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 #if !((defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)))
1037 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,12)
1041 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1043 struct packet_type *ptype;
1048 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1049 /* Never send packets back to the socket
1050 * they originated from - MvS (miquels@drinkel.ow.org)
1052 if ((ptype->dev == dev || !ptype->dev) &&
1053 (ptype->af_packet_priv == NULL ||
1054 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1055 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1059 /* skb->nh should be correctly
1060 set by sender, so that the second statement is
1061 just protection against buggy protocols.
1063 skb2->mac.raw = skb2->data;
1065 if (skb2->nh.raw < skb2->data ||
1066 skb2->nh.raw > skb2->tail) {
1067 if (net_ratelimit())
1068 printk(KERN_CRIT "protocol %04x is "
1070 skb2->protocol, dev->name);
1071 skb2->nh.raw = skb2->data;
1074 skb2->h.raw = skb2->nh.raw;
1075 skb2->pkt_type = PACKET_OUTGOING;
1076 ptype->func(skb2, skb->dev, ptype, skb->dev);
1083 * Invalidate hardware checksum when packet is to be mangled, and
1084 * complete checksum manually on outgoing path.
1086 int skb_checksum_help(struct sk_buff *skb, int inward)
1089 int ret = 0, offset = skb->h.raw - skb->data;
1092 skb->ip_summed = CHECKSUM_NONE;
1096 if (skb_cloned(skb)) {
1097 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1102 BUG_ON(offset > (int)skb->len);
1103 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1105 offset = skb->tail - skb->h.raw;
1106 BUG_ON(offset <= 0);
1107 BUG_ON(skb->csum + 2 > offset);
1109 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1110 skb->ip_summed = CHECKSUM_NONE;
1115 /* Take action when hardware reception checksum errors are detected. */
1117 void netdev_rx_csum_fault(struct net_device *dev)
1119 if (net_ratelimit()) {
1120 printk(KERN_ERR "%s: hw csum failure.\n",
1121 dev ? dev->name : "<unknown>");
1125 EXPORT_SYMBOL(netdev_rx_csum_fault);
1128 #ifdef CONFIG_HIGHMEM
1129 /* Actually, we should eliminate this check as soon as we know, that:
1130 * 1. IOMMU is present and allows to map all the memory.
1131 * 2. No high memory really exists on this machine.
1134 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1138 if (dev->features & NETIF_F_HIGHDMA)
1141 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1142 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1148 #define illegal_highdma(dev, skb) (0)
1151 /* Keep head the same: replace data */
1152 int __skb_linearize(struct sk_buff *skb, gfp_t gfp_mask)
1157 struct skb_shared_info *ninfo;
1158 int headerlen = skb->data - skb->head;
1159 int expand = (skb->tail + skb->data_len) - skb->end;
1161 if (skb_shared(skb))
1167 size = skb->end - skb->head + expand;
1168 size = SKB_DATA_ALIGN(size);
1169 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1173 /* Copy entire thing */
1174 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1178 ninfo = (struct skb_shared_info*)(data + size);
1179 atomic_set(&ninfo->dataref, 1);
1180 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1181 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1182 ninfo->ufo_size = skb_shinfo(skb)->ufo_size;
1183 ninfo->nr_frags = 0;
1184 ninfo->frag_list = NULL;
1186 /* Offset between the two in bytes */
1187 offset = data - skb->head;
1189 /* Free old data. */
1190 skb_release_data(skb);
1193 skb->end = data + size;
1195 /* Set up new pointers */
1196 skb->h.raw += offset;
1197 skb->nh.raw += offset;
1198 skb->mac.raw += offset;
1199 skb->tail += offset;
1200 skb->data += offset;
1202 /* We are no longer a clone, even if we were. */
1205 skb->tail += skb->data_len;
1210 #define HARD_TX_LOCK(dev, cpu) { \
1211 if ((dev->features & NETIF_F_LLTX) == 0) { \
1212 spin_lock(&dev->xmit_lock); \
1213 dev->xmit_lock_owner = cpu; \
1217 #define HARD_TX_UNLOCK(dev) { \
1218 if ((dev->features & NETIF_F_LLTX) == 0) { \
1219 dev->xmit_lock_owner = -1; \
1220 spin_unlock(&dev->xmit_lock); \
1225 * dev_queue_xmit - transmit a buffer
1226 * @skb: buffer to transmit
1228 * Queue a buffer for transmission to a network device. The caller must
1229 * have set the device and priority and built the buffer before calling
1230 * this function. The function can be called from an interrupt.
1232 * A negative errno code is returned on a failure. A success does not
1233 * guarantee the frame will be transmitted as it may be dropped due
1234 * to congestion or traffic shaping.
1236 * -----------------------------------------------------------------------------------
1237 * I notice this method can also return errors from the queue disciplines,
1238 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1241 * Regardless of the return value, the skb is consumed, so it is currently
1242 * difficult to retry a send to this method. (You can bump the ref count
1243 * before sending to hold a reference for retry if you are careful.)
1245 * When calling this method, interrupts MUST be enabled. This is because
1246 * the BH enable code must have IRQs enabled so that it will not deadlock.
1250 int dev_queue_xmit(struct sk_buff *skb)
1252 struct net_device *dev = skb->dev;
1256 if (skb_shinfo(skb)->frag_list &&
1257 !(dev->features & NETIF_F_FRAGLIST) &&
1258 __skb_linearize(skb, GFP_ATOMIC))
1261 /* Fragmented skb is linearized if device does not support SG,
1262 * or if at least one of fragments is in highmem and device
1263 * does not support DMA from it.
1265 if (skb_shinfo(skb)->nr_frags &&
1266 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1267 __skb_linearize(skb, GFP_ATOMIC))
1270 /* If packet is not checksummed and device does not support
1271 * checksumming for this protocol, complete checksumming here.
1273 if (skb->ip_summed == CHECKSUM_HW &&
1274 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1275 (!(dev->features & NETIF_F_IP_CSUM) ||
1276 skb->protocol != htons(ETH_P_IP))))
1277 if (skb_checksum_help(skb, 0))
1280 spin_lock_prefetch(&dev->queue_lock);
1282 /* Disable soft irqs for various locks below. Also
1283 * stops preemption for RCU.
1287 /* Updates of qdisc are serialized by queue_lock.
1288 * The struct Qdisc which is pointed to by qdisc is now a
1289 * rcu structure - it may be accessed without acquiring
1290 * a lock (but the structure may be stale.) The freeing of the
1291 * qdisc will be deferred until it's known that there are no
1292 * more references to it.
1294 * If the qdisc has an enqueue function, we still need to
1295 * hold the queue_lock before calling it, since queue_lock
1296 * also serializes access to the device queue.
1299 q = rcu_dereference(dev->qdisc);
1300 #ifdef CONFIG_NET_CLS_ACT
1301 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1304 /* Grab device queue */
1305 spin_lock(&dev->queue_lock);
1308 rc = q->enqueue(skb, q);
1310 spin_unlock(&dev->queue_lock);
1312 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1315 spin_unlock(&dev->queue_lock);
1318 /* The device has no queue. Common case for software devices:
1319 loopback, all the sorts of tunnels...
1321 Really, it is unlikely that xmit_lock protection is necessary here.
1322 (f.e. loopback and IP tunnels are clean ignoring statistics
1324 However, it is possible, that they rely on protection
1327 Check this and shot the lock. It is not prone from deadlocks.
1328 Either shot noqueue qdisc, it is even simpler 8)
1330 if (dev->flags & IFF_UP) {
1331 int cpu = smp_processor_id(); /* ok because BHs are off */
1333 if (dev->xmit_lock_owner != cpu) {
1335 HARD_TX_LOCK(dev, cpu);
1337 if (!netif_queue_stopped(dev)) {
1339 dev_queue_xmit_nit(skb, dev);
1342 if (!dev->hard_start_xmit(skb, dev)) {
1343 HARD_TX_UNLOCK(dev);
1347 HARD_TX_UNLOCK(dev);
1348 if (net_ratelimit())
1349 printk(KERN_CRIT "Virtual device %s asks to "
1350 "queue packet!\n", dev->name);
1352 /* Recursion is detected! It is possible,
1354 if (net_ratelimit())
1355 printk(KERN_CRIT "Dead loop on virtual device "
1356 "%s, fix it urgently!\n", dev->name);
1372 /*=======================================================================
1374 =======================================================================*/
1376 int netdev_max_backlog = 1000;
1377 int netdev_budget = 300;
1378 int weight_p = 64; /* old backlog weight */
1380 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1384 * netif_rx - post buffer to the network code
1385 * @skb: buffer to post
1387 * This function receives a packet from a device driver and queues it for
1388 * the upper (protocol) levels to process. It always succeeds. The buffer
1389 * may be dropped during processing for congestion control or by the
1393 * NET_RX_SUCCESS (no congestion)
1394 * NET_RX_CN_LOW (low congestion)
1395 * NET_RX_CN_MOD (moderate congestion)
1396 * NET_RX_CN_HIGH (high congestion)
1397 * NET_RX_DROP (packet was dropped)
1401 int netif_rx(struct sk_buff *skb)
1403 struct softnet_data *queue;
1404 unsigned long flags;
1406 /* if netpoll wants it, pretend we never saw it */
1407 if (netpoll_rx(skb))
1410 if (!skb->tstamp.off_sec)
1414 * The code is rearranged so that the path is the most
1415 * short when CPU is congested, but is still operating.
1417 local_irq_save(flags);
1418 queue = &__get_cpu_var(softnet_data);
1420 __get_cpu_var(netdev_rx_stat).total++;
1421 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1422 if (queue->input_pkt_queue.qlen) {
1425 __skb_queue_tail(&queue->input_pkt_queue, skb);
1426 local_irq_restore(flags);
1427 return NET_RX_SUCCESS;
1430 netif_rx_schedule(&queue->backlog_dev);
1434 __get_cpu_var(netdev_rx_stat).dropped++;
1435 local_irq_restore(flags);
1441 int netif_rx_ni(struct sk_buff *skb)
1446 err = netif_rx(skb);
1447 if (local_softirq_pending())
1454 EXPORT_SYMBOL(netif_rx_ni);
1456 static inline struct net_device *skb_bond(struct sk_buff *skb)
1458 struct net_device *dev = skb->dev;
1461 skb->dev = dev->master;
1466 static void net_tx_action(struct softirq_action *h)
1468 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1470 if (sd->completion_queue) {
1471 struct sk_buff *clist;
1473 local_irq_disable();
1474 clist = sd->completion_queue;
1475 sd->completion_queue = NULL;
1479 struct sk_buff *skb = clist;
1480 clist = clist->next;
1482 BUG_TRAP(!atomic_read(&skb->users));
1487 if (sd->output_queue) {
1488 struct net_device *head;
1490 local_irq_disable();
1491 head = sd->output_queue;
1492 sd->output_queue = NULL;
1496 struct net_device *dev = head;
1497 head = head->next_sched;
1499 smp_mb__before_clear_bit();
1500 clear_bit(__LINK_STATE_SCHED, &dev->state);
1502 if (spin_trylock(&dev->queue_lock)) {
1504 spin_unlock(&dev->queue_lock);
1506 netif_schedule(dev);
1512 static __inline__ int deliver_skb(struct sk_buff *skb,
1513 struct packet_type *pt_prev,
1514 struct net_device *orig_dev)
1516 atomic_inc(&skb->users);
1517 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1520 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1521 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1523 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1524 unsigned char *addr);
1525 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1527 static __inline__ int handle_bridge(struct sk_buff **pskb,
1528 struct packet_type **pt_prev, int *ret,
1529 struct net_device *orig_dev)
1531 struct net_bridge_port *port;
1533 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1534 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1538 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1542 return br_handle_frame_hook(port, pskb);
1545 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1548 #ifdef CONFIG_NET_CLS_ACT
1549 /* TODO: Maybe we should just force sch_ingress to be compiled in
1550 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1551 * a compare and 2 stores extra right now if we dont have it on
1552 * but have CONFIG_NET_CLS_ACT
1553 * NOTE: This doesnt stop any functionality; if you dont have
1554 * the ingress scheduler, you just cant add policies on ingress.
1557 static int ing_filter(struct sk_buff *skb)
1560 struct net_device *dev = skb->dev;
1561 int result = TC_ACT_OK;
1563 if (dev->qdisc_ingress) {
1564 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1565 if (MAX_RED_LOOP < ttl++) {
1566 printk("Redir loop detected Dropping packet (%s->%s)\n",
1567 skb->input_dev->name, skb->dev->name);
1571 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1573 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1575 spin_lock(&dev->ingress_lock);
1576 if ((q = dev->qdisc_ingress) != NULL)
1577 result = q->enqueue(skb, q);
1578 spin_unlock(&dev->ingress_lock);
1586 int netif_receive_skb(struct sk_buff *skb)
1588 struct packet_type *ptype, *pt_prev;
1589 struct net_device *orig_dev;
1590 int ret = NET_RX_DROP;
1591 unsigned short type;
1593 /* if we've gotten here through NAPI, check netpoll */
1594 if (skb->dev->poll && netpoll_rx(skb))
1597 if (!skb->tstamp.off_sec)
1600 if (!skb->input_dev)
1601 skb->input_dev = skb->dev;
1603 orig_dev = skb_bond(skb);
1605 __get_cpu_var(netdev_rx_stat).total++;
1607 skb->h.raw = skb->nh.raw = skb->data;
1608 skb->mac_len = skb->nh.raw - skb->mac.raw;
1614 #ifdef CONFIG_NET_CLS_ACT
1615 if (skb->tc_verd & TC_NCLS) {
1616 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1621 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1622 if (!ptype->dev || ptype->dev == skb->dev) {
1624 ret = deliver_skb(skb, pt_prev, orig_dev);
1629 #ifdef CONFIG_NET_CLS_ACT
1631 ret = deliver_skb(skb, pt_prev, orig_dev);
1632 pt_prev = NULL; /* noone else should process this after*/
1634 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1637 ret = ing_filter(skb);
1639 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1648 handle_diverter(skb);
1650 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1653 type = skb->protocol;
1654 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1655 if (ptype->type == type &&
1656 (!ptype->dev || ptype->dev == skb->dev)) {
1658 ret = deliver_skb(skb, pt_prev, orig_dev);
1664 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1667 /* Jamal, now you will not able to escape explaining
1668 * me how you were going to use this. :-)
1678 static int process_backlog(struct net_device *backlog_dev, int *budget)
1681 int quota = min(backlog_dev->quota, *budget);
1682 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1683 unsigned long start_time = jiffies;
1685 backlog_dev->weight = weight_p;
1687 struct sk_buff *skb;
1688 struct net_device *dev;
1690 local_irq_disable();
1691 skb = __skb_dequeue(&queue->input_pkt_queue);
1698 netif_receive_skb(skb);
1704 if (work >= quota || jiffies - start_time > 1)
1709 backlog_dev->quota -= work;
1714 backlog_dev->quota -= work;
1717 list_del(&backlog_dev->poll_list);
1718 smp_mb__before_clear_bit();
1719 netif_poll_enable(backlog_dev);
1725 static void net_rx_action(struct softirq_action *h)
1727 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1728 unsigned long start_time = jiffies;
1729 int budget = netdev_budget;
1732 local_irq_disable();
1734 while (!list_empty(&queue->poll_list)) {
1735 struct net_device *dev;
1737 if (budget <= 0 || jiffies - start_time > 1)
1742 dev = list_entry(queue->poll_list.next,
1743 struct net_device, poll_list);
1744 have = netpoll_poll_lock(dev);
1746 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1747 netpoll_poll_unlock(have);
1748 local_irq_disable();
1749 list_del(&dev->poll_list);
1750 list_add_tail(&dev->poll_list, &queue->poll_list);
1752 dev->quota += dev->weight;
1754 dev->quota = dev->weight;
1756 netpoll_poll_unlock(have);
1758 local_irq_disable();
1766 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1767 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1771 static gifconf_func_t * gifconf_list [NPROTO];
1774 * register_gifconf - register a SIOCGIF handler
1775 * @family: Address family
1776 * @gifconf: Function handler
1778 * Register protocol dependent address dumping routines. The handler
1779 * that is passed must not be freed or reused until it has been replaced
1780 * by another handler.
1782 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1784 if (family >= NPROTO)
1786 gifconf_list[family] = gifconf;
1792 * Map an interface index to its name (SIOCGIFNAME)
1796 * We need this ioctl for efficient implementation of the
1797 * if_indextoname() function required by the IPv6 API. Without
1798 * it, we would have to search all the interfaces to find a
1802 static int dev_ifname(struct ifreq __user *arg)
1804 struct net_device *dev;
1808 * Fetch the caller's info block.
1811 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1814 read_lock(&dev_base_lock);
1815 dev = __dev_get_by_index(ifr.ifr_ifindex);
1817 read_unlock(&dev_base_lock);
1821 strcpy(ifr.ifr_name, dev->name);
1822 read_unlock(&dev_base_lock);
1824 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1830 * Perform a SIOCGIFCONF call. This structure will change
1831 * size eventually, and there is nothing I can do about it.
1832 * Thus we will need a 'compatibility mode'.
1835 static int dev_ifconf(char __user *arg)
1838 struct net_device *dev;
1845 * Fetch the caller's info block.
1848 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1855 * Loop over the interfaces, and write an info block for each.
1859 for (dev = dev_base; dev; dev = dev->next) {
1860 if (vx_flags(VXF_HIDE_NETIF, 0) &&
1861 !dev_in_nx_info(dev, current->nx_info))
1863 for (i = 0; i < NPROTO; i++) {
1864 if (gifconf_list[i]) {
1867 done = gifconf_list[i](dev, NULL, 0);
1869 done = gifconf_list[i](dev, pos + total,
1879 * All done. Write the updated control block back to the caller.
1881 ifc.ifc_len = total;
1884 * Both BSD and Solaris return 0 here, so we do too.
1886 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1889 #ifdef CONFIG_PROC_FS
1891 * This is invoked by the /proc filesystem handler to display a device
1894 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1896 struct net_device *dev;
1899 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1901 return i == pos ? dev : NULL;
1904 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1906 read_lock(&dev_base_lock);
1907 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1910 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1913 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1916 void dev_seq_stop(struct seq_file *seq, void *v)
1918 read_unlock(&dev_base_lock);
1921 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1923 struct nx_info *nxi = current->nx_info;
1925 if (vx_flags(VXF_HIDE_NETIF, 0) && !dev_in_nx_info(dev, nxi))
1927 if (dev->get_stats) {
1928 struct net_device_stats *stats = dev->get_stats(dev);
1930 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1931 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1932 dev->name, stats->rx_bytes, stats->rx_packets,
1934 stats->rx_dropped + stats->rx_missed_errors,
1935 stats->rx_fifo_errors,
1936 stats->rx_length_errors + stats->rx_over_errors +
1937 stats->rx_crc_errors + stats->rx_frame_errors,
1938 stats->rx_compressed, stats->multicast,
1939 stats->tx_bytes, stats->tx_packets,
1940 stats->tx_errors, stats->tx_dropped,
1941 stats->tx_fifo_errors, stats->collisions,
1942 stats->tx_carrier_errors +
1943 stats->tx_aborted_errors +
1944 stats->tx_window_errors +
1945 stats->tx_heartbeat_errors,
1946 stats->tx_compressed);
1948 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1952 * Called from the PROCfs module. This now uses the new arbitrary sized
1953 * /proc/net interface to create /proc/net/dev
1955 static int dev_seq_show(struct seq_file *seq, void *v)
1957 if (v == SEQ_START_TOKEN)
1958 seq_puts(seq, "Inter-| Receive "
1960 " face |bytes packets errs drop fifo frame "
1961 "compressed multicast|bytes packets errs "
1962 "drop fifo colls carrier compressed\n");
1964 dev_seq_printf_stats(seq, v);
1968 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
1970 struct netif_rx_stats *rc = NULL;
1972 while (*pos < NR_CPUS)
1973 if (cpu_online(*pos)) {
1974 rc = &per_cpu(netdev_rx_stat, *pos);
1981 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
1983 return softnet_get_online(pos);
1986 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1989 return softnet_get_online(pos);
1992 static void softnet_seq_stop(struct seq_file *seq, void *v)
1996 static int softnet_seq_show(struct seq_file *seq, void *v)
1998 struct netif_rx_stats *s = v;
2000 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2001 s->total, s->dropped, s->time_squeeze, 0,
2002 0, 0, 0, 0, /* was fastroute */
2007 static struct seq_operations dev_seq_ops = {
2008 .start = dev_seq_start,
2009 .next = dev_seq_next,
2010 .stop = dev_seq_stop,
2011 .show = dev_seq_show,
2014 static int dev_seq_open(struct inode *inode, struct file *file)
2016 return seq_open(file, &dev_seq_ops);
2019 static struct file_operations dev_seq_fops = {
2020 .owner = THIS_MODULE,
2021 .open = dev_seq_open,
2023 .llseek = seq_lseek,
2024 .release = seq_release,
2027 static struct seq_operations softnet_seq_ops = {
2028 .start = softnet_seq_start,
2029 .next = softnet_seq_next,
2030 .stop = softnet_seq_stop,
2031 .show = softnet_seq_show,
2034 static int softnet_seq_open(struct inode *inode, struct file *file)
2036 return seq_open(file, &softnet_seq_ops);
2039 static struct file_operations softnet_seq_fops = {
2040 .owner = THIS_MODULE,
2041 .open = softnet_seq_open,
2043 .llseek = seq_lseek,
2044 .release = seq_release,
2048 extern int wireless_proc_init(void);
2050 #define wireless_proc_init() 0
2053 static int __init dev_proc_init(void)
2057 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2059 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2061 if (wireless_proc_init())
2067 proc_net_remove("softnet_stat");
2069 proc_net_remove("dev");
2073 #define dev_proc_init() 0
2074 #endif /* CONFIG_PROC_FS */
2078 * netdev_set_master - set up master/slave pair
2079 * @slave: slave device
2080 * @master: new master device
2082 * Changes the master device of the slave. Pass %NULL to break the
2083 * bonding. The caller must hold the RTNL semaphore. On a failure
2084 * a negative errno code is returned. On success the reference counts
2085 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2086 * function returns zero.
2088 int netdev_set_master(struct net_device *slave, struct net_device *master)
2090 struct net_device *old = slave->master;
2100 slave->master = master;
2108 slave->flags |= IFF_SLAVE;
2110 slave->flags &= ~IFF_SLAVE;
2112 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2117 * dev_set_promiscuity - update promiscuity count on a device
2121 * Add or remove promsicuity from a device. While the count in the device
2122 * remains above zero the interface remains promiscuous. Once it hits zero
2123 * the device reverts back to normal filtering operation. A negative inc
2124 * value is used to drop promiscuity on the device.
2126 void dev_set_promiscuity(struct net_device *dev, int inc)
2128 unsigned short old_flags = dev->flags;
2130 if ((dev->promiscuity += inc) == 0)
2131 dev->flags &= ~IFF_PROMISC;
2133 dev->flags |= IFF_PROMISC;
2134 if (dev->flags != old_flags) {
2136 printk(KERN_INFO "device %s %s promiscuous mode\n",
2137 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2143 * dev_set_allmulti - update allmulti count on a device
2147 * Add or remove reception of all multicast frames to a device. While the
2148 * count in the device remains above zero the interface remains listening
2149 * to all interfaces. Once it hits zero the device reverts back to normal
2150 * filtering operation. A negative @inc value is used to drop the counter
2151 * when releasing a resource needing all multicasts.
2154 void dev_set_allmulti(struct net_device *dev, int inc)
2156 unsigned short old_flags = dev->flags;
2158 dev->flags |= IFF_ALLMULTI;
2159 if ((dev->allmulti += inc) == 0)
2160 dev->flags &= ~IFF_ALLMULTI;
2161 if (dev->flags ^ old_flags)
2165 unsigned dev_get_flags(const struct net_device *dev)
2169 flags = (dev->flags & ~(IFF_PROMISC |
2172 (dev->gflags & (IFF_PROMISC |
2175 if (netif_running(dev) && netif_carrier_ok(dev))
2176 flags |= IFF_RUNNING;
2181 int dev_change_flags(struct net_device *dev, unsigned flags)
2184 int old_flags = dev->flags;
2187 * Set the flags on our device.
2190 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2191 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2193 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2197 * Load in the correct multicast list now the flags have changed.
2203 * Have we downed the interface. We handle IFF_UP ourselves
2204 * according to user attempts to set it, rather than blindly
2209 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2210 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2216 if (dev->flags & IFF_UP &&
2217 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2219 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2221 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2222 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2223 dev->gflags ^= IFF_PROMISC;
2224 dev_set_promiscuity(dev, inc);
2227 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2228 is important. Some (broken) drivers set IFF_PROMISC, when
2229 IFF_ALLMULTI is requested not asking us and not reporting.
2231 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2232 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2233 dev->gflags ^= IFF_ALLMULTI;
2234 dev_set_allmulti(dev, inc);
2237 if (old_flags ^ dev->flags)
2238 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2243 int dev_set_mtu(struct net_device *dev, int new_mtu)
2247 if (new_mtu == dev->mtu)
2250 /* MTU must be positive. */
2254 if (!netif_device_present(dev))
2258 if (dev->change_mtu)
2259 err = dev->change_mtu(dev, new_mtu);
2262 if (!err && dev->flags & IFF_UP)
2263 notifier_call_chain(&netdev_chain,
2264 NETDEV_CHANGEMTU, dev);
2268 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2272 if (!dev->set_mac_address)
2274 if (sa->sa_family != dev->type)
2276 if (!netif_device_present(dev))
2278 err = dev->set_mac_address(dev, sa);
2280 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2285 * Perform the SIOCxIFxxx calls.
2287 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2290 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2296 case SIOCGIFFLAGS: /* Get interface flags */
2297 ifr->ifr_flags = dev_get_flags(dev);
2300 case SIOCSIFFLAGS: /* Set interface flags */
2301 return dev_change_flags(dev, ifr->ifr_flags);
2303 case SIOCGIFMETRIC: /* Get the metric on the interface
2304 (currently unused) */
2305 ifr->ifr_metric = 0;
2308 case SIOCSIFMETRIC: /* Set the metric on the interface
2309 (currently unused) */
2312 case SIOCGIFMTU: /* Get the MTU of a device */
2313 ifr->ifr_mtu = dev->mtu;
2316 case SIOCSIFMTU: /* Set the MTU of a device */
2317 return dev_set_mtu(dev, ifr->ifr_mtu);
2321 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2323 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2324 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2325 ifr->ifr_hwaddr.sa_family = dev->type;
2329 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2331 case SIOCSIFHWBROADCAST:
2332 if (ifr->ifr_hwaddr.sa_family != dev->type)
2334 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2335 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2336 notifier_call_chain(&netdev_chain,
2337 NETDEV_CHANGEADDR, dev);
2341 ifr->ifr_map.mem_start = dev->mem_start;
2342 ifr->ifr_map.mem_end = dev->mem_end;
2343 ifr->ifr_map.base_addr = dev->base_addr;
2344 ifr->ifr_map.irq = dev->irq;
2345 ifr->ifr_map.dma = dev->dma;
2346 ifr->ifr_map.port = dev->if_port;
2350 if (dev->set_config) {
2351 if (!netif_device_present(dev))
2353 return dev->set_config(dev, &ifr->ifr_map);
2358 if (!dev->set_multicast_list ||
2359 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2361 if (!netif_device_present(dev))
2363 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2367 if (!dev->set_multicast_list ||
2368 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2370 if (!netif_device_present(dev))
2372 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2376 ifr->ifr_ifindex = dev->ifindex;
2380 ifr->ifr_qlen = dev->tx_queue_len;
2384 if (ifr->ifr_qlen < 0)
2386 dev->tx_queue_len = ifr->ifr_qlen;
2390 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2391 return dev_change_name(dev, ifr->ifr_newname);
2394 * Unknown or private ioctl
2398 if ((cmd >= SIOCDEVPRIVATE &&
2399 cmd <= SIOCDEVPRIVATE + 15) ||
2400 cmd == SIOCBONDENSLAVE ||
2401 cmd == SIOCBONDRELEASE ||
2402 cmd == SIOCBONDSETHWADDR ||
2403 cmd == SIOCBONDSLAVEINFOQUERY ||
2404 cmd == SIOCBONDINFOQUERY ||
2405 cmd == SIOCBONDCHANGEACTIVE ||
2406 cmd == SIOCGMIIPHY ||
2407 cmd == SIOCGMIIREG ||
2408 cmd == SIOCSMIIREG ||
2409 cmd == SIOCBRADDIF ||
2410 cmd == SIOCBRDELIF ||
2411 cmd == SIOCWANDEV) {
2413 if (dev->do_ioctl) {
2414 if (netif_device_present(dev))
2415 err = dev->do_ioctl(dev, ifr,
2428 * This function handles all "interface"-type I/O control requests. The actual
2429 * 'doing' part of this is dev_ifsioc above.
2433 * dev_ioctl - network device ioctl
2434 * @cmd: command to issue
2435 * @arg: pointer to a struct ifreq in user space
2437 * Issue ioctl functions to devices. This is normally called by the
2438 * user space syscall interfaces but can sometimes be useful for
2439 * other purposes. The return value is the return from the syscall if
2440 * positive or a negative errno code on error.
2443 int dev_ioctl(unsigned int cmd, void __user *arg)
2449 /* One special case: SIOCGIFCONF takes ifconf argument
2450 and requires shared lock, because it sleeps writing
2454 if (cmd == SIOCGIFCONF) {
2456 ret = dev_ifconf((char __user *) arg);
2460 if (cmd == SIOCGIFNAME)
2461 return dev_ifname((struct ifreq __user *)arg);
2463 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2466 ifr.ifr_name[IFNAMSIZ-1] = 0;
2468 colon = strchr(ifr.ifr_name, ':');
2473 * See which interface the caller is talking about.
2478 * These ioctl calls:
2479 * - can be done by all.
2480 * - atomic and do not require locking.
2491 dev_load(ifr.ifr_name);
2492 read_lock(&dev_base_lock);
2493 ret = dev_ifsioc(&ifr, cmd);
2494 read_unlock(&dev_base_lock);
2498 if (copy_to_user(arg, &ifr,
2499 sizeof(struct ifreq)))
2505 dev_load(ifr.ifr_name);
2507 ret = dev_ethtool(&ifr);
2512 if (copy_to_user(arg, &ifr,
2513 sizeof(struct ifreq)))
2519 * These ioctl calls:
2520 * - require superuser power.
2521 * - require strict serialization.
2527 if (!capable(CAP_NET_ADMIN))
2529 dev_load(ifr.ifr_name);
2531 ret = dev_ifsioc(&ifr, cmd);
2536 if (copy_to_user(arg, &ifr,
2537 sizeof(struct ifreq)))
2543 * These ioctl calls:
2544 * - require superuser power.
2545 * - require strict serialization.
2546 * - do not return a value
2556 case SIOCSIFHWBROADCAST:
2559 case SIOCBONDENSLAVE:
2560 case SIOCBONDRELEASE:
2561 case SIOCBONDSETHWADDR:
2562 case SIOCBONDCHANGEACTIVE:
2565 if (!capable(CAP_NET_ADMIN))
2568 case SIOCBONDSLAVEINFOQUERY:
2569 case SIOCBONDINFOQUERY:
2570 dev_load(ifr.ifr_name);
2572 ret = dev_ifsioc(&ifr, cmd);
2577 /* Get the per device memory space. We can add this but
2578 * currently do not support it */
2580 /* Set the per device memory buffer space.
2581 * Not applicable in our case */
2586 * Unknown or private ioctl.
2589 if (cmd == SIOCWANDEV ||
2590 (cmd >= SIOCDEVPRIVATE &&
2591 cmd <= SIOCDEVPRIVATE + 15)) {
2592 dev_load(ifr.ifr_name);
2594 ret = dev_ifsioc(&ifr, cmd);
2596 if (!ret && copy_to_user(arg, &ifr,
2597 sizeof(struct ifreq)))
2602 /* Take care of Wireless Extensions */
2603 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2604 /* If command is `set a parameter', or
2605 * `get the encoding parameters', check if
2606 * the user has the right to do it */
2607 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2608 if (!capable(CAP_NET_ADMIN))
2611 dev_load(ifr.ifr_name);
2613 /* Follow me in net/core/wireless.c */
2614 ret = wireless_process_ioctl(&ifr, cmd);
2616 if (IW_IS_GET(cmd) &&
2617 copy_to_user(arg, &ifr,
2618 sizeof(struct ifreq)))
2622 #endif /* WIRELESS_EXT */
2629 * dev_new_index - allocate an ifindex
2631 * Returns a suitable unique value for a new device interface
2632 * number. The caller must hold the rtnl semaphore or the
2633 * dev_base_lock to be sure it remains unique.
2635 static int dev_new_index(void)
2641 if (!__dev_get_by_index(ifindex))
2646 static int dev_boot_phase = 1;
2648 /* Delayed registration/unregisteration */
2649 static DEFINE_SPINLOCK(net_todo_list_lock);
2650 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2652 static inline void net_set_todo(struct net_device *dev)
2654 spin_lock(&net_todo_list_lock);
2655 list_add_tail(&dev->todo_list, &net_todo_list);
2656 spin_unlock(&net_todo_list_lock);
2660 * register_netdevice - register a network device
2661 * @dev: device to register
2663 * Take a completed network device structure and add it to the kernel
2664 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2665 * chain. 0 is returned on success. A negative errno code is returned
2666 * on a failure to set up the device, or if the name is a duplicate.
2668 * Callers must hold the rtnl semaphore. You may want
2669 * register_netdev() instead of this.
2672 * The locking appears insufficient to guarantee two parallel registers
2673 * will not get the same name.
2676 int register_netdevice(struct net_device *dev)
2678 struct hlist_head *head;
2679 struct hlist_node *p;
2682 BUG_ON(dev_boot_phase);
2685 /* When net_device's are persistent, this will be fatal. */
2686 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2688 spin_lock_init(&dev->queue_lock);
2689 spin_lock_init(&dev->xmit_lock);
2690 dev->xmit_lock_owner = -1;
2691 #ifdef CONFIG_NET_CLS_ACT
2692 spin_lock_init(&dev->ingress_lock);
2695 ret = alloc_divert_blk(dev);
2701 /* Init, if this function is available */
2703 ret = dev->init(dev);
2711 if (!dev_valid_name(dev->name)) {
2716 dev->ifindex = dev_new_index();
2717 if (dev->iflink == -1)
2718 dev->iflink = dev->ifindex;
2720 /* Check for existence of name */
2721 head = dev_name_hash(dev->name);
2722 hlist_for_each(p, head) {
2723 struct net_device *d
2724 = hlist_entry(p, struct net_device, name_hlist);
2725 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2731 /* Fix illegal SG+CSUM combinations. */
2732 if ((dev->features & NETIF_F_SG) &&
2733 !(dev->features & (NETIF_F_IP_CSUM |
2735 NETIF_F_HW_CSUM))) {
2736 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2738 dev->features &= ~NETIF_F_SG;
2741 /* TSO requires that SG is present as well. */
2742 if ((dev->features & NETIF_F_TSO) &&
2743 !(dev->features & NETIF_F_SG)) {
2744 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2746 dev->features &= ~NETIF_F_TSO;
2748 if (dev->features & NETIF_F_UFO) {
2749 if (!(dev->features & NETIF_F_HW_CSUM)) {
2750 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2751 "NETIF_F_HW_CSUM feature.\n",
2753 dev->features &= ~NETIF_F_UFO;
2755 if (!(dev->features & NETIF_F_SG)) {
2756 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2757 "NETIF_F_SG feature.\n",
2759 dev->features &= ~NETIF_F_UFO;
2764 * nil rebuild_header routine,
2765 * that should be never called and used as just bug trap.
2768 if (!dev->rebuild_header)
2769 dev->rebuild_header = default_rebuild_header;
2772 * Default initial state at registry is that the
2773 * device is present.
2776 set_bit(__LINK_STATE_PRESENT, &dev->state);
2779 dev_init_scheduler(dev);
2780 write_lock_bh(&dev_base_lock);
2782 dev_tail = &dev->next;
2783 hlist_add_head(&dev->name_hlist, head);
2784 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2786 dev->reg_state = NETREG_REGISTERING;
2787 write_unlock_bh(&dev_base_lock);
2789 /* Notify protocols, that a new device appeared. */
2790 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2792 /* Finish registration after unlock */
2799 free_divert_blk(dev);
2804 * register_netdev - register a network device
2805 * @dev: device to register
2807 * Take a completed network device structure and add it to the kernel
2808 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2809 * chain. 0 is returned on success. A negative errno code is returned
2810 * on a failure to set up the device, or if the name is a duplicate.
2812 * This is a wrapper around register_netdev that takes the rtnl semaphore
2813 * and expands the device name if you passed a format string to
2816 int register_netdev(struct net_device *dev)
2823 * If the name is a format string the caller wants us to do a
2826 if (strchr(dev->name, '%')) {
2827 err = dev_alloc_name(dev, dev->name);
2833 * Back compatibility hook. Kill this one in 2.5
2835 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2836 err = dev_alloc_name(dev, "eth%d");
2841 err = register_netdevice(dev);
2846 EXPORT_SYMBOL(register_netdev);
2849 * netdev_wait_allrefs - wait until all references are gone.
2851 * This is called when unregistering network devices.
2853 * Any protocol or device that holds a reference should register
2854 * for netdevice notification, and cleanup and put back the
2855 * reference if they receive an UNREGISTER event.
2856 * We can get stuck here if buggy protocols don't correctly
2859 static void netdev_wait_allrefs(struct net_device *dev)
2861 unsigned long rebroadcast_time, warning_time;
2863 rebroadcast_time = warning_time = jiffies;
2864 while (atomic_read(&dev->refcnt) != 0) {
2865 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2868 /* Rebroadcast unregister notification */
2869 notifier_call_chain(&netdev_chain,
2870 NETDEV_UNREGISTER, dev);
2872 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2874 /* We must not have linkwatch events
2875 * pending on unregister. If this
2876 * happens, we simply run the queue
2877 * unscheduled, resulting in a noop
2880 linkwatch_run_queue();
2885 rebroadcast_time = jiffies;
2890 if (time_after(jiffies, warning_time + 10 * HZ)) {
2891 printk(KERN_EMERG "unregister_netdevice: "
2892 "waiting for %s to become free. Usage "
2894 dev->name, atomic_read(&dev->refcnt));
2895 warning_time = jiffies;
2904 * register_netdevice(x1);
2905 * register_netdevice(x2);
2907 * unregister_netdevice(y1);
2908 * unregister_netdevice(y2);
2914 * We are invoked by rtnl_unlock() after it drops the semaphore.
2915 * This allows us to deal with problems:
2916 * 1) We can create/delete sysfs objects which invoke hotplug
2917 * without deadlocking with linkwatch via keventd.
2918 * 2) Since we run with the RTNL semaphore not held, we can sleep
2919 * safely in order to wait for the netdev refcnt to drop to zero.
2921 static DECLARE_MUTEX(net_todo_run_mutex);
2922 void netdev_run_todo(void)
2924 struct list_head list = LIST_HEAD_INIT(list);
2928 /* Need to guard against multiple cpu's getting out of order. */
2929 down(&net_todo_run_mutex);
2931 /* Not safe to do outside the semaphore. We must not return
2932 * until all unregister events invoked by the local processor
2933 * have been completed (either by this todo run, or one on
2936 if (list_empty(&net_todo_list))
2939 /* Snapshot list, allow later requests */
2940 spin_lock(&net_todo_list_lock);
2941 list_splice_init(&net_todo_list, &list);
2942 spin_unlock(&net_todo_list_lock);
2944 while (!list_empty(&list)) {
2945 struct net_device *dev
2946 = list_entry(list.next, struct net_device, todo_list);
2947 list_del(&dev->todo_list);
2949 switch(dev->reg_state) {
2950 case NETREG_REGISTERING:
2951 dev->reg_state = NETREG_REGISTERED;
2952 err = netdev_register_sysfs(dev);
2954 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2958 case NETREG_UNREGISTERING:
2959 netdev_unregister_sysfs(dev);
2960 dev->reg_state = NETREG_UNREGISTERED;
2962 netdev_wait_allrefs(dev);
2965 BUG_ON(atomic_read(&dev->refcnt));
2966 BUG_TRAP(!dev->ip_ptr);
2967 BUG_TRAP(!dev->ip6_ptr);
2968 BUG_TRAP(!dev->dn_ptr);
2971 /* It must be the very last action,
2972 * after this 'dev' may point to freed up memory.
2974 if (dev->destructor)
2975 dev->destructor(dev);
2979 printk(KERN_ERR "network todo '%s' but state %d\n",
2980 dev->name, dev->reg_state);
2986 up(&net_todo_run_mutex);
2990 * alloc_netdev - allocate network device
2991 * @sizeof_priv: size of private data to allocate space for
2992 * @name: device name format string
2993 * @setup: callback to initialize device
2995 * Allocates a struct net_device with private data area for driver use
2996 * and performs basic initialization.
2998 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
2999 void (*setup)(struct net_device *))
3002 struct net_device *dev;
3005 /* ensure 32-byte alignment of both the device and private area */
3006 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3007 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3009 p = kmalloc(alloc_size, GFP_KERNEL);
3011 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3014 memset(p, 0, alloc_size);
3016 dev = (struct net_device *)
3017 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3018 dev->padded = (char *)dev - (char *)p;
3021 dev->priv = netdev_priv(dev);
3024 strcpy(dev->name, name);
3027 EXPORT_SYMBOL(alloc_netdev);
3030 * free_netdev - free network device
3033 * This function does the last stage of destroying an allocated device
3034 * interface. The reference to the device object is released.
3035 * If this is the last reference then it will be freed.
3037 void free_netdev(struct net_device *dev)
3040 /* Compatiablity with error handling in drivers */
3041 if (dev->reg_state == NETREG_UNINITIALIZED) {
3042 kfree((char *)dev - dev->padded);
3046 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3047 dev->reg_state = NETREG_RELEASED;
3049 /* will free via class release */
3050 class_device_put(&dev->class_dev);
3052 kfree((char *)dev - dev->padded);
3056 /* Synchronize with packet receive processing. */
3057 void synchronize_net(void)
3064 * unregister_netdevice - remove device from the kernel
3067 * This function shuts down a device interface and removes it
3068 * from the kernel tables. On success 0 is returned, on a failure
3069 * a negative errno code is returned.
3071 * Callers must hold the rtnl semaphore. You may want
3072 * unregister_netdev() instead of this.
3075 int unregister_netdevice(struct net_device *dev)
3077 struct net_device *d, **dp;
3079 BUG_ON(dev_boot_phase);
3082 /* Some devices call without registering for initialization unwind. */
3083 if (dev->reg_state == NETREG_UNINITIALIZED) {
3084 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3085 "was registered\n", dev->name, dev);
3089 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3091 /* If device is running, close it first. */
3092 if (dev->flags & IFF_UP)
3095 /* And unlink it from device chain. */
3096 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3098 write_lock_bh(&dev_base_lock);
3099 hlist_del(&dev->name_hlist);
3100 hlist_del(&dev->index_hlist);
3101 if (dev_tail == &dev->next)
3104 write_unlock_bh(&dev_base_lock);
3109 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3114 dev->reg_state = NETREG_UNREGISTERING;
3118 /* Shutdown queueing discipline. */
3122 /* Notify protocols, that we are about to destroy
3123 this device. They should clean all the things.
3125 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3128 * Flush the multicast chain
3130 dev_mc_discard(dev);
3135 /* Notifier chain MUST detach us from master device. */
3136 BUG_TRAP(!dev->master);
3138 free_divert_blk(dev);
3140 /* Finish processing unregister after unlock */
3150 * unregister_netdev - remove device from the kernel
3153 * This function shuts down a device interface and removes it
3154 * from the kernel tables. On success 0 is returned, on a failure
3155 * a negative errno code is returned.
3157 * This is just a wrapper for unregister_netdevice that takes
3158 * the rtnl semaphore. In general you want to use this and not
3159 * unregister_netdevice.
3161 void unregister_netdev(struct net_device *dev)
3164 unregister_netdevice(dev);
3168 EXPORT_SYMBOL(unregister_netdev);
3170 #ifdef CONFIG_HOTPLUG_CPU
3171 static int dev_cpu_callback(struct notifier_block *nfb,
3172 unsigned long action,
3175 struct sk_buff **list_skb;
3176 struct net_device **list_net;
3177 struct sk_buff *skb;
3178 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3179 struct softnet_data *sd, *oldsd;
3181 if (action != CPU_DEAD)
3184 local_irq_disable();
3185 cpu = smp_processor_id();
3186 sd = &per_cpu(softnet_data, cpu);
3187 oldsd = &per_cpu(softnet_data, oldcpu);
3189 /* Find end of our completion_queue. */
3190 list_skb = &sd->completion_queue;
3192 list_skb = &(*list_skb)->next;
3193 /* Append completion queue from offline CPU. */
3194 *list_skb = oldsd->completion_queue;
3195 oldsd->completion_queue = NULL;
3197 /* Find end of our output_queue. */
3198 list_net = &sd->output_queue;
3200 list_net = &(*list_net)->next_sched;
3201 /* Append output queue from offline CPU. */
3202 *list_net = oldsd->output_queue;
3203 oldsd->output_queue = NULL;
3205 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3208 /* Process offline CPU's input_pkt_queue */
3209 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3214 #endif /* CONFIG_HOTPLUG_CPU */
3218 * Initialize the DEV module. At boot time this walks the device list and
3219 * unhooks any devices that fail to initialise (normally hardware not
3220 * present) and leaves us with a valid list of present and active devices.
3225 * This is called single threaded during boot, so no need
3226 * to take the rtnl semaphore.
3228 static int __init net_dev_init(void)
3230 int i, rc = -ENOMEM;
3232 BUG_ON(!dev_boot_phase);
3236 if (dev_proc_init())
3239 if (netdev_sysfs_init())
3242 INIT_LIST_HEAD(&ptype_all);
3243 for (i = 0; i < 16; i++)
3244 INIT_LIST_HEAD(&ptype_base[i]);
3246 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3247 INIT_HLIST_HEAD(&dev_name_head[i]);
3249 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3250 INIT_HLIST_HEAD(&dev_index_head[i]);
3253 * Initialise the packet receive queues.
3257 struct softnet_data *queue;
3259 queue = &per_cpu(softnet_data, i);
3260 skb_queue_head_init(&queue->input_pkt_queue);
3261 queue->completion_queue = NULL;
3262 INIT_LIST_HEAD(&queue->poll_list);
3263 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3264 queue->backlog_dev.weight = weight_p;
3265 queue->backlog_dev.poll = process_backlog;
3266 atomic_set(&queue->backlog_dev.refcnt, 1);
3271 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3272 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3274 hotcpu_notifier(dev_cpu_callback, 0);
3282 subsys_initcall(net_dev_init);
3284 EXPORT_SYMBOL(__dev_get_by_index);
3285 EXPORT_SYMBOL(__dev_get_by_name);
3286 EXPORT_SYMBOL(__dev_remove_pack);
3287 EXPORT_SYMBOL(__skb_linearize);
3288 EXPORT_SYMBOL(dev_valid_name);
3289 EXPORT_SYMBOL(dev_add_pack);
3290 EXPORT_SYMBOL(dev_alloc_name);
3291 EXPORT_SYMBOL(dev_close);
3292 EXPORT_SYMBOL(dev_get_by_flags);
3293 EXPORT_SYMBOL(dev_get_by_index);
3294 EXPORT_SYMBOL(dev_get_by_name);
3295 EXPORT_SYMBOL(dev_open);
3296 EXPORT_SYMBOL(dev_queue_xmit);
3297 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
3298 EXPORT_SYMBOL(dev_queue_xmit_nit);
3300 EXPORT_SYMBOL(dev_remove_pack);
3301 EXPORT_SYMBOL(dev_set_allmulti);
3302 EXPORT_SYMBOL(dev_set_promiscuity);
3303 EXPORT_SYMBOL(dev_change_flags);
3304 EXPORT_SYMBOL(dev_set_mtu);
3305 EXPORT_SYMBOL(dev_set_mac_address);
3306 EXPORT_SYMBOL(free_netdev);
3307 EXPORT_SYMBOL(netdev_boot_setup_check);
3308 EXPORT_SYMBOL(netdev_set_master);
3309 EXPORT_SYMBOL(netdev_state_change);
3310 EXPORT_SYMBOL(netif_receive_skb);
3311 EXPORT_SYMBOL(netif_rx);
3312 EXPORT_SYMBOL(register_gifconf);
3313 EXPORT_SYMBOL(register_netdevice);
3314 EXPORT_SYMBOL(register_netdevice_notifier);
3315 EXPORT_SYMBOL(skb_checksum_help);
3316 EXPORT_SYMBOL(synchronize_net);
3317 EXPORT_SYMBOL(unregister_netdevice);
3318 EXPORT_SYMBOL(unregister_netdevice_notifier);
3319 EXPORT_SYMBOL(net_enable_timestamp);
3320 EXPORT_SYMBOL(net_disable_timestamp);
3321 EXPORT_SYMBOL(dev_get_flags);
3323 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3324 EXPORT_SYMBOL(br_handle_frame_hook);
3325 EXPORT_SYMBOL(br_fdb_get_hook);
3326 EXPORT_SYMBOL(br_fdb_put_hook);
3330 EXPORT_SYMBOL(dev_load);
3333 EXPORT_PER_CPU_SYMBOL(softnet_data);