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 <asm/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 #ifdef CONFIG_NET_RADIO
112 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
113 #include <net/iw_handler.h>
114 #endif /* CONFIG_NET_RADIO */
115 #include <asm/current.h>
116 #include <linux/vs_base.h>
117 #include <linux/vs_network.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 spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
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 struct net_device **dev_tail = &dev_base;
189 rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
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 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 Return value is changed to int to prevent illegal usage in future.
527 It is still legal to use to check for device existence.
529 User should understand, that the result returned by this function
530 is meaningless, if it was not issued under rtnl semaphore.
534 * dev_get - test if a device exists
535 * @name: name to test for
537 * Test if a name exists. Returns true if the name is found. In order
538 * to be sure the name is not allocated or removed during the test the
539 * caller must hold the rtnl semaphore.
541 * This function exists only for back compatibility with older
544 int __dev_get(const char *name)
546 struct net_device *dev;
548 read_lock(&dev_base_lock);
549 dev = __dev_get_by_name(name);
550 read_unlock(&dev_base_lock);
555 * __dev_get_by_index - find a device by its ifindex
556 * @ifindex: index of device
558 * Search for an interface by index. Returns %NULL if the device
559 * is not found or a pointer to the device. The device has not
560 * had its reference counter increased so the caller must be careful
561 * about locking. The caller must hold either the RTNL semaphore
565 struct net_device *__dev_get_by_index(int ifindex)
567 struct hlist_node *p;
569 hlist_for_each(p, dev_index_hash(ifindex)) {
570 struct net_device *dev
571 = hlist_entry(p, struct net_device, index_hlist);
572 if (dev->ifindex == ifindex)
580 * dev_get_by_index - find a device by its ifindex
581 * @ifindex: index of device
583 * Search for an interface by index. Returns NULL if the device
584 * is not found or a pointer to the device. The device returned has
585 * had a reference added and the pointer is safe until the user calls
586 * dev_put to indicate they have finished with it.
589 struct net_device *dev_get_by_index(int ifindex)
591 struct net_device *dev;
593 read_lock(&dev_base_lock);
594 dev = __dev_get_by_index(ifindex);
597 read_unlock(&dev_base_lock);
602 * dev_getbyhwaddr - find a device by its hardware address
603 * @type: media type of device
604 * @ha: hardware address
606 * Search for an interface by MAC address. Returns NULL if the device
607 * is not found or a pointer to the device. The caller must hold the
608 * rtnl semaphore. The returned device has not had its ref count increased
609 * and the caller must therefore be careful about locking
612 * If the API was consistent this would be __dev_get_by_hwaddr
615 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
617 struct net_device *dev;
621 for (dev = dev_base; dev; dev = dev->next)
622 if (dev->type == type &&
623 !memcmp(dev->dev_addr, ha, dev->addr_len))
628 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
630 struct net_device *dev;
632 for (dev = dev_base; dev; dev = dev->next)
633 if (dev->type == type)
638 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
640 struct net_device *dev_getfirstbyhwtype(unsigned short type)
642 struct net_device *dev;
645 dev = __dev_getfirstbyhwtype(type);
652 EXPORT_SYMBOL(dev_getfirstbyhwtype);
655 * dev_get_by_flags - find any device with given flags
656 * @if_flags: IFF_* values
657 * @mask: bitmask of bits in if_flags to check
659 * Search for any interface with the given flags. Returns NULL if a device
660 * is not found or a pointer to the device. The device returned has
661 * had a reference added and the pointer is safe until the user calls
662 * dev_put to indicate they have finished with it.
665 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
667 struct net_device *dev;
669 read_lock(&dev_base_lock);
670 dev = __dev_get_by_flags(if_flags, mask);
673 read_unlock(&dev_base_lock);
678 * __dev_get_by_flags - find any device with given flags
679 * @if_flags: IFF_* values
680 * @mask: bitmask of bits in if_flags to check
682 * Search for any interface with the given flags. Returns NULL if a device
683 * is not found or a pointer to the device. The caller must hold either
684 * the RTNL semaphore or @dev_base_lock.
687 struct net_device *__dev_get_by_flags(unsigned short if_flags, unsigned short mask)
689 struct net_device *dev;
691 for (dev = dev_base; dev != NULL; dev = dev->next) {
692 if (((dev->flags ^ if_flags) & mask) == 0)
699 * dev_valid_name - check if name is okay for network device
702 * Network device names need to be valid file names to
703 * to allow sysfs to work
705 int dev_valid_name(const char *name)
707 return !(*name == '\0'
708 || !strcmp(name, ".")
709 || !strcmp(name, "..")
710 || strchr(name, '/'));
714 * dev_alloc_name - allocate a name for a device
716 * @name: name format string
718 * Passed a format string - eg "lt%d" it will try and find a suitable
719 * id. Not efficient for many devices, not called a lot. The caller
720 * must hold the dev_base or rtnl lock while allocating the name and
721 * adding the device in order to avoid duplicates. Returns the number
722 * of the unit assigned or a negative errno code.
725 int dev_alloc_name(struct net_device *dev, const char *name)
730 const int max_netdevices = 8*PAGE_SIZE;
732 struct net_device *d;
734 p = strnchr(name, IFNAMSIZ-1, '%');
737 * Verify the string as this thing may have come from
738 * the user. There must be either one "%d" and no other "%"
741 if (p[1] != 'd' || strchr(p + 2, '%'))
744 /* Use one page as a bit array of possible slots */
745 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
749 for (d = dev_base; d; d = d->next) {
750 if (!sscanf(d->name, name, &i))
752 if (i < 0 || i >= max_netdevices)
755 /* avoid cases where sscanf is not exact inverse of printf */
756 snprintf(buf, sizeof(buf), name, i);
757 if (!strncmp(buf, d->name, IFNAMSIZ))
761 i = find_first_zero_bit(inuse, max_netdevices);
762 free_page((unsigned long) inuse);
765 snprintf(buf, sizeof(buf), name, i);
766 if (!__dev_get_by_name(buf)) {
767 strlcpy(dev->name, buf, IFNAMSIZ);
771 /* It is possible to run out of possible slots
772 * when the name is long and there isn't enough space left
773 * for the digits, or if all bits are used.
780 * dev_change_name - change name of a device
782 * @newname: name (or format string) must be at least IFNAMSIZ
784 * Change name of a device, can pass format strings "eth%d".
787 int dev_change_name(struct net_device *dev, char *newname)
793 if (dev->flags & IFF_UP)
796 if (!dev_valid_name(newname))
799 if (strchr(newname, '%')) {
800 err = dev_alloc_name(dev, newname);
803 strcpy(newname, dev->name);
805 else if (__dev_get_by_name(newname))
808 strlcpy(dev->name, newname, IFNAMSIZ);
810 err = class_device_rename(&dev->class_dev, dev->name);
812 hlist_del(&dev->name_hlist);
813 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
814 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
821 * netdev_state_change - device changes state
822 * @dev: device to cause notification
824 * Called to indicate a device has changed state. This function calls
825 * the notifier chains for netdev_chain and sends a NEWLINK message
826 * to the routing socket.
828 void netdev_state_change(struct net_device *dev)
830 if (dev->flags & IFF_UP) {
831 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
832 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
837 * dev_load - load a network module
838 * @name: name of interface
840 * If a network interface is not present and the process has suitable
841 * privileges this function loads the module. If module loading is not
842 * available in this kernel then it becomes a nop.
845 void dev_load(const char *name)
847 struct net_device *dev;
849 read_lock(&dev_base_lock);
850 dev = __dev_get_by_name(name);
851 read_unlock(&dev_base_lock);
853 if (!dev && capable(CAP_SYS_MODULE))
854 request_module("%s", name);
857 static int default_rebuild_header(struct sk_buff *skb)
859 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
860 skb->dev ? skb->dev->name : "NULL!!!");
867 * dev_open - prepare an interface for use.
868 * @dev: device to open
870 * Takes a device from down to up state. The device's private open
871 * function is invoked and then the multicast lists are loaded. Finally
872 * the device is moved into the up state and a %NETDEV_UP message is
873 * sent to the netdev notifier chain.
875 * Calling this function on an active interface is a nop. On a failure
876 * a negative errno code is returned.
878 int dev_open(struct net_device *dev)
886 if (dev->flags & IFF_UP)
890 * Is it even present?
892 if (!netif_device_present(dev))
896 * Call device private open method
898 set_bit(__LINK_STATE_START, &dev->state);
900 ret = dev->open(dev);
902 clear_bit(__LINK_STATE_START, &dev->state);
906 * If it went open OK then:
913 dev->flags |= IFF_UP;
916 * Initialize multicasting status
921 * Wakeup transmit queue engine
926 * ... and announce new interface.
928 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
934 * dev_close - shutdown an interface.
935 * @dev: device to shutdown
937 * This function moves an active device into down state. A
938 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
939 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
942 int dev_close(struct net_device *dev)
944 if (!(dev->flags & IFF_UP))
948 * Tell people we are going down, so that they can
949 * prepare to death, when device is still operating.
951 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
955 clear_bit(__LINK_STATE_START, &dev->state);
957 /* Synchronize to scheduled poll. We cannot touch poll list,
958 * it can be even on different cpu. So just clear netif_running(),
959 * and wait when poll really will happen. Actually, the best place
960 * for this is inside dev->stop() after device stopped its irq
961 * engine, but this requires more changes in devices. */
963 smp_mb__after_clear_bit(); /* Commit netif_running(). */
964 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
966 current->state = TASK_INTERRUPTIBLE;
971 * Call the device specific close. This cannot fail.
972 * Only if device is UP
974 * We allow it to be called even after a DETACH hot-plug
981 * Device is now down.
984 dev->flags &= ~IFF_UP;
987 * Tell people we are down
989 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
996 * Device change register/unregister. These are not inline or static
997 * as we export them to the world.
1001 * register_netdevice_notifier - register a network notifier block
1004 * Register a notifier to be called when network device events occur.
1005 * The notifier passed is linked into the kernel structures and must
1006 * not be reused until it has been unregistered. A negative errno code
1007 * is returned on a failure.
1009 * When registered all registration and up events are replayed
1010 * to the new notifier to allow device to have a race free
1011 * view of the network device list.
1014 int register_netdevice_notifier(struct notifier_block *nb)
1016 struct net_device *dev;
1020 err = notifier_chain_register(&netdev_chain, nb);
1022 for (dev = dev_base; dev; dev = dev->next) {
1023 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1025 if (dev->flags & IFF_UP)
1026 nb->notifier_call(nb, NETDEV_UP, dev);
1034 * unregister_netdevice_notifier - unregister a network notifier block
1037 * Unregister a notifier previously registered by
1038 * register_netdevice_notifier(). The notifier is unlinked into the
1039 * kernel structures and may then be reused. A negative errno code
1040 * is returned on a failure.
1043 int unregister_netdevice_notifier(struct notifier_block *nb)
1045 return notifier_chain_unregister(&netdev_chain, nb);
1049 * call_netdevice_notifiers - call all network notifier blocks
1050 * @val: value passed unmodified to notifier function
1051 * @v: pointer passed unmodified to notifier function
1053 * Call all network notifier blocks. Parameters and return value
1054 * are as for notifier_call_chain().
1057 int call_netdevice_notifiers(unsigned long val, void *v)
1059 return notifier_call_chain(&netdev_chain, val, v);
1063 * Support routine. Sends outgoing frames to any network
1064 * taps currently in use.
1067 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1069 struct packet_type *ptype;
1070 net_timestamp(&skb->stamp);
1073 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1074 /* Never send packets back to the socket
1075 * they originated from - MvS (miquels@drinkel.ow.org)
1077 if ((ptype->dev == dev || !ptype->dev) &&
1078 (ptype->af_packet_priv == NULL ||
1079 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1080 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1084 /* skb->nh should be correctly
1085 set by sender, so that the second statement is
1086 just protection against buggy protocols.
1088 skb2->mac.raw = skb2->data;
1090 if (skb2->nh.raw < skb2->data ||
1091 skb2->nh.raw > skb2->tail) {
1092 if (net_ratelimit())
1093 printk(KERN_CRIT "protocol %04x is "
1095 skb2->protocol, dev->name);
1096 skb2->nh.raw = skb2->data;
1099 skb2->h.raw = skb2->nh.raw;
1100 skb2->pkt_type = PACKET_OUTGOING;
1101 ptype->func(skb2, skb->dev, ptype);
1108 * Invalidate hardware checksum when packet is to be mangled, and
1109 * complete checksum manually on outgoing path.
1111 int skb_checksum_help(struct sk_buff **pskb, int inward)
1114 int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1117 (*pskb)->ip_summed = CHECKSUM_NONE;
1121 if (skb_cloned(*pskb)) {
1122 ret = pskb_expand_head(*pskb, 0, 0, GFP_ATOMIC);
1127 if (offset > (int)(*pskb)->len)
1129 csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1131 offset = (*pskb)->tail - (*pskb)->h.raw;
1134 if ((*pskb)->csum + 2 > offset)
1137 *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1138 (*pskb)->ip_summed = CHECKSUM_NONE;
1143 #ifdef CONFIG_HIGHMEM
1144 /* Actually, we should eliminate this check as soon as we know, that:
1145 * 1. IOMMU is present and allows to map all the memory.
1146 * 2. No high memory really exists on this machine.
1149 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1153 if (dev->features & NETIF_F_HIGHDMA)
1156 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1157 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1163 #define illegal_highdma(dev, skb) (0)
1166 extern void skb_release_data(struct sk_buff *);
1168 /* Keep head the same: replace data */
1169 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1174 struct skb_shared_info *ninfo;
1175 int headerlen = skb->data - skb->head;
1176 int expand = (skb->tail + skb->data_len) - skb->end;
1178 if (skb_shared(skb))
1184 size = skb->end - skb->head + expand;
1185 size = SKB_DATA_ALIGN(size);
1186 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1190 /* Copy entire thing */
1191 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1195 ninfo = (struct skb_shared_info*)(data + size);
1196 atomic_set(&ninfo->dataref, 1);
1197 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1198 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1199 ninfo->nr_frags = 0;
1200 ninfo->frag_list = NULL;
1202 /* Offset between the two in bytes */
1203 offset = data - skb->head;
1205 /* Free old data. */
1206 skb_release_data(skb);
1209 skb->end = data + size;
1211 /* Set up new pointers */
1212 skb->h.raw += offset;
1213 skb->nh.raw += offset;
1214 skb->mac.raw += offset;
1215 skb->tail += offset;
1216 skb->data += offset;
1218 /* We are no longer a clone, even if we were. */
1221 skb->tail += skb->data_len;
1226 #define HARD_TX_LOCK(dev, cpu) { \
1227 if ((dev->features & NETIF_F_LLTX) == 0) { \
1228 spin_lock(&dev->xmit_lock); \
1229 dev->xmit_lock_owner = cpu; \
1233 #define HARD_TX_UNLOCK(dev) { \
1234 if ((dev->features & NETIF_F_LLTX) == 0) { \
1235 dev->xmit_lock_owner = -1; \
1236 spin_unlock(&dev->xmit_lock); \
1240 static inline void qdisc_run(struct net_device *dev)
1242 while (!netif_queue_stopped(dev) &&
1243 qdisc_restart(dev)<0)
1248 * dev_queue_xmit - transmit a buffer
1249 * @skb: buffer to transmit
1251 * Queue a buffer for transmission to a network device. The caller must
1252 * have set the device and priority and built the buffer before calling
1253 * this function. The function can be called from an interrupt.
1255 * A negative errno code is returned on a failure. A success does not
1256 * guarantee the frame will be transmitted as it may be dropped due
1257 * to congestion or traffic shaping.
1260 int dev_queue_xmit(struct sk_buff *skb)
1262 struct net_device *dev = skb->dev;
1266 if (skb_shinfo(skb)->frag_list &&
1267 !(dev->features & NETIF_F_FRAGLIST) &&
1268 __skb_linearize(skb, GFP_ATOMIC))
1271 /* Fragmented skb is linearized if device does not support SG,
1272 * or if at least one of fragments is in highmem and device
1273 * does not support DMA from it.
1275 if (skb_shinfo(skb)->nr_frags &&
1276 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1277 __skb_linearize(skb, GFP_ATOMIC))
1280 /* If packet is not checksummed and device does not support
1281 * checksumming for this protocol, complete checksumming here.
1283 if (skb->ip_summed == CHECKSUM_HW &&
1284 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1285 (!(dev->features & NETIF_F_IP_CSUM) ||
1286 skb->protocol != htons(ETH_P_IP))))
1287 if (skb_checksum_help(&skb, 0))
1291 /* Disable soft irqs for various locks below. Also
1292 * stops preemption for RCU.
1296 /* Updates of qdisc are serialized by queue_lock.
1297 * The struct Qdisc which is pointed to by qdisc is now a
1298 * rcu structure - it may be accessed without acquiring
1299 * a lock (but the structure may be stale.) The freeing of the
1300 * qdisc will be deferred until it's known that there are no
1301 * more references to it.
1303 * If the qdisc has an enqueue function, we still need to
1304 * hold the queue_lock before calling it, since queue_lock
1305 * also serializes access to the device queue.
1308 q = rcu_dereference(dev->qdisc);
1309 #ifdef CONFIG_NET_CLS_ACT
1310 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1313 /* Grab device queue */
1314 spin_lock(&dev->queue_lock);
1316 rc = q->enqueue(skb, q);
1320 spin_unlock(&dev->queue_lock);
1321 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1325 /* The device has no queue. Common case for software devices:
1326 loopback, all the sorts of tunnels...
1328 Really, it is unlikely that xmit_lock protection is necessary here.
1329 (f.e. loopback and IP tunnels are clean ignoring statistics
1331 However, it is possible, that they rely on protection
1334 Check this and shot the lock. It is not prone from deadlocks.
1335 Either shot noqueue qdisc, it is even simpler 8)
1337 if (dev->flags & IFF_UP) {
1338 int cpu = smp_processor_id(); /* ok because BHs are off */
1340 if (dev->xmit_lock_owner != cpu) {
1342 HARD_TX_LOCK(dev, cpu);
1344 if (!netif_queue_stopped(dev)) {
1346 dev_queue_xmit_nit(skb, dev);
1349 if (!dev->hard_start_xmit(skb, dev)) {
1350 HARD_TX_UNLOCK(dev);
1354 HARD_TX_UNLOCK(dev);
1355 if (net_ratelimit())
1356 printk(KERN_CRIT "Virtual device %s asks to "
1357 "queue packet!\n", dev->name);
1360 /* Recursion is detected! It is possible,
1362 if (net_ratelimit())
1363 printk(KERN_CRIT "Dead loop on virtual device "
1364 "%s, fix it urgently!\n", dev->name);
1377 /*=======================================================================
1379 =======================================================================*/
1381 int netdev_max_backlog = 300;
1382 int weight_p = 64; /* old backlog weight */
1383 /* These numbers are selected based on intuition and some
1384 * experimentatiom, if you have more scientific way of doing this
1385 * please go ahead and fix things.
1387 int no_cong_thresh = 10;
1392 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1395 #ifdef CONFIG_NET_HW_FLOWCONTROL
1396 atomic_t netdev_dropping = ATOMIC_INIT(0);
1397 static unsigned long netdev_fc_mask = 1;
1398 unsigned long netdev_fc_xoff;
1399 spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1403 void (*stimul)(struct net_device *);
1404 struct net_device *dev;
1405 } netdev_fc_slots[BITS_PER_LONG];
1407 int netdev_register_fc(struct net_device *dev,
1408 void (*stimul)(struct net_device *dev))
1411 unsigned long flags;
1413 spin_lock_irqsave(&netdev_fc_lock, flags);
1414 if (netdev_fc_mask != ~0UL) {
1415 bit = ffz(netdev_fc_mask);
1416 netdev_fc_slots[bit].stimul = stimul;
1417 netdev_fc_slots[bit].dev = dev;
1418 set_bit(bit, &netdev_fc_mask);
1419 clear_bit(bit, &netdev_fc_xoff);
1421 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1425 void netdev_unregister_fc(int bit)
1427 unsigned long flags;
1429 spin_lock_irqsave(&netdev_fc_lock, flags);
1431 netdev_fc_slots[bit].stimul = NULL;
1432 netdev_fc_slots[bit].dev = NULL;
1433 clear_bit(bit, &netdev_fc_mask);
1434 clear_bit(bit, &netdev_fc_xoff);
1436 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1439 static void netdev_wakeup(void)
1443 spin_lock(&netdev_fc_lock);
1444 xoff = netdev_fc_xoff;
1449 netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1451 spin_unlock(&netdev_fc_lock);
1455 static void get_sample_stats(int cpu)
1461 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1462 int blog = sd->input_pkt_queue.qlen;
1463 int avg_blog = sd->avg_blog;
1465 avg_blog = (avg_blog >> 1) + (blog >> 1);
1467 if (avg_blog > mod_cong) {
1468 /* Above moderate congestion levels. */
1469 sd->cng_level = NET_RX_CN_HIGH;
1472 rq = rd % netdev_max_backlog;
1473 if (rq < avg_blog) /* unlucky bastard */
1474 sd->cng_level = NET_RX_DROP;
1476 } else if (avg_blog > lo_cong) {
1477 sd->cng_level = NET_RX_CN_MOD;
1480 rq = rd % netdev_max_backlog;
1481 if (rq < avg_blog) /* unlucky bastard */
1482 sd->cng_level = NET_RX_CN_HIGH;
1484 } else if (avg_blog > no_cong)
1485 sd->cng_level = NET_RX_CN_LOW;
1486 else /* no congestion */
1487 sd->cng_level = NET_RX_SUCCESS;
1489 sd->avg_blog = avg_blog;
1492 #ifdef OFFLINE_SAMPLE
1493 static void sample_queue(unsigned long dummy)
1495 /* 10 ms 0r 1ms -- i don't care -- JHS */
1497 int cpu = smp_processor_id();
1499 get_sample_stats(cpu);
1500 next_tick += jiffies;
1501 mod_timer(&samp_timer, next_tick);
1507 * netif_rx - post buffer to the network code
1508 * @skb: buffer to post
1510 * This function receives a packet from a device driver and queues it for
1511 * the upper (protocol) levels to process. It always succeeds. The buffer
1512 * may be dropped during processing for congestion control or by the
1516 * NET_RX_SUCCESS (no congestion)
1517 * NET_RX_CN_LOW (low congestion)
1518 * NET_RX_CN_MOD (moderate congestion)
1519 * NET_RX_CN_HIGH (high congestion)
1520 * NET_RX_DROP (packet was dropped)
1524 int netif_rx(struct sk_buff *skb)
1527 struct softnet_data *queue;
1528 unsigned long flags;
1530 #ifdef CONFIG_NETPOLL
1531 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1537 if (!skb->stamp.tv_sec)
1538 net_timestamp(&skb->stamp);
1541 * The code is rearranged so that the path is the most
1542 * short when CPU is congested, but is still operating.
1544 local_irq_save(flags);
1545 this_cpu = smp_processor_id();
1546 queue = &__get_cpu_var(softnet_data);
1548 __get_cpu_var(netdev_rx_stat).total++;
1549 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1550 if (queue->input_pkt_queue.qlen) {
1551 if (queue->throttle)
1556 __skb_queue_tail(&queue->input_pkt_queue, skb);
1557 #ifndef OFFLINE_SAMPLE
1558 get_sample_stats(this_cpu);
1560 local_irq_restore(flags);
1561 return queue->cng_level;
1564 if (queue->throttle) {
1565 queue->throttle = 0;
1566 #ifdef CONFIG_NET_HW_FLOWCONTROL
1567 if (atomic_dec_and_test(&netdev_dropping))
1572 netif_rx_schedule(&queue->backlog_dev);
1576 if (!queue->throttle) {
1577 queue->throttle = 1;
1578 __get_cpu_var(netdev_rx_stat).throttled++;
1579 #ifdef CONFIG_NET_HW_FLOWCONTROL
1580 atomic_inc(&netdev_dropping);
1585 __get_cpu_var(netdev_rx_stat).dropped++;
1586 local_irq_restore(flags);
1592 static __inline__ void skb_bond(struct sk_buff *skb)
1594 struct net_device *dev = skb->dev;
1597 skb->real_dev = skb->dev;
1598 skb->dev = dev->master;
1602 static void net_tx_action(struct softirq_action *h)
1604 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1606 if (sd->completion_queue) {
1607 struct sk_buff *clist;
1609 local_irq_disable();
1610 clist = sd->completion_queue;
1611 sd->completion_queue = NULL;
1615 struct sk_buff *skb = clist;
1616 clist = clist->next;
1618 BUG_TRAP(!atomic_read(&skb->users));
1623 if (sd->output_queue) {
1624 struct net_device *head;
1626 local_irq_disable();
1627 head = sd->output_queue;
1628 sd->output_queue = NULL;
1632 struct net_device *dev = head;
1633 head = head->next_sched;
1635 smp_mb__before_clear_bit();
1636 clear_bit(__LINK_STATE_SCHED, &dev->state);
1638 if (spin_trylock(&dev->queue_lock)) {
1640 spin_unlock(&dev->queue_lock);
1642 netif_schedule(dev);
1648 static __inline__ int deliver_skb(struct sk_buff *skb,
1649 struct packet_type *pt_prev)
1651 atomic_inc(&skb->users);
1652 return pt_prev->func(skb, skb->dev, pt_prev);
1655 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1656 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1658 static __inline__ int handle_bridge(struct sk_buff **pskb,
1659 struct packet_type **pt_prev, int *ret)
1661 struct net_bridge_port *port;
1663 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1664 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1668 *ret = deliver_skb(*pskb, *pt_prev);
1672 return br_handle_frame_hook(port, pskb);
1675 #define handle_bridge(skb, pt_prev, ret) (0)
1678 #ifdef CONFIG_NET_CLS_ACT
1679 /* TODO: Maybe we should just force sch_ingress to be compiled in
1680 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1681 * a compare and 2 stores extra right now if we dont have it on
1682 * but have CONFIG_NET_CLS_ACT
1683 * NOTE: This doesnt stop any functionality; if you dont have
1684 * the ingress scheduler, you just cant add policies on ingress.
1687 int ing_filter(struct sk_buff *skb)
1690 struct net_device *dev = skb->dev;
1691 int result = TC_ACT_OK;
1693 if (dev->qdisc_ingress) {
1694 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1695 if (MAX_RED_LOOP < ttl++) {
1696 printk("Redir loop detected Dropping packet (%s->%s)\n",
1697 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1701 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1703 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1704 if (NULL == skb->input_dev) {
1705 skb->input_dev = skb->dev;
1706 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1708 spin_lock(&dev->ingress_lock);
1709 if ((q = dev->qdisc_ingress) != NULL)
1710 result = q->enqueue(skb, q);
1711 spin_unlock(&dev->ingress_lock);
1719 int netif_receive_skb(struct sk_buff *skb)
1721 struct packet_type *ptype, *pt_prev;
1722 int ret = NET_RX_DROP;
1723 unsigned short type;
1725 #ifdef CONFIG_NETPOLL
1726 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1732 if (!skb->stamp.tv_sec)
1733 net_timestamp(&skb->stamp);
1737 __get_cpu_var(netdev_rx_stat).total++;
1739 skb->h.raw = skb->nh.raw = skb->data;
1740 skb->mac_len = skb->nh.raw - skb->mac.raw;
1746 #ifdef CONFIG_NET_CLS_ACT
1747 if (skb->tc_verd & TC_NCLS) {
1748 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1753 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1754 if (!ptype->dev || ptype->dev == skb->dev) {
1756 ret = deliver_skb(skb, pt_prev);
1761 #ifdef CONFIG_NET_CLS_ACT
1763 ret = deliver_skb(skb, pt_prev);
1764 pt_prev = NULL; /* noone else should process this after*/
1766 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1769 ret = ing_filter(skb);
1771 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1780 handle_diverter(skb);
1782 if (handle_bridge(&skb, &pt_prev, &ret))
1785 type = skb->protocol;
1786 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1787 if (ptype->type == type &&
1788 (!ptype->dev || ptype->dev == skb->dev)) {
1790 ret = deliver_skb(skb, pt_prev);
1796 ret = pt_prev->func(skb, skb->dev, pt_prev);
1799 /* Jamal, now you will not able to escape explaining
1800 * me how you were going to use this. :-)
1810 static int process_backlog(struct net_device *backlog_dev, int *budget)
1813 int quota = min(backlog_dev->quota, *budget);
1814 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1815 unsigned long start_time = jiffies;
1818 struct sk_buff *skb;
1819 struct net_device *dev;
1821 local_irq_disable();
1822 skb = __skb_dequeue(&queue->input_pkt_queue);
1829 netif_receive_skb(skb);
1835 if (work >= quota || jiffies - start_time > 1)
1838 #ifdef CONFIG_NET_HW_FLOWCONTROL
1839 if (queue->throttle &&
1840 queue->input_pkt_queue.qlen < no_cong_thresh ) {
1841 queue->throttle = 0;
1842 if (atomic_dec_and_test(&netdev_dropping)) {
1850 backlog_dev->quota -= work;
1855 backlog_dev->quota -= work;
1858 list_del(&backlog_dev->poll_list);
1859 smp_mb__before_clear_bit();
1860 netif_poll_enable(backlog_dev);
1862 if (queue->throttle) {
1863 queue->throttle = 0;
1864 #ifdef CONFIG_NET_HW_FLOWCONTROL
1865 if (atomic_dec_and_test(&netdev_dropping))
1873 static void net_rx_action(struct softirq_action *h)
1875 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1876 unsigned long start_time = jiffies;
1877 int budget = netdev_max_backlog;
1880 local_irq_disable();
1882 while (!list_empty(&queue->poll_list)) {
1883 struct net_device *dev;
1885 if (budget <= 0 || jiffies - start_time > 1)
1890 dev = list_entry(queue->poll_list.next,
1891 struct net_device, poll_list);
1893 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1894 local_irq_disable();
1895 list_del(&dev->poll_list);
1896 list_add_tail(&dev->poll_list, &queue->poll_list);
1898 dev->quota += dev->weight;
1900 dev->quota = dev->weight;
1903 local_irq_disable();
1911 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1912 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1916 static gifconf_func_t * gifconf_list [NPROTO];
1919 * register_gifconf - register a SIOCGIF handler
1920 * @family: Address family
1921 * @gifconf: Function handler
1923 * Register protocol dependent address dumping routines. The handler
1924 * that is passed must not be freed or reused until it has been replaced
1925 * by another handler.
1927 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1929 if (family >= NPROTO)
1931 gifconf_list[family] = gifconf;
1937 * Map an interface index to its name (SIOCGIFNAME)
1941 * We need this ioctl for efficient implementation of the
1942 * if_indextoname() function required by the IPv6 API. Without
1943 * it, we would have to search all the interfaces to find a
1947 static int dev_ifname(struct ifreq __user *arg)
1949 struct net_device *dev;
1953 * Fetch the caller's info block.
1956 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1959 read_lock(&dev_base_lock);
1960 dev = __dev_get_by_index(ifr.ifr_ifindex);
1962 read_unlock(&dev_base_lock);
1966 strcpy(ifr.ifr_name, dev->name);
1967 read_unlock(&dev_base_lock);
1969 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1975 * Perform a SIOCGIFCONF call. This structure will change
1976 * size eventually, and there is nothing I can do about it.
1977 * Thus we will need a 'compatibility mode'.
1980 static int dev_ifconf(char __user *arg)
1983 struct net_device *dev;
1990 * Fetch the caller's info block.
1993 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2000 * Loop over the interfaces, and write an info block for each.
2004 for (dev = dev_base; dev; dev = dev->next) {
2005 if (vx_flags(VXF_HIDE_NETIF, 0) &&
2006 !dev_in_nx_info(dev, current->nx_info))
2008 for (i = 0; i < NPROTO; i++) {
2009 if (gifconf_list[i]) {
2012 done = gifconf_list[i](dev, NULL, 0);
2014 done = gifconf_list[i](dev, pos + total,
2024 * All done. Write the updated control block back to the caller.
2026 ifc.ifc_len = total;
2029 * Both BSD and Solaris return 0 here, so we do too.
2031 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2034 #ifdef CONFIG_PROC_FS
2036 * This is invoked by the /proc filesystem handler to display a device
2039 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2041 struct net_device *dev;
2044 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2046 return i == pos ? dev : NULL;
2049 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2051 read_lock(&dev_base_lock);
2052 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2055 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2058 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2061 void dev_seq_stop(struct seq_file *seq, void *v)
2063 read_unlock(&dev_base_lock);
2066 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2068 struct nx_info *nxi = current->nx_info;
2070 if (vx_flags(VXF_HIDE_NETIF, 0) && !dev_in_nx_info(dev, nxi))
2072 if (dev->get_stats) {
2073 struct net_device_stats *stats = dev->get_stats(dev);
2075 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2076 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2077 dev->name, stats->rx_bytes, stats->rx_packets,
2079 stats->rx_dropped + stats->rx_missed_errors,
2080 stats->rx_fifo_errors,
2081 stats->rx_length_errors + stats->rx_over_errors +
2082 stats->rx_crc_errors + stats->rx_frame_errors,
2083 stats->rx_compressed, stats->multicast,
2084 stats->tx_bytes, stats->tx_packets,
2085 stats->tx_errors, stats->tx_dropped,
2086 stats->tx_fifo_errors, stats->collisions,
2087 stats->tx_carrier_errors +
2088 stats->tx_aborted_errors +
2089 stats->tx_window_errors +
2090 stats->tx_heartbeat_errors,
2091 stats->tx_compressed);
2093 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2097 * Called from the PROCfs module. This now uses the new arbitrary sized
2098 * /proc/net interface to create /proc/net/dev
2100 static int dev_seq_show(struct seq_file *seq, void *v)
2102 if (v == SEQ_START_TOKEN)
2103 seq_puts(seq, "Inter-| Receive "
2105 " face |bytes packets errs drop fifo frame "
2106 "compressed multicast|bytes packets errs "
2107 "drop fifo colls carrier compressed\n");
2109 dev_seq_printf_stats(seq, v);
2113 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2115 struct netif_rx_stats *rc = NULL;
2117 while (*pos < NR_CPUS)
2118 if (cpu_online(*pos)) {
2119 rc = &per_cpu(netdev_rx_stat, *pos);
2126 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2128 return softnet_get_online(pos);
2131 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2134 return softnet_get_online(pos);
2137 static void softnet_seq_stop(struct seq_file *seq, void *v)
2141 static int softnet_seq_show(struct seq_file *seq, void *v)
2143 struct netif_rx_stats *s = v;
2145 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2146 s->total, s->dropped, s->time_squeeze, s->throttled,
2147 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2148 s->fastroute_deferred_out,
2150 s->fastroute_latency_reduction
2158 static struct seq_operations dev_seq_ops = {
2159 .start = dev_seq_start,
2160 .next = dev_seq_next,
2161 .stop = dev_seq_stop,
2162 .show = dev_seq_show,
2165 static int dev_seq_open(struct inode *inode, struct file *file)
2167 return seq_open(file, &dev_seq_ops);
2170 static struct file_operations dev_seq_fops = {
2171 .owner = THIS_MODULE,
2172 .open = dev_seq_open,
2174 .llseek = seq_lseek,
2175 .release = seq_release,
2178 static struct seq_operations softnet_seq_ops = {
2179 .start = softnet_seq_start,
2180 .next = softnet_seq_next,
2181 .stop = softnet_seq_stop,
2182 .show = softnet_seq_show,
2185 static int softnet_seq_open(struct inode *inode, struct file *file)
2187 return seq_open(file, &softnet_seq_ops);
2190 static struct file_operations softnet_seq_fops = {
2191 .owner = THIS_MODULE,
2192 .open = softnet_seq_open,
2194 .llseek = seq_lseek,
2195 .release = seq_release,
2199 extern int wireless_proc_init(void);
2201 #define wireless_proc_init() 0
2204 static int __init dev_proc_init(void)
2208 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2210 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2212 if (wireless_proc_init())
2218 proc_net_remove("softnet_stat");
2220 proc_net_remove("dev");
2224 #define dev_proc_init() 0
2225 #endif /* CONFIG_PROC_FS */
2229 * netdev_set_master - set up master/slave pair
2230 * @slave: slave device
2231 * @master: new master device
2233 * Changes the master device of the slave. Pass %NULL to break the
2234 * bonding. The caller must hold the RTNL semaphore. On a failure
2235 * a negative errno code is returned. On success the reference counts
2236 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2237 * function returns zero.
2239 int netdev_set_master(struct net_device *slave, struct net_device *master)
2241 struct net_device *old = slave->master;
2251 slave->master = master;
2259 slave->flags |= IFF_SLAVE;
2261 slave->flags &= ~IFF_SLAVE;
2263 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2268 * dev_set_promiscuity - update promiscuity count on a device
2272 * Add or remove promsicuity from a device. While the count in the device
2273 * remains above zero the interface remains promiscuous. Once it hits zero
2274 * the device reverts back to normal filtering operation. A negative inc
2275 * value is used to drop promiscuity on the device.
2277 void dev_set_promiscuity(struct net_device *dev, int inc)
2279 unsigned short old_flags = dev->flags;
2281 dev->flags |= IFF_PROMISC;
2282 if ((dev->promiscuity += inc) == 0)
2283 dev->flags &= ~IFF_PROMISC;
2284 if (dev->flags ^ old_flags) {
2286 printk(KERN_INFO "device %s %s promiscuous mode\n",
2287 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2293 * dev_set_allmulti - update allmulti count on a device
2297 * Add or remove reception of all multicast frames to a device. While the
2298 * count in the device remains above zero the interface remains listening
2299 * to all interfaces. Once it hits zero the device reverts back to normal
2300 * filtering operation. A negative @inc value is used to drop the counter
2301 * when releasing a resource needing all multicasts.
2304 void dev_set_allmulti(struct net_device *dev, int inc)
2306 unsigned short old_flags = dev->flags;
2308 dev->flags |= IFF_ALLMULTI;
2309 if ((dev->allmulti += inc) == 0)
2310 dev->flags &= ~IFF_ALLMULTI;
2311 if (dev->flags ^ old_flags)
2315 unsigned dev_get_flags(const struct net_device *dev)
2319 flags = (dev->flags & ~(IFF_PROMISC |
2322 (dev->gflags & (IFF_PROMISC |
2325 if (netif_running(dev) && netif_carrier_ok(dev))
2326 flags |= IFF_RUNNING;
2331 int dev_change_flags(struct net_device *dev, unsigned flags)
2334 int old_flags = dev->flags;
2337 * Set the flags on our device.
2340 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2341 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2343 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2347 * Load in the correct multicast list now the flags have changed.
2353 * Have we downed the interface. We handle IFF_UP ourselves
2354 * according to user attempts to set it, rather than blindly
2359 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2360 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2366 if (dev->flags & IFF_UP &&
2367 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2369 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2371 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2372 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2373 dev->gflags ^= IFF_PROMISC;
2374 dev_set_promiscuity(dev, inc);
2377 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2378 is important. Some (broken) drivers set IFF_PROMISC, when
2379 IFF_ALLMULTI is requested not asking us and not reporting.
2381 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2382 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2383 dev->gflags ^= IFF_ALLMULTI;
2384 dev_set_allmulti(dev, inc);
2387 if (old_flags ^ dev->flags)
2388 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2393 int dev_set_mtu(struct net_device *dev, int new_mtu)
2397 if (new_mtu == dev->mtu)
2400 /* MTU must be positive. */
2404 if (!netif_device_present(dev))
2408 if (dev->change_mtu)
2409 err = dev->change_mtu(dev, new_mtu);
2412 if (!err && dev->flags & IFF_UP)
2413 notifier_call_chain(&netdev_chain,
2414 NETDEV_CHANGEMTU, dev);
2420 * Perform the SIOCxIFxxx calls.
2422 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2425 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2431 case SIOCGIFFLAGS: /* Get interface flags */
2432 ifr->ifr_flags = dev_get_flags(dev);
2435 case SIOCSIFFLAGS: /* Set interface flags */
2436 return dev_change_flags(dev, ifr->ifr_flags);
2438 case SIOCGIFMETRIC: /* Get the metric on the interface
2439 (currently unused) */
2440 ifr->ifr_metric = 0;
2443 case SIOCSIFMETRIC: /* Set the metric on the interface
2444 (currently unused) */
2447 case SIOCGIFMTU: /* Get the MTU of a device */
2448 ifr->ifr_mtu = dev->mtu;
2451 case SIOCSIFMTU: /* Set the MTU of a device */
2452 return dev_set_mtu(dev, ifr->ifr_mtu);
2455 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2456 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2457 ifr->ifr_hwaddr.sa_family = dev->type;
2461 if (!dev->set_mac_address)
2463 if (ifr->ifr_hwaddr.sa_family != dev->type)
2465 if (!netif_device_present(dev))
2467 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2469 notifier_call_chain(&netdev_chain,
2470 NETDEV_CHANGEADDR, dev);
2473 case SIOCSIFHWBROADCAST:
2474 if (ifr->ifr_hwaddr.sa_family != dev->type)
2476 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2477 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2478 notifier_call_chain(&netdev_chain,
2479 NETDEV_CHANGEADDR, dev);
2483 ifr->ifr_map.mem_start = dev->mem_start;
2484 ifr->ifr_map.mem_end = dev->mem_end;
2485 ifr->ifr_map.base_addr = dev->base_addr;
2486 ifr->ifr_map.irq = dev->irq;
2487 ifr->ifr_map.dma = dev->dma;
2488 ifr->ifr_map.port = dev->if_port;
2492 if (dev->set_config) {
2493 if (!netif_device_present(dev))
2495 return dev->set_config(dev, &ifr->ifr_map);
2500 if (!dev->set_multicast_list ||
2501 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2503 if (!netif_device_present(dev))
2505 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2509 if (!dev->set_multicast_list ||
2510 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2512 if (!netif_device_present(dev))
2514 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2518 ifr->ifr_ifindex = dev->ifindex;
2522 ifr->ifr_qlen = dev->tx_queue_len;
2526 if (ifr->ifr_qlen < 0)
2528 dev->tx_queue_len = ifr->ifr_qlen;
2532 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2533 return dev_change_name(dev, ifr->ifr_newname);
2536 * Unknown or private ioctl
2540 if ((cmd >= SIOCDEVPRIVATE &&
2541 cmd <= SIOCDEVPRIVATE + 15) ||
2542 cmd == SIOCBONDENSLAVE ||
2543 cmd == SIOCBONDRELEASE ||
2544 cmd == SIOCBONDSETHWADDR ||
2545 cmd == SIOCBONDSLAVEINFOQUERY ||
2546 cmd == SIOCBONDINFOQUERY ||
2547 cmd == SIOCBONDCHANGEACTIVE ||
2548 cmd == SIOCGMIIPHY ||
2549 cmd == SIOCGMIIREG ||
2550 cmd == SIOCSMIIREG ||
2551 cmd == SIOCBRADDIF ||
2552 cmd == SIOCBRDELIF ||
2553 cmd == SIOCWANDEV) {
2555 if (dev->do_ioctl) {
2556 if (netif_device_present(dev))
2557 err = dev->do_ioctl(dev, ifr,
2570 * This function handles all "interface"-type I/O control requests. The actual
2571 * 'doing' part of this is dev_ifsioc above.
2575 * dev_ioctl - network device ioctl
2576 * @cmd: command to issue
2577 * @arg: pointer to a struct ifreq in user space
2579 * Issue ioctl functions to devices. This is normally called by the
2580 * user space syscall interfaces but can sometimes be useful for
2581 * other purposes. The return value is the return from the syscall if
2582 * positive or a negative errno code on error.
2585 int dev_ioctl(unsigned int cmd, void __user *arg)
2591 /* One special case: SIOCGIFCONF takes ifconf argument
2592 and requires shared lock, because it sleeps writing
2596 if (cmd == SIOCGIFCONF) {
2598 ret = dev_ifconf((char __user *) arg);
2602 if (cmd == SIOCGIFNAME)
2603 return dev_ifname((struct ifreq __user *)arg);
2605 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2608 ifr.ifr_name[IFNAMSIZ-1] = 0;
2610 colon = strchr(ifr.ifr_name, ':');
2615 * See which interface the caller is talking about.
2620 * These ioctl calls:
2621 * - can be done by all.
2622 * - atomic and do not require locking.
2633 dev_load(ifr.ifr_name);
2634 read_lock(&dev_base_lock);
2635 ret = dev_ifsioc(&ifr, cmd);
2636 read_unlock(&dev_base_lock);
2640 if (copy_to_user(arg, &ifr,
2641 sizeof(struct ifreq)))
2647 dev_load(ifr.ifr_name);
2649 ret = dev_ethtool(&ifr);
2654 if (copy_to_user(arg, &ifr,
2655 sizeof(struct ifreq)))
2661 * These ioctl calls:
2662 * - require superuser power.
2663 * - require strict serialization.
2669 if (!capable(CAP_NET_ADMIN))
2671 dev_load(ifr.ifr_name);
2673 ret = dev_ifsioc(&ifr, cmd);
2678 if (copy_to_user(arg, &ifr,
2679 sizeof(struct ifreq)))
2685 * These ioctl calls:
2686 * - require superuser power.
2687 * - require strict serialization.
2688 * - do not return a value
2698 case SIOCSIFHWBROADCAST:
2701 case SIOCBONDENSLAVE:
2702 case SIOCBONDRELEASE:
2703 case SIOCBONDSETHWADDR:
2704 case SIOCBONDSLAVEINFOQUERY:
2705 case SIOCBONDINFOQUERY:
2706 case SIOCBONDCHANGEACTIVE:
2709 if (!capable(CAP_NET_ADMIN))
2711 dev_load(ifr.ifr_name);
2713 ret = dev_ifsioc(&ifr, cmd);
2718 /* Get the per device memory space. We can add this but
2719 * currently do not support it */
2721 /* Set the per device memory buffer space.
2722 * Not applicable in our case */
2727 * Unknown or private ioctl.
2730 if (cmd == SIOCWANDEV ||
2731 (cmd >= SIOCDEVPRIVATE &&
2732 cmd <= SIOCDEVPRIVATE + 15)) {
2733 dev_load(ifr.ifr_name);
2735 ret = dev_ifsioc(&ifr, cmd);
2737 if (!ret && copy_to_user(arg, &ifr,
2738 sizeof(struct ifreq)))
2743 /* Take care of Wireless Extensions */
2744 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2745 /* If command is `set a parameter', or
2746 * `get the encoding parameters', check if
2747 * the user has the right to do it */
2748 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2749 if (!capable(CAP_NET_ADMIN))
2752 dev_load(ifr.ifr_name);
2754 /* Follow me in net/core/wireless.c */
2755 ret = wireless_process_ioctl(&ifr, cmd);
2757 if (!ret && IW_IS_GET(cmd) &&
2758 copy_to_user(arg, &ifr,
2759 sizeof(struct ifreq)))
2763 #endif /* WIRELESS_EXT */
2770 * dev_new_index - allocate an ifindex
2772 * Returns a suitable unique value for a new device interface
2773 * number. The caller must hold the rtnl semaphore or the
2774 * dev_base_lock to be sure it remains unique.
2776 int dev_new_index(void)
2782 if (!__dev_get_by_index(ifindex))
2787 static int dev_boot_phase = 1;
2789 /* Delayed registration/unregisteration */
2790 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2791 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2793 static inline void net_set_todo(struct net_device *dev)
2795 spin_lock(&net_todo_list_lock);
2796 list_add_tail(&dev->todo_list, &net_todo_list);
2797 spin_unlock(&net_todo_list_lock);
2801 * register_netdevice - register a network device
2802 * @dev: device to register
2804 * Take a completed network device structure and add it to the kernel
2805 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2806 * chain. 0 is returned on success. A negative errno code is returned
2807 * on a failure to set up the device, or if the name is a duplicate.
2809 * Callers must hold the rtnl semaphore. See the comment at the
2810 * end of Space.c for details about the locking. You may want
2811 * register_netdev() instead of this.
2814 * The locking appears insufficient to guarantee two parallel registers
2815 * will not get the same name.
2818 int register_netdevice(struct net_device *dev)
2820 struct hlist_head *head;
2821 struct hlist_node *p;
2824 BUG_ON(dev_boot_phase);
2827 /* When net_device's are persistent, this will be fatal. */
2828 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2830 spin_lock_init(&dev->queue_lock);
2831 spin_lock_init(&dev->xmit_lock);
2832 dev->xmit_lock_owner = -1;
2833 #ifdef CONFIG_NET_CLS_ACT
2834 spin_lock_init(&dev->ingress_lock);
2837 ret = alloc_divert_blk(dev);
2843 /* Init, if this function is available */
2845 ret = dev->init(dev);
2853 if (!dev_valid_name(dev->name)) {
2858 dev->ifindex = dev_new_index();
2859 if (dev->iflink == -1)
2860 dev->iflink = dev->ifindex;
2862 /* Check for existence of name */
2863 head = dev_name_hash(dev->name);
2864 hlist_for_each(p, head) {
2865 struct net_device *d
2866 = hlist_entry(p, struct net_device, name_hlist);
2867 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2873 /* Fix illegal SG+CSUM combinations. */
2874 if ((dev->features & NETIF_F_SG) &&
2875 !(dev->features & (NETIF_F_IP_CSUM |
2877 NETIF_F_HW_CSUM))) {
2878 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2880 dev->features &= ~NETIF_F_SG;
2884 * nil rebuild_header routine,
2885 * that should be never called and used as just bug trap.
2888 if (!dev->rebuild_header)
2889 dev->rebuild_header = default_rebuild_header;
2892 * Default initial state at registry is that the
2893 * device is present.
2896 set_bit(__LINK_STATE_PRESENT, &dev->state);
2899 dev_init_scheduler(dev);
2900 write_lock_bh(&dev_base_lock);
2902 dev_tail = &dev->next;
2903 hlist_add_head(&dev->name_hlist, head);
2904 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2906 dev->reg_state = NETREG_REGISTERING;
2907 write_unlock_bh(&dev_base_lock);
2909 /* Notify protocols, that a new device appeared. */
2910 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2912 /* Finish registration after unlock */
2919 free_divert_blk(dev);
2924 * netdev_wait_allrefs - wait until all references are gone.
2926 * This is called when unregistering network devices.
2928 * Any protocol or device that holds a reference should register
2929 * for netdevice notification, and cleanup and put back the
2930 * reference if they receive an UNREGISTER event.
2931 * We can get stuck here if buggy protocols don't correctly
2934 static void netdev_wait_allrefs(struct net_device *dev)
2936 unsigned long rebroadcast_time, warning_time;
2938 rebroadcast_time = warning_time = jiffies;
2939 while (atomic_read(&dev->refcnt) != 0) {
2940 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2943 /* Rebroadcast unregister notification */
2944 notifier_call_chain(&netdev_chain,
2945 NETDEV_UNREGISTER, dev);
2947 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2949 /* We must not have linkwatch events
2950 * pending on unregister. If this
2951 * happens, we simply run the queue
2952 * unscheduled, resulting in a noop
2955 linkwatch_run_queue();
2960 rebroadcast_time = jiffies;
2963 current->state = TASK_INTERRUPTIBLE;
2964 schedule_timeout(HZ / 4);
2966 if (time_after(jiffies, warning_time + 10 * HZ)) {
2967 printk(KERN_EMERG "unregister_netdevice: "
2968 "waiting for %s to become free. Usage "
2970 dev->name, atomic_read(&dev->refcnt));
2971 warning_time = jiffies;
2980 * register_netdevice(x1);
2981 * register_netdevice(x2);
2983 * unregister_netdevice(y1);
2984 * unregister_netdevice(y2);
2990 * We are invoked by rtnl_unlock() after it drops the semaphore.
2991 * This allows us to deal with problems:
2992 * 1) We can create/delete sysfs objects which invoke hotplug
2993 * without deadlocking with linkwatch via keventd.
2994 * 2) Since we run with the RTNL semaphore not held, we can sleep
2995 * safely in order to wait for the netdev refcnt to drop to zero.
2997 static DECLARE_MUTEX(net_todo_run_mutex);
2998 void netdev_run_todo(void)
3000 struct list_head list = LIST_HEAD_INIT(list);
3004 /* Need to guard against multiple cpu's getting out of order. */
3005 down(&net_todo_run_mutex);
3007 /* Not safe to do outside the semaphore. We must not return
3008 * until all unregister events invoked by the local processor
3009 * have been completed (either by this todo run, or one on
3012 if (list_empty(&net_todo_list))
3015 /* Snapshot list, allow later requests */
3016 spin_lock(&net_todo_list_lock);
3017 list_splice_init(&net_todo_list, &list);
3018 spin_unlock(&net_todo_list_lock);
3020 while (!list_empty(&list)) {
3021 struct net_device *dev
3022 = list_entry(list.next, struct net_device, todo_list);
3023 list_del(&dev->todo_list);
3025 switch(dev->reg_state) {
3026 case NETREG_REGISTERING:
3027 err = netdev_register_sysfs(dev);
3029 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3031 dev->reg_state = NETREG_REGISTERED;
3034 case NETREG_UNREGISTERING:
3035 netdev_unregister_sysfs(dev);
3036 dev->reg_state = NETREG_UNREGISTERED;
3038 netdev_wait_allrefs(dev);
3041 BUG_ON(atomic_read(&dev->refcnt));
3042 BUG_TRAP(!dev->ip_ptr);
3043 BUG_TRAP(!dev->ip6_ptr);
3044 BUG_TRAP(!dev->dn_ptr);
3047 /* It must be the very last action,
3048 * after this 'dev' may point to freed up memory.
3050 if (dev->destructor)
3051 dev->destructor(dev);
3055 printk(KERN_ERR "network todo '%s' but state %d\n",
3056 dev->name, dev->reg_state);
3062 up(&net_todo_run_mutex);
3066 * free_netdev - free network device
3069 * This function does the last stage of destroying an allocated device
3070 * interface. The reference to the device object is released.
3071 * If this is the last reference then it will be freed.
3073 void free_netdev(struct net_device *dev)
3076 /* Compatiablity with error handling in drivers */
3077 if (dev->reg_state == NETREG_UNINITIALIZED) {
3078 kfree((char *)dev - dev->padded);
3082 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3083 dev->reg_state = NETREG_RELEASED;
3085 /* will free via class release */
3086 class_device_put(&dev->class_dev);
3088 kfree((char *)dev - dev->padded);
3092 /* Synchronize with packet receive processing. */
3093 void synchronize_net(void)
3096 synchronize_kernel();
3100 * unregister_netdevice - remove device from the kernel
3103 * This function shuts down a device interface and removes it
3104 * from the kernel tables. On success 0 is returned, on a failure
3105 * a negative errno code is returned.
3107 * Callers must hold the rtnl semaphore. See the comment at the
3108 * end of Space.c for details about the locking. You may want
3109 * unregister_netdev() instead of this.
3112 int unregister_netdevice(struct net_device *dev)
3114 struct net_device *d, **dp;
3116 BUG_ON(dev_boot_phase);
3119 /* Some devices call without registering for initialization unwind. */
3120 if (dev->reg_state == NETREG_UNINITIALIZED) {
3121 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3122 "was registered\n", dev->name, dev);
3126 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3128 /* If device is running, close it first. */
3129 if (dev->flags & IFF_UP)
3132 /* And unlink it from device chain. */
3133 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3135 write_lock_bh(&dev_base_lock);
3136 hlist_del(&dev->name_hlist);
3137 hlist_del(&dev->index_hlist);
3138 if (dev_tail == &dev->next)
3141 write_unlock_bh(&dev_base_lock);
3146 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3151 dev->reg_state = NETREG_UNREGISTERING;
3155 /* Shutdown queueing discipline. */
3159 /* Notify protocols, that we are about to destroy
3160 this device. They should clean all the things.
3162 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3165 * Flush the multicast chain
3167 dev_mc_discard(dev);
3172 /* Notifier chain MUST detach us from master device. */
3173 BUG_TRAP(!dev->master);
3175 free_divert_blk(dev);
3177 /* Finish processing unregister after unlock */
3186 #ifdef CONFIG_HOTPLUG_CPU
3187 static int dev_cpu_callback(struct notifier_block *nfb,
3188 unsigned long action,
3191 struct sk_buff **list_skb;
3192 struct net_device **list_net;
3193 struct sk_buff *skb;
3194 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3195 struct softnet_data *sd, *oldsd;
3197 if (action != CPU_DEAD)
3200 local_irq_disable();
3201 cpu = smp_processor_id();
3202 sd = &per_cpu(softnet_data, cpu);
3203 oldsd = &per_cpu(softnet_data, oldcpu);
3205 /* Find end of our completion_queue. */
3206 list_skb = &sd->completion_queue;
3208 list_skb = &(*list_skb)->next;
3209 /* Append completion queue from offline CPU. */
3210 *list_skb = oldsd->completion_queue;
3211 oldsd->completion_queue = NULL;
3213 /* Find end of our output_queue. */
3214 list_net = &sd->output_queue;
3216 list_net = &(*list_net)->next_sched;
3217 /* Append output queue from offline CPU. */
3218 *list_net = oldsd->output_queue;
3219 oldsd->output_queue = NULL;
3221 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3224 /* Process offline CPU's input_pkt_queue */
3225 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3230 #endif /* CONFIG_HOTPLUG_CPU */
3234 * Initialize the DEV module. At boot time this walks the device list and
3235 * unhooks any devices that fail to initialise (normally hardware not
3236 * present) and leaves us with a valid list of present and active devices.
3241 * This is called single threaded during boot, so no need
3242 * to take the rtnl semaphore.
3244 static int __init net_dev_init(void)
3246 int i, rc = -ENOMEM;
3248 BUG_ON(!dev_boot_phase);
3252 if (dev_proc_init())
3255 if (netdev_sysfs_init())
3258 INIT_LIST_HEAD(&ptype_all);
3259 for (i = 0; i < 16; i++)
3260 INIT_LIST_HEAD(&ptype_base[i]);
3262 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3263 INIT_HLIST_HEAD(&dev_name_head[i]);
3265 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3266 INIT_HLIST_HEAD(&dev_index_head[i]);
3269 * Initialise the packet receive queues.
3272 for (i = 0; i < NR_CPUS; i++) {
3273 struct softnet_data *queue;
3275 queue = &per_cpu(softnet_data, i);
3276 skb_queue_head_init(&queue->input_pkt_queue);
3277 queue->throttle = 0;
3278 queue->cng_level = 0;
3279 queue->avg_blog = 10; /* arbitrary non-zero */
3280 queue->completion_queue = NULL;
3281 INIT_LIST_HEAD(&queue->poll_list);
3282 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3283 queue->backlog_dev.weight = weight_p;
3284 queue->backlog_dev.poll = process_backlog;
3285 atomic_set(&queue->backlog_dev.refcnt, 1);
3288 #ifdef OFFLINE_SAMPLE
3289 samp_timer.expires = jiffies + (10 * HZ);
3290 add_timer(&samp_timer);
3295 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3296 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3298 hotcpu_notifier(dev_cpu_callback, 0);
3306 subsys_initcall(net_dev_init);
3308 EXPORT_SYMBOL(__dev_get);
3309 EXPORT_SYMBOL(__dev_get_by_flags);
3310 EXPORT_SYMBOL(__dev_get_by_index);
3311 EXPORT_SYMBOL(__dev_get_by_name);
3312 EXPORT_SYMBOL(__dev_remove_pack);
3313 EXPORT_SYMBOL(__skb_linearize);
3314 EXPORT_SYMBOL(call_netdevice_notifiers);
3315 EXPORT_SYMBOL(dev_add_pack);
3316 EXPORT_SYMBOL(dev_alloc_name);
3317 EXPORT_SYMBOL(dev_close);
3318 EXPORT_SYMBOL(dev_get_by_flags);
3319 EXPORT_SYMBOL(dev_get_by_index);
3320 EXPORT_SYMBOL(dev_get_by_name);
3321 EXPORT_SYMBOL(dev_getbyhwaddr);
3322 EXPORT_SYMBOL(dev_ioctl);
3323 EXPORT_SYMBOL(dev_new_index);
3324 EXPORT_SYMBOL(dev_open);
3325 EXPORT_SYMBOL(dev_queue_xmit);
3326 EXPORT_SYMBOL(dev_queue_xmit_nit);
3327 EXPORT_SYMBOL(dev_remove_pack);
3328 EXPORT_SYMBOL(dev_set_allmulti);
3329 EXPORT_SYMBOL(dev_set_promiscuity);
3330 EXPORT_SYMBOL(dev_change_flags);
3331 EXPORT_SYMBOL(dev_change_name);
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);
3347 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3348 EXPORT_SYMBOL(br_handle_frame_hook);
3352 EXPORT_SYMBOL(dev_load);
3354 #ifdef CONFIG_NET_HW_FLOWCONTROL
3355 EXPORT_SYMBOL(netdev_dropping);
3356 EXPORT_SYMBOL(netdev_fc_xoff);
3357 EXPORT_SYMBOL(netdev_register_fc);
3358 EXPORT_SYMBOL(netdev_unregister_fc);
3361 #ifdef CONFIG_NET_CLS_ACT
3362 EXPORT_SYMBOL(ing_filter);
3366 EXPORT_PER_CPU_SYMBOL(softnet_data);