blob: be1d896cc5b94b96bbd10cd464564f12fcdc790e [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>
Randy Dunlap4fc268d2006-01-11 12:17:47 -080078#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#include <linux/config.h>
80#include <linux/cpu.h>
81#include <linux/types.h>
82#include <linux/kernel.h>
83#include <linux/sched.h>
84#include <linux/string.h>
85#include <linux/mm.h>
86#include <linux/socket.h>
87#include <linux/sockios.h>
88#include <linux/errno.h>
89#include <linux/interrupt.h>
90#include <linux/if_ether.h>
91#include <linux/netdevice.h>
92#include <linux/etherdevice.h>
93#include <linux/notifier.h>
94#include <linux/skbuff.h>
95#include <net/sock.h>
96#include <linux/rtnetlink.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/stat.h>
100#include <linux/if_bridge.h>
101#include <linux/divert.h>
102#include <net/dst.h>
103#include <net/pkt_sched.h>
104#include <net/checksum.h>
105#include <linux/highmem.h>
106#include <linux/init.h>
107#include <linux/kmod.h>
108#include <linux/module.h>
109#include <linux/kallsyms.h>
110#include <linux/netpoll.h>
111#include <linux/rcupdate.h>
112#include <linux/delay.h>
Adrian Bunkd86b5e02006-01-21 00:46:55 +0100113#include <linux/wireless.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114#include <net/iw_handler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115#include <asm/current.h>
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117/*
118 * The list of packet types we will receive (as opposed to discard)
119 * and the routines to invoke.
120 *
121 * Why 16. Because with 16 the only overlap we get on a hash of the
122 * low nibble of the protocol value is RARP/SNAP/X.25.
123 *
124 * NOTE: That is no longer true with the addition of VLAN tags. Not
125 * sure which should go first, but I bet it won't make much
126 * difference if we are running VLANs. The good news is that
127 * this protocol won't be in the list unless compiled in, so
128 * the average user (w/out VLANs) will not be adversly affected.
129 * --BLG
130 *
131 * 0800 IP
132 * 8100 802.1Q VLAN
133 * 0001 802.3
134 * 0002 AX.25
135 * 0004 802.2
136 * 8035 RARP
137 * 0005 SNAP
138 * 0805 X.25
139 * 0806 ARP
140 * 8137 IPX
141 * 0009 Localtalk
142 * 86DD IPv6
143 */
144
145static DEFINE_SPINLOCK(ptype_lock);
146static struct list_head ptype_base[16]; /* 16 way hashed list */
147static struct list_head ptype_all; /* Taps */
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149/*
150 * The @dev_base list is protected by @dev_base_lock and the rtln
151 * semaphore.
152 *
153 * Pure readers hold dev_base_lock for reading.
154 *
155 * Writers must hold the rtnl semaphore while they loop through the
156 * dev_base list, and hold dev_base_lock for writing when they do the
157 * actual updates. This allows pure readers to access the list even
158 * while a writer is preparing to update it.
159 *
160 * To put it another way, dev_base_lock is held for writing only to
161 * protect against pure readers; the rtnl semaphore provides the
162 * protection against other writers.
163 *
164 * See, for example usages, register_netdevice() and
165 * unregister_netdevice(), which must be called with the rtnl
166 * semaphore held.
167 */
168struct net_device *dev_base;
169static struct net_device **dev_tail = &dev_base;
170DEFINE_RWLOCK(dev_base_lock);
171
172EXPORT_SYMBOL(dev_base);
173EXPORT_SYMBOL(dev_base_lock);
174
175#define NETDEV_HASHBITS 8
176static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
177static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
178
179static inline struct hlist_head *dev_name_hash(const char *name)
180{
181 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
182 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
183}
184
185static inline struct hlist_head *dev_index_hash(int ifindex)
186{
187 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
188}
189
190/*
191 * Our notifier list
192 */
193
194static struct notifier_block *netdev_chain;
195
196/*
197 * Device drivers call our routines to queue packets here. We empty the
198 * queue in the local softnet handler.
199 */
Stephen Hemminger31aa02c2005-06-23 20:12:48 -0700200DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
202#ifdef CONFIG_SYSFS
203extern int netdev_sysfs_init(void);
204extern int netdev_register_sysfs(struct net_device *);
205extern void netdev_unregister_sysfs(struct net_device *);
206#else
207#define netdev_sysfs_init() (0)
208#define netdev_register_sysfs(dev) (0)
209#define netdev_unregister_sysfs(dev) do { } while(0)
210#endif
211
212
213/*******************************************************************************
214
215 Protocol management and registration routines
216
217*******************************************************************************/
218
219/*
220 * For efficiency
221 */
222
223int netdev_nit;
224
225/*
226 * Add a protocol ID to the list. Now that the input handler is
227 * smarter we can dispense with all the messy stuff that used to be
228 * here.
229 *
230 * BEWARE!!! Protocol handlers, mangling input packets,
231 * MUST BE last in hash buckets and checking protocol handlers
232 * MUST start from promiscuous ptype_all chain in net_bh.
233 * It is true now, do not change it.
234 * Explanation follows: if protocol handler, mangling packet, will
235 * be the first on list, it is not able to sense, that packet
236 * is cloned and should be copied-on-write, so that it will
237 * change it and subsequent readers will get broken packet.
238 * --ANK (980803)
239 */
240
241/**
242 * dev_add_pack - add packet handler
243 * @pt: packet type declaration
244 *
245 * Add a protocol handler to the networking stack. The passed &packet_type
246 * is linked into kernel lists and may not be freed until it has been
247 * removed from the kernel lists.
248 *
249 * This call does not sleep therefore it can not
250 * guarantee all CPU's that are in middle of receiving packets
251 * will see the new packet type (until the next received packet).
252 */
253
254void dev_add_pack(struct packet_type *pt)
255{
256 int hash;
257
258 spin_lock_bh(&ptype_lock);
259 if (pt->type == htons(ETH_P_ALL)) {
260 netdev_nit++;
261 list_add_rcu(&pt->list, &ptype_all);
262 } else {
263 hash = ntohs(pt->type) & 15;
264 list_add_rcu(&pt->list, &ptype_base[hash]);
265 }
266 spin_unlock_bh(&ptype_lock);
267}
268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269/**
270 * __dev_remove_pack - remove packet handler
271 * @pt: packet type declaration
272 *
273 * Remove a protocol handler that was previously added to the kernel
274 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
275 * from the kernel lists and can be freed or reused once this function
276 * returns.
277 *
278 * The packet type might still be in use by receivers
279 * and must not be freed until after all the CPU's have gone
280 * through a quiescent state.
281 */
282void __dev_remove_pack(struct packet_type *pt)
283{
284 struct list_head *head;
285 struct packet_type *pt1;
286
287 spin_lock_bh(&ptype_lock);
288
289 if (pt->type == htons(ETH_P_ALL)) {
290 netdev_nit--;
291 head = &ptype_all;
292 } else
293 head = &ptype_base[ntohs(pt->type) & 15];
294
295 list_for_each_entry(pt1, head, list) {
296 if (pt == pt1) {
297 list_del_rcu(&pt->list);
298 goto out;
299 }
300 }
301
302 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
303out:
304 spin_unlock_bh(&ptype_lock);
305}
306/**
307 * dev_remove_pack - remove packet handler
308 * @pt: packet type declaration
309 *
310 * Remove a protocol handler that was previously added to the kernel
311 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
312 * from the kernel lists and can be freed or reused once this function
313 * returns.
314 *
315 * This call sleeps to guarantee that no CPU is looking at the packet
316 * type after return.
317 */
318void dev_remove_pack(struct packet_type *pt)
319{
320 __dev_remove_pack(pt);
321
322 synchronize_net();
323}
324
325/******************************************************************************
326
327 Device Boot-time Settings Routines
328
329*******************************************************************************/
330
331/* Boot time configuration table */
332static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
333
334/**
335 * netdev_boot_setup_add - add new setup entry
336 * @name: name of the device
337 * @map: configured settings for the device
338 *
339 * Adds new setup entry to the dev_boot_setup list. The function
340 * returns 0 on error and 1 on success. This is a generic routine to
341 * all netdevices.
342 */
343static int netdev_boot_setup_add(char *name, struct ifmap *map)
344{
345 struct netdev_boot_setup *s;
346 int i;
347
348 s = dev_boot_setup;
349 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
350 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
351 memset(s[i].name, 0, sizeof(s[i].name));
352 strcpy(s[i].name, name);
353 memcpy(&s[i].map, map, sizeof(s[i].map));
354 break;
355 }
356 }
357
358 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
359}
360
361/**
362 * netdev_boot_setup_check - check boot time settings
363 * @dev: the netdevice
364 *
365 * Check boot time settings for the device.
366 * The found settings are set for the device to be used
367 * later in the device probing.
368 * Returns 0 if no settings found, 1 if they are.
369 */
370int netdev_boot_setup_check(struct net_device *dev)
371{
372 struct netdev_boot_setup *s = dev_boot_setup;
373 int i;
374
375 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
376 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
377 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
378 dev->irq = s[i].map.irq;
379 dev->base_addr = s[i].map.base_addr;
380 dev->mem_start = s[i].map.mem_start;
381 dev->mem_end = s[i].map.mem_end;
382 return 1;
383 }
384 }
385 return 0;
386}
387
388
389/**
390 * netdev_boot_base - get address from boot time settings
391 * @prefix: prefix for network device
392 * @unit: id for network device
393 *
394 * Check boot time settings for the base address of device.
395 * The found settings are set for the device to be used
396 * later in the device probing.
397 * Returns 0 if no settings found.
398 */
399unsigned long netdev_boot_base(const char *prefix, int unit)
400{
401 const struct netdev_boot_setup *s = dev_boot_setup;
402 char name[IFNAMSIZ];
403 int i;
404
405 sprintf(name, "%s%d", prefix, unit);
406
407 /*
408 * If device already registered then return base of 1
409 * to indicate not to probe for this interface
410 */
411 if (__dev_get_by_name(name))
412 return 1;
413
414 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
415 if (!strcmp(name, s[i].name))
416 return s[i].map.base_addr;
417 return 0;
418}
419
420/*
421 * Saves at boot time configured settings for any netdevice.
422 */
423int __init netdev_boot_setup(char *str)
424{
425 int ints[5];
426 struct ifmap map;
427
428 str = get_options(str, ARRAY_SIZE(ints), ints);
429 if (!str || !*str)
430 return 0;
431
432 /* Save settings */
433 memset(&map, 0, sizeof(map));
434 if (ints[0] > 0)
435 map.irq = ints[1];
436 if (ints[0] > 1)
437 map.base_addr = ints[2];
438 if (ints[0] > 2)
439 map.mem_start = ints[3];
440 if (ints[0] > 3)
441 map.mem_end = ints[4];
442
443 /* Add new entry to the list */
444 return netdev_boot_setup_add(str, &map);
445}
446
447__setup("netdev=", netdev_boot_setup);
448
449/*******************************************************************************
450
451 Device Interface Subroutines
452
453*******************************************************************************/
454
455/**
456 * __dev_get_by_name - find a device by its name
457 * @name: name to find
458 *
459 * Find an interface by name. Must be called under RTNL semaphore
460 * or @dev_base_lock. If the name is found a pointer to the device
461 * is returned. If the name is not found then %NULL is returned. The
462 * reference counters are not incremented so the caller must be
463 * careful with locks.
464 */
465
466struct net_device *__dev_get_by_name(const char *name)
467{
468 struct hlist_node *p;
469
470 hlist_for_each(p, dev_name_hash(name)) {
471 struct net_device *dev
472 = hlist_entry(p, struct net_device, name_hlist);
473 if (!strncmp(dev->name, name, IFNAMSIZ))
474 return dev;
475 }
476 return NULL;
477}
478
479/**
480 * dev_get_by_name - find a device by its name
481 * @name: name to find
482 *
483 * Find an interface by name. This can be called from any
484 * context and does its own locking. The returned handle has
485 * the usage count incremented and the caller must use dev_put() to
486 * release it when it is no longer needed. %NULL is returned if no
487 * matching device is found.
488 */
489
490struct net_device *dev_get_by_name(const char *name)
491{
492 struct net_device *dev;
493
494 read_lock(&dev_base_lock);
495 dev = __dev_get_by_name(name);
496 if (dev)
497 dev_hold(dev);
498 read_unlock(&dev_base_lock);
499 return dev;
500}
501
502/**
503 * __dev_get_by_index - find a device by its ifindex
504 * @ifindex: index of device
505 *
506 * Search for an interface by index. Returns %NULL if the device
507 * is not found or a pointer to the device. The device has not
508 * had its reference counter increased so the caller must be careful
509 * about locking. The caller must hold either the RTNL semaphore
510 * or @dev_base_lock.
511 */
512
513struct net_device *__dev_get_by_index(int ifindex)
514{
515 struct hlist_node *p;
516
517 hlist_for_each(p, dev_index_hash(ifindex)) {
518 struct net_device *dev
519 = hlist_entry(p, struct net_device, index_hlist);
520 if (dev->ifindex == ifindex)
521 return dev;
522 }
523 return NULL;
524}
525
526
527/**
528 * dev_get_by_index - find a device by its ifindex
529 * @ifindex: index of device
530 *
531 * Search for an interface by index. Returns NULL if the device
532 * is not found or a pointer to the device. The device returned has
533 * had a reference added and the pointer is safe until the user calls
534 * dev_put to indicate they have finished with it.
535 */
536
537struct net_device *dev_get_by_index(int ifindex)
538{
539 struct net_device *dev;
540
541 read_lock(&dev_base_lock);
542 dev = __dev_get_by_index(ifindex);
543 if (dev)
544 dev_hold(dev);
545 read_unlock(&dev_base_lock);
546 return dev;
547}
548
549/**
550 * dev_getbyhwaddr - find a device by its hardware address
551 * @type: media type of device
552 * @ha: hardware address
553 *
554 * Search for an interface by MAC address. Returns NULL if the device
555 * is not found or a pointer to the device. The caller must hold the
556 * rtnl semaphore. The returned device has not had its ref count increased
557 * and the caller must therefore be careful about locking
558 *
559 * BUGS:
560 * If the API was consistent this would be __dev_get_by_hwaddr
561 */
562
563struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
564{
565 struct net_device *dev;
566
567 ASSERT_RTNL();
568
569 for (dev = dev_base; dev; dev = dev->next)
570 if (dev->type == type &&
571 !memcmp(dev->dev_addr, ha, dev->addr_len))
572 break;
573 return dev;
574}
575
Jochen Friedrichcf309e32005-09-22 04:44:55 -0300576EXPORT_SYMBOL(dev_getbyhwaddr);
577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578struct net_device *dev_getfirstbyhwtype(unsigned short type)
579{
580 struct net_device *dev;
581
582 rtnl_lock();
583 for (dev = dev_base; dev; dev = dev->next) {
584 if (dev->type == type) {
585 dev_hold(dev);
586 break;
587 }
588 }
589 rtnl_unlock();
590 return dev;
591}
592
593EXPORT_SYMBOL(dev_getfirstbyhwtype);
594
595/**
596 * dev_get_by_flags - find any device with given flags
597 * @if_flags: IFF_* values
598 * @mask: bitmask of bits in if_flags to check
599 *
600 * Search for any interface with the given flags. Returns NULL if a device
601 * is not found or a pointer to the device. The device returned has
602 * had a reference added and the pointer is safe until the user calls
603 * dev_put to indicate they have finished with it.
604 */
605
606struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
607{
608 struct net_device *dev;
609
610 read_lock(&dev_base_lock);
611 for (dev = dev_base; dev != NULL; dev = dev->next) {
612 if (((dev->flags ^ if_flags) & mask) == 0) {
613 dev_hold(dev);
614 break;
615 }
616 }
617 read_unlock(&dev_base_lock);
618 return dev;
619}
620
621/**
622 * dev_valid_name - check if name is okay for network device
623 * @name: name string
624 *
625 * Network device names need to be valid file names to
626 * to allow sysfs to work
627 */
Mitch Williamsc2373ee2005-11-09 10:34:45 -0800628int dev_valid_name(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
630 return !(*name == '\0'
631 || !strcmp(name, ".")
632 || !strcmp(name, "..")
633 || strchr(name, '/'));
634}
635
636/**
637 * dev_alloc_name - allocate a name for a device
638 * @dev: device
639 * @name: name format string
640 *
641 * Passed a format string - eg "lt%d" it will try and find a suitable
642 * id. Not efficient for many devices, not called a lot. The caller
643 * must hold the dev_base or rtnl lock while allocating the name and
644 * adding the device in order to avoid duplicates. Returns the number
645 * of the unit assigned or a negative errno code.
646 */
647
648int dev_alloc_name(struct net_device *dev, const char *name)
649{
650 int i = 0;
651 char buf[IFNAMSIZ];
652 const char *p;
653 const int max_netdevices = 8*PAGE_SIZE;
654 long *inuse;
655 struct net_device *d;
656
657 p = strnchr(name, IFNAMSIZ-1, '%');
658 if (p) {
659 /*
660 * Verify the string as this thing may have come from
661 * the user. There must be either one "%d" and no other "%"
662 * characters.
663 */
664 if (p[1] != 'd' || strchr(p + 2, '%'))
665 return -EINVAL;
666
667 /* Use one page as a bit array of possible slots */
668 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
669 if (!inuse)
670 return -ENOMEM;
671
672 for (d = dev_base; d; d = d->next) {
673 if (!sscanf(d->name, name, &i))
674 continue;
675 if (i < 0 || i >= max_netdevices)
676 continue;
677
678 /* avoid cases where sscanf is not exact inverse of printf */
679 snprintf(buf, sizeof(buf), name, i);
680 if (!strncmp(buf, d->name, IFNAMSIZ))
681 set_bit(i, inuse);
682 }
683
684 i = find_first_zero_bit(inuse, max_netdevices);
685 free_page((unsigned long) inuse);
686 }
687
688 snprintf(buf, sizeof(buf), name, i);
689 if (!__dev_get_by_name(buf)) {
690 strlcpy(dev->name, buf, IFNAMSIZ);
691 return i;
692 }
693
694 /* It is possible to run out of possible slots
695 * when the name is long and there isn't enough space left
696 * for the digits, or if all bits are used.
697 */
698 return -ENFILE;
699}
700
701
702/**
703 * dev_change_name - change name of a device
704 * @dev: device
705 * @newname: name (or format string) must be at least IFNAMSIZ
706 *
707 * Change name of a device, can pass format strings "eth%d".
708 * for wildcarding.
709 */
710int dev_change_name(struct net_device *dev, char *newname)
711{
712 int err = 0;
713
714 ASSERT_RTNL();
715
716 if (dev->flags & IFF_UP)
717 return -EBUSY;
718
719 if (!dev_valid_name(newname))
720 return -EINVAL;
721
722 if (strchr(newname, '%')) {
723 err = dev_alloc_name(dev, newname);
724 if (err < 0)
725 return err;
726 strcpy(newname, dev->name);
727 }
728 else if (__dev_get_by_name(newname))
729 return -EEXIST;
730 else
731 strlcpy(dev->name, newname, IFNAMSIZ);
732
733 err = class_device_rename(&dev->class_dev, dev->name);
734 if (!err) {
735 hlist_del(&dev->name_hlist);
736 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
737 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
738 }
739
740 return err;
741}
742
743/**
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700744 * netdev_features_change - device changes fatures
745 * @dev: device to cause notification
746 *
747 * Called to indicate a device has changed features.
748 */
749void netdev_features_change(struct net_device *dev)
750{
751 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
752}
753EXPORT_SYMBOL(netdev_features_change);
754
755/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 * netdev_state_change - device changes state
757 * @dev: device to cause notification
758 *
759 * Called to indicate a device has changed state. This function calls
760 * the notifier chains for netdev_chain and sends a NEWLINK message
761 * to the routing socket.
762 */
763void netdev_state_change(struct net_device *dev)
764{
765 if (dev->flags & IFF_UP) {
766 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
767 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
768 }
769}
770
771/**
772 * dev_load - load a network module
773 * @name: name of interface
774 *
775 * If a network interface is not present and the process has suitable
776 * privileges this function loads the module. If module loading is not
777 * available in this kernel then it becomes a nop.
778 */
779
780void dev_load(const char *name)
781{
782 struct net_device *dev;
783
784 read_lock(&dev_base_lock);
785 dev = __dev_get_by_name(name);
786 read_unlock(&dev_base_lock);
787
788 if (!dev && capable(CAP_SYS_MODULE))
789 request_module("%s", name);
790}
791
792static int default_rebuild_header(struct sk_buff *skb)
793{
794 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
795 skb->dev ? skb->dev->name : "NULL!!!");
796 kfree_skb(skb);
797 return 1;
798}
799
800
801/**
802 * dev_open - prepare an interface for use.
803 * @dev: device to open
804 *
805 * Takes a device from down to up state. The device's private open
806 * function is invoked and then the multicast lists are loaded. Finally
807 * the device is moved into the up state and a %NETDEV_UP message is
808 * sent to the netdev notifier chain.
809 *
810 * Calling this function on an active interface is a nop. On a failure
811 * a negative errno code is returned.
812 */
813int dev_open(struct net_device *dev)
814{
815 int ret = 0;
816
817 /*
818 * Is it already up?
819 */
820
821 if (dev->flags & IFF_UP)
822 return 0;
823
824 /*
825 * Is it even present?
826 */
827 if (!netif_device_present(dev))
828 return -ENODEV;
829
830 /*
831 * Call device private open method
832 */
833 set_bit(__LINK_STATE_START, &dev->state);
834 if (dev->open) {
835 ret = dev->open(dev);
836 if (ret)
837 clear_bit(__LINK_STATE_START, &dev->state);
838 }
839
840 /*
841 * If it went open OK then:
842 */
843
844 if (!ret) {
845 /*
846 * Set the flags.
847 */
848 dev->flags |= IFF_UP;
849
850 /*
851 * Initialize multicasting status
852 */
853 dev_mc_upload(dev);
854
855 /*
856 * Wakeup transmit queue engine
857 */
858 dev_activate(dev);
859
860 /*
861 * ... and announce new interface.
862 */
863 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
864 }
865 return ret;
866}
867
868/**
869 * dev_close - shutdown an interface.
870 * @dev: device to shutdown
871 *
872 * This function moves an active device into down state. A
873 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
874 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
875 * chain.
876 */
877int dev_close(struct net_device *dev)
878{
879 if (!(dev->flags & IFF_UP))
880 return 0;
881
882 /*
883 * Tell people we are going down, so that they can
884 * prepare to death, when device is still operating.
885 */
886 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
887
888 dev_deactivate(dev);
889
890 clear_bit(__LINK_STATE_START, &dev->state);
891
892 /* Synchronize to scheduled poll. We cannot touch poll list,
893 * it can be even on different cpu. So just clear netif_running(),
894 * and wait when poll really will happen. Actually, the best place
895 * for this is inside dev->stop() after device stopped its irq
896 * engine, but this requires more changes in devices. */
897
898 smp_mb__after_clear_bit(); /* Commit netif_running(). */
899 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
900 /* No hurry. */
David S. Miller6192b542005-07-28 12:12:58 -0700901 msleep(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 }
903
904 /*
905 * Call the device specific close. This cannot fail.
906 * Only if device is UP
907 *
908 * We allow it to be called even after a DETACH hot-plug
909 * event.
910 */
911 if (dev->stop)
912 dev->stop(dev);
913
914 /*
915 * Device is now down.
916 */
917
918 dev->flags &= ~IFF_UP;
919
920 /*
921 * Tell people we are down
922 */
923 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
924
925 return 0;
926}
927
928
929/*
930 * Device change register/unregister. These are not inline or static
931 * as we export them to the world.
932 */
933
934/**
935 * register_netdevice_notifier - register a network notifier block
936 * @nb: notifier
937 *
938 * Register a notifier to be called when network device events occur.
939 * The notifier passed is linked into the kernel structures and must
940 * not be reused until it has been unregistered. A negative errno code
941 * is returned on a failure.
942 *
943 * When registered all registration and up events are replayed
944 * to the new notifier to allow device to have a race free
945 * view of the network device list.
946 */
947
948int register_netdevice_notifier(struct notifier_block *nb)
949{
950 struct net_device *dev;
951 int err;
952
953 rtnl_lock();
954 err = notifier_chain_register(&netdev_chain, nb);
955 if (!err) {
956 for (dev = dev_base; dev; dev = dev->next) {
957 nb->notifier_call(nb, NETDEV_REGISTER, dev);
958
959 if (dev->flags & IFF_UP)
960 nb->notifier_call(nb, NETDEV_UP, dev);
961 }
962 }
963 rtnl_unlock();
964 return err;
965}
966
967/**
968 * unregister_netdevice_notifier - unregister a network notifier block
969 * @nb: notifier
970 *
971 * Unregister a notifier previously registered by
972 * register_netdevice_notifier(). The notifier is unlinked into the
973 * kernel structures and may then be reused. A negative errno code
974 * is returned on a failure.
975 */
976
977int unregister_netdevice_notifier(struct notifier_block *nb)
978{
979 return notifier_chain_unregister(&netdev_chain, nb);
980}
981
982/**
983 * call_netdevice_notifiers - call all network notifier blocks
984 * @val: value passed unmodified to notifier function
985 * @v: pointer passed unmodified to notifier function
986 *
987 * Call all network notifier blocks. Parameters and return value
988 * are as for notifier_call_chain().
989 */
990
991int call_netdevice_notifiers(unsigned long val, void *v)
992{
993 return notifier_call_chain(&netdev_chain, val, v);
994}
995
996/* When > 0 there are consumers of rx skb time stamps */
997static atomic_t netstamp_needed = ATOMIC_INIT(0);
998
999void net_enable_timestamp(void)
1000{
1001 atomic_inc(&netstamp_needed);
1002}
1003
1004void net_disable_timestamp(void)
1005{
1006 atomic_dec(&netstamp_needed);
1007}
1008
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001009void __net_timestamp(struct sk_buff *skb)
1010{
1011 struct timeval tv;
1012
1013 do_gettimeofday(&tv);
1014 skb_set_timestamp(skb, &tv);
1015}
1016EXPORT_SYMBOL(__net_timestamp);
1017
1018static inline void net_timestamp(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019{
1020 if (atomic_read(&netstamp_needed))
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001021 __net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 else {
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001023 skb->tstamp.off_sec = 0;
1024 skb->tstamp.off_usec = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 }
1026}
1027
1028/*
1029 * Support routine. Sends outgoing frames to any network
1030 * taps currently in use.
1031 */
1032
1033void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1034{
1035 struct packet_type *ptype;
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001036
1037 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
1039 rcu_read_lock();
1040 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1041 /* Never send packets back to the socket
1042 * they originated from - MvS (miquels@drinkel.ow.org)
1043 */
1044 if ((ptype->dev == dev || !ptype->dev) &&
1045 (ptype->af_packet_priv == NULL ||
1046 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1047 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1048 if (!skb2)
1049 break;
1050
1051 /* skb->nh should be correctly
1052 set by sender, so that the second statement is
1053 just protection against buggy protocols.
1054 */
1055 skb2->mac.raw = skb2->data;
1056
1057 if (skb2->nh.raw < skb2->data ||
1058 skb2->nh.raw > skb2->tail) {
1059 if (net_ratelimit())
1060 printk(KERN_CRIT "protocol %04x is "
1061 "buggy, dev %s\n",
1062 skb2->protocol, dev->name);
1063 skb2->nh.raw = skb2->data;
1064 }
1065
1066 skb2->h.raw = skb2->nh.raw;
1067 skb2->pkt_type = PACKET_OUTGOING;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001068 ptype->func(skb2, skb->dev, ptype, skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 }
1070 }
1071 rcu_read_unlock();
1072}
1073
1074/*
1075 * Invalidate hardware checksum when packet is to be mangled, and
1076 * complete checksum manually on outgoing path.
1077 */
1078int skb_checksum_help(struct sk_buff *skb, int inward)
1079{
1080 unsigned int csum;
1081 int ret = 0, offset = skb->h.raw - skb->data;
1082
1083 if (inward) {
1084 skb->ip_summed = CHECKSUM_NONE;
1085 goto out;
1086 }
1087
1088 if (skb_cloned(skb)) {
1089 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1090 if (ret)
1091 goto out;
1092 }
1093
Kris Katterjohn09a62662006-01-08 22:24:28 -08001094 BUG_ON(offset > (int)skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1096
1097 offset = skb->tail - skb->h.raw;
Kris Katterjohn09a62662006-01-08 22:24:28 -08001098 BUG_ON(offset <= 0);
1099 BUG_ON(skb->csum + 2 > offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
1101 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1102 skb->ip_summed = CHECKSUM_NONE;
1103out:
1104 return ret;
1105}
1106
Herbert Xufb286bb2005-11-10 13:01:24 -08001107/* Take action when hardware reception checksum errors are detected. */
1108#ifdef CONFIG_BUG
1109void netdev_rx_csum_fault(struct net_device *dev)
1110{
1111 if (net_ratelimit()) {
Stephen Hemminger246a4212005-12-08 15:21:39 -08001112 printk(KERN_ERR "%s: hw csum failure.\n",
1113 dev ? dev->name : "<unknown>");
Herbert Xufb286bb2005-11-10 13:01:24 -08001114 dump_stack();
1115 }
1116}
1117EXPORT_SYMBOL(netdev_rx_csum_fault);
1118#endif
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120#ifdef CONFIG_HIGHMEM
1121/* Actually, we should eliminate this check as soon as we know, that:
1122 * 1. IOMMU is present and allows to map all the memory.
1123 * 2. No high memory really exists on this machine.
1124 */
1125
1126static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1127{
1128 int i;
1129
1130 if (dev->features & NETIF_F_HIGHDMA)
1131 return 0;
1132
1133 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1134 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1135 return 1;
1136
1137 return 0;
1138}
1139#else
1140#define illegal_highdma(dev, skb) (0)
1141#endif
1142
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143/* Keep head the same: replace data */
Al Virodd0fc662005-10-07 07:46:04 +01001144int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145{
1146 unsigned int size;
1147 u8 *data;
1148 long offset;
1149 struct skb_shared_info *ninfo;
1150 int headerlen = skb->data - skb->head;
1151 int expand = (skb->tail + skb->data_len) - skb->end;
1152
1153 if (skb_shared(skb))
1154 BUG();
1155
1156 if (expand <= 0)
1157 expand = 0;
1158
1159 size = skb->end - skb->head + expand;
1160 size = SKB_DATA_ALIGN(size);
1161 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1162 if (!data)
1163 return -ENOMEM;
1164
1165 /* Copy entire thing */
1166 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1167 BUG();
1168
1169 /* Set up shinfo */
1170 ninfo = (struct skb_shared_info*)(data + size);
1171 atomic_set(&ninfo->dataref, 1);
1172 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1173 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1174 ninfo->nr_frags = 0;
1175 ninfo->frag_list = NULL;
1176
1177 /* Offset between the two in bytes */
1178 offset = data - skb->head;
1179
1180 /* Free old data. */
1181 skb_release_data(skb);
1182
1183 skb->head = data;
1184 skb->end = data + size;
1185
1186 /* Set up new pointers */
1187 skb->h.raw += offset;
1188 skb->nh.raw += offset;
1189 skb->mac.raw += offset;
1190 skb->tail += offset;
1191 skb->data += offset;
1192
1193 /* We are no longer a clone, even if we were. */
1194 skb->cloned = 0;
1195
1196 skb->tail += skb->data_len;
1197 skb->data_len = 0;
1198 return 0;
1199}
1200
1201#define HARD_TX_LOCK(dev, cpu) { \
1202 if ((dev->features & NETIF_F_LLTX) == 0) { \
1203 spin_lock(&dev->xmit_lock); \
1204 dev->xmit_lock_owner = cpu; \
1205 } \
1206}
1207
1208#define HARD_TX_UNLOCK(dev) { \
1209 if ((dev->features & NETIF_F_LLTX) == 0) { \
1210 dev->xmit_lock_owner = -1; \
1211 spin_unlock(&dev->xmit_lock); \
1212 } \
1213}
1214
1215/**
1216 * dev_queue_xmit - transmit a buffer
1217 * @skb: buffer to transmit
1218 *
1219 * Queue a buffer for transmission to a network device. The caller must
1220 * have set the device and priority and built the buffer before calling
1221 * this function. The function can be called from an interrupt.
1222 *
1223 * A negative errno code is returned on a failure. A success does not
1224 * guarantee the frame will be transmitted as it may be dropped due
1225 * to congestion or traffic shaping.
Ben Greearaf191362005-04-24 20:12:36 -07001226 *
1227 * -----------------------------------------------------------------------------------
1228 * I notice this method can also return errors from the queue disciplines,
1229 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1230 * be positive.
1231 *
1232 * Regardless of the return value, the skb is consumed, so it is currently
1233 * difficult to retry a send to this method. (You can bump the ref count
1234 * before sending to hold a reference for retry if you are careful.)
1235 *
1236 * When calling this method, interrupts MUST be enabled. This is because
1237 * the BH enable code must have IRQs enabled so that it will not deadlock.
1238 * --BLG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 */
1240
1241int dev_queue_xmit(struct sk_buff *skb)
1242{
1243 struct net_device *dev = skb->dev;
1244 struct Qdisc *q;
1245 int rc = -ENOMEM;
1246
1247 if (skb_shinfo(skb)->frag_list &&
1248 !(dev->features & NETIF_F_FRAGLIST) &&
1249 __skb_linearize(skb, GFP_ATOMIC))
1250 goto out_kfree_skb;
1251
1252 /* Fragmented skb is linearized if device does not support SG,
1253 * or if at least one of fragments is in highmem and device
1254 * does not support DMA from it.
1255 */
1256 if (skb_shinfo(skb)->nr_frags &&
1257 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1258 __skb_linearize(skb, GFP_ATOMIC))
1259 goto out_kfree_skb;
1260
1261 /* If packet is not checksummed and device does not support
1262 * checksumming for this protocol, complete checksumming here.
1263 */
1264 if (skb->ip_summed == CHECKSUM_HW &&
1265 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1266 (!(dev->features & NETIF_F_IP_CSUM) ||
1267 skb->protocol != htons(ETH_P_IP))))
1268 if (skb_checksum_help(skb, 0))
1269 goto out_kfree_skb;
1270
Eric Dumazet2d7ceec2005-09-27 15:22:58 -07001271 spin_lock_prefetch(&dev->queue_lock);
1272
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 /* Disable soft irqs for various locks below. Also
1274 * stops preemption for RCU.
1275 */
1276 local_bh_disable();
1277
1278 /* Updates of qdisc are serialized by queue_lock.
1279 * The struct Qdisc which is pointed to by qdisc is now a
1280 * rcu structure - it may be accessed without acquiring
1281 * a lock (but the structure may be stale.) The freeing of the
1282 * qdisc will be deferred until it's known that there are no
1283 * more references to it.
1284 *
1285 * If the qdisc has an enqueue function, we still need to
1286 * hold the queue_lock before calling it, since queue_lock
1287 * also serializes access to the device queue.
1288 */
1289
1290 q = rcu_dereference(dev->qdisc);
1291#ifdef CONFIG_NET_CLS_ACT
1292 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1293#endif
1294 if (q->enqueue) {
1295 /* Grab device queue */
1296 spin_lock(&dev->queue_lock);
1297
1298 rc = q->enqueue(skb, q);
1299
1300 qdisc_run(dev);
1301
1302 spin_unlock(&dev->queue_lock);
1303 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1304 goto out;
1305 }
1306
1307 /* The device has no queue. Common case for software devices:
1308 loopback, all the sorts of tunnels...
1309
1310 Really, it is unlikely that xmit_lock protection is necessary here.
1311 (f.e. loopback and IP tunnels are clean ignoring statistics
1312 counters.)
1313 However, it is possible, that they rely on protection
1314 made by us here.
1315
1316 Check this and shot the lock. It is not prone from deadlocks.
1317 Either shot noqueue qdisc, it is even simpler 8)
1318 */
1319 if (dev->flags & IFF_UP) {
1320 int cpu = smp_processor_id(); /* ok because BHs are off */
1321
1322 if (dev->xmit_lock_owner != cpu) {
1323
1324 HARD_TX_LOCK(dev, cpu);
1325
1326 if (!netif_queue_stopped(dev)) {
1327 if (netdev_nit)
1328 dev_queue_xmit_nit(skb, dev);
1329
1330 rc = 0;
1331 if (!dev->hard_start_xmit(skb, dev)) {
1332 HARD_TX_UNLOCK(dev);
1333 goto out;
1334 }
1335 }
1336 HARD_TX_UNLOCK(dev);
1337 if (net_ratelimit())
1338 printk(KERN_CRIT "Virtual device %s asks to "
1339 "queue packet!\n", dev->name);
1340 } else {
1341 /* Recursion is detected! It is possible,
1342 * unfortunately */
1343 if (net_ratelimit())
1344 printk(KERN_CRIT "Dead loop on virtual device "
1345 "%s, fix it urgently!\n", dev->name);
1346 }
1347 }
1348
1349 rc = -ENETDOWN;
1350 local_bh_enable();
1351
1352out_kfree_skb:
1353 kfree_skb(skb);
1354 return rc;
1355out:
1356 local_bh_enable();
1357 return rc;
1358}
1359
1360
1361/*=======================================================================
1362 Receiver routines
1363 =======================================================================*/
1364
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001365int netdev_max_backlog = 1000;
1366int netdev_budget = 300;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367int weight_p = 64; /* old backlog weight */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
1369DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1370
1371
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372/**
1373 * netif_rx - post buffer to the network code
1374 * @skb: buffer to post
1375 *
1376 * This function receives a packet from a device driver and queues it for
1377 * the upper (protocol) levels to process. It always succeeds. The buffer
1378 * may be dropped during processing for congestion control or by the
1379 * protocol layers.
1380 *
1381 * return values:
1382 * NET_RX_SUCCESS (no congestion)
1383 * NET_RX_CN_LOW (low congestion)
1384 * NET_RX_CN_MOD (moderate congestion)
1385 * NET_RX_CN_HIGH (high congestion)
1386 * NET_RX_DROP (packet was dropped)
1387 *
1388 */
1389
1390int netif_rx(struct sk_buff *skb)
1391{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 struct softnet_data *queue;
1393 unsigned long flags;
1394
1395 /* if netpoll wants it, pretend we never saw it */
1396 if (netpoll_rx(skb))
1397 return NET_RX_DROP;
1398
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001399 if (!skb->tstamp.off_sec)
1400 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
1402 /*
1403 * The code is rearranged so that the path is the most
1404 * short when CPU is congested, but is still operating.
1405 */
1406 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 queue = &__get_cpu_var(softnet_data);
1408
1409 __get_cpu_var(netdev_rx_stat).total++;
1410 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1411 if (queue->input_pkt_queue.qlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412enqueue:
1413 dev_hold(skb->dev);
1414 __skb_queue_tail(&queue->input_pkt_queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 local_irq_restore(flags);
Stephen Hemminger34008d82005-06-23 20:10:00 -07001416 return NET_RX_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 }
1418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 netif_rx_schedule(&queue->backlog_dev);
1420 goto enqueue;
1421 }
1422
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 __get_cpu_var(netdev_rx_stat).dropped++;
1424 local_irq_restore(flags);
1425
1426 kfree_skb(skb);
1427 return NET_RX_DROP;
1428}
1429
1430int netif_rx_ni(struct sk_buff *skb)
1431{
1432 int err;
1433
1434 preempt_disable();
1435 err = netif_rx(skb);
1436 if (local_softirq_pending())
1437 do_softirq();
1438 preempt_enable();
1439
1440 return err;
1441}
1442
1443EXPORT_SYMBOL(netif_rx_ni);
1444
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001445static inline struct net_device *skb_bond(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446{
1447 struct net_device *dev = skb->dev;
1448
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001449 if (dev->master) {
1450 /*
1451 * On bonding slaves other than the currently active
1452 * slave, suppress duplicates except for 802.3ad
1453 * ETH_P_SLOW and alb non-mcast/bcast.
1454 */
1455 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1456 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1457 if (skb->pkt_type != PACKET_BROADCAST &&
1458 skb->pkt_type != PACKET_MULTICAST)
1459 goto keep;
1460 }
1461
1462 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1463 skb->protocol == __constant_htons(ETH_P_SLOW))
1464 goto keep;
1465
1466 kfree_skb(skb);
1467 return NULL;
1468 }
1469keep:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 skb->dev = dev->master;
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001471 }
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001472
1473 return dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474}
1475
1476static void net_tx_action(struct softirq_action *h)
1477{
1478 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1479
1480 if (sd->completion_queue) {
1481 struct sk_buff *clist;
1482
1483 local_irq_disable();
1484 clist = sd->completion_queue;
1485 sd->completion_queue = NULL;
1486 local_irq_enable();
1487
1488 while (clist) {
1489 struct sk_buff *skb = clist;
1490 clist = clist->next;
1491
1492 BUG_TRAP(!atomic_read(&skb->users));
1493 __kfree_skb(skb);
1494 }
1495 }
1496
1497 if (sd->output_queue) {
1498 struct net_device *head;
1499
1500 local_irq_disable();
1501 head = sd->output_queue;
1502 sd->output_queue = NULL;
1503 local_irq_enable();
1504
1505 while (head) {
1506 struct net_device *dev = head;
1507 head = head->next_sched;
1508
1509 smp_mb__before_clear_bit();
1510 clear_bit(__LINK_STATE_SCHED, &dev->state);
1511
1512 if (spin_trylock(&dev->queue_lock)) {
1513 qdisc_run(dev);
1514 spin_unlock(&dev->queue_lock);
1515 } else {
1516 netif_schedule(dev);
1517 }
1518 }
1519 }
1520}
1521
1522static __inline__ int deliver_skb(struct sk_buff *skb,
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001523 struct packet_type *pt_prev,
1524 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
1526 atomic_inc(&skb->users);
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001527 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528}
1529
1530#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1531int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1532struct net_bridge;
1533struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1534 unsigned char *addr);
1535void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1536
1537static __inline__ int handle_bridge(struct sk_buff **pskb,
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001538 struct packet_type **pt_prev, int *ret,
1539 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540{
1541 struct net_bridge_port *port;
1542
1543 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1544 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1545 return 0;
1546
1547 if (*pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001548 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 *pt_prev = NULL;
1550 }
1551
1552 return br_handle_frame_hook(port, pskb);
1553}
1554#else
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001555#define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556#endif
1557
1558#ifdef CONFIG_NET_CLS_ACT
1559/* TODO: Maybe we should just force sch_ingress to be compiled in
1560 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1561 * a compare and 2 stores extra right now if we dont have it on
1562 * but have CONFIG_NET_CLS_ACT
1563 * NOTE: This doesnt stop any functionality; if you dont have
1564 * the ingress scheduler, you just cant add policies on ingress.
1565 *
1566 */
1567static int ing_filter(struct sk_buff *skb)
1568{
1569 struct Qdisc *q;
1570 struct net_device *dev = skb->dev;
1571 int result = TC_ACT_OK;
1572
1573 if (dev->qdisc_ingress) {
1574 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1575 if (MAX_RED_LOOP < ttl++) {
1576 printk("Redir loop detected Dropping packet (%s->%s)\n",
David S. Miller86e65da2005-08-09 19:36:29 -07001577 skb->input_dev->name, skb->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 return TC_ACT_SHOT;
1579 }
1580
1581 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1582
1583 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
David S. Miller86e65da2005-08-09 19:36:29 -07001584
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 spin_lock(&dev->ingress_lock);
1586 if ((q = dev->qdisc_ingress) != NULL)
1587 result = q->enqueue(skb, q);
1588 spin_unlock(&dev->ingress_lock);
1589
1590 }
1591
1592 return result;
1593}
1594#endif
1595
1596int netif_receive_skb(struct sk_buff *skb)
1597{
1598 struct packet_type *ptype, *pt_prev;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001599 struct net_device *orig_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 int ret = NET_RX_DROP;
1601 unsigned short type;
1602
1603 /* if we've gotten here through NAPI, check netpoll */
1604 if (skb->dev->poll && netpoll_rx(skb))
1605 return NET_RX_DROP;
1606
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001607 if (!skb->tstamp.off_sec)
1608 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
David S. Miller86e65da2005-08-09 19:36:29 -07001610 if (!skb->input_dev)
1611 skb->input_dev = skb->dev;
1612
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001613 orig_dev = skb_bond(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001615 if (!orig_dev)
1616 return NET_RX_DROP;
1617
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 __get_cpu_var(netdev_rx_stat).total++;
1619
1620 skb->h.raw = skb->nh.raw = skb->data;
1621 skb->mac_len = skb->nh.raw - skb->mac.raw;
1622
1623 pt_prev = NULL;
1624
1625 rcu_read_lock();
1626
1627#ifdef CONFIG_NET_CLS_ACT
1628 if (skb->tc_verd & TC_NCLS) {
1629 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1630 goto ncls;
1631 }
1632#endif
1633
1634 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1635 if (!ptype->dev || ptype->dev == skb->dev) {
1636 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001637 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 pt_prev = ptype;
1639 }
1640 }
1641
1642#ifdef CONFIG_NET_CLS_ACT
1643 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001644 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 pt_prev = NULL; /* noone else should process this after*/
1646 } else {
1647 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1648 }
1649
1650 ret = ing_filter(skb);
1651
1652 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1653 kfree_skb(skb);
1654 goto out;
1655 }
1656
1657 skb->tc_verd = 0;
1658ncls:
1659#endif
1660
1661 handle_diverter(skb);
1662
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001663 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 goto out;
1665
1666 type = skb->protocol;
1667 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1668 if (ptype->type == type &&
1669 (!ptype->dev || ptype->dev == skb->dev)) {
1670 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001671 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 pt_prev = ptype;
1673 }
1674 }
1675
1676 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001677 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 } else {
1679 kfree_skb(skb);
1680 /* Jamal, now you will not able to escape explaining
1681 * me how you were going to use this. :-)
1682 */
1683 ret = NET_RX_DROP;
1684 }
1685
1686out:
1687 rcu_read_unlock();
1688 return ret;
1689}
1690
1691static int process_backlog(struct net_device *backlog_dev, int *budget)
1692{
1693 int work = 0;
1694 int quota = min(backlog_dev->quota, *budget);
1695 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1696 unsigned long start_time = jiffies;
1697
Stephen Hemmingere3876602005-06-08 14:56:01 -07001698 backlog_dev->weight = weight_p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 for (;;) {
1700 struct sk_buff *skb;
1701 struct net_device *dev;
1702
1703 local_irq_disable();
1704 skb = __skb_dequeue(&queue->input_pkt_queue);
1705 if (!skb)
1706 goto job_done;
1707 local_irq_enable();
1708
1709 dev = skb->dev;
1710
1711 netif_receive_skb(skb);
1712
1713 dev_put(dev);
1714
1715 work++;
1716
1717 if (work >= quota || jiffies - start_time > 1)
1718 break;
1719
1720 }
1721
1722 backlog_dev->quota -= work;
1723 *budget -= work;
1724 return -1;
1725
1726job_done:
1727 backlog_dev->quota -= work;
1728 *budget -= work;
1729
1730 list_del(&backlog_dev->poll_list);
1731 smp_mb__before_clear_bit();
1732 netif_poll_enable(backlog_dev);
1733
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 local_irq_enable();
1735 return 0;
1736}
1737
1738static void net_rx_action(struct softirq_action *h)
1739{
1740 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1741 unsigned long start_time = jiffies;
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001742 int budget = netdev_budget;
Matt Mackall53fb95d2005-08-11 19:27:43 -07001743 void *have;
1744
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 local_irq_disable();
1746
1747 while (!list_empty(&queue->poll_list)) {
1748 struct net_device *dev;
1749
1750 if (budget <= 0 || jiffies - start_time > 1)
1751 goto softnet_break;
1752
1753 local_irq_enable();
1754
1755 dev = list_entry(queue->poll_list.next,
1756 struct net_device, poll_list);
Matt Mackall53fb95d2005-08-11 19:27:43 -07001757 have = netpoll_poll_lock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758
1759 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
Matt Mackall53fb95d2005-08-11 19:27:43 -07001760 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 local_irq_disable();
1762 list_del(&dev->poll_list);
1763 list_add_tail(&dev->poll_list, &queue->poll_list);
1764 if (dev->quota < 0)
1765 dev->quota += dev->weight;
1766 else
1767 dev->quota = dev->weight;
1768 } else {
Matt Mackall53fb95d2005-08-11 19:27:43 -07001769 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 dev_put(dev);
1771 local_irq_disable();
1772 }
1773 }
1774out:
1775 local_irq_enable();
1776 return;
1777
1778softnet_break:
1779 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1780 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1781 goto out;
1782}
1783
1784static gifconf_func_t * gifconf_list [NPROTO];
1785
1786/**
1787 * register_gifconf - register a SIOCGIF handler
1788 * @family: Address family
1789 * @gifconf: Function handler
1790 *
1791 * Register protocol dependent address dumping routines. The handler
1792 * that is passed must not be freed or reused until it has been replaced
1793 * by another handler.
1794 */
1795int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1796{
1797 if (family >= NPROTO)
1798 return -EINVAL;
1799 gifconf_list[family] = gifconf;
1800 return 0;
1801}
1802
1803
1804/*
1805 * Map an interface index to its name (SIOCGIFNAME)
1806 */
1807
1808/*
1809 * We need this ioctl for efficient implementation of the
1810 * if_indextoname() function required by the IPv6 API. Without
1811 * it, we would have to search all the interfaces to find a
1812 * match. --pb
1813 */
1814
1815static int dev_ifname(struct ifreq __user *arg)
1816{
1817 struct net_device *dev;
1818 struct ifreq ifr;
1819
1820 /*
1821 * Fetch the caller's info block.
1822 */
1823
1824 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1825 return -EFAULT;
1826
1827 read_lock(&dev_base_lock);
1828 dev = __dev_get_by_index(ifr.ifr_ifindex);
1829 if (!dev) {
1830 read_unlock(&dev_base_lock);
1831 return -ENODEV;
1832 }
1833
1834 strcpy(ifr.ifr_name, dev->name);
1835 read_unlock(&dev_base_lock);
1836
1837 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1838 return -EFAULT;
1839 return 0;
1840}
1841
1842/*
1843 * Perform a SIOCGIFCONF call. This structure will change
1844 * size eventually, and there is nothing I can do about it.
1845 * Thus we will need a 'compatibility mode'.
1846 */
1847
1848static int dev_ifconf(char __user *arg)
1849{
1850 struct ifconf ifc;
1851 struct net_device *dev;
1852 char __user *pos;
1853 int len;
1854 int total;
1855 int i;
1856
1857 /*
1858 * Fetch the caller's info block.
1859 */
1860
1861 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1862 return -EFAULT;
1863
1864 pos = ifc.ifc_buf;
1865 len = ifc.ifc_len;
1866
1867 /*
1868 * Loop over the interfaces, and write an info block for each.
1869 */
1870
1871 total = 0;
1872 for (dev = dev_base; dev; dev = dev->next) {
1873 for (i = 0; i < NPROTO; i++) {
1874 if (gifconf_list[i]) {
1875 int done;
1876 if (!pos)
1877 done = gifconf_list[i](dev, NULL, 0);
1878 else
1879 done = gifconf_list[i](dev, pos + total,
1880 len - total);
1881 if (done < 0)
1882 return -EFAULT;
1883 total += done;
1884 }
1885 }
1886 }
1887
1888 /*
1889 * All done. Write the updated control block back to the caller.
1890 */
1891 ifc.ifc_len = total;
1892
1893 /*
1894 * Both BSD and Solaris return 0 here, so we do too.
1895 */
1896 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1897}
1898
1899#ifdef CONFIG_PROC_FS
1900/*
1901 * This is invoked by the /proc filesystem handler to display a device
1902 * in detail.
1903 */
1904static __inline__ struct net_device *dev_get_idx(loff_t pos)
1905{
1906 struct net_device *dev;
1907 loff_t i;
1908
1909 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1910
1911 return i == pos ? dev : NULL;
1912}
1913
1914void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1915{
1916 read_lock(&dev_base_lock);
1917 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1918}
1919
1920void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1921{
1922 ++*pos;
1923 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1924}
1925
1926void dev_seq_stop(struct seq_file *seq, void *v)
1927{
1928 read_unlock(&dev_base_lock);
1929}
1930
1931static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1932{
1933 if (dev->get_stats) {
1934 struct net_device_stats *stats = dev->get_stats(dev);
1935
1936 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1937 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1938 dev->name, stats->rx_bytes, stats->rx_packets,
1939 stats->rx_errors,
1940 stats->rx_dropped + stats->rx_missed_errors,
1941 stats->rx_fifo_errors,
1942 stats->rx_length_errors + stats->rx_over_errors +
1943 stats->rx_crc_errors + stats->rx_frame_errors,
1944 stats->rx_compressed, stats->multicast,
1945 stats->tx_bytes, stats->tx_packets,
1946 stats->tx_errors, stats->tx_dropped,
1947 stats->tx_fifo_errors, stats->collisions,
1948 stats->tx_carrier_errors +
1949 stats->tx_aborted_errors +
1950 stats->tx_window_errors +
1951 stats->tx_heartbeat_errors,
1952 stats->tx_compressed);
1953 } else
1954 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1955}
1956
1957/*
1958 * Called from the PROCfs module. This now uses the new arbitrary sized
1959 * /proc/net interface to create /proc/net/dev
1960 */
1961static int dev_seq_show(struct seq_file *seq, void *v)
1962{
1963 if (v == SEQ_START_TOKEN)
1964 seq_puts(seq, "Inter-| Receive "
1965 " | Transmit\n"
1966 " face |bytes packets errs drop fifo frame "
1967 "compressed multicast|bytes packets errs "
1968 "drop fifo colls carrier compressed\n");
1969 else
1970 dev_seq_printf_stats(seq, v);
1971 return 0;
1972}
1973
1974static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1975{
1976 struct netif_rx_stats *rc = NULL;
1977
1978 while (*pos < NR_CPUS)
1979 if (cpu_online(*pos)) {
1980 rc = &per_cpu(netdev_rx_stat, *pos);
1981 break;
1982 } else
1983 ++*pos;
1984 return rc;
1985}
1986
1987static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1988{
1989 return softnet_get_online(pos);
1990}
1991
1992static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1993{
1994 ++*pos;
1995 return softnet_get_online(pos);
1996}
1997
1998static void softnet_seq_stop(struct seq_file *seq, void *v)
1999{
2000}
2001
2002static int softnet_seq_show(struct seq_file *seq, void *v)
2003{
2004 struct netif_rx_stats *s = v;
2005
2006 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
Stephen Hemminger31aa02c2005-06-23 20:12:48 -07002007 s->total, s->dropped, s->time_squeeze, 0,
Stephen Hemmingerc1ebcdb2005-06-23 20:08:59 -07002008 0, 0, 0, 0, /* was fastroute */
2009 s->cpu_collision );
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 return 0;
2011}
2012
2013static struct seq_operations dev_seq_ops = {
2014 .start = dev_seq_start,
2015 .next = dev_seq_next,
2016 .stop = dev_seq_stop,
2017 .show = dev_seq_show,
2018};
2019
2020static int dev_seq_open(struct inode *inode, struct file *file)
2021{
2022 return seq_open(file, &dev_seq_ops);
2023}
2024
2025static struct file_operations dev_seq_fops = {
2026 .owner = THIS_MODULE,
2027 .open = dev_seq_open,
2028 .read = seq_read,
2029 .llseek = seq_lseek,
2030 .release = seq_release,
2031};
2032
2033static struct seq_operations softnet_seq_ops = {
2034 .start = softnet_seq_start,
2035 .next = softnet_seq_next,
2036 .stop = softnet_seq_stop,
2037 .show = softnet_seq_show,
2038};
2039
2040static int softnet_seq_open(struct inode *inode, struct file *file)
2041{
2042 return seq_open(file, &softnet_seq_ops);
2043}
2044
2045static struct file_operations softnet_seq_fops = {
2046 .owner = THIS_MODULE,
2047 .open = softnet_seq_open,
2048 .read = seq_read,
2049 .llseek = seq_lseek,
2050 .release = seq_release,
2051};
2052
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002053#ifdef CONFIG_WIRELESS_EXT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054extern int wireless_proc_init(void);
2055#else
2056#define wireless_proc_init() 0
2057#endif
2058
2059static int __init dev_proc_init(void)
2060{
2061 int rc = -ENOMEM;
2062
2063 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2064 goto out;
2065 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2066 goto out_dev;
2067 if (wireless_proc_init())
2068 goto out_softnet;
2069 rc = 0;
2070out:
2071 return rc;
2072out_softnet:
2073 proc_net_remove("softnet_stat");
2074out_dev:
2075 proc_net_remove("dev");
2076 goto out;
2077}
2078#else
2079#define dev_proc_init() 0
2080#endif /* CONFIG_PROC_FS */
2081
2082
2083/**
2084 * netdev_set_master - set up master/slave pair
2085 * @slave: slave device
2086 * @master: new master device
2087 *
2088 * Changes the master device of the slave. Pass %NULL to break the
2089 * bonding. The caller must hold the RTNL semaphore. On a failure
2090 * a negative errno code is returned. On success the reference counts
2091 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2092 * function returns zero.
2093 */
2094int netdev_set_master(struct net_device *slave, struct net_device *master)
2095{
2096 struct net_device *old = slave->master;
2097
2098 ASSERT_RTNL();
2099
2100 if (master) {
2101 if (old)
2102 return -EBUSY;
2103 dev_hold(master);
2104 }
2105
2106 slave->master = master;
2107
2108 synchronize_net();
2109
2110 if (old)
2111 dev_put(old);
2112
2113 if (master)
2114 slave->flags |= IFF_SLAVE;
2115 else
2116 slave->flags &= ~IFF_SLAVE;
2117
2118 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2119 return 0;
2120}
2121
2122/**
2123 * dev_set_promiscuity - update promiscuity count on a device
2124 * @dev: device
2125 * @inc: modifier
2126 *
2127 * Add or remove promsicuity from a device. While the count in the device
2128 * remains above zero the interface remains promiscuous. Once it hits zero
2129 * the device reverts back to normal filtering operation. A negative inc
2130 * value is used to drop promiscuity on the device.
2131 */
2132void dev_set_promiscuity(struct net_device *dev, int inc)
2133{
2134 unsigned short old_flags = dev->flags;
2135
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 if ((dev->promiscuity += inc) == 0)
2137 dev->flags &= ~IFF_PROMISC;
David Chau52609c02005-07-05 15:11:06 -07002138 else
2139 dev->flags |= IFF_PROMISC;
2140 if (dev->flags != old_flags) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 dev_mc_upload(dev);
2142 printk(KERN_INFO "device %s %s promiscuous mode\n",
2143 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2144 "left");
2145 }
2146}
2147
2148/**
2149 * dev_set_allmulti - update allmulti count on a device
2150 * @dev: device
2151 * @inc: modifier
2152 *
2153 * Add or remove reception of all multicast frames to a device. While the
2154 * count in the device remains above zero the interface remains listening
2155 * to all interfaces. Once it hits zero the device reverts back to normal
2156 * filtering operation. A negative @inc value is used to drop the counter
2157 * when releasing a resource needing all multicasts.
2158 */
2159
2160void dev_set_allmulti(struct net_device *dev, int inc)
2161{
2162 unsigned short old_flags = dev->flags;
2163
2164 dev->flags |= IFF_ALLMULTI;
2165 if ((dev->allmulti += inc) == 0)
2166 dev->flags &= ~IFF_ALLMULTI;
2167 if (dev->flags ^ old_flags)
2168 dev_mc_upload(dev);
2169}
2170
2171unsigned dev_get_flags(const struct net_device *dev)
2172{
2173 unsigned flags;
2174
2175 flags = (dev->flags & ~(IFF_PROMISC |
2176 IFF_ALLMULTI |
Stefan Rompfb00055a2006-03-20 17:09:11 -08002177 IFF_RUNNING |
2178 IFF_LOWER_UP |
2179 IFF_DORMANT)) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 (dev->gflags & (IFF_PROMISC |
2181 IFF_ALLMULTI));
2182
Stefan Rompfb00055a2006-03-20 17:09:11 -08002183 if (netif_running(dev)) {
2184 if (netif_oper_up(dev))
2185 flags |= IFF_RUNNING;
2186 if (netif_carrier_ok(dev))
2187 flags |= IFF_LOWER_UP;
2188 if (netif_dormant(dev))
2189 flags |= IFF_DORMANT;
2190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191
2192 return flags;
2193}
2194
2195int dev_change_flags(struct net_device *dev, unsigned flags)
2196{
2197 int ret;
2198 int old_flags = dev->flags;
2199
2200 /*
2201 * Set the flags on our device.
2202 */
2203
2204 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2205 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2206 IFF_AUTOMEDIA)) |
2207 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2208 IFF_ALLMULTI));
2209
2210 /*
2211 * Load in the correct multicast list now the flags have changed.
2212 */
2213
2214 dev_mc_upload(dev);
2215
2216 /*
2217 * Have we downed the interface. We handle IFF_UP ourselves
2218 * according to user attempts to set it, rather than blindly
2219 * setting it.
2220 */
2221
2222 ret = 0;
2223 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2224 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2225
2226 if (!ret)
2227 dev_mc_upload(dev);
2228 }
2229
2230 if (dev->flags & IFF_UP &&
2231 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2232 IFF_VOLATILE)))
2233 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2234
2235 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2236 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2237 dev->gflags ^= IFF_PROMISC;
2238 dev_set_promiscuity(dev, inc);
2239 }
2240
2241 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2242 is important. Some (broken) drivers set IFF_PROMISC, when
2243 IFF_ALLMULTI is requested not asking us and not reporting.
2244 */
2245 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2246 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2247 dev->gflags ^= IFF_ALLMULTI;
2248 dev_set_allmulti(dev, inc);
2249 }
2250
2251 if (old_flags ^ dev->flags)
2252 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2253
2254 return ret;
2255}
2256
2257int dev_set_mtu(struct net_device *dev, int new_mtu)
2258{
2259 int err;
2260
2261 if (new_mtu == dev->mtu)
2262 return 0;
2263
2264 /* MTU must be positive. */
2265 if (new_mtu < 0)
2266 return -EINVAL;
2267
2268 if (!netif_device_present(dev))
2269 return -ENODEV;
2270
2271 err = 0;
2272 if (dev->change_mtu)
2273 err = dev->change_mtu(dev, new_mtu);
2274 else
2275 dev->mtu = new_mtu;
2276 if (!err && dev->flags & IFF_UP)
2277 notifier_call_chain(&netdev_chain,
2278 NETDEV_CHANGEMTU, dev);
2279 return err;
2280}
2281
2282int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2283{
2284 int err;
2285
2286 if (!dev->set_mac_address)
2287 return -EOPNOTSUPP;
2288 if (sa->sa_family != dev->type)
2289 return -EINVAL;
2290 if (!netif_device_present(dev))
2291 return -ENODEV;
2292 err = dev->set_mac_address(dev, sa);
2293 if (!err)
2294 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2295 return err;
2296}
2297
2298/*
2299 * Perform the SIOCxIFxxx calls.
2300 */
2301static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2302{
2303 int err;
2304 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2305
2306 if (!dev)
2307 return -ENODEV;
2308
2309 switch (cmd) {
2310 case SIOCGIFFLAGS: /* Get interface flags */
2311 ifr->ifr_flags = dev_get_flags(dev);
2312 return 0;
2313
2314 case SIOCSIFFLAGS: /* Set interface flags */
2315 return dev_change_flags(dev, ifr->ifr_flags);
2316
2317 case SIOCGIFMETRIC: /* Get the metric on the interface
2318 (currently unused) */
2319 ifr->ifr_metric = 0;
2320 return 0;
2321
2322 case SIOCSIFMETRIC: /* Set the metric on the interface
2323 (currently unused) */
2324 return -EOPNOTSUPP;
2325
2326 case SIOCGIFMTU: /* Get the MTU of a device */
2327 ifr->ifr_mtu = dev->mtu;
2328 return 0;
2329
2330 case SIOCSIFMTU: /* Set the MTU of a device */
2331 return dev_set_mtu(dev, ifr->ifr_mtu);
2332
2333 case SIOCGIFHWADDR:
2334 if (!dev->addr_len)
2335 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2336 else
2337 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2338 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2339 ifr->ifr_hwaddr.sa_family = dev->type;
2340 return 0;
2341
2342 case SIOCSIFHWADDR:
2343 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2344
2345 case SIOCSIFHWBROADCAST:
2346 if (ifr->ifr_hwaddr.sa_family != dev->type)
2347 return -EINVAL;
2348 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2349 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2350 notifier_call_chain(&netdev_chain,
2351 NETDEV_CHANGEADDR, dev);
2352 return 0;
2353
2354 case SIOCGIFMAP:
2355 ifr->ifr_map.mem_start = dev->mem_start;
2356 ifr->ifr_map.mem_end = dev->mem_end;
2357 ifr->ifr_map.base_addr = dev->base_addr;
2358 ifr->ifr_map.irq = dev->irq;
2359 ifr->ifr_map.dma = dev->dma;
2360 ifr->ifr_map.port = dev->if_port;
2361 return 0;
2362
2363 case SIOCSIFMAP:
2364 if (dev->set_config) {
2365 if (!netif_device_present(dev))
2366 return -ENODEV;
2367 return dev->set_config(dev, &ifr->ifr_map);
2368 }
2369 return -EOPNOTSUPP;
2370
2371 case SIOCADDMULTI:
2372 if (!dev->set_multicast_list ||
2373 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2374 return -EINVAL;
2375 if (!netif_device_present(dev))
2376 return -ENODEV;
2377 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2378 dev->addr_len, 1);
2379
2380 case SIOCDELMULTI:
2381 if (!dev->set_multicast_list ||
2382 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2383 return -EINVAL;
2384 if (!netif_device_present(dev))
2385 return -ENODEV;
2386 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2387 dev->addr_len, 1);
2388
2389 case SIOCGIFINDEX:
2390 ifr->ifr_ifindex = dev->ifindex;
2391 return 0;
2392
2393 case SIOCGIFTXQLEN:
2394 ifr->ifr_qlen = dev->tx_queue_len;
2395 return 0;
2396
2397 case SIOCSIFTXQLEN:
2398 if (ifr->ifr_qlen < 0)
2399 return -EINVAL;
2400 dev->tx_queue_len = ifr->ifr_qlen;
2401 return 0;
2402
2403 case SIOCSIFNAME:
2404 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2405 return dev_change_name(dev, ifr->ifr_newname);
2406
2407 /*
2408 * Unknown or private ioctl
2409 */
2410
2411 default:
2412 if ((cmd >= SIOCDEVPRIVATE &&
2413 cmd <= SIOCDEVPRIVATE + 15) ||
2414 cmd == SIOCBONDENSLAVE ||
2415 cmd == SIOCBONDRELEASE ||
2416 cmd == SIOCBONDSETHWADDR ||
2417 cmd == SIOCBONDSLAVEINFOQUERY ||
2418 cmd == SIOCBONDINFOQUERY ||
2419 cmd == SIOCBONDCHANGEACTIVE ||
2420 cmd == SIOCGMIIPHY ||
2421 cmd == SIOCGMIIREG ||
2422 cmd == SIOCSMIIREG ||
2423 cmd == SIOCBRADDIF ||
2424 cmd == SIOCBRDELIF ||
2425 cmd == SIOCWANDEV) {
2426 err = -EOPNOTSUPP;
2427 if (dev->do_ioctl) {
2428 if (netif_device_present(dev))
2429 err = dev->do_ioctl(dev, ifr,
2430 cmd);
2431 else
2432 err = -ENODEV;
2433 }
2434 } else
2435 err = -EINVAL;
2436
2437 }
2438 return err;
2439}
2440
2441/*
2442 * This function handles all "interface"-type I/O control requests. The actual
2443 * 'doing' part of this is dev_ifsioc above.
2444 */
2445
2446/**
2447 * dev_ioctl - network device ioctl
2448 * @cmd: command to issue
2449 * @arg: pointer to a struct ifreq in user space
2450 *
2451 * Issue ioctl functions to devices. This is normally called by the
2452 * user space syscall interfaces but can sometimes be useful for
2453 * other purposes. The return value is the return from the syscall if
2454 * positive or a negative errno code on error.
2455 */
2456
2457int dev_ioctl(unsigned int cmd, void __user *arg)
2458{
2459 struct ifreq ifr;
2460 int ret;
2461 char *colon;
2462
2463 /* One special case: SIOCGIFCONF takes ifconf argument
2464 and requires shared lock, because it sleeps writing
2465 to user space.
2466 */
2467
2468 if (cmd == SIOCGIFCONF) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002469 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 ret = dev_ifconf((char __user *) arg);
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002471 rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 return ret;
2473 }
2474 if (cmd == SIOCGIFNAME)
2475 return dev_ifname((struct ifreq __user *)arg);
2476
2477 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2478 return -EFAULT;
2479
2480 ifr.ifr_name[IFNAMSIZ-1] = 0;
2481
2482 colon = strchr(ifr.ifr_name, ':');
2483 if (colon)
2484 *colon = 0;
2485
2486 /*
2487 * See which interface the caller is talking about.
2488 */
2489
2490 switch (cmd) {
2491 /*
2492 * These ioctl calls:
2493 * - can be done by all.
2494 * - atomic and do not require locking.
2495 * - return a value
2496 */
2497 case SIOCGIFFLAGS:
2498 case SIOCGIFMETRIC:
2499 case SIOCGIFMTU:
2500 case SIOCGIFHWADDR:
2501 case SIOCGIFSLAVE:
2502 case SIOCGIFMAP:
2503 case SIOCGIFINDEX:
2504 case SIOCGIFTXQLEN:
2505 dev_load(ifr.ifr_name);
2506 read_lock(&dev_base_lock);
2507 ret = dev_ifsioc(&ifr, cmd);
2508 read_unlock(&dev_base_lock);
2509 if (!ret) {
2510 if (colon)
2511 *colon = ':';
2512 if (copy_to_user(arg, &ifr,
2513 sizeof(struct ifreq)))
2514 ret = -EFAULT;
2515 }
2516 return ret;
2517
2518 case SIOCETHTOOL:
2519 dev_load(ifr.ifr_name);
2520 rtnl_lock();
2521 ret = dev_ethtool(&ifr);
2522 rtnl_unlock();
2523 if (!ret) {
2524 if (colon)
2525 *colon = ':';
2526 if (copy_to_user(arg, &ifr,
2527 sizeof(struct ifreq)))
2528 ret = -EFAULT;
2529 }
2530 return ret;
2531
2532 /*
2533 * These ioctl calls:
2534 * - require superuser power.
2535 * - require strict serialization.
2536 * - return a value
2537 */
2538 case SIOCGMIIPHY:
2539 case SIOCGMIIREG:
2540 case SIOCSIFNAME:
2541 if (!capable(CAP_NET_ADMIN))
2542 return -EPERM;
2543 dev_load(ifr.ifr_name);
2544 rtnl_lock();
2545 ret = dev_ifsioc(&ifr, cmd);
2546 rtnl_unlock();
2547 if (!ret) {
2548 if (colon)
2549 *colon = ':';
2550 if (copy_to_user(arg, &ifr,
2551 sizeof(struct ifreq)))
2552 ret = -EFAULT;
2553 }
2554 return ret;
2555
2556 /*
2557 * These ioctl calls:
2558 * - require superuser power.
2559 * - require strict serialization.
2560 * - do not return a value
2561 */
2562 case SIOCSIFFLAGS:
2563 case SIOCSIFMETRIC:
2564 case SIOCSIFMTU:
2565 case SIOCSIFMAP:
2566 case SIOCSIFHWADDR:
2567 case SIOCSIFSLAVE:
2568 case SIOCADDMULTI:
2569 case SIOCDELMULTI:
2570 case SIOCSIFHWBROADCAST:
2571 case SIOCSIFTXQLEN:
2572 case SIOCSMIIREG:
2573 case SIOCBONDENSLAVE:
2574 case SIOCBONDRELEASE:
2575 case SIOCBONDSETHWADDR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 case SIOCBONDCHANGEACTIVE:
2577 case SIOCBRADDIF:
2578 case SIOCBRDELIF:
2579 if (!capable(CAP_NET_ADMIN))
2580 return -EPERM;
Thomas Grafcabcac02006-01-24 12:46:33 -08002581 /* fall through */
2582 case SIOCBONDSLAVEINFOQUERY:
2583 case SIOCBONDINFOQUERY:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 dev_load(ifr.ifr_name);
2585 rtnl_lock();
2586 ret = dev_ifsioc(&ifr, cmd);
2587 rtnl_unlock();
2588 return ret;
2589
2590 case SIOCGIFMEM:
2591 /* Get the per device memory space. We can add this but
2592 * currently do not support it */
2593 case SIOCSIFMEM:
2594 /* Set the per device memory buffer space.
2595 * Not applicable in our case */
2596 case SIOCSIFLINK:
2597 return -EINVAL;
2598
2599 /*
2600 * Unknown or private ioctl.
2601 */
2602 default:
2603 if (cmd == SIOCWANDEV ||
2604 (cmd >= SIOCDEVPRIVATE &&
2605 cmd <= SIOCDEVPRIVATE + 15)) {
2606 dev_load(ifr.ifr_name);
2607 rtnl_lock();
2608 ret = dev_ifsioc(&ifr, cmd);
2609 rtnl_unlock();
2610 if (!ret && copy_to_user(arg, &ifr,
2611 sizeof(struct ifreq)))
2612 ret = -EFAULT;
2613 return ret;
2614 }
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002615#ifdef CONFIG_WIRELESS_EXT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 /* Take care of Wireless Extensions */
2617 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2618 /* If command is `set a parameter', or
2619 * `get the encoding parameters', check if
2620 * the user has the right to do it */
2621 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2622 if (!capable(CAP_NET_ADMIN))
2623 return -EPERM;
2624 }
2625 dev_load(ifr.ifr_name);
2626 rtnl_lock();
2627 /* Follow me in net/core/wireless.c */
2628 ret = wireless_process_ioctl(&ifr, cmd);
2629 rtnl_unlock();
2630 if (IW_IS_GET(cmd) &&
2631 copy_to_user(arg, &ifr,
2632 sizeof(struct ifreq)))
2633 ret = -EFAULT;
2634 return ret;
2635 }
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002636#endif /* CONFIG_WIRELESS_EXT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 return -EINVAL;
2638 }
2639}
2640
2641
2642/**
2643 * dev_new_index - allocate an ifindex
2644 *
2645 * Returns a suitable unique value for a new device interface
2646 * number. The caller must hold the rtnl semaphore or the
2647 * dev_base_lock to be sure it remains unique.
2648 */
2649static int dev_new_index(void)
2650{
2651 static int ifindex;
2652 for (;;) {
2653 if (++ifindex <= 0)
2654 ifindex = 1;
2655 if (!__dev_get_by_index(ifindex))
2656 return ifindex;
2657 }
2658}
2659
2660static int dev_boot_phase = 1;
2661
2662/* Delayed registration/unregisteration */
2663static DEFINE_SPINLOCK(net_todo_list_lock);
2664static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2665
2666static inline void net_set_todo(struct net_device *dev)
2667{
2668 spin_lock(&net_todo_list_lock);
2669 list_add_tail(&dev->todo_list, &net_todo_list);
2670 spin_unlock(&net_todo_list_lock);
2671}
2672
2673/**
2674 * register_netdevice - register a network device
2675 * @dev: device to register
2676 *
2677 * Take a completed network device structure and add it to the kernel
2678 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2679 * chain. 0 is returned on success. A negative errno code is returned
2680 * on a failure to set up the device, or if the name is a duplicate.
2681 *
2682 * Callers must hold the rtnl semaphore. You may want
2683 * register_netdev() instead of this.
2684 *
2685 * BUGS:
2686 * The locking appears insufficient to guarantee two parallel registers
2687 * will not get the same name.
2688 */
2689
2690int register_netdevice(struct net_device *dev)
2691{
2692 struct hlist_head *head;
2693 struct hlist_node *p;
2694 int ret;
2695
2696 BUG_ON(dev_boot_phase);
2697 ASSERT_RTNL();
2698
2699 /* When net_device's are persistent, this will be fatal. */
2700 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2701
2702 spin_lock_init(&dev->queue_lock);
2703 spin_lock_init(&dev->xmit_lock);
2704 dev->xmit_lock_owner = -1;
2705#ifdef CONFIG_NET_CLS_ACT
2706 spin_lock_init(&dev->ingress_lock);
2707#endif
2708
2709 ret = alloc_divert_blk(dev);
2710 if (ret)
2711 goto out;
2712
2713 dev->iflink = -1;
2714
2715 /* Init, if this function is available */
2716 if (dev->init) {
2717 ret = dev->init(dev);
2718 if (ret) {
2719 if (ret > 0)
2720 ret = -EIO;
2721 goto out_err;
2722 }
2723 }
2724
2725 if (!dev_valid_name(dev->name)) {
2726 ret = -EINVAL;
2727 goto out_err;
2728 }
2729
2730 dev->ifindex = dev_new_index();
2731 if (dev->iflink == -1)
2732 dev->iflink = dev->ifindex;
2733
2734 /* Check for existence of name */
2735 head = dev_name_hash(dev->name);
2736 hlist_for_each(p, head) {
2737 struct net_device *d
2738 = hlist_entry(p, struct net_device, name_hlist);
2739 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2740 ret = -EEXIST;
2741 goto out_err;
2742 }
2743 }
2744
2745 /* Fix illegal SG+CSUM combinations. */
2746 if ((dev->features & NETIF_F_SG) &&
2747 !(dev->features & (NETIF_F_IP_CSUM |
2748 NETIF_F_NO_CSUM |
2749 NETIF_F_HW_CSUM))) {
2750 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2751 dev->name);
2752 dev->features &= ~NETIF_F_SG;
2753 }
2754
2755 /* TSO requires that SG is present as well. */
2756 if ((dev->features & NETIF_F_TSO) &&
2757 !(dev->features & NETIF_F_SG)) {
2758 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2759 dev->name);
2760 dev->features &= ~NETIF_F_TSO;
2761 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002762 if (dev->features & NETIF_F_UFO) {
2763 if (!(dev->features & NETIF_F_HW_CSUM)) {
2764 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2765 "NETIF_F_HW_CSUM feature.\n",
2766 dev->name);
2767 dev->features &= ~NETIF_F_UFO;
2768 }
2769 if (!(dev->features & NETIF_F_SG)) {
2770 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2771 "NETIF_F_SG feature.\n",
2772 dev->name);
2773 dev->features &= ~NETIF_F_UFO;
2774 }
2775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
2777 /*
2778 * nil rebuild_header routine,
2779 * that should be never called and used as just bug trap.
2780 */
2781
2782 if (!dev->rebuild_header)
2783 dev->rebuild_header = default_rebuild_header;
2784
2785 /*
2786 * Default initial state at registry is that the
2787 * device is present.
2788 */
2789
2790 set_bit(__LINK_STATE_PRESENT, &dev->state);
2791
2792 dev->next = NULL;
2793 dev_init_scheduler(dev);
2794 write_lock_bh(&dev_base_lock);
2795 *dev_tail = dev;
2796 dev_tail = &dev->next;
2797 hlist_add_head(&dev->name_hlist, head);
2798 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2799 dev_hold(dev);
2800 dev->reg_state = NETREG_REGISTERING;
2801 write_unlock_bh(&dev_base_lock);
2802
2803 /* Notify protocols, that a new device appeared. */
2804 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2805
2806 /* Finish registration after unlock */
2807 net_set_todo(dev);
2808 ret = 0;
2809
2810out:
2811 return ret;
2812out_err:
2813 free_divert_blk(dev);
2814 goto out;
2815}
2816
2817/**
2818 * register_netdev - register a network device
2819 * @dev: device to register
2820 *
2821 * Take a completed network device structure and add it to the kernel
2822 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2823 * chain. 0 is returned on success. A negative errno code is returned
2824 * on a failure to set up the device, or if the name is a duplicate.
2825 *
2826 * This is a wrapper around register_netdev that takes the rtnl semaphore
2827 * and expands the device name if you passed a format string to
2828 * alloc_netdev.
2829 */
2830int register_netdev(struct net_device *dev)
2831{
2832 int err;
2833
2834 rtnl_lock();
2835
2836 /*
2837 * If the name is a format string the caller wants us to do a
2838 * name allocation.
2839 */
2840 if (strchr(dev->name, '%')) {
2841 err = dev_alloc_name(dev, dev->name);
2842 if (err < 0)
2843 goto out;
2844 }
2845
2846 /*
2847 * Back compatibility hook. Kill this one in 2.5
2848 */
2849 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2850 err = dev_alloc_name(dev, "eth%d");
2851 if (err < 0)
2852 goto out;
2853 }
2854
2855 err = register_netdevice(dev);
2856out:
2857 rtnl_unlock();
2858 return err;
2859}
2860EXPORT_SYMBOL(register_netdev);
2861
2862/*
2863 * netdev_wait_allrefs - wait until all references are gone.
2864 *
2865 * This is called when unregistering network devices.
2866 *
2867 * Any protocol or device that holds a reference should register
2868 * for netdevice notification, and cleanup and put back the
2869 * reference if they receive an UNREGISTER event.
2870 * We can get stuck here if buggy protocols don't correctly
2871 * call dev_put.
2872 */
2873static void netdev_wait_allrefs(struct net_device *dev)
2874{
2875 unsigned long rebroadcast_time, warning_time;
2876
2877 rebroadcast_time = warning_time = jiffies;
2878 while (atomic_read(&dev->refcnt) != 0) {
2879 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002880 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881
2882 /* Rebroadcast unregister notification */
2883 notifier_call_chain(&netdev_chain,
2884 NETDEV_UNREGISTER, dev);
2885
2886 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2887 &dev->state)) {
2888 /* We must not have linkwatch events
2889 * pending on unregister. If this
2890 * happens, we simply run the queue
2891 * unscheduled, resulting in a noop
2892 * for this device.
2893 */
2894 linkwatch_run_queue();
2895 }
2896
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002897 __rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898
2899 rebroadcast_time = jiffies;
2900 }
2901
2902 msleep(250);
2903
2904 if (time_after(jiffies, warning_time + 10 * HZ)) {
2905 printk(KERN_EMERG "unregister_netdevice: "
2906 "waiting for %s to become free. Usage "
2907 "count = %d\n",
2908 dev->name, atomic_read(&dev->refcnt));
2909 warning_time = jiffies;
2910 }
2911 }
2912}
2913
2914/* The sequence is:
2915 *
2916 * rtnl_lock();
2917 * ...
2918 * register_netdevice(x1);
2919 * register_netdevice(x2);
2920 * ...
2921 * unregister_netdevice(y1);
2922 * unregister_netdevice(y2);
2923 * ...
2924 * rtnl_unlock();
2925 * free_netdev(y1);
2926 * free_netdev(y2);
2927 *
2928 * We are invoked by rtnl_unlock() after it drops the semaphore.
2929 * This allows us to deal with problems:
2930 * 1) We can create/delete sysfs objects which invoke hotplug
2931 * without deadlocking with linkwatch via keventd.
2932 * 2) Since we run with the RTNL semaphore not held, we can sleep
2933 * safely in order to wait for the netdev refcnt to drop to zero.
2934 */
2935static DECLARE_MUTEX(net_todo_run_mutex);
2936void netdev_run_todo(void)
2937{
2938 struct list_head list = LIST_HEAD_INIT(list);
2939 int err;
2940
2941
2942 /* Need to guard against multiple cpu's getting out of order. */
2943 down(&net_todo_run_mutex);
2944
2945 /* Not safe to do outside the semaphore. We must not return
2946 * until all unregister events invoked by the local processor
2947 * have been completed (either by this todo run, or one on
2948 * another cpu).
2949 */
2950 if (list_empty(&net_todo_list))
2951 goto out;
2952
2953 /* Snapshot list, allow later requests */
2954 spin_lock(&net_todo_list_lock);
2955 list_splice_init(&net_todo_list, &list);
2956 spin_unlock(&net_todo_list_lock);
2957
2958 while (!list_empty(&list)) {
2959 struct net_device *dev
2960 = list_entry(list.next, struct net_device, todo_list);
2961 list_del(&dev->todo_list);
2962
2963 switch(dev->reg_state) {
2964 case NETREG_REGISTERING:
2965 err = netdev_register_sysfs(dev);
2966 if (err)
2967 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2968 dev->name, err);
2969 dev->reg_state = NETREG_REGISTERED;
2970 break;
2971
2972 case NETREG_UNREGISTERING:
2973 netdev_unregister_sysfs(dev);
2974 dev->reg_state = NETREG_UNREGISTERED;
2975
2976 netdev_wait_allrefs(dev);
2977
2978 /* paranoia */
2979 BUG_ON(atomic_read(&dev->refcnt));
2980 BUG_TRAP(!dev->ip_ptr);
2981 BUG_TRAP(!dev->ip6_ptr);
2982 BUG_TRAP(!dev->dn_ptr);
2983
2984
2985 /* It must be the very last action,
2986 * after this 'dev' may point to freed up memory.
2987 */
2988 if (dev->destructor)
2989 dev->destructor(dev);
2990 break;
2991
2992 default:
2993 printk(KERN_ERR "network todo '%s' but state %d\n",
2994 dev->name, dev->reg_state);
2995 break;
2996 }
2997 }
2998
2999out:
3000 up(&net_todo_run_mutex);
3001}
3002
3003/**
3004 * alloc_netdev - allocate network device
3005 * @sizeof_priv: size of private data to allocate space for
3006 * @name: device name format string
3007 * @setup: callback to initialize device
3008 *
3009 * Allocates a struct net_device with private data area for driver use
3010 * and performs basic initialization.
3011 */
3012struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3013 void (*setup)(struct net_device *))
3014{
3015 void *p;
3016 struct net_device *dev;
3017 int alloc_size;
3018
3019 /* ensure 32-byte alignment of both the device and private area */
3020 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3021 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3022
3023 p = kmalloc(alloc_size, GFP_KERNEL);
3024 if (!p) {
3025 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3026 return NULL;
3027 }
3028 memset(p, 0, alloc_size);
3029
3030 dev = (struct net_device *)
3031 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3032 dev->padded = (char *)dev - (char *)p;
3033
3034 if (sizeof_priv)
3035 dev->priv = netdev_priv(dev);
3036
3037 setup(dev);
3038 strcpy(dev->name, name);
3039 return dev;
3040}
3041EXPORT_SYMBOL(alloc_netdev);
3042
3043/**
3044 * free_netdev - free network device
3045 * @dev: device
3046 *
3047 * This function does the last stage of destroying an allocated device
3048 * interface. The reference to the device object is released.
3049 * If this is the last reference then it will be freed.
3050 */
3051void free_netdev(struct net_device *dev)
3052{
3053#ifdef CONFIG_SYSFS
3054 /* Compatiablity with error handling in drivers */
3055 if (dev->reg_state == NETREG_UNINITIALIZED) {
3056 kfree((char *)dev - dev->padded);
3057 return;
3058 }
3059
3060 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3061 dev->reg_state = NETREG_RELEASED;
3062
3063 /* will free via class release */
3064 class_device_put(&dev->class_dev);
3065#else
3066 kfree((char *)dev - dev->padded);
3067#endif
3068}
3069
3070/* Synchronize with packet receive processing. */
3071void synchronize_net(void)
3072{
3073 might_sleep();
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07003074 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075}
3076
3077/**
3078 * unregister_netdevice - remove device from the kernel
3079 * @dev: device
3080 *
3081 * This function shuts down a device interface and removes it
3082 * from the kernel tables. On success 0 is returned, on a failure
3083 * a negative errno code is returned.
3084 *
3085 * Callers must hold the rtnl semaphore. You may want
3086 * unregister_netdev() instead of this.
3087 */
3088
3089int unregister_netdevice(struct net_device *dev)
3090{
3091 struct net_device *d, **dp;
3092
3093 BUG_ON(dev_boot_phase);
3094 ASSERT_RTNL();
3095
3096 /* Some devices call without registering for initialization unwind. */
3097 if (dev->reg_state == NETREG_UNINITIALIZED) {
3098 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3099 "was registered\n", dev->name, dev);
3100 return -ENODEV;
3101 }
3102
3103 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3104
3105 /* If device is running, close it first. */
3106 if (dev->flags & IFF_UP)
3107 dev_close(dev);
3108
3109 /* And unlink it from device chain. */
3110 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3111 if (d == dev) {
3112 write_lock_bh(&dev_base_lock);
3113 hlist_del(&dev->name_hlist);
3114 hlist_del(&dev->index_hlist);
3115 if (dev_tail == &dev->next)
3116 dev_tail = dp;
3117 *dp = d->next;
3118 write_unlock_bh(&dev_base_lock);
3119 break;
3120 }
3121 }
3122 if (!d) {
3123 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3124 dev->name);
3125 return -ENODEV;
3126 }
3127
3128 dev->reg_state = NETREG_UNREGISTERING;
3129
3130 synchronize_net();
3131
3132 /* Shutdown queueing discipline. */
3133 dev_shutdown(dev);
3134
3135
3136 /* Notify protocols, that we are about to destroy
3137 this device. They should clean all the things.
3138 */
3139 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3140
3141 /*
3142 * Flush the multicast chain
3143 */
3144 dev_mc_discard(dev);
3145
3146 if (dev->uninit)
3147 dev->uninit(dev);
3148
3149 /* Notifier chain MUST detach us from master device. */
3150 BUG_TRAP(!dev->master);
3151
3152 free_divert_blk(dev);
3153
3154 /* Finish processing unregister after unlock */
3155 net_set_todo(dev);
3156
3157 synchronize_net();
3158
3159 dev_put(dev);
3160 return 0;
3161}
3162
3163/**
3164 * unregister_netdev - remove device from the kernel
3165 * @dev: device
3166 *
3167 * This function shuts down a device interface and removes it
3168 * from the kernel tables. On success 0 is returned, on a failure
3169 * a negative errno code is returned.
3170 *
3171 * This is just a wrapper for unregister_netdevice that takes
3172 * the rtnl semaphore. In general you want to use this and not
3173 * unregister_netdevice.
3174 */
3175void unregister_netdev(struct net_device *dev)
3176{
3177 rtnl_lock();
3178 unregister_netdevice(dev);
3179 rtnl_unlock();
3180}
3181
3182EXPORT_SYMBOL(unregister_netdev);
3183
3184#ifdef CONFIG_HOTPLUG_CPU
3185static int dev_cpu_callback(struct notifier_block *nfb,
3186 unsigned long action,
3187 void *ocpu)
3188{
3189 struct sk_buff **list_skb;
3190 struct net_device **list_net;
3191 struct sk_buff *skb;
3192 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3193 struct softnet_data *sd, *oldsd;
3194
3195 if (action != CPU_DEAD)
3196 return NOTIFY_OK;
3197
3198 local_irq_disable();
3199 cpu = smp_processor_id();
3200 sd = &per_cpu(softnet_data, cpu);
3201 oldsd = &per_cpu(softnet_data, oldcpu);
3202
3203 /* Find end of our completion_queue. */
3204 list_skb = &sd->completion_queue;
3205 while (*list_skb)
3206 list_skb = &(*list_skb)->next;
3207 /* Append completion queue from offline CPU. */
3208 *list_skb = oldsd->completion_queue;
3209 oldsd->completion_queue = NULL;
3210
3211 /* Find end of our output_queue. */
3212 list_net = &sd->output_queue;
3213 while (*list_net)
3214 list_net = &(*list_net)->next_sched;
3215 /* Append output queue from offline CPU. */
3216 *list_net = oldsd->output_queue;
3217 oldsd->output_queue = NULL;
3218
3219 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3220 local_irq_enable();
3221
3222 /* Process offline CPU's input_pkt_queue */
3223 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3224 netif_rx(skb);
3225
3226 return NOTIFY_OK;
3227}
3228#endif /* CONFIG_HOTPLUG_CPU */
3229
3230
3231/*
3232 * Initialize the DEV module. At boot time this walks the device list and
3233 * unhooks any devices that fail to initialise (normally hardware not
3234 * present) and leaves us with a valid list of present and active devices.
3235 *
3236 */
3237
3238/*
3239 * This is called single threaded during boot, so no need
3240 * to take the rtnl semaphore.
3241 */
3242static int __init net_dev_init(void)
3243{
3244 int i, rc = -ENOMEM;
3245
3246 BUG_ON(!dev_boot_phase);
3247
3248 net_random_init();
3249
3250 if (dev_proc_init())
3251 goto out;
3252
3253 if (netdev_sysfs_init())
3254 goto out;
3255
3256 INIT_LIST_HEAD(&ptype_all);
3257 for (i = 0; i < 16; i++)
3258 INIT_LIST_HEAD(&ptype_base[i]);
3259
3260 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3261 INIT_HLIST_HEAD(&dev_name_head[i]);
3262
3263 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3264 INIT_HLIST_HEAD(&dev_index_head[i]);
3265
3266 /*
3267 * Initialise the packet receive queues.
3268 */
3269
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08003270 for_each_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 struct softnet_data *queue;
3272
3273 queue = &per_cpu(softnet_data, i);
3274 skb_queue_head_init(&queue->input_pkt_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 queue->completion_queue = NULL;
3276 INIT_LIST_HEAD(&queue->poll_list);
3277 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3278 queue->backlog_dev.weight = weight_p;
3279 queue->backlog_dev.poll = process_backlog;
3280 atomic_set(&queue->backlog_dev.refcnt, 1);
3281 }
3282
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 dev_boot_phase = 0;
3284
3285 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3286 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3287
3288 hotcpu_notifier(dev_cpu_callback, 0);
3289 dst_init();
3290 dev_mcast_init();
3291 rc = 0;
3292out:
3293 return rc;
3294}
3295
3296subsys_initcall(net_dev_init);
3297
3298EXPORT_SYMBOL(__dev_get_by_index);
3299EXPORT_SYMBOL(__dev_get_by_name);
3300EXPORT_SYMBOL(__dev_remove_pack);
3301EXPORT_SYMBOL(__skb_linearize);
Mitch Williamsc2373ee2005-11-09 10:34:45 -08003302EXPORT_SYMBOL(dev_valid_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303EXPORT_SYMBOL(dev_add_pack);
3304EXPORT_SYMBOL(dev_alloc_name);
3305EXPORT_SYMBOL(dev_close);
3306EXPORT_SYMBOL(dev_get_by_flags);
3307EXPORT_SYMBOL(dev_get_by_index);
3308EXPORT_SYMBOL(dev_get_by_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309EXPORT_SYMBOL(dev_open);
3310EXPORT_SYMBOL(dev_queue_xmit);
3311EXPORT_SYMBOL(dev_remove_pack);
3312EXPORT_SYMBOL(dev_set_allmulti);
3313EXPORT_SYMBOL(dev_set_promiscuity);
3314EXPORT_SYMBOL(dev_change_flags);
3315EXPORT_SYMBOL(dev_set_mtu);
3316EXPORT_SYMBOL(dev_set_mac_address);
3317EXPORT_SYMBOL(free_netdev);
3318EXPORT_SYMBOL(netdev_boot_setup_check);
3319EXPORT_SYMBOL(netdev_set_master);
3320EXPORT_SYMBOL(netdev_state_change);
3321EXPORT_SYMBOL(netif_receive_skb);
3322EXPORT_SYMBOL(netif_rx);
3323EXPORT_SYMBOL(register_gifconf);
3324EXPORT_SYMBOL(register_netdevice);
3325EXPORT_SYMBOL(register_netdevice_notifier);
3326EXPORT_SYMBOL(skb_checksum_help);
3327EXPORT_SYMBOL(synchronize_net);
3328EXPORT_SYMBOL(unregister_netdevice);
3329EXPORT_SYMBOL(unregister_netdevice_notifier);
3330EXPORT_SYMBOL(net_enable_timestamp);
3331EXPORT_SYMBOL(net_disable_timestamp);
3332EXPORT_SYMBOL(dev_get_flags);
3333
3334#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3335EXPORT_SYMBOL(br_handle_frame_hook);
3336EXPORT_SYMBOL(br_fdb_get_hook);
3337EXPORT_SYMBOL(br_fdb_put_hook);
3338#endif
3339
3340#ifdef CONFIG_KMOD
3341EXPORT_SYMBOL(dev_load);
3342#endif
3343
3344EXPORT_PER_CPU_SYMBOL(softnet_data);