]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/core/dev.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[NET]: network dev read_mostly
[net-next-2.6.git] / net / core / dev.c
CommitLineData
1da177e4
LT		 1/*
2 * NET3 Protocol independent device support routines.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Derived from the non IP parts of dev.c 1.0.19
02c30a84 10 * Authors: Ross Biro
1da177e4
LT		 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 *
14 * Additional Authors:
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
21 *
22 * Changes:
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
34 * drivers
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
44 * call a packet.
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
58 * 1 device.
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
66 * the backlog queue.
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
73 */
74
75#include <asm/uaccess.h>
76#include <asm/system.h>
77#include <linux/bitops.h>
4fc268d2 78#include <linux/capability.h>
1da177e4
LT		 79#include <linux/cpu.h>
80#include <linux/types.h>
81#include <linux/kernel.h>
82#include <linux/sched.h>
4a3e2f71 83#include <linux/mutex.h>
1da177e4
LT		 84#include <linux/string.h>
85#include <linux/mm.h>
86#include <linux/socket.h>
87#include <linux/sockios.h>
88#include <linux/errno.h>
89#include <linux/interrupt.h>
90#include <linux/if_ether.h>
91#include <linux/netdevice.h>
92#include <linux/etherdevice.h>
93#include <linux/notifier.h>
94#include <linux/skbuff.h>
95#include <net/sock.h>
96#include <linux/rtnetlink.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/stat.h>
100#include <linux/if_bridge.h>
1da177e4
LT		 101#include <net/dst.h>
102#include <net/pkt_sched.h>
103#include <net/checksum.h>
104#include <linux/highmem.h>
105#include <linux/init.h>
106#include <linux/kmod.h>
107#include <linux/module.h>
108#include <linux/kallsyms.h>
109#include <linux/netpoll.h>
110#include <linux/rcupdate.h>
111#include <linux/delay.h>
d86b5e0e 112#include <linux/wireless.h>
1da177e4 113#include <net/iw_handler.h>
1da177e4 114#include <asm/current.h>
5bdb9886 115#include <linux/audit.h>
db217334 116#include <linux/dmaengine.h>
f6a78bfc 117#include <linux/err.h>
c7fa9d18 118#include <linux/ctype.h>
1da177e4 119
1da177e4
LT		 120/*
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
3041a069 131 * the average user (w/out VLANs) will not be adversely affected.
1da177e4
LT		 132 * --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);
6b2bedc3
SH		 149static struct list_head ptype_base[16] __read_mostly; /* 16 way hashed list */
150static struct list_head ptype_all __read_mostly; /* Taps */
1da177e4 151
db217334
CL		 152#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
1da177e4 158/*
3041a069 159 * The @dev_base list is protected by @dev_base_lock and the rtnl
1da177e4
LT		 160 * 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
f07d5b94 203static RAW_NOTIFIER_HEAD(netdev_chain);
1da177e4
LT		 204
205/*
206 * Device drivers call our routines to queue packets here. We empty the
207 * queue in the local softnet handler.
208 */
31aa02c5 209DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
1da177e4
LT		 210
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
60481264 232static int netdev_nit;
1da177e4
LT		 233
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 *
4ec93edb 258 * This call does not sleep therefore it can not
1da177e4
LT		 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
1da177e4
LT		 278/**
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
4ec93edb 285 * returns.
1da177e4
LT		 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);
4ec93edb 330
1da177e4
LT		 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
cf309e3f
JF		 585EXPORT_SYMBOL(dev_getbyhwaddr);
586
1da177e4
LT		 587struct 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
4ec93edb 610 * is not found or a pointer to the device. The device returned has
1da177e4
LT		 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
c7fa9d18
DM		 635 * to allow sysfs to work. We also disallow any kind of
636 * whitespace.
1da177e4 637 */
c2373ee9 638int dev_valid_name(const char *name)
1da177e4 639{
c7fa9d18
DM		 640 if (*name == '\0')
641 return 0;
b6fe17d6
SH		 642 if (strlen(name) >= IFNAMSIZ)
643 return 0;
c7fa9d18
DM		 644 if (!strcmp(name, ".") || !strcmp(name, ".."))
645 return 0;
646
647 while (*name) {
648 if (*name == '/' || isspace(*name))
649 return 0;
650 name++;
651 }
652 return 1;
1da177e4
LT		 653}
654
655/**
656 * dev_alloc_name - allocate a name for a device
657 * @dev: device
658 * @name: name format string
659 *
660 * Passed a format string - eg "lt%d" it will try and find a suitable
3041a069
SH		 661 * id. It scans list of devices to build up a free map, then chooses
662 * the first empty slot. The caller must hold the dev_base or rtnl lock
663 * while allocating the name and adding the device in order to avoid
664 * duplicates.
665 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
666 * Returns the number of the unit assigned or a negative errno code.
1da177e4
LT		 667 */
668
669int dev_alloc_name(struct net_device *dev, const char *name)
670{
671 int i = 0;
672 char buf[IFNAMSIZ];
673 const char *p;
674 const int max_netdevices = 8*PAGE_SIZE;
675 long *inuse;
676 struct net_device *d;
677
678 p = strnchr(name, IFNAMSIZ-1, '%');
679 if (p) {
680 /*
681 * Verify the string as this thing may have come from
682 * the user. There must be either one "%d" and no other "%"
683 * characters.
684 */
685 if (p[1] != 'd' || strchr(p + 2, '%'))
686 return -EINVAL;
687
688 /* Use one page as a bit array of possible slots */
689 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
690 if (!inuse)
691 return -ENOMEM;
692
693 for (d = dev_base; d; d = d->next) {
694 if (!sscanf(d->name, name, &i))
695 continue;
696 if (i < 0 || i >= max_netdevices)
697 continue;
698
699 /* avoid cases where sscanf is not exact inverse of printf */
700 snprintf(buf, sizeof(buf), name, i);
701 if (!strncmp(buf, d->name, IFNAMSIZ))
702 set_bit(i, inuse);
703 }
704
705 i = find_first_zero_bit(inuse, max_netdevices);
706 free_page((unsigned long) inuse);
707 }
708
709 snprintf(buf, sizeof(buf), name, i);
710 if (!__dev_get_by_name(buf)) {
711 strlcpy(dev->name, buf, IFNAMSIZ);
712 return i;
713 }
714
715 /* It is possible to run out of possible slots
716 * when the name is long and there isn't enough space left
717 * for the digits, or if all bits are used.
718 */
719 return -ENFILE;
720}
721
722
723/**
724 * dev_change_name - change name of a device
725 * @dev: device
726 * @newname: name (or format string) must be at least IFNAMSIZ
727 *
728 * Change name of a device, can pass format strings "eth%d".
729 * for wildcarding.
730 */
731int dev_change_name(struct net_device *dev, char *newname)
732{
733 int err = 0;
734
735 ASSERT_RTNL();
736
737 if (dev->flags & IFF_UP)
738 return -EBUSY;
739
740 if (!dev_valid_name(newname))
741 return -EINVAL;
742
743 if (strchr(newname, '%')) {
744 err = dev_alloc_name(dev, newname);
745 if (err < 0)
746 return err;
747 strcpy(newname, dev->name);
748 }
749 else if (__dev_get_by_name(newname))
750 return -EEXIST;
751 else
752 strlcpy(dev->name, newname, IFNAMSIZ);
753
92749821
EB		 754 device_rename(&dev->dev, dev->name);
755 hlist_del(&dev->name_hlist);
756 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
757 raw_notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
1da177e4
LT		 758
759 return err;
760}
761
d8a33ac4 762/**
3041a069 763 * netdev_features_change - device changes features
d8a33ac4
SH		 764 * @dev: device to cause notification
765 *
766 * Called to indicate a device has changed features.
767 */
768void netdev_features_change(struct net_device *dev)
769{
f07d5b94 770 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
d8a33ac4
SH		 771}
772EXPORT_SYMBOL(netdev_features_change);
773
1da177e4
LT		 774/**
775 * netdev_state_change - device changes state
776 * @dev: device to cause notification
777 *
778 * Called to indicate a device has changed state. This function calls
779 * the notifier chains for netdev_chain and sends a NEWLINK message
780 * to the routing socket.
781 */
782void netdev_state_change(struct net_device *dev)
783{
784 if (dev->flags & IFF_UP) {
f07d5b94 785 raw_notifier_call_chain(&netdev_chain,
e041c683 786 NETDEV_CHANGE, dev);
1da177e4
LT		 787 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
788 }
789}
790
791/**
792 * dev_load - load a network module
793 * @name: name of interface
794 *
795 * If a network interface is not present and the process has suitable
796 * privileges this function loads the module. If module loading is not
797 * available in this kernel then it becomes a nop.
798 */
799
800void dev_load(const char *name)
801{
4ec93edb 802 struct net_device *dev;
1da177e4
LT		 803
804 read_lock(&dev_base_lock);
805 dev = __dev_get_by_name(name);
806 read_unlock(&dev_base_lock);
807
808 if (!dev && capable(CAP_SYS_MODULE))
809 request_module("%s", name);
810}
811
812static int default_rebuild_header(struct sk_buff *skb)
813{
814 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
815 skb->dev ? skb->dev->name : "NULL!!!");
816 kfree_skb(skb);
817 return 1;
818}
819
820
821/**
822 * dev_open - prepare an interface for use.
823 * @dev: device to open
824 *
825 * Takes a device from down to up state. The device's private open
826 * function is invoked and then the multicast lists are loaded. Finally
827 * the device is moved into the up state and a %NETDEV_UP message is
828 * sent to the netdev notifier chain.
829 *
830 * Calling this function on an active interface is a nop. On a failure
831 * a negative errno code is returned.
832 */
833int dev_open(struct net_device *dev)
834{
835 int ret = 0;
836
837 /*
838 * Is it already up?
839 */
840
841 if (dev->flags & IFF_UP)
842 return 0;
843
844 /*
845 * Is it even present?
846 */
847 if (!netif_device_present(dev))
848 return -ENODEV;
849
850 /*
851 * Call device private open method
852 */
853 set_bit(__LINK_STATE_START, &dev->state);
854 if (dev->open) {
855 ret = dev->open(dev);
856 if (ret)
857 clear_bit(__LINK_STATE_START, &dev->state);
858 }
859
4ec93edb 860 /*
1da177e4
LT		 861 * If it went open OK then:
862 */
863
864 if (!ret) {
865 /*
866 * Set the flags.
867 */
868 dev->flags |= IFF_UP;
869
870 /*
871 * Initialize multicasting status
872 */
873 dev_mc_upload(dev);
874
875 /*
876 * Wakeup transmit queue engine
877 */
878 dev_activate(dev);
879
880 /*
881 * ... and announce new interface.
882 */
f07d5b94 883 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
1da177e4
LT		 884 }
885 return ret;
886}
887
888/**
889 * dev_close - shutdown an interface.
890 * @dev: device to shutdown
891 *
892 * This function moves an active device into down state. A
893 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
894 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
895 * chain.
896 */
897int dev_close(struct net_device *dev)
898{
899 if (!(dev->flags & IFF_UP))
900 return 0;
901
902 /*
903 * Tell people we are going down, so that they can
904 * prepare to death, when device is still operating.
905 */
f07d5b94 906 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
1da177e4
LT		 907
908 dev_deactivate(dev);
909
910 clear_bit(__LINK_STATE_START, &dev->state);
911
912 /* Synchronize to scheduled poll. We cannot touch poll list,
913 * it can be even on different cpu. So just clear netif_running(),
914 * and wait when poll really will happen. Actually, the best place
915 * for this is inside dev->stop() after device stopped its irq
916 * engine, but this requires more changes in devices. */
917
918 smp_mb__after_clear_bit(); /* Commit netif_running(). */
919 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
920 /* No hurry. */
6192b54b 921 msleep(1);
1da177e4
LT		 922 }
923
924 /*
925 * Call the device specific close. This cannot fail.
926 * Only if device is UP
927 *
928 * We allow it to be called even after a DETACH hot-plug
929 * event.
930 */
931 if (dev->stop)
932 dev->stop(dev);
933
934 /*
935 * Device is now down.
936 */
937
938 dev->flags &= ~IFF_UP;
939
940 /*
941 * Tell people we are down
942 */
f07d5b94 943 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1da177e4
LT		 944
945 return 0;
946}
947
948
949/*
950 * Device change register/unregister. These are not inline or static
951 * as we export them to the world.
952 */
953
954/**
955 * register_netdevice_notifier - register a network notifier block
956 * @nb: notifier
957 *
958 * Register a notifier to be called when network device events occur.
959 * The notifier passed is linked into the kernel structures and must
960 * not be reused until it has been unregistered. A negative errno code
961 * is returned on a failure.
962 *
963 * When registered all registration and up events are replayed
4ec93edb 964 * to the new notifier to allow device to have a race free
1da177e4
LT		 965 * view of the network device list.
966 */
967
968int register_netdevice_notifier(struct notifier_block *nb)
969{
970 struct net_device *dev;
971 int err;
972
973 rtnl_lock();
f07d5b94 974 err = raw_notifier_chain_register(&netdev_chain, nb);
1da177e4
LT		 975 if (!err) {
976 for (dev = dev_base; dev; dev = dev->next) {
977 nb->notifier_call(nb, NETDEV_REGISTER, dev);
978
4ec93edb 979 if (dev->flags & IFF_UP)
1da177e4
LT		 980 nb->notifier_call(nb, NETDEV_UP, dev);
981 }
982 }
983 rtnl_unlock();
984 return err;
985}
986
987/**
988 * unregister_netdevice_notifier - unregister a network notifier block
989 * @nb: notifier
990 *
991 * Unregister a notifier previously registered by
992 * register_netdevice_notifier(). The notifier is unlinked into the
993 * kernel structures and may then be reused. A negative errno code
994 * is returned on a failure.
995 */
996
997int unregister_netdevice_notifier(struct notifier_block *nb)
998{
9f514950
HX		 999 int err;
1000
1001 rtnl_lock();
f07d5b94 1002 err = raw_notifier_chain_unregister(&netdev_chain, nb);
9f514950
HX		 1003 rtnl_unlock();
1004 return err;
1da177e4
LT		 1005}
1006
1007/**
1008 * call_netdevice_notifiers - call all network notifier blocks
1009 * @val: value passed unmodified to notifier function
1010 * @v: pointer passed unmodified to notifier function
1011 *
1012 * Call all network notifier blocks. Parameters and return value
f07d5b94 1013 * are as for raw_notifier_call_chain().
1da177e4
LT		 1014 */
1015
1016int call_netdevice_notifiers(unsigned long val, void *v)
1017{
f07d5b94 1018 return raw_notifier_call_chain(&netdev_chain, val, v);
1da177e4
LT		 1019}
1020
1021/* When > 0 there are consumers of rx skb time stamps */
1022static atomic_t netstamp_needed = ATOMIC_INIT(0);
1023
1024void net_enable_timestamp(void)
1025{
1026 atomic_inc(&netstamp_needed);
1027}
1028
1029void net_disable_timestamp(void)
1030{
1031 atomic_dec(&netstamp_needed);
1032}
1033
a61bbcf2 1034static inline void net_timestamp(struct sk_buff *skb)
1da177e4
LT		 1035{
1036 if (atomic_read(&netstamp_needed))
a61bbcf2 1037 __net_timestamp(skb);
b7aa0bf7
ED		 1038 else
1039 skb->tstamp.tv64 = 0;
1da177e4
LT		 1040}
1041
1042/*
1043 * Support routine. Sends outgoing frames to any network
1044 * taps currently in use.
1045 */
1046
f6a78bfc 1047static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1da177e4
LT		 1048{
1049 struct packet_type *ptype;
a61bbcf2
PM		 1050
1051 net_timestamp(skb);
1da177e4
LT		 1052
1053 rcu_read_lock();
1054 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1055 /* Never send packets back to the socket
1056 * they originated from - MvS (miquels@drinkel.ow.org)
1057 */
1058 if ((ptype->dev == dev || !ptype->dev) &&
1059 (ptype->af_packet_priv == NULL ||
1060 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1061 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1062 if (!skb2)
1063 break;
1064
1065 /* skb->nh should be correctly
1066 set by sender, so that the second statement is
1067 just protection against buggy protocols.
1068 */
459a98ed 1069 skb_reset_mac_header(skb2);
1da177e4 1070
d56f90a7
ACM		 1071 if (skb_network_header(skb2) < skb2->data ||
1072 skb_network_header(skb2) > skb2->tail) {
1da177e4
LT		 1073 if (net_ratelimit())
1074 printk(KERN_CRIT "protocol %04x is "
1075 "buggy, dev %s\n",
1076 skb2->protocol, dev->name);
c1d2bbe1 1077 skb_reset_network_header(skb2);
1da177e4
LT		 1078 }
1079
1080 skb2->h.raw = skb2->nh.raw;
1081 skb2->pkt_type = PACKET_OUTGOING;
f2ccd8fa 1082 ptype->func(skb2, skb->dev, ptype, skb->dev);
1da177e4
LT		 1083 }
1084 }
1085 rcu_read_unlock();
1086}
1087
56079431
DV		 1088
1089void __netif_schedule(struct net_device *dev)
1090{
1091 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1092 unsigned long flags;
1093 struct softnet_data *sd;
1094
1095 local_irq_save(flags);
1096 sd = &__get_cpu_var(softnet_data);
1097 dev->next_sched = sd->output_queue;
1098 sd->output_queue = dev;
1099 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1100 local_irq_restore(flags);
1101 }
1102}
1103EXPORT_SYMBOL(__netif_schedule);
1104
1105void __netif_rx_schedule(struct net_device *dev)
1106{
1107 unsigned long flags;
1108
1109 local_irq_save(flags);
1110 dev_hold(dev);
1111 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1112 if (dev->quota < 0)
1113 dev->quota += dev->weight;
1114 else
1115 dev->quota = dev->weight;
1116 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1117 local_irq_restore(flags);
1118}
1119EXPORT_SYMBOL(__netif_rx_schedule);
1120
1121void dev_kfree_skb_any(struct sk_buff *skb)
1122{
1123 if (in_irq() || irqs_disabled())
1124 dev_kfree_skb_irq(skb);
1125 else
1126 dev_kfree_skb(skb);
1127}
1128EXPORT_SYMBOL(dev_kfree_skb_any);
1129
1130
1131/* Hot-plugging. */
1132void netif_device_detach(struct net_device *dev)
1133{
1134 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1135 netif_running(dev)) {
1136 netif_stop_queue(dev);
1137 }
1138}
1139EXPORT_SYMBOL(netif_device_detach);
1140
1141void netif_device_attach(struct net_device *dev)
1142{
1143 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1144 netif_running(dev)) {
1145 netif_wake_queue(dev);
4ec93edb 1146 __netdev_watchdog_up(dev);
56079431
DV		 1147 }
1148}
1149EXPORT_SYMBOL(netif_device_attach);
1150
1151
1da177e4
LT		 1152/*
1153 * Invalidate hardware checksum when packet is to be mangled, and
1154 * complete checksum manually on outgoing path.
1155 */
84fa7933 1156int skb_checksum_help(struct sk_buff *skb)
1da177e4 1157{
d3bc23e7 1158 __wsum csum;
1da177e4
LT		 1159 int ret = 0, offset = skb->h.raw - skb->data;
1160
84fa7933 1161 if (skb->ip_summed == CHECKSUM_COMPLETE)
a430a43d
HX		 1162 goto out_set_summed;
1163
1164 if (unlikely(skb_shinfo(skb)->gso_size)) {
a430a43d
HX		 1165 /* Let GSO fix up the checksum. */
1166 goto out_set_summed;
1da177e4
LT		 1167 }
1168
1169 if (skb_cloned(skb)) {
1170 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1171 if (ret)
1172 goto out;
1173 }
1174
09a62660 1175 BUG_ON(offset > (int)skb->len);
1da177e4
LT		 1176 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1177
1178 offset = skb->tail - skb->h.raw;
09a62660 1179 BUG_ON(offset <= 0);
ff1dcadb 1180 BUG_ON(skb->csum_offset + 2 > offset);
1da177e4 1181
ff1dcadb 1182 *(__sum16*)(skb->h.raw + skb->csum_offset) = csum_fold(csum);
a430a43d
HX		 1183
1184out_set_summed:
1da177e4 1185 skb->ip_summed = CHECKSUM_NONE;
4ec93edb 1186out:
1da177e4
LT		 1187 return ret;
1188}
1189
f6a78bfc
HX		 1190/**
1191 * skb_gso_segment - Perform segmentation on skb.
1192 * @skb: buffer to segment
576a30eb 1193 * @features: features for the output path (see dev->features)
f6a78bfc
HX		 1194 *
1195 * This function segments the given skb and returns a list of segments.
576a30eb
HX		 1196 *
1197 * It may return NULL if the skb requires no segmentation. This is
1198 * only possible when GSO is used for verifying header integrity.
f6a78bfc 1199 */
576a30eb 1200struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
f6a78bfc
HX		 1201{
1202 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1203 struct packet_type *ptype;
252e3346 1204 __be16 type = skb->protocol;
a430a43d 1205 int err;
f6a78bfc
HX		 1206
1207 BUG_ON(skb_shinfo(skb)->frag_list);
f6a78bfc 1208
459a98ed 1209 skb_reset_mac_header(skb);
d56f90a7 1210 skb->mac_len = skb->nh.raw - skb->mac.raw;
f6a78bfc
HX		 1211 __skb_pull(skb, skb->mac_len);
1212
84fa7933 1213 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
a430a43d
HX		 1214 if (skb_header_cloned(skb) &&
1215 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1216 return ERR_PTR(err);
1217 }
1218
f6a78bfc
HX		 1219 rcu_read_lock();
1220 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1221 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
84fa7933 1222 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
a430a43d
HX		 1223 err = ptype->gso_send_check(skb);
1224 segs = ERR_PTR(err);
1225 if (err || skb_gso_ok(skb, features))
1226 break;
d56f90a7
ACM		 1227 __skb_push(skb, (skb->data -
1228 skb_network_header(skb)));
a430a43d 1229 }
576a30eb 1230 segs = ptype->gso_segment(skb, features);
f6a78bfc
HX		 1231 break;
1232 }
1233 }
1234 rcu_read_unlock();
1235
98e399f8 1236 __skb_push(skb, skb->data - skb_mac_header(skb));
576a30eb 1237
f6a78bfc
HX		 1238 return segs;
1239}
1240
1241EXPORT_SYMBOL(skb_gso_segment);
1242
fb286bb2
HX		 1243/* Take action when hardware reception checksum errors are detected. */
1244#ifdef CONFIG_BUG
1245void netdev_rx_csum_fault(struct net_device *dev)
1246{
1247 if (net_ratelimit()) {
4ec93edb 1248 printk(KERN_ERR "%s: hw csum failure.\n",
246a4212 1249 dev ? dev->name : "<unknown>");
fb286bb2
HX		 1250 dump_stack();
1251 }
1252}
1253EXPORT_SYMBOL(netdev_rx_csum_fault);
1254#endif
1255
1da177e4
LT		 1256/* Actually, we should eliminate this check as soon as we know, that:
1257 * 1. IOMMU is present and allows to map all the memory.
1258 * 2. No high memory really exists on this machine.
1259 */
1260
1261static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1262{
3d3a8533 1263#ifdef CONFIG_HIGHMEM
1da177e4
LT		 1264 int i;
1265
1266 if (dev->features & NETIF_F_HIGHDMA)
1267 return 0;
1268
1269 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1270 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1271 return 1;
1272
3d3a8533 1273#endif
1da177e4
LT		 1274 return 0;
1275}
1da177e4 1276
f6a78bfc
HX		 1277struct dev_gso_cb {
1278 void (*destructor)(struct sk_buff *skb);
1279};
1280
1281#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1282
1283static void dev_gso_skb_destructor(struct sk_buff *skb)
1284{
1285 struct dev_gso_cb *cb;
1286
1287 do {
1288 struct sk_buff *nskb = skb->next;
1289
1290 skb->next = nskb->next;
1291 nskb->next = NULL;
1292 kfree_skb(nskb);
1293 } while (skb->next);
1294
1295 cb = DEV_GSO_CB(skb);
1296 if (cb->destructor)
1297 cb->destructor(skb);
1298}
1299
1300/**
1301 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1302 * @skb: buffer to segment
1303 *
1304 * This function segments the given skb and stores the list of segments
1305 * in skb->next.
1306 */
1307static int dev_gso_segment(struct sk_buff *skb)
1308{
1309 struct net_device *dev = skb->dev;
1310 struct sk_buff *segs;
576a30eb
HX		 1311 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1312 NETIF_F_SG : 0);
1313
1314 segs = skb_gso_segment(skb, features);
1315
1316 /* Verifying header integrity only. */
1317 if (!segs)
1318 return 0;
f6a78bfc 1319
f6a78bfc
HX		 1320 if (unlikely(IS_ERR(segs)))
1321 return PTR_ERR(segs);
1322
1323 skb->next = segs;
1324 DEV_GSO_CB(skb)->destructor = skb->destructor;
1325 skb->destructor = dev_gso_skb_destructor;
1326
1327 return 0;
1328}
1329
1330int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1331{
1332 if (likely(!skb->next)) {
1333 if (netdev_nit)
1334 dev_queue_xmit_nit(skb, dev);
1335
576a30eb
HX		 1336 if (netif_needs_gso(dev, skb)) {
1337 if (unlikely(dev_gso_segment(skb)))
1338 goto out_kfree_skb;
1339 if (skb->next)
1340 goto gso;
1341 }
f6a78bfc 1342
576a30eb 1343 return dev->hard_start_xmit(skb, dev);
f6a78bfc
HX		 1344 }
1345
576a30eb 1346gso:
f6a78bfc
HX		 1347 do {
1348 struct sk_buff *nskb = skb->next;
1349 int rc;
1350
1351 skb->next = nskb->next;
1352 nskb->next = NULL;
1353 rc = dev->hard_start_xmit(nskb, dev);
1354 if (unlikely(rc)) {
f54d9e8d 1355 nskb->next = skb->next;
f6a78bfc
HX		 1356 skb->next = nskb;
1357 return rc;
1358 }
f54d9e8d
MC		 1359 if (unlikely(netif_queue_stopped(dev) && skb->next))
1360 return NETDEV_TX_BUSY;
f6a78bfc 1361 } while (skb->next);
4ec93edb 1362
f6a78bfc
HX		 1363 skb->destructor = DEV_GSO_CB(skb)->destructor;
1364
1365out_kfree_skb:
1366 kfree_skb(skb);
1367 return 0;
1368}
1369
1da177e4
LT		 1370#define HARD_TX_LOCK(dev, cpu) { \
1371 if ((dev->features & NETIF_F_LLTX) == 0) { \
932ff279 1372 netif_tx_lock(dev); \
1da177e4
LT		 1373 } \
1374}
1375
1376#define HARD_TX_UNLOCK(dev) { \
1377 if ((dev->features & NETIF_F_LLTX) == 0) { \
932ff279 1378 netif_tx_unlock(dev); \
1da177e4
LT		 1379 } \
1380}
1381
1382/**
1383 * dev_queue_xmit - transmit a buffer
1384 * @skb: buffer to transmit
1385 *
1386 * Queue a buffer for transmission to a network device. The caller must
1387 * have set the device and priority and built the buffer before calling
1388 * this function. The function can be called from an interrupt.
1389 *
1390 * A negative errno code is returned on a failure. A success does not
1391 * guarantee the frame will be transmitted as it may be dropped due
1392 * to congestion or traffic shaping.
af191367
BG		 1393 *
1394 * -----------------------------------------------------------------------------------
1395 * I notice this method can also return errors from the queue disciplines,
1396 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1397 * be positive.
1398 *
1399 * Regardless of the return value, the skb is consumed, so it is currently
1400 * difficult to retry a send to this method. (You can bump the ref count
1401 * before sending to hold a reference for retry if you are careful.)
1402 *
1403 * When calling this method, interrupts MUST be enabled. This is because
1404 * the BH enable code must have IRQs enabled so that it will not deadlock.
1405 * --BLG
1da177e4
LT		 1406 */
1407
1408int dev_queue_xmit(struct sk_buff *skb)
1409{
1410 struct net_device *dev = skb->dev;
1411 struct Qdisc *q;
1412 int rc = -ENOMEM;
1413
f6a78bfc
HX		 1414 /* GSO will handle the following emulations directly. */
1415 if (netif_needs_gso(dev, skb))
1416 goto gso;
1417
1da177e4
LT		 1418 if (skb_shinfo(skb)->frag_list &&
1419 !(dev->features & NETIF_F_FRAGLIST) &&
364c6bad 1420 __skb_linearize(skb))
1da177e4
LT		 1421 goto out_kfree_skb;
1422
1423 /* Fragmented skb is linearized if device does not support SG,
1424 * or if at least one of fragments is in highmem and device
1425 * does not support DMA from it.
1426 */
1427 if (skb_shinfo(skb)->nr_frags &&
1428 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
364c6bad 1429 __skb_linearize(skb))
1da177e4
LT		 1430 goto out_kfree_skb;
1431
1432 /* If packet is not checksummed and device does not support
1433 * checksumming for this protocol, complete checksumming here.
1434 */
84fa7933 1435 if (skb->ip_summed == CHECKSUM_PARTIAL &&
8648b305 1436 (!(dev->features & NETIF_F_GEN_CSUM) &&
1da177e4
LT		 1437 (!(dev->features & NETIF_F_IP_CSUM) ||
1438 skb->protocol != htons(ETH_P_IP))))
4ec93edb
YH		 1439 if (skb_checksum_help(skb))
1440 goto out_kfree_skb;
1da177e4 1441
f6a78bfc 1442gso:
2d7ceece
ED		 1443 spin_lock_prefetch(&dev->queue_lock);
1444
4ec93edb
YH		 1445 /* Disable soft irqs for various locks below. Also
1446 * stops preemption for RCU.
1da177e4 1447 */
4ec93edb 1448 rcu_read_lock_bh();
1da177e4 1449
4ec93edb
YH		 1450 /* Updates of qdisc are serialized by queue_lock.
1451 * The struct Qdisc which is pointed to by qdisc is now a
1452 * rcu structure - it may be accessed without acquiring
1da177e4 1453 * a lock (but the structure may be stale.) The freeing of the
4ec93edb 1454 * qdisc will be deferred until it's known that there are no
1da177e4 1455 * more references to it.
4ec93edb
YH		 1456 *
1457 * If the qdisc has an enqueue function, we still need to
1da177e4
LT		 1458 * hold the queue_lock before calling it, since queue_lock
1459 * also serializes access to the device queue.
1460 */
1461
1462 q = rcu_dereference(dev->qdisc);
1463#ifdef CONFIG_NET_CLS_ACT
1464 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1465#endif
1466 if (q->enqueue) {
1467 /* Grab device queue */
1468 spin_lock(&dev->queue_lock);
85670cc1
PM		 1469 q = dev->qdisc;
1470 if (q->enqueue) {
1471 rc = q->enqueue(skb, q);
1472 qdisc_run(dev);
1473 spin_unlock(&dev->queue_lock);
1da177e4 1474
85670cc1
PM		 1475 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1476 goto out;
1477 }
1da177e4 1478 spin_unlock(&dev->queue_lock);
1da177e4
LT		 1479 }
1480
1481 /* The device has no queue. Common case for software devices:
1482 loopback, all the sorts of tunnels...
1483
932ff279
HX		 1484 Really, it is unlikely that netif_tx_lock protection is necessary
1485 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1da177e4
LT		 1486 counters.)
1487 However, it is possible, that they rely on protection
1488 made by us here.
1489
1490 Check this and shot the lock. It is not prone from deadlocks.
1491 Either shot noqueue qdisc, it is even simpler 8)
1492 */
1493 if (dev->flags & IFF_UP) {
1494 int cpu = smp_processor_id(); /* ok because BHs are off */
1495
1496 if (dev->xmit_lock_owner != cpu) {
1497
1498 HARD_TX_LOCK(dev, cpu);
1499
1500 if (!netif_queue_stopped(dev)) {
1da177e4 1501 rc = 0;
f6a78bfc 1502 if (!dev_hard_start_xmit(skb, dev)) {
1da177e4
LT		 1503 HARD_TX_UNLOCK(dev);
1504 goto out;
1505 }
1506 }
1507 HARD_TX_UNLOCK(dev);
1508 if (net_ratelimit())
1509 printk(KERN_CRIT "Virtual device %s asks to "
1510 "queue packet!\n", dev->name);
1511 } else {
1512 /* Recursion is detected! It is possible,
1513 * unfortunately */
1514 if (net_ratelimit())
1515 printk(KERN_CRIT "Dead loop on virtual device "
1516 "%s, fix it urgently!\n", dev->name);
1517 }
1518 }
1519
1520 rc = -ENETDOWN;
d4828d85 1521 rcu_read_unlock_bh();
1da177e4
LT		 1522
1523out_kfree_skb:
1524 kfree_skb(skb);
1525 return rc;
1526out:
d4828d85 1527 rcu_read_unlock_bh();
1da177e4
LT		 1528 return rc;
1529}
1530
1531
1532/*=======================================================================
1533 Receiver routines
1534 =======================================================================*/
1535
6b2bedc3
SH		 1536int netdev_max_backlog __read_mostly = 1000;
1537int netdev_budget __read_mostly = 300;
1538int weight_p __read_mostly = 64; /* old backlog weight */
1da177e4
LT		 1539
1540DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1541
1542
1da177e4
LT		 1543/**
1544 * netif_rx - post buffer to the network code
1545 * @skb: buffer to post
1546 *
1547 * This function receives a packet from a device driver and queues it for
1548 * the upper (protocol) levels to process. It always succeeds. The buffer
1549 * may be dropped during processing for congestion control or by the
1550 * protocol layers.
1551 *
1552 * return values:
1553 * NET_RX_SUCCESS (no congestion)
1554 * NET_RX_CN_LOW (low congestion)
1555 * NET_RX_CN_MOD (moderate congestion)
1556 * NET_RX_CN_HIGH (high congestion)
1557 * NET_RX_DROP (packet was dropped)
1558 *
1559 */
1560
1561int netif_rx(struct sk_buff *skb)
1562{
1da177e4
LT		 1563 struct softnet_data *queue;
1564 unsigned long flags;
1565
1566 /* if netpoll wants it, pretend we never saw it */
1567 if (netpoll_rx(skb))
1568 return NET_RX_DROP;
1569
b7aa0bf7 1570 if (!skb->tstamp.tv64)
a61bbcf2 1571 net_timestamp(skb);
1da177e4
LT		 1572
1573 /*
1574 * The code is rearranged so that the path is the most
1575 * short when CPU is congested, but is still operating.
1576 */
1577 local_irq_save(flags);
1da177e4
LT		 1578 queue = &__get_cpu_var(softnet_data);
1579
1580 __get_cpu_var(netdev_rx_stat).total++;
1581 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1582 if (queue->input_pkt_queue.qlen) {
1da177e4
LT		 1583enqueue:
1584 dev_hold(skb->dev);
1585 __skb_queue_tail(&queue->input_pkt_queue, skb);
1da177e4 1586 local_irq_restore(flags);
34008d8c 1587 return NET_RX_SUCCESS;
1da177e4
LT		 1588 }
1589
1da177e4
LT		 1590 netif_rx_schedule(&queue->backlog_dev);
1591 goto enqueue;
1592 }
1593
1da177e4
LT		 1594 __get_cpu_var(netdev_rx_stat).dropped++;
1595 local_irq_restore(flags);
1596
1597 kfree_skb(skb);
1598 return NET_RX_DROP;
1599}
1600
1601int netif_rx_ni(struct sk_buff *skb)
1602{
1603 int err;
1604
1605 preempt_disable();
1606 err = netif_rx(skb);
1607 if (local_softirq_pending())
1608 do_softirq();
1609 preempt_enable();
1610
1611 return err;
1612}
1613
1614EXPORT_SYMBOL(netif_rx_ni);
1615
f2ccd8fa 1616static inline struct net_device *skb_bond(struct sk_buff *skb)
1da177e4
LT		 1617{
1618 struct net_device *dev = skb->dev;
1619
8f903c70 1620 if (dev->master) {
7ea49ed7 1621 if (skb_bond_should_drop(skb)) {
8f903c70
JV		 1622 kfree_skb(skb);
1623 return NULL;
1624 }
1da177e4 1625 skb->dev = dev->master;
8f903c70 1626 }
f2ccd8fa
DM		 1627
1628 return dev;
1da177e4
LT		 1629}
1630
1631static void net_tx_action(struct softirq_action *h)
1632{
1633 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1634
1635 if (sd->completion_queue) {
1636 struct sk_buff *clist;
1637
1638 local_irq_disable();
1639 clist = sd->completion_queue;
1640 sd->completion_queue = NULL;
1641 local_irq_enable();
1642
1643 while (clist) {
1644 struct sk_buff *skb = clist;
1645 clist = clist->next;
1646
1647 BUG_TRAP(!atomic_read(&skb->users));
1648 __kfree_skb(skb);
1649 }
1650 }
1651
1652 if (sd->output_queue) {
1653 struct net_device *head;
1654
1655 local_irq_disable();
1656 head = sd->output_queue;
1657 sd->output_queue = NULL;
1658 local_irq_enable();
1659
1660 while (head) {
1661 struct net_device *dev = head;
1662 head = head->next_sched;
1663
1664 smp_mb__before_clear_bit();
1665 clear_bit(__LINK_STATE_SCHED, &dev->state);
1666
1667 if (spin_trylock(&dev->queue_lock)) {
1668 qdisc_run(dev);
1669 spin_unlock(&dev->queue_lock);
1670 } else {
1671 netif_schedule(dev);
1672 }
1673 }
1674 }
1675}
1676
6f05f629
SH		 1677static inline int deliver_skb(struct sk_buff *skb,
1678 struct packet_type *pt_prev,
1679 struct net_device *orig_dev)
1da177e4
LT		 1680{
1681 atomic_inc(&skb->users);
f2ccd8fa 1682 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT		 1683}
1684
1685#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1686int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1687struct net_bridge;
1688struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1689 unsigned char *addr);
1690void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1691
1692static __inline__ int handle_bridge(struct sk_buff **pskb,
f2ccd8fa
DM		 1693 struct packet_type **pt_prev, int *ret,
1694 struct net_device *orig_dev)
1da177e4
LT		 1695{
1696 struct net_bridge_port *port;
1697
1698 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1699 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1700 return 0;
1701
1702 if (*pt_prev) {
f2ccd8fa 1703 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1da177e4 1704 *pt_prev = NULL;
4ec93edb
YH		 1705 }
1706
1da177e4
LT		 1707 return br_handle_frame_hook(port, pskb);
1708}
1709#else
f2ccd8fa 1710#define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1da177e4
LT		 1711#endif
1712
1713#ifdef CONFIG_NET_CLS_ACT
1714/* TODO: Maybe we should just force sch_ingress to be compiled in
1715 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1716 * a compare and 2 stores extra right now if we dont have it on
1717 * but have CONFIG_NET_CLS_ACT
4ec93edb 1718 * NOTE: This doesnt stop any functionality; if you dont have
1da177e4
LT		 1719 * the ingress scheduler, you just cant add policies on ingress.
1720 *
1721 */
4ec93edb 1722static int ing_filter(struct sk_buff *skb)
1da177e4
LT		 1723{
1724 struct Qdisc *q;
1725 struct net_device *dev = skb->dev;
1726 int result = TC_ACT_OK;
4ec93edb 1727
1da177e4
LT		 1728 if (dev->qdisc_ingress) {
1729 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1730 if (MAX_RED_LOOP < ttl++) {
c01003c2
PM		 1731 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1732 skb->iif, skb->dev->ifindex);
1da177e4
LT		 1733 return TC_ACT_SHOT;
1734 }
1735
1736 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1737
1738 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
86e65da9 1739
035832a2 1740 spin_lock(&dev->queue_lock);
1da177e4
LT		 1741 if ((q = dev->qdisc_ingress) != NULL)
1742 result = q->enqueue(skb, q);
035832a2 1743 spin_unlock(&dev->queue_lock);
1da177e4
LT		 1744
1745 }
1746
1747 return result;
1748}
1749#endif
1750
1751int netif_receive_skb(struct sk_buff *skb)
1752{
1753 struct packet_type *ptype, *pt_prev;
f2ccd8fa 1754 struct net_device *orig_dev;
1da177e4 1755 int ret = NET_RX_DROP;
252e3346 1756 __be16 type;
1da177e4
LT		 1757
1758 /* if we've gotten here through NAPI, check netpoll */
1759 if (skb->dev->poll && netpoll_rx(skb))
1760 return NET_RX_DROP;
1761
b7aa0bf7 1762 if (!skb->tstamp.tv64)
a61bbcf2 1763 net_timestamp(skb);
1da177e4 1764
c01003c2
PM		 1765 if (!skb->iif)
1766 skb->iif = skb->dev->ifindex;
86e65da9 1767
f2ccd8fa 1768 orig_dev = skb_bond(skb);
1da177e4 1769
8f903c70
JV		 1770 if (!orig_dev)
1771 return NET_RX_DROP;
1772
1da177e4
LT		 1773 __get_cpu_var(netdev_rx_stat).total++;
1774
c1d2bbe1
ACM		 1775 skb_reset_network_header(skb);
1776 skb->h.raw = skb->data;
1da177e4
LT		 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) {
4ec93edb 1792 if (pt_prev)
f2ccd8fa 1793 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT		 1794 pt_prev = ptype;
1795 }
1796 }
1797
1798#ifdef CONFIG_NET_CLS_ACT
1799 if (pt_prev) {
f2ccd8fa 1800 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT		 1801 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
f2ccd8fa 1817 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1da177e4
LT		 1818 goto out;
1819
1820 type = skb->protocol;
1821 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1822 if (ptype->type == type &&
1823 (!ptype->dev || ptype->dev == skb->dev)) {
4ec93edb 1824 if (pt_prev)
f2ccd8fa 1825 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT		 1826 pt_prev = ptype;
1827 }
1828 }
1829
1830 if (pt_prev) {
f2ccd8fa 1831 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT		 1832 } else {
1833 kfree_skb(skb);
1834 /* Jamal, now you will not able to escape explaining
1835 * me how you were going to use this. :-)
1836 */
1837 ret = NET_RX_DROP;
1838 }
1839
1840out:
1841 rcu_read_unlock();
1842 return ret;
1843}
1844
1845static int process_backlog(struct net_device *backlog_dev, int *budget)
1846{
1847 int work = 0;
1848 int quota = min(backlog_dev->quota, *budget);
1849 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1850 unsigned long start_time = jiffies;
1851
e3876605 1852 backlog_dev->weight = weight_p;
1da177e4
LT		 1853 for (;;) {
1854 struct sk_buff *skb;
1855 struct net_device *dev;
1856
1857 local_irq_disable();
1858 skb = __skb_dequeue(&queue->input_pkt_queue);
1859 if (!skb)
1860 goto job_done;
1861 local_irq_enable();
1862
1863 dev = skb->dev;
1864
1865 netif_receive_skb(skb);
1866
1867 dev_put(dev);
1868
1869 work++;
1870
1871 if (work >= quota || jiffies - start_time > 1)
1872 break;
1873
1874 }
1875
1876 backlog_dev->quota -= work;
1877 *budget -= work;
1878 return -1;
1879
1880job_done:
1881 backlog_dev->quota -= work;
1882 *budget -= work;
1883
1884 list_del(&backlog_dev->poll_list);
1885 smp_mb__before_clear_bit();
1886 netif_poll_enable(backlog_dev);
1887
1da177e4
LT		 1888 local_irq_enable();
1889 return 0;
1890}
1891
1892static void net_rx_action(struct softirq_action *h)
1893{
1894 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1895 unsigned long start_time = jiffies;
51b0bded 1896 int budget = netdev_budget;
53fb95d3
MM		 1897 void *have;
1898
1da177e4
LT		 1899 local_irq_disable();
1900
1901 while (!list_empty(&queue->poll_list)) {
1902 struct net_device *dev;
1903
1904 if (budget <= 0 || jiffies - start_time > 1)
1905 goto softnet_break;
1906
1907 local_irq_enable();
1908
1909 dev = list_entry(queue->poll_list.next,
1910 struct net_device, poll_list);
53fb95d3 1911 have = netpoll_poll_lock(dev);
1da177e4
LT		 1912
1913 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
53fb95d3 1914 netpoll_poll_unlock(have);
1da177e4 1915 local_irq_disable();
8aca8a27 1916 list_move_tail(&dev->poll_list, &queue->poll_list);
1da177e4
LT		 1917 if (dev->quota < 0)
1918 dev->quota += dev->weight;
1919 else
1920 dev->quota = dev->weight;
1921 } else {
53fb95d3 1922 netpoll_poll_unlock(have);
1da177e4
LT		 1923 dev_put(dev);
1924 local_irq_disable();
1925 }
1926 }
1927out:
db217334
CL		 1928#ifdef CONFIG_NET_DMA
1929 /*
1930 * There may not be any more sk_buffs coming right now, so push
1931 * any pending DMA copies to hardware
1932 */
1933 if (net_dma_client) {
1934 struct dma_chan *chan;
1935 rcu_read_lock();
1936 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1937 dma_async_memcpy_issue_pending(chan);
1938 rcu_read_unlock();
1939 }
1940#endif
1da177e4
LT		 1941 local_irq_enable();
1942 return;
1943
1944softnet_break:
1945 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1946 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1947 goto out;
1948}
1949
1950static gifconf_func_t * gifconf_list [NPROTO];
1951
1952/**
1953 * register_gifconf - register a SIOCGIF handler
1954 * @family: Address family
1955 * @gifconf: Function handler
1956 *
1957 * Register protocol dependent address dumping routines. The handler
1958 * that is passed must not be freed or reused until it has been replaced
1959 * by another handler.
1960 */
1961int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1962{
1963 if (family >= NPROTO)
1964 return -EINVAL;
1965 gifconf_list[family] = gifconf;
1966 return 0;
1967}
1968
1969
1970/*
1971 * Map an interface index to its name (SIOCGIFNAME)
1972 */
1973
1974/*
1975 * We need this ioctl for efficient implementation of the
1976 * if_indextoname() function required by the IPv6 API. Without
1977 * it, we would have to search all the interfaces to find a
1978 * match. --pb
1979 */
1980
1981static int dev_ifname(struct ifreq __user *arg)
1982{
1983 struct net_device *dev;
1984 struct ifreq ifr;
1985
1986 /*
1987 * Fetch the caller's info block.
1988 */
1989
1990 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1991 return -EFAULT;
1992
1993 read_lock(&dev_base_lock);
1994 dev = __dev_get_by_index(ifr.ifr_ifindex);
1995 if (!dev) {
1996 read_unlock(&dev_base_lock);
1997 return -ENODEV;
1998 }
1999
2000 strcpy(ifr.ifr_name, dev->name);
2001 read_unlock(&dev_base_lock);
2002
2003 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2004 return -EFAULT;
2005 return 0;
2006}
2007
2008/*
2009 * Perform a SIOCGIFCONF call. This structure will change
2010 * size eventually, and there is nothing I can do about it.
2011 * Thus we will need a 'compatibility mode'.
2012 */
2013
2014static int dev_ifconf(char __user *arg)
2015{
2016 struct ifconf ifc;
2017 struct net_device *dev;
2018 char __user *pos;
2019 int len;
2020 int total;
2021 int i;
2022
2023 /*
2024 * Fetch the caller's info block.
2025 */
2026
2027 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2028 return -EFAULT;
2029
2030 pos = ifc.ifc_buf;
2031 len = ifc.ifc_len;
2032
2033 /*
2034 * Loop over the interfaces, and write an info block for each.
2035 */
2036
2037 total = 0;
2038 for (dev = dev_base; dev; dev = dev->next) {
2039 for (i = 0; i < NPROTO; i++) {
2040 if (gifconf_list[i]) {
2041 int done;
2042 if (!pos)
2043 done = gifconf_list[i](dev, NULL, 0);
2044 else
2045 done = gifconf_list[i](dev, pos + total,
2046 len - total);
2047 if (done < 0)
2048 return -EFAULT;
2049 total += done;
2050 }
2051 }
4ec93edb 2052 }
1da177e4
LT		 2053
2054 /*
2055 * All done. Write the updated control block back to the caller.
2056 */
2057 ifc.ifc_len = total;
2058
2059 /*
2060 * Both BSD and Solaris return 0 here, so we do too.
2061 */
2062 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2063}
2064
2065#ifdef CONFIG_PROC_FS
2066/*
2067 * This is invoked by the /proc filesystem handler to display a device
2068 * in detail.
2069 */
6f05f629 2070static struct net_device *dev_get_idx(loff_t pos)
1da177e4
LT		 2071{
2072 struct net_device *dev;
2073 loff_t i;
2074
2075 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2076
2077 return i == pos ? dev : NULL;
2078}
2079
2080void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2081{
2082 read_lock(&dev_base_lock);
2083 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2084}
2085
2086void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2087{
2088 ++*pos;
2089 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2090}
2091
2092void dev_seq_stop(struct seq_file *seq, void *v)
2093{
2094 read_unlock(&dev_base_lock);
2095}
2096
2097static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2098{
2099 if (dev->get_stats) {
2100 struct net_device_stats *stats = dev->get_stats(dev);
2101
2102 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2103 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2104 dev->name, stats->rx_bytes, stats->rx_packets,
2105 stats->rx_errors,
2106 stats->rx_dropped + stats->rx_missed_errors,
2107 stats->rx_fifo_errors,
2108 stats->rx_length_errors + stats->rx_over_errors +
2109 stats->rx_crc_errors + stats->rx_frame_errors,
2110 stats->rx_compressed, stats->multicast,
2111 stats->tx_bytes, stats->tx_packets,
2112 stats->tx_errors, stats->tx_dropped,
2113 stats->tx_fifo_errors, stats->collisions,
2114 stats->tx_carrier_errors +
2115 stats->tx_aborted_errors +
2116 stats->tx_window_errors +
2117 stats->tx_heartbeat_errors,
2118 stats->tx_compressed);
2119 } else
2120 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2121}
2122
2123/*
2124 * Called from the PROCfs module. This now uses the new arbitrary sized
2125 * /proc/net interface to create /proc/net/dev
2126 */
2127static int dev_seq_show(struct seq_file *seq, void *v)
2128{
2129 if (v == SEQ_START_TOKEN)
2130 seq_puts(seq, "Inter-| Receive "
2131 " | Transmit\n"
2132 " face |bytes packets errs drop fifo frame "
2133 "compressed multicast|bytes packets errs "
2134 "drop fifo colls carrier compressed\n");
2135 else
2136 dev_seq_printf_stats(seq, v);
2137 return 0;
2138}
2139
2140static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2141{
2142 struct netif_rx_stats *rc = NULL;
2143
2144 while (*pos < NR_CPUS)
4ec93edb 2145 if (cpu_online(*pos)) {
1da177e4
LT		 2146 rc = &per_cpu(netdev_rx_stat, *pos);
2147 break;
2148 } else
2149 ++*pos;
2150 return rc;
2151}
2152
2153static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2154{
2155 return softnet_get_online(pos);
2156}
2157
2158static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2159{
2160 ++*pos;
2161 return softnet_get_online(pos);
2162}
2163
2164static void softnet_seq_stop(struct seq_file *seq, void *v)
2165{
2166}
2167
2168static int softnet_seq_show(struct seq_file *seq, void *v)
2169{
2170 struct netif_rx_stats *s = v;
2171
2172 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
31aa02c5 2173 s->total, s->dropped, s->time_squeeze, 0,
c1ebcdb8
SH		 2174 0, 0, 0, 0, /* was fastroute */
2175 s->cpu_collision );
1da177e4
LT		 2176 return 0;
2177}
2178
2179static struct seq_operations dev_seq_ops = {
2180 .start = dev_seq_start,
2181 .next = dev_seq_next,
2182 .stop = dev_seq_stop,
2183 .show = dev_seq_show,
2184};
2185
2186static int dev_seq_open(struct inode *inode, struct file *file)
2187{
2188 return seq_open(file, &dev_seq_ops);
2189}
2190
9a32144e 2191static const struct file_operations dev_seq_fops = {
1da177e4
LT		 2192 .owner = THIS_MODULE,
2193 .open = dev_seq_open,
2194 .read = seq_read,
2195 .llseek = seq_lseek,
2196 .release = seq_release,
2197};
2198
2199static struct seq_operations softnet_seq_ops = {
2200 .start = softnet_seq_start,
2201 .next = softnet_seq_next,
2202 .stop = softnet_seq_stop,
2203 .show = softnet_seq_show,
2204};
2205
2206static int softnet_seq_open(struct inode *inode, struct file *file)
2207{
2208 return seq_open(file, &softnet_seq_ops);
2209}
2210
9a32144e 2211static const struct file_operations softnet_seq_fops = {
1da177e4
LT		 2212 .owner = THIS_MODULE,
2213 .open = softnet_seq_open,
2214 .read = seq_read,
2215 .llseek = seq_lseek,
2216 .release = seq_release,
2217};
2218
d86b5e0e 2219#ifdef CONFIG_WIRELESS_EXT
1da177e4
LT		 2220extern int wireless_proc_init(void);
2221#else
2222#define wireless_proc_init() 0
2223#endif
2224
2225static int __init dev_proc_init(void)
2226{
2227 int rc = -ENOMEM;
2228
2229 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2230 goto out;
2231 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2232 goto out_dev;
2233 if (wireless_proc_init())
2234 goto out_softnet;
2235 rc = 0;
2236out:
2237 return rc;
2238out_softnet:
2239 proc_net_remove("softnet_stat");
2240out_dev:
2241 proc_net_remove("dev");
2242 goto out;
2243}
2244#else
2245#define dev_proc_init() 0
2246#endif /* CONFIG_PROC_FS */
2247
2248
2249/**
2250 * netdev_set_master - set up master/slave pair
2251 * @slave: slave device
2252 * @master: new master device
2253 *
2254 * Changes the master device of the slave. Pass %NULL to break the
2255 * bonding. The caller must hold the RTNL semaphore. On a failure
2256 * a negative errno code is returned. On success the reference counts
2257 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2258 * function returns zero.
2259 */
2260int netdev_set_master(struct net_device *slave, struct net_device *master)
2261{
2262 struct net_device *old = slave->master;
2263
2264 ASSERT_RTNL();
2265
2266 if (master) {
2267 if (old)
2268 return -EBUSY;
2269 dev_hold(master);
2270 }
2271
2272 slave->master = master;
4ec93edb 2273
1da177e4
LT		 2274 synchronize_net();
2275
2276 if (old)
2277 dev_put(old);
2278
2279 if (master)
2280 slave->flags |= IFF_SLAVE;
2281 else
2282 slave->flags &= ~IFF_SLAVE;
2283
2284 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2285 return 0;
2286}
2287
2288/**
2289 * dev_set_promiscuity - update promiscuity count on a device
2290 * @dev: device
2291 * @inc: modifier
2292 *
3041a069 2293 * Add or remove promiscuity from a device. While the count in the device
1da177e4
LT		 2294 * remains above zero the interface remains promiscuous. Once it hits zero
2295 * the device reverts back to normal filtering operation. A negative inc
2296 * value is used to drop promiscuity on the device.
2297 */
2298void dev_set_promiscuity(struct net_device *dev, int inc)
2299{
2300 unsigned short old_flags = dev->flags;
2301
1da177e4
LT		 2302 if ((dev->promiscuity += inc) == 0)
2303 dev->flags &= ~IFF_PROMISC;
52609c0b
DC		 2304 else
2305 dev->flags |= IFF_PROMISC;
2306 if (dev->flags != old_flags) {
1da177e4
LT		 2307 dev_mc_upload(dev);
2308 printk(KERN_INFO "device %s %s promiscuous mode\n",
2309 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4ec93edb 2310 "left");
5bdb9886
SG		 2311 audit_log(current->audit_context, GFP_ATOMIC,
2312 AUDIT_ANOM_PROMISCUOUS,
2313 "dev=%s prom=%d old_prom=%d auid=%u",
2314 dev->name, (dev->flags & IFF_PROMISC),
2315 (old_flags & IFF_PROMISC),
4ec93edb 2316 audit_get_loginuid(current->audit_context));
1da177e4
LT		 2317 }
2318}
2319
2320/**
2321 * dev_set_allmulti - update allmulti count on a device
2322 * @dev: device
2323 * @inc: modifier
2324 *
2325 * Add or remove reception of all multicast frames to a device. While the
2326 * count in the device remains above zero the interface remains listening
2327 * to all interfaces. Once it hits zero the device reverts back to normal
2328 * filtering operation. A negative @inc value is used to drop the counter
2329 * when releasing a resource needing all multicasts.
2330 */
2331
2332void dev_set_allmulti(struct net_device *dev, int inc)
2333{
2334 unsigned short old_flags = dev->flags;
2335
2336 dev->flags |= IFF_ALLMULTI;
2337 if ((dev->allmulti += inc) == 0)
2338 dev->flags &= ~IFF_ALLMULTI;
2339 if (dev->flags ^ old_flags)
2340 dev_mc_upload(dev);
2341}
2342
2343unsigned dev_get_flags(const struct net_device *dev)
2344{
2345 unsigned flags;
2346
2347 flags = (dev->flags & ~(IFF_PROMISC |
2348 IFF_ALLMULTI |
b00055aa
SR		 2349 IFF_RUNNING |
2350 IFF_LOWER_UP |
2351 IFF_DORMANT)) |
1da177e4
LT		 2352 (dev->gflags & (IFF_PROMISC |
2353 IFF_ALLMULTI));
2354
b00055aa
SR		 2355 if (netif_running(dev)) {
2356 if (netif_oper_up(dev))
2357 flags |= IFF_RUNNING;
2358 if (netif_carrier_ok(dev))
2359 flags |= IFF_LOWER_UP;
2360 if (netif_dormant(dev))
2361 flags |= IFF_DORMANT;
2362 }
1da177e4
LT		 2363
2364 return flags;
2365}
2366
2367int dev_change_flags(struct net_device *dev, unsigned flags)
2368{
2369 int ret;
2370 int old_flags = dev->flags;
2371
2372 /*
2373 * Set the flags on our device.
2374 */
2375
2376 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2377 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2378 IFF_AUTOMEDIA)) |
2379 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2380 IFF_ALLMULTI));
2381
2382 /*
2383 * Load in the correct multicast list now the flags have changed.
2384 */
2385
2386 dev_mc_upload(dev);
2387
2388 /*
2389 * Have we downed the interface. We handle IFF_UP ourselves
2390 * according to user attempts to set it, rather than blindly
2391 * setting it.
2392 */
2393
2394 ret = 0;
2395 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2396 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2397
2398 if (!ret)
2399 dev_mc_upload(dev);
2400 }
2401
2402 if (dev->flags & IFF_UP &&
2403 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2404 IFF_VOLATILE)))
f07d5b94 2405 raw_notifier_call_chain(&netdev_chain,
e041c683 2406 NETDEV_CHANGE, dev);
1da177e4
LT		 2407
2408 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2409 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2410 dev->gflags ^= IFF_PROMISC;
2411 dev_set_promiscuity(dev, inc);
2412 }
2413
2414 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2415 is important. Some (broken) drivers set IFF_PROMISC, when
2416 IFF_ALLMULTI is requested not asking us and not reporting.
2417 */
2418 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2419 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2420 dev->gflags ^= IFF_ALLMULTI;
2421 dev_set_allmulti(dev, inc);
2422 }
2423
2424 if (old_flags ^ dev->flags)
2425 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2426
2427 return ret;
2428}
2429
2430int dev_set_mtu(struct net_device *dev, int new_mtu)
2431{
2432 int err;
2433
2434 if (new_mtu == dev->mtu)
2435 return 0;
2436
2437 /* MTU must be positive. */
2438 if (new_mtu < 0)
2439 return -EINVAL;
2440
2441 if (!netif_device_present(dev))
2442 return -ENODEV;
2443
2444 err = 0;
2445 if (dev->change_mtu)
2446 err = dev->change_mtu(dev, new_mtu);
2447 else
2448 dev->mtu = new_mtu;
2449 if (!err && dev->flags & IFF_UP)
f07d5b94 2450 raw_notifier_call_chain(&netdev_chain,
e041c683 2451 NETDEV_CHANGEMTU, dev);
1da177e4
LT		 2452 return err;
2453}
2454
2455int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2456{
2457 int err;
2458
2459 if (!dev->set_mac_address)
2460 return -EOPNOTSUPP;
2461 if (sa->sa_family != dev->type)
2462 return -EINVAL;
2463 if (!netif_device_present(dev))
2464 return -ENODEV;
2465 err = dev->set_mac_address(dev, sa);
2466 if (!err)
f07d5b94 2467 raw_notifier_call_chain(&netdev_chain,
e041c683 2468 NETDEV_CHANGEADDR, dev);
1da177e4
LT		 2469 return err;
2470}
2471
2472/*
2473 * Perform the SIOCxIFxxx calls.
2474 */
2475static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2476{
2477 int err;
2478 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2479
2480 if (!dev)
2481 return -ENODEV;
2482
2483 switch (cmd) {
2484 case SIOCGIFFLAGS: /* Get interface flags */
2485 ifr->ifr_flags = dev_get_flags(dev);
2486 return 0;
2487
2488 case SIOCSIFFLAGS: /* Set interface flags */
2489 return dev_change_flags(dev, ifr->ifr_flags);
2490
2491 case SIOCGIFMETRIC: /* Get the metric on the interface
2492 (currently unused) */
2493 ifr->ifr_metric = 0;
2494 return 0;
2495
2496 case SIOCSIFMETRIC: /* Set the metric on the interface
2497 (currently unused) */
2498 return -EOPNOTSUPP;
2499
2500 case SIOCGIFMTU: /* Get the MTU of a device */
2501 ifr->ifr_mtu = dev->mtu;
2502 return 0;
2503
2504 case SIOCSIFMTU: /* Set the MTU of a device */
2505 return dev_set_mtu(dev, ifr->ifr_mtu);
2506
2507 case SIOCGIFHWADDR:
2508 if (!dev->addr_len)
2509 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2510 else
2511 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2512 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2513 ifr->ifr_hwaddr.sa_family = dev->type;
2514 return 0;
2515
2516 case SIOCSIFHWADDR:
2517 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2518
2519 case SIOCSIFHWBROADCAST:
2520 if (ifr->ifr_hwaddr.sa_family != dev->type)
2521 return -EINVAL;
2522 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2523 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
f07d5b94 2524 raw_notifier_call_chain(&netdev_chain,
1da177e4
LT		 2525 NETDEV_CHANGEADDR, dev);
2526 return 0;
2527
2528 case SIOCGIFMAP:
2529 ifr->ifr_map.mem_start = dev->mem_start;
2530 ifr->ifr_map.mem_end = dev->mem_end;
2531 ifr->ifr_map.base_addr = dev->base_addr;
2532 ifr->ifr_map.irq = dev->irq;
2533 ifr->ifr_map.dma = dev->dma;
2534 ifr->ifr_map.port = dev->if_port;
2535 return 0;
2536
2537 case SIOCSIFMAP:
2538 if (dev->set_config) {
2539 if (!netif_device_present(dev))
2540 return -ENODEV;
2541 return dev->set_config(dev, &ifr->ifr_map);
2542 }
2543 return -EOPNOTSUPP;
2544
2545 case SIOCADDMULTI:
2546 if (!dev->set_multicast_list ||
2547 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2548 return -EINVAL;
2549 if (!netif_device_present(dev))
2550 return -ENODEV;
2551 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2552 dev->addr_len, 1);
2553
2554 case SIOCDELMULTI:
2555 if (!dev->set_multicast_list ||
2556 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2557 return -EINVAL;
2558 if (!netif_device_present(dev))
2559 return -ENODEV;
2560 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2561 dev->addr_len, 1);
2562
2563 case SIOCGIFINDEX:
2564 ifr->ifr_ifindex = dev->ifindex;
2565 return 0;
2566
2567 case SIOCGIFTXQLEN:
2568 ifr->ifr_qlen = dev->tx_queue_len;
2569 return 0;
2570
2571 case SIOCSIFTXQLEN:
2572 if (ifr->ifr_qlen < 0)
2573 return -EINVAL;
2574 dev->tx_queue_len = ifr->ifr_qlen;
2575 return 0;
2576
2577 case SIOCSIFNAME:
2578 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2579 return dev_change_name(dev, ifr->ifr_newname);
2580
2581 /*
2582 * Unknown or private ioctl
2583 */
2584
2585 default:
2586 if ((cmd >= SIOCDEVPRIVATE &&
2587 cmd <= SIOCDEVPRIVATE + 15) ||
2588 cmd == SIOCBONDENSLAVE ||
2589 cmd == SIOCBONDRELEASE ||
2590 cmd == SIOCBONDSETHWADDR ||
2591 cmd == SIOCBONDSLAVEINFOQUERY ||
2592 cmd == SIOCBONDINFOQUERY ||
2593 cmd == SIOCBONDCHANGEACTIVE ||
2594 cmd == SIOCGMIIPHY ||
2595 cmd == SIOCGMIIREG ||
2596 cmd == SIOCSMIIREG ||
2597 cmd == SIOCBRADDIF ||
2598 cmd == SIOCBRDELIF ||
2599 cmd == SIOCWANDEV) {
2600 err = -EOPNOTSUPP;
2601 if (dev->do_ioctl) {
2602 if (netif_device_present(dev))
2603 err = dev->do_ioctl(dev, ifr,
2604 cmd);
2605 else
2606 err = -ENODEV;
2607 }
2608 } else
2609 err = -EINVAL;
2610
2611 }
2612 return err;
2613}
2614
2615/*
2616 * This function handles all "interface"-type I/O control requests. The actual
2617 * 'doing' part of this is dev_ifsioc above.
2618 */
2619
2620/**
2621 * dev_ioctl - network device ioctl
2622 * @cmd: command to issue
2623 * @arg: pointer to a struct ifreq in user space
2624 *
2625 * Issue ioctl functions to devices. This is normally called by the
2626 * user space syscall interfaces but can sometimes be useful for
2627 * other purposes. The return value is the return from the syscall if
2628 * positive or a negative errno code on error.
2629 */
2630
2631int dev_ioctl(unsigned int cmd, void __user *arg)
2632{
2633 struct ifreq ifr;
2634 int ret;
2635 char *colon;
2636
2637 /* One special case: SIOCGIFCONF takes ifconf argument
2638 and requires shared lock, because it sleeps writing
2639 to user space.
2640 */
2641
2642 if (cmd == SIOCGIFCONF) {
6756ae4b 2643 rtnl_lock();
1da177e4 2644 ret = dev_ifconf((char __user *) arg);
6756ae4b 2645 rtnl_unlock();
1da177e4
LT		 2646 return ret;
2647 }
2648 if (cmd == SIOCGIFNAME)
2649 return dev_ifname((struct ifreq __user *)arg);
2650
2651 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2652 return -EFAULT;
2653
2654 ifr.ifr_name[IFNAMSIZ-1] = 0;
2655
2656 colon = strchr(ifr.ifr_name, ':');
2657 if (colon)
2658 *colon = 0;
2659
2660 /*
2661 * See which interface the caller is talking about.
2662 */
2663
2664 switch (cmd) {
2665 /*
2666 * These ioctl calls:
2667 * - can be done by all.
2668 * - atomic and do not require locking.
2669 * - return a value
2670 */
2671 case SIOCGIFFLAGS:
2672 case SIOCGIFMETRIC:
2673 case SIOCGIFMTU:
2674 case SIOCGIFHWADDR:
2675 case SIOCGIFSLAVE:
2676 case SIOCGIFMAP:
2677 case SIOCGIFINDEX:
2678 case SIOCGIFTXQLEN:
2679 dev_load(ifr.ifr_name);
2680 read_lock(&dev_base_lock);
2681 ret = dev_ifsioc(&ifr, cmd);
2682 read_unlock(&dev_base_lock);
2683 if (!ret) {
2684 if (colon)
2685 *colon = ':';
2686 if (copy_to_user(arg, &ifr,
2687 sizeof(struct ifreq)))
2688 ret = -EFAULT;
2689 }
2690 return ret;
2691
2692 case SIOCETHTOOL:
2693 dev_load(ifr.ifr_name);
2694 rtnl_lock();
2695 ret = dev_ethtool(&ifr);
2696 rtnl_unlock();
2697 if (!ret) {
2698 if (colon)
2699 *colon = ':';
2700 if (copy_to_user(arg, &ifr,
2701 sizeof(struct ifreq)))
2702 ret = -EFAULT;
2703 }
2704 return ret;
2705
2706 /*
2707 * These ioctl calls:
2708 * - require superuser power.
2709 * - require strict serialization.
2710 * - return a value
2711 */
2712 case SIOCGMIIPHY:
2713 case SIOCGMIIREG:
2714 case SIOCSIFNAME:
2715 if (!capable(CAP_NET_ADMIN))
2716 return -EPERM;
2717 dev_load(ifr.ifr_name);
2718 rtnl_lock();
2719 ret = dev_ifsioc(&ifr, cmd);
2720 rtnl_unlock();
2721 if (!ret) {
2722 if (colon)
2723 *colon = ':';
2724 if (copy_to_user(arg, &ifr,
2725 sizeof(struct ifreq)))
2726 ret = -EFAULT;
2727 }
2728 return ret;
2729
2730 /*
2731 * These ioctl calls:
2732 * - require superuser power.
2733 * - require strict serialization.
2734 * - do not return a value
2735 */
2736 case SIOCSIFFLAGS:
2737 case SIOCSIFMETRIC:
2738 case SIOCSIFMTU:
2739 case SIOCSIFMAP:
2740 case SIOCSIFHWADDR:
2741 case SIOCSIFSLAVE:
2742 case SIOCADDMULTI:
2743 case SIOCDELMULTI:
2744 case SIOCSIFHWBROADCAST:
2745 case SIOCSIFTXQLEN:
2746 case SIOCSMIIREG:
2747 case SIOCBONDENSLAVE:
2748 case SIOCBONDRELEASE:
2749 case SIOCBONDSETHWADDR:
1da177e4
LT		 2750 case SIOCBONDCHANGEACTIVE:
2751 case SIOCBRADDIF:
2752 case SIOCBRDELIF:
2753 if (!capable(CAP_NET_ADMIN))
2754 return -EPERM;
cabcac0b
TG		 2755 /* fall through */
2756 case SIOCBONDSLAVEINFOQUERY:
2757 case SIOCBONDINFOQUERY:
1da177e4
LT		 2758 dev_load(ifr.ifr_name);
2759 rtnl_lock();
2760 ret = dev_ifsioc(&ifr, cmd);
2761 rtnl_unlock();
2762 return ret;
2763
2764 case SIOCGIFMEM:
2765 /* Get the per device memory space. We can add this but
2766 * currently do not support it */
2767 case SIOCSIFMEM:
2768 /* Set the per device memory buffer space.
2769 * Not applicable in our case */
2770 case SIOCSIFLINK:
2771 return -EINVAL;
2772
2773 /*
2774 * Unknown or private ioctl.
2775 */
2776 default:
2777 if (cmd == SIOCWANDEV ||
2778 (cmd >= SIOCDEVPRIVATE &&
2779 cmd <= SIOCDEVPRIVATE + 15)) {
2780 dev_load(ifr.ifr_name);
2781 rtnl_lock();
2782 ret = dev_ifsioc(&ifr, cmd);
2783 rtnl_unlock();
2784 if (!ret && copy_to_user(arg, &ifr,
2785 sizeof(struct ifreq)))
2786 ret = -EFAULT;
2787 return ret;
2788 }
d86b5e0e 2789#ifdef CONFIG_WIRELESS_EXT
1da177e4
LT		 2790 /* Take care of Wireless Extensions */
2791 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2792 /* If command is `set a parameter', or
2793 * `get the encoding parameters', check if
2794 * the user has the right to do it */
a417016d
JT		 2795 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2796 || cmd == SIOCGIWENCODEEXT) {
1da177e4
LT		 2797 if (!capable(CAP_NET_ADMIN))
2798 return -EPERM;
2799 }
2800 dev_load(ifr.ifr_name);
2801 rtnl_lock();
2802 /* Follow me in net/core/wireless.c */
2803 ret = wireless_process_ioctl(&ifr, cmd);
2804 rtnl_unlock();
2805 if (IW_IS_GET(cmd) &&
2806 copy_to_user(arg, &ifr,
4ec93edb 2807 sizeof(struct ifreq)))
1da177e4
LT		 2808 ret = -EFAULT;
2809 return ret;
2810 }
d86b5e0e 2811#endif /* CONFIG_WIRELESS_EXT */
1da177e4
LT		 2812 return -EINVAL;
2813 }
2814}
2815
2816
2817/**
2818 * dev_new_index - allocate an ifindex
2819 *
2820 * Returns a suitable unique value for a new device interface
2821 * number. The caller must hold the rtnl semaphore or the
2822 * dev_base_lock to be sure it remains unique.
2823 */
2824static int dev_new_index(void)
2825{
2826 static int ifindex;
2827 for (;;) {
2828 if (++ifindex <= 0)
2829 ifindex = 1;
2830 if (!__dev_get_by_index(ifindex))
2831 return ifindex;
2832 }
2833}
2834
2835static int dev_boot_phase = 1;
2836
2837/* Delayed registration/unregisteration */
2838static DEFINE_SPINLOCK(net_todo_list_lock);
2839static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2840
6f05f629 2841static void net_set_todo(struct net_device *dev)
1da177e4
LT		 2842{
2843 spin_lock(&net_todo_list_lock);
2844 list_add_tail(&dev->todo_list, &net_todo_list);
2845 spin_unlock(&net_todo_list_lock);
2846}
2847
2848/**
2849 * register_netdevice - register a network device
2850 * @dev: device to register
2851 *
2852 * Take a completed network device structure and add it to the kernel
2853 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2854 * chain. 0 is returned on success. A negative errno code is returned
2855 * on a failure to set up the device, or if the name is a duplicate.
2856 *
2857 * Callers must hold the rtnl semaphore. You may want
2858 * register_netdev() instead of this.
2859 *
2860 * BUGS:
2861 * The locking appears insufficient to guarantee two parallel registers
2862 * will not get the same name.
2863 */
2864
2865int register_netdevice(struct net_device *dev)
2866{
2867 struct hlist_head *head;
2868 struct hlist_node *p;
2869 int ret;
2870
2871 BUG_ON(dev_boot_phase);
2872 ASSERT_RTNL();
2873
b17a7c17
SH		 2874 might_sleep();
2875
1da177e4
LT		 2876 /* When net_device's are persistent, this will be fatal. */
2877 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2878
2879 spin_lock_init(&dev->queue_lock);
932ff279 2880 spin_lock_init(&dev->_xmit_lock);
1da177e4
LT		 2881 dev->xmit_lock_owner = -1;
2882#ifdef CONFIG_NET_CLS_ACT
2883 spin_lock_init(&dev->ingress_lock);
2884#endif
2885
1da177e4
LT		 2886 dev->iflink = -1;
2887
2888 /* Init, if this function is available */
2889 if (dev->init) {
2890 ret = dev->init(dev);
2891 if (ret) {
2892 if (ret > 0)
2893 ret = -EIO;
90833aa4 2894 goto out;
1da177e4
LT		 2895 }
2896 }
4ec93edb 2897
1da177e4
LT		 2898 if (!dev_valid_name(dev->name)) {
2899 ret = -EINVAL;
90833aa4 2900 goto out;
1da177e4
LT		 2901 }
2902
2903 dev->ifindex = dev_new_index();
2904 if (dev->iflink == -1)
2905 dev->iflink = dev->ifindex;
2906
2907 /* Check for existence of name */
2908 head = dev_name_hash(dev->name);
2909 hlist_for_each(p, head) {
2910 struct net_device *d
2911 = hlist_entry(p, struct net_device, name_hlist);
2912 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2913 ret = -EEXIST;
4ec93edb 2914 goto out;
1da177e4 2915 }
4ec93edb 2916 }
1da177e4
LT		 2917
2918 /* Fix illegal SG+CSUM combinations. */
2919 if ((dev->features & NETIF_F_SG) &&
8648b305 2920 !(dev->features & NETIF_F_ALL_CSUM)) {
5a8da02b 2921 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
1da177e4
LT		 2922 dev->name);
2923 dev->features &= ~NETIF_F_SG;
2924 }
2925
2926 /* TSO requires that SG is present as well. */
2927 if ((dev->features & NETIF_F_TSO) &&
2928 !(dev->features & NETIF_F_SG)) {
5a8da02b 2929 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
1da177e4
LT		 2930 dev->name);
2931 dev->features &= ~NETIF_F_TSO;
2932 }
e89e9cf5
AR		 2933 if (dev->features & NETIF_F_UFO) {
2934 if (!(dev->features & NETIF_F_HW_CSUM)) {
2935 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2936 "NETIF_F_HW_CSUM feature.\n",
2937 dev->name);
2938 dev->features &= ~NETIF_F_UFO;
2939 }
2940 if (!(dev->features & NETIF_F_SG)) {
2941 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2942 "NETIF_F_SG feature.\n",
2943 dev->name);
2944 dev->features &= ~NETIF_F_UFO;
2945 }
2946 }
1da177e4
LT		 2947
2948 /*
2949 * nil rebuild_header routine,
2950 * that should be never called and used as just bug trap.
2951 */
2952
2953 if (!dev->rebuild_header)
2954 dev->rebuild_header = default_rebuild_header;
2955
b17a7c17
SH		 2956 ret = netdev_register_sysfs(dev);
2957 if (ret)
90833aa4 2958 goto out;
b17a7c17
SH		 2959 dev->reg_state = NETREG_REGISTERED;
2960
1da177e4
LT		 2961 /*
2962 * Default initial state at registry is that the
2963 * device is present.
2964 */
2965
2966 set_bit(__LINK_STATE_PRESENT, &dev->state);
2967
2968 dev->next = NULL;
2969 dev_init_scheduler(dev);
2970 write_lock_bh(&dev_base_lock);
2971 *dev_tail = dev;
2972 dev_tail = &dev->next;
2973 hlist_add_head(&dev->name_hlist, head);
2974 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2975 dev_hold(dev);
1da177e4
LT		 2976 write_unlock_bh(&dev_base_lock);
2977
2978 /* Notify protocols, that a new device appeared. */
f07d5b94 2979 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
1da177e4 2980
1da177e4
LT		 2981 ret = 0;
2982
2983out:
2984 return ret;
1da177e4
LT		 2985}
2986
2987/**
2988 * register_netdev - register a network device
2989 * @dev: device to register
2990 *
2991 * Take a completed network device structure and add it to the kernel
2992 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2993 * chain. 0 is returned on success. A negative errno code is returned
2994 * on a failure to set up the device, or if the name is a duplicate.
2995 *
2996 * This is a wrapper around register_netdev that takes the rtnl semaphore
2997 * and expands the device name if you passed a format string to
2998 * alloc_netdev.
2999 */
3000int register_netdev(struct net_device *dev)
3001{
3002 int err;
3003
3004 rtnl_lock();
3005
3006 /*
3007 * If the name is a format string the caller wants us to do a
3008 * name allocation.
3009 */
3010 if (strchr(dev->name, '%')) {
3011 err = dev_alloc_name(dev, dev->name);
3012 if (err < 0)
3013 goto out;
3014 }
4ec93edb 3015
1da177e4
LT		 3016 err = register_netdevice(dev);
3017out:
3018 rtnl_unlock();
3019 return err;
3020}
3021EXPORT_SYMBOL(register_netdev);
3022
3023/*
3024 * netdev_wait_allrefs - wait until all references are gone.
3025 *
3026 * This is called when unregistering network devices.
3027 *
3028 * Any protocol or device that holds a reference should register
3029 * for netdevice notification, and cleanup and put back the
3030 * reference if they receive an UNREGISTER event.
3031 * We can get stuck here if buggy protocols don't correctly
4ec93edb 3032 * call dev_put.
1da177e4
LT		 3033 */
3034static void netdev_wait_allrefs(struct net_device *dev)
3035{
3036 unsigned long rebroadcast_time, warning_time;
3037
3038 rebroadcast_time = warning_time = jiffies;
3039 while (atomic_read(&dev->refcnt) != 0) {
3040 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6756ae4b 3041 rtnl_lock();
1da177e4
LT		 3042
3043 /* Rebroadcast unregister notification */
f07d5b94 3044 raw_notifier_call_chain(&netdev_chain,
1da177e4
LT		 3045 NETDEV_UNREGISTER, dev);
3046
3047 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3048 &dev->state)) {
3049 /* We must not have linkwatch events
3050 * pending on unregister. If this
3051 * happens, we simply run the queue
3052 * unscheduled, resulting in a noop
3053 * for this device.
3054 */
3055 linkwatch_run_queue();
3056 }
3057
6756ae4b 3058 __rtnl_unlock();
1da177e4
LT		 3059
3060 rebroadcast_time = jiffies;
3061 }
3062
3063 msleep(250);
3064
3065 if (time_after(jiffies, warning_time + 10 * HZ)) {
3066 printk(KERN_EMERG "unregister_netdevice: "
3067 "waiting for %s to become free. Usage "
3068 "count = %d\n",
3069 dev->name, atomic_read(&dev->refcnt));
3070 warning_time = jiffies;
3071 }
3072 }
3073}
3074
3075/* The sequence is:
3076 *
3077 * rtnl_lock();
3078 * ...
3079 * register_netdevice(x1);
3080 * register_netdevice(x2);
3081 * ...
3082 * unregister_netdevice(y1);
3083 * unregister_netdevice(y2);
3084 * ...
3085 * rtnl_unlock();
3086 * free_netdev(y1);
3087 * free_netdev(y2);
3088 *
3089 * We are invoked by rtnl_unlock() after it drops the semaphore.
3090 * This allows us to deal with problems:
b17a7c17 3091 * 1) We can delete sysfs objects which invoke hotplug
1da177e4
LT		 3092 * without deadlocking with linkwatch via keventd.
3093 * 2) Since we run with the RTNL semaphore not held, we can sleep
3094 * safely in order to wait for the netdev refcnt to drop to zero.
3095 */
4a3e2f71 3096static DEFINE_MUTEX(net_todo_run_mutex);
1da177e4
LT		 3097void netdev_run_todo(void)
3098{
626ab0e6 3099 struct list_head list;
1da177e4
LT		 3100
3101 /* Need to guard against multiple cpu's getting out of order. */
4a3e2f71 3102 mutex_lock(&net_todo_run_mutex);
1da177e4
LT		 3103
3104 /* Not safe to do outside the semaphore. We must not return
3105 * until all unregister events invoked by the local processor
3106 * have been completed (either by this todo run, or one on
3107 * another cpu).
3108 */
3109 if (list_empty(&net_todo_list))
3110 goto out;
3111
3112 /* Snapshot list, allow later requests */
3113 spin_lock(&net_todo_list_lock);
626ab0e6 3114 list_replace_init(&net_todo_list, &list);
1da177e4 3115 spin_unlock(&net_todo_list_lock);
626ab0e6 3116
1da177e4
LT		 3117 while (!list_empty(&list)) {
3118 struct net_device *dev
3119 = list_entry(list.next, struct net_device, todo_list);
3120 list_del(&dev->todo_list);
3121
b17a7c17
SH		 3122 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3123 printk(KERN_ERR "network todo '%s' but state %d\n",
3124 dev->name, dev->reg_state);
3125 dump_stack();
3126 continue;
3127 }
1da177e4 3128
b17a7c17
SH		 3129 netdev_unregister_sysfs(dev);
3130 dev->reg_state = NETREG_UNREGISTERED;
1da177e4 3131
b17a7c17 3132 netdev_wait_allrefs(dev);
1da177e4 3133
b17a7c17
SH		 3134 /* paranoia */
3135 BUG_ON(atomic_read(&dev->refcnt));
3136 BUG_TRAP(!dev->ip_ptr);
3137 BUG_TRAP(!dev->ip6_ptr);
3138 BUG_TRAP(!dev->dn_ptr);
1da177e4 3139
b17a7c17
SH		 3140 /* It must be the very last action,
3141 * after this 'dev' may point to freed up memory.
3142 */
3143 if (dev->destructor)
3144 dev->destructor(dev);
1da177e4
LT		 3145 }
3146
3147out:
4a3e2f71 3148 mutex_unlock(&net_todo_run_mutex);
1da177e4
LT		 3149}
3150
3151/**
3152 * alloc_netdev - allocate network device
3153 * @sizeof_priv: size of private data to allocate space for
3154 * @name: device name format string
3155 * @setup: callback to initialize device
3156 *
3157 * Allocates a struct net_device with private data area for driver use
3158 * and performs basic initialization.
3159 */
3160struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3161 void (*setup)(struct net_device *))
3162{
3163 void *p;
3164 struct net_device *dev;
3165 int alloc_size;
3166
b6fe17d6
SH		 3167 BUG_ON(strlen(name) >= sizeof(dev->name));
3168
1da177e4
LT		 3169 /* ensure 32-byte alignment of both the device and private area */
3170 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3171 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3172
31380de9 3173 p = kzalloc(alloc_size, GFP_KERNEL);
1da177e4 3174 if (!p) {
b6fe17d6 3175 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
1da177e4
LT		 3176 return NULL;
3177 }
1da177e4
LT		 3178
3179 dev = (struct net_device *)
3180 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3181 dev->padded = (char *)dev - (char *)p;
3182
3183 if (sizeof_priv)
3184 dev->priv = netdev_priv(dev);
3185
3186 setup(dev);
3187 strcpy(dev->name, name);
3188 return dev;
3189}
3190EXPORT_SYMBOL(alloc_netdev);
3191
3192/**
3193 * free_netdev - free network device
3194 * @dev: device
3195 *
4ec93edb
YH		 3196 * This function does the last stage of destroying an allocated device
3197 * interface. The reference to the device object is released.
1da177e4
LT		 3198 * If this is the last reference then it will be freed.
3199 */
3200void free_netdev(struct net_device *dev)
3201{
3202#ifdef CONFIG_SYSFS
3041a069 3203 /* Compatibility with error handling in drivers */
1da177e4
LT		 3204 if (dev->reg_state == NETREG_UNINITIALIZED) {
3205 kfree((char *)dev - dev->padded);
3206 return;
3207 }
3208
3209 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3210 dev->reg_state = NETREG_RELEASED;
3211
43cb76d9
GKH		 3212 /* will free via device release */
3213 put_device(&dev->dev);
1da177e4
LT		 3214#else
3215 kfree((char *)dev - dev->padded);
3216#endif
3217}
4ec93edb 3218
1da177e4 3219/* Synchronize with packet receive processing. */
4ec93edb 3220void synchronize_net(void)
1da177e4
LT		 3221{
3222 might_sleep();
fbd568a3 3223 synchronize_rcu();
1da177e4
LT		 3224}
3225
3226/**
3227 * unregister_netdevice - remove device from the kernel
3228 * @dev: device
3229 *
3230 * This function shuts down a device interface and removes it
3231 * from the kernel tables. On success 0 is returned, on a failure
3232 * a negative errno code is returned.
3233 *
3234 * Callers must hold the rtnl semaphore. You may want
3235 * unregister_netdev() instead of this.
3236 */
3237
22f8cde5 3238void unregister_netdevice(struct net_device *dev)
1da177e4
LT		 3239{
3240 struct net_device *d, **dp;
3241
3242 BUG_ON(dev_boot_phase);
3243 ASSERT_RTNL();
3244
3245 /* Some devices call without registering for initialization unwind. */
3246 if (dev->reg_state == NETREG_UNINITIALIZED) {
3247 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3248 "was registered\n", dev->name, dev);
22f8cde5
SH		 3249
3250 WARN_ON(1);
3251 return;
1da177e4
LT		 3252 }
3253
3254 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3255
3256 /* If device is running, close it first. */
3257 if (dev->flags & IFF_UP)
3258 dev_close(dev);
3259
3260 /* And unlink it from device chain. */
3261 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3262 if (d == dev) {
3263 write_lock_bh(&dev_base_lock);
3264 hlist_del(&dev->name_hlist);
3265 hlist_del(&dev->index_hlist);
3266 if (dev_tail == &dev->next)
3267 dev_tail = dp;
3268 *dp = d->next;
3269 write_unlock_bh(&dev_base_lock);
3270 break;
3271 }
3272 }
22f8cde5 3273 BUG_ON(!d);
1da177e4
LT		 3274
3275 dev->reg_state = NETREG_UNREGISTERING;
3276
3277 synchronize_net();
3278
3279 /* Shutdown queueing discipline. */
3280 dev_shutdown(dev);
3281
4ec93edb 3282
1da177e4
LT		 3283 /* Notify protocols, that we are about to destroy
3284 this device. They should clean all the things.
3285 */
f07d5b94 3286 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
4ec93edb 3287
1da177e4
LT		 3288 /*
3289 * Flush the multicast chain
3290 */
3291 dev_mc_discard(dev);
3292
3293 if (dev->uninit)
3294 dev->uninit(dev);
3295
3296 /* Notifier chain MUST detach us from master device. */
3297 BUG_TRAP(!dev->master);
3298
1da177e4
LT		 3299 /* Finish processing unregister after unlock */
3300 net_set_todo(dev);
3301
3302 synchronize_net();
3303
3304 dev_put(dev);
1da177e4
LT		 3305}
3306
3307/**
3308 * unregister_netdev - remove device from the kernel
3309 * @dev: device
3310 *
3311 * This function shuts down a device interface and removes it
3312 * from the kernel tables. On success 0 is returned, on a failure
3313 * a negative errno code is returned.
3314 *
3315 * This is just a wrapper for unregister_netdevice that takes
3316 * the rtnl semaphore. In general you want to use this and not
3317 * unregister_netdevice.
3318 */
3319void unregister_netdev(struct net_device *dev)
3320{
3321 rtnl_lock();
3322 unregister_netdevice(dev);
3323 rtnl_unlock();
3324}
3325
3326EXPORT_SYMBOL(unregister_netdev);
3327
1da177e4
LT		 3328static int dev_cpu_callback(struct notifier_block *nfb,
3329 unsigned long action,
3330 void *ocpu)
3331{
3332 struct sk_buff **list_skb;
3333 struct net_device **list_net;
3334 struct sk_buff *skb;
3335 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3336 struct softnet_data *sd, *oldsd;
3337
3338 if (action != CPU_DEAD)
3339 return NOTIFY_OK;
3340
3341 local_irq_disable();
3342 cpu = smp_processor_id();
3343 sd = &per_cpu(softnet_data, cpu);
3344 oldsd = &per_cpu(softnet_data, oldcpu);
3345
3346 /* Find end of our completion_queue. */
3347 list_skb = &sd->completion_queue;
3348 while (*list_skb)
3349 list_skb = &(*list_skb)->next;
3350 /* Append completion queue from offline CPU. */
3351 *list_skb = oldsd->completion_queue;
3352 oldsd->completion_queue = NULL;
3353
3354 /* Find end of our output_queue. */
3355 list_net = &sd->output_queue;
3356 while (*list_net)
3357 list_net = &(*list_net)->next_sched;
3358 /* Append output queue from offline CPU. */
3359 *list_net = oldsd->output_queue;
3360 oldsd->output_queue = NULL;
3361
3362 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3363 local_irq_enable();
3364
3365 /* Process offline CPU's input_pkt_queue */
3366 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3367 netif_rx(skb);
3368
3369 return NOTIFY_OK;
3370}
1da177e4 3371
db217334
CL		 3372#ifdef CONFIG_NET_DMA
3373/**
3374 * net_dma_rebalance -
3375 * This is called when the number of channels allocated to the net_dma_client
3376 * changes. The net_dma_client tries to have one DMA channel per CPU.
3377 */
3378static void net_dma_rebalance(void)
3379{
3380 unsigned int cpu, i, n;
3381 struct dma_chan *chan;
3382
db217334
CL		 3383 if (net_dma_count == 0) {
3384 for_each_online_cpu(cpu)
29bbd72d 3385 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
db217334
CL		 3386 return;
3387 }
3388
3389 i = 0;
3390 cpu = first_cpu(cpu_online_map);
3391
3392 rcu_read_lock();
3393 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3394 n = ((num_online_cpus() / net_dma_count)
3395 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3396
3397 while(n) {
29bbd72d 3398 per_cpu(softnet_data, cpu).net_dma = chan;
db217334
CL		 3399 cpu = next_cpu(cpu, cpu_online_map);
3400 n--;
3401 }
3402 i++;
3403 }
3404 rcu_read_unlock();
db217334
CL		 3405}
3406
3407/**
3408 * netdev_dma_event - event callback for the net_dma_client
3409 * @client: should always be net_dma_client
f4b8ea78
RD		 3410 * @chan: DMA channel for the event
3411 * @event: event type
db217334
CL		 3412 */
3413static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3414 enum dma_event event)
3415{
3416 spin_lock(&net_dma_event_lock);
3417 switch (event) {
3418 case DMA_RESOURCE_ADDED:
3419 net_dma_count++;
3420 net_dma_rebalance();
3421 break;
3422 case DMA_RESOURCE_REMOVED:
3423 net_dma_count--;
3424 net_dma_rebalance();
3425 break;
3426 default:
3427 break;
3428 }
3429 spin_unlock(&net_dma_event_lock);
3430}
3431
3432/**
3433 * netdev_dma_regiser - register the networking subsystem as a DMA client
3434 */
3435static int __init netdev_dma_register(void)
3436{
3437 spin_lock_init(&net_dma_event_lock);
3438 net_dma_client = dma_async_client_register(netdev_dma_event);
3439 if (net_dma_client == NULL)
3440 return -ENOMEM;
3441
3442 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3443 return 0;
3444}
3445
3446#else
3447static int __init netdev_dma_register(void) { return -ENODEV; }
3448#endif /* CONFIG_NET_DMA */
1da177e4
LT		 3449
3450/*
3451 * Initialize the DEV module. At boot time this walks the device list and
3452 * unhooks any devices that fail to initialise (normally hardware not
3453 * present) and leaves us with a valid list of present and active devices.
3454 *
3455 */
3456
3457/*
3458 * This is called single threaded during boot, so no need
3459 * to take the rtnl semaphore.
3460 */
3461static int __init net_dev_init(void)
3462{
3463 int i, rc = -ENOMEM;
3464
3465 BUG_ON(!dev_boot_phase);
3466
1da177e4
LT		 3467 if (dev_proc_init())
3468 goto out;
3469
3470 if (netdev_sysfs_init())
3471 goto out;
3472
3473 INIT_LIST_HEAD(&ptype_all);
4ec93edb 3474 for (i = 0; i < 16; i++)
1da177e4
LT		 3475 INIT_LIST_HEAD(&ptype_base[i]);
3476
3477 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3478 INIT_HLIST_HEAD(&dev_name_head[i]);
3479
3480 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3481 INIT_HLIST_HEAD(&dev_index_head[i]);
3482
3483 /*
3484 * Initialise the packet receive queues.
3485 */
3486
6f912042 3487 for_each_possible_cpu(i) {
1da177e4
LT		 3488 struct softnet_data *queue;
3489
3490 queue = &per_cpu(softnet_data, i);
3491 skb_queue_head_init(&queue->input_pkt_queue);
1da177e4
LT		 3492 queue->completion_queue = NULL;
3493 INIT_LIST_HEAD(&queue->poll_list);
3494 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3495 queue->backlog_dev.weight = weight_p;
3496 queue->backlog_dev.poll = process_backlog;
3497 atomic_set(&queue->backlog_dev.refcnt, 1);
3498 }
3499
db217334
CL		 3500 netdev_dma_register();
3501
1da177e4
LT		 3502 dev_boot_phase = 0;
3503
3504 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3505 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3506
3507 hotcpu_notifier(dev_cpu_callback, 0);
3508 dst_init();
3509 dev_mcast_init();
3510 rc = 0;
3511out:
3512 return rc;
3513}
3514
3515subsys_initcall(net_dev_init);
3516
3517EXPORT_SYMBOL(__dev_get_by_index);
3518EXPORT_SYMBOL(__dev_get_by_name);
3519EXPORT_SYMBOL(__dev_remove_pack);
c2373ee9 3520EXPORT_SYMBOL(dev_valid_name);
1da177e4
LT		 3521EXPORT_SYMBOL(dev_add_pack);
3522EXPORT_SYMBOL(dev_alloc_name);
3523EXPORT_SYMBOL(dev_close);
3524EXPORT_SYMBOL(dev_get_by_flags);
3525EXPORT_SYMBOL(dev_get_by_index);
3526EXPORT_SYMBOL(dev_get_by_name);
1da177e4
LT		 3527EXPORT_SYMBOL(dev_open);
3528EXPORT_SYMBOL(dev_queue_xmit);
3529EXPORT_SYMBOL(dev_remove_pack);
3530EXPORT_SYMBOL(dev_set_allmulti);
3531EXPORT_SYMBOL(dev_set_promiscuity);
3532EXPORT_SYMBOL(dev_change_flags);
3533EXPORT_SYMBOL(dev_set_mtu);
3534EXPORT_SYMBOL(dev_set_mac_address);
3535EXPORT_SYMBOL(free_netdev);
3536EXPORT_SYMBOL(netdev_boot_setup_check);
3537EXPORT_SYMBOL(netdev_set_master);
3538EXPORT_SYMBOL(netdev_state_change);
3539EXPORT_SYMBOL(netif_receive_skb);
3540EXPORT_SYMBOL(netif_rx);
3541EXPORT_SYMBOL(register_gifconf);
3542EXPORT_SYMBOL(register_netdevice);
3543EXPORT_SYMBOL(register_netdevice_notifier);
3544EXPORT_SYMBOL(skb_checksum_help);
3545EXPORT_SYMBOL(synchronize_net);
3546EXPORT_SYMBOL(unregister_netdevice);
3547EXPORT_SYMBOL(unregister_netdevice_notifier);
3548EXPORT_SYMBOL(net_enable_timestamp);
3549EXPORT_SYMBOL(net_disable_timestamp);
3550EXPORT_SYMBOL(dev_get_flags);
3551
3552#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3553EXPORT_SYMBOL(br_handle_frame_hook);
3554EXPORT_SYMBOL(br_fdb_get_hook);
3555EXPORT_SYMBOL(br_fdb_put_hook);
3556#endif
3557
3558#ifdef CONFIG_KMOD
3559EXPORT_SYMBOL(dev_load);
3560#endif
3561
3562EXPORT_PER_CPU_SYMBOL(softnet_data);
Clone of davem's net-next-2.6 tree