2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/config.h>
80 #include <linux/cpu.h>
81 #include <linux/types.h>
82 #include <linux/kernel.h>
83 #include <linux/sched.h>
84 #include <linux/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/notifier.h>
94 #include <linux/skbuff.h>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
101 #include <linux/divert.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/kmod.h>
108 #include <linux/module.h>
109 #include <linux/kallsyms.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #ifdef CONFIG_NET_RADIO
114 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
115 #include <net/iw_handler.h>
116 #endif /* CONFIG_NET_RADIO */
117 #include <linux/vs_network.h>
118 #include <asm/current.h>
121 * The list of packet types we will receive (as opposed to discard)
122 * and the routines to invoke.
124 * Why 16. Because with 16 the only overlap we get on a hash of the
125 * low nibble of the protocol value is RARP/SNAP/X.25.
127 * NOTE: That is no longer true with the addition of VLAN tags. Not
128 * sure which should go first, but I bet it won't make much
129 * difference if we are running VLANs. The good news is that
130 * this protocol won't be in the list unless compiled in, so
131 * the average user (w/out VLANs) will not be adversly affected.
148 static DEFINE_SPINLOCK(ptype_lock);
149 static struct list_head ptype_base[16]; /* 16 way hashed list */
150 static struct list_head ptype_all; /* Taps */
153 * The @dev_base list is protected by @dev_base_lock and the rtln
156 * Pure readers hold dev_base_lock for reading.
158 * Writers must hold the rtnl semaphore while they loop through the
159 * dev_base list, and hold dev_base_lock for writing when they do the
160 * actual updates. This allows pure readers to access the list even
161 * while a writer is preparing to update it.
163 * To put it another way, dev_base_lock is held for writing only to
164 * protect against pure readers; the rtnl semaphore provides the
165 * protection against other writers.
167 * See, for example usages, register_netdevice() and
168 * unregister_netdevice(), which must be called with the rtnl
171 struct net_device *dev_base;
172 static struct net_device **dev_tail = &dev_base;
173 DEFINE_RWLOCK(dev_base_lock);
175 EXPORT_SYMBOL(dev_base);
176 EXPORT_SYMBOL(dev_base_lock);
178 #define NETDEV_HASHBITS 8
179 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
180 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
182 static inline struct hlist_head *dev_name_hash(const char *name)
184 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
185 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
188 static inline struct hlist_head *dev_index_hash(int ifindex)
190 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
197 static struct notifier_block *netdev_chain;
200 * Device drivers call our routines to queue packets here. We empty the
201 * queue in the local softnet handler.
203 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
206 extern int netdev_sysfs_init(void);
207 extern int netdev_register_sysfs(struct net_device *);
208 extern void netdev_unregister_sysfs(struct net_device *);
210 #define netdev_sysfs_init() (0)
211 #define netdev_register_sysfs(dev) (0)
212 #define netdev_unregister_sysfs(dev) do { } while(0)
216 /*******************************************************************************
218 Protocol management and registration routines
220 *******************************************************************************/
229 * Add a protocol ID to the list. Now that the input handler is
230 * smarter we can dispense with all the messy stuff that used to be
233 * BEWARE!!! Protocol handlers, mangling input packets,
234 * MUST BE last in hash buckets and checking protocol handlers
235 * MUST start from promiscuous ptype_all chain in net_bh.
236 * It is true now, do not change it.
237 * Explanation follows: if protocol handler, mangling packet, will
238 * be the first on list, it is not able to sense, that packet
239 * is cloned and should be copied-on-write, so that it will
240 * change it and subsequent readers will get broken packet.
245 * dev_add_pack - add packet handler
246 * @pt: packet type declaration
248 * Add a protocol handler to the networking stack. The passed &packet_type
249 * is linked into kernel lists and may not be freed until it has been
250 * removed from the kernel lists.
252 * This call does not sleep therefore it can not
253 * guarantee all CPU's that are in middle of receiving packets
254 * will see the new packet type (until the next received packet).
257 void dev_add_pack(struct packet_type *pt)
261 spin_lock_bh(&ptype_lock);
262 if (pt->type == htons(ETH_P_ALL)) {
264 list_add_rcu(&pt->list, &ptype_all);
266 hash = ntohs(pt->type) & 15;
267 list_add_rcu(&pt->list, &ptype_base[hash]);
269 spin_unlock_bh(&ptype_lock);
273 * __dev_remove_pack - remove packet handler
274 * @pt: packet type declaration
276 * Remove a protocol handler that was previously added to the kernel
277 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
278 * from the kernel lists and can be freed or reused once this function
281 * The packet type might still be in use by receivers
282 * and must not be freed until after all the CPU's have gone
283 * through a quiescent state.
285 void __dev_remove_pack(struct packet_type *pt)
287 struct list_head *head;
288 struct packet_type *pt1;
290 spin_lock_bh(&ptype_lock);
292 if (pt->type == htons(ETH_P_ALL)) {
296 head = &ptype_base[ntohs(pt->type) & 15];
298 list_for_each_entry(pt1, head, list) {
300 list_del_rcu(&pt->list);
305 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
307 spin_unlock_bh(&ptype_lock);
310 * dev_remove_pack - remove packet handler
311 * @pt: packet type declaration
313 * Remove a protocol handler that was previously added to the kernel
314 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
315 * from the kernel lists and can be freed or reused once this function
318 * This call sleeps to guarantee that no CPU is looking at the packet
321 void dev_remove_pack(struct packet_type *pt)
323 __dev_remove_pack(pt);
328 /******************************************************************************
330 Device Boot-time Settings Routines
332 *******************************************************************************/
334 /* Boot time configuration table */
335 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
338 * netdev_boot_setup_add - add new setup entry
339 * @name: name of the device
340 * @map: configured settings for the device
342 * Adds new setup entry to the dev_boot_setup list. The function
343 * returns 0 on error and 1 on success. This is a generic routine to
346 static int netdev_boot_setup_add(char *name, struct ifmap *map)
348 struct netdev_boot_setup *s;
352 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
353 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
354 memset(s[i].name, 0, sizeof(s[i].name));
355 strcpy(s[i].name, name);
356 memcpy(&s[i].map, map, sizeof(s[i].map));
361 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
365 * netdev_boot_setup_check - check boot time settings
366 * @dev: the netdevice
368 * Check boot time settings for the device.
369 * The found settings are set for the device to be used
370 * later in the device probing.
371 * Returns 0 if no settings found, 1 if they are.
373 int netdev_boot_setup_check(struct net_device *dev)
375 struct netdev_boot_setup *s = dev_boot_setup;
378 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
379 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
380 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
381 dev->irq = s[i].map.irq;
382 dev->base_addr = s[i].map.base_addr;
383 dev->mem_start = s[i].map.mem_start;
384 dev->mem_end = s[i].map.mem_end;
393 * netdev_boot_base - get address from boot time settings
394 * @prefix: prefix for network device
395 * @unit: id for network device
397 * Check boot time settings for the base address of device.
398 * The found settings are set for the device to be used
399 * later in the device probing.
400 * Returns 0 if no settings found.
402 unsigned long netdev_boot_base(const char *prefix, int unit)
404 const struct netdev_boot_setup *s = dev_boot_setup;
408 sprintf(name, "%s%d", prefix, unit);
411 * If device already registered then return base of 1
412 * to indicate not to probe for this interface
414 if (__dev_get_by_name(name))
417 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
418 if (!strcmp(name, s[i].name))
419 return s[i].map.base_addr;
424 * Saves at boot time configured settings for any netdevice.
426 int __init netdev_boot_setup(char *str)
431 str = get_options(str, ARRAY_SIZE(ints), ints);
436 memset(&map, 0, sizeof(map));
440 map.base_addr = ints[2];
442 map.mem_start = ints[3];
444 map.mem_end = ints[4];
446 /* Add new entry to the list */
447 return netdev_boot_setup_add(str, &map);
450 __setup("netdev=", netdev_boot_setup);
452 /*******************************************************************************
454 Device Interface Subroutines
456 *******************************************************************************/
459 * __dev_get_by_name - find a device by its name
460 * @name: name to find
462 * Find an interface by name. Must be called under RTNL semaphore
463 * or @dev_base_lock. If the name is found a pointer to the device
464 * is returned. If the name is not found then %NULL is returned. The
465 * reference counters are not incremented so the caller must be
466 * careful with locks.
469 struct net_device *__dev_get_by_name(const char *name)
471 struct hlist_node *p;
473 hlist_for_each(p, dev_name_hash(name)) {
474 struct net_device *dev
475 = hlist_entry(p, struct net_device, name_hlist);
476 if (!strncmp(dev->name, name, IFNAMSIZ))
483 * dev_get_by_name - find a device by its name
484 * @name: name to find
486 * Find an interface by name. This can be called from any
487 * context and does its own locking. The returned handle has
488 * the usage count incremented and the caller must use dev_put() to
489 * release it when it is no longer needed. %NULL is returned if no
490 * matching device is found.
493 struct net_device *dev_get_by_name(const char *name)
495 struct net_device *dev;
497 read_lock(&dev_base_lock);
498 dev = __dev_get_by_name(name);
501 read_unlock(&dev_base_lock);
506 * __dev_get_by_index - find a device by its ifindex
507 * @ifindex: index of device
509 * Search for an interface by index. Returns %NULL if the device
510 * is not found or a pointer to the device. The device has not
511 * had its reference counter increased so the caller must be careful
512 * about locking. The caller must hold either the RTNL semaphore
516 struct net_device *__dev_get_by_index(int ifindex)
518 struct hlist_node *p;
520 hlist_for_each(p, dev_index_hash(ifindex)) {
521 struct net_device *dev
522 = hlist_entry(p, struct net_device, index_hlist);
523 if (dev->ifindex == ifindex)
531 * dev_get_by_index - find a device by its ifindex
532 * @ifindex: index of device
534 * Search for an interface by index. Returns NULL if the device
535 * is not found or a pointer to the device. The device returned has
536 * had a reference added and the pointer is safe until the user calls
537 * dev_put to indicate they have finished with it.
540 struct net_device *dev_get_by_index(int ifindex)
542 struct net_device *dev;
544 read_lock(&dev_base_lock);
545 dev = __dev_get_by_index(ifindex);
548 read_unlock(&dev_base_lock);
553 * dev_getbyhwaddr - find a device by its hardware address
554 * @type: media type of device
555 * @ha: hardware address
557 * Search for an interface by MAC address. Returns NULL if the device
558 * is not found or a pointer to the device. The caller must hold the
559 * rtnl semaphore. The returned device has not had its ref count increased
560 * and the caller must therefore be careful about locking
563 * If the API was consistent this would be __dev_get_by_hwaddr
566 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
568 struct net_device *dev;
572 for (dev = dev_base; dev; dev = dev->next)
573 if (dev->type == type &&
574 !memcmp(dev->dev_addr, ha, dev->addr_len))
579 EXPORT_SYMBOL(dev_getbyhwaddr);
581 struct net_device *dev_getfirstbyhwtype(unsigned short type)
583 struct net_device *dev;
586 for (dev = dev_base; dev; dev = dev->next) {
587 if (dev->type == type) {
596 EXPORT_SYMBOL(dev_getfirstbyhwtype);
599 * dev_get_by_flags - find any device with given flags
600 * @if_flags: IFF_* values
601 * @mask: bitmask of bits in if_flags to check
603 * Search for any interface with the given flags. Returns NULL if a device
604 * is not found or a pointer to the device. The device returned has
605 * had a reference added and the pointer is safe until the user calls
606 * dev_put to indicate they have finished with it.
609 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
611 struct net_device *dev;
613 read_lock(&dev_base_lock);
614 for (dev = dev_base; dev != NULL; dev = dev->next) {
615 if (((dev->flags ^ if_flags) & mask) == 0) {
620 read_unlock(&dev_base_lock);
625 * dev_valid_name - check if name is okay for network device
628 * Network device names need to be valid file names to
629 * to allow sysfs to work
631 int dev_valid_name(const char *name)
633 return !(*name == '\0'
634 || !strcmp(name, ".")
635 || !strcmp(name, "..")
636 || strchr(name, '/'));
640 * dev_alloc_name - allocate a name for a device
642 * @name: name format string
644 * Passed a format string - eg "lt%d" it will try and find a suitable
645 * id. Not efficient for many devices, not called a lot. The caller
646 * must hold the dev_base or rtnl lock while allocating the name and
647 * adding the device in order to avoid duplicates. Returns the number
648 * of the unit assigned or a negative errno code.
651 int dev_alloc_name(struct net_device *dev, const char *name)
656 const int max_netdevices = 8*PAGE_SIZE;
658 struct net_device *d;
660 p = strnchr(name, IFNAMSIZ-1, '%');
663 * Verify the string as this thing may have come from
664 * the user. There must be either one "%d" and no other "%"
667 if (p[1] != 'd' || strchr(p + 2, '%'))
670 /* Use one page as a bit array of possible slots */
671 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
675 for (d = dev_base; d; d = d->next) {
676 if (!sscanf(d->name, name, &i))
678 if (i < 0 || i >= max_netdevices)
681 /* avoid cases where sscanf is not exact inverse of printf */
682 snprintf(buf, sizeof(buf), name, i);
683 if (!strncmp(buf, d->name, IFNAMSIZ))
687 i = find_first_zero_bit(inuse, max_netdevices);
688 free_page((unsigned long) inuse);
691 snprintf(buf, sizeof(buf), name, i);
692 if (!__dev_get_by_name(buf)) {
693 strlcpy(dev->name, buf, IFNAMSIZ);
697 /* It is possible to run out of possible slots
698 * when the name is long and there isn't enough space left
699 * for the digits, or if all bits are used.
706 * dev_change_name - change name of a device
708 * @newname: name (or format string) must be at least IFNAMSIZ
710 * Change name of a device, can pass format strings "eth%d".
713 int dev_change_name(struct net_device *dev, char *newname)
719 if (dev->flags & IFF_UP)
722 if (!dev_valid_name(newname))
725 if (strchr(newname, '%')) {
726 err = dev_alloc_name(dev, newname);
729 strcpy(newname, dev->name);
731 else if (__dev_get_by_name(newname))
734 strlcpy(dev->name, newname, IFNAMSIZ);
736 err = class_device_rename(&dev->class_dev, dev->name);
738 hlist_del(&dev->name_hlist);
739 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
740 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
747 * netdev_features_change - device changes fatures
748 * @dev: device to cause notification
750 * Called to indicate a device has changed features.
752 void netdev_features_change(struct net_device *dev)
754 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
756 EXPORT_SYMBOL(netdev_features_change);
759 * netdev_state_change - device changes state
760 * @dev: device to cause notification
762 * Called to indicate a device has changed state. This function calls
763 * the notifier chains for netdev_chain and sends a NEWLINK message
764 * to the routing socket.
766 void netdev_state_change(struct net_device *dev)
768 if (dev->flags & IFF_UP) {
769 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
770 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
775 * dev_load - load a network module
776 * @name: name of interface
778 * If a network interface is not present and the process has suitable
779 * privileges this function loads the module. If module loading is not
780 * available in this kernel then it becomes a nop.
783 void dev_load(const char *name)
785 struct net_device *dev;
787 read_lock(&dev_base_lock);
788 dev = __dev_get_by_name(name);
789 read_unlock(&dev_base_lock);
791 if (!dev && capable(CAP_SYS_MODULE))
792 request_module("%s", name);
795 static int default_rebuild_header(struct sk_buff *skb)
797 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
798 skb->dev ? skb->dev->name : "NULL!!!");
805 * dev_open - prepare an interface for use.
806 * @dev: device to open
808 * Takes a device from down to up state. The device's private open
809 * function is invoked and then the multicast lists are loaded. Finally
810 * the device is moved into the up state and a %NETDEV_UP message is
811 * sent to the netdev notifier chain.
813 * Calling this function on an active interface is a nop. On a failure
814 * a negative errno code is returned.
816 int dev_open(struct net_device *dev)
824 if (dev->flags & IFF_UP)
828 * Is it even present?
830 if (!netif_device_present(dev))
834 * Call device private open method
836 set_bit(__LINK_STATE_START, &dev->state);
838 ret = dev->open(dev);
840 clear_bit(__LINK_STATE_START, &dev->state);
844 * If it went open OK then:
851 dev->flags |= IFF_UP;
854 * Initialize multicasting status
859 * Wakeup transmit queue engine
864 * ... and announce new interface.
866 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
872 * dev_close - shutdown an interface.
873 * @dev: device to shutdown
875 * This function moves an active device into down state. A
876 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
877 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
880 int dev_close(struct net_device *dev)
882 if (!(dev->flags & IFF_UP))
886 * Tell people we are going down, so that they can
887 * prepare to death, when device is still operating.
889 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
893 clear_bit(__LINK_STATE_START, &dev->state);
895 /* Synchronize to scheduled poll. We cannot touch poll list,
896 * it can be even on different cpu. So just clear netif_running(),
897 * and wait when poll really will happen. Actually, the best place
898 * for this is inside dev->stop() after device stopped its irq
899 * engine, but this requires more changes in devices. */
901 smp_mb__after_clear_bit(); /* Commit netif_running(). */
902 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
908 * Call the device specific close. This cannot fail.
909 * Only if device is UP
911 * We allow it to be called even after a DETACH hot-plug
918 * Device is now down.
921 dev->flags &= ~IFF_UP;
924 * Tell people we are down
926 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
933 * Device change register/unregister. These are not inline or static
934 * as we export them to the world.
938 * register_netdevice_notifier - register a network notifier block
941 * Register a notifier to be called when network device events occur.
942 * The notifier passed is linked into the kernel structures and must
943 * not be reused until it has been unregistered. A negative errno code
944 * is returned on a failure.
946 * When registered all registration and up events are replayed
947 * to the new notifier to allow device to have a race free
948 * view of the network device list.
951 int register_netdevice_notifier(struct notifier_block *nb)
953 struct net_device *dev;
957 err = notifier_chain_register(&netdev_chain, nb);
959 for (dev = dev_base; dev; dev = dev->next) {
960 nb->notifier_call(nb, NETDEV_REGISTER, dev);
962 if (dev->flags & IFF_UP)
963 nb->notifier_call(nb, NETDEV_UP, dev);
971 * unregister_netdevice_notifier - unregister a network notifier block
974 * Unregister a notifier previously registered by
975 * register_netdevice_notifier(). The notifier is unlinked into the
976 * kernel structures and may then be reused. A negative errno code
977 * is returned on a failure.
980 int unregister_netdevice_notifier(struct notifier_block *nb)
982 return notifier_chain_unregister(&netdev_chain, nb);
986 * call_netdevice_notifiers - call all network notifier blocks
987 * @val: value passed unmodified to notifier function
988 * @v: pointer passed unmodified to notifier function
990 * Call all network notifier blocks. Parameters and return value
991 * are as for notifier_call_chain().
994 int call_netdevice_notifiers(unsigned long val, void *v)
996 return notifier_call_chain(&netdev_chain, val, v);
999 /* When > 0 there are consumers of rx skb time stamps */
1000 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1002 void net_enable_timestamp(void)
1004 atomic_inc(&netstamp_needed);
1007 void net_disable_timestamp(void)
1009 atomic_dec(&netstamp_needed);
1012 void __net_timestamp(struct sk_buff *skb)
1016 do_gettimeofday(&tv);
1017 skb_set_timestamp(skb, &tv);
1019 EXPORT_SYMBOL(__net_timestamp);
1021 static inline void net_timestamp(struct sk_buff *skb)
1023 if (atomic_read(&netstamp_needed))
1024 __net_timestamp(skb);
1026 skb->tstamp.off_sec = 0;
1027 skb->tstamp.off_usec = 0;
1032 * Support routine. Sends outgoing frames to any network
1033 * taps currently in use.
1036 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1038 struct packet_type *ptype;
1043 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1044 /* Never send packets back to the socket
1045 * they originated from - MvS (miquels@drinkel.ow.org)
1047 if ((ptype->dev == dev || !ptype->dev) &&
1048 (ptype->af_packet_priv == NULL ||
1049 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1050 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1054 /* skb->nh should be correctly
1055 set by sender, so that the second statement is
1056 just protection against buggy protocols.
1058 skb2->mac.raw = skb2->data;
1060 if (skb2->nh.raw < skb2->data ||
1061 skb2->nh.raw > skb2->tail) {
1062 if (net_ratelimit())
1063 printk(KERN_CRIT "protocol %04x is "
1065 skb2->protocol, dev->name);
1066 skb2->nh.raw = skb2->data;
1069 skb2->h.raw = skb2->nh.raw;
1070 skb2->pkt_type = PACKET_OUTGOING;
1071 ptype->func(skb2, skb->dev, ptype, skb->dev);
1078 * Invalidate hardware checksum when packet is to be mangled, and
1079 * complete checksum manually on outgoing path.
1081 int skb_checksum_help(struct sk_buff *skb, int inward)
1084 int ret = 0, offset = skb->h.raw - skb->data;
1087 skb->ip_summed = CHECKSUM_NONE;
1091 if (skb_cloned(skb)) {
1092 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1097 BUG_ON(offset > (int)skb->len);
1098 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1100 offset = skb->tail - skb->h.raw;
1101 BUG_ON(offset <= 0);
1102 BUG_ON(skb->csum + 2 > offset);
1104 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1105 skb->ip_summed = CHECKSUM_NONE;
1110 /* Take action when hardware reception checksum errors are detected. */
1112 void netdev_rx_csum_fault(struct net_device *dev)
1114 if (net_ratelimit()) {
1115 printk(KERN_ERR "%s: hw csum failure.\n",
1116 dev ? dev->name : "<unknown>");
1120 EXPORT_SYMBOL(netdev_rx_csum_fault);
1123 #ifdef CONFIG_HIGHMEM
1124 /* Actually, we should eliminate this check as soon as we know, that:
1125 * 1. IOMMU is present and allows to map all the memory.
1126 * 2. No high memory really exists on this machine.
1129 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1133 if (dev->features & NETIF_F_HIGHDMA)
1136 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1137 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1143 #define illegal_highdma(dev, skb) (0)
1146 /* Keep head the same: replace data */
1147 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1152 struct skb_shared_info *ninfo;
1153 int headerlen = skb->data - skb->head;
1154 int expand = (skb->tail + skb->data_len) - skb->end;
1156 if (skb_shared(skb))
1162 size = skb->end - skb->head + expand;
1163 size = SKB_DATA_ALIGN(size);
1164 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1168 /* Copy entire thing */
1169 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1173 ninfo = (struct skb_shared_info*)(data + size);
1174 atomic_set(&ninfo->dataref, 1);
1175 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1176 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1177 ninfo->ufo_size = skb_shinfo(skb)->ufo_size;
1178 ninfo->nr_frags = 0;
1179 ninfo->frag_list = NULL;
1181 /* Offset between the two in bytes */
1182 offset = data - skb->head;
1184 /* Free old data. */
1185 skb_release_data(skb);
1188 skb->end = data + size;
1190 /* Set up new pointers */
1191 skb->h.raw += offset;
1192 skb->nh.raw += offset;
1193 skb->mac.raw += offset;
1194 skb->tail += offset;
1195 skb->data += offset;
1197 /* We are no longer a clone, even if we were. */
1200 skb->tail += skb->data_len;
1205 #define HARD_TX_LOCK(dev, cpu) { \
1206 if ((dev->features & NETIF_F_LLTX) == 0) { \
1207 spin_lock(&dev->xmit_lock); \
1208 dev->xmit_lock_owner = cpu; \
1212 #define HARD_TX_UNLOCK(dev) { \
1213 if ((dev->features & NETIF_F_LLTX) == 0) { \
1214 dev->xmit_lock_owner = -1; \
1215 spin_unlock(&dev->xmit_lock); \
1220 * dev_queue_xmit - transmit a buffer
1221 * @skb: buffer to transmit
1223 * Queue a buffer for transmission to a network device. The caller must
1224 * have set the device and priority and built the buffer before calling
1225 * this function. The function can be called from an interrupt.
1227 * A negative errno code is returned on a failure. A success does not
1228 * guarantee the frame will be transmitted as it may be dropped due
1229 * to congestion or traffic shaping.
1231 * -----------------------------------------------------------------------------------
1232 * I notice this method can also return errors from the queue disciplines,
1233 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1236 * Regardless of the return value, the skb is consumed, so it is currently
1237 * difficult to retry a send to this method. (You can bump the ref count
1238 * before sending to hold a reference for retry if you are careful.)
1240 * When calling this method, interrupts MUST be enabled. This is because
1241 * the BH enable code must have IRQs enabled so that it will not deadlock.
1245 int dev_queue_xmit(struct sk_buff *skb)
1247 struct net_device *dev = skb->dev;
1251 if (skb_shinfo(skb)->frag_list &&
1252 !(dev->features & NETIF_F_FRAGLIST) &&
1253 __skb_linearize(skb, GFP_ATOMIC))
1256 /* Fragmented skb is linearized if device does not support SG,
1257 * or if at least one of fragments is in highmem and device
1258 * does not support DMA from it.
1260 if (skb_shinfo(skb)->nr_frags &&
1261 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1262 __skb_linearize(skb, GFP_ATOMIC))
1265 /* If packet is not checksummed and device does not support
1266 * checksumming for this protocol, complete checksumming here.
1268 if (skb->ip_summed == CHECKSUM_HW &&
1269 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1270 (!(dev->features & NETIF_F_IP_CSUM) ||
1271 skb->protocol != htons(ETH_P_IP))))
1272 if (skb_checksum_help(skb, 0))
1275 spin_lock_prefetch(&dev->queue_lock);
1277 /* Disable soft irqs for various locks below. Also
1278 * stops preemption for RCU.
1282 /* Updates of qdisc are serialized by queue_lock.
1283 * The struct Qdisc which is pointed to by qdisc is now a
1284 * rcu structure - it may be accessed without acquiring
1285 * a lock (but the structure may be stale.) The freeing of the
1286 * qdisc will be deferred until it's known that there are no
1287 * more references to it.
1289 * If the qdisc has an enqueue function, we still need to
1290 * hold the queue_lock before calling it, since queue_lock
1291 * also serializes access to the device queue.
1294 q = rcu_dereference(dev->qdisc);
1295 #ifdef CONFIG_NET_CLS_ACT
1296 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1299 /* Grab device queue */
1300 spin_lock(&dev->queue_lock);
1303 rc = q->enqueue(skb, q);
1305 spin_unlock(&dev->queue_lock);
1307 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1310 spin_unlock(&dev->queue_lock);
1313 /* The device has no queue. Common case for software devices:
1314 loopback, all the sorts of tunnels...
1316 Really, it is unlikely that xmit_lock protection is necessary here.
1317 (f.e. loopback and IP tunnels are clean ignoring statistics
1319 However, it is possible, that they rely on protection
1322 Check this and shot the lock. It is not prone from deadlocks.
1323 Either shot noqueue qdisc, it is even simpler 8)
1325 if (dev->flags & IFF_UP) {
1326 int cpu = smp_processor_id(); /* ok because BHs are off */
1328 if (dev->xmit_lock_owner != cpu) {
1330 HARD_TX_LOCK(dev, cpu);
1332 if (!netif_queue_stopped(dev)) {
1334 dev_queue_xmit_nit(skb, dev);
1337 if (!dev->hard_start_xmit(skb, dev)) {
1338 HARD_TX_UNLOCK(dev);
1342 HARD_TX_UNLOCK(dev);
1343 if (net_ratelimit())
1344 printk(KERN_CRIT "Virtual device %s asks to "
1345 "queue packet!\n", dev->name);
1347 /* Recursion is detected! It is possible,
1349 if (net_ratelimit())
1350 printk(KERN_CRIT "Dead loop on virtual device "
1351 "%s, fix it urgently!\n", dev->name);
1367 /*=======================================================================
1369 =======================================================================*/
1371 int netdev_max_backlog = 1000;
1372 int netdev_budget = 300;
1373 int weight_p = 64; /* old backlog weight */
1375 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1379 * netif_rx - post buffer to the network code
1380 * @skb: buffer to post
1382 * This function receives a packet from a device driver and queues it for
1383 * the upper (protocol) levels to process. It always succeeds. The buffer
1384 * may be dropped during processing for congestion control or by the
1388 * NET_RX_SUCCESS (no congestion)
1389 * NET_RX_CN_LOW (low congestion)
1390 * NET_RX_CN_MOD (moderate congestion)
1391 * NET_RX_CN_HIGH (high congestion)
1392 * NET_RX_DROP (packet was dropped)
1396 int netif_rx(struct sk_buff *skb)
1398 struct softnet_data *queue;
1399 unsigned long flags;
1401 /* if netpoll wants it, pretend we never saw it */
1402 if (netpoll_rx(skb))
1405 if (!skb->tstamp.off_sec)
1409 * The code is rearranged so that the path is the most
1410 * short when CPU is congested, but is still operating.
1412 local_irq_save(flags);
1413 queue = &__get_cpu_var(softnet_data);
1415 __get_cpu_var(netdev_rx_stat).total++;
1416 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1417 if (queue->input_pkt_queue.qlen) {
1420 __skb_queue_tail(&queue->input_pkt_queue, skb);
1421 local_irq_restore(flags);
1422 return NET_RX_SUCCESS;
1425 netif_rx_schedule(&queue->backlog_dev);
1429 __get_cpu_var(netdev_rx_stat).dropped++;
1430 local_irq_restore(flags);
1436 int netif_rx_ni(struct sk_buff *skb)
1441 err = netif_rx(skb);
1442 if (local_softirq_pending())
1449 EXPORT_SYMBOL(netif_rx_ni);
1451 static inline struct net_device *skb_bond(struct sk_buff *skb)
1453 struct net_device *dev = skb->dev;
1456 skb->dev = dev->master;
1461 static void net_tx_action(struct softirq_action *h)
1463 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1465 if (sd->completion_queue) {
1466 struct sk_buff *clist;
1468 local_irq_disable();
1469 clist = sd->completion_queue;
1470 sd->completion_queue = NULL;
1474 struct sk_buff *skb = clist;
1475 clist = clist->next;
1477 BUG_TRAP(!atomic_read(&skb->users));
1482 if (sd->output_queue) {
1483 struct net_device *head;
1485 local_irq_disable();
1486 head = sd->output_queue;
1487 sd->output_queue = NULL;
1491 struct net_device *dev = head;
1492 head = head->next_sched;
1494 smp_mb__before_clear_bit();
1495 clear_bit(__LINK_STATE_SCHED, &dev->state);
1497 if (spin_trylock(&dev->queue_lock)) {
1499 spin_unlock(&dev->queue_lock);
1501 netif_schedule(dev);
1507 static __inline__ int deliver_skb(struct sk_buff *skb,
1508 struct packet_type *pt_prev,
1509 struct net_device *orig_dev)
1511 atomic_inc(&skb->users);
1512 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1515 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1516 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1518 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1519 unsigned char *addr);
1520 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1522 static __inline__ int handle_bridge(struct sk_buff **pskb,
1523 struct packet_type **pt_prev, int *ret,
1524 struct net_device *orig_dev)
1526 struct net_bridge_port *port;
1528 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1529 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1533 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1537 return br_handle_frame_hook(port, pskb);
1540 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1543 #ifdef CONFIG_NET_CLS_ACT
1544 /* TODO: Maybe we should just force sch_ingress to be compiled in
1545 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1546 * a compare and 2 stores extra right now if we dont have it on
1547 * but have CONFIG_NET_CLS_ACT
1548 * NOTE: This doesnt stop any functionality; if you dont have
1549 * the ingress scheduler, you just cant add policies on ingress.
1552 static int ing_filter(struct sk_buff *skb)
1555 struct net_device *dev = skb->dev;
1556 int result = TC_ACT_OK;
1558 if (dev->qdisc_ingress) {
1559 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1560 if (MAX_RED_LOOP < ttl++) {
1561 printk("Redir loop detected Dropping packet (%s->%s)\n",
1562 skb->input_dev->name, skb->dev->name);
1566 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1568 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1570 spin_lock(&dev->ingress_lock);
1571 if ((q = dev->qdisc_ingress) != NULL)
1572 result = q->enqueue(skb, q);
1573 spin_unlock(&dev->ingress_lock);
1581 int netif_receive_skb(struct sk_buff *skb)
1583 struct packet_type *ptype, *pt_prev;
1584 struct net_device *orig_dev;
1585 int ret = NET_RX_DROP;
1586 unsigned short type;
1588 /* if we've gotten here through NAPI, check netpoll */
1589 if (skb->dev->poll && netpoll_rx(skb))
1592 if (!skb->tstamp.off_sec)
1595 if (!skb->input_dev)
1596 skb->input_dev = skb->dev;
1598 orig_dev = skb_bond(skb);
1600 __get_cpu_var(netdev_rx_stat).total++;
1602 skb->h.raw = skb->nh.raw = skb->data;
1603 skb->mac_len = skb->nh.raw - skb->mac.raw;
1609 #ifdef CONFIG_NET_CLS_ACT
1610 if (skb->tc_verd & TC_NCLS) {
1611 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1616 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1617 if (!ptype->dev || ptype->dev == skb->dev) {
1619 ret = deliver_skb(skb, pt_prev, orig_dev);
1624 #ifdef CONFIG_NET_CLS_ACT
1626 ret = deliver_skb(skb, pt_prev, orig_dev);
1627 pt_prev = NULL; /* noone else should process this after*/
1629 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1632 ret = ing_filter(skb);
1634 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1643 handle_diverter(skb);
1645 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1648 type = skb->protocol;
1649 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1650 if (ptype->type == type &&
1651 (!ptype->dev || ptype->dev == skb->dev)) {
1653 ret = deliver_skb(skb, pt_prev, orig_dev);
1659 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1662 /* Jamal, now you will not able to escape explaining
1663 * me how you were going to use this. :-)
1673 static int process_backlog(struct net_device *backlog_dev, int *budget)
1676 int quota = min(backlog_dev->quota, *budget);
1677 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1678 unsigned long start_time = jiffies;
1680 backlog_dev->weight = weight_p;
1682 struct sk_buff *skb;
1683 struct net_device *dev;
1685 local_irq_disable();
1686 skb = __skb_dequeue(&queue->input_pkt_queue);
1693 netif_receive_skb(skb);
1699 if (work >= quota || jiffies - start_time > 1)
1704 backlog_dev->quota -= work;
1709 backlog_dev->quota -= work;
1712 list_del(&backlog_dev->poll_list);
1713 smp_mb__before_clear_bit();
1714 netif_poll_enable(backlog_dev);
1720 static void net_rx_action(struct softirq_action *h)
1722 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1723 unsigned long start_time = jiffies;
1724 int budget = netdev_budget;
1727 local_irq_disable();
1729 while (!list_empty(&queue->poll_list)) {
1730 struct net_device *dev;
1732 if (budget <= 0 || jiffies - start_time > 1)
1737 dev = list_entry(queue->poll_list.next,
1738 struct net_device, poll_list);
1739 have = netpoll_poll_lock(dev);
1741 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1742 netpoll_poll_unlock(have);
1743 local_irq_disable();
1744 list_del(&dev->poll_list);
1745 list_add_tail(&dev->poll_list, &queue->poll_list);
1747 dev->quota += dev->weight;
1749 dev->quota = dev->weight;
1751 netpoll_poll_unlock(have);
1753 local_irq_disable();
1761 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1762 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1766 static gifconf_func_t * gifconf_list [NPROTO];
1769 * register_gifconf - register a SIOCGIF handler
1770 * @family: Address family
1771 * @gifconf: Function handler
1773 * Register protocol dependent address dumping routines. The handler
1774 * that is passed must not be freed or reused until it has been replaced
1775 * by another handler.
1777 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1779 if (family >= NPROTO)
1781 gifconf_list[family] = gifconf;
1787 * Map an interface index to its name (SIOCGIFNAME)
1791 * We need this ioctl for efficient implementation of the
1792 * if_indextoname() function required by the IPv6 API. Without
1793 * it, we would have to search all the interfaces to find a
1797 static int dev_ifname(struct ifreq __user *arg)
1799 struct net_device *dev;
1803 * Fetch the caller's info block.
1806 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1809 read_lock(&dev_base_lock);
1810 dev = __dev_get_by_index(ifr.ifr_ifindex);
1812 read_unlock(&dev_base_lock);
1816 strcpy(ifr.ifr_name, dev->name);
1817 read_unlock(&dev_base_lock);
1819 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1825 * Perform a SIOCGIFCONF call. This structure will change
1826 * size eventually, and there is nothing I can do about it.
1827 * Thus we will need a 'compatibility mode'.
1830 static int dev_ifconf(char __user *arg)
1833 struct net_device *dev;
1840 * Fetch the caller's info block.
1843 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1850 * Loop over the interfaces, and write an info block for each.
1854 for (dev = dev_base; dev; dev = dev->next) {
1855 if (vx_flags(VXF_HIDE_NETIF, 0) &&
1856 !dev_in_nx_info(dev, current->nx_info))
1858 for (i = 0; i < NPROTO; i++) {
1859 if (gifconf_list[i]) {
1862 done = gifconf_list[i](dev, NULL, 0);
1864 done = gifconf_list[i](dev, pos + total,
1874 * All done. Write the updated control block back to the caller.
1876 ifc.ifc_len = total;
1879 * Both BSD and Solaris return 0 here, so we do too.
1881 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1884 #ifdef CONFIG_PROC_FS
1886 * This is invoked by the /proc filesystem handler to display a device
1889 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1891 struct net_device *dev;
1894 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1896 return i == pos ? dev : NULL;
1899 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1901 read_lock(&dev_base_lock);
1902 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1905 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1908 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1911 void dev_seq_stop(struct seq_file *seq, void *v)
1913 read_unlock(&dev_base_lock);
1916 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1918 struct nx_info *nxi = current->nx_info;
1920 if (vx_flags(VXF_HIDE_NETIF, 0) && !dev_in_nx_info(dev, nxi))
1922 if (dev->get_stats) {
1923 struct net_device_stats *stats = dev->get_stats(dev);
1925 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1926 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1927 dev->name, stats->rx_bytes, stats->rx_packets,
1929 stats->rx_dropped + stats->rx_missed_errors,
1930 stats->rx_fifo_errors,
1931 stats->rx_length_errors + stats->rx_over_errors +
1932 stats->rx_crc_errors + stats->rx_frame_errors,
1933 stats->rx_compressed, stats->multicast,
1934 stats->tx_bytes, stats->tx_packets,
1935 stats->tx_errors, stats->tx_dropped,
1936 stats->tx_fifo_errors, stats->collisions,
1937 stats->tx_carrier_errors +
1938 stats->tx_aborted_errors +
1939 stats->tx_window_errors +
1940 stats->tx_heartbeat_errors,
1941 stats->tx_compressed);
1943 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1947 * Called from the PROCfs module. This now uses the new arbitrary sized
1948 * /proc/net interface to create /proc/net/dev
1950 static int dev_seq_show(struct seq_file *seq, void *v)
1952 if (v == SEQ_START_TOKEN)
1953 seq_puts(seq, "Inter-| Receive "
1955 " face |bytes packets errs drop fifo frame "
1956 "compressed multicast|bytes packets errs "
1957 "drop fifo colls carrier compressed\n");
1959 dev_seq_printf_stats(seq, v);
1963 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1965 struct netif_rx_stats *rc = NULL;
1967 while (*pos < NR_CPUS)
1968 if (cpu_online(*pos)) {
1969 rc = &per_cpu(netdev_rx_stat, *pos);
1976 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1978 return softnet_get_online(pos);
1981 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1984 return softnet_get_online(pos);
1987 static void softnet_seq_stop(struct seq_file *seq, void *v)
1991 static int softnet_seq_show(struct seq_file *seq, void *v)
1993 struct netif_rx_stats *s = v;
1995 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
1996 s->total, s->dropped, s->time_squeeze, 0,
1997 0, 0, 0, 0, /* was fastroute */
2002 static struct seq_operations dev_seq_ops = {
2003 .start = dev_seq_start,
2004 .next = dev_seq_next,
2005 .stop = dev_seq_stop,
2006 .show = dev_seq_show,
2009 static int dev_seq_open(struct inode *inode, struct file *file)
2011 return seq_open(file, &dev_seq_ops);
2014 static struct file_operations dev_seq_fops = {
2015 .owner = THIS_MODULE,
2016 .open = dev_seq_open,
2018 .llseek = seq_lseek,
2019 .release = seq_release,
2022 static struct seq_operations softnet_seq_ops = {
2023 .start = softnet_seq_start,
2024 .next = softnet_seq_next,
2025 .stop = softnet_seq_stop,
2026 .show = softnet_seq_show,
2029 static int softnet_seq_open(struct inode *inode, struct file *file)
2031 return seq_open(file, &softnet_seq_ops);
2034 static struct file_operations softnet_seq_fops = {
2035 .owner = THIS_MODULE,
2036 .open = softnet_seq_open,
2038 .llseek = seq_lseek,
2039 .release = seq_release,
2043 extern int wireless_proc_init(void);
2045 #define wireless_proc_init() 0
2048 static int __init dev_proc_init(void)
2052 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2054 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2056 if (wireless_proc_init())
2062 proc_net_remove("softnet_stat");
2064 proc_net_remove("dev");
2068 #define dev_proc_init() 0
2069 #endif /* CONFIG_PROC_FS */
2073 * netdev_set_master - set up master/slave pair
2074 * @slave: slave device
2075 * @master: new master device
2077 * Changes the master device of the slave. Pass %NULL to break the
2078 * bonding. The caller must hold the RTNL semaphore. On a failure
2079 * a negative errno code is returned. On success the reference counts
2080 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2081 * function returns zero.
2083 int netdev_set_master(struct net_device *slave, struct net_device *master)
2085 struct net_device *old = slave->master;
2095 slave->master = master;
2103 slave->flags |= IFF_SLAVE;
2105 slave->flags &= ~IFF_SLAVE;
2107 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2112 * dev_set_promiscuity - update promiscuity count on a device
2116 * Add or remove promsicuity from a device. While the count in the device
2117 * remains above zero the interface remains promiscuous. Once it hits zero
2118 * the device reverts back to normal filtering operation. A negative inc
2119 * value is used to drop promiscuity on the device.
2121 void dev_set_promiscuity(struct net_device *dev, int inc)
2123 unsigned short old_flags = dev->flags;
2125 if ((dev->promiscuity += inc) == 0)
2126 dev->flags &= ~IFF_PROMISC;
2128 dev->flags |= IFF_PROMISC;
2129 if (dev->flags != old_flags) {
2131 printk(KERN_INFO "device %s %s promiscuous mode\n",
2132 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2138 * dev_set_allmulti - update allmulti count on a device
2142 * Add or remove reception of all multicast frames to a device. While the
2143 * count in the device remains above zero the interface remains listening
2144 * to all interfaces. Once it hits zero the device reverts back to normal
2145 * filtering operation. A negative @inc value is used to drop the counter
2146 * when releasing a resource needing all multicasts.
2149 void dev_set_allmulti(struct net_device *dev, int inc)
2151 unsigned short old_flags = dev->flags;
2153 dev->flags |= IFF_ALLMULTI;
2154 if ((dev->allmulti += inc) == 0)
2155 dev->flags &= ~IFF_ALLMULTI;
2156 if (dev->flags ^ old_flags)
2160 unsigned dev_get_flags(const struct net_device *dev)
2164 flags = (dev->flags & ~(IFF_PROMISC |
2167 (dev->gflags & (IFF_PROMISC |
2170 if (netif_running(dev) && netif_carrier_ok(dev))
2171 flags |= IFF_RUNNING;
2176 int dev_change_flags(struct net_device *dev, unsigned flags)
2179 int old_flags = dev->flags;
2182 * Set the flags on our device.
2185 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2186 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2188 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2192 * Load in the correct multicast list now the flags have changed.
2198 * Have we downed the interface. We handle IFF_UP ourselves
2199 * according to user attempts to set it, rather than blindly
2204 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2205 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2211 if (dev->flags & IFF_UP &&
2212 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2214 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2216 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2217 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2218 dev->gflags ^= IFF_PROMISC;
2219 dev_set_promiscuity(dev, inc);
2222 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2223 is important. Some (broken) drivers set IFF_PROMISC, when
2224 IFF_ALLMULTI is requested not asking us and not reporting.
2226 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2227 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2228 dev->gflags ^= IFF_ALLMULTI;
2229 dev_set_allmulti(dev, inc);
2232 if (old_flags ^ dev->flags)
2233 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2238 int dev_set_mtu(struct net_device *dev, int new_mtu)
2242 if (new_mtu == dev->mtu)
2245 /* MTU must be positive. */
2249 if (!netif_device_present(dev))
2253 if (dev->change_mtu)
2254 err = dev->change_mtu(dev, new_mtu);
2257 if (!err && dev->flags & IFF_UP)
2258 notifier_call_chain(&netdev_chain,
2259 NETDEV_CHANGEMTU, dev);
2263 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2267 if (!dev->set_mac_address)
2269 if (sa->sa_family != dev->type)
2271 if (!netif_device_present(dev))
2273 err = dev->set_mac_address(dev, sa);
2275 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2280 * Perform the SIOCxIFxxx calls.
2282 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2285 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2291 case SIOCGIFFLAGS: /* Get interface flags */
2292 ifr->ifr_flags = dev_get_flags(dev);
2295 case SIOCSIFFLAGS: /* Set interface flags */
2296 return dev_change_flags(dev, ifr->ifr_flags);
2298 case SIOCGIFMETRIC: /* Get the metric on the interface
2299 (currently unused) */
2300 ifr->ifr_metric = 0;
2303 case SIOCSIFMETRIC: /* Set the metric on the interface
2304 (currently unused) */
2307 case SIOCGIFMTU: /* Get the MTU of a device */
2308 ifr->ifr_mtu = dev->mtu;
2311 case SIOCSIFMTU: /* Set the MTU of a device */
2312 return dev_set_mtu(dev, ifr->ifr_mtu);
2316 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2318 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2319 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2320 ifr->ifr_hwaddr.sa_family = dev->type;
2324 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2326 case SIOCSIFHWBROADCAST:
2327 if (ifr->ifr_hwaddr.sa_family != dev->type)
2329 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2330 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2331 notifier_call_chain(&netdev_chain,
2332 NETDEV_CHANGEADDR, dev);
2336 ifr->ifr_map.mem_start = dev->mem_start;
2337 ifr->ifr_map.mem_end = dev->mem_end;
2338 ifr->ifr_map.base_addr = dev->base_addr;
2339 ifr->ifr_map.irq = dev->irq;
2340 ifr->ifr_map.dma = dev->dma;
2341 ifr->ifr_map.port = dev->if_port;
2345 if (dev->set_config) {
2346 if (!netif_device_present(dev))
2348 return dev->set_config(dev, &ifr->ifr_map);
2353 if (!dev->set_multicast_list ||
2354 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2356 if (!netif_device_present(dev))
2358 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2362 if (!dev->set_multicast_list ||
2363 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2365 if (!netif_device_present(dev))
2367 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2371 ifr->ifr_ifindex = dev->ifindex;
2375 ifr->ifr_qlen = dev->tx_queue_len;
2379 if (ifr->ifr_qlen < 0)
2381 dev->tx_queue_len = ifr->ifr_qlen;
2385 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2386 return dev_change_name(dev, ifr->ifr_newname);
2389 * Unknown or private ioctl
2393 if ((cmd >= SIOCDEVPRIVATE &&
2394 cmd <= SIOCDEVPRIVATE + 15) ||
2395 cmd == SIOCBONDENSLAVE ||
2396 cmd == SIOCBONDRELEASE ||
2397 cmd == SIOCBONDSETHWADDR ||
2398 cmd == SIOCBONDSLAVEINFOQUERY ||
2399 cmd == SIOCBONDINFOQUERY ||
2400 cmd == SIOCBONDCHANGEACTIVE ||
2401 cmd == SIOCGMIIPHY ||
2402 cmd == SIOCGMIIREG ||
2403 cmd == SIOCSMIIREG ||
2404 cmd == SIOCBRADDIF ||
2405 cmd == SIOCBRDELIF ||
2406 cmd == SIOCWANDEV) {
2408 if (dev->do_ioctl) {
2409 if (netif_device_present(dev))
2410 err = dev->do_ioctl(dev, ifr,
2423 * This function handles all "interface"-type I/O control requests. The actual
2424 * 'doing' part of this is dev_ifsioc above.
2428 * dev_ioctl - network device ioctl
2429 * @cmd: command to issue
2430 * @arg: pointer to a struct ifreq in user space
2432 * Issue ioctl functions to devices. This is normally called by the
2433 * user space syscall interfaces but can sometimes be useful for
2434 * other purposes. The return value is the return from the syscall if
2435 * positive or a negative errno code on error.
2438 int dev_ioctl(unsigned int cmd, void __user *arg)
2444 /* One special case: SIOCGIFCONF takes ifconf argument
2445 and requires shared lock, because it sleeps writing
2449 if (cmd == SIOCGIFCONF) {
2451 ret = dev_ifconf((char __user *) arg);
2455 if (cmd == SIOCGIFNAME)
2456 return dev_ifname((struct ifreq __user *)arg);
2458 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2461 ifr.ifr_name[IFNAMSIZ-1] = 0;
2463 colon = strchr(ifr.ifr_name, ':');
2468 * See which interface the caller is talking about.
2473 * These ioctl calls:
2474 * - can be done by all.
2475 * - atomic and do not require locking.
2486 dev_load(ifr.ifr_name);
2487 read_lock(&dev_base_lock);
2488 ret = dev_ifsioc(&ifr, cmd);
2489 read_unlock(&dev_base_lock);
2493 if (copy_to_user(arg, &ifr,
2494 sizeof(struct ifreq)))
2500 dev_load(ifr.ifr_name);
2502 ret = dev_ethtool(&ifr);
2507 if (copy_to_user(arg, &ifr,
2508 sizeof(struct ifreq)))
2514 * These ioctl calls:
2515 * - require superuser power.
2516 * - require strict serialization.
2522 if (!capable(CAP_NET_ADMIN))
2524 dev_load(ifr.ifr_name);
2526 ret = dev_ifsioc(&ifr, cmd);
2531 if (copy_to_user(arg, &ifr,
2532 sizeof(struct ifreq)))
2538 * These ioctl calls:
2539 * - require superuser power.
2540 * - require strict serialization.
2541 * - do not return a value
2551 case SIOCSIFHWBROADCAST:
2554 case SIOCBONDENSLAVE:
2555 case SIOCBONDRELEASE:
2556 case SIOCBONDSETHWADDR:
2557 case SIOCBONDCHANGEACTIVE:
2560 if (!capable(CAP_NET_ADMIN))
2563 case SIOCBONDSLAVEINFOQUERY:
2564 case SIOCBONDINFOQUERY:
2565 dev_load(ifr.ifr_name);
2567 ret = dev_ifsioc(&ifr, cmd);
2572 /* Get the per device memory space. We can add this but
2573 * currently do not support it */
2575 /* Set the per device memory buffer space.
2576 * Not applicable in our case */
2581 * Unknown or private ioctl.
2584 if (cmd == SIOCWANDEV ||
2585 (cmd >= SIOCDEVPRIVATE &&
2586 cmd <= SIOCDEVPRIVATE + 15)) {
2587 dev_load(ifr.ifr_name);
2589 ret = dev_ifsioc(&ifr, cmd);
2591 if (!ret && copy_to_user(arg, &ifr,
2592 sizeof(struct ifreq)))
2597 /* Take care of Wireless Extensions */
2598 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2599 /* If command is `set a parameter', or
2600 * `get the encoding parameters', check if
2601 * the user has the right to do it */
2602 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2603 if (!capable(CAP_NET_ADMIN))
2606 dev_load(ifr.ifr_name);
2608 /* Follow me in net/core/wireless.c */
2609 ret = wireless_process_ioctl(&ifr, cmd);
2611 if (IW_IS_GET(cmd) &&
2612 copy_to_user(arg, &ifr,
2613 sizeof(struct ifreq)))
2617 #endif /* WIRELESS_EXT */
2624 * dev_new_index - allocate an ifindex
2626 * Returns a suitable unique value for a new device interface
2627 * number. The caller must hold the rtnl semaphore or the
2628 * dev_base_lock to be sure it remains unique.
2630 static int dev_new_index(void)
2636 if (!__dev_get_by_index(ifindex))
2641 static int dev_boot_phase = 1;
2643 /* Delayed registration/unregisteration */
2644 static DEFINE_SPINLOCK(net_todo_list_lock);
2645 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2647 static inline void net_set_todo(struct net_device *dev)
2649 spin_lock(&net_todo_list_lock);
2650 list_add_tail(&dev->todo_list, &net_todo_list);
2651 spin_unlock(&net_todo_list_lock);
2655 * register_netdevice - register a network device
2656 * @dev: device to register
2658 * Take a completed network device structure and add it to the kernel
2659 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2660 * chain. 0 is returned on success. A negative errno code is returned
2661 * on a failure to set up the device, or if the name is a duplicate.
2663 * Callers must hold the rtnl semaphore. You may want
2664 * register_netdev() instead of this.
2667 * The locking appears insufficient to guarantee two parallel registers
2668 * will not get the same name.
2671 int register_netdevice(struct net_device *dev)
2673 struct hlist_head *head;
2674 struct hlist_node *p;
2677 BUG_ON(dev_boot_phase);
2680 /* When net_device's are persistent, this will be fatal. */
2681 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2683 spin_lock_init(&dev->queue_lock);
2684 spin_lock_init(&dev->xmit_lock);
2685 dev->xmit_lock_owner = -1;
2686 #ifdef CONFIG_NET_CLS_ACT
2687 spin_lock_init(&dev->ingress_lock);
2690 ret = alloc_divert_blk(dev);
2696 /* Init, if this function is available */
2698 ret = dev->init(dev);
2706 if (!dev_valid_name(dev->name)) {
2711 dev->ifindex = dev_new_index();
2712 if (dev->iflink == -1)
2713 dev->iflink = dev->ifindex;
2715 /* Check for existence of name */
2716 head = dev_name_hash(dev->name);
2717 hlist_for_each(p, head) {
2718 struct net_device *d
2719 = hlist_entry(p, struct net_device, name_hlist);
2720 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2726 /* Fix illegal SG+CSUM combinations. */
2727 if ((dev->features & NETIF_F_SG) &&
2728 !(dev->features & (NETIF_F_IP_CSUM |
2730 NETIF_F_HW_CSUM))) {
2731 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2733 dev->features &= ~NETIF_F_SG;
2736 /* TSO requires that SG is present as well. */
2737 if ((dev->features & NETIF_F_TSO) &&
2738 !(dev->features & NETIF_F_SG)) {
2739 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2741 dev->features &= ~NETIF_F_TSO;
2743 if (dev->features & NETIF_F_UFO) {
2744 if (!(dev->features & NETIF_F_HW_CSUM)) {
2745 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2746 "NETIF_F_HW_CSUM feature.\n",
2748 dev->features &= ~NETIF_F_UFO;
2750 if (!(dev->features & NETIF_F_SG)) {
2751 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2752 "NETIF_F_SG feature.\n",
2754 dev->features &= ~NETIF_F_UFO;
2759 * nil rebuild_header routine,
2760 * that should be never called and used as just bug trap.
2763 if (!dev->rebuild_header)
2764 dev->rebuild_header = default_rebuild_header;
2767 * Default initial state at registry is that the
2768 * device is present.
2771 set_bit(__LINK_STATE_PRESENT, &dev->state);
2774 dev_init_scheduler(dev);
2775 write_lock_bh(&dev_base_lock);
2777 dev_tail = &dev->next;
2778 hlist_add_head(&dev->name_hlist, head);
2779 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2781 dev->reg_state = NETREG_REGISTERING;
2782 write_unlock_bh(&dev_base_lock);
2784 /* Notify protocols, that a new device appeared. */
2785 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2787 /* Finish registration after unlock */
2794 free_divert_blk(dev);
2799 * register_netdev - register a network device
2800 * @dev: device to register
2802 * Take a completed network device structure and add it to the kernel
2803 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2804 * chain. 0 is returned on success. A negative errno code is returned
2805 * on a failure to set up the device, or if the name is a duplicate.
2807 * This is a wrapper around register_netdev that takes the rtnl semaphore
2808 * and expands the device name if you passed a format string to
2811 int register_netdev(struct net_device *dev)
2818 * If the name is a format string the caller wants us to do a
2821 if (strchr(dev->name, '%')) {
2822 err = dev_alloc_name(dev, dev->name);
2828 * Back compatibility hook. Kill this one in 2.5
2830 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2831 err = dev_alloc_name(dev, "eth%d");
2836 err = register_netdevice(dev);
2841 EXPORT_SYMBOL(register_netdev);
2844 * netdev_wait_allrefs - wait until all references are gone.
2846 * This is called when unregistering network devices.
2848 * Any protocol or device that holds a reference should register
2849 * for netdevice notification, and cleanup and put back the
2850 * reference if they receive an UNREGISTER event.
2851 * We can get stuck here if buggy protocols don't correctly
2854 static void netdev_wait_allrefs(struct net_device *dev)
2856 unsigned long rebroadcast_time, warning_time;
2858 rebroadcast_time = warning_time = jiffies;
2859 while (atomic_read(&dev->refcnt) != 0) {
2860 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2863 /* Rebroadcast unregister notification */
2864 notifier_call_chain(&netdev_chain,
2865 NETDEV_UNREGISTER, dev);
2867 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2869 /* We must not have linkwatch events
2870 * pending on unregister. If this
2871 * happens, we simply run the queue
2872 * unscheduled, resulting in a noop
2875 linkwatch_run_queue();
2880 rebroadcast_time = jiffies;
2885 if (time_after(jiffies, warning_time + 10 * HZ)) {
2886 printk(KERN_EMERG "unregister_netdevice: "
2887 "waiting for %s to become free. Usage "
2889 dev->name, atomic_read(&dev->refcnt));
2890 warning_time = jiffies;
2899 * register_netdevice(x1);
2900 * register_netdevice(x2);
2902 * unregister_netdevice(y1);
2903 * unregister_netdevice(y2);
2909 * We are invoked by rtnl_unlock() after it drops the semaphore.
2910 * This allows us to deal with problems:
2911 * 1) We can create/delete sysfs objects which invoke hotplug
2912 * without deadlocking with linkwatch via keventd.
2913 * 2) Since we run with the RTNL semaphore not held, we can sleep
2914 * safely in order to wait for the netdev refcnt to drop to zero.
2916 static DECLARE_MUTEX(net_todo_run_mutex);
2917 void netdev_run_todo(void)
2919 struct list_head list = LIST_HEAD_INIT(list);
2923 /* Need to guard against multiple cpu's getting out of order. */
2924 down(&net_todo_run_mutex);
2926 /* Not safe to do outside the semaphore. We must not return
2927 * until all unregister events invoked by the local processor
2928 * have been completed (either by this todo run, or one on
2931 if (list_empty(&net_todo_list))
2934 /* Snapshot list, allow later requests */
2935 spin_lock(&net_todo_list_lock);
2936 list_splice_init(&net_todo_list, &list);
2937 spin_unlock(&net_todo_list_lock);
2939 while (!list_empty(&list)) {
2940 struct net_device *dev
2941 = list_entry(list.next, struct net_device, todo_list);
2942 list_del(&dev->todo_list);
2944 switch(dev->reg_state) {
2945 case NETREG_REGISTERING:
2946 dev->reg_state = NETREG_REGISTERED;
2947 err = netdev_register_sysfs(dev);
2949 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2953 case NETREG_UNREGISTERING:
2954 netdev_unregister_sysfs(dev);
2955 dev->reg_state = NETREG_UNREGISTERED;
2957 netdev_wait_allrefs(dev);
2960 BUG_ON(atomic_read(&dev->refcnt));
2961 BUG_TRAP(!dev->ip_ptr);
2962 BUG_TRAP(!dev->ip6_ptr);
2963 BUG_TRAP(!dev->dn_ptr);
2966 /* It must be the very last action,
2967 * after this 'dev' may point to freed up memory.
2969 if (dev->destructor)
2970 dev->destructor(dev);
2974 printk(KERN_ERR "network todo '%s' but state %d\n",
2975 dev->name, dev->reg_state);
2981 up(&net_todo_run_mutex);
2985 * alloc_netdev - allocate network device
2986 * @sizeof_priv: size of private data to allocate space for
2987 * @name: device name format string
2988 * @setup: callback to initialize device
2990 * Allocates a struct net_device with private data area for driver use
2991 * and performs basic initialization.
2993 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
2994 void (*setup)(struct net_device *))
2997 struct net_device *dev;
3000 /* ensure 32-byte alignment of both the device and private area */
3001 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3002 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3004 p = kmalloc(alloc_size, GFP_KERNEL);
3006 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3009 memset(p, 0, alloc_size);
3011 dev = (struct net_device *)
3012 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3013 dev->padded = (char *)dev - (char *)p;
3016 dev->priv = netdev_priv(dev);
3019 strcpy(dev->name, name);
3022 EXPORT_SYMBOL(alloc_netdev);
3025 * free_netdev - free network device
3028 * This function does the last stage of destroying an allocated device
3029 * interface. The reference to the device object is released.
3030 * If this is the last reference then it will be freed.
3032 void free_netdev(struct net_device *dev)
3035 /* Compatiablity with error handling in drivers */
3036 if (dev->reg_state == NETREG_UNINITIALIZED) {
3037 kfree((char *)dev - dev->padded);
3041 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3042 dev->reg_state = NETREG_RELEASED;
3044 /* will free via class release */
3045 class_device_put(&dev->class_dev);
3047 kfree((char *)dev - dev->padded);
3051 /* Synchronize with packet receive processing. */
3052 void synchronize_net(void)
3059 * unregister_netdevice - remove device from the kernel
3062 * This function shuts down a device interface and removes it
3063 * from the kernel tables. On success 0 is returned, on a failure
3064 * a negative errno code is returned.
3066 * Callers must hold the rtnl semaphore. You may want
3067 * unregister_netdev() instead of this.
3070 int unregister_netdevice(struct net_device *dev)
3072 struct net_device *d, **dp;
3074 BUG_ON(dev_boot_phase);
3077 /* Some devices call without registering for initialization unwind. */
3078 if (dev->reg_state == NETREG_UNINITIALIZED) {
3079 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3080 "was registered\n", dev->name, dev);
3084 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3086 /* If device is running, close it first. */
3087 if (dev->flags & IFF_UP)
3090 /* And unlink it from device chain. */
3091 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3093 write_lock_bh(&dev_base_lock);
3094 hlist_del(&dev->name_hlist);
3095 hlist_del(&dev->index_hlist);
3096 if (dev_tail == &dev->next)
3099 write_unlock_bh(&dev_base_lock);
3104 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3109 dev->reg_state = NETREG_UNREGISTERING;
3113 /* Shutdown queueing discipline. */
3117 /* Notify protocols, that we are about to destroy
3118 this device. They should clean all the things.
3120 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3123 * Flush the multicast chain
3125 dev_mc_discard(dev);
3130 /* Notifier chain MUST detach us from master device. */
3131 BUG_TRAP(!dev->master);
3133 free_divert_blk(dev);
3135 /* Finish processing unregister after unlock */
3145 * unregister_netdev - remove device from the kernel
3148 * This function shuts down a device interface and removes it
3149 * from the kernel tables. On success 0 is returned, on a failure
3150 * a negative errno code is returned.
3152 * This is just a wrapper for unregister_netdevice that takes
3153 * the rtnl semaphore. In general you want to use this and not
3154 * unregister_netdevice.
3156 void unregister_netdev(struct net_device *dev)
3159 unregister_netdevice(dev);
3163 EXPORT_SYMBOL(unregister_netdev);
3165 #ifdef CONFIG_HOTPLUG_CPU
3166 static int dev_cpu_callback(struct notifier_block *nfb,
3167 unsigned long action,
3170 struct sk_buff **list_skb;
3171 struct net_device **list_net;
3172 struct sk_buff *skb;
3173 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3174 struct softnet_data *sd, *oldsd;
3176 if (action != CPU_DEAD)
3179 local_irq_disable();
3180 cpu = smp_processor_id();
3181 sd = &per_cpu(softnet_data, cpu);
3182 oldsd = &per_cpu(softnet_data, oldcpu);
3184 /* Find end of our completion_queue. */
3185 list_skb = &sd->completion_queue;
3187 list_skb = &(*list_skb)->next;
3188 /* Append completion queue from offline CPU. */
3189 *list_skb = oldsd->completion_queue;
3190 oldsd->completion_queue = NULL;
3192 /* Find end of our output_queue. */
3193 list_net = &sd->output_queue;
3195 list_net = &(*list_net)->next_sched;
3196 /* Append output queue from offline CPU. */
3197 *list_net = oldsd->output_queue;
3198 oldsd->output_queue = NULL;
3200 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3203 /* Process offline CPU's input_pkt_queue */
3204 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3209 #endif /* CONFIG_HOTPLUG_CPU */
3213 * Initialize the DEV module. At boot time this walks the device list and
3214 * unhooks any devices that fail to initialise (normally hardware not
3215 * present) and leaves us with a valid list of present and active devices.
3220 * This is called single threaded during boot, so no need
3221 * to take the rtnl semaphore.
3223 static int __init net_dev_init(void)
3225 int i, rc = -ENOMEM;
3227 BUG_ON(!dev_boot_phase);
3231 if (dev_proc_init())
3234 if (netdev_sysfs_init())
3237 INIT_LIST_HEAD(&ptype_all);
3238 for (i = 0; i < 16; i++)
3239 INIT_LIST_HEAD(&ptype_base[i]);
3241 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3242 INIT_HLIST_HEAD(&dev_name_head[i]);
3244 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3245 INIT_HLIST_HEAD(&dev_index_head[i]);
3248 * Initialise the packet receive queues.
3252 struct softnet_data *queue;
3254 queue = &per_cpu(softnet_data, i);
3255 skb_queue_head_init(&queue->input_pkt_queue);
3256 queue->completion_queue = NULL;
3257 INIT_LIST_HEAD(&queue->poll_list);
3258 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3259 queue->backlog_dev.weight = weight_p;
3260 queue->backlog_dev.poll = process_backlog;
3261 atomic_set(&queue->backlog_dev.refcnt, 1);
3266 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3267 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3269 hotcpu_notifier(dev_cpu_callback, 0);
3277 subsys_initcall(net_dev_init);
3279 EXPORT_SYMBOL(__dev_get_by_index);
3280 EXPORT_SYMBOL(__dev_get_by_name);
3281 EXPORT_SYMBOL(__dev_remove_pack);
3282 EXPORT_SYMBOL(__skb_linearize);
3283 EXPORT_SYMBOL(dev_valid_name);
3284 EXPORT_SYMBOL(dev_add_pack);
3285 EXPORT_SYMBOL(dev_alloc_name);
3286 EXPORT_SYMBOL(dev_close);
3287 EXPORT_SYMBOL(dev_get_by_flags);
3288 EXPORT_SYMBOL(dev_get_by_index);
3289 EXPORT_SYMBOL(dev_get_by_name);
3290 EXPORT_SYMBOL(dev_open);
3291 EXPORT_SYMBOL(dev_queue_xmit);
3292 EXPORT_SYMBOL(dev_remove_pack);
3293 EXPORT_SYMBOL(dev_set_allmulti);
3294 EXPORT_SYMBOL(dev_set_promiscuity);
3295 EXPORT_SYMBOL(dev_change_flags);
3296 EXPORT_SYMBOL(dev_set_mtu);
3297 EXPORT_SYMBOL(dev_set_mac_address);
3298 EXPORT_SYMBOL(free_netdev);
3299 EXPORT_SYMBOL(netdev_boot_setup_check);
3300 EXPORT_SYMBOL(netdev_set_master);
3301 EXPORT_SYMBOL(netdev_state_change);
3302 EXPORT_SYMBOL(netif_receive_skb);
3303 EXPORT_SYMBOL(netif_rx);
3304 EXPORT_SYMBOL(register_gifconf);
3305 EXPORT_SYMBOL(register_netdevice);
3306 EXPORT_SYMBOL(register_netdevice_notifier);
3307 EXPORT_SYMBOL(skb_checksum_help);
3308 EXPORT_SYMBOL(synchronize_net);
3309 EXPORT_SYMBOL(unregister_netdevice);
3310 EXPORT_SYMBOL(unregister_netdevice_notifier);
3311 EXPORT_SYMBOL(net_enable_timestamp);
3312 EXPORT_SYMBOL(net_disable_timestamp);
3313 EXPORT_SYMBOL(dev_get_flags);
3315 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3316 EXPORT_SYMBOL(br_handle_frame_hook);
3317 EXPORT_SYMBOL(br_fdb_get_hook);
3318 EXPORT_SYMBOL(br_fdb_put_hook);
3322 EXPORT_SYMBOL(dev_load);
3325 EXPORT_PER_CPU_SYMBOL(softnet_data);