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>
117 /* This define, if set, will randomly drop a packet when congestion
118 * is more than moderate. It helps fairness in the multi-interface
119 * case when one of them is a hog, but it kills performance for the
120 * single interface case so it is off now by default.
124 /* Setting this will sample the queue lengths and thus congestion
125 * via a timer instead of as each packet is received.
127 #undef OFFLINE_SAMPLE
130 * The list of packet types we will receive (as opposed to discard)
131 * and the routines to invoke.
133 * Why 16. Because with 16 the only overlap we get on a hash of the
134 * low nibble of the protocol value is RARP/SNAP/X.25.
136 * NOTE: That is no longer true with the addition of VLAN tags. Not
137 * sure which should go first, but I bet it won't make much
138 * difference if we are running VLANs. The good news is that
139 * this protocol won't be in the list unless compiled in, so
140 * the average user (w/out VLANs) will not be adversly affected.
157 static spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
158 static struct list_head ptype_base[16]; /* 16 way hashed list */
159 static struct list_head ptype_all; /* Taps */
161 #ifdef OFFLINE_SAMPLE
162 static void sample_queue(unsigned long dummy);
163 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
167 * The @dev_base list is protected by @dev_base_lock and the rtln
170 * Pure readers hold dev_base_lock for reading.
172 * Writers must hold the rtnl semaphore while they loop through the
173 * dev_base list, and hold dev_base_lock for writing when they do the
174 * actual updates. This allows pure readers to access the list even
175 * while a writer is preparing to update it.
177 * To put it another way, dev_base_lock is held for writing only to
178 * protect against pure readers; the rtnl semaphore provides the
179 * protection against other writers.
181 * See, for example usages, register_netdevice() and
182 * unregister_netdevice(), which must be called with the rtnl
185 struct net_device *dev_base;
186 struct net_device **dev_tail = &dev_base;
187 rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
189 EXPORT_SYMBOL(dev_base);
190 EXPORT_SYMBOL(dev_base_lock);
192 #define NETDEV_HASHBITS 8
193 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
194 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
196 static inline struct hlist_head *dev_name_hash(const char *name)
198 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
199 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
202 static inline struct hlist_head *dev_index_hash(int ifindex)
204 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
211 static struct notifier_block *netdev_chain;
214 * Device drivers call our routines to queue packets here. We empty the
215 * queue in the local softnet handler.
217 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
219 #ifdef CONFIG_NET_FASTROUTE
220 int netdev_fastroute;
221 int netdev_fastroute_obstacles;
225 extern int netdev_sysfs_init(void);
226 extern int netdev_register_sysfs(struct net_device *);
227 extern void netdev_unregister_sysfs(struct net_device *);
229 #define netdev_sysfs_init() (0)
230 #define netdev_register_sysfs(dev) (0)
231 #define netdev_unregister_sysfs(dev) do { } while(0)
234 /* netdump function */
235 void (*netdump_func) (struct pt_regs *regs) = NULL;
237 /*******************************************************************************
239 Protocol management and registration routines
241 *******************************************************************************/
250 * Add a protocol ID to the list. Now that the input handler is
251 * smarter we can dispense with all the messy stuff that used to be
254 * BEWARE!!! Protocol handlers, mangling input packets,
255 * MUST BE last in hash buckets and checking protocol handlers
256 * MUST start from promiscuous ptype_all chain in net_bh.
257 * It is true now, do not change it.
258 * Explanation follows: if protocol handler, mangling packet, will
259 * be the first on list, it is not able to sense, that packet
260 * is cloned and should be copied-on-write, so that it will
261 * change it and subsequent readers will get broken packet.
266 * dev_add_pack - add packet handler
267 * @pt: packet type declaration
269 * Add a protocol handler to the networking stack. The passed &packet_type
270 * is linked into kernel lists and may not be freed until it has been
271 * removed from the kernel lists.
273 * This call does not sleep therefore it can not
274 * guarantee all CPU's that are in middle of receiving packets
275 * will see the new packet type (until the next received packet).
278 void dev_add_pack(struct packet_type *pt)
282 spin_lock_bh(&ptype_lock);
283 #ifdef CONFIG_NET_FASTROUTE
284 if (pt->af_packet_priv) {
285 netdev_fastroute_obstacles++;
286 dev_clear_fastroute(pt->dev);
289 if (pt->type == htons(ETH_P_ALL)) {
291 list_add_rcu(&pt->list, &ptype_all);
293 hash = ntohs(pt->type) & 15;
294 list_add_rcu(&pt->list, &ptype_base[hash]);
296 spin_unlock_bh(&ptype_lock);
299 extern void linkwatch_run_queue(void);
304 * __dev_remove_pack - remove packet handler
305 * @pt: packet type declaration
307 * Remove a protocol handler that was previously added to the kernel
308 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
309 * from the kernel lists and can be freed or reused once this function
312 * The packet type might still be in use by receivers
313 * and must not be freed until after all the CPU's have gone
314 * through a quiescent state.
316 void __dev_remove_pack(struct packet_type *pt)
318 struct list_head *head;
319 struct packet_type *pt1;
321 spin_lock_bh(&ptype_lock);
323 if (pt->type == htons(ETH_P_ALL)) {
327 head = &ptype_base[ntohs(pt->type) & 15];
329 list_for_each_entry(pt1, head, list) {
331 #ifdef CONFIG_NET_FASTROUTE
332 if (pt->af_packet_priv)
333 netdev_fastroute_obstacles--;
335 list_del_rcu(&pt->list);
340 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
342 spin_unlock_bh(&ptype_lock);
345 * dev_remove_pack - remove packet handler
346 * @pt: packet type declaration
348 * Remove a protocol handler that was previously added to the kernel
349 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
350 * from the kernel lists and can be freed or reused once this function
353 * This call sleeps to guarantee that no CPU is looking at the packet
356 void dev_remove_pack(struct packet_type *pt)
358 __dev_remove_pack(pt);
363 /******************************************************************************
365 Device Boot-time Settings Routines
367 *******************************************************************************/
369 /* Boot time configuration table */
370 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
373 * netdev_boot_setup_add - add new setup entry
374 * @name: name of the device
375 * @map: configured settings for the device
377 * Adds new setup entry to the dev_boot_setup list. The function
378 * returns 0 on error and 1 on success. This is a generic routine to
381 int netdev_boot_setup_add(char *name, struct ifmap *map)
383 struct netdev_boot_setup *s;
387 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
388 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
389 memset(s[i].name, 0, sizeof(s[i].name));
390 strcpy(s[i].name, name);
391 memcpy(&s[i].map, map, sizeof(s[i].map));
396 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
400 * netdev_boot_setup_check - check boot time settings
401 * @dev: the netdevice
403 * Check boot time settings for the device.
404 * The found settings are set for the device to be used
405 * later in the device probing.
406 * Returns 0 if no settings found, 1 if they are.
408 int netdev_boot_setup_check(struct net_device *dev)
410 struct netdev_boot_setup *s = dev_boot_setup;
413 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
414 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
415 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
416 dev->irq = s[i].map.irq;
417 dev->base_addr = s[i].map.base_addr;
418 dev->mem_start = s[i].map.mem_start;
419 dev->mem_end = s[i].map.mem_end;
428 * netdev_boot_base - get address from boot time settings
429 * @prefix: prefix for network device
430 * @unit: id for network device
432 * Check boot time settings for the base address of device.
433 * The found settings are set for the device to be used
434 * later in the device probing.
435 * Returns 0 if no settings found.
437 unsigned long netdev_boot_base(const char *prefix, int unit)
439 const struct netdev_boot_setup *s = dev_boot_setup;
443 sprintf(name, "%s%d", prefix, unit);
446 * If device already registered then return base of 1
447 * to indicate not to probe for this interface
449 if (__dev_get_by_name(name))
452 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
453 if (!strcmp(name, s[i].name))
454 return s[i].map.base_addr;
459 * Saves at boot time configured settings for any netdevice.
461 int __init netdev_boot_setup(char *str)
466 str = get_options(str, ARRAY_SIZE(ints), ints);
471 memset(&map, 0, sizeof(map));
475 map.base_addr = ints[2];
477 map.mem_start = ints[3];
479 map.mem_end = ints[4];
481 /* Add new entry to the list */
482 return netdev_boot_setup_add(str, &map);
485 __setup("netdev=", netdev_boot_setup);
487 /*******************************************************************************
489 Device Interface Subroutines
491 *******************************************************************************/
494 * __dev_get_by_name - find a device by its name
495 * @name: name to find
497 * Find an interface by name. Must be called under RTNL semaphore
498 * or @dev_base_lock. If the name is found a pointer to the device
499 * is returned. If the name is not found then %NULL is returned. The
500 * reference counters are not incremented so the caller must be
501 * careful with locks.
504 struct net_device *__dev_get_by_name(const char *name)
506 struct hlist_node *p;
508 hlist_for_each(p, dev_name_hash(name)) {
509 struct net_device *dev
510 = hlist_entry(p, struct net_device, name_hlist);
511 if (!strncmp(dev->name, name, IFNAMSIZ))
518 * dev_get_by_name - find a device by its name
519 * @name: name to find
521 * Find an interface by name. This can be called from any
522 * context and does its own locking. The returned handle has
523 * the usage count incremented and the caller must use dev_put() to
524 * release it when it is no longer needed. %NULL is returned if no
525 * matching device is found.
528 struct net_device *dev_get_by_name(const char *name)
530 struct net_device *dev;
532 read_lock(&dev_base_lock);
533 dev = __dev_get_by_name(name);
536 read_unlock(&dev_base_lock);
541 Return value is changed to int to prevent illegal usage in future.
542 It is still legal to use to check for device existence.
544 User should understand, that the result returned by this function
545 is meaningless, if it was not issued under rtnl semaphore.
549 * dev_get - test if a device exists
550 * @name: name to test for
552 * Test if a name exists. Returns true if the name is found. In order
553 * to be sure the name is not allocated or removed during the test the
554 * caller must hold the rtnl semaphore.
556 * This function exists only for back compatibility with older
559 int __dev_get(const char *name)
561 struct net_device *dev;
563 read_lock(&dev_base_lock);
564 dev = __dev_get_by_name(name);
565 read_unlock(&dev_base_lock);
570 * __dev_get_by_index - find a device by its ifindex
571 * @ifindex: index of device
573 * Search for an interface by index. Returns %NULL if the device
574 * is not found or a pointer to the device. The device has not
575 * had its reference counter increased so the caller must be careful
576 * about locking. The caller must hold either the RTNL semaphore
580 struct net_device *__dev_get_by_index(int ifindex)
582 struct hlist_node *p;
584 hlist_for_each(p, dev_index_hash(ifindex)) {
585 struct net_device *dev
586 = hlist_entry(p, struct net_device, index_hlist);
587 if (dev->ifindex == ifindex)
595 * dev_get_by_index - find a device by its ifindex
596 * @ifindex: index of device
598 * Search for an interface by index. Returns NULL if the device
599 * is not found or a pointer to the device. The device returned has
600 * had a reference added and the pointer is safe until the user calls
601 * dev_put to indicate they have finished with it.
604 struct net_device *dev_get_by_index(int ifindex)
606 struct net_device *dev;
608 read_lock(&dev_base_lock);
609 dev = __dev_get_by_index(ifindex);
612 read_unlock(&dev_base_lock);
617 * dev_getbyhwaddr - find a device by its hardware address
618 * @type: media type of device
619 * @ha: hardware address
621 * Search for an interface by MAC address. Returns NULL if the device
622 * is not found or a pointer to the device. The caller must hold the
623 * rtnl semaphore. The returned device has not had its ref count increased
624 * and the caller must therefore be careful about locking
627 * If the API was consistent this would be __dev_get_by_hwaddr
630 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
632 struct net_device *dev;
636 for (dev = dev_base; dev; dev = dev->next)
637 if (dev->type == type &&
638 !memcmp(dev->dev_addr, ha, dev->addr_len))
643 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
645 struct net_device *dev;
647 for (dev = dev_base; dev; dev = dev->next)
648 if (dev->type == type)
653 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
655 struct net_device *dev_getfirstbyhwtype(unsigned short type)
657 struct net_device *dev;
660 dev = __dev_getfirstbyhwtype(type);
667 EXPORT_SYMBOL(dev_getfirstbyhwtype);
670 * dev_get_by_flags - find any device with given flags
671 * @if_flags: IFF_* values
672 * @mask: bitmask of bits in if_flags to check
674 * Search for any interface with the given flags. Returns NULL if a device
675 * is not found or a pointer to the device. The device returned has
676 * had a reference added and the pointer is safe until the user calls
677 * dev_put to indicate they have finished with it.
680 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
682 struct net_device *dev;
684 read_lock(&dev_base_lock);
685 dev = __dev_get_by_flags(if_flags, mask);
688 read_unlock(&dev_base_lock);
693 * __dev_get_by_flags - find any device with given flags
694 * @if_flags: IFF_* values
695 * @mask: bitmask of bits in if_flags to check
697 * Search for any interface with the given flags. Returns NULL if a device
698 * is not found or a pointer to the device. The caller must hold either
699 * the RTNL semaphore or @dev_base_lock.
702 struct net_device *__dev_get_by_flags(unsigned short if_flags, unsigned short mask)
704 struct net_device *dev;
706 for (dev = dev_base; dev != NULL; dev = dev->next) {
707 if (((dev->flags ^ if_flags) & mask) == 0)
714 * dev_valid_name - check if name is okay for network device
717 * Network device names need to be valid file names to
718 * to allow sysfs to work
720 int dev_valid_name(const char *name)
722 return !(*name == '\0'
723 || !strcmp(name, ".")
724 || !strcmp(name, "..")
725 || strchr(name, '/'));
729 * dev_alloc_name - allocate a name for a device
731 * @name: name format string
733 * Passed a format string - eg "lt%d" it will try and find a suitable
734 * id. Not efficient for many devices, not called a lot. The caller
735 * must hold the dev_base or rtnl lock while allocating the name and
736 * adding the device in order to avoid duplicates. Returns the number
737 * of the unit assigned or a negative errno code.
740 int dev_alloc_name(struct net_device *dev, const char *name)
745 const int max_netdevices = 8*PAGE_SIZE;
747 struct net_device *d;
749 p = strnchr(name, IFNAMSIZ-1, '%');
752 * Verify the string as this thing may have come from
753 * the user. There must be either one "%d" and no other "%"
756 if (p[1] != 'd' || strchr(p + 2, '%'))
759 /* Use one page as a bit array of possible slots */
760 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
764 for (d = dev_base; d; d = d->next) {
765 if (!sscanf(d->name, name, &i))
767 if (i < 0 || i >= max_netdevices)
770 /* avoid cases where sscanf is not exact inverse of printf */
771 snprintf(buf, sizeof(buf), name, i);
772 if (!strncmp(buf, d->name, IFNAMSIZ))
776 i = find_first_zero_bit(inuse, max_netdevices);
777 free_page((unsigned long) inuse);
780 snprintf(buf, sizeof(buf), name, i);
781 if (!__dev_get_by_name(buf)) {
782 strlcpy(dev->name, buf, IFNAMSIZ);
786 /* It is possible to run out of possible slots
787 * when the name is long and there isn't enough space left
788 * for the digits, or if all bits are used.
795 * dev_change_name - change name of a device
797 * @newname: name (or format string) must be at least IFNAMSIZ
799 * Change name of a device, can pass format strings "eth%d".
802 int dev_change_name(struct net_device *dev, char *newname)
808 if (dev->flags & IFF_UP)
811 if (!dev_valid_name(newname))
814 if (strchr(newname, '%')) {
815 err = dev_alloc_name(dev, newname);
818 strcpy(newname, dev->name);
820 else if (__dev_get_by_name(newname))
823 strlcpy(dev->name, newname, IFNAMSIZ);
825 err = class_device_rename(&dev->class_dev, dev->name);
827 hlist_del(&dev->name_hlist);
828 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
829 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
836 * netdev_state_change - device changes state
837 * @dev: device to cause notification
839 * Called to indicate a device has changed state. This function calls
840 * the notifier chains for netdev_chain and sends a NEWLINK message
841 * to the routing socket.
843 void netdev_state_change(struct net_device *dev)
845 if (dev->flags & IFF_UP) {
846 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
847 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
852 * dev_load - load a network module
853 * @name: name of interface
855 * If a network interface is not present and the process has suitable
856 * privileges this function loads the module. If module loading is not
857 * available in this kernel then it becomes a nop.
860 void dev_load(const char *name)
862 struct net_device *dev;
864 read_lock(&dev_base_lock);
865 dev = __dev_get_by_name(name);
866 read_unlock(&dev_base_lock);
868 if (!dev && capable(CAP_SYS_MODULE))
869 request_module("%s", name);
872 static int default_rebuild_header(struct sk_buff *skb)
874 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
875 skb->dev ? skb->dev->name : "NULL!!!");
882 * Some old buggy device drivers change get_stats after registering
883 * the device. Try and trap them here.
884 * This can be elimnated when all devices are known fixed.
886 static inline int get_stats_changed(struct net_device *dev)
888 int changed = dev->last_stats != dev->get_stats;
889 dev->last_stats = dev->get_stats;
894 * dev_open - prepare an interface for use.
895 * @dev: device to open
897 * Takes a device from down to up state. The device's private open
898 * function is invoked and then the multicast lists are loaded. Finally
899 * the device is moved into the up state and a %NETDEV_UP message is
900 * sent to the netdev notifier chain.
902 * Calling this function on an active interface is a nop. On a failure
903 * a negative errno code is returned.
905 int dev_open(struct net_device *dev)
913 if (dev->flags & IFF_UP)
917 * Check for broken device drivers.
919 if (get_stats_changed(dev) && net_ratelimit()) {
920 printk(KERN_ERR "%s: driver changed get_stats after register\n",
925 * Is it even present?
927 if (!netif_device_present(dev))
931 * Call device private open method
933 set_bit(__LINK_STATE_START, &dev->state);
935 ret = dev->open(dev);
937 clear_bit(__LINK_STATE_START, &dev->state);
941 * Check for more broken device drivers.
943 if (get_stats_changed(dev) && net_ratelimit()) {
944 printk(KERN_ERR "%s: driver changed get_stats in open\n",
949 * If it went open OK then:
956 dev->flags |= IFF_UP;
959 * Initialize multicasting status
964 * Wakeup transmit queue engine
969 * ... and announce new interface.
971 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
976 #ifdef CONFIG_NET_FASTROUTE
978 static void dev_do_clear_fastroute(struct net_device *dev)
980 if (dev->accept_fastpath) {
983 for (i = 0; i <= NETDEV_FASTROUTE_HMASK; i++) {
984 struct dst_entry *dst;
986 write_lock_irq(&dev->fastpath_lock);
987 dst = dev->fastpath[i];
988 dev->fastpath[i] = NULL;
989 write_unlock_irq(&dev->fastpath_lock);
996 void dev_clear_fastroute(struct net_device *dev)
999 dev_do_clear_fastroute(dev);
1001 read_lock(&dev_base_lock);
1002 for (dev = dev_base; dev; dev = dev->next)
1003 dev_do_clear_fastroute(dev);
1004 read_unlock(&dev_base_lock);
1010 * dev_close - shutdown an interface.
1011 * @dev: device to shutdown
1013 * This function moves an active device into down state. A
1014 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1015 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1018 int dev_close(struct net_device *dev)
1020 if (!(dev->flags & IFF_UP))
1024 * Tell people we are going down, so that they can
1025 * prepare to death, when device is still operating.
1027 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
1029 dev_deactivate(dev);
1031 clear_bit(__LINK_STATE_START, &dev->state);
1033 /* Synchronize to scheduled poll. We cannot touch poll list,
1034 * it can be even on different cpu. So just clear netif_running(),
1035 * and wait when poll really will happen. Actually, the best place
1036 * for this is inside dev->stop() after device stopped its irq
1037 * engine, but this requires more changes in devices. */
1039 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1040 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
1042 current->state = TASK_INTERRUPTIBLE;
1043 schedule_timeout(1);
1047 * Call the device specific close. This cannot fail.
1048 * Only if device is UP
1050 * We allow it to be called even after a DETACH hot-plug
1057 * Device is now down.
1060 dev->flags &= ~IFF_UP;
1061 #ifdef CONFIG_NET_FASTROUTE
1062 dev_clear_fastroute(dev);
1066 * Tell people we are down
1068 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1075 * Device change register/unregister. These are not inline or static
1076 * as we export them to the world.
1080 * register_netdevice_notifier - register a network notifier block
1083 * Register a notifier to be called when network device events occur.
1084 * The notifier passed is linked into the kernel structures and must
1085 * not be reused until it has been unregistered. A negative errno code
1086 * is returned on a failure.
1088 * When registered all registration and up events are replayed
1089 * to the new notifier to allow device to have a race free
1090 * view of the network device list.
1093 int register_netdevice_notifier(struct notifier_block *nb)
1095 struct net_device *dev;
1099 err = notifier_chain_register(&netdev_chain, nb);
1101 for (dev = dev_base; dev; dev = dev->next) {
1102 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1104 if (dev->flags & IFF_UP)
1105 nb->notifier_call(nb, NETDEV_UP, dev);
1113 * unregister_netdevice_notifier - unregister a network notifier block
1116 * Unregister a notifier previously registered by
1117 * register_netdevice_notifier(). The notifier is unlinked into the
1118 * kernel structures and may then be reused. A negative errno code
1119 * is returned on a failure.
1122 int unregister_netdevice_notifier(struct notifier_block *nb)
1124 return notifier_chain_unregister(&netdev_chain, nb);
1128 * call_netdevice_notifiers - call all network notifier blocks
1129 * @val: value passed unmodified to notifier function
1130 * @v: pointer passed unmodified to notifier function
1132 * Call all network notifier blocks. Parameters and return value
1133 * are as for notifier_call_chain().
1136 int call_netdevice_notifiers(unsigned long val, void *v)
1138 return notifier_call_chain(&netdev_chain, val, v);
1142 * Support routine. Sends outgoing frames to any network
1143 * taps currently in use.
1146 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1148 struct packet_type *ptype;
1149 net_timestamp(&skb->stamp);
1152 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1153 /* Never send packets back to the socket
1154 * they originated from - MvS (miquels@drinkel.ow.org)
1156 if ((ptype->dev == dev || !ptype->dev) &&
1157 (ptype->af_packet_priv == NULL ||
1158 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1159 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1163 /* skb->nh should be correctly
1164 set by sender, so that the second statement is
1165 just protection against buggy protocols.
1167 skb2->mac.raw = skb2->data;
1169 if (skb2->nh.raw < skb2->data ||
1170 skb2->nh.raw > skb2->tail) {
1171 if (net_ratelimit())
1172 printk(KERN_CRIT "protocol %04x is "
1174 skb2->protocol, dev->name);
1175 skb2->nh.raw = skb2->data;
1178 skb2->h.raw = skb2->nh.raw;
1179 skb2->pkt_type = PACKET_OUTGOING;
1180 ptype->func(skb2, skb->dev, ptype);
1187 * Invalidate hardware checksum when packet is to be mangled, and
1188 * complete checksum manually on outgoing path.
1190 int skb_checksum_help(struct sk_buff **pskb, int inward)
1193 int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1196 (*pskb)->ip_summed = CHECKSUM_NONE;
1200 if (skb_shared(*pskb) || skb_cloned(*pskb)) {
1201 struct sk_buff *newskb = skb_copy(*pskb, GFP_ATOMIC);
1207 skb_set_owner_w(newskb, (*pskb)->sk);
1212 if (offset > (int)(*pskb)->len)
1214 csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1216 offset = (*pskb)->tail - (*pskb)->h.raw;
1219 if ((*pskb)->csum + 2 > offset)
1222 *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1223 (*pskb)->ip_summed = CHECKSUM_NONE;
1228 #ifdef CONFIG_HIGHMEM
1229 /* Actually, we should eliminate this check as soon as we know, that:
1230 * 1. IOMMU is present and allows to map all the memory.
1231 * 2. No high memory really exists on this machine.
1234 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1238 if (dev->features & NETIF_F_HIGHDMA)
1241 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1242 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1248 #define illegal_highdma(dev, skb) (0)
1251 extern void skb_release_data(struct sk_buff *);
1253 /* Keep head the same: replace data */
1254 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1259 struct skb_shared_info *ninfo;
1260 int headerlen = skb->data - skb->head;
1261 int expand = (skb->tail + skb->data_len) - skb->end;
1263 if (skb_shared(skb))
1269 size = skb->end - skb->head + expand;
1270 size = SKB_DATA_ALIGN(size);
1271 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1275 /* Copy entire thing */
1276 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1280 ninfo = (struct skb_shared_info*)(data + size);
1281 atomic_set(&ninfo->dataref, 1);
1282 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1283 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1284 ninfo->nr_frags = 0;
1285 ninfo->frag_list = NULL;
1287 /* Offset between the two in bytes */
1288 offset = data - skb->head;
1290 /* Free old data. */
1291 skb_release_data(skb);
1294 skb->end = data + size;
1296 /* Set up new pointers */
1297 skb->h.raw += offset;
1298 skb->nh.raw += offset;
1299 skb->mac.raw += offset;
1300 skb->tail += offset;
1301 skb->data += offset;
1303 /* We are no longer a clone, even if we were. */
1306 skb->tail += skb->data_len;
1311 #define HARD_TX_LOCK_BH(dev, cpu) { \
1312 if ((dev->features & NETIF_F_LLTX) == 0) { \
1313 spin_lock_bh(&dev->xmit_lock); \
1314 dev->xmit_lock_owner = cpu; \
1318 #define HARD_TX_UNLOCK_BH(dev) { \
1319 if ((dev->features & NETIF_F_LLTX) == 0) { \
1320 dev->xmit_lock_owner = -1; \
1321 spin_unlock_bh(&dev->xmit_lock); \
1326 * dev_queue_xmit - transmit a buffer
1327 * @skb: buffer to transmit
1329 * Queue a buffer for transmission to a network device. The caller must
1330 * have set the device and priority and built the buffer before calling
1331 * this function. The function can be called from an interrupt.
1333 * A negative errno code is returned on a failure. A success does not
1334 * guarantee the frame will be transmitted as it may be dropped due
1335 * to congestion or traffic shaping.
1338 int dev_queue_xmit(struct sk_buff *skb)
1340 struct net_device *dev = skb->dev;
1344 if (skb_shinfo(skb)->frag_list &&
1345 !(dev->features & NETIF_F_FRAGLIST) &&
1346 __skb_linearize(skb, GFP_ATOMIC))
1349 /* Fragmented skb is linearized if device does not support SG,
1350 * or if at least one of fragments is in highmem and device
1351 * does not support DMA from it.
1353 if (skb_shinfo(skb)->nr_frags &&
1354 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1355 __skb_linearize(skb, GFP_ATOMIC))
1358 /* If packet is not checksummed and device does not support
1359 * checksumming for this protocol, complete checksumming here.
1361 if (skb->ip_summed == CHECKSUM_HW &&
1362 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1363 (!(dev->features & NETIF_F_IP_CSUM) ||
1364 skb->protocol != htons(ETH_P_IP))))
1365 if (skb_checksum_help(&skb, 0))
1369 /* Updates of qdisc are serialized by queue_lock.
1370 * The struct Qdisc which is pointed to by qdisc is now a
1371 * rcu structure - it may be accessed without acquiring
1372 * a lock (but the structure may be stale.) The freeing of the
1373 * qdisc will be deferred until it's known that there are no
1374 * more references to it.
1376 * If the qdisc has an enqueue function, we still need to
1377 * hold the queue_lock before calling it, since queue_lock
1378 * also serializes access to the device queue.
1382 smp_read_barrier_depends();
1383 #ifdef CONFIG_NET_CLS_ACT
1384 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1387 /* Grab device queue */
1388 spin_lock_bh(&dev->queue_lock);
1390 rc = q->enqueue(skb, q);
1394 spin_unlock_bh(&dev->queue_lock);
1396 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1401 /* The device has no queue. Common case for software devices:
1402 loopback, all the sorts of tunnels...
1404 Really, it is unlikely that xmit_lock protection is necessary here.
1405 (f.e. loopback and IP tunnels are clean ignoring statistics
1407 However, it is possible, that they rely on protection
1410 Check this and shot the lock. It is not prone from deadlocks.
1411 Either shot noqueue qdisc, it is even simpler 8)
1413 if (dev->flags & IFF_UP) {
1414 int cpu = get_cpu();
1416 if (dev->xmit_lock_owner != cpu) {
1418 HARD_TX_LOCK_BH(dev, cpu);
1421 if (!netif_queue_stopped(dev)) {
1423 dev_queue_xmit_nit(skb, dev);
1426 if (!dev->hard_start_xmit(skb, dev)) {
1427 HARD_TX_UNLOCK_BH(dev);
1431 HARD_TX_UNLOCK_BH(dev);
1432 if (net_ratelimit())
1433 printk(KERN_CRIT "Virtual device %s asks to "
1434 "queue packet!\n", dev->name);
1438 /* Recursion is detected! It is possible,
1440 if (net_ratelimit())
1441 printk(KERN_CRIT "Dead loop on virtual device "
1442 "%s, fix it urgently!\n", dev->name);
1454 /*=======================================================================
1456 =======================================================================*/
1458 int netdev_max_backlog = 300;
1459 int weight_p = 64; /* old backlog weight */
1460 /* These numbers are selected based on intuition and some
1461 * experimentatiom, if you have more scientific way of doing this
1462 * please go ahead and fix things.
1464 int no_cong_thresh = 10;
1469 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1472 #ifdef CONFIG_NET_HW_FLOWCONTROL
1473 atomic_t netdev_dropping = ATOMIC_INIT(0);
1474 static unsigned long netdev_fc_mask = 1;
1475 unsigned long netdev_fc_xoff;
1476 spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1480 void (*stimul)(struct net_device *);
1481 struct net_device *dev;
1482 } netdev_fc_slots[BITS_PER_LONG];
1484 int netdev_register_fc(struct net_device *dev,
1485 void (*stimul)(struct net_device *dev))
1488 unsigned long flags;
1490 spin_lock_irqsave(&netdev_fc_lock, flags);
1491 if (netdev_fc_mask != ~0UL) {
1492 bit = ffz(netdev_fc_mask);
1493 netdev_fc_slots[bit].stimul = stimul;
1494 netdev_fc_slots[bit].dev = dev;
1495 set_bit(bit, &netdev_fc_mask);
1496 clear_bit(bit, &netdev_fc_xoff);
1498 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1502 void netdev_unregister_fc(int bit)
1504 unsigned long flags;
1506 spin_lock_irqsave(&netdev_fc_lock, flags);
1508 netdev_fc_slots[bit].stimul = NULL;
1509 netdev_fc_slots[bit].dev = NULL;
1510 clear_bit(bit, &netdev_fc_mask);
1511 clear_bit(bit, &netdev_fc_xoff);
1513 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1516 static void netdev_wakeup(void)
1520 spin_lock(&netdev_fc_lock);
1521 xoff = netdev_fc_xoff;
1526 netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1528 spin_unlock(&netdev_fc_lock);
1532 static void get_sample_stats(int cpu)
1538 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1539 int blog = sd->input_pkt_queue.qlen;
1540 int avg_blog = sd->avg_blog;
1542 avg_blog = (avg_blog >> 1) + (blog >> 1);
1544 if (avg_blog > mod_cong) {
1545 /* Above moderate congestion levels. */
1546 sd->cng_level = NET_RX_CN_HIGH;
1549 rq = rd % netdev_max_backlog;
1550 if (rq < avg_blog) /* unlucky bastard */
1551 sd->cng_level = NET_RX_DROP;
1553 } else if (avg_blog > lo_cong) {
1554 sd->cng_level = NET_RX_CN_MOD;
1557 rq = rd % netdev_max_backlog;
1558 if (rq < avg_blog) /* unlucky bastard */
1559 sd->cng_level = NET_RX_CN_HIGH;
1561 } else if (avg_blog > no_cong)
1562 sd->cng_level = NET_RX_CN_LOW;
1563 else /* no congestion */
1564 sd->cng_level = NET_RX_SUCCESS;
1566 sd->avg_blog = avg_blog;
1569 #ifdef OFFLINE_SAMPLE
1570 static void sample_queue(unsigned long dummy)
1572 /* 10 ms 0r 1ms -- i don't care -- JHS */
1574 int cpu = smp_processor_id();
1576 get_sample_stats(cpu);
1577 next_tick += jiffies;
1578 mod_timer(&samp_timer, next_tick);
1584 * netif_rx - post buffer to the network code
1585 * @skb: buffer to post
1587 * This function receives a packet from a device driver and queues it for
1588 * the upper (protocol) levels to process. It always succeeds. The buffer
1589 * may be dropped during processing for congestion control or by the
1593 * NET_RX_SUCCESS (no congestion)
1594 * NET_RX_CN_LOW (low congestion)
1595 * NET_RX_CN_MOD (moderate congestion)
1596 * NET_RX_CN_HIGH (high congestion)
1597 * NET_RX_DROP (packet was dropped)
1601 int netif_rx(struct sk_buff *skb)
1604 struct softnet_data *queue;
1605 unsigned long flags;
1607 #ifdef CONFIG_NETPOLL
1608 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1614 if (!skb->stamp.tv_sec)
1615 net_timestamp(&skb->stamp);
1618 * The code is rearranged so that the path is the most
1619 * short when CPU is congested, but is still operating.
1621 local_irq_save(flags);
1622 this_cpu = smp_processor_id();
1623 queue = &__get_cpu_var(softnet_data);
1625 __get_cpu_var(netdev_rx_stat).total++;
1626 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1627 if (queue->input_pkt_queue.qlen) {
1628 if (queue->throttle)
1633 __skb_queue_tail(&queue->input_pkt_queue, skb);
1634 #ifndef OFFLINE_SAMPLE
1635 get_sample_stats(this_cpu);
1637 local_irq_restore(flags);
1638 return queue->cng_level;
1641 if (queue->throttle) {
1642 queue->throttle = 0;
1643 #ifdef CONFIG_NET_HW_FLOWCONTROL
1644 if (atomic_dec_and_test(&netdev_dropping))
1649 netif_rx_schedule(&queue->backlog_dev);
1653 if (!queue->throttle) {
1654 queue->throttle = 1;
1655 __get_cpu_var(netdev_rx_stat).throttled++;
1656 #ifdef CONFIG_NET_HW_FLOWCONTROL
1657 atomic_inc(&netdev_dropping);
1662 __get_cpu_var(netdev_rx_stat).dropped++;
1663 local_irq_restore(flags);
1669 static __inline__ void skb_bond(struct sk_buff *skb)
1671 struct net_device *dev = skb->dev;
1674 skb->real_dev = skb->dev;
1675 skb->dev = dev->master;
1679 static void net_tx_action(struct softirq_action *h)
1681 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1683 if (sd->completion_queue) {
1684 struct sk_buff *clist;
1686 local_irq_disable();
1687 clist = sd->completion_queue;
1688 sd->completion_queue = NULL;
1692 struct sk_buff *skb = clist;
1693 clist = clist->next;
1695 BUG_TRAP(!atomic_read(&skb->users));
1700 if (sd->output_queue) {
1701 struct net_device *head;
1703 local_irq_disable();
1704 head = sd->output_queue;
1705 sd->output_queue = NULL;
1709 struct net_device *dev = head;
1710 head = head->next_sched;
1712 smp_mb__before_clear_bit();
1713 clear_bit(__LINK_STATE_SCHED, &dev->state);
1715 if (spin_trylock(&dev->queue_lock)) {
1717 spin_unlock(&dev->queue_lock);
1719 netif_schedule(dev);
1725 static __inline__ int deliver_skb(struct sk_buff *skb,
1726 struct packet_type *pt_prev, int last)
1728 atomic_inc(&skb->users);
1729 return pt_prev->func(skb, skb->dev, pt_prev);
1733 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1734 int (*br_handle_frame_hook)(struct sk_buff *skb);
1736 static __inline__ int handle_bridge(struct sk_buff *skb,
1737 struct packet_type *pt_prev)
1739 int ret = NET_RX_DROP;
1741 ret = deliver_skb(skb, pt_prev, 0);
1748 static inline int __handle_bridge(struct sk_buff *skb,
1749 struct packet_type **pt_prev, int *ret)
1751 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
1752 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1753 *ret = handle_bridge(skb, *pt_prev);
1754 if (br_handle_frame_hook(skb) == 0)
1764 #ifdef CONFIG_NET_CLS_ACT
1765 /* TODO: Maybe we should just force sch_ingress to be compiled in
1766 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1767 * a compare and 2 stores extra right now if we dont have it on
1768 * but have CONFIG_NET_CLS_ACT
1769 * NOTE: This doesnt stop any functionality; if you dont have
1770 * the ingress scheduler, you just cant add policies on ingress.
1773 int ing_filter(struct sk_buff *skb)
1776 struct net_device *dev = skb->dev;
1777 int result = TC_ACT_OK;
1779 if (dev->qdisc_ingress) {
1780 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1781 if (MAX_RED_LOOP < ttl++) {
1782 printk("Redir loop detected Dropping packet (%s->%s)\n",
1783 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1787 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1789 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1790 if (NULL == skb->input_dev) {
1791 skb->input_dev = skb->dev;
1792 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1794 spin_lock(&dev->ingress_lock);
1795 if ((q = dev->qdisc_ingress) != NULL)
1796 result = q->enqueue(skb, q);
1797 spin_unlock(&dev->ingress_lock);
1805 int netif_receive_skb(struct sk_buff *skb)
1807 struct packet_type *ptype, *pt_prev;
1808 int ret = NET_RX_DROP;
1809 unsigned short type;
1811 #ifdef CONFIG_NETPOLL
1812 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1818 if (!skb->stamp.tv_sec)
1819 net_timestamp(&skb->stamp);
1823 __get_cpu_var(netdev_rx_stat).total++;
1825 #ifdef CONFIG_NET_FASTROUTE
1826 if (skb->pkt_type == PACKET_FASTROUTE) {
1827 __get_cpu_var(netdev_rx_stat).fastroute_deferred_out++;
1828 return dev_queue_xmit(skb);
1832 skb->h.raw = skb->nh.raw = skb->data;
1833 skb->mac_len = skb->nh.raw - skb->mac.raw;
1836 #ifdef CONFIG_NET_CLS_ACT
1837 if (skb->tc_verd & TC_NCLS) {
1838 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1845 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1846 if (!ptype->dev || ptype->dev == skb->dev) {
1848 ret = deliver_skb(skb, pt_prev, 0);
1853 #ifdef CONFIG_NET_CLS_ACT
1855 atomic_inc(&skb->users);
1856 ret = pt_prev->func(skb, skb->dev, pt_prev);
1857 pt_prev = NULL; /* noone else should process this after*/
1859 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1862 ret = ing_filter(skb);
1864 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1873 handle_diverter(skb);
1875 if (__handle_bridge(skb, &pt_prev, &ret))
1878 type = skb->protocol;
1879 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1880 if (ptype->type == type &&
1881 (!ptype->dev || ptype->dev == skb->dev)) {
1883 ret = deliver_skb(skb, pt_prev, 0);
1889 ret = pt_prev->func(skb, skb->dev, pt_prev);
1892 /* Jamal, now you will not able to escape explaining
1893 * me how you were going to use this. :-)
1903 static int process_backlog(struct net_device *backlog_dev, int *budget)
1906 int quota = min(backlog_dev->quota, *budget);
1907 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1908 unsigned long start_time = jiffies;
1911 struct sk_buff *skb;
1912 struct net_device *dev;
1914 local_irq_disable();
1915 skb = __skb_dequeue(&queue->input_pkt_queue);
1922 netif_receive_skb(skb);
1928 if (work >= quota || jiffies - start_time > 1)
1931 #ifdef CONFIG_NET_HW_FLOWCONTROL
1932 if (queue->throttle &&
1933 queue->input_pkt_queue.qlen < no_cong_thresh ) {
1934 queue->throttle = 0;
1935 if (atomic_dec_and_test(&netdev_dropping)) {
1943 backlog_dev->quota -= work;
1948 backlog_dev->quota -= work;
1951 list_del(&backlog_dev->poll_list);
1952 smp_mb__before_clear_bit();
1953 netif_poll_enable(backlog_dev);
1955 if (queue->throttle) {
1956 queue->throttle = 0;
1957 #ifdef CONFIG_NET_HW_FLOWCONTROL
1958 if (atomic_dec_and_test(&netdev_dropping))
1966 static void net_rx_action(struct softirq_action *h)
1968 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1969 unsigned long start_time = jiffies;
1970 int budget = netdev_max_backlog;
1973 local_irq_disable();
1975 while (!list_empty(&queue->poll_list)) {
1976 struct net_device *dev;
1978 if (budget <= 0 || jiffies - start_time > 1)
1983 dev = list_entry(queue->poll_list.next,
1984 struct net_device, poll_list);
1986 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1987 local_irq_disable();
1988 list_del(&dev->poll_list);
1989 list_add_tail(&dev->poll_list, &queue->poll_list);
1991 dev->quota += dev->weight;
1993 dev->quota = dev->weight;
1996 local_irq_disable();
2004 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2005 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2009 static gifconf_func_t * gifconf_list [NPROTO];
2012 * register_gifconf - register a SIOCGIF handler
2013 * @family: Address family
2014 * @gifconf: Function handler
2016 * Register protocol dependent address dumping routines. The handler
2017 * that is passed must not be freed or reused until it has been replaced
2018 * by another handler.
2020 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2022 if (family >= NPROTO)
2024 gifconf_list[family] = gifconf;
2030 * Map an interface index to its name (SIOCGIFNAME)
2034 * We need this ioctl for efficient implementation of the
2035 * if_indextoname() function required by the IPv6 API. Without
2036 * it, we would have to search all the interfaces to find a
2040 static int dev_ifname(struct ifreq __user *arg)
2042 struct net_device *dev;
2046 * Fetch the caller's info block.
2049 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2052 read_lock(&dev_base_lock);
2053 dev = __dev_get_by_index(ifr.ifr_ifindex);
2055 read_unlock(&dev_base_lock);
2059 strcpy(ifr.ifr_name, dev->name);
2060 read_unlock(&dev_base_lock);
2062 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2068 * Perform a SIOCGIFCONF call. This structure will change
2069 * size eventually, and there is nothing I can do about it.
2070 * Thus we will need a 'compatibility mode'.
2073 static int dev_ifconf(char __user *arg)
2076 struct net_device *dev;
2083 * Fetch the caller's info block.
2086 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2093 * Loop over the interfaces, and write an info block for each.
2097 for (dev = dev_base; dev; dev = dev->next) {
2098 if (!dev_in_nx_info(dev, current->nx_info))
2100 for (i = 0; i < NPROTO; i++) {
2101 if (gifconf_list[i]) {
2104 done = gifconf_list[i](dev, NULL, 0);
2106 done = gifconf_list[i](dev, pos + total,
2116 * All done. Write the updated control block back to the caller.
2118 ifc.ifc_len = total;
2121 * Both BSD and Solaris return 0 here, so we do too.
2123 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2126 #ifdef CONFIG_PROC_FS
2128 * This is invoked by the /proc filesystem handler to display a device
2131 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2133 struct net_device *dev;
2136 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2138 return i == pos ? dev : NULL;
2141 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2143 read_lock(&dev_base_lock);
2144 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2147 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2150 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2153 void dev_seq_stop(struct seq_file *seq, void *v)
2155 read_unlock(&dev_base_lock);
2158 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2160 struct nx_info *nxi = current->nx_info;
2162 if (!dev_in_nx_info(dev, nxi))
2164 if (dev->get_stats) {
2165 struct net_device_stats *stats = dev->get_stats(dev);
2167 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2168 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2169 dev->name, stats->rx_bytes, stats->rx_packets,
2171 stats->rx_dropped + stats->rx_missed_errors,
2172 stats->rx_fifo_errors,
2173 stats->rx_length_errors + stats->rx_over_errors +
2174 stats->rx_crc_errors + stats->rx_frame_errors,
2175 stats->rx_compressed, stats->multicast,
2176 stats->tx_bytes, stats->tx_packets,
2177 stats->tx_errors, stats->tx_dropped,
2178 stats->tx_fifo_errors, stats->collisions,
2179 stats->tx_carrier_errors +
2180 stats->tx_aborted_errors +
2181 stats->tx_window_errors +
2182 stats->tx_heartbeat_errors,
2183 stats->tx_compressed);
2185 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2189 * Called from the PROCfs module. This now uses the new arbitrary sized
2190 * /proc/net interface to create /proc/net/dev
2192 static int dev_seq_show(struct seq_file *seq, void *v)
2194 if (v == SEQ_START_TOKEN)
2195 seq_puts(seq, "Inter-| Receive "
2197 " face |bytes packets errs drop fifo frame "
2198 "compressed multicast|bytes packets errs "
2199 "drop fifo colls carrier compressed\n");
2201 dev_seq_printf_stats(seq, v);
2205 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2207 struct netif_rx_stats *rc = NULL;
2209 while (*pos < NR_CPUS)
2210 if (cpu_online(*pos)) {
2211 rc = &per_cpu(netdev_rx_stat, *pos);
2218 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2220 return softnet_get_online(pos);
2223 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2226 return softnet_get_online(pos);
2229 static void softnet_seq_stop(struct seq_file *seq, void *v)
2233 static int softnet_seq_show(struct seq_file *seq, void *v)
2235 struct netif_rx_stats *s = v;
2237 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2238 s->total, s->dropped, s->time_squeeze, s->throttled,
2239 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2240 s->fastroute_deferred_out,
2242 s->fastroute_latency_reduction
2250 static struct seq_operations dev_seq_ops = {
2251 .start = dev_seq_start,
2252 .next = dev_seq_next,
2253 .stop = dev_seq_stop,
2254 .show = dev_seq_show,
2257 static int dev_seq_open(struct inode *inode, struct file *file)
2259 return seq_open(file, &dev_seq_ops);
2262 static struct file_operations dev_seq_fops = {
2263 .owner = THIS_MODULE,
2264 .open = dev_seq_open,
2266 .llseek = seq_lseek,
2267 .release = seq_release,
2270 static struct seq_operations softnet_seq_ops = {
2271 .start = softnet_seq_start,
2272 .next = softnet_seq_next,
2273 .stop = softnet_seq_stop,
2274 .show = softnet_seq_show,
2277 static int softnet_seq_open(struct inode *inode, struct file *file)
2279 return seq_open(file, &softnet_seq_ops);
2282 static struct file_operations softnet_seq_fops = {
2283 .owner = THIS_MODULE,
2284 .open = softnet_seq_open,
2286 .llseek = seq_lseek,
2287 .release = seq_release,
2291 extern int wireless_proc_init(void);
2293 #define wireless_proc_init() 0
2296 static int __init dev_proc_init(void)
2300 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2302 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2304 if (wireless_proc_init())
2310 proc_net_remove("softnet_stat");
2312 proc_net_remove("dev");
2316 #define dev_proc_init() 0
2317 #endif /* CONFIG_PROC_FS */
2321 * netdev_set_master - set up master/slave pair
2322 * @slave: slave device
2323 * @master: new master device
2325 * Changes the master device of the slave. Pass %NULL to break the
2326 * bonding. The caller must hold the RTNL semaphore. On a failure
2327 * a negative errno code is returned. On success the reference counts
2328 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2329 * function returns zero.
2331 int netdev_set_master(struct net_device *slave, struct net_device *master)
2333 struct net_device *old = slave->master;
2343 slave->master = master;
2351 slave->flags |= IFF_SLAVE;
2353 slave->flags &= ~IFF_SLAVE;
2355 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2360 * dev_set_promiscuity - update promiscuity count on a device
2364 * Add or remove promsicuity from a device. While the count in the device
2365 * remains above zero the interface remains promiscuous. Once it hits zero
2366 * the device reverts back to normal filtering operation. A negative inc
2367 * value is used to drop promiscuity on the device.
2369 void dev_set_promiscuity(struct net_device *dev, int inc)
2371 unsigned short old_flags = dev->flags;
2373 dev->flags |= IFF_PROMISC;
2374 if ((dev->promiscuity += inc) == 0)
2375 dev->flags &= ~IFF_PROMISC;
2376 if (dev->flags ^ old_flags) {
2377 #ifdef CONFIG_NET_FASTROUTE
2378 if (dev->flags & IFF_PROMISC) {
2379 netdev_fastroute_obstacles++;
2380 dev_clear_fastroute(dev);
2382 netdev_fastroute_obstacles--;
2385 printk(KERN_INFO "device %s %s promiscuous mode\n",
2386 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2392 * dev_set_allmulti - update allmulti count on a device
2396 * Add or remove reception of all multicast frames to a device. While the
2397 * count in the device remains above zero the interface remains listening
2398 * to all interfaces. Once it hits zero the device reverts back to normal
2399 * filtering operation. A negative @inc value is used to drop the counter
2400 * when releasing a resource needing all multicasts.
2403 void dev_set_allmulti(struct net_device *dev, int inc)
2405 unsigned short old_flags = dev->flags;
2407 dev->flags |= IFF_ALLMULTI;
2408 if ((dev->allmulti += inc) == 0)
2409 dev->flags &= ~IFF_ALLMULTI;
2410 if (dev->flags ^ old_flags)
2414 unsigned dev_get_flags(const struct net_device *dev)
2418 flags = (dev->flags & ~(IFF_PROMISC |
2421 (dev->gflags & (IFF_PROMISC |
2424 if (netif_running(dev) && netif_carrier_ok(dev))
2425 flags |= IFF_RUNNING;
2430 int dev_change_flags(struct net_device *dev, unsigned flags)
2433 int old_flags = dev->flags;
2436 * Set the flags on our device.
2439 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2440 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2442 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2446 * Load in the correct multicast list now the flags have changed.
2452 * Have we downed the interface. We handle IFF_UP ourselves
2453 * according to user attempts to set it, rather than blindly
2458 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2459 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2465 if (dev->flags & IFF_UP &&
2466 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2468 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2470 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2471 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2472 dev->gflags ^= IFF_PROMISC;
2473 dev_set_promiscuity(dev, inc);
2476 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2477 is important. Some (broken) drivers set IFF_PROMISC, when
2478 IFF_ALLMULTI is requested not asking us and not reporting.
2480 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2481 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2482 dev->gflags ^= IFF_ALLMULTI;
2483 dev_set_allmulti(dev, inc);
2486 if (old_flags ^ dev->flags)
2487 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2492 int dev_set_mtu(struct net_device *dev, int new_mtu)
2496 if (new_mtu == dev->mtu)
2499 /* MTU must be positive. */
2503 if (!netif_device_present(dev))
2507 if (dev->change_mtu)
2508 err = dev->change_mtu(dev, new_mtu);
2511 if (!err && dev->flags & IFF_UP)
2512 notifier_call_chain(&netdev_chain,
2513 NETDEV_CHANGEMTU, dev);
2519 * Perform the SIOCxIFxxx calls.
2521 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2524 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2530 case SIOCGIFFLAGS: /* Get interface flags */
2531 ifr->ifr_flags = dev_get_flags(dev);
2534 case SIOCSIFFLAGS: /* Set interface flags */
2535 return dev_change_flags(dev, ifr->ifr_flags);
2537 case SIOCGIFMETRIC: /* Get the metric on the interface
2538 (currently unused) */
2539 ifr->ifr_metric = 0;
2542 case SIOCSIFMETRIC: /* Set the metric on the interface
2543 (currently unused) */
2546 case SIOCGIFMTU: /* Get the MTU of a device */
2547 ifr->ifr_mtu = dev->mtu;
2550 case SIOCSIFMTU: /* Set the MTU of a device */
2551 return dev_set_mtu(dev, ifr->ifr_mtu);
2554 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2555 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2556 ifr->ifr_hwaddr.sa_family = dev->type;
2560 if (!dev->set_mac_address)
2562 if (ifr->ifr_hwaddr.sa_family != dev->type)
2564 if (!netif_device_present(dev))
2566 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2568 notifier_call_chain(&netdev_chain,
2569 NETDEV_CHANGEADDR, dev);
2572 case SIOCSIFHWBROADCAST:
2573 if (ifr->ifr_hwaddr.sa_family != dev->type)
2575 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2576 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2577 notifier_call_chain(&netdev_chain,
2578 NETDEV_CHANGEADDR, dev);
2582 ifr->ifr_map.mem_start = dev->mem_start;
2583 ifr->ifr_map.mem_end = dev->mem_end;
2584 ifr->ifr_map.base_addr = dev->base_addr;
2585 ifr->ifr_map.irq = dev->irq;
2586 ifr->ifr_map.dma = dev->dma;
2587 ifr->ifr_map.port = dev->if_port;
2591 if (dev->set_config) {
2592 if (!netif_device_present(dev))
2594 return dev->set_config(dev, &ifr->ifr_map);
2599 if (!dev->set_multicast_list ||
2600 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2602 if (!netif_device_present(dev))
2604 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2608 if (!dev->set_multicast_list ||
2609 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2611 if (!netif_device_present(dev))
2613 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2617 ifr->ifr_ifindex = dev->ifindex;
2621 ifr->ifr_qlen = dev->tx_queue_len;
2625 if (ifr->ifr_qlen < 0)
2627 dev->tx_queue_len = ifr->ifr_qlen;
2631 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2632 return dev_change_name(dev, ifr->ifr_newname);
2635 * Unknown or private ioctl
2639 if ((cmd >= SIOCDEVPRIVATE &&
2640 cmd <= SIOCDEVPRIVATE + 15) ||
2641 cmd == SIOCBONDENSLAVE ||
2642 cmd == SIOCBONDRELEASE ||
2643 cmd == SIOCBONDSETHWADDR ||
2644 cmd == SIOCBONDSLAVEINFOQUERY ||
2645 cmd == SIOCBONDINFOQUERY ||
2646 cmd == SIOCBONDCHANGEACTIVE ||
2647 cmd == SIOCGMIIPHY ||
2648 cmd == SIOCGMIIREG ||
2649 cmd == SIOCSMIIREG ||
2650 cmd == SIOCBRADDIF ||
2651 cmd == SIOCBRDELIF ||
2652 cmd == SIOCWANDEV) {
2654 if (dev->do_ioctl) {
2655 if (netif_device_present(dev))
2656 err = dev->do_ioctl(dev, ifr,
2669 * This function handles all "interface"-type I/O control requests. The actual
2670 * 'doing' part of this is dev_ifsioc above.
2674 * dev_ioctl - network device ioctl
2675 * @cmd: command to issue
2676 * @arg: pointer to a struct ifreq in user space
2678 * Issue ioctl functions to devices. This is normally called by the
2679 * user space syscall interfaces but can sometimes be useful for
2680 * other purposes. The return value is the return from the syscall if
2681 * positive or a negative errno code on error.
2684 int dev_ioctl(unsigned int cmd, void __user *arg)
2690 /* One special case: SIOCGIFCONF takes ifconf argument
2691 and requires shared lock, because it sleeps writing
2695 if (cmd == SIOCGIFCONF) {
2697 ret = dev_ifconf((char __user *) arg);
2701 if (cmd == SIOCGIFNAME)
2702 return dev_ifname((struct ifreq __user *)arg);
2704 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2707 ifr.ifr_name[IFNAMSIZ-1] = 0;
2709 colon = strchr(ifr.ifr_name, ':');
2714 * See which interface the caller is talking about.
2719 * These ioctl calls:
2720 * - can be done by all.
2721 * - atomic and do not require locking.
2732 dev_load(ifr.ifr_name);
2733 read_lock(&dev_base_lock);
2734 ret = dev_ifsioc(&ifr, cmd);
2735 read_unlock(&dev_base_lock);
2739 if (copy_to_user(arg, &ifr,
2740 sizeof(struct ifreq)))
2746 dev_load(ifr.ifr_name);
2748 ret = dev_ethtool(&ifr);
2753 if (copy_to_user(arg, &ifr,
2754 sizeof(struct ifreq)))
2760 * These ioctl calls:
2761 * - require superuser power.
2762 * - require strict serialization.
2768 if (!capable(CAP_NET_ADMIN))
2770 dev_load(ifr.ifr_name);
2772 ret = dev_ifsioc(&ifr, cmd);
2777 if (copy_to_user(arg, &ifr,
2778 sizeof(struct ifreq)))
2784 * These ioctl calls:
2785 * - require superuser power.
2786 * - require strict serialization.
2787 * - do not return a value
2797 case SIOCSIFHWBROADCAST:
2800 case SIOCBONDENSLAVE:
2801 case SIOCBONDRELEASE:
2802 case SIOCBONDSETHWADDR:
2803 case SIOCBONDSLAVEINFOQUERY:
2804 case SIOCBONDINFOQUERY:
2805 case SIOCBONDCHANGEACTIVE:
2808 if (!capable(CAP_NET_ADMIN))
2810 dev_load(ifr.ifr_name);
2812 ret = dev_ifsioc(&ifr, cmd);
2817 /* Get the per device memory space. We can add this but
2818 * currently do not support it */
2820 /* Set the per device memory buffer space.
2821 * Not applicable in our case */
2826 * Unknown or private ioctl.
2829 if (cmd == SIOCWANDEV ||
2830 (cmd >= SIOCDEVPRIVATE &&
2831 cmd <= SIOCDEVPRIVATE + 15)) {
2832 dev_load(ifr.ifr_name);
2834 ret = dev_ifsioc(&ifr, cmd);
2836 if (!ret && copy_to_user(arg, &ifr,
2837 sizeof(struct ifreq)))
2842 /* Take care of Wireless Extensions */
2843 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2844 /* If command is `set a parameter', or
2845 * `get the encoding parameters', check if
2846 * the user has the right to do it */
2847 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2848 if (!capable(CAP_NET_ADMIN))
2851 dev_load(ifr.ifr_name);
2853 /* Follow me in net/core/wireless.c */
2854 ret = wireless_process_ioctl(&ifr, cmd);
2856 if (!ret && IW_IS_GET(cmd) &&
2857 copy_to_user(arg, &ifr,
2858 sizeof(struct ifreq)))
2862 #endif /* WIRELESS_EXT */
2869 * dev_new_index - allocate an ifindex
2871 * Returns a suitable unique value for a new device interface
2872 * number. The caller must hold the rtnl semaphore or the
2873 * dev_base_lock to be sure it remains unique.
2875 int dev_new_index(void)
2881 if (!__dev_get_by_index(ifindex))
2886 static int dev_boot_phase = 1;
2888 /* Delayed registration/unregisteration */
2889 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2890 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2892 static inline void net_set_todo(struct net_device *dev)
2894 spin_lock(&net_todo_list_lock);
2895 list_add_tail(&dev->todo_list, &net_todo_list);
2896 spin_unlock(&net_todo_list_lock);
2900 * register_netdevice - register a network device
2901 * @dev: device to register
2903 * Take a completed network device structure and add it to the kernel
2904 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2905 * chain. 0 is returned on success. A negative errno code is returned
2906 * on a failure to set up the device, or if the name is a duplicate.
2908 * Callers must hold the rtnl semaphore. See the comment at the
2909 * end of Space.c for details about the locking. You may want
2910 * register_netdev() instead of this.
2913 * The locking appears insufficient to guarantee two parallel registers
2914 * will not get the same name.
2917 int register_netdevice(struct net_device *dev)
2919 struct hlist_head *head;
2920 struct hlist_node *p;
2923 BUG_ON(dev_boot_phase);
2926 /* When net_device's are persistent, this will be fatal. */
2927 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2929 spin_lock_init(&dev->queue_lock);
2930 spin_lock_init(&dev->xmit_lock);
2931 dev->xmit_lock_owner = -1;
2932 #ifdef CONFIG_NET_CLS_ACT
2933 spin_lock_init(&dev->ingress_lock);
2936 #ifdef CONFIG_NET_FASTROUTE
2937 dev->fastpath_lock = RW_LOCK_UNLOCKED;
2940 ret = alloc_divert_blk(dev);
2946 /* Init, if this function is available */
2948 ret = dev->init(dev);
2956 if (!dev_valid_name(dev->name)) {
2961 dev->ifindex = dev_new_index();
2962 if (dev->iflink == -1)
2963 dev->iflink = dev->ifindex;
2965 /* Check for existence of name */
2966 head = dev_name_hash(dev->name);
2967 hlist_for_each(p, head) {
2968 struct net_device *d
2969 = hlist_entry(p, struct net_device, name_hlist);
2970 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2976 /* Fix illegal SG+CSUM combinations. */
2977 if ((dev->features & NETIF_F_SG) &&
2978 !(dev->features & (NETIF_F_IP_CSUM |
2980 NETIF_F_HW_CSUM))) {
2981 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2983 dev->features &= ~NETIF_F_SG;
2987 * nil rebuild_header routine,
2988 * that should be never called and used as just bug trap.
2991 if (!dev->rebuild_header)
2992 dev->rebuild_header = default_rebuild_header;
2995 * Default initial state at registry is that the
2996 * device is present.
2999 set_bit(__LINK_STATE_PRESENT, &dev->state);
3002 dev_init_scheduler(dev);
3003 write_lock_bh(&dev_base_lock);
3005 dev_tail = &dev->next;
3006 hlist_add_head(&dev->name_hlist, head);
3007 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3009 dev->reg_state = NETREG_REGISTERING;
3010 write_unlock_bh(&dev_base_lock);
3012 /* Notify protocols, that a new device appeared. */
3013 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3015 /* Finish registration after unlock */
3022 free_divert_blk(dev);
3027 * netdev_wait_allrefs - wait until all references are gone.
3029 * This is called when unregistering network devices.
3031 * Any protocol or device that holds a reference should register
3032 * for netdevice notification, and cleanup and put back the
3033 * reference if they receive an UNREGISTER event.
3034 * We can get stuck here if buggy protocols don't correctly
3037 static void netdev_wait_allrefs(struct net_device *dev)
3039 unsigned long rebroadcast_time, warning_time;
3041 rebroadcast_time = warning_time = jiffies;
3042 while (atomic_read(&dev->refcnt) != 0) {
3043 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3047 /* Rebroadcast unregister notification */
3048 notifier_call_chain(&netdev_chain,
3049 NETDEV_UNREGISTER, dev);
3051 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3053 /* We must not have linkwatch events
3054 * pending on unregister. If this
3055 * happens, we simply run the queue
3056 * unscheduled, resulting in a noop
3059 linkwatch_run_queue();
3065 rebroadcast_time = jiffies;
3068 current->state = TASK_INTERRUPTIBLE;
3069 schedule_timeout(HZ / 4);
3071 if (time_after(jiffies, warning_time + 10 * HZ)) {
3072 printk(KERN_EMERG "unregister_netdevice: "
3073 "waiting for %s to become free. Usage "
3075 dev->name, atomic_read(&dev->refcnt));
3076 warning_time = jiffies;
3085 * register_netdevice(x1);
3086 * register_netdevice(x2);
3088 * unregister_netdevice(y1);
3089 * unregister_netdevice(y2);
3095 * We are invoked by rtnl_unlock() after it drops the semaphore.
3096 * This allows us to deal with problems:
3097 * 1) We can create/delete sysfs objects which invoke hotplug
3098 * without deadlocking with linkwatch via keventd.
3099 * 2) Since we run with the RTNL semaphore not held, we can sleep
3100 * safely in order to wait for the netdev refcnt to drop to zero.
3102 static DECLARE_MUTEX(net_todo_run_mutex);
3103 void netdev_run_todo(void)
3105 struct list_head list = LIST_HEAD_INIT(list);
3109 /* Need to guard against multiple cpu's getting out of order. */
3110 down(&net_todo_run_mutex);
3112 /* Not safe to do outside the semaphore. We must not return
3113 * until all unregister events invoked by the local processor
3114 * have been completed (either by this todo run, or one on
3117 if (list_empty(&net_todo_list))
3120 /* Snapshot list, allow later requests */
3121 spin_lock(&net_todo_list_lock);
3122 list_splice_init(&net_todo_list, &list);
3123 spin_unlock(&net_todo_list_lock);
3125 while (!list_empty(&list)) {
3126 struct net_device *dev
3127 = list_entry(list.next, struct net_device, todo_list);
3128 list_del(&dev->todo_list);
3130 switch(dev->reg_state) {
3131 case NETREG_REGISTERING:
3132 err = netdev_register_sysfs(dev);
3134 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3136 dev->reg_state = NETREG_REGISTERED;
3139 case NETREG_UNREGISTERING:
3140 netdev_unregister_sysfs(dev);
3141 dev->reg_state = NETREG_UNREGISTERED;
3143 netdev_wait_allrefs(dev);
3146 BUG_ON(atomic_read(&dev->refcnt));
3147 BUG_TRAP(!dev->ip_ptr);
3148 BUG_TRAP(!dev->ip6_ptr);
3149 BUG_TRAP(!dev->dn_ptr);
3152 /* It must be the very last action,
3153 * after this 'dev' may point to freed up memory.
3155 if (dev->destructor)
3156 dev->destructor(dev);
3160 printk(KERN_ERR "network todo '%s' but state %d\n",
3161 dev->name, dev->reg_state);
3167 up(&net_todo_run_mutex);
3171 * free_netdev - free network device
3174 * This function does the last stage of destroying an allocated device
3175 * interface. The reference to the device object is released.
3176 * If this is the last reference then it will be freed.
3178 void free_netdev(struct net_device *dev)
3181 /* Compatiablity with error handling in drivers */
3182 if (dev->reg_state == NETREG_UNINITIALIZED) {
3183 kfree((char *)dev - dev->padded);
3187 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3188 dev->reg_state = NETREG_RELEASED;
3190 /* will free via class release */
3191 class_device_put(&dev->class_dev);
3193 kfree((char *)dev - dev->padded);
3197 /* Synchronize with packet receive processing. */
3198 void synchronize_net(void)
3201 synchronize_kernel();
3205 * unregister_netdevice - remove device from the kernel
3208 * This function shuts down a device interface and removes it
3209 * from the kernel tables. On success 0 is returned, on a failure
3210 * a negative errno code is returned.
3212 * Callers must hold the rtnl semaphore. See the comment at the
3213 * end of Space.c for details about the locking. You may want
3214 * unregister_netdev() instead of this.
3217 int unregister_netdevice(struct net_device *dev)
3219 struct net_device *d, **dp;
3221 BUG_ON(dev_boot_phase);
3224 /* Some devices call without registering for initialization unwind. */
3225 if (dev->reg_state == NETREG_UNINITIALIZED) {
3226 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3227 "was registered\n", dev->name, dev);
3231 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3233 /* If device is running, close it first. */
3234 if (dev->flags & IFF_UP)
3237 /* And unlink it from device chain. */
3238 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3240 write_lock_bh(&dev_base_lock);
3241 hlist_del(&dev->name_hlist);
3242 hlist_del(&dev->index_hlist);
3243 if (dev_tail == &dev->next)
3246 write_unlock_bh(&dev_base_lock);
3251 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3256 dev->reg_state = NETREG_UNREGISTERING;
3260 #ifdef CONFIG_NET_FASTROUTE
3261 dev_clear_fastroute(dev);
3264 /* Shutdown queueing discipline. */
3268 /* Notify protocols, that we are about to destroy
3269 this device. They should clean all the things.
3271 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3274 * Flush the multicast chain
3276 dev_mc_discard(dev);
3281 /* Notifier chain MUST detach us from master device. */
3282 BUG_TRAP(!dev->master);
3284 free_divert_blk(dev);
3286 /* Finish processing unregister after unlock */
3295 #ifdef CONFIG_HOTPLUG_CPU
3296 static int dev_cpu_callback(struct notifier_block *nfb,
3297 unsigned long action,
3300 struct sk_buff **list_skb;
3301 struct net_device **list_net;
3302 struct sk_buff *skb;
3303 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3304 struct softnet_data *sd, *oldsd;
3306 if (action != CPU_DEAD)
3309 local_irq_disable();
3310 cpu = smp_processor_id();
3311 sd = &per_cpu(softnet_data, cpu);
3312 oldsd = &per_cpu(softnet_data, oldcpu);
3314 /* Find end of our completion_queue. */
3315 list_skb = &sd->completion_queue;
3317 list_skb = &(*list_skb)->next;
3318 /* Append completion queue from offline CPU. */
3319 *list_skb = oldsd->completion_queue;
3320 oldsd->completion_queue = NULL;
3322 /* Find end of our output_queue. */
3323 list_net = &sd->output_queue;
3325 list_net = &(*list_net)->next_sched;
3326 /* Append output queue from offline CPU. */
3327 *list_net = oldsd->output_queue;
3328 oldsd->output_queue = NULL;
3330 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3333 /* Process offline CPU's input_pkt_queue */
3334 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3339 #endif /* CONFIG_HOTPLUG_CPU */
3343 * Initialize the DEV module. At boot time this walks the device list and
3344 * unhooks any devices that fail to initialise (normally hardware not
3345 * present) and leaves us with a valid list of present and active devices.
3350 * This is called single threaded during boot, so no need
3351 * to take the rtnl semaphore.
3353 static int __init net_dev_init(void)
3355 int i, rc = -ENOMEM;
3357 BUG_ON(!dev_boot_phase);
3359 if (dev_proc_init())
3362 if (netdev_sysfs_init())
3365 INIT_LIST_HEAD(&ptype_all);
3366 for (i = 0; i < 16; i++)
3367 INIT_LIST_HEAD(&ptype_base[i]);
3369 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3370 INIT_HLIST_HEAD(&dev_name_head[i]);
3372 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3373 INIT_HLIST_HEAD(&dev_index_head[i]);
3376 * Initialise the packet receive queues.
3379 for (i = 0; i < NR_CPUS; i++) {
3380 struct softnet_data *queue;
3382 queue = &per_cpu(softnet_data, i);
3383 skb_queue_head_init(&queue->input_pkt_queue);
3384 queue->throttle = 0;
3385 queue->cng_level = 0;
3386 queue->avg_blog = 10; /* arbitrary non-zero */
3387 queue->completion_queue = NULL;
3388 INIT_LIST_HEAD(&queue->poll_list);
3389 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3390 queue->backlog_dev.weight = weight_p;
3391 queue->backlog_dev.poll = process_backlog;
3392 atomic_set(&queue->backlog_dev.refcnt, 1);
3395 #ifdef OFFLINE_SAMPLE
3396 samp_timer.expires = jiffies + (10 * HZ);
3397 add_timer(&samp_timer);
3402 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3403 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3405 hotcpu_notifier(dev_cpu_callback, 0);
3413 subsys_initcall(net_dev_init);
3415 EXPORT_SYMBOL(__dev_get);
3416 EXPORT_SYMBOL(__dev_get_by_flags);
3417 EXPORT_SYMBOL(__dev_get_by_index);
3418 EXPORT_SYMBOL(__dev_get_by_name);
3419 EXPORT_SYMBOL(__dev_remove_pack);
3420 EXPORT_SYMBOL(__skb_linearize);
3421 EXPORT_SYMBOL(call_netdevice_notifiers);
3422 EXPORT_SYMBOL(dev_add_pack);
3423 EXPORT_SYMBOL(dev_alloc_name);
3424 EXPORT_SYMBOL(dev_close);
3425 EXPORT_SYMBOL(dev_get_by_flags);
3426 EXPORT_SYMBOL(dev_get_by_index);
3427 EXPORT_SYMBOL(dev_get_by_name);
3428 EXPORT_SYMBOL(dev_getbyhwaddr);
3429 EXPORT_SYMBOL(dev_ioctl);
3430 EXPORT_SYMBOL(dev_new_index);
3431 EXPORT_SYMBOL(dev_open);
3432 EXPORT_SYMBOL(dev_queue_xmit);
3433 EXPORT_SYMBOL(dev_queue_xmit_nit);
3434 EXPORT_SYMBOL(dev_remove_pack);
3435 EXPORT_SYMBOL(dev_set_allmulti);
3436 EXPORT_SYMBOL(dev_set_promiscuity);
3437 EXPORT_SYMBOL(free_netdev);
3438 EXPORT_SYMBOL(netdev_boot_setup_check);
3439 EXPORT_SYMBOL(netdev_set_master);
3440 EXPORT_SYMBOL(netdev_state_change);
3441 EXPORT_SYMBOL(netif_receive_skb);
3442 EXPORT_SYMBOL(netif_rx);
3443 EXPORT_SYMBOL(register_gifconf);
3444 EXPORT_SYMBOL(register_netdevice);
3445 EXPORT_SYMBOL(register_netdevice_notifier);
3446 EXPORT_SYMBOL(skb_checksum_help);
3447 EXPORT_SYMBOL(synchronize_net);
3448 EXPORT_SYMBOL(unregister_netdevice);
3449 EXPORT_SYMBOL(unregister_netdevice_notifier);
3451 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3452 EXPORT_SYMBOL(br_handle_frame_hook);
3454 /* for 801q VLAN support */
3455 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
3456 EXPORT_SYMBOL(dev_change_flags);
3459 EXPORT_SYMBOL(dev_load);
3461 #ifdef CONFIG_NET_HW_FLOWCONTROL
3462 EXPORT_SYMBOL(netdev_dropping);
3463 EXPORT_SYMBOL(netdev_fc_xoff);
3464 EXPORT_SYMBOL(netdev_register_fc);
3465 EXPORT_SYMBOL(netdev_unregister_fc);
3467 #ifdef CONFIG_NET_FASTROUTE
3468 EXPORT_SYMBOL(netdev_fastroute);
3469 EXPORT_SYMBOL(netdev_fastroute_obstacles);
3472 #ifdef CONFIG_NET_CLS_ACT
3473 EXPORT_SYMBOL(ing_filter);
3477 EXPORT_PER_CPU_SYMBOL(softnet_data);