2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <asm/bitops.h>
78 #include <linux/config.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/notifier.h>
93 #include <linux/skbuff.h>
95 #include <linux/rtnetlink.h>
96 #include <linux/proc_fs.h>
97 #include <linux/seq_file.h>
98 #include <linux/stat.h>
99 #include <linux/if_bridge.h>
100 #include <linux/divert.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #ifdef CONFIG_NET_RADIO
112 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
113 #include <net/iw_handler.h>
114 #endif /* CONFIG_NET_RADIO */
115 #include <asm/current.h>
117 /* This define, if set, will randomly drop a packet when congestion
118 * is more than moderate. It helps fairness in the multi-interface
119 * case when one of them is a hog, but it kills performance for the
120 * single interface case so it is off now by default.
124 /* Setting this will sample the queue lengths and thus congestion
125 * via a timer instead of as each packet is received.
127 #undef OFFLINE_SAMPLE
130 * The list of packet types we will receive (as opposed to discard)
131 * and the routines to invoke.
133 * Why 16. Because with 16 the only overlap we get on a hash of the
134 * low nibble of the protocol value is RARP/SNAP/X.25.
136 * NOTE: That is no longer true with the addition of VLAN tags. Not
137 * sure which should go first, but I bet it won't make much
138 * difference if we are running VLANs. The good news is that
139 * this protocol won't be in the list unless compiled in, so
140 * the average user (w/out VLANs) will not be adversly affected.
157 static spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
158 static struct list_head ptype_base[16]; /* 16 way hashed list */
159 static struct list_head ptype_all; /* Taps */
161 #ifdef OFFLINE_SAMPLE
162 static void sample_queue(unsigned long dummy);
163 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
167 * The @dev_base list is protected by @dev_base_lock and the rtln
170 * Pure readers hold dev_base_lock for reading.
172 * Writers must hold the rtnl semaphore while they loop through the
173 * dev_base list, and hold dev_base_lock for writing when they do the
174 * actual updates. This allows pure readers to access the list even
175 * while a writer is preparing to update it.
177 * To put it another way, dev_base_lock is held for writing only to
178 * protect against pure readers; the rtnl semaphore provides the
179 * protection against other writers.
181 * See, for example usages, register_netdevice() and
182 * unregister_netdevice(), which must be called with the rtnl
185 struct net_device *dev_base;
186 struct net_device **dev_tail = &dev_base;
187 rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
189 EXPORT_SYMBOL(dev_base);
190 EXPORT_SYMBOL(dev_base_lock);
192 #define NETDEV_HASHBITS 8
193 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
194 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
196 static inline struct hlist_head *dev_name_hash(const char *name)
198 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
199 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
202 static inline struct hlist_head *dev_index_hash(int ifindex)
204 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
211 static struct notifier_block *netdev_chain;
214 * Device drivers call our routines to queue packets here. We empty the
215 * queue in the local softnet handler.
217 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
220 extern int netdev_sysfs_init(void);
221 extern int netdev_register_sysfs(struct net_device *);
222 extern void netdev_unregister_sysfs(struct net_device *);
224 #define netdev_sysfs_init() (0)
225 #define netdev_register_sysfs(dev) (0)
226 #define netdev_unregister_sysfs(dev) do { } while(0)
230 /*******************************************************************************
232 Protocol management and registration routines
234 *******************************************************************************/
243 * Add a protocol ID to the list. Now that the input handler is
244 * smarter we can dispense with all the messy stuff that used to be
247 * BEWARE!!! Protocol handlers, mangling input packets,
248 * MUST BE last in hash buckets and checking protocol handlers
249 * MUST start from promiscuous ptype_all chain in net_bh.
250 * It is true now, do not change it.
251 * Explanation follows: if protocol handler, mangling packet, will
252 * be the first on list, it is not able to sense, that packet
253 * is cloned and should be copied-on-write, so that it will
254 * change it and subsequent readers will get broken packet.
259 * dev_add_pack - add packet handler
260 * @pt: packet type declaration
262 * Add a protocol handler to the networking stack. The passed &packet_type
263 * is linked into kernel lists and may not be freed until it has been
264 * removed from the kernel lists.
266 * This call does not sleep therefore it can not
267 * guarantee all CPU's that are in middle of receiving packets
268 * will see the new packet type (until the next received packet).
271 void dev_add_pack(struct packet_type *pt)
275 spin_lock_bh(&ptype_lock);
276 if (pt->type == htons(ETH_P_ALL)) {
278 list_add_rcu(&pt->list, &ptype_all);
280 hash = ntohs(pt->type) & 15;
281 list_add_rcu(&pt->list, &ptype_base[hash]);
283 spin_unlock_bh(&ptype_lock);
286 extern void linkwatch_run_queue(void);
291 * __dev_remove_pack - remove packet handler
292 * @pt: packet type declaration
294 * Remove a protocol handler that was previously added to the kernel
295 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
296 * from the kernel lists and can be freed or reused once this function
299 * The packet type might still be in use by receivers
300 * and must not be freed until after all the CPU's have gone
301 * through a quiescent state.
303 void __dev_remove_pack(struct packet_type *pt)
305 struct list_head *head;
306 struct packet_type *pt1;
308 spin_lock_bh(&ptype_lock);
310 if (pt->type == htons(ETH_P_ALL)) {
314 head = &ptype_base[ntohs(pt->type) & 15];
316 list_for_each_entry(pt1, head, list) {
318 list_del_rcu(&pt->list);
323 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
325 spin_unlock_bh(&ptype_lock);
328 * dev_remove_pack - remove packet handler
329 * @pt: packet type declaration
331 * Remove a protocol handler that was previously added to the kernel
332 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
333 * from the kernel lists and can be freed or reused once this function
336 * This call sleeps to guarantee that no CPU is looking at the packet
339 void dev_remove_pack(struct packet_type *pt)
341 __dev_remove_pack(pt);
346 /******************************************************************************
348 Device Boot-time Settings Routines
350 *******************************************************************************/
352 /* Boot time configuration table */
353 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
356 * netdev_boot_setup_add - add new setup entry
357 * @name: name of the device
358 * @map: configured settings for the device
360 * Adds new setup entry to the dev_boot_setup list. The function
361 * returns 0 on error and 1 on success. This is a generic routine to
364 int netdev_boot_setup_add(char *name, struct ifmap *map)
366 struct netdev_boot_setup *s;
370 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
371 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
372 memset(s[i].name, 0, sizeof(s[i].name));
373 strcpy(s[i].name, name);
374 memcpy(&s[i].map, map, sizeof(s[i].map));
379 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
383 * netdev_boot_setup_check - check boot time settings
384 * @dev: the netdevice
386 * Check boot time settings for the device.
387 * The found settings are set for the device to be used
388 * later in the device probing.
389 * Returns 0 if no settings found, 1 if they are.
391 int netdev_boot_setup_check(struct net_device *dev)
393 struct netdev_boot_setup *s = dev_boot_setup;
396 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
397 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
398 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
399 dev->irq = s[i].map.irq;
400 dev->base_addr = s[i].map.base_addr;
401 dev->mem_start = s[i].map.mem_start;
402 dev->mem_end = s[i].map.mem_end;
411 * netdev_boot_base - get address from boot time settings
412 * @prefix: prefix for network device
413 * @unit: id for network device
415 * Check boot time settings for the base address of device.
416 * The found settings are set for the device to be used
417 * later in the device probing.
418 * Returns 0 if no settings found.
420 unsigned long netdev_boot_base(const char *prefix, int unit)
422 const struct netdev_boot_setup *s = dev_boot_setup;
426 sprintf(name, "%s%d", prefix, unit);
429 * If device already registered then return base of 1
430 * to indicate not to probe for this interface
432 if (__dev_get_by_name(name))
435 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
436 if (!strcmp(name, s[i].name))
437 return s[i].map.base_addr;
442 * Saves at boot time configured settings for any netdevice.
444 int __init netdev_boot_setup(char *str)
449 str = get_options(str, ARRAY_SIZE(ints), ints);
454 memset(&map, 0, sizeof(map));
458 map.base_addr = ints[2];
460 map.mem_start = ints[3];
462 map.mem_end = ints[4];
464 /* Add new entry to the list */
465 return netdev_boot_setup_add(str, &map);
468 __setup("netdev=", netdev_boot_setup);
470 /*******************************************************************************
472 Device Interface Subroutines
474 *******************************************************************************/
477 * __dev_get_by_name - find a device by its name
478 * @name: name to find
480 * Find an interface by name. Must be called under RTNL semaphore
481 * or @dev_base_lock. If the name is found a pointer to the device
482 * is returned. If the name is not found then %NULL is returned. The
483 * reference counters are not incremented so the caller must be
484 * careful with locks.
487 struct net_device *__dev_get_by_name(const char *name)
489 struct hlist_node *p;
491 hlist_for_each(p, dev_name_hash(name)) {
492 struct net_device *dev
493 = hlist_entry(p, struct net_device, name_hlist);
494 if (!strncmp(dev->name, name, IFNAMSIZ))
501 * dev_get_by_name - find a device by its name
502 * @name: name to find
504 * Find an interface by name. This can be called from any
505 * context and does its own locking. The returned handle has
506 * the usage count incremented and the caller must use dev_put() to
507 * release it when it is no longer needed. %NULL is returned if no
508 * matching device is found.
511 struct net_device *dev_get_by_name(const char *name)
513 struct net_device *dev;
515 read_lock(&dev_base_lock);
516 dev = __dev_get_by_name(name);
519 read_unlock(&dev_base_lock);
524 Return value is changed to int to prevent illegal usage in future.
525 It is still legal to use to check for device existence.
527 User should understand, that the result returned by this function
528 is meaningless, if it was not issued under rtnl semaphore.
532 * dev_get - test if a device exists
533 * @name: name to test for
535 * Test if a name exists. Returns true if the name is found. In order
536 * to be sure the name is not allocated or removed during the test the
537 * caller must hold the rtnl semaphore.
539 * This function exists only for back compatibility with older
542 int __dev_get(const char *name)
544 struct net_device *dev;
546 read_lock(&dev_base_lock);
547 dev = __dev_get_by_name(name);
548 read_unlock(&dev_base_lock);
553 * __dev_get_by_index - find a device by its ifindex
554 * @ifindex: index of device
556 * Search for an interface by index. Returns %NULL if the device
557 * is not found or a pointer to the device. The device has not
558 * had its reference counter increased so the caller must be careful
559 * about locking. The caller must hold either the RTNL semaphore
563 struct net_device *__dev_get_by_index(int ifindex)
565 struct hlist_node *p;
567 hlist_for_each(p, dev_index_hash(ifindex)) {
568 struct net_device *dev
569 = hlist_entry(p, struct net_device, index_hlist);
570 if (dev->ifindex == ifindex)
578 * dev_get_by_index - find a device by its ifindex
579 * @ifindex: index of device
581 * Search for an interface by index. Returns NULL if the device
582 * is not found or a pointer to the device. The device returned has
583 * had a reference added and the pointer is safe until the user calls
584 * dev_put to indicate they have finished with it.
587 struct net_device *dev_get_by_index(int ifindex)
589 struct net_device *dev;
591 read_lock(&dev_base_lock);
592 dev = __dev_get_by_index(ifindex);
595 read_unlock(&dev_base_lock);
600 * dev_getbyhwaddr - find a device by its hardware address
601 * @type: media type of device
602 * @ha: hardware address
604 * Search for an interface by MAC address. Returns NULL if the device
605 * is not found or a pointer to the device. The caller must hold the
606 * rtnl semaphore. The returned device has not had its ref count increased
607 * and the caller must therefore be careful about locking
610 * If the API was consistent this would be __dev_get_by_hwaddr
613 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
615 struct net_device *dev;
619 for (dev = dev_base; dev; dev = dev->next)
620 if (dev->type == type &&
621 !memcmp(dev->dev_addr, ha, dev->addr_len))
626 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
628 struct net_device *dev;
630 for (dev = dev_base; dev; dev = dev->next)
631 if (dev->type == type)
636 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
638 struct net_device *dev_getfirstbyhwtype(unsigned short type)
640 struct net_device *dev;
643 dev = __dev_getfirstbyhwtype(type);
650 EXPORT_SYMBOL(dev_getfirstbyhwtype);
653 * dev_get_by_flags - find any device with given flags
654 * @if_flags: IFF_* values
655 * @mask: bitmask of bits in if_flags to check
657 * Search for any interface with the given flags. Returns NULL if a device
658 * is not found or a pointer to the device. The device returned has
659 * had a reference added and the pointer is safe until the user calls
660 * dev_put to indicate they have finished with it.
663 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
665 struct net_device *dev;
667 read_lock(&dev_base_lock);
668 dev = __dev_get_by_flags(if_flags, mask);
671 read_unlock(&dev_base_lock);
676 * __dev_get_by_flags - find any device with given flags
677 * @if_flags: IFF_* values
678 * @mask: bitmask of bits in if_flags to check
680 * Search for any interface with the given flags. Returns NULL if a device
681 * is not found or a pointer to the device. The caller must hold either
682 * the RTNL semaphore or @dev_base_lock.
685 struct net_device *__dev_get_by_flags(unsigned short if_flags, unsigned short mask)
687 struct net_device *dev;
689 for (dev = dev_base; dev != NULL; dev = dev->next) {
690 if (((dev->flags ^ if_flags) & mask) == 0)
697 * dev_valid_name - check if name is okay for network device
700 * Network device names need to be valid file names to
701 * to allow sysfs to work
703 int dev_valid_name(const char *name)
705 return !(*name == '\0'
706 || !strcmp(name, ".")
707 || !strcmp(name, "..")
708 || strchr(name, '/'));
712 * dev_alloc_name - allocate a name for a device
714 * @name: name format string
716 * Passed a format string - eg "lt%d" it will try and find a suitable
717 * id. Not efficient for many devices, not called a lot. The caller
718 * must hold the dev_base or rtnl lock while allocating the name and
719 * adding the device in order to avoid duplicates. Returns the number
720 * of the unit assigned or a negative errno code.
723 int dev_alloc_name(struct net_device *dev, const char *name)
728 const int max_netdevices = 8*PAGE_SIZE;
730 struct net_device *d;
732 p = strnchr(name, IFNAMSIZ-1, '%');
735 * Verify the string as this thing may have come from
736 * the user. There must be either one "%d" and no other "%"
739 if (p[1] != 'd' || strchr(p + 2, '%'))
742 /* Use one page as a bit array of possible slots */
743 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
747 for (d = dev_base; d; d = d->next) {
748 if (!sscanf(d->name, name, &i))
750 if (i < 0 || i >= max_netdevices)
753 /* avoid cases where sscanf is not exact inverse of printf */
754 snprintf(buf, sizeof(buf), name, i);
755 if (!strncmp(buf, d->name, IFNAMSIZ))
759 i = find_first_zero_bit(inuse, max_netdevices);
760 free_page((unsigned long) inuse);
763 snprintf(buf, sizeof(buf), name, i);
764 if (!__dev_get_by_name(buf)) {
765 strlcpy(dev->name, buf, IFNAMSIZ);
769 /* It is possible to run out of possible slots
770 * when the name is long and there isn't enough space left
771 * for the digits, or if all bits are used.
778 * dev_change_name - change name of a device
780 * @newname: name (or format string) must be at least IFNAMSIZ
782 * Change name of a device, can pass format strings "eth%d".
785 int dev_change_name(struct net_device *dev, char *newname)
791 if (dev->flags & IFF_UP)
794 if (!dev_valid_name(newname))
797 if (strchr(newname, '%')) {
798 err = dev_alloc_name(dev, newname);
801 strcpy(newname, dev->name);
803 else if (__dev_get_by_name(newname))
806 strlcpy(dev->name, newname, IFNAMSIZ);
808 err = class_device_rename(&dev->class_dev, dev->name);
810 hlist_del(&dev->name_hlist);
811 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
812 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
819 struct work_struct work;
820 struct net_device *dev;
824 static char *NetEvtStr[16] = {
843 static void netdev_event_callback(void *data)
845 struct dev_tq *tq = (struct dev_tq *) data;
847 if (0 == rtnl_shlock_nowait()) {
848 notifier_call_chain(&netdev_chain, tq->event, tq->dev);
851 if (tq->dev->reg_state == NETREG_UNREGISTERED) {
852 printk(KERN_WARNING "%s: task for event %s canceled for device unregister\n",
853 __FUNCTION__, NetEvtStr[0xf & tq->event]);
855 schedule_delayed_work(&tq->work, 1);
864 * netdev_event - send an event for a network device
865 * @dev: pointer to net device
868 * Send an event about a network device to clients in the stack.
869 * Can be called from interrupt context.
870 * Currently useful events:
871 * NETDEV_REBOOT - A dial-on-demand device hung up.
873 void netdev_event(struct net_device *dev, int event)
877 if (dev->reg_state == NETREG_UNREGISTERED)
879 tq = kmalloc(sizeof(*tq), GFP_ATOMIC);
882 memset(tq, 0, sizeof(*tq));
886 INIT_WORK(&tq->work, netdev_event_callback, tq);
887 schedule_work(&tq->work);
891 * netdev_state_change - device changes state
892 * @dev: device to cause notification
894 * Called to indicate a device has changed state. This function calls
895 * the notifier chains for netdev_chain and sends a NEWLINK message
896 * to the routing socket.
898 void netdev_state_change(struct net_device *dev)
900 if (dev->flags & IFF_UP) {
901 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
902 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
907 * dev_load - load a network module
908 * @name: name of interface
910 * If a network interface is not present and the process has suitable
911 * privileges this function loads the module. If module loading is not
912 * available in this kernel then it becomes a nop.
915 void dev_load(const char *name)
917 struct net_device *dev;
919 read_lock(&dev_base_lock);
920 dev = __dev_get_by_name(name);
921 read_unlock(&dev_base_lock);
923 if (!dev && capable(CAP_SYS_MODULE))
924 request_module("%s", name);
927 static int default_rebuild_header(struct sk_buff *skb)
929 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
930 skb->dev ? skb->dev->name : "NULL!!!");
937 * Some old buggy device drivers change get_stats after registering
938 * the device. Try and trap them here.
939 * This can be elimnated when all devices are known fixed.
941 static inline int get_stats_changed(struct net_device *dev)
943 int changed = dev->last_stats != dev->get_stats;
944 dev->last_stats = dev->get_stats;
949 * dev_open - prepare an interface for use.
950 * @dev: device to open
952 * Takes a device from down to up state. The device's private open
953 * function is invoked and then the multicast lists are loaded. Finally
954 * the device is moved into the up state and a %NETDEV_UP message is
955 * sent to the netdev notifier chain.
957 * Calling this function on an active interface is a nop. On a failure
958 * a negative errno code is returned.
960 int dev_open(struct net_device *dev)
968 if (dev->flags & IFF_UP)
972 * Check for broken device drivers.
974 if (get_stats_changed(dev) && net_ratelimit()) {
975 printk(KERN_ERR "%s: driver changed get_stats after register\n",
980 * Is it even present?
982 if (!netif_device_present(dev))
986 * Call device private open method
988 set_bit(__LINK_STATE_START, &dev->state);
990 ret = dev->open(dev);
992 clear_bit(__LINK_STATE_START, &dev->state);
996 * Check for more broken device drivers.
998 if (get_stats_changed(dev) && net_ratelimit()) {
999 printk(KERN_ERR "%s: driver changed get_stats in open\n",
1004 * If it went open OK then:
1011 dev->flags |= IFF_UP;
1014 * Initialize multicasting status
1019 * Wakeup transmit queue engine
1024 * ... and announce new interface.
1026 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
1032 * dev_close - shutdown an interface.
1033 * @dev: device to shutdown
1035 * This function moves an active device into down state. A
1036 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1037 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1040 int dev_close(struct net_device *dev)
1042 if (!(dev->flags & IFF_UP))
1046 * Tell people we are going down, so that they can
1047 * prepare to death, when device is still operating.
1049 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
1051 dev_deactivate(dev);
1053 clear_bit(__LINK_STATE_START, &dev->state);
1055 /* Synchronize to scheduled poll. We cannot touch poll list,
1056 * it can be even on different cpu. So just clear netif_running(),
1057 * and wait when poll really will happen. Actually, the best place
1058 * for this is inside dev->stop() after device stopped its irq
1059 * engine, but this requires more changes in devices. */
1061 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1062 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
1064 current->state = TASK_INTERRUPTIBLE;
1065 schedule_timeout(1);
1069 * Call the device specific close. This cannot fail.
1070 * Only if device is UP
1072 * We allow it to be called even after a DETACH hot-plug
1079 * Device is now down.
1082 dev->flags &= ~IFF_UP;
1085 * Tell people we are down
1087 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1094 * Device change register/unregister. These are not inline or static
1095 * as we export them to the world.
1099 * register_netdevice_notifier - register a network notifier block
1102 * Register a notifier to be called when network device events occur.
1103 * The notifier passed is linked into the kernel structures and must
1104 * not be reused until it has been unregistered. A negative errno code
1105 * is returned on a failure.
1107 * When registered all registration and up events are replayed
1108 * to the new notifier to allow device to have a race free
1109 * view of the network device list.
1112 int register_netdevice_notifier(struct notifier_block *nb)
1114 struct net_device *dev;
1118 err = notifier_chain_register(&netdev_chain, nb);
1120 for (dev = dev_base; dev; dev = dev->next) {
1121 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1123 if (dev->flags & IFF_UP)
1124 nb->notifier_call(nb, NETDEV_UP, dev);
1132 * unregister_netdevice_notifier - unregister a network notifier block
1135 * Unregister a notifier previously registered by
1136 * register_netdevice_notifier(). The notifier is unlinked into the
1137 * kernel structures and may then be reused. A negative errno code
1138 * is returned on a failure.
1141 int unregister_netdevice_notifier(struct notifier_block *nb)
1143 return notifier_chain_unregister(&netdev_chain, nb);
1147 * call_netdevice_notifiers - call all network notifier blocks
1148 * @val: value passed unmodified to notifier function
1149 * @v: pointer passed unmodified to notifier function
1151 * Call all network notifier blocks. Parameters and return value
1152 * are as for notifier_call_chain().
1155 int call_netdevice_notifiers(unsigned long val, void *v)
1157 return notifier_call_chain(&netdev_chain, val, v);
1161 * Support routine. Sends outgoing frames to any network
1162 * taps currently in use.
1165 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1167 struct packet_type *ptype;
1168 net_timestamp(&skb->stamp);
1171 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1172 /* Never send packets back to the socket
1173 * they originated from - MvS (miquels@drinkel.ow.org)
1175 if ((ptype->dev == dev || !ptype->dev) &&
1176 (ptype->af_packet_priv == NULL ||
1177 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1178 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1182 /* skb->nh should be correctly
1183 set by sender, so that the second statement is
1184 just protection against buggy protocols.
1186 skb2->mac.raw = skb2->data;
1188 if (skb2->nh.raw < skb2->data ||
1189 skb2->nh.raw > skb2->tail) {
1190 if (net_ratelimit())
1191 printk(KERN_CRIT "protocol %04x is "
1193 skb2->protocol, dev->name);
1194 skb2->nh.raw = skb2->data;
1197 skb2->h.raw = skb2->nh.raw;
1198 skb2->pkt_type = PACKET_OUTGOING;
1199 ptype->func(skb2, skb->dev, ptype);
1206 * Invalidate hardware checksum when packet is to be mangled, and
1207 * complete checksum manually on outgoing path.
1209 int skb_checksum_help(struct sk_buff **pskb, int inward)
1212 int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1215 (*pskb)->ip_summed = CHECKSUM_NONE;
1219 if (skb_shared(*pskb) || skb_cloned(*pskb)) {
1220 struct sk_buff *newskb = skb_copy(*pskb, GFP_ATOMIC);
1226 skb_set_owner_w(newskb, (*pskb)->sk);
1231 if (offset > (int)(*pskb)->len)
1233 csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1235 offset = (*pskb)->tail - (*pskb)->h.raw;
1238 if ((*pskb)->csum + 2 > offset)
1241 *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1242 (*pskb)->ip_summed = CHECKSUM_NONE;
1247 #ifdef CONFIG_HIGHMEM
1248 /* Actually, we should eliminate this check as soon as we know, that:
1249 * 1. IOMMU is present and allows to map all the memory.
1250 * 2. No high memory really exists on this machine.
1253 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1257 if (dev->features & NETIF_F_HIGHDMA)
1260 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1261 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1267 #define illegal_highdma(dev, skb) (0)
1270 extern void skb_release_data(struct sk_buff *);
1272 /* Keep head the same: replace data */
1273 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1278 struct skb_shared_info *ninfo;
1279 int headerlen = skb->data - skb->head;
1280 int expand = (skb->tail + skb->data_len) - skb->end;
1282 if (skb_shared(skb))
1288 size = skb->end - skb->head + expand;
1289 size = SKB_DATA_ALIGN(size);
1290 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1294 /* Copy entire thing */
1295 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1299 ninfo = (struct skb_shared_info*)(data + size);
1300 atomic_set(&ninfo->dataref, 1);
1301 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1302 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1303 ninfo->nr_frags = 0;
1304 ninfo->frag_list = NULL;
1306 /* Offset between the two in bytes */
1307 offset = data - skb->head;
1309 /* Free old data. */
1310 skb_release_data(skb);
1313 skb->end = data + size;
1315 /* Set up new pointers */
1316 skb->h.raw += offset;
1317 skb->nh.raw += offset;
1318 skb->mac.raw += offset;
1319 skb->tail += offset;
1320 skb->data += offset;
1322 /* We are no longer a clone, even if we were. */
1325 skb->tail += skb->data_len;
1330 #define HARD_TX_LOCK_BH(dev, cpu) { \
1331 if ((dev->features & NETIF_F_LLTX) == 0) { \
1332 spin_lock_bh(&dev->xmit_lock); \
1333 dev->xmit_lock_owner = cpu; \
1337 #define HARD_TX_UNLOCK_BH(dev) { \
1338 if ((dev->features & NETIF_F_LLTX) == 0) { \
1339 dev->xmit_lock_owner = -1; \
1340 spin_unlock_bh(&dev->xmit_lock); \
1344 static inline void qdisc_run(struct net_device *dev)
1346 while (!netif_queue_stopped(dev) &&
1347 qdisc_restart(dev)<0)
1352 * dev_queue_xmit - transmit a buffer
1353 * @skb: buffer to transmit
1355 * Queue a buffer for transmission to a network device. The caller must
1356 * have set the device and priority and built the buffer before calling
1357 * this function. The function can be called from an interrupt.
1359 * A negative errno code is returned on a failure. A success does not
1360 * guarantee the frame will be transmitted as it may be dropped due
1361 * to congestion or traffic shaping.
1364 int dev_queue_xmit(struct sk_buff *skb)
1366 struct net_device *dev = skb->dev;
1370 if (skb_shinfo(skb)->frag_list &&
1371 !(dev->features & NETIF_F_FRAGLIST) &&
1372 __skb_linearize(skb, GFP_ATOMIC))
1375 /* Fragmented skb is linearized if device does not support SG,
1376 * or if at least one of fragments is in highmem and device
1377 * does not support DMA from it.
1379 if (skb_shinfo(skb)->nr_frags &&
1380 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1381 __skb_linearize(skb, GFP_ATOMIC))
1384 /* If packet is not checksummed and device does not support
1385 * checksumming for this protocol, complete checksumming here.
1387 if (skb->ip_summed == CHECKSUM_HW &&
1388 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1389 (!(dev->features & NETIF_F_IP_CSUM) ||
1390 skb->protocol != htons(ETH_P_IP))))
1391 if (skb_checksum_help(&skb, 0))
1395 /* Updates of qdisc are serialized by queue_lock.
1396 * The struct Qdisc which is pointed to by qdisc is now a
1397 * rcu structure - it may be accessed without acquiring
1398 * a lock (but the structure may be stale.) The freeing of the
1399 * qdisc will be deferred until it's known that there are no
1400 * more references to it.
1402 * If the qdisc has an enqueue function, we still need to
1403 * hold the queue_lock before calling it, since queue_lock
1404 * also serializes access to the device queue.
1408 smp_read_barrier_depends();
1409 #ifdef CONFIG_NET_CLS_ACT
1410 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1413 /* Grab device queue */
1414 spin_lock_bh(&dev->queue_lock);
1416 rc = q->enqueue(skb, q);
1420 spin_unlock_bh(&dev->queue_lock);
1422 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1427 /* The device has no queue. Common case for software devices:
1428 loopback, all the sorts of tunnels...
1430 Really, it is unlikely that xmit_lock protection is necessary here.
1431 (f.e. loopback and IP tunnels are clean ignoring statistics
1433 However, it is possible, that they rely on protection
1436 Check this and shot the lock. It is not prone from deadlocks.
1437 Either shot noqueue qdisc, it is even simpler 8)
1439 if (dev->flags & IFF_UP) {
1440 int cpu = get_cpu();
1442 if (dev->xmit_lock_owner != cpu) {
1444 HARD_TX_LOCK_BH(dev, cpu);
1447 if (!netif_queue_stopped(dev)) {
1449 dev_queue_xmit_nit(skb, dev);
1452 if (!dev->hard_start_xmit(skb, dev)) {
1453 HARD_TX_UNLOCK_BH(dev);
1457 HARD_TX_UNLOCK_BH(dev);
1458 if (net_ratelimit())
1459 printk(KERN_CRIT "Virtual device %s asks to "
1460 "queue packet!\n", dev->name);
1464 /* Recursion is detected! It is possible,
1466 if (net_ratelimit())
1467 printk(KERN_CRIT "Dead loop on virtual device "
1468 "%s, fix it urgently!\n", dev->name);
1480 /*=======================================================================
1482 =======================================================================*/
1484 int netdev_max_backlog = 300;
1485 int weight_p = 64; /* old backlog weight */
1486 /* These numbers are selected based on intuition and some
1487 * experimentatiom, if you have more scientific way of doing this
1488 * please go ahead and fix things.
1490 int no_cong_thresh = 10;
1495 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1498 #ifdef CONFIG_NET_HW_FLOWCONTROL
1499 atomic_t netdev_dropping = ATOMIC_INIT(0);
1500 static unsigned long netdev_fc_mask = 1;
1501 unsigned long netdev_fc_xoff;
1502 spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1506 void (*stimul)(struct net_device *);
1507 struct net_device *dev;
1508 } netdev_fc_slots[BITS_PER_LONG];
1510 int netdev_register_fc(struct net_device *dev,
1511 void (*stimul)(struct net_device *dev))
1514 unsigned long flags;
1516 spin_lock_irqsave(&netdev_fc_lock, flags);
1517 if (netdev_fc_mask != ~0UL) {
1518 bit = ffz(netdev_fc_mask);
1519 netdev_fc_slots[bit].stimul = stimul;
1520 netdev_fc_slots[bit].dev = dev;
1521 set_bit(bit, &netdev_fc_mask);
1522 clear_bit(bit, &netdev_fc_xoff);
1524 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1528 void netdev_unregister_fc(int bit)
1530 unsigned long flags;
1532 spin_lock_irqsave(&netdev_fc_lock, flags);
1534 netdev_fc_slots[bit].stimul = NULL;
1535 netdev_fc_slots[bit].dev = NULL;
1536 clear_bit(bit, &netdev_fc_mask);
1537 clear_bit(bit, &netdev_fc_xoff);
1539 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1542 static void netdev_wakeup(void)
1546 spin_lock(&netdev_fc_lock);
1547 xoff = netdev_fc_xoff;
1552 netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1554 spin_unlock(&netdev_fc_lock);
1558 static void get_sample_stats(int cpu)
1564 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1565 int blog = sd->input_pkt_queue.qlen;
1566 int avg_blog = sd->avg_blog;
1568 avg_blog = (avg_blog >> 1) + (blog >> 1);
1570 if (avg_blog > mod_cong) {
1571 /* Above moderate congestion levels. */
1572 sd->cng_level = NET_RX_CN_HIGH;
1575 rq = rd % netdev_max_backlog;
1576 if (rq < avg_blog) /* unlucky bastard */
1577 sd->cng_level = NET_RX_DROP;
1579 } else if (avg_blog > lo_cong) {
1580 sd->cng_level = NET_RX_CN_MOD;
1583 rq = rd % netdev_max_backlog;
1584 if (rq < avg_blog) /* unlucky bastard */
1585 sd->cng_level = NET_RX_CN_HIGH;
1587 } else if (avg_blog > no_cong)
1588 sd->cng_level = NET_RX_CN_LOW;
1589 else /* no congestion */
1590 sd->cng_level = NET_RX_SUCCESS;
1592 sd->avg_blog = avg_blog;
1595 #ifdef OFFLINE_SAMPLE
1596 static void sample_queue(unsigned long dummy)
1598 /* 10 ms 0r 1ms -- i don't care -- JHS */
1600 int cpu = smp_processor_id();
1602 get_sample_stats(cpu);
1603 next_tick += jiffies;
1604 mod_timer(&samp_timer, next_tick);
1610 * netif_rx - post buffer to the network code
1611 * @skb: buffer to post
1613 * This function receives a packet from a device driver and queues it for
1614 * the upper (protocol) levels to process. It always succeeds. The buffer
1615 * may be dropped during processing for congestion control or by the
1619 * NET_RX_SUCCESS (no congestion)
1620 * NET_RX_CN_LOW (low congestion)
1621 * NET_RX_CN_MOD (moderate congestion)
1622 * NET_RX_CN_HIGH (high congestion)
1623 * NET_RX_DROP (packet was dropped)
1627 int netif_rx(struct sk_buff *skb)
1630 struct softnet_data *queue;
1631 unsigned long flags;
1633 #ifdef CONFIG_NETPOLL_RX
1634 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1640 if (!skb->stamp.tv_sec)
1641 net_timestamp(&skb->stamp);
1644 * The code is rearranged so that the path is the most
1645 * short when CPU is congested, but is still operating.
1647 local_irq_save(flags);
1648 this_cpu = smp_processor_id();
1649 queue = &__get_cpu_var(softnet_data);
1651 __get_cpu_var(netdev_rx_stat).total++;
1652 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1653 if (queue->input_pkt_queue.qlen) {
1654 if (queue->throttle)
1659 __skb_queue_tail(&queue->input_pkt_queue, skb);
1660 #ifndef OFFLINE_SAMPLE
1661 get_sample_stats(this_cpu);
1663 local_irq_restore(flags);
1664 return queue->cng_level;
1667 if (queue->throttle) {
1668 queue->throttle = 0;
1669 #ifdef CONFIG_NET_HW_FLOWCONTROL
1670 if (atomic_dec_and_test(&netdev_dropping))
1675 netif_rx_schedule(&queue->backlog_dev);
1679 if (!queue->throttle) {
1680 queue->throttle = 1;
1681 __get_cpu_var(netdev_rx_stat).throttled++;
1682 #ifdef CONFIG_NET_HW_FLOWCONTROL
1683 atomic_inc(&netdev_dropping);
1688 __get_cpu_var(netdev_rx_stat).dropped++;
1689 local_irq_restore(flags);
1695 static __inline__ void skb_bond(struct sk_buff *skb)
1697 struct net_device *dev = skb->dev;
1700 skb->real_dev = skb->dev;
1701 skb->dev = dev->master;
1705 static void net_tx_action(struct softirq_action *h)
1707 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1709 if (sd->completion_queue) {
1710 struct sk_buff *clist;
1712 local_irq_disable();
1713 clist = sd->completion_queue;
1714 sd->completion_queue = NULL;
1718 struct sk_buff *skb = clist;
1719 clist = clist->next;
1721 BUG_TRAP(!atomic_read(&skb->users));
1726 if (sd->output_queue) {
1727 struct net_device *head;
1729 local_irq_disable();
1730 head = sd->output_queue;
1731 sd->output_queue = NULL;
1735 struct net_device *dev = head;
1736 head = head->next_sched;
1738 smp_mb__before_clear_bit();
1739 clear_bit(__LINK_STATE_SCHED, &dev->state);
1741 if (spin_trylock(&dev->queue_lock)) {
1743 spin_unlock(&dev->queue_lock);
1745 netif_schedule(dev);
1751 static __inline__ int deliver_skb(struct sk_buff *skb,
1752 struct packet_type *pt_prev, int last)
1754 atomic_inc(&skb->users);
1755 return pt_prev->func(skb, skb->dev, pt_prev);
1759 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1760 int (*br_handle_frame_hook)(struct sk_buff *skb);
1762 static __inline__ int handle_bridge(struct sk_buff *skb,
1763 struct packet_type *pt_prev)
1765 int ret = NET_RX_DROP;
1767 ret = deliver_skb(skb, pt_prev, 0);
1774 static inline int __handle_bridge(struct sk_buff *skb,
1775 struct packet_type **pt_prev, int *ret)
1777 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
1778 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1779 *ret = handle_bridge(skb, *pt_prev);
1780 if (br_handle_frame_hook(skb) == 0)
1790 #ifdef CONFIG_NET_CLS_ACT
1791 /* TODO: Maybe we should just force sch_ingress to be compiled in
1792 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1793 * a compare and 2 stores extra right now if we dont have it on
1794 * but have CONFIG_NET_CLS_ACT
1795 * NOTE: This doesnt stop any functionality; if you dont have
1796 * the ingress scheduler, you just cant add policies on ingress.
1799 int ing_filter(struct sk_buff *skb)
1802 struct net_device *dev = skb->dev;
1803 int result = TC_ACT_OK;
1805 if (dev->qdisc_ingress) {
1806 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1807 if (MAX_RED_LOOP < ttl++) {
1808 printk("Redir loop detected Dropping packet (%s->%s)\n",
1809 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1813 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1815 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1816 if (NULL == skb->input_dev) {
1817 skb->input_dev = skb->dev;
1818 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1820 spin_lock(&dev->ingress_lock);
1821 if ((q = dev->qdisc_ingress) != NULL)
1822 result = q->enqueue(skb, q);
1823 spin_unlock(&dev->ingress_lock);
1831 int netif_receive_skb(struct sk_buff *skb)
1833 struct packet_type *ptype, *pt_prev;
1834 int ret = NET_RX_DROP;
1835 unsigned short type;
1837 #ifdef CONFIG_NETPOLL_RX
1838 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1844 if (!skb->stamp.tv_sec)
1845 net_timestamp(&skb->stamp);
1849 __get_cpu_var(netdev_rx_stat).total++;
1851 skb->h.raw = skb->nh.raw = skb->data;
1852 skb->mac_len = skb->nh.raw - skb->mac.raw;
1855 #ifdef CONFIG_NET_CLS_ACT
1856 if (skb->tc_verd & TC_NCLS) {
1857 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1864 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1865 if (!ptype->dev || ptype->dev == skb->dev) {
1867 ret = deliver_skb(skb, pt_prev, 0);
1872 #ifdef CONFIG_NET_CLS_ACT
1874 atomic_inc(&skb->users);
1875 ret = pt_prev->func(skb, skb->dev, pt_prev);
1876 pt_prev = NULL; /* noone else should process this after*/
1878 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1881 ret = ing_filter(skb);
1883 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1892 handle_diverter(skb);
1894 if (__handle_bridge(skb, &pt_prev, &ret))
1897 type = skb->protocol;
1898 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1899 if (ptype->type == type &&
1900 (!ptype->dev || ptype->dev == skb->dev)) {
1902 ret = deliver_skb(skb, pt_prev, 0);
1908 ret = pt_prev->func(skb, skb->dev, pt_prev);
1911 /* Jamal, now you will not able to escape explaining
1912 * me how you were going to use this. :-)
1922 static int process_backlog(struct net_device *backlog_dev, int *budget)
1925 int quota = min(backlog_dev->quota, *budget);
1926 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1927 unsigned long start_time = jiffies;
1930 struct sk_buff *skb;
1931 struct net_device *dev;
1933 local_irq_disable();
1934 skb = __skb_dequeue(&queue->input_pkt_queue);
1941 netif_receive_skb(skb);
1947 if (work >= quota || jiffies - start_time > 1)
1950 #ifdef CONFIG_NET_HW_FLOWCONTROL
1951 if (queue->throttle &&
1952 queue->input_pkt_queue.qlen < no_cong_thresh ) {
1953 queue->throttle = 0;
1954 if (atomic_dec_and_test(&netdev_dropping)) {
1962 backlog_dev->quota -= work;
1967 backlog_dev->quota -= work;
1970 list_del(&backlog_dev->poll_list);
1971 smp_mb__before_clear_bit();
1972 netif_poll_enable(backlog_dev);
1974 if (queue->throttle) {
1975 queue->throttle = 0;
1976 #ifdef CONFIG_NET_HW_FLOWCONTROL
1977 if (atomic_dec_and_test(&netdev_dropping))
1985 static void net_rx_action(struct softirq_action *h)
1987 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1988 unsigned long start_time = jiffies;
1989 int budget = netdev_max_backlog;
1992 local_irq_disable();
1994 while (!list_empty(&queue->poll_list)) {
1995 struct net_device *dev;
1997 if (budget <= 0 || jiffies - start_time > 1)
2002 dev = list_entry(queue->poll_list.next,
2003 struct net_device, poll_list);
2005 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
2006 local_irq_disable();
2007 list_del(&dev->poll_list);
2008 list_add_tail(&dev->poll_list, &queue->poll_list);
2010 dev->quota += dev->weight;
2012 dev->quota = dev->weight;
2015 local_irq_disable();
2023 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2024 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2028 static gifconf_func_t * gifconf_list [NPROTO];
2031 * register_gifconf - register a SIOCGIF handler
2032 * @family: Address family
2033 * @gifconf: Function handler
2035 * Register protocol dependent address dumping routines. The handler
2036 * that is passed must not be freed or reused until it has been replaced
2037 * by another handler.
2039 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2041 if (family >= NPROTO)
2043 gifconf_list[family] = gifconf;
2049 * Map an interface index to its name (SIOCGIFNAME)
2053 * We need this ioctl for efficient implementation of the
2054 * if_indextoname() function required by the IPv6 API. Without
2055 * it, we would have to search all the interfaces to find a
2059 static int dev_ifname(struct ifreq __user *arg)
2061 struct net_device *dev;
2065 * Fetch the caller's info block.
2068 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2071 read_lock(&dev_base_lock);
2072 dev = __dev_get_by_index(ifr.ifr_ifindex);
2074 read_unlock(&dev_base_lock);
2078 strcpy(ifr.ifr_name, dev->name);
2079 read_unlock(&dev_base_lock);
2081 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2087 * Perform a SIOCGIFCONF call. This structure will change
2088 * size eventually, and there is nothing I can do about it.
2089 * Thus we will need a 'compatibility mode'.
2092 static int dev_ifconf(char __user *arg)
2095 struct net_device *dev;
2102 * Fetch the caller's info block.
2105 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2112 * Loop over the interfaces, and write an info block for each.
2116 for (dev = dev_base; dev; dev = dev->next) {
2117 for (i = 0; i < NPROTO; i++) {
2118 if (gifconf_list[i]) {
2121 done = gifconf_list[i](dev, NULL, 0);
2123 done = gifconf_list[i](dev, pos + total,
2133 * All done. Write the updated control block back to the caller.
2135 ifc.ifc_len = total;
2138 * Both BSD and Solaris return 0 here, so we do too.
2140 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2143 #ifdef CONFIG_PROC_FS
2145 * This is invoked by the /proc filesystem handler to display a device
2148 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2150 struct net_device *dev;
2153 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2155 return i == pos ? dev : NULL;
2158 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2160 read_lock(&dev_base_lock);
2161 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2164 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2167 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2170 void dev_seq_stop(struct seq_file *seq, void *v)
2172 read_unlock(&dev_base_lock);
2175 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2177 if (dev->get_stats) {
2178 struct net_device_stats *stats = dev->get_stats(dev);
2180 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2181 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2182 dev->name, stats->rx_bytes, stats->rx_packets,
2184 stats->rx_dropped + stats->rx_missed_errors,
2185 stats->rx_fifo_errors,
2186 stats->rx_length_errors + stats->rx_over_errors +
2187 stats->rx_crc_errors + stats->rx_frame_errors,
2188 stats->rx_compressed, stats->multicast,
2189 stats->tx_bytes, stats->tx_packets,
2190 stats->tx_errors, stats->tx_dropped,
2191 stats->tx_fifo_errors, stats->collisions,
2192 stats->tx_carrier_errors +
2193 stats->tx_aborted_errors +
2194 stats->tx_window_errors +
2195 stats->tx_heartbeat_errors,
2196 stats->tx_compressed);
2198 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2202 * Called from the PROCfs module. This now uses the new arbitrary sized
2203 * /proc/net interface to create /proc/net/dev
2205 static int dev_seq_show(struct seq_file *seq, void *v)
2207 if (v == SEQ_START_TOKEN)
2208 seq_puts(seq, "Inter-| Receive "
2210 " face |bytes packets errs drop fifo frame "
2211 "compressed multicast|bytes packets errs "
2212 "drop fifo colls carrier compressed\n");
2214 dev_seq_printf_stats(seq, v);
2218 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2220 struct netif_rx_stats *rc = NULL;
2222 while (*pos < NR_CPUS)
2223 if (cpu_online(*pos)) {
2224 rc = &per_cpu(netdev_rx_stat, *pos);
2231 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2233 return softnet_get_online(pos);
2236 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2239 return softnet_get_online(pos);
2242 static void softnet_seq_stop(struct seq_file *seq, void *v)
2246 static int softnet_seq_show(struct seq_file *seq, void *v)
2248 struct netif_rx_stats *s = v;
2250 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2251 s->total, s->dropped, s->time_squeeze, s->throttled,
2252 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2253 s->fastroute_deferred_out,
2255 s->fastroute_latency_reduction
2263 static struct seq_operations dev_seq_ops = {
2264 .start = dev_seq_start,
2265 .next = dev_seq_next,
2266 .stop = dev_seq_stop,
2267 .show = dev_seq_show,
2270 static int dev_seq_open(struct inode *inode, struct file *file)
2272 return seq_open(file, &dev_seq_ops);
2275 static struct file_operations dev_seq_fops = {
2276 .owner = THIS_MODULE,
2277 .open = dev_seq_open,
2279 .llseek = seq_lseek,
2280 .release = seq_release,
2283 static struct seq_operations softnet_seq_ops = {
2284 .start = softnet_seq_start,
2285 .next = softnet_seq_next,
2286 .stop = softnet_seq_stop,
2287 .show = softnet_seq_show,
2290 static int softnet_seq_open(struct inode *inode, struct file *file)
2292 return seq_open(file, &softnet_seq_ops);
2295 static struct file_operations softnet_seq_fops = {
2296 .owner = THIS_MODULE,
2297 .open = softnet_seq_open,
2299 .llseek = seq_lseek,
2300 .release = seq_release,
2304 extern int wireless_proc_init(void);
2306 #define wireless_proc_init() 0
2309 static int __init dev_proc_init(void)
2313 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2315 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2317 if (wireless_proc_init())
2323 proc_net_remove("softnet_stat");
2325 proc_net_remove("dev");
2329 #define dev_proc_init() 0
2330 #endif /* CONFIG_PROC_FS */
2334 * netdev_set_master - set up master/slave pair
2335 * @slave: slave device
2336 * @master: new master device
2338 * Changes the master device of the slave. Pass %NULL to break the
2339 * bonding. The caller must hold the RTNL semaphore. On a failure
2340 * a negative errno code is returned. On success the reference counts
2341 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2342 * function returns zero.
2344 int netdev_set_master(struct net_device *slave, struct net_device *master)
2346 struct net_device *old = slave->master;
2356 slave->master = master;
2364 slave->flags |= IFF_SLAVE;
2366 slave->flags &= ~IFF_SLAVE;
2368 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2373 * dev_set_promiscuity - update promiscuity count on a device
2377 * Add or remove promsicuity from a device. While the count in the device
2378 * remains above zero the interface remains promiscuous. Once it hits zero
2379 * the device reverts back to normal filtering operation. A negative inc
2380 * value is used to drop promiscuity on the device.
2382 void dev_set_promiscuity(struct net_device *dev, int inc)
2384 unsigned short old_flags = dev->flags;
2386 dev->flags |= IFF_PROMISC;
2387 if ((dev->promiscuity += inc) == 0)
2388 dev->flags &= ~IFF_PROMISC;
2389 if (dev->flags ^ old_flags) {
2391 printk(KERN_INFO "device %s %s promiscuous mode\n",
2392 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2398 * dev_set_allmulti - update allmulti count on a device
2402 * Add or remove reception of all multicast frames to a device. While the
2403 * count in the device remains above zero the interface remains listening
2404 * to all interfaces. Once it hits zero the device reverts back to normal
2405 * filtering operation. A negative @inc value is used to drop the counter
2406 * when releasing a resource needing all multicasts.
2409 void dev_set_allmulti(struct net_device *dev, int inc)
2411 unsigned short old_flags = dev->flags;
2413 dev->flags |= IFF_ALLMULTI;
2414 if ((dev->allmulti += inc) == 0)
2415 dev->flags &= ~IFF_ALLMULTI;
2416 if (dev->flags ^ old_flags)
2420 unsigned dev_get_flags(const struct net_device *dev)
2424 flags = (dev->flags & ~(IFF_PROMISC |
2427 (dev->gflags & (IFF_PROMISC |
2430 if (netif_running(dev) && netif_carrier_ok(dev))
2431 flags |= IFF_RUNNING;
2436 int dev_change_flags(struct net_device *dev, unsigned flags)
2439 int old_flags = dev->flags;
2442 * Set the flags on our device.
2445 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2446 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2448 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2452 * Load in the correct multicast list now the flags have changed.
2458 * Have we downed the interface. We handle IFF_UP ourselves
2459 * according to user attempts to set it, rather than blindly
2464 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2465 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2471 if (dev->flags & IFF_UP &&
2472 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2474 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2476 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2477 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2478 dev->gflags ^= IFF_PROMISC;
2479 dev_set_promiscuity(dev, inc);
2482 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2483 is important. Some (broken) drivers set IFF_PROMISC, when
2484 IFF_ALLMULTI is requested not asking us and not reporting.
2486 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2487 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2488 dev->gflags ^= IFF_ALLMULTI;
2489 dev_set_allmulti(dev, inc);
2492 if (old_flags ^ dev->flags)
2493 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2498 int dev_set_mtu(struct net_device *dev, int new_mtu)
2502 if (new_mtu == dev->mtu)
2505 /* MTU must be positive. */
2509 if (!netif_device_present(dev))
2513 if (dev->change_mtu)
2514 err = dev->change_mtu(dev, new_mtu);
2517 if (!err && dev->flags & IFF_UP)
2518 notifier_call_chain(&netdev_chain,
2519 NETDEV_CHANGEMTU, dev);
2525 * Perform the SIOCxIFxxx calls.
2527 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2530 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2536 case SIOCGIFFLAGS: /* Get interface flags */
2537 ifr->ifr_flags = dev_get_flags(dev);
2540 case SIOCSIFFLAGS: /* Set interface flags */
2541 return dev_change_flags(dev, ifr->ifr_flags);
2543 case SIOCGIFMETRIC: /* Get the metric on the interface
2544 (currently unused) */
2545 ifr->ifr_metric = 0;
2548 case SIOCSIFMETRIC: /* Set the metric on the interface
2549 (currently unused) */
2552 case SIOCGIFMTU: /* Get the MTU of a device */
2553 ifr->ifr_mtu = dev->mtu;
2556 case SIOCSIFMTU: /* Set the MTU of a device */
2557 return dev_set_mtu(dev, ifr->ifr_mtu);
2560 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2561 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2562 ifr->ifr_hwaddr.sa_family = dev->type;
2566 if (!dev->set_mac_address)
2568 if (ifr->ifr_hwaddr.sa_family != dev->type)
2570 if (!netif_device_present(dev))
2572 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2574 notifier_call_chain(&netdev_chain,
2575 NETDEV_CHANGEADDR, dev);
2578 case SIOCSIFHWBROADCAST:
2579 if (ifr->ifr_hwaddr.sa_family != dev->type)
2581 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2582 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2583 notifier_call_chain(&netdev_chain,
2584 NETDEV_CHANGEADDR, dev);
2588 ifr->ifr_map.mem_start = dev->mem_start;
2589 ifr->ifr_map.mem_end = dev->mem_end;
2590 ifr->ifr_map.base_addr = dev->base_addr;
2591 ifr->ifr_map.irq = dev->irq;
2592 ifr->ifr_map.dma = dev->dma;
2593 ifr->ifr_map.port = dev->if_port;
2597 if (dev->set_config) {
2598 if (!netif_device_present(dev))
2600 return dev->set_config(dev, &ifr->ifr_map);
2605 if (!dev->set_multicast_list ||
2606 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2608 if (!netif_device_present(dev))
2610 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2614 if (!dev->set_multicast_list ||
2615 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2617 if (!netif_device_present(dev))
2619 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2623 ifr->ifr_ifindex = dev->ifindex;
2627 ifr->ifr_qlen = dev->tx_queue_len;
2631 if (ifr->ifr_qlen < 0)
2633 dev->tx_queue_len = ifr->ifr_qlen;
2637 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2638 return dev_change_name(dev, ifr->ifr_newname);
2641 * Unknown or private ioctl
2645 if ((cmd >= SIOCDEVPRIVATE &&
2646 cmd <= SIOCDEVPRIVATE + 15) ||
2647 cmd == SIOCBONDENSLAVE ||
2648 cmd == SIOCBONDRELEASE ||
2649 cmd == SIOCBONDSETHWADDR ||
2650 cmd == SIOCBONDSLAVEINFOQUERY ||
2651 cmd == SIOCBONDINFOQUERY ||
2652 cmd == SIOCBONDCHANGEACTIVE ||
2653 cmd == SIOCGMIIPHY ||
2654 cmd == SIOCGMIIREG ||
2655 cmd == SIOCSMIIREG ||
2656 cmd == SIOCBRADDIF ||
2657 cmd == SIOCBRDELIF ||
2658 cmd == SIOCWANDEV) {
2660 if (dev->do_ioctl) {
2661 if (netif_device_present(dev))
2662 err = dev->do_ioctl(dev, ifr,
2675 * This function handles all "interface"-type I/O control requests. The actual
2676 * 'doing' part of this is dev_ifsioc above.
2680 * dev_ioctl - network device ioctl
2681 * @cmd: command to issue
2682 * @arg: pointer to a struct ifreq in user space
2684 * Issue ioctl functions to devices. This is normally called by the
2685 * user space syscall interfaces but can sometimes be useful for
2686 * other purposes. The return value is the return from the syscall if
2687 * positive or a negative errno code on error.
2690 int dev_ioctl(unsigned int cmd, void __user *arg)
2696 /* One special case: SIOCGIFCONF takes ifconf argument
2697 and requires shared lock, because it sleeps writing
2701 if (cmd == SIOCGIFCONF) {
2703 ret = dev_ifconf((char __user *) arg);
2707 if (cmd == SIOCGIFNAME)
2708 return dev_ifname((struct ifreq __user *)arg);
2710 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2713 ifr.ifr_name[IFNAMSIZ-1] = 0;
2715 colon = strchr(ifr.ifr_name, ':');
2720 * See which interface the caller is talking about.
2725 * These ioctl calls:
2726 * - can be done by all.
2727 * - atomic and do not require locking.
2738 dev_load(ifr.ifr_name);
2739 read_lock(&dev_base_lock);
2740 ret = dev_ifsioc(&ifr, cmd);
2741 read_unlock(&dev_base_lock);
2745 if (copy_to_user(arg, &ifr,
2746 sizeof(struct ifreq)))
2752 dev_load(ifr.ifr_name);
2754 ret = dev_ethtool(&ifr);
2759 if (copy_to_user(arg, &ifr,
2760 sizeof(struct ifreq)))
2766 * These ioctl calls:
2767 * - require superuser power.
2768 * - require strict serialization.
2774 if (!capable(CAP_NET_ADMIN))
2776 dev_load(ifr.ifr_name);
2778 ret = dev_ifsioc(&ifr, cmd);
2783 if (copy_to_user(arg, &ifr,
2784 sizeof(struct ifreq)))
2790 * These ioctl calls:
2791 * - require superuser power.
2792 * - require strict serialization.
2793 * - do not return a value
2803 case SIOCSIFHWBROADCAST:
2806 case SIOCBONDENSLAVE:
2807 case SIOCBONDRELEASE:
2808 case SIOCBONDSETHWADDR:
2809 case SIOCBONDSLAVEINFOQUERY:
2810 case SIOCBONDINFOQUERY:
2811 case SIOCBONDCHANGEACTIVE:
2814 if (!capable(CAP_NET_ADMIN))
2816 dev_load(ifr.ifr_name);
2818 ret = dev_ifsioc(&ifr, cmd);
2823 /* Get the per device memory space. We can add this but
2824 * currently do not support it */
2826 /* Set the per device memory buffer space.
2827 * Not applicable in our case */
2832 * Unknown or private ioctl.
2835 if (cmd == SIOCWANDEV ||
2836 (cmd >= SIOCDEVPRIVATE &&
2837 cmd <= SIOCDEVPRIVATE + 15)) {
2838 dev_load(ifr.ifr_name);
2840 ret = dev_ifsioc(&ifr, cmd);
2842 if (!ret && copy_to_user(arg, &ifr,
2843 sizeof(struct ifreq)))
2848 /* Take care of Wireless Extensions */
2849 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2850 /* If command is `set a parameter', or
2851 * `get the encoding parameters', check if
2852 * the user has the right to do it */
2853 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2854 if (!capable(CAP_NET_ADMIN))
2857 dev_load(ifr.ifr_name);
2859 /* Follow me in net/core/wireless.c */
2860 ret = wireless_process_ioctl(&ifr, cmd);
2862 if (!ret && IW_IS_GET(cmd) &&
2863 copy_to_user(arg, &ifr,
2864 sizeof(struct ifreq)))
2868 #endif /* WIRELESS_EXT */
2875 * dev_new_index - allocate an ifindex
2877 * Returns a suitable unique value for a new device interface
2878 * number. The caller must hold the rtnl semaphore or the
2879 * dev_base_lock to be sure it remains unique.
2881 int dev_new_index(void)
2887 if (!__dev_get_by_index(ifindex))
2892 static int dev_boot_phase = 1;
2894 /* Delayed registration/unregisteration */
2895 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2896 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2898 static inline void net_set_todo(struct net_device *dev)
2900 spin_lock(&net_todo_list_lock);
2901 list_add_tail(&dev->todo_list, &net_todo_list);
2902 spin_unlock(&net_todo_list_lock);
2906 * register_netdevice - register a network device
2907 * @dev: device to register
2909 * Take a completed network device structure and add it to the kernel
2910 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2911 * chain. 0 is returned on success. A negative errno code is returned
2912 * on a failure to set up the device, or if the name is a duplicate.
2914 * Callers must hold the rtnl semaphore. See the comment at the
2915 * end of Space.c for details about the locking. You may want
2916 * register_netdev() instead of this.
2919 * The locking appears insufficient to guarantee two parallel registers
2920 * will not get the same name.
2923 int register_netdevice(struct net_device *dev)
2925 struct hlist_head *head;
2926 struct hlist_node *p;
2929 BUG_ON(dev_boot_phase);
2932 /* When net_device's are persistent, this will be fatal. */
2933 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2935 spin_lock_init(&dev->queue_lock);
2936 spin_lock_init(&dev->xmit_lock);
2937 dev->xmit_lock_owner = -1;
2938 #ifdef CONFIG_NET_CLS_ACT
2939 spin_lock_init(&dev->ingress_lock);
2942 ret = alloc_divert_blk(dev);
2948 /* Init, if this function is available */
2950 ret = dev->init(dev);
2958 if (!dev_valid_name(dev->name)) {
2963 dev->ifindex = dev_new_index();
2964 if (dev->iflink == -1)
2965 dev->iflink = dev->ifindex;
2967 /* Check for existence of name */
2968 head = dev_name_hash(dev->name);
2969 hlist_for_each(p, head) {
2970 struct net_device *d
2971 = hlist_entry(p, struct net_device, name_hlist);
2972 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2978 /* Fix illegal SG+CSUM combinations. */
2979 if ((dev->features & NETIF_F_SG) &&
2980 !(dev->features & (NETIF_F_IP_CSUM |
2982 NETIF_F_HW_CSUM))) {
2983 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2985 dev->features &= ~NETIF_F_SG;
2989 * nil rebuild_header routine,
2990 * that should be never called and used as just bug trap.
2993 if (!dev->rebuild_header)
2994 dev->rebuild_header = default_rebuild_header;
2997 * Default initial state at registry is that the
2998 * device is present.
3001 set_bit(__LINK_STATE_PRESENT, &dev->state);
3004 dev_init_scheduler(dev);
3005 write_lock_bh(&dev_base_lock);
3007 dev_tail = &dev->next;
3008 hlist_add_head(&dev->name_hlist, head);
3009 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3011 dev->reg_state = NETREG_REGISTERING;
3012 write_unlock_bh(&dev_base_lock);
3014 /* Notify protocols, that a new device appeared. */
3015 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3017 /* Finish registration after unlock */
3024 free_divert_blk(dev);
3029 * netdev_wait_allrefs - wait until all references are gone.
3031 * This is called when unregistering network devices.
3033 * Any protocol or device that holds a reference should register
3034 * for netdevice notification, and cleanup and put back the
3035 * reference if they receive an UNREGISTER event.
3036 * We can get stuck here if buggy protocols don't correctly
3039 static void netdev_wait_allrefs(struct net_device *dev)
3041 unsigned long rebroadcast_time, warning_time;
3043 rebroadcast_time = warning_time = jiffies;
3044 while (atomic_read(&dev->refcnt) != 0) {
3045 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();
3065 rebroadcast_time = jiffies;
3068 current->state = TASK_INTERRUPTIBLE;
3069 schedule_timeout(HZ / 4);
3071 if (time_after(jiffies, warning_time + 10 * HZ)) {
3072 printk(KERN_EMERG "unregister_netdevice: "
3073 "waiting for %s to become free. Usage "
3075 dev->name, atomic_read(&dev->refcnt));
3076 warning_time = jiffies;
3085 * register_netdevice(x1);
3086 * register_netdevice(x2);
3088 * unregister_netdevice(y1);
3089 * unregister_netdevice(y2);
3095 * We are invoked by rtnl_unlock() after it drops the semaphore.
3096 * This allows us to deal with problems:
3097 * 1) We can create/delete sysfs objects which invoke hotplug
3098 * without deadlocking with linkwatch via keventd.
3099 * 2) Since we run with the RTNL semaphore not held, we can sleep
3100 * safely in order to wait for the netdev refcnt to drop to zero.
3102 static DECLARE_MUTEX(net_todo_run_mutex);
3103 void netdev_run_todo(void)
3105 struct list_head list = LIST_HEAD_INIT(list);
3109 /* Need to guard against multiple cpu's getting out of order. */
3110 down(&net_todo_run_mutex);
3112 /* Not safe to do outside the semaphore. We must not return
3113 * until all unregister events invoked by the local processor
3114 * have been completed (either by this todo run, or one on
3117 if (list_empty(&net_todo_list))
3120 /* Snapshot list, allow later requests */
3121 spin_lock(&net_todo_list_lock);
3122 list_splice_init(&net_todo_list, &list);
3123 spin_unlock(&net_todo_list_lock);
3125 while (!list_empty(&list)) {
3126 struct net_device *dev
3127 = list_entry(list.next, struct net_device, todo_list);
3128 list_del(&dev->todo_list);
3130 switch(dev->reg_state) {
3131 case NETREG_REGISTERING:
3132 err = netdev_register_sysfs(dev);
3134 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3136 dev->reg_state = NETREG_REGISTERED;
3139 case NETREG_UNREGISTERING:
3140 netdev_unregister_sysfs(dev);
3141 dev->reg_state = NETREG_UNREGISTERED;
3143 netdev_wait_allrefs(dev);
3146 BUG_ON(atomic_read(&dev->refcnt));
3147 BUG_TRAP(!dev->ip_ptr);
3148 BUG_TRAP(!dev->ip6_ptr);
3149 BUG_TRAP(!dev->dn_ptr);
3152 /* It must be the very last action,
3153 * after this 'dev' may point to freed up memory.
3155 if (dev->destructor)
3156 dev->destructor(dev);
3160 printk(KERN_ERR "network todo '%s' but state %d\n",
3161 dev->name, dev->reg_state);
3167 up(&net_todo_run_mutex);
3171 * free_netdev - free network device
3174 * This function does the last stage of destroying an allocated device
3175 * interface. The reference to the device object is released.
3176 * If this is the last reference then it will be freed.
3178 void free_netdev(struct net_device *dev)
3181 /* Compatiablity with error handling in drivers */
3182 if (dev->reg_state == NETREG_UNINITIALIZED) {
3183 kfree((char *)dev - dev->padded);
3187 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3188 dev->reg_state = NETREG_RELEASED;
3190 /* will free via class release */
3191 class_device_put(&dev->class_dev);
3193 kfree((char *)dev - dev->padded);
3197 /* Synchronize with packet receive processing. */
3198 void synchronize_net(void)
3201 synchronize_kernel();
3205 * unregister_netdevice - remove device from the kernel
3208 * This function shuts down a device interface and removes it
3209 * from the kernel tables. On success 0 is returned, on a failure
3210 * a negative errno code is returned.
3212 * Callers must hold the rtnl semaphore. See the comment at the
3213 * end of Space.c for details about the locking. You may want
3214 * unregister_netdev() instead of this.
3217 int unregister_netdevice(struct net_device *dev)
3219 struct net_device *d, **dp;
3221 BUG_ON(dev_boot_phase);
3224 /* Some devices call without registering for initialization unwind. */
3225 if (dev->reg_state == NETREG_UNINITIALIZED) {
3226 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3227 "was registered\n", dev->name, dev);
3231 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3233 /* If device is running, close it first. */
3234 if (dev->flags & IFF_UP)
3237 /* And unlink it from device chain. */
3238 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3240 write_lock_bh(&dev_base_lock);
3241 hlist_del(&dev->name_hlist);
3242 hlist_del(&dev->index_hlist);
3243 if (dev_tail == &dev->next)
3246 write_unlock_bh(&dev_base_lock);
3251 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3256 dev->reg_state = NETREG_UNREGISTERING;
3260 /* Shutdown queueing discipline. */
3264 /* Notify protocols, that we are about to destroy
3265 this device. They should clean all the things.
3267 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3270 * Flush the multicast chain
3272 dev_mc_discard(dev);
3277 /* Notifier chain MUST detach us from master device. */
3278 BUG_TRAP(!dev->master);
3280 free_divert_blk(dev);
3282 /* Finish processing unregister after unlock */
3291 #ifdef CONFIG_HOTPLUG_CPU
3292 static int dev_cpu_callback(struct notifier_block *nfb,
3293 unsigned long action,
3296 struct sk_buff **list_skb;
3297 struct net_device **list_net;
3298 struct sk_buff *skb;
3299 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3300 struct softnet_data *sd, *oldsd;
3302 if (action != CPU_DEAD)
3305 local_irq_disable();
3306 cpu = smp_processor_id();
3307 sd = &per_cpu(softnet_data, cpu);
3308 oldsd = &per_cpu(softnet_data, oldcpu);
3310 /* Find end of our completion_queue. */
3311 list_skb = &sd->completion_queue;
3313 list_skb = &(*list_skb)->next;
3314 /* Append completion queue from offline CPU. */
3315 *list_skb = oldsd->completion_queue;
3316 oldsd->completion_queue = NULL;
3318 /* Find end of our output_queue. */
3319 list_net = &sd->output_queue;
3321 list_net = &(*list_net)->next_sched;
3322 /* Append output queue from offline CPU. */
3323 *list_net = oldsd->output_queue;
3324 oldsd->output_queue = NULL;
3326 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3329 /* Process offline CPU's input_pkt_queue */
3330 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3335 #endif /* CONFIG_HOTPLUG_CPU */
3339 * Initialize the DEV module. At boot time this walks the device list and
3340 * unhooks any devices that fail to initialise (normally hardware not
3341 * present) and leaves us with a valid list of present and active devices.
3346 * This is called single threaded during boot, so no need
3347 * to take the rtnl semaphore.
3349 static int __init net_dev_init(void)
3351 int i, rc = -ENOMEM;
3353 BUG_ON(!dev_boot_phase);
3355 if (dev_proc_init())
3358 if (netdev_sysfs_init())
3361 INIT_LIST_HEAD(&ptype_all);
3362 for (i = 0; i < 16; i++)
3363 INIT_LIST_HEAD(&ptype_base[i]);
3365 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3366 INIT_HLIST_HEAD(&dev_name_head[i]);
3368 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3369 INIT_HLIST_HEAD(&dev_index_head[i]);
3372 * Initialise the packet receive queues.
3375 for (i = 0; i < NR_CPUS; i++) {
3376 struct softnet_data *queue;
3378 queue = &per_cpu(softnet_data, i);
3379 skb_queue_head_init(&queue->input_pkt_queue);
3380 queue->throttle = 0;
3381 queue->cng_level = 0;
3382 queue->avg_blog = 10; /* arbitrary non-zero */
3383 queue->completion_queue = NULL;
3384 INIT_LIST_HEAD(&queue->poll_list);
3385 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3386 queue->backlog_dev.weight = weight_p;
3387 queue->backlog_dev.poll = process_backlog;
3388 atomic_set(&queue->backlog_dev.refcnt, 1);
3391 #ifdef OFFLINE_SAMPLE
3392 samp_timer.expires = jiffies + (10 * HZ);
3393 add_timer(&samp_timer);
3398 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3399 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3401 hotcpu_notifier(dev_cpu_callback, 0);
3409 subsys_initcall(net_dev_init);
3411 static spinlock_t netdev_printk_lock = SPIN_LOCK_UNLOCKED;
3413 * __netdev_printk() - Log message with interface name, gated by message level
3414 * @sevlevel: severity level -- e.g., KERN_INFO
3415 * @netdev: net_device pointer
3416 * @msglevel: a standard message-level flag such as NETIF_MSG_PROBE.
3417 * Unless msglevel is NETIF_MSG_ALL, log the message only if
3418 * that flag is set in netdev->msg_enable.
3419 * @format: as with printk
3420 * @args: as with printk
3422 * Does the work for the netdev_printk macro.
3423 * For a lot of network drivers, the probe function looks like
3425 * netdev = alloc_netdev(...); // or alloc_etherdev(...)
3426 * SET_NETDEV_DEV(netdev, dev);
3428 * register_netdev(netdev);
3430 * netdev_printk and its wrappers (e.g., netdev_err) can be used as
3431 * soon as you have a valid net_device pointer -- e.g., from alloc_netdev,
3432 * alloc_etherdev, or init_etherdev. (Before that, use dev_printk and
3433 * its wrappers to report device errors.) It's common for an interface to
3434 * have a name like "eth%d" until the device is successfully configured,
3435 * and the call to register_netdev changes it to a "real" name like "eth0".
3437 * If the interface's reg_state is NETREG_REGISTERED, we assume that it has
3438 * been successfully set up in sysfs, and we prepend only the interface name
3439 * to the message -- e.g., "eth0: NIC Link is Down". The interface
3440 * name can be used to find eth0's driver, bus ID, etc. in sysfs.
3442 * For any other value of reg_state, we prepend the driver name and bus ID
3443 * as well as the (possibly incomplete) interface name -- e.g.,
3444 * "eth%d (e100 0000:00:03.0): Failed to map PCI address..."
3446 * Probe functions that alloc and register in one step (via init_etherdev),
3447 * or otherwise register the device before the probe completes successfully,
3448 * may need to take other steps to ensure that the failing device is clearly
3451 int __netdev_printk(const char *sevlevel, const struct net_device *netdev,
3452 int msglevel, const char *format, ...)
3454 if (!netdev || !format) {
3457 if (msglevel == NETIF_MSG_ALL || (netdev->msg_enable & msglevel)) {
3458 static char msg[512]; /* protected by netdev_printk_lock */
3459 unsigned long flags;
3461 struct device *dev = netdev->class_dev.dev;
3463 spin_lock_irqsave(&netdev_printk_lock, flags);
3464 va_start(args, format);
3465 vsnprintf(msg, 512, format, args);
3472 if (netdev->reg_state == NETREG_REGISTERED || !dev) {
3473 printk("%s%s: %s", sevlevel, netdev->name, msg);
3475 printk("%s%s (%s %s): %s", sevlevel, netdev->name,
3476 dev->driver->name, dev->bus_id, msg);
3478 spin_unlock_irqrestore(&netdev_printk_lock, flags);
3483 EXPORT_SYMBOL(__netdev_printk);
3484 EXPORT_SYMBOL(__dev_get);
3485 EXPORT_SYMBOL(__dev_get_by_flags);
3486 EXPORT_SYMBOL(__dev_get_by_index);
3487 EXPORT_SYMBOL(__dev_get_by_name);
3488 EXPORT_SYMBOL(__dev_remove_pack);
3489 EXPORT_SYMBOL(__skb_linearize);
3490 EXPORT_SYMBOL(call_netdevice_notifiers);
3491 EXPORT_SYMBOL(dev_add_pack);
3492 EXPORT_SYMBOL(dev_alloc_name);
3493 EXPORT_SYMBOL(dev_close);
3494 EXPORT_SYMBOL(dev_get_by_flags);
3495 EXPORT_SYMBOL(dev_get_by_index);
3496 EXPORT_SYMBOL(dev_get_by_name);
3497 EXPORT_SYMBOL(dev_getbyhwaddr);
3498 EXPORT_SYMBOL(dev_ioctl);
3499 EXPORT_SYMBOL(dev_new_index);
3500 EXPORT_SYMBOL(dev_open);
3501 EXPORT_SYMBOL(dev_queue_xmit);
3502 EXPORT_SYMBOL(dev_queue_xmit_nit);
3503 EXPORT_SYMBOL(dev_remove_pack);
3504 EXPORT_SYMBOL(dev_set_allmulti);
3505 EXPORT_SYMBOL(dev_set_promiscuity);
3506 EXPORT_SYMBOL(dev_change_flags);
3507 EXPORT_SYMBOL(dev_set_mtu);
3508 EXPORT_SYMBOL(free_netdev);
3509 EXPORT_SYMBOL(netdev_boot_setup_check);
3510 EXPORT_SYMBOL(netdev_set_master);
3511 EXPORT_SYMBOL(netdev_state_change);
3512 EXPORT_SYMBOL(netif_receive_skb);
3513 EXPORT_SYMBOL(netif_rx);
3514 EXPORT_SYMBOL(register_gifconf);
3515 EXPORT_SYMBOL(register_netdevice);
3516 EXPORT_SYMBOL(register_netdevice_notifier);
3517 EXPORT_SYMBOL(netdev_event);
3518 EXPORT_SYMBOL(skb_checksum_help);
3519 EXPORT_SYMBOL(synchronize_net);
3520 EXPORT_SYMBOL(unregister_netdevice);
3521 EXPORT_SYMBOL(unregister_netdevice_notifier);
3523 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3524 EXPORT_SYMBOL(br_handle_frame_hook);
3528 EXPORT_SYMBOL(dev_load);
3530 #ifdef CONFIG_NET_HW_FLOWCONTROL
3531 EXPORT_SYMBOL(netdev_dropping);
3532 EXPORT_SYMBOL(netdev_fc_xoff);
3533 EXPORT_SYMBOL(netdev_register_fc);
3534 EXPORT_SYMBOL(netdev_unregister_fc);
3537 #ifdef CONFIG_NET_CLS_ACT
3538 EXPORT_SYMBOL(ing_filter);
3542 EXPORT_PER_CPU_SYMBOL(softnet_data);