]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/core/dev.c
Linux-2.6.12-rc2
[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
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 *
14 * Additional Authors:
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
21 *
22 * Changes:
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
34 * drivers
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
44 * call a packet.
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
58 * 1 device.
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
66 * the backlog queue.
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
73 */
74
75#include <asm/uaccess.h>
76#include <asm/system.h>
77#include <linux/bitops.h>
78#include <linux/config.h>
79#include <linux/cpu.h>
80#include <linux/types.h>
81#include <linux/kernel.h>
82#include <linux/sched.h>
83#include <linux/string.h>
84#include <linux/mm.h>
85#include <linux/socket.h>
86#include <linux/sockios.h>
87#include <linux/errno.h>
88#include <linux/interrupt.h>
89#include <linux/if_ether.h>
90#include <linux/netdevice.h>
91#include <linux/etherdevice.h>
92#include <linux/notifier.h>
93#include <linux/skbuff.h>
94#include <net/sock.h>
95#include <linux/rtnetlink.h>
96#include <linux/proc_fs.h>
97#include <linux/seq_file.h>
98#include <linux/stat.h>
99#include <linux/if_bridge.h>
100#include <linux/divert.h>
101#include <net/dst.h>
102#include <net/pkt_sched.h>
103#include <net/checksum.h>
104#include <linux/highmem.h>
105#include <linux/init.h>
106#include <linux/kmod.h>
107#include <linux/module.h>
108#include <linux/kallsyms.h>
109#include <linux/netpoll.h>
110#include <linux/rcupdate.h>
111#include <linux/delay.h>
112#ifdef CONFIG_NET_RADIO
113#include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
114#include <net/iw_handler.h>
115#endif /* CONFIG_NET_RADIO */
116#include <asm/current.h>
117
118/* This define, if set, will randomly drop a packet when congestion
119 * is more than moderate. It helps fairness in the multi-interface
120 * case when one of them is a hog, but it kills performance for the
121 * single interface case so it is off now by default.
122 */
123#undef RAND_LIE
124
125/* Setting this will sample the queue lengths and thus congestion
126 * via a timer instead of as each packet is received.
127 */
128#undef OFFLINE_SAMPLE
129
130/*
131 * The list of packet types we will receive (as opposed to discard)
132 * and the routines to invoke.
133 *
134 * Why 16. Because with 16 the only overlap we get on a hash of the
135 * low nibble of the protocol value is RARP/SNAP/X.25.
136 *
137 * NOTE: That is no longer true with the addition of VLAN tags. Not
138 * sure which should go first, but I bet it won't make much
139 * difference if we are running VLANs. The good news is that
140 * this protocol won't be in the list unless compiled in, so
141 * the average user (w/out VLANs) will not be adversly affected.
142 * --BLG
143 *
144 * 0800 IP
145 * 8100 802.1Q VLAN
146 * 0001 802.3
147 * 0002 AX.25
148 * 0004 802.2
149 * 8035 RARP
150 * 0005 SNAP
151 * 0805 X.25
152 * 0806 ARP
153 * 8137 IPX
154 * 0009 Localtalk
155 * 86DD IPv6
156 */
157
158static DEFINE_SPINLOCK(ptype_lock);
159static struct list_head ptype_base[16]; /* 16 way hashed list */
160static struct list_head ptype_all; /* Taps */
161
162#ifdef OFFLINE_SAMPLE
163static void sample_queue(unsigned long dummy);
164static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
165#endif
166
167/*
168 * The @dev_base list is protected by @dev_base_lock and the rtln
169 * semaphore.
170 *
171 * Pure readers hold dev_base_lock for reading.
172 *
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
177 *
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
181 *
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
184 * semaphore held.
185 */
186struct net_device *dev_base;
187static struct net_device **dev_tail = &dev_base;
188DEFINE_RWLOCK(dev_base_lock);
189
190EXPORT_SYMBOL(dev_base);
191EXPORT_SYMBOL(dev_base_lock);
192
193#define NETDEV_HASHBITS 8
194static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
195static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
196
197static inline struct hlist_head *dev_name_hash(const char *name)
198{
199 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
200 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
201}
202
203static inline struct hlist_head *dev_index_hash(int ifindex)
204{
205 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
206}
207
208/*
209 * Our notifier list
210 */
211
212static struct notifier_block *netdev_chain;
213
214/*
215 * Device drivers call our routines to queue packets here. We empty the
216 * queue in the local softnet handler.
217 */
218DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
219
220#ifdef CONFIG_SYSFS
221extern int netdev_sysfs_init(void);
222extern int netdev_register_sysfs(struct net_device *);
223extern void netdev_unregister_sysfs(struct net_device *);
224#else
225#define netdev_sysfs_init() (0)
226#define netdev_register_sysfs(dev) (0)
227#define netdev_unregister_sysfs(dev) do { } while(0)
228#endif
229
230
231/*******************************************************************************
232
233 Protocol management and registration routines
234
235*******************************************************************************/
236
237/*
238 * For efficiency
239 */
240
241int netdev_nit;
242
243/*
244 * Add a protocol ID to the list. Now that the input handler is
245 * smarter we can dispense with all the messy stuff that used to be
246 * here.
247 *
248 * BEWARE!!! Protocol handlers, mangling input packets,
249 * MUST BE last in hash buckets and checking protocol handlers
250 * MUST start from promiscuous ptype_all chain in net_bh.
251 * It is true now, do not change it.
252 * Explanation follows: if protocol handler, mangling packet, will
253 * be the first on list, it is not able to sense, that packet
254 * is cloned and should be copied-on-write, so that it will
255 * change it and subsequent readers will get broken packet.
256 * --ANK (980803)
257 */
258
259/**
260 * dev_add_pack - add packet handler
261 * @pt: packet type declaration
262 *
263 * Add a protocol handler to the networking stack. The passed &packet_type
264 * is linked into kernel lists and may not be freed until it has been
265 * removed from the kernel lists.
266 *
267 * This call does not sleep therefore it can not
268 * guarantee all CPU's that are in middle of receiving packets
269 * will see the new packet type (until the next received packet).
270 */
271
272void dev_add_pack(struct packet_type *pt)
273{
274 int hash;
275
276 spin_lock_bh(&ptype_lock);
277 if (pt->type == htons(ETH_P_ALL)) {
278 netdev_nit++;
279 list_add_rcu(&pt->list, &ptype_all);
280 } else {
281 hash = ntohs(pt->type) & 15;
282 list_add_rcu(&pt->list, &ptype_base[hash]);
283 }
284 spin_unlock_bh(&ptype_lock);
285}
286
287extern void linkwatch_run_queue(void);
288
289
290
291/**
292 * __dev_remove_pack - remove packet handler
293 * @pt: packet type declaration
294 *
295 * Remove a protocol handler that was previously added to the kernel
296 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
297 * from the kernel lists and can be freed or reused once this function
298 * returns.
299 *
300 * The packet type might still be in use by receivers
301 * and must not be freed until after all the CPU's have gone
302 * through a quiescent state.
303 */
304void __dev_remove_pack(struct packet_type *pt)
305{
306 struct list_head *head;
307 struct packet_type *pt1;
308
309 spin_lock_bh(&ptype_lock);
310
311 if (pt->type == htons(ETH_P_ALL)) {
312 netdev_nit--;
313 head = &ptype_all;
314 } else
315 head = &ptype_base[ntohs(pt->type) & 15];
316
317 list_for_each_entry(pt1, head, list) {
318 if (pt == pt1) {
319 list_del_rcu(&pt->list);
320 goto out;
321 }
322 }
323
324 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
325out:
326 spin_unlock_bh(&ptype_lock);
327}
328/**
329 * dev_remove_pack - remove packet handler
330 * @pt: packet type declaration
331 *
332 * Remove a protocol handler that was previously added to the kernel
333 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
334 * from the kernel lists and can be freed or reused once this function
335 * returns.
336 *
337 * This call sleeps to guarantee that no CPU is looking at the packet
338 * type after return.
339 */
340void dev_remove_pack(struct packet_type *pt)
341{
342 __dev_remove_pack(pt);
343
344 synchronize_net();
345}
346
347/******************************************************************************
348
349 Device Boot-time Settings Routines
350
351*******************************************************************************/
352
353/* Boot time configuration table */
354static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
355
356/**
357 * netdev_boot_setup_add - add new setup entry
358 * @name: name of the device
359 * @map: configured settings for the device
360 *
361 * Adds new setup entry to the dev_boot_setup list. The function
362 * returns 0 on error and 1 on success. This is a generic routine to
363 * all netdevices.
364 */
365static int netdev_boot_setup_add(char *name, struct ifmap *map)
366{
367 struct netdev_boot_setup *s;
368 int i;
369
370 s = dev_boot_setup;
371 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
372 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
373 memset(s[i].name, 0, sizeof(s[i].name));
374 strcpy(s[i].name, name);
375 memcpy(&s[i].map, map, sizeof(s[i].map));
376 break;
377 }
378 }
379
380 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
381}
382
383/**
384 * netdev_boot_setup_check - check boot time settings
385 * @dev: the netdevice
386 *
387 * Check boot time settings for the device.
388 * The found settings are set for the device to be used
389 * later in the device probing.
390 * Returns 0 if no settings found, 1 if they are.
391 */
392int netdev_boot_setup_check(struct net_device *dev)
393{
394 struct netdev_boot_setup *s = dev_boot_setup;
395 int i;
396
397 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
398 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
399 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
400 dev->irq = s[i].map.irq;
401 dev->base_addr = s[i].map.base_addr;
402 dev->mem_start = s[i].map.mem_start;
403 dev->mem_end = s[i].map.mem_end;
404 return 1;
405 }
406 }
407 return 0;
408}
409
410
411/**
412 * netdev_boot_base - get address from boot time settings
413 * @prefix: prefix for network device
414 * @unit: id for network device
415 *
416 * Check boot time settings for the base address of device.
417 * The found settings are set for the device to be used
418 * later in the device probing.
419 * Returns 0 if no settings found.
420 */
421unsigned long netdev_boot_base(const char *prefix, int unit)
422{
423 const struct netdev_boot_setup *s = dev_boot_setup;
424 char name[IFNAMSIZ];
425 int i;
426
427 sprintf(name, "%s%d", prefix, unit);
428
429 /*
430 * If device already registered then return base of 1
431 * to indicate not to probe for this interface
432 */
433 if (__dev_get_by_name(name))
434 return 1;
435
436 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
437 if (!strcmp(name, s[i].name))
438 return s[i].map.base_addr;
439 return 0;
440}
441
442/*
443 * Saves at boot time configured settings for any netdevice.
444 */
445int __init netdev_boot_setup(char *str)
446{
447 int ints[5];
448 struct ifmap map;
449
450 str = get_options(str, ARRAY_SIZE(ints), ints);
451 if (!str || !*str)
452 return 0;
453
454 /* Save settings */
455 memset(&map, 0, sizeof(map));
456 if (ints[0] > 0)
457 map.irq = ints[1];
458 if (ints[0] > 1)
459 map.base_addr = ints[2];
460 if (ints[0] > 2)
461 map.mem_start = ints[3];
462 if (ints[0] > 3)
463 map.mem_end = ints[4];
464
465 /* Add new entry to the list */
466 return netdev_boot_setup_add(str, &map);
467}
468
469__setup("netdev=", netdev_boot_setup);
470
471/*******************************************************************************
472
473 Device Interface Subroutines
474
475*******************************************************************************/
476
477/**
478 * __dev_get_by_name - find a device by its name
479 * @name: name to find
480 *
481 * Find an interface by name. Must be called under RTNL semaphore
482 * or @dev_base_lock. If the name is found a pointer to the device
483 * is returned. If the name is not found then %NULL is returned. The
484 * reference counters are not incremented so the caller must be
485 * careful with locks.
486 */
487
488struct net_device *__dev_get_by_name(const char *name)
489{
490 struct hlist_node *p;
491
492 hlist_for_each(p, dev_name_hash(name)) {
493 struct net_device *dev
494 = hlist_entry(p, struct net_device, name_hlist);
495 if (!strncmp(dev->name, name, IFNAMSIZ))
496 return dev;
497 }
498 return NULL;
499}
500
501/**
502 * dev_get_by_name - find a device by its name
503 * @name: name to find
504 *
505 * Find an interface by name. This can be called from any
506 * context and does its own locking. The returned handle has
507 * the usage count incremented and the caller must use dev_put() to
508 * release it when it is no longer needed. %NULL is returned if no
509 * matching device is found.
510 */
511
512struct net_device *dev_get_by_name(const char *name)
513{
514 struct net_device *dev;
515
516 read_lock(&dev_base_lock);
517 dev = __dev_get_by_name(name);
518 if (dev)
519 dev_hold(dev);
520 read_unlock(&dev_base_lock);
521 return dev;
522}
523
524/**
525 * __dev_get_by_index - find a device by its ifindex
526 * @ifindex: index of device
527 *
528 * Search for an interface by index. Returns %NULL if the device
529 * is not found or a pointer to the device. The device has not
530 * had its reference counter increased so the caller must be careful
531 * about locking. The caller must hold either the RTNL semaphore
532 * or @dev_base_lock.
533 */
534
535struct net_device *__dev_get_by_index(int ifindex)
536{
537 struct hlist_node *p;
538
539 hlist_for_each(p, dev_index_hash(ifindex)) {
540 struct net_device *dev
541 = hlist_entry(p, struct net_device, index_hlist);
542 if (dev->ifindex == ifindex)
543 return dev;
544 }
545 return NULL;
546}
547
548
549/**
550 * dev_get_by_index - find a device by its ifindex
551 * @ifindex: index of device
552 *
553 * Search for an interface by index. Returns NULL if the device
554 * is not found or a pointer to the device. The device returned has
555 * had a reference added and the pointer is safe until the user calls
556 * dev_put to indicate they have finished with it.
557 */
558
559struct net_device *dev_get_by_index(int ifindex)
560{
561 struct net_device *dev;
562
563 read_lock(&dev_base_lock);
564 dev = __dev_get_by_index(ifindex);
565 if (dev)
566 dev_hold(dev);
567 read_unlock(&dev_base_lock);
568 return dev;
569}
570
571/**
572 * dev_getbyhwaddr - find a device by its hardware address
573 * @type: media type of device
574 * @ha: hardware address
575 *
576 * Search for an interface by MAC address. Returns NULL if the device
577 * is not found or a pointer to the device. The caller must hold the
578 * rtnl semaphore. The returned device has not had its ref count increased
579 * and the caller must therefore be careful about locking
580 *
581 * BUGS:
582 * If the API was consistent this would be __dev_get_by_hwaddr
583 */
584
585struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
586{
587 struct net_device *dev;
588
589 ASSERT_RTNL();
590
591 for (dev = dev_base; dev; dev = dev->next)
592 if (dev->type == type &&
593 !memcmp(dev->dev_addr, ha, dev->addr_len))
594 break;
595 return dev;
596}
597
598struct net_device *dev_getfirstbyhwtype(unsigned short type)
599{
600 struct net_device *dev;
601
602 rtnl_lock();
603 for (dev = dev_base; dev; dev = dev->next) {
604 if (dev->type == type) {
605 dev_hold(dev);
606 break;
607 }
608 }
609 rtnl_unlock();
610 return dev;
611}
612
613EXPORT_SYMBOL(dev_getfirstbyhwtype);
614
615/**
616 * dev_get_by_flags - find any device with given flags
617 * @if_flags: IFF_* values
618 * @mask: bitmask of bits in if_flags to check
619 *
620 * Search for any interface with the given flags. Returns NULL if a device
621 * is not found or a pointer to the device. The device returned has
622 * had a reference added and the pointer is safe until the user calls
623 * dev_put to indicate they have finished with it.
624 */
625
626struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
627{
628 struct net_device *dev;
629
630 read_lock(&dev_base_lock);
631 for (dev = dev_base; dev != NULL; dev = dev->next) {
632 if (((dev->flags ^ if_flags) & mask) == 0) {
633 dev_hold(dev);
634 break;
635 }
636 }
637 read_unlock(&dev_base_lock);
638 return dev;
639}
640
641/**
642 * dev_valid_name - check if name is okay for network device
643 * @name: name string
644 *
645 * Network device names need to be valid file names to
646 * to allow sysfs to work
647 */
648static int dev_valid_name(const char *name)
649{
650 return !(*name == '\0'
651 || !strcmp(name, ".")
652 || !strcmp(name, "..")
653 || strchr(name, '/'));
654}
655
656/**
657 * dev_alloc_name - allocate a name for a device
658 * @dev: device
659 * @name: name format string
660 *
661 * Passed a format string - eg "lt%d" it will try and find a suitable
662 * id. Not efficient for many devices, not called a lot. The caller
663 * must hold the dev_base or rtnl lock while allocating the name and
664 * adding the device in order to avoid duplicates. Returns the number
665 * of the unit assigned or a negative errno code.
666 */
667
668int dev_alloc_name(struct net_device *dev, const char *name)
669{
670 int i = 0;
671 char buf[IFNAMSIZ];
672 const char *p;
673 const int max_netdevices = 8*PAGE_SIZE;
674 long *inuse;
675 struct net_device *d;
676
677 p = strnchr(name, IFNAMSIZ-1, '%');
678 if (p) {
679 /*
680 * Verify the string as this thing may have come from
681 * the user. There must be either one "%d" and no other "%"
682 * characters.
683 */
684 if (p[1] != 'd' || strchr(p + 2, '%'))
685 return -EINVAL;
686
687 /* Use one page as a bit array of possible slots */
688 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
689 if (!inuse)
690 return -ENOMEM;
691
692 for (d = dev_base; d; d = d->next) {
693 if (!sscanf(d->name, name, &i))
694 continue;
695 if (i < 0 || i >= max_netdevices)
696 continue;
697
698 /* avoid cases where sscanf is not exact inverse of printf */
699 snprintf(buf, sizeof(buf), name, i);
700 if (!strncmp(buf, d->name, IFNAMSIZ))
701 set_bit(i, inuse);
702 }
703
704 i = find_first_zero_bit(inuse, max_netdevices);
705 free_page((unsigned long) inuse);
706 }
707
708 snprintf(buf, sizeof(buf), name, i);
709 if (!__dev_get_by_name(buf)) {
710 strlcpy(dev->name, buf, IFNAMSIZ);
711 return i;
712 }
713
714 /* It is possible to run out of possible slots
715 * when the name is long and there isn't enough space left
716 * for the digits, or if all bits are used.
717 */
718 return -ENFILE;
719}
720
721
722/**
723 * dev_change_name - change name of a device
724 * @dev: device
725 * @newname: name (or format string) must be at least IFNAMSIZ
726 *
727 * Change name of a device, can pass format strings "eth%d".
728 * for wildcarding.
729 */
730int dev_change_name(struct net_device *dev, char *newname)
731{
732 int err = 0;
733
734 ASSERT_RTNL();
735
736 if (dev->flags & IFF_UP)
737 return -EBUSY;
738
739 if (!dev_valid_name(newname))
740 return -EINVAL;
741
742 if (strchr(newname, '%')) {
743 err = dev_alloc_name(dev, newname);
744 if (err < 0)
745 return err;
746 strcpy(newname, dev->name);
747 }
748 else if (__dev_get_by_name(newname))
749 return -EEXIST;
750 else
751 strlcpy(dev->name, newname, IFNAMSIZ);
752
753 err = class_device_rename(&dev->class_dev, dev->name);
754 if (!err) {
755 hlist_del(&dev->name_hlist);
756 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
757 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
758 }
759
760 return err;
761}
762
763/**
764 * netdev_state_change - device changes state
765 * @dev: device to cause notification
766 *
767 * Called to indicate a device has changed state. This function calls
768 * the notifier chains for netdev_chain and sends a NEWLINK message
769 * to the routing socket.
770 */
771void netdev_state_change(struct net_device *dev)
772{
773 if (dev->flags & IFF_UP) {
774 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
775 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
776 }
777}
778
779/**
780 * dev_load - load a network module
781 * @name: name of interface
782 *
783 * If a network interface is not present and the process has suitable
784 * privileges this function loads the module. If module loading is not
785 * available in this kernel then it becomes a nop.
786 */
787
788void dev_load(const char *name)
789{
790 struct net_device *dev;
791
792 read_lock(&dev_base_lock);
793 dev = __dev_get_by_name(name);
794 read_unlock(&dev_base_lock);
795
796 if (!dev && capable(CAP_SYS_MODULE))
797 request_module("%s", name);
798}
799
800static int default_rebuild_header(struct sk_buff *skb)
801{
802 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
803 skb->dev ? skb->dev->name : "NULL!!!");
804 kfree_skb(skb);
805 return 1;
806}
807
808
809/**
810 * dev_open - prepare an interface for use.
811 * @dev: device to open
812 *
813 * Takes a device from down to up state. The device's private open
814 * function is invoked and then the multicast lists are loaded. Finally
815 * the device is moved into the up state and a %NETDEV_UP message is
816 * sent to the netdev notifier chain.
817 *
818 * Calling this function on an active interface is a nop. On a failure
819 * a negative errno code is returned.
820 */
821int dev_open(struct net_device *dev)
822{
823 int ret = 0;
824
825 /*
826 * Is it already up?
827 */
828
829 if (dev->flags & IFF_UP)
830 return 0;
831
832 /*
833 * Is it even present?
834 */
835 if (!netif_device_present(dev))
836 return -ENODEV;
837
838 /*
839 * Call device private open method
840 */
841 set_bit(__LINK_STATE_START, &dev->state);
842 if (dev->open) {
843 ret = dev->open(dev);
844 if (ret)
845 clear_bit(__LINK_STATE_START, &dev->state);
846 }
847
848 /*
849 * If it went open OK then:
850 */
851
852 if (!ret) {
853 /*
854 * Set the flags.
855 */
856 dev->flags |= IFF_UP;
857
858 /*
859 * Initialize multicasting status
860 */
861 dev_mc_upload(dev);
862
863 /*
864 * Wakeup transmit queue engine
865 */
866 dev_activate(dev);
867
868 /*
869 * ... and announce new interface.
870 */
871 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
872 }
873 return ret;
874}
875
876/**
877 * dev_close - shutdown an interface.
878 * @dev: device to shutdown
879 *
880 * This function moves an active device into down state. A
881 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
882 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
883 * chain.
884 */
885int dev_close(struct net_device *dev)
886{
887 if (!(dev->flags & IFF_UP))
888 return 0;
889
890 /*
891 * Tell people we are going down, so that they can
892 * prepare to death, when device is still operating.
893 */
894 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
895
896 dev_deactivate(dev);
897
898 clear_bit(__LINK_STATE_START, &dev->state);
899
900 /* Synchronize to scheduled poll. We cannot touch poll list,
901 * it can be even on different cpu. So just clear netif_running(),
902 * and wait when poll really will happen. Actually, the best place
903 * for this is inside dev->stop() after device stopped its irq
904 * engine, but this requires more changes in devices. */
905
906 smp_mb__after_clear_bit(); /* Commit netif_running(). */
907 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
908 /* No hurry. */
909 current->state = TASK_INTERRUPTIBLE;
910 schedule_timeout(1);
911 }
912
913 /*
914 * Call the device specific close. This cannot fail.
915 * Only if device is UP
916 *
917 * We allow it to be called even after a DETACH hot-plug
918 * event.
919 */
920 if (dev->stop)
921 dev->stop(dev);
922
923 /*
924 * Device is now down.
925 */
926
927 dev->flags &= ~IFF_UP;
928
929 /*
930 * Tell people we are down
931 */
932 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
933
934 return 0;
935}
936
937
938/*
939 * Device change register/unregister. These are not inline or static
940 * as we export them to the world.
941 */
942
943/**
944 * register_netdevice_notifier - register a network notifier block
945 * @nb: notifier
946 *
947 * Register a notifier to be called when network device events occur.
948 * The notifier passed is linked into the kernel structures and must
949 * not be reused until it has been unregistered. A negative errno code
950 * is returned on a failure.
951 *
952 * When registered all registration and up events are replayed
953 * to the new notifier to allow device to have a race free
954 * view of the network device list.
955 */
956
957int register_netdevice_notifier(struct notifier_block *nb)
958{
959 struct net_device *dev;
960 int err;
961
962 rtnl_lock();
963 err = notifier_chain_register(&netdev_chain, nb);
964 if (!err) {
965 for (dev = dev_base; dev; dev = dev->next) {
966 nb->notifier_call(nb, NETDEV_REGISTER, dev);
967
968 if (dev->flags & IFF_UP)
969 nb->notifier_call(nb, NETDEV_UP, dev);
970 }
971 }
972 rtnl_unlock();
973 return err;
974}
975
976/**
977 * unregister_netdevice_notifier - unregister a network notifier block
978 * @nb: notifier
979 *
980 * Unregister a notifier previously registered by
981 * register_netdevice_notifier(). The notifier is unlinked into the
982 * kernel structures and may then be reused. A negative errno code
983 * is returned on a failure.
984 */
985
986int unregister_netdevice_notifier(struct notifier_block *nb)
987{
988 return notifier_chain_unregister(&netdev_chain, nb);
989}
990
991/**
992 * call_netdevice_notifiers - call all network notifier blocks
993 * @val: value passed unmodified to notifier function
994 * @v: pointer passed unmodified to notifier function
995 *
996 * Call all network notifier blocks. Parameters and return value
997 * are as for notifier_call_chain().
998 */
999
1000int call_netdevice_notifiers(unsigned long val, void *v)
1001{
1002 return notifier_call_chain(&netdev_chain, val, v);
1003}
1004
1005/* When > 0 there are consumers of rx skb time stamps */
1006static atomic_t netstamp_needed = ATOMIC_INIT(0);
1007
1008void net_enable_timestamp(void)
1009{
1010 atomic_inc(&netstamp_needed);
1011}
1012
1013void net_disable_timestamp(void)
1014{
1015 atomic_dec(&netstamp_needed);
1016}
1017
1018static inline void net_timestamp(struct timeval *stamp)
1019{
1020 if (atomic_read(&netstamp_needed))
1021 do_gettimeofday(stamp);
1022 else {
1023 stamp->tv_sec = 0;
1024 stamp->tv_usec = 0;
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;
1036 net_timestamp(&skb->stamp);
1037
1038 rcu_read_lock();
1039 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1040 /* Never send packets back to the socket
1041 * they originated from - MvS (miquels@drinkel.ow.org)
1042 */
1043 if ((ptype->dev == dev || !ptype->dev) &&
1044 (ptype->af_packet_priv == NULL ||
1045 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1046 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1047 if (!skb2)
1048 break;
1049
1050 /* skb->nh should be correctly
1051 set by sender, so that the second statement is
1052 just protection against buggy protocols.
1053 */
1054 skb2->mac.raw = skb2->data;
1055
1056 if (skb2->nh.raw < skb2->data ||
1057 skb2->nh.raw > skb2->tail) {
1058 if (net_ratelimit())
1059 printk(KERN_CRIT "protocol %04x is "
1060 "buggy, dev %s\n",
1061 skb2->protocol, dev->name);
1062 skb2->nh.raw = skb2->data;
1063 }
1064
1065 skb2->h.raw = skb2->nh.raw;
1066 skb2->pkt_type = PACKET_OUTGOING;
1067 ptype->func(skb2, skb->dev, ptype);
1068 }
1069 }
1070 rcu_read_unlock();
1071}
1072
1073/*
1074 * Invalidate hardware checksum when packet is to be mangled, and
1075 * complete checksum manually on outgoing path.
1076 */
1077int skb_checksum_help(struct sk_buff *skb, int inward)
1078{
1079 unsigned int csum;
1080 int ret = 0, offset = skb->h.raw - skb->data;
1081
1082 if (inward) {
1083 skb->ip_summed = CHECKSUM_NONE;
1084 goto out;
1085 }
1086
1087 if (skb_cloned(skb)) {
1088 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1089 if (ret)
1090 goto out;
1091 }
1092
1093 if (offset > (int)skb->len)
1094 BUG();
1095 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1096
1097 offset = skb->tail - skb->h.raw;
1098 if (offset <= 0)
1099 BUG();
1100 if (skb->csum + 2 > offset)
1101 BUG();
1102
1103 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1104 skb->ip_summed = CHECKSUM_NONE;
1105out:
1106 return ret;
1107}
1108
1109#ifdef CONFIG_HIGHMEM
1110/* Actually, we should eliminate this check as soon as we know, that:
1111 * 1. IOMMU is present and allows to map all the memory.
1112 * 2. No high memory really exists on this machine.
1113 */
1114
1115static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1116{
1117 int i;
1118
1119 if (dev->features & NETIF_F_HIGHDMA)
1120 return 0;
1121
1122 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1123 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1124 return 1;
1125
1126 return 0;
1127}
1128#else
1129#define illegal_highdma(dev, skb) (0)
1130#endif
1131
1132extern void skb_release_data(struct sk_buff *);
1133
1134/* Keep head the same: replace data */
1135int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1136{
1137 unsigned int size;
1138 u8 *data;
1139 long offset;
1140 struct skb_shared_info *ninfo;
1141 int headerlen = skb->data - skb->head;
1142 int expand = (skb->tail + skb->data_len) - skb->end;
1143
1144 if (skb_shared(skb))
1145 BUG();
1146
1147 if (expand <= 0)
1148 expand = 0;
1149
1150 size = skb->end - skb->head + expand;
1151 size = SKB_DATA_ALIGN(size);
1152 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1153 if (!data)
1154 return -ENOMEM;
1155
1156 /* Copy entire thing */
1157 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1158 BUG();
1159
1160 /* Set up shinfo */
1161 ninfo = (struct skb_shared_info*)(data + size);
1162 atomic_set(&ninfo->dataref, 1);
1163 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1164 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1165 ninfo->nr_frags = 0;
1166 ninfo->frag_list = NULL;
1167
1168 /* Offset between the two in bytes */
1169 offset = data - skb->head;
1170
1171 /* Free old data. */
1172 skb_release_data(skb);
1173
1174 skb->head = data;
1175 skb->end = data + size;
1176
1177 /* Set up new pointers */
1178 skb->h.raw += offset;
1179 skb->nh.raw += offset;
1180 skb->mac.raw += offset;
1181 skb->tail += offset;
1182 skb->data += offset;
1183
1184 /* We are no longer a clone, even if we were. */
1185 skb->cloned = 0;
1186
1187 skb->tail += skb->data_len;
1188 skb->data_len = 0;
1189 return 0;
1190}
1191
1192#define HARD_TX_LOCK(dev, cpu) { \
1193 if ((dev->features & NETIF_F_LLTX) == 0) { \
1194 spin_lock(&dev->xmit_lock); \
1195 dev->xmit_lock_owner = cpu; \
1196 } \
1197}
1198
1199#define HARD_TX_UNLOCK(dev) { \
1200 if ((dev->features & NETIF_F_LLTX) == 0) { \
1201 dev->xmit_lock_owner = -1; \
1202 spin_unlock(&dev->xmit_lock); \
1203 } \
1204}
1205
1206/**
1207 * dev_queue_xmit - transmit a buffer
1208 * @skb: buffer to transmit
1209 *
1210 * Queue a buffer for transmission to a network device. The caller must
1211 * have set the device and priority and built the buffer before calling
1212 * this function. The function can be called from an interrupt.
1213 *
1214 * A negative errno code is returned on a failure. A success does not
1215 * guarantee the frame will be transmitted as it may be dropped due
1216 * to congestion or traffic shaping.
1217 */
1218
1219int dev_queue_xmit(struct sk_buff *skb)
1220{
1221 struct net_device *dev = skb->dev;
1222 struct Qdisc *q;
1223 int rc = -ENOMEM;
1224
1225 if (skb_shinfo(skb)->frag_list &&
1226 !(dev->features & NETIF_F_FRAGLIST) &&
1227 __skb_linearize(skb, GFP_ATOMIC))
1228 goto out_kfree_skb;
1229
1230 /* Fragmented skb is linearized if device does not support SG,
1231 * or if at least one of fragments is in highmem and device
1232 * does not support DMA from it.
1233 */
1234 if (skb_shinfo(skb)->nr_frags &&
1235 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1236 __skb_linearize(skb, GFP_ATOMIC))
1237 goto out_kfree_skb;
1238
1239 /* If packet is not checksummed and device does not support
1240 * checksumming for this protocol, complete checksumming here.
1241 */
1242 if (skb->ip_summed == CHECKSUM_HW &&
1243 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1244 (!(dev->features & NETIF_F_IP_CSUM) ||
1245 skb->protocol != htons(ETH_P_IP))))
1246 if (skb_checksum_help(skb, 0))
1247 goto out_kfree_skb;
1248
1249 /* Disable soft irqs for various locks below. Also
1250 * stops preemption for RCU.
1251 */
1252 local_bh_disable();
1253
1254 /* Updates of qdisc are serialized by queue_lock.
1255 * The struct Qdisc which is pointed to by qdisc is now a
1256 * rcu structure - it may be accessed without acquiring
1257 * a lock (but the structure may be stale.) The freeing of the
1258 * qdisc will be deferred until it's known that there are no
1259 * more references to it.
1260 *
1261 * If the qdisc has an enqueue function, we still need to
1262 * hold the queue_lock before calling it, since queue_lock
1263 * also serializes access to the device queue.
1264 */
1265
1266 q = rcu_dereference(dev->qdisc);
1267#ifdef CONFIG_NET_CLS_ACT
1268 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1269#endif
1270 if (q->enqueue) {
1271 /* Grab device queue */
1272 spin_lock(&dev->queue_lock);
1273
1274 rc = q->enqueue(skb, q);
1275
1276 qdisc_run(dev);
1277
1278 spin_unlock(&dev->queue_lock);
1279 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1280 goto out;
1281 }
1282
1283 /* The device has no queue. Common case for software devices:
1284 loopback, all the sorts of tunnels...
1285
1286 Really, it is unlikely that xmit_lock protection is necessary here.
1287 (f.e. loopback and IP tunnels are clean ignoring statistics
1288 counters.)
1289 However, it is possible, that they rely on protection
1290 made by us here.
1291
1292 Check this and shot the lock. It is not prone from deadlocks.
1293 Either shot noqueue qdisc, it is even simpler 8)
1294 */
1295 if (dev->flags & IFF_UP) {
1296 int cpu = smp_processor_id(); /* ok because BHs are off */
1297
1298 if (dev->xmit_lock_owner != cpu) {
1299
1300 HARD_TX_LOCK(dev, cpu);
1301
1302 if (!netif_queue_stopped(dev)) {
1303 if (netdev_nit)
1304 dev_queue_xmit_nit(skb, dev);
1305
1306 rc = 0;
1307 if (!dev->hard_start_xmit(skb, dev)) {
1308 HARD_TX_UNLOCK(dev);
1309 goto out;
1310 }
1311 }
1312 HARD_TX_UNLOCK(dev);
1313 if (net_ratelimit())
1314 printk(KERN_CRIT "Virtual device %s asks to "
1315 "queue packet!\n", dev->name);
1316 } else {
1317 /* Recursion is detected! It is possible,
1318 * unfortunately */
1319 if (net_ratelimit())
1320 printk(KERN_CRIT "Dead loop on virtual device "
1321 "%s, fix it urgently!\n", dev->name);
1322 }
1323 }
1324
1325 rc = -ENETDOWN;
1326 local_bh_enable();
1327
1328out_kfree_skb:
1329 kfree_skb(skb);
1330 return rc;
1331out:
1332 local_bh_enable();
1333 return rc;
1334}
1335
1336
1337/*=======================================================================
1338 Receiver routines
1339 =======================================================================*/
1340
1341int netdev_max_backlog = 300;
1342int weight_p = 64; /* old backlog weight */
1343/* These numbers are selected based on intuition and some
1344 * experimentatiom, if you have more scientific way of doing this
1345 * please go ahead and fix things.
1346 */
1347int no_cong_thresh = 10;
1348int no_cong = 20;
1349int lo_cong = 100;
1350int mod_cong = 290;
1351
1352DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1353
1354
1355static void get_sample_stats(int cpu)
1356{
1357#ifdef RAND_LIE
1358 unsigned long rd;
1359 int rq;
1360#endif
1361 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1362 int blog = sd->input_pkt_queue.qlen;
1363 int avg_blog = sd->avg_blog;
1364
1365 avg_blog = (avg_blog >> 1) + (blog >> 1);
1366
1367 if (avg_blog > mod_cong) {
1368 /* Above moderate congestion levels. */
1369 sd->cng_level = NET_RX_CN_HIGH;
1370#ifdef RAND_LIE
1371 rd = net_random();
1372 rq = rd % netdev_max_backlog;
1373 if (rq < avg_blog) /* unlucky bastard */
1374 sd->cng_level = NET_RX_DROP;
1375#endif
1376 } else if (avg_blog > lo_cong) {
1377 sd->cng_level = NET_RX_CN_MOD;
1378#ifdef RAND_LIE
1379 rd = net_random();
1380 rq = rd % netdev_max_backlog;
1381 if (rq < avg_blog) /* unlucky bastard */
1382 sd->cng_level = NET_RX_CN_HIGH;
1383#endif
1384 } else if (avg_blog > no_cong)
1385 sd->cng_level = NET_RX_CN_LOW;
1386 else /* no congestion */
1387 sd->cng_level = NET_RX_SUCCESS;
1388
1389 sd->avg_blog = avg_blog;
1390}
1391
1392#ifdef OFFLINE_SAMPLE
1393static void sample_queue(unsigned long dummy)
1394{
1395/* 10 ms 0r 1ms -- i don't care -- JHS */
1396 int next_tick = 1;
1397 int cpu = smp_processor_id();
1398
1399 get_sample_stats(cpu);
1400 next_tick += jiffies;
1401 mod_timer(&samp_timer, next_tick);
1402}
1403#endif
1404
1405
1406/**
1407 * netif_rx - post buffer to the network code
1408 * @skb: buffer to post
1409 *
1410 * This function receives a packet from a device driver and queues it for
1411 * the upper (protocol) levels to process. It always succeeds. The buffer
1412 * may be dropped during processing for congestion control or by the
1413 * protocol layers.
1414 *
1415 * return values:
1416 * NET_RX_SUCCESS (no congestion)
1417 * NET_RX_CN_LOW (low congestion)
1418 * NET_RX_CN_MOD (moderate congestion)
1419 * NET_RX_CN_HIGH (high congestion)
1420 * NET_RX_DROP (packet was dropped)
1421 *
1422 */
1423
1424int netif_rx(struct sk_buff *skb)
1425{
1426 int this_cpu;
1427 struct softnet_data *queue;
1428 unsigned long flags;
1429
1430 /* if netpoll wants it, pretend we never saw it */
1431 if (netpoll_rx(skb))
1432 return NET_RX_DROP;
1433
1434 if (!skb->stamp.tv_sec)
1435 net_timestamp(&skb->stamp);
1436
1437 /*
1438 * The code is rearranged so that the path is the most
1439 * short when CPU is congested, but is still operating.
1440 */
1441 local_irq_save(flags);
1442 this_cpu = smp_processor_id();
1443 queue = &__get_cpu_var(softnet_data);
1444
1445 __get_cpu_var(netdev_rx_stat).total++;
1446 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1447 if (queue->input_pkt_queue.qlen) {
1448 if (queue->throttle)
1449 goto drop;
1450
1451enqueue:
1452 dev_hold(skb->dev);
1453 __skb_queue_tail(&queue->input_pkt_queue, skb);
1454#ifndef OFFLINE_SAMPLE
1455 get_sample_stats(this_cpu);
1456#endif
1457 local_irq_restore(flags);
1458 return queue->cng_level;
1459 }
1460
1461 if (queue->throttle)
1462 queue->throttle = 0;
1463
1464 netif_rx_schedule(&queue->backlog_dev);
1465 goto enqueue;
1466 }
1467
1468 if (!queue->throttle) {
1469 queue->throttle = 1;
1470 __get_cpu_var(netdev_rx_stat).throttled++;
1471 }
1472
1473drop:
1474 __get_cpu_var(netdev_rx_stat).dropped++;
1475 local_irq_restore(flags);
1476
1477 kfree_skb(skb);
1478 return NET_RX_DROP;
1479}
1480
1481int netif_rx_ni(struct sk_buff *skb)
1482{
1483 int err;
1484
1485 preempt_disable();
1486 err = netif_rx(skb);
1487 if (local_softirq_pending())
1488 do_softirq();
1489 preempt_enable();
1490
1491 return err;
1492}
1493
1494EXPORT_SYMBOL(netif_rx_ni);
1495
1496static __inline__ void skb_bond(struct sk_buff *skb)
1497{
1498 struct net_device *dev = skb->dev;
1499
1500 if (dev->master) {
1501 skb->real_dev = skb->dev;
1502 skb->dev = dev->master;
1503 }
1504}
1505
1506static void net_tx_action(struct softirq_action *h)
1507{
1508 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1509
1510 if (sd->completion_queue) {
1511 struct sk_buff *clist;
1512
1513 local_irq_disable();
1514 clist = sd->completion_queue;
1515 sd->completion_queue = NULL;
1516 local_irq_enable();
1517
1518 while (clist) {
1519 struct sk_buff *skb = clist;
1520 clist = clist->next;
1521
1522 BUG_TRAP(!atomic_read(&skb->users));
1523 __kfree_skb(skb);
1524 }
1525 }
1526
1527 if (sd->output_queue) {
1528 struct net_device *head;
1529
1530 local_irq_disable();
1531 head = sd->output_queue;
1532 sd->output_queue = NULL;
1533 local_irq_enable();
1534
1535 while (head) {
1536 struct net_device *dev = head;
1537 head = head->next_sched;
1538
1539 smp_mb__before_clear_bit();
1540 clear_bit(__LINK_STATE_SCHED, &dev->state);
1541
1542 if (spin_trylock(&dev->queue_lock)) {
1543 qdisc_run(dev);
1544 spin_unlock(&dev->queue_lock);
1545 } else {
1546 netif_schedule(dev);
1547 }
1548 }
1549 }
1550}
1551
1552static __inline__ int deliver_skb(struct sk_buff *skb,
1553 struct packet_type *pt_prev)
1554{
1555 atomic_inc(&skb->users);
1556 return pt_prev->func(skb, skb->dev, pt_prev);
1557}
1558
1559#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1560int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1561struct net_bridge;
1562struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1563 unsigned char *addr);
1564void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1565
1566static __inline__ int handle_bridge(struct sk_buff **pskb,
1567 struct packet_type **pt_prev, int *ret)
1568{
1569 struct net_bridge_port *port;
1570
1571 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1572 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1573 return 0;
1574
1575 if (*pt_prev) {
1576 *ret = deliver_skb(*pskb, *pt_prev);
1577 *pt_prev = NULL;
1578 }
1579
1580 return br_handle_frame_hook(port, pskb);
1581}
1582#else
1583#define handle_bridge(skb, pt_prev, ret) (0)
1584#endif
1585
1586#ifdef CONFIG_NET_CLS_ACT
1587/* TODO: Maybe we should just force sch_ingress to be compiled in
1588 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1589 * a compare and 2 stores extra right now if we dont have it on
1590 * but have CONFIG_NET_CLS_ACT
1591 * NOTE: This doesnt stop any functionality; if you dont have
1592 * the ingress scheduler, you just cant add policies on ingress.
1593 *
1594 */
1595static int ing_filter(struct sk_buff *skb)
1596{
1597 struct Qdisc *q;
1598 struct net_device *dev = skb->dev;
1599 int result = TC_ACT_OK;
1600
1601 if (dev->qdisc_ingress) {
1602 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1603 if (MAX_RED_LOOP < ttl++) {
1604 printk("Redir loop detected Dropping packet (%s->%s)\n",
1605 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1606 return TC_ACT_SHOT;
1607 }
1608
1609 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1610
1611 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1612 if (NULL == skb->input_dev) {
1613 skb->input_dev = skb->dev;
1614 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1615 }
1616 spin_lock(&dev->ingress_lock);
1617 if ((q = dev->qdisc_ingress) != NULL)
1618 result = q->enqueue(skb, q);
1619 spin_unlock(&dev->ingress_lock);
1620
1621 }
1622
1623 return result;
1624}
1625#endif
1626
1627int netif_receive_skb(struct sk_buff *skb)
1628{
1629 struct packet_type *ptype, *pt_prev;
1630 int ret = NET_RX_DROP;
1631 unsigned short type;
1632
1633 /* if we've gotten here through NAPI, check netpoll */
1634 if (skb->dev->poll && netpoll_rx(skb))
1635 return NET_RX_DROP;
1636
1637 if (!skb->stamp.tv_sec)
1638 net_timestamp(&skb->stamp);
1639
1640 skb_bond(skb);
1641
1642 __get_cpu_var(netdev_rx_stat).total++;
1643
1644 skb->h.raw = skb->nh.raw = skb->data;
1645 skb->mac_len = skb->nh.raw - skb->mac.raw;
1646
1647 pt_prev = NULL;
1648
1649 rcu_read_lock();
1650
1651#ifdef CONFIG_NET_CLS_ACT
1652 if (skb->tc_verd & TC_NCLS) {
1653 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1654 goto ncls;
1655 }
1656#endif
1657
1658 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1659 if (!ptype->dev || ptype->dev == skb->dev) {
1660 if (pt_prev)
1661 ret = deliver_skb(skb, pt_prev);
1662 pt_prev = ptype;
1663 }
1664 }
1665
1666#ifdef CONFIG_NET_CLS_ACT
1667 if (pt_prev) {
1668 ret = deliver_skb(skb, pt_prev);
1669 pt_prev = NULL; /* noone else should process this after*/
1670 } else {
1671 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1672 }
1673
1674 ret = ing_filter(skb);
1675
1676 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1677 kfree_skb(skb);
1678 goto out;
1679 }
1680
1681 skb->tc_verd = 0;
1682ncls:
1683#endif
1684
1685 handle_diverter(skb);
1686
1687 if (handle_bridge(&skb, &pt_prev, &ret))
1688 goto out;
1689
1690 type = skb->protocol;
1691 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1692 if (ptype->type == type &&
1693 (!ptype->dev || ptype->dev == skb->dev)) {
1694 if (pt_prev)
1695 ret = deliver_skb(skb, pt_prev);
1696 pt_prev = ptype;
1697 }
1698 }
1699
1700 if (pt_prev) {
1701 ret = pt_prev->func(skb, skb->dev, pt_prev);
1702 } else {
1703 kfree_skb(skb);
1704 /* Jamal, now you will not able to escape explaining
1705 * me how you were going to use this. :-)
1706 */
1707 ret = NET_RX_DROP;
1708 }
1709
1710out:
1711 rcu_read_unlock();
1712 return ret;
1713}
1714
1715static int process_backlog(struct net_device *backlog_dev, int *budget)
1716{
1717 int work = 0;
1718 int quota = min(backlog_dev->quota, *budget);
1719 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1720 unsigned long start_time = jiffies;
1721
1722 for (;;) {
1723 struct sk_buff *skb;
1724 struct net_device *dev;
1725
1726 local_irq_disable();
1727 skb = __skb_dequeue(&queue->input_pkt_queue);
1728 if (!skb)
1729 goto job_done;
1730 local_irq_enable();
1731
1732 dev = skb->dev;
1733
1734 netif_receive_skb(skb);
1735
1736 dev_put(dev);
1737
1738 work++;
1739
1740 if (work >= quota || jiffies - start_time > 1)
1741 break;
1742
1743 }
1744
1745 backlog_dev->quota -= work;
1746 *budget -= work;
1747 return -1;
1748
1749job_done:
1750 backlog_dev->quota -= work;
1751 *budget -= work;
1752
1753 list_del(&backlog_dev->poll_list);
1754 smp_mb__before_clear_bit();
1755 netif_poll_enable(backlog_dev);
1756
1757 if (queue->throttle)
1758 queue->throttle = 0;
1759 local_irq_enable();
1760 return 0;
1761}
1762
1763static void net_rx_action(struct softirq_action *h)
1764{
1765 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1766 unsigned long start_time = jiffies;
1767 int budget = netdev_max_backlog;
1768
1769
1770 local_irq_disable();
1771
1772 while (!list_empty(&queue->poll_list)) {
1773 struct net_device *dev;
1774
1775 if (budget <= 0 || jiffies - start_time > 1)
1776 goto softnet_break;
1777
1778 local_irq_enable();
1779
1780 dev = list_entry(queue->poll_list.next,
1781 struct net_device, poll_list);
1782 netpoll_poll_lock(dev);
1783
1784 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1785 netpoll_poll_unlock(dev);
1786 local_irq_disable();
1787 list_del(&dev->poll_list);
1788 list_add_tail(&dev->poll_list, &queue->poll_list);
1789 if (dev->quota < 0)
1790 dev->quota += dev->weight;
1791 else
1792 dev->quota = dev->weight;
1793 } else {
1794 netpoll_poll_unlock(dev);
1795 dev_put(dev);
1796 local_irq_disable();
1797 }
1798 }
1799out:
1800 local_irq_enable();
1801 return;
1802
1803softnet_break:
1804 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1805 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1806 goto out;
1807}
1808
1809static gifconf_func_t * gifconf_list [NPROTO];
1810
1811/**
1812 * register_gifconf - register a SIOCGIF handler
1813 * @family: Address family
1814 * @gifconf: Function handler
1815 *
1816 * Register protocol dependent address dumping routines. The handler
1817 * that is passed must not be freed or reused until it has been replaced
1818 * by another handler.
1819 */
1820int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1821{
1822 if (family >= NPROTO)
1823 return -EINVAL;
1824 gifconf_list[family] = gifconf;
1825 return 0;
1826}
1827
1828
1829/*
1830 * Map an interface index to its name (SIOCGIFNAME)
1831 */
1832
1833/*
1834 * We need this ioctl for efficient implementation of the
1835 * if_indextoname() function required by the IPv6 API. Without
1836 * it, we would have to search all the interfaces to find a
1837 * match. --pb
1838 */
1839
1840static int dev_ifname(struct ifreq __user *arg)
1841{
1842 struct net_device *dev;
1843 struct ifreq ifr;
1844
1845 /*
1846 * Fetch the caller's info block.
1847 */
1848
1849 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1850 return -EFAULT;
1851
1852 read_lock(&dev_base_lock);
1853 dev = __dev_get_by_index(ifr.ifr_ifindex);
1854 if (!dev) {
1855 read_unlock(&dev_base_lock);
1856 return -ENODEV;
1857 }
1858
1859 strcpy(ifr.ifr_name, dev->name);
1860 read_unlock(&dev_base_lock);
1861
1862 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1863 return -EFAULT;
1864 return 0;
1865}
1866
1867/*
1868 * Perform a SIOCGIFCONF call. This structure will change
1869 * size eventually, and there is nothing I can do about it.
1870 * Thus we will need a 'compatibility mode'.
1871 */
1872
1873static int dev_ifconf(char __user *arg)
1874{
1875 struct ifconf ifc;
1876 struct net_device *dev;
1877 char __user *pos;
1878 int len;
1879 int total;
1880 int i;
1881
1882 /*
1883 * Fetch the caller's info block.
1884 */
1885
1886 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1887 return -EFAULT;
1888
1889 pos = ifc.ifc_buf;
1890 len = ifc.ifc_len;
1891
1892 /*
1893 * Loop over the interfaces, and write an info block for each.
1894 */
1895
1896 total = 0;
1897 for (dev = dev_base; dev; dev = dev->next) {
1898 for (i = 0; i < NPROTO; i++) {
1899 if (gifconf_list[i]) {
1900 int done;
1901 if (!pos)
1902 done = gifconf_list[i](dev, NULL, 0);
1903 else
1904 done = gifconf_list[i](dev, pos + total,
1905 len - total);
1906 if (done < 0)
1907 return -EFAULT;
1908 total += done;
1909 }
1910 }
1911 }
1912
1913 /*
1914 * All done. Write the updated control block back to the caller.
1915 */
1916 ifc.ifc_len = total;
1917
1918 /*
1919 * Both BSD and Solaris return 0 here, so we do too.
1920 */
1921 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1922}
1923
1924#ifdef CONFIG_PROC_FS
1925/*
1926 * This is invoked by the /proc filesystem handler to display a device
1927 * in detail.
1928 */
1929static __inline__ struct net_device *dev_get_idx(loff_t pos)
1930{
1931 struct net_device *dev;
1932 loff_t i;
1933
1934 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1935
1936 return i == pos ? dev : NULL;
1937}
1938
1939void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1940{
1941 read_lock(&dev_base_lock);
1942 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1943}
1944
1945void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1946{
1947 ++*pos;
1948 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1949}
1950
1951void dev_seq_stop(struct seq_file *seq, void *v)
1952{
1953 read_unlock(&dev_base_lock);
1954}
1955
1956static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1957{
1958 if (dev->get_stats) {
1959 struct net_device_stats *stats = dev->get_stats(dev);
1960
1961 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1962 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1963 dev->name, stats->rx_bytes, stats->rx_packets,
1964 stats->rx_errors,
1965 stats->rx_dropped + stats->rx_missed_errors,
1966 stats->rx_fifo_errors,
1967 stats->rx_length_errors + stats->rx_over_errors +
1968 stats->rx_crc_errors + stats->rx_frame_errors,
1969 stats->rx_compressed, stats->multicast,
1970 stats->tx_bytes, stats->tx_packets,
1971 stats->tx_errors, stats->tx_dropped,
1972 stats->tx_fifo_errors, stats->collisions,
1973 stats->tx_carrier_errors +
1974 stats->tx_aborted_errors +
1975 stats->tx_window_errors +
1976 stats->tx_heartbeat_errors,
1977 stats->tx_compressed);
1978 } else
1979 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1980}
1981
1982/*
1983 * Called from the PROCfs module. This now uses the new arbitrary sized
1984 * /proc/net interface to create /proc/net/dev
1985 */
1986static int dev_seq_show(struct seq_file *seq, void *v)
1987{
1988 if (v == SEQ_START_TOKEN)
1989 seq_puts(seq, "Inter-| Receive "
1990 " | Transmit\n"
1991 " face |bytes packets errs drop fifo frame "
1992 "compressed multicast|bytes packets errs "
1993 "drop fifo colls carrier compressed\n");
1994 else
1995 dev_seq_printf_stats(seq, v);
1996 return 0;
1997}
1998
1999static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2000{
2001 struct netif_rx_stats *rc = NULL;
2002
2003 while (*pos < NR_CPUS)
2004 if (cpu_online(*pos)) {
2005 rc = &per_cpu(netdev_rx_stat, *pos);
2006 break;
2007 } else
2008 ++*pos;
2009 return rc;
2010}
2011
2012static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2013{
2014 return softnet_get_online(pos);
2015}
2016
2017static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2018{
2019 ++*pos;
2020 return softnet_get_online(pos);
2021}
2022
2023static void softnet_seq_stop(struct seq_file *seq, void *v)
2024{
2025}
2026
2027static int softnet_seq_show(struct seq_file *seq, void *v)
2028{
2029 struct netif_rx_stats *s = v;
2030
2031 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2032 s->total, s->dropped, s->time_squeeze, s->throttled,
2033 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2034 s->fastroute_deferred_out,
2035#if 0
2036 s->fastroute_latency_reduction
2037#else
2038 s->cpu_collision
2039#endif
2040 );
2041 return 0;
2042}
2043
2044static struct seq_operations dev_seq_ops = {
2045 .start = dev_seq_start,
2046 .next = dev_seq_next,
2047 .stop = dev_seq_stop,
2048 .show = dev_seq_show,
2049};
2050
2051static int dev_seq_open(struct inode *inode, struct file *file)
2052{
2053 return seq_open(file, &dev_seq_ops);
2054}
2055
2056static struct file_operations dev_seq_fops = {
2057 .owner = THIS_MODULE,
2058 .open = dev_seq_open,
2059 .read = seq_read,
2060 .llseek = seq_lseek,
2061 .release = seq_release,
2062};
2063
2064static struct seq_operations softnet_seq_ops = {
2065 .start = softnet_seq_start,
2066 .next = softnet_seq_next,
2067 .stop = softnet_seq_stop,
2068 .show = softnet_seq_show,
2069};
2070
2071static int softnet_seq_open(struct inode *inode, struct file *file)
2072{
2073 return seq_open(file, &softnet_seq_ops);
2074}
2075
2076static struct file_operations softnet_seq_fops = {
2077 .owner = THIS_MODULE,
2078 .open = softnet_seq_open,
2079 .read = seq_read,
2080 .llseek = seq_lseek,
2081 .release = seq_release,
2082};
2083
2084#ifdef WIRELESS_EXT
2085extern int wireless_proc_init(void);
2086#else
2087#define wireless_proc_init() 0
2088#endif
2089
2090static int __init dev_proc_init(void)
2091{
2092 int rc = -ENOMEM;
2093
2094 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2095 goto out;
2096 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2097 goto out_dev;
2098 if (wireless_proc_init())
2099 goto out_softnet;
2100 rc = 0;
2101out:
2102 return rc;
2103out_softnet:
2104 proc_net_remove("softnet_stat");
2105out_dev:
2106 proc_net_remove("dev");
2107 goto out;
2108}
2109#else
2110#define dev_proc_init() 0
2111#endif /* CONFIG_PROC_FS */
2112
2113
2114/**
2115 * netdev_set_master - set up master/slave pair
2116 * @slave: slave device
2117 * @master: new master device
2118 *
2119 * Changes the master device of the slave. Pass %NULL to break the
2120 * bonding. The caller must hold the RTNL semaphore. On a failure
2121 * a negative errno code is returned. On success the reference counts
2122 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2123 * function returns zero.
2124 */
2125int netdev_set_master(struct net_device *slave, struct net_device *master)
2126{
2127 struct net_device *old = slave->master;
2128
2129 ASSERT_RTNL();
2130
2131 if (master) {
2132 if (old)
2133 return -EBUSY;
2134 dev_hold(master);
2135 }
2136
2137 slave->master = master;
2138
2139 synchronize_net();
2140
2141 if (old)
2142 dev_put(old);
2143
2144 if (master)
2145 slave->flags |= IFF_SLAVE;
2146 else
2147 slave->flags &= ~IFF_SLAVE;
2148
2149 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2150 return 0;
2151}
2152
2153/**
2154 * dev_set_promiscuity - update promiscuity count on a device
2155 * @dev: device
2156 * @inc: modifier
2157 *
2158 * Add or remove promsicuity from a device. While the count in the device
2159 * remains above zero the interface remains promiscuous. Once it hits zero
2160 * the device reverts back to normal filtering operation. A negative inc
2161 * value is used to drop promiscuity on the device.
2162 */
2163void dev_set_promiscuity(struct net_device *dev, int inc)
2164{
2165 unsigned short old_flags = dev->flags;
2166
2167 dev->flags |= IFF_PROMISC;
2168 if ((dev->promiscuity += inc) == 0)
2169 dev->flags &= ~IFF_PROMISC;
2170 if (dev->flags ^ old_flags) {
2171 dev_mc_upload(dev);
2172 printk(KERN_INFO "device %s %s promiscuous mode\n",
2173 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2174 "left");
2175 }
2176}
2177
2178/**
2179 * dev_set_allmulti - update allmulti count on a device
2180 * @dev: device
2181 * @inc: modifier
2182 *
2183 * Add or remove reception of all multicast frames to a device. While the
2184 * count in the device remains above zero the interface remains listening
2185 * to all interfaces. Once it hits zero the device reverts back to normal
2186 * filtering operation. A negative @inc value is used to drop the counter
2187 * when releasing a resource needing all multicasts.
2188 */
2189
2190void dev_set_allmulti(struct net_device *dev, int inc)
2191{
2192 unsigned short old_flags = dev->flags;
2193
2194 dev->flags |= IFF_ALLMULTI;
2195 if ((dev->allmulti += inc) == 0)
2196 dev->flags &= ~IFF_ALLMULTI;
2197 if (dev->flags ^ old_flags)
2198 dev_mc_upload(dev);
2199}
2200
2201unsigned dev_get_flags(const struct net_device *dev)
2202{
2203 unsigned flags;
2204
2205 flags = (dev->flags & ~(IFF_PROMISC |
2206 IFF_ALLMULTI |
2207 IFF_RUNNING)) |
2208 (dev->gflags & (IFF_PROMISC |
2209 IFF_ALLMULTI));
2210
2211 if (netif_running(dev) && netif_carrier_ok(dev))
2212 flags |= IFF_RUNNING;
2213
2214 return flags;
2215}
2216
2217int dev_change_flags(struct net_device *dev, unsigned flags)
2218{
2219 int ret;
2220 int old_flags = dev->flags;
2221
2222 /*
2223 * Set the flags on our device.
2224 */
2225
2226 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2227 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2228 IFF_AUTOMEDIA)) |
2229 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2230 IFF_ALLMULTI));
2231
2232 /*
2233 * Load in the correct multicast list now the flags have changed.
2234 */
2235
2236 dev_mc_upload(dev);
2237
2238 /*
2239 * Have we downed the interface. We handle IFF_UP ourselves
2240 * according to user attempts to set it, rather than blindly
2241 * setting it.
2242 */
2243
2244 ret = 0;
2245 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2246 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2247
2248 if (!ret)
2249 dev_mc_upload(dev);
2250 }
2251
2252 if (dev->flags & IFF_UP &&
2253 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2254 IFF_VOLATILE)))
2255 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2256
2257 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2258 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2259 dev->gflags ^= IFF_PROMISC;
2260 dev_set_promiscuity(dev, inc);
2261 }
2262
2263 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2264 is important. Some (broken) drivers set IFF_PROMISC, when
2265 IFF_ALLMULTI is requested not asking us and not reporting.
2266 */
2267 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2268 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2269 dev->gflags ^= IFF_ALLMULTI;
2270 dev_set_allmulti(dev, inc);
2271 }
2272
2273 if (old_flags ^ dev->flags)
2274 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2275
2276 return ret;
2277}
2278
2279int dev_set_mtu(struct net_device *dev, int new_mtu)
2280{
2281 int err;
2282
2283 if (new_mtu == dev->mtu)
2284 return 0;
2285
2286 /* MTU must be positive. */
2287 if (new_mtu < 0)
2288 return -EINVAL;
2289
2290 if (!netif_device_present(dev))
2291 return -ENODEV;
2292
2293 err = 0;
2294 if (dev->change_mtu)
2295 err = dev->change_mtu(dev, new_mtu);
2296 else
2297 dev->mtu = new_mtu;
2298 if (!err && dev->flags & IFF_UP)
2299 notifier_call_chain(&netdev_chain,
2300 NETDEV_CHANGEMTU, dev);
2301 return err;
2302}
2303
2304int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2305{
2306 int err;
2307
2308 if (!dev->set_mac_address)
2309 return -EOPNOTSUPP;
2310 if (sa->sa_family != dev->type)
2311 return -EINVAL;
2312 if (!netif_device_present(dev))
2313 return -ENODEV;
2314 err = dev->set_mac_address(dev, sa);
2315 if (!err)
2316 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2317 return err;
2318}
2319
2320/*
2321 * Perform the SIOCxIFxxx calls.
2322 */
2323static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2324{
2325 int err;
2326 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2327
2328 if (!dev)
2329 return -ENODEV;
2330
2331 switch (cmd) {
2332 case SIOCGIFFLAGS: /* Get interface flags */
2333 ifr->ifr_flags = dev_get_flags(dev);
2334 return 0;
2335
2336 case SIOCSIFFLAGS: /* Set interface flags */
2337 return dev_change_flags(dev, ifr->ifr_flags);
2338
2339 case SIOCGIFMETRIC: /* Get the metric on the interface
2340 (currently unused) */
2341 ifr->ifr_metric = 0;
2342 return 0;
2343
2344 case SIOCSIFMETRIC: /* Set the metric on the interface
2345 (currently unused) */
2346 return -EOPNOTSUPP;
2347
2348 case SIOCGIFMTU: /* Get the MTU of a device */
2349 ifr->ifr_mtu = dev->mtu;
2350 return 0;
2351
2352 case SIOCSIFMTU: /* Set the MTU of a device */
2353 return dev_set_mtu(dev, ifr->ifr_mtu);
2354
2355 case SIOCGIFHWADDR:
2356 if (!dev->addr_len)
2357 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2358 else
2359 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2360 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2361 ifr->ifr_hwaddr.sa_family = dev->type;
2362 return 0;
2363
2364 case SIOCSIFHWADDR:
2365 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2366
2367 case SIOCSIFHWBROADCAST:
2368 if (ifr->ifr_hwaddr.sa_family != dev->type)
2369 return -EINVAL;
2370 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2371 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2372 notifier_call_chain(&netdev_chain,
2373 NETDEV_CHANGEADDR, dev);
2374 return 0;
2375
2376 case SIOCGIFMAP:
2377 ifr->ifr_map.mem_start = dev->mem_start;
2378 ifr->ifr_map.mem_end = dev->mem_end;
2379 ifr->ifr_map.base_addr = dev->base_addr;
2380 ifr->ifr_map.irq = dev->irq;
2381 ifr->ifr_map.dma = dev->dma;
2382 ifr->ifr_map.port = dev->if_port;
2383 return 0;
2384
2385 case SIOCSIFMAP:
2386 if (dev->set_config) {
2387 if (!netif_device_present(dev))
2388 return -ENODEV;
2389 return dev->set_config(dev, &ifr->ifr_map);
2390 }
2391 return -EOPNOTSUPP;
2392
2393 case SIOCADDMULTI:
2394 if (!dev->set_multicast_list ||
2395 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2396 return -EINVAL;
2397 if (!netif_device_present(dev))
2398 return -ENODEV;
2399 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2400 dev->addr_len, 1);
2401
2402 case SIOCDELMULTI:
2403 if (!dev->set_multicast_list ||
2404 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2405 return -EINVAL;
2406 if (!netif_device_present(dev))
2407 return -ENODEV;
2408 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2409 dev->addr_len, 1);
2410
2411 case SIOCGIFINDEX:
2412 ifr->ifr_ifindex = dev->ifindex;
2413 return 0;
2414
2415 case SIOCGIFTXQLEN:
2416 ifr->ifr_qlen = dev->tx_queue_len;
2417 return 0;
2418
2419 case SIOCSIFTXQLEN:
2420 if (ifr->ifr_qlen < 0)
2421 return -EINVAL;
2422 dev->tx_queue_len = ifr->ifr_qlen;
2423 return 0;
2424
2425 case SIOCSIFNAME:
2426 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2427 return dev_change_name(dev, ifr->ifr_newname);
2428
2429 /*
2430 * Unknown or private ioctl
2431 */
2432
2433 default:
2434 if ((cmd >= SIOCDEVPRIVATE &&
2435 cmd <= SIOCDEVPRIVATE + 15) ||
2436 cmd == SIOCBONDENSLAVE ||
2437 cmd == SIOCBONDRELEASE ||
2438 cmd == SIOCBONDSETHWADDR ||
2439 cmd == SIOCBONDSLAVEINFOQUERY ||
2440 cmd == SIOCBONDINFOQUERY ||
2441 cmd == SIOCBONDCHANGEACTIVE ||
2442 cmd == SIOCGMIIPHY ||
2443 cmd == SIOCGMIIREG ||
2444 cmd == SIOCSMIIREG ||
2445 cmd == SIOCBRADDIF ||
2446 cmd == SIOCBRDELIF ||
2447 cmd == SIOCWANDEV) {
2448 err = -EOPNOTSUPP;
2449 if (dev->do_ioctl) {
2450 if (netif_device_present(dev))
2451 err = dev->do_ioctl(dev, ifr,
2452 cmd);
2453 else
2454 err = -ENODEV;
2455 }
2456 } else
2457 err = -EINVAL;
2458
2459 }
2460 return err;
2461}
2462
2463/*
2464 * This function handles all "interface"-type I/O control requests. The actual
2465 * 'doing' part of this is dev_ifsioc above.
2466 */
2467
2468/**
2469 * dev_ioctl - network device ioctl
2470 * @cmd: command to issue
2471 * @arg: pointer to a struct ifreq in user space
2472 *
2473 * Issue ioctl functions to devices. This is normally called by the
2474 * user space syscall interfaces but can sometimes be useful for
2475 * other purposes. The return value is the return from the syscall if
2476 * positive or a negative errno code on error.
2477 */
2478
2479int dev_ioctl(unsigned int cmd, void __user *arg)
2480{
2481 struct ifreq ifr;
2482 int ret;
2483 char *colon;
2484
2485 /* One special case: SIOCGIFCONF takes ifconf argument
2486 and requires shared lock, because it sleeps writing
2487 to user space.
2488 */
2489
2490 if (cmd == SIOCGIFCONF) {
2491 rtnl_shlock();
2492 ret = dev_ifconf((char __user *) arg);
2493 rtnl_shunlock();
2494 return ret;
2495 }
2496 if (cmd == SIOCGIFNAME)
2497 return dev_ifname((struct ifreq __user *)arg);
2498
2499 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2500 return -EFAULT;
2501
2502 ifr.ifr_name[IFNAMSIZ-1] = 0;
2503
2504 colon = strchr(ifr.ifr_name, ':');
2505 if (colon)
2506 *colon = 0;
2507
2508 /*
2509 * See which interface the caller is talking about.
2510 */
2511
2512 switch (cmd) {
2513 /*
2514 * These ioctl calls:
2515 * - can be done by all.
2516 * - atomic and do not require locking.
2517 * - return a value
2518 */
2519 case SIOCGIFFLAGS:
2520 case SIOCGIFMETRIC:
2521 case SIOCGIFMTU:
2522 case SIOCGIFHWADDR:
2523 case SIOCGIFSLAVE:
2524 case SIOCGIFMAP:
2525 case SIOCGIFINDEX:
2526 case SIOCGIFTXQLEN:
2527 dev_load(ifr.ifr_name);
2528 read_lock(&dev_base_lock);
2529 ret = dev_ifsioc(&ifr, cmd);
2530 read_unlock(&dev_base_lock);
2531 if (!ret) {
2532 if (colon)
2533 *colon = ':';
2534 if (copy_to_user(arg, &ifr,
2535 sizeof(struct ifreq)))
2536 ret = -EFAULT;
2537 }
2538 return ret;
2539
2540 case SIOCETHTOOL:
2541 dev_load(ifr.ifr_name);
2542 rtnl_lock();
2543 ret = dev_ethtool(&ifr);
2544 rtnl_unlock();
2545 if (!ret) {
2546 if (colon)
2547 *colon = ':';
2548 if (copy_to_user(arg, &ifr,
2549 sizeof(struct ifreq)))
2550 ret = -EFAULT;
2551 }
2552 return ret;
2553
2554 /*
2555 * These ioctl calls:
2556 * - require superuser power.
2557 * - require strict serialization.
2558 * - return a value
2559 */
2560 case SIOCGMIIPHY:
2561 case SIOCGMIIREG:
2562 case SIOCSIFNAME:
2563 if (!capable(CAP_NET_ADMIN))
2564 return -EPERM;
2565 dev_load(ifr.ifr_name);
2566 rtnl_lock();
2567 ret = dev_ifsioc(&ifr, cmd);
2568 rtnl_unlock();
2569 if (!ret) {
2570 if (colon)
2571 *colon = ':';
2572 if (copy_to_user(arg, &ifr,
2573 sizeof(struct ifreq)))
2574 ret = -EFAULT;
2575 }
2576 return ret;
2577
2578 /*
2579 * These ioctl calls:
2580 * - require superuser power.
2581 * - require strict serialization.
2582 * - do not return a value
2583 */
2584 case SIOCSIFFLAGS:
2585 case SIOCSIFMETRIC:
2586 case SIOCSIFMTU:
2587 case SIOCSIFMAP:
2588 case SIOCSIFHWADDR:
2589 case SIOCSIFSLAVE:
2590 case SIOCADDMULTI:
2591 case SIOCDELMULTI:
2592 case SIOCSIFHWBROADCAST:
2593 case SIOCSIFTXQLEN:
2594 case SIOCSMIIREG:
2595 case SIOCBONDENSLAVE:
2596 case SIOCBONDRELEASE:
2597 case SIOCBONDSETHWADDR:
2598 case SIOCBONDSLAVEINFOQUERY:
2599 case SIOCBONDINFOQUERY:
2600 case SIOCBONDCHANGEACTIVE:
2601 case SIOCBRADDIF:
2602 case SIOCBRDELIF:
2603 if (!capable(CAP_NET_ADMIN))
2604 return -EPERM;
2605 dev_load(ifr.ifr_name);
2606 rtnl_lock();
2607 ret = dev_ifsioc(&ifr, cmd);
2608 rtnl_unlock();
2609 return ret;
2610
2611 case SIOCGIFMEM:
2612 /* Get the per device memory space. We can add this but
2613 * currently do not support it */
2614 case SIOCSIFMEM:
2615 /* Set the per device memory buffer space.
2616 * Not applicable in our case */
2617 case SIOCSIFLINK:
2618 return -EINVAL;
2619
2620 /*
2621 * Unknown or private ioctl.
2622 */
2623 default:
2624 if (cmd == SIOCWANDEV ||
2625 (cmd >= SIOCDEVPRIVATE &&
2626 cmd <= SIOCDEVPRIVATE + 15)) {
2627 dev_load(ifr.ifr_name);
2628 rtnl_lock();
2629 ret = dev_ifsioc(&ifr, cmd);
2630 rtnl_unlock();
2631 if (!ret && copy_to_user(arg, &ifr,
2632 sizeof(struct ifreq)))
2633 ret = -EFAULT;
2634 return ret;
2635 }
2636#ifdef WIRELESS_EXT
2637 /* Take care of Wireless Extensions */
2638 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2639 /* If command is `set a parameter', or
2640 * `get the encoding parameters', check if
2641 * the user has the right to do it */
2642 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2643 if (!capable(CAP_NET_ADMIN))
2644 return -EPERM;
2645 }
2646 dev_load(ifr.ifr_name);
2647 rtnl_lock();
2648 /* Follow me in net/core/wireless.c */
2649 ret = wireless_process_ioctl(&ifr, cmd);
2650 rtnl_unlock();
2651 if (IW_IS_GET(cmd) &&
2652 copy_to_user(arg, &ifr,
2653 sizeof(struct ifreq)))
2654 ret = -EFAULT;
2655 return ret;
2656 }
2657#endif /* WIRELESS_EXT */
2658 return -EINVAL;
2659 }
2660}
2661
2662
2663/**
2664 * dev_new_index - allocate an ifindex
2665 *
2666 * Returns a suitable unique value for a new device interface
2667 * number. The caller must hold the rtnl semaphore or the
2668 * dev_base_lock to be sure it remains unique.
2669 */
2670static int dev_new_index(void)
2671{
2672 static int ifindex;
2673 for (;;) {
2674 if (++ifindex <= 0)
2675 ifindex = 1;
2676 if (!__dev_get_by_index(ifindex))
2677 return ifindex;
2678 }
2679}
2680
2681static int dev_boot_phase = 1;
2682
2683/* Delayed registration/unregisteration */
2684static DEFINE_SPINLOCK(net_todo_list_lock);
2685static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2686
2687static inline void net_set_todo(struct net_device *dev)
2688{
2689 spin_lock(&net_todo_list_lock);
2690 list_add_tail(&dev->todo_list, &net_todo_list);
2691 spin_unlock(&net_todo_list_lock);
2692}
2693
2694/**
2695 * register_netdevice - register a network device
2696 * @dev: device to register
2697 *
2698 * Take a completed network device structure and add it to the kernel
2699 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2700 * chain. 0 is returned on success. A negative errno code is returned
2701 * on a failure to set up the device, or if the name is a duplicate.
2702 *
2703 * Callers must hold the rtnl semaphore. You may want
2704 * register_netdev() instead of this.
2705 *
2706 * BUGS:
2707 * The locking appears insufficient to guarantee two parallel registers
2708 * will not get the same name.
2709 */
2710
2711int register_netdevice(struct net_device *dev)
2712{
2713 struct hlist_head *head;
2714 struct hlist_node *p;
2715 int ret;
2716
2717 BUG_ON(dev_boot_phase);
2718 ASSERT_RTNL();
2719
2720 /* When net_device's are persistent, this will be fatal. */
2721 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2722
2723 spin_lock_init(&dev->queue_lock);
2724 spin_lock_init(&dev->xmit_lock);
2725 dev->xmit_lock_owner = -1;
2726#ifdef CONFIG_NET_CLS_ACT
2727 spin_lock_init(&dev->ingress_lock);
2728#endif
2729
2730 ret = alloc_divert_blk(dev);
2731 if (ret)
2732 goto out;
2733
2734 dev->iflink = -1;
2735
2736 /* Init, if this function is available */
2737 if (dev->init) {
2738 ret = dev->init(dev);
2739 if (ret) {
2740 if (ret > 0)
2741 ret = -EIO;
2742 goto out_err;
2743 }
2744 }
2745
2746 if (!dev_valid_name(dev->name)) {
2747 ret = -EINVAL;
2748 goto out_err;
2749 }
2750
2751 dev->ifindex = dev_new_index();
2752 if (dev->iflink == -1)
2753 dev->iflink = dev->ifindex;
2754
2755 /* Check for existence of name */
2756 head = dev_name_hash(dev->name);
2757 hlist_for_each(p, head) {
2758 struct net_device *d
2759 = hlist_entry(p, struct net_device, name_hlist);
2760 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2761 ret = -EEXIST;
2762 goto out_err;
2763 }
2764 }
2765
2766 /* Fix illegal SG+CSUM combinations. */
2767 if ((dev->features & NETIF_F_SG) &&
2768 !(dev->features & (NETIF_F_IP_CSUM |
2769 NETIF_F_NO_CSUM |
2770 NETIF_F_HW_CSUM))) {
2771 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2772 dev->name);
2773 dev->features &= ~NETIF_F_SG;
2774 }
2775
2776 /* TSO requires that SG is present as well. */
2777 if ((dev->features & NETIF_F_TSO) &&
2778 !(dev->features & NETIF_F_SG)) {
2779 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2780 dev->name);
2781 dev->features &= ~NETIF_F_TSO;
2782 }
2783
2784 /*
2785 * nil rebuild_header routine,
2786 * that should be never called and used as just bug trap.
2787 */
2788
2789 if (!dev->rebuild_header)
2790 dev->rebuild_header = default_rebuild_header;
2791
2792 /*
2793 * Default initial state at registry is that the
2794 * device is present.
2795 */
2796
2797 set_bit(__LINK_STATE_PRESENT, &dev->state);
2798
2799 dev->next = NULL;
2800 dev_init_scheduler(dev);
2801 write_lock_bh(&dev_base_lock);
2802 *dev_tail = dev;
2803 dev_tail = &dev->next;
2804 hlist_add_head(&dev->name_hlist, head);
2805 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2806 dev_hold(dev);
2807 dev->reg_state = NETREG_REGISTERING;
2808 write_unlock_bh(&dev_base_lock);
2809
2810 /* Notify protocols, that a new device appeared. */
2811 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2812
2813 /* Finish registration after unlock */
2814 net_set_todo(dev);
2815 ret = 0;
2816
2817out:
2818 return ret;
2819out_err:
2820 free_divert_blk(dev);
2821 goto out;
2822}
2823
2824/**
2825 * register_netdev - register a network device
2826 * @dev: device to register
2827 *
2828 * Take a completed network device structure and add it to the kernel
2829 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2830 * chain. 0 is returned on success. A negative errno code is returned
2831 * on a failure to set up the device, or if the name is a duplicate.
2832 *
2833 * This is a wrapper around register_netdev that takes the rtnl semaphore
2834 * and expands the device name if you passed a format string to
2835 * alloc_netdev.
2836 */
2837int register_netdev(struct net_device *dev)
2838{
2839 int err;
2840
2841 rtnl_lock();
2842
2843 /*
2844 * If the name is a format string the caller wants us to do a
2845 * name allocation.
2846 */
2847 if (strchr(dev->name, '%')) {
2848 err = dev_alloc_name(dev, dev->name);
2849 if (err < 0)
2850 goto out;
2851 }
2852
2853 /*
2854 * Back compatibility hook. Kill this one in 2.5
2855 */
2856 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2857 err = dev_alloc_name(dev, "eth%d");
2858 if (err < 0)
2859 goto out;
2860 }
2861
2862 err = register_netdevice(dev);
2863out:
2864 rtnl_unlock();
2865 return err;
2866}
2867EXPORT_SYMBOL(register_netdev);
2868
2869/*
2870 * netdev_wait_allrefs - wait until all references are gone.
2871 *
2872 * This is called when unregistering network devices.
2873 *
2874 * Any protocol or device that holds a reference should register
2875 * for netdevice notification, and cleanup and put back the
2876 * reference if they receive an UNREGISTER event.
2877 * We can get stuck here if buggy protocols don't correctly
2878 * call dev_put.
2879 */
2880static void netdev_wait_allrefs(struct net_device *dev)
2881{
2882 unsigned long rebroadcast_time, warning_time;
2883
2884 rebroadcast_time = warning_time = jiffies;
2885 while (atomic_read(&dev->refcnt) != 0) {
2886 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2887 rtnl_shlock();
2888
2889 /* Rebroadcast unregister notification */
2890 notifier_call_chain(&netdev_chain,
2891 NETDEV_UNREGISTER, dev);
2892
2893 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2894 &dev->state)) {
2895 /* We must not have linkwatch events
2896 * pending on unregister. If this
2897 * happens, we simply run the queue
2898 * unscheduled, resulting in a noop
2899 * for this device.
2900 */
2901 linkwatch_run_queue();
2902 }
2903
2904 rtnl_shunlock();
2905
2906 rebroadcast_time = jiffies;
2907 }
2908
2909 msleep(250);
2910
2911 if (time_after(jiffies, warning_time + 10 * HZ)) {
2912 printk(KERN_EMERG "unregister_netdevice: "
2913 "waiting for %s to become free. Usage "
2914 "count = %d\n",
2915 dev->name, atomic_read(&dev->refcnt));
2916 warning_time = jiffies;
2917 }
2918 }
2919}
2920
2921/* The sequence is:
2922 *
2923 * rtnl_lock();
2924 * ...
2925 * register_netdevice(x1);
2926 * register_netdevice(x2);
2927 * ...
2928 * unregister_netdevice(y1);
2929 * unregister_netdevice(y2);
2930 * ...
2931 * rtnl_unlock();
2932 * free_netdev(y1);
2933 * free_netdev(y2);
2934 *
2935 * We are invoked by rtnl_unlock() after it drops the semaphore.
2936 * This allows us to deal with problems:
2937 * 1) We can create/delete sysfs objects which invoke hotplug
2938 * without deadlocking with linkwatch via keventd.
2939 * 2) Since we run with the RTNL semaphore not held, we can sleep
2940 * safely in order to wait for the netdev refcnt to drop to zero.
2941 */
2942static DECLARE_MUTEX(net_todo_run_mutex);
2943void netdev_run_todo(void)
2944{
2945 struct list_head list = LIST_HEAD_INIT(list);
2946 int err;
2947
2948
2949 /* Need to guard against multiple cpu's getting out of order. */
2950 down(&net_todo_run_mutex);
2951
2952 /* Not safe to do outside the semaphore. We must not return
2953 * until all unregister events invoked by the local processor
2954 * have been completed (either by this todo run, or one on
2955 * another cpu).
2956 */
2957 if (list_empty(&net_todo_list))
2958 goto out;
2959
2960 /* Snapshot list, allow later requests */
2961 spin_lock(&net_todo_list_lock);
2962 list_splice_init(&net_todo_list, &list);
2963 spin_unlock(&net_todo_list_lock);
2964
2965 while (!list_empty(&list)) {
2966 struct net_device *dev
2967 = list_entry(list.next, struct net_device, todo_list);
2968 list_del(&dev->todo_list);
2969
2970 switch(dev->reg_state) {
2971 case NETREG_REGISTERING:
2972 err = netdev_register_sysfs(dev);
2973 if (err)
2974 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2975 dev->name, err);
2976 dev->reg_state = NETREG_REGISTERED;
2977 break;
2978
2979 case NETREG_UNREGISTERING:
2980 netdev_unregister_sysfs(dev);
2981 dev->reg_state = NETREG_UNREGISTERED;
2982
2983 netdev_wait_allrefs(dev);
2984
2985 /* paranoia */
2986 BUG_ON(atomic_read(&dev->refcnt));
2987 BUG_TRAP(!dev->ip_ptr);
2988 BUG_TRAP(!dev->ip6_ptr);
2989 BUG_TRAP(!dev->dn_ptr);
2990
2991
2992 /* It must be the very last action,
2993 * after this 'dev' may point to freed up memory.
2994 */
2995 if (dev->destructor)
2996 dev->destructor(dev);
2997 break;
2998
2999 default:
3000 printk(KERN_ERR "network todo '%s' but state %d\n",
3001 dev->name, dev->reg_state);
3002 break;
3003 }
3004 }
3005
3006out:
3007 up(&net_todo_run_mutex);
3008}
3009
3010/**
3011 * alloc_netdev - allocate network device
3012 * @sizeof_priv: size of private data to allocate space for
3013 * @name: device name format string
3014 * @setup: callback to initialize device
3015 *
3016 * Allocates a struct net_device with private data area for driver use
3017 * and performs basic initialization.
3018 */
3019struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3020 void (*setup)(struct net_device *))
3021{
3022 void *p;
3023 struct net_device *dev;
3024 int alloc_size;
3025
3026 /* ensure 32-byte alignment of both the device and private area */
3027 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3028 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3029
3030 p = kmalloc(alloc_size, GFP_KERNEL);
3031 if (!p) {
3032 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3033 return NULL;
3034 }
3035 memset(p, 0, alloc_size);
3036
3037 dev = (struct net_device *)
3038 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3039 dev->padded = (char *)dev - (char *)p;
3040
3041 if (sizeof_priv)
3042 dev->priv = netdev_priv(dev);
3043
3044 setup(dev);
3045 strcpy(dev->name, name);
3046 return dev;
3047}
3048EXPORT_SYMBOL(alloc_netdev);
3049
3050/**
3051 * free_netdev - free network device
3052 * @dev: device
3053 *
3054 * This function does the last stage of destroying an allocated device
3055 * interface. The reference to the device object is released.
3056 * If this is the last reference then it will be freed.
3057 */
3058void free_netdev(struct net_device *dev)
3059{
3060#ifdef CONFIG_SYSFS
3061 /* Compatiablity with error handling in drivers */
3062 if (dev->reg_state == NETREG_UNINITIALIZED) {
3063 kfree((char *)dev - dev->padded);
3064 return;
3065 }
3066
3067 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3068 dev->reg_state = NETREG_RELEASED;
3069
3070 /* will free via class release */
3071 class_device_put(&dev->class_dev);
3072#else
3073 kfree((char *)dev - dev->padded);
3074#endif
3075}
3076
3077/* Synchronize with packet receive processing. */
3078void synchronize_net(void)
3079{
3080 might_sleep();
3081 synchronize_kernel();
3082}
3083
3084/**
3085 * unregister_netdevice - remove device from the kernel
3086 * @dev: device
3087 *
3088 * This function shuts down a device interface and removes it
3089 * from the kernel tables. On success 0 is returned, on a failure
3090 * a negative errno code is returned.
3091 *
3092 * Callers must hold the rtnl semaphore. You may want
3093 * unregister_netdev() instead of this.
3094 */
3095
3096int unregister_netdevice(struct net_device *dev)
3097{
3098 struct net_device *d, **dp;
3099
3100 BUG_ON(dev_boot_phase);
3101 ASSERT_RTNL();
3102
3103 /* Some devices call without registering for initialization unwind. */
3104 if (dev->reg_state == NETREG_UNINITIALIZED) {
3105 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3106 "was registered\n", dev->name, dev);
3107 return -ENODEV;
3108 }
3109
3110 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3111
3112 /* If device is running, close it first. */
3113 if (dev->flags & IFF_UP)
3114 dev_close(dev);
3115
3116 /* And unlink it from device chain. */
3117 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3118 if (d == dev) {
3119 write_lock_bh(&dev_base_lock);
3120 hlist_del(&dev->name_hlist);
3121 hlist_del(&dev->index_hlist);
3122 if (dev_tail == &dev->next)
3123 dev_tail = dp;
3124 *dp = d->next;
3125 write_unlock_bh(&dev_base_lock);
3126 break;
3127 }
3128 }
3129 if (!d) {
3130 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3131 dev->name);
3132 return -ENODEV;
3133 }
3134
3135 dev->reg_state = NETREG_UNREGISTERING;
3136
3137 synchronize_net();
3138
3139 /* Shutdown queueing discipline. */
3140 dev_shutdown(dev);
3141
3142
3143 /* Notify protocols, that we are about to destroy
3144 this device. They should clean all the things.
3145 */
3146 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3147
3148 /*
3149 * Flush the multicast chain
3150 */
3151 dev_mc_discard(dev);
3152
3153 if (dev->uninit)
3154 dev->uninit(dev);
3155
3156 /* Notifier chain MUST detach us from master device. */
3157 BUG_TRAP(!dev->master);
3158
3159 free_divert_blk(dev);
3160
3161 /* Finish processing unregister after unlock */
3162 net_set_todo(dev);
3163
3164 synchronize_net();
3165
3166 dev_put(dev);
3167 return 0;
3168}
3169
3170/**
3171 * unregister_netdev - remove device from the kernel
3172 * @dev: device
3173 *
3174 * This function shuts down a device interface and removes it
3175 * from the kernel tables. On success 0 is returned, on a failure
3176 * a negative errno code is returned.
3177 *
3178 * This is just a wrapper for unregister_netdevice that takes
3179 * the rtnl semaphore. In general you want to use this and not
3180 * unregister_netdevice.
3181 */
3182void unregister_netdev(struct net_device *dev)
3183{
3184 rtnl_lock();
3185 unregister_netdevice(dev);
3186 rtnl_unlock();
3187}
3188
3189EXPORT_SYMBOL(unregister_netdev);
3190
3191#ifdef CONFIG_HOTPLUG_CPU
3192static int dev_cpu_callback(struct notifier_block *nfb,
3193 unsigned long action,
3194 void *ocpu)
3195{
3196 struct sk_buff **list_skb;
3197 struct net_device **list_net;
3198 struct sk_buff *skb;
3199 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3200 struct softnet_data *sd, *oldsd;
3201
3202 if (action != CPU_DEAD)
3203 return NOTIFY_OK;
3204
3205 local_irq_disable();
3206 cpu = smp_processor_id();
3207 sd = &per_cpu(softnet_data, cpu);
3208 oldsd = &per_cpu(softnet_data, oldcpu);
3209
3210 /* Find end of our completion_queue. */
3211 list_skb = &sd->completion_queue;
3212 while (*list_skb)
3213 list_skb = &(*list_skb)->next;
3214 /* Append completion queue from offline CPU. */
3215 *list_skb = oldsd->completion_queue;
3216 oldsd->completion_queue = NULL;
3217
3218 /* Find end of our output_queue. */
3219 list_net = &sd->output_queue;
3220 while (*list_net)
3221 list_net = &(*list_net)->next_sched;
3222 /* Append output queue from offline CPU. */
3223 *list_net = oldsd->output_queue;
3224 oldsd->output_queue = NULL;
3225
3226 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3227 local_irq_enable();
3228
3229 /* Process offline CPU's input_pkt_queue */
3230 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3231 netif_rx(skb);
3232
3233 return NOTIFY_OK;
3234}
3235#endif /* CONFIG_HOTPLUG_CPU */
3236
3237
3238/*
3239 * Initialize the DEV module. At boot time this walks the device list and
3240 * unhooks any devices that fail to initialise (normally hardware not
3241 * present) and leaves us with a valid list of present and active devices.
3242 *
3243 */
3244
3245/*
3246 * This is called single threaded during boot, so no need
3247 * to take the rtnl semaphore.
3248 */
3249static int __init net_dev_init(void)
3250{
3251 int i, rc = -ENOMEM;
3252
3253 BUG_ON(!dev_boot_phase);
3254
3255 net_random_init();
3256
3257 if (dev_proc_init())
3258 goto out;
3259
3260 if (netdev_sysfs_init())
3261 goto out;
3262
3263 INIT_LIST_HEAD(&ptype_all);
3264 for (i = 0; i < 16; i++)
3265 INIT_LIST_HEAD(&ptype_base[i]);
3266
3267 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3268 INIT_HLIST_HEAD(&dev_name_head[i]);
3269
3270 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3271 INIT_HLIST_HEAD(&dev_index_head[i]);
3272
3273 /*
3274 * Initialise the packet receive queues.
3275 */
3276
3277 for (i = 0; i < NR_CPUS; i++) {
3278 struct softnet_data *queue;
3279
3280 queue = &per_cpu(softnet_data, i);
3281 skb_queue_head_init(&queue->input_pkt_queue);
3282 queue->throttle = 0;
3283 queue->cng_level = 0;
3284 queue->avg_blog = 10; /* arbitrary non-zero */
3285 queue->completion_queue = NULL;
3286 INIT_LIST_HEAD(&queue->poll_list);
3287 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3288 queue->backlog_dev.weight = weight_p;
3289 queue->backlog_dev.poll = process_backlog;
3290 atomic_set(&queue->backlog_dev.refcnt, 1);
3291 }
3292
3293#ifdef OFFLINE_SAMPLE
3294 samp_timer.expires = jiffies + (10 * HZ);
3295 add_timer(&samp_timer);
3296#endif
3297
3298 dev_boot_phase = 0;
3299
3300 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3301 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3302
3303 hotcpu_notifier(dev_cpu_callback, 0);
3304 dst_init();
3305 dev_mcast_init();
3306 rc = 0;
3307out:
3308 return rc;
3309}
3310
3311subsys_initcall(net_dev_init);
3312
3313EXPORT_SYMBOL(__dev_get_by_index);
3314EXPORT_SYMBOL(__dev_get_by_name);
3315EXPORT_SYMBOL(__dev_remove_pack);
3316EXPORT_SYMBOL(__skb_linearize);
3317EXPORT_SYMBOL(dev_add_pack);
3318EXPORT_SYMBOL(dev_alloc_name);
3319EXPORT_SYMBOL(dev_close);
3320EXPORT_SYMBOL(dev_get_by_flags);
3321EXPORT_SYMBOL(dev_get_by_index);
3322EXPORT_SYMBOL(dev_get_by_name);
3323EXPORT_SYMBOL(dev_ioctl);
3324EXPORT_SYMBOL(dev_open);
3325EXPORT_SYMBOL(dev_queue_xmit);
3326EXPORT_SYMBOL(dev_remove_pack);
3327EXPORT_SYMBOL(dev_set_allmulti);
3328EXPORT_SYMBOL(dev_set_promiscuity);
3329EXPORT_SYMBOL(dev_change_flags);
3330EXPORT_SYMBOL(dev_set_mtu);
3331EXPORT_SYMBOL(dev_set_mac_address);
3332EXPORT_SYMBOL(free_netdev);
3333EXPORT_SYMBOL(netdev_boot_setup_check);
3334EXPORT_SYMBOL(netdev_set_master);
3335EXPORT_SYMBOL(netdev_state_change);
3336EXPORT_SYMBOL(netif_receive_skb);
3337EXPORT_SYMBOL(netif_rx);
3338EXPORT_SYMBOL(register_gifconf);
3339EXPORT_SYMBOL(register_netdevice);
3340EXPORT_SYMBOL(register_netdevice_notifier);
3341EXPORT_SYMBOL(skb_checksum_help);
3342EXPORT_SYMBOL(synchronize_net);
3343EXPORT_SYMBOL(unregister_netdevice);
3344EXPORT_SYMBOL(unregister_netdevice_notifier);
3345EXPORT_SYMBOL(net_enable_timestamp);
3346EXPORT_SYMBOL(net_disable_timestamp);
3347EXPORT_SYMBOL(dev_get_flags);
3348
3349#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3350EXPORT_SYMBOL(br_handle_frame_hook);
3351EXPORT_SYMBOL(br_fdb_get_hook);
3352EXPORT_SYMBOL(br_fdb_put_hook);
3353#endif
3354
3355#ifdef CONFIG_KMOD
3356EXPORT_SYMBOL(dev_load);
3357#endif
3358
3359EXPORT_PER_CPU_SYMBOL(softnet_data);