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/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
97 #include <linux/rtnetlink.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <linux/stat.h>
101 #include <linux/if_bridge.h>
102 #include <linux/divert.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/kmod.h>
109 #include <linux/module.h>
110 #include <linux/kallsyms.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <linux/wireless.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/err.h>
119 #include <linux/vs_network.h>
123 #include <linux/tcp.h>
124 #include <linux/udp.h>
128 * The list of packet types we will receive (as opposed to discard)
129 * and the routines to invoke.
131 * Why 16. Because with 16 the only overlap we get on a hash of the
132 * low nibble of the protocol value is RARP/SNAP/X.25.
134 * NOTE: That is no longer true with the addition of VLAN tags. Not
135 * sure which should go first, but I bet it won't make much
136 * difference if we are running VLANs. The good news is that
137 * this protocol won't be in the list unless compiled in, so
138 * the average user (w/out VLANs) will not be adversely affected.
155 static DEFINE_SPINLOCK(ptype_lock);
156 static struct list_head ptype_base[16]; /* 16 way hashed list */
157 static struct list_head ptype_all; /* Taps */
160 * The @dev_base list is protected by @dev_base_lock and the rtnl
163 * Pure readers hold dev_base_lock for reading.
165 * Writers must hold the rtnl semaphore while they loop through the
166 * dev_base list, and hold dev_base_lock for writing when they do the
167 * actual updates. This allows pure readers to access the list even
168 * while a writer is preparing to update it.
170 * To put it another way, dev_base_lock is held for writing only to
171 * protect against pure readers; the rtnl semaphore provides the
172 * protection against other writers.
174 * See, for example usages, register_netdevice() and
175 * unregister_netdevice(), which must be called with the rtnl
178 struct net_device *dev_base;
179 static struct net_device **dev_tail = &dev_base;
180 DEFINE_RWLOCK(dev_base_lock);
182 EXPORT_SYMBOL(dev_base);
183 EXPORT_SYMBOL(dev_base_lock);
185 #define NETDEV_HASHBITS 8
186 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
187 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
189 static inline struct hlist_head *dev_name_hash(const char *name)
191 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
192 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
195 static inline struct hlist_head *dev_index_hash(int ifindex)
197 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
204 static RAW_NOTIFIER_HEAD(netdev_chain);
207 * Device drivers call our routines to queue packets here. We empty the
208 * queue in the local softnet handler.
210 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
213 extern int netdev_sysfs_init(void);
214 extern int netdev_register_sysfs(struct net_device *);
215 extern void netdev_unregister_sysfs(struct net_device *);
217 #define netdev_sysfs_init() (0)
218 #define netdev_register_sysfs(dev) (0)
219 #define netdev_unregister_sysfs(dev) do { } while(0)
223 /*******************************************************************************
225 Protocol management and registration routines
227 *******************************************************************************/
236 * Add a protocol ID to the list. Now that the input handler is
237 * smarter we can dispense with all the messy stuff that used to be
240 * BEWARE!!! Protocol handlers, mangling input packets,
241 * MUST BE last in hash buckets and checking protocol handlers
242 * MUST start from promiscuous ptype_all chain in net_bh.
243 * It is true now, do not change it.
244 * Explanation follows: if protocol handler, mangling packet, will
245 * be the first on list, it is not able to sense, that packet
246 * is cloned and should be copied-on-write, so that it will
247 * change it and subsequent readers will get broken packet.
252 * dev_add_pack - add packet handler
253 * @pt: packet type declaration
255 * Add a protocol handler to the networking stack. The passed &packet_type
256 * is linked into kernel lists and may not be freed until it has been
257 * removed from the kernel lists.
259 * This call does not sleep therefore it can not
260 * guarantee all CPU's that are in middle of receiving packets
261 * will see the new packet type (until the next received packet).
264 void dev_add_pack(struct packet_type *pt)
268 spin_lock_bh(&ptype_lock);
269 if (pt->type == htons(ETH_P_ALL)) {
271 list_add_rcu(&pt->list, &ptype_all);
273 hash = ntohs(pt->type) & 15;
274 list_add_rcu(&pt->list, &ptype_base[hash]);
276 spin_unlock_bh(&ptype_lock);
280 * __dev_remove_pack - remove packet handler
281 * @pt: packet type declaration
283 * Remove a protocol handler that was previously added to the kernel
284 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
285 * from the kernel lists and can be freed or reused once this function
288 * The packet type might still be in use by receivers
289 * and must not be freed until after all the CPU's have gone
290 * through a quiescent state.
292 void __dev_remove_pack(struct packet_type *pt)
294 struct list_head *head;
295 struct packet_type *pt1;
297 spin_lock_bh(&ptype_lock);
299 if (pt->type == htons(ETH_P_ALL)) {
303 head = &ptype_base[ntohs(pt->type) & 15];
305 list_for_each_entry(pt1, head, list) {
307 list_del_rcu(&pt->list);
312 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
314 spin_unlock_bh(&ptype_lock);
317 * dev_remove_pack - remove packet handler
318 * @pt: packet type declaration
320 * Remove a protocol handler that was previously added to the kernel
321 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
322 * from the kernel lists and can be freed or reused once this function
325 * This call sleeps to guarantee that no CPU is looking at the packet
328 void dev_remove_pack(struct packet_type *pt)
330 __dev_remove_pack(pt);
335 /******************************************************************************
337 Device Boot-time Settings Routines
339 *******************************************************************************/
341 /* Boot time configuration table */
342 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
345 * netdev_boot_setup_add - add new setup entry
346 * @name: name of the device
347 * @map: configured settings for the device
349 * Adds new setup entry to the dev_boot_setup list. The function
350 * returns 0 on error and 1 on success. This is a generic routine to
353 static int netdev_boot_setup_add(char *name, struct ifmap *map)
355 struct netdev_boot_setup *s;
359 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
360 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
361 memset(s[i].name, 0, sizeof(s[i].name));
362 strcpy(s[i].name, name);
363 memcpy(&s[i].map, map, sizeof(s[i].map));
368 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
372 * netdev_boot_setup_check - check boot time settings
373 * @dev: the netdevice
375 * Check boot time settings for the device.
376 * The found settings are set for the device to be used
377 * later in the device probing.
378 * Returns 0 if no settings found, 1 if they are.
380 int netdev_boot_setup_check(struct net_device *dev)
382 struct netdev_boot_setup *s = dev_boot_setup;
385 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
386 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
387 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
388 dev->irq = s[i].map.irq;
389 dev->base_addr = s[i].map.base_addr;
390 dev->mem_start = s[i].map.mem_start;
391 dev->mem_end = s[i].map.mem_end;
400 * netdev_boot_base - get address from boot time settings
401 * @prefix: prefix for network device
402 * @unit: id for network device
404 * Check boot time settings for the base address of device.
405 * The found settings are set for the device to be used
406 * later in the device probing.
407 * Returns 0 if no settings found.
409 unsigned long netdev_boot_base(const char *prefix, int unit)
411 const struct netdev_boot_setup *s = dev_boot_setup;
415 sprintf(name, "%s%d", prefix, unit);
418 * If device already registered then return base of 1
419 * to indicate not to probe for this interface
421 if (__dev_get_by_name(name))
424 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
425 if (!strcmp(name, s[i].name))
426 return s[i].map.base_addr;
431 * Saves at boot time configured settings for any netdevice.
433 int __init netdev_boot_setup(char *str)
438 str = get_options(str, ARRAY_SIZE(ints), ints);
443 memset(&map, 0, sizeof(map));
447 map.base_addr = ints[2];
449 map.mem_start = ints[3];
451 map.mem_end = ints[4];
453 /* Add new entry to the list */
454 return netdev_boot_setup_add(str, &map);
457 __setup("netdev=", netdev_boot_setup);
459 /*******************************************************************************
461 Device Interface Subroutines
463 *******************************************************************************/
466 * __dev_get_by_name - find a device by its name
467 * @name: name to find
469 * Find an interface by name. Must be called under RTNL semaphore
470 * or @dev_base_lock. If the name is found a pointer to the device
471 * is returned. If the name is not found then %NULL is returned. The
472 * reference counters are not incremented so the caller must be
473 * careful with locks.
476 struct net_device *__dev_get_by_name(const char *name)
478 struct hlist_node *p;
480 hlist_for_each(p, dev_name_hash(name)) {
481 struct net_device *dev
482 = hlist_entry(p, struct net_device, name_hlist);
483 if (!strncmp(dev->name, name, IFNAMSIZ))
490 * dev_get_by_name - find a device by its name
491 * @name: name to find
493 * Find an interface by name. This can be called from any
494 * context and does its own locking. The returned handle has
495 * the usage count incremented and the caller must use dev_put() to
496 * release it when it is no longer needed. %NULL is returned if no
497 * matching device is found.
500 struct net_device *dev_get_by_name(const char *name)
502 struct net_device *dev;
504 read_lock(&dev_base_lock);
505 dev = __dev_get_by_name(name);
508 read_unlock(&dev_base_lock);
513 * __dev_get_by_index - find a device by its ifindex
514 * @ifindex: index of device
516 * Search for an interface by index. Returns %NULL if the device
517 * is not found or a pointer to the device. The device has not
518 * had its reference counter increased so the caller must be careful
519 * about locking. The caller must hold either the RTNL semaphore
523 struct net_device *__dev_get_by_index(int ifindex)
525 struct hlist_node *p;
527 hlist_for_each(p, dev_index_hash(ifindex)) {
528 struct net_device *dev
529 = hlist_entry(p, struct net_device, index_hlist);
530 if (dev->ifindex == ifindex)
538 * dev_get_by_index - find a device by its ifindex
539 * @ifindex: index of device
541 * Search for an interface by index. Returns NULL if the device
542 * is not found or a pointer to the device. The device returned has
543 * had a reference added and the pointer is safe until the user calls
544 * dev_put to indicate they have finished with it.
547 struct net_device *dev_get_by_index(int ifindex)
549 struct net_device *dev;
551 read_lock(&dev_base_lock);
552 dev = __dev_get_by_index(ifindex);
555 read_unlock(&dev_base_lock);
560 * dev_getbyhwaddr - find a device by its hardware address
561 * @type: media type of device
562 * @ha: hardware address
564 * Search for an interface by MAC address. Returns NULL if the device
565 * is not found or a pointer to the device. The caller must hold the
566 * rtnl semaphore. The returned device has not had its ref count increased
567 * and the caller must therefore be careful about locking
570 * If the API was consistent this would be __dev_get_by_hwaddr
573 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
575 struct net_device *dev;
579 for (dev = dev_base; dev; dev = dev->next)
580 if (dev->type == type &&
581 !memcmp(dev->dev_addr, ha, dev->addr_len))
586 EXPORT_SYMBOL(dev_getbyhwaddr);
588 struct net_device *dev_getfirstbyhwtype(unsigned short type)
590 struct net_device *dev;
593 for (dev = dev_base; dev; dev = dev->next) {
594 if (dev->type == type) {
603 EXPORT_SYMBOL(dev_getfirstbyhwtype);
606 * dev_get_by_flags - find any device with given flags
607 * @if_flags: IFF_* values
608 * @mask: bitmask of bits in if_flags to check
610 * Search for any interface with the given flags. Returns NULL if a device
611 * is not found or a pointer to the device. The device returned has
612 * had a reference added and the pointer is safe until the user calls
613 * dev_put to indicate they have finished with it.
616 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
618 struct net_device *dev;
620 read_lock(&dev_base_lock);
621 for (dev = dev_base; dev != NULL; dev = dev->next) {
622 if (((dev->flags ^ if_flags) & mask) == 0) {
627 read_unlock(&dev_base_lock);
632 * dev_valid_name - check if name is okay for network device
635 * Network device names need to be valid file names to
636 * to allow sysfs to work
638 int dev_valid_name(const char *name)
640 return !(*name == '\0'
641 || !strcmp(name, ".")
642 || !strcmp(name, "..")
643 || strchr(name, '/'));
647 * dev_alloc_name - allocate a name for a device
649 * @name: name format string
651 * Passed a format string - eg "lt%d" it will try and find a suitable
652 * id. It scans list of devices to build up a free map, then chooses
653 * the first empty slot. The caller must hold the dev_base or rtnl lock
654 * while allocating the name and adding the device in order to avoid
656 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
657 * Returns the number of the unit assigned or a negative errno code.
660 int dev_alloc_name(struct net_device *dev, const char *name)
665 const int max_netdevices = 8*PAGE_SIZE;
667 struct net_device *d;
669 p = strnchr(name, IFNAMSIZ-1, '%');
672 * Verify the string as this thing may have come from
673 * the user. There must be either one "%d" and no other "%"
676 if (p[1] != 'd' || strchr(p + 2, '%'))
679 /* Use one page as a bit array of possible slots */
680 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
684 for (d = dev_base; d; d = d->next) {
685 if (!sscanf(d->name, name, &i))
687 if (i < 0 || i >= max_netdevices)
690 /* avoid cases where sscanf is not exact inverse of printf */
691 snprintf(buf, sizeof(buf), name, i);
692 if (!strncmp(buf, d->name, IFNAMSIZ))
696 i = find_first_zero_bit(inuse, max_netdevices);
697 free_page((unsigned long) inuse);
700 snprintf(buf, sizeof(buf), name, i);
701 if (!__dev_get_by_name(buf)) {
702 strlcpy(dev->name, buf, IFNAMSIZ);
706 /* It is possible to run out of possible slots
707 * when the name is long and there isn't enough space left
708 * for the digits, or if all bits are used.
715 * dev_change_name - change name of a device
717 * @newname: name (or format string) must be at least IFNAMSIZ
719 * Change name of a device, can pass format strings "eth%d".
722 int dev_change_name(struct net_device *dev, char *newname)
728 if (dev->flags & IFF_UP)
731 if (!dev_valid_name(newname))
734 if (strchr(newname, '%')) {
735 err = dev_alloc_name(dev, newname);
738 strcpy(newname, dev->name);
740 else if (__dev_get_by_name(newname))
743 strlcpy(dev->name, newname, IFNAMSIZ);
745 err = class_device_rename(&dev->class_dev, dev->name);
747 hlist_del(&dev->name_hlist);
748 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
749 raw_notifier_call_chain(&netdev_chain,
750 NETDEV_CHANGENAME, dev);
757 * netdev_features_change - device changes features
758 * @dev: device to cause notification
760 * Called to indicate a device has changed features.
762 void netdev_features_change(struct net_device *dev)
764 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
766 EXPORT_SYMBOL(netdev_features_change);
769 * netdev_state_change - device changes state
770 * @dev: device to cause notification
772 * Called to indicate a device has changed state. This function calls
773 * the notifier chains for netdev_chain and sends a NEWLINK message
774 * to the routing socket.
776 void netdev_state_change(struct net_device *dev)
778 if (dev->flags & IFF_UP) {
779 raw_notifier_call_chain(&netdev_chain,
781 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
786 * dev_load - load a network module
787 * @name: name of interface
789 * If a network interface is not present and the process has suitable
790 * privileges this function loads the module. If module loading is not
791 * available in this kernel then it becomes a nop.
794 void dev_load(const char *name)
796 struct net_device *dev;
798 read_lock(&dev_base_lock);
799 dev = __dev_get_by_name(name);
800 read_unlock(&dev_base_lock);
802 if (!dev && capable(CAP_SYS_MODULE))
803 request_module("%s", name);
806 static int default_rebuild_header(struct sk_buff *skb)
808 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
809 skb->dev ? skb->dev->name : "NULL!!!");
816 * dev_open - prepare an interface for use.
817 * @dev: device to open
819 * Takes a device from down to up state. The device's private open
820 * function is invoked and then the multicast lists are loaded. Finally
821 * the device is moved into the up state and a %NETDEV_UP message is
822 * sent to the netdev notifier chain.
824 * Calling this function on an active interface is a nop. On a failure
825 * a negative errno code is returned.
827 int dev_open(struct net_device *dev)
835 if (dev->flags & IFF_UP)
839 * Is it even present?
841 if (!netif_device_present(dev))
845 * Call device private open method
847 set_bit(__LINK_STATE_START, &dev->state);
849 ret = dev->open(dev);
851 clear_bit(__LINK_STATE_START, &dev->state);
855 * If it went open OK then:
862 dev->flags |= IFF_UP;
865 * Initialize multicasting status
870 * Wakeup transmit queue engine
875 * ... and announce new interface.
877 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
883 * dev_close - shutdown an interface.
884 * @dev: device to shutdown
886 * This function moves an active device into down state. A
887 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
888 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
891 int dev_close(struct net_device *dev)
893 if (!(dev->flags & IFF_UP))
897 * Tell people we are going down, so that they can
898 * prepare to death, when device is still operating.
900 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
904 clear_bit(__LINK_STATE_START, &dev->state);
906 /* Synchronize to scheduled poll. We cannot touch poll list,
907 * it can be even on different cpu. So just clear netif_running(),
908 * and wait when poll really will happen. Actually, the best place
909 * for this is inside dev->stop() after device stopped its irq
910 * engine, but this requires more changes in devices. */
912 smp_mb__after_clear_bit(); /* Commit netif_running(). */
913 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
919 * Call the device specific close. This cannot fail.
920 * Only if device is UP
922 * We allow it to be called even after a DETACH hot-plug
929 * Device is now down.
932 dev->flags &= ~IFF_UP;
935 * Tell people we are down
937 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
944 * Device change register/unregister. These are not inline or static
945 * as we export them to the world.
949 * register_netdevice_notifier - register a network notifier block
952 * Register a notifier to be called when network device events occur.
953 * The notifier passed is linked into the kernel structures and must
954 * not be reused until it has been unregistered. A negative errno code
955 * is returned on a failure.
957 * When registered all registration and up events are replayed
958 * to the new notifier to allow device to have a race free
959 * view of the network device list.
962 int register_netdevice_notifier(struct notifier_block *nb)
964 struct net_device *dev;
968 err = raw_notifier_chain_register(&netdev_chain, nb);
970 for (dev = dev_base; dev; dev = dev->next) {
971 nb->notifier_call(nb, NETDEV_REGISTER, dev);
973 if (dev->flags & IFF_UP)
974 nb->notifier_call(nb, NETDEV_UP, dev);
982 * unregister_netdevice_notifier - unregister a network notifier block
985 * Unregister a notifier previously registered by
986 * register_netdevice_notifier(). The notifier is unlinked into the
987 * kernel structures and may then be reused. A negative errno code
988 * is returned on a failure.
991 int unregister_netdevice_notifier(struct notifier_block *nb)
996 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1002 * call_netdevice_notifiers - call all network notifier blocks
1003 * @val: value passed unmodified to notifier function
1004 * @v: pointer passed unmodified to notifier function
1006 * Call all network notifier blocks. Parameters and return value
1007 * are as for raw_notifier_call_chain().
1010 int call_netdevice_notifiers(unsigned long val, void *v)
1012 return raw_notifier_call_chain(&netdev_chain, val, v);
1015 /* When > 0 there are consumers of rx skb time stamps */
1016 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1018 void net_enable_timestamp(void)
1020 atomic_inc(&netstamp_needed);
1023 void net_disable_timestamp(void)
1025 atomic_dec(&netstamp_needed);
1028 void __net_timestamp(struct sk_buff *skb)
1032 do_gettimeofday(&tv);
1033 skb_set_timestamp(skb, &tv);
1035 EXPORT_SYMBOL(__net_timestamp);
1037 static inline void net_timestamp(struct sk_buff *skb)
1039 if (atomic_read(&netstamp_needed))
1040 __net_timestamp(skb);
1042 skb->tstamp.off_sec = 0;
1043 skb->tstamp.off_usec = 0;
1048 * Support routine. Sends outgoing frames to any network
1049 * taps currently in use.
1052 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1054 struct packet_type *ptype;
1059 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1060 /* Never send packets back to the socket
1061 * they originated from - MvS (miquels@drinkel.ow.org)
1063 if ((ptype->dev == dev || !ptype->dev) &&
1064 (ptype->af_packet_priv == NULL ||
1065 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1066 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1070 /* skb->nh should be correctly
1071 set by sender, so that the second statement is
1072 just protection against buggy protocols.
1074 skb2->mac.raw = skb2->data;
1076 if (skb2->nh.raw < skb2->data ||
1077 skb2->nh.raw > skb2->tail) {
1078 if (net_ratelimit())
1079 printk(KERN_CRIT "protocol %04x is "
1081 skb2->protocol, dev->name);
1082 skb2->nh.raw = skb2->data;
1085 skb2->h.raw = skb2->nh.raw;
1086 skb2->pkt_type = PACKET_OUTGOING;
1087 ptype->func(skb2, skb->dev, ptype, skb->dev);
1094 void __netif_schedule(struct net_device *dev)
1096 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1097 unsigned long flags;
1098 struct softnet_data *sd;
1100 local_irq_save(flags);
1101 sd = &__get_cpu_var(softnet_data);
1102 dev->next_sched = sd->output_queue;
1103 sd->output_queue = dev;
1104 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1105 local_irq_restore(flags);
1108 EXPORT_SYMBOL(__netif_schedule);
1110 void __netif_rx_schedule(struct net_device *dev)
1112 unsigned long flags;
1114 local_irq_save(flags);
1116 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1118 dev->quota += dev->weight;
1120 dev->quota = dev->weight;
1121 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1122 local_irq_restore(flags);
1124 EXPORT_SYMBOL(__netif_rx_schedule);
1126 void dev_kfree_skb_any(struct sk_buff *skb)
1128 if (in_irq() || irqs_disabled())
1129 dev_kfree_skb_irq(skb);
1133 EXPORT_SYMBOL(dev_kfree_skb_any);
1137 void netif_device_detach(struct net_device *dev)
1139 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1140 netif_running(dev)) {
1141 netif_stop_queue(dev);
1144 EXPORT_SYMBOL(netif_device_detach);
1146 void netif_device_attach(struct net_device *dev)
1148 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1149 netif_running(dev)) {
1150 netif_wake_queue(dev);
1151 __netdev_watchdog_up(dev);
1154 EXPORT_SYMBOL(netif_device_attach);
1158 * Invalidate hardware checksum when packet is to be mangled, and
1159 * complete checksum manually on outgoing path.
1161 int skb_checksum_help(struct sk_buff *skb, int inward)
1164 int ret = 0, offset = skb->h.raw - skb->data;
1167 skb->ip_summed = CHECKSUM_NONE;
1171 if (skb_cloned(skb)) {
1172 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1177 BUG_ON(offset > (int)skb->len);
1178 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1180 offset = skb->tail - skb->h.raw;
1181 BUG_ON(offset <= 0);
1182 BUG_ON(skb->csum + 2 > offset);
1184 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1185 skb->ip_summed = CHECKSUM_NONE;
1191 * skb_gso_segment - Perform segmentation on skb.
1192 * @skb: buffer to segment
1193 * @features: features for the output path (see dev->features)
1195 * This function segments the given skb and returns a list of segments.
1197 * It may return NULL if the skb requires no segmentation. This is
1198 * only possible when GSO is used for verifying header integrity.
1200 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1202 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1203 struct packet_type *ptype;
1204 int type = skb->protocol;
1206 BUG_ON(skb_shinfo(skb)->frag_list);
1207 BUG_ON(skb->ip_summed != CHECKSUM_HW);
1209 skb->mac.raw = skb->data;
1210 skb->mac_len = skb->nh.raw - skb->data;
1211 __skb_pull(skb, skb->mac_len);
1214 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1215 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1216 segs = ptype->gso_segment(skb, features);
1222 __skb_push(skb, skb->data - skb->mac.raw);
1227 EXPORT_SYMBOL(skb_gso_segment);
1229 /* Take action when hardware reception checksum errors are detected. */
1231 void netdev_rx_csum_fault(struct net_device *dev)
1233 if (net_ratelimit()) {
1234 printk(KERN_ERR "%s: hw csum failure.\n",
1235 dev ? dev->name : "<unknown>");
1239 EXPORT_SYMBOL(netdev_rx_csum_fault);
1242 #ifdef CONFIG_HIGHMEM
1243 /* Actually, we should eliminate this check as soon as we know, that:
1244 * 1. IOMMU is present and allows to map all the memory.
1245 * 2. No high memory really exists on this machine.
1248 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1252 if (dev->features & NETIF_F_HIGHDMA)
1255 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1256 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1262 #define illegal_highdma(dev, skb) (0)
1266 void (*destructor)(struct sk_buff *skb);
1269 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1271 static void dev_gso_skb_destructor(struct sk_buff *skb)
1273 struct dev_gso_cb *cb;
1276 struct sk_buff *nskb = skb->next;
1278 skb->next = nskb->next;
1281 } while (skb->next);
1283 cb = DEV_GSO_CB(skb);
1285 cb->destructor(skb);
1289 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1290 * @skb: buffer to segment
1292 * This function segments the given skb and stores the list of segments
1295 static int dev_gso_segment(struct sk_buff *skb)
1297 struct net_device *dev = skb->dev;
1298 struct sk_buff *segs;
1299 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1302 segs = skb_gso_segment(skb, features);
1304 /* Verifying header integrity only. */
1308 if (unlikely(IS_ERR(segs)))
1309 return PTR_ERR(segs);
1312 DEV_GSO_CB(skb)->destructor = skb->destructor;
1313 skb->destructor = dev_gso_skb_destructor;
1318 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1320 if (likely(!skb->next)) {
1322 dev_queue_xmit_nit(skb, dev);
1324 if (netif_needs_gso(dev, skb)) {
1325 if (unlikely(dev_gso_segment(skb)))
1331 return dev->hard_start_xmit(skb, dev);
1336 struct sk_buff *nskb = skb->next;
1339 skb->next = nskb->next;
1341 rc = dev->hard_start_xmit(nskb, dev);
1343 nskb->next = skb->next;
1347 if (unlikely(netif_queue_stopped(dev) && skb->next))
1348 return NETDEV_TX_BUSY;
1349 } while (skb->next);
1351 skb->destructor = DEV_GSO_CB(skb)->destructor;
1358 #define HARD_TX_LOCK(dev, cpu) { \
1359 if ((dev->features & NETIF_F_LLTX) == 0) { \
1360 netif_tx_lock(dev); \
1364 #define HARD_TX_UNLOCK(dev) { \
1365 if ((dev->features & NETIF_F_LLTX) == 0) { \
1366 netif_tx_unlock(dev); \
1371 inline int skb_checksum_setup(struct sk_buff *skb)
1373 if (skb->proto_csum_blank) {
1374 if (skb->protocol != htons(ETH_P_IP))
1376 skb->h.raw = (unsigned char *)skb->nh.iph + 4*skb->nh.iph->ihl;
1377 if (skb->h.raw >= skb->tail)
1379 switch (skb->nh.iph->protocol) {
1381 skb->csum = offsetof(struct tcphdr, check);
1384 skb->csum = offsetof(struct udphdr, check);
1387 if (net_ratelimit())
1388 printk(KERN_ERR "Attempting to checksum a non-"
1389 "TCP/UDP packet, dropping a protocol"
1390 " %d packet", skb->nh.iph->protocol);
1393 if ((skb->h.raw + skb->csum + 2) > skb->tail)
1395 skb->ip_summed = CHECKSUM_HW;
1396 skb->proto_csum_blank = 0;
1403 inline int skb_checksum_setup(struct sk_buff *skb) { return 0; }
1408 * dev_queue_xmit - transmit a buffer
1409 * @skb: buffer to transmit
1411 * Queue a buffer for transmission to a network device. The caller must
1412 * have set the device and priority and built the buffer before calling
1413 * this function. The function can be called from an interrupt.
1415 * A negative errno code is returned on a failure. A success does not
1416 * guarantee the frame will be transmitted as it may be dropped due
1417 * to congestion or traffic shaping.
1419 * -----------------------------------------------------------------------------------
1420 * I notice this method can also return errors from the queue disciplines,
1421 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1424 * Regardless of the return value, the skb is consumed, so it is currently
1425 * difficult to retry a send to this method. (You can bump the ref count
1426 * before sending to hold a reference for retry if you are careful.)
1428 * When calling this method, interrupts MUST be enabled. This is because
1429 * the BH enable code must have IRQs enabled so that it will not deadlock.
1433 int dev_queue_xmit(struct sk_buff *skb)
1435 struct net_device *dev = skb->dev;
1439 /* If a checksum-deferred packet is forwarded to a device that needs a
1440 * checksum, correct the pointers and force checksumming.
1442 if (skb_checksum_setup(skb))
1445 /* GSO will handle the following emulations directly. */
1446 if (netif_needs_gso(dev, skb))
1449 if (skb_shinfo(skb)->frag_list &&
1450 !(dev->features & NETIF_F_FRAGLIST) &&
1451 __skb_linearize(skb))
1454 /* Fragmented skb is linearized if device does not support SG,
1455 * or if at least one of fragments is in highmem and device
1456 * does not support DMA from it.
1458 if (skb_shinfo(skb)->nr_frags &&
1459 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1460 __skb_linearize(skb))
1463 /* If packet is not checksummed and device does not support
1464 * checksumming for this protocol, complete checksumming here.
1466 if (skb->ip_summed == CHECKSUM_HW &&
1467 (!(dev->features & NETIF_F_GEN_CSUM) &&
1468 (!(dev->features & NETIF_F_IP_CSUM) ||
1469 skb->protocol != htons(ETH_P_IP))))
1470 if (skb_checksum_help(skb, 0))
1474 spin_lock_prefetch(&dev->queue_lock);
1476 /* Disable soft irqs for various locks below. Also
1477 * stops preemption for RCU.
1481 /* Updates of qdisc are serialized by queue_lock.
1482 * The struct Qdisc which is pointed to by qdisc is now a
1483 * rcu structure - it may be accessed without acquiring
1484 * a lock (but the structure may be stale.) The freeing of the
1485 * qdisc will be deferred until it's known that there are no
1486 * more references to it.
1488 * If the qdisc has an enqueue function, we still need to
1489 * hold the queue_lock before calling it, since queue_lock
1490 * also serializes access to the device queue.
1493 q = rcu_dereference(dev->qdisc);
1494 #ifdef CONFIG_NET_CLS_ACT
1495 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1498 /* Grab device queue */
1499 spin_lock(&dev->queue_lock);
1501 rc = q->enqueue(skb, q);
1505 spin_unlock(&dev->queue_lock);
1506 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1510 /* The device has no queue. Common case for software devices:
1511 loopback, all the sorts of tunnels...
1513 Really, it is unlikely that netif_tx_lock protection is necessary
1514 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1516 However, it is possible, that they rely on protection
1519 Check this and shot the lock. It is not prone from deadlocks.
1520 Either shot noqueue qdisc, it is even simpler 8)
1522 if (dev->flags & IFF_UP) {
1523 int cpu = smp_processor_id(); /* ok because BHs are off */
1525 if (dev->xmit_lock_owner != cpu) {
1527 HARD_TX_LOCK(dev, cpu);
1529 if (!netif_queue_stopped(dev)) {
1531 if (!dev_hard_start_xmit(skb, dev)) {
1532 HARD_TX_UNLOCK(dev);
1536 HARD_TX_UNLOCK(dev);
1537 if (net_ratelimit())
1538 printk(KERN_CRIT "Virtual device %s asks to "
1539 "queue packet!\n", dev->name);
1541 /* Recursion is detected! It is possible,
1543 if (net_ratelimit())
1544 printk(KERN_CRIT "Dead loop on virtual device "
1545 "%s, fix it urgently!\n", dev->name);
1550 rcu_read_unlock_bh();
1556 rcu_read_unlock_bh();
1561 /*=======================================================================
1563 =======================================================================*/
1565 int netdev_max_backlog = 1000;
1566 int netdev_budget = 300;
1567 int weight_p = 64; /* old backlog weight */
1569 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1573 * netif_rx - post buffer to the network code
1574 * @skb: buffer to post
1576 * This function receives a packet from a device driver and queues it for
1577 * the upper (protocol) levels to process. It always succeeds. The buffer
1578 * may be dropped during processing for congestion control or by the
1582 * NET_RX_SUCCESS (no congestion)
1583 * NET_RX_CN_LOW (low congestion)
1584 * NET_RX_CN_MOD (moderate congestion)
1585 * NET_RX_CN_HIGH (high congestion)
1586 * NET_RX_DROP (packet was dropped)
1590 int netif_rx(struct sk_buff *skb)
1592 struct softnet_data *queue;
1593 unsigned long flags;
1595 /* if netpoll wants it, pretend we never saw it */
1596 if (netpoll_rx(skb))
1599 if (!skb->tstamp.off_sec)
1603 * The code is rearranged so that the path is the most
1604 * short when CPU is congested, but is still operating.
1606 local_irq_save(flags);
1607 queue = &__get_cpu_var(softnet_data);
1609 __get_cpu_var(netdev_rx_stat).total++;
1610 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1611 if (queue->input_pkt_queue.qlen) {
1614 __skb_queue_tail(&queue->input_pkt_queue, skb);
1615 local_irq_restore(flags);
1616 return NET_RX_SUCCESS;
1619 netif_rx_schedule(&queue->backlog_dev);
1623 __get_cpu_var(netdev_rx_stat).dropped++;
1624 local_irq_restore(flags);
1630 int netif_rx_ni(struct sk_buff *skb)
1635 err = netif_rx(skb);
1636 if (local_softirq_pending())
1643 EXPORT_SYMBOL(netif_rx_ni);
1645 static inline struct net_device *skb_bond(struct sk_buff *skb)
1647 struct net_device *dev = skb->dev;
1651 * On bonding slaves other than the currently active
1652 * slave, suppress duplicates except for 802.3ad
1653 * ETH_P_SLOW and alb non-mcast/bcast.
1655 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1656 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1657 if (skb->pkt_type != PACKET_BROADCAST &&
1658 skb->pkt_type != PACKET_MULTICAST)
1662 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1663 skb->protocol == __constant_htons(ETH_P_SLOW))
1670 skb->dev = dev->master;
1676 static void net_tx_action(struct softirq_action *h)
1678 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1680 if (sd->completion_queue) {
1681 struct sk_buff *clist;
1683 local_irq_disable();
1684 clist = sd->completion_queue;
1685 sd->completion_queue = NULL;
1689 struct sk_buff *skb = clist;
1690 clist = clist->next;
1692 BUG_TRAP(!atomic_read(&skb->users));
1697 if (sd->output_queue) {
1698 struct net_device *head;
1700 local_irq_disable();
1701 head = sd->output_queue;
1702 sd->output_queue = NULL;
1706 struct net_device *dev = head;
1707 head = head->next_sched;
1709 smp_mb__before_clear_bit();
1710 clear_bit(__LINK_STATE_SCHED, &dev->state);
1712 if (spin_trylock(&dev->queue_lock)) {
1714 spin_unlock(&dev->queue_lock);
1716 netif_schedule(dev);
1722 static __inline__ int deliver_skb(struct sk_buff *skb,
1723 struct packet_type *pt_prev,
1724 struct net_device *orig_dev)
1726 atomic_inc(&skb->users);
1727 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1730 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1731 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1733 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1734 unsigned char *addr);
1735 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1737 static __inline__ int handle_bridge(struct sk_buff **pskb,
1738 struct packet_type **pt_prev, int *ret,
1739 struct net_device *orig_dev)
1741 struct net_bridge_port *port;
1743 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1744 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1748 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1752 return br_handle_frame_hook(port, pskb);
1755 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1758 #ifdef CONFIG_NET_CLS_ACT
1759 /* TODO: Maybe we should just force sch_ingress to be compiled in
1760 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1761 * a compare and 2 stores extra right now if we dont have it on
1762 * but have CONFIG_NET_CLS_ACT
1763 * NOTE: This doesnt stop any functionality; if you dont have
1764 * the ingress scheduler, you just cant add policies on ingress.
1767 static int ing_filter(struct sk_buff *skb)
1770 struct net_device *dev = skb->dev;
1771 int result = TC_ACT_OK;
1773 if (dev->qdisc_ingress) {
1774 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1775 if (MAX_RED_LOOP < ttl++) {
1776 printk("Redir loop detected Dropping packet (%s->%s)\n",
1777 skb->input_dev->name, skb->dev->name);
1781 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1783 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1785 spin_lock(&dev->ingress_lock);
1786 if ((q = dev->qdisc_ingress) != NULL)
1787 result = q->enqueue(skb, q);
1788 spin_unlock(&dev->ingress_lock);
1796 int netif_receive_skb(struct sk_buff *skb)
1798 struct packet_type *ptype, *pt_prev;
1799 struct net_device *orig_dev;
1800 int ret = NET_RX_DROP;
1801 unsigned short type;
1803 /* if we've gotten here through NAPI, check netpoll */
1804 if (skb->dev->poll && netpoll_rx(skb))
1807 if (!skb->tstamp.off_sec)
1810 if (!skb->input_dev)
1811 skb->input_dev = skb->dev;
1813 orig_dev = skb_bond(skb);
1818 __get_cpu_var(netdev_rx_stat).total++;
1820 skb->h.raw = skb->nh.raw = skb->data;
1821 skb->mac_len = skb->nh.raw - skb->mac.raw;
1827 #ifdef CONFIG_NET_CLS_ACT
1828 if (skb->tc_verd & TC_NCLS) {
1829 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1835 switch (skb->ip_summed) {
1836 case CHECKSUM_UNNECESSARY:
1837 skb->proto_data_valid = 1;
1840 /* XXX Implement me. */
1842 skb->proto_data_valid = 0;
1847 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1848 if (!ptype->dev || ptype->dev == skb->dev) {
1850 ret = deliver_skb(skb, pt_prev, orig_dev);
1855 #ifdef CONFIG_NET_CLS_ACT
1857 ret = deliver_skb(skb, pt_prev, orig_dev);
1858 pt_prev = NULL; /* noone else should process this after*/
1860 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1863 ret = ing_filter(skb);
1865 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1874 handle_diverter(skb);
1876 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1879 type = skb->protocol;
1880 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1881 if (ptype->type == type &&
1882 (!ptype->dev || ptype->dev == skb->dev)) {
1884 ret = deliver_skb(skb, pt_prev, orig_dev);
1890 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1893 /* Jamal, now you will not able to escape explaining
1894 * me how you were going to use this. :-)
1904 static int process_backlog(struct net_device *backlog_dev, int *budget)
1907 int quota = min(backlog_dev->quota, *budget);
1908 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1909 unsigned long start_time = jiffies;
1911 backlog_dev->weight = weight_p;
1913 struct sk_buff *skb;
1914 struct net_device *dev;
1916 local_irq_disable();
1917 skb = __skb_dequeue(&queue->input_pkt_queue);
1924 netif_receive_skb(skb);
1930 if (work >= quota || jiffies - start_time > 1)
1935 backlog_dev->quota -= work;
1940 backlog_dev->quota -= work;
1943 list_del(&backlog_dev->poll_list);
1944 smp_mb__before_clear_bit();
1945 netif_poll_enable(backlog_dev);
1951 static void net_rx_action(struct softirq_action *h)
1953 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1954 unsigned long start_time = jiffies;
1955 int budget = netdev_budget;
1958 local_irq_disable();
1960 while (!list_empty(&queue->poll_list)) {
1961 struct net_device *dev;
1963 if (budget <= 0 || jiffies - start_time > 1)
1968 dev = list_entry(queue->poll_list.next,
1969 struct net_device, poll_list);
1970 have = netpoll_poll_lock(dev);
1972 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1973 netpoll_poll_unlock(have);
1974 local_irq_disable();
1975 list_move_tail(&dev->poll_list, &queue->poll_list);
1977 dev->quota += dev->weight;
1979 dev->quota = dev->weight;
1981 netpoll_poll_unlock(have);
1983 local_irq_disable();
1991 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1992 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1996 static gifconf_func_t * gifconf_list [NPROTO];
1999 * register_gifconf - register a SIOCGIF handler
2000 * @family: Address family
2001 * @gifconf: Function handler
2003 * Register protocol dependent address dumping routines. The handler
2004 * that is passed must not be freed or reused until it has been replaced
2005 * by another handler.
2007 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2009 if (family >= NPROTO)
2011 gifconf_list[family] = gifconf;
2017 * Map an interface index to its name (SIOCGIFNAME)
2021 * We need this ioctl for efficient implementation of the
2022 * if_indextoname() function required by the IPv6 API. Without
2023 * it, we would have to search all the interfaces to find a
2027 static int dev_ifname(struct ifreq __user *arg)
2029 struct net_device *dev;
2033 * Fetch the caller's info block.
2036 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2039 read_lock(&dev_base_lock);
2040 dev = __dev_get_by_index(ifr.ifr_ifindex);
2042 read_unlock(&dev_base_lock);
2046 strcpy(ifr.ifr_name, dev->name);
2047 read_unlock(&dev_base_lock);
2049 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2055 * Perform a SIOCGIFCONF call. This structure will change
2056 * size eventually, and there is nothing I can do about it.
2057 * Thus we will need a 'compatibility mode'.
2060 static int dev_ifconf(char __user *arg)
2063 struct net_device *dev;
2070 * Fetch the caller's info block.
2073 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2080 * Loop over the interfaces, and write an info block for each.
2084 for (dev = dev_base; dev; dev = dev->next) {
2085 if (vx_flags(VXF_HIDE_NETIF, 0) &&
2086 !dev_in_nx_info(dev, current->nx_info))
2088 for (i = 0; i < NPROTO; i++) {
2089 if (gifconf_list[i]) {
2092 done = gifconf_list[i](dev, NULL, 0);
2094 done = gifconf_list[i](dev, pos + total,
2104 * All done. Write the updated control block back to the caller.
2106 ifc.ifc_len = total;
2109 * Both BSD and Solaris return 0 here, so we do too.
2111 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2114 #ifdef CONFIG_PROC_FS
2116 * This is invoked by the /proc filesystem handler to display a device
2119 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2121 struct net_device *dev;
2124 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2126 return i == pos ? dev : NULL;
2129 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2131 read_lock(&dev_base_lock);
2132 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2135 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2138 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2141 void dev_seq_stop(struct seq_file *seq, void *v)
2143 read_unlock(&dev_base_lock);
2146 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2148 struct nx_info *nxi = current->nx_info;
2150 if (vx_flags(VXF_HIDE_NETIF, 0) && !dev_in_nx_info(dev, nxi))
2152 if (dev->get_stats) {
2153 struct net_device_stats *stats = dev->get_stats(dev);
2155 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2156 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2157 dev->name, stats->rx_bytes, stats->rx_packets,
2159 stats->rx_dropped + stats->rx_missed_errors,
2160 stats->rx_fifo_errors,
2161 stats->rx_length_errors + stats->rx_over_errors +
2162 stats->rx_crc_errors + stats->rx_frame_errors,
2163 stats->rx_compressed, stats->multicast,
2164 stats->tx_bytes, stats->tx_packets,
2165 stats->tx_errors, stats->tx_dropped,
2166 stats->tx_fifo_errors, stats->collisions,
2167 stats->tx_carrier_errors +
2168 stats->tx_aborted_errors +
2169 stats->tx_window_errors +
2170 stats->tx_heartbeat_errors,
2171 stats->tx_compressed);
2173 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2177 * Called from the PROCfs module. This now uses the new arbitrary sized
2178 * /proc/net interface to create /proc/net/dev
2180 static int dev_seq_show(struct seq_file *seq, void *v)
2182 if (v == SEQ_START_TOKEN)
2183 seq_puts(seq, "Inter-| Receive "
2185 " face |bytes packets errs drop fifo frame "
2186 "compressed multicast|bytes packets errs "
2187 "drop fifo colls carrier compressed\n");
2189 dev_seq_printf_stats(seq, v);
2193 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2195 struct netif_rx_stats *rc = NULL;
2197 while (*pos < NR_CPUS)
2198 if (cpu_online(*pos)) {
2199 rc = &per_cpu(netdev_rx_stat, *pos);
2206 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2208 return softnet_get_online(pos);
2211 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2214 return softnet_get_online(pos);
2217 static void softnet_seq_stop(struct seq_file *seq, void *v)
2221 static int softnet_seq_show(struct seq_file *seq, void *v)
2223 struct netif_rx_stats *s = v;
2225 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2226 s->total, s->dropped, s->time_squeeze, 0,
2227 0, 0, 0, 0, /* was fastroute */
2232 static struct seq_operations dev_seq_ops = {
2233 .start = dev_seq_start,
2234 .next = dev_seq_next,
2235 .stop = dev_seq_stop,
2236 .show = dev_seq_show,
2239 static int dev_seq_open(struct inode *inode, struct file *file)
2241 return seq_open(file, &dev_seq_ops);
2244 static struct file_operations dev_seq_fops = {
2245 .owner = THIS_MODULE,
2246 .open = dev_seq_open,
2248 .llseek = seq_lseek,
2249 .release = seq_release,
2252 static struct seq_operations softnet_seq_ops = {
2253 .start = softnet_seq_start,
2254 .next = softnet_seq_next,
2255 .stop = softnet_seq_stop,
2256 .show = softnet_seq_show,
2259 static int softnet_seq_open(struct inode *inode, struct file *file)
2261 return seq_open(file, &softnet_seq_ops);
2264 static struct file_operations softnet_seq_fops = {
2265 .owner = THIS_MODULE,
2266 .open = softnet_seq_open,
2268 .llseek = seq_lseek,
2269 .release = seq_release,
2272 #ifdef CONFIG_WIRELESS_EXT
2273 extern int wireless_proc_init(void);
2275 #define wireless_proc_init() 0
2278 static int __init dev_proc_init(void)
2282 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2284 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2286 if (wireless_proc_init())
2292 proc_net_remove("softnet_stat");
2294 proc_net_remove("dev");
2298 #define dev_proc_init() 0
2299 #endif /* CONFIG_PROC_FS */
2303 * netdev_set_master - set up master/slave pair
2304 * @slave: slave device
2305 * @master: new master device
2307 * Changes the master device of the slave. Pass %NULL to break the
2308 * bonding. The caller must hold the RTNL semaphore. On a failure
2309 * a negative errno code is returned. On success the reference counts
2310 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2311 * function returns zero.
2313 int netdev_set_master(struct net_device *slave, struct net_device *master)
2315 struct net_device *old = slave->master;
2325 slave->master = master;
2333 slave->flags |= IFF_SLAVE;
2335 slave->flags &= ~IFF_SLAVE;
2337 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2342 * dev_set_promiscuity - update promiscuity count on a device
2346 * Add or remove promiscuity from a device. While the count in the device
2347 * remains above zero the interface remains promiscuous. Once it hits zero
2348 * the device reverts back to normal filtering operation. A negative inc
2349 * value is used to drop promiscuity on the device.
2351 void dev_set_promiscuity(struct net_device *dev, int inc)
2353 unsigned short old_flags = dev->flags;
2355 if ((dev->promiscuity += inc) == 0)
2356 dev->flags &= ~IFF_PROMISC;
2358 dev->flags |= IFF_PROMISC;
2359 if (dev->flags != old_flags) {
2361 printk(KERN_INFO "device %s %s promiscuous mode\n",
2362 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2364 audit_log(current->audit_context, GFP_ATOMIC,
2365 AUDIT_ANOM_PROMISCUOUS,
2366 "dev=%s prom=%d old_prom=%d auid=%u",
2367 dev->name, (dev->flags & IFF_PROMISC),
2368 (old_flags & IFF_PROMISC),
2369 audit_get_loginuid(current->audit_context));
2374 * dev_set_allmulti - update allmulti count on a device
2378 * Add or remove reception of all multicast frames to a device. While the
2379 * count in the device remains above zero the interface remains listening
2380 * to all interfaces. Once it hits zero the device reverts back to normal
2381 * filtering operation. A negative @inc value is used to drop the counter
2382 * when releasing a resource needing all multicasts.
2385 void dev_set_allmulti(struct net_device *dev, int inc)
2387 unsigned short old_flags = dev->flags;
2389 dev->flags |= IFF_ALLMULTI;
2390 if ((dev->allmulti += inc) == 0)
2391 dev->flags &= ~IFF_ALLMULTI;
2392 if (dev->flags ^ old_flags)
2396 unsigned dev_get_flags(const struct net_device *dev)
2400 flags = (dev->flags & ~(IFF_PROMISC |
2405 (dev->gflags & (IFF_PROMISC |
2408 if (netif_running(dev)) {
2409 if (netif_oper_up(dev))
2410 flags |= IFF_RUNNING;
2411 if (netif_carrier_ok(dev))
2412 flags |= IFF_LOWER_UP;
2413 if (netif_dormant(dev))
2414 flags |= IFF_DORMANT;
2420 int dev_change_flags(struct net_device *dev, unsigned flags)
2423 int old_flags = dev->flags;
2426 * Set the flags on our device.
2429 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2430 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2432 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2436 * Load in the correct multicast list now the flags have changed.
2442 * Have we downed the interface. We handle IFF_UP ourselves
2443 * according to user attempts to set it, rather than blindly
2448 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2449 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2455 if (dev->flags & IFF_UP &&
2456 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2458 raw_notifier_call_chain(&netdev_chain,
2459 NETDEV_CHANGE, dev);
2461 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2462 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2463 dev->gflags ^= IFF_PROMISC;
2464 dev_set_promiscuity(dev, inc);
2467 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2468 is important. Some (broken) drivers set IFF_PROMISC, when
2469 IFF_ALLMULTI is requested not asking us and not reporting.
2471 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2472 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2473 dev->gflags ^= IFF_ALLMULTI;
2474 dev_set_allmulti(dev, inc);
2477 if (old_flags ^ dev->flags)
2478 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2483 int dev_set_mtu(struct net_device *dev, int new_mtu)
2487 if (new_mtu == dev->mtu)
2490 /* MTU must be positive. */
2494 if (!netif_device_present(dev))
2498 if (dev->change_mtu)
2499 err = dev->change_mtu(dev, new_mtu);
2502 if (!err && dev->flags & IFF_UP)
2503 raw_notifier_call_chain(&netdev_chain,
2504 NETDEV_CHANGEMTU, dev);
2508 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2512 if (!dev->set_mac_address)
2514 if (sa->sa_family != dev->type)
2516 if (!netif_device_present(dev))
2518 err = dev->set_mac_address(dev, sa);
2520 raw_notifier_call_chain(&netdev_chain,
2521 NETDEV_CHANGEADDR, dev);
2526 * Perform the SIOCxIFxxx calls.
2528 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2531 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2537 case SIOCGIFFLAGS: /* Get interface flags */
2538 ifr->ifr_flags = dev_get_flags(dev);
2541 case SIOCSIFFLAGS: /* Set interface flags */
2542 return dev_change_flags(dev, ifr->ifr_flags);
2544 case SIOCGIFMETRIC: /* Get the metric on the interface
2545 (currently unused) */
2546 ifr->ifr_metric = 0;
2549 case SIOCSIFMETRIC: /* Set the metric on the interface
2550 (currently unused) */
2553 case SIOCGIFMTU: /* Get the MTU of a device */
2554 ifr->ifr_mtu = dev->mtu;
2557 case SIOCSIFMTU: /* Set the MTU of a device */
2558 return dev_set_mtu(dev, ifr->ifr_mtu);
2562 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2564 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2565 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2566 ifr->ifr_hwaddr.sa_family = dev->type;
2570 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2572 case SIOCSIFHWBROADCAST:
2573 if (ifr->ifr_hwaddr.sa_family != dev->type)
2575 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2576 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2577 raw_notifier_call_chain(&netdev_chain,
2578 NETDEV_CHANGEADDR, dev);
2582 ifr->ifr_map.mem_start = dev->mem_start;
2583 ifr->ifr_map.mem_end = dev->mem_end;
2584 ifr->ifr_map.base_addr = dev->base_addr;
2585 ifr->ifr_map.irq = dev->irq;
2586 ifr->ifr_map.dma = dev->dma;
2587 ifr->ifr_map.port = dev->if_port;
2591 if (dev->set_config) {
2592 if (!netif_device_present(dev))
2594 return dev->set_config(dev, &ifr->ifr_map);
2599 if (!dev->set_multicast_list ||
2600 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2602 if (!netif_device_present(dev))
2604 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2608 if (!dev->set_multicast_list ||
2609 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2611 if (!netif_device_present(dev))
2613 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2617 ifr->ifr_ifindex = dev->ifindex;
2621 ifr->ifr_qlen = dev->tx_queue_len;
2625 if (ifr->ifr_qlen < 0)
2627 dev->tx_queue_len = ifr->ifr_qlen;
2631 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2632 return dev_change_name(dev, ifr->ifr_newname);
2635 * Unknown or private ioctl
2639 if ((cmd >= SIOCDEVPRIVATE &&
2640 cmd <= SIOCDEVPRIVATE + 15) ||
2641 cmd == SIOCBONDENSLAVE ||
2642 cmd == SIOCBONDRELEASE ||
2643 cmd == SIOCBONDSETHWADDR ||
2644 cmd == SIOCBONDSLAVEINFOQUERY ||
2645 cmd == SIOCBONDINFOQUERY ||
2646 cmd == SIOCBONDCHANGEACTIVE ||
2647 cmd == SIOCGMIIPHY ||
2648 cmd == SIOCGMIIREG ||
2649 cmd == SIOCSMIIREG ||
2650 cmd == SIOCBRADDIF ||
2651 cmd == SIOCBRDELIF ||
2652 cmd == SIOCWANDEV) {
2654 if (dev->do_ioctl) {
2655 if (netif_device_present(dev))
2656 err = dev->do_ioctl(dev, ifr,
2669 * This function handles all "interface"-type I/O control requests. The actual
2670 * 'doing' part of this is dev_ifsioc above.
2674 * dev_ioctl - network device ioctl
2675 * @cmd: command to issue
2676 * @arg: pointer to a struct ifreq in user space
2678 * Issue ioctl functions to devices. This is normally called by the
2679 * user space syscall interfaces but can sometimes be useful for
2680 * other purposes. The return value is the return from the syscall if
2681 * positive or a negative errno code on error.
2684 int dev_ioctl(unsigned int cmd, void __user *arg)
2690 /* One special case: SIOCGIFCONF takes ifconf argument
2691 and requires shared lock, because it sleeps writing
2695 if (cmd == SIOCGIFCONF) {
2697 ret = dev_ifconf((char __user *) arg);
2701 if (cmd == SIOCGIFNAME)
2702 return dev_ifname((struct ifreq __user *)arg);
2704 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2707 ifr.ifr_name[IFNAMSIZ-1] = 0;
2709 colon = strchr(ifr.ifr_name, ':');
2714 * See which interface the caller is talking about.
2719 * These ioctl calls:
2720 * - can be done by all.
2721 * - atomic and do not require locking.
2732 dev_load(ifr.ifr_name);
2733 read_lock(&dev_base_lock);
2734 ret = dev_ifsioc(&ifr, cmd);
2735 read_unlock(&dev_base_lock);
2739 if (copy_to_user(arg, &ifr,
2740 sizeof(struct ifreq)))
2746 dev_load(ifr.ifr_name);
2748 ret = dev_ethtool(&ifr);
2753 if (copy_to_user(arg, &ifr,
2754 sizeof(struct ifreq)))
2760 * These ioctl calls:
2761 * - require superuser power.
2762 * - require strict serialization.
2768 if (!capable(CAP_NET_ADMIN))
2770 dev_load(ifr.ifr_name);
2772 ret = dev_ifsioc(&ifr, cmd);
2777 if (copy_to_user(arg, &ifr,
2778 sizeof(struct ifreq)))
2784 * These ioctl calls:
2785 * - require superuser power.
2786 * - require strict serialization.
2787 * - do not return a value
2797 case SIOCSIFHWBROADCAST:
2800 case SIOCBONDENSLAVE:
2801 case SIOCBONDRELEASE:
2802 case SIOCBONDSETHWADDR:
2803 case SIOCBONDCHANGEACTIVE:
2806 if (!capable(CAP_NET_ADMIN))
2809 case SIOCBONDSLAVEINFOQUERY:
2810 case SIOCBONDINFOQUERY:
2811 dev_load(ifr.ifr_name);
2813 ret = dev_ifsioc(&ifr, cmd);
2818 /* Get the per device memory space. We can add this but
2819 * currently do not support it */
2821 /* Set the per device memory buffer space.
2822 * Not applicable in our case */
2827 * Unknown or private ioctl.
2830 if (cmd == SIOCWANDEV ||
2831 (cmd >= SIOCDEVPRIVATE &&
2832 cmd <= SIOCDEVPRIVATE + 15)) {
2833 dev_load(ifr.ifr_name);
2835 ret = dev_ifsioc(&ifr, cmd);
2837 if (!ret && copy_to_user(arg, &ifr,
2838 sizeof(struct ifreq)))
2842 #ifdef CONFIG_WIRELESS_EXT
2843 /* Take care of Wireless Extensions */
2844 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2845 /* If command is `set a parameter', or
2846 * `get the encoding parameters', check if
2847 * the user has the right to do it */
2848 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2849 || cmd == SIOCGIWENCODEEXT) {
2850 if (!capable(CAP_NET_ADMIN))
2853 dev_load(ifr.ifr_name);
2855 /* Follow me in net/core/wireless.c */
2856 ret = wireless_process_ioctl(&ifr, cmd);
2858 if (IW_IS_GET(cmd) &&
2859 copy_to_user(arg, &ifr,
2860 sizeof(struct ifreq)))
2864 #endif /* CONFIG_WIRELESS_EXT */
2871 * dev_new_index - allocate an ifindex
2873 * Returns a suitable unique value for a new device interface
2874 * number. The caller must hold the rtnl semaphore or the
2875 * dev_base_lock to be sure it remains unique.
2877 static int dev_new_index(void)
2883 if (!__dev_get_by_index(ifindex))
2888 static int dev_boot_phase = 1;
2890 /* Delayed registration/unregisteration */
2891 static DEFINE_SPINLOCK(net_todo_list_lock);
2892 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2894 static inline void net_set_todo(struct net_device *dev)
2896 spin_lock(&net_todo_list_lock);
2897 list_add_tail(&dev->todo_list, &net_todo_list);
2898 spin_unlock(&net_todo_list_lock);
2902 * register_netdevice - register a network device
2903 * @dev: device to register
2905 * Take a completed network device structure and add it to the kernel
2906 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2907 * chain. 0 is returned on success. A negative errno code is returned
2908 * on a failure to set up the device, or if the name is a duplicate.
2910 * Callers must hold the rtnl semaphore. You may want
2911 * register_netdev() instead of this.
2914 * The locking appears insufficient to guarantee two parallel registers
2915 * will not get the same name.
2918 int register_netdevice(struct net_device *dev)
2920 struct hlist_head *head;
2921 struct hlist_node *p;
2924 BUG_ON(dev_boot_phase);
2929 /* When net_device's are persistent, this will be fatal. */
2930 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2932 spin_lock_init(&dev->queue_lock);
2933 spin_lock_init(&dev->_xmit_lock);
2934 dev->xmit_lock_owner = -1;
2935 #ifdef CONFIG_NET_CLS_ACT
2936 spin_lock_init(&dev->ingress_lock);
2939 ret = alloc_divert_blk(dev);
2945 /* Init, if this function is available */
2947 ret = dev->init(dev);
2955 if (!dev_valid_name(dev->name)) {
2960 dev->ifindex = dev_new_index();
2961 if (dev->iflink == -1)
2962 dev->iflink = dev->ifindex;
2964 /* Check for existence of name */
2965 head = dev_name_hash(dev->name);
2966 hlist_for_each(p, head) {
2967 struct net_device *d
2968 = hlist_entry(p, struct net_device, name_hlist);
2969 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2975 /* Fix illegal SG+CSUM combinations. */
2976 if ((dev->features & NETIF_F_SG) &&
2977 !(dev->features & NETIF_F_ALL_CSUM)) {
2978 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2980 dev->features &= ~NETIF_F_SG;
2983 /* TSO requires that SG is present as well. */
2984 if ((dev->features & NETIF_F_TSO) &&
2985 !(dev->features & NETIF_F_SG)) {
2986 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2988 dev->features &= ~NETIF_F_TSO;
2990 if (dev->features & NETIF_F_UFO) {
2991 if (!(dev->features & NETIF_F_HW_CSUM)) {
2992 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2993 "NETIF_F_HW_CSUM feature.\n",
2995 dev->features &= ~NETIF_F_UFO;
2997 if (!(dev->features & NETIF_F_SG)) {
2998 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2999 "NETIF_F_SG feature.\n",
3001 dev->features &= ~NETIF_F_UFO;
3006 * nil rebuild_header routine,
3007 * that should be never called and used as just bug trap.
3010 if (!dev->rebuild_header)
3011 dev->rebuild_header = default_rebuild_header;
3013 ret = netdev_register_sysfs(dev);
3016 dev->reg_state = NETREG_REGISTERED;
3019 * Default initial state at registry is that the
3020 * device is present.
3023 set_bit(__LINK_STATE_PRESENT, &dev->state);
3026 dev_init_scheduler(dev);
3027 write_lock_bh(&dev_base_lock);
3029 dev_tail = &dev->next;
3030 hlist_add_head(&dev->name_hlist, head);
3031 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3033 write_unlock_bh(&dev_base_lock);
3035 /* Notify protocols, that a new device appeared. */
3036 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3043 free_divert_blk(dev);
3048 * register_netdev - register a network device
3049 * @dev: device to register
3051 * Take a completed network device structure and add it to the kernel
3052 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3053 * chain. 0 is returned on success. A negative errno code is returned
3054 * on a failure to set up the device, or if the name is a duplicate.
3056 * This is a wrapper around register_netdev that takes the rtnl semaphore
3057 * and expands the device name if you passed a format string to
3060 int register_netdev(struct net_device *dev)
3067 * If the name is a format string the caller wants us to do a
3070 if (strchr(dev->name, '%')) {
3071 err = dev_alloc_name(dev, dev->name);
3077 * Back compatibility hook. Kill this one in 2.5
3079 if (dev->name[0] == 0 || dev->name[0] == ' ') {
3080 err = dev_alloc_name(dev, "eth%d");
3085 err = register_netdevice(dev);
3090 EXPORT_SYMBOL(register_netdev);
3093 * netdev_wait_allrefs - wait until all references are gone.
3095 * This is called when unregistering network devices.
3097 * Any protocol or device that holds a reference should register
3098 * for netdevice notification, and cleanup and put back the
3099 * reference if they receive an UNREGISTER event.
3100 * We can get stuck here if buggy protocols don't correctly
3103 static void netdev_wait_allrefs(struct net_device *dev)
3105 unsigned long rebroadcast_time, warning_time;
3107 rebroadcast_time = warning_time = jiffies;
3108 while (atomic_read(&dev->refcnt) != 0) {
3109 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3112 /* Rebroadcast unregister notification */
3113 raw_notifier_call_chain(&netdev_chain,
3114 NETDEV_UNREGISTER, dev);
3116 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3118 /* We must not have linkwatch events
3119 * pending on unregister. If this
3120 * happens, we simply run the queue
3121 * unscheduled, resulting in a noop
3124 linkwatch_run_queue();
3129 rebroadcast_time = jiffies;
3134 if (time_after(jiffies, warning_time + 10 * HZ)) {
3135 printk(KERN_EMERG "unregister_netdevice: "
3136 "waiting for %s to become free. Usage "
3138 dev->name, atomic_read(&dev->refcnt));
3139 warning_time = jiffies;
3148 * register_netdevice(x1);
3149 * register_netdevice(x2);
3151 * unregister_netdevice(y1);
3152 * unregister_netdevice(y2);
3158 * We are invoked by rtnl_unlock() after it drops the semaphore.
3159 * This allows us to deal with problems:
3160 * 1) We can delete sysfs objects which invoke hotplug
3161 * without deadlocking with linkwatch via keventd.
3162 * 2) Since we run with the RTNL semaphore not held, we can sleep
3163 * safely in order to wait for the netdev refcnt to drop to zero.
3165 static DEFINE_MUTEX(net_todo_run_mutex);
3166 void netdev_run_todo(void)
3168 struct list_head list = LIST_HEAD_INIT(list);
3170 /* Need to guard against multiple cpu's getting out of order. */
3171 mutex_lock(&net_todo_run_mutex);
3173 /* Not safe to do outside the semaphore. We must not return
3174 * until all unregister events invoked by the local processor
3175 * have been completed (either by this todo run, or one on
3178 if (list_empty(&net_todo_list))
3181 /* Snapshot list, allow later requests */
3182 spin_lock(&net_todo_list_lock);
3183 list_splice_init(&net_todo_list, &list);
3184 spin_unlock(&net_todo_list_lock);
3186 while (!list_empty(&list)) {
3187 struct net_device *dev
3188 = list_entry(list.next, struct net_device, todo_list);
3189 list_del(&dev->todo_list);
3191 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3192 printk(KERN_ERR "network todo '%s' but state %d\n",
3193 dev->name, dev->reg_state);
3198 netdev_unregister_sysfs(dev);
3199 dev->reg_state = NETREG_UNREGISTERED;
3201 netdev_wait_allrefs(dev);
3204 BUG_ON(atomic_read(&dev->refcnt));
3205 BUG_TRAP(!dev->ip_ptr);
3206 BUG_TRAP(!dev->ip6_ptr);
3207 BUG_TRAP(!dev->dn_ptr);
3209 /* It must be the very last action,
3210 * after this 'dev' may point to freed up memory.
3212 if (dev->destructor)
3213 dev->destructor(dev);
3217 mutex_unlock(&net_todo_run_mutex);
3221 * alloc_netdev - allocate network device
3222 * @sizeof_priv: size of private data to allocate space for
3223 * @name: device name format string
3224 * @setup: callback to initialize device
3226 * Allocates a struct net_device with private data area for driver use
3227 * and performs basic initialization.
3229 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3230 void (*setup)(struct net_device *))
3233 struct net_device *dev;
3236 /* ensure 32-byte alignment of both the device and private area */
3237 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3238 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3240 p = kzalloc(alloc_size, GFP_KERNEL);
3242 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3246 dev = (struct net_device *)
3247 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3248 dev->padded = (char *)dev - (char *)p;
3251 dev->priv = netdev_priv(dev);
3254 strcpy(dev->name, name);
3257 EXPORT_SYMBOL(alloc_netdev);
3260 * free_netdev - free network device
3263 * This function does the last stage of destroying an allocated device
3264 * interface. The reference to the device object is released.
3265 * If this is the last reference then it will be freed.
3267 void free_netdev(struct net_device *dev)
3270 /* Compatibility with error handling in drivers */
3271 if (dev->reg_state == NETREG_UNINITIALIZED) {
3272 kfree((char *)dev - dev->padded);
3276 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3277 dev->reg_state = NETREG_RELEASED;
3279 /* will free via class release */
3280 class_device_put(&dev->class_dev);
3282 kfree((char *)dev - dev->padded);
3286 /* Synchronize with packet receive processing. */
3287 void synchronize_net(void)
3294 * unregister_netdevice - remove device from the kernel
3297 * This function shuts down a device interface and removes it
3298 * from the kernel tables. On success 0 is returned, on a failure
3299 * a negative errno code is returned.
3301 * Callers must hold the rtnl semaphore. You may want
3302 * unregister_netdev() instead of this.
3305 int unregister_netdevice(struct net_device *dev)
3307 struct net_device *d, **dp;
3309 BUG_ON(dev_boot_phase);
3312 /* Some devices call without registering for initialization unwind. */
3313 if (dev->reg_state == NETREG_UNINITIALIZED) {
3314 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3315 "was registered\n", dev->name, dev);
3319 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3321 /* If device is running, close it first. */
3322 if (dev->flags & IFF_UP)
3325 /* And unlink it from device chain. */
3326 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3328 write_lock_bh(&dev_base_lock);
3329 hlist_del(&dev->name_hlist);
3330 hlist_del(&dev->index_hlist);
3331 if (dev_tail == &dev->next)
3334 write_unlock_bh(&dev_base_lock);
3339 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3344 dev->reg_state = NETREG_UNREGISTERING;
3348 /* Shutdown queueing discipline. */
3352 /* Notify protocols, that we are about to destroy
3353 this device. They should clean all the things.
3355 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3358 * Flush the multicast chain
3360 dev_mc_discard(dev);
3365 /* Notifier chain MUST detach us from master device. */
3366 BUG_TRAP(!dev->master);
3368 free_divert_blk(dev);
3370 /* Finish processing unregister after unlock */
3380 * unregister_netdev - remove device from the kernel
3383 * This function shuts down a device interface and removes it
3384 * from the kernel tables. On success 0 is returned, on a failure
3385 * a negative errno code is returned.
3387 * This is just a wrapper for unregister_netdevice that takes
3388 * the rtnl semaphore. In general you want to use this and not
3389 * unregister_netdevice.
3391 void unregister_netdev(struct net_device *dev)
3394 unregister_netdevice(dev);
3398 EXPORT_SYMBOL(unregister_netdev);
3400 #ifdef CONFIG_HOTPLUG_CPU
3401 static int dev_cpu_callback(struct notifier_block *nfb,
3402 unsigned long action,
3405 struct sk_buff **list_skb;
3406 struct net_device **list_net;
3407 struct sk_buff *skb;
3408 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3409 struct softnet_data *sd, *oldsd;
3411 if (action != CPU_DEAD)
3414 local_irq_disable();
3415 cpu = smp_processor_id();
3416 sd = &per_cpu(softnet_data, cpu);
3417 oldsd = &per_cpu(softnet_data, oldcpu);
3419 /* Find end of our completion_queue. */
3420 list_skb = &sd->completion_queue;
3422 list_skb = &(*list_skb)->next;
3423 /* Append completion queue from offline CPU. */
3424 *list_skb = oldsd->completion_queue;
3425 oldsd->completion_queue = NULL;
3427 /* Find end of our output_queue. */
3428 list_net = &sd->output_queue;
3430 list_net = &(*list_net)->next_sched;
3431 /* Append output queue from offline CPU. */
3432 *list_net = oldsd->output_queue;
3433 oldsd->output_queue = NULL;
3435 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3438 /* Process offline CPU's input_pkt_queue */
3439 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3444 #endif /* CONFIG_HOTPLUG_CPU */
3448 * Initialize the DEV module. At boot time this walks the device list and
3449 * unhooks any devices that fail to initialise (normally hardware not
3450 * present) and leaves us with a valid list of present and active devices.
3455 * This is called single threaded during boot, so no need
3456 * to take the rtnl semaphore.
3458 static int __init net_dev_init(void)
3460 int i, rc = -ENOMEM;
3462 BUG_ON(!dev_boot_phase);
3466 if (dev_proc_init())
3469 if (netdev_sysfs_init())
3472 INIT_LIST_HEAD(&ptype_all);
3473 for (i = 0; i < 16; i++)
3474 INIT_LIST_HEAD(&ptype_base[i]);
3476 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3477 INIT_HLIST_HEAD(&dev_name_head[i]);
3479 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3480 INIT_HLIST_HEAD(&dev_index_head[i]);
3483 * Initialise the packet receive queues.
3486 for_each_possible_cpu(i) {
3487 struct softnet_data *queue;
3489 queue = &per_cpu(softnet_data, i);
3490 skb_queue_head_init(&queue->input_pkt_queue);
3491 queue->completion_queue = NULL;
3492 INIT_LIST_HEAD(&queue->poll_list);
3493 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3494 queue->backlog_dev.weight = weight_p;
3495 queue->backlog_dev.poll = process_backlog;
3496 atomic_set(&queue->backlog_dev.refcnt, 1);
3501 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3502 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3504 hotcpu_notifier(dev_cpu_callback, 0);
3512 subsys_initcall(net_dev_init);
3514 EXPORT_SYMBOL(__dev_get_by_index);
3515 EXPORT_SYMBOL(__dev_get_by_name);
3516 EXPORT_SYMBOL(__dev_remove_pack);
3517 EXPORT_SYMBOL(dev_valid_name);
3518 EXPORT_SYMBOL(dev_add_pack);
3519 EXPORT_SYMBOL(dev_alloc_name);
3520 EXPORT_SYMBOL(dev_close);
3521 EXPORT_SYMBOL(dev_get_by_flags);
3522 EXPORT_SYMBOL(dev_get_by_index);
3523 EXPORT_SYMBOL(dev_get_by_name);
3524 EXPORT_SYMBOL(dev_open);
3525 EXPORT_SYMBOL(dev_queue_xmit);
3526 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
3527 EXPORT_SYMBOL(dev_queue_xmit_nit);
3529 EXPORT_SYMBOL(dev_remove_pack);
3530 EXPORT_SYMBOL(dev_set_allmulti);
3531 EXPORT_SYMBOL(dev_set_promiscuity);
3532 EXPORT_SYMBOL(dev_change_flags);
3533 EXPORT_SYMBOL(dev_set_mtu);
3534 EXPORT_SYMBOL(dev_set_mac_address);
3535 EXPORT_SYMBOL(free_netdev);
3536 EXPORT_SYMBOL(netdev_boot_setup_check);
3537 EXPORT_SYMBOL(netdev_set_master);
3538 EXPORT_SYMBOL(netdev_state_change);
3539 EXPORT_SYMBOL(netif_receive_skb);
3540 EXPORT_SYMBOL(netif_rx);
3541 EXPORT_SYMBOL(register_gifconf);
3542 EXPORT_SYMBOL(register_netdevice);
3543 EXPORT_SYMBOL(register_netdevice_notifier);
3544 EXPORT_SYMBOL(skb_checksum_help);
3545 EXPORT_SYMBOL(synchronize_net);
3546 EXPORT_SYMBOL(unregister_netdevice);
3547 EXPORT_SYMBOL(unregister_netdevice_notifier);
3548 EXPORT_SYMBOL(net_enable_timestamp);
3549 EXPORT_SYMBOL(net_disable_timestamp);
3550 EXPORT_SYMBOL(dev_get_flags);
3551 EXPORT_SYMBOL(skb_checksum_setup);
3553 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3554 EXPORT_SYMBOL(br_handle_frame_hook);
3555 EXPORT_SYMBOL(br_fdb_get_hook);
3556 EXPORT_SYMBOL(br_fdb_put_hook);
3560 EXPORT_SYMBOL(dev_load);
3563 EXPORT_PER_CPU_SYMBOL(softnet_data);