[NET]: Take RTNL when unregistering notifier
[deliverable/linux.git] / net / core / dev.c
1 /*
2 * NET3 Protocol independent device support routines.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Derived from the non IP parts of dev.c 1.0.19
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 *
14 * Additional Authors:
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
21 *
22 * Changes:
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
34 * drivers
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
44 * call a packet.
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
58 * 1 device.
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
66 * the backlog queue.
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
73 */
74
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/config.h>
80 #include <linux/cpu.h>
81 #include <linux/types.h>
82 #include <linux/kernel.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #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>
114 #include <linux/wireless.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117
118 /*
119 * The list of packet types we will receive (as opposed to discard)
120 * and the routines to invoke.
121 *
122 * Why 16. Because with 16 the only overlap we get on a hash of the
123 * low nibble of the protocol value is RARP/SNAP/X.25.
124 *
125 * NOTE: That is no longer true with the addition of VLAN tags. Not
126 * sure which should go first, but I bet it won't make much
127 * difference if we are running VLANs. The good news is that
128 * this protocol won't be in the list unless compiled in, so
129 * the average user (w/out VLANs) will not be adversly affected.
130 * --BLG
131 *
132 * 0800 IP
133 * 8100 802.1Q VLAN
134 * 0001 802.3
135 * 0002 AX.25
136 * 0004 802.2
137 * 8035 RARP
138 * 0005 SNAP
139 * 0805 X.25
140 * 0806 ARP
141 * 8137 IPX
142 * 0009 Localtalk
143 * 86DD IPv6
144 */
145
146 static DEFINE_SPINLOCK(ptype_lock);
147 static struct list_head ptype_base[16]; /* 16 way hashed list */
148 static struct list_head ptype_all; /* Taps */
149
150 /*
151 * The @dev_base list is protected by @dev_base_lock and the rtln
152 * semaphore.
153 *
154 * Pure readers hold dev_base_lock for reading.
155 *
156 * Writers must hold the rtnl semaphore while they loop through the
157 * dev_base list, and hold dev_base_lock for writing when they do the
158 * actual updates. This allows pure readers to access the list even
159 * while a writer is preparing to update it.
160 *
161 * To put it another way, dev_base_lock is held for writing only to
162 * protect against pure readers; the rtnl semaphore provides the
163 * protection against other writers.
164 *
165 * See, for example usages, register_netdevice() and
166 * unregister_netdevice(), which must be called with the rtnl
167 * semaphore held.
168 */
169 struct net_device *dev_base;
170 static struct net_device **dev_tail = &dev_base;
171 DEFINE_RWLOCK(dev_base_lock);
172
173 EXPORT_SYMBOL(dev_base);
174 EXPORT_SYMBOL(dev_base_lock);
175
176 #define NETDEV_HASHBITS 8
177 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
178 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
179
180 static inline struct hlist_head *dev_name_hash(const char *name)
181 {
182 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
183 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
184 }
185
186 static inline struct hlist_head *dev_index_hash(int ifindex)
187 {
188 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
189 }
190
191 /*
192 * Our notifier list
193 */
194
195 static struct notifier_block *netdev_chain;
196
197 /*
198 * Device drivers call our routines to queue packets here. We empty the
199 * queue in the local softnet handler.
200 */
201 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
202
203 #ifdef CONFIG_SYSFS
204 extern int netdev_sysfs_init(void);
205 extern int netdev_register_sysfs(struct net_device *);
206 extern void netdev_unregister_sysfs(struct net_device *);
207 #else
208 #define netdev_sysfs_init() (0)
209 #define netdev_register_sysfs(dev) (0)
210 #define netdev_unregister_sysfs(dev) do { } while(0)
211 #endif
212
213
214 /*******************************************************************************
215
216 Protocol management and registration routines
217
218 *******************************************************************************/
219
220 /*
221 * For efficiency
222 */
223
224 int netdev_nit;
225
226 /*
227 * Add a protocol ID to the list. Now that the input handler is
228 * smarter we can dispense with all the messy stuff that used to be
229 * here.
230 *
231 * BEWARE!!! Protocol handlers, mangling input packets,
232 * MUST BE last in hash buckets and checking protocol handlers
233 * MUST start from promiscuous ptype_all chain in net_bh.
234 * It is true now, do not change it.
235 * Explanation follows: if protocol handler, mangling packet, will
236 * be the first on list, it is not able to sense, that packet
237 * is cloned and should be copied-on-write, so that it will
238 * change it and subsequent readers will get broken packet.
239 * --ANK (980803)
240 */
241
242 /**
243 * dev_add_pack - add packet handler
244 * @pt: packet type declaration
245 *
246 * Add a protocol handler to the networking stack. The passed &packet_type
247 * is linked into kernel lists and may not be freed until it has been
248 * removed from the kernel lists.
249 *
250 * This call does not sleep therefore it can not
251 * guarantee all CPU's that are in middle of receiving packets
252 * will see the new packet type (until the next received packet).
253 */
254
255 void dev_add_pack(struct packet_type *pt)
256 {
257 int hash;
258
259 spin_lock_bh(&ptype_lock);
260 if (pt->type == htons(ETH_P_ALL)) {
261 netdev_nit++;
262 list_add_rcu(&pt->list, &ptype_all);
263 } else {
264 hash = ntohs(pt->type) & 15;
265 list_add_rcu(&pt->list, &ptype_base[hash]);
266 }
267 spin_unlock_bh(&ptype_lock);
268 }
269
270 /**
271 * __dev_remove_pack - remove packet handler
272 * @pt: packet type declaration
273 *
274 * Remove a protocol handler that was previously added to the kernel
275 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
276 * from the kernel lists and can be freed or reused once this function
277 * returns.
278 *
279 * The packet type might still be in use by receivers
280 * and must not be freed until after all the CPU's have gone
281 * through a quiescent state.
282 */
283 void __dev_remove_pack(struct packet_type *pt)
284 {
285 struct list_head *head;
286 struct packet_type *pt1;
287
288 spin_lock_bh(&ptype_lock);
289
290 if (pt->type == htons(ETH_P_ALL)) {
291 netdev_nit--;
292 head = &ptype_all;
293 } else
294 head = &ptype_base[ntohs(pt->type) & 15];
295
296 list_for_each_entry(pt1, head, list) {
297 if (pt == pt1) {
298 list_del_rcu(&pt->list);
299 goto out;
300 }
301 }
302
303 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
304 out:
305 spin_unlock_bh(&ptype_lock);
306 }
307 /**
308 * dev_remove_pack - remove packet handler
309 * @pt: packet type declaration
310 *
311 * Remove a protocol handler that was previously added to the kernel
312 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
313 * from the kernel lists and can be freed or reused once this function
314 * returns.
315 *
316 * This call sleeps to guarantee that no CPU is looking at the packet
317 * type after return.
318 */
319 void dev_remove_pack(struct packet_type *pt)
320 {
321 __dev_remove_pack(pt);
322
323 synchronize_net();
324 }
325
326 /******************************************************************************
327
328 Device Boot-time Settings Routines
329
330 *******************************************************************************/
331
332 /* Boot time configuration table */
333 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
334
335 /**
336 * netdev_boot_setup_add - add new setup entry
337 * @name: name of the device
338 * @map: configured settings for the device
339 *
340 * Adds new setup entry to the dev_boot_setup list. The function
341 * returns 0 on error and 1 on success. This is a generic routine to
342 * all netdevices.
343 */
344 static int netdev_boot_setup_add(char *name, struct ifmap *map)
345 {
346 struct netdev_boot_setup *s;
347 int i;
348
349 s = dev_boot_setup;
350 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
351 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
352 memset(s[i].name, 0, sizeof(s[i].name));
353 strcpy(s[i].name, name);
354 memcpy(&s[i].map, map, sizeof(s[i].map));
355 break;
356 }
357 }
358
359 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
360 }
361
362 /**
363 * netdev_boot_setup_check - check boot time settings
364 * @dev: the netdevice
365 *
366 * Check boot time settings for the device.
367 * The found settings are set for the device to be used
368 * later in the device probing.
369 * Returns 0 if no settings found, 1 if they are.
370 */
371 int netdev_boot_setup_check(struct net_device *dev)
372 {
373 struct netdev_boot_setup *s = dev_boot_setup;
374 int i;
375
376 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
377 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
378 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
379 dev->irq = s[i].map.irq;
380 dev->base_addr = s[i].map.base_addr;
381 dev->mem_start = s[i].map.mem_start;
382 dev->mem_end = s[i].map.mem_end;
383 return 1;
384 }
385 }
386 return 0;
387 }
388
389
390 /**
391 * netdev_boot_base - get address from boot time settings
392 * @prefix: prefix for network device
393 * @unit: id for network device
394 *
395 * Check boot time settings for the base address of device.
396 * The found settings are set for the device to be used
397 * later in the device probing.
398 * Returns 0 if no settings found.
399 */
400 unsigned long netdev_boot_base(const char *prefix, int unit)
401 {
402 const struct netdev_boot_setup *s = dev_boot_setup;
403 char name[IFNAMSIZ];
404 int i;
405
406 sprintf(name, "%s%d", prefix, unit);
407
408 /*
409 * If device already registered then return base of 1
410 * to indicate not to probe for this interface
411 */
412 if (__dev_get_by_name(name))
413 return 1;
414
415 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
416 if (!strcmp(name, s[i].name))
417 return s[i].map.base_addr;
418 return 0;
419 }
420
421 /*
422 * Saves at boot time configured settings for any netdevice.
423 */
424 int __init netdev_boot_setup(char *str)
425 {
426 int ints[5];
427 struct ifmap map;
428
429 str = get_options(str, ARRAY_SIZE(ints), ints);
430 if (!str || !*str)
431 return 0;
432
433 /* Save settings */
434 memset(&map, 0, sizeof(map));
435 if (ints[0] > 0)
436 map.irq = ints[1];
437 if (ints[0] > 1)
438 map.base_addr = ints[2];
439 if (ints[0] > 2)
440 map.mem_start = ints[3];
441 if (ints[0] > 3)
442 map.mem_end = ints[4];
443
444 /* Add new entry to the list */
445 return netdev_boot_setup_add(str, &map);
446 }
447
448 __setup("netdev=", netdev_boot_setup);
449
450 /*******************************************************************************
451
452 Device Interface Subroutines
453
454 *******************************************************************************/
455
456 /**
457 * __dev_get_by_name - find a device by its name
458 * @name: name to find
459 *
460 * Find an interface by name. Must be called under RTNL semaphore
461 * or @dev_base_lock. If the name is found a pointer to the device
462 * is returned. If the name is not found then %NULL is returned. The
463 * reference counters are not incremented so the caller must be
464 * careful with locks.
465 */
466
467 struct net_device *__dev_get_by_name(const char *name)
468 {
469 struct hlist_node *p;
470
471 hlist_for_each(p, dev_name_hash(name)) {
472 struct net_device *dev
473 = hlist_entry(p, struct net_device, name_hlist);
474 if (!strncmp(dev->name, name, IFNAMSIZ))
475 return dev;
476 }
477 return NULL;
478 }
479
480 /**
481 * dev_get_by_name - find a device by its name
482 * @name: name to find
483 *
484 * Find an interface by name. This can be called from any
485 * context and does its own locking. The returned handle has
486 * the usage count incremented and the caller must use dev_put() to
487 * release it when it is no longer needed. %NULL is returned if no
488 * matching device is found.
489 */
490
491 struct net_device *dev_get_by_name(const char *name)
492 {
493 struct net_device *dev;
494
495 read_lock(&dev_base_lock);
496 dev = __dev_get_by_name(name);
497 if (dev)
498 dev_hold(dev);
499 read_unlock(&dev_base_lock);
500 return dev;
501 }
502
503 /**
504 * __dev_get_by_index - find a device by its ifindex
505 * @ifindex: index of device
506 *
507 * Search for an interface by index. Returns %NULL if the device
508 * is not found or a pointer to the device. The device has not
509 * had its reference counter increased so the caller must be careful
510 * about locking. The caller must hold either the RTNL semaphore
511 * or @dev_base_lock.
512 */
513
514 struct net_device *__dev_get_by_index(int ifindex)
515 {
516 struct hlist_node *p;
517
518 hlist_for_each(p, dev_index_hash(ifindex)) {
519 struct net_device *dev
520 = hlist_entry(p, struct net_device, index_hlist);
521 if (dev->ifindex == ifindex)
522 return dev;
523 }
524 return NULL;
525 }
526
527
528 /**
529 * dev_get_by_index - find a device by its ifindex
530 * @ifindex: index of device
531 *
532 * Search for an interface by index. Returns NULL if the device
533 * is not found or a pointer to the device. The device returned has
534 * had a reference added and the pointer is safe until the user calls
535 * dev_put to indicate they have finished with it.
536 */
537
538 struct net_device *dev_get_by_index(int ifindex)
539 {
540 struct net_device *dev;
541
542 read_lock(&dev_base_lock);
543 dev = __dev_get_by_index(ifindex);
544 if (dev)
545 dev_hold(dev);
546 read_unlock(&dev_base_lock);
547 return dev;
548 }
549
550 /**
551 * dev_getbyhwaddr - find a device by its hardware address
552 * @type: media type of device
553 * @ha: hardware address
554 *
555 * Search for an interface by MAC address. Returns NULL if the device
556 * is not found or a pointer to the device. The caller must hold the
557 * rtnl semaphore. The returned device has not had its ref count increased
558 * and the caller must therefore be careful about locking
559 *
560 * BUGS:
561 * If the API was consistent this would be __dev_get_by_hwaddr
562 */
563
564 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
565 {
566 struct net_device *dev;
567
568 ASSERT_RTNL();
569
570 for (dev = dev_base; dev; dev = dev->next)
571 if (dev->type == type &&
572 !memcmp(dev->dev_addr, ha, dev->addr_len))
573 break;
574 return dev;
575 }
576
577 EXPORT_SYMBOL(dev_getbyhwaddr);
578
579 struct net_device *dev_getfirstbyhwtype(unsigned short type)
580 {
581 struct net_device *dev;
582
583 rtnl_lock();
584 for (dev = dev_base; dev; dev = dev->next) {
585 if (dev->type == type) {
586 dev_hold(dev);
587 break;
588 }
589 }
590 rtnl_unlock();
591 return dev;
592 }
593
594 EXPORT_SYMBOL(dev_getfirstbyhwtype);
595
596 /**
597 * dev_get_by_flags - find any device with given flags
598 * @if_flags: IFF_* values
599 * @mask: bitmask of bits in if_flags to check
600 *
601 * Search for any interface with the given flags. Returns NULL if a device
602 * is not found or a pointer to the device. The device returned has
603 * had a reference added and the pointer is safe until the user calls
604 * dev_put to indicate they have finished with it.
605 */
606
607 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
608 {
609 struct net_device *dev;
610
611 read_lock(&dev_base_lock);
612 for (dev = dev_base; dev != NULL; dev = dev->next) {
613 if (((dev->flags ^ if_flags) & mask) == 0) {
614 dev_hold(dev);
615 break;
616 }
617 }
618 read_unlock(&dev_base_lock);
619 return dev;
620 }
621
622 /**
623 * dev_valid_name - check if name is okay for network device
624 * @name: name string
625 *
626 * Network device names need to be valid file names to
627 * to allow sysfs to work
628 */
629 int dev_valid_name(const char *name)
630 {
631 return !(*name == '\0'
632 || !strcmp(name, ".")
633 || !strcmp(name, "..")
634 || strchr(name, '/'));
635 }
636
637 /**
638 * dev_alloc_name - allocate a name for a device
639 * @dev: device
640 * @name: name format string
641 *
642 * Passed a format string - eg "lt%d" it will try and find a suitable
643 * id. Not efficient for many devices, not called a lot. The caller
644 * must hold the dev_base or rtnl lock while allocating the name and
645 * adding the device in order to avoid duplicates. Returns the number
646 * of the unit assigned or a negative errno code.
647 */
648
649 int dev_alloc_name(struct net_device *dev, const char *name)
650 {
651 int i = 0;
652 char buf[IFNAMSIZ];
653 const char *p;
654 const int max_netdevices = 8*PAGE_SIZE;
655 long *inuse;
656 struct net_device *d;
657
658 p = strnchr(name, IFNAMSIZ-1, '%');
659 if (p) {
660 /*
661 * Verify the string as this thing may have come from
662 * the user. There must be either one "%d" and no other "%"
663 * characters.
664 */
665 if (p[1] != 'd' || strchr(p + 2, '%'))
666 return -EINVAL;
667
668 /* Use one page as a bit array of possible slots */
669 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
670 if (!inuse)
671 return -ENOMEM;
672
673 for (d = dev_base; d; d = d->next) {
674 if (!sscanf(d->name, name, &i))
675 continue;
676 if (i < 0 || i >= max_netdevices)
677 continue;
678
679 /* avoid cases where sscanf is not exact inverse of printf */
680 snprintf(buf, sizeof(buf), name, i);
681 if (!strncmp(buf, d->name, IFNAMSIZ))
682 set_bit(i, inuse);
683 }
684
685 i = find_first_zero_bit(inuse, max_netdevices);
686 free_page((unsigned long) inuse);
687 }
688
689 snprintf(buf, sizeof(buf), name, i);
690 if (!__dev_get_by_name(buf)) {
691 strlcpy(dev->name, buf, IFNAMSIZ);
692 return i;
693 }
694
695 /* It is possible to run out of possible slots
696 * when the name is long and there isn't enough space left
697 * for the digits, or if all bits are used.
698 */
699 return -ENFILE;
700 }
701
702
703 /**
704 * dev_change_name - change name of a device
705 * @dev: device
706 * @newname: name (or format string) must be at least IFNAMSIZ
707 *
708 * Change name of a device, can pass format strings "eth%d".
709 * for wildcarding.
710 */
711 int dev_change_name(struct net_device *dev, char *newname)
712 {
713 int err = 0;
714
715 ASSERT_RTNL();
716
717 if (dev->flags & IFF_UP)
718 return -EBUSY;
719
720 if (!dev_valid_name(newname))
721 return -EINVAL;
722
723 if (strchr(newname, '%')) {
724 err = dev_alloc_name(dev, newname);
725 if (err < 0)
726 return err;
727 strcpy(newname, dev->name);
728 }
729 else if (__dev_get_by_name(newname))
730 return -EEXIST;
731 else
732 strlcpy(dev->name, newname, IFNAMSIZ);
733
734 err = class_device_rename(&dev->class_dev, dev->name);
735 if (!err) {
736 hlist_del(&dev->name_hlist);
737 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
738 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
739 }
740
741 return err;
742 }
743
744 /**
745 * netdev_features_change - device changes fatures
746 * @dev: device to cause notification
747 *
748 * Called to indicate a device has changed features.
749 */
750 void netdev_features_change(struct net_device *dev)
751 {
752 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
753 }
754 EXPORT_SYMBOL(netdev_features_change);
755
756 /**
757 * netdev_state_change - device changes state
758 * @dev: device to cause notification
759 *
760 * Called to indicate a device has changed state. This function calls
761 * the notifier chains for netdev_chain and sends a NEWLINK message
762 * to the routing socket.
763 */
764 void netdev_state_change(struct net_device *dev)
765 {
766 if (dev->flags & IFF_UP) {
767 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
768 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
769 }
770 }
771
772 /**
773 * dev_load - load a network module
774 * @name: name of interface
775 *
776 * If a network interface is not present and the process has suitable
777 * privileges this function loads the module. If module loading is not
778 * available in this kernel then it becomes a nop.
779 */
780
781 void dev_load(const char *name)
782 {
783 struct net_device *dev;
784
785 read_lock(&dev_base_lock);
786 dev = __dev_get_by_name(name);
787 read_unlock(&dev_base_lock);
788
789 if (!dev && capable(CAP_SYS_MODULE))
790 request_module("%s", name);
791 }
792
793 static int default_rebuild_header(struct sk_buff *skb)
794 {
795 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
796 skb->dev ? skb->dev->name : "NULL!!!");
797 kfree_skb(skb);
798 return 1;
799 }
800
801
802 /**
803 * dev_open - prepare an interface for use.
804 * @dev: device to open
805 *
806 * Takes a device from down to up state. The device's private open
807 * function is invoked and then the multicast lists are loaded. Finally
808 * the device is moved into the up state and a %NETDEV_UP message is
809 * sent to the netdev notifier chain.
810 *
811 * Calling this function on an active interface is a nop. On a failure
812 * a negative errno code is returned.
813 */
814 int dev_open(struct net_device *dev)
815 {
816 int ret = 0;
817
818 /*
819 * Is it already up?
820 */
821
822 if (dev->flags & IFF_UP)
823 return 0;
824
825 /*
826 * Is it even present?
827 */
828 if (!netif_device_present(dev))
829 return -ENODEV;
830
831 /*
832 * Call device private open method
833 */
834 set_bit(__LINK_STATE_START, &dev->state);
835 if (dev->open) {
836 ret = dev->open(dev);
837 if (ret)
838 clear_bit(__LINK_STATE_START, &dev->state);
839 }
840
841 /*
842 * If it went open OK then:
843 */
844
845 if (!ret) {
846 /*
847 * Set the flags.
848 */
849 dev->flags |= IFF_UP;
850
851 /*
852 * Initialize multicasting status
853 */
854 dev_mc_upload(dev);
855
856 /*
857 * Wakeup transmit queue engine
858 */
859 dev_activate(dev);
860
861 /*
862 * ... and announce new interface.
863 */
864 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
865 }
866 return ret;
867 }
868
869 /**
870 * dev_close - shutdown an interface.
871 * @dev: device to shutdown
872 *
873 * This function moves an active device into down state. A
874 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
875 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
876 * chain.
877 */
878 int dev_close(struct net_device *dev)
879 {
880 if (!(dev->flags & IFF_UP))
881 return 0;
882
883 /*
884 * Tell people we are going down, so that they can
885 * prepare to death, when device is still operating.
886 */
887 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
888
889 dev_deactivate(dev);
890
891 clear_bit(__LINK_STATE_START, &dev->state);
892
893 /* Synchronize to scheduled poll. We cannot touch poll list,
894 * it can be even on different cpu. So just clear netif_running(),
895 * and wait when poll really will happen. Actually, the best place
896 * for this is inside dev->stop() after device stopped its irq
897 * engine, but this requires more changes in devices. */
898
899 smp_mb__after_clear_bit(); /* Commit netif_running(). */
900 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
901 /* No hurry. */
902 msleep(1);
903 }
904
905 /*
906 * Call the device specific close. This cannot fail.
907 * Only if device is UP
908 *
909 * We allow it to be called even after a DETACH hot-plug
910 * event.
911 */
912 if (dev->stop)
913 dev->stop(dev);
914
915 /*
916 * Device is now down.
917 */
918
919 dev->flags &= ~IFF_UP;
920
921 /*
922 * Tell people we are down
923 */
924 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
925
926 return 0;
927 }
928
929
930 /*
931 * Device change register/unregister. These are not inline or static
932 * as we export them to the world.
933 */
934
935 /**
936 * register_netdevice_notifier - register a network notifier block
937 * @nb: notifier
938 *
939 * Register a notifier to be called when network device events occur.
940 * The notifier passed is linked into the kernel structures and must
941 * not be reused until it has been unregistered. A negative errno code
942 * is returned on a failure.
943 *
944 * When registered all registration and up events are replayed
945 * to the new notifier to allow device to have a race free
946 * view of the network device list.
947 */
948
949 int register_netdevice_notifier(struct notifier_block *nb)
950 {
951 struct net_device *dev;
952 int err;
953
954 rtnl_lock();
955 err = notifier_chain_register(&netdev_chain, nb);
956 if (!err) {
957 for (dev = dev_base; dev; dev = dev->next) {
958 nb->notifier_call(nb, NETDEV_REGISTER, dev);
959
960 if (dev->flags & IFF_UP)
961 nb->notifier_call(nb, NETDEV_UP, dev);
962 }
963 }
964 rtnl_unlock();
965 return err;
966 }
967
968 /**
969 * unregister_netdevice_notifier - unregister a network notifier block
970 * @nb: notifier
971 *
972 * Unregister a notifier previously registered by
973 * register_netdevice_notifier(). The notifier is unlinked into the
974 * kernel structures and may then be reused. A negative errno code
975 * is returned on a failure.
976 */
977
978 int unregister_netdevice_notifier(struct notifier_block *nb)
979 {
980 int err;
981
982 rtnl_lock();
983 err = notifier_chain_unregister(&netdev_chain, nb);
984 rtnl_unlock();
985 return err;
986 }
987
988 /**
989 * call_netdevice_notifiers - call all network notifier blocks
990 * @val: value passed unmodified to notifier function
991 * @v: pointer passed unmodified to notifier function
992 *
993 * Call all network notifier blocks. Parameters and return value
994 * are as for notifier_call_chain().
995 */
996
997 int call_netdevice_notifiers(unsigned long val, void *v)
998 {
999 return notifier_call_chain(&netdev_chain, val, v);
1000 }
1001
1002 /* When > 0 there are consumers of rx skb time stamps */
1003 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1004
1005 void net_enable_timestamp(void)
1006 {
1007 atomic_inc(&netstamp_needed);
1008 }
1009
1010 void net_disable_timestamp(void)
1011 {
1012 atomic_dec(&netstamp_needed);
1013 }
1014
1015 void __net_timestamp(struct sk_buff *skb)
1016 {
1017 struct timeval tv;
1018
1019 do_gettimeofday(&tv);
1020 skb_set_timestamp(skb, &tv);
1021 }
1022 EXPORT_SYMBOL(__net_timestamp);
1023
1024 static inline void net_timestamp(struct sk_buff *skb)
1025 {
1026 if (atomic_read(&netstamp_needed))
1027 __net_timestamp(skb);
1028 else {
1029 skb->tstamp.off_sec = 0;
1030 skb->tstamp.off_usec = 0;
1031 }
1032 }
1033
1034 /*
1035 * Support routine. Sends outgoing frames to any network
1036 * taps currently in use.
1037 */
1038
1039 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1040 {
1041 struct packet_type *ptype;
1042
1043 net_timestamp(skb);
1044
1045 rcu_read_lock();
1046 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1047 /* Never send packets back to the socket
1048 * they originated from - MvS (miquels@drinkel.ow.org)
1049 */
1050 if ((ptype->dev == dev || !ptype->dev) &&
1051 (ptype->af_packet_priv == NULL ||
1052 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1053 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1054 if (!skb2)
1055 break;
1056
1057 /* skb->nh should be correctly
1058 set by sender, so that the second statement is
1059 just protection against buggy protocols.
1060 */
1061 skb2->mac.raw = skb2->data;
1062
1063 if (skb2->nh.raw < skb2->data ||
1064 skb2->nh.raw > skb2->tail) {
1065 if (net_ratelimit())
1066 printk(KERN_CRIT "protocol %04x is "
1067 "buggy, dev %s\n",
1068 skb2->protocol, dev->name);
1069 skb2->nh.raw = skb2->data;
1070 }
1071
1072 skb2->h.raw = skb2->nh.raw;
1073 skb2->pkt_type = PACKET_OUTGOING;
1074 ptype->func(skb2, skb->dev, ptype, skb->dev);
1075 }
1076 }
1077 rcu_read_unlock();
1078 }
1079
1080 /*
1081 * Invalidate hardware checksum when packet is to be mangled, and
1082 * complete checksum manually on outgoing path.
1083 */
1084 int skb_checksum_help(struct sk_buff *skb, int inward)
1085 {
1086 unsigned int csum;
1087 int ret = 0, offset = skb->h.raw - skb->data;
1088
1089 if (inward) {
1090 skb->ip_summed = CHECKSUM_NONE;
1091 goto out;
1092 }
1093
1094 if (skb_cloned(skb)) {
1095 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1096 if (ret)
1097 goto out;
1098 }
1099
1100 BUG_ON(offset > (int)skb->len);
1101 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1102
1103 offset = skb->tail - skb->h.raw;
1104 BUG_ON(offset <= 0);
1105 BUG_ON(skb->csum + 2 > offset);
1106
1107 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1108 skb->ip_summed = CHECKSUM_NONE;
1109 out:
1110 return ret;
1111 }
1112
1113 /* Take action when hardware reception checksum errors are detected. */
1114 #ifdef CONFIG_BUG
1115 void netdev_rx_csum_fault(struct net_device *dev)
1116 {
1117 if (net_ratelimit()) {
1118 printk(KERN_ERR "%s: hw csum failure.\n",
1119 dev ? dev->name : "<unknown>");
1120 dump_stack();
1121 }
1122 }
1123 EXPORT_SYMBOL(netdev_rx_csum_fault);
1124 #endif
1125
1126 #ifdef CONFIG_HIGHMEM
1127 /* Actually, we should eliminate this check as soon as we know, that:
1128 * 1. IOMMU is present and allows to map all the memory.
1129 * 2. No high memory really exists on this machine.
1130 */
1131
1132 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1133 {
1134 int i;
1135
1136 if (dev->features & NETIF_F_HIGHDMA)
1137 return 0;
1138
1139 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1140 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1141 return 1;
1142
1143 return 0;
1144 }
1145 #else
1146 #define illegal_highdma(dev, skb) (0)
1147 #endif
1148
1149 /* Keep head the same: replace data */
1150 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1151 {
1152 unsigned int size;
1153 u8 *data;
1154 long offset;
1155 struct skb_shared_info *ninfo;
1156 int headerlen = skb->data - skb->head;
1157 int expand = (skb->tail + skb->data_len) - skb->end;
1158
1159 if (skb_shared(skb))
1160 BUG();
1161
1162 if (expand <= 0)
1163 expand = 0;
1164
1165 size = skb->end - skb->head + expand;
1166 size = SKB_DATA_ALIGN(size);
1167 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1168 if (!data)
1169 return -ENOMEM;
1170
1171 /* Copy entire thing */
1172 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1173 BUG();
1174
1175 /* Set up shinfo */
1176 ninfo = (struct skb_shared_info*)(data + size);
1177 atomic_set(&ninfo->dataref, 1);
1178 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1179 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1180 ninfo->nr_frags = 0;
1181 ninfo->frag_list = NULL;
1182
1183 /* Offset between the two in bytes */
1184 offset = data - skb->head;
1185
1186 /* Free old data. */
1187 skb_release_data(skb);
1188
1189 skb->head = data;
1190 skb->end = data + size;
1191
1192 /* Set up new pointers */
1193 skb->h.raw += offset;
1194 skb->nh.raw += offset;
1195 skb->mac.raw += offset;
1196 skb->tail += offset;
1197 skb->data += offset;
1198
1199 /* We are no longer a clone, even if we were. */
1200 skb->cloned = 0;
1201
1202 skb->tail += skb->data_len;
1203 skb->data_len = 0;
1204 return 0;
1205 }
1206
1207 #define HARD_TX_LOCK(dev, cpu) { \
1208 if ((dev->features & NETIF_F_LLTX) == 0) { \
1209 spin_lock(&dev->xmit_lock); \
1210 dev->xmit_lock_owner = cpu; \
1211 } \
1212 }
1213
1214 #define HARD_TX_UNLOCK(dev) { \
1215 if ((dev->features & NETIF_F_LLTX) == 0) { \
1216 dev->xmit_lock_owner = -1; \
1217 spin_unlock(&dev->xmit_lock); \
1218 } \
1219 }
1220
1221 /**
1222 * dev_queue_xmit - transmit a buffer
1223 * @skb: buffer to transmit
1224 *
1225 * Queue a buffer for transmission to a network device. The caller must
1226 * have set the device and priority and built the buffer before calling
1227 * this function. The function can be called from an interrupt.
1228 *
1229 * A negative errno code is returned on a failure. A success does not
1230 * guarantee the frame will be transmitted as it may be dropped due
1231 * to congestion or traffic shaping.
1232 *
1233 * -----------------------------------------------------------------------------------
1234 * I notice this method can also return errors from the queue disciplines,
1235 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1236 * be positive.
1237 *
1238 * Regardless of the return value, the skb is consumed, so it is currently
1239 * difficult to retry a send to this method. (You can bump the ref count
1240 * before sending to hold a reference for retry if you are careful.)
1241 *
1242 * When calling this method, interrupts MUST be enabled. This is because
1243 * the BH enable code must have IRQs enabled so that it will not deadlock.
1244 * --BLG
1245 */
1246
1247 int dev_queue_xmit(struct sk_buff *skb)
1248 {
1249 struct net_device *dev = skb->dev;
1250 struct Qdisc *q;
1251 int rc = -ENOMEM;
1252
1253 if (skb_shinfo(skb)->frag_list &&
1254 !(dev->features & NETIF_F_FRAGLIST) &&
1255 __skb_linearize(skb, GFP_ATOMIC))
1256 goto out_kfree_skb;
1257
1258 /* Fragmented skb is linearized if device does not support SG,
1259 * or if at least one of fragments is in highmem and device
1260 * does not support DMA from it.
1261 */
1262 if (skb_shinfo(skb)->nr_frags &&
1263 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1264 __skb_linearize(skb, GFP_ATOMIC))
1265 goto out_kfree_skb;
1266
1267 /* If packet is not checksummed and device does not support
1268 * checksumming for this protocol, complete checksumming here.
1269 */
1270 if (skb->ip_summed == CHECKSUM_HW &&
1271 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1272 (!(dev->features & NETIF_F_IP_CSUM) ||
1273 skb->protocol != htons(ETH_P_IP))))
1274 if (skb_checksum_help(skb, 0))
1275 goto out_kfree_skb;
1276
1277 spin_lock_prefetch(&dev->queue_lock);
1278
1279 /* Disable soft irqs for various locks below. Also
1280 * stops preemption for RCU.
1281 */
1282 local_bh_disable();
1283
1284 /* Updates of qdisc are serialized by queue_lock.
1285 * The struct Qdisc which is pointed to by qdisc is now a
1286 * rcu structure - it may be accessed without acquiring
1287 * a lock (but the structure may be stale.) The freeing of the
1288 * qdisc will be deferred until it's known that there are no
1289 * more references to it.
1290 *
1291 * If the qdisc has an enqueue function, we still need to
1292 * hold the queue_lock before calling it, since queue_lock
1293 * also serializes access to the device queue.
1294 */
1295
1296 q = rcu_dereference(dev->qdisc);
1297 #ifdef CONFIG_NET_CLS_ACT
1298 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1299 #endif
1300 if (q->enqueue) {
1301 /* Grab device queue */
1302 spin_lock(&dev->queue_lock);
1303
1304 rc = q->enqueue(skb, q);
1305
1306 qdisc_run(dev);
1307
1308 spin_unlock(&dev->queue_lock);
1309 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1310 goto out;
1311 }
1312
1313 /* The device has no queue. Common case for software devices:
1314 loopback, all the sorts of tunnels...
1315
1316 Really, it is unlikely that xmit_lock protection is necessary here.
1317 (f.e. loopback and IP tunnels are clean ignoring statistics
1318 counters.)
1319 However, it is possible, that they rely on protection
1320 made by us here.
1321
1322 Check this and shot the lock. It is not prone from deadlocks.
1323 Either shot noqueue qdisc, it is even simpler 8)
1324 */
1325 if (dev->flags & IFF_UP) {
1326 int cpu = smp_processor_id(); /* ok because BHs are off */
1327
1328 if (dev->xmit_lock_owner != cpu) {
1329
1330 HARD_TX_LOCK(dev, cpu);
1331
1332 if (!netif_queue_stopped(dev)) {
1333 if (netdev_nit)
1334 dev_queue_xmit_nit(skb, dev);
1335
1336 rc = 0;
1337 if (!dev->hard_start_xmit(skb, dev)) {
1338 HARD_TX_UNLOCK(dev);
1339 goto out;
1340 }
1341 }
1342 HARD_TX_UNLOCK(dev);
1343 if (net_ratelimit())
1344 printk(KERN_CRIT "Virtual device %s asks to "
1345 "queue packet!\n", dev->name);
1346 } else {
1347 /* Recursion is detected! It is possible,
1348 * unfortunately */
1349 if (net_ratelimit())
1350 printk(KERN_CRIT "Dead loop on virtual device "
1351 "%s, fix it urgently!\n", dev->name);
1352 }
1353 }
1354
1355 rc = -ENETDOWN;
1356 local_bh_enable();
1357
1358 out_kfree_skb:
1359 kfree_skb(skb);
1360 return rc;
1361 out:
1362 local_bh_enable();
1363 return rc;
1364 }
1365
1366
1367 /*=======================================================================
1368 Receiver routines
1369 =======================================================================*/
1370
1371 int netdev_max_backlog = 1000;
1372 int netdev_budget = 300;
1373 int weight_p = 64; /* old backlog weight */
1374
1375 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1376
1377
1378 /**
1379 * netif_rx - post buffer to the network code
1380 * @skb: buffer to post
1381 *
1382 * This function receives a packet from a device driver and queues it for
1383 * the upper (protocol) levels to process. It always succeeds. The buffer
1384 * may be dropped during processing for congestion control or by the
1385 * protocol layers.
1386 *
1387 * return values:
1388 * NET_RX_SUCCESS (no congestion)
1389 * NET_RX_CN_LOW (low congestion)
1390 * NET_RX_CN_MOD (moderate congestion)
1391 * NET_RX_CN_HIGH (high congestion)
1392 * NET_RX_DROP (packet was dropped)
1393 *
1394 */
1395
1396 int netif_rx(struct sk_buff *skb)
1397 {
1398 struct softnet_data *queue;
1399 unsigned long flags;
1400
1401 /* if netpoll wants it, pretend we never saw it */
1402 if (netpoll_rx(skb))
1403 return NET_RX_DROP;
1404
1405 if (!skb->tstamp.off_sec)
1406 net_timestamp(skb);
1407
1408 /*
1409 * The code is rearranged so that the path is the most
1410 * short when CPU is congested, but is still operating.
1411 */
1412 local_irq_save(flags);
1413 queue = &__get_cpu_var(softnet_data);
1414
1415 __get_cpu_var(netdev_rx_stat).total++;
1416 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1417 if (queue->input_pkt_queue.qlen) {
1418 enqueue:
1419 dev_hold(skb->dev);
1420 __skb_queue_tail(&queue->input_pkt_queue, skb);
1421 local_irq_restore(flags);
1422 return NET_RX_SUCCESS;
1423 }
1424
1425 netif_rx_schedule(&queue->backlog_dev);
1426 goto enqueue;
1427 }
1428
1429 __get_cpu_var(netdev_rx_stat).dropped++;
1430 local_irq_restore(flags);
1431
1432 kfree_skb(skb);
1433 return NET_RX_DROP;
1434 }
1435
1436 int netif_rx_ni(struct sk_buff *skb)
1437 {
1438 int err;
1439
1440 preempt_disable();
1441 err = netif_rx(skb);
1442 if (local_softirq_pending())
1443 do_softirq();
1444 preempt_enable();
1445
1446 return err;
1447 }
1448
1449 EXPORT_SYMBOL(netif_rx_ni);
1450
1451 static inline struct net_device *skb_bond(struct sk_buff *skb)
1452 {
1453 struct net_device *dev = skb->dev;
1454
1455 if (dev->master) {
1456 /*
1457 * On bonding slaves other than the currently active
1458 * slave, suppress duplicates except for 802.3ad
1459 * ETH_P_SLOW and alb non-mcast/bcast.
1460 */
1461 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1462 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1463 if (skb->pkt_type != PACKET_BROADCAST &&
1464 skb->pkt_type != PACKET_MULTICAST)
1465 goto keep;
1466 }
1467
1468 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1469 skb->protocol == __constant_htons(ETH_P_SLOW))
1470 goto keep;
1471
1472 kfree_skb(skb);
1473 return NULL;
1474 }
1475 keep:
1476 skb->dev = dev->master;
1477 }
1478
1479 return dev;
1480 }
1481
1482 static void net_tx_action(struct softirq_action *h)
1483 {
1484 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1485
1486 if (sd->completion_queue) {
1487 struct sk_buff *clist;
1488
1489 local_irq_disable();
1490 clist = sd->completion_queue;
1491 sd->completion_queue = NULL;
1492 local_irq_enable();
1493
1494 while (clist) {
1495 struct sk_buff *skb = clist;
1496 clist = clist->next;
1497
1498 BUG_TRAP(!atomic_read(&skb->users));
1499 __kfree_skb(skb);
1500 }
1501 }
1502
1503 if (sd->output_queue) {
1504 struct net_device *head;
1505
1506 local_irq_disable();
1507 head = sd->output_queue;
1508 sd->output_queue = NULL;
1509 local_irq_enable();
1510
1511 while (head) {
1512 struct net_device *dev = head;
1513 head = head->next_sched;
1514
1515 smp_mb__before_clear_bit();
1516 clear_bit(__LINK_STATE_SCHED, &dev->state);
1517
1518 if (spin_trylock(&dev->queue_lock)) {
1519 qdisc_run(dev);
1520 spin_unlock(&dev->queue_lock);
1521 } else {
1522 netif_schedule(dev);
1523 }
1524 }
1525 }
1526 }
1527
1528 static __inline__ int deliver_skb(struct sk_buff *skb,
1529 struct packet_type *pt_prev,
1530 struct net_device *orig_dev)
1531 {
1532 atomic_inc(&skb->users);
1533 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1534 }
1535
1536 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1537 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1538 struct net_bridge;
1539 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1540 unsigned char *addr);
1541 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1542
1543 static __inline__ int handle_bridge(struct sk_buff **pskb,
1544 struct packet_type **pt_prev, int *ret,
1545 struct net_device *orig_dev)
1546 {
1547 struct net_bridge_port *port;
1548
1549 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1550 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1551 return 0;
1552
1553 if (*pt_prev) {
1554 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1555 *pt_prev = NULL;
1556 }
1557
1558 return br_handle_frame_hook(port, pskb);
1559 }
1560 #else
1561 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1562 #endif
1563
1564 #ifdef CONFIG_NET_CLS_ACT
1565 /* TODO: Maybe we should just force sch_ingress to be compiled in
1566 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1567 * a compare and 2 stores extra right now if we dont have it on
1568 * but have CONFIG_NET_CLS_ACT
1569 * NOTE: This doesnt stop any functionality; if you dont have
1570 * the ingress scheduler, you just cant add policies on ingress.
1571 *
1572 */
1573 static int ing_filter(struct sk_buff *skb)
1574 {
1575 struct Qdisc *q;
1576 struct net_device *dev = skb->dev;
1577 int result = TC_ACT_OK;
1578
1579 if (dev->qdisc_ingress) {
1580 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1581 if (MAX_RED_LOOP < ttl++) {
1582 printk("Redir loop detected Dropping packet (%s->%s)\n",
1583 skb->input_dev->name, skb->dev->name);
1584 return TC_ACT_SHOT;
1585 }
1586
1587 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1588
1589 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1590
1591 spin_lock(&dev->ingress_lock);
1592 if ((q = dev->qdisc_ingress) != NULL)
1593 result = q->enqueue(skb, q);
1594 spin_unlock(&dev->ingress_lock);
1595
1596 }
1597
1598 return result;
1599 }
1600 #endif
1601
1602 int netif_receive_skb(struct sk_buff *skb)
1603 {
1604 struct packet_type *ptype, *pt_prev;
1605 struct net_device *orig_dev;
1606 int ret = NET_RX_DROP;
1607 unsigned short type;
1608
1609 /* if we've gotten here through NAPI, check netpoll */
1610 if (skb->dev->poll && netpoll_rx(skb))
1611 return NET_RX_DROP;
1612
1613 if (!skb->tstamp.off_sec)
1614 net_timestamp(skb);
1615
1616 if (!skb->input_dev)
1617 skb->input_dev = skb->dev;
1618
1619 orig_dev = skb_bond(skb);
1620
1621 if (!orig_dev)
1622 return NET_RX_DROP;
1623
1624 __get_cpu_var(netdev_rx_stat).total++;
1625
1626 skb->h.raw = skb->nh.raw = skb->data;
1627 skb->mac_len = skb->nh.raw - skb->mac.raw;
1628
1629 pt_prev = NULL;
1630
1631 rcu_read_lock();
1632
1633 #ifdef CONFIG_NET_CLS_ACT
1634 if (skb->tc_verd & TC_NCLS) {
1635 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1636 goto ncls;
1637 }
1638 #endif
1639
1640 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1641 if (!ptype->dev || ptype->dev == skb->dev) {
1642 if (pt_prev)
1643 ret = deliver_skb(skb, pt_prev, orig_dev);
1644 pt_prev = ptype;
1645 }
1646 }
1647
1648 #ifdef CONFIG_NET_CLS_ACT
1649 if (pt_prev) {
1650 ret = deliver_skb(skb, pt_prev, orig_dev);
1651 pt_prev = NULL; /* noone else should process this after*/
1652 } else {
1653 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1654 }
1655
1656 ret = ing_filter(skb);
1657
1658 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1659 kfree_skb(skb);
1660 goto out;
1661 }
1662
1663 skb->tc_verd = 0;
1664 ncls:
1665 #endif
1666
1667 handle_diverter(skb);
1668
1669 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1670 goto out;
1671
1672 type = skb->protocol;
1673 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1674 if (ptype->type == type &&
1675 (!ptype->dev || ptype->dev == skb->dev)) {
1676 if (pt_prev)
1677 ret = deliver_skb(skb, pt_prev, orig_dev);
1678 pt_prev = ptype;
1679 }
1680 }
1681
1682 if (pt_prev) {
1683 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1684 } else {
1685 kfree_skb(skb);
1686 /* Jamal, now you will not able to escape explaining
1687 * me how you were going to use this. :-)
1688 */
1689 ret = NET_RX_DROP;
1690 }
1691
1692 out:
1693 rcu_read_unlock();
1694 return ret;
1695 }
1696
1697 static int process_backlog(struct net_device *backlog_dev, int *budget)
1698 {
1699 int work = 0;
1700 int quota = min(backlog_dev->quota, *budget);
1701 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1702 unsigned long start_time = jiffies;
1703
1704 backlog_dev->weight = weight_p;
1705 for (;;) {
1706 struct sk_buff *skb;
1707 struct net_device *dev;
1708
1709 local_irq_disable();
1710 skb = __skb_dequeue(&queue->input_pkt_queue);
1711 if (!skb)
1712 goto job_done;
1713 local_irq_enable();
1714
1715 dev = skb->dev;
1716
1717 netif_receive_skb(skb);
1718
1719 dev_put(dev);
1720
1721 work++;
1722
1723 if (work >= quota || jiffies - start_time > 1)
1724 break;
1725
1726 }
1727
1728 backlog_dev->quota -= work;
1729 *budget -= work;
1730 return -1;
1731
1732 job_done:
1733 backlog_dev->quota -= work;
1734 *budget -= work;
1735
1736 list_del(&backlog_dev->poll_list);
1737 smp_mb__before_clear_bit();
1738 netif_poll_enable(backlog_dev);
1739
1740 local_irq_enable();
1741 return 0;
1742 }
1743
1744 static void net_rx_action(struct softirq_action *h)
1745 {
1746 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1747 unsigned long start_time = jiffies;
1748 int budget = netdev_budget;
1749 void *have;
1750
1751 local_irq_disable();
1752
1753 while (!list_empty(&queue->poll_list)) {
1754 struct net_device *dev;
1755
1756 if (budget <= 0 || jiffies - start_time > 1)
1757 goto softnet_break;
1758
1759 local_irq_enable();
1760
1761 dev = list_entry(queue->poll_list.next,
1762 struct net_device, poll_list);
1763 have = netpoll_poll_lock(dev);
1764
1765 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1766 netpoll_poll_unlock(have);
1767 local_irq_disable();
1768 list_move_tail(&dev->poll_list, &queue->poll_list);
1769 if (dev->quota < 0)
1770 dev->quota += dev->weight;
1771 else
1772 dev->quota = dev->weight;
1773 } else {
1774 netpoll_poll_unlock(have);
1775 dev_put(dev);
1776 local_irq_disable();
1777 }
1778 }
1779 out:
1780 local_irq_enable();
1781 return;
1782
1783 softnet_break:
1784 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1785 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1786 goto out;
1787 }
1788
1789 static gifconf_func_t * gifconf_list [NPROTO];
1790
1791 /**
1792 * register_gifconf - register a SIOCGIF handler
1793 * @family: Address family
1794 * @gifconf: Function handler
1795 *
1796 * Register protocol dependent address dumping routines. The handler
1797 * that is passed must not be freed or reused until it has been replaced
1798 * by another handler.
1799 */
1800 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1801 {
1802 if (family >= NPROTO)
1803 return -EINVAL;
1804 gifconf_list[family] = gifconf;
1805 return 0;
1806 }
1807
1808
1809 /*
1810 * Map an interface index to its name (SIOCGIFNAME)
1811 */
1812
1813 /*
1814 * We need this ioctl for efficient implementation of the
1815 * if_indextoname() function required by the IPv6 API. Without
1816 * it, we would have to search all the interfaces to find a
1817 * match. --pb
1818 */
1819
1820 static int dev_ifname(struct ifreq __user *arg)
1821 {
1822 struct net_device *dev;
1823 struct ifreq ifr;
1824
1825 /*
1826 * Fetch the caller's info block.
1827 */
1828
1829 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1830 return -EFAULT;
1831
1832 read_lock(&dev_base_lock);
1833 dev = __dev_get_by_index(ifr.ifr_ifindex);
1834 if (!dev) {
1835 read_unlock(&dev_base_lock);
1836 return -ENODEV;
1837 }
1838
1839 strcpy(ifr.ifr_name, dev->name);
1840 read_unlock(&dev_base_lock);
1841
1842 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1843 return -EFAULT;
1844 return 0;
1845 }
1846
1847 /*
1848 * Perform a SIOCGIFCONF call. This structure will change
1849 * size eventually, and there is nothing I can do about it.
1850 * Thus we will need a 'compatibility mode'.
1851 */
1852
1853 static int dev_ifconf(char __user *arg)
1854 {
1855 struct ifconf ifc;
1856 struct net_device *dev;
1857 char __user *pos;
1858 int len;
1859 int total;
1860 int i;
1861
1862 /*
1863 * Fetch the caller's info block.
1864 */
1865
1866 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1867 return -EFAULT;
1868
1869 pos = ifc.ifc_buf;
1870 len = ifc.ifc_len;
1871
1872 /*
1873 * Loop over the interfaces, and write an info block for each.
1874 */
1875
1876 total = 0;
1877 for (dev = dev_base; dev; dev = dev->next) {
1878 for (i = 0; i < NPROTO; i++) {
1879 if (gifconf_list[i]) {
1880 int done;
1881 if (!pos)
1882 done = gifconf_list[i](dev, NULL, 0);
1883 else
1884 done = gifconf_list[i](dev, pos + total,
1885 len - total);
1886 if (done < 0)
1887 return -EFAULT;
1888 total += done;
1889 }
1890 }
1891 }
1892
1893 /*
1894 * All done. Write the updated control block back to the caller.
1895 */
1896 ifc.ifc_len = total;
1897
1898 /*
1899 * Both BSD and Solaris return 0 here, so we do too.
1900 */
1901 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1902 }
1903
1904 #ifdef CONFIG_PROC_FS
1905 /*
1906 * This is invoked by the /proc filesystem handler to display a device
1907 * in detail.
1908 */
1909 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1910 {
1911 struct net_device *dev;
1912 loff_t i;
1913
1914 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1915
1916 return i == pos ? dev : NULL;
1917 }
1918
1919 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1920 {
1921 read_lock(&dev_base_lock);
1922 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1923 }
1924
1925 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1926 {
1927 ++*pos;
1928 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1929 }
1930
1931 void dev_seq_stop(struct seq_file *seq, void *v)
1932 {
1933 read_unlock(&dev_base_lock);
1934 }
1935
1936 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1937 {
1938 if (dev->get_stats) {
1939 struct net_device_stats *stats = dev->get_stats(dev);
1940
1941 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1942 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1943 dev->name, stats->rx_bytes, stats->rx_packets,
1944 stats->rx_errors,
1945 stats->rx_dropped + stats->rx_missed_errors,
1946 stats->rx_fifo_errors,
1947 stats->rx_length_errors + stats->rx_over_errors +
1948 stats->rx_crc_errors + stats->rx_frame_errors,
1949 stats->rx_compressed, stats->multicast,
1950 stats->tx_bytes, stats->tx_packets,
1951 stats->tx_errors, stats->tx_dropped,
1952 stats->tx_fifo_errors, stats->collisions,
1953 stats->tx_carrier_errors +
1954 stats->tx_aborted_errors +
1955 stats->tx_window_errors +
1956 stats->tx_heartbeat_errors,
1957 stats->tx_compressed);
1958 } else
1959 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1960 }
1961
1962 /*
1963 * Called from the PROCfs module. This now uses the new arbitrary sized
1964 * /proc/net interface to create /proc/net/dev
1965 */
1966 static int dev_seq_show(struct seq_file *seq, void *v)
1967 {
1968 if (v == SEQ_START_TOKEN)
1969 seq_puts(seq, "Inter-| Receive "
1970 " | Transmit\n"
1971 " face |bytes packets errs drop fifo frame "
1972 "compressed multicast|bytes packets errs "
1973 "drop fifo colls carrier compressed\n");
1974 else
1975 dev_seq_printf_stats(seq, v);
1976 return 0;
1977 }
1978
1979 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1980 {
1981 struct netif_rx_stats *rc = NULL;
1982
1983 while (*pos < NR_CPUS)
1984 if (cpu_online(*pos)) {
1985 rc = &per_cpu(netdev_rx_stat, *pos);
1986 break;
1987 } else
1988 ++*pos;
1989 return rc;
1990 }
1991
1992 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1993 {
1994 return softnet_get_online(pos);
1995 }
1996
1997 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1998 {
1999 ++*pos;
2000 return softnet_get_online(pos);
2001 }
2002
2003 static void softnet_seq_stop(struct seq_file *seq, void *v)
2004 {
2005 }
2006
2007 static int softnet_seq_show(struct seq_file *seq, void *v)
2008 {
2009 struct netif_rx_stats *s = v;
2010
2011 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2012 s->total, s->dropped, s->time_squeeze, 0,
2013 0, 0, 0, 0, /* was fastroute */
2014 s->cpu_collision );
2015 return 0;
2016 }
2017
2018 static struct seq_operations dev_seq_ops = {
2019 .start = dev_seq_start,
2020 .next = dev_seq_next,
2021 .stop = dev_seq_stop,
2022 .show = dev_seq_show,
2023 };
2024
2025 static int dev_seq_open(struct inode *inode, struct file *file)
2026 {
2027 return seq_open(file, &dev_seq_ops);
2028 }
2029
2030 static struct file_operations dev_seq_fops = {
2031 .owner = THIS_MODULE,
2032 .open = dev_seq_open,
2033 .read = seq_read,
2034 .llseek = seq_lseek,
2035 .release = seq_release,
2036 };
2037
2038 static struct seq_operations softnet_seq_ops = {
2039 .start = softnet_seq_start,
2040 .next = softnet_seq_next,
2041 .stop = softnet_seq_stop,
2042 .show = softnet_seq_show,
2043 };
2044
2045 static int softnet_seq_open(struct inode *inode, struct file *file)
2046 {
2047 return seq_open(file, &softnet_seq_ops);
2048 }
2049
2050 static struct file_operations softnet_seq_fops = {
2051 .owner = THIS_MODULE,
2052 .open = softnet_seq_open,
2053 .read = seq_read,
2054 .llseek = seq_lseek,
2055 .release = seq_release,
2056 };
2057
2058 #ifdef CONFIG_WIRELESS_EXT
2059 extern int wireless_proc_init(void);
2060 #else
2061 #define wireless_proc_init() 0
2062 #endif
2063
2064 static int __init dev_proc_init(void)
2065 {
2066 int rc = -ENOMEM;
2067
2068 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2069 goto out;
2070 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2071 goto out_dev;
2072 if (wireless_proc_init())
2073 goto out_softnet;
2074 rc = 0;
2075 out:
2076 return rc;
2077 out_softnet:
2078 proc_net_remove("softnet_stat");
2079 out_dev:
2080 proc_net_remove("dev");
2081 goto out;
2082 }
2083 #else
2084 #define dev_proc_init() 0
2085 #endif /* CONFIG_PROC_FS */
2086
2087
2088 /**
2089 * netdev_set_master - set up master/slave pair
2090 * @slave: slave device
2091 * @master: new master device
2092 *
2093 * Changes the master device of the slave. Pass %NULL to break the
2094 * bonding. The caller must hold the RTNL semaphore. On a failure
2095 * a negative errno code is returned. On success the reference counts
2096 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2097 * function returns zero.
2098 */
2099 int netdev_set_master(struct net_device *slave, struct net_device *master)
2100 {
2101 struct net_device *old = slave->master;
2102
2103 ASSERT_RTNL();
2104
2105 if (master) {
2106 if (old)
2107 return -EBUSY;
2108 dev_hold(master);
2109 }
2110
2111 slave->master = master;
2112
2113 synchronize_net();
2114
2115 if (old)
2116 dev_put(old);
2117
2118 if (master)
2119 slave->flags |= IFF_SLAVE;
2120 else
2121 slave->flags &= ~IFF_SLAVE;
2122
2123 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2124 return 0;
2125 }
2126
2127 /**
2128 * dev_set_promiscuity - update promiscuity count on a device
2129 * @dev: device
2130 * @inc: modifier
2131 *
2132 * Add or remove promsicuity from a device. While the count in the device
2133 * remains above zero the interface remains promiscuous. Once it hits zero
2134 * the device reverts back to normal filtering operation. A negative inc
2135 * value is used to drop promiscuity on the device.
2136 */
2137 void dev_set_promiscuity(struct net_device *dev, int inc)
2138 {
2139 unsigned short old_flags = dev->flags;
2140
2141 if ((dev->promiscuity += inc) == 0)
2142 dev->flags &= ~IFF_PROMISC;
2143 else
2144 dev->flags |= IFF_PROMISC;
2145 if (dev->flags != old_flags) {
2146 dev_mc_upload(dev);
2147 printk(KERN_INFO "device %s %s promiscuous mode\n",
2148 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2149 "left");
2150 }
2151 }
2152
2153 /**
2154 * dev_set_allmulti - update allmulti count on a device
2155 * @dev: device
2156 * @inc: modifier
2157 *
2158 * Add or remove reception of all multicast frames to a device. While the
2159 * count in the device remains above zero the interface remains listening
2160 * to all interfaces. Once it hits zero the device reverts back to normal
2161 * filtering operation. A negative @inc value is used to drop the counter
2162 * when releasing a resource needing all multicasts.
2163 */
2164
2165 void dev_set_allmulti(struct net_device *dev, int inc)
2166 {
2167 unsigned short old_flags = dev->flags;
2168
2169 dev->flags |= IFF_ALLMULTI;
2170 if ((dev->allmulti += inc) == 0)
2171 dev->flags &= ~IFF_ALLMULTI;
2172 if (dev->flags ^ old_flags)
2173 dev_mc_upload(dev);
2174 }
2175
2176 unsigned dev_get_flags(const struct net_device *dev)
2177 {
2178 unsigned flags;
2179
2180 flags = (dev->flags & ~(IFF_PROMISC |
2181 IFF_ALLMULTI |
2182 IFF_RUNNING |
2183 IFF_LOWER_UP |
2184 IFF_DORMANT)) |
2185 (dev->gflags & (IFF_PROMISC |
2186 IFF_ALLMULTI));
2187
2188 if (netif_running(dev)) {
2189 if (netif_oper_up(dev))
2190 flags |= IFF_RUNNING;
2191 if (netif_carrier_ok(dev))
2192 flags |= IFF_LOWER_UP;
2193 if (netif_dormant(dev))
2194 flags |= IFF_DORMANT;
2195 }
2196
2197 return flags;
2198 }
2199
2200 int dev_change_flags(struct net_device *dev, unsigned flags)
2201 {
2202 int ret;
2203 int old_flags = dev->flags;
2204
2205 /*
2206 * Set the flags on our device.
2207 */
2208
2209 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2210 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2211 IFF_AUTOMEDIA)) |
2212 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2213 IFF_ALLMULTI));
2214
2215 /*
2216 * Load in the correct multicast list now the flags have changed.
2217 */
2218
2219 dev_mc_upload(dev);
2220
2221 /*
2222 * Have we downed the interface. We handle IFF_UP ourselves
2223 * according to user attempts to set it, rather than blindly
2224 * setting it.
2225 */
2226
2227 ret = 0;
2228 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2229 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2230
2231 if (!ret)
2232 dev_mc_upload(dev);
2233 }
2234
2235 if (dev->flags & IFF_UP &&
2236 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2237 IFF_VOLATILE)))
2238 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2239
2240 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2241 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2242 dev->gflags ^= IFF_PROMISC;
2243 dev_set_promiscuity(dev, inc);
2244 }
2245
2246 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2247 is important. Some (broken) drivers set IFF_PROMISC, when
2248 IFF_ALLMULTI is requested not asking us and not reporting.
2249 */
2250 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2251 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2252 dev->gflags ^= IFF_ALLMULTI;
2253 dev_set_allmulti(dev, inc);
2254 }
2255
2256 if (old_flags ^ dev->flags)
2257 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2258
2259 return ret;
2260 }
2261
2262 int dev_set_mtu(struct net_device *dev, int new_mtu)
2263 {
2264 int err;
2265
2266 if (new_mtu == dev->mtu)
2267 return 0;
2268
2269 /* MTU must be positive. */
2270 if (new_mtu < 0)
2271 return -EINVAL;
2272
2273 if (!netif_device_present(dev))
2274 return -ENODEV;
2275
2276 err = 0;
2277 if (dev->change_mtu)
2278 err = dev->change_mtu(dev, new_mtu);
2279 else
2280 dev->mtu = new_mtu;
2281 if (!err && dev->flags & IFF_UP)
2282 notifier_call_chain(&netdev_chain,
2283 NETDEV_CHANGEMTU, dev);
2284 return err;
2285 }
2286
2287 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2288 {
2289 int err;
2290
2291 if (!dev->set_mac_address)
2292 return -EOPNOTSUPP;
2293 if (sa->sa_family != dev->type)
2294 return -EINVAL;
2295 if (!netif_device_present(dev))
2296 return -ENODEV;
2297 err = dev->set_mac_address(dev, sa);
2298 if (!err)
2299 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2300 return err;
2301 }
2302
2303 /*
2304 * Perform the SIOCxIFxxx calls.
2305 */
2306 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2307 {
2308 int err;
2309 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2310
2311 if (!dev)
2312 return -ENODEV;
2313
2314 switch (cmd) {
2315 case SIOCGIFFLAGS: /* Get interface flags */
2316 ifr->ifr_flags = dev_get_flags(dev);
2317 return 0;
2318
2319 case SIOCSIFFLAGS: /* Set interface flags */
2320 return dev_change_flags(dev, ifr->ifr_flags);
2321
2322 case SIOCGIFMETRIC: /* Get the metric on the interface
2323 (currently unused) */
2324 ifr->ifr_metric = 0;
2325 return 0;
2326
2327 case SIOCSIFMETRIC: /* Set the metric on the interface
2328 (currently unused) */
2329 return -EOPNOTSUPP;
2330
2331 case SIOCGIFMTU: /* Get the MTU of a device */
2332 ifr->ifr_mtu = dev->mtu;
2333 return 0;
2334
2335 case SIOCSIFMTU: /* Set the MTU of a device */
2336 return dev_set_mtu(dev, ifr->ifr_mtu);
2337
2338 case SIOCGIFHWADDR:
2339 if (!dev->addr_len)
2340 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2341 else
2342 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2343 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2344 ifr->ifr_hwaddr.sa_family = dev->type;
2345 return 0;
2346
2347 case SIOCSIFHWADDR:
2348 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2349
2350 case SIOCSIFHWBROADCAST:
2351 if (ifr->ifr_hwaddr.sa_family != dev->type)
2352 return -EINVAL;
2353 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2354 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2355 notifier_call_chain(&netdev_chain,
2356 NETDEV_CHANGEADDR, dev);
2357 return 0;
2358
2359 case SIOCGIFMAP:
2360 ifr->ifr_map.mem_start = dev->mem_start;
2361 ifr->ifr_map.mem_end = dev->mem_end;
2362 ifr->ifr_map.base_addr = dev->base_addr;
2363 ifr->ifr_map.irq = dev->irq;
2364 ifr->ifr_map.dma = dev->dma;
2365 ifr->ifr_map.port = dev->if_port;
2366 return 0;
2367
2368 case SIOCSIFMAP:
2369 if (dev->set_config) {
2370 if (!netif_device_present(dev))
2371 return -ENODEV;
2372 return dev->set_config(dev, &ifr->ifr_map);
2373 }
2374 return -EOPNOTSUPP;
2375
2376 case SIOCADDMULTI:
2377 if (!dev->set_multicast_list ||
2378 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2379 return -EINVAL;
2380 if (!netif_device_present(dev))
2381 return -ENODEV;
2382 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2383 dev->addr_len, 1);
2384
2385 case SIOCDELMULTI:
2386 if (!dev->set_multicast_list ||
2387 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2388 return -EINVAL;
2389 if (!netif_device_present(dev))
2390 return -ENODEV;
2391 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2392 dev->addr_len, 1);
2393
2394 case SIOCGIFINDEX:
2395 ifr->ifr_ifindex = dev->ifindex;
2396 return 0;
2397
2398 case SIOCGIFTXQLEN:
2399 ifr->ifr_qlen = dev->tx_queue_len;
2400 return 0;
2401
2402 case SIOCSIFTXQLEN:
2403 if (ifr->ifr_qlen < 0)
2404 return -EINVAL;
2405 dev->tx_queue_len = ifr->ifr_qlen;
2406 return 0;
2407
2408 case SIOCSIFNAME:
2409 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2410 return dev_change_name(dev, ifr->ifr_newname);
2411
2412 /*
2413 * Unknown or private ioctl
2414 */
2415
2416 default:
2417 if ((cmd >= SIOCDEVPRIVATE &&
2418 cmd <= SIOCDEVPRIVATE + 15) ||
2419 cmd == SIOCBONDENSLAVE ||
2420 cmd == SIOCBONDRELEASE ||
2421 cmd == SIOCBONDSETHWADDR ||
2422 cmd == SIOCBONDSLAVEINFOQUERY ||
2423 cmd == SIOCBONDINFOQUERY ||
2424 cmd == SIOCBONDCHANGEACTIVE ||
2425 cmd == SIOCGMIIPHY ||
2426 cmd == SIOCGMIIREG ||
2427 cmd == SIOCSMIIREG ||
2428 cmd == SIOCBRADDIF ||
2429 cmd == SIOCBRDELIF ||
2430 cmd == SIOCWANDEV) {
2431 err = -EOPNOTSUPP;
2432 if (dev->do_ioctl) {
2433 if (netif_device_present(dev))
2434 err = dev->do_ioctl(dev, ifr,
2435 cmd);
2436 else
2437 err = -ENODEV;
2438 }
2439 } else
2440 err = -EINVAL;
2441
2442 }
2443 return err;
2444 }
2445
2446 /*
2447 * This function handles all "interface"-type I/O control requests. The actual
2448 * 'doing' part of this is dev_ifsioc above.
2449 */
2450
2451 /**
2452 * dev_ioctl - network device ioctl
2453 * @cmd: command to issue
2454 * @arg: pointer to a struct ifreq in user space
2455 *
2456 * Issue ioctl functions to devices. This is normally called by the
2457 * user space syscall interfaces but can sometimes be useful for
2458 * other purposes. The return value is the return from the syscall if
2459 * positive or a negative errno code on error.
2460 */
2461
2462 int dev_ioctl(unsigned int cmd, void __user *arg)
2463 {
2464 struct ifreq ifr;
2465 int ret;
2466 char *colon;
2467
2468 /* One special case: SIOCGIFCONF takes ifconf argument
2469 and requires shared lock, because it sleeps writing
2470 to user space.
2471 */
2472
2473 if (cmd == SIOCGIFCONF) {
2474 rtnl_lock();
2475 ret = dev_ifconf((char __user *) arg);
2476 rtnl_unlock();
2477 return ret;
2478 }
2479 if (cmd == SIOCGIFNAME)
2480 return dev_ifname((struct ifreq __user *)arg);
2481
2482 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2483 return -EFAULT;
2484
2485 ifr.ifr_name[IFNAMSIZ-1] = 0;
2486
2487 colon = strchr(ifr.ifr_name, ':');
2488 if (colon)
2489 *colon = 0;
2490
2491 /*
2492 * See which interface the caller is talking about.
2493 */
2494
2495 switch (cmd) {
2496 /*
2497 * These ioctl calls:
2498 * - can be done by all.
2499 * - atomic and do not require locking.
2500 * - return a value
2501 */
2502 case SIOCGIFFLAGS:
2503 case SIOCGIFMETRIC:
2504 case SIOCGIFMTU:
2505 case SIOCGIFHWADDR:
2506 case SIOCGIFSLAVE:
2507 case SIOCGIFMAP:
2508 case SIOCGIFINDEX:
2509 case SIOCGIFTXQLEN:
2510 dev_load(ifr.ifr_name);
2511 read_lock(&dev_base_lock);
2512 ret = dev_ifsioc(&ifr, cmd);
2513 read_unlock(&dev_base_lock);
2514 if (!ret) {
2515 if (colon)
2516 *colon = ':';
2517 if (copy_to_user(arg, &ifr,
2518 sizeof(struct ifreq)))
2519 ret = -EFAULT;
2520 }
2521 return ret;
2522
2523 case SIOCETHTOOL:
2524 dev_load(ifr.ifr_name);
2525 rtnl_lock();
2526 ret = dev_ethtool(&ifr);
2527 rtnl_unlock();
2528 if (!ret) {
2529 if (colon)
2530 *colon = ':';
2531 if (copy_to_user(arg, &ifr,
2532 sizeof(struct ifreq)))
2533 ret = -EFAULT;
2534 }
2535 return ret;
2536
2537 /*
2538 * These ioctl calls:
2539 * - require superuser power.
2540 * - require strict serialization.
2541 * - return a value
2542 */
2543 case SIOCGMIIPHY:
2544 case SIOCGMIIREG:
2545 case SIOCSIFNAME:
2546 if (!capable(CAP_NET_ADMIN))
2547 return -EPERM;
2548 dev_load(ifr.ifr_name);
2549 rtnl_lock();
2550 ret = dev_ifsioc(&ifr, cmd);
2551 rtnl_unlock();
2552 if (!ret) {
2553 if (colon)
2554 *colon = ':';
2555 if (copy_to_user(arg, &ifr,
2556 sizeof(struct ifreq)))
2557 ret = -EFAULT;
2558 }
2559 return ret;
2560
2561 /*
2562 * These ioctl calls:
2563 * - require superuser power.
2564 * - require strict serialization.
2565 * - do not return a value
2566 */
2567 case SIOCSIFFLAGS:
2568 case SIOCSIFMETRIC:
2569 case SIOCSIFMTU:
2570 case SIOCSIFMAP:
2571 case SIOCSIFHWADDR:
2572 case SIOCSIFSLAVE:
2573 case SIOCADDMULTI:
2574 case SIOCDELMULTI:
2575 case SIOCSIFHWBROADCAST:
2576 case SIOCSIFTXQLEN:
2577 case SIOCSMIIREG:
2578 case SIOCBONDENSLAVE:
2579 case SIOCBONDRELEASE:
2580 case SIOCBONDSETHWADDR:
2581 case SIOCBONDCHANGEACTIVE:
2582 case SIOCBRADDIF:
2583 case SIOCBRDELIF:
2584 if (!capable(CAP_NET_ADMIN))
2585 return -EPERM;
2586 /* fall through */
2587 case SIOCBONDSLAVEINFOQUERY:
2588 case SIOCBONDINFOQUERY:
2589 dev_load(ifr.ifr_name);
2590 rtnl_lock();
2591 ret = dev_ifsioc(&ifr, cmd);
2592 rtnl_unlock();
2593 return ret;
2594
2595 case SIOCGIFMEM:
2596 /* Get the per device memory space. We can add this but
2597 * currently do not support it */
2598 case SIOCSIFMEM:
2599 /* Set the per device memory buffer space.
2600 * Not applicable in our case */
2601 case SIOCSIFLINK:
2602 return -EINVAL;
2603
2604 /*
2605 * Unknown or private ioctl.
2606 */
2607 default:
2608 if (cmd == SIOCWANDEV ||
2609 (cmd >= SIOCDEVPRIVATE &&
2610 cmd <= SIOCDEVPRIVATE + 15)) {
2611 dev_load(ifr.ifr_name);
2612 rtnl_lock();
2613 ret = dev_ifsioc(&ifr, cmd);
2614 rtnl_unlock();
2615 if (!ret && copy_to_user(arg, &ifr,
2616 sizeof(struct ifreq)))
2617 ret = -EFAULT;
2618 return ret;
2619 }
2620 #ifdef CONFIG_WIRELESS_EXT
2621 /* Take care of Wireless Extensions */
2622 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2623 /* If command is `set a parameter', or
2624 * `get the encoding parameters', check if
2625 * the user has the right to do it */
2626 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2627 if (!capable(CAP_NET_ADMIN))
2628 return -EPERM;
2629 }
2630 dev_load(ifr.ifr_name);
2631 rtnl_lock();
2632 /* Follow me in net/core/wireless.c */
2633 ret = wireless_process_ioctl(&ifr, cmd);
2634 rtnl_unlock();
2635 if (IW_IS_GET(cmd) &&
2636 copy_to_user(arg, &ifr,
2637 sizeof(struct ifreq)))
2638 ret = -EFAULT;
2639 return ret;
2640 }
2641 #endif /* CONFIG_WIRELESS_EXT */
2642 return -EINVAL;
2643 }
2644 }
2645
2646
2647 /**
2648 * dev_new_index - allocate an ifindex
2649 *
2650 * Returns a suitable unique value for a new device interface
2651 * number. The caller must hold the rtnl semaphore or the
2652 * dev_base_lock to be sure it remains unique.
2653 */
2654 static int dev_new_index(void)
2655 {
2656 static int ifindex;
2657 for (;;) {
2658 if (++ifindex <= 0)
2659 ifindex = 1;
2660 if (!__dev_get_by_index(ifindex))
2661 return ifindex;
2662 }
2663 }
2664
2665 static int dev_boot_phase = 1;
2666
2667 /* Delayed registration/unregisteration */
2668 static DEFINE_SPINLOCK(net_todo_list_lock);
2669 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2670
2671 static inline void net_set_todo(struct net_device *dev)
2672 {
2673 spin_lock(&net_todo_list_lock);
2674 list_add_tail(&dev->todo_list, &net_todo_list);
2675 spin_unlock(&net_todo_list_lock);
2676 }
2677
2678 /**
2679 * register_netdevice - register a network device
2680 * @dev: device to register
2681 *
2682 * Take a completed network device structure and add it to the kernel
2683 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2684 * chain. 0 is returned on success. A negative errno code is returned
2685 * on a failure to set up the device, or if the name is a duplicate.
2686 *
2687 * Callers must hold the rtnl semaphore. You may want
2688 * register_netdev() instead of this.
2689 *
2690 * BUGS:
2691 * The locking appears insufficient to guarantee two parallel registers
2692 * will not get the same name.
2693 */
2694
2695 int register_netdevice(struct net_device *dev)
2696 {
2697 struct hlist_head *head;
2698 struct hlist_node *p;
2699 int ret;
2700
2701 BUG_ON(dev_boot_phase);
2702 ASSERT_RTNL();
2703
2704 /* When net_device's are persistent, this will be fatal. */
2705 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2706
2707 spin_lock_init(&dev->queue_lock);
2708 spin_lock_init(&dev->xmit_lock);
2709 dev->xmit_lock_owner = -1;
2710 #ifdef CONFIG_NET_CLS_ACT
2711 spin_lock_init(&dev->ingress_lock);
2712 #endif
2713
2714 ret = alloc_divert_blk(dev);
2715 if (ret)
2716 goto out;
2717
2718 dev->iflink = -1;
2719
2720 /* Init, if this function is available */
2721 if (dev->init) {
2722 ret = dev->init(dev);
2723 if (ret) {
2724 if (ret > 0)
2725 ret = -EIO;
2726 goto out_err;
2727 }
2728 }
2729
2730 if (!dev_valid_name(dev->name)) {
2731 ret = -EINVAL;
2732 goto out_err;
2733 }
2734
2735 dev->ifindex = dev_new_index();
2736 if (dev->iflink == -1)
2737 dev->iflink = dev->ifindex;
2738
2739 /* Check for existence of name */
2740 head = dev_name_hash(dev->name);
2741 hlist_for_each(p, head) {
2742 struct net_device *d
2743 = hlist_entry(p, struct net_device, name_hlist);
2744 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2745 ret = -EEXIST;
2746 goto out_err;
2747 }
2748 }
2749
2750 /* Fix illegal SG+CSUM combinations. */
2751 if ((dev->features & NETIF_F_SG) &&
2752 !(dev->features & (NETIF_F_IP_CSUM |
2753 NETIF_F_NO_CSUM |
2754 NETIF_F_HW_CSUM))) {
2755 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2756 dev->name);
2757 dev->features &= ~NETIF_F_SG;
2758 }
2759
2760 /* TSO requires that SG is present as well. */
2761 if ((dev->features & NETIF_F_TSO) &&
2762 !(dev->features & NETIF_F_SG)) {
2763 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2764 dev->name);
2765 dev->features &= ~NETIF_F_TSO;
2766 }
2767 if (dev->features & NETIF_F_UFO) {
2768 if (!(dev->features & NETIF_F_HW_CSUM)) {
2769 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2770 "NETIF_F_HW_CSUM feature.\n",
2771 dev->name);
2772 dev->features &= ~NETIF_F_UFO;
2773 }
2774 if (!(dev->features & NETIF_F_SG)) {
2775 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2776 "NETIF_F_SG feature.\n",
2777 dev->name);
2778 dev->features &= ~NETIF_F_UFO;
2779 }
2780 }
2781
2782 /*
2783 * nil rebuild_header routine,
2784 * that should be never called and used as just bug trap.
2785 */
2786
2787 if (!dev->rebuild_header)
2788 dev->rebuild_header = default_rebuild_header;
2789
2790 /*
2791 * Default initial state at registry is that the
2792 * device is present.
2793 */
2794
2795 set_bit(__LINK_STATE_PRESENT, &dev->state);
2796
2797 dev->next = NULL;
2798 dev_init_scheduler(dev);
2799 write_lock_bh(&dev_base_lock);
2800 *dev_tail = dev;
2801 dev_tail = &dev->next;
2802 hlist_add_head(&dev->name_hlist, head);
2803 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2804 dev_hold(dev);
2805 dev->reg_state = NETREG_REGISTERING;
2806 write_unlock_bh(&dev_base_lock);
2807
2808 /* Notify protocols, that a new device appeared. */
2809 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2810
2811 /* Finish registration after unlock */
2812 net_set_todo(dev);
2813 ret = 0;
2814
2815 out:
2816 return ret;
2817 out_err:
2818 free_divert_blk(dev);
2819 goto out;
2820 }
2821
2822 /**
2823 * register_netdev - register a network device
2824 * @dev: device to register
2825 *
2826 * Take a completed network device structure and add it to the kernel
2827 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2828 * chain. 0 is returned on success. A negative errno code is returned
2829 * on a failure to set up the device, or if the name is a duplicate.
2830 *
2831 * This is a wrapper around register_netdev that takes the rtnl semaphore
2832 * and expands the device name if you passed a format string to
2833 * alloc_netdev.
2834 */
2835 int register_netdev(struct net_device *dev)
2836 {
2837 int err;
2838
2839 rtnl_lock();
2840
2841 /*
2842 * If the name is a format string the caller wants us to do a
2843 * name allocation.
2844 */
2845 if (strchr(dev->name, '%')) {
2846 err = dev_alloc_name(dev, dev->name);
2847 if (err < 0)
2848 goto out;
2849 }
2850
2851 /*
2852 * Back compatibility hook. Kill this one in 2.5
2853 */
2854 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2855 err = dev_alloc_name(dev, "eth%d");
2856 if (err < 0)
2857 goto out;
2858 }
2859
2860 err = register_netdevice(dev);
2861 out:
2862 rtnl_unlock();
2863 return err;
2864 }
2865 EXPORT_SYMBOL(register_netdev);
2866
2867 /*
2868 * netdev_wait_allrefs - wait until all references are gone.
2869 *
2870 * This is called when unregistering network devices.
2871 *
2872 * Any protocol or device that holds a reference should register
2873 * for netdevice notification, and cleanup and put back the
2874 * reference if they receive an UNREGISTER event.
2875 * We can get stuck here if buggy protocols don't correctly
2876 * call dev_put.
2877 */
2878 static void netdev_wait_allrefs(struct net_device *dev)
2879 {
2880 unsigned long rebroadcast_time, warning_time;
2881
2882 rebroadcast_time = warning_time = jiffies;
2883 while (atomic_read(&dev->refcnt) != 0) {
2884 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2885 rtnl_lock();
2886
2887 /* Rebroadcast unregister notification */
2888 notifier_call_chain(&netdev_chain,
2889 NETDEV_UNREGISTER, dev);
2890
2891 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2892 &dev->state)) {
2893 /* We must not have linkwatch events
2894 * pending on unregister. If this
2895 * happens, we simply run the queue
2896 * unscheduled, resulting in a noop
2897 * for this device.
2898 */
2899 linkwatch_run_queue();
2900 }
2901
2902 __rtnl_unlock();
2903
2904 rebroadcast_time = jiffies;
2905 }
2906
2907 msleep(250);
2908
2909 if (time_after(jiffies, warning_time + 10 * HZ)) {
2910 printk(KERN_EMERG "unregister_netdevice: "
2911 "waiting for %s to become free. Usage "
2912 "count = %d\n",
2913 dev->name, atomic_read(&dev->refcnt));
2914 warning_time = jiffies;
2915 }
2916 }
2917 }
2918
2919 /* The sequence is:
2920 *
2921 * rtnl_lock();
2922 * ...
2923 * register_netdevice(x1);
2924 * register_netdevice(x2);
2925 * ...
2926 * unregister_netdevice(y1);
2927 * unregister_netdevice(y2);
2928 * ...
2929 * rtnl_unlock();
2930 * free_netdev(y1);
2931 * free_netdev(y2);
2932 *
2933 * We are invoked by rtnl_unlock() after it drops the semaphore.
2934 * This allows us to deal with problems:
2935 * 1) We can create/delete sysfs objects which invoke hotplug
2936 * without deadlocking with linkwatch via keventd.
2937 * 2) Since we run with the RTNL semaphore not held, we can sleep
2938 * safely in order to wait for the netdev refcnt to drop to zero.
2939 */
2940 static DEFINE_MUTEX(net_todo_run_mutex);
2941 void netdev_run_todo(void)
2942 {
2943 struct list_head list = LIST_HEAD_INIT(list);
2944 int err;
2945
2946
2947 /* Need to guard against multiple cpu's getting out of order. */
2948 mutex_lock(&net_todo_run_mutex);
2949
2950 /* Not safe to do outside the semaphore. We must not return
2951 * until all unregister events invoked by the local processor
2952 * have been completed (either by this todo run, or one on
2953 * another cpu).
2954 */
2955 if (list_empty(&net_todo_list))
2956 goto out;
2957
2958 /* Snapshot list, allow later requests */
2959 spin_lock(&net_todo_list_lock);
2960 list_splice_init(&net_todo_list, &list);
2961 spin_unlock(&net_todo_list_lock);
2962
2963 while (!list_empty(&list)) {
2964 struct net_device *dev
2965 = list_entry(list.next, struct net_device, todo_list);
2966 list_del(&dev->todo_list);
2967
2968 switch(dev->reg_state) {
2969 case NETREG_REGISTERING:
2970 err = netdev_register_sysfs(dev);
2971 if (err)
2972 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2973 dev->name, err);
2974 dev->reg_state = NETREG_REGISTERED;
2975 break;
2976
2977 case NETREG_UNREGISTERING:
2978 netdev_unregister_sysfs(dev);
2979 dev->reg_state = NETREG_UNREGISTERED;
2980
2981 netdev_wait_allrefs(dev);
2982
2983 /* paranoia */
2984 BUG_ON(atomic_read(&dev->refcnt));
2985 BUG_TRAP(!dev->ip_ptr);
2986 BUG_TRAP(!dev->ip6_ptr);
2987 BUG_TRAP(!dev->dn_ptr);
2988
2989
2990 /* It must be the very last action,
2991 * after this 'dev' may point to freed up memory.
2992 */
2993 if (dev->destructor)
2994 dev->destructor(dev);
2995 break;
2996
2997 default:
2998 printk(KERN_ERR "network todo '%s' but state %d\n",
2999 dev->name, dev->reg_state);
3000 break;
3001 }
3002 }
3003
3004 out:
3005 mutex_unlock(&net_todo_run_mutex);
3006 }
3007
3008 /**
3009 * alloc_netdev - allocate network device
3010 * @sizeof_priv: size of private data to allocate space for
3011 * @name: device name format string
3012 * @setup: callback to initialize device
3013 *
3014 * Allocates a struct net_device with private data area for driver use
3015 * and performs basic initialization.
3016 */
3017 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3018 void (*setup)(struct net_device *))
3019 {
3020 void *p;
3021 struct net_device *dev;
3022 int alloc_size;
3023
3024 /* ensure 32-byte alignment of both the device and private area */
3025 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3026 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3027
3028 p = kmalloc(alloc_size, GFP_KERNEL);
3029 if (!p) {
3030 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3031 return NULL;
3032 }
3033 memset(p, 0, alloc_size);
3034
3035 dev = (struct net_device *)
3036 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3037 dev->padded = (char *)dev - (char *)p;
3038
3039 if (sizeof_priv)
3040 dev->priv = netdev_priv(dev);
3041
3042 setup(dev);
3043 strcpy(dev->name, name);
3044 return dev;
3045 }
3046 EXPORT_SYMBOL(alloc_netdev);
3047
3048 /**
3049 * free_netdev - free network device
3050 * @dev: device
3051 *
3052 * This function does the last stage of destroying an allocated device
3053 * interface. The reference to the device object is released.
3054 * If this is the last reference then it will be freed.
3055 */
3056 void free_netdev(struct net_device *dev)
3057 {
3058 #ifdef CONFIG_SYSFS
3059 /* Compatiablity with error handling in drivers */
3060 if (dev->reg_state == NETREG_UNINITIALIZED) {
3061 kfree((char *)dev - dev->padded);
3062 return;
3063 }
3064
3065 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3066 dev->reg_state = NETREG_RELEASED;
3067
3068 /* will free via class release */
3069 class_device_put(&dev->class_dev);
3070 #else
3071 kfree((char *)dev - dev->padded);
3072 #endif
3073 }
3074
3075 /* Synchronize with packet receive processing. */
3076 void synchronize_net(void)
3077 {
3078 might_sleep();
3079 synchronize_rcu();
3080 }
3081
3082 /**
3083 * unregister_netdevice - remove device from the kernel
3084 * @dev: device
3085 *
3086 * This function shuts down a device interface and removes it
3087 * from the kernel tables. On success 0 is returned, on a failure
3088 * a negative errno code is returned.
3089 *
3090 * Callers must hold the rtnl semaphore. You may want
3091 * unregister_netdev() instead of this.
3092 */
3093
3094 int unregister_netdevice(struct net_device *dev)
3095 {
3096 struct net_device *d, **dp;
3097
3098 BUG_ON(dev_boot_phase);
3099 ASSERT_RTNL();
3100
3101 /* Some devices call without registering for initialization unwind. */
3102 if (dev->reg_state == NETREG_UNINITIALIZED) {
3103 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3104 "was registered\n", dev->name, dev);
3105 return -ENODEV;
3106 }
3107
3108 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3109
3110 /* If device is running, close it first. */
3111 if (dev->flags & IFF_UP)
3112 dev_close(dev);
3113
3114 /* And unlink it from device chain. */
3115 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3116 if (d == dev) {
3117 write_lock_bh(&dev_base_lock);
3118 hlist_del(&dev->name_hlist);
3119 hlist_del(&dev->index_hlist);
3120 if (dev_tail == &dev->next)
3121 dev_tail = dp;
3122 *dp = d->next;
3123 write_unlock_bh(&dev_base_lock);
3124 break;
3125 }
3126 }
3127 if (!d) {
3128 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3129 dev->name);
3130 return -ENODEV;
3131 }
3132
3133 dev->reg_state = NETREG_UNREGISTERING;
3134
3135 synchronize_net();
3136
3137 /* Shutdown queueing discipline. */
3138 dev_shutdown(dev);
3139
3140
3141 /* Notify protocols, that we are about to destroy
3142 this device. They should clean all the things.
3143 */
3144 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3145
3146 /*
3147 * Flush the multicast chain
3148 */
3149 dev_mc_discard(dev);
3150
3151 if (dev->uninit)
3152 dev->uninit(dev);
3153
3154 /* Notifier chain MUST detach us from master device. */
3155 BUG_TRAP(!dev->master);
3156
3157 free_divert_blk(dev);
3158
3159 /* Finish processing unregister after unlock */
3160 net_set_todo(dev);
3161
3162 synchronize_net();
3163
3164 dev_put(dev);
3165 return 0;
3166 }
3167
3168 /**
3169 * unregister_netdev - remove device from the kernel
3170 * @dev: device
3171 *
3172 * This function shuts down a device interface and removes it
3173 * from the kernel tables. On success 0 is returned, on a failure
3174 * a negative errno code is returned.
3175 *
3176 * This is just a wrapper for unregister_netdevice that takes
3177 * the rtnl semaphore. In general you want to use this and not
3178 * unregister_netdevice.
3179 */
3180 void unregister_netdev(struct net_device *dev)
3181 {
3182 rtnl_lock();
3183 unregister_netdevice(dev);
3184 rtnl_unlock();
3185 }
3186
3187 EXPORT_SYMBOL(unregister_netdev);
3188
3189 #ifdef CONFIG_HOTPLUG_CPU
3190 static int dev_cpu_callback(struct notifier_block *nfb,
3191 unsigned long action,
3192 void *ocpu)
3193 {
3194 struct sk_buff **list_skb;
3195 struct net_device **list_net;
3196 struct sk_buff *skb;
3197 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3198 struct softnet_data *sd, *oldsd;
3199
3200 if (action != CPU_DEAD)
3201 return NOTIFY_OK;
3202
3203 local_irq_disable();
3204 cpu = smp_processor_id();
3205 sd = &per_cpu(softnet_data, cpu);
3206 oldsd = &per_cpu(softnet_data, oldcpu);
3207
3208 /* Find end of our completion_queue. */
3209 list_skb = &sd->completion_queue;
3210 while (*list_skb)
3211 list_skb = &(*list_skb)->next;
3212 /* Append completion queue from offline CPU. */
3213 *list_skb = oldsd->completion_queue;
3214 oldsd->completion_queue = NULL;
3215
3216 /* Find end of our output_queue. */
3217 list_net = &sd->output_queue;
3218 while (*list_net)
3219 list_net = &(*list_net)->next_sched;
3220 /* Append output queue from offline CPU. */
3221 *list_net = oldsd->output_queue;
3222 oldsd->output_queue = NULL;
3223
3224 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3225 local_irq_enable();
3226
3227 /* Process offline CPU's input_pkt_queue */
3228 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3229 netif_rx(skb);
3230
3231 return NOTIFY_OK;
3232 }
3233 #endif /* CONFIG_HOTPLUG_CPU */
3234
3235
3236 /*
3237 * Initialize the DEV module. At boot time this walks the device list and
3238 * unhooks any devices that fail to initialise (normally hardware not
3239 * present) and leaves us with a valid list of present and active devices.
3240 *
3241 */
3242
3243 /*
3244 * This is called single threaded during boot, so no need
3245 * to take the rtnl semaphore.
3246 */
3247 static int __init net_dev_init(void)
3248 {
3249 int i, rc = -ENOMEM;
3250
3251 BUG_ON(!dev_boot_phase);
3252
3253 net_random_init();
3254
3255 if (dev_proc_init())
3256 goto out;
3257
3258 if (netdev_sysfs_init())
3259 goto out;
3260
3261 INIT_LIST_HEAD(&ptype_all);
3262 for (i = 0; i < 16; i++)
3263 INIT_LIST_HEAD(&ptype_base[i]);
3264
3265 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3266 INIT_HLIST_HEAD(&dev_name_head[i]);
3267
3268 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3269 INIT_HLIST_HEAD(&dev_index_head[i]);
3270
3271 /*
3272 * Initialise the packet receive queues.
3273 */
3274
3275 for_each_cpu(i) {
3276 struct softnet_data *queue;
3277
3278 queue = &per_cpu(softnet_data, i);
3279 skb_queue_head_init(&queue->input_pkt_queue);
3280 queue->completion_queue = NULL;
3281 INIT_LIST_HEAD(&queue->poll_list);
3282 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3283 queue->backlog_dev.weight = weight_p;
3284 queue->backlog_dev.poll = process_backlog;
3285 atomic_set(&queue->backlog_dev.refcnt, 1);
3286 }
3287
3288 dev_boot_phase = 0;
3289
3290 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3291 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3292
3293 hotcpu_notifier(dev_cpu_callback, 0);
3294 dst_init();
3295 dev_mcast_init();
3296 rc = 0;
3297 out:
3298 return rc;
3299 }
3300
3301 subsys_initcall(net_dev_init);
3302
3303 EXPORT_SYMBOL(__dev_get_by_index);
3304 EXPORT_SYMBOL(__dev_get_by_name);
3305 EXPORT_SYMBOL(__dev_remove_pack);
3306 EXPORT_SYMBOL(__skb_linearize);
3307 EXPORT_SYMBOL(dev_valid_name);
3308 EXPORT_SYMBOL(dev_add_pack);
3309 EXPORT_SYMBOL(dev_alloc_name);
3310 EXPORT_SYMBOL(dev_close);
3311 EXPORT_SYMBOL(dev_get_by_flags);
3312 EXPORT_SYMBOL(dev_get_by_index);
3313 EXPORT_SYMBOL(dev_get_by_name);
3314 EXPORT_SYMBOL(dev_open);
3315 EXPORT_SYMBOL(dev_queue_xmit);
3316 EXPORT_SYMBOL(dev_remove_pack);
3317 EXPORT_SYMBOL(dev_set_allmulti);
3318 EXPORT_SYMBOL(dev_set_promiscuity);
3319 EXPORT_SYMBOL(dev_change_flags);
3320 EXPORT_SYMBOL(dev_set_mtu);
3321 EXPORT_SYMBOL(dev_set_mac_address);
3322 EXPORT_SYMBOL(free_netdev);
3323 EXPORT_SYMBOL(netdev_boot_setup_check);
3324 EXPORT_SYMBOL(netdev_set_master);
3325 EXPORT_SYMBOL(netdev_state_change);
3326 EXPORT_SYMBOL(netif_receive_skb);
3327 EXPORT_SYMBOL(netif_rx);
3328 EXPORT_SYMBOL(register_gifconf);
3329 EXPORT_SYMBOL(register_netdevice);
3330 EXPORT_SYMBOL(register_netdevice_notifier);
3331 EXPORT_SYMBOL(skb_checksum_help);
3332 EXPORT_SYMBOL(synchronize_net);
3333 EXPORT_SYMBOL(unregister_netdevice);
3334 EXPORT_SYMBOL(unregister_netdevice_notifier);
3335 EXPORT_SYMBOL(net_enable_timestamp);
3336 EXPORT_SYMBOL(net_disable_timestamp);
3337 EXPORT_SYMBOL(dev_get_flags);
3338
3339 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3340 EXPORT_SYMBOL(br_handle_frame_hook);
3341 EXPORT_SYMBOL(br_fdb_get_hook);
3342 EXPORT_SYMBOL(br_fdb_put_hook);
3343 #endif
3344
3345 #ifdef CONFIG_KMOD
3346 EXPORT_SYMBOL(dev_load);
3347 #endif
3348
3349 EXPORT_PER_CPU_SYMBOL(softnet_data);
This page took 0.146394 seconds and 5 git commands to generate.