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 #if defined(CONFIG_MIPS_BRCM)
118 /* NOTE: This value should correspond to the value of
119 ** SIOCGIFTRANSSTART in bcmnet.h. Since bcmnet.h is not
120 ** visible from the linux/net/core directory,
121 ** we have to hardcode this value here */
122 #define SIOCGIFTRANSSTART (SIOCDEVPRIVATE + 3)
123 #define SIOCCIFSTATS (SIOCDEVPRIVATE + 7)
126 /* This define, if set, will randomly drop a packet when congestion
127 * is more than moderate. It helps fairness in the multi-interface
128 * case when one of them is a hog, but it kills performance for the
129 * single interface case so it is off now by default.
133 /* Setting this will sample the queue lengths and thus congestion
134 * via a timer instead of as each packet is received.
136 #undef OFFLINE_SAMPLE
139 * The list of packet types we will receive (as opposed to discard)
140 * and the routines to invoke.
142 * Why 16. Because with 16 the only overlap we get on a hash of the
143 * low nibble of the protocol value is RARP/SNAP/X.25.
145 * NOTE: That is no longer true with the addition of VLAN tags. Not
146 * sure which should go first, but I bet it won't make much
147 * difference if we are running VLANs. The good news is that
148 * this protocol won't be in the list unless compiled in, so
149 * the average user (w/out VLANs) will not be adversly affected.
166 static spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
167 static struct list_head ptype_base[16]; /* 16 way hashed list */
168 static struct list_head ptype_all; /* Taps */
170 #ifdef OFFLINE_SAMPLE
171 static void sample_queue(unsigned long dummy);
172 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
176 * The @dev_base list is protected by @dev_base_lock and the rtln
179 * Pure readers hold dev_base_lock for reading.
181 * Writers must hold the rtnl semaphore while they loop through the
182 * dev_base list, and hold dev_base_lock for writing when they do the
183 * actual updates. This allows pure readers to access the list even
184 * while a writer is preparing to update it.
186 * To put it another way, dev_base_lock is held for writing only to
187 * protect against pure readers; the rtnl semaphore provides the
188 * protection against other writers.
190 * See, for example usages, register_netdevice() and
191 * unregister_netdevice(), which must be called with the rtnl
194 struct net_device *dev_base;
195 struct net_device **dev_tail = &dev_base;
196 rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
198 EXPORT_SYMBOL(dev_base);
199 EXPORT_SYMBOL(dev_base_lock);
201 #define NETDEV_HASHBITS 8
202 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
203 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
205 static inline struct hlist_head *dev_name_hash(const char *name)
207 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
208 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
211 static inline struct hlist_head *dev_index_hash(int ifindex)
213 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
220 static struct notifier_block *netdev_chain;
223 * Device drivers call our routines to queue packets here. We empty the
224 * queue in the local softnet handler.
226 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
229 extern int netdev_sysfs_init(void);
230 extern int netdev_register_sysfs(struct net_device *);
231 extern void netdev_unregister_sysfs(struct net_device *);
233 #define netdev_sysfs_init() (0)
234 #define netdev_register_sysfs(dev) (0)
235 #define netdev_unregister_sysfs(dev) do { } while(0)
239 /*******************************************************************************
241 Protocol management and registration routines
243 *******************************************************************************/
252 * Add a protocol ID to the list. Now that the input handler is
253 * smarter we can dispense with all the messy stuff that used to be
256 * BEWARE!!! Protocol handlers, mangling input packets,
257 * MUST BE last in hash buckets and checking protocol handlers
258 * MUST start from promiscuous ptype_all chain in net_bh.
259 * It is true now, do not change it.
260 * Explanation follows: if protocol handler, mangling packet, will
261 * be the first on list, it is not able to sense, that packet
262 * is cloned and should be copied-on-write, so that it will
263 * change it and subsequent readers will get broken packet.
268 * dev_add_pack - add packet handler
269 * @pt: packet type declaration
271 * Add a protocol handler to the networking stack. The passed &packet_type
272 * is linked into kernel lists and may not be freed until it has been
273 * removed from the kernel lists.
275 * This call does not sleep therefore it can not
276 * guarantee all CPU's that are in middle of receiving packets
277 * will see the new packet type (until the next received packet).
280 void dev_add_pack(struct packet_type *pt)
284 spin_lock_bh(&ptype_lock);
285 if (pt->type == htons(ETH_P_ALL)) {
287 list_add_rcu(&pt->list, &ptype_all);
289 hash = ntohs(pt->type) & 15;
290 list_add_rcu(&pt->list, &ptype_base[hash]);
292 spin_unlock_bh(&ptype_lock);
295 extern void linkwatch_run_queue(void);
300 * __dev_remove_pack - remove packet handler
301 * @pt: packet type declaration
303 * Remove a protocol handler that was previously added to the kernel
304 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
305 * from the kernel lists and can be freed or reused once this function
308 * The packet type might still be in use by receivers
309 * and must not be freed until after all the CPU's have gone
310 * through a quiescent state.
312 void __dev_remove_pack(struct packet_type *pt)
314 struct list_head *head;
315 struct packet_type *pt1;
317 spin_lock_bh(&ptype_lock);
319 if (pt->type == htons(ETH_P_ALL)) {
323 head = &ptype_base[ntohs(pt->type) & 15];
325 list_for_each_entry(pt1, head, list) {
327 list_del_rcu(&pt->list);
332 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
334 spin_unlock_bh(&ptype_lock);
337 * dev_remove_pack - remove packet handler
338 * @pt: packet type declaration
340 * Remove a protocol handler that was previously added to the kernel
341 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
342 * from the kernel lists and can be freed or reused once this function
345 * This call sleeps to guarantee that no CPU is looking at the packet
348 void dev_remove_pack(struct packet_type *pt)
350 __dev_remove_pack(pt);
355 /******************************************************************************
357 Device Boot-time Settings Routines
359 *******************************************************************************/
361 /* Boot time configuration table */
362 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
365 * netdev_boot_setup_add - add new setup entry
366 * @name: name of the device
367 * @map: configured settings for the device
369 * Adds new setup entry to the dev_boot_setup list. The function
370 * returns 0 on error and 1 on success. This is a generic routine to
373 int netdev_boot_setup_add(char *name, struct ifmap *map)
375 struct netdev_boot_setup *s;
379 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
380 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
381 memset(s[i].name, 0, sizeof(s[i].name));
382 strcpy(s[i].name, name);
383 memcpy(&s[i].map, map, sizeof(s[i].map));
388 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
392 * netdev_boot_setup_check - check boot time settings
393 * @dev: the netdevice
395 * Check boot time settings for the device.
396 * The found settings are set for the device to be used
397 * later in the device probing.
398 * Returns 0 if no settings found, 1 if they are.
400 int netdev_boot_setup_check(struct net_device *dev)
402 struct netdev_boot_setup *s = dev_boot_setup;
405 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
406 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
407 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
408 dev->irq = s[i].map.irq;
409 dev->base_addr = s[i].map.base_addr;
410 dev->mem_start = s[i].map.mem_start;
411 dev->mem_end = s[i].map.mem_end;
420 * netdev_boot_base - get address from boot time settings
421 * @prefix: prefix for network device
422 * @unit: id for network device
424 * Check boot time settings for the base address of device.
425 * The found settings are set for the device to be used
426 * later in the device probing.
427 * Returns 0 if no settings found.
429 unsigned long netdev_boot_base(const char *prefix, int unit)
431 const struct netdev_boot_setup *s = dev_boot_setup;
435 sprintf(name, "%s%d", prefix, unit);
438 * If device already registered then return base of 1
439 * to indicate not to probe for this interface
441 if (__dev_get_by_name(name))
444 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
445 if (!strcmp(name, s[i].name))
446 return s[i].map.base_addr;
451 * Saves at boot time configured settings for any netdevice.
453 int __init netdev_boot_setup(char *str)
458 str = get_options(str, ARRAY_SIZE(ints), ints);
463 memset(&map, 0, sizeof(map));
467 map.base_addr = ints[2];
469 map.mem_start = ints[3];
471 map.mem_end = ints[4];
473 /* Add new entry to the list */
474 return netdev_boot_setup_add(str, &map);
477 __setup("netdev=", netdev_boot_setup);
479 /*******************************************************************************
481 Device Interface Subroutines
483 *******************************************************************************/
486 * __dev_get_by_name - find a device by its name
487 * @name: name to find
489 * Find an interface by name. Must be called under RTNL semaphore
490 * or @dev_base_lock. If the name is found a pointer to the device
491 * is returned. If the name is not found then %NULL is returned. The
492 * reference counters are not incremented so the caller must be
493 * careful with locks.
496 struct net_device *__dev_get_by_name(const char *name)
498 struct hlist_node *p;
500 hlist_for_each(p, dev_name_hash(name)) {
501 struct net_device *dev
502 = hlist_entry(p, struct net_device, name_hlist);
503 if (!strncmp(dev->name, name, IFNAMSIZ))
510 * dev_get_by_name - find a device by its name
511 * @name: name to find
513 * Find an interface by name. This can be called from any
514 * context and does its own locking. The returned handle has
515 * the usage count incremented and the caller must use dev_put() to
516 * release it when it is no longer needed. %NULL is returned if no
517 * matching device is found.
520 struct net_device *dev_get_by_name(const char *name)
522 struct net_device *dev;
524 read_lock(&dev_base_lock);
525 dev = __dev_get_by_name(name);
528 read_unlock(&dev_base_lock);
533 Return value is changed to int to prevent illegal usage in future.
534 It is still legal to use to check for device existence.
536 User should understand, that the result returned by this function
537 is meaningless, if it was not issued under rtnl semaphore.
541 * dev_get - test if a device exists
542 * @name: name to test for
544 * Test if a name exists. Returns true if the name is found. In order
545 * to be sure the name is not allocated or removed during the test the
546 * caller must hold the rtnl semaphore.
548 * This function exists only for back compatibility with older
551 int __dev_get(const char *name)
553 struct net_device *dev;
555 read_lock(&dev_base_lock);
556 dev = __dev_get_by_name(name);
557 read_unlock(&dev_base_lock);
562 * __dev_get_by_index - find a device by its ifindex
563 * @ifindex: index of device
565 * Search for an interface by index. Returns %NULL if the device
566 * is not found or a pointer to the device. The device has not
567 * had its reference counter increased so the caller must be careful
568 * about locking. The caller must hold either the RTNL semaphore
572 struct net_device *__dev_get_by_index(int ifindex)
574 struct hlist_node *p;
576 hlist_for_each(p, dev_index_hash(ifindex)) {
577 struct net_device *dev
578 = hlist_entry(p, struct net_device, index_hlist);
579 if (dev->ifindex == ifindex)
587 * dev_get_by_index - find a device by its ifindex
588 * @ifindex: index of device
590 * Search for an interface by index. Returns NULL if the device
591 * is not found or a pointer to the device. The device returned has
592 * had a reference added and the pointer is safe until the user calls
593 * dev_put to indicate they have finished with it.
596 struct net_device *dev_get_by_index(int ifindex)
598 struct net_device *dev;
600 read_lock(&dev_base_lock);
601 dev = __dev_get_by_index(ifindex);
604 read_unlock(&dev_base_lock);
609 * dev_getbyhwaddr - find a device by its hardware address
610 * @type: media type of device
611 * @ha: hardware address
613 * Search for an interface by MAC address. Returns NULL if the device
614 * is not found or a pointer to the device. The caller must hold the
615 * rtnl semaphore. The returned device has not had its ref count increased
616 * and the caller must therefore be careful about locking
619 * If the API was consistent this would be __dev_get_by_hwaddr
622 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
624 struct net_device *dev;
628 for (dev = dev_base; dev; dev = dev->next)
629 if (dev->type == type &&
630 !memcmp(dev->dev_addr, ha, dev->addr_len))
635 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
637 struct net_device *dev;
639 for (dev = dev_base; dev; dev = dev->next)
640 if (dev->type == type)
645 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
647 struct net_device *dev_getfirstbyhwtype(unsigned short type)
649 struct net_device *dev;
652 dev = __dev_getfirstbyhwtype(type);
659 EXPORT_SYMBOL(dev_getfirstbyhwtype);
662 * dev_get_by_flags - find any device with given flags
663 * @if_flags: IFF_* values
664 * @mask: bitmask of bits in if_flags to check
666 * Search for any interface with the given flags. Returns NULL if a device
667 * is not found or a pointer to the device. The device returned has
668 * had a reference added and the pointer is safe until the user calls
669 * dev_put to indicate they have finished with it.
672 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
674 struct net_device *dev;
676 read_lock(&dev_base_lock);
677 dev = __dev_get_by_flags(if_flags, mask);
680 read_unlock(&dev_base_lock);
685 * __dev_get_by_flags - find any device with given flags
686 * @if_flags: IFF_* values
687 * @mask: bitmask of bits in if_flags to check
689 * Search for any interface with the given flags. Returns NULL if a device
690 * is not found or a pointer to the device. The caller must hold either
691 * the RTNL semaphore or @dev_base_lock.
694 struct net_device *__dev_get_by_flags(unsigned short if_flags, unsigned short mask)
696 struct net_device *dev;
698 for (dev = dev_base; dev != NULL; dev = dev->next) {
699 if (((dev->flags ^ if_flags) & mask) == 0)
706 * dev_valid_name - check if name is okay for network device
709 * Network device names need to be valid file names to
710 * to allow sysfs to work
712 int dev_valid_name(const char *name)
714 return !(*name == '\0'
715 || !strcmp(name, ".")
716 || !strcmp(name, "..")
717 || strchr(name, '/'));
721 * dev_alloc_name - allocate a name for a device
723 * @name: name format string
725 * Passed a format string - eg "lt%d" it will try and find a suitable
726 * id. Not efficient for many devices, not called a lot. The caller
727 * must hold the dev_base or rtnl lock while allocating the name and
728 * adding the device in order to avoid duplicates. Returns the number
729 * of the unit assigned or a negative errno code.
732 int dev_alloc_name(struct net_device *dev, const char *name)
737 const int max_netdevices = 8*PAGE_SIZE;
739 struct net_device *d;
741 p = strnchr(name, IFNAMSIZ-1, '%');
744 * Verify the string as this thing may have come from
745 * the user. There must be either one "%d" and no other "%"
748 if (p[1] != 'd' || strchr(p + 2, '%'))
751 /* Use one page as a bit array of possible slots */
752 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
756 for (d = dev_base; d; d = d->next) {
757 if (!sscanf(d->name, name, &i))
759 if (i < 0 || i >= max_netdevices)
762 /* avoid cases where sscanf is not exact inverse of printf */
763 snprintf(buf, sizeof(buf), name, i);
764 if (!strncmp(buf, d->name, IFNAMSIZ))
768 i = find_first_zero_bit(inuse, max_netdevices);
769 free_page((unsigned long) inuse);
772 snprintf(buf, sizeof(buf), name, i);
773 if (!__dev_get_by_name(buf)) {
774 strlcpy(dev->name, buf, IFNAMSIZ);
778 /* It is possible to run out of possible slots
779 * when the name is long and there isn't enough space left
780 * for the digits, or if all bits are used.
787 * dev_change_name - change name of a device
789 * @newname: name (or format string) must be at least IFNAMSIZ
791 * Change name of a device, can pass format strings "eth%d".
794 int dev_change_name(struct net_device *dev, char *newname)
800 if (dev->flags & IFF_UP)
803 if (!dev_valid_name(newname))
806 if (strchr(newname, '%')) {
807 err = dev_alloc_name(dev, newname);
810 strcpy(newname, dev->name);
812 else if (__dev_get_by_name(newname))
815 strlcpy(dev->name, newname, IFNAMSIZ);
817 err = class_device_rename(&dev->class_dev, dev->name);
819 hlist_del(&dev->name_hlist);
820 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
821 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
828 * netdev_state_change - device changes state
829 * @dev: device to cause notification
831 * Called to indicate a device has changed state. This function calls
832 * the notifier chains for netdev_chain and sends a NEWLINK message
833 * to the routing socket.
835 void netdev_state_change(struct net_device *dev)
837 if (dev->flags & IFF_UP) {
838 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
839 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
844 * dev_load - load a network module
845 * @name: name of interface
847 * If a network interface is not present and the process has suitable
848 * privileges this function loads the module. If module loading is not
849 * available in this kernel then it becomes a nop.
852 void dev_load(const char *name)
854 struct net_device *dev;
856 read_lock(&dev_base_lock);
857 dev = __dev_get_by_name(name);
858 read_unlock(&dev_base_lock);
860 if (!dev && capable(CAP_SYS_MODULE))
861 request_module("%s", name);
864 static int default_rebuild_header(struct sk_buff *skb)
866 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
867 skb->dev ? skb->dev->name : "NULL!!!");
874 * Some old buggy device drivers change get_stats after registering
875 * the device. Try and trap them here.
876 * This can be elimnated when all devices are known fixed.
878 static inline int get_stats_changed(struct net_device *dev)
880 int changed = dev->last_stats != dev->get_stats;
881 dev->last_stats = dev->get_stats;
886 * dev_open - prepare an interface for use.
887 * @dev: device to open
889 * Takes a device from down to up state. The device's private open
890 * function is invoked and then the multicast lists are loaded. Finally
891 * the device is moved into the up state and a %NETDEV_UP message is
892 * sent to the netdev notifier chain.
894 * Calling this function on an active interface is a nop. On a failure
895 * a negative errno code is returned.
897 int dev_open(struct net_device *dev)
905 if (dev->flags & IFF_UP)
909 * Check for broken device drivers.
911 if (get_stats_changed(dev) && net_ratelimit()) {
912 printk(KERN_ERR "%s: driver changed get_stats after register\n",
917 * Is it even present?
919 if (!netif_device_present(dev))
923 * Call device private open method
925 set_bit(__LINK_STATE_START, &dev->state);
927 ret = dev->open(dev);
929 clear_bit(__LINK_STATE_START, &dev->state);
933 * Check for more broken device drivers.
935 if (get_stats_changed(dev) && net_ratelimit()) {
936 printk(KERN_ERR "%s: driver changed get_stats in open\n",
941 * If it went open OK then:
948 dev->flags |= IFF_UP;
951 * Initialize multicasting status
956 * Wakeup transmit queue engine
961 * ... and announce new interface.
963 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
969 * dev_close - shutdown an interface.
970 * @dev: device to shutdown
972 * This function moves an active device into down state. A
973 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
974 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
977 int dev_close(struct net_device *dev)
979 if (!(dev->flags & IFF_UP))
983 * Tell people we are going down, so that they can
984 * prepare to death, when device is still operating.
986 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
990 clear_bit(__LINK_STATE_START, &dev->state);
992 /* Synchronize to scheduled poll. We cannot touch poll list,
993 * it can be even on different cpu. So just clear netif_running(),
994 * and wait when poll really will happen. Actually, the best place
995 * for this is inside dev->stop() after device stopped its irq
996 * engine, but this requires more changes in devices. */
998 smp_mb__after_clear_bit(); /* Commit netif_running(). */
999 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
1001 current->state = TASK_INTERRUPTIBLE;
1002 schedule_timeout(1);
1006 * Call the device specific close. This cannot fail.
1007 * Only if device is UP
1009 * We allow it to be called even after a DETACH hot-plug
1016 * Device is now down.
1019 dev->flags &= ~IFF_UP;
1022 * Tell people we are down
1024 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1031 * Device change register/unregister. These are not inline or static
1032 * as we export them to the world.
1036 * register_netdevice_notifier - register a network notifier block
1039 * Register a notifier to be called when network device events occur.
1040 * The notifier passed is linked into the kernel structures and must
1041 * not be reused until it has been unregistered. A negative errno code
1042 * is returned on a failure.
1044 * When registered all registration and up events are replayed
1045 * to the new notifier to allow device to have a race free
1046 * view of the network device list.
1049 int register_netdevice_notifier(struct notifier_block *nb)
1051 struct net_device *dev;
1055 err = notifier_chain_register(&netdev_chain, nb);
1057 for (dev = dev_base; dev; dev = dev->next) {
1058 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1060 if (dev->flags & IFF_UP)
1061 nb->notifier_call(nb, NETDEV_UP, dev);
1069 * unregister_netdevice_notifier - unregister a network notifier block
1072 * Unregister a notifier previously registered by
1073 * register_netdevice_notifier(). The notifier is unlinked into the
1074 * kernel structures and may then be reused. A negative errno code
1075 * is returned on a failure.
1078 int unregister_netdevice_notifier(struct notifier_block *nb)
1080 return notifier_chain_unregister(&netdev_chain, nb);
1084 * call_netdevice_notifiers - call all network notifier blocks
1085 * @val: value passed unmodified to notifier function
1086 * @v: pointer passed unmodified to notifier function
1088 * Call all network notifier blocks. Parameters and return value
1089 * are as for notifier_call_chain().
1092 int call_netdevice_notifiers(unsigned long val, void *v)
1094 return notifier_call_chain(&netdev_chain, val, v);
1098 * Support routine. Sends outgoing frames to any network
1099 * taps currently in use.
1102 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1104 struct packet_type *ptype;
1105 net_timestamp(&skb->stamp);
1108 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1109 /* Never send packets back to the socket
1110 * they originated from - MvS (miquels@drinkel.ow.org)
1112 if ((ptype->dev == dev || !ptype->dev) &&
1113 (ptype->af_packet_priv == NULL ||
1114 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1115 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1119 /* skb->nh should be correctly
1120 set by sender, so that the second statement is
1121 just protection against buggy protocols.
1123 skb2->mac.raw = skb2->data;
1125 if (skb2->nh.raw < skb2->data ||
1126 skb2->nh.raw > skb2->tail) {
1127 if (net_ratelimit())
1128 printk(KERN_CRIT "protocol %04x is "
1130 skb2->protocol, dev->name);
1131 skb2->nh.raw = skb2->data;
1134 skb2->h.raw = skb2->nh.raw;
1135 skb2->pkt_type = PACKET_OUTGOING;
1136 ptype->func(skb2, skb->dev, ptype);
1143 * Invalidate hardware checksum when packet is to be mangled, and
1144 * complete checksum manually on outgoing path.
1146 int skb_checksum_help(struct sk_buff **pskb, int inward)
1149 int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1152 (*pskb)->ip_summed = CHECKSUM_NONE;
1156 if (skb_shared(*pskb) || skb_cloned(*pskb)) {
1157 struct sk_buff *newskb = skb_copy(*pskb, GFP_ATOMIC);
1163 skb_set_owner_w(newskb, (*pskb)->sk);
1168 if (offset > (int)(*pskb)->len)
1170 csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1172 offset = (*pskb)->tail - (*pskb)->h.raw;
1175 if ((*pskb)->csum + 2 > offset)
1178 *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1179 (*pskb)->ip_summed = CHECKSUM_NONE;
1184 #ifdef CONFIG_HIGHMEM
1185 /* Actually, we should eliminate this check as soon as we know, that:
1186 * 1. IOMMU is present and allows to map all the memory.
1187 * 2. No high memory really exists on this machine.
1190 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1194 if (dev->features & NETIF_F_HIGHDMA)
1197 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1198 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1204 #define illegal_highdma(dev, skb) (0)
1207 extern void skb_release_data(struct sk_buff *);
1209 /* Keep head the same: replace data */
1210 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1215 struct skb_shared_info *ninfo;
1216 int headerlen = skb->data - skb->head;
1217 int expand = (skb->tail + skb->data_len) - skb->end;
1219 if (skb_shared(skb))
1225 size = skb->end - skb->head + expand;
1226 size = SKB_DATA_ALIGN(size);
1227 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1231 /* Copy entire thing */
1232 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1236 ninfo = (struct skb_shared_info*)(data + size);
1237 atomic_set(&ninfo->dataref, 1);
1238 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1239 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1240 ninfo->nr_frags = 0;
1241 ninfo->frag_list = NULL;
1243 /* Offset between the two in bytes */
1244 offset = data - skb->head;
1246 /* Free old data. */
1247 skb_release_data(skb);
1250 skb->end = data + size;
1252 /* Set up new pointers */
1253 skb->h.raw += offset;
1254 skb->nh.raw += offset;
1255 skb->mac.raw += offset;
1256 skb->tail += offset;
1257 skb->data += offset;
1259 /* We are no longer a clone, even if we were. */
1261 #if defined(CONFIG_MIPS_BRCM)
1262 /* The data buffer of this skb is not pre-allocated any more
1263 * even the skb itself is pre-allocated. (songw)
1265 skb->retfreeq_data_prealloc = 0;
1268 skb->tail += skb->data_len;
1273 #define HARD_TX_LOCK_BH(dev, cpu) { \
1274 if ((dev->features & NETIF_F_LLTX) == 0) { \
1275 spin_lock_bh(&dev->xmit_lock); \
1276 dev->xmit_lock_owner = cpu; \
1280 #define HARD_TX_UNLOCK_BH(dev) { \
1281 if ((dev->features & NETIF_F_LLTX) == 0) { \
1282 dev->xmit_lock_owner = -1; \
1283 spin_unlock_bh(&dev->xmit_lock); \
1287 static inline void qdisc_run(struct net_device *dev)
1289 while (!netif_queue_stopped(dev) &&
1290 qdisc_restart(dev)<0)
1295 * dev_queue_xmit - transmit a buffer
1296 * @skb: buffer to transmit
1298 * Queue a buffer for transmission to a network device. The caller must
1299 * have set the device and priority and built the buffer before calling
1300 * this function. The function can be called from an interrupt.
1302 * A negative errno code is returned on a failure. A success does not
1303 * guarantee the frame will be transmitted as it may be dropped due
1304 * to congestion or traffic shaping.
1307 int dev_queue_xmit(struct sk_buff *skb)
1309 struct net_device *dev = skb->dev;
1313 if (skb_shinfo(skb)->frag_list &&
1314 !(dev->features & NETIF_F_FRAGLIST) &&
1315 __skb_linearize(skb, GFP_ATOMIC))
1318 /* Fragmented skb is linearized if device does not support SG,
1319 * or if at least one of fragments is in highmem and device
1320 * does not support DMA from it.
1322 if (skb_shinfo(skb)->nr_frags &&
1323 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1324 __skb_linearize(skb, GFP_ATOMIC))
1327 /* If packet is not checksummed and device does not support
1328 * checksumming for this protocol, complete checksumming here.
1330 if (skb->ip_summed == CHECKSUM_HW &&
1331 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1332 (!(dev->features & NETIF_F_IP_CSUM) ||
1333 skb->protocol != htons(ETH_P_IP))))
1334 if (skb_checksum_help(&skb, 0))
1338 /* Updates of qdisc are serialized by queue_lock.
1339 * The struct Qdisc which is pointed to by qdisc is now a
1340 * rcu structure - it may be accessed without acquiring
1341 * a lock (but the structure may be stale.) The freeing of the
1342 * qdisc will be deferred until it's known that there are no
1343 * more references to it.
1345 * If the qdisc has an enqueue function, we still need to
1346 * hold the queue_lock before calling it, since queue_lock
1347 * also serializes access to the device queue.
1351 smp_read_barrier_depends();
1352 #ifdef CONFIG_NET_CLS_ACT
1353 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1356 /* Grab device queue */
1357 spin_lock_bh(&dev->queue_lock);
1359 rc = q->enqueue(skb, q);
1363 spin_unlock_bh(&dev->queue_lock);
1365 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1370 /* The device has no queue. Common case for software devices:
1371 loopback, all the sorts of tunnels...
1373 Really, it is unlikely that xmit_lock protection is necessary here.
1374 (f.e. loopback and IP tunnels are clean ignoring statistics
1376 However, it is possible, that they rely on protection
1379 Check this and shot the lock. It is not prone from deadlocks.
1380 Either shot noqueue qdisc, it is even simpler 8)
1382 if (dev->flags & IFF_UP) {
1383 int cpu = get_cpu();
1385 if (dev->xmit_lock_owner != cpu) {
1387 HARD_TX_LOCK_BH(dev, cpu);
1390 if (!netif_queue_stopped(dev)) {
1392 dev_queue_xmit_nit(skb, dev);
1395 if (!dev->hard_start_xmit(skb, dev)) {
1396 HARD_TX_UNLOCK_BH(dev);
1400 HARD_TX_UNLOCK_BH(dev);
1401 if (net_ratelimit())
1402 printk(KERN_CRIT "Virtual device %s asks to "
1403 "queue packet!\n", dev->name);
1407 /* Recursion is detected! It is possible,
1409 if (net_ratelimit())
1410 printk(KERN_CRIT "Dead loop on virtual device "
1411 "%s, fix it urgently!\n", dev->name);
1423 /*=======================================================================
1425 =======================================================================*/
1427 int netdev_max_backlog = 300;
1428 int weight_p = 64; /* old backlog weight */
1429 /* These numbers are selected based on intuition and some
1430 * experimentatiom, if you have more scientific way of doing this
1431 * please go ahead and fix things.
1433 int no_cong_thresh = 10;
1438 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1441 #ifdef CONFIG_NET_HW_FLOWCONTROL
1442 atomic_t netdev_dropping = ATOMIC_INIT(0);
1443 static unsigned long netdev_fc_mask = 1;
1444 unsigned long netdev_fc_xoff;
1445 spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1449 void (*stimul)(struct net_device *);
1450 struct net_device *dev;
1451 } netdev_fc_slots[BITS_PER_LONG];
1453 int netdev_register_fc(struct net_device *dev,
1454 void (*stimul)(struct net_device *dev))
1457 unsigned long flags;
1459 spin_lock_irqsave(&netdev_fc_lock, flags);
1460 if (netdev_fc_mask != ~0UL) {
1461 bit = ffz(netdev_fc_mask);
1462 netdev_fc_slots[bit].stimul = stimul;
1463 netdev_fc_slots[bit].dev = dev;
1464 set_bit(bit, &netdev_fc_mask);
1465 clear_bit(bit, &netdev_fc_xoff);
1467 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1471 void netdev_unregister_fc(int bit)
1473 unsigned long flags;
1475 spin_lock_irqsave(&netdev_fc_lock, flags);
1477 netdev_fc_slots[bit].stimul = NULL;
1478 netdev_fc_slots[bit].dev = NULL;
1479 clear_bit(bit, &netdev_fc_mask);
1480 clear_bit(bit, &netdev_fc_xoff);
1482 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1485 static void netdev_wakeup(void)
1489 spin_lock(&netdev_fc_lock);
1490 xoff = netdev_fc_xoff;
1495 netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1497 spin_unlock(&netdev_fc_lock);
1501 static void get_sample_stats(int cpu)
1507 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1508 int blog = sd->input_pkt_queue.qlen;
1509 int avg_blog = sd->avg_blog;
1511 avg_blog = (avg_blog >> 1) + (blog >> 1);
1513 if (avg_blog > mod_cong) {
1514 /* Above moderate congestion levels. */
1515 sd->cng_level = NET_RX_CN_HIGH;
1518 rq = rd % netdev_max_backlog;
1519 if (rq < avg_blog) /* unlucky bastard */
1520 sd->cng_level = NET_RX_DROP;
1522 } else if (avg_blog > lo_cong) {
1523 sd->cng_level = NET_RX_CN_MOD;
1526 rq = rd % netdev_max_backlog;
1527 if (rq < avg_blog) /* unlucky bastard */
1528 sd->cng_level = NET_RX_CN_HIGH;
1530 } else if (avg_blog > no_cong)
1531 sd->cng_level = NET_RX_CN_LOW;
1532 else /* no congestion */
1533 sd->cng_level = NET_RX_SUCCESS;
1535 sd->avg_blog = avg_blog;
1538 #ifdef OFFLINE_SAMPLE
1539 static void sample_queue(unsigned long dummy)
1541 /* 10 ms 0r 1ms -- i don't care -- JHS */
1543 int cpu = smp_processor_id();
1545 get_sample_stats(cpu);
1546 next_tick += jiffies;
1547 mod_timer(&samp_timer, next_tick);
1553 * netif_rx - post buffer to the network code
1554 * @skb: buffer to post
1556 * This function receives a packet from a device driver and queues it for
1557 * the upper (protocol) levels to process. It always succeeds. The buffer
1558 * may be dropped during processing for congestion control or by the
1562 * NET_RX_SUCCESS (no congestion)
1563 * NET_RX_CN_LOW (low congestion)
1564 * NET_RX_CN_MOD (moderate congestion)
1565 * NET_RX_CN_HIGH (high congestion)
1566 * NET_RX_DROP (packet was dropped)
1570 int netif_rx(struct sk_buff *skb)
1573 struct softnet_data *queue;
1574 unsigned long flags;
1576 #ifdef CONFIG_NETPOLL_RX
1577 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1583 if (!skb->stamp.tv_sec)
1584 net_timestamp(&skb->stamp);
1587 * The code is rearranged so that the path is the most
1588 * short when CPU is congested, but is still operating.
1590 local_irq_save(flags);
1591 this_cpu = smp_processor_id();
1592 queue = &__get_cpu_var(softnet_data);
1594 __get_cpu_var(netdev_rx_stat).total++;
1595 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1596 if (queue->input_pkt_queue.qlen) {
1597 if (queue->throttle)
1602 __skb_queue_tail(&queue->input_pkt_queue, skb);
1603 #ifndef OFFLINE_SAMPLE
1604 get_sample_stats(this_cpu);
1606 local_irq_restore(flags);
1607 return queue->cng_level;
1610 if (queue->throttle) {
1611 queue->throttle = 0;
1612 #ifdef CONFIG_NET_HW_FLOWCONTROL
1613 if (atomic_dec_and_test(&netdev_dropping))
1618 netif_rx_schedule(&queue->backlog_dev);
1622 if (!queue->throttle) {
1623 queue->throttle = 1;
1624 __get_cpu_var(netdev_rx_stat).throttled++;
1625 #ifdef CONFIG_NET_HW_FLOWCONTROL
1626 atomic_inc(&netdev_dropping);
1631 __get_cpu_var(netdev_rx_stat).dropped++;
1632 local_irq_restore(flags);
1638 static __inline__ void skb_bond(struct sk_buff *skb)
1640 struct net_device *dev = skb->dev;
1643 skb->real_dev = skb->dev;
1644 skb->dev = dev->master;
1648 static void net_tx_action(struct softirq_action *h)
1650 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1652 if (sd->completion_queue) {
1653 struct sk_buff *clist;
1655 local_irq_disable();
1656 clist = sd->completion_queue;
1657 sd->completion_queue = NULL;
1661 struct sk_buff *skb = clist;
1662 clist = clist->next;
1664 BUG_TRAP(!atomic_read(&skb->users));
1669 if (sd->output_queue) {
1670 struct net_device *head;
1672 local_irq_disable();
1673 head = sd->output_queue;
1674 sd->output_queue = NULL;
1678 struct net_device *dev = head;
1679 head = head->next_sched;
1681 smp_mb__before_clear_bit();
1682 clear_bit(__LINK_STATE_SCHED, &dev->state);
1684 if (spin_trylock(&dev->queue_lock)) {
1686 spin_unlock(&dev->queue_lock);
1688 netif_schedule(dev);
1694 static __inline__ int deliver_skb(struct sk_buff *skb,
1695 struct packet_type *pt_prev, int last)
1697 atomic_inc(&skb->users);
1698 return pt_prev->func(skb, skb->dev, pt_prev);
1702 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1703 int (*br_handle_frame_hook)(struct sk_buff *skb);
1705 static __inline__ int handle_bridge(struct sk_buff *skb,
1706 struct packet_type *pt_prev)
1708 int ret = NET_RX_DROP;
1710 ret = deliver_skb(skb, pt_prev, 0);
1717 static inline int __handle_bridge(struct sk_buff *skb,
1718 struct packet_type **pt_prev, int *ret)
1720 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
1721 #if defined(CONFIG_MIPS_BRCM)
1722 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1723 if (skb->protocol == __constant_htons(ETH_P_PPP_SES)||
1724 skb->protocol == __constant_htons(ETH_P_PPP_DISC)) {
1725 if (!memcmp(skb->mac.ethernet->h_dest,skb->dev->dev_addr,ETH_ALEN))
1728 *ret = handle_bridge(skb, *pt_prev);
1729 if (br_handle_frame_hook(skb) == 0)
1736 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1737 *ret = handle_bridge(skb, *pt_prev);
1738 if (br_handle_frame_hook(skb) == 0)
1749 #ifdef CONFIG_NET_CLS_ACT
1750 /* TODO: Maybe we should just force sch_ingress to be compiled in
1751 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1752 * a compare and 2 stores extra right now if we dont have it on
1753 * but have CONFIG_NET_CLS_ACT
1754 * NOTE: This doesnt stop any functionality; if you dont have
1755 * the ingress scheduler, you just cant add policies on ingress.
1758 int ing_filter(struct sk_buff *skb)
1761 struct net_device *dev = skb->dev;
1762 int result = TC_ACT_OK;
1764 if (dev->qdisc_ingress) {
1765 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1766 if (MAX_RED_LOOP < ttl++) {
1767 printk("Redir loop detected Dropping packet (%s->%s)\n",
1768 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1772 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1774 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1775 if (NULL == skb->input_dev) {
1776 skb->input_dev = skb->dev;
1777 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1779 spin_lock(&dev->ingress_lock);
1780 if ((q = dev->qdisc_ingress) != NULL)
1781 result = q->enqueue(skb, q);
1782 spin_unlock(&dev->ingress_lock);
1790 int netif_receive_skb(struct sk_buff *skb)
1792 struct packet_type *ptype, *pt_prev;
1793 int ret = NET_RX_DROP;
1794 unsigned short type;
1796 #ifdef CONFIG_NETPOLL_RX
1797 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1803 if (!skb->stamp.tv_sec)
1804 net_timestamp(&skb->stamp);
1808 __get_cpu_var(netdev_rx_stat).total++;
1810 skb->h.raw = skb->nh.raw = skb->data;
1811 skb->mac_len = skb->nh.raw - skb->mac.raw;
1814 #ifdef CONFIG_NET_CLS_ACT
1815 if (skb->tc_verd & TC_NCLS) {
1816 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1823 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1824 if (!ptype->dev || ptype->dev == skb->dev) {
1826 ret = deliver_skb(skb, pt_prev, 0);
1831 #ifdef CONFIG_NET_CLS_ACT
1833 atomic_inc(&skb->users);
1834 ret = pt_prev->func(skb, skb->dev, pt_prev);
1835 pt_prev = NULL; /* noone else should process this after*/
1837 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1840 ret = ing_filter(skb);
1842 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1851 handle_diverter(skb);
1853 if (__handle_bridge(skb, &pt_prev, &ret))
1856 type = skb->protocol;
1857 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1858 if (ptype->type == type &&
1859 (!ptype->dev || ptype->dev == skb->dev)) {
1861 ret = deliver_skb(skb, pt_prev, 0);
1867 ret = pt_prev->func(skb, skb->dev, pt_prev);
1870 /* Jamal, now you will not able to escape explaining
1871 * me how you were going to use this. :-)
1881 static int process_backlog(struct net_device *backlog_dev, int *budget)
1884 int quota = min(backlog_dev->quota, *budget);
1885 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1886 unsigned long start_time = jiffies;
1889 struct sk_buff *skb;
1890 struct net_device *dev;
1892 local_irq_disable();
1893 skb = __skb_dequeue(&queue->input_pkt_queue);
1900 netif_receive_skb(skb);
1906 if (work >= quota || jiffies - start_time > 1)
1909 #ifdef CONFIG_NET_HW_FLOWCONTROL
1910 if (queue->throttle &&
1911 queue->input_pkt_queue.qlen < no_cong_thresh ) {
1912 queue->throttle = 0;
1913 if (atomic_dec_and_test(&netdev_dropping)) {
1921 backlog_dev->quota -= work;
1926 backlog_dev->quota -= work;
1929 list_del(&backlog_dev->poll_list);
1930 smp_mb__before_clear_bit();
1931 netif_poll_enable(backlog_dev);
1933 if (queue->throttle) {
1934 queue->throttle = 0;
1935 #ifdef CONFIG_NET_HW_FLOWCONTROL
1936 if (atomic_dec_and_test(&netdev_dropping))
1944 static void net_rx_action(struct softirq_action *h)
1946 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1947 unsigned long start_time = jiffies;
1948 int budget = netdev_max_backlog;
1951 local_irq_disable();
1953 while (!list_empty(&queue->poll_list)) {
1954 struct net_device *dev;
1956 if (budget <= 0 || jiffies - start_time > 1)
1961 dev = list_entry(queue->poll_list.next,
1962 struct net_device, poll_list);
1964 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1965 local_irq_disable();
1966 list_del(&dev->poll_list);
1967 list_add_tail(&dev->poll_list, &queue->poll_list);
1969 dev->quota += dev->weight;
1971 dev->quota = dev->weight;
1974 local_irq_disable();
1982 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1983 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1987 static gifconf_func_t * gifconf_list [NPROTO];
1990 * register_gifconf - register a SIOCGIF handler
1991 * @family: Address family
1992 * @gifconf: Function handler
1994 * Register protocol dependent address dumping routines. The handler
1995 * that is passed must not be freed or reused until it has been replaced
1996 * by another handler.
1998 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2000 if (family >= NPROTO)
2002 gifconf_list[family] = gifconf;
2008 * Map an interface index to its name (SIOCGIFNAME)
2012 * We need this ioctl for efficient implementation of the
2013 * if_indextoname() function required by the IPv6 API. Without
2014 * it, we would have to search all the interfaces to find a
2018 static int dev_ifname(struct ifreq __user *arg)
2020 struct net_device *dev;
2024 * Fetch the caller's info block.
2027 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2030 read_lock(&dev_base_lock);
2031 dev = __dev_get_by_index(ifr.ifr_ifindex);
2033 read_unlock(&dev_base_lock);
2037 strcpy(ifr.ifr_name, dev->name);
2038 read_unlock(&dev_base_lock);
2040 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2046 * Perform a SIOCGIFCONF call. This structure will change
2047 * size eventually, and there is nothing I can do about it.
2048 * Thus we will need a 'compatibility mode'.
2051 static int dev_ifconf(char __user *arg)
2054 struct net_device *dev;
2061 * Fetch the caller's info block.
2064 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2071 * Loop over the interfaces, and write an info block for each.
2075 for (dev = dev_base; dev; dev = dev->next) {
2076 for (i = 0; i < NPROTO; i++) {
2077 if (gifconf_list[i]) {
2080 done = gifconf_list[i](dev, NULL, 0);
2082 done = gifconf_list[i](dev, pos + total,
2092 * All done. Write the updated control block back to the caller.
2094 ifc.ifc_len = total;
2097 * Both BSD and Solaris return 0 here, so we do too.
2099 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2102 #ifdef CONFIG_PROC_FS
2104 * This is invoked by the /proc filesystem handler to display a device
2107 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2109 struct net_device *dev;
2112 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2114 return i == pos ? dev : NULL;
2117 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2119 read_lock(&dev_base_lock);
2120 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2123 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2126 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2129 void dev_seq_stop(struct seq_file *seq, void *v)
2131 read_unlock(&dev_base_lock);
2134 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2136 if (dev->get_stats) {
2137 struct net_device_stats *stats = dev->get_stats(dev);
2139 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2140 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2141 dev->name, stats->rx_bytes, stats->rx_packets,
2143 stats->rx_dropped + stats->rx_missed_errors,
2144 stats->rx_fifo_errors,
2145 stats->rx_length_errors + stats->rx_over_errors +
2146 stats->rx_crc_errors + stats->rx_frame_errors,
2147 stats->rx_compressed, stats->multicast,
2148 stats->tx_bytes, stats->tx_packets,
2149 stats->tx_errors, stats->tx_dropped,
2150 stats->tx_fifo_errors, stats->collisions,
2151 stats->tx_carrier_errors +
2152 stats->tx_aborted_errors +
2153 stats->tx_window_errors +
2154 stats->tx_heartbeat_errors,
2155 stats->tx_compressed);
2157 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2161 * Called from the PROCfs module. This now uses the new arbitrary sized
2162 * /proc/net interface to create /proc/net/dev
2164 static int dev_seq_show(struct seq_file *seq, void *v)
2166 if (v == SEQ_START_TOKEN)
2167 seq_puts(seq, "Inter-| Receive "
2169 " face |bytes packets errs drop fifo frame "
2170 "compressed multicast|bytes packets errs "
2171 "drop fifo colls carrier compressed\n");
2173 dev_seq_printf_stats(seq, v);
2177 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2179 struct netif_rx_stats *rc = NULL;
2181 while (*pos < NR_CPUS)
2182 if (cpu_online(*pos)) {
2183 rc = &per_cpu(netdev_rx_stat, *pos);
2190 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2192 return softnet_get_online(pos);
2195 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2198 return softnet_get_online(pos);
2201 static void softnet_seq_stop(struct seq_file *seq, void *v)
2205 static int softnet_seq_show(struct seq_file *seq, void *v)
2207 struct netif_rx_stats *s = v;
2209 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2210 s->total, s->dropped, s->time_squeeze, s->throttled,
2211 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2212 s->fastroute_deferred_out,
2214 s->fastroute_latency_reduction
2222 static struct seq_operations dev_seq_ops = {
2223 .start = dev_seq_start,
2224 .next = dev_seq_next,
2225 .stop = dev_seq_stop,
2226 .show = dev_seq_show,
2229 static int dev_seq_open(struct inode *inode, struct file *file)
2231 return seq_open(file, &dev_seq_ops);
2234 static struct file_operations dev_seq_fops = {
2235 .owner = THIS_MODULE,
2236 .open = dev_seq_open,
2238 .llseek = seq_lseek,
2239 .release = seq_release,
2242 static struct seq_operations softnet_seq_ops = {
2243 .start = softnet_seq_start,
2244 .next = softnet_seq_next,
2245 .stop = softnet_seq_stop,
2246 .show = softnet_seq_show,
2249 static int softnet_seq_open(struct inode *inode, struct file *file)
2251 return seq_open(file, &softnet_seq_ops);
2254 static struct file_operations softnet_seq_fops = {
2255 .owner = THIS_MODULE,
2256 .open = softnet_seq_open,
2258 .llseek = seq_lseek,
2259 .release = seq_release,
2263 extern int wireless_proc_init(void);
2265 #define wireless_proc_init() 0
2268 static int __init dev_proc_init(void)
2272 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2274 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2276 if (wireless_proc_init())
2282 proc_net_remove("softnet_stat");
2284 proc_net_remove("dev");
2288 #define dev_proc_init() 0
2289 #endif /* CONFIG_PROC_FS */
2293 * netdev_set_master - set up master/slave pair
2294 * @slave: slave device
2295 * @master: new master device
2297 * Changes the master device of the slave. Pass %NULL to break the
2298 * bonding. The caller must hold the RTNL semaphore. On a failure
2299 * a negative errno code is returned. On success the reference counts
2300 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2301 * function returns zero.
2303 int netdev_set_master(struct net_device *slave, struct net_device *master)
2305 struct net_device *old = slave->master;
2315 slave->master = master;
2323 slave->flags |= IFF_SLAVE;
2325 slave->flags &= ~IFF_SLAVE;
2327 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2332 * dev_set_promiscuity - update promiscuity count on a device
2336 * Add or remove promsicuity from a device. While the count in the device
2337 * remains above zero the interface remains promiscuous. Once it hits zero
2338 * the device reverts back to normal filtering operation. A negative inc
2339 * value is used to drop promiscuity on the device.
2341 void dev_set_promiscuity(struct net_device *dev, int inc)
2343 unsigned short old_flags = dev->flags;
2345 dev->flags |= IFF_PROMISC;
2346 if ((dev->promiscuity += inc) == 0)
2347 dev->flags &= ~IFF_PROMISC;
2348 if (dev->flags ^ old_flags) {
2350 printk(KERN_INFO "device %s %s promiscuous mode\n",
2351 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2357 * dev_set_allmulti - update allmulti count on a device
2361 * Add or remove reception of all multicast frames to a device. While the
2362 * count in the device remains above zero the interface remains listening
2363 * to all interfaces. Once it hits zero the device reverts back to normal
2364 * filtering operation. A negative @inc value is used to drop the counter
2365 * when releasing a resource needing all multicasts.
2368 void dev_set_allmulti(struct net_device *dev, int inc)
2370 unsigned short old_flags = dev->flags;
2372 dev->flags |= IFF_ALLMULTI;
2373 if ((dev->allmulti += inc) == 0)
2374 dev->flags &= ~IFF_ALLMULTI;
2375 if (dev->flags ^ old_flags)
2379 unsigned dev_get_flags(const struct net_device *dev)
2383 flags = (dev->flags & ~(IFF_PROMISC |
2386 (dev->gflags & (IFF_PROMISC |
2389 if (netif_running(dev) && netif_carrier_ok(dev))
2390 flags |= IFF_RUNNING;
2395 int dev_change_flags(struct net_device *dev, unsigned flags)
2398 int old_flags = dev->flags;
2401 * Set the flags on our device.
2404 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2405 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2407 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2411 * Load in the correct multicast list now the flags have changed.
2417 * Have we downed the interface. We handle IFF_UP ourselves
2418 * according to user attempts to set it, rather than blindly
2423 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2424 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2430 if (dev->flags & IFF_UP &&
2431 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2433 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2435 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2436 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2437 dev->gflags ^= IFF_PROMISC;
2438 dev_set_promiscuity(dev, inc);
2441 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2442 is important. Some (broken) drivers set IFF_PROMISC, when
2443 IFF_ALLMULTI is requested not asking us and not reporting.
2445 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2446 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2447 dev->gflags ^= IFF_ALLMULTI;
2448 dev_set_allmulti(dev, inc);
2451 if (old_flags ^ dev->flags)
2452 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2457 int dev_set_mtu(struct net_device *dev, int new_mtu)
2461 if (new_mtu == dev->mtu)
2464 /* MTU must be positive. */
2468 if (!netif_device_present(dev))
2472 if (dev->change_mtu)
2473 err = dev->change_mtu(dev, new_mtu);
2476 if (!err && dev->flags & IFF_UP)
2477 notifier_call_chain(&netdev_chain,
2478 NETDEV_CHANGEMTU, dev);
2484 * Perform the SIOCxIFxxx calls.
2486 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2489 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2495 case SIOCGIFFLAGS: /* Get interface flags */
2496 ifr->ifr_flags = dev_get_flags(dev);
2499 case SIOCSIFFLAGS: /* Set interface flags */
2500 return dev_change_flags(dev, ifr->ifr_flags);
2502 case SIOCGIFMETRIC: /* Get the metric on the interface
2503 (currently unused) */
2504 ifr->ifr_metric = 0;
2507 case SIOCSIFMETRIC: /* Set the metric on the interface
2508 (currently unused) */
2511 #if defined(CONFIG_MIPS_BRCM)
2512 case SIOCGIFTRANSSTART:
2513 ifr->ifr_ifru.ifru_ivalue = dev->trans_start;
2517 case SIOCGIFMTU: /* Get the MTU of a device */
2518 ifr->ifr_mtu = dev->mtu;
2521 #if defined(CONFIG_MIPS_BRCM)
2522 case SIOCCIFSTATS: /* Clean up the Stats of a device */
2524 struct net_device_stats * pStats = dev->get_stats(dev);
2525 memset(pStats, 0, sizeof(struct net_device_stats));
2530 case SIOCSIFMTU: /* Set the MTU of a device */
2531 return dev_set_mtu(dev, ifr->ifr_mtu);
2534 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2535 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2536 ifr->ifr_hwaddr.sa_family = dev->type;
2540 if (!dev->set_mac_address)
2542 if (ifr->ifr_hwaddr.sa_family != dev->type)
2544 if (!netif_device_present(dev))
2546 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2548 notifier_call_chain(&netdev_chain,
2549 NETDEV_CHANGEADDR, dev);
2552 case SIOCSIFHWBROADCAST:
2553 if (ifr->ifr_hwaddr.sa_family != dev->type)
2555 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2556 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2557 notifier_call_chain(&netdev_chain,
2558 NETDEV_CHANGEADDR, dev);
2562 ifr->ifr_map.mem_start = dev->mem_start;
2563 ifr->ifr_map.mem_end = dev->mem_end;
2564 ifr->ifr_map.base_addr = dev->base_addr;
2565 ifr->ifr_map.irq = dev->irq;
2566 ifr->ifr_map.dma = dev->dma;
2567 ifr->ifr_map.port = dev->if_port;
2571 if (dev->set_config) {
2572 if (!netif_device_present(dev))
2574 return dev->set_config(dev, &ifr->ifr_map);
2579 if (!dev->set_multicast_list ||
2580 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2582 if (!netif_device_present(dev))
2584 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2588 if (!dev->set_multicast_list ||
2589 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2591 if (!netif_device_present(dev))
2593 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2597 ifr->ifr_ifindex = dev->ifindex;
2601 ifr->ifr_qlen = dev->tx_queue_len;
2605 if (ifr->ifr_qlen < 0)
2607 dev->tx_queue_len = ifr->ifr_qlen;
2611 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2612 return dev_change_name(dev, ifr->ifr_newname);
2615 * Unknown or private ioctl
2619 if ((cmd >= SIOCDEVPRIVATE &&
2620 cmd <= SIOCDEVPRIVATE + 15) ||
2621 cmd == SIOCBONDENSLAVE ||
2622 cmd == SIOCBONDRELEASE ||
2623 cmd == SIOCBONDSETHWADDR ||
2624 cmd == SIOCBONDSLAVEINFOQUERY ||
2625 cmd == SIOCBONDINFOQUERY ||
2626 cmd == SIOCBONDCHANGEACTIVE ||
2627 cmd == SIOCGMIIPHY ||
2628 cmd == SIOCGMIIREG ||
2629 cmd == SIOCSMIIREG ||
2630 cmd == SIOCBRADDIF ||
2631 cmd == SIOCBRDELIF ||
2632 cmd == SIOCWANDEV) {
2634 if (dev->do_ioctl) {
2635 if (netif_device_present(dev))
2636 err = dev->do_ioctl(dev, ifr,
2649 * This function handles all "interface"-type I/O control requests. The actual
2650 * 'doing' part of this is dev_ifsioc above.
2654 * dev_ioctl - network device ioctl
2655 * @cmd: command to issue
2656 * @arg: pointer to a struct ifreq in user space
2658 * Issue ioctl functions to devices. This is normally called by the
2659 * user space syscall interfaces but can sometimes be useful for
2660 * other purposes. The return value is the return from the syscall if
2661 * positive or a negative errno code on error.
2664 int dev_ioctl(unsigned int cmd, void __user *arg)
2670 /* One special case: SIOCGIFCONF takes ifconf argument
2671 and requires shared lock, because it sleeps writing
2675 if (cmd == SIOCGIFCONF) {
2677 ret = dev_ifconf((char __user *) arg);
2681 if (cmd == SIOCGIFNAME)
2682 return dev_ifname((struct ifreq __user *)arg);
2684 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2687 ifr.ifr_name[IFNAMSIZ-1] = 0;
2689 colon = strchr(ifr.ifr_name, ':');
2694 * See which interface the caller is talking about.
2699 * These ioctl calls:
2700 * - can be done by all.
2701 * - atomic and do not require locking.
2712 dev_load(ifr.ifr_name);
2713 read_lock(&dev_base_lock);
2714 ret = dev_ifsioc(&ifr, cmd);
2715 read_unlock(&dev_base_lock);
2719 if (copy_to_user(arg, &ifr,
2720 sizeof(struct ifreq)))
2726 dev_load(ifr.ifr_name);
2728 ret = dev_ethtool(&ifr);
2733 if (copy_to_user(arg, &ifr,
2734 sizeof(struct ifreq)))
2740 * These ioctl calls:
2741 * - require superuser power.
2742 * - require strict serialization.
2748 if (!capable(CAP_NET_ADMIN))
2750 dev_load(ifr.ifr_name);
2752 ret = dev_ifsioc(&ifr, cmd);
2757 if (copy_to_user(arg, &ifr,
2758 sizeof(struct ifreq)))
2764 * These ioctl calls:
2765 * - require superuser power.
2766 * - require strict serialization.
2767 * - do not return a value
2777 case SIOCSIFHWBROADCAST:
2780 case SIOCBONDENSLAVE:
2781 case SIOCBONDRELEASE:
2782 case SIOCBONDSETHWADDR:
2783 case SIOCBONDSLAVEINFOQUERY:
2784 case SIOCBONDINFOQUERY:
2785 case SIOCBONDCHANGEACTIVE:
2788 if (!capable(CAP_NET_ADMIN))
2790 dev_load(ifr.ifr_name);
2792 ret = dev_ifsioc(&ifr, cmd);
2797 /* Get the per device memory space. We can add this but
2798 * currently do not support it */
2800 /* Set the per device memory buffer space.
2801 * Not applicable in our case */
2806 * Unknown or private ioctl.
2809 if (cmd == SIOCWANDEV ||
2810 (cmd >= SIOCDEVPRIVATE &&
2811 cmd <= SIOCDEVPRIVATE + 15)) {
2812 dev_load(ifr.ifr_name);
2814 ret = dev_ifsioc(&ifr, cmd);
2816 if (!ret && copy_to_user(arg, &ifr,
2817 sizeof(struct ifreq)))
2822 /* Take care of Wireless Extensions */
2823 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2824 /* If command is `set a parameter', or
2825 * `get the encoding parameters', check if
2826 * the user has the right to do it */
2827 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2828 if (!capable(CAP_NET_ADMIN))
2831 dev_load(ifr.ifr_name);
2833 /* Follow me in net/core/wireless.c */
2834 ret = wireless_process_ioctl(&ifr, cmd);
2836 if (!ret && IW_IS_GET(cmd) &&
2837 copy_to_user(arg, &ifr,
2838 sizeof(struct ifreq)))
2842 #endif /* WIRELESS_EXT */
2849 * dev_new_index - allocate an ifindex
2851 * Returns a suitable unique value for a new device interface
2852 * number. The caller must hold the rtnl semaphore or the
2853 * dev_base_lock to be sure it remains unique.
2855 int dev_new_index(void)
2861 if (!__dev_get_by_index(ifindex))
2866 static int dev_boot_phase = 1;
2868 /* Delayed registration/unregisteration */
2869 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2870 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2872 static inline void net_set_todo(struct net_device *dev)
2874 spin_lock(&net_todo_list_lock);
2875 list_add_tail(&dev->todo_list, &net_todo_list);
2876 spin_unlock(&net_todo_list_lock);
2880 * register_netdevice - register a network device
2881 * @dev: device to register
2883 * Take a completed network device structure and add it to the kernel
2884 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2885 * chain. 0 is returned on success. A negative errno code is returned
2886 * on a failure to set up the device, or if the name is a duplicate.
2888 * Callers must hold the rtnl semaphore. See the comment at the
2889 * end of Space.c for details about the locking. You may want
2890 * register_netdev() instead of this.
2893 * The locking appears insufficient to guarantee two parallel registers
2894 * will not get the same name.
2897 int register_netdevice(struct net_device *dev)
2899 struct hlist_head *head;
2900 struct hlist_node *p;
2903 BUG_ON(dev_boot_phase);
2906 /* When net_device's are persistent, this will be fatal. */
2907 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2909 spin_lock_init(&dev->queue_lock);
2910 spin_lock_init(&dev->xmit_lock);
2911 dev->xmit_lock_owner = -1;
2912 #ifdef CONFIG_NET_CLS_ACT
2913 spin_lock_init(&dev->ingress_lock);
2916 ret = alloc_divert_blk(dev);
2922 #if defined(CONFIG_MIPS_BRCM)
2923 /* Init dev->last_stats in case CONFIG_SYSFS is disabled */
2924 dev->last_stats = dev->get_stats;
2927 /* Init, if this function is available */
2929 ret = dev->init(dev);
2937 if (!dev_valid_name(dev->name)) {
2942 dev->ifindex = dev_new_index();
2943 if (dev->iflink == -1)
2944 dev->iflink = dev->ifindex;
2946 /* Check for existence of name */
2947 head = dev_name_hash(dev->name);
2948 hlist_for_each(p, head) {
2949 struct net_device *d
2950 = hlist_entry(p, struct net_device, name_hlist);
2951 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2957 /* Fix illegal SG+CSUM combinations. */
2958 if ((dev->features & NETIF_F_SG) &&
2959 !(dev->features & (NETIF_F_IP_CSUM |
2961 NETIF_F_HW_CSUM))) {
2962 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2964 dev->features &= ~NETIF_F_SG;
2968 * nil rebuild_header routine,
2969 * that should be never called and used as just bug trap.
2972 if (!dev->rebuild_header)
2973 dev->rebuild_header = default_rebuild_header;
2976 * Default initial state at registry is that the
2977 * device is present.
2980 set_bit(__LINK_STATE_PRESENT, &dev->state);
2983 dev_init_scheduler(dev);
2984 write_lock_bh(&dev_base_lock);
2986 dev_tail = &dev->next;
2987 hlist_add_head(&dev->name_hlist, head);
2988 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2990 dev->reg_state = NETREG_REGISTERING;
2991 write_unlock_bh(&dev_base_lock);
2993 /* Notify protocols, that a new device appeared. */
2994 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2996 /* Finish registration after unlock */
3003 free_divert_blk(dev);
3008 * netdev_wait_allrefs - wait until all references are gone.
3010 * This is called when unregistering network devices.
3012 * Any protocol or device that holds a reference should register
3013 * for netdevice notification, and cleanup and put back the
3014 * reference if they receive an UNREGISTER event.
3015 * We can get stuck here if buggy protocols don't correctly
3018 static void netdev_wait_allrefs(struct net_device *dev)
3020 unsigned long rebroadcast_time, warning_time;
3022 rebroadcast_time = warning_time = jiffies;
3023 while (atomic_read(&dev->refcnt) != 0) {
3024 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3027 /* Rebroadcast unregister notification */
3028 notifier_call_chain(&netdev_chain,
3029 NETDEV_UNREGISTER, dev);
3031 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3033 /* We must not have linkwatch events
3034 * pending on unregister. If this
3035 * happens, we simply run the queue
3036 * unscheduled, resulting in a noop
3039 linkwatch_run_queue();
3044 rebroadcast_time = jiffies;
3047 current->state = TASK_INTERRUPTIBLE;
3048 schedule_timeout(HZ / 4);
3050 if (time_after(jiffies, warning_time + 10 * HZ)) {
3051 printk(KERN_EMERG "unregister_netdevice: "
3052 "waiting for %s to become free. Usage "
3054 dev->name, atomic_read(&dev->refcnt));
3055 warning_time = jiffies;
3064 * register_netdevice(x1);
3065 * register_netdevice(x2);
3067 * unregister_netdevice(y1);
3068 * unregister_netdevice(y2);
3074 * We are invoked by rtnl_unlock() after it drops the semaphore.
3075 * This allows us to deal with problems:
3076 * 1) We can create/delete sysfs objects which invoke hotplug
3077 * without deadlocking with linkwatch via keventd.
3078 * 2) Since we run with the RTNL semaphore not held, we can sleep
3079 * safely in order to wait for the netdev refcnt to drop to zero.
3081 static DECLARE_MUTEX(net_todo_run_mutex);
3082 void netdev_run_todo(void)
3084 struct list_head list = LIST_HEAD_INIT(list);
3088 /* Need to guard against multiple cpu's getting out of order. */
3089 down(&net_todo_run_mutex);
3091 /* Not safe to do outside the semaphore. We must not return
3092 * until all unregister events invoked by the local processor
3093 * have been completed (either by this todo run, or one on
3096 if (list_empty(&net_todo_list))
3099 /* Snapshot list, allow later requests */
3100 spin_lock(&net_todo_list_lock);
3101 list_splice_init(&net_todo_list, &list);
3102 spin_unlock(&net_todo_list_lock);
3104 while (!list_empty(&list)) {
3105 struct net_device *dev
3106 = list_entry(list.next, struct net_device, todo_list);
3107 list_del(&dev->todo_list);
3109 switch(dev->reg_state) {
3110 case NETREG_REGISTERING:
3111 err = netdev_register_sysfs(dev);
3113 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3115 dev->reg_state = NETREG_REGISTERED;
3118 case NETREG_UNREGISTERING:
3119 netdev_unregister_sysfs(dev);
3120 dev->reg_state = NETREG_UNREGISTERED;
3122 netdev_wait_allrefs(dev);
3125 BUG_ON(atomic_read(&dev->refcnt));
3126 BUG_TRAP(!dev->ip_ptr);
3127 BUG_TRAP(!dev->ip6_ptr);
3128 BUG_TRAP(!dev->dn_ptr);
3131 /* It must be the very last action,
3132 * after this 'dev' may point to freed up memory.
3134 if (dev->destructor)
3135 dev->destructor(dev);
3139 printk(KERN_ERR "network todo '%s' but state %d\n",
3140 dev->name, dev->reg_state);
3146 up(&net_todo_run_mutex);
3150 * free_netdev - free network device
3153 * This function does the last stage of destroying an allocated device
3154 * interface. The reference to the device object is released.
3155 * If this is the last reference then it will be freed.
3157 void free_netdev(struct net_device *dev)
3160 /* Compatiablity with error handling in drivers */
3161 if (dev->reg_state == NETREG_UNINITIALIZED) {
3162 kfree((char *)dev - dev->padded);
3166 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3167 dev->reg_state = NETREG_RELEASED;
3169 /* will free via class release */
3170 class_device_put(&dev->class_dev);
3172 kfree((char *)dev - dev->padded);
3176 /* Synchronize with packet receive processing. */
3177 void synchronize_net(void)
3180 synchronize_kernel();
3184 * unregister_netdevice - remove device from the kernel
3187 * This function shuts down a device interface and removes it
3188 * from the kernel tables. On success 0 is returned, on a failure
3189 * a negative errno code is returned.
3191 * Callers must hold the rtnl semaphore. See the comment at the
3192 * end of Space.c for details about the locking. You may want
3193 * unregister_netdev() instead of this.
3196 int unregister_netdevice(struct net_device *dev)
3198 struct net_device *d, **dp;
3200 BUG_ON(dev_boot_phase);
3203 /* Some devices call without registering for initialization unwind. */
3204 if (dev->reg_state == NETREG_UNINITIALIZED) {
3205 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3206 "was registered\n", dev->name, dev);
3210 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3212 /* If device is running, close it first. */
3213 if (dev->flags & IFF_UP)
3216 /* And unlink it from device chain. */
3217 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3219 write_lock_bh(&dev_base_lock);
3220 hlist_del(&dev->name_hlist);
3221 hlist_del(&dev->index_hlist);
3222 if (dev_tail == &dev->next)
3225 write_unlock_bh(&dev_base_lock);
3230 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3235 dev->reg_state = NETREG_UNREGISTERING;
3239 /* Shutdown queueing discipline. */
3243 /* Notify protocols, that we are about to destroy
3244 this device. They should clean all the things.
3246 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3249 * Flush the multicast chain
3251 dev_mc_discard(dev);
3256 /* Notifier chain MUST detach us from master device. */
3257 BUG_TRAP(!dev->master);
3259 free_divert_blk(dev);
3261 /* Finish processing unregister after unlock */
3270 #ifdef CONFIG_HOTPLUG_CPU
3271 static int dev_cpu_callback(struct notifier_block *nfb,
3272 unsigned long action,
3275 struct sk_buff **list_skb;
3276 struct net_device **list_net;
3277 struct sk_buff *skb;
3278 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3279 struct softnet_data *sd, *oldsd;
3281 if (action != CPU_DEAD)
3284 local_irq_disable();
3285 cpu = smp_processor_id();
3286 sd = &per_cpu(softnet_data, cpu);
3287 oldsd = &per_cpu(softnet_data, oldcpu);
3289 /* Find end of our completion_queue. */
3290 list_skb = &sd->completion_queue;
3292 list_skb = &(*list_skb)->next;
3293 /* Append completion queue from offline CPU. */
3294 *list_skb = oldsd->completion_queue;
3295 oldsd->completion_queue = NULL;
3297 /* Find end of our output_queue. */
3298 list_net = &sd->output_queue;
3300 list_net = &(*list_net)->next_sched;
3301 /* Append output queue from offline CPU. */
3302 *list_net = oldsd->output_queue;
3303 oldsd->output_queue = NULL;
3305 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3308 /* Process offline CPU's input_pkt_queue */
3309 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3314 #endif /* CONFIG_HOTPLUG_CPU */
3318 * Initialize the DEV module. At boot time this walks the device list and
3319 * unhooks any devices that fail to initialise (normally hardware not
3320 * present) and leaves us with a valid list of present and active devices.
3325 * This is called single threaded during boot, so no need
3326 * to take the rtnl semaphore.
3328 static int __init net_dev_init(void)
3330 int i, rc = -ENOMEM;
3332 BUG_ON(!dev_boot_phase);
3334 if (dev_proc_init())
3337 if (netdev_sysfs_init())
3340 INIT_LIST_HEAD(&ptype_all);
3341 for (i = 0; i < 16; i++)
3342 INIT_LIST_HEAD(&ptype_base[i]);
3344 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3345 INIT_HLIST_HEAD(&dev_name_head[i]);
3347 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3348 INIT_HLIST_HEAD(&dev_index_head[i]);
3351 * Initialise the packet receive queues.
3354 for (i = 0; i < NR_CPUS; i++) {
3355 struct softnet_data *queue;
3357 queue = &per_cpu(softnet_data, i);
3358 skb_queue_head_init(&queue->input_pkt_queue);
3359 queue->throttle = 0;
3360 queue->cng_level = 0;
3361 queue->avg_blog = 10; /* arbitrary non-zero */
3362 queue->completion_queue = NULL;
3363 INIT_LIST_HEAD(&queue->poll_list);
3364 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3365 queue->backlog_dev.weight = weight_p;
3366 queue->backlog_dev.poll = process_backlog;
3367 atomic_set(&queue->backlog_dev.refcnt, 1);
3370 #ifdef OFFLINE_SAMPLE
3371 samp_timer.expires = jiffies + (10 * HZ);
3372 add_timer(&samp_timer);
3377 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3378 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3380 hotcpu_notifier(dev_cpu_callback, 0);
3388 subsys_initcall(net_dev_init);
3390 EXPORT_SYMBOL(__dev_get);
3391 EXPORT_SYMBOL(__dev_get_by_flags);
3392 EXPORT_SYMBOL(__dev_get_by_index);
3393 EXPORT_SYMBOL(__dev_get_by_name);
3394 EXPORT_SYMBOL(__dev_remove_pack);
3395 EXPORT_SYMBOL(__skb_linearize);
3396 EXPORT_SYMBOL(call_netdevice_notifiers);
3397 EXPORT_SYMBOL(dev_add_pack);
3398 EXPORT_SYMBOL(dev_alloc_name);
3399 EXPORT_SYMBOL(dev_close);
3400 EXPORT_SYMBOL(dev_get_by_flags);
3401 EXPORT_SYMBOL(dev_get_by_index);
3402 EXPORT_SYMBOL(dev_get_by_name);
3403 EXPORT_SYMBOL(dev_getbyhwaddr);
3404 EXPORT_SYMBOL(dev_ioctl);
3405 EXPORT_SYMBOL(dev_new_index);
3406 EXPORT_SYMBOL(dev_open);
3407 EXPORT_SYMBOL(dev_queue_xmit);
3408 EXPORT_SYMBOL(dev_queue_xmit_nit);
3409 EXPORT_SYMBOL(dev_remove_pack);
3410 EXPORT_SYMBOL(dev_set_allmulti);
3411 EXPORT_SYMBOL(dev_set_promiscuity);
3412 EXPORT_SYMBOL(dev_change_flags);
3413 EXPORT_SYMBOL(dev_set_mtu);
3414 EXPORT_SYMBOL(free_netdev);
3415 EXPORT_SYMBOL(netdev_boot_setup_check);
3416 EXPORT_SYMBOL(netdev_set_master);
3417 EXPORT_SYMBOL(netdev_state_change);
3418 EXPORT_SYMBOL(netif_receive_skb);
3419 EXPORT_SYMBOL(netif_rx);
3420 EXPORT_SYMBOL(register_gifconf);
3421 EXPORT_SYMBOL(register_netdevice);
3422 EXPORT_SYMBOL(register_netdevice_notifier);
3423 EXPORT_SYMBOL(skb_checksum_help);
3424 EXPORT_SYMBOL(synchronize_net);
3425 EXPORT_SYMBOL(unregister_netdevice);
3426 EXPORT_SYMBOL(unregister_netdevice_notifier);
3428 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3429 EXPORT_SYMBOL(br_handle_frame_hook);
3433 EXPORT_SYMBOL(dev_load);
3435 #ifdef CONFIG_NET_HW_FLOWCONTROL
3436 EXPORT_SYMBOL(netdev_dropping);
3437 EXPORT_SYMBOL(netdev_fc_xoff);
3438 EXPORT_SYMBOL(netdev_register_fc);
3439 EXPORT_SYMBOL(netdev_unregister_fc);
3442 #ifdef CONFIG_NET_CLS_ACT
3443 EXPORT_SYMBOL(ing_filter);
3447 EXPORT_PER_CPU_SYMBOL(softnet_data);