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
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 <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/config.h>
80 #include <linux/cpu.h>
81 #include <linux/types.h>
82 #include <linux/kernel.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
97 #include <linux/rtnetlink.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <linux/stat.h>
101 #include <linux/if_bridge.h>
102 #include <linux/divert.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/kmod.h>
109 #include <linux/module.h>
110 #include <linux/kallsyms.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <linux/wireless.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
121 * The list of packet types we will receive (as opposed to discard)
122 * and the routines to invoke.
124 * Why 16. Because with 16 the only overlap we get on a hash of the
125 * low nibble of the protocol value is RARP/SNAP/X.25.
127 * NOTE: That is no longer true with the addition of VLAN tags. Not
128 * sure which should go first, but I bet it won't make much
129 * difference if we are running VLANs. The good news is that
130 * this protocol won't be in the list unless compiled in, so
131 * the average user (w/out VLANs) will not be adversely affected.
148 static DEFINE_SPINLOCK(ptype_lock);
149 static struct list_head ptype_base[16]; /* 16 way hashed list */
150 static struct list_head ptype_all; /* Taps */
152 #ifdef CONFIG_NET_DMA
153 static struct dma_client *net_dma_client;
154 static unsigned int net_dma_count;
155 static spinlock_t net_dma_event_lock;
159 * The @dev_base list is protected by @dev_base_lock and the rtnl
162 * Pure readers hold dev_base_lock for reading.
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
177 struct net_device *dev_base;
178 static struct net_device **dev_tail = &dev_base;
179 DEFINE_RWLOCK(dev_base_lock);
181 EXPORT_SYMBOL(dev_base);
182 EXPORT_SYMBOL(dev_base_lock);
184 #define NETDEV_HASHBITS 8
185 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
186 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
188 static inline struct hlist_head *dev_name_hash(const char *name)
190 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
194 static inline struct hlist_head *dev_index_hash(int ifindex)
196 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
203 static RAW_NOTIFIER_HEAD(netdev_chain);
206 * Device drivers call our routines to queue packets here. We empty the
207 * queue in the local softnet handler.
209 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
212 extern int netdev_sysfs_init(void);
213 extern int netdev_register_sysfs(struct net_device *);
214 extern void netdev_unregister_sysfs(struct net_device *);
216 #define netdev_sysfs_init() (0)
217 #define netdev_register_sysfs(dev) (0)
218 #define netdev_unregister_sysfs(dev) do { } while(0)
222 /*******************************************************************************
224 Protocol management and registration routines
226 *******************************************************************************/
235 * Add a protocol ID to the list. Now that the input handler is
236 * smarter we can dispense with all the messy stuff that used to be
239 * BEWARE!!! Protocol handlers, mangling input packets,
240 * MUST BE last in hash buckets and checking protocol handlers
241 * MUST start from promiscuous ptype_all chain in net_bh.
242 * It is true now, do not change it.
243 * Explanation follows: if protocol handler, mangling packet, will
244 * be the first on list, it is not able to sense, that packet
245 * is cloned and should be copied-on-write, so that it will
246 * change it and subsequent readers will get broken packet.
251 * dev_add_pack - add packet handler
252 * @pt: packet type declaration
254 * Add a protocol handler to the networking stack. The passed &packet_type
255 * is linked into kernel lists and may not be freed until it has been
256 * removed from the kernel lists.
258 * This call does not sleep therefore it can not
259 * guarantee all CPU's that are in middle of receiving packets
260 * will see the new packet type (until the next received packet).
263 void dev_add_pack(struct packet_type *pt)
267 spin_lock_bh(&ptype_lock);
268 if (pt->type == htons(ETH_P_ALL)) {
270 list_add_rcu(&pt->list, &ptype_all);
272 hash = ntohs(pt->type) & 15;
273 list_add_rcu(&pt->list, &ptype_base[hash]);
275 spin_unlock_bh(&ptype_lock);
279 * __dev_remove_pack - remove packet handler
280 * @pt: packet type declaration
282 * Remove a protocol handler that was previously added to the kernel
283 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
284 * from the kernel lists and can be freed or reused once this function
287 * The packet type might still be in use by receivers
288 * and must not be freed until after all the CPU's have gone
289 * through a quiescent state.
291 void __dev_remove_pack(struct packet_type *pt)
293 struct list_head *head;
294 struct packet_type *pt1;
296 spin_lock_bh(&ptype_lock);
298 if (pt->type == htons(ETH_P_ALL)) {
302 head = &ptype_base[ntohs(pt->type) & 15];
304 list_for_each_entry(pt1, head, list) {
306 list_del_rcu(&pt->list);
311 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
313 spin_unlock_bh(&ptype_lock);
316 * dev_remove_pack - remove packet handler
317 * @pt: packet type declaration
319 * Remove a protocol handler that was previously added to the kernel
320 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
321 * from the kernel lists and can be freed or reused once this function
324 * This call sleeps to guarantee that no CPU is looking at the packet
327 void dev_remove_pack(struct packet_type *pt)
329 __dev_remove_pack(pt);
334 /******************************************************************************
336 Device Boot-time Settings Routines
338 *******************************************************************************/
340 /* Boot time configuration table */
341 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
344 * netdev_boot_setup_add - add new setup entry
345 * @name: name of the device
346 * @map: configured settings for the device
348 * Adds new setup entry to the dev_boot_setup list. The function
349 * returns 0 on error and 1 on success. This is a generic routine to
352 static int netdev_boot_setup_add(char *name, struct ifmap *map)
354 struct netdev_boot_setup *s;
358 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
359 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
360 memset(s[i].name, 0, sizeof(s[i].name));
361 strcpy(s[i].name, name);
362 memcpy(&s[i].map, map, sizeof(s[i].map));
367 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
371 * netdev_boot_setup_check - check boot time settings
372 * @dev: the netdevice
374 * Check boot time settings for the device.
375 * The found settings are set for the device to be used
376 * later in the device probing.
377 * Returns 0 if no settings found, 1 if they are.
379 int netdev_boot_setup_check(struct net_device *dev)
381 struct netdev_boot_setup *s = dev_boot_setup;
384 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
385 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
386 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
387 dev->irq = s[i].map.irq;
388 dev->base_addr = s[i].map.base_addr;
389 dev->mem_start = s[i].map.mem_start;
390 dev->mem_end = s[i].map.mem_end;
399 * netdev_boot_base - get address from boot time settings
400 * @prefix: prefix for network device
401 * @unit: id for network device
403 * Check boot time settings for the base address of device.
404 * The found settings are set for the device to be used
405 * later in the device probing.
406 * Returns 0 if no settings found.
408 unsigned long netdev_boot_base(const char *prefix, int unit)
410 const struct netdev_boot_setup *s = dev_boot_setup;
414 sprintf(name, "%s%d", prefix, unit);
417 * If device already registered then return base of 1
418 * to indicate not to probe for this interface
420 if (__dev_get_by_name(name))
423 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
424 if (!strcmp(name, s[i].name))
425 return s[i].map.base_addr;
430 * Saves at boot time configured settings for any netdevice.
432 int __init netdev_boot_setup(char *str)
437 str = get_options(str, ARRAY_SIZE(ints), ints);
442 memset(&map, 0, sizeof(map));
446 map.base_addr = ints[2];
448 map.mem_start = ints[3];
450 map.mem_end = ints[4];
452 /* Add new entry to the list */
453 return netdev_boot_setup_add(str, &map);
456 __setup("netdev=", netdev_boot_setup);
458 /*******************************************************************************
460 Device Interface Subroutines
462 *******************************************************************************/
465 * __dev_get_by_name - find a device by its name
466 * @name: name to find
468 * Find an interface by name. Must be called under RTNL semaphore
469 * or @dev_base_lock. If the name is found a pointer to the device
470 * is returned. If the name is not found then %NULL is returned. The
471 * reference counters are not incremented so the caller must be
472 * careful with locks.
475 struct net_device *__dev_get_by_name(const char *name)
477 struct hlist_node *p;
479 hlist_for_each(p, dev_name_hash(name)) {
480 struct net_device *dev
481 = hlist_entry(p, struct net_device, name_hlist);
482 if (!strncmp(dev->name, name, IFNAMSIZ))
489 * dev_get_by_name - find a device by its name
490 * @name: name to find
492 * Find an interface by name. This can be called from any
493 * context and does its own locking. The returned handle has
494 * the usage count incremented and the caller must use dev_put() to
495 * release it when it is no longer needed. %NULL is returned if no
496 * matching device is found.
499 struct net_device *dev_get_by_name(const char *name)
501 struct net_device *dev;
503 read_lock(&dev_base_lock);
504 dev = __dev_get_by_name(name);
507 read_unlock(&dev_base_lock);
512 * __dev_get_by_index - find a device by its ifindex
513 * @ifindex: index of device
515 * Search for an interface by index. Returns %NULL if the device
516 * is not found or a pointer to the device. The device has not
517 * had its reference counter increased so the caller must be careful
518 * about locking. The caller must hold either the RTNL semaphore
522 struct net_device *__dev_get_by_index(int ifindex)
524 struct hlist_node *p;
526 hlist_for_each(p, dev_index_hash(ifindex)) {
527 struct net_device *dev
528 = hlist_entry(p, struct net_device, index_hlist);
529 if (dev->ifindex == ifindex)
537 * dev_get_by_index - find a device by its ifindex
538 * @ifindex: index of device
540 * Search for an interface by index. Returns NULL if the device
541 * is not found or a pointer to the device. The device returned has
542 * had a reference added and the pointer is safe until the user calls
543 * dev_put to indicate they have finished with it.
546 struct net_device *dev_get_by_index(int ifindex)
548 struct net_device *dev;
550 read_lock(&dev_base_lock);
551 dev = __dev_get_by_index(ifindex);
554 read_unlock(&dev_base_lock);
559 * dev_getbyhwaddr - find a device by its hardware address
560 * @type: media type of device
561 * @ha: hardware address
563 * Search for an interface by MAC address. Returns NULL if the device
564 * is not found or a pointer to the device. The caller must hold the
565 * rtnl semaphore. The returned device has not had its ref count increased
566 * and the caller must therefore be careful about locking
569 * If the API was consistent this would be __dev_get_by_hwaddr
572 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
574 struct net_device *dev;
578 for (dev = dev_base; dev; dev = dev->next)
579 if (dev->type == type &&
580 !memcmp(dev->dev_addr, ha, dev->addr_len))
585 EXPORT_SYMBOL(dev_getbyhwaddr);
587 struct net_device *dev_getfirstbyhwtype(unsigned short type)
589 struct net_device *dev;
592 for (dev = dev_base; dev; dev = dev->next) {
593 if (dev->type == type) {
602 EXPORT_SYMBOL(dev_getfirstbyhwtype);
605 * dev_get_by_flags - find any device with given flags
606 * @if_flags: IFF_* values
607 * @mask: bitmask of bits in if_flags to check
609 * Search for any interface with the given flags. Returns NULL if a device
610 * is not found or a pointer to the device. The device returned has
611 * had a reference added and the pointer is safe until the user calls
612 * dev_put to indicate they have finished with it.
615 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
617 struct net_device *dev;
619 read_lock(&dev_base_lock);
620 for (dev = dev_base; dev != NULL; dev = dev->next) {
621 if (((dev->flags ^ if_flags) & mask) == 0) {
626 read_unlock(&dev_base_lock);
631 * dev_valid_name - check if name is okay for network device
634 * Network device names need to be valid file names to
635 * to allow sysfs to work
637 int dev_valid_name(const char *name)
639 return !(*name == '\0'
640 || !strcmp(name, ".")
641 || !strcmp(name, "..")
642 || strchr(name, '/'));
646 * dev_alloc_name - allocate a name for a device
648 * @name: name format string
650 * Passed a format string - eg "lt%d" it will try and find a suitable
651 * id. It scans list of devices to build up a free map, then chooses
652 * the first empty slot. The caller must hold the dev_base or rtnl lock
653 * while allocating the name and adding the device in order to avoid
655 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
656 * Returns the number of the unit assigned or a negative errno code.
659 int dev_alloc_name(struct net_device *dev, const char *name)
664 const int max_netdevices = 8*PAGE_SIZE;
666 struct net_device *d;
668 p = strnchr(name, IFNAMSIZ-1, '%');
671 * Verify the string as this thing may have come from
672 * the user. There must be either one "%d" and no other "%"
675 if (p[1] != 'd' || strchr(p + 2, '%'))
678 /* Use one page as a bit array of possible slots */
679 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
683 for (d = dev_base; d; d = d->next) {
684 if (!sscanf(d->name, name, &i))
686 if (i < 0 || i >= max_netdevices)
689 /* avoid cases where sscanf is not exact inverse of printf */
690 snprintf(buf, sizeof(buf), name, i);
691 if (!strncmp(buf, d->name, IFNAMSIZ))
695 i = find_first_zero_bit(inuse, max_netdevices);
696 free_page((unsigned long) inuse);
699 snprintf(buf, sizeof(buf), name, i);
700 if (!__dev_get_by_name(buf)) {
701 strlcpy(dev->name, buf, IFNAMSIZ);
705 /* It is possible to run out of possible slots
706 * when the name is long and there isn't enough space left
707 * for the digits, or if all bits are used.
714 * dev_change_name - change name of a device
716 * @newname: name (or format string) must be at least IFNAMSIZ
718 * Change name of a device, can pass format strings "eth%d".
721 int dev_change_name(struct net_device *dev, char *newname)
727 if (dev->flags & IFF_UP)
730 if (!dev_valid_name(newname))
733 if (strchr(newname, '%')) {
734 err = dev_alloc_name(dev, newname);
737 strcpy(newname, dev->name);
739 else if (__dev_get_by_name(newname))
742 strlcpy(dev->name, newname, IFNAMSIZ);
744 err = class_device_rename(&dev->class_dev, dev->name);
746 hlist_del(&dev->name_hlist);
747 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
748 raw_notifier_call_chain(&netdev_chain,
749 NETDEV_CHANGENAME, dev);
756 * netdev_features_change - device changes features
757 * @dev: device to cause notification
759 * Called to indicate a device has changed features.
761 void netdev_features_change(struct net_device *dev)
763 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
765 EXPORT_SYMBOL(netdev_features_change);
768 * netdev_state_change - device changes state
769 * @dev: device to cause notification
771 * Called to indicate a device has changed state. This function calls
772 * the notifier chains for netdev_chain and sends a NEWLINK message
773 * to the routing socket.
775 void netdev_state_change(struct net_device *dev)
777 if (dev->flags & IFF_UP) {
778 raw_notifier_call_chain(&netdev_chain,
780 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
785 * dev_load - load a network module
786 * @name: name of interface
788 * If a network interface is not present and the process has suitable
789 * privileges this function loads the module. If module loading is not
790 * available in this kernel then it becomes a nop.
793 void dev_load(const char *name)
795 struct net_device *dev;
797 read_lock(&dev_base_lock);
798 dev = __dev_get_by_name(name);
799 read_unlock(&dev_base_lock);
801 if (!dev && capable(CAP_SYS_MODULE))
802 request_module("%s", name);
805 static int default_rebuild_header(struct sk_buff *skb)
807 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
808 skb->dev ? skb->dev->name : "NULL!!!");
815 * dev_open - prepare an interface for use.
816 * @dev: device to open
818 * Takes a device from down to up state. The device's private open
819 * function is invoked and then the multicast lists are loaded. Finally
820 * the device is moved into the up state and a %NETDEV_UP message is
821 * sent to the netdev notifier chain.
823 * Calling this function on an active interface is a nop. On a failure
824 * a negative errno code is returned.
826 int dev_open(struct net_device *dev)
834 if (dev->flags & IFF_UP)
838 * Is it even present?
840 if (!netif_device_present(dev))
844 * Call device private open method
846 set_bit(__LINK_STATE_START, &dev->state);
848 ret = dev->open(dev);
850 clear_bit(__LINK_STATE_START, &dev->state);
854 * If it went open OK then:
861 dev->flags |= IFF_UP;
864 * Initialize multicasting status
869 * Wakeup transmit queue engine
874 * ... and announce new interface.
876 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
882 * dev_close - shutdown an interface.
883 * @dev: device to shutdown
885 * This function moves an active device into down state. A
886 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
887 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
890 int dev_close(struct net_device *dev)
892 if (!(dev->flags & IFF_UP))
896 * Tell people we are going down, so that they can
897 * prepare to death, when device is still operating.
899 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
903 clear_bit(__LINK_STATE_START, &dev->state);
905 /* Synchronize to scheduled poll. We cannot touch poll list,
906 * it can be even on different cpu. So just clear netif_running(),
907 * and wait when poll really will happen. Actually, the best place
908 * for this is inside dev->stop() after device stopped its irq
909 * engine, but this requires more changes in devices. */
911 smp_mb__after_clear_bit(); /* Commit netif_running(). */
912 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
918 * Call the device specific close. This cannot fail.
919 * Only if device is UP
921 * We allow it to be called even after a DETACH hot-plug
928 * Device is now down.
931 dev->flags &= ~IFF_UP;
934 * Tell people we are down
936 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
943 * Device change register/unregister. These are not inline or static
944 * as we export them to the world.
948 * register_netdevice_notifier - register a network notifier block
951 * Register a notifier to be called when network device events occur.
952 * The notifier passed is linked into the kernel structures and must
953 * not be reused until it has been unregistered. A negative errno code
954 * is returned on a failure.
956 * When registered all registration and up events are replayed
957 * to the new notifier to allow device to have a race free
958 * view of the network device list.
961 int register_netdevice_notifier(struct notifier_block *nb)
963 struct net_device *dev;
967 err = raw_notifier_chain_register(&netdev_chain, nb);
969 for (dev = dev_base; dev; dev = dev->next) {
970 nb->notifier_call(nb, NETDEV_REGISTER, dev);
972 if (dev->flags & IFF_UP)
973 nb->notifier_call(nb, NETDEV_UP, dev);
981 * unregister_netdevice_notifier - unregister a network notifier block
984 * Unregister a notifier previously registered by
985 * register_netdevice_notifier(). The notifier is unlinked into the
986 * kernel structures and may then be reused. A negative errno code
987 * is returned on a failure.
990 int unregister_netdevice_notifier(struct notifier_block *nb)
995 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1001 * call_netdevice_notifiers - call all network notifier blocks
1002 * @val: value passed unmodified to notifier function
1003 * @v: pointer passed unmodified to notifier function
1005 * Call all network notifier blocks. Parameters and return value
1006 * are as for raw_notifier_call_chain().
1009 int call_netdevice_notifiers(unsigned long val, void *v)
1011 return raw_notifier_call_chain(&netdev_chain, val, v);
1014 /* When > 0 there are consumers of rx skb time stamps */
1015 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1017 void net_enable_timestamp(void)
1019 atomic_inc(&netstamp_needed);
1022 void net_disable_timestamp(void)
1024 atomic_dec(&netstamp_needed);
1027 void __net_timestamp(struct sk_buff *skb)
1031 do_gettimeofday(&tv);
1032 skb_set_timestamp(skb, &tv);
1034 EXPORT_SYMBOL(__net_timestamp);
1036 static inline void net_timestamp(struct sk_buff *skb)
1038 if (atomic_read(&netstamp_needed))
1039 __net_timestamp(skb);
1041 skb->tstamp.off_sec = 0;
1042 skb->tstamp.off_usec = 0;
1047 * Support routine. Sends outgoing frames to any network
1048 * taps currently in use.
1051 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1053 struct packet_type *ptype;
1058 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1059 /* Never send packets back to the socket
1060 * they originated from - MvS (miquels@drinkel.ow.org)
1062 if ((ptype->dev == dev || !ptype->dev) &&
1063 (ptype->af_packet_priv == NULL ||
1064 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1065 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1069 /* skb->nh should be correctly
1070 set by sender, so that the second statement is
1071 just protection against buggy protocols.
1073 skb2->mac.raw = skb2->data;
1075 if (skb2->nh.raw < skb2->data ||
1076 skb2->nh.raw > skb2->tail) {
1077 if (net_ratelimit())
1078 printk(KERN_CRIT "protocol %04x is "
1080 skb2->protocol, dev->name);
1081 skb2->nh.raw = skb2->data;
1084 skb2->h.raw = skb2->nh.raw;
1085 skb2->pkt_type = PACKET_OUTGOING;
1086 ptype->func(skb2, skb->dev, ptype, skb->dev);
1093 void __netif_schedule(struct net_device *dev)
1095 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1096 unsigned long flags;
1097 struct softnet_data *sd;
1099 local_irq_save(flags);
1100 sd = &__get_cpu_var(softnet_data);
1101 dev->next_sched = sd->output_queue;
1102 sd->output_queue = dev;
1103 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1104 local_irq_restore(flags);
1107 EXPORT_SYMBOL(__netif_schedule);
1109 void __netif_rx_schedule(struct net_device *dev)
1111 unsigned long flags;
1113 local_irq_save(flags);
1115 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1117 dev->quota += dev->weight;
1119 dev->quota = dev->weight;
1120 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1121 local_irq_restore(flags);
1123 EXPORT_SYMBOL(__netif_rx_schedule);
1125 void dev_kfree_skb_any(struct sk_buff *skb)
1127 if (in_irq() || irqs_disabled())
1128 dev_kfree_skb_irq(skb);
1132 EXPORT_SYMBOL(dev_kfree_skb_any);
1136 void netif_device_detach(struct net_device *dev)
1138 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1139 netif_running(dev)) {
1140 netif_stop_queue(dev);
1143 EXPORT_SYMBOL(netif_device_detach);
1145 void netif_device_attach(struct net_device *dev)
1147 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1148 netif_running(dev)) {
1149 netif_wake_queue(dev);
1150 __netdev_watchdog_up(dev);
1153 EXPORT_SYMBOL(netif_device_attach);
1157 * Invalidate hardware checksum when packet is to be mangled, and
1158 * complete checksum manually on outgoing path.
1160 int skb_checksum_help(struct sk_buff *skb, int inward)
1163 int ret = 0, offset = skb->h.raw - skb->data;
1166 skb->ip_summed = CHECKSUM_NONE;
1170 if (skb_cloned(skb)) {
1171 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1176 BUG_ON(offset > (int)skb->len);
1177 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1179 offset = skb->tail - skb->h.raw;
1180 BUG_ON(offset <= 0);
1181 BUG_ON(skb->csum + 2 > offset);
1183 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1184 skb->ip_summed = CHECKSUM_NONE;
1189 /* Take action when hardware reception checksum errors are detected. */
1191 void netdev_rx_csum_fault(struct net_device *dev)
1193 if (net_ratelimit()) {
1194 printk(KERN_ERR "%s: hw csum failure.\n",
1195 dev ? dev->name : "<unknown>");
1199 EXPORT_SYMBOL(netdev_rx_csum_fault);
1202 #ifdef CONFIG_HIGHMEM
1203 /* Actually, we should eliminate this check as soon as we know, that:
1204 * 1. IOMMU is present and allows to map all the memory.
1205 * 2. No high memory really exists on this machine.
1208 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1212 if (dev->features & NETIF_F_HIGHDMA)
1215 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1216 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1222 #define illegal_highdma(dev, skb) (0)
1225 /* Keep head the same: replace data */
1226 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1231 struct skb_shared_info *ninfo;
1232 int headerlen = skb->data - skb->head;
1233 int expand = (skb->tail + skb->data_len) - skb->end;
1235 if (skb_shared(skb))
1241 size = skb->end - skb->head + expand;
1242 size = SKB_DATA_ALIGN(size);
1243 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1247 /* Copy entire thing */
1248 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1252 ninfo = (struct skb_shared_info*)(data + size);
1253 atomic_set(&ninfo->dataref, 1);
1254 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1255 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1256 ninfo->nr_frags = 0;
1257 ninfo->frag_list = NULL;
1259 /* Offset between the two in bytes */
1260 offset = data - skb->head;
1262 /* Free old data. */
1263 skb_release_data(skb);
1266 skb->end = data + size;
1268 /* Set up new pointers */
1269 skb->h.raw += offset;
1270 skb->nh.raw += offset;
1271 skb->mac.raw += offset;
1272 skb->tail += offset;
1273 skb->data += offset;
1275 /* We are no longer a clone, even if we were. */
1278 skb->tail += skb->data_len;
1283 #define HARD_TX_LOCK(dev, cpu) { \
1284 if ((dev->features & NETIF_F_LLTX) == 0) { \
1285 spin_lock(&dev->xmit_lock); \
1286 dev->xmit_lock_owner = cpu; \
1290 #define HARD_TX_UNLOCK(dev) { \
1291 if ((dev->features & NETIF_F_LLTX) == 0) { \
1292 dev->xmit_lock_owner = -1; \
1293 spin_unlock(&dev->xmit_lock); \
1298 * dev_queue_xmit - transmit a buffer
1299 * @skb: buffer to transmit
1301 * Queue a buffer for transmission to a network device. The caller must
1302 * have set the device and priority and built the buffer before calling
1303 * this function. The function can be called from an interrupt.
1305 * A negative errno code is returned on a failure. A success does not
1306 * guarantee the frame will be transmitted as it may be dropped due
1307 * to congestion or traffic shaping.
1309 * -----------------------------------------------------------------------------------
1310 * I notice this method can also return errors from the queue disciplines,
1311 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1314 * Regardless of the return value, the skb is consumed, so it is currently
1315 * difficult to retry a send to this method. (You can bump the ref count
1316 * before sending to hold a reference for retry if you are careful.)
1318 * When calling this method, interrupts MUST be enabled. This is because
1319 * the BH enable code must have IRQs enabled so that it will not deadlock.
1323 int dev_queue_xmit(struct sk_buff *skb)
1325 struct net_device *dev = skb->dev;
1329 if (skb_shinfo(skb)->frag_list &&
1330 !(dev->features & NETIF_F_FRAGLIST) &&
1331 __skb_linearize(skb, GFP_ATOMIC))
1334 /* Fragmented skb is linearized if device does not support SG,
1335 * or if at least one of fragments is in highmem and device
1336 * does not support DMA from it.
1338 if (skb_shinfo(skb)->nr_frags &&
1339 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1340 __skb_linearize(skb, GFP_ATOMIC))
1343 /* If packet is not checksummed and device does not support
1344 * checksumming for this protocol, complete checksumming here.
1346 if (skb->ip_summed == CHECKSUM_HW &&
1347 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1348 (!(dev->features & NETIF_F_IP_CSUM) ||
1349 skb->protocol != htons(ETH_P_IP))))
1350 if (skb_checksum_help(skb, 0))
1353 spin_lock_prefetch(&dev->queue_lock);
1355 /* Disable soft irqs for various locks below. Also
1356 * stops preemption for RCU.
1360 /* Updates of qdisc are serialized by queue_lock.
1361 * The struct Qdisc which is pointed to by qdisc is now a
1362 * rcu structure - it may be accessed without acquiring
1363 * a lock (but the structure may be stale.) The freeing of the
1364 * qdisc will be deferred until it's known that there are no
1365 * more references to it.
1367 * If the qdisc has an enqueue function, we still need to
1368 * hold the queue_lock before calling it, since queue_lock
1369 * also serializes access to the device queue.
1372 q = rcu_dereference(dev->qdisc);
1373 #ifdef CONFIG_NET_CLS_ACT
1374 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1377 /* Grab device queue */
1378 spin_lock(&dev->queue_lock);
1380 rc = q->enqueue(skb, q);
1384 spin_unlock(&dev->queue_lock);
1385 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1389 /* The device has no queue. Common case for software devices:
1390 loopback, all the sorts of tunnels...
1392 Really, it is unlikely that xmit_lock protection is necessary here.
1393 (f.e. loopback and IP tunnels are clean ignoring statistics
1395 However, it is possible, that they rely on protection
1398 Check this and shot the lock. It is not prone from deadlocks.
1399 Either shot noqueue qdisc, it is even simpler 8)
1401 if (dev->flags & IFF_UP) {
1402 int cpu = smp_processor_id(); /* ok because BHs are off */
1404 if (dev->xmit_lock_owner != cpu) {
1406 HARD_TX_LOCK(dev, cpu);
1408 if (!netif_queue_stopped(dev)) {
1410 dev_queue_xmit_nit(skb, dev);
1413 if (!dev->hard_start_xmit(skb, dev)) {
1414 HARD_TX_UNLOCK(dev);
1418 HARD_TX_UNLOCK(dev);
1419 if (net_ratelimit())
1420 printk(KERN_CRIT "Virtual device %s asks to "
1421 "queue packet!\n", dev->name);
1423 /* Recursion is detected! It is possible,
1425 if (net_ratelimit())
1426 printk(KERN_CRIT "Dead loop on virtual device "
1427 "%s, fix it urgently!\n", dev->name);
1443 /*=======================================================================
1445 =======================================================================*/
1447 int netdev_max_backlog = 1000;
1448 int netdev_budget = 300;
1449 int weight_p = 64; /* old backlog weight */
1451 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1455 * netif_rx - post buffer to the network code
1456 * @skb: buffer to post
1458 * This function receives a packet from a device driver and queues it for
1459 * the upper (protocol) levels to process. It always succeeds. The buffer
1460 * may be dropped during processing for congestion control or by the
1464 * NET_RX_SUCCESS (no congestion)
1465 * NET_RX_CN_LOW (low congestion)
1466 * NET_RX_CN_MOD (moderate congestion)
1467 * NET_RX_CN_HIGH (high congestion)
1468 * NET_RX_DROP (packet was dropped)
1472 int netif_rx(struct sk_buff *skb)
1474 struct softnet_data *queue;
1475 unsigned long flags;
1477 /* if netpoll wants it, pretend we never saw it */
1478 if (netpoll_rx(skb))
1481 if (!skb->tstamp.off_sec)
1485 * The code is rearranged so that the path is the most
1486 * short when CPU is congested, but is still operating.
1488 local_irq_save(flags);
1489 queue = &__get_cpu_var(softnet_data);
1491 __get_cpu_var(netdev_rx_stat).total++;
1492 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1493 if (queue->input_pkt_queue.qlen) {
1496 __skb_queue_tail(&queue->input_pkt_queue, skb);
1497 local_irq_restore(flags);
1498 return NET_RX_SUCCESS;
1501 netif_rx_schedule(&queue->backlog_dev);
1505 __get_cpu_var(netdev_rx_stat).dropped++;
1506 local_irq_restore(flags);
1512 int netif_rx_ni(struct sk_buff *skb)
1517 err = netif_rx(skb);
1518 if (local_softirq_pending())
1525 EXPORT_SYMBOL(netif_rx_ni);
1527 static inline struct net_device *skb_bond(struct sk_buff *skb)
1529 struct net_device *dev = skb->dev;
1533 * On bonding slaves other than the currently active
1534 * slave, suppress duplicates except for 802.3ad
1535 * ETH_P_SLOW and alb non-mcast/bcast.
1537 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1538 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1539 if (skb->pkt_type != PACKET_BROADCAST &&
1540 skb->pkt_type != PACKET_MULTICAST)
1544 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1545 skb->protocol == __constant_htons(ETH_P_SLOW))
1552 skb->dev = dev->master;
1558 static void net_tx_action(struct softirq_action *h)
1560 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1562 if (sd->completion_queue) {
1563 struct sk_buff *clist;
1565 local_irq_disable();
1566 clist = sd->completion_queue;
1567 sd->completion_queue = NULL;
1571 struct sk_buff *skb = clist;
1572 clist = clist->next;
1574 BUG_TRAP(!atomic_read(&skb->users));
1579 if (sd->output_queue) {
1580 struct net_device *head;
1582 local_irq_disable();
1583 head = sd->output_queue;
1584 sd->output_queue = NULL;
1588 struct net_device *dev = head;
1589 head = head->next_sched;
1591 smp_mb__before_clear_bit();
1592 clear_bit(__LINK_STATE_SCHED, &dev->state);
1594 if (spin_trylock(&dev->queue_lock)) {
1596 spin_unlock(&dev->queue_lock);
1598 netif_schedule(dev);
1604 static __inline__ int deliver_skb(struct sk_buff *skb,
1605 struct packet_type *pt_prev,
1606 struct net_device *orig_dev)
1608 atomic_inc(&skb->users);
1609 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1612 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1613 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1615 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1616 unsigned char *addr);
1617 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1619 static __inline__ int handle_bridge(struct sk_buff **pskb,
1620 struct packet_type **pt_prev, int *ret,
1621 struct net_device *orig_dev)
1623 struct net_bridge_port *port;
1625 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1626 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1630 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1634 return br_handle_frame_hook(port, pskb);
1637 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1640 #ifdef CONFIG_NET_CLS_ACT
1641 /* TODO: Maybe we should just force sch_ingress to be compiled in
1642 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1643 * a compare and 2 stores extra right now if we dont have it on
1644 * but have CONFIG_NET_CLS_ACT
1645 * NOTE: This doesnt stop any functionality; if you dont have
1646 * the ingress scheduler, you just cant add policies on ingress.
1649 static int ing_filter(struct sk_buff *skb)
1652 struct net_device *dev = skb->dev;
1653 int result = TC_ACT_OK;
1655 if (dev->qdisc_ingress) {
1656 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1657 if (MAX_RED_LOOP < ttl++) {
1658 printk("Redir loop detected Dropping packet (%s->%s)\n",
1659 skb->input_dev->name, skb->dev->name);
1663 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1665 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1667 spin_lock(&dev->ingress_lock);
1668 if ((q = dev->qdisc_ingress) != NULL)
1669 result = q->enqueue(skb, q);
1670 spin_unlock(&dev->ingress_lock);
1678 int netif_receive_skb(struct sk_buff *skb)
1680 struct packet_type *ptype, *pt_prev;
1681 struct net_device *orig_dev;
1682 int ret = NET_RX_DROP;
1683 unsigned short type;
1685 /* if we've gotten here through NAPI, check netpoll */
1686 if (skb->dev->poll && netpoll_rx(skb))
1689 if (!skb->tstamp.off_sec)
1692 if (!skb->input_dev)
1693 skb->input_dev = skb->dev;
1695 orig_dev = skb_bond(skb);
1700 __get_cpu_var(netdev_rx_stat).total++;
1702 skb->h.raw = skb->nh.raw = skb->data;
1703 skb->mac_len = skb->nh.raw - skb->mac.raw;
1709 #ifdef CONFIG_NET_CLS_ACT
1710 if (skb->tc_verd & TC_NCLS) {
1711 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1716 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1717 if (!ptype->dev || ptype->dev == skb->dev) {
1719 ret = deliver_skb(skb, pt_prev, orig_dev);
1724 #ifdef CONFIG_NET_CLS_ACT
1726 ret = deliver_skb(skb, pt_prev, orig_dev);
1727 pt_prev = NULL; /* noone else should process this after*/
1729 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1732 ret = ing_filter(skb);
1734 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1743 handle_diverter(skb);
1745 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1748 type = skb->protocol;
1749 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1750 if (ptype->type == type &&
1751 (!ptype->dev || ptype->dev == skb->dev)) {
1753 ret = deliver_skb(skb, pt_prev, orig_dev);
1759 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1762 /* Jamal, now you will not able to escape explaining
1763 * me how you were going to use this. :-)
1773 static int process_backlog(struct net_device *backlog_dev, int *budget)
1776 int quota = min(backlog_dev->quota, *budget);
1777 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1778 unsigned long start_time = jiffies;
1780 backlog_dev->weight = weight_p;
1782 struct sk_buff *skb;
1783 struct net_device *dev;
1785 local_irq_disable();
1786 skb = __skb_dequeue(&queue->input_pkt_queue);
1793 netif_receive_skb(skb);
1799 if (work >= quota || jiffies - start_time > 1)
1804 backlog_dev->quota -= work;
1809 backlog_dev->quota -= work;
1812 list_del(&backlog_dev->poll_list);
1813 smp_mb__before_clear_bit();
1814 netif_poll_enable(backlog_dev);
1820 static void net_rx_action(struct softirq_action *h)
1822 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1823 unsigned long start_time = jiffies;
1824 int budget = netdev_budget;
1827 local_irq_disable();
1829 while (!list_empty(&queue->poll_list)) {
1830 struct net_device *dev;
1832 if (budget <= 0 || jiffies - start_time > 1)
1837 dev = list_entry(queue->poll_list.next,
1838 struct net_device, poll_list);
1839 have = netpoll_poll_lock(dev);
1841 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1842 netpoll_poll_unlock(have);
1843 local_irq_disable();
1844 list_move_tail(&dev->poll_list, &queue->poll_list);
1846 dev->quota += dev->weight;
1848 dev->quota = dev->weight;
1850 netpoll_poll_unlock(have);
1852 local_irq_disable();
1856 #ifdef CONFIG_NET_DMA
1858 * There may not be any more sk_buffs coming right now, so push
1859 * any pending DMA copies to hardware
1861 if (net_dma_client) {
1862 struct dma_chan *chan;
1864 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1865 dma_async_memcpy_issue_pending(chan);
1873 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1874 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1878 static gifconf_func_t * gifconf_list [NPROTO];
1881 * register_gifconf - register a SIOCGIF handler
1882 * @family: Address family
1883 * @gifconf: Function handler
1885 * Register protocol dependent address dumping routines. The handler
1886 * that is passed must not be freed or reused until it has been replaced
1887 * by another handler.
1889 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1891 if (family >= NPROTO)
1893 gifconf_list[family] = gifconf;
1899 * Map an interface index to its name (SIOCGIFNAME)
1903 * We need this ioctl for efficient implementation of the
1904 * if_indextoname() function required by the IPv6 API. Without
1905 * it, we would have to search all the interfaces to find a
1909 static int dev_ifname(struct ifreq __user *arg)
1911 struct net_device *dev;
1915 * Fetch the caller's info block.
1918 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1921 read_lock(&dev_base_lock);
1922 dev = __dev_get_by_index(ifr.ifr_ifindex);
1924 read_unlock(&dev_base_lock);
1928 strcpy(ifr.ifr_name, dev->name);
1929 read_unlock(&dev_base_lock);
1931 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1937 * Perform a SIOCGIFCONF call. This structure will change
1938 * size eventually, and there is nothing I can do about it.
1939 * Thus we will need a 'compatibility mode'.
1942 static int dev_ifconf(char __user *arg)
1945 struct net_device *dev;
1952 * Fetch the caller's info block.
1955 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1962 * Loop over the interfaces, and write an info block for each.
1966 for (dev = dev_base; dev; dev = dev->next) {
1967 for (i = 0; i < NPROTO; i++) {
1968 if (gifconf_list[i]) {
1971 done = gifconf_list[i](dev, NULL, 0);
1973 done = gifconf_list[i](dev, pos + total,
1983 * All done. Write the updated control block back to the caller.
1985 ifc.ifc_len = total;
1988 * Both BSD and Solaris return 0 here, so we do too.
1990 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1993 #ifdef CONFIG_PROC_FS
1995 * This is invoked by the /proc filesystem handler to display a device
1998 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2000 struct net_device *dev;
2003 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2005 return i == pos ? dev : NULL;
2008 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2010 read_lock(&dev_base_lock);
2011 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2014 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2017 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2020 void dev_seq_stop(struct seq_file *seq, void *v)
2022 read_unlock(&dev_base_lock);
2025 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2027 if (dev->get_stats) {
2028 struct net_device_stats *stats = dev->get_stats(dev);
2030 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2031 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2032 dev->name, stats->rx_bytes, stats->rx_packets,
2034 stats->rx_dropped + stats->rx_missed_errors,
2035 stats->rx_fifo_errors,
2036 stats->rx_length_errors + stats->rx_over_errors +
2037 stats->rx_crc_errors + stats->rx_frame_errors,
2038 stats->rx_compressed, stats->multicast,
2039 stats->tx_bytes, stats->tx_packets,
2040 stats->tx_errors, stats->tx_dropped,
2041 stats->tx_fifo_errors, stats->collisions,
2042 stats->tx_carrier_errors +
2043 stats->tx_aborted_errors +
2044 stats->tx_window_errors +
2045 stats->tx_heartbeat_errors,
2046 stats->tx_compressed);
2048 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2052 * Called from the PROCfs module. This now uses the new arbitrary sized
2053 * /proc/net interface to create /proc/net/dev
2055 static int dev_seq_show(struct seq_file *seq, void *v)
2057 if (v == SEQ_START_TOKEN)
2058 seq_puts(seq, "Inter-| Receive "
2060 " face |bytes packets errs drop fifo frame "
2061 "compressed multicast|bytes packets errs "
2062 "drop fifo colls carrier compressed\n");
2064 dev_seq_printf_stats(seq, v);
2068 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2070 struct netif_rx_stats *rc = NULL;
2072 while (*pos < NR_CPUS)
2073 if (cpu_online(*pos)) {
2074 rc = &per_cpu(netdev_rx_stat, *pos);
2081 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2083 return softnet_get_online(pos);
2086 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2089 return softnet_get_online(pos);
2092 static void softnet_seq_stop(struct seq_file *seq, void *v)
2096 static int softnet_seq_show(struct seq_file *seq, void *v)
2098 struct netif_rx_stats *s = v;
2100 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2101 s->total, s->dropped, s->time_squeeze, 0,
2102 0, 0, 0, 0, /* was fastroute */
2107 static struct seq_operations dev_seq_ops = {
2108 .start = dev_seq_start,
2109 .next = dev_seq_next,
2110 .stop = dev_seq_stop,
2111 .show = dev_seq_show,
2114 static int dev_seq_open(struct inode *inode, struct file *file)
2116 return seq_open(file, &dev_seq_ops);
2119 static struct file_operations dev_seq_fops = {
2120 .owner = THIS_MODULE,
2121 .open = dev_seq_open,
2123 .llseek = seq_lseek,
2124 .release = seq_release,
2127 static struct seq_operations softnet_seq_ops = {
2128 .start = softnet_seq_start,
2129 .next = softnet_seq_next,
2130 .stop = softnet_seq_stop,
2131 .show = softnet_seq_show,
2134 static int softnet_seq_open(struct inode *inode, struct file *file)
2136 return seq_open(file, &softnet_seq_ops);
2139 static struct file_operations softnet_seq_fops = {
2140 .owner = THIS_MODULE,
2141 .open = softnet_seq_open,
2143 .llseek = seq_lseek,
2144 .release = seq_release,
2147 #ifdef CONFIG_WIRELESS_EXT
2148 extern int wireless_proc_init(void);
2150 #define wireless_proc_init() 0
2153 static int __init dev_proc_init(void)
2157 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2159 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2161 if (wireless_proc_init())
2167 proc_net_remove("softnet_stat");
2169 proc_net_remove("dev");
2173 #define dev_proc_init() 0
2174 #endif /* CONFIG_PROC_FS */
2178 * netdev_set_master - set up master/slave pair
2179 * @slave: slave device
2180 * @master: new master device
2182 * Changes the master device of the slave. Pass %NULL to break the
2183 * bonding. The caller must hold the RTNL semaphore. On a failure
2184 * a negative errno code is returned. On success the reference counts
2185 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2186 * function returns zero.
2188 int netdev_set_master(struct net_device *slave, struct net_device *master)
2190 struct net_device *old = slave->master;
2200 slave->master = master;
2208 slave->flags |= IFF_SLAVE;
2210 slave->flags &= ~IFF_SLAVE;
2212 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2217 * dev_set_promiscuity - update promiscuity count on a device
2221 * Add or remove promiscuity from a device. While the count in the device
2222 * remains above zero the interface remains promiscuous. Once it hits zero
2223 * the device reverts back to normal filtering operation. A negative inc
2224 * value is used to drop promiscuity on the device.
2226 void dev_set_promiscuity(struct net_device *dev, int inc)
2228 unsigned short old_flags = dev->flags;
2230 if ((dev->promiscuity += inc) == 0)
2231 dev->flags &= ~IFF_PROMISC;
2233 dev->flags |= IFF_PROMISC;
2234 if (dev->flags != old_flags) {
2236 printk(KERN_INFO "device %s %s promiscuous mode\n",
2237 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2239 audit_log(current->audit_context, GFP_ATOMIC,
2240 AUDIT_ANOM_PROMISCUOUS,
2241 "dev=%s prom=%d old_prom=%d auid=%u",
2242 dev->name, (dev->flags & IFF_PROMISC),
2243 (old_flags & IFF_PROMISC),
2244 audit_get_loginuid(current->audit_context));
2249 * dev_set_allmulti - update allmulti count on a device
2253 * Add or remove reception of all multicast frames to a device. While the
2254 * count in the device remains above zero the interface remains listening
2255 * to all interfaces. Once it hits zero the device reverts back to normal
2256 * filtering operation. A negative @inc value is used to drop the counter
2257 * when releasing a resource needing all multicasts.
2260 void dev_set_allmulti(struct net_device *dev, int inc)
2262 unsigned short old_flags = dev->flags;
2264 dev->flags |= IFF_ALLMULTI;
2265 if ((dev->allmulti += inc) == 0)
2266 dev->flags &= ~IFF_ALLMULTI;
2267 if (dev->flags ^ old_flags)
2271 unsigned dev_get_flags(const struct net_device *dev)
2275 flags = (dev->flags & ~(IFF_PROMISC |
2280 (dev->gflags & (IFF_PROMISC |
2283 if (netif_running(dev)) {
2284 if (netif_oper_up(dev))
2285 flags |= IFF_RUNNING;
2286 if (netif_carrier_ok(dev))
2287 flags |= IFF_LOWER_UP;
2288 if (netif_dormant(dev))
2289 flags |= IFF_DORMANT;
2295 int dev_change_flags(struct net_device *dev, unsigned flags)
2298 int old_flags = dev->flags;
2301 * Set the flags on our device.
2304 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2305 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2307 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2311 * Load in the correct multicast list now the flags have changed.
2317 * Have we downed the interface. We handle IFF_UP ourselves
2318 * according to user attempts to set it, rather than blindly
2323 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2324 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2330 if (dev->flags & IFF_UP &&
2331 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2333 raw_notifier_call_chain(&netdev_chain,
2334 NETDEV_CHANGE, dev);
2336 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2337 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2338 dev->gflags ^= IFF_PROMISC;
2339 dev_set_promiscuity(dev, inc);
2342 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2343 is important. Some (broken) drivers set IFF_PROMISC, when
2344 IFF_ALLMULTI is requested not asking us and not reporting.
2346 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2347 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2348 dev->gflags ^= IFF_ALLMULTI;
2349 dev_set_allmulti(dev, inc);
2352 if (old_flags ^ dev->flags)
2353 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2358 int dev_set_mtu(struct net_device *dev, int new_mtu)
2362 if (new_mtu == dev->mtu)
2365 /* MTU must be positive. */
2369 if (!netif_device_present(dev))
2373 if (dev->change_mtu)
2374 err = dev->change_mtu(dev, new_mtu);
2377 if (!err && dev->flags & IFF_UP)
2378 raw_notifier_call_chain(&netdev_chain,
2379 NETDEV_CHANGEMTU, dev);
2383 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2387 if (!dev->set_mac_address)
2389 if (sa->sa_family != dev->type)
2391 if (!netif_device_present(dev))
2393 err = dev->set_mac_address(dev, sa);
2395 raw_notifier_call_chain(&netdev_chain,
2396 NETDEV_CHANGEADDR, dev);
2401 * Perform the SIOCxIFxxx calls.
2403 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2406 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2412 case SIOCGIFFLAGS: /* Get interface flags */
2413 ifr->ifr_flags = dev_get_flags(dev);
2416 case SIOCSIFFLAGS: /* Set interface flags */
2417 return dev_change_flags(dev, ifr->ifr_flags);
2419 case SIOCGIFMETRIC: /* Get the metric on the interface
2420 (currently unused) */
2421 ifr->ifr_metric = 0;
2424 case SIOCSIFMETRIC: /* Set the metric on the interface
2425 (currently unused) */
2428 case SIOCGIFMTU: /* Get the MTU of a device */
2429 ifr->ifr_mtu = dev->mtu;
2432 case SIOCSIFMTU: /* Set the MTU of a device */
2433 return dev_set_mtu(dev, ifr->ifr_mtu);
2437 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2439 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2440 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2441 ifr->ifr_hwaddr.sa_family = dev->type;
2445 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2447 case SIOCSIFHWBROADCAST:
2448 if (ifr->ifr_hwaddr.sa_family != dev->type)
2450 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2451 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2452 raw_notifier_call_chain(&netdev_chain,
2453 NETDEV_CHANGEADDR, dev);
2457 ifr->ifr_map.mem_start = dev->mem_start;
2458 ifr->ifr_map.mem_end = dev->mem_end;
2459 ifr->ifr_map.base_addr = dev->base_addr;
2460 ifr->ifr_map.irq = dev->irq;
2461 ifr->ifr_map.dma = dev->dma;
2462 ifr->ifr_map.port = dev->if_port;
2466 if (dev->set_config) {
2467 if (!netif_device_present(dev))
2469 return dev->set_config(dev, &ifr->ifr_map);
2474 if (!dev->set_multicast_list ||
2475 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2477 if (!netif_device_present(dev))
2479 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2483 if (!dev->set_multicast_list ||
2484 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2486 if (!netif_device_present(dev))
2488 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2492 ifr->ifr_ifindex = dev->ifindex;
2496 ifr->ifr_qlen = dev->tx_queue_len;
2500 if (ifr->ifr_qlen < 0)
2502 dev->tx_queue_len = ifr->ifr_qlen;
2506 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2507 return dev_change_name(dev, ifr->ifr_newname);
2510 * Unknown or private ioctl
2514 if ((cmd >= SIOCDEVPRIVATE &&
2515 cmd <= SIOCDEVPRIVATE + 15) ||
2516 cmd == SIOCBONDENSLAVE ||
2517 cmd == SIOCBONDRELEASE ||
2518 cmd == SIOCBONDSETHWADDR ||
2519 cmd == SIOCBONDSLAVEINFOQUERY ||
2520 cmd == SIOCBONDINFOQUERY ||
2521 cmd == SIOCBONDCHANGEACTIVE ||
2522 cmd == SIOCGMIIPHY ||
2523 cmd == SIOCGMIIREG ||
2524 cmd == SIOCSMIIREG ||
2525 cmd == SIOCBRADDIF ||
2526 cmd == SIOCBRDELIF ||
2527 cmd == SIOCWANDEV) {
2529 if (dev->do_ioctl) {
2530 if (netif_device_present(dev))
2531 err = dev->do_ioctl(dev, ifr,
2544 * This function handles all "interface"-type I/O control requests. The actual
2545 * 'doing' part of this is dev_ifsioc above.
2549 * dev_ioctl - network device ioctl
2550 * @cmd: command to issue
2551 * @arg: pointer to a struct ifreq in user space
2553 * Issue ioctl functions to devices. This is normally called by the
2554 * user space syscall interfaces but can sometimes be useful for
2555 * other purposes. The return value is the return from the syscall if
2556 * positive or a negative errno code on error.
2559 int dev_ioctl(unsigned int cmd, void __user *arg)
2565 /* One special case: SIOCGIFCONF takes ifconf argument
2566 and requires shared lock, because it sleeps writing
2570 if (cmd == SIOCGIFCONF) {
2572 ret = dev_ifconf((char __user *) arg);
2576 if (cmd == SIOCGIFNAME)
2577 return dev_ifname((struct ifreq __user *)arg);
2579 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2582 ifr.ifr_name[IFNAMSIZ-1] = 0;
2584 colon = strchr(ifr.ifr_name, ':');
2589 * See which interface the caller is talking about.
2594 * These ioctl calls:
2595 * - can be done by all.
2596 * - atomic and do not require locking.
2607 dev_load(ifr.ifr_name);
2608 read_lock(&dev_base_lock);
2609 ret = dev_ifsioc(&ifr, cmd);
2610 read_unlock(&dev_base_lock);
2614 if (copy_to_user(arg, &ifr,
2615 sizeof(struct ifreq)))
2621 dev_load(ifr.ifr_name);
2623 ret = dev_ethtool(&ifr);
2628 if (copy_to_user(arg, &ifr,
2629 sizeof(struct ifreq)))
2635 * These ioctl calls:
2636 * - require superuser power.
2637 * - require strict serialization.
2643 if (!capable(CAP_NET_ADMIN))
2645 dev_load(ifr.ifr_name);
2647 ret = dev_ifsioc(&ifr, cmd);
2652 if (copy_to_user(arg, &ifr,
2653 sizeof(struct ifreq)))
2659 * These ioctl calls:
2660 * - require superuser power.
2661 * - require strict serialization.
2662 * - do not return a value
2672 case SIOCSIFHWBROADCAST:
2675 case SIOCBONDENSLAVE:
2676 case SIOCBONDRELEASE:
2677 case SIOCBONDSETHWADDR:
2678 case SIOCBONDCHANGEACTIVE:
2681 if (!capable(CAP_NET_ADMIN))
2684 case SIOCBONDSLAVEINFOQUERY:
2685 case SIOCBONDINFOQUERY:
2686 dev_load(ifr.ifr_name);
2688 ret = dev_ifsioc(&ifr, cmd);
2693 /* Get the per device memory space. We can add this but
2694 * currently do not support it */
2696 /* Set the per device memory buffer space.
2697 * Not applicable in our case */
2702 * Unknown or private ioctl.
2705 if (cmd == SIOCWANDEV ||
2706 (cmd >= SIOCDEVPRIVATE &&
2707 cmd <= SIOCDEVPRIVATE + 15)) {
2708 dev_load(ifr.ifr_name);
2710 ret = dev_ifsioc(&ifr, cmd);
2712 if (!ret && copy_to_user(arg, &ifr,
2713 sizeof(struct ifreq)))
2717 #ifdef CONFIG_WIRELESS_EXT
2718 /* Take care of Wireless Extensions */
2719 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2720 /* If command is `set a parameter', or
2721 * `get the encoding parameters', check if
2722 * the user has the right to do it */
2723 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2724 || cmd == SIOCGIWENCODEEXT) {
2725 if (!capable(CAP_NET_ADMIN))
2728 dev_load(ifr.ifr_name);
2730 /* Follow me in net/core/wireless.c */
2731 ret = wireless_process_ioctl(&ifr, cmd);
2733 if (IW_IS_GET(cmd) &&
2734 copy_to_user(arg, &ifr,
2735 sizeof(struct ifreq)))
2739 #endif /* CONFIG_WIRELESS_EXT */
2746 * dev_new_index - allocate an ifindex
2748 * Returns a suitable unique value for a new device interface
2749 * number. The caller must hold the rtnl semaphore or the
2750 * dev_base_lock to be sure it remains unique.
2752 static int dev_new_index(void)
2758 if (!__dev_get_by_index(ifindex))
2763 static int dev_boot_phase = 1;
2765 /* Delayed registration/unregisteration */
2766 static DEFINE_SPINLOCK(net_todo_list_lock);
2767 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2769 static inline void net_set_todo(struct net_device *dev)
2771 spin_lock(&net_todo_list_lock);
2772 list_add_tail(&dev->todo_list, &net_todo_list);
2773 spin_unlock(&net_todo_list_lock);
2777 * register_netdevice - register a network device
2778 * @dev: device to register
2780 * Take a completed network device structure and add it to the kernel
2781 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2782 * chain. 0 is returned on success. A negative errno code is returned
2783 * on a failure to set up the device, or if the name is a duplicate.
2785 * Callers must hold the rtnl semaphore. You may want
2786 * register_netdev() instead of this.
2789 * The locking appears insufficient to guarantee two parallel registers
2790 * will not get the same name.
2793 int register_netdevice(struct net_device *dev)
2795 struct hlist_head *head;
2796 struct hlist_node *p;
2799 BUG_ON(dev_boot_phase);
2804 /* When net_device's are persistent, this will be fatal. */
2805 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2807 spin_lock_init(&dev->queue_lock);
2808 spin_lock_init(&dev->xmit_lock);
2809 dev->xmit_lock_owner = -1;
2810 #ifdef CONFIG_NET_CLS_ACT
2811 spin_lock_init(&dev->ingress_lock);
2814 ret = alloc_divert_blk(dev);
2820 /* Init, if this function is available */
2822 ret = dev->init(dev);
2830 if (!dev_valid_name(dev->name)) {
2835 dev->ifindex = dev_new_index();
2836 if (dev->iflink == -1)
2837 dev->iflink = dev->ifindex;
2839 /* Check for existence of name */
2840 head = dev_name_hash(dev->name);
2841 hlist_for_each(p, head) {
2842 struct net_device *d
2843 = hlist_entry(p, struct net_device, name_hlist);
2844 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2850 /* Fix illegal SG+CSUM combinations. */
2851 if ((dev->features & NETIF_F_SG) &&
2852 !(dev->features & (NETIF_F_IP_CSUM |
2854 NETIF_F_HW_CSUM))) {
2855 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2857 dev->features &= ~NETIF_F_SG;
2860 /* TSO requires that SG is present as well. */
2861 if ((dev->features & NETIF_F_TSO) &&
2862 !(dev->features & NETIF_F_SG)) {
2863 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2865 dev->features &= ~NETIF_F_TSO;
2867 if (dev->features & NETIF_F_UFO) {
2868 if (!(dev->features & NETIF_F_HW_CSUM)) {
2869 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2870 "NETIF_F_HW_CSUM feature.\n",
2872 dev->features &= ~NETIF_F_UFO;
2874 if (!(dev->features & NETIF_F_SG)) {
2875 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2876 "NETIF_F_SG feature.\n",
2878 dev->features &= ~NETIF_F_UFO;
2883 * nil rebuild_header routine,
2884 * that should be never called and used as just bug trap.
2887 if (!dev->rebuild_header)
2888 dev->rebuild_header = default_rebuild_header;
2890 ret = netdev_register_sysfs(dev);
2893 dev->reg_state = NETREG_REGISTERED;
2896 * Default initial state at registry is that the
2897 * device is present.
2900 set_bit(__LINK_STATE_PRESENT, &dev->state);
2903 dev_init_scheduler(dev);
2904 write_lock_bh(&dev_base_lock);
2906 dev_tail = &dev->next;
2907 hlist_add_head(&dev->name_hlist, head);
2908 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2910 write_unlock_bh(&dev_base_lock);
2912 /* Notify protocols, that a new device appeared. */
2913 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2920 free_divert_blk(dev);
2925 * register_netdev - register a network device
2926 * @dev: device to register
2928 * Take a completed network device structure and add it to the kernel
2929 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2930 * chain. 0 is returned on success. A negative errno code is returned
2931 * on a failure to set up the device, or if the name is a duplicate.
2933 * This is a wrapper around register_netdev that takes the rtnl semaphore
2934 * and expands the device name if you passed a format string to
2937 int register_netdev(struct net_device *dev)
2944 * If the name is a format string the caller wants us to do a
2947 if (strchr(dev->name, '%')) {
2948 err = dev_alloc_name(dev, dev->name);
2954 * Back compatibility hook. Kill this one in 2.5
2956 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2957 err = dev_alloc_name(dev, "eth%d");
2962 err = register_netdevice(dev);
2967 EXPORT_SYMBOL(register_netdev);
2970 * netdev_wait_allrefs - wait until all references are gone.
2972 * This is called when unregistering network devices.
2974 * Any protocol or device that holds a reference should register
2975 * for netdevice notification, and cleanup and put back the
2976 * reference if they receive an UNREGISTER event.
2977 * We can get stuck here if buggy protocols don't correctly
2980 static void netdev_wait_allrefs(struct net_device *dev)
2982 unsigned long rebroadcast_time, warning_time;
2984 rebroadcast_time = warning_time = jiffies;
2985 while (atomic_read(&dev->refcnt) != 0) {
2986 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2989 /* Rebroadcast unregister notification */
2990 raw_notifier_call_chain(&netdev_chain,
2991 NETDEV_UNREGISTER, dev);
2993 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2995 /* We must not have linkwatch events
2996 * pending on unregister. If this
2997 * happens, we simply run the queue
2998 * unscheduled, resulting in a noop
3001 linkwatch_run_queue();
3006 rebroadcast_time = jiffies;
3011 if (time_after(jiffies, warning_time + 10 * HZ)) {
3012 printk(KERN_EMERG "unregister_netdevice: "
3013 "waiting for %s to become free. Usage "
3015 dev->name, atomic_read(&dev->refcnt));
3016 warning_time = jiffies;
3025 * register_netdevice(x1);
3026 * register_netdevice(x2);
3028 * unregister_netdevice(y1);
3029 * unregister_netdevice(y2);
3035 * We are invoked by rtnl_unlock() after it drops the semaphore.
3036 * This allows us to deal with problems:
3037 * 1) We can delete sysfs objects which invoke hotplug
3038 * without deadlocking with linkwatch via keventd.
3039 * 2) Since we run with the RTNL semaphore not held, we can sleep
3040 * safely in order to wait for the netdev refcnt to drop to zero.
3042 static DEFINE_MUTEX(net_todo_run_mutex);
3043 void netdev_run_todo(void)
3045 struct list_head list = LIST_HEAD_INIT(list);
3047 /* Need to guard against multiple cpu's getting out of order. */
3048 mutex_lock(&net_todo_run_mutex);
3050 /* Not safe to do outside the semaphore. We must not return
3051 * until all unregister events invoked by the local processor
3052 * have been completed (either by this todo run, or one on
3055 if (list_empty(&net_todo_list))
3058 /* Snapshot list, allow later requests */
3059 spin_lock(&net_todo_list_lock);
3060 list_splice_init(&net_todo_list, &list);
3061 spin_unlock(&net_todo_list_lock);
3063 while (!list_empty(&list)) {
3064 struct net_device *dev
3065 = list_entry(list.next, struct net_device, todo_list);
3066 list_del(&dev->todo_list);
3068 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3069 printk(KERN_ERR "network todo '%s' but state %d\n",
3070 dev->name, dev->reg_state);
3075 netdev_unregister_sysfs(dev);
3076 dev->reg_state = NETREG_UNREGISTERED;
3078 netdev_wait_allrefs(dev);
3081 BUG_ON(atomic_read(&dev->refcnt));
3082 BUG_TRAP(!dev->ip_ptr);
3083 BUG_TRAP(!dev->ip6_ptr);
3084 BUG_TRAP(!dev->dn_ptr);
3086 /* It must be the very last action,
3087 * after this 'dev' may point to freed up memory.
3089 if (dev->destructor)
3090 dev->destructor(dev);
3094 mutex_unlock(&net_todo_run_mutex);
3098 * alloc_netdev - allocate network device
3099 * @sizeof_priv: size of private data to allocate space for
3100 * @name: device name format string
3101 * @setup: callback to initialize device
3103 * Allocates a struct net_device with private data area for driver use
3104 * and performs basic initialization.
3106 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3107 void (*setup)(struct net_device *))
3110 struct net_device *dev;
3113 /* ensure 32-byte alignment of both the device and private area */
3114 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3115 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3117 p = kzalloc(alloc_size, GFP_KERNEL);
3119 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3123 dev = (struct net_device *)
3124 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3125 dev->padded = (char *)dev - (char *)p;
3128 dev->priv = netdev_priv(dev);
3131 strcpy(dev->name, name);
3134 EXPORT_SYMBOL(alloc_netdev);
3137 * free_netdev - free network device
3140 * This function does the last stage of destroying an allocated device
3141 * interface. The reference to the device object is released.
3142 * If this is the last reference then it will be freed.
3144 void free_netdev(struct net_device *dev)
3147 /* Compatibility with error handling in drivers */
3148 if (dev->reg_state == NETREG_UNINITIALIZED) {
3149 kfree((char *)dev - dev->padded);
3153 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3154 dev->reg_state = NETREG_RELEASED;
3156 /* will free via class release */
3157 class_device_put(&dev->class_dev);
3159 kfree((char *)dev - dev->padded);
3163 /* Synchronize with packet receive processing. */
3164 void synchronize_net(void)
3171 * unregister_netdevice - remove device from the kernel
3174 * This function shuts down a device interface and removes it
3175 * from the kernel tables. On success 0 is returned, on a failure
3176 * a negative errno code is returned.
3178 * Callers must hold the rtnl semaphore. You may want
3179 * unregister_netdev() instead of this.
3182 int unregister_netdevice(struct net_device *dev)
3184 struct net_device *d, **dp;
3186 BUG_ON(dev_boot_phase);
3189 /* Some devices call without registering for initialization unwind. */
3190 if (dev->reg_state == NETREG_UNINITIALIZED) {
3191 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3192 "was registered\n", dev->name, dev);
3196 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3198 /* If device is running, close it first. */
3199 if (dev->flags & IFF_UP)
3202 /* And unlink it from device chain. */
3203 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3205 write_lock_bh(&dev_base_lock);
3206 hlist_del(&dev->name_hlist);
3207 hlist_del(&dev->index_hlist);
3208 if (dev_tail == &dev->next)
3211 write_unlock_bh(&dev_base_lock);
3216 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3221 dev->reg_state = NETREG_UNREGISTERING;
3225 /* Shutdown queueing discipline. */
3229 /* Notify protocols, that we are about to destroy
3230 this device. They should clean all the things.
3232 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3235 * Flush the multicast chain
3237 dev_mc_discard(dev);
3242 /* Notifier chain MUST detach us from master device. */
3243 BUG_TRAP(!dev->master);
3245 free_divert_blk(dev);
3247 /* Finish processing unregister after unlock */
3257 * unregister_netdev - remove device from the kernel
3260 * This function shuts down a device interface and removes it
3261 * from the kernel tables. On success 0 is returned, on a failure
3262 * a negative errno code is returned.
3264 * This is just a wrapper for unregister_netdevice that takes
3265 * the rtnl semaphore. In general you want to use this and not
3266 * unregister_netdevice.
3268 void unregister_netdev(struct net_device *dev)
3271 unregister_netdevice(dev);
3275 EXPORT_SYMBOL(unregister_netdev);
3277 #ifdef CONFIG_HOTPLUG_CPU
3278 static int dev_cpu_callback(struct notifier_block *nfb,
3279 unsigned long action,
3282 struct sk_buff **list_skb;
3283 struct net_device **list_net;
3284 struct sk_buff *skb;
3285 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3286 struct softnet_data *sd, *oldsd;
3288 if (action != CPU_DEAD)
3291 local_irq_disable();
3292 cpu = smp_processor_id();
3293 sd = &per_cpu(softnet_data, cpu);
3294 oldsd = &per_cpu(softnet_data, oldcpu);
3296 /* Find end of our completion_queue. */
3297 list_skb = &sd->completion_queue;
3299 list_skb = &(*list_skb)->next;
3300 /* Append completion queue from offline CPU. */
3301 *list_skb = oldsd->completion_queue;
3302 oldsd->completion_queue = NULL;
3304 /* Find end of our output_queue. */
3305 list_net = &sd->output_queue;
3307 list_net = &(*list_net)->next_sched;
3308 /* Append output queue from offline CPU. */
3309 *list_net = oldsd->output_queue;
3310 oldsd->output_queue = NULL;
3312 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3315 /* Process offline CPU's input_pkt_queue */
3316 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3321 #endif /* CONFIG_HOTPLUG_CPU */
3323 #ifdef CONFIG_NET_DMA
3325 * net_dma_rebalance -
3326 * This is called when the number of channels allocated to the net_dma_client
3327 * changes. The net_dma_client tries to have one DMA channel per CPU.
3329 static void net_dma_rebalance(void)
3331 unsigned int cpu, i, n;
3332 struct dma_chan *chan;
3336 if (net_dma_count == 0) {
3337 for_each_online_cpu(cpu)
3338 rcu_assign_pointer(per_cpu(softnet_data.net_dma, cpu), NULL);
3339 unlock_cpu_hotplug();
3344 cpu = first_cpu(cpu_online_map);
3347 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3348 n = ((num_online_cpus() / net_dma_count)
3349 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3352 per_cpu(softnet_data.net_dma, cpu) = chan;
3353 cpu = next_cpu(cpu, cpu_online_map);
3360 unlock_cpu_hotplug();
3364 * netdev_dma_event - event callback for the net_dma_client
3365 * @client: should always be net_dma_client
3369 static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3370 enum dma_event event)
3372 spin_lock(&net_dma_event_lock);
3374 case DMA_RESOURCE_ADDED:
3376 net_dma_rebalance();
3378 case DMA_RESOURCE_REMOVED:
3380 net_dma_rebalance();
3385 spin_unlock(&net_dma_event_lock);
3389 * netdev_dma_regiser - register the networking subsystem as a DMA client
3391 static int __init netdev_dma_register(void)
3393 spin_lock_init(&net_dma_event_lock);
3394 net_dma_client = dma_async_client_register(netdev_dma_event);
3395 if (net_dma_client == NULL)
3398 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3403 static int __init netdev_dma_register(void) { return -ENODEV; }
3404 #endif /* CONFIG_NET_DMA */
3407 * Initialize the DEV module. At boot time this walks the device list and
3408 * unhooks any devices that fail to initialise (normally hardware not
3409 * present) and leaves us with a valid list of present and active devices.
3414 * This is called single threaded during boot, so no need
3415 * to take the rtnl semaphore.
3417 static int __init net_dev_init(void)
3419 int i, rc = -ENOMEM;
3421 BUG_ON(!dev_boot_phase);
3425 if (dev_proc_init())
3428 if (netdev_sysfs_init())
3431 INIT_LIST_HEAD(&ptype_all);
3432 for (i = 0; i < 16; i++)
3433 INIT_LIST_HEAD(&ptype_base[i]);
3435 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3436 INIT_HLIST_HEAD(&dev_name_head[i]);
3438 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3439 INIT_HLIST_HEAD(&dev_index_head[i]);
3442 * Initialise the packet receive queues.
3445 for_each_possible_cpu(i) {
3446 struct softnet_data *queue;
3448 queue = &per_cpu(softnet_data, i);
3449 skb_queue_head_init(&queue->input_pkt_queue);
3450 queue->completion_queue = NULL;
3451 INIT_LIST_HEAD(&queue->poll_list);
3452 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3453 queue->backlog_dev.weight = weight_p;
3454 queue->backlog_dev.poll = process_backlog;
3455 atomic_set(&queue->backlog_dev.refcnt, 1);
3458 netdev_dma_register();
3462 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3463 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3465 hotcpu_notifier(dev_cpu_callback, 0);
3473 subsys_initcall(net_dev_init);
3475 EXPORT_SYMBOL(__dev_get_by_index);
3476 EXPORT_SYMBOL(__dev_get_by_name);
3477 EXPORT_SYMBOL(__dev_remove_pack);
3478 EXPORT_SYMBOL(__skb_linearize);
3479 EXPORT_SYMBOL(dev_valid_name);
3480 EXPORT_SYMBOL(dev_add_pack);
3481 EXPORT_SYMBOL(dev_alloc_name);
3482 EXPORT_SYMBOL(dev_close);
3483 EXPORT_SYMBOL(dev_get_by_flags);
3484 EXPORT_SYMBOL(dev_get_by_index);
3485 EXPORT_SYMBOL(dev_get_by_name);
3486 EXPORT_SYMBOL(dev_open);
3487 EXPORT_SYMBOL(dev_queue_xmit);
3488 EXPORT_SYMBOL(dev_remove_pack);
3489 EXPORT_SYMBOL(dev_set_allmulti);
3490 EXPORT_SYMBOL(dev_set_promiscuity);
3491 EXPORT_SYMBOL(dev_change_flags);
3492 EXPORT_SYMBOL(dev_set_mtu);
3493 EXPORT_SYMBOL(dev_set_mac_address);
3494 EXPORT_SYMBOL(free_netdev);
3495 EXPORT_SYMBOL(netdev_boot_setup_check);
3496 EXPORT_SYMBOL(netdev_set_master);
3497 EXPORT_SYMBOL(netdev_state_change);
3498 EXPORT_SYMBOL(netif_receive_skb);
3499 EXPORT_SYMBOL(netif_rx);
3500 EXPORT_SYMBOL(register_gifconf);
3501 EXPORT_SYMBOL(register_netdevice);
3502 EXPORT_SYMBOL(register_netdevice_notifier);
3503 EXPORT_SYMBOL(skb_checksum_help);
3504 EXPORT_SYMBOL(synchronize_net);
3505 EXPORT_SYMBOL(unregister_netdevice);
3506 EXPORT_SYMBOL(unregister_netdevice_notifier);
3507 EXPORT_SYMBOL(net_enable_timestamp);
3508 EXPORT_SYMBOL(net_disable_timestamp);
3509 EXPORT_SYMBOL(dev_get_flags);
3511 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3512 EXPORT_SYMBOL(br_handle_frame_hook);
3513 EXPORT_SYMBOL(br_fdb_get_hook);
3514 EXPORT_SYMBOL(br_fdb_put_hook);
3518 EXPORT_SYMBOL(dev_load);
3521 EXPORT_PER_CPU_SYMBOL(softnet_data);