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
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
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
158 static DEFINE_SPINLOCK(ptype_lock);
159 static struct list_head ptype_base[16]; /* 16 way hashed list */
160 static struct list_head ptype_all; /* Taps */
161
162 #ifdef OFFLINE_SAMPLE
163 static void sample_queue(unsigned long dummy);
164 static 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 */
186 struct net_device *dev_base;
187 static struct net_device **dev_tail = &dev_base;
188 DEFINE_RWLOCK(dev_base_lock);
189
190 EXPORT_SYMBOL(dev_base);
191 EXPORT_SYMBOL(dev_base_lock);
192
193 #define NETDEV_HASHBITS 8
194 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
195 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
196
197 static 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
203 static 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
212 static 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 */
218 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
219
220 #ifdef CONFIG_SYSFS
221 extern int netdev_sysfs_init(void);
222 extern int netdev_register_sysfs(struct net_device *);
223 extern 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
241 int 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
272 void 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
287 extern 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 */
304 void __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);
325 out:
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 */
340 void 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 */
354 static 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 */
365 static 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 */
392 int 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 */
421 unsigned 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 */
445 int __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
488 struct 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
512 struct 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
535 struct 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
559 struct 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
585 struct 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
598 struct 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
613 EXPORT_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
626 struct 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 */
648 static 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
668 int 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 */
730 int 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 */
771 void 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
788 void 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
800 static 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 */
821 int 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 */
885 int 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
957 int 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
986 int 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
1000 int 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 */
1006 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1007
1008 void net_enable_timestamp(void)
1009 {
1010 atomic_inc(&netstamp_needed);
1011 }
1012
1013 void net_disable_timestamp(void)
1014 {
1015 atomic_dec(&netstamp_needed);
1016 }
1017
1018 static 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
1033 void 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 */
1077 int 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;
1105 out:
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
1115 static 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
1132 extern void skb_release_data(struct sk_buff *);
1133
1134 /* Keep head the same: replace data */
1135 int __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.
1217 */
1218
1219 int dev_queue_xmit(struct sk_buff *skb)
1220 {
1221 struct net_device *dev = skb->dev;
1222 struct Qdisc *q;
1223 int rc = -ENOMEM;
1224
1225 if (skb_shinfo(skb)->frag_list &&
1226 !(dev->features & NETIF_F_FRAGLIST) &&
1227 __skb_linearize(skb, GFP_ATOMIC))
1228 goto out_kfree_skb;
1229
1230 /* Fragmented skb is linearized if device does not support SG,
1231 * or if at least one of fragments is in highmem and device
1232 * does not support DMA from it.
1233 */
1234 if (skb_shinfo(skb)->nr_frags &&
1235 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1236 __skb_linearize(skb, GFP_ATOMIC))
1237 goto out_kfree_skb;
1238
1239 /* If packet is not checksummed and device does not support
1240 * checksumming for this protocol, complete checksumming here.
1241 */
1242 if (skb->ip_summed == CHECKSUM_HW &&
1243 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1244 (!(dev->features & NETIF_F_IP_CSUM) ||
1245 skb->protocol != htons(ETH_P_IP))))
1246 if (skb_checksum_help(skb, 0))
1247 goto out_kfree_skb;
1248
1249 /* Disable soft irqs for various locks below. Also
1250 * stops preemption for RCU.
1251 */
1252 local_bh_disable();
1253
1254 /* Updates of qdisc are serialized by queue_lock.
1255 * The struct Qdisc which is pointed to by qdisc is now a
1256 * rcu structure - it may be accessed without acquiring
1257 * a lock (but the structure may be stale.) The freeing of the
1258 * qdisc will be deferred until it's known that there are no
1259 * more references to it.
1260 *
1261 * If the qdisc has an enqueue function, we still need to
1262 * hold the queue_lock before calling it, since queue_lock
1263 * also serializes access to the device queue.
1264 */
1265
1266 q = rcu_dereference(dev->qdisc);
1267 #ifdef CONFIG_NET_CLS_ACT
1268 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1269 #endif
1270 if (q->enqueue) {
1271 /* Grab device queue */
1272 spin_lock(&dev->queue_lock);
1273
1274 rc = q->enqueue(skb, q);
1275
1276 qdisc_run(dev);
1277
1278 spin_unlock(&dev->queue_lock);
1279 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1280 goto out;
1281 }
1282
1283 /* The device has no queue. Common case for software devices:
1284 loopback, all the sorts of tunnels...
1285
1286 Really, it is unlikely that xmit_lock protection is necessary here.
1287 (f.e. loopback and IP tunnels are clean ignoring statistics
1288 counters.)
1289 However, it is possible, that they rely on protection
1290 made by us here.
1291
1292 Check this and shot the lock. It is not prone from deadlocks.
1293 Either shot noqueue qdisc, it is even simpler 8)
1294 */
1295 if (dev->flags & IFF_UP) {
1296 int cpu = smp_processor_id(); /* ok because BHs are off */
1297
1298 if (dev->xmit_lock_owner != cpu) {
1299
1300 HARD_TX_LOCK(dev, cpu);
1301
1302 if (!netif_queue_stopped(dev)) {
1303 if (netdev_nit)
1304 dev_queue_xmit_nit(skb, dev);
1305
1306 rc = 0;
1307 if (!dev->hard_start_xmit(skb, dev)) {
1308 HARD_TX_UNLOCK(dev);
1309 goto out;
1310 }
1311 }
1312 HARD_TX_UNLOCK(dev);
1313 if (net_ratelimit())
1314 printk(KERN_CRIT "Virtual device %s asks to "
1315 "queue packet!\n", dev->name);
1316 } else {
1317 /* Recursion is detected! It is possible,
1318 * unfortunately */
1319 if (net_ratelimit())
1320 printk(KERN_CRIT "Dead loop on virtual device "
1321 "%s, fix it urgently!\n", dev->name);
1322 }
1323 }
1324
1325 rc = -ENETDOWN;
1326 local_bh_enable();
1327
1328 out_kfree_skb:
1329 kfree_skb(skb);
1330 return rc;
1331 out:
1332 local_bh_enable();
1333 return rc;
1334 }
1335
1336
1337 /*=======================================================================
1338 Receiver routines
1339 =======================================================================*/
1340
1341 int netdev_max_backlog = 300;
1342 int weight_p = 64; /* old backlog weight */
1343 /* These numbers are selected based on intuition and some
1344 * experimentatiom, if you have more scientific way of doing this
1345 * please go ahead and fix things.
1346 */
1347 int no_cong_thresh = 10;
1348 int no_cong = 20;
1349 int lo_cong = 100;
1350 int mod_cong = 290;
1351
1352 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1353
1354
1355 static void get_sample_stats(int cpu)
1356 {
1357 #ifdef RAND_LIE
1358 unsigned long rd;
1359 int rq;
1360 #endif
1361 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1362 int blog = sd->input_pkt_queue.qlen;
1363 int avg_blog = sd->avg_blog;
1364
1365 avg_blog = (avg_blog >> 1) + (blog >> 1);
1366
1367 if (avg_blog > mod_cong) {
1368 /* Above moderate congestion levels. */
1369 sd->cng_level = NET_RX_CN_HIGH;
1370 #ifdef RAND_LIE
1371 rd = net_random();
1372 rq = rd % netdev_max_backlog;
1373 if (rq < avg_blog) /* unlucky bastard */
1374 sd->cng_level = NET_RX_DROP;
1375 #endif
1376 } else if (avg_blog > lo_cong) {
1377 sd->cng_level = NET_RX_CN_MOD;
1378 #ifdef RAND_LIE
1379 rd = net_random();
1380 rq = rd % netdev_max_backlog;
1381 if (rq < avg_blog) /* unlucky bastard */
1382 sd->cng_level = NET_RX_CN_HIGH;
1383 #endif
1384 } else if (avg_blog > no_cong)
1385 sd->cng_level = NET_RX_CN_LOW;
1386 else /* no congestion */
1387 sd->cng_level = NET_RX_SUCCESS;
1388
1389 sd->avg_blog = avg_blog;
1390 }
1391
1392 #ifdef OFFLINE_SAMPLE
1393 static void sample_queue(unsigned long dummy)
1394 {
1395 /* 10 ms 0r 1ms -- i don't care -- JHS */
1396 int next_tick = 1;
1397 int cpu = smp_processor_id();
1398
1399 get_sample_stats(cpu);
1400 next_tick += jiffies;
1401 mod_timer(&samp_timer, next_tick);
1402 }
1403 #endif
1404
1405
1406 /**
1407 * netif_rx - post buffer to the network code
1408 * @skb: buffer to post
1409 *
1410 * This function receives a packet from a device driver and queues it for
1411 * the upper (protocol) levels to process. It always succeeds. The buffer
1412 * may be dropped during processing for congestion control or by the
1413 * protocol layers.
1414 *
1415 * return values:
1416 * NET_RX_SUCCESS (no congestion)
1417 * NET_RX_CN_LOW (low congestion)
1418 * NET_RX_CN_MOD (moderate congestion)
1419 * NET_RX_CN_HIGH (high congestion)
1420 * NET_RX_DROP (packet was dropped)
1421 *
1422 */
1423
1424 int netif_rx(struct sk_buff *skb)
1425 {
1426 int this_cpu;
1427 struct softnet_data *queue;
1428 unsigned long flags;
1429
1430 #ifdef CONFIG_NETPOLL
1431 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1432 kfree_skb(skb);
1433 return NET_RX_DROP;
1434 }
1435 #endif
1436
1437 if (!skb->stamp.tv_sec)
1438 net_timestamp(&skb->stamp);
1439
1440 /*
1441 * The code is rearranged so that the path is the most
1442 * short when CPU is congested, but is still operating.
1443 */
1444 local_irq_save(flags);
1445 this_cpu = smp_processor_id();
1446 queue = &__get_cpu_var(softnet_data);
1447
1448 __get_cpu_var(netdev_rx_stat).total++;
1449 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1450 if (queue->input_pkt_queue.qlen) {
1451 if (queue->throttle)
1452 goto drop;
1453
1454 enqueue:
1455 dev_hold(skb->dev);
1456 __skb_queue_tail(&queue->input_pkt_queue, skb);
1457 #ifndef OFFLINE_SAMPLE
1458 get_sample_stats(this_cpu);
1459 #endif
1460 local_irq_restore(flags);
1461 return queue->cng_level;
1462 }
1463
1464 if (queue->throttle)
1465 queue->throttle = 0;
1466
1467 netif_rx_schedule(&queue->backlog_dev);
1468 goto enqueue;
1469 }
1470
1471 if (!queue->throttle) {
1472 queue->throttle = 1;
1473 __get_cpu_var(netdev_rx_stat).throttled++;
1474 }
1475
1476 drop:
1477 __get_cpu_var(netdev_rx_stat).dropped++;
1478 local_irq_restore(flags);
1479
1480 kfree_skb(skb);
1481 return NET_RX_DROP;
1482 }
1483
1484 int netif_rx_ni(struct sk_buff *skb)
1485 {
1486 int err;
1487
1488 preempt_disable();
1489 err = netif_rx(skb);
1490 if (local_softirq_pending())
1491 do_softirq();
1492 preempt_enable();
1493
1494 return err;
1495 }
1496
1497 EXPORT_SYMBOL(netif_rx_ni);
1498
1499 static __inline__ void skb_bond(struct sk_buff *skb)
1500 {
1501 struct net_device *dev = skb->dev;
1502
1503 if (dev->master) {
1504 skb->real_dev = skb->dev;
1505 skb->dev = dev->master;
1506 }
1507 }
1508
1509 static void net_tx_action(struct softirq_action *h)
1510 {
1511 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1512
1513 if (sd->completion_queue) {
1514 struct sk_buff *clist;
1515
1516 local_irq_disable();
1517 clist = sd->completion_queue;
1518 sd->completion_queue = NULL;
1519 local_irq_enable();
1520
1521 while (clist) {
1522 struct sk_buff *skb = clist;
1523 clist = clist->next;
1524
1525 BUG_TRAP(!atomic_read(&skb->users));
1526 __kfree_skb(skb);
1527 }
1528 }
1529
1530 if (sd->output_queue) {
1531 struct net_device *head;
1532
1533 local_irq_disable();
1534 head = sd->output_queue;
1535 sd->output_queue = NULL;
1536 local_irq_enable();
1537
1538 while (head) {
1539 struct net_device *dev = head;
1540 head = head->next_sched;
1541
1542 smp_mb__before_clear_bit();
1543 clear_bit(__LINK_STATE_SCHED, &dev->state);
1544
1545 if (spin_trylock(&dev->queue_lock)) {
1546 qdisc_run(dev);
1547 spin_unlock(&dev->queue_lock);
1548 } else {
1549 netif_schedule(dev);
1550 }
1551 }
1552 }
1553 }
1554
1555 static __inline__ int deliver_skb(struct sk_buff *skb,
1556 struct packet_type *pt_prev)
1557 {
1558 atomic_inc(&skb->users);
1559 return pt_prev->func(skb, skb->dev, pt_prev);
1560 }
1561
1562 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1563 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1564
1565 static __inline__ int handle_bridge(struct sk_buff **pskb,
1566 struct packet_type **pt_prev, int *ret)
1567 {
1568 struct net_bridge_port *port;
1569
1570 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1571 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1572 return 0;
1573
1574 if (*pt_prev) {
1575 *ret = deliver_skb(*pskb, *pt_prev);
1576 *pt_prev = NULL;
1577 }
1578
1579 return br_handle_frame_hook(port, pskb);
1580 }
1581 #else
1582 #define handle_bridge(skb, pt_prev, ret) (0)
1583 #endif
1584
1585 #ifdef CONFIG_NET_CLS_ACT
1586 /* TODO: Maybe we should just force sch_ingress to be compiled in
1587 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1588 * a compare and 2 stores extra right now if we dont have it on
1589 * but have CONFIG_NET_CLS_ACT
1590 * NOTE: This doesnt stop any functionality; if you dont have
1591 * the ingress scheduler, you just cant add policies on ingress.
1592 *
1593 */
1594 static int ing_filter(struct sk_buff *skb)
1595 {
1596 struct Qdisc *q;
1597 struct net_device *dev = skb->dev;
1598 int result = TC_ACT_OK;
1599
1600 if (dev->qdisc_ingress) {
1601 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1602 if (MAX_RED_LOOP < ttl++) {
1603 printk("Redir loop detected Dropping packet (%s->%s)\n",
1604 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1605 return TC_ACT_SHOT;
1606 }
1607
1608 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1609
1610 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1611 if (NULL == skb->input_dev) {
1612 skb->input_dev = skb->dev;
1613 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1614 }
1615 spin_lock(&dev->ingress_lock);
1616 if ((q = dev->qdisc_ingress) != NULL)
1617 result = q->enqueue(skb, q);
1618 spin_unlock(&dev->ingress_lock);
1619
1620 }
1621
1622 return result;
1623 }
1624 #endif
1625
1626 int netif_receive_skb(struct sk_buff *skb)
1627 {
1628 struct packet_type *ptype, *pt_prev;
1629 int ret = NET_RX_DROP;
1630 unsigned short type;
1631
1632 #ifdef CONFIG_NETPOLL
1633 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1634 kfree_skb(skb);
1635 return NET_RX_DROP;
1636 }
1637 #endif
1638
1639 if (!skb->stamp.tv_sec)
1640 net_timestamp(&skb->stamp);
1641
1642 skb_bond(skb);
1643
1644 __get_cpu_var(netdev_rx_stat).total++;
1645
1646 skb->h.raw = skb->nh.raw = skb->data;
1647 skb->mac_len = skb->nh.raw - skb->mac.raw;
1648
1649 pt_prev = NULL;
1650
1651 rcu_read_lock();
1652
1653 #ifdef CONFIG_NET_CLS_ACT
1654 if (skb->tc_verd & TC_NCLS) {
1655 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1656 goto ncls;
1657 }
1658 #endif
1659
1660 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1661 if (!ptype->dev || ptype->dev == skb->dev) {
1662 if (pt_prev)
1663 ret = deliver_skb(skb, pt_prev);
1664 pt_prev = ptype;
1665 }
1666 }
1667
1668 #ifdef CONFIG_NET_CLS_ACT
1669 if (pt_prev) {
1670 ret = deliver_skb(skb, pt_prev);
1671 pt_prev = NULL; /* noone else should process this after*/
1672 } else {
1673 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1674 }
1675
1676 ret = ing_filter(skb);
1677
1678 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1679 kfree_skb(skb);
1680 goto out;
1681 }
1682
1683 skb->tc_verd = 0;
1684 ncls:
1685 #endif
1686
1687 handle_diverter(skb);
1688
1689 if (handle_bridge(&skb, &pt_prev, &ret))
1690 goto out;
1691
1692 type = skb->protocol;
1693 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1694 if (ptype->type == type &&
1695 (!ptype->dev || ptype->dev == skb->dev)) {
1696 if (pt_prev)
1697 ret = deliver_skb(skb, pt_prev);
1698 pt_prev = ptype;
1699 }
1700 }
1701
1702 if (pt_prev) {
1703 ret = pt_prev->func(skb, skb->dev, pt_prev);
1704 } else {
1705 kfree_skb(skb);
1706 /* Jamal, now you will not able to escape explaining
1707 * me how you were going to use this. :-)
1708 */
1709 ret = NET_RX_DROP;
1710 }
1711
1712 out:
1713 rcu_read_unlock();
1714 return ret;
1715 }
1716
1717 static int process_backlog(struct net_device *backlog_dev, int *budget)
1718 {
1719 int work = 0;
1720 int quota = min(backlog_dev->quota, *budget);
1721 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1722 unsigned long start_time = jiffies;
1723
1724 for (;;) {
1725 struct sk_buff *skb;
1726 struct net_device *dev;
1727
1728 local_irq_disable();
1729 skb = __skb_dequeue(&queue->input_pkt_queue);
1730 if (!skb)
1731 goto job_done;
1732 local_irq_enable();
1733
1734 dev = skb->dev;
1735
1736 netif_receive_skb(skb);
1737
1738 dev_put(dev);
1739
1740 work++;
1741
1742 if (work >= quota || jiffies - start_time > 1)
1743 break;
1744
1745 }
1746
1747 backlog_dev->quota -= work;
1748 *budget -= work;
1749 return -1;
1750
1751 job_done:
1752 backlog_dev->quota -= work;
1753 *budget -= work;
1754
1755 list_del(&backlog_dev->poll_list);
1756 smp_mb__before_clear_bit();
1757 netif_poll_enable(backlog_dev);
1758
1759 if (queue->throttle)
1760 queue->throttle = 0;
1761 local_irq_enable();
1762 return 0;
1763 }
1764
1765 static void net_rx_action(struct softirq_action *h)
1766 {
1767 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1768 unsigned long start_time = jiffies;
1769 int budget = netdev_max_backlog;
1770
1771
1772 local_irq_disable();
1773
1774 while (!list_empty(&queue->poll_list)) {
1775 struct net_device *dev;
1776
1777 if (budget <= 0 || jiffies - start_time > 1)
1778 goto softnet_break;
1779
1780 local_irq_enable();
1781
1782 dev = list_entry(queue->poll_list.next,
1783 struct net_device, poll_list);
1784
1785 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1786 local_irq_disable();
1787 list_del(&dev->poll_list);
1788 list_add_tail(&dev->poll_list, &queue->poll_list);
1789 if (dev->quota < 0)
1790 dev->quota += dev->weight;
1791 else
1792 dev->quota = dev->weight;
1793 } else {
1794 dev_put(dev);
1795 local_irq_disable();
1796 }
1797 }
1798 out:
1799 local_irq_enable();
1800 return;
1801
1802 softnet_break:
1803 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1804 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1805 goto out;
1806 }
1807
1808 static gifconf_func_t * gifconf_list [NPROTO];
1809
1810 /**
1811 * register_gifconf - register a SIOCGIF handler
1812 * @family: Address family
1813 * @gifconf: Function handler
1814 *
1815 * Register protocol dependent address dumping routines. The handler
1816 * that is passed must not be freed or reused until it has been replaced
1817 * by another handler.
1818 */
1819 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1820 {
1821 if (family >= NPROTO)
1822 return -EINVAL;
1823 gifconf_list[family] = gifconf;
1824 return 0;
1825 }
1826
1827
1828 /*
1829 * Map an interface index to its name (SIOCGIFNAME)
1830 */
1831
1832 /*
1833 * We need this ioctl for efficient implementation of the
1834 * if_indextoname() function required by the IPv6 API. Without
1835 * it, we would have to search all the interfaces to find a
1836 * match. --pb
1837 */
1838
1839 static int dev_ifname(struct ifreq __user *arg)
1840 {
1841 struct net_device *dev;
1842 struct ifreq ifr;
1843
1844 /*
1845 * Fetch the caller's info block.
1846 */
1847
1848 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1849 return -EFAULT;
1850
1851 read_lock(&dev_base_lock);
1852 dev = __dev_get_by_index(ifr.ifr_ifindex);
1853 if (!dev) {
1854 read_unlock(&dev_base_lock);
1855 return -ENODEV;
1856 }
1857
1858 strcpy(ifr.ifr_name, dev->name);
1859 read_unlock(&dev_base_lock);
1860
1861 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1862 return -EFAULT;
1863 return 0;
1864 }
1865
1866 /*
1867 * Perform a SIOCGIFCONF call. This structure will change
1868 * size eventually, and there is nothing I can do about it.
1869 * Thus we will need a 'compatibility mode'.
1870 */
1871
1872 static int dev_ifconf(char __user *arg)
1873 {
1874 struct ifconf ifc;
1875 struct net_device *dev;
1876 char __user *pos;
1877 int len;
1878 int total;
1879 int i;
1880
1881 /*
1882 * Fetch the caller's info block.
1883 */
1884
1885 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1886 return -EFAULT;
1887
1888 pos = ifc.ifc_buf;
1889 len = ifc.ifc_len;
1890
1891 /*
1892 * Loop over the interfaces, and write an info block for each.
1893 */
1894
1895 total = 0;
1896 for (dev = dev_base; dev; dev = dev->next) {
1897 for (i = 0; i < NPROTO; i++) {
1898 if (gifconf_list[i]) {
1899 int done;
1900 if (!pos)
1901 done = gifconf_list[i](dev, NULL, 0);
1902 else
1903 done = gifconf_list[i](dev, pos + total,
1904 len - total);
1905 if (done < 0)
1906 return -EFAULT;
1907 total += done;
1908 }
1909 }
1910 }
1911
1912 /*
1913 * All done. Write the updated control block back to the caller.
1914 */
1915 ifc.ifc_len = total;
1916
1917 /*
1918 * Both BSD and Solaris return 0 here, so we do too.
1919 */
1920 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1921 }
1922
1923 #ifdef CONFIG_PROC_FS
1924 /*
1925 * This is invoked by the /proc filesystem handler to display a device
1926 * in detail.
1927 */
1928 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1929 {
1930 struct net_device *dev;
1931 loff_t i;
1932
1933 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1934
1935 return i == pos ? dev : NULL;
1936 }
1937
1938 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1939 {
1940 read_lock(&dev_base_lock);
1941 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1942 }
1943
1944 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1945 {
1946 ++*pos;
1947 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1948 }
1949
1950 void dev_seq_stop(struct seq_file *seq, void *v)
1951 {
1952 read_unlock(&dev_base_lock);
1953 }
1954
1955 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1956 {
1957 if (dev->get_stats) {
1958 struct net_device_stats *stats = dev->get_stats(dev);
1959
1960 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1961 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1962 dev->name, stats->rx_bytes, stats->rx_packets,
1963 stats->rx_errors,
1964 stats->rx_dropped + stats->rx_missed_errors,
1965 stats->rx_fifo_errors,
1966 stats->rx_length_errors + stats->rx_over_errors +
1967 stats->rx_crc_errors + stats->rx_frame_errors,
1968 stats->rx_compressed, stats->multicast,
1969 stats->tx_bytes, stats->tx_packets,
1970 stats->tx_errors, stats->tx_dropped,
1971 stats->tx_fifo_errors, stats->collisions,
1972 stats->tx_carrier_errors +
1973 stats->tx_aborted_errors +
1974 stats->tx_window_errors +
1975 stats->tx_heartbeat_errors,
1976 stats->tx_compressed);
1977 } else
1978 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1979 }
1980
1981 /*
1982 * Called from the PROCfs module. This now uses the new arbitrary sized
1983 * /proc/net interface to create /proc/net/dev
1984 */
1985 static int dev_seq_show(struct seq_file *seq, void *v)
1986 {
1987 if (v == SEQ_START_TOKEN)
1988 seq_puts(seq, "Inter-| Receive "
1989 " | Transmit\n"
1990 " face |bytes packets errs drop fifo frame "
1991 "compressed multicast|bytes packets errs "
1992 "drop fifo colls carrier compressed\n");
1993 else
1994 dev_seq_printf_stats(seq, v);
1995 return 0;
1996 }
1997
1998 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1999 {
2000 struct netif_rx_stats *rc = NULL;
2001
2002 while (*pos < NR_CPUS)
2003 if (cpu_online(*pos)) {
2004 rc = &per_cpu(netdev_rx_stat, *pos);
2005 break;
2006 } else
2007 ++*pos;
2008 return rc;
2009 }
2010
2011 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2012 {
2013 return softnet_get_online(pos);
2014 }
2015
2016 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2017 {
2018 ++*pos;
2019 return softnet_get_online(pos);
2020 }
2021
2022 static void softnet_seq_stop(struct seq_file *seq, void *v)
2023 {
2024 }
2025
2026 static int softnet_seq_show(struct seq_file *seq, void *v)
2027 {
2028 struct netif_rx_stats *s = v;
2029
2030 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2031 s->total, s->dropped, s->time_squeeze, s->throttled,
2032 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2033 s->fastroute_deferred_out,
2034 #if 0
2035 s->fastroute_latency_reduction
2036 #else
2037 s->cpu_collision
2038 #endif
2039 );
2040 return 0;
2041 }
2042
2043 static struct seq_operations dev_seq_ops = {
2044 .start = dev_seq_start,
2045 .next = dev_seq_next,
2046 .stop = dev_seq_stop,
2047 .show = dev_seq_show,
2048 };
2049
2050 static int dev_seq_open(struct inode *inode, struct file *file)
2051 {
2052 return seq_open(file, &dev_seq_ops);
2053 }
2054
2055 static struct file_operations dev_seq_fops = {
2056 .owner = THIS_MODULE,
2057 .open = dev_seq_open,
2058 .read = seq_read,
2059 .llseek = seq_lseek,
2060 .release = seq_release,
2061 };
2062
2063 static struct seq_operations softnet_seq_ops = {
2064 .start = softnet_seq_start,
2065 .next = softnet_seq_next,
2066 .stop = softnet_seq_stop,
2067 .show = softnet_seq_show,
2068 };
2069
2070 static int softnet_seq_open(struct inode *inode, struct file *file)
2071 {
2072 return seq_open(file, &softnet_seq_ops);
2073 }
2074
2075 static struct file_operations softnet_seq_fops = {
2076 .owner = THIS_MODULE,
2077 .open = softnet_seq_open,
2078 .read = seq_read,
2079 .llseek = seq_lseek,
2080 .release = seq_release,
2081 };
2082
2083 #ifdef WIRELESS_EXT
2084 extern int wireless_proc_init(void);
2085 #else
2086 #define wireless_proc_init() 0
2087 #endif
2088
2089 static int __init dev_proc_init(void)
2090 {
2091 int rc = -ENOMEM;
2092
2093 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2094 goto out;
2095 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2096 goto out_dev;
2097 if (wireless_proc_init())
2098 goto out_softnet;
2099 rc = 0;
2100 out:
2101 return rc;
2102 out_softnet:
2103 proc_net_remove("softnet_stat");
2104 out_dev:
2105 proc_net_remove("dev");
2106 goto out;
2107 }
2108 #else
2109 #define dev_proc_init() 0
2110 #endif /* CONFIG_PROC_FS */
2111
2112
2113 /**
2114 * netdev_set_master - set up master/slave pair
2115 * @slave: slave device
2116 * @master: new master device
2117 *
2118 * Changes the master device of the slave. Pass %NULL to break the
2119 * bonding. The caller must hold the RTNL semaphore. On a failure
2120 * a negative errno code is returned. On success the reference counts
2121 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2122 * function returns zero.
2123 */
2124 int netdev_set_master(struct net_device *slave, struct net_device *master)
2125 {
2126 struct net_device *old = slave->master;
2127
2128 ASSERT_RTNL();
2129
2130 if (master) {
2131 if (old)
2132 return -EBUSY;
2133 dev_hold(master);
2134 }
2135
2136 slave->master = master;
2137
2138 synchronize_net();
2139
2140 if (old)
2141 dev_put(old);
2142
2143 if (master)
2144 slave->flags |= IFF_SLAVE;
2145 else
2146 slave->flags &= ~IFF_SLAVE;
2147
2148 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2149 return 0;
2150 }
2151
2152 /**
2153 * dev_set_promiscuity - update promiscuity count on a device
2154 * @dev: device
2155 * @inc: modifier
2156 *
2157 * Add or remove promsicuity from a device. While the count in the device
2158 * remains above zero the interface remains promiscuous. Once it hits zero
2159 * the device reverts back to normal filtering operation. A negative inc
2160 * value is used to drop promiscuity on the device.
2161 */
2162 void dev_set_promiscuity(struct net_device *dev, int inc)
2163 {
2164 unsigned short old_flags = dev->flags;
2165
2166 dev->flags |= IFF_PROMISC;
2167 if ((dev->promiscuity += inc) == 0)
2168 dev->flags &= ~IFF_PROMISC;
2169 if (dev->flags ^ old_flags) {
2170 dev_mc_upload(dev);
2171 printk(KERN_INFO "device %s %s promiscuous mode\n",
2172 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2173 "left");
2174 }
2175 }
2176
2177 /**
2178 * dev_set_allmulti - update allmulti count on a device
2179 * @dev: device
2180 * @inc: modifier
2181 *
2182 * Add or remove reception of all multicast frames to a device. While the
2183 * count in the device remains above zero the interface remains listening
2184 * to all interfaces. Once it hits zero the device reverts back to normal
2185 * filtering operation. A negative @inc value is used to drop the counter
2186 * when releasing a resource needing all multicasts.
2187 */
2188
2189 void dev_set_allmulti(struct net_device *dev, int inc)
2190 {
2191 unsigned short old_flags = dev->flags;
2192
2193 dev->flags |= IFF_ALLMULTI;
2194 if ((dev->allmulti += inc) == 0)
2195 dev->flags &= ~IFF_ALLMULTI;
2196 if (dev->flags ^ old_flags)
2197 dev_mc_upload(dev);
2198 }
2199
2200 unsigned dev_get_flags(const struct net_device *dev)
2201 {
2202 unsigned flags;
2203
2204 flags = (dev->flags & ~(IFF_PROMISC |
2205 IFF_ALLMULTI |
2206 IFF_RUNNING)) |
2207 (dev->gflags & (IFF_PROMISC |
2208 IFF_ALLMULTI));
2209
2210 if (netif_running(dev) && netif_carrier_ok(dev))
2211 flags |= IFF_RUNNING;
2212
2213 return flags;
2214 }
2215
2216 int dev_change_flags(struct net_device *dev, unsigned flags)
2217 {
2218 int ret;
2219 int old_flags = dev->flags;
2220
2221 /*
2222 * Set the flags on our device.
2223 */
2224
2225 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2226 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2227 IFF_AUTOMEDIA)) |
2228 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2229 IFF_ALLMULTI));
2230
2231 /*
2232 * Load in the correct multicast list now the flags have changed.
2233 */
2234
2235 dev_mc_upload(dev);
2236
2237 /*
2238 * Have we downed the interface. We handle IFF_UP ourselves
2239 * according to user attempts to set it, rather than blindly
2240 * setting it.
2241 */
2242
2243 ret = 0;
2244 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2245 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2246
2247 if (!ret)
2248 dev_mc_upload(dev);
2249 }
2250
2251 if (dev->flags & IFF_UP &&
2252 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2253 IFF_VOLATILE)))
2254 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2255
2256 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2257 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2258 dev->gflags ^= IFF_PROMISC;
2259 dev_set_promiscuity(dev, inc);
2260 }
2261
2262 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2263 is important. Some (broken) drivers set IFF_PROMISC, when
2264 IFF_ALLMULTI is requested not asking us and not reporting.
2265 */
2266 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2267 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2268 dev->gflags ^= IFF_ALLMULTI;
2269 dev_set_allmulti(dev, inc);
2270 }
2271
2272 if (old_flags ^ dev->flags)
2273 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2274
2275 return ret;
2276 }
2277
2278 int dev_set_mtu(struct net_device *dev, int new_mtu)
2279 {
2280 int err;
2281
2282 if (new_mtu == dev->mtu)
2283 return 0;
2284
2285 /* MTU must be positive. */
2286 if (new_mtu < 0)
2287 return -EINVAL;
2288
2289 if (!netif_device_present(dev))
2290 return -ENODEV;
2291
2292 err = 0;
2293 if (dev->change_mtu)
2294 err = dev->change_mtu(dev, new_mtu);
2295 else
2296 dev->mtu = new_mtu;
2297 if (!err && dev->flags & IFF_UP)
2298 notifier_call_chain(&netdev_chain,
2299 NETDEV_CHANGEMTU, dev);
2300 return err;
2301 }
2302
2303
2304 /*
2305 * Perform the SIOCxIFxxx calls.
2306 */
2307 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2308 {
2309 int err;
2310 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2311
2312 if (!dev)
2313 return -ENODEV;
2314
2315 switch (cmd) {
2316 case SIOCGIFFLAGS: /* Get interface flags */
2317 ifr->ifr_flags = dev_get_flags(dev);
2318 return 0;
2319
2320 case SIOCSIFFLAGS: /* Set interface flags */
2321 return dev_change_flags(dev, ifr->ifr_flags);
2322
2323 case SIOCGIFMETRIC: /* Get the metric on the interface
2324 (currently unused) */
2325 ifr->ifr_metric = 0;
2326 return 0;
2327
2328 case SIOCSIFMETRIC: /* Set the metric on the interface
2329 (currently unused) */
2330 return -EOPNOTSUPP;
2331
2332 case SIOCGIFMTU: /* Get the MTU of a device */
2333 ifr->ifr_mtu = dev->mtu;
2334 return 0;
2335
2336 case SIOCSIFMTU: /* Set the MTU of a device */
2337 return dev_set_mtu(dev, ifr->ifr_mtu);
2338
2339 case SIOCGIFHWADDR:
2340 if (!dev->addr_len)
2341 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2342 else
2343 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2344 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2345 ifr->ifr_hwaddr.sa_family = dev->type;
2346 return 0;
2347
2348 case SIOCSIFHWADDR:
2349 if (!dev->set_mac_address)
2350 return -EOPNOTSUPP;
2351 if (ifr->ifr_hwaddr.sa_family != dev->type)
2352 return -EINVAL;
2353 if (!netif_device_present(dev))
2354 return -ENODEV;
2355 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2356 if (!err)
2357 notifier_call_chain(&netdev_chain,
2358 NETDEV_CHANGEADDR, dev);
2359 return err;
2360
2361 case SIOCSIFHWBROADCAST:
2362 if (ifr->ifr_hwaddr.sa_family != dev->type)
2363 return -EINVAL;
2364 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2365 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2366 notifier_call_chain(&netdev_chain,
2367 NETDEV_CHANGEADDR, dev);
2368 return 0;
2369
2370 case SIOCGIFMAP:
2371 ifr->ifr_map.mem_start = dev->mem_start;
2372 ifr->ifr_map.mem_end = dev->mem_end;
2373 ifr->ifr_map.base_addr = dev->base_addr;
2374 ifr->ifr_map.irq = dev->irq;
2375 ifr->ifr_map.dma = dev->dma;
2376 ifr->ifr_map.port = dev->if_port;
2377 return 0;
2378
2379 case SIOCSIFMAP:
2380 if (dev->set_config) {
2381 if (!netif_device_present(dev))
2382 return -ENODEV;
2383 return dev->set_config(dev, &ifr->ifr_map);
2384 }
2385 return -EOPNOTSUPP;
2386
2387 case SIOCADDMULTI:
2388 if (!dev->set_multicast_list ||
2389 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2390 return -EINVAL;
2391 if (!netif_device_present(dev))
2392 return -ENODEV;
2393 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2394 dev->addr_len, 1);
2395
2396 case SIOCDELMULTI:
2397 if (!dev->set_multicast_list ||
2398 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2399 return -EINVAL;
2400 if (!netif_device_present(dev))
2401 return -ENODEV;
2402 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2403 dev->addr_len, 1);
2404
2405 case SIOCGIFINDEX:
2406 ifr->ifr_ifindex = dev->ifindex;
2407 return 0;
2408
2409 case SIOCGIFTXQLEN:
2410 ifr->ifr_qlen = dev->tx_queue_len;
2411 return 0;
2412
2413 case SIOCSIFTXQLEN:
2414 if (ifr->ifr_qlen < 0)
2415 return -EINVAL;
2416 dev->tx_queue_len = ifr->ifr_qlen;
2417 return 0;
2418
2419 case SIOCSIFNAME:
2420 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2421 return dev_change_name(dev, ifr->ifr_newname);
2422
2423 /*
2424 * Unknown or private ioctl
2425 */
2426
2427 default:
2428 if ((cmd >= SIOCDEVPRIVATE &&
2429 cmd <= SIOCDEVPRIVATE + 15) ||
2430 cmd == SIOCBONDENSLAVE ||
2431 cmd == SIOCBONDRELEASE ||
2432 cmd == SIOCBONDSETHWADDR ||
2433 cmd == SIOCBONDSLAVEINFOQUERY ||
2434 cmd == SIOCBONDINFOQUERY ||
2435 cmd == SIOCBONDCHANGEACTIVE ||
2436 cmd == SIOCGMIIPHY ||
2437 cmd == SIOCGMIIREG ||
2438 cmd == SIOCSMIIREG ||
2439 cmd == SIOCBRADDIF ||
2440 cmd == SIOCBRDELIF ||
2441 cmd == SIOCWANDEV) {
2442 err = -EOPNOTSUPP;
2443 if (dev->do_ioctl) {
2444 if (netif_device_present(dev))
2445 err = dev->do_ioctl(dev, ifr,
2446 cmd);
2447 else
2448 err = -ENODEV;
2449 }
2450 } else
2451 err = -EINVAL;
2452
2453 }
2454 return err;
2455 }
2456
2457 /*
2458 * This function handles all "interface"-type I/O control requests. The actual
2459 * 'doing' part of this is dev_ifsioc above.
2460 */
2461
2462 /**
2463 * dev_ioctl - network device ioctl
2464 * @cmd: command to issue
2465 * @arg: pointer to a struct ifreq in user space
2466 *
2467 * Issue ioctl functions to devices. This is normally called by the
2468 * user space syscall interfaces but can sometimes be useful for
2469 * other purposes. The return value is the return from the syscall if
2470 * positive or a negative errno code on error.
2471 */
2472
2473 int dev_ioctl(unsigned int cmd, void __user *arg)
2474 {
2475 struct ifreq ifr;
2476 int ret;
2477 char *colon;
2478
2479 /* One special case: SIOCGIFCONF takes ifconf argument
2480 and requires shared lock, because it sleeps writing
2481 to user space.
2482 */
2483
2484 if (cmd == SIOCGIFCONF) {
2485 rtnl_shlock();
2486 ret = dev_ifconf((char __user *) arg);
2487 rtnl_shunlock();
2488 return ret;
2489 }
2490 if (cmd == SIOCGIFNAME)
2491 return dev_ifname((struct ifreq __user *)arg);
2492
2493 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2494 return -EFAULT;
2495
2496 ifr.ifr_name[IFNAMSIZ-1] = 0;
2497
2498 colon = strchr(ifr.ifr_name, ':');
2499 if (colon)
2500 *colon = 0;
2501
2502 /*
2503 * See which interface the caller is talking about.
2504 */
2505
2506 switch (cmd) {
2507 /*
2508 * These ioctl calls:
2509 * - can be done by all.
2510 * - atomic and do not require locking.
2511 * - return a value
2512 */
2513 case SIOCGIFFLAGS:
2514 case SIOCGIFMETRIC:
2515 case SIOCGIFMTU:
2516 case SIOCGIFHWADDR:
2517 case SIOCGIFSLAVE:
2518 case SIOCGIFMAP:
2519 case SIOCGIFINDEX:
2520 case SIOCGIFTXQLEN:
2521 dev_load(ifr.ifr_name);
2522 read_lock(&dev_base_lock);
2523 ret = dev_ifsioc(&ifr, cmd);
2524 read_unlock(&dev_base_lock);
2525 if (!ret) {
2526 if (colon)
2527 *colon = ':';
2528 if (copy_to_user(arg, &ifr,
2529 sizeof(struct ifreq)))
2530 ret = -EFAULT;
2531 }
2532 return ret;
2533
2534 case SIOCETHTOOL:
2535 dev_load(ifr.ifr_name);
2536 rtnl_lock();
2537 ret = dev_ethtool(&ifr);
2538 rtnl_unlock();
2539 if (!ret) {
2540 if (colon)
2541 *colon = ':';
2542 if (copy_to_user(arg, &ifr,
2543 sizeof(struct ifreq)))
2544 ret = -EFAULT;
2545 }
2546 return ret;
2547
2548 /*
2549 * These ioctl calls:
2550 * - require superuser power.
2551 * - require strict serialization.
2552 * - return a value
2553 */
2554 case SIOCGMIIPHY:
2555 case SIOCGMIIREG:
2556 case SIOCSIFNAME:
2557 if (!capable(CAP_NET_ADMIN))
2558 return -EPERM;
2559 dev_load(ifr.ifr_name);
2560 rtnl_lock();
2561 ret = dev_ifsioc(&ifr, cmd);
2562 rtnl_unlock();
2563 if (!ret) {
2564 if (colon)
2565 *colon = ':';
2566 if (copy_to_user(arg, &ifr,
2567 sizeof(struct ifreq)))
2568 ret = -EFAULT;
2569 }
2570 return ret;
2571
2572 /*
2573 * These ioctl calls:
2574 * - require superuser power.
2575 * - require strict serialization.
2576 * - do not return a value
2577 */
2578 case SIOCSIFFLAGS:
2579 case SIOCSIFMETRIC:
2580 case SIOCSIFMTU:
2581 case SIOCSIFMAP:
2582 case SIOCSIFHWADDR:
2583 case SIOCSIFSLAVE:
2584 case SIOCADDMULTI:
2585 case SIOCDELMULTI:
2586 case SIOCSIFHWBROADCAST:
2587 case SIOCSIFTXQLEN:
2588 case SIOCSMIIREG:
2589 case SIOCBONDENSLAVE:
2590 case SIOCBONDRELEASE:
2591 case SIOCBONDSETHWADDR:
2592 case SIOCBONDSLAVEINFOQUERY:
2593 case SIOCBONDINFOQUERY:
2594 case SIOCBONDCHANGEACTIVE:
2595 case SIOCBRADDIF:
2596 case SIOCBRDELIF:
2597 if (!capable(CAP_NET_ADMIN))
2598 return -EPERM;
2599 dev_load(ifr.ifr_name);
2600 rtnl_lock();
2601 ret = dev_ifsioc(&ifr, cmd);
2602 rtnl_unlock();
2603 return ret;
2604
2605 case SIOCGIFMEM:
2606 /* Get the per device memory space. We can add this but
2607 * currently do not support it */
2608 case SIOCSIFMEM:
2609 /* Set the per device memory buffer space.
2610 * Not applicable in our case */
2611 case SIOCSIFLINK:
2612 return -EINVAL;
2613
2614 /*
2615 * Unknown or private ioctl.
2616 */
2617 default:
2618 if (cmd == SIOCWANDEV ||
2619 (cmd >= SIOCDEVPRIVATE &&
2620 cmd <= SIOCDEVPRIVATE + 15)) {
2621 dev_load(ifr.ifr_name);
2622 rtnl_lock();
2623 ret = dev_ifsioc(&ifr, cmd);
2624 rtnl_unlock();
2625 if (!ret && copy_to_user(arg, &ifr,
2626 sizeof(struct ifreq)))
2627 ret = -EFAULT;
2628 return ret;
2629 }
2630 #ifdef WIRELESS_EXT
2631 /* Take care of Wireless Extensions */
2632 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2633 /* If command is `set a parameter', or
2634 * `get the encoding parameters', check if
2635 * the user has the right to do it */
2636 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2637 if (!capable(CAP_NET_ADMIN))
2638 return -EPERM;
2639 }
2640 dev_load(ifr.ifr_name);
2641 rtnl_lock();
2642 /* Follow me in net/core/wireless.c */
2643 ret = wireless_process_ioctl(&ifr, cmd);
2644 rtnl_unlock();
2645 if (IW_IS_GET(cmd) &&
2646 copy_to_user(arg, &ifr,
2647 sizeof(struct ifreq)))
2648 ret = -EFAULT;
2649 return ret;
2650 }
2651 #endif /* WIRELESS_EXT */
2652 return -EINVAL;
2653 }
2654 }
2655
2656
2657 /**
2658 * dev_new_index - allocate an ifindex
2659 *
2660 * Returns a suitable unique value for a new device interface
2661 * number. The caller must hold the rtnl semaphore or the
2662 * dev_base_lock to be sure it remains unique.
2663 */
2664 static int dev_new_index(void)
2665 {
2666 static int ifindex;
2667 for (;;) {
2668 if (++ifindex <= 0)
2669 ifindex = 1;
2670 if (!__dev_get_by_index(ifindex))
2671 return ifindex;
2672 }
2673 }
2674
2675 static int dev_boot_phase = 1;
2676
2677 /* Delayed registration/unregisteration */
2678 static DEFINE_SPINLOCK(net_todo_list_lock);
2679 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2680
2681 static inline void net_set_todo(struct net_device *dev)
2682 {
2683 spin_lock(&net_todo_list_lock);
2684 list_add_tail(&dev->todo_list, &net_todo_list);
2685 spin_unlock(&net_todo_list_lock);
2686 }
2687
2688 /**
2689 * register_netdevice - register a network device
2690 * @dev: device to register
2691 *
2692 * Take a completed network device structure and add it to the kernel
2693 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2694 * chain. 0 is returned on success. A negative errno code is returned
2695 * on a failure to set up the device, or if the name is a duplicate.
2696 *
2697 * Callers must hold the rtnl semaphore. You may want
2698 * register_netdev() instead of this.
2699 *
2700 * BUGS:
2701 * The locking appears insufficient to guarantee two parallel registers
2702 * will not get the same name.
2703 */
2704
2705 int register_netdevice(struct net_device *dev)
2706 {
2707 struct hlist_head *head;
2708 struct hlist_node *p;
2709 int ret;
2710
2711 BUG_ON(dev_boot_phase);
2712 ASSERT_RTNL();
2713
2714 /* When net_device's are persistent, this will be fatal. */
2715 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2716
2717 spin_lock_init(&dev->queue_lock);
2718 spin_lock_init(&dev->xmit_lock);
2719 dev->xmit_lock_owner = -1;
2720 #ifdef CONFIG_NET_CLS_ACT
2721 spin_lock_init(&dev->ingress_lock);
2722 #endif
2723
2724 ret = alloc_divert_blk(dev);
2725 if (ret)
2726 goto out;
2727
2728 dev->iflink = -1;
2729
2730 /* Init, if this function is available */
2731 if (dev->init) {
2732 ret = dev->init(dev);
2733 if (ret) {
2734 if (ret > 0)
2735 ret = -EIO;
2736 goto out_err;
2737 }
2738 }
2739
2740 if (!dev_valid_name(dev->name)) {
2741 ret = -EINVAL;
2742 goto out_err;
2743 }
2744
2745 dev->ifindex = dev_new_index();
2746 if (dev->iflink == -1)
2747 dev->iflink = dev->ifindex;
2748
2749 /* Check for existence of name */
2750 head = dev_name_hash(dev->name);
2751 hlist_for_each(p, head) {
2752 struct net_device *d
2753 = hlist_entry(p, struct net_device, name_hlist);
2754 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2755 ret = -EEXIST;
2756 goto out_err;
2757 }
2758 }
2759
2760 /* Fix illegal SG+CSUM combinations. */
2761 if ((dev->features & NETIF_F_SG) &&
2762 !(dev->features & (NETIF_F_IP_CSUM |
2763 NETIF_F_NO_CSUM |
2764 NETIF_F_HW_CSUM))) {
2765 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2766 dev->name);
2767 dev->features &= ~NETIF_F_SG;
2768 }
2769
2770 /* TSO requires that SG is present as well. */
2771 if ((dev->features & NETIF_F_TSO) &&
2772 !(dev->features & NETIF_F_SG)) {
2773 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2774 dev->name);
2775 dev->features &= ~NETIF_F_TSO;
2776 }
2777
2778 /*
2779 * nil rebuild_header routine,
2780 * that should be never called and used as just bug trap.
2781 */
2782
2783 if (!dev->rebuild_header)
2784 dev->rebuild_header = default_rebuild_header;
2785
2786 /*
2787 * Default initial state at registry is that the
2788 * device is present.
2789 */
2790
2791 set_bit(__LINK_STATE_PRESENT, &dev->state);
2792
2793 dev->next = NULL;
2794 dev_init_scheduler(dev);
2795 write_lock_bh(&dev_base_lock);
2796 *dev_tail = dev;
2797 dev_tail = &dev->next;
2798 hlist_add_head(&dev->name_hlist, head);
2799 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2800 dev_hold(dev);
2801 dev->reg_state = NETREG_REGISTERING;
2802 write_unlock_bh(&dev_base_lock);
2803
2804 /* Notify protocols, that a new device appeared. */
2805 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2806
2807 /* Finish registration after unlock */
2808 net_set_todo(dev);
2809 ret = 0;
2810
2811 out:
2812 return ret;
2813 out_err:
2814 free_divert_blk(dev);
2815 goto out;
2816 }
2817
2818 /**
2819 * register_netdev - register a network device
2820 * @dev: device to register
2821 *
2822 * Take a completed network device structure and add it to the kernel
2823 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2824 * chain. 0 is returned on success. A negative errno code is returned
2825 * on a failure to set up the device, or if the name is a duplicate.
2826 *
2827 * This is a wrapper around register_netdev that takes the rtnl semaphore
2828 * and expands the device name if you passed a format string to
2829 * alloc_netdev.
2830 */
2831 int register_netdev(struct net_device *dev)
2832 {
2833 int err;
2834
2835 rtnl_lock();
2836
2837 /*
2838 * If the name is a format string the caller wants us to do a
2839 * name allocation.
2840 */
2841 if (strchr(dev->name, '%')) {
2842 err = dev_alloc_name(dev, dev->name);
2843 if (err < 0)
2844 goto out;
2845 }
2846
2847 /*
2848 * Back compatibility hook. Kill this one in 2.5
2849 */
2850 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2851 err = dev_alloc_name(dev, "eth%d");
2852 if (err < 0)
2853 goto out;
2854 }
2855
2856 err = register_netdevice(dev);
2857 out:
2858 rtnl_unlock();
2859 return err;
2860 }
2861 EXPORT_SYMBOL(register_netdev);
2862
2863 /*
2864 * netdev_wait_allrefs - wait until all references are gone.
2865 *
2866 * This is called when unregistering network devices.
2867 *
2868 * Any protocol or device that holds a reference should register
2869 * for netdevice notification, and cleanup and put back the
2870 * reference if they receive an UNREGISTER event.
2871 * We can get stuck here if buggy protocols don't correctly
2872 * call dev_put.
2873 */
2874 static void netdev_wait_allrefs(struct net_device *dev)
2875 {
2876 unsigned long rebroadcast_time, warning_time;
2877
2878 rebroadcast_time = warning_time = jiffies;
2879 while (atomic_read(&dev->refcnt) != 0) {
2880 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2881 rtnl_shlock();
2882
2883 /* Rebroadcast unregister notification */
2884 notifier_call_chain(&netdev_chain,
2885 NETDEV_UNREGISTER, dev);
2886
2887 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2888 &dev->state)) {
2889 /* We must not have linkwatch events
2890 * pending on unregister. If this
2891 * happens, we simply run the queue
2892 * unscheduled, resulting in a noop
2893 * for this device.
2894 */
2895 linkwatch_run_queue();
2896 }
2897
2898 rtnl_shunlock();
2899
2900 rebroadcast_time = jiffies;
2901 }
2902
2903 msleep(250);
2904
2905 if (time_after(jiffies, warning_time + 10 * HZ)) {
2906 printk(KERN_EMERG "unregister_netdevice: "
2907 "waiting for %s to become free. Usage "
2908 "count = %d\n",
2909 dev->name, atomic_read(&dev->refcnt));
2910 warning_time = jiffies;
2911 }
2912 }
2913 }
2914
2915 /* The sequence is:
2916 *
2917 * rtnl_lock();
2918 * ...
2919 * register_netdevice(x1);
2920 * register_netdevice(x2);
2921 * ...
2922 * unregister_netdevice(y1);
2923 * unregister_netdevice(y2);
2924 * ...
2925 * rtnl_unlock();
2926 * free_netdev(y1);
2927 * free_netdev(y2);
2928 *
2929 * We are invoked by rtnl_unlock() after it drops the semaphore.
2930 * This allows us to deal with problems:
2931 * 1) We can create/delete sysfs objects which invoke hotplug
2932 * without deadlocking with linkwatch via keventd.
2933 * 2) Since we run with the RTNL semaphore not held, we can sleep
2934 * safely in order to wait for the netdev refcnt to drop to zero.
2935 */
2936 static DECLARE_MUTEX(net_todo_run_mutex);
2937 void netdev_run_todo(void)
2938 {
2939 struct list_head list = LIST_HEAD_INIT(list);
2940 int err;
2941
2942
2943 /* Need to guard against multiple cpu's getting out of order. */
2944 down(&net_todo_run_mutex);
2945
2946 /* Not safe to do outside the semaphore. We must not return
2947 * until all unregister events invoked by the local processor
2948 * have been completed (either by this todo run, or one on
2949 * another cpu).
2950 */
2951 if (list_empty(&net_todo_list))
2952 goto out;
2953
2954 /* Snapshot list, allow later requests */
2955 spin_lock(&net_todo_list_lock);
2956 list_splice_init(&net_todo_list, &list);
2957 spin_unlock(&net_todo_list_lock);
2958
2959 while (!list_empty(&list)) {
2960 struct net_device *dev
2961 = list_entry(list.next, struct net_device, todo_list);
2962 list_del(&dev->todo_list);
2963
2964 switch(dev->reg_state) {
2965 case NETREG_REGISTERING:
2966 err = netdev_register_sysfs(dev);
2967 if (err)
2968 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2969 dev->name, err);
2970 dev->reg_state = NETREG_REGISTERED;
2971 break;
2972
2973 case NETREG_UNREGISTERING:
2974 netdev_unregister_sysfs(dev);
2975 dev->reg_state = NETREG_UNREGISTERED;
2976
2977 netdev_wait_allrefs(dev);
2978
2979 /* paranoia */
2980 BUG_ON(atomic_read(&dev->refcnt));
2981 BUG_TRAP(!dev->ip_ptr);
2982 BUG_TRAP(!dev->ip6_ptr);
2983 BUG_TRAP(!dev->dn_ptr);
2984
2985
2986 /* It must be the very last action,
2987 * after this 'dev' may point to freed up memory.
2988 */
2989 if (dev->destructor)
2990 dev->destructor(dev);
2991 break;
2992
2993 default:
2994 printk(KERN_ERR "network todo '%s' but state %d\n",
2995 dev->name, dev->reg_state);
2996 break;
2997 }
2998 }
2999
3000 out:
3001 up(&net_todo_run_mutex);
3002 }
3003
3004 /**
3005 * alloc_netdev - allocate network device
3006 * @sizeof_priv: size of private data to allocate space for
3007 * @name: device name format string
3008 * @setup: callback to initialize device
3009 *
3010 * Allocates a struct net_device with private data area for driver use
3011 * and performs basic initialization.
3012 */
3013 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3014 void (*setup)(struct net_device *))
3015 {
3016 void *p;
3017 struct net_device *dev;
3018 int alloc_size;
3019
3020 /* ensure 32-byte alignment of both the device and private area */
3021 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3022 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3023
3024 p = kmalloc(alloc_size, GFP_KERNEL);
3025 if (!p) {
3026 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3027 return NULL;
3028 }
3029 memset(p, 0, alloc_size);
3030
3031 dev = (struct net_device *)
3032 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3033 dev->padded = (char *)dev - (char *)p;
3034
3035 if (sizeof_priv)
3036 dev->priv = netdev_priv(dev);
3037
3038 setup(dev);
3039 strcpy(dev->name, name);
3040 return dev;
3041 }
3042 EXPORT_SYMBOL(alloc_netdev);
3043
3044 /**
3045 * free_netdev - free network device
3046 * @dev: device
3047 *
3048 * This function does the last stage of destroying an allocated device
3049 * interface. The reference to the device object is released.
3050 * If this is the last reference then it will be freed.
3051 */
3052 void free_netdev(struct net_device *dev)
3053 {
3054 #ifdef CONFIG_SYSFS
3055 /* Compatiablity with error handling in drivers */
3056 if (dev->reg_state == NETREG_UNINITIALIZED) {
3057 kfree((char *)dev - dev->padded);
3058 return;
3059 }
3060
3061 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3062 dev->reg_state = NETREG_RELEASED;
3063
3064 /* will free via class release */
3065 class_device_put(&dev->class_dev);
3066 #else
3067 kfree((char *)dev - dev->padded);
3068 #endif
3069 }
3070
3071 /* Synchronize with packet receive processing. */
3072 void synchronize_net(void)
3073 {
3074 might_sleep();
3075 synchronize_kernel();
3076 }
3077
3078 /**
3079 * unregister_netdevice - remove device from the kernel
3080 * @dev: device
3081 *
3082 * This function shuts down a device interface and removes it
3083 * from the kernel tables. On success 0 is returned, on a failure
3084 * a negative errno code is returned.
3085 *
3086 * Callers must hold the rtnl semaphore. You may want
3087 * unregister_netdev() instead of this.
3088 */
3089
3090 int unregister_netdevice(struct net_device *dev)
3091 {
3092 struct net_device *d, **dp;
3093
3094 BUG_ON(dev_boot_phase);
3095 ASSERT_RTNL();
3096
3097 /* Some devices call without registering for initialization unwind. */
3098 if (dev->reg_state == NETREG_UNINITIALIZED) {
3099 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3100 "was registered\n", dev->name, dev);
3101 return -ENODEV;
3102 }
3103
3104 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3105
3106 /* If device is running, close it first. */
3107 if (dev->flags & IFF_UP)
3108 dev_close(dev);
3109
3110 /* And unlink it from device chain. */
3111 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3112 if (d == dev) {
3113 write_lock_bh(&dev_base_lock);
3114 hlist_del(&dev->name_hlist);
3115 hlist_del(&dev->index_hlist);
3116 if (dev_tail == &dev->next)
3117 dev_tail = dp;
3118 *dp = d->next;
3119 write_unlock_bh(&dev_base_lock);
3120 break;
3121 }
3122 }
3123 if (!d) {
3124 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3125 dev->name);
3126 return -ENODEV;
3127 }
3128
3129 dev->reg_state = NETREG_UNREGISTERING;
3130
3131 synchronize_net();
3132
3133 /* Shutdown queueing discipline. */
3134 dev_shutdown(dev);
3135
3136
3137 /* Notify protocols, that we are about to destroy
3138 this device. They should clean all the things.
3139 */
3140 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3141
3142 /*
3143 * Flush the multicast chain
3144 */
3145 dev_mc_discard(dev);
3146
3147 if (dev->uninit)
3148 dev->uninit(dev);
3149
3150 /* Notifier chain MUST detach us from master device. */
3151 BUG_TRAP(!dev->master);
3152
3153 free_divert_blk(dev);
3154
3155 /* Finish processing unregister after unlock */
3156 net_set_todo(dev);
3157
3158 synchronize_net();
3159
3160 dev_put(dev);
3161 return 0;
3162 }
3163
3164 /**
3165 * unregister_netdev - remove device from the kernel
3166 * @dev: device
3167 *
3168 * This function shuts down a device interface and removes it
3169 * from the kernel tables. On success 0 is returned, on a failure
3170 * a negative errno code is returned.
3171 *
3172 * This is just a wrapper for unregister_netdevice that takes
3173 * the rtnl semaphore. In general you want to use this and not
3174 * unregister_netdevice.
3175 */
3176 void unregister_netdev(struct net_device *dev)
3177 {
3178 rtnl_lock();
3179 unregister_netdevice(dev);
3180 rtnl_unlock();
3181 }
3182
3183 EXPORT_SYMBOL(unregister_netdev);
3184
3185 #ifdef CONFIG_HOTPLUG_CPU
3186 static int dev_cpu_callback(struct notifier_block *nfb,
3187 unsigned long action,
3188 void *ocpu)
3189 {
3190 struct sk_buff **list_skb;
3191 struct net_device **list_net;
3192 struct sk_buff *skb;
3193 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3194 struct softnet_data *sd, *oldsd;
3195
3196 if (action != CPU_DEAD)
3197 return NOTIFY_OK;
3198
3199 local_irq_disable();
3200 cpu = smp_processor_id();
3201 sd = &per_cpu(softnet_data, cpu);
3202 oldsd = &per_cpu(softnet_data, oldcpu);
3203
3204 /* Find end of our completion_queue. */
3205 list_skb = &sd->completion_queue;
3206 while (*list_skb)
3207 list_skb = &(*list_skb)->next;
3208 /* Append completion queue from offline CPU. */
3209 *list_skb = oldsd->completion_queue;
3210 oldsd->completion_queue = NULL;
3211
3212 /* Find end of our output_queue. */
3213 list_net = &sd->output_queue;
3214 while (*list_net)
3215 list_net = &(*list_net)->next_sched;
3216 /* Append output queue from offline CPU. */
3217 *list_net = oldsd->output_queue;
3218 oldsd->output_queue = NULL;
3219
3220 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3221 local_irq_enable();
3222
3223 /* Process offline CPU's input_pkt_queue */
3224 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3225 netif_rx(skb);
3226
3227 return NOTIFY_OK;
3228 }
3229 #endif /* CONFIG_HOTPLUG_CPU */
3230
3231
3232 /*
3233 * Initialize the DEV module. At boot time this walks the device list and
3234 * unhooks any devices that fail to initialise (normally hardware not
3235 * present) and leaves us with a valid list of present and active devices.
3236 *
3237 */
3238
3239 /*
3240 * This is called single threaded during boot, so no need
3241 * to take the rtnl semaphore.
3242 */
3243 static int __init net_dev_init(void)
3244 {
3245 int i, rc = -ENOMEM;
3246
3247 BUG_ON(!dev_boot_phase);
3248
3249 net_random_init();
3250
3251 if (dev_proc_init())
3252 goto out;
3253
3254 if (netdev_sysfs_init())
3255 goto out;
3256
3257 INIT_LIST_HEAD(&ptype_all);
3258 for (i = 0; i < 16; i++)
3259 INIT_LIST_HEAD(&ptype_base[i]);
3260
3261 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3262 INIT_HLIST_HEAD(&dev_name_head[i]);
3263
3264 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3265 INIT_HLIST_HEAD(&dev_index_head[i]);
3266
3267 /*
3268 * Initialise the packet receive queues.
3269 */
3270
3271 for (i = 0; i < NR_CPUS; i++) {
3272 struct softnet_data *queue;
3273
3274 queue = &per_cpu(softnet_data, i);
3275 skb_queue_head_init(&queue->input_pkt_queue);
3276 queue->throttle = 0;
3277 queue->cng_level = 0;
3278 queue->avg_blog = 10; /* arbitrary non-zero */
3279 queue->completion_queue = NULL;
3280 INIT_LIST_HEAD(&queue->poll_list);
3281 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3282 queue->backlog_dev.weight = weight_p;
3283 queue->backlog_dev.poll = process_backlog;
3284 atomic_set(&queue->backlog_dev.refcnt, 1);
3285 }
3286
3287 #ifdef OFFLINE_SAMPLE
3288 samp_timer.expires = jiffies + (10 * HZ);
3289 add_timer(&samp_timer);
3290 #endif
3291
3292 dev_boot_phase = 0;
3293
3294 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3295 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3296
3297 hotcpu_notifier(dev_cpu_callback, 0);
3298 dst_init();
3299 dev_mcast_init();
3300 rc = 0;
3301 out:
3302 return rc;
3303 }
3304
3305 subsys_initcall(net_dev_init);
3306
3307 EXPORT_SYMBOL(__dev_get_by_index);
3308 EXPORT_SYMBOL(__dev_get_by_name);
3309 EXPORT_SYMBOL(__dev_remove_pack);
3310 EXPORT_SYMBOL(__skb_linearize);
3311 EXPORT_SYMBOL(dev_add_pack);
3312 EXPORT_SYMBOL(dev_alloc_name);
3313 EXPORT_SYMBOL(dev_close);
3314 EXPORT_SYMBOL(dev_get_by_flags);
3315 EXPORT_SYMBOL(dev_get_by_index);
3316 EXPORT_SYMBOL(dev_get_by_name);
3317 EXPORT_SYMBOL(dev_ioctl);
3318 EXPORT_SYMBOL(dev_open);
3319 EXPORT_SYMBOL(dev_queue_xmit);
3320 EXPORT_SYMBOL(dev_remove_pack);
3321 EXPORT_SYMBOL(dev_set_allmulti);
3322 EXPORT_SYMBOL(dev_set_promiscuity);
3323 EXPORT_SYMBOL(dev_change_flags);
3324 EXPORT_SYMBOL(dev_set_mtu);
3325 EXPORT_SYMBOL(free_netdev);
3326 EXPORT_SYMBOL(netdev_boot_setup_check);
3327 EXPORT_SYMBOL(netdev_set_master);
3328 EXPORT_SYMBOL(netdev_state_change);
3329 EXPORT_SYMBOL(netif_receive_skb);
3330 EXPORT_SYMBOL(netif_rx);
3331 EXPORT_SYMBOL(register_gifconf);
3332 EXPORT_SYMBOL(register_netdevice);
3333 EXPORT_SYMBOL(register_netdevice_notifier);
3334 EXPORT_SYMBOL(skb_checksum_help);
3335 EXPORT_SYMBOL(synchronize_net);
3336 EXPORT_SYMBOL(unregister_netdevice);
3337 EXPORT_SYMBOL(unregister_netdevice_notifier);
3338 EXPORT_SYMBOL(net_enable_timestamp);
3339 EXPORT_SYMBOL(net_disable_timestamp);
3340
3341 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3342 EXPORT_SYMBOL(br_handle_frame_hook);
3343 #endif
3344
3345 #ifdef CONFIG_KMOD
3346 EXPORT_SYMBOL(dev_load);
3347 #endif
3348
3349 EXPORT_PER_CPU_SYMBOL(softnet_data);
3350
|
This page was automatically generated by the
LXR engine.
|