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>
117 /* This define, if set, will randomly drop a packet when congestion
118 * is more than moderate. It helps fairness in the multi-interface
119 * case when one of them is a hog, but it kills performance for the
120 * single interface case so it is off now by default.
124 /* Setting this will sample the queue lengths and thus congestion
125 * via a timer instead of as each packet is received.
127 #undef OFFLINE_SAMPLE
130 * The list of packet types we will receive (as opposed to discard)
131 * and the routines to invoke.
133 * Why 16. Because with 16 the only overlap we get on a hash of the
134 * low nibble of the protocol value is RARP/SNAP/X.25.
136 * NOTE: That is no longer true with the addition of VLAN tags. Not
137 * sure which should go first, but I bet it won't make much
138 * difference if we are running VLANs. The good news is that
139 * this protocol won't be in the list unless compiled in, so
140 * the average user (w/out VLANs) will not be adversly affected.
157 static spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
158 static struct list_head ptype_base[16]; /* 16 way hashed list */
159 static struct list_head ptype_all; /* Taps */
161 #ifdef OFFLINE_SAMPLE
162 static void sample_queue(unsigned long dummy);
163 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
167 * The @dev_base list is protected by @dev_base_lock and the rtln
170 * Pure readers hold dev_base_lock for reading.
172 * Writers must hold the rtnl semaphore while they loop through the
173 * dev_base list, and hold dev_base_lock for writing when they do the
174 * actual updates. This allows pure readers to access the list even
175 * while a writer is preparing to update it.
177 * To put it another way, dev_base_lock is held for writing only to
178 * protect against pure readers; the rtnl semaphore provides the
179 * protection against other writers.
181 * See, for example usages, register_netdevice() and
182 * unregister_netdevice(), which must be called with the rtnl
185 struct net_device *dev_base;
186 struct net_device **dev_tail = &dev_base;
187 rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
189 EXPORT_SYMBOL(dev_base);
190 EXPORT_SYMBOL(dev_base_lock);
192 #define NETDEV_HASHBITS 8
193 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
194 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
196 static inline struct hlist_head *dev_name_hash(const char *name)
198 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
199 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
202 static inline struct hlist_head *dev_index_hash(int ifindex)
204 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
211 static struct notifier_block *netdev_chain;
214 * Device drivers call our routines to queue packets here. We empty the
215 * queue in the local softnet handler.
217 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
220 extern int netdev_sysfs_init(void);
221 extern int netdev_register_sysfs(struct net_device *);
222 extern void netdev_unregister_sysfs(struct net_device *);
224 #define netdev_sysfs_init() (0)
225 #define netdev_register_sysfs(dev) (0)
226 #define netdev_unregister_sysfs(dev) do { } while(0)
230 /*******************************************************************************
232 Protocol management and registration routines
234 *******************************************************************************/
243 * Add a protocol ID to the list. Now that the input handler is
244 * smarter we can dispense with all the messy stuff that used to be
247 * BEWARE!!! Protocol handlers, mangling input packets,
248 * MUST BE last in hash buckets and checking protocol handlers
249 * MUST start from promiscuous ptype_all chain in net_bh.
250 * It is true now, do not change it.
251 * Explanation follows: if protocol handler, mangling packet, will
252 * be the first on list, it is not able to sense, that packet
253 * is cloned and should be copied-on-write, so that it will
254 * change it and subsequent readers will get broken packet.
259 * dev_add_pack - add packet handler
260 * @pt: packet type declaration
262 * Add a protocol handler to the networking stack. The passed &packet_type
263 * is linked into kernel lists and may not be freed until it has been
264 * removed from the kernel lists.
266 * This call does not sleep therefore it can not
267 * guarantee all CPU's that are in middle of receiving packets
268 * will see the new packet type (until the next received packet).
271 void dev_add_pack(struct packet_type *pt)
275 spin_lock_bh(&ptype_lock);
276 if (pt->type == htons(ETH_P_ALL)) {
278 list_add_rcu(&pt->list, &ptype_all);
280 hash = ntohs(pt->type) & 15;
281 list_add_rcu(&pt->list, &ptype_base[hash]);
283 spin_unlock_bh(&ptype_lock);
286 extern void linkwatch_run_queue(void);
291 * __dev_remove_pack - remove packet handler
292 * @pt: packet type declaration
294 * Remove a protocol handler that was previously added to the kernel
295 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
296 * from the kernel lists and can be freed or reused once this function
299 * The packet type might still be in use by receivers
300 * and must not be freed until after all the CPU's have gone
301 * through a quiescent state.
303 void __dev_remove_pack(struct packet_type *pt)
305 struct list_head *head;
306 struct packet_type *pt1;
308 spin_lock_bh(&ptype_lock);
310 if (pt->type == htons(ETH_P_ALL)) {
314 head = &ptype_base[ntohs(pt->type) & 15];
316 list_for_each_entry(pt1, head, list) {
318 list_del_rcu(&pt->list);
323 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
325 spin_unlock_bh(&ptype_lock);
328 * dev_remove_pack - remove packet handler
329 * @pt: packet type declaration
331 * Remove a protocol handler that was previously added to the kernel
332 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
333 * from the kernel lists and can be freed or reused once this function
336 * This call sleeps to guarantee that no CPU is looking at the packet
339 void dev_remove_pack(struct packet_type *pt)
341 __dev_remove_pack(pt);
346 /******************************************************************************
348 Device Boot-time Settings Routines
350 *******************************************************************************/
352 /* Boot time configuration table */
353 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
356 * netdev_boot_setup_add - add new setup entry
357 * @name: name of the device
358 * @map: configured settings for the device
360 * Adds new setup entry to the dev_boot_setup list. The function
361 * returns 0 on error and 1 on success. This is a generic routine to
364 int netdev_boot_setup_add(char *name, struct ifmap *map)
366 struct netdev_boot_setup *s;
370 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
371 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
372 memset(s[i].name, 0, sizeof(s[i].name));
373 strcpy(s[i].name, name);
374 memcpy(&s[i].map, map, sizeof(s[i].map));
379 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
383 * netdev_boot_setup_check - check boot time settings
384 * @dev: the netdevice
386 * Check boot time settings for the device.
387 * The found settings are set for the device to be used
388 * later in the device probing.
389 * Returns 0 if no settings found, 1 if they are.
391 int netdev_boot_setup_check(struct net_device *dev)
393 struct netdev_boot_setup *s = dev_boot_setup;
396 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
397 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
398 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
399 dev->irq = s[i].map.irq;
400 dev->base_addr = s[i].map.base_addr;
401 dev->mem_start = s[i].map.mem_start;
402 dev->mem_end = s[i].map.mem_end;
411 * netdev_boot_base - get address from boot time settings
412 * @prefix: prefix for network device
413 * @unit: id for network device
415 * Check boot time settings for the base address of device.
416 * The found settings are set for the device to be used
417 * later in the device probing.
418 * Returns 0 if no settings found.
420 unsigned long netdev_boot_base(const char *prefix, int unit)
422 const struct netdev_boot_setup *s = dev_boot_setup;
426 sprintf(name, "%s%d", prefix, unit);
429 * If device already registered then return base of 1
430 * to indicate not to probe for this interface
432 if (__dev_get_by_name(name))
435 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
436 if (!strcmp(name, s[i].name))
437 return s[i].map.base_addr;
442 * Saves at boot time configured settings for any netdevice.
444 int __init netdev_boot_setup(char *str)
449 str = get_options(str, ARRAY_SIZE(ints), ints);
454 memset(&map, 0, sizeof(map));
458 map.base_addr = ints[2];
460 map.mem_start = ints[3];
462 map.mem_end = ints[4];
464 /* Add new entry to the list */
465 return netdev_boot_setup_add(str, &map);
468 __setup("netdev=", netdev_boot_setup);
470 /*******************************************************************************
472 Device Interface Subroutines
474 *******************************************************************************/
477 * __dev_get_by_name - find a device by its name
478 * @name: name to find
480 * Find an interface by name. Must be called under RTNL semaphore
481 * or @dev_base_lock. If the name is found a pointer to the device
482 * is returned. If the name is not found then %NULL is returned. The
483 * reference counters are not incremented so the caller must be
484 * careful with locks.
487 struct net_device *__dev_get_by_name(const char *name)
489 struct hlist_node *p;
491 hlist_for_each(p, dev_name_hash(name)) {
492 struct net_device *dev
493 = hlist_entry(p, struct net_device, name_hlist);
494 if (!strncmp(dev->name, name, IFNAMSIZ))
501 * dev_get_by_name - find a device by its name
502 * @name: name to find
504 * Find an interface by name. This can be called from any
505 * context and does its own locking. The returned handle has
506 * the usage count incremented and the caller must use dev_put() to
507 * release it when it is no longer needed. %NULL is returned if no
508 * matching device is found.
511 struct net_device *dev_get_by_name(const char *name)
513 struct net_device *dev;
515 read_lock(&dev_base_lock);
516 dev = __dev_get_by_name(name);
519 read_unlock(&dev_base_lock);
524 Return value is changed to int to prevent illegal usage in future.
525 It is still legal to use to check for device existence.
527 User should understand, that the result returned by this function
528 is meaningless, if it was not issued under rtnl semaphore.
532 * dev_get - test if a device exists
533 * @name: name to test for
535 * Test if a name exists. Returns true if the name is found. In order
536 * to be sure the name is not allocated or removed during the test the
537 * caller must hold the rtnl semaphore.
539 * This function exists only for back compatibility with older
542 int __dev_get(const char *name)
544 struct net_device *dev;
546 read_lock(&dev_base_lock);
547 dev = __dev_get_by_name(name);
548 read_unlock(&dev_base_lock);
553 * __dev_get_by_index - find a device by its ifindex
554 * @ifindex: index of device
556 * Search for an interface by index. Returns %NULL if the device
557 * is not found or a pointer to the device. The device has not
558 * had its reference counter increased so the caller must be careful
559 * about locking. The caller must hold either the RTNL semaphore
563 struct net_device *__dev_get_by_index(int ifindex)
565 struct hlist_node *p;
567 hlist_for_each(p, dev_index_hash(ifindex)) {
568 struct net_device *dev
569 = hlist_entry(p, struct net_device, index_hlist);
570 if (dev->ifindex == ifindex)
578 * dev_get_by_index - find a device by its ifindex
579 * @ifindex: index of device
581 * Search for an interface by index. Returns NULL if the device
582 * is not found or a pointer to the device. The device returned has
583 * had a reference added and the pointer is safe until the user calls
584 * dev_put to indicate they have finished with it.
587 struct net_device *dev_get_by_index(int ifindex)
589 struct net_device *dev;
591 read_lock(&dev_base_lock);
592 dev = __dev_get_by_index(ifindex);
595 read_unlock(&dev_base_lock);
600 * dev_getbyhwaddr - find a device by its hardware address
601 * @type: media type of device
602 * @ha: hardware address
604 * Search for an interface by MAC address. Returns NULL if the device
605 * is not found or a pointer to the device. The caller must hold the
606 * rtnl semaphore. The returned device has not had its ref count increased
607 * and the caller must therefore be careful about locking
610 * If the API was consistent this would be __dev_get_by_hwaddr
613 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
615 struct net_device *dev;
619 for (dev = dev_base; dev; dev = dev->next)
620 if (dev->type == type &&
621 !memcmp(dev->dev_addr, ha, dev->addr_len))
626 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
628 struct net_device *dev;
630 for (dev = dev_base; dev; dev = dev->next)
631 if (dev->type == type)
636 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
638 struct net_device *dev_getfirstbyhwtype(unsigned short type)
640 struct net_device *dev;
643 dev = __dev_getfirstbyhwtype(type);
650 EXPORT_SYMBOL(dev_getfirstbyhwtype);
653 * dev_get_by_flags - find any device with given flags
654 * @if_flags: IFF_* values
655 * @mask: bitmask of bits in if_flags to check
657 * Search for any interface with the given flags. Returns NULL if a device
658 * is not found or a pointer to the device. The device returned has
659 * had a reference added and the pointer is safe until the user calls
660 * dev_put to indicate they have finished with it.
663 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
665 struct net_device *dev;
667 read_lock(&dev_base_lock);
668 dev = __dev_get_by_flags(if_flags, mask);
671 read_unlock(&dev_base_lock);
676 * __dev_get_by_flags - find any device with given flags
677 * @if_flags: IFF_* values
678 * @mask: bitmask of bits in if_flags to check
680 * Search for any interface with the given flags. Returns NULL if a device
681 * is not found or a pointer to the device. The caller must hold either
682 * the RTNL semaphore or @dev_base_lock.
685 struct net_device *__dev_get_by_flags(unsigned short if_flags, unsigned short mask)
687 struct net_device *dev;
689 for (dev = dev_base; dev != NULL; dev = dev->next) {
690 if (((dev->flags ^ if_flags) & mask) == 0)
697 * dev_valid_name - check if name is okay for network device
700 * Network device names need to be valid file names to
701 * to allow sysfs to work
703 int dev_valid_name(const char *name)
705 return !(*name == '\0'
706 || !strcmp(name, ".")
707 || !strcmp(name, "..")
708 || strchr(name, '/'));
712 * dev_alloc_name - allocate a name for a device
714 * @name: name format string
716 * Passed a format string - eg "lt%d" it will try and find a suitable
717 * id. Not efficient for many devices, not called a lot. The caller
718 * must hold the dev_base or rtnl lock while allocating the name and
719 * adding the device in order to avoid duplicates. Returns the number
720 * of the unit assigned or a negative errno code.
723 int dev_alloc_name(struct net_device *dev, const char *name)
728 const int max_netdevices = 8*PAGE_SIZE;
730 struct net_device *d;
732 p = strnchr(name, IFNAMSIZ-1, '%');
735 * Verify the string as this thing may have come from
736 * the user. There must be either one "%d" and no other "%"
739 if (p[1] != 'd' || strchr(p + 2, '%'))
742 /* Use one page as a bit array of possible slots */
743 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
747 for (d = dev_base; d; d = d->next) {
748 if (!sscanf(d->name, name, &i))
750 if (i < 0 || i >= max_netdevices)
753 /* avoid cases where sscanf is not exact inverse of printf */
754 snprintf(buf, sizeof(buf), name, i);
755 if (!strncmp(buf, d->name, IFNAMSIZ))
759 i = find_first_zero_bit(inuse, max_netdevices);
760 free_page((unsigned long) inuse);
763 snprintf(buf, sizeof(buf), name, i);
764 if (!__dev_get_by_name(buf)) {
765 strlcpy(dev->name, buf, IFNAMSIZ);
769 /* It is possible to run out of possible slots
770 * when the name is long and there isn't enough space left
771 * for the digits, or if all bits are used.
778 * dev_change_name - change name of a device
780 * @newname: name (or format string) must be at least IFNAMSIZ
782 * Change name of a device, can pass format strings "eth%d".
785 int dev_change_name(struct net_device *dev, char *newname)
791 if (dev->flags & IFF_UP)
794 if (!dev_valid_name(newname))
797 if (strchr(newname, '%')) {
798 err = dev_alloc_name(dev, newname);
801 strcpy(newname, dev->name);
803 else if (__dev_get_by_name(newname))
806 strlcpy(dev->name, newname, IFNAMSIZ);
808 err = class_device_rename(&dev->class_dev, dev->name);
810 hlist_del(&dev->name_hlist);
811 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
812 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
819 * netdev_state_change - device changes state
820 * @dev: device to cause notification
822 * Called to indicate a device has changed state. This function calls
823 * the notifier chains for netdev_chain and sends a NEWLINK message
824 * to the routing socket.
826 void netdev_state_change(struct net_device *dev)
828 if (dev->flags & IFF_UP) {
829 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
830 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
835 * dev_load - load a network module
836 * @name: name of interface
838 * If a network interface is not present and the process has suitable
839 * privileges this function loads the module. If module loading is not
840 * available in this kernel then it becomes a nop.
843 void dev_load(const char *name)
845 struct net_device *dev;
847 read_lock(&dev_base_lock);
848 dev = __dev_get_by_name(name);
849 read_unlock(&dev_base_lock);
851 if (!dev && capable(CAP_SYS_MODULE))
852 request_module("%s", name);
855 static int default_rebuild_header(struct sk_buff *skb)
857 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
858 skb->dev ? skb->dev->name : "NULL!!!");
865 * Some old buggy device drivers change get_stats after registering
866 * the device. Try and trap them here.
867 * This can be elimnated when all devices are known fixed.
869 static inline int get_stats_changed(struct net_device *dev)
871 int changed = dev->last_stats != dev->get_stats;
872 dev->last_stats = dev->get_stats;
877 * dev_open - prepare an interface for use.
878 * @dev: device to open
880 * Takes a device from down to up state. The device's private open
881 * function is invoked and then the multicast lists are loaded. Finally
882 * the device is moved into the up state and a %NETDEV_UP message is
883 * sent to the netdev notifier chain.
885 * Calling this function on an active interface is a nop. On a failure
886 * a negative errno code is returned.
888 int dev_open(struct net_device *dev)
896 if (dev->flags & IFF_UP)
900 * Check for broken device drivers.
902 if (get_stats_changed(dev) && net_ratelimit()) {
903 printk(KERN_ERR "%s: driver changed get_stats after register\n",
908 * Is it even present?
910 if (!netif_device_present(dev))
914 * Call device private open method
916 set_bit(__LINK_STATE_START, &dev->state);
918 ret = dev->open(dev);
920 clear_bit(__LINK_STATE_START, &dev->state);
924 * Check for more broken device drivers.
926 if (get_stats_changed(dev) && net_ratelimit()) {
927 printk(KERN_ERR "%s: driver changed get_stats in open\n",
932 * If it went open OK then:
939 dev->flags |= IFF_UP;
942 * Initialize multicasting status
947 * Wakeup transmit queue engine
952 * ... and announce new interface.
954 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
960 * dev_close - shutdown an interface.
961 * @dev: device to shutdown
963 * This function moves an active device into down state. A
964 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
965 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
968 int dev_close(struct net_device *dev)
970 if (!(dev->flags & IFF_UP))
974 * Tell people we are going down, so that they can
975 * prepare to death, when device is still operating.
977 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
981 clear_bit(__LINK_STATE_START, &dev->state);
983 /* Synchronize to scheduled poll. We cannot touch poll list,
984 * it can be even on different cpu. So just clear netif_running(),
985 * and wait when poll really will happen. Actually, the best place
986 * for this is inside dev->stop() after device stopped its irq
987 * engine, but this requires more changes in devices. */
989 smp_mb__after_clear_bit(); /* Commit netif_running(). */
990 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
992 current->state = TASK_INTERRUPTIBLE;
997 * Call the device specific close. This cannot fail.
998 * Only if device is UP
1000 * We allow it to be called even after a DETACH hot-plug
1007 * Device is now down.
1010 dev->flags &= ~IFF_UP;
1013 * Tell people we are down
1015 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1022 * Device change register/unregister. These are not inline or static
1023 * as we export them to the world.
1027 * register_netdevice_notifier - register a network notifier block
1030 * Register a notifier to be called when network device events occur.
1031 * The notifier passed is linked into the kernel structures and must
1032 * not be reused until it has been unregistered. A negative errno code
1033 * is returned on a failure.
1035 * When registered all registration and up events are replayed
1036 * to the new notifier to allow device to have a race free
1037 * view of the network device list.
1040 int register_netdevice_notifier(struct notifier_block *nb)
1042 struct net_device *dev;
1046 err = notifier_chain_register(&netdev_chain, nb);
1048 for (dev = dev_base; dev; dev = dev->next) {
1049 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1051 if (dev->flags & IFF_UP)
1052 nb->notifier_call(nb, NETDEV_UP, dev);
1060 * unregister_netdevice_notifier - unregister a network notifier block
1063 * Unregister a notifier previously registered by
1064 * register_netdevice_notifier(). The notifier is unlinked into the
1065 * kernel structures and may then be reused. A negative errno code
1066 * is returned on a failure.
1069 int unregister_netdevice_notifier(struct notifier_block *nb)
1071 return notifier_chain_unregister(&netdev_chain, nb);
1075 * call_netdevice_notifiers - call all network notifier blocks
1076 * @val: value passed unmodified to notifier function
1077 * @v: pointer passed unmodified to notifier function
1079 * Call all network notifier blocks. Parameters and return value
1080 * are as for notifier_call_chain().
1083 int call_netdevice_notifiers(unsigned long val, void *v)
1085 return notifier_call_chain(&netdev_chain, val, v);
1089 * Support routine. Sends outgoing frames to any network
1090 * taps currently in use.
1093 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1095 struct packet_type *ptype;
1096 net_timestamp(&skb->stamp);
1099 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1100 /* Never send packets back to the socket
1101 * they originated from - MvS (miquels@drinkel.ow.org)
1103 if ((ptype->dev == dev || !ptype->dev) &&
1104 (ptype->af_packet_priv == NULL ||
1105 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1106 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1110 /* skb->nh should be correctly
1111 set by sender, so that the second statement is
1112 just protection against buggy protocols.
1114 skb2->mac.raw = skb2->data;
1116 if (skb2->nh.raw < skb2->data ||
1117 skb2->nh.raw > skb2->tail) {
1118 if (net_ratelimit())
1119 printk(KERN_CRIT "protocol %04x is "
1121 skb2->protocol, dev->name);
1122 skb2->nh.raw = skb2->data;
1125 skb2->h.raw = skb2->nh.raw;
1126 skb2->pkt_type = PACKET_OUTGOING;
1127 ptype->func(skb2, skb->dev, ptype);
1134 * Invalidate hardware checksum when packet is to be mangled, and
1135 * complete checksum manually on outgoing path.
1137 int skb_checksum_help(struct sk_buff **pskb, int inward)
1140 int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1143 (*pskb)->ip_summed = CHECKSUM_NONE;
1147 if (skb_shared(*pskb) || skb_cloned(*pskb)) {
1148 struct sk_buff *newskb = skb_copy(*pskb, GFP_ATOMIC);
1154 skb_set_owner_w(newskb, (*pskb)->sk);
1159 if (offset > (int)(*pskb)->len)
1161 csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1163 offset = (*pskb)->tail - (*pskb)->h.raw;
1166 if ((*pskb)->csum + 2 > offset)
1169 *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1170 (*pskb)->ip_summed = CHECKSUM_NONE;
1175 #ifdef CONFIG_HIGHMEM
1176 /* Actually, we should eliminate this check as soon as we know, that:
1177 * 1. IOMMU is present and allows to map all the memory.
1178 * 2. No high memory really exists on this machine.
1181 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1185 if (dev->features & NETIF_F_HIGHDMA)
1188 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1189 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1195 #define illegal_highdma(dev, skb) (0)
1198 extern void skb_release_data(struct sk_buff *);
1200 /* Keep head the same: replace data */
1201 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1206 struct skb_shared_info *ninfo;
1207 int headerlen = skb->data - skb->head;
1208 int expand = (skb->tail + skb->data_len) - skb->end;
1210 if (skb_shared(skb))
1216 size = skb->end - skb->head + expand;
1217 size = SKB_DATA_ALIGN(size);
1218 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1222 /* Copy entire thing */
1223 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1227 ninfo = (struct skb_shared_info*)(data + size);
1228 atomic_set(&ninfo->dataref, 1);
1229 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1230 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1231 ninfo->nr_frags = 0;
1232 ninfo->frag_list = NULL;
1234 /* Offset between the two in bytes */
1235 offset = data - skb->head;
1237 /* Free old data. */
1238 skb_release_data(skb);
1241 skb->end = data + size;
1243 /* Set up new pointers */
1244 skb->h.raw += offset;
1245 skb->nh.raw += offset;
1246 skb->mac.raw += offset;
1247 skb->tail += offset;
1248 skb->data += offset;
1250 /* We are no longer a clone, even if we were. */
1253 skb->tail += skb->data_len;
1258 #define HARD_TX_LOCK_BH(dev, cpu) { \
1259 if ((dev->features & NETIF_F_LLTX) == 0) { \
1260 spin_lock_bh(&dev->xmit_lock); \
1261 dev->xmit_lock_owner = cpu; \
1265 #define HARD_TX_UNLOCK_BH(dev) { \
1266 if ((dev->features & NETIF_F_LLTX) == 0) { \
1267 dev->xmit_lock_owner = -1; \
1268 spin_unlock_bh(&dev->xmit_lock); \
1272 static inline void qdisc_run(struct net_device *dev)
1274 while (!netif_queue_stopped(dev) &&
1275 qdisc_restart(dev)<0)
1280 * dev_queue_xmit - transmit a buffer
1281 * @skb: buffer to transmit
1283 * Queue a buffer for transmission to a network device. The caller must
1284 * have set the device and priority and built the buffer before calling
1285 * this function. The function can be called from an interrupt.
1287 * A negative errno code is returned on a failure. A success does not
1288 * guarantee the frame will be transmitted as it may be dropped due
1289 * to congestion or traffic shaping.
1292 int dev_queue_xmit(struct sk_buff *skb)
1294 struct net_device *dev = skb->dev;
1298 if (skb_shinfo(skb)->frag_list &&
1299 !(dev->features & NETIF_F_FRAGLIST) &&
1300 __skb_linearize(skb, GFP_ATOMIC))
1303 /* Fragmented skb is linearized if device does not support SG,
1304 * or if at least one of fragments is in highmem and device
1305 * does not support DMA from it.
1307 if (skb_shinfo(skb)->nr_frags &&
1308 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1309 __skb_linearize(skb, GFP_ATOMIC))
1312 /* If packet is not checksummed and device does not support
1313 * checksumming for this protocol, complete checksumming here.
1315 if (skb->ip_summed == CHECKSUM_HW &&
1316 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1317 (!(dev->features & NETIF_F_IP_CSUM) ||
1318 skb->protocol != htons(ETH_P_IP))))
1319 if (skb_checksum_help(&skb, 0))
1323 /* Updates of qdisc are serialized by queue_lock.
1324 * The struct Qdisc which is pointed to by qdisc is now a
1325 * rcu structure - it may be accessed without acquiring
1326 * a lock (but the structure may be stale.) The freeing of the
1327 * qdisc will be deferred until it's known that there are no
1328 * more references to it.
1330 * If the qdisc has an enqueue function, we still need to
1331 * hold the queue_lock before calling it, since queue_lock
1332 * also serializes access to the device queue.
1336 smp_read_barrier_depends();
1337 #ifdef CONFIG_NET_CLS_ACT
1338 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1341 /* Grab device queue */
1342 spin_lock_bh(&dev->queue_lock);
1344 rc = q->enqueue(skb, q);
1348 spin_unlock_bh(&dev->queue_lock);
1350 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1355 /* The device has no queue. Common case for software devices:
1356 loopback, all the sorts of tunnels...
1358 Really, it is unlikely that xmit_lock protection is necessary here.
1359 (f.e. loopback and IP tunnels are clean ignoring statistics
1361 However, it is possible, that they rely on protection
1364 Check this and shot the lock. It is not prone from deadlocks.
1365 Either shot noqueue qdisc, it is even simpler 8)
1367 if (dev->flags & IFF_UP) {
1368 int cpu = get_cpu();
1370 if (dev->xmit_lock_owner != cpu) {
1372 HARD_TX_LOCK_BH(dev, cpu);
1375 if (!netif_queue_stopped(dev)) {
1377 dev_queue_xmit_nit(skb, dev);
1380 if (!dev->hard_start_xmit(skb, dev)) {
1381 HARD_TX_UNLOCK_BH(dev);
1385 HARD_TX_UNLOCK_BH(dev);
1386 if (net_ratelimit())
1387 printk(KERN_CRIT "Virtual device %s asks to "
1388 "queue packet!\n", dev->name);
1392 /* Recursion is detected! It is possible,
1394 if (net_ratelimit())
1395 printk(KERN_CRIT "Dead loop on virtual device "
1396 "%s, fix it urgently!\n", dev->name);
1408 /*=======================================================================
1410 =======================================================================*/
1412 int netdev_max_backlog = 300;
1413 int weight_p = 64; /* old backlog weight */
1414 /* These numbers are selected based on intuition and some
1415 * experimentatiom, if you have more scientific way of doing this
1416 * please go ahead and fix things.
1418 int no_cong_thresh = 10;
1423 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1426 #ifdef CONFIG_NET_HW_FLOWCONTROL
1427 atomic_t netdev_dropping = ATOMIC_INIT(0);
1428 static unsigned long netdev_fc_mask = 1;
1429 unsigned long netdev_fc_xoff;
1430 spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1434 void (*stimul)(struct net_device *);
1435 struct net_device *dev;
1436 } netdev_fc_slots[BITS_PER_LONG];
1438 int netdev_register_fc(struct net_device *dev,
1439 void (*stimul)(struct net_device *dev))
1442 unsigned long flags;
1444 spin_lock_irqsave(&netdev_fc_lock, flags);
1445 if (netdev_fc_mask != ~0UL) {
1446 bit = ffz(netdev_fc_mask);
1447 netdev_fc_slots[bit].stimul = stimul;
1448 netdev_fc_slots[bit].dev = dev;
1449 set_bit(bit, &netdev_fc_mask);
1450 clear_bit(bit, &netdev_fc_xoff);
1452 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1456 void netdev_unregister_fc(int bit)
1458 unsigned long flags;
1460 spin_lock_irqsave(&netdev_fc_lock, flags);
1462 netdev_fc_slots[bit].stimul = NULL;
1463 netdev_fc_slots[bit].dev = NULL;
1464 clear_bit(bit, &netdev_fc_mask);
1465 clear_bit(bit, &netdev_fc_xoff);
1467 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1470 static void netdev_wakeup(void)
1474 spin_lock(&netdev_fc_lock);
1475 xoff = netdev_fc_xoff;
1480 netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1482 spin_unlock(&netdev_fc_lock);
1486 static void get_sample_stats(int cpu)
1492 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1493 int blog = sd->input_pkt_queue.qlen;
1494 int avg_blog = sd->avg_blog;
1496 avg_blog = (avg_blog >> 1) + (blog >> 1);
1498 if (avg_blog > mod_cong) {
1499 /* Above moderate congestion levels. */
1500 sd->cng_level = NET_RX_CN_HIGH;
1503 rq = rd % netdev_max_backlog;
1504 if (rq < avg_blog) /* unlucky bastard */
1505 sd->cng_level = NET_RX_DROP;
1507 } else if (avg_blog > lo_cong) {
1508 sd->cng_level = NET_RX_CN_MOD;
1511 rq = rd % netdev_max_backlog;
1512 if (rq < avg_blog) /* unlucky bastard */
1513 sd->cng_level = NET_RX_CN_HIGH;
1515 } else if (avg_blog > no_cong)
1516 sd->cng_level = NET_RX_CN_LOW;
1517 else /* no congestion */
1518 sd->cng_level = NET_RX_SUCCESS;
1520 sd->avg_blog = avg_blog;
1523 #ifdef OFFLINE_SAMPLE
1524 static void sample_queue(unsigned long dummy)
1526 /* 10 ms 0r 1ms -- i don't care -- JHS */
1528 int cpu = smp_processor_id();
1530 get_sample_stats(cpu);
1531 next_tick += jiffies;
1532 mod_timer(&samp_timer, next_tick);
1538 * netif_rx - post buffer to the network code
1539 * @skb: buffer to post
1541 * This function receives a packet from a device driver and queues it for
1542 * the upper (protocol) levels to process. It always succeeds. The buffer
1543 * may be dropped during processing for congestion control or by the
1547 * NET_RX_SUCCESS (no congestion)
1548 * NET_RX_CN_LOW (low congestion)
1549 * NET_RX_CN_MOD (moderate congestion)
1550 * NET_RX_CN_HIGH (high congestion)
1551 * NET_RX_DROP (packet was dropped)
1555 int netif_rx(struct sk_buff *skb)
1558 struct softnet_data *queue;
1559 unsigned long flags;
1561 #ifdef CONFIG_NETPOLL_RX
1562 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1568 if (!skb->stamp.tv_sec)
1569 net_timestamp(&skb->stamp);
1572 * The code is rearranged so that the path is the most
1573 * short when CPU is congested, but is still operating.
1575 local_irq_save(flags);
1576 this_cpu = smp_processor_id();
1577 queue = &__get_cpu_var(softnet_data);
1579 __get_cpu_var(netdev_rx_stat).total++;
1580 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1581 if (queue->input_pkt_queue.qlen) {
1582 if (queue->throttle)
1587 __skb_queue_tail(&queue->input_pkt_queue, skb);
1588 #ifndef OFFLINE_SAMPLE
1589 get_sample_stats(this_cpu);
1591 local_irq_restore(flags);
1592 return queue->cng_level;
1595 if (queue->throttle) {
1596 queue->throttle = 0;
1597 #ifdef CONFIG_NET_HW_FLOWCONTROL
1598 if (atomic_dec_and_test(&netdev_dropping))
1603 netif_rx_schedule(&queue->backlog_dev);
1607 if (!queue->throttle) {
1608 queue->throttle = 1;
1609 __get_cpu_var(netdev_rx_stat).throttled++;
1610 #ifdef CONFIG_NET_HW_FLOWCONTROL
1611 atomic_inc(&netdev_dropping);
1616 __get_cpu_var(netdev_rx_stat).dropped++;
1617 local_irq_restore(flags);
1623 static __inline__ void skb_bond(struct sk_buff *skb)
1625 struct net_device *dev = skb->dev;
1628 skb->real_dev = skb->dev;
1629 skb->dev = dev->master;
1633 static void net_tx_action(struct softirq_action *h)
1635 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1637 if (sd->completion_queue) {
1638 struct sk_buff *clist;
1640 local_irq_disable();
1641 clist = sd->completion_queue;
1642 sd->completion_queue = NULL;
1646 struct sk_buff *skb = clist;
1647 clist = clist->next;
1649 BUG_TRAP(!atomic_read(&skb->users));
1654 if (sd->output_queue) {
1655 struct net_device *head;
1657 local_irq_disable();
1658 head = sd->output_queue;
1659 sd->output_queue = NULL;
1663 struct net_device *dev = head;
1664 head = head->next_sched;
1666 smp_mb__before_clear_bit();
1667 clear_bit(__LINK_STATE_SCHED, &dev->state);
1669 if (spin_trylock(&dev->queue_lock)) {
1671 spin_unlock(&dev->queue_lock);
1673 netif_schedule(dev);
1679 static __inline__ int deliver_skb(struct sk_buff *skb,
1680 struct packet_type *pt_prev, int last)
1682 atomic_inc(&skb->users);
1683 return pt_prev->func(skb, skb->dev, pt_prev);
1687 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1688 int (*br_handle_frame_hook)(struct sk_buff *skb);
1690 static __inline__ int handle_bridge(struct sk_buff *skb,
1691 struct packet_type *pt_prev)
1693 int ret = NET_RX_DROP;
1695 ret = deliver_skb(skb, pt_prev, 0);
1702 static inline int __handle_bridge(struct sk_buff *skb,
1703 struct packet_type **pt_prev, int *ret)
1705 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
1706 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1707 *ret = handle_bridge(skb, *pt_prev);
1708 if (br_handle_frame_hook(skb) == 0)
1718 #ifdef CONFIG_NET_CLS_ACT
1719 /* TODO: Maybe we should just force sch_ingress to be compiled in
1720 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1721 * a compare and 2 stores extra right now if we dont have it on
1722 * but have CONFIG_NET_CLS_ACT
1723 * NOTE: This doesnt stop any functionality; if you dont have
1724 * the ingress scheduler, you just cant add policies on ingress.
1727 int ing_filter(struct sk_buff *skb)
1730 struct net_device *dev = skb->dev;
1731 int result = TC_ACT_OK;
1733 if (dev->qdisc_ingress) {
1734 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1735 if (MAX_RED_LOOP < ttl++) {
1736 printk("Redir loop detected Dropping packet (%s->%s)\n",
1737 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1741 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1743 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1744 if (NULL == skb->input_dev) {
1745 skb->input_dev = skb->dev;
1746 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1748 spin_lock(&dev->ingress_lock);
1749 if ((q = dev->qdisc_ingress) != NULL)
1750 result = q->enqueue(skb, q);
1751 spin_unlock(&dev->ingress_lock);
1759 int netif_receive_skb(struct sk_buff *skb)
1761 struct packet_type *ptype, *pt_prev;
1762 int ret = NET_RX_DROP;
1763 unsigned short type;
1765 #ifdef CONFIG_NETPOLL_RX
1766 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1772 if (!skb->stamp.tv_sec)
1773 net_timestamp(&skb->stamp);
1777 __get_cpu_var(netdev_rx_stat).total++;
1779 skb->h.raw = skb->nh.raw = skb->data;
1780 skb->mac_len = skb->nh.raw - skb->mac.raw;
1783 #ifdef CONFIG_NET_CLS_ACT
1784 if (skb->tc_verd & TC_NCLS) {
1785 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1792 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1793 if (!ptype->dev || ptype->dev == skb->dev) {
1795 ret = deliver_skb(skb, pt_prev, 0);
1800 #ifdef CONFIG_NET_CLS_ACT
1802 atomic_inc(&skb->users);
1803 ret = pt_prev->func(skb, skb->dev, pt_prev);
1804 pt_prev = NULL; /* noone else should process this after*/
1806 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1809 ret = ing_filter(skb);
1811 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1820 handle_diverter(skb);
1822 if (__handle_bridge(skb, &pt_prev, &ret))
1825 type = skb->protocol;
1826 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1827 if (ptype->type == type &&
1828 (!ptype->dev || ptype->dev == skb->dev)) {
1830 ret = deliver_skb(skb, pt_prev, 0);
1836 ret = pt_prev->func(skb, skb->dev, pt_prev);
1839 /* Jamal, now you will not able to escape explaining
1840 * me how you were going to use this. :-)
1850 static int process_backlog(struct net_device *backlog_dev, int *budget)
1853 int quota = min(backlog_dev->quota, *budget);
1854 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1855 unsigned long start_time = jiffies;
1858 struct sk_buff *skb;
1859 struct net_device *dev;
1861 local_irq_disable();
1862 skb = __skb_dequeue(&queue->input_pkt_queue);
1869 netif_receive_skb(skb);
1875 if (work >= quota || jiffies - start_time > 1)
1878 #ifdef CONFIG_NET_HW_FLOWCONTROL
1879 if (queue->throttle &&
1880 queue->input_pkt_queue.qlen < no_cong_thresh ) {
1881 queue->throttle = 0;
1882 if (atomic_dec_and_test(&netdev_dropping)) {
1890 backlog_dev->quota -= work;
1895 backlog_dev->quota -= work;
1898 list_del(&backlog_dev->poll_list);
1899 smp_mb__before_clear_bit();
1900 netif_poll_enable(backlog_dev);
1902 if (queue->throttle) {
1903 queue->throttle = 0;
1904 #ifdef CONFIG_NET_HW_FLOWCONTROL
1905 if (atomic_dec_and_test(&netdev_dropping))
1913 static void net_rx_action(struct softirq_action *h)
1915 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1916 unsigned long start_time = jiffies;
1917 int budget = netdev_max_backlog;
1920 local_irq_disable();
1922 while (!list_empty(&queue->poll_list)) {
1923 struct net_device *dev;
1925 if (budget <= 0 || jiffies - start_time > 1)
1930 dev = list_entry(queue->poll_list.next,
1931 struct net_device, poll_list);
1933 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1934 local_irq_disable();
1935 list_del(&dev->poll_list);
1936 list_add_tail(&dev->poll_list, &queue->poll_list);
1938 dev->quota += dev->weight;
1940 dev->quota = dev->weight;
1943 local_irq_disable();
1951 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1952 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1956 static gifconf_func_t * gifconf_list [NPROTO];
1959 * register_gifconf - register a SIOCGIF handler
1960 * @family: Address family
1961 * @gifconf: Function handler
1963 * Register protocol dependent address dumping routines. The handler
1964 * that is passed must not be freed or reused until it has been replaced
1965 * by another handler.
1967 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1969 if (family >= NPROTO)
1971 gifconf_list[family] = gifconf;
1977 * Map an interface index to its name (SIOCGIFNAME)
1981 * We need this ioctl for efficient implementation of the
1982 * if_indextoname() function required by the IPv6 API. Without
1983 * it, we would have to search all the interfaces to find a
1987 static int dev_ifname(struct ifreq __user *arg)
1989 struct net_device *dev;
1993 * Fetch the caller's info block.
1996 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1999 read_lock(&dev_base_lock);
2000 dev = __dev_get_by_index(ifr.ifr_ifindex);
2002 read_unlock(&dev_base_lock);
2006 strcpy(ifr.ifr_name, dev->name);
2007 read_unlock(&dev_base_lock);
2009 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2015 * Perform a SIOCGIFCONF call. This structure will change
2016 * size eventually, and there is nothing I can do about it.
2017 * Thus we will need a 'compatibility mode'.
2020 static int dev_ifconf(char __user *arg)
2023 struct net_device *dev;
2030 * Fetch the caller's info block.
2033 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2040 * Loop over the interfaces, and write an info block for each.
2044 for (dev = dev_base; dev; dev = dev->next) {
2045 for (i = 0; i < NPROTO; i++) {
2046 if (gifconf_list[i]) {
2049 done = gifconf_list[i](dev, NULL, 0);
2051 done = gifconf_list[i](dev, pos + total,
2061 * All done. Write the updated control block back to the caller.
2063 ifc.ifc_len = total;
2066 * Both BSD and Solaris return 0 here, so we do too.
2068 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2071 #ifdef CONFIG_PROC_FS
2073 * This is invoked by the /proc filesystem handler to display a device
2076 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2078 struct net_device *dev;
2081 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2083 return i == pos ? dev : NULL;
2086 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2088 read_lock(&dev_base_lock);
2089 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2092 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2095 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2098 void dev_seq_stop(struct seq_file *seq, void *v)
2100 read_unlock(&dev_base_lock);
2103 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2105 if (dev->get_stats) {
2106 struct net_device_stats *stats = dev->get_stats(dev);
2108 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2109 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2110 dev->name, stats->rx_bytes, stats->rx_packets,
2112 stats->rx_dropped + stats->rx_missed_errors,
2113 stats->rx_fifo_errors,
2114 stats->rx_length_errors + stats->rx_over_errors +
2115 stats->rx_crc_errors + stats->rx_frame_errors,
2116 stats->rx_compressed, stats->multicast,
2117 stats->tx_bytes, stats->tx_packets,
2118 stats->tx_errors, stats->tx_dropped,
2119 stats->tx_fifo_errors, stats->collisions,
2120 stats->tx_carrier_errors +
2121 stats->tx_aborted_errors +
2122 stats->tx_window_errors +
2123 stats->tx_heartbeat_errors,
2124 stats->tx_compressed);
2126 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2130 * Called from the PROCfs module. This now uses the new arbitrary sized
2131 * /proc/net interface to create /proc/net/dev
2133 static int dev_seq_show(struct seq_file *seq, void *v)
2135 if (v == SEQ_START_TOKEN)
2136 seq_puts(seq, "Inter-| Receive "
2138 " face |bytes packets errs drop fifo frame "
2139 "compressed multicast|bytes packets errs "
2140 "drop fifo colls carrier compressed\n");
2142 dev_seq_printf_stats(seq, v);
2146 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2148 struct netif_rx_stats *rc = NULL;
2150 while (*pos < NR_CPUS)
2151 if (cpu_online(*pos)) {
2152 rc = &per_cpu(netdev_rx_stat, *pos);
2159 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2161 return softnet_get_online(pos);
2164 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2167 return softnet_get_online(pos);
2170 static void softnet_seq_stop(struct seq_file *seq, void *v)
2174 static int softnet_seq_show(struct seq_file *seq, void *v)
2176 struct netif_rx_stats *s = v;
2178 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2179 s->total, s->dropped, s->time_squeeze, s->throttled,
2180 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2181 s->fastroute_deferred_out,
2183 s->fastroute_latency_reduction
2191 static struct seq_operations dev_seq_ops = {
2192 .start = dev_seq_start,
2193 .next = dev_seq_next,
2194 .stop = dev_seq_stop,
2195 .show = dev_seq_show,
2198 static int dev_seq_open(struct inode *inode, struct file *file)
2200 return seq_open(file, &dev_seq_ops);
2203 static struct file_operations dev_seq_fops = {
2204 .owner = THIS_MODULE,
2205 .open = dev_seq_open,
2207 .llseek = seq_lseek,
2208 .release = seq_release,
2211 static struct seq_operations softnet_seq_ops = {
2212 .start = softnet_seq_start,
2213 .next = softnet_seq_next,
2214 .stop = softnet_seq_stop,
2215 .show = softnet_seq_show,
2218 static int softnet_seq_open(struct inode *inode, struct file *file)
2220 return seq_open(file, &softnet_seq_ops);
2223 static struct file_operations softnet_seq_fops = {
2224 .owner = THIS_MODULE,
2225 .open = softnet_seq_open,
2227 .llseek = seq_lseek,
2228 .release = seq_release,
2232 extern int wireless_proc_init(void);
2234 #define wireless_proc_init() 0
2237 static int __init dev_proc_init(void)
2241 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2243 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2245 if (wireless_proc_init())
2251 proc_net_remove("softnet_stat");
2253 proc_net_remove("dev");
2257 #define dev_proc_init() 0
2258 #endif /* CONFIG_PROC_FS */
2262 * netdev_set_master - set up master/slave pair
2263 * @slave: slave device
2264 * @master: new master device
2266 * Changes the master device of the slave. Pass %NULL to break the
2267 * bonding. The caller must hold the RTNL semaphore. On a failure
2268 * a negative errno code is returned. On success the reference counts
2269 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2270 * function returns zero.
2272 int netdev_set_master(struct net_device *slave, struct net_device *master)
2274 struct net_device *old = slave->master;
2284 slave->master = master;
2292 slave->flags |= IFF_SLAVE;
2294 slave->flags &= ~IFF_SLAVE;
2296 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2301 * dev_set_promiscuity - update promiscuity count on a device
2305 * Add or remove promsicuity from a device. While the count in the device
2306 * remains above zero the interface remains promiscuous. Once it hits zero
2307 * the device reverts back to normal filtering operation. A negative inc
2308 * value is used to drop promiscuity on the device.
2310 void dev_set_promiscuity(struct net_device *dev, int inc)
2312 unsigned short old_flags = dev->flags;
2314 dev->flags |= IFF_PROMISC;
2315 if ((dev->promiscuity += inc) == 0)
2316 dev->flags &= ~IFF_PROMISC;
2317 if (dev->flags ^ old_flags) {
2319 printk(KERN_INFO "device %s %s promiscuous mode\n",
2320 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2326 * dev_set_allmulti - update allmulti count on a device
2330 * Add or remove reception of all multicast frames to a device. While the
2331 * count in the device remains above zero the interface remains listening
2332 * to all interfaces. Once it hits zero the device reverts back to normal
2333 * filtering operation. A negative @inc value is used to drop the counter
2334 * when releasing a resource needing all multicasts.
2337 void dev_set_allmulti(struct net_device *dev, int inc)
2339 unsigned short old_flags = dev->flags;
2341 dev->flags |= IFF_ALLMULTI;
2342 if ((dev->allmulti += inc) == 0)
2343 dev->flags &= ~IFF_ALLMULTI;
2344 if (dev->flags ^ old_flags)
2348 unsigned dev_get_flags(const struct net_device *dev)
2352 flags = (dev->flags & ~(IFF_PROMISC |
2355 (dev->gflags & (IFF_PROMISC |
2358 if (netif_running(dev) && netif_carrier_ok(dev))
2359 flags |= IFF_RUNNING;
2364 int dev_change_flags(struct net_device *dev, unsigned flags)
2367 int old_flags = dev->flags;
2370 * Set the flags on our device.
2373 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2374 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2376 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2380 * Load in the correct multicast list now the flags have changed.
2386 * Have we downed the interface. We handle IFF_UP ourselves
2387 * according to user attempts to set it, rather than blindly
2392 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2393 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2399 if (dev->flags & IFF_UP &&
2400 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2402 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2404 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2405 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2406 dev->gflags ^= IFF_PROMISC;
2407 dev_set_promiscuity(dev, inc);
2410 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2411 is important. Some (broken) drivers set IFF_PROMISC, when
2412 IFF_ALLMULTI is requested not asking us and not reporting.
2414 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2415 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2416 dev->gflags ^= IFF_ALLMULTI;
2417 dev_set_allmulti(dev, inc);
2420 if (old_flags ^ dev->flags)
2421 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2426 int dev_set_mtu(struct net_device *dev, int new_mtu)
2430 if (new_mtu == dev->mtu)
2433 /* MTU must be positive. */
2437 if (!netif_device_present(dev))
2441 if (dev->change_mtu)
2442 err = dev->change_mtu(dev, new_mtu);
2445 if (!err && dev->flags & IFF_UP)
2446 notifier_call_chain(&netdev_chain,
2447 NETDEV_CHANGEMTU, dev);
2453 * Perform the SIOCxIFxxx calls.
2455 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2458 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2464 case SIOCGIFFLAGS: /* Get interface flags */
2465 ifr->ifr_flags = dev_get_flags(dev);
2468 case SIOCSIFFLAGS: /* Set interface flags */
2469 return dev_change_flags(dev, ifr->ifr_flags);
2471 case SIOCGIFMETRIC: /* Get the metric on the interface
2472 (currently unused) */
2473 ifr->ifr_metric = 0;
2476 case SIOCSIFMETRIC: /* Set the metric on the interface
2477 (currently unused) */
2480 case SIOCGIFMTU: /* Get the MTU of a device */
2481 ifr->ifr_mtu = dev->mtu;
2484 case SIOCSIFMTU: /* Set the MTU of a device */
2485 return dev_set_mtu(dev, ifr->ifr_mtu);
2488 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2489 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2490 ifr->ifr_hwaddr.sa_family = dev->type;
2494 if (!dev->set_mac_address)
2496 if (ifr->ifr_hwaddr.sa_family != dev->type)
2498 if (!netif_device_present(dev))
2500 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2502 notifier_call_chain(&netdev_chain,
2503 NETDEV_CHANGEADDR, dev);
2506 case SIOCSIFHWBROADCAST:
2507 if (ifr->ifr_hwaddr.sa_family != dev->type)
2509 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2510 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2511 notifier_call_chain(&netdev_chain,
2512 NETDEV_CHANGEADDR, dev);
2516 ifr->ifr_map.mem_start = dev->mem_start;
2517 ifr->ifr_map.mem_end = dev->mem_end;
2518 ifr->ifr_map.base_addr = dev->base_addr;
2519 ifr->ifr_map.irq = dev->irq;
2520 ifr->ifr_map.dma = dev->dma;
2521 ifr->ifr_map.port = dev->if_port;
2525 if (dev->set_config) {
2526 if (!netif_device_present(dev))
2528 return dev->set_config(dev, &ifr->ifr_map);
2533 if (!dev->set_multicast_list ||
2534 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2536 if (!netif_device_present(dev))
2538 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2542 if (!dev->set_multicast_list ||
2543 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2545 if (!netif_device_present(dev))
2547 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2551 ifr->ifr_ifindex = dev->ifindex;
2555 ifr->ifr_qlen = dev->tx_queue_len;
2559 if (ifr->ifr_qlen < 0)
2561 dev->tx_queue_len = ifr->ifr_qlen;
2565 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2566 return dev_change_name(dev, ifr->ifr_newname);
2569 * Unknown or private ioctl
2573 if ((cmd >= SIOCDEVPRIVATE &&
2574 cmd <= SIOCDEVPRIVATE + 15) ||
2575 cmd == SIOCBONDENSLAVE ||
2576 cmd == SIOCBONDRELEASE ||
2577 cmd == SIOCBONDSETHWADDR ||
2578 cmd == SIOCBONDSLAVEINFOQUERY ||
2579 cmd == SIOCBONDINFOQUERY ||
2580 cmd == SIOCBONDCHANGEACTIVE ||
2581 cmd == SIOCGMIIPHY ||
2582 cmd == SIOCGMIIREG ||
2583 cmd == SIOCSMIIREG ||
2584 cmd == SIOCBRADDIF ||
2585 cmd == SIOCBRDELIF ||
2586 cmd == SIOCWANDEV) {
2588 if (dev->do_ioctl) {
2589 if (netif_device_present(dev))
2590 err = dev->do_ioctl(dev, ifr,
2603 * This function handles all "interface"-type I/O control requests. The actual
2604 * 'doing' part of this is dev_ifsioc above.
2608 * dev_ioctl - network device ioctl
2609 * @cmd: command to issue
2610 * @arg: pointer to a struct ifreq in user space
2612 * Issue ioctl functions to devices. This is normally called by the
2613 * user space syscall interfaces but can sometimes be useful for
2614 * other purposes. The return value is the return from the syscall if
2615 * positive or a negative errno code on error.
2618 int dev_ioctl(unsigned int cmd, void __user *arg)
2624 /* One special case: SIOCGIFCONF takes ifconf argument
2625 and requires shared lock, because it sleeps writing
2629 if (cmd == SIOCGIFCONF) {
2631 ret = dev_ifconf((char __user *) arg);
2635 if (cmd == SIOCGIFNAME)
2636 return dev_ifname((struct ifreq __user *)arg);
2638 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2641 ifr.ifr_name[IFNAMSIZ-1] = 0;
2643 colon = strchr(ifr.ifr_name, ':');
2648 * See which interface the caller is talking about.
2653 * These ioctl calls:
2654 * - can be done by all.
2655 * - atomic and do not require locking.
2666 dev_load(ifr.ifr_name);
2667 read_lock(&dev_base_lock);
2668 ret = dev_ifsioc(&ifr, cmd);
2669 read_unlock(&dev_base_lock);
2673 if (copy_to_user(arg, &ifr,
2674 sizeof(struct ifreq)))
2680 dev_load(ifr.ifr_name);
2682 ret = dev_ethtool(&ifr);
2687 if (copy_to_user(arg, &ifr,
2688 sizeof(struct ifreq)))
2694 * These ioctl calls:
2695 * - require superuser power.
2696 * - require strict serialization.
2702 if (!capable(CAP_NET_ADMIN))
2704 dev_load(ifr.ifr_name);
2706 ret = dev_ifsioc(&ifr, cmd);
2711 if (copy_to_user(arg, &ifr,
2712 sizeof(struct ifreq)))
2718 * These ioctl calls:
2719 * - require superuser power.
2720 * - require strict serialization.
2721 * - do not return a value
2731 case SIOCSIFHWBROADCAST:
2734 case SIOCBONDENSLAVE:
2735 case SIOCBONDRELEASE:
2736 case SIOCBONDSETHWADDR:
2737 case SIOCBONDSLAVEINFOQUERY:
2738 case SIOCBONDINFOQUERY:
2739 case SIOCBONDCHANGEACTIVE:
2742 if (!capable(CAP_NET_ADMIN))
2744 dev_load(ifr.ifr_name);
2746 ret = dev_ifsioc(&ifr, cmd);
2751 /* Get the per device memory space. We can add this but
2752 * currently do not support it */
2754 /* Set the per device memory buffer space.
2755 * Not applicable in our case */
2760 * Unknown or private ioctl.
2763 if (cmd == SIOCWANDEV ||
2764 (cmd >= SIOCDEVPRIVATE &&
2765 cmd <= SIOCDEVPRIVATE + 15)) {
2766 dev_load(ifr.ifr_name);
2768 ret = dev_ifsioc(&ifr, cmd);
2770 if (!ret && copy_to_user(arg, &ifr,
2771 sizeof(struct ifreq)))
2776 /* Take care of Wireless Extensions */
2777 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2778 /* If command is `set a parameter', or
2779 * `get the encoding parameters', check if
2780 * the user has the right to do it */
2781 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2782 if (!capable(CAP_NET_ADMIN))
2785 dev_load(ifr.ifr_name);
2787 /* Follow me in net/core/wireless.c */
2788 ret = wireless_process_ioctl(&ifr, cmd);
2790 if (!ret && IW_IS_GET(cmd) &&
2791 copy_to_user(arg, &ifr,
2792 sizeof(struct ifreq)))
2796 #endif /* WIRELESS_EXT */
2803 * dev_new_index - allocate an ifindex
2805 * Returns a suitable unique value for a new device interface
2806 * number. The caller must hold the rtnl semaphore or the
2807 * dev_base_lock to be sure it remains unique.
2809 int dev_new_index(void)
2815 if (!__dev_get_by_index(ifindex))
2820 static int dev_boot_phase = 1;
2822 /* Delayed registration/unregisteration */
2823 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2824 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2826 static inline void net_set_todo(struct net_device *dev)
2828 spin_lock(&net_todo_list_lock);
2829 list_add_tail(&dev->todo_list, &net_todo_list);
2830 spin_unlock(&net_todo_list_lock);
2834 * register_netdevice - register a network device
2835 * @dev: device to register
2837 * Take a completed network device structure and add it to the kernel
2838 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2839 * chain. 0 is returned on success. A negative errno code is returned
2840 * on a failure to set up the device, or if the name is a duplicate.
2842 * Callers must hold the rtnl semaphore. See the comment at the
2843 * end of Space.c for details about the locking. You may want
2844 * register_netdev() instead of this.
2847 * The locking appears insufficient to guarantee two parallel registers
2848 * will not get the same name.
2851 int register_netdevice(struct net_device *dev)
2853 struct hlist_head *head;
2854 struct hlist_node *p;
2857 BUG_ON(dev_boot_phase);
2860 /* When net_device's are persistent, this will be fatal. */
2861 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2863 spin_lock_init(&dev->queue_lock);
2864 spin_lock_init(&dev->xmit_lock);
2865 dev->xmit_lock_owner = -1;
2866 #ifdef CONFIG_NET_CLS_ACT
2867 spin_lock_init(&dev->ingress_lock);
2870 ret = alloc_divert_blk(dev);
2876 /* Init, if this function is available */
2878 ret = dev->init(dev);
2886 if (!dev_valid_name(dev->name)) {
2891 dev->ifindex = dev_new_index();
2892 if (dev->iflink == -1)
2893 dev->iflink = dev->ifindex;
2895 /* Check for existence of name */
2896 head = dev_name_hash(dev->name);
2897 hlist_for_each(p, head) {
2898 struct net_device *d
2899 = hlist_entry(p, struct net_device, name_hlist);
2900 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2906 /* Fix illegal SG+CSUM combinations. */
2907 if ((dev->features & NETIF_F_SG) &&
2908 !(dev->features & (NETIF_F_IP_CSUM |
2910 NETIF_F_HW_CSUM))) {
2911 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2913 dev->features &= ~NETIF_F_SG;
2917 * nil rebuild_header routine,
2918 * that should be never called and used as just bug trap.
2921 if (!dev->rebuild_header)
2922 dev->rebuild_header = default_rebuild_header;
2925 * Default initial state at registry is that the
2926 * device is present.
2929 set_bit(__LINK_STATE_PRESENT, &dev->state);
2932 dev_init_scheduler(dev);
2933 write_lock_bh(&dev_base_lock);
2935 dev_tail = &dev->next;
2936 hlist_add_head(&dev->name_hlist, head);
2937 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2939 dev->reg_state = NETREG_REGISTERING;
2940 write_unlock_bh(&dev_base_lock);
2942 /* Notify protocols, that a new device appeared. */
2943 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2945 /* Finish registration after unlock */
2952 free_divert_blk(dev);
2957 * netdev_wait_allrefs - wait until all references are gone.
2959 * This is called when unregistering network devices.
2961 * Any protocol or device that holds a reference should register
2962 * for netdevice notification, and cleanup and put back the
2963 * reference if they receive an UNREGISTER event.
2964 * We can get stuck here if buggy protocols don't correctly
2967 static void netdev_wait_allrefs(struct net_device *dev)
2969 unsigned long rebroadcast_time, warning_time;
2971 rebroadcast_time = warning_time = jiffies;
2972 while (atomic_read(&dev->refcnt) != 0) {
2973 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2976 /* Rebroadcast unregister notification */
2977 notifier_call_chain(&netdev_chain,
2978 NETDEV_UNREGISTER, dev);
2980 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2982 /* We must not have linkwatch events
2983 * pending on unregister. If this
2984 * happens, we simply run the queue
2985 * unscheduled, resulting in a noop
2988 linkwatch_run_queue();
2993 rebroadcast_time = jiffies;
2996 current->state = TASK_INTERRUPTIBLE;
2997 schedule_timeout(HZ / 4);
2999 if (time_after(jiffies, warning_time + 10 * HZ)) {
3000 printk(KERN_EMERG "unregister_netdevice: "
3001 "waiting for %s to become free. Usage "
3003 dev->name, atomic_read(&dev->refcnt));
3004 warning_time = jiffies;
3013 * register_netdevice(x1);
3014 * register_netdevice(x2);
3016 * unregister_netdevice(y1);
3017 * unregister_netdevice(y2);
3023 * We are invoked by rtnl_unlock() after it drops the semaphore.
3024 * This allows us to deal with problems:
3025 * 1) We can create/delete sysfs objects which invoke hotplug
3026 * without deadlocking with linkwatch via keventd.
3027 * 2) Since we run with the RTNL semaphore not held, we can sleep
3028 * safely in order to wait for the netdev refcnt to drop to zero.
3030 static DECLARE_MUTEX(net_todo_run_mutex);
3031 void netdev_run_todo(void)
3033 struct list_head list = LIST_HEAD_INIT(list);
3037 /* Need to guard against multiple cpu's getting out of order. */
3038 down(&net_todo_run_mutex);
3040 /* Not safe to do outside the semaphore. We must not return
3041 * until all unregister events invoked by the local processor
3042 * have been completed (either by this todo run, or one on
3045 if (list_empty(&net_todo_list))
3048 /* Snapshot list, allow later requests */
3049 spin_lock(&net_todo_list_lock);
3050 list_splice_init(&net_todo_list, &list);
3051 spin_unlock(&net_todo_list_lock);
3053 while (!list_empty(&list)) {
3054 struct net_device *dev
3055 = list_entry(list.next, struct net_device, todo_list);
3056 list_del(&dev->todo_list);
3058 switch(dev->reg_state) {
3059 case NETREG_REGISTERING:
3060 err = netdev_register_sysfs(dev);
3062 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3064 dev->reg_state = NETREG_REGISTERED;
3067 case NETREG_UNREGISTERING:
3068 netdev_unregister_sysfs(dev);
3069 dev->reg_state = NETREG_UNREGISTERED;
3071 netdev_wait_allrefs(dev);
3074 BUG_ON(atomic_read(&dev->refcnt));
3075 BUG_TRAP(!dev->ip_ptr);
3076 BUG_TRAP(!dev->ip6_ptr);
3077 BUG_TRAP(!dev->dn_ptr);
3080 /* It must be the very last action,
3081 * after this 'dev' may point to freed up memory.
3083 if (dev->destructor)
3084 dev->destructor(dev);
3088 printk(KERN_ERR "network todo '%s' but state %d\n",
3089 dev->name, dev->reg_state);
3095 up(&net_todo_run_mutex);
3099 * free_netdev - free network device
3102 * This function does the last stage of destroying an allocated device
3103 * interface. The reference to the device object is released.
3104 * If this is the last reference then it will be freed.
3106 void free_netdev(struct net_device *dev)
3109 /* Compatiablity with error handling in drivers */
3110 if (dev->reg_state == NETREG_UNINITIALIZED) {
3111 kfree((char *)dev - dev->padded);
3115 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3116 dev->reg_state = NETREG_RELEASED;
3118 /* will free via class release */
3119 class_device_put(&dev->class_dev);
3121 kfree((char *)dev - dev->padded);
3125 /* Synchronize with packet receive processing. */
3126 void synchronize_net(void)
3129 synchronize_kernel();
3133 * unregister_netdevice - remove device from the kernel
3136 * This function shuts down a device interface and removes it
3137 * from the kernel tables. On success 0 is returned, on a failure
3138 * a negative errno code is returned.
3140 * Callers must hold the rtnl semaphore. See the comment at the
3141 * end of Space.c for details about the locking. You may want
3142 * unregister_netdev() instead of this.
3145 int unregister_netdevice(struct net_device *dev)
3147 struct net_device *d, **dp;
3149 BUG_ON(dev_boot_phase);
3152 /* Some devices call without registering for initialization unwind. */
3153 if (dev->reg_state == NETREG_UNINITIALIZED) {
3154 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3155 "was registered\n", dev->name, dev);
3159 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3161 /* If device is running, close it first. */
3162 if (dev->flags & IFF_UP)
3165 /* And unlink it from device chain. */
3166 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3168 write_lock_bh(&dev_base_lock);
3169 hlist_del(&dev->name_hlist);
3170 hlist_del(&dev->index_hlist);
3171 if (dev_tail == &dev->next)
3174 write_unlock_bh(&dev_base_lock);
3179 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3184 dev->reg_state = NETREG_UNREGISTERING;
3188 /* Shutdown queueing discipline. */
3192 /* Notify protocols, that we are about to destroy
3193 this device. They should clean all the things.
3195 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3198 * Flush the multicast chain
3200 dev_mc_discard(dev);
3205 /* Notifier chain MUST detach us from master device. */
3206 BUG_TRAP(!dev->master);
3208 free_divert_blk(dev);
3210 /* Finish processing unregister after unlock */
3219 #ifdef CONFIG_HOTPLUG_CPU
3220 static int dev_cpu_callback(struct notifier_block *nfb,
3221 unsigned long action,
3224 struct sk_buff **list_skb;
3225 struct net_device **list_net;
3226 struct sk_buff *skb;
3227 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3228 struct softnet_data *sd, *oldsd;
3230 if (action != CPU_DEAD)
3233 local_irq_disable();
3234 cpu = smp_processor_id();
3235 sd = &per_cpu(softnet_data, cpu);
3236 oldsd = &per_cpu(softnet_data, oldcpu);
3238 /* Find end of our completion_queue. */
3239 list_skb = &sd->completion_queue;
3241 list_skb = &(*list_skb)->next;
3242 /* Append completion queue from offline CPU. */
3243 *list_skb = oldsd->completion_queue;
3244 oldsd->completion_queue = NULL;
3246 /* Find end of our output_queue. */
3247 list_net = &sd->output_queue;
3249 list_net = &(*list_net)->next_sched;
3250 /* Append output queue from offline CPU. */
3251 *list_net = oldsd->output_queue;
3252 oldsd->output_queue = NULL;
3254 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3257 /* Process offline CPU's input_pkt_queue */
3258 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3263 #endif /* CONFIG_HOTPLUG_CPU */
3267 * Initialize the DEV module. At boot time this walks the device list and
3268 * unhooks any devices that fail to initialise (normally hardware not
3269 * present) and leaves us with a valid list of present and active devices.
3274 * This is called single threaded during boot, so no need
3275 * to take the rtnl semaphore.
3277 static int __init net_dev_init(void)
3279 int i, rc = -ENOMEM;
3281 BUG_ON(!dev_boot_phase);
3283 if (dev_proc_init())
3286 if (netdev_sysfs_init())
3289 INIT_LIST_HEAD(&ptype_all);
3290 for (i = 0; i < 16; i++)
3291 INIT_LIST_HEAD(&ptype_base[i]);
3293 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3294 INIT_HLIST_HEAD(&dev_name_head[i]);
3296 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3297 INIT_HLIST_HEAD(&dev_index_head[i]);
3300 * Initialise the packet receive queues.
3303 for (i = 0; i < NR_CPUS; i++) {
3304 struct softnet_data *queue;
3306 queue = &per_cpu(softnet_data, i);
3307 skb_queue_head_init(&queue->input_pkt_queue);
3308 queue->throttle = 0;
3309 queue->cng_level = 0;
3310 queue->avg_blog = 10; /* arbitrary non-zero */
3311 queue->completion_queue = NULL;
3312 INIT_LIST_HEAD(&queue->poll_list);
3313 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3314 queue->backlog_dev.weight = weight_p;
3315 queue->backlog_dev.poll = process_backlog;
3316 atomic_set(&queue->backlog_dev.refcnt, 1);
3319 #ifdef OFFLINE_SAMPLE
3320 samp_timer.expires = jiffies + (10 * HZ);
3321 add_timer(&samp_timer);
3326 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3327 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3329 hotcpu_notifier(dev_cpu_callback, 0);
3337 subsys_initcall(net_dev_init);
3339 EXPORT_SYMBOL(__dev_get);
3340 EXPORT_SYMBOL(__dev_get_by_flags);
3341 EXPORT_SYMBOL(__dev_get_by_index);
3342 EXPORT_SYMBOL(__dev_get_by_name);
3343 EXPORT_SYMBOL(__dev_remove_pack);
3344 EXPORT_SYMBOL(__skb_linearize);
3345 EXPORT_SYMBOL(call_netdevice_notifiers);
3346 EXPORT_SYMBOL(dev_add_pack);
3347 EXPORT_SYMBOL(dev_alloc_name);
3348 EXPORT_SYMBOL(dev_close);
3349 EXPORT_SYMBOL(dev_get_by_flags);
3350 EXPORT_SYMBOL(dev_get_by_index);
3351 EXPORT_SYMBOL(dev_get_by_name);
3352 EXPORT_SYMBOL(dev_getbyhwaddr);
3353 EXPORT_SYMBOL(dev_ioctl);
3354 EXPORT_SYMBOL(dev_new_index);
3355 EXPORT_SYMBOL(dev_open);
3356 EXPORT_SYMBOL(dev_queue_xmit);
3357 EXPORT_SYMBOL(dev_queue_xmit_nit);
3358 EXPORT_SYMBOL(dev_remove_pack);
3359 EXPORT_SYMBOL(dev_set_allmulti);
3360 EXPORT_SYMBOL(dev_set_promiscuity);
3361 EXPORT_SYMBOL(dev_change_flags);
3362 EXPORT_SYMBOL(dev_set_mtu);
3363 EXPORT_SYMBOL(free_netdev);
3364 EXPORT_SYMBOL(netdev_boot_setup_check);
3365 EXPORT_SYMBOL(netdev_set_master);
3366 EXPORT_SYMBOL(netdev_state_change);
3367 EXPORT_SYMBOL(netif_receive_skb);
3368 EXPORT_SYMBOL(netif_rx);
3369 EXPORT_SYMBOL(register_gifconf);
3370 EXPORT_SYMBOL(register_netdevice);
3371 EXPORT_SYMBOL(register_netdevice_notifier);
3372 EXPORT_SYMBOL(skb_checksum_help);
3373 EXPORT_SYMBOL(synchronize_net);
3374 EXPORT_SYMBOL(unregister_netdevice);
3375 EXPORT_SYMBOL(unregister_netdevice_notifier);
3377 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3378 EXPORT_SYMBOL(br_handle_frame_hook);
3382 EXPORT_SYMBOL(dev_load);
3384 #ifdef CONFIG_NET_HW_FLOWCONTROL
3385 EXPORT_SYMBOL(netdev_dropping);
3386 EXPORT_SYMBOL(netdev_fc_xoff);
3387 EXPORT_SYMBOL(netdev_register_fc);
3388 EXPORT_SYMBOL(netdev_unregister_fc);
3391 #ifdef CONFIG_NET_CLS_ACT
3392 EXPORT_SYMBOL(ing_filter);
3396 EXPORT_PER_CPU_SYMBOL(softnet_data);