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 #ifdef CONFIG_HIGHMEM
1112 /* Actually, we should eliminate this check as soon as we know, that:
1113 * 1. IOMMU is present and allows to map all the memory.
1114 * 2. No high memory really exists on this machine.
1117 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1121 if (dev->features & NETIF_F_HIGHDMA)
1124 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1125 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1131 #define illegal_highdma(dev, skb) (0)
1134 /* Keep head the same: replace data */
1135 int __skb_linearize(struct sk_buff *skb, unsigned int __nocast gfp_mask)
1140 struct skb_shared_info *ninfo;
1141 int headerlen = skb->data - skb->head;
1142 int expand = (skb->tail + skb->data_len) - skb->end;
1144 if (skb_shared(skb))
1150 size = skb->end - skb->head + expand;
1151 size = SKB_DATA_ALIGN(size);
1152 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1156 /* Copy entire thing */
1157 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1161 ninfo = (struct skb_shared_info*)(data + size);
1162 atomic_set(&ninfo->dataref, 1);
1163 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1164 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1165 ninfo->nr_frags = 0;
1166 ninfo->frag_list = NULL;
1168 /* Offset between the two in bytes */
1169 offset = data - skb->head;
1171 /* Free old data. */
1172 skb_release_data(skb);
1175 skb->end = data + size;
1177 /* Set up new pointers */
1178 skb->h.raw += offset;
1179 skb->nh.raw += offset;
1180 skb->mac.raw += offset;
1181 skb->tail += offset;
1182 skb->data += offset;
1184 /* We are no longer a clone, even if we were. */
1187 skb->tail += skb->data_len;
1192 #define HARD_TX_LOCK(dev, cpu) { \
1193 if ((dev->features & NETIF_F_LLTX) == 0) { \
1194 spin_lock(&dev->xmit_lock); \
1195 dev->xmit_lock_owner = cpu; \
1199 #define HARD_TX_UNLOCK(dev) { \
1200 if ((dev->features & NETIF_F_LLTX) == 0) { \
1201 dev->xmit_lock_owner = -1; \
1202 spin_unlock(&dev->xmit_lock); \
1207 * dev_queue_xmit - transmit a buffer
1208 * @skb: buffer to transmit
1210 * Queue a buffer for transmission to a network device. The caller must
1211 * have set the device and priority and built the buffer before calling
1212 * this function. The function can be called from an interrupt.
1214 * A negative errno code is returned on a failure. A success does not
1215 * guarantee the frame will be transmitted as it may be dropped due
1216 * to congestion or traffic shaping.
1218 * -----------------------------------------------------------------------------------
1219 * I notice this method can also return errors from the queue disciplines,
1220 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1223 * Regardless of the return value, the skb is consumed, so it is currently
1224 * difficult to retry a send to this method. (You can bump the ref count
1225 * before sending to hold a reference for retry if you are careful.)
1227 * When calling this method, interrupts MUST be enabled. This is because
1228 * the BH enable code must have IRQs enabled so that it will not deadlock.
1232 int dev_queue_xmit(struct sk_buff *skb)
1234 struct net_device *dev = skb->dev;
1238 if (skb_shinfo(skb)->frag_list &&
1239 !(dev->features & NETIF_F_FRAGLIST) &&
1240 __skb_linearize(skb, GFP_ATOMIC))
1243 /* Fragmented skb is linearized if device does not support SG,
1244 * or if at least one of fragments is in highmem and device
1245 * does not support DMA from it.
1247 if (skb_shinfo(skb)->nr_frags &&
1248 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1249 __skb_linearize(skb, GFP_ATOMIC))
1252 /* If packet is not checksummed and device does not support
1253 * checksumming for this protocol, complete checksumming here.
1255 if (skb->ip_summed == CHECKSUM_HW &&
1256 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1257 (!(dev->features & NETIF_F_IP_CSUM) ||
1258 skb->protocol != htons(ETH_P_IP))))
1259 if (skb_checksum_help(skb, 0))
1262 /* Disable soft irqs for various locks below. Also
1263 * stops preemption for RCU.
1267 /* Updates of qdisc are serialized by queue_lock.
1268 * The struct Qdisc which is pointed to by qdisc is now a
1269 * rcu structure - it may be accessed without acquiring
1270 * a lock (but the structure may be stale.) The freeing of the
1271 * qdisc will be deferred until it's known that there are no
1272 * more references to it.
1274 * If the qdisc has an enqueue function, we still need to
1275 * hold the queue_lock before calling it, since queue_lock
1276 * also serializes access to the device queue.
1279 q = rcu_dereference(dev->qdisc);
1280 #ifdef CONFIG_NET_CLS_ACT
1281 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1284 /* Grab device queue */
1285 spin_lock(&dev->queue_lock);
1287 rc = q->enqueue(skb, q);
1291 spin_unlock(&dev->queue_lock);
1292 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1296 /* The device has no queue. Common case for software devices:
1297 loopback, all the sorts of tunnels...
1299 Really, it is unlikely that xmit_lock protection is necessary here.
1300 (f.e. loopback and IP tunnels are clean ignoring statistics
1302 However, it is possible, that they rely on protection
1305 Check this and shot the lock. It is not prone from deadlocks.
1306 Either shot noqueue qdisc, it is even simpler 8)
1308 if (dev->flags & IFF_UP) {
1309 int cpu = smp_processor_id(); /* ok because BHs are off */
1311 if (dev->xmit_lock_owner != cpu) {
1313 HARD_TX_LOCK(dev, cpu);
1315 if (!netif_queue_stopped(dev)) {
1317 dev_queue_xmit_nit(skb, dev);
1320 if (!dev->hard_start_xmit(skb, dev)) {
1321 HARD_TX_UNLOCK(dev);
1325 HARD_TX_UNLOCK(dev);
1326 if (net_ratelimit())
1327 printk(KERN_CRIT "Virtual device %s asks to "
1328 "queue packet!\n", dev->name);
1330 /* Recursion is detected! It is possible,
1332 if (net_ratelimit())
1333 printk(KERN_CRIT "Dead loop on virtual device "
1334 "%s, fix it urgently!\n", dev->name);
1350 /*=======================================================================
1352 =======================================================================*/
1354 int netdev_max_backlog = 1000;
1355 int netdev_budget = 300;
1356 int weight_p = 64; /* old backlog weight */
1358 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1362 * netif_rx - post buffer to the network code
1363 * @skb: buffer to post
1365 * This function receives a packet from a device driver and queues it for
1366 * the upper (protocol) levels to process. It always succeeds. The buffer
1367 * may be dropped during processing for congestion control or by the
1371 * NET_RX_SUCCESS (no congestion)
1372 * NET_RX_CN_LOW (low congestion)
1373 * NET_RX_CN_MOD (moderate congestion)
1374 * NET_RX_CN_HIGH (high congestion)
1375 * NET_RX_DROP (packet was dropped)
1379 int netif_rx(struct sk_buff *skb)
1381 struct softnet_data *queue;
1382 unsigned long flags;
1384 /* if netpoll wants it, pretend we never saw it */
1385 if (netpoll_rx(skb))
1388 if (!skb->tstamp.off_sec)
1392 * The code is rearranged so that the path is the most
1393 * short when CPU is congested, but is still operating.
1395 local_irq_save(flags);
1396 queue = &__get_cpu_var(softnet_data);
1398 __get_cpu_var(netdev_rx_stat).total++;
1399 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1400 if (queue->input_pkt_queue.qlen) {
1403 __skb_queue_tail(&queue->input_pkt_queue, skb);
1404 local_irq_restore(flags);
1405 return NET_RX_SUCCESS;
1408 netif_rx_schedule(&queue->backlog_dev);
1412 __get_cpu_var(netdev_rx_stat).dropped++;
1413 local_irq_restore(flags);
1419 int netif_rx_ni(struct sk_buff *skb)
1424 err = netif_rx(skb);
1425 if (local_softirq_pending())
1432 EXPORT_SYMBOL(netif_rx_ni);
1434 static inline struct net_device *skb_bond(struct sk_buff *skb)
1436 struct net_device *dev = skb->dev;
1439 skb->dev = dev->master;
1444 static void net_tx_action(struct softirq_action *h)
1446 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1448 if (sd->completion_queue) {
1449 struct sk_buff *clist;
1451 local_irq_disable();
1452 clist = sd->completion_queue;
1453 sd->completion_queue = NULL;
1457 struct sk_buff *skb = clist;
1458 clist = clist->next;
1460 BUG_TRAP(!atomic_read(&skb->users));
1465 if (sd->output_queue) {
1466 struct net_device *head;
1468 local_irq_disable();
1469 head = sd->output_queue;
1470 sd->output_queue = NULL;
1474 struct net_device *dev = head;
1475 head = head->next_sched;
1477 smp_mb__before_clear_bit();
1478 clear_bit(__LINK_STATE_SCHED, &dev->state);
1480 if (spin_trylock(&dev->queue_lock)) {
1482 spin_unlock(&dev->queue_lock);
1484 netif_schedule(dev);
1490 static __inline__ int deliver_skb(struct sk_buff *skb,
1491 struct packet_type *pt_prev,
1492 struct net_device *orig_dev)
1494 atomic_inc(&skb->users);
1495 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1498 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1499 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1501 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1502 unsigned char *addr);
1503 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1505 static __inline__ int handle_bridge(struct sk_buff **pskb,
1506 struct packet_type **pt_prev, int *ret,
1507 struct net_device *orig_dev)
1509 struct net_bridge_port *port;
1511 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1512 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1516 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1520 return br_handle_frame_hook(port, pskb);
1523 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1526 #ifdef CONFIG_NET_CLS_ACT
1527 /* TODO: Maybe we should just force sch_ingress to be compiled in
1528 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1529 * a compare and 2 stores extra right now if we dont have it on
1530 * but have CONFIG_NET_CLS_ACT
1531 * NOTE: This doesnt stop any functionality; if you dont have
1532 * the ingress scheduler, you just cant add policies on ingress.
1535 static int ing_filter(struct sk_buff *skb)
1538 struct net_device *dev = skb->dev;
1539 int result = TC_ACT_OK;
1541 if (dev->qdisc_ingress) {
1542 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1543 if (MAX_RED_LOOP < ttl++) {
1544 printk("Redir loop detected Dropping packet (%s->%s)\n",
1545 skb->input_dev->name, skb->dev->name);
1549 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1551 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1553 spin_lock(&dev->ingress_lock);
1554 if ((q = dev->qdisc_ingress) != NULL)
1555 result = q->enqueue(skb, q);
1556 spin_unlock(&dev->ingress_lock);
1564 int netif_receive_skb(struct sk_buff *skb)
1566 struct packet_type *ptype, *pt_prev;
1567 struct net_device *orig_dev;
1568 int ret = NET_RX_DROP;
1569 unsigned short type;
1571 /* if we've gotten here through NAPI, check netpoll */
1572 if (skb->dev->poll && netpoll_rx(skb))
1575 if (!skb->tstamp.off_sec)
1578 if (!skb->input_dev)
1579 skb->input_dev = skb->dev;
1581 orig_dev = skb_bond(skb);
1583 __get_cpu_var(netdev_rx_stat).total++;
1585 skb->h.raw = skb->nh.raw = skb->data;
1586 skb->mac_len = skb->nh.raw - skb->mac.raw;
1592 #ifdef CONFIG_NET_CLS_ACT
1593 if (skb->tc_verd & TC_NCLS) {
1594 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1599 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1600 if (!ptype->dev || ptype->dev == skb->dev) {
1602 ret = deliver_skb(skb, pt_prev, orig_dev);
1607 #ifdef CONFIG_NET_CLS_ACT
1609 ret = deliver_skb(skb, pt_prev, orig_dev);
1610 pt_prev = NULL; /* noone else should process this after*/
1612 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1615 ret = ing_filter(skb);
1617 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1626 handle_diverter(skb);
1628 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1631 type = skb->protocol;
1632 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1633 if (ptype->type == type &&
1634 (!ptype->dev || ptype->dev == skb->dev)) {
1636 ret = deliver_skb(skb, pt_prev, orig_dev);
1642 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1645 /* Jamal, now you will not able to escape explaining
1646 * me how you were going to use this. :-)
1656 static int process_backlog(struct net_device *backlog_dev, int *budget)
1659 int quota = min(backlog_dev->quota, *budget);
1660 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1661 unsigned long start_time = jiffies;
1663 backlog_dev->weight = weight_p;
1665 struct sk_buff *skb;
1666 struct net_device *dev;
1668 local_irq_disable();
1669 skb = __skb_dequeue(&queue->input_pkt_queue);
1676 netif_receive_skb(skb);
1682 if (work >= quota || jiffies - start_time > 1)
1687 backlog_dev->quota -= work;
1692 backlog_dev->quota -= work;
1695 list_del(&backlog_dev->poll_list);
1696 smp_mb__before_clear_bit();
1697 netif_poll_enable(backlog_dev);
1703 static void net_rx_action(struct softirq_action *h)
1705 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1706 unsigned long start_time = jiffies;
1707 int budget = netdev_budget;
1710 local_irq_disable();
1712 while (!list_empty(&queue->poll_list)) {
1713 struct net_device *dev;
1715 if (budget <= 0 || jiffies - start_time > 1)
1720 dev = list_entry(queue->poll_list.next,
1721 struct net_device, poll_list);
1722 have = netpoll_poll_lock(dev);
1724 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1725 netpoll_poll_unlock(have);
1726 local_irq_disable();
1727 list_del(&dev->poll_list);
1728 list_add_tail(&dev->poll_list, &queue->poll_list);
1730 dev->quota += dev->weight;
1732 dev->quota = dev->weight;
1734 netpoll_poll_unlock(have);
1736 local_irq_disable();
1744 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1745 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1749 static gifconf_func_t * gifconf_list [NPROTO];
1752 * register_gifconf - register a SIOCGIF handler
1753 * @family: Address family
1754 * @gifconf: Function handler
1756 * Register protocol dependent address dumping routines. The handler
1757 * that is passed must not be freed or reused until it has been replaced
1758 * by another handler.
1760 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1762 if (family >= NPROTO)
1764 gifconf_list[family] = gifconf;
1770 * Map an interface index to its name (SIOCGIFNAME)
1774 * We need this ioctl for efficient implementation of the
1775 * if_indextoname() function required by the IPv6 API. Without
1776 * it, we would have to search all the interfaces to find a
1780 static int dev_ifname(struct ifreq __user *arg)
1782 struct net_device *dev;
1786 * Fetch the caller's info block.
1789 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1792 read_lock(&dev_base_lock);
1793 dev = __dev_get_by_index(ifr.ifr_ifindex);
1795 read_unlock(&dev_base_lock);
1799 strcpy(ifr.ifr_name, dev->name);
1800 read_unlock(&dev_base_lock);
1802 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1808 * Perform a SIOCGIFCONF call. This structure will change
1809 * size eventually, and there is nothing I can do about it.
1810 * Thus we will need a 'compatibility mode'.
1813 static int dev_ifconf(char __user *arg)
1816 struct net_device *dev;
1823 * Fetch the caller's info block.
1826 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1833 * Loop over the interfaces, and write an info block for each.
1837 for (dev = dev_base; dev; dev = dev->next) {
1838 for (i = 0; i < NPROTO; i++) {
1839 if (gifconf_list[i]) {
1842 done = gifconf_list[i](dev, NULL, 0);
1844 done = gifconf_list[i](dev, pos + total,
1854 * All done. Write the updated control block back to the caller.
1856 ifc.ifc_len = total;
1859 * Both BSD and Solaris return 0 here, so we do too.
1861 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1864 #ifdef CONFIG_PROC_FS
1866 * This is invoked by the /proc filesystem handler to display a device
1869 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1871 struct net_device *dev;
1874 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1876 return i == pos ? dev : NULL;
1879 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1881 read_lock(&dev_base_lock);
1882 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1885 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1888 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1891 void dev_seq_stop(struct seq_file *seq, void *v)
1893 read_unlock(&dev_base_lock);
1896 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1898 if (dev->get_stats) {
1899 struct net_device_stats *stats = dev->get_stats(dev);
1901 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1902 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1903 dev->name, stats->rx_bytes, stats->rx_packets,
1905 stats->rx_dropped + stats->rx_missed_errors,
1906 stats->rx_fifo_errors,
1907 stats->rx_length_errors + stats->rx_over_errors +
1908 stats->rx_crc_errors + stats->rx_frame_errors,
1909 stats->rx_compressed, stats->multicast,
1910 stats->tx_bytes, stats->tx_packets,
1911 stats->tx_errors, stats->tx_dropped,
1912 stats->tx_fifo_errors, stats->collisions,
1913 stats->tx_carrier_errors +
1914 stats->tx_aborted_errors +
1915 stats->tx_window_errors +
1916 stats->tx_heartbeat_errors,
1917 stats->tx_compressed);
1919 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1923 * Called from the PROCfs module. This now uses the new arbitrary sized
1924 * /proc/net interface to create /proc/net/dev
1926 static int dev_seq_show(struct seq_file *seq, void *v)
1928 if (v == SEQ_START_TOKEN)
1929 seq_puts(seq, "Inter-| Receive "
1931 " face |bytes packets errs drop fifo frame "
1932 "compressed multicast|bytes packets errs "
1933 "drop fifo colls carrier compressed\n");
1935 dev_seq_printf_stats(seq, v);
1939 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1941 struct netif_rx_stats *rc = NULL;
1943 while (*pos < NR_CPUS)
1944 if (cpu_online(*pos)) {
1945 rc = &per_cpu(netdev_rx_stat, *pos);
1952 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1954 return softnet_get_online(pos);
1957 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1960 return softnet_get_online(pos);
1963 static void softnet_seq_stop(struct seq_file *seq, void *v)
1967 static int softnet_seq_show(struct seq_file *seq, void *v)
1969 struct netif_rx_stats *s = v;
1971 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
1972 s->total, s->dropped, s->time_squeeze, 0,
1973 0, 0, 0, 0, /* was fastroute */
1978 static struct seq_operations dev_seq_ops = {
1979 .start = dev_seq_start,
1980 .next = dev_seq_next,
1981 .stop = dev_seq_stop,
1982 .show = dev_seq_show,
1985 static int dev_seq_open(struct inode *inode, struct file *file)
1987 return seq_open(file, &dev_seq_ops);
1990 static struct file_operations dev_seq_fops = {
1991 .owner = THIS_MODULE,
1992 .open = dev_seq_open,
1994 .llseek = seq_lseek,
1995 .release = seq_release,
1998 static struct seq_operations softnet_seq_ops = {
1999 .start = softnet_seq_start,
2000 .next = softnet_seq_next,
2001 .stop = softnet_seq_stop,
2002 .show = softnet_seq_show,
2005 static int softnet_seq_open(struct inode *inode, struct file *file)
2007 return seq_open(file, &softnet_seq_ops);
2010 static struct file_operations softnet_seq_fops = {
2011 .owner = THIS_MODULE,
2012 .open = softnet_seq_open,
2014 .llseek = seq_lseek,
2015 .release = seq_release,
2019 extern int wireless_proc_init(void);
2021 #define wireless_proc_init() 0
2024 static int __init dev_proc_init(void)
2028 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2030 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2032 if (wireless_proc_init())
2038 proc_net_remove("softnet_stat");
2040 proc_net_remove("dev");
2044 #define dev_proc_init() 0
2045 #endif /* CONFIG_PROC_FS */
2049 * netdev_set_master - set up master/slave pair
2050 * @slave: slave device
2051 * @master: new master device
2053 * Changes the master device of the slave. Pass %NULL to break the
2054 * bonding. The caller must hold the RTNL semaphore. On a failure
2055 * a negative errno code is returned. On success the reference counts
2056 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2057 * function returns zero.
2059 int netdev_set_master(struct net_device *slave, struct net_device *master)
2061 struct net_device *old = slave->master;
2071 slave->master = master;
2079 slave->flags |= IFF_SLAVE;
2081 slave->flags &= ~IFF_SLAVE;
2083 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2088 * dev_set_promiscuity - update promiscuity count on a device
2092 * Add or remove promsicuity from a device. While the count in the device
2093 * remains above zero the interface remains promiscuous. Once it hits zero
2094 * the device reverts back to normal filtering operation. A negative inc
2095 * value is used to drop promiscuity on the device.
2097 void dev_set_promiscuity(struct net_device *dev, int inc)
2099 unsigned short old_flags = dev->flags;
2101 if ((dev->promiscuity += inc) == 0)
2102 dev->flags &= ~IFF_PROMISC;
2104 dev->flags |= IFF_PROMISC;
2105 if (dev->flags != old_flags) {
2107 printk(KERN_INFO "device %s %s promiscuous mode\n",
2108 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2114 * dev_set_allmulti - update allmulti count on a device
2118 * Add or remove reception of all multicast frames to a device. While the
2119 * count in the device remains above zero the interface remains listening
2120 * to all interfaces. Once it hits zero the device reverts back to normal
2121 * filtering operation. A negative @inc value is used to drop the counter
2122 * when releasing a resource needing all multicasts.
2125 void dev_set_allmulti(struct net_device *dev, int inc)
2127 unsigned short old_flags = dev->flags;
2129 dev->flags |= IFF_ALLMULTI;
2130 if ((dev->allmulti += inc) == 0)
2131 dev->flags &= ~IFF_ALLMULTI;
2132 if (dev->flags ^ old_flags)
2136 unsigned dev_get_flags(const struct net_device *dev)
2140 flags = (dev->flags & ~(IFF_PROMISC |
2143 (dev->gflags & (IFF_PROMISC |
2146 if (netif_running(dev) && netif_carrier_ok(dev))
2147 flags |= IFF_RUNNING;
2152 int dev_change_flags(struct net_device *dev, unsigned flags)
2155 int old_flags = dev->flags;
2158 * Set the flags on our device.
2161 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2162 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2164 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2168 * Load in the correct multicast list now the flags have changed.
2174 * Have we downed the interface. We handle IFF_UP ourselves
2175 * according to user attempts to set it, rather than blindly
2180 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2181 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2187 if (dev->flags & IFF_UP &&
2188 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2190 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2192 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2193 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2194 dev->gflags ^= IFF_PROMISC;
2195 dev_set_promiscuity(dev, inc);
2198 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2199 is important. Some (broken) drivers set IFF_PROMISC, when
2200 IFF_ALLMULTI is requested not asking us and not reporting.
2202 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2203 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2204 dev->gflags ^= IFF_ALLMULTI;
2205 dev_set_allmulti(dev, inc);
2208 if (old_flags ^ dev->flags)
2209 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2214 int dev_set_mtu(struct net_device *dev, int new_mtu)
2218 if (new_mtu == dev->mtu)
2221 /* MTU must be positive. */
2225 if (!netif_device_present(dev))
2229 if (dev->change_mtu)
2230 err = dev->change_mtu(dev, new_mtu);
2233 if (!err && dev->flags & IFF_UP)
2234 notifier_call_chain(&netdev_chain,
2235 NETDEV_CHANGEMTU, dev);
2239 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2243 if (!dev->set_mac_address)
2245 if (sa->sa_family != dev->type)
2247 if (!netif_device_present(dev))
2249 err = dev->set_mac_address(dev, sa);
2251 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2256 * Perform the SIOCxIFxxx calls.
2258 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2261 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2267 case SIOCGIFFLAGS: /* Get interface flags */
2268 ifr->ifr_flags = dev_get_flags(dev);
2271 case SIOCSIFFLAGS: /* Set interface flags */
2272 return dev_change_flags(dev, ifr->ifr_flags);
2274 case SIOCGIFMETRIC: /* Get the metric on the interface
2275 (currently unused) */
2276 ifr->ifr_metric = 0;
2279 case SIOCSIFMETRIC: /* Set the metric on the interface
2280 (currently unused) */
2283 case SIOCGIFMTU: /* Get the MTU of a device */
2284 ifr->ifr_mtu = dev->mtu;
2287 case SIOCSIFMTU: /* Set the MTU of a device */
2288 return dev_set_mtu(dev, ifr->ifr_mtu);
2292 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2294 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2295 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2296 ifr->ifr_hwaddr.sa_family = dev->type;
2300 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2302 case SIOCSIFHWBROADCAST:
2303 if (ifr->ifr_hwaddr.sa_family != dev->type)
2305 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2306 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2307 notifier_call_chain(&netdev_chain,
2308 NETDEV_CHANGEADDR, dev);
2312 ifr->ifr_map.mem_start = dev->mem_start;
2313 ifr->ifr_map.mem_end = dev->mem_end;
2314 ifr->ifr_map.base_addr = dev->base_addr;
2315 ifr->ifr_map.irq = dev->irq;
2316 ifr->ifr_map.dma = dev->dma;
2317 ifr->ifr_map.port = dev->if_port;
2321 if (dev->set_config) {
2322 if (!netif_device_present(dev))
2324 return dev->set_config(dev, &ifr->ifr_map);
2329 if (!dev->set_multicast_list ||
2330 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2332 if (!netif_device_present(dev))
2334 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2338 if (!dev->set_multicast_list ||
2339 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2341 if (!netif_device_present(dev))
2343 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2347 ifr->ifr_ifindex = dev->ifindex;
2351 ifr->ifr_qlen = dev->tx_queue_len;
2355 if (ifr->ifr_qlen < 0)
2357 dev->tx_queue_len = ifr->ifr_qlen;
2361 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2362 return dev_change_name(dev, ifr->ifr_newname);
2365 * Unknown or private ioctl
2369 if ((cmd >= SIOCDEVPRIVATE &&
2370 cmd <= SIOCDEVPRIVATE + 15) ||
2371 cmd == SIOCBONDENSLAVE ||
2372 cmd == SIOCBONDRELEASE ||
2373 cmd == SIOCBONDSETHWADDR ||
2374 cmd == SIOCBONDSLAVEINFOQUERY ||
2375 cmd == SIOCBONDINFOQUERY ||
2376 cmd == SIOCBONDCHANGEACTIVE ||
2377 cmd == SIOCGMIIPHY ||
2378 cmd == SIOCGMIIREG ||
2379 cmd == SIOCSMIIREG ||
2380 cmd == SIOCBRADDIF ||
2381 cmd == SIOCBRDELIF ||
2382 cmd == SIOCWANDEV) {
2384 if (dev->do_ioctl) {
2385 if (netif_device_present(dev))
2386 err = dev->do_ioctl(dev, ifr,
2399 * This function handles all "interface"-type I/O control requests. The actual
2400 * 'doing' part of this is dev_ifsioc above.
2404 * dev_ioctl - network device ioctl
2405 * @cmd: command to issue
2406 * @arg: pointer to a struct ifreq in user space
2408 * Issue ioctl functions to devices. This is normally called by the
2409 * user space syscall interfaces but can sometimes be useful for
2410 * other purposes. The return value is the return from the syscall if
2411 * positive or a negative errno code on error.
2414 int dev_ioctl(unsigned int cmd, void __user *arg)
2420 /* One special case: SIOCGIFCONF takes ifconf argument
2421 and requires shared lock, because it sleeps writing
2425 if (cmd == SIOCGIFCONF) {
2427 ret = dev_ifconf((char __user *) arg);
2431 if (cmd == SIOCGIFNAME)
2432 return dev_ifname((struct ifreq __user *)arg);
2434 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2437 ifr.ifr_name[IFNAMSIZ-1] = 0;
2439 colon = strchr(ifr.ifr_name, ':');
2444 * See which interface the caller is talking about.
2449 * These ioctl calls:
2450 * - can be done by all.
2451 * - atomic and do not require locking.
2462 dev_load(ifr.ifr_name);
2463 read_lock(&dev_base_lock);
2464 ret = dev_ifsioc(&ifr, cmd);
2465 read_unlock(&dev_base_lock);
2469 if (copy_to_user(arg, &ifr,
2470 sizeof(struct ifreq)))
2476 dev_load(ifr.ifr_name);
2478 ret = dev_ethtool(&ifr);
2483 if (copy_to_user(arg, &ifr,
2484 sizeof(struct ifreq)))
2490 * These ioctl calls:
2491 * - require superuser power.
2492 * - require strict serialization.
2498 if (!capable(CAP_NET_ADMIN))
2500 dev_load(ifr.ifr_name);
2502 ret = dev_ifsioc(&ifr, cmd);
2507 if (copy_to_user(arg, &ifr,
2508 sizeof(struct ifreq)))
2514 * These ioctl calls:
2515 * - require superuser power.
2516 * - require strict serialization.
2517 * - do not return a value
2527 case SIOCSIFHWBROADCAST:
2530 case SIOCBONDENSLAVE:
2531 case SIOCBONDRELEASE:
2532 case SIOCBONDSETHWADDR:
2533 case SIOCBONDSLAVEINFOQUERY:
2534 case SIOCBONDINFOQUERY:
2535 case SIOCBONDCHANGEACTIVE:
2538 if (!capable(CAP_NET_ADMIN))
2540 dev_load(ifr.ifr_name);
2542 ret = dev_ifsioc(&ifr, cmd);
2547 /* Get the per device memory space. We can add this but
2548 * currently do not support it */
2550 /* Set the per device memory buffer space.
2551 * Not applicable in our case */
2556 * Unknown or private ioctl.
2559 if (cmd == SIOCWANDEV ||
2560 (cmd >= SIOCDEVPRIVATE &&
2561 cmd <= SIOCDEVPRIVATE + 15)) {
2562 dev_load(ifr.ifr_name);
2564 ret = dev_ifsioc(&ifr, cmd);
2566 if (!ret && copy_to_user(arg, &ifr,
2567 sizeof(struct ifreq)))
2572 /* Take care of Wireless Extensions */
2573 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2574 /* If command is `set a parameter', or
2575 * `get the encoding parameters', check if
2576 * the user has the right to do it */
2577 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2578 if (!capable(CAP_NET_ADMIN))
2581 dev_load(ifr.ifr_name);
2583 /* Follow me in net/core/wireless.c */
2584 ret = wireless_process_ioctl(&ifr, cmd);
2586 if (IW_IS_GET(cmd) &&
2587 copy_to_user(arg, &ifr,
2588 sizeof(struct ifreq)))
2592 #endif /* WIRELESS_EXT */
2599 * dev_new_index - allocate an ifindex
2601 * Returns a suitable unique value for a new device interface
2602 * number. The caller must hold the rtnl semaphore or the
2603 * dev_base_lock to be sure it remains unique.
2605 static int dev_new_index(void)
2611 if (!__dev_get_by_index(ifindex))
2616 static int dev_boot_phase = 1;
2618 /* Delayed registration/unregisteration */
2619 static DEFINE_SPINLOCK(net_todo_list_lock);
2620 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2622 static inline void net_set_todo(struct net_device *dev)
2624 spin_lock(&net_todo_list_lock);
2625 list_add_tail(&dev->todo_list, &net_todo_list);
2626 spin_unlock(&net_todo_list_lock);
2630 * register_netdevice - register a network device
2631 * @dev: device to register
2633 * Take a completed network device structure and add it to the kernel
2634 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2635 * chain. 0 is returned on success. A negative errno code is returned
2636 * on a failure to set up the device, or if the name is a duplicate.
2638 * Callers must hold the rtnl semaphore. You may want
2639 * register_netdev() instead of this.
2642 * The locking appears insufficient to guarantee two parallel registers
2643 * will not get the same name.
2646 int register_netdevice(struct net_device *dev)
2648 struct hlist_head *head;
2649 struct hlist_node *p;
2652 BUG_ON(dev_boot_phase);
2655 /* When net_device's are persistent, this will be fatal. */
2656 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2658 spin_lock_init(&dev->queue_lock);
2659 spin_lock_init(&dev->xmit_lock);
2660 dev->xmit_lock_owner = -1;
2661 #ifdef CONFIG_NET_CLS_ACT
2662 spin_lock_init(&dev->ingress_lock);
2665 ret = alloc_divert_blk(dev);
2671 /* Init, if this function is available */
2673 ret = dev->init(dev);
2681 if (!dev_valid_name(dev->name)) {
2686 dev->ifindex = dev_new_index();
2687 if (dev->iflink == -1)
2688 dev->iflink = dev->ifindex;
2690 /* Check for existence of name */
2691 head = dev_name_hash(dev->name);
2692 hlist_for_each(p, head) {
2693 struct net_device *d
2694 = hlist_entry(p, struct net_device, name_hlist);
2695 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2701 /* Fix illegal SG+CSUM combinations. */
2702 if ((dev->features & NETIF_F_SG) &&
2703 !(dev->features & (NETIF_F_IP_CSUM |
2705 NETIF_F_HW_CSUM))) {
2706 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2708 dev->features &= ~NETIF_F_SG;
2711 /* TSO requires that SG is present as well. */
2712 if ((dev->features & NETIF_F_TSO) &&
2713 !(dev->features & NETIF_F_SG)) {
2714 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2716 dev->features &= ~NETIF_F_TSO;
2720 * nil rebuild_header routine,
2721 * that should be never called and used as just bug trap.
2724 if (!dev->rebuild_header)
2725 dev->rebuild_header = default_rebuild_header;
2728 * Default initial state at registry is that the
2729 * device is present.
2732 set_bit(__LINK_STATE_PRESENT, &dev->state);
2735 dev_init_scheduler(dev);
2736 write_lock_bh(&dev_base_lock);
2738 dev_tail = &dev->next;
2739 hlist_add_head(&dev->name_hlist, head);
2740 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2742 dev->reg_state = NETREG_REGISTERING;
2743 write_unlock_bh(&dev_base_lock);
2745 /* Notify protocols, that a new device appeared. */
2746 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2748 /* Finish registration after unlock */
2755 free_divert_blk(dev);
2760 * register_netdev - register a network device
2761 * @dev: device to register
2763 * Take a completed network device structure and add it to the kernel
2764 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2765 * chain. 0 is returned on success. A negative errno code is returned
2766 * on a failure to set up the device, or if the name is a duplicate.
2768 * This is a wrapper around register_netdev that takes the rtnl semaphore
2769 * and expands the device name if you passed a format string to
2772 int register_netdev(struct net_device *dev)
2779 * If the name is a format string the caller wants us to do a
2782 if (strchr(dev->name, '%')) {
2783 err = dev_alloc_name(dev, dev->name);
2789 * Back compatibility hook. Kill this one in 2.5
2791 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2792 err = dev_alloc_name(dev, "eth%d");
2797 err = register_netdevice(dev);
2802 EXPORT_SYMBOL(register_netdev);
2805 * netdev_wait_allrefs - wait until all references are gone.
2807 * This is called when unregistering network devices.
2809 * Any protocol or device that holds a reference should register
2810 * for netdevice notification, and cleanup and put back the
2811 * reference if they receive an UNREGISTER event.
2812 * We can get stuck here if buggy protocols don't correctly
2815 static void netdev_wait_allrefs(struct net_device *dev)
2817 unsigned long rebroadcast_time, warning_time;
2819 rebroadcast_time = warning_time = jiffies;
2820 while (atomic_read(&dev->refcnt) != 0) {
2821 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2824 /* Rebroadcast unregister notification */
2825 notifier_call_chain(&netdev_chain,
2826 NETDEV_UNREGISTER, dev);
2828 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2830 /* We must not have linkwatch events
2831 * pending on unregister. If this
2832 * happens, we simply run the queue
2833 * unscheduled, resulting in a noop
2836 linkwatch_run_queue();
2841 rebroadcast_time = jiffies;
2846 if (time_after(jiffies, warning_time + 10 * HZ)) {
2847 printk(KERN_EMERG "unregister_netdevice: "
2848 "waiting for %s to become free. Usage "
2850 dev->name, atomic_read(&dev->refcnt));
2851 warning_time = jiffies;
2860 * register_netdevice(x1);
2861 * register_netdevice(x2);
2863 * unregister_netdevice(y1);
2864 * unregister_netdevice(y2);
2870 * We are invoked by rtnl_unlock() after it drops the semaphore.
2871 * This allows us to deal with problems:
2872 * 1) We can create/delete sysfs objects which invoke hotplug
2873 * without deadlocking with linkwatch via keventd.
2874 * 2) Since we run with the RTNL semaphore not held, we can sleep
2875 * safely in order to wait for the netdev refcnt to drop to zero.
2877 static DECLARE_MUTEX(net_todo_run_mutex);
2878 void netdev_run_todo(void)
2880 struct list_head list = LIST_HEAD_INIT(list);
2884 /* Need to guard against multiple cpu's getting out of order. */
2885 down(&net_todo_run_mutex);
2887 /* Not safe to do outside the semaphore. We must not return
2888 * until all unregister events invoked by the local processor
2889 * have been completed (either by this todo run, or one on
2892 if (list_empty(&net_todo_list))
2895 /* Snapshot list, allow later requests */
2896 spin_lock(&net_todo_list_lock);
2897 list_splice_init(&net_todo_list, &list);
2898 spin_unlock(&net_todo_list_lock);
2900 while (!list_empty(&list)) {
2901 struct net_device *dev
2902 = list_entry(list.next, struct net_device, todo_list);
2903 list_del(&dev->todo_list);
2905 switch(dev->reg_state) {
2906 case NETREG_REGISTERING:
2907 err = netdev_register_sysfs(dev);
2909 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2911 dev->reg_state = NETREG_REGISTERED;
2914 case NETREG_UNREGISTERING:
2915 netdev_unregister_sysfs(dev);
2916 dev->reg_state = NETREG_UNREGISTERED;
2918 netdev_wait_allrefs(dev);
2921 BUG_ON(atomic_read(&dev->refcnt));
2922 BUG_TRAP(!dev->ip_ptr);
2923 BUG_TRAP(!dev->ip6_ptr);
2924 BUG_TRAP(!dev->dn_ptr);
2927 /* It must be the very last action,
2928 * after this 'dev' may point to freed up memory.
2930 if (dev->destructor)
2931 dev->destructor(dev);
2935 printk(KERN_ERR "network todo '%s' but state %d\n",
2936 dev->name, dev->reg_state);
2942 up(&net_todo_run_mutex);
2946 * alloc_netdev - allocate network device
2947 * @sizeof_priv: size of private data to allocate space for
2948 * @name: device name format string
2949 * @setup: callback to initialize device
2951 * Allocates a struct net_device with private data area for driver use
2952 * and performs basic initialization.
2954 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
2955 void (*setup)(struct net_device *))
2958 struct net_device *dev;
2961 /* ensure 32-byte alignment of both the device and private area */
2962 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
2963 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
2965 p = kmalloc(alloc_size, GFP_KERNEL);
2967 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
2970 memset(p, 0, alloc_size);
2972 dev = (struct net_device *)
2973 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
2974 dev->padded = (char *)dev - (char *)p;
2977 dev->priv = netdev_priv(dev);
2980 strcpy(dev->name, name);
2983 EXPORT_SYMBOL(alloc_netdev);
2986 * free_netdev - free network device
2989 * This function does the last stage of destroying an allocated device
2990 * interface. The reference to the device object is released.
2991 * If this is the last reference then it will be freed.
2993 void free_netdev(struct net_device *dev)
2996 /* Compatiablity with error handling in drivers */
2997 if (dev->reg_state == NETREG_UNINITIALIZED) {
2998 kfree((char *)dev - dev->padded);
3002 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3003 dev->reg_state = NETREG_RELEASED;
3005 /* will free via class release */
3006 class_device_put(&dev->class_dev);
3008 kfree((char *)dev - dev->padded);
3012 /* Synchronize with packet receive processing. */
3013 void synchronize_net(void)
3020 * unregister_netdevice - remove device from the kernel
3023 * This function shuts down a device interface and removes it
3024 * from the kernel tables. On success 0 is returned, on a failure
3025 * a negative errno code is returned.
3027 * Callers must hold the rtnl semaphore. You may want
3028 * unregister_netdev() instead of this.
3031 int unregister_netdevice(struct net_device *dev)
3033 struct net_device *d, **dp;
3035 BUG_ON(dev_boot_phase);
3038 /* Some devices call without registering for initialization unwind. */
3039 if (dev->reg_state == NETREG_UNINITIALIZED) {
3040 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3041 "was registered\n", dev->name, dev);
3045 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3047 /* If device is running, close it first. */
3048 if (dev->flags & IFF_UP)
3051 /* And unlink it from device chain. */
3052 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3054 write_lock_bh(&dev_base_lock);
3055 hlist_del(&dev->name_hlist);
3056 hlist_del(&dev->index_hlist);
3057 if (dev_tail == &dev->next)
3060 write_unlock_bh(&dev_base_lock);
3065 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3070 dev->reg_state = NETREG_UNREGISTERING;
3074 /* Shutdown queueing discipline. */
3078 /* Notify protocols, that we are about to destroy
3079 this device. They should clean all the things.
3081 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3084 * Flush the multicast chain
3086 dev_mc_discard(dev);
3091 /* Notifier chain MUST detach us from master device. */
3092 BUG_TRAP(!dev->master);
3094 free_divert_blk(dev);
3096 /* Finish processing unregister after unlock */
3106 * unregister_netdev - remove device from the kernel
3109 * This function shuts down a device interface and removes it
3110 * from the kernel tables. On success 0 is returned, on a failure
3111 * a negative errno code is returned.
3113 * This is just a wrapper for unregister_netdevice that takes
3114 * the rtnl semaphore. In general you want to use this and not
3115 * unregister_netdevice.
3117 void unregister_netdev(struct net_device *dev)
3120 unregister_netdevice(dev);
3124 EXPORT_SYMBOL(unregister_netdev);
3126 #ifdef CONFIG_HOTPLUG_CPU
3127 static int dev_cpu_callback(struct notifier_block *nfb,
3128 unsigned long action,
3131 struct sk_buff **list_skb;
3132 struct net_device **list_net;
3133 struct sk_buff *skb;
3134 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3135 struct softnet_data *sd, *oldsd;
3137 if (action != CPU_DEAD)
3140 local_irq_disable();
3141 cpu = smp_processor_id();
3142 sd = &per_cpu(softnet_data, cpu);
3143 oldsd = &per_cpu(softnet_data, oldcpu);
3145 /* Find end of our completion_queue. */
3146 list_skb = &sd->completion_queue;
3148 list_skb = &(*list_skb)->next;
3149 /* Append completion queue from offline CPU. */
3150 *list_skb = oldsd->completion_queue;
3151 oldsd->completion_queue = NULL;
3153 /* Find end of our output_queue. */
3154 list_net = &sd->output_queue;
3156 list_net = &(*list_net)->next_sched;
3157 /* Append output queue from offline CPU. */
3158 *list_net = oldsd->output_queue;
3159 oldsd->output_queue = NULL;
3161 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3164 /* Process offline CPU's input_pkt_queue */
3165 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3170 #endif /* CONFIG_HOTPLUG_CPU */
3174 * Initialize the DEV module. At boot time this walks the device list and
3175 * unhooks any devices that fail to initialise (normally hardware not
3176 * present) and leaves us with a valid list of present and active devices.
3181 * This is called single threaded during boot, so no need
3182 * to take the rtnl semaphore.
3184 static int __init net_dev_init(void)
3186 int i, rc = -ENOMEM;
3188 BUG_ON(!dev_boot_phase);
3192 if (dev_proc_init())
3195 if (netdev_sysfs_init())
3198 INIT_LIST_HEAD(&ptype_all);
3199 for (i = 0; i < 16; i++)
3200 INIT_LIST_HEAD(&ptype_base[i]);
3202 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3203 INIT_HLIST_HEAD(&dev_name_head[i]);
3205 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3206 INIT_HLIST_HEAD(&dev_index_head[i]);
3209 * Initialise the packet receive queues.
3212 for (i = 0; i < NR_CPUS; i++) {
3213 struct softnet_data *queue;
3215 queue = &per_cpu(softnet_data, i);
3216 skb_queue_head_init(&queue->input_pkt_queue);
3217 queue->completion_queue = NULL;
3218 INIT_LIST_HEAD(&queue->poll_list);
3219 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3220 queue->backlog_dev.weight = weight_p;
3221 queue->backlog_dev.poll = process_backlog;
3222 atomic_set(&queue->backlog_dev.refcnt, 1);
3227 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3228 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3230 hotcpu_notifier(dev_cpu_callback, 0);
3238 subsys_initcall(net_dev_init);
3240 EXPORT_SYMBOL(__dev_get_by_index);
3241 EXPORT_SYMBOL(__dev_get_by_name);
3242 EXPORT_SYMBOL(__dev_remove_pack);
3243 EXPORT_SYMBOL(__skb_linearize);
3244 EXPORT_SYMBOL(dev_add_pack);
3245 EXPORT_SYMBOL(dev_alloc_name);
3246 EXPORT_SYMBOL(dev_close);
3247 EXPORT_SYMBOL(dev_get_by_flags);
3248 EXPORT_SYMBOL(dev_get_by_index);
3249 EXPORT_SYMBOL(dev_get_by_name);
3250 EXPORT_SYMBOL(dev_ioctl);
3251 EXPORT_SYMBOL(dev_open);
3252 EXPORT_SYMBOL(dev_queue_xmit);
3253 EXPORT_SYMBOL(dev_remove_pack);
3254 EXPORT_SYMBOL(dev_set_allmulti);
3255 EXPORT_SYMBOL(dev_set_promiscuity);
3256 EXPORT_SYMBOL(dev_change_flags);
3257 EXPORT_SYMBOL(dev_set_mtu);
3258 EXPORT_SYMBOL(dev_set_mac_address);
3259 EXPORT_SYMBOL(free_netdev);
3260 EXPORT_SYMBOL(netdev_boot_setup_check);
3261 EXPORT_SYMBOL(netdev_set_master);
3262 EXPORT_SYMBOL(netdev_state_change);
3263 EXPORT_SYMBOL(netif_receive_skb);
3264 EXPORT_SYMBOL(netif_rx);
3265 EXPORT_SYMBOL(register_gifconf);
3266 EXPORT_SYMBOL(register_netdevice);
3267 EXPORT_SYMBOL(register_netdevice_notifier);
3268 EXPORT_SYMBOL(skb_checksum_help);
3269 EXPORT_SYMBOL(synchronize_net);
3270 EXPORT_SYMBOL(unregister_netdevice);
3271 EXPORT_SYMBOL(unregister_netdevice_notifier);
3272 EXPORT_SYMBOL(net_enable_timestamp);
3273 EXPORT_SYMBOL(net_disable_timestamp);
3274 EXPORT_SYMBOL(dev_get_flags);
3276 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3277 EXPORT_SYMBOL(br_handle_frame_hook);
3278 EXPORT_SYMBOL(br_fdb_get_hook);
3279 EXPORT_SYMBOL(br_fdb_put_hook);
3283 EXPORT_SYMBOL(dev_load);
3286 EXPORT_PER_CPU_SYMBOL(softnet_data);