blob: 9fe96cde3e1956e57dcfca598d74b3bb63530333 [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/cpu.h>
80#include <linux/types.h>
81#include <linux/kernel.h>
82#include <linux/sched.h>
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -080083#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070084#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>
Steve Grubb5bdb9882005-12-03 08:39:35 -0500116#include <linux/audit.h>
Chris Leechdb217332006-06-17 21:24:58 -0700117#include <linux/dmaengine.h>
Herbert Xuf6a78bf2006-06-22 02:57:17 -0700118#include <linux/err.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120/*
121 * The list of packet types we will receive (as opposed to discard)
122 * and the routines to invoke.
123 *
124 * Why 16. Because with 16 the only overlap we get on a hash of the
125 * low nibble of the protocol value is RARP/SNAP/X.25.
126 *
127 * NOTE: That is no longer true with the addition of VLAN tags. Not
128 * sure which should go first, but I bet it won't make much
129 * difference if we are running VLANs. The good news is that
130 * this protocol won't be in the list unless compiled in, so
Stephen Hemminger3041a062006-05-26 13:25:24 -0700131 * the average user (w/out VLANs) will not be adversely affected.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 * --BLG
133 *
134 * 0800 IP
135 * 8100 802.1Q VLAN
136 * 0001 802.3
137 * 0002 AX.25
138 * 0004 802.2
139 * 8035 RARP
140 * 0005 SNAP
141 * 0805 X.25
142 * 0806 ARP
143 * 8137 IPX
144 * 0009 Localtalk
145 * 86DD IPv6
146 */
147
148static DEFINE_SPINLOCK(ptype_lock);
149static struct list_head ptype_base[16]; /* 16 way hashed list */
150static struct list_head ptype_all; /* Taps */
151
Chris Leechdb217332006-06-17 21:24:58 -0700152#ifdef CONFIG_NET_DMA
153static struct dma_client *net_dma_client;
154static unsigned int net_dma_count;
155static spinlock_t net_dma_event_lock;
156#endif
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158/*
Stephen Hemminger3041a062006-05-26 13:25:24 -0700159 * The @dev_base list is protected by @dev_base_lock and the rtnl
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 * semaphore.
161 *
162 * Pure readers hold dev_base_lock for reading.
163 *
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
168 *
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
172 *
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
175 * semaphore held.
176 */
177struct net_device *dev_base;
178static struct net_device **dev_tail = &dev_base;
179DEFINE_RWLOCK(dev_base_lock);
180
181EXPORT_SYMBOL(dev_base);
182EXPORT_SYMBOL(dev_base_lock);
183
184#define NETDEV_HASHBITS 8
185static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
186static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
187
188static inline struct hlist_head *dev_name_hash(const char *name)
189{
190 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
192}
193
194static inline struct hlist_head *dev_index_hash(int ifindex)
195{
196 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
197}
198
199/*
200 * Our notifier list
201 */
202
Alan Sternf07d5b92006-05-09 15:23:03 -0700203static RAW_NOTIFIER_HEAD(netdev_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
205/*
206 * Device drivers call our routines to queue packets here. We empty the
207 * queue in the local softnet handler.
208 */
Stephen Hemminger31aa02c2005-06-23 20:12:48 -0700209DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211#ifdef CONFIG_SYSFS
212extern int netdev_sysfs_init(void);
213extern int netdev_register_sysfs(struct net_device *);
214extern void netdev_unregister_sysfs(struct net_device *);
215#else
216#define netdev_sysfs_init() (0)
217#define netdev_register_sysfs(dev) (0)
218#define netdev_unregister_sysfs(dev) do { } while(0)
219#endif
220
221
222/*******************************************************************************
223
224 Protocol management and registration routines
225
226*******************************************************************************/
227
228/*
229 * For efficiency
230 */
231
Adrian Bunk60481262006-06-25 23:58:10 -0700232static int netdev_nit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
234/*
235 * Add a protocol ID to the list. Now that the input handler is
236 * smarter we can dispense with all the messy stuff that used to be
237 * here.
238 *
239 * BEWARE!!! Protocol handlers, mangling input packets,
240 * MUST BE last in hash buckets and checking protocol handlers
241 * MUST start from promiscuous ptype_all chain in net_bh.
242 * It is true now, do not change it.
243 * Explanation follows: if protocol handler, mangling packet, will
244 * be the first on list, it is not able to sense, that packet
245 * is cloned and should be copied-on-write, so that it will
246 * change it and subsequent readers will get broken packet.
247 * --ANK (980803)
248 */
249
250/**
251 * dev_add_pack - add packet handler
252 * @pt: packet type declaration
253 *
254 * Add a protocol handler to the networking stack. The passed &packet_type
255 * is linked into kernel lists and may not be freed until it has been
256 * removed from the kernel lists.
257 *
258 * This call does not sleep therefore it can not
259 * guarantee all CPU's that are in middle of receiving packets
260 * will see the new packet type (until the next received packet).
261 */
262
263void dev_add_pack(struct packet_type *pt)
264{
265 int hash;
266
267 spin_lock_bh(&ptype_lock);
268 if (pt->type == htons(ETH_P_ALL)) {
269 netdev_nit++;
270 list_add_rcu(&pt->list, &ptype_all);
271 } else {
272 hash = ntohs(pt->type) & 15;
273 list_add_rcu(&pt->list, &ptype_base[hash]);
274 }
275 spin_unlock_bh(&ptype_lock);
276}
277
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278/**
279 * __dev_remove_pack - remove packet handler
280 * @pt: packet type declaration
281 *
282 * Remove a protocol handler that was previously added to the kernel
283 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
284 * from the kernel lists and can be freed or reused once this function
285 * returns.
286 *
287 * The packet type might still be in use by receivers
288 * and must not be freed until after all the CPU's have gone
289 * through a quiescent state.
290 */
291void __dev_remove_pack(struct packet_type *pt)
292{
293 struct list_head *head;
294 struct packet_type *pt1;
295
296 spin_lock_bh(&ptype_lock);
297
298 if (pt->type == htons(ETH_P_ALL)) {
299 netdev_nit--;
300 head = &ptype_all;
301 } else
302 head = &ptype_base[ntohs(pt->type) & 15];
303
304 list_for_each_entry(pt1, head, list) {
305 if (pt == pt1) {
306 list_del_rcu(&pt->list);
307 goto out;
308 }
309 }
310
311 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
312out:
313 spin_unlock_bh(&ptype_lock);
314}
315/**
316 * dev_remove_pack - remove packet handler
317 * @pt: packet type declaration
318 *
319 * Remove a protocol handler that was previously added to the kernel
320 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
321 * from the kernel lists and can be freed or reused once this function
322 * returns.
323 *
324 * This call sleeps to guarantee that no CPU is looking at the packet
325 * type after return.
326 */
327void dev_remove_pack(struct packet_type *pt)
328{
329 __dev_remove_pack(pt);
330
331 synchronize_net();
332}
333
334/******************************************************************************
335
336 Device Boot-time Settings Routines
337
338*******************************************************************************/
339
340/* Boot time configuration table */
341static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
342
343/**
344 * netdev_boot_setup_add - add new setup entry
345 * @name: name of the device
346 * @map: configured settings for the device
347 *
348 * Adds new setup entry to the dev_boot_setup list. The function
349 * returns 0 on error and 1 on success. This is a generic routine to
350 * all netdevices.
351 */
352static int netdev_boot_setup_add(char *name, struct ifmap *map)
353{
354 struct netdev_boot_setup *s;
355 int i;
356
357 s = dev_boot_setup;
358 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
359 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
360 memset(s[i].name, 0, sizeof(s[i].name));
361 strcpy(s[i].name, name);
362 memcpy(&s[i].map, map, sizeof(s[i].map));
363 break;
364 }
365 }
366
367 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
368}
369
370/**
371 * netdev_boot_setup_check - check boot time settings
372 * @dev: the netdevice
373 *
374 * Check boot time settings for the device.
375 * The found settings are set for the device to be used
376 * later in the device probing.
377 * Returns 0 if no settings found, 1 if they are.
378 */
379int netdev_boot_setup_check(struct net_device *dev)
380{
381 struct netdev_boot_setup *s = dev_boot_setup;
382 int i;
383
384 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
385 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
386 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
387 dev->irq = s[i].map.irq;
388 dev->base_addr = s[i].map.base_addr;
389 dev->mem_start = s[i].map.mem_start;
390 dev->mem_end = s[i].map.mem_end;
391 return 1;
392 }
393 }
394 return 0;
395}
396
397
398/**
399 * netdev_boot_base - get address from boot time settings
400 * @prefix: prefix for network device
401 * @unit: id for network device
402 *
403 * Check boot time settings for the base address of device.
404 * The found settings are set for the device to be used
405 * later in the device probing.
406 * Returns 0 if no settings found.
407 */
408unsigned long netdev_boot_base(const char *prefix, int unit)
409{
410 const struct netdev_boot_setup *s = dev_boot_setup;
411 char name[IFNAMSIZ];
412 int i;
413
414 sprintf(name, "%s%d", prefix, unit);
415
416 /*
417 * If device already registered then return base of 1
418 * to indicate not to probe for this interface
419 */
420 if (__dev_get_by_name(name))
421 return 1;
422
423 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
424 if (!strcmp(name, s[i].name))
425 return s[i].map.base_addr;
426 return 0;
427}
428
429/*
430 * Saves at boot time configured settings for any netdevice.
431 */
432int __init netdev_boot_setup(char *str)
433{
434 int ints[5];
435 struct ifmap map;
436
437 str = get_options(str, ARRAY_SIZE(ints), ints);
438 if (!str || !*str)
439 return 0;
440
441 /* Save settings */
442 memset(&map, 0, sizeof(map));
443 if (ints[0] > 0)
444 map.irq = ints[1];
445 if (ints[0] > 1)
446 map.base_addr = ints[2];
447 if (ints[0] > 2)
448 map.mem_start = ints[3];
449 if (ints[0] > 3)
450 map.mem_end = ints[4];
451
452 /* Add new entry to the list */
453 return netdev_boot_setup_add(str, &map);
454}
455
456__setup("netdev=", netdev_boot_setup);
457
458/*******************************************************************************
459
460 Device Interface Subroutines
461
462*******************************************************************************/
463
464/**
465 * __dev_get_by_name - find a device by its name
466 * @name: name to find
467 *
468 * Find an interface by name. Must be called under RTNL semaphore
469 * or @dev_base_lock. If the name is found a pointer to the device
470 * is returned. If the name is not found then %NULL is returned. The
471 * reference counters are not incremented so the caller must be
472 * careful with locks.
473 */
474
475struct net_device *__dev_get_by_name(const char *name)
476{
477 struct hlist_node *p;
478
479 hlist_for_each(p, dev_name_hash(name)) {
480 struct net_device *dev
481 = hlist_entry(p, struct net_device, name_hlist);
482 if (!strncmp(dev->name, name, IFNAMSIZ))
483 return dev;
484 }
485 return NULL;
486}
487
488/**
489 * dev_get_by_name - find a device by its name
490 * @name: name to find
491 *
492 * Find an interface by name. This can be called from any
493 * context and does its own locking. The returned handle has
494 * the usage count incremented and the caller must use dev_put() to
495 * release it when it is no longer needed. %NULL is returned if no
496 * matching device is found.
497 */
498
499struct net_device *dev_get_by_name(const char *name)
500{
501 struct net_device *dev;
502
503 read_lock(&dev_base_lock);
504 dev = __dev_get_by_name(name);
505 if (dev)
506 dev_hold(dev);
507 read_unlock(&dev_base_lock);
508 return dev;
509}
510
511/**
512 * __dev_get_by_index - find a device by its ifindex
513 * @ifindex: index of device
514 *
515 * Search for an interface by index. Returns %NULL if the device
516 * is not found or a pointer to the device. The device has not
517 * had its reference counter increased so the caller must be careful
518 * about locking. The caller must hold either the RTNL semaphore
519 * or @dev_base_lock.
520 */
521
522struct net_device *__dev_get_by_index(int ifindex)
523{
524 struct hlist_node *p;
525
526 hlist_for_each(p, dev_index_hash(ifindex)) {
527 struct net_device *dev
528 = hlist_entry(p, struct net_device, index_hlist);
529 if (dev->ifindex == ifindex)
530 return dev;
531 }
532 return NULL;
533}
534
535
536/**
537 * dev_get_by_index - find a device by its ifindex
538 * @ifindex: index of device
539 *
540 * Search for an interface by index. Returns NULL if the device
541 * is not found or a pointer to the device. The device returned has
542 * had a reference added and the pointer is safe until the user calls
543 * dev_put to indicate they have finished with it.
544 */
545
546struct net_device *dev_get_by_index(int ifindex)
547{
548 struct net_device *dev;
549
550 read_lock(&dev_base_lock);
551 dev = __dev_get_by_index(ifindex);
552 if (dev)
553 dev_hold(dev);
554 read_unlock(&dev_base_lock);
555 return dev;
556}
557
558/**
559 * dev_getbyhwaddr - find a device by its hardware address
560 * @type: media type of device
561 * @ha: hardware address
562 *
563 * Search for an interface by MAC address. Returns NULL if the device
564 * is not found or a pointer to the device. The caller must hold the
565 * rtnl semaphore. The returned device has not had its ref count increased
566 * and the caller must therefore be careful about locking
567 *
568 * BUGS:
569 * If the API was consistent this would be __dev_get_by_hwaddr
570 */
571
572struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
573{
574 struct net_device *dev;
575
576 ASSERT_RTNL();
577
578 for (dev = dev_base; dev; dev = dev->next)
579 if (dev->type == type &&
580 !memcmp(dev->dev_addr, ha, dev->addr_len))
581 break;
582 return dev;
583}
584
Jochen Friedrichcf309e32005-09-22 04:44:55 -0300585EXPORT_SYMBOL(dev_getbyhwaddr);
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587struct net_device *dev_getfirstbyhwtype(unsigned short type)
588{
589 struct net_device *dev;
590
591 rtnl_lock();
592 for (dev = dev_base; dev; dev = dev->next) {
593 if (dev->type == type) {
594 dev_hold(dev);
595 break;
596 }
597 }
598 rtnl_unlock();
599 return dev;
600}
601
602EXPORT_SYMBOL(dev_getfirstbyhwtype);
603
604/**
605 * dev_get_by_flags - find any device with given flags
606 * @if_flags: IFF_* values
607 * @mask: bitmask of bits in if_flags to check
608 *
609 * Search for any interface with the given flags. Returns NULL if a device
610 * is not found or a pointer to the device. The device returned has
611 * had a reference added and the pointer is safe until the user calls
612 * dev_put to indicate they have finished with it.
613 */
614
615struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
616{
617 struct net_device *dev;
618
619 read_lock(&dev_base_lock);
620 for (dev = dev_base; dev != NULL; dev = dev->next) {
621 if (((dev->flags ^ if_flags) & mask) == 0) {
622 dev_hold(dev);
623 break;
624 }
625 }
626 read_unlock(&dev_base_lock);
627 return dev;
628}
629
630/**
631 * dev_valid_name - check if name is okay for network device
632 * @name: name string
633 *
634 * Network device names need to be valid file names to
635 * to allow sysfs to work
636 */
Mitch Williamsc2373ee2005-11-09 10:34:45 -0800637int dev_valid_name(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638{
639 return !(*name == '\0'
640 || !strcmp(name, ".")
641 || !strcmp(name, "..")
642 || strchr(name, '/'));
643}
644
645/**
646 * dev_alloc_name - allocate a name for a device
647 * @dev: device
648 * @name: name format string
649 *
650 * Passed a format string - eg "lt%d" it will try and find a suitable
Stephen Hemminger3041a062006-05-26 13:25:24 -0700651 * id. It scans list of devices to build up a free map, then chooses
652 * the first empty slot. The caller must hold the dev_base or rtnl lock
653 * while allocating the name and adding the device in order to avoid
654 * duplicates.
655 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
656 * Returns the number of the unit assigned or a negative errno code.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 */
658
659int dev_alloc_name(struct net_device *dev, const char *name)
660{
661 int i = 0;
662 char buf[IFNAMSIZ];
663 const char *p;
664 const int max_netdevices = 8*PAGE_SIZE;
665 long *inuse;
666 struct net_device *d;
667
668 p = strnchr(name, IFNAMSIZ-1, '%');
669 if (p) {
670 /*
671 * Verify the string as this thing may have come from
672 * the user. There must be either one "%d" and no other "%"
673 * characters.
674 */
675 if (p[1] != 'd' || strchr(p + 2, '%'))
676 return -EINVAL;
677
678 /* Use one page as a bit array of possible slots */
679 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
680 if (!inuse)
681 return -ENOMEM;
682
683 for (d = dev_base; d; d = d->next) {
684 if (!sscanf(d->name, name, &i))
685 continue;
686 if (i < 0 || i >= max_netdevices)
687 continue;
688
689 /* avoid cases where sscanf is not exact inverse of printf */
690 snprintf(buf, sizeof(buf), name, i);
691 if (!strncmp(buf, d->name, IFNAMSIZ))
692 set_bit(i, inuse);
693 }
694
695 i = find_first_zero_bit(inuse, max_netdevices);
696 free_page((unsigned long) inuse);
697 }
698
699 snprintf(buf, sizeof(buf), name, i);
700 if (!__dev_get_by_name(buf)) {
701 strlcpy(dev->name, buf, IFNAMSIZ);
702 return i;
703 }
704
705 /* It is possible to run out of possible slots
706 * when the name is long and there isn't enough space left
707 * for the digits, or if all bits are used.
708 */
709 return -ENFILE;
710}
711
712
713/**
714 * dev_change_name - change name of a device
715 * @dev: device
716 * @newname: name (or format string) must be at least IFNAMSIZ
717 *
718 * Change name of a device, can pass format strings "eth%d".
719 * for wildcarding.
720 */
721int dev_change_name(struct net_device *dev, char *newname)
722{
723 int err = 0;
724
725 ASSERT_RTNL();
726
727 if (dev->flags & IFF_UP)
728 return -EBUSY;
729
730 if (!dev_valid_name(newname))
731 return -EINVAL;
732
733 if (strchr(newname, '%')) {
734 err = dev_alloc_name(dev, newname);
735 if (err < 0)
736 return err;
737 strcpy(newname, dev->name);
738 }
739 else if (__dev_get_by_name(newname))
740 return -EEXIST;
741 else
742 strlcpy(dev->name, newname, IFNAMSIZ);
743
744 err = class_device_rename(&dev->class_dev, dev->name);
745 if (!err) {
746 hlist_del(&dev->name_hlist);
747 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
Alan Sternf07d5b92006-05-09 15:23:03 -0700748 raw_notifier_call_chain(&netdev_chain,
Alan Sterne041c682006-03-27 01:16:30 -0800749 NETDEV_CHANGENAME, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
751
752 return err;
753}
754
755/**
Stephen Hemminger3041a062006-05-26 13:25:24 -0700756 * netdev_features_change - device changes features
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700757 * @dev: device to cause notification
758 *
759 * Called to indicate a device has changed features.
760 */
761void netdev_features_change(struct net_device *dev)
762{
Alan Sternf07d5b92006-05-09 15:23:03 -0700763 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700764}
765EXPORT_SYMBOL(netdev_features_change);
766
767/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 * netdev_state_change - device changes state
769 * @dev: device to cause notification
770 *
771 * Called to indicate a device has changed state. This function calls
772 * the notifier chains for netdev_chain and sends a NEWLINK message
773 * to the routing socket.
774 */
775void netdev_state_change(struct net_device *dev)
776{
777 if (dev->flags & IFF_UP) {
Alan Sternf07d5b92006-05-09 15:23:03 -0700778 raw_notifier_call_chain(&netdev_chain,
Alan Sterne041c682006-03-27 01:16:30 -0800779 NETDEV_CHANGE, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
781 }
782}
783
784/**
785 * dev_load - load a network module
786 * @name: name of interface
787 *
788 * If a network interface is not present and the process has suitable
789 * privileges this function loads the module. If module loading is not
790 * available in this kernel then it becomes a nop.
791 */
792
793void dev_load(const char *name)
794{
795 struct net_device *dev;
796
797 read_lock(&dev_base_lock);
798 dev = __dev_get_by_name(name);
799 read_unlock(&dev_base_lock);
800
801 if (!dev && capable(CAP_SYS_MODULE))
802 request_module("%s", name);
803}
804
805static int default_rebuild_header(struct sk_buff *skb)
806{
807 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
808 skb->dev ? skb->dev->name : "NULL!!!");
809 kfree_skb(skb);
810 return 1;
811}
812
813
814/**
815 * dev_open - prepare an interface for use.
816 * @dev: device to open
817 *
818 * Takes a device from down to up state. The device's private open
819 * function is invoked and then the multicast lists are loaded. Finally
820 * the device is moved into the up state and a %NETDEV_UP message is
821 * sent to the netdev notifier chain.
822 *
823 * Calling this function on an active interface is a nop. On a failure
824 * a negative errno code is returned.
825 */
826int dev_open(struct net_device *dev)
827{
828 int ret = 0;
829
830 /*
831 * Is it already up?
832 */
833
834 if (dev->flags & IFF_UP)
835 return 0;
836
837 /*
838 * Is it even present?
839 */
840 if (!netif_device_present(dev))
841 return -ENODEV;
842
843 /*
844 * Call device private open method
845 */
846 set_bit(__LINK_STATE_START, &dev->state);
847 if (dev->open) {
848 ret = dev->open(dev);
849 if (ret)
850 clear_bit(__LINK_STATE_START, &dev->state);
851 }
852
853 /*
854 * If it went open OK then:
855 */
856
857 if (!ret) {
858 /*
859 * Set the flags.
860 */
861 dev->flags |= IFF_UP;
862
863 /*
864 * Initialize multicasting status
865 */
866 dev_mc_upload(dev);
867
868 /*
869 * Wakeup transmit queue engine
870 */
871 dev_activate(dev);
872
873 /*
874 * ... and announce new interface.
875 */
Alan Sternf07d5b92006-05-09 15:23:03 -0700876 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 }
878 return ret;
879}
880
881/**
882 * dev_close - shutdown an interface.
883 * @dev: device to shutdown
884 *
885 * This function moves an active device into down state. A
886 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
887 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
888 * chain.
889 */
890int dev_close(struct net_device *dev)
891{
892 if (!(dev->flags & IFF_UP))
893 return 0;
894
895 /*
896 * Tell people we are going down, so that they can
897 * prepare to death, when device is still operating.
898 */
Alan Sternf07d5b92006-05-09 15:23:03 -0700899 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901 dev_deactivate(dev);
902
903 clear_bit(__LINK_STATE_START, &dev->state);
904
905 /* Synchronize to scheduled poll. We cannot touch poll list,
906 * it can be even on different cpu. So just clear netif_running(),
907 * and wait when poll really will happen. Actually, the best place
908 * for this is inside dev->stop() after device stopped its irq
909 * engine, but this requires more changes in devices. */
910
911 smp_mb__after_clear_bit(); /* Commit netif_running(). */
912 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
913 /* No hurry. */
David S. Miller6192b542005-07-28 12:12:58 -0700914 msleep(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 }
916
917 /*
918 * Call the device specific close. This cannot fail.
919 * Only if device is UP
920 *
921 * We allow it to be called even after a DETACH hot-plug
922 * event.
923 */
924 if (dev->stop)
925 dev->stop(dev);
926
927 /*
928 * Device is now down.
929 */
930
931 dev->flags &= ~IFF_UP;
932
933 /*
934 * Tell people we are down
935 */
Alan Sternf07d5b92006-05-09 15:23:03 -0700936 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937
938 return 0;
939}
940
941
942/*
943 * Device change register/unregister. These are not inline or static
944 * as we export them to the world.
945 */
946
947/**
948 * register_netdevice_notifier - register a network notifier block
949 * @nb: notifier
950 *
951 * Register a notifier to be called when network device events occur.
952 * The notifier passed is linked into the kernel structures and must
953 * not be reused until it has been unregistered. A negative errno code
954 * is returned on a failure.
955 *
956 * When registered all registration and up events are replayed
957 * to the new notifier to allow device to have a race free
958 * view of the network device list.
959 */
960
961int register_netdevice_notifier(struct notifier_block *nb)
962{
963 struct net_device *dev;
964 int err;
965
966 rtnl_lock();
Alan Sternf07d5b92006-05-09 15:23:03 -0700967 err = raw_notifier_chain_register(&netdev_chain, nb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 if (!err) {
969 for (dev = dev_base; dev; dev = dev->next) {
970 nb->notifier_call(nb, NETDEV_REGISTER, dev);
971
972 if (dev->flags & IFF_UP)
973 nb->notifier_call(nb, NETDEV_UP, dev);
974 }
975 }
976 rtnl_unlock();
977 return err;
978}
979
980/**
981 * unregister_netdevice_notifier - unregister a network notifier block
982 * @nb: notifier
983 *
984 * Unregister a notifier previously registered by
985 * register_netdevice_notifier(). The notifier is unlinked into the
986 * kernel structures and may then be reused. A negative errno code
987 * is returned on a failure.
988 */
989
990int unregister_netdevice_notifier(struct notifier_block *nb)
991{
Herbert Xu9f514952006-03-25 01:24:25 -0800992 int err;
993
994 rtnl_lock();
Alan Sternf07d5b92006-05-09 15:23:03 -0700995 err = raw_notifier_chain_unregister(&netdev_chain, nb);
Herbert Xu9f514952006-03-25 01:24:25 -0800996 rtnl_unlock();
997 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998}
999
1000/**
1001 * call_netdevice_notifiers - call all network notifier blocks
1002 * @val: value passed unmodified to notifier function
1003 * @v: pointer passed unmodified to notifier function
1004 *
1005 * Call all network notifier blocks. Parameters and return value
Alan Sternf07d5b92006-05-09 15:23:03 -07001006 * are as for raw_notifier_call_chain().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 */
1008
1009int call_netdevice_notifiers(unsigned long val, void *v)
1010{
Alan Sternf07d5b92006-05-09 15:23:03 -07001011 return raw_notifier_call_chain(&netdev_chain, val, v);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012}
1013
1014/* When > 0 there are consumers of rx skb time stamps */
1015static atomic_t netstamp_needed = ATOMIC_INIT(0);
1016
1017void net_enable_timestamp(void)
1018{
1019 atomic_inc(&netstamp_needed);
1020}
1021
1022void net_disable_timestamp(void)
1023{
1024 atomic_dec(&netstamp_needed);
1025}
1026
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001027void __net_timestamp(struct sk_buff *skb)
1028{
1029 struct timeval tv;
1030
1031 do_gettimeofday(&tv);
1032 skb_set_timestamp(skb, &tv);
1033}
1034EXPORT_SYMBOL(__net_timestamp);
1035
1036static inline void net_timestamp(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037{
1038 if (atomic_read(&netstamp_needed))
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001039 __net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 else {
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001041 skb->tstamp.off_sec = 0;
1042 skb->tstamp.off_usec = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 }
1044}
1045
1046/*
1047 * Support routine. Sends outgoing frames to any network
1048 * taps currently in use.
1049 */
1050
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001051static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052{
1053 struct packet_type *ptype;
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001054
1055 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
1057 rcu_read_lock();
1058 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1059 /* Never send packets back to the socket
1060 * they originated from - MvS (miquels@drinkel.ow.org)
1061 */
1062 if ((ptype->dev == dev || !ptype->dev) &&
1063 (ptype->af_packet_priv == NULL ||
1064 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1065 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1066 if (!skb2)
1067 break;
1068
1069 /* skb->nh should be correctly
1070 set by sender, so that the second statement is
1071 just protection against buggy protocols.
1072 */
1073 skb2->mac.raw = skb2->data;
1074
1075 if (skb2->nh.raw < skb2->data ||
1076 skb2->nh.raw > skb2->tail) {
1077 if (net_ratelimit())
1078 printk(KERN_CRIT "protocol %04x is "
1079 "buggy, dev %s\n",
1080 skb2->protocol, dev->name);
1081 skb2->nh.raw = skb2->data;
1082 }
1083
1084 skb2->h.raw = skb2->nh.raw;
1085 skb2->pkt_type = PACKET_OUTGOING;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001086 ptype->func(skb2, skb->dev, ptype, skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 }
1088 }
1089 rcu_read_unlock();
1090}
1091
Denis Vlasenko56079432006-03-29 15:57:29 -08001092
1093void __netif_schedule(struct net_device *dev)
1094{
1095 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1096 unsigned long flags;
1097 struct softnet_data *sd;
1098
1099 local_irq_save(flags);
1100 sd = &__get_cpu_var(softnet_data);
1101 dev->next_sched = sd->output_queue;
1102 sd->output_queue = dev;
1103 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1104 local_irq_restore(flags);
1105 }
1106}
1107EXPORT_SYMBOL(__netif_schedule);
1108
1109void __netif_rx_schedule(struct net_device *dev)
1110{
1111 unsigned long flags;
1112
1113 local_irq_save(flags);
1114 dev_hold(dev);
1115 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1116 if (dev->quota < 0)
1117 dev->quota += dev->weight;
1118 else
1119 dev->quota = dev->weight;
1120 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1121 local_irq_restore(flags);
1122}
1123EXPORT_SYMBOL(__netif_rx_schedule);
1124
1125void dev_kfree_skb_any(struct sk_buff *skb)
1126{
1127 if (in_irq() || irqs_disabled())
1128 dev_kfree_skb_irq(skb);
1129 else
1130 dev_kfree_skb(skb);
1131}
1132EXPORT_SYMBOL(dev_kfree_skb_any);
1133
1134
1135/* Hot-plugging. */
1136void netif_device_detach(struct net_device *dev)
1137{
1138 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1139 netif_running(dev)) {
1140 netif_stop_queue(dev);
1141 }
1142}
1143EXPORT_SYMBOL(netif_device_detach);
1144
1145void netif_device_attach(struct net_device *dev)
1146{
1147 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1148 netif_running(dev)) {
1149 netif_wake_queue(dev);
1150 __netdev_watchdog_up(dev);
1151 }
1152}
1153EXPORT_SYMBOL(netif_device_attach);
1154
1155
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156/*
1157 * Invalidate hardware checksum when packet is to be mangled, and
1158 * complete checksum manually on outgoing path.
1159 */
1160int skb_checksum_help(struct sk_buff *skb, int inward)
1161{
1162 unsigned int csum;
1163 int ret = 0, offset = skb->h.raw - skb->data;
1164
Herbert Xua430a432006-07-08 13:34:56 -07001165 if (inward)
1166 goto out_set_summed;
1167
1168 if (unlikely(skb_shinfo(skb)->gso_size)) {
Herbert Xua430a432006-07-08 13:34:56 -07001169 /* Let GSO fix up the checksum. */
1170 goto out_set_summed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 }
1172
1173 if (skb_cloned(skb)) {
1174 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1175 if (ret)
1176 goto out;
1177 }
1178
Kris Katterjohn09a62662006-01-08 22:24:28 -08001179 BUG_ON(offset > (int)skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1181
1182 offset = skb->tail - skb->h.raw;
Kris Katterjohn09a62662006-01-08 22:24:28 -08001183 BUG_ON(offset <= 0);
1184 BUG_ON(skb->csum + 2 > offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
1186 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
Herbert Xua430a432006-07-08 13:34:56 -07001187
1188out_set_summed:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 skb->ip_summed = CHECKSUM_NONE;
1190out:
1191 return ret;
1192}
1193
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001194/**
1195 * skb_gso_segment - Perform segmentation on skb.
1196 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07001197 * @features: features for the output path (see dev->features)
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001198 *
1199 * This function segments the given skb and returns a list of segments.
Herbert Xu576a30e2006-06-27 13:22:38 -07001200 *
1201 * It may return NULL if the skb requires no segmentation. This is
1202 * only possible when GSO is used for verifying header integrity.
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001203 */
Herbert Xu576a30e2006-06-27 13:22:38 -07001204struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001205{
1206 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1207 struct packet_type *ptype;
1208 int type = skb->protocol;
Herbert Xua430a432006-07-08 13:34:56 -07001209 int err;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001210
1211 BUG_ON(skb_shinfo(skb)->frag_list);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001212
1213 skb->mac.raw = skb->data;
1214 skb->mac_len = skb->nh.raw - skb->data;
1215 __skb_pull(skb, skb->mac_len);
1216
Herbert Xua430a432006-07-08 13:34:56 -07001217 if (unlikely(skb->ip_summed != CHECKSUM_HW)) {
Herbert Xua430a432006-07-08 13:34:56 -07001218 if (skb_header_cloned(skb) &&
1219 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1220 return ERR_PTR(err);
1221 }
1222
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001223 rcu_read_lock();
1224 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1225 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
Herbert Xua430a432006-07-08 13:34:56 -07001226 if (unlikely(skb->ip_summed != CHECKSUM_HW)) {
1227 err = ptype->gso_send_check(skb);
1228 segs = ERR_PTR(err);
1229 if (err || skb_gso_ok(skb, features))
1230 break;
1231 __skb_push(skb, skb->data - skb->nh.raw);
1232 }
Herbert Xu576a30e2006-06-27 13:22:38 -07001233 segs = ptype->gso_segment(skb, features);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001234 break;
1235 }
1236 }
1237 rcu_read_unlock();
1238
Herbert Xu576a30e2006-06-27 13:22:38 -07001239 __skb_push(skb, skb->data - skb->mac.raw);
1240
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001241 return segs;
1242}
1243
1244EXPORT_SYMBOL(skb_gso_segment);
1245
Herbert Xufb286bb2005-11-10 13:01:24 -08001246/* Take action when hardware reception checksum errors are detected. */
1247#ifdef CONFIG_BUG
1248void netdev_rx_csum_fault(struct net_device *dev)
1249{
1250 if (net_ratelimit()) {
Stephen Hemminger246a4212005-12-08 15:21:39 -08001251 printk(KERN_ERR "%s: hw csum failure.\n",
1252 dev ? dev->name : "<unknown>");
Herbert Xufb286bb2005-11-10 13:01:24 -08001253 dump_stack();
1254 }
1255}
1256EXPORT_SYMBOL(netdev_rx_csum_fault);
1257#endif
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259/* Actually, we should eliminate this check as soon as we know, that:
1260 * 1. IOMMU is present and allows to map all the memory.
1261 * 2. No high memory really exists on this machine.
1262 */
1263
1264static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1265{
Herbert Xu3d3a8532006-06-27 13:33:10 -07001266#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 int i;
1268
1269 if (dev->features & NETIF_F_HIGHDMA)
1270 return 0;
1271
1272 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1273 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1274 return 1;
1275
Herbert Xu3d3a8532006-06-27 13:33:10 -07001276#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 return 0;
1278}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001280struct dev_gso_cb {
1281 void (*destructor)(struct sk_buff *skb);
1282};
1283
1284#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1285
1286static void dev_gso_skb_destructor(struct sk_buff *skb)
1287{
1288 struct dev_gso_cb *cb;
1289
1290 do {
1291 struct sk_buff *nskb = skb->next;
1292
1293 skb->next = nskb->next;
1294 nskb->next = NULL;
1295 kfree_skb(nskb);
1296 } while (skb->next);
1297
1298 cb = DEV_GSO_CB(skb);
1299 if (cb->destructor)
1300 cb->destructor(skb);
1301}
1302
1303/**
1304 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1305 * @skb: buffer to segment
1306 *
1307 * This function segments the given skb and stores the list of segments
1308 * in skb->next.
1309 */
1310static int dev_gso_segment(struct sk_buff *skb)
1311{
1312 struct net_device *dev = skb->dev;
1313 struct sk_buff *segs;
Herbert Xu576a30e2006-06-27 13:22:38 -07001314 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1315 NETIF_F_SG : 0);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001316
Herbert Xu576a30e2006-06-27 13:22:38 -07001317 segs = skb_gso_segment(skb, features);
1318
1319 /* Verifying header integrity only. */
1320 if (!segs)
1321 return 0;
1322
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001323 if (unlikely(IS_ERR(segs)))
1324 return PTR_ERR(segs);
1325
1326 skb->next = segs;
1327 DEV_GSO_CB(skb)->destructor = skb->destructor;
1328 skb->destructor = dev_gso_skb_destructor;
1329
1330 return 0;
1331}
1332
1333int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1334{
1335 if (likely(!skb->next)) {
1336 if (netdev_nit)
1337 dev_queue_xmit_nit(skb, dev);
1338
Herbert Xu576a30e2006-06-27 13:22:38 -07001339 if (netif_needs_gso(dev, skb)) {
1340 if (unlikely(dev_gso_segment(skb)))
1341 goto out_kfree_skb;
1342 if (skb->next)
1343 goto gso;
1344 }
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001345
Herbert Xu576a30e2006-06-27 13:22:38 -07001346 return dev->hard_start_xmit(skb, dev);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001347 }
1348
Herbert Xu576a30e2006-06-27 13:22:38 -07001349gso:
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001350 do {
1351 struct sk_buff *nskb = skb->next;
1352 int rc;
1353
1354 skb->next = nskb->next;
1355 nskb->next = NULL;
1356 rc = dev->hard_start_xmit(nskb, dev);
1357 if (unlikely(rc)) {
Michael Chanf54d9e82006-06-25 23:57:04 -07001358 nskb->next = skb->next;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001359 skb->next = nskb;
1360 return rc;
1361 }
Michael Chanf54d9e82006-06-25 23:57:04 -07001362 if (unlikely(netif_queue_stopped(dev) && skb->next))
1363 return NETDEV_TX_BUSY;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001364 } while (skb->next);
1365
1366 skb->destructor = DEV_GSO_CB(skb)->destructor;
1367
1368out_kfree_skb:
1369 kfree_skb(skb);
1370 return 0;
1371}
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373#define HARD_TX_LOCK(dev, cpu) { \
1374 if ((dev->features & NETIF_F_LLTX) == 0) { \
Herbert Xu932ff272006-06-09 12:20:56 -07001375 netif_tx_lock(dev); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 } \
1377}
1378
1379#define HARD_TX_UNLOCK(dev) { \
1380 if ((dev->features & NETIF_F_LLTX) == 0) { \
Herbert Xu932ff272006-06-09 12:20:56 -07001381 netif_tx_unlock(dev); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 } \
1383}
1384
1385/**
1386 * dev_queue_xmit - transmit a buffer
1387 * @skb: buffer to transmit
1388 *
1389 * Queue a buffer for transmission to a network device. The caller must
1390 * have set the device and priority and built the buffer before calling
1391 * this function. The function can be called from an interrupt.
1392 *
1393 * A negative errno code is returned on a failure. A success does not
1394 * guarantee the frame will be transmitted as it may be dropped due
1395 * to congestion or traffic shaping.
Ben Greearaf191362005-04-24 20:12:36 -07001396 *
1397 * -----------------------------------------------------------------------------------
1398 * I notice this method can also return errors from the queue disciplines,
1399 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1400 * be positive.
1401 *
1402 * Regardless of the return value, the skb is consumed, so it is currently
1403 * difficult to retry a send to this method. (You can bump the ref count
1404 * before sending to hold a reference for retry if you are careful.)
1405 *
1406 * When calling this method, interrupts MUST be enabled. This is because
1407 * the BH enable code must have IRQs enabled so that it will not deadlock.
1408 * --BLG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 */
1410
1411int dev_queue_xmit(struct sk_buff *skb)
1412{
1413 struct net_device *dev = skb->dev;
1414 struct Qdisc *q;
1415 int rc = -ENOMEM;
1416
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001417 /* GSO will handle the following emulations directly. */
1418 if (netif_needs_gso(dev, skb))
1419 goto gso;
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 if (skb_shinfo(skb)->frag_list &&
1422 !(dev->features & NETIF_F_FRAGLIST) &&
Herbert Xu364c6ba2006-06-09 16:10:40 -07001423 __skb_linearize(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 goto out_kfree_skb;
1425
1426 /* Fragmented skb is linearized if device does not support SG,
1427 * or if at least one of fragments is in highmem and device
1428 * does not support DMA from it.
1429 */
1430 if (skb_shinfo(skb)->nr_frags &&
1431 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
Herbert Xu364c6ba2006-06-09 16:10:40 -07001432 __skb_linearize(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 goto out_kfree_skb;
1434
1435 /* If packet is not checksummed and device does not support
1436 * checksumming for this protocol, complete checksumming here.
1437 */
1438 if (skb->ip_summed == CHECKSUM_HW &&
Herbert Xu8648b302006-06-17 22:06:05 -07001439 (!(dev->features & NETIF_F_GEN_CSUM) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 (!(dev->features & NETIF_F_IP_CSUM) ||
1441 skb->protocol != htons(ETH_P_IP))))
1442 if (skb_checksum_help(skb, 0))
1443 goto out_kfree_skb;
1444
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001445gso:
Eric Dumazet2d7ceec2005-09-27 15:22:58 -07001446 spin_lock_prefetch(&dev->queue_lock);
1447
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 /* Disable soft irqs for various locks below. Also
1449 * stops preemption for RCU.
1450 */
Herbert Xud4828d82006-06-22 02:28:18 -07001451 rcu_read_lock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453 /* Updates of qdisc are serialized by queue_lock.
1454 * The struct Qdisc which is pointed to by qdisc is now a
1455 * rcu structure - it may be accessed without acquiring
1456 * a lock (but the structure may be stale.) The freeing of the
1457 * qdisc will be deferred until it's known that there are no
1458 * more references to it.
1459 *
1460 * If the qdisc has an enqueue function, we still need to
1461 * hold the queue_lock before calling it, since queue_lock
1462 * also serializes access to the device queue.
1463 */
1464
1465 q = rcu_dereference(dev->qdisc);
1466#ifdef CONFIG_NET_CLS_ACT
1467 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1468#endif
1469 if (q->enqueue) {
1470 /* Grab device queue */
1471 spin_lock(&dev->queue_lock);
1472
1473 rc = q->enqueue(skb, q);
1474
1475 qdisc_run(dev);
1476
1477 spin_unlock(&dev->queue_lock);
1478 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1479 goto out;
1480 }
1481
1482 /* The device has no queue. Common case for software devices:
1483 loopback, all the sorts of tunnels...
1484
Herbert Xu932ff272006-06-09 12:20:56 -07001485 Really, it is unlikely that netif_tx_lock protection is necessary
1486 here. (f.e. loopback and IP tunnels are clean ignoring statistics
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 counters.)
1488 However, it is possible, that they rely on protection
1489 made by us here.
1490
1491 Check this and shot the lock. It is not prone from deadlocks.
1492 Either shot noqueue qdisc, it is even simpler 8)
1493 */
1494 if (dev->flags & IFF_UP) {
1495 int cpu = smp_processor_id(); /* ok because BHs are off */
1496
1497 if (dev->xmit_lock_owner != cpu) {
1498
1499 HARD_TX_LOCK(dev, cpu);
1500
1501 if (!netif_queue_stopped(dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 rc = 0;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001503 if (!dev_hard_start_xmit(skb, dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 HARD_TX_UNLOCK(dev);
1505 goto out;
1506 }
1507 }
1508 HARD_TX_UNLOCK(dev);
1509 if (net_ratelimit())
1510 printk(KERN_CRIT "Virtual device %s asks to "
1511 "queue packet!\n", dev->name);
1512 } else {
1513 /* Recursion is detected! It is possible,
1514 * unfortunately */
1515 if (net_ratelimit())
1516 printk(KERN_CRIT "Dead loop on virtual device "
1517 "%s, fix it urgently!\n", dev->name);
1518 }
1519 }
1520
1521 rc = -ENETDOWN;
Herbert Xud4828d82006-06-22 02:28:18 -07001522 rcu_read_unlock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
1524out_kfree_skb:
1525 kfree_skb(skb);
1526 return rc;
1527out:
Herbert Xud4828d82006-06-22 02:28:18 -07001528 rcu_read_unlock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 return rc;
1530}
1531
1532
1533/*=======================================================================
1534 Receiver routines
1535 =======================================================================*/
1536
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001537int netdev_max_backlog = 1000;
1538int netdev_budget = 300;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539int weight_p = 64; /* old backlog weight */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
1541DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1542
1543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544/**
1545 * netif_rx - post buffer to the network code
1546 * @skb: buffer to post
1547 *
1548 * This function receives a packet from a device driver and queues it for
1549 * the upper (protocol) levels to process. It always succeeds. The buffer
1550 * may be dropped during processing for congestion control or by the
1551 * protocol layers.
1552 *
1553 * return values:
1554 * NET_RX_SUCCESS (no congestion)
1555 * NET_RX_CN_LOW (low congestion)
1556 * NET_RX_CN_MOD (moderate congestion)
1557 * NET_RX_CN_HIGH (high congestion)
1558 * NET_RX_DROP (packet was dropped)
1559 *
1560 */
1561
1562int netif_rx(struct sk_buff *skb)
1563{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 struct softnet_data *queue;
1565 unsigned long flags;
1566
1567 /* if netpoll wants it, pretend we never saw it */
1568 if (netpoll_rx(skb))
1569 return NET_RX_DROP;
1570
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001571 if (!skb->tstamp.off_sec)
1572 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
1574 /*
1575 * The code is rearranged so that the path is the most
1576 * short when CPU is congested, but is still operating.
1577 */
1578 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 queue = &__get_cpu_var(softnet_data);
1580
1581 __get_cpu_var(netdev_rx_stat).total++;
1582 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1583 if (queue->input_pkt_queue.qlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584enqueue:
1585 dev_hold(skb->dev);
1586 __skb_queue_tail(&queue->input_pkt_queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 local_irq_restore(flags);
Stephen Hemminger34008d82005-06-23 20:10:00 -07001588 return NET_RX_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 }
1590
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 netif_rx_schedule(&queue->backlog_dev);
1592 goto enqueue;
1593 }
1594
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 __get_cpu_var(netdev_rx_stat).dropped++;
1596 local_irq_restore(flags);
1597
1598 kfree_skb(skb);
1599 return NET_RX_DROP;
1600}
1601
1602int netif_rx_ni(struct sk_buff *skb)
1603{
1604 int err;
1605
1606 preempt_disable();
1607 err = netif_rx(skb);
1608 if (local_softirq_pending())
1609 do_softirq();
1610 preempt_enable();
1611
1612 return err;
1613}
1614
1615EXPORT_SYMBOL(netif_rx_ni);
1616
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001617static inline struct net_device *skb_bond(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618{
1619 struct net_device *dev = skb->dev;
1620
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001621 if (dev->master) {
David S. Miller7ea49ed2006-08-14 17:08:36 -07001622 if (skb_bond_should_drop(skb)) {
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001623 kfree_skb(skb);
1624 return NULL;
1625 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 skb->dev = dev->master;
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001627 }
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001628
1629 return dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630}
1631
1632static void net_tx_action(struct softirq_action *h)
1633{
1634 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1635
1636 if (sd->completion_queue) {
1637 struct sk_buff *clist;
1638
1639 local_irq_disable();
1640 clist = sd->completion_queue;
1641 sd->completion_queue = NULL;
1642 local_irq_enable();
1643
1644 while (clist) {
1645 struct sk_buff *skb = clist;
1646 clist = clist->next;
1647
1648 BUG_TRAP(!atomic_read(&skb->users));
1649 __kfree_skb(skb);
1650 }
1651 }
1652
1653 if (sd->output_queue) {
1654 struct net_device *head;
1655
1656 local_irq_disable();
1657 head = sd->output_queue;
1658 sd->output_queue = NULL;
1659 local_irq_enable();
1660
1661 while (head) {
1662 struct net_device *dev = head;
1663 head = head->next_sched;
1664
1665 smp_mb__before_clear_bit();
1666 clear_bit(__LINK_STATE_SCHED, &dev->state);
1667
1668 if (spin_trylock(&dev->queue_lock)) {
1669 qdisc_run(dev);
1670 spin_unlock(&dev->queue_lock);
1671 } else {
1672 netif_schedule(dev);
1673 }
1674 }
1675 }
1676}
1677
1678static __inline__ int deliver_skb(struct sk_buff *skb,
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001679 struct packet_type *pt_prev,
1680 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681{
1682 atomic_inc(&skb->users);
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001683 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684}
1685
1686#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1687int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1688struct net_bridge;
1689struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1690 unsigned char *addr);
1691void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1692
1693static __inline__ int handle_bridge(struct sk_buff **pskb,
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001694 struct packet_type **pt_prev, int *ret,
1695 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696{
1697 struct net_bridge_port *port;
1698
1699 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1700 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1701 return 0;
1702
1703 if (*pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001704 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 *pt_prev = NULL;
1706 }
1707
1708 return br_handle_frame_hook(port, pskb);
1709}
1710#else
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001711#define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712#endif
1713
1714#ifdef CONFIG_NET_CLS_ACT
1715/* TODO: Maybe we should just force sch_ingress to be compiled in
1716 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1717 * a compare and 2 stores extra right now if we dont have it on
1718 * but have CONFIG_NET_CLS_ACT
1719 * NOTE: This doesnt stop any functionality; if you dont have
1720 * the ingress scheduler, you just cant add policies on ingress.
1721 *
1722 */
1723static int ing_filter(struct sk_buff *skb)
1724{
1725 struct Qdisc *q;
1726 struct net_device *dev = skb->dev;
1727 int result = TC_ACT_OK;
1728
1729 if (dev->qdisc_ingress) {
1730 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1731 if (MAX_RED_LOOP < ttl++) {
Stephen Hemminger5a8da022006-07-07 16:54:05 -07001732 printk(KERN_WARNING "Redir loop detected Dropping packet (%s->%s)\n",
David S. Miller86e65da2005-08-09 19:36:29 -07001733 skb->input_dev->name, skb->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 return TC_ACT_SHOT;
1735 }
1736
1737 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1738
1739 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
David S. Miller86e65da2005-08-09 19:36:29 -07001740
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 spin_lock(&dev->ingress_lock);
1742 if ((q = dev->qdisc_ingress) != NULL)
1743 result = q->enqueue(skb, q);
1744 spin_unlock(&dev->ingress_lock);
1745
1746 }
1747
1748 return result;
1749}
1750#endif
1751
1752int netif_receive_skb(struct sk_buff *skb)
1753{
1754 struct packet_type *ptype, *pt_prev;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001755 struct net_device *orig_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 int ret = NET_RX_DROP;
1757 unsigned short type;
1758
1759 /* if we've gotten here through NAPI, check netpoll */
1760 if (skb->dev->poll && netpoll_rx(skb))
1761 return NET_RX_DROP;
1762
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001763 if (!skb->tstamp.off_sec)
1764 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
David S. Miller86e65da2005-08-09 19:36:29 -07001766 if (!skb->input_dev)
1767 skb->input_dev = skb->dev;
1768
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001769 orig_dev = skb_bond(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001771 if (!orig_dev)
1772 return NET_RX_DROP;
1773
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 __get_cpu_var(netdev_rx_stat).total++;
1775
1776 skb->h.raw = skb->nh.raw = skb->data;
1777 skb->mac_len = skb->nh.raw - skb->mac.raw;
1778
1779 pt_prev = NULL;
1780
1781 rcu_read_lock();
1782
1783#ifdef CONFIG_NET_CLS_ACT
1784 if (skb->tc_verd & TC_NCLS) {
1785 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1786 goto ncls;
1787 }
1788#endif
1789
1790 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1791 if (!ptype->dev || ptype->dev == skb->dev) {
1792 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001793 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 pt_prev = ptype;
1795 }
1796 }
1797
1798#ifdef CONFIG_NET_CLS_ACT
1799 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001800 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 pt_prev = NULL; /* noone else should process this after*/
1802 } else {
1803 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1804 }
1805
1806 ret = ing_filter(skb);
1807
1808 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1809 kfree_skb(skb);
1810 goto out;
1811 }
1812
1813 skb->tc_verd = 0;
1814ncls:
1815#endif
1816
1817 handle_diverter(skb);
1818
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001819 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 goto out;
1821
1822 type = skb->protocol;
1823 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1824 if (ptype->type == type &&
1825 (!ptype->dev || ptype->dev == skb->dev)) {
1826 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001827 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 pt_prev = ptype;
1829 }
1830 }
1831
1832 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001833 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 } else {
1835 kfree_skb(skb);
1836 /* Jamal, now you will not able to escape explaining
1837 * me how you were going to use this. :-)
1838 */
1839 ret = NET_RX_DROP;
1840 }
1841
1842out:
1843 rcu_read_unlock();
1844 return ret;
1845}
1846
1847static int process_backlog(struct net_device *backlog_dev, int *budget)
1848{
1849 int work = 0;
1850 int quota = min(backlog_dev->quota, *budget);
1851 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1852 unsigned long start_time = jiffies;
1853
Stephen Hemmingere3876602005-06-08 14:56:01 -07001854 backlog_dev->weight = weight_p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 for (;;) {
1856 struct sk_buff *skb;
1857 struct net_device *dev;
1858
1859 local_irq_disable();
1860 skb = __skb_dequeue(&queue->input_pkt_queue);
1861 if (!skb)
1862 goto job_done;
1863 local_irq_enable();
1864
1865 dev = skb->dev;
1866
1867 netif_receive_skb(skb);
1868
1869 dev_put(dev);
1870
1871 work++;
1872
1873 if (work >= quota || jiffies - start_time > 1)
1874 break;
1875
1876 }
1877
1878 backlog_dev->quota -= work;
1879 *budget -= work;
1880 return -1;
1881
1882job_done:
1883 backlog_dev->quota -= work;
1884 *budget -= work;
1885
1886 list_del(&backlog_dev->poll_list);
1887 smp_mb__before_clear_bit();
1888 netif_poll_enable(backlog_dev);
1889
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 local_irq_enable();
1891 return 0;
1892}
1893
1894static void net_rx_action(struct softirq_action *h)
1895{
1896 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1897 unsigned long start_time = jiffies;
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001898 int budget = netdev_budget;
Matt Mackall53fb95d2005-08-11 19:27:43 -07001899 void *have;
1900
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 local_irq_disable();
1902
1903 while (!list_empty(&queue->poll_list)) {
1904 struct net_device *dev;
1905
1906 if (budget <= 0 || jiffies - start_time > 1)
1907 goto softnet_break;
1908
1909 local_irq_enable();
1910
1911 dev = list_entry(queue->poll_list.next,
1912 struct net_device, poll_list);
Matt Mackall53fb95d2005-08-11 19:27:43 -07001913 have = netpoll_poll_lock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
1915 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
Matt Mackall53fb95d2005-08-11 19:27:43 -07001916 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 local_irq_disable();
Stephen Hemminger8aca8a22006-03-20 22:26:39 -08001918 list_move_tail(&dev->poll_list, &queue->poll_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 if (dev->quota < 0)
1920 dev->quota += dev->weight;
1921 else
1922 dev->quota = dev->weight;
1923 } else {
Matt Mackall53fb95d2005-08-11 19:27:43 -07001924 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 dev_put(dev);
1926 local_irq_disable();
1927 }
1928 }
1929out:
Chris Leechdb217332006-06-17 21:24:58 -07001930#ifdef CONFIG_NET_DMA
1931 /*
1932 * There may not be any more sk_buffs coming right now, so push
1933 * any pending DMA copies to hardware
1934 */
1935 if (net_dma_client) {
1936 struct dma_chan *chan;
1937 rcu_read_lock();
1938 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1939 dma_async_memcpy_issue_pending(chan);
1940 rcu_read_unlock();
1941 }
1942#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 local_irq_enable();
1944 return;
1945
1946softnet_break:
1947 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1948 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1949 goto out;
1950}
1951
1952static gifconf_func_t * gifconf_list [NPROTO];
1953
1954/**
1955 * register_gifconf - register a SIOCGIF handler
1956 * @family: Address family
1957 * @gifconf: Function handler
1958 *
1959 * Register protocol dependent address dumping routines. The handler
1960 * that is passed must not be freed or reused until it has been replaced
1961 * by another handler.
1962 */
1963int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1964{
1965 if (family >= NPROTO)
1966 return -EINVAL;
1967 gifconf_list[family] = gifconf;
1968 return 0;
1969}
1970
1971
1972/*
1973 * Map an interface index to its name (SIOCGIFNAME)
1974 */
1975
1976/*
1977 * We need this ioctl for efficient implementation of the
1978 * if_indextoname() function required by the IPv6 API. Without
1979 * it, we would have to search all the interfaces to find a
1980 * match. --pb
1981 */
1982
1983static int dev_ifname(struct ifreq __user *arg)
1984{
1985 struct net_device *dev;
1986 struct ifreq ifr;
1987
1988 /*
1989 * Fetch the caller's info block.
1990 */
1991
1992 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1993 return -EFAULT;
1994
1995 read_lock(&dev_base_lock);
1996 dev = __dev_get_by_index(ifr.ifr_ifindex);
1997 if (!dev) {
1998 read_unlock(&dev_base_lock);
1999 return -ENODEV;
2000 }
2001
2002 strcpy(ifr.ifr_name, dev->name);
2003 read_unlock(&dev_base_lock);
2004
2005 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2006 return -EFAULT;
2007 return 0;
2008}
2009
2010/*
2011 * Perform a SIOCGIFCONF call. This structure will change
2012 * size eventually, and there is nothing I can do about it.
2013 * Thus we will need a 'compatibility mode'.
2014 */
2015
2016static int dev_ifconf(char __user *arg)
2017{
2018 struct ifconf ifc;
2019 struct net_device *dev;
2020 char __user *pos;
2021 int len;
2022 int total;
2023 int i;
2024
2025 /*
2026 * Fetch the caller's info block.
2027 */
2028
2029 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2030 return -EFAULT;
2031
2032 pos = ifc.ifc_buf;
2033 len = ifc.ifc_len;
2034
2035 /*
2036 * Loop over the interfaces, and write an info block for each.
2037 */
2038
2039 total = 0;
2040 for (dev = dev_base; dev; dev = dev->next) {
2041 for (i = 0; i < NPROTO; i++) {
2042 if (gifconf_list[i]) {
2043 int done;
2044 if (!pos)
2045 done = gifconf_list[i](dev, NULL, 0);
2046 else
2047 done = gifconf_list[i](dev, pos + total,
2048 len - total);
2049 if (done < 0)
2050 return -EFAULT;
2051 total += done;
2052 }
2053 }
2054 }
2055
2056 /*
2057 * All done. Write the updated control block back to the caller.
2058 */
2059 ifc.ifc_len = total;
2060
2061 /*
2062 * Both BSD and Solaris return 0 here, so we do too.
2063 */
2064 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2065}
2066
2067#ifdef CONFIG_PROC_FS
2068/*
2069 * This is invoked by the /proc filesystem handler to display a device
2070 * in detail.
2071 */
2072static __inline__ struct net_device *dev_get_idx(loff_t pos)
2073{
2074 struct net_device *dev;
2075 loff_t i;
2076
2077 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2078
2079 return i == pos ? dev : NULL;
2080}
2081
2082void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2083{
2084 read_lock(&dev_base_lock);
2085 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2086}
2087
2088void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2089{
2090 ++*pos;
2091 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2092}
2093
2094void dev_seq_stop(struct seq_file *seq, void *v)
2095{
2096 read_unlock(&dev_base_lock);
2097}
2098
2099static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2100{
2101 if (dev->get_stats) {
2102 struct net_device_stats *stats = dev->get_stats(dev);
2103
2104 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2105 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2106 dev->name, stats->rx_bytes, stats->rx_packets,
2107 stats->rx_errors,
2108 stats->rx_dropped + stats->rx_missed_errors,
2109 stats->rx_fifo_errors,
2110 stats->rx_length_errors + stats->rx_over_errors +
2111 stats->rx_crc_errors + stats->rx_frame_errors,
2112 stats->rx_compressed, stats->multicast,
2113 stats->tx_bytes, stats->tx_packets,
2114 stats->tx_errors, stats->tx_dropped,
2115 stats->tx_fifo_errors, stats->collisions,
2116 stats->tx_carrier_errors +
2117 stats->tx_aborted_errors +
2118 stats->tx_window_errors +
2119 stats->tx_heartbeat_errors,
2120 stats->tx_compressed);
2121 } else
2122 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2123}
2124
2125/*
2126 * Called from the PROCfs module. This now uses the new arbitrary sized
2127 * /proc/net interface to create /proc/net/dev
2128 */
2129static int dev_seq_show(struct seq_file *seq, void *v)
2130{
2131 if (v == SEQ_START_TOKEN)
2132 seq_puts(seq, "Inter-| Receive "
2133 " | Transmit\n"
2134 " face |bytes packets errs drop fifo frame "
2135 "compressed multicast|bytes packets errs "
2136 "drop fifo colls carrier compressed\n");
2137 else
2138 dev_seq_printf_stats(seq, v);
2139 return 0;
2140}
2141
2142static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2143{
2144 struct netif_rx_stats *rc = NULL;
2145
2146 while (*pos < NR_CPUS)
2147 if (cpu_online(*pos)) {
2148 rc = &per_cpu(netdev_rx_stat, *pos);
2149 break;
2150 } else
2151 ++*pos;
2152 return rc;
2153}
2154
2155static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2156{
2157 return softnet_get_online(pos);
2158}
2159
2160static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2161{
2162 ++*pos;
2163 return softnet_get_online(pos);
2164}
2165
2166static void softnet_seq_stop(struct seq_file *seq, void *v)
2167{
2168}
2169
2170static int softnet_seq_show(struct seq_file *seq, void *v)
2171{
2172 struct netif_rx_stats *s = v;
2173
2174 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
Stephen Hemminger31aa02c2005-06-23 20:12:48 -07002175 s->total, s->dropped, s->time_squeeze, 0,
Stephen Hemmingerc1ebcdb2005-06-23 20:08:59 -07002176 0, 0, 0, 0, /* was fastroute */
2177 s->cpu_collision );
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 return 0;
2179}
2180
2181static struct seq_operations dev_seq_ops = {
2182 .start = dev_seq_start,
2183 .next = dev_seq_next,
2184 .stop = dev_seq_stop,
2185 .show = dev_seq_show,
2186};
2187
2188static int dev_seq_open(struct inode *inode, struct file *file)
2189{
2190 return seq_open(file, &dev_seq_ops);
2191}
2192
2193static struct file_operations dev_seq_fops = {
2194 .owner = THIS_MODULE,
2195 .open = dev_seq_open,
2196 .read = seq_read,
2197 .llseek = seq_lseek,
2198 .release = seq_release,
2199};
2200
2201static struct seq_operations softnet_seq_ops = {
2202 .start = softnet_seq_start,
2203 .next = softnet_seq_next,
2204 .stop = softnet_seq_stop,
2205 .show = softnet_seq_show,
2206};
2207
2208static int softnet_seq_open(struct inode *inode, struct file *file)
2209{
2210 return seq_open(file, &softnet_seq_ops);
2211}
2212
2213static struct file_operations softnet_seq_fops = {
2214 .owner = THIS_MODULE,
2215 .open = softnet_seq_open,
2216 .read = seq_read,
2217 .llseek = seq_lseek,
2218 .release = seq_release,
2219};
2220
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002221#ifdef CONFIG_WIRELESS_EXT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222extern int wireless_proc_init(void);
2223#else
2224#define wireless_proc_init() 0
2225#endif
2226
2227static int __init dev_proc_init(void)
2228{
2229 int rc = -ENOMEM;
2230
2231 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2232 goto out;
2233 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2234 goto out_dev;
2235 if (wireless_proc_init())
2236 goto out_softnet;
2237 rc = 0;
2238out:
2239 return rc;
2240out_softnet:
2241 proc_net_remove("softnet_stat");
2242out_dev:
2243 proc_net_remove("dev");
2244 goto out;
2245}
2246#else
2247#define dev_proc_init() 0
2248#endif /* CONFIG_PROC_FS */
2249
2250
2251/**
2252 * netdev_set_master - set up master/slave pair
2253 * @slave: slave device
2254 * @master: new master device
2255 *
2256 * Changes the master device of the slave. Pass %NULL to break the
2257 * bonding. The caller must hold the RTNL semaphore. On a failure
2258 * a negative errno code is returned. On success the reference counts
2259 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2260 * function returns zero.
2261 */
2262int netdev_set_master(struct net_device *slave, struct net_device *master)
2263{
2264 struct net_device *old = slave->master;
2265
2266 ASSERT_RTNL();
2267
2268 if (master) {
2269 if (old)
2270 return -EBUSY;
2271 dev_hold(master);
2272 }
2273
2274 slave->master = master;
2275
2276 synchronize_net();
2277
2278 if (old)
2279 dev_put(old);
2280
2281 if (master)
2282 slave->flags |= IFF_SLAVE;
2283 else
2284 slave->flags &= ~IFF_SLAVE;
2285
2286 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2287 return 0;
2288}
2289
2290/**
2291 * dev_set_promiscuity - update promiscuity count on a device
2292 * @dev: device
2293 * @inc: modifier
2294 *
Stephen Hemminger3041a062006-05-26 13:25:24 -07002295 * Add or remove promiscuity from a device. While the count in the device
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 * remains above zero the interface remains promiscuous. Once it hits zero
2297 * the device reverts back to normal filtering operation. A negative inc
2298 * value is used to drop promiscuity on the device.
2299 */
2300void dev_set_promiscuity(struct net_device *dev, int inc)
2301{
2302 unsigned short old_flags = dev->flags;
2303
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 if ((dev->promiscuity += inc) == 0)
2305 dev->flags &= ~IFF_PROMISC;
David Chau52609c02005-07-05 15:11:06 -07002306 else
2307 dev->flags |= IFF_PROMISC;
2308 if (dev->flags != old_flags) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 dev_mc_upload(dev);
2310 printk(KERN_INFO "device %s %s promiscuous mode\n",
2311 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2312 "left");
Steve Grubb5bdb9882005-12-03 08:39:35 -05002313 audit_log(current->audit_context, GFP_ATOMIC,
2314 AUDIT_ANOM_PROMISCUOUS,
2315 "dev=%s prom=%d old_prom=%d auid=%u",
2316 dev->name, (dev->flags & IFF_PROMISC),
2317 (old_flags & IFF_PROMISC),
2318 audit_get_loginuid(current->audit_context));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 }
2320}
2321
2322/**
2323 * dev_set_allmulti - update allmulti count on a device
2324 * @dev: device
2325 * @inc: modifier
2326 *
2327 * Add or remove reception of all multicast frames to a device. While the
2328 * count in the device remains above zero the interface remains listening
2329 * to all interfaces. Once it hits zero the device reverts back to normal
2330 * filtering operation. A negative @inc value is used to drop the counter
2331 * when releasing a resource needing all multicasts.
2332 */
2333
2334void dev_set_allmulti(struct net_device *dev, int inc)
2335{
2336 unsigned short old_flags = dev->flags;
2337
2338 dev->flags |= IFF_ALLMULTI;
2339 if ((dev->allmulti += inc) == 0)
2340 dev->flags &= ~IFF_ALLMULTI;
2341 if (dev->flags ^ old_flags)
2342 dev_mc_upload(dev);
2343}
2344
2345unsigned dev_get_flags(const struct net_device *dev)
2346{
2347 unsigned flags;
2348
2349 flags = (dev->flags & ~(IFF_PROMISC |
2350 IFF_ALLMULTI |
Stefan Rompfb00055a2006-03-20 17:09:11 -08002351 IFF_RUNNING |
2352 IFF_LOWER_UP |
2353 IFF_DORMANT)) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 (dev->gflags & (IFF_PROMISC |
2355 IFF_ALLMULTI));
2356
Stefan Rompfb00055a2006-03-20 17:09:11 -08002357 if (netif_running(dev)) {
2358 if (netif_oper_up(dev))
2359 flags |= IFF_RUNNING;
2360 if (netif_carrier_ok(dev))
2361 flags |= IFF_LOWER_UP;
2362 if (netif_dormant(dev))
2363 flags |= IFF_DORMANT;
2364 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
2366 return flags;
2367}
2368
2369int dev_change_flags(struct net_device *dev, unsigned flags)
2370{
2371 int ret;
2372 int old_flags = dev->flags;
2373
2374 /*
2375 * Set the flags on our device.
2376 */
2377
2378 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2379 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2380 IFF_AUTOMEDIA)) |
2381 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2382 IFF_ALLMULTI));
2383
2384 /*
2385 * Load in the correct multicast list now the flags have changed.
2386 */
2387
2388 dev_mc_upload(dev);
2389
2390 /*
2391 * Have we downed the interface. We handle IFF_UP ourselves
2392 * according to user attempts to set it, rather than blindly
2393 * setting it.
2394 */
2395
2396 ret = 0;
2397 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2398 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2399
2400 if (!ret)
2401 dev_mc_upload(dev);
2402 }
2403
2404 if (dev->flags & IFF_UP &&
2405 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2406 IFF_VOLATILE)))
Alan Sternf07d5b92006-05-09 15:23:03 -07002407 raw_notifier_call_chain(&netdev_chain,
Alan Sterne041c682006-03-27 01:16:30 -08002408 NETDEV_CHANGE, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
2410 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2411 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2412 dev->gflags ^= IFF_PROMISC;
2413 dev_set_promiscuity(dev, inc);
2414 }
2415
2416 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2417 is important. Some (broken) drivers set IFF_PROMISC, when
2418 IFF_ALLMULTI is requested not asking us and not reporting.
2419 */
2420 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2421 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2422 dev->gflags ^= IFF_ALLMULTI;
2423 dev_set_allmulti(dev, inc);
2424 }
2425
2426 if (old_flags ^ dev->flags)
2427 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2428
2429 return ret;
2430}
2431
2432int dev_set_mtu(struct net_device *dev, int new_mtu)
2433{
2434 int err;
2435
2436 if (new_mtu == dev->mtu)
2437 return 0;
2438
2439 /* MTU must be positive. */
2440 if (new_mtu < 0)
2441 return -EINVAL;
2442
2443 if (!netif_device_present(dev))
2444 return -ENODEV;
2445
2446 err = 0;
2447 if (dev->change_mtu)
2448 err = dev->change_mtu(dev, new_mtu);
2449 else
2450 dev->mtu = new_mtu;
2451 if (!err && dev->flags & IFF_UP)
Alan Sternf07d5b92006-05-09 15:23:03 -07002452 raw_notifier_call_chain(&netdev_chain,
Alan Sterne041c682006-03-27 01:16:30 -08002453 NETDEV_CHANGEMTU, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 return err;
2455}
2456
2457int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2458{
2459 int err;
2460
2461 if (!dev->set_mac_address)
2462 return -EOPNOTSUPP;
2463 if (sa->sa_family != dev->type)
2464 return -EINVAL;
2465 if (!netif_device_present(dev))
2466 return -ENODEV;
2467 err = dev->set_mac_address(dev, sa);
2468 if (!err)
Alan Sternf07d5b92006-05-09 15:23:03 -07002469 raw_notifier_call_chain(&netdev_chain,
Alan Sterne041c682006-03-27 01:16:30 -08002470 NETDEV_CHANGEADDR, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 return err;
2472}
2473
2474/*
2475 * Perform the SIOCxIFxxx calls.
2476 */
2477static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2478{
2479 int err;
2480 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2481
2482 if (!dev)
2483 return -ENODEV;
2484
2485 switch (cmd) {
2486 case SIOCGIFFLAGS: /* Get interface flags */
2487 ifr->ifr_flags = dev_get_flags(dev);
2488 return 0;
2489
2490 case SIOCSIFFLAGS: /* Set interface flags */
2491 return dev_change_flags(dev, ifr->ifr_flags);
2492
2493 case SIOCGIFMETRIC: /* Get the metric on the interface
2494 (currently unused) */
2495 ifr->ifr_metric = 0;
2496 return 0;
2497
2498 case SIOCSIFMETRIC: /* Set the metric on the interface
2499 (currently unused) */
2500 return -EOPNOTSUPP;
2501
2502 case SIOCGIFMTU: /* Get the MTU of a device */
2503 ifr->ifr_mtu = dev->mtu;
2504 return 0;
2505
2506 case SIOCSIFMTU: /* Set the MTU of a device */
2507 return dev_set_mtu(dev, ifr->ifr_mtu);
2508
2509 case SIOCGIFHWADDR:
2510 if (!dev->addr_len)
2511 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2512 else
2513 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2514 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2515 ifr->ifr_hwaddr.sa_family = dev->type;
2516 return 0;
2517
2518 case SIOCSIFHWADDR:
2519 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2520
2521 case SIOCSIFHWBROADCAST:
2522 if (ifr->ifr_hwaddr.sa_family != dev->type)
2523 return -EINVAL;
2524 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2525 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
Alan Sternf07d5b92006-05-09 15:23:03 -07002526 raw_notifier_call_chain(&netdev_chain,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 NETDEV_CHANGEADDR, dev);
2528 return 0;
2529
2530 case SIOCGIFMAP:
2531 ifr->ifr_map.mem_start = dev->mem_start;
2532 ifr->ifr_map.mem_end = dev->mem_end;
2533 ifr->ifr_map.base_addr = dev->base_addr;
2534 ifr->ifr_map.irq = dev->irq;
2535 ifr->ifr_map.dma = dev->dma;
2536 ifr->ifr_map.port = dev->if_port;
2537 return 0;
2538
2539 case SIOCSIFMAP:
2540 if (dev->set_config) {
2541 if (!netif_device_present(dev))
2542 return -ENODEV;
2543 return dev->set_config(dev, &ifr->ifr_map);
2544 }
2545 return -EOPNOTSUPP;
2546
2547 case SIOCADDMULTI:
2548 if (!dev->set_multicast_list ||
2549 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2550 return -EINVAL;
2551 if (!netif_device_present(dev))
2552 return -ENODEV;
2553 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2554 dev->addr_len, 1);
2555
2556 case SIOCDELMULTI:
2557 if (!dev->set_multicast_list ||
2558 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2559 return -EINVAL;
2560 if (!netif_device_present(dev))
2561 return -ENODEV;
2562 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2563 dev->addr_len, 1);
2564
2565 case SIOCGIFINDEX:
2566 ifr->ifr_ifindex = dev->ifindex;
2567 return 0;
2568
2569 case SIOCGIFTXQLEN:
2570 ifr->ifr_qlen = dev->tx_queue_len;
2571 return 0;
2572
2573 case SIOCSIFTXQLEN:
2574 if (ifr->ifr_qlen < 0)
2575 return -EINVAL;
2576 dev->tx_queue_len = ifr->ifr_qlen;
2577 return 0;
2578
2579 case SIOCSIFNAME:
2580 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2581 return dev_change_name(dev, ifr->ifr_newname);
2582
2583 /*
2584 * Unknown or private ioctl
2585 */
2586
2587 default:
2588 if ((cmd >= SIOCDEVPRIVATE &&
2589 cmd <= SIOCDEVPRIVATE + 15) ||
2590 cmd == SIOCBONDENSLAVE ||
2591 cmd == SIOCBONDRELEASE ||
2592 cmd == SIOCBONDSETHWADDR ||
2593 cmd == SIOCBONDSLAVEINFOQUERY ||
2594 cmd == SIOCBONDINFOQUERY ||
2595 cmd == SIOCBONDCHANGEACTIVE ||
2596 cmd == SIOCGMIIPHY ||
2597 cmd == SIOCGMIIREG ||
2598 cmd == SIOCSMIIREG ||
2599 cmd == SIOCBRADDIF ||
2600 cmd == SIOCBRDELIF ||
2601 cmd == SIOCWANDEV) {
2602 err = -EOPNOTSUPP;
2603 if (dev->do_ioctl) {
2604 if (netif_device_present(dev))
2605 err = dev->do_ioctl(dev, ifr,
2606 cmd);
2607 else
2608 err = -ENODEV;
2609 }
2610 } else
2611 err = -EINVAL;
2612
2613 }
2614 return err;
2615}
2616
2617/*
2618 * This function handles all "interface"-type I/O control requests. The actual
2619 * 'doing' part of this is dev_ifsioc above.
2620 */
2621
2622/**
2623 * dev_ioctl - network device ioctl
2624 * @cmd: command to issue
2625 * @arg: pointer to a struct ifreq in user space
2626 *
2627 * Issue ioctl functions to devices. This is normally called by the
2628 * user space syscall interfaces but can sometimes be useful for
2629 * other purposes. The return value is the return from the syscall if
2630 * positive or a negative errno code on error.
2631 */
2632
2633int dev_ioctl(unsigned int cmd, void __user *arg)
2634{
2635 struct ifreq ifr;
2636 int ret;
2637 char *colon;
2638
2639 /* One special case: SIOCGIFCONF takes ifconf argument
2640 and requires shared lock, because it sleeps writing
2641 to user space.
2642 */
2643
2644 if (cmd == SIOCGIFCONF) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002645 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 ret = dev_ifconf((char __user *) arg);
Stephen Hemminger6756ae42006-03-20 22:23:58 -08002647 rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 return ret;
2649 }
2650 if (cmd == SIOCGIFNAME)
2651 return dev_ifname((struct ifreq __user *)arg);
2652
2653 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2654 return -EFAULT;
2655
2656 ifr.ifr_name[IFNAMSIZ-1] = 0;
2657
2658 colon = strchr(ifr.ifr_name, ':');
2659 if (colon)
2660 *colon = 0;
2661
2662 /*
2663 * See which interface the caller is talking about.
2664 */
2665
2666 switch (cmd) {
2667 /*
2668 * These ioctl calls:
2669 * - can be done by all.
2670 * - atomic and do not require locking.
2671 * - return a value
2672 */
2673 case SIOCGIFFLAGS:
2674 case SIOCGIFMETRIC:
2675 case SIOCGIFMTU:
2676 case SIOCGIFHWADDR:
2677 case SIOCGIFSLAVE:
2678 case SIOCGIFMAP:
2679 case SIOCGIFINDEX:
2680 case SIOCGIFTXQLEN:
2681 dev_load(ifr.ifr_name);
2682 read_lock(&dev_base_lock);
2683 ret = dev_ifsioc(&ifr, cmd);
2684 read_unlock(&dev_base_lock);
2685 if (!ret) {
2686 if (colon)
2687 *colon = ':';
2688 if (copy_to_user(arg, &ifr,
2689 sizeof(struct ifreq)))
2690 ret = -EFAULT;
2691 }
2692 return ret;
2693
2694 case SIOCETHTOOL:
2695 dev_load(ifr.ifr_name);
2696 rtnl_lock();
2697 ret = dev_ethtool(&ifr);
2698 rtnl_unlock();
2699 if (!ret) {
2700 if (colon)
2701 *colon = ':';
2702 if (copy_to_user(arg, &ifr,
2703 sizeof(struct ifreq)))
2704 ret = -EFAULT;
2705 }
2706 return ret;
2707
2708 /*
2709 * These ioctl calls:
2710 * - require superuser power.
2711 * - require strict serialization.
2712 * - return a value
2713 */
2714 case SIOCGMIIPHY:
2715 case SIOCGMIIREG:
2716 case SIOCSIFNAME:
2717 if (!capable(CAP_NET_ADMIN))
2718 return -EPERM;
2719 dev_load(ifr.ifr_name);
2720 rtnl_lock();
2721 ret = dev_ifsioc(&ifr, cmd);
2722 rtnl_unlock();
2723 if (!ret) {
2724 if (colon)
2725 *colon = ':';
2726 if (copy_to_user(arg, &ifr,
2727 sizeof(struct ifreq)))
2728 ret = -EFAULT;
2729 }
2730 return ret;
2731
2732 /*
2733 * These ioctl calls:
2734 * - require superuser power.
2735 * - require strict serialization.
2736 * - do not return a value
2737 */
2738 case SIOCSIFFLAGS:
2739 case SIOCSIFMETRIC:
2740 case SIOCSIFMTU:
2741 case SIOCSIFMAP:
2742 case SIOCSIFHWADDR:
2743 case SIOCSIFSLAVE:
2744 case SIOCADDMULTI:
2745 case SIOCDELMULTI:
2746 case SIOCSIFHWBROADCAST:
2747 case SIOCSIFTXQLEN:
2748 case SIOCSMIIREG:
2749 case SIOCBONDENSLAVE:
2750 case SIOCBONDRELEASE:
2751 case SIOCBONDSETHWADDR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 case SIOCBONDCHANGEACTIVE:
2753 case SIOCBRADDIF:
2754 case SIOCBRDELIF:
2755 if (!capable(CAP_NET_ADMIN))
2756 return -EPERM;
Thomas Grafcabcac02006-01-24 12:46:33 -08002757 /* fall through */
2758 case SIOCBONDSLAVEINFOQUERY:
2759 case SIOCBONDINFOQUERY:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 dev_load(ifr.ifr_name);
2761 rtnl_lock();
2762 ret = dev_ifsioc(&ifr, cmd);
2763 rtnl_unlock();
2764 return ret;
2765
2766 case SIOCGIFMEM:
2767 /* Get the per device memory space. We can add this but
2768 * currently do not support it */
2769 case SIOCSIFMEM:
2770 /* Set the per device memory buffer space.
2771 * Not applicable in our case */
2772 case SIOCSIFLINK:
2773 return -EINVAL;
2774
2775 /*
2776 * Unknown or private ioctl.
2777 */
2778 default:
2779 if (cmd == SIOCWANDEV ||
2780 (cmd >= SIOCDEVPRIVATE &&
2781 cmd <= SIOCDEVPRIVATE + 15)) {
2782 dev_load(ifr.ifr_name);
2783 rtnl_lock();
2784 ret = dev_ifsioc(&ifr, cmd);
2785 rtnl_unlock();
2786 if (!ret && copy_to_user(arg, &ifr,
2787 sizeof(struct ifreq)))
2788 ret = -EFAULT;
2789 return ret;
2790 }
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002791#ifdef CONFIG_WIRELESS_EXT
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 /* Take care of Wireless Extensions */
2793 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2794 /* If command is `set a parameter', or
2795 * `get the encoding parameters', check if
2796 * the user has the right to do it */
Jean Tourrilhesa4170162006-04-04 15:53:43 -07002797 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2798 || cmd == SIOCGIWENCODEEXT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 if (!capable(CAP_NET_ADMIN))
2800 return -EPERM;
2801 }
2802 dev_load(ifr.ifr_name);
2803 rtnl_lock();
2804 /* Follow me in net/core/wireless.c */
2805 ret = wireless_process_ioctl(&ifr, cmd);
2806 rtnl_unlock();
2807 if (IW_IS_GET(cmd) &&
2808 copy_to_user(arg, &ifr,
2809 sizeof(struct ifreq)))
2810 ret = -EFAULT;
2811 return ret;
2812 }
Adrian Bunkd86b5e02006-01-21 00:46:55 +01002813#endif /* CONFIG_WIRELESS_EXT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 return -EINVAL;
2815 }
2816}
2817
2818
2819/**
2820 * dev_new_index - allocate an ifindex
2821 *
2822 * Returns a suitable unique value for a new device interface
2823 * number. The caller must hold the rtnl semaphore or the
2824 * dev_base_lock to be sure it remains unique.
2825 */
2826static int dev_new_index(void)
2827{
2828 static int ifindex;
2829 for (;;) {
2830 if (++ifindex <= 0)
2831 ifindex = 1;
2832 if (!__dev_get_by_index(ifindex))
2833 return ifindex;
2834 }
2835}
2836
2837static int dev_boot_phase = 1;
2838
2839/* Delayed registration/unregisteration */
2840static DEFINE_SPINLOCK(net_todo_list_lock);
2841static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2842
2843static inline void net_set_todo(struct net_device *dev)
2844{
2845 spin_lock(&net_todo_list_lock);
2846 list_add_tail(&dev->todo_list, &net_todo_list);
2847 spin_unlock(&net_todo_list_lock);
2848}
2849
2850/**
2851 * register_netdevice - register a network device
2852 * @dev: device to register
2853 *
2854 * Take a completed network device structure and add it to the kernel
2855 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2856 * chain. 0 is returned on success. A negative errno code is returned
2857 * on a failure to set up the device, or if the name is a duplicate.
2858 *
2859 * Callers must hold the rtnl semaphore. You may want
2860 * register_netdev() instead of this.
2861 *
2862 * BUGS:
2863 * The locking appears insufficient to guarantee two parallel registers
2864 * will not get the same name.
2865 */
2866
2867int register_netdevice(struct net_device *dev)
2868{
2869 struct hlist_head *head;
2870 struct hlist_node *p;
2871 int ret;
2872
2873 BUG_ON(dev_boot_phase);
2874 ASSERT_RTNL();
2875
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07002876 might_sleep();
2877
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 /* When net_device's are persistent, this will be fatal. */
2879 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2880
2881 spin_lock_init(&dev->queue_lock);
Herbert Xu932ff272006-06-09 12:20:56 -07002882 spin_lock_init(&dev->_xmit_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 dev->xmit_lock_owner = -1;
2884#ifdef CONFIG_NET_CLS_ACT
2885 spin_lock_init(&dev->ingress_lock);
2886#endif
2887
2888 ret = alloc_divert_blk(dev);
2889 if (ret)
2890 goto out;
2891
2892 dev->iflink = -1;
2893
2894 /* Init, if this function is available */
2895 if (dev->init) {
2896 ret = dev->init(dev);
2897 if (ret) {
2898 if (ret > 0)
2899 ret = -EIO;
2900 goto out_err;
2901 }
2902 }
2903
2904 if (!dev_valid_name(dev->name)) {
2905 ret = -EINVAL;
2906 goto out_err;
2907 }
2908
2909 dev->ifindex = dev_new_index();
2910 if (dev->iflink == -1)
2911 dev->iflink = dev->ifindex;
2912
2913 /* Check for existence of name */
2914 head = dev_name_hash(dev->name);
2915 hlist_for_each(p, head) {
2916 struct net_device *d
2917 = hlist_entry(p, struct net_device, name_hlist);
2918 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2919 ret = -EEXIST;
2920 goto out_err;
2921 }
2922 }
2923
2924 /* Fix illegal SG+CSUM combinations. */
2925 if ((dev->features & NETIF_F_SG) &&
Herbert Xu8648b302006-06-17 22:06:05 -07002926 !(dev->features & NETIF_F_ALL_CSUM)) {
Stephen Hemminger5a8da022006-07-07 16:54:05 -07002927 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 dev->name);
2929 dev->features &= ~NETIF_F_SG;
2930 }
2931
2932 /* TSO requires that SG is present as well. */
2933 if ((dev->features & NETIF_F_TSO) &&
2934 !(dev->features & NETIF_F_SG)) {
Stephen Hemminger5a8da022006-07-07 16:54:05 -07002935 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936 dev->name);
2937 dev->features &= ~NETIF_F_TSO;
2938 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07002939 if (dev->features & NETIF_F_UFO) {
2940 if (!(dev->features & NETIF_F_HW_CSUM)) {
2941 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2942 "NETIF_F_HW_CSUM feature.\n",
2943 dev->name);
2944 dev->features &= ~NETIF_F_UFO;
2945 }
2946 if (!(dev->features & NETIF_F_SG)) {
2947 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2948 "NETIF_F_SG feature.\n",
2949 dev->name);
2950 dev->features &= ~NETIF_F_UFO;
2951 }
2952 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953
2954 /*
2955 * nil rebuild_header routine,
2956 * that should be never called and used as just bug trap.
2957 */
2958
2959 if (!dev->rebuild_header)
2960 dev->rebuild_header = default_rebuild_header;
2961
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07002962 ret = netdev_register_sysfs(dev);
2963 if (ret)
2964 goto out_err;
2965 dev->reg_state = NETREG_REGISTERED;
2966
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 /*
2968 * Default initial state at registry is that the
2969 * device is present.
2970 */
2971
2972 set_bit(__LINK_STATE_PRESENT, &dev->state);
2973
2974 dev->next = NULL;
2975 dev_init_scheduler(dev);
2976 write_lock_bh(&dev_base_lock);
2977 *dev_tail = dev;
2978 dev_tail = &dev->next;
2979 hlist_add_head(&dev->name_hlist, head);
2980 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2981 dev_hold(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 write_unlock_bh(&dev_base_lock);
2983
2984 /* Notify protocols, that a new device appeared. */
Alan Sternf07d5b92006-05-09 15:23:03 -07002985 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 ret = 0;
2988
2989out:
2990 return ret;
2991out_err:
2992 free_divert_blk(dev);
2993 goto out;
2994}
2995
2996/**
2997 * register_netdev - register a network device
2998 * @dev: device to register
2999 *
3000 * Take a completed network device structure and add it to the kernel
3001 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3002 * chain. 0 is returned on success. A negative errno code is returned
3003 * on a failure to set up the device, or if the name is a duplicate.
3004 *
3005 * This is a wrapper around register_netdev that takes the rtnl semaphore
3006 * and expands the device name if you passed a format string to
3007 * alloc_netdev.
3008 */
3009int register_netdev(struct net_device *dev)
3010{
3011 int err;
3012
3013 rtnl_lock();
3014
3015 /*
3016 * If the name is a format string the caller wants us to do a
3017 * name allocation.
3018 */
3019 if (strchr(dev->name, '%')) {
3020 err = dev_alloc_name(dev, dev->name);
3021 if (err < 0)
3022 goto out;
3023 }
3024
3025 /*
3026 * Back compatibility hook. Kill this one in 2.5
3027 */
3028 if (dev->name[0] == 0 || dev->name[0] == ' ') {
3029 err = dev_alloc_name(dev, "eth%d");
3030 if (err < 0)
3031 goto out;
3032 }
3033
3034 err = register_netdevice(dev);
3035out:
3036 rtnl_unlock();
3037 return err;
3038}
3039EXPORT_SYMBOL(register_netdev);
3040
3041/*
3042 * netdev_wait_allrefs - wait until all references are gone.
3043 *
3044 * This is called when unregistering network devices.
3045 *
3046 * Any protocol or device that holds a reference should register
3047 * for netdevice notification, and cleanup and put back the
3048 * reference if they receive an UNREGISTER event.
3049 * We can get stuck here if buggy protocols don't correctly
3050 * call dev_put.
3051 */
3052static void netdev_wait_allrefs(struct net_device *dev)
3053{
3054 unsigned long rebroadcast_time, warning_time;
3055
3056 rebroadcast_time = warning_time = jiffies;
3057 while (atomic_read(&dev->refcnt) != 0) {
3058 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08003059 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060
3061 /* Rebroadcast unregister notification */
Alan Sternf07d5b92006-05-09 15:23:03 -07003062 raw_notifier_call_chain(&netdev_chain,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 NETDEV_UNREGISTER, dev);
3064
3065 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3066 &dev->state)) {
3067 /* We must not have linkwatch events
3068 * pending on unregister. If this
3069 * happens, we simply run the queue
3070 * unscheduled, resulting in a noop
3071 * for this device.
3072 */
3073 linkwatch_run_queue();
3074 }
3075
Stephen Hemminger6756ae42006-03-20 22:23:58 -08003076 __rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077
3078 rebroadcast_time = jiffies;
3079 }
3080
3081 msleep(250);
3082
3083 if (time_after(jiffies, warning_time + 10 * HZ)) {
3084 printk(KERN_EMERG "unregister_netdevice: "
3085 "waiting for %s to become free. Usage "
3086 "count = %d\n",
3087 dev->name, atomic_read(&dev->refcnt));
3088 warning_time = jiffies;
3089 }
3090 }
3091}
3092
3093/* The sequence is:
3094 *
3095 * rtnl_lock();
3096 * ...
3097 * register_netdevice(x1);
3098 * register_netdevice(x2);
3099 * ...
3100 * unregister_netdevice(y1);
3101 * unregister_netdevice(y2);
3102 * ...
3103 * rtnl_unlock();
3104 * free_netdev(y1);
3105 * free_netdev(y2);
3106 *
3107 * We are invoked by rtnl_unlock() after it drops the semaphore.
3108 * This allows us to deal with problems:
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003109 * 1) We can delete sysfs objects which invoke hotplug
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110 * without deadlocking with linkwatch via keventd.
3111 * 2) Since we run with the RTNL semaphore not held, we can sleep
3112 * safely in order to wait for the netdev refcnt to drop to zero.
3113 */
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08003114static DEFINE_MUTEX(net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115void netdev_run_todo(void)
3116{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003117 struct list_head list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118
3119 /* Need to guard against multiple cpu's getting out of order. */
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08003120 mutex_lock(&net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121
3122 /* Not safe to do outside the semaphore. We must not return
3123 * until all unregister events invoked by the local processor
3124 * have been completed (either by this todo run, or one on
3125 * another cpu).
3126 */
3127 if (list_empty(&net_todo_list))
3128 goto out;
3129
3130 /* Snapshot list, allow later requests */
3131 spin_lock(&net_todo_list_lock);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003132 list_replace_init(&net_todo_list, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133 spin_unlock(&net_todo_list_lock);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003134
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 while (!list_empty(&list)) {
3136 struct net_device *dev
3137 = list_entry(list.next, struct net_device, todo_list);
3138 list_del(&dev->todo_list);
3139
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003140 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141 printk(KERN_ERR "network todo '%s' but state %d\n",
3142 dev->name, dev->reg_state);
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003143 dump_stack();
3144 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 }
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003146
3147 netdev_unregister_sysfs(dev);
3148 dev->reg_state = NETREG_UNREGISTERED;
3149
3150 netdev_wait_allrefs(dev);
3151
3152 /* paranoia */
3153 BUG_ON(atomic_read(&dev->refcnt));
3154 BUG_TRAP(!dev->ip_ptr);
3155 BUG_TRAP(!dev->ip6_ptr);
3156 BUG_TRAP(!dev->dn_ptr);
3157
3158 /* It must be the very last action,
3159 * after this 'dev' may point to freed up memory.
3160 */
3161 if (dev->destructor)
3162 dev->destructor(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 }
3164
3165out:
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08003166 mutex_unlock(&net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167}
3168
3169/**
3170 * alloc_netdev - allocate network device
3171 * @sizeof_priv: size of private data to allocate space for
3172 * @name: device name format string
3173 * @setup: callback to initialize device
3174 *
3175 * Allocates a struct net_device with private data area for driver use
3176 * and performs basic initialization.
3177 */
3178struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3179 void (*setup)(struct net_device *))
3180{
3181 void *p;
3182 struct net_device *dev;
3183 int alloc_size;
3184
3185 /* ensure 32-byte alignment of both the device and private area */
3186 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3187 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3188
Paolo 'Blaisorblade' Giarrusso31380de2006-04-06 22:38:28 -07003189 p = kzalloc(alloc_size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190 if (!p) {
3191 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3192 return NULL;
3193 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194
3195 dev = (struct net_device *)
3196 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3197 dev->padded = (char *)dev - (char *)p;
3198
3199 if (sizeof_priv)
3200 dev->priv = netdev_priv(dev);
3201
3202 setup(dev);
3203 strcpy(dev->name, name);
3204 return dev;
3205}
3206EXPORT_SYMBOL(alloc_netdev);
3207
3208/**
3209 * free_netdev - free network device
3210 * @dev: device
3211 *
3212 * This function does the last stage of destroying an allocated device
3213 * interface. The reference to the device object is released.
3214 * If this is the last reference then it will be freed.
3215 */
3216void free_netdev(struct net_device *dev)
3217{
3218#ifdef CONFIG_SYSFS
Stephen Hemminger3041a062006-05-26 13:25:24 -07003219 /* Compatibility with error handling in drivers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220 if (dev->reg_state == NETREG_UNINITIALIZED) {
3221 kfree((char *)dev - dev->padded);
3222 return;
3223 }
3224
3225 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3226 dev->reg_state = NETREG_RELEASED;
3227
3228 /* will free via class release */
3229 class_device_put(&dev->class_dev);
3230#else
3231 kfree((char *)dev - dev->padded);
3232#endif
3233}
3234
3235/* Synchronize with packet receive processing. */
3236void synchronize_net(void)
3237{
3238 might_sleep();
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07003239 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240}
3241
3242/**
3243 * unregister_netdevice - remove device from the kernel
3244 * @dev: device
3245 *
3246 * This function shuts down a device interface and removes it
3247 * from the kernel tables. On success 0 is returned, on a failure
3248 * a negative errno code is returned.
3249 *
3250 * Callers must hold the rtnl semaphore. You may want
3251 * unregister_netdev() instead of this.
3252 */
3253
3254int unregister_netdevice(struct net_device *dev)
3255{
3256 struct net_device *d, **dp;
3257
3258 BUG_ON(dev_boot_phase);
3259 ASSERT_RTNL();
3260
3261 /* Some devices call without registering for initialization unwind. */
3262 if (dev->reg_state == NETREG_UNINITIALIZED) {
3263 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3264 "was registered\n", dev->name, dev);
3265 return -ENODEV;
3266 }
3267
3268 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3269
3270 /* If device is running, close it first. */
3271 if (dev->flags & IFF_UP)
3272 dev_close(dev);
3273
3274 /* And unlink it from device chain. */
3275 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3276 if (d == dev) {
3277 write_lock_bh(&dev_base_lock);
3278 hlist_del(&dev->name_hlist);
3279 hlist_del(&dev->index_hlist);
3280 if (dev_tail == &dev->next)
3281 dev_tail = dp;
3282 *dp = d->next;
3283 write_unlock_bh(&dev_base_lock);
3284 break;
3285 }
3286 }
3287 if (!d) {
3288 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3289 dev->name);
3290 return -ENODEV;
3291 }
3292
3293 dev->reg_state = NETREG_UNREGISTERING;
3294
3295 synchronize_net();
3296
3297 /* Shutdown queueing discipline. */
3298 dev_shutdown(dev);
3299
3300
3301 /* Notify protocols, that we are about to destroy
3302 this device. They should clean all the things.
3303 */
Alan Sternf07d5b92006-05-09 15:23:03 -07003304 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305
3306 /*
3307 * Flush the multicast chain
3308 */
3309 dev_mc_discard(dev);
3310
3311 if (dev->uninit)
3312 dev->uninit(dev);
3313
3314 /* Notifier chain MUST detach us from master device. */
3315 BUG_TRAP(!dev->master);
3316
3317 free_divert_blk(dev);
3318
3319 /* Finish processing unregister after unlock */
3320 net_set_todo(dev);
3321
3322 synchronize_net();
3323
3324 dev_put(dev);
3325 return 0;
3326}
3327
3328/**
3329 * unregister_netdev - remove device from the kernel
3330 * @dev: device
3331 *
3332 * This function shuts down a device interface and removes it
3333 * from the kernel tables. On success 0 is returned, on a failure
3334 * a negative errno code is returned.
3335 *
3336 * This is just a wrapper for unregister_netdevice that takes
3337 * the rtnl semaphore. In general you want to use this and not
3338 * unregister_netdevice.
3339 */
3340void unregister_netdev(struct net_device *dev)
3341{
3342 rtnl_lock();
3343 unregister_netdevice(dev);
3344 rtnl_unlock();
3345}
3346
3347EXPORT_SYMBOL(unregister_netdev);
3348
3349#ifdef CONFIG_HOTPLUG_CPU
3350static int dev_cpu_callback(struct notifier_block *nfb,
3351 unsigned long action,
3352 void *ocpu)
3353{
3354 struct sk_buff **list_skb;
3355 struct net_device **list_net;
3356 struct sk_buff *skb;
3357 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3358 struct softnet_data *sd, *oldsd;
3359
3360 if (action != CPU_DEAD)
3361 return NOTIFY_OK;
3362
3363 local_irq_disable();
3364 cpu = smp_processor_id();
3365 sd = &per_cpu(softnet_data, cpu);
3366 oldsd = &per_cpu(softnet_data, oldcpu);
3367
3368 /* Find end of our completion_queue. */
3369 list_skb = &sd->completion_queue;
3370 while (*list_skb)
3371 list_skb = &(*list_skb)->next;
3372 /* Append completion queue from offline CPU. */
3373 *list_skb = oldsd->completion_queue;
3374 oldsd->completion_queue = NULL;
3375
3376 /* Find end of our output_queue. */
3377 list_net = &sd->output_queue;
3378 while (*list_net)
3379 list_net = &(*list_net)->next_sched;
3380 /* Append output queue from offline CPU. */
3381 *list_net = oldsd->output_queue;
3382 oldsd->output_queue = NULL;
3383
3384 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3385 local_irq_enable();
3386
3387 /* Process offline CPU's input_pkt_queue */
3388 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3389 netif_rx(skb);
3390
3391 return NOTIFY_OK;
3392}
3393#endif /* CONFIG_HOTPLUG_CPU */
3394
Chris Leechdb217332006-06-17 21:24:58 -07003395#ifdef CONFIG_NET_DMA
3396/**
3397 * net_dma_rebalance -
3398 * This is called when the number of channels allocated to the net_dma_client
3399 * changes. The net_dma_client tries to have one DMA channel per CPU.
3400 */
3401static void net_dma_rebalance(void)
3402{
3403 unsigned int cpu, i, n;
3404 struct dma_chan *chan;
3405
Chris Leechdb217332006-06-17 21:24:58 -07003406 if (net_dma_count == 0) {
3407 for_each_online_cpu(cpu)
Alexey Dobriyan29bbd722006-08-02 15:02:31 -07003408 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
Chris Leechdb217332006-06-17 21:24:58 -07003409 return;
3410 }
3411
3412 i = 0;
3413 cpu = first_cpu(cpu_online_map);
3414
3415 rcu_read_lock();
3416 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3417 n = ((num_online_cpus() / net_dma_count)
3418 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3419
3420 while(n) {
Alexey Dobriyan29bbd722006-08-02 15:02:31 -07003421 per_cpu(softnet_data, cpu).net_dma = chan;
Chris Leechdb217332006-06-17 21:24:58 -07003422 cpu = next_cpu(cpu, cpu_online_map);
3423 n--;
3424 }
3425 i++;
3426 }
3427 rcu_read_unlock();
Chris Leechdb217332006-06-17 21:24:58 -07003428}
3429
3430/**
3431 * netdev_dma_event - event callback for the net_dma_client
3432 * @client: should always be net_dma_client
Randy Dunlapf4b8ea72006-06-22 16:00:11 -07003433 * @chan: DMA channel for the event
3434 * @event: event type
Chris Leechdb217332006-06-17 21:24:58 -07003435 */
3436static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3437 enum dma_event event)
3438{
3439 spin_lock(&net_dma_event_lock);
3440 switch (event) {
3441 case DMA_RESOURCE_ADDED:
3442 net_dma_count++;
3443 net_dma_rebalance();
3444 break;
3445 case DMA_RESOURCE_REMOVED:
3446 net_dma_count--;
3447 net_dma_rebalance();
3448 break;
3449 default:
3450 break;
3451 }
3452 spin_unlock(&net_dma_event_lock);
3453}
3454
3455/**
3456 * netdev_dma_regiser - register the networking subsystem as a DMA client
3457 */
3458static int __init netdev_dma_register(void)
3459{
3460 spin_lock_init(&net_dma_event_lock);
3461 net_dma_client = dma_async_client_register(netdev_dma_event);
3462 if (net_dma_client == NULL)
3463 return -ENOMEM;
3464
3465 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3466 return 0;
3467}
3468
3469#else
3470static int __init netdev_dma_register(void) { return -ENODEV; }
3471#endif /* CONFIG_NET_DMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472
3473/*
3474 * Initialize the DEV module. At boot time this walks the device list and
3475 * unhooks any devices that fail to initialise (normally hardware not
3476 * present) and leaves us with a valid list of present and active devices.
3477 *
3478 */
3479
3480/*
3481 * This is called single threaded during boot, so no need
3482 * to take the rtnl semaphore.
3483 */
3484static int __init net_dev_init(void)
3485{
3486 int i, rc = -ENOMEM;
3487
3488 BUG_ON(!dev_boot_phase);
3489
3490 net_random_init();
3491
3492 if (dev_proc_init())
3493 goto out;
3494
3495 if (netdev_sysfs_init())
3496 goto out;
3497
3498 INIT_LIST_HEAD(&ptype_all);
3499 for (i = 0; i < 16; i++)
3500 INIT_LIST_HEAD(&ptype_base[i]);
3501
3502 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3503 INIT_HLIST_HEAD(&dev_name_head[i]);
3504
3505 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3506 INIT_HLIST_HEAD(&dev_index_head[i]);
3507
3508 /*
3509 * Initialise the packet receive queues.
3510 */
3511
KAMEZAWA Hiroyuki6f912042006-04-10 22:52:50 -07003512 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513 struct softnet_data *queue;
3514
3515 queue = &per_cpu(softnet_data, i);
3516 skb_queue_head_init(&queue->input_pkt_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 queue->completion_queue = NULL;
3518 INIT_LIST_HEAD(&queue->poll_list);
3519 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3520 queue->backlog_dev.weight = weight_p;
3521 queue->backlog_dev.poll = process_backlog;
3522 atomic_set(&queue->backlog_dev.refcnt, 1);
3523 }
3524
Chris Leechdb217332006-06-17 21:24:58 -07003525 netdev_dma_register();
3526
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527 dev_boot_phase = 0;
3528
3529 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3530 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3531
3532 hotcpu_notifier(dev_cpu_callback, 0);
3533 dst_init();
3534 dev_mcast_init();
3535 rc = 0;
3536out:
3537 return rc;
3538}
3539
3540subsys_initcall(net_dev_init);
3541
3542EXPORT_SYMBOL(__dev_get_by_index);
3543EXPORT_SYMBOL(__dev_get_by_name);
3544EXPORT_SYMBOL(__dev_remove_pack);
Mitch Williamsc2373ee2005-11-09 10:34:45 -08003545EXPORT_SYMBOL(dev_valid_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546EXPORT_SYMBOL(dev_add_pack);
3547EXPORT_SYMBOL(dev_alloc_name);
3548EXPORT_SYMBOL(dev_close);
3549EXPORT_SYMBOL(dev_get_by_flags);
3550EXPORT_SYMBOL(dev_get_by_index);
3551EXPORT_SYMBOL(dev_get_by_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552EXPORT_SYMBOL(dev_open);
3553EXPORT_SYMBOL(dev_queue_xmit);
3554EXPORT_SYMBOL(dev_remove_pack);
3555EXPORT_SYMBOL(dev_set_allmulti);
3556EXPORT_SYMBOL(dev_set_promiscuity);
3557EXPORT_SYMBOL(dev_change_flags);
3558EXPORT_SYMBOL(dev_set_mtu);
3559EXPORT_SYMBOL(dev_set_mac_address);
3560EXPORT_SYMBOL(free_netdev);
3561EXPORT_SYMBOL(netdev_boot_setup_check);
3562EXPORT_SYMBOL(netdev_set_master);
3563EXPORT_SYMBOL(netdev_state_change);
3564EXPORT_SYMBOL(netif_receive_skb);
3565EXPORT_SYMBOL(netif_rx);
3566EXPORT_SYMBOL(register_gifconf);
3567EXPORT_SYMBOL(register_netdevice);
3568EXPORT_SYMBOL(register_netdevice_notifier);
3569EXPORT_SYMBOL(skb_checksum_help);
3570EXPORT_SYMBOL(synchronize_net);
3571EXPORT_SYMBOL(unregister_netdevice);
3572EXPORT_SYMBOL(unregister_netdevice_notifier);
3573EXPORT_SYMBOL(net_enable_timestamp);
3574EXPORT_SYMBOL(net_disable_timestamp);
3575EXPORT_SYMBOL(dev_get_flags);
3576
3577#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3578EXPORT_SYMBOL(br_handle_frame_hook);
3579EXPORT_SYMBOL(br_fdb_get_hook);
3580EXPORT_SYMBOL(br_fdb_put_hook);
3581#endif
3582
3583#ifdef CONFIG_KMOD
3584EXPORT_SYMBOL(dev_load);
3585#endif
3586
3587EXPORT_PER_CPU_SYMBOL(softnet_data);