2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <asm/bitops.h>
78 #include <linux/config.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/notifier.h>
93 #include <linux/skbuff.h>
95 #include <linux/rtnetlink.h>
96 #include <linux/proc_fs.h>
97 #include <linux/seq_file.h>
98 #include <linux/stat.h>
99 #include <linux/if_bridge.h>
100 #include <linux/divert.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #ifdef CONFIG_NET_RADIO
112 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
113 #include <net/iw_handler.h>
114 #endif /* CONFIG_NET_RADIO */
115 #include <asm/current.h>
117 /* This define, if set, will randomly drop a packet when congestion
118 * is more than moderate. It helps fairness in the multi-interface
119 * case when one of them is a hog, but it kills performance for the
120 * single interface case so it is off now by default.
124 /* Setting this will sample the queue lengths and thus congestion
125 * via a timer instead of as each packet is received.
127 #undef OFFLINE_SAMPLE
130 * The list of packet types we will receive (as opposed to discard)
131 * and the routines to invoke.
133 * Why 16. Because with 16 the only overlap we get on a hash of the
134 * low nibble of the protocol value is RARP/SNAP/X.25.
136 * NOTE: That is no longer true with the addition of VLAN tags. Not
137 * sure which should go first, but I bet it won't make much
138 * difference if we are running VLANs. The good news is that
139 * this protocol won't be in the list unless compiled in, so
140 * the average user (w/out VLANs) will not be adversly affected.
157 static spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
158 static struct list_head ptype_base[16]; /* 16 way hashed list */
159 static struct list_head ptype_all; /* Taps */
161 #ifdef OFFLINE_SAMPLE
162 static void sample_queue(unsigned long dummy);
163 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
167 * The @dev_base list is protected by @dev_base_lock and the rtln
170 * Pure readers hold dev_base_lock for reading.
172 * Writers must hold the rtnl semaphore while they loop through the
173 * dev_base list, and hold dev_base_lock for writing when they do the
174 * actual updates. This allows pure readers to access the list even
175 * while a writer is preparing to update it.
177 * To put it another way, dev_base_lock is held for writing only to
178 * protect against pure readers; the rtnl semaphore provides the
179 * protection against other writers.
181 * See, for example usages, register_netdevice() and
182 * unregister_netdevice(), which must be called with the rtnl
185 struct net_device *dev_base;
186 struct net_device **dev_tail = &dev_base;
187 rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
189 EXPORT_SYMBOL(dev_base);
190 EXPORT_SYMBOL(dev_base_lock);
192 #define NETDEV_HASHBITS 8
193 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
194 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
196 static inline struct hlist_head *dev_name_hash(const char *name)
198 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
199 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
202 static inline struct hlist_head *dev_index_hash(int ifindex)
204 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
211 static struct notifier_block *netdev_chain;
214 * Device drivers call our routines to queue packets here. We empty the
215 * queue in the local softnet handler.
217 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
220 extern int netdev_sysfs_init(void);
221 extern int netdev_register_sysfs(struct net_device *);
222 extern void netdev_unregister_sysfs(struct net_device *);
224 #define netdev_sysfs_init() (0)
225 #define netdev_register_sysfs(dev) (0)
226 #define netdev_unregister_sysfs(dev) do { } while(0)
229 /* netdump function */
230 void (*netdump_func) (struct pt_regs *regs) = NULL;
232 /*******************************************************************************
234 Protocol management and registration routines
236 *******************************************************************************/
245 * Add a protocol ID to the list. Now that the input handler is
246 * smarter we can dispense with all the messy stuff that used to be
249 * BEWARE!!! Protocol handlers, mangling input packets,
250 * MUST BE last in hash buckets and checking protocol handlers
251 * MUST start from promiscuous ptype_all chain in net_bh.
252 * It is true now, do not change it.
253 * Explanation follows: if protocol handler, mangling packet, will
254 * be the first on list, it is not able to sense, that packet
255 * is cloned and should be copied-on-write, so that it will
256 * change it and subsequent readers will get broken packet.
261 * dev_add_pack - add packet handler
262 * @pt: packet type declaration
264 * Add a protocol handler to the networking stack. The passed &packet_type
265 * is linked into kernel lists and may not be freed until it has been
266 * removed from the kernel lists.
268 * This call does not sleep therefore it can not
269 * guarantee all CPU's that are in middle of receiving packets
270 * will see the new packet type (until the next received packet).
273 void dev_add_pack(struct packet_type *pt)
277 spin_lock_bh(&ptype_lock);
278 if (pt->type == htons(ETH_P_ALL)) {
280 list_add_rcu(&pt->list, &ptype_all);
282 hash = ntohs(pt->type) & 15;
283 list_add_rcu(&pt->list, &ptype_base[hash]);
285 spin_unlock_bh(&ptype_lock);
288 extern void linkwatch_run_queue(void);
293 * __dev_remove_pack - remove packet handler
294 * @pt: packet type declaration
296 * Remove a protocol handler that was previously added to the kernel
297 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
298 * from the kernel lists and can be freed or reused once this function
301 * The packet type might still be in use by receivers
302 * and must not be freed until after all the CPU's have gone
303 * through a quiescent state.
305 void __dev_remove_pack(struct packet_type *pt)
307 struct list_head *head;
308 struct packet_type *pt1;
310 spin_lock_bh(&ptype_lock);
312 if (pt->type == htons(ETH_P_ALL)) {
316 head = &ptype_base[ntohs(pt->type) & 15];
318 list_for_each_entry(pt1, head, list) {
320 list_del_rcu(&pt->list);
325 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
327 spin_unlock_bh(&ptype_lock);
330 * dev_remove_pack - remove packet handler
331 * @pt: packet type declaration
333 * Remove a protocol handler that was previously added to the kernel
334 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
335 * from the kernel lists and can be freed or reused once this function
338 * This call sleeps to guarantee that no CPU is looking at the packet
341 void dev_remove_pack(struct packet_type *pt)
343 __dev_remove_pack(pt);
348 /******************************************************************************
350 Device Boot-time Settings Routines
352 *******************************************************************************/
354 /* Boot time configuration table */
355 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
358 * netdev_boot_setup_add - add new setup entry
359 * @name: name of the device
360 * @map: configured settings for the device
362 * Adds new setup entry to the dev_boot_setup list. The function
363 * returns 0 on error and 1 on success. This is a generic routine to
366 int netdev_boot_setup_add(char *name, struct ifmap *map)
368 struct netdev_boot_setup *s;
372 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
373 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
374 memset(s[i].name, 0, sizeof(s[i].name));
375 strcpy(s[i].name, name);
376 memcpy(&s[i].map, map, sizeof(s[i].map));
381 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
385 * netdev_boot_setup_check - check boot time settings
386 * @dev: the netdevice
388 * Check boot time settings for the device.
389 * The found settings are set for the device to be used
390 * later in the device probing.
391 * Returns 0 if no settings found, 1 if they are.
393 int netdev_boot_setup_check(struct net_device *dev)
395 struct netdev_boot_setup *s = dev_boot_setup;
398 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
399 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
400 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
401 dev->irq = s[i].map.irq;
402 dev->base_addr = s[i].map.base_addr;
403 dev->mem_start = s[i].map.mem_start;
404 dev->mem_end = s[i].map.mem_end;
413 * netdev_boot_base - get address from boot time settings
414 * @prefix: prefix for network device
415 * @unit: id for network device
417 * Check boot time settings for the base address of device.
418 * The found settings are set for the device to be used
419 * later in the device probing.
420 * Returns 0 if no settings found.
422 unsigned long netdev_boot_base(const char *prefix, int unit)
424 const struct netdev_boot_setup *s = dev_boot_setup;
428 sprintf(name, "%s%d", prefix, unit);
431 * If device already registered then return base of 1
432 * to indicate not to probe for this interface
434 if (__dev_get_by_name(name))
437 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
438 if (!strcmp(name, s[i].name))
439 return s[i].map.base_addr;
444 * Saves at boot time configured settings for any netdevice.
446 int __init netdev_boot_setup(char *str)
451 str = get_options(str, ARRAY_SIZE(ints), ints);
456 memset(&map, 0, sizeof(map));
460 map.base_addr = ints[2];
462 map.mem_start = ints[3];
464 map.mem_end = ints[4];
466 /* Add new entry to the list */
467 return netdev_boot_setup_add(str, &map);
470 __setup("netdev=", netdev_boot_setup);
472 /*******************************************************************************
474 Device Interface Subroutines
476 *******************************************************************************/
479 * __dev_get_by_name - find a device by its name
480 * @name: name to find
482 * Find an interface by name. Must be called under RTNL semaphore
483 * or @dev_base_lock. If the name is found a pointer to the device
484 * is returned. If the name is not found then %NULL is returned. The
485 * reference counters are not incremented so the caller must be
486 * careful with locks.
489 struct net_device *__dev_get_by_name(const char *name)
491 struct hlist_node *p;
493 hlist_for_each(p, dev_name_hash(name)) {
494 struct net_device *dev
495 = hlist_entry(p, struct net_device, name_hlist);
496 if (!strncmp(dev->name, name, IFNAMSIZ))
503 * dev_get_by_name - find a device by its name
504 * @name: name to find
506 * Find an interface by name. This can be called from any
507 * context and does its own locking. The returned handle has
508 * the usage count incremented and the caller must use dev_put() to
509 * release it when it is no longer needed. %NULL is returned if no
510 * matching device is found.
513 struct net_device *dev_get_by_name(const char *name)
515 struct net_device *dev;
517 read_lock(&dev_base_lock);
518 dev = __dev_get_by_name(name);
521 read_unlock(&dev_base_lock);
526 Return value is changed to int to prevent illegal usage in future.
527 It is still legal to use to check for device existence.
529 User should understand, that the result returned by this function
530 is meaningless, if it was not issued under rtnl semaphore.
534 * dev_get - test if a device exists
535 * @name: name to test for
537 * Test if a name exists. Returns true if the name is found. In order
538 * to be sure the name is not allocated or removed during the test the
539 * caller must hold the rtnl semaphore.
541 * This function exists only for back compatibility with older
544 int __dev_get(const char *name)
546 struct net_device *dev;
548 read_lock(&dev_base_lock);
549 dev = __dev_get_by_name(name);
550 read_unlock(&dev_base_lock);
555 * __dev_get_by_index - find a device by its ifindex
556 * @ifindex: index of device
558 * Search for an interface by index. Returns %NULL if the device
559 * is not found or a pointer to the device. The device has not
560 * had its reference counter increased so the caller must be careful
561 * about locking. The caller must hold either the RTNL semaphore
565 struct net_device *__dev_get_by_index(int ifindex)
567 struct hlist_node *p;
569 hlist_for_each(p, dev_index_hash(ifindex)) {
570 struct net_device *dev
571 = hlist_entry(p, struct net_device, index_hlist);
572 if (dev->ifindex == ifindex)
580 * dev_get_by_index - find a device by its ifindex
581 * @ifindex: index of device
583 * Search for an interface by index. Returns NULL if the device
584 * is not found or a pointer to the device. The device returned has
585 * had a reference added and the pointer is safe until the user calls
586 * dev_put to indicate they have finished with it.
589 struct net_device *dev_get_by_index(int ifindex)
591 struct net_device *dev;
593 read_lock(&dev_base_lock);
594 dev = __dev_get_by_index(ifindex);
597 read_unlock(&dev_base_lock);
602 * dev_getbyhwaddr - find a device by its hardware address
603 * @type: media type of device
604 * @ha: hardware address
606 * Search for an interface by MAC address. Returns NULL if the device
607 * is not found or a pointer to the device. The caller must hold the
608 * rtnl semaphore. The returned device has not had its ref count increased
609 * and the caller must therefore be careful about locking
612 * If the API was consistent this would be __dev_get_by_hwaddr
615 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
617 struct net_device *dev;
621 for (dev = dev_base; dev; dev = dev->next)
622 if (dev->type == type &&
623 !memcmp(dev->dev_addr, ha, dev->addr_len))
628 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
630 struct net_device *dev;
632 for (dev = dev_base; dev; dev = dev->next)
633 if (dev->type == type)
638 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
640 struct net_device *dev_getfirstbyhwtype(unsigned short type)
642 struct net_device *dev;
645 dev = __dev_getfirstbyhwtype(type);
652 EXPORT_SYMBOL(dev_getfirstbyhwtype);
655 * dev_get_by_flags - find any device with given flags
656 * @if_flags: IFF_* values
657 * @mask: bitmask of bits in if_flags to check
659 * Search for any interface with the given flags. Returns NULL if a device
660 * is not found or a pointer to the device. The device returned has
661 * had a reference added and the pointer is safe until the user calls
662 * dev_put to indicate they have finished with it.
665 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
667 struct net_device *dev;
669 read_lock(&dev_base_lock);
670 dev = __dev_get_by_flags(if_flags, mask);
673 read_unlock(&dev_base_lock);
678 * __dev_get_by_flags - find any device with given flags
679 * @if_flags: IFF_* values
680 * @mask: bitmask of bits in if_flags to check
682 * Search for any interface with the given flags. Returns NULL if a device
683 * is not found or a pointer to the device. The caller must hold either
684 * the RTNL semaphore or @dev_base_lock.
687 struct net_device *__dev_get_by_flags(unsigned short if_flags, unsigned short mask)
689 struct net_device *dev;
691 for (dev = dev_base; dev != NULL; dev = dev->next) {
692 if (((dev->flags ^ if_flags) & mask) == 0)
699 * dev_valid_name - check if name is okay for network device
702 * Network device names need to be valid file names to
703 * to allow sysfs to work
705 int dev_valid_name(const char *name)
707 return !(*name == '\0'
708 || !strcmp(name, ".")
709 || !strcmp(name, "..")
710 || strchr(name, '/'));
714 * dev_alloc_name - allocate a name for a device
716 * @name: name format string
718 * Passed a format string - eg "lt%d" it will try and find a suitable
719 * id. Not efficient for many devices, not called a lot. The caller
720 * must hold the dev_base or rtnl lock while allocating the name and
721 * adding the device in order to avoid duplicates. Returns the number
722 * of the unit assigned or a negative errno code.
725 int dev_alloc_name(struct net_device *dev, const char *name)
730 const int max_netdevices = 8*PAGE_SIZE;
732 struct net_device *d;
734 p = strnchr(name, IFNAMSIZ-1, '%');
737 * Verify the string as this thing may have come from
738 * the user. There must be either one "%d" and no other "%"
741 if (p[1] != 'd' || strchr(p + 2, '%'))
744 /* Use one page as a bit array of possible slots */
745 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
749 for (d = dev_base; d; d = d->next) {
750 if (!sscanf(d->name, name, &i))
752 if (i < 0 || i >= max_netdevices)
755 /* avoid cases where sscanf is not exact inverse of printf */
756 snprintf(buf, sizeof(buf), name, i);
757 if (!strncmp(buf, d->name, IFNAMSIZ))
761 i = find_first_zero_bit(inuse, max_netdevices);
762 free_page((unsigned long) inuse);
765 snprintf(buf, sizeof(buf), name, i);
766 if (!__dev_get_by_name(buf)) {
767 strlcpy(dev->name, buf, IFNAMSIZ);
771 /* It is possible to run out of possible slots
772 * when the name is long and there isn't enough space left
773 * for the digits, or if all bits are used.
780 * dev_change_name - change name of a device
782 * @newname: name (or format string) must be at least IFNAMSIZ
784 * Change name of a device, can pass format strings "eth%d".
787 int dev_change_name(struct net_device *dev, char *newname)
793 if (dev->flags & IFF_UP)
796 if (!dev_valid_name(newname))
799 if (strchr(newname, '%')) {
800 err = dev_alloc_name(dev, newname);
803 strcpy(newname, dev->name);
805 else if (__dev_get_by_name(newname))
808 strlcpy(dev->name, newname, IFNAMSIZ);
810 err = class_device_rename(&dev->class_dev, dev->name);
812 hlist_del(&dev->name_hlist);
813 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
814 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
821 * netdev_state_change - device changes state
822 * @dev: device to cause notification
824 * Called to indicate a device has changed state. This function calls
825 * the notifier chains for netdev_chain and sends a NEWLINK message
826 * to the routing socket.
828 void netdev_state_change(struct net_device *dev)
830 if (dev->flags & IFF_UP) {
831 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
832 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
837 * dev_load - load a network module
838 * @name: name of interface
840 * If a network interface is not present and the process has suitable
841 * privileges this function loads the module. If module loading is not
842 * available in this kernel then it becomes a nop.
845 void dev_load(const char *name)
847 struct net_device *dev;
849 read_lock(&dev_base_lock);
850 dev = __dev_get_by_name(name);
851 read_unlock(&dev_base_lock);
853 if (!dev && capable(CAP_SYS_MODULE))
854 request_module("%s", name);
857 static int default_rebuild_header(struct sk_buff *skb)
859 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
860 skb->dev ? skb->dev->name : "NULL!!!");
867 * Some old buggy device drivers change get_stats after registering
868 * the device. Try and trap them here.
869 * This can be elimnated when all devices are known fixed.
871 static inline int get_stats_changed(struct net_device *dev)
873 int changed = dev->last_stats != dev->get_stats;
874 dev->last_stats = dev->get_stats;
879 * dev_open - prepare an interface for use.
880 * @dev: device to open
882 * Takes a device from down to up state. The device's private open
883 * function is invoked and then the multicast lists are loaded. Finally
884 * the device is moved into the up state and a %NETDEV_UP message is
885 * sent to the netdev notifier chain.
887 * Calling this function on an active interface is a nop. On a failure
888 * a negative errno code is returned.
890 int dev_open(struct net_device *dev)
898 if (dev->flags & IFF_UP)
902 * Check for broken device drivers.
904 if (get_stats_changed(dev) && net_ratelimit()) {
905 printk(KERN_ERR "%s: driver changed get_stats after register\n",
910 * Is it even present?
912 if (!netif_device_present(dev))
916 * Call device private open method
918 set_bit(__LINK_STATE_START, &dev->state);
920 ret = dev->open(dev);
922 clear_bit(__LINK_STATE_START, &dev->state);
926 * Check for more broken device drivers.
928 if (get_stats_changed(dev) && net_ratelimit()) {
929 printk(KERN_ERR "%s: driver changed get_stats in open\n",
934 * If it went open OK then:
941 dev->flags |= IFF_UP;
944 * Initialize multicasting status
949 * Wakeup transmit queue engine
954 * ... and announce new interface.
956 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
962 * dev_close - shutdown an interface.
963 * @dev: device to shutdown
965 * This function moves an active device into down state. A
966 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
967 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
970 int dev_close(struct net_device *dev)
972 if (!(dev->flags & IFF_UP))
976 * Tell people we are going down, so that they can
977 * prepare to death, when device is still operating.
979 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
983 clear_bit(__LINK_STATE_START, &dev->state);
985 /* Synchronize to scheduled poll. We cannot touch poll list,
986 * it can be even on different cpu. So just clear netif_running(),
987 * and wait when poll really will happen. Actually, the best place
988 * for this is inside dev->stop() after device stopped its irq
989 * engine, but this requires more changes in devices. */
991 smp_mb__after_clear_bit(); /* Commit netif_running(). */
992 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
994 current->state = TASK_INTERRUPTIBLE;
999 * Call the device specific close. This cannot fail.
1000 * Only if device is UP
1002 * We allow it to be called even after a DETACH hot-plug
1009 * Device is now down.
1012 dev->flags &= ~IFF_UP;
1015 * Tell people we are down
1017 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1024 * Device change register/unregister. These are not inline or static
1025 * as we export them to the world.
1029 * register_netdevice_notifier - register a network notifier block
1032 * Register a notifier to be called when network device events occur.
1033 * The notifier passed is linked into the kernel structures and must
1034 * not be reused until it has been unregistered. A negative errno code
1035 * is returned on a failure.
1037 * When registered all registration and up events are replayed
1038 * to the new notifier to allow device to have a race free
1039 * view of the network device list.
1042 int register_netdevice_notifier(struct notifier_block *nb)
1044 struct net_device *dev;
1048 err = notifier_chain_register(&netdev_chain, nb);
1050 for (dev = dev_base; dev; dev = dev->next) {
1051 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1053 if (dev->flags & IFF_UP)
1054 nb->notifier_call(nb, NETDEV_UP, dev);
1062 * unregister_netdevice_notifier - unregister a network notifier block
1065 * Unregister a notifier previously registered by
1066 * register_netdevice_notifier(). The notifier is unlinked into the
1067 * kernel structures and may then be reused. A negative errno code
1068 * is returned on a failure.
1071 int unregister_netdevice_notifier(struct notifier_block *nb)
1073 return notifier_chain_unregister(&netdev_chain, nb);
1077 * call_netdevice_notifiers - call all network notifier blocks
1078 * @val: value passed unmodified to notifier function
1079 * @v: pointer passed unmodified to notifier function
1081 * Call all network notifier blocks. Parameters and return value
1082 * are as for notifier_call_chain().
1085 int call_netdevice_notifiers(unsigned long val, void *v)
1087 return notifier_call_chain(&netdev_chain, val, v);
1091 * Support routine. Sends outgoing frames to any network
1092 * taps currently in use.
1095 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1097 struct packet_type *ptype;
1098 net_timestamp(&skb->stamp);
1101 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1102 /* Never send packets back to the socket
1103 * they originated from - MvS (miquels@drinkel.ow.org)
1105 if ((ptype->dev == dev || !ptype->dev) &&
1106 (ptype->af_packet_priv == NULL ||
1107 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1108 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1112 /* skb->nh should be correctly
1113 set by sender, so that the second statement is
1114 just protection against buggy protocols.
1116 skb2->mac.raw = skb2->data;
1118 if (skb2->nh.raw < skb2->data ||
1119 skb2->nh.raw > skb2->tail) {
1120 if (net_ratelimit())
1121 printk(KERN_CRIT "protocol %04x is "
1123 skb2->protocol, dev->name);
1124 skb2->nh.raw = skb2->data;
1127 skb2->h.raw = skb2->nh.raw;
1128 skb2->pkt_type = PACKET_OUTGOING;
1129 ptype->func(skb2, skb->dev, ptype);
1136 * Invalidate hardware checksum when packet is to be mangled, and
1137 * complete checksum manually on outgoing path.
1139 int skb_checksum_help(struct sk_buff **pskb, int inward)
1142 int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1145 (*pskb)->ip_summed = CHECKSUM_NONE;
1149 if (skb_shared(*pskb) || skb_cloned(*pskb)) {
1150 struct sk_buff *newskb = skb_copy(*pskb, GFP_ATOMIC);
1156 skb_set_owner_w(newskb, (*pskb)->sk);
1161 if (offset > (int)(*pskb)->len)
1163 csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1165 offset = (*pskb)->tail - (*pskb)->h.raw;
1168 if ((*pskb)->csum + 2 > offset)
1171 *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1172 (*pskb)->ip_summed = CHECKSUM_NONE;
1177 #ifdef CONFIG_HIGHMEM
1178 /* Actually, we should eliminate this check as soon as we know, that:
1179 * 1. IOMMU is present and allows to map all the memory.
1180 * 2. No high memory really exists on this machine.
1183 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1187 if (dev->features & NETIF_F_HIGHDMA)
1190 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1191 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1197 #define illegal_highdma(dev, skb) (0)
1200 extern void skb_release_data(struct sk_buff *);
1202 /* Keep head the same: replace data */
1203 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1208 struct skb_shared_info *ninfo;
1209 int headerlen = skb->data - skb->head;
1210 int expand = (skb->tail + skb->data_len) - skb->end;
1212 if (skb_shared(skb))
1218 size = skb->end - skb->head + expand;
1219 size = SKB_DATA_ALIGN(size);
1220 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1224 /* Copy entire thing */
1225 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1229 ninfo = (struct skb_shared_info*)(data + size);
1230 atomic_set(&ninfo->dataref, 1);
1231 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1232 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1233 ninfo->nr_frags = 0;
1234 ninfo->frag_list = NULL;
1236 /* Offset between the two in bytes */
1237 offset = data - skb->head;
1239 /* Free old data. */
1240 skb_release_data(skb);
1243 skb->end = data + size;
1245 /* Set up new pointers */
1246 skb->h.raw += offset;
1247 skb->nh.raw += offset;
1248 skb->mac.raw += offset;
1249 skb->tail += offset;
1250 skb->data += offset;
1252 /* We are no longer a clone, even if we were. */
1255 skb->tail += skb->data_len;
1260 #define HARD_TX_LOCK_BH(dev, cpu) { \
1261 if ((dev->features & NETIF_F_LLTX) == 0) { \
1262 spin_lock_bh(&dev->xmit_lock); \
1263 dev->xmit_lock_owner = cpu; \
1267 #define HARD_TX_UNLOCK_BH(dev) { \
1268 if ((dev->features & NETIF_F_LLTX) == 0) { \
1269 dev->xmit_lock_owner = -1; \
1270 spin_unlock_bh(&dev->xmit_lock); \
1274 static inline void qdisc_run(struct net_device *dev)
1276 while (!netif_queue_stopped(dev) &&
1277 qdisc_restart(dev)<0)
1282 * dev_queue_xmit - transmit a buffer
1283 * @skb: buffer to transmit
1285 * Queue a buffer for transmission to a network device. The caller must
1286 * have set the device and priority and built the buffer before calling
1287 * this function. The function can be called from an interrupt.
1289 * A negative errno code is returned on a failure. A success does not
1290 * guarantee the frame will be transmitted as it may be dropped due
1291 * to congestion or traffic shaping.
1294 int dev_queue_xmit(struct sk_buff *skb)
1296 struct net_device *dev = skb->dev;
1300 if (skb_shinfo(skb)->frag_list &&
1301 !(dev->features & NETIF_F_FRAGLIST) &&
1302 __skb_linearize(skb, GFP_ATOMIC))
1305 /* Fragmented skb is linearized if device does not support SG,
1306 * or if at least one of fragments is in highmem and device
1307 * does not support DMA from it.
1309 if (skb_shinfo(skb)->nr_frags &&
1310 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1311 __skb_linearize(skb, GFP_ATOMIC))
1314 /* If packet is not checksummed and device does not support
1315 * checksumming for this protocol, complete checksumming here.
1317 if (skb->ip_summed == CHECKSUM_HW &&
1318 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1319 (!(dev->features & NETIF_F_IP_CSUM) ||
1320 skb->protocol != htons(ETH_P_IP))))
1321 if (skb_checksum_help(&skb, 0))
1325 /* Updates of qdisc are serialized by queue_lock.
1326 * The struct Qdisc which is pointed to by qdisc is now a
1327 * rcu structure - it may be accessed without acquiring
1328 * a lock (but the structure may be stale.) The freeing of the
1329 * qdisc will be deferred until it's known that there are no
1330 * more references to it.
1332 * If the qdisc has an enqueue function, we still need to
1333 * hold the queue_lock before calling it, since queue_lock
1334 * also serializes access to the device queue.
1338 smp_read_barrier_depends();
1339 #ifdef CONFIG_NET_CLS_ACT
1340 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1343 /* Grab device queue */
1344 spin_lock_bh(&dev->queue_lock);
1346 rc = q->enqueue(skb, q);
1350 spin_unlock_bh(&dev->queue_lock);
1352 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1357 /* The device has no queue. Common case for software devices:
1358 loopback, all the sorts of tunnels...
1360 Really, it is unlikely that xmit_lock protection is necessary here.
1361 (f.e. loopback and IP tunnels are clean ignoring statistics
1363 However, it is possible, that they rely on protection
1366 Check this and shot the lock. It is not prone from deadlocks.
1367 Either shot noqueue qdisc, it is even simpler 8)
1369 if (dev->flags & IFF_UP) {
1370 int cpu = get_cpu();
1372 if (dev->xmit_lock_owner != cpu) {
1374 HARD_TX_LOCK_BH(dev, cpu);
1377 if (!netif_queue_stopped(dev)) {
1379 dev_queue_xmit_nit(skb, dev);
1382 if (!dev->hard_start_xmit(skb, dev)) {
1383 HARD_TX_UNLOCK_BH(dev);
1387 HARD_TX_UNLOCK_BH(dev);
1388 if (net_ratelimit())
1389 printk(KERN_CRIT "Virtual device %s asks to "
1390 "queue packet!\n", dev->name);
1394 /* Recursion is detected! It is possible,
1396 if (net_ratelimit())
1397 printk(KERN_CRIT "Dead loop on virtual device "
1398 "%s, fix it urgently!\n", dev->name);
1410 /*=======================================================================
1412 =======================================================================*/
1414 int netdev_max_backlog = 300;
1415 int weight_p = 64; /* old backlog weight */
1416 /* These numbers are selected based on intuition and some
1417 * experimentatiom, if you have more scientific way of doing this
1418 * please go ahead and fix things.
1420 int no_cong_thresh = 10;
1425 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1428 #ifdef CONFIG_NET_HW_FLOWCONTROL
1429 atomic_t netdev_dropping = ATOMIC_INIT(0);
1430 static unsigned long netdev_fc_mask = 1;
1431 unsigned long netdev_fc_xoff;
1432 spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1436 void (*stimul)(struct net_device *);
1437 struct net_device *dev;
1438 } netdev_fc_slots[BITS_PER_LONG];
1440 int netdev_register_fc(struct net_device *dev,
1441 void (*stimul)(struct net_device *dev))
1444 unsigned long flags;
1446 spin_lock_irqsave(&netdev_fc_lock, flags);
1447 if (netdev_fc_mask != ~0UL) {
1448 bit = ffz(netdev_fc_mask);
1449 netdev_fc_slots[bit].stimul = stimul;
1450 netdev_fc_slots[bit].dev = dev;
1451 set_bit(bit, &netdev_fc_mask);
1452 clear_bit(bit, &netdev_fc_xoff);
1454 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1458 void netdev_unregister_fc(int bit)
1460 unsigned long flags;
1462 spin_lock_irqsave(&netdev_fc_lock, flags);
1464 netdev_fc_slots[bit].stimul = NULL;
1465 netdev_fc_slots[bit].dev = NULL;
1466 clear_bit(bit, &netdev_fc_mask);
1467 clear_bit(bit, &netdev_fc_xoff);
1469 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1472 static void netdev_wakeup(void)
1476 spin_lock(&netdev_fc_lock);
1477 xoff = netdev_fc_xoff;
1482 netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1484 spin_unlock(&netdev_fc_lock);
1488 static void get_sample_stats(int cpu)
1494 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1495 int blog = sd->input_pkt_queue.qlen;
1496 int avg_blog = sd->avg_blog;
1498 avg_blog = (avg_blog >> 1) + (blog >> 1);
1500 if (avg_blog > mod_cong) {
1501 /* Above moderate congestion levels. */
1502 sd->cng_level = NET_RX_CN_HIGH;
1505 rq = rd % netdev_max_backlog;
1506 if (rq < avg_blog) /* unlucky bastard */
1507 sd->cng_level = NET_RX_DROP;
1509 } else if (avg_blog > lo_cong) {
1510 sd->cng_level = NET_RX_CN_MOD;
1513 rq = rd % netdev_max_backlog;
1514 if (rq < avg_blog) /* unlucky bastard */
1515 sd->cng_level = NET_RX_CN_HIGH;
1517 } else if (avg_blog > no_cong)
1518 sd->cng_level = NET_RX_CN_LOW;
1519 else /* no congestion */
1520 sd->cng_level = NET_RX_SUCCESS;
1522 sd->avg_blog = avg_blog;
1525 #ifdef OFFLINE_SAMPLE
1526 static void sample_queue(unsigned long dummy)
1528 /* 10 ms 0r 1ms -- i don't care -- JHS */
1530 int cpu = smp_processor_id();
1532 get_sample_stats(cpu);
1533 next_tick += jiffies;
1534 mod_timer(&samp_timer, next_tick);
1540 * netif_rx - post buffer to the network code
1541 * @skb: buffer to post
1543 * This function receives a packet from a device driver and queues it for
1544 * the upper (protocol) levels to process. It always succeeds. The buffer
1545 * may be dropped during processing for congestion control or by the
1549 * NET_RX_SUCCESS (no congestion)
1550 * NET_RX_CN_LOW (low congestion)
1551 * NET_RX_CN_MOD (moderate congestion)
1552 * NET_RX_CN_HIGH (high congestion)
1553 * NET_RX_DROP (packet was dropped)
1557 int netif_rx(struct sk_buff *skb)
1560 struct softnet_data *queue;
1561 unsigned long flags;
1563 #ifdef CONFIG_NETPOLL
1564 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1570 if (!skb->stamp.tv_sec)
1571 net_timestamp(&skb->stamp);
1574 * The code is rearranged so that the path is the most
1575 * short when CPU is congested, but is still operating.
1577 local_irq_save(flags);
1578 this_cpu = smp_processor_id();
1579 queue = &__get_cpu_var(softnet_data);
1581 __get_cpu_var(netdev_rx_stat).total++;
1582 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1583 if (queue->input_pkt_queue.qlen) {
1584 if (queue->throttle)
1589 __skb_queue_tail(&queue->input_pkt_queue, skb);
1590 #ifndef OFFLINE_SAMPLE
1591 get_sample_stats(this_cpu);
1593 local_irq_restore(flags);
1594 return queue->cng_level;
1597 if (queue->throttle) {
1598 queue->throttle = 0;
1599 #ifdef CONFIG_NET_HW_FLOWCONTROL
1600 if (atomic_dec_and_test(&netdev_dropping))
1605 netif_rx_schedule(&queue->backlog_dev);
1609 if (!queue->throttle) {
1610 queue->throttle = 1;
1611 __get_cpu_var(netdev_rx_stat).throttled++;
1612 #ifdef CONFIG_NET_HW_FLOWCONTROL
1613 atomic_inc(&netdev_dropping);
1618 __get_cpu_var(netdev_rx_stat).dropped++;
1619 local_irq_restore(flags);
1625 static __inline__ void skb_bond(struct sk_buff *skb)
1627 struct net_device *dev = skb->dev;
1630 skb->real_dev = skb->dev;
1631 skb->dev = dev->master;
1635 static void net_tx_action(struct softirq_action *h)
1637 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1639 if (sd->completion_queue) {
1640 struct sk_buff *clist;
1642 local_irq_disable();
1643 clist = sd->completion_queue;
1644 sd->completion_queue = NULL;
1648 struct sk_buff *skb = clist;
1649 clist = clist->next;
1651 BUG_TRAP(!atomic_read(&skb->users));
1656 if (sd->output_queue) {
1657 struct net_device *head;
1659 local_irq_disable();
1660 head = sd->output_queue;
1661 sd->output_queue = NULL;
1665 struct net_device *dev = head;
1666 head = head->next_sched;
1668 smp_mb__before_clear_bit();
1669 clear_bit(__LINK_STATE_SCHED, &dev->state);
1671 if (spin_trylock(&dev->queue_lock)) {
1673 spin_unlock(&dev->queue_lock);
1675 netif_schedule(dev);
1681 static __inline__ int deliver_skb(struct sk_buff *skb,
1682 struct packet_type *pt_prev, int last)
1684 atomic_inc(&skb->users);
1685 return pt_prev->func(skb, skb->dev, pt_prev);
1689 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1690 int (*br_handle_frame_hook)(struct sk_buff *skb);
1692 static __inline__ int handle_bridge(struct sk_buff *skb,
1693 struct packet_type *pt_prev)
1695 int ret = NET_RX_DROP;
1697 ret = deliver_skb(skb, pt_prev, 0);
1704 static inline int __handle_bridge(struct sk_buff *skb,
1705 struct packet_type **pt_prev, int *ret)
1707 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
1708 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1709 *ret = handle_bridge(skb, *pt_prev);
1710 if (br_handle_frame_hook(skb) == 0)
1720 #ifdef CONFIG_NET_CLS_ACT
1721 /* TODO: Maybe we should just force sch_ingress to be compiled in
1722 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1723 * a compare and 2 stores extra right now if we dont have it on
1724 * but have CONFIG_NET_CLS_ACT
1725 * NOTE: This doesnt stop any functionality; if you dont have
1726 * the ingress scheduler, you just cant add policies on ingress.
1729 int ing_filter(struct sk_buff *skb)
1732 struct net_device *dev = skb->dev;
1733 int result = TC_ACT_OK;
1735 if (dev->qdisc_ingress) {
1736 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1737 if (MAX_RED_LOOP < ttl++) {
1738 printk("Redir loop detected Dropping packet (%s->%s)\n",
1739 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1743 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1745 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1746 if (NULL == skb->input_dev) {
1747 skb->input_dev = skb->dev;
1748 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1750 spin_lock(&dev->ingress_lock);
1751 if ((q = dev->qdisc_ingress) != NULL)
1752 result = q->enqueue(skb, q);
1753 spin_unlock(&dev->ingress_lock);
1761 int netif_receive_skb(struct sk_buff *skb)
1763 struct packet_type *ptype, *pt_prev;
1764 int ret = NET_RX_DROP;
1765 unsigned short type;
1767 #ifdef CONFIG_NETPOLL
1768 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1774 if (!skb->stamp.tv_sec)
1775 net_timestamp(&skb->stamp);
1779 __get_cpu_var(netdev_rx_stat).total++;
1781 skb->h.raw = skb->nh.raw = skb->data;
1782 skb->mac_len = skb->nh.raw - skb->mac.raw;
1785 #ifdef CONFIG_NET_CLS_ACT
1786 if (skb->tc_verd & TC_NCLS) {
1787 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1794 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1795 if (!ptype->dev || ptype->dev == skb->dev) {
1797 ret = deliver_skb(skb, pt_prev, 0);
1802 #ifdef CONFIG_NET_CLS_ACT
1804 atomic_inc(&skb->users);
1805 ret = pt_prev->func(skb, skb->dev, pt_prev);
1806 pt_prev = NULL; /* noone else should process this after*/
1808 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1811 ret = ing_filter(skb);
1813 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1822 handle_diverter(skb);
1824 if (__handle_bridge(skb, &pt_prev, &ret))
1827 type = skb->protocol;
1828 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1829 if (ptype->type == type &&
1830 (!ptype->dev || ptype->dev == skb->dev)) {
1832 ret = deliver_skb(skb, pt_prev, 0);
1838 ret = pt_prev->func(skb, skb->dev, pt_prev);
1841 /* Jamal, now you will not able to escape explaining
1842 * me how you were going to use this. :-)
1852 static int process_backlog(struct net_device *backlog_dev, int *budget)
1855 int quota = min(backlog_dev->quota, *budget);
1856 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1857 unsigned long start_time = jiffies;
1860 struct sk_buff *skb;
1861 struct net_device *dev;
1863 local_irq_disable();
1864 skb = __skb_dequeue(&queue->input_pkt_queue);
1871 netif_receive_skb(skb);
1877 if (work >= quota || jiffies - start_time > 1)
1880 #ifdef CONFIG_NET_HW_FLOWCONTROL
1881 if (queue->throttle &&
1882 queue->input_pkt_queue.qlen < no_cong_thresh ) {
1883 queue->throttle = 0;
1884 if (atomic_dec_and_test(&netdev_dropping)) {
1892 backlog_dev->quota -= work;
1897 backlog_dev->quota -= work;
1900 list_del(&backlog_dev->poll_list);
1901 smp_mb__before_clear_bit();
1902 netif_poll_enable(backlog_dev);
1904 if (queue->throttle) {
1905 queue->throttle = 0;
1906 #ifdef CONFIG_NET_HW_FLOWCONTROL
1907 if (atomic_dec_and_test(&netdev_dropping))
1915 static void net_rx_action(struct softirq_action *h)
1917 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1918 unsigned long start_time = jiffies;
1919 int budget = netdev_max_backlog;
1922 local_irq_disable();
1924 while (!list_empty(&queue->poll_list)) {
1925 struct net_device *dev;
1927 if (budget <= 0 || jiffies - start_time > 1)
1932 dev = list_entry(queue->poll_list.next,
1933 struct net_device, poll_list);
1935 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1936 local_irq_disable();
1937 list_del(&dev->poll_list);
1938 list_add_tail(&dev->poll_list, &queue->poll_list);
1940 dev->quota += dev->weight;
1942 dev->quota = dev->weight;
1945 local_irq_disable();
1953 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1954 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1958 static gifconf_func_t * gifconf_list [NPROTO];
1961 * register_gifconf - register a SIOCGIF handler
1962 * @family: Address family
1963 * @gifconf: Function handler
1965 * Register protocol dependent address dumping routines. The handler
1966 * that is passed must not be freed or reused until it has been replaced
1967 * by another handler.
1969 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1971 if (family >= NPROTO)
1973 gifconf_list[family] = gifconf;
1979 * Map an interface index to its name (SIOCGIFNAME)
1983 * We need this ioctl for efficient implementation of the
1984 * if_indextoname() function required by the IPv6 API. Without
1985 * it, we would have to search all the interfaces to find a
1989 static int dev_ifname(struct ifreq __user *arg)
1991 struct net_device *dev;
1995 * Fetch the caller's info block.
1998 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2001 read_lock(&dev_base_lock);
2002 dev = __dev_get_by_index(ifr.ifr_ifindex);
2004 read_unlock(&dev_base_lock);
2008 strcpy(ifr.ifr_name, dev->name);
2009 read_unlock(&dev_base_lock);
2011 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2017 * Perform a SIOCGIFCONF call. This structure will change
2018 * size eventually, and there is nothing I can do about it.
2019 * Thus we will need a 'compatibility mode'.
2022 static int dev_ifconf(char __user *arg)
2025 struct net_device *dev;
2032 * Fetch the caller's info block.
2035 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2042 * Loop over the interfaces, and write an info block for each.
2046 for (dev = dev_base; dev; dev = dev->next) {
2047 for (i = 0; i < NPROTO; i++) {
2048 if (gifconf_list[i]) {
2051 done = gifconf_list[i](dev, NULL, 0);
2053 done = gifconf_list[i](dev, pos + total,
2063 * All done. Write the updated control block back to the caller.
2065 ifc.ifc_len = total;
2068 * Both BSD and Solaris return 0 here, so we do too.
2070 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2073 #ifdef CONFIG_PROC_FS
2075 * This is invoked by the /proc filesystem handler to display a device
2078 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2080 struct net_device *dev;
2083 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2085 return i == pos ? dev : NULL;
2088 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2090 read_lock(&dev_base_lock);
2091 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2094 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2097 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2100 void dev_seq_stop(struct seq_file *seq, void *v)
2102 read_unlock(&dev_base_lock);
2105 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2107 if (dev->get_stats) {
2108 struct net_device_stats *stats = dev->get_stats(dev);
2110 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2111 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2112 dev->name, stats->rx_bytes, stats->rx_packets,
2114 stats->rx_dropped + stats->rx_missed_errors,
2115 stats->rx_fifo_errors,
2116 stats->rx_length_errors + stats->rx_over_errors +
2117 stats->rx_crc_errors + stats->rx_frame_errors,
2118 stats->rx_compressed, stats->multicast,
2119 stats->tx_bytes, stats->tx_packets,
2120 stats->tx_errors, stats->tx_dropped,
2121 stats->tx_fifo_errors, stats->collisions,
2122 stats->tx_carrier_errors +
2123 stats->tx_aborted_errors +
2124 stats->tx_window_errors +
2125 stats->tx_heartbeat_errors,
2126 stats->tx_compressed);
2128 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2132 * Called from the PROCfs module. This now uses the new arbitrary sized
2133 * /proc/net interface to create /proc/net/dev
2135 static int dev_seq_show(struct seq_file *seq, void *v)
2137 if (v == SEQ_START_TOKEN)
2138 seq_puts(seq, "Inter-| Receive "
2140 " face |bytes packets errs drop fifo frame "
2141 "compressed multicast|bytes packets errs "
2142 "drop fifo colls carrier compressed\n");
2144 dev_seq_printf_stats(seq, v);
2148 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2150 struct netif_rx_stats *rc = NULL;
2152 while (*pos < NR_CPUS)
2153 if (cpu_online(*pos)) {
2154 rc = &per_cpu(netdev_rx_stat, *pos);
2161 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2163 return softnet_get_online(pos);
2166 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2169 return softnet_get_online(pos);
2172 static void softnet_seq_stop(struct seq_file *seq, void *v)
2176 static int softnet_seq_show(struct seq_file *seq, void *v)
2178 struct netif_rx_stats *s = v;
2180 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2181 s->total, s->dropped, s->time_squeeze, s->throttled,
2182 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2183 s->fastroute_deferred_out,
2185 s->fastroute_latency_reduction
2193 static struct seq_operations dev_seq_ops = {
2194 .start = dev_seq_start,
2195 .next = dev_seq_next,
2196 .stop = dev_seq_stop,
2197 .show = dev_seq_show,
2200 static int dev_seq_open(struct inode *inode, struct file *file)
2202 return seq_open(file, &dev_seq_ops);
2205 static struct file_operations dev_seq_fops = {
2206 .owner = THIS_MODULE,
2207 .open = dev_seq_open,
2209 .llseek = seq_lseek,
2210 .release = seq_release,
2213 static struct seq_operations softnet_seq_ops = {
2214 .start = softnet_seq_start,
2215 .next = softnet_seq_next,
2216 .stop = softnet_seq_stop,
2217 .show = softnet_seq_show,
2220 static int softnet_seq_open(struct inode *inode, struct file *file)
2222 return seq_open(file, &softnet_seq_ops);
2225 static struct file_operations softnet_seq_fops = {
2226 .owner = THIS_MODULE,
2227 .open = softnet_seq_open,
2229 .llseek = seq_lseek,
2230 .release = seq_release,
2234 extern int wireless_proc_init(void);
2236 #define wireless_proc_init() 0
2239 static int __init dev_proc_init(void)
2243 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2245 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2247 if (wireless_proc_init())
2253 proc_net_remove("softnet_stat");
2255 proc_net_remove("dev");
2259 #define dev_proc_init() 0
2260 #endif /* CONFIG_PROC_FS */
2264 * netdev_set_master - set up master/slave pair
2265 * @slave: slave device
2266 * @master: new master device
2268 * Changes the master device of the slave. Pass %NULL to break the
2269 * bonding. The caller must hold the RTNL semaphore. On a failure
2270 * a negative errno code is returned. On success the reference counts
2271 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2272 * function returns zero.
2274 int netdev_set_master(struct net_device *slave, struct net_device *master)
2276 struct net_device *old = slave->master;
2286 slave->master = master;
2294 slave->flags |= IFF_SLAVE;
2296 slave->flags &= ~IFF_SLAVE;
2298 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2303 * dev_set_promiscuity - update promiscuity count on a device
2307 * Add or remove promsicuity from a device. While the count in the device
2308 * remains above zero the interface remains promiscuous. Once it hits zero
2309 * the device reverts back to normal filtering operation. A negative inc
2310 * value is used to drop promiscuity on the device.
2312 void dev_set_promiscuity(struct net_device *dev, int inc)
2314 unsigned short old_flags = dev->flags;
2316 dev->flags |= IFF_PROMISC;
2317 if ((dev->promiscuity += inc) == 0)
2318 dev->flags &= ~IFF_PROMISC;
2319 if (dev->flags ^ old_flags) {
2321 printk(KERN_INFO "device %s %s promiscuous mode\n",
2322 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2328 * dev_set_allmulti - update allmulti count on a device
2332 * Add or remove reception of all multicast frames to a device. While the
2333 * count in the device remains above zero the interface remains listening
2334 * to all interfaces. Once it hits zero the device reverts back to normal
2335 * filtering operation. A negative @inc value is used to drop the counter
2336 * when releasing a resource needing all multicasts.
2339 void dev_set_allmulti(struct net_device *dev, int inc)
2341 unsigned short old_flags = dev->flags;
2343 dev->flags |= IFF_ALLMULTI;
2344 if ((dev->allmulti += inc) == 0)
2345 dev->flags &= ~IFF_ALLMULTI;
2346 if (dev->flags ^ old_flags)
2350 unsigned dev_get_flags(const struct net_device *dev)
2354 flags = (dev->flags & ~(IFF_PROMISC |
2357 (dev->gflags & (IFF_PROMISC |
2360 if (netif_running(dev) && netif_carrier_ok(dev))
2361 flags |= IFF_RUNNING;
2366 int dev_change_flags(struct net_device *dev, unsigned flags)
2369 int old_flags = dev->flags;
2372 * Set the flags on our device.
2375 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2376 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2378 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2382 * Load in the correct multicast list now the flags have changed.
2388 * Have we downed the interface. We handle IFF_UP ourselves
2389 * according to user attempts to set it, rather than blindly
2394 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2395 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2401 if (dev->flags & IFF_UP &&
2402 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2404 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2406 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2407 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2408 dev->gflags ^= IFF_PROMISC;
2409 dev_set_promiscuity(dev, inc);
2412 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2413 is important. Some (broken) drivers set IFF_PROMISC, when
2414 IFF_ALLMULTI is requested not asking us and not reporting.
2416 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2417 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2418 dev->gflags ^= IFF_ALLMULTI;
2419 dev_set_allmulti(dev, inc);
2422 if (old_flags ^ dev->flags)
2423 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2428 int dev_set_mtu(struct net_device *dev, int new_mtu)
2432 if (new_mtu == dev->mtu)
2435 /* MTU must be positive. */
2439 if (!netif_device_present(dev))
2443 if (dev->change_mtu)
2444 err = dev->change_mtu(dev, new_mtu);
2447 if (!err && dev->flags & IFF_UP)
2448 notifier_call_chain(&netdev_chain,
2449 NETDEV_CHANGEMTU, dev);
2455 * Perform the SIOCxIFxxx calls.
2457 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2460 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2466 case SIOCGIFFLAGS: /* Get interface flags */
2467 ifr->ifr_flags = dev_get_flags(dev);
2470 case SIOCSIFFLAGS: /* Set interface flags */
2471 return dev_change_flags(dev, ifr->ifr_flags);
2473 case SIOCGIFMETRIC: /* Get the metric on the interface
2474 (currently unused) */
2475 ifr->ifr_metric = 0;
2478 case SIOCSIFMETRIC: /* Set the metric on the interface
2479 (currently unused) */
2482 case SIOCGIFMTU: /* Get the MTU of a device */
2483 ifr->ifr_mtu = dev->mtu;
2486 case SIOCSIFMTU: /* Set the MTU of a device */
2487 return dev_set_mtu(dev, ifr->ifr_mtu);
2490 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2491 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2492 ifr->ifr_hwaddr.sa_family = dev->type;
2496 if (!dev->set_mac_address)
2498 if (ifr->ifr_hwaddr.sa_family != dev->type)
2500 if (!netif_device_present(dev))
2502 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2504 notifier_call_chain(&netdev_chain,
2505 NETDEV_CHANGEADDR, dev);
2508 case SIOCSIFHWBROADCAST:
2509 if (ifr->ifr_hwaddr.sa_family != dev->type)
2511 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2512 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2513 notifier_call_chain(&netdev_chain,
2514 NETDEV_CHANGEADDR, dev);
2518 ifr->ifr_map.mem_start = dev->mem_start;
2519 ifr->ifr_map.mem_end = dev->mem_end;
2520 ifr->ifr_map.base_addr = dev->base_addr;
2521 ifr->ifr_map.irq = dev->irq;
2522 ifr->ifr_map.dma = dev->dma;
2523 ifr->ifr_map.port = dev->if_port;
2527 if (dev->set_config) {
2528 if (!netif_device_present(dev))
2530 return dev->set_config(dev, &ifr->ifr_map);
2535 if (!dev->set_multicast_list ||
2536 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2538 if (!netif_device_present(dev))
2540 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2544 if (!dev->set_multicast_list ||
2545 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2547 if (!netif_device_present(dev))
2549 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2553 ifr->ifr_ifindex = dev->ifindex;
2557 ifr->ifr_qlen = dev->tx_queue_len;
2561 if (ifr->ifr_qlen < 0)
2563 dev->tx_queue_len = ifr->ifr_qlen;
2567 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2568 return dev_change_name(dev, ifr->ifr_newname);
2571 * Unknown or private ioctl
2575 if ((cmd >= SIOCDEVPRIVATE &&
2576 cmd <= SIOCDEVPRIVATE + 15) ||
2577 cmd == SIOCBONDENSLAVE ||
2578 cmd == SIOCBONDRELEASE ||
2579 cmd == SIOCBONDSETHWADDR ||
2580 cmd == SIOCBONDSLAVEINFOQUERY ||
2581 cmd == SIOCBONDINFOQUERY ||
2582 cmd == SIOCBONDCHANGEACTIVE ||
2583 cmd == SIOCGMIIPHY ||
2584 cmd == SIOCGMIIREG ||
2585 cmd == SIOCSMIIREG ||
2586 cmd == SIOCBRADDIF ||
2587 cmd == SIOCBRDELIF ||
2588 cmd == SIOCWANDEV) {
2590 if (dev->do_ioctl) {
2591 if (netif_device_present(dev))
2592 err = dev->do_ioctl(dev, ifr,
2605 * This function handles all "interface"-type I/O control requests. The actual
2606 * 'doing' part of this is dev_ifsioc above.
2610 * dev_ioctl - network device ioctl
2611 * @cmd: command to issue
2612 * @arg: pointer to a struct ifreq in user space
2614 * Issue ioctl functions to devices. This is normally called by the
2615 * user space syscall interfaces but can sometimes be useful for
2616 * other purposes. The return value is the return from the syscall if
2617 * positive or a negative errno code on error.
2620 int dev_ioctl(unsigned int cmd, void __user *arg)
2626 /* One special case: SIOCGIFCONF takes ifconf argument
2627 and requires shared lock, because it sleeps writing
2631 if (cmd == SIOCGIFCONF) {
2633 ret = dev_ifconf((char __user *) arg);
2637 if (cmd == SIOCGIFNAME)
2638 return dev_ifname((struct ifreq __user *)arg);
2640 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2643 ifr.ifr_name[IFNAMSIZ-1] = 0;
2645 colon = strchr(ifr.ifr_name, ':');
2650 * See which interface the caller is talking about.
2655 * These ioctl calls:
2656 * - can be done by all.
2657 * - atomic and do not require locking.
2668 dev_load(ifr.ifr_name);
2669 read_lock(&dev_base_lock);
2670 ret = dev_ifsioc(&ifr, cmd);
2671 read_unlock(&dev_base_lock);
2675 if (copy_to_user(arg, &ifr,
2676 sizeof(struct ifreq)))
2682 dev_load(ifr.ifr_name);
2684 ret = dev_ethtool(&ifr);
2689 if (copy_to_user(arg, &ifr,
2690 sizeof(struct ifreq)))
2696 * These ioctl calls:
2697 * - require superuser power.
2698 * - require strict serialization.
2704 if (!capable(CAP_NET_ADMIN))
2706 dev_load(ifr.ifr_name);
2708 ret = dev_ifsioc(&ifr, cmd);
2713 if (copy_to_user(arg, &ifr,
2714 sizeof(struct ifreq)))
2720 * These ioctl calls:
2721 * - require superuser power.
2722 * - require strict serialization.
2723 * - do not return a value
2733 case SIOCSIFHWBROADCAST:
2736 case SIOCBONDENSLAVE:
2737 case SIOCBONDRELEASE:
2738 case SIOCBONDSETHWADDR:
2739 case SIOCBONDSLAVEINFOQUERY:
2740 case SIOCBONDINFOQUERY:
2741 case SIOCBONDCHANGEACTIVE:
2744 if (!capable(CAP_NET_ADMIN))
2746 dev_load(ifr.ifr_name);
2748 ret = dev_ifsioc(&ifr, cmd);
2753 /* Get the per device memory space. We can add this but
2754 * currently do not support it */
2756 /* Set the per device memory buffer space.
2757 * Not applicable in our case */
2762 * Unknown or private ioctl.
2765 if (cmd == SIOCWANDEV ||
2766 (cmd >= SIOCDEVPRIVATE &&
2767 cmd <= SIOCDEVPRIVATE + 15)) {
2768 dev_load(ifr.ifr_name);
2770 ret = dev_ifsioc(&ifr, cmd);
2772 if (!ret && copy_to_user(arg, &ifr,
2773 sizeof(struct ifreq)))
2778 /* Take care of Wireless Extensions */
2779 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2780 /* If command is `set a parameter', or
2781 * `get the encoding parameters', check if
2782 * the user has the right to do it */
2783 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2784 if (!capable(CAP_NET_ADMIN))
2787 dev_load(ifr.ifr_name);
2789 /* Follow me in net/core/wireless.c */
2790 ret = wireless_process_ioctl(&ifr, cmd);
2792 if (!ret && IW_IS_GET(cmd) &&
2793 copy_to_user(arg, &ifr,
2794 sizeof(struct ifreq)))
2798 #endif /* WIRELESS_EXT */
2805 * dev_new_index - allocate an ifindex
2807 * Returns a suitable unique value for a new device interface
2808 * number. The caller must hold the rtnl semaphore or the
2809 * dev_base_lock to be sure it remains unique.
2811 int dev_new_index(void)
2817 if (!__dev_get_by_index(ifindex))
2822 static int dev_boot_phase = 1;
2824 /* Delayed registration/unregisteration */
2825 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2826 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2828 static inline void net_set_todo(struct net_device *dev)
2830 spin_lock(&net_todo_list_lock);
2831 list_add_tail(&dev->todo_list, &net_todo_list);
2832 spin_unlock(&net_todo_list_lock);
2836 * register_netdevice - register a network device
2837 * @dev: device to register
2839 * Take a completed network device structure and add it to the kernel
2840 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2841 * chain. 0 is returned on success. A negative errno code is returned
2842 * on a failure to set up the device, or if the name is a duplicate.
2844 * Callers must hold the rtnl semaphore. See the comment at the
2845 * end of Space.c for details about the locking. You may want
2846 * register_netdev() instead of this.
2849 * The locking appears insufficient to guarantee two parallel registers
2850 * will not get the same name.
2853 int register_netdevice(struct net_device *dev)
2855 struct hlist_head *head;
2856 struct hlist_node *p;
2859 BUG_ON(dev_boot_phase);
2862 /* When net_device's are persistent, this will be fatal. */
2863 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2865 spin_lock_init(&dev->queue_lock);
2866 spin_lock_init(&dev->xmit_lock);
2867 dev->xmit_lock_owner = -1;
2868 #ifdef CONFIG_NET_CLS_ACT
2869 spin_lock_init(&dev->ingress_lock);
2872 ret = alloc_divert_blk(dev);
2878 /* Init, if this function is available */
2880 ret = dev->init(dev);
2888 if (!dev_valid_name(dev->name)) {
2893 dev->ifindex = dev_new_index();
2894 if (dev->iflink == -1)
2895 dev->iflink = dev->ifindex;
2897 /* Check for existence of name */
2898 head = dev_name_hash(dev->name);
2899 hlist_for_each(p, head) {
2900 struct net_device *d
2901 = hlist_entry(p, struct net_device, name_hlist);
2902 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2908 /* Fix illegal SG+CSUM combinations. */
2909 if ((dev->features & NETIF_F_SG) &&
2910 !(dev->features & (NETIF_F_IP_CSUM |
2912 NETIF_F_HW_CSUM))) {
2913 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2915 dev->features &= ~NETIF_F_SG;
2919 * nil rebuild_header routine,
2920 * that should be never called and used as just bug trap.
2923 if (!dev->rebuild_header)
2924 dev->rebuild_header = default_rebuild_header;
2927 * Default initial state at registry is that the
2928 * device is present.
2931 set_bit(__LINK_STATE_PRESENT, &dev->state);
2934 dev_init_scheduler(dev);
2935 write_lock_bh(&dev_base_lock);
2937 dev_tail = &dev->next;
2938 hlist_add_head(&dev->name_hlist, head);
2939 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2941 dev->reg_state = NETREG_REGISTERING;
2942 write_unlock_bh(&dev_base_lock);
2944 /* Notify protocols, that a new device appeared. */
2945 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2947 /* Finish registration after unlock */
2954 free_divert_blk(dev);
2959 * netdev_wait_allrefs - wait until all references are gone.
2961 * This is called when unregistering network devices.
2963 * Any protocol or device that holds a reference should register
2964 * for netdevice notification, and cleanup and put back the
2965 * reference if they receive an UNREGISTER event.
2966 * We can get stuck here if buggy protocols don't correctly
2969 static void netdev_wait_allrefs(struct net_device *dev)
2971 unsigned long rebroadcast_time, warning_time;
2973 rebroadcast_time = warning_time = jiffies;
2974 while (atomic_read(&dev->refcnt) != 0) {
2975 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2978 /* Rebroadcast unregister notification */
2979 notifier_call_chain(&netdev_chain,
2980 NETDEV_UNREGISTER, dev);
2982 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2984 /* We must not have linkwatch events
2985 * pending on unregister. If this
2986 * happens, we simply run the queue
2987 * unscheduled, resulting in a noop
2990 linkwatch_run_queue();
2995 rebroadcast_time = jiffies;
2998 current->state = TASK_INTERRUPTIBLE;
2999 schedule_timeout(HZ / 4);
3001 if (time_after(jiffies, warning_time + 10 * HZ)) {
3002 printk(KERN_EMERG "unregister_netdevice: "
3003 "waiting for %s to become free. Usage "
3005 dev->name, atomic_read(&dev->refcnt));
3006 warning_time = jiffies;
3015 * register_netdevice(x1);
3016 * register_netdevice(x2);
3018 * unregister_netdevice(y1);
3019 * unregister_netdevice(y2);
3025 * We are invoked by rtnl_unlock() after it drops the semaphore.
3026 * This allows us to deal with problems:
3027 * 1) We can create/delete sysfs objects which invoke hotplug
3028 * without deadlocking with linkwatch via keventd.
3029 * 2) Since we run with the RTNL semaphore not held, we can sleep
3030 * safely in order to wait for the netdev refcnt to drop to zero.
3032 static DECLARE_MUTEX(net_todo_run_mutex);
3033 void netdev_run_todo(void)
3035 struct list_head list = LIST_HEAD_INIT(list);
3039 /* Need to guard against multiple cpu's getting out of order. */
3040 down(&net_todo_run_mutex);
3042 /* Not safe to do outside the semaphore. We must not return
3043 * until all unregister events invoked by the local processor
3044 * have been completed (either by this todo run, or one on
3047 if (list_empty(&net_todo_list))
3050 /* Snapshot list, allow later requests */
3051 spin_lock(&net_todo_list_lock);
3052 list_splice_init(&net_todo_list, &list);
3053 spin_unlock(&net_todo_list_lock);
3055 while (!list_empty(&list)) {
3056 struct net_device *dev
3057 = list_entry(list.next, struct net_device, todo_list);
3058 list_del(&dev->todo_list);
3060 switch(dev->reg_state) {
3061 case NETREG_REGISTERING:
3062 err = netdev_register_sysfs(dev);
3064 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3066 dev->reg_state = NETREG_REGISTERED;
3069 case NETREG_UNREGISTERING:
3070 netdev_unregister_sysfs(dev);
3071 dev->reg_state = NETREG_UNREGISTERED;
3073 netdev_wait_allrefs(dev);
3076 BUG_ON(atomic_read(&dev->refcnt));
3077 BUG_TRAP(!dev->ip_ptr);
3078 BUG_TRAP(!dev->ip6_ptr);
3079 BUG_TRAP(!dev->dn_ptr);
3082 /* It must be the very last action,
3083 * after this 'dev' may point to freed up memory.
3085 if (dev->destructor)
3086 dev->destructor(dev);
3090 printk(KERN_ERR "network todo '%s' but state %d\n",
3091 dev->name, dev->reg_state);
3097 up(&net_todo_run_mutex);
3101 * free_netdev - free network device
3104 * This function does the last stage of destroying an allocated device
3105 * interface. The reference to the device object is released.
3106 * If this is the last reference then it will be freed.
3108 void free_netdev(struct net_device *dev)
3111 /* Compatiablity with error handling in drivers */
3112 if (dev->reg_state == NETREG_UNINITIALIZED) {
3113 kfree((char *)dev - dev->padded);
3117 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3118 dev->reg_state = NETREG_RELEASED;
3120 /* will free via class release */
3121 class_device_put(&dev->class_dev);
3123 kfree((char *)dev - dev->padded);
3127 /* Synchronize with packet receive processing. */
3128 void synchronize_net(void)
3131 synchronize_kernel();
3135 * unregister_netdevice - remove device from the kernel
3138 * This function shuts down a device interface and removes it
3139 * from the kernel tables. On success 0 is returned, on a failure
3140 * a negative errno code is returned.
3142 * Callers must hold the rtnl semaphore. See the comment at the
3143 * end of Space.c for details about the locking. You may want
3144 * unregister_netdev() instead of this.
3147 int unregister_netdevice(struct net_device *dev)
3149 struct net_device *d, **dp;
3151 BUG_ON(dev_boot_phase);
3154 /* Some devices call without registering for initialization unwind. */
3155 if (dev->reg_state == NETREG_UNINITIALIZED) {
3156 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3157 "was registered\n", dev->name, dev);
3161 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3163 /* If device is running, close it first. */
3164 if (dev->flags & IFF_UP)
3167 /* And unlink it from device chain. */
3168 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3170 write_lock_bh(&dev_base_lock);
3171 hlist_del(&dev->name_hlist);
3172 hlist_del(&dev->index_hlist);
3173 if (dev_tail == &dev->next)
3176 write_unlock_bh(&dev_base_lock);
3181 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3186 dev->reg_state = NETREG_UNREGISTERING;
3190 /* Shutdown queueing discipline. */
3194 /* Notify protocols, that we are about to destroy
3195 this device. They should clean all the things.
3197 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3200 * Flush the multicast chain
3202 dev_mc_discard(dev);
3207 /* Notifier chain MUST detach us from master device. */
3208 BUG_TRAP(!dev->master);
3210 free_divert_blk(dev);
3212 /* Finish processing unregister after unlock */
3221 #ifdef CONFIG_HOTPLUG_CPU
3222 static int dev_cpu_callback(struct notifier_block *nfb,
3223 unsigned long action,
3226 struct sk_buff **list_skb;
3227 struct net_device **list_net;
3228 struct sk_buff *skb;
3229 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3230 struct softnet_data *sd, *oldsd;
3232 if (action != CPU_DEAD)
3235 local_irq_disable();
3236 cpu = smp_processor_id();
3237 sd = &per_cpu(softnet_data, cpu);
3238 oldsd = &per_cpu(softnet_data, oldcpu);
3240 /* Find end of our completion_queue. */
3241 list_skb = &sd->completion_queue;
3243 list_skb = &(*list_skb)->next;
3244 /* Append completion queue from offline CPU. */
3245 *list_skb = oldsd->completion_queue;
3246 oldsd->completion_queue = NULL;
3248 /* Find end of our output_queue. */
3249 list_net = &sd->output_queue;
3251 list_net = &(*list_net)->next_sched;
3252 /* Append output queue from offline CPU. */
3253 *list_net = oldsd->output_queue;
3254 oldsd->output_queue = NULL;
3256 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3259 /* Process offline CPU's input_pkt_queue */
3260 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3265 #endif /* CONFIG_HOTPLUG_CPU */
3269 * Initialize the DEV module. At boot time this walks the device list and
3270 * unhooks any devices that fail to initialise (normally hardware not
3271 * present) and leaves us with a valid list of present and active devices.
3276 * This is called single threaded during boot, so no need
3277 * to take the rtnl semaphore.
3279 static int __init net_dev_init(void)
3281 int i, rc = -ENOMEM;
3283 BUG_ON(!dev_boot_phase);
3285 if (dev_proc_init())
3288 if (netdev_sysfs_init())
3291 INIT_LIST_HEAD(&ptype_all);
3292 for (i = 0; i < 16; i++)
3293 INIT_LIST_HEAD(&ptype_base[i]);
3295 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3296 INIT_HLIST_HEAD(&dev_name_head[i]);
3298 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3299 INIT_HLIST_HEAD(&dev_index_head[i]);
3302 * Initialise the packet receive queues.
3305 for (i = 0; i < NR_CPUS; i++) {
3306 struct softnet_data *queue;
3308 queue = &per_cpu(softnet_data, i);
3309 skb_queue_head_init(&queue->input_pkt_queue);
3310 queue->throttle = 0;
3311 queue->cng_level = 0;
3312 queue->avg_blog = 10; /* arbitrary non-zero */
3313 queue->completion_queue = NULL;
3314 INIT_LIST_HEAD(&queue->poll_list);
3315 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3316 queue->backlog_dev.weight = weight_p;
3317 queue->backlog_dev.poll = process_backlog;
3318 atomic_set(&queue->backlog_dev.refcnt, 1);
3321 #ifdef OFFLINE_SAMPLE
3322 samp_timer.expires = jiffies + (10 * HZ);
3323 add_timer(&samp_timer);
3328 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3329 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3331 hotcpu_notifier(dev_cpu_callback, 0);
3339 subsys_initcall(net_dev_init);
3341 EXPORT_SYMBOL(__dev_get);
3342 EXPORT_SYMBOL(__dev_get_by_flags);
3343 EXPORT_SYMBOL(__dev_get_by_index);
3344 EXPORT_SYMBOL(__dev_get_by_name);
3345 EXPORT_SYMBOL(__dev_remove_pack);
3346 EXPORT_SYMBOL(__skb_linearize);
3347 EXPORT_SYMBOL(call_netdevice_notifiers);
3348 EXPORT_SYMBOL(dev_add_pack);
3349 EXPORT_SYMBOL(dev_alloc_name);
3350 EXPORT_SYMBOL(dev_close);
3351 EXPORT_SYMBOL(dev_get_by_flags);
3352 EXPORT_SYMBOL(dev_get_by_index);
3353 EXPORT_SYMBOL(dev_get_by_name);
3354 EXPORT_SYMBOL(dev_getbyhwaddr);
3355 EXPORT_SYMBOL(dev_ioctl);
3356 EXPORT_SYMBOL(dev_new_index);
3357 EXPORT_SYMBOL(dev_open);
3358 EXPORT_SYMBOL(dev_queue_xmit);
3359 EXPORT_SYMBOL(dev_queue_xmit_nit);
3360 EXPORT_SYMBOL(dev_remove_pack);
3361 EXPORT_SYMBOL(dev_set_allmulti);
3362 EXPORT_SYMBOL(dev_set_promiscuity);
3363 EXPORT_SYMBOL(dev_change_flags);
3364 EXPORT_SYMBOL(dev_set_mtu);
3365 EXPORT_SYMBOL(free_netdev);
3366 EXPORT_SYMBOL(netdev_boot_setup_check);
3367 EXPORT_SYMBOL(netdev_set_master);
3368 EXPORT_SYMBOL(netdev_state_change);
3369 EXPORT_SYMBOL(netif_receive_skb);
3370 EXPORT_SYMBOL(netif_rx);
3371 EXPORT_SYMBOL(register_gifconf);
3372 EXPORT_SYMBOL(register_netdevice);
3373 EXPORT_SYMBOL(register_netdevice_notifier);
3374 EXPORT_SYMBOL(skb_checksum_help);
3375 EXPORT_SYMBOL(synchronize_net);
3376 EXPORT_SYMBOL(unregister_netdevice);
3377 EXPORT_SYMBOL(unregister_netdevice_notifier);
3379 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3380 EXPORT_SYMBOL(br_handle_frame_hook);
3384 EXPORT_SYMBOL(dev_load);
3386 #ifdef CONFIG_NET_HW_FLOWCONTROL
3387 EXPORT_SYMBOL(netdev_dropping);
3388 EXPORT_SYMBOL(netdev_fc_xoff);
3389 EXPORT_SYMBOL(netdev_register_fc);
3390 EXPORT_SYMBOL(netdev_unregister_fc);
3393 #ifdef CONFIG_NET_CLS_ACT
3394 EXPORT_SYMBOL(ing_filter);
3398 EXPORT_PER_CPU_SYMBOL(softnet_data);