blob: a3ed53cc4af863f5c75b8dfc5b768c1d0148ef3b [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. */
David S. Miller6192b542005-07-28 12:12:58 -0700904 msleep(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 }
906
907 /*
908 * Call the device specific close. This cannot fail.
909 * Only if device is UP
910 *
911 * We allow it to be called even after a DETACH hot-plug
912 * event.
913 */
914 if (dev->stop)
915 dev->stop(dev);
916
917 /*
918 * Device is now down.
919 */
920
921 dev->flags &= ~IFF_UP;
922
923 /*
924 * Tell people we are down
925 */
926 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
927
928 return 0;
929}
930
931
932/*
933 * Device change register/unregister. These are not inline or static
934 * as we export them to the world.
935 */
936
937/**
938 * register_netdevice_notifier - register a network notifier block
939 * @nb: notifier
940 *
941 * Register a notifier to be called when network device events occur.
942 * The notifier passed is linked into the kernel structures and must
943 * not be reused until it has been unregistered. A negative errno code
944 * is returned on a failure.
945 *
946 * When registered all registration and up events are replayed
947 * to the new notifier to allow device to have a race free
948 * view of the network device list.
949 */
950
951int register_netdevice_notifier(struct notifier_block *nb)
952{
953 struct net_device *dev;
954 int err;
955
956 rtnl_lock();
957 err = notifier_chain_register(&netdev_chain, nb);
958 if (!err) {
959 for (dev = dev_base; dev; dev = dev->next) {
960 nb->notifier_call(nb, NETDEV_REGISTER, dev);
961
962 if (dev->flags & IFF_UP)
963 nb->notifier_call(nb, NETDEV_UP, dev);
964 }
965 }
966 rtnl_unlock();
967 return err;
968}
969
970/**
971 * unregister_netdevice_notifier - unregister a network notifier block
972 * @nb: notifier
973 *
974 * Unregister a notifier previously registered by
975 * register_netdevice_notifier(). The notifier is unlinked into the
976 * kernel structures and may then be reused. A negative errno code
977 * is returned on a failure.
978 */
979
980int unregister_netdevice_notifier(struct notifier_block *nb)
981{
982 return notifier_chain_unregister(&netdev_chain, nb);
983}
984
985/**
986 * call_netdevice_notifiers - call all network notifier blocks
987 * @val: value passed unmodified to notifier function
988 * @v: pointer passed unmodified to notifier function
989 *
990 * Call all network notifier blocks. Parameters and return value
991 * are as for notifier_call_chain().
992 */
993
994int call_netdevice_notifiers(unsigned long val, void *v)
995{
996 return notifier_call_chain(&netdev_chain, val, v);
997}
998
999/* When > 0 there are consumers of rx skb time stamps */
1000static atomic_t netstamp_needed = ATOMIC_INIT(0);
1001
1002void net_enable_timestamp(void)
1003{
1004 atomic_inc(&netstamp_needed);
1005}
1006
1007void net_disable_timestamp(void)
1008{
1009 atomic_dec(&netstamp_needed);
1010}
1011
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001012void __net_timestamp(struct sk_buff *skb)
1013{
1014 struct timeval tv;
1015
1016 do_gettimeofday(&tv);
1017 skb_set_timestamp(skb, &tv);
1018}
1019EXPORT_SYMBOL(__net_timestamp);
1020
1021static inline void net_timestamp(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022{
1023 if (atomic_read(&netstamp_needed))
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001024 __net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 else {
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001026 skb->tstamp.off_sec = 0;
1027 skb->tstamp.off_usec = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 }
1029}
1030
1031/*
1032 * Support routine. Sends outgoing frames to any network
1033 * taps currently in use.
1034 */
1035
1036void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1037{
1038 struct packet_type *ptype;
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001039
1040 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 rcu_read_lock();
1043 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1044 /* Never send packets back to the socket
1045 * they originated from - MvS (miquels@drinkel.ow.org)
1046 */
1047 if ((ptype->dev == dev || !ptype->dev) &&
1048 (ptype->af_packet_priv == NULL ||
1049 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1050 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1051 if (!skb2)
1052 break;
1053
1054 /* skb->nh should be correctly
1055 set by sender, so that the second statement is
1056 just protection against buggy protocols.
1057 */
1058 skb2->mac.raw = skb2->data;
1059
1060 if (skb2->nh.raw < skb2->data ||
1061 skb2->nh.raw > skb2->tail) {
1062 if (net_ratelimit())
1063 printk(KERN_CRIT "protocol %04x is "
1064 "buggy, dev %s\n",
1065 skb2->protocol, dev->name);
1066 skb2->nh.raw = skb2->data;
1067 }
1068
1069 skb2->h.raw = skb2->nh.raw;
1070 skb2->pkt_type = PACKET_OUTGOING;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001071 ptype->func(skb2, skb->dev, ptype, skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
1073 }
1074 rcu_read_unlock();
1075}
1076
1077/*
1078 * Invalidate hardware checksum when packet is to be mangled, and
1079 * complete checksum manually on outgoing path.
1080 */
1081int skb_checksum_help(struct sk_buff *skb, int inward)
1082{
1083 unsigned int csum;
1084 int ret = 0, offset = skb->h.raw - skb->data;
1085
1086 if (inward) {
1087 skb->ip_summed = CHECKSUM_NONE;
1088 goto out;
1089 }
1090
1091 if (skb_cloned(skb)) {
1092 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1093 if (ret)
1094 goto out;
1095 }
1096
1097 if (offset > (int)skb->len)
1098 BUG();
1099 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1100
1101 offset = skb->tail - skb->h.raw;
1102 if (offset <= 0)
1103 BUG();
1104 if (skb->csum + 2 > offset)
1105 BUG();
1106
1107 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1108 skb->ip_summed = CHECKSUM_NONE;
1109out:
1110 return ret;
1111}
1112
1113#ifdef CONFIG_HIGHMEM
1114/* Actually, we should eliminate this check as soon as we know, that:
1115 * 1. IOMMU is present and allows to map all the memory.
1116 * 2. No high memory really exists on this machine.
1117 */
1118
1119static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1120{
1121 int i;
1122
1123 if (dev->features & NETIF_F_HIGHDMA)
1124 return 0;
1125
1126 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1127 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1128 return 1;
1129
1130 return 0;
1131}
1132#else
1133#define illegal_highdma(dev, skb) (0)
1134#endif
1135
1136extern void skb_release_data(struct sk_buff *);
1137
1138/* Keep head the same: replace data */
Victor Fusco86a76ca2005-07-08 14:57:47 -07001139int __skb_linearize(struct sk_buff *skb, unsigned int __nocast gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140{
1141 unsigned int size;
1142 u8 *data;
1143 long offset;
1144 struct skb_shared_info *ninfo;
1145 int headerlen = skb->data - skb->head;
1146 int expand = (skb->tail + skb->data_len) - skb->end;
1147
1148 if (skb_shared(skb))
1149 BUG();
1150
1151 if (expand <= 0)
1152 expand = 0;
1153
1154 size = skb->end - skb->head + expand;
1155 size = SKB_DATA_ALIGN(size);
1156 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1157 if (!data)
1158 return -ENOMEM;
1159
1160 /* Copy entire thing */
1161 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1162 BUG();
1163
1164 /* Set up shinfo */
1165 ninfo = (struct skb_shared_info*)(data + size);
1166 atomic_set(&ninfo->dataref, 1);
1167 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1168 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1169 ninfo->nr_frags = 0;
1170 ninfo->frag_list = NULL;
1171
1172 /* Offset between the two in bytes */
1173 offset = data - skb->head;
1174
1175 /* Free old data. */
1176 skb_release_data(skb);
1177
1178 skb->head = data;
1179 skb->end = data + size;
1180
1181 /* Set up new pointers */
1182 skb->h.raw += offset;
1183 skb->nh.raw += offset;
1184 skb->mac.raw += offset;
1185 skb->tail += offset;
1186 skb->data += offset;
1187
1188 /* We are no longer a clone, even if we were. */
1189 skb->cloned = 0;
1190
1191 skb->tail += skb->data_len;
1192 skb->data_len = 0;
1193 return 0;
1194}
1195
1196#define HARD_TX_LOCK(dev, cpu) { \
1197 if ((dev->features & NETIF_F_LLTX) == 0) { \
1198 spin_lock(&dev->xmit_lock); \
1199 dev->xmit_lock_owner = cpu; \
1200 } \
1201}
1202
1203#define HARD_TX_UNLOCK(dev) { \
1204 if ((dev->features & NETIF_F_LLTX) == 0) { \
1205 dev->xmit_lock_owner = -1; \
1206 spin_unlock(&dev->xmit_lock); \
1207 } \
1208}
1209
1210/**
1211 * dev_queue_xmit - transmit a buffer
1212 * @skb: buffer to transmit
1213 *
1214 * Queue a buffer for transmission to a network device. The caller must
1215 * have set the device and priority and built the buffer before calling
1216 * this function. The function can be called from an interrupt.
1217 *
1218 * A negative errno code is returned on a failure. A success does not
1219 * guarantee the frame will be transmitted as it may be dropped due
1220 * to congestion or traffic shaping.
Ben Greearaf191362005-04-24 20:12:36 -07001221 *
1222 * -----------------------------------------------------------------------------------
1223 * I notice this method can also return errors from the queue disciplines,
1224 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1225 * be positive.
1226 *
1227 * Regardless of the return value, the skb is consumed, so it is currently
1228 * difficult to retry a send to this method. (You can bump the ref count
1229 * before sending to hold a reference for retry if you are careful.)
1230 *
1231 * When calling this method, interrupts MUST be enabled. This is because
1232 * the BH enable code must have IRQs enabled so that it will not deadlock.
1233 * --BLG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 */
1235
1236int dev_queue_xmit(struct sk_buff *skb)
1237{
1238 struct net_device *dev = skb->dev;
1239 struct Qdisc *q;
1240 int rc = -ENOMEM;
1241
1242 if (skb_shinfo(skb)->frag_list &&
1243 !(dev->features & NETIF_F_FRAGLIST) &&
1244 __skb_linearize(skb, GFP_ATOMIC))
1245 goto out_kfree_skb;
1246
1247 /* Fragmented skb is linearized if device does not support SG,
1248 * or if at least one of fragments is in highmem and device
1249 * does not support DMA from it.
1250 */
1251 if (skb_shinfo(skb)->nr_frags &&
1252 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1253 __skb_linearize(skb, GFP_ATOMIC))
1254 goto out_kfree_skb;
1255
1256 /* If packet is not checksummed and device does not support
1257 * checksumming for this protocol, complete checksumming here.
1258 */
1259 if (skb->ip_summed == CHECKSUM_HW &&
1260 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1261 (!(dev->features & NETIF_F_IP_CSUM) ||
1262 skb->protocol != htons(ETH_P_IP))))
1263 if (skb_checksum_help(skb, 0))
1264 goto out_kfree_skb;
1265
1266 /* Disable soft irqs for various locks below. Also
1267 * stops preemption for RCU.
1268 */
1269 local_bh_disable();
1270
1271 /* Updates of qdisc are serialized by queue_lock.
1272 * The struct Qdisc which is pointed to by qdisc is now a
1273 * rcu structure - it may be accessed without acquiring
1274 * a lock (but the structure may be stale.) The freeing of the
1275 * qdisc will be deferred until it's known that there are no
1276 * more references to it.
1277 *
1278 * If the qdisc has an enqueue function, we still need to
1279 * hold the queue_lock before calling it, since queue_lock
1280 * also serializes access to the device queue.
1281 */
1282
1283 q = rcu_dereference(dev->qdisc);
1284#ifdef CONFIG_NET_CLS_ACT
1285 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1286#endif
1287 if (q->enqueue) {
1288 /* Grab device queue */
1289 spin_lock(&dev->queue_lock);
1290
1291 rc = q->enqueue(skb, q);
1292
1293 qdisc_run(dev);
1294
1295 spin_unlock(&dev->queue_lock);
1296 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1297 goto out;
1298 }
1299
1300 /* The device has no queue. Common case for software devices:
1301 loopback, all the sorts of tunnels...
1302
1303 Really, it is unlikely that xmit_lock protection is necessary here.
1304 (f.e. loopback and IP tunnels are clean ignoring statistics
1305 counters.)
1306 However, it is possible, that they rely on protection
1307 made by us here.
1308
1309 Check this and shot the lock. It is not prone from deadlocks.
1310 Either shot noqueue qdisc, it is even simpler 8)
1311 */
1312 if (dev->flags & IFF_UP) {
1313 int cpu = smp_processor_id(); /* ok because BHs are off */
1314
1315 if (dev->xmit_lock_owner != cpu) {
1316
1317 HARD_TX_LOCK(dev, cpu);
1318
1319 if (!netif_queue_stopped(dev)) {
1320 if (netdev_nit)
1321 dev_queue_xmit_nit(skb, dev);
1322
1323 rc = 0;
1324 if (!dev->hard_start_xmit(skb, dev)) {
1325 HARD_TX_UNLOCK(dev);
1326 goto out;
1327 }
1328 }
1329 HARD_TX_UNLOCK(dev);
1330 if (net_ratelimit())
1331 printk(KERN_CRIT "Virtual device %s asks to "
1332 "queue packet!\n", dev->name);
1333 } else {
1334 /* Recursion is detected! It is possible,
1335 * unfortunately */
1336 if (net_ratelimit())
1337 printk(KERN_CRIT "Dead loop on virtual device "
1338 "%s, fix it urgently!\n", dev->name);
1339 }
1340 }
1341
1342 rc = -ENETDOWN;
1343 local_bh_enable();
1344
1345out_kfree_skb:
1346 kfree_skb(skb);
1347 return rc;
1348out:
1349 local_bh_enable();
1350 return rc;
1351}
1352
1353
1354/*=======================================================================
1355 Receiver routines
1356 =======================================================================*/
1357
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001358int netdev_max_backlog = 1000;
1359int netdev_budget = 300;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360int weight_p = 64; /* old backlog weight */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
1362DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1363
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365/**
1366 * netif_rx - post buffer to the network code
1367 * @skb: buffer to post
1368 *
1369 * This function receives a packet from a device driver and queues it for
1370 * the upper (protocol) levels to process. It always succeeds. The buffer
1371 * may be dropped during processing for congestion control or by the
1372 * protocol layers.
1373 *
1374 * return values:
1375 * NET_RX_SUCCESS (no congestion)
1376 * NET_RX_CN_LOW (low congestion)
1377 * NET_RX_CN_MOD (moderate congestion)
1378 * NET_RX_CN_HIGH (high congestion)
1379 * NET_RX_DROP (packet was dropped)
1380 *
1381 */
1382
1383int netif_rx(struct sk_buff *skb)
1384{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 struct softnet_data *queue;
1386 unsigned long flags;
1387
1388 /* if netpoll wants it, pretend we never saw it */
1389 if (netpoll_rx(skb))
1390 return NET_RX_DROP;
1391
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001392 if (!skb->tstamp.off_sec)
1393 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
1395 /*
1396 * The code is rearranged so that the path is the most
1397 * short when CPU is congested, but is still operating.
1398 */
1399 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 queue = &__get_cpu_var(softnet_data);
1401
1402 __get_cpu_var(netdev_rx_stat).total++;
1403 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1404 if (queue->input_pkt_queue.qlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405enqueue:
1406 dev_hold(skb->dev);
1407 __skb_queue_tail(&queue->input_pkt_queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 local_irq_restore(flags);
Stephen Hemminger34008d82005-06-23 20:10:00 -07001409 return NET_RX_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 }
1411
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 netif_rx_schedule(&queue->backlog_dev);
1413 goto enqueue;
1414 }
1415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 __get_cpu_var(netdev_rx_stat).dropped++;
1417 local_irq_restore(flags);
1418
1419 kfree_skb(skb);
1420 return NET_RX_DROP;
1421}
1422
1423int netif_rx_ni(struct sk_buff *skb)
1424{
1425 int err;
1426
1427 preempt_disable();
1428 err = netif_rx(skb);
1429 if (local_softirq_pending())
1430 do_softirq();
1431 preempt_enable();
1432
1433 return err;
1434}
1435
1436EXPORT_SYMBOL(netif_rx_ni);
1437
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001438static inline struct net_device *skb_bond(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439{
1440 struct net_device *dev = skb->dev;
1441
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001442 if (dev->master)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 skb->dev = dev->master;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001444
1445 return dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446}
1447
1448static void net_tx_action(struct softirq_action *h)
1449{
1450 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1451
1452 if (sd->completion_queue) {
1453 struct sk_buff *clist;
1454
1455 local_irq_disable();
1456 clist = sd->completion_queue;
1457 sd->completion_queue = NULL;
1458 local_irq_enable();
1459
1460 while (clist) {
1461 struct sk_buff *skb = clist;
1462 clist = clist->next;
1463
1464 BUG_TRAP(!atomic_read(&skb->users));
1465 __kfree_skb(skb);
1466 }
1467 }
1468
1469 if (sd->output_queue) {
1470 struct net_device *head;
1471
1472 local_irq_disable();
1473 head = sd->output_queue;
1474 sd->output_queue = NULL;
1475 local_irq_enable();
1476
1477 while (head) {
1478 struct net_device *dev = head;
1479 head = head->next_sched;
1480
1481 smp_mb__before_clear_bit();
1482 clear_bit(__LINK_STATE_SCHED, &dev->state);
1483
1484 if (spin_trylock(&dev->queue_lock)) {
1485 qdisc_run(dev);
1486 spin_unlock(&dev->queue_lock);
1487 } else {
1488 netif_schedule(dev);
1489 }
1490 }
1491 }
1492}
1493
1494static __inline__ int deliver_skb(struct sk_buff *skb,
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001495 struct packet_type *pt_prev,
1496 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497{
1498 atomic_inc(&skb->users);
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001499 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500}
1501
1502#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1503int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1504struct net_bridge;
1505struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1506 unsigned char *addr);
1507void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1508
1509static __inline__ int handle_bridge(struct sk_buff **pskb,
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001510 struct packet_type **pt_prev, int *ret,
1511 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512{
1513 struct net_bridge_port *port;
1514
1515 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1516 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1517 return 0;
1518
1519 if (*pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001520 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 *pt_prev = NULL;
1522 }
1523
1524 return br_handle_frame_hook(port, pskb);
1525}
1526#else
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001527#define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528#endif
1529
1530#ifdef CONFIG_NET_CLS_ACT
1531/* TODO: Maybe we should just force sch_ingress to be compiled in
1532 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1533 * a compare and 2 stores extra right now if we dont have it on
1534 * but have CONFIG_NET_CLS_ACT
1535 * NOTE: This doesnt stop any functionality; if you dont have
1536 * the ingress scheduler, you just cant add policies on ingress.
1537 *
1538 */
1539static int ing_filter(struct sk_buff *skb)
1540{
1541 struct Qdisc *q;
1542 struct net_device *dev = skb->dev;
1543 int result = TC_ACT_OK;
1544
1545 if (dev->qdisc_ingress) {
1546 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1547 if (MAX_RED_LOOP < ttl++) {
1548 printk("Redir loop detected Dropping packet (%s->%s)\n",
David S. Miller86e65da2005-08-09 19:36:29 -07001549 skb->input_dev->name, skb->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 return TC_ACT_SHOT;
1551 }
1552
1553 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1554
1555 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
David S. Miller86e65da2005-08-09 19:36:29 -07001556
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 spin_lock(&dev->ingress_lock);
1558 if ((q = dev->qdisc_ingress) != NULL)
1559 result = q->enqueue(skb, q);
1560 spin_unlock(&dev->ingress_lock);
1561
1562 }
1563
1564 return result;
1565}
1566#endif
1567
1568int netif_receive_skb(struct sk_buff *skb)
1569{
1570 struct packet_type *ptype, *pt_prev;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001571 struct net_device *orig_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 int ret = NET_RX_DROP;
1573 unsigned short type;
1574
1575 /* if we've gotten here through NAPI, check netpoll */
1576 if (skb->dev->poll && netpoll_rx(skb))
1577 return NET_RX_DROP;
1578
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001579 if (!skb->tstamp.off_sec)
1580 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
David S. Miller86e65da2005-08-09 19:36:29 -07001582 if (!skb->input_dev)
1583 skb->input_dev = skb->dev;
1584
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001585 orig_dev = skb_bond(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
1587 __get_cpu_var(netdev_rx_stat).total++;
1588
1589 skb->h.raw = skb->nh.raw = skb->data;
1590 skb->mac_len = skb->nh.raw - skb->mac.raw;
1591
1592 pt_prev = NULL;
1593
1594 rcu_read_lock();
1595
1596#ifdef CONFIG_NET_CLS_ACT
1597 if (skb->tc_verd & TC_NCLS) {
1598 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1599 goto ncls;
1600 }
1601#endif
1602
1603 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1604 if (!ptype->dev || ptype->dev == skb->dev) {
1605 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001606 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 pt_prev = ptype;
1608 }
1609 }
1610
1611#ifdef CONFIG_NET_CLS_ACT
1612 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001613 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 pt_prev = NULL; /* noone else should process this after*/
1615 } else {
1616 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1617 }
1618
1619 ret = ing_filter(skb);
1620
1621 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1622 kfree_skb(skb);
1623 goto out;
1624 }
1625
1626 skb->tc_verd = 0;
1627ncls:
1628#endif
1629
1630 handle_diverter(skb);
1631
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001632 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 goto out;
1634
1635 type = skb->protocol;
1636 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1637 if (ptype->type == type &&
1638 (!ptype->dev || ptype->dev == skb->dev)) {
1639 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001640 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 pt_prev = ptype;
1642 }
1643 }
1644
1645 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001646 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 } else {
1648 kfree_skb(skb);
1649 /* Jamal, now you will not able to escape explaining
1650 * me how you were going to use this. :-)
1651 */
1652 ret = NET_RX_DROP;
1653 }
1654
1655out:
1656 rcu_read_unlock();
1657 return ret;
1658}
1659
1660static int process_backlog(struct net_device *backlog_dev, int *budget)
1661{
1662 int work = 0;
1663 int quota = min(backlog_dev->quota, *budget);
1664 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1665 unsigned long start_time = jiffies;
1666
Stephen Hemmingere3876602005-06-08 14:56:01 -07001667 backlog_dev->weight = weight_p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 for (;;) {
1669 struct sk_buff *skb;
1670 struct net_device *dev;
1671
1672 local_irq_disable();
1673 skb = __skb_dequeue(&queue->input_pkt_queue);
1674 if (!skb)
1675 goto job_done;
1676 local_irq_enable();
1677
1678 dev = skb->dev;
1679
1680 netif_receive_skb(skb);
1681
1682 dev_put(dev);
1683
1684 work++;
1685
1686 if (work >= quota || jiffies - start_time > 1)
1687 break;
1688
1689 }
1690
1691 backlog_dev->quota -= work;
1692 *budget -= work;
1693 return -1;
1694
1695job_done:
1696 backlog_dev->quota -= work;
1697 *budget -= work;
1698
1699 list_del(&backlog_dev->poll_list);
1700 smp_mb__before_clear_bit();
1701 netif_poll_enable(backlog_dev);
1702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 local_irq_enable();
1704 return 0;
1705}
1706
1707static void net_rx_action(struct softirq_action *h)
1708{
1709 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1710 unsigned long start_time = jiffies;
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07001711 int budget = netdev_budget;
Matt Mackall53fb95d2005-08-11 19:27:43 -07001712 void *have;
1713
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 local_irq_disable();
1715
1716 while (!list_empty(&queue->poll_list)) {
1717 struct net_device *dev;
1718
1719 if (budget <= 0 || jiffies - start_time > 1)
1720 goto softnet_break;
1721
1722 local_irq_enable();
1723
1724 dev = list_entry(queue->poll_list.next,
1725 struct net_device, poll_list);
Matt Mackall53fb95d2005-08-11 19:27:43 -07001726 have = netpoll_poll_lock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
1728 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
Matt Mackall53fb95d2005-08-11 19:27:43 -07001729 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 local_irq_disable();
1731 list_del(&dev->poll_list);
1732 list_add_tail(&dev->poll_list, &queue->poll_list);
1733 if (dev->quota < 0)
1734 dev->quota += dev->weight;
1735 else
1736 dev->quota = dev->weight;
1737 } else {
Matt Mackall53fb95d2005-08-11 19:27:43 -07001738 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 dev_put(dev);
1740 local_irq_disable();
1741 }
1742 }
1743out:
1744 local_irq_enable();
1745 return;
1746
1747softnet_break:
1748 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1749 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1750 goto out;
1751}
1752
1753static gifconf_func_t * gifconf_list [NPROTO];
1754
1755/**
1756 * register_gifconf - register a SIOCGIF handler
1757 * @family: Address family
1758 * @gifconf: Function handler
1759 *
1760 * Register protocol dependent address dumping routines. The handler
1761 * that is passed must not be freed or reused until it has been replaced
1762 * by another handler.
1763 */
1764int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1765{
1766 if (family >= NPROTO)
1767 return -EINVAL;
1768 gifconf_list[family] = gifconf;
1769 return 0;
1770}
1771
1772
1773/*
1774 * Map an interface index to its name (SIOCGIFNAME)
1775 */
1776
1777/*
1778 * We need this ioctl for efficient implementation of the
1779 * if_indextoname() function required by the IPv6 API. Without
1780 * it, we would have to search all the interfaces to find a
1781 * match. --pb
1782 */
1783
1784static int dev_ifname(struct ifreq __user *arg)
1785{
1786 struct net_device *dev;
1787 struct ifreq ifr;
1788
1789 /*
1790 * Fetch the caller's info block.
1791 */
1792
1793 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1794 return -EFAULT;
1795
1796 read_lock(&dev_base_lock);
1797 dev = __dev_get_by_index(ifr.ifr_ifindex);
1798 if (!dev) {
1799 read_unlock(&dev_base_lock);
1800 return -ENODEV;
1801 }
1802
1803 strcpy(ifr.ifr_name, dev->name);
1804 read_unlock(&dev_base_lock);
1805
1806 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1807 return -EFAULT;
1808 return 0;
1809}
1810
1811/*
1812 * Perform a SIOCGIFCONF call. This structure will change
1813 * size eventually, and there is nothing I can do about it.
1814 * Thus we will need a 'compatibility mode'.
1815 */
1816
1817static int dev_ifconf(char __user *arg)
1818{
1819 struct ifconf ifc;
1820 struct net_device *dev;
1821 char __user *pos;
1822 int len;
1823 int total;
1824 int i;
1825
1826 /*
1827 * Fetch the caller's info block.
1828 */
1829
1830 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1831 return -EFAULT;
1832
1833 pos = ifc.ifc_buf;
1834 len = ifc.ifc_len;
1835
1836 /*
1837 * Loop over the interfaces, and write an info block for each.
1838 */
1839
1840 total = 0;
1841 for (dev = dev_base; dev; dev = dev->next) {
1842 for (i = 0; i < NPROTO; i++) {
1843 if (gifconf_list[i]) {
1844 int done;
1845 if (!pos)
1846 done = gifconf_list[i](dev, NULL, 0);
1847 else
1848 done = gifconf_list[i](dev, pos + total,
1849 len - total);
1850 if (done < 0)
1851 return -EFAULT;
1852 total += done;
1853 }
1854 }
1855 }
1856
1857 /*
1858 * All done. Write the updated control block back to the caller.
1859 */
1860 ifc.ifc_len = total;
1861
1862 /*
1863 * Both BSD and Solaris return 0 here, so we do too.
1864 */
1865 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1866}
1867
1868#ifdef CONFIG_PROC_FS
1869/*
1870 * This is invoked by the /proc filesystem handler to display a device
1871 * in detail.
1872 */
1873static __inline__ struct net_device *dev_get_idx(loff_t pos)
1874{
1875 struct net_device *dev;
1876 loff_t i;
1877
1878 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1879
1880 return i == pos ? dev : NULL;
1881}
1882
1883void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1884{
1885 read_lock(&dev_base_lock);
1886 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1887}
1888
1889void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1890{
1891 ++*pos;
1892 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1893}
1894
1895void dev_seq_stop(struct seq_file *seq, void *v)
1896{
1897 read_unlock(&dev_base_lock);
1898}
1899
1900static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1901{
1902 if (dev->get_stats) {
1903 struct net_device_stats *stats = dev->get_stats(dev);
1904
1905 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1906 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1907 dev->name, stats->rx_bytes, stats->rx_packets,
1908 stats->rx_errors,
1909 stats->rx_dropped + stats->rx_missed_errors,
1910 stats->rx_fifo_errors,
1911 stats->rx_length_errors + stats->rx_over_errors +
1912 stats->rx_crc_errors + stats->rx_frame_errors,
1913 stats->rx_compressed, stats->multicast,
1914 stats->tx_bytes, stats->tx_packets,
1915 stats->tx_errors, stats->tx_dropped,
1916 stats->tx_fifo_errors, stats->collisions,
1917 stats->tx_carrier_errors +
1918 stats->tx_aborted_errors +
1919 stats->tx_window_errors +
1920 stats->tx_heartbeat_errors,
1921 stats->tx_compressed);
1922 } else
1923 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1924}
1925
1926/*
1927 * Called from the PROCfs module. This now uses the new arbitrary sized
1928 * /proc/net interface to create /proc/net/dev
1929 */
1930static int dev_seq_show(struct seq_file *seq, void *v)
1931{
1932 if (v == SEQ_START_TOKEN)
1933 seq_puts(seq, "Inter-| Receive "
1934 " | Transmit\n"
1935 " face |bytes packets errs drop fifo frame "
1936 "compressed multicast|bytes packets errs "
1937 "drop fifo colls carrier compressed\n");
1938 else
1939 dev_seq_printf_stats(seq, v);
1940 return 0;
1941}
1942
1943static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1944{
1945 struct netif_rx_stats *rc = NULL;
1946
1947 while (*pos < NR_CPUS)
1948 if (cpu_online(*pos)) {
1949 rc = &per_cpu(netdev_rx_stat, *pos);
1950 break;
1951 } else
1952 ++*pos;
1953 return rc;
1954}
1955
1956static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1957{
1958 return softnet_get_online(pos);
1959}
1960
1961static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1962{
1963 ++*pos;
1964 return softnet_get_online(pos);
1965}
1966
1967static void softnet_seq_stop(struct seq_file *seq, void *v)
1968{
1969}
1970
1971static int softnet_seq_show(struct seq_file *seq, void *v)
1972{
1973 struct netif_rx_stats *s = v;
1974
1975 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
Stephen Hemminger31aa02c2005-06-23 20:12:48 -07001976 s->total, s->dropped, s->time_squeeze, 0,
Stephen Hemmingerc1ebcdb2005-06-23 20:08:59 -07001977 0, 0, 0, 0, /* was fastroute */
1978 s->cpu_collision );
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 return 0;
1980}
1981
1982static struct seq_operations dev_seq_ops = {
1983 .start = dev_seq_start,
1984 .next = dev_seq_next,
1985 .stop = dev_seq_stop,
1986 .show = dev_seq_show,
1987};
1988
1989static int dev_seq_open(struct inode *inode, struct file *file)
1990{
1991 return seq_open(file, &dev_seq_ops);
1992}
1993
1994static struct file_operations dev_seq_fops = {
1995 .owner = THIS_MODULE,
1996 .open = dev_seq_open,
1997 .read = seq_read,
1998 .llseek = seq_lseek,
1999 .release = seq_release,
2000};
2001
2002static struct seq_operations softnet_seq_ops = {
2003 .start = softnet_seq_start,
2004 .next = softnet_seq_next,
2005 .stop = softnet_seq_stop,
2006 .show = softnet_seq_show,
2007};
2008
2009static int softnet_seq_open(struct inode *inode, struct file *file)
2010{
2011 return seq_open(file, &softnet_seq_ops);
2012}
2013
2014static struct file_operations softnet_seq_fops = {
2015 .owner = THIS_MODULE,
2016 .open = softnet_seq_open,
2017 .read = seq_read,
2018 .llseek = seq_lseek,
2019 .release = seq_release,
2020};
2021
2022#ifdef WIRELESS_EXT
2023extern int wireless_proc_init(void);
2024#else
2025#define wireless_proc_init() 0
2026#endif
2027
2028static int __init dev_proc_init(void)
2029{
2030 int rc = -ENOMEM;
2031
2032 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2033 goto out;
2034 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2035 goto out_dev;
2036 if (wireless_proc_init())
2037 goto out_softnet;
2038 rc = 0;
2039out:
2040 return rc;
2041out_softnet:
2042 proc_net_remove("softnet_stat");
2043out_dev:
2044 proc_net_remove("dev");
2045 goto out;
2046}
2047#else
2048#define dev_proc_init() 0
2049#endif /* CONFIG_PROC_FS */
2050
2051
2052/**
2053 * netdev_set_master - set up master/slave pair
2054 * @slave: slave device
2055 * @master: new master device
2056 *
2057 * Changes the master device of the slave. Pass %NULL to break the
2058 * bonding. The caller must hold the RTNL semaphore. On a failure
2059 * a negative errno code is returned. On success the reference counts
2060 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2061 * function returns zero.
2062 */
2063int netdev_set_master(struct net_device *slave, struct net_device *master)
2064{
2065 struct net_device *old = slave->master;
2066
2067 ASSERT_RTNL();
2068
2069 if (master) {
2070 if (old)
2071 return -EBUSY;
2072 dev_hold(master);
2073 }
2074
2075 slave->master = master;
2076
2077 synchronize_net();
2078
2079 if (old)
2080 dev_put(old);
2081
2082 if (master)
2083 slave->flags |= IFF_SLAVE;
2084 else
2085 slave->flags &= ~IFF_SLAVE;
2086
2087 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2088 return 0;
2089}
2090
2091/**
2092 * dev_set_promiscuity - update promiscuity count on a device
2093 * @dev: device
2094 * @inc: modifier
2095 *
2096 * Add or remove promsicuity from a device. While the count in the device
2097 * remains above zero the interface remains promiscuous. Once it hits zero
2098 * the device reverts back to normal filtering operation. A negative inc
2099 * value is used to drop promiscuity on the device.
2100 */
2101void dev_set_promiscuity(struct net_device *dev, int inc)
2102{
2103 unsigned short old_flags = dev->flags;
2104
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 if ((dev->promiscuity += inc) == 0)
2106 dev->flags &= ~IFF_PROMISC;
David Chau52609c02005-07-05 15:11:06 -07002107 else
2108 dev->flags |= IFF_PROMISC;
2109 if (dev->flags != old_flags) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 dev_mc_upload(dev);
2111 printk(KERN_INFO "device %s %s promiscuous mode\n",
2112 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2113 "left");
2114 }
2115}
2116
2117/**
2118 * dev_set_allmulti - update allmulti count on a device
2119 * @dev: device
2120 * @inc: modifier
2121 *
2122 * Add or remove reception of all multicast frames to a device. While the
2123 * count in the device remains above zero the interface remains listening
2124 * to all interfaces. Once it hits zero the device reverts back to normal
2125 * filtering operation. A negative @inc value is used to drop the counter
2126 * when releasing a resource needing all multicasts.
2127 */
2128
2129void dev_set_allmulti(struct net_device *dev, int inc)
2130{
2131 unsigned short old_flags = dev->flags;
2132
2133 dev->flags |= IFF_ALLMULTI;
2134 if ((dev->allmulti += inc) == 0)
2135 dev->flags &= ~IFF_ALLMULTI;
2136 if (dev->flags ^ old_flags)
2137 dev_mc_upload(dev);
2138}
2139
2140unsigned dev_get_flags(const struct net_device *dev)
2141{
2142 unsigned flags;
2143
2144 flags = (dev->flags & ~(IFF_PROMISC |
2145 IFF_ALLMULTI |
2146 IFF_RUNNING)) |
2147 (dev->gflags & (IFF_PROMISC |
2148 IFF_ALLMULTI));
2149
2150 if (netif_running(dev) && netif_carrier_ok(dev))
2151 flags |= IFF_RUNNING;
2152
2153 return flags;
2154}
2155
2156int dev_change_flags(struct net_device *dev, unsigned flags)
2157{
2158 int ret;
2159 int old_flags = dev->flags;
2160
2161 /*
2162 * Set the flags on our device.
2163 */
2164
2165 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2166 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2167 IFF_AUTOMEDIA)) |
2168 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2169 IFF_ALLMULTI));
2170
2171 /*
2172 * Load in the correct multicast list now the flags have changed.
2173 */
2174
2175 dev_mc_upload(dev);
2176
2177 /*
2178 * Have we downed the interface. We handle IFF_UP ourselves
2179 * according to user attempts to set it, rather than blindly
2180 * setting it.
2181 */
2182
2183 ret = 0;
2184 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2185 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2186
2187 if (!ret)
2188 dev_mc_upload(dev);
2189 }
2190
2191 if (dev->flags & IFF_UP &&
2192 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2193 IFF_VOLATILE)))
2194 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2195
2196 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2197 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2198 dev->gflags ^= IFF_PROMISC;
2199 dev_set_promiscuity(dev, inc);
2200 }
2201
2202 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2203 is important. Some (broken) drivers set IFF_PROMISC, when
2204 IFF_ALLMULTI is requested not asking us and not reporting.
2205 */
2206 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2207 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2208 dev->gflags ^= IFF_ALLMULTI;
2209 dev_set_allmulti(dev, inc);
2210 }
2211
2212 if (old_flags ^ dev->flags)
2213 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2214
2215 return ret;
2216}
2217
2218int dev_set_mtu(struct net_device *dev, int new_mtu)
2219{
2220 int err;
2221
2222 if (new_mtu == dev->mtu)
2223 return 0;
2224
2225 /* MTU must be positive. */
2226 if (new_mtu < 0)
2227 return -EINVAL;
2228
2229 if (!netif_device_present(dev))
2230 return -ENODEV;
2231
2232 err = 0;
2233 if (dev->change_mtu)
2234 err = dev->change_mtu(dev, new_mtu);
2235 else
2236 dev->mtu = new_mtu;
2237 if (!err && dev->flags & IFF_UP)
2238 notifier_call_chain(&netdev_chain,
2239 NETDEV_CHANGEMTU, dev);
2240 return err;
2241}
2242
2243int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2244{
2245 int err;
2246
2247 if (!dev->set_mac_address)
2248 return -EOPNOTSUPP;
2249 if (sa->sa_family != dev->type)
2250 return -EINVAL;
2251 if (!netif_device_present(dev))
2252 return -ENODEV;
2253 err = dev->set_mac_address(dev, sa);
2254 if (!err)
2255 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2256 return err;
2257}
2258
2259/*
2260 * Perform the SIOCxIFxxx calls.
2261 */
2262static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2263{
2264 int err;
2265 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2266
2267 if (!dev)
2268 return -ENODEV;
2269
2270 switch (cmd) {
2271 case SIOCGIFFLAGS: /* Get interface flags */
2272 ifr->ifr_flags = dev_get_flags(dev);
2273 return 0;
2274
2275 case SIOCSIFFLAGS: /* Set interface flags */
2276 return dev_change_flags(dev, ifr->ifr_flags);
2277
2278 case SIOCGIFMETRIC: /* Get the metric on the interface
2279 (currently unused) */
2280 ifr->ifr_metric = 0;
2281 return 0;
2282
2283 case SIOCSIFMETRIC: /* Set the metric on the interface
2284 (currently unused) */
2285 return -EOPNOTSUPP;
2286
2287 case SIOCGIFMTU: /* Get the MTU of a device */
2288 ifr->ifr_mtu = dev->mtu;
2289 return 0;
2290
2291 case SIOCSIFMTU: /* Set the MTU of a device */
2292 return dev_set_mtu(dev, ifr->ifr_mtu);
2293
2294 case SIOCGIFHWADDR:
2295 if (!dev->addr_len)
2296 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2297 else
2298 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2299 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2300 ifr->ifr_hwaddr.sa_family = dev->type;
2301 return 0;
2302
2303 case SIOCSIFHWADDR:
2304 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2305
2306 case SIOCSIFHWBROADCAST:
2307 if (ifr->ifr_hwaddr.sa_family != dev->type)
2308 return -EINVAL;
2309 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2310 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2311 notifier_call_chain(&netdev_chain,
2312 NETDEV_CHANGEADDR, dev);
2313 return 0;
2314
2315 case SIOCGIFMAP:
2316 ifr->ifr_map.mem_start = dev->mem_start;
2317 ifr->ifr_map.mem_end = dev->mem_end;
2318 ifr->ifr_map.base_addr = dev->base_addr;
2319 ifr->ifr_map.irq = dev->irq;
2320 ifr->ifr_map.dma = dev->dma;
2321 ifr->ifr_map.port = dev->if_port;
2322 return 0;
2323
2324 case SIOCSIFMAP:
2325 if (dev->set_config) {
2326 if (!netif_device_present(dev))
2327 return -ENODEV;
2328 return dev->set_config(dev, &ifr->ifr_map);
2329 }
2330 return -EOPNOTSUPP;
2331
2332 case SIOCADDMULTI:
2333 if (!dev->set_multicast_list ||
2334 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2335 return -EINVAL;
2336 if (!netif_device_present(dev))
2337 return -ENODEV;
2338 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2339 dev->addr_len, 1);
2340
2341 case SIOCDELMULTI:
2342 if (!dev->set_multicast_list ||
2343 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2344 return -EINVAL;
2345 if (!netif_device_present(dev))
2346 return -ENODEV;
2347 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2348 dev->addr_len, 1);
2349
2350 case SIOCGIFINDEX:
2351 ifr->ifr_ifindex = dev->ifindex;
2352 return 0;
2353
2354 case SIOCGIFTXQLEN:
2355 ifr->ifr_qlen = dev->tx_queue_len;
2356 return 0;
2357
2358 case SIOCSIFTXQLEN:
2359 if (ifr->ifr_qlen < 0)
2360 return -EINVAL;
2361 dev->tx_queue_len = ifr->ifr_qlen;
2362 return 0;
2363
2364 case SIOCSIFNAME:
2365 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2366 return dev_change_name(dev, ifr->ifr_newname);
2367
2368 /*
2369 * Unknown or private ioctl
2370 */
2371
2372 default:
2373 if ((cmd >= SIOCDEVPRIVATE &&
2374 cmd <= SIOCDEVPRIVATE + 15) ||
2375 cmd == SIOCBONDENSLAVE ||
2376 cmd == SIOCBONDRELEASE ||
2377 cmd == SIOCBONDSETHWADDR ||
2378 cmd == SIOCBONDSLAVEINFOQUERY ||
2379 cmd == SIOCBONDINFOQUERY ||
2380 cmd == SIOCBONDCHANGEACTIVE ||
2381 cmd == SIOCGMIIPHY ||
2382 cmd == SIOCGMIIREG ||
2383 cmd == SIOCSMIIREG ||
2384 cmd == SIOCBRADDIF ||
2385 cmd == SIOCBRDELIF ||
2386 cmd == SIOCWANDEV) {
2387 err = -EOPNOTSUPP;
2388 if (dev->do_ioctl) {
2389 if (netif_device_present(dev))
2390 err = dev->do_ioctl(dev, ifr,
2391 cmd);
2392 else
2393 err = -ENODEV;
2394 }
2395 } else
2396 err = -EINVAL;
2397
2398 }
2399 return err;
2400}
2401
2402/*
2403 * This function handles all "interface"-type I/O control requests. The actual
2404 * 'doing' part of this is dev_ifsioc above.
2405 */
2406
2407/**
2408 * dev_ioctl - network device ioctl
2409 * @cmd: command to issue
2410 * @arg: pointer to a struct ifreq in user space
2411 *
2412 * Issue ioctl functions to devices. This is normally called by the
2413 * user space syscall interfaces but can sometimes be useful for
2414 * other purposes. The return value is the return from the syscall if
2415 * positive or a negative errno code on error.
2416 */
2417
2418int dev_ioctl(unsigned int cmd, void __user *arg)
2419{
2420 struct ifreq ifr;
2421 int ret;
2422 char *colon;
2423
2424 /* One special case: SIOCGIFCONF takes ifconf argument
2425 and requires shared lock, because it sleeps writing
2426 to user space.
2427 */
2428
2429 if (cmd == SIOCGIFCONF) {
2430 rtnl_shlock();
2431 ret = dev_ifconf((char __user *) arg);
2432 rtnl_shunlock();
2433 return ret;
2434 }
2435 if (cmd == SIOCGIFNAME)
2436 return dev_ifname((struct ifreq __user *)arg);
2437
2438 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2439 return -EFAULT;
2440
2441 ifr.ifr_name[IFNAMSIZ-1] = 0;
2442
2443 colon = strchr(ifr.ifr_name, ':');
2444 if (colon)
2445 *colon = 0;
2446
2447 /*
2448 * See which interface the caller is talking about.
2449 */
2450
2451 switch (cmd) {
2452 /*
2453 * These ioctl calls:
2454 * - can be done by all.
2455 * - atomic and do not require locking.
2456 * - return a value
2457 */
2458 case SIOCGIFFLAGS:
2459 case SIOCGIFMETRIC:
2460 case SIOCGIFMTU:
2461 case SIOCGIFHWADDR:
2462 case SIOCGIFSLAVE:
2463 case SIOCGIFMAP:
2464 case SIOCGIFINDEX:
2465 case SIOCGIFTXQLEN:
2466 dev_load(ifr.ifr_name);
2467 read_lock(&dev_base_lock);
2468 ret = dev_ifsioc(&ifr, cmd);
2469 read_unlock(&dev_base_lock);
2470 if (!ret) {
2471 if (colon)
2472 *colon = ':';
2473 if (copy_to_user(arg, &ifr,
2474 sizeof(struct ifreq)))
2475 ret = -EFAULT;
2476 }
2477 return ret;
2478
2479 case SIOCETHTOOL:
2480 dev_load(ifr.ifr_name);
2481 rtnl_lock();
2482 ret = dev_ethtool(&ifr);
2483 rtnl_unlock();
2484 if (!ret) {
2485 if (colon)
2486 *colon = ':';
2487 if (copy_to_user(arg, &ifr,
2488 sizeof(struct ifreq)))
2489 ret = -EFAULT;
2490 }
2491 return ret;
2492
2493 /*
2494 * These ioctl calls:
2495 * - require superuser power.
2496 * - require strict serialization.
2497 * - return a value
2498 */
2499 case SIOCGMIIPHY:
2500 case SIOCGMIIREG:
2501 case SIOCSIFNAME:
2502 if (!capable(CAP_NET_ADMIN))
2503 return -EPERM;
2504 dev_load(ifr.ifr_name);
2505 rtnl_lock();
2506 ret = dev_ifsioc(&ifr, cmd);
2507 rtnl_unlock();
2508 if (!ret) {
2509 if (colon)
2510 *colon = ':';
2511 if (copy_to_user(arg, &ifr,
2512 sizeof(struct ifreq)))
2513 ret = -EFAULT;
2514 }
2515 return ret;
2516
2517 /*
2518 * These ioctl calls:
2519 * - require superuser power.
2520 * - require strict serialization.
2521 * - do not return a value
2522 */
2523 case SIOCSIFFLAGS:
2524 case SIOCSIFMETRIC:
2525 case SIOCSIFMTU:
2526 case SIOCSIFMAP:
2527 case SIOCSIFHWADDR:
2528 case SIOCSIFSLAVE:
2529 case SIOCADDMULTI:
2530 case SIOCDELMULTI:
2531 case SIOCSIFHWBROADCAST:
2532 case SIOCSIFTXQLEN:
2533 case SIOCSMIIREG:
2534 case SIOCBONDENSLAVE:
2535 case SIOCBONDRELEASE:
2536 case SIOCBONDSETHWADDR:
2537 case SIOCBONDSLAVEINFOQUERY:
2538 case SIOCBONDINFOQUERY:
2539 case SIOCBONDCHANGEACTIVE:
2540 case SIOCBRADDIF:
2541 case SIOCBRDELIF:
2542 if (!capable(CAP_NET_ADMIN))
2543 return -EPERM;
2544 dev_load(ifr.ifr_name);
2545 rtnl_lock();
2546 ret = dev_ifsioc(&ifr, cmd);
2547 rtnl_unlock();
2548 return ret;
2549
2550 case SIOCGIFMEM:
2551 /* Get the per device memory space. We can add this but
2552 * currently do not support it */
2553 case SIOCSIFMEM:
2554 /* Set the per device memory buffer space.
2555 * Not applicable in our case */
2556 case SIOCSIFLINK:
2557 return -EINVAL;
2558
2559 /*
2560 * Unknown or private ioctl.
2561 */
2562 default:
2563 if (cmd == SIOCWANDEV ||
2564 (cmd >= SIOCDEVPRIVATE &&
2565 cmd <= SIOCDEVPRIVATE + 15)) {
2566 dev_load(ifr.ifr_name);
2567 rtnl_lock();
2568 ret = dev_ifsioc(&ifr, cmd);
2569 rtnl_unlock();
2570 if (!ret && copy_to_user(arg, &ifr,
2571 sizeof(struct ifreq)))
2572 ret = -EFAULT;
2573 return ret;
2574 }
2575#ifdef WIRELESS_EXT
2576 /* Take care of Wireless Extensions */
2577 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2578 /* If command is `set a parameter', or
2579 * `get the encoding parameters', check if
2580 * the user has the right to do it */
2581 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2582 if (!capable(CAP_NET_ADMIN))
2583 return -EPERM;
2584 }
2585 dev_load(ifr.ifr_name);
2586 rtnl_lock();
2587 /* Follow me in net/core/wireless.c */
2588 ret = wireless_process_ioctl(&ifr, cmd);
2589 rtnl_unlock();
2590 if (IW_IS_GET(cmd) &&
2591 copy_to_user(arg, &ifr,
2592 sizeof(struct ifreq)))
2593 ret = -EFAULT;
2594 return ret;
2595 }
2596#endif /* WIRELESS_EXT */
2597 return -EINVAL;
2598 }
2599}
2600
2601
2602/**
2603 * dev_new_index - allocate an ifindex
2604 *
2605 * Returns a suitable unique value for a new device interface
2606 * number. The caller must hold the rtnl semaphore or the
2607 * dev_base_lock to be sure it remains unique.
2608 */
2609static int dev_new_index(void)
2610{
2611 static int ifindex;
2612 for (;;) {
2613 if (++ifindex <= 0)
2614 ifindex = 1;
2615 if (!__dev_get_by_index(ifindex))
2616 return ifindex;
2617 }
2618}
2619
2620static int dev_boot_phase = 1;
2621
2622/* Delayed registration/unregisteration */
2623static DEFINE_SPINLOCK(net_todo_list_lock);
2624static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2625
2626static inline void net_set_todo(struct net_device *dev)
2627{
2628 spin_lock(&net_todo_list_lock);
2629 list_add_tail(&dev->todo_list, &net_todo_list);
2630 spin_unlock(&net_todo_list_lock);
2631}
2632
2633/**
2634 * register_netdevice - register a network device
2635 * @dev: device to register
2636 *
2637 * Take a completed network device structure and add it to the kernel
2638 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2639 * chain. 0 is returned on success. A negative errno code is returned
2640 * on a failure to set up the device, or if the name is a duplicate.
2641 *
2642 * Callers must hold the rtnl semaphore. You may want
2643 * register_netdev() instead of this.
2644 *
2645 * BUGS:
2646 * The locking appears insufficient to guarantee two parallel registers
2647 * will not get the same name.
2648 */
2649
2650int register_netdevice(struct net_device *dev)
2651{
2652 struct hlist_head *head;
2653 struct hlist_node *p;
2654 int ret;
2655
2656 BUG_ON(dev_boot_phase);
2657 ASSERT_RTNL();
2658
2659 /* When net_device's are persistent, this will be fatal. */
2660 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2661
2662 spin_lock_init(&dev->queue_lock);
2663 spin_lock_init(&dev->xmit_lock);
2664 dev->xmit_lock_owner = -1;
2665#ifdef CONFIG_NET_CLS_ACT
2666 spin_lock_init(&dev->ingress_lock);
2667#endif
2668
2669 ret = alloc_divert_blk(dev);
2670 if (ret)
2671 goto out;
2672
2673 dev->iflink = -1;
2674
2675 /* Init, if this function is available */
2676 if (dev->init) {
2677 ret = dev->init(dev);
2678 if (ret) {
2679 if (ret > 0)
2680 ret = -EIO;
2681 goto out_err;
2682 }
2683 }
2684
2685 if (!dev_valid_name(dev->name)) {
2686 ret = -EINVAL;
2687 goto out_err;
2688 }
2689
2690 dev->ifindex = dev_new_index();
2691 if (dev->iflink == -1)
2692 dev->iflink = dev->ifindex;
2693
2694 /* Check for existence of name */
2695 head = dev_name_hash(dev->name);
2696 hlist_for_each(p, head) {
2697 struct net_device *d
2698 = hlist_entry(p, struct net_device, name_hlist);
2699 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2700 ret = -EEXIST;
2701 goto out_err;
2702 }
2703 }
2704
2705 /* Fix illegal SG+CSUM combinations. */
2706 if ((dev->features & NETIF_F_SG) &&
2707 !(dev->features & (NETIF_F_IP_CSUM |
2708 NETIF_F_NO_CSUM |
2709 NETIF_F_HW_CSUM))) {
2710 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2711 dev->name);
2712 dev->features &= ~NETIF_F_SG;
2713 }
2714
2715 /* TSO requires that SG is present as well. */
2716 if ((dev->features & NETIF_F_TSO) &&
2717 !(dev->features & NETIF_F_SG)) {
2718 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2719 dev->name);
2720 dev->features &= ~NETIF_F_TSO;
2721 }
2722
2723 /*
2724 * nil rebuild_header routine,
2725 * that should be never called and used as just bug trap.
2726 */
2727
2728 if (!dev->rebuild_header)
2729 dev->rebuild_header = default_rebuild_header;
2730
2731 /*
2732 * Default initial state at registry is that the
2733 * device is present.
2734 */
2735
2736 set_bit(__LINK_STATE_PRESENT, &dev->state);
2737
2738 dev->next = NULL;
2739 dev_init_scheduler(dev);
2740 write_lock_bh(&dev_base_lock);
2741 *dev_tail = dev;
2742 dev_tail = &dev->next;
2743 hlist_add_head(&dev->name_hlist, head);
2744 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2745 dev_hold(dev);
2746 dev->reg_state = NETREG_REGISTERING;
2747 write_unlock_bh(&dev_base_lock);
2748
2749 /* Notify protocols, that a new device appeared. */
2750 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2751
2752 /* Finish registration after unlock */
2753 net_set_todo(dev);
2754 ret = 0;
2755
2756out:
2757 return ret;
2758out_err:
2759 free_divert_blk(dev);
2760 goto out;
2761}
2762
2763/**
2764 * register_netdev - register a network device
2765 * @dev: device to register
2766 *
2767 * Take a completed network device structure and add it to the kernel
2768 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2769 * chain. 0 is returned on success. A negative errno code is returned
2770 * on a failure to set up the device, or if the name is a duplicate.
2771 *
2772 * This is a wrapper around register_netdev that takes the rtnl semaphore
2773 * and expands the device name if you passed a format string to
2774 * alloc_netdev.
2775 */
2776int register_netdev(struct net_device *dev)
2777{
2778 int err;
2779
2780 rtnl_lock();
2781
2782 /*
2783 * If the name is a format string the caller wants us to do a
2784 * name allocation.
2785 */
2786 if (strchr(dev->name, '%')) {
2787 err = dev_alloc_name(dev, dev->name);
2788 if (err < 0)
2789 goto out;
2790 }
2791
2792 /*
2793 * Back compatibility hook. Kill this one in 2.5
2794 */
2795 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2796 err = dev_alloc_name(dev, "eth%d");
2797 if (err < 0)
2798 goto out;
2799 }
2800
2801 err = register_netdevice(dev);
2802out:
2803 rtnl_unlock();
2804 return err;
2805}
2806EXPORT_SYMBOL(register_netdev);
2807
2808/*
2809 * netdev_wait_allrefs - wait until all references are gone.
2810 *
2811 * This is called when unregistering network devices.
2812 *
2813 * Any protocol or device that holds a reference should register
2814 * for netdevice notification, and cleanup and put back the
2815 * reference if they receive an UNREGISTER event.
2816 * We can get stuck here if buggy protocols don't correctly
2817 * call dev_put.
2818 */
2819static void netdev_wait_allrefs(struct net_device *dev)
2820{
2821 unsigned long rebroadcast_time, warning_time;
2822
2823 rebroadcast_time = warning_time = jiffies;
2824 while (atomic_read(&dev->refcnt) != 0) {
2825 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2826 rtnl_shlock();
2827
2828 /* Rebroadcast unregister notification */
2829 notifier_call_chain(&netdev_chain,
2830 NETDEV_UNREGISTER, dev);
2831
2832 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2833 &dev->state)) {
2834 /* We must not have linkwatch events
2835 * pending on unregister. If this
2836 * happens, we simply run the queue
2837 * unscheduled, resulting in a noop
2838 * for this device.
2839 */
2840 linkwatch_run_queue();
2841 }
2842
2843 rtnl_shunlock();
2844
2845 rebroadcast_time = jiffies;
2846 }
2847
2848 msleep(250);
2849
2850 if (time_after(jiffies, warning_time + 10 * HZ)) {
2851 printk(KERN_EMERG "unregister_netdevice: "
2852 "waiting for %s to become free. Usage "
2853 "count = %d\n",
2854 dev->name, atomic_read(&dev->refcnt));
2855 warning_time = jiffies;
2856 }
2857 }
2858}
2859
2860/* The sequence is:
2861 *
2862 * rtnl_lock();
2863 * ...
2864 * register_netdevice(x1);
2865 * register_netdevice(x2);
2866 * ...
2867 * unregister_netdevice(y1);
2868 * unregister_netdevice(y2);
2869 * ...
2870 * rtnl_unlock();
2871 * free_netdev(y1);
2872 * free_netdev(y2);
2873 *
2874 * We are invoked by rtnl_unlock() after it drops the semaphore.
2875 * This allows us to deal with problems:
2876 * 1) We can create/delete sysfs objects which invoke hotplug
2877 * without deadlocking with linkwatch via keventd.
2878 * 2) Since we run with the RTNL semaphore not held, we can sleep
2879 * safely in order to wait for the netdev refcnt to drop to zero.
2880 */
2881static DECLARE_MUTEX(net_todo_run_mutex);
2882void netdev_run_todo(void)
2883{
2884 struct list_head list = LIST_HEAD_INIT(list);
2885 int err;
2886
2887
2888 /* Need to guard against multiple cpu's getting out of order. */
2889 down(&net_todo_run_mutex);
2890
2891 /* Not safe to do outside the semaphore. We must not return
2892 * until all unregister events invoked by the local processor
2893 * have been completed (either by this todo run, or one on
2894 * another cpu).
2895 */
2896 if (list_empty(&net_todo_list))
2897 goto out;
2898
2899 /* Snapshot list, allow later requests */
2900 spin_lock(&net_todo_list_lock);
2901 list_splice_init(&net_todo_list, &list);
2902 spin_unlock(&net_todo_list_lock);
2903
2904 while (!list_empty(&list)) {
2905 struct net_device *dev
2906 = list_entry(list.next, struct net_device, todo_list);
2907 list_del(&dev->todo_list);
2908
2909 switch(dev->reg_state) {
2910 case NETREG_REGISTERING:
2911 err = netdev_register_sysfs(dev);
2912 if (err)
2913 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2914 dev->name, err);
2915 dev->reg_state = NETREG_REGISTERED;
2916 break;
2917
2918 case NETREG_UNREGISTERING:
2919 netdev_unregister_sysfs(dev);
2920 dev->reg_state = NETREG_UNREGISTERED;
2921
2922 netdev_wait_allrefs(dev);
2923
2924 /* paranoia */
2925 BUG_ON(atomic_read(&dev->refcnt));
2926 BUG_TRAP(!dev->ip_ptr);
2927 BUG_TRAP(!dev->ip6_ptr);
2928 BUG_TRAP(!dev->dn_ptr);
2929
2930
2931 /* It must be the very last action,
2932 * after this 'dev' may point to freed up memory.
2933 */
2934 if (dev->destructor)
2935 dev->destructor(dev);
2936 break;
2937
2938 default:
2939 printk(KERN_ERR "network todo '%s' but state %d\n",
2940 dev->name, dev->reg_state);
2941 break;
2942 }
2943 }
2944
2945out:
2946 up(&net_todo_run_mutex);
2947}
2948
2949/**
2950 * alloc_netdev - allocate network device
2951 * @sizeof_priv: size of private data to allocate space for
2952 * @name: device name format string
2953 * @setup: callback to initialize device
2954 *
2955 * Allocates a struct net_device with private data area for driver use
2956 * and performs basic initialization.
2957 */
2958struct net_device *alloc_netdev(int sizeof_priv, const char *name,
2959 void (*setup)(struct net_device *))
2960{
2961 void *p;
2962 struct net_device *dev;
2963 int alloc_size;
2964
2965 /* ensure 32-byte alignment of both the device and private area */
2966 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
2967 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
2968
2969 p = kmalloc(alloc_size, GFP_KERNEL);
2970 if (!p) {
2971 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
2972 return NULL;
2973 }
2974 memset(p, 0, alloc_size);
2975
2976 dev = (struct net_device *)
2977 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
2978 dev->padded = (char *)dev - (char *)p;
2979
2980 if (sizeof_priv)
2981 dev->priv = netdev_priv(dev);
2982
2983 setup(dev);
2984 strcpy(dev->name, name);
2985 return dev;
2986}
2987EXPORT_SYMBOL(alloc_netdev);
2988
2989/**
2990 * free_netdev - free network device
2991 * @dev: device
2992 *
2993 * This function does the last stage of destroying an allocated device
2994 * interface. The reference to the device object is released.
2995 * If this is the last reference then it will be freed.
2996 */
2997void free_netdev(struct net_device *dev)
2998{
2999#ifdef CONFIG_SYSFS
3000 /* Compatiablity with error handling in drivers */
3001 if (dev->reg_state == NETREG_UNINITIALIZED) {
3002 kfree((char *)dev - dev->padded);
3003 return;
3004 }
3005
3006 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3007 dev->reg_state = NETREG_RELEASED;
3008
3009 /* will free via class release */
3010 class_device_put(&dev->class_dev);
3011#else
3012 kfree((char *)dev - dev->padded);
3013#endif
3014}
3015
3016/* Synchronize with packet receive processing. */
3017void synchronize_net(void)
3018{
3019 might_sleep();
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07003020 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021}
3022
3023/**
3024 * unregister_netdevice - remove device from the kernel
3025 * @dev: device
3026 *
3027 * This function shuts down a device interface and removes it
3028 * from the kernel tables. On success 0 is returned, on a failure
3029 * a negative errno code is returned.
3030 *
3031 * Callers must hold the rtnl semaphore. You may want
3032 * unregister_netdev() instead of this.
3033 */
3034
3035int unregister_netdevice(struct net_device *dev)
3036{
3037 struct net_device *d, **dp;
3038
3039 BUG_ON(dev_boot_phase);
3040 ASSERT_RTNL();
3041
3042 /* Some devices call without registering for initialization unwind. */
3043 if (dev->reg_state == NETREG_UNINITIALIZED) {
3044 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3045 "was registered\n", dev->name, dev);
3046 return -ENODEV;
3047 }
3048
3049 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3050
3051 /* If device is running, close it first. */
3052 if (dev->flags & IFF_UP)
3053 dev_close(dev);
3054
3055 /* And unlink it from device chain. */
3056 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3057 if (d == dev) {
3058 write_lock_bh(&dev_base_lock);
3059 hlist_del(&dev->name_hlist);
3060 hlist_del(&dev->index_hlist);
3061 if (dev_tail == &dev->next)
3062 dev_tail = dp;
3063 *dp = d->next;
3064 write_unlock_bh(&dev_base_lock);
3065 break;
3066 }
3067 }
3068 if (!d) {
3069 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3070 dev->name);
3071 return -ENODEV;
3072 }
3073
3074 dev->reg_state = NETREG_UNREGISTERING;
3075
3076 synchronize_net();
3077
3078 /* Shutdown queueing discipline. */
3079 dev_shutdown(dev);
3080
3081
3082 /* Notify protocols, that we are about to destroy
3083 this device. They should clean all the things.
3084 */
3085 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3086
3087 /*
3088 * Flush the multicast chain
3089 */
3090 dev_mc_discard(dev);
3091
3092 if (dev->uninit)
3093 dev->uninit(dev);
3094
3095 /* Notifier chain MUST detach us from master device. */
3096 BUG_TRAP(!dev->master);
3097
3098 free_divert_blk(dev);
3099
3100 /* Finish processing unregister after unlock */
3101 net_set_todo(dev);
3102
3103 synchronize_net();
3104
3105 dev_put(dev);
3106 return 0;
3107}
3108
3109/**
3110 * unregister_netdev - remove device from the kernel
3111 * @dev: device
3112 *
3113 * This function shuts down a device interface and removes it
3114 * from the kernel tables. On success 0 is returned, on a failure
3115 * a negative errno code is returned.
3116 *
3117 * This is just a wrapper for unregister_netdevice that takes
3118 * the rtnl semaphore. In general you want to use this and not
3119 * unregister_netdevice.
3120 */
3121void unregister_netdev(struct net_device *dev)
3122{
3123 rtnl_lock();
3124 unregister_netdevice(dev);
3125 rtnl_unlock();
3126}
3127
3128EXPORT_SYMBOL(unregister_netdev);
3129
3130#ifdef CONFIG_HOTPLUG_CPU
3131static int dev_cpu_callback(struct notifier_block *nfb,
3132 unsigned long action,
3133 void *ocpu)
3134{
3135 struct sk_buff **list_skb;
3136 struct net_device **list_net;
3137 struct sk_buff *skb;
3138 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3139 struct softnet_data *sd, *oldsd;
3140
3141 if (action != CPU_DEAD)
3142 return NOTIFY_OK;
3143
3144 local_irq_disable();
3145 cpu = smp_processor_id();
3146 sd = &per_cpu(softnet_data, cpu);
3147 oldsd = &per_cpu(softnet_data, oldcpu);
3148
3149 /* Find end of our completion_queue. */
3150 list_skb = &sd->completion_queue;
3151 while (*list_skb)
3152 list_skb = &(*list_skb)->next;
3153 /* Append completion queue from offline CPU. */
3154 *list_skb = oldsd->completion_queue;
3155 oldsd->completion_queue = NULL;
3156
3157 /* Find end of our output_queue. */
3158 list_net = &sd->output_queue;
3159 while (*list_net)
3160 list_net = &(*list_net)->next_sched;
3161 /* Append output queue from offline CPU. */
3162 *list_net = oldsd->output_queue;
3163 oldsd->output_queue = NULL;
3164
3165 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3166 local_irq_enable();
3167
3168 /* Process offline CPU's input_pkt_queue */
3169 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3170 netif_rx(skb);
3171
3172 return NOTIFY_OK;
3173}
3174#endif /* CONFIG_HOTPLUG_CPU */
3175
3176
3177/*
3178 * Initialize the DEV module. At boot time this walks the device list and
3179 * unhooks any devices that fail to initialise (normally hardware not
3180 * present) and leaves us with a valid list of present and active devices.
3181 *
3182 */
3183
3184/*
3185 * This is called single threaded during boot, so no need
3186 * to take the rtnl semaphore.
3187 */
3188static int __init net_dev_init(void)
3189{
3190 int i, rc = -ENOMEM;
3191
3192 BUG_ON(!dev_boot_phase);
3193
3194 net_random_init();
3195
3196 if (dev_proc_init())
3197 goto out;
3198
3199 if (netdev_sysfs_init())
3200 goto out;
3201
3202 INIT_LIST_HEAD(&ptype_all);
3203 for (i = 0; i < 16; i++)
3204 INIT_LIST_HEAD(&ptype_base[i]);
3205
3206 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3207 INIT_HLIST_HEAD(&dev_name_head[i]);
3208
3209 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3210 INIT_HLIST_HEAD(&dev_index_head[i]);
3211
3212 /*
3213 * Initialise the packet receive queues.
3214 */
3215
3216 for (i = 0; i < NR_CPUS; i++) {
3217 struct softnet_data *queue;
3218
3219 queue = &per_cpu(softnet_data, i);
3220 skb_queue_head_init(&queue->input_pkt_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221 queue->completion_queue = NULL;
3222 INIT_LIST_HEAD(&queue->poll_list);
3223 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3224 queue->backlog_dev.weight = weight_p;
3225 queue->backlog_dev.poll = process_backlog;
3226 atomic_set(&queue->backlog_dev.refcnt, 1);
3227 }
3228
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229 dev_boot_phase = 0;
3230
3231 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3232 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3233
3234 hotcpu_notifier(dev_cpu_callback, 0);
3235 dst_init();
3236 dev_mcast_init();
3237 rc = 0;
3238out:
3239 return rc;
3240}
3241
3242subsys_initcall(net_dev_init);
3243
3244EXPORT_SYMBOL(__dev_get_by_index);
3245EXPORT_SYMBOL(__dev_get_by_name);
3246EXPORT_SYMBOL(__dev_remove_pack);
3247EXPORT_SYMBOL(__skb_linearize);
3248EXPORT_SYMBOL(dev_add_pack);
3249EXPORT_SYMBOL(dev_alloc_name);
3250EXPORT_SYMBOL(dev_close);
3251EXPORT_SYMBOL(dev_get_by_flags);
3252EXPORT_SYMBOL(dev_get_by_index);
3253EXPORT_SYMBOL(dev_get_by_name);
3254EXPORT_SYMBOL(dev_ioctl);
3255EXPORT_SYMBOL(dev_open);
3256EXPORT_SYMBOL(dev_queue_xmit);
3257EXPORT_SYMBOL(dev_remove_pack);
3258EXPORT_SYMBOL(dev_set_allmulti);
3259EXPORT_SYMBOL(dev_set_promiscuity);
3260EXPORT_SYMBOL(dev_change_flags);
3261EXPORT_SYMBOL(dev_set_mtu);
3262EXPORT_SYMBOL(dev_set_mac_address);
3263EXPORT_SYMBOL(free_netdev);
3264EXPORT_SYMBOL(netdev_boot_setup_check);
3265EXPORT_SYMBOL(netdev_set_master);
3266EXPORT_SYMBOL(netdev_state_change);
3267EXPORT_SYMBOL(netif_receive_skb);
3268EXPORT_SYMBOL(netif_rx);
3269EXPORT_SYMBOL(register_gifconf);
3270EXPORT_SYMBOL(register_netdevice);
3271EXPORT_SYMBOL(register_netdevice_notifier);
3272EXPORT_SYMBOL(skb_checksum_help);
3273EXPORT_SYMBOL(synchronize_net);
3274EXPORT_SYMBOL(unregister_netdevice);
3275EXPORT_SYMBOL(unregister_netdevice_notifier);
3276EXPORT_SYMBOL(net_enable_timestamp);
3277EXPORT_SYMBOL(net_disable_timestamp);
3278EXPORT_SYMBOL(dev_get_flags);
3279
3280#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3281EXPORT_SYMBOL(br_handle_frame_hook);
3282EXPORT_SYMBOL(br_fdb_get_hook);
3283EXPORT_SYMBOL(br_fdb_put_hook);
3284#endif
3285
3286#ifdef CONFIG_KMOD
3287EXPORT_SYMBOL(dev_load);
3288#endif
3289
3290EXPORT_PER_CPU_SYMBOL(softnet_data);