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>
118 #include <linux/vs_network.h>
121 * The list of packet types we will receive (as opposed to discard)
122 * and the routines to invoke.
124 * Why 16. Because with 16 the only overlap we get on a hash of the
125 * low nibble of the protocol value is RARP/SNAP/X.25.
127 * NOTE: That is no longer true with the addition of VLAN tags. Not
128 * sure which should go first, but I bet it won't make much
129 * difference if we are running VLANs. The good news is that
130 * this protocol won't be in the list unless compiled in, so
131 * the average user (w/out VLANs) will not be adversely affected.
148 static DEFINE_SPINLOCK(ptype_lock);
149 static struct list_head ptype_base[16]; /* 16 way hashed list */
150 static struct list_head ptype_all; /* Taps */
153 * The @dev_base list is protected by @dev_base_lock and the rtnl
156 * Pure readers hold dev_base_lock for reading.
158 * Writers must hold the rtnl semaphore while they loop through the
159 * dev_base list, and hold dev_base_lock for writing when they do the
160 * actual updates. This allows pure readers to access the list even
161 * while a writer is preparing to update it.
163 * To put it another way, dev_base_lock is held for writing only to
164 * protect against pure readers; the rtnl semaphore provides the
165 * protection against other writers.
167 * See, for example usages, register_netdevice() and
168 * unregister_netdevice(), which must be called with the rtnl
171 struct net_device *dev_base;
172 static struct net_device **dev_tail = &dev_base;
173 DEFINE_RWLOCK(dev_base_lock);
175 EXPORT_SYMBOL(dev_base);
176 EXPORT_SYMBOL(dev_base_lock);
178 #define NETDEV_HASHBITS 8
179 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
180 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
182 static inline struct hlist_head *dev_name_hash(const char *name)
184 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
185 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
188 static inline struct hlist_head *dev_index_hash(int ifindex)
190 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
197 static RAW_NOTIFIER_HEAD(netdev_chain);
200 * Device drivers call our routines to queue packets here. We empty the
201 * queue in the local softnet handler.
203 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
206 extern int netdev_sysfs_init(void);
207 extern int netdev_register_sysfs(struct net_device *);
208 extern void netdev_unregister_sysfs(struct net_device *);
210 #define netdev_sysfs_init() (0)
211 #define netdev_register_sysfs(dev) (0)
212 #define netdev_unregister_sysfs(dev) do { } while(0)
216 /*******************************************************************************
218 Protocol management and registration routines
220 *******************************************************************************/
229 * Add a protocol ID to the list. Now that the input handler is
230 * smarter we can dispense with all the messy stuff that used to be
233 * BEWARE!!! Protocol handlers, mangling input packets,
234 * MUST BE last in hash buckets and checking protocol handlers
235 * MUST start from promiscuous ptype_all chain in net_bh.
236 * It is true now, do not change it.
237 * Explanation follows: if protocol handler, mangling packet, will
238 * be the first on list, it is not able to sense, that packet
239 * is cloned and should be copied-on-write, so that it will
240 * change it and subsequent readers will get broken packet.
245 * dev_add_pack - add packet handler
246 * @pt: packet type declaration
248 * Add a protocol handler to the networking stack. The passed &packet_type
249 * is linked into kernel lists and may not be freed until it has been
250 * removed from the kernel lists.
252 * This call does not sleep therefore it can not
253 * guarantee all CPU's that are in middle of receiving packets
254 * will see the new packet type (until the next received packet).
257 void dev_add_pack(struct packet_type *pt)
261 spin_lock_bh(&ptype_lock);
262 if (pt->type == htons(ETH_P_ALL)) {
264 list_add_rcu(&pt->list, &ptype_all);
266 hash = ntohs(pt->type) & 15;
267 list_add_rcu(&pt->list, &ptype_base[hash]);
269 spin_unlock_bh(&ptype_lock);
273 * __dev_remove_pack - remove packet handler
274 * @pt: packet type declaration
276 * Remove a protocol handler that was previously added to the kernel
277 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
278 * from the kernel lists and can be freed or reused once this function
281 * The packet type might still be in use by receivers
282 * and must not be freed until after all the CPU's have gone
283 * through a quiescent state.
285 void __dev_remove_pack(struct packet_type *pt)
287 struct list_head *head;
288 struct packet_type *pt1;
290 spin_lock_bh(&ptype_lock);
292 if (pt->type == htons(ETH_P_ALL)) {
296 head = &ptype_base[ntohs(pt->type) & 15];
298 list_for_each_entry(pt1, head, list) {
300 list_del_rcu(&pt->list);
305 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
307 spin_unlock_bh(&ptype_lock);
310 * dev_remove_pack - remove packet handler
311 * @pt: packet type declaration
313 * Remove a protocol handler that was previously added to the kernel
314 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
315 * from the kernel lists and can be freed or reused once this function
318 * This call sleeps to guarantee that no CPU is looking at the packet
321 void dev_remove_pack(struct packet_type *pt)
323 __dev_remove_pack(pt);
328 /******************************************************************************
330 Device Boot-time Settings Routines
332 *******************************************************************************/
334 /* Boot time configuration table */
335 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
338 * netdev_boot_setup_add - add new setup entry
339 * @name: name of the device
340 * @map: configured settings for the device
342 * Adds new setup entry to the dev_boot_setup list. The function
343 * returns 0 on error and 1 on success. This is a generic routine to
346 static int netdev_boot_setup_add(char *name, struct ifmap *map)
348 struct netdev_boot_setup *s;
352 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
353 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
354 memset(s[i].name, 0, sizeof(s[i].name));
355 strcpy(s[i].name, name);
356 memcpy(&s[i].map, map, sizeof(s[i].map));
361 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
365 * netdev_boot_setup_check - check boot time settings
366 * @dev: the netdevice
368 * Check boot time settings for the device.
369 * The found settings are set for the device to be used
370 * later in the device probing.
371 * Returns 0 if no settings found, 1 if they are.
373 int netdev_boot_setup_check(struct net_device *dev)
375 struct netdev_boot_setup *s = dev_boot_setup;
378 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
379 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
380 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
381 dev->irq = s[i].map.irq;
382 dev->base_addr = s[i].map.base_addr;
383 dev->mem_start = s[i].map.mem_start;
384 dev->mem_end = s[i].map.mem_end;
393 * netdev_boot_base - get address from boot time settings
394 * @prefix: prefix for network device
395 * @unit: id for network device
397 * Check boot time settings for the base address of device.
398 * The found settings are set for the device to be used
399 * later in the device probing.
400 * Returns 0 if no settings found.
402 unsigned long netdev_boot_base(const char *prefix, int unit)
404 const struct netdev_boot_setup *s = dev_boot_setup;
408 sprintf(name, "%s%d", prefix, unit);
411 * If device already registered then return base of 1
412 * to indicate not to probe for this interface
414 if (__dev_get_by_name(name))
417 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
418 if (!strcmp(name, s[i].name))
419 return s[i].map.base_addr;
424 * Saves at boot time configured settings for any netdevice.
426 int __init netdev_boot_setup(char *str)
431 str = get_options(str, ARRAY_SIZE(ints), ints);
436 memset(&map, 0, sizeof(map));
440 map.base_addr = ints[2];
442 map.mem_start = ints[3];
444 map.mem_end = ints[4];
446 /* Add new entry to the list */
447 return netdev_boot_setup_add(str, &map);
450 __setup("netdev=", netdev_boot_setup);
452 /*******************************************************************************
454 Device Interface Subroutines
456 *******************************************************************************/
459 * __dev_get_by_name - find a device by its name
460 * @name: name to find
462 * Find an interface by name. Must be called under RTNL semaphore
463 * or @dev_base_lock. If the name is found a pointer to the device
464 * is returned. If the name is not found then %NULL is returned. The
465 * reference counters are not incremented so the caller must be
466 * careful with locks.
469 struct net_device *__dev_get_by_name(const char *name)
471 struct hlist_node *p;
473 hlist_for_each(p, dev_name_hash(name)) {
474 struct net_device *dev
475 = hlist_entry(p, struct net_device, name_hlist);
476 if (!strncmp(dev->name, name, IFNAMSIZ))
483 * dev_get_by_name - find a device by its name
484 * @name: name to find
486 * Find an interface by name. This can be called from any
487 * context and does its own locking. The returned handle has
488 * the usage count incremented and the caller must use dev_put() to
489 * release it when it is no longer needed. %NULL is returned if no
490 * matching device is found.
493 struct net_device *dev_get_by_name(const char *name)
495 struct net_device *dev;
497 read_lock(&dev_base_lock);
498 dev = __dev_get_by_name(name);
501 read_unlock(&dev_base_lock);
506 * __dev_get_by_index - find a device by its ifindex
507 * @ifindex: index of device
509 * Search for an interface by index. Returns %NULL if the device
510 * is not found or a pointer to the device. The device has not
511 * had its reference counter increased so the caller must be careful
512 * about locking. The caller must hold either the RTNL semaphore
516 struct net_device *__dev_get_by_index(int ifindex)
518 struct hlist_node *p;
520 hlist_for_each(p, dev_index_hash(ifindex)) {
521 struct net_device *dev
522 = hlist_entry(p, struct net_device, index_hlist);
523 if (dev->ifindex == ifindex)
531 * dev_get_by_index - find a device by its ifindex
532 * @ifindex: index of device
534 * Search for an interface by index. Returns NULL if the device
535 * is not found or a pointer to the device. The device returned has
536 * had a reference added and the pointer is safe until the user calls
537 * dev_put to indicate they have finished with it.
540 struct net_device *dev_get_by_index(int ifindex)
542 struct net_device *dev;
544 read_lock(&dev_base_lock);
545 dev = __dev_get_by_index(ifindex);
548 read_unlock(&dev_base_lock);
553 * dev_getbyhwaddr - find a device by its hardware address
554 * @type: media type of device
555 * @ha: hardware address
557 * Search for an interface by MAC address. Returns NULL if the device
558 * is not found or a pointer to the device. The caller must hold the
559 * rtnl semaphore. The returned device has not had its ref count increased
560 * and the caller must therefore be careful about locking
563 * If the API was consistent this would be __dev_get_by_hwaddr
566 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
568 struct net_device *dev;
572 for (dev = dev_base; dev; dev = dev->next)
573 if (dev->type == type &&
574 !memcmp(dev->dev_addr, ha, dev->addr_len))
579 EXPORT_SYMBOL(dev_getbyhwaddr);
581 struct net_device *dev_getfirstbyhwtype(unsigned short type)
583 struct net_device *dev;
586 for (dev = dev_base; dev; dev = dev->next) {
587 if (dev->type == type) {
596 EXPORT_SYMBOL(dev_getfirstbyhwtype);
599 * dev_get_by_flags - find any device with given flags
600 * @if_flags: IFF_* values
601 * @mask: bitmask of bits in if_flags to check
603 * Search for any interface with the given flags. Returns NULL if a device
604 * is not found or a pointer to the device. The device returned has
605 * had a reference added and the pointer is safe until the user calls
606 * dev_put to indicate they have finished with it.
609 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
611 struct net_device *dev;
613 read_lock(&dev_base_lock);
614 for (dev = dev_base; dev != NULL; dev = dev->next) {
615 if (((dev->flags ^ if_flags) & mask) == 0) {
620 read_unlock(&dev_base_lock);
625 * dev_valid_name - check if name is okay for network device
628 * Network device names need to be valid file names to
629 * to allow sysfs to work
631 int dev_valid_name(const char *name)
633 return !(*name == '\0'
634 || !strcmp(name, ".")
635 || !strcmp(name, "..")
636 || strchr(name, '/'));
640 * dev_alloc_name - allocate a name for a device
642 * @name: name format string
644 * Passed a format string - eg "lt%d" it will try and find a suitable
645 * id. It scans list of devices to build up a free map, then chooses
646 * the first empty slot. The caller must hold the dev_base or rtnl lock
647 * while allocating the name and adding the device in order to avoid
649 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
650 * Returns the number of the unit assigned or a negative errno code.
653 int dev_alloc_name(struct net_device *dev, const char *name)
658 const int max_netdevices = 8*PAGE_SIZE;
660 struct net_device *d;
662 p = strnchr(name, IFNAMSIZ-1, '%');
665 * Verify the string as this thing may have come from
666 * the user. There must be either one "%d" and no other "%"
669 if (p[1] != 'd' || strchr(p + 2, '%'))
672 /* Use one page as a bit array of possible slots */
673 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
677 for (d = dev_base; d; d = d->next) {
678 if (!sscanf(d->name, name, &i))
680 if (i < 0 || i >= max_netdevices)
683 /* avoid cases where sscanf is not exact inverse of printf */
684 snprintf(buf, sizeof(buf), name, i);
685 if (!strncmp(buf, d->name, IFNAMSIZ))
689 i = find_first_zero_bit(inuse, max_netdevices);
690 free_page((unsigned long) inuse);
693 snprintf(buf, sizeof(buf), name, i);
694 if (!__dev_get_by_name(buf)) {
695 strlcpy(dev->name, buf, IFNAMSIZ);
699 /* It is possible to run out of possible slots
700 * when the name is long and there isn't enough space left
701 * for the digits, or if all bits are used.
708 * dev_change_name - change name of a device
710 * @newname: name (or format string) must be at least IFNAMSIZ
712 * Change name of a device, can pass format strings "eth%d".
715 int dev_change_name(struct net_device *dev, char *newname)
721 if (dev->flags & IFF_UP)
724 if (!dev_valid_name(newname))
727 if (strchr(newname, '%')) {
728 err = dev_alloc_name(dev, newname);
731 strcpy(newname, dev->name);
733 else if (__dev_get_by_name(newname))
736 strlcpy(dev->name, newname, IFNAMSIZ);
738 err = class_device_rename(&dev->class_dev, dev->name);
740 hlist_del(&dev->name_hlist);
741 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
742 raw_notifier_call_chain(&netdev_chain,
743 NETDEV_CHANGENAME, dev);
750 * netdev_features_change - device changes features
751 * @dev: device to cause notification
753 * Called to indicate a device has changed features.
755 void netdev_features_change(struct net_device *dev)
757 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
759 EXPORT_SYMBOL(netdev_features_change);
762 * netdev_state_change - device changes state
763 * @dev: device to cause notification
765 * Called to indicate a device has changed state. This function calls
766 * the notifier chains for netdev_chain and sends a NEWLINK message
767 * to the routing socket.
769 void netdev_state_change(struct net_device *dev)
771 if (dev->flags & IFF_UP) {
772 raw_notifier_call_chain(&netdev_chain,
774 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
779 * dev_load - load a network module
780 * @name: name of interface
782 * If a network interface is not present and the process has suitable
783 * privileges this function loads the module. If module loading is not
784 * available in this kernel then it becomes a nop.
787 void dev_load(const char *name)
789 struct net_device *dev;
791 read_lock(&dev_base_lock);
792 dev = __dev_get_by_name(name);
793 read_unlock(&dev_base_lock);
795 if (!dev && capable(CAP_SYS_MODULE))
796 request_module("%s", name);
799 static int default_rebuild_header(struct sk_buff *skb)
801 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
802 skb->dev ? skb->dev->name : "NULL!!!");
809 * dev_open - prepare an interface for use.
810 * @dev: device to open
812 * Takes a device from down to up state. The device's private open
813 * function is invoked and then the multicast lists are loaded. Finally
814 * the device is moved into the up state and a %NETDEV_UP message is
815 * sent to the netdev notifier chain.
817 * Calling this function on an active interface is a nop. On a failure
818 * a negative errno code is returned.
820 int dev_open(struct net_device *dev)
828 if (dev->flags & IFF_UP)
832 * Is it even present?
834 if (!netif_device_present(dev))
838 * Call device private open method
840 set_bit(__LINK_STATE_START, &dev->state);
842 ret = dev->open(dev);
844 clear_bit(__LINK_STATE_START, &dev->state);
848 * If it went open OK then:
855 dev->flags |= IFF_UP;
858 * Initialize multicasting status
863 * Wakeup transmit queue engine
868 * ... and announce new interface.
870 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
876 * dev_close - shutdown an interface.
877 * @dev: device to shutdown
879 * This function moves an active device into down state. A
880 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
881 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
884 int dev_close(struct net_device *dev)
886 if (!(dev->flags & IFF_UP))
890 * Tell people we are going down, so that they can
891 * prepare to death, when device is still operating.
893 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
897 clear_bit(__LINK_STATE_START, &dev->state);
899 /* Synchronize to scheduled poll. We cannot touch poll list,
900 * it can be even on different cpu. So just clear netif_running(),
901 * and wait when poll really will happen. Actually, the best place
902 * for this is inside dev->stop() after device stopped its irq
903 * engine, but this requires more changes in devices. */
905 smp_mb__after_clear_bit(); /* Commit netif_running(). */
906 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
912 * Call the device specific close. This cannot fail.
913 * Only if device is UP
915 * We allow it to be called even after a DETACH hot-plug
922 * Device is now down.
925 dev->flags &= ~IFF_UP;
928 * Tell people we are down
930 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
937 * Device change register/unregister. These are not inline or static
938 * as we export them to the world.
942 * register_netdevice_notifier - register a network notifier block
945 * Register a notifier to be called when network device events occur.
946 * The notifier passed is linked into the kernel structures and must
947 * not be reused until it has been unregistered. A negative errno code
948 * is returned on a failure.
950 * When registered all registration and up events are replayed
951 * to the new notifier to allow device to have a race free
952 * view of the network device list.
955 int register_netdevice_notifier(struct notifier_block *nb)
957 struct net_device *dev;
961 err = raw_notifier_chain_register(&netdev_chain, nb);
963 for (dev = dev_base; dev; dev = dev->next) {
964 nb->notifier_call(nb, NETDEV_REGISTER, dev);
966 if (dev->flags & IFF_UP)
967 nb->notifier_call(nb, NETDEV_UP, dev);
975 * unregister_netdevice_notifier - unregister a network notifier block
978 * Unregister a notifier previously registered by
979 * register_netdevice_notifier(). The notifier is unlinked into the
980 * kernel structures and may then be reused. A negative errno code
981 * is returned on a failure.
984 int unregister_netdevice_notifier(struct notifier_block *nb)
989 err = raw_notifier_chain_unregister(&netdev_chain, nb);
995 * call_netdevice_notifiers - call all network notifier blocks
996 * @val: value passed unmodified to notifier function
997 * @v: pointer passed unmodified to notifier function
999 * Call all network notifier blocks. Parameters and return value
1000 * are as for raw_notifier_call_chain().
1003 int call_netdevice_notifiers(unsigned long val, void *v)
1005 return raw_notifier_call_chain(&netdev_chain, val, v);
1008 /* When > 0 there are consumers of rx skb time stamps */
1009 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1011 void net_enable_timestamp(void)
1013 atomic_inc(&netstamp_needed);
1016 void net_disable_timestamp(void)
1018 atomic_dec(&netstamp_needed);
1021 void __net_timestamp(struct sk_buff *skb)
1025 do_gettimeofday(&tv);
1026 skb_set_timestamp(skb, &tv);
1028 EXPORT_SYMBOL(__net_timestamp);
1030 static inline void net_timestamp(struct sk_buff *skb)
1032 if (atomic_read(&netstamp_needed))
1033 __net_timestamp(skb);
1035 skb->tstamp.off_sec = 0;
1036 skb->tstamp.off_usec = 0;
1041 * Support routine. Sends outgoing frames to any network
1042 * taps currently in use.
1045 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1047 struct packet_type *ptype;
1052 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1053 /* Never send packets back to the socket
1054 * they originated from - MvS (miquels@drinkel.ow.org)
1056 if ((ptype->dev == dev || !ptype->dev) &&
1057 (ptype->af_packet_priv == NULL ||
1058 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1059 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1063 /* skb->nh should be correctly
1064 set by sender, so that the second statement is
1065 just protection against buggy protocols.
1067 skb2->mac.raw = skb2->data;
1069 if (skb2->nh.raw < skb2->data ||
1070 skb2->nh.raw > skb2->tail) {
1071 if (net_ratelimit())
1072 printk(KERN_CRIT "protocol %04x is "
1074 skb2->protocol, dev->name);
1075 skb2->nh.raw = skb2->data;
1078 skb2->h.raw = skb2->nh.raw;
1079 skb2->pkt_type = PACKET_OUTGOING;
1080 ptype->func(skb2, skb->dev, ptype, skb->dev);
1087 void __netif_schedule(struct net_device *dev)
1089 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1090 unsigned long flags;
1091 struct softnet_data *sd;
1093 local_irq_save(flags);
1094 sd = &__get_cpu_var(softnet_data);
1095 dev->next_sched = sd->output_queue;
1096 sd->output_queue = dev;
1097 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1098 local_irq_restore(flags);
1101 EXPORT_SYMBOL(__netif_schedule);
1103 void __netif_rx_schedule(struct net_device *dev)
1105 unsigned long flags;
1107 local_irq_save(flags);
1109 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1111 dev->quota += dev->weight;
1113 dev->quota = dev->weight;
1114 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1115 local_irq_restore(flags);
1117 EXPORT_SYMBOL(__netif_rx_schedule);
1119 void dev_kfree_skb_any(struct sk_buff *skb)
1121 if (in_irq() || irqs_disabled())
1122 dev_kfree_skb_irq(skb);
1126 EXPORT_SYMBOL(dev_kfree_skb_any);
1130 void netif_device_detach(struct net_device *dev)
1132 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1133 netif_running(dev)) {
1134 netif_stop_queue(dev);
1137 EXPORT_SYMBOL(netif_device_detach);
1139 void netif_device_attach(struct net_device *dev)
1141 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1142 netif_running(dev)) {
1143 netif_wake_queue(dev);
1144 __netdev_watchdog_up(dev);
1147 EXPORT_SYMBOL(netif_device_attach);
1151 * Invalidate hardware checksum when packet is to be mangled, and
1152 * complete checksum manually on outgoing path.
1154 int skb_checksum_help(struct sk_buff *skb, int inward)
1157 int ret = 0, offset = skb->h.raw - skb->data;
1160 skb->ip_summed = CHECKSUM_NONE;
1164 if (skb_cloned(skb)) {
1165 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1170 BUG_ON(offset > (int)skb->len);
1171 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1173 offset = skb->tail - skb->h.raw;
1174 BUG_ON(offset <= 0);
1175 BUG_ON(skb->csum + 2 > offset);
1177 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1178 skb->ip_summed = CHECKSUM_NONE;
1183 /* Take action when hardware reception checksum errors are detected. */
1185 void netdev_rx_csum_fault(struct net_device *dev)
1187 if (net_ratelimit()) {
1188 printk(KERN_ERR "%s: hw csum failure.\n",
1189 dev ? dev->name : "<unknown>");
1193 EXPORT_SYMBOL(netdev_rx_csum_fault);
1196 #ifdef CONFIG_HIGHMEM
1197 /* Actually, we should eliminate this check as soon as we know, that:
1198 * 1. IOMMU is present and allows to map all the memory.
1199 * 2. No high memory really exists on this machine.
1202 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1206 if (dev->features & NETIF_F_HIGHDMA)
1209 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1210 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1216 #define illegal_highdma(dev, skb) (0)
1219 /* Keep head the same: replace data */
1220 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1225 struct skb_shared_info *ninfo;
1226 int headerlen = skb->data - skb->head;
1227 int expand = (skb->tail + skb->data_len) - skb->end;
1229 if (skb_shared(skb))
1235 size = skb->end - skb->head + expand;
1236 size = SKB_DATA_ALIGN(size);
1237 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1241 /* Copy entire thing */
1242 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1246 ninfo = (struct skb_shared_info*)(data + size);
1247 atomic_set(&ninfo->dataref, 1);
1248 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1249 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1250 ninfo->nr_frags = 0;
1251 ninfo->frag_list = NULL;
1253 /* Offset between the two in bytes */
1254 offset = data - skb->head;
1256 /* Free old data. */
1257 skb_release_data(skb);
1260 skb->end = data + size;
1262 /* Set up new pointers */
1263 skb->h.raw += offset;
1264 skb->nh.raw += offset;
1265 skb->mac.raw += offset;
1266 skb->tail += offset;
1267 skb->data += offset;
1269 /* We are no longer a clone, even if we were. */
1272 skb->tail += skb->data_len;
1277 #define HARD_TX_LOCK(dev, cpu) { \
1278 if ((dev->features & NETIF_F_LLTX) == 0) { \
1279 spin_lock(&dev->xmit_lock); \
1280 dev->xmit_lock_owner = cpu; \
1284 #define HARD_TX_UNLOCK(dev) { \
1285 if ((dev->features & NETIF_F_LLTX) == 0) { \
1286 dev->xmit_lock_owner = -1; \
1287 spin_unlock(&dev->xmit_lock); \
1292 * dev_queue_xmit - transmit a buffer
1293 * @skb: buffer to transmit
1295 * Queue a buffer for transmission to a network device. The caller must
1296 * have set the device and priority and built the buffer before calling
1297 * this function. The function can be called from an interrupt.
1299 * A negative errno code is returned on a failure. A success does not
1300 * guarantee the frame will be transmitted as it may be dropped due
1301 * to congestion or traffic shaping.
1303 * -----------------------------------------------------------------------------------
1304 * I notice this method can also return errors from the queue disciplines,
1305 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1308 * Regardless of the return value, the skb is consumed, so it is currently
1309 * difficult to retry a send to this method. (You can bump the ref count
1310 * before sending to hold a reference for retry if you are careful.)
1312 * When calling this method, interrupts MUST be enabled. This is because
1313 * the BH enable code must have IRQs enabled so that it will not deadlock.
1317 int dev_queue_xmit(struct sk_buff *skb)
1319 struct net_device *dev = skb->dev;
1323 if (skb_shinfo(skb)->frag_list &&
1324 !(dev->features & NETIF_F_FRAGLIST) &&
1325 __skb_linearize(skb, GFP_ATOMIC))
1328 /* Fragmented skb is linearized if device does not support SG,
1329 * or if at least one of fragments is in highmem and device
1330 * does not support DMA from it.
1332 if (skb_shinfo(skb)->nr_frags &&
1333 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1334 __skb_linearize(skb, GFP_ATOMIC))
1337 /* If packet is not checksummed and device does not support
1338 * checksumming for this protocol, complete checksumming here.
1340 if (skb->ip_summed == CHECKSUM_HW &&
1341 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1342 (!(dev->features & NETIF_F_IP_CSUM) ||
1343 skb->protocol != htons(ETH_P_IP))))
1344 if (skb_checksum_help(skb, 0))
1347 spin_lock_prefetch(&dev->queue_lock);
1349 /* Disable soft irqs for various locks below. Also
1350 * stops preemption for RCU.
1354 /* Updates of qdisc are serialized by queue_lock.
1355 * The struct Qdisc which is pointed to by qdisc is now a
1356 * rcu structure - it may be accessed without acquiring
1357 * a lock (but the structure may be stale.) The freeing of the
1358 * qdisc will be deferred until it's known that there are no
1359 * more references to it.
1361 * If the qdisc has an enqueue function, we still need to
1362 * hold the queue_lock before calling it, since queue_lock
1363 * also serializes access to the device queue.
1366 q = rcu_dereference(dev->qdisc);
1367 #ifdef CONFIG_NET_CLS_ACT
1368 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1371 /* Grab device queue */
1372 spin_lock(&dev->queue_lock);
1374 rc = q->enqueue(skb, q);
1378 spin_unlock(&dev->queue_lock);
1379 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1383 /* The device has no queue. Common case for software devices:
1384 loopback, all the sorts of tunnels...
1386 Really, it is unlikely that xmit_lock protection is necessary here.
1387 (f.e. loopback and IP tunnels are clean ignoring statistics
1389 However, it is possible, that they rely on protection
1392 Check this and shot the lock. It is not prone from deadlocks.
1393 Either shot noqueue qdisc, it is even simpler 8)
1395 if (dev->flags & IFF_UP) {
1396 int cpu = smp_processor_id(); /* ok because BHs are off */
1398 if (dev->xmit_lock_owner != cpu) {
1400 HARD_TX_LOCK(dev, cpu);
1402 if (!netif_queue_stopped(dev)) {
1404 dev_queue_xmit_nit(skb, dev);
1407 if (!dev->hard_start_xmit(skb, dev)) {
1408 HARD_TX_UNLOCK(dev);
1412 HARD_TX_UNLOCK(dev);
1413 if (net_ratelimit())
1414 printk(KERN_CRIT "Virtual device %s asks to "
1415 "queue packet!\n", dev->name);
1417 /* Recursion is detected! It is possible,
1419 if (net_ratelimit())
1420 printk(KERN_CRIT "Dead loop on virtual device "
1421 "%s, fix it urgently!\n", dev->name);
1437 /*=======================================================================
1439 =======================================================================*/
1441 int netdev_max_backlog = 1000;
1442 int netdev_budget = 300;
1443 int weight_p = 64; /* old backlog weight */
1445 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1449 * netif_rx - post buffer to the network code
1450 * @skb: buffer to post
1452 * This function receives a packet from a device driver and queues it for
1453 * the upper (protocol) levels to process. It always succeeds. The buffer
1454 * may be dropped during processing for congestion control or by the
1458 * NET_RX_SUCCESS (no congestion)
1459 * NET_RX_CN_LOW (low congestion)
1460 * NET_RX_CN_MOD (moderate congestion)
1461 * NET_RX_CN_HIGH (high congestion)
1462 * NET_RX_DROP (packet was dropped)
1466 int netif_rx(struct sk_buff *skb)
1468 struct softnet_data *queue;
1469 unsigned long flags;
1471 /* if netpoll wants it, pretend we never saw it */
1472 if (netpoll_rx(skb))
1475 if (!skb->tstamp.off_sec)
1479 * The code is rearranged so that the path is the most
1480 * short when CPU is congested, but is still operating.
1482 local_irq_save(flags);
1483 queue = &__get_cpu_var(softnet_data);
1485 __get_cpu_var(netdev_rx_stat).total++;
1486 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1487 if (queue->input_pkt_queue.qlen) {
1490 __skb_queue_tail(&queue->input_pkt_queue, skb);
1491 local_irq_restore(flags);
1492 return NET_RX_SUCCESS;
1495 netif_rx_schedule(&queue->backlog_dev);
1499 __get_cpu_var(netdev_rx_stat).dropped++;
1500 local_irq_restore(flags);
1506 int netif_rx_ni(struct sk_buff *skb)
1511 err = netif_rx(skb);
1512 if (local_softirq_pending())
1519 EXPORT_SYMBOL(netif_rx_ni);
1521 static inline struct net_device *skb_bond(struct sk_buff *skb)
1523 struct net_device *dev = skb->dev;
1527 * On bonding slaves other than the currently active
1528 * slave, suppress duplicates except for 802.3ad
1529 * ETH_P_SLOW and alb non-mcast/bcast.
1531 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1532 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1533 if (skb->pkt_type != PACKET_BROADCAST &&
1534 skb->pkt_type != PACKET_MULTICAST)
1538 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1539 skb->protocol == __constant_htons(ETH_P_SLOW))
1546 skb->dev = dev->master;
1552 static void net_tx_action(struct softirq_action *h)
1554 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1556 if (sd->completion_queue) {
1557 struct sk_buff *clist;
1559 local_irq_disable();
1560 clist = sd->completion_queue;
1561 sd->completion_queue = NULL;
1565 struct sk_buff *skb = clist;
1566 clist = clist->next;
1568 BUG_TRAP(!atomic_read(&skb->users));
1573 if (sd->output_queue) {
1574 struct net_device *head;
1576 local_irq_disable();
1577 head = sd->output_queue;
1578 sd->output_queue = NULL;
1582 struct net_device *dev = head;
1583 head = head->next_sched;
1585 smp_mb__before_clear_bit();
1586 clear_bit(__LINK_STATE_SCHED, &dev->state);
1588 if (spin_trylock(&dev->queue_lock)) {
1590 spin_unlock(&dev->queue_lock);
1592 netif_schedule(dev);
1598 static __inline__ int deliver_skb(struct sk_buff *skb,
1599 struct packet_type *pt_prev,
1600 struct net_device *orig_dev)
1602 atomic_inc(&skb->users);
1603 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1606 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1607 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1609 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1610 unsigned char *addr);
1611 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1613 static __inline__ int handle_bridge(struct sk_buff **pskb,
1614 struct packet_type **pt_prev, int *ret,
1615 struct net_device *orig_dev)
1617 struct net_bridge_port *port;
1619 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1620 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1624 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1628 return br_handle_frame_hook(port, pskb);
1631 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1634 #ifdef CONFIG_NET_CLS_ACT
1635 /* TODO: Maybe we should just force sch_ingress to be compiled in
1636 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1637 * a compare and 2 stores extra right now if we dont have it on
1638 * but have CONFIG_NET_CLS_ACT
1639 * NOTE: This doesnt stop any functionality; if you dont have
1640 * the ingress scheduler, you just cant add policies on ingress.
1643 static int ing_filter(struct sk_buff *skb)
1646 struct net_device *dev = skb->dev;
1647 int result = TC_ACT_OK;
1649 if (dev->qdisc_ingress) {
1650 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1651 if (MAX_RED_LOOP < ttl++) {
1652 printk("Redir loop detected Dropping packet (%s->%s)\n",
1653 skb->input_dev->name, skb->dev->name);
1657 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1659 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1661 spin_lock(&dev->ingress_lock);
1662 if ((q = dev->qdisc_ingress) != NULL)
1663 result = q->enqueue(skb, q);
1664 spin_unlock(&dev->ingress_lock);
1672 int netif_receive_skb(struct sk_buff *skb)
1674 struct packet_type *ptype, *pt_prev;
1675 struct net_device *orig_dev;
1676 int ret = NET_RX_DROP;
1677 unsigned short type;
1679 /* if we've gotten here through NAPI, check netpoll */
1680 if (skb->dev->poll && netpoll_rx(skb))
1683 if (!skb->tstamp.off_sec)
1686 if (!skb->input_dev)
1687 skb->input_dev = skb->dev;
1689 orig_dev = skb_bond(skb);
1694 __get_cpu_var(netdev_rx_stat).total++;
1696 skb->h.raw = skb->nh.raw = skb->data;
1697 skb->mac_len = skb->nh.raw - skb->mac.raw;
1703 #ifdef CONFIG_NET_CLS_ACT
1704 if (skb->tc_verd & TC_NCLS) {
1705 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1710 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1711 if (!ptype->dev || ptype->dev == skb->dev) {
1713 ret = deliver_skb(skb, pt_prev, orig_dev);
1718 #ifdef CONFIG_NET_CLS_ACT
1720 ret = deliver_skb(skb, pt_prev, orig_dev);
1721 pt_prev = NULL; /* noone else should process this after*/
1723 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1726 ret = ing_filter(skb);
1728 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1737 handle_diverter(skb);
1739 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1742 type = skb->protocol;
1743 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1744 if (ptype->type == type &&
1745 (!ptype->dev || ptype->dev == skb->dev)) {
1747 ret = deliver_skb(skb, pt_prev, orig_dev);
1753 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1756 /* Jamal, now you will not able to escape explaining
1757 * me how you were going to use this. :-)
1767 static int process_backlog(struct net_device *backlog_dev, int *budget)
1770 int quota = min(backlog_dev->quota, *budget);
1771 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1772 unsigned long start_time = jiffies;
1774 backlog_dev->weight = weight_p;
1776 struct sk_buff *skb;
1777 struct net_device *dev;
1779 local_irq_disable();
1780 skb = __skb_dequeue(&queue->input_pkt_queue);
1787 netif_receive_skb(skb);
1793 if (work >= quota || jiffies - start_time > 1)
1798 backlog_dev->quota -= work;
1803 backlog_dev->quota -= work;
1806 list_del(&backlog_dev->poll_list);
1807 smp_mb__before_clear_bit();
1808 netif_poll_enable(backlog_dev);
1814 static void net_rx_action(struct softirq_action *h)
1816 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1817 unsigned long start_time = jiffies;
1818 int budget = netdev_budget;
1821 local_irq_disable();
1823 while (!list_empty(&queue->poll_list)) {
1824 struct net_device *dev;
1826 if (budget <= 0 || jiffies - start_time > 1)
1831 dev = list_entry(queue->poll_list.next,
1832 struct net_device, poll_list);
1833 have = netpoll_poll_lock(dev);
1835 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1836 netpoll_poll_unlock(have);
1837 local_irq_disable();
1838 list_move_tail(&dev->poll_list, &queue->poll_list);
1840 dev->quota += dev->weight;
1842 dev->quota = dev->weight;
1844 netpoll_poll_unlock(have);
1846 local_irq_disable();
1854 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1855 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1859 static gifconf_func_t * gifconf_list [NPROTO];
1862 * register_gifconf - register a SIOCGIF handler
1863 * @family: Address family
1864 * @gifconf: Function handler
1866 * Register protocol dependent address dumping routines. The handler
1867 * that is passed must not be freed or reused until it has been replaced
1868 * by another handler.
1870 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1872 if (family >= NPROTO)
1874 gifconf_list[family] = gifconf;
1880 * Map an interface index to its name (SIOCGIFNAME)
1884 * We need this ioctl for efficient implementation of the
1885 * if_indextoname() function required by the IPv6 API. Without
1886 * it, we would have to search all the interfaces to find a
1890 static int dev_ifname(struct ifreq __user *arg)
1892 struct net_device *dev;
1896 * Fetch the caller's info block.
1899 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1902 read_lock(&dev_base_lock);
1903 dev = __dev_get_by_index(ifr.ifr_ifindex);
1905 read_unlock(&dev_base_lock);
1909 strcpy(ifr.ifr_name, dev->name);
1910 read_unlock(&dev_base_lock);
1912 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1918 * Perform a SIOCGIFCONF call. This structure will change
1919 * size eventually, and there is nothing I can do about it.
1920 * Thus we will need a 'compatibility mode'.
1923 static int dev_ifconf(char __user *arg)
1926 struct net_device *dev;
1933 * Fetch the caller's info block.
1936 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1943 * Loop over the interfaces, and write an info block for each.
1947 for (dev = dev_base; dev; dev = dev->next) {
1948 if (vx_flags(VXF_HIDE_NETIF, 0) &&
1949 !dev_in_nx_info(dev, current->nx_info))
1951 for (i = 0; i < NPROTO; i++) {
1952 if (gifconf_list[i]) {
1955 done = gifconf_list[i](dev, NULL, 0);
1957 done = gifconf_list[i](dev, pos + total,
1967 * All done. Write the updated control block back to the caller.
1969 ifc.ifc_len = total;
1972 * Both BSD and Solaris return 0 here, so we do too.
1974 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1977 #ifdef CONFIG_PROC_FS
1979 * This is invoked by the /proc filesystem handler to display a device
1982 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1984 struct net_device *dev;
1987 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1989 return i == pos ? dev : NULL;
1992 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1994 read_lock(&dev_base_lock);
1995 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1998 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2001 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2004 void dev_seq_stop(struct seq_file *seq, void *v)
2006 read_unlock(&dev_base_lock);
2009 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2011 struct nx_info *nxi = current->nx_info;
2013 if (vx_flags(VXF_HIDE_NETIF, 0) && !dev_in_nx_info(dev, nxi))
2015 if (dev->get_stats) {
2016 struct net_device_stats *stats = dev->get_stats(dev);
2018 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2019 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2020 dev->name, stats->rx_bytes, stats->rx_packets,
2022 stats->rx_dropped + stats->rx_missed_errors,
2023 stats->rx_fifo_errors,
2024 stats->rx_length_errors + stats->rx_over_errors +
2025 stats->rx_crc_errors + stats->rx_frame_errors,
2026 stats->rx_compressed, stats->multicast,
2027 stats->tx_bytes, stats->tx_packets,
2028 stats->tx_errors, stats->tx_dropped,
2029 stats->tx_fifo_errors, stats->collisions,
2030 stats->tx_carrier_errors +
2031 stats->tx_aborted_errors +
2032 stats->tx_window_errors +
2033 stats->tx_heartbeat_errors,
2034 stats->tx_compressed);
2036 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2040 * Called from the PROCfs module. This now uses the new arbitrary sized
2041 * /proc/net interface to create /proc/net/dev
2043 static int dev_seq_show(struct seq_file *seq, void *v)
2045 if (v == SEQ_START_TOKEN)
2046 seq_puts(seq, "Inter-| Receive "
2048 " face |bytes packets errs drop fifo frame "
2049 "compressed multicast|bytes packets errs "
2050 "drop fifo colls carrier compressed\n");
2052 dev_seq_printf_stats(seq, v);
2056 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2058 struct netif_rx_stats *rc = NULL;
2060 while (*pos < NR_CPUS)
2061 if (cpu_online(*pos)) {
2062 rc = &per_cpu(netdev_rx_stat, *pos);
2069 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2071 return softnet_get_online(pos);
2074 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2077 return softnet_get_online(pos);
2080 static void softnet_seq_stop(struct seq_file *seq, void *v)
2084 static int softnet_seq_show(struct seq_file *seq, void *v)
2086 struct netif_rx_stats *s = v;
2088 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2089 s->total, s->dropped, s->time_squeeze, 0,
2090 0, 0, 0, 0, /* was fastroute */
2095 static struct seq_operations dev_seq_ops = {
2096 .start = dev_seq_start,
2097 .next = dev_seq_next,
2098 .stop = dev_seq_stop,
2099 .show = dev_seq_show,
2102 static int dev_seq_open(struct inode *inode, struct file *file)
2104 return seq_open(file, &dev_seq_ops);
2107 static struct file_operations dev_seq_fops = {
2108 .owner = THIS_MODULE,
2109 .open = dev_seq_open,
2111 .llseek = seq_lseek,
2112 .release = seq_release,
2115 static struct seq_operations softnet_seq_ops = {
2116 .start = softnet_seq_start,
2117 .next = softnet_seq_next,
2118 .stop = softnet_seq_stop,
2119 .show = softnet_seq_show,
2122 static int softnet_seq_open(struct inode *inode, struct file *file)
2124 return seq_open(file, &softnet_seq_ops);
2127 static struct file_operations softnet_seq_fops = {
2128 .owner = THIS_MODULE,
2129 .open = softnet_seq_open,
2131 .llseek = seq_lseek,
2132 .release = seq_release,
2135 #ifdef CONFIG_WIRELESS_EXT
2136 extern int wireless_proc_init(void);
2138 #define wireless_proc_init() 0
2141 static int __init dev_proc_init(void)
2145 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2147 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2149 if (wireless_proc_init())
2155 proc_net_remove("softnet_stat");
2157 proc_net_remove("dev");
2161 #define dev_proc_init() 0
2162 #endif /* CONFIG_PROC_FS */
2166 * netdev_set_master - set up master/slave pair
2167 * @slave: slave device
2168 * @master: new master device
2170 * Changes the master device of the slave. Pass %NULL to break the
2171 * bonding. The caller must hold the RTNL semaphore. On a failure
2172 * a negative errno code is returned. On success the reference counts
2173 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2174 * function returns zero.
2176 int netdev_set_master(struct net_device *slave, struct net_device *master)
2178 struct net_device *old = slave->master;
2188 slave->master = master;
2196 slave->flags |= IFF_SLAVE;
2198 slave->flags &= ~IFF_SLAVE;
2200 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2205 * dev_set_promiscuity - update promiscuity count on a device
2209 * Add or remove promiscuity from a device. While the count in the device
2210 * remains above zero the interface remains promiscuous. Once it hits zero
2211 * the device reverts back to normal filtering operation. A negative inc
2212 * value is used to drop promiscuity on the device.
2214 void dev_set_promiscuity(struct net_device *dev, int inc)
2216 unsigned short old_flags = dev->flags;
2218 if ((dev->promiscuity += inc) == 0)
2219 dev->flags &= ~IFF_PROMISC;
2221 dev->flags |= IFF_PROMISC;
2222 if (dev->flags != old_flags) {
2224 printk(KERN_INFO "device %s %s promiscuous mode\n",
2225 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2227 audit_log(current->audit_context, GFP_ATOMIC,
2228 AUDIT_ANOM_PROMISCUOUS,
2229 "dev=%s prom=%d old_prom=%d auid=%u",
2230 dev->name, (dev->flags & IFF_PROMISC),
2231 (old_flags & IFF_PROMISC),
2232 audit_get_loginuid(current->audit_context));
2237 * dev_set_allmulti - update allmulti count on a device
2241 * Add or remove reception of all multicast frames to a device. While the
2242 * count in the device remains above zero the interface remains listening
2243 * to all interfaces. Once it hits zero the device reverts back to normal
2244 * filtering operation. A negative @inc value is used to drop the counter
2245 * when releasing a resource needing all multicasts.
2248 void dev_set_allmulti(struct net_device *dev, int inc)
2250 unsigned short old_flags = dev->flags;
2252 dev->flags |= IFF_ALLMULTI;
2253 if ((dev->allmulti += inc) == 0)
2254 dev->flags &= ~IFF_ALLMULTI;
2255 if (dev->flags ^ old_flags)
2259 unsigned dev_get_flags(const struct net_device *dev)
2263 flags = (dev->flags & ~(IFF_PROMISC |
2268 (dev->gflags & (IFF_PROMISC |
2271 if (netif_running(dev)) {
2272 if (netif_oper_up(dev))
2273 flags |= IFF_RUNNING;
2274 if (netif_carrier_ok(dev))
2275 flags |= IFF_LOWER_UP;
2276 if (netif_dormant(dev))
2277 flags |= IFF_DORMANT;
2283 int dev_change_flags(struct net_device *dev, unsigned flags)
2286 int old_flags = dev->flags;
2289 * Set the flags on our device.
2292 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2293 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2295 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2299 * Load in the correct multicast list now the flags have changed.
2305 * Have we downed the interface. We handle IFF_UP ourselves
2306 * according to user attempts to set it, rather than blindly
2311 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2312 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2318 if (dev->flags & IFF_UP &&
2319 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2321 raw_notifier_call_chain(&netdev_chain,
2322 NETDEV_CHANGE, dev);
2324 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2325 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2326 dev->gflags ^= IFF_PROMISC;
2327 dev_set_promiscuity(dev, inc);
2330 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2331 is important. Some (broken) drivers set IFF_PROMISC, when
2332 IFF_ALLMULTI is requested not asking us and not reporting.
2334 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2335 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2336 dev->gflags ^= IFF_ALLMULTI;
2337 dev_set_allmulti(dev, inc);
2340 if (old_flags ^ dev->flags)
2341 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2346 int dev_set_mtu(struct net_device *dev, int new_mtu)
2350 if (new_mtu == dev->mtu)
2353 /* MTU must be positive. */
2357 if (!netif_device_present(dev))
2361 if (dev->change_mtu)
2362 err = dev->change_mtu(dev, new_mtu);
2365 if (!err && dev->flags & IFF_UP)
2366 raw_notifier_call_chain(&netdev_chain,
2367 NETDEV_CHANGEMTU, dev);
2371 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2375 if (!dev->set_mac_address)
2377 if (sa->sa_family != dev->type)
2379 if (!netif_device_present(dev))
2381 err = dev->set_mac_address(dev, sa);
2383 raw_notifier_call_chain(&netdev_chain,
2384 NETDEV_CHANGEADDR, dev);
2389 * Perform the SIOCxIFxxx calls.
2391 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2394 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2400 case SIOCGIFFLAGS: /* Get interface flags */
2401 ifr->ifr_flags = dev_get_flags(dev);
2404 case SIOCSIFFLAGS: /* Set interface flags */
2405 return dev_change_flags(dev, ifr->ifr_flags);
2407 case SIOCGIFMETRIC: /* Get the metric on the interface
2408 (currently unused) */
2409 ifr->ifr_metric = 0;
2412 case SIOCSIFMETRIC: /* Set the metric on the interface
2413 (currently unused) */
2416 case SIOCGIFMTU: /* Get the MTU of a device */
2417 ifr->ifr_mtu = dev->mtu;
2420 case SIOCSIFMTU: /* Set the MTU of a device */
2421 return dev_set_mtu(dev, ifr->ifr_mtu);
2425 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2427 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2428 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2429 ifr->ifr_hwaddr.sa_family = dev->type;
2433 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2435 case SIOCSIFHWBROADCAST:
2436 if (ifr->ifr_hwaddr.sa_family != dev->type)
2438 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2439 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2440 raw_notifier_call_chain(&netdev_chain,
2441 NETDEV_CHANGEADDR, dev);
2445 ifr->ifr_map.mem_start = dev->mem_start;
2446 ifr->ifr_map.mem_end = dev->mem_end;
2447 ifr->ifr_map.base_addr = dev->base_addr;
2448 ifr->ifr_map.irq = dev->irq;
2449 ifr->ifr_map.dma = dev->dma;
2450 ifr->ifr_map.port = dev->if_port;
2454 if (dev->set_config) {
2455 if (!netif_device_present(dev))
2457 return dev->set_config(dev, &ifr->ifr_map);
2462 if (!dev->set_multicast_list ||
2463 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2465 if (!netif_device_present(dev))
2467 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2471 if (!dev->set_multicast_list ||
2472 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2474 if (!netif_device_present(dev))
2476 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2480 ifr->ifr_ifindex = dev->ifindex;
2484 ifr->ifr_qlen = dev->tx_queue_len;
2488 if (ifr->ifr_qlen < 0)
2490 dev->tx_queue_len = ifr->ifr_qlen;
2494 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2495 return dev_change_name(dev, ifr->ifr_newname);
2498 * Unknown or private ioctl
2502 if ((cmd >= SIOCDEVPRIVATE &&
2503 cmd <= SIOCDEVPRIVATE + 15) ||
2504 cmd == SIOCBONDENSLAVE ||
2505 cmd == SIOCBONDRELEASE ||
2506 cmd == SIOCBONDSETHWADDR ||
2507 cmd == SIOCBONDSLAVEINFOQUERY ||
2508 cmd == SIOCBONDINFOQUERY ||
2509 cmd == SIOCBONDCHANGEACTIVE ||
2510 cmd == SIOCGMIIPHY ||
2511 cmd == SIOCGMIIREG ||
2512 cmd == SIOCSMIIREG ||
2513 cmd == SIOCBRADDIF ||
2514 cmd == SIOCBRDELIF ||
2515 cmd == SIOCWANDEV) {
2517 if (dev->do_ioctl) {
2518 if (netif_device_present(dev))
2519 err = dev->do_ioctl(dev, ifr,
2532 * This function handles all "interface"-type I/O control requests. The actual
2533 * 'doing' part of this is dev_ifsioc above.
2537 * dev_ioctl - network device ioctl
2538 * @cmd: command to issue
2539 * @arg: pointer to a struct ifreq in user space
2541 * Issue ioctl functions to devices. This is normally called by the
2542 * user space syscall interfaces but can sometimes be useful for
2543 * other purposes. The return value is the return from the syscall if
2544 * positive or a negative errno code on error.
2547 int dev_ioctl(unsigned int cmd, void __user *arg)
2553 /* One special case: SIOCGIFCONF takes ifconf argument
2554 and requires shared lock, because it sleeps writing
2558 if (cmd == SIOCGIFCONF) {
2560 ret = dev_ifconf((char __user *) arg);
2564 if (cmd == SIOCGIFNAME)
2565 return dev_ifname((struct ifreq __user *)arg);
2567 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2570 ifr.ifr_name[IFNAMSIZ-1] = 0;
2572 colon = strchr(ifr.ifr_name, ':');
2577 * See which interface the caller is talking about.
2582 * These ioctl calls:
2583 * - can be done by all.
2584 * - atomic and do not require locking.
2595 dev_load(ifr.ifr_name);
2596 read_lock(&dev_base_lock);
2597 ret = dev_ifsioc(&ifr, cmd);
2598 read_unlock(&dev_base_lock);
2602 if (copy_to_user(arg, &ifr,
2603 sizeof(struct ifreq)))
2609 dev_load(ifr.ifr_name);
2611 ret = dev_ethtool(&ifr);
2616 if (copy_to_user(arg, &ifr,
2617 sizeof(struct ifreq)))
2623 * These ioctl calls:
2624 * - require superuser power.
2625 * - require strict serialization.
2631 if (!capable(CAP_NET_ADMIN))
2633 dev_load(ifr.ifr_name);
2635 ret = dev_ifsioc(&ifr, cmd);
2640 if (copy_to_user(arg, &ifr,
2641 sizeof(struct ifreq)))
2647 * These ioctl calls:
2648 * - require superuser power.
2649 * - require strict serialization.
2650 * - do not return a value
2660 case SIOCSIFHWBROADCAST:
2663 case SIOCBONDENSLAVE:
2664 case SIOCBONDRELEASE:
2665 case SIOCBONDSETHWADDR:
2666 case SIOCBONDCHANGEACTIVE:
2669 if (!capable(CAP_NET_ADMIN))
2672 case SIOCBONDSLAVEINFOQUERY:
2673 case SIOCBONDINFOQUERY:
2674 dev_load(ifr.ifr_name);
2676 ret = dev_ifsioc(&ifr, cmd);
2681 /* Get the per device memory space. We can add this but
2682 * currently do not support it */
2684 /* Set the per device memory buffer space.
2685 * Not applicable in our case */
2690 * Unknown or private ioctl.
2693 if (cmd == SIOCWANDEV ||
2694 (cmd >= SIOCDEVPRIVATE &&
2695 cmd <= SIOCDEVPRIVATE + 15)) {
2696 dev_load(ifr.ifr_name);
2698 ret = dev_ifsioc(&ifr, cmd);
2700 if (!ret && copy_to_user(arg, &ifr,
2701 sizeof(struct ifreq)))
2705 #ifdef CONFIG_WIRELESS_EXT
2706 /* Take care of Wireless Extensions */
2707 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2708 /* If command is `set a parameter', or
2709 * `get the encoding parameters', check if
2710 * the user has the right to do it */
2711 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2712 || cmd == SIOCGIWENCODEEXT) {
2713 if (!capable(CAP_NET_ADMIN))
2716 dev_load(ifr.ifr_name);
2718 /* Follow me in net/core/wireless.c */
2719 ret = wireless_process_ioctl(&ifr, cmd);
2721 if (IW_IS_GET(cmd) &&
2722 copy_to_user(arg, &ifr,
2723 sizeof(struct ifreq)))
2727 #endif /* CONFIG_WIRELESS_EXT */
2734 * dev_new_index - allocate an ifindex
2736 * Returns a suitable unique value for a new device interface
2737 * number. The caller must hold the rtnl semaphore or the
2738 * dev_base_lock to be sure it remains unique.
2740 static int dev_new_index(void)
2746 if (!__dev_get_by_index(ifindex))
2751 static int dev_boot_phase = 1;
2753 /* Delayed registration/unregisteration */
2754 static DEFINE_SPINLOCK(net_todo_list_lock);
2755 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2757 static inline void net_set_todo(struct net_device *dev)
2759 spin_lock(&net_todo_list_lock);
2760 list_add_tail(&dev->todo_list, &net_todo_list);
2761 spin_unlock(&net_todo_list_lock);
2765 * register_netdevice - register a network device
2766 * @dev: device to register
2768 * Take a completed network device structure and add it to the kernel
2769 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2770 * chain. 0 is returned on success. A negative errno code is returned
2771 * on a failure to set up the device, or if the name is a duplicate.
2773 * Callers must hold the rtnl semaphore. You may want
2774 * register_netdev() instead of this.
2777 * The locking appears insufficient to guarantee two parallel registers
2778 * will not get the same name.
2781 int register_netdevice(struct net_device *dev)
2783 struct hlist_head *head;
2784 struct hlist_node *p;
2787 BUG_ON(dev_boot_phase);
2792 /* When net_device's are persistent, this will be fatal. */
2793 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2795 spin_lock_init(&dev->queue_lock);
2796 spin_lock_init(&dev->xmit_lock);
2797 dev->xmit_lock_owner = -1;
2798 #ifdef CONFIG_NET_CLS_ACT
2799 spin_lock_init(&dev->ingress_lock);
2802 ret = alloc_divert_blk(dev);
2808 /* Init, if this function is available */
2810 ret = dev->init(dev);
2818 if (!dev_valid_name(dev->name)) {
2823 dev->ifindex = dev_new_index();
2824 if (dev->iflink == -1)
2825 dev->iflink = dev->ifindex;
2827 /* Check for existence of name */
2828 head = dev_name_hash(dev->name);
2829 hlist_for_each(p, head) {
2830 struct net_device *d
2831 = hlist_entry(p, struct net_device, name_hlist);
2832 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2838 /* Fix illegal SG+CSUM combinations. */
2839 if ((dev->features & NETIF_F_SG) &&
2840 !(dev->features & (NETIF_F_IP_CSUM |
2842 NETIF_F_HW_CSUM))) {
2843 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2845 dev->features &= ~NETIF_F_SG;
2848 /* TSO requires that SG is present as well. */
2849 if ((dev->features & NETIF_F_TSO) &&
2850 !(dev->features & NETIF_F_SG)) {
2851 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2853 dev->features &= ~NETIF_F_TSO;
2855 if (dev->features & NETIF_F_UFO) {
2856 if (!(dev->features & NETIF_F_HW_CSUM)) {
2857 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2858 "NETIF_F_HW_CSUM feature.\n",
2860 dev->features &= ~NETIF_F_UFO;
2862 if (!(dev->features & NETIF_F_SG)) {
2863 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2864 "NETIF_F_SG feature.\n",
2866 dev->features &= ~NETIF_F_UFO;
2871 * nil rebuild_header routine,
2872 * that should be never called and used as just bug trap.
2875 if (!dev->rebuild_header)
2876 dev->rebuild_header = default_rebuild_header;
2878 ret = netdev_register_sysfs(dev);
2881 dev->reg_state = NETREG_REGISTERED;
2884 * Default initial state at registry is that the
2885 * device is present.
2888 set_bit(__LINK_STATE_PRESENT, &dev->state);
2891 dev_init_scheduler(dev);
2892 write_lock_bh(&dev_base_lock);
2894 dev_tail = &dev->next;
2895 hlist_add_head(&dev->name_hlist, head);
2896 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2898 write_unlock_bh(&dev_base_lock);
2900 /* Notify protocols, that a new device appeared. */
2901 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2908 free_divert_blk(dev);
2913 * register_netdev - register a network device
2914 * @dev: device to register
2916 * Take a completed network device structure and add it to the kernel
2917 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2918 * chain. 0 is returned on success. A negative errno code is returned
2919 * on a failure to set up the device, or if the name is a duplicate.
2921 * This is a wrapper around register_netdev that takes the rtnl semaphore
2922 * and expands the device name if you passed a format string to
2925 int register_netdev(struct net_device *dev)
2932 * If the name is a format string the caller wants us to do a
2935 if (strchr(dev->name, '%')) {
2936 err = dev_alloc_name(dev, dev->name);
2942 * Back compatibility hook. Kill this one in 2.5
2944 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2945 err = dev_alloc_name(dev, "eth%d");
2950 err = register_netdevice(dev);
2955 EXPORT_SYMBOL(register_netdev);
2958 * netdev_wait_allrefs - wait until all references are gone.
2960 * This is called when unregistering network devices.
2962 * Any protocol or device that holds a reference should register
2963 * for netdevice notification, and cleanup and put back the
2964 * reference if they receive an UNREGISTER event.
2965 * We can get stuck here if buggy protocols don't correctly
2968 static void netdev_wait_allrefs(struct net_device *dev)
2970 unsigned long rebroadcast_time, warning_time;
2972 rebroadcast_time = warning_time = jiffies;
2973 while (atomic_read(&dev->refcnt) != 0) {
2974 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2977 /* Rebroadcast unregister notification */
2978 raw_notifier_call_chain(&netdev_chain,
2979 NETDEV_UNREGISTER, dev);
2981 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2983 /* We must not have linkwatch events
2984 * pending on unregister. If this
2985 * happens, we simply run the queue
2986 * unscheduled, resulting in a noop
2989 linkwatch_run_queue();
2994 rebroadcast_time = jiffies;
2999 if (time_after(jiffies, warning_time + 10 * HZ)) {
3000 printk(KERN_EMERG "unregister_netdevice: "
3001 "waiting for %s to become free. Usage "
3003 dev->name, atomic_read(&dev->refcnt));
3004 warning_time = jiffies;
3013 * register_netdevice(x1);
3014 * register_netdevice(x2);
3016 * unregister_netdevice(y1);
3017 * unregister_netdevice(y2);
3023 * We are invoked by rtnl_unlock() after it drops the semaphore.
3024 * This allows us to deal with problems:
3025 * 1) We can delete sysfs objects which invoke hotplug
3026 * without deadlocking with linkwatch via keventd.
3027 * 2) Since we run with the RTNL semaphore not held, we can sleep
3028 * safely in order to wait for the netdev refcnt to drop to zero.
3030 static DEFINE_MUTEX(net_todo_run_mutex);
3031 void netdev_run_todo(void)
3033 struct list_head list = LIST_HEAD_INIT(list);
3035 /* Need to guard against multiple cpu's getting out of order. */
3036 mutex_lock(&net_todo_run_mutex);
3038 /* Not safe to do outside the semaphore. We must not return
3039 * until all unregister events invoked by the local processor
3040 * have been completed (either by this todo run, or one on
3043 if (list_empty(&net_todo_list))
3046 /* Snapshot list, allow later requests */
3047 spin_lock(&net_todo_list_lock);
3048 list_splice_init(&net_todo_list, &list);
3049 spin_unlock(&net_todo_list_lock);
3051 while (!list_empty(&list)) {
3052 struct net_device *dev
3053 = list_entry(list.next, struct net_device, todo_list);
3054 list_del(&dev->todo_list);
3056 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3057 printk(KERN_ERR "network todo '%s' but state %d\n",
3058 dev->name, dev->reg_state);
3063 netdev_unregister_sysfs(dev);
3064 dev->reg_state = NETREG_UNREGISTERED;
3066 netdev_wait_allrefs(dev);
3069 BUG_ON(atomic_read(&dev->refcnt));
3070 BUG_TRAP(!dev->ip_ptr);
3071 BUG_TRAP(!dev->ip6_ptr);
3072 BUG_TRAP(!dev->dn_ptr);
3074 /* It must be the very last action,
3075 * after this 'dev' may point to freed up memory.
3077 if (dev->destructor)
3078 dev->destructor(dev);
3082 mutex_unlock(&net_todo_run_mutex);
3086 * alloc_netdev - allocate network device
3087 * @sizeof_priv: size of private data to allocate space for
3088 * @name: device name format string
3089 * @setup: callback to initialize device
3091 * Allocates a struct net_device with private data area for driver use
3092 * and performs basic initialization.
3094 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3095 void (*setup)(struct net_device *))
3098 struct net_device *dev;
3101 /* ensure 32-byte alignment of both the device and private area */
3102 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3103 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3105 p = kzalloc(alloc_size, GFP_KERNEL);
3107 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3111 dev = (struct net_device *)
3112 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3113 dev->padded = (char *)dev - (char *)p;
3116 dev->priv = netdev_priv(dev);
3119 strcpy(dev->name, name);
3122 EXPORT_SYMBOL(alloc_netdev);
3125 * free_netdev - free network device
3128 * This function does the last stage of destroying an allocated device
3129 * interface. The reference to the device object is released.
3130 * If this is the last reference then it will be freed.
3132 void free_netdev(struct net_device *dev)
3135 /* Compatibility with error handling in drivers */
3136 if (dev->reg_state == NETREG_UNINITIALIZED) {
3137 kfree((char *)dev - dev->padded);
3141 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3142 dev->reg_state = NETREG_RELEASED;
3144 /* will free via class release */
3145 class_device_put(&dev->class_dev);
3147 kfree((char *)dev - dev->padded);
3151 /* Synchronize with packet receive processing. */
3152 void synchronize_net(void)
3159 * unregister_netdevice - remove device from the kernel
3162 * This function shuts down a device interface and removes it
3163 * from the kernel tables. On success 0 is returned, on a failure
3164 * a negative errno code is returned.
3166 * Callers must hold the rtnl semaphore. You may want
3167 * unregister_netdev() instead of this.
3170 int unregister_netdevice(struct net_device *dev)
3172 struct net_device *d, **dp;
3174 BUG_ON(dev_boot_phase);
3177 /* Some devices call without registering for initialization unwind. */
3178 if (dev->reg_state == NETREG_UNINITIALIZED) {
3179 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3180 "was registered\n", dev->name, dev);
3184 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3186 /* If device is running, close it first. */
3187 if (dev->flags & IFF_UP)
3190 /* And unlink it from device chain. */
3191 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3193 write_lock_bh(&dev_base_lock);
3194 hlist_del(&dev->name_hlist);
3195 hlist_del(&dev->index_hlist);
3196 if (dev_tail == &dev->next)
3199 write_unlock_bh(&dev_base_lock);
3204 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3209 dev->reg_state = NETREG_UNREGISTERING;
3213 /* Shutdown queueing discipline. */
3217 /* Notify protocols, that we are about to destroy
3218 this device. They should clean all the things.
3220 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3223 * Flush the multicast chain
3225 dev_mc_discard(dev);
3230 /* Notifier chain MUST detach us from master device. */
3231 BUG_TRAP(!dev->master);
3233 free_divert_blk(dev);
3235 /* Finish processing unregister after unlock */
3245 * unregister_netdev - remove device from the kernel
3248 * This function shuts down a device interface and removes it
3249 * from the kernel tables. On success 0 is returned, on a failure
3250 * a negative errno code is returned.
3252 * This is just a wrapper for unregister_netdevice that takes
3253 * the rtnl semaphore. In general you want to use this and not
3254 * unregister_netdevice.
3256 void unregister_netdev(struct net_device *dev)
3259 unregister_netdevice(dev);
3263 EXPORT_SYMBOL(unregister_netdev);
3265 #ifdef CONFIG_HOTPLUG_CPU
3266 static int dev_cpu_callback(struct notifier_block *nfb,
3267 unsigned long action,
3270 struct sk_buff **list_skb;
3271 struct net_device **list_net;
3272 struct sk_buff *skb;
3273 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3274 struct softnet_data *sd, *oldsd;
3276 if (action != CPU_DEAD)
3279 local_irq_disable();
3280 cpu = smp_processor_id();
3281 sd = &per_cpu(softnet_data, cpu);
3282 oldsd = &per_cpu(softnet_data, oldcpu);
3284 /* Find end of our completion_queue. */
3285 list_skb = &sd->completion_queue;
3287 list_skb = &(*list_skb)->next;
3288 /* Append completion queue from offline CPU. */
3289 *list_skb = oldsd->completion_queue;
3290 oldsd->completion_queue = NULL;
3292 /* Find end of our output_queue. */
3293 list_net = &sd->output_queue;
3295 list_net = &(*list_net)->next_sched;
3296 /* Append output queue from offline CPU. */
3297 *list_net = oldsd->output_queue;
3298 oldsd->output_queue = NULL;
3300 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3303 /* Process offline CPU's input_pkt_queue */
3304 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3309 #endif /* CONFIG_HOTPLUG_CPU */
3313 * Initialize the DEV module. At boot time this walks the device list and
3314 * unhooks any devices that fail to initialise (normally hardware not
3315 * present) and leaves us with a valid list of present and active devices.
3320 * This is called single threaded during boot, so no need
3321 * to take the rtnl semaphore.
3323 static int __init net_dev_init(void)
3325 int i, rc = -ENOMEM;
3327 BUG_ON(!dev_boot_phase);
3331 if (dev_proc_init())
3334 if (netdev_sysfs_init())
3337 INIT_LIST_HEAD(&ptype_all);
3338 for (i = 0; i < 16; i++)
3339 INIT_LIST_HEAD(&ptype_base[i]);
3341 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3342 INIT_HLIST_HEAD(&dev_name_head[i]);
3344 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3345 INIT_HLIST_HEAD(&dev_index_head[i]);
3348 * Initialise the packet receive queues.
3351 for_each_possible_cpu(i) {
3352 struct softnet_data *queue;
3354 queue = &per_cpu(softnet_data, i);
3355 skb_queue_head_init(&queue->input_pkt_queue);
3356 queue->completion_queue = NULL;
3357 INIT_LIST_HEAD(&queue->poll_list);
3358 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3359 queue->backlog_dev.weight = weight_p;
3360 queue->backlog_dev.poll = process_backlog;
3361 atomic_set(&queue->backlog_dev.refcnt, 1);
3366 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3367 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3369 hotcpu_notifier(dev_cpu_callback, 0);
3377 subsys_initcall(net_dev_init);
3379 EXPORT_SYMBOL(__dev_get_by_index);
3380 EXPORT_SYMBOL(__dev_get_by_name);
3381 EXPORT_SYMBOL(__dev_remove_pack);
3382 EXPORT_SYMBOL(__skb_linearize);
3383 EXPORT_SYMBOL(dev_valid_name);
3384 EXPORT_SYMBOL(dev_add_pack);
3385 EXPORT_SYMBOL(dev_alloc_name);
3386 EXPORT_SYMBOL(dev_close);
3387 EXPORT_SYMBOL(dev_get_by_flags);
3388 EXPORT_SYMBOL(dev_get_by_index);
3389 EXPORT_SYMBOL(dev_get_by_name);
3390 EXPORT_SYMBOL(dev_open);
3391 EXPORT_SYMBOL(dev_queue_xmit);
3392 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
3393 EXPORT_SYMBOL(dev_queue_xmit_nit);
3395 EXPORT_SYMBOL(dev_remove_pack);
3396 EXPORT_SYMBOL(dev_set_allmulti);
3397 EXPORT_SYMBOL(dev_set_promiscuity);
3398 EXPORT_SYMBOL(dev_change_flags);
3399 EXPORT_SYMBOL(dev_set_mtu);
3400 EXPORT_SYMBOL(dev_set_mac_address);
3401 EXPORT_SYMBOL(free_netdev);
3402 EXPORT_SYMBOL(netdev_boot_setup_check);
3403 EXPORT_SYMBOL(netdev_set_master);
3404 EXPORT_SYMBOL(netdev_state_change);
3405 EXPORT_SYMBOL(netif_receive_skb);
3406 EXPORT_SYMBOL(netif_rx);
3407 EXPORT_SYMBOL(register_gifconf);
3408 EXPORT_SYMBOL(register_netdevice);
3409 EXPORT_SYMBOL(register_netdevice_notifier);
3410 EXPORT_SYMBOL(skb_checksum_help);
3411 EXPORT_SYMBOL(synchronize_net);
3412 EXPORT_SYMBOL(unregister_netdevice);
3413 EXPORT_SYMBOL(unregister_netdevice_notifier);
3414 EXPORT_SYMBOL(net_enable_timestamp);
3415 EXPORT_SYMBOL(net_disable_timestamp);
3416 EXPORT_SYMBOL(dev_get_flags);
3418 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3419 EXPORT_SYMBOL(br_handle_frame_hook);
3420 EXPORT_SYMBOL(br_fdb_get_hook);
3421 EXPORT_SYMBOL(br_fdb_put_hook);
3425 EXPORT_SYMBOL(dev_load);
3428 EXPORT_PER_CPU_SYMBOL(softnet_data);