[ARM] 3349/1: ixp4xx exp bus defines
[linux-2.6/libata-dev.git] / net / core / dev.c
blob08dec6eb922bc6c971f952c4c396bf23e7c7d292
1 /*
2 * NET3 Protocol independent device support routines.
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.
9 * Derived from the non IP parts of dev.c 1.0.19
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
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>
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
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/config.h>
80 #include <linux/cpu.h>
81 #include <linux/types.h>
82 #include <linux/kernel.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
86 #include <linux/mm.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/sock.h>
97 #include <linux/rtnetlink.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <linux/stat.h>
101 #include <linux/if_bridge.h>
102 #include <linux/divert.h>
103 #include <net/dst.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/kmod.h>
109 #include <linux/module.h>
110 #include <linux/kallsyms.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <linux/wireless.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
119 * The list of packet types we will receive (as opposed to discard)
120 * and the routines to invoke.
122 * Why 16. Because with 16 the only overlap we get on a hash of the
123 * low nibble of the protocol value is RARP/SNAP/X.25.
125 * NOTE: That is no longer true with the addition of VLAN tags. Not
126 * sure which should go first, but I bet it won't make much
127 * difference if we are running VLANs. The good news is that
128 * this protocol won't be in the list unless compiled in, so
129 * the average user (w/out VLANs) will not be adversly affected.
130 * --BLG
132 * 0800 IP
133 * 8100 802.1Q VLAN
134 * 0001 802.3
135 * 0002 AX.25
136 * 0004 802.2
137 * 8035 RARP
138 * 0005 SNAP
139 * 0805 X.25
140 * 0806 ARP
141 * 8137 IPX
142 * 0009 Localtalk
143 * 86DD IPv6
146 static DEFINE_SPINLOCK(ptype_lock);
147 static struct list_head ptype_base[16]; /* 16 way hashed list */
148 static struct list_head ptype_all; /* Taps */
151 * The @dev_base list is protected by @dev_base_lock and the rtln
152 * semaphore.
154 * Pure readers hold dev_base_lock for reading.
156 * Writers must hold the rtnl semaphore while they loop through the
157 * dev_base list, and hold dev_base_lock for writing when they do the
158 * actual updates. This allows pure readers to access the list even
159 * while a writer is preparing to update it.
161 * To put it another way, dev_base_lock is held for writing only to
162 * protect against pure readers; the rtnl semaphore provides the
163 * protection against other writers.
165 * See, for example usages, register_netdevice() and
166 * unregister_netdevice(), which must be called with the rtnl
167 * semaphore held.
169 struct net_device *dev_base;
170 static struct net_device **dev_tail = &dev_base;
171 DEFINE_RWLOCK(dev_base_lock);
173 EXPORT_SYMBOL(dev_base);
174 EXPORT_SYMBOL(dev_base_lock);
176 #define NETDEV_HASHBITS 8
177 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
178 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
180 static inline struct hlist_head *dev_name_hash(const char *name)
182 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
183 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
186 static inline struct hlist_head *dev_index_hash(int ifindex)
188 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
192 * Our notifier list
195 static struct notifier_block *netdev_chain;
198 * Device drivers call our routines to queue packets here. We empty the
199 * queue in the local softnet handler.
201 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
203 #ifdef CONFIG_SYSFS
204 extern int netdev_sysfs_init(void);
205 extern int netdev_register_sysfs(struct net_device *);
206 extern void netdev_unregister_sysfs(struct net_device *);
207 #else
208 #define netdev_sysfs_init() (0)
209 #define netdev_register_sysfs(dev) (0)
210 #define netdev_unregister_sysfs(dev) do { } while(0)
211 #endif
214 /*******************************************************************************
216 Protocol management and registration routines
218 *******************************************************************************/
221 * For efficiency
224 int netdev_nit;
227 * Add a protocol ID to the list. Now that the input handler is
228 * smarter we can dispense with all the messy stuff that used to be
229 * here.
231 * BEWARE!!! Protocol handlers, mangling input packets,
232 * MUST BE last in hash buckets and checking protocol handlers
233 * MUST start from promiscuous ptype_all chain in net_bh.
234 * It is true now, do not change it.
235 * Explanation follows: if protocol handler, mangling packet, will
236 * be the first on list, it is not able to sense, that packet
237 * is cloned and should be copied-on-write, so that it will
238 * change it and subsequent readers will get broken packet.
239 * --ANK (980803)
243 * dev_add_pack - add packet handler
244 * @pt: packet type declaration
246 * Add a protocol handler to the networking stack. The passed &packet_type
247 * is linked into kernel lists and may not be freed until it has been
248 * removed from the kernel lists.
250 * This call does not sleep therefore it can not
251 * guarantee all CPU's that are in middle of receiving packets
252 * will see the new packet type (until the next received packet).
255 void dev_add_pack(struct packet_type *pt)
257 int hash;
259 spin_lock_bh(&ptype_lock);
260 if (pt->type == htons(ETH_P_ALL)) {
261 netdev_nit++;
262 list_add_rcu(&pt->list, &ptype_all);
263 } else {
264 hash = ntohs(pt->type) & 15;
265 list_add_rcu(&pt->list, &ptype_base[hash]);
267 spin_unlock_bh(&ptype_lock);
271 * __dev_remove_pack - remove packet handler
272 * @pt: packet type declaration
274 * Remove a protocol handler that was previously added to the kernel
275 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
276 * from the kernel lists and can be freed or reused once this function
277 * returns.
279 * The packet type might still be in use by receivers
280 * and must not be freed until after all the CPU's have gone
281 * through a quiescent state.
283 void __dev_remove_pack(struct packet_type *pt)
285 struct list_head *head;
286 struct packet_type *pt1;
288 spin_lock_bh(&ptype_lock);
290 if (pt->type == htons(ETH_P_ALL)) {
291 netdev_nit--;
292 head = &ptype_all;
293 } else
294 head = &ptype_base[ntohs(pt->type) & 15];
296 list_for_each_entry(pt1, head, list) {
297 if (pt == pt1) {
298 list_del_rcu(&pt->list);
299 goto out;
303 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
304 out:
305 spin_unlock_bh(&ptype_lock);
308 * dev_remove_pack - remove packet handler
309 * @pt: packet type declaration
311 * Remove a protocol handler that was previously added to the kernel
312 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
313 * from the kernel lists and can be freed or reused once this function
314 * returns.
316 * This call sleeps to guarantee that no CPU is looking at the packet
317 * type after return.
319 void dev_remove_pack(struct packet_type *pt)
321 __dev_remove_pack(pt);
323 synchronize_net();
326 /******************************************************************************
328 Device Boot-time Settings Routines
330 *******************************************************************************/
332 /* Boot time configuration table */
333 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
336 * netdev_boot_setup_add - add new setup entry
337 * @name: name of the device
338 * @map: configured settings for the device
340 * Adds new setup entry to the dev_boot_setup list. The function
341 * returns 0 on error and 1 on success. This is a generic routine to
342 * all netdevices.
344 static int netdev_boot_setup_add(char *name, struct ifmap *map)
346 struct netdev_boot_setup *s;
347 int i;
349 s = dev_boot_setup;
350 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
351 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
352 memset(s[i].name, 0, sizeof(s[i].name));
353 strcpy(s[i].name, name);
354 memcpy(&s[i].map, map, sizeof(s[i].map));
355 break;
359 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
363 * netdev_boot_setup_check - check boot time settings
364 * @dev: the netdevice
366 * Check boot time settings for the device.
367 * The found settings are set for the device to be used
368 * later in the device probing.
369 * Returns 0 if no settings found, 1 if they are.
371 int netdev_boot_setup_check(struct net_device *dev)
373 struct netdev_boot_setup *s = dev_boot_setup;
374 int i;
376 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
377 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
378 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
379 dev->irq = s[i].map.irq;
380 dev->base_addr = s[i].map.base_addr;
381 dev->mem_start = s[i].map.mem_start;
382 dev->mem_end = s[i].map.mem_end;
383 return 1;
386 return 0;
391 * netdev_boot_base - get address from boot time settings
392 * @prefix: prefix for network device
393 * @unit: id for network device
395 * Check boot time settings for the base address of device.
396 * The found settings are set for the device to be used
397 * later in the device probing.
398 * Returns 0 if no settings found.
400 unsigned long netdev_boot_base(const char *prefix, int unit)
402 const struct netdev_boot_setup *s = dev_boot_setup;
403 char name[IFNAMSIZ];
404 int i;
406 sprintf(name, "%s%d", prefix, unit);
409 * If device already registered then return base of 1
410 * to indicate not to probe for this interface
412 if (__dev_get_by_name(name))
413 return 1;
415 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
416 if (!strcmp(name, s[i].name))
417 return s[i].map.base_addr;
418 return 0;
422 * Saves at boot time configured settings for any netdevice.
424 int __init netdev_boot_setup(char *str)
426 int ints[5];
427 struct ifmap map;
429 str = get_options(str, ARRAY_SIZE(ints), ints);
430 if (!str || !*str)
431 return 0;
433 /* Save settings */
434 memset(&map, 0, sizeof(map));
435 if (ints[0] > 0)
436 map.irq = ints[1];
437 if (ints[0] > 1)
438 map.base_addr = ints[2];
439 if (ints[0] > 2)
440 map.mem_start = ints[3];
441 if (ints[0] > 3)
442 map.mem_end = ints[4];
444 /* Add new entry to the list */
445 return netdev_boot_setup_add(str, &map);
448 __setup("netdev=", netdev_boot_setup);
450 /*******************************************************************************
452 Device Interface Subroutines
454 *******************************************************************************/
457 * __dev_get_by_name - find a device by its name
458 * @name: name to find
460 * Find an interface by name. Must be called under RTNL semaphore
461 * or @dev_base_lock. If the name is found a pointer to the device
462 * is returned. If the name is not found then %NULL is returned. The
463 * reference counters are not incremented so the caller must be
464 * careful with locks.
467 struct net_device *__dev_get_by_name(const char *name)
469 struct hlist_node *p;
471 hlist_for_each(p, dev_name_hash(name)) {
472 struct net_device *dev
473 = hlist_entry(p, struct net_device, name_hlist);
474 if (!strncmp(dev->name, name, IFNAMSIZ))
475 return dev;
477 return NULL;
481 * dev_get_by_name - find a device by its name
482 * @name: name to find
484 * Find an interface by name. This can be called from any
485 * context and does its own locking. The returned handle has
486 * the usage count incremented and the caller must use dev_put() to
487 * release it when it is no longer needed. %NULL is returned if no
488 * matching device is found.
491 struct net_device *dev_get_by_name(const char *name)
493 struct net_device *dev;
495 read_lock(&dev_base_lock);
496 dev = __dev_get_by_name(name);
497 if (dev)
498 dev_hold(dev);
499 read_unlock(&dev_base_lock);
500 return dev;
504 * __dev_get_by_index - find a device by its ifindex
505 * @ifindex: index of device
507 * Search for an interface by index. Returns %NULL if the device
508 * is not found or a pointer to the device. The device has not
509 * had its reference counter increased so the caller must be careful
510 * about locking. The caller must hold either the RTNL semaphore
511 * or @dev_base_lock.
514 struct net_device *__dev_get_by_index(int ifindex)
516 struct hlist_node *p;
518 hlist_for_each(p, dev_index_hash(ifindex)) {
519 struct net_device *dev
520 = hlist_entry(p, struct net_device, index_hlist);
521 if (dev->ifindex == ifindex)
522 return dev;
524 return NULL;
529 * dev_get_by_index - find a device by its ifindex
530 * @ifindex: index of device
532 * Search for an interface by index. Returns NULL if the device
533 * is not found or a pointer to the device. The device returned has
534 * had a reference added and the pointer is safe until the user calls
535 * dev_put to indicate they have finished with it.
538 struct net_device *dev_get_by_index(int ifindex)
540 struct net_device *dev;
542 read_lock(&dev_base_lock);
543 dev = __dev_get_by_index(ifindex);
544 if (dev)
545 dev_hold(dev);
546 read_unlock(&dev_base_lock);
547 return dev;
551 * dev_getbyhwaddr - find a device by its hardware address
552 * @type: media type of device
553 * @ha: hardware address
555 * Search for an interface by MAC address. Returns NULL if the device
556 * is not found or a pointer to the device. The caller must hold the
557 * rtnl semaphore. The returned device has not had its ref count increased
558 * and the caller must therefore be careful about locking
560 * BUGS:
561 * If the API was consistent this would be __dev_get_by_hwaddr
564 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
566 struct net_device *dev;
568 ASSERT_RTNL();
570 for (dev = dev_base; dev; dev = dev->next)
571 if (dev->type == type &&
572 !memcmp(dev->dev_addr, ha, dev->addr_len))
573 break;
574 return dev;
577 EXPORT_SYMBOL(dev_getbyhwaddr);
579 struct net_device *dev_getfirstbyhwtype(unsigned short type)
581 struct net_device *dev;
583 rtnl_lock();
584 for (dev = dev_base; dev; dev = dev->next) {
585 if (dev->type == type) {
586 dev_hold(dev);
587 break;
590 rtnl_unlock();
591 return dev;
594 EXPORT_SYMBOL(dev_getfirstbyhwtype);
597 * dev_get_by_flags - find any device with given flags
598 * @if_flags: IFF_* values
599 * @mask: bitmask of bits in if_flags to check
601 * Search for any interface with the given flags. Returns NULL if a device
602 * is not found or a pointer to the device. The device returned has
603 * had a reference added and the pointer is safe until the user calls
604 * dev_put to indicate they have finished with it.
607 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
609 struct net_device *dev;
611 read_lock(&dev_base_lock);
612 for (dev = dev_base; dev != NULL; dev = dev->next) {
613 if (((dev->flags ^ if_flags) & mask) == 0) {
614 dev_hold(dev);
615 break;
618 read_unlock(&dev_base_lock);
619 return dev;
623 * dev_valid_name - check if name is okay for network device
624 * @name: name string
626 * Network device names need to be valid file names to
627 * to allow sysfs to work
629 int dev_valid_name(const char *name)
631 return !(*name == '\0'
632 || !strcmp(name, ".")
633 || !strcmp(name, "..")
634 || strchr(name, '/'));
638 * dev_alloc_name - allocate a name for a device
639 * @dev: device
640 * @name: name format string
642 * Passed a format string - eg "lt%d" it will try and find a suitable
643 * id. Not efficient for many devices, not called a lot. The caller
644 * must hold the dev_base or rtnl lock while allocating the name and
645 * adding the device in order to avoid duplicates. Returns the number
646 * of the unit assigned or a negative errno code.
649 int dev_alloc_name(struct net_device *dev, const char *name)
651 int i = 0;
652 char buf[IFNAMSIZ];
653 const char *p;
654 const int max_netdevices = 8*PAGE_SIZE;
655 long *inuse;
656 struct net_device *d;
658 p = strnchr(name, IFNAMSIZ-1, '%');
659 if (p) {
661 * Verify the string as this thing may have come from
662 * the user. There must be either one "%d" and no other "%"
663 * characters.
665 if (p[1] != 'd' || strchr(p + 2, '%'))
666 return -EINVAL;
668 /* Use one page as a bit array of possible slots */
669 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
670 if (!inuse)
671 return -ENOMEM;
673 for (d = dev_base; d; d = d->next) {
674 if (!sscanf(d->name, name, &i))
675 continue;
676 if (i < 0 || i >= max_netdevices)
677 continue;
679 /* avoid cases where sscanf is not exact inverse of printf */
680 snprintf(buf, sizeof(buf), name, i);
681 if (!strncmp(buf, d->name, IFNAMSIZ))
682 set_bit(i, inuse);
685 i = find_first_zero_bit(inuse, max_netdevices);
686 free_page((unsigned long) inuse);
689 snprintf(buf, sizeof(buf), name, i);
690 if (!__dev_get_by_name(buf)) {
691 strlcpy(dev->name, buf, IFNAMSIZ);
692 return i;
695 /* It is possible to run out of possible slots
696 * when the name is long and there isn't enough space left
697 * for the digits, or if all bits are used.
699 return -ENFILE;
704 * dev_change_name - change name of a device
705 * @dev: device
706 * @newname: name (or format string) must be at least IFNAMSIZ
708 * Change name of a device, can pass format strings "eth%d".
709 * for wildcarding.
711 int dev_change_name(struct net_device *dev, char *newname)
713 int err = 0;
715 ASSERT_RTNL();
717 if (dev->flags & IFF_UP)
718 return -EBUSY;
720 if (!dev_valid_name(newname))
721 return -EINVAL;
723 if (strchr(newname, '%')) {
724 err = dev_alloc_name(dev, newname);
725 if (err < 0)
726 return err;
727 strcpy(newname, dev->name);
729 else if (__dev_get_by_name(newname))
730 return -EEXIST;
731 else
732 strlcpy(dev->name, newname, IFNAMSIZ);
734 err = class_device_rename(&dev->class_dev, dev->name);
735 if (!err) {
736 hlist_del(&dev->name_hlist);
737 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
738 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
741 return err;
745 * netdev_features_change - device changes fatures
746 * @dev: device to cause notification
748 * Called to indicate a device has changed features.
750 void netdev_features_change(struct net_device *dev)
752 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
754 EXPORT_SYMBOL(netdev_features_change);
757 * netdev_state_change - device changes state
758 * @dev: device to cause notification
760 * Called to indicate a device has changed state. This function calls
761 * the notifier chains for netdev_chain and sends a NEWLINK message
762 * to the routing socket.
764 void netdev_state_change(struct net_device *dev)
766 if (dev->flags & IFF_UP) {
767 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
768 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
773 * dev_load - load a network module
774 * @name: name of interface
776 * If a network interface is not present and the process has suitable
777 * privileges this function loads the module. If module loading is not
778 * available in this kernel then it becomes a nop.
781 void dev_load(const char *name)
783 struct net_device *dev;
785 read_lock(&dev_base_lock);
786 dev = __dev_get_by_name(name);
787 read_unlock(&dev_base_lock);
789 if (!dev && capable(CAP_SYS_MODULE))
790 request_module("%s", name);
793 static int default_rebuild_header(struct sk_buff *skb)
795 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
796 skb->dev ? skb->dev->name : "NULL!!!");
797 kfree_skb(skb);
798 return 1;
803 * dev_open - prepare an interface for use.
804 * @dev: device to open
806 * Takes a device from down to up state. The device's private open
807 * function is invoked and then the multicast lists are loaded. Finally
808 * the device is moved into the up state and a %NETDEV_UP message is
809 * sent to the netdev notifier chain.
811 * Calling this function on an active interface is a nop. On a failure
812 * a negative errno code is returned.
814 int dev_open(struct net_device *dev)
816 int ret = 0;
819 * Is it already up?
822 if (dev->flags & IFF_UP)
823 return 0;
826 * Is it even present?
828 if (!netif_device_present(dev))
829 return -ENODEV;
832 * Call device private open method
834 set_bit(__LINK_STATE_START, &dev->state);
835 if (dev->open) {
836 ret = dev->open(dev);
837 if (ret)
838 clear_bit(__LINK_STATE_START, &dev->state);
842 * If it went open OK then:
845 if (!ret) {
847 * Set the flags.
849 dev->flags |= IFF_UP;
852 * Initialize multicasting status
854 dev_mc_upload(dev);
857 * Wakeup transmit queue engine
859 dev_activate(dev);
862 * ... and announce new interface.
864 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
866 return ret;
870 * dev_close - shutdown an interface.
871 * @dev: device to shutdown
873 * This function moves an active device into down state. A
874 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
875 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
876 * chain.
878 int dev_close(struct net_device *dev)
880 if (!(dev->flags & IFF_UP))
881 return 0;
884 * Tell people we are going down, so that they can
885 * prepare to death, when device is still operating.
887 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
889 dev_deactivate(dev);
891 clear_bit(__LINK_STATE_START, &dev->state);
893 /* Synchronize to scheduled poll. We cannot touch poll list,
894 * it can be even on different cpu. So just clear netif_running(),
895 * and wait when poll really will happen. Actually, the best place
896 * for this is inside dev->stop() after device stopped its irq
897 * engine, but this requires more changes in devices. */
899 smp_mb__after_clear_bit(); /* Commit netif_running(). */
900 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
901 /* No hurry. */
902 msleep(1);
906 * Call the device specific close. This cannot fail.
907 * Only if device is UP
909 * We allow it to be called even after a DETACH hot-plug
910 * event.
912 if (dev->stop)
913 dev->stop(dev);
916 * Device is now down.
919 dev->flags &= ~IFF_UP;
922 * Tell people we are down
924 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
926 return 0;
931 * Device change register/unregister. These are not inline or static
932 * as we export them to the world.
936 * register_netdevice_notifier - register a network notifier block
937 * @nb: notifier
939 * Register a notifier to be called when network device events occur.
940 * The notifier passed is linked into the kernel structures and must
941 * not be reused until it has been unregistered. A negative errno code
942 * is returned on a failure.
944 * When registered all registration and up events are replayed
945 * to the new notifier to allow device to have a race free
946 * view of the network device list.
949 int register_netdevice_notifier(struct notifier_block *nb)
951 struct net_device *dev;
952 int err;
954 rtnl_lock();
955 err = notifier_chain_register(&netdev_chain, nb);
956 if (!err) {
957 for (dev = dev_base; dev; dev = dev->next) {
958 nb->notifier_call(nb, NETDEV_REGISTER, dev);
960 if (dev->flags & IFF_UP)
961 nb->notifier_call(nb, NETDEV_UP, dev);
964 rtnl_unlock();
965 return err;
969 * unregister_netdevice_notifier - unregister a network notifier block
970 * @nb: notifier
972 * Unregister a notifier previously registered by
973 * register_netdevice_notifier(). The notifier is unlinked into the
974 * kernel structures and may then be reused. A negative errno code
975 * is returned on a failure.
978 int unregister_netdevice_notifier(struct notifier_block *nb)
980 return notifier_chain_unregister(&netdev_chain, nb);
984 * call_netdevice_notifiers - call all network notifier blocks
985 * @val: value passed unmodified to notifier function
986 * @v: pointer passed unmodified to notifier function
988 * Call all network notifier blocks. Parameters and return value
989 * are as for notifier_call_chain().
992 int call_netdevice_notifiers(unsigned long val, void *v)
994 return notifier_call_chain(&netdev_chain, val, v);
997 /* When > 0 there are consumers of rx skb time stamps */
998 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1000 void net_enable_timestamp(void)
1002 atomic_inc(&netstamp_needed);
1005 void net_disable_timestamp(void)
1007 atomic_dec(&netstamp_needed);
1010 void __net_timestamp(struct sk_buff *skb)
1012 struct timeval tv;
1014 do_gettimeofday(&tv);
1015 skb_set_timestamp(skb, &tv);
1017 EXPORT_SYMBOL(__net_timestamp);
1019 static inline void net_timestamp(struct sk_buff *skb)
1021 if (atomic_read(&netstamp_needed))
1022 __net_timestamp(skb);
1023 else {
1024 skb->tstamp.off_sec = 0;
1025 skb->tstamp.off_usec = 0;
1030 * Support routine. Sends outgoing frames to any network
1031 * taps currently in use.
1034 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1036 struct packet_type *ptype;
1038 net_timestamp(skb);
1040 rcu_read_lock();
1041 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1042 /* Never send packets back to the socket
1043 * they originated from - MvS (miquels@drinkel.ow.org)
1045 if ((ptype->dev == dev || !ptype->dev) &&
1046 (ptype->af_packet_priv == NULL ||
1047 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1048 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1049 if (!skb2)
1050 break;
1052 /* skb->nh should be correctly
1053 set by sender, so that the second statement is
1054 just protection against buggy protocols.
1056 skb2->mac.raw = skb2->data;
1058 if (skb2->nh.raw < skb2->data ||
1059 skb2->nh.raw > skb2->tail) {
1060 if (net_ratelimit())
1061 printk(KERN_CRIT "protocol %04x is "
1062 "buggy, dev %s\n",
1063 skb2->protocol, dev->name);
1064 skb2->nh.raw = skb2->data;
1067 skb2->h.raw = skb2->nh.raw;
1068 skb2->pkt_type = PACKET_OUTGOING;
1069 ptype->func(skb2, skb->dev, ptype, skb->dev);
1072 rcu_read_unlock();
1076 * Invalidate hardware checksum when packet is to be mangled, and
1077 * complete checksum manually on outgoing path.
1079 int skb_checksum_help(struct sk_buff *skb, int inward)
1081 unsigned int csum;
1082 int ret = 0, offset = skb->h.raw - skb->data;
1084 if (inward) {
1085 skb->ip_summed = CHECKSUM_NONE;
1086 goto out;
1089 if (skb_cloned(skb)) {
1090 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1091 if (ret)
1092 goto out;
1095 BUG_ON(offset > (int)skb->len);
1096 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1098 offset = skb->tail - skb->h.raw;
1099 BUG_ON(offset <= 0);
1100 BUG_ON(skb->csum + 2 > offset);
1102 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1103 skb->ip_summed = CHECKSUM_NONE;
1104 out:
1105 return ret;
1108 /* Take action when hardware reception checksum errors are detected. */
1109 #ifdef CONFIG_BUG
1110 void netdev_rx_csum_fault(struct net_device *dev)
1112 if (net_ratelimit()) {
1113 printk(KERN_ERR "%s: hw csum failure.\n",
1114 dev ? dev->name : "<unknown>");
1115 dump_stack();
1118 EXPORT_SYMBOL(netdev_rx_csum_fault);
1119 #endif
1121 #ifdef CONFIG_HIGHMEM
1122 /* Actually, we should eliminate this check as soon as we know, that:
1123 * 1. IOMMU is present and allows to map all the memory.
1124 * 2. No high memory really exists on this machine.
1127 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1129 int i;
1131 if (dev->features & NETIF_F_HIGHDMA)
1132 return 0;
1134 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1135 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1136 return 1;
1138 return 0;
1140 #else
1141 #define illegal_highdma(dev, skb) (0)
1142 #endif
1144 /* Keep head the same: replace data */
1145 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1147 unsigned int size;
1148 u8 *data;
1149 long offset;
1150 struct skb_shared_info *ninfo;
1151 int headerlen = skb->data - skb->head;
1152 int expand = (skb->tail + skb->data_len) - skb->end;
1154 if (skb_shared(skb))
1155 BUG();
1157 if (expand <= 0)
1158 expand = 0;
1160 size = skb->end - skb->head + expand;
1161 size = SKB_DATA_ALIGN(size);
1162 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1163 if (!data)
1164 return -ENOMEM;
1166 /* Copy entire thing */
1167 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1168 BUG();
1170 /* Set up shinfo */
1171 ninfo = (struct skb_shared_info*)(data + size);
1172 atomic_set(&ninfo->dataref, 1);
1173 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1174 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1175 ninfo->nr_frags = 0;
1176 ninfo->frag_list = NULL;
1178 /* Offset between the two in bytes */
1179 offset = data - skb->head;
1181 /* Free old data. */
1182 skb_release_data(skb);
1184 skb->head = data;
1185 skb->end = data + size;
1187 /* Set up new pointers */
1188 skb->h.raw += offset;
1189 skb->nh.raw += offset;
1190 skb->mac.raw += offset;
1191 skb->tail += offset;
1192 skb->data += offset;
1194 /* We are no longer a clone, even if we were. */
1195 skb->cloned = 0;
1197 skb->tail += skb->data_len;
1198 skb->data_len = 0;
1199 return 0;
1202 #define HARD_TX_LOCK(dev, cpu) { \
1203 if ((dev->features & NETIF_F_LLTX) == 0) { \
1204 spin_lock(&dev->xmit_lock); \
1205 dev->xmit_lock_owner = cpu; \
1209 #define HARD_TX_UNLOCK(dev) { \
1210 if ((dev->features & NETIF_F_LLTX) == 0) { \
1211 dev->xmit_lock_owner = -1; \
1212 spin_unlock(&dev->xmit_lock); \
1217 * dev_queue_xmit - transmit a buffer
1218 * @skb: buffer to transmit
1220 * Queue a buffer for transmission to a network device. The caller must
1221 * have set the device and priority and built the buffer before calling
1222 * this function. The function can be called from an interrupt.
1224 * A negative errno code is returned on a failure. A success does not
1225 * guarantee the frame will be transmitted as it may be dropped due
1226 * to congestion or traffic shaping.
1228 * -----------------------------------------------------------------------------------
1229 * I notice this method can also return errors from the queue disciplines,
1230 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1231 * be positive.
1233 * Regardless of the return value, the skb is consumed, so it is currently
1234 * difficult to retry a send to this method. (You can bump the ref count
1235 * before sending to hold a reference for retry if you are careful.)
1237 * When calling this method, interrupts MUST be enabled. This is because
1238 * the BH enable code must have IRQs enabled so that it will not deadlock.
1239 * --BLG
1242 int dev_queue_xmit(struct sk_buff *skb)
1244 struct net_device *dev = skb->dev;
1245 struct Qdisc *q;
1246 int rc = -ENOMEM;
1248 if (skb_shinfo(skb)->frag_list &&
1249 !(dev->features & NETIF_F_FRAGLIST) &&
1250 __skb_linearize(skb, GFP_ATOMIC))
1251 goto out_kfree_skb;
1253 /* Fragmented skb is linearized if device does not support SG,
1254 * or if at least one of fragments is in highmem and device
1255 * does not support DMA from it.
1257 if (skb_shinfo(skb)->nr_frags &&
1258 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1259 __skb_linearize(skb, GFP_ATOMIC))
1260 goto out_kfree_skb;
1262 /* If packet is not checksummed and device does not support
1263 * checksumming for this protocol, complete checksumming here.
1265 if (skb->ip_summed == CHECKSUM_HW &&
1266 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1267 (!(dev->features & NETIF_F_IP_CSUM) ||
1268 skb->protocol != htons(ETH_P_IP))))
1269 if (skb_checksum_help(skb, 0))
1270 goto out_kfree_skb;
1272 spin_lock_prefetch(&dev->queue_lock);
1274 /* Disable soft irqs for various locks below. Also
1275 * stops preemption for RCU.
1277 local_bh_disable();
1279 /* Updates of qdisc are serialized by queue_lock.
1280 * The struct Qdisc which is pointed to by qdisc is now a
1281 * rcu structure - it may be accessed without acquiring
1282 * a lock (but the structure may be stale.) The freeing of the
1283 * qdisc will be deferred until it's known that there are no
1284 * more references to it.
1286 * If the qdisc has an enqueue function, we still need to
1287 * hold the queue_lock before calling it, since queue_lock
1288 * also serializes access to the device queue.
1291 q = rcu_dereference(dev->qdisc);
1292 #ifdef CONFIG_NET_CLS_ACT
1293 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1294 #endif
1295 if (q->enqueue) {
1296 /* Grab device queue */
1297 spin_lock(&dev->queue_lock);
1299 rc = q->enqueue(skb, q);
1301 qdisc_run(dev);
1303 spin_unlock(&dev->queue_lock);
1304 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1305 goto out;
1308 /* The device has no queue. Common case for software devices:
1309 loopback, all the sorts of tunnels...
1311 Really, it is unlikely that xmit_lock protection is necessary here.
1312 (f.e. loopback and IP tunnels are clean ignoring statistics
1313 counters.)
1314 However, it is possible, that they rely on protection
1315 made by us here.
1317 Check this and shot the lock. It is not prone from deadlocks.
1318 Either shot noqueue qdisc, it is even simpler 8)
1320 if (dev->flags & IFF_UP) {
1321 int cpu = smp_processor_id(); /* ok because BHs are off */
1323 if (dev->xmit_lock_owner != cpu) {
1325 HARD_TX_LOCK(dev, cpu);
1327 if (!netif_queue_stopped(dev)) {
1328 if (netdev_nit)
1329 dev_queue_xmit_nit(skb, dev);
1331 rc = 0;
1332 if (!dev->hard_start_xmit(skb, dev)) {
1333 HARD_TX_UNLOCK(dev);
1334 goto out;
1337 HARD_TX_UNLOCK(dev);
1338 if (net_ratelimit())
1339 printk(KERN_CRIT "Virtual device %s asks to "
1340 "queue packet!\n", dev->name);
1341 } else {
1342 /* Recursion is detected! It is possible,
1343 * unfortunately */
1344 if (net_ratelimit())
1345 printk(KERN_CRIT "Dead loop on virtual device "
1346 "%s, fix it urgently!\n", dev->name);
1350 rc = -ENETDOWN;
1351 local_bh_enable();
1353 out_kfree_skb:
1354 kfree_skb(skb);
1355 return rc;
1356 out:
1357 local_bh_enable();
1358 return rc;
1362 /*=======================================================================
1363 Receiver routines
1364 =======================================================================*/
1366 int netdev_max_backlog = 1000;
1367 int netdev_budget = 300;
1368 int weight_p = 64; /* old backlog weight */
1370 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1374 * netif_rx - post buffer to the network code
1375 * @skb: buffer to post
1377 * This function receives a packet from a device driver and queues it for
1378 * the upper (protocol) levels to process. It always succeeds. The buffer
1379 * may be dropped during processing for congestion control or by the
1380 * protocol layers.
1382 * return values:
1383 * NET_RX_SUCCESS (no congestion)
1384 * NET_RX_CN_LOW (low congestion)
1385 * NET_RX_CN_MOD (moderate congestion)
1386 * NET_RX_CN_HIGH (high congestion)
1387 * NET_RX_DROP (packet was dropped)
1391 int netif_rx(struct sk_buff *skb)
1393 struct softnet_data *queue;
1394 unsigned long flags;
1396 /* if netpoll wants it, pretend we never saw it */
1397 if (netpoll_rx(skb))
1398 return NET_RX_DROP;
1400 if (!skb->tstamp.off_sec)
1401 net_timestamp(skb);
1404 * The code is rearranged so that the path is the most
1405 * short when CPU is congested, but is still operating.
1407 local_irq_save(flags);
1408 queue = &__get_cpu_var(softnet_data);
1410 __get_cpu_var(netdev_rx_stat).total++;
1411 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1412 if (queue->input_pkt_queue.qlen) {
1413 enqueue:
1414 dev_hold(skb->dev);
1415 __skb_queue_tail(&queue->input_pkt_queue, skb);
1416 local_irq_restore(flags);
1417 return NET_RX_SUCCESS;
1420 netif_rx_schedule(&queue->backlog_dev);
1421 goto enqueue;
1424 __get_cpu_var(netdev_rx_stat).dropped++;
1425 local_irq_restore(flags);
1427 kfree_skb(skb);
1428 return NET_RX_DROP;
1431 int netif_rx_ni(struct sk_buff *skb)
1433 int err;
1435 preempt_disable();
1436 err = netif_rx(skb);
1437 if (local_softirq_pending())
1438 do_softirq();
1439 preempt_enable();
1441 return err;
1444 EXPORT_SYMBOL(netif_rx_ni);
1446 static inline struct net_device *skb_bond(struct sk_buff *skb)
1448 struct net_device *dev = skb->dev;
1450 if (dev->master) {
1452 * On bonding slaves other than the currently active
1453 * slave, suppress duplicates except for 802.3ad
1454 * ETH_P_SLOW and alb non-mcast/bcast.
1456 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1457 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1458 if (skb->pkt_type != PACKET_BROADCAST &&
1459 skb->pkt_type != PACKET_MULTICAST)
1460 goto keep;
1463 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1464 skb->protocol == __constant_htons(ETH_P_SLOW))
1465 goto keep;
1467 kfree_skb(skb);
1468 return NULL;
1470 keep:
1471 skb->dev = dev->master;
1474 return dev;
1477 static void net_tx_action(struct softirq_action *h)
1479 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1481 if (sd->completion_queue) {
1482 struct sk_buff *clist;
1484 local_irq_disable();
1485 clist = sd->completion_queue;
1486 sd->completion_queue = NULL;
1487 local_irq_enable();
1489 while (clist) {
1490 struct sk_buff *skb = clist;
1491 clist = clist->next;
1493 BUG_TRAP(!atomic_read(&skb->users));
1494 __kfree_skb(skb);
1498 if (sd->output_queue) {
1499 struct net_device *head;
1501 local_irq_disable();
1502 head = sd->output_queue;
1503 sd->output_queue = NULL;
1504 local_irq_enable();
1506 while (head) {
1507 struct net_device *dev = head;
1508 head = head->next_sched;
1510 smp_mb__before_clear_bit();
1511 clear_bit(__LINK_STATE_SCHED, &dev->state);
1513 if (spin_trylock(&dev->queue_lock)) {
1514 qdisc_run(dev);
1515 spin_unlock(&dev->queue_lock);
1516 } else {
1517 netif_schedule(dev);
1523 static __inline__ int deliver_skb(struct sk_buff *skb,
1524 struct packet_type *pt_prev,
1525 struct net_device *orig_dev)
1527 atomic_inc(&skb->users);
1528 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1531 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1532 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1533 struct net_bridge;
1534 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1535 unsigned char *addr);
1536 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1538 static __inline__ int handle_bridge(struct sk_buff **pskb,
1539 struct packet_type **pt_prev, int *ret,
1540 struct net_device *orig_dev)
1542 struct net_bridge_port *port;
1544 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1545 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1546 return 0;
1548 if (*pt_prev) {
1549 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1550 *pt_prev = NULL;
1553 return br_handle_frame_hook(port, pskb);
1555 #else
1556 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1557 #endif
1559 #ifdef CONFIG_NET_CLS_ACT
1560 /* TODO: Maybe we should just force sch_ingress to be compiled in
1561 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1562 * a compare and 2 stores extra right now if we dont have it on
1563 * but have CONFIG_NET_CLS_ACT
1564 * NOTE: This doesnt stop any functionality; if you dont have
1565 * the ingress scheduler, you just cant add policies on ingress.
1568 static int ing_filter(struct sk_buff *skb)
1570 struct Qdisc *q;
1571 struct net_device *dev = skb->dev;
1572 int result = TC_ACT_OK;
1574 if (dev->qdisc_ingress) {
1575 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1576 if (MAX_RED_LOOP < ttl++) {
1577 printk("Redir loop detected Dropping packet (%s->%s)\n",
1578 skb->input_dev->name, skb->dev->name);
1579 return TC_ACT_SHOT;
1582 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1584 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1586 spin_lock(&dev->ingress_lock);
1587 if ((q = dev->qdisc_ingress) != NULL)
1588 result = q->enqueue(skb, q);
1589 spin_unlock(&dev->ingress_lock);
1593 return result;
1595 #endif
1597 int netif_receive_skb(struct sk_buff *skb)
1599 struct packet_type *ptype, *pt_prev;
1600 struct net_device *orig_dev;
1601 int ret = NET_RX_DROP;
1602 unsigned short type;
1604 /* if we've gotten here through NAPI, check netpoll */
1605 if (skb->dev->poll && netpoll_rx(skb))
1606 return NET_RX_DROP;
1608 if (!skb->tstamp.off_sec)
1609 net_timestamp(skb);
1611 if (!skb->input_dev)
1612 skb->input_dev = skb->dev;
1614 orig_dev = skb_bond(skb);
1616 if (!orig_dev)
1617 return NET_RX_DROP;
1619 __get_cpu_var(netdev_rx_stat).total++;
1621 skb->h.raw = skb->nh.raw = skb->data;
1622 skb->mac_len = skb->nh.raw - skb->mac.raw;
1624 pt_prev = NULL;
1626 rcu_read_lock();
1628 #ifdef CONFIG_NET_CLS_ACT
1629 if (skb->tc_verd & TC_NCLS) {
1630 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1631 goto ncls;
1633 #endif
1635 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1636 if (!ptype->dev || ptype->dev == skb->dev) {
1637 if (pt_prev)
1638 ret = deliver_skb(skb, pt_prev, orig_dev);
1639 pt_prev = ptype;
1643 #ifdef CONFIG_NET_CLS_ACT
1644 if (pt_prev) {
1645 ret = deliver_skb(skb, pt_prev, orig_dev);
1646 pt_prev = NULL; /* noone else should process this after*/
1647 } else {
1648 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1651 ret = ing_filter(skb);
1653 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1654 kfree_skb(skb);
1655 goto out;
1658 skb->tc_verd = 0;
1659 ncls:
1660 #endif
1662 handle_diverter(skb);
1664 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1665 goto out;
1667 type = skb->protocol;
1668 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1669 if (ptype->type == type &&
1670 (!ptype->dev || ptype->dev == skb->dev)) {
1671 if (pt_prev)
1672 ret = deliver_skb(skb, pt_prev, orig_dev);
1673 pt_prev = ptype;
1677 if (pt_prev) {
1678 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1679 } else {
1680 kfree_skb(skb);
1681 /* Jamal, now you will not able to escape explaining
1682 * me how you were going to use this. :-)
1684 ret = NET_RX_DROP;
1687 out:
1688 rcu_read_unlock();
1689 return ret;
1692 static int process_backlog(struct net_device *backlog_dev, int *budget)
1694 int work = 0;
1695 int quota = min(backlog_dev->quota, *budget);
1696 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1697 unsigned long start_time = jiffies;
1699 backlog_dev->weight = weight_p;
1700 for (;;) {
1701 struct sk_buff *skb;
1702 struct net_device *dev;
1704 local_irq_disable();
1705 skb = __skb_dequeue(&queue->input_pkt_queue);
1706 if (!skb)
1707 goto job_done;
1708 local_irq_enable();
1710 dev = skb->dev;
1712 netif_receive_skb(skb);
1714 dev_put(dev);
1716 work++;
1718 if (work >= quota || jiffies - start_time > 1)
1719 break;
1723 backlog_dev->quota -= work;
1724 *budget -= work;
1725 return -1;
1727 job_done:
1728 backlog_dev->quota -= work;
1729 *budget -= work;
1731 list_del(&backlog_dev->poll_list);
1732 smp_mb__before_clear_bit();
1733 netif_poll_enable(backlog_dev);
1735 local_irq_enable();
1736 return 0;
1739 static void net_rx_action(struct softirq_action *h)
1741 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1742 unsigned long start_time = jiffies;
1743 int budget = netdev_budget;
1744 void *have;
1746 local_irq_disable();
1748 while (!list_empty(&queue->poll_list)) {
1749 struct net_device *dev;
1751 if (budget <= 0 || jiffies - start_time > 1)
1752 goto softnet_break;
1754 local_irq_enable();
1756 dev = list_entry(queue->poll_list.next,
1757 struct net_device, poll_list);
1758 have = netpoll_poll_lock(dev);
1760 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1761 netpoll_poll_unlock(have);
1762 local_irq_disable();
1763 list_move_tail(&dev->poll_list, &queue->poll_list);
1764 if (dev->quota < 0)
1765 dev->quota += dev->weight;
1766 else
1767 dev->quota = dev->weight;
1768 } else {
1769 netpoll_poll_unlock(have);
1770 dev_put(dev);
1771 local_irq_disable();
1774 out:
1775 local_irq_enable();
1776 return;
1778 softnet_break:
1779 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1780 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1781 goto out;
1784 static gifconf_func_t * gifconf_list [NPROTO];
1787 * register_gifconf - register a SIOCGIF handler
1788 * @family: Address family
1789 * @gifconf: Function handler
1791 * Register protocol dependent address dumping routines. The handler
1792 * that is passed must not be freed or reused until it has been replaced
1793 * by another handler.
1795 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1797 if (family >= NPROTO)
1798 return -EINVAL;
1799 gifconf_list[family] = gifconf;
1800 return 0;
1805 * Map an interface index to its name (SIOCGIFNAME)
1809 * We need this ioctl for efficient implementation of the
1810 * if_indextoname() function required by the IPv6 API. Without
1811 * it, we would have to search all the interfaces to find a
1812 * match. --pb
1815 static int dev_ifname(struct ifreq __user *arg)
1817 struct net_device *dev;
1818 struct ifreq ifr;
1821 * Fetch the caller's info block.
1824 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1825 return -EFAULT;
1827 read_lock(&dev_base_lock);
1828 dev = __dev_get_by_index(ifr.ifr_ifindex);
1829 if (!dev) {
1830 read_unlock(&dev_base_lock);
1831 return -ENODEV;
1834 strcpy(ifr.ifr_name, dev->name);
1835 read_unlock(&dev_base_lock);
1837 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1838 return -EFAULT;
1839 return 0;
1843 * Perform a SIOCGIFCONF call. This structure will change
1844 * size eventually, and there is nothing I can do about it.
1845 * Thus we will need a 'compatibility mode'.
1848 static int dev_ifconf(char __user *arg)
1850 struct ifconf ifc;
1851 struct net_device *dev;
1852 char __user *pos;
1853 int len;
1854 int total;
1855 int i;
1858 * Fetch the caller's info block.
1861 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1862 return -EFAULT;
1864 pos = ifc.ifc_buf;
1865 len = ifc.ifc_len;
1868 * Loop over the interfaces, and write an info block for each.
1871 total = 0;
1872 for (dev = dev_base; dev; dev = dev->next) {
1873 for (i = 0; i < NPROTO; i++) {
1874 if (gifconf_list[i]) {
1875 int done;
1876 if (!pos)
1877 done = gifconf_list[i](dev, NULL, 0);
1878 else
1879 done = gifconf_list[i](dev, pos + total,
1880 len - total);
1881 if (done < 0)
1882 return -EFAULT;
1883 total += done;
1889 * All done. Write the updated control block back to the caller.
1891 ifc.ifc_len = total;
1894 * Both BSD and Solaris return 0 here, so we do too.
1896 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1899 #ifdef CONFIG_PROC_FS
1901 * This is invoked by the /proc filesystem handler to display a device
1902 * in detail.
1904 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1906 struct net_device *dev;
1907 loff_t i;
1909 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1911 return i == pos ? dev : NULL;
1914 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1916 read_lock(&dev_base_lock);
1917 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1920 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1922 ++*pos;
1923 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1926 void dev_seq_stop(struct seq_file *seq, void *v)
1928 read_unlock(&dev_base_lock);
1931 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1933 if (dev->get_stats) {
1934 struct net_device_stats *stats = dev->get_stats(dev);
1936 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1937 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1938 dev->name, stats->rx_bytes, stats->rx_packets,
1939 stats->rx_errors,
1940 stats->rx_dropped + stats->rx_missed_errors,
1941 stats->rx_fifo_errors,
1942 stats->rx_length_errors + stats->rx_over_errors +
1943 stats->rx_crc_errors + stats->rx_frame_errors,
1944 stats->rx_compressed, stats->multicast,
1945 stats->tx_bytes, stats->tx_packets,
1946 stats->tx_errors, stats->tx_dropped,
1947 stats->tx_fifo_errors, stats->collisions,
1948 stats->tx_carrier_errors +
1949 stats->tx_aborted_errors +
1950 stats->tx_window_errors +
1951 stats->tx_heartbeat_errors,
1952 stats->tx_compressed);
1953 } else
1954 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1958 * Called from the PROCfs module. This now uses the new arbitrary sized
1959 * /proc/net interface to create /proc/net/dev
1961 static int dev_seq_show(struct seq_file *seq, void *v)
1963 if (v == SEQ_START_TOKEN)
1964 seq_puts(seq, "Inter-| Receive "
1965 " | Transmit\n"
1966 " face |bytes packets errs drop fifo frame "
1967 "compressed multicast|bytes packets errs "
1968 "drop fifo colls carrier compressed\n");
1969 else
1970 dev_seq_printf_stats(seq, v);
1971 return 0;
1974 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1976 struct netif_rx_stats *rc = NULL;
1978 while (*pos < NR_CPUS)
1979 if (cpu_online(*pos)) {
1980 rc = &per_cpu(netdev_rx_stat, *pos);
1981 break;
1982 } else
1983 ++*pos;
1984 return rc;
1987 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1989 return softnet_get_online(pos);
1992 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1994 ++*pos;
1995 return softnet_get_online(pos);
1998 static void softnet_seq_stop(struct seq_file *seq, void *v)
2002 static int softnet_seq_show(struct seq_file *seq, void *v)
2004 struct netif_rx_stats *s = v;
2006 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2007 s->total, s->dropped, s->time_squeeze, 0,
2008 0, 0, 0, 0, /* was fastroute */
2009 s->cpu_collision );
2010 return 0;
2013 static struct seq_operations dev_seq_ops = {
2014 .start = dev_seq_start,
2015 .next = dev_seq_next,
2016 .stop = dev_seq_stop,
2017 .show = dev_seq_show,
2020 static int dev_seq_open(struct inode *inode, struct file *file)
2022 return seq_open(file, &dev_seq_ops);
2025 static struct file_operations dev_seq_fops = {
2026 .owner = THIS_MODULE,
2027 .open = dev_seq_open,
2028 .read = seq_read,
2029 .llseek = seq_lseek,
2030 .release = seq_release,
2033 static struct seq_operations softnet_seq_ops = {
2034 .start = softnet_seq_start,
2035 .next = softnet_seq_next,
2036 .stop = softnet_seq_stop,
2037 .show = softnet_seq_show,
2040 static int softnet_seq_open(struct inode *inode, struct file *file)
2042 return seq_open(file, &softnet_seq_ops);
2045 static struct file_operations softnet_seq_fops = {
2046 .owner = THIS_MODULE,
2047 .open = softnet_seq_open,
2048 .read = seq_read,
2049 .llseek = seq_lseek,
2050 .release = seq_release,
2053 #ifdef CONFIG_WIRELESS_EXT
2054 extern int wireless_proc_init(void);
2055 #else
2056 #define wireless_proc_init() 0
2057 #endif
2059 static int __init dev_proc_init(void)
2061 int rc = -ENOMEM;
2063 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2064 goto out;
2065 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2066 goto out_dev;
2067 if (wireless_proc_init())
2068 goto out_softnet;
2069 rc = 0;
2070 out:
2071 return rc;
2072 out_softnet:
2073 proc_net_remove("softnet_stat");
2074 out_dev:
2075 proc_net_remove("dev");
2076 goto out;
2078 #else
2079 #define dev_proc_init() 0
2080 #endif /* CONFIG_PROC_FS */
2084 * netdev_set_master - set up master/slave pair
2085 * @slave: slave device
2086 * @master: new master device
2088 * Changes the master device of the slave. Pass %NULL to break the
2089 * bonding. The caller must hold the RTNL semaphore. On a failure
2090 * a negative errno code is returned. On success the reference counts
2091 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2092 * function returns zero.
2094 int netdev_set_master(struct net_device *slave, struct net_device *master)
2096 struct net_device *old = slave->master;
2098 ASSERT_RTNL();
2100 if (master) {
2101 if (old)
2102 return -EBUSY;
2103 dev_hold(master);
2106 slave->master = master;
2108 synchronize_net();
2110 if (old)
2111 dev_put(old);
2113 if (master)
2114 slave->flags |= IFF_SLAVE;
2115 else
2116 slave->flags &= ~IFF_SLAVE;
2118 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2119 return 0;
2123 * dev_set_promiscuity - update promiscuity count on a device
2124 * @dev: device
2125 * @inc: modifier
2127 * Add or remove promsicuity from a device. While the count in the device
2128 * remains above zero the interface remains promiscuous. Once it hits zero
2129 * the device reverts back to normal filtering operation. A negative inc
2130 * value is used to drop promiscuity on the device.
2132 void dev_set_promiscuity(struct net_device *dev, int inc)
2134 unsigned short old_flags = dev->flags;
2136 if ((dev->promiscuity += inc) == 0)
2137 dev->flags &= ~IFF_PROMISC;
2138 else
2139 dev->flags |= IFF_PROMISC;
2140 if (dev->flags != old_flags) {
2141 dev_mc_upload(dev);
2142 printk(KERN_INFO "device %s %s promiscuous mode\n",
2143 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2144 "left");
2149 * dev_set_allmulti - update allmulti count on a device
2150 * @dev: device
2151 * @inc: modifier
2153 * Add or remove reception of all multicast frames to a device. While the
2154 * count in the device remains above zero the interface remains listening
2155 * to all interfaces. Once it hits zero the device reverts back to normal
2156 * filtering operation. A negative @inc value is used to drop the counter
2157 * when releasing a resource needing all multicasts.
2160 void dev_set_allmulti(struct net_device *dev, int inc)
2162 unsigned short old_flags = dev->flags;
2164 dev->flags |= IFF_ALLMULTI;
2165 if ((dev->allmulti += inc) == 0)
2166 dev->flags &= ~IFF_ALLMULTI;
2167 if (dev->flags ^ old_flags)
2168 dev_mc_upload(dev);
2171 unsigned dev_get_flags(const struct net_device *dev)
2173 unsigned flags;
2175 flags = (dev->flags & ~(IFF_PROMISC |
2176 IFF_ALLMULTI |
2177 IFF_RUNNING |
2178 IFF_LOWER_UP |
2179 IFF_DORMANT)) |
2180 (dev->gflags & (IFF_PROMISC |
2181 IFF_ALLMULTI));
2183 if (netif_running(dev)) {
2184 if (netif_oper_up(dev))
2185 flags |= IFF_RUNNING;
2186 if (netif_carrier_ok(dev))
2187 flags |= IFF_LOWER_UP;
2188 if (netif_dormant(dev))
2189 flags |= IFF_DORMANT;
2192 return flags;
2195 int dev_change_flags(struct net_device *dev, unsigned flags)
2197 int ret;
2198 int old_flags = dev->flags;
2201 * Set the flags on our device.
2204 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2205 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2206 IFF_AUTOMEDIA)) |
2207 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2208 IFF_ALLMULTI));
2211 * Load in the correct multicast list now the flags have changed.
2214 dev_mc_upload(dev);
2217 * Have we downed the interface. We handle IFF_UP ourselves
2218 * according to user attempts to set it, rather than blindly
2219 * setting it.
2222 ret = 0;
2223 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2224 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2226 if (!ret)
2227 dev_mc_upload(dev);
2230 if (dev->flags & IFF_UP &&
2231 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2232 IFF_VOLATILE)))
2233 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2235 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2236 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2237 dev->gflags ^= IFF_PROMISC;
2238 dev_set_promiscuity(dev, inc);
2241 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2242 is important. Some (broken) drivers set IFF_PROMISC, when
2243 IFF_ALLMULTI is requested not asking us and not reporting.
2245 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2246 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2247 dev->gflags ^= IFF_ALLMULTI;
2248 dev_set_allmulti(dev, inc);
2251 if (old_flags ^ dev->flags)
2252 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2254 return ret;
2257 int dev_set_mtu(struct net_device *dev, int new_mtu)
2259 int err;
2261 if (new_mtu == dev->mtu)
2262 return 0;
2264 /* MTU must be positive. */
2265 if (new_mtu < 0)
2266 return -EINVAL;
2268 if (!netif_device_present(dev))
2269 return -ENODEV;
2271 err = 0;
2272 if (dev->change_mtu)
2273 err = dev->change_mtu(dev, new_mtu);
2274 else
2275 dev->mtu = new_mtu;
2276 if (!err && dev->flags & IFF_UP)
2277 notifier_call_chain(&netdev_chain,
2278 NETDEV_CHANGEMTU, dev);
2279 return err;
2282 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2284 int err;
2286 if (!dev->set_mac_address)
2287 return -EOPNOTSUPP;
2288 if (sa->sa_family != dev->type)
2289 return -EINVAL;
2290 if (!netif_device_present(dev))
2291 return -ENODEV;
2292 err = dev->set_mac_address(dev, sa);
2293 if (!err)
2294 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2295 return err;
2299 * Perform the SIOCxIFxxx calls.
2301 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2303 int err;
2304 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2306 if (!dev)
2307 return -ENODEV;
2309 switch (cmd) {
2310 case SIOCGIFFLAGS: /* Get interface flags */
2311 ifr->ifr_flags = dev_get_flags(dev);
2312 return 0;
2314 case SIOCSIFFLAGS: /* Set interface flags */
2315 return dev_change_flags(dev, ifr->ifr_flags);
2317 case SIOCGIFMETRIC: /* Get the metric on the interface
2318 (currently unused) */
2319 ifr->ifr_metric = 0;
2320 return 0;
2322 case SIOCSIFMETRIC: /* Set the metric on the interface
2323 (currently unused) */
2324 return -EOPNOTSUPP;
2326 case SIOCGIFMTU: /* Get the MTU of a device */
2327 ifr->ifr_mtu = dev->mtu;
2328 return 0;
2330 case SIOCSIFMTU: /* Set the MTU of a device */
2331 return dev_set_mtu(dev, ifr->ifr_mtu);
2333 case SIOCGIFHWADDR:
2334 if (!dev->addr_len)
2335 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2336 else
2337 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2338 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2339 ifr->ifr_hwaddr.sa_family = dev->type;
2340 return 0;
2342 case SIOCSIFHWADDR:
2343 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2345 case SIOCSIFHWBROADCAST:
2346 if (ifr->ifr_hwaddr.sa_family != dev->type)
2347 return -EINVAL;
2348 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2349 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2350 notifier_call_chain(&netdev_chain,
2351 NETDEV_CHANGEADDR, dev);
2352 return 0;
2354 case SIOCGIFMAP:
2355 ifr->ifr_map.mem_start = dev->mem_start;
2356 ifr->ifr_map.mem_end = dev->mem_end;
2357 ifr->ifr_map.base_addr = dev->base_addr;
2358 ifr->ifr_map.irq = dev->irq;
2359 ifr->ifr_map.dma = dev->dma;
2360 ifr->ifr_map.port = dev->if_port;
2361 return 0;
2363 case SIOCSIFMAP:
2364 if (dev->set_config) {
2365 if (!netif_device_present(dev))
2366 return -ENODEV;
2367 return dev->set_config(dev, &ifr->ifr_map);
2369 return -EOPNOTSUPP;
2371 case SIOCADDMULTI:
2372 if (!dev->set_multicast_list ||
2373 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2374 return -EINVAL;
2375 if (!netif_device_present(dev))
2376 return -ENODEV;
2377 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2378 dev->addr_len, 1);
2380 case SIOCDELMULTI:
2381 if (!dev->set_multicast_list ||
2382 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2383 return -EINVAL;
2384 if (!netif_device_present(dev))
2385 return -ENODEV;
2386 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2387 dev->addr_len, 1);
2389 case SIOCGIFINDEX:
2390 ifr->ifr_ifindex = dev->ifindex;
2391 return 0;
2393 case SIOCGIFTXQLEN:
2394 ifr->ifr_qlen = dev->tx_queue_len;
2395 return 0;
2397 case SIOCSIFTXQLEN:
2398 if (ifr->ifr_qlen < 0)
2399 return -EINVAL;
2400 dev->tx_queue_len = ifr->ifr_qlen;
2401 return 0;
2403 case SIOCSIFNAME:
2404 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2405 return dev_change_name(dev, ifr->ifr_newname);
2408 * Unknown or private ioctl
2411 default:
2412 if ((cmd >= SIOCDEVPRIVATE &&
2413 cmd <= SIOCDEVPRIVATE + 15) ||
2414 cmd == SIOCBONDENSLAVE ||
2415 cmd == SIOCBONDRELEASE ||
2416 cmd == SIOCBONDSETHWADDR ||
2417 cmd == SIOCBONDSLAVEINFOQUERY ||
2418 cmd == SIOCBONDINFOQUERY ||
2419 cmd == SIOCBONDCHANGEACTIVE ||
2420 cmd == SIOCGMIIPHY ||
2421 cmd == SIOCGMIIREG ||
2422 cmd == SIOCSMIIREG ||
2423 cmd == SIOCBRADDIF ||
2424 cmd == SIOCBRDELIF ||
2425 cmd == SIOCWANDEV) {
2426 err = -EOPNOTSUPP;
2427 if (dev->do_ioctl) {
2428 if (netif_device_present(dev))
2429 err = dev->do_ioctl(dev, ifr,
2430 cmd);
2431 else
2432 err = -ENODEV;
2434 } else
2435 err = -EINVAL;
2438 return err;
2442 * This function handles all "interface"-type I/O control requests. The actual
2443 * 'doing' part of this is dev_ifsioc above.
2447 * dev_ioctl - network device ioctl
2448 * @cmd: command to issue
2449 * @arg: pointer to a struct ifreq in user space
2451 * Issue ioctl functions to devices. This is normally called by the
2452 * user space syscall interfaces but can sometimes be useful for
2453 * other purposes. The return value is the return from the syscall if
2454 * positive or a negative errno code on error.
2457 int dev_ioctl(unsigned int cmd, void __user *arg)
2459 struct ifreq ifr;
2460 int ret;
2461 char *colon;
2463 /* One special case: SIOCGIFCONF takes ifconf argument
2464 and requires shared lock, because it sleeps writing
2465 to user space.
2468 if (cmd == SIOCGIFCONF) {
2469 rtnl_lock();
2470 ret = dev_ifconf((char __user *) arg);
2471 rtnl_unlock();
2472 return ret;
2474 if (cmd == SIOCGIFNAME)
2475 return dev_ifname((struct ifreq __user *)arg);
2477 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2478 return -EFAULT;
2480 ifr.ifr_name[IFNAMSIZ-1] = 0;
2482 colon = strchr(ifr.ifr_name, ':');
2483 if (colon)
2484 *colon = 0;
2487 * See which interface the caller is talking about.
2490 switch (cmd) {
2492 * These ioctl calls:
2493 * - can be done by all.
2494 * - atomic and do not require locking.
2495 * - return a value
2497 case SIOCGIFFLAGS:
2498 case SIOCGIFMETRIC:
2499 case SIOCGIFMTU:
2500 case SIOCGIFHWADDR:
2501 case SIOCGIFSLAVE:
2502 case SIOCGIFMAP:
2503 case SIOCGIFINDEX:
2504 case SIOCGIFTXQLEN:
2505 dev_load(ifr.ifr_name);
2506 read_lock(&dev_base_lock);
2507 ret = dev_ifsioc(&ifr, cmd);
2508 read_unlock(&dev_base_lock);
2509 if (!ret) {
2510 if (colon)
2511 *colon = ':';
2512 if (copy_to_user(arg, &ifr,
2513 sizeof(struct ifreq)))
2514 ret = -EFAULT;
2516 return ret;
2518 case SIOCETHTOOL:
2519 dev_load(ifr.ifr_name);
2520 rtnl_lock();
2521 ret = dev_ethtool(&ifr);
2522 rtnl_unlock();
2523 if (!ret) {
2524 if (colon)
2525 *colon = ':';
2526 if (copy_to_user(arg, &ifr,
2527 sizeof(struct ifreq)))
2528 ret = -EFAULT;
2530 return ret;
2533 * These ioctl calls:
2534 * - require superuser power.
2535 * - require strict serialization.
2536 * - return a value
2538 case SIOCGMIIPHY:
2539 case SIOCGMIIREG:
2540 case SIOCSIFNAME:
2541 if (!capable(CAP_NET_ADMIN))
2542 return -EPERM;
2543 dev_load(ifr.ifr_name);
2544 rtnl_lock();
2545 ret = dev_ifsioc(&ifr, cmd);
2546 rtnl_unlock();
2547 if (!ret) {
2548 if (colon)
2549 *colon = ':';
2550 if (copy_to_user(arg, &ifr,
2551 sizeof(struct ifreq)))
2552 ret = -EFAULT;
2554 return ret;
2557 * These ioctl calls:
2558 * - require superuser power.
2559 * - require strict serialization.
2560 * - do not return a value
2562 case SIOCSIFFLAGS:
2563 case SIOCSIFMETRIC:
2564 case SIOCSIFMTU:
2565 case SIOCSIFMAP:
2566 case SIOCSIFHWADDR:
2567 case SIOCSIFSLAVE:
2568 case SIOCADDMULTI:
2569 case SIOCDELMULTI:
2570 case SIOCSIFHWBROADCAST:
2571 case SIOCSIFTXQLEN:
2572 case SIOCSMIIREG:
2573 case SIOCBONDENSLAVE:
2574 case SIOCBONDRELEASE:
2575 case SIOCBONDSETHWADDR:
2576 case SIOCBONDCHANGEACTIVE:
2577 case SIOCBRADDIF:
2578 case SIOCBRDELIF:
2579 if (!capable(CAP_NET_ADMIN))
2580 return -EPERM;
2581 /* fall through */
2582 case SIOCBONDSLAVEINFOQUERY:
2583 case SIOCBONDINFOQUERY:
2584 dev_load(ifr.ifr_name);
2585 rtnl_lock();
2586 ret = dev_ifsioc(&ifr, cmd);
2587 rtnl_unlock();
2588 return ret;
2590 case SIOCGIFMEM:
2591 /* Get the per device memory space. We can add this but
2592 * currently do not support it */
2593 case SIOCSIFMEM:
2594 /* Set the per device memory buffer space.
2595 * Not applicable in our case */
2596 case SIOCSIFLINK:
2597 return -EINVAL;
2600 * Unknown or private ioctl.
2602 default:
2603 if (cmd == SIOCWANDEV ||
2604 (cmd >= SIOCDEVPRIVATE &&
2605 cmd <= SIOCDEVPRIVATE + 15)) {
2606 dev_load(ifr.ifr_name);
2607 rtnl_lock();
2608 ret = dev_ifsioc(&ifr, cmd);
2609 rtnl_unlock();
2610 if (!ret && copy_to_user(arg, &ifr,
2611 sizeof(struct ifreq)))
2612 ret = -EFAULT;
2613 return ret;
2615 #ifdef CONFIG_WIRELESS_EXT
2616 /* Take care of Wireless Extensions */
2617 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2618 /* If command is `set a parameter', or
2619 * `get the encoding parameters', check if
2620 * the user has the right to do it */
2621 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2622 if (!capable(CAP_NET_ADMIN))
2623 return -EPERM;
2625 dev_load(ifr.ifr_name);
2626 rtnl_lock();
2627 /* Follow me in net/core/wireless.c */
2628 ret = wireless_process_ioctl(&ifr, cmd);
2629 rtnl_unlock();
2630 if (IW_IS_GET(cmd) &&
2631 copy_to_user(arg, &ifr,
2632 sizeof(struct ifreq)))
2633 ret = -EFAULT;
2634 return ret;
2636 #endif /* CONFIG_WIRELESS_EXT */
2637 return -EINVAL;
2643 * dev_new_index - allocate an ifindex
2645 * Returns a suitable unique value for a new device interface
2646 * number. The caller must hold the rtnl semaphore or the
2647 * dev_base_lock to be sure it remains unique.
2649 static int dev_new_index(void)
2651 static int ifindex;
2652 for (;;) {
2653 if (++ifindex <= 0)
2654 ifindex = 1;
2655 if (!__dev_get_by_index(ifindex))
2656 return ifindex;
2660 static int dev_boot_phase = 1;
2662 /* Delayed registration/unregisteration */
2663 static DEFINE_SPINLOCK(net_todo_list_lock);
2664 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2666 static inline void net_set_todo(struct net_device *dev)
2668 spin_lock(&net_todo_list_lock);
2669 list_add_tail(&dev->todo_list, &net_todo_list);
2670 spin_unlock(&net_todo_list_lock);
2674 * register_netdevice - register a network device
2675 * @dev: device to register
2677 * Take a completed network device structure and add it to the kernel
2678 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2679 * chain. 0 is returned on success. A negative errno code is returned
2680 * on a failure to set up the device, or if the name is a duplicate.
2682 * Callers must hold the rtnl semaphore. You may want
2683 * register_netdev() instead of this.
2685 * BUGS:
2686 * The locking appears insufficient to guarantee two parallel registers
2687 * will not get the same name.
2690 int register_netdevice(struct net_device *dev)
2692 struct hlist_head *head;
2693 struct hlist_node *p;
2694 int ret;
2696 BUG_ON(dev_boot_phase);
2697 ASSERT_RTNL();
2699 /* When net_device's are persistent, this will be fatal. */
2700 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2702 spin_lock_init(&dev->queue_lock);
2703 spin_lock_init(&dev->xmit_lock);
2704 dev->xmit_lock_owner = -1;
2705 #ifdef CONFIG_NET_CLS_ACT
2706 spin_lock_init(&dev->ingress_lock);
2707 #endif
2709 ret = alloc_divert_blk(dev);
2710 if (ret)
2711 goto out;
2713 dev->iflink = -1;
2715 /* Init, if this function is available */
2716 if (dev->init) {
2717 ret = dev->init(dev);
2718 if (ret) {
2719 if (ret > 0)
2720 ret = -EIO;
2721 goto out_err;
2725 if (!dev_valid_name(dev->name)) {
2726 ret = -EINVAL;
2727 goto out_err;
2730 dev->ifindex = dev_new_index();
2731 if (dev->iflink == -1)
2732 dev->iflink = dev->ifindex;
2734 /* Check for existence of name */
2735 head = dev_name_hash(dev->name);
2736 hlist_for_each(p, head) {
2737 struct net_device *d
2738 = hlist_entry(p, struct net_device, name_hlist);
2739 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2740 ret = -EEXIST;
2741 goto out_err;
2745 /* Fix illegal SG+CSUM combinations. */
2746 if ((dev->features & NETIF_F_SG) &&
2747 !(dev->features & (NETIF_F_IP_CSUM |
2748 NETIF_F_NO_CSUM |
2749 NETIF_F_HW_CSUM))) {
2750 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2751 dev->name);
2752 dev->features &= ~NETIF_F_SG;
2755 /* TSO requires that SG is present as well. */
2756 if ((dev->features & NETIF_F_TSO) &&
2757 !(dev->features & NETIF_F_SG)) {
2758 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2759 dev->name);
2760 dev->features &= ~NETIF_F_TSO;
2762 if (dev->features & NETIF_F_UFO) {
2763 if (!(dev->features & NETIF_F_HW_CSUM)) {
2764 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2765 "NETIF_F_HW_CSUM feature.\n",
2766 dev->name);
2767 dev->features &= ~NETIF_F_UFO;
2769 if (!(dev->features & NETIF_F_SG)) {
2770 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2771 "NETIF_F_SG feature.\n",
2772 dev->name);
2773 dev->features &= ~NETIF_F_UFO;
2778 * nil rebuild_header routine,
2779 * that should be never called and used as just bug trap.
2782 if (!dev->rebuild_header)
2783 dev->rebuild_header = default_rebuild_header;
2786 * Default initial state at registry is that the
2787 * device is present.
2790 set_bit(__LINK_STATE_PRESENT, &dev->state);
2792 dev->next = NULL;
2793 dev_init_scheduler(dev);
2794 write_lock_bh(&dev_base_lock);
2795 *dev_tail = dev;
2796 dev_tail = &dev->next;
2797 hlist_add_head(&dev->name_hlist, head);
2798 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2799 dev_hold(dev);
2800 dev->reg_state = NETREG_REGISTERING;
2801 write_unlock_bh(&dev_base_lock);
2803 /* Notify protocols, that a new device appeared. */
2804 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2806 /* Finish registration after unlock */
2807 net_set_todo(dev);
2808 ret = 0;
2810 out:
2811 return ret;
2812 out_err:
2813 free_divert_blk(dev);
2814 goto out;
2818 * register_netdev - register a network device
2819 * @dev: device to register
2821 * Take a completed network device structure and add it to the kernel
2822 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2823 * chain. 0 is returned on success. A negative errno code is returned
2824 * on a failure to set up the device, or if the name is a duplicate.
2826 * This is a wrapper around register_netdev that takes the rtnl semaphore
2827 * and expands the device name if you passed a format string to
2828 * alloc_netdev.
2830 int register_netdev(struct net_device *dev)
2832 int err;
2834 rtnl_lock();
2837 * If the name is a format string the caller wants us to do a
2838 * name allocation.
2840 if (strchr(dev->name, '%')) {
2841 err = dev_alloc_name(dev, dev->name);
2842 if (err < 0)
2843 goto out;
2847 * Back compatibility hook. Kill this one in 2.5
2849 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2850 err = dev_alloc_name(dev, "eth%d");
2851 if (err < 0)
2852 goto out;
2855 err = register_netdevice(dev);
2856 out:
2857 rtnl_unlock();
2858 return err;
2860 EXPORT_SYMBOL(register_netdev);
2863 * netdev_wait_allrefs - wait until all references are gone.
2865 * This is called when unregistering network devices.
2867 * Any protocol or device that holds a reference should register
2868 * for netdevice notification, and cleanup and put back the
2869 * reference if they receive an UNREGISTER event.
2870 * We can get stuck here if buggy protocols don't correctly
2871 * call dev_put.
2873 static void netdev_wait_allrefs(struct net_device *dev)
2875 unsigned long rebroadcast_time, warning_time;
2877 rebroadcast_time = warning_time = jiffies;
2878 while (atomic_read(&dev->refcnt) != 0) {
2879 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2880 rtnl_lock();
2882 /* Rebroadcast unregister notification */
2883 notifier_call_chain(&netdev_chain,
2884 NETDEV_UNREGISTER, dev);
2886 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2887 &dev->state)) {
2888 /* We must not have linkwatch events
2889 * pending on unregister. If this
2890 * happens, we simply run the queue
2891 * unscheduled, resulting in a noop
2892 * for this device.
2894 linkwatch_run_queue();
2897 __rtnl_unlock();
2899 rebroadcast_time = jiffies;
2902 msleep(250);
2904 if (time_after(jiffies, warning_time + 10 * HZ)) {
2905 printk(KERN_EMERG "unregister_netdevice: "
2906 "waiting for %s to become free. Usage "
2907 "count = %d\n",
2908 dev->name, atomic_read(&dev->refcnt));
2909 warning_time = jiffies;
2914 /* The sequence is:
2916 * rtnl_lock();
2917 * ...
2918 * register_netdevice(x1);
2919 * register_netdevice(x2);
2920 * ...
2921 * unregister_netdevice(y1);
2922 * unregister_netdevice(y2);
2923 * ...
2924 * rtnl_unlock();
2925 * free_netdev(y1);
2926 * free_netdev(y2);
2928 * We are invoked by rtnl_unlock() after it drops the semaphore.
2929 * This allows us to deal with problems:
2930 * 1) We can create/delete sysfs objects which invoke hotplug
2931 * without deadlocking with linkwatch via keventd.
2932 * 2) Since we run with the RTNL semaphore not held, we can sleep
2933 * safely in order to wait for the netdev refcnt to drop to zero.
2935 static DEFINE_MUTEX(net_todo_run_mutex);
2936 void netdev_run_todo(void)
2938 struct list_head list = LIST_HEAD_INIT(list);
2939 int err;
2942 /* Need to guard against multiple cpu's getting out of order. */
2943 mutex_lock(&net_todo_run_mutex);
2945 /* Not safe to do outside the semaphore. We must not return
2946 * until all unregister events invoked by the local processor
2947 * have been completed (either by this todo run, or one on
2948 * another cpu).
2950 if (list_empty(&net_todo_list))
2951 goto out;
2953 /* Snapshot list, allow later requests */
2954 spin_lock(&net_todo_list_lock);
2955 list_splice_init(&net_todo_list, &list);
2956 spin_unlock(&net_todo_list_lock);
2958 while (!list_empty(&list)) {
2959 struct net_device *dev
2960 = list_entry(list.next, struct net_device, todo_list);
2961 list_del(&dev->todo_list);
2963 switch(dev->reg_state) {
2964 case NETREG_REGISTERING:
2965 err = netdev_register_sysfs(dev);
2966 if (err)
2967 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2968 dev->name, err);
2969 dev->reg_state = NETREG_REGISTERED;
2970 break;
2972 case NETREG_UNREGISTERING:
2973 netdev_unregister_sysfs(dev);
2974 dev->reg_state = NETREG_UNREGISTERED;
2976 netdev_wait_allrefs(dev);
2978 /* paranoia */
2979 BUG_ON(atomic_read(&dev->refcnt));
2980 BUG_TRAP(!dev->ip_ptr);
2981 BUG_TRAP(!dev->ip6_ptr);
2982 BUG_TRAP(!dev->dn_ptr);
2985 /* It must be the very last action,
2986 * after this 'dev' may point to freed up memory.
2988 if (dev->destructor)
2989 dev->destructor(dev);
2990 break;
2992 default:
2993 printk(KERN_ERR "network todo '%s' but state %d\n",
2994 dev->name, dev->reg_state);
2995 break;
2999 out:
3000 mutex_unlock(&net_todo_run_mutex);
3004 * alloc_netdev - allocate network device
3005 * @sizeof_priv: size of private data to allocate space for
3006 * @name: device name format string
3007 * @setup: callback to initialize device
3009 * Allocates a struct net_device with private data area for driver use
3010 * and performs basic initialization.
3012 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3013 void (*setup)(struct net_device *))
3015 void *p;
3016 struct net_device *dev;
3017 int alloc_size;
3019 /* ensure 32-byte alignment of both the device and private area */
3020 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3021 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3023 p = kmalloc(alloc_size, GFP_KERNEL);
3024 if (!p) {
3025 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3026 return NULL;
3028 memset(p, 0, alloc_size);
3030 dev = (struct net_device *)
3031 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3032 dev->padded = (char *)dev - (char *)p;
3034 if (sizeof_priv)
3035 dev->priv = netdev_priv(dev);
3037 setup(dev);
3038 strcpy(dev->name, name);
3039 return dev;
3041 EXPORT_SYMBOL(alloc_netdev);
3044 * free_netdev - free network device
3045 * @dev: device
3047 * This function does the last stage of destroying an allocated device
3048 * interface. The reference to the device object is released.
3049 * If this is the last reference then it will be freed.
3051 void free_netdev(struct net_device *dev)
3053 #ifdef CONFIG_SYSFS
3054 /* Compatiablity with error handling in drivers */
3055 if (dev->reg_state == NETREG_UNINITIALIZED) {
3056 kfree((char *)dev - dev->padded);
3057 return;
3060 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3061 dev->reg_state = NETREG_RELEASED;
3063 /* will free via class release */
3064 class_device_put(&dev->class_dev);
3065 #else
3066 kfree((char *)dev - dev->padded);
3067 #endif
3070 /* Synchronize with packet receive processing. */
3071 void synchronize_net(void)
3073 might_sleep();
3074 synchronize_rcu();
3078 * unregister_netdevice - remove device from the kernel
3079 * @dev: device
3081 * This function shuts down a device interface and removes it
3082 * from the kernel tables. On success 0 is returned, on a failure
3083 * a negative errno code is returned.
3085 * Callers must hold the rtnl semaphore. You may want
3086 * unregister_netdev() instead of this.
3089 int unregister_netdevice(struct net_device *dev)
3091 struct net_device *d, **dp;
3093 BUG_ON(dev_boot_phase);
3094 ASSERT_RTNL();
3096 /* Some devices call without registering for initialization unwind. */
3097 if (dev->reg_state == NETREG_UNINITIALIZED) {
3098 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3099 "was registered\n", dev->name, dev);
3100 return -ENODEV;
3103 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3105 /* If device is running, close it first. */
3106 if (dev->flags & IFF_UP)
3107 dev_close(dev);
3109 /* And unlink it from device chain. */
3110 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3111 if (d == dev) {
3112 write_lock_bh(&dev_base_lock);
3113 hlist_del(&dev->name_hlist);
3114 hlist_del(&dev->index_hlist);
3115 if (dev_tail == &dev->next)
3116 dev_tail = dp;
3117 *dp = d->next;
3118 write_unlock_bh(&dev_base_lock);
3119 break;
3122 if (!d) {
3123 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3124 dev->name);
3125 return -ENODEV;
3128 dev->reg_state = NETREG_UNREGISTERING;
3130 synchronize_net();
3132 /* Shutdown queueing discipline. */
3133 dev_shutdown(dev);
3136 /* Notify protocols, that we are about to destroy
3137 this device. They should clean all the things.
3139 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3142 * Flush the multicast chain
3144 dev_mc_discard(dev);
3146 if (dev->uninit)
3147 dev->uninit(dev);
3149 /* Notifier chain MUST detach us from master device. */
3150 BUG_TRAP(!dev->master);
3152 free_divert_blk(dev);
3154 /* Finish processing unregister after unlock */
3155 net_set_todo(dev);
3157 synchronize_net();
3159 dev_put(dev);
3160 return 0;
3164 * unregister_netdev - remove device from the kernel
3165 * @dev: device
3167 * This function shuts down a device interface and removes it
3168 * from the kernel tables. On success 0 is returned, on a failure
3169 * a negative errno code is returned.
3171 * This is just a wrapper for unregister_netdevice that takes
3172 * the rtnl semaphore. In general you want to use this and not
3173 * unregister_netdevice.
3175 void unregister_netdev(struct net_device *dev)
3177 rtnl_lock();
3178 unregister_netdevice(dev);
3179 rtnl_unlock();
3182 EXPORT_SYMBOL(unregister_netdev);
3184 #ifdef CONFIG_HOTPLUG_CPU
3185 static int dev_cpu_callback(struct notifier_block *nfb,
3186 unsigned long action,
3187 void *ocpu)
3189 struct sk_buff **list_skb;
3190 struct net_device **list_net;
3191 struct sk_buff *skb;
3192 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3193 struct softnet_data *sd, *oldsd;
3195 if (action != CPU_DEAD)
3196 return NOTIFY_OK;
3198 local_irq_disable();
3199 cpu = smp_processor_id();
3200 sd = &per_cpu(softnet_data, cpu);
3201 oldsd = &per_cpu(softnet_data, oldcpu);
3203 /* Find end of our completion_queue. */
3204 list_skb = &sd->completion_queue;
3205 while (*list_skb)
3206 list_skb = &(*list_skb)->next;
3207 /* Append completion queue from offline CPU. */
3208 *list_skb = oldsd->completion_queue;
3209 oldsd->completion_queue = NULL;
3211 /* Find end of our output_queue. */
3212 list_net = &sd->output_queue;
3213 while (*list_net)
3214 list_net = &(*list_net)->next_sched;
3215 /* Append output queue from offline CPU. */
3216 *list_net = oldsd->output_queue;
3217 oldsd->output_queue = NULL;
3219 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3220 local_irq_enable();
3222 /* Process offline CPU's input_pkt_queue */
3223 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3224 netif_rx(skb);
3226 return NOTIFY_OK;
3228 #endif /* CONFIG_HOTPLUG_CPU */
3232 * Initialize the DEV module. At boot time this walks the device list and
3233 * unhooks any devices that fail to initialise (normally hardware not
3234 * present) and leaves us with a valid list of present and active devices.
3239 * This is called single threaded during boot, so no need
3240 * to take the rtnl semaphore.
3242 static int __init net_dev_init(void)
3244 int i, rc = -ENOMEM;
3246 BUG_ON(!dev_boot_phase);
3248 net_random_init();
3250 if (dev_proc_init())
3251 goto out;
3253 if (netdev_sysfs_init())
3254 goto out;
3256 INIT_LIST_HEAD(&ptype_all);
3257 for (i = 0; i < 16; i++)
3258 INIT_LIST_HEAD(&ptype_base[i]);
3260 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3261 INIT_HLIST_HEAD(&dev_name_head[i]);
3263 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3264 INIT_HLIST_HEAD(&dev_index_head[i]);
3267 * Initialise the packet receive queues.
3270 for_each_cpu(i) {
3271 struct softnet_data *queue;
3273 queue = &per_cpu(softnet_data, i);
3274 skb_queue_head_init(&queue->input_pkt_queue);
3275 queue->completion_queue = NULL;
3276 INIT_LIST_HEAD(&queue->poll_list);
3277 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3278 queue->backlog_dev.weight = weight_p;
3279 queue->backlog_dev.poll = process_backlog;
3280 atomic_set(&queue->backlog_dev.refcnt, 1);
3283 dev_boot_phase = 0;
3285 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3286 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3288 hotcpu_notifier(dev_cpu_callback, 0);
3289 dst_init();
3290 dev_mcast_init();
3291 rc = 0;
3292 out:
3293 return rc;
3296 subsys_initcall(net_dev_init);
3298 EXPORT_SYMBOL(__dev_get_by_index);
3299 EXPORT_SYMBOL(__dev_get_by_name);
3300 EXPORT_SYMBOL(__dev_remove_pack);
3301 EXPORT_SYMBOL(__skb_linearize);
3302 EXPORT_SYMBOL(dev_valid_name);
3303 EXPORT_SYMBOL(dev_add_pack);
3304 EXPORT_SYMBOL(dev_alloc_name);
3305 EXPORT_SYMBOL(dev_close);
3306 EXPORT_SYMBOL(dev_get_by_flags);
3307 EXPORT_SYMBOL(dev_get_by_index);
3308 EXPORT_SYMBOL(dev_get_by_name);
3309 EXPORT_SYMBOL(dev_open);
3310 EXPORT_SYMBOL(dev_queue_xmit);
3311 EXPORT_SYMBOL(dev_remove_pack);
3312 EXPORT_SYMBOL(dev_set_allmulti);
3313 EXPORT_SYMBOL(dev_set_promiscuity);
3314 EXPORT_SYMBOL(dev_change_flags);
3315 EXPORT_SYMBOL(dev_set_mtu);
3316 EXPORT_SYMBOL(dev_set_mac_address);
3317 EXPORT_SYMBOL(free_netdev);
3318 EXPORT_SYMBOL(netdev_boot_setup_check);
3319 EXPORT_SYMBOL(netdev_set_master);
3320 EXPORT_SYMBOL(netdev_state_change);
3321 EXPORT_SYMBOL(netif_receive_skb);
3322 EXPORT_SYMBOL(netif_rx);
3323 EXPORT_SYMBOL(register_gifconf);
3324 EXPORT_SYMBOL(register_netdevice);
3325 EXPORT_SYMBOL(register_netdevice_notifier);
3326 EXPORT_SYMBOL(skb_checksum_help);
3327 EXPORT_SYMBOL(synchronize_net);
3328 EXPORT_SYMBOL(unregister_netdevice);
3329 EXPORT_SYMBOL(unregister_netdevice_notifier);
3330 EXPORT_SYMBOL(net_enable_timestamp);
3331 EXPORT_SYMBOL(net_disable_timestamp);
3332 EXPORT_SYMBOL(dev_get_flags);
3334 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3335 EXPORT_SYMBOL(br_handle_frame_hook);
3336 EXPORT_SYMBOL(br_fdb_get_hook);
3337 EXPORT_SYMBOL(br_fdb_put_hook);
3338 #endif
3340 #ifdef CONFIG_KMOD
3341 EXPORT_SYMBOL(dev_load);
3342 #endif
3344 EXPORT_PER_CPU_SYMBOL(softnet_data);