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
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/config.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/notifier.h>
93 #include <linux/skbuff.h>
95 #include <linux/rtnetlink.h>
96 #include <linux/proc_fs.h>
97 #include <linux/seq_file.h>
98 #include <linux/stat.h>
99 #include <linux/if_bridge.h>
100 #include <linux/divert.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #ifdef CONFIG_NET_RADIO
113 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
114 #include <net/iw_handler.h>
115 #endif /* CONFIG_NET_RADIO */
116 #include <linux/vs_network.h>
117 #include <asm/current.h>
119 /* This define, if set, will randomly drop a packet when congestion
120 * is more than moderate. It helps fairness in the multi-interface
121 * case when one of them is a hog, but it kills performance for the
122 * single interface case so it is off now by default.
126 /* Setting this will sample the queue lengths and thus congestion
127 * via a timer instead of as each packet is received.
129 #undef OFFLINE_SAMPLE
132 * The list of packet types we will receive (as opposed to discard)
133 * and the routines to invoke.
135 * Why 16. Because with 16 the only overlap we get on a hash of the
136 * low nibble of the protocol value is RARP/SNAP/X.25.
138 * NOTE: That is no longer true with the addition of VLAN tags. Not
139 * sure which should go first, but I bet it won't make much
140 * difference if we are running VLANs. The good news is that
141 * this protocol won't be in the list unless compiled in, so
142 * the average user (w/out VLANs) will not be adversly affected.
159 static DEFINE_SPINLOCK(ptype_lock);
160 static struct list_head ptype_base[16]; /* 16 way hashed list */
161 static struct list_head ptype_all; /* Taps */
163 #ifdef OFFLINE_SAMPLE
164 static void sample_queue(unsigned long dummy);
165 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
169 * The @dev_base list is protected by @dev_base_lock and the rtln
172 * Pure readers hold dev_base_lock for reading.
174 * Writers must hold the rtnl semaphore while they loop through the
175 * dev_base list, and hold dev_base_lock for writing when they do the
176 * actual updates. This allows pure readers to access the list even
177 * while a writer is preparing to update it.
179 * To put it another way, dev_base_lock is held for writing only to
180 * protect against pure readers; the rtnl semaphore provides the
181 * protection against other writers.
183 * See, for example usages, register_netdevice() and
184 * unregister_netdevice(), which must be called with the rtnl
187 struct net_device *dev_base;
188 static struct net_device **dev_tail = &dev_base;
189 DEFINE_RWLOCK(dev_base_lock);
191 EXPORT_SYMBOL(dev_base);
192 EXPORT_SYMBOL(dev_base_lock);
194 #define NETDEV_HASHBITS 8
195 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
196 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
198 static inline struct hlist_head *dev_name_hash(const char *name)
200 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
201 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
204 static inline struct hlist_head *dev_index_hash(int ifindex)
206 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
213 static struct notifier_block *netdev_chain;
216 * Device drivers call our routines to queue packets here. We empty the
217 * queue in the local softnet handler.
219 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
222 extern int netdev_sysfs_init(void);
223 extern int netdev_register_sysfs(struct net_device *);
224 extern void netdev_unregister_sysfs(struct net_device *);
226 #define netdev_sysfs_init() (0)
227 #define netdev_register_sysfs(dev) (0)
228 #define netdev_unregister_sysfs(dev) do { } while(0)
232 /*******************************************************************************
234 Protocol management and registration routines
236 *******************************************************************************/
245 * Add a protocol ID to the list. Now that the input handler is
246 * smarter we can dispense with all the messy stuff that used to be
249 * BEWARE!!! Protocol handlers, mangling input packets,
250 * MUST BE last in hash buckets and checking protocol handlers
251 * MUST start from promiscuous ptype_all chain in net_bh.
252 * It is true now, do not change it.
253 * Explanation follows: if protocol handler, mangling packet, will
254 * be the first on list, it is not able to sense, that packet
255 * is cloned and should be copied-on-write, so that it will
256 * change it and subsequent readers will get broken packet.
261 * dev_add_pack - add packet handler
262 * @pt: packet type declaration
264 * Add a protocol handler to the networking stack. The passed &packet_type
265 * is linked into kernel lists and may not be freed until it has been
266 * removed from the kernel lists.
268 * This call does not sleep therefore it can not
269 * guarantee all CPU's that are in middle of receiving packets
270 * will see the new packet type (until the next received packet).
273 void dev_add_pack(struct packet_type *pt)
277 spin_lock_bh(&ptype_lock);
278 if (pt->type == htons(ETH_P_ALL)) {
280 list_add_rcu(&pt->list, &ptype_all);
282 hash = ntohs(pt->type) & 15;
283 list_add_rcu(&pt->list, &ptype_base[hash]);
285 spin_unlock_bh(&ptype_lock);
288 extern void linkwatch_run_queue(void);
293 * __dev_remove_pack - remove packet handler
294 * @pt: packet type declaration
296 * Remove a protocol handler that was previously added to the kernel
297 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
298 * from the kernel lists and can be freed or reused once this function
301 * The packet type might still be in use by receivers
302 * and must not be freed until after all the CPU's have gone
303 * through a quiescent state.
305 void __dev_remove_pack(struct packet_type *pt)
307 struct list_head *head;
308 struct packet_type *pt1;
310 spin_lock_bh(&ptype_lock);
312 if (pt->type == htons(ETH_P_ALL)) {
316 head = &ptype_base[ntohs(pt->type) & 15];
318 list_for_each_entry(pt1, head, list) {
320 list_del_rcu(&pt->list);
325 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
327 spin_unlock_bh(&ptype_lock);
330 * dev_remove_pack - remove packet handler
331 * @pt: packet type declaration
333 * Remove a protocol handler that was previously added to the kernel
334 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
335 * from the kernel lists and can be freed or reused once this function
338 * This call sleeps to guarantee that no CPU is looking at the packet
341 void dev_remove_pack(struct packet_type *pt)
343 __dev_remove_pack(pt);
348 /******************************************************************************
350 Device Boot-time Settings Routines
352 *******************************************************************************/
354 /* Boot time configuration table */
355 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
358 * netdev_boot_setup_add - add new setup entry
359 * @name: name of the device
360 * @map: configured settings for the device
362 * Adds new setup entry to the dev_boot_setup list. The function
363 * returns 0 on error and 1 on success. This is a generic routine to
366 static int netdev_boot_setup_add(char *name, struct ifmap *map)
368 struct netdev_boot_setup *s;
372 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
373 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
374 memset(s[i].name, 0, sizeof(s[i].name));
375 strcpy(s[i].name, name);
376 memcpy(&s[i].map, map, sizeof(s[i].map));
381 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
385 * netdev_boot_setup_check - check boot time settings
386 * @dev: the netdevice
388 * Check boot time settings for the device.
389 * The found settings are set for the device to be used
390 * later in the device probing.
391 * Returns 0 if no settings found, 1 if they are.
393 int netdev_boot_setup_check(struct net_device *dev)
395 struct netdev_boot_setup *s = dev_boot_setup;
398 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
399 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
400 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
401 dev->irq = s[i].map.irq;
402 dev->base_addr = s[i].map.base_addr;
403 dev->mem_start = s[i].map.mem_start;
404 dev->mem_end = s[i].map.mem_end;
413 * netdev_boot_base - get address from boot time settings
414 * @prefix: prefix for network device
415 * @unit: id for network device
417 * Check boot time settings for the base address of device.
418 * The found settings are set for the device to be used
419 * later in the device probing.
420 * Returns 0 if no settings found.
422 unsigned long netdev_boot_base(const char *prefix, int unit)
424 const struct netdev_boot_setup *s = dev_boot_setup;
428 sprintf(name, "%s%d", prefix, unit);
431 * If device already registered then return base of 1
432 * to indicate not to probe for this interface
434 if (__dev_get_by_name(name))
437 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
438 if (!strcmp(name, s[i].name))
439 return s[i].map.base_addr;
444 * Saves at boot time configured settings for any netdevice.
446 int __init netdev_boot_setup(char *str)
451 str = get_options(str, ARRAY_SIZE(ints), ints);
456 memset(&map, 0, sizeof(map));
460 map.base_addr = ints[2];
462 map.mem_start = ints[3];
464 map.mem_end = ints[4];
466 /* Add new entry to the list */
467 return netdev_boot_setup_add(str, &map);
470 __setup("netdev=", netdev_boot_setup);
472 /*******************************************************************************
474 Device Interface Subroutines
476 *******************************************************************************/
479 * __dev_get_by_name - find a device by its name
480 * @name: name to find
482 * Find an interface by name. Must be called under RTNL semaphore
483 * or @dev_base_lock. If the name is found a pointer to the device
484 * is returned. If the name is not found then %NULL is returned. The
485 * reference counters are not incremented so the caller must be
486 * careful with locks.
489 struct net_device *__dev_get_by_name(const char *name)
491 struct hlist_node *p;
493 hlist_for_each(p, dev_name_hash(name)) {
494 struct net_device *dev
495 = hlist_entry(p, struct net_device, name_hlist);
496 if (!strncmp(dev->name, name, IFNAMSIZ))
503 * dev_get_by_name - find a device by its name
504 * @name: name to find
506 * Find an interface by name. This can be called from any
507 * context and does its own locking. The returned handle has
508 * the usage count incremented and the caller must use dev_put() to
509 * release it when it is no longer needed. %NULL is returned if no
510 * matching device is found.
513 struct net_device *dev_get_by_name(const char *name)
515 struct net_device *dev;
517 read_lock(&dev_base_lock);
518 dev = __dev_get_by_name(name);
521 read_unlock(&dev_base_lock);
526 * __dev_get_by_index - find a device by its ifindex
527 * @ifindex: index of device
529 * Search for an interface by index. Returns %NULL if the device
530 * is not found or a pointer to the device. The device has not
531 * had its reference counter increased so the caller must be careful
532 * about locking. The caller must hold either the RTNL semaphore
536 struct net_device *__dev_get_by_index(int ifindex)
538 struct hlist_node *p;
540 hlist_for_each(p, dev_index_hash(ifindex)) {
541 struct net_device *dev
542 = hlist_entry(p, struct net_device, index_hlist);
543 if (dev->ifindex == ifindex)
551 * dev_get_by_index - find a device by its ifindex
552 * @ifindex: index of device
554 * Search for an interface by index. Returns NULL if the device
555 * is not found or a pointer to the device. The device returned has
556 * had a reference added and the pointer is safe until the user calls
557 * dev_put to indicate they have finished with it.
560 struct net_device *dev_get_by_index(int ifindex)
562 struct net_device *dev;
564 read_lock(&dev_base_lock);
565 dev = __dev_get_by_index(ifindex);
568 read_unlock(&dev_base_lock);
573 * dev_getbyhwaddr - find a device by its hardware address
574 * @type: media type of device
575 * @ha: hardware address
577 * Search for an interface by MAC address. Returns NULL if the device
578 * is not found or a pointer to the device. The caller must hold the
579 * rtnl semaphore. The returned device has not had its ref count increased
580 * and the caller must therefore be careful about locking
583 * If the API was consistent this would be __dev_get_by_hwaddr
586 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
588 struct net_device *dev;
592 for (dev = dev_base; dev; dev = dev->next)
593 if (dev->type == type &&
594 !memcmp(dev->dev_addr, ha, dev->addr_len))
599 struct net_device *dev_getfirstbyhwtype(unsigned short type)
601 struct net_device *dev;
604 for (dev = dev_base; dev; dev = dev->next) {
605 if (dev->type == type) {
614 EXPORT_SYMBOL(dev_getfirstbyhwtype);
617 * dev_get_by_flags - find any device with given flags
618 * @if_flags: IFF_* values
619 * @mask: bitmask of bits in if_flags to check
621 * Search for any interface with the given flags. Returns NULL if a device
622 * is not found or a pointer to the device. The device returned has
623 * had a reference added and the pointer is safe until the user calls
624 * dev_put to indicate they have finished with it.
627 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
629 struct net_device *dev;
631 read_lock(&dev_base_lock);
632 for (dev = dev_base; dev != NULL; dev = dev->next) {
633 if (((dev->flags ^ if_flags) & mask) == 0) {
638 read_unlock(&dev_base_lock);
643 * dev_valid_name - check if name is okay for network device
646 * Network device names need to be valid file names to
647 * to allow sysfs to work
649 static int dev_valid_name(const char *name)
651 return !(*name == '\0'
652 || !strcmp(name, ".")
653 || !strcmp(name, "..")
654 || strchr(name, '/'));
658 * dev_alloc_name - allocate a name for a device
660 * @name: name format string
662 * Passed a format string - eg "lt%d" it will try and find a suitable
663 * id. Not efficient for many devices, not called a lot. The caller
664 * must hold the dev_base or rtnl lock while allocating the name and
665 * adding the device in order to avoid duplicates. Returns the number
666 * of the unit assigned or a negative errno code.
669 int dev_alloc_name(struct net_device *dev, const char *name)
674 const int max_netdevices = 8*PAGE_SIZE;
676 struct net_device *d;
678 p = strnchr(name, IFNAMSIZ-1, '%');
681 * Verify the string as this thing may have come from
682 * the user. There must be either one "%d" and no other "%"
685 if (p[1] != 'd' || strchr(p + 2, '%'))
688 /* Use one page as a bit array of possible slots */
689 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
693 for (d = dev_base; d; d = d->next) {
694 if (!sscanf(d->name, name, &i))
696 if (i < 0 || i >= max_netdevices)
699 /* avoid cases where sscanf is not exact inverse of printf */
700 snprintf(buf, sizeof(buf), name, i);
701 if (!strncmp(buf, d->name, IFNAMSIZ))
705 i = find_first_zero_bit(inuse, max_netdevices);
706 free_page((unsigned long) inuse);
709 snprintf(buf, sizeof(buf), name, i);
710 if (!__dev_get_by_name(buf)) {
711 strlcpy(dev->name, buf, IFNAMSIZ);
715 /* It is possible to run out of possible slots
716 * when the name is long and there isn't enough space left
717 * for the digits, or if all bits are used.
724 * dev_change_name - change name of a device
726 * @newname: name (or format string) must be at least IFNAMSIZ
728 * Change name of a device, can pass format strings "eth%d".
731 int dev_change_name(struct net_device *dev, char *newname)
737 if (dev->flags & IFF_UP)
740 if (!dev_valid_name(newname))
743 if (strchr(newname, '%')) {
744 err = dev_alloc_name(dev, newname);
747 strcpy(newname, dev->name);
749 else if (__dev_get_by_name(newname))
752 strlcpy(dev->name, newname, IFNAMSIZ);
754 err = class_device_rename(&dev->class_dev, dev->name);
756 hlist_del(&dev->name_hlist);
757 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
758 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
765 * netdev_state_change - device changes state
766 * @dev: device to cause notification
768 * Called to indicate a device has changed state. This function calls
769 * the notifier chains for netdev_chain and sends a NEWLINK message
770 * to the routing socket.
772 void netdev_state_change(struct net_device *dev)
774 if (dev->flags & IFF_UP) {
775 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
776 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
781 * dev_load - load a network module
782 * @name: name of interface
784 * If a network interface is not present and the process has suitable
785 * privileges this function loads the module. If module loading is not
786 * available in this kernel then it becomes a nop.
789 void dev_load(const char *name)
791 struct net_device *dev;
793 read_lock(&dev_base_lock);
794 dev = __dev_get_by_name(name);
795 read_unlock(&dev_base_lock);
797 if (!dev && capable(CAP_SYS_MODULE))
798 request_module("%s", name);
801 static int default_rebuild_header(struct sk_buff *skb)
803 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
804 skb->dev ? skb->dev->name : "NULL!!!");
811 * dev_open - prepare an interface for use.
812 * @dev: device to open
814 * Takes a device from down to up state. The device's private open
815 * function is invoked and then the multicast lists are loaded. Finally
816 * the device is moved into the up state and a %NETDEV_UP message is
817 * sent to the netdev notifier chain.
819 * Calling this function on an active interface is a nop. On a failure
820 * a negative errno code is returned.
822 int dev_open(struct net_device *dev)
830 if (dev->flags & IFF_UP)
834 * Is it even present?
836 if (!netif_device_present(dev))
840 * Call device private open method
842 set_bit(__LINK_STATE_START, &dev->state);
844 ret = dev->open(dev);
846 clear_bit(__LINK_STATE_START, &dev->state);
850 * If it went open OK then:
857 dev->flags |= IFF_UP;
860 * Initialize multicasting status
865 * Wakeup transmit queue engine
870 * ... and announce new interface.
872 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
878 * dev_close - shutdown an interface.
879 * @dev: device to shutdown
881 * This function moves an active device into down state. A
882 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
883 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
886 int dev_close(struct net_device *dev)
888 if (!(dev->flags & IFF_UP))
892 * Tell people we are going down, so that they can
893 * prepare to death, when device is still operating.
895 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
899 clear_bit(__LINK_STATE_START, &dev->state);
901 /* Synchronize to scheduled poll. We cannot touch poll list,
902 * it can be even on different cpu. So just clear netif_running(),
903 * and wait when poll really will happen. Actually, the best place
904 * for this is inside dev->stop() after device stopped its irq
905 * engine, but this requires more changes in devices. */
907 smp_mb__after_clear_bit(); /* Commit netif_running(). */
908 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
910 current->state = TASK_INTERRUPTIBLE;
915 * Call the device specific close. This cannot fail.
916 * Only if device is UP
918 * We allow it to be called even after a DETACH hot-plug
925 * Device is now down.
928 dev->flags &= ~IFF_UP;
931 * Tell people we are down
933 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
940 * Device change register/unregister. These are not inline or static
941 * as we export them to the world.
945 * register_netdevice_notifier - register a network notifier block
948 * Register a notifier to be called when network device events occur.
949 * The notifier passed is linked into the kernel structures and must
950 * not be reused until it has been unregistered. A negative errno code
951 * is returned on a failure.
953 * When registered all registration and up events are replayed
954 * to the new notifier to allow device to have a race free
955 * view of the network device list.
958 int register_netdevice_notifier(struct notifier_block *nb)
960 struct net_device *dev;
964 err = notifier_chain_register(&netdev_chain, nb);
966 for (dev = dev_base; dev; dev = dev->next) {
967 nb->notifier_call(nb, NETDEV_REGISTER, dev);
969 if (dev->flags & IFF_UP)
970 nb->notifier_call(nb, NETDEV_UP, dev);
978 * unregister_netdevice_notifier - unregister a network notifier block
981 * Unregister a notifier previously registered by
982 * register_netdevice_notifier(). The notifier is unlinked into the
983 * kernel structures and may then be reused. A negative errno code
984 * is returned on a failure.
987 int unregister_netdevice_notifier(struct notifier_block *nb)
989 return notifier_chain_unregister(&netdev_chain, nb);
993 * call_netdevice_notifiers - call all network notifier blocks
994 * @val: value passed unmodified to notifier function
995 * @v: pointer passed unmodified to notifier function
997 * Call all network notifier blocks. Parameters and return value
998 * are as for notifier_call_chain().
1001 int call_netdevice_notifiers(unsigned long val, void *v)
1003 return notifier_call_chain(&netdev_chain, val, v);
1006 /* When > 0 there are consumers of rx skb time stamps */
1007 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1009 void net_enable_timestamp(void)
1011 atomic_inc(&netstamp_needed);
1014 void net_disable_timestamp(void)
1016 atomic_dec(&netstamp_needed);
1019 static inline void net_timestamp(struct timeval *stamp)
1021 if (atomic_read(&netstamp_needed))
1022 do_gettimeofday(stamp);
1030 * Support routine. Sends outgoing frames to any network
1031 * taps currently in use.
1034 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1036 struct packet_type *ptype;
1037 net_timestamp(&skb->stamp);
1040 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1041 /* Never send packets back to the socket
1042 * they originated from - MvS (miquels@drinkel.ow.org)
1044 if ((ptype->dev == dev || !ptype->dev) &&
1045 (ptype->af_packet_priv == NULL ||
1046 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1047 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1051 /* skb->nh should be correctly
1052 set by sender, so that the second statement is
1053 just protection against buggy protocols.
1055 skb2->mac.raw = skb2->data;
1057 if (skb2->nh.raw < skb2->data ||
1058 skb2->nh.raw > skb2->tail) {
1059 if (net_ratelimit())
1060 printk(KERN_CRIT "protocol %04x is "
1062 skb2->protocol, dev->name);
1063 skb2->nh.raw = skb2->data;
1066 skb2->h.raw = skb2->nh.raw;
1067 skb2->pkt_type = PACKET_OUTGOING;
1068 ptype->func(skb2, skb->dev, ptype);
1075 * Invalidate hardware checksum when packet is to be mangled, and
1076 * complete checksum manually on outgoing path.
1078 int skb_checksum_help(struct sk_buff *skb, int inward)
1081 int ret = 0, offset = skb->h.raw - skb->data;
1084 skb->ip_summed = CHECKSUM_NONE;
1088 if (skb_cloned(skb)) {
1089 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1094 if (offset > (int)skb->len)
1096 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1098 offset = skb->tail - skb->h.raw;
1101 if (skb->csum + 2 > offset)
1104 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1105 skb->ip_summed = CHECKSUM_NONE;
1110 #ifdef CONFIG_HIGHMEM
1111 /* Actually, we should eliminate this check as soon as we know, that:
1112 * 1. IOMMU is present and allows to map all the memory.
1113 * 2. No high memory really exists on this machine.
1116 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1120 if (dev->features & NETIF_F_HIGHDMA)
1123 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1124 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1130 #define illegal_highdma(dev, skb) (0)
1133 extern void skb_release_data(struct sk_buff *);
1135 /* Keep head the same: replace data */
1136 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1141 struct skb_shared_info *ninfo;
1142 int headerlen = skb->data - skb->head;
1143 int expand = (skb->tail + skb->data_len) - skb->end;
1145 if (skb_shared(skb))
1151 size = skb->end - skb->head + expand;
1152 size = SKB_DATA_ALIGN(size);
1153 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1157 /* Copy entire thing */
1158 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1162 ninfo = (struct skb_shared_info*)(data + size);
1163 atomic_set(&ninfo->dataref, 1);
1164 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1165 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1166 ninfo->nr_frags = 0;
1167 ninfo->frag_list = NULL;
1169 /* Offset between the two in bytes */
1170 offset = data - skb->head;
1172 /* Free old data. */
1173 skb_release_data(skb);
1176 skb->end = data + size;
1178 /* Set up new pointers */
1179 skb->h.raw += offset;
1180 skb->nh.raw += offset;
1181 skb->mac.raw += offset;
1182 skb->tail += offset;
1183 skb->data += offset;
1185 /* We are no longer a clone, even if we were. */
1188 skb->tail += skb->data_len;
1193 #define HARD_TX_LOCK(dev, cpu) { \
1194 if ((dev->features & NETIF_F_LLTX) == 0) { \
1195 spin_lock(&dev->xmit_lock); \
1196 dev->xmit_lock_owner = cpu; \
1200 #define HARD_TX_UNLOCK(dev) { \
1201 if ((dev->features & NETIF_F_LLTX) == 0) { \
1202 dev->xmit_lock_owner = -1; \
1203 spin_unlock(&dev->xmit_lock); \
1208 * dev_queue_xmit - transmit a buffer
1209 * @skb: buffer to transmit
1211 * Queue a buffer for transmission to a network device. The caller must
1212 * have set the device and priority and built the buffer before calling
1213 * this function. The function can be called from an interrupt.
1215 * A negative errno code is returned on a failure. A success does not
1216 * guarantee the frame will be transmitted as it may be dropped due
1217 * to congestion or traffic shaping.
1220 int dev_queue_xmit(struct sk_buff *skb)
1222 struct net_device *dev = skb->dev;
1226 if (skb_shinfo(skb)->frag_list &&
1227 !(dev->features & NETIF_F_FRAGLIST) &&
1228 __skb_linearize(skb, GFP_ATOMIC))
1231 /* Fragmented skb is linearized if device does not support SG,
1232 * or if at least one of fragments is in highmem and device
1233 * does not support DMA from it.
1235 if (skb_shinfo(skb)->nr_frags &&
1236 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1237 __skb_linearize(skb, GFP_ATOMIC))
1240 /* If packet is not checksummed and device does not support
1241 * checksumming for this protocol, complete checksumming here.
1243 if (skb->ip_summed == CHECKSUM_HW &&
1244 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1245 (!(dev->features & NETIF_F_IP_CSUM) ||
1246 skb->protocol != htons(ETH_P_IP))))
1247 if (skb_checksum_help(skb, 0))
1250 /* Disable soft irqs for various locks below. Also
1251 * stops preemption for RCU.
1255 /* Updates of qdisc are serialized by queue_lock.
1256 * The struct Qdisc which is pointed to by qdisc is now a
1257 * rcu structure - it may be accessed without acquiring
1258 * a lock (but the structure may be stale.) The freeing of the
1259 * qdisc will be deferred until it's known that there are no
1260 * more references to it.
1262 * If the qdisc has an enqueue function, we still need to
1263 * hold the queue_lock before calling it, since queue_lock
1264 * also serializes access to the device queue.
1267 q = rcu_dereference(dev->qdisc);
1268 #ifdef CONFIG_NET_CLS_ACT
1269 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1272 /* Grab device queue */
1273 spin_lock(&dev->queue_lock);
1275 rc = q->enqueue(skb, q);
1279 spin_unlock(&dev->queue_lock);
1280 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1284 /* The device has no queue. Common case for software devices:
1285 loopback, all the sorts of tunnels...
1287 Really, it is unlikely that xmit_lock protection is necessary here.
1288 (f.e. loopback and IP tunnels are clean ignoring statistics
1290 However, it is possible, that they rely on protection
1293 Check this and shot the lock. It is not prone from deadlocks.
1294 Either shot noqueue qdisc, it is even simpler 8)
1296 if (dev->flags & IFF_UP) {
1297 int cpu = smp_processor_id(); /* ok because BHs are off */
1299 if (dev->xmit_lock_owner != cpu) {
1301 HARD_TX_LOCK(dev, cpu);
1303 if (!netif_queue_stopped(dev)) {
1305 dev_queue_xmit_nit(skb, dev);
1308 if (!dev->hard_start_xmit(skb, dev)) {
1309 HARD_TX_UNLOCK(dev);
1313 HARD_TX_UNLOCK(dev);
1314 if (net_ratelimit())
1315 printk(KERN_CRIT "Virtual device %s asks to "
1316 "queue packet!\n", dev->name);
1318 /* Recursion is detected! It is possible,
1320 if (net_ratelimit())
1321 printk(KERN_CRIT "Dead loop on virtual device "
1322 "%s, fix it urgently!\n", dev->name);
1338 /*=======================================================================
1340 =======================================================================*/
1342 int netdev_max_backlog = 300;
1343 int weight_p = 64; /* old backlog weight */
1344 /* These numbers are selected based on intuition and some
1345 * experimentatiom, if you have more scientific way of doing this
1346 * please go ahead and fix things.
1348 int no_cong_thresh = 10;
1353 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1356 static void get_sample_stats(int cpu)
1362 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1363 int blog = sd->input_pkt_queue.qlen;
1364 int avg_blog = sd->avg_blog;
1366 avg_blog = (avg_blog >> 1) + (blog >> 1);
1368 if (avg_blog > mod_cong) {
1369 /* Above moderate congestion levels. */
1370 sd->cng_level = NET_RX_CN_HIGH;
1373 rq = rd % netdev_max_backlog;
1374 if (rq < avg_blog) /* unlucky bastard */
1375 sd->cng_level = NET_RX_DROP;
1377 } else if (avg_blog > lo_cong) {
1378 sd->cng_level = NET_RX_CN_MOD;
1381 rq = rd % netdev_max_backlog;
1382 if (rq < avg_blog) /* unlucky bastard */
1383 sd->cng_level = NET_RX_CN_HIGH;
1385 } else if (avg_blog > no_cong)
1386 sd->cng_level = NET_RX_CN_LOW;
1387 else /* no congestion */
1388 sd->cng_level = NET_RX_SUCCESS;
1390 sd->avg_blog = avg_blog;
1393 #ifdef OFFLINE_SAMPLE
1394 static void sample_queue(unsigned long dummy)
1396 /* 10 ms 0r 1ms -- i don't care -- JHS */
1398 int cpu = smp_processor_id();
1400 get_sample_stats(cpu);
1401 next_tick += jiffies;
1402 mod_timer(&samp_timer, next_tick);
1408 * netif_rx - post buffer to the network code
1409 * @skb: buffer to post
1411 * This function receives a packet from a device driver and queues it for
1412 * the upper (protocol) levels to process. It always succeeds. The buffer
1413 * may be dropped during processing for congestion control or by the
1417 * NET_RX_SUCCESS (no congestion)
1418 * NET_RX_CN_LOW (low congestion)
1419 * NET_RX_CN_MOD (moderate congestion)
1420 * NET_RX_CN_HIGH (high congestion)
1421 * NET_RX_DROP (packet was dropped)
1425 int netif_rx(struct sk_buff *skb)
1428 struct softnet_data *queue;
1429 unsigned long flags;
1431 #ifdef CONFIG_NETPOLL
1432 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1438 if (!skb->stamp.tv_sec)
1439 net_timestamp(&skb->stamp);
1442 * The code is rearranged so that the path is the most
1443 * short when CPU is congested, but is still operating.
1445 local_irq_save(flags);
1446 this_cpu = smp_processor_id();
1447 queue = &__get_cpu_var(softnet_data);
1449 __get_cpu_var(netdev_rx_stat).total++;
1450 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1451 if (queue->input_pkt_queue.qlen) {
1452 if (queue->throttle)
1457 __skb_queue_tail(&queue->input_pkt_queue, skb);
1458 #ifndef OFFLINE_SAMPLE
1459 get_sample_stats(this_cpu);
1461 local_irq_restore(flags);
1462 return queue->cng_level;
1465 if (queue->throttle)
1466 queue->throttle = 0;
1468 netif_rx_schedule(&queue->backlog_dev);
1472 if (!queue->throttle) {
1473 queue->throttle = 1;
1474 __get_cpu_var(netdev_rx_stat).throttled++;
1478 __get_cpu_var(netdev_rx_stat).dropped++;
1479 local_irq_restore(flags);
1485 int netif_rx_ni(struct sk_buff *skb)
1490 err = netif_rx(skb);
1491 if (local_softirq_pending())
1498 EXPORT_SYMBOL(netif_rx_ni);
1500 static __inline__ void skb_bond(struct sk_buff *skb)
1502 struct net_device *dev = skb->dev;
1505 skb->real_dev = skb->dev;
1506 skb->dev = dev->master;
1510 static void net_tx_action(struct softirq_action *h)
1512 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1514 if (sd->completion_queue) {
1515 struct sk_buff *clist;
1517 local_irq_disable();
1518 clist = sd->completion_queue;
1519 sd->completion_queue = NULL;
1523 struct sk_buff *skb = clist;
1524 clist = clist->next;
1526 BUG_TRAP(!atomic_read(&skb->users));
1531 if (sd->output_queue) {
1532 struct net_device *head;
1534 local_irq_disable();
1535 head = sd->output_queue;
1536 sd->output_queue = NULL;
1540 struct net_device *dev = head;
1541 head = head->next_sched;
1543 smp_mb__before_clear_bit();
1544 clear_bit(__LINK_STATE_SCHED, &dev->state);
1546 if (spin_trylock(&dev->queue_lock)) {
1548 spin_unlock(&dev->queue_lock);
1550 netif_schedule(dev);
1556 static __inline__ int deliver_skb(struct sk_buff *skb,
1557 struct packet_type *pt_prev)
1559 atomic_inc(&skb->users);
1560 return pt_prev->func(skb, skb->dev, pt_prev);
1563 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1564 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1566 static __inline__ int handle_bridge(struct sk_buff **pskb,
1567 struct packet_type **pt_prev, int *ret)
1569 struct net_bridge_port *port;
1571 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1572 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1576 *ret = deliver_skb(*pskb, *pt_prev);
1580 return br_handle_frame_hook(port, pskb);
1583 #define handle_bridge(skb, pt_prev, ret) (0)
1586 #ifdef CONFIG_NET_CLS_ACT
1587 /* TODO: Maybe we should just force sch_ingress to be compiled in
1588 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1589 * a compare and 2 stores extra right now if we dont have it on
1590 * but have CONFIG_NET_CLS_ACT
1591 * NOTE: This doesnt stop any functionality; if you dont have
1592 * the ingress scheduler, you just cant add policies on ingress.
1595 static int ing_filter(struct sk_buff *skb)
1598 struct net_device *dev = skb->dev;
1599 int result = TC_ACT_OK;
1601 if (dev->qdisc_ingress) {
1602 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1603 if (MAX_RED_LOOP < ttl++) {
1604 printk("Redir loop detected Dropping packet (%s->%s)\n",
1605 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1609 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1611 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1612 if (NULL == skb->input_dev) {
1613 skb->input_dev = skb->dev;
1614 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1616 spin_lock(&dev->ingress_lock);
1617 if ((q = dev->qdisc_ingress) != NULL)
1618 result = q->enqueue(skb, q);
1619 spin_unlock(&dev->ingress_lock);
1627 int netif_receive_skb(struct sk_buff *skb)
1629 struct packet_type *ptype, *pt_prev;
1630 int ret = NET_RX_DROP;
1631 unsigned short type;
1633 #ifdef CONFIG_NETPOLL
1634 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1640 if (!skb->stamp.tv_sec)
1641 net_timestamp(&skb->stamp);
1645 __get_cpu_var(netdev_rx_stat).total++;
1647 skb->h.raw = skb->nh.raw = skb->data;
1648 skb->mac_len = skb->nh.raw - skb->mac.raw;
1654 #ifdef CONFIG_NET_CLS_ACT
1655 if (skb->tc_verd & TC_NCLS) {
1656 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1661 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1662 if (!ptype->dev || ptype->dev == skb->dev) {
1664 ret = deliver_skb(skb, pt_prev);
1669 #ifdef CONFIG_NET_CLS_ACT
1671 ret = deliver_skb(skb, pt_prev);
1672 pt_prev = NULL; /* noone else should process this after*/
1674 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1677 ret = ing_filter(skb);
1679 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1688 handle_diverter(skb);
1690 if (handle_bridge(&skb, &pt_prev, &ret))
1693 type = skb->protocol;
1694 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1695 if (ptype->type == type &&
1696 (!ptype->dev || ptype->dev == skb->dev)) {
1698 ret = deliver_skb(skb, pt_prev);
1704 ret = pt_prev->func(skb, skb->dev, pt_prev);
1707 /* Jamal, now you will not able to escape explaining
1708 * me how you were going to use this. :-)
1718 static int process_backlog(struct net_device *backlog_dev, int *budget)
1721 int quota = min(backlog_dev->quota, *budget);
1722 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1723 unsigned long start_time = jiffies;
1726 struct sk_buff *skb;
1727 struct net_device *dev;
1729 local_irq_disable();
1730 skb = __skb_dequeue(&queue->input_pkt_queue);
1737 netif_receive_skb(skb);
1743 if (work >= quota || jiffies - start_time > 1)
1748 backlog_dev->quota -= work;
1753 backlog_dev->quota -= work;
1756 list_del(&backlog_dev->poll_list);
1757 smp_mb__before_clear_bit();
1758 netif_poll_enable(backlog_dev);
1760 if (queue->throttle)
1761 queue->throttle = 0;
1766 static void net_rx_action(struct softirq_action *h)
1768 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1769 unsigned long start_time = jiffies;
1770 int budget = netdev_max_backlog;
1773 local_irq_disable();
1775 while (!list_empty(&queue->poll_list)) {
1776 struct net_device *dev;
1778 if (budget <= 0 || jiffies - start_time > 1)
1783 dev = list_entry(queue->poll_list.next,
1784 struct net_device, poll_list);
1786 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1787 local_irq_disable();
1788 list_del(&dev->poll_list);
1789 list_add_tail(&dev->poll_list, &queue->poll_list);
1791 dev->quota += dev->weight;
1793 dev->quota = dev->weight;
1796 local_irq_disable();
1804 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1805 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1809 static gifconf_func_t * gifconf_list [NPROTO];
1812 * register_gifconf - register a SIOCGIF handler
1813 * @family: Address family
1814 * @gifconf: Function handler
1816 * Register protocol dependent address dumping routines. The handler
1817 * that is passed must not be freed or reused until it has been replaced
1818 * by another handler.
1820 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1822 if (family >= NPROTO)
1824 gifconf_list[family] = gifconf;
1830 * Map an interface index to its name (SIOCGIFNAME)
1834 * We need this ioctl for efficient implementation of the
1835 * if_indextoname() function required by the IPv6 API. Without
1836 * it, we would have to search all the interfaces to find a
1840 static int dev_ifname(struct ifreq __user *arg)
1842 struct net_device *dev;
1846 * Fetch the caller's info block.
1849 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1852 read_lock(&dev_base_lock);
1853 dev = __dev_get_by_index(ifr.ifr_ifindex);
1855 read_unlock(&dev_base_lock);
1859 strcpy(ifr.ifr_name, dev->name);
1860 read_unlock(&dev_base_lock);
1862 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1868 * Perform a SIOCGIFCONF call. This structure will change
1869 * size eventually, and there is nothing I can do about it.
1870 * Thus we will need a 'compatibility mode'.
1873 static int dev_ifconf(char __user *arg)
1876 struct net_device *dev;
1883 * Fetch the caller's info block.
1886 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1893 * Loop over the interfaces, and write an info block for each.
1897 for (dev = dev_base; dev; dev = dev->next) {
1898 if (vx_flags(VXF_HIDE_NETIF, 0) &&
1899 !dev_in_nx_info(dev, current->nx_info))
1901 for (i = 0; i < NPROTO; i++) {
1902 if (gifconf_list[i]) {
1905 done = gifconf_list[i](dev, NULL, 0);
1907 done = gifconf_list[i](dev, pos + total,
1917 * All done. Write the updated control block back to the caller.
1919 ifc.ifc_len = total;
1922 * Both BSD and Solaris return 0 here, so we do too.
1924 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1927 #ifdef CONFIG_PROC_FS
1929 * This is invoked by the /proc filesystem handler to display a device
1932 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1934 struct net_device *dev;
1937 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1939 return i == pos ? dev : NULL;
1942 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1944 read_lock(&dev_base_lock);
1945 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1948 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1951 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1954 void dev_seq_stop(struct seq_file *seq, void *v)
1956 read_unlock(&dev_base_lock);
1959 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1961 struct nx_info *nxi = current->nx_info;
1963 if (vx_flags(VXF_HIDE_NETIF, 0) && !dev_in_nx_info(dev, nxi))
1965 if (dev->get_stats) {
1966 struct net_device_stats *stats = dev->get_stats(dev);
1968 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1969 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1970 dev->name, stats->rx_bytes, stats->rx_packets,
1972 stats->rx_dropped + stats->rx_missed_errors,
1973 stats->rx_fifo_errors,
1974 stats->rx_length_errors + stats->rx_over_errors +
1975 stats->rx_crc_errors + stats->rx_frame_errors,
1976 stats->rx_compressed, stats->multicast,
1977 stats->tx_bytes, stats->tx_packets,
1978 stats->tx_errors, stats->tx_dropped,
1979 stats->tx_fifo_errors, stats->collisions,
1980 stats->tx_carrier_errors +
1981 stats->tx_aborted_errors +
1982 stats->tx_window_errors +
1983 stats->tx_heartbeat_errors,
1984 stats->tx_compressed);
1986 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
1990 * Called from the PROCfs module. This now uses the new arbitrary sized
1991 * /proc/net interface to create /proc/net/dev
1993 static int dev_seq_show(struct seq_file *seq, void *v)
1995 if (v == SEQ_START_TOKEN)
1996 seq_puts(seq, "Inter-| Receive "
1998 " face |bytes packets errs drop fifo frame "
1999 "compressed multicast|bytes packets errs "
2000 "drop fifo colls carrier compressed\n");
2002 dev_seq_printf_stats(seq, v);
2006 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2008 struct netif_rx_stats *rc = NULL;
2010 while (*pos < NR_CPUS)
2011 if (cpu_online(*pos)) {
2012 rc = &per_cpu(netdev_rx_stat, *pos);
2019 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2021 return softnet_get_online(pos);
2024 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2027 return softnet_get_online(pos);
2030 static void softnet_seq_stop(struct seq_file *seq, void *v)
2034 static int softnet_seq_show(struct seq_file *seq, void *v)
2036 struct netif_rx_stats *s = v;
2038 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2039 s->total, s->dropped, s->time_squeeze, s->throttled,
2040 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2041 s->fastroute_deferred_out,
2043 s->fastroute_latency_reduction
2051 static struct seq_operations dev_seq_ops = {
2052 .start = dev_seq_start,
2053 .next = dev_seq_next,
2054 .stop = dev_seq_stop,
2055 .show = dev_seq_show,
2058 static int dev_seq_open(struct inode *inode, struct file *file)
2060 return seq_open(file, &dev_seq_ops);
2063 static struct file_operations dev_seq_fops = {
2064 .owner = THIS_MODULE,
2065 .open = dev_seq_open,
2067 .llseek = seq_lseek,
2068 .release = seq_release,
2071 static struct seq_operations softnet_seq_ops = {
2072 .start = softnet_seq_start,
2073 .next = softnet_seq_next,
2074 .stop = softnet_seq_stop,
2075 .show = softnet_seq_show,
2078 static int softnet_seq_open(struct inode *inode, struct file *file)
2080 return seq_open(file, &softnet_seq_ops);
2083 static struct file_operations softnet_seq_fops = {
2084 .owner = THIS_MODULE,
2085 .open = softnet_seq_open,
2087 .llseek = seq_lseek,
2088 .release = seq_release,
2092 extern int wireless_proc_init(void);
2094 #define wireless_proc_init() 0
2097 static int __init dev_proc_init(void)
2101 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2103 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2105 if (wireless_proc_init())
2111 proc_net_remove("softnet_stat");
2113 proc_net_remove("dev");
2117 #define dev_proc_init() 0
2118 #endif /* CONFIG_PROC_FS */
2122 * netdev_set_master - set up master/slave pair
2123 * @slave: slave device
2124 * @master: new master device
2126 * Changes the master device of the slave. Pass %NULL to break the
2127 * bonding. The caller must hold the RTNL semaphore. On a failure
2128 * a negative errno code is returned. On success the reference counts
2129 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2130 * function returns zero.
2132 int netdev_set_master(struct net_device *slave, struct net_device *master)
2134 struct net_device *old = slave->master;
2144 slave->master = master;
2152 slave->flags |= IFF_SLAVE;
2154 slave->flags &= ~IFF_SLAVE;
2156 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2161 * dev_set_promiscuity - update promiscuity count on a device
2165 * Add or remove promsicuity from a device. While the count in the device
2166 * remains above zero the interface remains promiscuous. Once it hits zero
2167 * the device reverts back to normal filtering operation. A negative inc
2168 * value is used to drop promiscuity on the device.
2170 void dev_set_promiscuity(struct net_device *dev, int inc)
2172 unsigned short old_flags = dev->flags;
2174 dev->flags |= IFF_PROMISC;
2175 if ((dev->promiscuity += inc) == 0)
2176 dev->flags &= ~IFF_PROMISC;
2177 if (dev->flags ^ old_flags) {
2179 printk(KERN_INFO "device %s %s promiscuous mode\n",
2180 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2186 * dev_set_allmulti - update allmulti count on a device
2190 * Add or remove reception of all multicast frames to a device. While the
2191 * count in the device remains above zero the interface remains listening
2192 * to all interfaces. Once it hits zero the device reverts back to normal
2193 * filtering operation. A negative @inc value is used to drop the counter
2194 * when releasing a resource needing all multicasts.
2197 void dev_set_allmulti(struct net_device *dev, int inc)
2199 unsigned short old_flags = dev->flags;
2201 dev->flags |= IFF_ALLMULTI;
2202 if ((dev->allmulti += inc) == 0)
2203 dev->flags &= ~IFF_ALLMULTI;
2204 if (dev->flags ^ old_flags)
2208 unsigned dev_get_flags(const struct net_device *dev)
2212 flags = (dev->flags & ~(IFF_PROMISC |
2215 (dev->gflags & (IFF_PROMISC |
2218 if (netif_running(dev) && netif_carrier_ok(dev))
2219 flags |= IFF_RUNNING;
2224 int dev_change_flags(struct net_device *dev, unsigned flags)
2227 int old_flags = dev->flags;
2230 * Set the flags on our device.
2233 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2234 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2236 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2240 * Load in the correct multicast list now the flags have changed.
2246 * Have we downed the interface. We handle IFF_UP ourselves
2247 * according to user attempts to set it, rather than blindly
2252 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2253 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2259 if (dev->flags & IFF_UP &&
2260 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2262 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2264 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2265 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2266 dev->gflags ^= IFF_PROMISC;
2267 dev_set_promiscuity(dev, inc);
2270 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2271 is important. Some (broken) drivers set IFF_PROMISC, when
2272 IFF_ALLMULTI is requested not asking us and not reporting.
2274 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2275 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2276 dev->gflags ^= IFF_ALLMULTI;
2277 dev_set_allmulti(dev, inc);
2280 if (old_flags ^ dev->flags)
2281 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2286 int dev_set_mtu(struct net_device *dev, int new_mtu)
2290 if (new_mtu == dev->mtu)
2293 /* MTU must be positive. */
2297 if (!netif_device_present(dev))
2301 if (dev->change_mtu)
2302 err = dev->change_mtu(dev, new_mtu);
2305 if (!err && dev->flags & IFF_UP)
2306 notifier_call_chain(&netdev_chain,
2307 NETDEV_CHANGEMTU, dev);
2313 * Perform the SIOCxIFxxx calls.
2315 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2318 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2324 case SIOCGIFFLAGS: /* Get interface flags */
2325 ifr->ifr_flags = dev_get_flags(dev);
2328 case SIOCSIFFLAGS: /* Set interface flags */
2329 return dev_change_flags(dev, ifr->ifr_flags);
2331 case SIOCGIFMETRIC: /* Get the metric on the interface
2332 (currently unused) */
2333 ifr->ifr_metric = 0;
2336 case SIOCSIFMETRIC: /* Set the metric on the interface
2337 (currently unused) */
2340 case SIOCGIFMTU: /* Get the MTU of a device */
2341 ifr->ifr_mtu = dev->mtu;
2344 case SIOCSIFMTU: /* Set the MTU of a device */
2345 return dev_set_mtu(dev, ifr->ifr_mtu);
2349 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2351 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2352 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2353 ifr->ifr_hwaddr.sa_family = dev->type;
2357 if (!dev->set_mac_address)
2359 if (ifr->ifr_hwaddr.sa_family != dev->type)
2361 if (!netif_device_present(dev))
2363 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2365 notifier_call_chain(&netdev_chain,
2366 NETDEV_CHANGEADDR, dev);
2369 case SIOCSIFHWBROADCAST:
2370 if (ifr->ifr_hwaddr.sa_family != dev->type)
2372 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2373 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2374 notifier_call_chain(&netdev_chain,
2375 NETDEV_CHANGEADDR, dev);
2379 ifr->ifr_map.mem_start = dev->mem_start;
2380 ifr->ifr_map.mem_end = dev->mem_end;
2381 ifr->ifr_map.base_addr = dev->base_addr;
2382 ifr->ifr_map.irq = dev->irq;
2383 ifr->ifr_map.dma = dev->dma;
2384 ifr->ifr_map.port = dev->if_port;
2388 if (dev->set_config) {
2389 if (!netif_device_present(dev))
2391 return dev->set_config(dev, &ifr->ifr_map);
2396 if (!dev->set_multicast_list ||
2397 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2399 if (!netif_device_present(dev))
2401 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2405 if (!dev->set_multicast_list ||
2406 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2408 if (!netif_device_present(dev))
2410 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2414 ifr->ifr_ifindex = dev->ifindex;
2418 ifr->ifr_qlen = dev->tx_queue_len;
2422 if (ifr->ifr_qlen < 0)
2424 dev->tx_queue_len = ifr->ifr_qlen;
2428 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2429 return dev_change_name(dev, ifr->ifr_newname);
2432 * Unknown or private ioctl
2436 if ((cmd >= SIOCDEVPRIVATE &&
2437 cmd <= SIOCDEVPRIVATE + 15) ||
2438 cmd == SIOCBONDENSLAVE ||
2439 cmd == SIOCBONDRELEASE ||
2440 cmd == SIOCBONDSETHWADDR ||
2441 cmd == SIOCBONDSLAVEINFOQUERY ||
2442 cmd == SIOCBONDINFOQUERY ||
2443 cmd == SIOCBONDCHANGEACTIVE ||
2444 cmd == SIOCGMIIPHY ||
2445 cmd == SIOCGMIIREG ||
2446 cmd == SIOCSMIIREG ||
2447 cmd == SIOCBRADDIF ||
2448 cmd == SIOCBRDELIF ||
2449 cmd == SIOCWANDEV) {
2451 if (dev->do_ioctl) {
2452 if (netif_device_present(dev))
2453 err = dev->do_ioctl(dev, ifr,
2466 * This function handles all "interface"-type I/O control requests. The actual
2467 * 'doing' part of this is dev_ifsioc above.
2471 * dev_ioctl - network device ioctl
2472 * @cmd: command to issue
2473 * @arg: pointer to a struct ifreq in user space
2475 * Issue ioctl functions to devices. This is normally called by the
2476 * user space syscall interfaces but can sometimes be useful for
2477 * other purposes. The return value is the return from the syscall if
2478 * positive or a negative errno code on error.
2481 int dev_ioctl(unsigned int cmd, void __user *arg)
2487 /* One special case: SIOCGIFCONF takes ifconf argument
2488 and requires shared lock, because it sleeps writing
2492 if (cmd == SIOCGIFCONF) {
2494 ret = dev_ifconf((char __user *) arg);
2498 if (cmd == SIOCGIFNAME)
2499 return dev_ifname((struct ifreq __user *)arg);
2501 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2504 ifr.ifr_name[IFNAMSIZ-1] = 0;
2506 colon = strchr(ifr.ifr_name, ':');
2511 * See which interface the caller is talking about.
2516 * These ioctl calls:
2517 * - can be done by all.
2518 * - atomic and do not require locking.
2529 dev_load(ifr.ifr_name);
2530 read_lock(&dev_base_lock);
2531 ret = dev_ifsioc(&ifr, cmd);
2532 read_unlock(&dev_base_lock);
2536 if (copy_to_user(arg, &ifr,
2537 sizeof(struct ifreq)))
2543 dev_load(ifr.ifr_name);
2545 ret = dev_ethtool(&ifr);
2550 if (copy_to_user(arg, &ifr,
2551 sizeof(struct ifreq)))
2557 * These ioctl calls:
2558 * - require superuser power.
2559 * - require strict serialization.
2565 if (!capable(CAP_NET_ADMIN))
2567 dev_load(ifr.ifr_name);
2569 ret = dev_ifsioc(&ifr, cmd);
2574 if (copy_to_user(arg, &ifr,
2575 sizeof(struct ifreq)))
2581 * These ioctl calls:
2582 * - require superuser power.
2583 * - require strict serialization.
2584 * - do not return a value
2594 case SIOCSIFHWBROADCAST:
2597 case SIOCBONDENSLAVE:
2598 case SIOCBONDRELEASE:
2599 case SIOCBONDSETHWADDR:
2600 case SIOCBONDSLAVEINFOQUERY:
2601 case SIOCBONDINFOQUERY:
2602 case SIOCBONDCHANGEACTIVE:
2605 if (!capable(CAP_NET_ADMIN))
2607 dev_load(ifr.ifr_name);
2609 ret = dev_ifsioc(&ifr, cmd);
2614 /* Get the per device memory space. We can add this but
2615 * currently do not support it */
2617 /* Set the per device memory buffer space.
2618 * Not applicable in our case */
2623 * Unknown or private ioctl.
2626 if (cmd == SIOCWANDEV ||
2627 (cmd >= SIOCDEVPRIVATE &&
2628 cmd <= SIOCDEVPRIVATE + 15)) {
2629 dev_load(ifr.ifr_name);
2631 ret = dev_ifsioc(&ifr, cmd);
2633 if (!ret && copy_to_user(arg, &ifr,
2634 sizeof(struct ifreq)))
2639 /* Take care of Wireless Extensions */
2640 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2641 /* If command is `set a parameter', or
2642 * `get the encoding parameters', check if
2643 * the user has the right to do it */
2644 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2645 if (!capable(CAP_NET_ADMIN))
2648 dev_load(ifr.ifr_name);
2650 /* Follow me in net/core/wireless.c */
2651 ret = wireless_process_ioctl(&ifr, cmd);
2653 if (IW_IS_GET(cmd) &&
2654 copy_to_user(arg, &ifr,
2655 sizeof(struct ifreq)))
2659 #endif /* WIRELESS_EXT */
2666 * dev_new_index - allocate an ifindex
2668 * Returns a suitable unique value for a new device interface
2669 * number. The caller must hold the rtnl semaphore or the
2670 * dev_base_lock to be sure it remains unique.
2672 static int dev_new_index(void)
2678 if (!__dev_get_by_index(ifindex))
2683 static int dev_boot_phase = 1;
2685 /* Delayed registration/unregisteration */
2686 static DEFINE_SPINLOCK(net_todo_list_lock);
2687 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2689 static inline void net_set_todo(struct net_device *dev)
2691 spin_lock(&net_todo_list_lock);
2692 list_add_tail(&dev->todo_list, &net_todo_list);
2693 spin_unlock(&net_todo_list_lock);
2697 * register_netdevice - register a network device
2698 * @dev: device to register
2700 * Take a completed network device structure and add it to the kernel
2701 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2702 * chain. 0 is returned on success. A negative errno code is returned
2703 * on a failure to set up the device, or if the name is a duplicate.
2705 * Callers must hold the rtnl semaphore. You may want
2706 * register_netdev() instead of this.
2709 * The locking appears insufficient to guarantee two parallel registers
2710 * will not get the same name.
2713 int register_netdevice(struct net_device *dev)
2715 struct hlist_head *head;
2716 struct hlist_node *p;
2719 BUG_ON(dev_boot_phase);
2722 /* When net_device's are persistent, this will be fatal. */
2723 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2725 spin_lock_init(&dev->queue_lock);
2726 spin_lock_init(&dev->xmit_lock);
2727 dev->xmit_lock_owner = -1;
2728 #ifdef CONFIG_NET_CLS_ACT
2729 spin_lock_init(&dev->ingress_lock);
2732 ret = alloc_divert_blk(dev);
2738 /* Init, if this function is available */
2740 ret = dev->init(dev);
2748 if (!dev_valid_name(dev->name)) {
2753 dev->ifindex = dev_new_index();
2754 if (dev->iflink == -1)
2755 dev->iflink = dev->ifindex;
2757 /* Check for existence of name */
2758 head = dev_name_hash(dev->name);
2759 hlist_for_each(p, head) {
2760 struct net_device *d
2761 = hlist_entry(p, struct net_device, name_hlist);
2762 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2768 /* Fix illegal SG+CSUM combinations. */
2769 if ((dev->features & NETIF_F_SG) &&
2770 !(dev->features & (NETIF_F_IP_CSUM |
2772 NETIF_F_HW_CSUM))) {
2773 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2775 dev->features &= ~NETIF_F_SG;
2778 /* TSO requires that SG is present as well. */
2779 if ((dev->features & NETIF_F_TSO) &&
2780 !(dev->features & NETIF_F_SG)) {
2781 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2783 dev->features &= ~NETIF_F_TSO;
2787 * nil rebuild_header routine,
2788 * that should be never called and used as just bug trap.
2791 if (!dev->rebuild_header)
2792 dev->rebuild_header = default_rebuild_header;
2795 * Default initial state at registry is that the
2796 * device is present.
2799 set_bit(__LINK_STATE_PRESENT, &dev->state);
2802 dev_init_scheduler(dev);
2803 write_lock_bh(&dev_base_lock);
2805 dev_tail = &dev->next;
2806 hlist_add_head(&dev->name_hlist, head);
2807 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2809 dev->reg_state = NETREG_REGISTERING;
2810 write_unlock_bh(&dev_base_lock);
2812 /* Notify protocols, that a new device appeared. */
2813 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2815 /* Finish registration after unlock */
2822 free_divert_blk(dev);
2827 * register_netdev - register a network device
2828 * @dev: device to register
2830 * Take a completed network device structure and add it to the kernel
2831 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2832 * chain. 0 is returned on success. A negative errno code is returned
2833 * on a failure to set up the device, or if the name is a duplicate.
2835 * This is a wrapper around register_netdev that takes the rtnl semaphore
2836 * and expands the device name if you passed a format string to
2839 int register_netdev(struct net_device *dev)
2846 * If the name is a format string the caller wants us to do a
2849 if (strchr(dev->name, '%')) {
2850 err = dev_alloc_name(dev, dev->name);
2856 * Back compatibility hook. Kill this one in 2.5
2858 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2859 err = dev_alloc_name(dev, "eth%d");
2864 err = register_netdevice(dev);
2869 EXPORT_SYMBOL(register_netdev);
2872 * netdev_wait_allrefs - wait until all references are gone.
2874 * This is called when unregistering network devices.
2876 * Any protocol or device that holds a reference should register
2877 * for netdevice notification, and cleanup and put back the
2878 * reference if they receive an UNREGISTER event.
2879 * We can get stuck here if buggy protocols don't correctly
2882 static void netdev_wait_allrefs(struct net_device *dev)
2884 unsigned long rebroadcast_time, warning_time;
2886 rebroadcast_time = warning_time = jiffies;
2887 while (atomic_read(&dev->refcnt) != 0) {
2888 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2891 /* Rebroadcast unregister notification */
2892 notifier_call_chain(&netdev_chain,
2893 NETDEV_UNREGISTER, dev);
2895 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2897 /* We must not have linkwatch events
2898 * pending on unregister. If this
2899 * happens, we simply run the queue
2900 * unscheduled, resulting in a noop
2903 linkwatch_run_queue();
2908 rebroadcast_time = jiffies;
2913 if (time_after(jiffies, warning_time + 10 * HZ)) {
2914 printk(KERN_EMERG "unregister_netdevice: "
2915 "waiting for %s to become free. Usage "
2917 dev->name, atomic_read(&dev->refcnt));
2918 warning_time = jiffies;
2927 * register_netdevice(x1);
2928 * register_netdevice(x2);
2930 * unregister_netdevice(y1);
2931 * unregister_netdevice(y2);
2937 * We are invoked by rtnl_unlock() after it drops the semaphore.
2938 * This allows us to deal with problems:
2939 * 1) We can create/delete sysfs objects which invoke hotplug
2940 * without deadlocking with linkwatch via keventd.
2941 * 2) Since we run with the RTNL semaphore not held, we can sleep
2942 * safely in order to wait for the netdev refcnt to drop to zero.
2944 static DECLARE_MUTEX(net_todo_run_mutex);
2945 void netdev_run_todo(void)
2947 struct list_head list = LIST_HEAD_INIT(list);
2951 /* Need to guard against multiple cpu's getting out of order. */
2952 down(&net_todo_run_mutex);
2954 /* Not safe to do outside the semaphore. We must not return
2955 * until all unregister events invoked by the local processor
2956 * have been completed (either by this todo run, or one on
2959 if (list_empty(&net_todo_list))
2962 /* Snapshot list, allow later requests */
2963 spin_lock(&net_todo_list_lock);
2964 list_splice_init(&net_todo_list, &list);
2965 spin_unlock(&net_todo_list_lock);
2967 while (!list_empty(&list)) {
2968 struct net_device *dev
2969 = list_entry(list.next, struct net_device, todo_list);
2970 list_del(&dev->todo_list);
2972 switch(dev->reg_state) {
2973 case NETREG_REGISTERING:
2974 err = netdev_register_sysfs(dev);
2976 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
2978 dev->reg_state = NETREG_REGISTERED;
2981 case NETREG_UNREGISTERING:
2982 netdev_unregister_sysfs(dev);
2983 dev->reg_state = NETREG_UNREGISTERED;
2985 netdev_wait_allrefs(dev);
2988 BUG_ON(atomic_read(&dev->refcnt));
2989 BUG_TRAP(!dev->ip_ptr);
2990 BUG_TRAP(!dev->ip6_ptr);
2991 BUG_TRAP(!dev->dn_ptr);
2994 /* It must be the very last action,
2995 * after this 'dev' may point to freed up memory.
2997 if (dev->destructor)
2998 dev->destructor(dev);
3002 printk(KERN_ERR "network todo '%s' but state %d\n",
3003 dev->name, dev->reg_state);
3009 up(&net_todo_run_mutex);
3013 * alloc_netdev - allocate network device
3014 * @sizeof_priv: size of private data to allocate space for
3015 * @name: device name format string
3016 * @setup: callback to initialize device
3018 * Allocates a struct net_device with private data area for driver use
3019 * and performs basic initialization.
3021 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3022 void (*setup)(struct net_device *))
3025 struct net_device *dev;
3028 /* ensure 32-byte alignment of both the device and private area */
3029 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3030 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3032 p = kmalloc(alloc_size, GFP_KERNEL);
3034 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3037 memset(p, 0, alloc_size);
3039 dev = (struct net_device *)
3040 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3041 dev->padded = (char *)dev - (char *)p;
3044 dev->priv = netdev_priv(dev);
3047 strcpy(dev->name, name);
3050 EXPORT_SYMBOL(alloc_netdev);
3053 * free_netdev - free network device
3056 * This function does the last stage of destroying an allocated device
3057 * interface. The reference to the device object is released.
3058 * If this is the last reference then it will be freed.
3060 void free_netdev(struct net_device *dev)
3063 /* Compatiablity with error handling in drivers */
3064 if (dev->reg_state == NETREG_UNINITIALIZED) {
3065 kfree((char *)dev - dev->padded);
3069 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3070 dev->reg_state = NETREG_RELEASED;
3072 /* will free via class release */
3073 class_device_put(&dev->class_dev);
3075 kfree((char *)dev - dev->padded);
3079 /* Synchronize with packet receive processing. */
3080 void synchronize_net(void)
3083 synchronize_kernel();
3087 * unregister_netdevice - remove device from the kernel
3090 * This function shuts down a device interface and removes it
3091 * from the kernel tables. On success 0 is returned, on a failure
3092 * a negative errno code is returned.
3094 * Callers must hold the rtnl semaphore. You may want
3095 * unregister_netdev() instead of this.
3098 int unregister_netdevice(struct net_device *dev)
3100 struct net_device *d, **dp;
3102 BUG_ON(dev_boot_phase);
3105 /* Some devices call without registering for initialization unwind. */
3106 if (dev->reg_state == NETREG_UNINITIALIZED) {
3107 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3108 "was registered\n", dev->name, dev);
3112 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3114 /* If device is running, close it first. */
3115 if (dev->flags & IFF_UP)
3118 /* And unlink it from device chain. */
3119 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3121 write_lock_bh(&dev_base_lock);
3122 hlist_del(&dev->name_hlist);
3123 hlist_del(&dev->index_hlist);
3124 if (dev_tail == &dev->next)
3127 write_unlock_bh(&dev_base_lock);
3132 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3137 dev->reg_state = NETREG_UNREGISTERING;
3141 /* Shutdown queueing discipline. */
3145 /* Notify protocols, that we are about to destroy
3146 this device. They should clean all the things.
3148 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3151 * Flush the multicast chain
3153 dev_mc_discard(dev);
3158 /* Notifier chain MUST detach us from master device. */
3159 BUG_TRAP(!dev->master);
3161 free_divert_blk(dev);
3163 /* Finish processing unregister after unlock */
3173 * unregister_netdev - remove device from the kernel
3176 * This function shuts down a device interface and removes it
3177 * from the kernel tables. On success 0 is returned, on a failure
3178 * a negative errno code is returned.
3180 * This is just a wrapper for unregister_netdevice that takes
3181 * the rtnl semaphore. In general you want to use this and not
3182 * unregister_netdevice.
3184 void unregister_netdev(struct net_device *dev)
3187 unregister_netdevice(dev);
3191 EXPORT_SYMBOL(unregister_netdev);
3193 #ifdef CONFIG_HOTPLUG_CPU
3194 static int dev_cpu_callback(struct notifier_block *nfb,
3195 unsigned long action,
3198 struct sk_buff **list_skb;
3199 struct net_device **list_net;
3200 struct sk_buff *skb;
3201 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3202 struct softnet_data *sd, *oldsd;
3204 if (action != CPU_DEAD)
3207 local_irq_disable();
3208 cpu = smp_processor_id();
3209 sd = &per_cpu(softnet_data, cpu);
3210 oldsd = &per_cpu(softnet_data, oldcpu);
3212 /* Find end of our completion_queue. */
3213 list_skb = &sd->completion_queue;
3215 list_skb = &(*list_skb)->next;
3216 /* Append completion queue from offline CPU. */
3217 *list_skb = oldsd->completion_queue;
3218 oldsd->completion_queue = NULL;
3220 /* Find end of our output_queue. */
3221 list_net = &sd->output_queue;
3223 list_net = &(*list_net)->next_sched;
3224 /* Append output queue from offline CPU. */
3225 *list_net = oldsd->output_queue;
3226 oldsd->output_queue = NULL;
3228 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3231 /* Process offline CPU's input_pkt_queue */
3232 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3237 #endif /* CONFIG_HOTPLUG_CPU */
3241 * Initialize the DEV module. At boot time this walks the device list and
3242 * unhooks any devices that fail to initialise (normally hardware not
3243 * present) and leaves us with a valid list of present and active devices.
3248 * This is called single threaded during boot, so no need
3249 * to take the rtnl semaphore.
3251 static int __init net_dev_init(void)
3253 int i, rc = -ENOMEM;
3255 BUG_ON(!dev_boot_phase);
3259 if (dev_proc_init())
3262 if (netdev_sysfs_init())
3265 INIT_LIST_HEAD(&ptype_all);
3266 for (i = 0; i < 16; i++)
3267 INIT_LIST_HEAD(&ptype_base[i]);
3269 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3270 INIT_HLIST_HEAD(&dev_name_head[i]);
3272 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3273 INIT_HLIST_HEAD(&dev_index_head[i]);
3276 * Initialise the packet receive queues.
3279 for (i = 0; i < NR_CPUS; i++) {
3280 struct softnet_data *queue;
3282 queue = &per_cpu(softnet_data, i);
3283 skb_queue_head_init(&queue->input_pkt_queue);
3284 queue->throttle = 0;
3285 queue->cng_level = 0;
3286 queue->avg_blog = 10; /* arbitrary non-zero */
3287 queue->completion_queue = NULL;
3288 INIT_LIST_HEAD(&queue->poll_list);
3289 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3290 queue->backlog_dev.weight = weight_p;
3291 queue->backlog_dev.poll = process_backlog;
3292 atomic_set(&queue->backlog_dev.refcnt, 1);
3295 #ifdef OFFLINE_SAMPLE
3296 samp_timer.expires = jiffies + (10 * HZ);
3297 add_timer(&samp_timer);
3302 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3303 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3305 hotcpu_notifier(dev_cpu_callback, 0);
3313 subsys_initcall(net_dev_init);
3315 EXPORT_SYMBOL(__dev_get_by_index);
3316 EXPORT_SYMBOL(__dev_get_by_name);
3317 EXPORT_SYMBOL(__dev_remove_pack);
3318 EXPORT_SYMBOL(__skb_linearize);
3319 EXPORT_SYMBOL(dev_add_pack);
3320 EXPORT_SYMBOL(dev_alloc_name);
3321 EXPORT_SYMBOL(dev_close);
3322 EXPORT_SYMBOL(dev_get_by_flags);
3323 EXPORT_SYMBOL(dev_get_by_index);
3324 EXPORT_SYMBOL(dev_get_by_name);
3325 EXPORT_SYMBOL(dev_ioctl);
3326 EXPORT_SYMBOL(dev_open);
3327 EXPORT_SYMBOL(dev_queue_xmit);
3328 EXPORT_SYMBOL(dev_remove_pack);
3329 EXPORT_SYMBOL(dev_set_allmulti);
3330 EXPORT_SYMBOL(dev_set_promiscuity);
3331 EXPORT_SYMBOL(dev_change_flags);
3332 EXPORT_SYMBOL(dev_set_mtu);
3333 EXPORT_SYMBOL(free_netdev);
3334 EXPORT_SYMBOL(netdev_boot_setup_check);
3335 EXPORT_SYMBOL(netdev_set_master);
3336 EXPORT_SYMBOL(netdev_state_change);
3337 EXPORT_SYMBOL(netif_receive_skb);
3338 EXPORT_SYMBOL(netif_rx);
3339 EXPORT_SYMBOL(register_gifconf);
3340 EXPORT_SYMBOL(register_netdevice);
3341 EXPORT_SYMBOL(register_netdevice_notifier);
3342 EXPORT_SYMBOL(skb_checksum_help);
3343 EXPORT_SYMBOL(synchronize_net);
3344 EXPORT_SYMBOL(unregister_netdevice);
3345 EXPORT_SYMBOL(unregister_netdevice_notifier);
3346 EXPORT_SYMBOL(net_enable_timestamp);
3347 EXPORT_SYMBOL(net_disable_timestamp);
3349 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3350 EXPORT_SYMBOL(br_handle_frame_hook);
3354 EXPORT_SYMBOL(dev_load);
3357 EXPORT_PER_CPU_SYMBOL(softnet_data);