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/mutex.h>
85 #include <linux/string.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>
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>
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>
117 #include <linux/audit.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 adversely 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 rtnl
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 RAW_NOTIFIER_HEAD(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. It scans list of devices to build up a free map, then chooses
645 * the first empty slot. The caller must hold the dev_base or rtnl lock
646 * while allocating the name and adding the device in order to avoid
648 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
649 * Returns the number of the unit assigned or a negative errno code.
652 int dev_alloc_name(struct net_device *dev, const char *name)
657 const int max_netdevices = 8*PAGE_SIZE;
659 struct net_device *d;
661 p = strnchr(name, IFNAMSIZ-1, '%');
664 * Verify the string as this thing may have come from
665 * the user. There must be either one "%d" and no other "%"
668 if (p[1] != 'd' || strchr(p + 2, '%'))
671 /* Use one page as a bit array of possible slots */
672 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
676 for (d = dev_base; d; d = d->next) {
677 if (!sscanf(d->name, name, &i))
679 if (i < 0 || i >= max_netdevices)
682 /* avoid cases where sscanf is not exact inverse of printf */
683 snprintf(buf, sizeof(buf), name, i);
684 if (!strncmp(buf, d->name, IFNAMSIZ))
688 i = find_first_zero_bit(inuse, max_netdevices);
689 free_page((unsigned long) inuse);
692 snprintf(buf, sizeof(buf), name, i);
693 if (!__dev_get_by_name(buf)) {
694 strlcpy(dev->name, buf, IFNAMSIZ);
698 /* It is possible to run out of possible slots
699 * when the name is long and there isn't enough space left
700 * for the digits, or if all bits are used.
707 * dev_change_name - change name of a device
709 * @newname: name (or format string) must be at least IFNAMSIZ
711 * Change name of a device, can pass format strings "eth%d".
714 int dev_change_name(struct net_device *dev, char *newname)
720 if (dev->flags & IFF_UP)
723 if (!dev_valid_name(newname))
726 if (strchr(newname, '%')) {
727 err = dev_alloc_name(dev, newname);
730 strcpy(newname, dev->name);
732 else if (__dev_get_by_name(newname))
735 strlcpy(dev->name, newname, IFNAMSIZ);
737 err = class_device_rename(&dev->class_dev, dev->name);
739 hlist_del(&dev->name_hlist);
740 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
741 raw_notifier_call_chain(&netdev_chain,
742 NETDEV_CHANGENAME, dev);
749 * netdev_features_change - device changes features
750 * @dev: device to cause notification
752 * Called to indicate a device has changed features.
754 void netdev_features_change(struct net_device *dev)
756 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
758 EXPORT_SYMBOL(netdev_features_change);
761 * netdev_state_change - device changes state
762 * @dev: device to cause notification
764 * Called to indicate a device has changed state. This function calls
765 * the notifier chains for netdev_chain and sends a NEWLINK message
766 * to the routing socket.
768 void netdev_state_change(struct net_device *dev)
770 if (dev->flags & IFF_UP) {
771 raw_notifier_call_chain(&netdev_chain,
773 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
778 * dev_load - load a network module
779 * @name: name of interface
781 * If a network interface is not present and the process has suitable
782 * privileges this function loads the module. If module loading is not
783 * available in this kernel then it becomes a nop.
786 void dev_load(const char *name)
788 struct net_device *dev;
790 read_lock(&dev_base_lock);
791 dev = __dev_get_by_name(name);
792 read_unlock(&dev_base_lock);
794 if (!dev && capable(CAP_SYS_MODULE))
795 request_module("%s", name);
798 static int default_rebuild_header(struct sk_buff *skb)
800 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
801 skb->dev ? skb->dev->name : "NULL!!!");
808 * dev_open - prepare an interface for use.
809 * @dev: device to open
811 * Takes a device from down to up state. The device's private open
812 * function is invoked and then the multicast lists are loaded. Finally
813 * the device is moved into the up state and a %NETDEV_UP message is
814 * sent to the netdev notifier chain.
816 * Calling this function on an active interface is a nop. On a failure
817 * a negative errno code is returned.
819 int dev_open(struct net_device *dev)
827 if (dev->flags & IFF_UP)
831 * Is it even present?
833 if (!netif_device_present(dev))
837 * Call device private open method
839 set_bit(__LINK_STATE_START, &dev->state);
841 ret = dev->open(dev);
843 clear_bit(__LINK_STATE_START, &dev->state);
847 * If it went open OK then:
854 dev->flags |= IFF_UP;
857 * Initialize multicasting status
862 * Wakeup transmit queue engine
867 * ... and announce new interface.
869 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
875 * dev_close - shutdown an interface.
876 * @dev: device to shutdown
878 * This function moves an active device into down state. A
879 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
880 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
883 int dev_close(struct net_device *dev)
885 if (!(dev->flags & IFF_UP))
889 * Tell people we are going down, so that they can
890 * prepare to death, when device is still operating.
892 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
896 clear_bit(__LINK_STATE_START, &dev->state);
898 /* Synchronize to scheduled poll. We cannot touch poll list,
899 * it can be even on different cpu. So just clear netif_running(),
900 * and wait when poll really will happen. Actually, the best place
901 * for this is inside dev->stop() after device stopped its irq
902 * engine, but this requires more changes in devices. */
904 smp_mb__after_clear_bit(); /* Commit netif_running(). */
905 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
911 * Call the device specific close. This cannot fail.
912 * Only if device is UP
914 * We allow it to be called even after a DETACH hot-plug
921 * Device is now down.
924 dev->flags &= ~IFF_UP;
927 * Tell people we are down
929 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
936 * Device change register/unregister. These are not inline or static
937 * as we export them to the world.
941 * register_netdevice_notifier - register a network notifier block
944 * Register a notifier to be called when network device events occur.
945 * The notifier passed is linked into the kernel structures and must
946 * not be reused until it has been unregistered. A negative errno code
947 * is returned on a failure.
949 * When registered all registration and up events are replayed
950 * to the new notifier to allow device to have a race free
951 * view of the network device list.
954 int register_netdevice_notifier(struct notifier_block *nb)
956 struct net_device *dev;
960 err = raw_notifier_chain_register(&netdev_chain, nb);
962 for (dev = dev_base; dev; dev = dev->next) {
963 nb->notifier_call(nb, NETDEV_REGISTER, dev);
965 if (dev->flags & IFF_UP)
966 nb->notifier_call(nb, NETDEV_UP, dev);
974 * unregister_netdevice_notifier - unregister a network notifier block
977 * Unregister a notifier previously registered by
978 * register_netdevice_notifier(). The notifier is unlinked into the
979 * kernel structures and may then be reused. A negative errno code
980 * is returned on a failure.
983 int unregister_netdevice_notifier(struct notifier_block *nb)
988 err = raw_notifier_chain_unregister(&netdev_chain, nb);
994 * call_netdevice_notifiers - call all network notifier blocks
995 * @val: value passed unmodified to notifier function
996 * @v: pointer passed unmodified to notifier function
998 * Call all network notifier blocks. Parameters and return value
999 * are as for raw_notifier_call_chain().
1002 int call_netdevice_notifiers(unsigned long val, void *v)
1004 return raw_notifier_call_chain(&netdev_chain, val, v);
1007 /* When > 0 there are consumers of rx skb time stamps */
1008 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1010 void net_enable_timestamp(void)
1012 atomic_inc(&netstamp_needed);
1015 void net_disable_timestamp(void)
1017 atomic_dec(&netstamp_needed);
1020 void __net_timestamp(struct sk_buff *skb)
1024 do_gettimeofday(&tv);
1025 skb_set_timestamp(skb, &tv);
1027 EXPORT_SYMBOL(__net_timestamp);
1029 static inline void net_timestamp(struct sk_buff *skb)
1031 if (atomic_read(&netstamp_needed))
1032 __net_timestamp(skb);
1034 skb->tstamp.off_sec = 0;
1035 skb->tstamp.off_usec = 0;
1040 * Support routine. Sends outgoing frames to any network
1041 * taps currently in use.
1044 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1046 struct packet_type *ptype;
1051 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1052 /* Never send packets back to the socket
1053 * they originated from - MvS (miquels@drinkel.ow.org)
1055 if ((ptype->dev == dev || !ptype->dev) &&
1056 (ptype->af_packet_priv == NULL ||
1057 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1058 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1062 /* skb->nh should be correctly
1063 set by sender, so that the second statement is
1064 just protection against buggy protocols.
1066 skb2->mac.raw = skb2->data;
1068 if (skb2->nh.raw < skb2->data ||
1069 skb2->nh.raw > skb2->tail) {
1070 if (net_ratelimit())
1071 printk(KERN_CRIT "protocol %04x is "
1073 skb2->protocol, dev->name);
1074 skb2->nh.raw = skb2->data;
1077 skb2->h.raw = skb2->nh.raw;
1078 skb2->pkt_type = PACKET_OUTGOING;
1079 ptype->func(skb2, skb->dev, ptype, skb->dev);
1086 void __netif_schedule(struct net_device *dev)
1088 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1089 unsigned long flags;
1090 struct softnet_data *sd;
1092 local_irq_save(flags);
1093 sd = &__get_cpu_var(softnet_data);
1094 dev->next_sched = sd->output_queue;
1095 sd->output_queue = dev;
1096 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1097 local_irq_restore(flags);
1100 EXPORT_SYMBOL(__netif_schedule);
1102 void __netif_rx_schedule(struct net_device *dev)
1104 unsigned long flags;
1106 local_irq_save(flags);
1108 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1110 dev->quota += dev->weight;
1112 dev->quota = dev->weight;
1113 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1114 local_irq_restore(flags);
1116 EXPORT_SYMBOL(__netif_rx_schedule);
1118 void dev_kfree_skb_any(struct sk_buff *skb)
1120 if (in_irq() || irqs_disabled())
1121 dev_kfree_skb_irq(skb);
1125 EXPORT_SYMBOL(dev_kfree_skb_any);
1129 void netif_device_detach(struct net_device *dev)
1131 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1132 netif_running(dev)) {
1133 netif_stop_queue(dev);
1136 EXPORT_SYMBOL(netif_device_detach);
1138 void netif_device_attach(struct net_device *dev)
1140 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1141 netif_running(dev)) {
1142 netif_wake_queue(dev);
1143 __netdev_watchdog_up(dev);
1146 EXPORT_SYMBOL(netif_device_attach);
1150 * Invalidate hardware checksum when packet is to be mangled, and
1151 * complete checksum manually on outgoing path.
1153 int skb_checksum_help(struct sk_buff *skb, int inward)
1156 int ret = 0, offset = skb->h.raw - skb->data;
1159 skb->ip_summed = CHECKSUM_NONE;
1163 if (skb_cloned(skb)) {
1164 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1169 BUG_ON(offset > (int)skb->len);
1170 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1172 offset = skb->tail - skb->h.raw;
1173 BUG_ON(offset <= 0);
1174 BUG_ON(skb->csum + 2 > offset);
1176 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1177 skb->ip_summed = CHECKSUM_NONE;
1182 /* Take action when hardware reception checksum errors are detected. */
1184 void netdev_rx_csum_fault(struct net_device *dev)
1186 if (net_ratelimit()) {
1187 printk(KERN_ERR "%s: hw csum failure.\n",
1188 dev ? dev->name : "<unknown>");
1192 EXPORT_SYMBOL(netdev_rx_csum_fault);
1195 #ifdef CONFIG_HIGHMEM
1196 /* Actually, we should eliminate this check as soon as we know, that:
1197 * 1. IOMMU is present and allows to map all the memory.
1198 * 2. No high memory really exists on this machine.
1201 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1205 if (dev->features & NETIF_F_HIGHDMA)
1208 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1209 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1215 #define illegal_highdma(dev, skb) (0)
1218 /* Keep head the same: replace data */
1219 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1224 struct skb_shared_info *ninfo;
1225 int headerlen = skb->data - skb->head;
1226 int expand = (skb->tail + skb->data_len) - skb->end;
1228 if (skb_shared(skb))
1234 size = skb->end - skb->head + expand;
1235 size = SKB_DATA_ALIGN(size);
1236 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1240 /* Copy entire thing */
1241 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1245 ninfo = (struct skb_shared_info*)(data + size);
1246 atomic_set(&ninfo->dataref, 1);
1247 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1248 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1249 ninfo->nr_frags = 0;
1250 ninfo->frag_list = NULL;
1252 /* Offset between the two in bytes */
1253 offset = data - skb->head;
1255 /* Free old data. */
1256 skb_release_data(skb);
1259 skb->end = data + size;
1261 /* Set up new pointers */
1262 skb->h.raw += offset;
1263 skb->nh.raw += offset;
1264 skb->mac.raw += offset;
1265 skb->tail += offset;
1266 skb->data += offset;
1268 /* We are no longer a clone, even if we were. */
1271 skb->tail += skb->data_len;
1276 #define HARD_TX_LOCK(dev, cpu) { \
1277 if ((dev->features & NETIF_F_LLTX) == 0) { \
1278 spin_lock(&dev->xmit_lock); \
1279 dev->xmit_lock_owner = cpu; \
1283 #define HARD_TX_UNLOCK(dev) { \
1284 if ((dev->features & NETIF_F_LLTX) == 0) { \
1285 dev->xmit_lock_owner = -1; \
1286 spin_unlock(&dev->xmit_lock); \
1291 * dev_queue_xmit - transmit a buffer
1292 * @skb: buffer to transmit
1294 * Queue a buffer for transmission to a network device. The caller must
1295 * have set the device and priority and built the buffer before calling
1296 * this function. The function can be called from an interrupt.
1298 * A negative errno code is returned on a failure. A success does not
1299 * guarantee the frame will be transmitted as it may be dropped due
1300 * to congestion or traffic shaping.
1302 * -----------------------------------------------------------------------------------
1303 * I notice this method can also return errors from the queue disciplines,
1304 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1307 * Regardless of the return value, the skb is consumed, so it is currently
1308 * difficult to retry a send to this method. (You can bump the ref count
1309 * before sending to hold a reference for retry if you are careful.)
1311 * When calling this method, interrupts MUST be enabled. This is because
1312 * the BH enable code must have IRQs enabled so that it will not deadlock.
1316 int dev_queue_xmit(struct sk_buff *skb)
1318 struct net_device *dev = skb->dev;
1322 if (skb_shinfo(skb)->frag_list &&
1323 !(dev->features & NETIF_F_FRAGLIST) &&
1324 __skb_linearize(skb, GFP_ATOMIC))
1327 /* Fragmented skb is linearized if device does not support SG,
1328 * or if at least one of fragments is in highmem and device
1329 * does not support DMA from it.
1331 if (skb_shinfo(skb)->nr_frags &&
1332 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1333 __skb_linearize(skb, GFP_ATOMIC))
1336 /* If packet is not checksummed and device does not support
1337 * checksumming for this protocol, complete checksumming here.
1339 if (skb->ip_summed == CHECKSUM_HW &&
1340 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1341 (!(dev->features & NETIF_F_IP_CSUM) ||
1342 skb->protocol != htons(ETH_P_IP))))
1343 if (skb_checksum_help(skb, 0))
1346 spin_lock_prefetch(&dev->queue_lock);
1348 /* Disable soft irqs for various locks below. Also
1349 * stops preemption for RCU.
1353 /* Updates of qdisc are serialized by queue_lock.
1354 * The struct Qdisc which is pointed to by qdisc is now a
1355 * rcu structure - it may be accessed without acquiring
1356 * a lock (but the structure may be stale.) The freeing of the
1357 * qdisc will be deferred until it's known that there are no
1358 * more references to it.
1360 * If the qdisc has an enqueue function, we still need to
1361 * hold the queue_lock before calling it, since queue_lock
1362 * also serializes access to the device queue.
1365 q = rcu_dereference(dev->qdisc);
1366 #ifdef CONFIG_NET_CLS_ACT
1367 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1370 /* Grab device queue */
1371 spin_lock(&dev->queue_lock);
1373 rc = q->enqueue(skb, q);
1377 spin_unlock(&dev->queue_lock);
1378 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1382 /* The device has no queue. Common case for software devices:
1383 loopback, all the sorts of tunnels...
1385 Really, it is unlikely that xmit_lock protection is necessary here.
1386 (f.e. loopback and IP tunnels are clean ignoring statistics
1388 However, it is possible, that they rely on protection
1391 Check this and shot the lock. It is not prone from deadlocks.
1392 Either shot noqueue qdisc, it is even simpler 8)
1394 if (dev->flags & IFF_UP) {
1395 int cpu = smp_processor_id(); /* ok because BHs are off */
1397 if (dev->xmit_lock_owner != cpu) {
1399 HARD_TX_LOCK(dev, cpu);
1401 if (!netif_queue_stopped(dev)) {
1403 dev_queue_xmit_nit(skb, dev);
1406 if (!dev->hard_start_xmit(skb, dev)) {
1407 HARD_TX_UNLOCK(dev);
1411 HARD_TX_UNLOCK(dev);
1412 if (net_ratelimit())
1413 printk(KERN_CRIT "Virtual device %s asks to "
1414 "queue packet!\n", dev->name);
1416 /* Recursion is detected! It is possible,
1418 if (net_ratelimit())
1419 printk(KERN_CRIT "Dead loop on virtual device "
1420 "%s, fix it urgently!\n", dev->name);
1436 /*=======================================================================
1438 =======================================================================*/
1440 int netdev_max_backlog = 1000;
1441 int netdev_budget = 300;
1442 int weight_p = 64; /* old backlog weight */
1444 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1448 * netif_rx - post buffer to the network code
1449 * @skb: buffer to post
1451 * This function receives a packet from a device driver and queues it for
1452 * the upper (protocol) levels to process. It always succeeds. The buffer
1453 * may be dropped during processing for congestion control or by the
1457 * NET_RX_SUCCESS (no congestion)
1458 * NET_RX_CN_LOW (low congestion)
1459 * NET_RX_CN_MOD (moderate congestion)
1460 * NET_RX_CN_HIGH (high congestion)
1461 * NET_RX_DROP (packet was dropped)
1465 int netif_rx(struct sk_buff *skb)
1467 struct softnet_data *queue;
1468 unsigned long flags;
1470 /* if netpoll wants it, pretend we never saw it */
1471 if (netpoll_rx(skb))
1474 if (!skb->tstamp.off_sec)
1478 * The code is rearranged so that the path is the most
1479 * short when CPU is congested, but is still operating.
1481 local_irq_save(flags);
1482 queue = &__get_cpu_var(softnet_data);
1484 __get_cpu_var(netdev_rx_stat).total++;
1485 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1486 if (queue->input_pkt_queue.qlen) {
1489 __skb_queue_tail(&queue->input_pkt_queue, skb);
1490 local_irq_restore(flags);
1491 return NET_RX_SUCCESS;
1494 netif_rx_schedule(&queue->backlog_dev);
1498 __get_cpu_var(netdev_rx_stat).dropped++;
1499 local_irq_restore(flags);
1505 int netif_rx_ni(struct sk_buff *skb)
1510 err = netif_rx(skb);
1511 if (local_softirq_pending())
1518 EXPORT_SYMBOL(netif_rx_ni);
1520 static inline struct net_device *skb_bond(struct sk_buff *skb)
1522 struct net_device *dev = skb->dev;
1526 * On bonding slaves other than the currently active
1527 * slave, suppress duplicates except for 802.3ad
1528 * ETH_P_SLOW and alb non-mcast/bcast.
1530 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1531 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1532 if (skb->pkt_type != PACKET_BROADCAST &&
1533 skb->pkt_type != PACKET_MULTICAST)
1537 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1538 skb->protocol == __constant_htons(ETH_P_SLOW))
1545 skb->dev = dev->master;
1551 static void net_tx_action(struct softirq_action *h)
1553 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1555 if (sd->completion_queue) {
1556 struct sk_buff *clist;
1558 local_irq_disable();
1559 clist = sd->completion_queue;
1560 sd->completion_queue = NULL;
1564 struct sk_buff *skb = clist;
1565 clist = clist->next;
1567 BUG_TRAP(!atomic_read(&skb->users));
1572 if (sd->output_queue) {
1573 struct net_device *head;
1575 local_irq_disable();
1576 head = sd->output_queue;
1577 sd->output_queue = NULL;
1581 struct net_device *dev = head;
1582 head = head->next_sched;
1584 smp_mb__before_clear_bit();
1585 clear_bit(__LINK_STATE_SCHED, &dev->state);
1587 if (spin_trylock(&dev->queue_lock)) {
1589 spin_unlock(&dev->queue_lock);
1591 netif_schedule(dev);
1597 static __inline__ int deliver_skb(struct sk_buff *skb,
1598 struct packet_type *pt_prev,
1599 struct net_device *orig_dev)
1601 atomic_inc(&skb->users);
1602 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1605 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1606 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1608 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1609 unsigned char *addr);
1610 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1612 static __inline__ int handle_bridge(struct sk_buff **pskb,
1613 struct packet_type **pt_prev, int *ret,
1614 struct net_device *orig_dev)
1616 struct net_bridge_port *port;
1618 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1619 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1623 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1627 return br_handle_frame_hook(port, pskb);
1630 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1633 #ifdef CONFIG_NET_CLS_ACT
1634 /* TODO: Maybe we should just force sch_ingress to be compiled in
1635 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1636 * a compare and 2 stores extra right now if we dont have it on
1637 * but have CONFIG_NET_CLS_ACT
1638 * NOTE: This doesnt stop any functionality; if you dont have
1639 * the ingress scheduler, you just cant add policies on ingress.
1642 static int ing_filter(struct sk_buff *skb)
1645 struct net_device *dev = skb->dev;
1646 int result = TC_ACT_OK;
1648 if (dev->qdisc_ingress) {
1649 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1650 if (MAX_RED_LOOP < ttl++) {
1651 printk("Redir loop detected Dropping packet (%s->%s)\n",
1652 skb->input_dev->name, skb->dev->name);
1656 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1658 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1660 spin_lock(&dev->ingress_lock);
1661 if ((q = dev->qdisc_ingress) != NULL)
1662 result = q->enqueue(skb, q);
1663 spin_unlock(&dev->ingress_lock);
1671 int netif_receive_skb(struct sk_buff *skb)
1673 struct packet_type *ptype, *pt_prev;
1674 struct net_device *orig_dev;
1675 int ret = NET_RX_DROP;
1676 unsigned short type;
1678 /* if we've gotten here through NAPI, check netpoll */
1679 if (skb->dev->poll && netpoll_rx(skb))
1682 if (!skb->tstamp.off_sec)
1685 if (!skb->input_dev)
1686 skb->input_dev = skb->dev;
1688 orig_dev = skb_bond(skb);
1693 __get_cpu_var(netdev_rx_stat).total++;
1695 skb->h.raw = skb->nh.raw = skb->data;
1696 skb->mac_len = skb->nh.raw - skb->mac.raw;
1702 #ifdef CONFIG_NET_CLS_ACT
1703 if (skb->tc_verd & TC_NCLS) {
1704 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1709 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1710 if (!ptype->dev || ptype->dev == skb->dev) {
1712 ret = deliver_skb(skb, pt_prev, orig_dev);
1717 #ifdef CONFIG_NET_CLS_ACT
1719 ret = deliver_skb(skb, pt_prev, orig_dev);
1720 pt_prev = NULL; /* noone else should process this after*/
1722 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1725 ret = ing_filter(skb);
1727 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1736 handle_diverter(skb);
1738 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1741 type = skb->protocol;
1742 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1743 if (ptype->type == type &&
1744 (!ptype->dev || ptype->dev == skb->dev)) {
1746 ret = deliver_skb(skb, pt_prev, orig_dev);
1752 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1755 /* Jamal, now you will not able to escape explaining
1756 * me how you were going to use this. :-)
1766 static int process_backlog(struct net_device *backlog_dev, int *budget)
1769 int quota = min(backlog_dev->quota, *budget);
1770 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1771 unsigned long start_time = jiffies;
1773 backlog_dev->weight = weight_p;
1775 struct sk_buff *skb;
1776 struct net_device *dev;
1778 local_irq_disable();
1779 skb = __skb_dequeue(&queue->input_pkt_queue);
1786 netif_receive_skb(skb);
1792 if (work >= quota || jiffies - start_time > 1)
1797 backlog_dev->quota -= work;
1802 backlog_dev->quota -= work;
1805 list_del(&backlog_dev->poll_list);
1806 smp_mb__before_clear_bit();
1807 netif_poll_enable(backlog_dev);
1813 static void net_rx_action(struct softirq_action *h)
1815 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1816 unsigned long start_time = jiffies;
1817 int budget = netdev_budget;
1820 local_irq_disable();
1822 while (!list_empty(&queue->poll_list)) {
1823 struct net_device *dev;
1825 if (budget <= 0 || jiffies - start_time > 1)
1830 dev = list_entry(queue->poll_list.next,
1831 struct net_device, poll_list);
1832 have = netpoll_poll_lock(dev);
1834 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1835 netpoll_poll_unlock(have);
1836 local_irq_disable();
1837 list_move_tail(&dev->poll_list, &queue->poll_list);
1839 dev->quota += dev->weight;
1841 dev->quota = dev->weight;
1843 netpoll_poll_unlock(have);
1845 local_irq_disable();
1853 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1854 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1858 static gifconf_func_t * gifconf_list [NPROTO];
1861 * register_gifconf - register a SIOCGIF handler
1862 * @family: Address family
1863 * @gifconf: Function handler
1865 * Register protocol dependent address dumping routines. The handler
1866 * that is passed must not be freed or reused until it has been replaced
1867 * by another handler.
1869 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1871 if (family >= NPROTO)
1873 gifconf_list[family] = gifconf;
1879 * Map an interface index to its name (SIOCGIFNAME)
1883 * We need this ioctl for efficient implementation of the
1884 * if_indextoname() function required by the IPv6 API. Without
1885 * it, we would have to search all the interfaces to find a
1889 static int dev_ifname(struct ifreq __user *arg)
1891 struct net_device *dev;
1895 * Fetch the caller's info block.
1898 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1901 read_lock(&dev_base_lock);
1902 dev = __dev_get_by_index(ifr.ifr_ifindex);
1904 read_unlock(&dev_base_lock);
1908 strcpy(ifr.ifr_name, dev->name);
1909 read_unlock(&dev_base_lock);
1911 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1917 * Perform a SIOCGIFCONF call. This structure will change
1918 * size eventually, and there is nothing I can do about it.
1919 * Thus we will need a 'compatibility mode'.
1922 static int dev_ifconf(char __user *arg)
1925 struct net_device *dev;
1932 * Fetch the caller's info block.
1935 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1942 * Loop over the interfaces, and write an info block for each.
1946 for (dev = dev_base; dev; dev = dev->next) {
1947 for (i = 0; i < NPROTO; i++) {
1948 if (gifconf_list[i]) {
1951 done = gifconf_list[i](dev, NULL, 0);
1953 done = gifconf_list[i](dev, pos + total,
1963 * All done. Write the updated control block back to the caller.
1965 ifc.ifc_len = total;
1968 * Both BSD and Solaris return 0 here, so we do too.
1970 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1973 #ifdef CONFIG_PROC_FS
1975 * This is invoked by the /proc filesystem handler to display a device
1978 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1980 struct net_device *dev;
1983 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1985 return i == pos ? dev : NULL;
1988 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1990 read_lock(&dev_base_lock);
1991 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1994 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1997 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2000 void dev_seq_stop(struct seq_file *seq, void *v)
2002 read_unlock(&dev_base_lock);
2005 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2007 if (dev->get_stats) {
2008 struct net_device_stats *stats = dev->get_stats(dev);
2010 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2011 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2012 dev->name, stats->rx_bytes, stats->rx_packets,
2014 stats->rx_dropped + stats->rx_missed_errors,
2015 stats->rx_fifo_errors,
2016 stats->rx_length_errors + stats->rx_over_errors +
2017 stats->rx_crc_errors + stats->rx_frame_errors,
2018 stats->rx_compressed, stats->multicast,
2019 stats->tx_bytes, stats->tx_packets,
2020 stats->tx_errors, stats->tx_dropped,
2021 stats->tx_fifo_errors, stats->collisions,
2022 stats->tx_carrier_errors +
2023 stats->tx_aborted_errors +
2024 stats->tx_window_errors +
2025 stats->tx_heartbeat_errors,
2026 stats->tx_compressed);
2028 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2032 * Called from the PROCfs module. This now uses the new arbitrary sized
2033 * /proc/net interface to create /proc/net/dev
2035 static int dev_seq_show(struct seq_file *seq, void *v)
2037 if (v == SEQ_START_TOKEN)
2038 seq_puts(seq, "Inter-| Receive "
2040 " face |bytes packets errs drop fifo frame "
2041 "compressed multicast|bytes packets errs "
2042 "drop fifo colls carrier compressed\n");
2044 dev_seq_printf_stats(seq, v);
2048 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2050 struct netif_rx_stats *rc = NULL;
2052 while (*pos < NR_CPUS)
2053 if (cpu_online(*pos)) {
2054 rc = &per_cpu(netdev_rx_stat, *pos);
2061 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2063 return softnet_get_online(pos);
2066 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2069 return softnet_get_online(pos);
2072 static void softnet_seq_stop(struct seq_file *seq, void *v)
2076 static int softnet_seq_show(struct seq_file *seq, void *v)
2078 struct netif_rx_stats *s = v;
2080 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2081 s->total, s->dropped, s->time_squeeze, 0,
2082 0, 0, 0, 0, /* was fastroute */
2087 static struct seq_operations dev_seq_ops = {
2088 .start = dev_seq_start,
2089 .next = dev_seq_next,
2090 .stop = dev_seq_stop,
2091 .show = dev_seq_show,
2094 static int dev_seq_open(struct inode *inode, struct file *file)
2096 return seq_open(file, &dev_seq_ops);
2099 static struct file_operations dev_seq_fops = {
2100 .owner = THIS_MODULE,
2101 .open = dev_seq_open,
2103 .llseek = seq_lseek,
2104 .release = seq_release,
2107 static struct seq_operations softnet_seq_ops = {
2108 .start = softnet_seq_start,
2109 .next = softnet_seq_next,
2110 .stop = softnet_seq_stop,
2111 .show = softnet_seq_show,
2114 static int softnet_seq_open(struct inode *inode, struct file *file)
2116 return seq_open(file, &softnet_seq_ops);
2119 static struct file_operations softnet_seq_fops = {
2120 .owner = THIS_MODULE,
2121 .open = softnet_seq_open,
2123 .llseek = seq_lseek,
2124 .release = seq_release,
2127 #ifdef CONFIG_WIRELESS_EXT
2128 extern int wireless_proc_init(void);
2130 #define wireless_proc_init() 0
2133 static int __init dev_proc_init(void)
2137 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2139 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2141 if (wireless_proc_init())
2147 proc_net_remove("softnet_stat");
2149 proc_net_remove("dev");
2153 #define dev_proc_init() 0
2154 #endif /* CONFIG_PROC_FS */
2158 * netdev_set_master - set up master/slave pair
2159 * @slave: slave device
2160 * @master: new master device
2162 * Changes the master device of the slave. Pass %NULL to break the
2163 * bonding. The caller must hold the RTNL semaphore. On a failure
2164 * a negative errno code is returned. On success the reference counts
2165 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2166 * function returns zero.
2168 int netdev_set_master(struct net_device *slave, struct net_device *master)
2170 struct net_device *old = slave->master;
2180 slave->master = master;
2188 slave->flags |= IFF_SLAVE;
2190 slave->flags &= ~IFF_SLAVE;
2192 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2197 * dev_set_promiscuity - update promiscuity count on a device
2201 * Add or remove promiscuity from a device. While the count in the device
2202 * remains above zero the interface remains promiscuous. Once it hits zero
2203 * the device reverts back to normal filtering operation. A negative inc
2204 * value is used to drop promiscuity on the device.
2206 void dev_set_promiscuity(struct net_device *dev, int inc)
2208 unsigned short old_flags = dev->flags;
2210 if ((dev->promiscuity += inc) == 0)
2211 dev->flags &= ~IFF_PROMISC;
2213 dev->flags |= IFF_PROMISC;
2214 if (dev->flags != old_flags) {
2216 printk(KERN_INFO "device %s %s promiscuous mode\n",
2217 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2219 audit_log(current->audit_context, GFP_ATOMIC,
2220 AUDIT_ANOM_PROMISCUOUS,
2221 "dev=%s prom=%d old_prom=%d auid=%u",
2222 dev->name, (dev->flags & IFF_PROMISC),
2223 (old_flags & IFF_PROMISC),
2224 audit_get_loginuid(current->audit_context));
2229 * dev_set_allmulti - update allmulti count on a device
2233 * Add or remove reception of all multicast frames to a device. While the
2234 * count in the device remains above zero the interface remains listening
2235 * to all interfaces. Once it hits zero the device reverts back to normal
2236 * filtering operation. A negative @inc value is used to drop the counter
2237 * when releasing a resource needing all multicasts.
2240 void dev_set_allmulti(struct net_device *dev, int inc)
2242 unsigned short old_flags = dev->flags;
2244 dev->flags |= IFF_ALLMULTI;
2245 if ((dev->allmulti += inc) == 0)
2246 dev->flags &= ~IFF_ALLMULTI;
2247 if (dev->flags ^ old_flags)
2251 unsigned dev_get_flags(const struct net_device *dev)
2255 flags = (dev->flags & ~(IFF_PROMISC |
2260 (dev->gflags & (IFF_PROMISC |
2263 if (netif_running(dev)) {
2264 if (netif_oper_up(dev))
2265 flags |= IFF_RUNNING;
2266 if (netif_carrier_ok(dev))
2267 flags |= IFF_LOWER_UP;
2268 if (netif_dormant(dev))
2269 flags |= IFF_DORMANT;
2275 int dev_change_flags(struct net_device *dev, unsigned flags)
2278 int old_flags = dev->flags;
2281 * Set the flags on our device.
2284 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2285 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2287 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2291 * Load in the correct multicast list now the flags have changed.
2297 * Have we downed the interface. We handle IFF_UP ourselves
2298 * according to user attempts to set it, rather than blindly
2303 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2304 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2310 if (dev->flags & IFF_UP &&
2311 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2313 raw_notifier_call_chain(&netdev_chain,
2314 NETDEV_CHANGE, dev);
2316 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2317 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2318 dev->gflags ^= IFF_PROMISC;
2319 dev_set_promiscuity(dev, inc);
2322 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2323 is important. Some (broken) drivers set IFF_PROMISC, when
2324 IFF_ALLMULTI is requested not asking us and not reporting.
2326 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2327 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2328 dev->gflags ^= IFF_ALLMULTI;
2329 dev_set_allmulti(dev, inc);
2332 if (old_flags ^ dev->flags)
2333 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2338 int dev_set_mtu(struct net_device *dev, int new_mtu)
2342 if (new_mtu == dev->mtu)
2345 /* MTU must be positive. */
2349 if (!netif_device_present(dev))
2353 if (dev->change_mtu)
2354 err = dev->change_mtu(dev, new_mtu);
2357 if (!err && dev->flags & IFF_UP)
2358 raw_notifier_call_chain(&netdev_chain,
2359 NETDEV_CHANGEMTU, dev);
2363 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2367 if (!dev->set_mac_address)
2369 if (sa->sa_family != dev->type)
2371 if (!netif_device_present(dev))
2373 err = dev->set_mac_address(dev, sa);
2375 raw_notifier_call_chain(&netdev_chain,
2376 NETDEV_CHANGEADDR, dev);
2381 * Perform the SIOCxIFxxx calls.
2383 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2386 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2392 case SIOCGIFFLAGS: /* Get interface flags */
2393 ifr->ifr_flags = dev_get_flags(dev);
2396 case SIOCSIFFLAGS: /* Set interface flags */
2397 return dev_change_flags(dev, ifr->ifr_flags);
2399 case SIOCGIFMETRIC: /* Get the metric on the interface
2400 (currently unused) */
2401 ifr->ifr_metric = 0;
2404 case SIOCSIFMETRIC: /* Set the metric on the interface
2405 (currently unused) */
2408 case SIOCGIFMTU: /* Get the MTU of a device */
2409 ifr->ifr_mtu = dev->mtu;
2412 case SIOCSIFMTU: /* Set the MTU of a device */
2413 return dev_set_mtu(dev, ifr->ifr_mtu);
2417 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2419 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2420 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2421 ifr->ifr_hwaddr.sa_family = dev->type;
2425 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2427 case SIOCSIFHWBROADCAST:
2428 if (ifr->ifr_hwaddr.sa_family != dev->type)
2430 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2431 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2432 raw_notifier_call_chain(&netdev_chain,
2433 NETDEV_CHANGEADDR, dev);
2437 ifr->ifr_map.mem_start = dev->mem_start;
2438 ifr->ifr_map.mem_end = dev->mem_end;
2439 ifr->ifr_map.base_addr = dev->base_addr;
2440 ifr->ifr_map.irq = dev->irq;
2441 ifr->ifr_map.dma = dev->dma;
2442 ifr->ifr_map.port = dev->if_port;
2446 if (dev->set_config) {
2447 if (!netif_device_present(dev))
2449 return dev->set_config(dev, &ifr->ifr_map);
2454 if (!dev->set_multicast_list ||
2455 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2457 if (!netif_device_present(dev))
2459 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2463 if (!dev->set_multicast_list ||
2464 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2466 if (!netif_device_present(dev))
2468 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2472 ifr->ifr_ifindex = dev->ifindex;
2476 ifr->ifr_qlen = dev->tx_queue_len;
2480 if (ifr->ifr_qlen < 0)
2482 dev->tx_queue_len = ifr->ifr_qlen;
2486 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2487 return dev_change_name(dev, ifr->ifr_newname);
2490 * Unknown or private ioctl
2494 if ((cmd >= SIOCDEVPRIVATE &&
2495 cmd <= SIOCDEVPRIVATE + 15) ||
2496 cmd == SIOCBONDENSLAVE ||
2497 cmd == SIOCBONDRELEASE ||
2498 cmd == SIOCBONDSETHWADDR ||
2499 cmd == SIOCBONDSLAVEINFOQUERY ||
2500 cmd == SIOCBONDINFOQUERY ||
2501 cmd == SIOCBONDCHANGEACTIVE ||
2502 cmd == SIOCGMIIPHY ||
2503 cmd == SIOCGMIIREG ||
2504 cmd == SIOCSMIIREG ||
2505 cmd == SIOCBRADDIF ||
2506 cmd == SIOCBRDELIF ||
2507 cmd == SIOCWANDEV) {
2509 if (dev->do_ioctl) {
2510 if (netif_device_present(dev))
2511 err = dev->do_ioctl(dev, ifr,
2524 * This function handles all "interface"-type I/O control requests. The actual
2525 * 'doing' part of this is dev_ifsioc above.
2529 * dev_ioctl - network device ioctl
2530 * @cmd: command to issue
2531 * @arg: pointer to a struct ifreq in user space
2533 * Issue ioctl functions to devices. This is normally called by the
2534 * user space syscall interfaces but can sometimes be useful for
2535 * other purposes. The return value is the return from the syscall if
2536 * positive or a negative errno code on error.
2539 int dev_ioctl(unsigned int cmd, void __user *arg)
2545 /* One special case: SIOCGIFCONF takes ifconf argument
2546 and requires shared lock, because it sleeps writing
2550 if (cmd == SIOCGIFCONF) {
2552 ret = dev_ifconf((char __user *) arg);
2556 if (cmd == SIOCGIFNAME)
2557 return dev_ifname((struct ifreq __user *)arg);
2559 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2562 ifr.ifr_name[IFNAMSIZ-1] = 0;
2564 colon = strchr(ifr.ifr_name, ':');
2569 * See which interface the caller is talking about.
2574 * These ioctl calls:
2575 * - can be done by all.
2576 * - atomic and do not require locking.
2587 dev_load(ifr.ifr_name);
2588 read_lock(&dev_base_lock);
2589 ret = dev_ifsioc(&ifr, cmd);
2590 read_unlock(&dev_base_lock);
2594 if (copy_to_user(arg, &ifr,
2595 sizeof(struct ifreq)))
2601 dev_load(ifr.ifr_name);
2603 ret = dev_ethtool(&ifr);
2608 if (copy_to_user(arg, &ifr,
2609 sizeof(struct ifreq)))
2615 * These ioctl calls:
2616 * - require superuser power.
2617 * - require strict serialization.
2623 if (!capable(CAP_NET_ADMIN))
2625 dev_load(ifr.ifr_name);
2627 ret = dev_ifsioc(&ifr, cmd);
2632 if (copy_to_user(arg, &ifr,
2633 sizeof(struct ifreq)))
2639 * These ioctl calls:
2640 * - require superuser power.
2641 * - require strict serialization.
2642 * - do not return a value
2652 case SIOCSIFHWBROADCAST:
2655 case SIOCBONDENSLAVE:
2656 case SIOCBONDRELEASE:
2657 case SIOCBONDSETHWADDR:
2658 case SIOCBONDCHANGEACTIVE:
2661 if (!capable(CAP_NET_ADMIN))
2664 case SIOCBONDSLAVEINFOQUERY:
2665 case SIOCBONDINFOQUERY:
2666 dev_load(ifr.ifr_name);
2668 ret = dev_ifsioc(&ifr, cmd);
2673 /* Get the per device memory space. We can add this but
2674 * currently do not support it */
2676 /* Set the per device memory buffer space.
2677 * Not applicable in our case */
2682 * Unknown or private ioctl.
2685 if (cmd == SIOCWANDEV ||
2686 (cmd >= SIOCDEVPRIVATE &&
2687 cmd <= SIOCDEVPRIVATE + 15)) {
2688 dev_load(ifr.ifr_name);
2690 ret = dev_ifsioc(&ifr, cmd);
2692 if (!ret && copy_to_user(arg, &ifr,
2693 sizeof(struct ifreq)))
2697 #ifdef CONFIG_WIRELESS_EXT
2698 /* Take care of Wireless Extensions */
2699 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2700 /* If command is `set a parameter', or
2701 * `get the encoding parameters', check if
2702 * the user has the right to do it */
2703 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2704 || cmd == SIOCGIWENCODEEXT) {
2705 if (!capable(CAP_NET_ADMIN))
2708 dev_load(ifr.ifr_name);
2710 /* Follow me in net/core/wireless.c */
2711 ret = wireless_process_ioctl(&ifr, cmd);
2713 if (IW_IS_GET(cmd) &&
2714 copy_to_user(arg, &ifr,
2715 sizeof(struct ifreq)))
2719 #endif /* CONFIG_WIRELESS_EXT */
2726 * dev_new_index - allocate an ifindex
2728 * Returns a suitable unique value for a new device interface
2729 * number. The caller must hold the rtnl semaphore or the
2730 * dev_base_lock to be sure it remains unique.
2732 static int dev_new_index(void)
2738 if (!__dev_get_by_index(ifindex))
2743 static int dev_boot_phase = 1;
2745 /* Delayed registration/unregisteration */
2746 static DEFINE_SPINLOCK(net_todo_list_lock);
2747 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2749 static inline void net_set_todo(struct net_device *dev)
2751 spin_lock(&net_todo_list_lock);
2752 list_add_tail(&dev->todo_list, &net_todo_list);
2753 spin_unlock(&net_todo_list_lock);
2757 * register_netdevice - register a network device
2758 * @dev: device to register
2760 * Take a completed network device structure and add it to the kernel
2761 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2762 * chain. 0 is returned on success. A negative errno code is returned
2763 * on a failure to set up the device, or if the name is a duplicate.
2765 * Callers must hold the rtnl semaphore. You may want
2766 * register_netdev() instead of this.
2769 * The locking appears insufficient to guarantee two parallel registers
2770 * will not get the same name.
2773 int register_netdevice(struct net_device *dev)
2775 struct hlist_head *head;
2776 struct hlist_node *p;
2779 BUG_ON(dev_boot_phase);
2784 /* When net_device's are persistent, this will be fatal. */
2785 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2787 spin_lock_init(&dev->queue_lock);
2788 spin_lock_init(&dev->xmit_lock);
2789 dev->xmit_lock_owner = -1;
2790 #ifdef CONFIG_NET_CLS_ACT
2791 spin_lock_init(&dev->ingress_lock);
2794 ret = alloc_divert_blk(dev);
2800 /* Init, if this function is available */
2802 ret = dev->init(dev);
2810 if (!dev_valid_name(dev->name)) {
2815 dev->ifindex = dev_new_index();
2816 if (dev->iflink == -1)
2817 dev->iflink = dev->ifindex;
2819 /* Check for existence of name */
2820 head = dev_name_hash(dev->name);
2821 hlist_for_each(p, head) {
2822 struct net_device *d
2823 = hlist_entry(p, struct net_device, name_hlist);
2824 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2830 /* Fix illegal SG+CSUM combinations. */
2831 if ((dev->features & NETIF_F_SG) &&
2832 !(dev->features & (NETIF_F_IP_CSUM |
2834 NETIF_F_HW_CSUM))) {
2835 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2837 dev->features &= ~NETIF_F_SG;
2840 /* TSO requires that SG is present as well. */
2841 if ((dev->features & NETIF_F_TSO) &&
2842 !(dev->features & NETIF_F_SG)) {
2843 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2845 dev->features &= ~NETIF_F_TSO;
2847 if (dev->features & NETIF_F_UFO) {
2848 if (!(dev->features & NETIF_F_HW_CSUM)) {
2849 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2850 "NETIF_F_HW_CSUM feature.\n",
2852 dev->features &= ~NETIF_F_UFO;
2854 if (!(dev->features & NETIF_F_SG)) {
2855 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2856 "NETIF_F_SG feature.\n",
2858 dev->features &= ~NETIF_F_UFO;
2863 * nil rebuild_header routine,
2864 * that should be never called and used as just bug trap.
2867 if (!dev->rebuild_header)
2868 dev->rebuild_header = default_rebuild_header;
2870 ret = netdev_register_sysfs(dev);
2873 dev->reg_state = NETREG_REGISTERED;
2876 * Default initial state at registry is that the
2877 * device is present.
2880 set_bit(__LINK_STATE_PRESENT, &dev->state);
2883 dev_init_scheduler(dev);
2884 write_lock_bh(&dev_base_lock);
2886 dev_tail = &dev->next;
2887 hlist_add_head(&dev->name_hlist, head);
2888 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2890 write_unlock_bh(&dev_base_lock);
2892 /* Notify protocols, that a new device appeared. */
2893 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2900 free_divert_blk(dev);
2905 * register_netdev - register a network device
2906 * @dev: device to register
2908 * Take a completed network device structure and add it to the kernel
2909 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2910 * chain. 0 is returned on success. A negative errno code is returned
2911 * on a failure to set up the device, or if the name is a duplicate.
2913 * This is a wrapper around register_netdev that takes the rtnl semaphore
2914 * and expands the device name if you passed a format string to
2917 int register_netdev(struct net_device *dev)
2924 * If the name is a format string the caller wants us to do a
2927 if (strchr(dev->name, '%')) {
2928 err = dev_alloc_name(dev, dev->name);
2934 * Back compatibility hook. Kill this one in 2.5
2936 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2937 err = dev_alloc_name(dev, "eth%d");
2942 err = register_netdevice(dev);
2947 EXPORT_SYMBOL(register_netdev);
2950 * netdev_wait_allrefs - wait until all references are gone.
2952 * This is called when unregistering network devices.
2954 * Any protocol or device that holds a reference should register
2955 * for netdevice notification, and cleanup and put back the
2956 * reference if they receive an UNREGISTER event.
2957 * We can get stuck here if buggy protocols don't correctly
2960 static void netdev_wait_allrefs(struct net_device *dev)
2962 unsigned long rebroadcast_time, warning_time;
2964 rebroadcast_time = warning_time = jiffies;
2965 while (atomic_read(&dev->refcnt) != 0) {
2966 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2969 /* Rebroadcast unregister notification */
2970 raw_notifier_call_chain(&netdev_chain,
2971 NETDEV_UNREGISTER, dev);
2973 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2975 /* We must not have linkwatch events
2976 * pending on unregister. If this
2977 * happens, we simply run the queue
2978 * unscheduled, resulting in a noop
2981 linkwatch_run_queue();
2986 rebroadcast_time = jiffies;
2991 if (time_after(jiffies, warning_time + 10 * HZ)) {
2992 printk(KERN_EMERG "unregister_netdevice: "
2993 "waiting for %s to become free. Usage "
2995 dev->name, atomic_read(&dev->refcnt));
2996 warning_time = jiffies;
3005 * register_netdevice(x1);
3006 * register_netdevice(x2);
3008 * unregister_netdevice(y1);
3009 * unregister_netdevice(y2);
3015 * We are invoked by rtnl_unlock() after it drops the semaphore.
3016 * This allows us to deal with problems:
3017 * 1) We can delete sysfs objects which invoke hotplug
3018 * without deadlocking with linkwatch via keventd.
3019 * 2) Since we run with the RTNL semaphore not held, we can sleep
3020 * safely in order to wait for the netdev refcnt to drop to zero.
3022 static DEFINE_MUTEX(net_todo_run_mutex);
3023 void netdev_run_todo(void)
3025 struct list_head list = LIST_HEAD_INIT(list);
3027 /* Need to guard against multiple cpu's getting out of order. */
3028 mutex_lock(&net_todo_run_mutex);
3030 /* Not safe to do outside the semaphore. We must not return
3031 * until all unregister events invoked by the local processor
3032 * have been completed (either by this todo run, or one on
3035 if (list_empty(&net_todo_list))
3038 /* Snapshot list, allow later requests */
3039 spin_lock(&net_todo_list_lock);
3040 list_splice_init(&net_todo_list, &list);
3041 spin_unlock(&net_todo_list_lock);
3043 while (!list_empty(&list)) {
3044 struct net_device *dev
3045 = list_entry(list.next, struct net_device, todo_list);
3046 list_del(&dev->todo_list);
3048 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3049 printk(KERN_ERR "network todo '%s' but state %d\n",
3050 dev->name, dev->reg_state);
3055 netdev_unregister_sysfs(dev);
3056 dev->reg_state = NETREG_UNREGISTERED;
3058 netdev_wait_allrefs(dev);
3061 BUG_ON(atomic_read(&dev->refcnt));
3062 BUG_TRAP(!dev->ip_ptr);
3063 BUG_TRAP(!dev->ip6_ptr);
3064 BUG_TRAP(!dev->dn_ptr);
3066 /* It must be the very last action,
3067 * after this 'dev' may point to freed up memory.
3069 if (dev->destructor)
3070 dev->destructor(dev);
3074 mutex_unlock(&net_todo_run_mutex);
3078 * alloc_netdev - allocate network device
3079 * @sizeof_priv: size of private data to allocate space for
3080 * @name: device name format string
3081 * @setup: callback to initialize device
3083 * Allocates a struct net_device with private data area for driver use
3084 * and performs basic initialization.
3086 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3087 void (*setup)(struct net_device *))
3090 struct net_device *dev;
3093 /* ensure 32-byte alignment of both the device and private area */
3094 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3095 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3097 p = kzalloc(alloc_size, GFP_KERNEL);
3099 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3103 dev = (struct net_device *)
3104 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3105 dev->padded = (char *)dev - (char *)p;
3108 dev->priv = netdev_priv(dev);
3111 strcpy(dev->name, name);
3114 EXPORT_SYMBOL(alloc_netdev);
3117 * free_netdev - free network device
3120 * This function does the last stage of destroying an allocated device
3121 * interface. The reference to the device object is released.
3122 * If this is the last reference then it will be freed.
3124 void free_netdev(struct net_device *dev)
3127 /* Compatibility with error handling in drivers */
3128 if (dev->reg_state == NETREG_UNINITIALIZED) {
3129 kfree((char *)dev - dev->padded);
3133 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3134 dev->reg_state = NETREG_RELEASED;
3136 /* will free via class release */
3137 class_device_put(&dev->class_dev);
3139 kfree((char *)dev - dev->padded);
3143 /* Synchronize with packet receive processing. */
3144 void synchronize_net(void)
3151 * unregister_netdevice - remove device from the kernel
3154 * This function shuts down a device interface and removes it
3155 * from the kernel tables. On success 0 is returned, on a failure
3156 * a negative errno code is returned.
3158 * Callers must hold the rtnl semaphore. You may want
3159 * unregister_netdev() instead of this.
3162 int unregister_netdevice(struct net_device *dev)
3164 struct net_device *d, **dp;
3166 BUG_ON(dev_boot_phase);
3169 /* Some devices call without registering for initialization unwind. */
3170 if (dev->reg_state == NETREG_UNINITIALIZED) {
3171 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3172 "was registered\n", dev->name, dev);
3176 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3178 /* If device is running, close it first. */
3179 if (dev->flags & IFF_UP)
3182 /* And unlink it from device chain. */
3183 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3185 write_lock_bh(&dev_base_lock);
3186 hlist_del(&dev->name_hlist);
3187 hlist_del(&dev->index_hlist);
3188 if (dev_tail == &dev->next)
3191 write_unlock_bh(&dev_base_lock);
3196 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3201 dev->reg_state = NETREG_UNREGISTERING;
3205 /* Shutdown queueing discipline. */
3209 /* Notify protocols, that we are about to destroy
3210 this device. They should clean all the things.
3212 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3215 * Flush the multicast chain
3217 dev_mc_discard(dev);
3222 /* Notifier chain MUST detach us from master device. */
3223 BUG_TRAP(!dev->master);
3225 free_divert_blk(dev);
3227 /* Finish processing unregister after unlock */
3237 * unregister_netdev - remove device from the kernel
3240 * This function shuts down a device interface and removes it
3241 * from the kernel tables. On success 0 is returned, on a failure
3242 * a negative errno code is returned.
3244 * This is just a wrapper for unregister_netdevice that takes
3245 * the rtnl semaphore. In general you want to use this and not
3246 * unregister_netdevice.
3248 void unregister_netdev(struct net_device *dev)
3251 unregister_netdevice(dev);
3255 EXPORT_SYMBOL(unregister_netdev);
3257 #ifdef CONFIG_HOTPLUG_CPU
3258 static int dev_cpu_callback(struct notifier_block *nfb,
3259 unsigned long action,
3262 struct sk_buff **list_skb;
3263 struct net_device **list_net;
3264 struct sk_buff *skb;
3265 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3266 struct softnet_data *sd, *oldsd;
3268 if (action != CPU_DEAD)
3271 local_irq_disable();
3272 cpu = smp_processor_id();
3273 sd = &per_cpu(softnet_data, cpu);
3274 oldsd = &per_cpu(softnet_data, oldcpu);
3276 /* Find end of our completion_queue. */
3277 list_skb = &sd->completion_queue;
3279 list_skb = &(*list_skb)->next;
3280 /* Append completion queue from offline CPU. */
3281 *list_skb = oldsd->completion_queue;
3282 oldsd->completion_queue = NULL;
3284 /* Find end of our output_queue. */
3285 list_net = &sd->output_queue;
3287 list_net = &(*list_net)->next_sched;
3288 /* Append output queue from offline CPU. */
3289 *list_net = oldsd->output_queue;
3290 oldsd->output_queue = NULL;
3292 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3295 /* Process offline CPU's input_pkt_queue */
3296 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3301 #endif /* CONFIG_HOTPLUG_CPU */
3305 * Initialize the DEV module. At boot time this walks the device list and
3306 * unhooks any devices that fail to initialise (normally hardware not
3307 * present) and leaves us with a valid list of present and active devices.
3312 * This is called single threaded during boot, so no need
3313 * to take the rtnl semaphore.
3315 static int __init net_dev_init(void)
3317 int i, rc = -ENOMEM;
3319 BUG_ON(!dev_boot_phase);
3323 if (dev_proc_init())
3326 if (netdev_sysfs_init())
3329 INIT_LIST_HEAD(&ptype_all);
3330 for (i = 0; i < 16; i++)
3331 INIT_LIST_HEAD(&ptype_base[i]);
3333 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3334 INIT_HLIST_HEAD(&dev_name_head[i]);
3336 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3337 INIT_HLIST_HEAD(&dev_index_head[i]);
3340 * Initialise the packet receive queues.
3343 for_each_possible_cpu(i) {
3344 struct softnet_data *queue;
3346 queue = &per_cpu(softnet_data, i);
3347 skb_queue_head_init(&queue->input_pkt_queue);
3348 queue->completion_queue = NULL;
3349 INIT_LIST_HEAD(&queue->poll_list);
3350 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3351 queue->backlog_dev.weight = weight_p;
3352 queue->backlog_dev.poll = process_backlog;
3353 atomic_set(&queue->backlog_dev.refcnt, 1);
3358 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3359 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3361 hotcpu_notifier(dev_cpu_callback, 0);
3369 subsys_initcall(net_dev_init);
3371 EXPORT_SYMBOL(__dev_get_by_index);
3372 EXPORT_SYMBOL(__dev_get_by_name);
3373 EXPORT_SYMBOL(__dev_remove_pack);
3374 EXPORT_SYMBOL(__skb_linearize);
3375 EXPORT_SYMBOL(dev_valid_name);
3376 EXPORT_SYMBOL(dev_add_pack);
3377 EXPORT_SYMBOL(dev_alloc_name);
3378 EXPORT_SYMBOL(dev_close);
3379 EXPORT_SYMBOL(dev_get_by_flags);
3380 EXPORT_SYMBOL(dev_get_by_index);
3381 EXPORT_SYMBOL(dev_get_by_name);
3382 EXPORT_SYMBOL(dev_open);
3383 EXPORT_SYMBOL(dev_queue_xmit);
3384 EXPORT_SYMBOL(dev_remove_pack);
3385 EXPORT_SYMBOL(dev_set_allmulti);
3386 EXPORT_SYMBOL(dev_set_promiscuity);
3387 EXPORT_SYMBOL(dev_change_flags);
3388 EXPORT_SYMBOL(dev_set_mtu);
3389 EXPORT_SYMBOL(dev_set_mac_address);
3390 EXPORT_SYMBOL(free_netdev);
3391 EXPORT_SYMBOL(netdev_boot_setup_check);
3392 EXPORT_SYMBOL(netdev_set_master);
3393 EXPORT_SYMBOL(netdev_state_change);
3394 EXPORT_SYMBOL(netif_receive_skb);
3395 EXPORT_SYMBOL(netif_rx);
3396 EXPORT_SYMBOL(register_gifconf);
3397 EXPORT_SYMBOL(register_netdevice);
3398 EXPORT_SYMBOL(register_netdevice_notifier);
3399 EXPORT_SYMBOL(skb_checksum_help);
3400 EXPORT_SYMBOL(synchronize_net);
3401 EXPORT_SYMBOL(unregister_netdevice);
3402 EXPORT_SYMBOL(unregister_netdevice_notifier);
3403 EXPORT_SYMBOL(net_enable_timestamp);
3404 EXPORT_SYMBOL(net_disable_timestamp);
3405 EXPORT_SYMBOL(dev_get_flags);
3407 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3408 EXPORT_SYMBOL(br_handle_frame_hook);
3409 EXPORT_SYMBOL(br_fdb_get_hook);
3410 EXPORT_SYMBOL(br_fdb_put_hook);
3414 EXPORT_SYMBOL(dev_load);
3417 EXPORT_PER_CPU_SYMBOL(softnet_data);