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. */
1252 #if defined(CONFIG_MIPS_BRCM)
1253 /* The data buffer of this skb is not pre-allocated any more
1254 * even the skb itself is pre-allocated. (songw)
1256 skb->retfreeq_data_prealloc = 0;
1259 skb->tail += skb->data_len;
1264 #define HARD_TX_LOCK_BH(dev, cpu) { \
1265 if ((dev->features & NETIF_F_LLTX) == 0) { \
1266 spin_lock_bh(&dev->xmit_lock); \
1267 dev->xmit_lock_owner = cpu; \
1271 #define HARD_TX_UNLOCK_BH(dev) { \
1272 if ((dev->features & NETIF_F_LLTX) == 0) { \
1273 dev->xmit_lock_owner = -1; \
1274 spin_unlock_bh(&dev->xmit_lock); \
1278 static inline void qdisc_run(struct net_device *dev)
1280 while (!netif_queue_stopped(dev) &&
1281 qdisc_restart(dev)<0)
1286 * dev_queue_xmit - transmit a buffer
1287 * @skb: buffer to transmit
1289 * Queue a buffer for transmission to a network device. The caller must
1290 * have set the device and priority and built the buffer before calling
1291 * this function. The function can be called from an interrupt.
1293 * A negative errno code is returned on a failure. A success does not
1294 * guarantee the frame will be transmitted as it may be dropped due
1295 * to congestion or traffic shaping.
1298 int dev_queue_xmit(struct sk_buff *skb)
1300 struct net_device *dev = skb->dev;
1304 if (skb_shinfo(skb)->frag_list &&
1305 !(dev->features & NETIF_F_FRAGLIST) &&
1306 __skb_linearize(skb, GFP_ATOMIC))
1309 /* Fragmented skb is linearized if device does not support SG,
1310 * or if at least one of fragments is in highmem and device
1311 * does not support DMA from it.
1313 if (skb_shinfo(skb)->nr_frags &&
1314 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1315 __skb_linearize(skb, GFP_ATOMIC))
1318 /* If packet is not checksummed and device does not support
1319 * checksumming for this protocol, complete checksumming here.
1321 if (skb->ip_summed == CHECKSUM_HW &&
1322 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1323 (!(dev->features & NETIF_F_IP_CSUM) ||
1324 skb->protocol != htons(ETH_P_IP))))
1325 if (skb_checksum_help(&skb, 0))
1329 /* Updates of qdisc are serialized by queue_lock.
1330 * The struct Qdisc which is pointed to by qdisc is now a
1331 * rcu structure - it may be accessed without acquiring
1332 * a lock (but the structure may be stale.) The freeing of the
1333 * qdisc will be deferred until it's known that there are no
1334 * more references to it.
1336 * If the qdisc has an enqueue function, we still need to
1337 * hold the queue_lock before calling it, since queue_lock
1338 * also serializes access to the device queue.
1342 smp_read_barrier_depends();
1343 #ifdef CONFIG_NET_CLS_ACT
1344 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1347 /* Grab device queue */
1348 spin_lock_bh(&dev->queue_lock);
1350 rc = q->enqueue(skb, q);
1354 spin_unlock_bh(&dev->queue_lock);
1356 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1361 /* The device has no queue. Common case for software devices:
1362 loopback, all the sorts of tunnels...
1364 Really, it is unlikely that xmit_lock protection is necessary here.
1365 (f.e. loopback and IP tunnels are clean ignoring statistics
1367 However, it is possible, that they rely on protection
1370 Check this and shot the lock. It is not prone from deadlocks.
1371 Either shot noqueue qdisc, it is even simpler 8)
1373 if (dev->flags & IFF_UP) {
1374 int cpu = get_cpu();
1376 if (dev->xmit_lock_owner != cpu) {
1378 HARD_TX_LOCK_BH(dev, cpu);
1381 if (!netif_queue_stopped(dev)) {
1383 dev_queue_xmit_nit(skb, dev);
1386 if (!dev->hard_start_xmit(skb, dev)) {
1387 HARD_TX_UNLOCK_BH(dev);
1391 HARD_TX_UNLOCK_BH(dev);
1392 if (net_ratelimit())
1393 printk(KERN_CRIT "Virtual device %s asks to "
1394 "queue packet!\n", dev->name);
1398 /* Recursion is detected! It is possible,
1400 if (net_ratelimit())
1401 printk(KERN_CRIT "Dead loop on virtual device "
1402 "%s, fix it urgently!\n", dev->name);
1414 /*=======================================================================
1416 =======================================================================*/
1418 int netdev_max_backlog = 300;
1419 int weight_p = 64; /* old backlog weight */
1420 /* These numbers are selected based on intuition and some
1421 * experimentatiom, if you have more scientific way of doing this
1422 * please go ahead and fix things.
1424 int no_cong_thresh = 10;
1429 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1432 #ifdef CONFIG_NET_HW_FLOWCONTROL
1433 atomic_t netdev_dropping = ATOMIC_INIT(0);
1434 static unsigned long netdev_fc_mask = 1;
1435 unsigned long netdev_fc_xoff;
1436 spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1440 void (*stimul)(struct net_device *);
1441 struct net_device *dev;
1442 } netdev_fc_slots[BITS_PER_LONG];
1444 int netdev_register_fc(struct net_device *dev,
1445 void (*stimul)(struct net_device *dev))
1448 unsigned long flags;
1450 spin_lock_irqsave(&netdev_fc_lock, flags);
1451 if (netdev_fc_mask != ~0UL) {
1452 bit = ffz(netdev_fc_mask);
1453 netdev_fc_slots[bit].stimul = stimul;
1454 netdev_fc_slots[bit].dev = dev;
1455 set_bit(bit, &netdev_fc_mask);
1456 clear_bit(bit, &netdev_fc_xoff);
1458 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1462 void netdev_unregister_fc(int bit)
1464 unsigned long flags;
1466 spin_lock_irqsave(&netdev_fc_lock, flags);
1468 netdev_fc_slots[bit].stimul = NULL;
1469 netdev_fc_slots[bit].dev = NULL;
1470 clear_bit(bit, &netdev_fc_mask);
1471 clear_bit(bit, &netdev_fc_xoff);
1473 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1476 static void netdev_wakeup(void)
1480 spin_lock(&netdev_fc_lock);
1481 xoff = netdev_fc_xoff;
1486 netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1488 spin_unlock(&netdev_fc_lock);
1492 static void get_sample_stats(int cpu)
1498 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1499 int blog = sd->input_pkt_queue.qlen;
1500 int avg_blog = sd->avg_blog;
1502 avg_blog = (avg_blog >> 1) + (blog >> 1);
1504 if (avg_blog > mod_cong) {
1505 /* Above moderate congestion levels. */
1506 sd->cng_level = NET_RX_CN_HIGH;
1509 rq = rd % netdev_max_backlog;
1510 if (rq < avg_blog) /* unlucky bastard */
1511 sd->cng_level = NET_RX_DROP;
1513 } else if (avg_blog > lo_cong) {
1514 sd->cng_level = NET_RX_CN_MOD;
1517 rq = rd % netdev_max_backlog;
1518 if (rq < avg_blog) /* unlucky bastard */
1519 sd->cng_level = NET_RX_CN_HIGH;
1521 } else if (avg_blog > no_cong)
1522 sd->cng_level = NET_RX_CN_LOW;
1523 else /* no congestion */
1524 sd->cng_level = NET_RX_SUCCESS;
1526 sd->avg_blog = avg_blog;
1529 #ifdef OFFLINE_SAMPLE
1530 static void sample_queue(unsigned long dummy)
1532 /* 10 ms 0r 1ms -- i don't care -- JHS */
1534 int cpu = smp_processor_id();
1536 get_sample_stats(cpu);
1537 next_tick += jiffies;
1538 mod_timer(&samp_timer, next_tick);
1544 * netif_rx - post buffer to the network code
1545 * @skb: buffer to post
1547 * This function receives a packet from a device driver and queues it for
1548 * the upper (protocol) levels to process. It always succeeds. The buffer
1549 * may be dropped during processing for congestion control or by the
1553 * NET_RX_SUCCESS (no congestion)
1554 * NET_RX_CN_LOW (low congestion)
1555 * NET_RX_CN_MOD (moderate congestion)
1556 * NET_RX_CN_HIGH (high congestion)
1557 * NET_RX_DROP (packet was dropped)
1561 int netif_rx(struct sk_buff *skb)
1564 struct softnet_data *queue;
1565 unsigned long flags;
1567 #ifdef CONFIG_NETPOLL_RX
1568 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1574 if (!skb->stamp.tv_sec)
1575 net_timestamp(&skb->stamp);
1578 * The code is rearranged so that the path is the most
1579 * short when CPU is congested, but is still operating.
1581 local_irq_save(flags);
1582 this_cpu = smp_processor_id();
1583 queue = &__get_cpu_var(softnet_data);
1585 __get_cpu_var(netdev_rx_stat).total++;
1586 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1587 if (queue->input_pkt_queue.qlen) {
1588 if (queue->throttle)
1593 __skb_queue_tail(&queue->input_pkt_queue, skb);
1594 #ifndef OFFLINE_SAMPLE
1595 get_sample_stats(this_cpu);
1597 local_irq_restore(flags);
1598 return queue->cng_level;
1601 if (queue->throttle) {
1602 queue->throttle = 0;
1603 #ifdef CONFIG_NET_HW_FLOWCONTROL
1604 if (atomic_dec_and_test(&netdev_dropping))
1609 netif_rx_schedule(&queue->backlog_dev);
1613 if (!queue->throttle) {
1614 queue->throttle = 1;
1615 __get_cpu_var(netdev_rx_stat).throttled++;
1616 #ifdef CONFIG_NET_HW_FLOWCONTROL
1617 atomic_inc(&netdev_dropping);
1622 __get_cpu_var(netdev_rx_stat).dropped++;
1623 local_irq_restore(flags);
1629 static __inline__ void skb_bond(struct sk_buff *skb)
1631 struct net_device *dev = skb->dev;
1634 skb->real_dev = skb->dev;
1635 skb->dev = dev->master;
1639 static void net_tx_action(struct softirq_action *h)
1641 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1643 if (sd->completion_queue) {
1644 struct sk_buff *clist;
1646 local_irq_disable();
1647 clist = sd->completion_queue;
1648 sd->completion_queue = NULL;
1652 struct sk_buff *skb = clist;
1653 clist = clist->next;
1655 BUG_TRAP(!atomic_read(&skb->users));
1660 if (sd->output_queue) {
1661 struct net_device *head;
1663 local_irq_disable();
1664 head = sd->output_queue;
1665 sd->output_queue = NULL;
1669 struct net_device *dev = head;
1670 head = head->next_sched;
1672 smp_mb__before_clear_bit();
1673 clear_bit(__LINK_STATE_SCHED, &dev->state);
1675 if (spin_trylock(&dev->queue_lock)) {
1677 spin_unlock(&dev->queue_lock);
1679 netif_schedule(dev);
1685 static __inline__ int deliver_skb(struct sk_buff *skb,
1686 struct packet_type *pt_prev, int last)
1688 atomic_inc(&skb->users);
1689 return pt_prev->func(skb, skb->dev, pt_prev);
1693 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1694 int (*br_handle_frame_hook)(struct sk_buff *skb);
1696 static __inline__ int handle_bridge(struct sk_buff *skb,
1697 struct packet_type *pt_prev)
1699 int ret = NET_RX_DROP;
1701 ret = deliver_skb(skb, pt_prev, 0);
1708 static inline int __handle_bridge(struct sk_buff *skb,
1709 struct packet_type **pt_prev, int *ret)
1711 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
1712 #if defined(CONFIG_MIPS_BRCM)
1713 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1714 if (skb->protocol == __constant_htons(ETH_P_PPP_SES)||
1715 skb->protocol == __constant_htons(ETH_P_PPP_DISC)) {
1716 if (!memcmp(skb->mac.ethernet->h_dest,skb->dev->dev_addr,ETH_ALEN))
1719 *ret = handle_bridge(skb, *pt_prev);
1720 if (br_handle_frame_hook(skb) == 0)
1727 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1728 *ret = handle_bridge(skb, *pt_prev);
1729 if (br_handle_frame_hook(skb) == 0)
1740 #ifdef CONFIG_NET_CLS_ACT
1741 /* TODO: Maybe we should just force sch_ingress to be compiled in
1742 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1743 * a compare and 2 stores extra right now if we dont have it on
1744 * but have CONFIG_NET_CLS_ACT
1745 * NOTE: This doesnt stop any functionality; if you dont have
1746 * the ingress scheduler, you just cant add policies on ingress.
1749 int ing_filter(struct sk_buff *skb)
1752 struct net_device *dev = skb->dev;
1753 int result = TC_ACT_OK;
1755 if (dev->qdisc_ingress) {
1756 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1757 if (MAX_RED_LOOP < ttl++) {
1758 printk("Redir loop detected Dropping packet (%s->%s)\n",
1759 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1763 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1765 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1766 if (NULL == skb->input_dev) {
1767 skb->input_dev = skb->dev;
1768 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1770 spin_lock(&dev->ingress_lock);
1771 if ((q = dev->qdisc_ingress) != NULL)
1772 result = q->enqueue(skb, q);
1773 spin_unlock(&dev->ingress_lock);
1781 int netif_receive_skb(struct sk_buff *skb)
1783 struct packet_type *ptype, *pt_prev;
1784 int ret = NET_RX_DROP;
1785 unsigned short type;
1787 #ifdef CONFIG_NETPOLL_RX
1788 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1794 if (!skb->stamp.tv_sec)
1795 net_timestamp(&skb->stamp);
1799 __get_cpu_var(netdev_rx_stat).total++;
1801 skb->h.raw = skb->nh.raw = skb->data;
1802 skb->mac_len = skb->nh.raw - skb->mac.raw;
1805 #ifdef CONFIG_NET_CLS_ACT
1806 if (skb->tc_verd & TC_NCLS) {
1807 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1814 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1815 if (!ptype->dev || ptype->dev == skb->dev) {
1817 ret = deliver_skb(skb, pt_prev, 0);
1822 #ifdef CONFIG_NET_CLS_ACT
1824 atomic_inc(&skb->users);
1825 ret = pt_prev->func(skb, skb->dev, pt_prev);
1826 pt_prev = NULL; /* noone else should process this after*/
1828 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1831 ret = ing_filter(skb);
1833 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1842 handle_diverter(skb);
1844 if (__handle_bridge(skb, &pt_prev, &ret))
1847 type = skb->protocol;
1848 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1849 if (ptype->type == type &&
1850 (!ptype->dev || ptype->dev == skb->dev)) {
1852 ret = deliver_skb(skb, pt_prev, 0);
1858 ret = pt_prev->func(skb, skb->dev, pt_prev);
1861 /* Jamal, now you will not able to escape explaining
1862 * me how you were going to use this. :-)
1872 static int process_backlog(struct net_device *backlog_dev, int *budget)
1875 int quota = min(backlog_dev->quota, *budget);
1876 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1877 unsigned long start_time = jiffies;
1880 struct sk_buff *skb;
1881 struct net_device *dev;
1883 local_irq_disable();
1884 skb = __skb_dequeue(&queue->input_pkt_queue);
1891 netif_receive_skb(skb);
1897 if (work >= quota || jiffies - start_time > 1)
1900 #ifdef CONFIG_NET_HW_FLOWCONTROL
1901 if (queue->throttle &&
1902 queue->input_pkt_queue.qlen < no_cong_thresh ) {
1903 queue->throttle = 0;
1904 if (atomic_dec_and_test(&netdev_dropping)) {
1912 backlog_dev->quota -= work;
1917 backlog_dev->quota -= work;
1920 list_del(&backlog_dev->poll_list);
1921 smp_mb__before_clear_bit();
1922 netif_poll_enable(backlog_dev);
1924 if (queue->throttle) {
1925 queue->throttle = 0;
1926 #ifdef CONFIG_NET_HW_FLOWCONTROL
1927 if (atomic_dec_and_test(&netdev_dropping))
1935 static void net_rx_action(struct softirq_action *h)
1937 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1938 unsigned long start_time = jiffies;
1939 int budget = netdev_max_backlog;
1942 local_irq_disable();
1944 while (!list_empty(&queue->poll_list)) {
1945 struct net_device *dev;
1947 if (budget <= 0 || jiffies - start_time > 1)
1952 dev = list_entry(queue->poll_list.next,
1953 struct net_device, poll_list);
1955 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1956 local_irq_disable();
1957 list_del(&dev->poll_list);
1958 list_add_tail(&dev->poll_list, &queue->poll_list);
1960 dev->quota += dev->weight;
1962 dev->quota = dev->weight;
1965 local_irq_disable();
1973 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1974 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1978 static gifconf_func_t * gifconf_list [NPROTO];
1981 * register_gifconf - register a SIOCGIF handler
1982 * @family: Address family
1983 * @gifconf: Function handler
1985 * Register protocol dependent address dumping routines. The handler
1986 * that is passed must not be freed or reused until it has been replaced
1987 * by another handler.
1989 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1991 if (family >= NPROTO)
1993 gifconf_list[family] = gifconf;
1999 * Map an interface index to its name (SIOCGIFNAME)
2003 * We need this ioctl for efficient implementation of the
2004 * if_indextoname() function required by the IPv6 API. Without
2005 * it, we would have to search all the interfaces to find a
2009 static int dev_ifname(struct ifreq __user *arg)
2011 struct net_device *dev;
2015 * Fetch the caller's info block.
2018 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2021 read_lock(&dev_base_lock);
2022 dev = __dev_get_by_index(ifr.ifr_ifindex);
2024 read_unlock(&dev_base_lock);
2028 strcpy(ifr.ifr_name, dev->name);
2029 read_unlock(&dev_base_lock);
2031 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2037 * Perform a SIOCGIFCONF call. This structure will change
2038 * size eventually, and there is nothing I can do about it.
2039 * Thus we will need a 'compatibility mode'.
2042 static int dev_ifconf(char __user *arg)
2045 struct net_device *dev;
2052 * Fetch the caller's info block.
2055 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2062 * Loop over the interfaces, and write an info block for each.
2066 for (dev = dev_base; dev; dev = dev->next) {
2067 for (i = 0; i < NPROTO; i++) {
2068 if (gifconf_list[i]) {
2071 done = gifconf_list[i](dev, NULL, 0);
2073 done = gifconf_list[i](dev, pos + total,
2083 * All done. Write the updated control block back to the caller.
2085 ifc.ifc_len = total;
2088 * Both BSD and Solaris return 0 here, so we do too.
2090 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2093 #ifdef CONFIG_PROC_FS
2095 * This is invoked by the /proc filesystem handler to display a device
2098 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2100 struct net_device *dev;
2103 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2105 return i == pos ? dev : NULL;
2108 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2110 read_lock(&dev_base_lock);
2111 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2114 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2117 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2120 void dev_seq_stop(struct seq_file *seq, void *v)
2122 read_unlock(&dev_base_lock);
2125 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2127 if (dev->get_stats) {
2128 struct net_device_stats *stats = dev->get_stats(dev);
2130 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2131 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2132 dev->name, stats->rx_bytes, stats->rx_packets,
2134 stats->rx_dropped + stats->rx_missed_errors,
2135 stats->rx_fifo_errors,
2136 stats->rx_length_errors + stats->rx_over_errors +
2137 stats->rx_crc_errors + stats->rx_frame_errors,
2138 stats->rx_compressed, stats->multicast,
2139 stats->tx_bytes, stats->tx_packets,
2140 stats->tx_errors, stats->tx_dropped,
2141 stats->tx_fifo_errors, stats->collisions,
2142 stats->tx_carrier_errors +
2143 stats->tx_aborted_errors +
2144 stats->tx_window_errors +
2145 stats->tx_heartbeat_errors,
2146 stats->tx_compressed);
2148 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2152 * Called from the PROCfs module. This now uses the new arbitrary sized
2153 * /proc/net interface to create /proc/net/dev
2155 static int dev_seq_show(struct seq_file *seq, void *v)
2157 if (v == SEQ_START_TOKEN)
2158 seq_puts(seq, "Inter-| Receive "
2160 " face |bytes packets errs drop fifo frame "
2161 "compressed multicast|bytes packets errs "
2162 "drop fifo colls carrier compressed\n");
2164 dev_seq_printf_stats(seq, v);
2168 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2170 struct netif_rx_stats *rc = NULL;
2172 while (*pos < NR_CPUS)
2173 if (cpu_online(*pos)) {
2174 rc = &per_cpu(netdev_rx_stat, *pos);
2181 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2183 return softnet_get_online(pos);
2186 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2189 return softnet_get_online(pos);
2192 static void softnet_seq_stop(struct seq_file *seq, void *v)
2196 static int softnet_seq_show(struct seq_file *seq, void *v)
2198 struct netif_rx_stats *s = v;
2200 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2201 s->total, s->dropped, s->time_squeeze, s->throttled,
2202 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2203 s->fastroute_deferred_out,
2205 s->fastroute_latency_reduction
2213 static struct seq_operations dev_seq_ops = {
2214 .start = dev_seq_start,
2215 .next = dev_seq_next,
2216 .stop = dev_seq_stop,
2217 .show = dev_seq_show,
2220 static int dev_seq_open(struct inode *inode, struct file *file)
2222 return seq_open(file, &dev_seq_ops);
2225 static struct file_operations dev_seq_fops = {
2226 .owner = THIS_MODULE,
2227 .open = dev_seq_open,
2229 .llseek = seq_lseek,
2230 .release = seq_release,
2233 static struct seq_operations softnet_seq_ops = {
2234 .start = softnet_seq_start,
2235 .next = softnet_seq_next,
2236 .stop = softnet_seq_stop,
2237 .show = softnet_seq_show,
2240 static int softnet_seq_open(struct inode *inode, struct file *file)
2242 return seq_open(file, &softnet_seq_ops);
2245 static struct file_operations softnet_seq_fops = {
2246 .owner = THIS_MODULE,
2247 .open = softnet_seq_open,
2249 .llseek = seq_lseek,
2250 .release = seq_release,
2254 extern int wireless_proc_init(void);
2256 #define wireless_proc_init() 0
2259 static int __init dev_proc_init(void)
2263 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2265 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2267 if (wireless_proc_init())
2273 proc_net_remove("softnet_stat");
2275 proc_net_remove("dev");
2279 #define dev_proc_init() 0
2280 #endif /* CONFIG_PROC_FS */
2284 * netdev_set_master - set up master/slave pair
2285 * @slave: slave device
2286 * @master: new master device
2288 * Changes the master device of the slave. Pass %NULL to break the
2289 * bonding. The caller must hold the RTNL semaphore. On a failure
2290 * a negative errno code is returned. On success the reference counts
2291 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2292 * function returns zero.
2294 int netdev_set_master(struct net_device *slave, struct net_device *master)
2296 struct net_device *old = slave->master;
2306 slave->master = master;
2314 slave->flags |= IFF_SLAVE;
2316 slave->flags &= ~IFF_SLAVE;
2318 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2323 * dev_set_promiscuity - update promiscuity count on a device
2327 * Add or remove promsicuity from a device. While the count in the device
2328 * remains above zero the interface remains promiscuous. Once it hits zero
2329 * the device reverts back to normal filtering operation. A negative inc
2330 * value is used to drop promiscuity on the device.
2332 void dev_set_promiscuity(struct net_device *dev, int inc)
2334 unsigned short old_flags = dev->flags;
2336 dev->flags |= IFF_PROMISC;
2337 if ((dev->promiscuity += inc) == 0)
2338 dev->flags &= ~IFF_PROMISC;
2339 if (dev->flags ^ old_flags) {
2341 printk(KERN_INFO "device %s %s promiscuous mode\n",
2342 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2348 * dev_set_allmulti - update allmulti count on a device
2352 * Add or remove reception of all multicast frames to a device. While the
2353 * count in the device remains above zero the interface remains listening
2354 * to all interfaces. Once it hits zero the device reverts back to normal
2355 * filtering operation. A negative @inc value is used to drop the counter
2356 * when releasing a resource needing all multicasts.
2359 void dev_set_allmulti(struct net_device *dev, int inc)
2361 unsigned short old_flags = dev->flags;
2363 dev->flags |= IFF_ALLMULTI;
2364 if ((dev->allmulti += inc) == 0)
2365 dev->flags &= ~IFF_ALLMULTI;
2366 if (dev->flags ^ old_flags)
2370 unsigned dev_get_flags(const struct net_device *dev)
2374 flags = (dev->flags & ~(IFF_PROMISC |
2377 (dev->gflags & (IFF_PROMISC |
2380 if (netif_running(dev) && netif_carrier_ok(dev))
2381 flags |= IFF_RUNNING;
2386 int dev_change_flags(struct net_device *dev, unsigned flags)
2389 int old_flags = dev->flags;
2392 * Set the flags on our device.
2395 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2396 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2398 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2402 * Load in the correct multicast list now the flags have changed.
2408 * Have we downed the interface. We handle IFF_UP ourselves
2409 * according to user attempts to set it, rather than blindly
2414 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2415 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2421 if (dev->flags & IFF_UP &&
2422 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2424 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2426 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2427 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2428 dev->gflags ^= IFF_PROMISC;
2429 dev_set_promiscuity(dev, inc);
2432 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2433 is important. Some (broken) drivers set IFF_PROMISC, when
2434 IFF_ALLMULTI is requested not asking us and not reporting.
2436 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2437 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2438 dev->gflags ^= IFF_ALLMULTI;
2439 dev_set_allmulti(dev, inc);
2442 if (old_flags ^ dev->flags)
2443 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2448 int dev_set_mtu(struct net_device *dev, int new_mtu)
2452 if (new_mtu == dev->mtu)
2455 /* MTU must be positive. */
2459 if (!netif_device_present(dev))
2463 if (dev->change_mtu)
2464 err = dev->change_mtu(dev, new_mtu);
2467 if (!err && dev->flags & IFF_UP)
2468 notifier_call_chain(&netdev_chain,
2469 NETDEV_CHANGEMTU, dev);
2475 * Perform the SIOCxIFxxx calls.
2477 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2480 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2486 case SIOCGIFFLAGS: /* Get interface flags */
2487 ifr->ifr_flags = dev_get_flags(dev);
2490 case SIOCSIFFLAGS: /* Set interface flags */
2491 return dev_change_flags(dev, ifr->ifr_flags);
2493 case SIOCGIFMETRIC: /* Get the metric on the interface
2494 (currently unused) */
2495 ifr->ifr_metric = 0;
2498 case SIOCSIFMETRIC: /* Set the metric on the interface
2499 (currently unused) */
2502 #if defined(CONFIG_MIPS_BRCM)
2503 case SIOCGIFTRANSSTART:
2504 ifr->ifr_ifru.ifru_ivalue = dev->trans_start;
2508 case SIOCGIFMTU: /* Get the MTU of a device */
2509 ifr->ifr_mtu = dev->mtu;
2512 #if defined(CONFIG_MIPS_BRCM)
2513 case SIOCCIFSTATS: /* Clean up the Stats of a device */
2515 struct net_device_stats * pStats = dev->get_stats(dev);
2516 memset(pStats, 0, sizeof(struct net_device_stats));
2521 case SIOCSIFMTU: /* Set the MTU of a device */
2522 return dev_set_mtu(dev, ifr->ifr_mtu);
2525 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2526 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2527 ifr->ifr_hwaddr.sa_family = dev->type;
2531 if (!dev->set_mac_address)
2533 if (ifr->ifr_hwaddr.sa_family != dev->type)
2535 if (!netif_device_present(dev))
2537 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2539 notifier_call_chain(&netdev_chain,
2540 NETDEV_CHANGEADDR, dev);
2543 case SIOCSIFHWBROADCAST:
2544 if (ifr->ifr_hwaddr.sa_family != dev->type)
2546 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2547 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2548 notifier_call_chain(&netdev_chain,
2549 NETDEV_CHANGEADDR, dev);
2553 ifr->ifr_map.mem_start = dev->mem_start;
2554 ifr->ifr_map.mem_end = dev->mem_end;
2555 ifr->ifr_map.base_addr = dev->base_addr;
2556 ifr->ifr_map.irq = dev->irq;
2557 ifr->ifr_map.dma = dev->dma;
2558 ifr->ifr_map.port = dev->if_port;
2562 if (dev->set_config) {
2563 if (!netif_device_present(dev))
2565 return dev->set_config(dev, &ifr->ifr_map);
2570 if (!dev->set_multicast_list ||
2571 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2573 if (!netif_device_present(dev))
2575 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2579 if (!dev->set_multicast_list ||
2580 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2582 if (!netif_device_present(dev))
2584 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2588 ifr->ifr_ifindex = dev->ifindex;
2592 ifr->ifr_qlen = dev->tx_queue_len;
2596 if (ifr->ifr_qlen < 0)
2598 dev->tx_queue_len = ifr->ifr_qlen;
2602 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2603 return dev_change_name(dev, ifr->ifr_newname);
2606 * Unknown or private ioctl
2610 if ((cmd >= SIOCDEVPRIVATE &&
2611 cmd <= SIOCDEVPRIVATE + 15) ||
2612 cmd == SIOCBONDENSLAVE ||
2613 cmd == SIOCBONDRELEASE ||
2614 cmd == SIOCBONDSETHWADDR ||
2615 cmd == SIOCBONDSLAVEINFOQUERY ||
2616 cmd == SIOCBONDINFOQUERY ||
2617 cmd == SIOCBONDCHANGEACTIVE ||
2618 cmd == SIOCGMIIPHY ||
2619 cmd == SIOCGMIIREG ||
2620 cmd == SIOCSMIIREG ||
2621 cmd == SIOCBRADDIF ||
2622 cmd == SIOCBRDELIF ||
2623 cmd == SIOCWANDEV) {
2625 if (dev->do_ioctl) {
2626 if (netif_device_present(dev))
2627 err = dev->do_ioctl(dev, ifr,
2640 * This function handles all "interface"-type I/O control requests. The actual
2641 * 'doing' part of this is dev_ifsioc above.
2645 * dev_ioctl - network device ioctl
2646 * @cmd: command to issue
2647 * @arg: pointer to a struct ifreq in user space
2649 * Issue ioctl functions to devices. This is normally called by the
2650 * user space syscall interfaces but can sometimes be useful for
2651 * other purposes. The return value is the return from the syscall if
2652 * positive or a negative errno code on error.
2655 int dev_ioctl(unsigned int cmd, void __user *arg)
2661 /* One special case: SIOCGIFCONF takes ifconf argument
2662 and requires shared lock, because it sleeps writing
2666 if (cmd == SIOCGIFCONF) {
2668 ret = dev_ifconf((char __user *) arg);
2672 if (cmd == SIOCGIFNAME)
2673 return dev_ifname((struct ifreq __user *)arg);
2675 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2678 ifr.ifr_name[IFNAMSIZ-1] = 0;
2680 colon = strchr(ifr.ifr_name, ':');
2685 * See which interface the caller is talking about.
2690 * These ioctl calls:
2691 * - can be done by all.
2692 * - atomic and do not require locking.
2703 #if defined(CONFIG_MIPS_BRCM)
2704 case SIOCGIFTRANSSTART:
2706 dev_load(ifr.ifr_name);
2707 read_lock(&dev_base_lock);
2708 ret = dev_ifsioc(&ifr, cmd);
2709 read_unlock(&dev_base_lock);
2713 if (copy_to_user(arg, &ifr,
2714 sizeof(struct ifreq)))
2720 dev_load(ifr.ifr_name);
2722 ret = dev_ethtool(&ifr);
2727 if (copy_to_user(arg, &ifr,
2728 sizeof(struct ifreq)))
2734 * These ioctl calls:
2735 * - require superuser power.
2736 * - require strict serialization.
2742 if (!capable(CAP_NET_ADMIN))
2744 dev_load(ifr.ifr_name);
2746 ret = dev_ifsioc(&ifr, cmd);
2751 if (copy_to_user(arg, &ifr,
2752 sizeof(struct ifreq)))
2758 * These ioctl calls:
2759 * - require superuser power.
2760 * - require strict serialization.
2761 * - do not return a value
2771 case SIOCSIFHWBROADCAST:
2774 case SIOCBONDENSLAVE:
2775 case SIOCBONDRELEASE:
2776 case SIOCBONDSETHWADDR:
2777 case SIOCBONDSLAVEINFOQUERY:
2778 case SIOCBONDINFOQUERY:
2779 case SIOCBONDCHANGEACTIVE:
2782 if (!capable(CAP_NET_ADMIN))
2784 dev_load(ifr.ifr_name);
2786 ret = dev_ifsioc(&ifr, cmd);
2791 /* Get the per device memory space. We can add this but
2792 * currently do not support it */
2794 /* Set the per device memory buffer space.
2795 * Not applicable in our case */
2800 * Unknown or private ioctl.
2803 if (cmd == SIOCWANDEV ||
2804 #if defined(CONFIG_MIPS_BRCM)
2805 cmd == SIOCCIFSTATS ||
2807 (cmd >= SIOCDEVPRIVATE &&
2808 cmd <= SIOCDEVPRIVATE + 15)) {
2809 dev_load(ifr.ifr_name);
2811 ret = dev_ifsioc(&ifr, cmd);
2813 if (!ret && copy_to_user(arg, &ifr,
2814 sizeof(struct ifreq)))
2819 /* Take care of Wireless Extensions */
2820 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2821 /* If command is `set a parameter', or
2822 * `get the encoding parameters', check if
2823 * the user has the right to do it */
2824 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2825 if (!capable(CAP_NET_ADMIN))
2828 dev_load(ifr.ifr_name);
2830 /* Follow me in net/core/wireless.c */
2831 ret = wireless_process_ioctl(&ifr, cmd);
2833 if (!ret && IW_IS_GET(cmd) &&
2834 copy_to_user(arg, &ifr,
2835 sizeof(struct ifreq)))
2839 #endif /* WIRELESS_EXT */
2846 * dev_new_index - allocate an ifindex
2848 * Returns a suitable unique value for a new device interface
2849 * number. The caller must hold the rtnl semaphore or the
2850 * dev_base_lock to be sure it remains unique.
2852 int dev_new_index(void)
2858 if (!__dev_get_by_index(ifindex))
2863 static int dev_boot_phase = 1;
2865 /* Delayed registration/unregisteration */
2866 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2867 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2869 static inline void net_set_todo(struct net_device *dev)
2871 spin_lock(&net_todo_list_lock);
2872 list_add_tail(&dev->todo_list, &net_todo_list);
2873 spin_unlock(&net_todo_list_lock);
2877 * register_netdevice - register a network device
2878 * @dev: device to register
2880 * Take a completed network device structure and add it to the kernel
2881 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2882 * chain. 0 is returned on success. A negative errno code is returned
2883 * on a failure to set up the device, or if the name is a duplicate.
2885 * Callers must hold the rtnl semaphore. See the comment at the
2886 * end of Space.c for details about the locking. You may want
2887 * register_netdev() instead of this.
2890 * The locking appears insufficient to guarantee two parallel registers
2891 * will not get the same name.
2894 int register_netdevice(struct net_device *dev)
2896 struct hlist_head *head;
2897 struct hlist_node *p;
2900 BUG_ON(dev_boot_phase);
2903 /* When net_device's are persistent, this will be fatal. */
2904 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2906 spin_lock_init(&dev->queue_lock);
2907 spin_lock_init(&dev->xmit_lock);
2908 dev->xmit_lock_owner = -1;
2909 #ifdef CONFIG_NET_CLS_ACT
2910 spin_lock_init(&dev->ingress_lock);
2913 ret = alloc_divert_blk(dev);
2919 #if defined(CONFIG_MIPS_BRCM)
2920 /* Init dev->last_stats in case CONFIG_SYSFS is disabled */
2921 dev->last_stats = dev->get_stats;
2924 /* Init, if this function is available */
2926 ret = dev->init(dev);
2934 if (!dev_valid_name(dev->name)) {
2939 dev->ifindex = dev_new_index();
2940 if (dev->iflink == -1)
2941 dev->iflink = dev->ifindex;
2943 /* Check for existence of name */
2944 head = dev_name_hash(dev->name);
2945 hlist_for_each(p, head) {
2946 struct net_device *d
2947 = hlist_entry(p, struct net_device, name_hlist);
2948 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2954 /* Fix illegal SG+CSUM combinations. */
2955 if ((dev->features & NETIF_F_SG) &&
2956 !(dev->features & (NETIF_F_IP_CSUM |
2958 NETIF_F_HW_CSUM))) {
2959 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2961 dev->features &= ~NETIF_F_SG;
2965 * nil rebuild_header routine,
2966 * that should be never called and used as just bug trap.
2969 if (!dev->rebuild_header)
2970 dev->rebuild_header = default_rebuild_header;
2973 * Default initial state at registry is that the
2974 * device is present.
2977 set_bit(__LINK_STATE_PRESENT, &dev->state);
2980 dev_init_scheduler(dev);
2981 write_lock_bh(&dev_base_lock);
2983 dev_tail = &dev->next;
2984 hlist_add_head(&dev->name_hlist, head);
2985 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2987 dev->reg_state = NETREG_REGISTERING;
2988 write_unlock_bh(&dev_base_lock);
2990 /* Notify protocols, that a new device appeared. */
2991 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2993 /* Finish registration after unlock */
3000 free_divert_blk(dev);
3005 * netdev_wait_allrefs - wait until all references are gone.
3007 * This is called when unregistering network devices.
3009 * Any protocol or device that holds a reference should register
3010 * for netdevice notification, and cleanup and put back the
3011 * reference if they receive an UNREGISTER event.
3012 * We can get stuck here if buggy protocols don't correctly
3015 static void netdev_wait_allrefs(struct net_device *dev)
3017 unsigned long rebroadcast_time, warning_time;
3019 rebroadcast_time = warning_time = jiffies;
3020 while (atomic_read(&dev->refcnt) != 0) {
3021 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3024 /* Rebroadcast unregister notification */
3025 notifier_call_chain(&netdev_chain,
3026 NETDEV_UNREGISTER, dev);
3028 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3030 /* We must not have linkwatch events
3031 * pending on unregister. If this
3032 * happens, we simply run the queue
3033 * unscheduled, resulting in a noop
3036 linkwatch_run_queue();
3041 rebroadcast_time = jiffies;
3044 current->state = TASK_INTERRUPTIBLE;
3045 schedule_timeout(HZ / 4);
3047 if (time_after(jiffies, warning_time + 10 * HZ)) {
3048 printk(KERN_EMERG "unregister_netdevice: "
3049 "waiting for %s to become free. Usage "
3051 dev->name, atomic_read(&dev->refcnt));
3052 #if defined(SUPPORT_TC)
3053 /* decrement on ppp net work device reference count (refcnt)
3054 * in sch_generic.c could result -1 and it should be reset to 0
3056 if (atomic_read(&dev->refcnt) == -1 && strstr(dev->name, "ppp_")) {
3057 printk("dev->name = %s, dev->refcnt=%d\n", dev->name, atomic_read(&dev->refcnt));
3058 atomic_set(&dev->refcnt, 0);
3059 printk("after reset to 0, dev->refcnt=%d\n", atomic_read(&dev->refcnt));
3063 #if defined(CONFIG_MIPS_BRCM)
3064 /* if eth? device reference counter goes to negative, reset to 0 and get out the while loop */
3065 if (atomic_read(&dev->refcnt) < 0 && strstr(dev->name, "eth")) {
3066 atomic_set(&dev->refcnt, 0);
3067 printk("Reset to 0, dev->refcnt=%d.\n", atomic_read(&dev->refcnt));
3071 warning_time = jiffies;
3080 * register_netdevice(x1);
3081 * register_netdevice(x2);
3083 * unregister_netdevice(y1);
3084 * unregister_netdevice(y2);
3090 * We are invoked by rtnl_unlock() after it drops the semaphore.
3091 * This allows us to deal with problems:
3092 * 1) We can create/delete sysfs objects which invoke hotplug
3093 * without deadlocking with linkwatch via keventd.
3094 * 2) Since we run with the RTNL semaphore not held, we can sleep
3095 * safely in order to wait for the netdev refcnt to drop to zero.
3097 static DECLARE_MUTEX(net_todo_run_mutex);
3098 void netdev_run_todo(void)
3100 struct list_head list = LIST_HEAD_INIT(list);
3104 /* Need to guard against multiple cpu's getting out of order. */
3105 down(&net_todo_run_mutex);
3107 /* Not safe to do outside the semaphore. We must not return
3108 * until all unregister events invoked by the local processor
3109 * have been completed (either by this todo run, or one on
3112 if (list_empty(&net_todo_list))
3115 /* Snapshot list, allow later requests */
3116 spin_lock(&net_todo_list_lock);
3117 list_splice_init(&net_todo_list, &list);
3118 spin_unlock(&net_todo_list_lock);
3120 while (!list_empty(&list)) {
3121 struct net_device *dev
3122 = list_entry(list.next, struct net_device, todo_list);
3123 list_del(&dev->todo_list);
3125 switch(dev->reg_state) {
3126 case NETREG_REGISTERING:
3127 err = netdev_register_sysfs(dev);
3129 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3131 dev->reg_state = NETREG_REGISTERED;
3134 case NETREG_UNREGISTERING:
3135 netdev_unregister_sysfs(dev);
3136 dev->reg_state = NETREG_UNREGISTERED;
3138 netdev_wait_allrefs(dev);
3141 BUG_ON(atomic_read(&dev->refcnt));
3142 BUG_TRAP(!dev->ip_ptr);
3143 BUG_TRAP(!dev->ip6_ptr);
3144 BUG_TRAP(!dev->dn_ptr);
3147 /* It must be the very last action,
3148 * after this 'dev' may point to freed up memory.
3150 if (dev->destructor)
3151 dev->destructor(dev);
3155 printk(KERN_ERR "network todo '%s' but state %d\n",
3156 dev->name, dev->reg_state);
3162 up(&net_todo_run_mutex);
3166 * free_netdev - free network device
3169 * This function does the last stage of destroying an allocated device
3170 * interface. The reference to the device object is released.
3171 * If this is the last reference then it will be freed.
3173 void free_netdev(struct net_device *dev)
3176 /* Compatiablity with error handling in drivers */
3177 if (dev->reg_state == NETREG_UNINITIALIZED) {
3178 kfree((char *)dev - dev->padded);
3182 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3183 dev->reg_state = NETREG_RELEASED;
3185 /* will free via class release */
3186 class_device_put(&dev->class_dev);
3188 kfree((char *)dev - dev->padded);
3192 /* Synchronize with packet receive processing. */
3193 void synchronize_net(void)
3196 synchronize_kernel();
3200 * unregister_netdevice - remove device from the kernel
3203 * This function shuts down a device interface and removes it
3204 * from the kernel tables. On success 0 is returned, on a failure
3205 * a negative errno code is returned.
3207 * Callers must hold the rtnl semaphore. See the comment at the
3208 * end of Space.c for details about the locking. You may want
3209 * unregister_netdev() instead of this.
3212 int unregister_netdevice(struct net_device *dev)
3214 struct net_device *d, **dp;
3216 BUG_ON(dev_boot_phase);
3219 /* Some devices call without registering for initialization unwind. */
3220 if (dev->reg_state == NETREG_UNINITIALIZED) {
3221 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3222 "was registered\n", dev->name, dev);
3226 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3228 /* If device is running, close it first. */
3229 if (dev->flags & IFF_UP)
3232 /* And unlink it from device chain. */
3233 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3235 write_lock_bh(&dev_base_lock);
3236 hlist_del(&dev->name_hlist);
3237 hlist_del(&dev->index_hlist);
3238 if (dev_tail == &dev->next)
3241 write_unlock_bh(&dev_base_lock);
3246 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3251 dev->reg_state = NETREG_UNREGISTERING;
3255 /* Shutdown queueing discipline. */
3259 /* Notify protocols, that we are about to destroy
3260 this device. They should clean all the things.
3262 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3265 * Flush the multicast chain
3267 dev_mc_discard(dev);
3272 /* Notifier chain MUST detach us from master device. */
3273 BUG_TRAP(!dev->master);
3275 free_divert_blk(dev);
3277 /* Finish processing unregister after unlock */
3286 #ifdef CONFIG_HOTPLUG_CPU
3287 static int dev_cpu_callback(struct notifier_block *nfb,
3288 unsigned long action,
3291 struct sk_buff **list_skb;
3292 struct net_device **list_net;
3293 struct sk_buff *skb;
3294 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3295 struct softnet_data *sd, *oldsd;
3297 if (action != CPU_DEAD)
3300 local_irq_disable();
3301 cpu = smp_processor_id();
3302 sd = &per_cpu(softnet_data, cpu);
3303 oldsd = &per_cpu(softnet_data, oldcpu);
3305 /* Find end of our completion_queue. */
3306 list_skb = &sd->completion_queue;
3308 list_skb = &(*list_skb)->next;
3309 /* Append completion queue from offline CPU. */
3310 *list_skb = oldsd->completion_queue;
3311 oldsd->completion_queue = NULL;
3313 /* Find end of our output_queue. */
3314 list_net = &sd->output_queue;
3316 list_net = &(*list_net)->next_sched;
3317 /* Append output queue from offline CPU. */
3318 *list_net = oldsd->output_queue;
3319 oldsd->output_queue = NULL;
3321 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3324 /* Process offline CPU's input_pkt_queue */
3325 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3330 #endif /* CONFIG_HOTPLUG_CPU */
3334 * Initialize the DEV module. At boot time this walks the device list and
3335 * unhooks any devices that fail to initialise (normally hardware not
3336 * present) and leaves us with a valid list of present and active devices.
3341 * This is called single threaded during boot, so no need
3342 * to take the rtnl semaphore.
3344 static int __init net_dev_init(void)
3346 int i, rc = -ENOMEM;
3348 BUG_ON(!dev_boot_phase);
3350 if (dev_proc_init())
3353 if (netdev_sysfs_init())
3356 INIT_LIST_HEAD(&ptype_all);
3357 for (i = 0; i < 16; i++)
3358 INIT_LIST_HEAD(&ptype_base[i]);
3360 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3361 INIT_HLIST_HEAD(&dev_name_head[i]);
3363 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3364 INIT_HLIST_HEAD(&dev_index_head[i]);
3367 * Initialise the packet receive queues.
3370 for (i = 0; i < NR_CPUS; i++) {
3371 struct softnet_data *queue;
3373 queue = &per_cpu(softnet_data, i);
3374 skb_queue_head_init(&queue->input_pkt_queue);
3375 queue->throttle = 0;
3376 queue->cng_level = 0;
3377 queue->avg_blog = 10; /* arbitrary non-zero */
3378 queue->completion_queue = NULL;
3379 INIT_LIST_HEAD(&queue->poll_list);
3380 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3381 queue->backlog_dev.weight = weight_p;
3382 queue->backlog_dev.poll = process_backlog;
3383 atomic_set(&queue->backlog_dev.refcnt, 1);
3386 #ifdef OFFLINE_SAMPLE
3387 samp_timer.expires = jiffies + (10 * HZ);
3388 add_timer(&samp_timer);
3393 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3394 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3396 hotcpu_notifier(dev_cpu_callback, 0);
3404 subsys_initcall(net_dev_init);
3406 EXPORT_SYMBOL(__dev_get);
3407 EXPORT_SYMBOL(__dev_get_by_flags);
3408 EXPORT_SYMBOL(__dev_get_by_index);
3409 EXPORT_SYMBOL(__dev_get_by_name);
3410 EXPORT_SYMBOL(__dev_remove_pack);
3411 EXPORT_SYMBOL(__skb_linearize);
3412 EXPORT_SYMBOL(call_netdevice_notifiers);
3413 EXPORT_SYMBOL(dev_add_pack);
3414 EXPORT_SYMBOL(dev_alloc_name);
3415 EXPORT_SYMBOL(dev_close);
3416 EXPORT_SYMBOL(dev_get_by_flags);
3417 EXPORT_SYMBOL(dev_get_by_index);
3418 EXPORT_SYMBOL(dev_get_by_name);
3419 EXPORT_SYMBOL(dev_getbyhwaddr);
3420 EXPORT_SYMBOL(dev_ioctl);
3421 EXPORT_SYMBOL(dev_new_index);
3422 EXPORT_SYMBOL(dev_open);
3423 EXPORT_SYMBOL(dev_queue_xmit);
3424 EXPORT_SYMBOL(dev_queue_xmit_nit);
3425 EXPORT_SYMBOL(dev_remove_pack);
3426 EXPORT_SYMBOL(dev_set_allmulti);
3427 EXPORT_SYMBOL(dev_set_promiscuity);
3428 EXPORT_SYMBOL(dev_change_flags);
3429 EXPORT_SYMBOL(dev_set_mtu);
3430 EXPORT_SYMBOL(free_netdev);
3431 EXPORT_SYMBOL(netdev_boot_setup_check);
3432 EXPORT_SYMBOL(netdev_set_master);
3433 EXPORT_SYMBOL(netdev_state_change);
3434 EXPORT_SYMBOL(netif_receive_skb);
3435 EXPORT_SYMBOL(netif_rx);
3436 EXPORT_SYMBOL(register_gifconf);
3437 EXPORT_SYMBOL(register_netdevice);
3438 EXPORT_SYMBOL(register_netdevice_notifier);
3439 EXPORT_SYMBOL(skb_checksum_help);
3440 EXPORT_SYMBOL(synchronize_net);
3441 EXPORT_SYMBOL(unregister_netdevice);
3442 EXPORT_SYMBOL(unregister_netdevice_notifier);
3444 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3445 EXPORT_SYMBOL(br_handle_frame_hook);
3449 EXPORT_SYMBOL(dev_load);
3451 #ifdef CONFIG_NET_HW_FLOWCONTROL
3452 EXPORT_SYMBOL(netdev_dropping);
3453 EXPORT_SYMBOL(netdev_fc_xoff);
3454 EXPORT_SYMBOL(netdev_register_fc);
3455 EXPORT_SYMBOL(netdev_unregister_fc);
3458 #ifdef CONFIG_NET_CLS_ACT
3459 EXPORT_SYMBOL(ing_filter);
3463 EXPORT_PER_CPU_SYMBOL(softnet_data);