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 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 BLOCKING_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. 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 blocking_notifier_call_chain(&netdev_chain,
740 NETDEV_CHANGENAME, dev);
747 * netdev_features_change - device changes fatures
748 * @dev: device to cause notification
750 * Called to indicate a device has changed features.
752 void netdev_features_change(struct net_device *dev)
754 blocking_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
756 EXPORT_SYMBOL(netdev_features_change);
759 * netdev_state_change - device changes state
760 * @dev: device to cause notification
762 * Called to indicate a device has changed state. This function calls
763 * the notifier chains for netdev_chain and sends a NEWLINK message
764 * to the routing socket.
766 void netdev_state_change(struct net_device *dev)
768 if (dev->flags & IFF_UP) {
769 blocking_notifier_call_chain(&netdev_chain,
771 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
776 * dev_load - load a network module
777 * @name: name of interface
779 * If a network interface is not present and the process has suitable
780 * privileges this function loads the module. If module loading is not
781 * available in this kernel then it becomes a nop.
784 void dev_load(const char *name)
786 struct net_device *dev;
788 read_lock(&dev_base_lock);
789 dev = __dev_get_by_name(name);
790 read_unlock(&dev_base_lock);
792 if (!dev && capable(CAP_SYS_MODULE))
793 request_module("%s", name);
796 static int default_rebuild_header(struct sk_buff *skb)
798 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
799 skb->dev ? skb->dev->name : "NULL!!!");
806 * dev_open - prepare an interface for use.
807 * @dev: device to open
809 * Takes a device from down to up state. The device's private open
810 * function is invoked and then the multicast lists are loaded. Finally
811 * the device is moved into the up state and a %NETDEV_UP message is
812 * sent to the netdev notifier chain.
814 * Calling this function on an active interface is a nop. On a failure
815 * a negative errno code is returned.
817 int dev_open(struct net_device *dev)
825 if (dev->flags & IFF_UP)
829 * Is it even present?
831 if (!netif_device_present(dev))
835 * Call device private open method
837 set_bit(__LINK_STATE_START, &dev->state);
839 ret = dev->open(dev);
841 clear_bit(__LINK_STATE_START, &dev->state);
845 * If it went open OK then:
852 dev->flags |= IFF_UP;
855 * Initialize multicasting status
860 * Wakeup transmit queue engine
865 * ... and announce new interface.
867 blocking_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
873 * dev_close - shutdown an interface.
874 * @dev: device to shutdown
876 * This function moves an active device into down state. A
877 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
878 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
881 int dev_close(struct net_device *dev)
883 if (!(dev->flags & IFF_UP))
887 * Tell people we are going down, so that they can
888 * prepare to death, when device is still operating.
890 blocking_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
894 clear_bit(__LINK_STATE_START, &dev->state);
896 /* Synchronize to scheduled poll. We cannot touch poll list,
897 * it can be even on different cpu. So just clear netif_running(),
898 * and wait when poll really will happen. Actually, the best place
899 * for this is inside dev->stop() after device stopped its irq
900 * engine, but this requires more changes in devices. */
902 smp_mb__after_clear_bit(); /* Commit netif_running(). */
903 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
909 * Call the device specific close. This cannot fail.
910 * Only if device is UP
912 * We allow it to be called even after a DETACH hot-plug
919 * Device is now down.
922 dev->flags &= ~IFF_UP;
925 * Tell people we are down
927 blocking_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
934 * Device change register/unregister. These are not inline or static
935 * as we export them to the world.
939 * register_netdevice_notifier - register a network notifier block
942 * Register a notifier to be called when network device events occur.
943 * The notifier passed is linked into the kernel structures and must
944 * not be reused until it has been unregistered. A negative errno code
945 * is returned on a failure.
947 * When registered all registration and up events are replayed
948 * to the new notifier to allow device to have a race free
949 * view of the network device list.
952 int register_netdevice_notifier(struct notifier_block *nb)
954 struct net_device *dev;
958 err = blocking_notifier_chain_register(&netdev_chain, nb);
960 for (dev = dev_base; dev; dev = dev->next) {
961 nb->notifier_call(nb, NETDEV_REGISTER, dev);
963 if (dev->flags & IFF_UP)
964 nb->notifier_call(nb, NETDEV_UP, dev);
972 * unregister_netdevice_notifier - unregister a network notifier block
975 * Unregister a notifier previously registered by
976 * register_netdevice_notifier(). The notifier is unlinked into the
977 * kernel structures and may then be reused. A negative errno code
978 * is returned on a failure.
981 int unregister_netdevice_notifier(struct notifier_block *nb)
986 err = blocking_notifier_chain_unregister(&netdev_chain, nb);
992 * call_netdevice_notifiers - call all network notifier blocks
993 * @val: value passed unmodified to notifier function
994 * @v: pointer passed unmodified to notifier function
996 * Call all network notifier blocks. Parameters and return value
997 * are as for blocking_notifier_call_chain().
1000 int call_netdevice_notifiers(unsigned long val, void *v)
1002 return blocking_notifier_call_chain(&netdev_chain, val, v);
1005 /* When > 0 there are consumers of rx skb time stamps */
1006 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1008 void net_enable_timestamp(void)
1010 atomic_inc(&netstamp_needed);
1013 void net_disable_timestamp(void)
1015 atomic_dec(&netstamp_needed);
1018 void __net_timestamp(struct sk_buff *skb)
1022 do_gettimeofday(&tv);
1023 skb_set_timestamp(skb, &tv);
1025 EXPORT_SYMBOL(__net_timestamp);
1027 static inline void net_timestamp(struct sk_buff *skb)
1029 if (atomic_read(&netstamp_needed))
1030 __net_timestamp(skb);
1032 skb->tstamp.off_sec = 0;
1033 skb->tstamp.off_usec = 0;
1038 * Support routine. Sends outgoing frames to any network
1039 * taps currently in use.
1042 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1044 struct packet_type *ptype;
1049 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1050 /* Never send packets back to the socket
1051 * they originated from - MvS (miquels@drinkel.ow.org)
1053 if ((ptype->dev == dev || !ptype->dev) &&
1054 (ptype->af_packet_priv == NULL ||
1055 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1056 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1060 /* skb->nh should be correctly
1061 set by sender, so that the second statement is
1062 just protection against buggy protocols.
1064 skb2->mac.raw = skb2->data;
1066 if (skb2->nh.raw < skb2->data ||
1067 skb2->nh.raw > skb2->tail) {
1068 if (net_ratelimit())
1069 printk(KERN_CRIT "protocol %04x is "
1071 skb2->protocol, dev->name);
1072 skb2->nh.raw = skb2->data;
1075 skb2->h.raw = skb2->nh.raw;
1076 skb2->pkt_type = PACKET_OUTGOING;
1077 ptype->func(skb2, skb->dev, ptype, skb->dev);
1084 void __netif_schedule(struct net_device *dev)
1086 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1087 unsigned long flags;
1088 struct softnet_data *sd;
1090 local_irq_save(flags);
1091 sd = &__get_cpu_var(softnet_data);
1092 dev->next_sched = sd->output_queue;
1093 sd->output_queue = dev;
1094 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1095 local_irq_restore(flags);
1098 EXPORT_SYMBOL(__netif_schedule);
1100 void __netif_rx_schedule(struct net_device *dev)
1102 unsigned long flags;
1104 local_irq_save(flags);
1106 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1108 dev->quota += dev->weight;
1110 dev->quota = dev->weight;
1111 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1112 local_irq_restore(flags);
1114 EXPORT_SYMBOL(__netif_rx_schedule);
1116 void dev_kfree_skb_any(struct sk_buff *skb)
1118 if (in_irq() || irqs_disabled())
1119 dev_kfree_skb_irq(skb);
1123 EXPORT_SYMBOL(dev_kfree_skb_any);
1127 void netif_device_detach(struct net_device *dev)
1129 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1130 netif_running(dev)) {
1131 netif_stop_queue(dev);
1134 EXPORT_SYMBOL(netif_device_detach);
1136 void netif_device_attach(struct net_device *dev)
1138 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1139 netif_running(dev)) {
1140 netif_wake_queue(dev);
1141 __netdev_watchdog_up(dev);
1144 EXPORT_SYMBOL(netif_device_attach);
1148 * Invalidate hardware checksum when packet is to be mangled, and
1149 * complete checksum manually on outgoing path.
1151 int skb_checksum_help(struct sk_buff *skb, int inward)
1154 int ret = 0, offset = skb->h.raw - skb->data;
1157 skb->ip_summed = CHECKSUM_NONE;
1161 if (skb_cloned(skb)) {
1162 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1167 BUG_ON(offset > (int)skb->len);
1168 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1170 offset = skb->tail - skb->h.raw;
1171 BUG_ON(offset <= 0);
1172 BUG_ON(skb->csum + 2 > offset);
1174 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1175 skb->ip_summed = CHECKSUM_NONE;
1180 /* Take action when hardware reception checksum errors are detected. */
1182 void netdev_rx_csum_fault(struct net_device *dev)
1184 if (net_ratelimit()) {
1185 printk(KERN_ERR "%s: hw csum failure.\n",
1186 dev ? dev->name : "<unknown>");
1190 EXPORT_SYMBOL(netdev_rx_csum_fault);
1193 #ifdef CONFIG_HIGHMEM
1194 /* Actually, we should eliminate this check as soon as we know, that:
1195 * 1. IOMMU is present and allows to map all the memory.
1196 * 2. No high memory really exists on this machine.
1199 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1203 if (dev->features & NETIF_F_HIGHDMA)
1206 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1207 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1213 #define illegal_highdma(dev, skb) (0)
1216 /* Keep head the same: replace data */
1217 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1222 struct skb_shared_info *ninfo;
1223 int headerlen = skb->data - skb->head;
1224 int expand = (skb->tail + skb->data_len) - skb->end;
1226 if (skb_shared(skb))
1232 size = skb->end - skb->head + expand;
1233 size = SKB_DATA_ALIGN(size);
1234 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1238 /* Copy entire thing */
1239 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1243 ninfo = (struct skb_shared_info*)(data + size);
1244 atomic_set(&ninfo->dataref, 1);
1245 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1246 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1247 ninfo->nr_frags = 0;
1248 ninfo->frag_list = NULL;
1250 /* Offset between the two in bytes */
1251 offset = data - skb->head;
1253 /* Free old data. */
1254 skb_release_data(skb);
1257 skb->end = data + size;
1259 /* Set up new pointers */
1260 skb->h.raw += offset;
1261 skb->nh.raw += offset;
1262 skb->mac.raw += offset;
1263 skb->tail += offset;
1264 skb->data += offset;
1266 /* We are no longer a clone, even if we were. */
1269 skb->tail += skb->data_len;
1274 #define HARD_TX_LOCK(dev, cpu) { \
1275 if ((dev->features & NETIF_F_LLTX) == 0) { \
1276 spin_lock(&dev->xmit_lock); \
1277 dev->xmit_lock_owner = cpu; \
1281 #define HARD_TX_UNLOCK(dev) { \
1282 if ((dev->features & NETIF_F_LLTX) == 0) { \
1283 dev->xmit_lock_owner = -1; \
1284 spin_unlock(&dev->xmit_lock); \
1289 * dev_queue_xmit - transmit a buffer
1290 * @skb: buffer to transmit
1292 * Queue a buffer for transmission to a network device. The caller must
1293 * have set the device and priority and built the buffer before calling
1294 * this function. The function can be called from an interrupt.
1296 * A negative errno code is returned on a failure. A success does not
1297 * guarantee the frame will be transmitted as it may be dropped due
1298 * to congestion or traffic shaping.
1300 * -----------------------------------------------------------------------------------
1301 * I notice this method can also return errors from the queue disciplines,
1302 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1305 * Regardless of the return value, the skb is consumed, so it is currently
1306 * difficult to retry a send to this method. (You can bump the ref count
1307 * before sending to hold a reference for retry if you are careful.)
1309 * When calling this method, interrupts MUST be enabled. This is because
1310 * the BH enable code must have IRQs enabled so that it will not deadlock.
1314 int dev_queue_xmit(struct sk_buff *skb)
1316 struct net_device *dev = skb->dev;
1320 if (skb_shinfo(skb)->frag_list &&
1321 !(dev->features & NETIF_F_FRAGLIST) &&
1322 __skb_linearize(skb, GFP_ATOMIC))
1325 /* Fragmented skb is linearized if device does not support SG,
1326 * or if at least one of fragments is in highmem and device
1327 * does not support DMA from it.
1329 if (skb_shinfo(skb)->nr_frags &&
1330 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1331 __skb_linearize(skb, GFP_ATOMIC))
1334 /* If packet is not checksummed and device does not support
1335 * checksumming for this protocol, complete checksumming here.
1337 if (skb->ip_summed == CHECKSUM_HW &&
1338 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1339 (!(dev->features & NETIF_F_IP_CSUM) ||
1340 skb->protocol != htons(ETH_P_IP))))
1341 if (skb_checksum_help(skb, 0))
1344 spin_lock_prefetch(&dev->queue_lock);
1346 /* Disable soft irqs for various locks below. Also
1347 * stops preemption for RCU.
1351 /* Updates of qdisc are serialized by queue_lock.
1352 * The struct Qdisc which is pointed to by qdisc is now a
1353 * rcu structure - it may be accessed without acquiring
1354 * a lock (but the structure may be stale.) The freeing of the
1355 * qdisc will be deferred until it's known that there are no
1356 * more references to it.
1358 * If the qdisc has an enqueue function, we still need to
1359 * hold the queue_lock before calling it, since queue_lock
1360 * also serializes access to the device queue.
1363 q = rcu_dereference(dev->qdisc);
1364 #ifdef CONFIG_NET_CLS_ACT
1365 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1368 /* Grab device queue */
1369 spin_lock(&dev->queue_lock);
1371 rc = q->enqueue(skb, q);
1375 spin_unlock(&dev->queue_lock);
1376 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1380 /* The device has no queue. Common case for software devices:
1381 loopback, all the sorts of tunnels...
1383 Really, it is unlikely that xmit_lock protection is necessary here.
1384 (f.e. loopback and IP tunnels are clean ignoring statistics
1386 However, it is possible, that they rely on protection
1389 Check this and shot the lock. It is not prone from deadlocks.
1390 Either shot noqueue qdisc, it is even simpler 8)
1392 if (dev->flags & IFF_UP) {
1393 int cpu = smp_processor_id(); /* ok because BHs are off */
1395 if (dev->xmit_lock_owner != cpu) {
1397 HARD_TX_LOCK(dev, cpu);
1399 if (!netif_queue_stopped(dev)) {
1401 dev_queue_xmit_nit(skb, dev);
1404 if (!dev->hard_start_xmit(skb, dev)) {
1405 HARD_TX_UNLOCK(dev);
1409 HARD_TX_UNLOCK(dev);
1410 if (net_ratelimit())
1411 printk(KERN_CRIT "Virtual device %s asks to "
1412 "queue packet!\n", dev->name);
1414 /* Recursion is detected! It is possible,
1416 if (net_ratelimit())
1417 printk(KERN_CRIT "Dead loop on virtual device "
1418 "%s, fix it urgently!\n", dev->name);
1434 /*=======================================================================
1436 =======================================================================*/
1438 int netdev_max_backlog = 1000;
1439 int netdev_budget = 300;
1440 int weight_p = 64; /* old backlog weight */
1442 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1446 * netif_rx - post buffer to the network code
1447 * @skb: buffer to post
1449 * This function receives a packet from a device driver and queues it for
1450 * the upper (protocol) levels to process. It always succeeds. The buffer
1451 * may be dropped during processing for congestion control or by the
1455 * NET_RX_SUCCESS (no congestion)
1456 * NET_RX_CN_LOW (low congestion)
1457 * NET_RX_CN_MOD (moderate congestion)
1458 * NET_RX_CN_HIGH (high congestion)
1459 * NET_RX_DROP (packet was dropped)
1463 int netif_rx(struct sk_buff *skb)
1465 struct softnet_data *queue;
1466 unsigned long flags;
1468 /* if netpoll wants it, pretend we never saw it */
1469 if (netpoll_rx(skb))
1472 if (!skb->tstamp.off_sec)
1476 * The code is rearranged so that the path is the most
1477 * short when CPU is congested, but is still operating.
1479 local_irq_save(flags);
1480 queue = &__get_cpu_var(softnet_data);
1482 __get_cpu_var(netdev_rx_stat).total++;
1483 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1484 if (queue->input_pkt_queue.qlen) {
1487 __skb_queue_tail(&queue->input_pkt_queue, skb);
1488 local_irq_restore(flags);
1489 return NET_RX_SUCCESS;
1492 netif_rx_schedule(&queue->backlog_dev);
1496 __get_cpu_var(netdev_rx_stat).dropped++;
1497 local_irq_restore(flags);
1503 int netif_rx_ni(struct sk_buff *skb)
1508 err = netif_rx(skb);
1509 if (local_softirq_pending())
1516 EXPORT_SYMBOL(netif_rx_ni);
1518 static inline struct net_device *skb_bond(struct sk_buff *skb)
1520 struct net_device *dev = skb->dev;
1524 * On bonding slaves other than the currently active
1525 * slave, suppress duplicates except for 802.3ad
1526 * ETH_P_SLOW and alb non-mcast/bcast.
1528 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1529 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1530 if (skb->pkt_type != PACKET_BROADCAST &&
1531 skb->pkt_type != PACKET_MULTICAST)
1535 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1536 skb->protocol == __constant_htons(ETH_P_SLOW))
1543 skb->dev = dev->master;
1549 static void net_tx_action(struct softirq_action *h)
1551 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1553 if (sd->completion_queue) {
1554 struct sk_buff *clist;
1556 local_irq_disable();
1557 clist = sd->completion_queue;
1558 sd->completion_queue = NULL;
1562 struct sk_buff *skb = clist;
1563 clist = clist->next;
1565 BUG_TRAP(!atomic_read(&skb->users));
1570 if (sd->output_queue) {
1571 struct net_device *head;
1573 local_irq_disable();
1574 head = sd->output_queue;
1575 sd->output_queue = NULL;
1579 struct net_device *dev = head;
1580 head = head->next_sched;
1582 smp_mb__before_clear_bit();
1583 clear_bit(__LINK_STATE_SCHED, &dev->state);
1585 if (spin_trylock(&dev->queue_lock)) {
1587 spin_unlock(&dev->queue_lock);
1589 netif_schedule(dev);
1595 static __inline__ int deliver_skb(struct sk_buff *skb,
1596 struct packet_type *pt_prev,
1597 struct net_device *orig_dev)
1599 atomic_inc(&skb->users);
1600 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1603 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1604 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1606 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1607 unsigned char *addr);
1608 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1610 static __inline__ int handle_bridge(struct sk_buff **pskb,
1611 struct packet_type **pt_prev, int *ret,
1612 struct net_device *orig_dev)
1614 struct net_bridge_port *port;
1616 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1617 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1621 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1625 return br_handle_frame_hook(port, pskb);
1628 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1631 #ifdef CONFIG_NET_CLS_ACT
1632 /* TODO: Maybe we should just force sch_ingress to be compiled in
1633 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1634 * a compare and 2 stores extra right now if we dont have it on
1635 * but have CONFIG_NET_CLS_ACT
1636 * NOTE: This doesnt stop any functionality; if you dont have
1637 * the ingress scheduler, you just cant add policies on ingress.
1640 static int ing_filter(struct sk_buff *skb)
1643 struct net_device *dev = skb->dev;
1644 int result = TC_ACT_OK;
1646 if (dev->qdisc_ingress) {
1647 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1648 if (MAX_RED_LOOP < ttl++) {
1649 printk("Redir loop detected Dropping packet (%s->%s)\n",
1650 skb->input_dev->name, skb->dev->name);
1654 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1656 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1658 spin_lock(&dev->ingress_lock);
1659 if ((q = dev->qdisc_ingress) != NULL)
1660 result = q->enqueue(skb, q);
1661 spin_unlock(&dev->ingress_lock);
1669 int netif_receive_skb(struct sk_buff *skb)
1671 struct packet_type *ptype, *pt_prev;
1672 struct net_device *orig_dev;
1673 int ret = NET_RX_DROP;
1674 unsigned short type;
1676 /* if we've gotten here through NAPI, check netpoll */
1677 if (skb->dev->poll && netpoll_rx(skb))
1680 if (!skb->tstamp.off_sec)
1683 if (!skb->input_dev)
1684 skb->input_dev = skb->dev;
1686 orig_dev = skb_bond(skb);
1691 __get_cpu_var(netdev_rx_stat).total++;
1693 skb->h.raw = skb->nh.raw = skb->data;
1694 skb->mac_len = skb->nh.raw - skb->mac.raw;
1700 #ifdef CONFIG_NET_CLS_ACT
1701 if (skb->tc_verd & TC_NCLS) {
1702 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1707 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1708 if (!ptype->dev || ptype->dev == skb->dev) {
1710 ret = deliver_skb(skb, pt_prev, orig_dev);
1715 #ifdef CONFIG_NET_CLS_ACT
1717 ret = deliver_skb(skb, pt_prev, orig_dev);
1718 pt_prev = NULL; /* noone else should process this after*/
1720 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1723 ret = ing_filter(skb);
1725 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1734 handle_diverter(skb);
1736 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1739 type = skb->protocol;
1740 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1741 if (ptype->type == type &&
1742 (!ptype->dev || ptype->dev == skb->dev)) {
1744 ret = deliver_skb(skb, pt_prev, orig_dev);
1750 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1753 /* Jamal, now you will not able to escape explaining
1754 * me how you were going to use this. :-)
1764 static int process_backlog(struct net_device *backlog_dev, int *budget)
1767 int quota = min(backlog_dev->quota, *budget);
1768 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1769 unsigned long start_time = jiffies;
1771 backlog_dev->weight = weight_p;
1773 struct sk_buff *skb;
1774 struct net_device *dev;
1776 local_irq_disable();
1777 skb = __skb_dequeue(&queue->input_pkt_queue);
1784 netif_receive_skb(skb);
1790 if (work >= quota || jiffies - start_time > 1)
1795 backlog_dev->quota -= work;
1800 backlog_dev->quota -= work;
1803 list_del(&backlog_dev->poll_list);
1804 smp_mb__before_clear_bit();
1805 netif_poll_enable(backlog_dev);
1811 static void net_rx_action(struct softirq_action *h)
1813 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1814 unsigned long start_time = jiffies;
1815 int budget = netdev_budget;
1818 local_irq_disable();
1820 while (!list_empty(&queue->poll_list)) {
1821 struct net_device *dev;
1823 if (budget <= 0 || jiffies - start_time > 1)
1828 dev = list_entry(queue->poll_list.next,
1829 struct net_device, poll_list);
1830 have = netpoll_poll_lock(dev);
1832 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1833 netpoll_poll_unlock(have);
1834 local_irq_disable();
1835 list_move_tail(&dev->poll_list, &queue->poll_list);
1837 dev->quota += dev->weight;
1839 dev->quota = dev->weight;
1841 netpoll_poll_unlock(have);
1843 local_irq_disable();
1851 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1852 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1856 static gifconf_func_t * gifconf_list [NPROTO];
1859 * register_gifconf - register a SIOCGIF handler
1860 * @family: Address family
1861 * @gifconf: Function handler
1863 * Register protocol dependent address dumping routines. The handler
1864 * that is passed must not be freed or reused until it has been replaced
1865 * by another handler.
1867 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1869 if (family >= NPROTO)
1871 gifconf_list[family] = gifconf;
1877 * Map an interface index to its name (SIOCGIFNAME)
1881 * We need this ioctl for efficient implementation of the
1882 * if_indextoname() function required by the IPv6 API. Without
1883 * it, we would have to search all the interfaces to find a
1887 static int dev_ifname(struct ifreq __user *arg)
1889 struct net_device *dev;
1893 * Fetch the caller's info block.
1896 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1899 read_lock(&dev_base_lock);
1900 dev = __dev_get_by_index(ifr.ifr_ifindex);
1902 read_unlock(&dev_base_lock);
1906 strcpy(ifr.ifr_name, dev->name);
1907 read_unlock(&dev_base_lock);
1909 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1915 * Perform a SIOCGIFCONF call. This structure will change
1916 * size eventually, and there is nothing I can do about it.
1917 * Thus we will need a 'compatibility mode'.
1920 static int dev_ifconf(char __user *arg)
1923 struct net_device *dev;
1930 * Fetch the caller's info block.
1933 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1940 * Loop over the interfaces, and write an info block for each.
1944 for (dev = dev_base; dev; dev = dev->next) {
1945 for (i = 0; i < NPROTO; i++) {
1946 if (gifconf_list[i]) {
1949 done = gifconf_list[i](dev, NULL, 0);
1951 done = gifconf_list[i](dev, pos + total,
1961 * All done. Write the updated control block back to the caller.
1963 ifc.ifc_len = total;
1966 * Both BSD and Solaris return 0 here, so we do too.
1968 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1971 #ifdef CONFIG_PROC_FS
1973 * This is invoked by the /proc filesystem handler to display a device
1976 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1978 struct net_device *dev;
1981 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1983 return i == pos ? dev : NULL;
1986 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1988 read_lock(&dev_base_lock);
1989 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1992 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1995 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1998 void dev_seq_stop(struct seq_file *seq, void *v)
2000 read_unlock(&dev_base_lock);
2003 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2005 if (dev->get_stats) {
2006 struct net_device_stats *stats = dev->get_stats(dev);
2008 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2009 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2010 dev->name, stats->rx_bytes, stats->rx_packets,
2012 stats->rx_dropped + stats->rx_missed_errors,
2013 stats->rx_fifo_errors,
2014 stats->rx_length_errors + stats->rx_over_errors +
2015 stats->rx_crc_errors + stats->rx_frame_errors,
2016 stats->rx_compressed, stats->multicast,
2017 stats->tx_bytes, stats->tx_packets,
2018 stats->tx_errors, stats->tx_dropped,
2019 stats->tx_fifo_errors, stats->collisions,
2020 stats->tx_carrier_errors +
2021 stats->tx_aborted_errors +
2022 stats->tx_window_errors +
2023 stats->tx_heartbeat_errors,
2024 stats->tx_compressed);
2026 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2030 * Called from the PROCfs module. This now uses the new arbitrary sized
2031 * /proc/net interface to create /proc/net/dev
2033 static int dev_seq_show(struct seq_file *seq, void *v)
2035 if (v == SEQ_START_TOKEN)
2036 seq_puts(seq, "Inter-| Receive "
2038 " face |bytes packets errs drop fifo frame "
2039 "compressed multicast|bytes packets errs "
2040 "drop fifo colls carrier compressed\n");
2042 dev_seq_printf_stats(seq, v);
2046 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2048 struct netif_rx_stats *rc = NULL;
2050 while (*pos < NR_CPUS)
2051 if (cpu_online(*pos)) {
2052 rc = &per_cpu(netdev_rx_stat, *pos);
2059 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2061 return softnet_get_online(pos);
2064 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2067 return softnet_get_online(pos);
2070 static void softnet_seq_stop(struct seq_file *seq, void *v)
2074 static int softnet_seq_show(struct seq_file *seq, void *v)
2076 struct netif_rx_stats *s = v;
2078 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2079 s->total, s->dropped, s->time_squeeze, 0,
2080 0, 0, 0, 0, /* was fastroute */
2085 static struct seq_operations dev_seq_ops = {
2086 .start = dev_seq_start,
2087 .next = dev_seq_next,
2088 .stop = dev_seq_stop,
2089 .show = dev_seq_show,
2092 static int dev_seq_open(struct inode *inode, struct file *file)
2094 return seq_open(file, &dev_seq_ops);
2097 static struct file_operations dev_seq_fops = {
2098 .owner = THIS_MODULE,
2099 .open = dev_seq_open,
2101 .llseek = seq_lseek,
2102 .release = seq_release,
2105 static struct seq_operations softnet_seq_ops = {
2106 .start = softnet_seq_start,
2107 .next = softnet_seq_next,
2108 .stop = softnet_seq_stop,
2109 .show = softnet_seq_show,
2112 static int softnet_seq_open(struct inode *inode, struct file *file)
2114 return seq_open(file, &softnet_seq_ops);
2117 static struct file_operations softnet_seq_fops = {
2118 .owner = THIS_MODULE,
2119 .open = softnet_seq_open,
2121 .llseek = seq_lseek,
2122 .release = seq_release,
2125 #ifdef CONFIG_WIRELESS_EXT
2126 extern int wireless_proc_init(void);
2128 #define wireless_proc_init() 0
2131 static int __init dev_proc_init(void)
2135 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2137 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2139 if (wireless_proc_init())
2145 proc_net_remove("softnet_stat");
2147 proc_net_remove("dev");
2151 #define dev_proc_init() 0
2152 #endif /* CONFIG_PROC_FS */
2156 * netdev_set_master - set up master/slave pair
2157 * @slave: slave device
2158 * @master: new master device
2160 * Changes the master device of the slave. Pass %NULL to break the
2161 * bonding. The caller must hold the RTNL semaphore. On a failure
2162 * a negative errno code is returned. On success the reference counts
2163 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2164 * function returns zero.
2166 int netdev_set_master(struct net_device *slave, struct net_device *master)
2168 struct net_device *old = slave->master;
2178 slave->master = master;
2186 slave->flags |= IFF_SLAVE;
2188 slave->flags &= ~IFF_SLAVE;
2190 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2195 * dev_set_promiscuity - update promiscuity count on a device
2199 * Add or remove promsicuity from a device. While the count in the device
2200 * remains above zero the interface remains promiscuous. Once it hits zero
2201 * the device reverts back to normal filtering operation. A negative inc
2202 * value is used to drop promiscuity on the device.
2204 void dev_set_promiscuity(struct net_device *dev, int inc)
2206 unsigned short old_flags = dev->flags;
2208 if ((dev->promiscuity += inc) == 0)
2209 dev->flags &= ~IFF_PROMISC;
2211 dev->flags |= IFF_PROMISC;
2212 if (dev->flags != old_flags) {
2214 printk(KERN_INFO "device %s %s promiscuous mode\n",
2215 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2217 audit_log(current->audit_context, GFP_ATOMIC,
2218 AUDIT_ANOM_PROMISCUOUS,
2219 "dev=%s prom=%d old_prom=%d auid=%u",
2220 dev->name, (dev->flags & IFF_PROMISC),
2221 (old_flags & IFF_PROMISC),
2222 audit_get_loginuid(current->audit_context));
2227 * dev_set_allmulti - update allmulti count on a device
2231 * Add or remove reception of all multicast frames to a device. While the
2232 * count in the device remains above zero the interface remains listening
2233 * to all interfaces. Once it hits zero the device reverts back to normal
2234 * filtering operation. A negative @inc value is used to drop the counter
2235 * when releasing a resource needing all multicasts.
2238 void dev_set_allmulti(struct net_device *dev, int inc)
2240 unsigned short old_flags = dev->flags;
2242 dev->flags |= IFF_ALLMULTI;
2243 if ((dev->allmulti += inc) == 0)
2244 dev->flags &= ~IFF_ALLMULTI;
2245 if (dev->flags ^ old_flags)
2249 unsigned dev_get_flags(const struct net_device *dev)
2253 flags = (dev->flags & ~(IFF_PROMISC |
2258 (dev->gflags & (IFF_PROMISC |
2261 if (netif_running(dev)) {
2262 if (netif_oper_up(dev))
2263 flags |= IFF_RUNNING;
2264 if (netif_carrier_ok(dev))
2265 flags |= IFF_LOWER_UP;
2266 if (netif_dormant(dev))
2267 flags |= IFF_DORMANT;
2273 int dev_change_flags(struct net_device *dev, unsigned flags)
2276 int old_flags = dev->flags;
2279 * Set the flags on our device.
2282 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2283 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2285 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2289 * Load in the correct multicast list now the flags have changed.
2295 * Have we downed the interface. We handle IFF_UP ourselves
2296 * according to user attempts to set it, rather than blindly
2301 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2302 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2308 if (dev->flags & IFF_UP &&
2309 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2311 blocking_notifier_call_chain(&netdev_chain,
2312 NETDEV_CHANGE, dev);
2314 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2315 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2316 dev->gflags ^= IFF_PROMISC;
2317 dev_set_promiscuity(dev, inc);
2320 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2321 is important. Some (broken) drivers set IFF_PROMISC, when
2322 IFF_ALLMULTI is requested not asking us and not reporting.
2324 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2325 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2326 dev->gflags ^= IFF_ALLMULTI;
2327 dev_set_allmulti(dev, inc);
2330 if (old_flags ^ dev->flags)
2331 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2336 int dev_set_mtu(struct net_device *dev, int new_mtu)
2340 if (new_mtu == dev->mtu)
2343 /* MTU must be positive. */
2347 if (!netif_device_present(dev))
2351 if (dev->change_mtu)
2352 err = dev->change_mtu(dev, new_mtu);
2355 if (!err && dev->flags & IFF_UP)
2356 blocking_notifier_call_chain(&netdev_chain,
2357 NETDEV_CHANGEMTU, dev);
2361 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2365 if (!dev->set_mac_address)
2367 if (sa->sa_family != dev->type)
2369 if (!netif_device_present(dev))
2371 err = dev->set_mac_address(dev, sa);
2373 blocking_notifier_call_chain(&netdev_chain,
2374 NETDEV_CHANGEADDR, dev);
2379 * Perform the SIOCxIFxxx calls.
2381 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2384 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2390 case SIOCGIFFLAGS: /* Get interface flags */
2391 ifr->ifr_flags = dev_get_flags(dev);
2394 case SIOCSIFFLAGS: /* Set interface flags */
2395 return dev_change_flags(dev, ifr->ifr_flags);
2397 case SIOCGIFMETRIC: /* Get the metric on the interface
2398 (currently unused) */
2399 ifr->ifr_metric = 0;
2402 case SIOCSIFMETRIC: /* Set the metric on the interface
2403 (currently unused) */
2406 case SIOCGIFMTU: /* Get the MTU of a device */
2407 ifr->ifr_mtu = dev->mtu;
2410 case SIOCSIFMTU: /* Set the MTU of a device */
2411 return dev_set_mtu(dev, ifr->ifr_mtu);
2415 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2417 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2418 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2419 ifr->ifr_hwaddr.sa_family = dev->type;
2423 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2425 case SIOCSIFHWBROADCAST:
2426 if (ifr->ifr_hwaddr.sa_family != dev->type)
2428 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2429 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2430 blocking_notifier_call_chain(&netdev_chain,
2431 NETDEV_CHANGEADDR, dev);
2435 ifr->ifr_map.mem_start = dev->mem_start;
2436 ifr->ifr_map.mem_end = dev->mem_end;
2437 ifr->ifr_map.base_addr = dev->base_addr;
2438 ifr->ifr_map.irq = dev->irq;
2439 ifr->ifr_map.dma = dev->dma;
2440 ifr->ifr_map.port = dev->if_port;
2444 if (dev->set_config) {
2445 if (!netif_device_present(dev))
2447 return dev->set_config(dev, &ifr->ifr_map);
2452 if (!dev->set_multicast_list ||
2453 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2455 if (!netif_device_present(dev))
2457 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2461 if (!dev->set_multicast_list ||
2462 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2464 if (!netif_device_present(dev))
2466 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2470 ifr->ifr_ifindex = dev->ifindex;
2474 ifr->ifr_qlen = dev->tx_queue_len;
2478 if (ifr->ifr_qlen < 0)
2480 dev->tx_queue_len = ifr->ifr_qlen;
2484 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2485 return dev_change_name(dev, ifr->ifr_newname);
2488 * Unknown or private ioctl
2492 if ((cmd >= SIOCDEVPRIVATE &&
2493 cmd <= SIOCDEVPRIVATE + 15) ||
2494 cmd == SIOCBONDENSLAVE ||
2495 cmd == SIOCBONDRELEASE ||
2496 cmd == SIOCBONDSETHWADDR ||
2497 cmd == SIOCBONDSLAVEINFOQUERY ||
2498 cmd == SIOCBONDINFOQUERY ||
2499 cmd == SIOCBONDCHANGEACTIVE ||
2500 cmd == SIOCGMIIPHY ||
2501 cmd == SIOCGMIIREG ||
2502 cmd == SIOCSMIIREG ||
2503 cmd == SIOCBRADDIF ||
2504 cmd == SIOCBRDELIF ||
2505 cmd == SIOCWANDEV) {
2507 if (dev->do_ioctl) {
2508 if (netif_device_present(dev))
2509 err = dev->do_ioctl(dev, ifr,
2522 * This function handles all "interface"-type I/O control requests. The actual
2523 * 'doing' part of this is dev_ifsioc above.
2527 * dev_ioctl - network device ioctl
2528 * @cmd: command to issue
2529 * @arg: pointer to a struct ifreq in user space
2531 * Issue ioctl functions to devices. This is normally called by the
2532 * user space syscall interfaces but can sometimes be useful for
2533 * other purposes. The return value is the return from the syscall if
2534 * positive or a negative errno code on error.
2537 int dev_ioctl(unsigned int cmd, void __user *arg)
2543 /* One special case: SIOCGIFCONF takes ifconf argument
2544 and requires shared lock, because it sleeps writing
2548 if (cmd == SIOCGIFCONF) {
2550 ret = dev_ifconf((char __user *) arg);
2554 if (cmd == SIOCGIFNAME)
2555 return dev_ifname((struct ifreq __user *)arg);
2557 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2560 ifr.ifr_name[IFNAMSIZ-1] = 0;
2562 colon = strchr(ifr.ifr_name, ':');
2567 * See which interface the caller is talking about.
2572 * These ioctl calls:
2573 * - can be done by all.
2574 * - atomic and do not require locking.
2585 dev_load(ifr.ifr_name);
2586 read_lock(&dev_base_lock);
2587 ret = dev_ifsioc(&ifr, cmd);
2588 read_unlock(&dev_base_lock);
2592 if (copy_to_user(arg, &ifr,
2593 sizeof(struct ifreq)))
2599 dev_load(ifr.ifr_name);
2601 ret = dev_ethtool(&ifr);
2606 if (copy_to_user(arg, &ifr,
2607 sizeof(struct ifreq)))
2613 * These ioctl calls:
2614 * - require superuser power.
2615 * - require strict serialization.
2621 if (!capable(CAP_NET_ADMIN))
2623 dev_load(ifr.ifr_name);
2625 ret = dev_ifsioc(&ifr, cmd);
2630 if (copy_to_user(arg, &ifr,
2631 sizeof(struct ifreq)))
2637 * These ioctl calls:
2638 * - require superuser power.
2639 * - require strict serialization.
2640 * - do not return a value
2650 case SIOCSIFHWBROADCAST:
2653 case SIOCBONDENSLAVE:
2654 case SIOCBONDRELEASE:
2655 case SIOCBONDSETHWADDR:
2656 case SIOCBONDCHANGEACTIVE:
2659 if (!capable(CAP_NET_ADMIN))
2662 case SIOCBONDSLAVEINFOQUERY:
2663 case SIOCBONDINFOQUERY:
2664 dev_load(ifr.ifr_name);
2666 ret = dev_ifsioc(&ifr, cmd);
2671 /* Get the per device memory space. We can add this but
2672 * currently do not support it */
2674 /* Set the per device memory buffer space.
2675 * Not applicable in our case */
2680 * Unknown or private ioctl.
2683 if (cmd == SIOCWANDEV ||
2684 (cmd >= SIOCDEVPRIVATE &&
2685 cmd <= SIOCDEVPRIVATE + 15)) {
2686 dev_load(ifr.ifr_name);
2688 ret = dev_ifsioc(&ifr, cmd);
2690 if (!ret && copy_to_user(arg, &ifr,
2691 sizeof(struct ifreq)))
2695 #ifdef CONFIG_WIRELESS_EXT
2696 /* Take care of Wireless Extensions */
2697 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2698 /* If command is `set a parameter', or
2699 * `get the encoding parameters', check if
2700 * the user has the right to do it */
2701 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2702 if (!capable(CAP_NET_ADMIN))
2705 dev_load(ifr.ifr_name);
2707 /* Follow me in net/core/wireless.c */
2708 ret = wireless_process_ioctl(&ifr, cmd);
2710 if (IW_IS_GET(cmd) &&
2711 copy_to_user(arg, &ifr,
2712 sizeof(struct ifreq)))
2716 #endif /* CONFIG_WIRELESS_EXT */
2723 * dev_new_index - allocate an ifindex
2725 * Returns a suitable unique value for a new device interface
2726 * number. The caller must hold the rtnl semaphore or the
2727 * dev_base_lock to be sure it remains unique.
2729 static int dev_new_index(void)
2735 if (!__dev_get_by_index(ifindex))
2740 static int dev_boot_phase = 1;
2742 /* Delayed registration/unregisteration */
2743 static DEFINE_SPINLOCK(net_todo_list_lock);
2744 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2746 static inline void net_set_todo(struct net_device *dev)
2748 spin_lock(&net_todo_list_lock);
2749 list_add_tail(&dev->todo_list, &net_todo_list);
2750 spin_unlock(&net_todo_list_lock);
2754 * register_netdevice - register a network device
2755 * @dev: device to register
2757 * Take a completed network device structure and add it to the kernel
2758 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2759 * chain. 0 is returned on success. A negative errno code is returned
2760 * on a failure to set up the device, or if the name is a duplicate.
2762 * Callers must hold the rtnl semaphore. You may want
2763 * register_netdev() instead of this.
2766 * The locking appears insufficient to guarantee two parallel registers
2767 * will not get the same name.
2770 int register_netdevice(struct net_device *dev)
2772 struct hlist_head *head;
2773 struct hlist_node *p;
2776 BUG_ON(dev_boot_phase);
2779 /* When net_device's are persistent, this will be fatal. */
2780 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2782 spin_lock_init(&dev->queue_lock);
2783 spin_lock_init(&dev->xmit_lock);
2784 dev->xmit_lock_owner = -1;
2785 #ifdef CONFIG_NET_CLS_ACT
2786 spin_lock_init(&dev->ingress_lock);
2789 ret = alloc_divert_blk(dev);
2795 /* Init, if this function is available */
2797 ret = dev->init(dev);
2805 if (!dev_valid_name(dev->name)) {
2810 dev->ifindex = dev_new_index();
2811 if (dev->iflink == -1)
2812 dev->iflink = dev->ifindex;
2814 /* Check for existence of name */
2815 head = dev_name_hash(dev->name);
2816 hlist_for_each(p, head) {
2817 struct net_device *d
2818 = hlist_entry(p, struct net_device, name_hlist);
2819 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2825 /* Fix illegal SG+CSUM combinations. */
2826 if ((dev->features & NETIF_F_SG) &&
2827 !(dev->features & (NETIF_F_IP_CSUM |
2829 NETIF_F_HW_CSUM))) {
2830 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2832 dev->features &= ~NETIF_F_SG;
2835 /* TSO requires that SG is present as well. */
2836 if ((dev->features & NETIF_F_TSO) &&
2837 !(dev->features & NETIF_F_SG)) {
2838 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2840 dev->features &= ~NETIF_F_TSO;
2842 if (dev->features & NETIF_F_UFO) {
2843 if (!(dev->features & NETIF_F_HW_CSUM)) {
2844 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2845 "NETIF_F_HW_CSUM feature.\n",
2847 dev->features &= ~NETIF_F_UFO;
2849 if (!(dev->features & NETIF_F_SG)) {
2850 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2851 "NETIF_F_SG feature.\n",
2853 dev->features &= ~NETIF_F_UFO;
2858 * nil rebuild_header routine,
2859 * that should be never called and used as just bug trap.
2862 if (!dev->rebuild_header)
2863 dev->rebuild_header = default_rebuild_header;
2866 * Default initial state at registry is that the
2867 * device is present.
2870 set_bit(__LINK_STATE_PRESENT, &dev->state);
2873 dev_init_scheduler(dev);
2874 write_lock_bh(&dev_base_lock);
2876 dev_tail = &dev->next;
2877 hlist_add_head(&dev->name_hlist, head);
2878 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2880 dev->reg_state = NETREG_REGISTERING;
2881 write_unlock_bh(&dev_base_lock);
2883 /* Notify protocols, that a new device appeared. */
2884 blocking_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2886 /* Finish registration after unlock */
2893 free_divert_blk(dev);
2898 * register_netdev - register a network device
2899 * @dev: device to register
2901 * Take a completed network device structure and add it to the kernel
2902 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2903 * chain. 0 is returned on success. A negative errno code is returned
2904 * on a failure to set up the device, or if the name is a duplicate.
2906 * This is a wrapper around register_netdev that takes the rtnl semaphore
2907 * and expands the device name if you passed a format string to
2910 int register_netdev(struct net_device *dev)
2917 * If the name is a format string the caller wants us to do a
2920 if (strchr(dev->name, '%')) {
2921 err = dev_alloc_name(dev, dev->name);
2927 * Back compatibility hook. Kill this one in 2.5
2929 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2930 err = dev_alloc_name(dev, "eth%d");
2935 err = register_netdevice(dev);
2940 EXPORT_SYMBOL(register_netdev);
2943 * netdev_wait_allrefs - wait until all references are gone.
2945 * This is called when unregistering network devices.
2947 * Any protocol or device that holds a reference should register
2948 * for netdevice notification, and cleanup and put back the
2949 * reference if they receive an UNREGISTER event.
2950 * We can get stuck here if buggy protocols don't correctly
2953 static void netdev_wait_allrefs(struct net_device *dev)
2955 unsigned long rebroadcast_time, warning_time;
2957 rebroadcast_time = warning_time = jiffies;
2958 while (atomic_read(&dev->refcnt) != 0) {
2959 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2962 /* Rebroadcast unregister notification */
2963 blocking_notifier_call_chain(&netdev_chain,
2964 NETDEV_UNREGISTER, dev);
2966 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2968 /* We must not have linkwatch events
2969 * pending on unregister. If this
2970 * happens, we simply run the queue
2971 * unscheduled, resulting in a noop
2974 linkwatch_run_queue();
2979 rebroadcast_time = jiffies;
2984 if (time_after(jiffies, warning_time + 10 * HZ)) {
2985 printk(KERN_EMERG "unregister_netdevice: "
2986 "waiting for %s to become free. Usage "
2988 dev->name, atomic_read(&dev->refcnt));
2989 warning_time = jiffies;
2998 * register_netdevice(x1);
2999 * register_netdevice(x2);
3001 * unregister_netdevice(y1);
3002 * unregister_netdevice(y2);
3008 * We are invoked by rtnl_unlock() after it drops the semaphore.
3009 * This allows us to deal with problems:
3010 * 1) We can create/delete sysfs objects which invoke hotplug
3011 * without deadlocking with linkwatch via keventd.
3012 * 2) Since we run with the RTNL semaphore not held, we can sleep
3013 * safely in order to wait for the netdev refcnt to drop to zero.
3015 static DEFINE_MUTEX(net_todo_run_mutex);
3016 void netdev_run_todo(void)
3018 struct list_head list = LIST_HEAD_INIT(list);
3022 /* Need to guard against multiple cpu's getting out of order. */
3023 mutex_lock(&net_todo_run_mutex);
3025 /* Not safe to do outside the semaphore. We must not return
3026 * until all unregister events invoked by the local processor
3027 * have been completed (either by this todo run, or one on
3030 if (list_empty(&net_todo_list))
3033 /* Snapshot list, allow later requests */
3034 spin_lock(&net_todo_list_lock);
3035 list_splice_init(&net_todo_list, &list);
3036 spin_unlock(&net_todo_list_lock);
3038 while (!list_empty(&list)) {
3039 struct net_device *dev
3040 = list_entry(list.next, struct net_device, todo_list);
3041 list_del(&dev->todo_list);
3043 switch(dev->reg_state) {
3044 case NETREG_REGISTERING:
3045 dev->reg_state = NETREG_REGISTERED;
3046 err = netdev_register_sysfs(dev);
3048 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3052 case NETREG_UNREGISTERING:
3053 netdev_unregister_sysfs(dev);
3054 dev->reg_state = NETREG_UNREGISTERED;
3056 netdev_wait_allrefs(dev);
3059 BUG_ON(atomic_read(&dev->refcnt));
3060 BUG_TRAP(!dev->ip_ptr);
3061 BUG_TRAP(!dev->ip6_ptr);
3062 BUG_TRAP(!dev->dn_ptr);
3065 /* It must be the very last action,
3066 * after this 'dev' may point to freed up memory.
3068 if (dev->destructor)
3069 dev->destructor(dev);
3073 printk(KERN_ERR "network todo '%s' but state %d\n",
3074 dev->name, dev->reg_state);
3080 mutex_unlock(&net_todo_run_mutex);
3084 * alloc_netdev - allocate network device
3085 * @sizeof_priv: size of private data to allocate space for
3086 * @name: device name format string
3087 * @setup: callback to initialize device
3089 * Allocates a struct net_device with private data area for driver use
3090 * and performs basic initialization.
3092 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3093 void (*setup)(struct net_device *))
3096 struct net_device *dev;
3099 /* ensure 32-byte alignment of both the device and private area */
3100 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3101 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3103 p = kzalloc(alloc_size, GFP_KERNEL);
3105 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3109 dev = (struct net_device *)
3110 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3111 dev->padded = (char *)dev - (char *)p;
3114 dev->priv = netdev_priv(dev);
3117 strcpy(dev->name, name);
3120 EXPORT_SYMBOL(alloc_netdev);
3123 * free_netdev - free network device
3126 * This function does the last stage of destroying an allocated device
3127 * interface. The reference to the device object is released.
3128 * If this is the last reference then it will be freed.
3130 void free_netdev(struct net_device *dev)
3133 /* Compatiablity with error handling in drivers */
3134 if (dev->reg_state == NETREG_UNINITIALIZED) {
3135 kfree((char *)dev - dev->padded);
3139 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3140 dev->reg_state = NETREG_RELEASED;
3142 /* will free via class release */
3143 class_device_put(&dev->class_dev);
3145 kfree((char *)dev - dev->padded);
3149 /* Synchronize with packet receive processing. */
3150 void synchronize_net(void)
3157 * unregister_netdevice - remove device from the kernel
3160 * This function shuts down a device interface and removes it
3161 * from the kernel tables. On success 0 is returned, on a failure
3162 * a negative errno code is returned.
3164 * Callers must hold the rtnl semaphore. You may want
3165 * unregister_netdev() instead of this.
3168 int unregister_netdevice(struct net_device *dev)
3170 struct net_device *d, **dp;
3172 BUG_ON(dev_boot_phase);
3175 /* Some devices call without registering for initialization unwind. */
3176 if (dev->reg_state == NETREG_UNINITIALIZED) {
3177 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3178 "was registered\n", dev->name, dev);
3182 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3184 /* If device is running, close it first. */
3185 if (dev->flags & IFF_UP)
3188 /* And unlink it from device chain. */
3189 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3191 write_lock_bh(&dev_base_lock);
3192 hlist_del(&dev->name_hlist);
3193 hlist_del(&dev->index_hlist);
3194 if (dev_tail == &dev->next)
3197 write_unlock_bh(&dev_base_lock);
3202 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3207 dev->reg_state = NETREG_UNREGISTERING;
3211 /* Shutdown queueing discipline. */
3215 /* Notify protocols, that we are about to destroy
3216 this device. They should clean all the things.
3218 blocking_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3221 * Flush the multicast chain
3223 dev_mc_discard(dev);
3228 /* Notifier chain MUST detach us from master device. */
3229 BUG_TRAP(!dev->master);
3231 free_divert_blk(dev);
3233 /* Finish processing unregister after unlock */
3243 * unregister_netdev - remove device from the kernel
3246 * This function shuts down a device interface and removes it
3247 * from the kernel tables. On success 0 is returned, on a failure
3248 * a negative errno code is returned.
3250 * This is just a wrapper for unregister_netdevice that takes
3251 * the rtnl semaphore. In general you want to use this and not
3252 * unregister_netdevice.
3254 void unregister_netdev(struct net_device *dev)
3257 unregister_netdevice(dev);
3261 EXPORT_SYMBOL(unregister_netdev);
3263 #ifdef CONFIG_HOTPLUG_CPU
3264 static int dev_cpu_callback(struct notifier_block *nfb,
3265 unsigned long action,
3268 struct sk_buff **list_skb;
3269 struct net_device **list_net;
3270 struct sk_buff *skb;
3271 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3272 struct softnet_data *sd, *oldsd;
3274 if (action != CPU_DEAD)
3277 local_irq_disable();
3278 cpu = smp_processor_id();
3279 sd = &per_cpu(softnet_data, cpu);
3280 oldsd = &per_cpu(softnet_data, oldcpu);
3282 /* Find end of our completion_queue. */
3283 list_skb = &sd->completion_queue;
3285 list_skb = &(*list_skb)->next;
3286 /* Append completion queue from offline CPU. */
3287 *list_skb = oldsd->completion_queue;
3288 oldsd->completion_queue = NULL;
3290 /* Find end of our output_queue. */
3291 list_net = &sd->output_queue;
3293 list_net = &(*list_net)->next_sched;
3294 /* Append output queue from offline CPU. */
3295 *list_net = oldsd->output_queue;
3296 oldsd->output_queue = NULL;
3298 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3301 /* Process offline CPU's input_pkt_queue */
3302 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3307 #endif /* CONFIG_HOTPLUG_CPU */
3311 * Initialize the DEV module. At boot time this walks the device list and
3312 * unhooks any devices that fail to initialise (normally hardware not
3313 * present) and leaves us with a valid list of present and active devices.
3318 * This is called single threaded during boot, so no need
3319 * to take the rtnl semaphore.
3321 static int __init net_dev_init(void)
3323 int i, rc = -ENOMEM;
3325 BUG_ON(!dev_boot_phase);
3329 if (dev_proc_init())
3332 if (netdev_sysfs_init())
3335 INIT_LIST_HEAD(&ptype_all);
3336 for (i = 0; i < 16; i++)
3337 INIT_LIST_HEAD(&ptype_base[i]);
3339 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3340 INIT_HLIST_HEAD(&dev_name_head[i]);
3342 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3343 INIT_HLIST_HEAD(&dev_index_head[i]);
3346 * Initialise the packet receive queues.
3349 for_each_possible_cpu(i) {
3350 struct softnet_data *queue;
3352 queue = &per_cpu(softnet_data, i);
3353 skb_queue_head_init(&queue->input_pkt_queue);
3354 queue->completion_queue = NULL;
3355 INIT_LIST_HEAD(&queue->poll_list);
3356 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3357 queue->backlog_dev.weight = weight_p;
3358 queue->backlog_dev.poll = process_backlog;
3359 atomic_set(&queue->backlog_dev.refcnt, 1);
3364 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3365 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3367 hotcpu_notifier(dev_cpu_callback, 0);
3375 subsys_initcall(net_dev_init);
3377 EXPORT_SYMBOL(__dev_get_by_index);
3378 EXPORT_SYMBOL(__dev_get_by_name);
3379 EXPORT_SYMBOL(__dev_remove_pack);
3380 EXPORT_SYMBOL(__skb_linearize);
3381 EXPORT_SYMBOL(dev_valid_name);
3382 EXPORT_SYMBOL(dev_add_pack);
3383 EXPORT_SYMBOL(dev_alloc_name);
3384 EXPORT_SYMBOL(dev_close);
3385 EXPORT_SYMBOL(dev_get_by_flags);
3386 EXPORT_SYMBOL(dev_get_by_index);
3387 EXPORT_SYMBOL(dev_get_by_name);
3388 EXPORT_SYMBOL(dev_open);
3389 EXPORT_SYMBOL(dev_queue_xmit);
3390 EXPORT_SYMBOL(dev_remove_pack);
3391 EXPORT_SYMBOL(dev_set_allmulti);
3392 EXPORT_SYMBOL(dev_set_promiscuity);
3393 EXPORT_SYMBOL(dev_change_flags);
3394 EXPORT_SYMBOL(dev_set_mtu);
3395 EXPORT_SYMBOL(dev_set_mac_address);
3396 EXPORT_SYMBOL(free_netdev);
3397 EXPORT_SYMBOL(netdev_boot_setup_check);
3398 EXPORT_SYMBOL(netdev_set_master);
3399 EXPORT_SYMBOL(netdev_state_change);
3400 EXPORT_SYMBOL(netif_receive_skb);
3401 EXPORT_SYMBOL(netif_rx);
3402 EXPORT_SYMBOL(register_gifconf);
3403 EXPORT_SYMBOL(register_netdevice);
3404 EXPORT_SYMBOL(register_netdevice_notifier);
3405 EXPORT_SYMBOL(skb_checksum_help);
3406 EXPORT_SYMBOL(synchronize_net);
3407 EXPORT_SYMBOL(unregister_netdevice);
3408 EXPORT_SYMBOL(unregister_netdevice_notifier);
3409 EXPORT_SYMBOL(net_enable_timestamp);
3410 EXPORT_SYMBOL(net_disable_timestamp);
3411 EXPORT_SYMBOL(dev_get_flags);
3413 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3414 EXPORT_SYMBOL(br_handle_frame_hook);
3415 EXPORT_SYMBOL(br_fdb_get_hook);
3416 EXPORT_SYMBOL(br_fdb_put_hook);
3420 EXPORT_SYMBOL(dev_load);
3423 EXPORT_PER_CPU_SYMBOL(softnet_data);