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 struct notifier_block *netdev_chain;
199 * Device drivers call our routines to queue packets here. We empty the
200 * queue in the local softnet handler.
202 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
205 extern int netdev_sysfs_init(void);
206 extern int netdev_register_sysfs(struct net_device *);
207 extern void netdev_unregister_sysfs(struct net_device *);
209 #define netdev_sysfs_init() (0)
210 #define netdev_register_sysfs(dev) (0)
211 #define netdev_unregister_sysfs(dev) do { } while(0)
215 /*******************************************************************************
217 Protocol management and registration routines
219 *******************************************************************************/
228 * Add a protocol ID to the list. Now that the input handler is
229 * smarter we can dispense with all the messy stuff that used to be
232 * BEWARE!!! Protocol handlers, mangling input packets,
233 * MUST BE last in hash buckets and checking protocol handlers
234 * MUST start from promiscuous ptype_all chain in net_bh.
235 * It is true now, do not change it.
236 * Explanation follows: if protocol handler, mangling packet, will
237 * be the first on list, it is not able to sense, that packet
238 * is cloned and should be copied-on-write, so that it will
239 * change it and subsequent readers will get broken packet.
244 * dev_add_pack - add packet handler
245 * @pt: packet type declaration
247 * Add a protocol handler to the networking stack. The passed &packet_type
248 * is linked into kernel lists and may not be freed until it has been
249 * removed from the kernel lists.
251 * This call does not sleep therefore it can not
252 * guarantee all CPU's that are in middle of receiving packets
253 * will see the new packet type (until the next received packet).
256 void dev_add_pack(struct packet_type *pt)
260 spin_lock_bh(&ptype_lock);
261 if (pt->type == htons(ETH_P_ALL)) {
263 list_add_rcu(&pt->list, &ptype_all);
265 hash = ntohs(pt->type) & 15;
266 list_add_rcu(&pt->list, &ptype_base[hash]);
268 spin_unlock_bh(&ptype_lock);
272 * __dev_remove_pack - remove packet handler
273 * @pt: packet type declaration
275 * Remove a protocol handler that was previously added to the kernel
276 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
277 * from the kernel lists and can be freed or reused once this function
280 * The packet type might still be in use by receivers
281 * and must not be freed until after all the CPU's have gone
282 * through a quiescent state.
284 void __dev_remove_pack(struct packet_type *pt)
286 struct list_head *head;
287 struct packet_type *pt1;
289 spin_lock_bh(&ptype_lock);
291 if (pt->type == htons(ETH_P_ALL)) {
295 head = &ptype_base[ntohs(pt->type) & 15];
297 list_for_each_entry(pt1, head, list) {
299 list_del_rcu(&pt->list);
304 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
306 spin_unlock_bh(&ptype_lock);
309 * dev_remove_pack - remove packet handler
310 * @pt: packet type declaration
312 * Remove a protocol handler that was previously added to the kernel
313 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
314 * from the kernel lists and can be freed or reused once this function
317 * This call sleeps to guarantee that no CPU is looking at the packet
320 void dev_remove_pack(struct packet_type *pt)
322 __dev_remove_pack(pt);
327 /******************************************************************************
329 Device Boot-time Settings Routines
331 *******************************************************************************/
333 /* Boot time configuration table */
334 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
337 * netdev_boot_setup_add - add new setup entry
338 * @name: name of the device
339 * @map: configured settings for the device
341 * Adds new setup entry to the dev_boot_setup list. The function
342 * returns 0 on error and 1 on success. This is a generic routine to
345 static int netdev_boot_setup_add(char *name, struct ifmap *map)
347 struct netdev_boot_setup *s;
351 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
352 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
353 memset(s[i].name, 0, sizeof(s[i].name));
354 strcpy(s[i].name, name);
355 memcpy(&s[i].map, map, sizeof(s[i].map));
360 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
364 * netdev_boot_setup_check - check boot time settings
365 * @dev: the netdevice
367 * Check boot time settings for the device.
368 * The found settings are set for the device to be used
369 * later in the device probing.
370 * Returns 0 if no settings found, 1 if they are.
372 int netdev_boot_setup_check(struct net_device *dev)
374 struct netdev_boot_setup *s = dev_boot_setup;
377 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
378 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
379 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
380 dev->irq = s[i].map.irq;
381 dev->base_addr = s[i].map.base_addr;
382 dev->mem_start = s[i].map.mem_start;
383 dev->mem_end = s[i].map.mem_end;
392 * netdev_boot_base - get address from boot time settings
393 * @prefix: prefix for network device
394 * @unit: id for network device
396 * Check boot time settings for the base address of device.
397 * The found settings are set for the device to be used
398 * later in the device probing.
399 * Returns 0 if no settings found.
401 unsigned long netdev_boot_base(const char *prefix, int unit)
403 const struct netdev_boot_setup *s = dev_boot_setup;
407 sprintf(name, "%s%d", prefix, unit);
410 * If device already registered then return base of 1
411 * to indicate not to probe for this interface
413 if (__dev_get_by_name(name))
416 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
417 if (!strcmp(name, s[i].name))
418 return s[i].map.base_addr;
423 * Saves at boot time configured settings for any netdevice.
425 int __init netdev_boot_setup(char *str)
430 str = get_options(str, ARRAY_SIZE(ints), ints);
435 memset(&map, 0, sizeof(map));
439 map.base_addr = ints[2];
441 map.mem_start = ints[3];
443 map.mem_end = ints[4];
445 /* Add new entry to the list */
446 return netdev_boot_setup_add(str, &map);
449 __setup("netdev=", netdev_boot_setup);
451 /*******************************************************************************
453 Device Interface Subroutines
455 *******************************************************************************/
458 * __dev_get_by_name - find a device by its name
459 * @name: name to find
461 * Find an interface by name. Must be called under RTNL semaphore
462 * or @dev_base_lock. If the name is found a pointer to the device
463 * is returned. If the name is not found then %NULL is returned. The
464 * reference counters are not incremented so the caller must be
465 * careful with locks.
468 struct net_device *__dev_get_by_name(const char *name)
470 struct hlist_node *p;
472 hlist_for_each(p, dev_name_hash(name)) {
473 struct net_device *dev
474 = hlist_entry(p, struct net_device, name_hlist);
475 if (!strncmp(dev->name, name, IFNAMSIZ))
482 * dev_get_by_name - find a device by its name
483 * @name: name to find
485 * Find an interface by name. This can be called from any
486 * context and does its own locking. The returned handle has
487 * the usage count incremented and the caller must use dev_put() to
488 * release it when it is no longer needed. %NULL is returned if no
489 * matching device is found.
492 struct net_device *dev_get_by_name(const char *name)
494 struct net_device *dev;
496 read_lock(&dev_base_lock);
497 dev = __dev_get_by_name(name);
500 read_unlock(&dev_base_lock);
505 * __dev_get_by_index - find a device by its ifindex
506 * @ifindex: index of device
508 * Search for an interface by index. Returns %NULL if the device
509 * is not found or a pointer to the device. The device has not
510 * had its reference counter increased so the caller must be careful
511 * about locking. The caller must hold either the RTNL semaphore
515 struct net_device *__dev_get_by_index(int ifindex)
517 struct hlist_node *p;
519 hlist_for_each(p, dev_index_hash(ifindex)) {
520 struct net_device *dev
521 = hlist_entry(p, struct net_device, index_hlist);
522 if (dev->ifindex == ifindex)
530 * dev_get_by_index - find a device by its ifindex
531 * @ifindex: index of device
533 * Search for an interface by index. Returns NULL if the device
534 * is not found or a pointer to the device. The device returned has
535 * had a reference added and the pointer is safe until the user calls
536 * dev_put to indicate they have finished with it.
539 struct net_device *dev_get_by_index(int ifindex)
541 struct net_device *dev;
543 read_lock(&dev_base_lock);
544 dev = __dev_get_by_index(ifindex);
547 read_unlock(&dev_base_lock);
552 * dev_getbyhwaddr - find a device by its hardware address
553 * @type: media type of device
554 * @ha: hardware address
556 * Search for an interface by MAC address. Returns NULL if the device
557 * is not found or a pointer to the device. The caller must hold the
558 * rtnl semaphore. The returned device has not had its ref count increased
559 * and the caller must therefore be careful about locking
562 * If the API was consistent this would be __dev_get_by_hwaddr
565 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
567 struct net_device *dev;
571 for (dev = dev_base; dev; dev = dev->next)
572 if (dev->type == type &&
573 !memcmp(dev->dev_addr, ha, dev->addr_len))
578 EXPORT_SYMBOL(dev_getbyhwaddr);
580 struct net_device *dev_getfirstbyhwtype(unsigned short type)
582 struct net_device *dev;
585 for (dev = dev_base; dev; dev = dev->next) {
586 if (dev->type == type) {
595 EXPORT_SYMBOL(dev_getfirstbyhwtype);
598 * dev_get_by_flags - find any device with given flags
599 * @if_flags: IFF_* values
600 * @mask: bitmask of bits in if_flags to check
602 * Search for any interface with the given flags. Returns NULL if a device
603 * is not found or a pointer to the device. The device returned has
604 * had a reference added and the pointer is safe until the user calls
605 * dev_put to indicate they have finished with it.
608 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
610 struct net_device *dev;
612 read_lock(&dev_base_lock);
613 for (dev = dev_base; dev != NULL; dev = dev->next) {
614 if (((dev->flags ^ if_flags) & mask) == 0) {
619 read_unlock(&dev_base_lock);
624 * dev_valid_name - check if name is okay for network device
627 * Network device names need to be valid file names to
628 * to allow sysfs to work
630 int dev_valid_name(const char *name)
632 return !(*name == '\0'
633 || !strcmp(name, ".")
634 || !strcmp(name, "..")
635 || strchr(name, '/'));
639 * dev_alloc_name - allocate a name for a device
641 * @name: name format string
643 * Passed a format string - eg "lt%d" it will try and find a suitable
644 * id. Not efficient for many devices, not called a lot. The caller
645 * must hold the dev_base or rtnl lock while allocating the name and
646 * adding the device in order to avoid duplicates. Returns the number
647 * of the unit assigned or a negative errno code.
650 int dev_alloc_name(struct net_device *dev, const char *name)
655 const int max_netdevices = 8*PAGE_SIZE;
657 struct net_device *d;
659 p = strnchr(name, IFNAMSIZ-1, '%');
662 * Verify the string as this thing may have come from
663 * the user. There must be either one "%d" and no other "%"
666 if (p[1] != 'd' || strchr(p + 2, '%'))
669 /* Use one page as a bit array of possible slots */
670 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
674 for (d = dev_base; d; d = d->next) {
675 if (!sscanf(d->name, name, &i))
677 if (i < 0 || i >= max_netdevices)
680 /* avoid cases where sscanf is not exact inverse of printf */
681 snprintf(buf, sizeof(buf), name, i);
682 if (!strncmp(buf, d->name, IFNAMSIZ))
686 i = find_first_zero_bit(inuse, max_netdevices);
687 free_page((unsigned long) inuse);
690 snprintf(buf, sizeof(buf), name, i);
691 if (!__dev_get_by_name(buf)) {
692 strlcpy(dev->name, buf, IFNAMSIZ);
696 /* It is possible to run out of possible slots
697 * when the name is long and there isn't enough space left
698 * for the digits, or if all bits are used.
705 * dev_change_name - change name of a device
707 * @newname: name (or format string) must be at least IFNAMSIZ
709 * Change name of a device, can pass format strings "eth%d".
712 int dev_change_name(struct net_device *dev, char *newname)
718 if (dev->flags & IFF_UP)
721 if (!dev_valid_name(newname))
724 if (strchr(newname, '%')) {
725 err = dev_alloc_name(dev, newname);
728 strcpy(newname, dev->name);
730 else if (__dev_get_by_name(newname))
733 strlcpy(dev->name, newname, IFNAMSIZ);
735 err = class_device_rename(&dev->class_dev, dev->name);
737 hlist_del(&dev->name_hlist);
738 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
739 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
746 * netdev_features_change - device changes fatures
747 * @dev: device to cause notification
749 * Called to indicate a device has changed features.
751 void netdev_features_change(struct net_device *dev)
753 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
755 EXPORT_SYMBOL(netdev_features_change);
758 * netdev_state_change - device changes state
759 * @dev: device to cause notification
761 * Called to indicate a device has changed state. This function calls
762 * the notifier chains for netdev_chain and sends a NEWLINK message
763 * to the routing socket.
765 void netdev_state_change(struct net_device *dev)
767 if (dev->flags & IFF_UP) {
768 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
769 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
774 * dev_load - load a network module
775 * @name: name of interface
777 * If a network interface is not present and the process has suitable
778 * privileges this function loads the module. If module loading is not
779 * available in this kernel then it becomes a nop.
782 void dev_load(const char *name)
784 struct net_device *dev;
786 read_lock(&dev_base_lock);
787 dev = __dev_get_by_name(name);
788 read_unlock(&dev_base_lock);
790 if (!dev && capable(CAP_SYS_MODULE))
791 request_module("%s", name);
794 static int default_rebuild_header(struct sk_buff *skb)
796 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
797 skb->dev ? skb->dev->name : "NULL!!!");
804 * dev_open - prepare an interface for use.
805 * @dev: device to open
807 * Takes a device from down to up state. The device's private open
808 * function is invoked and then the multicast lists are loaded. Finally
809 * the device is moved into the up state and a %NETDEV_UP message is
810 * sent to the netdev notifier chain.
812 * Calling this function on an active interface is a nop. On a failure
813 * a negative errno code is returned.
815 int dev_open(struct net_device *dev)
823 if (dev->flags & IFF_UP)
827 * Is it even present?
829 if (!netif_device_present(dev))
833 * Call device private open method
835 set_bit(__LINK_STATE_START, &dev->state);
837 ret = dev->open(dev);
839 clear_bit(__LINK_STATE_START, &dev->state);
843 * If it went open OK then:
850 dev->flags |= IFF_UP;
853 * Initialize multicasting status
858 * Wakeup transmit queue engine
863 * ... and announce new interface.
865 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
871 * dev_close - shutdown an interface.
872 * @dev: device to shutdown
874 * This function moves an active device into down state. A
875 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
876 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
879 int dev_close(struct net_device *dev)
881 if (!(dev->flags & IFF_UP))
885 * Tell people we are going down, so that they can
886 * prepare to death, when device is still operating.
888 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
892 clear_bit(__LINK_STATE_START, &dev->state);
894 /* Synchronize to scheduled poll. We cannot touch poll list,
895 * it can be even on different cpu. So just clear netif_running(),
896 * and wait when poll really will happen. Actually, the best place
897 * for this is inside dev->stop() after device stopped its irq
898 * engine, but this requires more changes in devices. */
900 smp_mb__after_clear_bit(); /* Commit netif_running(). */
901 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
907 * Call the device specific close. This cannot fail.
908 * Only if device is UP
910 * We allow it to be called even after a DETACH hot-plug
917 * Device is now down.
920 dev->flags &= ~IFF_UP;
923 * Tell people we are down
925 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
932 * Device change register/unregister. These are not inline or static
933 * as we export them to the world.
937 * register_netdevice_notifier - register a network notifier block
940 * Register a notifier to be called when network device events occur.
941 * The notifier passed is linked into the kernel structures and must
942 * not be reused until it has been unregistered. A negative errno code
943 * is returned on a failure.
945 * When registered all registration and up events are replayed
946 * to the new notifier to allow device to have a race free
947 * view of the network device list.
950 int register_netdevice_notifier(struct notifier_block *nb)
952 struct net_device *dev;
956 err = notifier_chain_register(&netdev_chain, nb);
958 for (dev = dev_base; dev; dev = dev->next) {
959 nb->notifier_call(nb, NETDEV_REGISTER, dev);
961 if (dev->flags & IFF_UP)
962 nb->notifier_call(nb, NETDEV_UP, dev);
970 * unregister_netdevice_notifier - unregister a network notifier block
973 * Unregister a notifier previously registered by
974 * register_netdevice_notifier(). The notifier is unlinked into the
975 * kernel structures and may then be reused. A negative errno code
976 * is returned on a failure.
979 int unregister_netdevice_notifier(struct notifier_block *nb)
984 err = notifier_chain_unregister(&netdev_chain, nb);
990 * call_netdevice_notifiers - call all network notifier blocks
991 * @val: value passed unmodified to notifier function
992 * @v: pointer passed unmodified to notifier function
994 * Call all network notifier blocks. Parameters and return value
995 * are as for notifier_call_chain().
998 int call_netdevice_notifiers(unsigned long val, void *v)
1000 return notifier_call_chain(&netdev_chain, val, v);
1003 /* When > 0 there are consumers of rx skb time stamps */
1004 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1006 void net_enable_timestamp(void)
1008 atomic_inc(&netstamp_needed);
1011 void net_disable_timestamp(void)
1013 atomic_dec(&netstamp_needed);
1016 void __net_timestamp(struct sk_buff *skb)
1020 do_gettimeofday(&tv);
1021 skb_set_timestamp(skb, &tv);
1023 EXPORT_SYMBOL(__net_timestamp);
1025 static inline void net_timestamp(struct sk_buff *skb)
1027 if (atomic_read(&netstamp_needed))
1028 __net_timestamp(skb);
1030 skb->tstamp.off_sec = 0;
1031 skb->tstamp.off_usec = 0;
1036 * Support routine. Sends outgoing frames to any network
1037 * taps currently in use.
1040 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1042 struct packet_type *ptype;
1047 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1048 /* Never send packets back to the socket
1049 * they originated from - MvS (miquels@drinkel.ow.org)
1051 if ((ptype->dev == dev || !ptype->dev) &&
1052 (ptype->af_packet_priv == NULL ||
1053 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1054 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1058 /* skb->nh should be correctly
1059 set by sender, so that the second statement is
1060 just protection against buggy protocols.
1062 skb2->mac.raw = skb2->data;
1064 if (skb2->nh.raw < skb2->data ||
1065 skb2->nh.raw > skb2->tail) {
1066 if (net_ratelimit())
1067 printk(KERN_CRIT "protocol %04x is "
1069 skb2->protocol, dev->name);
1070 skb2->nh.raw = skb2->data;
1073 skb2->h.raw = skb2->nh.raw;
1074 skb2->pkt_type = PACKET_OUTGOING;
1075 ptype->func(skb2, skb->dev, ptype, skb->dev);
1082 * Invalidate hardware checksum when packet is to be mangled, and
1083 * complete checksum manually on outgoing path.
1085 int skb_checksum_help(struct sk_buff *skb, int inward)
1088 int ret = 0, offset = skb->h.raw - skb->data;
1091 skb->ip_summed = CHECKSUM_NONE;
1095 if (skb_cloned(skb)) {
1096 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1101 BUG_ON(offset > (int)skb->len);
1102 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1104 offset = skb->tail - skb->h.raw;
1105 BUG_ON(offset <= 0);
1106 BUG_ON(skb->csum + 2 > offset);
1108 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1109 skb->ip_summed = CHECKSUM_NONE;
1114 /* Take action when hardware reception checksum errors are detected. */
1116 void netdev_rx_csum_fault(struct net_device *dev)
1118 if (net_ratelimit()) {
1119 printk(KERN_ERR "%s: hw csum failure.\n",
1120 dev ? dev->name : "<unknown>");
1124 EXPORT_SYMBOL(netdev_rx_csum_fault);
1127 #ifdef CONFIG_HIGHMEM
1128 /* Actually, we should eliminate this check as soon as we know, that:
1129 * 1. IOMMU is present and allows to map all the memory.
1130 * 2. No high memory really exists on this machine.
1133 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1137 if (dev->features & NETIF_F_HIGHDMA)
1140 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1141 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1147 #define illegal_highdma(dev, skb) (0)
1150 /* Keep head the same: replace data */
1151 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1156 struct skb_shared_info *ninfo;
1157 int headerlen = skb->data - skb->head;
1158 int expand = (skb->tail + skb->data_len) - skb->end;
1160 if (skb_shared(skb))
1166 size = skb->end - skb->head + expand;
1167 size = SKB_DATA_ALIGN(size);
1168 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1172 /* Copy entire thing */
1173 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1177 ninfo = (struct skb_shared_info*)(data + size);
1178 atomic_set(&ninfo->dataref, 1);
1179 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1180 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1181 ninfo->nr_frags = 0;
1182 ninfo->frag_list = NULL;
1184 /* Offset between the two in bytes */
1185 offset = data - skb->head;
1187 /* Free old data. */
1188 skb_release_data(skb);
1191 skb->end = data + size;
1193 /* Set up new pointers */
1194 skb->h.raw += offset;
1195 skb->nh.raw += offset;
1196 skb->mac.raw += offset;
1197 skb->tail += offset;
1198 skb->data += offset;
1200 /* We are no longer a clone, even if we were. */
1203 skb->tail += skb->data_len;
1208 #define HARD_TX_LOCK(dev, cpu) { \
1209 if ((dev->features & NETIF_F_LLTX) == 0) { \
1210 spin_lock(&dev->xmit_lock); \
1211 dev->xmit_lock_owner = cpu; \
1215 #define HARD_TX_UNLOCK(dev) { \
1216 if ((dev->features & NETIF_F_LLTX) == 0) { \
1217 dev->xmit_lock_owner = -1; \
1218 spin_unlock(&dev->xmit_lock); \
1223 * dev_queue_xmit - transmit a buffer
1224 * @skb: buffer to transmit
1226 * Queue a buffer for transmission to a network device. The caller must
1227 * have set the device and priority and built the buffer before calling
1228 * this function. The function can be called from an interrupt.
1230 * A negative errno code is returned on a failure. A success does not
1231 * guarantee the frame will be transmitted as it may be dropped due
1232 * to congestion or traffic shaping.
1234 * -----------------------------------------------------------------------------------
1235 * I notice this method can also return errors from the queue disciplines,
1236 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1239 * Regardless of the return value, the skb is consumed, so it is currently
1240 * difficult to retry a send to this method. (You can bump the ref count
1241 * before sending to hold a reference for retry if you are careful.)
1243 * When calling this method, interrupts MUST be enabled. This is because
1244 * the BH enable code must have IRQs enabled so that it will not deadlock.
1248 int dev_queue_xmit(struct sk_buff *skb)
1250 struct net_device *dev = skb->dev;
1254 if (skb_shinfo(skb)->frag_list &&
1255 !(dev->features & NETIF_F_FRAGLIST) &&
1256 __skb_linearize(skb, GFP_ATOMIC))
1259 /* Fragmented skb is linearized if device does not support SG,
1260 * or if at least one of fragments is in highmem and device
1261 * does not support DMA from it.
1263 if (skb_shinfo(skb)->nr_frags &&
1264 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1265 __skb_linearize(skb, GFP_ATOMIC))
1268 /* If packet is not checksummed and device does not support
1269 * checksumming for this protocol, complete checksumming here.
1271 if (skb->ip_summed == CHECKSUM_HW &&
1272 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1273 (!(dev->features & NETIF_F_IP_CSUM) ||
1274 skb->protocol != htons(ETH_P_IP))))
1275 if (skb_checksum_help(skb, 0))
1278 spin_lock_prefetch(&dev->queue_lock);
1280 /* Disable soft irqs for various locks below. Also
1281 * stops preemption for RCU.
1285 /* Updates of qdisc are serialized by queue_lock.
1286 * The struct Qdisc which is pointed to by qdisc is now a
1287 * rcu structure - it may be accessed without acquiring
1288 * a lock (but the structure may be stale.) The freeing of the
1289 * qdisc will be deferred until it's known that there are no
1290 * more references to it.
1292 * If the qdisc has an enqueue function, we still need to
1293 * hold the queue_lock before calling it, since queue_lock
1294 * also serializes access to the device queue.
1297 q = rcu_dereference(dev->qdisc);
1298 #ifdef CONFIG_NET_CLS_ACT
1299 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1302 /* Grab device queue */
1303 spin_lock(&dev->queue_lock);
1305 rc = q->enqueue(skb, q);
1309 spin_unlock(&dev->queue_lock);
1310 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1314 /* The device has no queue. Common case for software devices:
1315 loopback, all the sorts of tunnels...
1317 Really, it is unlikely that xmit_lock protection is necessary here.
1318 (f.e. loopback and IP tunnels are clean ignoring statistics
1320 However, it is possible, that they rely on protection
1323 Check this and shot the lock. It is not prone from deadlocks.
1324 Either shot noqueue qdisc, it is even simpler 8)
1326 if (dev->flags & IFF_UP) {
1327 int cpu = smp_processor_id(); /* ok because BHs are off */
1329 if (dev->xmit_lock_owner != cpu) {
1331 HARD_TX_LOCK(dev, cpu);
1333 if (!netif_queue_stopped(dev)) {
1335 dev_queue_xmit_nit(skb, dev);
1338 if (!dev->hard_start_xmit(skb, dev)) {
1339 HARD_TX_UNLOCK(dev);
1343 HARD_TX_UNLOCK(dev);
1344 if (net_ratelimit())
1345 printk(KERN_CRIT "Virtual device %s asks to "
1346 "queue packet!\n", dev->name);
1348 /* Recursion is detected! It is possible,
1350 if (net_ratelimit())
1351 printk(KERN_CRIT "Dead loop on virtual device "
1352 "%s, fix it urgently!\n", dev->name);
1368 /*=======================================================================
1370 =======================================================================*/
1372 int netdev_max_backlog = 1000;
1373 int netdev_budget = 300;
1374 int weight_p = 64; /* old backlog weight */
1376 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1380 * netif_rx - post buffer to the network code
1381 * @skb: buffer to post
1383 * This function receives a packet from a device driver and queues it for
1384 * the upper (protocol) levels to process. It always succeeds. The buffer
1385 * may be dropped during processing for congestion control or by the
1389 * NET_RX_SUCCESS (no congestion)
1390 * NET_RX_CN_LOW (low congestion)
1391 * NET_RX_CN_MOD (moderate congestion)
1392 * NET_RX_CN_HIGH (high congestion)
1393 * NET_RX_DROP (packet was dropped)
1397 int netif_rx(struct sk_buff *skb)
1399 struct softnet_data *queue;
1400 unsigned long flags;
1402 /* if netpoll wants it, pretend we never saw it */
1403 if (netpoll_rx(skb))
1406 if (!skb->tstamp.off_sec)
1410 * The code is rearranged so that the path is the most
1411 * short when CPU is congested, but is still operating.
1413 local_irq_save(flags);
1414 queue = &__get_cpu_var(softnet_data);
1416 __get_cpu_var(netdev_rx_stat).total++;
1417 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1418 if (queue->input_pkt_queue.qlen) {
1421 __skb_queue_tail(&queue->input_pkt_queue, skb);
1422 local_irq_restore(flags);
1423 return NET_RX_SUCCESS;
1426 netif_rx_schedule(&queue->backlog_dev);
1430 __get_cpu_var(netdev_rx_stat).dropped++;
1431 local_irq_restore(flags);
1437 int netif_rx_ni(struct sk_buff *skb)
1442 err = netif_rx(skb);
1443 if (local_softirq_pending())
1450 EXPORT_SYMBOL(netif_rx_ni);
1452 static inline struct net_device *skb_bond(struct sk_buff *skb)
1454 struct net_device *dev = skb->dev;
1458 * On bonding slaves other than the currently active
1459 * slave, suppress duplicates except for 802.3ad
1460 * ETH_P_SLOW and alb non-mcast/bcast.
1462 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1463 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1464 if (skb->pkt_type != PACKET_BROADCAST &&
1465 skb->pkt_type != PACKET_MULTICAST)
1469 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1470 skb->protocol == __constant_htons(ETH_P_SLOW))
1477 skb->dev = dev->master;
1483 static void net_tx_action(struct softirq_action *h)
1485 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1487 if (sd->completion_queue) {
1488 struct sk_buff *clist;
1490 local_irq_disable();
1491 clist = sd->completion_queue;
1492 sd->completion_queue = NULL;
1496 struct sk_buff *skb = clist;
1497 clist = clist->next;
1499 BUG_TRAP(!atomic_read(&skb->users));
1504 if (sd->output_queue) {
1505 struct net_device *head;
1507 local_irq_disable();
1508 head = sd->output_queue;
1509 sd->output_queue = NULL;
1513 struct net_device *dev = head;
1514 head = head->next_sched;
1516 smp_mb__before_clear_bit();
1517 clear_bit(__LINK_STATE_SCHED, &dev->state);
1519 if (spin_trylock(&dev->queue_lock)) {
1521 spin_unlock(&dev->queue_lock);
1523 netif_schedule(dev);
1529 static __inline__ int deliver_skb(struct sk_buff *skb,
1530 struct packet_type *pt_prev,
1531 struct net_device *orig_dev)
1533 atomic_inc(&skb->users);
1534 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1537 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1538 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1540 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1541 unsigned char *addr);
1542 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1544 static __inline__ int handle_bridge(struct sk_buff **pskb,
1545 struct packet_type **pt_prev, int *ret,
1546 struct net_device *orig_dev)
1548 struct net_bridge_port *port;
1550 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1551 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1555 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1559 return br_handle_frame_hook(port, pskb);
1562 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1565 #ifdef CONFIG_NET_CLS_ACT
1566 /* TODO: Maybe we should just force sch_ingress to be compiled in
1567 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1568 * a compare and 2 stores extra right now if we dont have it on
1569 * but have CONFIG_NET_CLS_ACT
1570 * NOTE: This doesnt stop any functionality; if you dont have
1571 * the ingress scheduler, you just cant add policies on ingress.
1574 static int ing_filter(struct sk_buff *skb)
1577 struct net_device *dev = skb->dev;
1578 int result = TC_ACT_OK;
1580 if (dev->qdisc_ingress) {
1581 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1582 if (MAX_RED_LOOP < ttl++) {
1583 printk("Redir loop detected Dropping packet (%s->%s)\n",
1584 skb->input_dev->name, skb->dev->name);
1588 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1590 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1592 spin_lock(&dev->ingress_lock);
1593 if ((q = dev->qdisc_ingress) != NULL)
1594 result = q->enqueue(skb, q);
1595 spin_unlock(&dev->ingress_lock);
1603 int netif_receive_skb(struct sk_buff *skb)
1605 struct packet_type *ptype, *pt_prev;
1606 struct net_device *orig_dev;
1607 int ret = NET_RX_DROP;
1608 unsigned short type;
1610 /* if we've gotten here through NAPI, check netpoll */
1611 if (skb->dev->poll && netpoll_rx(skb))
1614 if (!skb->tstamp.off_sec)
1617 if (!skb->input_dev)
1618 skb->input_dev = skb->dev;
1620 orig_dev = skb_bond(skb);
1625 __get_cpu_var(netdev_rx_stat).total++;
1627 skb->h.raw = skb->nh.raw = skb->data;
1628 skb->mac_len = skb->nh.raw - skb->mac.raw;
1634 #ifdef CONFIG_NET_CLS_ACT
1635 if (skb->tc_verd & TC_NCLS) {
1636 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1641 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1642 if (!ptype->dev || ptype->dev == skb->dev) {
1644 ret = deliver_skb(skb, pt_prev, orig_dev);
1649 #ifdef CONFIG_NET_CLS_ACT
1651 ret = deliver_skb(skb, pt_prev, orig_dev);
1652 pt_prev = NULL; /* noone else should process this after*/
1654 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1657 ret = ing_filter(skb);
1659 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1668 handle_diverter(skb);
1670 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1673 type = skb->protocol;
1674 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1675 if (ptype->type == type &&
1676 (!ptype->dev || ptype->dev == skb->dev)) {
1678 ret = deliver_skb(skb, pt_prev, orig_dev);
1684 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1687 /* Jamal, now you will not able to escape explaining
1688 * me how you were going to use this. :-)
1698 static int process_backlog(struct net_device *backlog_dev, int *budget)
1701 int quota = min(backlog_dev->quota, *budget);
1702 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1703 unsigned long start_time = jiffies;
1705 backlog_dev->weight = weight_p;
1707 struct sk_buff *skb;
1708 struct net_device *dev;
1710 local_irq_disable();
1711 skb = __skb_dequeue(&queue->input_pkt_queue);
1718 netif_receive_skb(skb);
1724 if (work >= quota || jiffies - start_time > 1)
1729 backlog_dev->quota -= work;
1734 backlog_dev->quota -= work;
1737 list_del(&backlog_dev->poll_list);
1738 smp_mb__before_clear_bit();
1739 netif_poll_enable(backlog_dev);
1745 static void net_rx_action(struct softirq_action *h)
1747 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1748 unsigned long start_time = jiffies;
1749 int budget = netdev_budget;
1752 local_irq_disable();
1754 while (!list_empty(&queue->poll_list)) {
1755 struct net_device *dev;
1757 if (budget <= 0 || jiffies - start_time > 1)
1762 dev = list_entry(queue->poll_list.next,
1763 struct net_device, poll_list);
1764 have = netpoll_poll_lock(dev);
1766 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1767 netpoll_poll_unlock(have);
1768 local_irq_disable();
1769 list_move_tail(&dev->poll_list, &queue->poll_list);
1771 dev->quota += dev->weight;
1773 dev->quota = dev->weight;
1775 netpoll_poll_unlock(have);
1777 local_irq_disable();
1785 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1786 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1790 static gifconf_func_t * gifconf_list [NPROTO];
1793 * register_gifconf - register a SIOCGIF handler
1794 * @family: Address family
1795 * @gifconf: Function handler
1797 * Register protocol dependent address dumping routines. The handler
1798 * that is passed must not be freed or reused until it has been replaced
1799 * by another handler.
1801 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1803 if (family >= NPROTO)
1805 gifconf_list[family] = gifconf;
1811 * Map an interface index to its name (SIOCGIFNAME)
1815 * We need this ioctl for efficient implementation of the
1816 * if_indextoname() function required by the IPv6 API. Without
1817 * it, we would have to search all the interfaces to find a
1821 static int dev_ifname(struct ifreq __user *arg)
1823 struct net_device *dev;
1827 * Fetch the caller's info block.
1830 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1833 read_lock(&dev_base_lock);
1834 dev = __dev_get_by_index(ifr.ifr_ifindex);
1836 read_unlock(&dev_base_lock);
1840 strcpy(ifr.ifr_name, dev->name);
1841 read_unlock(&dev_base_lock);
1843 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1849 * Perform a SIOCGIFCONF call. This structure will change
1850 * size eventually, and there is nothing I can do about it.
1851 * Thus we will need a 'compatibility mode'.
1854 static int dev_ifconf(char __user *arg)
1857 struct net_device *dev;
1864 * Fetch the caller's info block.
1867 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1874 * Loop over the interfaces, and write an info block for each.
1878 for (dev = dev_base; dev; dev = dev->next) {
1879 for (i = 0; i < NPROTO; i++) {
1880 if (gifconf_list[i]) {
1883 done = gifconf_list[i](dev, NULL, 0);
1885 done = gifconf_list[i](dev, pos + total,
1895 * All done. Write the updated control block back to the caller.
1897 ifc.ifc_len = total;
1900 * Both BSD and Solaris return 0 here, so we do too.
1902 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1905 #ifdef CONFIG_PROC_FS
1907 * This is invoked by the /proc filesystem handler to display a device
1910 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1912 struct net_device *dev;
1915 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1917 return i == pos ? dev : NULL;
1920 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1922 read_lock(&dev_base_lock);
1923 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1926 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1929 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1932 void dev_seq_stop(struct seq_file *seq, void *v)
1934 read_unlock(&dev_base_lock);
1937 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1939 if (dev->get_stats) {
1940 struct net_device_stats *stats = dev->get_stats(dev);
1942 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1943 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1944 dev->name, stats->rx_bytes, stats->rx_packets,
1946 stats->rx_dropped + stats->rx_missed_errors,
1947 stats->rx_fifo_errors,
1948 stats->rx_length_errors + stats->rx_over_errors +
1949 stats->rx_crc_errors + stats->rx_frame_errors,
1950 stats->rx_compressed, stats->multicast,
1951 stats->tx_bytes, stats->tx_packets,
1952 stats->tx_errors, stats->tx_dropped,
1953 stats->tx_fifo_errors, stats->collisions,
1954 stats->tx_carrier_errors +
1955 stats->tx_aborted_errors +
1956 stats->tx_window_errors +
1957 stats->tx_heartbeat_errors,
1958 stats->tx_compressed);
1960 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1964 * Called from the PROCfs module. This now uses the new arbitrary sized
1965 * /proc/net interface to create /proc/net/dev
1967 static int dev_seq_show(struct seq_file *seq, void *v)
1969 if (v == SEQ_START_TOKEN)
1970 seq_puts(seq, "Inter-| Receive "
1972 " face |bytes packets errs drop fifo frame "
1973 "compressed multicast|bytes packets errs "
1974 "drop fifo colls carrier compressed\n");
1976 dev_seq_printf_stats(seq, v);
1980 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1982 struct netif_rx_stats *rc = NULL;
1984 while (*pos < NR_CPUS)
1985 if (cpu_online(*pos)) {
1986 rc = &per_cpu(netdev_rx_stat, *pos);
1993 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1995 return softnet_get_online(pos);
1998 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2001 return softnet_get_online(pos);
2004 static void softnet_seq_stop(struct seq_file *seq, void *v)
2008 static int softnet_seq_show(struct seq_file *seq, void *v)
2010 struct netif_rx_stats *s = v;
2012 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2013 s->total, s->dropped, s->time_squeeze, 0,
2014 0, 0, 0, 0, /* was fastroute */
2019 static struct seq_operations dev_seq_ops = {
2020 .start = dev_seq_start,
2021 .next = dev_seq_next,
2022 .stop = dev_seq_stop,
2023 .show = dev_seq_show,
2026 static int dev_seq_open(struct inode *inode, struct file *file)
2028 return seq_open(file, &dev_seq_ops);
2031 static struct file_operations dev_seq_fops = {
2032 .owner = THIS_MODULE,
2033 .open = dev_seq_open,
2035 .llseek = seq_lseek,
2036 .release = seq_release,
2039 static struct seq_operations softnet_seq_ops = {
2040 .start = softnet_seq_start,
2041 .next = softnet_seq_next,
2042 .stop = softnet_seq_stop,
2043 .show = softnet_seq_show,
2046 static int softnet_seq_open(struct inode *inode, struct file *file)
2048 return seq_open(file, &softnet_seq_ops);
2051 static struct file_operations softnet_seq_fops = {
2052 .owner = THIS_MODULE,
2053 .open = softnet_seq_open,
2055 .llseek = seq_lseek,
2056 .release = seq_release,
2059 #ifdef CONFIG_WIRELESS_EXT
2060 extern int wireless_proc_init(void);
2062 #define wireless_proc_init() 0
2065 static int __init dev_proc_init(void)
2069 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2071 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2073 if (wireless_proc_init())
2079 proc_net_remove("softnet_stat");
2081 proc_net_remove("dev");
2085 #define dev_proc_init() 0
2086 #endif /* CONFIG_PROC_FS */
2090 * netdev_set_master - set up master/slave pair
2091 * @slave: slave device
2092 * @master: new master device
2094 * Changes the master device of the slave. Pass %NULL to break the
2095 * bonding. The caller must hold the RTNL semaphore. On a failure
2096 * a negative errno code is returned. On success the reference counts
2097 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2098 * function returns zero.
2100 int netdev_set_master(struct net_device *slave, struct net_device *master)
2102 struct net_device *old = slave->master;
2112 slave->master = master;
2120 slave->flags |= IFF_SLAVE;
2122 slave->flags &= ~IFF_SLAVE;
2124 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2129 * dev_set_promiscuity - update promiscuity count on a device
2133 * Add or remove promsicuity from a device. While the count in the device
2134 * remains above zero the interface remains promiscuous. Once it hits zero
2135 * the device reverts back to normal filtering operation. A negative inc
2136 * value is used to drop promiscuity on the device.
2138 void dev_set_promiscuity(struct net_device *dev, int inc)
2140 unsigned short old_flags = dev->flags;
2142 if ((dev->promiscuity += inc) == 0)
2143 dev->flags &= ~IFF_PROMISC;
2145 dev->flags |= IFF_PROMISC;
2146 if (dev->flags != old_flags) {
2148 printk(KERN_INFO "device %s %s promiscuous mode\n",
2149 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2151 audit_log(current->audit_context, GFP_ATOMIC,
2152 AUDIT_ANOM_PROMISCUOUS,
2153 "dev=%s prom=%d old_prom=%d auid=%u",
2154 dev->name, (dev->flags & IFF_PROMISC),
2155 (old_flags & IFF_PROMISC),
2156 audit_get_loginuid(current->audit_context));
2161 * dev_set_allmulti - update allmulti count on a device
2165 * Add or remove reception of all multicast frames to a device. While the
2166 * count in the device remains above zero the interface remains listening
2167 * to all interfaces. Once it hits zero the device reverts back to normal
2168 * filtering operation. A negative @inc value is used to drop the counter
2169 * when releasing a resource needing all multicasts.
2172 void dev_set_allmulti(struct net_device *dev, int inc)
2174 unsigned short old_flags = dev->flags;
2176 dev->flags |= IFF_ALLMULTI;
2177 if ((dev->allmulti += inc) == 0)
2178 dev->flags &= ~IFF_ALLMULTI;
2179 if (dev->flags ^ old_flags)
2183 unsigned dev_get_flags(const struct net_device *dev)
2187 flags = (dev->flags & ~(IFF_PROMISC |
2192 (dev->gflags & (IFF_PROMISC |
2195 if (netif_running(dev)) {
2196 if (netif_oper_up(dev))
2197 flags |= IFF_RUNNING;
2198 if (netif_carrier_ok(dev))
2199 flags |= IFF_LOWER_UP;
2200 if (netif_dormant(dev))
2201 flags |= IFF_DORMANT;
2207 int dev_change_flags(struct net_device *dev, unsigned flags)
2210 int old_flags = dev->flags;
2213 * Set the flags on our device.
2216 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2217 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2219 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2223 * Load in the correct multicast list now the flags have changed.
2229 * Have we downed the interface. We handle IFF_UP ourselves
2230 * according to user attempts to set it, rather than blindly
2235 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2236 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2242 if (dev->flags & IFF_UP &&
2243 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2245 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2247 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2248 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2249 dev->gflags ^= IFF_PROMISC;
2250 dev_set_promiscuity(dev, inc);
2253 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2254 is important. Some (broken) drivers set IFF_PROMISC, when
2255 IFF_ALLMULTI is requested not asking us and not reporting.
2257 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2258 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2259 dev->gflags ^= IFF_ALLMULTI;
2260 dev_set_allmulti(dev, inc);
2263 if (old_flags ^ dev->flags)
2264 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2269 int dev_set_mtu(struct net_device *dev, int new_mtu)
2273 if (new_mtu == dev->mtu)
2276 /* MTU must be positive. */
2280 if (!netif_device_present(dev))
2284 if (dev->change_mtu)
2285 err = dev->change_mtu(dev, new_mtu);
2288 if (!err && dev->flags & IFF_UP)
2289 notifier_call_chain(&netdev_chain,
2290 NETDEV_CHANGEMTU, dev);
2294 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2298 if (!dev->set_mac_address)
2300 if (sa->sa_family != dev->type)
2302 if (!netif_device_present(dev))
2304 err = dev->set_mac_address(dev, sa);
2306 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2311 * Perform the SIOCxIFxxx calls.
2313 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2316 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2322 case SIOCGIFFLAGS: /* Get interface flags */
2323 ifr->ifr_flags = dev_get_flags(dev);
2326 case SIOCSIFFLAGS: /* Set interface flags */
2327 return dev_change_flags(dev, ifr->ifr_flags);
2329 case SIOCGIFMETRIC: /* Get the metric on the interface
2330 (currently unused) */
2331 ifr->ifr_metric = 0;
2334 case SIOCSIFMETRIC: /* Set the metric on the interface
2335 (currently unused) */
2338 case SIOCGIFMTU: /* Get the MTU of a device */
2339 ifr->ifr_mtu = dev->mtu;
2342 case SIOCSIFMTU: /* Set the MTU of a device */
2343 return dev_set_mtu(dev, ifr->ifr_mtu);
2347 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2349 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2350 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2351 ifr->ifr_hwaddr.sa_family = dev->type;
2355 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2357 case SIOCSIFHWBROADCAST:
2358 if (ifr->ifr_hwaddr.sa_family != dev->type)
2360 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2361 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2362 notifier_call_chain(&netdev_chain,
2363 NETDEV_CHANGEADDR, dev);
2367 ifr->ifr_map.mem_start = dev->mem_start;
2368 ifr->ifr_map.mem_end = dev->mem_end;
2369 ifr->ifr_map.base_addr = dev->base_addr;
2370 ifr->ifr_map.irq = dev->irq;
2371 ifr->ifr_map.dma = dev->dma;
2372 ifr->ifr_map.port = dev->if_port;
2376 if (dev->set_config) {
2377 if (!netif_device_present(dev))
2379 return dev->set_config(dev, &ifr->ifr_map);
2384 if (!dev->set_multicast_list ||
2385 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2387 if (!netif_device_present(dev))
2389 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2393 if (!dev->set_multicast_list ||
2394 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2396 if (!netif_device_present(dev))
2398 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2402 ifr->ifr_ifindex = dev->ifindex;
2406 ifr->ifr_qlen = dev->tx_queue_len;
2410 if (ifr->ifr_qlen < 0)
2412 dev->tx_queue_len = ifr->ifr_qlen;
2416 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2417 return dev_change_name(dev, ifr->ifr_newname);
2420 * Unknown or private ioctl
2424 if ((cmd >= SIOCDEVPRIVATE &&
2425 cmd <= SIOCDEVPRIVATE + 15) ||
2426 cmd == SIOCBONDENSLAVE ||
2427 cmd == SIOCBONDRELEASE ||
2428 cmd == SIOCBONDSETHWADDR ||
2429 cmd == SIOCBONDSLAVEINFOQUERY ||
2430 cmd == SIOCBONDINFOQUERY ||
2431 cmd == SIOCBONDCHANGEACTIVE ||
2432 cmd == SIOCGMIIPHY ||
2433 cmd == SIOCGMIIREG ||
2434 cmd == SIOCSMIIREG ||
2435 cmd == SIOCBRADDIF ||
2436 cmd == SIOCBRDELIF ||
2437 cmd == SIOCWANDEV) {
2439 if (dev->do_ioctl) {
2440 if (netif_device_present(dev))
2441 err = dev->do_ioctl(dev, ifr,
2454 * This function handles all "interface"-type I/O control requests. The actual
2455 * 'doing' part of this is dev_ifsioc above.
2459 * dev_ioctl - network device ioctl
2460 * @cmd: command to issue
2461 * @arg: pointer to a struct ifreq in user space
2463 * Issue ioctl functions to devices. This is normally called by the
2464 * user space syscall interfaces but can sometimes be useful for
2465 * other purposes. The return value is the return from the syscall if
2466 * positive or a negative errno code on error.
2469 int dev_ioctl(unsigned int cmd, void __user *arg)
2475 /* One special case: SIOCGIFCONF takes ifconf argument
2476 and requires shared lock, because it sleeps writing
2480 if (cmd == SIOCGIFCONF) {
2482 ret = dev_ifconf((char __user *) arg);
2486 if (cmd == SIOCGIFNAME)
2487 return dev_ifname((struct ifreq __user *)arg);
2489 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2492 ifr.ifr_name[IFNAMSIZ-1] = 0;
2494 colon = strchr(ifr.ifr_name, ':');
2499 * See which interface the caller is talking about.
2504 * These ioctl calls:
2505 * - can be done by all.
2506 * - atomic and do not require locking.
2517 dev_load(ifr.ifr_name);
2518 read_lock(&dev_base_lock);
2519 ret = dev_ifsioc(&ifr, cmd);
2520 read_unlock(&dev_base_lock);
2524 if (copy_to_user(arg, &ifr,
2525 sizeof(struct ifreq)))
2531 dev_load(ifr.ifr_name);
2533 ret = dev_ethtool(&ifr);
2538 if (copy_to_user(arg, &ifr,
2539 sizeof(struct ifreq)))
2545 * These ioctl calls:
2546 * - require superuser power.
2547 * - require strict serialization.
2553 if (!capable(CAP_NET_ADMIN))
2555 dev_load(ifr.ifr_name);
2557 ret = dev_ifsioc(&ifr, cmd);
2562 if (copy_to_user(arg, &ifr,
2563 sizeof(struct ifreq)))
2569 * These ioctl calls:
2570 * - require superuser power.
2571 * - require strict serialization.
2572 * - do not return a value
2582 case SIOCSIFHWBROADCAST:
2585 case SIOCBONDENSLAVE:
2586 case SIOCBONDRELEASE:
2587 case SIOCBONDSETHWADDR:
2588 case SIOCBONDCHANGEACTIVE:
2591 if (!capable(CAP_NET_ADMIN))
2594 case SIOCBONDSLAVEINFOQUERY:
2595 case SIOCBONDINFOQUERY:
2596 dev_load(ifr.ifr_name);
2598 ret = dev_ifsioc(&ifr, cmd);
2603 /* Get the per device memory space. We can add this but
2604 * currently do not support it */
2606 /* Set the per device memory buffer space.
2607 * Not applicable in our case */
2612 * Unknown or private ioctl.
2615 if (cmd == SIOCWANDEV ||
2616 (cmd >= SIOCDEVPRIVATE &&
2617 cmd <= SIOCDEVPRIVATE + 15)) {
2618 dev_load(ifr.ifr_name);
2620 ret = dev_ifsioc(&ifr, cmd);
2622 if (!ret && copy_to_user(arg, &ifr,
2623 sizeof(struct ifreq)))
2627 #ifdef CONFIG_WIRELESS_EXT
2628 /* Take care of Wireless Extensions */
2629 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2630 /* If command is `set a parameter', or
2631 * `get the encoding parameters', check if
2632 * the user has the right to do it */
2633 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2634 if (!capable(CAP_NET_ADMIN))
2637 dev_load(ifr.ifr_name);
2639 /* Follow me in net/core/wireless.c */
2640 ret = wireless_process_ioctl(&ifr, cmd);
2642 if (IW_IS_GET(cmd) &&
2643 copy_to_user(arg, &ifr,
2644 sizeof(struct ifreq)))
2648 #endif /* CONFIG_WIRELESS_EXT */
2655 * dev_new_index - allocate an ifindex
2657 * Returns a suitable unique value for a new device interface
2658 * number. The caller must hold the rtnl semaphore or the
2659 * dev_base_lock to be sure it remains unique.
2661 static int dev_new_index(void)
2667 if (!__dev_get_by_index(ifindex))
2672 static int dev_boot_phase = 1;
2674 /* Delayed registration/unregisteration */
2675 static DEFINE_SPINLOCK(net_todo_list_lock);
2676 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2678 static inline void net_set_todo(struct net_device *dev)
2680 spin_lock(&net_todo_list_lock);
2681 list_add_tail(&dev->todo_list, &net_todo_list);
2682 spin_unlock(&net_todo_list_lock);
2686 * register_netdevice - register a network device
2687 * @dev: device to register
2689 * Take a completed network device structure and add it to the kernel
2690 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2691 * chain. 0 is returned on success. A negative errno code is returned
2692 * on a failure to set up the device, or if the name is a duplicate.
2694 * Callers must hold the rtnl semaphore. You may want
2695 * register_netdev() instead of this.
2698 * The locking appears insufficient to guarantee two parallel registers
2699 * will not get the same name.
2702 int register_netdevice(struct net_device *dev)
2704 struct hlist_head *head;
2705 struct hlist_node *p;
2708 BUG_ON(dev_boot_phase);
2711 /* When net_device's are persistent, this will be fatal. */
2712 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2714 spin_lock_init(&dev->queue_lock);
2715 spin_lock_init(&dev->xmit_lock);
2716 dev->xmit_lock_owner = -1;
2717 #ifdef CONFIG_NET_CLS_ACT
2718 spin_lock_init(&dev->ingress_lock);
2721 ret = alloc_divert_blk(dev);
2727 /* Init, if this function is available */
2729 ret = dev->init(dev);
2737 if (!dev_valid_name(dev->name)) {
2742 dev->ifindex = dev_new_index();
2743 if (dev->iflink == -1)
2744 dev->iflink = dev->ifindex;
2746 /* Check for existence of name */
2747 head = dev_name_hash(dev->name);
2748 hlist_for_each(p, head) {
2749 struct net_device *d
2750 = hlist_entry(p, struct net_device, name_hlist);
2751 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2757 /* Fix illegal SG+CSUM combinations. */
2758 if ((dev->features & NETIF_F_SG) &&
2759 !(dev->features & (NETIF_F_IP_CSUM |
2761 NETIF_F_HW_CSUM))) {
2762 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2764 dev->features &= ~NETIF_F_SG;
2767 /* TSO requires that SG is present as well. */
2768 if ((dev->features & NETIF_F_TSO) &&
2769 !(dev->features & NETIF_F_SG)) {
2770 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2772 dev->features &= ~NETIF_F_TSO;
2774 if (dev->features & NETIF_F_UFO) {
2775 if (!(dev->features & NETIF_F_HW_CSUM)) {
2776 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2777 "NETIF_F_HW_CSUM feature.\n",
2779 dev->features &= ~NETIF_F_UFO;
2781 if (!(dev->features & NETIF_F_SG)) {
2782 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2783 "NETIF_F_SG feature.\n",
2785 dev->features &= ~NETIF_F_UFO;
2790 * nil rebuild_header routine,
2791 * that should be never called and used as just bug trap.
2794 if (!dev->rebuild_header)
2795 dev->rebuild_header = default_rebuild_header;
2798 * Default initial state at registry is that the
2799 * device is present.
2802 set_bit(__LINK_STATE_PRESENT, &dev->state);
2805 dev_init_scheduler(dev);
2806 write_lock_bh(&dev_base_lock);
2808 dev_tail = &dev->next;
2809 hlist_add_head(&dev->name_hlist, head);
2810 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2812 dev->reg_state = NETREG_REGISTERING;
2813 write_unlock_bh(&dev_base_lock);
2815 /* Notify protocols, that a new device appeared. */
2816 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2818 /* Finish registration after unlock */
2825 free_divert_blk(dev);
2830 * register_netdev - register a network device
2831 * @dev: device to register
2833 * Take a completed network device structure and add it to the kernel
2834 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2835 * chain. 0 is returned on success. A negative errno code is returned
2836 * on a failure to set up the device, or if the name is a duplicate.
2838 * This is a wrapper around register_netdev that takes the rtnl semaphore
2839 * and expands the device name if you passed a format string to
2842 int register_netdev(struct net_device *dev)
2849 * If the name is a format string the caller wants us to do a
2852 if (strchr(dev->name, '%')) {
2853 err = dev_alloc_name(dev, dev->name);
2859 * Back compatibility hook. Kill this one in 2.5
2861 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2862 err = dev_alloc_name(dev, "eth%d");
2867 err = register_netdevice(dev);
2872 EXPORT_SYMBOL(register_netdev);
2875 * netdev_wait_allrefs - wait until all references are gone.
2877 * This is called when unregistering network devices.
2879 * Any protocol or device that holds a reference should register
2880 * for netdevice notification, and cleanup and put back the
2881 * reference if they receive an UNREGISTER event.
2882 * We can get stuck here if buggy protocols don't correctly
2885 static void netdev_wait_allrefs(struct net_device *dev)
2887 unsigned long rebroadcast_time, warning_time;
2889 rebroadcast_time = warning_time = jiffies;
2890 while (atomic_read(&dev->refcnt) != 0) {
2891 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2894 /* Rebroadcast unregister notification */
2895 notifier_call_chain(&netdev_chain,
2896 NETDEV_UNREGISTER, dev);
2898 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2900 /* We must not have linkwatch events
2901 * pending on unregister. If this
2902 * happens, we simply run the queue
2903 * unscheduled, resulting in a noop
2906 linkwatch_run_queue();
2911 rebroadcast_time = jiffies;
2916 if (time_after(jiffies, warning_time + 10 * HZ)) {
2917 printk(KERN_EMERG "unregister_netdevice: "
2918 "waiting for %s to become free. Usage "
2920 dev->name, atomic_read(&dev->refcnt));
2921 warning_time = jiffies;
2930 * register_netdevice(x1);
2931 * register_netdevice(x2);
2933 * unregister_netdevice(y1);
2934 * unregister_netdevice(y2);
2940 * We are invoked by rtnl_unlock() after it drops the semaphore.
2941 * This allows us to deal with problems:
2942 * 1) We can create/delete sysfs objects which invoke hotplug
2943 * without deadlocking with linkwatch via keventd.
2944 * 2) Since we run with the RTNL semaphore not held, we can sleep
2945 * safely in order to wait for the netdev refcnt to drop to zero.
2947 static DEFINE_MUTEX(net_todo_run_mutex);
2948 void netdev_run_todo(void)
2950 struct list_head list = LIST_HEAD_INIT(list);
2954 /* Need to guard against multiple cpu's getting out of order. */
2955 mutex_lock(&net_todo_run_mutex);
2957 /* Not safe to do outside the semaphore. We must not return
2958 * until all unregister events invoked by the local processor
2959 * have been completed (either by this todo run, or one on
2962 if (list_empty(&net_todo_list))
2965 /* Snapshot list, allow later requests */
2966 spin_lock(&net_todo_list_lock);
2967 list_splice_init(&net_todo_list, &list);
2968 spin_unlock(&net_todo_list_lock);
2970 while (!list_empty(&list)) {
2971 struct net_device *dev
2972 = list_entry(list.next, struct net_device, todo_list);
2973 list_del(&dev->todo_list);
2975 switch(dev->reg_state) {
2976 case NETREG_REGISTERING:
2977 err = netdev_register_sysfs(dev);
2979 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2981 dev->reg_state = NETREG_REGISTERED;
2984 case NETREG_UNREGISTERING:
2985 netdev_unregister_sysfs(dev);
2986 dev->reg_state = NETREG_UNREGISTERED;
2988 netdev_wait_allrefs(dev);
2991 BUG_ON(atomic_read(&dev->refcnt));
2992 BUG_TRAP(!dev->ip_ptr);
2993 BUG_TRAP(!dev->ip6_ptr);
2994 BUG_TRAP(!dev->dn_ptr);
2997 /* It must be the very last action,
2998 * after this 'dev' may point to freed up memory.
3000 if (dev->destructor)
3001 dev->destructor(dev);
3005 printk(KERN_ERR "network todo '%s' but state %d\n",
3006 dev->name, dev->reg_state);
3012 mutex_unlock(&net_todo_run_mutex);
3016 * alloc_netdev - allocate network device
3017 * @sizeof_priv: size of private data to allocate space for
3018 * @name: device name format string
3019 * @setup: callback to initialize device
3021 * Allocates a struct net_device with private data area for driver use
3022 * and performs basic initialization.
3024 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3025 void (*setup)(struct net_device *))
3028 struct net_device *dev;
3031 /* ensure 32-byte alignment of both the device and private area */
3032 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3033 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3035 p = kmalloc(alloc_size, GFP_KERNEL);
3037 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3040 memset(p, 0, alloc_size);
3042 dev = (struct net_device *)
3043 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3044 dev->padded = (char *)dev - (char *)p;
3047 dev->priv = netdev_priv(dev);
3050 strcpy(dev->name, name);
3053 EXPORT_SYMBOL(alloc_netdev);
3056 * free_netdev - free network device
3059 * This function does the last stage of destroying an allocated device
3060 * interface. The reference to the device object is released.
3061 * If this is the last reference then it will be freed.
3063 void free_netdev(struct net_device *dev)
3066 /* Compatiablity with error handling in drivers */
3067 if (dev->reg_state == NETREG_UNINITIALIZED) {
3068 kfree((char *)dev - dev->padded);
3072 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3073 dev->reg_state = NETREG_RELEASED;
3075 /* will free via class release */
3076 class_device_put(&dev->class_dev);
3078 kfree((char *)dev - dev->padded);
3082 /* Synchronize with packet receive processing. */
3083 void synchronize_net(void)
3090 * unregister_netdevice - remove device from the kernel
3093 * This function shuts down a device interface and removes it
3094 * from the kernel tables. On success 0 is returned, on a failure
3095 * a negative errno code is returned.
3097 * Callers must hold the rtnl semaphore. You may want
3098 * unregister_netdev() instead of this.
3101 int unregister_netdevice(struct net_device *dev)
3103 struct net_device *d, **dp;
3105 BUG_ON(dev_boot_phase);
3108 /* Some devices call without registering for initialization unwind. */
3109 if (dev->reg_state == NETREG_UNINITIALIZED) {
3110 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3111 "was registered\n", dev->name, dev);
3115 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3117 /* If device is running, close it first. */
3118 if (dev->flags & IFF_UP)
3121 /* And unlink it from device chain. */
3122 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3124 write_lock_bh(&dev_base_lock);
3125 hlist_del(&dev->name_hlist);
3126 hlist_del(&dev->index_hlist);
3127 if (dev_tail == &dev->next)
3130 write_unlock_bh(&dev_base_lock);
3135 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3140 dev->reg_state = NETREG_UNREGISTERING;
3144 /* Shutdown queueing discipline. */
3148 /* Notify protocols, that we are about to destroy
3149 this device. They should clean all the things.
3151 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3154 * Flush the multicast chain
3156 dev_mc_discard(dev);
3161 /* Notifier chain MUST detach us from master device. */
3162 BUG_TRAP(!dev->master);
3164 free_divert_blk(dev);
3166 /* Finish processing unregister after unlock */
3176 * unregister_netdev - remove device from the kernel
3179 * This function shuts down a device interface and removes it
3180 * from the kernel tables. On success 0 is returned, on a failure
3181 * a negative errno code is returned.
3183 * This is just a wrapper for unregister_netdevice that takes
3184 * the rtnl semaphore. In general you want to use this and not
3185 * unregister_netdevice.
3187 void unregister_netdev(struct net_device *dev)
3190 unregister_netdevice(dev);
3194 EXPORT_SYMBOL(unregister_netdev);
3196 #ifdef CONFIG_HOTPLUG_CPU
3197 static int dev_cpu_callback(struct notifier_block *nfb,
3198 unsigned long action,
3201 struct sk_buff **list_skb;
3202 struct net_device **list_net;
3203 struct sk_buff *skb;
3204 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3205 struct softnet_data *sd, *oldsd;
3207 if (action != CPU_DEAD)
3210 local_irq_disable();
3211 cpu = smp_processor_id();
3212 sd = &per_cpu(softnet_data, cpu);
3213 oldsd = &per_cpu(softnet_data, oldcpu);
3215 /* Find end of our completion_queue. */
3216 list_skb = &sd->completion_queue;
3218 list_skb = &(*list_skb)->next;
3219 /* Append completion queue from offline CPU. */
3220 *list_skb = oldsd->completion_queue;
3221 oldsd->completion_queue = NULL;
3223 /* Find end of our output_queue. */
3224 list_net = &sd->output_queue;
3226 list_net = &(*list_net)->next_sched;
3227 /* Append output queue from offline CPU. */
3228 *list_net = oldsd->output_queue;
3229 oldsd->output_queue = NULL;
3231 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3234 /* Process offline CPU's input_pkt_queue */
3235 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3240 #endif /* CONFIG_HOTPLUG_CPU */
3244 * Initialize the DEV module. At boot time this walks the device list and
3245 * unhooks any devices that fail to initialise (normally hardware not
3246 * present) and leaves us with a valid list of present and active devices.
3251 * This is called single threaded during boot, so no need
3252 * to take the rtnl semaphore.
3254 static int __init net_dev_init(void)
3256 int i, rc = -ENOMEM;
3258 BUG_ON(!dev_boot_phase);
3262 if (dev_proc_init())
3265 if (netdev_sysfs_init())
3268 INIT_LIST_HEAD(&ptype_all);
3269 for (i = 0; i < 16; i++)
3270 INIT_LIST_HEAD(&ptype_base[i]);
3272 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3273 INIT_HLIST_HEAD(&dev_name_head[i]);
3275 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3276 INIT_HLIST_HEAD(&dev_index_head[i]);
3279 * Initialise the packet receive queues.
3283 struct softnet_data *queue;
3285 queue = &per_cpu(softnet_data, i);
3286 skb_queue_head_init(&queue->input_pkt_queue);
3287 queue->completion_queue = NULL;
3288 INIT_LIST_HEAD(&queue->poll_list);
3289 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3290 queue->backlog_dev.weight = weight_p;
3291 queue->backlog_dev.poll = process_backlog;
3292 atomic_set(&queue->backlog_dev.refcnt, 1);
3297 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3298 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3300 hotcpu_notifier(dev_cpu_callback, 0);
3308 subsys_initcall(net_dev_init);
3310 EXPORT_SYMBOL(__dev_get_by_index);
3311 EXPORT_SYMBOL(__dev_get_by_name);
3312 EXPORT_SYMBOL(__dev_remove_pack);
3313 EXPORT_SYMBOL(__skb_linearize);
3314 EXPORT_SYMBOL(dev_valid_name);
3315 EXPORT_SYMBOL(dev_add_pack);
3316 EXPORT_SYMBOL(dev_alloc_name);
3317 EXPORT_SYMBOL(dev_close);
3318 EXPORT_SYMBOL(dev_get_by_flags);
3319 EXPORT_SYMBOL(dev_get_by_index);
3320 EXPORT_SYMBOL(dev_get_by_name);
3321 EXPORT_SYMBOL(dev_open);
3322 EXPORT_SYMBOL(dev_queue_xmit);
3323 EXPORT_SYMBOL(dev_remove_pack);
3324 EXPORT_SYMBOL(dev_set_allmulti);
3325 EXPORT_SYMBOL(dev_set_promiscuity);
3326 EXPORT_SYMBOL(dev_change_flags);
3327 EXPORT_SYMBOL(dev_set_mtu);
3328 EXPORT_SYMBOL(dev_set_mac_address);
3329 EXPORT_SYMBOL(free_netdev);
3330 EXPORT_SYMBOL(netdev_boot_setup_check);
3331 EXPORT_SYMBOL(netdev_set_master);
3332 EXPORT_SYMBOL(netdev_state_change);
3333 EXPORT_SYMBOL(netif_receive_skb);
3334 EXPORT_SYMBOL(netif_rx);
3335 EXPORT_SYMBOL(register_gifconf);
3336 EXPORT_SYMBOL(register_netdevice);
3337 EXPORT_SYMBOL(register_netdevice_notifier);
3338 EXPORT_SYMBOL(skb_checksum_help);
3339 EXPORT_SYMBOL(synchronize_net);
3340 EXPORT_SYMBOL(unregister_netdevice);
3341 EXPORT_SYMBOL(unregister_netdevice_notifier);
3342 EXPORT_SYMBOL(net_enable_timestamp);
3343 EXPORT_SYMBOL(net_disable_timestamp);
3344 EXPORT_SYMBOL(dev_get_flags);
3346 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3347 EXPORT_SYMBOL(br_handle_frame_hook);
3348 EXPORT_SYMBOL(br_fdb_get_hook);
3349 EXPORT_SYMBOL(br_fdb_put_hook);
3353 EXPORT_SYMBOL(dev_load);
3356 EXPORT_PER_CPU_SYMBOL(softnet_data);