOTE: That is no longer true with the addition of VLAN tags. Not
127 * sure which should go first, but I bet it won't make much
128 * difference if we are running VLANs. The good news is that
129 * this protocol won't be in the list unless compiled in, so
130 * the average user (w/out VLANs) will not be adversly affected.
131 * --BLG
132 *
133 * 0800 IP
134 * 8100 802.1Q VLAN
135 * 0001 802.3
136 * 0002 AX.25
137 * 0004 802.2
138 * 8035 RARP
139 * 0005 SNAP
140 * 0805 X.25
141 * 0806 ARP
142 * 8137 IPX
143 * 0009 Localtalk
144 * 86DD IPv6
145 */
146
147static DEFINE_SPINLOCK(ptype_lock);
148static struct list_head ptype_base[16]; /* 16 way hashed list */
149static struct list_head ptype_all; /* Taps */
150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151/*
152 * The @dev_base list is protected by @dev_base_lock and the rtln
153 * semaphore.
154 *
155 * Pure readers hold dev_base_lock for reading.
156 *
157 * Writers must hold the rtnl semaphore while they loop through the
158 * dev_base list, and hold dev_base_lock for writing when they do the
159 * actual updates. This allows pure readers to access the list even
160 * while a writer is preparing to update it.
161 *
162 * To put it another way, dev_base_lock is held for writing only to
163 * protect against pure readers; the rtnl semaphore provides the
164 * protection against other writers.
165 *
166 * See, for example usages, register_netdevice() and
167 * unregister_netdevice(), which must be called with the rtnl
168 * semaphore held.
169 */
170struct net_device *dev_base;
171static struct net_device **dev_tail = &dev_base;
172DEFINE_RWLOCK(dev_base_lock);
173
174EXPORT_SYMBOL(dev_base);
175EXPORT_SYMBOL(dev_base_lock);
176
177#define NETDEV_HASHBITS 8
178static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
179static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
180
181static inline struct hlist_head *dev_name_hash(const char *name)
182{
183 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
184 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
185}
186
187static inline struct hlist_head *dev_index_hash(int ifindex)
188{
189 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
190}
191
192/*
193 * Our notifier list
194 */
195
196static struct notifier_block *netdev_chain;
197
198/*
199 * Device drivers call our routines to queue packets here. We empty the
200 * queue in the local softnet handler.
201 */
Stephen Hemminger31aa02c2005-06-23 20:12:48 -0700202DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
204#ifdef CONFIG_SYSFS
205extern int netdev_sysfs_init(void);
206extern int netdev_register_sysfs(struct net_device *);
207extern void netdev_unregister_sysfs(struct net_device *);
208#else
209#define netdev_sysfs_init() (0)
210#define netdev_register_sysfs(dev) (0)
211#define netdev_unregister_sysfs(dev) do { } while(0)
212#endif
213
214
215/*******************************************************************************
216
217 Protocol management and registration routines
218
219*******************************************************************************/
220
221/*
222 * For efficiency
223 */
224
225int netdev_nit;
226
227/*
228 * Add a protocol ID to the list. Now that the input handler is
229 * smarter we can dispense with all the messy stuff that used to be
230 * here.
231 *
232 * BEWARE!!! Protocol handlers, mangling input packets,
233 * MUST BE last in hash buckets and checking protocol handlers
234 * MUST start from promiscuous ptype_all chain in net_bh.
235 * It is true now, do not change it.
236 * Explanation follows: if protocol handler, mangling packet, will
237 * be the first on list, it is not able to sense, that packet
238 * is cloned and should be copied-on-write, so that it will
239 * change it and subsequent readers will get broken packet.
240 * --ANK (980803)
241 */
242
243/**
244 * dev_add_pack - add packet handler
245 * @pt: packet type declaration
246 *
247 * Add a protocol handler to the networking stack. The passed &packet_type
248 * is linked into kernel lists and may not be freed until it has been
249 * removed from the kernel lists.
250 *
251 * This call does not sleep therefore it can not
252 * guarantee all CPU's that are in middle of receiving packets
253 * will see the new packet type (until the next received packet).
254 */
255
256void dev_add_pack(struct packet_type *pt)
257{
258 int hash;
259
260 spin_lock_bh(&ptype_lock);
261 if (pt->type == htons(ETH_P_ALL)) {
262 netdev_nit++;
263 list_add_rcu(&pt->list, &ptype_all);
264 } else {
265 hash = ntohs(pt->type) & 15;
266 list_add_rcu(&pt->list, &ptype_base[hash]);
267 }
268 spin_unlock_bh(&ptype_lock);
269}
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271/**
272 * __dev_remove_pack - remove packet handler
273 * @pt: packet type declaration
274 *
275 * Remove a protocol handler that was previously added to the kernel
276 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
277 * from the kernel lists and can be freed or reused once this function
278 * returns.
279 *
280 * The packet type might still be in use by receivers
281 * and must not be freed until after all the CPU's have gone
282 * through a quiescent state.
283 */
284void __dev_remove_pack(struct packet_type *pt)
285{
286 struct list_head *head;
287 struct packet_type *pt1;
288
289 spin_lock_bh(&ptype_lock);
290
291 if (pt->type == htons(ETH_P_ALL)) {
292 netdev_nit--;
293 head = &ptype_all;
294 } else
295 head = &ptype_base[ntohs(pt->type) & 15];
296
297 list_for_each_entry(pt1, head, list) {
298 if (pt == pt1) {
299 list_del_rcu(&pt->list);
300 goto out;
301 }
302 }
303
304 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
305out:
306 spin_unlock_bh(&ptype_lock);
307}
308/**
309 * dev_remove_pack - remove packet handler
310 * @pt: packet type declaration
311 *
312 * Remove a protocol handler that was previously added to the kernel
313 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
314 * from the kernel lists and can be freed or reused once this function
315 * returns.
316 *
317 * This call sleeps to guarantee that no CPU is looking at the packet
318 * type after return.
319 */
320void dev_remove_pack(struct packet_type *pt)
321{
322 __dev_remove_pack(pt);
323
324 synchronize_net();
325}
326
327/******************************************************************************
328
329 Device Boot-time Settings Routines
330
331*******************************************************************************/
332
333/* Boot time configuration table */
334static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
335
336/**
337 * netdev_boot_setup_add - add new setup entry
338 * @name: name of the device
339 * @map: configured settings for the device
340 *
341 * Adds new setup entry to the dev_boot_setup list. The function
342 * returns 0 on error and 1 on success. This is a generic routine to
343 * all netdevices.
344 */
345static int netdev_boot_setup_add(char *name, struct ifmap *map)
346{
347 struct netdev_boot_setup *s;
348 int i;
349
350 s = dev_boot_setup;
351 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
352 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
353 memset(s[i].name, 0, sizeof(s[i].name));
354 strcpy(s[i].name, name);
355 memcpy(&s[i].map, map, sizeof(s[i].map));
356 break;
357 }
358 }
359
360 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
361}
362
363/**
364 * netdev_boot_setup_check - check boot time settings
365 * @dev: the netdevice
366 *
367 * Check boot time settings for the device.
368 * The found settings are set for the device to be used
369 * later in the device probing.
370 * Returns 0 if no settings found, 1 if they are.
371 */
372int netdev_boot_setup_check(struct net_device *dev)
373{
374 struct netdev_boot_setup *s = dev_boot_setup;
375 int i;
376
377 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
378 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
379 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
380 dev->irq = s[i].map.irq;
381 dev->base_addr = s[i].map.base_addr;
382 dev->mem_start = s[i].map.mem_start;
383 dev->mem_end = s[i].map.mem_end;
384 return 1;
385 }
386 }
387 return 0;
388}
389
390
391/**
392 * netdev_boot_base - get address from boot time settings
393 * @prefix: prefix for network device
394 * @unit: id for network device
395 *
396 * Check boot time settings for the base address of device.
397 * The found settings are set for the device to be used
398 * later in the device probing.
399 * Returns 0 if no settings found.
400 */
401unsigned long netdev_boot_base(const char *prefix, int unit)
402{
403 const struct netdev_boot_setup *s = dev_boot_setup;
404 char name[IFNAMSIZ];
405 int i;
406
407 sprintf(name, "%s%d", prefix, unit);
408
409 /*
410 * If device already registered then return base of 1
411 * to indicate not to probe for this interface
412 */
413 if (__dev_get_by_name(name))
414 return 1;
415
416 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
417 if (!strcmp(name, s[i].name))
418 return s[i].map.base_addr;
419 return 0;
420}
421
422/*
423 * Saves at boot time configured settings for any netdevice.
424 */
425int __init netdev_boot_setup(char *str)
426{
427 int ints[5];
428 struct ifmap map;
429
430 str = get_options(str, ARRAY_SIZE(ints), ints);
431 if (!str || !*str)
432 return 0;
433
434 /* Save settings */
435 memset(&map, 0, sizeof(map));
436 if (ints[0] > 0)
437 map.irq = ints[1];
438 if (ints[0] > 1)
439 map.base_addr = ints[2];
440 if (ints[0] > 2)
441 map.mem_start = ints[3];
442 if (ints[0] > 3)
443 map.mem_end = ints[4];
444
445 /* Add new entry to the list */
446 return netdev_boot_setup_add(str, &map);
447}
448
449__setup("netdev=", netdev_boot_setup);
450
451/*******************************************************************************
452
453 Device Interface Subroutines
454
455*******************************************************************************/
456
457/**
458 * __dev_get_by_name - find a device by its name
459 * @name: name to find
460 *
461 * Find an interface by name. Must be called under RTNL semaphore
462 * or @dev_base_lock. If the name is found a pointer to the device
463 * is returned. If the name is not found then %NULL is returned. The
464 * reference counters are not incremented so the caller must be
465 * careful with locks.
466 */
467
468struct net_device *__dev_get_by_name(const char *name)
469{
470 struct hlist_node *p;
471
472 hlist_for_each(p, dev_name_hash(name)) {
473 struct net_device *dev
474 = hlist_entry(p, struct net_device, name_hlist);
475 if (!strncmp(dev->name, name, IFNAMSIZ))
476 return dev;
477 }
478 return NULL;
479}
480
481/**
482 * dev_get_by_name - find a device by its name
483 * @name: name to find
484 *
485 * Find an interface by name. This can be called from any
486 * context and does its own locking. The returned handle has
487 * the usage count incremented and the caller must use dev_put() to
488 * release it when it is no longer needed. %NULL is returned if no
489 * matching device is found.
490 */
491
492struct net_device *dev_get_by_name(const char *name)
493{
494 struct net_device *dev;
495
496 read_lock(&dev_base_lock);
497 dev = __dev_get_by_name(name);
498 if (dev)
499 dev_hold(dev);
500 read_unlock(&dev_base_lock);
501 return dev;
502}
503
504/**
505 * __dev_get_by_index - find a device by its ifindex
506 * @ifindex: index of device
507 *
508 * Search for an interface by index. Returns %NULL if the device
509 * is not found or a pointer to the device. The device has not
510 * had its reference counter increased so the caller must be careful
511 * about locking. The caller must hold either the RTNL semaphore
512 * or @dev_base_lock.
513 */
514
515struct net_device *__dev_get_by_index(int ifindex)
516{
517 struct hlist_node *p;
518
519 hlist_for_each(p, dev_index_hash(ifindex)) {
520 struct net_device *dev
521 = hlist_entry(p, struct net_device, index_hlist);
522 if (dev->ifindex == ifindex)
523 return dev;
524 }
525 return NULL;
526}
527
528
529/**
530 * dev_get_by_index - find a device by its ifindex
531 * @ifindex: index of device
532 *
533 * Search for an interface by index. Returns NULL if the device
534 * is not found or a pointer to the device. The device returned has
535 * had a reference added and the pointer is safe until the user calls
536 * dev_put to indicate they have finished with it.
537 */
538
539struct net_device *dev_get_by_index(int ifindex)
540{
541 struct net_device *dev;
542
543 read_lock(&dev_base_lock);
544 dev = __dev_get_by_index(ifindex);
545 if (dev)
546 dev_hold(dev);
547 read_unlock(&dev_base_lock);
548 return dev;
549}
550
551/**
552 * dev_getbyhwaddr - find a device by its hardware address
553 * @type: media type of device
554 * @ha: hardware address
555 *
556 * Search for an interface by MAC address. Returns NULL if the device
557 * is not found or a pointer to the device. The caller must hold the
558 * rtnl semaphore. The returned device has not had its ref count increased
559 * and the caller must therefore be careful about locking
560 *
561 * BUGS:
562 * If the API was consistent this would be __dev_get_by_hwaddr
563 */
564
565struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
566{
567 struct net_device *dev;
568
569 ASSERT_RTNL();
570
571 for (dev = dev_base; dev; dev = dev->next)
572 if (dev->type == type &&
573 !memcmp(dev->dev_addr, ha, dev->addr_len))
574 break;
575 return dev;
576}
577
Jochen Friedrichcf309e32005-09-22 04:44:55 -0300578EXPORT_SYMBOL(dev_getbyhwaddr);
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580struct net_device *dev_getfirstbyhwtype(unsigned short type)
581{
582 struct net_device *dev;
583
584 rtnl_lock();
585 for (dev = dev_base; dev; dev = dev->next) {
586 if (dev->type == type) {
587 dev_hold(dev);
588 break;
589 }
590 }
591 rtnl_unlock();
592 return dev;
593}
594
595EXPORT_SYMBOL(dev_getfirstbyhwtype);
596
597/**
598 * dev_get_by_flags - find any device with given flags
599 * @if_flags: IFF_* values
600 * @mask: bitmask of bits in if_flags to check
601 *
602 * Search for any interface with the given flags. Returns NULL if a device
603 * is not found or a pointer to the device. The device returned has
604 * had a reference added and the pointer is safe until the user calls
605 * dev_put to indicate they have finished with it.
606 */
607
608struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
609{
610 struct net_device *dev;
611
612 read_lock(&dev_base_lock);
613 for (dev = dev_base; dev != NULL; dev = dev->next) {
614 if (((dev->flags ^ if_flags) & mask) == 0) {
615 dev_hold(dev);
616 break;
617 }
618 }
619 read_unlock(&dev_base_lock);
620 return dev;
621}
622
623/**
624 * dev_valid_name - check if name is okay for network device
625 * @name: name string
626 *
627 * Network device names need to be valid file names to
628 * to allow sysfs to work
629 */
Mitch Williamsc2373ee2005-11-09 10:34:45 -0800630int dev_valid_name(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
632 return !(*name == '\0'
633 || !strcmp(name, ".")
634 || !strcmp(name, "..")
635 || strchr(name, '/'));
636}
637
638/**
639 * dev_alloc_name - allocate a name for a device
640 * @dev: device
641 * @name: name format string
642 *
643 * Passed a format string - eg "lt%d" it will try and find a suitable
644 * id. Not efficient for many devices, not called a lot. The caller
645 * must hold the dev_base or rtnl lock while allocating the name and
646 * adding the device in order to avoid duplicates. Returns the number
647 * of the unit assigned or a negative errno code.
648 */
649
650int dev_alloc_name(struct net_device *dev, const char *name)
651{
652 int i = 0;
653 char buf[IFNAMSIZ];
654 const char *p;
655 const int max_netdevices = 8*PAGE_SIZE;
656 long *inuse;
657 struct net_device *d;
658
659 p = strnchr(name, IFNAMSIZ-1, '%');
660 if (p) {
661 /*
662 * Verify the string as this thing may have come from
663 * the user. There must be either one "%d" and no other "%"
664 * characters.
665 */
666 if (p[1] != 'd' || strchr(p + 2, '%'))
667 return -EINVAL;
668
669 /* Use one page as a bit array of possible slots */
670 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
671 if (!inuse)
672 return -ENOMEM;
673
674 for (d = dev_base; d; d = d->next) {
675 if (!sscanf(d->name, name, &i))
676 continue;
677 if (i < 0 || i >= max_netdevices)
678 continue;
679
680 /* avoid cases where sscanf is not exact inverse of printf */
681 snprintf(buf, sizeof(buf), name, i);
682 if (!strncmp(buf, d->name, IFNAMSIZ))
683 set_bit(i, inuse);
684 }
685
686 i = find_first_zero_bit(inuse, max_netdevices);
687 free_page((unsigned long) inuse);
688 }
689
690 snprintf(buf, sizeof(buf), name, i);
691 if (!__dev_get_by_name(buf)) {
692 strlcpy(dev->name, buf, IFNAMSIZ);
693 return i;
694 }
695
696 /* It is possible to run out of possible slots
697 * when the name is long and there isn't enough space left
698 * for the digits, or if all bits are used.
699 */
700 return -ENFILE;
701}
702
703
704/**
705 * dev_change_name - change name of a device
706 * @dev: device
707 * @newname: name (or format string) must be at least IFNAMSIZ
708 *
709 * Change name of a device, can pass format strings "eth%d".
710 * for wildcarding.
711 */
712int dev_change_name(struct net_device *dev, char *newname)
713{
714 int err = 0;
715
716 ASSERT_RTNL();
717
718 if (dev->flags & IFF_UP)
719 return -EBUSY;
720
721 if (!dev_valid_name(newname))
722 return -EINVAL;
723
724 if (strchr(newname, '%')) {
725 err = dev_alloc_name(dev, newname);
726 if (err < 0)
727 return err;
728 strcpy(newname, dev->name);
729 }
730 else if (__dev_get_by_name(newname))
731 return -EEXIST;
732 else
733 strlcpy(dev->name, newname, IFNAMSIZ);
734
735 err = class_device_rename(&dev->class_dev, dev->name);
736 if (!err) {
737 hlist_del(&dev->name_hlist);
738 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
739 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
740 }
741
742 return err;
743}
744
745/**
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700746 * netdev_features_change - device changes fatures
747 * @dev: device to cause notification
748 *
749 * Called to indicate a device has changed features.
750 */
751void netdev_features_change(struct net_device *dev)
752{
753 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
754}
755EXPORT_SYMBOL(netdev_features_change);
756
757/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 * netdev_state_change - device changes state
759 * @dev: device to cause notification
760 *
761 * Called to indicate a device has changed state. This function calls
762 * the notifier chains for netdev_chain and sends a NEWLINK message
763 * to the routing socket.
764 */
765void netdev_state_change(struct net_device *dev)
766{
767 if (dev->flags & IFF_UP) {
768 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
769 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
770 }
771}
772
773/**
774 * dev_load - load a network module
775 * @name: name of interface
776 *
777 * If a network interface is not present and the process has suitable
778 * privileges this function loads the module. If module loading is not
779 * available in this kernel then it becomes a nop.
780 */
781
782void dev_load(const char *name)
783{
784 struct net_device *dev;
785
786 read_lock(&dev_base_lock);
787 dev = __dev_get_by_name(name);
788 read_unlock(&dev_base_lock);
789
790 if (!dev && capable(CAP_SYS_MODULE))
791 request_module("%s", name);
792}
793
794static int default_rebuild_header(struct sk_buff *skb)
795{
796 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
797 skb->dev ? skb->dev->name : "NULL!!!");
798 kfree_skb(skb);
799 return 1;
800}
801
802
803/**
804 * dev_open - prepare an interface for use.
805 * @dev: device to open
806 *
807 * Takes a device from down to up state. The device's private open
808 * function is invoked and then the multicast lists are loaded. Finally
809 * the device is moved into the up state and a %NETDEV_UP message is
810 * sent to the netdev notifier chain.
811 *
812 * Calling this function on an active interface is a nop. On a failure
813 * a negative errno code is returned.
814 */
815int dev_open(struct net_device *dev)
816{
817 int ret = 0;
818
819 /*
820 * Is it already up?
821 */
822
823 if (dev->flags & IFF_UP)
824 return 0;
825
826 /*
827 * Is it even present?
828 */
829 if (!netif_device_present(dev))
830 return -ENODEV;
831
832 /*
833 * Call device private open method
834 */
835 set_bit(__LINK_STATE_START, &dev->state);
836 if (dev->open) {
837 ret = dev->open(dev);
838 if (ret)
839 clear_bit(__LINK_STATE_START, &dev->state);
840 }
841
842 /*
843 * If it went open OK then:
844 */
845
846 if (!ret) {
847 /*
848 * Set the flags.
849 */
850 dev->flags |= IFF_UP;
851
852 /*
853 * Initialize multicasting status
854 */
855 dev_mc_upload(dev);
856
857 /*
858 * Wakeup transmit queue engine
859 */
860 dev_activate(dev);
861
862 /*
863 * ... and announce new interface.
864 */
865 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
866 }
867 return ret;
868}
869
870/**
871 * dev_close - shutdown an interface.
872 * @dev: device to shutdown
873 *
874 * This function moves an active device into down state. A
875 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
876 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
877 * chain.
878 */
879int dev_close(struct net_device *dev)
880{
881 if (!(dev->flags & IFF_UP))
882 return 0;
883
884 /*
885 * Tell people we are going down, so that they can
886 * prepare to death, when device is still operating.
887 */
888 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
889
890 dev_deactivate(dev);
891
892 clear_bit(__LINK_STATE_START, &dev->state);
893
894 /* Synchronize to scheduled poll. We cannot touch poll list,
895 * it can be even on different cpu. So just clear netif_running(),
896 * and wait when poll really will happen. Actually, the best place
897 * for this is inside dev->stop() after device stopped its irq
898 * engine, but this requires more changes in devices. */
899
900 smp_mb__after_clear_bit(); /* Commit netif_running(). */
901 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
902 /* No hurry. */
David S. Miller6192b542005-07-28 12:12:58 -0700903 msleep(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 }
905
906 /*
907 * Call the device specific close. This cannot fail.
908 * Only if device is UP
909 *
910 * We allow it to be called even after a DETACH hot-plug
911 * event.
912 */
913 if (dev->stop)
914 dev->stop(dev);
915
916 /*
917 * Device is now down.
918 */
919
920 dev->flags &= ~IFF_UP;
921
922 /*
923 * Tell people we are down
924 */
925 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
926
927 return 0;
928}
929
930
931/*
932 * Device change register/unregister. These are not inline or static
933 * as we export them to the world.
934 */
935
936/**
937 * register_netdevice_notifier - register a network notifier block
938 * @nb: notifier
939 *
940 * Register a notifier to be called when network device events occur.
941 * The notifier passed is linked into the kernel structures and must
942 * not be reused until it has been unregistered. A negative errno code
943 * is returned on a failure.
944 *
945 * When registered all registration and up events are replayed
946 * to the new notifier to allow device to have a race free
947 * view of the network device list.
948 */
949
950int register_netdevice_notifier(struct notifier_block *nb)
951{
952 struct net_device *dev;
953 int err;
954
955 rtnl_lock();
956 err = notifier_chain_register(&netdev_chain, nb);
957 if (!err) {
958 for (dev = dev_base; dev; dev = dev->next) {
959 nb->notifier_call(nb, NETDEV_REGISTER, dev);
960
961 if (dev->flags & IFF_UP)
962 nb->notifier_call(nb, NETDEV_UP, dev);
963 }
964 }
965 rtnl_unlock();
966 return err;
967}
968
969/**
970 * unregister_netdevice_notifier - unregister a network notifier block
971 * @nb: notifier
972 *
973 * Unregister a notifier previously registered by
974 * register_netdevice_notifier(). The notifier is unlinked into the
975 * kernel structures and may then be reused. A negative errno code
976 * is returned on a failure.
977 */
978
979int unregister_netdevice_notifier(struct notifier_block *nb)
980{
Herbert Xu9f514952006-03-25 01:24:25 -0800981 int err;
982
983 rtnl_lock();
984 err = notifier_chain_unregister(&netdev_chain, nb);
985 rtnl_unlock();
986 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987}
988
989/**
990 * call_netdevice_notifiers - call all network notifier blocks
991 * @val: value passed unmodified to notifier function
992 * @v: pointer passed unmodified to notifier function
993 *
994 * Call all network notifier blocks. Parameters and return value
995 * are as for notifier_call_chain().
996 */
997
998int call_netdevice_notifiers(unsigned long val, void *v)
999{
1000 return notifier_call_chain(&netdev_chain, val, v);
1001}
1002
1003/* When > 0 there are consumers of rx skb time stamps */
1004static atomic_t netstamp_needed = ATOMIC_INIT(0);
1005
1006void net_enable_timestamp(void)
1007{
1008 atomic_inc(&netstamp_needed);
1009}
1010
1011void net_disable_timestamp(void)
1012{
1013 atomic_dec(&netstamp_needed);
1014}
1015
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001016void __net_timestamp(struct sk_buff *skb)
1017{
1018 struct timeval tv;
1019
1020 do_gettimeofday(&tv);
1021 skb_set_timestamp(skb, &tv);
1022}
1023EXPORT_SYMBOL(__net_timestamp);
1024
1025static inline void net_timestamp(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
1027 if (atomic_read(&netstamp_needed))
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001028 __net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 else {
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001030 skb->tstamp.off_sec = 0;
1031 skb->tstamp.off_usec = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 }
1033}
1034
1035/*
1036 * Support routine. Sends outgoing frames to any network
1037 * taps currently in use.
1038 */
1039
1040void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1041{
1042 struct packet_type *ptype;
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001043
1044 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
1046 rcu_read_lock();
1047 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1048 /* Never send packets back to the socket
1049 * they originated from - MvS (miquels@drinkel.ow.org)
1050 */
1051 if ((ptype->dev == dev || !ptype->dev) &&
1052 (ptype->af_packet_priv == NULL ||
1053 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1054 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1055 if (!skb2)
1056 break;
1057
1058 /* skb->nh should be correctly
1059 set by sender, so that the second statement is
1060 just protection against buggy protocols.
1061 */
1062 skb2->mac.raw = skb2->data;
1063
1064 if (skb2->nh.raw < skb2->data ||
1065 skb2->nh.raw > skb2->tail) {
1066 if (net_ratelimit())
1067 printk(KERN_CRIT "protocol %04x is "
1068 "buggy, dev %s\n",
1069 skb2->protocol, dev->name);
1070 skb2->nh.raw = skb2->data;
1071 }
1072
1073 skb2->h.raw = skb2->nh.raw;
1074 skb2->pkt_type = PACKET_OUTGOING;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001075 ptype->func(skb2, skb->dev, ptype, skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 }
1077 }
1078 rcu_read_unlock();
1079}
1080
1081/*
1082 * Invalidate hardware checksum when packet is to be mangled, and
1083 * complete checksum manually on outgoing path.
1084 */
1085int skb_checksum_help(struct sk_buff *skb, int inward)
1086{
1087 unsigned int csum;
1088 int ret = 0, offset = skb->h.raw - skb->data;
1089
1090 if (inward) {
1091 skb->ip_summed = CHECKSUM_NONE;
1092 goto out;
1093 }
1094
1095 if (skb_cloned(skb)) {
1096 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1097 if (ret)
1098 goto out;
1099 }
1100
Kris Katterjohn09a62662006-01-08 22:24:28 -08001101 BUG_ON(offset > (int)skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1103
1104 offset = skb->tail - skb->h.raw;
Kris Katterjohn09a62662006-01-08 22:24:28 -08001105 BUG_ON(offset <= 0);
1106 BUG_ON(skb->csum + 2 > offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
1108 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1109 skb->ip_summed = CHECKSUM_NONE;
1110out:
1111 return ret;
1112}
1113
Herbert Xufb286bb2005-11-10 13:01:24 -08001114/* Take action when hardware reception checksum errors are detected. */
1115#ifdef CONFIG_BUG
1116void netdev_rx_csum_fault(struct net_device *dev)
1117{
1118 if (net_ratelimit()) {
Stephen Hemminger246a4212005-12-08 15:21:39 -08001119 printk(KERN_ERR "%s: hw csum failure.\n",
1120 dev ? dev->name : "<unknown>");
Herbert Xufb286bb2005-11-10 13:01:24 -08001121 dump_stack();
1122 }
1123}
1124EXPORT_SYMBOL(netdev_rx_csum_fault);
1125#endif
1126
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127#ifdef CONFIG_HIGHMEM
1128/* Actually, we should eliminate this check as soon as we know, that:
1129 * 1. IOMMU is present and allows to map all the memory.
1130 * 2. No high memory really exists on this machine.
1131 */
1132
1133static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1134{
1135 int i;
1136
1137 if (dev->features & NETIF_F_HIGHDMA)
1138 return 0;
1139
1140 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1141 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1142 return 1;
1143
1144 return 0;
1145}
1146#else
1147#define illegal_highdma(dev, skb) (0)
1148#endif
1149
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150/* Keep head the same: replace data */
Al Virodd0fc662005-10-07 07:46:04 +01001151int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152{
1153 unsigned int size;
1154 u8 *data;
1155 long offset;
1156 struct skb_shared_info *ninfo;
1157 int headerlen = skb->data - skb->head;
1158 int expand = (skb->tail + skb->data_len) - skb->end;
1159
1160 if (skb_shared(skb))
1161 BUG();
1162
1163 if (expand <= 0)
1164 expand = 0;
1165
1166 size = skb->end - skb->head + expand;
1167 size = SKB_DATA_ALIGN(size);
1168 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1169 if (!data)
1170 return -ENOMEM;
1171
1172 /* Copy entire thing */
1173 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1174 BUG();
1175
1176 /* Set up shinfo */
1177 ninfo = (struct skb_shared_info*)(data + size);
1178 atomic_set(&ninfo->dataref, 1);
1179 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1180 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1181 ninfo->nr_frags = 0;
1182 ninfo->frag_list = NULL;
1183
1184 /* Offset between the two in bytes */
1185 offset = data - skb->head;
1186
1187 /* Free old data. */
1188 skb_release_data(skb);
1189
1190 skb->head = data;
1191 skb->end = data + size;
1192
1193 /* Set up new pointers */
1194 skb->h.raw += offset;
1195 skb->nh.raw += offset;
1196 skb->mac.raw += offset;
1197 skb->tail += offset;
1198 skb->data += offset;
1199
1200 /* We are no longer a clone, even if we were. */
1201 skb->cloned = 0;
1202
1203 skb->tail += skb->data_len;
1204 skb->data_len = 0;
1205 return 0;
1206}
1207
1208#define HARD_TX_LOCK(dev, cpu) { \
1209 if ((dev->features & NETIF_F_LLTX) == 0) { \
1210 spin_lock(&dev->xmit_lock); \
1211 dev->xmit_lock_owner = cpu; \
1212 } \
1213}
1214
1215#define HARD_TX_UNLOCK(dev) { \
1216 if ((dev->features & NETIF_F_LLTX) == 0) { \
1217 dev->xmit_lock_owner = -1; \
1218 spin_unlock(&dev->xmit_lock); \
1219 } \
1220}
1221
1222/**
1223 * dev_queue_xmit - transmit a buffer
1224 * @skb: buffer to transmit
1225 *
1226 * Queue a buffer for transmission to a network device. The caller must
1227 * have set the device and priority and built the buffer before calling
1228 * this function. The function can be called from an interrupt.
1229 *
1230 * A negative errno code is returned on a failure. A success does not
1231 * guarantee the frame will be transmitted as it may be dropped due
1232 * to congestion or traffic shaping.
Ben Greearaf191362005-04-24 20:12:36 -07001233 *
1234 * -----------------------------------------------------------------------------------
1235 * I notice this method can also return errors from the queue disciplines,
1236 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1237 * be positive.
1238 *
1239 * Regardless of the return value, the skb is consumed, so it is currently
1240 * difficult to retry a send to this method. (You can bump the ref count
1241 * before sending to hold a reference for retry if you are careful.)
1242 *
1243 * When calling this method, interrupts MUST be enabled. This is because
1244 * the BH enable code must have IRQs enabled so that it will not deadlock.
1245 * --BLG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 */
1247
1248int dev_queue_xmit(struct sk_buff *skb)
1249{
1250 struct net_device *dev = skb->dev;
1251 struct Qdisc *q;
1252 int rc = -ENOMEM;
1253
1254 if (skb_shinfo(skb)->frag_list &&
1255 !(dev->features & NETIF_F_FRAGLIST) &&
1256 __skb_linearize(skb, GFP_ATOMIC))
1257 goto out_kfree_skb;
1258
1259 /* Fragmented skb is linearized if device does not support SG,
1260 * or if at least one of fragments is in highmem and device
1261 * does not support DMA from it.
1262 */
1263 if (skb_shinfo(skb)->nr_frags &&
1264 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1265 __skb_linearize(skb, GFP_ATOMIC))
1266 goto out_kfree_skb;
1267
1268 /* If packet is not checksummed and device does not support
1269 * checksumming for this protocol, complete checksumming here.
1270 */
1271 if (skb->ip_summed == CHECKSUM_HW &&
1272 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1273 (!(dev->features & NETIF_F_IP_CSUM) ||
1274 skb->protocol != htons(ETH_P_IP))))
1275 if (skb_checksum_help(skb, 0))
1276 goto out_kfree_skb;
1277
Eric Dumazet2d7ceec2005-09-27 15:22:58 -07001278 spin_lock_prefetch(&dev->queue_lock);
1279
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 /* Disable soft irqs for various locks below. Also
1281 * stops preemption for RCU.
1282 */
1283 local_bh_disable();
1284
1285 /* Updates of qdisc are serialized by queue_lock.
1286 * The struct Qdisc which is pointed to by qdisc is now a
1287 * rcu structure - it may be accessed without acquiring
1288 * a lock (but the structure may be stale.) The freeing of the
1289 * qdisc will be deferred until it's known that there are no
1290 * more references to it.
1291 *
1292 * If the qdisc has an enqueue function, we still need to
1293 * hold the queue_lock before calling it, since queue_lock
1294 * also serializes access to the device queue.
1295 */
1296
1297 q = rcu_dereference(dev->qdisc);
1298#ifdef CONFIG_NET_CLS_ACT
1299 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1300#endif
1301 if (q->enqueue) {
1302 /* Grab device queue */
1303 spin_lock(&dev->queue_lock);
1304
1305 rc = q->enqueue(skb, q);
1306
1307 qdisc_run(dev);
1308
1309 spin_unlock(&dev->queue_lock);
1310 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1311 goto out;
1312 }
1313
1314 /* The device has no queue. Common case for software devices:
1315 loopback, all the sorts of tunnels...
1316
1317 Really, it is unlikely that xmit_lock protection is necessary here.
1318 (f.e. loopback and IP tunnels are clean ignoring statistics
1319 counters.)
1320 However, it is possible, that they rely on protection
1321 made by us here.
1322
1323 Check this and shot the lock. It is not prone from deadlocks.
1324 Either shot noqueue qdisc, it is even simpler 8)
1325 */
1326 if (dev->flags & IFF_UP) {
1327 int cpu = smp_processor_id(); /* ok because BHs are off */
1328
1329 if (dev->xmit_lock_owner != cpu) {
1330
1331 HARD_TX_LOCK(dev, cpu);
1332
1333 if (!netif_queue_stopped(dev)) {
1334 if (netdev_nit)
1335 dev_queue_xmit_nit(skb, dev);
1336
1337 rc = 0;
1338 if (!dev->hard_start_xmit(skb, dev)) {
1339 HARD_TX_UNLOCK(dev);
1340 goto out;
1341 }
1342 }
1343 HARD_TX_UNLOCK(dev);
1344 if (net_ratelimit())
1345 printk(KERN_CRIT "Virtual device %s asks to "
1346 "queue packet!\n", dev->name);
1347 } else {
1348 /* Recursion is detected! It is possible,
1349 * unfortunately */
1350 if (net_ratelimit())
1351 printk(KERN_CRIT "Dead loop on virtual device "
1352 "%s, fix it urgently!\n", dev->name);
1353 }
1354 }
1355
1356 rc = -ENETDOWN;
1357 local_bh_enable();
1358
1359out_kfree_skb:
1360 kfree_skb(skb);
1361 return rc;
1362out:
1363 local_bh_enable();
1364 return rc;
1365}
1366
1367
1368/*=======================================================================
1369 Receiver routines
1370 =======================================================================*/
1371
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001372int netdev_max_backlog = 1000;
1373int netdev_budget = 300;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374int weight_p = 64; /* old backlog weight */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
1376DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1377
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379/**
1380 * netif_rx - post buffer to the network code
1381 * @skb: buffer to post
1382 *
1383 * This function receives a packet from a device driver and queues it for
1384 * the upper (protocol) levels to process. It always succeeds. The buffer
1385 * may be dropped during processing for congestion control or by the
1386 * protocol layers.
1387 *
1388 * return values:
1389 * NET_RX_SUCCESS (no congestion)
1390 * NET_RX_CN_LOW (low congestion)
1391 * NET_RX_CN_MOD (moderate congestion)
1392 * NET_RX_CN_HIGH (high congestion)
1393 * NET_RX_DROP (packet was dropped)
1394 *
1395 */
1396
1397int netif_rx(struct sk_buff *skb)
1398{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 struct softnet_data *queue;
1400 unsigned long flags;
1401
1402 /* if netpoll wants it, pretend we never saw it */
1403 if (netpoll_rx(skb))
1404 return NET_RX_DROP;
1405
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001406 if (!skb->tstamp.off_sec)
1407 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
1409 /*
1410 * The code is rearranged so that the path is the most
1411 * short when CPU is congested, but is still operating.
1412 */
1413 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 queue = &__get_cpu_var(softnet_data);
1415
1416 __get_cpu_var(netdev_rx_stat).total++;
1417 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1418 if (queue->input_pkt_queue.qlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419enqueue:
1420 dev_hold(skb->dev);
1421 __skb_queue_tail(&queue->input_pkt_queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 local_irq_restore(flags);
Stephen Hemminger34008d82005-06-23 20:10:00 -07001423 return NET_RX_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 }
1425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 netif_rx_schedule(&queue->backlog_dev);
1427 goto enqueue;
1428 }
1429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 __get_cpu_var(netdev_rx_stat).dropped++;
1431 local_irq_restore(flags);
1432
1433 kfree_skb(skb);
1434 return NET_RX_DROP;
1435}
1436
1437int netif_rx_ni(struct sk_buff *skb)
1438{
1439 int err;
1440
1441 preempt_disable();
1442 err = netif_rx(skb);
1443 if (local_softirq_pending())
1444 do_softirq();
1445 preempt_enable();
1446
1447 return err;
1448}
1449
1450EXPORT_SYMBOL(netif_rx_ni);
1451
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001452static inline struct net_device *skb_bond(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
1454 struct net_device *dev = skb->dev;
1455
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001456 if (dev->master) {
1457 /*
1458 * On bonding slaves other than the currently active
1459 * slave, suppress duplicates except for 802.3ad
1460 * ETH_P_SLOW and alb non-mcast/bcast.
1461 */
1462 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1463 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1464 if (skb->pkt_type != PACKET_BROADCAST &&
1465 skb->pkt_type != PACKET_MULTICAST)
1466 goto keep;
1467 }
1468
1469 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1470 skb->protocol == __constant_htons(ETH_P_SLOW))
1471 goto keep;
1472
1473 kfree_skb(skb);
1474 return NULL;
1475 }
1476keep:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 skb->dev = dev->master;
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001478 }
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001479
1480 return dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481}
1482
1483static void net_tx_action(struct softirq_action *h)
1484{
1485 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1486
1487 if (sd->completion_queue) {
1488 struct sk_buff *clist;
1489
1490 local_irq_disable();
1491 clist = sd->completion_queue;
1492 sd->completion_queue = NULL;
1493 local_irq_enable();
1494
1495 while (clist) {
1496 struct sk_buff *skb = clist;
1497 clist = clist->next;
1498
1499 BUG_TRAP(!atomic_read(&skb->users));
1500 __kfree_skb(skb);
1501 }
1502 }
1503
1504 if (sd->output_queue) {
1505 struct net_device *head;
1506
1507 local_irq_disable();
1508 head = sd->output_queue;
1509 sd->output_queue = NULL;
1510 local_irq_enable();
1511
1512 while (head) {
1513 struct net_device *dev = head;
1514 head = head->next_sched;
1515
1516 smp_mb__before_clear_bit();
1517 clear_bit(__LINK_STATE_SCHED, &dev->state);
1518
1519 if (spin_trylock(&dev->queue_lock)) {
1520 qdisc_run(dev);
1521 spin_unlock(&dev->queue_lock);
1522 } else {
1523 netif_schedule(dev);
1524 }
1525 }
1526 }
1527}
1528
1529static __inline__ int deliver_skb(struct sk_buff *skb,
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001530 struct packet_type *pt_prev,
1531 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532{
1533 atomic_inc(&skb->users);
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001534 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535}
1536
1537#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1538int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1539struct net_bridge;
1540struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1541 unsigned char *addr);
1542void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1543
1544static __inline__ int handle_bridge(struct sk_buff **pskb,
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001545 struct packet_type **pt_prev, int *ret,
1546 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547{
1548 struct net_bridge_port *port;
1549
1550 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1551 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1552 return 0;
1553
1554 if (*pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001555 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 *pt_prev = NULL;
1557 }
1558
1559 return br_handle_frame_hook(port, pskb);
1560}
1561#else
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001562#define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563#endif
1564
1565#ifdef CONFIG_NET_CLS_ACT
1566/* TODO: Maybe we should just force sch_ingress to be compiled in
1567 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1568 * a compare and 2 stores extra right now if we dont have it on
1569 * but have CONFIG_NET_CLS_ACT
1570 * NOTE: This doesnt stop any functionality; if you dont have
1571 * the ingress scheduler, you just cant add policies on ingress.
1572 *
1573 */
1574static int ing_filter(struct sk_buff *skb)
1575{
1576 struct Qdisc *q;
1577 struct net_device *dev = skb->dev;
1578 int result = TC_ACT_OK;
1579
1580 if (dev->qdisc_ingress) {
1581 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1582 if (MAX_RED_LOOP < ttl++) {
1583 printk("Redir loop detected Dropping packet (%s->%s)\n",
David S. Miller86e65da2005-08-09 19:36:29 -07001584 skb->input_dev->name, skb->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 return TC_ACT_SHOT;
1586 }
1587
1588 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1589
1590 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
David S. Miller86e65da2005-08-09 19:36:29 -07001591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 spin_lock(&dev->ingress_lock);
1593 if ((q = dev->qdisc_ingress) != NULL)
1594 result = q->enqueue(skb, q);
1595 spin_unlock(&dev->ingress_lock);
1596
1597 }
1598
1599 return result;
1600}
1601#endif
1602
1603int netif_receive_skb(struct sk_buff *skb)
1604{
1605 struct packet_type *ptype, *pt_prev;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001606 struct net_device *orig_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 int ret = NET_RX_DROP;
1608 unsigned short type;
1609
1610 /* if we've gotten here through NAPI, check netpoll */
1611 if (skb->dev->poll && netpoll_rx(skb))
1612 return NET_RX_DROP;
1613
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001614 if (!skb->tstamp.off_sec)
1615 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
David S. Miller86e65da2005-08-09 19:36:29 -07001617 if (!skb->input_dev)
1618 skb->input_dev = skb->dev;
1619
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001620 orig_dev = skb_bond(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001622 if (!orig_dev)
1623 return NET_RX_DROP;
1624
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 __get_cpu_var(netdev_rx_stat).total++;
1626
1627 skb->h.raw = skb->nh.raw = skb->data;
1628 skb->mac_len = skb->nh.raw - skb->mac.raw;
1629
1630 pt_prev = NULL;
1631
1632 rcu_read_lock();
1633
1634#ifdef CONFIG_NET_CLS_ACT
1635 if (skb->tc_verd & TC_NCLS) {
1636 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1637 goto ncls;
1638 }
1639#endif
1640
1641 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1642 if (!ptype->dev || ptype->dev == skb->dev) {
1643 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001644 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 pt_prev = ptype;
1646 }
1647 }
1648
1649#ifdef CONFIG_NET_CLS_ACT
1650 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001651 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 pt_prev = NULL; /* noone else should process this after*/
1653 } else {
1654 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1655 }
1656
1657 ret = ing_filter(skb);
1658
1659 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1660 kfree_skb(skb);
1661 goto out;
1662 }
1663
1664 skb->tc_verd = 0;
1665ncls:
1666#endif
1667
1668 handle_diverter(skb);
1669
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001670 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 goto out;
1672
1673 type = skb->protocol;
1674 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1675 if (ptype->type == type &&
1676 (!ptype->dev || ptype->dev == skb->dev)) {
1677 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001678 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 pt_prev = ptype;
1680 }
1681 }
1682
1683 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001684 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 } else {
1686 kfree_skb(skb);
1687 /* Jamal, now you will not able to escape explaining
1688 * me how you were going to use this. :-)
1689 */
1690 ret = NET_RX_DROP;
1691 }
1692
1693out:
1694 rcu_read_unlock();
1695 return ret;
1696}
1697
1698static int process_backlog(struct net_device *backlog_dev, int *budget)
1699{
1700 int work = 0;
1701 int quota = min(backlog_dev->quota, *budget);
1702 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1703 unsigned long start_time = jiffies;
1704
Stephen Hemmingere3876602005-06-08 14:56:01 -07001705 backlog_dev->weight = weight_p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 for (;;) {
1707 struct sk_buff *skb;
1708 struct net_device *dev;
1709
1710 local_irq_disable();
1711 skb = __skb_dequeue(&queue->input_pkt_queue);
1712 if (!skb)
1713 goto job_done;
1714 local_irq_enable();
1715
1716 dev = skb->dev;
1717
1718 netif_receive_skb(skb);
1719
1720 dev_put(dev);
1721
1722 work++;
1723
1724 if (work >= quota || jiffies - start_time > 1)
1725 break;
1726
1727 }
1728
1729 backlog_dev->quota -= work;
1730 *budget -= work;
1731 return -1;
1732
1733job_done:
1734 backlog_dev->quota -= work;
1735 *budget -= work;
1736
1737 list_del(&backlog_dev->poll_list);
1738 smp_mb__before_clear_bit();
1739 netif_poll_enable(backlog_dev);
1740
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 local_irq_enable();
1742 return 0;
1743}
1744
1745static void net_rx_action(struct softirq_action *h)
1746{
1747 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1748 unsigned long start_time = jiffies;
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001749 int budget = netdev_budget;
Matt Mackall53fb95d2005-08-11 19:27:43 -07001750 void *have;
1751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 local_irq_disable();
1753
1754 while (!list_empty(&queue->poll_list)) {
1755 struct net_device *dev;
1756
1757 if (budget <= 0 || jiffies - start_time > 1)
1758 goto softnet_break;
1759
1760 local_irq_enable();
1761
1762 dev = list_entry(queue->poll_list.next,
1763 struct net_device, poll_list);
Matt Mackall53fb95d2005-08-11 19:27:43 -07001764 have = netpoll_poll_lock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
1766 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
Matt Mackall53fb95d2005-08-11 19:27:43 -07001767 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 local_irq_disable();
Stephen Hemminger8aca8a22006-03-20 22:26:39 -08001769 list_move_tail(&dev->poll_list, &queue->poll_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 if (dev->quota < 0)
1771 dev->quota += dev->weight;
1772 else
1773 dev->quota = dev->weight;
1774 } else {
Matt Mackall53fb95d2005-08-11 19:27:43 -07001775 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 dev_put(dev);
1777 local_irq_disable();
1778 }
1779 }
1780out:
1781 local_irq_enable();
1782 return;
1783
1784softnet_break:
1785 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1786 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1787 goto out;
1788}
1789
1790static gifconf_func_t * gifconf_list [NPROTO];
1791
1792/**
1793 * register_gifconf - register a SIOCGIF handler
1794 * @family: Address family
1795 * @gifconf: Function handler
1796 *
1797 * Register protocol dependent address dumping routines. The handler
1798 * that is passed must not be freed or reused until it has been replaced
1799 * by another handler.
1800 */
1801int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1802{
1803 if (family >= NPROTO)
1804 return -EINVAL;
1805 gifconf_list[family] = gifconf;
1806 return 0;
1807}
1808
1809
1810/*
1811 * Map an interface index to its name (SIOCGIFNAME)
1812 */
1813
1814/*
1815 * We need this ioctl for efficient implementation of the
1816 * if_indextoname() function required by the IPv6 API. Without
1817 * it, we would have to search all the interfaces to find a
1818 * match. --pb
1819 */
1820
1821static int dev_ifname(struct ifreq __user *arg)
1822{
1823 struct net_device *dev;
1824 struct ifreq ifr;
1825
1826 /*
1827 * Fetch the caller's info block.
1828 */
1829
1830 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1831 return -EFAULT;
1832
1833 read_lock(&dev_base_lock);
1834 dev = __dev_get_by_index(ifr.ifr_ifindex);
1835 if (!dev) {
1836 read_unlock(&dev_base_lock);
1837 return -ENODEV;
1838 }
1839
1840 strcpy(ifr.ifr_name, dev->name);
1841 read_unlock(&dev_base_lock);
1842
1843 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1844 return -EFAULT;
1845 return 0;
1846}
1847
1848/*
1849 * Perform a SIOCGIFCONF call. This structure will change
1850 * size eventually, and there is nothing I can do about it.
1851 * Thus we will need a 'compatibility mode'.
1852 */
1853
1854static int dev_ifconf(char __user *arg)
1855{
1856 struct ifconf ifc;
1857 struct net_device *dev;
1858 char __user *pos;
1859 int len;
1860 int total;
1861 int i;
1862
1863 /*
1864 * Fetch the caller's info block.
1865 */
1866
1867 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1868 return -EFAULT;
1869
1870 pos = ifc.ifc_buf;
1871 len = ifc.ifc_len;
1872
1873 /*
1874 * Loop over the interfaces, and write an info block for each.
1875 */
1876
1877 total = 0;
1878 for (dev = dev_base; dev; dev = dev->next) {
1879 for (i = 0; i < NPROTO; i++) {
1880 if (gifconf_list[i]) {
1881 int done;
1882 if (!pos)
1883 done = gifconf_list[i](dev, NULL, 0);
1884 else
1885 done = gifconf_list[i](dev, pos + total,
1886 len - total);
1887 if (done < 0)
1888 return -EFAULT;
1889 total += done;
1890 }
1891 }
1892 }
1893
1894 /*
1895 * All done. Write the updated control block back to the caller.
1896 */
1897 ifc.ifc_len = total;
1898
1899 /*
1900 * Both BSD and Solaris return 0 here, so we do too.
1901 */
1902 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1903}
1904
1905#ifdef CONFIG_PROC_FS
1906/*
1907 * This is invoked by the /proc filesystem handler to display a device
1908 * in detail.
1909 */
1910static __inline__ struct net_device *dev_get_idx(loff_t pos)
1911{
1912 struct net_device *dev;
1913 loff_t i;
1914
1915 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1916
1917 return i == pos ? dev : NULL;
1918}
1919
1920void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1921{
1922 read_lock(&dev_base_lock);
1923 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1924}
1925
1926void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1927{
1928 ++*pos;
1929 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1930}
1931
1932void dev_seq_stop(struct seq_file *seq, void *v)
1933{
1934 read_unlock(&dev_base_lock);
1935}
1936
1937static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1938{
1939 if (dev->get_stats) {
1940 struct net_device_stats *stats = dev->get_stats(dev);
1941
1942 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1943 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1944 dev->name, stats->rx_bytes, stats->rx_packets,
1945 stats->rx_errors,
1946 stats->rx_dropped + stats->rx_missed_errors,
1947 stats->rx_fifo_errors,
1948 stats->rx_length_errors + stats->rx_over_errors +
1949 stats->rx_crc_errors + stats->rx_frame_errors,
1950 stats->rx_compressed, stats->multicast,
1951 stats->tx_bytes, stats->tx_packets,
1952 stats->tx_errors, stats->tx_dropped,
1953 stats->tx_fifo_errors, stats->collisions,
1954 stats->tx_carrier_errors +
1955 stats->tx_aborted_errors +
1956 stats->tx_window_errors +
1957 stats->tx_heartbeat_errors,
1958 stats->tx_compressed);
1959 } else
1960 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1961}
1962
1963/*
1964 * Called from the PROCfs module. This now uses the new arbitrary sized
1965 * /proc/net interface to create /proc/net/dev
1966 */
1967static int dev_seq_show(struct seq_file *seq, void *v)
1968{
1969 if (v == SEQ_START_TOKEN)
1970 seq_puts(seq, "Inter-| Receive "
1971 " | Transmit\n"
1972 " face |bytes packets errs drop fifo frame "
1973 "compressed multicast|bytes packets errs "
1974 "drop fifo colls carrier compressed\n");
1975 else
1976 dev_seq_printf_stats(seq, v);
1977 return 0;
1978}
1979
1980static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1981{
1982 struct netif_rx_stats *rc = NULL;
1983
1984 while (*pos < NR_CPUS)
1985 if (cpu_online(*pos)) {
1986 rc = &per_cpu(netdev_rx_stat, *pos);
1987 break;
1988 } else
1989 ++*pos;
1990 return rc;
1991}
1992
1993static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1994{
1995 return softnet_get_online(pos);
1996}
1997
1998static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1999{
2000 ++*pos;
2001 return softnet_get_online(pos);
2002}
2003
2004static void softnet_seq_stop(struct seq_file *seq, void *v)
2005{
2006}
2007
2008static int softnet_seq_show(struct seq_file *seq, void *v)
2009{
2010 struct netif_rx_stats *s = v;
2011
2012 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
Stephen Hemminger31aa02c2005-06-23 20:12:48 -07002013 s->total, s->dropped, s->time_squeeze, 0,
Stephen Hemmingerc1ebcdb2005-06-23 20:08:59 -07002014 0, 0, 0, 0, /* was fastroute */
2015 s->cpu_collision );
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 return 0;
2017}
2018
2019static struct seq_operations dev_seq_ops = {
2020 .start = dev_seq_start,
2021 .next = dev_seq_next,
2022 .stop = dev_seq_stop,
2023 .show = dev_seq_show,
2024};
2025
2026static int dev_seq_open(struct inode *inode, struct file *file)
2027{
2028 return seq_open(file, &dev_seq_ops);
2029}
2030
2031static struct file_operations dev_seq_fops = {
2032 .owner = THIS_MODULE,
2033 .open = dev_seq_open,
2034 .read = seq_read,
2035 .llseek = seq_lseek,
2036 .release = seq_release,
2037};
2038
2039static struct seq_operations softnet_seq_ops = {
2040 .start = softnet_seq_start,
2041 .next = softnet_seq_next,
2042 .stop = softnet_seq_stop,
2043 .show = softnet_seq_show,
2044};
2045
2046static int softnet_seq_open(struct inode *inode, struct file *file)
2047{
2048 return seq_open(file, &softnet_seq_ops);
2049}
2050
2051static struct file_operations softnet_seq_fops = {
2052 .owner = THIS_MODULE,
2053 .open = softnet_seq_open,
2054 .read = seq_read,
2055 .llseek = seq_lseek,
2056 .release = seq_release,
2057};
2058
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002059#ifdef CONFIG_WIRELESS_EXT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060extern int wireless_proc_init(void);
2061#else
2062#define wireless_proc_init() 0
2063#endif
2064
2065static int __init dev_proc_init(void)
2066{
2067 int rc = -ENOMEM;
2068
2069 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2070 goto out;
2071 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2072 goto out_dev;
2073 if (wireless_proc_init())
2074 goto out_softnet;
2075 rc = 0;
2076out:
2077 return rc;
2078out_softnet:
2079 proc_net_remove("softnet_stat");
2080out_dev:
2081 proc_net_remove("dev");
2082 goto out;
2083}
2084#else
2085#define dev_proc_init() 0
2086#endif /* CONFIG_PROC_FS */
2087
2088
2089/**
2090 * netdev_set_master - set up master/slave pair
2091 * @slave: slave device
2092 * @master: new master device
2093 *
2094 * Changes the master device of the slave. Pass %NULL to break the
2095 * bonding. The caller must hold the RTNL semaphore. On a failure
2096 * a negative errno code is returned. On success the reference counts
2097 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2098 * function returns zero.
2099 */
2100int netdev_set_master(struct net_device *slave, struct net_device *master)
2101{
2102 struct net_device *old = slave->master;
2103
2104 ASSERT_RTNL();
2105
2106 if (master) {
2107 if (old)
2108 return -EBUSY;
2109 dev_hold(master);
2110 }
2111
2112 slave->master = master;
2113
2114 synchronize_net();
2115
2116 if (old)
2117 dev_put(old);
2118
2119 if (master)
2120 slave->flags |= IFF_SLAVE;
2121 else
2122 slave->flags &= ~IFF_SLAVE;
2123
2124 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2125 return 0;
2126}
2127
2128/**
2129 * dev_set_promiscuity - update promiscuity count on a device
2130 * @dev: device
2131 * @inc: modifier
2132 *
2133 * Add or remove promsicuity from a device. While the count in the device
2134 * remains above zero the interface remains promiscuous. Once it hits zero
2135 * the device reverts back to normal filtering operation. A negative inc
2136 * value is used to drop promiscuity on the device.
2137 */
2138void dev_set_promiscuity(struct net_device *dev, int inc)
2139{
2140 unsigned short old_flags = dev->flags;
2141
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 if ((dev->promiscuity += inc) == 0)
2143 dev->flags &= ~IFF_PROMISC;
David Chau52609c02005-07-05 15:11:06 -07002144 else
2145 dev->flags |= IFF_PROMISC;
2146 if (dev->flags != old_flags) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 dev_mc_upload(dev);
2148 printk(KERN_INFO "device %s %s promiscuous mode\n",
2149 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2150 "left");
Steve Grubb5bdb9882005-12-03 08:39:35 -05002151 audit_log(current->audit_context, GFP_ATOMIC,
2152 AUDIT_ANOM_PROMISCUOUS,
2153 "dev=%s prom=%d old_prom=%d auid=%u",
2154 dev->name, (dev->flags & IFF_PROMISC),
2155 (old_flags & IFF_PROMISC),
2156 audit_get_loginuid(current->audit_context));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 }
2158}
2159
2160/**
2161 * dev_set_allmulti - update allmulti count on a device
2162 * @dev: device
2163 * @inc: modifier
2164 *
2165 * Add or remove reception of all multicast frames to a device. While the
2166 * count in the device remains above zero the interface remains listening
2167 * to all interfaces. Once it hits zero the device reverts back to normal
2168 * filtering operation. A negative @inc value is used to drop the counter
2169 * when releasing a resource needing all multicasts.
2170 */
2171
2172void dev_set_allmulti(struct net_device *dev, int inc)
2173{
2174 unsigned short old_flags = dev->flags;
2175
2176 dev->flags |= IFF_ALLMULTI;
2177 if ((dev->allmulti += inc) == 0)
2178 dev->flags &= ~IFF_ALLMULTI;
2179 if (dev->flags ^ old_flags)
2180 dev_mc_upload(dev);
2181}
2182
2183unsigned dev_get_flags(const struct net_device *dev)
2184{
2185 unsigned flags;
2186
2187 flags = (dev->flags & ~(IFF_PROMISC |
2188 IFF_ALLMULTI |
Stefan Rompfb00055a2006-03-20 17:09:11 -08002189 IFF_RUNNING |
2190 IFF_LOWER_UP |
2191 IFF_DORMANT)) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 (dev->gflags & (IFF_PROMISC |
2193 IFF_ALLMULTI));
2194
Stefan Rompfb00055a2006-03-20 17:09:11 -08002195 if (netif_running(dev)) {
2196 if (netif_oper_up(dev))
2197 flags |= IFF_RUNNING;
2198 if (netif_carrier_ok(dev))
2199 flags |= IFF_LOWER_UP;
2200 if (netif_dormant(dev))
2201 flags |= IFF_DORMANT;
2202 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
2204 return flags;
2205}
2206
2207int dev_change_flags(struct net_device *dev, unsigned flags)
2208{
2209 int ret;
2210 int old_flags = dev->flags;
2211
2212 /*
2213 * Set the flags on our device.
2214 */
2215
2216 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2217 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2218 IFF_AUTOMEDIA)) |
2219 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2220 IFF_ALLMULTI));
2221
2222 /*
2223 * Load in the correct multicast list now the flags have changed.
2224 */
2225
2226 dev_mc_upload(dev);
2227
2228 /*
2229 * Have we downed the interface. We handle IFF_UP ourselves
2230 * according to user attempts to set it, rather than blindly
2231 * setting it.
2232 */
2233
2234 ret = 0;
2235 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2236 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2237
2238 if (!ret)
2239 dev_mc_upload(dev);
2240 }
2241
2242 if (dev->flags & IFF_UP &&
2243 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2244 IFF_VOLATILE)))
2245 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2246
2247 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2248 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2249 dev->gflags ^= IFF_PROMISC;
2250 dev_set_promiscuity(dev, inc);
2251 }
2252
2253 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2254 is important. Some (broken) drivers set IFF_PROMISC, when
2255 IFF_ALLMULTI is requested not asking us and not reporting.
2256 */
2257 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2258 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2259 dev->gflags ^= IFF_ALLMULTI;
2260 dev_set_allmulti(dev, inc);
2261 }
2262
2263 if (old_flags ^ dev->flags)
2264 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2265
2266 return ret;
2267}
2268
2269int dev_set_mtu(struct net_device *dev, int new_mtu)
2270{
2271 int err;
2272
2273 if (new_mtu == dev->mtu)
2274 return 0;
2275
2276 /* MTU must be positive. */
2277 if (new_mtu < 0)
2278 return -EINVAL;
2279
2280 if (!netif_device_present(dev))
2281 return -ENODEV;
2282
2283 err = 0;
2284 if (dev->change_mtu)
2285 err = dev->change_mtu(dev, new_mtu);
2286 else
2287 dev->mtu = new_mtu;
2288 if (!err && dev->flags & IFF_UP)
2289 notifier_call_chain(&netdev_chain,
2290 NETDEV_CHANGEMTU, dev);
2291 return err;
2292}
2293
2294int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2295{
2296 int err;
2297
2298 if (!dev->set_mac_address)
2299 return -EOPNOTSUPP;
2300 if (sa->sa_family != dev->type)
2301 return -EINVAL;
2302 if (!netif_device_present(dev))
2303 return -ENODEV;
2304 err = dev->set_mac_address(dev, sa);
2305 if (!err)
2306 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2307 return err;
2308}
2309
2310/*
2311 * Perform the SIOCxIFxxx calls.
2312 */
2313static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2314{
2315 int err;
2316 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2317
2318 if (!dev)
2319 return -ENODEV;
2320
2321 switch (cmd) {
2322 case SIOCGIFFLAGS: /* Get interface flags */
2323 ifr->ifr_flags = dev_get_flags(dev);
2324 return 0;
2325
2326 case SIOCSIFFLAGS: /* Set interface flags */
2327 return dev_change_flags(dev, ifr->ifr_flags);
2328
2329 case SIOCGIFMETRIC: /* Get the metric on the interface
2330 (currently unused) */
2331 ifr->ifr_metric = 0;
2332 return 0;
2333
2334 case SIOCSIFMETRIC: /* Set the metric on the interface
2335 (currently unused) */
2336 return -EOPNOTSUPP;
2337
2338 case SIOCGIFMTU: /* Get the MTU of a device */
2339 ifr->ifr_mtu = dev->mtu;
2340 return 0;
2341
2342 case SIOCSIFMTU: /* Set the MTU of a device */
2343 return dev_set_mtu(dev, ifr->ifr_mtu);
2344
2345 case SIOCGIFHWADDR:
2346 if (!dev->addr_len)
2347 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2348 else
2349 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2350 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2351 ifr->ifr_hwaddr.sa_family = dev->type;
2352 return 0;
2353
2354 case SIOCSIFHWADDR:
2355 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2356
2357 case SIOCSIFHWBROADCAST:
2358 if (ifr->ifr_hwaddr.sa_family != dev->type)
2359 return -EINVAL;
2360 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2361 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2362 notifier_call_chain(&netdev_chain,
2363 NETDEV_CHANGEADDR, dev);
2364 return 0;
2365
2366 case SIOCGIFMAP:
2367 ifr->ifr_map.mem_start = dev->mem_start;
2368 ifr->ifr_map.mem_end = dev->mem_end;
2369 ifr->ifr_map.base_addr = dev->base_addr;
2370 ifr->ifr_map.irq = dev->irq;
2371 ifr->ifr_map.dma = dev->dma;
2372 ifr->ifr_map.port = dev->if_port;
2373 return 0;
2374
2375 case SIOCSIFMAP:
2376 if (dev->set_config) {
2377 if (!netif_device_present(dev))
2378 return -ENODEV;
2379 return dev->set_config(dev, &ifr->ifr_map);
2380 }
2381 return -EOPNOTSUPP;
2382
2383 case SIOCADDMULTI:
2384 if (!dev->set_multicast_list ||
2385 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2386 return -EINVAL;
2387 if (!netif_device_present(dev))
2388 return -ENODEV;
2389 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2390 dev->addr_len, 1);
2391
2392 case SIOCDELMULTI:
2393 if (!dev->set_multicast_list ||
2394 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2395 return -EINVAL;
2396 if (!netif_device_present(dev))
2397 return -ENODEV;
2398 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2399 dev->addr_len, 1);
2400
2401 case SIOCGIFINDEX:
2402 ifr->ifr_ifindex = dev->ifindex;
2403 return 0;
2404
2405 case SIOCGIFTXQLEN:
2406 ifr->ifr_qlen = dev->tx_queue_len;
2407 return 0;
2408
2409 case SIOCSIFTXQLEN:
2410 if (ifr->ifr_qlen < 0)
2411 return -EINVAL;
2412 dev->tx_queue_len = ifr->ifr_qlen;
2413 return 0;
2414
2415 case SIOCSIFNAME:
2416 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2417 return dev_change_name(dev, ifr->ifr_newname);
2418
2419 /*
2420 * Unknown or private ioctl
2421 */
2422
2423 default:
2424 if ((cmd >= SIOCDEVPRIVATE &&
2425 cmd <= SIOCDEVPRIVATE + 15) ||
2426 cmd == SIOCBONDENSLAVE ||
2427 cmd == SIOCBONDRELEASE ||
2428 cmd == SIOCBONDSETHWADDR ||
2429 cmd == SIOCBONDSLAVEINFOQUERY ||
2430 cmd == SIOCBONDINFOQUERY ||
2431 cmd == SIOCBONDCHANGEACTIVE ||
2432 cmd == SIOCGMIIPHY ||
2433 cmd == SIOCGMIIREG ||
2434 cmd == SIOCSMIIREG ||
2435 cmd == SIOCBRADDIF ||
2436 cmd == SIOCBRDELIF ||
2437 cmd == SIOCWANDEV) {
2438 err = -EOPNOTSUPP;
2439 if (dev->do_ioctl) {
2440 if (netif_device_present(dev))
2441 err = dev->do_ioctl(dev, ifr,
2442 cmd);
2443 else
2444 err = -ENODEV;
2445 }
2446 } else
2447 err = -EINVAL;
2448
2449 }
2450 return err;
2451}
2452
2453/*
2454 * This function handles all "interface"-type I/O control requests. The actual
2455 * 'doing' part of this is dev_ifsioc above.
2456 */
2457
2458/**
2459 * dev_ioctl - network device ioctl
2460 * @cmd: command to issue
2461 * @arg: pointer to a struct ifreq in user space
2462 *
2463 * Issue ioctl functions to devices. This is normally called by the
2464 * user space syscall interfaces but can sometimes be useful for
2465 * other purposes. The return value is the return from the syscall if
2466 * positive or a negative errno code on error.
2467 */
2468
2469int dev_ioctl(unsigned int cmd, void __user *arg)
2470{
2471 struct ifreq ifr;
2472 int ret;
2473 char *colon;
2474
2475 /* One special case: SIOCGIFCONF takes ifconf argument
2476 and requires shared lock, because it sleeps writing
2477 to user space.
2478 */
2479
2480 if (cmd == SIOCGIFCONF) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002481 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 ret = dev_ifconf((char __user *) arg);
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002483 rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 return ret;
2485 }
2486 if (cmd == SIOCGIFNAME)
2487 return dev_ifname((struct ifreq __user *)arg);
2488
2489 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2490 return -EFAULT;
2491
2492 ifr.ifr_name[IFNAMSIZ-1] = 0;
2493
2494 colon = strchr(ifr.ifr_name, ':');
2495 if (colon)
2496 *colon = 0;
2497
2498 /*
2499 * See which interface the caller is talking about.
2500 */
2501
2502 switch (cmd) {
2503 /*
2504 * These ioctl calls:
2505 * - can be done by all.
2506 * - atomic and do not require locking.
2507 * - return a value
2508 */
2509 case SIOCGIFFLAGS:
2510 case SIOCGIFMETRIC:
2511 case SIOCGIFMTU:
2512 case SIOCGIFHWADDR:
2513 case SIOCGIFSLAVE:
2514 case SIOCGIFMAP:
2515 case SIOCGIFINDEX:
2516 case SIOCGIFTXQLEN:
2517 dev_load(ifr.ifr_name);
2518 read_lock(&dev_base_lock);
2519 ret = dev_ifsioc(&ifr, cmd);
2520 read_unlock(&dev_base_lock);
2521 if (!ret) {
2522 if (colon)
2523 *colon = ':';
2524 if (copy_to_user(arg, &ifr,
2525 sizeof(struct ifreq)))
2526 ret = -EFAULT;
2527 }
2528 return ret;
2529
2530 case SIOCETHTOOL:
2531 dev_load(ifr.ifr_name);
2532 rtnl_lock();
2533 ret = dev_ethtool(&ifr);
2534 rtnl_unlock();
2535 if (!ret) {
2536 if (colon)
2537 *colon = ':';
2538 if (copy_to_user(arg, &ifr,
2539 sizeof(struct ifreq)))
2540 ret = -EFAULT;
2541 }
2542 return ret;
2543
2544 /*
2545 * These ioctl calls:
2546 * - require superuser power.
2547 * - require strict serialization.
2548 * - return a value
2549 */
2550 case SIOCGMIIPHY:
2551 case SIOCGMIIREG:
2552 case SIOCSIFNAME:
2553 if (!capable(CAP_NET_ADMIN))
2554 return -EPERM;
2555 dev_load(ifr.ifr_name);
2556 rtnl_lock();
2557 ret = dev_ifsioc(&ifr, cmd);
2558 rtnl_unlock();
2559 if (!ret) {
2560 if (colon)
2561 *colon = ':';
2562 if (copy_to_user(arg, &ifr,
2563 sizeof(struct ifreq)))
2564 ret = -EFAULT;
2565 }
2566 return ret;
2567
2568 /*
2569 * These ioctl calls:
2570 * - require superuser power.
2571 * - require strict serialization.
2572 * - do not return a value
2573 */
2574 case SIOCSIFFLAGS:
2575 case SIOCSIFMETRIC:
2576 case SIOCSIFMTU:
2577 case SIOCSIFMAP:
2578 case SIOCSIFHWADDR:
2579 case SIOCSIFSLAVE:
2580 case SIOCADDMULTI:
2581 case SIOCDELMULTI:
2582 case SIOCSIFHWBROADCAST:
2583 case SIOCSIFTXQLEN:
2584 case SIOCSMIIREG:
2585 case SIOCBONDENSLAVE:
2586 case SIOCBONDRELEASE:
2587 case SIOCBONDSETHWADDR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 case SIOCBONDCHANGEACTIVE:
2589 case SIOCBRADDIF:
2590 case SIOCBRDELIF:
2591 if (!capable(CAP_NET_ADMIN))
2592 return -EPERM;
Thomas Grafcabcac02006-01-24 12:46:33 -08002593 /* fall through */
2594 case SIOCBONDSLAVEINFOQUERY:
2595 case SIOCBONDINFOQUERY:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 dev_load(ifr.ifr_name);
2597 rtnl_lock();
2598 ret = dev_ifsioc(&ifr, cmd);
2599 rtnl_unlock();
2600 return ret;
2601
2602 case SIOCGIFMEM:
2603 /* Get the per device memory space. We can add this but
2604 * currently do not support it */
2605 case SIOCSIFMEM:
2606 /* Set the per device memory buffer space.
2607 * Not applicable in our case */
2608 case SIOCSIFLINK:
2609 return -EINVAL;
2610
2611 /*
2612 * Unknown or private ioctl.
2613 */
2614 default:
2615 if (cmd == SIOCWANDEV ||
2616 (cmd >= SIOCDEVPRIVATE &&
2617 cmd <= SIOCDEVPRIVATE + 15)) {
2618 dev_load(ifr.ifr_name);
2619 rtnl_lock();
2620 ret = dev_ifsioc(&ifr, cmd);
2621 rtnl_unlock();
2622 if (!ret && copy_to_user(arg, &ifr,
2623 sizeof(struct ifreq)))
2624 ret = -EFAULT;
2625 return ret;
2626 }
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002627#ifdef CONFIG_WIRELESS_EXT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 /* Take care of Wireless Extensions */
2629 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2630 /* If command is `set a parameter', or
2631 * `get the encoding parameters', check if
2632 * the user has the right to do it */
2633 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2634 if (!capable(CAP_NET_ADMIN))
2635 return -EPERM;
2636 }
2637 dev_load(ifr.ifr_name);
2638 rtnl_lock();
2639 /* Follow me in net/core/wireless.c */
2640 ret = wireless_process_ioctl(&ifr, cmd);
2641 rtnl_unlock();
2642 if (IW_IS_GET(cmd) &&
2643 copy_to_user(arg, &ifr,
2644 sizeof(struct ifreq)))
2645 ret = -EFAULT;
2646 return ret;
2647 }
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002648#endif /* CONFIG_WIRELESS_EXT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 return -EINVAL;
2650 }
2651}
2652
2653
2654/**
2655 * dev_new_index - allocate an ifindex
2656 *
2657 * Returns a suitable unique value for a new device interface
2658 * number. The caller must hold the rtnl semaphore or the
2659 * dev_base_lock to be sure it remains unique.
2660 */
2661static int dev_new_index(void)
2662{
2663 static int ifindex;
2664 for (;;) {
2665 if (++ifindex <= 0)
2666 ifindex = 1;
2667 if (!__dev_get_by_index(ifindex))
2668 return ifindex;
2669 }
2670}
2671
2672static int dev_boot_phase = 1;
2673
2674/* Delayed registration/unregisteration */
2675static DEFINE_SPINLOCK(net_todo_list_lock);
2676static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2677
2678static inline void net_set_todo(struct net_device *dev)
2679{
2680 spin_lock(&net_todo_list_lock);
2681 list_add_tail(&dev->todo_list, &net_todo_list);
2682 spin_unlock(&net_todo_list_lock);
2683}
2684
2685/**
2686 * register_netdevice - register a network device
2687 * @dev: device to register
2688 *
2689 * Take a completed network device structure and add it to the kernel
2690 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2691 * chain. 0 is returned on success. A negative errno code is returned
2692 * on a failure to set up the device, or if the name is a duplicate.
2693 *
2694 * Callers must hold the rtnl semaphore. You may want
2695 * register_netdev() instead of this.
2696 *
2697 * BUGS:
2698 * The locking appears insufficient to guarantee two parallel registers
2699 * will not get the same name.
2700 */
2701
2702int register_netdevice(struct net_device *dev)
2703{
2704 struct hlist_head *head;
2705 struct hlist_node *p;
2706 int ret;
2707
2708 BUG_ON(dev_boot_phase);
2709 ASSERT_RTNL();
2710
2711 /* When net_device's are persistent, this will be fatal. */
2712 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2713
2714 spin_lock_init(&dev->queue_lock);
2715 spin_lock_init(&dev->xmit_lock);
2716 dev->xmit_lock_owner = -1;
2717#ifdef CONFIG_NET_CLS_ACT
2718 spin_lock_init(&dev->ingress_lock);
2719#endif
2720
2721 ret = alloc_divert_blk(dev);
2722 if (ret)
2723 goto out;
2724
2725 dev->iflink = -1;
2726
2727 /* Init, if this function is available */
2728 if (dev->init) {
2729 ret = dev->init(dev);
2730 if (ret) {
2731 if (ret > 0)
2732 ret = -EIO;
2733 goto out_err;
2734 }
2735 }
2736
2737 if (!dev_valid_name(dev->name)) {
2738 ret = -EINVAL;
2739 goto out_err;
2740 }
2741
2742 dev->ifindex = dev_new_index();
2743 if (dev->iflink == -1)
2744 dev->iflink = dev->ifindex;
2745
2746 /* Check for existence of name */
2747 head = dev_name_hash(dev->name);
2748 hlist_for_each(p, head) {
2749 struct net_device *d
2750 = hlist_entry(p, struct net_device, name_hlist);
2751 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2752 ret = -EEXIST;
2753 goto out_err;
2754 }
2755 }
2756
2757 /* Fix illegal SG+CSUM combinations. */
2758 if ((dev->features & NETIF_F_SG) &&
2759 !(dev->features & (NETIF_F_IP_CSUM |
2760 NETIF_F_NO_CSUM |
2761 NETIF_F_HW_CSUM))) {
2762 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2763 dev->name);
2764 dev->features &= ~NETIF_F_SG;
2765 }
2766
2767 /* TSO requires that SG is present as well. */
2768 if ((dev->features & NETIF_F_TSO) &&
2769 !(dev->features & NETIF_F_SG)) {
2770 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2771 dev->name);
2772 dev->features &= ~NETIF_F_TSO;
2773 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002774 if (dev->features & NETIF_F_UFO) {
2775 if (!(dev->features & NETIF_F_HW_CSUM)) {
2776 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2777 "NETIF_F_HW_CSUM feature.\n",
2778 dev->name);
2779 dev->features &= ~NETIF_F_UFO;
2780 }
2781 if (!(dev->features & NETIF_F_SG)) {
2782 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2783 "NETIF_F_SG feature.\n",
2784 dev->name);
2785 dev->features &= ~NETIF_F_UFO;
2786 }
2787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788
2789 /*
2790 * nil rebuild_header routine,
2791 * that should be never called and used as just bug trap.
2792 */
2793
2794 if (!dev->rebuild_header)
2795 dev->rebuild_header = default_rebuild_header;
2796
2797 /*
2798 * Default initial state at registry is that the
2799 * device is present.
2800 */
2801
2802 set_bit(__LINK_STATE_PRESENT, &dev->state);
2803
2804 dev->next = NULL;
2805 dev_init_scheduler(dev);
2806 write_lock_bh(&dev_base_lock);
2807 *dev_tail = dev;
2808 dev_tail = &dev->next;
2809 hlist_add_head(&dev->name_hlist, head);
2810 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2811 dev_hold(dev);
2812 dev->reg_state = NETREG_REGISTERING;
2813 write_unlock_bh(&dev_base_lock);
2814
2815 /* Notify protocols, that a new device appeared. */
2816 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2817
2818 /* Finish registration after unlock */
2819 net_set_todo(dev);
2820 ret = 0;
2821
2822out:
2823 return ret;
2824out_err:
2825 free_divert_blk(dev);
2826 goto out;
2827}
2828
2829/**
2830 * register_netdev - register a network device
2831 * @dev: device to register
2832 *
2833 * Take a completed network device structure and add it to the kernel
2834 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2835 * chain. 0 is returned on success. A negative errno code is returned
2836 * on a failure to set up the device, or if the name is a duplicate.
2837 *
2838 * This is a wrapper around register_netdev that takes the rtnl semaphore
2839 * and expands the device name if you passed a format string to
2840 * alloc_netdev.
2841 */
2842int register_netdev(struct net_device *dev)
2843{
2844 int err;
2845
2846 rtnl_lock();
2847
2848 /*
2849 * If the name is a format string the caller wants us to do a
2850 * name allocation.
2851 */
2852 if (strchr(dev->name, '%')) {
2853 err = dev_alloc_name(dev, dev->name);
2854 if (err < 0)
2855 goto out;
2856 }
2857
2858 /*
2859 * Back compatibility hook. Kill this one in 2.5
2860 */
2861 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2862 err = dev_alloc_name(dev, "eth%d");
2863 if (err < 0)
2864 goto out;
2865 }
2866
2867 err = register_netdevice(dev);
2868out:
2869 rtnl_unlock();
2870 return err;
2871}
2872EXPORT_SYMBOL(register_netdev);
2873
2874/*
2875 * netdev_wait_allrefs - wait until all references are gone.
2876 *
2877 * This is called when unregistering network devices.
2878 *
2879 * Any protocol or device that holds a reference should register
2880 * for netdevice notification, and cleanup and put back the
2881 * reference if they receive an UNREGISTER event.
2882 * We can get stuck here if buggy protocols don't correctly
2883 * call dev_put.
2884 */
2885static void netdev_wait_allrefs(struct net_device *dev)
2886{
2887 unsigned long rebroadcast_time, warning_time;
2888
2889 rebroadcast_time = warning_time = jiffies;
2890 while (atomic_read(&dev->refcnt) != 0) {
2891 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002892 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893
2894 /* Rebroadcast unregister notification */
2895 notifier_call_chain(&netdev_chain,
2896 NETDEV_UNREGISTER, dev);
2897
2898 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2899 &dev->state)) {
2900 /* We must not have linkwatch events
2901 * pending on unregister. If this
2902 * happens, we simply run the queue
2903 * unscheduled, resulting in a noop
2904 * for this device.
2905 */
2906 linkwatch_run_queue();
2907 }
2908
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002909 __rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910
2911 rebroadcast_time = jiffies;
2912 }
2913
2914 msleep(250);
2915
2916 if (time_after(jiffies, warning_time + 10 * HZ)) {
2917 printk(KERN_EMERG "unregister_netdevice: "
2918 "waiting for %s to become free. Usage "
2919 "count = %d\n",
2920 dev->name, atomic_read(&dev->refcnt));
2921 warning_time = jiffies;
2922 }
2923 }
2924}
2925
2926/* The sequence is:
2927 *
2928 * rtnl_lock();
2929 * ...
2930 * register_netdevice(x1);
2931 * register_netdevice(x2);
2932 * ...
2933 * unregister_netdevice(y1);
2934 * unregister_netdevice(y2);
2935 * ...
2936 * rtnl_unlock();
2937 * free_netdev(y1);
2938 * free_netdev(y2);
2939 *
2940 * We are invoked by rtnl_unlock() after it drops the semaphore.
2941 * This allows us to deal with problems:
2942 * 1) We can create/delete sysfs objects which invoke hotplug
2943 * without deadlocking with linkwatch via keventd.
2944 * 2) Since we run with the RTNL semaphore not held, we can sleep
2945 * safely in order to wait for the netdev refcnt to drop to zero.
2946 */
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08002947static DEFINE_MUTEX(net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948void netdev_run_todo(void)
2949{
2950 struct list_head list = LIST_HEAD_INIT(list);
2951 int err;
2952
2953
2954 /* Need to guard against multiple cpu's getting out of order. */
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08002955 mutex_lock(&net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956
2957 /* Not safe to do outside the semaphore. We must not return
2958 * until all unregister events invoked by the local processor
2959 * have been completed (either by this todo run, or one on
2960 * another cpu).
2961 */
2962 if (list_empty(&net_todo_list))
2963 goto out;
2964
2965 /* Snapshot list, allow later requests */
2966 spin_lock(&net_todo_list_lock);
2967 list_splice_init(&net_todo_list, &list);
2968 spin_unlock(&net_todo_list_lock);
2969
2970 while (!list_empty(&list)) {
2971 struct net_device *dev
2972 = list_entry(list.next, struct net_device, todo_list);
2973 list_del(&dev->todo_list);
2974
2975 switch(dev->reg_state) {
2976 case NETREG_REGISTERING:
2977 err = netdev_register_sysfs(dev);
2978 if (err)
2979 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2980 dev->name, err);
2981 dev->reg_state = NETREG_REGISTERED;
2982 break;
2983
2984 case NETREG_UNREGISTERING:
2985 netdev_unregister_sysfs(dev);
2986 dev->reg_state = NETREG_UNREGISTERED;
2987
2988 netdev_wait_allrefs(dev);
2989
2990 /* paranoia */
2991 BUG_ON(atomic_read(&dev->refcnt));
2992 BUG_TRAP(!dev->ip_ptr);
2993 BUG_TRAP(!dev->ip6_ptr);
2994 BUG_TRAP(!dev->dn_ptr);
2995
2996
2997 /* It must be the very last action,
2998 * after this 'dev' may point to freed up memory.
2999 */
3000 if (dev->destructor)
3001 dev->destructor(dev);
3002 break;
3003
3004 default:
3005 printk(KERN_ERR "network todo '%s' but state %d\n",
3006 dev->name, dev->reg_state);
3007 break;
3008 }
3009 }
3010
3011out:
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08003012 mutex_unlock(&net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013}
3014
3015/**
3016 * alloc_netdev - allocate network device
3017 * @sizeof_priv: size of private data to allocate space for
3018 * @name: device name format string
3019 * @setup: callback to initialize device
3020 *
3021 * Allocates a struct net_device with private data area for driver use
3022 * and performs basic initialization.
3023 */
3024struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3025 void (*setup)(struct net_device *))
3026{
3027 void *p;
3028 struct net_device *dev;
3029 int alloc_size;
3030
3031 /* ensure 32-byte alignment of both the device and private area */
3032 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3033 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3034
3035 p = kmalloc(alloc_size, GFP_KERNEL);
3036 if (!p) {
3037 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3038 return NULL;
3039 }
3040 memset(p, 0, alloc_size);
3041
3042 dev = (struct net_device *)
3043 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3044 dev->padded = (char *)dev - (char *)p;
3045
3046 if (sizeof_priv)
3047 dev->priv = netdev_priv(dev);
3048
3049 setup(dev);
3050 strcpy(dev->name, name);
3051 return dev;
3052}
3053EXPORT_SYMBOL(alloc_netdev);
3054
3055/**
3056 * free_netdev - free network device
3057 * @dev: device
3058 *
3059 * This function does the last stage of destroying an allocated device
3060 * interface. The reference to the device object is released.
3061 * If this is the last reference then it will be freed.
3062 */
3063void free_netdev(struct net_device *dev)
3064{
3065#ifdef CONFIG_SYSFS
3066 /* Compatiablity with error handling in drivers */
3067 if (dev->reg_state == NETREG_UNINITIALIZED) {
3068 kfree((char *)dev - dev->padded);
3069 return;
3070 }
3071
3072 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3073 dev->reg_state = NETREG_RELEASED;
3074
3075 /* will free via class release */
3076 class_device_put(&dev->class_dev);
3077#else
3078 kfree((char *)dev - dev->padded);
3079#endif
3080}
3081
3082/* Synchronize with packet receive processing. */
3083void synchronize_net(void)
3084{
3085 might_sleep();
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07003086 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087}
3088
3089/**
3090 * unregister_netdevice - remove device from the kernel
3091 * @dev: device
3092 *
3093 * This function shuts down a device interface and removes it
3094 * from the kernel tables. On success 0 is returned, on a failure
3095 * a negative errno code is returned.
3096 *
3097 * Callers must hold the rtnl semaphore. You may want
3098 * unregister_netdev() instead of this.
3099 */
3100
3101int unregister_netdevice(struct net_device *dev)
3102{
3103 struct net_device *d, **dp;
3104
3105 BUG_ON(dev_boot_phase);
3106 ASSERT_RTNL();
3107
3108 /* Some devices call without registering for initialization unwind. */
3109 if (dev->reg_state == NETREG_UNINITIALIZED) {
3110 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3111 "was registered\n", dev->name, dev);
3112 return -ENODEV;
3113 }
3114
3115 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3116
3117 /* If device is running, close it first. */
3118 if (dev->flags & IFF_UP)
3119 dev_close(dev);
3120
3121 /* And unlink it from device chain. */
3122 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3123 if (d == dev) {
3124 write_lock_bh(&dev_base_lock);
3125 hlist_del(&dev->name_hlist);
3126 hlist_del(&dev->index_hlist);
3127 if (dev_tail == &dev->next)
3128 dev_tail = dp;
3129 *dp = d->next;
3130 write_unlock_bh(&dev_base_lock);
3131 break;
3132 }
3133 }
3134 if (!d) {
3135 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3136 dev->name);
3137 return -ENODEV;
3138 }
3139
3140 dev->reg_state = NETREG_UNREGISTERING;
3141
3142 synchronize_net();
3143
3144 /* Shutdown queueing discipline. */
3145 dev_shutdown(dev);
3146
3147
3148 /* Notify protocols, that we are about to destroy
3149 this device. They should clean all the things.
3150 */
3151 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3152
3153 /*
3154 * Flush the multicast chain
3155 */
3156 dev_mc_discard(dev);
3157
3158 if (dev->uninit)
3159 dev->uninit(dev);
3160
3161 /* Notifier chain MUST detach us from master device. */
3162 BUG_TRAP(!dev->master);
3163
3164 free_divert_blk(dev);
3165
3166 /* Finish processing unregister after unlock */
3167 net_set_todo(dev);
3168
3169 synchronize_net();
3170
3171 dev_put(dev);
3172 return 0;
3173}
3174
3175/**
3176 * unregister_netdev - remove device from the kernel
3177 * @dev: device
3178 *
3179 * This function shuts down a device interface and removes it
3180 * from the kernel tables. On success 0 is returned, on a failure
3181 * a negative errno code is returned.
3182 *
3183 * This is just a wrapper for unregister_netdevice that takes
3184 * the rtnl semaphore. In general you want to use this and not
3185 * unregister_netdevice.
3186 */
3187void unregister_netdev(struct net_device *dev)
3188{
3189 rtnl_lock();
3190 unregister_netdevice(dev);
3191 rtnl_unlock();
3192}
3193
3194EXPORT_SYMBOL(unregister_netdev);
3195
3196#ifdef CONFIG_HOTPLUG_CPU
3197static int dev_cpu_callback(struct notifier_block *nfb,
3198 unsigned long action,
3199 void *ocpu)
3200{
3201 struct sk_buff **list_skb;
3202 struct net_device **list_net;
3203 struct sk_buff *skb;
3204 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3205 struct softnet_data *sd, *oldsd;
3206
3207 if (action != CPU_DEAD)
3208 return NOTIFY_OK;
3209
3210 local_irq_disable();
3211 cpu = smp_processor_id();
3212 sd = &per_cpu(softnet_data, cpu);
3213 oldsd = &per_cpu(softnet_data, oldcpu);
3214
3215 /* Find end of our completion_queue. */
3216 list_skb = &sd->completion_queue;
3217 while (*list_skb)
3218 list_skb = &(*list_skb)->next;
3219 /* Append completion queue from offline CPU. */
3220 *list_skb = oldsd->completion_queue;
3221 oldsd->completion_queue = NULL;
3222
3223 /* Find end of our output_queue. */
3224 list_net = &sd->output_queue;
3225 while (*list_net)
3226 list_net = &(*list_net)->next_sched;
3227 /* Append output queue from offline CPU. */
3228 *list_net = oldsd->output_queue;
3229 oldsd->output_queue = NULL;
3230
3231 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3232 local_irq_enable();
3233
3234 /* Process offline CPU's input_pkt_queue */
3235 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3236 netif_rx(skb);
3237
3238 return NOTIFY_OK;
3239}
3240#endif /* CONFIG_HOTPLUG_CPU */
3241
3242
3243/*
3244 * Initialize the DEV module. At boot time this walks the device list and
3245 * unhooks any devices that fail to initialise (normally hardware not
3246 * present) and leaves us with a valid list of present and active devices.
3247 *
3248 */
3249
3250/*
3251 * This is called single threaded during boot, so no need
3252 * to take the rtnl semaphore.
3253 */
3254static int __init net_dev_init(void)
3255{
3256 int i, rc = -ENOMEM;
3257
3258 BUG_ON(!dev_boot_phase);
3259
3260 net_random_init();
3261
3262 if (dev_proc_init())
3263 goto out;
3264
3265 if (netdev_sysfs_init())
3266 goto out;
3267
3268 INIT_LIST_HEAD(&ptype_all);
3269 for (i = 0; i < 16; i++)
3270 INIT_LIST_HEAD(&ptype_base[i]);
3271
3272 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3273 INIT_HLIST_HEAD(&dev_name_head[i]);
3274
3275 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3276 INIT_HLIST_HEAD(&dev_index_head[i]);
3277
3278 /*
3279 * Initialise the packet receive queues.
3280 */
3281
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08003282 for_each_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 struct softnet_data *queue;
3284
3285 queue = &per_cpu(softnet_data, i);
3286 skb_queue_head_init(&queue->input_pkt_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287 queue->completion_queue = NULL;
3288 INIT_LIST_HEAD(&queue->poll_list);
3289 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3290 queue->backlog_dev.weight = weight_p;
3291 queue->backlog_dev.poll = process_backlog;
3292 atomic_set(&queue->backlog_dev.refcnt, 1);
3293 }
3294
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 dev_boot_phase = 0;
3296
3297 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3298 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3299
3300 hotcpu_notifier(dev_cpu_callback, 0);
3301 dst_init();
3302 dev_mcast_init();
3303 rc = 0;
3304out:
3305 return rc;
3306}
3307
3308subsys_initcall(net_dev_init);
3309
3310EXPORT_SYMBOL(__dev_get_by_index);
3311EXPORT_SYMBOL(__dev_get_by_name);
3312EXPORT_SYMBOL(__dev_remove_pack);
3313EXPORT_SYMBOL(__skb_linearize);
Mitch Williamsc2373ee2005-11-09 10:34:45 -08003314EXPORT_SYMBOL(dev_valid_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315EXPORT_SYMBOL(dev_add_pack);
3316EXPORT_SYMBOL(dev_alloc_name);
3317EXPORT_SYMBOL(dev_close);
3318EXPORT_SYMBOL(dev_get_by_flags);
3319EXPORT_SYMBOL(dev_get_by_index);
3320EXPORT_SYMBOL(dev_get_by_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321EXPORT_SYMBOL(dev_open);
3322EXPORT_SYMBOL(dev_queue_xmit);
3323EXPORT_SYMBOL(dev_remove_pack);
3324EXPORT_SYMBOL(dev_set_allmulti);
3325EXPORT_SYMBOL(dev_set_promiscuity);
3326EXPORT_SYMBOL(dev_change_flags);
3327EXPORT_SYMBOL(dev_set_mtu);
3328EXPORT_SYMBOL(dev_set_mac_address);
3329EXPORT_SYMBOL(free_netdev);
3330EXPORT_SYMBOL(netdev_boot_setup_check);
3331EXPORT_SYMBOL(netdev_set_master);
3332EXPORT_SYMBOL(netdev_state_change);
3333EXPORT_SYMBOL(netif_receive_skb);
3334EXPORT_SYMBOL(netif_rx);
3335EXPORT_SYMBOL(register_gifconf);
3336EXPORT_SYMBOL(register_netdevice);
3337EXPORT_SYMBOL(register_netdevice_notifier);
3338EXPORT_SYMBOL(skb_checksum_help);
3339EXPORT_SYMBOL(synchronize_net);
3340EXPORT_SYMBOL(unregister_netdevice);
3341EXPORT_SYMBOL(unregister_netdevice_notifier);
3342EXPORT_SYMBOL(net_enable_timestamp);
3343EXPORT_SYMBOL(net_disable_timestamp);
3344EXPORT_SYMBOL(dev_get_flags);
3345
3346#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3347EXPORT_SYMBOL(br_handle_frame_hook);
3348EXPORT_SYMBOL(br_fdb_get_hook);
3349EXPORT_SYMBOL(br_fdb_put_hook);
3350#endif
3351
3352#ifdef CONFIG_KMOD
3353EXPORT_SYMBOL(dev_load);
3354#endif
3355
3356EXPORT_PER_CPU_SYMBOL(softnet_data);