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 #ifdef CONFIG_NET_RADIO
111 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
112 #include <net/iw_handler.h>
113 #endif /* CONFIG_NET_RADIO */
114 #include <asm/current.h>
116 /* This define, if set, will randomly drop a packet when congestion
117 * is more than moderate. It helps fairness in the multi-interface
118 * case when one of them is a hog, but it kills performance for the
119 * single interface case so it is off now by default.
123 /* Setting this will sample the queue lengths and thus congestion
124 * via a timer instead of as each packet is received.
126 #undef OFFLINE_SAMPLE
129 * The list of packet types we will receive (as opposed to discard)
130 * and the routines to invoke.
132 * Why 16. Because with 16 the only overlap we get on a hash of the
133 * low nibble of the protocol value is RARP/SNAP/X.25.
135 * NOTE: That is no longer true with the addition of VLAN tags. Not
136 * sure which should go first, but I bet it won't make much
137 * difference if we are running VLANs. The good news is that
138 * this protocol won't be in the list unless compiled in, so
139 * the average user (w/out VLANs) will not be adversly affected.
156 static spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
157 static struct list_head ptype_base[16]; /* 16 way hashed list */
158 static struct list_head ptype_all; /* Taps */
160 #ifdef OFFLINE_SAMPLE
161 static void sample_queue(unsigned long dummy);
162 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
166 * The @dev_base list is protected by @dev_base_lock and the rtln
169 * Pure readers hold dev_base_lock for reading.
171 * Writers must hold the rtnl semaphore while they loop through the
172 * dev_base list, and hold dev_base_lock for writing when they do the
173 * actual updates. This allows pure readers to access the list even
174 * while a writer is preparing to update it.
176 * To put it another way, dev_base_lock is held for writing only to
177 * protect against pure readers; the rtnl semaphore provides the
178 * protection against other writers.
180 * See, for example usages, register_netdevice() and
181 * unregister_netdevice(), which must be called with the rtnl
184 struct net_device *dev_base;
185 struct net_device **dev_tail = &dev_base;
186 rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
188 EXPORT_SYMBOL(dev_base);
189 EXPORT_SYMBOL(dev_base_lock);
191 #define NETDEV_HASHBITS 8
192 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
193 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
195 static inline struct hlist_head *dev_name_hash(const char *name)
197 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
198 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
201 static inline struct hlist_head *dev_index_hash(int ifindex)
203 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
210 static struct notifier_block *netdev_chain;
213 * Device drivers call our routines to queue packets here. We empty the
214 * queue in the local softnet handler.
216 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
218 #ifdef CONFIG_NET_FASTROUTE
219 int netdev_fastroute;
220 int netdev_fastroute_obstacles;
224 extern int netdev_sysfs_init(void);
225 extern int netdev_register_sysfs(struct net_device *);
226 extern void netdev_unregister_sysfs(struct net_device *);
228 #define netdev_sysfs_init() (0)
229 #define netdev_register_sysfs(dev) (0)
230 #define netdev_unregister_sysfs(dev) do { } while(0)
234 /*******************************************************************************
236 Protocol management and registration routines
238 *******************************************************************************/
247 * Add a protocol ID to the list. Now that the input handler is
248 * smarter we can dispense with all the messy stuff that used to be
251 * BEWARE!!! Protocol handlers, mangling input packets,
252 * MUST BE last in hash buckets and checking protocol handlers
253 * MUST start from promiscuous ptype_all chain in net_bh.
254 * It is true now, do not change it.
255 * Explanation follows: if protocol handler, mangling packet, will
256 * be the first on list, it is not able to sense, that packet
257 * is cloned and should be copied-on-write, so that it will
258 * change it and subsequent readers will get broken packet.
263 * dev_add_pack - add packet handler
264 * @pt: packet type declaration
266 * Add a protocol handler to the networking stack. The passed &packet_type
267 * is linked into kernel lists and may not be freed until it has been
268 * removed from the kernel lists.
270 * This call does not sleep therefore it can not
271 * guarantee all CPU's that are in middle of receiving packets
272 * will see the new packet type (until the next received packet).
275 void dev_add_pack(struct packet_type *pt)
279 spin_lock_bh(&ptype_lock);
280 #ifdef CONFIG_NET_FASTROUTE
281 if (pt->af_packet_priv) {
282 netdev_fastroute_obstacles++;
283 dev_clear_fastroute(pt->dev);
286 if (pt->type == htons(ETH_P_ALL)) {
288 list_add_rcu(&pt->list, &ptype_all);
290 hash = ntohs(pt->type) & 15;
291 list_add_rcu(&pt->list, &ptype_base[hash]);
293 spin_unlock_bh(&ptype_lock);
296 extern void linkwatch_run_queue(void);
301 * __dev_remove_pack - remove packet handler
302 * @pt: packet type declaration
304 * Remove a protocol handler that was previously added to the kernel
305 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
306 * from the kernel lists and can be freed or reused once this function
309 * The packet type might still be in use by receivers
310 * and must not be freed until after all the CPU's have gone
311 * through a quiescent state.
313 void __dev_remove_pack(struct packet_type *pt)
315 struct list_head *head;
316 struct packet_type *pt1;
318 spin_lock_bh(&ptype_lock);
320 if (pt->type == htons(ETH_P_ALL)) {
324 head = &ptype_base[ntohs(pt->type) & 15];
326 list_for_each_entry(pt1, head, list) {
328 #ifdef CONFIG_NET_FASTROUTE
329 if (pt->af_packet_priv)
330 netdev_fastroute_obstacles--;
332 list_del_rcu(&pt->list);
337 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
339 spin_unlock_bh(&ptype_lock);
342 * dev_remove_pack - remove packet handler
343 * @pt: packet type declaration
345 * Remove a protocol handler that was previously added to the kernel
346 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
347 * from the kernel lists and can be freed or reused once this function
350 * This call sleeps to guarantee that no CPU is looking at the packet
353 void dev_remove_pack(struct packet_type *pt)
355 __dev_remove_pack(pt);
360 /******************************************************************************
362 Device Boot-time Settings Routines
364 *******************************************************************************/
366 /* Boot time configuration table */
367 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
370 * netdev_boot_setup_add - add new setup entry
371 * @name: name of the device
372 * @map: configured settings for the device
374 * Adds new setup entry to the dev_boot_setup list. The function
375 * returns 0 on error and 1 on success. This is a generic routine to
378 int netdev_boot_setup_add(char *name, struct ifmap *map)
380 struct netdev_boot_setup *s;
384 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
385 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
386 memset(s[i].name, 0, sizeof(s[i].name));
387 strcpy(s[i].name, name);
388 memcpy(&s[i].map, map, sizeof(s[i].map));
393 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
397 * netdev_boot_setup_check - check boot time settings
398 * @dev: the netdevice
400 * Check boot time settings for the device.
401 * The found settings are set for the device to be used
402 * later in the device probing.
403 * Returns 0 if no settings found, 1 if they are.
405 int netdev_boot_setup_check(struct net_device *dev)
407 struct netdev_boot_setup *s = dev_boot_setup;
410 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
411 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
412 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
413 dev->irq = s[i].map.irq;
414 dev->base_addr = s[i].map.base_addr;
415 dev->mem_start = s[i].map.mem_start;
416 dev->mem_end = s[i].map.mem_end;
425 * netdev_boot_base - get address from boot time settings
426 * @prefix: prefix for network device
427 * @unit: id for network device
429 * Check boot time settings for the base address of device.
430 * The found settings are set for the device to be used
431 * later in the device probing.
432 * Returns 0 if no settings found.
434 unsigned long netdev_boot_base(const char *prefix, int unit)
436 const struct netdev_boot_setup *s = dev_boot_setup;
440 sprintf(name, "%s%d", prefix, unit);
443 * If device already registered then return base of 1
444 * to indicate not to probe for this interface
446 if (__dev_get_by_name(name))
449 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
450 if (!strcmp(name, s[i].name))
451 return s[i].map.base_addr;
456 * Saves at boot time configured settings for any netdevice.
458 int __init netdev_boot_setup(char *str)
463 str = get_options(str, ARRAY_SIZE(ints), ints);
468 memset(&map, 0, sizeof(map));
472 map.base_addr = ints[2];
474 map.mem_start = ints[3];
476 map.mem_end = ints[4];
478 /* Add new entry to the list */
479 return netdev_boot_setup_add(str, &map);
482 __setup("netdev=", netdev_boot_setup);
484 /*******************************************************************************
486 Device Interface Subroutines
488 *******************************************************************************/
491 * __dev_get_by_name - find a device by its name
492 * @name: name to find
494 * Find an interface by name. Must be called under RTNL semaphore
495 * or @dev_base_lock. If the name is found a pointer to the device
496 * is returned. If the name is not found then %NULL is returned. The
497 * reference counters are not incremented so the caller must be
498 * careful with locks.
501 struct net_device *__dev_get_by_name(const char *name)
503 struct hlist_node *p;
505 hlist_for_each(p, dev_name_hash(name)) {
506 struct net_device *dev
507 = hlist_entry(p, struct net_device, name_hlist);
508 if (!strncmp(dev->name, name, IFNAMSIZ))
515 * dev_get_by_name - find a device by its name
516 * @name: name to find
518 * Find an interface by name. This can be called from any
519 * context and does its own locking. The returned handle has
520 * the usage count incremented and the caller must use dev_put() to
521 * release it when it is no longer needed. %NULL is returned if no
522 * matching device is found.
525 struct net_device *dev_get_by_name(const char *name)
527 struct net_device *dev;
529 read_lock(&dev_base_lock);
530 dev = __dev_get_by_name(name);
533 read_unlock(&dev_base_lock);
538 Return value is changed to int to prevent illegal usage in future.
539 It is still legal to use to check for device existence.
541 User should understand, that the result returned by this function
542 is meaningless, if it was not issued under rtnl semaphore.
546 * dev_get - test if a device exists
547 * @name: name to test for
549 * Test if a name exists. Returns true if the name is found. In order
550 * to be sure the name is not allocated or removed during the test the
551 * caller must hold the rtnl semaphore.
553 * This function exists only for back compatibility with older
556 int __dev_get(const char *name)
558 struct net_device *dev;
560 read_lock(&dev_base_lock);
561 dev = __dev_get_by_name(name);
562 read_unlock(&dev_base_lock);
567 * __dev_get_by_index - find a device by its ifindex
568 * @ifindex: index of device
570 * Search for an interface by index. Returns %NULL if the device
571 * is not found or a pointer to the device. The device has not
572 * had its reference counter increased so the caller must be careful
573 * about locking. The caller must hold either the RTNL semaphore
577 struct net_device *__dev_get_by_index(int ifindex)
579 struct hlist_node *p;
581 hlist_for_each(p, dev_index_hash(ifindex)) {
582 struct net_device *dev
583 = hlist_entry(p, struct net_device, index_hlist);
584 if (dev->ifindex == ifindex)
592 * dev_get_by_index - find a device by its ifindex
593 * @ifindex: index of device
595 * Search for an interface by index. Returns NULL if the device
596 * is not found or a pointer to the device. The device returned has
597 * had a reference added and the pointer is safe until the user calls
598 * dev_put to indicate they have finished with it.
601 struct net_device *dev_get_by_index(int ifindex)
603 struct net_device *dev;
605 read_lock(&dev_base_lock);
606 dev = __dev_get_by_index(ifindex);
609 read_unlock(&dev_base_lock);
614 * dev_getbyhwaddr - find a device by its hardware address
615 * @type: media type of device
616 * @ha: hardware address
618 * Search for an interface by MAC address. Returns NULL if the device
619 * is not found or a pointer to the device. The caller must hold the
620 * rtnl semaphore. The returned device has not had its ref count increased
621 * and the caller must therefore be careful about locking
624 * If the API was consistent this would be __dev_get_by_hwaddr
627 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
629 struct net_device *dev;
633 for (dev = dev_base; dev; dev = dev->next)
634 if (dev->type == type &&
635 !memcmp(dev->dev_addr, ha, dev->addr_len))
640 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
642 struct net_device *dev;
644 for (dev = dev_base; dev; dev = dev->next)
645 if (dev->type == type)
650 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
652 struct net_device *dev_getfirstbyhwtype(unsigned short type)
654 struct net_device *dev;
657 dev = __dev_getfirstbyhwtype(type);
664 EXPORT_SYMBOL(dev_getfirstbyhwtype);
667 * dev_get_by_flags - find any device with given flags
668 * @if_flags: IFF_* values
669 * @mask: bitmask of bits in if_flags to check
671 * Search for any interface with the given flags. Returns NULL if a device
672 * is not found or a pointer to the device. The device returned has
673 * had a reference added and the pointer is safe until the user calls
674 * dev_put to indicate they have finished with it.
677 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
679 struct net_device *dev;
681 read_lock(&dev_base_lock);
682 dev = __dev_get_by_flags(if_flags, mask);
685 read_unlock(&dev_base_lock);
690 * __dev_get_by_flags - find any device with given flags
691 * @if_flags: IFF_* values
692 * @mask: bitmask of bits in if_flags to check
694 * Search for any interface with the given flags. Returns NULL if a device
695 * is not found or a pointer to the device. The caller must hold either
696 * the RTNL semaphore or @dev_base_lock.
699 struct net_device *__dev_get_by_flags(unsigned short if_flags, unsigned short mask)
701 struct net_device *dev;
703 for (dev = dev_base; dev != NULL; dev = dev->next) {
704 if (((dev->flags ^ if_flags) & mask) == 0)
711 * dev_valid_name - check if name is okay for network device
714 * Network device names need to be valid file names to
715 * to allow sysfs to work
717 int dev_valid_name(const char *name)
719 return !(*name == '\0'
720 || !strcmp(name, ".")
721 || !strcmp(name, "..")
722 || strchr(name, '/'));
726 * dev_alloc_name - allocate a name for a device
728 * @name: name format string
730 * Passed a format string - eg "lt%d" it will try and find a suitable
731 * id. Not efficient for many devices, not called a lot. The caller
732 * must hold the dev_base or rtnl lock while allocating the name and
733 * adding the device in order to avoid duplicates. Returns the number
734 * of the unit assigned or a negative errno code.
737 int dev_alloc_name(struct net_device *dev, const char *name)
742 const int max_netdevices = 8*PAGE_SIZE;
744 struct net_device *d;
746 p = strnchr(name, IFNAMSIZ-1, '%');
749 * Verify the string as this thing may have come from
750 * the user. There must be either one "%d" and no other "%"
753 if (p[1] != 'd' || strchr(p + 2, '%'))
756 /* Use one page as a bit array of possible slots */
757 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
761 for (d = dev_base; d; d = d->next) {
762 if (!sscanf(d->name, name, &i))
764 if (i < 0 || i >= max_netdevices)
767 /* avoid cases where sscanf is not exact inverse of printf */
768 snprintf(buf, sizeof(buf), name, i);
769 if (!strncmp(buf, d->name, IFNAMSIZ))
773 i = find_first_zero_bit(inuse, max_netdevices);
774 free_page((unsigned long) inuse);
777 snprintf(buf, sizeof(buf), name, i);
778 if (!__dev_get_by_name(buf)) {
779 strlcpy(dev->name, buf, IFNAMSIZ);
783 /* It is possible to run out of possible slots
784 * when the name is long and there isn't enough space left
785 * for the digits, or if all bits are used.
792 * dev_change_name - change name of a device
794 * @newname: name (or format string) must be at least IFNAMSIZ
796 * Change name of a device, can pass format strings "eth%d".
799 int dev_change_name(struct net_device *dev, char *newname)
805 if (dev->flags & IFF_UP)
808 if (!dev_valid_name(newname))
811 if (strchr(newname, '%')) {
812 err = dev_alloc_name(dev, newname);
815 strcpy(newname, dev->name);
817 else if (__dev_get_by_name(newname))
820 strlcpy(dev->name, newname, IFNAMSIZ);
822 err = class_device_rename(&dev->class_dev, dev->name);
824 hlist_del(&dev->name_hlist);
825 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
826 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
833 * netdev_state_change - device changes state
834 * @dev: device to cause notification
836 * Called to indicate a device has changed state. This function calls
837 * the notifier chains for netdev_chain and sends a NEWLINK message
838 * to the routing socket.
840 void netdev_state_change(struct net_device *dev)
842 if (dev->flags & IFF_UP) {
843 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
844 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
849 * dev_load - load a network module
850 * @name: name of interface
852 * If a network interface is not present and the process has suitable
853 * privileges this function loads the module. If module loading is not
854 * available in this kernel then it becomes a nop.
857 void dev_load(const char *name)
859 struct net_device *dev;
861 read_lock(&dev_base_lock);
862 dev = __dev_get_by_name(name);
863 read_unlock(&dev_base_lock);
865 if (!dev && capable(CAP_SYS_MODULE))
866 request_module("%s", name);
869 static int default_rebuild_header(struct sk_buff *skb)
871 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
872 skb->dev ? skb->dev->name : "NULL!!!");
879 * Some old buggy device drivers change get_stats after registering
880 * the device. Try and trap them here.
881 * This can be elimnated when all devices are known fixed.
883 static inline int get_stats_changed(struct net_device *dev)
885 int changed = dev->last_stats != dev->get_stats;
886 dev->last_stats = dev->get_stats;
891 * dev_open - prepare an interface for use.
892 * @dev: device to open
894 * Takes a device from down to up state. The device's private open
895 * function is invoked and then the multicast lists are loaded. Finally
896 * the device is moved into the up state and a %NETDEV_UP message is
897 * sent to the netdev notifier chain.
899 * Calling this function on an active interface is a nop. On a failure
900 * a negative errno code is returned.
902 int dev_open(struct net_device *dev)
910 if (dev->flags & IFF_UP)
914 * Check for broken device drivers.
916 if (get_stats_changed(dev) && net_ratelimit()) {
917 printk(KERN_ERR "%s: driver changed get_stats after register\n",
922 * Is it even present?
924 if (!netif_device_present(dev))
928 * Call device private open method
930 set_bit(__LINK_STATE_START, &dev->state);
932 ret = dev->open(dev);
934 clear_bit(__LINK_STATE_START, &dev->state);
938 * Check for more broken device drivers.
940 if (get_stats_changed(dev) && net_ratelimit()) {
941 printk(KERN_ERR "%s: driver changed get_stats in open\n",
946 * If it went open OK then:
953 dev->flags |= IFF_UP;
956 * Initialize multicasting status
961 * Wakeup transmit queue engine
966 * ... and announce new interface.
968 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
973 #ifdef CONFIG_NET_FASTROUTE
975 static void dev_do_clear_fastroute(struct net_device *dev)
977 if (dev->accept_fastpath) {
980 for (i = 0; i <= NETDEV_FASTROUTE_HMASK; i++) {
981 struct dst_entry *dst;
983 write_lock_irq(&dev->fastpath_lock);
984 dst = dev->fastpath[i];
985 dev->fastpath[i] = NULL;
986 write_unlock_irq(&dev->fastpath_lock);
993 void dev_clear_fastroute(struct net_device *dev)
996 dev_do_clear_fastroute(dev);
998 read_lock(&dev_base_lock);
999 for (dev = dev_base; dev; dev = dev->next)
1000 dev_do_clear_fastroute(dev);
1001 read_unlock(&dev_base_lock);
1007 * dev_close - shutdown an interface.
1008 * @dev: device to shutdown
1010 * This function moves an active device into down state. A
1011 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1012 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1015 int dev_close(struct net_device *dev)
1017 if (!(dev->flags & IFF_UP))
1021 * Tell people we are going down, so that they can
1022 * prepare to death, when device is still operating.
1024 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
1026 dev_deactivate(dev);
1028 clear_bit(__LINK_STATE_START, &dev->state);
1030 /* Synchronize to scheduled poll. We cannot touch poll list,
1031 * it can be even on different cpu. So just clear netif_running(),
1032 * and wait when poll really will happen. Actually, the best place
1033 * for this is inside dev->stop() after device stopped its irq
1034 * engine, but this requires more changes in devices. */
1036 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1037 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
1039 current->state = TASK_INTERRUPTIBLE;
1040 schedule_timeout(1);
1044 * Call the device specific close. This cannot fail.
1045 * Only if device is UP
1047 * We allow it to be called even after a DETACH hot-plug
1054 * Device is now down.
1057 dev->flags &= ~IFF_UP;
1058 #ifdef CONFIG_NET_FASTROUTE
1059 dev_clear_fastroute(dev);
1063 * Tell people we are down
1065 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1072 * Device change register/unregister. These are not inline or static
1073 * as we export them to the world.
1077 * register_netdevice_notifier - register a network notifier block
1080 * Register a notifier to be called when network device events occur.
1081 * The notifier passed is linked into the kernel structures and must
1082 * not be reused until it has been unregistered. A negative errno code
1083 * is returned on a failure.
1085 * When registered all registration and up events are replayed
1086 * to the new notifier to allow device to have a race free
1087 * view of the network device list.
1090 int register_netdevice_notifier(struct notifier_block *nb)
1092 struct net_device *dev;
1096 err = notifier_chain_register(&netdev_chain, nb);
1098 for (dev = dev_base; dev; dev = dev->next) {
1099 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1101 if (dev->flags & IFF_UP)
1102 nb->notifier_call(nb, NETDEV_UP, dev);
1110 * unregister_netdevice_notifier - unregister a network notifier block
1113 * Unregister a notifier previously registered by
1114 * register_netdevice_notifier(). The notifier is unlinked into the
1115 * kernel structures and may then be reused. A negative errno code
1116 * is returned on a failure.
1119 int unregister_netdevice_notifier(struct notifier_block *nb)
1121 return notifier_chain_unregister(&netdev_chain, nb);
1125 * call_netdevice_notifiers - call all network notifier blocks
1126 * @val: value passed unmodified to notifier function
1127 * @v: pointer passed unmodified to notifier function
1129 * Call all network notifier blocks. Parameters and return value
1130 * are as for notifier_call_chain().
1133 int call_netdevice_notifiers(unsigned long val, void *v)
1135 return notifier_call_chain(&netdev_chain, val, v);
1139 * Support routine. Sends outgoing frames to any network
1140 * taps currently in use.
1143 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1145 struct packet_type *ptype;
1146 net_timestamp(&skb->stamp);
1149 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1150 /* Never send packets back to the socket
1151 * they originated from - MvS (miquels@drinkel.ow.org)
1153 if ((ptype->dev == dev || !ptype->dev) &&
1154 (ptype->af_packet_priv == NULL ||
1155 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1156 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1160 /* skb->nh should be correctly
1161 set by sender, so that the second statement is
1162 just protection against buggy protocols.
1164 skb2->mac.raw = skb2->data;
1166 if (skb2->nh.raw < skb2->data ||
1167 skb2->nh.raw > skb2->tail) {
1168 if (net_ratelimit())
1169 printk(KERN_CRIT "protocol %04x is "
1171 skb2->protocol, dev->name);
1172 skb2->nh.raw = skb2->data;
1175 skb2->h.raw = skb2->nh.raw;
1176 skb2->pkt_type = PACKET_OUTGOING;
1177 ptype->func(skb2, skb->dev, ptype);
1184 * Invalidate hardware checksum when packet is to be mangled, and
1185 * complete checksum manually on outgoing path.
1187 int skb_checksum_help(struct sk_buff **pskb, int inward)
1190 int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1193 (*pskb)->ip_summed = CHECKSUM_NONE;
1197 if (skb_shared(*pskb) || skb_cloned(*pskb)) {
1198 struct sk_buff *newskb = skb_copy(*pskb, GFP_ATOMIC);
1204 skb_set_owner_w(newskb, (*pskb)->sk);
1209 if (offset > (int)(*pskb)->len)
1211 csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1213 offset = (*pskb)->tail - (*pskb)->h.raw;
1216 if ((*pskb)->csum + 2 > offset)
1219 *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1220 (*pskb)->ip_summed = CHECKSUM_NONE;
1225 #ifdef CONFIG_HIGHMEM
1226 /* Actually, we should eliminate this check as soon as we know, that:
1227 * 1. IOMMU is present and allows to map all the memory.
1228 * 2. No high memory really exists on this machine.
1231 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1235 if (dev->features & NETIF_F_HIGHDMA)
1238 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1239 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1245 #define illegal_highdma(dev, skb) (0)
1248 extern void skb_release_data(struct sk_buff *);
1250 /* Keep head the same: replace data */
1251 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1256 struct skb_shared_info *ninfo;
1257 int headerlen = skb->data - skb->head;
1258 int expand = (skb->tail + skb->data_len) - skb->end;
1260 if (skb_shared(skb))
1266 size = skb->end - skb->head + expand;
1267 size = SKB_DATA_ALIGN(size);
1268 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1272 /* Copy entire thing */
1273 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1277 ninfo = (struct skb_shared_info*)(data + size);
1278 atomic_set(&ninfo->dataref, 1);
1279 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1280 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1281 ninfo->nr_frags = 0;
1282 ninfo->frag_list = NULL;
1284 /* Offset between the two in bytes */
1285 offset = data - skb->head;
1287 /* Free old data. */
1288 skb_release_data(skb);
1291 skb->end = data + size;
1293 /* Set up new pointers */
1294 skb->h.raw += offset;
1295 skb->nh.raw += offset;
1296 skb->mac.raw += offset;
1297 skb->tail += offset;
1298 skb->data += offset;
1300 /* We are no longer a clone, even if we were. */
1303 skb->tail += skb->data_len;
1309 * dev_queue_xmit - transmit a buffer
1310 * @skb: buffer to transmit
1312 * Queue a buffer for transmission to a network device. The caller must
1313 * have set the device and priority and built the buffer before calling
1314 * this function. The function can be called from an interrupt.
1316 * A negative errno code is returned on a failure. A success does not
1317 * guarantee the frame will be transmitted as it may be dropped due
1318 * to congestion or traffic shaping.
1321 int dev_queue_xmit(struct sk_buff *skb)
1323 struct net_device *dev = skb->dev;
1327 if (skb_shinfo(skb)->frag_list &&
1328 !(dev->features & NETIF_F_FRAGLIST) &&
1329 __skb_linearize(skb, GFP_ATOMIC))
1332 /* Fragmented skb is linearized if device does not support SG,
1333 * or if at least one of fragments is in highmem and device
1334 * does not support DMA from it.
1336 if (skb_shinfo(skb)->nr_frags &&
1337 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1338 __skb_linearize(skb, GFP_ATOMIC))
1341 /* If packet is not checksummed and device does not support
1342 * checksumming for this protocol, complete checksumming here.
1344 if (skb->ip_summed == CHECKSUM_HW &&
1345 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1346 (!(dev->features & NETIF_F_IP_CSUM) ||
1347 skb->protocol != htons(ETH_P_IP))))
1348 if (skb_checksum_help(&skb, 0))
1351 /* Grab device queue */
1352 spin_lock_bh(&dev->queue_lock);
1355 rc = q->enqueue(skb, q);
1359 spin_unlock_bh(&dev->queue_lock);
1360 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1364 /* The device has no queue. Common case for software devices:
1365 loopback, all the sorts of tunnels...
1367 Really, it is unlikely that xmit_lock protection is necessary here.
1368 (f.e. loopback and IP tunnels are clean ignoring statistics
1370 However, it is possible, that they rely on protection
1373 Check this and shot the lock. It is not prone from deadlocks.
1374 Either shot noqueue qdisc, it is even simpler 8)
1376 if (dev->flags & IFF_UP) {
1377 int cpu = smp_processor_id();
1379 if (dev->xmit_lock_owner != cpu) {
1381 * The spin_lock effectivly does a preempt lock, but
1382 * we are about to drop that...
1385 spin_unlock(&dev->queue_lock);
1386 spin_lock(&dev->xmit_lock);
1387 dev->xmit_lock_owner = cpu;
1390 if (!netif_queue_stopped(dev)) {
1392 dev_queue_xmit_nit(skb, dev);
1395 if (!dev->hard_start_xmit(skb, dev)) {
1396 dev->xmit_lock_owner = -1;
1397 spin_unlock_bh(&dev->xmit_lock);
1401 dev->xmit_lock_owner = -1;
1402 spin_unlock_bh(&dev->xmit_lock);
1403 if (net_ratelimit())
1404 printk(KERN_CRIT "Virtual device %s asks to "
1405 "queue packet!\n", dev->name);
1408 /* Recursion is detected! It is possible,
1410 if (net_ratelimit())
1411 printk(KERN_CRIT "Dead loop on virtual device "
1412 "%s, fix it urgently!\n", dev->name);
1415 spin_unlock_bh(&dev->queue_lock);
1425 /*=======================================================================
1427 =======================================================================*/
1429 int netdev_max_backlog = 300;
1430 int weight_p = 64; /* old backlog weight */
1431 /* These numbers are selected based on intuition and some
1432 * experimentatiom, if you have more scientific way of doing this
1433 * please go ahead and fix things.
1435 int no_cong_thresh = 10;
1440 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1443 #ifdef CONFIG_NET_HW_FLOWCONTROL
1444 atomic_t netdev_dropping = ATOMIC_INIT(0);
1445 static unsigned long netdev_fc_mask = 1;
1446 unsigned long netdev_fc_xoff;
1447 spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1451 void (*stimul)(struct net_device *);
1452 struct net_device *dev;
1453 } netdev_fc_slots[BITS_PER_LONG];
1455 int netdev_register_fc(struct net_device *dev,
1456 void (*stimul)(struct net_device *dev))
1459 unsigned long flags;
1461 spin_lock_irqsave(&netdev_fc_lock, flags);
1462 if (netdev_fc_mask != ~0UL) {
1463 bit = ffz(netdev_fc_mask);
1464 netdev_fc_slots[bit].stimul = stimul;
1465 netdev_fc_slots[bit].dev = dev;
1466 set_bit(bit, &netdev_fc_mask);
1467 clear_bit(bit, &netdev_fc_xoff);
1469 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1473 void netdev_unregister_fc(int bit)
1475 unsigned long flags;
1477 spin_lock_irqsave(&netdev_fc_lock, flags);
1479 netdev_fc_slots[bit].stimul = NULL;
1480 netdev_fc_slots[bit].dev = NULL;
1481 clear_bit(bit, &netdev_fc_mask);
1482 clear_bit(bit, &netdev_fc_xoff);
1484 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1487 static void netdev_wakeup(void)
1491 spin_lock(&netdev_fc_lock);
1492 xoff = netdev_fc_xoff;
1497 netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1499 spin_unlock(&netdev_fc_lock);
1503 static void get_sample_stats(int cpu)
1509 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1510 int blog = sd->input_pkt_queue.qlen;
1511 int avg_blog = sd->avg_blog;
1513 avg_blog = (avg_blog >> 1) + (blog >> 1);
1515 if (avg_blog > mod_cong) {
1516 /* Above moderate congestion levels. */
1517 sd->cng_level = NET_RX_CN_HIGH;
1520 rq = rd % netdev_max_backlog;
1521 if (rq < avg_blog) /* unlucky bastard */
1522 sd->cng_level = NET_RX_DROP;
1524 } else if (avg_blog > lo_cong) {
1525 sd->cng_level = NET_RX_CN_MOD;
1528 rq = rd % netdev_max_backlog;
1529 if (rq < avg_blog) /* unlucky bastard */
1530 sd->cng_level = NET_RX_CN_HIGH;
1532 } else if (avg_blog > no_cong)
1533 sd->cng_level = NET_RX_CN_LOW;
1534 else /* no congestion */
1535 sd->cng_level = NET_RX_SUCCESS;
1537 sd->avg_blog = avg_blog;
1540 #ifdef OFFLINE_SAMPLE
1541 static void sample_queue(unsigned long dummy)
1543 /* 10 ms 0r 1ms -- i don't care -- JHS */
1545 int cpu = smp_processor_id();
1547 get_sample_stats(cpu);
1548 next_tick += jiffies;
1549 mod_timer(&samp_timer, next_tick);
1555 * netif_rx - post buffer to the network code
1556 * @skb: buffer to post
1558 * This function receives a packet from a device driver and queues it for
1559 * the upper (protocol) levels to process. It always succeeds. The buffer
1560 * may be dropped during processing for congestion control or by the
1564 * NET_RX_SUCCESS (no congestion)
1565 * NET_RX_CN_LOW (low congestion)
1566 * NET_RX_CN_MOD (moderate congestion)
1567 * NET_RX_CN_HIGH (high congestion)
1568 * NET_RX_DROP (packet was dropped)
1572 int netif_rx(struct sk_buff *skb)
1575 struct softnet_data *queue;
1576 unsigned long flags;
1578 #ifdef CONFIG_NETPOLL_RX
1579 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1585 if (!skb->stamp.tv_sec)
1586 net_timestamp(&skb->stamp);
1589 * The code is rearranged so that the path is the most
1590 * short when CPU is congested, but is still operating.
1592 local_irq_save(flags);
1593 this_cpu = smp_processor_id();
1594 queue = &__get_cpu_var(softnet_data);
1596 __get_cpu_var(netdev_rx_stat).total++;
1597 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1598 if (queue->input_pkt_queue.qlen) {
1599 if (queue->throttle)
1604 __skb_queue_tail(&queue->input_pkt_queue, skb);
1605 #ifndef OFFLINE_SAMPLE
1606 get_sample_stats(this_cpu);
1608 local_irq_restore(flags);
1609 return queue->cng_level;
1612 if (queue->throttle) {
1613 queue->throttle = 0;
1614 #ifdef CONFIG_NET_HW_FLOWCONTROL
1615 if (atomic_dec_and_test(&netdev_dropping))
1620 netif_rx_schedule(&queue->backlog_dev);
1624 if (!queue->throttle) {
1625 queue->throttle = 1;
1626 __get_cpu_var(netdev_rx_stat).throttled++;
1627 #ifdef CONFIG_NET_HW_FLOWCONTROL
1628 atomic_inc(&netdev_dropping);
1633 __get_cpu_var(netdev_rx_stat).dropped++;
1634 local_irq_restore(flags);
1640 static __inline__ void skb_bond(struct sk_buff *skb)
1642 struct net_device *dev = skb->dev;
1645 skb->real_dev = skb->dev;
1646 skb->dev = dev->master;
1650 static void net_tx_action(struct softirq_action *h)
1652 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1654 if (sd->completion_queue) {
1655 struct sk_buff *clist;
1657 local_irq_disable();
1658 clist = sd->completion_queue;
1659 sd->completion_queue = NULL;
1663 struct sk_buff *skb = clist;
1664 clist = clist->next;
1666 BUG_TRAP(!atomic_read(&skb->users));
1671 if (sd->output_queue) {
1672 struct net_device *head;
1674 local_irq_disable();
1675 head = sd->output_queue;
1676 sd->output_queue = NULL;
1680 struct net_device *dev = head;
1681 head = head->next_sched;
1683 smp_mb__before_clear_bit();
1684 clear_bit(__LINK_STATE_SCHED, &dev->state);
1686 if (spin_trylock(&dev->queue_lock)) {
1688 spin_unlock(&dev->queue_lock);
1690 netif_schedule(dev);
1696 static __inline__ int deliver_skb(struct sk_buff *skb,
1697 struct packet_type *pt_prev, int last)
1699 atomic_inc(&skb->users);
1700 return pt_prev->func(skb, skb->dev, pt_prev);
1704 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1705 int (*br_handle_frame_hook)(struct sk_buff *skb);
1707 static __inline__ int handle_bridge(struct sk_buff *skb,
1708 struct packet_type *pt_prev)
1710 int ret = NET_RX_DROP;
1712 ret = deliver_skb(skb, pt_prev, 0);
1719 static inline int __handle_bridge(struct sk_buff *skb,
1720 struct packet_type **pt_prev, int *ret)
1722 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
1723 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1724 *ret = handle_bridge(skb, *pt_prev);
1725 if (br_handle_frame_hook(skb) == 0)
1734 int netif_receive_skb(struct sk_buff *skb)
1736 struct packet_type *ptype, *pt_prev;
1737 int ret = NET_RX_DROP;
1738 unsigned short type;
1740 #ifdef CONFIG_NETPOLL_RX
1741 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1747 if (!skb->stamp.tv_sec)
1748 net_timestamp(&skb->stamp);
1752 __get_cpu_var(netdev_rx_stat).total++;
1754 #ifdef CONFIG_NET_FASTROUTE
1755 if (skb->pkt_type == PACKET_FASTROUTE) {
1756 __get_cpu_var(netdev_rx_stat).fastroute_deferred_out++;
1757 return dev_queue_xmit(skb);
1761 skb->h.raw = skb->nh.raw = skb->data;
1762 skb->mac_len = skb->nh.raw - skb->mac.raw;
1766 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1767 if (!ptype->dev || ptype->dev == skb->dev) {
1769 ret = deliver_skb(skb, pt_prev, 0);
1774 handle_diverter(skb);
1776 if (__handle_bridge(skb, &pt_prev, &ret))
1779 type = skb->protocol;
1780 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1781 if (ptype->type == type &&
1782 (!ptype->dev || ptype->dev == skb->dev)) {
1784 ret = deliver_skb(skb, pt_prev, 0);
1790 ret = pt_prev->func(skb, skb->dev, pt_prev);
1793 /* Jamal, now you will not able to escape explaining
1794 * me how you were going to use this. :-)
1804 static int process_backlog(struct net_device *backlog_dev, int *budget)
1807 int quota = min(backlog_dev->quota, *budget);
1808 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1809 unsigned long start_time = jiffies;
1812 struct sk_buff *skb;
1813 struct net_device *dev;
1815 local_irq_disable();
1816 skb = __skb_dequeue(&queue->input_pkt_queue);
1823 netif_receive_skb(skb);
1829 if (work >= quota || jiffies - start_time > 1)
1832 #ifdef CONFIG_NET_HW_FLOWCONTROL
1833 if (queue->throttle &&
1834 queue->input_pkt_queue.qlen < no_cong_thresh ) {
1835 queue->throttle = 0;
1836 if (atomic_dec_and_test(&netdev_dropping)) {
1844 backlog_dev->quota -= work;
1849 backlog_dev->quota -= work;
1852 list_del(&backlog_dev->poll_list);
1853 smp_mb__before_clear_bit();
1854 netif_poll_enable(backlog_dev);
1856 if (queue->throttle) {
1857 queue->throttle = 0;
1858 #ifdef CONFIG_NET_HW_FLOWCONTROL
1859 if (atomic_dec_and_test(&netdev_dropping))
1867 static void net_rx_action(struct softirq_action *h)
1869 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1870 unsigned long start_time = jiffies;
1871 int budget = netdev_max_backlog;
1874 local_irq_disable();
1876 while (!list_empty(&queue->poll_list)) {
1877 struct net_device *dev;
1879 if (budget <= 0 || jiffies - start_time > 1)
1884 dev = list_entry(queue->poll_list.next,
1885 struct net_device, poll_list);
1887 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1888 local_irq_disable();
1889 list_del(&dev->poll_list);
1890 list_add_tail(&dev->poll_list, &queue->poll_list);
1892 dev->quota += dev->weight;
1894 dev->quota = dev->weight;
1897 local_irq_disable();
1905 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1906 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1910 static gifconf_func_t * gifconf_list [NPROTO];
1913 * register_gifconf - register a SIOCGIF handler
1914 * @family: Address family
1915 * @gifconf: Function handler
1917 * Register protocol dependent address dumping routines. The handler
1918 * that is passed must not be freed or reused until it has been replaced
1919 * by another handler.
1921 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1923 if (family >= NPROTO)
1925 gifconf_list[family] = gifconf;
1931 * Map an interface index to its name (SIOCGIFNAME)
1935 * We need this ioctl for efficient implementation of the
1936 * if_indextoname() function required by the IPv6 API. Without
1937 * it, we would have to search all the interfaces to find a
1941 static int dev_ifname(struct ifreq __user *arg)
1943 struct net_device *dev;
1947 * Fetch the caller's info block.
1950 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1953 read_lock(&dev_base_lock);
1954 dev = __dev_get_by_index(ifr.ifr_ifindex);
1956 read_unlock(&dev_base_lock);
1960 strcpy(ifr.ifr_name, dev->name);
1961 read_unlock(&dev_base_lock);
1963 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1969 * Perform a SIOCGIFCONF call. This structure will change
1970 * size eventually, and there is nothing I can do about it.
1971 * Thus we will need a 'compatibility mode'.
1974 static int dev_ifconf(char __user *arg)
1977 struct net_device *dev;
1984 * Fetch the caller's info block.
1987 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1994 * Loop over the interfaces, and write an info block for each.
1998 for (dev = dev_base; dev; dev = dev->next) {
1999 if (!dev_in_nx_info(dev, current->nx_info))
2001 for (i = 0; i < NPROTO; i++) {
2002 if (gifconf_list[i]) {
2005 done = gifconf_list[i](dev, NULL, 0);
2007 done = gifconf_list[i](dev, pos + total,
2017 * All done. Write the updated control block back to the caller.
2019 ifc.ifc_len = total;
2022 * Both BSD and Solaris return 0 here, so we do too.
2024 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2027 #ifdef CONFIG_PROC_FS
2029 * This is invoked by the /proc filesystem handler to display a device
2032 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2034 struct net_device *dev;
2037 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2039 return i == pos ? dev : NULL;
2042 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2044 read_lock(&dev_base_lock);
2045 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2048 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2051 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2054 void dev_seq_stop(struct seq_file *seq, void *v)
2056 read_unlock(&dev_base_lock);
2059 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2061 struct nx_info *nxi = current->nx_info;
2063 if (!dev_in_nx_info(dev, nxi))
2065 if (dev->get_stats) {
2066 struct net_device_stats *stats = dev->get_stats(dev);
2068 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2069 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2070 dev->name, stats->rx_bytes, stats->rx_packets,
2072 stats->rx_dropped + stats->rx_missed_errors,
2073 stats->rx_fifo_errors,
2074 stats->rx_length_errors + stats->rx_over_errors +
2075 stats->rx_crc_errors + stats->rx_frame_errors,
2076 stats->rx_compressed, stats->multicast,
2077 stats->tx_bytes, stats->tx_packets,
2078 stats->tx_errors, stats->tx_dropped,
2079 stats->tx_fifo_errors, stats->collisions,
2080 stats->tx_carrier_errors +
2081 stats->tx_aborted_errors +
2082 stats->tx_window_errors +
2083 stats->tx_heartbeat_errors,
2084 stats->tx_compressed);
2086 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2090 * Called from the PROCfs module. This now uses the new arbitrary sized
2091 * /proc/net interface to create /proc/net/dev
2093 static int dev_seq_show(struct seq_file *seq, void *v)
2095 if (v == SEQ_START_TOKEN)
2096 seq_puts(seq, "Inter-| Receive "
2098 " face |bytes packets errs drop fifo frame "
2099 "compressed multicast|bytes packets errs "
2100 "drop fifo colls carrier compressed\n");
2102 dev_seq_printf_stats(seq, v);
2106 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2108 struct netif_rx_stats *rc = NULL;
2110 while (*pos < NR_CPUS)
2111 if (cpu_online(*pos)) {
2112 rc = &per_cpu(netdev_rx_stat, *pos);
2119 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2121 return softnet_get_online(pos);
2124 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2127 return softnet_get_online(pos);
2130 static void softnet_seq_stop(struct seq_file *seq, void *v)
2134 static int softnet_seq_show(struct seq_file *seq, void *v)
2136 struct netif_rx_stats *s = v;
2138 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2139 s->total, s->dropped, s->time_squeeze, s->throttled,
2140 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2141 s->fastroute_deferred_out,
2143 s->fastroute_latency_reduction
2151 static struct seq_operations dev_seq_ops = {
2152 .start = dev_seq_start,
2153 .next = dev_seq_next,
2154 .stop = dev_seq_stop,
2155 .show = dev_seq_show,
2158 static int dev_seq_open(struct inode *inode, struct file *file)
2160 return seq_open(file, &dev_seq_ops);
2163 static struct file_operations dev_seq_fops = {
2164 .owner = THIS_MODULE,
2165 .open = dev_seq_open,
2167 .llseek = seq_lseek,
2168 .release = seq_release,
2171 static struct seq_operations softnet_seq_ops = {
2172 .start = softnet_seq_start,
2173 .next = softnet_seq_next,
2174 .stop = softnet_seq_stop,
2175 .show = softnet_seq_show,
2178 static int softnet_seq_open(struct inode *inode, struct file *file)
2180 return seq_open(file, &softnet_seq_ops);
2183 static struct file_operations softnet_seq_fops = {
2184 .owner = THIS_MODULE,
2185 .open = softnet_seq_open,
2187 .llseek = seq_lseek,
2188 .release = seq_release,
2192 extern int wireless_proc_init(void);
2194 #define wireless_proc_init() 0
2197 static int __init dev_proc_init(void)
2201 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2203 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2205 if (wireless_proc_init())
2211 proc_net_remove("softnet_stat");
2213 proc_net_remove("dev");
2217 #define dev_proc_init() 0
2218 #endif /* CONFIG_PROC_FS */
2222 * netdev_set_master - set up master/slave pair
2223 * @slave: slave device
2224 * @master: new master device
2226 * Changes the master device of the slave. Pass %NULL to break the
2227 * bonding. The caller must hold the RTNL semaphore. On a failure
2228 * a negative errno code is returned. On success the reference counts
2229 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2230 * function returns zero.
2232 int netdev_set_master(struct net_device *slave, struct net_device *master)
2234 struct net_device *old = slave->master;
2244 slave->master = master;
2252 slave->flags |= IFF_SLAVE;
2254 slave->flags &= ~IFF_SLAVE;
2256 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2261 * dev_set_promiscuity - update promiscuity count on a device
2265 * Add or remove promsicuity from a device. While the count in the device
2266 * remains above zero the interface remains promiscuous. Once it hits zero
2267 * the device reverts back to normal filtering operation. A negative inc
2268 * value is used to drop promiscuity on the device.
2270 void dev_set_promiscuity(struct net_device *dev, int inc)
2272 unsigned short old_flags = dev->flags;
2274 dev->flags |= IFF_PROMISC;
2275 if ((dev->promiscuity += inc) == 0)
2276 dev->flags &= ~IFF_PROMISC;
2277 if (dev->flags ^ old_flags) {
2278 #ifdef CONFIG_NET_FASTROUTE
2279 if (dev->flags & IFF_PROMISC) {
2280 netdev_fastroute_obstacles++;
2281 dev_clear_fastroute(dev);
2283 netdev_fastroute_obstacles--;
2286 printk(KERN_INFO "device %s %s promiscuous mode\n",
2287 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2293 * dev_set_allmulti - update allmulti count on a device
2297 * Add or remove reception of all multicast frames to a device. While the
2298 * count in the device remains above zero the interface remains listening
2299 * to all interfaces. Once it hits zero the device reverts back to normal
2300 * filtering operation. A negative @inc value is used to drop the counter
2301 * when releasing a resource needing all multicasts.
2304 void dev_set_allmulti(struct net_device *dev, int inc)
2306 unsigned short old_flags = dev->flags;
2308 dev->flags |= IFF_ALLMULTI;
2309 if ((dev->allmulti += inc) == 0)
2310 dev->flags &= ~IFF_ALLMULTI;
2311 if (dev->flags ^ old_flags)
2315 unsigned dev_get_flags(const struct net_device *dev)
2319 flags = (dev->flags & ~(IFF_PROMISC |
2322 (dev->gflags & (IFF_PROMISC |
2325 if (netif_running(dev) && netif_carrier_ok(dev))
2326 flags |= IFF_RUNNING;
2331 int dev_change_flags(struct net_device *dev, unsigned flags)
2334 int old_flags = dev->flags;
2337 * Set the flags on our device.
2340 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2341 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2343 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2347 * Load in the correct multicast list now the flags have changed.
2353 * Have we downed the interface. We handle IFF_UP ourselves
2354 * according to user attempts to set it, rather than blindly
2359 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2360 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2366 if (dev->flags & IFF_UP &&
2367 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2369 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2371 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2372 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2373 dev->gflags ^= IFF_PROMISC;
2374 dev_set_promiscuity(dev, inc);
2377 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2378 is important. Some (broken) drivers set IFF_PROMISC, when
2379 IFF_ALLMULTI is requested not asking us and not reporting.
2381 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2382 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2383 dev->gflags ^= IFF_ALLMULTI;
2384 dev_set_allmulti(dev, inc);
2387 if (old_flags ^ dev->flags)
2388 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2393 int dev_set_mtu(struct net_device *dev, int new_mtu)
2397 if (new_mtu == dev->mtu)
2400 /* MTU must be positive. */
2404 if (!netif_device_present(dev))
2408 if (dev->change_mtu)
2409 err = dev->change_mtu(dev, new_mtu);
2412 if (!err && dev->flags & IFF_UP)
2413 notifier_call_chain(&netdev_chain,
2414 NETDEV_CHANGEMTU, dev);
2420 * Perform the SIOCxIFxxx calls.
2422 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2425 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2431 case SIOCGIFFLAGS: /* Get interface flags */
2432 ifr->ifr_flags = dev_get_flags(dev);
2435 case SIOCSIFFLAGS: /* Set interface flags */
2436 return dev_change_flags(dev, ifr->ifr_flags);
2438 case SIOCGIFMETRIC: /* Get the metric on the interface
2439 (currently unused) */
2440 ifr->ifr_metric = 0;
2443 case SIOCSIFMETRIC: /* Set the metric on the interface
2444 (currently unused) */
2447 case SIOCGIFMTU: /* Get the MTU of a device */
2448 ifr->ifr_mtu = dev->mtu;
2451 case SIOCSIFMTU: /* Set the MTU of a device */
2452 return dev_set_mtu(dev, ifr->ifr_mtu);
2455 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2456 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2457 ifr->ifr_hwaddr.sa_family = dev->type;
2461 if (!dev->set_mac_address)
2463 if (ifr->ifr_hwaddr.sa_family != dev->type)
2465 if (!netif_device_present(dev))
2467 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2469 notifier_call_chain(&netdev_chain,
2470 NETDEV_CHANGEADDR, dev);
2473 case SIOCSIFHWBROADCAST:
2474 if (ifr->ifr_hwaddr.sa_family != dev->type)
2476 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2477 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2478 notifier_call_chain(&netdev_chain,
2479 NETDEV_CHANGEADDR, dev);
2483 ifr->ifr_map.mem_start = dev->mem_start;
2484 ifr->ifr_map.mem_end = dev->mem_end;
2485 ifr->ifr_map.base_addr = dev->base_addr;
2486 ifr->ifr_map.irq = dev->irq;
2487 ifr->ifr_map.dma = dev->dma;
2488 ifr->ifr_map.port = dev->if_port;
2492 if (dev->set_config) {
2493 if (!netif_device_present(dev))
2495 return dev->set_config(dev, &ifr->ifr_map);
2500 if (!dev->set_multicast_list ||
2501 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2503 if (!netif_device_present(dev))
2505 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2509 if (!dev->set_multicast_list ||
2510 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2512 if (!netif_device_present(dev))
2514 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2518 ifr->ifr_ifindex = dev->ifindex;
2522 ifr->ifr_qlen = dev->tx_queue_len;
2526 if (ifr->ifr_qlen < 0)
2528 dev->tx_queue_len = ifr->ifr_qlen;
2532 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2533 return dev_change_name(dev, ifr->ifr_newname);
2536 * Unknown or private ioctl
2540 if ((cmd >= SIOCDEVPRIVATE &&
2541 cmd <= SIOCDEVPRIVATE + 15) ||
2542 cmd == SIOCBONDENSLAVE ||
2543 cmd == SIOCBONDRELEASE ||
2544 cmd == SIOCBONDSETHWADDR ||
2545 cmd == SIOCBONDSLAVEINFOQUERY ||
2546 cmd == SIOCBONDINFOQUERY ||
2547 cmd == SIOCBONDCHANGEACTIVE ||
2548 cmd == SIOCGMIIPHY ||
2549 cmd == SIOCGMIIREG ||
2550 cmd == SIOCSMIIREG ||
2551 cmd == SIOCBRADDIF ||
2552 cmd == SIOCBRDELIF ||
2553 cmd == SIOCWANDEV) {
2555 if (dev->do_ioctl) {
2556 if (netif_device_present(dev))
2557 err = dev->do_ioctl(dev, ifr,
2570 * This function handles all "interface"-type I/O control requests. The actual
2571 * 'doing' part of this is dev_ifsioc above.
2575 * dev_ioctl - network device ioctl
2576 * @cmd: command to issue
2577 * @arg: pointer to a struct ifreq in user space
2579 * Issue ioctl functions to devices. This is normally called by the
2580 * user space syscall interfaces but can sometimes be useful for
2581 * other purposes. The return value is the return from the syscall if
2582 * positive or a negative errno code on error.
2585 int dev_ioctl(unsigned int cmd, void __user *arg)
2591 /* One special case: SIOCGIFCONF takes ifconf argument
2592 and requires shared lock, because it sleeps writing
2596 if (cmd == SIOCGIFCONF) {
2598 ret = dev_ifconf((char __user *) arg);
2602 if (cmd == SIOCGIFNAME)
2603 return dev_ifname((struct ifreq __user *)arg);
2605 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2608 ifr.ifr_name[IFNAMSIZ-1] = 0;
2610 colon = strchr(ifr.ifr_name, ':');
2615 * See which interface the caller is talking about.
2620 * These ioctl calls:
2621 * - can be done by all.
2622 * - atomic and do not require locking.
2633 dev_load(ifr.ifr_name);
2634 read_lock(&dev_base_lock);
2635 ret = dev_ifsioc(&ifr, cmd);
2636 read_unlock(&dev_base_lock);
2640 if (copy_to_user(arg, &ifr,
2641 sizeof(struct ifreq)))
2647 dev_load(ifr.ifr_name);
2649 ret = dev_ethtool(&ifr);
2654 if (copy_to_user(arg, &ifr,
2655 sizeof(struct ifreq)))
2661 * These ioctl calls:
2662 * - require superuser power.
2663 * - require strict serialization.
2669 if (!capable(CAP_NET_ADMIN))
2671 dev_load(ifr.ifr_name);
2673 ret = dev_ifsioc(&ifr, cmd);
2678 if (copy_to_user(arg, &ifr,
2679 sizeof(struct ifreq)))
2685 * These ioctl calls:
2686 * - require superuser power.
2687 * - require strict serialization.
2688 * - do not return a value
2698 case SIOCSIFHWBROADCAST:
2701 case SIOCBONDENSLAVE:
2702 case SIOCBONDRELEASE:
2703 case SIOCBONDSETHWADDR:
2704 case SIOCBONDSLAVEINFOQUERY:
2705 case SIOCBONDINFOQUERY:
2706 case SIOCBONDCHANGEACTIVE:
2709 if (!capable(CAP_NET_ADMIN))
2711 dev_load(ifr.ifr_name);
2713 ret = dev_ifsioc(&ifr, cmd);
2718 /* Get the per device memory space. We can add this but
2719 * currently do not support it */
2721 /* Set the per device memory buffer space.
2722 * Not applicable in our case */
2727 * Unknown or private ioctl.
2730 if (cmd == SIOCWANDEV ||
2731 (cmd >= SIOCDEVPRIVATE &&
2732 cmd <= SIOCDEVPRIVATE + 15)) {
2733 dev_load(ifr.ifr_name);
2735 ret = dev_ifsioc(&ifr, cmd);
2737 if (!ret && copy_to_user(arg, &ifr,
2738 sizeof(struct ifreq)))
2743 /* Take care of Wireless Extensions */
2744 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2745 /* If command is `set a parameter', or
2746 * `get the encoding parameters', check if
2747 * the user has the right to do it */
2748 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2749 if (!capable(CAP_NET_ADMIN))
2752 dev_load(ifr.ifr_name);
2754 /* Follow me in net/core/wireless.c */
2755 ret = wireless_process_ioctl(&ifr, cmd);
2757 if (!ret && IW_IS_GET(cmd) &&
2758 copy_to_user(arg, &ifr,
2759 sizeof(struct ifreq)))
2763 #endif /* WIRELESS_EXT */
2770 * dev_new_index - allocate an ifindex
2772 * Returns a suitable unique value for a new device interface
2773 * number. The caller must hold the rtnl semaphore or the
2774 * dev_base_lock to be sure it remains unique.
2776 int dev_new_index(void)
2782 if (!__dev_get_by_index(ifindex))
2787 static int dev_boot_phase = 1;
2789 /* Delayed registration/unregisteration */
2790 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2791 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2793 static inline void net_set_todo(struct net_device *dev)
2795 spin_lock(&net_todo_list_lock);
2796 list_add_tail(&dev->todo_list, &net_todo_list);
2797 spin_unlock(&net_todo_list_lock);
2801 * register_netdevice - register a network device
2802 * @dev: device to register
2804 * Take a completed network device structure and add it to the kernel
2805 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2806 * chain. 0 is returned on success. A negative errno code is returned
2807 * on a failure to set up the device, or if the name is a duplicate.
2809 * Callers must hold the rtnl semaphore. See the comment at the
2810 * end of Space.c for details about the locking. You may want
2811 * register_netdev() instead of this.
2814 * The locking appears insufficient to guarantee two parallel registers
2815 * will not get the same name.
2818 int register_netdevice(struct net_device *dev)
2820 struct hlist_head *head;
2821 struct hlist_node *p;
2824 BUG_ON(dev_boot_phase);
2827 /* When net_device's are persistent, this will be fatal. */
2828 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2830 spin_lock_init(&dev->queue_lock);
2831 spin_lock_init(&dev->xmit_lock);
2832 dev->xmit_lock_owner = -1;
2833 #ifdef CONFIG_NET_FASTROUTE
2834 dev->fastpath_lock = RW_LOCK_UNLOCKED;
2837 ret = alloc_divert_blk(dev);
2843 /* Init, if this function is available */
2845 ret = dev->init(dev);
2853 if (!dev_valid_name(dev->name)) {
2858 dev->ifindex = dev_new_index();
2859 if (dev->iflink == -1)
2860 dev->iflink = dev->ifindex;
2862 /* Check for existence of name */
2863 head = dev_name_hash(dev->name);
2864 hlist_for_each(p, head) {
2865 struct net_device *d
2866 = hlist_entry(p, struct net_device, name_hlist);
2867 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2873 /* Fix illegal SG+CSUM combinations. */
2874 if ((dev->features & NETIF_F_SG) &&
2875 !(dev->features & (NETIF_F_IP_CSUM |
2877 NETIF_F_HW_CSUM))) {
2878 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2880 dev->features &= ~NETIF_F_SG;
2884 * nil rebuild_header routine,
2885 * that should be never called and used as just bug trap.
2888 if (!dev->rebuild_header)
2889 dev->rebuild_header = default_rebuild_header;
2892 * Default initial state at registry is that the
2893 * device is present.
2896 set_bit(__LINK_STATE_PRESENT, &dev->state);
2899 dev_init_scheduler(dev);
2900 write_lock_bh(&dev_base_lock);
2902 dev_tail = &dev->next;
2903 hlist_add_head(&dev->name_hlist, head);
2904 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2906 dev->reg_state = NETREG_REGISTERING;
2907 write_unlock_bh(&dev_base_lock);
2909 /* Notify protocols, that a new device appeared. */
2910 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2912 /* Finish registration after unlock */
2919 free_divert_blk(dev);
2924 * netdev_wait_allrefs - wait until all references are gone.
2926 * This is called when unregistering network devices.
2928 * Any protocol or device that holds a reference should register
2929 * for netdevice notification, and cleanup and put back the
2930 * reference if they receive an UNREGISTER event.
2931 * We can get stuck here if buggy protocols don't correctly
2934 static void netdev_wait_allrefs(struct net_device *dev)
2936 unsigned long rebroadcast_time, warning_time;
2938 rebroadcast_time = warning_time = jiffies;
2939 while (atomic_read(&dev->refcnt) != 0) {
2940 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2944 /* Rebroadcast unregister notification */
2945 notifier_call_chain(&netdev_chain,
2946 NETDEV_UNREGISTER, dev);
2948 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2950 /* We must not have linkwatch events
2951 * pending on unregister. If this
2952 * happens, we simply run the queue
2953 * unscheduled, resulting in a noop
2956 linkwatch_run_queue();
2962 rebroadcast_time = jiffies;
2965 current->state = TASK_INTERRUPTIBLE;
2966 schedule_timeout(HZ / 4);
2968 if (time_after(jiffies, warning_time + 10 * HZ)) {
2969 printk(KERN_EMERG "unregister_netdevice: "
2970 "waiting for %s to become free. Usage "
2972 dev->name, atomic_read(&dev->refcnt));
2973 warning_time = jiffies;
2982 * register_netdevice(x1);
2983 * register_netdevice(x2);
2985 * unregister_netdevice(y1);
2986 * unregister_netdevice(y2);
2992 * We are invoked by rtnl_unlock() after it drops the semaphore.
2993 * This allows us to deal with problems:
2994 * 1) We can create/delete sysfs objects which invoke hotplug
2995 * without deadlocking with linkwatch via keventd.
2996 * 2) Since we run with the RTNL semaphore not held, we can sleep
2997 * safely in order to wait for the netdev refcnt to drop to zero.
2999 static DECLARE_MUTEX(net_todo_run_mutex);
3000 void netdev_run_todo(void)
3002 struct list_head list = LIST_HEAD_INIT(list);
3006 /* Need to guard against multiple cpu's getting out of order. */
3007 down(&net_todo_run_mutex);
3009 /* Not safe to do outside the semaphore. We must not return
3010 * until all unregister events invoked by the local processor
3011 * have been completed (either by this todo run, or one on
3014 if (list_empty(&net_todo_list))
3017 /* Snapshot list, allow later requests */
3018 spin_lock(&net_todo_list_lock);
3019 list_splice_init(&net_todo_list, &list);
3020 spin_unlock(&net_todo_list_lock);
3022 while (!list_empty(&list)) {
3023 struct net_device *dev
3024 = list_entry(list.next, struct net_device, todo_list);
3025 list_del(&dev->todo_list);
3027 switch(dev->reg_state) {
3028 case NETREG_REGISTERING:
3029 err = netdev_register_sysfs(dev);
3031 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3033 dev->reg_state = NETREG_REGISTERED;
3036 case NETREG_UNREGISTERING:
3037 netdev_unregister_sysfs(dev);
3038 dev->reg_state = NETREG_UNREGISTERED;
3040 netdev_wait_allrefs(dev);
3043 BUG_ON(atomic_read(&dev->refcnt));
3044 BUG_TRAP(!dev->ip_ptr);
3045 BUG_TRAP(!dev->ip6_ptr);
3046 BUG_TRAP(!dev->dn_ptr);
3049 /* It must be the very last action,
3050 * after this 'dev' may point to freed up memory.
3052 if (dev->destructor)
3053 dev->destructor(dev);
3057 printk(KERN_ERR "network todo '%s' but state %d\n",
3058 dev->name, dev->reg_state);
3064 up(&net_todo_run_mutex);
3068 * free_netdev - free network device
3071 * This function does the last stage of destroying an allocated device
3072 * interface. The reference to the device object is released.
3073 * If this is the last reference then it will be freed.
3075 void free_netdev(struct net_device *dev)
3078 /* Compatiablity with error handling in drivers */
3079 if (dev->reg_state == NETREG_UNINITIALIZED) {
3080 kfree((char *)dev - dev->padded);
3084 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3085 dev->reg_state = NETREG_RELEASED;
3087 /* will free via class release */
3088 class_device_put(&dev->class_dev);
3090 kfree((char *)dev - dev->padded);
3094 /* Synchronize with packet receive processing. */
3095 void synchronize_net(void)
3098 synchronize_kernel();
3102 * unregister_netdevice - remove device from the kernel
3105 * This function shuts down a device interface and removes it
3106 * from the kernel tables. On success 0 is returned, on a failure
3107 * a negative errno code is returned.
3109 * Callers must hold the rtnl semaphore. See the comment at the
3110 * end of Space.c for details about the locking. You may want
3111 * unregister_netdev() instead of this.
3114 int unregister_netdevice(struct net_device *dev)
3116 struct net_device *d, **dp;
3118 BUG_ON(dev_boot_phase);
3121 /* Some devices call without registering for initialization unwind. */
3122 if (dev->reg_state == NETREG_UNINITIALIZED) {
3123 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3124 "was registered\n", dev->name, dev);
3128 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3130 /* If device is running, close it first. */
3131 if (dev->flags & IFF_UP)
3134 /* And unlink it from device chain. */
3135 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3137 write_lock_bh(&dev_base_lock);
3138 hlist_del(&dev->name_hlist);
3139 hlist_del(&dev->index_hlist);
3140 if (dev_tail == &dev->next)
3143 write_unlock_bh(&dev_base_lock);
3148 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3153 dev->reg_state = NETREG_UNREGISTERING;
3157 #ifdef CONFIG_NET_FASTROUTE
3158 dev_clear_fastroute(dev);
3161 /* Shutdown queueing discipline. */
3165 /* Notify protocols, that we are about to destroy
3166 this device. They should clean all the things.
3168 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3171 * Flush the multicast chain
3173 dev_mc_discard(dev);
3178 /* Notifier chain MUST detach us from master device. */
3179 BUG_TRAP(!dev->master);
3181 free_divert_blk(dev);
3183 /* Finish processing unregister after unlock */
3190 #ifdef CONFIG_HOTPLUG_CPU
3191 static int dev_cpu_callback(struct notifier_block *nfb,
3192 unsigned long action,
3195 struct sk_buff **list_skb;
3196 struct net_device **list_net;
3197 struct sk_buff *skb;
3198 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3199 struct softnet_data *sd, *oldsd;
3201 if (action != CPU_DEAD)
3204 local_irq_disable();
3205 cpu = smp_processor_id();
3206 sd = &per_cpu(softnet_data, cpu);
3207 oldsd = &per_cpu(softnet_data, oldcpu);
3209 /* Find end of our completion_queue. */
3210 list_skb = &sd->completion_queue;
3212 list_skb = &(*list_skb)->next;
3213 /* Append completion queue from offline CPU. */
3214 *list_skb = oldsd->completion_queue;
3215 oldsd->completion_queue = NULL;
3217 /* Find end of our output_queue. */
3218 list_net = &sd->output_queue;
3220 list_net = &(*list_net)->next_sched;
3221 /* Append output queue from offline CPU. */
3222 *list_net = oldsd->output_queue;
3223 oldsd->output_queue = NULL;
3225 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3228 /* Process offline CPU's input_pkt_queue */
3229 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3234 #endif /* CONFIG_HOTPLUG_CPU */
3238 * Initialize the DEV module. At boot time this walks the device list and
3239 * unhooks any devices that fail to initialise (normally hardware not
3240 * present) and leaves us with a valid list of present and active devices.
3245 * This is called single threaded during boot, so no need
3246 * to take the rtnl semaphore.
3248 static int __init net_dev_init(void)
3250 int i, rc = -ENOMEM;
3252 BUG_ON(!dev_boot_phase);
3254 if (dev_proc_init())
3257 if (netdev_sysfs_init())
3260 INIT_LIST_HEAD(&ptype_all);
3261 for (i = 0; i < 16; i++)
3262 INIT_LIST_HEAD(&ptype_base[i]);
3264 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3265 INIT_HLIST_HEAD(&dev_name_head[i]);
3267 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3268 INIT_HLIST_HEAD(&dev_index_head[i]);
3271 * Initialise the packet receive queues.
3274 for (i = 0; i < NR_CPUS; i++) {
3275 struct softnet_data *queue;
3277 queue = &per_cpu(softnet_data, i);
3278 skb_queue_head_init(&queue->input_pkt_queue);
3279 queue->throttle = 0;
3280 queue->cng_level = 0;
3281 queue->avg_blog = 10; /* arbitrary non-zero */
3282 queue->completion_queue = NULL;
3283 INIT_LIST_HEAD(&queue->poll_list);
3284 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3285 queue->backlog_dev.weight = weight_p;
3286 queue->backlog_dev.poll = process_backlog;
3287 atomic_set(&queue->backlog_dev.refcnt, 1);
3290 #ifdef OFFLINE_SAMPLE
3291 samp_timer.expires = jiffies + (10 * HZ);
3292 add_timer(&samp_timer);
3297 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3298 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3300 hotcpu_notifier(dev_cpu_callback, 0);
3308 subsys_initcall(net_dev_init);
3310 EXPORT_SYMBOL(__dev_get);
3311 EXPORT_SYMBOL(__dev_get_by_flags);
3312 EXPORT_SYMBOL(__dev_get_by_index);
3313 EXPORT_SYMBOL(__dev_get_by_name);
3314 EXPORT_SYMBOL(__dev_remove_pack);
3315 EXPORT_SYMBOL(__skb_linearize);
3316 EXPORT_SYMBOL(call_netdevice_notifiers);
3317 EXPORT_SYMBOL(dev_add_pack);
3318 EXPORT_SYMBOL(dev_alloc_name);
3319 EXPORT_SYMBOL(dev_close);
3320 EXPORT_SYMBOL(dev_get_by_flags);
3321 EXPORT_SYMBOL(dev_get_by_index);
3322 EXPORT_SYMBOL(dev_get_by_name);
3323 EXPORT_SYMBOL(dev_getbyhwaddr);
3324 EXPORT_SYMBOL(dev_ioctl);
3325 EXPORT_SYMBOL(dev_new_index);
3326 EXPORT_SYMBOL(dev_open);
3327 EXPORT_SYMBOL(dev_queue_xmit);
3328 EXPORT_SYMBOL(dev_queue_xmit_nit);
3329 EXPORT_SYMBOL(dev_remove_pack);
3330 EXPORT_SYMBOL(dev_set_allmulti);
3331 EXPORT_SYMBOL(dev_set_promiscuity);
3332 EXPORT_SYMBOL(free_netdev);
3333 EXPORT_SYMBOL(netdev_boot_setup_check);
3334 EXPORT_SYMBOL(netdev_set_master);
3335 EXPORT_SYMBOL(netdev_state_change);
3336 EXPORT_SYMBOL(netif_receive_skb);
3337 EXPORT_SYMBOL(netif_rx);
3338 EXPORT_SYMBOL(register_gifconf);
3339 EXPORT_SYMBOL(register_netdevice);
3340 EXPORT_SYMBOL(register_netdevice_notifier);
3341 EXPORT_SYMBOL(skb_checksum_help);
3342 EXPORT_SYMBOL(synchronize_net);
3343 EXPORT_SYMBOL(unregister_netdevice);
3344 EXPORT_SYMBOL(unregister_netdevice_notifier);
3346 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3347 EXPORT_SYMBOL(br_handle_frame_hook);
3349 /* for 801q VLAN support */
3350 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
3351 EXPORT_SYMBOL(dev_change_flags);
3354 EXPORT_SYMBOL(dev_load);
3356 #ifdef CONFIG_NET_HW_FLOWCONTROL
3357 EXPORT_SYMBOL(netdev_dropping);
3358 EXPORT_SYMBOL(netdev_fc_xoff);
3359 EXPORT_SYMBOL(netdev_register_fc);
3360 EXPORT_SYMBOL(netdev_unregister_fc);
3362 #ifdef CONFIG_NET_FASTROUTE
3363 EXPORT_SYMBOL(netdev_fastroute);
3364 EXPORT_SYMBOL(netdev_fastroute_obstacles);
3367 EXPORT_PER_CPU_SYMBOL(softnet_data);