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);
271 * __dev_remove_pack - remove packet handler
272 * @pt: packet type declaration
274 * Remove a protocol handler that was previously added to the kernel
275 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
276 * from the kernel lists and can be freed or reused once this function
279 * The packet type might still be in use by receivers
280 * and must not be freed until after all the CPU's have gone
281 * through a quiescent state.
283 void __dev_remove_pack(struct packet_type *pt)
285 struct list_head *head;
286 struct packet_type *pt1;
288 spin_lock_bh(&ptype_lock);
290 if (pt->type == htons(ETH_P_ALL)) {
294 head = &ptype_base[ntohs(pt->type) & 15];
296 list_for_each_entry(pt1, head, list) {
298 list_del_rcu(&pt->list);
303 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
305 spin_unlock_bh(&ptype_lock);
308 * dev_remove_pack - remove packet handler
309 * @pt: packet type declaration
311 * Remove a protocol handler that was previously added to the kernel
312 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
313 * from the kernel lists and can be freed or reused once this function
316 * This call sleeps to guarantee that no CPU is looking at the packet
319 void dev_remove_pack(struct packet_type *pt)
321 __dev_remove_pack(pt);
326 /******************************************************************************
328 Device Boot-time Settings Routines
330 *******************************************************************************/
332 /* Boot time configuration table */
333 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
336 * netdev_boot_setup_add - add new setup entry
337 * @name: name of the device
338 * @map: configured settings for the device
340 * Adds new setup entry to the dev_boot_setup list. The function
341 * returns 0 on error and 1 on success. This is a generic routine to
344 static int netdev_boot_setup_add(char *name, struct ifmap *map)
346 struct netdev_boot_setup *s;
350 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
351 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
352 memset(s[i].name, 0, sizeof(s[i].name));
353 strcpy(s[i].name, name);
354 memcpy(&s[i].map, map, sizeof(s[i].map));
359 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
363 * netdev_boot_setup_check - check boot time settings
364 * @dev: the netdevice
366 * Check boot time settings for the device.
367 * The found settings are set for the device to be used
368 * later in the device probing.
369 * Returns 0 if no settings found, 1 if they are.
371 int netdev_boot_setup_check(struct net_device *dev)
373 struct netdev_boot_setup *s = dev_boot_setup;
376 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
377 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
378 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
379 dev->irq = s[i].map.irq;
380 dev->base_addr = s[i].map.base_addr;
381 dev->mem_start = s[i].map.mem_start;
382 dev->mem_end = s[i].map.mem_end;
391 * netdev_boot_base - get address from boot time settings
392 * @prefix: prefix for network device
393 * @unit: id for network device
395 * Check boot time settings for the base address of device.
396 * The found settings are set for the device to be used
397 * later in the device probing.
398 * Returns 0 if no settings found.
400 unsigned long netdev_boot_base(const char *prefix, int unit)
402 const struct netdev_boot_setup *s = dev_boot_setup;
406 sprintf(name, "%s%d", prefix, unit);
409 * If device already registered then return base of 1
410 * to indicate not to probe for this interface
412 if (__dev_get_by_name(name))
415 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
416 if (!strcmp(name, s[i].name))
417 return s[i].map.base_addr;
422 * Saves at boot time configured settings for any netdevice.
424 int __init netdev_boot_setup(char *str)
429 str = get_options(str, ARRAY_SIZE(ints), ints);
434 memset(&map, 0, sizeof(map));
438 map.base_addr = ints[2];
440 map.mem_start = ints[3];
442 map.mem_end = ints[4];
444 /* Add new entry to the list */
445 return netdev_boot_setup_add(str, &map);
448 __setup("netdev=", netdev_boot_setup);
450 /*******************************************************************************
452 Device Interface Subroutines
454 *******************************************************************************/
457 * __dev_get_by_name - find a device by its name
458 * @name: name to find
460 * Find an interface by name. Must be called under RTNL semaphore
461 * or @dev_base_lock. If the name is found a pointer to the device
462 * is returned. If the name is not found then %NULL is returned. The
463 * reference counters are not incremented so the caller must be
464 * careful with locks.
467 struct net_device *__dev_get_by_name(const char *name)
469 struct hlist_node *p;
471 hlist_for_each(p, dev_name_hash(name)) {
472 struct net_device *dev
473 = hlist_entry(p, struct net_device, name_hlist);
474 if (!strncmp(dev->name, name, IFNAMSIZ))
481 * dev_get_by_name - find a device by its name
482 * @name: name to find
484 * Find an interface by name. This can be called from any
485 * context and does its own locking. The returned handle has
486 * the usage count incremented and the caller must use dev_put() to
487 * release it when it is no longer needed. %NULL is returned if no
488 * matching device is found.
491 struct net_device *dev_get_by_name(const char *name)
493 struct net_device *dev;
495 read_lock(&dev_base_lock);
496 dev = __dev_get_by_name(name);
499 read_unlock(&dev_base_lock);
504 * __dev_get_by_index - find a device by its ifindex
505 * @ifindex: index of device
507 * Search for an interface by index. Returns %NULL if the device
508 * is not found or a pointer to the device. The device has not
509 * had its reference counter increased so the caller must be careful
510 * about locking. The caller must hold either the RTNL semaphore
514 struct net_device *__dev_get_by_index(int ifindex)
516 struct hlist_node *p;
518 hlist_for_each(p, dev_index_hash(ifindex)) {
519 struct net_device *dev
520 = hlist_entry(p, struct net_device, index_hlist);
521 if (dev->ifindex == ifindex)
529 * dev_get_by_index - find a device by its ifindex
530 * @ifindex: index of device
532 * Search for an interface by index. Returns NULL if the device
533 * is not found or a pointer to the device. The device returned has
534 * had a reference added and the pointer is safe until the user calls
535 * dev_put to indicate they have finished with it.
538 struct net_device *dev_get_by_index(int ifindex)
540 struct net_device *dev;
542 read_lock(&dev_base_lock);
543 dev = __dev_get_by_index(ifindex);
546 read_unlock(&dev_base_lock);
551 * dev_getbyhwaddr - find a device by its hardware address
552 * @type: media type of device
553 * @ha: hardware address
555 * Search for an interface by MAC address. Returns NULL if the device
556 * is not found or a pointer to the device. The caller must hold the
557 * rtnl semaphore. The returned device has not had its ref count increased
558 * and the caller must therefore be careful about locking
561 * If the API was consistent this would be __dev_get_by_hwaddr
564 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
566 struct net_device *dev;
570 for (dev = dev_base; dev; dev = dev->next)
571 if (dev->type == type &&
572 !memcmp(dev->dev_addr, ha, dev->addr_len))
577 EXPORT_SYMBOL(dev_getbyhwaddr);
579 struct net_device *dev_getfirstbyhwtype(unsigned short type)
581 struct net_device *dev;
584 for (dev = dev_base; dev; dev = dev->next) {
585 if (dev->type == type) {
594 EXPORT_SYMBOL(dev_getfirstbyhwtype);
597 * dev_get_by_flags - find any device with given flags
598 * @if_flags: IFF_* values
599 * @mask: bitmask of bits in if_flags to check
601 * Search for any interface with the given flags. Returns NULL if a device
602 * is not found or a pointer to the device. The device returned has
603 * had a reference added and the pointer is safe until the user calls
604 * dev_put to indicate they have finished with it.
607 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
609 struct net_device *dev;
611 read_lock(&dev_base_lock);
612 for (dev = dev_base; dev != NULL; dev = dev->next) {
613 if (((dev->flags ^ if_flags) & mask) == 0) {
618 read_unlock(&dev_base_lock);
623 * dev_valid_name - check if name is okay for network device
626 * Network device names need to be valid file names to
627 * to allow sysfs to work
629 static int dev_valid_name(const char *name)
631 return !(*name == '\0'
632 || !strcmp(name, ".")
633 || !strcmp(name, "..")
634 || strchr(name, '/'));
638 * dev_alloc_name - allocate a name for a device
640 * @name: name format string
642 * Passed a format string - eg "lt%d" it will try and find a suitable
643 * id. Not efficient for many devices, not called a lot. The caller
644 * must hold the dev_base or rtnl lock while allocating the name and
645 * adding the device in order to avoid duplicates. Returns the number
646 * of the unit assigned or a negative errno code.
649 int dev_alloc_name(struct net_device *dev, const char *name)
654 const int max_netdevices = 8*PAGE_SIZE;
656 struct net_device *d;
658 p = strnchr(name, IFNAMSIZ-1, '%');
661 * Verify the string as this thing may have come from
662 * the user. There must be either one "%d" and no other "%"
665 if (p[1] != 'd' || strchr(p + 2, '%'))
668 /* Use one page as a bit array of possible slots */
669 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
673 for (d = dev_base; d; d = d->next) {
674 if (!sscanf(d->name, name, &i))
676 if (i < 0 || i >= max_netdevices)
679 /* avoid cases where sscanf is not exact inverse of printf */
680 snprintf(buf, sizeof(buf), name, i);
681 if (!strncmp(buf, d->name, IFNAMSIZ))
685 i = find_first_zero_bit(inuse, max_netdevices);
686 free_page((unsigned long) inuse);
689 snprintf(buf, sizeof(buf), name, i);
690 if (!__dev_get_by_name(buf)) {
691 strlcpy(dev->name, buf, IFNAMSIZ);
695 /* It is possible to run out of possible slots
696 * when the name is long and there isn't enough space left
697 * for the digits, or if all bits are used.
704 * dev_change_name - change name of a device
706 * @newname: name (or format string) must be at least IFNAMSIZ
708 * Change name of a device, can pass format strings "eth%d".
711 int dev_change_name(struct net_device *dev, char *newname)
717 if (dev->flags & IFF_UP)
720 if (!dev_valid_name(newname))
723 if (strchr(newname, '%')) {
724 err = dev_alloc_name(dev, newname);
727 strcpy(newname, dev->name);
729 else if (__dev_get_by_name(newname))
732 strlcpy(dev->name, newname, IFNAMSIZ);
734 err = class_device_rename(&dev->class_dev, dev->name);
736 hlist_del(&dev->name_hlist);
737 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
738 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
745 * netdev_features_change - device changes fatures
746 * @dev: device to cause notification
748 * Called to indicate a device has changed features.
750 void netdev_features_change(struct net_device *dev)
752 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
754 EXPORT_SYMBOL(netdev_features_change);
757 * netdev_state_change - device changes state
758 * @dev: device to cause notification
760 * Called to indicate a device has changed state. This function calls
761 * the notifier chains for netdev_chain and sends a NEWLINK message
762 * to the routing socket.
764 void netdev_state_change(struct net_device *dev)
766 if (dev->flags & IFF_UP) {
767 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
768 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
773 * dev_load - load a network module
774 * @name: name of interface
776 * If a network interface is not present and the process has suitable
777 * privileges this function loads the module. If module loading is not
778 * available in this kernel then it becomes a nop.
781 void dev_load(const char *name)
783 struct net_device *dev;
785 read_lock(&dev_base_lock);
786 dev = __dev_get_by_name(name);
787 read_unlock(&dev_base_lock);
789 if (!dev && capable(CAP_SYS_MODULE))
790 request_module("%s", name);
793 static int default_rebuild_header(struct sk_buff *skb)
795 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
796 skb->dev ? skb->dev->name : "NULL!!!");
803 * dev_open - prepare an interface for use.
804 * @dev: device to open
806 * Takes a device from down to up state. The device's private open
807 * function is invoked and then the multicast lists are loaded. Finally
808 * the device is moved into the up state and a %NETDEV_UP message is
809 * sent to the netdev notifier chain.
811 * Calling this function on an active interface is a nop. On a failure
812 * a negative errno code is returned.
814 int dev_open(struct net_device *dev)
822 if (dev->flags & IFF_UP)
826 * Is it even present?
828 if (!netif_device_present(dev))
832 * Call device private open method
834 set_bit(__LINK_STATE_START, &dev->state);
836 ret = dev->open(dev);
838 clear_bit(__LINK_STATE_START, &dev->state);
842 * If it went open OK then:
849 dev->flags |= IFF_UP;
852 * Initialize multicasting status
857 * Wakeup transmit queue engine
862 * ... and announce new interface.
864 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
870 * dev_close - shutdown an interface.
871 * @dev: device to shutdown
873 * This function moves an active device into down state. A
874 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
875 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
878 int dev_close(struct net_device *dev)
880 if (!(dev->flags & IFF_UP))
884 * Tell people we are going down, so that they can
885 * prepare to death, when device is still operating.
887 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
891 clear_bit(__LINK_STATE_START, &dev->state);
893 /* Synchronize to scheduled poll. We cannot touch poll list,
894 * it can be even on different cpu. So just clear netif_running(),
895 * and wait when poll really will happen. Actually, the best place
896 * for this is inside dev->stop() after device stopped its irq
897 * engine, but this requires more changes in devices. */
899 smp_mb__after_clear_bit(); /* Commit netif_running(). */
900 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
906 * Call the device specific close. This cannot fail.
907 * Only if device is UP
909 * We allow it to be called even after a DETACH hot-plug
916 * Device is now down.
919 dev->flags &= ~IFF_UP;
922 * Tell people we are down
924 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
931 * Device change register/unregister. These are not inline or static
932 * as we export them to the world.
936 * register_netdevice_notifier - register a network notifier block
939 * Register a notifier to be called when network device events occur.
940 * The notifier passed is linked into the kernel structures and must
941 * not be reused until it has been unregistered. A negative errno code
942 * is returned on a failure.
944 * When registered all registration and up events are replayed
945 * to the new notifier to allow device to have a race free
946 * view of the network device list.
949 int register_netdevice_notifier(struct notifier_block *nb)
951 struct net_device *dev;
955 err = notifier_chain_register(&netdev_chain, nb);
957 for (dev = dev_base; dev; dev = dev->next) {
958 nb->notifier_call(nb, NETDEV_REGISTER, dev);
960 if (dev->flags & IFF_UP)
961 nb->notifier_call(nb, NETDEV_UP, dev);
969 * unregister_netdevice_notifier - unregister a network notifier block
972 * Unregister a notifier previously registered by
973 * register_netdevice_notifier(). The notifier is unlinked into the
974 * kernel structures and may then be reused. A negative errno code
975 * is returned on a failure.
978 int unregister_netdevice_notifier(struct notifier_block *nb)
980 return notifier_chain_unregister(&netdev_chain, nb);
984 * call_netdevice_notifiers - call all network notifier blocks
985 * @val: value passed unmodified to notifier function
986 * @v: pointer passed unmodified to notifier function
988 * Call all network notifier blocks. Parameters and return value
989 * are as for notifier_call_chain().
992 int call_netdevice_notifiers(unsigned long val, void *v)
994 return notifier_call_chain(&netdev_chain, val, v);
997 /* When > 0 there are consumers of rx skb time stamps */
998 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1000 void net_enable_timestamp(void)
1002 atomic_inc(&netstamp_needed);
1005 void net_disable_timestamp(void)
1007 atomic_dec(&netstamp_needed);
1010 void __net_timestamp(struct sk_buff *skb)
1014 do_gettimeofday(&tv);
1015 skb_set_timestamp(skb, &tv);
1017 EXPORT_SYMBOL(__net_timestamp);
1019 static inline void net_timestamp(struct sk_buff *skb)
1021 if (atomic_read(&netstamp_needed))
1022 __net_timestamp(skb);
1024 skb->tstamp.off_sec = 0;
1025 skb->tstamp.off_usec = 0;
1030 * Support routine. Sends outgoing frames to any network
1031 * taps currently in use.
1034 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1036 struct packet_type *ptype;
1041 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1042 /* Never send packets back to the socket
1043 * they originated from - MvS (miquels@drinkel.ow.org)
1045 if ((ptype->dev == dev || !ptype->dev) &&
1046 (ptype->af_packet_priv == NULL ||
1047 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1048 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1052 /* skb->nh should be correctly
1053 set by sender, so that the second statement is
1054 just protection against buggy protocols.
1056 skb2->mac.raw = skb2->data;
1058 if (skb2->nh.raw < skb2->data ||
1059 skb2->nh.raw > skb2->tail) {
1060 if (net_ratelimit())
1061 printk(KERN_CRIT "protocol %04x is "
1063 skb2->protocol, dev->name);
1064 skb2->nh.raw = skb2->data;
1067 skb2->h.raw = skb2->nh.raw;
1068 skb2->pkt_type = PACKET_OUTGOING;
1069 ptype->func(skb2, skb->dev, ptype, skb->dev);
1076 * Invalidate hardware checksum when packet is to be mangled, and
1077 * complete checksum manually on outgoing path.
1079 int skb_checksum_help(struct sk_buff *skb, int inward)
1082 int ret = 0, offset = skb->h.raw - skb->data;
1085 skb->ip_summed = CHECKSUM_NONE;
1089 if (skb_cloned(skb)) {
1090 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1095 if (offset > (int)skb->len)
1097 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1099 offset = skb->tail - skb->h.raw;
1102 if (skb->csum + 2 > offset)
1105 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1106 skb->ip_summed = CHECKSUM_NONE;
1111 /* Take action when hardware reception checksum errors are detected. */
1113 void netdev_rx_csum_fault(struct net_device *dev)
1115 if (net_ratelimit()) {
1116 printk(KERN_ERR "%s: hw csum failure.\n", dev->name);
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" :
2128 * dev_set_allmulti - update allmulti count on a device
2132 * Add or remove reception of all multicast frames to a device. While the
2133 * count in the device remains above zero the interface remains listening
2134 * to all interfaces. Once it hits zero the device reverts back to normal
2135 * filtering operation. A negative @inc value is used to drop the counter
2136 * when releasing a resource needing all multicasts.
2139 void dev_set_allmulti(struct net_device *dev, int inc)
2141 unsigned short old_flags = dev->flags;
2143 dev->flags |= IFF_ALLMULTI;
2144 if ((dev->allmulti += inc) == 0)
2145 dev->flags &= ~IFF_ALLMULTI;
2146 if (dev->flags ^ old_flags)
2150 unsigned dev_get_flags(const struct net_device *dev)
2154 flags = (dev->flags & ~(IFF_PROMISC |
2157 (dev->gflags & (IFF_PROMISC |
2160 if (netif_running(dev) && netif_carrier_ok(dev))
2161 flags |= IFF_RUNNING;
2166 int dev_change_flags(struct net_device *dev, unsigned flags)
2169 int old_flags = dev->flags;
2172 * Set the flags on our device.
2175 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2176 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2178 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2182 * Load in the correct multicast list now the flags have changed.
2188 * Have we downed the interface. We handle IFF_UP ourselves
2189 * according to user attempts to set it, rather than blindly
2194 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2195 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2201 if (dev->flags & IFF_UP &&
2202 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2204 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2206 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2207 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2208 dev->gflags ^= IFF_PROMISC;
2209 dev_set_promiscuity(dev, inc);
2212 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2213 is important. Some (broken) drivers set IFF_PROMISC, when
2214 IFF_ALLMULTI is requested not asking us and not reporting.
2216 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2217 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2218 dev->gflags ^= IFF_ALLMULTI;
2219 dev_set_allmulti(dev, inc);
2222 if (old_flags ^ dev->flags)
2223 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2228 int dev_set_mtu(struct net_device *dev, int new_mtu)
2232 if (new_mtu == dev->mtu)
2235 /* MTU must be positive. */
2239 if (!netif_device_present(dev))
2243 if (dev->change_mtu)
2244 err = dev->change_mtu(dev, new_mtu);
2247 if (!err && dev->flags & IFF_UP)
2248 notifier_call_chain(&netdev_chain,
2249 NETDEV_CHANGEMTU, dev);
2253 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2257 if (!dev->set_mac_address)
2259 if (sa->sa_family != dev->type)
2261 if (!netif_device_present(dev))
2263 err = dev->set_mac_address(dev, sa);
2265 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2270 * Perform the SIOCxIFxxx calls.
2272 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2275 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2281 case SIOCGIFFLAGS: /* Get interface flags */
2282 ifr->ifr_flags = dev_get_flags(dev);
2285 case SIOCSIFFLAGS: /* Set interface flags */
2286 return dev_change_flags(dev, ifr->ifr_flags);
2288 case SIOCGIFMETRIC: /* Get the metric on the interface
2289 (currently unused) */
2290 ifr->ifr_metric = 0;
2293 case SIOCSIFMETRIC: /* Set the metric on the interface
2294 (currently unused) */
2297 case SIOCGIFMTU: /* Get the MTU of a device */
2298 ifr->ifr_mtu = dev->mtu;
2301 case SIOCSIFMTU: /* Set the MTU of a device */
2302 return dev_set_mtu(dev, ifr->ifr_mtu);
2306 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2308 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2309 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2310 ifr->ifr_hwaddr.sa_family = dev->type;
2314 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2316 case SIOCSIFHWBROADCAST:
2317 if (ifr->ifr_hwaddr.sa_family != dev->type)
2319 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2320 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2321 notifier_call_chain(&netdev_chain,
2322 NETDEV_CHANGEADDR, dev);
2326 ifr->ifr_map.mem_start = dev->mem_start;
2327 ifr->ifr_map.mem_end = dev->mem_end;
2328 ifr->ifr_map.base_addr = dev->base_addr;
2329 ifr->ifr_map.irq = dev->irq;
2330 ifr->ifr_map.dma = dev->dma;
2331 ifr->ifr_map.port = dev->if_port;
2335 if (dev->set_config) {
2336 if (!netif_device_present(dev))
2338 return dev->set_config(dev, &ifr->ifr_map);
2343 if (!dev->set_multicast_list ||
2344 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2346 if (!netif_device_present(dev))
2348 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2352 if (!dev->set_multicast_list ||
2353 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2355 if (!netif_device_present(dev))
2357 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2361 ifr->ifr_ifindex = dev->ifindex;
2365 ifr->ifr_qlen = dev->tx_queue_len;
2369 if (ifr->ifr_qlen < 0)
2371 dev->tx_queue_len = ifr->ifr_qlen;
2375 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2376 return dev_change_name(dev, ifr->ifr_newname);
2379 * Unknown or private ioctl
2383 if ((cmd >= SIOCDEVPRIVATE &&
2384 cmd <= SIOCDEVPRIVATE + 15) ||
2385 cmd == SIOCBONDENSLAVE ||
2386 cmd == SIOCBONDRELEASE ||
2387 cmd == SIOCBONDSETHWADDR ||
2388 cmd == SIOCBONDSLAVEINFOQUERY ||
2389 cmd == SIOCBONDINFOQUERY ||
2390 cmd == SIOCBONDCHANGEACTIVE ||
2391 cmd == SIOCGMIIPHY ||
2392 cmd == SIOCGMIIREG ||
2393 cmd == SIOCSMIIREG ||
2394 cmd == SIOCBRADDIF ||
2395 cmd == SIOCBRDELIF ||
2396 cmd == SIOCWANDEV) {
2398 if (dev->do_ioctl) {
2399 if (netif_device_present(dev))
2400 err = dev->do_ioctl(dev, ifr,
2413 * This function handles all "interface"-type I/O control requests. The actual
2414 * 'doing' part of this is dev_ifsioc above.
2418 * dev_ioctl - network device ioctl
2419 * @cmd: command to issue
2420 * @arg: pointer to a struct ifreq in user space
2422 * Issue ioctl functions to devices. This is normally called by the
2423 * user space syscall interfaces but can sometimes be useful for
2424 * other purposes. The return value is the return from the syscall if
2425 * positive or a negative errno code on error.
2428 int dev_ioctl(unsigned int cmd, void __user *arg)
2434 /* One special case: SIOCGIFCONF takes ifconf argument
2435 and requires shared lock, because it sleeps writing
2439 if (cmd == SIOCGIFCONF) {
2441 ret = dev_ifconf((char __user *) arg);
2445 if (cmd == SIOCGIFNAME)
2446 return dev_ifname((struct ifreq __user *)arg);
2448 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2451 ifr.ifr_name[IFNAMSIZ-1] = 0;
2453 colon = strchr(ifr.ifr_name, ':');
2458 * See which interface the caller is talking about.
2463 * These ioctl calls:
2464 * - can be done by all.
2465 * - atomic and do not require locking.
2476 dev_load(ifr.ifr_name);
2477 read_lock(&dev_base_lock);
2478 ret = dev_ifsioc(&ifr, cmd);
2479 read_unlock(&dev_base_lock);
2483 if (copy_to_user(arg, &ifr,
2484 sizeof(struct ifreq)))
2490 dev_load(ifr.ifr_name);
2492 ret = dev_ethtool(&ifr);
2497 if (copy_to_user(arg, &ifr,
2498 sizeof(struct ifreq)))
2504 * These ioctl calls:
2505 * - require superuser power.
2506 * - require strict serialization.
2512 if (!capable(CAP_NET_ADMIN))
2514 dev_load(ifr.ifr_name);
2516 ret = dev_ifsioc(&ifr, cmd);
2521 if (copy_to_user(arg, &ifr,
2522 sizeof(struct ifreq)))
2528 * These ioctl calls:
2529 * - require superuser power.
2530 * - require strict serialization.
2531 * - do not return a value
2541 case SIOCSIFHWBROADCAST:
2544 case SIOCBONDENSLAVE:
2545 case SIOCBONDRELEASE:
2546 case SIOCBONDSETHWADDR:
2547 case SIOCBONDSLAVEINFOQUERY:
2548 case SIOCBONDINFOQUERY:
2549 case SIOCBONDCHANGEACTIVE:
2552 if (!capable(CAP_NET_ADMIN))
2554 dev_load(ifr.ifr_name);
2556 ret = dev_ifsioc(&ifr, cmd);
2561 /* Get the per device memory space. We can add this but
2562 * currently do not support it */
2564 /* Set the per device memory buffer space.
2565 * Not applicable in our case */
2570 * Unknown or private ioctl.
2573 if (cmd == SIOCWANDEV ||
2574 (cmd >= SIOCDEVPRIVATE &&
2575 cmd <= SIOCDEVPRIVATE + 15)) {
2576 dev_load(ifr.ifr_name);
2578 ret = dev_ifsioc(&ifr, cmd);
2580 if (!ret && copy_to_user(arg, &ifr,
2581 sizeof(struct ifreq)))
2586 /* Take care of Wireless Extensions */
2587 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2588 /* If command is `set a parameter', or
2589 * `get the encoding parameters', check if
2590 * the user has the right to do it */
2591 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2592 if (!capable(CAP_NET_ADMIN))
2595 dev_load(ifr.ifr_name);
2597 /* Follow me in net/core/wireless.c */
2598 ret = wireless_process_ioctl(&ifr, cmd);
2600 if (IW_IS_GET(cmd) &&
2601 copy_to_user(arg, &ifr,
2602 sizeof(struct ifreq)))
2606 #endif /* WIRELESS_EXT */
2613 * dev_new_index - allocate an ifindex
2615 * Returns a suitable unique value for a new device interface
2616 * number. The caller must hold the rtnl semaphore or the
2617 * dev_base_lock to be sure it remains unique.
2619 static int dev_new_index(void)
2625 if (!__dev_get_by_index(ifindex))
2630 static int dev_boot_phase = 1;
2632 /* Delayed registration/unregisteration */
2633 static DEFINE_SPINLOCK(net_todo_list_lock);
2634 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2636 static inline void net_set_todo(struct net_device *dev)
2638 spin_lock(&net_todo_list_lock);
2639 list_add_tail(&dev->todo_list, &net_todo_list);
2640 spin_unlock(&net_todo_list_lock);
2644 * register_netdevice - register a network device
2645 * @dev: device to register
2647 * Take a completed network device structure and add it to the kernel
2648 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2649 * chain. 0 is returned on success. A negative errno code is returned
2650 * on a failure to set up the device, or if the name is a duplicate.
2652 * Callers must hold the rtnl semaphore. You may want
2653 * register_netdev() instead of this.
2656 * The locking appears insufficient to guarantee two parallel registers
2657 * will not get the same name.
2660 int register_netdevice(struct net_device *dev)
2662 struct hlist_head *head;
2663 struct hlist_node *p;
2666 BUG_ON(dev_boot_phase);
2669 /* When net_device's are persistent, this will be fatal. */
2670 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2672 spin_lock_init(&dev->queue_lock);
2673 spin_lock_init(&dev->xmit_lock);
2674 dev->xmit_lock_owner = -1;
2675 #ifdef CONFIG_NET_CLS_ACT
2676 spin_lock_init(&dev->ingress_lock);
2679 ret = alloc_divert_blk(dev);
2685 /* Init, if this function is available */
2687 ret = dev->init(dev);
2695 if (!dev_valid_name(dev->name)) {
2700 dev->ifindex = dev_new_index();
2701 if (dev->iflink == -1)
2702 dev->iflink = dev->ifindex;
2704 /* Check for existence of name */
2705 head = dev_name_hash(dev->name);
2706 hlist_for_each(p, head) {
2707 struct net_device *d
2708 = hlist_entry(p, struct net_device, name_hlist);
2709 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2715 /* Fix illegal SG+CSUM combinations. */
2716 if ((dev->features & NETIF_F_SG) &&
2717 !(dev->features & (NETIF_F_IP_CSUM |
2719 NETIF_F_HW_CSUM))) {
2720 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2722 dev->features &= ~NETIF_F_SG;
2725 /* TSO requires that SG is present as well. */
2726 if ((dev->features & NETIF_F_TSO) &&
2727 !(dev->features & NETIF_F_SG)) {
2728 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2730 dev->features &= ~NETIF_F_TSO;
2732 if (dev->features & NETIF_F_UFO) {
2733 if (!(dev->features & NETIF_F_HW_CSUM)) {
2734 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2735 "NETIF_F_HW_CSUM feature.\n",
2737 dev->features &= ~NETIF_F_UFO;
2739 if (!(dev->features & NETIF_F_SG)) {
2740 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2741 "NETIF_F_SG feature.\n",
2743 dev->features &= ~NETIF_F_UFO;
2748 * nil rebuild_header routine,
2749 * that should be never called and used as just bug trap.
2752 if (!dev->rebuild_header)
2753 dev->rebuild_header = default_rebuild_header;
2756 * Default initial state at registry is that the
2757 * device is present.
2760 set_bit(__LINK_STATE_PRESENT, &dev->state);
2763 dev_init_scheduler(dev);
2764 write_lock_bh(&dev_base_lock);
2766 dev_tail = &dev->next;
2767 hlist_add_head(&dev->name_hlist, head);
2768 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2770 dev->reg_state = NETREG_REGISTERING;
2771 write_unlock_bh(&dev_base_lock);
2773 /* Notify protocols, that a new device appeared. */
2774 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2776 /* Finish registration after unlock */
2783 free_divert_blk(dev);
2788 * register_netdev - register a network device
2789 * @dev: device to register
2791 * Take a completed network device structure and add it to the kernel
2792 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2793 * chain. 0 is returned on success. A negative errno code is returned
2794 * on a failure to set up the device, or if the name is a duplicate.
2796 * This is a wrapper around register_netdev that takes the rtnl semaphore
2797 * and expands the device name if you passed a format string to
2800 int register_netdev(struct net_device *dev)
2807 * If the name is a format string the caller wants us to do a
2810 if (strchr(dev->name, '%')) {
2811 err = dev_alloc_name(dev, dev->name);
2817 * Back compatibility hook. Kill this one in 2.5
2819 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2820 err = dev_alloc_name(dev, "eth%d");
2825 err = register_netdevice(dev);
2830 EXPORT_SYMBOL(register_netdev);
2833 * netdev_wait_allrefs - wait until all references are gone.
2835 * This is called when unregistering network devices.
2837 * Any protocol or device that holds a reference should register
2838 * for netdevice notification, and cleanup and put back the
2839 * reference if they receive an UNREGISTER event.
2840 * We can get stuck here if buggy protocols don't correctly
2843 static void netdev_wait_allrefs(struct net_device *dev)
2845 unsigned long rebroadcast_time, warning_time;
2847 rebroadcast_time = warning_time = jiffies;
2848 while (atomic_read(&dev->refcnt) != 0) {
2849 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2852 /* Rebroadcast unregister notification */
2853 notifier_call_chain(&netdev_chain,
2854 NETDEV_UNREGISTER, dev);
2856 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2858 /* We must not have linkwatch events
2859 * pending on unregister. If this
2860 * happens, we simply run the queue
2861 * unscheduled, resulting in a noop
2864 linkwatch_run_queue();
2869 rebroadcast_time = jiffies;
2874 if (time_after(jiffies, warning_time + 10 * HZ)) {
2875 printk(KERN_EMERG "unregister_netdevice: "
2876 "waiting for %s to become free. Usage "
2878 dev->name, atomic_read(&dev->refcnt));
2879 warning_time = jiffies;
2888 * register_netdevice(x1);
2889 * register_netdevice(x2);
2891 * unregister_netdevice(y1);
2892 * unregister_netdevice(y2);
2898 * We are invoked by rtnl_unlock() after it drops the semaphore.
2899 * This allows us to deal with problems:
2900 * 1) We can create/delete sysfs objects which invoke hotplug
2901 * without deadlocking with linkwatch via keventd.
2902 * 2) Since we run with the RTNL semaphore not held, we can sleep
2903 * safely in order to wait for the netdev refcnt to drop to zero.
2905 static DECLARE_MUTEX(net_todo_run_mutex);
2906 void netdev_run_todo(void)
2908 struct list_head list = LIST_HEAD_INIT(list);
2912 /* Need to guard against multiple cpu's getting out of order. */
2913 down(&net_todo_run_mutex);
2915 /* Not safe to do outside the semaphore. We must not return
2916 * until all unregister events invoked by the local processor
2917 * have been completed (either by this todo run, or one on
2920 if (list_empty(&net_todo_list))
2923 /* Snapshot list, allow later requests */
2924 spin_lock(&net_todo_list_lock);
2925 list_splice_init(&net_todo_list, &list);
2926 spin_unlock(&net_todo_list_lock);
2928 while (!list_empty(&list)) {
2929 struct net_device *dev
2930 = list_entry(list.next, struct net_device, todo_list);
2931 list_del(&dev->todo_list);
2933 switch(dev->reg_state) {
2934 case NETREG_REGISTERING:
2935 err = netdev_register_sysfs(dev);
2937 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2939 dev->reg_state = NETREG_REGISTERED;
2942 case NETREG_UNREGISTERING:
2943 netdev_unregister_sysfs(dev);
2944 dev->reg_state = NETREG_UNREGISTERED;
2946 netdev_wait_allrefs(dev);
2949 BUG_ON(atomic_read(&dev->refcnt));
2950 BUG_TRAP(!dev->ip_ptr);
2951 BUG_TRAP(!dev->ip6_ptr);
2952 BUG_TRAP(!dev->dn_ptr);
2955 /* It must be the very last action,
2956 * after this 'dev' may point to freed up memory.
2958 if (dev->destructor)
2959 dev->destructor(dev);
2963 printk(KERN_ERR "network todo '%s' but state %d\n",
2964 dev->name, dev->reg_state);
2970 up(&net_todo_run_mutex);
2974 * alloc_netdev - allocate network device
2975 * @sizeof_priv: size of private data to allocate space for
2976 * @name: device name format string
2977 * @setup: callback to initialize device
2979 * Allocates a struct net_device with private data area for driver use
2980 * and performs basic initialization.
2982 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
2983 void (*setup)(struct net_device *))
2986 struct net_device *dev;
2989 /* ensure 32-byte alignment of both the device and private area */
2990 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
2991 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
2993 p = kmalloc(alloc_size, GFP_KERNEL);
2995 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
2998 memset(p, 0, alloc_size);
3000 dev = (struct net_device *)
3001 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3002 dev->padded = (char *)dev - (char *)p;
3005 dev->priv = netdev_priv(dev);
3008 strcpy(dev->name, name);
3011 EXPORT_SYMBOL(alloc_netdev);
3014 * free_netdev - free network device
3017 * This function does the last stage of destroying an allocated device
3018 * interface. The reference to the device object is released.
3019 * If this is the last reference then it will be freed.
3021 void free_netdev(struct net_device *dev)
3024 /* Compatiablity with error handling in drivers */
3025 if (dev->reg_state == NETREG_UNINITIALIZED) {
3026 kfree((char *)dev - dev->padded);
3030 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3031 dev->reg_state = NETREG_RELEASED;
3033 /* will free via class release */
3034 class_device_put(&dev->class_dev);
3036 kfree((char *)dev - dev->padded);
3040 /* Synchronize with packet receive processing. */
3041 void synchronize_net(void)
3048 * unregister_netdevice - remove device from the kernel
3051 * This function shuts down a device interface and removes it
3052 * from the kernel tables. On success 0 is returned, on a failure
3053 * a negative errno code is returned.
3055 * Callers must hold the rtnl semaphore. You may want
3056 * unregister_netdev() instead of this.
3059 int unregister_netdevice(struct net_device *dev)
3061 struct net_device *d, **dp;
3063 BUG_ON(dev_boot_phase);
3066 /* Some devices call without registering for initialization unwind. */
3067 if (dev->reg_state == NETREG_UNINITIALIZED) {
3068 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3069 "was registered\n", dev->name, dev);
3073 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3075 /* If device is running, close it first. */
3076 if (dev->flags & IFF_UP)
3079 /* And unlink it from device chain. */
3080 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3082 write_lock_bh(&dev_base_lock);
3083 hlist_del(&dev->name_hlist);
3084 hlist_del(&dev->index_hlist);
3085 if (dev_tail == &dev->next)
3088 write_unlock_bh(&dev_base_lock);
3093 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3098 dev->reg_state = NETREG_UNREGISTERING;
3102 /* Shutdown queueing discipline. */
3106 /* Notify protocols, that we are about to destroy
3107 this device. They should clean all the things.
3109 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3112 * Flush the multicast chain
3114 dev_mc_discard(dev);
3119 /* Notifier chain MUST detach us from master device. */
3120 BUG_TRAP(!dev->master);
3122 free_divert_blk(dev);
3124 /* Finish processing unregister after unlock */
3134 * unregister_netdev - remove device from the kernel
3137 * This function shuts down a device interface and removes it
3138 * from the kernel tables. On success 0 is returned, on a failure
3139 * a negative errno code is returned.
3141 * This is just a wrapper for unregister_netdevice that takes
3142 * the rtnl semaphore. In general you want to use this and not
3143 * unregister_netdevice.
3145 void unregister_netdev(struct net_device *dev)
3148 unregister_netdevice(dev);
3152 EXPORT_SYMBOL(unregister_netdev);
3154 #ifdef CONFIG_HOTPLUG_CPU
3155 static int dev_cpu_callback(struct notifier_block *nfb,
3156 unsigned long action,
3159 struct sk_buff **list_skb;
3160 struct net_device **list_net;
3161 struct sk_buff *skb;
3162 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3163 struct softnet_data *sd, *oldsd;
3165 if (action != CPU_DEAD)
3168 local_irq_disable();
3169 cpu = smp_processor_id();
3170 sd = &per_cpu(softnet_data, cpu);
3171 oldsd = &per_cpu(softnet_data, oldcpu);
3173 /* Find end of our completion_queue. */
3174 list_skb = &sd->completion_queue;
3176 list_skb = &(*list_skb)->next;
3177 /* Append completion queue from offline CPU. */
3178 *list_skb = oldsd->completion_queue;
3179 oldsd->completion_queue = NULL;
3181 /* Find end of our output_queue. */
3182 list_net = &sd->output_queue;
3184 list_net = &(*list_net)->next_sched;
3185 /* Append output queue from offline CPU. */
3186 *list_net = oldsd->output_queue;
3187 oldsd->output_queue = NULL;
3189 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3192 /* Process offline CPU's input_pkt_queue */
3193 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3198 #endif /* CONFIG_HOTPLUG_CPU */
3202 * Initialize the DEV module. At boot time this walks the device list and
3203 * unhooks any devices that fail to initialise (normally hardware not
3204 * present) and leaves us with a valid list of present and active devices.
3209 * This is called single threaded during boot, so no need
3210 * to take the rtnl semaphore.
3212 static int __init net_dev_init(void)
3214 int i, rc = -ENOMEM;
3216 BUG_ON(!dev_boot_phase);
3220 if (dev_proc_init())
3223 if (netdev_sysfs_init())
3226 INIT_LIST_HEAD(&ptype_all);
3227 for (i = 0; i < 16; i++)
3228 INIT_LIST_HEAD(&ptype_base[i]);
3230 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3231 INIT_HLIST_HEAD(&dev_name_head[i]);
3233 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3234 INIT_HLIST_HEAD(&dev_index_head[i]);
3237 * Initialise the packet receive queues.
3240 for (i = 0; i < NR_CPUS; i++) {
3241 struct softnet_data *queue;
3243 queue = &per_cpu(softnet_data, i);
3244 skb_queue_head_init(&queue->input_pkt_queue);
3245 queue->completion_queue = NULL;
3246 INIT_LIST_HEAD(&queue->poll_list);
3247 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3248 queue->backlog_dev.weight = weight_p;
3249 queue->backlog_dev.poll = process_backlog;
3250 atomic_set(&queue->backlog_dev.refcnt, 1);
3255 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3256 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3258 hotcpu_notifier(dev_cpu_callback, 0);
3266 subsys_initcall(net_dev_init);
3268 EXPORT_SYMBOL(__dev_get_by_index);
3269 EXPORT_SYMBOL(__dev_get_by_name);
3270 EXPORT_SYMBOL(__dev_remove_pack);
3271 EXPORT_SYMBOL(__skb_linearize);
3272 EXPORT_SYMBOL(dev_add_pack);
3273 EXPORT_SYMBOL(dev_alloc_name);
3274 EXPORT_SYMBOL(dev_close);
3275 EXPORT_SYMBOL(dev_get_by_flags);
3276 EXPORT_SYMBOL(dev_get_by_index);
3277 EXPORT_SYMBOL(dev_get_by_name);
3278 EXPORT_SYMBOL(dev_ioctl);
3279 EXPORT_SYMBOL(dev_open);
3280 EXPORT_SYMBOL(dev_queue_xmit);
3281 EXPORT_SYMBOL(dev_remove_pack);
3282 EXPORT_SYMBOL(dev_set_allmulti);
3283 EXPORT_SYMBOL(dev_set_promiscuity);
3284 EXPORT_SYMBOL(dev_change_flags);
3285 EXPORT_SYMBOL(dev_set_mtu);
3286 EXPORT_SYMBOL(dev_set_mac_address);
3287 EXPORT_SYMBOL(free_netdev);
3288 EXPORT_SYMBOL(netdev_boot_setup_check);
3289 EXPORT_SYMBOL(netdev_set_master);
3290 EXPORT_SYMBOL(netdev_state_change);
3291 EXPORT_SYMBOL(netif_receive_skb);
3292 EXPORT_SYMBOL(netif_rx);
3293 EXPORT_SYMBOL(register_gifconf);
3294 EXPORT_SYMBOL(register_netdevice);
3295 EXPORT_SYMBOL(register_netdevice_notifier);
3296 EXPORT_SYMBOL(skb_checksum_help);
3297 EXPORT_SYMBOL(synchronize_net);
3298 EXPORT_SYMBOL(unregister_netdevice);
3299 EXPORT_SYMBOL(unregister_netdevice_notifier);
3300 EXPORT_SYMBOL(net_enable_timestamp);
3301 EXPORT_SYMBOL(net_disable_timestamp);
3302 EXPORT_SYMBOL(dev_get_flags);
3304 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3305 EXPORT_SYMBOL(br_handle_frame_hook);
3306 EXPORT_SYMBOL(br_fdb_get_hook);
3307 EXPORT_SYMBOL(br_fdb_put_hook);
3311 EXPORT_SYMBOL(dev_load);
3314 EXPORT_PER_CPU_SYMBOL(softnet_data);