blob: 4f1ae2efe8724d7114af60ae4ca5d1d12eda5bf2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * NET3 Protocol independent device support routines.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Derived from the non IP parts of dev.c 1.0.19
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 *
14 * Additional Authors:
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
21 *
22 * Changes:
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
34 * drivers
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
44 * call a packet.
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
58 * 1 device.
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
66 * the backlog queue.
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
73 */
74
75#include <asm/uaccess.h>
76#include <asm/system.h>
77#include <linux/bitops.h>
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/**
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700764 * netdev_features_change - device changes fatures
765 * @dev: device to cause notification
766 *
767 * Called to indicate a device has changed features.
768 */
769void netdev_features_change(struct net_device *dev)
770{
771 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
772}
773EXPORT_SYMBOL(netdev_features_change);
774
775/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 * netdev_state_change - device changes state
777 * @dev: device to cause notification
778 *
779 * Called to indicate a device has changed state. This function calls
780 * the notifier chains for netdev_chain and sends a NEWLINK message
781 * to the routing socket.
782 */
783void netdev_state_change(struct net_device *dev)
784{
785 if (dev->flags & IFF_UP) {
786 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
787 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
788 }
789}
790
791/**
792 * dev_load - load a network module
793 * @name: name of interface
794 *
795 * If a network interface is not present and the process has suitable
796 * privileges this function loads the module. If module loading is not
797 * available in this kernel then it becomes a nop.
798 */
799
800void dev_load(const char *name)
801{
802 struct net_device *dev;
803
804 read_lock(&dev_base_lock);
805 dev = __dev_get_by_name(name);
806 read_unlock(&dev_base_lock);
807
808 if (!dev && capable(CAP_SYS_MODULE))
809 request_module("%s", name);
810}
811
812static int default_rebuild_header(struct sk_buff *skb)
813{
814 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
815 skb->dev ? skb->dev->name : "NULL!!!");
816 kfree_skb(skb);
817 return 1;
818}
819
820
821/**
822 * dev_open - prepare an interface for use.
823 * @dev: device to open
824 *
825 * Takes a device from down to up state. The device's private open
826 * function is invoked and then the multicast lists are loaded. Finally
827 * the device is moved into the up state and a %NETDEV_UP message is
828 * sent to the netdev notifier chain.
829 *
830 * Calling this function on an active interface is a nop. On a failure
831 * a negative errno code is returned.
832 */
833int dev_open(struct net_device *dev)
834{
835 int ret = 0;
836
837 /*
838 * Is it already up?
839 */
840
841 if (dev->flags & IFF_UP)
842 return 0;
843
844 /*
845 * Is it even present?
846 */
847 if (!netif_device_present(dev))
848 return -ENODEV;
849
850 /*
851 * Call device private open method
852 */
853 set_bit(__LINK_STATE_START, &dev->state);
854 if (dev->open) {
855 ret = dev->open(dev);
856 if (ret)
857 clear_bit(__LINK_STATE_START, &dev->state);
858 }
859
860 /*
861 * If it went open OK then:
862 */
863
864 if (!ret) {
865 /*
866 * Set the flags.
867 */
868 dev->flags |= IFF_UP;
869
870 /*
871 * Initialize multicasting status
872 */
873 dev_mc_upload(dev);
874
875 /*
876 * Wakeup transmit queue engine
877 */
878 dev_activate(dev);
879
880 /*
881 * ... and announce new interface.
882 */
883 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
884 }
885 return ret;
886}
887
888/**
889 * dev_close - shutdown an interface.
890 * @dev: device to shutdown
891 *
892 * This function moves an active device into down state. A
893 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
894 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
895 * chain.
896 */
897int dev_close(struct net_device *dev)
898{
899 if (!(dev->flags & IFF_UP))
900 return 0;
901
902 /*
903 * Tell people we are going down, so that they can
904 * prepare to death, when device is still operating.
905 */
906 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
907
908 dev_deactivate(dev);
909
910 clear_bit(__LINK_STATE_START, &dev->state);
911
912 /* Synchronize to scheduled poll. We cannot touch poll list,
913 * it can be even on different cpu. So just clear netif_running(),
914 * and wait when poll really will happen. Actually, the best place
915 * for this is inside dev->stop() after device stopped its irq
916 * engine, but this requires more changes in devices. */
917
918 smp_mb__after_clear_bit(); /* Commit netif_running(). */
919 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
920 /* No hurry. */
921 current->state = TASK_INTERRUPTIBLE;
922 schedule_timeout(1);
923 }
924
925 /*
926 * Call the device specific close. This cannot fail.
927 * Only if device is UP
928 *
929 * We allow it to be called even after a DETACH hot-plug
930 * event.
931 */
932 if (dev->stop)
933 dev->stop(dev);
934
935 /*
936 * Device is now down.
937 */
938
939 dev->flags &= ~IFF_UP;
940
941 /*
942 * Tell people we are down
943 */
944 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
945
946 return 0;
947}
948
949
950/*
951 * Device change register/unregister. These are not inline or static
952 * as we export them to the world.
953 */
954
955/**
956 * register_netdevice_notifier - register a network notifier block
957 * @nb: notifier
958 *
959 * Register a notifier to be called when network device events occur.
960 * The notifier passed is linked into the kernel structures and must
961 * not be reused until it has been unregistered. A negative errno code
962 * is returned on a failure.
963 *
964 * When registered all registration and up events are replayed
965 * to the new notifier to allow device to have a race free
966 * view of the network device list.
967 */
968
969int register_netdevice_notifier(struct notifier_block *nb)
970{
971 struct net_device *dev;
972 int err;
973
974 rtnl_lock();
975 err = notifier_chain_register(&netdev_chain, nb);
976 if (!err) {
977 for (dev = dev_base; dev; dev = dev->next) {
978 nb->notifier_call(nb, NETDEV_REGISTER, dev);
979
980 if (dev->flags & IFF_UP)
981 nb->notifier_call(nb, NETDEV_UP, dev);
982 }
983 }
984 rtnl_unlock();
985 return err;
986}
987
988/**
989 * unregister_netdevice_notifier - unregister a network notifier block
990 * @nb: notifier
991 *
992 * Unregister a notifier previously registered by
993 * register_netdevice_notifier(). The notifier is unlinked into the
994 * kernel structures and may then be reused. A negative errno code
995 * is returned on a failure.
996 */
997
998int unregister_netdevice_notifier(struct notifier_block *nb)
999{
1000 return notifier_chain_unregister(&netdev_chain, nb);
1001}
1002
1003/**
1004 * call_netdevice_notifiers - call all network notifier blocks
1005 * @val: value passed unmodified to notifier function
1006 * @v: pointer passed unmodified to notifier function
1007 *
1008 * Call all network notifier blocks. Parameters and return value
1009 * are as for notifier_call_chain().
1010 */
1011
1012int call_netdevice_notifiers(unsigned long val, void *v)
1013{
1014 return notifier_call_chain(&netdev_chain, val, v);
1015}
1016
1017/* When > 0 there are consumers of rx skb time stamps */
1018static atomic_t netstamp_needed = ATOMIC_INIT(0);
1019
1020void net_enable_timestamp(void)
1021{
1022 atomic_inc(&netstamp_needed);
1023}
1024
1025void net_disable_timestamp(void)
1026{
1027 atomic_dec(&netstamp_needed);
1028}
1029
1030static inline void net_timestamp(struct timeval *stamp)
1031{
1032 if (atomic_read(&netstamp_needed))
1033 do_gettimeofday(stamp);
1034 else {
1035 stamp->tv_sec = 0;
1036 stamp->tv_usec = 0;
1037 }
1038}
1039
1040/*
1041 * Support routine. Sends outgoing frames to any network
1042 * taps currently in use.
1043 */
1044
1045void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1046{
1047 struct packet_type *ptype;
1048 net_timestamp(&skb->stamp);
1049
1050 rcu_read_lock();
1051 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1052 /* Never send packets back to the socket
1053 * they originated from - MvS (miquels@drinkel.ow.org)
1054 */
1055 if ((ptype->dev == dev || !ptype->dev) &&
1056 (ptype->af_packet_priv == NULL ||
1057 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1058 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1059 if (!skb2)
1060 break;
1061
1062 /* skb->nh should be correctly
1063 set by sender, so that the second statement is
1064 just protection against buggy protocols.
1065 */
1066 skb2->mac.raw = skb2->data;
1067
1068 if (skb2->nh.raw < skb2->data ||
1069 skb2->nh.raw > skb2->tail) {
1070 if (net_ratelimit())
1071 printk(KERN_CRIT "protocol %04x is "
1072 "buggy, dev %s\n",
1073 skb2->protocol, dev->name);
1074 skb2->nh.raw = skb2->data;
1075 }
1076
1077 skb2->h.raw = skb2->nh.raw;
1078 skb2->pkt_type = PACKET_OUTGOING;
1079 ptype->func(skb2, skb->dev, ptype);
1080 }
1081 }
1082 rcu_read_unlock();
1083}
1084
1085/*
1086 * Invalidate hardware checksum when packet is to be mangled, and
1087 * complete checksum manually on outgoing path.
1088 */
1089int skb_checksum_help(struct sk_buff *skb, int inward)
1090{
1091 unsigned int csum;
1092 int ret = 0, offset = skb->h.raw - skb->data;
1093
1094 if (inward) {
1095 skb->ip_summed = CHECKSUM_NONE;
1096 goto out;
1097 }
1098
1099 if (skb_cloned(skb)) {
1100 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1101 if (ret)
1102 goto out;
1103 }
1104
1105 if (offset > (int)skb->len)
1106 BUG();
1107 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1108
1109 offset = skb->tail - skb->h.raw;
1110 if (offset <= 0)
1111 BUG();
1112 if (skb->csum + 2 > offset)
1113 BUG();
1114
1115 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1116 skb->ip_summed = CHECKSUM_NONE;
1117out:
1118 return ret;
1119}
1120
1121#ifdef CONFIG_HIGHMEM
1122/* Actually, we should eliminate this check as soon as we know, that:
1123 * 1. IOMMU is present and allows to map all the memory.
1124 * 2. No high memory really exists on this machine.
1125 */
1126
1127static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1128{
1129 int i;
1130
1131 if (dev->features & NETIF_F_HIGHDMA)
1132 return 0;
1133
1134 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1135 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1136 return 1;
1137
1138 return 0;
1139}
1140#else
1141#define illegal_highdma(dev, skb) (0)
1142#endif
1143
1144extern void skb_release_data(struct sk_buff *);
1145
1146/* Keep head the same: replace data */
1147int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1148{
1149 unsigned int size;
1150 u8 *data;
1151 long offset;
1152 struct skb_shared_info *ninfo;
1153 int headerlen = skb->data - skb->head;
1154 int expand = (skb->tail + skb->data_len) - skb->end;
1155
1156 if (skb_shared(skb))
1157 BUG();
1158
1159 if (expand <= 0)
1160 expand = 0;
1161
1162 size = skb->end - skb->head + expand;
1163 size = SKB_DATA_ALIGN(size);
1164 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1165 if (!data)
1166 return -ENOMEM;
1167
1168 /* Copy entire thing */
1169 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1170 BUG();
1171
1172 /* Set up shinfo */
1173 ninfo = (struct skb_shared_info*)(data + size);
1174 atomic_set(&ninfo->dataref, 1);
1175 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1176 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1177 ninfo->nr_frags = 0;
1178 ninfo->frag_list = NULL;
1179
1180 /* Offset between the two in bytes */
1181 offset = data - skb->head;
1182
1183 /* Free old data. */
1184 skb_release_data(skb);
1185
1186 skb->head = data;
1187 skb->end = data + size;
1188
1189 /* Set up new pointers */
1190 skb->h.raw += offset;
1191 skb->nh.raw += offset;
1192 skb->mac.raw += offset;
1193 skb->tail += offset;
1194 skb->data += offset;
1195
1196 /* We are no longer a clone, even if we were. */
1197 skb->cloned = 0;
1198
1199 skb->tail += skb->data_len;
1200 skb->data_len = 0;
1201 return 0;
1202}
1203
1204#define HARD_TX_LOCK(dev, cpu) { \
1205 if ((dev->features & NETIF_F_LLTX) == 0) { \
1206 spin_lock(&dev->xmit_lock); \
1207 dev->xmit_lock_owner = cpu; \
1208 } \
1209}
1210
1211#define HARD_TX_UNLOCK(dev) { \
1212 if ((dev->features & NETIF_F_LLTX) == 0) { \
1213 dev->xmit_lock_owner = -1; \
1214 spin_unlock(&dev->xmit_lock); \
1215 } \
1216}
1217
1218/**
1219 * dev_queue_xmit - transmit a buffer
1220 * @skb: buffer to transmit
1221 *
1222 * Queue a buffer for transmission to a network device. The caller must
1223 * have set the device and priority and built the buffer before calling
1224 * this function. The function can be called from an interrupt.
1225 *
1226 * A negative errno code is returned on a failure. A success does not
1227 * guarantee the frame will be transmitted as it may be dropped due
1228 * to congestion or traffic shaping.
Ben Greearaf191362005-04-24 20:12:36 -07001229 *
1230 * -----------------------------------------------------------------------------------
1231 * I notice this method can also return errors from the queue disciplines,
1232 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1233 * be positive.
1234 *
1235 * Regardless of the return value, the skb is consumed, so it is currently
1236 * difficult to retry a send to this method. (You can bump the ref count
1237 * before sending to hold a reference for retry if you are careful.)
1238 *
1239 * When calling this method, interrupts MUST be enabled. This is because
1240 * the BH enable code must have IRQs enabled so that it will not deadlock.
1241 * --BLG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 */
1243
1244int dev_queue_xmit(struct sk_buff *skb)
1245{
1246 struct net_device *dev = skb->dev;
1247 struct Qdisc *q;
1248 int rc = -ENOMEM;
1249
1250 if (skb_shinfo(skb)->frag_list &&
1251 !(dev->features & NETIF_F_FRAGLIST) &&
1252 __skb_linearize(skb, GFP_ATOMIC))
1253 goto out_kfree_skb;
1254
1255 /* Fragmented skb is linearized if device does not support SG,
1256 * or if at least one of fragments is in highmem and device
1257 * does not support DMA from it.
1258 */
1259 if (skb_shinfo(skb)->nr_frags &&
1260 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1261 __skb_linearize(skb, GFP_ATOMIC))
1262 goto out_kfree_skb;
1263
1264 /* If packet is not checksummed and device does not support
1265 * checksumming for this protocol, complete checksumming here.
1266 */
1267 if (skb->ip_summed == CHECKSUM_HW &&
1268 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1269 (!(dev->features & NETIF_F_IP_CSUM) ||
1270 skb->protocol != htons(ETH_P_IP))))
1271 if (skb_checksum_help(skb, 0))
1272 goto out_kfree_skb;
1273
1274 /* Disable soft irqs for various locks below. Also
1275 * stops preemption for RCU.
1276 */
1277 local_bh_disable();
1278
1279 /* Updates of qdisc are serialized by queue_lock.
1280 * The struct Qdisc which is pointed to by qdisc is now a
1281 * rcu structure - it may be accessed without acquiring
1282 * a lock (but the structure may be stale.) The freeing of the
1283 * qdisc will be deferred until it's known that there are no
1284 * more references to it.
1285 *
1286 * If the qdisc has an enqueue function, we still need to
1287 * hold the queue_lock before calling it, since queue_lock
1288 * also serializes access to the device queue.
1289 */
1290
1291 q = rcu_dereference(dev->qdisc);
1292#ifdef CONFIG_NET_CLS_ACT
1293 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1294#endif
1295 if (q->enqueue) {
1296 /* Grab device queue */
1297 spin_lock(&dev->queue_lock);
1298
1299 rc = q->enqueue(skb, q);
1300
1301 qdisc_run(dev);
1302
1303 spin_unlock(&dev->queue_lock);
1304 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1305 goto out;
1306 }
1307
1308 /* The device has no queue. Common case for software devices:
1309 loopback, all the sorts of tunnels...
1310
1311 Really, it is unlikely that xmit_lock protection is necessary here.
1312 (f.e. loopback and IP tunnels are clean ignoring statistics
1313 counters.)
1314 However, it is possible, that they rely on protection
1315 made by us here.
1316
1317 Check this and shot the lock. It is not prone from deadlocks.
1318 Either shot noqueue qdisc, it is even simpler 8)
1319 */
1320 if (dev->flags & IFF_UP) {
1321 int cpu = smp_processor_id(); /* ok because BHs are off */
1322
1323 if (dev->xmit_lock_owner != cpu) {
1324
1325 HARD_TX_LOCK(dev, cpu);
1326
1327 if (!netif_queue_stopped(dev)) {
1328 if (netdev_nit)
1329 dev_queue_xmit_nit(skb, dev);
1330
1331 rc = 0;
1332 if (!dev->hard_start_xmit(skb, dev)) {
1333 HARD_TX_UNLOCK(dev);
1334 goto out;
1335 }
1336 }
1337 HARD_TX_UNLOCK(dev);
1338 if (net_ratelimit())
1339 printk(KERN_CRIT "Virtual device %s asks to "
1340 "queue packet!\n", dev->name);
1341 } else {
1342 /* Recursion is detected! It is possible,
1343 * unfortunately */
1344 if (net_ratelimit())
1345 printk(KERN_CRIT "Dead loop on virtual device "
1346 "%s, fix it urgently!\n", dev->name);
1347 }
1348 }
1349
1350 rc = -ENETDOWN;
1351 local_bh_enable();
1352
1353out_kfree_skb:
1354 kfree_skb(skb);
1355 return rc;
1356out:
1357 local_bh_enable();
1358 return rc;
1359}
1360
1361
1362/*=======================================================================
1363 Receiver routines
1364 =======================================================================*/
1365
1366int netdev_max_backlog = 300;
1367int weight_p = 64; /* old backlog weight */
1368/* These numbers are selected based on intuition and some
1369 * experimentatiom, if you have more scientific way of doing this
1370 * please go ahead and fix things.
1371 */
1372int no_cong_thresh = 10;
1373int no_cong = 20;
1374int lo_cong = 100;
1375int mod_cong = 290;
1376
1377DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1378
1379
1380static void get_sample_stats(int cpu)
1381{
1382#ifdef RAND_LIE
1383 unsigned long rd;
1384 int rq;
1385#endif
1386 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1387 int blog = sd->input_pkt_queue.qlen;
1388 int avg_blog = sd->avg_blog;
1389
1390 avg_blog = (avg_blog >> 1) + (blog >> 1);
1391
1392 if (avg_blog > mod_cong) {
1393 /* Above moderate congestion levels. */
1394 sd->cng_level = NET_RX_CN_HIGH;
1395#ifdef RAND_LIE
1396 rd = net_random();
1397 rq = rd % netdev_max_backlog;
1398 if (rq < avg_blog) /* unlucky bastard */
1399 sd->cng_level = NET_RX_DROP;
1400#endif
1401 } else if (avg_blog > lo_cong) {
1402 sd->cng_level = NET_RX_CN_MOD;
1403#ifdef RAND_LIE
1404 rd = net_random();
1405 rq = rd % netdev_max_backlog;
1406 if (rq < avg_blog) /* unlucky bastard */
1407 sd->cng_level = NET_RX_CN_HIGH;
1408#endif
1409 } else if (avg_blog > no_cong)
1410 sd->cng_level = NET_RX_CN_LOW;
1411 else /* no congestion */
1412 sd->cng_level = NET_RX_SUCCESS;
1413
1414 sd->avg_blog = avg_blog;
1415}
1416
1417#ifdef OFFLINE_SAMPLE
1418static void sample_queue(unsigned long dummy)
1419{
1420/* 10 ms 0r 1ms -- i don't care -- JHS */
1421 int next_tick = 1;
1422 int cpu = smp_processor_id();
1423
1424 get_sample_stats(cpu);
1425 next_tick += jiffies;
1426 mod_timer(&samp_timer, next_tick);
1427}
1428#endif
1429
1430
1431/**
1432 * netif_rx - post buffer to the network code
1433 * @skb: buffer to post
1434 *
1435 * This function receives a packet from a device driver and queues it for
1436 * the upper (protocol) levels to process. It always succeeds. The buffer
1437 * may be dropped during processing for congestion control or by the
1438 * protocol layers.
1439 *
1440 * return values:
1441 * NET_RX_SUCCESS (no congestion)
1442 * NET_RX_CN_LOW (low congestion)
1443 * NET_RX_CN_MOD (moderate congestion)
1444 * NET_RX_CN_HIGH (high congestion)
1445 * NET_RX_DROP (packet was dropped)
1446 *
1447 */
1448
1449int netif_rx(struct sk_buff *skb)
1450{
1451 int this_cpu;
1452 struct softnet_data *queue;
1453 unsigned long flags;
1454
1455 /* if netpoll wants it, pretend we never saw it */
1456 if (netpoll_rx(skb))
1457 return NET_RX_DROP;
1458
1459 if (!skb->stamp.tv_sec)
1460 net_timestamp(&skb->stamp);
1461
1462 /*
1463 * The code is rearranged so that the path is the most
1464 * short when CPU is congested, but is still operating.
1465 */
1466 local_irq_save(flags);
1467 this_cpu = smp_processor_id();
1468 queue = &__get_cpu_var(softnet_data);
1469
1470 __get_cpu_var(netdev_rx_stat).total++;
1471 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1472 if (queue->input_pkt_queue.qlen) {
1473 if (queue->throttle)
1474 goto drop;
1475
1476enqueue:
1477 dev_hold(skb->dev);
1478 __skb_queue_tail(&queue->input_pkt_queue, skb);
1479#ifndef OFFLINE_SAMPLE
1480 get_sample_stats(this_cpu);
1481#endif
1482 local_irq_restore(flags);
1483 return queue->cng_level;
1484 }
1485
1486 if (queue->throttle)
1487 queue->throttle = 0;
1488
1489 netif_rx_schedule(&queue->backlog_dev);
1490 goto enqueue;
1491 }
1492
1493 if (!queue->throttle) {
1494 queue->throttle = 1;
1495 __get_cpu_var(netdev_rx_stat).throttled++;
1496 }
1497
1498drop:
1499 __get_cpu_var(netdev_rx_stat).dropped++;
1500 local_irq_restore(flags);
1501
1502 kfree_skb(skb);
1503 return NET_RX_DROP;
1504}
1505
1506int netif_rx_ni(struct sk_buff *skb)
1507{
1508 int err;
1509
1510 preempt_disable();
1511 err = netif_rx(skb);
1512 if (local_softirq_pending())
1513 do_softirq();
1514 preempt_enable();
1515
1516 return err;
1517}
1518
1519EXPORT_SYMBOL(netif_rx_ni);
1520
1521static __inline__ void skb_bond(struct sk_buff *skb)
1522{
1523 struct net_device *dev = skb->dev;
1524
1525 if (dev->master) {
1526 skb->real_dev = skb->dev;
1527 skb->dev = dev->master;
1528 }
1529}
1530
1531static void net_tx_action(struct softirq_action *h)
1532{
1533 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1534
1535 if (sd->completion_queue) {
1536 struct sk_buff *clist;
1537
1538 local_irq_disable();
1539 clist = sd->completion_queue;
1540 sd->completion_queue = NULL;
1541 local_irq_enable();
1542
1543 while (clist) {
1544 struct sk_buff *skb = clist;
1545 clist = clist->next;
1546
1547 BUG_TRAP(!atomic_read(&skb->users));
1548 __kfree_skb(skb);
1549 }
1550 }
1551
1552 if (sd->output_queue) {
1553 struct net_device *head;
1554
1555 local_irq_disable();
1556 head = sd->output_queue;
1557 sd->output_queue = NULL;
1558 local_irq_enable();
1559
1560 while (head) {
1561 struct net_device *dev = head;
1562 head = head->next_sched;
1563
1564 smp_mb__before_clear_bit();
1565 clear_bit(__LINK_STATE_SCHED, &dev->state);
1566
1567 if (spin_trylock(&dev->queue_lock)) {
1568 qdisc_run(dev);
1569 spin_unlock(&dev->queue_lock);
1570 } else {
1571 netif_schedule(dev);
1572 }
1573 }
1574 }
1575}
1576
1577static __inline__ int deliver_skb(struct sk_buff *skb,
1578 struct packet_type *pt_prev)
1579{
1580 atomic_inc(&skb->users);
1581 return pt_prev->func(skb, skb->dev, pt_prev);
1582}
1583
1584#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1585int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1586struct net_bridge;
1587struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1588 unsigned char *addr);
1589void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1590
1591static __inline__ int handle_bridge(struct sk_buff **pskb,
1592 struct packet_type **pt_prev, int *ret)
1593{
1594 struct net_bridge_port *port;
1595
1596 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1597 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1598 return 0;
1599
1600 if (*pt_prev) {
1601 *ret = deliver_skb(*pskb, *pt_prev);
1602 *pt_prev = NULL;
1603 }
1604
1605 return br_handle_frame_hook(port, pskb);
1606}
1607#else
1608#define handle_bridge(skb, pt_prev, ret) (0)
1609#endif
1610
1611#ifdef CONFIG_NET_CLS_ACT
1612/* TODO: Maybe we should just force sch_ingress to be compiled in
1613 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1614 * a compare and 2 stores extra right now if we dont have it on
1615 * but have CONFIG_NET_CLS_ACT
1616 * NOTE: This doesnt stop any functionality; if you dont have
1617 * the ingress scheduler, you just cant add policies on ingress.
1618 *
1619 */
1620static int ing_filter(struct sk_buff *skb)
1621{
1622 struct Qdisc *q;
1623 struct net_device *dev = skb->dev;
1624 int result = TC_ACT_OK;
1625
1626 if (dev->qdisc_ingress) {
1627 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1628 if (MAX_RED_LOOP < ttl++) {
1629 printk("Redir loop detected Dropping packet (%s->%s)\n",
1630 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1631 return TC_ACT_SHOT;
1632 }
1633
1634 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1635
1636 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1637 if (NULL == skb->input_dev) {
1638 skb->input_dev = skb->dev;
1639 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1640 }
1641 spin_lock(&dev->ingress_lock);
1642 if ((q = dev->qdisc_ingress) != NULL)
1643 result = q->enqueue(skb, q);
1644 spin_unlock(&dev->ingress_lock);
1645
1646 }
1647
1648 return result;
1649}
1650#endif
1651
1652int netif_receive_skb(struct sk_buff *skb)
1653{
1654 struct packet_type *ptype, *pt_prev;
1655 int ret = NET_RX_DROP;
1656 unsigned short type;
1657
1658 /* if we've gotten here through NAPI, check netpoll */
1659 if (skb->dev->poll && netpoll_rx(skb))
1660 return NET_RX_DROP;
1661
1662 if (!skb->stamp.tv_sec)
1663 net_timestamp(&skb->stamp);
1664
1665 skb_bond(skb);
1666
1667 __get_cpu_var(netdev_rx_stat).total++;
1668
1669 skb->h.raw = skb->nh.raw = skb->data;
1670 skb->mac_len = skb->nh.raw - skb->mac.raw;
1671
1672 pt_prev = NULL;
1673
1674 rcu_read_lock();
1675
1676#ifdef CONFIG_NET_CLS_ACT
1677 if (skb->tc_verd & TC_NCLS) {
1678 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1679 goto ncls;
1680 }
1681#endif
1682
1683 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1684 if (!ptype->dev || ptype->dev == skb->dev) {
1685 if (pt_prev)
1686 ret = deliver_skb(skb, pt_prev);
1687 pt_prev = ptype;
1688 }
1689 }
1690
1691#ifdef CONFIG_NET_CLS_ACT
1692 if (pt_prev) {
1693 ret = deliver_skb(skb, pt_prev);
1694 pt_prev = NULL; /* noone else should process this after*/
1695 } else {
1696 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1697 }
1698
1699 ret = ing_filter(skb);
1700
1701 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1702 kfree_skb(skb);
1703 goto out;
1704 }
1705
1706 skb->tc_verd = 0;
1707ncls:
1708#endif
1709
1710 handle_diverter(skb);
1711
1712 if (handle_bridge(&skb, &pt_prev, &ret))
1713 goto out;
1714
1715 type = skb->protocol;
1716 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1717 if (ptype->type == type &&
1718 (!ptype->dev || ptype->dev == skb->dev)) {
1719 if (pt_prev)
1720 ret = deliver_skb(skb, pt_prev);
1721 pt_prev = ptype;
1722 }
1723 }
1724
1725 if (pt_prev) {
1726 ret = pt_prev->func(skb, skb->dev, pt_prev);
1727 } else {
1728 kfree_skb(skb);
1729 /* Jamal, now you will not able to escape explaining
1730 * me how you were going to use this. :-)
1731 */
1732 ret = NET_RX_DROP;
1733 }
1734
1735out:
1736 rcu_read_unlock();
1737 return ret;
1738}
1739
1740static int process_backlog(struct net_device *backlog_dev, int *budget)
1741{
1742 int work = 0;
1743 int quota = min(backlog_dev->quota, *budget);
1744 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1745 unsigned long start_time = jiffies;
1746
Stephen Hemmingere3876602005-06-08 14:56:01 -07001747 backlog_dev->weight = weight_p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 for (;;) {
1749 struct sk_buff *skb;
1750 struct net_device *dev;
1751
1752 local_irq_disable();
1753 skb = __skb_dequeue(&queue->input_pkt_queue);
1754 if (!skb)
1755 goto job_done;
1756 local_irq_enable();
1757
1758 dev = skb->dev;
1759
1760 netif_receive_skb(skb);
1761
1762 dev_put(dev);
1763
1764 work++;
1765
1766 if (work >= quota || jiffies - start_time > 1)
1767 break;
1768
1769 }
1770
1771 backlog_dev->quota -= work;
1772 *budget -= work;
1773 return -1;
1774
1775job_done:
1776 backlog_dev->quota -= work;
1777 *budget -= work;
1778
1779 list_del(&backlog_dev->poll_list);
1780 smp_mb__before_clear_bit();
1781 netif_poll_enable(backlog_dev);
1782
1783 if (queue->throttle)
1784 queue->throttle = 0;
1785 local_irq_enable();
1786 return 0;
1787}
1788
1789static void net_rx_action(struct softirq_action *h)
1790{
1791 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1792 unsigned long start_time = jiffies;
1793 int budget = netdev_max_backlog;
1794
1795
1796 local_irq_disable();
1797
1798 while (!list_empty(&queue->poll_list)) {
1799 struct net_device *dev;
1800
1801 if (budget <= 0 || jiffies - start_time > 1)
1802 goto softnet_break;
1803
1804 local_irq_enable();
1805
1806 dev = list_entry(queue->poll_list.next,
1807 struct net_device, poll_list);
1808 netpoll_poll_lock(dev);
1809
1810 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1811 netpoll_poll_unlock(dev);
1812 local_irq_disable();
1813 list_del(&dev->poll_list);
1814 list_add_tail(&dev->poll_list, &queue->poll_list);
1815 if (dev->quota < 0)
1816 dev->quota += dev->weight;
1817 else
1818 dev->quota = dev->weight;
1819 } else {
1820 netpoll_poll_unlock(dev);
1821 dev_put(dev);
1822 local_irq_disable();
1823 }
1824 }
1825out:
1826 local_irq_enable();
1827 return;
1828
1829softnet_break:
1830 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1831 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1832 goto out;
1833}
1834
1835static gifconf_func_t * gifconf_list [NPROTO];
1836
1837/**
1838 * register_gifconf - register a SIOCGIF handler
1839 * @family: Address family
1840 * @gifconf: Function handler
1841 *
1842 * Register protocol dependent address dumping routines. The handler
1843 * that is passed must not be freed or reused until it has been replaced
1844 * by another handler.
1845 */
1846int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1847{
1848 if (family >= NPROTO)
1849 return -EINVAL;
1850 gifconf_list[family] = gifconf;
1851 return 0;
1852}
1853
1854
1855/*
1856 * Map an interface index to its name (SIOCGIFNAME)
1857 */
1858
1859/*
1860 * We need this ioctl for efficient implementation of the
1861 * if_indextoname() function required by the IPv6 API. Without
1862 * it, we would have to search all the interfaces to find a
1863 * match. --pb
1864 */
1865
1866static int dev_ifname(struct ifreq __user *arg)
1867{
1868 struct net_device *dev;
1869 struct ifreq ifr;
1870
1871 /*
1872 * Fetch the caller's info block.
1873 */
1874
1875 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1876 return -EFAULT;
1877
1878 read_lock(&dev_base_lock);
1879 dev = __dev_get_by_index(ifr.ifr_ifindex);
1880 if (!dev) {
1881 read_unlock(&dev_base_lock);
1882 return -ENODEV;
1883 }
1884
1885 strcpy(ifr.ifr_name, dev->name);
1886 read_unlock(&dev_base_lock);
1887
1888 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1889 return -EFAULT;
1890 return 0;
1891}
1892
1893/*
1894 * Perform a SIOCGIFCONF call. This structure will change
1895 * size eventually, and there is nothing I can do about it.
1896 * Thus we will need a 'compatibility mode'.
1897 */
1898
1899static int dev_ifconf(char __user *arg)
1900{
1901 struct ifconf ifc;
1902 struct net_device *dev;
1903 char __user *pos;
1904 int len;
1905 int total;
1906 int i;
1907
1908 /*
1909 * Fetch the caller's info block.
1910 */
1911
1912 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1913 return -EFAULT;
1914
1915 pos = ifc.ifc_buf;
1916 len = ifc.ifc_len;
1917
1918 /*
1919 * Loop over the interfaces, and write an info block for each.
1920 */
1921
1922 total = 0;
1923 for (dev = dev_base; dev; dev = dev->next) {
1924 for (i = 0; i < NPROTO; i++) {
1925 if (gifconf_list[i]) {
1926 int done;
1927 if (!pos)
1928 done = gifconf_list[i](dev, NULL, 0);
1929 else
1930 done = gifconf_list[i](dev, pos + total,
1931 len - total);
1932 if (done < 0)
1933 return -EFAULT;
1934 total += done;
1935 }
1936 }
1937 }
1938
1939 /*
1940 * All done. Write the updated control block back to the caller.
1941 */
1942 ifc.ifc_len = total;
1943
1944 /*
1945 * Both BSD and Solaris return 0 here, so we do too.
1946 */
1947 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1948}
1949
1950#ifdef CONFIG_PROC_FS
1951/*
1952 * This is invoked by the /proc filesystem handler to display a device
1953 * in detail.
1954 */
1955static __inline__ struct net_device *dev_get_idx(loff_t pos)
1956{
1957 struct net_device *dev;
1958 loff_t i;
1959
1960 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1961
1962 return i == pos ? dev : NULL;
1963}
1964
1965void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1966{
1967 read_lock(&dev_base_lock);
1968 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1969}
1970
1971void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1972{
1973 ++*pos;
1974 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1975}
1976
1977void dev_seq_stop(struct seq_file *seq, void *v)
1978{
1979 read_unlock(&dev_base_lock);
1980}
1981
1982static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1983{
1984 if (dev->get_stats) {
1985 struct net_device_stats *stats = dev->get_stats(dev);
1986
1987 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1988 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1989 dev->name, stats->rx_bytes, stats->rx_packets,
1990 stats->rx_errors,
1991 stats->rx_dropped + stats->rx_missed_errors,
1992 stats->rx_fifo_errors,
1993 stats->rx_length_errors + stats->rx_over_errors +
1994 stats->rx_crc_errors + stats->rx_frame_errors,
1995 stats->rx_compressed, stats->multicast,
1996 stats->tx_bytes, stats->tx_packets,
1997 stats->tx_errors, stats->tx_dropped,
1998 stats->tx_fifo_errors, stats->collisions,
1999 stats->tx_carrier_errors +
2000 stats->tx_aborted_errors +
2001 stats->tx_window_errors +
2002 stats->tx_heartbeat_errors,
2003 stats->tx_compressed);
2004 } else
2005 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2006}
2007
2008/*
2009 * Called from the PROCfs module. This now uses the new arbitrary sized
2010 * /proc/net interface to create /proc/net/dev
2011 */
2012static int dev_seq_show(struct seq_file *seq, void *v)
2013{
2014 if (v == SEQ_START_TOKEN)
2015 seq_puts(seq, "Inter-| Receive "
2016 " | Transmit\n"
2017 " face |bytes packets errs drop fifo frame "
2018 "compressed multicast|bytes packets errs "
2019 "drop fifo colls carrier compressed\n");
2020 else
2021 dev_seq_printf_stats(seq, v);
2022 return 0;
2023}
2024
2025static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2026{
2027 struct netif_rx_stats *rc = NULL;
2028
2029 while (*pos < NR_CPUS)
2030 if (cpu_online(*pos)) {
2031 rc = &per_cpu(netdev_rx_stat, *pos);
2032 break;
2033 } else
2034 ++*pos;
2035 return rc;
2036}
2037
2038static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2039{
2040 return softnet_get_online(pos);
2041}
2042
2043static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2044{
2045 ++*pos;
2046 return softnet_get_online(pos);
2047}
2048
2049static void softnet_seq_stop(struct seq_file *seq, void *v)
2050{
2051}
2052
2053static int softnet_seq_show(struct seq_file *seq, void *v)
2054{
2055 struct netif_rx_stats *s = v;
2056
2057 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2058 s->total, s->dropped, s->time_squeeze, s->throttled,
Stephen Hemmingerc1ebcdb2005-06-23 20:08:59 -07002059 0, 0, 0, 0, /* was fastroute */
2060 s->cpu_collision );
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 return 0;
2062}
2063
2064static struct seq_operations dev_seq_ops = {
2065 .start = dev_seq_start,
2066 .next = dev_seq_next,
2067 .stop = dev_seq_stop,
2068 .show = dev_seq_show,
2069};
2070
2071static int dev_seq_open(struct inode *inode, struct file *file)
2072{
2073 return seq_open(file, &dev_seq_ops);
2074}
2075
2076static struct file_operations dev_seq_fops = {
2077 .owner = THIS_MODULE,
2078 .open = dev_seq_open,
2079 .read = seq_read,
2080 .llseek = seq_lseek,
2081 .release = seq_release,
2082};
2083
2084static struct seq_operations softnet_seq_ops = {
2085 .start = softnet_seq_start,
2086 .next = softnet_seq_next,
2087 .stop = softnet_seq_stop,
2088 .show = softnet_seq_show,
2089};
2090
2091static int softnet_seq_open(struct inode *inode, struct file *file)
2092{
2093 return seq_open(file, &softnet_seq_ops);
2094}
2095
2096static struct file_operations softnet_seq_fops = {
2097 .owner = THIS_MODULE,
2098 .open = softnet_seq_open,
2099 .read = seq_read,
2100 .llseek = seq_lseek,
2101 .release = seq_release,
2102};
2103
2104#ifdef WIRELESS_EXT
2105extern int wireless_proc_init(void);
2106#else
2107#define wireless_proc_init() 0
2108#endif
2109
2110static int __init dev_proc_init(void)
2111{
2112 int rc = -ENOMEM;
2113
2114 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2115 goto out;
2116 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2117 goto out_dev;
2118 if (wireless_proc_init())
2119 goto out_softnet;
2120 rc = 0;
2121out:
2122 return rc;
2123out_softnet:
2124 proc_net_remove("softnet_stat");
2125out_dev:
2126 proc_net_remove("dev");
2127 goto out;
2128}
2129#else
2130#define dev_proc_init() 0
2131#endif /* CONFIG_PROC_FS */
2132
2133
2134/**
2135 * netdev_set_master - set up master/slave pair
2136 * @slave: slave device
2137 * @master: new master device
2138 *
2139 * Changes the master device of the slave. Pass %NULL to break the
2140 * bonding. The caller must hold the RTNL semaphore. On a failure
2141 * a negative errno code is returned. On success the reference counts
2142 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2143 * function returns zero.
2144 */
2145int netdev_set_master(struct net_device *slave, struct net_device *master)
2146{
2147 struct net_device *old = slave->master;
2148
2149 ASSERT_RTNL();
2150
2151 if (master) {
2152 if (old)
2153 return -EBUSY;
2154 dev_hold(master);
2155 }
2156
2157 slave->master = master;
2158
2159 synchronize_net();
2160
2161 if (old)
2162 dev_put(old);
2163
2164 if (master)
2165 slave->flags |= IFF_SLAVE;
2166 else
2167 slave->flags &= ~IFF_SLAVE;
2168
2169 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2170 return 0;
2171}
2172
2173/**
2174 * dev_set_promiscuity - update promiscuity count on a device
2175 * @dev: device
2176 * @inc: modifier
2177 *
2178 * Add or remove promsicuity from a device. While the count in the device
2179 * remains above zero the interface remains promiscuous. Once it hits zero
2180 * the device reverts back to normal filtering operation. A negative inc
2181 * value is used to drop promiscuity on the device.
2182 */
2183void dev_set_promiscuity(struct net_device *dev, int inc)
2184{
2185 unsigned short old_flags = dev->flags;
2186
2187 dev->flags |= IFF_PROMISC;
2188 if ((dev->promiscuity += inc) == 0)
2189 dev->flags &= ~IFF_PROMISC;
2190 if (dev->flags ^ old_flags) {
2191 dev_mc_upload(dev);
2192 printk(KERN_INFO "device %s %s promiscuous mode\n",
2193 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2194 "left");
2195 }
2196}
2197
2198/**
2199 * dev_set_allmulti - update allmulti count on a device
2200 * @dev: device
2201 * @inc: modifier
2202 *
2203 * Add or remove reception of all multicast frames to a device. While the
2204 * count in the device remains above zero the interface remains listening
2205 * to all interfaces. Once it hits zero the device reverts back to normal
2206 * filtering operation. A negative @inc value is used to drop the counter
2207 * when releasing a resource needing all multicasts.
2208 */
2209
2210void dev_set_allmulti(struct net_device *dev, int inc)
2211{
2212 unsigned short old_flags = dev->flags;
2213
2214 dev->flags |= IFF_ALLMULTI;
2215 if ((dev->allmulti += inc) == 0)
2216 dev->flags &= ~IFF_ALLMULTI;
2217 if (dev->flags ^ old_flags)
2218 dev_mc_upload(dev);
2219}
2220
2221unsigned dev_get_flags(const struct net_device *dev)
2222{
2223 unsigned flags;
2224
2225 flags = (dev->flags & ~(IFF_PROMISC |
2226 IFF_ALLMULTI |
2227 IFF_RUNNING)) |
2228 (dev->gflags & (IFF_PROMISC |
2229 IFF_ALLMULTI));
2230
2231 if (netif_running(dev) && netif_carrier_ok(dev))
2232 flags |= IFF_RUNNING;
2233
2234 return flags;
2235}
2236
2237int dev_change_flags(struct net_device *dev, unsigned flags)
2238{
2239 int ret;
2240 int old_flags = dev->flags;
2241
2242 /*
2243 * Set the flags on our device.
2244 */
2245
2246 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2247 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2248 IFF_AUTOMEDIA)) |
2249 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2250 IFF_ALLMULTI));
2251
2252 /*
2253 * Load in the correct multicast list now the flags have changed.
2254 */
2255
2256 dev_mc_upload(dev);
2257
2258 /*
2259 * Have we downed the interface. We handle IFF_UP ourselves
2260 * according to user attempts to set it, rather than blindly
2261 * setting it.
2262 */
2263
2264 ret = 0;
2265 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2266 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2267
2268 if (!ret)
2269 dev_mc_upload(dev);
2270 }
2271
2272 if (dev->flags & IFF_UP &&
2273 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2274 IFF_VOLATILE)))
2275 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2276
2277 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2278 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2279 dev->gflags ^= IFF_PROMISC;
2280 dev_set_promiscuity(dev, inc);
2281 }
2282
2283 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2284 is important. Some (broken) drivers set IFF_PROMISC, when
2285 IFF_ALLMULTI is requested not asking us and not reporting.
2286 */
2287 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2288 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2289 dev->gflags ^= IFF_ALLMULTI;
2290 dev_set_allmulti(dev, inc);
2291 }
2292
2293 if (old_flags ^ dev->flags)
2294 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2295
2296 return ret;
2297}
2298
2299int dev_set_mtu(struct net_device *dev, int new_mtu)
2300{
2301 int err;
2302
2303 if (new_mtu == dev->mtu)
2304 return 0;
2305
2306 /* MTU must be positive. */
2307 if (new_mtu < 0)
2308 return -EINVAL;
2309
2310 if (!netif_device_present(dev))
2311 return -ENODEV;
2312
2313 err = 0;
2314 if (dev->change_mtu)
2315 err = dev->change_mtu(dev, new_mtu);
2316 else
2317 dev->mtu = new_mtu;
2318 if (!err && dev->flags & IFF_UP)
2319 notifier_call_chain(&netdev_chain,
2320 NETDEV_CHANGEMTU, dev);
2321 return err;
2322}
2323
2324int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2325{
2326 int err;
2327
2328 if (!dev->set_mac_address)
2329 return -EOPNOTSUPP;
2330 if (sa->sa_family != dev->type)
2331 return -EINVAL;
2332 if (!netif_device_present(dev))
2333 return -ENODEV;
2334 err = dev->set_mac_address(dev, sa);
2335 if (!err)
2336 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2337 return err;
2338}
2339
2340/*
2341 * Perform the SIOCxIFxxx calls.
2342 */
2343static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2344{
2345 int err;
2346 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2347
2348 if (!dev)
2349 return -ENODEV;
2350
2351 switch (cmd) {
2352 case SIOCGIFFLAGS: /* Get interface flags */
2353 ifr->ifr_flags = dev_get_flags(dev);
2354 return 0;
2355
2356 case SIOCSIFFLAGS: /* Set interface flags */
2357 return dev_change_flags(dev, ifr->ifr_flags);
2358
2359 case SIOCGIFMETRIC: /* Get the metric on the interface
2360 (currently unused) */
2361 ifr->ifr_metric = 0;
2362 return 0;
2363
2364 case SIOCSIFMETRIC: /* Set the metric on the interface
2365 (currently unused) */
2366 return -EOPNOTSUPP;
2367
2368 case SIOCGIFMTU: /* Get the MTU of a device */
2369 ifr->ifr_mtu = dev->mtu;
2370 return 0;
2371
2372 case SIOCSIFMTU: /* Set the MTU of a device */
2373 return dev_set_mtu(dev, ifr->ifr_mtu);
2374
2375 case SIOCGIFHWADDR:
2376 if (!dev->addr_len)
2377 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2378 else
2379 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2380 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2381 ifr->ifr_hwaddr.sa_family = dev->type;
2382 return 0;
2383
2384 case SIOCSIFHWADDR:
2385 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2386
2387 case SIOCSIFHWBROADCAST:
2388 if (ifr->ifr_hwaddr.sa_family != dev->type)
2389 return -EINVAL;
2390 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2391 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2392 notifier_call_chain(&netdev_chain,
2393 NETDEV_CHANGEADDR, dev);
2394 return 0;
2395
2396 case SIOCGIFMAP:
2397 ifr->ifr_map.mem_start = dev->mem_start;
2398 ifr->ifr_map.mem_end = dev->mem_end;
2399 ifr->ifr_map.base_addr = dev->base_addr;
2400 ifr->ifr_map.irq = dev->irq;
2401 ifr->ifr_map.dma = dev->dma;
2402 ifr->ifr_map.port = dev->if_port;
2403 return 0;
2404
2405 case SIOCSIFMAP:
2406 if (dev->set_config) {
2407 if (!netif_device_present(dev))
2408 return -ENODEV;
2409 return dev->set_config(dev, &ifr->ifr_map);
2410 }
2411 return -EOPNOTSUPP;
2412
2413 case SIOCADDMULTI:
2414 if (!dev->set_multicast_list ||
2415 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2416 return -EINVAL;
2417 if (!netif_device_present(dev))
2418 return -ENODEV;
2419 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2420 dev->addr_len, 1);
2421
2422 case SIOCDELMULTI:
2423 if (!dev->set_multicast_list ||
2424 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2425 return -EINVAL;
2426 if (!netif_device_present(dev))
2427 return -ENODEV;
2428 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2429 dev->addr_len, 1);
2430
2431 case SIOCGIFINDEX:
2432 ifr->ifr_ifindex = dev->ifindex;
2433 return 0;
2434
2435 case SIOCGIFTXQLEN:
2436 ifr->ifr_qlen = dev->tx_queue_len;
2437 return 0;
2438
2439 case SIOCSIFTXQLEN:
2440 if (ifr->ifr_qlen < 0)
2441 return -EINVAL;
2442 dev->tx_queue_len = ifr->ifr_qlen;
2443 return 0;
2444
2445 case SIOCSIFNAME:
2446 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2447 return dev_change_name(dev, ifr->ifr_newname);
2448
2449 /*
2450 * Unknown or private ioctl
2451 */
2452
2453 default:
2454 if ((cmd >= SIOCDEVPRIVATE &&
2455 cmd <= SIOCDEVPRIVATE + 15) ||
2456 cmd == SIOCBONDENSLAVE ||
2457 cmd == SIOCBONDRELEASE ||
2458 cmd == SIOCBONDSETHWADDR ||
2459 cmd == SIOCBONDSLAVEINFOQUERY ||
2460 cmd == SIOCBONDINFOQUERY ||
2461 cmd == SIOCBONDCHANGEACTIVE ||
2462 cmd == SIOCGMIIPHY ||
2463 cmd == SIOCGMIIREG ||
2464 cmd == SIOCSMIIREG ||
2465 cmd == SIOCBRADDIF ||
2466 cmd == SIOCBRDELIF ||
2467 cmd == SIOCWANDEV) {
2468 err = -EOPNOTSUPP;
2469 if (dev->do_ioctl) {
2470 if (netif_device_present(dev))
2471 err = dev->do_ioctl(dev, ifr,
2472 cmd);
2473 else
2474 err = -ENODEV;
2475 }
2476 } else
2477 err = -EINVAL;
2478
2479 }
2480 return err;
2481}
2482
2483/*
2484 * This function handles all "interface"-type I/O control requests. The actual
2485 * 'doing' part of this is dev_ifsioc above.
2486 */
2487
2488/**
2489 * dev_ioctl - network device ioctl
2490 * @cmd: command to issue
2491 * @arg: pointer to a struct ifreq in user space
2492 *
2493 * Issue ioctl functions to devices. This is normally called by the
2494 * user space syscall interfaces but can sometimes be useful for
2495 * other purposes. The return value is the return from the syscall if
2496 * positive or a negative errno code on error.
2497 */
2498
2499int dev_ioctl(unsigned int cmd, void __user *arg)
2500{
2501 struct ifreq ifr;
2502 int ret;
2503 char *colon;
2504
2505 /* One special case: SIOCGIFCONF takes ifconf argument
2506 and requires shared lock, because it sleeps writing
2507 to user space.
2508 */
2509
2510 if (cmd == SIOCGIFCONF) {
2511 rtnl_shlock();
2512 ret = dev_ifconf((char __user *) arg);
2513 rtnl_shunlock();
2514 return ret;
2515 }
2516 if (cmd == SIOCGIFNAME)
2517 return dev_ifname((struct ifreq __user *)arg);
2518
2519 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2520 return -EFAULT;
2521
2522 ifr.ifr_name[IFNAMSIZ-1] = 0;
2523
2524 colon = strchr(ifr.ifr_name, ':');
2525 if (colon)
2526 *colon = 0;
2527
2528 /*
2529 * See which interface the caller is talking about.
2530 */
2531
2532 switch (cmd) {
2533 /*
2534 * These ioctl calls:
2535 * - can be done by all.
2536 * - atomic and do not require locking.
2537 * - return a value
2538 */
2539 case SIOCGIFFLAGS:
2540 case SIOCGIFMETRIC:
2541 case SIOCGIFMTU:
2542 case SIOCGIFHWADDR:
2543 case SIOCGIFSLAVE:
2544 case SIOCGIFMAP:
2545 case SIOCGIFINDEX:
2546 case SIOCGIFTXQLEN:
2547 dev_load(ifr.ifr_name);
2548 read_lock(&dev_base_lock);
2549 ret = dev_ifsioc(&ifr, cmd);
2550 read_unlock(&dev_base_lock);
2551 if (!ret) {
2552 if (colon)
2553 *colon = ':';
2554 if (copy_to_user(arg, &ifr,
2555 sizeof(struct ifreq)))
2556 ret = -EFAULT;
2557 }
2558 return ret;
2559
2560 case SIOCETHTOOL:
2561 dev_load(ifr.ifr_name);
2562 rtnl_lock();
2563 ret = dev_ethtool(&ifr);
2564 rtnl_unlock();
2565 if (!ret) {
2566 if (colon)
2567 *colon = ':';
2568 if (copy_to_user(arg, &ifr,
2569 sizeof(struct ifreq)))
2570 ret = -EFAULT;
2571 }
2572 return ret;
2573
2574 /*
2575 * These ioctl calls:
2576 * - require superuser power.
2577 * - require strict serialization.
2578 * - return a value
2579 */
2580 case SIOCGMIIPHY:
2581 case SIOCGMIIREG:
2582 case SIOCSIFNAME:
2583 if (!capable(CAP_NET_ADMIN))
2584 return -EPERM;
2585 dev_load(ifr.ifr_name);
2586 rtnl_lock();
2587 ret = dev_ifsioc(&ifr, cmd);
2588 rtnl_unlock();
2589 if (!ret) {
2590 if (colon)
2591 *colon = ':';
2592 if (copy_to_user(arg, &ifr,
2593 sizeof(struct ifreq)))
2594 ret = -EFAULT;
2595 }
2596 return ret;
2597
2598 /*
2599 * These ioctl calls:
2600 * - require superuser power.
2601 * - require strict serialization.
2602 * - do not return a value
2603 */
2604 case SIOCSIFFLAGS:
2605 case SIOCSIFMETRIC:
2606 case SIOCSIFMTU:
2607 case SIOCSIFMAP:
2608 case SIOCSIFHWADDR:
2609 case SIOCSIFSLAVE:
2610 case SIOCADDMULTI:
2611 case SIOCDELMULTI:
2612 case SIOCSIFHWBROADCAST:
2613 case SIOCSIFTXQLEN:
2614 case SIOCSMIIREG:
2615 case SIOCBONDENSLAVE:
2616 case SIOCBONDRELEASE:
2617 case SIOCBONDSETHWADDR:
2618 case SIOCBONDSLAVEINFOQUERY:
2619 case SIOCBONDINFOQUERY:
2620 case SIOCBONDCHANGEACTIVE:
2621 case SIOCBRADDIF:
2622 case SIOCBRDELIF:
2623 if (!capable(CAP_NET_ADMIN))
2624 return -EPERM;
2625 dev_load(ifr.ifr_name);
2626 rtnl_lock();
2627 ret = dev_ifsioc(&ifr, cmd);
2628 rtnl_unlock();
2629 return ret;
2630
2631 case SIOCGIFMEM:
2632 /* Get the per device memory space. We can add this but
2633 * currently do not support it */
2634 case SIOCSIFMEM:
2635 /* Set the per device memory buffer space.
2636 * Not applicable in our case */
2637 case SIOCSIFLINK:
2638 return -EINVAL;
2639
2640 /*
2641 * Unknown or private ioctl.
2642 */
2643 default:
2644 if (cmd == SIOCWANDEV ||
2645 (cmd >= SIOCDEVPRIVATE &&
2646 cmd <= SIOCDEVPRIVATE + 15)) {
2647 dev_load(ifr.ifr_name);
2648 rtnl_lock();
2649 ret = dev_ifsioc(&ifr, cmd);
2650 rtnl_unlock();
2651 if (!ret && copy_to_user(arg, &ifr,
2652 sizeof(struct ifreq)))
2653 ret = -EFAULT;
2654 return ret;
2655 }
2656#ifdef WIRELESS_EXT
2657 /* Take care of Wireless Extensions */
2658 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2659 /* If command is `set a parameter', or
2660 * `get the encoding parameters', check if
2661 * the user has the right to do it */
2662 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2663 if (!capable(CAP_NET_ADMIN))
2664 return -EPERM;
2665 }
2666 dev_load(ifr.ifr_name);
2667 rtnl_lock();
2668 /* Follow me in net/core/wireless.c */
2669 ret = wireless_process_ioctl(&ifr, cmd);
2670 rtnl_unlock();
2671 if (IW_IS_GET(cmd) &&
2672 copy_to_user(arg, &ifr,
2673 sizeof(struct ifreq)))
2674 ret = -EFAULT;
2675 return ret;
2676 }
2677#endif /* WIRELESS_EXT */
2678 return -EINVAL;
2679 }
2680}
2681
2682
2683/**
2684 * dev_new_index - allocate an ifindex
2685 *
2686 * Returns a suitable unique value for a new device interface
2687 * number. The caller must hold the rtnl semaphore or the
2688 * dev_base_lock to be sure it remains unique.
2689 */
2690static int dev_new_index(void)
2691{
2692 static int ifindex;
2693 for (;;) {
2694 if (++ifindex <= 0)
2695 ifindex = 1;
2696 if (!__dev_get_by_index(ifindex))
2697 return ifindex;
2698 }
2699}
2700
2701static int dev_boot_phase = 1;
2702
2703/* Delayed registration/unregisteration */
2704static DEFINE_SPINLOCK(net_todo_list_lock);
2705static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2706
2707static inline void net_set_todo(struct net_device *dev)
2708{
2709 spin_lock(&net_todo_list_lock);
2710 list_add_tail(&dev->todo_list, &net_todo_list);
2711 spin_unlock(&net_todo_list_lock);
2712}
2713
2714/**
2715 * register_netdevice - register a network device
2716 * @dev: device to register
2717 *
2718 * Take a completed network device structure and add it to the kernel
2719 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2720 * chain. 0 is returned on success. A negative errno code is returned
2721 * on a failure to set up the device, or if the name is a duplicate.
2722 *
2723 * Callers must hold the rtnl semaphore. You may want
2724 * register_netdev() instead of this.
2725 *
2726 * BUGS:
2727 * The locking appears insufficient to guarantee two parallel registers
2728 * will not get the same name.
2729 */
2730
2731int register_netdevice(struct net_device *dev)
2732{
2733 struct hlist_head *head;
2734 struct hlist_node *p;
2735 int ret;
2736
2737 BUG_ON(dev_boot_phase);
2738 ASSERT_RTNL();
2739
2740 /* When net_device's are persistent, this will be fatal. */
2741 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2742
2743 spin_lock_init(&dev->queue_lock);
2744 spin_lock_init(&dev->xmit_lock);
2745 dev->xmit_lock_owner = -1;
2746#ifdef CONFIG_NET_CLS_ACT
2747 spin_lock_init(&dev->ingress_lock);
2748#endif
2749
2750 ret = alloc_divert_blk(dev);
2751 if (ret)
2752 goto out;
2753
2754 dev->iflink = -1;
2755
2756 /* Init, if this function is available */
2757 if (dev->init) {
2758 ret = dev->init(dev);
2759 if (ret) {
2760 if (ret > 0)
2761 ret = -EIO;
2762 goto out_err;
2763 }
2764 }
2765
2766 if (!dev_valid_name(dev->name)) {
2767 ret = -EINVAL;
2768 goto out_err;
2769 }
2770
2771 dev->ifindex = dev_new_index();
2772 if (dev->iflink == -1)
2773 dev->iflink = dev->ifindex;
2774
2775 /* Check for existence of name */
2776 head = dev_name_hash(dev->name);
2777 hlist_for_each(p, head) {
2778 struct net_device *d
2779 = hlist_entry(p, struct net_device, name_hlist);
2780 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2781 ret = -EEXIST;
2782 goto out_err;
2783 }
2784 }
2785
2786 /* Fix illegal SG+CSUM combinations. */
2787 if ((dev->features & NETIF_F_SG) &&
2788 !(dev->features & (NETIF_F_IP_CSUM |
2789 NETIF_F_NO_CSUM |
2790 NETIF_F_HW_CSUM))) {
2791 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2792 dev->name);
2793 dev->features &= ~NETIF_F_SG;
2794 }
2795
2796 /* TSO requires that SG is present as well. */
2797 if ((dev->features & NETIF_F_TSO) &&
2798 !(dev->features & NETIF_F_SG)) {
2799 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2800 dev->name);
2801 dev->features &= ~NETIF_F_TSO;
2802 }
2803
2804 /*
2805 * nil rebuild_header routine,
2806 * that should be never called and used as just bug trap.
2807 */
2808
2809 if (!dev->rebuild_header)
2810 dev->rebuild_header = default_rebuild_header;
2811
2812 /*
2813 * Default initial state at registry is that the
2814 * device is present.
2815 */
2816
2817 set_bit(__LINK_STATE_PRESENT, &dev->state);
2818
2819 dev->next = NULL;
2820 dev_init_scheduler(dev);
2821 write_lock_bh(&dev_base_lock);
2822 *dev_tail = dev;
2823 dev_tail = &dev->next;
2824 hlist_add_head(&dev->name_hlist, head);
2825 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2826 dev_hold(dev);
2827 dev->reg_state = NETREG_REGISTERING;
2828 write_unlock_bh(&dev_base_lock);
2829
2830 /* Notify protocols, that a new device appeared. */
2831 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2832
2833 /* Finish registration after unlock */
2834 net_set_todo(dev);
2835 ret = 0;
2836
2837out:
2838 return ret;
2839out_err:
2840 free_divert_blk(dev);
2841 goto out;
2842}
2843
2844/**
2845 * register_netdev - register a network device
2846 * @dev: device to register
2847 *
2848 * Take a completed network device structure and add it to the kernel
2849 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2850 * chain. 0 is returned on success. A negative errno code is returned
2851 * on a failure to set up the device, or if the name is a duplicate.
2852 *
2853 * This is a wrapper around register_netdev that takes the rtnl semaphore
2854 * and expands the device name if you passed a format string to
2855 * alloc_netdev.
2856 */
2857int register_netdev(struct net_device *dev)
2858{
2859 int err;
2860
2861 rtnl_lock();
2862
2863 /*
2864 * If the name is a format string the caller wants us to do a
2865 * name allocation.
2866 */
2867 if (strchr(dev->name, '%')) {
2868 err = dev_alloc_name(dev, dev->name);
2869 if (err < 0)
2870 goto out;
2871 }
2872
2873 /*
2874 * Back compatibility hook. Kill this one in 2.5
2875 */
2876 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2877 err = dev_alloc_name(dev, "eth%d");
2878 if (err < 0)
2879 goto out;
2880 }
2881
2882 err = register_netdevice(dev);
2883out:
2884 rtnl_unlock();
2885 return err;
2886}
2887EXPORT_SYMBOL(register_netdev);
2888
2889/*
2890 * netdev_wait_allrefs - wait until all references are gone.
2891 *
2892 * This is called when unregistering network devices.
2893 *
2894 * Any protocol or device that holds a reference should register
2895 * for netdevice notification, and cleanup and put back the
2896 * reference if they receive an UNREGISTER event.
2897 * We can get stuck here if buggy protocols don't correctly
2898 * call dev_put.
2899 */
2900static void netdev_wait_allrefs(struct net_device *dev)
2901{
2902 unsigned long rebroadcast_time, warning_time;
2903
2904 rebroadcast_time = warning_time = jiffies;
2905 while (atomic_read(&dev->refcnt) != 0) {
2906 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2907 rtnl_shlock();
2908
2909 /* Rebroadcast unregister notification */
2910 notifier_call_chain(&netdev_chain,
2911 NETDEV_UNREGISTER, dev);
2912
2913 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2914 &dev->state)) {
2915 /* We must not have linkwatch events
2916 * pending on unregister. If this
2917 * happens, we simply run the queue
2918 * unscheduled, resulting in a noop
2919 * for this device.
2920 */
2921 linkwatch_run_queue();
2922 }
2923
2924 rtnl_shunlock();
2925
2926 rebroadcast_time = jiffies;
2927 }
2928
2929 msleep(250);
2930
2931 if (time_after(jiffies, warning_time + 10 * HZ)) {
2932 printk(KERN_EMERG "unregister_netdevice: "
2933 "waiting for %s to become free. Usage "
2934 "count = %d\n",
2935 dev->name, atomic_read(&dev->refcnt));
2936 warning_time = jiffies;
2937 }
2938 }
2939}
2940
2941/* The sequence is:
2942 *
2943 * rtnl_lock();
2944 * ...
2945 * register_netdevice(x1);
2946 * register_netdevice(x2);
2947 * ...
2948 * unregister_netdevice(y1);
2949 * unregister_netdevice(y2);
2950 * ...
2951 * rtnl_unlock();
2952 * free_netdev(y1);
2953 * free_netdev(y2);
2954 *
2955 * We are invoked by rtnl_unlock() after it drops the semaphore.
2956 * This allows us to deal with problems:
2957 * 1) We can create/delete sysfs objects which invoke hotplug
2958 * without deadlocking with linkwatch via keventd.
2959 * 2) Since we run with the RTNL semaphore not held, we can sleep
2960 * safely in order to wait for the netdev refcnt to drop to zero.
2961 */
2962static DECLARE_MUTEX(net_todo_run_mutex);
2963void netdev_run_todo(void)
2964{
2965 struct list_head list = LIST_HEAD_INIT(list);
2966 int err;
2967
2968
2969 /* Need to guard against multiple cpu's getting out of order. */
2970 down(&net_todo_run_mutex);
2971
2972 /* Not safe to do outside the semaphore. We must not return
2973 * until all unregister events invoked by the local processor
2974 * have been completed (either by this todo run, or one on
2975 * another cpu).
2976 */
2977 if (list_empty(&net_todo_list))
2978 goto out;
2979
2980 /* Snapshot list, allow later requests */
2981 spin_lock(&net_todo_list_lock);
2982 list_splice_init(&net_todo_list, &list);
2983 spin_unlock(&net_todo_list_lock);
2984
2985 while (!list_empty(&list)) {
2986 struct net_device *dev
2987 = list_entry(list.next, struct net_device, todo_list);
2988 list_del(&dev->todo_list);
2989
2990 switch(dev->reg_state) {
2991 case NETREG_REGISTERING:
2992 err = netdev_register_sysfs(dev);
2993 if (err)
2994 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2995 dev->name, err);
2996 dev->reg_state = NETREG_REGISTERED;
2997 break;
2998
2999 case NETREG_UNREGISTERING:
3000 netdev_unregister_sysfs(dev);
3001 dev->reg_state = NETREG_UNREGISTERED;
3002
3003 netdev_wait_allrefs(dev);
3004
3005 /* paranoia */
3006 BUG_ON(atomic_read(&dev->refcnt));
3007 BUG_TRAP(!dev->ip_ptr);
3008 BUG_TRAP(!dev->ip6_ptr);
3009 BUG_TRAP(!dev->dn_ptr);
3010
3011
3012 /* It must be the very last action,
3013 * after this 'dev' may point to freed up memory.
3014 */
3015 if (dev->destructor)
3016 dev->destructor(dev);
3017 break;
3018
3019 default:
3020 printk(KERN_ERR "network todo '%s' but state %d\n",
3021 dev->name, dev->reg_state);
3022 break;
3023 }
3024 }
3025
3026out:
3027 up(&net_todo_run_mutex);
3028}
3029
3030/**
3031 * alloc_netdev - allocate network device
3032 * @sizeof_priv: size of private data to allocate space for
3033 * @name: device name format string
3034 * @setup: callback to initialize device
3035 *
3036 * Allocates a struct net_device with private data area for driver use
3037 * and performs basic initialization.
3038 */
3039struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3040 void (*setup)(struct net_device *))
3041{
3042 void *p;
3043 struct net_device *dev;
3044 int alloc_size;
3045
3046 /* ensure 32-byte alignment of both the device and private area */
3047 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3048 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3049
3050 p = kmalloc(alloc_size, GFP_KERNEL);
3051 if (!p) {
3052 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3053 return NULL;
3054 }
3055 memset(p, 0, alloc_size);
3056
3057 dev = (struct net_device *)
3058 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3059 dev->padded = (char *)dev - (char *)p;
3060
3061 if (sizeof_priv)
3062 dev->priv = netdev_priv(dev);
3063
3064 setup(dev);
3065 strcpy(dev->name, name);
3066 return dev;
3067}
3068EXPORT_SYMBOL(alloc_netdev);
3069
3070/**
3071 * free_netdev - free network device
3072 * @dev: device
3073 *
3074 * This function does the last stage of destroying an allocated device
3075 * interface. The reference to the device object is released.
3076 * If this is the last reference then it will be freed.
3077 */
3078void free_netdev(struct net_device *dev)
3079{
3080#ifdef CONFIG_SYSFS
3081 /* Compatiablity with error handling in drivers */
3082 if (dev->reg_state == NETREG_UNINITIALIZED) {
3083 kfree((char *)dev - dev->padded);
3084 return;
3085 }
3086
3087 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3088 dev->reg_state = NETREG_RELEASED;
3089
3090 /* will free via class release */
3091 class_device_put(&dev->class_dev);
3092#else
3093 kfree((char *)dev - dev->padded);
3094#endif
3095}
3096
3097/* Synchronize with packet receive processing. */
3098void synchronize_net(void)
3099{
3100 might_sleep();
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07003101 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102}
3103
3104/**
3105 * unregister_netdevice - remove device from the kernel
3106 * @dev: device
3107 *
3108 * This function shuts down a device interface and removes it
3109 * from the kernel tables. On success 0 is returned, on a failure
3110 * a negative errno code is returned.
3111 *
3112 * Callers must hold the rtnl semaphore. You may want
3113 * unregister_netdev() instead of this.
3114 */
3115
3116int unregister_netdevice(struct net_device *dev)
3117{
3118 struct net_device *d, **dp;
3119
3120 BUG_ON(dev_boot_phase);
3121 ASSERT_RTNL();
3122
3123 /* Some devices call without registering for initialization unwind. */
3124 if (dev->reg_state == NETREG_UNINITIALIZED) {
3125 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3126 "was registered\n", dev->name, dev);
3127 return -ENODEV;
3128 }
3129
3130 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3131
3132 /* If device is running, close it first. */
3133 if (dev->flags & IFF_UP)
3134 dev_close(dev);
3135
3136 /* And unlink it from device chain. */
3137 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3138 if (d == dev) {
3139 write_lock_bh(&dev_base_lock);
3140 hlist_del(&dev->name_hlist);
3141 hlist_del(&dev->index_hlist);
3142 if (dev_tail == &dev->next)
3143 dev_tail = dp;
3144 *dp = d->next;
3145 write_unlock_bh(&dev_base_lock);
3146 break;
3147 }
3148 }
3149 if (!d) {
3150 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3151 dev->name);
3152 return -ENODEV;
3153 }
3154
3155 dev->reg_state = NETREG_UNREGISTERING;
3156
3157 synchronize_net();
3158
3159 /* Shutdown queueing discipline. */
3160 dev_shutdown(dev);
3161
3162
3163 /* Notify protocols, that we are about to destroy
3164 this device. They should clean all the things.
3165 */
3166 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3167
3168 /*
3169 * Flush the multicast chain
3170 */
3171 dev_mc_discard(dev);
3172
3173 if (dev->uninit)
3174 dev->uninit(dev);
3175
3176 /* Notifier chain MUST detach us from master device. */
3177 BUG_TRAP(!dev->master);
3178
3179 free_divert_blk(dev);
3180
3181 /* Finish processing unregister after unlock */
3182 net_set_todo(dev);
3183
3184 synchronize_net();
3185
3186 dev_put(dev);
3187 return 0;
3188}
3189
3190/**
3191 * unregister_netdev - remove device from the kernel
3192 * @dev: device
3193 *
3194 * This function shuts down a device interface and removes it
3195 * from the kernel tables. On success 0 is returned, on a failure
3196 * a negative errno code is returned.
3197 *
3198 * This is just a wrapper for unregister_netdevice that takes
3199 * the rtnl semaphore. In general you want to use this and not
3200 * unregister_netdevice.
3201 */
3202void unregister_netdev(struct net_device *dev)
3203{
3204 rtnl_lock();
3205 unregister_netdevice(dev);
3206 rtnl_unlock();
3207}
3208
3209EXPORT_SYMBOL(unregister_netdev);
3210
3211#ifdef CONFIG_HOTPLUG_CPU
3212static int dev_cpu_callback(struct notifier_block *nfb,
3213 unsigned long action,
3214 void *ocpu)
3215{
3216 struct sk_buff **list_skb;
3217 struct net_device **list_net;
3218 struct sk_buff *skb;
3219 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3220 struct softnet_data *sd, *oldsd;
3221
3222 if (action != CPU_DEAD)
3223 return NOTIFY_OK;
3224
3225 local_irq_disable();
3226 cpu = smp_processor_id();
3227 sd = &per_cpu(softnet_data, cpu);
3228 oldsd = &per_cpu(softnet_data, oldcpu);
3229
3230 /* Find end of our completion_queue. */
3231 list_skb = &sd->completion_queue;
3232 while (*list_skb)
3233 list_skb = &(*list_skb)->next;
3234 /* Append completion queue from offline CPU. */
3235 *list_skb = oldsd->completion_queue;
3236 oldsd->completion_queue = NULL;
3237
3238 /* Find end of our output_queue. */
3239 list_net = &sd->output_queue;
3240 while (*list_net)
3241 list_net = &(*list_net)->next_sched;
3242 /* Append output queue from offline CPU. */
3243 *list_net = oldsd->output_queue;
3244 oldsd->output_queue = NULL;
3245
3246 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3247 local_irq_enable();
3248
3249 /* Process offline CPU's input_pkt_queue */
3250 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3251 netif_rx(skb);
3252
3253 return NOTIFY_OK;
3254}
3255#endif /* CONFIG_HOTPLUG_CPU */
3256
3257
3258/*
3259 * Initialize the DEV module. At boot time this walks the device list and
3260 * unhooks any devices that fail to initialise (normally hardware not
3261 * present) and leaves us with a valid list of present and active devices.
3262 *
3263 */
3264
3265/*
3266 * This is called single threaded during boot, so no need
3267 * to take the rtnl semaphore.
3268 */
3269static int __init net_dev_init(void)
3270{
3271 int i, rc = -ENOMEM;
3272
3273 BUG_ON(!dev_boot_phase);
3274
3275 net_random_init();
3276
3277 if (dev_proc_init())
3278 goto out;
3279
3280 if (netdev_sysfs_init())
3281 goto out;
3282
3283 INIT_LIST_HEAD(&ptype_all);
3284 for (i = 0; i < 16; i++)
3285 INIT_LIST_HEAD(&ptype_base[i]);
3286
3287 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3288 INIT_HLIST_HEAD(&dev_name_head[i]);
3289
3290 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3291 INIT_HLIST_HEAD(&dev_index_head[i]);
3292
3293 /*
3294 * Initialise the packet receive queues.
3295 */
3296
3297 for (i = 0; i < NR_CPUS; i++) {
3298 struct softnet_data *queue;
3299
3300 queue = &per_cpu(softnet_data, i);
3301 skb_queue_head_init(&queue->input_pkt_queue);
3302 queue->throttle = 0;
3303 queue->cng_level = 0;
3304 queue->avg_blog = 10; /* arbitrary non-zero */
3305 queue->completion_queue = NULL;
3306 INIT_LIST_HEAD(&queue->poll_list);
3307 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3308 queue->backlog_dev.weight = weight_p;
3309 queue->backlog_dev.poll = process_backlog;
3310 atomic_set(&queue->backlog_dev.refcnt, 1);
3311 }
3312
3313#ifdef OFFLINE_SAMPLE
3314 samp_timer.expires = jiffies + (10 * HZ);
3315 add_timer(&samp_timer);
3316#endif
3317
3318 dev_boot_phase = 0;
3319
3320 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3321 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3322
3323 hotcpu_notifier(dev_cpu_callback, 0);
3324 dst_init();
3325 dev_mcast_init();
3326 rc = 0;
3327out:
3328 return rc;
3329}
3330
3331subsys_initcall(net_dev_init);
3332
3333EXPORT_SYMBOL(__dev_get_by_index);
3334EXPORT_SYMBOL(__dev_get_by_name);
3335EXPORT_SYMBOL(__dev_remove_pack);
3336EXPORT_SYMBOL(__skb_linearize);
3337EXPORT_SYMBOL(dev_add_pack);
3338EXPORT_SYMBOL(dev_alloc_name);
3339EXPORT_SYMBOL(dev_close);
3340EXPORT_SYMBOL(dev_get_by_flags);
3341EXPORT_SYMBOL(dev_get_by_index);
3342EXPORT_SYMBOL(dev_get_by_name);
3343EXPORT_SYMBOL(dev_ioctl);
3344EXPORT_SYMBOL(dev_open);
3345EXPORT_SYMBOL(dev_queue_xmit);
3346EXPORT_SYMBOL(dev_remove_pack);
3347EXPORT_SYMBOL(dev_set_allmulti);
3348EXPORT_SYMBOL(dev_set_promiscuity);
3349EXPORT_SYMBOL(dev_change_flags);
3350EXPORT_SYMBOL(dev_set_mtu);
3351EXPORT_SYMBOL(dev_set_mac_address);
3352EXPORT_SYMBOL(free_netdev);
3353EXPORT_SYMBOL(netdev_boot_setup_check);
3354EXPORT_SYMBOL(netdev_set_master);
3355EXPORT_SYMBOL(netdev_state_change);
3356EXPORT_SYMBOL(netif_receive_skb);
3357EXPORT_SYMBOL(netif_rx);
3358EXPORT_SYMBOL(register_gifconf);
3359EXPORT_SYMBOL(register_netdevice);
3360EXPORT_SYMBOL(register_netdevice_notifier);
3361EXPORT_SYMBOL(skb_checksum_help);
3362EXPORT_SYMBOL(synchronize_net);
3363EXPORT_SYMBOL(unregister_netdevice);
3364EXPORT_SYMBOL(unregister_netdevice_notifier);
3365EXPORT_SYMBOL(net_enable_timestamp);
3366EXPORT_SYMBOL(net_disable_timestamp);
3367EXPORT_SYMBOL(dev_get_flags);
3368
3369#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3370EXPORT_SYMBOL(br_handle_frame_hook);
3371EXPORT_SYMBOL(br_fdb_get_hook);
3372EXPORT_SYMBOL(br_fdb_put_hook);
3373#endif
3374
3375#ifdef CONFIG_KMOD
3376EXPORT_SYMBOL(dev_load);
3377#endif
3378
3379EXPORT_PER_CPU_SYMBOL(softnet_data);