]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/core/dev.c
Automatic merge of 'for-linus' branch from
[mirror_ubuntu-bionic-kernel.git] / net / core / dev.c
CommitLineData
1da177e4
LT
1/*
2 * NET3 Protocol independent device support routines.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Derived from the non IP parts of dev.c 1.0.19
02c30a84 10 * Authors: Ross Biro
1da177e4
LT
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 *
14 * Additional Authors:
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
21 *
22 * Changes:
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
34 * drivers
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
44 * call a packet.
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
58 * 1 device.
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
66 * the backlog queue.
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
73 */
74
75#include <asm/uaccess.h>
76#include <asm/system.h>
77#include <linux/bitops.h>
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
118/* This define, if set, will randomly drop a packet when congestion
119 * is more than moderate. It helps fairness in the multi-interface
120 * case when one of them is a hog, but it kills performance for the
121 * single interface case so it is off now by default.
122 */
123#undef RAND_LIE
124
125/* Setting this will sample the queue lengths and thus congestion
126 * via a timer instead of as each packet is received.
127 */
128#undef OFFLINE_SAMPLE
129
130/*
131 * The list of packet types we will receive (as opposed to discard)
132 * and the routines to invoke.
133 *
134 * Why 16. Because with 16 the only overlap we get on a hash of the
135 * low nibble of the protocol value is RARP/SNAP/X.25.
136 *
137 * NOTE: That is no longer true with the addition of VLAN tags. Not
138 * sure which should go first, but I bet it won't make much
139 * difference if we are running VLANs. The good news is that
140 * this protocol won't be in the list unless compiled in, so
141 * the average user (w/out VLANs) will not be adversly affected.
142 * --BLG
143 *
144 * 0800 IP
145 * 8100 802.1Q VLAN
146 * 0001 802.3
147 * 0002 AX.25
148 * 0004 802.2
149 * 8035 RARP
150 * 0005 SNAP
151 * 0805 X.25
152 * 0806 ARP
153 * 8137 IPX
154 * 0009 Localtalk
155 * 86DD IPv6
156 */
157
158static DEFINE_SPINLOCK(ptype_lock);
159static struct list_head ptype_base[16]; /* 16 way hashed list */
160static struct list_head ptype_all; /* Taps */
161
162#ifdef OFFLINE_SAMPLE
163static void sample_queue(unsigned long dummy);
164static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
165#endif
166
167/*
168 * The @dev_base list is protected by @dev_base_lock and the rtln
169 * semaphore.
170 *
171 * Pure readers hold dev_base_lock for reading.
172 *
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
177 *
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
181 *
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
184 * semaphore held.
185 */
186struct net_device *dev_base;
187static struct net_device **dev_tail = &dev_base;
188DEFINE_RWLOCK(dev_base_lock);
189
190EXPORT_SYMBOL(dev_base);
191EXPORT_SYMBOL(dev_base_lock);
192
193#define NETDEV_HASHBITS 8
194static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
195static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
196
197static inline struct hlist_head *dev_name_hash(const char *name)
198{
199 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
200 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
201}
202
203static inline struct hlist_head *dev_index_hash(int ifindex)
204{
205 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
206}
207
208/*
209 * Our notifier list
210 */
211
212static struct notifier_block *netdev_chain;
213
214/*
215 * Device drivers call our routines to queue packets here. We empty the
216 * queue in the local softnet handler.
217 */
218DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
219
220#ifdef CONFIG_SYSFS
221extern int netdev_sysfs_init(void);
222extern int netdev_register_sysfs(struct net_device *);
223extern void netdev_unregister_sysfs(struct net_device *);
224#else
225#define netdev_sysfs_init() (0)
226#define netdev_register_sysfs(dev) (0)
227#define netdev_unregister_sysfs(dev) do { } while(0)
228#endif
229
230
231/*******************************************************************************
232
233 Protocol management and registration routines
234
235*******************************************************************************/
236
237/*
238 * For efficiency
239 */
240
241int netdev_nit;
242
243/*
244 * Add a protocol ID to the list. Now that the input handler is
245 * smarter we can dispense with all the messy stuff that used to be
246 * here.
247 *
248 * BEWARE!!! Protocol handlers, mangling input packets,
249 * MUST BE last in hash buckets and checking protocol handlers
250 * MUST start from promiscuous ptype_all chain in net_bh.
251 * It is true now, do not change it.
252 * Explanation follows: if protocol handler, mangling packet, will
253 * be the first on list, it is not able to sense, that packet
254 * is cloned and should be copied-on-write, so that it will
255 * change it and subsequent readers will get broken packet.
256 * --ANK (980803)
257 */
258
259/**
260 * dev_add_pack - add packet handler
261 * @pt: packet type declaration
262 *
263 * Add a protocol handler to the networking stack. The passed &packet_type
264 * is linked into kernel lists and may not be freed until it has been
265 * removed from the kernel lists.
266 *
267 * This call does not sleep therefore it can not
268 * guarantee all CPU's that are in middle of receiving packets
269 * will see the new packet type (until the next received packet).
270 */
271
272void dev_add_pack(struct packet_type *pt)
273{
274 int hash;
275
276 spin_lock_bh(&ptype_lock);
277 if (pt->type == htons(ETH_P_ALL)) {
278 netdev_nit++;
279 list_add_rcu(&pt->list, &ptype_all);
280 } else {
281 hash = ntohs(pt->type) & 15;
282 list_add_rcu(&pt->list, &ptype_base[hash]);
283 }
284 spin_unlock_bh(&ptype_lock);
285}
286
287extern void linkwatch_run_queue(void);
288
289
290
291/**
292 * __dev_remove_pack - remove packet handler
293 * @pt: packet type declaration
294 *
295 * Remove a protocol handler that was previously added to the kernel
296 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
297 * from the kernel lists and can be freed or reused once this function
298 * returns.
299 *
300 * The packet type might still be in use by receivers
301 * and must not be freed until after all the CPU's have gone
302 * through a quiescent state.
303 */
304void __dev_remove_pack(struct packet_type *pt)
305{
306 struct list_head *head;
307 struct packet_type *pt1;
308
309 spin_lock_bh(&ptype_lock);
310
311 if (pt->type == htons(ETH_P_ALL)) {
312 netdev_nit--;
313 head = &ptype_all;
314 } else
315 head = &ptype_base[ntohs(pt->type) & 15];
316
317 list_for_each_entry(pt1, head, list) {
318 if (pt == pt1) {
319 list_del_rcu(&pt->list);
320 goto out;
321 }
322 }
323
324 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
325out:
326 spin_unlock_bh(&ptype_lock);
327}
328/**
329 * dev_remove_pack - remove packet handler
330 * @pt: packet type declaration
331 *
332 * Remove a protocol handler that was previously added to the kernel
333 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
334 * from the kernel lists and can be freed or reused once this function
335 * returns.
336 *
337 * This call sleeps to guarantee that no CPU is looking at the packet
338 * type after return.
339 */
340void dev_remove_pack(struct packet_type *pt)
341{
342 __dev_remove_pack(pt);
343
344 synchronize_net();
345}
346
347/******************************************************************************
348
349 Device Boot-time Settings Routines
350
351*******************************************************************************/
352
353/* Boot time configuration table */
354static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
355
356/**
357 * netdev_boot_setup_add - add new setup entry
358 * @name: name of the device
359 * @map: configured settings for the device
360 *
361 * Adds new setup entry to the dev_boot_setup list. The function
362 * returns 0 on error and 1 on success. This is a generic routine to
363 * all netdevices.
364 */
365static int netdev_boot_setup_add(char *name, struct ifmap *map)
366{
367 struct netdev_boot_setup *s;
368 int i;
369
370 s = dev_boot_setup;
371 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
372 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
373 memset(s[i].name, 0, sizeof(s[i].name));
374 strcpy(s[i].name, name);
375 memcpy(&s[i].map, map, sizeof(s[i].map));
376 break;
377 }
378 }
379
380 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
381}
382
383/**
384 * netdev_boot_setup_check - check boot time settings
385 * @dev: the netdevice
386 *
387 * Check boot time settings for the device.
388 * The found settings are set for the device to be used
389 * later in the device probing.
390 * Returns 0 if no settings found, 1 if they are.
391 */
392int netdev_boot_setup_check(struct net_device *dev)
393{
394 struct netdev_boot_setup *s = dev_boot_setup;
395 int i;
396
397 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
398 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
399 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
400 dev->irq = s[i].map.irq;
401 dev->base_addr = s[i].map.base_addr;
402 dev->mem_start = s[i].map.mem_start;
403 dev->mem_end = s[i].map.mem_end;
404 return 1;
405 }
406 }
407 return 0;
408}
409
410
411/**
412 * netdev_boot_base - get address from boot time settings
413 * @prefix: prefix for network device
414 * @unit: id for network device
415 *
416 * Check boot time settings for the base address of device.
417 * The found settings are set for the device to be used
418 * later in the device probing.
419 * Returns 0 if no settings found.
420 */
421unsigned long netdev_boot_base(const char *prefix, int unit)
422{
423 const struct netdev_boot_setup *s = dev_boot_setup;
424 char name[IFNAMSIZ];
425 int i;
426
427 sprintf(name, "%s%d", prefix, unit);
428
429 /*
430 * If device already registered then return base of 1
431 * to indicate not to probe for this interface
432 */
433 if (__dev_get_by_name(name))
434 return 1;
435
436 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
437 if (!strcmp(name, s[i].name))
438 return s[i].map.base_addr;
439 return 0;
440}
441
442/*
443 * Saves at boot time configured settings for any netdevice.
444 */
445int __init netdev_boot_setup(char *str)
446{
447 int ints[5];
448 struct ifmap map;
449
450 str = get_options(str, ARRAY_SIZE(ints), ints);
451 if (!str || !*str)
452 return 0;
453
454 /* Save settings */
455 memset(&map, 0, sizeof(map));
456 if (ints[0] > 0)
457 map.irq = ints[1];
458 if (ints[0] > 1)
459 map.base_addr = ints[2];
460 if (ints[0] > 2)
461 map.mem_start = ints[3];
462 if (ints[0] > 3)
463 map.mem_end = ints[4];
464
465 /* Add new entry to the list */
466 return netdev_boot_setup_add(str, &map);
467}
468
469__setup("netdev=", netdev_boot_setup);
470
471/*******************************************************************************
472
473 Device Interface Subroutines
474
475*******************************************************************************/
476
477/**
478 * __dev_get_by_name - find a device by its name
479 * @name: name to find
480 *
481 * Find an interface by name. Must be called under RTNL semaphore
482 * or @dev_base_lock. If the name is found a pointer to the device
483 * is returned. If the name is not found then %NULL is returned. The
484 * reference counters are not incremented so the caller must be
485 * careful with locks.
486 */
487
488struct net_device *__dev_get_by_name(const char *name)
489{
490 struct hlist_node *p;
491
492 hlist_for_each(p, dev_name_hash(name)) {
493 struct net_device *dev
494 = hlist_entry(p, struct net_device, name_hlist);
495 if (!strncmp(dev->name, name, IFNAMSIZ))
496 return dev;
497 }
498 return NULL;
499}
500
501/**
502 * dev_get_by_name - find a device by its name
503 * @name: name to find
504 *
505 * Find an interface by name. This can be called from any
506 * context and does its own locking. The returned handle has
507 * the usage count incremented and the caller must use dev_put() to
508 * release it when it is no longer needed. %NULL is returned if no
509 * matching device is found.
510 */
511
512struct net_device *dev_get_by_name(const char *name)
513{
514 struct net_device *dev;
515
516 read_lock(&dev_base_lock);
517 dev = __dev_get_by_name(name);
518 if (dev)
519 dev_hold(dev);
520 read_unlock(&dev_base_lock);
521 return dev;
522}
523
524/**
525 * __dev_get_by_index - find a device by its ifindex
526 * @ifindex: index of device
527 *
528 * Search for an interface by index. Returns %NULL if the device
529 * is not found or a pointer to the device. The device has not
530 * had its reference counter increased so the caller must be careful
531 * about locking. The caller must hold either the RTNL semaphore
532 * or @dev_base_lock.
533 */
534
535struct net_device *__dev_get_by_index(int ifindex)
536{
537 struct hlist_node *p;
538
539 hlist_for_each(p, dev_index_hash(ifindex)) {
540 struct net_device *dev
541 = hlist_entry(p, struct net_device, index_hlist);
542 if (dev->ifindex == ifindex)
543 return dev;
544 }
545 return NULL;
546}
547
548
549/**
550 * dev_get_by_index - find a device by its ifindex
551 * @ifindex: index of device
552 *
553 * Search for an interface by index. Returns NULL if the device
554 * is not found or a pointer to the device. The device returned has
555 * had a reference added and the pointer is safe until the user calls
556 * dev_put to indicate they have finished with it.
557 */
558
559struct net_device *dev_get_by_index(int ifindex)
560{
561 struct net_device *dev;
562
563 read_lock(&dev_base_lock);
564 dev = __dev_get_by_index(ifindex);
565 if (dev)
566 dev_hold(dev);
567 read_unlock(&dev_base_lock);
568 return dev;
569}
570
571/**
572 * dev_getbyhwaddr - find a device by its hardware address
573 * @type: media type of device
574 * @ha: hardware address
575 *
576 * Search for an interface by MAC address. Returns NULL if the device
577 * is not found or a pointer to the device. The caller must hold the
578 * rtnl semaphore. The returned device has not had its ref count increased
579 * and the caller must therefore be careful about locking
580 *
581 * BUGS:
582 * If the API was consistent this would be __dev_get_by_hwaddr
583 */
584
585struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
586{
587 struct net_device *dev;
588
589 ASSERT_RTNL();
590
591 for (dev = dev_base; dev; dev = dev->next)
592 if (dev->type == type &&
593 !memcmp(dev->dev_addr, ha, dev->addr_len))
594 break;
595 return dev;
596}
597
598struct net_device *dev_getfirstbyhwtype(unsigned short type)
599{
600 struct net_device *dev;
601
602 rtnl_lock();
603 for (dev = dev_base; dev; dev = dev->next) {
604 if (dev->type == type) {
605 dev_hold(dev);
606 break;
607 }
608 }
609 rtnl_unlock();
610 return dev;
611}
612
613EXPORT_SYMBOL(dev_getfirstbyhwtype);
614
615/**
616 * dev_get_by_flags - find any device with given flags
617 * @if_flags: IFF_* values
618 * @mask: bitmask of bits in if_flags to check
619 *
620 * Search for any interface with the given flags. Returns NULL if a device
621 * is not found or a pointer to the device. The device returned has
622 * had a reference added and the pointer is safe until the user calls
623 * dev_put to indicate they have finished with it.
624 */
625
626struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
627{
628 struct net_device *dev;
629
630 read_lock(&dev_base_lock);
631 for (dev = dev_base; dev != NULL; dev = dev->next) {
632 if (((dev->flags ^ if_flags) & mask) == 0) {
633 dev_hold(dev);
634 break;
635 }
636 }
637 read_unlock(&dev_base_lock);
638 return dev;
639}
640
641/**
642 * dev_valid_name - check if name is okay for network device
643 * @name: name string
644 *
645 * Network device names need to be valid file names to
646 * to allow sysfs to work
647 */
648static int dev_valid_name(const char *name)
649{
650 return !(*name == '\0'
651 || !strcmp(name, ".")
652 || !strcmp(name, "..")
653 || strchr(name, '/'));
654}
655
656/**
657 * dev_alloc_name - allocate a name for a device
658 * @dev: device
659 * @name: name format string
660 *
661 * Passed a format string - eg "lt%d" it will try and find a suitable
662 * id. Not efficient for many devices, not called a lot. The caller
663 * must hold the dev_base or rtnl lock while allocating the name and
664 * adding the device in order to avoid duplicates. Returns the number
665 * of the unit assigned or a negative errno code.
666 */
667
668int dev_alloc_name(struct net_device *dev, const char *name)
669{
670 int i = 0;
671 char buf[IFNAMSIZ];
672 const char *p;
673 const int max_netdevices = 8*PAGE_SIZE;
674 long *inuse;
675 struct net_device *d;
676
677 p = strnchr(name, IFNAMSIZ-1, '%');
678 if (p) {
679 /*
680 * Verify the string as this thing may have come from
681 * the user. There must be either one "%d" and no other "%"
682 * characters.
683 */
684 if (p[1] != 'd' || strchr(p + 2, '%'))
685 return -EINVAL;
686
687 /* Use one page as a bit array of possible slots */
688 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
689 if (!inuse)
690 return -ENOMEM;
691
692 for (d = dev_base; d; d = d->next) {
693 if (!sscanf(d->name, name, &i))
694 continue;
695 if (i < 0 || i >= max_netdevices)
696 continue;
697
698 /* avoid cases where sscanf is not exact inverse of printf */
699 snprintf(buf, sizeof(buf), name, i);
700 if (!strncmp(buf, d->name, IFNAMSIZ))
701 set_bit(i, inuse);
702 }
703
704 i = find_first_zero_bit(inuse, max_netdevices);
705 free_page((unsigned long) inuse);
706 }
707
708 snprintf(buf, sizeof(buf), name, i);
709 if (!__dev_get_by_name(buf)) {
710 strlcpy(dev->name, buf, IFNAMSIZ);
711 return i;
712 }
713
714 /* It is possible to run out of possible slots
715 * when the name is long and there isn't enough space left
716 * for the digits, or if all bits are used.
717 */
718 return -ENFILE;
719}
720
721
722/**
723 * dev_change_name - change name of a device
724 * @dev: device
725 * @newname: name (or format string) must be at least IFNAMSIZ
726 *
727 * Change name of a device, can pass format strings "eth%d".
728 * for wildcarding.
729 */
730int dev_change_name(struct net_device *dev, char *newname)
731{
732 int err = 0;
733
734 ASSERT_RTNL();
735
736 if (dev->flags & IFF_UP)
737 return -EBUSY;
738
739 if (!dev_valid_name(newname))
740 return -EINVAL;
741
742 if (strchr(newname, '%')) {
743 err = dev_alloc_name(dev, newname);
744 if (err < 0)
745 return err;
746 strcpy(newname, dev->name);
747 }
748 else if (__dev_get_by_name(newname))
749 return -EEXIST;
750 else
751 strlcpy(dev->name, newname, IFNAMSIZ);
752
753 err = class_device_rename(&dev->class_dev, dev->name);
754 if (!err) {
755 hlist_del(&dev->name_hlist);
756 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
757 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
758 }
759
760 return err;
761}
762
763/**
764 * netdev_state_change - device changes state
765 * @dev: device to cause notification
766 *
767 * Called to indicate a device has changed state. This function calls
768 * the notifier chains for netdev_chain and sends a NEWLINK message
769 * to the routing socket.
770 */
771void netdev_state_change(struct net_device *dev)
772{
773 if (dev->flags & IFF_UP) {
774 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
775 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
776 }
777}
778
779/**
780 * dev_load - load a network module
781 * @name: name of interface
782 *
783 * If a network interface is not present and the process has suitable
784 * privileges this function loads the module. If module loading is not
785 * available in this kernel then it becomes a nop.
786 */
787
788void dev_load(const char *name)
789{
790 struct net_device *dev;
791
792 read_lock(&dev_base_lock);
793 dev = __dev_get_by_name(name);
794 read_unlock(&dev_base_lock);
795
796 if (!dev && capable(CAP_SYS_MODULE))
797 request_module("%s", name);
798}
799
800static int default_rebuild_header(struct sk_buff *skb)
801{
802 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
803 skb->dev ? skb->dev->name : "NULL!!!");
804 kfree_skb(skb);
805 return 1;
806}
807
808
809/**
810 * dev_open - prepare an interface for use.
811 * @dev: device to open
812 *
813 * Takes a device from down to up state. The device's private open
814 * function is invoked and then the multicast lists are loaded. Finally
815 * the device is moved into the up state and a %NETDEV_UP message is
816 * sent to the netdev notifier chain.
817 *
818 * Calling this function on an active interface is a nop. On a failure
819 * a negative errno code is returned.
820 */
821int dev_open(struct net_device *dev)
822{
823 int ret = 0;
824
825 /*
826 * Is it already up?
827 */
828
829 if (dev->flags & IFF_UP)
830 return 0;
831
832 /*
833 * Is it even present?
834 */
835 if (!netif_device_present(dev))
836 return -ENODEV;
837
838 /*
839 * Call device private open method
840 */
841 set_bit(__LINK_STATE_START, &dev->state);
842 if (dev->open) {
843 ret = dev->open(dev);
844 if (ret)
845 clear_bit(__LINK_STATE_START, &dev->state);
846 }
847
848 /*
849 * If it went open OK then:
850 */
851
852 if (!ret) {
853 /*
854 * Set the flags.
855 */
856 dev->flags |= IFF_UP;
857
858 /*
859 * Initialize multicasting status
860 */
861 dev_mc_upload(dev);
862
863 /*
864 * Wakeup transmit queue engine
865 */
866 dev_activate(dev);
867
868 /*
869 * ... and announce new interface.
870 */
871 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
872 }
873 return ret;
874}
875
876/**
877 * dev_close - shutdown an interface.
878 * @dev: device to shutdown
879 *
880 * This function moves an active device into down state. A
881 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
882 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
883 * chain.
884 */
885int dev_close(struct net_device *dev)
886{
887 if (!(dev->flags & IFF_UP))
888 return 0;
889
890 /*
891 * Tell people we are going down, so that they can
892 * prepare to death, when device is still operating.
893 */
894 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
895
896 dev_deactivate(dev);
897
898 clear_bit(__LINK_STATE_START, &dev->state);
899
900 /* Synchronize to scheduled poll. We cannot touch poll list,
901 * it can be even on different cpu. So just clear netif_running(),
902 * and wait when poll really will happen. Actually, the best place
903 * for this is inside dev->stop() after device stopped its irq
904 * engine, but this requires more changes in devices. */
905
906 smp_mb__after_clear_bit(); /* Commit netif_running(). */
907 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
908 /* No hurry. */
909 current->state = TASK_INTERRUPTIBLE;
910 schedule_timeout(1);
911 }
912
913 /*
914 * Call the device specific close. This cannot fail.
915 * Only if device is UP
916 *
917 * We allow it to be called even after a DETACH hot-plug
918 * event.
919 */
920 if (dev->stop)
921 dev->stop(dev);
922
923 /*
924 * Device is now down.
925 */
926
927 dev->flags &= ~IFF_UP;
928
929 /*
930 * Tell people we are down
931 */
932 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
933
934 return 0;
935}
936
937
938/*
939 * Device change register/unregister. These are not inline or static
940 * as we export them to the world.
941 */
942
943/**
944 * register_netdevice_notifier - register a network notifier block
945 * @nb: notifier
946 *
947 * Register a notifier to be called when network device events occur.
948 * The notifier passed is linked into the kernel structures and must
949 * not be reused until it has been unregistered. A negative errno code
950 * is returned on a failure.
951 *
952 * When registered all registration and up events are replayed
953 * to the new notifier to allow device to have a race free
954 * view of the network device list.
955 */
956
957int register_netdevice_notifier(struct notifier_block *nb)
958{
959 struct net_device *dev;
960 int err;
961
962 rtnl_lock();
963 err = notifier_chain_register(&netdev_chain, nb);
964 if (!err) {
965 for (dev = dev_base; dev; dev = dev->next) {
966 nb->notifier_call(nb, NETDEV_REGISTER, dev);
967
968 if (dev->flags & IFF_UP)
969 nb->notifier_call(nb, NETDEV_UP, dev);
970 }
971 }
972 rtnl_unlock();
973 return err;
974}
975
976/**
977 * unregister_netdevice_notifier - unregister a network notifier block
978 * @nb: notifier
979 *
980 * Unregister a notifier previously registered by
981 * register_netdevice_notifier(). The notifier is unlinked into the
982 * kernel structures and may then be reused. A negative errno code
983 * is returned on a failure.
984 */
985
986int unregister_netdevice_notifier(struct notifier_block *nb)
987{
988 return notifier_chain_unregister(&netdev_chain, nb);
989}
990
991/**
992 * call_netdevice_notifiers - call all network notifier blocks
993 * @val: value passed unmodified to notifier function
994 * @v: pointer passed unmodified to notifier function
995 *
996 * Call all network notifier blocks. Parameters and return value
997 * are as for notifier_call_chain().
998 */
999
1000int call_netdevice_notifiers(unsigned long val, void *v)
1001{
1002 return notifier_call_chain(&netdev_chain, val, v);
1003}
1004
1005/* When > 0 there are consumers of rx skb time stamps */
1006static atomic_t netstamp_needed = ATOMIC_INIT(0);
1007
1008void net_enable_timestamp(void)
1009{
1010 atomic_inc(&netstamp_needed);
1011}
1012
1013void net_disable_timestamp(void)
1014{
1015 atomic_dec(&netstamp_needed);
1016}
1017
1018static inline void net_timestamp(struct timeval *stamp)
1019{
1020 if (atomic_read(&netstamp_needed))
1021 do_gettimeofday(stamp);
1022 else {
1023 stamp->tv_sec = 0;
1024 stamp->tv_usec = 0;
1025 }
1026}
1027
1028/*
1029 * Support routine. Sends outgoing frames to any network
1030 * taps currently in use.
1031 */
1032
1033void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1034{
1035 struct packet_type *ptype;
1036 net_timestamp(&skb->stamp);
1037
1038 rcu_read_lock();
1039 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1040 /* Never send packets back to the socket
1041 * they originated from - MvS (miquels@drinkel.ow.org)
1042 */
1043 if ((ptype->dev == dev || !ptype->dev) &&
1044 (ptype->af_packet_priv == NULL ||
1045 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1046 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1047 if (!skb2)
1048 break;
1049
1050 /* skb->nh should be correctly
1051 set by sender, so that the second statement is
1052 just protection against buggy protocols.
1053 */
1054 skb2->mac.raw = skb2->data;
1055
1056 if (skb2->nh.raw < skb2->data ||
1057 skb2->nh.raw > skb2->tail) {
1058 if (net_ratelimit())
1059 printk(KERN_CRIT "protocol %04x is "
1060 "buggy, dev %s\n",
1061 skb2->protocol, dev->name);
1062 skb2->nh.raw = skb2->data;
1063 }
1064
1065 skb2->h.raw = skb2->nh.raw;
1066 skb2->pkt_type = PACKET_OUTGOING;
1067 ptype->func(skb2, skb->dev, ptype);
1068 }
1069 }
1070 rcu_read_unlock();
1071}
1072
1073/*
1074 * Invalidate hardware checksum when packet is to be mangled, and
1075 * complete checksum manually on outgoing path.
1076 */
1077int skb_checksum_help(struct sk_buff *skb, int inward)
1078{
1079 unsigned int csum;
1080 int ret = 0, offset = skb->h.raw - skb->data;
1081
1082 if (inward) {
1083 skb->ip_summed = CHECKSUM_NONE;
1084 goto out;
1085 }
1086
1087 if (skb_cloned(skb)) {
1088 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1089 if (ret)
1090 goto out;
1091 }
1092
1093 if (offset > (int)skb->len)
1094 BUG();
1095 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1096
1097 offset = skb->tail - skb->h.raw;
1098 if (offset <= 0)
1099 BUG();
1100 if (skb->csum + 2 > offset)
1101 BUG();
1102
1103 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1104 skb->ip_summed = CHECKSUM_NONE;
1105out:
1106 return ret;
1107}
1108
1109#ifdef CONFIG_HIGHMEM
1110/* Actually, we should eliminate this check as soon as we know, that:
1111 * 1. IOMMU is present and allows to map all the memory.
1112 * 2. No high memory really exists on this machine.
1113 */
1114
1115static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1116{
1117 int i;
1118
1119 if (dev->features & NETIF_F_HIGHDMA)
1120 return 0;
1121
1122 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1123 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1124 return 1;
1125
1126 return 0;
1127}
1128#else
1129#define illegal_highdma(dev, skb) (0)
1130#endif
1131
1132extern void skb_release_data(struct sk_buff *);
1133
1134/* Keep head the same: replace data */
1135int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1136{
1137 unsigned int size;
1138 u8 *data;
1139 long offset;
1140 struct skb_shared_info *ninfo;
1141 int headerlen = skb->data - skb->head;
1142 int expand = (skb->tail + skb->data_len) - skb->end;
1143
1144 if (skb_shared(skb))
1145 BUG();
1146
1147 if (expand <= 0)
1148 expand = 0;
1149
1150 size = skb->end - skb->head + expand;
1151 size = SKB_DATA_ALIGN(size);
1152 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1153 if (!data)
1154 return -ENOMEM;
1155
1156 /* Copy entire thing */
1157 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1158 BUG();
1159
1160 /* Set up shinfo */
1161 ninfo = (struct skb_shared_info*)(data + size);
1162 atomic_set(&ninfo->dataref, 1);
1163 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1164 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1165 ninfo->nr_frags = 0;
1166 ninfo->frag_list = NULL;
1167
1168 /* Offset between the two in bytes */
1169 offset = data - skb->head;
1170
1171 /* Free old data. */
1172 skb_release_data(skb);
1173
1174 skb->head = data;
1175 skb->end = data + size;
1176
1177 /* Set up new pointers */
1178 skb->h.raw += offset;
1179 skb->nh.raw += offset;
1180 skb->mac.raw += offset;
1181 skb->tail += offset;
1182 skb->data += offset;
1183
1184 /* We are no longer a clone, even if we were. */
1185 skb->cloned = 0;
1186
1187 skb->tail += skb->data_len;
1188 skb->data_len = 0;
1189 return 0;
1190}
1191
1192#define HARD_TX_LOCK(dev, cpu) { \
1193 if ((dev->features & NETIF_F_LLTX) == 0) { \
1194 spin_lock(&dev->xmit_lock); \
1195 dev->xmit_lock_owner = cpu; \
1196 } \
1197}
1198
1199#define HARD_TX_UNLOCK(dev) { \
1200 if ((dev->features & NETIF_F_LLTX) == 0) { \
1201 dev->xmit_lock_owner = -1; \
1202 spin_unlock(&dev->xmit_lock); \
1203 } \
1204}
1205
1206/**
1207 * dev_queue_xmit - transmit a buffer
1208 * @skb: buffer to transmit
1209 *
1210 * Queue a buffer for transmission to a network device. The caller must
1211 * have set the device and priority and built the buffer before calling
1212 * this function. The function can be called from an interrupt.
1213 *
1214 * A negative errno code is returned on a failure. A success does not
1215 * guarantee the frame will be transmitted as it may be dropped due
1216 * to congestion or traffic shaping.
af191367
BG
1217 *
1218 * -----------------------------------------------------------------------------------
1219 * I notice this method can also return errors from the queue disciplines,
1220 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1221 * be positive.
1222 *
1223 * Regardless of the return value, the skb is consumed, so it is currently
1224 * difficult to retry a send to this method. (You can bump the ref count
1225 * before sending to hold a reference for retry if you are careful.)
1226 *
1227 * When calling this method, interrupts MUST be enabled. This is because
1228 * the BH enable code must have IRQs enabled so that it will not deadlock.
1229 * --BLG
1da177e4
LT
1230 */
1231
1232int dev_queue_xmit(struct sk_buff *skb)
1233{
1234 struct net_device *dev = skb->dev;
1235 struct Qdisc *q;
1236 int rc = -ENOMEM;
1237
1238 if (skb_shinfo(skb)->frag_list &&
1239 !(dev->features & NETIF_F_FRAGLIST) &&
1240 __skb_linearize(skb, GFP_ATOMIC))
1241 goto out_kfree_skb;
1242
1243 /* Fragmented skb is linearized if device does not support SG,
1244 * or if at least one of fragments is in highmem and device
1245 * does not support DMA from it.
1246 */
1247 if (skb_shinfo(skb)->nr_frags &&
1248 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1249 __skb_linearize(skb, GFP_ATOMIC))
1250 goto out_kfree_skb;
1251
1252 /* If packet is not checksummed and device does not support
1253 * checksumming for this protocol, complete checksumming here.
1254 */
1255 if (skb->ip_summed == CHECKSUM_HW &&
1256 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1257 (!(dev->features & NETIF_F_IP_CSUM) ||
1258 skb->protocol != htons(ETH_P_IP))))
1259 if (skb_checksum_help(skb, 0))
1260 goto out_kfree_skb;
1261
1262 /* Disable soft irqs for various locks below. Also
1263 * stops preemption for RCU.
1264 */
1265 local_bh_disable();
1266
1267 /* Updates of qdisc are serialized by queue_lock.
1268 * The struct Qdisc which is pointed to by qdisc is now a
1269 * rcu structure - it may be accessed without acquiring
1270 * a lock (but the structure may be stale.) The freeing of the
1271 * qdisc will be deferred until it's known that there are no
1272 * more references to it.
1273 *
1274 * If the qdisc has an enqueue function, we still need to
1275 * hold the queue_lock before calling it, since queue_lock
1276 * also serializes access to the device queue.
1277 */
1278
1279 q = rcu_dereference(dev->qdisc);
1280#ifdef CONFIG_NET_CLS_ACT
1281 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1282#endif
1283 if (q->enqueue) {
1284 /* Grab device queue */
1285 spin_lock(&dev->queue_lock);
1286
1287 rc = q->enqueue(skb, q);
1288
1289 qdisc_run(dev);
1290
1291 spin_unlock(&dev->queue_lock);
1292 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1293 goto out;
1294 }
1295
1296 /* The device has no queue. Common case for software devices:
1297 loopback, all the sorts of tunnels...
1298
1299 Really, it is unlikely that xmit_lock protection is necessary here.
1300 (f.e. loopback and IP tunnels are clean ignoring statistics
1301 counters.)
1302 However, it is possible, that they rely on protection
1303 made by us here.
1304
1305 Check this and shot the lock. It is not prone from deadlocks.
1306 Either shot noqueue qdisc, it is even simpler 8)
1307 */
1308 if (dev->flags & IFF_UP) {
1309 int cpu = smp_processor_id(); /* ok because BHs are off */
1310
1311 if (dev->xmit_lock_owner != cpu) {
1312
1313 HARD_TX_LOCK(dev, cpu);
1314
1315 if (!netif_queue_stopped(dev)) {
1316 if (netdev_nit)
1317 dev_queue_xmit_nit(skb, dev);
1318
1319 rc = 0;
1320 if (!dev->hard_start_xmit(skb, dev)) {
1321 HARD_TX_UNLOCK(dev);
1322 goto out;
1323 }
1324 }
1325 HARD_TX_UNLOCK(dev);
1326 if (net_ratelimit())
1327 printk(KERN_CRIT "Virtual device %s asks to "
1328 "queue packet!\n", dev->name);
1329 } else {
1330 /* Recursion is detected! It is possible,
1331 * unfortunately */
1332 if (net_ratelimit())
1333 printk(KERN_CRIT "Dead loop on virtual device "
1334 "%s, fix it urgently!\n", dev->name);
1335 }
1336 }
1337
1338 rc = -ENETDOWN;
1339 local_bh_enable();
1340
1341out_kfree_skb:
1342 kfree_skb(skb);
1343 return rc;
1344out:
1345 local_bh_enable();
1346 return rc;
1347}
1348
1349
1350/*=======================================================================
1351 Receiver routines
1352 =======================================================================*/
1353
1354int netdev_max_backlog = 300;
1355int weight_p = 64; /* old backlog weight */
1356/* These numbers are selected based on intuition and some
1357 * experimentatiom, if you have more scientific way of doing this
1358 * please go ahead and fix things.
1359 */
1360int no_cong_thresh = 10;
1361int no_cong = 20;
1362int lo_cong = 100;
1363int mod_cong = 290;
1364
1365DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1366
1367
1368static void get_sample_stats(int cpu)
1369{
1370#ifdef RAND_LIE
1371 unsigned long rd;
1372 int rq;
1373#endif
1374 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1375 int blog = sd->input_pkt_queue.qlen;
1376 int avg_blog = sd->avg_blog;
1377
1378 avg_blog = (avg_blog >> 1) + (blog >> 1);
1379
1380 if (avg_blog > mod_cong) {
1381 /* Above moderate congestion levels. */
1382 sd->cng_level = NET_RX_CN_HIGH;
1383#ifdef RAND_LIE
1384 rd = net_random();
1385 rq = rd % netdev_max_backlog;
1386 if (rq < avg_blog) /* unlucky bastard */
1387 sd->cng_level = NET_RX_DROP;
1388#endif
1389 } else if (avg_blog > lo_cong) {
1390 sd->cng_level = NET_RX_CN_MOD;
1391#ifdef RAND_LIE
1392 rd = net_random();
1393 rq = rd % netdev_max_backlog;
1394 if (rq < avg_blog) /* unlucky bastard */
1395 sd->cng_level = NET_RX_CN_HIGH;
1396#endif
1397 } else if (avg_blog > no_cong)
1398 sd->cng_level = NET_RX_CN_LOW;
1399 else /* no congestion */
1400 sd->cng_level = NET_RX_SUCCESS;
1401
1402 sd->avg_blog = avg_blog;
1403}
1404
1405#ifdef OFFLINE_SAMPLE
1406static void sample_queue(unsigned long dummy)
1407{
1408/* 10 ms 0r 1ms -- i don't care -- JHS */
1409 int next_tick = 1;
1410 int cpu = smp_processor_id();
1411
1412 get_sample_stats(cpu);
1413 next_tick += jiffies;
1414 mod_timer(&samp_timer, next_tick);
1415}
1416#endif
1417
1418
1419/**
1420 * netif_rx - post buffer to the network code
1421 * @skb: buffer to post
1422 *
1423 * This function receives a packet from a device driver and queues it for
1424 * the upper (protocol) levels to process. It always succeeds. The buffer
1425 * may be dropped during processing for congestion control or by the
1426 * protocol layers.
1427 *
1428 * return values:
1429 * NET_RX_SUCCESS (no congestion)
1430 * NET_RX_CN_LOW (low congestion)
1431 * NET_RX_CN_MOD (moderate congestion)
1432 * NET_RX_CN_HIGH (high congestion)
1433 * NET_RX_DROP (packet was dropped)
1434 *
1435 */
1436
1437int netif_rx(struct sk_buff *skb)
1438{
1439 int this_cpu;
1440 struct softnet_data *queue;
1441 unsigned long flags;
1442
1443 /* if netpoll wants it, pretend we never saw it */
1444 if (netpoll_rx(skb))
1445 return NET_RX_DROP;
1446
1447 if (!skb->stamp.tv_sec)
1448 net_timestamp(&skb->stamp);
1449
1450 /*
1451 * The code is rearranged so that the path is the most
1452 * short when CPU is congested, but is still operating.
1453 */
1454 local_irq_save(flags);
1455 this_cpu = smp_processor_id();
1456 queue = &__get_cpu_var(softnet_data);
1457
1458 __get_cpu_var(netdev_rx_stat).total++;
1459 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1460 if (queue->input_pkt_queue.qlen) {
1461 if (queue->throttle)
1462 goto drop;
1463
1464enqueue:
1465 dev_hold(skb->dev);
1466 __skb_queue_tail(&queue->input_pkt_queue, skb);
1467#ifndef OFFLINE_SAMPLE
1468 get_sample_stats(this_cpu);
1469#endif
1470 local_irq_restore(flags);
1471 return queue->cng_level;
1472 }
1473
1474 if (queue->throttle)
1475 queue->throttle = 0;
1476
1477 netif_rx_schedule(&queue->backlog_dev);
1478 goto enqueue;
1479 }
1480
1481 if (!queue->throttle) {
1482 queue->throttle = 1;
1483 __get_cpu_var(netdev_rx_stat).throttled++;
1484 }
1485
1486drop:
1487 __get_cpu_var(netdev_rx_stat).dropped++;
1488 local_irq_restore(flags);
1489
1490 kfree_skb(skb);
1491 return NET_RX_DROP;
1492}
1493
1494int netif_rx_ni(struct sk_buff *skb)
1495{
1496 int err;
1497
1498 preempt_disable();
1499 err = netif_rx(skb);
1500 if (local_softirq_pending())
1501 do_softirq();
1502 preempt_enable();
1503
1504 return err;
1505}
1506
1507EXPORT_SYMBOL(netif_rx_ni);
1508
1509static __inline__ void skb_bond(struct sk_buff *skb)
1510{
1511 struct net_device *dev = skb->dev;
1512
1513 if (dev->master) {
1514 skb->real_dev = skb->dev;
1515 skb->dev = dev->master;
1516 }
1517}
1518
1519static void net_tx_action(struct softirq_action *h)
1520{
1521 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1522
1523 if (sd->completion_queue) {
1524 struct sk_buff *clist;
1525
1526 local_irq_disable();
1527 clist = sd->completion_queue;
1528 sd->completion_queue = NULL;
1529 local_irq_enable();
1530
1531 while (clist) {
1532 struct sk_buff *skb = clist;
1533 clist = clist->next;
1534
1535 BUG_TRAP(!atomic_read(&skb->users));
1536 __kfree_skb(skb);
1537 }
1538 }
1539
1540 if (sd->output_queue) {
1541 struct net_device *head;
1542
1543 local_irq_disable();
1544 head = sd->output_queue;
1545 sd->output_queue = NULL;
1546 local_irq_enable();
1547
1548 while (head) {
1549 struct net_device *dev = head;
1550 head = head->next_sched;
1551
1552 smp_mb__before_clear_bit();
1553 clear_bit(__LINK_STATE_SCHED, &dev->state);
1554
1555 if (spin_trylock(&dev->queue_lock)) {
1556 qdisc_run(dev);
1557 spin_unlock(&dev->queue_lock);
1558 } else {
1559 netif_schedule(dev);
1560 }
1561 }
1562 }
1563}
1564
1565static __inline__ int deliver_skb(struct sk_buff *skb,
1566 struct packet_type *pt_prev)
1567{
1568 atomic_inc(&skb->users);
1569 return pt_prev->func(skb, skb->dev, pt_prev);
1570}
1571
1572#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1573int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1574struct net_bridge;
1575struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1576 unsigned char *addr);
1577void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1578
1579static __inline__ int handle_bridge(struct sk_buff **pskb,
1580 struct packet_type **pt_prev, int *ret)
1581{
1582 struct net_bridge_port *port;
1583
1584 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1585 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1586 return 0;
1587
1588 if (*pt_prev) {
1589 *ret = deliver_skb(*pskb, *pt_prev);
1590 *pt_prev = NULL;
1591 }
1592
1593 return br_handle_frame_hook(port, pskb);
1594}
1595#else
1596#define handle_bridge(skb, pt_prev, ret) (0)
1597#endif
1598
1599#ifdef CONFIG_NET_CLS_ACT
1600/* TODO: Maybe we should just force sch_ingress to be compiled in
1601 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1602 * a compare and 2 stores extra right now if we dont have it on
1603 * but have CONFIG_NET_CLS_ACT
1604 * NOTE: This doesnt stop any functionality; if you dont have
1605 * the ingress scheduler, you just cant add policies on ingress.
1606 *
1607 */
1608static int ing_filter(struct sk_buff *skb)
1609{
1610 struct Qdisc *q;
1611 struct net_device *dev = skb->dev;
1612 int result = TC_ACT_OK;
1613
1614 if (dev->qdisc_ingress) {
1615 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1616 if (MAX_RED_LOOP < ttl++) {
1617 printk("Redir loop detected Dropping packet (%s->%s)\n",
1618 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1619 return TC_ACT_SHOT;
1620 }
1621
1622 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1623
1624 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1625 if (NULL == skb->input_dev) {
1626 skb->input_dev = skb->dev;
1627 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1628 }
1629 spin_lock(&dev->ingress_lock);
1630 if ((q = dev->qdisc_ingress) != NULL)
1631 result = q->enqueue(skb, q);
1632 spin_unlock(&dev->ingress_lock);
1633
1634 }
1635
1636 return result;
1637}
1638#endif
1639
1640int netif_receive_skb(struct sk_buff *skb)
1641{
1642 struct packet_type *ptype, *pt_prev;
1643 int ret = NET_RX_DROP;
1644 unsigned short type;
1645
1646 /* if we've gotten here through NAPI, check netpoll */
1647 if (skb->dev->poll && netpoll_rx(skb))
1648 return NET_RX_DROP;
1649
1650 if (!skb->stamp.tv_sec)
1651 net_timestamp(&skb->stamp);
1652
1653 skb_bond(skb);
1654
1655 __get_cpu_var(netdev_rx_stat).total++;
1656
1657 skb->h.raw = skb->nh.raw = skb->data;
1658 skb->mac_len = skb->nh.raw - skb->mac.raw;
1659
1660 pt_prev = NULL;
1661
1662 rcu_read_lock();
1663
1664#ifdef CONFIG_NET_CLS_ACT
1665 if (skb->tc_verd & TC_NCLS) {
1666 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1667 goto ncls;
1668 }
1669#endif
1670
1671 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1672 if (!ptype->dev || ptype->dev == skb->dev) {
1673 if (pt_prev)
1674 ret = deliver_skb(skb, pt_prev);
1675 pt_prev = ptype;
1676 }
1677 }
1678
1679#ifdef CONFIG_NET_CLS_ACT
1680 if (pt_prev) {
1681 ret = deliver_skb(skb, pt_prev);
1682 pt_prev = NULL; /* noone else should process this after*/
1683 } else {
1684 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1685 }
1686
1687 ret = ing_filter(skb);
1688
1689 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1690 kfree_skb(skb);
1691 goto out;
1692 }
1693
1694 skb->tc_verd = 0;
1695ncls:
1696#endif
1697
1698 handle_diverter(skb);
1699
1700 if (handle_bridge(&skb, &pt_prev, &ret))
1701 goto out;
1702
1703 type = skb->protocol;
1704 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1705 if (ptype->type == type &&
1706 (!ptype->dev || ptype->dev == skb->dev)) {
1707 if (pt_prev)
1708 ret = deliver_skb(skb, pt_prev);
1709 pt_prev = ptype;
1710 }
1711 }
1712
1713 if (pt_prev) {
1714 ret = pt_prev->func(skb, skb->dev, pt_prev);
1715 } else {
1716 kfree_skb(skb);
1717 /* Jamal, now you will not able to escape explaining
1718 * me how you were going to use this. :-)
1719 */
1720 ret = NET_RX_DROP;
1721 }
1722
1723out:
1724 rcu_read_unlock();
1725 return ret;
1726}
1727
1728static int process_backlog(struct net_device *backlog_dev, int *budget)
1729{
1730 int work = 0;
1731 int quota = min(backlog_dev->quota, *budget);
1732 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1733 unsigned long start_time = jiffies;
1734
1735 for (;;) {
1736 struct sk_buff *skb;
1737 struct net_device *dev;
1738
1739 local_irq_disable();
1740 skb = __skb_dequeue(&queue->input_pkt_queue);
1741 if (!skb)
1742 goto job_done;
1743 local_irq_enable();
1744
1745 dev = skb->dev;
1746
1747 netif_receive_skb(skb);
1748
1749 dev_put(dev);
1750
1751 work++;
1752
1753 if (work >= quota || jiffies - start_time > 1)
1754 break;
1755
1756 }
1757
1758 backlog_dev->quota -= work;
1759 *budget -= work;
1760 return -1;
1761
1762job_done:
1763 backlog_dev->quota -= work;
1764 *budget -= work;
1765
1766 list_del(&backlog_dev->poll_list);
1767 smp_mb__before_clear_bit();
1768 netif_poll_enable(backlog_dev);
1769
1770 if (queue->throttle)
1771 queue->throttle = 0;
1772 local_irq_enable();
1773 return 0;
1774}
1775
1776static void net_rx_action(struct softirq_action *h)
1777{
1778 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1779 unsigned long start_time = jiffies;
1780 int budget = netdev_max_backlog;
1781
1782
1783 local_irq_disable();
1784
1785 while (!list_empty(&queue->poll_list)) {
1786 struct net_device *dev;
1787
1788 if (budget <= 0 || jiffies - start_time > 1)
1789 goto softnet_break;
1790
1791 local_irq_enable();
1792
1793 dev = list_entry(queue->poll_list.next,
1794 struct net_device, poll_list);
1795 netpoll_poll_lock(dev);
1796
1797 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1798 netpoll_poll_unlock(dev);
1799 local_irq_disable();
1800 list_del(&dev->poll_list);
1801 list_add_tail(&dev->poll_list, &queue->poll_list);
1802 if (dev->quota < 0)
1803 dev->quota += dev->weight;
1804 else
1805 dev->quota = dev->weight;
1806 } else {
1807 netpoll_poll_unlock(dev);
1808 dev_put(dev);
1809 local_irq_disable();
1810 }
1811 }
1812out:
1813 local_irq_enable();
1814 return;
1815
1816softnet_break:
1817 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1818 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1819 goto out;
1820}
1821
1822static gifconf_func_t * gifconf_list [NPROTO];
1823
1824/**
1825 * register_gifconf - register a SIOCGIF handler
1826 * @family: Address family
1827 * @gifconf: Function handler
1828 *
1829 * Register protocol dependent address dumping routines. The handler
1830 * that is passed must not be freed or reused until it has been replaced
1831 * by another handler.
1832 */
1833int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1834{
1835 if (family >= NPROTO)
1836 return -EINVAL;
1837 gifconf_list[family] = gifconf;
1838 return 0;
1839}
1840
1841
1842/*
1843 * Map an interface index to its name (SIOCGIFNAME)
1844 */
1845
1846/*
1847 * We need this ioctl for efficient implementation of the
1848 * if_indextoname() function required by the IPv6 API. Without
1849 * it, we would have to search all the interfaces to find a
1850 * match. --pb
1851 */
1852
1853static int dev_ifname(struct ifreq __user *arg)
1854{
1855 struct net_device *dev;
1856 struct ifreq ifr;
1857
1858 /*
1859 * Fetch the caller's info block.
1860 */
1861
1862 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1863 return -EFAULT;
1864
1865 read_lock(&dev_base_lock);
1866 dev = __dev_get_by_index(ifr.ifr_ifindex);
1867 if (!dev) {
1868 read_unlock(&dev_base_lock);
1869 return -ENODEV;
1870 }
1871
1872 strcpy(ifr.ifr_name, dev->name);
1873 read_unlock(&dev_base_lock);
1874
1875 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1876 return -EFAULT;
1877 return 0;
1878}
1879
1880/*
1881 * Perform a SIOCGIFCONF call. This structure will change
1882 * size eventually, and there is nothing I can do about it.
1883 * Thus we will need a 'compatibility mode'.
1884 */
1885
1886static int dev_ifconf(char __user *arg)
1887{
1888 struct ifconf ifc;
1889 struct net_device *dev;
1890 char __user *pos;
1891 int len;
1892 int total;
1893 int i;
1894
1895 /*
1896 * Fetch the caller's info block.
1897 */
1898
1899 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1900 return -EFAULT;
1901
1902 pos = ifc.ifc_buf;
1903 len = ifc.ifc_len;
1904
1905 /*
1906 * Loop over the interfaces, and write an info block for each.
1907 */
1908
1909 total = 0;
1910 for (dev = dev_base; dev; dev = dev->next) {
1911 for (i = 0; i < NPROTO; i++) {
1912 if (gifconf_list[i]) {
1913 int done;
1914 if (!pos)
1915 done = gifconf_list[i](dev, NULL, 0);
1916 else
1917 done = gifconf_list[i](dev, pos + total,
1918 len - total);
1919 if (done < 0)
1920 return -EFAULT;
1921 total += done;
1922 }
1923 }
1924 }
1925
1926 /*
1927 * All done. Write the updated control block back to the caller.
1928 */
1929 ifc.ifc_len = total;
1930
1931 /*
1932 * Both BSD and Solaris return 0 here, so we do too.
1933 */
1934 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1935}
1936
1937#ifdef CONFIG_PROC_FS
1938/*
1939 * This is invoked by the /proc filesystem handler to display a device
1940 * in detail.
1941 */
1942static __inline__ struct net_device *dev_get_idx(loff_t pos)
1943{
1944 struct net_device *dev;
1945 loff_t i;
1946
1947 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1948
1949 return i == pos ? dev : NULL;
1950}
1951
1952void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1953{
1954 read_lock(&dev_base_lock);
1955 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1956}
1957
1958void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1959{
1960 ++*pos;
1961 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1962}
1963
1964void dev_seq_stop(struct seq_file *seq, void *v)
1965{
1966 read_unlock(&dev_base_lock);
1967}
1968
1969static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1970{
1971 if (dev->get_stats) {
1972 struct net_device_stats *stats = dev->get_stats(dev);
1973
1974 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1975 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1976 dev->name, stats->rx_bytes, stats->rx_packets,
1977 stats->rx_errors,
1978 stats->rx_dropped + stats->rx_missed_errors,
1979 stats->rx_fifo_errors,
1980 stats->rx_length_errors + stats->rx_over_errors +
1981 stats->rx_crc_errors + stats->rx_frame_errors,
1982 stats->rx_compressed, stats->multicast,
1983 stats->tx_bytes, stats->tx_packets,
1984 stats->tx_errors, stats->tx_dropped,
1985 stats->tx_fifo_errors, stats->collisions,
1986 stats->tx_carrier_errors +
1987 stats->tx_aborted_errors +
1988 stats->tx_window_errors +
1989 stats->tx_heartbeat_errors,
1990 stats->tx_compressed);
1991 } else
1992 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1993}
1994
1995/*
1996 * Called from the PROCfs module. This now uses the new arbitrary sized
1997 * /proc/net interface to create /proc/net/dev
1998 */
1999static int dev_seq_show(struct seq_file *seq, void *v)
2000{
2001 if (v == SEQ_START_TOKEN)
2002 seq_puts(seq, "Inter-| Receive "
2003 " | Transmit\n"
2004 " face |bytes packets errs drop fifo frame "
2005 "compressed multicast|bytes packets errs "
2006 "drop fifo colls carrier compressed\n");
2007 else
2008 dev_seq_printf_stats(seq, v);
2009 return 0;
2010}
2011
2012static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2013{
2014 struct netif_rx_stats *rc = NULL;
2015
2016 while (*pos < NR_CPUS)
2017 if (cpu_online(*pos)) {
2018 rc = &per_cpu(netdev_rx_stat, *pos);
2019 break;
2020 } else
2021 ++*pos;
2022 return rc;
2023}
2024
2025static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2026{
2027 return softnet_get_online(pos);
2028}
2029
2030static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2031{
2032 ++*pos;
2033 return softnet_get_online(pos);
2034}
2035
2036static void softnet_seq_stop(struct seq_file *seq, void *v)
2037{
2038}
2039
2040static int softnet_seq_show(struct seq_file *seq, void *v)
2041{
2042 struct netif_rx_stats *s = v;
2043
2044 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2045 s->total, s->dropped, s->time_squeeze, s->throttled,
2046 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2047 s->fastroute_deferred_out,
2048#if 0
2049 s->fastroute_latency_reduction
2050#else
2051 s->cpu_collision
2052#endif
2053 );
2054 return 0;
2055}
2056
2057static struct seq_operations dev_seq_ops = {
2058 .start = dev_seq_start,
2059 .next = dev_seq_next,
2060 .stop = dev_seq_stop,
2061 .show = dev_seq_show,
2062};
2063
2064static int dev_seq_open(struct inode *inode, struct file *file)
2065{
2066 return seq_open(file, &dev_seq_ops);
2067}
2068
2069static struct file_operations dev_seq_fops = {
2070 .owner = THIS_MODULE,
2071 .open = dev_seq_open,
2072 .read = seq_read,
2073 .llseek = seq_lseek,
2074 .release = seq_release,
2075};
2076
2077static struct seq_operations softnet_seq_ops = {
2078 .start = softnet_seq_start,
2079 .next = softnet_seq_next,
2080 .stop = softnet_seq_stop,
2081 .show = softnet_seq_show,
2082};
2083
2084static int softnet_seq_open(struct inode *inode, struct file *file)
2085{
2086 return seq_open(file, &softnet_seq_ops);
2087}
2088
2089static struct file_operations softnet_seq_fops = {
2090 .owner = THIS_MODULE,
2091 .open = softnet_seq_open,
2092 .read = seq_read,
2093 .llseek = seq_lseek,
2094 .release = seq_release,
2095};
2096
2097#ifdef WIRELESS_EXT
2098extern int wireless_proc_init(void);
2099#else
2100#define wireless_proc_init() 0
2101#endif
2102
2103static int __init dev_proc_init(void)
2104{
2105 int rc = -ENOMEM;
2106
2107 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2108 goto out;
2109 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2110 goto out_dev;
2111 if (wireless_proc_init())
2112 goto out_softnet;
2113 rc = 0;
2114out:
2115 return rc;
2116out_softnet:
2117 proc_net_remove("softnet_stat");
2118out_dev:
2119 proc_net_remove("dev");
2120 goto out;
2121}
2122#else
2123#define dev_proc_init() 0
2124#endif /* CONFIG_PROC_FS */
2125
2126
2127/**
2128 * netdev_set_master - set up master/slave pair
2129 * @slave: slave device
2130 * @master: new master device
2131 *
2132 * Changes the master device of the slave. Pass %NULL to break the
2133 * bonding. The caller must hold the RTNL semaphore. On a failure
2134 * a negative errno code is returned. On success the reference counts
2135 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2136 * function returns zero.
2137 */
2138int netdev_set_master(struct net_device *slave, struct net_device *master)
2139{
2140 struct net_device *old = slave->master;
2141
2142 ASSERT_RTNL();
2143
2144 if (master) {
2145 if (old)
2146 return -EBUSY;
2147 dev_hold(master);
2148 }
2149
2150 slave->master = master;
2151
2152 synchronize_net();
2153
2154 if (old)
2155 dev_put(old);
2156
2157 if (master)
2158 slave->flags |= IFF_SLAVE;
2159 else
2160 slave->flags &= ~IFF_SLAVE;
2161
2162 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2163 return 0;
2164}
2165
2166/**
2167 * dev_set_promiscuity - update promiscuity count on a device
2168 * @dev: device
2169 * @inc: modifier
2170 *
2171 * Add or remove promsicuity from a device. While the count in the device
2172 * remains above zero the interface remains promiscuous. Once it hits zero
2173 * the device reverts back to normal filtering operation. A negative inc
2174 * value is used to drop promiscuity on the device.
2175 */
2176void dev_set_promiscuity(struct net_device *dev, int inc)
2177{
2178 unsigned short old_flags = dev->flags;
2179
2180 dev->flags |= IFF_PROMISC;
2181 if ((dev->promiscuity += inc) == 0)
2182 dev->flags &= ~IFF_PROMISC;
2183 if (dev->flags ^ old_flags) {
2184 dev_mc_upload(dev);
2185 printk(KERN_INFO "device %s %s promiscuous mode\n",
2186 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2187 "left");
2188 }
2189}
2190
2191/**
2192 * dev_set_allmulti - update allmulti count on a device
2193 * @dev: device
2194 * @inc: modifier
2195 *
2196 * Add or remove reception of all multicast frames to a device. While the
2197 * count in the device remains above zero the interface remains listening
2198 * to all interfaces. Once it hits zero the device reverts back to normal
2199 * filtering operation. A negative @inc value is used to drop the counter
2200 * when releasing a resource needing all multicasts.
2201 */
2202
2203void dev_set_allmulti(struct net_device *dev, int inc)
2204{
2205 unsigned short old_flags = dev->flags;
2206
2207 dev->flags |= IFF_ALLMULTI;
2208 if ((dev->allmulti += inc) == 0)
2209 dev->flags &= ~IFF_ALLMULTI;
2210 if (dev->flags ^ old_flags)
2211 dev_mc_upload(dev);
2212}
2213
2214unsigned dev_get_flags(const struct net_device *dev)
2215{
2216 unsigned flags;
2217
2218 flags = (dev->flags & ~(IFF_PROMISC |
2219 IFF_ALLMULTI |
2220 IFF_RUNNING)) |
2221 (dev->gflags & (IFF_PROMISC |
2222 IFF_ALLMULTI));
2223
2224 if (netif_running(dev) && netif_carrier_ok(dev))
2225 flags |= IFF_RUNNING;
2226
2227 return flags;
2228}
2229
2230int dev_change_flags(struct net_device *dev, unsigned flags)
2231{
2232 int ret;
2233 int old_flags = dev->flags;
2234
2235 /*
2236 * Set the flags on our device.
2237 */
2238
2239 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2240 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2241 IFF_AUTOMEDIA)) |
2242 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2243 IFF_ALLMULTI));
2244
2245 /*
2246 * Load in the correct multicast list now the flags have changed.
2247 */
2248
2249 dev_mc_upload(dev);
2250
2251 /*
2252 * Have we downed the interface. We handle IFF_UP ourselves
2253 * according to user attempts to set it, rather than blindly
2254 * setting it.
2255 */
2256
2257 ret = 0;
2258 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2259 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2260
2261 if (!ret)
2262 dev_mc_upload(dev);
2263 }
2264
2265 if (dev->flags & IFF_UP &&
2266 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2267 IFF_VOLATILE)))
2268 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2269
2270 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2271 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2272 dev->gflags ^= IFF_PROMISC;
2273 dev_set_promiscuity(dev, inc);
2274 }
2275
2276 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2277 is important. Some (broken) drivers set IFF_PROMISC, when
2278 IFF_ALLMULTI is requested not asking us and not reporting.
2279 */
2280 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2281 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2282 dev->gflags ^= IFF_ALLMULTI;
2283 dev_set_allmulti(dev, inc);
2284 }
2285
2286 if (old_flags ^ dev->flags)
2287 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2288
2289 return ret;
2290}
2291
2292int dev_set_mtu(struct net_device *dev, int new_mtu)
2293{
2294 int err;
2295
2296 if (new_mtu == dev->mtu)
2297 return 0;
2298
2299 /* MTU must be positive. */
2300 if (new_mtu < 0)
2301 return -EINVAL;
2302
2303 if (!netif_device_present(dev))
2304 return -ENODEV;
2305
2306 err = 0;
2307 if (dev->change_mtu)
2308 err = dev->change_mtu(dev, new_mtu);
2309 else
2310 dev->mtu = new_mtu;
2311 if (!err && dev->flags & IFF_UP)
2312 notifier_call_chain(&netdev_chain,
2313 NETDEV_CHANGEMTU, dev);
2314 return err;
2315}
2316
2317int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2318{
2319 int err;
2320
2321 if (!dev->set_mac_address)
2322 return -EOPNOTSUPP;
2323 if (sa->sa_family != dev->type)
2324 return -EINVAL;
2325 if (!netif_device_present(dev))
2326 return -ENODEV;
2327 err = dev->set_mac_address(dev, sa);
2328 if (!err)
2329 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2330 return err;
2331}
2332
2333/*
2334 * Perform the SIOCxIFxxx calls.
2335 */
2336static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2337{
2338 int err;
2339 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2340
2341 if (!dev)
2342 return -ENODEV;
2343
2344 switch (cmd) {
2345 case SIOCGIFFLAGS: /* Get interface flags */
2346 ifr->ifr_flags = dev_get_flags(dev);
2347 return 0;
2348
2349 case SIOCSIFFLAGS: /* Set interface flags */
2350 return dev_change_flags(dev, ifr->ifr_flags);
2351
2352 case SIOCGIFMETRIC: /* Get the metric on the interface
2353 (currently unused) */
2354 ifr->ifr_metric = 0;
2355 return 0;
2356
2357 case SIOCSIFMETRIC: /* Set the metric on the interface
2358 (currently unused) */
2359 return -EOPNOTSUPP;
2360
2361 case SIOCGIFMTU: /* Get the MTU of a device */
2362 ifr->ifr_mtu = dev->mtu;
2363 return 0;
2364
2365 case SIOCSIFMTU: /* Set the MTU of a device */
2366 return dev_set_mtu(dev, ifr->ifr_mtu);
2367
2368 case SIOCGIFHWADDR:
2369 if (!dev->addr_len)
2370 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2371 else
2372 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2373 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2374 ifr->ifr_hwaddr.sa_family = dev->type;
2375 return 0;
2376
2377 case SIOCSIFHWADDR:
2378 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2379
2380 case SIOCSIFHWBROADCAST:
2381 if (ifr->ifr_hwaddr.sa_family != dev->type)
2382 return -EINVAL;
2383 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2384 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2385 notifier_call_chain(&netdev_chain,
2386 NETDEV_CHANGEADDR, dev);
2387 return 0;
2388
2389 case SIOCGIFMAP:
2390 ifr->ifr_map.mem_start = dev->mem_start;
2391 ifr->ifr_map.mem_end = dev->mem_end;
2392 ifr->ifr_map.base_addr = dev->base_addr;
2393 ifr->ifr_map.irq = dev->irq;
2394 ifr->ifr_map.dma = dev->dma;
2395 ifr->ifr_map.port = dev->if_port;
2396 return 0;
2397
2398 case SIOCSIFMAP:
2399 if (dev->set_config) {
2400 if (!netif_device_present(dev))
2401 return -ENODEV;
2402 return dev->set_config(dev, &ifr->ifr_map);
2403 }
2404 return -EOPNOTSUPP;
2405
2406 case SIOCADDMULTI:
2407 if (!dev->set_multicast_list ||
2408 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2409 return -EINVAL;
2410 if (!netif_device_present(dev))
2411 return -ENODEV;
2412 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2413 dev->addr_len, 1);
2414
2415 case SIOCDELMULTI:
2416 if (!dev->set_multicast_list ||
2417 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2418 return -EINVAL;
2419 if (!netif_device_present(dev))
2420 return -ENODEV;
2421 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2422 dev->addr_len, 1);
2423
2424 case SIOCGIFINDEX:
2425 ifr->ifr_ifindex = dev->ifindex;
2426 return 0;
2427
2428 case SIOCGIFTXQLEN:
2429 ifr->ifr_qlen = dev->tx_queue_len;
2430 return 0;
2431
2432 case SIOCSIFTXQLEN:
2433 if (ifr->ifr_qlen < 0)
2434 return -EINVAL;
2435 dev->tx_queue_len = ifr->ifr_qlen;
2436 return 0;
2437
2438 case SIOCSIFNAME:
2439 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2440 return dev_change_name(dev, ifr->ifr_newname);
2441
2442 /*
2443 * Unknown or private ioctl
2444 */
2445
2446 default:
2447 if ((cmd >= SIOCDEVPRIVATE &&
2448 cmd <= SIOCDEVPRIVATE + 15) ||
2449 cmd == SIOCBONDENSLAVE ||
2450 cmd == SIOCBONDRELEASE ||
2451 cmd == SIOCBONDSETHWADDR ||
2452 cmd == SIOCBONDSLAVEINFOQUERY ||
2453 cmd == SIOCBONDINFOQUERY ||
2454 cmd == SIOCBONDCHANGEACTIVE ||
2455 cmd == SIOCGMIIPHY ||
2456 cmd == SIOCGMIIREG ||
2457 cmd == SIOCSMIIREG ||
2458 cmd == SIOCBRADDIF ||
2459 cmd == SIOCBRDELIF ||
2460 cmd == SIOCWANDEV) {
2461 err = -EOPNOTSUPP;
2462 if (dev->do_ioctl) {
2463 if (netif_device_present(dev))
2464 err = dev->do_ioctl(dev, ifr,
2465 cmd);
2466 else
2467 err = -ENODEV;
2468 }
2469 } else
2470 err = -EINVAL;
2471
2472 }
2473 return err;
2474}
2475
2476/*
2477 * This function handles all "interface"-type I/O control requests. The actual
2478 * 'doing' part of this is dev_ifsioc above.
2479 */
2480
2481/**
2482 * dev_ioctl - network device ioctl
2483 * @cmd: command to issue
2484 * @arg: pointer to a struct ifreq in user space
2485 *
2486 * Issue ioctl functions to devices. This is normally called by the
2487 * user space syscall interfaces but can sometimes be useful for
2488 * other purposes. The return value is the return from the syscall if
2489 * positive or a negative errno code on error.
2490 */
2491
2492int dev_ioctl(unsigned int cmd, void __user *arg)
2493{
2494 struct ifreq ifr;
2495 int ret;
2496 char *colon;
2497
2498 /* One special case: SIOCGIFCONF takes ifconf argument
2499 and requires shared lock, because it sleeps writing
2500 to user space.
2501 */
2502
2503 if (cmd == SIOCGIFCONF) {
2504 rtnl_shlock();
2505 ret = dev_ifconf((char __user *) arg);
2506 rtnl_shunlock();
2507 return ret;
2508 }
2509 if (cmd == SIOCGIFNAME)
2510 return dev_ifname((struct ifreq __user *)arg);
2511
2512 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2513 return -EFAULT;
2514
2515 ifr.ifr_name[IFNAMSIZ-1] = 0;
2516
2517 colon = strchr(ifr.ifr_name, ':');
2518 if (colon)
2519 *colon = 0;
2520
2521 /*
2522 * See which interface the caller is talking about.
2523 */
2524
2525 switch (cmd) {
2526 /*
2527 * These ioctl calls:
2528 * - can be done by all.
2529 * - atomic and do not require locking.
2530 * - return a value
2531 */
2532 case SIOCGIFFLAGS:
2533 case SIOCGIFMETRIC:
2534 case SIOCGIFMTU:
2535 case SIOCGIFHWADDR:
2536 case SIOCGIFSLAVE:
2537 case SIOCGIFMAP:
2538 case SIOCGIFINDEX:
2539 case SIOCGIFTXQLEN:
2540 dev_load(ifr.ifr_name);
2541 read_lock(&dev_base_lock);
2542 ret = dev_ifsioc(&ifr, cmd);
2543 read_unlock(&dev_base_lock);
2544 if (!ret) {
2545 if (colon)
2546 *colon = ':';
2547 if (copy_to_user(arg, &ifr,
2548 sizeof(struct ifreq)))
2549 ret = -EFAULT;
2550 }
2551 return ret;
2552
2553 case SIOCETHTOOL:
2554 dev_load(ifr.ifr_name);
2555 rtnl_lock();
2556 ret = dev_ethtool(&ifr);
2557 rtnl_unlock();
2558 if (!ret) {
2559 if (colon)
2560 *colon = ':';
2561 if (copy_to_user(arg, &ifr,
2562 sizeof(struct ifreq)))
2563 ret = -EFAULT;
2564 }
2565 return ret;
2566
2567 /*
2568 * These ioctl calls:
2569 * - require superuser power.
2570 * - require strict serialization.
2571 * - return a value
2572 */
2573 case SIOCGMIIPHY:
2574 case SIOCGMIIREG:
2575 case SIOCSIFNAME:
2576 if (!capable(CAP_NET_ADMIN))
2577 return -EPERM;
2578 dev_load(ifr.ifr_name);
2579 rtnl_lock();
2580 ret = dev_ifsioc(&ifr, cmd);
2581 rtnl_unlock();
2582 if (!ret) {
2583 if (colon)
2584 *colon = ':';
2585 if (copy_to_user(arg, &ifr,
2586 sizeof(struct ifreq)))
2587 ret = -EFAULT;
2588 }
2589 return ret;
2590
2591 /*
2592 * These ioctl calls:
2593 * - require superuser power.
2594 * - require strict serialization.
2595 * - do not return a value
2596 */
2597 case SIOCSIFFLAGS:
2598 case SIOCSIFMETRIC:
2599 case SIOCSIFMTU:
2600 case SIOCSIFMAP:
2601 case SIOCSIFHWADDR:
2602 case SIOCSIFSLAVE:
2603 case SIOCADDMULTI:
2604 case SIOCDELMULTI:
2605 case SIOCSIFHWBROADCAST:
2606 case SIOCSIFTXQLEN:
2607 case SIOCSMIIREG:
2608 case SIOCBONDENSLAVE:
2609 case SIOCBONDRELEASE:
2610 case SIOCBONDSETHWADDR:
2611 case SIOCBONDSLAVEINFOQUERY:
2612 case SIOCBONDINFOQUERY:
2613 case SIOCBONDCHANGEACTIVE:
2614 case SIOCBRADDIF:
2615 case SIOCBRDELIF:
2616 if (!capable(CAP_NET_ADMIN))
2617 return -EPERM;
2618 dev_load(ifr.ifr_name);
2619 rtnl_lock();
2620 ret = dev_ifsioc(&ifr, cmd);
2621 rtnl_unlock();
2622 return ret;
2623
2624 case SIOCGIFMEM:
2625 /* Get the per device memory space. We can add this but
2626 * currently do not support it */
2627 case SIOCSIFMEM:
2628 /* Set the per device memory buffer space.
2629 * Not applicable in our case */
2630 case SIOCSIFLINK:
2631 return -EINVAL;
2632
2633 /*
2634 * Unknown or private ioctl.
2635 */
2636 default:
2637 if (cmd == SIOCWANDEV ||
2638 (cmd >= SIOCDEVPRIVATE &&
2639 cmd <= SIOCDEVPRIVATE + 15)) {
2640 dev_load(ifr.ifr_name);
2641 rtnl_lock();
2642 ret = dev_ifsioc(&ifr, cmd);
2643 rtnl_unlock();
2644 if (!ret && copy_to_user(arg, &ifr,
2645 sizeof(struct ifreq)))
2646 ret = -EFAULT;
2647 return ret;
2648 }
2649#ifdef WIRELESS_EXT
2650 /* Take care of Wireless Extensions */
2651 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2652 /* If command is `set a parameter', or
2653 * `get the encoding parameters', check if
2654 * the user has the right to do it */
2655 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2656 if (!capable(CAP_NET_ADMIN))
2657 return -EPERM;
2658 }
2659 dev_load(ifr.ifr_name);
2660 rtnl_lock();
2661 /* Follow me in net/core/wireless.c */
2662 ret = wireless_process_ioctl(&ifr, cmd);
2663 rtnl_unlock();
2664 if (IW_IS_GET(cmd) &&
2665 copy_to_user(arg, &ifr,
2666 sizeof(struct ifreq)))
2667 ret = -EFAULT;
2668 return ret;
2669 }
2670#endif /* WIRELESS_EXT */
2671 return -EINVAL;
2672 }
2673}
2674
2675
2676/**
2677 * dev_new_index - allocate an ifindex
2678 *
2679 * Returns a suitable unique value for a new device interface
2680 * number. The caller must hold the rtnl semaphore or the
2681 * dev_base_lock to be sure it remains unique.
2682 */
2683static int dev_new_index(void)
2684{
2685 static int ifindex;
2686 for (;;) {
2687 if (++ifindex <= 0)
2688 ifindex = 1;
2689 if (!__dev_get_by_index(ifindex))
2690 return ifindex;
2691 }
2692}
2693
2694static int dev_boot_phase = 1;
2695
2696/* Delayed registration/unregisteration */
2697static DEFINE_SPINLOCK(net_todo_list_lock);
2698static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2699
2700static inline void net_set_todo(struct net_device *dev)
2701{
2702 spin_lock(&net_todo_list_lock);
2703 list_add_tail(&dev->todo_list, &net_todo_list);
2704 spin_unlock(&net_todo_list_lock);
2705}
2706
2707/**
2708 * register_netdevice - register a network device
2709 * @dev: device to register
2710 *
2711 * Take a completed network device structure and add it to the kernel
2712 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2713 * chain. 0 is returned on success. A negative errno code is returned
2714 * on a failure to set up the device, or if the name is a duplicate.
2715 *
2716 * Callers must hold the rtnl semaphore. You may want
2717 * register_netdev() instead of this.
2718 *
2719 * BUGS:
2720 * The locking appears insufficient to guarantee two parallel registers
2721 * will not get the same name.
2722 */
2723
2724int register_netdevice(struct net_device *dev)
2725{
2726 struct hlist_head *head;
2727 struct hlist_node *p;
2728 int ret;
2729
2730 BUG_ON(dev_boot_phase);
2731 ASSERT_RTNL();
2732
2733 /* When net_device's are persistent, this will be fatal. */
2734 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2735
2736 spin_lock_init(&dev->queue_lock);
2737 spin_lock_init(&dev->xmit_lock);
2738 dev->xmit_lock_owner = -1;
2739#ifdef CONFIG_NET_CLS_ACT
2740 spin_lock_init(&dev->ingress_lock);
2741#endif
2742
2743 ret = alloc_divert_blk(dev);
2744 if (ret)
2745 goto out;
2746
2747 dev->iflink = -1;
2748
2749 /* Init, if this function is available */
2750 if (dev->init) {
2751 ret = dev->init(dev);
2752 if (ret) {
2753 if (ret > 0)
2754 ret = -EIO;
2755 goto out_err;
2756 }
2757 }
2758
2759 if (!dev_valid_name(dev->name)) {
2760 ret = -EINVAL;
2761 goto out_err;
2762 }
2763
2764 dev->ifindex = dev_new_index();
2765 if (dev->iflink == -1)
2766 dev->iflink = dev->ifindex;
2767
2768 /* Check for existence of name */
2769 head = dev_name_hash(dev->name);
2770 hlist_for_each(p, head) {
2771 struct net_device *d
2772 = hlist_entry(p, struct net_device, name_hlist);
2773 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2774 ret = -EEXIST;
2775 goto out_err;
2776 }
2777 }
2778
2779 /* Fix illegal SG+CSUM combinations. */
2780 if ((dev->features & NETIF_F_SG) &&
2781 !(dev->features & (NETIF_F_IP_CSUM |
2782 NETIF_F_NO_CSUM |
2783 NETIF_F_HW_CSUM))) {
2784 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2785 dev->name);
2786 dev->features &= ~NETIF_F_SG;
2787 }
2788
2789 /* TSO requires that SG is present as well. */
2790 if ((dev->features & NETIF_F_TSO) &&
2791 !(dev->features & NETIF_F_SG)) {
2792 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2793 dev->name);
2794 dev->features &= ~NETIF_F_TSO;
2795 }
2796
2797 /*
2798 * nil rebuild_header routine,
2799 * that should be never called and used as just bug trap.
2800 */
2801
2802 if (!dev->rebuild_header)
2803 dev->rebuild_header = default_rebuild_header;
2804
2805 /*
2806 * Default initial state at registry is that the
2807 * device is present.
2808 */
2809
2810 set_bit(__LINK_STATE_PRESENT, &dev->state);
2811
2812 dev->next = NULL;
2813 dev_init_scheduler(dev);
2814 write_lock_bh(&dev_base_lock);
2815 *dev_tail = dev;
2816 dev_tail = &dev->next;
2817 hlist_add_head(&dev->name_hlist, head);
2818 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2819 dev_hold(dev);
2820 dev->reg_state = NETREG_REGISTERING;
2821 write_unlock_bh(&dev_base_lock);
2822
2823 /* Notify protocols, that a new device appeared. */
2824 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2825
2826 /* Finish registration after unlock */
2827 net_set_todo(dev);
2828 ret = 0;
2829
2830out:
2831 return ret;
2832out_err:
2833 free_divert_blk(dev);
2834 goto out;
2835}
2836
2837/**
2838 * register_netdev - register a network device
2839 * @dev: device to register
2840 *
2841 * Take a completed network device structure and add it to the kernel
2842 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2843 * chain. 0 is returned on success. A negative errno code is returned
2844 * on a failure to set up the device, or if the name is a duplicate.
2845 *
2846 * This is a wrapper around register_netdev that takes the rtnl semaphore
2847 * and expands the device name if you passed a format string to
2848 * alloc_netdev.
2849 */
2850int register_netdev(struct net_device *dev)
2851{
2852 int err;
2853
2854 rtnl_lock();
2855
2856 /*
2857 * If the name is a format string the caller wants us to do a
2858 * name allocation.
2859 */
2860 if (strchr(dev->name, '%')) {
2861 err = dev_alloc_name(dev, dev->name);
2862 if (err < 0)
2863 goto out;
2864 }
2865
2866 /*
2867 * Back compatibility hook. Kill this one in 2.5
2868 */
2869 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2870 err = dev_alloc_name(dev, "eth%d");
2871 if (err < 0)
2872 goto out;
2873 }
2874
2875 err = register_netdevice(dev);
2876out:
2877 rtnl_unlock();
2878 return err;
2879}
2880EXPORT_SYMBOL(register_netdev);
2881
2882/*
2883 * netdev_wait_allrefs - wait until all references are gone.
2884 *
2885 * This is called when unregistering network devices.
2886 *
2887 * Any protocol or device that holds a reference should register
2888 * for netdevice notification, and cleanup and put back the
2889 * reference if they receive an UNREGISTER event.
2890 * We can get stuck here if buggy protocols don't correctly
2891 * call dev_put.
2892 */
2893static void netdev_wait_allrefs(struct net_device *dev)
2894{
2895 unsigned long rebroadcast_time, warning_time;
2896
2897 rebroadcast_time = warning_time = jiffies;
2898 while (atomic_read(&dev->refcnt) != 0) {
2899 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2900 rtnl_shlock();
2901
2902 /* Rebroadcast unregister notification */
2903 notifier_call_chain(&netdev_chain,
2904 NETDEV_UNREGISTER, dev);
2905
2906 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2907 &dev->state)) {
2908 /* We must not have linkwatch events
2909 * pending on unregister. If this
2910 * happens, we simply run the queue
2911 * unscheduled, resulting in a noop
2912 * for this device.
2913 */
2914 linkwatch_run_queue();
2915 }
2916
2917 rtnl_shunlock();
2918
2919 rebroadcast_time = jiffies;
2920 }
2921
2922 msleep(250);
2923
2924 if (time_after(jiffies, warning_time + 10 * HZ)) {
2925 printk(KERN_EMERG "unregister_netdevice: "
2926 "waiting for %s to become free. Usage "
2927 "count = %d\n",
2928 dev->name, atomic_read(&dev->refcnt));
2929 warning_time = jiffies;
2930 }
2931 }
2932}
2933
2934/* The sequence is:
2935 *
2936 * rtnl_lock();
2937 * ...
2938 * register_netdevice(x1);
2939 * register_netdevice(x2);
2940 * ...
2941 * unregister_netdevice(y1);
2942 * unregister_netdevice(y2);
2943 * ...
2944 * rtnl_unlock();
2945 * free_netdev(y1);
2946 * free_netdev(y2);
2947 *
2948 * We are invoked by rtnl_unlock() after it drops the semaphore.
2949 * This allows us to deal with problems:
2950 * 1) We can create/delete sysfs objects which invoke hotplug
2951 * without deadlocking with linkwatch via keventd.
2952 * 2) Since we run with the RTNL semaphore not held, we can sleep
2953 * safely in order to wait for the netdev refcnt to drop to zero.
2954 */
2955static DECLARE_MUTEX(net_todo_run_mutex);
2956void netdev_run_todo(void)
2957{
2958 struct list_head list = LIST_HEAD_INIT(list);
2959 int err;
2960
2961
2962 /* Need to guard against multiple cpu's getting out of order. */
2963 down(&net_todo_run_mutex);
2964
2965 /* Not safe to do outside the semaphore. We must not return
2966 * until all unregister events invoked by the local processor
2967 * have been completed (either by this todo run, or one on
2968 * another cpu).
2969 */
2970 if (list_empty(&net_todo_list))
2971 goto out;
2972
2973 /* Snapshot list, allow later requests */
2974 spin_lock(&net_todo_list_lock);
2975 list_splice_init(&net_todo_list, &list);
2976 spin_unlock(&net_todo_list_lock);
2977
2978 while (!list_empty(&list)) {
2979 struct net_device *dev
2980 = list_entry(list.next, struct net_device, todo_list);
2981 list_del(&dev->todo_list);
2982
2983 switch(dev->reg_state) {
2984 case NETREG_REGISTERING:
2985 err = netdev_register_sysfs(dev);
2986 if (err)
2987 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2988 dev->name, err);
2989 dev->reg_state = NETREG_REGISTERED;
2990 break;
2991
2992 case NETREG_UNREGISTERING:
2993 netdev_unregister_sysfs(dev);
2994 dev->reg_state = NETREG_UNREGISTERED;
2995
2996 netdev_wait_allrefs(dev);
2997
2998 /* paranoia */
2999 BUG_ON(atomic_read(&dev->refcnt));
3000 BUG_TRAP(!dev->ip_ptr);
3001 BUG_TRAP(!dev->ip6_ptr);
3002 BUG_TRAP(!dev->dn_ptr);
3003
3004
3005 /* It must be the very last action,
3006 * after this 'dev' may point to freed up memory.
3007 */
3008 if (dev->destructor)
3009 dev->destructor(dev);
3010 break;
3011
3012 default:
3013 printk(KERN_ERR "network todo '%s' but state %d\n",
3014 dev->name, dev->reg_state);
3015 break;
3016 }
3017 }
3018
3019out:
3020 up(&net_todo_run_mutex);
3021}
3022
3023/**
3024 * alloc_netdev - allocate network device
3025 * @sizeof_priv: size of private data to allocate space for
3026 * @name: device name format string
3027 * @setup: callback to initialize device
3028 *
3029 * Allocates a struct net_device with private data area for driver use
3030 * and performs basic initialization.
3031 */
3032struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3033 void (*setup)(struct net_device *))
3034{
3035 void *p;
3036 struct net_device *dev;
3037 int alloc_size;
3038
3039 /* ensure 32-byte alignment of both the device and private area */
3040 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3041 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3042
3043 p = kmalloc(alloc_size, GFP_KERNEL);
3044 if (!p) {
3045 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3046 return NULL;
3047 }
3048 memset(p, 0, alloc_size);
3049
3050 dev = (struct net_device *)
3051 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3052 dev->padded = (char *)dev - (char *)p;
3053
3054 if (sizeof_priv)
3055 dev->priv = netdev_priv(dev);
3056
3057 setup(dev);
3058 strcpy(dev->name, name);
3059 return dev;
3060}
3061EXPORT_SYMBOL(alloc_netdev);
3062
3063/**
3064 * free_netdev - free network device
3065 * @dev: device
3066 *
3067 * This function does the last stage of destroying an allocated device
3068 * interface. The reference to the device object is released.
3069 * If this is the last reference then it will be freed.
3070 */
3071void free_netdev(struct net_device *dev)
3072{
3073#ifdef CONFIG_SYSFS
3074 /* Compatiablity with error handling in drivers */
3075 if (dev->reg_state == NETREG_UNINITIALIZED) {
3076 kfree((char *)dev - dev->padded);
3077 return;
3078 }
3079
3080 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3081 dev->reg_state = NETREG_RELEASED;
3082
3083 /* will free via class release */
3084 class_device_put(&dev->class_dev);
3085#else
3086 kfree((char *)dev - dev->padded);
3087#endif
3088}
3089
3090/* Synchronize with packet receive processing. */
3091void synchronize_net(void)
3092{
3093 might_sleep();
fbd568a3 3094 synchronize_rcu();
1da177e4
LT
3095}
3096
3097/**
3098 * unregister_netdevice - remove device from the kernel
3099 * @dev: device
3100 *
3101 * This function shuts down a device interface and removes it
3102 * from the kernel tables. On success 0 is returned, on a failure
3103 * a negative errno code is returned.
3104 *
3105 * Callers must hold the rtnl semaphore. You may want
3106 * unregister_netdev() instead of this.
3107 */
3108
3109int unregister_netdevice(struct net_device *dev)
3110{
3111 struct net_device *d, **dp;
3112
3113 BUG_ON(dev_boot_phase);
3114 ASSERT_RTNL();
3115
3116 /* Some devices call without registering for initialization unwind. */
3117 if (dev->reg_state == NETREG_UNINITIALIZED) {
3118 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3119 "was registered\n", dev->name, dev);
3120 return -ENODEV;
3121 }
3122
3123 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3124
3125 /* If device is running, close it first. */
3126 if (dev->flags & IFF_UP)
3127 dev_close(dev);
3128
3129 /* And unlink it from device chain. */
3130 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3131 if (d == dev) {
3132 write_lock_bh(&dev_base_lock);
3133 hlist_del(&dev->name_hlist);
3134 hlist_del(&dev->index_hlist);
3135 if (dev_tail == &dev->next)
3136 dev_tail = dp;
3137 *dp = d->next;
3138 write_unlock_bh(&dev_base_lock);
3139 break;
3140 }
3141 }
3142 if (!d) {
3143 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3144 dev->name);
3145 return -ENODEV;
3146 }
3147
3148 dev->reg_state = NETREG_UNREGISTERING;
3149
3150 synchronize_net();
3151
3152 /* Shutdown queueing discipline. */
3153 dev_shutdown(dev);
3154
3155
3156 /* Notify protocols, that we are about to destroy
3157 this device. They should clean all the things.
3158 */
3159 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3160
3161 /*
3162 * Flush the multicast chain
3163 */
3164 dev_mc_discard(dev);
3165
3166 if (dev->uninit)
3167 dev->uninit(dev);
3168
3169 /* Notifier chain MUST detach us from master device. */
3170 BUG_TRAP(!dev->master);
3171
3172 free_divert_blk(dev);
3173
3174 /* Finish processing unregister after unlock */
3175 net_set_todo(dev);
3176
3177 synchronize_net();
3178
3179 dev_put(dev);
3180 return 0;
3181}
3182
3183/**
3184 * unregister_netdev - remove device from the kernel
3185 * @dev: device
3186 *
3187 * This function shuts down a device interface and removes it
3188 * from the kernel tables. On success 0 is returned, on a failure
3189 * a negative errno code is returned.
3190 *
3191 * This is just a wrapper for unregister_netdevice that takes
3192 * the rtnl semaphore. In general you want to use this and not
3193 * unregister_netdevice.
3194 */
3195void unregister_netdev(struct net_device *dev)
3196{
3197 rtnl_lock();
3198 unregister_netdevice(dev);
3199 rtnl_unlock();
3200}
3201
3202EXPORT_SYMBOL(unregister_netdev);
3203
3204#ifdef CONFIG_HOTPLUG_CPU
3205static int dev_cpu_callback(struct notifier_block *nfb,
3206 unsigned long action,
3207 void *ocpu)
3208{
3209 struct sk_buff **list_skb;
3210 struct net_device **list_net;
3211 struct sk_buff *skb;
3212 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3213 struct softnet_data *sd, *oldsd;
3214
3215 if (action != CPU_DEAD)
3216 return NOTIFY_OK;
3217
3218 local_irq_disable();
3219 cpu = smp_processor_id();
3220 sd = &per_cpu(softnet_data, cpu);
3221 oldsd = &per_cpu(softnet_data, oldcpu);
3222
3223 /* Find end of our completion_queue. */
3224 list_skb = &sd->completion_queue;
3225 while (*list_skb)
3226 list_skb = &(*list_skb)->next;
3227 /* Append completion queue from offline CPU. */
3228 *list_skb = oldsd->completion_queue;
3229 oldsd->completion_queue = NULL;
3230
3231 /* Find end of our output_queue. */
3232 list_net = &sd->output_queue;
3233 while (*list_net)
3234 list_net = &(*list_net)->next_sched;
3235 /* Append output queue from offline CPU. */
3236 *list_net = oldsd->output_queue;
3237 oldsd->output_queue = NULL;
3238
3239 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3240 local_irq_enable();
3241
3242 /* Process offline CPU's input_pkt_queue */
3243 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3244 netif_rx(skb);
3245
3246 return NOTIFY_OK;
3247}
3248#endif /* CONFIG_HOTPLUG_CPU */
3249
3250
3251/*
3252 * Initialize the DEV module. At boot time this walks the device list and
3253 * unhooks any devices that fail to initialise (normally hardware not
3254 * present) and leaves us with a valid list of present and active devices.
3255 *
3256 */
3257
3258/*
3259 * This is called single threaded during boot, so no need
3260 * to take the rtnl semaphore.
3261 */
3262static int __init net_dev_init(void)
3263{
3264 int i, rc = -ENOMEM;
3265
3266 BUG_ON(!dev_boot_phase);
3267
3268 net_random_init();
3269
3270 if (dev_proc_init())
3271 goto out;
3272
3273 if (netdev_sysfs_init())
3274 goto out;
3275
3276 INIT_LIST_HEAD(&ptype_all);
3277 for (i = 0; i < 16; i++)
3278 INIT_LIST_HEAD(&ptype_base[i]);
3279
3280 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3281 INIT_HLIST_HEAD(&dev_name_head[i]);
3282
3283 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3284 INIT_HLIST_HEAD(&dev_index_head[i]);
3285
3286 /*
3287 * Initialise the packet receive queues.
3288 */
3289
3290 for (i = 0; i < NR_CPUS; i++) {
3291 struct softnet_data *queue;
3292
3293 queue = &per_cpu(softnet_data, i);
3294 skb_queue_head_init(&queue->input_pkt_queue);
3295 queue->throttle = 0;
3296 queue->cng_level = 0;
3297 queue->avg_blog = 10; /* arbitrary non-zero */
3298 queue->completion_queue = NULL;
3299 INIT_LIST_HEAD(&queue->poll_list);
3300 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3301 queue->backlog_dev.weight = weight_p;
3302 queue->backlog_dev.poll = process_backlog;
3303 atomic_set(&queue->backlog_dev.refcnt, 1);
3304 }
3305
3306#ifdef OFFLINE_SAMPLE
3307 samp_timer.expires = jiffies + (10 * HZ);
3308 add_timer(&samp_timer);
3309#endif
3310
3311 dev_boot_phase = 0;
3312
3313 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3314 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3315
3316 hotcpu_notifier(dev_cpu_callback, 0);
3317 dst_init();
3318 dev_mcast_init();
3319 rc = 0;
3320out:
3321 return rc;
3322}
3323
3324subsys_initcall(net_dev_init);
3325
3326EXPORT_SYMBOL(__dev_get_by_index);
3327EXPORT_SYMBOL(__dev_get_by_name);
3328EXPORT_SYMBOL(__dev_remove_pack);
3329EXPORT_SYMBOL(__skb_linearize);
3330EXPORT_SYMBOL(dev_add_pack);
3331EXPORT_SYMBOL(dev_alloc_name);
3332EXPORT_SYMBOL(dev_close);
3333EXPORT_SYMBOL(dev_get_by_flags);
3334EXPORT_SYMBOL(dev_get_by_index);
3335EXPORT_SYMBOL(dev_get_by_name);
3336EXPORT_SYMBOL(dev_ioctl);
3337EXPORT_SYMBOL(dev_open);
3338EXPORT_SYMBOL(dev_queue_xmit);
3339EXPORT_SYMBOL(dev_remove_pack);
3340EXPORT_SYMBOL(dev_set_allmulti);
3341EXPORT_SYMBOL(dev_set_promiscuity);
3342EXPORT_SYMBOL(dev_change_flags);
3343EXPORT_SYMBOL(dev_set_mtu);
3344EXPORT_SYMBOL(dev_set_mac_address);
3345EXPORT_SYMBOL(free_netdev);
3346EXPORT_SYMBOL(netdev_boot_setup_check);
3347EXPORT_SYMBOL(netdev_set_master);
3348EXPORT_SYMBOL(netdev_state_change);
3349EXPORT_SYMBOL(netif_receive_skb);
3350EXPORT_SYMBOL(netif_rx);
3351EXPORT_SYMBOL(register_gifconf);
3352EXPORT_SYMBOL(register_netdevice);
3353EXPORT_SYMBOL(register_netdevice_notifier);
3354EXPORT_SYMBOL(skb_checksum_help);
3355EXPORT_SYMBOL(synchronize_net);
3356EXPORT_SYMBOL(unregister_netdevice);
3357EXPORT_SYMBOL(unregister_netdevice_notifier);
3358EXPORT_SYMBOL(net_enable_timestamp);
3359EXPORT_SYMBOL(net_disable_timestamp);
3360EXPORT_SYMBOL(dev_get_flags);
3361
3362#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3363EXPORT_SYMBOL(br_handle_frame_hook);
3364EXPORT_SYMBOL(br_fdb_get_hook);
3365EXPORT_SYMBOL(br_fdb_put_hook);
3366#endif
3367
3368#ifdef CONFIG_KMOD
3369EXPORT_SYMBOL(dev_load);
3370#endif
3371
3372EXPORT_PER_CPU_SYMBOL(softnet_data);