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 <asm/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);
1006 * Support routine. Sends outgoing frames to any network
1007 * taps currently in use.
1010 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1012 struct packet_type *ptype;
1013 net_timestamp(&skb->stamp);
1016 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1017 /* Never send packets back to the socket
1018 * they originated from - MvS (miquels@drinkel.ow.org)
1020 if ((ptype->dev == dev || !ptype->dev) &&
1021 (ptype->af_packet_priv == NULL ||
1022 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1023 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1027 /* skb->nh should be correctly
1028 set by sender, so that the second statement is
1029 just protection against buggy protocols.
1031 skb2->mac.raw = skb2->data;
1033 if (skb2->nh.raw < skb2->data ||
1034 skb2->nh.raw > skb2->tail) {
1035 if (net_ratelimit())
1036 printk(KERN_CRIT "protocol %04x is "
1038 skb2->protocol, dev->name);
1039 skb2->nh.raw = skb2->data;
1042 skb2->h.raw = skb2->nh.raw;
1043 skb2->pkt_type = PACKET_OUTGOING;
1044 ptype->func(skb2, skb->dev, ptype);
1051 * Invalidate hardware checksum when packet is to be mangled, and
1052 * complete checksum manually on outgoing path.
1054 int skb_checksum_help(struct sk_buff **pskb, int inward)
1057 int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1060 (*pskb)->ip_summed = CHECKSUM_NONE;
1064 if (skb_cloned(*pskb)) {
1065 ret = pskb_expand_head(*pskb, 0, 0, GFP_ATOMIC);
1070 if (offset > (int)(*pskb)->len)
1072 csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1074 offset = (*pskb)->tail - (*pskb)->h.raw;
1077 if ((*pskb)->csum + 2 > offset)
1080 *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1081 (*pskb)->ip_summed = CHECKSUM_NONE;
1086 #ifdef CONFIG_HIGHMEM
1087 /* Actually, we should eliminate this check as soon as we know, that:
1088 * 1. IOMMU is present and allows to map all the memory.
1089 * 2. No high memory really exists on this machine.
1092 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1096 if (dev->features & NETIF_F_HIGHDMA)
1099 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1100 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1106 #define illegal_highdma(dev, skb) (0)
1109 extern void skb_release_data(struct sk_buff *);
1111 /* Keep head the same: replace data */
1112 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1117 struct skb_shared_info *ninfo;
1118 int headerlen = skb->data - skb->head;
1119 int expand = (skb->tail + skb->data_len) - skb->end;
1121 if (skb_shared(skb))
1127 size = skb->end - skb->head + expand;
1128 size = SKB_DATA_ALIGN(size);
1129 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1133 /* Copy entire thing */
1134 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1138 ninfo = (struct skb_shared_info*)(data + size);
1139 atomic_set(&ninfo->dataref, 1);
1140 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1141 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1142 ninfo->nr_frags = 0;
1143 ninfo->frag_list = NULL;
1145 /* Offset between the two in bytes */
1146 offset = data - skb->head;
1148 /* Free old data. */
1149 skb_release_data(skb);
1152 skb->end = data + size;
1154 /* Set up new pointers */
1155 skb->h.raw += offset;
1156 skb->nh.raw += offset;
1157 skb->mac.raw += offset;
1158 skb->tail += offset;
1159 skb->data += offset;
1161 /* We are no longer a clone, even if we were. */
1164 skb->tail += skb->data_len;
1169 #define HARD_TX_LOCK(dev, cpu) { \
1170 if ((dev->features & NETIF_F_LLTX) == 0) { \
1171 spin_lock(&dev->xmit_lock); \
1172 dev->xmit_lock_owner = cpu; \
1176 #define HARD_TX_UNLOCK(dev) { \
1177 if ((dev->features & NETIF_F_LLTX) == 0) { \
1178 dev->xmit_lock_owner = -1; \
1179 spin_unlock(&dev->xmit_lock); \
1183 static inline void qdisc_run(struct net_device *dev)
1185 while (!netif_queue_stopped(dev) &&
1186 qdisc_restart(dev)<0)
1191 * dev_queue_xmit - transmit a buffer
1192 * @skb: buffer to transmit
1194 * Queue a buffer for transmission to a network device. The caller must
1195 * have set the device and priority and built the buffer before calling
1196 * this function. The function can be called from an interrupt.
1198 * A negative errno code is returned on a failure. A success does not
1199 * guarantee the frame will be transmitted as it may be dropped due
1200 * to congestion or traffic shaping.
1203 int dev_queue_xmit(struct sk_buff *skb)
1205 struct net_device *dev = skb->dev;
1209 if (skb_shinfo(skb)->frag_list &&
1210 !(dev->features & NETIF_F_FRAGLIST) &&
1211 __skb_linearize(skb, GFP_ATOMIC))
1214 /* Fragmented skb is linearized if device does not support SG,
1215 * or if at least one of fragments is in highmem and device
1216 * does not support DMA from it.
1218 if (skb_shinfo(skb)->nr_frags &&
1219 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1220 __skb_linearize(skb, GFP_ATOMIC))
1223 /* If packet is not checksummed and device does not support
1224 * checksumming for this protocol, complete checksumming here.
1226 if (skb->ip_summed == CHECKSUM_HW &&
1227 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1228 (!(dev->features & NETIF_F_IP_CSUM) ||
1229 skb->protocol != htons(ETH_P_IP))))
1230 if (skb_checksum_help(&skb, 0))
1234 /* Disable soft irqs for various locks below. Also
1235 * stops preemption for RCU.
1239 /* Updates of qdisc are serialized by queue_lock.
1240 * The struct Qdisc which is pointed to by qdisc is now a
1241 * rcu structure - it may be accessed without acquiring
1242 * a lock (but the structure may be stale.) The freeing of the
1243 * qdisc will be deferred until it's known that there are no
1244 * more references to it.
1246 * If the qdisc has an enqueue function, we still need to
1247 * hold the queue_lock before calling it, since queue_lock
1248 * also serializes access to the device queue.
1251 q = rcu_dereference(dev->qdisc);
1252 #ifdef CONFIG_NET_CLS_ACT
1253 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1256 /* Grab device queue */
1257 spin_lock(&dev->queue_lock);
1259 rc = q->enqueue(skb, q);
1263 spin_unlock(&dev->queue_lock);
1264 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1268 /* The device has no queue. Common case for software devices:
1269 loopback, all the sorts of tunnels...
1271 Really, it is unlikely that xmit_lock protection is necessary here.
1272 (f.e. loopback and IP tunnels are clean ignoring statistics
1274 However, it is possible, that they rely on protection
1277 Check this and shot the lock. It is not prone from deadlocks.
1278 Either shot noqueue qdisc, it is even simpler 8)
1280 if (dev->flags & IFF_UP) {
1281 int cpu = smp_processor_id(); /* ok because BHs are off */
1283 if (dev->xmit_lock_owner != cpu) {
1285 HARD_TX_LOCK(dev, cpu);
1287 if (!netif_queue_stopped(dev)) {
1289 dev_queue_xmit_nit(skb, dev);
1292 if (!dev->hard_start_xmit(skb, dev)) {
1293 HARD_TX_UNLOCK(dev);
1297 HARD_TX_UNLOCK(dev);
1298 if (net_ratelimit())
1299 printk(KERN_CRIT "Virtual device %s asks to "
1300 "queue packet!\n", dev->name);
1303 /* Recursion is detected! It is possible,
1305 if (net_ratelimit())
1306 printk(KERN_CRIT "Dead loop on virtual device "
1307 "%s, fix it urgently!\n", dev->name);
1320 /*=======================================================================
1322 =======================================================================*/
1324 int netdev_max_backlog = 300;
1325 int weight_p = 64; /* old backlog weight */
1326 /* These numbers are selected based on intuition and some
1327 * experimentatiom, if you have more scientific way of doing this
1328 * please go ahead and fix things.
1330 int no_cong_thresh = 10;
1335 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1338 static void get_sample_stats(int cpu)
1344 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1345 int blog = sd->input_pkt_queue.qlen;
1346 int avg_blog = sd->avg_blog;
1348 avg_blog = (avg_blog >> 1) + (blog >> 1);
1350 if (avg_blog > mod_cong) {
1351 /* Above moderate congestion levels. */
1352 sd->cng_level = NET_RX_CN_HIGH;
1355 rq = rd % netdev_max_backlog;
1356 if (rq < avg_blog) /* unlucky bastard */
1357 sd->cng_level = NET_RX_DROP;
1359 } else if (avg_blog > lo_cong) {
1360 sd->cng_level = NET_RX_CN_MOD;
1363 rq = rd % netdev_max_backlog;
1364 if (rq < avg_blog) /* unlucky bastard */
1365 sd->cng_level = NET_RX_CN_HIGH;
1367 } else if (avg_blog > no_cong)
1368 sd->cng_level = NET_RX_CN_LOW;
1369 else /* no congestion */
1370 sd->cng_level = NET_RX_SUCCESS;
1372 sd->avg_blog = avg_blog;
1375 #ifdef OFFLINE_SAMPLE
1376 static void sample_queue(unsigned long dummy)
1378 /* 10 ms 0r 1ms -- i don't care -- JHS */
1380 int cpu = smp_processor_id();
1382 get_sample_stats(cpu);
1383 next_tick += jiffies;
1384 mod_timer(&samp_timer, next_tick);
1390 * netif_rx - post buffer to the network code
1391 * @skb: buffer to post
1393 * This function receives a packet from a device driver and queues it for
1394 * the upper (protocol) levels to process. It always succeeds. The buffer
1395 * may be dropped during processing for congestion control or by the
1399 * NET_RX_SUCCESS (no congestion)
1400 * NET_RX_CN_LOW (low congestion)
1401 * NET_RX_CN_MOD (moderate congestion)
1402 * NET_RX_CN_HIGH (high congestion)
1403 * NET_RX_DROP (packet was dropped)
1407 int netif_rx(struct sk_buff *skb)
1410 struct softnet_data *queue;
1411 unsigned long flags;
1413 #ifdef CONFIG_NETPOLL
1414 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1420 if (!skb->stamp.tv_sec)
1421 net_timestamp(&skb->stamp);
1424 * The code is rearranged so that the path is the most
1425 * short when CPU is congested, but is still operating.
1427 local_irq_save(flags);
1428 this_cpu = smp_processor_id();
1429 queue = &__get_cpu_var(softnet_data);
1431 __get_cpu_var(netdev_rx_stat).total++;
1432 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1433 if (queue->input_pkt_queue.qlen) {
1434 if (queue->throttle)
1439 __skb_queue_tail(&queue->input_pkt_queue, skb);
1440 #ifndef OFFLINE_SAMPLE
1441 get_sample_stats(this_cpu);
1443 local_irq_restore(flags);
1444 return queue->cng_level;
1447 if (queue->throttle)
1448 queue->throttle = 0;
1450 netif_rx_schedule(&queue->backlog_dev);
1454 if (!queue->throttle) {
1455 queue->throttle = 1;
1456 __get_cpu_var(netdev_rx_stat).throttled++;
1460 __get_cpu_var(netdev_rx_stat).dropped++;
1461 local_irq_restore(flags);
1467 static __inline__ void skb_bond(struct sk_buff *skb)
1469 struct net_device *dev = skb->dev;
1472 skb->real_dev = skb->dev;
1473 skb->dev = dev->master;
1477 static void net_tx_action(struct softirq_action *h)
1479 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1481 if (sd->completion_queue) {
1482 struct sk_buff *clist;
1484 local_irq_disable();
1485 clist = sd->completion_queue;
1486 sd->completion_queue = NULL;
1490 struct sk_buff *skb = clist;
1491 clist = clist->next;
1493 BUG_TRAP(!atomic_read(&skb->users));
1498 if (sd->output_queue) {
1499 struct net_device *head;
1501 local_irq_disable();
1502 head = sd->output_queue;
1503 sd->output_queue = NULL;
1507 struct net_device *dev = head;
1508 head = head->next_sched;
1510 smp_mb__before_clear_bit();
1511 clear_bit(__LINK_STATE_SCHED, &dev->state);
1513 if (spin_trylock(&dev->queue_lock)) {
1515 spin_unlock(&dev->queue_lock);
1517 netif_schedule(dev);
1523 static __inline__ int deliver_skb(struct sk_buff *skb,
1524 struct packet_type *pt_prev)
1526 atomic_inc(&skb->users);
1527 return pt_prev->func(skb, skb->dev, pt_prev);
1530 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1531 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1533 static __inline__ int handle_bridge(struct sk_buff **pskb,
1534 struct packet_type **pt_prev, int *ret)
1536 struct net_bridge_port *port;
1538 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1539 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1543 *ret = deliver_skb(*pskb, *pt_prev);
1547 return br_handle_frame_hook(port, pskb);
1550 #define handle_bridge(skb, pt_prev, ret) (0)
1553 #ifdef CONFIG_NET_CLS_ACT
1554 /* TODO: Maybe we should just force sch_ingress to be compiled in
1555 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1556 * a compare and 2 stores extra right now if we dont have it on
1557 * but have CONFIG_NET_CLS_ACT
1558 * NOTE: This doesnt stop any functionality; if you dont have
1559 * the ingress scheduler, you just cant add policies on ingress.
1562 int ing_filter(struct sk_buff *skb)
1565 struct net_device *dev = skb->dev;
1566 int result = TC_ACT_OK;
1568 if (dev->qdisc_ingress) {
1569 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1570 if (MAX_RED_LOOP < ttl++) {
1571 printk("Redir loop detected Dropping packet (%s->%s)\n",
1572 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1576 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1578 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1579 if (NULL == skb->input_dev) {
1580 skb->input_dev = skb->dev;
1581 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1583 spin_lock(&dev->ingress_lock);
1584 if ((q = dev->qdisc_ingress) != NULL)
1585 result = q->enqueue(skb, q);
1586 spin_unlock(&dev->ingress_lock);
1594 int netif_receive_skb(struct sk_buff *skb)
1596 struct packet_type *ptype, *pt_prev;
1597 int ret = NET_RX_DROP;
1598 unsigned short type;
1600 #ifdef CONFIG_NETPOLL
1601 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1607 if (!skb->stamp.tv_sec)
1608 net_timestamp(&skb->stamp);
1612 __get_cpu_var(netdev_rx_stat).total++;
1614 skb->h.raw = skb->nh.raw = skb->data;
1615 skb->mac_len = skb->nh.raw - skb->mac.raw;
1621 #ifdef CONFIG_NET_CLS_ACT
1622 if (skb->tc_verd & TC_NCLS) {
1623 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1628 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1629 if (!ptype->dev || ptype->dev == skb->dev) {
1631 ret = deliver_skb(skb, pt_prev);
1636 #ifdef CONFIG_NET_CLS_ACT
1638 ret = deliver_skb(skb, pt_prev);
1639 pt_prev = NULL; /* noone else should process this after*/
1641 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1644 ret = ing_filter(skb);
1646 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1655 handle_diverter(skb);
1657 if (handle_bridge(&skb, &pt_prev, &ret))
1660 type = skb->protocol;
1661 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1662 if (ptype->type == type &&
1663 (!ptype->dev || ptype->dev == skb->dev)) {
1665 ret = deliver_skb(skb, pt_prev);
1671 ret = pt_prev->func(skb, skb->dev, pt_prev);
1674 /* Jamal, now you will not able to escape explaining
1675 * me how you were going to use this. :-)
1685 static int process_backlog(struct net_device *backlog_dev, int *budget)
1688 int quota = min(backlog_dev->quota, *budget);
1689 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1690 unsigned long start_time = jiffies;
1693 struct sk_buff *skb;
1694 struct net_device *dev;
1696 local_irq_disable();
1697 skb = __skb_dequeue(&queue->input_pkt_queue);
1704 netif_receive_skb(skb);
1710 if (work >= quota || jiffies - start_time > 1)
1715 backlog_dev->quota -= work;
1720 backlog_dev->quota -= work;
1723 list_del(&backlog_dev->poll_list);
1724 smp_mb__before_clear_bit();
1725 netif_poll_enable(backlog_dev);
1727 if (queue->throttle)
1728 queue->throttle = 0;
1733 static void net_rx_action(struct softirq_action *h)
1735 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1736 unsigned long start_time = jiffies;
1737 int budget = netdev_max_backlog;
1740 local_irq_disable();
1742 while (!list_empty(&queue->poll_list)) {
1743 struct net_device *dev;
1745 if (budget <= 0 || jiffies - start_time > 1)
1750 dev = list_entry(queue->poll_list.next,
1751 struct net_device, poll_list);
1753 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1754 local_irq_disable();
1755 list_del(&dev->poll_list);
1756 list_add_tail(&dev->poll_list, &queue->poll_list);
1758 dev->quota += dev->weight;
1760 dev->quota = dev->weight;
1763 local_irq_disable();
1771 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1772 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1776 static gifconf_func_t * gifconf_list [NPROTO];
1779 * register_gifconf - register a SIOCGIF handler
1780 * @family: Address family
1781 * @gifconf: Function handler
1783 * Register protocol dependent address dumping routines. The handler
1784 * that is passed must not be freed or reused until it has been replaced
1785 * by another handler.
1787 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1789 if (family >= NPROTO)
1791 gifconf_list[family] = gifconf;
1797 * Map an interface index to its name (SIOCGIFNAME)
1801 * We need this ioctl for efficient implementation of the
1802 * if_indextoname() function required by the IPv6 API. Without
1803 * it, we would have to search all the interfaces to find a
1807 static int dev_ifname(struct ifreq __user *arg)
1809 struct net_device *dev;
1813 * Fetch the caller's info block.
1816 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1819 read_lock(&dev_base_lock);
1820 dev = __dev_get_by_index(ifr.ifr_ifindex);
1822 read_unlock(&dev_base_lock);
1826 strcpy(ifr.ifr_name, dev->name);
1827 read_unlock(&dev_base_lock);
1829 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1835 * Perform a SIOCGIFCONF call. This structure will change
1836 * size eventually, and there is nothing I can do about it.
1837 * Thus we will need a 'compatibility mode'.
1840 static int dev_ifconf(char __user *arg)
1843 struct net_device *dev;
1850 * Fetch the caller's info block.
1853 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1860 * Loop over the interfaces, and write an info block for each.
1864 for (dev = dev_base; dev; dev = dev->next) {
1865 for (i = 0; i < NPROTO; i++) {
1866 if (gifconf_list[i]) {
1869 done = gifconf_list[i](dev, NULL, 0);
1871 done = gifconf_list[i](dev, pos + total,
1881 * All done. Write the updated control block back to the caller.
1883 ifc.ifc_len = total;
1886 * Both BSD and Solaris return 0 here, so we do too.
1888 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1891 #ifdef CONFIG_PROC_FS
1893 * This is invoked by the /proc filesystem handler to display a device
1896 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1898 struct net_device *dev;
1901 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1903 return i == pos ? dev : NULL;
1906 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1908 read_lock(&dev_base_lock);
1909 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1912 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1915 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1918 void dev_seq_stop(struct seq_file *seq, void *v)
1920 read_unlock(&dev_base_lock);
1923 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1925 if (dev->get_stats) {
1926 struct net_device_stats *stats = dev->get_stats(dev);
1928 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1929 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1930 dev->name, stats->rx_bytes, stats->rx_packets,
1932 stats->rx_dropped + stats->rx_missed_errors,
1933 stats->rx_fifo_errors,
1934 stats->rx_length_errors + stats->rx_over_errors +
1935 stats->rx_crc_errors + stats->rx_frame_errors,
1936 stats->rx_compressed, stats->multicast,
1937 stats->tx_bytes, stats->tx_packets,
1938 stats->tx_errors, stats->tx_dropped,
1939 stats->tx_fifo_errors, stats->collisions,
1940 stats->tx_carrier_errors +
1941 stats->tx_aborted_errors +
1942 stats->tx_window_errors +
1943 stats->tx_heartbeat_errors,
1944 stats->tx_compressed);
1946 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1950 * Called from the PROCfs module. This now uses the new arbitrary sized
1951 * /proc/net interface to create /proc/net/dev
1953 static int dev_seq_show(struct seq_file *seq, void *v)
1955 if (v == SEQ_START_TOKEN)
1956 seq_puts(seq, "Inter-| Receive "
1958 " face |bytes packets errs drop fifo frame "
1959 "compressed multicast|bytes packets errs "
1960 "drop fifo colls carrier compressed\n");
1962 dev_seq_printf_stats(seq, v);
1966 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1968 struct netif_rx_stats *rc = NULL;
1970 while (*pos < NR_CPUS)
1971 if (cpu_online(*pos)) {
1972 rc = &per_cpu(netdev_rx_stat, *pos);
1979 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1981 return softnet_get_online(pos);
1984 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1987 return softnet_get_online(pos);
1990 static void softnet_seq_stop(struct seq_file *seq, void *v)
1994 static int softnet_seq_show(struct seq_file *seq, void *v)
1996 struct netif_rx_stats *s = v;
1998 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
1999 s->total, s->dropped, s->time_squeeze, s->throttled,
2000 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2001 s->fastroute_deferred_out,
2003 s->fastroute_latency_reduction
2011 static struct seq_operations dev_seq_ops = {
2012 .start = dev_seq_start,
2013 .next = dev_seq_next,
2014 .stop = dev_seq_stop,
2015 .show = dev_seq_show,
2018 static int dev_seq_open(struct inode *inode, struct file *file)
2020 return seq_open(file, &dev_seq_ops);
2023 static struct file_operations dev_seq_fops = {
2024 .owner = THIS_MODULE,
2025 .open = dev_seq_open,
2027 .llseek = seq_lseek,
2028 .release = seq_release,
2031 static struct seq_operations softnet_seq_ops = {
2032 .start = softnet_seq_start,
2033 .next = softnet_seq_next,
2034 .stop = softnet_seq_stop,
2035 .show = softnet_seq_show,
2038 static int softnet_seq_open(struct inode *inode, struct file *file)
2040 return seq_open(file, &softnet_seq_ops);
2043 static struct file_operations softnet_seq_fops = {
2044 .owner = THIS_MODULE,
2045 .open = softnet_seq_open,
2047 .llseek = seq_lseek,
2048 .release = seq_release,
2052 extern int wireless_proc_init(void);
2054 #define wireless_proc_init() 0
2057 static int __init dev_proc_init(void)
2061 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2063 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2065 if (wireless_proc_init())
2071 proc_net_remove("softnet_stat");
2073 proc_net_remove("dev");
2077 #define dev_proc_init() 0
2078 #endif /* CONFIG_PROC_FS */
2082 * netdev_set_master - set up master/slave pair
2083 * @slave: slave device
2084 * @master: new master device
2086 * Changes the master device of the slave. Pass %NULL to break the
2087 * bonding. The caller must hold the RTNL semaphore. On a failure
2088 * a negative errno code is returned. On success the reference counts
2089 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2090 * function returns zero.
2092 int netdev_set_master(struct net_device *slave, struct net_device *master)
2094 struct net_device *old = slave->master;
2104 slave->master = master;
2112 slave->flags |= IFF_SLAVE;
2114 slave->flags &= ~IFF_SLAVE;
2116 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2121 * dev_set_promiscuity - update promiscuity count on a device
2125 * Add or remove promsicuity from a device. While the count in the device
2126 * remains above zero the interface remains promiscuous. Once it hits zero
2127 * the device reverts back to normal filtering operation. A negative inc
2128 * value is used to drop promiscuity on the device.
2130 void dev_set_promiscuity(struct net_device *dev, int inc)
2132 unsigned short old_flags = dev->flags;
2134 dev->flags |= IFF_PROMISC;
2135 if ((dev->promiscuity += inc) == 0)
2136 dev->flags &= ~IFF_PROMISC;
2137 if (dev->flags ^ old_flags) {
2139 printk(KERN_INFO "device %s %s promiscuous mode\n",
2140 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2146 * dev_set_allmulti - update allmulti count on a device
2150 * Add or remove reception of all multicast frames to a device. While the
2151 * count in the device remains above zero the interface remains listening
2152 * to all interfaces. Once it hits zero the device reverts back to normal
2153 * filtering operation. A negative @inc value is used to drop the counter
2154 * when releasing a resource needing all multicasts.
2157 void dev_set_allmulti(struct net_device *dev, int inc)
2159 unsigned short old_flags = dev->flags;
2161 dev->flags |= IFF_ALLMULTI;
2162 if ((dev->allmulti += inc) == 0)
2163 dev->flags &= ~IFF_ALLMULTI;
2164 if (dev->flags ^ old_flags)
2168 unsigned dev_get_flags(const struct net_device *dev)
2172 flags = (dev->flags & ~(IFF_PROMISC |
2175 (dev->gflags & (IFF_PROMISC |
2178 if (netif_running(dev) && netif_carrier_ok(dev))
2179 flags |= IFF_RUNNING;
2184 int dev_change_flags(struct net_device *dev, unsigned flags)
2187 int old_flags = dev->flags;
2190 * Set the flags on our device.
2193 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2194 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2196 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2200 * Load in the correct multicast list now the flags have changed.
2206 * Have we downed the interface. We handle IFF_UP ourselves
2207 * according to user attempts to set it, rather than blindly
2212 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2213 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2219 if (dev->flags & IFF_UP &&
2220 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2222 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2224 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2225 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2226 dev->gflags ^= IFF_PROMISC;
2227 dev_set_promiscuity(dev, inc);
2230 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2231 is important. Some (broken) drivers set IFF_PROMISC, when
2232 IFF_ALLMULTI is requested not asking us and not reporting.
2234 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2235 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2236 dev->gflags ^= IFF_ALLMULTI;
2237 dev_set_allmulti(dev, inc);
2240 if (old_flags ^ dev->flags)
2241 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2246 int dev_set_mtu(struct net_device *dev, int new_mtu)
2250 if (new_mtu == dev->mtu)
2253 /* MTU must be positive. */
2257 if (!netif_device_present(dev))
2261 if (dev->change_mtu)
2262 err = dev->change_mtu(dev, new_mtu);
2265 if (!err && dev->flags & IFF_UP)
2266 notifier_call_chain(&netdev_chain,
2267 NETDEV_CHANGEMTU, dev);
2273 * Perform the SIOCxIFxxx calls.
2275 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2278 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2284 case SIOCGIFFLAGS: /* Get interface flags */
2285 ifr->ifr_flags = dev_get_flags(dev);
2288 case SIOCSIFFLAGS: /* Set interface flags */
2289 return dev_change_flags(dev, ifr->ifr_flags);
2291 case SIOCGIFMETRIC: /* Get the metric on the interface
2292 (currently unused) */
2293 ifr->ifr_metric = 0;
2296 case SIOCSIFMETRIC: /* Set the metric on the interface
2297 (currently unused) */
2300 case SIOCGIFMTU: /* Get the MTU of a device */
2301 ifr->ifr_mtu = dev->mtu;
2304 case SIOCSIFMTU: /* Set the MTU of a device */
2305 return dev_set_mtu(dev, ifr->ifr_mtu);
2308 if ((size_t) dev->addr_len > sizeof ifr->ifr_hwaddr.sa_data)
2310 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2311 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, dev->addr_len);
2312 ifr->ifr_hwaddr.sa_family = dev->type;
2316 if (!dev->set_mac_address)
2318 if (ifr->ifr_hwaddr.sa_family != dev->type)
2320 if (!netif_device_present(dev))
2322 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2324 notifier_call_chain(&netdev_chain,
2325 NETDEV_CHANGEADDR, dev);
2328 case SIOCSIFHWBROADCAST:
2329 if (ifr->ifr_hwaddr.sa_family != dev->type)
2331 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2332 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2333 notifier_call_chain(&netdev_chain,
2334 NETDEV_CHANGEADDR, dev);
2338 ifr->ifr_map.mem_start = dev->mem_start;
2339 ifr->ifr_map.mem_end = dev->mem_end;
2340 ifr->ifr_map.base_addr = dev->base_addr;
2341 ifr->ifr_map.irq = dev->irq;
2342 ifr->ifr_map.dma = dev->dma;
2343 ifr->ifr_map.port = dev->if_port;
2347 if (dev->set_config) {
2348 if (!netif_device_present(dev))
2350 return dev->set_config(dev, &ifr->ifr_map);
2355 if (!dev->set_multicast_list ||
2356 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2358 if (!netif_device_present(dev))
2360 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2364 if (!dev->set_multicast_list ||
2365 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2367 if (!netif_device_present(dev))
2369 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2373 ifr->ifr_ifindex = dev->ifindex;
2377 ifr->ifr_qlen = dev->tx_queue_len;
2381 if (ifr->ifr_qlen < 0)
2383 dev->tx_queue_len = ifr->ifr_qlen;
2387 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2388 return dev_change_name(dev, ifr->ifr_newname);
2391 * Unknown or private ioctl
2395 if ((cmd >= SIOCDEVPRIVATE &&
2396 cmd <= SIOCDEVPRIVATE + 15) ||
2397 cmd == SIOCBONDENSLAVE ||
2398 cmd == SIOCBONDRELEASE ||
2399 cmd == SIOCBONDSETHWADDR ||
2400 cmd == SIOCBONDSLAVEINFOQUERY ||
2401 cmd == SIOCBONDINFOQUERY ||
2402 cmd == SIOCBONDCHANGEACTIVE ||
2403 cmd == SIOCGMIIPHY ||
2404 cmd == SIOCGMIIREG ||
2405 cmd == SIOCSMIIREG ||
2406 cmd == SIOCBRADDIF ||
2407 cmd == SIOCBRDELIF ||
2408 cmd == SIOCWANDEV) {
2410 if (dev->do_ioctl) {
2411 if (netif_device_present(dev))
2412 err = dev->do_ioctl(dev, ifr,
2425 * This function handles all "interface"-type I/O control requests. The actual
2426 * 'doing' part of this is dev_ifsioc above.
2430 * dev_ioctl - network device ioctl
2431 * @cmd: command to issue
2432 * @arg: pointer to a struct ifreq in user space
2434 * Issue ioctl functions to devices. This is normally called by the
2435 * user space syscall interfaces but can sometimes be useful for
2436 * other purposes. The return value is the return from the syscall if
2437 * positive or a negative errno code on error.
2440 int dev_ioctl(unsigned int cmd, void __user *arg)
2446 /* One special case: SIOCGIFCONF takes ifconf argument
2447 and requires shared lock, because it sleeps writing
2451 if (cmd == SIOCGIFCONF) {
2453 ret = dev_ifconf((char __user *) arg);
2457 if (cmd == SIOCGIFNAME)
2458 return dev_ifname((struct ifreq __user *)arg);
2460 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2463 ifr.ifr_name[IFNAMSIZ-1] = 0;
2465 colon = strchr(ifr.ifr_name, ':');
2470 * See which interface the caller is talking about.
2475 * These ioctl calls:
2476 * - can be done by all.
2477 * - atomic and do not require locking.
2488 dev_load(ifr.ifr_name);
2489 read_lock(&dev_base_lock);
2490 ret = dev_ifsioc(&ifr, cmd);
2491 read_unlock(&dev_base_lock);
2495 if (copy_to_user(arg, &ifr,
2496 sizeof(struct ifreq)))
2502 dev_load(ifr.ifr_name);
2504 ret = dev_ethtool(&ifr);
2509 if (copy_to_user(arg, &ifr,
2510 sizeof(struct ifreq)))
2516 * These ioctl calls:
2517 * - require superuser power.
2518 * - require strict serialization.
2524 if (!capable(CAP_NET_ADMIN))
2526 dev_load(ifr.ifr_name);
2528 ret = dev_ifsioc(&ifr, cmd);
2533 if (copy_to_user(arg, &ifr,
2534 sizeof(struct ifreq)))
2540 * These ioctl calls:
2541 * - require superuser power.
2542 * - require strict serialization.
2543 * - do not return a value
2553 case SIOCSIFHWBROADCAST:
2556 case SIOCBONDENSLAVE:
2557 case SIOCBONDRELEASE:
2558 case SIOCBONDSETHWADDR:
2559 case SIOCBONDSLAVEINFOQUERY:
2560 case SIOCBONDINFOQUERY:
2561 case SIOCBONDCHANGEACTIVE:
2564 if (!capable(CAP_NET_ADMIN))
2566 dev_load(ifr.ifr_name);
2568 ret = dev_ifsioc(&ifr, cmd);
2573 /* Get the per device memory space. We can add this but
2574 * currently do not support it */
2576 /* Set the per device memory buffer space.
2577 * Not applicable in our case */
2582 * Unknown or private ioctl.
2585 if (cmd == SIOCWANDEV ||
2586 (cmd >= SIOCDEVPRIVATE &&
2587 cmd <= SIOCDEVPRIVATE + 15)) {
2588 dev_load(ifr.ifr_name);
2590 ret = dev_ifsioc(&ifr, cmd);
2592 if (!ret && copy_to_user(arg, &ifr,
2593 sizeof(struct ifreq)))
2598 /* Take care of Wireless Extensions */
2599 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2600 /* If command is `set a parameter', or
2601 * `get the encoding parameters', check if
2602 * the user has the right to do it */
2603 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2604 if (!capable(CAP_NET_ADMIN))
2607 dev_load(ifr.ifr_name);
2609 /* Follow me in net/core/wireless.c */
2610 ret = wireless_process_ioctl(&ifr, cmd);
2612 if (!ret && IW_IS_GET(cmd) &&
2613 copy_to_user(arg, &ifr,
2614 sizeof(struct ifreq)))
2618 #endif /* WIRELESS_EXT */
2625 * dev_new_index - allocate an ifindex
2627 * Returns a suitable unique value for a new device interface
2628 * number. The caller must hold the rtnl semaphore or the
2629 * dev_base_lock to be sure it remains unique.
2631 static int dev_new_index(void)
2637 if (!__dev_get_by_index(ifindex))
2642 static int dev_boot_phase = 1;
2644 /* Delayed registration/unregisteration */
2645 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2646 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2648 static inline void net_set_todo(struct net_device *dev)
2650 spin_lock(&net_todo_list_lock);
2651 list_add_tail(&dev->todo_list, &net_todo_list);
2652 spin_unlock(&net_todo_list_lock);
2656 * register_netdevice - register a network device
2657 * @dev: device to register
2659 * Take a completed network device structure and add it to the kernel
2660 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2661 * chain. 0 is returned on success. A negative errno code is returned
2662 * on a failure to set up the device, or if the name is a duplicate.
2664 * Callers must hold the rtnl semaphore. See the comment at the
2665 * end of Space.c for details about the locking. You may want
2666 * register_netdev() instead of this.
2669 * The locking appears insufficient to guarantee two parallel registers
2670 * will not get the same name.
2673 int register_netdevice(struct net_device *dev)
2675 struct hlist_head *head;
2676 struct hlist_node *p;
2679 BUG_ON(dev_boot_phase);
2682 /* When net_device's are persistent, this will be fatal. */
2683 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2685 spin_lock_init(&dev->queue_lock);
2686 spin_lock_init(&dev->xmit_lock);
2687 dev->xmit_lock_owner = -1;
2688 #ifdef CONFIG_NET_CLS_ACT
2689 spin_lock_init(&dev->ingress_lock);
2692 ret = alloc_divert_blk(dev);
2698 /* Init, if this function is available */
2700 ret = dev->init(dev);
2708 if (!dev_valid_name(dev->name)) {
2713 dev->ifindex = dev_new_index();
2714 if (dev->iflink == -1)
2715 dev->iflink = dev->ifindex;
2717 /* Check for existence of name */
2718 head = dev_name_hash(dev->name);
2719 hlist_for_each(p, head) {
2720 struct net_device *d
2721 = hlist_entry(p, struct net_device, name_hlist);
2722 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2728 /* Fix illegal SG+CSUM combinations. */
2729 if ((dev->features & NETIF_F_SG) &&
2730 !(dev->features & (NETIF_F_IP_CSUM |
2732 NETIF_F_HW_CSUM))) {
2733 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2735 dev->features &= ~NETIF_F_SG;
2739 * nil rebuild_header routine,
2740 * that should be never called and used as just bug trap.
2743 if (!dev->rebuild_header)
2744 dev->rebuild_header = default_rebuild_header;
2747 * Default initial state at registry is that the
2748 * device is present.
2751 set_bit(__LINK_STATE_PRESENT, &dev->state);
2754 dev_init_scheduler(dev);
2755 write_lock_bh(&dev_base_lock);
2757 dev_tail = &dev->next;
2758 hlist_add_head(&dev->name_hlist, head);
2759 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2761 dev->reg_state = NETREG_REGISTERING;
2762 write_unlock_bh(&dev_base_lock);
2764 /* Notify protocols, that a new device appeared. */
2765 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2767 /* Finish registration after unlock */
2774 free_divert_blk(dev);
2779 * netdev_wait_allrefs - wait until all references are gone.
2781 * This is called when unregistering network devices.
2783 * Any protocol or device that holds a reference should register
2784 * for netdevice notification, and cleanup and put back the
2785 * reference if they receive an UNREGISTER event.
2786 * We can get stuck here if buggy protocols don't correctly
2789 static void netdev_wait_allrefs(struct net_device *dev)
2791 unsigned long rebroadcast_time, warning_time;
2793 rebroadcast_time = warning_time = jiffies;
2794 while (atomic_read(&dev->refcnt) != 0) {
2795 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2798 /* Rebroadcast unregister notification */
2799 notifier_call_chain(&netdev_chain,
2800 NETDEV_UNREGISTER, dev);
2802 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2804 /* We must not have linkwatch events
2805 * pending on unregister. If this
2806 * happens, we simply run the queue
2807 * unscheduled, resulting in a noop
2810 linkwatch_run_queue();
2815 rebroadcast_time = jiffies;
2818 current->state = TASK_INTERRUPTIBLE;
2819 schedule_timeout(HZ / 4);
2821 if (time_after(jiffies, warning_time + 10 * HZ)) {
2822 printk(KERN_EMERG "unregister_netdevice: "
2823 "waiting for %s to become free. Usage "
2825 dev->name, atomic_read(&dev->refcnt));
2826 warning_time = jiffies;
2835 * register_netdevice(x1);
2836 * register_netdevice(x2);
2838 * unregister_netdevice(y1);
2839 * unregister_netdevice(y2);
2845 * We are invoked by rtnl_unlock() after it drops the semaphore.
2846 * This allows us to deal with problems:
2847 * 1) We can create/delete sysfs objects which invoke hotplug
2848 * without deadlocking with linkwatch via keventd.
2849 * 2) Since we run with the RTNL semaphore not held, we can sleep
2850 * safely in order to wait for the netdev refcnt to drop to zero.
2852 static DECLARE_MUTEX(net_todo_run_mutex);
2853 void netdev_run_todo(void)
2855 struct list_head list = LIST_HEAD_INIT(list);
2859 /* Need to guard against multiple cpu's getting out of order. */
2860 down(&net_todo_run_mutex);
2862 /* Not safe to do outside the semaphore. We must not return
2863 * until all unregister events invoked by the local processor
2864 * have been completed (either by this todo run, or one on
2867 if (list_empty(&net_todo_list))
2870 /* Snapshot list, allow later requests */
2871 spin_lock(&net_todo_list_lock);
2872 list_splice_init(&net_todo_list, &list);
2873 spin_unlock(&net_todo_list_lock);
2875 while (!list_empty(&list)) {
2876 struct net_device *dev
2877 = list_entry(list.next, struct net_device, todo_list);
2878 list_del(&dev->todo_list);
2880 switch(dev->reg_state) {
2881 case NETREG_REGISTERING:
2882 err = netdev_register_sysfs(dev);
2884 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2886 dev->reg_state = NETREG_REGISTERED;
2889 case NETREG_UNREGISTERING:
2890 netdev_unregister_sysfs(dev);
2891 dev->reg_state = NETREG_UNREGISTERED;
2893 netdev_wait_allrefs(dev);
2896 BUG_ON(atomic_read(&dev->refcnt));
2897 BUG_TRAP(!dev->ip_ptr);
2898 BUG_TRAP(!dev->ip6_ptr);
2899 BUG_TRAP(!dev->dn_ptr);
2902 /* It must be the very last action,
2903 * after this 'dev' may point to freed up memory.
2905 if (dev->destructor)
2906 dev->destructor(dev);
2910 printk(KERN_ERR "network todo '%s' but state %d\n",
2911 dev->name, dev->reg_state);
2917 up(&net_todo_run_mutex);
2921 * free_netdev - free network device
2924 * This function does the last stage of destroying an allocated device
2925 * interface. The reference to the device object is released.
2926 * If this is the last reference then it will be freed.
2928 void free_netdev(struct net_device *dev)
2931 /* Compatiablity with error handling in drivers */
2932 if (dev->reg_state == NETREG_UNINITIALIZED) {
2933 kfree((char *)dev - dev->padded);
2937 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
2938 dev->reg_state = NETREG_RELEASED;
2940 /* will free via class release */
2941 class_device_put(&dev->class_dev);
2943 kfree((char *)dev - dev->padded);
2947 /* Synchronize with packet receive processing. */
2948 void synchronize_net(void)
2951 synchronize_kernel();
2955 * unregister_netdevice - remove device from the kernel
2958 * This function shuts down a device interface and removes it
2959 * from the kernel tables. On success 0 is returned, on a failure
2960 * a negative errno code is returned.
2962 * Callers must hold the rtnl semaphore. See the comment at the
2963 * end of Space.c for details about the locking. You may want
2964 * unregister_netdev() instead of this.
2967 int unregister_netdevice(struct net_device *dev)
2969 struct net_device *d, **dp;
2971 BUG_ON(dev_boot_phase);
2974 /* Some devices call without registering for initialization unwind. */
2975 if (dev->reg_state == NETREG_UNINITIALIZED) {
2976 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
2977 "was registered\n", dev->name, dev);
2981 BUG_ON(dev->reg_state != NETREG_REGISTERED);
2983 /* If device is running, close it first. */
2984 if (dev->flags & IFF_UP)
2987 /* And unlink it from device chain. */
2988 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
2990 write_lock_bh(&dev_base_lock);
2991 hlist_del(&dev->name_hlist);
2992 hlist_del(&dev->index_hlist);
2993 if (dev_tail == &dev->next)
2996 write_unlock_bh(&dev_base_lock);
3001 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3006 dev->reg_state = NETREG_UNREGISTERING;
3010 /* Shutdown queueing discipline. */
3014 /* Notify protocols, that we are about to destroy
3015 this device. They should clean all the things.
3017 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3020 * Flush the multicast chain
3022 dev_mc_discard(dev);
3027 /* Notifier chain MUST detach us from master device. */
3028 BUG_TRAP(!dev->master);
3030 free_divert_blk(dev);
3032 /* Finish processing unregister after unlock */
3041 #ifdef CONFIG_HOTPLUG_CPU
3042 static int dev_cpu_callback(struct notifier_block *nfb,
3043 unsigned long action,
3046 struct sk_buff **list_skb;
3047 struct net_device **list_net;
3048 struct sk_buff *skb;
3049 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3050 struct softnet_data *sd, *oldsd;
3052 if (action != CPU_DEAD)
3055 local_irq_disable();
3056 cpu = smp_processor_id();
3057 sd = &per_cpu(softnet_data, cpu);
3058 oldsd = &per_cpu(softnet_data, oldcpu);
3060 /* Find end of our completion_queue. */
3061 list_skb = &sd->completion_queue;
3063 list_skb = &(*list_skb)->next;
3064 /* Append completion queue from offline CPU. */
3065 *list_skb = oldsd->completion_queue;
3066 oldsd->completion_queue = NULL;
3068 /* Find end of our output_queue. */
3069 list_net = &sd->output_queue;
3071 list_net = &(*list_net)->next_sched;
3072 /* Append output queue from offline CPU. */
3073 *list_net = oldsd->output_queue;
3074 oldsd->output_queue = NULL;
3076 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3079 /* Process offline CPU's input_pkt_queue */
3080 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3085 #endif /* CONFIG_HOTPLUG_CPU */
3089 * Initialize the DEV module. At boot time this walks the device list and
3090 * unhooks any devices that fail to initialise (normally hardware not
3091 * present) and leaves us with a valid list of present and active devices.
3096 * This is called single threaded during boot, so no need
3097 * to take the rtnl semaphore.
3099 static int __init net_dev_init(void)
3101 int i, rc = -ENOMEM;
3103 BUG_ON(!dev_boot_phase);
3107 if (dev_proc_init())
3110 if (netdev_sysfs_init())
3113 INIT_LIST_HEAD(&ptype_all);
3114 for (i = 0; i < 16; i++)
3115 INIT_LIST_HEAD(&ptype_base[i]);
3117 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3118 INIT_HLIST_HEAD(&dev_name_head[i]);
3120 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3121 INIT_HLIST_HEAD(&dev_index_head[i]);
3124 * Initialise the packet receive queues.
3127 for (i = 0; i < NR_CPUS; i++) {
3128 struct softnet_data *queue;
3130 queue = &per_cpu(softnet_data, i);
3131 skb_queue_head_init(&queue->input_pkt_queue);
3132 queue->throttle = 0;
3133 queue->cng_level = 0;
3134 queue->avg_blog = 10; /* arbitrary non-zero */
3135 queue->completion_queue = NULL;
3136 INIT_LIST_HEAD(&queue->poll_list);
3137 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3138 queue->backlog_dev.weight = weight_p;
3139 queue->backlog_dev.poll = process_backlog;
3140 atomic_set(&queue->backlog_dev.refcnt, 1);
3143 #ifdef OFFLINE_SAMPLE
3144 samp_timer.expires = jiffies + (10 * HZ);
3145 add_timer(&samp_timer);
3150 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3151 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3153 hotcpu_notifier(dev_cpu_callback, 0);
3161 subsys_initcall(net_dev_init);
3163 EXPORT_SYMBOL(__dev_get_by_index);
3164 EXPORT_SYMBOL(__dev_get_by_name);
3165 EXPORT_SYMBOL(__dev_remove_pack);
3166 EXPORT_SYMBOL(__skb_linearize);
3167 EXPORT_SYMBOL(call_netdevice_notifiers);
3168 EXPORT_SYMBOL(dev_add_pack);
3169 EXPORT_SYMBOL(dev_alloc_name);
3170 EXPORT_SYMBOL(dev_close);
3171 EXPORT_SYMBOL(dev_get_by_flags);
3172 EXPORT_SYMBOL(dev_get_by_index);
3173 EXPORT_SYMBOL(dev_get_by_name);
3174 EXPORT_SYMBOL(dev_ioctl);
3175 EXPORT_SYMBOL(dev_open);
3176 EXPORT_SYMBOL(dev_queue_xmit);
3177 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
3178 EXPORT_SYMBOL(dev_queue_xmit_nit);
3180 EXPORT_SYMBOL(dev_remove_pack);
3181 EXPORT_SYMBOL(dev_set_allmulti);
3182 EXPORT_SYMBOL(dev_set_promiscuity);
3183 EXPORT_SYMBOL(dev_change_flags);
3184 EXPORT_SYMBOL(dev_change_name);
3185 EXPORT_SYMBOL(dev_set_mtu);
3186 EXPORT_SYMBOL(free_netdev);
3187 EXPORT_SYMBOL(netdev_boot_setup_check);
3188 EXPORT_SYMBOL(netdev_set_master);
3189 EXPORT_SYMBOL(netdev_state_change);
3190 EXPORT_SYMBOL(netif_receive_skb);
3191 EXPORT_SYMBOL(netif_rx);
3192 EXPORT_SYMBOL(register_gifconf);
3193 EXPORT_SYMBOL(register_netdevice);
3194 EXPORT_SYMBOL(register_netdevice_notifier);
3195 EXPORT_SYMBOL(skb_checksum_help);
3196 EXPORT_SYMBOL(synchronize_net);
3197 EXPORT_SYMBOL(unregister_netdevice);
3198 EXPORT_SYMBOL(unregister_netdevice_notifier);
3200 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3201 EXPORT_SYMBOL(br_handle_frame_hook);
3205 EXPORT_SYMBOL(dev_load);
3208 #ifdef CONFIG_NET_CLS_ACT
3209 EXPORT_SYMBOL(ing_filter);
3213 EXPORT_PER_CPU_SYMBOL(softnet_data);