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/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/notifier.h>
94 #include <linux/skbuff.h>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
101 #include <linux/divert.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/kmod.h>
108 #include <linux/module.h>
109 #include <linux/kallsyms.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #ifdef CONFIG_NET_RADIO
114 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
115 #include <net/iw_handler.h>
116 #endif /* CONFIG_NET_RADIO */
117 #include <asm/current.h>
118 #include <linux/audit.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 adversly 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 */
153 * The @dev_base list is protected by @dev_base_lock and the rtln
156 * Pure readers hold dev_base_lock for reading.
158 * Writers must hold the rtnl semaphore while they loop through the
159 * dev_base list, and hold dev_base_lock for writing when they do the
160 * actual updates. This allows pure readers to access the list even
161 * while a writer is preparing to update it.
163 * To put it another way, dev_base_lock is held for writing only to
164 * protect against pure readers; the rtnl semaphore provides the
165 * protection against other writers.
167 * See, for example usages, register_netdevice() and
168 * unregister_netdevice(), which must be called with the rtnl
171 struct net_device *dev_base;
172 static struct net_device **dev_tail = &dev_base;
173 DEFINE_RWLOCK(dev_base_lock);
175 EXPORT_SYMBOL(dev_base);
176 EXPORT_SYMBOL(dev_base_lock);
178 #define NETDEV_HASHBITS 8
179 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
180 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
182 static inline struct hlist_head *dev_name_hash(const char *name)
184 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
185 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
188 static inline struct hlist_head *dev_index_hash(int ifindex)
190 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
197 static struct notifier_block *netdev_chain;
200 * Device drivers call our routines to queue packets here. We empty the
201 * queue in the local softnet handler.
203 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
206 extern int netdev_sysfs_init(void);
207 extern int netdev_register_sysfs(struct net_device *);
208 extern void netdev_unregister_sysfs(struct net_device *);
210 #define netdev_sysfs_init() (0)
211 #define netdev_register_sysfs(dev) (0)
212 #define netdev_unregister_sysfs(dev) do { } while(0)
216 /*******************************************************************************
218 Protocol management and registration routines
220 *******************************************************************************/
229 * Add a protocol ID to the list. Now that the input handler is
230 * smarter we can dispense with all the messy stuff that used to be
233 * BEWARE!!! Protocol handlers, mangling input packets,
234 * MUST BE last in hash buckets and checking protocol handlers
235 * MUST start from promiscuous ptype_all chain in net_bh.
236 * It is true now, do not change it.
237 * Explanation follows: if protocol handler, mangling packet, will
238 * be the first on list, it is not able to sense, that packet
239 * is cloned and should be copied-on-write, so that it will
240 * change it and subsequent readers will get broken packet.
245 * dev_add_pack - add packet handler
246 * @pt: packet type declaration
248 * Add a protocol handler to the networking stack. The passed &packet_type
249 * is linked into kernel lists and may not be freed until it has been
250 * removed from the kernel lists.
252 * This call does not sleep therefore it can not
253 * guarantee all CPU's that are in middle of receiving packets
254 * will see the new packet type (until the next received packet).
257 void dev_add_pack(struct packet_type *pt)
261 spin_lock_bh(&ptype_lock);
262 if (pt->type == htons(ETH_P_ALL)) {
264 list_add_rcu(&pt->list, &ptype_all);
266 hash = ntohs(pt->type) & 15;
267 list_add_rcu(&pt->list, &ptype_base[hash]);
269 spin_unlock_bh(&ptype_lock);
273 * __dev_remove_pack - remove packet handler
274 * @pt: packet type declaration
276 * Remove a protocol handler that was previously added to the kernel
277 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
278 * from the kernel lists and can be freed or reused once this function
281 * The packet type might still be in use by receivers
282 * and must not be freed until after all the CPU's have gone
283 * through a quiescent state.
285 void __dev_remove_pack(struct packet_type *pt)
287 struct list_head *head;
288 struct packet_type *pt1;
290 spin_lock_bh(&ptype_lock);
292 if (pt->type == htons(ETH_P_ALL)) {
296 head = &ptype_base[ntohs(pt->type) & 15];
298 list_for_each_entry(pt1, head, list) {
300 list_del_rcu(&pt->list);
305 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
307 spin_unlock_bh(&ptype_lock);
310 * dev_remove_pack - remove packet handler
311 * @pt: packet type declaration
313 * Remove a protocol handler that was previously added to the kernel
314 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
315 * from the kernel lists and can be freed or reused once this function
318 * This call sleeps to guarantee that no CPU is looking at the packet
321 void dev_remove_pack(struct packet_type *pt)
323 __dev_remove_pack(pt);
328 /******************************************************************************
330 Device Boot-time Settings Routines
332 *******************************************************************************/
334 /* Boot time configuration table */
335 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
338 * netdev_boot_setup_add - add new setup entry
339 * @name: name of the device
340 * @map: configured settings for the device
342 * Adds new setup entry to the dev_boot_setup list. The function
343 * returns 0 on error and 1 on success. This is a generic routine to
346 static int netdev_boot_setup_add(char *name, struct ifmap *map)
348 struct netdev_boot_setup *s;
352 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
353 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
354 memset(s[i].name, 0, sizeof(s[i].name));
355 strcpy(s[i].name, name);
356 memcpy(&s[i].map, map, sizeof(s[i].map));
361 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
365 * netdev_boot_setup_check - check boot time settings
366 * @dev: the netdevice
368 * Check boot time settings for the device.
369 * The found settings are set for the device to be used
370 * later in the device probing.
371 * Returns 0 if no settings found, 1 if they are.
373 int netdev_boot_setup_check(struct net_device *dev)
375 struct netdev_boot_setup *s = dev_boot_setup;
378 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
379 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
380 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
381 dev->irq = s[i].map.irq;
382 dev->base_addr = s[i].map.base_addr;
383 dev->mem_start = s[i].map.mem_start;
384 dev->mem_end = s[i].map.mem_end;
393 * netdev_boot_base - get address from boot time settings
394 * @prefix: prefix for network device
395 * @unit: id for network device
397 * Check boot time settings for the base address of device.
398 * The found settings are set for the device to be used
399 * later in the device probing.
400 * Returns 0 if no settings found.
402 unsigned long netdev_boot_base(const char *prefix, int unit)
404 const struct netdev_boot_setup *s = dev_boot_setup;
408 sprintf(name, "%s%d", prefix, unit);
411 * If device already registered then return base of 1
412 * to indicate not to probe for this interface
414 if (__dev_get_by_name(name))
417 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
418 if (!strcmp(name, s[i].name))
419 return s[i].map.base_addr;
424 * Saves at boot time configured settings for any netdevice.
426 int __init netdev_boot_setup(char *str)
431 str = get_options(str, ARRAY_SIZE(ints), ints);
436 memset(&map, 0, sizeof(map));
440 map.base_addr = ints[2];
442 map.mem_start = ints[3];
444 map.mem_end = ints[4];
446 /* Add new entry to the list */
447 return netdev_boot_setup_add(str, &map);
450 __setup("netdev=", netdev_boot_setup);
452 /*******************************************************************************
454 Device Interface Subroutines
456 *******************************************************************************/
459 * __dev_get_by_name - find a device by its name
460 * @name: name to find
462 * Find an interface by name. Must be called under RTNL semaphore
463 * or @dev_base_lock. If the name is found a pointer to the device
464 * is returned. If the name is not found then %NULL is returned. The
465 * reference counters are not incremented so the caller must be
466 * careful with locks.
469 struct net_device *__dev_get_by_name(const char *name)
471 struct hlist_node *p;
473 hlist_for_each(p, dev_name_hash(name)) {
474 struct net_device *dev
475 = hlist_entry(p, struct net_device, name_hlist);
476 if (!strncmp(dev->name, name, IFNAMSIZ))
483 * dev_get_by_name - find a device by its name
484 * @name: name to find
486 * Find an interface by name. This can be called from any
487 * context and does its own locking. The returned handle has
488 * the usage count incremented and the caller must use dev_put() to
489 * release it when it is no longer needed. %NULL is returned if no
490 * matching device is found.
493 struct net_device *dev_get_by_name(const char *name)
495 struct net_device *dev;
497 read_lock(&dev_base_lock);
498 dev = __dev_get_by_name(name);
501 read_unlock(&dev_base_lock);
506 * __dev_get_by_index - find a device by its ifindex
507 * @ifindex: index of device
509 * Search for an interface by index. Returns %NULL if the device
510 * is not found or a pointer to the device. The device has not
511 * had its reference counter increased so the caller must be careful
512 * about locking. The caller must hold either the RTNL semaphore
516 struct net_device *__dev_get_by_index(int ifindex)
518 struct hlist_node *p;
520 hlist_for_each(p, dev_index_hash(ifindex)) {
521 struct net_device *dev
522 = hlist_entry(p, struct net_device, index_hlist);
523 if (dev->ifindex == ifindex)
531 * dev_get_by_index - find a device by its ifindex
532 * @ifindex: index of device
534 * Search for an interface by index. Returns NULL if the device
535 * is not found or a pointer to the device. The device returned has
536 * had a reference added and the pointer is safe until the user calls
537 * dev_put to indicate they have finished with it.
540 struct net_device *dev_get_by_index(int ifindex)
542 struct net_device *dev;
544 read_lock(&dev_base_lock);
545 dev = __dev_get_by_index(ifindex);
548 read_unlock(&dev_base_lock);
553 * dev_getbyhwaddr - find a device by its hardware address
554 * @type: media type of device
555 * @ha: hardware address
557 * Search for an interface by MAC address. Returns NULL if the device
558 * is not found or a pointer to the device. The caller must hold the
559 * rtnl semaphore. The returned device has not had its ref count increased
560 * and the caller must therefore be careful about locking
563 * If the API was consistent this would be __dev_get_by_hwaddr
566 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
568 struct net_device *dev;
572 for (dev = dev_base; dev; dev = dev->next)
573 if (dev->type == type &&
574 !memcmp(dev->dev_addr, ha, dev->addr_len))
579 EXPORT_SYMBOL(dev_getbyhwaddr);
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 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 void __net_timestamp(struct sk_buff *skb)
1016 do_gettimeofday(&tv);
1017 skb_set_timestamp(skb, &tv);
1019 EXPORT_SYMBOL(__net_timestamp);
1021 static inline void net_timestamp(struct sk_buff *skb)
1023 if (atomic_read(&netstamp_needed))
1024 __net_timestamp(skb);
1026 skb->tstamp.off_sec = 0;
1027 skb->tstamp.off_usec = 0;
1032 * Support routine. Sends outgoing frames to any network
1033 * taps currently in use.
1036 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1038 struct packet_type *ptype;
1043 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1044 /* Never send packets back to the socket
1045 * they originated from - MvS (miquels@drinkel.ow.org)
1047 if ((ptype->dev == dev || !ptype->dev) &&
1048 (ptype->af_packet_priv == NULL ||
1049 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1050 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1054 /* skb->nh should be correctly
1055 set by sender, so that the second statement is
1056 just protection against buggy protocols.
1058 skb2->mac.raw = skb2->data;
1060 if (skb2->nh.raw < skb2->data ||
1061 skb2->nh.raw > skb2->tail) {
1062 if (net_ratelimit())
1063 printk(KERN_CRIT "protocol %04x is "
1065 skb2->protocol, dev->name);
1066 skb2->nh.raw = skb2->data;
1069 skb2->h.raw = skb2->nh.raw;
1070 skb2->pkt_type = PACKET_OUTGOING;
1071 ptype->func(skb2, skb->dev, ptype, skb->dev);
1078 * Invalidate hardware checksum when packet is to be mangled, and
1079 * complete checksum manually on outgoing path.
1081 int skb_checksum_help(struct sk_buff *skb, int inward)
1084 int ret = 0, offset = skb->h.raw - skb->data;
1087 skb->ip_summed = CHECKSUM_NONE;
1091 if (skb_cloned(skb)) {
1092 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1097 BUG_ON(offset > (int)skb->len);
1098 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1100 offset = skb->tail - skb->h.raw;
1101 BUG_ON(offset <= 0);
1102 BUG_ON(skb->csum + 2 > offset);
1104 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1105 skb->ip_summed = CHECKSUM_NONE;
1110 /* Take action when hardware reception checksum errors are detected. */
1112 void netdev_rx_csum_fault(struct net_device *dev)
1114 if (net_ratelimit()) {
1115 printk(KERN_ERR "%s: hw csum failure.\n",
1116 dev ? dev->name : "<unknown>");
1120 EXPORT_SYMBOL(netdev_rx_csum_fault);
1123 #ifdef CONFIG_HIGHMEM
1124 /* Actually, we should eliminate this check as soon as we know, that:
1125 * 1. IOMMU is present and allows to map all the memory.
1126 * 2. No high memory really exists on this machine.
1129 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1133 if (dev->features & NETIF_F_HIGHDMA)
1136 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1137 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1143 #define illegal_highdma(dev, skb) (0)
1146 /* Keep head the same: replace data */
1147 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1152 struct skb_shared_info *ninfo;
1153 int headerlen = skb->data - skb->head;
1154 int expand = (skb->tail + skb->data_len) - skb->end;
1156 if (skb_shared(skb))
1162 size = skb->end - skb->head + expand;
1163 size = SKB_DATA_ALIGN(size);
1164 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1168 /* Copy entire thing */
1169 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1173 ninfo = (struct skb_shared_info*)(data + size);
1174 atomic_set(&ninfo->dataref, 1);
1175 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1176 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1177 ninfo->nr_frags = 0;
1178 ninfo->frag_list = NULL;
1180 /* Offset between the two in bytes */
1181 offset = data - skb->head;
1183 /* Free old data. */
1184 skb_release_data(skb);
1187 skb->end = data + size;
1189 /* Set up new pointers */
1190 skb->h.raw += offset;
1191 skb->nh.raw += offset;
1192 skb->mac.raw += offset;
1193 skb->tail += offset;
1194 skb->data += offset;
1196 /* We are no longer a clone, even if we were. */
1199 skb->tail += skb->data_len;
1204 #define HARD_TX_LOCK(dev, cpu) { \
1205 if ((dev->features & NETIF_F_LLTX) == 0) { \
1206 spin_lock(&dev->xmit_lock); \
1207 dev->xmit_lock_owner = cpu; \
1211 #define HARD_TX_UNLOCK(dev) { \
1212 if ((dev->features & NETIF_F_LLTX) == 0) { \
1213 dev->xmit_lock_owner = -1; \
1214 spin_unlock(&dev->xmit_lock); \
1219 * dev_queue_xmit - transmit a buffer
1220 * @skb: buffer to transmit
1222 * Queue a buffer for transmission to a network device. The caller must
1223 * have set the device and priority and built the buffer before calling
1224 * this function. The function can be called from an interrupt.
1226 * A negative errno code is returned on a failure. A success does not
1227 * guarantee the frame will be transmitted as it may be dropped due
1228 * to congestion or traffic shaping.
1230 * -----------------------------------------------------------------------------------
1231 * I notice this method can also return errors from the queue disciplines,
1232 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1235 * Regardless of the return value, the skb is consumed, so it is currently
1236 * difficult to retry a send to this method. (You can bump the ref count
1237 * before sending to hold a reference for retry if you are careful.)
1239 * When calling this method, interrupts MUST be enabled. This is because
1240 * the BH enable code must have IRQs enabled so that it will not deadlock.
1244 int dev_queue_xmit(struct sk_buff *skb)
1246 struct net_device *dev = skb->dev;
1250 if (skb_shinfo(skb)->frag_list &&
1251 !(dev->features & NETIF_F_FRAGLIST) &&
1252 __skb_linearize(skb, GFP_ATOMIC))
1255 /* Fragmented skb is linearized if device does not support SG,
1256 * or if at least one of fragments is in highmem and device
1257 * does not support DMA from it.
1259 if (skb_shinfo(skb)->nr_frags &&
1260 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1261 __skb_linearize(skb, GFP_ATOMIC))
1264 /* If packet is not checksummed and device does not support
1265 * checksumming for this protocol, complete checksumming here.
1267 if (skb->ip_summed == CHECKSUM_HW &&
1268 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1269 (!(dev->features & NETIF_F_IP_CSUM) ||
1270 skb->protocol != htons(ETH_P_IP))))
1271 if (skb_checksum_help(skb, 0))
1274 spin_lock_prefetch(&dev->queue_lock);
1276 /* Disable soft irqs for various locks below. Also
1277 * stops preemption for RCU.
1281 /* Updates of qdisc are serialized by queue_lock.
1282 * The struct Qdisc which is pointed to by qdisc is now a
1283 * rcu structure - it may be accessed without acquiring
1284 * a lock (but the structure may be stale.) The freeing of the
1285 * qdisc will be deferred until it's known that there are no
1286 * more references to it.
1288 * If the qdisc has an enqueue function, we still need to
1289 * hold the queue_lock before calling it, since queue_lock
1290 * also serializes access to the device queue.
1293 q = rcu_dereference(dev->qdisc);
1294 #ifdef CONFIG_NET_CLS_ACT
1295 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1298 /* Grab device queue */
1299 spin_lock(&dev->queue_lock);
1301 rc = q->enqueue(skb, q);
1305 spin_unlock(&dev->queue_lock);
1306 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1310 /* The device has no queue. Common case for software devices:
1311 loopback, all the sorts of tunnels...
1313 Really, it is unlikely that xmit_lock protection is necessary here.
1314 (f.e. loopback and IP tunnels are clean ignoring statistics
1316 However, it is possible, that they rely on protection
1319 Check this and shot the lock. It is not prone from deadlocks.
1320 Either shot noqueue qdisc, it is even simpler 8)
1322 if (dev->flags & IFF_UP) {
1323 int cpu = smp_processor_id(); /* ok because BHs are off */
1325 if (dev->xmit_lock_owner != cpu) {
1327 HARD_TX_LOCK(dev, cpu);
1329 if (!netif_queue_stopped(dev)) {
1331 dev_queue_xmit_nit(skb, dev);
1334 if (!dev->hard_start_xmit(skb, dev)) {
1335 HARD_TX_UNLOCK(dev);
1339 HARD_TX_UNLOCK(dev);
1340 if (net_ratelimit())
1341 printk(KERN_CRIT "Virtual device %s asks to "
1342 "queue packet!\n", dev->name);
1344 /* Recursion is detected! It is possible,
1346 if (net_ratelimit())
1347 printk(KERN_CRIT "Dead loop on virtual device "
1348 "%s, fix it urgently!\n", dev->name);
1364 /*=======================================================================
1366 =======================================================================*/
1368 int netdev_max_backlog = 1000;
1369 int netdev_budget = 300;
1370 int weight_p = 64; /* old backlog weight */
1372 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1376 * netif_rx - post buffer to the network code
1377 * @skb: buffer to post
1379 * This function receives a packet from a device driver and queues it for
1380 * the upper (protocol) levels to process. It always succeeds. The buffer
1381 * may be dropped during processing for congestion control or by the
1385 * NET_RX_SUCCESS (no congestion)
1386 * NET_RX_CN_LOW (low congestion)
1387 * NET_RX_CN_MOD (moderate congestion)
1388 * NET_RX_CN_HIGH (high congestion)
1389 * NET_RX_DROP (packet was dropped)
1393 int netif_rx(struct sk_buff *skb)
1395 struct softnet_data *queue;
1396 unsigned long flags;
1398 /* if netpoll wants it, pretend we never saw it */
1399 if (netpoll_rx(skb))
1402 if (!skb->tstamp.off_sec)
1406 * The code is rearranged so that the path is the most
1407 * short when CPU is congested, but is still operating.
1409 local_irq_save(flags);
1410 queue = &__get_cpu_var(softnet_data);
1412 __get_cpu_var(netdev_rx_stat).total++;
1413 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1414 if (queue->input_pkt_queue.qlen) {
1417 __skb_queue_tail(&queue->input_pkt_queue, skb);
1418 local_irq_restore(flags);
1419 return NET_RX_SUCCESS;
1422 netif_rx_schedule(&queue->backlog_dev);
1426 __get_cpu_var(netdev_rx_stat).dropped++;
1427 local_irq_restore(flags);
1433 int netif_rx_ni(struct sk_buff *skb)
1438 err = netif_rx(skb);
1439 if (local_softirq_pending())
1446 EXPORT_SYMBOL(netif_rx_ni);
1448 static inline struct net_device *skb_bond(struct sk_buff *skb)
1450 struct net_device *dev = skb->dev;
1453 skb->dev = dev->master;
1458 static void net_tx_action(struct softirq_action *h)
1460 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1462 if (sd->completion_queue) {
1463 struct sk_buff *clist;
1465 local_irq_disable();
1466 clist = sd->completion_queue;
1467 sd->completion_queue = NULL;
1471 struct sk_buff *skb = clist;
1472 clist = clist->next;
1474 BUG_TRAP(!atomic_read(&skb->users));
1479 if (sd->output_queue) {
1480 struct net_device *head;
1482 local_irq_disable();
1483 head = sd->output_queue;
1484 sd->output_queue = NULL;
1488 struct net_device *dev = head;
1489 head = head->next_sched;
1491 smp_mb__before_clear_bit();
1492 clear_bit(__LINK_STATE_SCHED, &dev->state);
1494 if (spin_trylock(&dev->queue_lock)) {
1496 spin_unlock(&dev->queue_lock);
1498 netif_schedule(dev);
1504 static __inline__ int deliver_skb(struct sk_buff *skb,
1505 struct packet_type *pt_prev,
1506 struct net_device *orig_dev)
1508 atomic_inc(&skb->users);
1509 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1512 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1513 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1515 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1516 unsigned char *addr);
1517 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1519 static __inline__ int handle_bridge(struct sk_buff **pskb,
1520 struct packet_type **pt_prev, int *ret,
1521 struct net_device *orig_dev)
1523 struct net_bridge_port *port;
1525 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1526 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1530 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1534 return br_handle_frame_hook(port, pskb);
1537 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1540 #ifdef CONFIG_NET_CLS_ACT
1541 /* TODO: Maybe we should just force sch_ingress to be compiled in
1542 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1543 * a compare and 2 stores extra right now if we dont have it on
1544 * but have CONFIG_NET_CLS_ACT
1545 * NOTE: This doesnt stop any functionality; if you dont have
1546 * the ingress scheduler, you just cant add policies on ingress.
1549 static int ing_filter(struct sk_buff *skb)
1552 struct net_device *dev = skb->dev;
1553 int result = TC_ACT_OK;
1555 if (dev->qdisc_ingress) {
1556 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1557 if (MAX_RED_LOOP < ttl++) {
1558 printk("Redir loop detected Dropping packet (%s->%s)\n",
1559 skb->input_dev->name, skb->dev->name);
1563 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1565 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1567 spin_lock(&dev->ingress_lock);
1568 if ((q = dev->qdisc_ingress) != NULL)
1569 result = q->enqueue(skb, q);
1570 spin_unlock(&dev->ingress_lock);
1578 int netif_receive_skb(struct sk_buff *skb)
1580 struct packet_type *ptype, *pt_prev;
1581 struct net_device *orig_dev;
1582 int ret = NET_RX_DROP;
1583 unsigned short type;
1585 /* if we've gotten here through NAPI, check netpoll */
1586 if (skb->dev->poll && netpoll_rx(skb))
1589 if (!skb->tstamp.off_sec)
1592 if (!skb->input_dev)
1593 skb->input_dev = skb->dev;
1595 orig_dev = skb_bond(skb);
1597 __get_cpu_var(netdev_rx_stat).total++;
1599 skb->h.raw = skb->nh.raw = skb->data;
1600 skb->mac_len = skb->nh.raw - skb->mac.raw;
1606 #ifdef CONFIG_NET_CLS_ACT
1607 if (skb->tc_verd & TC_NCLS) {
1608 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1613 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1614 if (!ptype->dev || ptype->dev == skb->dev) {
1616 ret = deliver_skb(skb, pt_prev, orig_dev);
1621 #ifdef CONFIG_NET_CLS_ACT
1623 ret = deliver_skb(skb, pt_prev, orig_dev);
1624 pt_prev = NULL; /* noone else should process this after*/
1626 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1629 ret = ing_filter(skb);
1631 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1640 handle_diverter(skb);
1642 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1645 type = skb->protocol;
1646 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1647 if (ptype->type == type &&
1648 (!ptype->dev || ptype->dev == skb->dev)) {
1650 ret = deliver_skb(skb, pt_prev, orig_dev);
1656 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1659 /* Jamal, now you will not able to escape explaining
1660 * me how you were going to use this. :-)
1670 static int process_backlog(struct net_device *backlog_dev, int *budget)
1673 int quota = min(backlog_dev->quota, *budget);
1674 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1675 unsigned long start_time = jiffies;
1677 backlog_dev->weight = weight_p;
1679 struct sk_buff *skb;
1680 struct net_device *dev;
1682 local_irq_disable();
1683 skb = __skb_dequeue(&queue->input_pkt_queue);
1690 netif_receive_skb(skb);
1696 if (work >= quota || jiffies - start_time > 1)
1701 backlog_dev->quota -= work;
1706 backlog_dev->quota -= work;
1709 list_del(&backlog_dev->poll_list);
1710 smp_mb__before_clear_bit();
1711 netif_poll_enable(backlog_dev);
1717 static void net_rx_action(struct softirq_action *h)
1719 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1720 unsigned long start_time = jiffies;
1721 int budget = netdev_budget;
1724 local_irq_disable();
1726 while (!list_empty(&queue->poll_list)) {
1727 struct net_device *dev;
1729 if (budget <= 0 || jiffies - start_time > 1)
1734 dev = list_entry(queue->poll_list.next,
1735 struct net_device, poll_list);
1736 have = netpoll_poll_lock(dev);
1738 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1739 netpoll_poll_unlock(have);
1740 local_irq_disable();
1741 list_del(&dev->poll_list);
1742 list_add_tail(&dev->poll_list, &queue->poll_list);
1744 dev->quota += dev->weight;
1746 dev->quota = dev->weight;
1748 netpoll_poll_unlock(have);
1750 local_irq_disable();
1758 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1759 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1763 static gifconf_func_t * gifconf_list [NPROTO];
1766 * register_gifconf - register a SIOCGIF handler
1767 * @family: Address family
1768 * @gifconf: Function handler
1770 * Register protocol dependent address dumping routines. The handler
1771 * that is passed must not be freed or reused until it has been replaced
1772 * by another handler.
1774 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1776 if (family >= NPROTO)
1778 gifconf_list[family] = gifconf;
1784 * Map an interface index to its name (SIOCGIFNAME)
1788 * We need this ioctl for efficient implementation of the
1789 * if_indextoname() function required by the IPv6 API. Without
1790 * it, we would have to search all the interfaces to find a
1794 static int dev_ifname(struct ifreq __user *arg)
1796 struct net_device *dev;
1800 * Fetch the caller's info block.
1803 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1806 read_lock(&dev_base_lock);
1807 dev = __dev_get_by_index(ifr.ifr_ifindex);
1809 read_unlock(&dev_base_lock);
1813 strcpy(ifr.ifr_name, dev->name);
1814 read_unlock(&dev_base_lock);
1816 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1822 * Perform a SIOCGIFCONF call. This structure will change
1823 * size eventually, and there is nothing I can do about it.
1824 * Thus we will need a 'compatibility mode'.
1827 static int dev_ifconf(char __user *arg)
1830 struct net_device *dev;
1837 * Fetch the caller's info block.
1840 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1847 * Loop over the interfaces, and write an info block for each.
1851 for (dev = dev_base; dev; dev = dev->next) {
1852 for (i = 0; i < NPROTO; i++) {
1853 if (gifconf_list[i]) {
1856 done = gifconf_list[i](dev, NULL, 0);
1858 done = gifconf_list[i](dev, pos + total,
1868 * All done. Write the updated control block back to the caller.
1870 ifc.ifc_len = total;
1873 * Both BSD and Solaris return 0 here, so we do too.
1875 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1878 #ifdef CONFIG_PROC_FS
1880 * This is invoked by the /proc filesystem handler to display a device
1883 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1885 struct net_device *dev;
1888 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1890 return i == pos ? dev : NULL;
1893 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1895 read_lock(&dev_base_lock);
1896 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1899 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1902 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1905 void dev_seq_stop(struct seq_file *seq, void *v)
1907 read_unlock(&dev_base_lock);
1910 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1912 if (dev->get_stats) {
1913 struct net_device_stats *stats = dev->get_stats(dev);
1915 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1916 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1917 dev->name, stats->rx_bytes, stats->rx_packets,
1919 stats->rx_dropped + stats->rx_missed_errors,
1920 stats->rx_fifo_errors,
1921 stats->rx_length_errors + stats->rx_over_errors +
1922 stats->rx_crc_errors + stats->rx_frame_errors,
1923 stats->rx_compressed, stats->multicast,
1924 stats->tx_bytes, stats->tx_packets,
1925 stats->tx_errors, stats->tx_dropped,
1926 stats->tx_fifo_errors, stats->collisions,
1927 stats->tx_carrier_errors +
1928 stats->tx_aborted_errors +
1929 stats->tx_window_errors +
1930 stats->tx_heartbeat_errors,
1931 stats->tx_compressed);
1933 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1937 * Called from the PROCfs module. This now uses the new arbitrary sized
1938 * /proc/net interface to create /proc/net/dev
1940 static int dev_seq_show(struct seq_file *seq, void *v)
1942 if (v == SEQ_START_TOKEN)
1943 seq_puts(seq, "Inter-| Receive "
1945 " face |bytes packets errs drop fifo frame "
1946 "compressed multicast|bytes packets errs "
1947 "drop fifo colls carrier compressed\n");
1949 dev_seq_printf_stats(seq, v);
1953 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1955 struct netif_rx_stats *rc = NULL;
1957 while (*pos < NR_CPUS)
1958 if (cpu_online(*pos)) {
1959 rc = &per_cpu(netdev_rx_stat, *pos);
1966 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1968 return softnet_get_online(pos);
1971 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1974 return softnet_get_online(pos);
1977 static void softnet_seq_stop(struct seq_file *seq, void *v)
1981 static int softnet_seq_show(struct seq_file *seq, void *v)
1983 struct netif_rx_stats *s = v;
1985 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
1986 s->total, s->dropped, s->time_squeeze, 0,
1987 0, 0, 0, 0, /* was fastroute */
1992 static struct seq_operations dev_seq_ops = {
1993 .start = dev_seq_start,
1994 .next = dev_seq_next,
1995 .stop = dev_seq_stop,
1996 .show = dev_seq_show,
1999 static int dev_seq_open(struct inode *inode, struct file *file)
2001 return seq_open(file, &dev_seq_ops);
2004 static struct file_operations dev_seq_fops = {
2005 .owner = THIS_MODULE,
2006 .open = dev_seq_open,
2008 .llseek = seq_lseek,
2009 .release = seq_release,
2012 static struct seq_operations softnet_seq_ops = {
2013 .start = softnet_seq_start,
2014 .next = softnet_seq_next,
2015 .stop = softnet_seq_stop,
2016 .show = softnet_seq_show,
2019 static int softnet_seq_open(struct inode *inode, struct file *file)
2021 return seq_open(file, &softnet_seq_ops);
2024 static struct file_operations softnet_seq_fops = {
2025 .owner = THIS_MODULE,
2026 .open = softnet_seq_open,
2028 .llseek = seq_lseek,
2029 .release = seq_release,
2033 extern int wireless_proc_init(void);
2035 #define wireless_proc_init() 0
2038 static int __init dev_proc_init(void)
2042 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2044 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2046 if (wireless_proc_init())
2052 proc_net_remove("softnet_stat");
2054 proc_net_remove("dev");
2058 #define dev_proc_init() 0
2059 #endif /* CONFIG_PROC_FS */
2063 * netdev_set_master - set up master/slave pair
2064 * @slave: slave device
2065 * @master: new master device
2067 * Changes the master device of the slave. Pass %NULL to break the
2068 * bonding. The caller must hold the RTNL semaphore. On a failure
2069 * a negative errno code is returned. On success the reference counts
2070 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2071 * function returns zero.
2073 int netdev_set_master(struct net_device *slave, struct net_device *master)
2075 struct net_device *old = slave->master;
2085 slave->master = master;
2093 slave->flags |= IFF_SLAVE;
2095 slave->flags &= ~IFF_SLAVE;
2097 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2102 * dev_set_promiscuity - update promiscuity count on a device
2106 * Add or remove promsicuity from a device. While the count in the device
2107 * remains above zero the interface remains promiscuous. Once it hits zero
2108 * the device reverts back to normal filtering operation. A negative inc
2109 * value is used to drop promiscuity on the device.
2111 void dev_set_promiscuity(struct net_device *dev, int inc)
2113 unsigned short old_flags = dev->flags;
2115 if ((dev->promiscuity += inc) == 0)
2116 dev->flags &= ~IFF_PROMISC;
2118 dev->flags |= IFF_PROMISC;
2119 if (dev->flags != old_flags) {
2121 printk(KERN_INFO "device %s %s promiscuous mode\n",
2122 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2124 audit_log(current->audit_context, GFP_ATOMIC,
2125 AUDIT_ANOM_PROMISCUOUS,
2126 "dev=%s prom=%d old_prom=%d auid=%u",
2127 dev->name, (dev->flags & IFF_PROMISC),
2128 (old_flags & IFF_PROMISC),
2129 audit_get_loginuid(current->audit_context));
2134 * dev_set_allmulti - update allmulti count on a device
2138 * Add or remove reception of all multicast frames to a device. While the
2139 * count in the device remains above zero the interface remains listening
2140 * to all interfaces. Once it hits zero the device reverts back to normal
2141 * filtering operation. A negative @inc value is used to drop the counter
2142 * when releasing a resource needing all multicasts.
2145 void dev_set_allmulti(struct net_device *dev, int inc)
2147 unsigned short old_flags = dev->flags;
2149 dev->flags |= IFF_ALLMULTI;
2150 if ((dev->allmulti += inc) == 0)
2151 dev->flags &= ~IFF_ALLMULTI;
2152 if (dev->flags ^ old_flags)
2156 unsigned dev_get_flags(const struct net_device *dev)
2160 flags = (dev->flags & ~(IFF_PROMISC |
2163 (dev->gflags & (IFF_PROMISC |
2166 if (netif_running(dev) && netif_carrier_ok(dev))
2167 flags |= IFF_RUNNING;
2172 int dev_change_flags(struct net_device *dev, unsigned flags)
2175 int old_flags = dev->flags;
2178 * Set the flags on our device.
2181 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2182 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2184 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2188 * Load in the correct multicast list now the flags have changed.
2194 * Have we downed the interface. We handle IFF_UP ourselves
2195 * according to user attempts to set it, rather than blindly
2200 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2201 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2207 if (dev->flags & IFF_UP &&
2208 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2210 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2212 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2213 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2214 dev->gflags ^= IFF_PROMISC;
2215 dev_set_promiscuity(dev, inc);
2218 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2219 is important. Some (broken) drivers set IFF_PROMISC, when
2220 IFF_ALLMULTI is requested not asking us and not reporting.
2222 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2223 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2224 dev->gflags ^= IFF_ALLMULTI;
2225 dev_set_allmulti(dev, inc);
2228 if (old_flags ^ dev->flags)
2229 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2234 int dev_set_mtu(struct net_device *dev, int new_mtu)
2238 if (new_mtu == dev->mtu)
2241 /* MTU must be positive. */
2245 if (!netif_device_present(dev))
2249 if (dev->change_mtu)
2250 err = dev->change_mtu(dev, new_mtu);
2253 if (!err && dev->flags & IFF_UP)
2254 notifier_call_chain(&netdev_chain,
2255 NETDEV_CHANGEMTU, dev);
2259 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2263 if (!dev->set_mac_address)
2265 if (sa->sa_family != dev->type)
2267 if (!netif_device_present(dev))
2269 err = dev->set_mac_address(dev, sa);
2271 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2276 * Perform the SIOCxIFxxx calls.
2278 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2281 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2287 case SIOCGIFFLAGS: /* Get interface flags */
2288 ifr->ifr_flags = dev_get_flags(dev);
2291 case SIOCSIFFLAGS: /* Set interface flags */
2292 return dev_change_flags(dev, ifr->ifr_flags);
2294 case SIOCGIFMETRIC: /* Get the metric on the interface
2295 (currently unused) */
2296 ifr->ifr_metric = 0;
2299 case SIOCSIFMETRIC: /* Set the metric on the interface
2300 (currently unused) */
2303 case SIOCGIFMTU: /* Get the MTU of a device */
2304 ifr->ifr_mtu = dev->mtu;
2307 case SIOCSIFMTU: /* Set the MTU of a device */
2308 return dev_set_mtu(dev, ifr->ifr_mtu);
2312 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2314 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2315 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2316 ifr->ifr_hwaddr.sa_family = dev->type;
2320 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2322 case SIOCSIFHWBROADCAST:
2323 if (ifr->ifr_hwaddr.sa_family != dev->type)
2325 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2326 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2327 notifier_call_chain(&netdev_chain,
2328 NETDEV_CHANGEADDR, dev);
2332 ifr->ifr_map.mem_start = dev->mem_start;
2333 ifr->ifr_map.mem_end = dev->mem_end;
2334 ifr->ifr_map.base_addr = dev->base_addr;
2335 ifr->ifr_map.irq = dev->irq;
2336 ifr->ifr_map.dma = dev->dma;
2337 ifr->ifr_map.port = dev->if_port;
2341 if (dev->set_config) {
2342 if (!netif_device_present(dev))
2344 return dev->set_config(dev, &ifr->ifr_map);
2349 if (!dev->set_multicast_list ||
2350 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2352 if (!netif_device_present(dev))
2354 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2358 if (!dev->set_multicast_list ||
2359 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2361 if (!netif_device_present(dev))
2363 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2367 ifr->ifr_ifindex = dev->ifindex;
2371 ifr->ifr_qlen = dev->tx_queue_len;
2375 if (ifr->ifr_qlen < 0)
2377 dev->tx_queue_len = ifr->ifr_qlen;
2381 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2382 return dev_change_name(dev, ifr->ifr_newname);
2385 * Unknown or private ioctl
2389 if ((cmd >= SIOCDEVPRIVATE &&
2390 cmd <= SIOCDEVPRIVATE + 15) ||
2391 cmd == SIOCBONDENSLAVE ||
2392 cmd == SIOCBONDRELEASE ||
2393 cmd == SIOCBONDSETHWADDR ||
2394 cmd == SIOCBONDSLAVEINFOQUERY ||
2395 cmd == SIOCBONDINFOQUERY ||
2396 cmd == SIOCBONDCHANGEACTIVE ||
2397 cmd == SIOCGMIIPHY ||
2398 cmd == SIOCGMIIREG ||
2399 cmd == SIOCSMIIREG ||
2400 cmd == SIOCBRADDIF ||
2401 cmd == SIOCBRDELIF ||
2402 cmd == SIOCWANDEV) {
2404 if (dev->do_ioctl) {
2405 if (netif_device_present(dev))
2406 err = dev->do_ioctl(dev, ifr,
2419 * This function handles all "interface"-type I/O control requests. The actual
2420 * 'doing' part of this is dev_ifsioc above.
2424 * dev_ioctl - network device ioctl
2425 * @cmd: command to issue
2426 * @arg: pointer to a struct ifreq in user space
2428 * Issue ioctl functions to devices. This is normally called by the
2429 * user space syscall interfaces but can sometimes be useful for
2430 * other purposes. The return value is the return from the syscall if
2431 * positive or a negative errno code on error.
2434 int dev_ioctl(unsigned int cmd, void __user *arg)
2440 /* One special case: SIOCGIFCONF takes ifconf argument
2441 and requires shared lock, because it sleeps writing
2445 if (cmd == SIOCGIFCONF) {
2447 ret = dev_ifconf((char __user *) arg);
2451 if (cmd == SIOCGIFNAME)
2452 return dev_ifname((struct ifreq __user *)arg);
2454 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2457 ifr.ifr_name[IFNAMSIZ-1] = 0;
2459 colon = strchr(ifr.ifr_name, ':');
2464 * See which interface the caller is talking about.
2469 * These ioctl calls:
2470 * - can be done by all.
2471 * - atomic and do not require locking.
2482 dev_load(ifr.ifr_name);
2483 read_lock(&dev_base_lock);
2484 ret = dev_ifsioc(&ifr, cmd);
2485 read_unlock(&dev_base_lock);
2489 if (copy_to_user(arg, &ifr,
2490 sizeof(struct ifreq)))
2496 dev_load(ifr.ifr_name);
2498 ret = dev_ethtool(&ifr);
2503 if (copy_to_user(arg, &ifr,
2504 sizeof(struct ifreq)))
2510 * These ioctl calls:
2511 * - require superuser power.
2512 * - require strict serialization.
2518 if (!capable(CAP_NET_ADMIN))
2520 dev_load(ifr.ifr_name);
2522 ret = dev_ifsioc(&ifr, cmd);
2527 if (copy_to_user(arg, &ifr,
2528 sizeof(struct ifreq)))
2534 * These ioctl calls:
2535 * - require superuser power.
2536 * - require strict serialization.
2537 * - do not return a value
2547 case SIOCSIFHWBROADCAST:
2550 case SIOCBONDENSLAVE:
2551 case SIOCBONDRELEASE:
2552 case SIOCBONDSETHWADDR:
2553 case SIOCBONDCHANGEACTIVE:
2556 if (!capable(CAP_NET_ADMIN))
2559 case SIOCBONDSLAVEINFOQUERY:
2560 case SIOCBONDINFOQUERY:
2561 dev_load(ifr.ifr_name);
2563 ret = dev_ifsioc(&ifr, cmd);
2568 /* Get the per device memory space. We can add this but
2569 * currently do not support it */
2571 /* Set the per device memory buffer space.
2572 * Not applicable in our case */
2577 * Unknown or private ioctl.
2580 if (cmd == SIOCWANDEV ||
2581 (cmd >= SIOCDEVPRIVATE &&
2582 cmd <= SIOCDEVPRIVATE + 15)) {
2583 dev_load(ifr.ifr_name);
2585 ret = dev_ifsioc(&ifr, cmd);
2587 if (!ret && copy_to_user(arg, &ifr,
2588 sizeof(struct ifreq)))
2593 /* Take care of Wireless Extensions */
2594 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2595 /* If command is `set a parameter', or
2596 * `get the encoding parameters', check if
2597 * the user has the right to do it */
2598 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2599 if (!capable(CAP_NET_ADMIN))
2602 dev_load(ifr.ifr_name);
2604 /* Follow me in net/core/wireless.c */
2605 ret = wireless_process_ioctl(&ifr, cmd);
2607 if (IW_IS_GET(cmd) &&
2608 copy_to_user(arg, &ifr,
2609 sizeof(struct ifreq)))
2613 #endif /* WIRELESS_EXT */
2620 * dev_new_index - allocate an ifindex
2622 * Returns a suitable unique value for a new device interface
2623 * number. The caller must hold the rtnl semaphore or the
2624 * dev_base_lock to be sure it remains unique.
2626 static int dev_new_index(void)
2632 if (!__dev_get_by_index(ifindex))
2637 static int dev_boot_phase = 1;
2639 /* Delayed registration/unregisteration */
2640 static DEFINE_SPINLOCK(net_todo_list_lock);
2641 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2643 static inline void net_set_todo(struct net_device *dev)
2645 spin_lock(&net_todo_list_lock);
2646 list_add_tail(&dev->todo_list, &net_todo_list);
2647 spin_unlock(&net_todo_list_lock);
2651 * register_netdevice - register a network device
2652 * @dev: device to register
2654 * Take a completed network device structure and add it to the kernel
2655 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2656 * chain. 0 is returned on success. A negative errno code is returned
2657 * on a failure to set up the device, or if the name is a duplicate.
2659 * Callers must hold the rtnl semaphore. You may want
2660 * register_netdev() instead of this.
2663 * The locking appears insufficient to guarantee two parallel registers
2664 * will not get the same name.
2667 int register_netdevice(struct net_device *dev)
2669 struct hlist_head *head;
2670 struct hlist_node *p;
2673 BUG_ON(dev_boot_phase);
2676 /* When net_device's are persistent, this will be fatal. */
2677 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2679 spin_lock_init(&dev->queue_lock);
2680 spin_lock_init(&dev->xmit_lock);
2681 dev->xmit_lock_owner = -1;
2682 #ifdef CONFIG_NET_CLS_ACT
2683 spin_lock_init(&dev->ingress_lock);
2686 ret = alloc_divert_blk(dev);
2692 /* Init, if this function is available */
2694 ret = dev->init(dev);
2702 if (!dev_valid_name(dev->name)) {
2707 dev->ifindex = dev_new_index();
2708 if (dev->iflink == -1)
2709 dev->iflink = dev->ifindex;
2711 /* Check for existence of name */
2712 head = dev_name_hash(dev->name);
2713 hlist_for_each(p, head) {
2714 struct net_device *d
2715 = hlist_entry(p, struct net_device, name_hlist);
2716 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2722 /* Fix illegal SG+CSUM combinations. */
2723 if ((dev->features & NETIF_F_SG) &&
2724 !(dev->features & (NETIF_F_IP_CSUM |
2726 NETIF_F_HW_CSUM))) {
2727 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2729 dev->features &= ~NETIF_F_SG;
2732 /* TSO requires that SG is present as well. */
2733 if ((dev->features & NETIF_F_TSO) &&
2734 !(dev->features & NETIF_F_SG)) {
2735 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2737 dev->features &= ~NETIF_F_TSO;
2739 if (dev->features & NETIF_F_UFO) {
2740 if (!(dev->features & NETIF_F_HW_CSUM)) {
2741 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2742 "NETIF_F_HW_CSUM feature.\n",
2744 dev->features &= ~NETIF_F_UFO;
2746 if (!(dev->features & NETIF_F_SG)) {
2747 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2748 "NETIF_F_SG feature.\n",
2750 dev->features &= ~NETIF_F_UFO;
2755 * nil rebuild_header routine,
2756 * that should be never called and used as just bug trap.
2759 if (!dev->rebuild_header)
2760 dev->rebuild_header = default_rebuild_header;
2763 * Default initial state at registry is that the
2764 * device is present.
2767 set_bit(__LINK_STATE_PRESENT, &dev->state);
2770 dev_init_scheduler(dev);
2771 write_lock_bh(&dev_base_lock);
2773 dev_tail = &dev->next;
2774 hlist_add_head(&dev->name_hlist, head);
2775 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2777 dev->reg_state = NETREG_REGISTERING;
2778 write_unlock_bh(&dev_base_lock);
2780 /* Notify protocols, that a new device appeared. */
2781 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2783 /* Finish registration after unlock */
2790 free_divert_blk(dev);
2795 * register_netdev - register a network device
2796 * @dev: device to register
2798 * Take a completed network device structure and add it to the kernel
2799 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2800 * chain. 0 is returned on success. A negative errno code is returned
2801 * on a failure to set up the device, or if the name is a duplicate.
2803 * This is a wrapper around register_netdev that takes the rtnl semaphore
2804 * and expands the device name if you passed a format string to
2807 int register_netdev(struct net_device *dev)
2814 * If the name is a format string the caller wants us to do a
2817 if (strchr(dev->name, '%')) {
2818 err = dev_alloc_name(dev, dev->name);
2824 * Back compatibility hook. Kill this one in 2.5
2826 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2827 err = dev_alloc_name(dev, "eth%d");
2832 err = register_netdevice(dev);
2837 EXPORT_SYMBOL(register_netdev);
2840 * netdev_wait_allrefs - wait until all references are gone.
2842 * This is called when unregistering network devices.
2844 * Any protocol or device that holds a reference should register
2845 * for netdevice notification, and cleanup and put back the
2846 * reference if they receive an UNREGISTER event.
2847 * We can get stuck here if buggy protocols don't correctly
2850 static void netdev_wait_allrefs(struct net_device *dev)
2852 unsigned long rebroadcast_time, warning_time;
2854 rebroadcast_time = warning_time = jiffies;
2855 while (atomic_read(&dev->refcnt) != 0) {
2856 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2859 /* Rebroadcast unregister notification */
2860 notifier_call_chain(&netdev_chain,
2861 NETDEV_UNREGISTER, dev);
2863 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2865 /* We must not have linkwatch events
2866 * pending on unregister. If this
2867 * happens, we simply run the queue
2868 * unscheduled, resulting in a noop
2871 linkwatch_run_queue();
2876 rebroadcast_time = jiffies;
2881 if (time_after(jiffies, warning_time + 10 * HZ)) {
2882 printk(KERN_EMERG "unregister_netdevice: "
2883 "waiting for %s to become free. Usage "
2885 dev->name, atomic_read(&dev->refcnt));
2886 warning_time = jiffies;
2895 * register_netdevice(x1);
2896 * register_netdevice(x2);
2898 * unregister_netdevice(y1);
2899 * unregister_netdevice(y2);
2905 * We are invoked by rtnl_unlock() after it drops the semaphore.
2906 * This allows us to deal with problems:
2907 * 1) We can create/delete sysfs objects which invoke hotplug
2908 * without deadlocking with linkwatch via keventd.
2909 * 2) Since we run with the RTNL semaphore not held, we can sleep
2910 * safely in order to wait for the netdev refcnt to drop to zero.
2912 static DECLARE_MUTEX(net_todo_run_mutex);
2913 void netdev_run_todo(void)
2915 struct list_head list = LIST_HEAD_INIT(list);
2919 /* Need to guard against multiple cpu's getting out of order. */
2920 down(&net_todo_run_mutex);
2922 /* Not safe to do outside the semaphore. We must not return
2923 * until all unregister events invoked by the local processor
2924 * have been completed (either by this todo run, or one on
2927 if (list_empty(&net_todo_list))
2930 /* Snapshot list, allow later requests */
2931 spin_lock(&net_todo_list_lock);
2932 list_splice_init(&net_todo_list, &list);
2933 spin_unlock(&net_todo_list_lock);
2935 while (!list_empty(&list)) {
2936 struct net_device *dev
2937 = list_entry(list.next, struct net_device, todo_list);
2938 list_del(&dev->todo_list);
2940 switch(dev->reg_state) {
2941 case NETREG_REGISTERING:
2942 err = netdev_register_sysfs(dev);
2944 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2946 dev->reg_state = NETREG_REGISTERED;
2949 case NETREG_UNREGISTERING:
2950 netdev_unregister_sysfs(dev);
2951 dev->reg_state = NETREG_UNREGISTERED;
2953 netdev_wait_allrefs(dev);
2956 BUG_ON(atomic_read(&dev->refcnt));
2957 BUG_TRAP(!dev->ip_ptr);
2958 BUG_TRAP(!dev->ip6_ptr);
2959 BUG_TRAP(!dev->dn_ptr);
2962 /* It must be the very last action,
2963 * after this 'dev' may point to freed up memory.
2965 if (dev->destructor)
2966 dev->destructor(dev);
2970 printk(KERN_ERR "network todo '%s' but state %d\n",
2971 dev->name, dev->reg_state);
2977 up(&net_todo_run_mutex);
2981 * alloc_netdev - allocate network device
2982 * @sizeof_priv: size of private data to allocate space for
2983 * @name: device name format string
2984 * @setup: callback to initialize device
2986 * Allocates a struct net_device with private data area for driver use
2987 * and performs basic initialization.
2989 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
2990 void (*setup)(struct net_device *))
2993 struct net_device *dev;
2996 /* ensure 32-byte alignment of both the device and private area */
2997 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
2998 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3000 p = kmalloc(alloc_size, GFP_KERNEL);
3002 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3005 memset(p, 0, alloc_size);
3007 dev = (struct net_device *)
3008 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3009 dev->padded = (char *)dev - (char *)p;
3012 dev->priv = netdev_priv(dev);
3015 strcpy(dev->name, name);
3018 EXPORT_SYMBOL(alloc_netdev);
3021 * free_netdev - free network device
3024 * This function does the last stage of destroying an allocated device
3025 * interface. The reference to the device object is released.
3026 * If this is the last reference then it will be freed.
3028 void free_netdev(struct net_device *dev)
3031 /* Compatiablity with error handling in drivers */
3032 if (dev->reg_state == NETREG_UNINITIALIZED) {
3033 kfree((char *)dev - dev->padded);
3037 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3038 dev->reg_state = NETREG_RELEASED;
3040 /* will free via class release */
3041 class_device_put(&dev->class_dev);
3043 kfree((char *)dev - dev->padded);
3047 /* Synchronize with packet receive processing. */
3048 void synchronize_net(void)
3055 * unregister_netdevice - remove device from the kernel
3058 * This function shuts down a device interface and removes it
3059 * from the kernel tables. On success 0 is returned, on a failure
3060 * a negative errno code is returned.
3062 * Callers must hold the rtnl semaphore. You may want
3063 * unregister_netdev() instead of this.
3066 int unregister_netdevice(struct net_device *dev)
3068 struct net_device *d, **dp;
3070 BUG_ON(dev_boot_phase);
3073 /* Some devices call without registering for initialization unwind. */
3074 if (dev->reg_state == NETREG_UNINITIALIZED) {
3075 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3076 "was registered\n", dev->name, dev);
3080 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3082 /* If device is running, close it first. */
3083 if (dev->flags & IFF_UP)
3086 /* And unlink it from device chain. */
3087 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3089 write_lock_bh(&dev_base_lock);
3090 hlist_del(&dev->name_hlist);
3091 hlist_del(&dev->index_hlist);
3092 if (dev_tail == &dev->next)
3095 write_unlock_bh(&dev_base_lock);
3100 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3105 dev->reg_state = NETREG_UNREGISTERING;
3109 /* Shutdown queueing discipline. */
3113 /* Notify protocols, that we are about to destroy
3114 this device. They should clean all the things.
3116 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3119 * Flush the multicast chain
3121 dev_mc_discard(dev);
3126 /* Notifier chain MUST detach us from master device. */
3127 BUG_TRAP(!dev->master);
3129 free_divert_blk(dev);
3131 /* Finish processing unregister after unlock */
3141 * unregister_netdev - remove device from the kernel
3144 * This function shuts down a device interface and removes it
3145 * from the kernel tables. On success 0 is returned, on a failure
3146 * a negative errno code is returned.
3148 * This is just a wrapper for unregister_netdevice that takes
3149 * the rtnl semaphore. In general you want to use this and not
3150 * unregister_netdevice.
3152 void unregister_netdev(struct net_device *dev)
3155 unregister_netdevice(dev);
3159 EXPORT_SYMBOL(unregister_netdev);
3161 #ifdef CONFIG_HOTPLUG_CPU
3162 static int dev_cpu_callback(struct notifier_block *nfb,
3163 unsigned long action,
3166 struct sk_buff **list_skb;
3167 struct net_device **list_net;
3168 struct sk_buff *skb;
3169 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3170 struct softnet_data *sd, *oldsd;
3172 if (action != CPU_DEAD)
3175 local_irq_disable();
3176 cpu = smp_processor_id();
3177 sd = &per_cpu(softnet_data, cpu);
3178 oldsd = &per_cpu(softnet_data, oldcpu);
3180 /* Find end of our completion_queue. */
3181 list_skb = &sd->completion_queue;
3183 list_skb = &(*list_skb)->next;
3184 /* Append completion queue from offline CPU. */
3185 *list_skb = oldsd->completion_queue;
3186 oldsd->completion_queue = NULL;
3188 /* Find end of our output_queue. */
3189 list_net = &sd->output_queue;
3191 list_net = &(*list_net)->next_sched;
3192 /* Append output queue from offline CPU. */
3193 *list_net = oldsd->output_queue;
3194 oldsd->output_queue = NULL;
3196 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3199 /* Process offline CPU's input_pkt_queue */
3200 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3205 #endif /* CONFIG_HOTPLUG_CPU */
3209 * Initialize the DEV module. At boot time this walks the device list and
3210 * unhooks any devices that fail to initialise (normally hardware not
3211 * present) and leaves us with a valid list of present and active devices.
3216 * This is called single threaded during boot, so no need
3217 * to take the rtnl semaphore.
3219 static int __init net_dev_init(void)
3221 int i, rc = -ENOMEM;
3223 BUG_ON(!dev_boot_phase);
3227 if (dev_proc_init())
3230 if (netdev_sysfs_init())
3233 INIT_LIST_HEAD(&ptype_all);
3234 for (i = 0; i < 16; i++)
3235 INIT_LIST_HEAD(&ptype_base[i]);
3237 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3238 INIT_HLIST_HEAD(&dev_name_head[i]);
3240 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3241 INIT_HLIST_HEAD(&dev_index_head[i]);
3244 * Initialise the packet receive queues.
3248 struct softnet_data *queue;
3250 queue = &per_cpu(softnet_data, i);
3251 skb_queue_head_init(&queue->input_pkt_queue);
3252 queue->completion_queue = NULL;
3253 INIT_LIST_HEAD(&queue->poll_list);
3254 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3255 queue->backlog_dev.weight = weight_p;
3256 queue->backlog_dev.poll = process_backlog;
3257 atomic_set(&queue->backlog_dev.refcnt, 1);
3262 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3263 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3265 hotcpu_notifier(dev_cpu_callback, 0);
3273 subsys_initcall(net_dev_init);
3275 EXPORT_SYMBOL(__dev_get_by_index);
3276 EXPORT_SYMBOL(__dev_get_by_name);
3277 EXPORT_SYMBOL(__dev_remove_pack);
3278 EXPORT_SYMBOL(__skb_linearize);
3279 EXPORT_SYMBOL(dev_valid_name);
3280 EXPORT_SYMBOL(dev_add_pack);
3281 EXPORT_SYMBOL(dev_alloc_name);
3282 EXPORT_SYMBOL(dev_close);
3283 EXPORT_SYMBOL(dev_get_by_flags);
3284 EXPORT_SYMBOL(dev_get_by_index);
3285 EXPORT_SYMBOL(dev_get_by_name);
3286 EXPORT_SYMBOL(dev_open);
3287 EXPORT_SYMBOL(dev_queue_xmit);
3288 EXPORT_SYMBOL(dev_remove_pack);
3289 EXPORT_SYMBOL(dev_set_allmulti);
3290 EXPORT_SYMBOL(dev_set_promiscuity);
3291 EXPORT_SYMBOL(dev_change_flags);
3292 EXPORT_SYMBOL(dev_set_mtu);
3293 EXPORT_SYMBOL(dev_set_mac_address);
3294 EXPORT_SYMBOL(free_netdev);
3295 EXPORT_SYMBOL(netdev_boot_setup_check);
3296 EXPORT_SYMBOL(netdev_set_master);
3297 EXPORT_SYMBOL(netdev_state_change);
3298 EXPORT_SYMBOL(netif_receive_skb);
3299 EXPORT_SYMBOL(netif_rx);
3300 EXPORT_SYMBOL(register_gifconf);
3301 EXPORT_SYMBOL(register_netdevice);
3302 EXPORT_SYMBOL(register_netdevice_notifier);
3303 EXPORT_SYMBOL(skb_checksum_help);
3304 EXPORT_SYMBOL(synchronize_net);
3305 EXPORT_SYMBOL(unregister_netdevice);
3306 EXPORT_SYMBOL(unregister_netdevice_notifier);
3307 EXPORT_SYMBOL(net_enable_timestamp);
3308 EXPORT_SYMBOL(net_disable_timestamp);
3309 EXPORT_SYMBOL(dev_get_flags);
3311 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3312 EXPORT_SYMBOL(br_handle_frame_hook);
3313 EXPORT_SYMBOL(br_fdb_get_hook);
3314 EXPORT_SYMBOL(br_fdb_put_hook);
3318 EXPORT_SYMBOL(dev_load);
3321 EXPORT_PER_CPU_SYMBOL(softnet_data);