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
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
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>
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
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
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
51 * Rudi Cilibrasi : Pass the right thing to
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
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
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/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/notifier.h>
94 #include <linux/skbuff.h>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
101 #include <linux/divert.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/kmod.h>
108 #include <linux/module.h>
109 #include <linux/kallsyms.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #ifdef CONFIG_NET_RADIO
114 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
115 #include <net/iw_handler.h>
116 #endif /* CONFIG_NET_RADIO */
117 #include <asm/current.h>
120 * The list of packet types we will receive (as opposed to discard)
121 * and the routines to invoke.
123 * Why 16. Because with 16 the only overlap we get on a hash of the
124 * low nibble of the protocol value is RARP/SNAP/X.25.
126 * NOTE: That is no longer true with the addition of VLAN tags. Not
127 * sure which should go first, but I bet it won't make much
128 * difference if we are running VLANs. The good news is that
129 * this protocol won't be in the list unless compiled in, so
130 * the average user (w/out VLANs) will not be adversly affected.
147 static DEFINE_SPINLOCK(ptype_lock);
148 static struct list_head ptype_base[16]; /* 16 way hashed list */
149 static struct list_head ptype_all; /* Taps */
152 * The @dev_base list is protected by @dev_base_lock and the rtln
155 * Pure readers hold dev_base_lock for reading.
157 * Writers must hold the rtnl semaphore while they loop through the
158 * dev_base list, and hold dev_base_lock for writing when they do the
159 * actual updates. This allows pure readers to access the list even
160 * while a writer is preparing to update it.
162 * To put it another way, dev_base_lock is held for writing only to
163 * protect against pure readers; the rtnl semaphore provides the
164 * protection against other writers.
166 * See, for example usages, register_netdevice() and
167 * unregister_netdevice(), which must be called with the rtnl
170 struct net_device *dev_base;
171 static struct net_device **dev_tail = &dev_base;
172 DEFINE_RWLOCK(dev_base_lock);
174 EXPORT_SYMBOL(dev_base);
175 EXPORT_SYMBOL(dev_base_lock);
177 #define NETDEV_HASHBITS 8
178 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
179 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
181 static inline struct hlist_head *dev_name_hash(const char *name)
183 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
184 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
187 static inline struct hlist_head *dev_index_hash(int ifindex)
189 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
196 static struct notifier_block *netdev_chain;
199 * Device drivers call our routines to queue packets here. We empty the
200 * queue in the local softnet handler.
202 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
205 extern int netdev_sysfs_init(void);
206 extern int netdev_register_sysfs(struct net_device *);
207 extern void netdev_unregister_sysfs(struct net_device *);
209 #define netdev_sysfs_init() (0)
210 #define netdev_register_sysfs(dev) (0)
211 #define netdev_unregister_sysfs(dev) do { } while(0)
215 /*******************************************************************************
217 Protocol management and registration routines
219 *******************************************************************************/
228 * Add a protocol ID to the list. Now that the input handler is
229 * smarter we can dispense with all the messy stuff that used to be
232 * BEWARE!!! Protocol handlers, mangling input packets,
233 * MUST BE last in hash buckets and checking protocol handlers
234 * MUST start from promiscuous ptype_all chain in net_bh.
235 * It is true now, do not change it.
236 * Explanation follows: if protocol handler, mangling packet, will
237 * be the first on list, it is not able to sense, that packet
238 * is cloned and should be copied-on-write, so that it will
239 * change it and subsequent readers will get broken packet.
244 * dev_add_pack - add packet handler
245 * @pt: packet type declaration
247 * Add a protocol handler to the networking stack. The passed &packet_type
248 * is linked into kernel lists and may not be freed until it has been
249 * removed from the kernel lists.
251 * This call does not sleep therefore it can not
252 * guarantee all CPU's that are in middle of receiving packets
253 * will see the new packet type (until the next received packet).
256 void dev_add_pack(struct packet_type *pt)
260 spin_lock_bh(&ptype_lock);
261 if (pt->type == htons(ETH_P_ALL)) {
263 list_add_rcu(&pt->list, &ptype_all);
265 hash = ntohs(pt->type) & 15;
266 list_add_rcu(&pt->list, &ptype_base[hash]);
268 spin_unlock_bh(&ptype_lock);
272 * __dev_remove_pack - remove packet handler
273 * @pt: packet type declaration
275 * Remove a protocol handler that was previously added to the kernel
276 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
277 * from the kernel lists and can be freed or reused once this function
280 * The packet type might still be in use by receivers
281 * and must not be freed until after all the CPU's have gone
282 * through a quiescent state.
284 void __dev_remove_pack(struct packet_type *pt)
286 struct list_head *head;
287 struct packet_type *pt1;
289 spin_lock_bh(&ptype_lock);
291 if (pt->type == htons(ETH_P_ALL)) {
295 head = &ptype_base[ntohs(pt->type) & 15];
297 list_for_each_entry(pt1, head, list) {
299 list_del_rcu(&pt->list);
304 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
306 spin_unlock_bh(&ptype_lock);
309 * dev_remove_pack - remove packet handler
310 * @pt: packet type declaration
312 * Remove a protocol handler that was previously added to the kernel
313 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
314 * from the kernel lists and can be freed or reused once this function
317 * This call sleeps to guarantee that no CPU is looking at the packet
320 void dev_remove_pack(struct packet_type *pt)
322 __dev_remove_pack(pt);
327 /******************************************************************************
329 Device Boot-time Settings Routines
331 *******************************************************************************/
333 /* Boot time configuration table */
334 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
337 * netdev_boot_setup_add - add new setup entry
338 * @name: name of the device
339 * @map: configured settings for the device
341 * Adds new setup entry to the dev_boot_setup list. The function
342 * returns 0 on error and 1 on success. This is a generic routine to
345 static int netdev_boot_setup_add(char *name, struct ifmap *map)
347 struct netdev_boot_setup *s;
351 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
352 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
353 memset(s[i].name, 0, sizeof(s[i].name));
354 strcpy(s[i].name, name);
355 memcpy(&s[i].map, map, sizeof(s[i].map));
360 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
364 * netdev_boot_setup_check - check boot time settings
365 * @dev: the netdevice
367 * Check boot time settings for the device.
368 * The found settings are set for the device to be used
369 * later in the device probing.
370 * Returns 0 if no settings found, 1 if they are.
372 int netdev_boot_setup_check(struct net_device *dev)
374 struct netdev_boot_setup *s = dev_boot_setup;
377 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
378 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
379 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
380 dev->irq = s[i].map.irq;
381 dev->base_addr = s[i].map.base_addr;
382 dev->mem_start = s[i].map.mem_start;
383 dev->mem_end = s[i].map.mem_end;
392 * netdev_boot_base - get address from boot time settings
393 * @prefix: prefix for network device
394 * @unit: id for network device
396 * Check boot time settings for the base address of device.
397 * The found settings are set for the device to be used
398 * later in the device probing.
399 * Returns 0 if no settings found.
401 unsigned long netdev_boot_base(const char *prefix, int unit)
403 const struct netdev_boot_setup *s = dev_boot_setup;
407 sprintf(name, "%s%d", prefix, unit);
410 * If device already registered then return base of 1
411 * to indicate not to probe for this interface
413 if (__dev_get_by_name(name))
416 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
417 if (!strcmp(name, s[i].name))
418 return s[i].map.base_addr;
423 * Saves at boot time configured settings for any netdevice.
425 int __init netdev_boot_setup(char *str)
430 str = get_options(str, ARRAY_SIZE(ints), ints);
435 memset(&map, 0, sizeof(map));
439 map.base_addr = ints[2];
441 map.mem_start = ints[3];
443 map.mem_end = ints[4];
445 /* Add new entry to the list */
446 return netdev_boot_setup_add(str, &map);
449 __setup("netdev=", netdev_boot_setup);
451 /*******************************************************************************
453 Device Interface Subroutines
455 *******************************************************************************/
458 * __dev_get_by_name - find a device by its name
459 * @name: name to find
461 * Find an interface by name. Must be called under RTNL semaphore
462 * or @dev_base_lock. If the name is found a pointer to the device
463 * is returned. If the name is not found then %NULL is returned. The
464 * reference counters are not incremented so the caller must be
465 * careful with locks.
468 struct net_device *__dev_get_by_name(const char *name)
470 struct hlist_node *p;
472 hlist_for_each(p, dev_name_hash(name)) {
473 struct net_device *dev
474 = hlist_entry(p, struct net_device, name_hlist);
475 if (!strncmp(dev->name, name, IFNAMSIZ))
482 * dev_get_by_name - find a device by its name
483 * @name: name to find
485 * Find an interface by name. This can be called from any
486 * context and does its own locking. The returned handle has
487 * the usage count incremented and the caller must use dev_put() to
488 * release it when it is no longer needed. %NULL is returned if no
489 * matching device is found.
492 struct net_device *dev_get_by_name(const char *name)
494 struct net_device *dev;
496 read_lock(&dev_base_lock);
497 dev = __dev_get_by_name(name);
500 read_unlock(&dev_base_lock);
505 * __dev_get_by_index - find a device by its ifindex
506 * @ifindex: index of device
508 * Search for an interface by index. Returns %NULL if the device
509 * is not found or a pointer to the device. The device has not
510 * had its reference counter increased so the caller must be careful
511 * about locking. The caller must hold either the RTNL semaphore
515 struct net_device *__dev_get_by_index(int ifindex)
517 struct hlist_node *p;
519 hlist_for_each(p, dev_index_hash(ifindex)) {
520 struct net_device *dev
521 = hlist_entry(p, struct net_device, index_hlist);
522 if (dev->ifindex == ifindex)
530 * dev_get_by_index - find a device by its ifindex
531 * @ifindex: index of device
533 * Search for an interface by index. Returns NULL if the device
534 * is not found or a pointer to the device. The device returned has
535 * had a reference added and the pointer is safe until the user calls
536 * dev_put to indicate they have finished with it.
539 struct net_device *dev_get_by_index(int ifindex)
541 struct net_device *dev;
543 read_lock(&dev_base_lock);
544 dev = __dev_get_by_index(ifindex);
547 read_unlock(&dev_base_lock);
552 * dev_getbyhwaddr - find a device by its hardware address
553 * @type: media type of device
554 * @ha: hardware address
556 * Search for an interface by MAC address. Returns NULL if the device
557 * is not found or a pointer to the device. The caller must hold the
558 * rtnl semaphore. The returned device has not had its ref count increased
559 * and the caller must therefore be careful about locking
562 * If the API was consistent this would be __dev_get_by_hwaddr
565 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
567 struct net_device *dev;
571 for (dev = dev_base; dev; dev = dev->next)
572 if (dev->type == type &&
573 !memcmp(dev->dev_addr, ha, dev->addr_len))
578 EXPORT_SYMBOL(dev_getbyhwaddr);
580 struct net_device *dev_getfirstbyhwtype(unsigned short type)
582 struct net_device *dev;
585 for (dev = dev_base; dev; dev = dev->next) {
586 if (dev->type == type) {
595 EXPORT_SYMBOL(dev_getfirstbyhwtype);
598 * dev_get_by_flags - find any device with given flags
599 * @if_flags: IFF_* values
600 * @mask: bitmask of bits in if_flags to check
602 * Search for any interface with the given flags. Returns NULL if a device
603 * is not found or a pointer to the device. The device returned has
604 * had a reference added and the pointer is safe until the user calls
605 * dev_put to indicate they have finished with it.
608 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
610 struct net_device *dev;
612 read_lock(&dev_base_lock);
613 for (dev = dev_base; dev != NULL; dev = dev->next) {
614 if (((dev->flags ^ if_flags) & mask) == 0) {
619 read_unlock(&dev_base_lock);
624 * dev_valid_name - check if name is okay for network device
627 * Network device names need to be valid file names to
628 * to allow sysfs to work
630 int dev_valid_name(const char *name)
632 return !(*name == '\0'
633 || !strcmp(name, ".")
634 || !strcmp(name, "..")
635 || strchr(name, '/'));
639 * dev_alloc_name - allocate a name for a device
641 * @name: name format string
643 * Passed a format string - eg "lt%d" it will try and find a suitable
644 * id. Not efficient for many devices, not called a lot. The caller
645 * must hold the dev_base or rtnl lock while allocating the name and
646 * adding the device in order to avoid duplicates. Returns the number
647 * of the unit assigned or a negative errno code.
650 int dev_alloc_name(struct net_device *dev, const char *name)
655 const int max_netdevices = 8*PAGE_SIZE;
657 struct net_device *d;
659 p = strnchr(name, IFNAMSIZ-1, '%');
662 * Verify the string as this thing may have come from
663 * the user. There must be either one "%d" and no other "%"
666 if (p[1] != 'd' || strchr(p + 2, '%'))
669 /* Use one page as a bit array of possible slots */
670 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
674 for (d = dev_base; d; d = d->next) {
675 if (!sscanf(d->name, name, &i))
677 if (i < 0 || i >= max_netdevices)
680 /* avoid cases where sscanf is not exact inverse of printf */
681 snprintf(buf, sizeof(buf), name, i);
682 if (!strncmp(buf, d->name, IFNAMSIZ))
686 i = find_first_zero_bit(inuse, max_netdevices);
687 free_page((unsigned long) inuse);
690 snprintf(buf, sizeof(buf), name, i);
691 if (!__dev_get_by_name(buf)) {
692 strlcpy(dev->name, buf, IFNAMSIZ);
696 /* It is possible to run out of possible slots
697 * when the name is long and there isn't enough space left
698 * for the digits, or if all bits are used.
705 * dev_change_name - change name of a device
707 * @newname: name (or format string) must be at least IFNAMSIZ
709 * Change name of a device, can pass format strings "eth%d".
712 int dev_change_name(struct net_device *dev, char *newname)
718 if (dev->flags & IFF_UP)
721 if (!dev_valid_name(newname))
724 if (strchr(newname, '%')) {
725 err = dev_alloc_name(dev, newname);
728 strcpy(newname, dev->name);
730 else if (__dev_get_by_name(newname))
733 strlcpy(dev->name, newname, IFNAMSIZ);
735 err = class_device_rename(&dev->class_dev, dev->name);
737 hlist_del(&dev->name_hlist);
738 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
739 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
746 * netdev_features_change - device changes fatures
747 * @dev: device to cause notification
749 * Called to indicate a device has changed features.
751 void netdev_features_change(struct net_device *dev)
753 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
755 EXPORT_SYMBOL(netdev_features_change);
758 * netdev_state_change - device changes state
759 * @dev: device to cause notification
761 * Called to indicate a device has changed state. This function calls
762 * the notifier chains for netdev_chain and sends a NEWLINK message
763 * to the routing socket.
765 void netdev_state_change(struct net_device *dev)
767 if (dev->flags & IFF_UP) {
768 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
769 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
774 * dev_load - load a network module
775 * @name: name of interface
777 * If a network interface is not present and the process has suitable
778 * privileges this function loads the module. If module loading is not
779 * available in this kernel then it becomes a nop.
782 void dev_load(const char *name)
784 struct net_device *dev;
786 read_lock(&dev_base_lock);
787 dev = __dev_get_by_name(name);
788 read_unlock(&dev_base_lock);
790 if (!dev && capable(CAP_SYS_MODULE))
791 request_module("%s", name);
794 static int default_rebuild_header(struct sk_buff *skb)
796 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
797 skb->dev ? skb->dev->name : "NULL!!!");
804 * dev_open - prepare an interface for use.
805 * @dev: device to open
807 * Takes a device from down to up state. The device's private open
808 * function is invoked and then the multicast lists are loaded. Finally
809 * the device is moved into the up state and a %NETDEV_UP message is
810 * sent to the netdev notifier chain.
812 * Calling this function on an active interface is a nop. On a failure
813 * a negative errno code is returned.
815 int dev_open(struct net_device *dev)
823 if (dev->flags & IFF_UP)
827 * Is it even present?
829 if (!netif_device_present(dev))
833 * Call device private open method
835 set_bit(__LINK_STATE_START, &dev->state);
837 ret = dev->open(dev);
839 clear_bit(__LINK_STATE_START, &dev->state);
843 * If it went open OK then:
850 dev->flags |= IFF_UP;
853 * Initialize multicasting status
858 * Wakeup transmit queue engine
863 * ... and announce new interface.
865 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
871 * dev_close - shutdown an interface.
872 * @dev: device to shutdown
874 * This function moves an active device into down state. A
875 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
876 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
879 int dev_close(struct net_device *dev)
881 if (!(dev->flags & IFF_UP))
885 * Tell people we are going down, so that they can
886 * prepare to death, when device is still operating.
888 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
892 clear_bit(__LINK_STATE_START, &dev->state);
894 /* Synchronize to scheduled poll. We cannot touch poll list,
895 * it can be even on different cpu. So just clear netif_running(),
896 * and wait when poll really will happen. Actually, the best place
897 * for this is inside dev->stop() after device stopped its irq
898 * engine, but this requires more changes in devices. */
900 smp_mb__after_clear_bit(); /* Commit netif_running(). */
901 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
907 * Call the device specific close. This cannot fail.
908 * Only if device is UP
910 * We allow it to be called even after a DETACH hot-plug
917 * Device is now down.
920 dev->flags &= ~IFF_UP;
923 * Tell people we are down
925 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
932 * Device change register/unregister. These are not inline or static
933 * as we export them to the world.
937 * register_netdevice_notifier - register a network notifier block
940 * Register a notifier to be called when network device events occur.
941 * The notifier passed is linked into the kernel structures and must
942 * not be reused until it has been unregistered. A negative errno code
943 * is returned on a failure.
945 * When registered all registration and up events are replayed
946 * to the new notifier to allow device to have a race free
947 * view of the network device list.
950 int register_netdevice_notifier(struct notifier_block *nb)
952 struct net_device *dev;
956 err = notifier_chain_register(&netdev_chain, nb);
958 for (dev = dev_base; dev; dev = dev->next) {
959 nb->notifier_call(nb, NETDEV_REGISTER, dev);
961 if (dev->flags & IFF_UP)
962 nb->notifier_call(nb, NETDEV_UP, dev);
970 * unregister_netdevice_notifier - unregister a network notifier block
973 * Unregister a notifier previously registered by
974 * register_netdevice_notifier(). The notifier is unlinked into the
975 * kernel structures and may then be reused. A negative errno code
976 * is returned on a failure.
979 int unregister_netdevice_notifier(struct notifier_block *nb)
981 return notifier_chain_unregister(&netdev_chain, nb);
985 * call_netdevice_notifiers - call all network notifier blocks
986 * @val: value passed unmodified to notifier function
987 * @v: pointer passed unmodified to notifier function
989 * Call all network notifier blocks. Parameters and return value
990 * are as for notifier_call_chain().
993 int call_netdevice_notifiers(unsigned long val, void *v)
995 return notifier_call_chain(&netdev_chain, val, v);
998 /* When > 0 there are consumers of rx skb time stamps */
999 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1001 void net_enable_timestamp(void)
1003 atomic_inc(&netstamp_needed);
1006 void net_disable_timestamp(void)
1008 atomic_dec(&netstamp_needed);
1011 void __net_timestamp(struct sk_buff *skb)
1015 do_gettimeofday(&tv);
1016 skb_set_timestamp(skb, &tv);
1018 EXPORT_SYMBOL(__net_timestamp);
1020 static inline void net_timestamp(struct sk_buff *skb)
1022 if (atomic_read(&netstamp_needed))
1023 __net_timestamp(skb);
1025 skb->tstamp.off_sec = 0;
1026 skb->tstamp.off_usec = 0;
1031 * Support routine. Sends outgoing frames to any network
1032 * taps currently in use.
1035 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1037 struct packet_type *ptype;
1042 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1043 /* Never send packets back to the socket
1044 * they originated from - MvS (miquels@drinkel.ow.org)
1046 if ((ptype->dev == dev || !ptype->dev) &&
1047 (ptype->af_packet_priv == NULL ||
1048 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1049 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1053 /* skb->nh should be correctly
1054 set by sender, so that the second statement is
1055 just protection against buggy protocols.
1057 skb2->mac.raw = skb2->data;
1059 if (skb2->nh.raw < skb2->data ||
1060 skb2->nh.raw > skb2->tail) {
1061 if (net_ratelimit())
1062 printk(KERN_CRIT "protocol %04x is "
1064 skb2->protocol, dev->name);
1065 skb2->nh.raw = skb2->data;
1068 skb2->h.raw = skb2->nh.raw;
1069 skb2->pkt_type = PACKET_OUTGOING;
1070 ptype->func(skb2, skb->dev, ptype, skb->dev);
1077 * Invalidate hardware checksum when packet is to be mangled, and
1078 * complete checksum manually on outgoing path.
1080 int skb_checksum_help(struct sk_buff *skb, int inward)
1083 int ret = 0, offset = skb->h.raw - skb->data;
1086 skb->ip_summed = CHECKSUM_NONE;
1090 if (skb_cloned(skb)) {
1091 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1096 BUG_ON(offset > (int)skb->len);
1097 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1099 offset = skb->tail - skb->h.raw;
1100 BUG_ON(offset <= 0);
1101 BUG_ON(skb->csum + 2 > offset);
1103 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1104 skb->ip_summed = CHECKSUM_NONE;
1109 /* Take action when hardware reception checksum errors are detected. */
1111 void netdev_rx_csum_fault(struct net_device *dev)
1113 if (net_ratelimit()) {
1114 printk(KERN_ERR "%s: hw csum failure.\n",
1115 dev ? dev->name : "<unknown>");
1119 EXPORT_SYMBOL(netdev_rx_csum_fault);
1122 #ifdef CONFIG_HIGHMEM
1123 /* Actually, we should eliminate this check as soon as we know, that:
1124 * 1. IOMMU is present and allows to map all the memory.
1125 * 2. No high memory really exists on this machine.
1128 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1132 if (dev->features & NETIF_F_HIGHDMA)
1135 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1136 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1142 #define illegal_highdma(dev, skb) (0)
1145 /* Keep head the same: replace data */
1146 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1151 struct skb_shared_info *ninfo;
1152 int headerlen = skb->data - skb->head;
1153 int expand = (skb->tail + skb->data_len) - skb->end;
1155 if (skb_shared(skb))
1161 size = skb->end - skb->head + expand;
1162 size = SKB_DATA_ALIGN(size);
1163 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1167 /* Copy entire thing */
1168 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1172 ninfo = (struct skb_shared_info*)(data + size);
1173 atomic_set(&ninfo->dataref, 1);
1174 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1175 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1176 ninfo->nr_frags = 0;
1177 ninfo->frag_list = NULL;
1179 /* Offset between the two in bytes */
1180 offset = data - skb->head;
1182 /* Free old data. */
1183 skb_release_data(skb);
1186 skb->end = data + size;
1188 /* Set up new pointers */
1189 skb->h.raw += offset;
1190 skb->nh.raw += offset;
1191 skb->mac.raw += offset;
1192 skb->tail += offset;
1193 skb->data += offset;
1195 /* We are no longer a clone, even if we were. */
1198 skb->tail += skb->data_len;
1203 #define HARD_TX_LOCK(dev, cpu) { \
1204 if ((dev->features & NETIF_F_LLTX) == 0) { \
1205 spin_lock(&dev->xmit_lock); \
1206 dev->xmit_lock_owner = cpu; \
1210 #define HARD_TX_UNLOCK(dev) { \
1211 if ((dev->features & NETIF_F_LLTX) == 0) { \
1212 dev->xmit_lock_owner = -1; \
1213 spin_unlock(&dev->xmit_lock); \
1218 * dev_queue_xmit - transmit a buffer
1219 * @skb: buffer to transmit
1221 * Queue a buffer for transmission to a network device. The caller must
1222 * have set the device and priority and built the buffer before calling
1223 * this function. The function can be called from an interrupt.
1225 * A negative errno code is returned on a failure. A success does not
1226 * guarantee the frame will be transmitted as it may be dropped due
1227 * to congestion or traffic shaping.
1229 * -----------------------------------------------------------------------------------
1230 * I notice this method can also return errors from the queue disciplines,
1231 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1234 * Regardless of the return value, the skb is consumed, so it is currently
1235 * difficult to retry a send to this method. (You can bump the ref count
1236 * before sending to hold a reference for retry if you are careful.)
1238 * When calling this method, interrupts MUST be enabled. This is because
1239 * the BH enable code must have IRQs enabled so that it will not deadlock.
1243 int dev_queue_xmit(struct sk_buff *skb)
1245 struct net_device *dev = skb->dev;
1249 if (skb_shinfo(skb)->frag_list &&
1250 !(dev->features & NETIF_F_FRAGLIST) &&
1251 __skb_linearize(skb, GFP_ATOMIC))
1254 /* Fragmented skb is linearized if device does not support SG,
1255 * or if at least one of fragments is in highmem and device
1256 * does not support DMA from it.
1258 if (skb_shinfo(skb)->nr_frags &&
1259 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1260 __skb_linearize(skb, GFP_ATOMIC))
1263 /* If packet is not checksummed and device does not support
1264 * checksumming for this protocol, complete checksumming here.
1266 if (skb->ip_summed == CHECKSUM_HW &&
1267 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1268 (!(dev->features & NETIF_F_IP_CSUM) ||
1269 skb->protocol != htons(ETH_P_IP))))
1270 if (skb_checksum_help(skb, 0))
1273 spin_lock_prefetch(&dev->queue_lock);
1275 /* Disable soft irqs for various locks below. Also
1276 * stops preemption for RCU.
1280 /* Updates of qdisc are serialized by queue_lock.
1281 * The struct Qdisc which is pointed to by qdisc is now a
1282 * rcu structure - it may be accessed without acquiring
1283 * a lock (but the structure may be stale.) The freeing of the
1284 * qdisc will be deferred until it's known that there are no
1285 * more references to it.
1287 * If the qdisc has an enqueue function, we still need to
1288 * hold the queue_lock before calling it, since queue_lock
1289 * also serializes access to the device queue.
1292 q = rcu_dereference(dev->qdisc);
1293 #ifdef CONFIG_NET_CLS_ACT
1294 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1297 /* Grab device queue */
1298 spin_lock(&dev->queue_lock);
1300 rc = q->enqueue(skb, q);
1304 spin_unlock(&dev->queue_lock);
1305 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1309 /* The device has no queue. Common case for software devices:
1310 loopback, all the sorts of tunnels...
1312 Really, it is unlikely that xmit_lock protection is necessary here.
1313 (f.e. loopback and IP tunnels are clean ignoring statistics
1315 However, it is possible, that they rely on protection
1318 Check this and shot the lock. It is not prone from deadlocks.
1319 Either shot noqueue qdisc, it is even simpler 8)
1321 if (dev->flags & IFF_UP) {
1322 int cpu = smp_processor_id(); /* ok because BHs are off */
1324 if (dev->xmit_lock_owner != cpu) {
1326 HARD_TX_LOCK(dev, cpu);
1328 if (!netif_queue_stopped(dev)) {
1330 dev_queue_xmit_nit(skb, dev);
1333 if (!dev->hard_start_xmit(skb, dev)) {
1334 HARD_TX_UNLOCK(dev);
1338 HARD_TX_UNLOCK(dev);
1339 if (net_ratelimit())
1340 printk(KERN_CRIT "Virtual device %s asks to "
1341 "queue packet!\n", dev->name);
1343 /* Recursion is detected! It is possible,
1345 if (net_ratelimit())
1346 printk(KERN_CRIT "Dead loop on virtual device "
1347 "%s, fix it urgently!\n", dev->name);
1363 /*=======================================================================
1365 =======================================================================*/
1367 int netdev_max_backlog = 1000;
1368 int netdev_budget = 300;
1369 int weight_p = 64; /* old backlog weight */
1371 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1375 * netif_rx - post buffer to the network code
1376 * @skb: buffer to post
1378 * This function receives a packet from a device driver and queues it for
1379 * the upper (protocol) levels to process. It always succeeds. The buffer
1380 * may be dropped during processing for congestion control or by the
1384 * NET_RX_SUCCESS (no congestion)
1385 * NET_RX_CN_LOW (low congestion)
1386 * NET_RX_CN_MOD (moderate congestion)
1387 * NET_RX_CN_HIGH (high congestion)
1388 * NET_RX_DROP (packet was dropped)
1392 int netif_rx(struct sk_buff *skb)
1394 struct softnet_data *queue;
1395 unsigned long flags;
1397 /* if netpoll wants it, pretend we never saw it */
1398 if (netpoll_rx(skb))
1401 if (!skb->tstamp.off_sec)
1405 * The code is rearranged so that the path is the most
1406 * short when CPU is congested, but is still operating.
1408 local_irq_save(flags);
1409 queue = &__get_cpu_var(softnet_data);
1411 __get_cpu_var(netdev_rx_stat).total++;
1412 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1413 if (queue->input_pkt_queue.qlen) {
1416 __skb_queue_tail(&queue->input_pkt_queue, skb);
1417 local_irq_restore(flags);
1418 return NET_RX_SUCCESS;
1421 netif_rx_schedule(&queue->backlog_dev);
1425 __get_cpu_var(netdev_rx_stat).dropped++;
1426 local_irq_restore(flags);
1432 int netif_rx_ni(struct sk_buff *skb)
1437 err = netif_rx(skb);
1438 if (local_softirq_pending())
1445 EXPORT_SYMBOL(netif_rx_ni);
1447 static inline struct net_device *skb_bond(struct sk_buff *skb)
1449 struct net_device *dev = skb->dev;
1452 skb->dev = dev->master;
1457 static void net_tx_action(struct softirq_action *h)
1459 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1461 if (sd->completion_queue) {
1462 struct sk_buff *clist;
1464 local_irq_disable();
1465 clist = sd->completion_queue;
1466 sd->completion_queue = NULL;
1470 struct sk_buff *skb = clist;
1471 clist = clist->next;
1473 BUG_TRAP(!atomic_read(&skb->users));
1478 if (sd->output_queue) {
1479 struct net_device *head;
1481 local_irq_disable();
1482 head = sd->output_queue;
1483 sd->output_queue = NULL;
1487 struct net_device *dev = head;
1488 head = head->next_sched;
1490 smp_mb__before_clear_bit();
1491 clear_bit(__LINK_STATE_SCHED, &dev->state);
1493 if (spin_trylock(&dev->queue_lock)) {
1495 spin_unlock(&dev->queue_lock);
1497 netif_schedule(dev);
1503 static __inline__ int deliver_skb(struct sk_buff *skb,
1504 struct packet_type *pt_prev,
1505 struct net_device *orig_dev)
1507 atomic_inc(&skb->users);
1508 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1511 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1512 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1514 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1515 unsigned char *addr);
1516 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1518 static __inline__ int handle_bridge(struct sk_buff **pskb,
1519 struct packet_type **pt_prev, int *ret,
1520 struct net_device *orig_dev)
1522 struct net_bridge_port *port;
1524 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1525 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1529 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1533 return br_handle_frame_hook(port, pskb);
1536 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1539 #ifdef CONFIG_NET_CLS_ACT
1540 /* TODO: Maybe we should just force sch_ingress to be compiled in
1541 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1542 * a compare and 2 stores extra right now if we dont have it on
1543 * but have CONFIG_NET_CLS_ACT
1544 * NOTE: This doesnt stop any functionality; if you dont have
1545 * the ingress scheduler, you just cant add policies on ingress.
1548 static int ing_filter(struct sk_buff *skb)
1551 struct net_device *dev = skb->dev;
1552 int result = TC_ACT_OK;
1554 if (dev->qdisc_ingress) {
1555 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1556 if (MAX_RED_LOOP < ttl++) {
1557 printk("Redir loop detected Dropping packet (%s->%s)\n",
1558 skb->input_dev->name, skb->dev->name);
1562 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1564 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1566 spin_lock(&dev->ingress_lock);
1567 if ((q = dev->qdisc_ingress) != NULL)
1568 result = q->enqueue(skb, q);
1569 spin_unlock(&dev->ingress_lock);
1577 int netif_receive_skb(struct sk_buff *skb)
1579 struct packet_type *ptype, *pt_prev;
1580 struct net_device *orig_dev;
1581 int ret = NET_RX_DROP;
1582 unsigned short type;
1584 /* if we've gotten here through NAPI, check netpoll */
1585 if (skb->dev->poll && netpoll_rx(skb))
1588 if (!skb->tstamp.off_sec)
1591 if (!skb->input_dev)
1592 skb->input_dev = skb->dev;
1594 orig_dev = skb_bond(skb);
1596 __get_cpu_var(netdev_rx_stat).total++;
1598 skb->h.raw = skb->nh.raw = skb->data;
1599 skb->mac_len = skb->nh.raw - skb->mac.raw;
1605 #ifdef CONFIG_NET_CLS_ACT
1606 if (skb->tc_verd & TC_NCLS) {
1607 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1612 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1613 if (!ptype->dev || ptype->dev == skb->dev) {
1615 ret = deliver_skb(skb, pt_prev, orig_dev);
1620 #ifdef CONFIG_NET_CLS_ACT
1622 ret = deliver_skb(skb, pt_prev, orig_dev);
1623 pt_prev = NULL; /* noone else should process this after*/
1625 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1628 ret = ing_filter(skb);
1630 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1639 handle_diverter(skb);
1641 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1644 type = skb->protocol;
1645 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1646 if (ptype->type == type &&
1647 (!ptype->dev || ptype->dev == skb->dev)) {
1649 ret = deliver_skb(skb, pt_prev, orig_dev);
1655 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1658 /* Jamal, now you will not able to escape explaining
1659 * me how you were going to use this. :-)
1669 static int process_backlog(struct net_device *backlog_dev, int *budget)
1672 int quota = min(backlog_dev->quota, *budget);
1673 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1674 unsigned long start_time = jiffies;
1676 backlog_dev->weight = weight_p;
1678 struct sk_buff *skb;
1679 struct net_device *dev;
1681 local_irq_disable();
1682 skb = __skb_dequeue(&queue->input_pkt_queue);
1689 netif_receive_skb(skb);
1695 if (work >= quota || jiffies - start_time > 1)
1700 backlog_dev->quota -= work;
1705 backlog_dev->quota -= work;
1708 list_del(&backlog_dev->poll_list);
1709 smp_mb__before_clear_bit();
1710 netif_poll_enable(backlog_dev);
1716 static void net_rx_action(struct softirq_action *h)
1718 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1719 unsigned long start_time = jiffies;
1720 int budget = netdev_budget;
1723 local_irq_disable();
1725 while (!list_empty(&queue->poll_list)) {
1726 struct net_device *dev;
1728 if (budget <= 0 || jiffies - start_time > 1)
1733 dev = list_entry(queue->poll_list.next,
1734 struct net_device, poll_list);
1735 have = netpoll_poll_lock(dev);
1737 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1738 netpoll_poll_unlock(have);
1739 local_irq_disable();
1740 list_del(&dev->poll_list);
1741 list_add_tail(&dev->poll_list, &queue->poll_list);
1743 dev->quota += dev->weight;
1745 dev->quota = dev->weight;
1747 netpoll_poll_unlock(have);
1749 local_irq_disable();
1757 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1758 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1762 static gifconf_func_t * gifconf_list [NPROTO];
1765 * register_gifconf - register a SIOCGIF handler
1766 * @family: Address family
1767 * @gifconf: Function handler
1769 * Register protocol dependent address dumping routines. The handler
1770 * that is passed must not be freed or reused until it has been replaced
1771 * by another handler.
1773 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1775 if (family >= NPROTO)
1777 gifconf_list[family] = gifconf;
1783 * Map an interface index to its name (SIOCGIFNAME)
1787 * We need this ioctl for efficient implementation of the
1788 * if_indextoname() function required by the IPv6 API. Without
1789 * it, we would have to search all the interfaces to find a
1793 static int dev_ifname(struct ifreq __user *arg)
1795 struct net_device *dev;
1799 * Fetch the caller's info block.
1802 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1805 read_lock(&dev_base_lock);
1806 dev = __dev_get_by_index(ifr.ifr_ifindex);
1808 read_unlock(&dev_base_lock);
1812 strcpy(ifr.ifr_name, dev->name);
1813 read_unlock(&dev_base_lock);
1815 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1821 * Perform a SIOCGIFCONF call. This structure will change
1822 * size eventually, and there is nothing I can do about it.
1823 * Thus we will need a 'compatibility mode'.
1826 static int dev_ifconf(char __user *arg)
1829 struct net_device *dev;
1836 * Fetch the caller's info block.
1839 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1846 * Loop over the interfaces, and write an info block for each.
1850 for (dev = dev_base; dev; dev = dev->next) {
1851 for (i = 0; i < NPROTO; i++) {
1852 if (gifconf_list[i]) {
1855 done = gifconf_list[i](dev, NULL, 0);
1857 done = gifconf_list[i](dev, pos + total,
1867 * All done. Write the updated control block back to the caller.
1869 ifc.ifc_len = total;
1872 * Both BSD and Solaris return 0 here, so we do too.
1874 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1877 #ifdef CONFIG_PROC_FS
1879 * This is invoked by the /proc filesystem handler to display a device
1882 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1884 struct net_device *dev;
1887 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1889 return i == pos ? dev : NULL;
1892 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1894 read_lock(&dev_base_lock);
1895 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1898 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1901 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1904 void dev_seq_stop(struct seq_file *seq, void *v)
1906 read_unlock(&dev_base_lock);
1909 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1911 if (dev->get_stats) {
1912 struct net_device_stats *stats = dev->get_stats(dev);
1914 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1915 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1916 dev->name, stats->rx_bytes, stats->rx_packets,
1918 stats->rx_dropped + stats->rx_missed_errors,
1919 stats->rx_fifo_errors,
1920 stats->rx_length_errors + stats->rx_over_errors +
1921 stats->rx_crc_errors + stats->rx_frame_errors,
1922 stats->rx_compressed, stats->multicast,
1923 stats->tx_bytes, stats->tx_packets,
1924 stats->tx_errors, stats->tx_dropped,
1925 stats->tx_fifo_errors, stats->collisions,
1926 stats->tx_carrier_errors +
1927 stats->tx_aborted_errors +
1928 stats->tx_window_errors +
1929 stats->tx_heartbeat_errors,
1930 stats->tx_compressed);
1932 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1936 * Called from the PROCfs module. This now uses the new arbitrary sized
1937 * /proc/net interface to create /proc/net/dev
1939 static int dev_seq_show(struct seq_file *seq, void *v)
1941 if (v == SEQ_START_TOKEN)
1942 seq_puts(seq, "Inter-| Receive "
1944 " face |bytes packets errs drop fifo frame "
1945 "compressed multicast|bytes packets errs "
1946 "drop fifo colls carrier compressed\n");
1948 dev_seq_printf_stats(seq, v);
1952 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1954 struct netif_rx_stats *rc = NULL;
1956 while (*pos < NR_CPUS)
1957 if (cpu_online(*pos)) {
1958 rc = &per_cpu(netdev_rx_stat, *pos);
1965 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1967 return softnet_get_online(pos);
1970 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1973 return softnet_get_online(pos);
1976 static void softnet_seq_stop(struct seq_file *seq, void *v)
1980 static int softnet_seq_show(struct seq_file *seq, void *v)
1982 struct netif_rx_stats *s = v;
1984 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
1985 s->total, s->dropped, s->time_squeeze, 0,
1986 0, 0, 0, 0, /* was fastroute */
1991 static struct seq_operations dev_seq_ops = {
1992 .start = dev_seq_start,
1993 .next = dev_seq_next,
1994 .stop = dev_seq_stop,
1995 .show = dev_seq_show,
1998 static int dev_seq_open(struct inode *inode, struct file *file)
2000 return seq_open(file, &dev_seq_ops);
2003 static struct file_operations dev_seq_fops = {
2004 .owner = THIS_MODULE,
2005 .open = dev_seq_open,
2007 .llseek = seq_lseek,
2008 .release = seq_release,
2011 static struct seq_operations softnet_seq_ops = {
2012 .start = softnet_seq_start,
2013 .next = softnet_seq_next,
2014 .stop = softnet_seq_stop,
2015 .show = softnet_seq_show,
2018 static int softnet_seq_open(struct inode *inode, struct file *file)
2020 return seq_open(file, &softnet_seq_ops);
2023 static struct file_operations softnet_seq_fops = {
2024 .owner = THIS_MODULE,
2025 .open = softnet_seq_open,
2027 .llseek = seq_lseek,
2028 .release = seq_release,
2032 extern int wireless_proc_init(void);
2034 #define wireless_proc_init() 0
2037 static int __init dev_proc_init(void)
2041 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2043 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2045 if (wireless_proc_init())
2051 proc_net_remove("softnet_stat");
2053 proc_net_remove("dev");
2057 #define dev_proc_init() 0
2058 #endif /* CONFIG_PROC_FS */
2062 * netdev_set_master - set up master/slave pair
2063 * @slave: slave device
2064 * @master: new master device
2066 * Changes the master device of the slave. Pass %NULL to break the
2067 * bonding. The caller must hold the RTNL semaphore. On a failure
2068 * a negative errno code is returned. On success the reference counts
2069 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2070 * function returns zero.
2072 int netdev_set_master(struct net_device *slave, struct net_device *master)
2074 struct net_device *old = slave->master;
2084 slave->master = master;
2092 slave->flags |= IFF_SLAVE;
2094 slave->flags &= ~IFF_SLAVE;
2096 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2101 * dev_set_promiscuity - update promiscuity count on a device
2105 * Add or remove promsicuity from a device. While the count in the device
2106 * remains above zero the interface remains promiscuous. Once it hits zero
2107 * the device reverts back to normal filtering operation. A negative inc
2108 * value is used to drop promiscuity on the device.
2110 void dev_set_promiscuity(struct net_device *dev, int inc)
2112 unsigned short old_flags = dev->flags;
2114 if ((dev->promiscuity += inc) == 0)
2115 dev->flags &= ~IFF_PROMISC;
2117 dev->flags |= IFF_PROMISC;
2118 if (dev->flags != old_flags) {
2120 printk(KERN_INFO "device %s %s promiscuous mode\n",
2121 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2127 * dev_set_allmulti - update allmulti count on a device
2131 * Add or remove reception of all multicast frames to a device. While the
2132 * count in the device remains above zero the interface remains listening
2133 * to all interfaces. Once it hits zero the device reverts back to normal
2134 * filtering operation. A negative @inc value is used to drop the counter
2135 * when releasing a resource needing all multicasts.
2138 void dev_set_allmulti(struct net_device *dev, int inc)
2140 unsigned short old_flags = dev->flags;
2142 dev->flags |= IFF_ALLMULTI;
2143 if ((dev->allmulti += inc) == 0)
2144 dev->flags &= ~IFF_ALLMULTI;
2145 if (dev->flags ^ old_flags)
2149 unsigned dev_get_flags(const struct net_device *dev)
2153 flags = (dev->flags & ~(IFF_PROMISC |
2156 (dev->gflags & (IFF_PROMISC |
2159 if (netif_running(dev) && netif_carrier_ok(dev))
2160 flags |= IFF_RUNNING;
2165 int dev_change_flags(struct net_device *dev, unsigned flags)
2168 int old_flags = dev->flags;
2171 * Set the flags on our device.
2174 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2175 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2177 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2181 * Load in the correct multicast list now the flags have changed.
2187 * Have we downed the interface. We handle IFF_UP ourselves
2188 * according to user attempts to set it, rather than blindly
2193 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2194 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2200 if (dev->flags & IFF_UP &&
2201 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2203 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2205 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2206 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2207 dev->gflags ^= IFF_PROMISC;
2208 dev_set_promiscuity(dev, inc);
2211 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2212 is important. Some (broken) drivers set IFF_PROMISC, when
2213 IFF_ALLMULTI is requested not asking us and not reporting.
2215 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2216 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2217 dev->gflags ^= IFF_ALLMULTI;
2218 dev_set_allmulti(dev, inc);
2221 if (old_flags ^ dev->flags)
2222 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2227 int dev_set_mtu(struct net_device *dev, int new_mtu)
2231 if (new_mtu == dev->mtu)
2234 /* MTU must be positive. */
2238 if (!netif_device_present(dev))
2242 if (dev->change_mtu)
2243 err = dev->change_mtu(dev, new_mtu);
2246 if (!err && dev->flags & IFF_UP)
2247 notifier_call_chain(&netdev_chain,
2248 NETDEV_CHANGEMTU, dev);
2252 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2256 if (!dev->set_mac_address)
2258 if (sa->sa_family != dev->type)
2260 if (!netif_device_present(dev))
2262 err = dev->set_mac_address(dev, sa);
2264 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2269 * Perform the SIOCxIFxxx calls.
2271 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2274 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2280 case SIOCGIFFLAGS: /* Get interface flags */
2281 ifr->ifr_flags = dev_get_flags(dev);
2284 case SIOCSIFFLAGS: /* Set interface flags */
2285 return dev_change_flags(dev, ifr->ifr_flags);
2287 case SIOCGIFMETRIC: /* Get the metric on the interface
2288 (currently unused) */
2289 ifr->ifr_metric = 0;
2292 case SIOCSIFMETRIC: /* Set the metric on the interface
2293 (currently unused) */
2296 case SIOCGIFMTU: /* Get the MTU of a device */
2297 ifr->ifr_mtu = dev->mtu;
2300 case SIOCSIFMTU: /* Set the MTU of a device */
2301 return dev_set_mtu(dev, ifr->ifr_mtu);
2305 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2307 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2308 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2309 ifr->ifr_hwaddr.sa_family = dev->type;
2313 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2315 case SIOCSIFHWBROADCAST:
2316 if (ifr->ifr_hwaddr.sa_family != dev->type)
2318 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2319 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2320 notifier_call_chain(&netdev_chain,
2321 NETDEV_CHANGEADDR, dev);
2325 ifr->ifr_map.mem_start = dev->mem_start;
2326 ifr->ifr_map.mem_end = dev->mem_end;
2327 ifr->ifr_map.base_addr = dev->base_addr;
2328 ifr->ifr_map.irq = dev->irq;
2329 ifr->ifr_map.dma = dev->dma;
2330 ifr->ifr_map.port = dev->if_port;
2334 if (dev->set_config) {
2335 if (!netif_device_present(dev))
2337 return dev->set_config(dev, &ifr->ifr_map);
2342 if (!dev->set_multicast_list ||
2343 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2345 if (!netif_device_present(dev))
2347 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2351 if (!dev->set_multicast_list ||
2352 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2354 if (!netif_device_present(dev))
2356 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2360 ifr->ifr_ifindex = dev->ifindex;
2364 ifr->ifr_qlen = dev->tx_queue_len;
2368 if (ifr->ifr_qlen < 0)
2370 dev->tx_queue_len = ifr->ifr_qlen;
2374 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2375 return dev_change_name(dev, ifr->ifr_newname);
2378 * Unknown or private ioctl
2382 if ((cmd >= SIOCDEVPRIVATE &&
2383 cmd <= SIOCDEVPRIVATE + 15) ||
2384 cmd == SIOCBONDENSLAVE ||
2385 cmd == SIOCBONDRELEASE ||
2386 cmd == SIOCBONDSETHWADDR ||
2387 cmd == SIOCBONDSLAVEINFOQUERY ||
2388 cmd == SIOCBONDINFOQUERY ||
2389 cmd == SIOCBONDCHANGEACTIVE ||
2390 cmd == SIOCGMIIPHY ||
2391 cmd == SIOCGMIIREG ||
2392 cmd == SIOCSMIIREG ||
2393 cmd == SIOCBRADDIF ||
2394 cmd == SIOCBRDELIF ||
2395 cmd == SIOCWANDEV) {
2397 if (dev->do_ioctl) {
2398 if (netif_device_present(dev))
2399 err = dev->do_ioctl(dev, ifr,
2412 * This function handles all "interface"-type I/O control requests. The actual
2413 * 'doing' part of this is dev_ifsioc above.
2417 * dev_ioctl - network device ioctl
2418 * @cmd: command to issue
2419 * @arg: pointer to a struct ifreq in user space
2421 * Issue ioctl functions to devices. This is normally called by the
2422 * user space syscall interfaces but can sometimes be useful for
2423 * other purposes. The return value is the return from the syscall if
2424 * positive or a negative errno code on error.
2427 int dev_ioctl(unsigned int cmd, void __user *arg)
2433 /* One special case: SIOCGIFCONF takes ifconf argument
2434 and requires shared lock, because it sleeps writing
2438 if (cmd == SIOCGIFCONF) {
2440 ret = dev_ifconf((char __user *) arg);
2444 if (cmd == SIOCGIFNAME)
2445 return dev_ifname((struct ifreq __user *)arg);
2447 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2450 ifr.ifr_name[IFNAMSIZ-1] = 0;
2452 colon = strchr(ifr.ifr_name, ':');
2457 * See which interface the caller is talking about.
2462 * These ioctl calls:
2463 * - can be done by all.
2464 * - atomic and do not require locking.
2475 dev_load(ifr.ifr_name);
2476 read_lock(&dev_base_lock);
2477 ret = dev_ifsioc(&ifr, cmd);
2478 read_unlock(&dev_base_lock);
2482 if (copy_to_user(arg, &ifr,
2483 sizeof(struct ifreq)))
2489 dev_load(ifr.ifr_name);
2491 ret = dev_ethtool(&ifr);
2496 if (copy_to_user(arg, &ifr,
2497 sizeof(struct ifreq)))
2503 * These ioctl calls:
2504 * - require superuser power.
2505 * - require strict serialization.
2511 if (!capable(CAP_NET_ADMIN))
2513 dev_load(ifr.ifr_name);
2515 ret = dev_ifsioc(&ifr, cmd);
2520 if (copy_to_user(arg, &ifr,
2521 sizeof(struct ifreq)))
2527 * These ioctl calls:
2528 * - require superuser power.
2529 * - require strict serialization.
2530 * - do not return a value
2540 case SIOCSIFHWBROADCAST:
2543 case SIOCBONDENSLAVE:
2544 case SIOCBONDRELEASE:
2545 case SIOCBONDSETHWADDR:
2546 case SIOCBONDCHANGEACTIVE:
2549 if (!capable(CAP_NET_ADMIN))
2552 case SIOCBONDSLAVEINFOQUERY:
2553 case SIOCBONDINFOQUERY:
2554 dev_load(ifr.ifr_name);
2556 ret = dev_ifsioc(&ifr, cmd);
2561 /* Get the per device memory space. We can add this but
2562 * currently do not support it */
2564 /* Set the per device memory buffer space.
2565 * Not applicable in our case */
2570 * Unknown or private ioctl.
2573 if (cmd == SIOCWANDEV ||
2574 (cmd >= SIOCDEVPRIVATE &&
2575 cmd <= SIOCDEVPRIVATE + 15)) {
2576 dev_load(ifr.ifr_name);
2578 ret = dev_ifsioc(&ifr, cmd);
2580 if (!ret && copy_to_user(arg, &ifr,
2581 sizeof(struct ifreq)))
2586 /* Take care of Wireless Extensions */
2587 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2588 /* If command is `set a parameter', or
2589 * `get the encoding parameters', check if
2590 * the user has the right to do it */
2591 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2592 if (!capable(CAP_NET_ADMIN))
2595 dev_load(ifr.ifr_name);
2597 /* Follow me in net/core/wireless.c */
2598 ret = wireless_process_ioctl(&ifr, cmd);
2600 if (IW_IS_GET(cmd) &&
2601 copy_to_user(arg, &ifr,
2602 sizeof(struct ifreq)))
2606 #endif /* WIRELESS_EXT */
2613 * dev_new_index - allocate an ifindex
2615 * Returns a suitable unique value for a new device interface
2616 * number. The caller must hold the rtnl semaphore or the
2617 * dev_base_lock to be sure it remains unique.
2619 static int dev_new_index(void)
2625 if (!__dev_get_by_index(ifindex))
2630 static int dev_boot_phase = 1;
2632 /* Delayed registration/unregisteration */
2633 static DEFINE_SPINLOCK(net_todo_list_lock);
2634 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2636 static inline void net_set_todo(struct net_device *dev)
2638 spin_lock(&net_todo_list_lock);
2639 list_add_tail(&dev->todo_list, &net_todo_list);
2640 spin_unlock(&net_todo_list_lock);
2644 * register_netdevice - register a network device
2645 * @dev: device to register
2647 * Take a completed network device structure and add it to the kernel
2648 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2649 * chain. 0 is returned on success. A negative errno code is returned
2650 * on a failure to set up the device, or if the name is a duplicate.
2652 * Callers must hold the rtnl semaphore. You may want
2653 * register_netdev() instead of this.
2656 * The locking appears insufficient to guarantee two parallel registers
2657 * will not get the same name.
2660 int register_netdevice(struct net_device *dev)
2662 struct hlist_head *head;
2663 struct hlist_node *p;
2666 BUG_ON(dev_boot_phase);
2669 /* When net_device's are persistent, this will be fatal. */
2670 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2672 spin_lock_init(&dev->queue_lock);
2673 spin_lock_init(&dev->xmit_lock);
2674 dev->xmit_lock_owner = -1;
2675 #ifdef CONFIG_NET_CLS_ACT
2676 spin_lock_init(&dev->ingress_lock);
2679 ret = alloc_divert_blk(dev);
2685 /* Init, if this function is available */
2687 ret = dev->init(dev);
2695 if (!dev_valid_name(dev->name)) {
2700 dev->ifindex = dev_new_index();
2701 if (dev->iflink == -1)
2702 dev->iflink = dev->ifindex;
2704 /* Check for existence of name */
2705 head = dev_name_hash(dev->name);
2706 hlist_for_each(p, head) {
2707 struct net_device *d
2708 = hlist_entry(p, struct net_device, name_hlist);
2709 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2715 /* Fix illegal SG+CSUM combinations. */
2716 if ((dev->features & NETIF_F_SG) &&
2717 !(dev->features & (NETIF_F_IP_CSUM |
2719 NETIF_F_HW_CSUM))) {
2720 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2722 dev->features &= ~NETIF_F_SG;
2725 /* TSO requires that SG is present as well. */
2726 if ((dev->features & NETIF_F_TSO) &&
2727 !(dev->features & NETIF_F_SG)) {
2728 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2730 dev->features &= ~NETIF_F_TSO;
2732 if (dev->features & NETIF_F_UFO) {
2733 if (!(dev->features & NETIF_F_HW_CSUM)) {
2734 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2735 "NETIF_F_HW_CSUM feature.\n",
2737 dev->features &= ~NETIF_F_UFO;
2739 if (!(dev->features & NETIF_F_SG)) {
2740 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2741 "NETIF_F_SG feature.\n",
2743 dev->features &= ~NETIF_F_UFO;
2748 * nil rebuild_header routine,
2749 * that should be never called and used as just bug trap.
2752 if (!dev->rebuild_header)
2753 dev->rebuild_header = default_rebuild_header;
2756 * Default initial state at registry is that the
2757 * device is present.
2760 set_bit(__LINK_STATE_PRESENT, &dev->state);
2763 dev_init_scheduler(dev);
2764 write_lock_bh(&dev_base_lock);
2766 dev_tail = &dev->next;
2767 hlist_add_head(&dev->name_hlist, head);
2768 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2770 dev->reg_state = NETREG_REGISTERING;
2771 write_unlock_bh(&dev_base_lock);
2773 /* Notify protocols, that a new device appeared. */
2774 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2776 /* Finish registration after unlock */
2783 free_divert_blk(dev);
2788 * register_netdev - register a network device
2789 * @dev: device to register
2791 * Take a completed network device structure and add it to the kernel
2792 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2793 * chain. 0 is returned on success. A negative errno code is returned
2794 * on a failure to set up the device, or if the name is a duplicate.
2796 * This is a wrapper around register_netdev that takes the rtnl semaphore
2797 * and expands the device name if you passed a format string to
2800 int register_netdev(struct net_device *dev)
2807 * If the name is a format string the caller wants us to do a
2810 if (strchr(dev->name, '%')) {
2811 err = dev_alloc_name(dev, dev->name);
2817 * Back compatibility hook. Kill this one in 2.5
2819 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2820 err = dev_alloc_name(dev, "eth%d");
2825 err = register_netdevice(dev);
2830 EXPORT_SYMBOL(register_netdev);
2833 * netdev_wait_allrefs - wait until all references are gone.
2835 * This is called when unregistering network devices.
2837 * Any protocol or device that holds a reference should register
2838 * for netdevice notification, and cleanup and put back the
2839 * reference if they receive an UNREGISTER event.
2840 * We can get stuck here if buggy protocols don't correctly
2843 static void netdev_wait_allrefs(struct net_device *dev)
2845 unsigned long rebroadcast_time, warning_time;
2847 rebroadcast_time = warning_time = jiffies;
2848 while (atomic_read(&dev->refcnt) != 0) {
2849 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2852 /* Rebroadcast unregister notification */
2853 notifier_call_chain(&netdev_chain,
2854 NETDEV_UNREGISTER, dev);
2856 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2858 /* We must not have linkwatch events
2859 * pending on unregister. If this
2860 * happens, we simply run the queue
2861 * unscheduled, resulting in a noop
2864 linkwatch_run_queue();
2869 rebroadcast_time = jiffies;
2874 if (time_after(jiffies, warning_time + 10 * HZ)) {
2875 printk(KERN_EMERG "unregister_netdevice: "
2876 "waiting for %s to become free. Usage "
2878 dev->name, atomic_read(&dev->refcnt));
2879 warning_time = jiffies;
2888 * register_netdevice(x1);
2889 * register_netdevice(x2);
2891 * unregister_netdevice(y1);
2892 * unregister_netdevice(y2);
2898 * We are invoked by rtnl_unlock() after it drops the semaphore.
2899 * This allows us to deal with problems:
2900 * 1) We can create/delete sysfs objects which invoke hotplug
2901 * without deadlocking with linkwatch via keventd.
2902 * 2) Since we run with the RTNL semaphore not held, we can sleep
2903 * safely in order to wait for the netdev refcnt to drop to zero.
2905 static DECLARE_MUTEX(net_todo_run_mutex);
2906 void netdev_run_todo(void)
2908 struct list_head list = LIST_HEAD_INIT(list);
2912 /* Need to guard against multiple cpu's getting out of order. */
2913 down(&net_todo_run_mutex);
2915 /* Not safe to do outside the semaphore. We must not return
2916 * until all unregister events invoked by the local processor
2917 * have been completed (either by this todo run, or one on
2920 if (list_empty(&net_todo_list))
2923 /* Snapshot list, allow later requests */
2924 spin_lock(&net_todo_list_lock);
2925 list_splice_init(&net_todo_list, &list);
2926 spin_unlock(&net_todo_list_lock);
2928 while (!list_empty(&list)) {
2929 struct net_device *dev
2930 = list_entry(list.next, struct net_device, todo_list);
2931 list_del(&dev->todo_list);
2933 switch(dev->reg_state) {
2934 case NETREG_REGISTERING:
2935 err = netdev_register_sysfs(dev);
2937 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2939 dev->reg_state = NETREG_REGISTERED;
2942 case NETREG_UNREGISTERING:
2943 netdev_unregister_sysfs(dev);
2944 dev->reg_state = NETREG_UNREGISTERED;
2946 netdev_wait_allrefs(dev);
2949 BUG_ON(atomic_read(&dev->refcnt));
2950 BUG_TRAP(!dev->ip_ptr);
2951 BUG_TRAP(!dev->ip6_ptr);
2952 BUG_TRAP(!dev->dn_ptr);
2955 /* It must be the very last action,
2956 * after this 'dev' may point to freed up memory.
2958 if (dev->destructor)
2959 dev->destructor(dev);
2963 printk(KERN_ERR "network todo '%s' but state %d\n",
2964 dev->name, dev->reg_state);
2970 up(&net_todo_run_mutex);
2974 * alloc_netdev - allocate network device
2975 * @sizeof_priv: size of private data to allocate space for
2976 * @name: device name format string
2977 * @setup: callback to initialize device
2979 * Allocates a struct net_device with private data area for driver use
2980 * and performs basic initialization.
2982 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
2983 void (*setup)(struct net_device *))
2986 struct net_device *dev;
2989 /* ensure 32-byte alignment of both the device and private area */
2990 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
2991 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
2993 p = kmalloc(alloc_size, GFP_KERNEL);
2995 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
2998 memset(p, 0, alloc_size);
3000 dev = (struct net_device *)
3001 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3002 dev->padded = (char *)dev - (char *)p;
3005 dev->priv = netdev_priv(dev);
3008 strcpy(dev->name, name);
3011 EXPORT_SYMBOL(alloc_netdev);
3014 * free_netdev - free network device
3017 * This function does the last stage of destroying an allocated device
3018 * interface. The reference to the device object is released.
3019 * If this is the last reference then it will be freed.
3021 void free_netdev(struct net_device *dev)
3024 /* Compatiablity with error handling in drivers */
3025 if (dev->reg_state == NETREG_UNINITIALIZED) {
3026 kfree((char *)dev - dev->padded);
3030 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3031 dev->reg_state = NETREG_RELEASED;
3033 /* will free via class release */
3034 class_device_put(&dev->class_dev);
3036 kfree((char *)dev - dev->padded);
3040 /* Synchronize with packet receive processing. */
3041 void synchronize_net(void)
3048 * unregister_netdevice - remove device from the kernel
3051 * This function shuts down a device interface and removes it
3052 * from the kernel tables. On success 0 is returned, on a failure
3053 * a negative errno code is returned.
3055 * Callers must hold the rtnl semaphore. You may want
3056 * unregister_netdev() instead of this.
3059 int unregister_netdevice(struct net_device *dev)
3061 struct net_device *d, **dp;
3063 BUG_ON(dev_boot_phase);
3066 /* Some devices call without registering for initialization unwind. */
3067 if (dev->reg_state == NETREG_UNINITIALIZED) {
3068 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3069 "was registered\n", dev->name, dev);
3073 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3075 /* If device is running, close it first. */
3076 if (dev->flags & IFF_UP)
3079 /* And unlink it from device chain. */
3080 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3082 write_lock_bh(&dev_base_lock);
3083 hlist_del(&dev->name_hlist);
3084 hlist_del(&dev->index_hlist);
3085 if (dev_tail == &dev->next)
3088 write_unlock_bh(&dev_base_lock);
3093 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3098 dev->reg_state = NETREG_UNREGISTERING;
3102 /* Shutdown queueing discipline. */
3106 /* Notify protocols, that we are about to destroy
3107 this device. They should clean all the things.
3109 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3112 * Flush the multicast chain
3114 dev_mc_discard(dev);
3119 /* Notifier chain MUST detach us from master device. */
3120 BUG_TRAP(!dev->master);
3122 free_divert_blk(dev);
3124 /* Finish processing unregister after unlock */
3134 * unregister_netdev - remove device from the kernel
3137 * This function shuts down a device interface and removes it
3138 * from the kernel tables. On success 0 is returned, on a failure
3139 * a negative errno code is returned.
3141 * This is just a wrapper for unregister_netdevice that takes
3142 * the rtnl semaphore. In general you want to use this and not
3143 * unregister_netdevice.
3145 void unregister_netdev(struct net_device *dev)
3148 unregister_netdevice(dev);
3152 EXPORT_SYMBOL(unregister_netdev);
3154 #ifdef CONFIG_HOTPLUG_CPU
3155 static int dev_cpu_callback(struct notifier_block *nfb,
3156 unsigned long action,
3159 struct sk_buff **list_skb;
3160 struct net_device **list_net;
3161 struct sk_buff *skb;
3162 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3163 struct softnet_data *sd, *oldsd;
3165 if (action != CPU_DEAD)
3168 local_irq_disable();
3169 cpu = smp_processor_id();
3170 sd = &per_cpu(softnet_data, cpu);
3171 oldsd = &per_cpu(softnet_data, oldcpu);
3173 /* Find end of our completion_queue. */
3174 list_skb = &sd->completion_queue;
3176 list_skb = &(*list_skb)->next;
3177 /* Append completion queue from offline CPU. */
3178 *list_skb = oldsd->completion_queue;
3179 oldsd->completion_queue = NULL;
3181 /* Find end of our output_queue. */
3182 list_net = &sd->output_queue;
3184 list_net = &(*list_net)->next_sched;
3185 /* Append output queue from offline CPU. */
3186 *list_net = oldsd->output_queue;
3187 oldsd->output_queue = NULL;
3189 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3192 /* Process offline CPU's input_pkt_queue */
3193 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3198 #endif /* CONFIG_HOTPLUG_CPU */
3202 * Initialize the DEV module. At boot time this walks the device list and
3203 * unhooks any devices that fail to initialise (normally hardware not
3204 * present) and leaves us with a valid list of present and active devices.
3209 * This is called single threaded during boot, so no need
3210 * to take the rtnl semaphore.
3212 static int __init net_dev_init(void)
3214 int i, rc = -ENOMEM;
3216 BUG_ON(!dev_boot_phase);
3220 if (dev_proc_init())
3223 if (netdev_sysfs_init())
3226 INIT_LIST_HEAD(&ptype_all);
3227 for (i = 0; i < 16; i++)
3228 INIT_LIST_HEAD(&ptype_base[i]);
3230 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3231 INIT_HLIST_HEAD(&dev_name_head[i]);
3233 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3234 INIT_HLIST_HEAD(&dev_index_head[i]);
3237 * Initialise the packet receive queues.
3241 struct softnet_data *queue;
3243 queue = &per_cpu(softnet_data, i);
3244 skb_queue_head_init(&queue->input_pkt_queue);
3245 queue->completion_queue = NULL;
3246 INIT_LIST_HEAD(&queue->poll_list);
3247 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3248 queue->backlog_dev.weight = weight_p;
3249 queue->backlog_dev.poll = process_backlog;
3250 atomic_set(&queue->backlog_dev.refcnt, 1);
3255 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3256 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3258 hotcpu_notifier(dev_cpu_callback, 0);
3266 subsys_initcall(net_dev_init);
3268 EXPORT_SYMBOL(__dev_get_by_index);
3269 EXPORT_SYMBOL(__dev_get_by_name);
3270 EXPORT_SYMBOL(__dev_remove_pack);
3271 EXPORT_SYMBOL(__skb_linearize);
3272 EXPORT_SYMBOL(dev_valid_name);
3273 EXPORT_SYMBOL(dev_add_pack);
3274 EXPORT_SYMBOL(dev_alloc_name);
3275 EXPORT_SYMBOL(dev_close);
3276 EXPORT_SYMBOL(dev_get_by_flags);
3277 EXPORT_SYMBOL(dev_get_by_index);
3278 EXPORT_SYMBOL(dev_get_by_name);
3279 EXPORT_SYMBOL(dev_open);
3280 EXPORT_SYMBOL(dev_queue_xmit);
3281 EXPORT_SYMBOL(dev_remove_pack);
3282 EXPORT_SYMBOL(dev_set_allmulti);
3283 EXPORT_SYMBOL(dev_set_promiscuity);
3284 EXPORT_SYMBOL(dev_change_flags);
3285 EXPORT_SYMBOL(dev_set_mtu);
3286 EXPORT_SYMBOL(dev_set_mac_address);
3287 EXPORT_SYMBOL(free_netdev);
3288 EXPORT_SYMBOL(netdev_boot_setup_check);
3289 EXPORT_SYMBOL(netdev_set_master);
3290 EXPORT_SYMBOL(netdev_state_change);
3291 EXPORT_SYMBOL(netif_receive_skb);
3292 EXPORT_SYMBOL(netif_rx);
3293 EXPORT_SYMBOL(register_gifconf);
3294 EXPORT_SYMBOL(register_netdevice);
3295 EXPORT_SYMBOL(register_netdevice_notifier);
3296 EXPORT_SYMBOL(skb_checksum_help);
3297 EXPORT_SYMBOL(synchronize_net);
3298 EXPORT_SYMBOL(unregister_netdevice);
3299 EXPORT_SYMBOL(unregister_netdevice_notifier);
3300 EXPORT_SYMBOL(net_enable_timestamp);
3301 EXPORT_SYMBOL(net_disable_timestamp);
3302 EXPORT_SYMBOL(dev_get_flags);
3304 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3305 EXPORT_SYMBOL(br_handle_frame_hook);
3306 EXPORT_SYMBOL(br_fdb_get_hook);
3307 EXPORT_SYMBOL(br_fdb_put_hook);
3311 EXPORT_SYMBOL(dev_load);
3314 EXPORT_PER_CPU_SYMBOL(softnet_data);