]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/core/dev.c
[WIRELESS]: Fix config dependencies.
[net-next-2.6.git] / net / core / dev.c
CommitLineData
1da177e4
LT
1/*
2 * NET3 Protocol independent device support routines.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Derived from the non IP parts of dev.c 1.0.19
02c30a84 10 * Authors: Ross Biro
1da177e4
LT
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 *
14 * Additional Authors:
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
21 *
22 * Changes:
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
34 * drivers
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
44 * call a packet.
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
58 * 1 device.
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
66 * the backlog queue.
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
73 */
74
75#include <asm/uaccess.h>
76#include <asm/system.h>
77#include <linux/bitops.h>
4fc268d2 78#include <linux/capability.h>
1da177e4
LT
79#include <linux/config.h>
80#include <linux/cpu.h>
81#include <linux/types.h>
82#include <linux/kernel.h>
83#include <linux/sched.h>
4a3e2f71 84#include <linux/mutex.h>
1da177e4
LT
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>
d86b5e0e 114#include <linux/wireless.h>
1da177e4 115#include <net/iw_handler.h>
1da177e4
LT
116#include <asm/current.h>
117
1da177e4
LT
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
146static DEFINE_SPINLOCK(ptype_lock);
147static struct list_head ptype_base[16]; /* 16 way hashed list */
148static struct list_head ptype_all; /* Taps */
149
1da177e4
LT
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 */
169struct net_device *dev_base;
170static struct net_device **dev_tail = &dev_base;
171DEFINE_RWLOCK(dev_base_lock);
172
173EXPORT_SYMBOL(dev_base);
174EXPORT_SYMBOL(dev_base_lock);
175
176#define NETDEV_HASHBITS 8
177static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
178static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
179
180static 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
186static 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
195static 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 */
31aa02c5 201DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
1da177e4
LT
202
203#ifdef CONFIG_SYSFS
204extern int netdev_sysfs_init(void);
205extern int netdev_register_sysfs(struct net_device *);
206extern 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
224int 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
255void 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
1da177e4
LT
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 */
283void __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);
304out:
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 */
319void 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 */
333static 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 */
344static 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 */
371int 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 */
400unsigned 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 */
424int __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
467struct 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
491struct 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
514struct 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
538struct 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
564struct 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
cf309e3f
JF
577EXPORT_SYMBOL(dev_getbyhwaddr);
578
1da177e4
LT
579struct 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
594EXPORT_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
607struct 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 */
c2373ee9 629int dev_valid_name(const char *name)
1da177e4
LT
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
649int 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 */
711int 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
d8a33ac4
SH
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 */
750void netdev_features_change(struct net_device *dev)
751{
752 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
753}
754EXPORT_SYMBOL(netdev_features_change);
755
1da177e4
LT
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 */
764void 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
781void 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
793static 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 */
814int 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 */
878int 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. */
6192b54b 902 msleep(1);
1da177e4
LT
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
949int 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
978int unregister_netdevice_notifier(struct notifier_block *nb)
979{
980 return notifier_chain_unregister(&netdev_chain, nb);
981}
982
983/**
984 * call_netdevice_notifiers - call all network notifier blocks
985 * @val: value passed unmodified to notifier function
986 * @v: pointer passed unmodified to notifier function
987 *
988 * Call all network notifier blocks. Parameters and return value
989 * are as for notifier_call_chain().
990 */
991
992int call_netdevice_notifiers(unsigned long val, void *v)
993{
994 return notifier_call_chain(&netdev_chain, val, v);
995}
996
997/* When > 0 there are consumers of rx skb time stamps */
998static atomic_t netstamp_needed = ATOMIC_INIT(0);
999
1000void net_enable_timestamp(void)
1001{
1002 atomic_inc(&netstamp_needed);
1003}
1004
1005void net_disable_timestamp(void)
1006{
1007 atomic_dec(&netstamp_needed);
1008}
1009
a61bbcf2
PM
1010void __net_timestamp(struct sk_buff *skb)
1011{
1012 struct timeval tv;
1013
1014 do_gettimeofday(&tv);
1015 skb_set_timestamp(skb, &tv);
1016}
1017EXPORT_SYMBOL(__net_timestamp);
1018
1019static inline void net_timestamp(struct sk_buff *skb)
1da177e4
LT
1020{
1021 if (atomic_read(&netstamp_needed))
a61bbcf2 1022 __net_timestamp(skb);
1da177e4 1023 else {
a61bbcf2
PM
1024 skb->tstamp.off_sec = 0;
1025 skb->tstamp.off_usec = 0;
1da177e4
LT
1026 }
1027}
1028
1029/*
1030 * Support routine. Sends outgoing frames to any network
1031 * taps currently in use.
1032 */
1033
1034void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1035{
1036 struct packet_type *ptype;
a61bbcf2
PM
1037
1038 net_timestamp(skb);
1da177e4
LT
1039
1040 rcu_read_lock();
1041 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1042 /* Never send packets back to the socket
1043 * they originated from - MvS (miquels@drinkel.ow.org)
1044 */
1045 if ((ptype->dev == dev || !ptype->dev) &&
1046 (ptype->af_packet_priv == NULL ||
1047 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1048 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1049 if (!skb2)
1050 break;
1051
1052 /* skb->nh should be correctly
1053 set by sender, so that the second statement is
1054 just protection against buggy protocols.
1055 */
1056 skb2->mac.raw = skb2->data;
1057
1058 if (skb2->nh.raw < skb2->data ||
1059 skb2->nh.raw > skb2->tail) {
1060 if (net_ratelimit())
1061 printk(KERN_CRIT "protocol %04x is "
1062 "buggy, dev %s\n",
1063 skb2->protocol, dev->name);
1064 skb2->nh.raw = skb2->data;
1065 }
1066
1067 skb2->h.raw = skb2->nh.raw;
1068 skb2->pkt_type = PACKET_OUTGOING;
f2ccd8fa 1069 ptype->func(skb2, skb->dev, ptype, skb->dev);
1da177e4
LT
1070 }
1071 }
1072 rcu_read_unlock();
1073}
1074
1075/*
1076 * Invalidate hardware checksum when packet is to be mangled, and
1077 * complete checksum manually on outgoing path.
1078 */
1079int skb_checksum_help(struct sk_buff *skb, int inward)
1080{
1081 unsigned int csum;
1082 int ret = 0, offset = skb->h.raw - skb->data;
1083
1084 if (inward) {
1085 skb->ip_summed = CHECKSUM_NONE;
1086 goto out;
1087 }
1088
1089 if (skb_cloned(skb)) {
1090 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1091 if (ret)
1092 goto out;
1093 }
1094
09a62660 1095 BUG_ON(offset > (int)skb->len);
1da177e4
LT
1096 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1097
1098 offset = skb->tail - skb->h.raw;
09a62660
KK
1099 BUG_ON(offset <= 0);
1100 BUG_ON(skb->csum + 2 > offset);
1da177e4
LT
1101
1102 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1103 skb->ip_summed = CHECKSUM_NONE;
1104out:
1105 return ret;
1106}
1107
fb286bb2
HX
1108/* Take action when hardware reception checksum errors are detected. */
1109#ifdef CONFIG_BUG
1110void netdev_rx_csum_fault(struct net_device *dev)
1111{
1112 if (net_ratelimit()) {
246a4212
SH
1113 printk(KERN_ERR "%s: hw csum failure.\n",
1114 dev ? dev->name : "<unknown>");
fb286bb2
HX
1115 dump_stack();
1116 }
1117}
1118EXPORT_SYMBOL(netdev_rx_csum_fault);
1119#endif
1120
1da177e4
LT
1121#ifdef CONFIG_HIGHMEM
1122/* Actually, we should eliminate this check as soon as we know, that:
1123 * 1. IOMMU is present and allows to map all the memory.
1124 * 2. No high memory really exists on this machine.
1125 */
1126
1127static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1128{
1129 int i;
1130
1131 if (dev->features & NETIF_F_HIGHDMA)
1132 return 0;
1133
1134 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1135 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1136 return 1;
1137
1138 return 0;
1139}
1140#else
1141#define illegal_highdma(dev, skb) (0)
1142#endif
1143
1da177e4 1144/* Keep head the same: replace data */
dd0fc66f 1145int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1da177e4
LT
1146{
1147 unsigned int size;
1148 u8 *data;
1149 long offset;
1150 struct skb_shared_info *ninfo;
1151 int headerlen = skb->data - skb->head;
1152 int expand = (skb->tail + skb->data_len) - skb->end;
1153
1154 if (skb_shared(skb))
1155 BUG();
1156
1157 if (expand <= 0)
1158 expand = 0;
1159
1160 size = skb->end - skb->head + expand;
1161 size = SKB_DATA_ALIGN(size);
1162 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1163 if (!data)
1164 return -ENOMEM;
1165
1166 /* Copy entire thing */
1167 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1168 BUG();
1169
1170 /* Set up shinfo */
1171 ninfo = (struct skb_shared_info*)(data + size);
1172 atomic_set(&ninfo->dataref, 1);
1173 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1174 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1175 ninfo->nr_frags = 0;
1176 ninfo->frag_list = NULL;
1177
1178 /* Offset between the two in bytes */
1179 offset = data - skb->head;
1180
1181 /* Free old data. */
1182 skb_release_data(skb);
1183
1184 skb->head = data;
1185 skb->end = data + size;
1186
1187 /* Set up new pointers */
1188 skb->h.raw += offset;
1189 skb->nh.raw += offset;
1190 skb->mac.raw += offset;
1191 skb->tail += offset;
1192 skb->data += offset;
1193
1194 /* We are no longer a clone, even if we were. */
1195 skb->cloned = 0;
1196
1197 skb->tail += skb->data_len;
1198 skb->data_len = 0;
1199 return 0;
1200}
1201
1202#define HARD_TX_LOCK(dev, cpu) { \
1203 if ((dev->features & NETIF_F_LLTX) == 0) { \
1204 spin_lock(&dev->xmit_lock); \
1205 dev->xmit_lock_owner = cpu; \
1206 } \
1207}
1208
1209#define HARD_TX_UNLOCK(dev) { \
1210 if ((dev->features & NETIF_F_LLTX) == 0) { \
1211 dev->xmit_lock_owner = -1; \
1212 spin_unlock(&dev->xmit_lock); \
1213 } \
1214}
1215
1216/**
1217 * dev_queue_xmit - transmit a buffer
1218 * @skb: buffer to transmit
1219 *
1220 * Queue a buffer for transmission to a network device. The caller must
1221 * have set the device and priority and built the buffer before calling
1222 * this function. The function can be called from an interrupt.
1223 *
1224 * A negative errno code is returned on a failure. A success does not
1225 * guarantee the frame will be transmitted as it may be dropped due
1226 * to congestion or traffic shaping.
af191367
BG
1227 *
1228 * -----------------------------------------------------------------------------------
1229 * I notice this method can also return errors from the queue disciplines,
1230 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1231 * be positive.
1232 *
1233 * Regardless of the return value, the skb is consumed, so it is currently
1234 * difficult to retry a send to this method. (You can bump the ref count
1235 * before sending to hold a reference for retry if you are careful.)
1236 *
1237 * When calling this method, interrupts MUST be enabled. This is because
1238 * the BH enable code must have IRQs enabled so that it will not deadlock.
1239 * --BLG
1da177e4
LT
1240 */
1241
1242int dev_queue_xmit(struct sk_buff *skb)
1243{
1244 struct net_device *dev = skb->dev;
1245 struct Qdisc *q;
1246 int rc = -ENOMEM;
1247
1248 if (skb_shinfo(skb)->frag_list &&
1249 !(dev->features & NETIF_F_FRAGLIST) &&
1250 __skb_linearize(skb, GFP_ATOMIC))
1251 goto out_kfree_skb;
1252
1253 /* Fragmented skb is linearized if device does not support SG,
1254 * or if at least one of fragments is in highmem and device
1255 * does not support DMA from it.
1256 */
1257 if (skb_shinfo(skb)->nr_frags &&
1258 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1259 __skb_linearize(skb, GFP_ATOMIC))
1260 goto out_kfree_skb;
1261
1262 /* If packet is not checksummed and device does not support
1263 * checksumming for this protocol, complete checksumming here.
1264 */
1265 if (skb->ip_summed == CHECKSUM_HW &&
1266 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1267 (!(dev->features & NETIF_F_IP_CSUM) ||
1268 skb->protocol != htons(ETH_P_IP))))
1269 if (skb_checksum_help(skb, 0))
1270 goto out_kfree_skb;
1271
2d7ceece
ED
1272 spin_lock_prefetch(&dev->queue_lock);
1273
1da177e4
LT
1274 /* Disable soft irqs for various locks below. Also
1275 * stops preemption for RCU.
1276 */
1277 local_bh_disable();
1278
1279 /* Updates of qdisc are serialized by queue_lock.
1280 * The struct Qdisc which is pointed to by qdisc is now a
1281 * rcu structure - it may be accessed without acquiring
1282 * a lock (but the structure may be stale.) The freeing of the
1283 * qdisc will be deferred until it's known that there are no
1284 * more references to it.
1285 *
1286 * If the qdisc has an enqueue function, we still need to
1287 * hold the queue_lock before calling it, since queue_lock
1288 * also serializes access to the device queue.
1289 */
1290
1291 q = rcu_dereference(dev->qdisc);
1292#ifdef CONFIG_NET_CLS_ACT
1293 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1294#endif
1295 if (q->enqueue) {
1296 /* Grab device queue */
1297 spin_lock(&dev->queue_lock);
1298
1299 rc = q->enqueue(skb, q);
1300
1301 qdisc_run(dev);
1302
1303 spin_unlock(&dev->queue_lock);
1304 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1305 goto out;
1306 }
1307
1308 /* The device has no queue. Common case for software devices:
1309 loopback, all the sorts of tunnels...
1310
1311 Really, it is unlikely that xmit_lock protection is necessary here.
1312 (f.e. loopback and IP tunnels are clean ignoring statistics
1313 counters.)
1314 However, it is possible, that they rely on protection
1315 made by us here.
1316
1317 Check this and shot the lock. It is not prone from deadlocks.
1318 Either shot noqueue qdisc, it is even simpler 8)
1319 */
1320 if (dev->flags & IFF_UP) {
1321 int cpu = smp_processor_id(); /* ok because BHs are off */
1322
1323 if (dev->xmit_lock_owner != cpu) {
1324
1325 HARD_TX_LOCK(dev, cpu);
1326
1327 if (!netif_queue_stopped(dev)) {
1328 if (netdev_nit)
1329 dev_queue_xmit_nit(skb, dev);
1330
1331 rc = 0;
1332 if (!dev->hard_start_xmit(skb, dev)) {
1333 HARD_TX_UNLOCK(dev);
1334 goto out;
1335 }
1336 }
1337 HARD_TX_UNLOCK(dev);
1338 if (net_ratelimit())
1339 printk(KERN_CRIT "Virtual device %s asks to "
1340 "queue packet!\n", dev->name);
1341 } else {
1342 /* Recursion is detected! It is possible,
1343 * unfortunately */
1344 if (net_ratelimit())
1345 printk(KERN_CRIT "Dead loop on virtual device "
1346 "%s, fix it urgently!\n", dev->name);
1347 }
1348 }
1349
1350 rc = -ENETDOWN;
1351 local_bh_enable();
1352
1353out_kfree_skb:
1354 kfree_skb(skb);
1355 return rc;
1356out:
1357 local_bh_enable();
1358 return rc;
1359}
1360
1361
1362/*=======================================================================
1363 Receiver routines
1364 =======================================================================*/
1365
51b0bded
SH
1366int netdev_max_backlog = 1000;
1367int netdev_budget = 300;
1da177e4 1368int weight_p = 64; /* old backlog weight */
1da177e4
LT
1369
1370DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1371
1372
1da177e4
LT
1373/**
1374 * netif_rx - post buffer to the network code
1375 * @skb: buffer to post
1376 *
1377 * This function receives a packet from a device driver and queues it for
1378 * the upper (protocol) levels to process. It always succeeds. The buffer
1379 * may be dropped during processing for congestion control or by the
1380 * protocol layers.
1381 *
1382 * return values:
1383 * NET_RX_SUCCESS (no congestion)
1384 * NET_RX_CN_LOW (low congestion)
1385 * NET_RX_CN_MOD (moderate congestion)
1386 * NET_RX_CN_HIGH (high congestion)
1387 * NET_RX_DROP (packet was dropped)
1388 *
1389 */
1390
1391int netif_rx(struct sk_buff *skb)
1392{
1da177e4
LT
1393 struct softnet_data *queue;
1394 unsigned long flags;
1395
1396 /* if netpoll wants it, pretend we never saw it */
1397 if (netpoll_rx(skb))
1398 return NET_RX_DROP;
1399
a61bbcf2
PM
1400 if (!skb->tstamp.off_sec)
1401 net_timestamp(skb);
1da177e4
LT
1402
1403 /*
1404 * The code is rearranged so that the path is the most
1405 * short when CPU is congested, but is still operating.
1406 */
1407 local_irq_save(flags);
1da177e4
LT
1408 queue = &__get_cpu_var(softnet_data);
1409
1410 __get_cpu_var(netdev_rx_stat).total++;
1411 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1412 if (queue->input_pkt_queue.qlen) {
1da177e4
LT
1413enqueue:
1414 dev_hold(skb->dev);
1415 __skb_queue_tail(&queue->input_pkt_queue, skb);
1da177e4 1416 local_irq_restore(flags);
34008d8c 1417 return NET_RX_SUCCESS;
1da177e4
LT
1418 }
1419
1da177e4
LT
1420 netif_rx_schedule(&queue->backlog_dev);
1421 goto enqueue;
1422 }
1423
1da177e4
LT
1424 __get_cpu_var(netdev_rx_stat).dropped++;
1425 local_irq_restore(flags);
1426
1427 kfree_skb(skb);
1428 return NET_RX_DROP;
1429}
1430
1431int netif_rx_ni(struct sk_buff *skb)
1432{
1433 int err;
1434
1435 preempt_disable();
1436 err = netif_rx(skb);
1437 if (local_softirq_pending())
1438 do_softirq();
1439 preempt_enable();
1440
1441 return err;
1442}
1443
1444EXPORT_SYMBOL(netif_rx_ni);
1445
f2ccd8fa 1446static inline struct net_device *skb_bond(struct sk_buff *skb)
1da177e4
LT
1447{
1448 struct net_device *dev = skb->dev;
1449
8f903c70
JV
1450 if (dev->master) {
1451 /*
1452 * On bonding slaves other than the currently active
1453 * slave, suppress duplicates except for 802.3ad
1454 * ETH_P_SLOW and alb non-mcast/bcast.
1455 */
1456 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1457 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1458 if (skb->pkt_type != PACKET_BROADCAST &&
1459 skb->pkt_type != PACKET_MULTICAST)
1460 goto keep;
1461 }
1462
1463 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1464 skb->protocol == __constant_htons(ETH_P_SLOW))
1465 goto keep;
1466
1467 kfree_skb(skb);
1468 return NULL;
1469 }
1470keep:
1da177e4 1471 skb->dev = dev->master;
8f903c70 1472 }
f2ccd8fa
DM
1473
1474 return dev;
1da177e4
LT
1475}
1476
1477static void net_tx_action(struct softirq_action *h)
1478{
1479 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1480
1481 if (sd->completion_queue) {
1482 struct sk_buff *clist;
1483
1484 local_irq_disable();
1485 clist = sd->completion_queue;
1486 sd->completion_queue = NULL;
1487 local_irq_enable();
1488
1489 while (clist) {
1490 struct sk_buff *skb = clist;
1491 clist = clist->next;
1492
1493 BUG_TRAP(!atomic_read(&skb->users));
1494 __kfree_skb(skb);
1495 }
1496 }
1497
1498 if (sd->output_queue) {
1499 struct net_device *head;
1500
1501 local_irq_disable();
1502 head = sd->output_queue;
1503 sd->output_queue = NULL;
1504 local_irq_enable();
1505
1506 while (head) {
1507 struct net_device *dev = head;
1508 head = head->next_sched;
1509
1510 smp_mb__before_clear_bit();
1511 clear_bit(__LINK_STATE_SCHED, &dev->state);
1512
1513 if (spin_trylock(&dev->queue_lock)) {
1514 qdisc_run(dev);
1515 spin_unlock(&dev->queue_lock);
1516 } else {
1517 netif_schedule(dev);
1518 }
1519 }
1520 }
1521}
1522
1523static __inline__ int deliver_skb(struct sk_buff *skb,
f2ccd8fa
DM
1524 struct packet_type *pt_prev,
1525 struct net_device *orig_dev)
1da177e4
LT
1526{
1527 atomic_inc(&skb->users);
f2ccd8fa 1528 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT
1529}
1530
1531#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1532int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1533struct net_bridge;
1534struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1535 unsigned char *addr);
1536void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1537
1538static __inline__ int handle_bridge(struct sk_buff **pskb,
f2ccd8fa
DM
1539 struct packet_type **pt_prev, int *ret,
1540 struct net_device *orig_dev)
1da177e4
LT
1541{
1542 struct net_bridge_port *port;
1543
1544 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1545 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1546 return 0;
1547
1548 if (*pt_prev) {
f2ccd8fa 1549 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1da177e4
LT
1550 *pt_prev = NULL;
1551 }
1552
1553 return br_handle_frame_hook(port, pskb);
1554}
1555#else
f2ccd8fa 1556#define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1da177e4
LT
1557#endif
1558
1559#ifdef CONFIG_NET_CLS_ACT
1560/* TODO: Maybe we should just force sch_ingress to be compiled in
1561 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1562 * a compare and 2 stores extra right now if we dont have it on
1563 * but have CONFIG_NET_CLS_ACT
1564 * NOTE: This doesnt stop any functionality; if you dont have
1565 * the ingress scheduler, you just cant add policies on ingress.
1566 *
1567 */
1568static int ing_filter(struct sk_buff *skb)
1569{
1570 struct Qdisc *q;
1571 struct net_device *dev = skb->dev;
1572 int result = TC_ACT_OK;
1573
1574 if (dev->qdisc_ingress) {
1575 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1576 if (MAX_RED_LOOP < ttl++) {
1577 printk("Redir loop detected Dropping packet (%s->%s)\n",
86e65da9 1578 skb->input_dev->name, skb->dev->name);
1da177e4
LT
1579 return TC_ACT_SHOT;
1580 }
1581
1582 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1583
1584 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
86e65da9 1585
1da177e4
LT
1586 spin_lock(&dev->ingress_lock);
1587 if ((q = dev->qdisc_ingress) != NULL)
1588 result = q->enqueue(skb, q);
1589 spin_unlock(&dev->ingress_lock);
1590
1591 }
1592
1593 return result;
1594}
1595#endif
1596
1597int netif_receive_skb(struct sk_buff *skb)
1598{
1599 struct packet_type *ptype, *pt_prev;
f2ccd8fa 1600 struct net_device *orig_dev;
1da177e4
LT
1601 int ret = NET_RX_DROP;
1602 unsigned short type;
1603
1604 /* if we've gotten here through NAPI, check netpoll */
1605 if (skb->dev->poll && netpoll_rx(skb))
1606 return NET_RX_DROP;
1607
a61bbcf2
PM
1608 if (!skb->tstamp.off_sec)
1609 net_timestamp(skb);
1da177e4 1610
86e65da9
DM
1611 if (!skb->input_dev)
1612 skb->input_dev = skb->dev;
1613
f2ccd8fa 1614 orig_dev = skb_bond(skb);
1da177e4 1615
8f903c70
JV
1616 if (!orig_dev)
1617 return NET_RX_DROP;
1618
1da177e4
LT
1619 __get_cpu_var(netdev_rx_stat).total++;
1620
1621 skb->h.raw = skb->nh.raw = skb->data;
1622 skb->mac_len = skb->nh.raw - skb->mac.raw;
1623
1624 pt_prev = NULL;
1625
1626 rcu_read_lock();
1627
1628#ifdef CONFIG_NET_CLS_ACT
1629 if (skb->tc_verd & TC_NCLS) {
1630 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1631 goto ncls;
1632 }
1633#endif
1634
1635 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1636 if (!ptype->dev || ptype->dev == skb->dev) {
1637 if (pt_prev)
f2ccd8fa 1638 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
1639 pt_prev = ptype;
1640 }
1641 }
1642
1643#ifdef CONFIG_NET_CLS_ACT
1644 if (pt_prev) {
f2ccd8fa 1645 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
1646 pt_prev = NULL; /* noone else should process this after*/
1647 } else {
1648 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1649 }
1650
1651 ret = ing_filter(skb);
1652
1653 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1654 kfree_skb(skb);
1655 goto out;
1656 }
1657
1658 skb->tc_verd = 0;
1659ncls:
1660#endif
1661
1662 handle_diverter(skb);
1663
f2ccd8fa 1664 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1da177e4
LT
1665 goto out;
1666
1667 type = skb->protocol;
1668 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1669 if (ptype->type == type &&
1670 (!ptype->dev || ptype->dev == skb->dev)) {
1671 if (pt_prev)
f2ccd8fa 1672 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
1673 pt_prev = ptype;
1674 }
1675 }
1676
1677 if (pt_prev) {
f2ccd8fa 1678 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT
1679 } else {
1680 kfree_skb(skb);
1681 /* Jamal, now you will not able to escape explaining
1682 * me how you were going to use this. :-)
1683 */
1684 ret = NET_RX_DROP;
1685 }
1686
1687out:
1688 rcu_read_unlock();
1689 return ret;
1690}
1691
1692static int process_backlog(struct net_device *backlog_dev, int *budget)
1693{
1694 int work = 0;
1695 int quota = min(backlog_dev->quota, *budget);
1696 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1697 unsigned long start_time = jiffies;
1698
e3876605 1699 backlog_dev->weight = weight_p;
1da177e4
LT
1700 for (;;) {
1701 struct sk_buff *skb;
1702 struct net_device *dev;
1703
1704 local_irq_disable();
1705 skb = __skb_dequeue(&queue->input_pkt_queue);
1706 if (!skb)
1707 goto job_done;
1708 local_irq_enable();
1709
1710 dev = skb->dev;
1711
1712 netif_receive_skb(skb);
1713
1714 dev_put(dev);
1715
1716 work++;
1717
1718 if (work >= quota || jiffies - start_time > 1)
1719 break;
1720
1721 }
1722
1723 backlog_dev->quota -= work;
1724 *budget -= work;
1725 return -1;
1726
1727job_done:
1728 backlog_dev->quota -= work;
1729 *budget -= work;
1730
1731 list_del(&backlog_dev->poll_list);
1732 smp_mb__before_clear_bit();
1733 netif_poll_enable(backlog_dev);
1734
1da177e4
LT
1735 local_irq_enable();
1736 return 0;
1737}
1738
1739static void net_rx_action(struct softirq_action *h)
1740{
1741 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1742 unsigned long start_time = jiffies;
51b0bded 1743 int budget = netdev_budget;
53fb95d3
MM
1744 void *have;
1745
1da177e4
LT
1746 local_irq_disable();
1747
1748 while (!list_empty(&queue->poll_list)) {
1749 struct net_device *dev;
1750
1751 if (budget <= 0 || jiffies - start_time > 1)
1752 goto softnet_break;
1753
1754 local_irq_enable();
1755
1756 dev = list_entry(queue->poll_list.next,
1757 struct net_device, poll_list);
53fb95d3 1758 have = netpoll_poll_lock(dev);
1da177e4
LT
1759
1760 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
53fb95d3 1761 netpoll_poll_unlock(have);
1da177e4 1762 local_irq_disable();
8aca8a27 1763 list_move_tail(&dev->poll_list, &queue->poll_list);
1da177e4
LT
1764 if (dev->quota < 0)
1765 dev->quota += dev->weight;
1766 else
1767 dev->quota = dev->weight;
1768 } else {
53fb95d3 1769 netpoll_poll_unlock(have);
1da177e4
LT
1770 dev_put(dev);
1771 local_irq_disable();
1772 }
1773 }
1774out:
1775 local_irq_enable();
1776 return;
1777
1778softnet_break:
1779 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1780 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1781 goto out;
1782}
1783
1784static gifconf_func_t * gifconf_list [NPROTO];
1785
1786/**
1787 * register_gifconf - register a SIOCGIF handler
1788 * @family: Address family
1789 * @gifconf: Function handler
1790 *
1791 * Register protocol dependent address dumping routines. The handler
1792 * that is passed must not be freed or reused until it has been replaced
1793 * by another handler.
1794 */
1795int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1796{
1797 if (family >= NPROTO)
1798 return -EINVAL;
1799 gifconf_list[family] = gifconf;
1800 return 0;
1801}
1802
1803
1804/*
1805 * Map an interface index to its name (SIOCGIFNAME)
1806 */
1807
1808/*
1809 * We need this ioctl for efficient implementation of the
1810 * if_indextoname() function required by the IPv6 API. Without
1811 * it, we would have to search all the interfaces to find a
1812 * match. --pb
1813 */
1814
1815static int dev_ifname(struct ifreq __user *arg)
1816{
1817 struct net_device *dev;
1818 struct ifreq ifr;
1819
1820 /*
1821 * Fetch the caller's info block.
1822 */
1823
1824 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1825 return -EFAULT;
1826
1827 read_lock(&dev_base_lock);
1828 dev = __dev_get_by_index(ifr.ifr_ifindex);
1829 if (!dev) {
1830 read_unlock(&dev_base_lock);
1831 return -ENODEV;
1832 }
1833
1834 strcpy(ifr.ifr_name, dev->name);
1835 read_unlock(&dev_base_lock);
1836
1837 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1838 return -EFAULT;
1839 return 0;
1840}
1841
1842/*
1843 * Perform a SIOCGIFCONF call. This structure will change
1844 * size eventually, and there is nothing I can do about it.
1845 * Thus we will need a 'compatibility mode'.
1846 */
1847
1848static int dev_ifconf(char __user *arg)
1849{
1850 struct ifconf ifc;
1851 struct net_device *dev;
1852 char __user *pos;
1853 int len;
1854 int total;
1855 int i;
1856
1857 /*
1858 * Fetch the caller's info block.
1859 */
1860
1861 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1862 return -EFAULT;
1863
1864 pos = ifc.ifc_buf;
1865 len = ifc.ifc_len;
1866
1867 /*
1868 * Loop over the interfaces, and write an info block for each.
1869 */
1870
1871 total = 0;
1872 for (dev = dev_base; dev; dev = dev->next) {
1873 for (i = 0; i < NPROTO; i++) {
1874 if (gifconf_list[i]) {
1875 int done;
1876 if (!pos)
1877 done = gifconf_list[i](dev, NULL, 0);
1878 else
1879 done = gifconf_list[i](dev, pos + total,
1880 len - total);
1881 if (done < 0)
1882 return -EFAULT;
1883 total += done;
1884 }
1885 }
1886 }
1887
1888 /*
1889 * All done. Write the updated control block back to the caller.
1890 */
1891 ifc.ifc_len = total;
1892
1893 /*
1894 * Both BSD and Solaris return 0 here, so we do too.
1895 */
1896 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1897}
1898
1899#ifdef CONFIG_PROC_FS
1900/*
1901 * This is invoked by the /proc filesystem handler to display a device
1902 * in detail.
1903 */
1904static __inline__ struct net_device *dev_get_idx(loff_t pos)
1905{
1906 struct net_device *dev;
1907 loff_t i;
1908
1909 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1910
1911 return i == pos ? dev : NULL;
1912}
1913
1914void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1915{
1916 read_lock(&dev_base_lock);
1917 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1918}
1919
1920void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1921{
1922 ++*pos;
1923 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1924}
1925
1926void dev_seq_stop(struct seq_file *seq, void *v)
1927{
1928 read_unlock(&dev_base_lock);
1929}
1930
1931static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1932{
1933 if (dev->get_stats) {
1934 struct net_device_stats *stats = dev->get_stats(dev);
1935
1936 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1937 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1938 dev->name, stats->rx_bytes, stats->rx_packets,
1939 stats->rx_errors,
1940 stats->rx_dropped + stats->rx_missed_errors,
1941 stats->rx_fifo_errors,
1942 stats->rx_length_errors + stats->rx_over_errors +
1943 stats->rx_crc_errors + stats->rx_frame_errors,
1944 stats->rx_compressed, stats->multicast,
1945 stats->tx_bytes, stats->tx_packets,
1946 stats->tx_errors, stats->tx_dropped,
1947 stats->tx_fifo_errors, stats->collisions,
1948 stats->tx_carrier_errors +
1949 stats->tx_aborted_errors +
1950 stats->tx_window_errors +
1951 stats->tx_heartbeat_errors,
1952 stats->tx_compressed);
1953 } else
1954 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1955}
1956
1957/*
1958 * Called from the PROCfs module. This now uses the new arbitrary sized
1959 * /proc/net interface to create /proc/net/dev
1960 */
1961static int dev_seq_show(struct seq_file *seq, void *v)
1962{
1963 if (v == SEQ_START_TOKEN)
1964 seq_puts(seq, "Inter-| Receive "
1965 " | Transmit\n"
1966 " face |bytes packets errs drop fifo frame "
1967 "compressed multicast|bytes packets errs "
1968 "drop fifo colls carrier compressed\n");
1969 else
1970 dev_seq_printf_stats(seq, v);
1971 return 0;
1972}
1973
1974static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1975{
1976 struct netif_rx_stats *rc = NULL;
1977
1978 while (*pos < NR_CPUS)
1979 if (cpu_online(*pos)) {
1980 rc = &per_cpu(netdev_rx_stat, *pos);
1981 break;
1982 } else
1983 ++*pos;
1984 return rc;
1985}
1986
1987static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1988{
1989 return softnet_get_online(pos);
1990}
1991
1992static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1993{
1994 ++*pos;
1995 return softnet_get_online(pos);
1996}
1997
1998static void softnet_seq_stop(struct seq_file *seq, void *v)
1999{
2000}
2001
2002static int softnet_seq_show(struct seq_file *seq, void *v)
2003{
2004 struct netif_rx_stats *s = v;
2005
2006 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
31aa02c5 2007 s->total, s->dropped, s->time_squeeze, 0,
c1ebcdb8
SH
2008 0, 0, 0, 0, /* was fastroute */
2009 s->cpu_collision );
1da177e4
LT
2010 return 0;
2011}
2012
2013static struct seq_operations dev_seq_ops = {
2014 .start = dev_seq_start,
2015 .next = dev_seq_next,
2016 .stop = dev_seq_stop,
2017 .show = dev_seq_show,
2018};
2019
2020static int dev_seq_open(struct inode *inode, struct file *file)
2021{
2022 return seq_open(file, &dev_seq_ops);
2023}
2024
2025static struct file_operations dev_seq_fops = {
2026 .owner = THIS_MODULE,
2027 .open = dev_seq_open,
2028 .read = seq_read,
2029 .llseek = seq_lseek,
2030 .release = seq_release,
2031};
2032
2033static struct seq_operations softnet_seq_ops = {
2034 .start = softnet_seq_start,
2035 .next = softnet_seq_next,
2036 .stop = softnet_seq_stop,
2037 .show = softnet_seq_show,
2038};
2039
2040static int softnet_seq_open(struct inode *inode, struct file *file)
2041{
2042 return seq_open(file, &softnet_seq_ops);
2043}
2044
2045static struct file_operations softnet_seq_fops = {
2046 .owner = THIS_MODULE,
2047 .open = softnet_seq_open,
2048 .read = seq_read,
2049 .llseek = seq_lseek,
2050 .release = seq_release,
2051};
2052
d86b5e0e 2053#ifdef CONFIG_WIRELESS_EXT
1da177e4
LT
2054extern int wireless_proc_init(void);
2055#else
2056#define wireless_proc_init() 0
2057#endif
2058
2059static int __init dev_proc_init(void)
2060{
2061 int rc = -ENOMEM;
2062
2063 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2064 goto out;
2065 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2066 goto out_dev;
2067 if (wireless_proc_init())
2068 goto out_softnet;
2069 rc = 0;
2070out:
2071 return rc;
2072out_softnet:
2073 proc_net_remove("softnet_stat");
2074out_dev:
2075 proc_net_remove("dev");
2076 goto out;
2077}
2078#else
2079#define dev_proc_init() 0
2080#endif /* CONFIG_PROC_FS */
2081
2082
2083/**
2084 * netdev_set_master - set up master/slave pair
2085 * @slave: slave device
2086 * @master: new master device
2087 *
2088 * Changes the master device of the slave. Pass %NULL to break the
2089 * bonding. The caller must hold the RTNL semaphore. On a failure
2090 * a negative errno code is returned. On success the reference counts
2091 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2092 * function returns zero.
2093 */
2094int netdev_set_master(struct net_device *slave, struct net_device *master)
2095{
2096 struct net_device *old = slave->master;
2097
2098 ASSERT_RTNL();
2099
2100 if (master) {
2101 if (old)
2102 return -EBUSY;
2103 dev_hold(master);
2104 }
2105
2106 slave->master = master;
2107
2108 synchronize_net();
2109
2110 if (old)
2111 dev_put(old);
2112
2113 if (master)
2114 slave->flags |= IFF_SLAVE;
2115 else
2116 slave->flags &= ~IFF_SLAVE;
2117
2118 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2119 return 0;
2120}
2121
2122/**
2123 * dev_set_promiscuity - update promiscuity count on a device
2124 * @dev: device
2125 * @inc: modifier
2126 *
2127 * Add or remove promsicuity from a device. While the count in the device
2128 * remains above zero the interface remains promiscuous. Once it hits zero
2129 * the device reverts back to normal filtering operation. A negative inc
2130 * value is used to drop promiscuity on the device.
2131 */
2132void dev_set_promiscuity(struct net_device *dev, int inc)
2133{
2134 unsigned short old_flags = dev->flags;
2135
1da177e4
LT
2136 if ((dev->promiscuity += inc) == 0)
2137 dev->flags &= ~IFF_PROMISC;
52609c0b
DC
2138 else
2139 dev->flags |= IFF_PROMISC;
2140 if (dev->flags != old_flags) {
1da177e4
LT
2141 dev_mc_upload(dev);
2142 printk(KERN_INFO "device %s %s promiscuous mode\n",
2143 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2144 "left");
2145 }
2146}
2147
2148/**
2149 * dev_set_allmulti - update allmulti count on a device
2150 * @dev: device
2151 * @inc: modifier
2152 *
2153 * Add or remove reception of all multicast frames to a device. While the
2154 * count in the device remains above zero the interface remains listening
2155 * to all interfaces. Once it hits zero the device reverts back to normal
2156 * filtering operation. A negative @inc value is used to drop the counter
2157 * when releasing a resource needing all multicasts.
2158 */
2159
2160void dev_set_allmulti(struct net_device *dev, int inc)
2161{
2162 unsigned short old_flags = dev->flags;
2163
2164 dev->flags |= IFF_ALLMULTI;
2165 if ((dev->allmulti += inc) == 0)
2166 dev->flags &= ~IFF_ALLMULTI;
2167 if (dev->flags ^ old_flags)
2168 dev_mc_upload(dev);
2169}
2170
2171unsigned dev_get_flags(const struct net_device *dev)
2172{
2173 unsigned flags;
2174
2175 flags = (dev->flags & ~(IFF_PROMISC |
2176 IFF_ALLMULTI |
b00055aa
SR
2177 IFF_RUNNING |
2178 IFF_LOWER_UP |
2179 IFF_DORMANT)) |
1da177e4
LT
2180 (dev->gflags & (IFF_PROMISC |
2181 IFF_ALLMULTI));
2182
b00055aa
SR
2183 if (netif_running(dev)) {
2184 if (netif_oper_up(dev))
2185 flags |= IFF_RUNNING;
2186 if (netif_carrier_ok(dev))
2187 flags |= IFF_LOWER_UP;
2188 if (netif_dormant(dev))
2189 flags |= IFF_DORMANT;
2190 }
1da177e4
LT
2191
2192 return flags;
2193}
2194
2195int dev_change_flags(struct net_device *dev, unsigned flags)
2196{
2197 int ret;
2198 int old_flags = dev->flags;
2199
2200 /*
2201 * Set the flags on our device.
2202 */
2203
2204 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2205 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2206 IFF_AUTOMEDIA)) |
2207 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2208 IFF_ALLMULTI));
2209
2210 /*
2211 * Load in the correct multicast list now the flags have changed.
2212 */
2213
2214 dev_mc_upload(dev);
2215
2216 /*
2217 * Have we downed the interface. We handle IFF_UP ourselves
2218 * according to user attempts to set it, rather than blindly
2219 * setting it.
2220 */
2221
2222 ret = 0;
2223 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2224 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2225
2226 if (!ret)
2227 dev_mc_upload(dev);
2228 }
2229
2230 if (dev->flags & IFF_UP &&
2231 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2232 IFF_VOLATILE)))
2233 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2234
2235 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2236 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2237 dev->gflags ^= IFF_PROMISC;
2238 dev_set_promiscuity(dev, inc);
2239 }
2240
2241 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2242 is important. Some (broken) drivers set IFF_PROMISC, when
2243 IFF_ALLMULTI is requested not asking us and not reporting.
2244 */
2245 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2246 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2247 dev->gflags ^= IFF_ALLMULTI;
2248 dev_set_allmulti(dev, inc);
2249 }
2250
2251 if (old_flags ^ dev->flags)
2252 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2253
2254 return ret;
2255}
2256
2257int dev_set_mtu(struct net_device *dev, int new_mtu)
2258{
2259 int err;
2260
2261 if (new_mtu == dev->mtu)
2262 return 0;
2263
2264 /* MTU must be positive. */
2265 if (new_mtu < 0)
2266 return -EINVAL;
2267
2268 if (!netif_device_present(dev))
2269 return -ENODEV;
2270
2271 err = 0;
2272 if (dev->change_mtu)
2273 err = dev->change_mtu(dev, new_mtu);
2274 else
2275 dev->mtu = new_mtu;
2276 if (!err && dev->flags & IFF_UP)
2277 notifier_call_chain(&netdev_chain,
2278 NETDEV_CHANGEMTU, dev);
2279 return err;
2280}
2281
2282int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2283{
2284 int err;
2285
2286 if (!dev->set_mac_address)
2287 return -EOPNOTSUPP;
2288 if (sa->sa_family != dev->type)
2289 return -EINVAL;
2290 if (!netif_device_present(dev))
2291 return -ENODEV;
2292 err = dev->set_mac_address(dev, sa);
2293 if (!err)
2294 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2295 return err;
2296}
2297
2298/*
2299 * Perform the SIOCxIFxxx calls.
2300 */
2301static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2302{
2303 int err;
2304 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2305
2306 if (!dev)
2307 return -ENODEV;
2308
2309 switch (cmd) {
2310 case SIOCGIFFLAGS: /* Get interface flags */
2311 ifr->ifr_flags = dev_get_flags(dev);
2312 return 0;
2313
2314 case SIOCSIFFLAGS: /* Set interface flags */
2315 return dev_change_flags(dev, ifr->ifr_flags);
2316
2317 case SIOCGIFMETRIC: /* Get the metric on the interface
2318 (currently unused) */
2319 ifr->ifr_metric = 0;
2320 return 0;
2321
2322 case SIOCSIFMETRIC: /* Set the metric on the interface
2323 (currently unused) */
2324 return -EOPNOTSUPP;
2325
2326 case SIOCGIFMTU: /* Get the MTU of a device */
2327 ifr->ifr_mtu = dev->mtu;
2328 return 0;
2329
2330 case SIOCSIFMTU: /* Set the MTU of a device */
2331 return dev_set_mtu(dev, ifr->ifr_mtu);
2332
2333 case SIOCGIFHWADDR:
2334 if (!dev->addr_len)
2335 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2336 else
2337 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2338 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2339 ifr->ifr_hwaddr.sa_family = dev->type;
2340 return 0;
2341
2342 case SIOCSIFHWADDR:
2343 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2344
2345 case SIOCSIFHWBROADCAST:
2346 if (ifr->ifr_hwaddr.sa_family != dev->type)
2347 return -EINVAL;
2348 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2349 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2350 notifier_call_chain(&netdev_chain,
2351 NETDEV_CHANGEADDR, dev);
2352 return 0;
2353
2354 case SIOCGIFMAP:
2355 ifr->ifr_map.mem_start = dev->mem_start;
2356 ifr->ifr_map.mem_end = dev->mem_end;
2357 ifr->ifr_map.base_addr = dev->base_addr;
2358 ifr->ifr_map.irq = dev->irq;
2359 ifr->ifr_map.dma = dev->dma;
2360 ifr->ifr_map.port = dev->if_port;
2361 return 0;
2362
2363 case SIOCSIFMAP:
2364 if (dev->set_config) {
2365 if (!netif_device_present(dev))
2366 return -ENODEV;
2367 return dev->set_config(dev, &ifr->ifr_map);
2368 }
2369 return -EOPNOTSUPP;
2370
2371 case SIOCADDMULTI:
2372 if (!dev->set_multicast_list ||
2373 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2374 return -EINVAL;
2375 if (!netif_device_present(dev))
2376 return -ENODEV;
2377 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2378 dev->addr_len, 1);
2379
2380 case SIOCDELMULTI:
2381 if (!dev->set_multicast_list ||
2382 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2383 return -EINVAL;
2384 if (!netif_device_present(dev))
2385 return -ENODEV;
2386 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2387 dev->addr_len, 1);
2388
2389 case SIOCGIFINDEX:
2390 ifr->ifr_ifindex = dev->ifindex;
2391 return 0;
2392
2393 case SIOCGIFTXQLEN:
2394 ifr->ifr_qlen = dev->tx_queue_len;
2395 return 0;
2396
2397 case SIOCSIFTXQLEN:
2398 if (ifr->ifr_qlen < 0)
2399 return -EINVAL;
2400 dev->tx_queue_len = ifr->ifr_qlen;
2401 return 0;
2402
2403 case SIOCSIFNAME:
2404 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2405 return dev_change_name(dev, ifr->ifr_newname);
2406
2407 /*
2408 * Unknown or private ioctl
2409 */
2410
2411 default:
2412 if ((cmd >= SIOCDEVPRIVATE &&
2413 cmd <= SIOCDEVPRIVATE + 15) ||
2414 cmd == SIOCBONDENSLAVE ||
2415 cmd == SIOCBONDRELEASE ||
2416 cmd == SIOCBONDSETHWADDR ||
2417 cmd == SIOCBONDSLAVEINFOQUERY ||
2418 cmd == SIOCBONDINFOQUERY ||
2419 cmd == SIOCBONDCHANGEACTIVE ||
2420 cmd == SIOCGMIIPHY ||
2421 cmd == SIOCGMIIREG ||
2422 cmd == SIOCSMIIREG ||
2423 cmd == SIOCBRADDIF ||
2424 cmd == SIOCBRDELIF ||
2425 cmd == SIOCWANDEV) {
2426 err = -EOPNOTSUPP;
2427 if (dev->do_ioctl) {
2428 if (netif_device_present(dev))
2429 err = dev->do_ioctl(dev, ifr,
2430 cmd);
2431 else
2432 err = -ENODEV;
2433 }
2434 } else
2435 err = -EINVAL;
2436
2437 }
2438 return err;
2439}
2440
2441/*
2442 * This function handles all "interface"-type I/O control requests. The actual
2443 * 'doing' part of this is dev_ifsioc above.
2444 */
2445
2446/**
2447 * dev_ioctl - network device ioctl
2448 * @cmd: command to issue
2449 * @arg: pointer to a struct ifreq in user space
2450 *
2451 * Issue ioctl functions to devices. This is normally called by the
2452 * user space syscall interfaces but can sometimes be useful for
2453 * other purposes. The return value is the return from the syscall if
2454 * positive or a negative errno code on error.
2455 */
2456
2457int dev_ioctl(unsigned int cmd, void __user *arg)
2458{
2459 struct ifreq ifr;
2460 int ret;
2461 char *colon;
2462
2463 /* One special case: SIOCGIFCONF takes ifconf argument
2464 and requires shared lock, because it sleeps writing
2465 to user space.
2466 */
2467
2468 if (cmd == SIOCGIFCONF) {
6756ae4b 2469 rtnl_lock();
1da177e4 2470 ret = dev_ifconf((char __user *) arg);
6756ae4b 2471 rtnl_unlock();
1da177e4
LT
2472 return ret;
2473 }
2474 if (cmd == SIOCGIFNAME)
2475 return dev_ifname((struct ifreq __user *)arg);
2476
2477 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2478 return -EFAULT;
2479
2480 ifr.ifr_name[IFNAMSIZ-1] = 0;
2481
2482 colon = strchr(ifr.ifr_name, ':');
2483 if (colon)
2484 *colon = 0;
2485
2486 /*
2487 * See which interface the caller is talking about.
2488 */
2489
2490 switch (cmd) {
2491 /*
2492 * These ioctl calls:
2493 * - can be done by all.
2494 * - atomic and do not require locking.
2495 * - return a value
2496 */
2497 case SIOCGIFFLAGS:
2498 case SIOCGIFMETRIC:
2499 case SIOCGIFMTU:
2500 case SIOCGIFHWADDR:
2501 case SIOCGIFSLAVE:
2502 case SIOCGIFMAP:
2503 case SIOCGIFINDEX:
2504 case SIOCGIFTXQLEN:
2505 dev_load(ifr.ifr_name);
2506 read_lock(&dev_base_lock);
2507 ret = dev_ifsioc(&ifr, cmd);
2508 read_unlock(&dev_base_lock);
2509 if (!ret) {
2510 if (colon)
2511 *colon = ':';
2512 if (copy_to_user(arg, &ifr,
2513 sizeof(struct ifreq)))
2514 ret = -EFAULT;
2515 }
2516 return ret;
2517
2518 case SIOCETHTOOL:
2519 dev_load(ifr.ifr_name);
2520 rtnl_lock();
2521 ret = dev_ethtool(&ifr);
2522 rtnl_unlock();
2523 if (!ret) {
2524 if (colon)
2525 *colon = ':';
2526 if (copy_to_user(arg, &ifr,
2527 sizeof(struct ifreq)))
2528 ret = -EFAULT;
2529 }
2530 return ret;
2531
2532 /*
2533 * These ioctl calls:
2534 * - require superuser power.
2535 * - require strict serialization.
2536 * - return a value
2537 */
2538 case SIOCGMIIPHY:
2539 case SIOCGMIIREG:
2540 case SIOCSIFNAME:
2541 if (!capable(CAP_NET_ADMIN))
2542 return -EPERM;
2543 dev_load(ifr.ifr_name);
2544 rtnl_lock();
2545 ret = dev_ifsioc(&ifr, cmd);
2546 rtnl_unlock();
2547 if (!ret) {
2548 if (colon)
2549 *colon = ':';
2550 if (copy_to_user(arg, &ifr,
2551 sizeof(struct ifreq)))
2552 ret = -EFAULT;
2553 }
2554 return ret;
2555
2556 /*
2557 * These ioctl calls:
2558 * - require superuser power.
2559 * - require strict serialization.
2560 * - do not return a value
2561 */
2562 case SIOCSIFFLAGS:
2563 case SIOCSIFMETRIC:
2564 case SIOCSIFMTU:
2565 case SIOCSIFMAP:
2566 case SIOCSIFHWADDR:
2567 case SIOCSIFSLAVE:
2568 case SIOCADDMULTI:
2569 case SIOCDELMULTI:
2570 case SIOCSIFHWBROADCAST:
2571 case SIOCSIFTXQLEN:
2572 case SIOCSMIIREG:
2573 case SIOCBONDENSLAVE:
2574 case SIOCBONDRELEASE:
2575 case SIOCBONDSETHWADDR:
1da177e4
LT
2576 case SIOCBONDCHANGEACTIVE:
2577 case SIOCBRADDIF:
2578 case SIOCBRDELIF:
2579 if (!capable(CAP_NET_ADMIN))
2580 return -EPERM;
cabcac0b
TG
2581 /* fall through */
2582 case SIOCBONDSLAVEINFOQUERY:
2583 case SIOCBONDINFOQUERY:
1da177e4
LT
2584 dev_load(ifr.ifr_name);
2585 rtnl_lock();
2586 ret = dev_ifsioc(&ifr, cmd);
2587 rtnl_unlock();
2588 return ret;
2589
2590 case SIOCGIFMEM:
2591 /* Get the per device memory space. We can add this but
2592 * currently do not support it */
2593 case SIOCSIFMEM:
2594 /* Set the per device memory buffer space.
2595 * Not applicable in our case */
2596 case SIOCSIFLINK:
2597 return -EINVAL;
2598
2599 /*
2600 * Unknown or private ioctl.
2601 */
2602 default:
2603 if (cmd == SIOCWANDEV ||
2604 (cmd >= SIOCDEVPRIVATE &&
2605 cmd <= SIOCDEVPRIVATE + 15)) {
2606 dev_load(ifr.ifr_name);
2607 rtnl_lock();
2608 ret = dev_ifsioc(&ifr, cmd);
2609 rtnl_unlock();
2610 if (!ret && copy_to_user(arg, &ifr,
2611 sizeof(struct ifreq)))
2612 ret = -EFAULT;
2613 return ret;
2614 }
d86b5e0e 2615#ifdef CONFIG_WIRELESS_EXT
1da177e4
LT
2616 /* Take care of Wireless Extensions */
2617 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2618 /* If command is `set a parameter', or
2619 * `get the encoding parameters', check if
2620 * the user has the right to do it */
2621 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2622 if (!capable(CAP_NET_ADMIN))
2623 return -EPERM;
2624 }
2625 dev_load(ifr.ifr_name);
2626 rtnl_lock();
2627 /* Follow me in net/core/wireless.c */
2628 ret = wireless_process_ioctl(&ifr, cmd);
2629 rtnl_unlock();
2630 if (IW_IS_GET(cmd) &&
2631 copy_to_user(arg, &ifr,
2632 sizeof(struct ifreq)))
2633 ret = -EFAULT;
2634 return ret;
2635 }
d86b5e0e 2636#endif /* CONFIG_WIRELESS_EXT */
1da177e4
LT
2637 return -EINVAL;
2638 }
2639}
2640
2641
2642/**
2643 * dev_new_index - allocate an ifindex
2644 *
2645 * Returns a suitable unique value for a new device interface
2646 * number. The caller must hold the rtnl semaphore or the
2647 * dev_base_lock to be sure it remains unique.
2648 */
2649static int dev_new_index(void)
2650{
2651 static int ifindex;
2652 for (;;) {
2653 if (++ifindex <= 0)
2654 ifindex = 1;
2655 if (!__dev_get_by_index(ifindex))
2656 return ifindex;
2657 }
2658}
2659
2660static int dev_boot_phase = 1;
2661
2662/* Delayed registration/unregisteration */
2663static DEFINE_SPINLOCK(net_todo_list_lock);
2664static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2665
2666static inline void net_set_todo(struct net_device *dev)
2667{
2668 spin_lock(&net_todo_list_lock);
2669 list_add_tail(&dev->todo_list, &net_todo_list);
2670 spin_unlock(&net_todo_list_lock);
2671}
2672
2673/**
2674 * register_netdevice - register a network device
2675 * @dev: device to register
2676 *
2677 * Take a completed network device structure and add it to the kernel
2678 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2679 * chain. 0 is returned on success. A negative errno code is returned
2680 * on a failure to set up the device, or if the name is a duplicate.
2681 *
2682 * Callers must hold the rtnl semaphore. You may want
2683 * register_netdev() instead of this.
2684 *
2685 * BUGS:
2686 * The locking appears insufficient to guarantee two parallel registers
2687 * will not get the same name.
2688 */
2689
2690int register_netdevice(struct net_device *dev)
2691{
2692 struct hlist_head *head;
2693 struct hlist_node *p;
2694 int ret;
2695
2696 BUG_ON(dev_boot_phase);
2697 ASSERT_RTNL();
2698
2699 /* When net_device's are persistent, this will be fatal. */
2700 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2701
2702 spin_lock_init(&dev->queue_lock);
2703 spin_lock_init(&dev->xmit_lock);
2704 dev->xmit_lock_owner = -1;
2705#ifdef CONFIG_NET_CLS_ACT
2706 spin_lock_init(&dev->ingress_lock);
2707#endif
2708
2709 ret = alloc_divert_blk(dev);
2710 if (ret)
2711 goto out;
2712
2713 dev->iflink = -1;
2714
2715 /* Init, if this function is available */
2716 if (dev->init) {
2717 ret = dev->init(dev);
2718 if (ret) {
2719 if (ret > 0)
2720 ret = -EIO;
2721 goto out_err;
2722 }
2723 }
2724
2725 if (!dev_valid_name(dev->name)) {
2726 ret = -EINVAL;
2727 goto out_err;
2728 }
2729
2730 dev->ifindex = dev_new_index();
2731 if (dev->iflink == -1)
2732 dev->iflink = dev->ifindex;
2733
2734 /* Check for existence of name */
2735 head = dev_name_hash(dev->name);
2736 hlist_for_each(p, head) {
2737 struct net_device *d
2738 = hlist_entry(p, struct net_device, name_hlist);
2739 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2740 ret = -EEXIST;
2741 goto out_err;
2742 }
2743 }
2744
2745 /* Fix illegal SG+CSUM combinations. */
2746 if ((dev->features & NETIF_F_SG) &&
2747 !(dev->features & (NETIF_F_IP_CSUM |
2748 NETIF_F_NO_CSUM |
2749 NETIF_F_HW_CSUM))) {
2750 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2751 dev->name);
2752 dev->features &= ~NETIF_F_SG;
2753 }
2754
2755 /* TSO requires that SG is present as well. */
2756 if ((dev->features & NETIF_F_TSO) &&
2757 !(dev->features & NETIF_F_SG)) {
2758 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2759 dev->name);
2760 dev->features &= ~NETIF_F_TSO;
2761 }
e89e9cf5
AR
2762 if (dev->features & NETIF_F_UFO) {
2763 if (!(dev->features & NETIF_F_HW_CSUM)) {
2764 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2765 "NETIF_F_HW_CSUM feature.\n",
2766 dev->name);
2767 dev->features &= ~NETIF_F_UFO;
2768 }
2769 if (!(dev->features & NETIF_F_SG)) {
2770 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2771 "NETIF_F_SG feature.\n",
2772 dev->name);
2773 dev->features &= ~NETIF_F_UFO;
2774 }
2775 }
1da177e4
LT
2776
2777 /*
2778 * nil rebuild_header routine,
2779 * that should be never called and used as just bug trap.
2780 */
2781
2782 if (!dev->rebuild_header)
2783 dev->rebuild_header = default_rebuild_header;
2784
2785 /*
2786 * Default initial state at registry is that the
2787 * device is present.
2788 */
2789
2790 set_bit(__LINK_STATE_PRESENT, &dev->state);
2791
2792 dev->next = NULL;
2793 dev_init_scheduler(dev);
2794 write_lock_bh(&dev_base_lock);
2795 *dev_tail = dev;
2796 dev_tail = &dev->next;
2797 hlist_add_head(&dev->name_hlist, head);
2798 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2799 dev_hold(dev);
2800 dev->reg_state = NETREG_REGISTERING;
2801 write_unlock_bh(&dev_base_lock);
2802
2803 /* Notify protocols, that a new device appeared. */
2804 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2805
2806 /* Finish registration after unlock */
2807 net_set_todo(dev);
2808 ret = 0;
2809
2810out:
2811 return ret;
2812out_err:
2813 free_divert_blk(dev);
2814 goto out;
2815}
2816
2817/**
2818 * register_netdev - register a network device
2819 * @dev: device to register
2820 *
2821 * Take a completed network device structure and add it to the kernel
2822 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2823 * chain. 0 is returned on success. A negative errno code is returned
2824 * on a failure to set up the device, or if the name is a duplicate.
2825 *
2826 * This is a wrapper around register_netdev that takes the rtnl semaphore
2827 * and expands the device name if you passed a format string to
2828 * alloc_netdev.
2829 */
2830int register_netdev(struct net_device *dev)
2831{
2832 int err;
2833
2834 rtnl_lock();
2835
2836 /*
2837 * If the name is a format string the caller wants us to do a
2838 * name allocation.
2839 */
2840 if (strchr(dev->name, '%')) {
2841 err = dev_alloc_name(dev, dev->name);
2842 if (err < 0)
2843 goto out;
2844 }
2845
2846 /*
2847 * Back compatibility hook. Kill this one in 2.5
2848 */
2849 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2850 err = dev_alloc_name(dev, "eth%d");
2851 if (err < 0)
2852 goto out;
2853 }
2854
2855 err = register_netdevice(dev);
2856out:
2857 rtnl_unlock();
2858 return err;
2859}
2860EXPORT_SYMBOL(register_netdev);
2861
2862/*
2863 * netdev_wait_allrefs - wait until all references are gone.
2864 *
2865 * This is called when unregistering network devices.
2866 *
2867 * Any protocol or device that holds a reference should register
2868 * for netdevice notification, and cleanup and put back the
2869 * reference if they receive an UNREGISTER event.
2870 * We can get stuck here if buggy protocols don't correctly
2871 * call dev_put.
2872 */
2873static void netdev_wait_allrefs(struct net_device *dev)
2874{
2875 unsigned long rebroadcast_time, warning_time;
2876
2877 rebroadcast_time = warning_time = jiffies;
2878 while (atomic_read(&dev->refcnt) != 0) {
2879 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6756ae4b 2880 rtnl_lock();
1da177e4
LT
2881
2882 /* Rebroadcast unregister notification */
2883 notifier_call_chain(&netdev_chain,
2884 NETDEV_UNREGISTER, dev);
2885
2886 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2887 &dev->state)) {
2888 /* We must not have linkwatch events
2889 * pending on unregister. If this
2890 * happens, we simply run the queue
2891 * unscheduled, resulting in a noop
2892 * for this device.
2893 */
2894 linkwatch_run_queue();
2895 }
2896
6756ae4b 2897 __rtnl_unlock();
1da177e4
LT
2898
2899 rebroadcast_time = jiffies;
2900 }
2901
2902 msleep(250);
2903
2904 if (time_after(jiffies, warning_time + 10 * HZ)) {
2905 printk(KERN_EMERG "unregister_netdevice: "
2906 "waiting for %s to become free. Usage "
2907 "count = %d\n",
2908 dev->name, atomic_read(&dev->refcnt));
2909 warning_time = jiffies;
2910 }
2911 }
2912}
2913
2914/* The sequence is:
2915 *
2916 * rtnl_lock();
2917 * ...
2918 * register_netdevice(x1);
2919 * register_netdevice(x2);
2920 * ...
2921 * unregister_netdevice(y1);
2922 * unregister_netdevice(y2);
2923 * ...
2924 * rtnl_unlock();
2925 * free_netdev(y1);
2926 * free_netdev(y2);
2927 *
2928 * We are invoked by rtnl_unlock() after it drops the semaphore.
2929 * This allows us to deal with problems:
2930 * 1) We can create/delete sysfs objects which invoke hotplug
2931 * without deadlocking with linkwatch via keventd.
2932 * 2) Since we run with the RTNL semaphore not held, we can sleep
2933 * safely in order to wait for the netdev refcnt to drop to zero.
2934 */
4a3e2f71 2935static DEFINE_MUTEX(net_todo_run_mutex);
1da177e4
LT
2936void netdev_run_todo(void)
2937{
2938 struct list_head list = LIST_HEAD_INIT(list);
2939 int err;
2940
2941
2942 /* Need to guard against multiple cpu's getting out of order. */
4a3e2f71 2943 mutex_lock(&net_todo_run_mutex);
1da177e4
LT
2944
2945 /* Not safe to do outside the semaphore. We must not return
2946 * until all unregister events invoked by the local processor
2947 * have been completed (either by this todo run, or one on
2948 * another cpu).
2949 */
2950 if (list_empty(&net_todo_list))
2951 goto out;
2952
2953 /* Snapshot list, allow later requests */
2954 spin_lock(&net_todo_list_lock);
2955 list_splice_init(&net_todo_list, &list);
2956 spin_unlock(&net_todo_list_lock);
2957
2958 while (!list_empty(&list)) {
2959 struct net_device *dev
2960 = list_entry(list.next, struct net_device, todo_list);
2961 list_del(&dev->todo_list);
2962
2963 switch(dev->reg_state) {
2964 case NETREG_REGISTERING:
2965 err = netdev_register_sysfs(dev);
2966 if (err)
2967 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2968 dev->name, err);
2969 dev->reg_state = NETREG_REGISTERED;
2970 break;
2971
2972 case NETREG_UNREGISTERING:
2973 netdev_unregister_sysfs(dev);
2974 dev->reg_state = NETREG_UNREGISTERED;
2975
2976 netdev_wait_allrefs(dev);
2977
2978 /* paranoia */
2979 BUG_ON(atomic_read(&dev->refcnt));
2980 BUG_TRAP(!dev->ip_ptr);
2981 BUG_TRAP(!dev->ip6_ptr);
2982 BUG_TRAP(!dev->dn_ptr);
2983
2984
2985 /* It must be the very last action,
2986 * after this 'dev' may point to freed up memory.
2987 */
2988 if (dev->destructor)
2989 dev->destructor(dev);
2990 break;
2991
2992 default:
2993 printk(KERN_ERR "network todo '%s' but state %d\n",
2994 dev->name, dev->reg_state);
2995 break;
2996 }
2997 }
2998
2999out:
4a3e2f71 3000 mutex_unlock(&net_todo_run_mutex);
1da177e4
LT
3001}
3002
3003/**
3004 * alloc_netdev - allocate network device
3005 * @sizeof_priv: size of private data to allocate space for
3006 * @name: device name format string
3007 * @setup: callback to initialize device
3008 *
3009 * Allocates a struct net_device with private data area for driver use
3010 * and performs basic initialization.
3011 */
3012struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3013 void (*setup)(struct net_device *))
3014{
3015 void *p;
3016 struct net_device *dev;
3017 int alloc_size;
3018
3019 /* ensure 32-byte alignment of both the device and private area */
3020 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3021 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3022
3023 p = kmalloc(alloc_size, GFP_KERNEL);
3024 if (!p) {
3025 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3026 return NULL;
3027 }
3028 memset(p, 0, alloc_size);
3029
3030 dev = (struct net_device *)
3031 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3032 dev->padded = (char *)dev - (char *)p;
3033
3034 if (sizeof_priv)
3035 dev->priv = netdev_priv(dev);
3036
3037 setup(dev);
3038 strcpy(dev->name, name);
3039 return dev;
3040}
3041EXPORT_SYMBOL(alloc_netdev);
3042
3043/**
3044 * free_netdev - free network device
3045 * @dev: device
3046 *
3047 * This function does the last stage of destroying an allocated device
3048 * interface. The reference to the device object is released.
3049 * If this is the last reference then it will be freed.
3050 */
3051void free_netdev(struct net_device *dev)
3052{
3053#ifdef CONFIG_SYSFS
3054 /* Compatiablity with error handling in drivers */
3055 if (dev->reg_state == NETREG_UNINITIALIZED) {
3056 kfree((char *)dev - dev->padded);
3057 return;
3058 }
3059
3060 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3061 dev->reg_state = NETREG_RELEASED;
3062
3063 /* will free via class release */
3064 class_device_put(&dev->class_dev);
3065#else
3066 kfree((char *)dev - dev->padded);
3067#endif
3068}
3069
3070/* Synchronize with packet receive processing. */
3071void synchronize_net(void)
3072{
3073 might_sleep();
fbd568a3 3074 synchronize_rcu();
1da177e4
LT
3075}
3076
3077/**
3078 * unregister_netdevice - remove device from the kernel
3079 * @dev: device
3080 *
3081 * This function shuts down a device interface and removes it
3082 * from the kernel tables. On success 0 is returned, on a failure
3083 * a negative errno code is returned.
3084 *
3085 * Callers must hold the rtnl semaphore. You may want
3086 * unregister_netdev() instead of this.
3087 */
3088
3089int unregister_netdevice(struct net_device *dev)
3090{
3091 struct net_device *d, **dp;
3092
3093 BUG_ON(dev_boot_phase);
3094 ASSERT_RTNL();
3095
3096 /* Some devices call without registering for initialization unwind. */
3097 if (dev->reg_state == NETREG_UNINITIALIZED) {
3098 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3099 "was registered\n", dev->name, dev);
3100 return -ENODEV;
3101 }
3102
3103 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3104
3105 /* If device is running, close it first. */
3106 if (dev->flags & IFF_UP)
3107 dev_close(dev);
3108
3109 /* And unlink it from device chain. */
3110 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3111 if (d == dev) {
3112 write_lock_bh(&dev_base_lock);
3113 hlist_del(&dev->name_hlist);
3114 hlist_del(&dev->index_hlist);
3115 if (dev_tail == &dev->next)
3116 dev_tail = dp;
3117 *dp = d->next;
3118 write_unlock_bh(&dev_base_lock);
3119 break;
3120 }
3121 }
3122 if (!d) {
3123 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3124 dev->name);
3125 return -ENODEV;
3126 }
3127
3128 dev->reg_state = NETREG_UNREGISTERING;
3129
3130 synchronize_net();
3131
3132 /* Shutdown queueing discipline. */
3133 dev_shutdown(dev);
3134
3135
3136 /* Notify protocols, that we are about to destroy
3137 this device. They should clean all the things.
3138 */
3139 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3140
3141 /*
3142 * Flush the multicast chain
3143 */
3144 dev_mc_discard(dev);
3145
3146 if (dev->uninit)
3147 dev->uninit(dev);
3148
3149 /* Notifier chain MUST detach us from master device. */
3150 BUG_TRAP(!dev->master);
3151
3152 free_divert_blk(dev);
3153
3154 /* Finish processing unregister after unlock */
3155 net_set_todo(dev);
3156
3157 synchronize_net();
3158
3159 dev_put(dev);
3160 return 0;
3161}
3162
3163/**
3164 * unregister_netdev - remove device from the kernel
3165 * @dev: device
3166 *
3167 * This function shuts down a device interface and removes it
3168 * from the kernel tables. On success 0 is returned, on a failure
3169 * a negative errno code is returned.
3170 *
3171 * This is just a wrapper for unregister_netdevice that takes
3172 * the rtnl semaphore. In general you want to use this and not
3173 * unregister_netdevice.
3174 */
3175void unregister_netdev(struct net_device *dev)
3176{
3177 rtnl_lock();
3178 unregister_netdevice(dev);
3179 rtnl_unlock();
3180}
3181
3182EXPORT_SYMBOL(unregister_netdev);
3183
3184#ifdef CONFIG_HOTPLUG_CPU
3185static int dev_cpu_callback(struct notifier_block *nfb,
3186 unsigned long action,
3187 void *ocpu)
3188{
3189 struct sk_buff **list_skb;
3190 struct net_device **list_net;
3191 struct sk_buff *skb;
3192 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3193 struct softnet_data *sd, *oldsd;
3194
3195 if (action != CPU_DEAD)
3196 return NOTIFY_OK;
3197
3198 local_irq_disable();
3199 cpu = smp_processor_id();
3200 sd = &per_cpu(softnet_data, cpu);
3201 oldsd = &per_cpu(softnet_data, oldcpu);
3202
3203 /* Find end of our completion_queue. */
3204 list_skb = &sd->completion_queue;
3205 while (*list_skb)
3206 list_skb = &(*list_skb)->next;
3207 /* Append completion queue from offline CPU. */
3208 *list_skb = oldsd->completion_queue;
3209 oldsd->completion_queue = NULL;
3210
3211 /* Find end of our output_queue. */
3212 list_net = &sd->output_queue;
3213 while (*list_net)
3214 list_net = &(*list_net)->next_sched;
3215 /* Append output queue from offline CPU. */
3216 *list_net = oldsd->output_queue;
3217 oldsd->output_queue = NULL;
3218
3219 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3220 local_irq_enable();
3221
3222 /* Process offline CPU's input_pkt_queue */
3223 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3224 netif_rx(skb);
3225
3226 return NOTIFY_OK;
3227}
3228#endif /* CONFIG_HOTPLUG_CPU */
3229
3230
3231/*
3232 * Initialize the DEV module. At boot time this walks the device list and
3233 * unhooks any devices that fail to initialise (normally hardware not
3234 * present) and leaves us with a valid list of present and active devices.
3235 *
3236 */
3237
3238/*
3239 * This is called single threaded during boot, so no need
3240 * to take the rtnl semaphore.
3241 */
3242static int __init net_dev_init(void)
3243{
3244 int i, rc = -ENOMEM;
3245
3246 BUG_ON(!dev_boot_phase);
3247
3248 net_random_init();
3249
3250 if (dev_proc_init())
3251 goto out;
3252
3253 if (netdev_sysfs_init())
3254 goto out;
3255
3256 INIT_LIST_HEAD(&ptype_all);
3257 for (i = 0; i < 16; i++)
3258 INIT_LIST_HEAD(&ptype_base[i]);
3259
3260 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3261 INIT_HLIST_HEAD(&dev_name_head[i]);
3262
3263 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3264 INIT_HLIST_HEAD(&dev_index_head[i]);
3265
3266 /*
3267 * Initialise the packet receive queues.
3268 */
3269
88a2a4ac 3270 for_each_cpu(i) {
1da177e4
LT
3271 struct softnet_data *queue;
3272
3273 queue = &per_cpu(softnet_data, i);
3274 skb_queue_head_init(&queue->input_pkt_queue);
1da177e4
LT
3275 queue->completion_queue = NULL;
3276 INIT_LIST_HEAD(&queue->poll_list);
3277 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3278 queue->backlog_dev.weight = weight_p;
3279 queue->backlog_dev.poll = process_backlog;
3280 atomic_set(&queue->backlog_dev.refcnt, 1);
3281 }
3282
1da177e4
LT
3283 dev_boot_phase = 0;
3284
3285 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3286 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3287
3288 hotcpu_notifier(dev_cpu_callback, 0);
3289 dst_init();
3290 dev_mcast_init();
3291 rc = 0;
3292out:
3293 return rc;
3294}
3295
3296subsys_initcall(net_dev_init);
3297
3298EXPORT_SYMBOL(__dev_get_by_index);
3299EXPORT_SYMBOL(__dev_get_by_name);
3300EXPORT_SYMBOL(__dev_remove_pack);
3301EXPORT_SYMBOL(__skb_linearize);
c2373ee9 3302EXPORT_SYMBOL(dev_valid_name);
1da177e4
LT
3303EXPORT_SYMBOL(dev_add_pack);
3304EXPORT_SYMBOL(dev_alloc_name);
3305EXPORT_SYMBOL(dev_close);
3306EXPORT_SYMBOL(dev_get_by_flags);
3307EXPORT_SYMBOL(dev_get_by_index);
3308EXPORT_SYMBOL(dev_get_by_name);
1da177e4
LT
3309EXPORT_SYMBOL(dev_open);
3310EXPORT_SYMBOL(dev_queue_xmit);
3311EXPORT_SYMBOL(dev_remove_pack);
3312EXPORT_SYMBOL(dev_set_allmulti);
3313EXPORT_SYMBOL(dev_set_promiscuity);
3314EXPORT_SYMBOL(dev_change_flags);
3315EXPORT_SYMBOL(dev_set_mtu);
3316EXPORT_SYMBOL(dev_set_mac_address);
3317EXPORT_SYMBOL(free_netdev);
3318EXPORT_SYMBOL(netdev_boot_setup_check);
3319EXPORT_SYMBOL(netdev_set_master);
3320EXPORT_SYMBOL(netdev_state_change);
3321EXPORT_SYMBOL(netif_receive_skb);
3322EXPORT_SYMBOL(netif_rx);
3323EXPORT_SYMBOL(register_gifconf);
3324EXPORT_SYMBOL(register_netdevice);
3325EXPORT_SYMBOL(register_netdevice_notifier);
3326EXPORT_SYMBOL(skb_checksum_help);
3327EXPORT_SYMBOL(synchronize_net);
3328EXPORT_SYMBOL(unregister_netdevice);
3329EXPORT_SYMBOL(unregister_netdevice_notifier);
3330EXPORT_SYMBOL(net_enable_timestamp);
3331EXPORT_SYMBOL(net_disable_timestamp);
3332EXPORT_SYMBOL(dev_get_flags);
3333
3334#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3335EXPORT_SYMBOL(br_handle_frame_hook);
3336EXPORT_SYMBOL(br_fdb_get_hook);
3337EXPORT_SYMBOL(br_fdb_put_hook);
3338#endif
3339
3340#ifdef CONFIG_KMOD
3341EXPORT_SYMBOL(dev_load);
3342#endif
3343
3344EXPORT_PER_CPU_SYMBOL(softnet_data);