2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <asm/bitops.h>
78 #include <linux/config.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/notifier.h>
93 #include <linux/skbuff.h>
95 #include <linux/rtnetlink.h>
96 #include <linux/proc_fs.h>
97 #include <linux/seq_file.h>
98 #include <linux/stat.h>
99 #include <linux/if_bridge.h>
100 #include <linux/divert.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #ifdef CONFIG_NET_RADIO
112 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
113 #include <net/iw_handler.h>
114 #endif /* CONFIG_NET_RADIO */
115 #include <asm/current.h>
116 #include <linux/vs_network.h>
118 /* This define, if set, will randomly drop a packet when congestion
119 * is more than moderate. It helps fairness in the multi-interface
120 * case when one of them is a hog, but it kills performance for the
121 * single interface case so it is off now by default.
125 /* Setting this will sample the queue lengths and thus congestion
126 * via a timer instead of as each packet is received.
128 #undef OFFLINE_SAMPLE
131 * The list of packet types we will receive (as opposed to discard)
132 * and the routines to invoke.
134 * Why 16. Because with 16 the only overlap we get on a hash of the
135 * low nibble of the protocol value is RARP/SNAP/X.25.
137 * NOTE: That is no longer true with the addition of VLAN tags. Not
138 * sure which should go first, but I bet it won't make much
139 * difference if we are running VLANs. The good news is that
140 * this protocol won't be in the list unless compiled in, so
141 * the average user (w/out VLANs) will not be adversly affected.
158 static spinlock_t ptype_lock = SPIN_LOCK_UNLOCKED;
159 static struct list_head ptype_base[16]; /* 16 way hashed list */
160 static struct list_head ptype_all; /* Taps */
162 #ifdef OFFLINE_SAMPLE
163 static void sample_queue(unsigned long dummy);
164 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
168 * The @dev_base list is protected by @dev_base_lock and the rtln
171 * Pure readers hold dev_base_lock for reading.
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
186 struct net_device *dev_base;
187 struct net_device **dev_tail = &dev_base;
188 rwlock_t dev_base_lock = RW_LOCK_UNLOCKED;
190 EXPORT_SYMBOL(dev_base);
191 EXPORT_SYMBOL(dev_base_lock);
193 #define NETDEV_HASHBITS 8
194 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
195 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
197 static inline struct hlist_head *dev_name_hash(const char *name)
199 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
200 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
203 static inline struct hlist_head *dev_index_hash(int ifindex)
205 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
212 static struct notifier_block *netdev_chain;
215 * Device drivers call our routines to queue packets here. We empty the
216 * queue in the local softnet handler.
218 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
221 extern int netdev_sysfs_init(void);
222 extern int netdev_register_sysfs(struct net_device *);
223 extern void netdev_unregister_sysfs(struct net_device *);
225 #define netdev_sysfs_init() (0)
226 #define netdev_register_sysfs(dev) (0)
227 #define netdev_unregister_sysfs(dev) do { } while(0)
231 /*******************************************************************************
233 Protocol management and registration routines
235 *******************************************************************************/
244 * Add a protocol ID to the list. Now that the input handler is
245 * smarter we can dispense with all the messy stuff that used to be
248 * BEWARE!!! Protocol handlers, mangling input packets,
249 * MUST BE last in hash buckets and checking protocol handlers
250 * MUST start from promiscuous ptype_all chain in net_bh.
251 * It is true now, do not change it.
252 * Explanation follows: if protocol handler, mangling packet, will
253 * be the first on list, it is not able to sense, that packet
254 * is cloned and should be copied-on-write, so that it will
255 * change it and subsequent readers will get broken packet.
260 * dev_add_pack - add packet handler
261 * @pt: packet type declaration
263 * Add a protocol handler to the networking stack. The passed &packet_type
264 * is linked into kernel lists and may not be freed until it has been
265 * removed from the kernel lists.
267 * This call does not sleep therefore it can not
268 * guarantee all CPU's that are in middle of receiving packets
269 * will see the new packet type (until the next received packet).
272 void dev_add_pack(struct packet_type *pt)
276 spin_lock_bh(&ptype_lock);
277 if (pt->type == htons(ETH_P_ALL)) {
279 list_add_rcu(&pt->list, &ptype_all);
281 hash = ntohs(pt->type) & 15;
282 list_add_rcu(&pt->list, &ptype_base[hash]);
284 spin_unlock_bh(&ptype_lock);
287 extern void linkwatch_run_queue(void);
292 * __dev_remove_pack - remove packet handler
293 * @pt: packet type declaration
295 * Remove a protocol handler that was previously added to the kernel
296 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
297 * from the kernel lists and can be freed or reused once this function
300 * The packet type might still be in use by receivers
301 * and must not be freed until after all the CPU's have gone
302 * through a quiescent state.
304 void __dev_remove_pack(struct packet_type *pt)
306 struct list_head *head;
307 struct packet_type *pt1;
309 spin_lock_bh(&ptype_lock);
311 if (pt->type == htons(ETH_P_ALL)) {
315 head = &ptype_base[ntohs(pt->type) & 15];
317 list_for_each_entry(pt1, head, list) {
319 list_del_rcu(&pt->list);
324 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
326 spin_unlock_bh(&ptype_lock);
329 * dev_remove_pack - remove packet handler
330 * @pt: packet type declaration
332 * Remove a protocol handler that was previously added to the kernel
333 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
334 * from the kernel lists and can be freed or reused once this function
337 * This call sleeps to guarantee that no CPU is looking at the packet
340 void dev_remove_pack(struct packet_type *pt)
342 __dev_remove_pack(pt);
347 /******************************************************************************
349 Device Boot-time Settings Routines
351 *******************************************************************************/
353 /* Boot time configuration table */
354 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
357 * netdev_boot_setup_add - add new setup entry
358 * @name: name of the device
359 * @map: configured settings for the device
361 * Adds new setup entry to the dev_boot_setup list. The function
362 * returns 0 on error and 1 on success. This is a generic routine to
365 int netdev_boot_setup_add(char *name, struct ifmap *map)
367 struct netdev_boot_setup *s;
371 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
372 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
373 memset(s[i].name, 0, sizeof(s[i].name));
374 strcpy(s[i].name, name);
375 memcpy(&s[i].map, map, sizeof(s[i].map));
380 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
384 * netdev_boot_setup_check - check boot time settings
385 * @dev: the netdevice
387 * Check boot time settings for the device.
388 * The found settings are set for the device to be used
389 * later in the device probing.
390 * Returns 0 if no settings found, 1 if they are.
392 int netdev_boot_setup_check(struct net_device *dev)
394 struct netdev_boot_setup *s = dev_boot_setup;
397 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
398 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
399 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
400 dev->irq = s[i].map.irq;
401 dev->base_addr = s[i].map.base_addr;
402 dev->mem_start = s[i].map.mem_start;
403 dev->mem_end = s[i].map.mem_end;
412 * netdev_boot_base - get address from boot time settings
413 * @prefix: prefix for network device
414 * @unit: id for network device
416 * Check boot time settings for the base address of device.
417 * The found settings are set for the device to be used
418 * later in the device probing.
419 * Returns 0 if no settings found.
421 unsigned long netdev_boot_base(const char *prefix, int unit)
423 const struct netdev_boot_setup *s = dev_boot_setup;
427 sprintf(name, "%s%d", prefix, unit);
430 * If device already registered then return base of 1
431 * to indicate not to probe for this interface
433 if (__dev_get_by_name(name))
436 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
437 if (!strcmp(name, s[i].name))
438 return s[i].map.base_addr;
443 * Saves at boot time configured settings for any netdevice.
445 int __init netdev_boot_setup(char *str)
450 str = get_options(str, ARRAY_SIZE(ints), ints);
455 memset(&map, 0, sizeof(map));
459 map.base_addr = ints[2];
461 map.mem_start = ints[3];
463 map.mem_end = ints[4];
465 /* Add new entry to the list */
466 return netdev_boot_setup_add(str, &map);
469 __setup("netdev=", netdev_boot_setup);
471 /*******************************************************************************
473 Device Interface Subroutines
475 *******************************************************************************/
478 * __dev_get_by_name - find a device by its name
479 * @name: name to find
481 * Find an interface by name. Must be called under RTNL semaphore
482 * or @dev_base_lock. If the name is found a pointer to the device
483 * is returned. If the name is not found then %NULL is returned. The
484 * reference counters are not incremented so the caller must be
485 * careful with locks.
488 struct net_device *__dev_get_by_name(const char *name)
490 struct hlist_node *p;
492 hlist_for_each(p, dev_name_hash(name)) {
493 struct net_device *dev
494 = hlist_entry(p, struct net_device, name_hlist);
495 if (!strncmp(dev->name, name, IFNAMSIZ))
502 * dev_get_by_name - find a device by its name
503 * @name: name to find
505 * Find an interface by name. This can be called from any
506 * context and does its own locking. The returned handle has
507 * the usage count incremented and the caller must use dev_put() to
508 * release it when it is no longer needed. %NULL is returned if no
509 * matching device is found.
512 struct net_device *dev_get_by_name(const char *name)
514 struct net_device *dev;
516 read_lock(&dev_base_lock);
517 dev = __dev_get_by_name(name);
520 read_unlock(&dev_base_lock);
525 Return value is changed to int to prevent illegal usage in future.
526 It is still legal to use to check for device existence.
528 User should understand, that the result returned by this function
529 is meaningless, if it was not issued under rtnl semaphore.
533 * dev_get - test if a device exists
534 * @name: name to test for
536 * Test if a name exists. Returns true if the name is found. In order
537 * to be sure the name is not allocated or removed during the test the
538 * caller must hold the rtnl semaphore.
540 * This function exists only for back compatibility with older
543 int __dev_get(const char *name)
545 struct net_device *dev;
547 read_lock(&dev_base_lock);
548 dev = __dev_get_by_name(name);
549 read_unlock(&dev_base_lock);
554 * __dev_get_by_index - find a device by its ifindex
555 * @ifindex: index of device
557 * Search for an interface by index. Returns %NULL if the device
558 * is not found or a pointer to the device. The device has not
559 * had its reference counter increased so the caller must be careful
560 * about locking. The caller must hold either the RTNL semaphore
564 struct net_device *__dev_get_by_index(int ifindex)
566 struct hlist_node *p;
568 hlist_for_each(p, dev_index_hash(ifindex)) {
569 struct net_device *dev
570 = hlist_entry(p, struct net_device, index_hlist);
571 if (dev->ifindex == ifindex)
579 * dev_get_by_index - find a device by its ifindex
580 * @ifindex: index of device
582 * Search for an interface by index. Returns NULL if the device
583 * is not found or a pointer to the device. The device returned has
584 * had a reference added and the pointer is safe until the user calls
585 * dev_put to indicate they have finished with it.
588 struct net_device *dev_get_by_index(int ifindex)
590 struct net_device *dev;
592 read_lock(&dev_base_lock);
593 dev = __dev_get_by_index(ifindex);
596 read_unlock(&dev_base_lock);
601 * dev_getbyhwaddr - find a device by its hardware address
602 * @type: media type of device
603 * @ha: hardware address
605 * Search for an interface by MAC address. Returns NULL if the device
606 * is not found or a pointer to the device. The caller must hold the
607 * rtnl semaphore. The returned device has not had its ref count increased
608 * and the caller must therefore be careful about locking
611 * If the API was consistent this would be __dev_get_by_hwaddr
614 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
616 struct net_device *dev;
620 for (dev = dev_base; dev; dev = dev->next)
621 if (dev->type == type &&
622 !memcmp(dev->dev_addr, ha, dev->addr_len))
627 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
629 struct net_device *dev;
631 for (dev = dev_base; dev; dev = dev->next)
632 if (dev->type == type)
637 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
639 struct net_device *dev_getfirstbyhwtype(unsigned short type)
641 struct net_device *dev;
644 dev = __dev_getfirstbyhwtype(type);
651 EXPORT_SYMBOL(dev_getfirstbyhwtype);
654 * dev_get_by_flags - find any device with given flags
655 * @if_flags: IFF_* values
656 * @mask: bitmask of bits in if_flags to check
658 * Search for any interface with the given flags. Returns NULL if a device
659 * is not found or a pointer to the device. The device returned has
660 * had a reference added and the pointer is safe until the user calls
661 * dev_put to indicate they have finished with it.
664 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
666 struct net_device *dev;
668 read_lock(&dev_base_lock);
669 dev = __dev_get_by_flags(if_flags, mask);
672 read_unlock(&dev_base_lock);
677 * __dev_get_by_flags - find any device with given flags
678 * @if_flags: IFF_* values
679 * @mask: bitmask of bits in if_flags to check
681 * Search for any interface with the given flags. Returns NULL if a device
682 * is not found or a pointer to the device. The caller must hold either
683 * the RTNL semaphore or @dev_base_lock.
686 struct net_device *__dev_get_by_flags(unsigned short if_flags, unsigned short mask)
688 struct net_device *dev;
690 for (dev = dev_base; dev != NULL; dev = dev->next) {
691 if (((dev->flags ^ if_flags) & mask) == 0)
698 * dev_valid_name - check if name is okay for network device
701 * Network device names need to be valid file names to
702 * to allow sysfs to work
704 int dev_valid_name(const char *name)
706 return !(*name == '\0'
707 || !strcmp(name, ".")
708 || !strcmp(name, "..")
709 || strchr(name, '/'));
713 * dev_alloc_name - allocate a name for a device
715 * @name: name format string
717 * Passed a format string - eg "lt%d" it will try and find a suitable
718 * id. Not efficient for many devices, not called a lot. The caller
719 * must hold the dev_base or rtnl lock while allocating the name and
720 * adding the device in order to avoid duplicates. Returns the number
721 * of the unit assigned or a negative errno code.
724 int dev_alloc_name(struct net_device *dev, const char *name)
729 const int max_netdevices = 8*PAGE_SIZE;
731 struct net_device *d;
733 p = strnchr(name, IFNAMSIZ-1, '%');
736 * Verify the string as this thing may have come from
737 * the user. There must be either one "%d" and no other "%"
740 if (p[1] != 'd' || strchr(p + 2, '%'))
743 /* Use one page as a bit array of possible slots */
744 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
748 for (d = dev_base; d; d = d->next) {
749 if (!sscanf(d->name, name, &i))
751 if (i < 0 || i >= max_netdevices)
754 /* avoid cases where sscanf is not exact inverse of printf */
755 snprintf(buf, sizeof(buf), name, i);
756 if (!strncmp(buf, d->name, IFNAMSIZ))
760 i = find_first_zero_bit(inuse, max_netdevices);
761 free_page((unsigned long) inuse);
764 snprintf(buf, sizeof(buf), name, i);
765 if (!__dev_get_by_name(buf)) {
766 strlcpy(dev->name, buf, IFNAMSIZ);
770 /* It is possible to run out of possible slots
771 * when the name is long and there isn't enough space left
772 * for the digits, or if all bits are used.
779 * dev_change_name - change name of a device
781 * @newname: name (or format string) must be at least IFNAMSIZ
783 * Change name of a device, can pass format strings "eth%d".
786 int dev_change_name(struct net_device *dev, char *newname)
792 if (dev->flags & IFF_UP)
795 if (!dev_valid_name(newname))
798 if (strchr(newname, '%')) {
799 err = dev_alloc_name(dev, newname);
802 strcpy(newname, dev->name);
804 else if (__dev_get_by_name(newname))
807 strlcpy(dev->name, newname, IFNAMSIZ);
809 err = class_device_rename(&dev->class_dev, dev->name);
811 hlist_del(&dev->name_hlist);
812 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
813 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
820 * netdev_state_change - device changes state
821 * @dev: device to cause notification
823 * Called to indicate a device has changed state. This function calls
824 * the notifier chains for netdev_chain and sends a NEWLINK message
825 * to the routing socket.
827 void netdev_state_change(struct net_device *dev)
829 if (dev->flags & IFF_UP) {
830 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
831 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
836 * dev_load - load a network module
837 * @name: name of interface
839 * If a network interface is not present and the process has suitable
840 * privileges this function loads the module. If module loading is not
841 * available in this kernel then it becomes a nop.
844 void dev_load(const char *name)
846 struct net_device *dev;
848 read_lock(&dev_base_lock);
849 dev = __dev_get_by_name(name);
850 read_unlock(&dev_base_lock);
852 if (!dev && capable(CAP_SYS_MODULE))
853 request_module("%s", name);
856 static int default_rebuild_header(struct sk_buff *skb)
858 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
859 skb->dev ? skb->dev->name : "NULL!!!");
866 * Some old buggy device drivers change get_stats after registering
867 * the device. Try and trap them here.
868 * This can be elimnated when all devices are known fixed.
870 static inline int get_stats_changed(struct net_device *dev)
872 int changed = dev->last_stats != dev->get_stats;
873 dev->last_stats = dev->get_stats;
878 * dev_open - prepare an interface for use.
879 * @dev: device to open
881 * Takes a device from down to up state. The device's private open
882 * function is invoked and then the multicast lists are loaded. Finally
883 * the device is moved into the up state and a %NETDEV_UP message is
884 * sent to the netdev notifier chain.
886 * Calling this function on an active interface is a nop. On a failure
887 * a negative errno code is returned.
889 int dev_open(struct net_device *dev)
897 if (dev->flags & IFF_UP)
901 * Check for broken device drivers.
903 if (get_stats_changed(dev) && net_ratelimit()) {
904 printk(KERN_ERR "%s: driver changed get_stats after register\n",
909 * Is it even present?
911 if (!netif_device_present(dev))
915 * Call device private open method
917 set_bit(__LINK_STATE_START, &dev->state);
919 ret = dev->open(dev);
921 clear_bit(__LINK_STATE_START, &dev->state);
925 * Check for more broken device drivers.
927 if (get_stats_changed(dev) && net_ratelimit()) {
928 printk(KERN_ERR "%s: driver changed get_stats in open\n",
933 * If it went open OK then:
940 dev->flags |= IFF_UP;
943 * Initialize multicasting status
948 * Wakeup transmit queue engine
953 * ... and announce new interface.
955 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
961 * dev_close - shutdown an interface.
962 * @dev: device to shutdown
964 * This function moves an active device into down state. A
965 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
966 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
969 int dev_close(struct net_device *dev)
971 if (!(dev->flags & IFF_UP))
975 * Tell people we are going down, so that they can
976 * prepare to death, when device is still operating.
978 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
982 clear_bit(__LINK_STATE_START, &dev->state);
984 /* Synchronize to scheduled poll. We cannot touch poll list,
985 * it can be even on different cpu. So just clear netif_running(),
986 * and wait when poll really will happen. Actually, the best place
987 * for this is inside dev->stop() after device stopped its irq
988 * engine, but this requires more changes in devices. */
990 smp_mb__after_clear_bit(); /* Commit netif_running(). */
991 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
993 current->state = TASK_INTERRUPTIBLE;
998 * Call the device specific close. This cannot fail.
999 * Only if device is UP
1001 * We allow it to be called even after a DETACH hot-plug
1008 * Device is now down.
1011 dev->flags &= ~IFF_UP;
1014 * Tell people we are down
1016 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1023 * Device change register/unregister. These are not inline or static
1024 * as we export them to the world.
1028 * register_netdevice_notifier - register a network notifier block
1031 * Register a notifier to be called when network device events occur.
1032 * The notifier passed is linked into the kernel structures and must
1033 * not be reused until it has been unregistered. A negative errno code
1034 * is returned on a failure.
1036 * When registered all registration and up events are replayed
1037 * to the new notifier to allow device to have a race free
1038 * view of the network device list.
1041 int register_netdevice_notifier(struct notifier_block *nb)
1043 struct net_device *dev;
1047 err = notifier_chain_register(&netdev_chain, nb);
1049 for (dev = dev_base; dev; dev = dev->next) {
1050 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1052 if (dev->flags & IFF_UP)
1053 nb->notifier_call(nb, NETDEV_UP, dev);
1061 * unregister_netdevice_notifier - unregister a network notifier block
1064 * Unregister a notifier previously registered by
1065 * register_netdevice_notifier(). The notifier is unlinked into the
1066 * kernel structures and may then be reused. A negative errno code
1067 * is returned on a failure.
1070 int unregister_netdevice_notifier(struct notifier_block *nb)
1072 return notifier_chain_unregister(&netdev_chain, nb);
1076 * call_netdevice_notifiers - call all network notifier blocks
1077 * @val: value passed unmodified to notifier function
1078 * @v: pointer passed unmodified to notifier function
1080 * Call all network notifier blocks. Parameters and return value
1081 * are as for notifier_call_chain().
1084 int call_netdevice_notifiers(unsigned long val, void *v)
1086 return notifier_call_chain(&netdev_chain, val, v);
1090 * Support routine. Sends outgoing frames to any network
1091 * taps currently in use.
1094 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1096 struct packet_type *ptype;
1097 net_timestamp(&skb->stamp);
1100 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1101 /* Never send packets back to the socket
1102 * they originated from - MvS (miquels@drinkel.ow.org)
1104 if ((ptype->dev == dev || !ptype->dev) &&
1105 (ptype->af_packet_priv == NULL ||
1106 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1107 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1111 /* skb->nh should be correctly
1112 set by sender, so that the second statement is
1113 just protection against buggy protocols.
1115 skb2->mac.raw = skb2->data;
1117 if (skb2->nh.raw < skb2->data ||
1118 skb2->nh.raw > skb2->tail) {
1119 if (net_ratelimit())
1120 printk(KERN_CRIT "protocol %04x is "
1122 skb2->protocol, dev->name);
1123 skb2->nh.raw = skb2->data;
1126 skb2->h.raw = skb2->nh.raw;
1127 skb2->pkt_type = PACKET_OUTGOING;
1128 ptype->func(skb2, skb->dev, ptype);
1135 * Invalidate hardware checksum when packet is to be mangled, and
1136 * complete checksum manually on outgoing path.
1138 int skb_checksum_help(struct sk_buff **pskb, int inward)
1141 int ret = 0, offset = (*pskb)->h.raw - (*pskb)->data;
1144 (*pskb)->ip_summed = CHECKSUM_NONE;
1148 if (skb_shared(*pskb) || skb_cloned(*pskb)) {
1149 struct sk_buff *newskb = skb_copy(*pskb, GFP_ATOMIC);
1155 skb_set_owner_w(newskb, (*pskb)->sk);
1160 if (offset > (int)(*pskb)->len)
1162 csum = skb_checksum(*pskb, offset, (*pskb)->len-offset, 0);
1164 offset = (*pskb)->tail - (*pskb)->h.raw;
1167 if ((*pskb)->csum + 2 > offset)
1170 *(u16*)((*pskb)->h.raw + (*pskb)->csum) = csum_fold(csum);
1171 (*pskb)->ip_summed = CHECKSUM_NONE;
1176 #ifdef CONFIG_HIGHMEM
1177 /* Actually, we should eliminate this check as soon as we know, that:
1178 * 1. IOMMU is present and allows to map all the memory.
1179 * 2. No high memory really exists on this machine.
1182 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1186 if (dev->features & NETIF_F_HIGHDMA)
1189 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1190 if (skb_shinfo(skb)->frags[i].page >= highmem_start_page)
1196 #define illegal_highdma(dev, skb) (0)
1199 extern void skb_release_data(struct sk_buff *);
1201 /* Keep head the same: replace data */
1202 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1207 struct skb_shared_info *ninfo;
1208 int headerlen = skb->data - skb->head;
1209 int expand = (skb->tail + skb->data_len) - skb->end;
1211 if (skb_shared(skb))
1217 size = skb->end - skb->head + expand;
1218 size = SKB_DATA_ALIGN(size);
1219 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1223 /* Copy entire thing */
1224 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1228 ninfo = (struct skb_shared_info*)(data + size);
1229 atomic_set(&ninfo->dataref, 1);
1230 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1231 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1232 ninfo->nr_frags = 0;
1233 ninfo->frag_list = NULL;
1235 /* Offset between the two in bytes */
1236 offset = data - skb->head;
1238 /* Free old data. */
1239 skb_release_data(skb);
1242 skb->end = data + size;
1244 /* Set up new pointers */
1245 skb->h.raw += offset;
1246 skb->nh.raw += offset;
1247 skb->mac.raw += offset;
1248 skb->tail += offset;
1249 skb->data += offset;
1251 /* We are no longer a clone, even if we were. */
1254 skb->tail += skb->data_len;
1259 #define HARD_TX_LOCK_BH(dev, cpu) { \
1260 if ((dev->features & NETIF_F_LLTX) == 0) { \
1261 spin_lock_bh(&dev->xmit_lock); \
1262 dev->xmit_lock_owner = cpu; \
1266 #define HARD_TX_UNLOCK_BH(dev) { \
1267 if ((dev->features & NETIF_F_LLTX) == 0) { \
1268 dev->xmit_lock_owner = -1; \
1269 spin_unlock_bh(&dev->xmit_lock); \
1273 static inline void qdisc_run(struct net_device *dev)
1275 while (!netif_queue_stopped(dev) &&
1276 qdisc_restart(dev)<0)
1281 * dev_queue_xmit - transmit a buffer
1282 * @skb: buffer to transmit
1284 * Queue a buffer for transmission to a network device. The caller must
1285 * have set the device and priority and built the buffer before calling
1286 * this function. The function can be called from an interrupt.
1288 * A negative errno code is returned on a failure. A success does not
1289 * guarantee the frame will be transmitted as it may be dropped due
1290 * to congestion or traffic shaping.
1293 int dev_queue_xmit(struct sk_buff *skb)
1295 struct net_device *dev = skb->dev;
1299 if (skb_shinfo(skb)->frag_list &&
1300 !(dev->features & NETIF_F_FRAGLIST) &&
1301 __skb_linearize(skb, GFP_ATOMIC))
1304 /* Fragmented skb is linearized if device does not support SG,
1305 * or if at least one of fragments is in highmem and device
1306 * does not support DMA from it.
1308 if (skb_shinfo(skb)->nr_frags &&
1309 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1310 __skb_linearize(skb, GFP_ATOMIC))
1313 /* If packet is not checksummed and device does not support
1314 * checksumming for this protocol, complete checksumming here.
1316 if (skb->ip_summed == CHECKSUM_HW &&
1317 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1318 (!(dev->features & NETIF_F_IP_CSUM) ||
1319 skb->protocol != htons(ETH_P_IP))))
1320 if (skb_checksum_help(&skb, 0))
1324 /* Updates of qdisc are serialized by queue_lock.
1325 * The struct Qdisc which is pointed to by qdisc is now a
1326 * rcu structure - it may be accessed without acquiring
1327 * a lock (but the structure may be stale.) The freeing of the
1328 * qdisc will be deferred until it's known that there are no
1329 * more references to it.
1331 * If the qdisc has an enqueue function, we still need to
1332 * hold the queue_lock before calling it, since queue_lock
1333 * also serializes access to the device queue.
1337 smp_read_barrier_depends();
1338 #ifdef CONFIG_NET_CLS_ACT
1339 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1342 /* Grab device queue */
1343 spin_lock_bh(&dev->queue_lock);
1345 rc = q->enqueue(skb, q);
1349 spin_unlock_bh(&dev->queue_lock);
1351 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1356 /* The device has no queue. Common case for software devices:
1357 loopback, all the sorts of tunnels...
1359 Really, it is unlikely that xmit_lock protection is necessary here.
1360 (f.e. loopback and IP tunnels are clean ignoring statistics
1362 However, it is possible, that they rely on protection
1365 Check this and shot the lock. It is not prone from deadlocks.
1366 Either shot noqueue qdisc, it is even simpler 8)
1368 if (dev->flags & IFF_UP) {
1369 int cpu = get_cpu();
1371 if (dev->xmit_lock_owner != cpu) {
1373 HARD_TX_LOCK_BH(dev, cpu);
1376 if (!netif_queue_stopped(dev)) {
1378 dev_queue_xmit_nit(skb, dev);
1381 if (!dev->hard_start_xmit(skb, dev)) {
1382 HARD_TX_UNLOCK_BH(dev);
1386 HARD_TX_UNLOCK_BH(dev);
1387 if (net_ratelimit())
1388 printk(KERN_CRIT "Virtual device %s asks to "
1389 "queue packet!\n", dev->name);
1393 /* Recursion is detected! It is possible,
1395 if (net_ratelimit())
1396 printk(KERN_CRIT "Dead loop on virtual device "
1397 "%s, fix it urgently!\n", dev->name);
1409 /*=======================================================================
1411 =======================================================================*/
1413 int netdev_max_backlog = 300;
1414 int weight_p = 64; /* old backlog weight */
1415 /* These numbers are selected based on intuition and some
1416 * experimentatiom, if you have more scientific way of doing this
1417 * please go ahead and fix things.
1419 int no_cong_thresh = 10;
1424 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1427 #ifdef CONFIG_NET_HW_FLOWCONTROL
1428 atomic_t netdev_dropping = ATOMIC_INIT(0);
1429 static unsigned long netdev_fc_mask = 1;
1430 unsigned long netdev_fc_xoff;
1431 spinlock_t netdev_fc_lock = SPIN_LOCK_UNLOCKED;
1435 void (*stimul)(struct net_device *);
1436 struct net_device *dev;
1437 } netdev_fc_slots[BITS_PER_LONG];
1439 int netdev_register_fc(struct net_device *dev,
1440 void (*stimul)(struct net_device *dev))
1443 unsigned long flags;
1445 spin_lock_irqsave(&netdev_fc_lock, flags);
1446 if (netdev_fc_mask != ~0UL) {
1447 bit = ffz(netdev_fc_mask);
1448 netdev_fc_slots[bit].stimul = stimul;
1449 netdev_fc_slots[bit].dev = dev;
1450 set_bit(bit, &netdev_fc_mask);
1451 clear_bit(bit, &netdev_fc_xoff);
1453 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1457 void netdev_unregister_fc(int bit)
1459 unsigned long flags;
1461 spin_lock_irqsave(&netdev_fc_lock, flags);
1463 netdev_fc_slots[bit].stimul = NULL;
1464 netdev_fc_slots[bit].dev = NULL;
1465 clear_bit(bit, &netdev_fc_mask);
1466 clear_bit(bit, &netdev_fc_xoff);
1468 spin_unlock_irqrestore(&netdev_fc_lock, flags);
1471 static void netdev_wakeup(void)
1475 spin_lock(&netdev_fc_lock);
1476 xoff = netdev_fc_xoff;
1481 netdev_fc_slots[i].stimul(netdev_fc_slots[i].dev);
1483 spin_unlock(&netdev_fc_lock);
1487 static void get_sample_stats(int cpu)
1493 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1494 int blog = sd->input_pkt_queue.qlen;
1495 int avg_blog = sd->avg_blog;
1497 avg_blog = (avg_blog >> 1) + (blog >> 1);
1499 if (avg_blog > mod_cong) {
1500 /* Above moderate congestion levels. */
1501 sd->cng_level = NET_RX_CN_HIGH;
1504 rq = rd % netdev_max_backlog;
1505 if (rq < avg_blog) /* unlucky bastard */
1506 sd->cng_level = NET_RX_DROP;
1508 } else if (avg_blog > lo_cong) {
1509 sd->cng_level = NET_RX_CN_MOD;
1512 rq = rd % netdev_max_backlog;
1513 if (rq < avg_blog) /* unlucky bastard */
1514 sd->cng_level = NET_RX_CN_HIGH;
1516 } else if (avg_blog > no_cong)
1517 sd->cng_level = NET_RX_CN_LOW;
1518 else /* no congestion */
1519 sd->cng_level = NET_RX_SUCCESS;
1521 sd->avg_blog = avg_blog;
1524 #ifdef OFFLINE_SAMPLE
1525 static void sample_queue(unsigned long dummy)
1527 /* 10 ms 0r 1ms -- i don't care -- JHS */
1529 int cpu = smp_processor_id();
1531 get_sample_stats(cpu);
1532 next_tick += jiffies;
1533 mod_timer(&samp_timer, next_tick);
1539 * netif_rx - post buffer to the network code
1540 * @skb: buffer to post
1542 * This function receives a packet from a device driver and queues it for
1543 * the upper (protocol) levels to process. It always succeeds. The buffer
1544 * may be dropped during processing for congestion control or by the
1548 * NET_RX_SUCCESS (no congestion)
1549 * NET_RX_CN_LOW (low congestion)
1550 * NET_RX_CN_MOD (moderate congestion)
1551 * NET_RX_CN_HIGH (high congestion)
1552 * NET_RX_DROP (packet was dropped)
1556 int netif_rx(struct sk_buff *skb)
1559 struct softnet_data *queue;
1560 unsigned long flags;
1562 #ifdef CONFIG_NETPOLL_RX
1563 if (skb->dev->netpoll_rx && netpoll_rx(skb)) {
1569 if (!skb->stamp.tv_sec)
1570 net_timestamp(&skb->stamp);
1573 * The code is rearranged so that the path is the most
1574 * short when CPU is congested, but is still operating.
1576 local_irq_save(flags);
1577 this_cpu = smp_processor_id();
1578 queue = &__get_cpu_var(softnet_data);
1580 __get_cpu_var(netdev_rx_stat).total++;
1581 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1582 if (queue->input_pkt_queue.qlen) {
1583 if (queue->throttle)
1588 __skb_queue_tail(&queue->input_pkt_queue, skb);
1589 #ifndef OFFLINE_SAMPLE
1590 get_sample_stats(this_cpu);
1592 local_irq_restore(flags);
1593 return queue->cng_level;
1596 if (queue->throttle) {
1597 queue->throttle = 0;
1598 #ifdef CONFIG_NET_HW_FLOWCONTROL
1599 if (atomic_dec_and_test(&netdev_dropping))
1604 netif_rx_schedule(&queue->backlog_dev);
1608 if (!queue->throttle) {
1609 queue->throttle = 1;
1610 __get_cpu_var(netdev_rx_stat).throttled++;
1611 #ifdef CONFIG_NET_HW_FLOWCONTROL
1612 atomic_inc(&netdev_dropping);
1617 __get_cpu_var(netdev_rx_stat).dropped++;
1618 local_irq_restore(flags);
1624 static __inline__ void skb_bond(struct sk_buff *skb)
1626 struct net_device *dev = skb->dev;
1629 skb->real_dev = skb->dev;
1630 skb->dev = dev->master;
1634 static void net_tx_action(struct softirq_action *h)
1636 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1638 if (sd->completion_queue) {
1639 struct sk_buff *clist;
1641 local_irq_disable();
1642 clist = sd->completion_queue;
1643 sd->completion_queue = NULL;
1647 struct sk_buff *skb = clist;
1648 clist = clist->next;
1650 BUG_TRAP(!atomic_read(&skb->users));
1655 if (sd->output_queue) {
1656 struct net_device *head;
1658 local_irq_disable();
1659 head = sd->output_queue;
1660 sd->output_queue = NULL;
1664 struct net_device *dev = head;
1665 head = head->next_sched;
1667 smp_mb__before_clear_bit();
1668 clear_bit(__LINK_STATE_SCHED, &dev->state);
1670 if (spin_trylock(&dev->queue_lock)) {
1672 spin_unlock(&dev->queue_lock);
1674 netif_schedule(dev);
1680 static __inline__ int deliver_skb(struct sk_buff *skb,
1681 struct packet_type *pt_prev, int last)
1683 atomic_inc(&skb->users);
1684 return pt_prev->func(skb, skb->dev, pt_prev);
1688 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1689 int (*br_handle_frame_hook)(struct sk_buff *skb);
1691 static __inline__ int handle_bridge(struct sk_buff *skb,
1692 struct packet_type *pt_prev)
1694 int ret = NET_RX_DROP;
1696 ret = deliver_skb(skb, pt_prev, 0);
1703 static inline int __handle_bridge(struct sk_buff *skb,
1704 struct packet_type **pt_prev, int *ret)
1706 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
1707 if (skb->dev->br_port && skb->pkt_type != PACKET_LOOPBACK) {
1708 *ret = handle_bridge(skb, *pt_prev);
1709 if (br_handle_frame_hook(skb) == 0)
1719 #ifdef CONFIG_NET_CLS_ACT
1720 /* TODO: Maybe we should just force sch_ingress to be compiled in
1721 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1722 * a compare and 2 stores extra right now if we dont have it on
1723 * but have CONFIG_NET_CLS_ACT
1724 * NOTE: This doesnt stop any functionality; if you dont have
1725 * the ingress scheduler, you just cant add policies on ingress.
1728 int ing_filter(struct sk_buff *skb)
1731 struct net_device *dev = skb->dev;
1732 int result = TC_ACT_OK;
1734 if (dev->qdisc_ingress) {
1735 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1736 if (MAX_RED_LOOP < ttl++) {
1737 printk("Redir loop detected Dropping packet (%s->%s)\n",
1738 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1742 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1744 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1745 if (NULL == skb->input_dev) {
1746 skb->input_dev = skb->dev;
1747 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1749 spin_lock(&dev->ingress_lock);
1750 if ((q = dev->qdisc_ingress) != NULL)
1751 result = q->enqueue(skb, q);
1752 spin_unlock(&dev->ingress_lock);
1760 int netif_receive_skb(struct sk_buff *skb)
1762 struct packet_type *ptype, *pt_prev;
1763 int ret = NET_RX_DROP;
1764 unsigned short type;
1766 #ifdef CONFIG_NETPOLL_RX
1767 if (skb->dev->netpoll_rx && skb->dev->poll && netpoll_rx(skb)) {
1773 if (!skb->stamp.tv_sec)
1774 net_timestamp(&skb->stamp);
1778 __get_cpu_var(netdev_rx_stat).total++;
1780 skb->h.raw = skb->nh.raw = skb->data;
1781 skb->mac_len = skb->nh.raw - skb->mac.raw;
1784 #ifdef CONFIG_NET_CLS_ACT
1785 if (skb->tc_verd & TC_NCLS) {
1786 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1793 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1794 if (!ptype->dev || ptype->dev == skb->dev) {
1796 ret = deliver_skb(skb, pt_prev, 0);
1801 #ifdef CONFIG_NET_CLS_ACT
1803 atomic_inc(&skb->users);
1804 ret = pt_prev->func(skb, skb->dev, pt_prev);
1805 pt_prev = NULL; /* noone else should process this after*/
1807 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1810 ret = ing_filter(skb);
1812 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1821 handle_diverter(skb);
1823 if (__handle_bridge(skb, &pt_prev, &ret))
1826 type = skb->protocol;
1827 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1828 if (ptype->type == type &&
1829 (!ptype->dev || ptype->dev == skb->dev)) {
1831 ret = deliver_skb(skb, pt_prev, 0);
1837 ret = pt_prev->func(skb, skb->dev, pt_prev);
1840 /* Jamal, now you will not able to escape explaining
1841 * me how you were going to use this. :-)
1851 static int process_backlog(struct net_device *backlog_dev, int *budget)
1854 int quota = min(backlog_dev->quota, *budget);
1855 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1856 unsigned long start_time = jiffies;
1859 struct sk_buff *skb;
1860 struct net_device *dev;
1862 local_irq_disable();
1863 skb = __skb_dequeue(&queue->input_pkt_queue);
1870 netif_receive_skb(skb);
1876 if (work >= quota || jiffies - start_time > 1)
1879 #ifdef CONFIG_NET_HW_FLOWCONTROL
1880 if (queue->throttle &&
1881 queue->input_pkt_queue.qlen < no_cong_thresh ) {
1882 queue->throttle = 0;
1883 if (atomic_dec_and_test(&netdev_dropping)) {
1891 backlog_dev->quota -= work;
1896 backlog_dev->quota -= work;
1899 list_del(&backlog_dev->poll_list);
1900 smp_mb__before_clear_bit();
1901 netif_poll_enable(backlog_dev);
1903 if (queue->throttle) {
1904 queue->throttle = 0;
1905 #ifdef CONFIG_NET_HW_FLOWCONTROL
1906 if (atomic_dec_and_test(&netdev_dropping))
1914 static void net_rx_action(struct softirq_action *h)
1916 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1917 unsigned long start_time = jiffies;
1918 int budget = netdev_max_backlog;
1921 local_irq_disable();
1923 while (!list_empty(&queue->poll_list)) {
1924 struct net_device *dev;
1926 if (budget <= 0 || jiffies - start_time > 1)
1931 dev = list_entry(queue->poll_list.next,
1932 struct net_device, poll_list);
1934 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1935 local_irq_disable();
1936 list_del(&dev->poll_list);
1937 list_add_tail(&dev->poll_list, &queue->poll_list);
1939 dev->quota += dev->weight;
1941 dev->quota = dev->weight;
1944 local_irq_disable();
1952 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1953 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1957 static gifconf_func_t * gifconf_list [NPROTO];
1960 * register_gifconf - register a SIOCGIF handler
1961 * @family: Address family
1962 * @gifconf: Function handler
1964 * Register protocol dependent address dumping routines. The handler
1965 * that is passed must not be freed or reused until it has been replaced
1966 * by another handler.
1968 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1970 if (family >= NPROTO)
1972 gifconf_list[family] = gifconf;
1978 * Map an interface index to its name (SIOCGIFNAME)
1982 * We need this ioctl for efficient implementation of the
1983 * if_indextoname() function required by the IPv6 API. Without
1984 * it, we would have to search all the interfaces to find a
1988 static int dev_ifname(struct ifreq __user *arg)
1990 struct net_device *dev;
1994 * Fetch the caller's info block.
1997 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2000 read_lock(&dev_base_lock);
2001 dev = __dev_get_by_index(ifr.ifr_ifindex);
2003 read_unlock(&dev_base_lock);
2007 strcpy(ifr.ifr_name, dev->name);
2008 read_unlock(&dev_base_lock);
2010 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2016 * Perform a SIOCGIFCONF call. This structure will change
2017 * size eventually, and there is nothing I can do about it.
2018 * Thus we will need a 'compatibility mode'.
2021 static int dev_ifconf(char __user *arg)
2024 struct net_device *dev;
2031 * Fetch the caller's info block.
2034 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2041 * Loop over the interfaces, and write an info block for each.
2045 for (dev = dev_base; dev; dev = dev->next) {
2046 if (!dev_in_nx_info(dev, current->nx_info))
2048 for (i = 0; i < NPROTO; i++) {
2049 if (gifconf_list[i]) {
2052 done = gifconf_list[i](dev, NULL, 0);
2054 done = gifconf_list[i](dev, pos + total,
2064 * All done. Write the updated control block back to the caller.
2066 ifc.ifc_len = total;
2069 * Both BSD and Solaris return 0 here, so we do too.
2071 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2074 #ifdef CONFIG_PROC_FS
2076 * This is invoked by the /proc filesystem handler to display a device
2079 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2081 struct net_device *dev;
2084 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2086 return i == pos ? dev : NULL;
2089 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2091 read_lock(&dev_base_lock);
2092 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2095 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2098 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2101 void dev_seq_stop(struct seq_file *seq, void *v)
2103 read_unlock(&dev_base_lock);
2106 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2108 struct nx_info *nxi = current->nx_info;
2110 if (!dev_in_nx_info(dev, nxi))
2112 if (dev->get_stats) {
2113 struct net_device_stats *stats = dev->get_stats(dev);
2115 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2116 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2117 dev->name, stats->rx_bytes, stats->rx_packets,
2119 stats->rx_dropped + stats->rx_missed_errors,
2120 stats->rx_fifo_errors,
2121 stats->rx_length_errors + stats->rx_over_errors +
2122 stats->rx_crc_errors + stats->rx_frame_errors,
2123 stats->rx_compressed, stats->multicast,
2124 stats->tx_bytes, stats->tx_packets,
2125 stats->tx_errors, stats->tx_dropped,
2126 stats->tx_fifo_errors, stats->collisions,
2127 stats->tx_carrier_errors +
2128 stats->tx_aborted_errors +
2129 stats->tx_window_errors +
2130 stats->tx_heartbeat_errors,
2131 stats->tx_compressed);
2133 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2137 * Called from the PROCfs module. This now uses the new arbitrary sized
2138 * /proc/net interface to create /proc/net/dev
2140 static int dev_seq_show(struct seq_file *seq, void *v)
2142 if (v == SEQ_START_TOKEN)
2143 seq_puts(seq, "Inter-| Receive "
2145 " face |bytes packets errs drop fifo frame "
2146 "compressed multicast|bytes packets errs "
2147 "drop fifo colls carrier compressed\n");
2149 dev_seq_printf_stats(seq, v);
2153 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2155 struct netif_rx_stats *rc = NULL;
2157 while (*pos < NR_CPUS)
2158 if (cpu_online(*pos)) {
2159 rc = &per_cpu(netdev_rx_stat, *pos);
2166 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2168 return softnet_get_online(pos);
2171 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2174 return softnet_get_online(pos);
2177 static void softnet_seq_stop(struct seq_file *seq, void *v)
2181 static int softnet_seq_show(struct seq_file *seq, void *v)
2183 struct netif_rx_stats *s = v;
2185 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2186 s->total, s->dropped, s->time_squeeze, s->throttled,
2187 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2188 s->fastroute_deferred_out,
2190 s->fastroute_latency_reduction
2198 static struct seq_operations dev_seq_ops = {
2199 .start = dev_seq_start,
2200 .next = dev_seq_next,
2201 .stop = dev_seq_stop,
2202 .show = dev_seq_show,
2205 static int dev_seq_open(struct inode *inode, struct file *file)
2207 return seq_open(file, &dev_seq_ops);
2210 static struct file_operations dev_seq_fops = {
2211 .owner = THIS_MODULE,
2212 .open = dev_seq_open,
2214 .llseek = seq_lseek,
2215 .release = seq_release,
2218 static struct seq_operations softnet_seq_ops = {
2219 .start = softnet_seq_start,
2220 .next = softnet_seq_next,
2221 .stop = softnet_seq_stop,
2222 .show = softnet_seq_show,
2225 static int softnet_seq_open(struct inode *inode, struct file *file)
2227 return seq_open(file, &softnet_seq_ops);
2230 static struct file_operations softnet_seq_fops = {
2231 .owner = THIS_MODULE,
2232 .open = softnet_seq_open,
2234 .llseek = seq_lseek,
2235 .release = seq_release,
2239 extern int wireless_proc_init(void);
2241 #define wireless_proc_init() 0
2244 static int __init dev_proc_init(void)
2248 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2250 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2252 if (wireless_proc_init())
2258 proc_net_remove("softnet_stat");
2260 proc_net_remove("dev");
2264 #define dev_proc_init() 0
2265 #endif /* CONFIG_PROC_FS */
2269 * netdev_set_master - set up master/slave pair
2270 * @slave: slave device
2271 * @master: new master device
2273 * Changes the master device of the slave. Pass %NULL to break the
2274 * bonding. The caller must hold the RTNL semaphore. On a failure
2275 * a negative errno code is returned. On success the reference counts
2276 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2277 * function returns zero.
2279 int netdev_set_master(struct net_device *slave, struct net_device *master)
2281 struct net_device *old = slave->master;
2291 slave->master = master;
2299 slave->flags |= IFF_SLAVE;
2301 slave->flags &= ~IFF_SLAVE;
2303 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2308 * dev_set_promiscuity - update promiscuity count on a device
2312 * Add or remove promsicuity from a device. While the count in the device
2313 * remains above zero the interface remains promiscuous. Once it hits zero
2314 * the device reverts back to normal filtering operation. A negative inc
2315 * value is used to drop promiscuity on the device.
2317 void dev_set_promiscuity(struct net_device *dev, int inc)
2319 unsigned short old_flags = dev->flags;
2321 dev->flags |= IFF_PROMISC;
2322 if ((dev->promiscuity += inc) == 0)
2323 dev->flags &= ~IFF_PROMISC;
2324 if (dev->flags ^ old_flags) {
2326 printk(KERN_INFO "device %s %s promiscuous mode\n",
2327 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2333 * dev_set_allmulti - update allmulti count on a device
2337 * Add or remove reception of all multicast frames to a device. While the
2338 * count in the device remains above zero the interface remains listening
2339 * to all interfaces. Once it hits zero the device reverts back to normal
2340 * filtering operation. A negative @inc value is used to drop the counter
2341 * when releasing a resource needing all multicasts.
2344 void dev_set_allmulti(struct net_device *dev, int inc)
2346 unsigned short old_flags = dev->flags;
2348 dev->flags |= IFF_ALLMULTI;
2349 if ((dev->allmulti += inc) == 0)
2350 dev->flags &= ~IFF_ALLMULTI;
2351 if (dev->flags ^ old_flags)
2355 unsigned dev_get_flags(const struct net_device *dev)
2359 flags = (dev->flags & ~(IFF_PROMISC |
2362 (dev->gflags & (IFF_PROMISC |
2365 if (netif_running(dev) && netif_carrier_ok(dev))
2366 flags |= IFF_RUNNING;
2371 int dev_change_flags(struct net_device *dev, unsigned flags)
2374 int old_flags = dev->flags;
2377 * Set the flags on our device.
2380 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2381 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2383 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2387 * Load in the correct multicast list now the flags have changed.
2393 * Have we downed the interface. We handle IFF_UP ourselves
2394 * according to user attempts to set it, rather than blindly
2399 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2400 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2406 if (dev->flags & IFF_UP &&
2407 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2409 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2411 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2412 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2413 dev->gflags ^= IFF_PROMISC;
2414 dev_set_promiscuity(dev, inc);
2417 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2418 is important. Some (broken) drivers set IFF_PROMISC, when
2419 IFF_ALLMULTI is requested not asking us and not reporting.
2421 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2422 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2423 dev->gflags ^= IFF_ALLMULTI;
2424 dev_set_allmulti(dev, inc);
2427 if (old_flags ^ dev->flags)
2428 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2433 int dev_set_mtu(struct net_device *dev, int new_mtu)
2437 if (new_mtu == dev->mtu)
2440 /* MTU must be positive. */
2444 if (!netif_device_present(dev))
2448 if (dev->change_mtu)
2449 err = dev->change_mtu(dev, new_mtu);
2452 if (!err && dev->flags & IFF_UP)
2453 notifier_call_chain(&netdev_chain,
2454 NETDEV_CHANGEMTU, dev);
2460 * Perform the SIOCxIFxxx calls.
2462 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2465 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2471 case SIOCGIFFLAGS: /* Get interface flags */
2472 ifr->ifr_flags = dev_get_flags(dev);
2475 case SIOCSIFFLAGS: /* Set interface flags */
2476 return dev_change_flags(dev, ifr->ifr_flags);
2478 case SIOCGIFMETRIC: /* Get the metric on the interface
2479 (currently unused) */
2480 ifr->ifr_metric = 0;
2483 case SIOCSIFMETRIC: /* Set the metric on the interface
2484 (currently unused) */
2487 case SIOCGIFMTU: /* Get the MTU of a device */
2488 ifr->ifr_mtu = dev->mtu;
2491 case SIOCSIFMTU: /* Set the MTU of a device */
2492 return dev_set_mtu(dev, ifr->ifr_mtu);
2495 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2496 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2497 ifr->ifr_hwaddr.sa_family = dev->type;
2501 if (!dev->set_mac_address)
2503 if (ifr->ifr_hwaddr.sa_family != dev->type)
2505 if (!netif_device_present(dev))
2507 err = dev->set_mac_address(dev, &ifr->ifr_hwaddr);
2509 notifier_call_chain(&netdev_chain,
2510 NETDEV_CHANGEADDR, dev);
2513 case SIOCSIFHWBROADCAST:
2514 if (ifr->ifr_hwaddr.sa_family != dev->type)
2516 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2517 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2518 notifier_call_chain(&netdev_chain,
2519 NETDEV_CHANGEADDR, dev);
2523 ifr->ifr_map.mem_start = dev->mem_start;
2524 ifr->ifr_map.mem_end = dev->mem_end;
2525 ifr->ifr_map.base_addr = dev->base_addr;
2526 ifr->ifr_map.irq = dev->irq;
2527 ifr->ifr_map.dma = dev->dma;
2528 ifr->ifr_map.port = dev->if_port;
2532 if (dev->set_config) {
2533 if (!netif_device_present(dev))
2535 return dev->set_config(dev, &ifr->ifr_map);
2540 if (!dev->set_multicast_list ||
2541 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2543 if (!netif_device_present(dev))
2545 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2549 if (!dev->set_multicast_list ||
2550 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2552 if (!netif_device_present(dev))
2554 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2558 ifr->ifr_ifindex = dev->ifindex;
2562 ifr->ifr_qlen = dev->tx_queue_len;
2566 if (ifr->ifr_qlen < 0)
2568 dev->tx_queue_len = ifr->ifr_qlen;
2572 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2573 return dev_change_name(dev, ifr->ifr_newname);
2576 * Unknown or private ioctl
2580 if ((cmd >= SIOCDEVPRIVATE &&
2581 cmd <= SIOCDEVPRIVATE + 15) ||
2582 cmd == SIOCBONDENSLAVE ||
2583 cmd == SIOCBONDRELEASE ||
2584 cmd == SIOCBONDSETHWADDR ||
2585 cmd == SIOCBONDSLAVEINFOQUERY ||
2586 cmd == SIOCBONDINFOQUERY ||
2587 cmd == SIOCBONDCHANGEACTIVE ||
2588 cmd == SIOCGMIIPHY ||
2589 cmd == SIOCGMIIREG ||
2590 cmd == SIOCSMIIREG ||
2591 cmd == SIOCBRADDIF ||
2592 cmd == SIOCBRDELIF ||
2593 cmd == SIOCWANDEV) {
2595 if (dev->do_ioctl) {
2596 if (netif_device_present(dev))
2597 err = dev->do_ioctl(dev, ifr,
2610 * This function handles all "interface"-type I/O control requests. The actual
2611 * 'doing' part of this is dev_ifsioc above.
2615 * dev_ioctl - network device ioctl
2616 * @cmd: command to issue
2617 * @arg: pointer to a struct ifreq in user space
2619 * Issue ioctl functions to devices. This is normally called by the
2620 * user space syscall interfaces but can sometimes be useful for
2621 * other purposes. The return value is the return from the syscall if
2622 * positive or a negative errno code on error.
2625 int dev_ioctl(unsigned int cmd, void __user *arg)
2631 /* One special case: SIOCGIFCONF takes ifconf argument
2632 and requires shared lock, because it sleeps writing
2636 if (cmd == SIOCGIFCONF) {
2638 ret = dev_ifconf((char __user *) arg);
2642 if (cmd == SIOCGIFNAME)
2643 return dev_ifname((struct ifreq __user *)arg);
2645 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2648 ifr.ifr_name[IFNAMSIZ-1] = 0;
2650 colon = strchr(ifr.ifr_name, ':');
2655 * See which interface the caller is talking about.
2660 * These ioctl calls:
2661 * - can be done by all.
2662 * - atomic and do not require locking.
2673 dev_load(ifr.ifr_name);
2674 read_lock(&dev_base_lock);
2675 ret = dev_ifsioc(&ifr, cmd);
2676 read_unlock(&dev_base_lock);
2680 if (copy_to_user(arg, &ifr,
2681 sizeof(struct ifreq)))
2687 dev_load(ifr.ifr_name);
2689 ret = dev_ethtool(&ifr);
2694 if (copy_to_user(arg, &ifr,
2695 sizeof(struct ifreq)))
2701 * These ioctl calls:
2702 * - require superuser power.
2703 * - require strict serialization.
2709 if (!capable(CAP_NET_ADMIN))
2711 dev_load(ifr.ifr_name);
2713 ret = dev_ifsioc(&ifr, cmd);
2718 if (copy_to_user(arg, &ifr,
2719 sizeof(struct ifreq)))
2725 * These ioctl calls:
2726 * - require superuser power.
2727 * - require strict serialization.
2728 * - do not return a value
2738 case SIOCSIFHWBROADCAST:
2741 case SIOCBONDENSLAVE:
2742 case SIOCBONDRELEASE:
2743 case SIOCBONDSETHWADDR:
2744 case SIOCBONDSLAVEINFOQUERY:
2745 case SIOCBONDINFOQUERY:
2746 case SIOCBONDCHANGEACTIVE:
2749 if (!capable(CAP_NET_ADMIN))
2751 dev_load(ifr.ifr_name);
2753 ret = dev_ifsioc(&ifr, cmd);
2758 /* Get the per device memory space. We can add this but
2759 * currently do not support it */
2761 /* Set the per device memory buffer space.
2762 * Not applicable in our case */
2767 * Unknown or private ioctl.
2770 if (cmd == SIOCWANDEV ||
2771 (cmd >= SIOCDEVPRIVATE &&
2772 cmd <= SIOCDEVPRIVATE + 15)) {
2773 dev_load(ifr.ifr_name);
2775 ret = dev_ifsioc(&ifr, cmd);
2777 if (!ret && copy_to_user(arg, &ifr,
2778 sizeof(struct ifreq)))
2783 /* Take care of Wireless Extensions */
2784 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2785 /* If command is `set a parameter', or
2786 * `get the encoding parameters', check if
2787 * the user has the right to do it */
2788 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2789 if (!capable(CAP_NET_ADMIN))
2792 dev_load(ifr.ifr_name);
2794 /* Follow me in net/core/wireless.c */
2795 ret = wireless_process_ioctl(&ifr, cmd);
2797 if (!ret && IW_IS_GET(cmd) &&
2798 copy_to_user(arg, &ifr,
2799 sizeof(struct ifreq)))
2803 #endif /* WIRELESS_EXT */
2810 * dev_new_index - allocate an ifindex
2812 * Returns a suitable unique value for a new device interface
2813 * number. The caller must hold the rtnl semaphore or the
2814 * dev_base_lock to be sure it remains unique.
2816 int dev_new_index(void)
2822 if (!__dev_get_by_index(ifindex))
2827 static int dev_boot_phase = 1;
2829 /* Delayed registration/unregisteration */
2830 static spinlock_t net_todo_list_lock = SPIN_LOCK_UNLOCKED;
2831 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2833 static inline void net_set_todo(struct net_device *dev)
2835 spin_lock(&net_todo_list_lock);
2836 list_add_tail(&dev->todo_list, &net_todo_list);
2837 spin_unlock(&net_todo_list_lock);
2841 * register_netdevice - register a network device
2842 * @dev: device to register
2844 * Take a completed network device structure and add it to the kernel
2845 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2846 * chain. 0 is returned on success. A negative errno code is returned
2847 * on a failure to set up the device, or if the name is a duplicate.
2849 * Callers must hold the rtnl semaphore. See the comment at the
2850 * end of Space.c for details about the locking. You may want
2851 * register_netdev() instead of this.
2854 * The locking appears insufficient to guarantee two parallel registers
2855 * will not get the same name.
2858 int register_netdevice(struct net_device *dev)
2860 struct hlist_head *head;
2861 struct hlist_node *p;
2864 BUG_ON(dev_boot_phase);
2867 /* When net_device's are persistent, this will be fatal. */
2868 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2870 spin_lock_init(&dev->queue_lock);
2871 spin_lock_init(&dev->xmit_lock);
2872 dev->xmit_lock_owner = -1;
2873 #ifdef CONFIG_NET_CLS_ACT
2874 spin_lock_init(&dev->ingress_lock);
2877 ret = alloc_divert_blk(dev);
2883 /* Init, if this function is available */
2885 ret = dev->init(dev);
2893 if (!dev_valid_name(dev->name)) {
2898 dev->ifindex = dev_new_index();
2899 if (dev->iflink == -1)
2900 dev->iflink = dev->ifindex;
2902 /* Check for existence of name */
2903 head = dev_name_hash(dev->name);
2904 hlist_for_each(p, head) {
2905 struct net_device *d
2906 = hlist_entry(p, struct net_device, name_hlist);
2907 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2913 /* Fix illegal SG+CSUM combinations. */
2914 if ((dev->features & NETIF_F_SG) &&
2915 !(dev->features & (NETIF_F_IP_CSUM |
2917 NETIF_F_HW_CSUM))) {
2918 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2920 dev->features &= ~NETIF_F_SG;
2924 * nil rebuild_header routine,
2925 * that should be never called and used as just bug trap.
2928 if (!dev->rebuild_header)
2929 dev->rebuild_header = default_rebuild_header;
2932 * Default initial state at registry is that the
2933 * device is present.
2936 set_bit(__LINK_STATE_PRESENT, &dev->state);
2939 dev_init_scheduler(dev);
2940 write_lock_bh(&dev_base_lock);
2942 dev_tail = &dev->next;
2943 hlist_add_head(&dev->name_hlist, head);
2944 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2946 dev->reg_state = NETREG_REGISTERING;
2947 write_unlock_bh(&dev_base_lock);
2949 /* Notify protocols, that a new device appeared. */
2950 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2952 /* Finish registration after unlock */
2959 free_divert_blk(dev);
2964 * netdev_wait_allrefs - wait until all references are gone.
2966 * This is called when unregistering network devices.
2968 * Any protocol or device that holds a reference should register
2969 * for netdevice notification, and cleanup and put back the
2970 * reference if they receive an UNREGISTER event.
2971 * We can get stuck here if buggy protocols don't correctly
2974 static void netdev_wait_allrefs(struct net_device *dev)
2976 unsigned long rebroadcast_time, warning_time;
2978 rebroadcast_time = warning_time = jiffies;
2979 while (atomic_read(&dev->refcnt) != 0) {
2980 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2983 /* Rebroadcast unregister notification */
2984 notifier_call_chain(&netdev_chain,
2985 NETDEV_UNREGISTER, dev);
2987 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2989 /* We must not have linkwatch events
2990 * pending on unregister. If this
2991 * happens, we simply run the queue
2992 * unscheduled, resulting in a noop
2995 linkwatch_run_queue();
3000 rebroadcast_time = jiffies;
3003 current->state = TASK_INTERRUPTIBLE;
3004 schedule_timeout(HZ / 4);
3006 if (time_after(jiffies, warning_time + 10 * HZ)) {
3007 printk(KERN_EMERG "unregister_netdevice: "
3008 "waiting for %s to become free. Usage "
3010 dev->name, atomic_read(&dev->refcnt));
3011 warning_time = jiffies;
3020 * register_netdevice(x1);
3021 * register_netdevice(x2);
3023 * unregister_netdevice(y1);
3024 * unregister_netdevice(y2);
3030 * We are invoked by rtnl_unlock() after it drops the semaphore.
3031 * This allows us to deal with problems:
3032 * 1) We can create/delete sysfs objects which invoke hotplug
3033 * without deadlocking with linkwatch via keventd.
3034 * 2) Since we run with the RTNL semaphore not held, we can sleep
3035 * safely in order to wait for the netdev refcnt to drop to zero.
3037 static DECLARE_MUTEX(net_todo_run_mutex);
3038 void netdev_run_todo(void)
3040 struct list_head list = LIST_HEAD_INIT(list);
3044 /* Need to guard against multiple cpu's getting out of order. */
3045 down(&net_todo_run_mutex);
3047 /* Not safe to do outside the semaphore. We must not return
3048 * until all unregister events invoked by the local processor
3049 * have been completed (either by this todo run, or one on
3052 if (list_empty(&net_todo_list))
3055 /* Snapshot list, allow later requests */
3056 spin_lock(&net_todo_list_lock);
3057 list_splice_init(&net_todo_list, &list);
3058 spin_unlock(&net_todo_list_lock);
3060 while (!list_empty(&list)) {
3061 struct net_device *dev
3062 = list_entry(list.next, struct net_device, todo_list);
3063 list_del(&dev->todo_list);
3065 switch(dev->reg_state) {
3066 case NETREG_REGISTERING:
3067 err = netdev_register_sysfs(dev);
3069 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3071 dev->reg_state = NETREG_REGISTERED;
3074 case NETREG_UNREGISTERING:
3075 netdev_unregister_sysfs(dev);
3076 dev->reg_state = NETREG_UNREGISTERED;
3078 netdev_wait_allrefs(dev);
3081 BUG_ON(atomic_read(&dev->refcnt));
3082 BUG_TRAP(!dev->ip_ptr);
3083 BUG_TRAP(!dev->ip6_ptr);
3084 BUG_TRAP(!dev->dn_ptr);
3087 /* It must be the very last action,
3088 * after this 'dev' may point to freed up memory.
3090 if (dev->destructor)
3091 dev->destructor(dev);
3095 printk(KERN_ERR "network todo '%s' but state %d\n",
3096 dev->name, dev->reg_state);
3102 up(&net_todo_run_mutex);
3106 * free_netdev - free network device
3109 * This function does the last stage of destroying an allocated device
3110 * interface. The reference to the device object is released.
3111 * If this is the last reference then it will be freed.
3113 void free_netdev(struct net_device *dev)
3116 /* Compatiablity with error handling in drivers */
3117 if (dev->reg_state == NETREG_UNINITIALIZED) {
3118 kfree((char *)dev - dev->padded);
3122 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3123 dev->reg_state = NETREG_RELEASED;
3125 /* will free via class release */
3126 class_device_put(&dev->class_dev);
3128 kfree((char *)dev - dev->padded);
3132 /* Synchronize with packet receive processing. */
3133 void synchronize_net(void)
3136 synchronize_kernel();
3140 * unregister_netdevice - remove device from the kernel
3143 * This function shuts down a device interface and removes it
3144 * from the kernel tables. On success 0 is returned, on a failure
3145 * a negative errno code is returned.
3147 * Callers must hold the rtnl semaphore. See the comment at the
3148 * end of Space.c for details about the locking. You may want
3149 * unregister_netdev() instead of this.
3152 int unregister_netdevice(struct net_device *dev)
3154 struct net_device *d, **dp;
3156 BUG_ON(dev_boot_phase);
3159 /* Some devices call without registering for initialization unwind. */
3160 if (dev->reg_state == NETREG_UNINITIALIZED) {
3161 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3162 "was registered\n", dev->name, dev);
3166 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3168 /* If device is running, close it first. */
3169 if (dev->flags & IFF_UP)
3172 /* And unlink it from device chain. */
3173 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3175 write_lock_bh(&dev_base_lock);
3176 hlist_del(&dev->name_hlist);
3177 hlist_del(&dev->index_hlist);
3178 if (dev_tail == &dev->next)
3181 write_unlock_bh(&dev_base_lock);
3186 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3191 dev->reg_state = NETREG_UNREGISTERING;
3195 /* Shutdown queueing discipline. */
3199 /* Notify protocols, that we are about to destroy
3200 this device. They should clean all the things.
3202 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3205 * Flush the multicast chain
3207 dev_mc_discard(dev);
3212 /* Notifier chain MUST detach us from master device. */
3213 BUG_TRAP(!dev->master);
3215 free_divert_blk(dev);
3217 /* Finish processing unregister after unlock */
3226 #ifdef CONFIG_HOTPLUG_CPU
3227 static int dev_cpu_callback(struct notifier_block *nfb,
3228 unsigned long action,
3231 struct sk_buff **list_skb;
3232 struct net_device **list_net;
3233 struct sk_buff *skb;
3234 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3235 struct softnet_data *sd, *oldsd;
3237 if (action != CPU_DEAD)
3240 local_irq_disable();
3241 cpu = smp_processor_id();
3242 sd = &per_cpu(softnet_data, cpu);
3243 oldsd = &per_cpu(softnet_data, oldcpu);
3245 /* Find end of our completion_queue. */
3246 list_skb = &sd->completion_queue;
3248 list_skb = &(*list_skb)->next;
3249 /* Append completion queue from offline CPU. */
3250 *list_skb = oldsd->completion_queue;
3251 oldsd->completion_queue = NULL;
3253 /* Find end of our output_queue. */
3254 list_net = &sd->output_queue;
3256 list_net = &(*list_net)->next_sched;
3257 /* Append output queue from offline CPU. */
3258 *list_net = oldsd->output_queue;
3259 oldsd->output_queue = NULL;
3261 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3264 /* Process offline CPU's input_pkt_queue */
3265 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3270 #endif /* CONFIG_HOTPLUG_CPU */
3274 * Initialize the DEV module. At boot time this walks the device list and
3275 * unhooks any devices that fail to initialise (normally hardware not
3276 * present) and leaves us with a valid list of present and active devices.
3281 * This is called single threaded during boot, so no need
3282 * to take the rtnl semaphore.
3284 static int __init net_dev_init(void)
3286 int i, rc = -ENOMEM;
3288 BUG_ON(!dev_boot_phase);
3290 if (dev_proc_init())
3293 if (netdev_sysfs_init())
3296 INIT_LIST_HEAD(&ptype_all);
3297 for (i = 0; i < 16; i++)
3298 INIT_LIST_HEAD(&ptype_base[i]);
3300 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3301 INIT_HLIST_HEAD(&dev_name_head[i]);
3303 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3304 INIT_HLIST_HEAD(&dev_index_head[i]);
3307 * Initialise the packet receive queues.
3310 for (i = 0; i < NR_CPUS; i++) {
3311 struct softnet_data *queue;
3313 queue = &per_cpu(softnet_data, i);
3314 skb_queue_head_init(&queue->input_pkt_queue);
3315 queue->throttle = 0;
3316 queue->cng_level = 0;
3317 queue->avg_blog = 10; /* arbitrary non-zero */
3318 queue->completion_queue = NULL;
3319 INIT_LIST_HEAD(&queue->poll_list);
3320 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3321 queue->backlog_dev.weight = weight_p;
3322 queue->backlog_dev.poll = process_backlog;
3323 atomic_set(&queue->backlog_dev.refcnt, 1);
3326 #ifdef OFFLINE_SAMPLE
3327 samp_timer.expires = jiffies + (10 * HZ);
3328 add_timer(&samp_timer);
3333 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3334 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3336 hotcpu_notifier(dev_cpu_callback, 0);
3344 subsys_initcall(net_dev_init);
3346 EXPORT_SYMBOL(__dev_get);
3347 EXPORT_SYMBOL(__dev_get_by_flags);
3348 EXPORT_SYMBOL(__dev_get_by_index);
3349 EXPORT_SYMBOL(__dev_get_by_name);
3350 EXPORT_SYMBOL(__dev_remove_pack);
3351 EXPORT_SYMBOL(__skb_linearize);
3352 EXPORT_SYMBOL(call_netdevice_notifiers);
3353 EXPORT_SYMBOL(dev_add_pack);
3354 EXPORT_SYMBOL(dev_alloc_name);
3355 EXPORT_SYMBOL(dev_close);
3356 EXPORT_SYMBOL(dev_get_by_flags);
3357 EXPORT_SYMBOL(dev_get_by_index);
3358 EXPORT_SYMBOL(dev_get_by_name);
3359 EXPORT_SYMBOL(dev_getbyhwaddr);
3360 EXPORT_SYMBOL(dev_ioctl);
3361 EXPORT_SYMBOL(dev_new_index);
3362 EXPORT_SYMBOL(dev_open);
3363 EXPORT_SYMBOL(dev_queue_xmit);
3364 EXPORT_SYMBOL(dev_queue_xmit_nit);
3365 EXPORT_SYMBOL(dev_remove_pack);
3366 EXPORT_SYMBOL(dev_set_allmulti);
3367 EXPORT_SYMBOL(dev_set_promiscuity);
3368 EXPORT_SYMBOL(dev_change_flags);
3369 EXPORT_SYMBOL(dev_set_mtu);
3370 EXPORT_SYMBOL(free_netdev);
3371 EXPORT_SYMBOL(netdev_boot_setup_check);
3372 EXPORT_SYMBOL(netdev_set_master);
3373 EXPORT_SYMBOL(netdev_state_change);
3374 EXPORT_SYMBOL(netif_receive_skb);
3375 EXPORT_SYMBOL(netif_rx);
3376 EXPORT_SYMBOL(register_gifconf);
3377 EXPORT_SYMBOL(register_netdevice);
3378 EXPORT_SYMBOL(register_netdevice_notifier);
3379 EXPORT_SYMBOL(skb_checksum_help);
3380 EXPORT_SYMBOL(synchronize_net);
3381 EXPORT_SYMBOL(unregister_netdevice);
3382 EXPORT_SYMBOL(unregister_netdevice_notifier);
3384 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3385 EXPORT_SYMBOL(br_handle_frame_hook);
3389 EXPORT_SYMBOL(dev_load);
3391 #ifdef CONFIG_NET_HW_FLOWCONTROL
3392 EXPORT_SYMBOL(netdev_dropping);
3393 EXPORT_SYMBOL(netdev_fc_xoff);
3394 EXPORT_SYMBOL(netdev_register_fc);
3395 EXPORT_SYMBOL(netdev_unregister_fc);
3398 #ifdef CONFIG_NET_CLS_ACT
3399 EXPORT_SYMBOL(ing_filter);
3403 EXPORT_PER_CPU_SYMBOL(softnet_data);