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
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
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/config.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/notifier.h>
93 #include <linux/skbuff.h>
95 #include <linux/rtnetlink.h>
96 #include <linux/proc_fs.h>
97 #include <linux/seq_file.h>
98 #include <linux/stat.h>
99 #include <linux/if_bridge.h>
100 #include <linux/divert.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #ifdef CONFIG_NET_RADIO
112 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
113 #include <net/iw_handler.h>
114 #endif /* CONFIG_NET_RADIO */
115 #include <asm/current.h>
116 #include <linux/vs_network.h>
118 /* This define, if set, will randomly drop a packet when congestion
119 * is more than moderate. It helps fairness in the multi-interface
120 * case when one of them is a hog, but it kills performance for the
121 * single interface case so it is off now by default.
125 /* Setting this will sample the queue lengths and thus congestion
126 * via a timer instead of as each packet is received.
128 #undef OFFLINE_SAMPLE
131 * The list of packet types we will receive (as opposed to discard)
132 * and the routines to invoke.
134 * Why 16. Because with 16 the only overlap we get on a hash of the
135 * low nibble of the protocol value is RARP/SNAP/X.25.
137 * NOTE: That is no longer true with the addition of VLAN tags. Not
138 * sure which should go first, but I bet it won't make much
139 * difference if we are running VLANs. The good news is that
140 * this protocol won't be in the list unless compiled in, so
141 * the average user (w/out VLANs) will not be adversly affected.
158 static spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
159 static struct list_head ptype_base[16]; /* 16 way hashed list */
160 static struct list_head ptype_all; /* Taps */
162 #ifdef OFFLINE_SAMPLE
163 static void sample_queue(unsigned long dummy);
164 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
168 * The @dev_base list is protected by @dev_base_lock and the rtln
171 * Pure readers hold dev_base_lock for reading.
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
186 struct net_device *dev_base;
187 struct net_device **dev_tail = &dev_base;
188 rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
190 EXPORT_SYMBOL(dev_base);
191 EXPORT_SYMBOL(dev_base_lock);
193 #define NETDEV_HASHBITS 8
194 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
195 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
197 static inline struct hlist_head *dev_name_hash(const char *name)
199 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
200 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
203 static inline struct hlist_head *dev_index_hash(int ifindex)
205 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
212 static struct notifier_block *netdev_chain;
215 * Device drivers call our routines to queue packets here. We empty the
216 * queue in the local softnet handler.
218 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
221 extern int netdev_sysfs_init(void);
222 extern int netdev_register_sysfs(struct net_device *);
223 extern void netdev_unregister_sysfs(struct net_device *);
225 #define netdev_sysfs_init() (0)
226 #define netdev_register_sysfs(dev) (0)
227 #define netdev_unregister_sysfs(dev) do { } while(0)
231 /*******************************************************************************
233 Protocol management and registration routines
235 *******************************************************************************/
244 * Add a protocol ID to the list. Now that the input handler is
245 * smarter we can dispense with all the messy stuff that used to be
248 * BEWARE!!! Protocol handlers, mangling input packets,
249 * MUST BE last in hash buckets and checking protocol handlers
250 * MUST start from promiscuous ptype_all chain in net_bh.
251 * It is true now, do not change it.
252 * Explanation follows: if protocol handler, mangling packet, will
253 * be the first on list, it is not able to sense, that packet
254 * is cloned and should be copied-on-write, so that it will
255 * change it and subsequent readers will get broken packet.
260 * dev_add_pack - add packet handler
261 * @pt: packet type declaration
263 * Add a protocol handler to the networking stack. The passed &packet_type
264 * is linked into kernel lists and may not be freed until it has been
265 * removed from the kernel lists.
267 * This call does not sleep therefore it can not
268 * guarantee all CPU's that are in middle of receiving packets
269 * will see the new packet type (until the next received packet).
272 void dev_add_pack(struct packet_type *pt)
276 spin_lock_bh(&ptype_lock);
277 if (pt->type == htons(ETH_P_ALL)) {
279 list_add_rcu(&pt->list, &ptype_all);
281 hash = ntohs(pt->type) & 15;
282 list_add_rcu(&pt->list, &ptype_base[hash]);
284 spin_unlock_bh(&ptype_lock);
287 extern void linkwatch_run_queue(void);
292 * __dev_remove_pack - remove packet handler
293 * @pt: packet type declaration
295 * Remove a protocol handler that was previously added to the kernel
296 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
297 * from the kernel lists and can be freed or reused once this function
300 * The packet type might still be in use by receivers
301 * and must not be freed until after all the CPU's have gone
302 * through a quiescent state.
304 void __dev_remove_pack(struct packet_type *pt)
306 struct list_head *head;
307 struct packet_type *pt1;
309 spin_lock_bh(&ptype_lock);
311 if (pt->type == htons(ETH_P_ALL)) {
315 head = &ptype_base[ntohs(pt->type) & 15];
317 list_for_each_entry(pt1, head, list) {
319 list_del_rcu(&pt->list);
324 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
326 spin_unlock_bh(&ptype_lock);
329 * dev_remove_pack - remove packet handler
330 * @pt: packet type declaration
332 * Remove a protocol handler that was previously added to the kernel
333 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
334 * from the kernel lists and can be freed or reused once this function
337 * This call sleeps to guarantee that no CPU is looking at the packet
340 void dev_remove_pack(struct packet_type *pt)
342 __dev_remove_pack(pt);
347 /******************************************************************************
349 Device Boot-time Settings Routines
351 *******************************************************************************/
353 /* Boot time configuration table */
354 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
357 * netdev_boot_setup_add - add new setup entry
358 * @name: name of the device
359 * @map: configured settings for the device
361 * Adds new setup entry to the dev_boot_setup list. The function
362 * returns 0 on error and 1 on success. This is a generic routine to
365 int netdev_boot_setup_add(char *name, struct ifmap *map)
367 struct netdev_boot_setup *s;
371 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
372 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
373 memset(s[i].name, 0, sizeof(s[i].name));
374 strcpy(s[i].name, name);
375 memcpy(&s[i].map, map, sizeof(s[i].map));
380 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
384 * netdev_boot_setup_check - check boot time settings
385 * @dev: the netdevice
387 * Check boot time settings for the device.
388 * The found settings are set for the device to be used
389 * later in the device probing.
390 * Returns 0 if no settings found, 1 if they are.
392 int netdev_boot_setup_check(struct net_device *dev)
394 struct netdev_boot_setup *s = dev_boot_setup;
397 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
398 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
399 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
400 dev->irq = s[i].map.irq;
401 dev->base_addr = s[i].map.base_addr;
402 dev->mem_start = s[i].map.mem_start;
403 dev->mem_end = s[i].map.mem_end;
412 * netdev_boot_base - get address from boot time settings
413 * @prefix: prefix for network device
414 * @unit: id for network device
416 * Check boot time settings for the base address of device.
417 * The found settings are set for the device to be used
418 * later in the device probing.
419 * Returns 0 if no settings found.
421 unsigned long netdev_boot_base(const char *prefix, int unit)
423 const struct netdev_boot_setup *s = dev_boot_setup;
427 sprintf(name, "%s%d", prefix, unit);
430 * If device already registered then return base of 1
431 * to indicate not to probe for this interface
433 if (__dev_get_by_name(name))
436 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
437 if (!strcmp(name, s[i].name))
438 return s[i].map.base_addr;
443 * Saves at boot time configured settings for any netdevice.
445 int __init netdev_boot_setup(char *str)
450 str = get_options(str, ARRAY_SIZE(ints), ints);
455 memset(&map, 0, sizeof(map));
459 map.base_addr = ints[2];
461 map.mem_start = ints[3];
463 map.mem_end = ints[4];
465 /* Add new entry to the list */
466 return netdev_boot_setup_add(str, &map);
469 __setup("netdev=", netdev_boot_setup);
471 /*******************************************************************************
473 Device Interface Subroutines
475 *******************************************************************************/
478 * __dev_get_by_name - find a device by its name
479 * @name: name to find
481 * Find an interface by name. Must be called under RTNL semaphore
482 * or @dev_base_lock. If the name is found a pointer to the device
483 * is returned. If the name is not found then %NULL is returned. The
484 * reference counters are not incremented so the caller must be
485 * careful with locks.
488 struct net_device *__dev_get_by_name(const char *name)
490 struct hlist_node *p;
492 hlist_for_each(p, dev_name_hash(name)) {
493 struct net_device *dev
494 = hlist_entry(p, struct net_device, name_hlist);
495 if (!strncmp(dev->name, name, IFNAMSIZ))
502 * dev_get_by_name - find a device by its name
503 * @name: name to find
505 * Find an interface by name. This can be called from any
506 * context and does its own locking. The returned handle has
507 * the usage count incremented and the caller must use dev_put() to
508 * release it when it is no longer needed. %NULL is returned if no
509 * matching device is found.
512 struct net_device *dev_get_by_name(const char *name)
514 struct net_device *dev;
516 read_lock(&dev_base_lock);
517 dev = __dev_get_by_name(name);
520 read_unlock(&dev_base_lock);
525 * __dev_get_by_index - find a device by its ifindex
526 * @ifindex: index of device
528 * Search for an interface by index. Returns %NULL if the device
529 * is not found or a pointer to the device. The device has not
530 * had its reference counter increased so the caller must be careful
531 * about locking. The caller must hold either the RTNL semaphore
535 struct net_device *__dev_get_by_index(int ifindex)
537 struct hlist_node *p;
539 hlist_for_each(p, dev_index_hash(ifindex)) {
540 struct net_device *dev
541 = hlist_entry(p, struct net_device, index_hlist);
542 if (dev->ifindex == ifindex)
550 * dev_get_by_index - find a device by its ifindex
551 * @ifindex: index of device
553 * Search for an interface by index. Returns NULL if the device
554 * is not found or a pointer to the device. The device returned has
555 * had a reference added and the pointer is safe until the user calls
556 * dev_put to indicate they have finished with it.
559 struct net_device *dev_get_by_index(int ifindex)
561 struct net_device *dev;
563 read_lock(&dev_base_lock);
564 dev = __dev_get_by_index(ifindex);
567 read_unlock(&dev_base_lock);
572 * dev_getbyhwaddr - find a device by its hardware address
573 * @type: media type of device
574 * @ha: hardware address
576 * Search for an interface by MAC address. Returns NULL if the device
577 * is not found or a pointer to the device. The caller must hold the
578 * rtnl semaphore. The returned device has not had its ref count increased
579 * and the caller must therefore be careful about locking
582 * If the API was consistent this would be __dev_get_by_hwaddr
585 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
587 struct net_device *dev;
591 for (dev = dev_base; dev; dev = dev->next)
592 if (dev->type == type &&
593 !memcmp(dev->dev_addr, ha, dev->addr_len))
598 struct net_device *dev_getfirstbyhwtype(unsigned short type)
600 struct net_device *dev;
603 for (dev = dev_base; dev; dev = dev->next) {
604 if (dev->type == type) {
613 EXPORT_SYMBOL(dev_getfirstbyhwtype);
616 * dev_get_by_flags - find any device with given flags
617 * @if_flags: IFF_* values
618 * @mask: bitmask of bits in if_flags to check
620 * Search for any interface with the given flags. Returns NULL if a device
621 * is not found or a pointer to the device. The device returned has
622 * had a reference added and the pointer is safe until the user calls
623 * dev_put to indicate they have finished with it.
626 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
628 struct net_device *dev;
630 read_lock(&dev_base_lock);
631 for (dev = dev_base; dev != NULL; dev = dev->next) {
632 if (((dev->flags ^ if_flags) & mask) == 0) {
637 read_unlock(&dev_base_lock);
642 * dev_valid_name - check if name is okay for network device
645 * Network device names need to be valid file names to
646 * to allow sysfs to work
648 int dev_valid_name(const char *name)
650 return !(*name == '\0'
651 || !strcmp(name, ".")
652 || !strcmp(name, "..")
653 || strchr(name, '/'));
657 * dev_alloc_name - allocate a name for a device
659 * @name: name format string
661 * Passed a format string - eg "lt%d" it will try and find a suitable
662 * id. Not efficient for many devices, not called a lot. The caller
663 * must hold the dev_base or rtnl lock while allocating the name and
664 * adding the device in order to avoid duplicates. Returns the number
665 * of the unit assigned or a negative errno code.
668 int dev_alloc_name(struct net_device *dev, const char *name)
673 const int max_netdevices = 8*PAGE_SIZE;
675 struct net_device *d;
677 p = strnchr(name, IFNAMSIZ-1, '%');
680 * Verify the string as this thing may have come from
681 * the user. There must be either one "%d" and no other "%"
684 if (p[1] != 'd' || strchr(p + 2, '%'))
687 /* Use one page as a bit array of possible slots */
688 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
692 for (d = dev_base; d; d = d->next) {
693 if (!sscanf(d->name, name, &i))
695 if (i < 0 || i >= max_netdevices)
698 /* avoid cases where sscanf is not exact inverse of printf */
699 snprintf(buf, sizeof(buf), name, i);
700 if (!strncmp(buf, d->name, IFNAMSIZ))
704 i = find_first_zero_bit(inuse, max_netdevices);
705 free_page((unsigned long) inuse);
708 snprintf(buf, sizeof(buf), name, i);
709 if (!__dev_get_by_name(buf)) {
710 strlcpy(dev->name, buf, IFNAMSIZ);
714 /* It is possible to run out of possible slots
715 * when the name is long and there isn't enough space left
716 * for the digits, or if all bits are used.
723 * dev_change_name - change name of a device
725 * @newname: name (or format string) must be at least IFNAMSIZ
727 * Change name of a device, can pass format strings "eth%d".
730 int dev_change_name(struct net_device *dev, char *newname)
736 if (dev->flags & IFF_UP)
739 if (!dev_valid_name(newname))
742 if (strchr(newname, '%')) {
743 err = dev_alloc_name(dev, newname);
746 strcpy(newname, dev->name);
748 else if (__dev_get_by_name(newname))
751 strlcpy(dev->name, newname, IFNAMSIZ);
753 err = class_device_rename(&dev->class_dev, dev->name);
755 hlist_del(&dev->name_hlist);
756 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
757 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
764 * netdev_state_change - device changes state
765 * @dev: device to cause notification
767 * Called to indicate a device has changed state. This function calls
768 * the notifier chains for netdev_chain and sends a NEWLINK message
769 * to the routing socket.
771 void netdev_state_change(struct net_device *dev)
773 if (dev->flags & IFF_UP) {
774 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
775 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
780 * dev_load - load a network module
781 * @name: name of interface
783 * If a network interface is not present and the process has suitable
784 * privileges this function loads the module. If module loading is not
785 * available in this kernel then it becomes a nop.
788 void dev_load(const char *name)
790 struct net_device *dev;
792 read_lock(&dev_base_lock);
793 dev = __dev_get_by_name(name);
794 read_unlock(&dev_base_lock);
796 if (!dev && capable(CAP_SYS_MODULE))
797 request_module("%s", name);
800 static int default_rebuild_header(struct sk_buff *skb)
802 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
803 skb->dev ? skb->dev->name : "NULL!!!");
810 * dev_open - prepare an interface for use.
811 * @dev: device to open
813 * Takes a device from down to up state. The device's private open
814 * function is invoked and then the multicast lists are loaded. Finally
815 * the device is moved into the up state and a %NETDEV_UP message is
816 * sent to the netdev notifier chain.
818 * Calling this function on an active interface is a nop. On a failure
819 * a negative errno code is returned.
821 int dev_open(struct net_device *dev)
829 if (dev->flags & IFF_UP)
833 * Is it even present?
835 if (!netif_device_present(dev))
839 * Call device private open method
841 set_bit(__LINK_STATE_START, &dev->state);
843 ret = dev->open(dev);
845 clear_bit(__LINK_STATE_START, &dev->state);
849 * If it went open OK then:
856 dev->flags |= IFF_UP;
859 * Initialize multicasting status
864 * Wakeup transmit queue engine
869 * ... and announce new interface.
871 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
877 * dev_close - shutdown an interface.
878 * @dev: device to shutdown
880 * This function moves an active device into down state. A
881 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
882 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
885 int dev_close(struct net_device *dev)
887 if (!(dev->flags & IFF_UP))
891 * Tell people we are going down, so that they can
892 * prepare to death, when device is still operating.
894 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
898 clear_bit(__LINK_STATE_START, &dev->state);
900 /* Synchronize to scheduled poll. We cannot touch poll list,
901 * it can be even on different cpu. So just clear netif_running(),
902 * and wait when poll really will happen. Actually, the best place
903 * for this is inside dev->stop() after device stopped its irq
904 * engine, but this requires more changes in devices. */
906 smp_mb__after_clear_bit(); /* Commit netif_running(). */
907 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
909 current->state = TASK_INTERRUPTIBLE;
914 * Call the device specific close. This cannot fail.
915 * Only if device is UP
917 * We allow it to be called even after a DETACH hot-plug
924 * Device is now down.
927 dev->flags &= ~IFF_UP;
930 * Tell people we are down
932 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
939 * Device change register/unregister. These are not inline or static
940 * as we export them to the world.
944 * register_netdevice_notifier - register a network notifier block
947 * Register a notifier to be called when network device events occur.
948 * The notifier passed is linked into the kernel structures and must
949 * not be reused until it has been unregistered. A negative errno code
950 * is returned on a failure.
952 * When registered all registration and up events are replayed
953 * to the new notifier to allow device to have a race free
954 * view of the network device list.
957 int register_netdevice_notifier(struct notifier_block *nb)
959 struct net_device *dev;
963 err = notifier_chain_register(&netdev_chain, nb);
965 for (dev = dev_base; dev; dev = dev->next) {
966 nb->notifier_call(nb, NETDEV_REGISTER, dev);
968 if (dev->flags & IFF_UP)
969 nb->notifier_call(nb, NETDEV_UP, dev);
977 * unregister_netdevice_notifier - unregister a network notifier block
980 * Unregister a notifier previously registered by
981 * register_netdevice_notifier(). The notifier is unlinked into the
982 * kernel structures and may then be reused. A negative errno code
983 * is returned on a failure.
986 int unregister_netdevice_notifier(struct notifier_block *nb)
988 return notifier_chain_unregister(&netdev_chain, nb);
992 * call_netdevice_notifiers - call all network notifier blocks
993 * @val: value passed unmodified to notifier function
994 * @v: pointer passed unmodified to notifier function
996 * Call all network notifier blocks. Parameters and return value
997 * are as for notifier_call_chain().
1000 int call_netdevice_notifiers(unsigned long val, void *v)
1002 return notifier_call_chain(&netdev_chain, val, v);
1005 /* When > 0 there are consumers of rx skb time stamps */
1006 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1008 void net_enable_timestamp(void)
1010 atomic_inc(&netstamp_needed);
1013 void net_disable_timestamp(void)
1015 atomic_dec(&netstamp_needed);
1018 static inline void net_timestamp(struct timeval *stamp)
1020 if (atomic_read(&netstamp_needed))
1021 do_gettimeofday(stamp);
1029 * Support routine. Sends outgoing frames to any network
1030 * taps currently in use.
1033 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1035 struct packet_type *ptype;
1036 net_timestamp(&skb->stamp);
1039 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1040 /* Never send packets back to the socket
1041 * they originated from - MvS (miquels@drinkel.ow.org)
1043 if ((ptype->dev == dev || !ptype->dev) &&
1044 (ptype->af_packet_priv == NULL ||
1045 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1046 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1050 /* skb->nh should be correctly
1051 set by sender, so that the second statement is
1052 just protection against buggy protocols.
1054 skb2->mac.raw = skb2->data;
1056 if (skb2->nh.raw < skb2->data ||
1057 skb2->nh.raw > skb2->tail) {
1058 if (net_ratelimit())
1059 printk(KERN_CRIT "protocol %04x is "
1061 skb2->protocol, dev->name);
1062 skb2->nh.raw = skb2->data;
1065 skb2->h.raw = skb2->nh.raw;
1066 skb2->pkt_type = PACKET_OUTGOING;
1067 ptype->func(skb2, skb->dev, ptype);
1074 * Invalidate hardware checksum when packet is to be mangled, and
1075 * complete checksum manually on outgoing path.
1077 int skb_checksum_help(struct sk_buff *skb, int inward)
1080 int ret = 0, offset = skb->h.raw - skb->data;
1083 skb->ip_summed = CHECKSUM_NONE;
1087 if (skb_cloned(skb)) {
1088 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1093 if (offset > (int)skb->len)
1095 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1097 offset = skb->tail - skb->h.raw;
1100 if (skb->csum + 2 > offset)
1103 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1104 skb->ip_summed = CHECKSUM_NONE;
1109 #ifdef CONFIG_HIGHMEM
1110 /* Actually, we should eliminate this check as soon as we know, that:
1111 * 1. IOMMU is present and allows to map all the memory.
1112 * 2. No high memory really exists on this machine.
1115 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1119 if (dev->features & NETIF_F_HIGHDMA)
1122 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1123 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1129 #define illegal_highdma(dev, skb) (0)
1132 extern void skb_release_data(struct sk_buff *);
1134 /* Keep head the same: replace data */
1135 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1140 struct skb_shared_info *ninfo;
1141 int headerlen = skb->data - skb->head;
1142 int expand = (skb->tail + skb->data_len) - skb->end;
1144 if (skb_shared(skb))
1150 size = skb->end - skb->head + expand;
1151 size = SKB_DATA_ALIGN(size);
1152 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1156 /* Copy entire thing */
1157 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1161 ninfo = (struct skb_shared_info*)(data + size);
1162 atomic_set(&ninfo->dataref, 1);
1163 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1164 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1165 ninfo->nr_frags = 0;
1166 ninfo->frag_list = NULL;
1168 /* Offset between the two in bytes */
1169 offset = data - skb->head;
1171 /* Free old data. */
1172 skb_release_data(skb);
1175 skb->end = data + size;
1177 /* Set up new pointers */
1178 skb->h.raw += offset;
1179 skb->nh.raw += offset;
1180 skb->mac.raw += offset;
1181 skb->tail += offset;
1182 skb->data += offset;
1184 /* We are no longer a clone, even if we were. */
1187 skb->tail += skb->data_len;
1192 #define HARD_TX_LOCK(dev, cpu) { \
1193 if ((dev->features & NETIF_F_LLTX) == 0) { \
1194 spin_lock(&dev->xmit_lock); \
1195 dev->xmit_lock_owner = cpu; \
1199 #define HARD_TX_UNLOCK(dev) { \
1200 if ((dev->features & NETIF_F_LLTX) == 0) { \
1201 dev->xmit_lock_owner = -1; \
1202 spin_unlock(&dev->xmit_lock); \
1207 * dev_queue_xmit - transmit a buffer
1208 * @skb: buffer to transmit
1210 * Queue a buffer for transmission to a network device. The caller must
1211 * have set the device and priority and built the buffer before calling
1212 * this function. The function can be called from an interrupt.
1214 * A negative errno code is returned on a failure. A success does not
1215 * guarantee the frame will be transmitted as it may be dropped due
1216 * to congestion or traffic shaping.
1219 int dev_queue_xmit(struct sk_buff *skb)
1221 struct net_device *dev = skb->dev;
1225 if (skb_shinfo(skb)->frag_list &&
1226 !(dev->features & NETIF_F_FRAGLIST) &&
1227 __skb_linearize(skb, GFP_ATOMIC))
1230 /* Fragmented skb is linearized if device does not support SG,
1231 * or if at least one of fragments is in highmem and device
1232 * does not support DMA from it.
1234 if (skb_shinfo(skb)->nr_frags &&
1235 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1236 __skb_linearize(skb, GFP_ATOMIC))
1239 /* If packet is not checksummed and device does not support
1240 * checksumming for this protocol, complete checksumming here.
1242 if (skb->ip_summed == CHECKSUM_HW &&
1243 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1244 (!(dev->features & NETIF_F_IP_CSUM) ||
1245 skb->protocol != htons(ETH_P_IP))))
1246 if (skb_checksum_help(skb, 0))
1249 /* Disable soft irqs for various locks below. Also
1250 * stops preemption for RCU.
1254 /* Updates of qdisc are serialized by queue_lock.
1255 * The struct Qdisc which is pointed to by qdisc is now a
1256 * rcu structure - it may be accessed without acquiring
1257 * a lock (but the structure may be stale.) The freeing of the
1258 * qdisc will be deferred until it's known that there are no
1259 * more references to it.
1261 * If the qdisc has an enqueue function, we still need to
1262 * hold the queue_lock before calling it, since queue_lock
1263 * also serializes access to the device queue.
1266 q = rcu_dereference(dev->qdisc);
1267 #ifdef CONFIG_NET_CLS_ACT
1268 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1271 /* Grab device queue */
1272 spin_lock(&dev->queue_lock);
1274 rc = q->enqueue(skb, q);
1278 spin_unlock(&dev->queue_lock);
1279 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1283 /* The device has no queue. Common case for software devices:
1284 loopback, all the sorts of tunnels...
1286 Really, it is unlikely that xmit_lock protection is necessary here.
1287 (f.e. loopback and IP tunnels are clean ignoring statistics
1289 However, it is possible, that they rely on protection
1292 Check this and shot the lock. It is not prone from deadlocks.
1293 Either shot noqueue qdisc, it is even simpler 8)
1295 if (dev->flags & IFF_UP) {
1296 int cpu = smp_processor_id(); /* ok because BHs are off */
1298 if (dev->xmit_lock_owner != cpu) {
1300 HARD_TX_LOCK(dev, cpu);
1302 if (!netif_queue_stopped(dev)) {
1304 dev_queue_xmit_nit(skb, dev);
1307 if (!dev->hard_start_xmit(skb, dev)) {
1308 HARD_TX_UNLOCK(dev);
1312 HARD_TX_UNLOCK(dev);
1313 if (net_ratelimit())
1314 printk(KERN_CRIT "Virtual device %s asks to "
1315 "queue packet!\n", dev->name);
1317 /* Recursion is detected! It is possible,
1319 if (net_ratelimit())
1320 printk(KERN_CRIT "Dead loop on virtual device "
1321 "%s, fix it urgently!\n", dev->name);
1337 /*=======================================================================
1339 =======================================================================*/
1341 int netdev_max_backlog = 300;
1342 int weight_p = 64; /* old backlog weight */
1343 /* These numbers are selected based on intuition and some
1344 * experimentatiom, if you have more scientific way of doing this
1345 * please go ahead and fix things.
1347 int no_cong_thresh = 10;
1352 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1355 static void get_sample_stats(int cpu)
1361 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1362 int blog = sd->input_pkt_queue.qlen;
1363 int avg_blog = sd->avg_blog;
1365 avg_blog = (avg_blog >> 1) + (blog >> 1);
1367 if (avg_blog > mod_cong) {
1368 /* Above moderate congestion levels. */
1369 sd->cng_level = NET_RX_CN_HIGH;
1372 rq = rd % netdev_max_backlog;
1373 if (rq < avg_blog) /* unlucky bastard */
1374 sd->cng_level = NET_RX_DROP;
1376 } else if (avg_blog > lo_cong) {
1377 sd->cng_level = NET_RX_CN_MOD;
1380 rq = rd % netdev_max_backlog;
1381 if (rq < avg_blog) /* unlucky bastard */
1382 sd->cng_level = NET_RX_CN_HIGH;
1384 } else if (avg_blog > no_cong)
1385 sd->cng_level = NET_RX_CN_LOW;
1386 else /* no congestion */
1387 sd->cng_level = NET_RX_SUCCESS;
1389 sd->avg_blog = avg_blog;
1392 #ifdef OFFLINE_SAMPLE
1393 static void sample_queue(unsigned long dummy)
1395 /* 10 ms 0r 1ms -- i don't care -- JHS */
1397 int cpu = smp_processor_id();
1399 get_sample_stats(cpu);
1400 next_tick += jiffies;
1401 mod_timer(&samp_timer, next_tick);
1407 * netif_rx - post buffer to the network code
1408 * @skb: buffer to post
1410 * This function receives a packet from a device driver and queues it for
1411 * the upper (protocol) levels to process. It always succeeds. The buffer
1412 * may be dropped during processing for congestion control or by the
1416 * NET_RX_SUCCESS (no congestion)
1417 * NET_RX_CN_LOW (low congestion)
1418 * NET_RX_CN_MOD (moderate congestion)
1419 * NET_RX_CN_HIGH (high congestion)
1420 * NET_RX_DROP (packet was dropped)
1424 int netif_rx(struct sk_buff *skb)
1427 struct softnet_data *queue;
1428 unsigned long flags;
1430 #ifdef CONFIG_NETPOLL
1431 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1437 if (!skb->stamp.tv_sec)
1438 net_timestamp(&skb->stamp);
1441 * The code is rearranged so that the path is the most
1442 * short when CPU is congested, but is still operating.
1444 local_irq_save(flags);
1445 this_cpu = smp_processor_id();
1446 queue = &__get_cpu_var(softnet_data);
1448 __get_cpu_var(netdev_rx_stat).total++;
1449 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1450 if (queue->input_pkt_queue.qlen) {
1451 if (queue->throttle)
1456 __skb_queue_tail(&queue->input_pkt_queue, skb);
1457 #ifndef OFFLINE_SAMPLE
1458 get_sample_stats(this_cpu);
1460 local_irq_restore(flags);
1461 return queue->cng_level;
1464 if (queue->throttle)
1465 queue->throttle = 0;
1467 netif_rx_schedule(&queue->backlog_dev);
1471 if (!queue->throttle) {
1472 queue->throttle = 1;
1473 __get_cpu_var(netdev_rx_stat).throttled++;
1477 __get_cpu_var(netdev_rx_stat).dropped++;
1478 local_irq_restore(flags);
1484 int netif_rx_ni(struct sk_buff *skb)
1489 err = netif_rx(skb);
1490 if (softirq_pending(smp_processor_id()))
1497 EXPORT_SYMBOL(netif_rx_ni);
1499 static __inline__ void skb_bond(struct sk_buff *skb)
1501 struct net_device *dev = skb->dev;
1504 skb->real_dev = skb->dev;
1505 skb->dev = dev->master;
1509 static void net_tx_action(struct softirq_action *h)
1511 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1513 if (sd->completion_queue) {
1514 struct sk_buff *clist;
1516 local_irq_disable();
1517 clist = sd->completion_queue;
1518 sd->completion_queue = NULL;
1522 struct sk_buff *skb = clist;
1523 clist = clist->next;
1525 BUG_TRAP(!atomic_read(&skb->users));
1530 if (sd->output_queue) {
1531 struct net_device *head;
1533 local_irq_disable();
1534 head = sd->output_queue;
1535 sd->output_queue = NULL;
1539 struct net_device *dev = head;
1540 head = head->next_sched;
1542 smp_mb__before_clear_bit();
1543 clear_bit(__LINK_STATE_SCHED, &dev->state);
1545 if (spin_trylock(&dev->queue_lock)) {
1547 spin_unlock(&dev->queue_lock);
1549 netif_schedule(dev);
1555 static __inline__ int deliver_skb(struct sk_buff *skb,
1556 struct packet_type *pt_prev)
1558 atomic_inc(&skb->users);
1559 return pt_prev->func(skb, skb->dev, pt_prev);
1562 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1563 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1565 static __inline__ int handle_bridge(struct sk_buff **pskb,
1566 struct packet_type **pt_prev, int *ret)
1568 struct net_bridge_port *port;
1570 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1571 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1575 *ret = deliver_skb(*pskb, *pt_prev);
1579 return br_handle_frame_hook(port, pskb);
1582 #define handle_bridge(skb, pt_prev, ret) (0)
1585 #ifdef CONFIG_NET_CLS_ACT
1586 /* TODO: Maybe we should just force sch_ingress to be compiled in
1587 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1588 * a compare and 2 stores extra right now if we dont have it on
1589 * but have CONFIG_NET_CLS_ACT
1590 * NOTE: This doesnt stop any functionality; if you dont have
1591 * the ingress scheduler, you just cant add policies on ingress.
1594 int ing_filter(struct sk_buff *skb)
1597 struct net_device *dev = skb->dev;
1598 int result = TC_ACT_OK;
1600 if (dev->qdisc_ingress) {
1601 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1602 if (MAX_RED_LOOP < ttl++) {
1603 printk("Redir loop detected Dropping packet (%s->%s)\n",
1604 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1608 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1610 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1611 if (NULL == skb->input_dev) {
1612 skb->input_dev = skb->dev;
1613 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1615 spin_lock(&dev->ingress_lock);
1616 if ((q = dev->qdisc_ingress) != NULL)
1617 result = q->enqueue(skb, q);
1618 spin_unlock(&dev->ingress_lock);
1626 int netif_receive_skb(struct sk_buff *skb)
1628 struct packet_type *ptype, *pt_prev;
1629 int ret = NET_RX_DROP;
1630 unsigned short type;
1632 #ifdef CONFIG_NETPOLL
1633 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1639 if (!skb->stamp.tv_sec)
1640 net_timestamp(&skb->stamp);
1644 __get_cpu_var(netdev_rx_stat).total++;
1646 skb->h.raw = skb->nh.raw = skb->data;
1647 skb->mac_len = skb->nh.raw - skb->mac.raw;
1653 #ifdef CONFIG_NET_CLS_ACT
1654 if (skb->tc_verd & TC_NCLS) {
1655 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1660 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1661 if (!ptype->dev || ptype->dev == skb->dev) {
1663 ret = deliver_skb(skb, pt_prev);
1668 #ifdef CONFIG_NET_CLS_ACT
1670 ret = deliver_skb(skb, pt_prev);
1671 pt_prev = NULL; /* noone else should process this after*/
1673 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1676 ret = ing_filter(skb);
1678 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1687 handle_diverter(skb);
1689 if (handle_bridge(&skb, &pt_prev, &ret))
1692 type = skb->protocol;
1693 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1694 if (ptype->type == type &&
1695 (!ptype->dev || ptype->dev == skb->dev)) {
1697 ret = deliver_skb(skb, pt_prev);
1703 ret = pt_prev->func(skb, skb->dev, pt_prev);
1706 /* Jamal, now you will not able to escape explaining
1707 * me how you were going to use this. :-)
1717 static int process_backlog(struct net_device *backlog_dev, int *budget)
1720 int quota = min(backlog_dev->quota, *budget);
1721 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1722 unsigned long start_time = jiffies;
1725 struct sk_buff *skb;
1726 struct net_device *dev;
1728 local_irq_disable();
1729 skb = __skb_dequeue(&queue->input_pkt_queue);
1736 netif_receive_skb(skb);
1742 if (work >= quota || jiffies - start_time > 1)
1747 backlog_dev->quota -= work;
1752 backlog_dev->quota -= work;
1755 list_del(&backlog_dev->poll_list);
1756 smp_mb__before_clear_bit();
1757 netif_poll_enable(backlog_dev);
1759 if (queue->throttle)
1760 queue->throttle = 0;
1765 static void net_rx_action(struct softirq_action *h)
1767 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1768 unsigned long start_time = jiffies;
1769 int budget = netdev_max_backlog;
1772 local_irq_disable();
1774 while (!list_empty(&queue->poll_list)) {
1775 struct net_device *dev;
1777 if (budget <= 0 || jiffies - start_time > 1)
1782 dev = list_entry(queue->poll_list.next,
1783 struct net_device, poll_list);
1785 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1786 local_irq_disable();
1787 list_del(&dev->poll_list);
1788 list_add_tail(&dev->poll_list, &queue->poll_list);
1790 dev->quota += dev->weight;
1792 dev->quota = dev->weight;
1795 local_irq_disable();
1803 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1804 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1808 static gifconf_func_t * gifconf_list [NPROTO];
1811 * register_gifconf - register a SIOCGIF handler
1812 * @family: Address family
1813 * @gifconf: Function handler
1815 * Register protocol dependent address dumping routines. The handler
1816 * that is passed must not be freed or reused until it has been replaced
1817 * by another handler.
1819 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1821 if (family >= NPROTO)
1823 gifconf_list[family] = gifconf;
1829 * Map an interface index to its name (SIOCGIFNAME)
1833 * We need this ioctl for efficient implementation of the
1834 * if_indextoname() function required by the IPv6 API. Without
1835 * it, we would have to search all the interfaces to find a
1839 static int dev_ifname(struct ifreq __user *arg)
1841 struct net_device *dev;
1845 * Fetch the caller's info block.
1848 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1851 read_lock(&dev_base_lock);
1852 dev = __dev_get_by_index(ifr.ifr_ifindex);
1854 read_unlock(&dev_base_lock);
1858 strcpy(ifr.ifr_name, dev->name);
1859 read_unlock(&dev_base_lock);
1861 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1867 * Perform a SIOCGIFCONF call. This structure will change
1868 * size eventually, and there is nothing I can do about it.
1869 * Thus we will need a 'compatibility mode'.
1872 static int dev_ifconf(char __user *arg)
1875 struct net_device *dev;
1882 * Fetch the caller's info block.
1885 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1892 * Loop over the interfaces, and write an info block for each.
1896 for (dev = dev_base; dev; dev = dev->next) {
1897 for (i = 0; i < NPROTO; i++) {
1898 if (gifconf_list[i]) {
1901 done = gifconf_list[i](dev, NULL, 0);
1903 done = gifconf_list[i](dev, pos + total,
1913 * All done. Write the updated control block back to the caller.
1915 ifc.ifc_len = total;
1918 * Both BSD and Solaris return 0 here, so we do too.
1920 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1923 #ifdef CONFIG_PROC_FS
1925 * This is invoked by the /proc filesystem handler to display a device
1928 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1930 struct net_device *dev;
1933 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1935 return i == pos ? dev : NULL;
1938 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1940 read_lock(&dev_base_lock);
1941 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1944 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1947 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1950 void dev_seq_stop(struct seq_file *seq, void *v)
1952 read_unlock(&dev_base_lock);
1955 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1957 if (dev->get_stats) {
1958 struct net_device_stats *stats = dev->get_stats(dev);
1960 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1961 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1962 dev->name, stats->rx_bytes, stats->rx_packets,
1964 stats->rx_dropped + stats->rx_missed_errors,
1965 stats->rx_fifo_errors,
1966 stats->rx_length_errors + stats->rx_over_errors +
1967 stats->rx_crc_errors + stats->rx_frame_errors,
1968 stats->rx_compressed, stats->multicast,
1969 stats->tx_bytes, stats->tx_packets,
1970 stats->tx_errors, stats->tx_dropped,
1971 stats->tx_fifo_errors, stats->collisions,
1972 stats->tx_carrier_errors +
1973 stats->tx_aborted_errors +
1974 stats->tx_window_errors +
1975 stats->tx_heartbeat_errors,
1976 stats->tx_compressed);
1978 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1982 * Called from the PROCfs module. This now uses the new arbitrary sized
1983 * /proc/net interface to create /proc/net/dev
1985 static int dev_seq_show(struct seq_file *seq, void *v)
1987 if (v == SEQ_START_TOKEN)
1988 seq_puts(seq, "Inter-| Receive "
1990 " face |bytes packets errs drop fifo frame "
1991 "compressed multicast|bytes packets errs "
1992 "drop fifo colls carrier compressed\n");
1994 dev_seq_printf_stats(seq, v);
1998 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2000 struct netif_rx_stats *rc = NULL;
2002 while (*pos < NR_CPUS)
2003 if (cpu_online(*pos)) {
2004 rc = &per_cpu(netdev_rx_stat, *pos);
2011 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2013 return softnet_get_online(pos);
2016 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2019 return softnet_get_online(pos);
2022 static void softnet_seq_stop(struct seq_file *seq, void *v)
2026 static int softnet_seq_show(struct seq_file *seq, void *v)
2028 struct netif_rx_stats *s = v;
2030 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2031 s->total, s->dropped, s->time_squeeze, s->throttled,
2032 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2033 s->fastroute_deferred_out,
2035 s->fastroute_latency_reduction
2043 static struct seq_operations dev_seq_ops = {
2044 .start = dev_seq_start,
2045 .next = dev_seq_next,
2046 .stop = dev_seq_stop,
2047 .show = dev_seq_show,
2050 static int dev_seq_open(struct inode *inode, struct file *file)
2052 return seq_open(file, &dev_seq_ops);
2055 static struct file_operations dev_seq_fops = {
2056 .owner = THIS_MODULE,
2057 .open = dev_seq_open,
2059 .llseek = seq_lseek,
2060 .release = seq_release,
2063 static struct seq_operations softnet_seq_ops = {
2064 .start = softnet_seq_start,
2065 .next = softnet_seq_next,
2066 .stop = softnet_seq_stop,
2067 .show = softnet_seq_show,
2070 static int softnet_seq_open(struct inode *inode, struct file *file)
2072 return seq_open(file, &softnet_seq_ops);
2075 static struct file_operations softnet_seq_fops = {
2076 .owner = THIS_MODULE,
2077 .open = softnet_seq_open,
2079 .llseek = seq_lseek,
2080 .release = seq_release,
2084 extern int wireless_proc_init(void);
2086 #define wireless_proc_init() 0
2089 static int __init dev_proc_init(void)
2093 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2095 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2097 if (wireless_proc_init())
2103 proc_net_remove("softnet_stat");
2105 proc_net_remove("dev");
2109 #define dev_proc_init() 0
2110 #endif /* CONFIG_PROC_FS */
2114 * netdev_set_master - set up master/slave pair
2115 * @slave: slave device
2116 * @master: new master device
2118 * Changes the master device of the slave. Pass %NULL to break the
2119 * bonding. The caller must hold the RTNL semaphore. On a failure
2120 * a negative errno code is returned. On success the reference counts
2121 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2122 * function returns zero.
2124 int netdev_set_master(struct net_device *slave, struct net_device *master)
2126 struct net_device *old = slave->master;
2136 slave->master = master;
2144 slave->flags |= IFF_SLAVE;
2146 slave->flags &= ~IFF_SLAVE;
2148 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2153 * dev_set_promiscuity - update promiscuity count on a device
2157 * Add or remove promsicuity from a device. While the count in the device
2158 * remains above zero the interface remains promiscuous. Once it hits zero
2159 * the device reverts back to normal filtering operation. A negative inc
2160 * value is used to drop promiscuity on the device.
2162 void dev_set_promiscuity(struct net_device *dev, int inc)
2164 unsigned short old_flags = dev->flags;
2166 dev->flags |= IFF_PROMISC;
2167 if ((dev->promiscuity += inc) == 0)
2168 dev->flags &= ~IFF_PROMISC;
2169 if (dev->flags ^ old_flags) {
2171 printk(KERN_INFO "device %s %s promiscuous mode\n",
2172 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2178 * dev_set_allmulti - update allmulti count on a device
2182 * Add or remove reception of all multicast frames to a device. While the
2183 * count in the device remains above zero the interface remains listening
2184 * to all interfaces. Once it hits zero the device reverts back to normal
2185 * filtering operation. A negative @inc value is used to drop the counter
2186 * when releasing a resource needing all multicasts.
2189 void dev_set_allmulti(struct net_device *dev, int inc)
2191 unsigned short old_flags = dev->flags;
2193 dev->flags |= IFF_ALLMULTI;
2194 if ((dev->allmulti += inc) == 0)
2195 dev->flags &= ~IFF_ALLMULTI;
2196 if (dev->flags ^ old_flags)
2200 unsigned dev_get_flags(const struct net_device *dev)
2204 flags = (dev->flags & ~(IFF_PROMISC |
2207 (dev->gflags & (IFF_PROMISC |
2210 if (netif_running(dev) && netif_carrier_ok(dev))
2211 flags |= IFF_RUNNING;
2216 int dev_change_flags(struct net_device *dev, unsigned flags)
2219 int old_flags = dev->flags;
2222 * Set the flags on our device.
2225 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2226 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2228 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2232 * Load in the correct multicast list now the flags have changed.
2238 * Have we downed the interface. We handle IFF_UP ourselves
2239 * according to user attempts to set it, rather than blindly
2244 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2245 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2251 if (dev->flags & IFF_UP &&
2252 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2254 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2256 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2257 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2258 dev->gflags ^= IFF_PROMISC;
2259 dev_set_promiscuity(dev, inc);
2262 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2263 is important. Some (broken) drivers set IFF_PROMISC, when
2264 IFF_ALLMULTI is requested not asking us and not reporting.
2266 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2267 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2268 dev->gflags ^= IFF_ALLMULTI;
2269 dev_set_allmulti(dev, inc);
2272 if (old_flags ^ dev->flags)
2273 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2278 int dev_set_mtu(struct net_device *dev, int new_mtu)
2282 if (new_mtu == dev->mtu)
2285 /* MTU must be positive. */
2289 if (!netif_device_present(dev))
2293 if (dev->change_mtu)
2294 err = dev->change_mtu(dev, new_mtu);
2297 if (!err && dev->flags & IFF_UP)
2298 notifier_call_chain(&netdev_chain,
2299 NETDEV_CHANGEMTU, dev);
2305 * Perform the SIOCxIFxxx calls.
2307 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2310 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2316 case SIOCGIFFLAGS: /* Get interface flags */
2317 ifr->ifr_flags = dev_get_flags(dev);
2320 case SIOCSIFFLAGS: /* Set interface flags */
2321 return dev_change_flags(dev, ifr->ifr_flags);
2323 case SIOCGIFMETRIC: /* Get the metric on the interface
2324 (currently unused) */
2325 ifr->ifr_metric = 0;
2328 case SIOCSIFMETRIC: /* Set the metric on the interface
2329 (currently unused) */
2332 case SIOCGIFMTU: /* Get the MTU of a device */
2333 ifr->ifr_mtu = dev->mtu;
2336 case SIOCSIFMTU: /* Set the MTU of a device */
2337 return dev_set_mtu(dev, ifr->ifr_mtu);
2341 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2343 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2344 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2345 ifr->ifr_hwaddr.sa_family = dev->type;
2349 if (!dev->set_mac_address)
2351 if (ifr->ifr_hwaddr.sa_family != dev->type)
2353 if (!netif_device_present(dev))
2355 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2357 notifier_call_chain(&netdev_chain,
2358 NETDEV_CHANGEADDR, dev);
2361 case SIOCSIFHWBROADCAST:
2362 if (ifr->ifr_hwaddr.sa_family != dev->type)
2364 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2365 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2366 notifier_call_chain(&netdev_chain,
2367 NETDEV_CHANGEADDR, dev);
2371 ifr->ifr_map.mem_start = dev->mem_start;
2372 ifr->ifr_map.mem_end = dev->mem_end;
2373 ifr->ifr_map.base_addr = dev->base_addr;
2374 ifr->ifr_map.irq = dev->irq;
2375 ifr->ifr_map.dma = dev->dma;
2376 ifr->ifr_map.port = dev->if_port;
2380 if (dev->set_config) {
2381 if (!netif_device_present(dev))
2383 return dev->set_config(dev, &ifr->ifr_map);
2388 if (!dev->set_multicast_list ||
2389 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2391 if (!netif_device_present(dev))
2393 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2397 if (!dev->set_multicast_list ||
2398 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2400 if (!netif_device_present(dev))
2402 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2406 ifr->ifr_ifindex = dev->ifindex;
2410 ifr->ifr_qlen = dev->tx_queue_len;
2414 if (ifr->ifr_qlen < 0)
2416 dev->tx_queue_len = ifr->ifr_qlen;
2420 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2421 return dev_change_name(dev, ifr->ifr_newname);
2424 * Unknown or private ioctl
2428 if ((cmd >= SIOCDEVPRIVATE &&
2429 cmd <= SIOCDEVPRIVATE + 15) ||
2430 cmd == SIOCBONDENSLAVE ||
2431 cmd == SIOCBONDRELEASE ||
2432 cmd == SIOCBONDSETHWADDR ||
2433 cmd == SIOCBONDSLAVEINFOQUERY ||
2434 cmd == SIOCBONDINFOQUERY ||
2435 cmd == SIOCBONDCHANGEACTIVE ||
2436 cmd == SIOCGMIIPHY ||
2437 cmd == SIOCGMIIREG ||
2438 cmd == SIOCSMIIREG ||
2439 cmd == SIOCBRADDIF ||
2440 cmd == SIOCBRDELIF ||
2441 cmd == SIOCWANDEV) {
2443 if (dev->do_ioctl) {
2444 if (netif_device_present(dev))
2445 err = dev->do_ioctl(dev, ifr,
2458 * This function handles all "interface"-type I/O control requests. The actual
2459 * 'doing' part of this is dev_ifsioc above.
2463 * dev_ioctl - network device ioctl
2464 * @cmd: command to issue
2465 * @arg: pointer to a struct ifreq in user space
2467 * Issue ioctl functions to devices. This is normally called by the
2468 * user space syscall interfaces but can sometimes be useful for
2469 * other purposes. The return value is the return from the syscall if
2470 * positive or a negative errno code on error.
2473 int dev_ioctl(unsigned int cmd, void __user *arg)
2479 /* One special case: SIOCGIFCONF takes ifconf argument
2480 and requires shared lock, because it sleeps writing
2484 if (cmd == SIOCGIFCONF) {
2486 ret = dev_ifconf((char __user *) arg);
2490 if (cmd == SIOCGIFNAME)
2491 return dev_ifname((struct ifreq __user *)arg);
2493 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2496 ifr.ifr_name[IFNAMSIZ-1] = 0;
2498 colon = strchr(ifr.ifr_name, ':');
2503 * See which interface the caller is talking about.
2508 * These ioctl calls:
2509 * - can be done by all.
2510 * - atomic and do not require locking.
2521 dev_load(ifr.ifr_name);
2522 read_lock(&dev_base_lock);
2523 ret = dev_ifsioc(&ifr, cmd);
2524 read_unlock(&dev_base_lock);
2528 if (copy_to_user(arg, &ifr,
2529 sizeof(struct ifreq)))
2535 dev_load(ifr.ifr_name);
2537 ret = dev_ethtool(&ifr);
2542 if (copy_to_user(arg, &ifr,
2543 sizeof(struct ifreq)))
2549 * These ioctl calls:
2550 * - require superuser power.
2551 * - require strict serialization.
2557 if (!capable(CAP_NET_ADMIN))
2559 dev_load(ifr.ifr_name);
2561 ret = dev_ifsioc(&ifr, cmd);
2566 if (copy_to_user(arg, &ifr,
2567 sizeof(struct ifreq)))
2573 * These ioctl calls:
2574 * - require superuser power.
2575 * - require strict serialization.
2576 * - do not return a value
2586 case SIOCSIFHWBROADCAST:
2589 case SIOCBONDENSLAVE:
2590 case SIOCBONDRELEASE:
2591 case SIOCBONDSETHWADDR:
2592 case SIOCBONDSLAVEINFOQUERY:
2593 case SIOCBONDINFOQUERY:
2594 case SIOCBONDCHANGEACTIVE:
2597 if (!capable(CAP_NET_ADMIN))
2599 dev_load(ifr.ifr_name);
2601 ret = dev_ifsioc(&ifr, cmd);
2606 /* Get the per device memory space. We can add this but
2607 * currently do not support it */
2609 /* Set the per device memory buffer space.
2610 * Not applicable in our case */
2615 * Unknown or private ioctl.
2618 if (cmd == SIOCWANDEV ||
2619 (cmd >= SIOCDEVPRIVATE &&
2620 cmd <= SIOCDEVPRIVATE + 15)) {
2621 dev_load(ifr.ifr_name);
2623 ret = dev_ifsioc(&ifr, cmd);
2625 if (!ret && copy_to_user(arg, &ifr,
2626 sizeof(struct ifreq)))
2631 /* Take care of Wireless Extensions */
2632 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2633 /* If command is `set a parameter', or
2634 * `get the encoding parameters', check if
2635 * the user has the right to do it */
2636 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2637 if (!capable(CAP_NET_ADMIN))
2640 dev_load(ifr.ifr_name);
2642 /* Follow me in net/core/wireless.c */
2643 ret = wireless_process_ioctl(&ifr, cmd);
2645 if (IW_IS_GET(cmd) &&
2646 copy_to_user(arg, &ifr,
2647 sizeof(struct ifreq)))
2651 #endif /* WIRELESS_EXT */
2658 * dev_new_index - allocate an ifindex
2660 * Returns a suitable unique value for a new device interface
2661 * number. The caller must hold the rtnl semaphore or the
2662 * dev_base_lock to be sure it remains unique.
2664 static int dev_new_index(void)
2670 if (!__dev_get_by_index(ifindex))
2675 static int dev_boot_phase = 1;
2677 /* Delayed registration/unregisteration */
2678 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2679 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2681 static inline void net_set_todo(struct net_device *dev)
2683 spin_lock(&net_todo_list_lock);
2684 list_add_tail(&dev->todo_list, &net_todo_list);
2685 spin_unlock(&net_todo_list_lock);
2689 * register_netdevice - register a network device
2690 * @dev: device to register
2692 * Take a completed network device structure and add it to the kernel
2693 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2694 * chain. 0 is returned on success. A negative errno code is returned
2695 * on a failure to set up the device, or if the name is a duplicate.
2697 * Callers must hold the rtnl semaphore. See the comment at the
2698 * end of Space.c for details about the locking. You may want
2699 * register_netdev() instead of this.
2702 * The locking appears insufficient to guarantee two parallel registers
2703 * will not get the same name.
2706 int register_netdevice(struct net_device *dev)
2708 struct hlist_head *head;
2709 struct hlist_node *p;
2712 BUG_ON(dev_boot_phase);
2715 /* When net_device's are persistent, this will be fatal. */
2716 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2718 spin_lock_init(&dev->queue_lock);
2719 spin_lock_init(&dev->xmit_lock);
2720 dev->xmit_lock_owner = -1;
2721 #ifdef CONFIG_NET_CLS_ACT
2722 spin_lock_init(&dev->ingress_lock);
2725 ret = alloc_divert_blk(dev);
2731 /* Init, if this function is available */
2733 ret = dev->init(dev);
2741 if (!dev_valid_name(dev->name)) {
2746 dev->ifindex = dev_new_index();
2747 if (dev->iflink == -1)
2748 dev->iflink = dev->ifindex;
2750 /* Check for existence of name */
2751 head = dev_name_hash(dev->name);
2752 hlist_for_each(p, head) {
2753 struct net_device *d
2754 = hlist_entry(p, struct net_device, name_hlist);
2755 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2761 /* Fix illegal SG+CSUM combinations. */
2762 if ((dev->features & NETIF_F_SG) &&
2763 !(dev->features & (NETIF_F_IP_CSUM |
2765 NETIF_F_HW_CSUM))) {
2766 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2768 dev->features &= ~NETIF_F_SG;
2771 /* TSO requires that SG is present as well. */
2772 if ((dev->features & NETIF_F_TSO) &&
2773 !(dev->features & NETIF_F_SG)) {
2774 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2776 dev->features &= ~NETIF_F_TSO;
2780 * nil rebuild_header routine,
2781 * that should be never called and used as just bug trap.
2784 if (!dev->rebuild_header)
2785 dev->rebuild_header = default_rebuild_header;
2788 * Default initial state at registry is that the
2789 * device is present.
2792 set_bit(__LINK_STATE_PRESENT, &dev->state);
2795 dev_init_scheduler(dev);
2796 write_lock_bh(&dev_base_lock);
2798 dev_tail = &dev->next;
2799 hlist_add_head(&dev->name_hlist, head);
2800 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2802 dev->reg_state = NETREG_REGISTERING;
2803 write_unlock_bh(&dev_base_lock);
2805 /* Notify protocols, that a new device appeared. */
2806 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2808 /* Finish registration after unlock */
2815 free_divert_blk(dev);
2820 * netdev_wait_allrefs - wait until all references are gone.
2822 * This is called when unregistering network devices.
2824 * Any protocol or device that holds a reference should register
2825 * for netdevice notification, and cleanup and put back the
2826 * reference if they receive an UNREGISTER event.
2827 * We can get stuck here if buggy protocols don't correctly
2830 static void netdev_wait_allrefs(struct net_device *dev)
2832 unsigned long rebroadcast_time, warning_time;
2834 rebroadcast_time = warning_time = jiffies;
2835 while (atomic_read(&dev->refcnt) != 0) {
2836 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2839 /* Rebroadcast unregister notification */
2840 notifier_call_chain(&netdev_chain,
2841 NETDEV_UNREGISTER, dev);
2843 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2845 /* We must not have linkwatch events
2846 * pending on unregister. If this
2847 * happens, we simply run the queue
2848 * unscheduled, resulting in a noop
2851 linkwatch_run_queue();
2856 rebroadcast_time = jiffies;
2859 current->state = TASK_INTERRUPTIBLE;
2860 schedule_timeout(HZ / 4);
2862 if (time_after(jiffies, warning_time + 10 * HZ)) {
2863 printk(KERN_EMERG "unregister_netdevice: "
2864 "waiting for %s to become free. Usage "
2866 dev->name, atomic_read(&dev->refcnt));
2867 warning_time = jiffies;
2876 * register_netdevice(x1);
2877 * register_netdevice(x2);
2879 * unregister_netdevice(y1);
2880 * unregister_netdevice(y2);
2886 * We are invoked by rtnl_unlock() after it drops the semaphore.
2887 * This allows us to deal with problems:
2888 * 1) We can create/delete sysfs objects which invoke hotplug
2889 * without deadlocking with linkwatch via keventd.
2890 * 2) Since we run with the RTNL semaphore not held, we can sleep
2891 * safely in order to wait for the netdev refcnt to drop to zero.
2893 static DECLARE_MUTEX(net_todo_run_mutex);
2894 void netdev_run_todo(void)
2896 struct list_head list = LIST_HEAD_INIT(list);
2900 /* Need to guard against multiple cpu's getting out of order. */
2901 down(&net_todo_run_mutex);
2903 /* Not safe to do outside the semaphore. We must not return
2904 * until all unregister events invoked by the local processor
2905 * have been completed (either by this todo run, or one on
2908 if (list_empty(&net_todo_list))
2911 /* Snapshot list, allow later requests */
2912 spin_lock(&net_todo_list_lock);
2913 list_splice_init(&net_todo_list, &list);
2914 spin_unlock(&net_todo_list_lock);
2916 while (!list_empty(&list)) {
2917 struct net_device *dev
2918 = list_entry(list.next, struct net_device, todo_list);
2919 list_del(&dev->todo_list);
2921 switch(dev->reg_state) {
2922 case NETREG_REGISTERING:
2923 err = netdev_register_sysfs(dev);
2925 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2927 dev->reg_state = NETREG_REGISTERED;
2930 case NETREG_UNREGISTERING:
2931 netdev_unregister_sysfs(dev);
2932 dev->reg_state = NETREG_UNREGISTERED;
2934 netdev_wait_allrefs(dev);
2937 BUG_ON(atomic_read(&dev->refcnt));
2938 BUG_TRAP(!dev->ip_ptr);
2939 BUG_TRAP(!dev->ip6_ptr);
2940 BUG_TRAP(!dev->dn_ptr);
2943 /* It must be the very last action,
2944 * after this 'dev' may point to freed up memory.
2946 if (dev->destructor)
2947 dev->destructor(dev);
2951 printk(KERN_ERR "network todo '%s' but state %d\n",
2952 dev->name, dev->reg_state);
2958 up(&net_todo_run_mutex);
2962 * free_netdev - free network device
2965 * This function does the last stage of destroying an allocated device
2966 * interface. The reference to the device object is released.
2967 * If this is the last reference then it will be freed.
2969 void free_netdev(struct net_device *dev)
2972 /* Compatiablity with error handling in drivers */
2973 if (dev->reg_state == NETREG_UNINITIALIZED) {
2974 kfree((char *)dev - dev->padded);
2978 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
2979 dev->reg_state = NETREG_RELEASED;
2981 /* will free via class release */
2982 class_device_put(&dev->class_dev);
2984 kfree((char *)dev - dev->padded);
2988 /* Synchronize with packet receive processing. */
2989 void synchronize_net(void)
2992 synchronize_kernel();
2996 * unregister_netdevice - remove device from the kernel
2999 * This function shuts down a device interface and removes it
3000 * from the kernel tables. On success 0 is returned, on a failure
3001 * a negative errno code is returned.
3003 * Callers must hold the rtnl semaphore. See the comment at the
3004 * end of Space.c for details about the locking. You may want
3005 * unregister_netdev() instead of this.
3008 int unregister_netdevice(struct net_device *dev)
3010 struct net_device *d, **dp;
3012 BUG_ON(dev_boot_phase);
3015 /* Some devices call without registering for initialization unwind. */
3016 if (dev->reg_state == NETREG_UNINITIALIZED) {
3017 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3018 "was registered\n", dev->name, dev);
3022 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3024 /* If device is running, close it first. */
3025 if (dev->flags & IFF_UP)
3028 /* And unlink it from device chain. */
3029 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3031 write_lock_bh(&dev_base_lock);
3032 hlist_del(&dev->name_hlist);
3033 hlist_del(&dev->index_hlist);
3034 if (dev_tail == &dev->next)
3037 write_unlock_bh(&dev_base_lock);
3042 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3047 dev->reg_state = NETREG_UNREGISTERING;
3051 /* Shutdown queueing discipline. */
3055 /* Notify protocols, that we are about to destroy
3056 this device. They should clean all the things.
3058 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3061 * Flush the multicast chain
3063 dev_mc_discard(dev);
3068 /* Notifier chain MUST detach us from master device. */
3069 BUG_TRAP(!dev->master);
3071 free_divert_blk(dev);
3073 /* Finish processing unregister after unlock */
3082 #ifdef CONFIG_HOTPLUG_CPU
3083 static int dev_cpu_callback(struct notifier_block *nfb,
3084 unsigned long action,
3087 struct sk_buff **list_skb;
3088 struct net_device **list_net;
3089 struct sk_buff *skb;
3090 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3091 struct softnet_data *sd, *oldsd;
3093 if (action != CPU_DEAD)
3096 local_irq_disable();
3097 cpu = smp_processor_id();
3098 sd = &per_cpu(softnet_data, cpu);
3099 oldsd = &per_cpu(softnet_data, oldcpu);
3101 /* Find end of our completion_queue. */
3102 list_skb = &sd->completion_queue;
3104 list_skb = &(*list_skb)->next;
3105 /* Append completion queue from offline CPU. */
3106 *list_skb = oldsd->completion_queue;
3107 oldsd->completion_queue = NULL;
3109 /* Find end of our output_queue. */
3110 list_net = &sd->output_queue;
3112 list_net = &(*list_net)->next_sched;
3113 /* Append output queue from offline CPU. */
3114 *list_net = oldsd->output_queue;
3115 oldsd->output_queue = NULL;
3117 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3120 /* Process offline CPU's input_pkt_queue */
3121 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3126 #endif /* CONFIG_HOTPLUG_CPU */
3130 * Initialize the DEV module. At boot time this walks the device list and
3131 * unhooks any devices that fail to initialise (normally hardware not
3132 * present) and leaves us with a valid list of present and active devices.
3137 * This is called single threaded during boot, so no need
3138 * to take the rtnl semaphore.
3140 static int __init net_dev_init(void)
3142 int i, rc = -ENOMEM;
3144 BUG_ON(!dev_boot_phase);
3148 if (dev_proc_init())
3151 if (netdev_sysfs_init())
3154 INIT_LIST_HEAD(&ptype_all);
3155 for (i = 0; i < 16; i++)
3156 INIT_LIST_HEAD(&ptype_base[i]);
3158 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3159 INIT_HLIST_HEAD(&dev_name_head[i]);
3161 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3162 INIT_HLIST_HEAD(&dev_index_head[i]);
3165 * Initialise the packet receive queues.
3168 for (i = 0; i < NR_CPUS; i++) {
3169 struct softnet_data *queue;
3171 queue = &per_cpu(softnet_data, i);
3172 skb_queue_head_init(&queue->input_pkt_queue);
3173 queue->throttle = 0;
3174 queue->cng_level = 0;
3175 queue->avg_blog = 10; /* arbitrary non-zero */
3176 queue->completion_queue = NULL;
3177 INIT_LIST_HEAD(&queue->poll_list);
3178 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3179 queue->backlog_dev.weight = weight_p;
3180 queue->backlog_dev.poll = process_backlog;
3181 atomic_set(&queue->backlog_dev.refcnt, 1);
3184 #ifdef OFFLINE_SAMPLE
3185 samp_timer.expires = jiffies + (10 * HZ);
3186 add_timer(&samp_timer);
3191 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3192 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3194 hotcpu_notifier(dev_cpu_callback, 0);
3202 subsys_initcall(net_dev_init);
3204 EXPORT_SYMBOL(__dev_get_by_index);
3205 EXPORT_SYMBOL(__dev_get_by_name);
3206 EXPORT_SYMBOL(__dev_remove_pack);
3207 EXPORT_SYMBOL(__skb_linearize);
3208 EXPORT_SYMBOL(dev_add_pack);
3209 EXPORT_SYMBOL(dev_alloc_name);
3210 EXPORT_SYMBOL(dev_close);
3211 EXPORT_SYMBOL(dev_get_by_flags);
3212 EXPORT_SYMBOL(dev_get_by_index);
3213 EXPORT_SYMBOL(dev_get_by_name);
3214 EXPORT_SYMBOL(dev_ioctl);
3215 EXPORT_SYMBOL(dev_open);
3216 EXPORT_SYMBOL(dev_queue_xmit);
3217 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
3218 EXPORT_SYMBOL(dev_queue_xmit_nit);
3220 EXPORT_SYMBOL(dev_remove_pack);
3221 EXPORT_SYMBOL(dev_set_allmulti);
3222 EXPORT_SYMBOL(dev_set_promiscuity);
3223 EXPORT_SYMBOL(dev_change_flags);
3224 EXPORT_SYMBOL(dev_change_name);
3225 EXPORT_SYMBOL(dev_set_mtu);
3226 EXPORT_SYMBOL(free_netdev);
3227 EXPORT_SYMBOL(netdev_boot_setup_check);
3228 EXPORT_SYMBOL(netdev_set_master);
3229 EXPORT_SYMBOL(netdev_state_change);
3230 EXPORT_SYMBOL(netif_receive_skb);
3231 EXPORT_SYMBOL(netif_rx);
3232 EXPORT_SYMBOL(register_gifconf);
3233 EXPORT_SYMBOL(register_netdevice);
3234 EXPORT_SYMBOL(register_netdevice_notifier);
3235 EXPORT_SYMBOL(skb_checksum_help);
3236 EXPORT_SYMBOL(synchronize_net);
3237 EXPORT_SYMBOL(unregister_netdevice);
3238 EXPORT_SYMBOL(unregister_netdevice_notifier);
3239 EXPORT_SYMBOL(net_enable_timestamp);
3240 EXPORT_SYMBOL(net_disable_timestamp);
3242 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3243 EXPORT_SYMBOL(br_handle_frame_hook);
3247 EXPORT_SYMBOL(dev_load);
3250 #ifdef CONFIG_NET_CLS_ACT
3251 EXPORT_SYMBOL(ing_filter);
3255 EXPORT_PER_CPU_SYMBOL(softnet_data);