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 #include <linux/vserver/network.h>
112 #ifdef CONFIG_NET_RADIO
113 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
114 #include <net/iw_handler.h>
115 #endif /* CONFIG_NET_RADIO */
116 #include <asm/current.h>
118 /* This define, if set, will randomly drop a packet when congestion
119 * is more than moderate. It helps fairness in the multi-interface
120 * case when one of them is a hog, but it kills performance for the
121 * single interface case so it is off now by default.
125 /* Setting this will sample the queue lengths and thus congestion
126 * via a timer instead of as each packet is received.
128 #undef OFFLINE_SAMPLE
131 * The list of packet types we will receive (as opposed to discard)
132 * and the routines to invoke.
134 * Why 16. Because with 16 the only overlap we get on a hash of the
135 * low nibble of the protocol value is RARP/SNAP/X.25.
137 * NOTE: That is no longer true with the addition of VLAN tags. Not
138 * sure which should go first, but I bet it won't make much
139 * difference if we are running VLANs. The good news is that
140 * this protocol won't be in the list unless compiled in, so
141 * the average user (w/out VLANs) will not be adversly affected.
158 static spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
159 static struct list_head ptype_base[16]; /* 16 way hashed list */
160 static struct list_head ptype_all; /* Taps */
162 #ifdef OFFLINE_SAMPLE
163 static void sample_queue(unsigned long dummy);
164 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
168 * The @dev_base list is protected by @dev_base_lock and the rtln
171 * Pure readers hold dev_base_lock for reading.
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
186 struct net_device *dev_base;
187 struct net_device **dev_tail = &dev_base;
188 rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
190 EXPORT_SYMBOL(dev_base);
191 EXPORT_SYMBOL(dev_base_lock);
193 #define NETDEV_HASHBITS 8
194 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
195 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
197 static inline struct hlist_head *dev_name_hash(const char *name)
199 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
200 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
203 static inline struct hlist_head *dev_index_hash(int ifindex)
205 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
212 static struct notifier_block *netdev_chain;
215 * Device drivers call our routines to queue packets here. We empty the
216 * queue in the local softnet handler.
218 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
220 #ifdef CONFIG_NET_FASTROUTE
221 int netdev_fastroute;
222 int netdev_fastroute_obstacles;
226 extern int netdev_sysfs_init(void);
227 extern int netdev_register_sysfs(struct net_device *);
228 extern void netdev_unregister_sysfs(struct net_device *);
230 #define netdev_sysfs_init() (0)
231 #define netdev_register_sysfs(dev) (0)
232 #define netdev_unregister_sysfs(dev) do { } while(0)
235 /* netdump function */
236 void (*netdump_func) (struct pt_regs *regs) = NULL;
238 /*******************************************************************************
240 Protocol management and registration routines
242 *******************************************************************************/
251 * Add a protocol ID to the list. Now that the input handler is
252 * smarter we can dispense with all the messy stuff that used to be
255 * BEWARE!!! Protocol handlers, mangling input packets,
256 * MUST BE last in hash buckets and checking protocol handlers
257 * MUST start from promiscuous ptype_all chain in net_bh.
258 * It is true now, do not change it.
259 * Explanation follows: if protocol handler, mangling packet, will
260 * be the first on list, it is not able to sense, that packet
261 * is cloned and should be copied-on-write, so that it will
262 * change it and subsequent readers will get broken packet.
267 * dev_add_pack - add packet handler
268 * @pt: packet type declaration
270 * Add a protocol handler to the networking stack. The passed &packet_type
271 * is linked into kernel lists and may not be freed until it has been
272 * removed from the kernel lists.
274 * This call does not sleep therefore it can not
275 * guarantee all CPU's that are in middle of receiving packets
276 * will see the new packet type (until the next received packet).
279 void dev_add_pack(struct packet_type *pt)
283 spin_lock_bh(&ptype_lock);
284 #ifdef CONFIG_NET_FASTROUTE
285 if (pt->af_packet_priv) {
286 netdev_fastroute_obstacles++;
287 dev_clear_fastroute(pt->dev);
290 if (pt->type == htons(ETH_P_ALL)) {
292 list_add_rcu(&pt->list, &ptype_all);
294 hash = ntohs(pt->type) & 15;
295 list_add_rcu(&pt->list, &ptype_base[hash]);
297 spin_unlock_bh(&ptype_lock);
300 extern void linkwatch_run_queue(void);
305 * __dev_remove_pack - remove packet handler
306 * @pt: packet type declaration
308 * Remove a protocol handler that was previously added to the kernel
309 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
310 * from the kernel lists and can be freed or reused once this function
313 * The packet type might still be in use by receivers
314 * and must not be freed until after all the CPU's have gone
315 * through a quiescent state.
317 void __dev_remove_pack(struct packet_type *pt)
319 struct list_head *head;
320 struct packet_type *pt1;
322 spin_lock_bh(&ptype_lock);
324 if (pt->type == htons(ETH_P_ALL)) {
328 head = &ptype_base[ntohs(pt->type) & 15];
330 list_for_each_entry(pt1, head, list) {
332 #ifdef CONFIG_NET_FASTROUTE
333 if (pt->af_packet_priv)
334 netdev_fastroute_obstacles--;
336 list_del_rcu(&pt->list);
341 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
343 spin_unlock_bh(&ptype_lock);
346 * dev_remove_pack - remove packet handler
347 * @pt: packet type declaration
349 * Remove a protocol handler that was previously added to the kernel
350 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
351 * from the kernel lists and can be freed or reused once this function
354 * This call sleeps to guarantee that no CPU is looking at the packet
357 void dev_remove_pack(struct packet_type *pt)
359 __dev_remove_pack(pt);
364 /******************************************************************************
366 Device Boot-time Settings Routines
368 *******************************************************************************/
370 /* Boot time configuration table */
371 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
374 * netdev_boot_setup_add - add new setup entry
375 * @name: name of the device
376 * @map: configured settings for the device
378 * Adds new setup entry to the dev_boot_setup list. The function
379 * returns 0 on error and 1 on success. This is a generic routine to
382 int netdev_boot_setup_add(char *name, struct ifmap *map)
384 struct netdev_boot_setup *s;
388 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
389 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
390 memset(s[i].name, 0, sizeof(s[i].name));
391 strcpy(s[i].name, name);
392 memcpy(&s[i].map, map, sizeof(s[i].map));
397 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
401 * netdev_boot_setup_check - check boot time settings
402 * @dev: the netdevice
404 * Check boot time settings for the device.
405 * The found settings are set for the device to be used
406 * later in the device probing.
407 * Returns 0 if no settings found, 1 if they are.
409 int netdev_boot_setup_check(struct net_device *dev)
411 struct netdev_boot_setup *s = dev_boot_setup;
414 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
415 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
416 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
417 dev->irq = s[i].map.irq;
418 dev->base_addr = s[i].map.base_addr;
419 dev->mem_start = s[i].map.mem_start;
420 dev->mem_end = s[i].map.mem_end;
429 * netdev_boot_base - get address from boot time settings
430 * @prefix: prefix for network device
431 * @unit: id for network device
433 * Check boot time settings for the base address of device.
434 * The found settings are set for the device to be used
435 * later in the device probing.
436 * Returns 0 if no settings found.
438 unsigned long netdev_boot_base(const char *prefix, int unit)
440 const struct netdev_boot_setup *s = dev_boot_setup;
444 sprintf(name, "%s%d", prefix, unit);
447 * If device already registered then return base of 1
448 * to indicate not to probe for this interface
450 if (__dev_get_by_name(name))
453 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
454 if (!strcmp(name, s[i].name))
455 return s[i].map.base_addr;
460 * Saves at boot time configured settings for any netdevice.
462 int __init netdev_boot_setup(char *str)
467 str = get_options(str, ARRAY_SIZE(ints), ints);
472 memset(&map, 0, sizeof(map));
476 map.base_addr = ints[2];
478 map.mem_start = ints[3];
480 map.mem_end = ints[4];
482 /* Add new entry to the list */
483 return netdev_boot_setup_add(str, &map);
486 __setup("netdev=", netdev_boot_setup);
488 /*******************************************************************************
490 Device Interface Subroutines
492 *******************************************************************************/
495 * __dev_get_by_name - find a device by its name
496 * @name: name to find
498 * Find an interface by name. Must be called under RTNL semaphore
499 * or @dev_base_lock. If the name is found a pointer to the device
500 * is returned. If the name is not found then %NULL is returned. The
501 * reference counters are not incremented so the caller must be
502 * careful with locks.
505 struct net_device *__dev_get_by_name(const char *name)
507 struct hlist_node *p;
509 hlist_for_each(p, dev_name_hash(name)) {
510 struct net_device *dev
511 = hlist_entry(p, struct net_device, name_hlist);
512 if (!strncmp(dev->name, name, IFNAMSIZ))
519 * dev_get_by_name - find a device by its name
520 * @name: name to find
522 * Find an interface by name. This can be called from any
523 * context and does its own locking. The returned handle has
524 * the usage count incremented and the caller must use dev_put() to
525 * release it when it is no longer needed. %NULL is returned if no
526 * matching device is found.
529 struct net_device *dev_get_by_name(const char *name)
531 struct net_device *dev;
533 read_lock(&dev_base_lock);
534 dev = __dev_get_by_name(name);
537 read_unlock(&dev_base_lock);
542 Return value is changed to int to prevent illegal usage in future.
543 It is still legal to use to check for device existence.
545 User should understand, that the result returned by this function
546 is meaningless, if it was not issued under rtnl semaphore.
550 * dev_get - test if a device exists
551 * @name: name to test for
553 * Test if a name exists. Returns true if the name is found. In order
554 * to be sure the name is not allocated or removed during the test the
555 * caller must hold the rtnl semaphore.
557 * This function exists only for back compatibility with older
560 int __dev_get(const char *name)
562 struct net_device *dev;
564 read_lock(&dev_base_lock);
565 dev = __dev_get_by_name(name);
566 read_unlock(&dev_base_lock);
571 * __dev_get_by_index - find a device by its ifindex
572 * @ifindex: index of device
574 * Search for an interface by index. Returns %NULL if the device
575 * is not found or a pointer to the device. The device has not
576 * had its reference counter increased so the caller must be careful
577 * about locking. The caller must hold either the RTNL semaphore
581 struct net_device *__dev_get_by_index(int ifindex)
583 struct hlist_node *p;
585 hlist_for_each(p, dev_index_hash(ifindex)) {
586 struct net_device *dev
587 = hlist_entry(p, struct net_device, index_hlist);
588 if (dev->ifindex == ifindex)
596 * dev_get_by_index - find a device by its ifindex
597 * @ifindex: index of device
599 * Search for an interface by index. Returns NULL if the device
600 * is not found or a pointer to the device. The device returned has
601 * had a reference added and the pointer is safe until the user calls
602 * dev_put to indicate they have finished with it.
605 struct net_device *dev_get_by_index(int ifindex)
607 struct net_device *dev;
609 read_lock(&dev_base_lock);
610 dev = __dev_get_by_index(ifindex);
613 read_unlock(&dev_base_lock);
618 * dev_getbyhwaddr - find a device by its hardware address
619 * @type: media type of device
620 * @ha: hardware address
622 * Search for an interface by MAC address. Returns NULL if the device
623 * is not found or a pointer to the device. The caller must hold the
624 * rtnl semaphore. The returned device has not had its ref count increased
625 * and the caller must therefore be careful about locking
628 * If the API was consistent this would be __dev_get_by_hwaddr
631 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
633 struct net_device *dev;
637 for (dev = dev_base; dev; dev = dev->next)
638 if (dev->type == type &&
639 !memcmp(dev->dev_addr, ha, dev->addr_len))
644 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
646 struct net_device *dev;
648 for (dev = dev_base; dev; dev = dev->next)
649 if (dev->type == type)
654 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
656 struct net_device *dev_getfirstbyhwtype(unsigned short type)
658 struct net_device *dev;
661 dev = __dev_getfirstbyhwtype(type);
668 EXPORT_SYMBOL(dev_getfirstbyhwtype);
671 * dev_get_by_flags - find any device with given flags
672 * @if_flags: IFF_* values
673 * @mask: bitmask of bits in if_flags to check
675 * Search for any interface with the given flags. Returns NULL if a device
676 * is not found or a pointer to the device. The device returned has
677 * had a reference added and the pointer is safe until the user calls
678 * dev_put to indicate they have finished with it.
681 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
683 struct net_device *dev;
685 read_lock(&dev_base_lock);
686 dev = __dev_get_by_flags(if_flags, mask);
689 read_unlock(&dev_base_lock);
694 * __dev_get_by_flags - find any device with given flags
695 * @if_flags: IFF_* values
696 * @mask: bitmask of bits in if_flags to check
698 * Search for any interface with the given flags. Returns NULL if a device
699 * is not found or a pointer to the device. The caller must hold either
700 * the RTNL semaphore or @dev_base_lock.
703 struct net_device *__dev_get_by_flags(unsigned short if_flags, unsigned short mask)
705 struct net_device *dev;
707 for (dev = dev_base; dev != NULL; dev = dev->next) {
708 if (((dev->flags ^ if_flags) & mask) == 0)
715 * dev_valid_name - check if name is okay for network device
718 * Network device names need to be valid file names to
719 * to allow sysfs to work
721 int dev_valid_name(const char *name)
723 return !(*name == '\0'
724 || !strcmp(name, ".")
725 || !strcmp(name, "..")
726 || strchr(name, '/'));
730 * dev_alloc_name - allocate a name for a device
732 * @name: name format string
734 * Passed a format string - eg "lt%d" it will try and find a suitable
735 * id. Not efficient for many devices, not called a lot. The caller
736 * must hold the dev_base or rtnl lock while allocating the name and
737 * adding the device in order to avoid duplicates. Returns the number
738 * of the unit assigned or a negative errno code.
741 int dev_alloc_name(struct net_device *dev, const char *name)
746 const int max_netdevices = 8*PAGE_SIZE;
748 struct net_device *d;
750 p = strnchr(name, IFNAMSIZ-1, '%');
753 * Verify the string as this thing may have come from
754 * the user. There must be either one "%d" and no other "%"
757 if (p[1] != 'd' || strchr(p + 2, '%'))
760 /* Use one page as a bit array of possible slots */
761 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
765 for (d = dev_base; d; d = d->next) {
766 if (!sscanf(d->name, name, &i))
768 if (i < 0 || i >= max_netdevices)
771 /* avoid cases where sscanf is not exact inverse of printf */
772 snprintf(buf, sizeof(buf), name, i);
773 if (!strncmp(buf, d->name, IFNAMSIZ))
777 i = find_first_zero_bit(inuse, max_netdevices);
778 free_page((unsigned long) inuse);
781 snprintf(buf, sizeof(buf), name, i);
782 if (!__dev_get_by_name(buf)) {
783 strlcpy(dev->name, buf, IFNAMSIZ);
787 /* It is possible to run out of possible slots
788 * when the name is long and there isn't enough space left
789 * for the digits, or if all bits are used.
796 * dev_change_name - change name of a device
798 * @newname: name (or format string) must be at least IFNAMSIZ
800 * Change name of a device, can pass format strings "eth%d".
803 int dev_change_name(struct net_device *dev, char *newname)
809 if (dev->flags & IFF_UP)
812 if (!dev_valid_name(newname))
815 if (strchr(newname, '%')) {
816 err = dev_alloc_name(dev, newname);
819 strcpy(newname, dev->name);
821 else if (__dev_get_by_name(newname))
824 strlcpy(dev->name, newname, IFNAMSIZ);
826 err = class_device_rename(&dev->class_dev, dev->name);
828 hlist_del(&dev->name_hlist);
829 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
830 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
837 * netdev_state_change - device changes state
838 * @dev: device to cause notification
840 * Called to indicate a device has changed state. This function calls
841 * the notifier chains for netdev_chain and sends a NEWLINK message
842 * to the routing socket.
844 void netdev_state_change(struct net_device *dev)
846 if (dev->flags & IFF_UP) {
847 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
848 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
853 * dev_load - load a network module
854 * @name: name of interface
856 * If a network interface is not present and the process has suitable
857 * privileges this function loads the module. If module loading is not
858 * available in this kernel then it becomes a nop.
861 void dev_load(const char *name)
863 struct net_device *dev;
865 read_lock(&dev_base_lock);
866 dev = __dev_get_by_name(name);
867 read_unlock(&dev_base_lock);
869 if (!dev && capable(CAP_SYS_MODULE))
870 request_module("%s", name);
873 static int default_rebuild_header(struct sk_buff *skb)
875 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
876 skb->dev ? skb->dev->name : "NULL!!!");
883 * Some old buggy device drivers change get_stats after registering
884 * the device. Try and trap them here.
885 * This can be elimnated when all devices are known fixed.
887 static inline int get_stats_changed(struct net_device *dev)
889 int changed = dev->last_stats != dev->get_stats;
890 dev->last_stats = dev->get_stats;
895 * dev_open - prepare an interface for use.
896 * @dev: device to open
898 * Takes a device from down to up state. The device's private open
899 * function is invoked and then the multicast lists are loaded. Finally
900 * the device is moved into the up state and a %NETDEV_UP message is
901 * sent to the netdev notifier chain.
903 * Calling this function on an active interface is a nop. On a failure
904 * a negative errno code is returned.
906 int dev_open(struct net_device *dev)
914 if (dev->flags & IFF_UP)
918 * Check for broken device drivers.
920 if (get_stats_changed(dev) && net_ratelimit()) {
921 printk(KERN_ERR "%s: driver changed get_stats after register\n",
926 * Is it even present?
928 if (!netif_device_present(dev))
932 * Call device private open method
934 set_bit(__LINK_STATE_START, &dev->state);
936 ret = dev->open(dev);
938 clear_bit(__LINK_STATE_START, &dev->state);
942 * Check for more broken device drivers.
944 if (get_stats_changed(dev) && net_ratelimit()) {
945 printk(KERN_ERR "%s: driver changed get_stats in open\n",
950 * If it went open OK then:
957 dev->flags |= IFF_UP;
960 * Initialize multicasting status
965 * Wakeup transmit queue engine
970 * ... and announce new interface.
972 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
977 #ifdef CONFIG_NET_FASTROUTE
979 static void dev_do_clear_fastroute(struct net_device *dev)
981 if (dev->accept_fastpath) {
984 for (i = 0; i <= NETDEV_FASTROUTE_HMASK; i++) {
985 struct dst_entry *dst;
987 write_lock_irq(&dev->fastpath_lock);
988 dst = dev->fastpath[i];
989 dev->fastpath[i] = NULL;
990 write_unlock_irq(&dev->fastpath_lock);
997 void dev_clear_fastroute(struct net_device *dev)
1000 dev_do_clear_fastroute(dev);
1002 read_lock(&dev_base_lock);
1003 for (dev = dev_base; dev; dev = dev->next)
1004 dev_do_clear_fastroute(dev);
1005 read_unlock(&dev_base_lock);
1011 * dev_close - shutdown an interface.
1012 * @dev: device to shutdown
1014 * This function moves an active device into down state. A
1015 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1016 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1019 int dev_close(struct net_device *dev)
1021 if (!(dev->flags & IFF_UP))
1025 * Tell people we are going down, so that they can
1026 * prepare to death, when device is still operating.
1028 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
1030 dev_deactivate(dev);
1032 clear_bit(__LINK_STATE_START, &dev->state);
1034 /* Synchronize to scheduled poll. We cannot touch poll list,
1035 * it can be even on different cpu. So just clear netif_running(),
1036 * and wait when poll really will happen. Actually, the best place
1037 * for this is inside dev->stop() after device stopped its irq
1038 * engine, but this requires more changes in devices. */
1040 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1041 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
1043 current->state = TASK_INTERRUPTIBLE;
1044 schedule_timeout(1);
1048 * Call the device specific close. This cannot fail.
1049 * Only if device is UP
1051 * We allow it to be called even after a DETACH hot-plug
1058 * Device is now down.
1061 dev->flags &= ~IFF_UP;
1062 #ifdef CONFIG_NET_FASTROUTE
1063 dev_clear_fastroute(dev);
1067 * Tell people we are down
1069 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1076 * Device change register/unregister. These are not inline or static
1077 * as we export them to the world.
1081 * register_netdevice_notifier - register a network notifier block
1084 * Register a notifier to be called when network device events occur.
1085 * The notifier passed is linked into the kernel structures and must
1086 * not be reused until it has been unregistered. A negative errno code
1087 * is returned on a failure.
1089 * When registered all registration and up events are replayed
1090 * to the new notifier to allow device to have a race free
1091 * view of the network device list.
1094 int register_netdevice_notifier(struct notifier_block *nb)
1096 struct net_device *dev;
1100 err = notifier_chain_register(&netdev_chain, nb);
1102 for (dev = dev_base; dev; dev = dev->next) {
1103 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1105 if (dev->flags & IFF_UP)
1106 nb->notifier_call(nb, NETDEV_UP, dev);
1114 * unregister_netdevice_notifier - unregister a network notifier block
1117 * Unregister a notifier previously registered by
1118 * register_netdevice_notifier(). The notifier is unlinked into the
1119 * kernel structures and may then be reused. A negative errno code
1120 * is returned on a failure.
1123 int unregister_netdevice_notifier(struct notifier_block *nb)
1125 return notifier_chain_unregister(&netdev_chain, nb);
1129 * call_netdevice_notifiers - call all network notifier blocks
1130 * @val: value passed unmodified to notifier function
1131 * @v: pointer passed unmodified to notifier function
1133 * Call all network notifier blocks. Parameters and return value
1134 * are as for notifier_call_chain().
1137 int call_netdevice_notifiers(unsigned long val, void *v)
1139 return notifier_call_chain(&netdev_chain, val, v);
1143 * Support routine. Sends outgoing frames to any network
1144 * taps currently in use.
1147 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1149 struct packet_type *ptype;
1150 net_timestamp(&skb->stamp);
1153 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1154 /* Never send packets back to the socket
1155 * they originated from - MvS (miquels@drinkel.ow.org)
1157 if ((ptype->dev == dev || !ptype->dev) &&
1158 (ptype->af_packet_priv == NULL ||
1159 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1160 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1164 /* skb->nh should be correctly
1165 set by sender, so that the second statement is
1166 just protection against buggy protocols.
1168 skb2->mac.raw = skb2->data;
1170 if (skb2->nh.raw < skb2->data ||
1171 skb2->nh.raw > skb2->tail) {
1172 if (net_ratelimit())
1173 printk(KERN_CRIT "protocol %04x is "
1175 skb2->protocol, dev->name);
1176 skb2->nh.raw = skb2->data;
1179 skb2->h.raw = skb2->nh.raw;
1180 skb2->pkt_type = PACKET_OUTGOING;
1181 ptype->func(skb2, skb->dev, ptype);
1188 * Invalidate hardware checksum when packet is to be mangled, and
1189 * complete checksum manually on outgoing path.
1191 int skb_checksum_help(struct sk_buff **pskb, int inward)
1194 int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1197 (*pskb)->ip_summed = CHECKSUM_NONE;
1201 if (skb_shared(*pskb) || skb_cloned(*pskb)) {
1202 struct sk_buff *newskb = skb_copy(*pskb, GFP_ATOMIC);
1208 skb_set_owner_w(newskb, (*pskb)->sk);
1213 if (offset > (int)(*pskb)->len)
1215 csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1217 offset = (*pskb)->tail - (*pskb)->h.raw;
1220 if ((*pskb)->csum + 2 > offset)
1223 *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1224 (*pskb)->ip_summed = CHECKSUM_NONE;
1229 #ifdef CONFIG_HIGHMEM
1230 /* Actually, we should eliminate this check as soon as we know, that:
1231 * 1. IOMMU is present and allows to map all the memory.
1232 * 2. No high memory really exists on this machine.
1235 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1239 if (dev->features & NETIF_F_HIGHDMA)
1242 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1243 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1249 #define illegal_highdma(dev, skb) (0)
1252 extern void skb_release_data(struct sk_buff *);
1254 /* Keep head the same: replace data */
1255 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1260 struct skb_shared_info *ninfo;
1261 int headerlen = skb->data - skb->head;
1262 int expand = (skb->tail + skb->data_len) - skb->end;
1264 if (skb_shared(skb))
1270 size = skb->end - skb->head + expand;
1271 size = SKB_DATA_ALIGN(size);
1272 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1276 /* Copy entire thing */
1277 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1281 ninfo = (struct skb_shared_info*)(data + size);
1282 atomic_set(&ninfo->dataref, 1);
1283 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1284 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1285 ninfo->nr_frags = 0;
1286 ninfo->frag_list = NULL;
1288 /* Offset between the two in bytes */
1289 offset = data - skb->head;
1291 /* Free old data. */
1292 skb_release_data(skb);
1295 skb->end = data + size;
1297 /* Set up new pointers */
1298 skb->h.raw += offset;
1299 skb->nh.raw += offset;
1300 skb->mac.raw += offset;
1301 skb->tail += offset;
1302 skb->data += offset;
1304 /* We are no longer a clone, even if we were. */
1307 skb->tail += skb->data_len;
1312 #define HARD_TX_LOCK_BH(dev, cpu) { \
1313 if ((dev->features & NETIF_F_LLTX) == 0) { \
1314 spin_lock_bh(&dev->xmit_lock); \
1315 dev->xmit_lock_owner = cpu; \
1319 #define HARD_TX_UNLOCK_BH(dev) { \
1320 if ((dev->features & NETIF_F_LLTX) == 0) { \
1321 dev->xmit_lock_owner = -1; \
1322 spin_unlock_bh(&dev->xmit_lock); \
1327 * dev_queue_xmit - transmit a buffer
1328 * @skb: buffer to transmit
1330 * Queue a buffer for transmission to a network device. The caller must
1331 * have set the device and priority and built the buffer before calling
1332 * this function. The function can be called from an interrupt.
1334 * A negative errno code is returned on a failure. A success does not
1335 * guarantee the frame will be transmitted as it may be dropped due
1336 * to congestion or traffic shaping.
1339 int dev_queue_xmit(struct sk_buff *skb)
1341 struct net_device *dev = skb->dev;
1345 if (skb_shinfo(skb)->frag_list &&
1346 !(dev->features & NETIF_F_FRAGLIST) &&
1347 __skb_linearize(skb, GFP_ATOMIC))
1350 /* Fragmented skb is linearized if device does not support SG,
1351 * or if at least one of fragments is in highmem and device
1352 * does not support DMA from it.
1354 if (skb_shinfo(skb)->nr_frags &&
1355 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1356 __skb_linearize(skb, GFP_ATOMIC))
1359 /* If packet is not checksummed and device does not support
1360 * checksumming for this protocol, complete checksumming here.
1362 if (skb->ip_summed == CHECKSUM_HW &&
1363 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1364 (!(dev->features & NETIF_F_IP_CSUM) ||
1365 skb->protocol != htons(ETH_P_IP))))
1366 if (skb_checksum_help(&skb, 0))
1370 /* Updates of qdisc are serialized by queue_lock.
1371 * The struct Qdisc which is pointed to by qdisc is now a
1372 * rcu structure - it may be accessed without acquiring
1373 * a lock (but the structure may be stale.) The freeing of the
1374 * qdisc will be deferred until it's known that there are no
1375 * more references to it.
1377 * If the qdisc has an enqueue function, we still need to
1378 * hold the queue_lock before calling it, since queue_lock
1379 * also serializes access to the device queue.
1383 smp_read_barrier_depends();
1384 #ifdef CONFIG_NET_CLS_ACT
1385 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1388 /* Grab device queue */
1389 spin_lock_bh(&dev->queue_lock);
1391 rc = q->enqueue(skb, q);
1395 spin_unlock_bh(&dev->queue_lock);
1397 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1402 /* The device has no queue. Common case for software devices:
1403 loopback, all the sorts of tunnels...
1405 Really, it is unlikely that xmit_lock protection is necessary here.
1406 (f.e. loopback and IP tunnels are clean ignoring statistics
1408 However, it is possible, that they rely on protection
1411 Check this and shot the lock. It is not prone from deadlocks.
1412 Either shot noqueue qdisc, it is even simpler 8)
1414 if (dev->flags & IFF_UP) {
1415 int cpu = get_cpu();
1417 if (dev->xmit_lock_owner != cpu) {
1419 HARD_TX_LOCK_BH(dev, cpu);
1422 if (!netif_queue_stopped(dev)) {
1424 dev_queue_xmit_nit(skb, dev);
1427 if (!dev->hard_start_xmit(skb, dev)) {
1428 HARD_TX_UNLOCK_BH(dev);
1432 HARD_TX_UNLOCK_BH(dev);
1433 if (net_ratelimit())
1434 printk(KERN_CRIT "Virtual device %s asks to "
1435 "queue packet!\n", dev->name);
1439 /* Recursion is detected! It is possible,
1441 if (net_ratelimit())
1442 printk(KERN_CRIT "Dead loop on virtual device "
1443 "%s, fix it urgently!\n", dev->name);
1455 /*=======================================================================
1457 =======================================================================*/
1459 int netdev_max_backlog = 300;
1460 int weight_p = 64; /* old backlog weight */
1461 /* These numbers are selected based on intuition and some
1462 * experimentatiom, if you have more scientific way of doing this
1463 * please go ahead and fix things.
1465 int no_cong_thresh = 10;
1470 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1473 #ifdef CONFIG_NET_HW_FLOWCONTROL
1474 atomic_t netdev_dropping = ATOMIC_INIT(0);
1475 static unsigned long netdev_fc_mask = 1;
1476 unsigned long netdev_fc_xoff;
1477 spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1481 void (*stimul)(struct net_device *);
1482 struct net_device *dev;
1483 } netdev_fc_slots[BITS_PER_LONG];
1485 int netdev_register_fc(struct net_device *dev,
1486 void (*stimul)(struct net_device *dev))
1489 unsigned long flags;
1491 spin_lock_irqsave(&netdev_fc_lock, flags);
1492 if (netdev_fc_mask != ~0UL) {
1493 bit = ffz(netdev_fc_mask);
1494 netdev_fc_slots[bit].stimul = stimul;
1495 netdev_fc_slots[bit].dev = dev;
1496 set_bit(bit, &netdev_fc_mask);
1497 clear_bit(bit, &netdev_fc_xoff);
1499 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1503 void netdev_unregister_fc(int bit)
1505 unsigned long flags;
1507 spin_lock_irqsave(&netdev_fc_lock, flags);
1509 netdev_fc_slots[bit].stimul = NULL;
1510 netdev_fc_slots[bit].dev = NULL;
1511 clear_bit(bit, &netdev_fc_mask);
1512 clear_bit(bit, &netdev_fc_xoff);
1514 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1517 static void netdev_wakeup(void)
1521 spin_lock(&netdev_fc_lock);
1522 xoff = netdev_fc_xoff;
1527 netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1529 spin_unlock(&netdev_fc_lock);
1533 static void get_sample_stats(int cpu)
1539 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1540 int blog = sd->input_pkt_queue.qlen;
1541 int avg_blog = sd->avg_blog;
1543 avg_blog = (avg_blog >> 1) + (blog >> 1);
1545 if (avg_blog > mod_cong) {
1546 /* Above moderate congestion levels. */
1547 sd->cng_level = NET_RX_CN_HIGH;
1550 rq = rd % netdev_max_backlog;
1551 if (rq < avg_blog) /* unlucky bastard */
1552 sd->cng_level = NET_RX_DROP;
1554 } else if (avg_blog > lo_cong) {
1555 sd->cng_level = NET_RX_CN_MOD;
1558 rq = rd % netdev_max_backlog;
1559 if (rq < avg_blog) /* unlucky bastard */
1560 sd->cng_level = NET_RX_CN_HIGH;
1562 } else if (avg_blog > no_cong)
1563 sd->cng_level = NET_RX_CN_LOW;
1564 else /* no congestion */
1565 sd->cng_level = NET_RX_SUCCESS;
1567 sd->avg_blog = avg_blog;
1570 #ifdef OFFLINE_SAMPLE
1571 static void sample_queue(unsigned long dummy)
1573 /* 10 ms 0r 1ms -- i don't care -- JHS */
1575 int cpu = smp_processor_id();
1577 get_sample_stats(cpu);
1578 next_tick += jiffies;
1579 mod_timer(&samp_timer, next_tick);
1585 * netif_rx - post buffer to the network code
1586 * @skb: buffer to post
1588 * This function receives a packet from a device driver and queues it for
1589 * the upper (protocol) levels to process. It always succeeds. The buffer
1590 * may be dropped during processing for congestion control or by the
1594 * NET_RX_SUCCESS (no congestion)
1595 * NET_RX_CN_LOW (low congestion)
1596 * NET_RX_CN_MOD (moderate congestion)
1597 * NET_RX_CN_HIGH (high congestion)
1598 * NET_RX_DROP (packet was dropped)
1602 int netif_rx(struct sk_buff *skb)
1605 struct softnet_data *queue;
1606 unsigned long flags;
1608 #ifdef CONFIG_NETPOLL
1609 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1615 if (!skb->stamp.tv_sec)
1616 net_timestamp(&skb->stamp);
1619 * The code is rearranged so that the path is the most
1620 * short when CPU is congested, but is still operating.
1622 local_irq_save(flags);
1623 this_cpu = smp_processor_id();
1624 queue = &__get_cpu_var(softnet_data);
1626 __get_cpu_var(netdev_rx_stat).total++;
1627 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1628 if (queue->input_pkt_queue.qlen) {
1629 if (queue->throttle)
1634 __skb_queue_tail(&queue->input_pkt_queue, skb);
1635 #ifndef OFFLINE_SAMPLE
1636 get_sample_stats(this_cpu);
1638 local_irq_restore(flags);
1639 return queue->cng_level;
1642 if (queue->throttle) {
1643 queue->throttle = 0;
1644 #ifdef CONFIG_NET_HW_FLOWCONTROL
1645 if (atomic_dec_and_test(&netdev_dropping))
1650 netif_rx_schedule(&queue->backlog_dev);
1654 if (!queue->throttle) {
1655 queue->throttle = 1;
1656 __get_cpu_var(netdev_rx_stat).throttled++;
1657 #ifdef CONFIG_NET_HW_FLOWCONTROL
1658 atomic_inc(&netdev_dropping);
1663 __get_cpu_var(netdev_rx_stat).dropped++;
1664 local_irq_restore(flags);
1670 static __inline__ void skb_bond(struct sk_buff *skb)
1672 struct net_device *dev = skb->dev;
1675 skb->real_dev = skb->dev;
1676 skb->dev = dev->master;
1680 static void net_tx_action(struct softirq_action *h)
1682 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1684 if (sd->completion_queue) {
1685 struct sk_buff *clist;
1687 local_irq_disable();
1688 clist = sd->completion_queue;
1689 sd->completion_queue = NULL;
1693 struct sk_buff *skb = clist;
1694 clist = clist->next;
1696 BUG_TRAP(!atomic_read(&skb->users));
1701 if (sd->output_queue) {
1702 struct net_device *head;
1704 local_irq_disable();
1705 head = sd->output_queue;
1706 sd->output_queue = NULL;
1710 struct net_device *dev = head;
1711 head = head->next_sched;
1713 smp_mb__before_clear_bit();
1714 clear_bit(__LINK_STATE_SCHED, &dev->state);
1716 if (spin_trylock(&dev->queue_lock)) {
1718 spin_unlock(&dev->queue_lock);
1720 netif_schedule(dev);
1726 static __inline__ int deliver_skb(struct sk_buff *skb,
1727 struct packet_type *pt_prev, int last)
1729 atomic_inc(&skb->users);
1730 return pt_prev->func(skb, skb->dev, pt_prev);
1734 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1735 int (*br_handle_frame_hook)(struct sk_buff *skb);
1737 static __inline__ int handle_bridge(struct sk_buff *skb,
1738 struct packet_type *pt_prev)
1740 int ret = NET_RX_DROP;
1742 ret = deliver_skb(skb, pt_prev, 0);
1749 static inline int __handle_bridge(struct sk_buff *skb,
1750 struct packet_type **pt_prev, int *ret)
1752 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
1753 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1754 *ret = handle_bridge(skb, *pt_prev);
1755 if (br_handle_frame_hook(skb) == 0)
1765 #ifdef CONFIG_NET_CLS_ACT
1766 /* TODO: Maybe we should just force sch_ingress to be compiled in
1767 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1768 * a compare and 2 stores extra right now if we dont have it on
1769 * but have CONFIG_NET_CLS_ACT
1770 * NOTE: This doesnt stop any functionality; if you dont have
1771 * the ingress scheduler, you just cant add policies on ingress.
1774 int ing_filter(struct sk_buff *skb)
1777 struct net_device *dev = skb->dev;
1778 int result = TC_ACT_OK;
1780 if (dev->qdisc_ingress) {
1781 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1782 if (MAX_RED_LOOP < ttl++) {
1783 printk("Redir loop detected Dropping packet (%s->%s)\n",
1784 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1788 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1790 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1791 if (NULL == skb->input_dev) {
1792 skb->input_dev = skb->dev;
1793 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1795 spin_lock(&dev->ingress_lock);
1796 if ((q = dev->qdisc_ingress) != NULL)
1797 result = q->enqueue(skb, q);
1798 spin_unlock(&dev->ingress_lock);
1806 int netif_receive_skb(struct sk_buff *skb)
1808 struct packet_type *ptype, *pt_prev;
1809 int ret = NET_RX_DROP;
1810 unsigned short type;
1812 #ifdef CONFIG_NETPOLL
1813 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1819 if (!skb->stamp.tv_sec)
1820 net_timestamp(&skb->stamp);
1824 __get_cpu_var(netdev_rx_stat).total++;
1826 #ifdef CONFIG_NET_FASTROUTE
1827 if (skb->pkt_type == PACKET_FASTROUTE) {
1828 __get_cpu_var(netdev_rx_stat).fastroute_deferred_out++;
1829 return dev_queue_xmit(skb);
1833 skb->h.raw = skb->nh.raw = skb->data;
1834 skb->mac_len = skb->nh.raw - skb->mac.raw;
1837 #ifdef CONFIG_NET_CLS_ACT
1838 if (skb->tc_verd & TC_NCLS) {
1839 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1846 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1847 if (!ptype->dev || ptype->dev == skb->dev) {
1849 ret = deliver_skb(skb, pt_prev, 0);
1854 #ifdef CONFIG_NET_CLS_ACT
1856 atomic_inc(&skb->users);
1857 ret = pt_prev->func(skb, skb->dev, pt_prev);
1858 pt_prev = NULL; /* noone else should process this after*/
1860 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1863 ret = ing_filter(skb);
1865 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1874 handle_diverter(skb);
1876 if (__handle_bridge(skb, &pt_prev, &ret))
1879 type = skb->protocol;
1880 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1881 if (ptype->type == type &&
1882 (!ptype->dev || ptype->dev == skb->dev)) {
1884 ret = deliver_skb(skb, pt_prev, 0);
1890 ret = pt_prev->func(skb, skb->dev, pt_prev);
1893 /* Jamal, now you will not able to escape explaining
1894 * me how you were going to use this. :-)
1904 static int process_backlog(struct net_device *backlog_dev, int *budget)
1907 int quota = min(backlog_dev->quota, *budget);
1908 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1909 unsigned long start_time = jiffies;
1912 struct sk_buff *skb;
1913 struct net_device *dev;
1915 local_irq_disable();
1916 skb = __skb_dequeue(&queue->input_pkt_queue);
1923 netif_receive_skb(skb);
1929 if (work >= quota || jiffies - start_time > 1)
1932 #ifdef CONFIG_NET_HW_FLOWCONTROL
1933 if (queue->throttle &&
1934 queue->input_pkt_queue.qlen < no_cong_thresh ) {
1935 queue->throttle = 0;
1936 if (atomic_dec_and_test(&netdev_dropping)) {
1944 backlog_dev->quota -= work;
1949 backlog_dev->quota -= work;
1952 list_del(&backlog_dev->poll_list);
1953 smp_mb__before_clear_bit();
1954 netif_poll_enable(backlog_dev);
1956 if (queue->throttle) {
1957 queue->throttle = 0;
1958 #ifdef CONFIG_NET_HW_FLOWCONTROL
1959 if (atomic_dec_and_test(&netdev_dropping))
1967 static void net_rx_action(struct softirq_action *h)
1969 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1970 unsigned long start_time = jiffies;
1971 int budget = netdev_max_backlog;
1974 local_irq_disable();
1976 while (!list_empty(&queue->poll_list)) {
1977 struct net_device *dev;
1979 if (budget <= 0 || jiffies - start_time > 1)
1984 dev = list_entry(queue->poll_list.next,
1985 struct net_device, poll_list);
1987 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1988 local_irq_disable();
1989 list_del(&dev->poll_list);
1990 list_add_tail(&dev->poll_list, &queue->poll_list);
1992 dev->quota += dev->weight;
1994 dev->quota = dev->weight;
1997 local_irq_disable();
2005 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2006 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2010 static gifconf_func_t * gifconf_list [NPROTO];
2013 * register_gifconf - register a SIOCGIF handler
2014 * @family: Address family
2015 * @gifconf: Function handler
2017 * Register protocol dependent address dumping routines. The handler
2018 * that is passed must not be freed or reused until it has been replaced
2019 * by another handler.
2021 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2023 if (family >= NPROTO)
2025 gifconf_list[family] = gifconf;
2031 * Map an interface index to its name (SIOCGIFNAME)
2035 * We need this ioctl for efficient implementation of the
2036 * if_indextoname() function required by the IPv6 API. Without
2037 * it, we would have to search all the interfaces to find a
2041 static int dev_ifname(struct ifreq __user *arg)
2043 struct net_device *dev;
2047 * Fetch the caller's info block.
2050 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2053 read_lock(&dev_base_lock);
2054 dev = __dev_get_by_index(ifr.ifr_ifindex);
2056 read_unlock(&dev_base_lock);
2060 strcpy(ifr.ifr_name, dev->name);
2061 read_unlock(&dev_base_lock);
2063 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2069 * Perform a SIOCGIFCONF call. This structure will change
2070 * size eventually, and there is nothing I can do about it.
2071 * Thus we will need a 'compatibility mode'.
2074 static int dev_ifconf(char __user *arg)
2077 struct net_device *dev;
2084 * Fetch the caller's info block.
2087 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2094 * Loop over the interfaces, and write an info block for each.
2098 for (dev = dev_base; dev; dev = dev->next) {
2099 if (!dev_in_nx_info(dev, current->nx_info))
2101 for (i = 0; i < NPROTO; i++) {
2102 if (gifconf_list[i]) {
2105 done = gifconf_list[i](dev, NULL, 0);
2107 done = gifconf_list[i](dev, pos + total,
2117 * All done. Write the updated control block back to the caller.
2119 ifc.ifc_len = total;
2122 * Both BSD and Solaris return 0 here, so we do too.
2124 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2127 #ifdef CONFIG_PROC_FS
2129 * This is invoked by the /proc filesystem handler to display a device
2132 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2134 struct net_device *dev;
2137 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2139 return i == pos ? dev : NULL;
2142 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2144 read_lock(&dev_base_lock);
2145 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2148 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2151 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2154 void dev_seq_stop(struct seq_file *seq, void *v)
2156 read_unlock(&dev_base_lock);
2159 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2161 struct nx_info *nxi = current->nx_info;
2163 if (!dev_in_nx_info(dev, nxi))
2165 if (dev->get_stats) {
2166 struct net_device_stats *stats = dev->get_stats(dev);
2168 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2169 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2170 dev->name, stats->rx_bytes, stats->rx_packets,
2172 stats->rx_dropped + stats->rx_missed_errors,
2173 stats->rx_fifo_errors,
2174 stats->rx_length_errors + stats->rx_over_errors +
2175 stats->rx_crc_errors + stats->rx_frame_errors,
2176 stats->rx_compressed, stats->multicast,
2177 stats->tx_bytes, stats->tx_packets,
2178 stats->tx_errors, stats->tx_dropped,
2179 stats->tx_fifo_errors, stats->collisions,
2180 stats->tx_carrier_errors +
2181 stats->tx_aborted_errors +
2182 stats->tx_window_errors +
2183 stats->tx_heartbeat_errors,
2184 stats->tx_compressed);
2186 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2190 * Called from the PROCfs module. This now uses the new arbitrary sized
2191 * /proc/net interface to create /proc/net/dev
2193 static int dev_seq_show(struct seq_file *seq, void *v)
2195 if (v == SEQ_START_TOKEN)
2196 seq_puts(seq, "Inter-| Receive "
2198 " face |bytes packets errs drop fifo frame "
2199 "compressed multicast|bytes packets errs "
2200 "drop fifo colls carrier compressed\n");
2202 dev_seq_printf_stats(seq, v);
2206 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2208 struct netif_rx_stats *rc = NULL;
2210 while (*pos < NR_CPUS)
2211 if (cpu_online(*pos)) {
2212 rc = &per_cpu(netdev_rx_stat, *pos);
2219 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2221 return softnet_get_online(pos);
2224 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2227 return softnet_get_online(pos);
2230 static void softnet_seq_stop(struct seq_file *seq, void *v)
2234 static int softnet_seq_show(struct seq_file *seq, void *v)
2236 struct netif_rx_stats *s = v;
2238 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2239 s->total, s->dropped, s->time_squeeze, s->throttled,
2240 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2241 s->fastroute_deferred_out,
2243 s->fastroute_latency_reduction
2251 static struct seq_operations dev_seq_ops = {
2252 .start = dev_seq_start,
2253 .next = dev_seq_next,
2254 .stop = dev_seq_stop,
2255 .show = dev_seq_show,
2258 static int dev_seq_open(struct inode *inode, struct file *file)
2260 return seq_open(file, &dev_seq_ops);
2263 static struct file_operations dev_seq_fops = {
2264 .owner = THIS_MODULE,
2265 .open = dev_seq_open,
2267 .llseek = seq_lseek,
2268 .release = seq_release,
2271 static struct seq_operations softnet_seq_ops = {
2272 .start = softnet_seq_start,
2273 .next = softnet_seq_next,
2274 .stop = softnet_seq_stop,
2275 .show = softnet_seq_show,
2278 static int softnet_seq_open(struct inode *inode, struct file *file)
2280 return seq_open(file, &softnet_seq_ops);
2283 static struct file_operations softnet_seq_fops = {
2284 .owner = THIS_MODULE,
2285 .open = softnet_seq_open,
2287 .llseek = seq_lseek,
2288 .release = seq_release,
2292 extern int wireless_proc_init(void);
2294 #define wireless_proc_init() 0
2297 static int __init dev_proc_init(void)
2301 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2303 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2305 if (wireless_proc_init())
2311 proc_net_remove("softnet_stat");
2313 proc_net_remove("dev");
2317 #define dev_proc_init() 0
2318 #endif /* CONFIG_PROC_FS */
2322 * netdev_set_master - set up master/slave pair
2323 * @slave: slave device
2324 * @master: new master device
2326 * Changes the master device of the slave. Pass %NULL to break the
2327 * bonding. The caller must hold the RTNL semaphore. On a failure
2328 * a negative errno code is returned. On success the reference counts
2329 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2330 * function returns zero.
2332 int netdev_set_master(struct net_device *slave, struct net_device *master)
2334 struct net_device *old = slave->master;
2344 slave->master = master;
2352 slave->flags |= IFF_SLAVE;
2354 slave->flags &= ~IFF_SLAVE;
2356 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2361 * dev_set_promiscuity - update promiscuity count on a device
2365 * Add or remove promsicuity from a device. While the count in the device
2366 * remains above zero the interface remains promiscuous. Once it hits zero
2367 * the device reverts back to normal filtering operation. A negative inc
2368 * value is used to drop promiscuity on the device.
2370 void dev_set_promiscuity(struct net_device *dev, int inc)
2372 unsigned short old_flags = dev->flags;
2374 dev->flags |= IFF_PROMISC;
2375 if ((dev->promiscuity += inc) == 0)
2376 dev->flags &= ~IFF_PROMISC;
2377 if (dev->flags ^ old_flags) {
2378 #ifdef CONFIG_NET_FASTROUTE
2379 if (dev->flags & IFF_PROMISC) {
2380 netdev_fastroute_obstacles++;
2381 dev_clear_fastroute(dev);
2383 netdev_fastroute_obstacles--;
2386 printk(KERN_INFO "device %s %s promiscuous mode\n",
2387 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2393 * dev_set_allmulti - update allmulti count on a device
2397 * Add or remove reception of all multicast frames to a device. While the
2398 * count in the device remains above zero the interface remains listening
2399 * to all interfaces. Once it hits zero the device reverts back to normal
2400 * filtering operation. A negative @inc value is used to drop the counter
2401 * when releasing a resource needing all multicasts.
2404 void dev_set_allmulti(struct net_device *dev, int inc)
2406 unsigned short old_flags = dev->flags;
2408 dev->flags |= IFF_ALLMULTI;
2409 if ((dev->allmulti += inc) == 0)
2410 dev->flags &= ~IFF_ALLMULTI;
2411 if (dev->flags ^ old_flags)
2415 unsigned dev_get_flags(const struct net_device *dev)
2419 flags = (dev->flags & ~(IFF_PROMISC |
2422 (dev->gflags & (IFF_PROMISC |
2425 if (netif_running(dev) && netif_carrier_ok(dev))
2426 flags |= IFF_RUNNING;
2431 int dev_change_flags(struct net_device *dev, unsigned flags)
2434 int old_flags = dev->flags;
2437 * Set the flags on our device.
2440 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2441 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2443 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2447 * Load in the correct multicast list now the flags have changed.
2453 * Have we downed the interface. We handle IFF_UP ourselves
2454 * according to user attempts to set it, rather than blindly
2459 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2460 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2466 if (dev->flags & IFF_UP &&
2467 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2469 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2471 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2472 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2473 dev->gflags ^= IFF_PROMISC;
2474 dev_set_promiscuity(dev, inc);
2477 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2478 is important. Some (broken) drivers set IFF_PROMISC, when
2479 IFF_ALLMULTI is requested not asking us and not reporting.
2481 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2482 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2483 dev->gflags ^= IFF_ALLMULTI;
2484 dev_set_allmulti(dev, inc);
2487 if (old_flags ^ dev->flags)
2488 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2493 int dev_set_mtu(struct net_device *dev, int new_mtu)
2497 if (new_mtu == dev->mtu)
2500 /* MTU must be positive. */
2504 if (!netif_device_present(dev))
2508 if (dev->change_mtu)
2509 err = dev->change_mtu(dev, new_mtu);
2512 if (!err && dev->flags & IFF_UP)
2513 notifier_call_chain(&netdev_chain,
2514 NETDEV_CHANGEMTU, dev);
2520 * Perform the SIOCxIFxxx calls.
2522 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2525 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2531 case SIOCGIFFLAGS: /* Get interface flags */
2532 ifr->ifr_flags = dev_get_flags(dev);
2535 case SIOCSIFFLAGS: /* Set interface flags */
2536 return dev_change_flags(dev, ifr->ifr_flags);
2538 case SIOCGIFMETRIC: /* Get the metric on the interface
2539 (currently unused) */
2540 ifr->ifr_metric = 0;
2543 case SIOCSIFMETRIC: /* Set the metric on the interface
2544 (currently unused) */
2547 case SIOCGIFMTU: /* Get the MTU of a device */
2548 ifr->ifr_mtu = dev->mtu;
2551 case SIOCSIFMTU: /* Set the MTU of a device */
2552 return dev_set_mtu(dev, ifr->ifr_mtu);
2555 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2556 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2557 ifr->ifr_hwaddr.sa_family = dev->type;
2561 if (!dev->set_mac_address)
2563 if (ifr->ifr_hwaddr.sa_family != dev->type)
2565 if (!netif_device_present(dev))
2567 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2569 notifier_call_chain(&netdev_chain,
2570 NETDEV_CHANGEADDR, dev);
2573 case SIOCSIFHWBROADCAST:
2574 if (ifr->ifr_hwaddr.sa_family != dev->type)
2576 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2577 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2578 notifier_call_chain(&netdev_chain,
2579 NETDEV_CHANGEADDR, dev);
2583 ifr->ifr_map.mem_start = dev->mem_start;
2584 ifr->ifr_map.mem_end = dev->mem_end;
2585 ifr->ifr_map.base_addr = dev->base_addr;
2586 ifr->ifr_map.irq = dev->irq;
2587 ifr->ifr_map.dma = dev->dma;
2588 ifr->ifr_map.port = dev->if_port;
2592 if (dev->set_config) {
2593 if (!netif_device_present(dev))
2595 return dev->set_config(dev, &ifr->ifr_map);
2600 if (!dev->set_multicast_list ||
2601 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2603 if (!netif_device_present(dev))
2605 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2609 if (!dev->set_multicast_list ||
2610 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2612 if (!netif_device_present(dev))
2614 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2618 ifr->ifr_ifindex = dev->ifindex;
2622 ifr->ifr_qlen = dev->tx_queue_len;
2626 if (ifr->ifr_qlen < 0)
2628 dev->tx_queue_len = ifr->ifr_qlen;
2632 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2633 return dev_change_name(dev, ifr->ifr_newname);
2636 * Unknown or private ioctl
2640 if ((cmd >= SIOCDEVPRIVATE &&
2641 cmd <= SIOCDEVPRIVATE + 15) ||
2642 cmd == SIOCBONDENSLAVE ||
2643 cmd == SIOCBONDRELEASE ||
2644 cmd == SIOCBONDSETHWADDR ||
2645 cmd == SIOCBONDSLAVEINFOQUERY ||
2646 cmd == SIOCBONDINFOQUERY ||
2647 cmd == SIOCBONDCHANGEACTIVE ||
2648 cmd == SIOCGMIIPHY ||
2649 cmd == SIOCGMIIREG ||
2650 cmd == SIOCSMIIREG ||
2651 cmd == SIOCBRADDIF ||
2652 cmd == SIOCBRDELIF ||
2653 cmd == SIOCWANDEV) {
2655 if (dev->do_ioctl) {
2656 if (netif_device_present(dev))
2657 err = dev->do_ioctl(dev, ifr,
2670 * This function handles all "interface"-type I/O control requests. The actual
2671 * 'doing' part of this is dev_ifsioc above.
2675 * dev_ioctl - network device ioctl
2676 * @cmd: command to issue
2677 * @arg: pointer to a struct ifreq in user space
2679 * Issue ioctl functions to devices. This is normally called by the
2680 * user space syscall interfaces but can sometimes be useful for
2681 * other purposes. The return value is the return from the syscall if
2682 * positive or a negative errno code on error.
2685 int dev_ioctl(unsigned int cmd, void __user *arg)
2691 /* One special case: SIOCGIFCONF takes ifconf argument
2692 and requires shared lock, because it sleeps writing
2696 if (cmd == SIOCGIFCONF) {
2698 ret = dev_ifconf((char __user *) arg);
2702 if (cmd == SIOCGIFNAME)
2703 return dev_ifname((struct ifreq __user *)arg);
2705 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2708 ifr.ifr_name[IFNAMSIZ-1] = 0;
2710 colon = strchr(ifr.ifr_name, ':');
2715 * See which interface the caller is talking about.
2720 * These ioctl calls:
2721 * - can be done by all.
2722 * - atomic and do not require locking.
2733 dev_load(ifr.ifr_name);
2734 read_lock(&dev_base_lock);
2735 ret = dev_ifsioc(&ifr, cmd);
2736 read_unlock(&dev_base_lock);
2740 if (copy_to_user(arg, &ifr,
2741 sizeof(struct ifreq)))
2747 dev_load(ifr.ifr_name);
2749 ret = dev_ethtool(&ifr);
2754 if (copy_to_user(arg, &ifr,
2755 sizeof(struct ifreq)))
2761 * These ioctl calls:
2762 * - require superuser power.
2763 * - require strict serialization.
2769 if (!capable(CAP_NET_ADMIN))
2771 dev_load(ifr.ifr_name);
2773 ret = dev_ifsioc(&ifr, cmd);
2778 if (copy_to_user(arg, &ifr,
2779 sizeof(struct ifreq)))
2785 * These ioctl calls:
2786 * - require superuser power.
2787 * - require strict serialization.
2788 * - do not return a value
2798 case SIOCSIFHWBROADCAST:
2801 case SIOCBONDENSLAVE:
2802 case SIOCBONDRELEASE:
2803 case SIOCBONDSETHWADDR:
2804 case SIOCBONDSLAVEINFOQUERY:
2805 case SIOCBONDINFOQUERY:
2806 case SIOCBONDCHANGEACTIVE:
2809 if (!capable(CAP_NET_ADMIN))
2811 dev_load(ifr.ifr_name);
2813 ret = dev_ifsioc(&ifr, cmd);
2818 /* Get the per device memory space. We can add this but
2819 * currently do not support it */
2821 /* Set the per device memory buffer space.
2822 * Not applicable in our case */
2827 * Unknown or private ioctl.
2830 if (cmd == SIOCWANDEV ||
2831 (cmd >= SIOCDEVPRIVATE &&
2832 cmd <= SIOCDEVPRIVATE + 15)) {
2833 dev_load(ifr.ifr_name);
2835 ret = dev_ifsioc(&ifr, cmd);
2837 if (!ret && copy_to_user(arg, &ifr,
2838 sizeof(struct ifreq)))
2843 /* Take care of Wireless Extensions */
2844 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2845 /* If command is `set a parameter', or
2846 * `get the encoding parameters', check if
2847 * the user has the right to do it */
2848 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2849 if (!capable(CAP_NET_ADMIN))
2852 dev_load(ifr.ifr_name);
2854 /* Follow me in net/core/wireless.c */
2855 ret = wireless_process_ioctl(&ifr, cmd);
2857 if (!ret && IW_IS_GET(cmd) &&
2858 copy_to_user(arg, &ifr,
2859 sizeof(struct ifreq)))
2863 #endif /* WIRELESS_EXT */
2870 * dev_new_index - allocate an ifindex
2872 * Returns a suitable unique value for a new device interface
2873 * number. The caller must hold the rtnl semaphore or the
2874 * dev_base_lock to be sure it remains unique.
2876 int dev_new_index(void)
2882 if (!__dev_get_by_index(ifindex))
2887 static int dev_boot_phase = 1;
2889 /* Delayed registration/unregisteration */
2890 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2891 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2893 static inline void net_set_todo(struct net_device *dev)
2895 spin_lock(&net_todo_list_lock);
2896 list_add_tail(&dev->todo_list, &net_todo_list);
2897 spin_unlock(&net_todo_list_lock);
2901 * register_netdevice - register a network device
2902 * @dev: device to register
2904 * Take a completed network device structure and add it to the kernel
2905 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2906 * chain. 0 is returned on success. A negative errno code is returned
2907 * on a failure to set up the device, or if the name is a duplicate.
2909 * Callers must hold the rtnl semaphore. See the comment at the
2910 * end of Space.c for details about the locking. You may want
2911 * register_netdev() instead of this.
2914 * The locking appears insufficient to guarantee two parallel registers
2915 * will not get the same name.
2918 int register_netdevice(struct net_device *dev)
2920 struct hlist_head *head;
2921 struct hlist_node *p;
2924 BUG_ON(dev_boot_phase);
2927 /* When net_device's are persistent, this will be fatal. */
2928 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2930 spin_lock_init(&dev->queue_lock);
2931 spin_lock_init(&dev->xmit_lock);
2932 dev->xmit_lock_owner = -1;
2933 #ifdef CONFIG_NET_CLS_ACT
2934 spin_lock_init(&dev->ingress_lock);
2937 #ifdef CONFIG_NET_FASTROUTE
2938 dev->fastpath_lock = RW_LOCK_UNLOCKED;
2941 ret = alloc_divert_blk(dev);
2947 /* Init, if this function is available */
2949 ret = dev->init(dev);
2957 if (!dev_valid_name(dev->name)) {
2962 dev->ifindex = dev_new_index();
2963 if (dev->iflink == -1)
2964 dev->iflink = dev->ifindex;
2966 /* Check for existence of name */
2967 head = dev_name_hash(dev->name);
2968 hlist_for_each(p, head) {
2969 struct net_device *d
2970 = hlist_entry(p, struct net_device, name_hlist);
2971 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2977 /* Fix illegal SG+CSUM combinations. */
2978 if ((dev->features & NETIF_F_SG) &&
2979 !(dev->features & (NETIF_F_IP_CSUM |
2981 NETIF_F_HW_CSUM))) {
2982 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2984 dev->features &= ~NETIF_F_SG;
2988 * nil rebuild_header routine,
2989 * that should be never called and used as just bug trap.
2992 if (!dev->rebuild_header)
2993 dev->rebuild_header = default_rebuild_header;
2996 * Default initial state at registry is that the
2997 * device is present.
3000 set_bit(__LINK_STATE_PRESENT, &dev->state);
3003 dev_init_scheduler(dev);
3004 write_lock_bh(&dev_base_lock);
3006 dev_tail = &dev->next;
3007 hlist_add_head(&dev->name_hlist, head);
3008 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3010 dev->reg_state = NETREG_REGISTERING;
3011 write_unlock_bh(&dev_base_lock);
3013 /* Notify protocols, that a new device appeared. */
3014 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3016 /* Finish registration after unlock */
3023 free_divert_blk(dev);
3028 * netdev_wait_allrefs - wait until all references are gone.
3030 * This is called when unregistering network devices.
3032 * Any protocol or device that holds a reference should register
3033 * for netdevice notification, and cleanup and put back the
3034 * reference if they receive an UNREGISTER event.
3035 * We can get stuck here if buggy protocols don't correctly
3038 static void netdev_wait_allrefs(struct net_device *dev)
3040 unsigned long rebroadcast_time, warning_time;
3042 rebroadcast_time = warning_time = jiffies;
3043 while (atomic_read(&dev->refcnt) != 0) {
3044 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3048 /* Rebroadcast unregister notification */
3049 notifier_call_chain(&netdev_chain,
3050 NETDEV_UNREGISTER, dev);
3052 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3054 /* We must not have linkwatch events
3055 * pending on unregister. If this
3056 * happens, we simply run the queue
3057 * unscheduled, resulting in a noop
3060 linkwatch_run_queue();
3066 rebroadcast_time = jiffies;
3069 current->state = TASK_INTERRUPTIBLE;
3070 schedule_timeout(HZ / 4);
3072 if (time_after(jiffies, warning_time + 10 * HZ)) {
3073 printk(KERN_EMERG "unregister_netdevice: "
3074 "waiting for %s to become free. Usage "
3076 dev->name, atomic_read(&dev->refcnt));
3077 warning_time = jiffies;
3086 * register_netdevice(x1);
3087 * register_netdevice(x2);
3089 * unregister_netdevice(y1);
3090 * unregister_netdevice(y2);
3096 * We are invoked by rtnl_unlock() after it drops the semaphore.
3097 * This allows us to deal with problems:
3098 * 1) We can create/delete sysfs objects which invoke hotplug
3099 * without deadlocking with linkwatch via keventd.
3100 * 2) Since we run with the RTNL semaphore not held, we can sleep
3101 * safely in order to wait for the netdev refcnt to drop to zero.
3103 static DECLARE_MUTEX(net_todo_run_mutex);
3104 void netdev_run_todo(void)
3106 struct list_head list = LIST_HEAD_INIT(list);
3110 /* Need to guard against multiple cpu's getting out of order. */
3111 down(&net_todo_run_mutex);
3113 /* Not safe to do outside the semaphore. We must not return
3114 * until all unregister events invoked by the local processor
3115 * have been completed (either by this todo run, or one on
3118 if (list_empty(&net_todo_list))
3121 /* Snapshot list, allow later requests */
3122 spin_lock(&net_todo_list_lock);
3123 list_splice_init(&net_todo_list, &list);
3124 spin_unlock(&net_todo_list_lock);
3126 while (!list_empty(&list)) {
3127 struct net_device *dev
3128 = list_entry(list.next, struct net_device, todo_list);
3129 list_del(&dev->todo_list);
3131 switch(dev->reg_state) {
3132 case NETREG_REGISTERING:
3133 err = netdev_register_sysfs(dev);
3135 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3137 dev->reg_state = NETREG_REGISTERED;
3140 case NETREG_UNREGISTERING:
3141 netdev_unregister_sysfs(dev);
3142 dev->reg_state = NETREG_UNREGISTERED;
3144 netdev_wait_allrefs(dev);
3147 BUG_ON(atomic_read(&dev->refcnt));
3148 BUG_TRAP(!dev->ip_ptr);
3149 BUG_TRAP(!dev->ip6_ptr);
3150 BUG_TRAP(!dev->dn_ptr);
3153 /* It must be the very last action,
3154 * after this 'dev' may point to freed up memory.
3156 if (dev->destructor)
3157 dev->destructor(dev);
3161 printk(KERN_ERR "network todo '%s' but state %d\n",
3162 dev->name, dev->reg_state);
3168 up(&net_todo_run_mutex);
3172 * free_netdev - free network device
3175 * This function does the last stage of destroying an allocated device
3176 * interface. The reference to the device object is released.
3177 * If this is the last reference then it will be freed.
3179 void free_netdev(struct net_device *dev)
3182 /* Compatiablity with error handling in drivers */
3183 if (dev->reg_state == NETREG_UNINITIALIZED) {
3184 kfree((char *)dev - dev->padded);
3188 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3189 dev->reg_state = NETREG_RELEASED;
3191 /* will free via class release */
3192 class_device_put(&dev->class_dev);
3194 kfree((char *)dev - dev->padded);
3198 /* Synchronize with packet receive processing. */
3199 void synchronize_net(void)
3202 synchronize_kernel();
3206 * unregister_netdevice - remove device from the kernel
3209 * This function shuts down a device interface and removes it
3210 * from the kernel tables. On success 0 is returned, on a failure
3211 * a negative errno code is returned.
3213 * Callers must hold the rtnl semaphore. See the comment at the
3214 * end of Space.c for details about the locking. You may want
3215 * unregister_netdev() instead of this.
3218 int unregister_netdevice(struct net_device *dev)
3220 struct net_device *d, **dp;
3222 BUG_ON(dev_boot_phase);
3225 /* Some devices call without registering for initialization unwind. */
3226 if (dev->reg_state == NETREG_UNINITIALIZED) {
3227 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3228 "was registered\n", dev->name, dev);
3232 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3234 /* If device is running, close it first. */
3235 if (dev->flags & IFF_UP)
3238 /* And unlink it from device chain. */
3239 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3241 write_lock_bh(&dev_base_lock);
3242 hlist_del(&dev->name_hlist);
3243 hlist_del(&dev->index_hlist);
3244 if (dev_tail == &dev->next)
3247 write_unlock_bh(&dev_base_lock);
3252 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3257 dev->reg_state = NETREG_UNREGISTERING;
3261 #ifdef CONFIG_NET_FASTROUTE
3262 dev_clear_fastroute(dev);
3265 /* Shutdown queueing discipline. */
3269 /* Notify protocols, that we are about to destroy
3270 this device. They should clean all the things.
3272 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3275 * Flush the multicast chain
3277 dev_mc_discard(dev);
3282 /* Notifier chain MUST detach us from master device. */
3283 BUG_TRAP(!dev->master);
3285 free_divert_blk(dev);
3287 /* Finish processing unregister after unlock */
3296 #ifdef CONFIG_HOTPLUG_CPU
3297 static int dev_cpu_callback(struct notifier_block *nfb,
3298 unsigned long action,
3301 struct sk_buff **list_skb;
3302 struct net_device **list_net;
3303 struct sk_buff *skb;
3304 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3305 struct softnet_data *sd, *oldsd;
3307 if (action != CPU_DEAD)
3310 local_irq_disable();
3311 cpu = smp_processor_id();
3312 sd = &per_cpu(softnet_data, cpu);
3313 oldsd = &per_cpu(softnet_data, oldcpu);
3315 /* Find end of our completion_queue. */
3316 list_skb = &sd->completion_queue;
3318 list_skb = &(*list_skb)->next;
3319 /* Append completion queue from offline CPU. */
3320 *list_skb = oldsd->completion_queue;
3321 oldsd->completion_queue = NULL;
3323 /* Find end of our output_queue. */
3324 list_net = &sd->output_queue;
3326 list_net = &(*list_net)->next_sched;
3327 /* Append output queue from offline CPU. */
3328 *list_net = oldsd->output_queue;
3329 oldsd->output_queue = NULL;
3331 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3334 /* Process offline CPU's input_pkt_queue */
3335 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3340 #endif /* CONFIG_HOTPLUG_CPU */
3344 * Initialize the DEV module. At boot time this walks the device list and
3345 * unhooks any devices that fail to initialise (normally hardware not
3346 * present) and leaves us with a valid list of present and active devices.
3351 * This is called single threaded during boot, so no need
3352 * to take the rtnl semaphore.
3354 static int __init net_dev_init(void)
3356 int i, rc = -ENOMEM;
3358 BUG_ON(!dev_boot_phase);
3360 if (dev_proc_init())
3363 if (netdev_sysfs_init())
3366 INIT_LIST_HEAD(&ptype_all);
3367 for (i = 0; i < 16; i++)
3368 INIT_LIST_HEAD(&ptype_base[i]);
3370 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3371 INIT_HLIST_HEAD(&dev_name_head[i]);
3373 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3374 INIT_HLIST_HEAD(&dev_index_head[i]);
3377 * Initialise the packet receive queues.
3380 for (i = 0; i < NR_CPUS; i++) {
3381 struct softnet_data *queue;
3383 queue = &per_cpu(softnet_data, i);
3384 skb_queue_head_init(&queue->input_pkt_queue);
3385 queue->throttle = 0;
3386 queue->cng_level = 0;
3387 queue->avg_blog = 10; /* arbitrary non-zero */
3388 queue->completion_queue = NULL;
3389 INIT_LIST_HEAD(&queue->poll_list);
3390 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3391 queue->backlog_dev.weight = weight_p;
3392 queue->backlog_dev.poll = process_backlog;
3393 atomic_set(&queue->backlog_dev.refcnt, 1);
3396 #ifdef OFFLINE_SAMPLE
3397 samp_timer.expires = jiffies + (10 * HZ);
3398 add_timer(&samp_timer);
3403 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3404 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3406 hotcpu_notifier(dev_cpu_callback, 0);
3414 subsys_initcall(net_dev_init);
3416 EXPORT_SYMBOL(__dev_get);
3417 EXPORT_SYMBOL(__dev_get_by_flags);
3418 EXPORT_SYMBOL(__dev_get_by_index);
3419 EXPORT_SYMBOL(__dev_get_by_name);
3420 EXPORT_SYMBOL(__dev_remove_pack);
3421 EXPORT_SYMBOL(__skb_linearize);
3422 EXPORT_SYMBOL(call_netdevice_notifiers);
3423 EXPORT_SYMBOL(dev_add_pack);
3424 EXPORT_SYMBOL(dev_alloc_name);
3425 EXPORT_SYMBOL(dev_close);
3426 EXPORT_SYMBOL(dev_get_by_flags);
3427 EXPORT_SYMBOL(dev_get_by_index);
3428 EXPORT_SYMBOL(dev_get_by_name);
3429 EXPORT_SYMBOL(dev_getbyhwaddr);
3430 EXPORT_SYMBOL(dev_ioctl);
3431 EXPORT_SYMBOL(dev_new_index);
3432 EXPORT_SYMBOL(dev_open);
3433 EXPORT_SYMBOL(dev_queue_xmit);
3434 EXPORT_SYMBOL(dev_queue_xmit_nit);
3435 EXPORT_SYMBOL(dev_remove_pack);
3436 EXPORT_SYMBOL(dev_set_allmulti);
3437 EXPORT_SYMBOL(dev_set_promiscuity);
3438 EXPORT_SYMBOL(free_netdev);
3439 EXPORT_SYMBOL(netdev_boot_setup_check);
3440 EXPORT_SYMBOL(netdev_set_master);
3441 EXPORT_SYMBOL(netdev_state_change);
3442 EXPORT_SYMBOL(netif_receive_skb);
3443 EXPORT_SYMBOL(netif_rx);
3444 EXPORT_SYMBOL(register_gifconf);
3445 EXPORT_SYMBOL(register_netdevice);
3446 EXPORT_SYMBOL(register_netdevice_notifier);
3447 EXPORT_SYMBOL(skb_checksum_help);
3448 EXPORT_SYMBOL(synchronize_net);
3449 EXPORT_SYMBOL(unregister_netdevice);
3450 EXPORT_SYMBOL(unregister_netdevice_notifier);
3452 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3453 EXPORT_SYMBOL(br_handle_frame_hook);
3455 /* for 801q VLAN support */
3456 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
3457 EXPORT_SYMBOL(dev_change_flags);
3460 EXPORT_SYMBOL(dev_load);
3462 #ifdef CONFIG_NET_HW_FLOWCONTROL
3463 EXPORT_SYMBOL(netdev_dropping);
3464 EXPORT_SYMBOL(netdev_fc_xoff);
3465 EXPORT_SYMBOL(netdev_register_fc);
3466 EXPORT_SYMBOL(netdev_unregister_fc);
3468 #ifdef CONFIG_NET_FASTROUTE
3469 EXPORT_SYMBOL(netdev_fastroute);
3470 EXPORT_SYMBOL(netdev_fastroute_obstacles);
3473 #ifdef CONFIG_NET_CLS_ACT
3474 EXPORT_SYMBOL(ing_filter);
3478 EXPORT_PER_CPU_SYMBOL(softnet_data);