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/config.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/notifier.h>
93 #include <linux/skbuff.h>
95 #include <linux/rtnetlink.h>
96 #include <linux/proc_fs.h>
97 #include <linux/seq_file.h>
98 #include <linux/stat.h>
99 #include <linux/if_bridge.h>
100 #include <linux/divert.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #ifdef CONFIG_NET_RADIO
113 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
114 #include <net/iw_handler.h>
115 #endif /* CONFIG_NET_RADIO */
116 #include <asm/current.h>
119 * The list of packet types we will receive (as opposed to discard)
120 * and the routines to invoke.
122 * Why 16. Because with 16 the only overlap we get on a hash of the
123 * low nibble of the protocol value is RARP/SNAP/X.25.
125 * NOTE: That is no longer true with the addition of VLAN tags. Not
126 * sure which should go first, but I bet it won't make much
127 * difference if we are running VLANs. The good news is that
128 * this protocol won't be in the list unless compiled in, so
129 * the average user (w/out VLANs) will not be adversly affected.
146 static DEFINE_SPINLOCK(ptype_lock);
147 static struct list_head ptype_base[16]; /* 16 way hashed list */
148 static struct list_head ptype_all; /* Taps */
151 * The @dev_base list is protected by @dev_base_lock and the rtln
154 * Pure readers hold dev_base_lock for reading.
156 * Writers must hold the rtnl semaphore while they loop through the
157 * dev_base list, and hold dev_base_lock for writing when they do the
158 * actual updates. This allows pure readers to access the list even
159 * while a writer is preparing to update it.
161 * To put it another way, dev_base_lock is held for writing only to
162 * protect against pure readers; the rtnl semaphore provides the
163 * protection against other writers.
165 * See, for example usages, register_netdevice() and
166 * unregister_netdevice(), which must be called with the rtnl
169 struct net_device *dev_base;
170 static struct net_device **dev_tail = &dev_base;
171 DEFINE_RWLOCK(dev_base_lock);
173 EXPORT_SYMBOL(dev_base);
174 EXPORT_SYMBOL(dev_base_lock);
176 #define NETDEV_HASHBITS 8
177 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
178 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
180 static inline struct hlist_head *dev_name_hash(const char *name)
182 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
183 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
186 static inline struct hlist_head *dev_index_hash(int ifindex)
188 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
195 static struct notifier_block *netdev_chain;
198 * Device drivers call our routines to queue packets here. We empty the
199 * queue in the local softnet handler.
201 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
204 extern int netdev_sysfs_init(void);
205 extern int netdev_register_sysfs(struct net_device *);
206 extern void netdev_unregister_sysfs(struct net_device *);
208 #define netdev_sysfs_init() (0)
209 #define netdev_register_sysfs(dev) (0)
210 #define netdev_unregister_sysfs(dev) do { } while(0)
214 /*******************************************************************************
216 Protocol management and registration routines
218 *******************************************************************************/
227 * Add a protocol ID to the list. Now that the input handler is
228 * smarter we can dispense with all the messy stuff that used to be
231 * BEWARE!!! Protocol handlers, mangling input packets,
232 * MUST BE last in hash buckets and checking protocol handlers
233 * MUST start from promiscuous ptype_all chain in net_bh.
234 * It is true now, do not change it.
235 * Explanation follows: if protocol handler, mangling packet, will
236 * be the first on list, it is not able to sense, that packet
237 * is cloned and should be copied-on-write, so that it will
238 * change it and subsequent readers will get broken packet.
243 * dev_add_pack - add packet handler
244 * @pt: packet type declaration
246 * Add a protocol handler to the networking stack. The passed &packet_type
247 * is linked into kernel lists and may not be freed until it has been
248 * removed from the kernel lists.
250 * This call does not sleep therefore it can not
251 * guarantee all CPU's that are in middle of receiving packets
252 * will see the new packet type (until the next received packet).
255 void dev_add_pack(struct packet_type *pt)
259 spin_lock_bh(&ptype_lock);
260 if (pt->type == htons(ETH_P_ALL)) {
262 list_add_rcu(&pt->list, &ptype_all);
264 hash = ntohs(pt->type) & 15;
265 list_add_rcu(&pt->list, &ptype_base[hash]);
267 spin_unlock_bh(&ptype_lock);
270 extern void linkwatch_run_queue(void);
275 * __dev_remove_pack - remove packet handler
276 * @pt: packet type declaration
278 * Remove a protocol handler that was previously added to the kernel
279 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
280 * from the kernel lists and can be freed or reused once this function
283 * The packet type might still be in use by receivers
284 * and must not be freed until after all the CPU's have gone
285 * through a quiescent state.
287 void __dev_remove_pack(struct packet_type *pt)
289 struct list_head *head;
290 struct packet_type *pt1;
292 spin_lock_bh(&ptype_lock);
294 if (pt->type == htons(ETH_P_ALL)) {
298 head = &ptype_base[ntohs(pt->type) & 15];
300 list_for_each_entry(pt1, head, list) {
302 list_del_rcu(&pt->list);
307 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
309 spin_unlock_bh(&ptype_lock);
312 * dev_remove_pack - remove packet handler
313 * @pt: packet type declaration
315 * Remove a protocol handler that was previously added to the kernel
316 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
317 * from the kernel lists and can be freed or reused once this function
320 * This call sleeps to guarantee that no CPU is looking at the packet
323 void dev_remove_pack(struct packet_type *pt)
325 __dev_remove_pack(pt);
330 /******************************************************************************
332 Device Boot-time Settings Routines
334 *******************************************************************************/
336 /* Boot time configuration table */
337 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
340 * netdev_boot_setup_add - add new setup entry
341 * @name: name of the device
342 * @map: configured settings for the device
344 * Adds new setup entry to the dev_boot_setup list. The function
345 * returns 0 on error and 1 on success. This is a generic routine to
348 static int netdev_boot_setup_add(char *name, struct ifmap *map)
350 struct netdev_boot_setup *s;
354 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
355 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
356 memset(s[i].name, 0, sizeof(s[i].name));
357 strcpy(s[i].name, name);
358 memcpy(&s[i].map, map, sizeof(s[i].map));
363 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
367 * netdev_boot_setup_check - check boot time settings
368 * @dev: the netdevice
370 * Check boot time settings for the device.
371 * The found settings are set for the device to be used
372 * later in the device probing.
373 * Returns 0 if no settings found, 1 if they are.
375 int netdev_boot_setup_check(struct net_device *dev)
377 struct netdev_boot_setup *s = dev_boot_setup;
380 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
381 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
382 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
383 dev->irq = s[i].map.irq;
384 dev->base_addr = s[i].map.base_addr;
385 dev->mem_start = s[i].map.mem_start;
386 dev->mem_end = s[i].map.mem_end;
395 * netdev_boot_base - get address from boot time settings
396 * @prefix: prefix for network device
397 * @unit: id for network device
399 * Check boot time settings for the base address of device.
400 * The found settings are set for the device to be used
401 * later in the device probing.
402 * Returns 0 if no settings found.
404 unsigned long netdev_boot_base(const char *prefix, int unit)
406 const struct netdev_boot_setup *s = dev_boot_setup;
410 sprintf(name, "%s%d", prefix, unit);
413 * If device already registered then return base of 1
414 * to indicate not to probe for this interface
416 if (__dev_get_by_name(name))
419 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
420 if (!strcmp(name, s[i].name))
421 return s[i].map.base_addr;
426 * Saves at boot time configured settings for any netdevice.
428 int __init netdev_boot_setup(char *str)
433 str = get_options(str, ARRAY_SIZE(ints), ints);
438 memset(&map, 0, sizeof(map));
442 map.base_addr = ints[2];
444 map.mem_start = ints[3];
446 map.mem_end = ints[4];
448 /* Add new entry to the list */
449 return netdev_boot_setup_add(str, &map);
452 __setup("netdev=", netdev_boot_setup);
454 /*******************************************************************************
456 Device Interface Subroutines
458 *******************************************************************************/
461 * __dev_get_by_name - find a device by its name
462 * @name: name to find
464 * Find an interface by name. Must be called under RTNL semaphore
465 * or @dev_base_lock. If the name is found a pointer to the device
466 * is returned. If the name is not found then %NULL is returned. The
467 * reference counters are not incremented so the caller must be
468 * careful with locks.
471 struct net_device *__dev_get_by_name(const char *name)
473 struct hlist_node *p;
475 hlist_for_each(p, dev_name_hash(name)) {
476 struct net_device *dev
477 = hlist_entry(p, struct net_device, name_hlist);
478 if (!strncmp(dev->name, name, IFNAMSIZ))
485 * dev_get_by_name - find a device by its name
486 * @name: name to find
488 * Find an interface by name. This can be called from any
489 * context and does its own locking. The returned handle has
490 * the usage count incremented and the caller must use dev_put() to
491 * release it when it is no longer needed. %NULL is returned if no
492 * matching device is found.
495 struct net_device *dev_get_by_name(const char *name)
497 struct net_device *dev;
499 read_lock(&dev_base_lock);
500 dev = __dev_get_by_name(name);
503 read_unlock(&dev_base_lock);
508 * __dev_get_by_index - find a device by its ifindex
509 * @ifindex: index of device
511 * Search for an interface by index. Returns %NULL if the device
512 * is not found or a pointer to the device. The device has not
513 * had its reference counter increased so the caller must be careful
514 * about locking. The caller must hold either the RTNL semaphore
518 struct net_device *__dev_get_by_index(int ifindex)
520 struct hlist_node *p;
522 hlist_for_each(p, dev_index_hash(ifindex)) {
523 struct net_device *dev
524 = hlist_entry(p, struct net_device, index_hlist);
525 if (dev->ifindex == ifindex)
533 * dev_get_by_index - find a device by its ifindex
534 * @ifindex: index of device
536 * Search for an interface by index. Returns NULL if the device
537 * is not found or a pointer to the device. The device returned has
538 * had a reference added and the pointer is safe until the user calls
539 * dev_put to indicate they have finished with it.
542 struct net_device *dev_get_by_index(int ifindex)
544 struct net_device *dev;
546 read_lock(&dev_base_lock);
547 dev = __dev_get_by_index(ifindex);
550 read_unlock(&dev_base_lock);
555 * dev_getbyhwaddr - find a device by its hardware address
556 * @type: media type of device
557 * @ha: hardware address
559 * Search for an interface by MAC address. Returns NULL if the device
560 * is not found or a pointer to the device. The caller must hold the
561 * rtnl semaphore. The returned device has not had its ref count increased
562 * and the caller must therefore be careful about locking
565 * If the API was consistent this would be __dev_get_by_hwaddr
568 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
570 struct net_device *dev;
574 for (dev = dev_base; dev; dev = dev->next)
575 if (dev->type == type &&
576 !memcmp(dev->dev_addr, ha, dev->addr_len))
581 struct net_device *dev_getfirstbyhwtype(unsigned short type)
583 struct net_device *dev;
586 for (dev = dev_base; dev; dev = dev->next) {
587 if (dev->type == type) {
596 EXPORT_SYMBOL(dev_getfirstbyhwtype);
599 * dev_get_by_flags - find any device with given flags
600 * @if_flags: IFF_* values
601 * @mask: bitmask of bits in if_flags to check
603 * Search for any interface with the given flags. Returns NULL if a device
604 * is not found or a pointer to the device. The device returned has
605 * had a reference added and the pointer is safe until the user calls
606 * dev_put to indicate they have finished with it.
609 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
611 struct net_device *dev;
613 read_lock(&dev_base_lock);
614 for (dev = dev_base; dev != NULL; dev = dev->next) {
615 if (((dev->flags ^ if_flags) & mask) == 0) {
620 read_unlock(&dev_base_lock);
625 * dev_valid_name - check if name is okay for network device
628 * Network device names need to be valid file names to
629 * to allow sysfs to work
631 static int dev_valid_name(const char *name)
633 return !(*name == '\0'
634 || !strcmp(name, ".")
635 || !strcmp(name, "..")
636 || strchr(name, '/'));
640 * dev_alloc_name - allocate a name for a device
642 * @name: name format string
644 * Passed a format string - eg "lt%d" it will try and find a suitable
645 * id. Not efficient for many devices, not called a lot. The caller
646 * must hold the dev_base or rtnl lock while allocating the name and
647 * adding the device in order to avoid duplicates. Returns the number
648 * of the unit assigned or a negative errno code.
651 int dev_alloc_name(struct net_device *dev, const char *name)
656 const int max_netdevices = 8*PAGE_SIZE;
658 struct net_device *d;
660 p = strnchr(name, IFNAMSIZ-1, '%');
663 * Verify the string as this thing may have come from
664 * the user. There must be either one "%d" and no other "%"
667 if (p[1] != 'd' || strchr(p + 2, '%'))
670 /* Use one page as a bit array of possible slots */
671 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
675 for (d = dev_base; d; d = d->next) {
676 if (!sscanf(d->name, name, &i))
678 if (i < 0 || i >= max_netdevices)
681 /* avoid cases where sscanf is not exact inverse of printf */
682 snprintf(buf, sizeof(buf), name, i);
683 if (!strncmp(buf, d->name, IFNAMSIZ))
687 i = find_first_zero_bit(inuse, max_netdevices);
688 free_page((unsigned long) inuse);
691 snprintf(buf, sizeof(buf), name, i);
692 if (!__dev_get_by_name(buf)) {
693 strlcpy(dev->name, buf, IFNAMSIZ);
697 /* It is possible to run out of possible slots
698 * when the name is long and there isn't enough space left
699 * for the digits, or if all bits are used.
706 * dev_change_name - change name of a device
708 * @newname: name (or format string) must be at least IFNAMSIZ
710 * Change name of a device, can pass format strings "eth%d".
713 int dev_change_name(struct net_device *dev, char *newname)
719 if (dev->flags & IFF_UP)
722 if (!dev_valid_name(newname))
725 if (strchr(newname, '%')) {
726 err = dev_alloc_name(dev, newname);
729 strcpy(newname, dev->name);
731 else if (__dev_get_by_name(newname))
734 strlcpy(dev->name, newname, IFNAMSIZ);
736 err = class_device_rename(&dev->class_dev, dev->name);
738 hlist_del(&dev->name_hlist);
739 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
740 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
747 * netdev_features_change - device changes fatures
748 * @dev: device to cause notification
750 * Called to indicate a device has changed features.
752 void netdev_features_change(struct net_device *dev)
754 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
756 EXPORT_SYMBOL(netdev_features_change);
759 * netdev_state_change - device changes state
760 * @dev: device to cause notification
762 * Called to indicate a device has changed state. This function calls
763 * the notifier chains for netdev_chain and sends a NEWLINK message
764 * to the routing socket.
766 void netdev_state_change(struct net_device *dev)
768 if (dev->flags & IFF_UP) {
769 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
770 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
775 * dev_load - load a network module
776 * @name: name of interface
778 * If a network interface is not present and the process has suitable
779 * privileges this function loads the module. If module loading is not
780 * available in this kernel then it becomes a nop.
783 void dev_load(const char *name)
785 struct net_device *dev;
787 read_lock(&dev_base_lock);
788 dev = __dev_get_by_name(name);
789 read_unlock(&dev_base_lock);
791 if (!dev && capable(CAP_SYS_MODULE))
792 request_module("%s", name);
795 static int default_rebuild_header(struct sk_buff *skb)
797 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
798 skb->dev ? skb->dev->name : "NULL!!!");
805 * dev_open - prepare an interface for use.
806 * @dev: device to open
808 * Takes a device from down to up state. The device's private open
809 * function is invoked and then the multicast lists are loaded. Finally
810 * the device is moved into the up state and a %NETDEV_UP message is
811 * sent to the netdev notifier chain.
813 * Calling this function on an active interface is a nop. On a failure
814 * a negative errno code is returned.
816 int dev_open(struct net_device *dev)
824 if (dev->flags & IFF_UP)
828 * Is it even present?
830 if (!netif_device_present(dev))
834 * Call device private open method
836 set_bit(__LINK_STATE_START, &dev->state);
838 ret = dev->open(dev);
840 clear_bit(__LINK_STATE_START, &dev->state);
844 * If it went open OK then:
851 dev->flags |= IFF_UP;
854 * Initialize multicasting status
859 * Wakeup transmit queue engine
864 * ... and announce new interface.
866 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
872 * dev_close - shutdown an interface.
873 * @dev: device to shutdown
875 * This function moves an active device into down state. A
876 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
877 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
880 int dev_close(struct net_device *dev)
882 if (!(dev->flags & IFF_UP))
886 * Tell people we are going down, so that they can
887 * prepare to death, when device is still operating.
889 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
893 clear_bit(__LINK_STATE_START, &dev->state);
895 /* Synchronize to scheduled poll. We cannot touch poll list,
896 * it can be even on different cpu. So just clear netif_running(),
897 * and wait when poll really will happen. Actually, the best place
898 * for this is inside dev->stop() after device stopped its irq
899 * engine, but this requires more changes in devices. */
901 smp_mb__after_clear_bit(); /* Commit netif_running(). */
902 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
908 * Call the device specific close. This cannot fail.
909 * Only if device is UP
911 * We allow it to be called even after a DETACH hot-plug
918 * Device is now down.
921 dev->flags &= ~IFF_UP;
924 * Tell people we are down
926 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
933 * Device change register/unregister. These are not inline or static
934 * as we export them to the world.
938 * register_netdevice_notifier - register a network notifier block
941 * Register a notifier to be called when network device events occur.
942 * The notifier passed is linked into the kernel structures and must
943 * not be reused until it has been unregistered. A negative errno code
944 * is returned on a failure.
946 * When registered all registration and up events are replayed
947 * to the new notifier to allow device to have a race free
948 * view of the network device list.
951 int register_netdevice_notifier(struct notifier_block *nb)
953 struct net_device *dev;
957 err = notifier_chain_register(&netdev_chain, nb);
959 for (dev = dev_base; dev; dev = dev->next) {
960 nb->notifier_call(nb, NETDEV_REGISTER, dev);
962 if (dev->flags & IFF_UP)
963 nb->notifier_call(nb, NETDEV_UP, dev);
971 * unregister_netdevice_notifier - unregister a network notifier block
974 * Unregister a notifier previously registered by
975 * register_netdevice_notifier(). The notifier is unlinked into the
976 * kernel structures and may then be reused. A negative errno code
977 * is returned on a failure.
980 int unregister_netdevice_notifier(struct notifier_block *nb)
982 return notifier_chain_unregister(&netdev_chain, nb);
986 * call_netdevice_notifiers - call all network notifier blocks
987 * @val: value passed unmodified to notifier function
988 * @v: pointer passed unmodified to notifier function
990 * Call all network notifier blocks. Parameters and return value
991 * are as for notifier_call_chain().
994 int call_netdevice_notifiers(unsigned long val, void *v)
996 return notifier_call_chain(&netdev_chain, val, v);
999 /* When > 0 there are consumers of rx skb time stamps */
1000 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1002 void net_enable_timestamp(void)
1004 atomic_inc(&netstamp_needed);
1007 void net_disable_timestamp(void)
1009 atomic_dec(&netstamp_needed);
1012 static inline void net_timestamp(struct timeval *stamp)
1014 if (atomic_read(&netstamp_needed))
1015 do_gettimeofday(stamp);
1023 * Support routine. Sends outgoing frames to any network
1024 * taps currently in use.
1027 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1029 struct packet_type *ptype;
1030 net_timestamp(&skb->stamp);
1033 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1034 /* Never send packets back to the socket
1035 * they originated from - MvS (miquels@drinkel.ow.org)
1037 if ((ptype->dev == dev || !ptype->dev) &&
1038 (ptype->af_packet_priv == NULL ||
1039 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1040 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1044 /* skb->nh should be correctly
1045 set by sender, so that the second statement is
1046 just protection against buggy protocols.
1048 skb2->mac.raw = skb2->data;
1050 if (skb2->nh.raw < skb2->data ||
1051 skb2->nh.raw > skb2->tail) {
1052 if (net_ratelimit())
1053 printk(KERN_CRIT "protocol %04x is "
1055 skb2->protocol, dev->name);
1056 skb2->nh.raw = skb2->data;
1059 skb2->h.raw = skb2->nh.raw;
1060 skb2->pkt_type = PACKET_OUTGOING;
1061 ptype->func(skb2, skb->dev, ptype);
1068 * Invalidate hardware checksum when packet is to be mangled, and
1069 * complete checksum manually on outgoing path.
1071 int skb_checksum_help(struct sk_buff *skb, int inward)
1074 int ret = 0, offset = skb->h.raw - skb->data;
1077 skb->ip_summed = CHECKSUM_NONE;
1081 if (skb_cloned(skb)) {
1082 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1087 if (offset > (int)skb->len)
1089 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1091 offset = skb->tail - skb->h.raw;
1094 if (skb->csum + 2 > offset)
1097 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1098 skb->ip_summed = CHECKSUM_NONE;
1103 #ifdef CONFIG_HIGHMEM
1104 /* Actually, we should eliminate this check as soon as we know, that:
1105 * 1. IOMMU is present and allows to map all the memory.
1106 * 2. No high memory really exists on this machine.
1109 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1113 if (dev->features & NETIF_F_HIGHDMA)
1116 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1117 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1123 #define illegal_highdma(dev, skb) (0)
1126 extern void skb_release_data(struct sk_buff *);
1128 /* Keep head the same: replace data */
1129 int __skb_linearize(struct sk_buff *skb, unsigned int __nocast gfp_mask)
1134 struct skb_shared_info *ninfo;
1135 int headerlen = skb->data - skb->head;
1136 int expand = (skb->tail + skb->data_len) - skb->end;
1138 if (skb_shared(skb))
1144 size = skb->end - skb->head + expand;
1145 size = SKB_DATA_ALIGN(size);
1146 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1150 /* Copy entire thing */
1151 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1155 ninfo = (struct skb_shared_info*)(data + size);
1156 atomic_set(&ninfo->dataref, 1);
1157 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1158 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1159 ninfo->nr_frags = 0;
1160 ninfo->frag_list = NULL;
1162 /* Offset between the two in bytes */
1163 offset = data - skb->head;
1165 /* Free old data. */
1166 skb_release_data(skb);
1169 skb->end = data + size;
1171 /* Set up new pointers */
1172 skb->h.raw += offset;
1173 skb->nh.raw += offset;
1174 skb->mac.raw += offset;
1175 skb->tail += offset;
1176 skb->data += offset;
1178 /* We are no longer a clone, even if we were. */
1181 skb->tail += skb->data_len;
1186 #define HARD_TX_LOCK(dev, cpu) { \
1187 if ((dev->features & NETIF_F_LLTX) == 0) { \
1188 spin_lock(&dev->xmit_lock); \
1189 dev->xmit_lock_owner = cpu; \
1193 #define HARD_TX_UNLOCK(dev) { \
1194 if ((dev->features & NETIF_F_LLTX) == 0) { \
1195 dev->xmit_lock_owner = -1; \
1196 spin_unlock(&dev->xmit_lock); \
1201 * dev_queue_xmit - transmit a buffer
1202 * @skb: buffer to transmit
1204 * Queue a buffer for transmission to a network device. The caller must
1205 * have set the device and priority and built the buffer before calling
1206 * this function. The function can be called from an interrupt.
1208 * A negative errno code is returned on a failure. A success does not
1209 * guarantee the frame will be transmitted as it may be dropped due
1210 * to congestion or traffic shaping.
1212 * -----------------------------------------------------------------------------------
1213 * I notice this method can also return errors from the queue disciplines,
1214 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1217 * Regardless of the return value, the skb is consumed, so it is currently
1218 * difficult to retry a send to this method. (You can bump the ref count
1219 * before sending to hold a reference for retry if you are careful.)
1221 * When calling this method, interrupts MUST be enabled. This is because
1222 * the BH enable code must have IRQs enabled so that it will not deadlock.
1226 int dev_queue_xmit(struct sk_buff *skb)
1228 struct net_device *dev = skb->dev;
1232 if (skb_shinfo(skb)->frag_list &&
1233 !(dev->features & NETIF_F_FRAGLIST) &&
1234 __skb_linearize(skb, GFP_ATOMIC))
1237 /* Fragmented skb is linearized if device does not support SG,
1238 * or if at least one of fragments is in highmem and device
1239 * does not support DMA from it.
1241 if (skb_shinfo(skb)->nr_frags &&
1242 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1243 __skb_linearize(skb, GFP_ATOMIC))
1246 /* If packet is not checksummed and device does not support
1247 * checksumming for this protocol, complete checksumming here.
1249 if (skb->ip_summed == CHECKSUM_HW &&
1250 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1251 (!(dev->features & NETIF_F_IP_CSUM) ||
1252 skb->protocol != htons(ETH_P_IP))))
1253 if (skb_checksum_help(skb, 0))
1256 /* Disable soft irqs for various locks below. Also
1257 * stops preemption for RCU.
1261 /* Updates of qdisc are serialized by queue_lock.
1262 * The struct Qdisc which is pointed to by qdisc is now a
1263 * rcu structure - it may be accessed without acquiring
1264 * a lock (but the structure may be stale.) The freeing of the
1265 * qdisc will be deferred until it's known that there are no
1266 * more references to it.
1268 * If the qdisc has an enqueue function, we still need to
1269 * hold the queue_lock before calling it, since queue_lock
1270 * also serializes access to the device queue.
1273 q = rcu_dereference(dev->qdisc);
1274 #ifdef CONFIG_NET_CLS_ACT
1275 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1278 /* Grab device queue */
1279 spin_lock(&dev->queue_lock);
1281 rc = q->enqueue(skb, q);
1285 spin_unlock(&dev->queue_lock);
1286 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1290 /* The device has no queue. Common case for software devices:
1291 loopback, all the sorts of tunnels...
1293 Really, it is unlikely that xmit_lock protection is necessary here.
1294 (f.e. loopback and IP tunnels are clean ignoring statistics
1296 However, it is possible, that they rely on protection
1299 Check this and shot the lock. It is not prone from deadlocks.
1300 Either shot noqueue qdisc, it is even simpler 8)
1302 if (dev->flags & IFF_UP) {
1303 int cpu = smp_processor_id(); /* ok because BHs are off */
1305 if (dev->xmit_lock_owner != cpu) {
1307 HARD_TX_LOCK(dev, cpu);
1309 if (!netif_queue_stopped(dev)) {
1311 dev_queue_xmit_nit(skb, dev);
1314 if (!dev->hard_start_xmit(skb, dev)) {
1315 HARD_TX_UNLOCK(dev);
1319 HARD_TX_UNLOCK(dev);
1320 if (net_ratelimit())
1321 printk(KERN_CRIT "Virtual device %s asks to "
1322 "queue packet!\n", dev->name);
1324 /* Recursion is detected! It is possible,
1326 if (net_ratelimit())
1327 printk(KERN_CRIT "Dead loop on virtual device "
1328 "%s, fix it urgently!\n", dev->name);
1344 /*=======================================================================
1346 =======================================================================*/
1348 int netdev_max_backlog = 1000;
1349 int netdev_budget = 300;
1350 int weight_p = 64; /* old backlog weight */
1352 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1356 * netif_rx - post buffer to the network code
1357 * @skb: buffer to post
1359 * This function receives a packet from a device driver and queues it for
1360 * the upper (protocol) levels to process. It always succeeds. The buffer
1361 * may be dropped during processing for congestion control or by the
1365 * NET_RX_SUCCESS (no congestion)
1366 * NET_RX_CN_LOW (low congestion)
1367 * NET_RX_CN_MOD (moderate congestion)
1368 * NET_RX_CN_HIGH (high congestion)
1369 * NET_RX_DROP (packet was dropped)
1373 int netif_rx(struct sk_buff *skb)
1375 struct softnet_data *queue;
1376 unsigned long flags;
1378 /* if netpoll wants it, pretend we never saw it */
1379 if (netpoll_rx(skb))
1382 if (!skb->stamp.tv_sec)
1383 net_timestamp(&skb->stamp);
1386 * The code is rearranged so that the path is the most
1387 * short when CPU is congested, but is still operating.
1389 local_irq_save(flags);
1390 queue = &__get_cpu_var(softnet_data);
1392 __get_cpu_var(netdev_rx_stat).total++;
1393 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1394 if (queue->input_pkt_queue.qlen) {
1397 __skb_queue_tail(&queue->input_pkt_queue, skb);
1398 local_irq_restore(flags);
1399 return NET_RX_SUCCESS;
1402 netif_rx_schedule(&queue->backlog_dev);
1406 __get_cpu_var(netdev_rx_stat).dropped++;
1407 local_irq_restore(flags);
1413 int netif_rx_ni(struct sk_buff *skb)
1418 err = netif_rx(skb);
1419 if (local_softirq_pending())
1426 EXPORT_SYMBOL(netif_rx_ni);
1428 static __inline__ void skb_bond(struct sk_buff *skb)
1430 struct net_device *dev = skb->dev;
1433 skb->real_dev = skb->dev;
1434 skb->dev = dev->master;
1438 static void net_tx_action(struct softirq_action *h)
1440 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1442 if (sd->completion_queue) {
1443 struct sk_buff *clist;
1445 local_irq_disable();
1446 clist = sd->completion_queue;
1447 sd->completion_queue = NULL;
1451 struct sk_buff *skb = clist;
1452 clist = clist->next;
1454 BUG_TRAP(!atomic_read(&skb->users));
1459 if (sd->output_queue) {
1460 struct net_device *head;
1462 local_irq_disable();
1463 head = sd->output_queue;
1464 sd->output_queue = NULL;
1468 struct net_device *dev = head;
1469 head = head->next_sched;
1471 smp_mb__before_clear_bit();
1472 clear_bit(__LINK_STATE_SCHED, &dev->state);
1474 if (spin_trylock(&dev->queue_lock)) {
1476 spin_unlock(&dev->queue_lock);
1478 netif_schedule(dev);
1484 static __inline__ int deliver_skb(struct sk_buff *skb,
1485 struct packet_type *pt_prev)
1487 atomic_inc(&skb->users);
1488 return pt_prev->func(skb, skb->dev, pt_prev);
1491 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1492 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1494 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1495 unsigned char *addr);
1496 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1498 static __inline__ int handle_bridge(struct sk_buff **pskb,
1499 struct packet_type **pt_prev, int *ret)
1501 struct net_bridge_port *port;
1503 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1504 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1508 *ret = deliver_skb(*pskb, *pt_prev);
1512 return br_handle_frame_hook(port, pskb);
1515 #define handle_bridge(skb, pt_prev, ret) (0)
1518 #ifdef CONFIG_NET_CLS_ACT
1519 /* TODO: Maybe we should just force sch_ingress to be compiled in
1520 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1521 * a compare and 2 stores extra right now if we dont have it on
1522 * but have CONFIG_NET_CLS_ACT
1523 * NOTE: This doesnt stop any functionality; if you dont have
1524 * the ingress scheduler, you just cant add policies on ingress.
1527 static int ing_filter(struct sk_buff *skb)
1530 struct net_device *dev = skb->dev;
1531 int result = TC_ACT_OK;
1533 if (dev->qdisc_ingress) {
1534 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1535 if (MAX_RED_LOOP < ttl++) {
1536 printk("Redir loop detected Dropping packet (%s->%s)\n",
1537 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1541 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1543 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1544 if (NULL == skb->input_dev) {
1545 skb->input_dev = skb->dev;
1546 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1548 spin_lock(&dev->ingress_lock);
1549 if ((q = dev->qdisc_ingress) != NULL)
1550 result = q->enqueue(skb, q);
1551 spin_unlock(&dev->ingress_lock);
1559 int netif_receive_skb(struct sk_buff *skb)
1561 struct packet_type *ptype, *pt_prev;
1562 int ret = NET_RX_DROP;
1563 unsigned short type;
1565 /* if we've gotten here through NAPI, check netpoll */
1566 if (skb->dev->poll && netpoll_rx(skb))
1569 if (!skb->stamp.tv_sec)
1570 net_timestamp(&skb->stamp);
1574 __get_cpu_var(netdev_rx_stat).total++;
1576 skb->h.raw = skb->nh.raw = skb->data;
1577 skb->mac_len = skb->nh.raw - skb->mac.raw;
1583 #ifdef CONFIG_NET_CLS_ACT
1584 if (skb->tc_verd & TC_NCLS) {
1585 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1590 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1591 if (!ptype->dev || ptype->dev == skb->dev) {
1593 ret = deliver_skb(skb, pt_prev);
1598 #ifdef CONFIG_NET_CLS_ACT
1600 ret = deliver_skb(skb, pt_prev);
1601 pt_prev = NULL; /* noone else should process this after*/
1603 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1606 ret = ing_filter(skb);
1608 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1617 handle_diverter(skb);
1619 if (handle_bridge(&skb, &pt_prev, &ret))
1622 type = skb->protocol;
1623 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1624 if (ptype->type == type &&
1625 (!ptype->dev || ptype->dev == skb->dev)) {
1627 ret = deliver_skb(skb, pt_prev);
1633 ret = pt_prev->func(skb, skb->dev, pt_prev);
1636 /* Jamal, now you will not able to escape explaining
1637 * me how you were going to use this. :-)
1647 static int process_backlog(struct net_device *backlog_dev, int *budget)
1650 int quota = min(backlog_dev->quota, *budget);
1651 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1652 unsigned long start_time = jiffies;
1654 backlog_dev->weight = weight_p;
1656 struct sk_buff *skb;
1657 struct net_device *dev;
1659 local_irq_disable();
1660 skb = __skb_dequeue(&queue->input_pkt_queue);
1667 netif_receive_skb(skb);
1673 if (work >= quota || jiffies - start_time > 1)
1678 backlog_dev->quota -= work;
1683 backlog_dev->quota -= work;
1686 list_del(&backlog_dev->poll_list);
1687 smp_mb__before_clear_bit();
1688 netif_poll_enable(backlog_dev);
1694 static void net_rx_action(struct softirq_action *h)
1696 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1697 unsigned long start_time = jiffies;
1698 int budget = netdev_budget;
1700 local_irq_disable();
1702 while (!list_empty(&queue->poll_list)) {
1703 struct net_device *dev;
1705 if (budget <= 0 || jiffies - start_time > 1)
1710 dev = list_entry(queue->poll_list.next,
1711 struct net_device, poll_list);
1712 netpoll_poll_lock(dev);
1714 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1715 netpoll_poll_unlock(dev);
1716 local_irq_disable();
1717 list_del(&dev->poll_list);
1718 list_add_tail(&dev->poll_list, &queue->poll_list);
1720 dev->quota += dev->weight;
1722 dev->quota = dev->weight;
1724 netpoll_poll_unlock(dev);
1726 local_irq_disable();
1734 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1735 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1739 static gifconf_func_t * gifconf_list [NPROTO];
1742 * register_gifconf - register a SIOCGIF handler
1743 * @family: Address family
1744 * @gifconf: Function handler
1746 * Register protocol dependent address dumping routines. The handler
1747 * that is passed must not be freed or reused until it has been replaced
1748 * by another handler.
1750 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1752 if (family >= NPROTO)
1754 gifconf_list[family] = gifconf;
1760 * Map an interface index to its name (SIOCGIFNAME)
1764 * We need this ioctl for efficient implementation of the
1765 * if_indextoname() function required by the IPv6 API. Without
1766 * it, we would have to search all the interfaces to find a
1770 static int dev_ifname(struct ifreq __user *arg)
1772 struct net_device *dev;
1776 * Fetch the caller's info block.
1779 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1782 read_lock(&dev_base_lock);
1783 dev = __dev_get_by_index(ifr.ifr_ifindex);
1785 read_unlock(&dev_base_lock);
1789 strcpy(ifr.ifr_name, dev->name);
1790 read_unlock(&dev_base_lock);
1792 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1798 * Perform a SIOCGIFCONF call. This structure will change
1799 * size eventually, and there is nothing I can do about it.
1800 * Thus we will need a 'compatibility mode'.
1803 static int dev_ifconf(char __user *arg)
1806 struct net_device *dev;
1813 * Fetch the caller's info block.
1816 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1823 * Loop over the interfaces, and write an info block for each.
1827 for (dev = dev_base; dev; dev = dev->next) {
1828 for (i = 0; i < NPROTO; i++) {
1829 if (gifconf_list[i]) {
1832 done = gifconf_list[i](dev, NULL, 0);
1834 done = gifconf_list[i](dev, pos + total,
1844 * All done. Write the updated control block back to the caller.
1846 ifc.ifc_len = total;
1849 * Both BSD and Solaris return 0 here, so we do too.
1851 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1854 #ifdef CONFIG_PROC_FS
1856 * This is invoked by the /proc filesystem handler to display a device
1859 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1861 struct net_device *dev;
1864 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1866 return i == pos ? dev : NULL;
1869 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1871 read_lock(&dev_base_lock);
1872 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1875 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1878 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1881 void dev_seq_stop(struct seq_file *seq, void *v)
1883 read_unlock(&dev_base_lock);
1886 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1888 if (dev->get_stats) {
1889 struct net_device_stats *stats = dev->get_stats(dev);
1891 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1892 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1893 dev->name, stats->rx_bytes, stats->rx_packets,
1895 stats->rx_dropped + stats->rx_missed_errors,
1896 stats->rx_fifo_errors,
1897 stats->rx_length_errors + stats->rx_over_errors +
1898 stats->rx_crc_errors + stats->rx_frame_errors,
1899 stats->rx_compressed, stats->multicast,
1900 stats->tx_bytes, stats->tx_packets,
1901 stats->tx_errors, stats->tx_dropped,
1902 stats->tx_fifo_errors, stats->collisions,
1903 stats->tx_carrier_errors +
1904 stats->tx_aborted_errors +
1905 stats->tx_window_errors +
1906 stats->tx_heartbeat_errors,
1907 stats->tx_compressed);
1909 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1913 * Called from the PROCfs module. This now uses the new arbitrary sized
1914 * /proc/net interface to create /proc/net/dev
1916 static int dev_seq_show(struct seq_file *seq, void *v)
1918 if (v == SEQ_START_TOKEN)
1919 seq_puts(seq, "Inter-| Receive "
1921 " face |bytes packets errs drop fifo frame "
1922 "compressed multicast|bytes packets errs "
1923 "drop fifo colls carrier compressed\n");
1925 dev_seq_printf_stats(seq, v);
1929 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1931 struct netif_rx_stats *rc = NULL;
1933 while (*pos < NR_CPUS)
1934 if (cpu_online(*pos)) {
1935 rc = &per_cpu(netdev_rx_stat, *pos);
1942 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1944 return softnet_get_online(pos);
1947 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1950 return softnet_get_online(pos);
1953 static void softnet_seq_stop(struct seq_file *seq, void *v)
1957 static int softnet_seq_show(struct seq_file *seq, void *v)
1959 struct netif_rx_stats *s = v;
1961 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
1962 s->total, s->dropped, s->time_squeeze, 0,
1963 0, 0, 0, 0, /* was fastroute */
1968 static struct seq_operations dev_seq_ops = {
1969 .start = dev_seq_start,
1970 .next = dev_seq_next,
1971 .stop = dev_seq_stop,
1972 .show = dev_seq_show,
1975 static int dev_seq_open(struct inode *inode, struct file *file)
1977 return seq_open(file, &dev_seq_ops);
1980 static struct file_operations dev_seq_fops = {
1981 .owner = THIS_MODULE,
1982 .open = dev_seq_open,
1984 .llseek = seq_lseek,
1985 .release = seq_release,
1988 static struct seq_operations softnet_seq_ops = {
1989 .start = softnet_seq_start,
1990 .next = softnet_seq_next,
1991 .stop = softnet_seq_stop,
1992 .show = softnet_seq_show,
1995 static int softnet_seq_open(struct inode *inode, struct file *file)
1997 return seq_open(file, &softnet_seq_ops);
2000 static struct file_operations softnet_seq_fops = {
2001 .owner = THIS_MODULE,
2002 .open = softnet_seq_open,
2004 .llseek = seq_lseek,
2005 .release = seq_release,
2009 extern int wireless_proc_init(void);
2011 #define wireless_proc_init() 0
2014 static int __init dev_proc_init(void)
2018 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2020 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2022 if (wireless_proc_init())
2028 proc_net_remove("softnet_stat");
2030 proc_net_remove("dev");
2034 #define dev_proc_init() 0
2035 #endif /* CONFIG_PROC_FS */
2039 * netdev_set_master - set up master/slave pair
2040 * @slave: slave device
2041 * @master: new master device
2043 * Changes the master device of the slave. Pass %NULL to break the
2044 * bonding. The caller must hold the RTNL semaphore. On a failure
2045 * a negative errno code is returned. On success the reference counts
2046 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2047 * function returns zero.
2049 int netdev_set_master(struct net_device *slave, struct net_device *master)
2051 struct net_device *old = slave->master;
2061 slave->master = master;
2069 slave->flags |= IFF_SLAVE;
2071 slave->flags &= ~IFF_SLAVE;
2073 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2078 * dev_set_promiscuity - update promiscuity count on a device
2082 * Add or remove promsicuity from a device. While the count in the device
2083 * remains above zero the interface remains promiscuous. Once it hits zero
2084 * the device reverts back to normal filtering operation. A negative inc
2085 * value is used to drop promiscuity on the device.
2087 void dev_set_promiscuity(struct net_device *dev, int inc)
2089 unsigned short old_flags = dev->flags;
2091 if ((dev->promiscuity += inc) == 0)
2092 dev->flags &= ~IFF_PROMISC;
2094 dev->flags |= IFF_PROMISC;
2095 if (dev->flags != old_flags) {
2097 printk(KERN_INFO "device %s %s promiscuous mode\n",
2098 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2104 * dev_set_allmulti - update allmulti count on a device
2108 * Add or remove reception of all multicast frames to a device. While the
2109 * count in the device remains above zero the interface remains listening
2110 * to all interfaces. Once it hits zero the device reverts back to normal
2111 * filtering operation. A negative @inc value is used to drop the counter
2112 * when releasing a resource needing all multicasts.
2115 void dev_set_allmulti(struct net_device *dev, int inc)
2117 unsigned short old_flags = dev->flags;
2119 dev->flags |= IFF_ALLMULTI;
2120 if ((dev->allmulti += inc) == 0)
2121 dev->flags &= ~IFF_ALLMULTI;
2122 if (dev->flags ^ old_flags)
2126 unsigned dev_get_flags(const struct net_device *dev)
2130 flags = (dev->flags & ~(IFF_PROMISC |
2133 (dev->gflags & (IFF_PROMISC |
2136 if (netif_running(dev) && netif_carrier_ok(dev))
2137 flags |= IFF_RUNNING;
2142 int dev_change_flags(struct net_device *dev, unsigned flags)
2145 int old_flags = dev->flags;
2148 * Set the flags on our device.
2151 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2152 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2154 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2158 * Load in the correct multicast list now the flags have changed.
2164 * Have we downed the interface. We handle IFF_UP ourselves
2165 * according to user attempts to set it, rather than blindly
2170 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2171 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2177 if (dev->flags & IFF_UP &&
2178 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2180 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2182 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2183 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2184 dev->gflags ^= IFF_PROMISC;
2185 dev_set_promiscuity(dev, inc);
2188 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2189 is important. Some (broken) drivers set IFF_PROMISC, when
2190 IFF_ALLMULTI is requested not asking us and not reporting.
2192 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2193 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2194 dev->gflags ^= IFF_ALLMULTI;
2195 dev_set_allmulti(dev, inc);
2198 if (old_flags ^ dev->flags)
2199 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2204 int dev_set_mtu(struct net_device *dev, int new_mtu)
2208 if (new_mtu == dev->mtu)
2211 /* MTU must be positive. */
2215 if (!netif_device_present(dev))
2219 if (dev->change_mtu)
2220 err = dev->change_mtu(dev, new_mtu);
2223 if (!err && dev->flags & IFF_UP)
2224 notifier_call_chain(&netdev_chain,
2225 NETDEV_CHANGEMTU, dev);
2229 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2233 if (!dev->set_mac_address)
2235 if (sa->sa_family != dev->type)
2237 if (!netif_device_present(dev))
2239 err = dev->set_mac_address(dev, sa);
2241 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2246 * Perform the SIOCxIFxxx calls.
2248 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2251 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2257 case SIOCGIFFLAGS: /* Get interface flags */
2258 ifr->ifr_flags = dev_get_flags(dev);
2261 case SIOCSIFFLAGS: /* Set interface flags */
2262 return dev_change_flags(dev, ifr->ifr_flags);
2264 case SIOCGIFMETRIC: /* Get the metric on the interface
2265 (currently unused) */
2266 ifr->ifr_metric = 0;
2269 case SIOCSIFMETRIC: /* Set the metric on the interface
2270 (currently unused) */
2273 case SIOCGIFMTU: /* Get the MTU of a device */
2274 ifr->ifr_mtu = dev->mtu;
2277 case SIOCSIFMTU: /* Set the MTU of a device */
2278 return dev_set_mtu(dev, ifr->ifr_mtu);
2282 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2284 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2285 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2286 ifr->ifr_hwaddr.sa_family = dev->type;
2290 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2292 case SIOCSIFHWBROADCAST:
2293 if (ifr->ifr_hwaddr.sa_family != dev->type)
2295 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2296 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2297 notifier_call_chain(&netdev_chain,
2298 NETDEV_CHANGEADDR, dev);
2302 ifr->ifr_map.mem_start = dev->mem_start;
2303 ifr->ifr_map.mem_end = dev->mem_end;
2304 ifr->ifr_map.base_addr = dev->base_addr;
2305 ifr->ifr_map.irq = dev->irq;
2306 ifr->ifr_map.dma = dev->dma;
2307 ifr->ifr_map.port = dev->if_port;
2311 if (dev->set_config) {
2312 if (!netif_device_present(dev))
2314 return dev->set_config(dev, &ifr->ifr_map);
2319 if (!dev->set_multicast_list ||
2320 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2322 if (!netif_device_present(dev))
2324 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2328 if (!dev->set_multicast_list ||
2329 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2331 if (!netif_device_present(dev))
2333 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2337 ifr->ifr_ifindex = dev->ifindex;
2341 ifr->ifr_qlen = dev->tx_queue_len;
2345 if (ifr->ifr_qlen < 0)
2347 dev->tx_queue_len = ifr->ifr_qlen;
2351 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2352 return dev_change_name(dev, ifr->ifr_newname);
2355 * Unknown or private ioctl
2359 if ((cmd >= SIOCDEVPRIVATE &&
2360 cmd <= SIOCDEVPRIVATE + 15) ||
2361 cmd == SIOCBONDENSLAVE ||
2362 cmd == SIOCBONDRELEASE ||
2363 cmd == SIOCBONDSETHWADDR ||
2364 cmd == SIOCBONDSLAVEINFOQUERY ||
2365 cmd == SIOCBONDINFOQUERY ||
2366 cmd == SIOCBONDCHANGEACTIVE ||
2367 cmd == SIOCGMIIPHY ||
2368 cmd == SIOCGMIIREG ||
2369 cmd == SIOCSMIIREG ||
2370 cmd == SIOCBRADDIF ||
2371 cmd == SIOCBRDELIF ||
2372 cmd == SIOCWANDEV) {
2374 if (dev->do_ioctl) {
2375 if (netif_device_present(dev))
2376 err = dev->do_ioctl(dev, ifr,
2389 * This function handles all "interface"-type I/O control requests. The actual
2390 * 'doing' part of this is dev_ifsioc above.
2394 * dev_ioctl - network device ioctl
2395 * @cmd: command to issue
2396 * @arg: pointer to a struct ifreq in user space
2398 * Issue ioctl functions to devices. This is normally called by the
2399 * user space syscall interfaces but can sometimes be useful for
2400 * other purposes. The return value is the return from the syscall if
2401 * positive or a negative errno code on error.
2404 int dev_ioctl(unsigned int cmd, void __user *arg)
2410 /* One special case: SIOCGIFCONF takes ifconf argument
2411 and requires shared lock, because it sleeps writing
2415 if (cmd == SIOCGIFCONF) {
2417 ret = dev_ifconf((char __user *) arg);
2421 if (cmd == SIOCGIFNAME)
2422 return dev_ifname((struct ifreq __user *)arg);
2424 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2427 ifr.ifr_name[IFNAMSIZ-1] = 0;
2429 colon = strchr(ifr.ifr_name, ':');
2434 * See which interface the caller is talking about.
2439 * These ioctl calls:
2440 * - can be done by all.
2441 * - atomic and do not require locking.
2452 dev_load(ifr.ifr_name);
2453 read_lock(&dev_base_lock);
2454 ret = dev_ifsioc(&ifr, cmd);
2455 read_unlock(&dev_base_lock);
2459 if (copy_to_user(arg, &ifr,
2460 sizeof(struct ifreq)))
2466 dev_load(ifr.ifr_name);
2468 ret = dev_ethtool(&ifr);
2473 if (copy_to_user(arg, &ifr,
2474 sizeof(struct ifreq)))
2480 * These ioctl calls:
2481 * - require superuser power.
2482 * - require strict serialization.
2488 if (!capable(CAP_NET_ADMIN))
2490 dev_load(ifr.ifr_name);
2492 ret = dev_ifsioc(&ifr, cmd);
2497 if (copy_to_user(arg, &ifr,
2498 sizeof(struct ifreq)))
2504 * These ioctl calls:
2505 * - require superuser power.
2506 * - require strict serialization.
2507 * - do not return a value
2517 case SIOCSIFHWBROADCAST:
2520 case SIOCBONDENSLAVE:
2521 case SIOCBONDRELEASE:
2522 case SIOCBONDSETHWADDR:
2523 case SIOCBONDSLAVEINFOQUERY:
2524 case SIOCBONDINFOQUERY:
2525 case SIOCBONDCHANGEACTIVE:
2528 if (!capable(CAP_NET_ADMIN))
2530 dev_load(ifr.ifr_name);
2532 ret = dev_ifsioc(&ifr, cmd);
2537 /* Get the per device memory space. We can add this but
2538 * currently do not support it */
2540 /* Set the per device memory buffer space.
2541 * Not applicable in our case */
2546 * Unknown or private ioctl.
2549 if (cmd == SIOCWANDEV ||
2550 (cmd >= SIOCDEVPRIVATE &&
2551 cmd <= SIOCDEVPRIVATE + 15)) {
2552 dev_load(ifr.ifr_name);
2554 ret = dev_ifsioc(&ifr, cmd);
2556 if (!ret && copy_to_user(arg, &ifr,
2557 sizeof(struct ifreq)))
2562 /* Take care of Wireless Extensions */
2563 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2564 /* If command is `set a parameter', or
2565 * `get the encoding parameters', check if
2566 * the user has the right to do it */
2567 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2568 if (!capable(CAP_NET_ADMIN))
2571 dev_load(ifr.ifr_name);
2573 /* Follow me in net/core/wireless.c */
2574 ret = wireless_process_ioctl(&ifr, cmd);
2576 if (IW_IS_GET(cmd) &&
2577 copy_to_user(arg, &ifr,
2578 sizeof(struct ifreq)))
2582 #endif /* WIRELESS_EXT */
2589 * dev_new_index - allocate an ifindex
2591 * Returns a suitable unique value for a new device interface
2592 * number. The caller must hold the rtnl semaphore or the
2593 * dev_base_lock to be sure it remains unique.
2595 static int dev_new_index(void)
2601 if (!__dev_get_by_index(ifindex))
2606 static int dev_boot_phase = 1;
2608 /* Delayed registration/unregisteration */
2609 static DEFINE_SPINLOCK(net_todo_list_lock);
2610 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2612 static inline void net_set_todo(struct net_device *dev)
2614 spin_lock(&net_todo_list_lock);
2615 list_add_tail(&dev->todo_list, &net_todo_list);
2616 spin_unlock(&net_todo_list_lock);
2620 * register_netdevice - register a network device
2621 * @dev: device to register
2623 * Take a completed network device structure and add it to the kernel
2624 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2625 * chain. 0 is returned on success. A negative errno code is returned
2626 * on a failure to set up the device, or if the name is a duplicate.
2628 * Callers must hold the rtnl semaphore. You may want
2629 * register_netdev() instead of this.
2632 * The locking appears insufficient to guarantee two parallel registers
2633 * will not get the same name.
2636 int register_netdevice(struct net_device *dev)
2638 struct hlist_head *head;
2639 struct hlist_node *p;
2642 BUG_ON(dev_boot_phase);
2645 /* When net_device's are persistent, this will be fatal. */
2646 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2648 spin_lock_init(&dev->queue_lock);
2649 spin_lock_init(&dev->xmit_lock);
2650 dev->xmit_lock_owner = -1;
2651 #ifdef CONFIG_NET_CLS_ACT
2652 spin_lock_init(&dev->ingress_lock);
2655 ret = alloc_divert_blk(dev);
2661 /* Init, if this function is available */
2663 ret = dev->init(dev);
2671 if (!dev_valid_name(dev->name)) {
2676 dev->ifindex = dev_new_index();
2677 if (dev->iflink == -1)
2678 dev->iflink = dev->ifindex;
2680 /* Check for existence of name */
2681 head = dev_name_hash(dev->name);
2682 hlist_for_each(p, head) {
2683 struct net_device *d
2684 = hlist_entry(p, struct net_device, name_hlist);
2685 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2691 /* Fix illegal SG+CSUM combinations. */
2692 if ((dev->features & NETIF_F_SG) &&
2693 !(dev->features & (NETIF_F_IP_CSUM |
2695 NETIF_F_HW_CSUM))) {
2696 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2698 dev->features &= ~NETIF_F_SG;
2701 /* TSO requires that SG is present as well. */
2702 if ((dev->features & NETIF_F_TSO) &&
2703 !(dev->features & NETIF_F_SG)) {
2704 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2706 dev->features &= ~NETIF_F_TSO;
2710 * nil rebuild_header routine,
2711 * that should be never called and used as just bug trap.
2714 if (!dev->rebuild_header)
2715 dev->rebuild_header = default_rebuild_header;
2718 * Default initial state at registry is that the
2719 * device is present.
2722 set_bit(__LINK_STATE_PRESENT, &dev->state);
2725 dev_init_scheduler(dev);
2726 write_lock_bh(&dev_base_lock);
2728 dev_tail = &dev->next;
2729 hlist_add_head(&dev->name_hlist, head);
2730 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2732 dev->reg_state = NETREG_REGISTERING;
2733 write_unlock_bh(&dev_base_lock);
2735 /* Notify protocols, that a new device appeared. */
2736 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2738 /* Finish registration after unlock */
2745 free_divert_blk(dev);
2750 * register_netdev - register a network device
2751 * @dev: device to register
2753 * Take a completed network device structure and add it to the kernel
2754 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2755 * chain. 0 is returned on success. A negative errno code is returned
2756 * on a failure to set up the device, or if the name is a duplicate.
2758 * This is a wrapper around register_netdev that takes the rtnl semaphore
2759 * and expands the device name if you passed a format string to
2762 int register_netdev(struct net_device *dev)
2769 * If the name is a format string the caller wants us to do a
2772 if (strchr(dev->name, '%')) {
2773 err = dev_alloc_name(dev, dev->name);
2779 * Back compatibility hook. Kill this one in 2.5
2781 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2782 err = dev_alloc_name(dev, "eth%d");
2787 err = register_netdevice(dev);
2792 EXPORT_SYMBOL(register_netdev);
2795 * netdev_wait_allrefs - wait until all references are gone.
2797 * This is called when unregistering network devices.
2799 * Any protocol or device that holds a reference should register
2800 * for netdevice notification, and cleanup and put back the
2801 * reference if they receive an UNREGISTER event.
2802 * We can get stuck here if buggy protocols don't correctly
2805 static void netdev_wait_allrefs(struct net_device *dev)
2807 unsigned long rebroadcast_time, warning_time;
2809 rebroadcast_time = warning_time = jiffies;
2810 while (atomic_read(&dev->refcnt) != 0) {
2811 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2814 /* Rebroadcast unregister notification */
2815 notifier_call_chain(&netdev_chain,
2816 NETDEV_UNREGISTER, dev);
2818 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2820 /* We must not have linkwatch events
2821 * pending on unregister. If this
2822 * happens, we simply run the queue
2823 * unscheduled, resulting in a noop
2826 linkwatch_run_queue();
2831 rebroadcast_time = jiffies;
2836 if (time_after(jiffies, warning_time + 10 * HZ)) {
2837 printk(KERN_EMERG "unregister_netdevice: "
2838 "waiting for %s to become free. Usage "
2840 dev->name, atomic_read(&dev->refcnt));
2841 warning_time = jiffies;
2850 * register_netdevice(x1);
2851 * register_netdevice(x2);
2853 * unregister_netdevice(y1);
2854 * unregister_netdevice(y2);
2860 * We are invoked by rtnl_unlock() after it drops the semaphore.
2861 * This allows us to deal with problems:
2862 * 1) We can create/delete sysfs objects which invoke hotplug
2863 * without deadlocking with linkwatch via keventd.
2864 * 2) Since we run with the RTNL semaphore not held, we can sleep
2865 * safely in order to wait for the netdev refcnt to drop to zero.
2867 static DECLARE_MUTEX(net_todo_run_mutex);
2868 void netdev_run_todo(void)
2870 struct list_head list = LIST_HEAD_INIT(list);
2874 /* Need to guard against multiple cpu's getting out of order. */
2875 down(&net_todo_run_mutex);
2877 /* Not safe to do outside the semaphore. We must not return
2878 * until all unregister events invoked by the local processor
2879 * have been completed (either by this todo run, or one on
2882 if (list_empty(&net_todo_list))
2885 /* Snapshot list, allow later requests */
2886 spin_lock(&net_todo_list_lock);
2887 list_splice_init(&net_todo_list, &list);
2888 spin_unlock(&net_todo_list_lock);
2890 while (!list_empty(&list)) {
2891 struct net_device *dev
2892 = list_entry(list.next, struct net_device, todo_list);
2893 list_del(&dev->todo_list);
2895 switch(dev->reg_state) {
2896 case NETREG_REGISTERING:
2897 err = netdev_register_sysfs(dev);
2899 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2901 dev->reg_state = NETREG_REGISTERED;
2904 case NETREG_UNREGISTERING:
2905 netdev_unregister_sysfs(dev);
2906 dev->reg_state = NETREG_UNREGISTERED;
2908 netdev_wait_allrefs(dev);
2911 BUG_ON(atomic_read(&dev->refcnt));
2912 BUG_TRAP(!dev->ip_ptr);
2913 BUG_TRAP(!dev->ip6_ptr);
2914 BUG_TRAP(!dev->dn_ptr);
2917 /* It must be the very last action,
2918 * after this 'dev' may point to freed up memory.
2920 if (dev->destructor)
2921 dev->destructor(dev);
2925 printk(KERN_ERR "network todo '%s' but state %d\n",
2926 dev->name, dev->reg_state);
2932 up(&net_todo_run_mutex);
2936 * alloc_netdev - allocate network device
2937 * @sizeof_priv: size of private data to allocate space for
2938 * @name: device name format string
2939 * @setup: callback to initialize device
2941 * Allocates a struct net_device with private data area for driver use
2942 * and performs basic initialization.
2944 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
2945 void (*setup)(struct net_device *))
2948 struct net_device *dev;
2951 /* ensure 32-byte alignment of both the device and private area */
2952 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
2953 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
2955 p = kmalloc(alloc_size, GFP_KERNEL);
2957 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
2960 memset(p, 0, alloc_size);
2962 dev = (struct net_device *)
2963 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
2964 dev->padded = (char *)dev - (char *)p;
2967 dev->priv = netdev_priv(dev);
2970 strcpy(dev->name, name);
2973 EXPORT_SYMBOL(alloc_netdev);
2976 * free_netdev - free network device
2979 * This function does the last stage of destroying an allocated device
2980 * interface. The reference to the device object is released.
2981 * If this is the last reference then it will be freed.
2983 void free_netdev(struct net_device *dev)
2986 /* Compatiablity with error handling in drivers */
2987 if (dev->reg_state == NETREG_UNINITIALIZED) {
2988 kfree((char *)dev - dev->padded);
2992 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
2993 dev->reg_state = NETREG_RELEASED;
2995 /* will free via class release */
2996 class_device_put(&dev->class_dev);
2998 kfree((char *)dev - dev->padded);
3002 /* Synchronize with packet receive processing. */
3003 void synchronize_net(void)
3010 * unregister_netdevice - remove device from the kernel
3013 * This function shuts down a device interface and removes it
3014 * from the kernel tables. On success 0 is returned, on a failure
3015 * a negative errno code is returned.
3017 * Callers must hold the rtnl semaphore. You may want
3018 * unregister_netdev() instead of this.
3021 int unregister_netdevice(struct net_device *dev)
3023 struct net_device *d, **dp;
3025 BUG_ON(dev_boot_phase);
3028 /* Some devices call without registering for initialization unwind. */
3029 if (dev->reg_state == NETREG_UNINITIALIZED) {
3030 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3031 "was registered\n", dev->name, dev);
3035 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3037 /* If device is running, close it first. */
3038 if (dev->flags & IFF_UP)
3041 /* And unlink it from device chain. */
3042 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3044 write_lock_bh(&dev_base_lock);
3045 hlist_del(&dev->name_hlist);
3046 hlist_del(&dev->index_hlist);
3047 if (dev_tail == &dev->next)
3050 write_unlock_bh(&dev_base_lock);
3055 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3060 dev->reg_state = NETREG_UNREGISTERING;
3064 /* Shutdown queueing discipline. */
3068 /* Notify protocols, that we are about to destroy
3069 this device. They should clean all the things.
3071 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3074 * Flush the multicast chain
3076 dev_mc_discard(dev);
3081 /* Notifier chain MUST detach us from master device. */
3082 BUG_TRAP(!dev->master);
3084 free_divert_blk(dev);
3086 /* Finish processing unregister after unlock */
3096 * unregister_netdev - remove device from the kernel
3099 * This function shuts down a device interface and removes it
3100 * from the kernel tables. On success 0 is returned, on a failure
3101 * a negative errno code is returned.
3103 * This is just a wrapper for unregister_netdevice that takes
3104 * the rtnl semaphore. In general you want to use this and not
3105 * unregister_netdevice.
3107 void unregister_netdev(struct net_device *dev)
3110 unregister_netdevice(dev);
3114 EXPORT_SYMBOL(unregister_netdev);
3116 #ifdef CONFIG_HOTPLUG_CPU
3117 static int dev_cpu_callback(struct notifier_block *nfb,
3118 unsigned long action,
3121 struct sk_buff **list_skb;
3122 struct net_device **list_net;
3123 struct sk_buff *skb;
3124 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3125 struct softnet_data *sd, *oldsd;
3127 if (action != CPU_DEAD)
3130 local_irq_disable();
3131 cpu = smp_processor_id();
3132 sd = &per_cpu(softnet_data, cpu);
3133 oldsd = &per_cpu(softnet_data, oldcpu);
3135 /* Find end of our completion_queue. */
3136 list_skb = &sd->completion_queue;
3138 list_skb = &(*list_skb)->next;
3139 /* Append completion queue from offline CPU. */
3140 *list_skb = oldsd->completion_queue;
3141 oldsd->completion_queue = NULL;
3143 /* Find end of our output_queue. */
3144 list_net = &sd->output_queue;
3146 list_net = &(*list_net)->next_sched;
3147 /* Append output queue from offline CPU. */
3148 *list_net = oldsd->output_queue;
3149 oldsd->output_queue = NULL;
3151 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3154 /* Process offline CPU's input_pkt_queue */
3155 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3160 #endif /* CONFIG_HOTPLUG_CPU */
3164 * Initialize the DEV module. At boot time this walks the device list and
3165 * unhooks any devices that fail to initialise (normally hardware not
3166 * present) and leaves us with a valid list of present and active devices.
3171 * This is called single threaded during boot, so no need
3172 * to take the rtnl semaphore.
3174 static int __init net_dev_init(void)
3176 int i, rc = -ENOMEM;
3178 BUG_ON(!dev_boot_phase);
3182 if (dev_proc_init())
3185 if (netdev_sysfs_init())
3188 INIT_LIST_HEAD(&ptype_all);
3189 for (i = 0; i < 16; i++)
3190 INIT_LIST_HEAD(&ptype_base[i]);
3192 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3193 INIT_HLIST_HEAD(&dev_name_head[i]);
3195 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3196 INIT_HLIST_HEAD(&dev_index_head[i]);
3199 * Initialise the packet receive queues.
3202 for (i = 0; i < NR_CPUS; i++) {
3203 struct softnet_data *queue;
3205 queue = &per_cpu(softnet_data, i);
3206 skb_queue_head_init(&queue->input_pkt_queue);
3207 queue->completion_queue = NULL;
3208 INIT_LIST_HEAD(&queue->poll_list);
3209 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3210 queue->backlog_dev.weight = weight_p;
3211 queue->backlog_dev.poll = process_backlog;
3212 atomic_set(&queue->backlog_dev.refcnt, 1);
3217 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3218 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3220 hotcpu_notifier(dev_cpu_callback, 0);
3228 subsys_initcall(net_dev_init);
3230 EXPORT_SYMBOL(__dev_get_by_index);
3231 EXPORT_SYMBOL(__dev_get_by_name);
3232 EXPORT_SYMBOL(__dev_remove_pack);
3233 EXPORT_SYMBOL(__skb_linearize);
3234 EXPORT_SYMBOL(dev_add_pack);
3235 EXPORT_SYMBOL(dev_alloc_name);
3236 EXPORT_SYMBOL(dev_close);
3237 EXPORT_SYMBOL(dev_get_by_flags);
3238 EXPORT_SYMBOL(dev_get_by_index);
3239 EXPORT_SYMBOL(dev_get_by_name);
3240 EXPORT_SYMBOL(dev_ioctl);
3241 EXPORT_SYMBOL(dev_open);
3242 EXPORT_SYMBOL(dev_queue_xmit);
3243 EXPORT_SYMBOL(dev_remove_pack);
3244 EXPORT_SYMBOL(dev_set_allmulti);
3245 EXPORT_SYMBOL(dev_set_promiscuity);
3246 EXPORT_SYMBOL(dev_change_flags);
3247 EXPORT_SYMBOL(dev_set_mtu);
3248 EXPORT_SYMBOL(dev_set_mac_address);
3249 EXPORT_SYMBOL(free_netdev);
3250 EXPORT_SYMBOL(netdev_boot_setup_check);
3251 EXPORT_SYMBOL(netdev_set_master);
3252 EXPORT_SYMBOL(netdev_state_change);
3253 EXPORT_SYMBOL(netif_receive_skb);
3254 EXPORT_SYMBOL(netif_rx);
3255 EXPORT_SYMBOL(register_gifconf);
3256 EXPORT_SYMBOL(register_netdevice);
3257 EXPORT_SYMBOL(register_netdevice_notifier);
3258 EXPORT_SYMBOL(skb_checksum_help);
3259 EXPORT_SYMBOL(synchronize_net);
3260 EXPORT_SYMBOL(unregister_netdevice);
3261 EXPORT_SYMBOL(unregister_netdevice_notifier);
3262 EXPORT_SYMBOL(net_enable_timestamp);
3263 EXPORT_SYMBOL(net_disable_timestamp);
3264 EXPORT_SYMBOL(dev_get_flags);
3266 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3267 EXPORT_SYMBOL(br_handle_frame_hook);
3268 EXPORT_SYMBOL(br_fdb_get_hook);
3269 EXPORT_SYMBOL(br_fdb_put_hook);
3273 EXPORT_SYMBOL(dev_load);
3276 EXPORT_PER_CPU_SYMBOL(softnet_data);