]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/core/dev.c
[IRDA] sem2mutex: drivers/net/irda
[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>
84#include <linux/string.h>
85#include <linux/mm.h>
86#include <linux/socket.h>
87#include <linux/sockios.h>
88#include <linux/errno.h>
89#include <linux/interrupt.h>
90#include <linux/if_ether.h>
91#include <linux/netdevice.h>
92#include <linux/etherdevice.h>
93#include <linux/notifier.h>
94#include <linux/skbuff.h>
95#include <net/sock.h>
96#include <linux/rtnetlink.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/stat.h>
100#include <linux/if_bridge.h>
101#include <linux/divert.h>
102#include <net/dst.h>
103#include <net/pkt_sched.h>
104#include <net/checksum.h>
105#include <linux/highmem.h>
106#include <linux/init.h>
107#include <linux/kmod.h>
108#include <linux/module.h>
109#include <linux/kallsyms.h>
110#include <linux/netpoll.h>
111#include <linux/rcupdate.h>
112#include <linux/delay.h>
d86b5e0e 113#include <linux/wireless.h>
1da177e4 114#include <net/iw_handler.h>
1da177e4
LT
115#include <asm/current.h>
116
1da177e4
LT
117/*
118 * The list of packet types we will receive (as opposed to discard)
119 * and the routines to invoke.
120 *
121 * Why 16. Because with 16 the only overlap we get on a hash of the
122 * low nibble of the protocol value is RARP/SNAP/X.25.
123 *
124 * NOTE: That is no longer true with the addition of VLAN tags. Not
125 * sure which should go first, but I bet it won't make much
126 * difference if we are running VLANs. The good news is that
127 * this protocol won't be in the list unless compiled in, so
128 * the average user (w/out VLANs) will not be adversly affected.
129 * --BLG
130 *
131 * 0800 IP
132 * 8100 802.1Q VLAN
133 * 0001 802.3
134 * 0002 AX.25
135 * 0004 802.2
136 * 8035 RARP
137 * 0005 SNAP
138 * 0805 X.25
139 * 0806 ARP
140 * 8137 IPX
141 * 0009 Localtalk
142 * 86DD IPv6
143 */
144
145static DEFINE_SPINLOCK(ptype_lock);
146static struct list_head ptype_base[16]; /* 16 way hashed list */
147static struct list_head ptype_all; /* Taps */
148
1da177e4
LT
149/*
150 * The @dev_base list is protected by @dev_base_lock and the rtln
151 * semaphore.
152 *
153 * Pure readers hold dev_base_lock for reading.
154 *
155 * Writers must hold the rtnl semaphore while they loop through the
156 * dev_base list, and hold dev_base_lock for writing when they do the
157 * actual updates. This allows pure readers to access the list even
158 * while a writer is preparing to update it.
159 *
160 * To put it another way, dev_base_lock is held for writing only to
161 * protect against pure readers; the rtnl semaphore provides the
162 * protection against other writers.
163 *
164 * See, for example usages, register_netdevice() and
165 * unregister_netdevice(), which must be called with the rtnl
166 * semaphore held.
167 */
168struct net_device *dev_base;
169static struct net_device **dev_tail = &dev_base;
170DEFINE_RWLOCK(dev_base_lock);
171
172EXPORT_SYMBOL(dev_base);
173EXPORT_SYMBOL(dev_base_lock);
174
175#define NETDEV_HASHBITS 8
176static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
177static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
178
179static inline struct hlist_head *dev_name_hash(const char *name)
180{
181 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
182 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
183}
184
185static inline struct hlist_head *dev_index_hash(int ifindex)
186{
187 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
188}
189
190/*
191 * Our notifier list
192 */
193
194static struct notifier_block *netdev_chain;
195
196/*
197 * Device drivers call our routines to queue packets here. We empty the
198 * queue in the local softnet handler.
199 */
31aa02c5 200DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
1da177e4
LT
201
202#ifdef CONFIG_SYSFS
203extern int netdev_sysfs_init(void);
204extern int netdev_register_sysfs(struct net_device *);
205extern void netdev_unregister_sysfs(struct net_device *);
206#else
207#define netdev_sysfs_init() (0)
208#define netdev_register_sysfs(dev) (0)
209#define netdev_unregister_sysfs(dev) do { } while(0)
210#endif
211
212
213/*******************************************************************************
214
215 Protocol management and registration routines
216
217*******************************************************************************/
218
219/*
220 * For efficiency
221 */
222
223int netdev_nit;
224
225/*
226 * Add a protocol ID to the list. Now that the input handler is
227 * smarter we can dispense with all the messy stuff that used to be
228 * here.
229 *
230 * BEWARE!!! Protocol handlers, mangling input packets,
231 * MUST BE last in hash buckets and checking protocol handlers
232 * MUST start from promiscuous ptype_all chain in net_bh.
233 * It is true now, do not change it.
234 * Explanation follows: if protocol handler, mangling packet, will
235 * be the first on list, it is not able to sense, that packet
236 * is cloned and should be copied-on-write, so that it will
237 * change it and subsequent readers will get broken packet.
238 * --ANK (980803)
239 */
240
241/**
242 * dev_add_pack - add packet handler
243 * @pt: packet type declaration
244 *
245 * Add a protocol handler to the networking stack. The passed &packet_type
246 * is linked into kernel lists and may not be freed until it has been
247 * removed from the kernel lists.
248 *
249 * This call does not sleep therefore it can not
250 * guarantee all CPU's that are in middle of receiving packets
251 * will see the new packet type (until the next received packet).
252 */
253
254void dev_add_pack(struct packet_type *pt)
255{
256 int hash;
257
258 spin_lock_bh(&ptype_lock);
259 if (pt->type == htons(ETH_P_ALL)) {
260 netdev_nit++;
261 list_add_rcu(&pt->list, &ptype_all);
262 } else {
263 hash = ntohs(pt->type) & 15;
264 list_add_rcu(&pt->list, &ptype_base[hash]);
265 }
266 spin_unlock_bh(&ptype_lock);
267}
268
1da177e4
LT
269/**
270 * __dev_remove_pack - remove packet handler
271 * @pt: packet type declaration
272 *
273 * Remove a protocol handler that was previously added to the kernel
274 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
275 * from the kernel lists and can be freed or reused once this function
276 * returns.
277 *
278 * The packet type might still be in use by receivers
279 * and must not be freed until after all the CPU's have gone
280 * through a quiescent state.
281 */
282void __dev_remove_pack(struct packet_type *pt)
283{
284 struct list_head *head;
285 struct packet_type *pt1;
286
287 spin_lock_bh(&ptype_lock);
288
289 if (pt->type == htons(ETH_P_ALL)) {
290 netdev_nit--;
291 head = &ptype_all;
292 } else
293 head = &ptype_base[ntohs(pt->type) & 15];
294
295 list_for_each_entry(pt1, head, list) {
296 if (pt == pt1) {
297 list_del_rcu(&pt->list);
298 goto out;
299 }
300 }
301
302 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
303out:
304 spin_unlock_bh(&ptype_lock);
305}
306/**
307 * dev_remove_pack - remove packet handler
308 * @pt: packet type declaration
309 *
310 * Remove a protocol handler that was previously added to the kernel
311 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
312 * from the kernel lists and can be freed or reused once this function
313 * returns.
314 *
315 * This call sleeps to guarantee that no CPU is looking at the packet
316 * type after return.
317 */
318void dev_remove_pack(struct packet_type *pt)
319{
320 __dev_remove_pack(pt);
321
322 synchronize_net();
323}
324
325/******************************************************************************
326
327 Device Boot-time Settings Routines
328
329*******************************************************************************/
330
331/* Boot time configuration table */
332static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
333
334/**
335 * netdev_boot_setup_add - add new setup entry
336 * @name: name of the device
337 * @map: configured settings for the device
338 *
339 * Adds new setup entry to the dev_boot_setup list. The function
340 * returns 0 on error and 1 on success. This is a generic routine to
341 * all netdevices.
342 */
343static int netdev_boot_setup_add(char *name, struct ifmap *map)
344{
345 struct netdev_boot_setup *s;
346 int i;
347
348 s = dev_boot_setup;
349 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
350 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
351 memset(s[i].name, 0, sizeof(s[i].name));
352 strcpy(s[i].name, name);
353 memcpy(&s[i].map, map, sizeof(s[i].map));
354 break;
355 }
356 }
357
358 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
359}
360
361/**
362 * netdev_boot_setup_check - check boot time settings
363 * @dev: the netdevice
364 *
365 * Check boot time settings for the device.
366 * The found settings are set for the device to be used
367 * later in the device probing.
368 * Returns 0 if no settings found, 1 if they are.
369 */
370int netdev_boot_setup_check(struct net_device *dev)
371{
372 struct netdev_boot_setup *s = dev_boot_setup;
373 int i;
374
375 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
376 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
377 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
378 dev->irq = s[i].map.irq;
379 dev->base_addr = s[i].map.base_addr;
380 dev->mem_start = s[i].map.mem_start;
381 dev->mem_end = s[i].map.mem_end;
382 return 1;
383 }
384 }
385 return 0;
386}
387
388
389/**
390 * netdev_boot_base - get address from boot time settings
391 * @prefix: prefix for network device
392 * @unit: id for network device
393 *
394 * Check boot time settings for the base address of device.
395 * The found settings are set for the device to be used
396 * later in the device probing.
397 * Returns 0 if no settings found.
398 */
399unsigned long netdev_boot_base(const char *prefix, int unit)
400{
401 const struct netdev_boot_setup *s = dev_boot_setup;
402 char name[IFNAMSIZ];
403 int i;
404
405 sprintf(name, "%s%d", prefix, unit);
406
407 /*
408 * If device already registered then return base of 1
409 * to indicate not to probe for this interface
410 */
411 if (__dev_get_by_name(name))
412 return 1;
413
414 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
415 if (!strcmp(name, s[i].name))
416 return s[i].map.base_addr;
417 return 0;
418}
419
420/*
421 * Saves at boot time configured settings for any netdevice.
422 */
423int __init netdev_boot_setup(char *str)
424{
425 int ints[5];
426 struct ifmap map;
427
428 str = get_options(str, ARRAY_SIZE(ints), ints);
429 if (!str || !*str)
430 return 0;
431
432 /* Save settings */
433 memset(&map, 0, sizeof(map));
434 if (ints[0] > 0)
435 map.irq = ints[1];
436 if (ints[0] > 1)
437 map.base_addr = ints[2];
438 if (ints[0] > 2)
439 map.mem_start = ints[3];
440 if (ints[0] > 3)
441 map.mem_end = ints[4];
442
443 /* Add new entry to the list */
444 return netdev_boot_setup_add(str, &map);
445}
446
447__setup("netdev=", netdev_boot_setup);
448
449/*******************************************************************************
450
451 Device Interface Subroutines
452
453*******************************************************************************/
454
455/**
456 * __dev_get_by_name - find a device by its name
457 * @name: name to find
458 *
459 * Find an interface by name. Must be called under RTNL semaphore
460 * or @dev_base_lock. If the name is found a pointer to the device
461 * is returned. If the name is not found then %NULL is returned. The
462 * reference counters are not incremented so the caller must be
463 * careful with locks.
464 */
465
466struct net_device *__dev_get_by_name(const char *name)
467{
468 struct hlist_node *p;
469
470 hlist_for_each(p, dev_name_hash(name)) {
471 struct net_device *dev
472 = hlist_entry(p, struct net_device, name_hlist);
473 if (!strncmp(dev->name, name, IFNAMSIZ))
474 return dev;
475 }
476 return NULL;
477}
478
479/**
480 * dev_get_by_name - find a device by its name
481 * @name: name to find
482 *
483 * Find an interface by name. This can be called from any
484 * context and does its own locking. The returned handle has
485 * the usage count incremented and the caller must use dev_put() to
486 * release it when it is no longer needed. %NULL is returned if no
487 * matching device is found.
488 */
489
490struct net_device *dev_get_by_name(const char *name)
491{
492 struct net_device *dev;
493
494 read_lock(&dev_base_lock);
495 dev = __dev_get_by_name(name);
496 if (dev)
497 dev_hold(dev);
498 read_unlock(&dev_base_lock);
499 return dev;
500}
501
502/**
503 * __dev_get_by_index - find a device by its ifindex
504 * @ifindex: index of device
505 *
506 * Search for an interface by index. Returns %NULL if the device
507 * is not found or a pointer to the device. The device has not
508 * had its reference counter increased so the caller must be careful
509 * about locking. The caller must hold either the RTNL semaphore
510 * or @dev_base_lock.
511 */
512
513struct net_device *__dev_get_by_index(int ifindex)
514{
515 struct hlist_node *p;
516
517 hlist_for_each(p, dev_index_hash(ifindex)) {
518 struct net_device *dev
519 = hlist_entry(p, struct net_device, index_hlist);
520 if (dev->ifindex == ifindex)
521 return dev;
522 }
523 return NULL;
524}
525
526
527/**
528 * dev_get_by_index - find a device by its ifindex
529 * @ifindex: index of device
530 *
531 * Search for an interface by index. Returns NULL if the device
532 * is not found or a pointer to the device. The device returned has
533 * had a reference added and the pointer is safe until the user calls
534 * dev_put to indicate they have finished with it.
535 */
536
537struct net_device *dev_get_by_index(int ifindex)
538{
539 struct net_device *dev;
540
541 read_lock(&dev_base_lock);
542 dev = __dev_get_by_index(ifindex);
543 if (dev)
544 dev_hold(dev);
545 read_unlock(&dev_base_lock);
546 return dev;
547}
548
549/**
550 * dev_getbyhwaddr - find a device by its hardware address
551 * @type: media type of device
552 * @ha: hardware address
553 *
554 * Search for an interface by MAC address. Returns NULL if the device
555 * is not found or a pointer to the device. The caller must hold the
556 * rtnl semaphore. The returned device has not had its ref count increased
557 * and the caller must therefore be careful about locking
558 *
559 * BUGS:
560 * If the API was consistent this would be __dev_get_by_hwaddr
561 */
562
563struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
564{
565 struct net_device *dev;
566
567 ASSERT_RTNL();
568
569 for (dev = dev_base; dev; dev = dev->next)
570 if (dev->type == type &&
571 !memcmp(dev->dev_addr, ha, dev->addr_len))
572 break;
573 return dev;
574}
575
cf309e3f
JF
576EXPORT_SYMBOL(dev_getbyhwaddr);
577
1da177e4
LT
578struct net_device *dev_getfirstbyhwtype(unsigned short type)
579{
580 struct net_device *dev;
581
582 rtnl_lock();
583 for (dev = dev_base; dev; dev = dev->next) {
584 if (dev->type == type) {
585 dev_hold(dev);
586 break;
587 }
588 }
589 rtnl_unlock();
590 return dev;
591}
592
593EXPORT_SYMBOL(dev_getfirstbyhwtype);
594
595/**
596 * dev_get_by_flags - find any device with given flags
597 * @if_flags: IFF_* values
598 * @mask: bitmask of bits in if_flags to check
599 *
600 * Search for any interface with the given flags. Returns NULL if a device
601 * is not found or a pointer to the device. The device returned has
602 * had a reference added and the pointer is safe until the user calls
603 * dev_put to indicate they have finished with it.
604 */
605
606struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
607{
608 struct net_device *dev;
609
610 read_lock(&dev_base_lock);
611 for (dev = dev_base; dev != NULL; dev = dev->next) {
612 if (((dev->flags ^ if_flags) & mask) == 0) {
613 dev_hold(dev);
614 break;
615 }
616 }
617 read_unlock(&dev_base_lock);
618 return dev;
619}
620
621/**
622 * dev_valid_name - check if name is okay for network device
623 * @name: name string
624 *
625 * Network device names need to be valid file names to
626 * to allow sysfs to work
627 */
c2373ee9 628int dev_valid_name(const char *name)
1da177e4
LT
629{
630 return !(*name == '\0'
631 || !strcmp(name, ".")
632 || !strcmp(name, "..")
633 || strchr(name, '/'));
634}
635
636/**
637 * dev_alloc_name - allocate a name for a device
638 * @dev: device
639 * @name: name format string
640 *
641 * Passed a format string - eg "lt%d" it will try and find a suitable
642 * id. Not efficient for many devices, not called a lot. The caller
643 * must hold the dev_base or rtnl lock while allocating the name and
644 * adding the device in order to avoid duplicates. Returns the number
645 * of the unit assigned or a negative errno code.
646 */
647
648int dev_alloc_name(struct net_device *dev, const char *name)
649{
650 int i = 0;
651 char buf[IFNAMSIZ];
652 const char *p;
653 const int max_netdevices = 8*PAGE_SIZE;
654 long *inuse;
655 struct net_device *d;
656
657 p = strnchr(name, IFNAMSIZ-1, '%');
658 if (p) {
659 /*
660 * Verify the string as this thing may have come from
661 * the user. There must be either one "%d" and no other "%"
662 * characters.
663 */
664 if (p[1] != 'd' || strchr(p + 2, '%'))
665 return -EINVAL;
666
667 /* Use one page as a bit array of possible slots */
668 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
669 if (!inuse)
670 return -ENOMEM;
671
672 for (d = dev_base; d; d = d->next) {
673 if (!sscanf(d->name, name, &i))
674 continue;
675 if (i < 0 || i >= max_netdevices)
676 continue;
677
678 /* avoid cases where sscanf is not exact inverse of printf */
679 snprintf(buf, sizeof(buf), name, i);
680 if (!strncmp(buf, d->name, IFNAMSIZ))
681 set_bit(i, inuse);
682 }
683
684 i = find_first_zero_bit(inuse, max_netdevices);
685 free_page((unsigned long) inuse);
686 }
687
688 snprintf(buf, sizeof(buf), name, i);
689 if (!__dev_get_by_name(buf)) {
690 strlcpy(dev->name, buf, IFNAMSIZ);
691 return i;
692 }
693
694 /* It is possible to run out of possible slots
695 * when the name is long and there isn't enough space left
696 * for the digits, or if all bits are used.
697 */
698 return -ENFILE;
699}
700
701
702/**
703 * dev_change_name - change name of a device
704 * @dev: device
705 * @newname: name (or format string) must be at least IFNAMSIZ
706 *
707 * Change name of a device, can pass format strings "eth%d".
708 * for wildcarding.
709 */
710int dev_change_name(struct net_device *dev, char *newname)
711{
712 int err = 0;
713
714 ASSERT_RTNL();
715
716 if (dev->flags & IFF_UP)
717 return -EBUSY;
718
719 if (!dev_valid_name(newname))
720 return -EINVAL;
721
722 if (strchr(newname, '%')) {
723 err = dev_alloc_name(dev, newname);
724 if (err < 0)
725 return err;
726 strcpy(newname, dev->name);
727 }
728 else if (__dev_get_by_name(newname))
729 return -EEXIST;
730 else
731 strlcpy(dev->name, newname, IFNAMSIZ);
732
733 err = class_device_rename(&dev->class_dev, dev->name);
734 if (!err) {
735 hlist_del(&dev->name_hlist);
736 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
737 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
738 }
739
740 return err;
741}
742
d8a33ac4
SH
743/**
744 * netdev_features_change - device changes fatures
745 * @dev: device to cause notification
746 *
747 * Called to indicate a device has changed features.
748 */
749void netdev_features_change(struct net_device *dev)
750{
751 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
752}
753EXPORT_SYMBOL(netdev_features_change);
754
1da177e4
LT
755/**
756 * netdev_state_change - device changes state
757 * @dev: device to cause notification
758 *
759 * Called to indicate a device has changed state. This function calls
760 * the notifier chains for netdev_chain and sends a NEWLINK message
761 * to the routing socket.
762 */
763void netdev_state_change(struct net_device *dev)
764{
765 if (dev->flags & IFF_UP) {
766 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
767 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
768 }
769}
770
771/**
772 * dev_load - load a network module
773 * @name: name of interface
774 *
775 * If a network interface is not present and the process has suitable
776 * privileges this function loads the module. If module loading is not
777 * available in this kernel then it becomes a nop.
778 */
779
780void dev_load(const char *name)
781{
782 struct net_device *dev;
783
784 read_lock(&dev_base_lock);
785 dev = __dev_get_by_name(name);
786 read_unlock(&dev_base_lock);
787
788 if (!dev && capable(CAP_SYS_MODULE))
789 request_module("%s", name);
790}
791
792static int default_rebuild_header(struct sk_buff *skb)
793{
794 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
795 skb->dev ? skb->dev->name : "NULL!!!");
796 kfree_skb(skb);
797 return 1;
798}
799
800
801/**
802 * dev_open - prepare an interface for use.
803 * @dev: device to open
804 *
805 * Takes a device from down to up state. The device's private open
806 * function is invoked and then the multicast lists are loaded. Finally
807 * the device is moved into the up state and a %NETDEV_UP message is
808 * sent to the netdev notifier chain.
809 *
810 * Calling this function on an active interface is a nop. On a failure
811 * a negative errno code is returned.
812 */
813int dev_open(struct net_device *dev)
814{
815 int ret = 0;
816
817 /*
818 * Is it already up?
819 */
820
821 if (dev->flags & IFF_UP)
822 return 0;
823
824 /*
825 * Is it even present?
826 */
827 if (!netif_device_present(dev))
828 return -ENODEV;
829
830 /*
831 * Call device private open method
832 */
833 set_bit(__LINK_STATE_START, &dev->state);
834 if (dev->open) {
835 ret = dev->open(dev);
836 if (ret)
837 clear_bit(__LINK_STATE_START, &dev->state);
838 }
839
840 /*
841 * If it went open OK then:
842 */
843
844 if (!ret) {
845 /*
846 * Set the flags.
847 */
848 dev->flags |= IFF_UP;
849
850 /*
851 * Initialize multicasting status
852 */
853 dev_mc_upload(dev);
854
855 /*
856 * Wakeup transmit queue engine
857 */
858 dev_activate(dev);
859
860 /*
861 * ... and announce new interface.
862 */
863 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
864 }
865 return ret;
866}
867
868/**
869 * dev_close - shutdown an interface.
870 * @dev: device to shutdown
871 *
872 * This function moves an active device into down state. A
873 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
874 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
875 * chain.
876 */
877int dev_close(struct net_device *dev)
878{
879 if (!(dev->flags & IFF_UP))
880 return 0;
881
882 /*
883 * Tell people we are going down, so that they can
884 * prepare to death, when device is still operating.
885 */
886 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
887
888 dev_deactivate(dev);
889
890 clear_bit(__LINK_STATE_START, &dev->state);
891
892 /* Synchronize to scheduled poll. We cannot touch poll list,
893 * it can be even on different cpu. So just clear netif_running(),
894 * and wait when poll really will happen. Actually, the best place
895 * for this is inside dev->stop() after device stopped its irq
896 * engine, but this requires more changes in devices. */
897
898 smp_mb__after_clear_bit(); /* Commit netif_running(). */
899 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
900 /* No hurry. */
6192b54b 901 msleep(1);
1da177e4
LT
902 }
903
904 /*
905 * Call the device specific close. This cannot fail.
906 * Only if device is UP
907 *
908 * We allow it to be called even after a DETACH hot-plug
909 * event.
910 */
911 if (dev->stop)
912 dev->stop(dev);
913
914 /*
915 * Device is now down.
916 */
917
918 dev->flags &= ~IFF_UP;
919
920 /*
921 * Tell people we are down
922 */
923 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
924
925 return 0;
926}
927
928
929/*
930 * Device change register/unregister. These are not inline or static
931 * as we export them to the world.
932 */
933
934/**
935 * register_netdevice_notifier - register a network notifier block
936 * @nb: notifier
937 *
938 * Register a notifier to be called when network device events occur.
939 * The notifier passed is linked into the kernel structures and must
940 * not be reused until it has been unregistered. A negative errno code
941 * is returned on a failure.
942 *
943 * When registered all registration and up events are replayed
944 * to the new notifier to allow device to have a race free
945 * view of the network device list.
946 */
947
948int register_netdevice_notifier(struct notifier_block *nb)
949{
950 struct net_device *dev;
951 int err;
952
953 rtnl_lock();
954 err = notifier_chain_register(&netdev_chain, nb);
955 if (!err) {
956 for (dev = dev_base; dev; dev = dev->next) {
957 nb->notifier_call(nb, NETDEV_REGISTER, dev);
958
959 if (dev->flags & IFF_UP)
960 nb->notifier_call(nb, NETDEV_UP, dev);
961 }
962 }
963 rtnl_unlock();
964 return err;
965}
966
967/**
968 * unregister_netdevice_notifier - unregister a network notifier block
969 * @nb: notifier
970 *
971 * Unregister a notifier previously registered by
972 * register_netdevice_notifier(). The notifier is unlinked into the
973 * kernel structures and may then be reused. A negative errno code
974 * is returned on a failure.
975 */
976
977int unregister_netdevice_notifier(struct notifier_block *nb)
978{
979 return notifier_chain_unregister(&netdev_chain, nb);
980}
981
982/**
983 * call_netdevice_notifiers - call all network notifier blocks
984 * @val: value passed unmodified to notifier function
985 * @v: pointer passed unmodified to notifier function
986 *
987 * Call all network notifier blocks. Parameters and return value
988 * are as for notifier_call_chain().
989 */
990
991int call_netdevice_notifiers(unsigned long val, void *v)
992{
993 return notifier_call_chain(&netdev_chain, val, v);
994}
995
996/* When > 0 there are consumers of rx skb time stamps */
997static atomic_t netstamp_needed = ATOMIC_INIT(0);
998
999void net_enable_timestamp(void)
1000{
1001 atomic_inc(&netstamp_needed);
1002}
1003
1004void net_disable_timestamp(void)
1005{
1006 atomic_dec(&netstamp_needed);
1007}
1008
a61bbcf2
PM
1009void __net_timestamp(struct sk_buff *skb)
1010{
1011 struct timeval tv;
1012
1013 do_gettimeofday(&tv);
1014 skb_set_timestamp(skb, &tv);
1015}
1016EXPORT_SYMBOL(__net_timestamp);
1017
1018static inline void net_timestamp(struct sk_buff *skb)
1da177e4
LT
1019{
1020 if (atomic_read(&netstamp_needed))
a61bbcf2 1021 __net_timestamp(skb);
1da177e4 1022 else {
a61bbcf2
PM
1023 skb->tstamp.off_sec = 0;
1024 skb->tstamp.off_usec = 0;
1da177e4
LT
1025 }
1026}
1027
1028/*
1029 * Support routine. Sends outgoing frames to any network
1030 * taps currently in use.
1031 */
1032
1033void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1034{
1035 struct packet_type *ptype;
a61bbcf2
PM
1036
1037 net_timestamp(skb);
1da177e4
LT
1038
1039 rcu_read_lock();
1040 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1041 /* Never send packets back to the socket
1042 * they originated from - MvS (miquels@drinkel.ow.org)
1043 */
1044 if ((ptype->dev == dev || !ptype->dev) &&
1045 (ptype->af_packet_priv == NULL ||
1046 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1047 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1048 if (!skb2)
1049 break;
1050
1051 /* skb->nh should be correctly
1052 set by sender, so that the second statement is
1053 just protection against buggy protocols.
1054 */
1055 skb2->mac.raw = skb2->data;
1056
1057 if (skb2->nh.raw < skb2->data ||
1058 skb2->nh.raw > skb2->tail) {
1059 if (net_ratelimit())
1060 printk(KERN_CRIT "protocol %04x is "
1061 "buggy, dev %s\n",
1062 skb2->protocol, dev->name);
1063 skb2->nh.raw = skb2->data;
1064 }
1065
1066 skb2->h.raw = skb2->nh.raw;
1067 skb2->pkt_type = PACKET_OUTGOING;
f2ccd8fa 1068 ptype->func(skb2, skb->dev, ptype, skb->dev);
1da177e4
LT
1069 }
1070 }
1071 rcu_read_unlock();
1072}
1073
1074/*
1075 * Invalidate hardware checksum when packet is to be mangled, and
1076 * complete checksum manually on outgoing path.
1077 */
1078int skb_checksum_help(struct sk_buff *skb, int inward)
1079{
1080 unsigned int csum;
1081 int ret = 0, offset = skb->h.raw - skb->data;
1082
1083 if (inward) {
1084 skb->ip_summed = CHECKSUM_NONE;
1085 goto out;
1086 }
1087
1088 if (skb_cloned(skb)) {
1089 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1090 if (ret)
1091 goto out;
1092 }
1093
09a62660 1094 BUG_ON(offset > (int)skb->len);
1da177e4
LT
1095 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1096
1097 offset = skb->tail - skb->h.raw;
09a62660
KK
1098 BUG_ON(offset <= 0);
1099 BUG_ON(skb->csum + 2 > offset);
1da177e4
LT
1100
1101 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1102 skb->ip_summed = CHECKSUM_NONE;
1103out:
1104 return ret;
1105}
1106
fb286bb2
HX
1107/* Take action when hardware reception checksum errors are detected. */
1108#ifdef CONFIG_BUG
1109void netdev_rx_csum_fault(struct net_device *dev)
1110{
1111 if (net_ratelimit()) {
246a4212
SH
1112 printk(KERN_ERR "%s: hw csum failure.\n",
1113 dev ? dev->name : "<unknown>");
fb286bb2
HX
1114 dump_stack();
1115 }
1116}
1117EXPORT_SYMBOL(netdev_rx_csum_fault);
1118#endif
1119
1da177e4
LT
1120#ifdef CONFIG_HIGHMEM
1121/* Actually, we should eliminate this check as soon as we know, that:
1122 * 1. IOMMU is present and allows to map all the memory.
1123 * 2. No high memory really exists on this machine.
1124 */
1125
1126static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1127{
1128 int i;
1129
1130 if (dev->features & NETIF_F_HIGHDMA)
1131 return 0;
1132
1133 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1134 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1135 return 1;
1136
1137 return 0;
1138}
1139#else
1140#define illegal_highdma(dev, skb) (0)
1141#endif
1142
1da177e4 1143/* Keep head the same: replace data */
dd0fc66f 1144int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1da177e4
LT
1145{
1146 unsigned int size;
1147 u8 *data;
1148 long offset;
1149 struct skb_shared_info *ninfo;
1150 int headerlen = skb->data - skb->head;
1151 int expand = (skb->tail + skb->data_len) - skb->end;
1152
1153 if (skb_shared(skb))
1154 BUG();
1155
1156 if (expand <= 0)
1157 expand = 0;
1158
1159 size = skb->end - skb->head + expand;
1160 size = SKB_DATA_ALIGN(size);
1161 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1162 if (!data)
1163 return -ENOMEM;
1164
1165 /* Copy entire thing */
1166 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1167 BUG();
1168
1169 /* Set up shinfo */
1170 ninfo = (struct skb_shared_info*)(data + size);
1171 atomic_set(&ninfo->dataref, 1);
1172 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1173 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1174 ninfo->nr_frags = 0;
1175 ninfo->frag_list = NULL;
1176
1177 /* Offset between the two in bytes */
1178 offset = data - skb->head;
1179
1180 /* Free old data. */
1181 skb_release_data(skb);
1182
1183 skb->head = data;
1184 skb->end = data + size;
1185
1186 /* Set up new pointers */
1187 skb->h.raw += offset;
1188 skb->nh.raw += offset;
1189 skb->mac.raw += offset;
1190 skb->tail += offset;
1191 skb->data += offset;
1192
1193 /* We are no longer a clone, even if we were. */
1194 skb->cloned = 0;
1195
1196 skb->tail += skb->data_len;
1197 skb->data_len = 0;
1198 return 0;
1199}
1200
1201#define HARD_TX_LOCK(dev, cpu) { \
1202 if ((dev->features & NETIF_F_LLTX) == 0) { \
1203 spin_lock(&dev->xmit_lock); \
1204 dev->xmit_lock_owner = cpu; \
1205 } \
1206}
1207
1208#define HARD_TX_UNLOCK(dev) { \
1209 if ((dev->features & NETIF_F_LLTX) == 0) { \
1210 dev->xmit_lock_owner = -1; \
1211 spin_unlock(&dev->xmit_lock); \
1212 } \
1213}
1214
1215/**
1216 * dev_queue_xmit - transmit a buffer
1217 * @skb: buffer to transmit
1218 *
1219 * Queue a buffer for transmission to a network device. The caller must
1220 * have set the device and priority and built the buffer before calling
1221 * this function. The function can be called from an interrupt.
1222 *
1223 * A negative errno code is returned on a failure. A success does not
1224 * guarantee the frame will be transmitted as it may be dropped due
1225 * to congestion or traffic shaping.
af191367
BG
1226 *
1227 * -----------------------------------------------------------------------------------
1228 * I notice this method can also return errors from the queue disciplines,
1229 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1230 * be positive.
1231 *
1232 * Regardless of the return value, the skb is consumed, so it is currently
1233 * difficult to retry a send to this method. (You can bump the ref count
1234 * before sending to hold a reference for retry if you are careful.)
1235 *
1236 * When calling this method, interrupts MUST be enabled. This is because
1237 * the BH enable code must have IRQs enabled so that it will not deadlock.
1238 * --BLG
1da177e4
LT
1239 */
1240
1241int dev_queue_xmit(struct sk_buff *skb)
1242{
1243 struct net_device *dev = skb->dev;
1244 struct Qdisc *q;
1245 int rc = -ENOMEM;
1246
1247 if (skb_shinfo(skb)->frag_list &&
1248 !(dev->features & NETIF_F_FRAGLIST) &&
1249 __skb_linearize(skb, GFP_ATOMIC))
1250 goto out_kfree_skb;
1251
1252 /* Fragmented skb is linearized if device does not support SG,
1253 * or if at least one of fragments is in highmem and device
1254 * does not support DMA from it.
1255 */
1256 if (skb_shinfo(skb)->nr_frags &&
1257 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1258 __skb_linearize(skb, GFP_ATOMIC))
1259 goto out_kfree_skb;
1260
1261 /* If packet is not checksummed and device does not support
1262 * checksumming for this protocol, complete checksumming here.
1263 */
1264 if (skb->ip_summed == CHECKSUM_HW &&
1265 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1266 (!(dev->features & NETIF_F_IP_CSUM) ||
1267 skb->protocol != htons(ETH_P_IP))))
1268 if (skb_checksum_help(skb, 0))
1269 goto out_kfree_skb;
1270
2d7ceece
ED
1271 spin_lock_prefetch(&dev->queue_lock);
1272
1da177e4
LT
1273 /* Disable soft irqs for various locks below. Also
1274 * stops preemption for RCU.
1275 */
1276 local_bh_disable();
1277
1278 /* Updates of qdisc are serialized by queue_lock.
1279 * The struct Qdisc which is pointed to by qdisc is now a
1280 * rcu structure - it may be accessed without acquiring
1281 * a lock (but the structure may be stale.) The freeing of the
1282 * qdisc will be deferred until it's known that there are no
1283 * more references to it.
1284 *
1285 * If the qdisc has an enqueue function, we still need to
1286 * hold the queue_lock before calling it, since queue_lock
1287 * also serializes access to the device queue.
1288 */
1289
1290 q = rcu_dereference(dev->qdisc);
1291#ifdef CONFIG_NET_CLS_ACT
1292 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1293#endif
1294 if (q->enqueue) {
1295 /* Grab device queue */
1296 spin_lock(&dev->queue_lock);
1297
1298 rc = q->enqueue(skb, q);
1299
1300 qdisc_run(dev);
1301
1302 spin_unlock(&dev->queue_lock);
1303 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1304 goto out;
1305 }
1306
1307 /* The device has no queue. Common case for software devices:
1308 loopback, all the sorts of tunnels...
1309
1310 Really, it is unlikely that xmit_lock protection is necessary here.
1311 (f.e. loopback and IP tunnels are clean ignoring statistics
1312 counters.)
1313 However, it is possible, that they rely on protection
1314 made by us here.
1315
1316 Check this and shot the lock. It is not prone from deadlocks.
1317 Either shot noqueue qdisc, it is even simpler 8)
1318 */
1319 if (dev->flags & IFF_UP) {
1320 int cpu = smp_processor_id(); /* ok because BHs are off */
1321
1322 if (dev->xmit_lock_owner != cpu) {
1323
1324 HARD_TX_LOCK(dev, cpu);
1325
1326 if (!netif_queue_stopped(dev)) {
1327 if (netdev_nit)
1328 dev_queue_xmit_nit(skb, dev);
1329
1330 rc = 0;
1331 if (!dev->hard_start_xmit(skb, dev)) {
1332 HARD_TX_UNLOCK(dev);
1333 goto out;
1334 }
1335 }
1336 HARD_TX_UNLOCK(dev);
1337 if (net_ratelimit())
1338 printk(KERN_CRIT "Virtual device %s asks to "
1339 "queue packet!\n", dev->name);
1340 } else {
1341 /* Recursion is detected! It is possible,
1342 * unfortunately */
1343 if (net_ratelimit())
1344 printk(KERN_CRIT "Dead loop on virtual device "
1345 "%s, fix it urgently!\n", dev->name);
1346 }
1347 }
1348
1349 rc = -ENETDOWN;
1350 local_bh_enable();
1351
1352out_kfree_skb:
1353 kfree_skb(skb);
1354 return rc;
1355out:
1356 local_bh_enable();
1357 return rc;
1358}
1359
1360
1361/*=======================================================================
1362 Receiver routines
1363 =======================================================================*/
1364
51b0bded
SH
1365int netdev_max_backlog = 1000;
1366int netdev_budget = 300;
1da177e4 1367int weight_p = 64; /* old backlog weight */
1da177e4
LT
1368
1369DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1370
1371
1da177e4
LT
1372/**
1373 * netif_rx - post buffer to the network code
1374 * @skb: buffer to post
1375 *
1376 * This function receives a packet from a device driver and queues it for
1377 * the upper (protocol) levels to process. It always succeeds. The buffer
1378 * may be dropped during processing for congestion control or by the
1379 * protocol layers.
1380 *
1381 * return values:
1382 * NET_RX_SUCCESS (no congestion)
1383 * NET_RX_CN_LOW (low congestion)
1384 * NET_RX_CN_MOD (moderate congestion)
1385 * NET_RX_CN_HIGH (high congestion)
1386 * NET_RX_DROP (packet was dropped)
1387 *
1388 */
1389
1390int netif_rx(struct sk_buff *skb)
1391{
1da177e4
LT
1392 struct softnet_data *queue;
1393 unsigned long flags;
1394
1395 /* if netpoll wants it, pretend we never saw it */
1396 if (netpoll_rx(skb))
1397 return NET_RX_DROP;
1398
a61bbcf2
PM
1399 if (!skb->tstamp.off_sec)
1400 net_timestamp(skb);
1da177e4
LT
1401
1402 /*
1403 * The code is rearranged so that the path is the most
1404 * short when CPU is congested, but is still operating.
1405 */
1406 local_irq_save(flags);
1da177e4
LT
1407 queue = &__get_cpu_var(softnet_data);
1408
1409 __get_cpu_var(netdev_rx_stat).total++;
1410 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1411 if (queue->input_pkt_queue.qlen) {
1da177e4
LT
1412enqueue:
1413 dev_hold(skb->dev);
1414 __skb_queue_tail(&queue->input_pkt_queue, skb);
1da177e4 1415 local_irq_restore(flags);
34008d8c 1416 return NET_RX_SUCCESS;
1da177e4
LT
1417 }
1418
1da177e4
LT
1419 netif_rx_schedule(&queue->backlog_dev);
1420 goto enqueue;
1421 }
1422
1da177e4
LT
1423 __get_cpu_var(netdev_rx_stat).dropped++;
1424 local_irq_restore(flags);
1425
1426 kfree_skb(skb);
1427 return NET_RX_DROP;
1428}
1429
1430int netif_rx_ni(struct sk_buff *skb)
1431{
1432 int err;
1433
1434 preempt_disable();
1435 err = netif_rx(skb);
1436 if (local_softirq_pending())
1437 do_softirq();
1438 preempt_enable();
1439
1440 return err;
1441}
1442
1443EXPORT_SYMBOL(netif_rx_ni);
1444
f2ccd8fa 1445static inline struct net_device *skb_bond(struct sk_buff *skb)
1da177e4
LT
1446{
1447 struct net_device *dev = skb->dev;
1448
8f903c70
JV
1449 if (dev->master) {
1450 /*
1451 * On bonding slaves other than the currently active
1452 * slave, suppress duplicates except for 802.3ad
1453 * ETH_P_SLOW and alb non-mcast/bcast.
1454 */
1455 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1456 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1457 if (skb->pkt_type != PACKET_BROADCAST &&
1458 skb->pkt_type != PACKET_MULTICAST)
1459 goto keep;
1460 }
1461
1462 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1463 skb->protocol == __constant_htons(ETH_P_SLOW))
1464 goto keep;
1465
1466 kfree_skb(skb);
1467 return NULL;
1468 }
1469keep:
1da177e4 1470 skb->dev = dev->master;
8f903c70 1471 }
f2ccd8fa
DM
1472
1473 return dev;
1da177e4
LT
1474}
1475
1476static void net_tx_action(struct softirq_action *h)
1477{
1478 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1479
1480 if (sd->completion_queue) {
1481 struct sk_buff *clist;
1482
1483 local_irq_disable();
1484 clist = sd->completion_queue;
1485 sd->completion_queue = NULL;
1486 local_irq_enable();
1487
1488 while (clist) {
1489 struct sk_buff *skb = clist;
1490 clist = clist->next;
1491
1492 BUG_TRAP(!atomic_read(&skb->users));
1493 __kfree_skb(skb);
1494 }
1495 }
1496
1497 if (sd->output_queue) {
1498 struct net_device *head;
1499
1500 local_irq_disable();
1501 head = sd->output_queue;
1502 sd->output_queue = NULL;
1503 local_irq_enable();
1504
1505 while (head) {
1506 struct net_device *dev = head;
1507 head = head->next_sched;
1508
1509 smp_mb__before_clear_bit();
1510 clear_bit(__LINK_STATE_SCHED, &dev->state);
1511
1512 if (spin_trylock(&dev->queue_lock)) {
1513 qdisc_run(dev);
1514 spin_unlock(&dev->queue_lock);
1515 } else {
1516 netif_schedule(dev);
1517 }
1518 }
1519 }
1520}
1521
1522static __inline__ int deliver_skb(struct sk_buff *skb,
f2ccd8fa
DM
1523 struct packet_type *pt_prev,
1524 struct net_device *orig_dev)
1da177e4
LT
1525{
1526 atomic_inc(&skb->users);
f2ccd8fa 1527 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT
1528}
1529
1530#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1531int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1532struct net_bridge;
1533struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1534 unsigned char *addr);
1535void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1536
1537static __inline__ int handle_bridge(struct sk_buff **pskb,
f2ccd8fa
DM
1538 struct packet_type **pt_prev, int *ret,
1539 struct net_device *orig_dev)
1da177e4
LT
1540{
1541 struct net_bridge_port *port;
1542
1543 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1544 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1545 return 0;
1546
1547 if (*pt_prev) {
f2ccd8fa 1548 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1da177e4
LT
1549 *pt_prev = NULL;
1550 }
1551
1552 return br_handle_frame_hook(port, pskb);
1553}
1554#else
f2ccd8fa 1555#define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1da177e4
LT
1556#endif
1557
1558#ifdef CONFIG_NET_CLS_ACT
1559/* TODO: Maybe we should just force sch_ingress to be compiled in
1560 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1561 * a compare and 2 stores extra right now if we dont have it on
1562 * but have CONFIG_NET_CLS_ACT
1563 * NOTE: This doesnt stop any functionality; if you dont have
1564 * the ingress scheduler, you just cant add policies on ingress.
1565 *
1566 */
1567static int ing_filter(struct sk_buff *skb)
1568{
1569 struct Qdisc *q;
1570 struct net_device *dev = skb->dev;
1571 int result = TC_ACT_OK;
1572
1573 if (dev->qdisc_ingress) {
1574 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1575 if (MAX_RED_LOOP < ttl++) {
1576 printk("Redir loop detected Dropping packet (%s->%s)\n",
86e65da9 1577 skb->input_dev->name, skb->dev->name);
1da177e4
LT
1578 return TC_ACT_SHOT;
1579 }
1580
1581 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1582
1583 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
86e65da9 1584
1da177e4
LT
1585 spin_lock(&dev->ingress_lock);
1586 if ((q = dev->qdisc_ingress) != NULL)
1587 result = q->enqueue(skb, q);
1588 spin_unlock(&dev->ingress_lock);
1589
1590 }
1591
1592 return result;
1593}
1594#endif
1595
1596int netif_receive_skb(struct sk_buff *skb)
1597{
1598 struct packet_type *ptype, *pt_prev;
f2ccd8fa 1599 struct net_device *orig_dev;
1da177e4
LT
1600 int ret = NET_RX_DROP;
1601 unsigned short type;
1602
1603 /* if we've gotten here through NAPI, check netpoll */
1604 if (skb->dev->poll && netpoll_rx(skb))
1605 return NET_RX_DROP;
1606
a61bbcf2
PM
1607 if (!skb->tstamp.off_sec)
1608 net_timestamp(skb);
1da177e4 1609
86e65da9
DM
1610 if (!skb->input_dev)
1611 skb->input_dev = skb->dev;
1612
f2ccd8fa 1613 orig_dev = skb_bond(skb);
1da177e4 1614
8f903c70
JV
1615 if (!orig_dev)
1616 return NET_RX_DROP;
1617
1da177e4
LT
1618 __get_cpu_var(netdev_rx_stat).total++;
1619
1620 skb->h.raw = skb->nh.raw = skb->data;
1621 skb->mac_len = skb->nh.raw - skb->mac.raw;
1622
1623 pt_prev = NULL;
1624
1625 rcu_read_lock();
1626
1627#ifdef CONFIG_NET_CLS_ACT
1628 if (skb->tc_verd & TC_NCLS) {
1629 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1630 goto ncls;
1631 }
1632#endif
1633
1634 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1635 if (!ptype->dev || ptype->dev == skb->dev) {
1636 if (pt_prev)
f2ccd8fa 1637 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
1638 pt_prev = ptype;
1639 }
1640 }
1641
1642#ifdef CONFIG_NET_CLS_ACT
1643 if (pt_prev) {
f2ccd8fa 1644 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
1645 pt_prev = NULL; /* noone else should process this after*/
1646 } else {
1647 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1648 }
1649
1650 ret = ing_filter(skb);
1651
1652 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1653 kfree_skb(skb);
1654 goto out;
1655 }
1656
1657 skb->tc_verd = 0;
1658ncls:
1659#endif
1660
1661 handle_diverter(skb);
1662
f2ccd8fa 1663 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1da177e4
LT
1664 goto out;
1665
1666 type = skb->protocol;
1667 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1668 if (ptype->type == type &&
1669 (!ptype->dev || ptype->dev == skb->dev)) {
1670 if (pt_prev)
f2ccd8fa 1671 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
1672 pt_prev = ptype;
1673 }
1674 }
1675
1676 if (pt_prev) {
f2ccd8fa 1677 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT
1678 } else {
1679 kfree_skb(skb);
1680 /* Jamal, now you will not able to escape explaining
1681 * me how you were going to use this. :-)
1682 */
1683 ret = NET_RX_DROP;
1684 }
1685
1686out:
1687 rcu_read_unlock();
1688 return ret;
1689}
1690
1691static int process_backlog(struct net_device *backlog_dev, int *budget)
1692{
1693 int work = 0;
1694 int quota = min(backlog_dev->quota, *budget);
1695 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1696 unsigned long start_time = jiffies;
1697
e3876605 1698 backlog_dev->weight = weight_p;
1da177e4
LT
1699 for (;;) {
1700 struct sk_buff *skb;
1701 struct net_device *dev;
1702
1703 local_irq_disable();
1704 skb = __skb_dequeue(&queue->input_pkt_queue);
1705 if (!skb)
1706 goto job_done;
1707 local_irq_enable();
1708
1709 dev = skb->dev;
1710
1711 netif_receive_skb(skb);
1712
1713 dev_put(dev);
1714
1715 work++;
1716
1717 if (work >= quota || jiffies - start_time > 1)
1718 break;
1719
1720 }
1721
1722 backlog_dev->quota -= work;
1723 *budget -= work;
1724 return -1;
1725
1726job_done:
1727 backlog_dev->quota -= work;
1728 *budget -= work;
1729
1730 list_del(&backlog_dev->poll_list);
1731 smp_mb__before_clear_bit();
1732 netif_poll_enable(backlog_dev);
1733
1da177e4
LT
1734 local_irq_enable();
1735 return 0;
1736}
1737
1738static void net_rx_action(struct softirq_action *h)
1739{
1740 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1741 unsigned long start_time = jiffies;
51b0bded 1742 int budget = netdev_budget;
53fb95d3
MM
1743 void *have;
1744
1da177e4
LT
1745 local_irq_disable();
1746
1747 while (!list_empty(&queue->poll_list)) {
1748 struct net_device *dev;
1749
1750 if (budget <= 0 || jiffies - start_time > 1)
1751 goto softnet_break;
1752
1753 local_irq_enable();
1754
1755 dev = list_entry(queue->poll_list.next,
1756 struct net_device, poll_list);
53fb95d3 1757 have = netpoll_poll_lock(dev);
1da177e4
LT
1758
1759 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
53fb95d3 1760 netpoll_poll_unlock(have);
1da177e4 1761 local_irq_disable();
8aca8a27 1762 list_move_tail(&dev->poll_list, &queue->poll_list);
1da177e4
LT
1763 if (dev->quota < 0)
1764 dev->quota += dev->weight;
1765 else
1766 dev->quota = dev->weight;
1767 } else {
53fb95d3 1768 netpoll_poll_unlock(have);
1da177e4
LT
1769 dev_put(dev);
1770 local_irq_disable();
1771 }
1772 }
1773out:
1774 local_irq_enable();
1775 return;
1776
1777softnet_break:
1778 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1779 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1780 goto out;
1781}
1782
1783static gifconf_func_t * gifconf_list [NPROTO];
1784
1785/**
1786 * register_gifconf - register a SIOCGIF handler
1787 * @family: Address family
1788 * @gifconf: Function handler
1789 *
1790 * Register protocol dependent address dumping routines. The handler
1791 * that is passed must not be freed or reused until it has been replaced
1792 * by another handler.
1793 */
1794int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1795{
1796 if (family >= NPROTO)
1797 return -EINVAL;
1798 gifconf_list[family] = gifconf;
1799 return 0;
1800}
1801
1802
1803/*
1804 * Map an interface index to its name (SIOCGIFNAME)
1805 */
1806
1807/*
1808 * We need this ioctl for efficient implementation of the
1809 * if_indextoname() function required by the IPv6 API. Without
1810 * it, we would have to search all the interfaces to find a
1811 * match. --pb
1812 */
1813
1814static int dev_ifname(struct ifreq __user *arg)
1815{
1816 struct net_device *dev;
1817 struct ifreq ifr;
1818
1819 /*
1820 * Fetch the caller's info block.
1821 */
1822
1823 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1824 return -EFAULT;
1825
1826 read_lock(&dev_base_lock);
1827 dev = __dev_get_by_index(ifr.ifr_ifindex);
1828 if (!dev) {
1829 read_unlock(&dev_base_lock);
1830 return -ENODEV;
1831 }
1832
1833 strcpy(ifr.ifr_name, dev->name);
1834 read_unlock(&dev_base_lock);
1835
1836 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1837 return -EFAULT;
1838 return 0;
1839}
1840
1841/*
1842 * Perform a SIOCGIFCONF call. This structure will change
1843 * size eventually, and there is nothing I can do about it.
1844 * Thus we will need a 'compatibility mode'.
1845 */
1846
1847static int dev_ifconf(char __user *arg)
1848{
1849 struct ifconf ifc;
1850 struct net_device *dev;
1851 char __user *pos;
1852 int len;
1853 int total;
1854 int i;
1855
1856 /*
1857 * Fetch the caller's info block.
1858 */
1859
1860 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1861 return -EFAULT;
1862
1863 pos = ifc.ifc_buf;
1864 len = ifc.ifc_len;
1865
1866 /*
1867 * Loop over the interfaces, and write an info block for each.
1868 */
1869
1870 total = 0;
1871 for (dev = dev_base; dev; dev = dev->next) {
1872 for (i = 0; i < NPROTO; i++) {
1873 if (gifconf_list[i]) {
1874 int done;
1875 if (!pos)
1876 done = gifconf_list[i](dev, NULL, 0);
1877 else
1878 done = gifconf_list[i](dev, pos + total,
1879 len - total);
1880 if (done < 0)
1881 return -EFAULT;
1882 total += done;
1883 }
1884 }
1885 }
1886
1887 /*
1888 * All done. Write the updated control block back to the caller.
1889 */
1890 ifc.ifc_len = total;
1891
1892 /*
1893 * Both BSD and Solaris return 0 here, so we do too.
1894 */
1895 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1896}
1897
1898#ifdef CONFIG_PROC_FS
1899/*
1900 * This is invoked by the /proc filesystem handler to display a device
1901 * in detail.
1902 */
1903static __inline__ struct net_device *dev_get_idx(loff_t pos)
1904{
1905 struct net_device *dev;
1906 loff_t i;
1907
1908 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1909
1910 return i == pos ? dev : NULL;
1911}
1912
1913void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1914{
1915 read_lock(&dev_base_lock);
1916 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1917}
1918
1919void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1920{
1921 ++*pos;
1922 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1923}
1924
1925void dev_seq_stop(struct seq_file *seq, void *v)
1926{
1927 read_unlock(&dev_base_lock);
1928}
1929
1930static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1931{
1932 if (dev->get_stats) {
1933 struct net_device_stats *stats = dev->get_stats(dev);
1934
1935 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1936 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1937 dev->name, stats->rx_bytes, stats->rx_packets,
1938 stats->rx_errors,
1939 stats->rx_dropped + stats->rx_missed_errors,
1940 stats->rx_fifo_errors,
1941 stats->rx_length_errors + stats->rx_over_errors +
1942 stats->rx_crc_errors + stats->rx_frame_errors,
1943 stats->rx_compressed, stats->multicast,
1944 stats->tx_bytes, stats->tx_packets,
1945 stats->tx_errors, stats->tx_dropped,
1946 stats->tx_fifo_errors, stats->collisions,
1947 stats->tx_carrier_errors +
1948 stats->tx_aborted_errors +
1949 stats->tx_window_errors +
1950 stats->tx_heartbeat_errors,
1951 stats->tx_compressed);
1952 } else
1953 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1954}
1955
1956/*
1957 * Called from the PROCfs module. This now uses the new arbitrary sized
1958 * /proc/net interface to create /proc/net/dev
1959 */
1960static int dev_seq_show(struct seq_file *seq, void *v)
1961{
1962 if (v == SEQ_START_TOKEN)
1963 seq_puts(seq, "Inter-| Receive "
1964 " | Transmit\n"
1965 " face |bytes packets errs drop fifo frame "
1966 "compressed multicast|bytes packets errs "
1967 "drop fifo colls carrier compressed\n");
1968 else
1969 dev_seq_printf_stats(seq, v);
1970 return 0;
1971}
1972
1973static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1974{
1975 struct netif_rx_stats *rc = NULL;
1976
1977 while (*pos < NR_CPUS)
1978 if (cpu_online(*pos)) {
1979 rc = &per_cpu(netdev_rx_stat, *pos);
1980 break;
1981 } else
1982 ++*pos;
1983 return rc;
1984}
1985
1986static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1987{
1988 return softnet_get_online(pos);
1989}
1990
1991static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1992{
1993 ++*pos;
1994 return softnet_get_online(pos);
1995}
1996
1997static void softnet_seq_stop(struct seq_file *seq, void *v)
1998{
1999}
2000
2001static int softnet_seq_show(struct seq_file *seq, void *v)
2002{
2003 struct netif_rx_stats *s = v;
2004
2005 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
31aa02c5 2006 s->total, s->dropped, s->time_squeeze, 0,
c1ebcdb8
SH
2007 0, 0, 0, 0, /* was fastroute */
2008 s->cpu_collision );
1da177e4
LT
2009 return 0;
2010}
2011
2012static struct seq_operations dev_seq_ops = {
2013 .start = dev_seq_start,
2014 .next = dev_seq_next,
2015 .stop = dev_seq_stop,
2016 .show = dev_seq_show,
2017};
2018
2019static int dev_seq_open(struct inode *inode, struct file *file)
2020{
2021 return seq_open(file, &dev_seq_ops);
2022}
2023
2024static struct file_operations dev_seq_fops = {
2025 .owner = THIS_MODULE,
2026 .open = dev_seq_open,
2027 .read = seq_read,
2028 .llseek = seq_lseek,
2029 .release = seq_release,
2030};
2031
2032static struct seq_operations softnet_seq_ops = {
2033 .start = softnet_seq_start,
2034 .next = softnet_seq_next,
2035 .stop = softnet_seq_stop,
2036 .show = softnet_seq_show,
2037};
2038
2039static int softnet_seq_open(struct inode *inode, struct file *file)
2040{
2041 return seq_open(file, &softnet_seq_ops);
2042}
2043
2044static struct file_operations softnet_seq_fops = {
2045 .owner = THIS_MODULE,
2046 .open = softnet_seq_open,
2047 .read = seq_read,
2048 .llseek = seq_lseek,
2049 .release = seq_release,
2050};
2051
d86b5e0e 2052#ifdef CONFIG_WIRELESS_EXT
1da177e4
LT
2053extern int wireless_proc_init(void);
2054#else
2055#define wireless_proc_init() 0
2056#endif
2057
2058static int __init dev_proc_init(void)
2059{
2060 int rc = -ENOMEM;
2061
2062 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2063 goto out;
2064 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2065 goto out_dev;
2066 if (wireless_proc_init())
2067 goto out_softnet;
2068 rc = 0;
2069out:
2070 return rc;
2071out_softnet:
2072 proc_net_remove("softnet_stat");
2073out_dev:
2074 proc_net_remove("dev");
2075 goto out;
2076}
2077#else
2078#define dev_proc_init() 0
2079#endif /* CONFIG_PROC_FS */
2080
2081
2082/**
2083 * netdev_set_master - set up master/slave pair
2084 * @slave: slave device
2085 * @master: new master device
2086 *
2087 * Changes the master device of the slave. Pass %NULL to break the
2088 * bonding. The caller must hold the RTNL semaphore. On a failure
2089 * a negative errno code is returned. On success the reference counts
2090 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2091 * function returns zero.
2092 */
2093int netdev_set_master(struct net_device *slave, struct net_device *master)
2094{
2095 struct net_device *old = slave->master;
2096
2097 ASSERT_RTNL();
2098
2099 if (master) {
2100 if (old)
2101 return -EBUSY;
2102 dev_hold(master);
2103 }
2104
2105 slave->master = master;
2106
2107 synchronize_net();
2108
2109 if (old)
2110 dev_put(old);
2111
2112 if (master)
2113 slave->flags |= IFF_SLAVE;
2114 else
2115 slave->flags &= ~IFF_SLAVE;
2116
2117 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2118 return 0;
2119}
2120
2121/**
2122 * dev_set_promiscuity - update promiscuity count on a device
2123 * @dev: device
2124 * @inc: modifier
2125 *
2126 * Add or remove promsicuity from a device. While the count in the device
2127 * remains above zero the interface remains promiscuous. Once it hits zero
2128 * the device reverts back to normal filtering operation. A negative inc
2129 * value is used to drop promiscuity on the device.
2130 */
2131void dev_set_promiscuity(struct net_device *dev, int inc)
2132{
2133 unsigned short old_flags = dev->flags;
2134
1da177e4
LT
2135 if ((dev->promiscuity += inc) == 0)
2136 dev->flags &= ~IFF_PROMISC;
52609c0b
DC
2137 else
2138 dev->flags |= IFF_PROMISC;
2139 if (dev->flags != old_flags) {
1da177e4
LT
2140 dev_mc_upload(dev);
2141 printk(KERN_INFO "device %s %s promiscuous mode\n",
2142 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2143 "left");
2144 }
2145}
2146
2147/**
2148 * dev_set_allmulti - update allmulti count on a device
2149 * @dev: device
2150 * @inc: modifier
2151 *
2152 * Add or remove reception of all multicast frames to a device. While the
2153 * count in the device remains above zero the interface remains listening
2154 * to all interfaces. Once it hits zero the device reverts back to normal
2155 * filtering operation. A negative @inc value is used to drop the counter
2156 * when releasing a resource needing all multicasts.
2157 */
2158
2159void dev_set_allmulti(struct net_device *dev, int inc)
2160{
2161 unsigned short old_flags = dev->flags;
2162
2163 dev->flags |= IFF_ALLMULTI;
2164 if ((dev->allmulti += inc) == 0)
2165 dev->flags &= ~IFF_ALLMULTI;
2166 if (dev->flags ^ old_flags)
2167 dev_mc_upload(dev);
2168}
2169
2170unsigned dev_get_flags(const struct net_device *dev)
2171{
2172 unsigned flags;
2173
2174 flags = (dev->flags & ~(IFF_PROMISC |
2175 IFF_ALLMULTI |
b00055aa
SR
2176 IFF_RUNNING |
2177 IFF_LOWER_UP |
2178 IFF_DORMANT)) |
1da177e4
LT
2179 (dev->gflags & (IFF_PROMISC |
2180 IFF_ALLMULTI));
2181
b00055aa
SR
2182 if (netif_running(dev)) {
2183 if (netif_oper_up(dev))
2184 flags |= IFF_RUNNING;
2185 if (netif_carrier_ok(dev))
2186 flags |= IFF_LOWER_UP;
2187 if (netif_dormant(dev))
2188 flags |= IFF_DORMANT;
2189 }
1da177e4
LT
2190
2191 return flags;
2192}
2193
2194int dev_change_flags(struct net_device *dev, unsigned flags)
2195{
2196 int ret;
2197 int old_flags = dev->flags;
2198
2199 /*
2200 * Set the flags on our device.
2201 */
2202
2203 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2204 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2205 IFF_AUTOMEDIA)) |
2206 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2207 IFF_ALLMULTI));
2208
2209 /*
2210 * Load in the correct multicast list now the flags have changed.
2211 */
2212
2213 dev_mc_upload(dev);
2214
2215 /*
2216 * Have we downed the interface. We handle IFF_UP ourselves
2217 * according to user attempts to set it, rather than blindly
2218 * setting it.
2219 */
2220
2221 ret = 0;
2222 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2223 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2224
2225 if (!ret)
2226 dev_mc_upload(dev);
2227 }
2228
2229 if (dev->flags & IFF_UP &&
2230 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2231 IFF_VOLATILE)))
2232 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2233
2234 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2235 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2236 dev->gflags ^= IFF_PROMISC;
2237 dev_set_promiscuity(dev, inc);
2238 }
2239
2240 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2241 is important. Some (broken) drivers set IFF_PROMISC, when
2242 IFF_ALLMULTI is requested not asking us and not reporting.
2243 */
2244 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2245 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2246 dev->gflags ^= IFF_ALLMULTI;
2247 dev_set_allmulti(dev, inc);
2248 }
2249
2250 if (old_flags ^ dev->flags)
2251 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2252
2253 return ret;
2254}
2255
2256int dev_set_mtu(struct net_device *dev, int new_mtu)
2257{
2258 int err;
2259
2260 if (new_mtu == dev->mtu)
2261 return 0;
2262
2263 /* MTU must be positive. */
2264 if (new_mtu < 0)
2265 return -EINVAL;
2266
2267 if (!netif_device_present(dev))
2268 return -ENODEV;
2269
2270 err = 0;
2271 if (dev->change_mtu)
2272 err = dev->change_mtu(dev, new_mtu);
2273 else
2274 dev->mtu = new_mtu;
2275 if (!err && dev->flags & IFF_UP)
2276 notifier_call_chain(&netdev_chain,
2277 NETDEV_CHANGEMTU, dev);
2278 return err;
2279}
2280
2281int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2282{
2283 int err;
2284
2285 if (!dev->set_mac_address)
2286 return -EOPNOTSUPP;
2287 if (sa->sa_family != dev->type)
2288 return -EINVAL;
2289 if (!netif_device_present(dev))
2290 return -ENODEV;
2291 err = dev->set_mac_address(dev, sa);
2292 if (!err)
2293 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2294 return err;
2295}
2296
2297/*
2298 * Perform the SIOCxIFxxx calls.
2299 */
2300static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2301{
2302 int err;
2303 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2304
2305 if (!dev)
2306 return -ENODEV;
2307
2308 switch (cmd) {
2309 case SIOCGIFFLAGS: /* Get interface flags */
2310 ifr->ifr_flags = dev_get_flags(dev);
2311 return 0;
2312
2313 case SIOCSIFFLAGS: /* Set interface flags */
2314 return dev_change_flags(dev, ifr->ifr_flags);
2315
2316 case SIOCGIFMETRIC: /* Get the metric on the interface
2317 (currently unused) */
2318 ifr->ifr_metric = 0;
2319 return 0;
2320
2321 case SIOCSIFMETRIC: /* Set the metric on the interface
2322 (currently unused) */
2323 return -EOPNOTSUPP;
2324
2325 case SIOCGIFMTU: /* Get the MTU of a device */
2326 ifr->ifr_mtu = dev->mtu;
2327 return 0;
2328
2329 case SIOCSIFMTU: /* Set the MTU of a device */
2330 return dev_set_mtu(dev, ifr->ifr_mtu);
2331
2332 case SIOCGIFHWADDR:
2333 if (!dev->addr_len)
2334 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2335 else
2336 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2337 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2338 ifr->ifr_hwaddr.sa_family = dev->type;
2339 return 0;
2340
2341 case SIOCSIFHWADDR:
2342 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2343
2344 case SIOCSIFHWBROADCAST:
2345 if (ifr->ifr_hwaddr.sa_family != dev->type)
2346 return -EINVAL;
2347 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2348 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2349 notifier_call_chain(&netdev_chain,
2350 NETDEV_CHANGEADDR, dev);
2351 return 0;
2352
2353 case SIOCGIFMAP:
2354 ifr->ifr_map.mem_start = dev->mem_start;
2355 ifr->ifr_map.mem_end = dev->mem_end;
2356 ifr->ifr_map.base_addr = dev->base_addr;
2357 ifr->ifr_map.irq = dev->irq;
2358 ifr->ifr_map.dma = dev->dma;
2359 ifr->ifr_map.port = dev->if_port;
2360 return 0;
2361
2362 case SIOCSIFMAP:
2363 if (dev->set_config) {
2364 if (!netif_device_present(dev))
2365 return -ENODEV;
2366 return dev->set_config(dev, &ifr->ifr_map);
2367 }
2368 return -EOPNOTSUPP;
2369
2370 case SIOCADDMULTI:
2371 if (!dev->set_multicast_list ||
2372 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2373 return -EINVAL;
2374 if (!netif_device_present(dev))
2375 return -ENODEV;
2376 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2377 dev->addr_len, 1);
2378
2379 case SIOCDELMULTI:
2380 if (!dev->set_multicast_list ||
2381 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2382 return -EINVAL;
2383 if (!netif_device_present(dev))
2384 return -ENODEV;
2385 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2386 dev->addr_len, 1);
2387
2388 case SIOCGIFINDEX:
2389 ifr->ifr_ifindex = dev->ifindex;
2390 return 0;
2391
2392 case SIOCGIFTXQLEN:
2393 ifr->ifr_qlen = dev->tx_queue_len;
2394 return 0;
2395
2396 case SIOCSIFTXQLEN:
2397 if (ifr->ifr_qlen < 0)
2398 return -EINVAL;
2399 dev->tx_queue_len = ifr->ifr_qlen;
2400 return 0;
2401
2402 case SIOCSIFNAME:
2403 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2404 return dev_change_name(dev, ifr->ifr_newname);
2405
2406 /*
2407 * Unknown or private ioctl
2408 */
2409
2410 default:
2411 if ((cmd >= SIOCDEVPRIVATE &&
2412 cmd <= SIOCDEVPRIVATE + 15) ||
2413 cmd == SIOCBONDENSLAVE ||
2414 cmd == SIOCBONDRELEASE ||
2415 cmd == SIOCBONDSETHWADDR ||
2416 cmd == SIOCBONDSLAVEINFOQUERY ||
2417 cmd == SIOCBONDINFOQUERY ||
2418 cmd == SIOCBONDCHANGEACTIVE ||
2419 cmd == SIOCGMIIPHY ||
2420 cmd == SIOCGMIIREG ||
2421 cmd == SIOCSMIIREG ||
2422 cmd == SIOCBRADDIF ||
2423 cmd == SIOCBRDELIF ||
2424 cmd == SIOCWANDEV) {
2425 err = -EOPNOTSUPP;
2426 if (dev->do_ioctl) {
2427 if (netif_device_present(dev))
2428 err = dev->do_ioctl(dev, ifr,
2429 cmd);
2430 else
2431 err = -ENODEV;
2432 }
2433 } else
2434 err = -EINVAL;
2435
2436 }
2437 return err;
2438}
2439
2440/*
2441 * This function handles all "interface"-type I/O control requests. The actual
2442 * 'doing' part of this is dev_ifsioc above.
2443 */
2444
2445/**
2446 * dev_ioctl - network device ioctl
2447 * @cmd: command to issue
2448 * @arg: pointer to a struct ifreq in user space
2449 *
2450 * Issue ioctl functions to devices. This is normally called by the
2451 * user space syscall interfaces but can sometimes be useful for
2452 * other purposes. The return value is the return from the syscall if
2453 * positive or a negative errno code on error.
2454 */
2455
2456int dev_ioctl(unsigned int cmd, void __user *arg)
2457{
2458 struct ifreq ifr;
2459 int ret;
2460 char *colon;
2461
2462 /* One special case: SIOCGIFCONF takes ifconf argument
2463 and requires shared lock, because it sleeps writing
2464 to user space.
2465 */
2466
2467 if (cmd == SIOCGIFCONF) {
6756ae4b 2468 rtnl_lock();
1da177e4 2469 ret = dev_ifconf((char __user *) arg);
6756ae4b 2470 rtnl_unlock();
1da177e4
LT
2471 return ret;
2472 }
2473 if (cmd == SIOCGIFNAME)
2474 return dev_ifname((struct ifreq __user *)arg);
2475
2476 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2477 return -EFAULT;
2478
2479 ifr.ifr_name[IFNAMSIZ-1] = 0;
2480
2481 colon = strchr(ifr.ifr_name, ':');
2482 if (colon)
2483 *colon = 0;
2484
2485 /*
2486 * See which interface the caller is talking about.
2487 */
2488
2489 switch (cmd) {
2490 /*
2491 * These ioctl calls:
2492 * - can be done by all.
2493 * - atomic and do not require locking.
2494 * - return a value
2495 */
2496 case SIOCGIFFLAGS:
2497 case SIOCGIFMETRIC:
2498 case SIOCGIFMTU:
2499 case SIOCGIFHWADDR:
2500 case SIOCGIFSLAVE:
2501 case SIOCGIFMAP:
2502 case SIOCGIFINDEX:
2503 case SIOCGIFTXQLEN:
2504 dev_load(ifr.ifr_name);
2505 read_lock(&dev_base_lock);
2506 ret = dev_ifsioc(&ifr, cmd);
2507 read_unlock(&dev_base_lock);
2508 if (!ret) {
2509 if (colon)
2510 *colon = ':';
2511 if (copy_to_user(arg, &ifr,
2512 sizeof(struct ifreq)))
2513 ret = -EFAULT;
2514 }
2515 return ret;
2516
2517 case SIOCETHTOOL:
2518 dev_load(ifr.ifr_name);
2519 rtnl_lock();
2520 ret = dev_ethtool(&ifr);
2521 rtnl_unlock();
2522 if (!ret) {
2523 if (colon)
2524 *colon = ':';
2525 if (copy_to_user(arg, &ifr,
2526 sizeof(struct ifreq)))
2527 ret = -EFAULT;
2528 }
2529 return ret;
2530
2531 /*
2532 * These ioctl calls:
2533 * - require superuser power.
2534 * - require strict serialization.
2535 * - return a value
2536 */
2537 case SIOCGMIIPHY:
2538 case SIOCGMIIREG:
2539 case SIOCSIFNAME:
2540 if (!capable(CAP_NET_ADMIN))
2541 return -EPERM;
2542 dev_load(ifr.ifr_name);
2543 rtnl_lock();
2544 ret = dev_ifsioc(&ifr, cmd);
2545 rtnl_unlock();
2546 if (!ret) {
2547 if (colon)
2548 *colon = ':';
2549 if (copy_to_user(arg, &ifr,
2550 sizeof(struct ifreq)))
2551 ret = -EFAULT;
2552 }
2553 return ret;
2554
2555 /*
2556 * These ioctl calls:
2557 * - require superuser power.
2558 * - require strict serialization.
2559 * - do not return a value
2560 */
2561 case SIOCSIFFLAGS:
2562 case SIOCSIFMETRIC:
2563 case SIOCSIFMTU:
2564 case SIOCSIFMAP:
2565 case SIOCSIFHWADDR:
2566 case SIOCSIFSLAVE:
2567 case SIOCADDMULTI:
2568 case SIOCDELMULTI:
2569 case SIOCSIFHWBROADCAST:
2570 case SIOCSIFTXQLEN:
2571 case SIOCSMIIREG:
2572 case SIOCBONDENSLAVE:
2573 case SIOCBONDRELEASE:
2574 case SIOCBONDSETHWADDR:
1da177e4
LT
2575 case SIOCBONDCHANGEACTIVE:
2576 case SIOCBRADDIF:
2577 case SIOCBRDELIF:
2578 if (!capable(CAP_NET_ADMIN))
2579 return -EPERM;
cabcac0b
TG
2580 /* fall through */
2581 case SIOCBONDSLAVEINFOQUERY:
2582 case SIOCBONDINFOQUERY:
1da177e4
LT
2583 dev_load(ifr.ifr_name);
2584 rtnl_lock();
2585 ret = dev_ifsioc(&ifr, cmd);
2586 rtnl_unlock();
2587 return ret;
2588
2589 case SIOCGIFMEM:
2590 /* Get the per device memory space. We can add this but
2591 * currently do not support it */
2592 case SIOCSIFMEM:
2593 /* Set the per device memory buffer space.
2594 * Not applicable in our case */
2595 case SIOCSIFLINK:
2596 return -EINVAL;
2597
2598 /*
2599 * Unknown or private ioctl.
2600 */
2601 default:
2602 if (cmd == SIOCWANDEV ||
2603 (cmd >= SIOCDEVPRIVATE &&
2604 cmd <= SIOCDEVPRIVATE + 15)) {
2605 dev_load(ifr.ifr_name);
2606 rtnl_lock();
2607 ret = dev_ifsioc(&ifr, cmd);
2608 rtnl_unlock();
2609 if (!ret && copy_to_user(arg, &ifr,
2610 sizeof(struct ifreq)))
2611 ret = -EFAULT;
2612 return ret;
2613 }
d86b5e0e 2614#ifdef CONFIG_WIRELESS_EXT
1da177e4
LT
2615 /* Take care of Wireless Extensions */
2616 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2617 /* If command is `set a parameter', or
2618 * `get the encoding parameters', check if
2619 * the user has the right to do it */
2620 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2621 if (!capable(CAP_NET_ADMIN))
2622 return -EPERM;
2623 }
2624 dev_load(ifr.ifr_name);
2625 rtnl_lock();
2626 /* Follow me in net/core/wireless.c */
2627 ret = wireless_process_ioctl(&ifr, cmd);
2628 rtnl_unlock();
2629 if (IW_IS_GET(cmd) &&
2630 copy_to_user(arg, &ifr,
2631 sizeof(struct ifreq)))
2632 ret = -EFAULT;
2633 return ret;
2634 }
d86b5e0e 2635#endif /* CONFIG_WIRELESS_EXT */
1da177e4
LT
2636 return -EINVAL;
2637 }
2638}
2639
2640
2641/**
2642 * dev_new_index - allocate an ifindex
2643 *
2644 * Returns a suitable unique value for a new device interface
2645 * number. The caller must hold the rtnl semaphore or the
2646 * dev_base_lock to be sure it remains unique.
2647 */
2648static int dev_new_index(void)
2649{
2650 static int ifindex;
2651 for (;;) {
2652 if (++ifindex <= 0)
2653 ifindex = 1;
2654 if (!__dev_get_by_index(ifindex))
2655 return ifindex;
2656 }
2657}
2658
2659static int dev_boot_phase = 1;
2660
2661/* Delayed registration/unregisteration */
2662static DEFINE_SPINLOCK(net_todo_list_lock);
2663static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2664
2665static inline void net_set_todo(struct net_device *dev)
2666{
2667 spin_lock(&net_todo_list_lock);
2668 list_add_tail(&dev->todo_list, &net_todo_list);
2669 spin_unlock(&net_todo_list_lock);
2670}
2671
2672/**
2673 * register_netdevice - register a network device
2674 * @dev: device to register
2675 *
2676 * Take a completed network device structure and add it to the kernel
2677 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2678 * chain. 0 is returned on success. A negative errno code is returned
2679 * on a failure to set up the device, or if the name is a duplicate.
2680 *
2681 * Callers must hold the rtnl semaphore. You may want
2682 * register_netdev() instead of this.
2683 *
2684 * BUGS:
2685 * The locking appears insufficient to guarantee two parallel registers
2686 * will not get the same name.
2687 */
2688
2689int register_netdevice(struct net_device *dev)
2690{
2691 struct hlist_head *head;
2692 struct hlist_node *p;
2693 int ret;
2694
2695 BUG_ON(dev_boot_phase);
2696 ASSERT_RTNL();
2697
2698 /* When net_device's are persistent, this will be fatal. */
2699 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2700
2701 spin_lock_init(&dev->queue_lock);
2702 spin_lock_init(&dev->xmit_lock);
2703 dev->xmit_lock_owner = -1;
2704#ifdef CONFIG_NET_CLS_ACT
2705 spin_lock_init(&dev->ingress_lock);
2706#endif
2707
2708 ret = alloc_divert_blk(dev);
2709 if (ret)
2710 goto out;
2711
2712 dev->iflink = -1;
2713
2714 /* Init, if this function is available */
2715 if (dev->init) {
2716 ret = dev->init(dev);
2717 if (ret) {
2718 if (ret > 0)
2719 ret = -EIO;
2720 goto out_err;
2721 }
2722 }
2723
2724 if (!dev_valid_name(dev->name)) {
2725 ret = -EINVAL;
2726 goto out_err;
2727 }
2728
2729 dev->ifindex = dev_new_index();
2730 if (dev->iflink == -1)
2731 dev->iflink = dev->ifindex;
2732
2733 /* Check for existence of name */
2734 head = dev_name_hash(dev->name);
2735 hlist_for_each(p, head) {
2736 struct net_device *d
2737 = hlist_entry(p, struct net_device, name_hlist);
2738 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2739 ret = -EEXIST;
2740 goto out_err;
2741 }
2742 }
2743
2744 /* Fix illegal SG+CSUM combinations. */
2745 if ((dev->features & NETIF_F_SG) &&
2746 !(dev->features & (NETIF_F_IP_CSUM |
2747 NETIF_F_NO_CSUM |
2748 NETIF_F_HW_CSUM))) {
2749 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2750 dev->name);
2751 dev->features &= ~NETIF_F_SG;
2752 }
2753
2754 /* TSO requires that SG is present as well. */
2755 if ((dev->features & NETIF_F_TSO) &&
2756 !(dev->features & NETIF_F_SG)) {
2757 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2758 dev->name);
2759 dev->features &= ~NETIF_F_TSO;
2760 }
e89e9cf5
AR
2761 if (dev->features & NETIF_F_UFO) {
2762 if (!(dev->features & NETIF_F_HW_CSUM)) {
2763 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2764 "NETIF_F_HW_CSUM feature.\n",
2765 dev->name);
2766 dev->features &= ~NETIF_F_UFO;
2767 }
2768 if (!(dev->features & NETIF_F_SG)) {
2769 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2770 "NETIF_F_SG feature.\n",
2771 dev->name);
2772 dev->features &= ~NETIF_F_UFO;
2773 }
2774 }
1da177e4
LT
2775
2776 /*
2777 * nil rebuild_header routine,
2778 * that should be never called and used as just bug trap.
2779 */
2780
2781 if (!dev->rebuild_header)
2782 dev->rebuild_header = default_rebuild_header;
2783
2784 /*
2785 * Default initial state at registry is that the
2786 * device is present.
2787 */
2788
2789 set_bit(__LINK_STATE_PRESENT, &dev->state);
2790
2791 dev->next = NULL;
2792 dev_init_scheduler(dev);
2793 write_lock_bh(&dev_base_lock);
2794 *dev_tail = dev;
2795 dev_tail = &dev->next;
2796 hlist_add_head(&dev->name_hlist, head);
2797 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2798 dev_hold(dev);
2799 dev->reg_state = NETREG_REGISTERING;
2800 write_unlock_bh(&dev_base_lock);
2801
2802 /* Notify protocols, that a new device appeared. */
2803 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2804
2805 /* Finish registration after unlock */
2806 net_set_todo(dev);
2807 ret = 0;
2808
2809out:
2810 return ret;
2811out_err:
2812 free_divert_blk(dev);
2813 goto out;
2814}
2815
2816/**
2817 * register_netdev - register a network device
2818 * @dev: device to register
2819 *
2820 * Take a completed network device structure and add it to the kernel
2821 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2822 * chain. 0 is returned on success. A negative errno code is returned
2823 * on a failure to set up the device, or if the name is a duplicate.
2824 *
2825 * This is a wrapper around register_netdev that takes the rtnl semaphore
2826 * and expands the device name if you passed a format string to
2827 * alloc_netdev.
2828 */
2829int register_netdev(struct net_device *dev)
2830{
2831 int err;
2832
2833 rtnl_lock();
2834
2835 /*
2836 * If the name is a format string the caller wants us to do a
2837 * name allocation.
2838 */
2839 if (strchr(dev->name, '%')) {
2840 err = dev_alloc_name(dev, dev->name);
2841 if (err < 0)
2842 goto out;
2843 }
2844
2845 /*
2846 * Back compatibility hook. Kill this one in 2.5
2847 */
2848 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2849 err = dev_alloc_name(dev, "eth%d");
2850 if (err < 0)
2851 goto out;
2852 }
2853
2854 err = register_netdevice(dev);
2855out:
2856 rtnl_unlock();
2857 return err;
2858}
2859EXPORT_SYMBOL(register_netdev);
2860
2861/*
2862 * netdev_wait_allrefs - wait until all references are gone.
2863 *
2864 * This is called when unregistering network devices.
2865 *
2866 * Any protocol or device that holds a reference should register
2867 * for netdevice notification, and cleanup and put back the
2868 * reference if they receive an UNREGISTER event.
2869 * We can get stuck here if buggy protocols don't correctly
2870 * call dev_put.
2871 */
2872static void netdev_wait_allrefs(struct net_device *dev)
2873{
2874 unsigned long rebroadcast_time, warning_time;
2875
2876 rebroadcast_time = warning_time = jiffies;
2877 while (atomic_read(&dev->refcnt) != 0) {
2878 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6756ae4b 2879 rtnl_lock();
1da177e4
LT
2880
2881 /* Rebroadcast unregister notification */
2882 notifier_call_chain(&netdev_chain,
2883 NETDEV_UNREGISTER, dev);
2884
2885 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2886 &dev->state)) {
2887 /* We must not have linkwatch events
2888 * pending on unregister. If this
2889 * happens, we simply run the queue
2890 * unscheduled, resulting in a noop
2891 * for this device.
2892 */
2893 linkwatch_run_queue();
2894 }
2895
6756ae4b 2896 __rtnl_unlock();
1da177e4
LT
2897
2898 rebroadcast_time = jiffies;
2899 }
2900
2901 msleep(250);
2902
2903 if (time_after(jiffies, warning_time + 10 * HZ)) {
2904 printk(KERN_EMERG "unregister_netdevice: "
2905 "waiting for %s to become free. Usage "
2906 "count = %d\n",
2907 dev->name, atomic_read(&dev->refcnt));
2908 warning_time = jiffies;
2909 }
2910 }
2911}
2912
2913/* The sequence is:
2914 *
2915 * rtnl_lock();
2916 * ...
2917 * register_netdevice(x1);
2918 * register_netdevice(x2);
2919 * ...
2920 * unregister_netdevice(y1);
2921 * unregister_netdevice(y2);
2922 * ...
2923 * rtnl_unlock();
2924 * free_netdev(y1);
2925 * free_netdev(y2);
2926 *
2927 * We are invoked by rtnl_unlock() after it drops the semaphore.
2928 * This allows us to deal with problems:
2929 * 1) We can create/delete sysfs objects which invoke hotplug
2930 * without deadlocking with linkwatch via keventd.
2931 * 2) Since we run with the RTNL semaphore not held, we can sleep
2932 * safely in order to wait for the netdev refcnt to drop to zero.
2933 */
2934static DECLARE_MUTEX(net_todo_run_mutex);
2935void netdev_run_todo(void)
2936{
2937 struct list_head list = LIST_HEAD_INIT(list);
2938 int err;
2939
2940
2941 /* Need to guard against multiple cpu's getting out of order. */
2942 down(&net_todo_run_mutex);
2943
2944 /* Not safe to do outside the semaphore. We must not return
2945 * until all unregister events invoked by the local processor
2946 * have been completed (either by this todo run, or one on
2947 * another cpu).
2948 */
2949 if (list_empty(&net_todo_list))
2950 goto out;
2951
2952 /* Snapshot list, allow later requests */
2953 spin_lock(&net_todo_list_lock);
2954 list_splice_init(&net_todo_list, &list);
2955 spin_unlock(&net_todo_list_lock);
2956
2957 while (!list_empty(&list)) {
2958 struct net_device *dev
2959 = list_entry(list.next, struct net_device, todo_list);
2960 list_del(&dev->todo_list);
2961
2962 switch(dev->reg_state) {
2963 case NETREG_REGISTERING:
2964 err = netdev_register_sysfs(dev);
2965 if (err)
2966 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2967 dev->name, err);
2968 dev->reg_state = NETREG_REGISTERED;
2969 break;
2970
2971 case NETREG_UNREGISTERING:
2972 netdev_unregister_sysfs(dev);
2973 dev->reg_state = NETREG_UNREGISTERED;
2974
2975 netdev_wait_allrefs(dev);
2976
2977 /* paranoia */
2978 BUG_ON(atomic_read(&dev->refcnt));
2979 BUG_TRAP(!dev->ip_ptr);
2980 BUG_TRAP(!dev->ip6_ptr);
2981 BUG_TRAP(!dev->dn_ptr);
2982
2983
2984 /* It must be the very last action,
2985 * after this 'dev' may point to freed up memory.
2986 */
2987 if (dev->destructor)
2988 dev->destructor(dev);
2989 break;
2990
2991 default:
2992 printk(KERN_ERR "network todo '%s' but state %d\n",
2993 dev->name, dev->reg_state);
2994 break;
2995 }
2996 }
2997
2998out:
2999 up(&net_todo_run_mutex);
3000}
3001
3002/**
3003 * alloc_netdev - allocate network device
3004 * @sizeof_priv: size of private data to allocate space for
3005 * @name: device name format string
3006 * @setup: callback to initialize device
3007 *
3008 * Allocates a struct net_device with private data area for driver use
3009 * and performs basic initialization.
3010 */
3011struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3012 void (*setup)(struct net_device *))
3013{
3014 void *p;
3015 struct net_device *dev;
3016 int alloc_size;
3017
3018 /* ensure 32-byte alignment of both the device and private area */
3019 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3020 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3021
3022 p = kmalloc(alloc_size, GFP_KERNEL);
3023 if (!p) {
3024 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3025 return NULL;
3026 }
3027 memset(p, 0, alloc_size);
3028
3029 dev = (struct net_device *)
3030 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3031 dev->padded = (char *)dev - (char *)p;
3032
3033 if (sizeof_priv)
3034 dev->priv = netdev_priv(dev);
3035
3036 setup(dev);
3037 strcpy(dev->name, name);
3038 return dev;
3039}
3040EXPORT_SYMBOL(alloc_netdev);
3041
3042/**
3043 * free_netdev - free network device
3044 * @dev: device
3045 *
3046 * This function does the last stage of destroying an allocated device
3047 * interface. The reference to the device object is released.
3048 * If this is the last reference then it will be freed.
3049 */
3050void free_netdev(struct net_device *dev)
3051{
3052#ifdef CONFIG_SYSFS
3053 /* Compatiablity with error handling in drivers */
3054 if (dev->reg_state == NETREG_UNINITIALIZED) {
3055 kfree((char *)dev - dev->padded);
3056 return;
3057 }
3058
3059 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3060 dev->reg_state = NETREG_RELEASED;
3061
3062 /* will free via class release */
3063 class_device_put(&dev->class_dev);
3064#else
3065 kfree((char *)dev - dev->padded);
3066#endif
3067}
3068
3069/* Synchronize with packet receive processing. */
3070void synchronize_net(void)
3071{
3072 might_sleep();
fbd568a3 3073 synchronize_rcu();
1da177e4
LT
3074}
3075
3076/**
3077 * unregister_netdevice - remove device from the kernel
3078 * @dev: device
3079 *
3080 * This function shuts down a device interface and removes it
3081 * from the kernel tables. On success 0 is returned, on a failure
3082 * a negative errno code is returned.
3083 *
3084 * Callers must hold the rtnl semaphore. You may want
3085 * unregister_netdev() instead of this.
3086 */
3087
3088int unregister_netdevice(struct net_device *dev)
3089{
3090 struct net_device *d, **dp;
3091
3092 BUG_ON(dev_boot_phase);
3093 ASSERT_RTNL();
3094
3095 /* Some devices call without registering for initialization unwind. */
3096 if (dev->reg_state == NETREG_UNINITIALIZED) {
3097 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3098 "was registered\n", dev->name, dev);
3099 return -ENODEV;
3100 }
3101
3102 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3103
3104 /* If device is running, close it first. */
3105 if (dev->flags & IFF_UP)
3106 dev_close(dev);
3107
3108 /* And unlink it from device chain. */
3109 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3110 if (d == dev) {
3111 write_lock_bh(&dev_base_lock);
3112 hlist_del(&dev->name_hlist);
3113 hlist_del(&dev->index_hlist);
3114 if (dev_tail == &dev->next)
3115 dev_tail = dp;
3116 *dp = d->next;
3117 write_unlock_bh(&dev_base_lock);
3118 break;
3119 }
3120 }
3121 if (!d) {
3122 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3123 dev->name);
3124 return -ENODEV;
3125 }
3126
3127 dev->reg_state = NETREG_UNREGISTERING;
3128
3129 synchronize_net();
3130
3131 /* Shutdown queueing discipline. */
3132 dev_shutdown(dev);
3133
3134
3135 /* Notify protocols, that we are about to destroy
3136 this device. They should clean all the things.
3137 */
3138 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3139
3140 /*
3141 * Flush the multicast chain
3142 */
3143 dev_mc_discard(dev);
3144
3145 if (dev->uninit)
3146 dev->uninit(dev);
3147
3148 /* Notifier chain MUST detach us from master device. */
3149 BUG_TRAP(!dev->master);
3150
3151 free_divert_blk(dev);
3152
3153 /* Finish processing unregister after unlock */
3154 net_set_todo(dev);
3155
3156 synchronize_net();
3157
3158 dev_put(dev);
3159 return 0;
3160}
3161
3162/**
3163 * unregister_netdev - remove device from the kernel
3164 * @dev: device
3165 *
3166 * This function shuts down a device interface and removes it
3167 * from the kernel tables. On success 0 is returned, on a failure
3168 * a negative errno code is returned.
3169 *
3170 * This is just a wrapper for unregister_netdevice that takes
3171 * the rtnl semaphore. In general you want to use this and not
3172 * unregister_netdevice.
3173 */
3174void unregister_netdev(struct net_device *dev)
3175{
3176 rtnl_lock();
3177 unregister_netdevice(dev);
3178 rtnl_unlock();
3179}
3180
3181EXPORT_SYMBOL(unregister_netdev);
3182
3183#ifdef CONFIG_HOTPLUG_CPU
3184static int dev_cpu_callback(struct notifier_block *nfb,
3185 unsigned long action,
3186 void *ocpu)
3187{
3188 struct sk_buff **list_skb;
3189 struct net_device **list_net;
3190 struct sk_buff *skb;
3191 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3192 struct softnet_data *sd, *oldsd;
3193
3194 if (action != CPU_DEAD)
3195 return NOTIFY_OK;
3196
3197 local_irq_disable();
3198 cpu = smp_processor_id();
3199 sd = &per_cpu(softnet_data, cpu);
3200 oldsd = &per_cpu(softnet_data, oldcpu);
3201
3202 /* Find end of our completion_queue. */
3203 list_skb = &sd->completion_queue;
3204 while (*list_skb)
3205 list_skb = &(*list_skb)->next;
3206 /* Append completion queue from offline CPU. */
3207 *list_skb = oldsd->completion_queue;
3208 oldsd->completion_queue = NULL;
3209
3210 /* Find end of our output_queue. */
3211 list_net = &sd->output_queue;
3212 while (*list_net)
3213 list_net = &(*list_net)->next_sched;
3214 /* Append output queue from offline CPU. */
3215 *list_net = oldsd->output_queue;
3216 oldsd->output_queue = NULL;
3217
3218 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3219 local_irq_enable();
3220
3221 /* Process offline CPU's input_pkt_queue */
3222 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3223 netif_rx(skb);
3224
3225 return NOTIFY_OK;
3226}
3227#endif /* CONFIG_HOTPLUG_CPU */
3228
3229
3230/*
3231 * Initialize the DEV module. At boot time this walks the device list and
3232 * unhooks any devices that fail to initialise (normally hardware not
3233 * present) and leaves us with a valid list of present and active devices.
3234 *
3235 */
3236
3237/*
3238 * This is called single threaded during boot, so no need
3239 * to take the rtnl semaphore.
3240 */
3241static int __init net_dev_init(void)
3242{
3243 int i, rc = -ENOMEM;
3244
3245 BUG_ON(!dev_boot_phase);
3246
3247 net_random_init();
3248
3249 if (dev_proc_init())
3250 goto out;
3251
3252 if (netdev_sysfs_init())
3253 goto out;
3254
3255 INIT_LIST_HEAD(&ptype_all);
3256 for (i = 0; i < 16; i++)
3257 INIT_LIST_HEAD(&ptype_base[i]);
3258
3259 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3260 INIT_HLIST_HEAD(&dev_name_head[i]);
3261
3262 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3263 INIT_HLIST_HEAD(&dev_index_head[i]);
3264
3265 /*
3266 * Initialise the packet receive queues.
3267 */
3268
88a2a4ac 3269 for_each_cpu(i) {
1da177e4
LT
3270 struct softnet_data *queue;
3271
3272 queue = &per_cpu(softnet_data, i);
3273 skb_queue_head_init(&queue->input_pkt_queue);
1da177e4
LT
3274 queue->completion_queue = NULL;
3275 INIT_LIST_HEAD(&queue->poll_list);
3276 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3277 queue->backlog_dev.weight = weight_p;
3278 queue->backlog_dev.poll = process_backlog;
3279 atomic_set(&queue->backlog_dev.refcnt, 1);
3280 }
3281
1da177e4
LT
3282 dev_boot_phase = 0;
3283
3284 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3285 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3286
3287 hotcpu_notifier(dev_cpu_callback, 0);
3288 dst_init();
3289 dev_mcast_init();
3290 rc = 0;
3291out:
3292 return rc;
3293}
3294
3295subsys_initcall(net_dev_init);
3296
3297EXPORT_SYMBOL(__dev_get_by_index);
3298EXPORT_SYMBOL(__dev_get_by_name);
3299EXPORT_SYMBOL(__dev_remove_pack);
3300EXPORT_SYMBOL(__skb_linearize);
c2373ee9 3301EXPORT_SYMBOL(dev_valid_name);
1da177e4
LT
3302EXPORT_SYMBOL(dev_add_pack);
3303EXPORT_SYMBOL(dev_alloc_name);
3304EXPORT_SYMBOL(dev_close);
3305EXPORT_SYMBOL(dev_get_by_flags);
3306EXPORT_SYMBOL(dev_get_by_index);
3307EXPORT_SYMBOL(dev_get_by_name);
1da177e4
LT
3308EXPORT_SYMBOL(dev_open);
3309EXPORT_SYMBOL(dev_queue_xmit);
3310EXPORT_SYMBOL(dev_remove_pack);
3311EXPORT_SYMBOL(dev_set_allmulti);
3312EXPORT_SYMBOL(dev_set_promiscuity);
3313EXPORT_SYMBOL(dev_change_flags);
3314EXPORT_SYMBOL(dev_set_mtu);
3315EXPORT_SYMBOL(dev_set_mac_address);
3316EXPORT_SYMBOL(free_netdev);
3317EXPORT_SYMBOL(netdev_boot_setup_check);
3318EXPORT_SYMBOL(netdev_set_master);
3319EXPORT_SYMBOL(netdev_state_change);
3320EXPORT_SYMBOL(netif_receive_skb);
3321EXPORT_SYMBOL(netif_rx);
3322EXPORT_SYMBOL(register_gifconf);
3323EXPORT_SYMBOL(register_netdevice);
3324EXPORT_SYMBOL(register_netdevice_notifier);
3325EXPORT_SYMBOL(skb_checksum_help);
3326EXPORT_SYMBOL(synchronize_net);
3327EXPORT_SYMBOL(unregister_netdevice);
3328EXPORT_SYMBOL(unregister_netdevice_notifier);
3329EXPORT_SYMBOL(net_enable_timestamp);
3330EXPORT_SYMBOL(net_disable_timestamp);
3331EXPORT_SYMBOL(dev_get_flags);
3332
3333#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3334EXPORT_SYMBOL(br_handle_frame_hook);
3335EXPORT_SYMBOL(br_fdb_get_hook);
3336EXPORT_SYMBOL(br_fdb_put_hook);
3337#endif
3338
3339#ifdef CONFIG_KMOD
3340EXPORT_SYMBOL(dev_load);
3341#endif
3342
3343EXPORT_PER_CPU_SYMBOL(softnet_data);