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 #if !((defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)))
1055 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1057 struct packet_type *ptype;
1062 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1063 /* Never send packets back to the socket
1064 * they originated from - MvS (miquels@drinkel.ow.org)
1066 if ((ptype->dev == dev || !ptype->dev) &&
1067 (ptype->af_packet_priv == NULL ||
1068 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1069 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1073 /* skb->nh should be correctly
1074 set by sender, so that the second statement is
1075 just protection against buggy protocols.
1077 skb2->mac.raw = skb2->data;
1079 if (skb2->nh.raw < skb2->data ||
1080 skb2->nh.raw > skb2->tail) {
1081 if (net_ratelimit())
1082 printk(KERN_CRIT "protocol %04x is "
1084 skb2->protocol, dev->name);
1085 skb2->nh.raw = skb2->data;
1088 skb2->h.raw = skb2->nh.raw;
1089 skb2->pkt_type = PACKET_OUTGOING;
1090 ptype->func(skb2, skb->dev, ptype, skb->dev);
1097 void __netif_schedule(struct net_device *dev)
1099 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1100 unsigned long flags;
1101 struct softnet_data *sd;
1103 local_irq_save(flags);
1104 sd = &__get_cpu_var(softnet_data);
1105 dev->next_sched = sd->output_queue;
1106 sd->output_queue = dev;
1107 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1108 local_irq_restore(flags);
1111 EXPORT_SYMBOL(__netif_schedule);
1113 void __netif_rx_schedule(struct net_device *dev)
1115 unsigned long flags;
1117 local_irq_save(flags);
1119 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1121 dev->quota += dev->weight;
1123 dev->quota = dev->weight;
1124 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1125 local_irq_restore(flags);
1127 EXPORT_SYMBOL(__netif_rx_schedule);
1129 void dev_kfree_skb_any(struct sk_buff *skb)
1131 if (in_irq() || irqs_disabled())
1132 dev_kfree_skb_irq(skb);
1136 EXPORT_SYMBOL(dev_kfree_skb_any);
1140 void netif_device_detach(struct net_device *dev)
1142 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1143 netif_running(dev)) {
1144 netif_stop_queue(dev);
1147 EXPORT_SYMBOL(netif_device_detach);
1149 void netif_device_attach(struct net_device *dev)
1151 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1152 netif_running(dev)) {
1153 netif_wake_queue(dev);
1154 __netdev_watchdog_up(dev);
1157 EXPORT_SYMBOL(netif_device_attach);
1161 * Invalidate hardware checksum when packet is to be mangled, and
1162 * complete checksum manually on outgoing path.
1164 int skb_checksum_help(struct sk_buff *skb, int inward)
1167 int ret = 0, offset = skb->h.raw - skb->data;
1170 skb->ip_summed = CHECKSUM_NONE;
1174 if (skb_cloned(skb)) {
1175 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1180 BUG_ON(offset > (int)skb->len);
1181 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1183 offset = skb->tail - skb->h.raw;
1184 BUG_ON(offset <= 0);
1185 BUG_ON(skb->csum + 2 > offset);
1187 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1188 skb->ip_summed = CHECKSUM_NONE;
1194 * skb_gso_segment - Perform segmentation on skb.
1195 * @skb: buffer to segment
1196 * @features: features for the output path (see dev->features)
1198 * This function segments the given skb and returns a list of segments.
1200 * It may return NULL if the skb requires no segmentation. This is
1201 * only possible when GSO is used for verifying header integrity.
1203 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1205 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1206 struct packet_type *ptype;
1207 int type = skb->protocol;
1209 BUG_ON(skb_shinfo(skb)->frag_list);
1210 BUG_ON(skb->ip_summed != CHECKSUM_HW);
1212 skb->mac.raw = skb->data;
1213 skb->mac_len = skb->nh.raw - skb->data;
1214 __skb_pull(skb, skb->mac_len);
1217 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1218 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1219 segs = ptype->gso_segment(skb, features);
1225 __skb_push(skb, skb->data - skb->mac.raw);
1230 EXPORT_SYMBOL(skb_gso_segment);
1232 /* Take action when hardware reception checksum errors are detected. */
1234 void netdev_rx_csum_fault(struct net_device *dev)
1236 if (net_ratelimit()) {
1237 printk(KERN_ERR "%s: hw csum failure.\n",
1238 dev ? dev->name : "<unknown>");
1242 EXPORT_SYMBOL(netdev_rx_csum_fault);
1245 #ifdef CONFIG_HIGHMEM
1246 /* Actually, we should eliminate this check as soon as we know, that:
1247 * 1. IOMMU is present and allows to map all the memory.
1248 * 2. No high memory really exists on this machine.
1251 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1255 if (dev->features & NETIF_F_HIGHDMA)
1258 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1259 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1265 #define illegal_highdma(dev, skb) (0)
1269 void (*destructor)(struct sk_buff *skb);
1272 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1274 static void dev_gso_skb_destructor(struct sk_buff *skb)
1276 struct dev_gso_cb *cb;
1279 struct sk_buff *nskb = skb->next;
1281 skb->next = nskb->next;
1284 } while (skb->next);
1286 cb = DEV_GSO_CB(skb);
1288 cb->destructor(skb);
1292 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1293 * @skb: buffer to segment
1295 * This function segments the given skb and stores the list of segments
1298 static int dev_gso_segment(struct sk_buff *skb)
1300 struct net_device *dev = skb->dev;
1301 struct sk_buff *segs;
1302 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1305 segs = skb_gso_segment(skb, features);
1307 /* Verifying header integrity only. */
1311 if (unlikely(IS_ERR(segs)))
1312 return PTR_ERR(segs);
1315 DEV_GSO_CB(skb)->destructor = skb->destructor;
1316 skb->destructor = dev_gso_skb_destructor;
1321 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1323 if (likely(!skb->next)) {
1325 dev_queue_xmit_nit(skb, dev);
1327 if (netif_needs_gso(dev, skb)) {
1328 if (unlikely(dev_gso_segment(skb)))
1334 return dev->hard_start_xmit(skb, dev);
1339 struct sk_buff *nskb = skb->next;
1342 skb->next = nskb->next;
1344 rc = dev->hard_start_xmit(nskb, dev);
1346 nskb->next = skb->next;
1350 if (unlikely(netif_queue_stopped(dev) && skb->next))
1351 return NETDEV_TX_BUSY;
1352 } while (skb->next);
1354 skb->destructor = DEV_GSO_CB(skb)->destructor;
1361 #define HARD_TX_LOCK(dev, cpu) { \
1362 if ((dev->features & NETIF_F_LLTX) == 0) { \
1363 netif_tx_lock(dev); \
1367 #define HARD_TX_UNLOCK(dev) { \
1368 if ((dev->features & NETIF_F_LLTX) == 0) { \
1369 netif_tx_unlock(dev); \
1374 inline int skb_checksum_setup(struct sk_buff *skb)
1376 if (skb->proto_csum_blank) {
1377 if (skb->protocol != htons(ETH_P_IP))
1379 skb->h.raw = (unsigned char *)skb->nh.iph + 4*skb->nh.iph->ihl;
1380 if (skb->h.raw >= skb->tail)
1382 switch (skb->nh.iph->protocol) {
1384 skb->csum = offsetof(struct tcphdr, check);
1387 skb->csum = offsetof(struct udphdr, check);
1390 if (net_ratelimit())
1391 printk(KERN_ERR "Attempting to checksum a non-"
1392 "TCP/UDP packet, dropping a protocol"
1393 " %d packet", skb->nh.iph->protocol);
1396 if ((skb->h.raw + skb->csum + 2) > skb->tail)
1398 skb->ip_summed = CHECKSUM_HW;
1399 skb->proto_csum_blank = 0;
1406 inline int skb_checksum_setup(struct sk_buff *skb) { return 0; }
1411 * dev_queue_xmit - transmit a buffer
1412 * @skb: buffer to transmit
1414 * Queue a buffer for transmission to a network device. The caller must
1415 * have set the device and priority and built the buffer before calling
1416 * this function. The function can be called from an interrupt.
1418 * A negative errno code is returned on a failure. A success does not
1419 * guarantee the frame will be transmitted as it may be dropped due
1420 * to congestion or traffic shaping.
1422 * -----------------------------------------------------------------------------------
1423 * I notice this method can also return errors from the queue disciplines,
1424 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1427 * Regardless of the return value, the skb is consumed, so it is currently
1428 * difficult to retry a send to this method. (You can bump the ref count
1429 * before sending to hold a reference for retry if you are careful.)
1431 * When calling this method, interrupts MUST be enabled. This is because
1432 * the BH enable code must have IRQs enabled so that it will not deadlock.
1436 int dev_queue_xmit(struct sk_buff *skb)
1438 struct net_device *dev = skb->dev;
1442 /* If a checksum-deferred packet is forwarded to a device that needs a
1443 * checksum, correct the pointers and force checksumming.
1445 if (skb_checksum_setup(skb))
1448 /* GSO will handle the following emulations directly. */
1449 if (netif_needs_gso(dev, skb))
1452 if (skb_shinfo(skb)->frag_list &&
1453 !(dev->features & NETIF_F_FRAGLIST) &&
1454 __skb_linearize(skb))
1457 /* Fragmented skb is linearized if device does not support SG,
1458 * or if at least one of fragments is in highmem and device
1459 * does not support DMA from it.
1461 if (skb_shinfo(skb)->nr_frags &&
1462 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1463 __skb_linearize(skb))
1466 /* If packet is not checksummed and device does not support
1467 * checksumming for this protocol, complete checksumming here.
1469 if (skb->ip_summed == CHECKSUM_HW &&
1470 (!(dev->features & NETIF_F_GEN_CSUM) &&
1471 (!(dev->features & NETIF_F_IP_CSUM) ||
1472 skb->protocol != htons(ETH_P_IP))))
1473 if (skb_checksum_help(skb, 0))
1477 spin_lock_prefetch(&dev->queue_lock);
1479 /* Disable soft irqs for various locks below. Also
1480 * stops preemption for RCU.
1484 /* Updates of qdisc are serialized by queue_lock.
1485 * The struct Qdisc which is pointed to by qdisc is now a
1486 * rcu structure - it may be accessed without acquiring
1487 * a lock (but the structure may be stale.) The freeing of the
1488 * qdisc will be deferred until it's known that there are no
1489 * more references to it.
1491 * If the qdisc has an enqueue function, we still need to
1492 * hold the queue_lock before calling it, since queue_lock
1493 * also serializes access to the device queue.
1496 q = rcu_dereference(dev->qdisc);
1497 #ifdef CONFIG_NET_CLS_ACT
1498 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1501 /* Grab device queue */
1502 spin_lock(&dev->queue_lock);
1504 rc = q->enqueue(skb, q);
1508 spin_unlock(&dev->queue_lock);
1509 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1513 /* The device has no queue. Common case for software devices:
1514 loopback, all the sorts of tunnels...
1516 Really, it is unlikely that netif_tx_lock protection is necessary
1517 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1519 However, it is possible, that they rely on protection
1522 Check this and shot the lock. It is not prone from deadlocks.
1523 Either shot noqueue qdisc, it is even simpler 8)
1525 if (dev->flags & IFF_UP) {
1526 int cpu = smp_processor_id(); /* ok because BHs are off */
1528 if (dev->xmit_lock_owner != cpu) {
1530 HARD_TX_LOCK(dev, cpu);
1532 if (!netif_queue_stopped(dev)) {
1534 if (!dev_hard_start_xmit(skb, dev)) {
1535 HARD_TX_UNLOCK(dev);
1539 HARD_TX_UNLOCK(dev);
1540 if (net_ratelimit())
1541 printk(KERN_CRIT "Virtual device %s asks to "
1542 "queue packet!\n", dev->name);
1544 /* Recursion is detected! It is possible,
1546 if (net_ratelimit())
1547 printk(KERN_CRIT "Dead loop on virtual device "
1548 "%s, fix it urgently!\n", dev->name);
1553 rcu_read_unlock_bh();
1559 rcu_read_unlock_bh();
1564 /*=======================================================================
1566 =======================================================================*/
1568 int netdev_max_backlog = 1000;
1569 int netdev_budget = 300;
1570 int weight_p = 64; /* old backlog weight */
1572 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1576 * netif_rx - post buffer to the network code
1577 * @skb: buffer to post
1579 * This function receives a packet from a device driver and queues it for
1580 * the upper (protocol) levels to process. It always succeeds. The buffer
1581 * may be dropped during processing for congestion control or by the
1585 * NET_RX_SUCCESS (no congestion)
1586 * NET_RX_CN_LOW (low congestion)
1587 * NET_RX_CN_MOD (moderate congestion)
1588 * NET_RX_CN_HIGH (high congestion)
1589 * NET_RX_DROP (packet was dropped)
1593 int netif_rx(struct sk_buff *skb)
1595 struct softnet_data *queue;
1596 unsigned long flags;
1598 /* if netpoll wants it, pretend we never saw it */
1599 if (netpoll_rx(skb))
1602 if (!skb->tstamp.off_sec)
1606 * The code is rearranged so that the path is the most
1607 * short when CPU is congested, but is still operating.
1609 local_irq_save(flags);
1610 queue = &__get_cpu_var(softnet_data);
1612 __get_cpu_var(netdev_rx_stat).total++;
1613 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1614 if (queue->input_pkt_queue.qlen) {
1617 __skb_queue_tail(&queue->input_pkt_queue, skb);
1618 local_irq_restore(flags);
1619 return NET_RX_SUCCESS;
1622 netif_rx_schedule(&queue->backlog_dev);
1626 __get_cpu_var(netdev_rx_stat).dropped++;
1627 local_irq_restore(flags);
1633 int netif_rx_ni(struct sk_buff *skb)
1638 err = netif_rx(skb);
1639 if (local_softirq_pending())
1646 EXPORT_SYMBOL(netif_rx_ni);
1648 static inline struct net_device *skb_bond(struct sk_buff *skb)
1650 struct net_device *dev = skb->dev;
1654 * On bonding slaves other than the currently active
1655 * slave, suppress duplicates except for 802.3ad
1656 * ETH_P_SLOW and alb non-mcast/bcast.
1658 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1659 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1660 if (skb->pkt_type != PACKET_BROADCAST &&
1661 skb->pkt_type != PACKET_MULTICAST)
1665 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1666 skb->protocol == __constant_htons(ETH_P_SLOW))
1673 skb->dev = dev->master;
1679 static void net_tx_action(struct softirq_action *h)
1681 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1683 if (sd->completion_queue) {
1684 struct sk_buff *clist;
1686 local_irq_disable();
1687 clist = sd->completion_queue;
1688 sd->completion_queue = NULL;
1692 struct sk_buff *skb = clist;
1693 clist = clist->next;
1695 BUG_TRAP(!atomic_read(&skb->users));
1700 if (sd->output_queue) {
1701 struct net_device *head;
1703 local_irq_disable();
1704 head = sd->output_queue;
1705 sd->output_queue = NULL;
1709 struct net_device *dev = head;
1710 head = head->next_sched;
1712 smp_mb__before_clear_bit();
1713 clear_bit(__LINK_STATE_SCHED, &dev->state);
1715 if (spin_trylock(&dev->queue_lock)) {
1717 spin_unlock(&dev->queue_lock);
1719 netif_schedule(dev);
1725 static __inline__ int deliver_skb(struct sk_buff *skb,
1726 struct packet_type *pt_prev,
1727 struct net_device *orig_dev)
1729 atomic_inc(&skb->users);
1730 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1733 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1734 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1736 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1737 unsigned char *addr);
1738 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1740 static __inline__ int handle_bridge(struct sk_buff **pskb,
1741 struct packet_type **pt_prev, int *ret,
1742 struct net_device *orig_dev)
1744 struct net_bridge_port *port;
1746 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1747 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1751 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1755 return br_handle_frame_hook(port, pskb);
1758 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1761 #ifdef CONFIG_NET_CLS_ACT
1762 /* TODO: Maybe we should just force sch_ingress to be compiled in
1763 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1764 * a compare and 2 stores extra right now if we dont have it on
1765 * but have CONFIG_NET_CLS_ACT
1766 * NOTE: This doesnt stop any functionality; if you dont have
1767 * the ingress scheduler, you just cant add policies on ingress.
1770 static int ing_filter(struct sk_buff *skb)
1773 struct net_device *dev = skb->dev;
1774 int result = TC_ACT_OK;
1776 if (dev->qdisc_ingress) {
1777 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1778 if (MAX_RED_LOOP < ttl++) {
1779 printk("Redir loop detected Dropping packet (%s->%s)\n",
1780 skb->input_dev->name, skb->dev->name);
1784 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1786 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1788 spin_lock(&dev->ingress_lock);
1789 if ((q = dev->qdisc_ingress) != NULL)
1790 result = q->enqueue(skb, q);
1791 spin_unlock(&dev->ingress_lock);
1799 int netif_receive_skb(struct sk_buff *skb)
1801 struct packet_type *ptype, *pt_prev;
1802 struct net_device *orig_dev;
1803 int ret = NET_RX_DROP;
1804 unsigned short type;
1806 /* if we've gotten here through NAPI, check netpoll */
1807 if (skb->dev->poll && netpoll_rx(skb))
1810 if (!skb->tstamp.off_sec)
1813 if (!skb->input_dev)
1814 skb->input_dev = skb->dev;
1816 orig_dev = skb_bond(skb);
1821 __get_cpu_var(netdev_rx_stat).total++;
1823 skb->h.raw = skb->nh.raw = skb->data;
1824 skb->mac_len = skb->nh.raw - skb->mac.raw;
1830 #ifdef CONFIG_NET_CLS_ACT
1831 if (skb->tc_verd & TC_NCLS) {
1832 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1838 switch (skb->ip_summed) {
1839 case CHECKSUM_UNNECESSARY:
1840 skb->proto_data_valid = 1;
1843 /* XXX Implement me. */
1845 skb->proto_data_valid = 0;
1850 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1851 if (!ptype->dev || ptype->dev == skb->dev) {
1853 ret = deliver_skb(skb, pt_prev, orig_dev);
1858 #ifdef CONFIG_NET_CLS_ACT
1860 ret = deliver_skb(skb, pt_prev, orig_dev);
1861 pt_prev = NULL; /* noone else should process this after*/
1863 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1866 ret = ing_filter(skb);
1868 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1877 handle_diverter(skb);
1879 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1882 type = skb->protocol;
1883 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1884 if (ptype->type == type &&
1885 (!ptype->dev || ptype->dev == skb->dev)) {
1887 ret = deliver_skb(skb, pt_prev, orig_dev);
1893 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1896 /* Jamal, now you will not able to escape explaining
1897 * me how you were going to use this. :-)
1907 static int process_backlog(struct net_device *backlog_dev, int *budget)
1910 int quota = min(backlog_dev->quota, *budget);
1911 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1912 unsigned long start_time = jiffies;
1914 backlog_dev->weight = weight_p;
1916 struct sk_buff *skb;
1917 struct net_device *dev;
1919 local_irq_disable();
1920 skb = __skb_dequeue(&queue->input_pkt_queue);
1927 netif_receive_skb(skb);
1933 if (work >= quota || jiffies - start_time > 1)
1938 backlog_dev->quota -= work;
1943 backlog_dev->quota -= work;
1946 list_del(&backlog_dev->poll_list);
1947 smp_mb__before_clear_bit();
1948 netif_poll_enable(backlog_dev);
1954 static void net_rx_action(struct softirq_action *h)
1956 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1957 unsigned long start_time = jiffies;
1958 int budget = netdev_budget;
1961 local_irq_disable();
1963 while (!list_empty(&queue->poll_list)) {
1964 struct net_device *dev;
1966 if (budget <= 0 || jiffies - start_time > 1)
1971 dev = list_entry(queue->poll_list.next,
1972 struct net_device, poll_list);
1973 have = netpoll_poll_lock(dev);
1975 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1976 netpoll_poll_unlock(have);
1977 local_irq_disable();
1978 list_move_tail(&dev->poll_list, &queue->poll_list);
1980 dev->quota += dev->weight;
1982 dev->quota = dev->weight;
1984 netpoll_poll_unlock(have);
1986 local_irq_disable();
1994 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1995 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1999 static gifconf_func_t * gifconf_list [NPROTO];
2002 * register_gifconf - register a SIOCGIF handler
2003 * @family: Address family
2004 * @gifconf: Function handler
2006 * Register protocol dependent address dumping routines. The handler
2007 * that is passed must not be freed or reused until it has been replaced
2008 * by another handler.
2010 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2012 if (family >= NPROTO)
2014 gifconf_list[family] = gifconf;
2020 * Map an interface index to its name (SIOCGIFNAME)
2024 * We need this ioctl for efficient implementation of the
2025 * if_indextoname() function required by the IPv6 API. Without
2026 * it, we would have to search all the interfaces to find a
2030 static int dev_ifname(struct ifreq __user *arg)
2032 struct net_device *dev;
2036 * Fetch the caller's info block.
2039 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2042 read_lock(&dev_base_lock);
2043 dev = __dev_get_by_index(ifr.ifr_ifindex);
2045 read_unlock(&dev_base_lock);
2049 strcpy(ifr.ifr_name, dev->name);
2050 read_unlock(&dev_base_lock);
2052 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2058 * Perform a SIOCGIFCONF call. This structure will change
2059 * size eventually, and there is nothing I can do about it.
2060 * Thus we will need a 'compatibility mode'.
2063 static int dev_ifconf(char __user *arg)
2066 struct net_device *dev;
2073 * Fetch the caller's info block.
2076 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2083 * Loop over the interfaces, and write an info block for each.
2087 for (dev = dev_base; dev; dev = dev->next) {
2088 if (vx_flags(VXF_HIDE_NETIF, 0) &&
2089 !dev_in_nx_info(dev, current->nx_info))
2091 for (i = 0; i < NPROTO; i++) {
2092 if (gifconf_list[i]) {
2095 done = gifconf_list[i](dev, NULL, 0);
2097 done = gifconf_list[i](dev, pos + total,
2107 * All done. Write the updated control block back to the caller.
2109 ifc.ifc_len = total;
2112 * Both BSD and Solaris return 0 here, so we do too.
2114 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2117 #ifdef CONFIG_PROC_FS
2119 * This is invoked by the /proc filesystem handler to display a device
2122 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2124 struct net_device *dev;
2127 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2129 return i == pos ? dev : NULL;
2132 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2134 read_lock(&dev_base_lock);
2135 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2138 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2141 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2144 void dev_seq_stop(struct seq_file *seq, void *v)
2146 read_unlock(&dev_base_lock);
2149 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2151 struct nx_info *nxi = current->nx_info;
2153 if (vx_flags(VXF_HIDE_NETIF, 0) && !dev_in_nx_info(dev, nxi))
2155 if (dev->get_stats) {
2156 struct net_device_stats *stats = dev->get_stats(dev);
2158 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2159 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2160 dev->name, stats->rx_bytes, stats->rx_packets,
2162 stats->rx_dropped + stats->rx_missed_errors,
2163 stats->rx_fifo_errors,
2164 stats->rx_length_errors + stats->rx_over_errors +
2165 stats->rx_crc_errors + stats->rx_frame_errors,
2166 stats->rx_compressed, stats->multicast,
2167 stats->tx_bytes, stats->tx_packets,
2168 stats->tx_errors, stats->tx_dropped,
2169 stats->tx_fifo_errors, stats->collisions,
2170 stats->tx_carrier_errors +
2171 stats->tx_aborted_errors +
2172 stats->tx_window_errors +
2173 stats->tx_heartbeat_errors,
2174 stats->tx_compressed);
2176 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2180 * Called from the PROCfs module. This now uses the new arbitrary sized
2181 * /proc/net interface to create /proc/net/dev
2183 static int dev_seq_show(struct seq_file *seq, void *v)
2185 if (v == SEQ_START_TOKEN)
2186 seq_puts(seq, "Inter-| Receive "
2188 " face |bytes packets errs drop fifo frame "
2189 "compressed multicast|bytes packets errs "
2190 "drop fifo colls carrier compressed\n");
2192 dev_seq_printf_stats(seq, v);
2196 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2198 struct netif_rx_stats *rc = NULL;
2200 while (*pos < NR_CPUS)
2201 if (cpu_online(*pos)) {
2202 rc = &per_cpu(netdev_rx_stat, *pos);
2209 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2211 return softnet_get_online(pos);
2214 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2217 return softnet_get_online(pos);
2220 static void softnet_seq_stop(struct seq_file *seq, void *v)
2224 static int softnet_seq_show(struct seq_file *seq, void *v)
2226 struct netif_rx_stats *s = v;
2228 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2229 s->total, s->dropped, s->time_squeeze, 0,
2230 0, 0, 0, 0, /* was fastroute */
2235 static struct seq_operations dev_seq_ops = {
2236 .start = dev_seq_start,
2237 .next = dev_seq_next,
2238 .stop = dev_seq_stop,
2239 .show = dev_seq_show,
2242 static int dev_seq_open(struct inode *inode, struct file *file)
2244 return seq_open(file, &dev_seq_ops);
2247 static struct file_operations dev_seq_fops = {
2248 .owner = THIS_MODULE,
2249 .open = dev_seq_open,
2251 .llseek = seq_lseek,
2252 .release = seq_release,
2255 static struct seq_operations softnet_seq_ops = {
2256 .start = softnet_seq_start,
2257 .next = softnet_seq_next,
2258 .stop = softnet_seq_stop,
2259 .show = softnet_seq_show,
2262 static int softnet_seq_open(struct inode *inode, struct file *file)
2264 return seq_open(file, &softnet_seq_ops);
2267 static struct file_operations softnet_seq_fops = {
2268 .owner = THIS_MODULE,
2269 .open = softnet_seq_open,
2271 .llseek = seq_lseek,
2272 .release = seq_release,
2275 #ifdef CONFIG_WIRELESS_EXT
2276 extern int wireless_proc_init(void);
2278 #define wireless_proc_init() 0
2281 static int __init dev_proc_init(void)
2285 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2287 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2289 if (wireless_proc_init())
2295 proc_net_remove("softnet_stat");
2297 proc_net_remove("dev");
2301 #define dev_proc_init() 0
2302 #endif /* CONFIG_PROC_FS */
2306 * netdev_set_master - set up master/slave pair
2307 * @slave: slave device
2308 * @master: new master device
2310 * Changes the master device of the slave. Pass %NULL to break the
2311 * bonding. The caller must hold the RTNL semaphore. On a failure
2312 * a negative errno code is returned. On success the reference counts
2313 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2314 * function returns zero.
2316 int netdev_set_master(struct net_device *slave, struct net_device *master)
2318 struct net_device *old = slave->master;
2328 slave->master = master;
2336 slave->flags |= IFF_SLAVE;
2338 slave->flags &= ~IFF_SLAVE;
2340 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2345 * dev_set_promiscuity - update promiscuity count on a device
2349 * Add or remove promiscuity from a device. While the count in the device
2350 * remains above zero the interface remains promiscuous. Once it hits zero
2351 * the device reverts back to normal filtering operation. A negative inc
2352 * value is used to drop promiscuity on the device.
2354 void dev_set_promiscuity(struct net_device *dev, int inc)
2356 unsigned short old_flags = dev->flags;
2358 if ((dev->promiscuity += inc) == 0)
2359 dev->flags &= ~IFF_PROMISC;
2361 dev->flags |= IFF_PROMISC;
2362 if (dev->flags != old_flags) {
2364 printk(KERN_INFO "device %s %s promiscuous mode\n",
2365 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2367 audit_log(current->audit_context, GFP_ATOMIC,
2368 AUDIT_ANOM_PROMISCUOUS,
2369 "dev=%s prom=%d old_prom=%d auid=%u",
2370 dev->name, (dev->flags & IFF_PROMISC),
2371 (old_flags & IFF_PROMISC),
2372 audit_get_loginuid(current->audit_context));
2377 * dev_set_allmulti - update allmulti count on a device
2381 * Add or remove reception of all multicast frames to a device. While the
2382 * count in the device remains above zero the interface remains listening
2383 * to all interfaces. Once it hits zero the device reverts back to normal
2384 * filtering operation. A negative @inc value is used to drop the counter
2385 * when releasing a resource needing all multicasts.
2388 void dev_set_allmulti(struct net_device *dev, int inc)
2390 unsigned short old_flags = dev->flags;
2392 dev->flags |= IFF_ALLMULTI;
2393 if ((dev->allmulti += inc) == 0)
2394 dev->flags &= ~IFF_ALLMULTI;
2395 if (dev->flags ^ old_flags)
2399 unsigned dev_get_flags(const struct net_device *dev)
2403 flags = (dev->flags & ~(IFF_PROMISC |
2408 (dev->gflags & (IFF_PROMISC |
2411 if (netif_running(dev)) {
2412 if (netif_oper_up(dev))
2413 flags |= IFF_RUNNING;
2414 if (netif_carrier_ok(dev))
2415 flags |= IFF_LOWER_UP;
2416 if (netif_dormant(dev))
2417 flags |= IFF_DORMANT;
2423 int dev_change_flags(struct net_device *dev, unsigned flags)
2426 int old_flags = dev->flags;
2429 * Set the flags on our device.
2432 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2433 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2435 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2439 * Load in the correct multicast list now the flags have changed.
2445 * Have we downed the interface. We handle IFF_UP ourselves
2446 * according to user attempts to set it, rather than blindly
2451 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2452 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2458 if (dev->flags & IFF_UP &&
2459 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2461 raw_notifier_call_chain(&netdev_chain,
2462 NETDEV_CHANGE, dev);
2464 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2465 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2466 dev->gflags ^= IFF_PROMISC;
2467 dev_set_promiscuity(dev, inc);
2470 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2471 is important. Some (broken) drivers set IFF_PROMISC, when
2472 IFF_ALLMULTI is requested not asking us and not reporting.
2474 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2475 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2476 dev->gflags ^= IFF_ALLMULTI;
2477 dev_set_allmulti(dev, inc);
2480 if (old_flags ^ dev->flags)
2481 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2486 int dev_set_mtu(struct net_device *dev, int new_mtu)
2490 if (new_mtu == dev->mtu)
2493 /* MTU must be positive. */
2497 if (!netif_device_present(dev))
2501 if (dev->change_mtu)
2502 err = dev->change_mtu(dev, new_mtu);
2505 if (!err && dev->flags & IFF_UP)
2506 raw_notifier_call_chain(&netdev_chain,
2507 NETDEV_CHANGEMTU, dev);
2511 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2515 if (!dev->set_mac_address)
2517 if (sa->sa_family != dev->type)
2519 if (!netif_device_present(dev))
2521 err = dev->set_mac_address(dev, sa);
2523 raw_notifier_call_chain(&netdev_chain,
2524 NETDEV_CHANGEADDR, dev);
2529 * Perform the SIOCxIFxxx calls.
2531 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2534 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2540 case SIOCGIFFLAGS: /* Get interface flags */
2541 ifr->ifr_flags = dev_get_flags(dev);
2544 case SIOCSIFFLAGS: /* Set interface flags */
2545 return dev_change_flags(dev, ifr->ifr_flags);
2547 case SIOCGIFMETRIC: /* Get the metric on the interface
2548 (currently unused) */
2549 ifr->ifr_metric = 0;
2552 case SIOCSIFMETRIC: /* Set the metric on the interface
2553 (currently unused) */
2556 case SIOCGIFMTU: /* Get the MTU of a device */
2557 ifr->ifr_mtu = dev->mtu;
2560 case SIOCSIFMTU: /* Set the MTU of a device */
2561 return dev_set_mtu(dev, ifr->ifr_mtu);
2565 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2567 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2568 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2569 ifr->ifr_hwaddr.sa_family = dev->type;
2573 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2575 case SIOCSIFHWBROADCAST:
2576 if (ifr->ifr_hwaddr.sa_family != dev->type)
2578 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2579 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2580 raw_notifier_call_chain(&netdev_chain,
2581 NETDEV_CHANGEADDR, dev);
2585 ifr->ifr_map.mem_start = dev->mem_start;
2586 ifr->ifr_map.mem_end = dev->mem_end;
2587 ifr->ifr_map.base_addr = dev->base_addr;
2588 ifr->ifr_map.irq = dev->irq;
2589 ifr->ifr_map.dma = dev->dma;
2590 ifr->ifr_map.port = dev->if_port;
2594 if (dev->set_config) {
2595 if (!netif_device_present(dev))
2597 return dev->set_config(dev, &ifr->ifr_map);
2602 if (!dev->set_multicast_list ||
2603 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2605 if (!netif_device_present(dev))
2607 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2611 if (!dev->set_multicast_list ||
2612 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2614 if (!netif_device_present(dev))
2616 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2620 ifr->ifr_ifindex = dev->ifindex;
2624 ifr->ifr_qlen = dev->tx_queue_len;
2628 if (ifr->ifr_qlen < 0)
2630 dev->tx_queue_len = ifr->ifr_qlen;
2634 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2635 return dev_change_name(dev, ifr->ifr_newname);
2638 * Unknown or private ioctl
2642 if ((cmd >= SIOCDEVPRIVATE &&
2643 cmd <= SIOCDEVPRIVATE + 15) ||
2644 cmd == SIOCBONDENSLAVE ||
2645 cmd == SIOCBONDRELEASE ||
2646 cmd == SIOCBONDSETHWADDR ||
2647 cmd == SIOCBONDSLAVEINFOQUERY ||
2648 cmd == SIOCBONDINFOQUERY ||
2649 cmd == SIOCBONDCHANGEACTIVE ||
2650 cmd == SIOCGMIIPHY ||
2651 cmd == SIOCGMIIREG ||
2652 cmd == SIOCSMIIREG ||
2653 cmd == SIOCBRADDIF ||
2654 cmd == SIOCBRDELIF ||
2655 cmd == SIOCWANDEV) {
2657 if (dev->do_ioctl) {
2658 if (netif_device_present(dev))
2659 err = dev->do_ioctl(dev, ifr,
2672 * This function handles all "interface"-type I/O control requests. The actual
2673 * 'doing' part of this is dev_ifsioc above.
2677 * dev_ioctl - network device ioctl
2678 * @cmd: command to issue
2679 * @arg: pointer to a struct ifreq in user space
2681 * Issue ioctl functions to devices. This is normally called by the
2682 * user space syscall interfaces but can sometimes be useful for
2683 * other purposes. The return value is the return from the syscall if
2684 * positive or a negative errno code on error.
2687 int dev_ioctl(unsigned int cmd, void __user *arg)
2693 /* One special case: SIOCGIFCONF takes ifconf argument
2694 and requires shared lock, because it sleeps writing
2698 if (cmd == SIOCGIFCONF) {
2700 ret = dev_ifconf((char __user *) arg);
2704 if (cmd == SIOCGIFNAME)
2705 return dev_ifname((struct ifreq __user *)arg);
2707 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2710 ifr.ifr_name[IFNAMSIZ-1] = 0;
2712 colon = strchr(ifr.ifr_name, ':');
2717 * See which interface the caller is talking about.
2722 * These ioctl calls:
2723 * - can be done by all.
2724 * - atomic and do not require locking.
2735 dev_load(ifr.ifr_name);
2736 read_lock(&dev_base_lock);
2737 ret = dev_ifsioc(&ifr, cmd);
2738 read_unlock(&dev_base_lock);
2742 if (copy_to_user(arg, &ifr,
2743 sizeof(struct ifreq)))
2749 dev_load(ifr.ifr_name);
2751 ret = dev_ethtool(&ifr);
2756 if (copy_to_user(arg, &ifr,
2757 sizeof(struct ifreq)))
2763 * These ioctl calls:
2764 * - require superuser power.
2765 * - require strict serialization.
2771 if (!capable(CAP_NET_ADMIN))
2773 dev_load(ifr.ifr_name);
2775 ret = dev_ifsioc(&ifr, cmd);
2780 if (copy_to_user(arg, &ifr,
2781 sizeof(struct ifreq)))
2787 * These ioctl calls:
2788 * - require superuser power.
2789 * - require strict serialization.
2790 * - do not return a value
2800 case SIOCSIFHWBROADCAST:
2803 case SIOCBONDENSLAVE:
2804 case SIOCBONDRELEASE:
2805 case SIOCBONDSETHWADDR:
2806 case SIOCBONDCHANGEACTIVE:
2809 if (!capable(CAP_NET_ADMIN))
2812 case SIOCBONDSLAVEINFOQUERY:
2813 case SIOCBONDINFOQUERY:
2814 dev_load(ifr.ifr_name);
2816 ret = dev_ifsioc(&ifr, cmd);
2821 /* Get the per device memory space. We can add this but
2822 * currently do not support it */
2824 /* Set the per device memory buffer space.
2825 * Not applicable in our case */
2830 * Unknown or private ioctl.
2833 if (cmd == SIOCWANDEV ||
2834 (cmd >= SIOCDEVPRIVATE &&
2835 cmd <= SIOCDEVPRIVATE + 15)) {
2836 dev_load(ifr.ifr_name);
2838 ret = dev_ifsioc(&ifr, cmd);
2840 if (!ret && copy_to_user(arg, &ifr,
2841 sizeof(struct ifreq)))
2845 #ifdef CONFIG_WIRELESS_EXT
2846 /* Take care of Wireless Extensions */
2847 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2848 /* If command is `set a parameter', or
2849 * `get the encoding parameters', check if
2850 * the user has the right to do it */
2851 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2852 || cmd == SIOCGIWENCODEEXT) {
2853 if (!capable(CAP_NET_ADMIN))
2856 dev_load(ifr.ifr_name);
2858 /* Follow me in net/core/wireless.c */
2859 ret = wireless_process_ioctl(&ifr, cmd);
2861 if (IW_IS_GET(cmd) &&
2862 copy_to_user(arg, &ifr,
2863 sizeof(struct ifreq)))
2867 #endif /* CONFIG_WIRELESS_EXT */
2874 * dev_new_index - allocate an ifindex
2876 * Returns a suitable unique value for a new device interface
2877 * number. The caller must hold the rtnl semaphore or the
2878 * dev_base_lock to be sure it remains unique.
2880 static int dev_new_index(void)
2886 if (!__dev_get_by_index(ifindex))
2891 static int dev_boot_phase = 1;
2893 /* Delayed registration/unregisteration */
2894 static DEFINE_SPINLOCK(net_todo_list_lock);
2895 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2897 static inline void net_set_todo(struct net_device *dev)
2899 spin_lock(&net_todo_list_lock);
2900 list_add_tail(&dev->todo_list, &net_todo_list);
2901 spin_unlock(&net_todo_list_lock);
2905 * register_netdevice - register a network device
2906 * @dev: device to register
2908 * Take a completed network device structure and add it to the kernel
2909 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2910 * chain. 0 is returned on success. A negative errno code is returned
2911 * on a failure to set up the device, or if the name is a duplicate.
2913 * Callers must hold the rtnl semaphore. You may want
2914 * register_netdev() instead of this.
2917 * The locking appears insufficient to guarantee two parallel registers
2918 * will not get the same name.
2921 int register_netdevice(struct net_device *dev)
2923 struct hlist_head *head;
2924 struct hlist_node *p;
2927 BUG_ON(dev_boot_phase);
2932 /* When net_device's are persistent, this will be fatal. */
2933 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2935 spin_lock_init(&dev->queue_lock);
2936 spin_lock_init(&dev->_xmit_lock);
2937 dev->xmit_lock_owner = -1;
2938 #ifdef CONFIG_NET_CLS_ACT
2939 spin_lock_init(&dev->ingress_lock);
2942 ret = alloc_divert_blk(dev);
2948 /* Init, if this function is available */
2950 ret = dev->init(dev);
2958 if (!dev_valid_name(dev->name)) {
2963 dev->ifindex = dev_new_index();
2964 if (dev->iflink == -1)
2965 dev->iflink = dev->ifindex;
2967 /* Check for existence of name */
2968 head = dev_name_hash(dev->name);
2969 hlist_for_each(p, head) {
2970 struct net_device *d
2971 = hlist_entry(p, struct net_device, name_hlist);
2972 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2978 /* Fix illegal SG+CSUM combinations. */
2979 if ((dev->features & NETIF_F_SG) &&
2980 !(dev->features & NETIF_F_ALL_CSUM)) {
2981 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2983 dev->features &= ~NETIF_F_SG;
2986 /* TSO requires that SG is present as well. */
2987 if ((dev->features & NETIF_F_TSO) &&
2988 !(dev->features & NETIF_F_SG)) {
2989 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2991 dev->features &= ~NETIF_F_TSO;
2993 if (dev->features & NETIF_F_UFO) {
2994 if (!(dev->features & NETIF_F_HW_CSUM)) {
2995 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2996 "NETIF_F_HW_CSUM feature.\n",
2998 dev->features &= ~NETIF_F_UFO;
3000 if (!(dev->features & NETIF_F_SG)) {
3001 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3002 "NETIF_F_SG feature.\n",
3004 dev->features &= ~NETIF_F_UFO;
3009 * nil rebuild_header routine,
3010 * that should be never called and used as just bug trap.
3013 if (!dev->rebuild_header)
3014 dev->rebuild_header = default_rebuild_header;
3016 ret = netdev_register_sysfs(dev);
3019 dev->reg_state = NETREG_REGISTERED;
3022 * Default initial state at registry is that the
3023 * device is present.
3026 set_bit(__LINK_STATE_PRESENT, &dev->state);
3029 dev_init_scheduler(dev);
3030 write_lock_bh(&dev_base_lock);
3032 dev_tail = &dev->next;
3033 hlist_add_head(&dev->name_hlist, head);
3034 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3036 write_unlock_bh(&dev_base_lock);
3038 /* Notify protocols, that a new device appeared. */
3039 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3046 free_divert_blk(dev);
3051 * register_netdev - register a network device
3052 * @dev: device to register
3054 * Take a completed network device structure and add it to the kernel
3055 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3056 * chain. 0 is returned on success. A negative errno code is returned
3057 * on a failure to set up the device, or if the name is a duplicate.
3059 * This is a wrapper around register_netdev that takes the rtnl semaphore
3060 * and expands the device name if you passed a format string to
3063 int register_netdev(struct net_device *dev)
3070 * If the name is a format string the caller wants us to do a
3073 if (strchr(dev->name, '%')) {
3074 err = dev_alloc_name(dev, dev->name);
3080 * Back compatibility hook. Kill this one in 2.5
3082 if (dev->name[0] == 0 || dev->name[0] == ' ') {
3083 err = dev_alloc_name(dev, "eth%d");
3088 err = register_netdevice(dev);
3093 EXPORT_SYMBOL(register_netdev);
3096 * netdev_wait_allrefs - wait until all references are gone.
3098 * This is called when unregistering network devices.
3100 * Any protocol or device that holds a reference should register
3101 * for netdevice notification, and cleanup and put back the
3102 * reference if they receive an UNREGISTER event.
3103 * We can get stuck here if buggy protocols don't correctly
3106 static void netdev_wait_allrefs(struct net_device *dev)
3108 unsigned long rebroadcast_time, warning_time;
3110 rebroadcast_time = warning_time = jiffies;
3111 while (atomic_read(&dev->refcnt) != 0) {
3112 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3115 /* Rebroadcast unregister notification */
3116 raw_notifier_call_chain(&netdev_chain,
3117 NETDEV_UNREGISTER, dev);
3119 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3121 /* We must not have linkwatch events
3122 * pending on unregister. If this
3123 * happens, we simply run the queue
3124 * unscheduled, resulting in a noop
3127 linkwatch_run_queue();
3132 rebroadcast_time = jiffies;
3137 if (time_after(jiffies, warning_time + 10 * HZ)) {
3138 printk(KERN_EMERG "unregister_netdevice: "
3139 "waiting for %s to become free. Usage "
3141 dev->name, atomic_read(&dev->refcnt));
3142 warning_time = jiffies;
3151 * register_netdevice(x1);
3152 * register_netdevice(x2);
3154 * unregister_netdevice(y1);
3155 * unregister_netdevice(y2);
3161 * We are invoked by rtnl_unlock() after it drops the semaphore.
3162 * This allows us to deal with problems:
3163 * 1) We can delete sysfs objects which invoke hotplug
3164 * without deadlocking with linkwatch via keventd.
3165 * 2) Since we run with the RTNL semaphore not held, we can sleep
3166 * safely in order to wait for the netdev refcnt to drop to zero.
3168 static DEFINE_MUTEX(net_todo_run_mutex);
3169 void netdev_run_todo(void)
3171 struct list_head list = LIST_HEAD_INIT(list);
3173 /* Need to guard against multiple cpu's getting out of order. */
3174 mutex_lock(&net_todo_run_mutex);
3176 /* Not safe to do outside the semaphore. We must not return
3177 * until all unregister events invoked by the local processor
3178 * have been completed (either by this todo run, or one on
3181 if (list_empty(&net_todo_list))
3184 /* Snapshot list, allow later requests */
3185 spin_lock(&net_todo_list_lock);
3186 list_splice_init(&net_todo_list, &list);
3187 spin_unlock(&net_todo_list_lock);
3189 while (!list_empty(&list)) {
3190 struct net_device *dev
3191 = list_entry(list.next, struct net_device, todo_list);
3192 list_del(&dev->todo_list);
3194 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3195 printk(KERN_ERR "network todo '%s' but state %d\n",
3196 dev->name, dev->reg_state);
3201 netdev_unregister_sysfs(dev);
3202 dev->reg_state = NETREG_UNREGISTERED;
3204 netdev_wait_allrefs(dev);
3207 BUG_ON(atomic_read(&dev->refcnt));
3208 BUG_TRAP(!dev->ip_ptr);
3209 BUG_TRAP(!dev->ip6_ptr);
3210 BUG_TRAP(!dev->dn_ptr);
3212 /* It must be the very last action,
3213 * after this 'dev' may point to freed up memory.
3215 if (dev->destructor)
3216 dev->destructor(dev);
3220 mutex_unlock(&net_todo_run_mutex);
3224 * alloc_netdev - allocate network device
3225 * @sizeof_priv: size of private data to allocate space for
3226 * @name: device name format string
3227 * @setup: callback to initialize device
3229 * Allocates a struct net_device with private data area for driver use
3230 * and performs basic initialization.
3232 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3233 void (*setup)(struct net_device *))
3236 struct net_device *dev;
3239 /* ensure 32-byte alignment of both the device and private area */
3240 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3241 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3243 p = kzalloc(alloc_size, GFP_KERNEL);
3245 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3249 dev = (struct net_device *)
3250 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3251 dev->padded = (char *)dev - (char *)p;
3254 dev->priv = netdev_priv(dev);
3257 strcpy(dev->name, name);
3260 EXPORT_SYMBOL(alloc_netdev);
3263 * free_netdev - free network device
3266 * This function does the last stage of destroying an allocated device
3267 * interface. The reference to the device object is released.
3268 * If this is the last reference then it will be freed.
3270 void free_netdev(struct net_device *dev)
3273 /* Compatibility with error handling in drivers */
3274 if (dev->reg_state == NETREG_UNINITIALIZED) {
3275 kfree((char *)dev - dev->padded);
3279 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3280 dev->reg_state = NETREG_RELEASED;
3282 /* will free via class release */
3283 class_device_put(&dev->class_dev);
3285 kfree((char *)dev - dev->padded);
3289 /* Synchronize with packet receive processing. */
3290 void synchronize_net(void)
3297 * unregister_netdevice - remove device from the kernel
3300 * This function shuts down a device interface and removes it
3301 * from the kernel tables. On success 0 is returned, on a failure
3302 * a negative errno code is returned.
3304 * Callers must hold the rtnl semaphore. You may want
3305 * unregister_netdev() instead of this.
3308 int unregister_netdevice(struct net_device *dev)
3310 struct net_device *d, **dp;
3312 BUG_ON(dev_boot_phase);
3315 /* Some devices call without registering for initialization unwind. */
3316 if (dev->reg_state == NETREG_UNINITIALIZED) {
3317 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3318 "was registered\n", dev->name, dev);
3322 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3324 /* If device is running, close it first. */
3325 if (dev->flags & IFF_UP)
3328 /* And unlink it from device chain. */
3329 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3331 write_lock_bh(&dev_base_lock);
3332 hlist_del(&dev->name_hlist);
3333 hlist_del(&dev->index_hlist);
3334 if (dev_tail == &dev->next)
3337 write_unlock_bh(&dev_base_lock);
3342 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3347 dev->reg_state = NETREG_UNREGISTERING;
3351 /* Shutdown queueing discipline. */
3355 /* Notify protocols, that we are about to destroy
3356 this device. They should clean all the things.
3358 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3361 * Flush the multicast chain
3363 dev_mc_discard(dev);
3368 /* Notifier chain MUST detach us from master device. */
3369 BUG_TRAP(!dev->master);
3371 free_divert_blk(dev);
3373 /* Finish processing unregister after unlock */
3383 * unregister_netdev - remove device from the kernel
3386 * This function shuts down a device interface and removes it
3387 * from the kernel tables. On success 0 is returned, on a failure
3388 * a negative errno code is returned.
3390 * This is just a wrapper for unregister_netdevice that takes
3391 * the rtnl semaphore. In general you want to use this and not
3392 * unregister_netdevice.
3394 void unregister_netdev(struct net_device *dev)
3397 unregister_netdevice(dev);
3401 EXPORT_SYMBOL(unregister_netdev);
3403 #ifdef CONFIG_HOTPLUG_CPU
3404 static int dev_cpu_callback(struct notifier_block *nfb,
3405 unsigned long action,
3408 struct sk_buff **list_skb;
3409 struct net_device **list_net;
3410 struct sk_buff *skb;
3411 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3412 struct softnet_data *sd, *oldsd;
3414 if (action != CPU_DEAD)
3417 local_irq_disable();
3418 cpu = smp_processor_id();
3419 sd = &per_cpu(softnet_data, cpu);
3420 oldsd = &per_cpu(softnet_data, oldcpu);
3422 /* Find end of our completion_queue. */
3423 list_skb = &sd->completion_queue;
3425 list_skb = &(*list_skb)->next;
3426 /* Append completion queue from offline CPU. */
3427 *list_skb = oldsd->completion_queue;
3428 oldsd->completion_queue = NULL;
3430 /* Find end of our output_queue. */
3431 list_net = &sd->output_queue;
3433 list_net = &(*list_net)->next_sched;
3434 /* Append output queue from offline CPU. */
3435 *list_net = oldsd->output_queue;
3436 oldsd->output_queue = NULL;
3438 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3441 /* Process offline CPU's input_pkt_queue */
3442 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3447 #endif /* CONFIG_HOTPLUG_CPU */
3451 * Initialize the DEV module. At boot time this walks the device list and
3452 * unhooks any devices that fail to initialise (normally hardware not
3453 * present) and leaves us with a valid list of present and active devices.
3458 * This is called single threaded during boot, so no need
3459 * to take the rtnl semaphore.
3461 static int __init net_dev_init(void)
3463 int i, rc = -ENOMEM;
3465 BUG_ON(!dev_boot_phase);
3469 if (dev_proc_init())
3472 if (netdev_sysfs_init())
3475 INIT_LIST_HEAD(&ptype_all);
3476 for (i = 0; i < 16; i++)
3477 INIT_LIST_HEAD(&ptype_base[i]);
3479 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3480 INIT_HLIST_HEAD(&dev_name_head[i]);
3482 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3483 INIT_HLIST_HEAD(&dev_index_head[i]);
3486 * Initialise the packet receive queues.
3489 for_each_possible_cpu(i) {
3490 struct softnet_data *queue;
3492 queue = &per_cpu(softnet_data, i);
3493 skb_queue_head_init(&queue->input_pkt_queue);
3494 queue->completion_queue = NULL;
3495 INIT_LIST_HEAD(&queue->poll_list);
3496 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3497 queue->backlog_dev.weight = weight_p;
3498 queue->backlog_dev.poll = process_backlog;
3499 atomic_set(&queue->backlog_dev.refcnt, 1);
3504 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3505 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3507 hotcpu_notifier(dev_cpu_callback, 0);
3515 subsys_initcall(net_dev_init);
3517 EXPORT_SYMBOL(__dev_get_by_index);
3518 EXPORT_SYMBOL(__dev_get_by_name);
3519 EXPORT_SYMBOL(__dev_remove_pack);
3520 EXPORT_SYMBOL(dev_valid_name);
3521 EXPORT_SYMBOL(dev_add_pack);
3522 EXPORT_SYMBOL(dev_alloc_name);
3523 EXPORT_SYMBOL(dev_close);
3524 EXPORT_SYMBOL(dev_get_by_flags);
3525 EXPORT_SYMBOL(dev_get_by_index);
3526 EXPORT_SYMBOL(dev_get_by_name);
3527 EXPORT_SYMBOL(dev_open);
3528 EXPORT_SYMBOL(dev_queue_xmit);
3529 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
3530 EXPORT_SYMBOL(dev_queue_xmit_nit);
3532 EXPORT_SYMBOL(dev_remove_pack);
3533 EXPORT_SYMBOL(dev_set_allmulti);
3534 EXPORT_SYMBOL(dev_set_promiscuity);
3535 EXPORT_SYMBOL(dev_change_flags);
3536 EXPORT_SYMBOL(dev_set_mtu);
3537 EXPORT_SYMBOL(dev_set_mac_address);
3538 EXPORT_SYMBOL(free_netdev);
3539 EXPORT_SYMBOL(netdev_boot_setup_check);
3540 EXPORT_SYMBOL(netdev_set_master);
3541 EXPORT_SYMBOL(netdev_state_change);
3542 EXPORT_SYMBOL(netif_receive_skb);
3543 EXPORT_SYMBOL(netif_rx);
3544 EXPORT_SYMBOL(register_gifconf);
3545 EXPORT_SYMBOL(register_netdevice);
3546 EXPORT_SYMBOL(register_netdevice_notifier);
3547 EXPORT_SYMBOL(skb_checksum_help);
3548 EXPORT_SYMBOL(synchronize_net);
3549 EXPORT_SYMBOL(unregister_netdevice);
3550 EXPORT_SYMBOL(unregister_netdevice_notifier);
3551 EXPORT_SYMBOL(net_enable_timestamp);
3552 EXPORT_SYMBOL(net_disable_timestamp);
3553 EXPORT_SYMBOL(dev_get_flags);
3554 EXPORT_SYMBOL(skb_checksum_setup);
3556 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3557 EXPORT_SYMBOL(br_handle_frame_hook);
3558 EXPORT_SYMBOL(br_fdb_get_hook);
3559 EXPORT_SYMBOL(br_fdb_put_hook);
3563 EXPORT_SYMBOL(dev_load);
3566 EXPORT_PER_CPU_SYMBOL(softnet_data);