blob: ff9dc029233a850cfa3985525c89d1da761601f8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * NET3 Protocol independent device support routines.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Derived from the non IP parts of dev.c 1.0.19
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 *
14 * Additional Authors:
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
21 *
22 * Changes:
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
34 * drivers
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
44 * call a packet.
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
58 * 1 device.
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
66 * the backlog queue.
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
73 */
74
75#include <asm/uaccess.h>
76#include <asm/system.h>
77#include <linux/bitops.h>
78#include <linux/config.h>
79#include <linux/cpu.h>
80#include <linux/types.h>
81#include <linux/kernel.h>
82#include <linux/sched.h>
83#include <linux/string.h>
84#include <linux/mm.h>
85#include <linux/socket.h>
86#include <linux/sockios.h>
87#include <linux/errno.h>
88#include <linux/interrupt.h>
89#include <linux/if_ether.h>
90#include <linux/netdevice.h>
91#include <linux/etherdevice.h>
92#include <linux/notifier.h>
93#include <linux/skbuff.h>
94#include <net/sock.h>
95#include <linux/rtnetlink.h>
96#include <linux/proc_fs.h>
97#include <linux/seq_file.h>
98#include <linux/stat.h>
99#include <linux/if_bridge.h>
100#include <linux/divert.h>
101#include <net/dst.h>
102#include <net/pkt_sched.h>
103#include <net/checksum.h>
104#include <linux/highmem.h>
105#include <linux/init.h>
106#include <linux/kmod.h>
107#include <linux/module.h>
108#include <linux/kallsyms.h>
109#include <linux/netpoll.h>
110#include <linux/rcupdate.h>
111#include <linux/delay.h>
112#ifdef CONFIG_NET_RADIO
113#include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
114#include <net/iw_handler.h>
115#endif /* CONFIG_NET_RADIO */
116#include <asm/current.h>
117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118/*
119 * The list of packet types we will receive (as opposed to discard)
120 * and the routines to invoke.
121 *
122 * Why 16. Because with 16 the only overlap we get on a hash of the
123 * low nibble of the protocol value is RARP/SNAP/X.25.
124 *
125 * NOTE: That is no longer true with the addition of VLAN tags. Not
126 * sure which should go first, but I bet it won't make much
127 * difference if we are running VLANs. The good news is that
128 * this protocol won't be in the list unless compiled in, so
129 * the average user (w/out VLANs) will not be adversly affected.
130 * --BLG
131 *
132 * 0800 IP
133 * 8100 802.1Q VLAN
134 * 0001 802.3
135 * 0002 AX.25
136 * 0004 802.2
137 * 8035 RARP
138 * 0005 SNAP
139 * 0805 X.25
140 * 0806 ARP
141 * 8137 IPX
142 * 0009 Localtalk
143 * 86DD IPv6
144 */
145
146static DEFINE_SPINLOCK(ptype_lock);
147static struct list_head ptype_base[16]; /* 16 way hashed list */
148static struct list_head ptype_all; /* Taps */
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150/*
151 * The @dev_base list is protected by @dev_base_lock and the rtln
152 * semaphore.
153 *
154 * Pure readers hold dev_base_lock for reading.
155 *
156 * Writers must hold the rtnl semaphore while they loop through the
157 * dev_base list, and hold dev_base_lock for writing when they do the
158 * actual updates. This allows pure readers to access the list even
159 * while a writer is preparing to update it.
160 *
161 * To put it another way, dev_base_lock is held for writing only to
162 * protect against pure readers; the rtnl semaphore provides the
163 * protection against other writers.
164 *
165 * See, for example usages, register_netdevice() and
166 * unregister_netdevice(), which must be called with the rtnl
167 * semaphore held.
168 */
169struct net_device *dev_base;
170static struct net_device **dev_tail = &dev_base;
171DEFINE_RWLOCK(dev_base_lock);
172
173EXPORT_SYMBOL(dev_base);
174EXPORT_SYMBOL(dev_base_lock);
175
176#define NETDEV_HASHBITS 8
177static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
178static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
179
180static inline struct hlist_head *dev_name_hash(const char *name)
181{
182 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
183 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
184}
185
186static inline struct hlist_head *dev_index_hash(int ifindex)
187{
188 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
189}
190
191/*
192 * Our notifier list
193 */
194
195static struct notifier_block *netdev_chain;
196
197/*
198 * Device drivers call our routines to queue packets here. We empty the
199 * queue in the local softnet handler.
200 */
Stephen Hemminger31aa02c2005-06-23 20:12:48 -0700201DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
203#ifdef CONFIG_SYSFS
204extern int netdev_sysfs_init(void);
205extern int netdev_register_sysfs(struct net_device *);
206extern void netdev_unregister_sysfs(struct net_device *);
207#else
208#define netdev_sysfs_init() (0)
209#define netdev_register_sysfs(dev) (0)
210#define netdev_unregister_sysfs(dev) do { } while(0)
211#endif
212
213
214/*******************************************************************************
215
216 Protocol management and registration routines
217
218*******************************************************************************/
219
220/*
221 * For efficiency
222 */
223
224int netdev_nit;
225
226/*
227 * Add a protocol ID to the list. Now that the input handler is
228 * smarter we can dispense with all the messy stuff that used to be
229 * here.
230 *
231 * BEWARE!!! Protocol handlers, mangling input packets,
232 * MUST BE last in hash buckets and checking protocol handlers
233 * MUST start from promiscuous ptype_all chain in net_bh.
234 * It is true now, do not change it.
235 * Explanation follows: if protocol handler, mangling packet, will
236 * be the first on list, it is not able to sense, that packet
237 * is cloned and should be copied-on-write, so that it will
238 * change it and subsequent readers will get broken packet.
239 * --ANK (980803)
240 */
241
242/**
243 * dev_add_pack - add packet handler
244 * @pt: packet type declaration
245 *
246 * Add a protocol handler to the networking stack. The passed &packet_type
247 * is linked into kernel lists and may not be freed until it has been
248 * removed from the kernel lists.
249 *
250 * This call does not sleep therefore it can not
251 * guarantee all CPU's that are in middle of receiving packets
252 * will see the new packet type (until the next received packet).
253 */
254
255void dev_add_pack(struct packet_type *pt)
256{
257 int hash;
258
259 spin_lock_bh(&ptype_lock);
260 if (pt->type == htons(ETH_P_ALL)) {
261 netdev_nit++;
262 list_add_rcu(&pt->list, &ptype_all);
263 } else {
264 hash = ntohs(pt->type) & 15;
265 list_add_rcu(&pt->list, &ptype_base[hash]);
266 }
267 spin_unlock_bh(&ptype_lock);
268}
269
270extern void linkwatch_run_queue(void);
271
272
273
274/**
275 * __dev_remove_pack - remove packet handler
276 * @pt: packet type declaration
277 *
278 * Remove a protocol handler that was previously added to the kernel
279 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
280 * from the kernel lists and can be freed or reused once this function
281 * returns.
282 *
283 * The packet type might still be in use by receivers
284 * and must not be freed until after all the CPU's have gone
285 * through a quiescent state.
286 */
287void __dev_remove_pack(struct packet_type *pt)
288{
289 struct list_head *head;
290 struct packet_type *pt1;
291
292 spin_lock_bh(&ptype_lock);
293
294 if (pt->type == htons(ETH_P_ALL)) {
295 netdev_nit--;
296 head = &ptype_all;
297 } else
298 head = &ptype_base[ntohs(pt->type) & 15];
299
300 list_for_each_entry(pt1, head, list) {
301 if (pt == pt1) {
302 list_del_rcu(&pt->list);
303 goto out;
304 }
305 }
306
307 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
308out:
309 spin_unlock_bh(&ptype_lock);
310}
311/**
312 * dev_remove_pack - remove packet handler
313 * @pt: packet type declaration
314 *
315 * Remove a protocol handler that was previously added to the kernel
316 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
317 * from the kernel lists and can be freed or reused once this function
318 * returns.
319 *
320 * This call sleeps to guarantee that no CPU is looking at the packet
321 * type after return.
322 */
323void dev_remove_pack(struct packet_type *pt)
324{
325 __dev_remove_pack(pt);
326
327 synchronize_net();
328}
329
330/******************************************************************************
331
332 Device Boot-time Settings Routines
333
334*******************************************************************************/
335
336/* Boot time configuration table */
337static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
338
339/**
340 * netdev_boot_setup_add - add new setup entry
341 * @name: name of the device
342 * @map: configured settings for the device
343 *
344 * Adds new setup entry to the dev_boot_setup list. The function
345 * returns 0 on error and 1 on success. This is a generic routine to
346 * all netdevices.
347 */
348static int netdev_boot_setup_add(char *name, struct ifmap *map)
349{
350 struct netdev_boot_setup *s;
351 int i;
352
353 s = dev_boot_setup;
354 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
355 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
356 memset(s[i].name, 0, sizeof(s[i].name));
357 strcpy(s[i].name, name);
358 memcpy(&s[i].map, map, sizeof(s[i].map));
359 break;
360 }
361 }
362
363 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
364}
365
366/**
367 * netdev_boot_setup_check - check boot time settings
368 * @dev: the netdevice
369 *
370 * Check boot time settings for the device.
371 * The found settings are set for the device to be used
372 * later in the device probing.
373 * Returns 0 if no settings found, 1 if they are.
374 */
375int netdev_boot_setup_check(struct net_device *dev)
376{
377 struct netdev_boot_setup *s = dev_boot_setup;
378 int i;
379
380 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
381 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
382 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
383 dev->irq = s[i].map.irq;
384 dev->base_addr = s[i].map.base_addr;
385 dev->mem_start = s[i].map.mem_start;
386 dev->mem_end = s[i].map.mem_end;
387 return 1;
388 }
389 }
390 return 0;
391}
392
393
394/**
395 * netdev_boot_base - get address from boot time settings
396 * @prefix: prefix for network device
397 * @unit: id for network device
398 *
399 * Check boot time settings for the base address of device.
400 * The found settings are set for the device to be used
401 * later in the device probing.
402 * Returns 0 if no settings found.
403 */
404unsigned long netdev_boot_base(const char *prefix, int unit)
405{
406 const struct netdev_boot_setup *s = dev_boot_setup;
407 char name[IFNAMSIZ];
408 int i;
409
410 sprintf(name, "%s%d", prefix, unit);
411
412 /*
413 * If device already registered then return base of 1
414 * to indicate not to probe for this interface
415 */
416 if (__dev_get_by_name(name))
417 return 1;
418
419 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
420 if (!strcmp(name, s[i].name))
421 return s[i].map.base_addr;
422 return 0;
423}
424
425/*
426 * Saves at boot time configured settings for any netdevice.
427 */
428int __init netdev_boot_setup(char *str)
429{
430 int ints[5];
431 struct ifmap map;
432
433 str = get_options(str, ARRAY_SIZE(ints), ints);
434 if (!str || !*str)
435 return 0;
436
437 /* Save settings */
438 memset(&map, 0, sizeof(map));
439 if (ints[0] > 0)
440 map.irq = ints[1];
441 if (ints[0] > 1)
442 map.base_addr = ints[2];
443 if (ints[0] > 2)
444 map.mem_start = ints[3];
445 if (ints[0] > 3)
446 map.mem_end = ints[4];
447
448 /* Add new entry to the list */
449 return netdev_boot_setup_add(str, &map);
450}
451
452__setup("netdev=", netdev_boot_setup);
453
454/*******************************************************************************
455
456 Device Interface Subroutines
457
458*******************************************************************************/
459
460/**
461 * __dev_get_by_name - find a device by its name
462 * @name: name to find
463 *
464 * Find an interface by name. Must be called under RTNL semaphore
465 * or @dev_base_lock. If the name is found a pointer to the device
466 * is returned. If the name is not found then %NULL is returned. The
467 * reference counters are not incremented so the caller must be
468 * careful with locks.
469 */
470
471struct net_device *__dev_get_by_name(const char *name)
472{
473 struct hlist_node *p;
474
475 hlist_for_each(p, dev_name_hash(name)) {
476 struct net_device *dev
477 = hlist_entry(p, struct net_device, name_hlist);
478 if (!strncmp(dev->name, name, IFNAMSIZ))
479 return dev;
480 }
481 return NULL;
482}
483
484/**
485 * dev_get_by_name - find a device by its name
486 * @name: name to find
487 *
488 * Find an interface by name. This can be called from any
489 * context and does its own locking. The returned handle has
490 * the usage count incremented and the caller must use dev_put() to
491 * release it when it is no longer needed. %NULL is returned if no
492 * matching device is found.
493 */
494
495struct net_device *dev_get_by_name(const char *name)
496{
497 struct net_device *dev;
498
499 read_lock(&dev_base_lock);
500 dev = __dev_get_by_name(name);
501 if (dev)
502 dev_hold(dev);
503 read_unlock(&dev_base_lock);
504 return dev;
505}
506
507/**
508 * __dev_get_by_index - find a device by its ifindex
509 * @ifindex: index of device
510 *
511 * Search for an interface by index. Returns %NULL if the device
512 * is not found or a pointer to the device. The device has not
513 * had its reference counter increased so the caller must be careful
514 * about locking. The caller must hold either the RTNL semaphore
515 * or @dev_base_lock.
516 */
517
518struct net_device *__dev_get_by_index(int ifindex)
519{
520 struct hlist_node *p;
521
522 hlist_for_each(p, dev_index_hash(ifindex)) {
523 struct net_device *dev
524 = hlist_entry(p, struct net_device, index_hlist);
525 if (dev->ifindex == ifindex)
526 return dev;
527 }
528 return NULL;
529}
530
531
532/**
533 * dev_get_by_index - find a device by its ifindex
534 * @ifindex: index of device
535 *
536 * Search for an interface by index. Returns NULL if the device
537 * is not found or a pointer to the device. The device returned has
538 * had a reference added and the pointer is safe until the user calls
539 * dev_put to indicate they have finished with it.
540 */
541
542struct net_device *dev_get_by_index(int ifindex)
543{
544 struct net_device *dev;
545
546 read_lock(&dev_base_lock);
547 dev = __dev_get_by_index(ifindex);
548 if (dev)
549 dev_hold(dev);
550 read_unlock(&dev_base_lock);
551 return dev;
552}
553
554/**
555 * dev_getbyhwaddr - find a device by its hardware address
556 * @type: media type of device
557 * @ha: hardware address
558 *
559 * Search for an interface by MAC address. Returns NULL if the device
560 * is not found or a pointer to the device. The caller must hold the
561 * rtnl semaphore. The returned device has not had its ref count increased
562 * and the caller must therefore be careful about locking
563 *
564 * BUGS:
565 * If the API was consistent this would be __dev_get_by_hwaddr
566 */
567
568struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
569{
570 struct net_device *dev;
571
572 ASSERT_RTNL();
573
574 for (dev = dev_base; dev; dev = dev->next)
575 if (dev->type == type &&
576 !memcmp(dev->dev_addr, ha, dev->addr_len))
577 break;
578 return dev;
579}
580
581struct net_device *dev_getfirstbyhwtype(unsigned short type)
582{
583 struct net_device *dev;
584
585 rtnl_lock();
586 for (dev = dev_base; dev; dev = dev->next) {
587 if (dev->type == type) {
588 dev_hold(dev);
589 break;
590 }
591 }
592 rtnl_unlock();
593 return dev;
594}
595
596EXPORT_SYMBOL(dev_getfirstbyhwtype);
597
598/**
599 * dev_get_by_flags - find any device with given flags
600 * @if_flags: IFF_* values
601 * @mask: bitmask of bits in if_flags to check
602 *
603 * Search for any interface with the given flags. Returns NULL if a device
604 * is not found or a pointer to the device. The device returned has
605 * had a reference added and the pointer is safe until the user calls
606 * dev_put to indicate they have finished with it.
607 */
608
609struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
610{
611 struct net_device *dev;
612
613 read_lock(&dev_base_lock);
614 for (dev = dev_base; dev != NULL; dev = dev->next) {
615 if (((dev->flags ^ if_flags) & mask) == 0) {
616 dev_hold(dev);
617 break;
618 }
619 }
620 read_unlock(&dev_base_lock);
621 return dev;
622}
623
624/**
625 * dev_valid_name - check if name is okay for network device
626 * @name: name string
627 *
628 * Network device names need to be valid file names to
629 * to allow sysfs to work
630 */
631static int dev_valid_name(const char *name)
632{
633 return !(*name == '\0'
634 || !strcmp(name, ".")
635 || !strcmp(name, "..")
636 || strchr(name, '/'));
637}
638
639/**
640 * dev_alloc_name - allocate a name for a device
641 * @dev: device
642 * @name: name format string
643 *
644 * Passed a format string - eg "lt%d" it will try and find a suitable
645 * id. Not efficient for many devices, not called a lot. The caller
646 * must hold the dev_base or rtnl lock while allocating the name and
647 * adding the device in order to avoid duplicates. Returns the number
648 * of the unit assigned or a negative errno code.
649 */
650
651int dev_alloc_name(struct net_device *dev, const char *name)
652{
653 int i = 0;
654 char buf[IFNAMSIZ];
655 const char *p;
656 const int max_netdevices = 8*PAGE_SIZE;
657 long *inuse;
658 struct net_device *d;
659
660 p = strnchr(name, IFNAMSIZ-1, '%');
661 if (p) {
662 /*
663 * Verify the string as this thing may have come from
664 * the user. There must be either one "%d" and no other "%"
665 * characters.
666 */
667 if (p[1] != 'd' || strchr(p + 2, '%'))
668 return -EINVAL;
669
670 /* Use one page as a bit array of possible slots */
671 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
672 if (!inuse)
673 return -ENOMEM;
674
675 for (d = dev_base; d; d = d->next) {
676 if (!sscanf(d->name, name, &i))
677 continue;
678 if (i < 0 || i >= max_netdevices)
679 continue;
680
681 /* avoid cases where sscanf is not exact inverse of printf */
682 snprintf(buf, sizeof(buf), name, i);
683 if (!strncmp(buf, d->name, IFNAMSIZ))
684 set_bit(i, inuse);
685 }
686
687 i = find_first_zero_bit(inuse, max_netdevices);
688 free_page((unsigned long) inuse);
689 }
690
691 snprintf(buf, sizeof(buf), name, i);
692 if (!__dev_get_by_name(buf)) {
693 strlcpy(dev->name, buf, IFNAMSIZ);
694 return i;
695 }
696
697 /* It is possible to run out of possible slots
698 * when the name is long and there isn't enough space left
699 * for the digits, or if all bits are used.
700 */
701 return -ENFILE;
702}
703
704
705/**
706 * dev_change_name - change name of a device
707 * @dev: device
708 * @newname: name (or format string) must be at least IFNAMSIZ
709 *
710 * Change name of a device, can pass format strings "eth%d".
711 * for wildcarding.
712 */
713int dev_change_name(struct net_device *dev, char *newname)
714{
715 int err = 0;
716
717 ASSERT_RTNL();
718
719 if (dev->flags & IFF_UP)
720 return -EBUSY;
721
722 if (!dev_valid_name(newname))
723 return -EINVAL;
724
725 if (strchr(newname, '%')) {
726 err = dev_alloc_name(dev, newname);
727 if (err < 0)
728 return err;
729 strcpy(newname, dev->name);
730 }
731 else if (__dev_get_by_name(newname))
732 return -EEXIST;
733 else
734 strlcpy(dev->name, newname, IFNAMSIZ);
735
736 err = class_device_rename(&dev->class_dev, dev->name);
737 if (!err) {
738 hlist_del(&dev->name_hlist);
739 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
740 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
741 }
742
743 return err;
744}
745
746/**
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700747 * netdev_features_change - device changes fatures
748 * @dev: device to cause notification
749 *
750 * Called to indicate a device has changed features.
751 */
752void netdev_features_change(struct net_device *dev)
753{
754 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
755}
756EXPORT_SYMBOL(netdev_features_change);
757
758/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 * netdev_state_change - device changes state
760 * @dev: device to cause notification
761 *
762 * Called to indicate a device has changed state. This function calls
763 * the notifier chains for netdev_chain and sends a NEWLINK message
764 * to the routing socket.
765 */
766void netdev_state_change(struct net_device *dev)
767{
768 if (dev->flags & IFF_UP) {
769 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
770 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
771 }
772}
773
774/**
775 * dev_load - load a network module
776 * @name: name of interface
777 *
778 * If a network interface is not present and the process has suitable
779 * privileges this function loads the module. If module loading is not
780 * available in this kernel then it becomes a nop.
781 */
782
783void dev_load(const char *name)
784{
785 struct net_device *dev;
786
787 read_lock(&dev_base_lock);
788 dev = __dev_get_by_name(name);
789 read_unlock(&dev_base_lock);
790
791 if (!dev && capable(CAP_SYS_MODULE))
792 request_module("%s", name);
793}
794
795static int default_rebuild_header(struct sk_buff *skb)
796{
797 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
798 skb->dev ? skb->dev->name : "NULL!!!");
799 kfree_skb(skb);
800 return 1;
801}
802
803
804/**
805 * dev_open - prepare an interface for use.
806 * @dev: device to open
807 *
808 * Takes a device from down to up state. The device's private open
809 * function is invoked and then the multicast lists are loaded. Finally
810 * the device is moved into the up state and a %NETDEV_UP message is
811 * sent to the netdev notifier chain.
812 *
813 * Calling this function on an active interface is a nop. On a failure
814 * a negative errno code is returned.
815 */
816int dev_open(struct net_device *dev)
817{
818 int ret = 0;
819
820 /*
821 * Is it already up?
822 */
823
824 if (dev->flags & IFF_UP)
825 return 0;
826
827 /*
828 * Is it even present?
829 */
830 if (!netif_device_present(dev))
831 return -ENODEV;
832
833 /*
834 * Call device private open method
835 */
836 set_bit(__LINK_STATE_START, &dev->state);
837 if (dev->open) {
838 ret = dev->open(dev);
839 if (ret)
840 clear_bit(__LINK_STATE_START, &dev->state);
841 }
842
843 /*
844 * If it went open OK then:
845 */
846
847 if (!ret) {
848 /*
849 * Set the flags.
850 */
851 dev->flags |= IFF_UP;
852
853 /*
854 * Initialize multicasting status
855 */
856 dev_mc_upload(dev);
857
858 /*
859 * Wakeup transmit queue engine
860 */
861 dev_activate(dev);
862
863 /*
864 * ... and announce new interface.
865 */
866 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
867 }
868 return ret;
869}
870
871/**
872 * dev_close - shutdown an interface.
873 * @dev: device to shutdown
874 *
875 * This function moves an active device into down state. A
876 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
877 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
878 * chain.
879 */
880int dev_close(struct net_device *dev)
881{
882 if (!(dev->flags & IFF_UP))
883 return 0;
884
885 /*
886 * Tell people we are going down, so that they can
887 * prepare to death, when device is still operating.
888 */
889 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
890
891 dev_deactivate(dev);
892
893 clear_bit(__LINK_STATE_START, &dev->state);
894
895 /* Synchronize to scheduled poll. We cannot touch poll list,
896 * it can be even on different cpu. So just clear netif_running(),
897 * and wait when poll really will happen. Actually, the best place
898 * for this is inside dev->stop() after device stopped its irq
899 * engine, but this requires more changes in devices. */
900
901 smp_mb__after_clear_bit(); /* Commit netif_running(). */
902 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
903 /* No hurry. */
904 current->state = TASK_INTERRUPTIBLE;
905 schedule_timeout(1);
906 }
907
908 /*
909 * Call the device specific close. This cannot fail.
910 * Only if device is UP
911 *
912 * We allow it to be called even after a DETACH hot-plug
913 * event.
914 */
915 if (dev->stop)
916 dev->stop(dev);
917
918 /*
919 * Device is now down.
920 */
921
922 dev->flags &= ~IFF_UP;
923
924 /*
925 * Tell people we are down
926 */
927 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
928
929 return 0;
930}
931
932
933/*
934 * Device change register/unregister. These are not inline or static
935 * as we export them to the world.
936 */
937
938/**
939 * register_netdevice_notifier - register a network notifier block
940 * @nb: notifier
941 *
942 * Register a notifier to be called when network device events occur.
943 * The notifier passed is linked into the kernel structures and must
944 * not be reused until it has been unregistered. A negative errno code
945 * is returned on a failure.
946 *
947 * When registered all registration and up events are replayed
948 * to the new notifier to allow device to have a race free
949 * view of the network device list.
950 */
951
952int register_netdevice_notifier(struct notifier_block *nb)
953{
954 struct net_device *dev;
955 int err;
956
957 rtnl_lock();
958 err = notifier_chain_register(&netdev_chain, nb);
959 if (!err) {
960 for (dev = dev_base; dev; dev = dev->next) {
961 nb->notifier_call(nb, NETDEV_REGISTER, dev);
962
963 if (dev->flags & IFF_UP)
964 nb->notifier_call(nb, NETDEV_UP, dev);
965 }
966 }
967 rtnl_unlock();
968 return err;
969}
970
971/**
972 * unregister_netdevice_notifier - unregister a network notifier block
973 * @nb: notifier
974 *
975 * Unregister a notifier previously registered by
976 * register_netdevice_notifier(). The notifier is unlinked into the
977 * kernel structures and may then be reused. A negative errno code
978 * is returned on a failure.
979 */
980
981int unregister_netdevice_notifier(struct notifier_block *nb)
982{
983 return notifier_chain_unregister(&netdev_chain, nb);
984}
985
986/**
987 * call_netdevice_notifiers - call all network notifier blocks
988 * @val: value passed unmodified to notifier function
989 * @v: pointer passed unmodified to notifier function
990 *
991 * Call all network notifier blocks. Parameters and return value
992 * are as for notifier_call_chain().
993 */
994
995int call_netdevice_notifiers(unsigned long val, void *v)
996{
997 return notifier_call_chain(&netdev_chain, val, v);
998}
999
1000/* When > 0 there are consumers of rx skb time stamps */
1001static atomic_t netstamp_needed = ATOMIC_INIT(0);
1002
1003void net_enable_timestamp(void)
1004{
1005 atomic_inc(&netstamp_needed);
1006}
1007
1008void net_disable_timestamp(void)
1009{
1010 atomic_dec(&netstamp_needed);
1011}
1012
1013static inline void net_timestamp(struct timeval *stamp)
1014{
1015 if (atomic_read(&netstamp_needed))
1016 do_gettimeofday(stamp);
1017 else {
1018 stamp->tv_sec = 0;
1019 stamp->tv_usec = 0;
1020 }
1021}
1022
1023/*
1024 * Support routine. Sends outgoing frames to any network
1025 * taps currently in use.
1026 */
1027
1028void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1029{
1030 struct packet_type *ptype;
1031 net_timestamp(&skb->stamp);
1032
1033 rcu_read_lock();
1034 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1035 /* Never send packets back to the socket
1036 * they originated from - MvS (miquels@drinkel.ow.org)
1037 */
1038 if ((ptype->dev == dev || !ptype->dev) &&
1039 (ptype->af_packet_priv == NULL ||
1040 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1041 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1042 if (!skb2)
1043 break;
1044
1045 /* skb->nh should be correctly
1046 set by sender, so that the second statement is
1047 just protection against buggy protocols.
1048 */
1049 skb2->mac.raw = skb2->data;
1050
1051 if (skb2->nh.raw < skb2->data ||
1052 skb2->nh.raw > skb2->tail) {
1053 if (net_ratelimit())
1054 printk(KERN_CRIT "protocol %04x is "
1055 "buggy, dev %s\n",
1056 skb2->protocol, dev->name);
1057 skb2->nh.raw = skb2->data;
1058 }
1059
1060 skb2->h.raw = skb2->nh.raw;
1061 skb2->pkt_type = PACKET_OUTGOING;
1062 ptype->func(skb2, skb->dev, ptype);
1063 }
1064 }
1065 rcu_read_unlock();
1066}
1067
1068/*
1069 * Invalidate hardware checksum when packet is to be mangled, and
1070 * complete checksum manually on outgoing path.
1071 */
1072int skb_checksum_help(struct sk_buff *skb, int inward)
1073{
1074 unsigned int csum;
1075 int ret = 0, offset = skb->h.raw - skb->data;
1076
1077 if (inward) {
1078 skb->ip_summed = CHECKSUM_NONE;
1079 goto out;
1080 }
1081
1082 if (skb_cloned(skb)) {
1083 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1084 if (ret)
1085 goto out;
1086 }
1087
1088 if (offset > (int)skb->len)
1089 BUG();
1090 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1091
1092 offset = skb->tail - skb->h.raw;
1093 if (offset <= 0)
1094 BUG();
1095 if (skb->csum + 2 > offset)
1096 BUG();
1097
1098 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1099 skb->ip_summed = CHECKSUM_NONE;
1100out:
1101 return ret;
1102}
1103
1104#ifdef CONFIG_HIGHMEM
1105/* Actually, we should eliminate this check as soon as we know, that:
1106 * 1. IOMMU is present and allows to map all the memory.
1107 * 2. No high memory really exists on this machine.
1108 */
1109
1110static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1111{
1112 int i;
1113
1114 if (dev->features & NETIF_F_HIGHDMA)
1115 return 0;
1116
1117 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1118 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1119 return 1;
1120
1121 return 0;
1122}
1123#else
1124#define illegal_highdma(dev, skb) (0)
1125#endif
1126
1127extern void skb_release_data(struct sk_buff *);
1128
1129/* Keep head the same: replace data */
Victor Fusco86a76ca2005-07-08 14:57:47 -07001130int __skb_linearize(struct sk_buff *skb, unsigned int __nocast gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131{
1132 unsigned int size;
1133 u8 *data;
1134 long offset;
1135 struct skb_shared_info *ninfo;
1136 int headerlen = skb->data - skb->head;
1137 int expand = (skb->tail + skb->data_len) - skb->end;
1138
1139 if (skb_shared(skb))
1140 BUG();
1141
1142 if (expand <= 0)
1143 expand = 0;
1144
1145 size = skb->end - skb->head + expand;
1146 size = SKB_DATA_ALIGN(size);
1147 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1148 if (!data)
1149 return -ENOMEM;
1150
1151 /* Copy entire thing */
1152 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1153 BUG();
1154
1155 /* Set up shinfo */
1156 ninfo = (struct skb_shared_info*)(data + size);
1157 atomic_set(&ninfo->dataref, 1);
1158 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1159 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1160 ninfo->nr_frags = 0;
1161 ninfo->frag_list = NULL;
1162
1163 /* Offset between the two in bytes */
1164 offset = data - skb->head;
1165
1166 /* Free old data. */
1167 skb_release_data(skb);
1168
1169 skb->head = data;
1170 skb->end = data + size;
1171
1172 /* Set up new pointers */
1173 skb->h.raw += offset;
1174 skb->nh.raw += offset;
1175 skb->mac.raw += offset;
1176 skb->tail += offset;
1177 skb->data += offset;
1178
1179 /* We are no longer a clone, even if we were. */
1180 skb->cloned = 0;
1181
1182 skb->tail += skb->data_len;
1183 skb->data_len = 0;
1184 return 0;
1185}
1186
1187#define HARD_TX_LOCK(dev, cpu) { \
1188 if ((dev->features & NETIF_F_LLTX) == 0) { \
1189 spin_lock(&dev->xmit_lock); \
1190 dev->xmit_lock_owner = cpu; \
1191 } \
1192}
1193
1194#define HARD_TX_UNLOCK(dev) { \
1195 if ((dev->features & NETIF_F_LLTX) == 0) { \
1196 dev->xmit_lock_owner = -1; \
1197 spin_unlock(&dev->xmit_lock); \
1198 } \
1199}
1200
1201/**
1202 * dev_queue_xmit - transmit a buffer
1203 * @skb: buffer to transmit
1204 *
1205 * Queue a buffer for transmission to a network device. The caller must
1206 * have set the device and priority and built the buffer before calling
1207 * this function. The function can be called from an interrupt.
1208 *
1209 * A negative errno code is returned on a failure. A success does not
1210 * guarantee the frame will be transmitted as it may be dropped due
1211 * to congestion or traffic shaping.
Ben Greearaf191362005-04-24 20:12:36 -07001212 *
1213 * -----------------------------------------------------------------------------------
1214 * I notice this method can also return errors from the queue disciplines,
1215 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1216 * be positive.
1217 *
1218 * Regardless of the return value, the skb is consumed, so it is currently
1219 * difficult to retry a send to this method. (You can bump the ref count
1220 * before sending to hold a reference for retry if you are careful.)
1221 *
1222 * When calling this method, interrupts MUST be enabled. This is because
1223 * the BH enable code must have IRQs enabled so that it will not deadlock.
1224 * --BLG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 */
1226
1227int dev_queue_xmit(struct sk_buff *skb)
1228{
1229 struct net_device *dev = skb->dev;
1230 struct Qdisc *q;
1231 int rc = -ENOMEM;
1232
1233 if (skb_shinfo(skb)->frag_list &&
1234 !(dev->features & NETIF_F_FRAGLIST) &&
1235 __skb_linearize(skb, GFP_ATOMIC))
1236 goto out_kfree_skb;
1237
1238 /* Fragmented skb is linearized if device does not support SG,
1239 * or if at least one of fragments is in highmem and device
1240 * does not support DMA from it.
1241 */
1242 if (skb_shinfo(skb)->nr_frags &&
1243 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1244 __skb_linearize(skb, GFP_ATOMIC))
1245 goto out_kfree_skb;
1246
1247 /* If packet is not checksummed and device does not support
1248 * checksumming for this protocol, complete checksumming here.
1249 */
1250 if (skb->ip_summed == CHECKSUM_HW &&
1251 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1252 (!(dev->features & NETIF_F_IP_CSUM) ||
1253 skb->protocol != htons(ETH_P_IP))))
1254 if (skb_checksum_help(skb, 0))
1255 goto out_kfree_skb;
1256
1257 /* Disable soft irqs for various locks below. Also
1258 * stops preemption for RCU.
1259 */
1260 local_bh_disable();
1261
1262 /* Updates of qdisc are serialized by queue_lock.
1263 * The struct Qdisc which is pointed to by qdisc is now a
1264 * rcu structure - it may be accessed without acquiring
1265 * a lock (but the structure may be stale.) The freeing of the
1266 * qdisc will be deferred until it's known that there are no
1267 * more references to it.
1268 *
1269 * If the qdisc has an enqueue function, we still need to
1270 * hold the queue_lock before calling it, since queue_lock
1271 * also serializes access to the device queue.
1272 */
1273
1274 q = rcu_dereference(dev->qdisc);
1275#ifdef CONFIG_NET_CLS_ACT
1276 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1277#endif
1278 if (q->enqueue) {
1279 /* Grab device queue */
1280 spin_lock(&dev->queue_lock);
1281
1282 rc = q->enqueue(skb, q);
1283
1284 qdisc_run(dev);
1285
1286 spin_unlock(&dev->queue_lock);
1287 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1288 goto out;
1289 }
1290
1291 /* The device has no queue. Common case for software devices:
1292 loopback, all the sorts of tunnels...
1293
1294 Really, it is unlikely that xmit_lock protection is necessary here.
1295 (f.e. loopback and IP tunnels are clean ignoring statistics
1296 counters.)
1297 However, it is possible, that they rely on protection
1298 made by us here.
1299
1300 Check this and shot the lock. It is not prone from deadlocks.
1301 Either shot noqueue qdisc, it is even simpler 8)
1302 */
1303 if (dev->flags & IFF_UP) {
1304 int cpu = smp_processor_id(); /* ok because BHs are off */
1305
1306 if (dev->xmit_lock_owner != cpu) {
1307
1308 HARD_TX_LOCK(dev, cpu);
1309
1310 if (!netif_queue_stopped(dev)) {
1311 if (netdev_nit)
1312 dev_queue_xmit_nit(skb, dev);
1313
1314 rc = 0;
1315 if (!dev->hard_start_xmit(skb, dev)) {
1316 HARD_TX_UNLOCK(dev);
1317 goto out;
1318 }
1319 }
1320 HARD_TX_UNLOCK(dev);
1321 if (net_ratelimit())
1322 printk(KERN_CRIT "Virtual device %s asks to "
1323 "queue packet!\n", dev->name);
1324 } else {
1325 /* Recursion is detected! It is possible,
1326 * unfortunately */
1327 if (net_ratelimit())
1328 printk(KERN_CRIT "Dead loop on virtual device "
1329 "%s, fix it urgently!\n", dev->name);
1330 }
1331 }
1332
1333 rc = -ENETDOWN;
1334 local_bh_enable();
1335
1336out_kfree_skb:
1337 kfree_skb(skb);
1338 return rc;
1339out:
1340 local_bh_enable();
1341 return rc;
1342}
1343
1344
1345/*=======================================================================
1346 Receiver routines
1347 =======================================================================*/
1348
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001349int netdev_max_backlog = 1000;
1350int netdev_budget = 300;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351int weight_p = 64; /* old backlog weight */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
1353DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1354
1355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356/**
1357 * netif_rx - post buffer to the network code
1358 * @skb: buffer to post
1359 *
1360 * This function receives a packet from a device driver and queues it for
1361 * the upper (protocol) levels to process. It always succeeds. The buffer
1362 * may be dropped during processing for congestion control or by the
1363 * protocol layers.
1364 *
1365 * return values:
1366 * NET_RX_SUCCESS (no congestion)
1367 * NET_RX_CN_LOW (low congestion)
1368 * NET_RX_CN_MOD (moderate congestion)
1369 * NET_RX_CN_HIGH (high congestion)
1370 * NET_RX_DROP (packet was dropped)
1371 *
1372 */
1373
1374int netif_rx(struct sk_buff *skb)
1375{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 struct softnet_data *queue;
1377 unsigned long flags;
1378
1379 /* if netpoll wants it, pretend we never saw it */
1380 if (netpoll_rx(skb))
1381 return NET_RX_DROP;
1382
1383 if (!skb->stamp.tv_sec)
1384 net_timestamp(&skb->stamp);
1385
1386 /*
1387 * The code is rearranged so that the path is the most
1388 * short when CPU is congested, but is still operating.
1389 */
1390 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 queue = &__get_cpu_var(softnet_data);
1392
1393 __get_cpu_var(netdev_rx_stat).total++;
1394 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1395 if (queue->input_pkt_queue.qlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396enqueue:
1397 dev_hold(skb->dev);
1398 __skb_queue_tail(&queue->input_pkt_queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 local_irq_restore(flags);
Stephen Hemminger34008d82005-06-23 20:10:00 -07001400 return NET_RX_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 }
1402
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 netif_rx_schedule(&queue->backlog_dev);
1404 goto enqueue;
1405 }
1406
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 __get_cpu_var(netdev_rx_stat).dropped++;
1408 local_irq_restore(flags);
1409
1410 kfree_skb(skb);
1411 return NET_RX_DROP;
1412}
1413
1414int netif_rx_ni(struct sk_buff *skb)
1415{
1416 int err;
1417
1418 preempt_disable();
1419 err = netif_rx(skb);
1420 if (local_softirq_pending())
1421 do_softirq();
1422 preempt_enable();
1423
1424 return err;
1425}
1426
1427EXPORT_SYMBOL(netif_rx_ni);
1428
1429static __inline__ void skb_bond(struct sk_buff *skb)
1430{
1431 struct net_device *dev = skb->dev;
1432
1433 if (dev->master) {
1434 skb->real_dev = skb->dev;
1435 skb->dev = dev->master;
1436 }
1437}
1438
1439static void net_tx_action(struct softirq_action *h)
1440{
1441 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1442
1443 if (sd->completion_queue) {
1444 struct sk_buff *clist;
1445
1446 local_irq_disable();
1447 clist = sd->completion_queue;
1448 sd->completion_queue = NULL;
1449 local_irq_enable();
1450
1451 while (clist) {
1452 struct sk_buff *skb = clist;
1453 clist = clist->next;
1454
1455 BUG_TRAP(!atomic_read(&skb->users));
1456 __kfree_skb(skb);
1457 }
1458 }
1459
1460 if (sd->output_queue) {
1461 struct net_device *head;
1462
1463 local_irq_disable();
1464 head = sd->output_queue;
1465 sd->output_queue = NULL;
1466 local_irq_enable();
1467
1468 while (head) {
1469 struct net_device *dev = head;
1470 head = head->next_sched;
1471
1472 smp_mb__before_clear_bit();
1473 clear_bit(__LINK_STATE_SCHED, &dev->state);
1474
1475 if (spin_trylock(&dev->queue_lock)) {
1476 qdisc_run(dev);
1477 spin_unlock(&dev->queue_lock);
1478 } else {
1479 netif_schedule(dev);
1480 }
1481 }
1482 }
1483}
1484
1485static __inline__ int deliver_skb(struct sk_buff *skb,
1486 struct packet_type *pt_prev)
1487{
1488 atomic_inc(&skb->users);
1489 return pt_prev->func(skb, skb->dev, pt_prev);
1490}
1491
1492#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1493int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1494struct net_bridge;
1495struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1496 unsigned char *addr);
1497void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1498
1499static __inline__ int handle_bridge(struct sk_buff **pskb,
1500 struct packet_type **pt_prev, int *ret)
1501{
1502 struct net_bridge_port *port;
1503
1504 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1505 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1506 return 0;
1507
1508 if (*pt_prev) {
1509 *ret = deliver_skb(*pskb, *pt_prev);
1510 *pt_prev = NULL;
1511 }
1512
1513 return br_handle_frame_hook(port, pskb);
1514}
1515#else
1516#define handle_bridge(skb, pt_prev, ret) (0)
1517#endif
1518
1519#ifdef CONFIG_NET_CLS_ACT
1520/* TODO: Maybe we should just force sch_ingress to be compiled in
1521 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1522 * a compare and 2 stores extra right now if we dont have it on
1523 * but have CONFIG_NET_CLS_ACT
1524 * NOTE: This doesnt stop any functionality; if you dont have
1525 * the ingress scheduler, you just cant add policies on ingress.
1526 *
1527 */
1528static int ing_filter(struct sk_buff *skb)
1529{
1530 struct Qdisc *q;
1531 struct net_device *dev = skb->dev;
1532 int result = TC_ACT_OK;
1533
1534 if (dev->qdisc_ingress) {
1535 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1536 if (MAX_RED_LOOP < ttl++) {
1537 printk("Redir loop detected Dropping packet (%s->%s)\n",
1538 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1539 return TC_ACT_SHOT;
1540 }
1541
1542 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1543
1544 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1545 if (NULL == skb->input_dev) {
1546 skb->input_dev = skb->dev;
1547 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1548 }
1549 spin_lock(&dev->ingress_lock);
1550 if ((q = dev->qdisc_ingress) != NULL)
1551 result = q->enqueue(skb, q);
1552 spin_unlock(&dev->ingress_lock);
1553
1554 }
1555
1556 return result;
1557}
1558#endif
1559
1560int netif_receive_skb(struct sk_buff *skb)
1561{
1562 struct packet_type *ptype, *pt_prev;
1563 int ret = NET_RX_DROP;
1564 unsigned short type;
1565
1566 /* if we've gotten here through NAPI, check netpoll */
1567 if (skb->dev->poll && netpoll_rx(skb))
1568 return NET_RX_DROP;
1569
1570 if (!skb->stamp.tv_sec)
1571 net_timestamp(&skb->stamp);
1572
1573 skb_bond(skb);
1574
1575 __get_cpu_var(netdev_rx_stat).total++;
1576
1577 skb->h.raw = skb->nh.raw = skb->data;
1578 skb->mac_len = skb->nh.raw - skb->mac.raw;
1579
1580 pt_prev = NULL;
1581
1582 rcu_read_lock();
1583
1584#ifdef CONFIG_NET_CLS_ACT
1585 if (skb->tc_verd & TC_NCLS) {
1586 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1587 goto ncls;
1588 }
1589#endif
1590
1591 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1592 if (!ptype->dev || ptype->dev == skb->dev) {
1593 if (pt_prev)
1594 ret = deliver_skb(skb, pt_prev);
1595 pt_prev = ptype;
1596 }
1597 }
1598
1599#ifdef CONFIG_NET_CLS_ACT
1600 if (pt_prev) {
1601 ret = deliver_skb(skb, pt_prev);
1602 pt_prev = NULL; /* noone else should process this after*/
1603 } else {
1604 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1605 }
1606
1607 ret = ing_filter(skb);
1608
1609 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1610 kfree_skb(skb);
1611 goto out;
1612 }
1613
1614 skb->tc_verd = 0;
1615ncls:
1616#endif
1617
1618 handle_diverter(skb);
1619
1620 if (handle_bridge(&skb, &pt_prev, &ret))
1621 goto out;
1622
1623 type = skb->protocol;
1624 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1625 if (ptype->type == type &&
1626 (!ptype->dev || ptype->dev == skb->dev)) {
1627 if (pt_prev)
1628 ret = deliver_skb(skb, pt_prev);
1629 pt_prev = ptype;
1630 }
1631 }
1632
1633 if (pt_prev) {
1634 ret = pt_prev->func(skb, skb->dev, pt_prev);
1635 } else {
1636 kfree_skb(skb);
1637 /* Jamal, now you will not able to escape explaining
1638 * me how you were going to use this. :-)
1639 */
1640 ret = NET_RX_DROP;
1641 }
1642
1643out:
1644 rcu_read_unlock();
1645 return ret;
1646}
1647
1648static int process_backlog(struct net_device *backlog_dev, int *budget)
1649{
1650 int work = 0;
1651 int quota = min(backlog_dev->quota, *budget);
1652 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1653 unsigned long start_time = jiffies;
1654
Stephen Hemmingere3876602005-06-08 14:56:01 -07001655 backlog_dev->weight = weight_p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 for (;;) {
1657 struct sk_buff *skb;
1658 struct net_device *dev;
1659
1660 local_irq_disable();
1661 skb = __skb_dequeue(&queue->input_pkt_queue);
1662 if (!skb)
1663 goto job_done;
1664 local_irq_enable();
1665
1666 dev = skb->dev;
1667
1668 netif_receive_skb(skb);
1669
1670 dev_put(dev);
1671
1672 work++;
1673
1674 if (work >= quota || jiffies - start_time > 1)
1675 break;
1676
1677 }
1678
1679 backlog_dev->quota -= work;
1680 *budget -= work;
1681 return -1;
1682
1683job_done:
1684 backlog_dev->quota -= work;
1685 *budget -= work;
1686
1687 list_del(&backlog_dev->poll_list);
1688 smp_mb__before_clear_bit();
1689 netif_poll_enable(backlog_dev);
1690
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 local_irq_enable();
1692 return 0;
1693}
1694
1695static void net_rx_action(struct softirq_action *h)
1696{
1697 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1698 unsigned long start_time = jiffies;
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001699 int budget = netdev_budget;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
1701 local_irq_disable();
1702
1703 while (!list_empty(&queue->poll_list)) {
1704 struct net_device *dev;
1705
1706 if (budget <= 0 || jiffies - start_time > 1)
1707 goto softnet_break;
1708
1709 local_irq_enable();
1710
1711 dev = list_entry(queue->poll_list.next,
1712 struct net_device, poll_list);
1713 netpoll_poll_lock(dev);
1714
1715 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1716 netpoll_poll_unlock(dev);
1717 local_irq_disable();
1718 list_del(&dev->poll_list);
1719 list_add_tail(&dev->poll_list, &queue->poll_list);
1720 if (dev->quota < 0)
1721 dev->quota += dev->weight;
1722 else
1723 dev->quota = dev->weight;
1724 } else {
1725 netpoll_poll_unlock(dev);
1726 dev_put(dev);
1727 local_irq_disable();
1728 }
1729 }
1730out:
1731 local_irq_enable();
1732 return;
1733
1734softnet_break:
1735 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1736 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1737 goto out;
1738}
1739
1740static gifconf_func_t * gifconf_list [NPROTO];
1741
1742/**
1743 * register_gifconf - register a SIOCGIF handler
1744 * @family: Address family
1745 * @gifconf: Function handler
1746 *
1747 * Register protocol dependent address dumping routines. The handler
1748 * that is passed must not be freed or reused until it has been replaced
1749 * by another handler.
1750 */
1751int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1752{
1753 if (family >= NPROTO)
1754 return -EINVAL;
1755 gifconf_list[family] = gifconf;
1756 return 0;
1757}
1758
1759
1760/*
1761 * Map an interface index to its name (SIOCGIFNAME)
1762 */
1763
1764/*
1765 * We need this ioctl for efficient implementation of the
1766 * if_indextoname() function required by the IPv6 API. Without
1767 * it, we would have to search all the interfaces to find a
1768 * match. --pb
1769 */
1770
1771static int dev_ifname(struct ifreq __user *arg)
1772{
1773 struct net_device *dev;
1774 struct ifreq ifr;
1775
1776 /*
1777 * Fetch the caller's info block.
1778 */
1779
1780 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1781 return -EFAULT;
1782
1783 read_lock(&dev_base_lock);
1784 dev = __dev_get_by_index(ifr.ifr_ifindex);
1785 if (!dev) {
1786 read_unlock(&dev_base_lock);
1787 return -ENODEV;
1788 }
1789
1790 strcpy(ifr.ifr_name, dev->name);
1791 read_unlock(&dev_base_lock);
1792
1793 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1794 return -EFAULT;
1795 return 0;
1796}
1797
1798/*
1799 * Perform a SIOCGIFCONF call. This structure will change
1800 * size eventually, and there is nothing I can do about it.
1801 * Thus we will need a 'compatibility mode'.
1802 */
1803
1804static int dev_ifconf(char __user *arg)
1805{
1806 struct ifconf ifc;
1807 struct net_device *dev;
1808 char __user *pos;
1809 int len;
1810 int total;
1811 int i;
1812
1813 /*
1814 * Fetch the caller's info block.
1815 */
1816
1817 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1818 return -EFAULT;
1819
1820 pos = ifc.ifc_buf;
1821 len = ifc.ifc_len;
1822
1823 /*
1824 * Loop over the interfaces, and write an info block for each.
1825 */
1826
1827 total = 0;
1828 for (dev = dev_base; dev; dev = dev->next) {
1829 for (i = 0; i < NPROTO; i++) {
1830 if (gifconf_list[i]) {
1831 int done;
1832 if (!pos)
1833 done = gifconf_list[i](dev, NULL, 0);
1834 else
1835 done = gifconf_list[i](dev, pos + total,
1836 len - total);
1837 if (done < 0)
1838 return -EFAULT;
1839 total += done;
1840 }
1841 }
1842 }
1843
1844 /*
1845 * All done. Write the updated control block back to the caller.
1846 */
1847 ifc.ifc_len = total;
1848
1849 /*
1850 * Both BSD and Solaris return 0 here, so we do too.
1851 */
1852 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1853}
1854
1855#ifdef CONFIG_PROC_FS
1856/*
1857 * This is invoked by the /proc filesystem handler to display a device
1858 * in detail.
1859 */
1860static __inline__ struct net_device *dev_get_idx(loff_t pos)
1861{
1862 struct net_device *dev;
1863 loff_t i;
1864
1865 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1866
1867 return i == pos ? dev : NULL;
1868}
1869
1870void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1871{
1872 read_lock(&dev_base_lock);
1873 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1874}
1875
1876void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1877{
1878 ++*pos;
1879 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1880}
1881
1882void dev_seq_stop(struct seq_file *seq, void *v)
1883{
1884 read_unlock(&dev_base_lock);
1885}
1886
1887static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1888{
1889 if (dev->get_stats) {
1890 struct net_device_stats *stats = dev->get_stats(dev);
1891
1892 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1893 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1894 dev->name, stats->rx_bytes, stats->rx_packets,
1895 stats->rx_errors,
1896 stats->rx_dropped + stats->rx_missed_errors,
1897 stats->rx_fifo_errors,
1898 stats->rx_length_errors + stats->rx_over_errors +
1899 stats->rx_crc_errors + stats->rx_frame_errors,
1900 stats->rx_compressed, stats->multicast,
1901 stats->tx_bytes, stats->tx_packets,
1902 stats->tx_errors, stats->tx_dropped,
1903 stats->tx_fifo_errors, stats->collisions,
1904 stats->tx_carrier_errors +
1905 stats->tx_aborted_errors +
1906 stats->tx_window_errors +
1907 stats->tx_heartbeat_errors,
1908 stats->tx_compressed);
1909 } else
1910 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1911}
1912
1913/*
1914 * Called from the PROCfs module. This now uses the new arbitrary sized
1915 * /proc/net interface to create /proc/net/dev
1916 */
1917static int dev_seq_show(struct seq_file *seq, void *v)
1918{
1919 if (v == SEQ_START_TOKEN)
1920 seq_puts(seq, "Inter-| Receive "
1921 " | Transmit\n"
1922 " face |bytes packets errs drop fifo frame "
1923 "compressed multicast|bytes packets errs "
1924 "drop fifo colls carrier compressed\n");
1925 else
1926 dev_seq_printf_stats(seq, v);
1927 return 0;
1928}
1929
1930static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1931{
1932 struct netif_rx_stats *rc = NULL;
1933
1934 while (*pos < NR_CPUS)
1935 if (cpu_online(*pos)) {
1936 rc = &per_cpu(netdev_rx_stat, *pos);
1937 break;
1938 } else
1939 ++*pos;
1940 return rc;
1941}
1942
1943static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1944{
1945 return softnet_get_online(pos);
1946}
1947
1948static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1949{
1950 ++*pos;
1951 return softnet_get_online(pos);
1952}
1953
1954static void softnet_seq_stop(struct seq_file *seq, void *v)
1955{
1956}
1957
1958static int softnet_seq_show(struct seq_file *seq, void *v)
1959{
1960 struct netif_rx_stats *s = v;
1961
1962 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
Stephen Hemminger31aa02c2005-06-23 20:12:48 -07001963 s->total, s->dropped, s->time_squeeze, 0,
Stephen Hemmingerc1ebcdb2005-06-23 20:08:59 -07001964 0, 0, 0, 0, /* was fastroute */
1965 s->cpu_collision );
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 return 0;
1967}
1968
1969static struct seq_operations dev_seq_ops = {
1970 .start = dev_seq_start,
1971 .next = dev_seq_next,
1972 .stop = dev_seq_stop,
1973 .show = dev_seq_show,
1974};
1975
1976static int dev_seq_open(struct inode *inode, struct file *file)
1977{
1978 return seq_open(file, &dev_seq_ops);
1979}
1980
1981static struct file_operations dev_seq_fops = {
1982 .owner = THIS_MODULE,
1983 .open = dev_seq_open,
1984 .read = seq_read,
1985 .llseek = seq_lseek,
1986 .release = seq_release,
1987};
1988
1989static struct seq_operations softnet_seq_ops = {
1990 .start = softnet_seq_start,
1991 .next = softnet_seq_next,
1992 .stop = softnet_seq_stop,
1993 .show = softnet_seq_show,
1994};
1995
1996static int softnet_seq_open(struct inode *inode, struct file *file)
1997{
1998 return seq_open(file, &softnet_seq_ops);
1999}
2000
2001static struct file_operations softnet_seq_fops = {
2002 .owner = THIS_MODULE,
2003 .open = softnet_seq_open,
2004 .read = seq_read,
2005 .llseek = seq_lseek,
2006 .release = seq_release,
2007};
2008
2009#ifdef WIRELESS_EXT
2010extern int wireless_proc_init(void);
2011#else
2012#define wireless_proc_init() 0
2013#endif
2014
2015static int __init dev_proc_init(void)
2016{
2017 int rc = -ENOMEM;
2018
2019 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2020 goto out;
2021 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2022 goto out_dev;
2023 if (wireless_proc_init())
2024 goto out_softnet;
2025 rc = 0;
2026out:
2027 return rc;
2028out_softnet:
2029 proc_net_remove("softnet_stat");
2030out_dev:
2031 proc_net_remove("dev");
2032 goto out;
2033}
2034#else
2035#define dev_proc_init() 0
2036#endif /* CONFIG_PROC_FS */
2037
2038
2039/**
2040 * netdev_set_master - set up master/slave pair
2041 * @slave: slave device
2042 * @master: new master device
2043 *
2044 * Changes the master device of the slave. Pass %NULL to break the
2045 * bonding. The caller must hold the RTNL semaphore. On a failure
2046 * a negative errno code is returned. On success the reference counts
2047 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2048 * function returns zero.
2049 */
2050int netdev_set_master(struct net_device *slave, struct net_device *master)
2051{
2052 struct net_device *old = slave->master;
2053
2054 ASSERT_RTNL();
2055
2056 if (master) {
2057 if (old)
2058 return -EBUSY;
2059 dev_hold(master);
2060 }
2061
2062 slave->master = master;
2063
2064 synchronize_net();
2065
2066 if (old)
2067 dev_put(old);
2068
2069 if (master)
2070 slave->flags |= IFF_SLAVE;
2071 else
2072 slave->flags &= ~IFF_SLAVE;
2073
2074 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2075 return 0;
2076}
2077
2078/**
2079 * dev_set_promiscuity - update promiscuity count on a device
2080 * @dev: device
2081 * @inc: modifier
2082 *
2083 * Add or remove promsicuity from a device. While the count in the device
2084 * remains above zero the interface remains promiscuous. Once it hits zero
2085 * the device reverts back to normal filtering operation. A negative inc
2086 * value is used to drop promiscuity on the device.
2087 */
2088void dev_set_promiscuity(struct net_device *dev, int inc)
2089{
2090 unsigned short old_flags = dev->flags;
2091
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 if ((dev->promiscuity += inc) == 0)
2093 dev->flags &= ~IFF_PROMISC;
David Chau52609c02005-07-05 15:11:06 -07002094 else
2095 dev->flags |= IFF_PROMISC;
2096 if (dev->flags != old_flags) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 dev_mc_upload(dev);
2098 printk(KERN_INFO "device %s %s promiscuous mode\n",
2099 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2100 "left");
2101 }
2102}
2103
2104/**
2105 * dev_set_allmulti - update allmulti count on a device
2106 * @dev: device
2107 * @inc: modifier
2108 *
2109 * Add or remove reception of all multicast frames to a device. While the
2110 * count in the device remains above zero the interface remains listening
2111 * to all interfaces. Once it hits zero the device reverts back to normal
2112 * filtering operation. A negative @inc value is used to drop the counter
2113 * when releasing a resource needing all multicasts.
2114 */
2115
2116void dev_set_allmulti(struct net_device *dev, int inc)
2117{
2118 unsigned short old_flags = dev->flags;
2119
2120 dev->flags |= IFF_ALLMULTI;
2121 if ((dev->allmulti += inc) == 0)
2122 dev->flags &= ~IFF_ALLMULTI;
2123 if (dev->flags ^ old_flags)
2124 dev_mc_upload(dev);
2125}
2126
2127unsigned dev_get_flags(const struct net_device *dev)
2128{
2129 unsigned flags;
2130
2131 flags = (dev->flags & ~(IFF_PROMISC |
2132 IFF_ALLMULTI |
2133 IFF_RUNNING)) |
2134 (dev->gflags & (IFF_PROMISC |
2135 IFF_ALLMULTI));
2136
2137 if (netif_running(dev) && netif_carrier_ok(dev))
2138 flags |= IFF_RUNNING;
2139
2140 return flags;
2141}
2142
2143int dev_change_flags(struct net_device *dev, unsigned flags)
2144{
2145 int ret;
2146 int old_flags = dev->flags;
2147
2148 /*
2149 * Set the flags on our device.
2150 */
2151
2152 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2153 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2154 IFF_AUTOMEDIA)) |
2155 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2156 IFF_ALLMULTI));
2157
2158 /*
2159 * Load in the correct multicast list now the flags have changed.
2160 */
2161
2162 dev_mc_upload(dev);
2163
2164 /*
2165 * Have we downed the interface. We handle IFF_UP ourselves
2166 * according to user attempts to set it, rather than blindly
2167 * setting it.
2168 */
2169
2170 ret = 0;
2171 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2172 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2173
2174 if (!ret)
2175 dev_mc_upload(dev);
2176 }
2177
2178 if (dev->flags & IFF_UP &&
2179 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2180 IFF_VOLATILE)))
2181 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2182
2183 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2184 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2185 dev->gflags ^= IFF_PROMISC;
2186 dev_set_promiscuity(dev, inc);
2187 }
2188
2189 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2190 is important. Some (broken) drivers set IFF_PROMISC, when
2191 IFF_ALLMULTI is requested not asking us and not reporting.
2192 */
2193 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2194 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2195 dev->gflags ^= IFF_ALLMULTI;
2196 dev_set_allmulti(dev, inc);
2197 }
2198
2199 if (old_flags ^ dev->flags)
2200 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2201
2202 return ret;
2203}
2204
2205int dev_set_mtu(struct net_device *dev, int new_mtu)
2206{
2207 int err;
2208
2209 if (new_mtu == dev->mtu)
2210 return 0;
2211
2212 /* MTU must be positive. */
2213 if (new_mtu < 0)
2214 return -EINVAL;
2215
2216 if (!netif_device_present(dev))
2217 return -ENODEV;
2218
2219 err = 0;
2220 if (dev->change_mtu)
2221 err = dev->change_mtu(dev, new_mtu);
2222 else
2223 dev->mtu = new_mtu;
2224 if (!err && dev->flags & IFF_UP)
2225 notifier_call_chain(&netdev_chain,
2226 NETDEV_CHANGEMTU, dev);
2227 return err;
2228}
2229
2230int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2231{
2232 int err;
2233
2234 if (!dev->set_mac_address)
2235 return -EOPNOTSUPP;
2236 if (sa->sa_family != dev->type)
2237 return -EINVAL;
2238 if (!netif_device_present(dev))
2239 return -ENODEV;
2240 err = dev->set_mac_address(dev, sa);
2241 if (!err)
2242 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2243 return err;
2244}
2245
2246/*
2247 * Perform the SIOCxIFxxx calls.
2248 */
2249static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2250{
2251 int err;
2252 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2253
2254 if (!dev)
2255 return -ENODEV;
2256
2257 switch (cmd) {
2258 case SIOCGIFFLAGS: /* Get interface flags */
2259 ifr->ifr_flags = dev_get_flags(dev);
2260 return 0;
2261
2262 case SIOCSIFFLAGS: /* Set interface flags */
2263 return dev_change_flags(dev, ifr->ifr_flags);
2264
2265 case SIOCGIFMETRIC: /* Get the metric on the interface
2266 (currently unused) */
2267 ifr->ifr_metric = 0;
2268 return 0;
2269
2270 case SIOCSIFMETRIC: /* Set the metric on the interface
2271 (currently unused) */
2272 return -EOPNOTSUPP;
2273
2274 case SIOCGIFMTU: /* Get the MTU of a device */
2275 ifr->ifr_mtu = dev->mtu;
2276 return 0;
2277
2278 case SIOCSIFMTU: /* Set the MTU of a device */
2279 return dev_set_mtu(dev, ifr->ifr_mtu);
2280
2281 case SIOCGIFHWADDR:
2282 if (!dev->addr_len)
2283 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2284 else
2285 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2286 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2287 ifr->ifr_hwaddr.sa_family = dev->type;
2288 return 0;
2289
2290 case SIOCSIFHWADDR:
2291 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2292
2293 case SIOCSIFHWBROADCAST:
2294 if (ifr->ifr_hwaddr.sa_family != dev->type)
2295 return -EINVAL;
2296 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2297 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2298 notifier_call_chain(&netdev_chain,
2299 NETDEV_CHANGEADDR, dev);
2300 return 0;
2301
2302 case SIOCGIFMAP:
2303 ifr->ifr_map.mem_start = dev->mem_start;
2304 ifr->ifr_map.mem_end = dev->mem_end;
2305 ifr->ifr_map.base_addr = dev->base_addr;
2306 ifr->ifr_map.irq = dev->irq;
2307 ifr->ifr_map.dma = dev->dma;
2308 ifr->ifr_map.port = dev->if_port;
2309 return 0;
2310
2311 case SIOCSIFMAP:
2312 if (dev->set_config) {
2313 if (!netif_device_present(dev))
2314 return -ENODEV;
2315 return dev->set_config(dev, &ifr->ifr_map);
2316 }
2317 return -EOPNOTSUPP;
2318
2319 case SIOCADDMULTI:
2320 if (!dev->set_multicast_list ||
2321 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2322 return -EINVAL;
2323 if (!netif_device_present(dev))
2324 return -ENODEV;
2325 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2326 dev->addr_len, 1);
2327
2328 case SIOCDELMULTI:
2329 if (!dev->set_multicast_list ||
2330 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2331 return -EINVAL;
2332 if (!netif_device_present(dev))
2333 return -ENODEV;
2334 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2335 dev->addr_len, 1);
2336
2337 case SIOCGIFINDEX:
2338 ifr->ifr_ifindex = dev->ifindex;
2339 return 0;
2340
2341 case SIOCGIFTXQLEN:
2342 ifr->ifr_qlen = dev->tx_queue_len;
2343 return 0;
2344
2345 case SIOCSIFTXQLEN:
2346 if (ifr->ifr_qlen < 0)
2347 return -EINVAL;
2348 dev->tx_queue_len = ifr->ifr_qlen;
2349 return 0;
2350
2351 case SIOCSIFNAME:
2352 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2353 return dev_change_name(dev, ifr->ifr_newname);
2354
2355 /*
2356 * Unknown or private ioctl
2357 */
2358
2359 default:
2360 if ((cmd >= SIOCDEVPRIVATE &&
2361 cmd <= SIOCDEVPRIVATE + 15) ||
2362 cmd == SIOCBONDENSLAVE ||
2363 cmd == SIOCBONDRELEASE ||
2364 cmd == SIOCBONDSETHWADDR ||
2365 cmd == SIOCBONDSLAVEINFOQUERY ||
2366 cmd == SIOCBONDINFOQUERY ||
2367 cmd == SIOCBONDCHANGEACTIVE ||
2368 cmd == SIOCGMIIPHY ||
2369 cmd == SIOCGMIIREG ||
2370 cmd == SIOCSMIIREG ||
2371 cmd == SIOCBRADDIF ||
2372 cmd == SIOCBRDELIF ||
2373 cmd == SIOCWANDEV) {
2374 err = -EOPNOTSUPP;
2375 if (dev->do_ioctl) {
2376 if (netif_device_present(dev))
2377 err = dev->do_ioctl(dev, ifr,
2378 cmd);
2379 else
2380 err = -ENODEV;
2381 }
2382 } else
2383 err = -EINVAL;
2384
2385 }
2386 return err;
2387}
2388
2389/*
2390 * This function handles all "interface"-type I/O control requests. The actual
2391 * 'doing' part of this is dev_ifsioc above.
2392 */
2393
2394/**
2395 * dev_ioctl - network device ioctl
2396 * @cmd: command to issue
2397 * @arg: pointer to a struct ifreq in user space
2398 *
2399 * Issue ioctl functions to devices. This is normally called by the
2400 * user space syscall interfaces but can sometimes be useful for
2401 * other purposes. The return value is the return from the syscall if
2402 * positive or a negative errno code on error.
2403 */
2404
2405int dev_ioctl(unsigned int cmd, void __user *arg)
2406{
2407 struct ifreq ifr;
2408 int ret;
2409 char *colon;
2410
2411 /* One special case: SIOCGIFCONF takes ifconf argument
2412 and requires shared lock, because it sleeps writing
2413 to user space.
2414 */
2415
2416 if (cmd == SIOCGIFCONF) {
2417 rtnl_shlock();
2418 ret = dev_ifconf((char __user *) arg);
2419 rtnl_shunlock();
2420 return ret;
2421 }
2422 if (cmd == SIOCGIFNAME)
2423 return dev_ifname((struct ifreq __user *)arg);
2424
2425 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2426 return -EFAULT;
2427
2428 ifr.ifr_name[IFNAMSIZ-1] = 0;
2429
2430 colon = strchr(ifr.ifr_name, ':');
2431 if (colon)
2432 *colon = 0;
2433
2434 /*
2435 * See which interface the caller is talking about.
2436 */
2437
2438 switch (cmd) {
2439 /*
2440 * These ioctl calls:
2441 * - can be done by all.
2442 * - atomic and do not require locking.
2443 * - return a value
2444 */
2445 case SIOCGIFFLAGS:
2446 case SIOCGIFMETRIC:
2447 case SIOCGIFMTU:
2448 case SIOCGIFHWADDR:
2449 case SIOCGIFSLAVE:
2450 case SIOCGIFMAP:
2451 case SIOCGIFINDEX:
2452 case SIOCGIFTXQLEN:
2453 dev_load(ifr.ifr_name);
2454 read_lock(&dev_base_lock);
2455 ret = dev_ifsioc(&ifr, cmd);
2456 read_unlock(&dev_base_lock);
2457 if (!ret) {
2458 if (colon)
2459 *colon = ':';
2460 if (copy_to_user(arg, &ifr,
2461 sizeof(struct ifreq)))
2462 ret = -EFAULT;
2463 }
2464 return ret;
2465
2466 case SIOCETHTOOL:
2467 dev_load(ifr.ifr_name);
2468 rtnl_lock();
2469 ret = dev_ethtool(&ifr);
2470 rtnl_unlock();
2471 if (!ret) {
2472 if (colon)
2473 *colon = ':';
2474 if (copy_to_user(arg, &ifr,
2475 sizeof(struct ifreq)))
2476 ret = -EFAULT;
2477 }
2478 return ret;
2479
2480 /*
2481 * These ioctl calls:
2482 * - require superuser power.
2483 * - require strict serialization.
2484 * - return a value
2485 */
2486 case SIOCGMIIPHY:
2487 case SIOCGMIIREG:
2488 case SIOCSIFNAME:
2489 if (!capable(CAP_NET_ADMIN))
2490 return -EPERM;
2491 dev_load(ifr.ifr_name);
2492 rtnl_lock();
2493 ret = dev_ifsioc(&ifr, cmd);
2494 rtnl_unlock();
2495 if (!ret) {
2496 if (colon)
2497 *colon = ':';
2498 if (copy_to_user(arg, &ifr,
2499 sizeof(struct ifreq)))
2500 ret = -EFAULT;
2501 }
2502 return ret;
2503
2504 /*
2505 * These ioctl calls:
2506 * - require superuser power.
2507 * - require strict serialization.
2508 * - do not return a value
2509 */
2510 case SIOCSIFFLAGS:
2511 case SIOCSIFMETRIC:
2512 case SIOCSIFMTU:
2513 case SIOCSIFMAP:
2514 case SIOCSIFHWADDR:
2515 case SIOCSIFSLAVE:
2516 case SIOCADDMULTI:
2517 case SIOCDELMULTI:
2518 case SIOCSIFHWBROADCAST:
2519 case SIOCSIFTXQLEN:
2520 case SIOCSMIIREG:
2521 case SIOCBONDENSLAVE:
2522 case SIOCBONDRELEASE:
2523 case SIOCBONDSETHWADDR:
2524 case SIOCBONDSLAVEINFOQUERY:
2525 case SIOCBONDINFOQUERY:
2526 case SIOCBONDCHANGEACTIVE:
2527 case SIOCBRADDIF:
2528 case SIOCBRDELIF:
2529 if (!capable(CAP_NET_ADMIN))
2530 return -EPERM;
2531 dev_load(ifr.ifr_name);
2532 rtnl_lock();
2533 ret = dev_ifsioc(&ifr, cmd);
2534 rtnl_unlock();
2535 return ret;
2536
2537 case SIOCGIFMEM:
2538 /* Get the per device memory space. We can add this but
2539 * currently do not support it */
2540 case SIOCSIFMEM:
2541 /* Set the per device memory buffer space.
2542 * Not applicable in our case */
2543 case SIOCSIFLINK:
2544 return -EINVAL;
2545
2546 /*
2547 * Unknown or private ioctl.
2548 */
2549 default:
2550 if (cmd == SIOCWANDEV ||
2551 (cmd >= SIOCDEVPRIVATE &&
2552 cmd <= SIOCDEVPRIVATE + 15)) {
2553 dev_load(ifr.ifr_name);
2554 rtnl_lock();
2555 ret = dev_ifsioc(&ifr, cmd);
2556 rtnl_unlock();
2557 if (!ret && copy_to_user(arg, &ifr,
2558 sizeof(struct ifreq)))
2559 ret = -EFAULT;
2560 return ret;
2561 }
2562#ifdef WIRELESS_EXT
2563 /* Take care of Wireless Extensions */
2564 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2565 /* If command is `set a parameter', or
2566 * `get the encoding parameters', check if
2567 * the user has the right to do it */
2568 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2569 if (!capable(CAP_NET_ADMIN))
2570 return -EPERM;
2571 }
2572 dev_load(ifr.ifr_name);
2573 rtnl_lock();
2574 /* Follow me in net/core/wireless.c */
2575 ret = wireless_process_ioctl(&ifr, cmd);
2576 rtnl_unlock();
2577 if (IW_IS_GET(cmd) &&
2578 copy_to_user(arg, &ifr,
2579 sizeof(struct ifreq)))
2580 ret = -EFAULT;
2581 return ret;
2582 }
2583#endif /* WIRELESS_EXT */
2584 return -EINVAL;
2585 }
2586}
2587
2588
2589/**
2590 * dev_new_index - allocate an ifindex
2591 *
2592 * Returns a suitable unique value for a new device interface
2593 * number. The caller must hold the rtnl semaphore or the
2594 * dev_base_lock to be sure it remains unique.
2595 */
2596static int dev_new_index(void)
2597{
2598 static int ifindex;
2599 for (;;) {
2600 if (++ifindex <= 0)
2601 ifindex = 1;
2602 if (!__dev_get_by_index(ifindex))
2603 return ifindex;
2604 }
2605}
2606
2607static int dev_boot_phase = 1;
2608
2609/* Delayed registration/unregisteration */
2610static DEFINE_SPINLOCK(net_todo_list_lock);
2611static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2612
2613static inline void net_set_todo(struct net_device *dev)
2614{
2615 spin_lock(&net_todo_list_lock);
2616 list_add_tail(&dev->todo_list, &net_todo_list);
2617 spin_unlock(&net_todo_list_lock);
2618}
2619
2620/**
2621 * register_netdevice - register a network device
2622 * @dev: device to register
2623 *
2624 * Take a completed network device structure and add it to the kernel
2625 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2626 * chain. 0 is returned on success. A negative errno code is returned
2627 * on a failure to set up the device, or if the name is a duplicate.
2628 *
2629 * Callers must hold the rtnl semaphore. You may want
2630 * register_netdev() instead of this.
2631 *
2632 * BUGS:
2633 * The locking appears insufficient to guarantee two parallel registers
2634 * will not get the same name.
2635 */
2636
2637int register_netdevice(struct net_device *dev)
2638{
2639 struct hlist_head *head;
2640 struct hlist_node *p;
2641 int ret;
2642
2643 BUG_ON(dev_boot_phase);
2644 ASSERT_RTNL();
2645
2646 /* When net_device's are persistent, this will be fatal. */
2647 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2648
2649 spin_lock_init(&dev->queue_lock);
2650 spin_lock_init(&dev->xmit_lock);
2651 dev->xmit_lock_owner = -1;
2652#ifdef CONFIG_NET_CLS_ACT
2653 spin_lock_init(&dev->ingress_lock);
2654#endif
2655
2656 ret = alloc_divert_blk(dev);
2657 if (ret)
2658 goto out;
2659
2660 dev->iflink = -1;
2661
2662 /* Init, if this function is available */
2663 if (dev->init) {
2664 ret = dev->init(dev);
2665 if (ret) {
2666 if (ret > 0)
2667 ret = -EIO;
2668 goto out_err;
2669 }
2670 }
2671
2672 if (!dev_valid_name(dev->name)) {
2673 ret = -EINVAL;
2674 goto out_err;
2675 }
2676
2677 dev->ifindex = dev_new_index();
2678 if (dev->iflink == -1)
2679 dev->iflink = dev->ifindex;
2680
2681 /* Check for existence of name */
2682 head = dev_name_hash(dev->name);
2683 hlist_for_each(p, head) {
2684 struct net_device *d
2685 = hlist_entry(p, struct net_device, name_hlist);
2686 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2687 ret = -EEXIST;
2688 goto out_err;
2689 }
2690 }
2691
2692 /* Fix illegal SG+CSUM combinations. */
2693 if ((dev->features & NETIF_F_SG) &&
2694 !(dev->features & (NETIF_F_IP_CSUM |
2695 NETIF_F_NO_CSUM |
2696 NETIF_F_HW_CSUM))) {
2697 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2698 dev->name);
2699 dev->features &= ~NETIF_F_SG;
2700 }
2701
2702 /* TSO requires that SG is present as well. */
2703 if ((dev->features & NETIF_F_TSO) &&
2704 !(dev->features & NETIF_F_SG)) {
2705 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2706 dev->name);
2707 dev->features &= ~NETIF_F_TSO;
2708 }
2709
2710 /*
2711 * nil rebuild_header routine,
2712 * that should be never called and used as just bug trap.
2713 */
2714
2715 if (!dev->rebuild_header)
2716 dev->rebuild_header = default_rebuild_header;
2717
2718 /*
2719 * Default initial state at registry is that the
2720 * device is present.
2721 */
2722
2723 set_bit(__LINK_STATE_PRESENT, &dev->state);
2724
2725 dev->next = NULL;
2726 dev_init_scheduler(dev);
2727 write_lock_bh(&dev_base_lock);
2728 *dev_tail = dev;
2729 dev_tail = &dev->next;
2730 hlist_add_head(&dev->name_hlist, head);
2731 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2732 dev_hold(dev);
2733 dev->reg_state = NETREG_REGISTERING;
2734 write_unlock_bh(&dev_base_lock);
2735
2736 /* Notify protocols, that a new device appeared. */
2737 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2738
2739 /* Finish registration after unlock */
2740 net_set_todo(dev);
2741 ret = 0;
2742
2743out:
2744 return ret;
2745out_err:
2746 free_divert_blk(dev);
2747 goto out;
2748}
2749
2750/**
2751 * register_netdev - register a network device
2752 * @dev: device to register
2753 *
2754 * Take a completed network device structure and add it to the kernel
2755 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2756 * chain. 0 is returned on success. A negative errno code is returned
2757 * on a failure to set up the device, or if the name is a duplicate.
2758 *
2759 * This is a wrapper around register_netdev that takes the rtnl semaphore
2760 * and expands the device name if you passed a format string to
2761 * alloc_netdev.
2762 */
2763int register_netdev(struct net_device *dev)
2764{
2765 int err;
2766
2767 rtnl_lock();
2768
2769 /*
2770 * If the name is a format string the caller wants us to do a
2771 * name allocation.
2772 */
2773 if (strchr(dev->name, '%')) {
2774 err = dev_alloc_name(dev, dev->name);
2775 if (err < 0)
2776 goto out;
2777 }
2778
2779 /*
2780 * Back compatibility hook. Kill this one in 2.5
2781 */
2782 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2783 err = dev_alloc_name(dev, "eth%d");
2784 if (err < 0)
2785 goto out;
2786 }
2787
2788 err = register_netdevice(dev);
2789out:
2790 rtnl_unlock();
2791 return err;
2792}
2793EXPORT_SYMBOL(register_netdev);
2794
2795/*
2796 * netdev_wait_allrefs - wait until all references are gone.
2797 *
2798 * This is called when unregistering network devices.
2799 *
2800 * Any protocol or device that holds a reference should register
2801 * for netdevice notification, and cleanup and put back the
2802 * reference if they receive an UNREGISTER event.
2803 * We can get stuck here if buggy protocols don't correctly
2804 * call dev_put.
2805 */
2806static void netdev_wait_allrefs(struct net_device *dev)
2807{
2808 unsigned long rebroadcast_time, warning_time;
2809
2810 rebroadcast_time = warning_time = jiffies;
2811 while (atomic_read(&dev->refcnt) != 0) {
2812 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2813 rtnl_shlock();
2814
2815 /* Rebroadcast unregister notification */
2816 notifier_call_chain(&netdev_chain,
2817 NETDEV_UNREGISTER, dev);
2818
2819 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2820 &dev->state)) {
2821 /* We must not have linkwatch events
2822 * pending on unregister. If this
2823 * happens, we simply run the queue
2824 * unscheduled, resulting in a noop
2825 * for this device.
2826 */
2827 linkwatch_run_queue();
2828 }
2829
2830 rtnl_shunlock();
2831
2832 rebroadcast_time = jiffies;
2833 }
2834
2835 msleep(250);
2836
2837 if (time_after(jiffies, warning_time + 10 * HZ)) {
2838 printk(KERN_EMERG "unregister_netdevice: "
2839 "waiting for %s to become free. Usage "
2840 "count = %d\n",
2841 dev->name, atomic_read(&dev->refcnt));
2842 warning_time = jiffies;
2843 }
2844 }
2845}
2846
2847/* The sequence is:
2848 *
2849 * rtnl_lock();
2850 * ...
2851 * register_netdevice(x1);
2852 * register_netdevice(x2);
2853 * ...
2854 * unregister_netdevice(y1);
2855 * unregister_netdevice(y2);
2856 * ...
2857 * rtnl_unlock();
2858 * free_netdev(y1);
2859 * free_netdev(y2);
2860 *
2861 * We are invoked by rtnl_unlock() after it drops the semaphore.
2862 * This allows us to deal with problems:
2863 * 1) We can create/delete sysfs objects which invoke hotplug
2864 * without deadlocking with linkwatch via keventd.
2865 * 2) Since we run with the RTNL semaphore not held, we can sleep
2866 * safely in order to wait for the netdev refcnt to drop to zero.
2867 */
2868static DECLARE_MUTEX(net_todo_run_mutex);
2869void netdev_run_todo(void)
2870{
2871 struct list_head list = LIST_HEAD_INIT(list);
2872 int err;
2873
2874
2875 /* Need to guard against multiple cpu's getting out of order. */
2876 down(&net_todo_run_mutex);
2877
2878 /* Not safe to do outside the semaphore. We must not return
2879 * until all unregister events invoked by the local processor
2880 * have been completed (either by this todo run, or one on
2881 * another cpu).
2882 */
2883 if (list_empty(&net_todo_list))
2884 goto out;
2885
2886 /* Snapshot list, allow later requests */
2887 spin_lock(&net_todo_list_lock);
2888 list_splice_init(&net_todo_list, &list);
2889 spin_unlock(&net_todo_list_lock);
2890
2891 while (!list_empty(&list)) {
2892 struct net_device *dev
2893 = list_entry(list.next, struct net_device, todo_list);
2894 list_del(&dev->todo_list);
2895
2896 switch(dev->reg_state) {
2897 case NETREG_REGISTERING:
2898 err = netdev_register_sysfs(dev);
2899 if (err)
2900 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2901 dev->name, err);
2902 dev->reg_state = NETREG_REGISTERED;
2903 break;
2904
2905 case NETREG_UNREGISTERING:
2906 netdev_unregister_sysfs(dev);
2907 dev->reg_state = NETREG_UNREGISTERED;
2908
2909 netdev_wait_allrefs(dev);
2910
2911 /* paranoia */
2912 BUG_ON(atomic_read(&dev->refcnt));
2913 BUG_TRAP(!dev->ip_ptr);
2914 BUG_TRAP(!dev->ip6_ptr);
2915 BUG_TRAP(!dev->dn_ptr);
2916
2917
2918 /* It must be the very last action,
2919 * after this 'dev' may point to freed up memory.
2920 */
2921 if (dev->destructor)
2922 dev->destructor(dev);
2923 break;
2924
2925 default:
2926 printk(KERN_ERR "network todo '%s' but state %d\n",
2927 dev->name, dev->reg_state);
2928 break;
2929 }
2930 }
2931
2932out:
2933 up(&net_todo_run_mutex);
2934}
2935
2936/**
2937 * alloc_netdev - allocate network device
2938 * @sizeof_priv: size of private data to allocate space for
2939 * @name: device name format string
2940 * @setup: callback to initialize device
2941 *
2942 * Allocates a struct net_device with private data area for driver use
2943 * and performs basic initialization.
2944 */
2945struct net_device *alloc_netdev(int sizeof_priv, const char *name,
2946 void (*setup)(struct net_device *))
2947{
2948 void *p;
2949 struct net_device *dev;
2950 int alloc_size;
2951
2952 /* ensure 32-byte alignment of both the device and private area */
2953 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
2954 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
2955
2956 p = kmalloc(alloc_size, GFP_KERNEL);
2957 if (!p) {
2958 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
2959 return NULL;
2960 }
2961 memset(p, 0, alloc_size);
2962
2963 dev = (struct net_device *)
2964 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
2965 dev->padded = (char *)dev - (char *)p;
2966
2967 if (sizeof_priv)
2968 dev->priv = netdev_priv(dev);
2969
2970 setup(dev);
2971 strcpy(dev->name, name);
2972 return dev;
2973}
2974EXPORT_SYMBOL(alloc_netdev);
2975
2976/**
2977 * free_netdev - free network device
2978 * @dev: device
2979 *
2980 * This function does the last stage of destroying an allocated device
2981 * interface. The reference to the device object is released.
2982 * If this is the last reference then it will be freed.
2983 */
2984void free_netdev(struct net_device *dev)
2985{
2986#ifdef CONFIG_SYSFS
2987 /* Compatiablity with error handling in drivers */
2988 if (dev->reg_state == NETREG_UNINITIALIZED) {
2989 kfree((char *)dev - dev->padded);
2990 return;
2991 }
2992
2993 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
2994 dev->reg_state = NETREG_RELEASED;
2995
2996 /* will free via class release */
2997 class_device_put(&dev->class_dev);
2998#else
2999 kfree((char *)dev - dev->padded);
3000#endif
3001}
3002
3003/* Synchronize with packet receive processing. */
3004void synchronize_net(void)
3005{
3006 might_sleep();
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07003007 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008}
3009
3010/**
3011 * unregister_netdevice - remove device from the kernel
3012 * @dev: device
3013 *
3014 * This function shuts down a device interface and removes it
3015 * from the kernel tables. On success 0 is returned, on a failure
3016 * a negative errno code is returned.
3017 *
3018 * Callers must hold the rtnl semaphore. You may want
3019 * unregister_netdev() instead of this.
3020 */
3021
3022int unregister_netdevice(struct net_device *dev)
3023{
3024 struct net_device *d, **dp;
3025
3026 BUG_ON(dev_boot_phase);
3027 ASSERT_RTNL();
3028
3029 /* Some devices call without registering for initialization unwind. */
3030 if (dev->reg_state == NETREG_UNINITIALIZED) {
3031 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3032 "was registered\n", dev->name, dev);
3033 return -ENODEV;
3034 }
3035
3036 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3037
3038 /* If device is running, close it first. */
3039 if (dev->flags & IFF_UP)
3040 dev_close(dev);
3041
3042 /* And unlink it from device chain. */
3043 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3044 if (d == dev) {
3045 write_lock_bh(&dev_base_lock);
3046 hlist_del(&dev->name_hlist);
3047 hlist_del(&dev->index_hlist);
3048 if (dev_tail == &dev->next)
3049 dev_tail = dp;
3050 *dp = d->next;
3051 write_unlock_bh(&dev_base_lock);
3052 break;
3053 }
3054 }
3055 if (!d) {
3056 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3057 dev->name);
3058 return -ENODEV;
3059 }
3060
3061 dev->reg_state = NETREG_UNREGISTERING;
3062
3063 synchronize_net();
3064
3065 /* Shutdown queueing discipline. */
3066 dev_shutdown(dev);
3067
3068
3069 /* Notify protocols, that we are about to destroy
3070 this device. They should clean all the things.
3071 */
3072 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3073
3074 /*
3075 * Flush the multicast chain
3076 */
3077 dev_mc_discard(dev);
3078
3079 if (dev->uninit)
3080 dev->uninit(dev);
3081
3082 /* Notifier chain MUST detach us from master device. */
3083 BUG_TRAP(!dev->master);
3084
3085 free_divert_blk(dev);
3086
3087 /* Finish processing unregister after unlock */
3088 net_set_todo(dev);
3089
3090 synchronize_net();
3091
3092 dev_put(dev);
3093 return 0;
3094}
3095
3096/**
3097 * unregister_netdev - remove device from the kernel
3098 * @dev: device
3099 *
3100 * This function shuts down a device interface and removes it
3101 * from the kernel tables. On success 0 is returned, on a failure
3102 * a negative errno code is returned.
3103 *
3104 * This is just a wrapper for unregister_netdevice that takes
3105 * the rtnl semaphore. In general you want to use this and not
3106 * unregister_netdevice.
3107 */
3108void unregister_netdev(struct net_device *dev)
3109{
3110 rtnl_lock();
3111 unregister_netdevice(dev);
3112 rtnl_unlock();
3113}
3114
3115EXPORT_SYMBOL(unregister_netdev);
3116
3117#ifdef CONFIG_HOTPLUG_CPU
3118static int dev_cpu_callback(struct notifier_block *nfb,
3119 unsigned long action,
3120 void *ocpu)
3121{
3122 struct sk_buff **list_skb;
3123 struct net_device **list_net;
3124 struct sk_buff *skb;
3125 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3126 struct softnet_data *sd, *oldsd;
3127
3128 if (action != CPU_DEAD)
3129 return NOTIFY_OK;
3130
3131 local_irq_disable();
3132 cpu = smp_processor_id();
3133 sd = &per_cpu(softnet_data, cpu);
3134 oldsd = &per_cpu(softnet_data, oldcpu);
3135
3136 /* Find end of our completion_queue. */
3137 list_skb = &sd->completion_queue;
3138 while (*list_skb)
3139 list_skb = &(*list_skb)->next;
3140 /* Append completion queue from offline CPU. */
3141 *list_skb = oldsd->completion_queue;
3142 oldsd->completion_queue = NULL;
3143
3144 /* Find end of our output_queue. */
3145 list_net = &sd->output_queue;
3146 while (*list_net)
3147 list_net = &(*list_net)->next_sched;
3148 /* Append output queue from offline CPU. */
3149 *list_net = oldsd->output_queue;
3150 oldsd->output_queue = NULL;
3151
3152 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3153 local_irq_enable();
3154
3155 /* Process offline CPU's input_pkt_queue */
3156 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3157 netif_rx(skb);
3158
3159 return NOTIFY_OK;
3160}
3161#endif /* CONFIG_HOTPLUG_CPU */
3162
3163
3164/*
3165 * Initialize the DEV module. At boot time this walks the device list and
3166 * unhooks any devices that fail to initialise (normally hardware not
3167 * present) and leaves us with a valid list of present and active devices.
3168 *
3169 */
3170
3171/*
3172 * This is called single threaded during boot, so no need
3173 * to take the rtnl semaphore.
3174 */
3175static int __init net_dev_init(void)
3176{
3177 int i, rc = -ENOMEM;
3178
3179 BUG_ON(!dev_boot_phase);
3180
3181 net_random_init();
3182
3183 if (dev_proc_init())
3184 goto out;
3185
3186 if (netdev_sysfs_init())
3187 goto out;
3188
3189 INIT_LIST_HEAD(&ptype_all);
3190 for (i = 0; i < 16; i++)
3191 INIT_LIST_HEAD(&ptype_base[i]);
3192
3193 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3194 INIT_HLIST_HEAD(&dev_name_head[i]);
3195
3196 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3197 INIT_HLIST_HEAD(&dev_index_head[i]);
3198
3199 /*
3200 * Initialise the packet receive queues.
3201 */
3202
3203 for (i = 0; i < NR_CPUS; i++) {
3204 struct softnet_data *queue;
3205
3206 queue = &per_cpu(softnet_data, i);
3207 skb_queue_head_init(&queue->input_pkt_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208 queue->completion_queue = NULL;
3209 INIT_LIST_HEAD(&queue->poll_list);
3210 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3211 queue->backlog_dev.weight = weight_p;
3212 queue->backlog_dev.poll = process_backlog;
3213 atomic_set(&queue->backlog_dev.refcnt, 1);
3214 }
3215
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 dev_boot_phase = 0;
3217
3218 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3219 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3220
3221 hotcpu_notifier(dev_cpu_callback, 0);
3222 dst_init();
3223 dev_mcast_init();
3224 rc = 0;
3225out:
3226 return rc;
3227}
3228
3229subsys_initcall(net_dev_init);
3230
3231EXPORT_SYMBOL(__dev_get_by_index);
3232EXPORT_SYMBOL(__dev_get_by_name);
3233EXPORT_SYMBOL(__dev_remove_pack);
3234EXPORT_SYMBOL(__skb_linearize);
3235EXPORT_SYMBOL(dev_add_pack);
3236EXPORT_SYMBOL(dev_alloc_name);
3237EXPORT_SYMBOL(dev_close);
3238EXPORT_SYMBOL(dev_get_by_flags);
3239EXPORT_SYMBOL(dev_get_by_index);
3240EXPORT_SYMBOL(dev_get_by_name);
3241EXPORT_SYMBOL(dev_ioctl);
3242EXPORT_SYMBOL(dev_open);
3243EXPORT_SYMBOL(dev_queue_xmit);
3244EXPORT_SYMBOL(dev_remove_pack);
3245EXPORT_SYMBOL(dev_set_allmulti);
3246EXPORT_SYMBOL(dev_set_promiscuity);
3247EXPORT_SYMBOL(dev_change_flags);
3248EXPORT_SYMBOL(dev_set_mtu);
3249EXPORT_SYMBOL(dev_set_mac_address);
3250EXPORT_SYMBOL(free_netdev);
3251EXPORT_SYMBOL(netdev_boot_setup_check);
3252EXPORT_SYMBOL(netdev_set_master);
3253EXPORT_SYMBOL(netdev_state_change);
3254EXPORT_SYMBOL(netif_receive_skb);
3255EXPORT_SYMBOL(netif_rx);
3256EXPORT_SYMBOL(register_gifconf);
3257EXPORT_SYMBOL(register_netdevice);
3258EXPORT_SYMBOL(register_netdevice_notifier);
3259EXPORT_SYMBOL(skb_checksum_help);
3260EXPORT_SYMBOL(synchronize_net);
3261EXPORT_SYMBOL(unregister_netdevice);
3262EXPORT_SYMBOL(unregister_netdevice_notifier);
3263EXPORT_SYMBOL(net_enable_timestamp);
3264EXPORT_SYMBOL(net_disable_timestamp);
3265EXPORT_SYMBOL(dev_get_flags);
3266
3267#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3268EXPORT_SYMBOL(br_handle_frame_hook);
3269EXPORT_SYMBOL(br_fdb_get_hook);
3270EXPORT_SYMBOL(br_fdb_put_hook);
3271#endif
3272
3273#ifdef CONFIG_KMOD
3274EXPORT_SYMBOL(dev_load);
3275#endif
3276
3277EXPORT_PER_CPU_SYMBOL(softnet_data);