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
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/mutex.h>
84 #include <linux/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/notifier.h>
94 #include <linux/skbuff.h>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <linux/wireless.h>
113 #include <net/iw_handler.h>
114 #include <asm/current.h>
115 #include <linux/audit.h>
116 #include <linux/dmaengine.h>
117 #include <linux/err.h>
118 #include <linux/ctype.h>
121 * The list of packet types we will receive (as opposed to discard)
122 * and the routines to invoke.
124 * Why 16. Because with 16 the only overlap we get on a hash of the
125 * low nibble of the protocol value is RARP/SNAP/X.25.
127 * NOTE: That is no longer true with the addition of VLAN tags. Not
128 * sure which should go first, but I bet it won't make much
129 * difference if we are running VLANs. The good news is that
130 * this protocol won't be in the list unless compiled in, so
131 * the average user (w/out VLANs) will not be adversely affected.
148 static DEFINE_SPINLOCK(ptype_lock);
149 static struct list_head ptype_base[16] __read_mostly; /* 16 way hashed list */
150 static struct list_head ptype_all __read_mostly; /* Taps */
152 #ifdef CONFIG_NET_DMA
153 static struct dma_client *net_dma_client;
154 static unsigned int net_dma_count;
155 static spinlock_t net_dma_event_lock;
159 * The @dev_base list is protected by @dev_base_lock and the rtnl
162 * Pure readers hold dev_base_lock for reading.
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
177 struct net_device *dev_base;
178 static struct net_device **dev_tail = &dev_base;
179 DEFINE_RWLOCK(dev_base_lock);
181 EXPORT_SYMBOL(dev_base);
182 EXPORT_SYMBOL(dev_base_lock);
184 #define NETDEV_HASHBITS 8
185 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
186 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
188 static inline struct hlist_head *dev_name_hash(const char *name)
190 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
194 static inline struct hlist_head *dev_index_hash(int ifindex)
196 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
203 static RAW_NOTIFIER_HEAD(netdev_chain);
206 * Device drivers call our routines to queue packets here. We empty the
207 * queue in the local softnet handler.
209 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
212 extern int netdev_sysfs_init(void);
213 extern int netdev_register_sysfs(struct net_device *);
214 extern void netdev_unregister_sysfs(struct net_device *);
216 #define netdev_sysfs_init() (0)
217 #define netdev_register_sysfs(dev) (0)
218 #define netdev_unregister_sysfs(dev) do { } while(0)
222 /*******************************************************************************
224 Protocol management and registration routines
226 *******************************************************************************/
232 static int netdev_nit;
235 * Add a protocol ID to the list. Now that the input handler is
236 * smarter we can dispense with all the messy stuff that used to be
239 * BEWARE!!! Protocol handlers, mangling input packets,
240 * MUST BE last in hash buckets and checking protocol handlers
241 * MUST start from promiscuous ptype_all chain in net_bh.
242 * It is true now, do not change it.
243 * Explanation follows: if protocol handler, mangling packet, will
244 * be the first on list, it is not able to sense, that packet
245 * is cloned and should be copied-on-write, so that it will
246 * change it and subsequent readers will get broken packet.
251 * dev_add_pack - add packet handler
252 * @pt: packet type declaration
254 * Add a protocol handler to the networking stack. The passed &packet_type
255 * is linked into kernel lists and may not be freed until it has been
256 * removed from the kernel lists.
258 * This call does not sleep therefore it can not
259 * guarantee all CPU's that are in middle of receiving packets
260 * will see the new packet type (until the next received packet).
263 void dev_add_pack(struct packet_type *pt)
267 spin_lock_bh(&ptype_lock);
268 if (pt->type == htons(ETH_P_ALL)) {
270 list_add_rcu(&pt->list, &ptype_all);
272 hash = ntohs(pt->type) & 15;
273 list_add_rcu(&pt->list, &ptype_base[hash]);
275 spin_unlock_bh(&ptype_lock);
279 * __dev_remove_pack - remove packet handler
280 * @pt: packet type declaration
282 * Remove a protocol handler that was previously added to the kernel
283 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
284 * from the kernel lists and can be freed or reused once this function
287 * The packet type might still be in use by receivers
288 * and must not be freed until after all the CPU's have gone
289 * through a quiescent state.
291 void __dev_remove_pack(struct packet_type *pt)
293 struct list_head *head;
294 struct packet_type *pt1;
296 spin_lock_bh(&ptype_lock);
298 if (pt->type == htons(ETH_P_ALL)) {
302 head = &ptype_base[ntohs(pt->type) & 15];
304 list_for_each_entry(pt1, head, list) {
306 list_del_rcu(&pt->list);
311 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
313 spin_unlock_bh(&ptype_lock);
316 * dev_remove_pack - remove packet handler
317 * @pt: packet type declaration
319 * Remove a protocol handler that was previously added to the kernel
320 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
321 * from the kernel lists and can be freed or reused once this function
324 * This call sleeps to guarantee that no CPU is looking at the packet
327 void dev_remove_pack(struct packet_type *pt)
329 __dev_remove_pack(pt);
334 /******************************************************************************
336 Device Boot-time Settings Routines
338 *******************************************************************************/
340 /* Boot time configuration table */
341 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
344 * netdev_boot_setup_add - add new setup entry
345 * @name: name of the device
346 * @map: configured settings for the device
348 * Adds new setup entry to the dev_boot_setup list. The function
349 * returns 0 on error and 1 on success. This is a generic routine to
352 static int netdev_boot_setup_add(char *name, struct ifmap *map)
354 struct netdev_boot_setup *s;
358 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
359 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
360 memset(s[i].name, 0, sizeof(s[i].name));
361 strcpy(s[i].name, name);
362 memcpy(&s[i].map, map, sizeof(s[i].map));
367 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
371 * netdev_boot_setup_check - check boot time settings
372 * @dev: the netdevice
374 * Check boot time settings for the device.
375 * The found settings are set for the device to be used
376 * later in the device probing.
377 * Returns 0 if no settings found, 1 if they are.
379 int netdev_boot_setup_check(struct net_device *dev)
381 struct netdev_boot_setup *s = dev_boot_setup;
384 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
385 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
386 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
387 dev->irq = s[i].map.irq;
388 dev->base_addr = s[i].map.base_addr;
389 dev->mem_start = s[i].map.mem_start;
390 dev->mem_end = s[i].map.mem_end;
399 * netdev_boot_base - get address from boot time settings
400 * @prefix: prefix for network device
401 * @unit: id for network device
403 * Check boot time settings for the base address of device.
404 * The found settings are set for the device to be used
405 * later in the device probing.
406 * Returns 0 if no settings found.
408 unsigned long netdev_boot_base(const char *prefix, int unit)
410 const struct netdev_boot_setup *s = dev_boot_setup;
414 sprintf(name, "%s%d", prefix, unit);
417 * If device already registered then return base of 1
418 * to indicate not to probe for this interface
420 if (__dev_get_by_name(name))
423 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
424 if (!strcmp(name, s[i].name))
425 return s[i].map.base_addr;
430 * Saves at boot time configured settings for any netdevice.
432 int __init netdev_boot_setup(char *str)
437 str = get_options(str, ARRAY_SIZE(ints), ints);
442 memset(&map, 0, sizeof(map));
446 map.base_addr = ints[2];
448 map.mem_start = ints[3];
450 map.mem_end = ints[4];
452 /* Add new entry to the list */
453 return netdev_boot_setup_add(str, &map);
456 __setup("netdev=", netdev_boot_setup);
458 /*******************************************************************************
460 Device Interface Subroutines
462 *******************************************************************************/
465 * __dev_get_by_name - find a device by its name
466 * @name: name to find
468 * Find an interface by name. Must be called under RTNL semaphore
469 * or @dev_base_lock. If the name is found a pointer to the device
470 * is returned. If the name is not found then %NULL is returned. The
471 * reference counters are not incremented so the caller must be
472 * careful with locks.
475 struct net_device *__dev_get_by_name(const char *name)
477 struct hlist_node *p;
479 hlist_for_each(p, dev_name_hash(name)) {
480 struct net_device *dev
481 = hlist_entry(p, struct net_device, name_hlist);
482 if (!strncmp(dev->name, name, IFNAMSIZ))
489 * dev_get_by_name - find a device by its name
490 * @name: name to find
492 * Find an interface by name. This can be called from any
493 * context and does its own locking. The returned handle has
494 * the usage count incremented and the caller must use dev_put() to
495 * release it when it is no longer needed. %NULL is returned if no
496 * matching device is found.
499 struct net_device *dev_get_by_name(const char *name)
501 struct net_device *dev;
503 read_lock(&dev_base_lock);
504 dev = __dev_get_by_name(name);
507 read_unlock(&dev_base_lock);
512 * __dev_get_by_index - find a device by its ifindex
513 * @ifindex: index of device
515 * Search for an interface by index. Returns %NULL if the device
516 * is not found or a pointer to the device. The device has not
517 * had its reference counter increased so the caller must be careful
518 * about locking. The caller must hold either the RTNL semaphore
522 struct net_device *__dev_get_by_index(int ifindex)
524 struct hlist_node *p;
526 hlist_for_each(p, dev_index_hash(ifindex)) {
527 struct net_device *dev
528 = hlist_entry(p, struct net_device, index_hlist);
529 if (dev->ifindex == ifindex)
537 * dev_get_by_index - find a device by its ifindex
538 * @ifindex: index of device
540 * Search for an interface by index. Returns NULL if the device
541 * is not found or a pointer to the device. The device returned has
542 * had a reference added and the pointer is safe until the user calls
543 * dev_put to indicate they have finished with it.
546 struct net_device *dev_get_by_index(int ifindex)
548 struct net_device *dev;
550 read_lock(&dev_base_lock);
551 dev = __dev_get_by_index(ifindex);
554 read_unlock(&dev_base_lock);
559 * dev_getbyhwaddr - find a device by its hardware address
560 * @type: media type of device
561 * @ha: hardware address
563 * Search for an interface by MAC address. Returns NULL if the device
564 * is not found or a pointer to the device. The caller must hold the
565 * rtnl semaphore. The returned device has not had its ref count increased
566 * and the caller must therefore be careful about locking
569 * If the API was consistent this would be __dev_get_by_hwaddr
572 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
574 struct net_device *dev;
578 for (dev = dev_base; dev; dev = dev->next)
579 if (dev->type == type &&
580 !memcmp(dev->dev_addr, ha, dev->addr_len))
585 EXPORT_SYMBOL(dev_getbyhwaddr);
587 struct net_device *dev_getfirstbyhwtype(unsigned short type)
589 struct net_device *dev;
592 for (dev = dev_base; dev; dev = dev->next) {
593 if (dev->type == type) {
602 EXPORT_SYMBOL(dev_getfirstbyhwtype);
605 * dev_get_by_flags - find any device with given flags
606 * @if_flags: IFF_* values
607 * @mask: bitmask of bits in if_flags to check
609 * Search for any interface with the given flags. Returns NULL if a device
610 * is not found or a pointer to the device. The device returned has
611 * had a reference added and the pointer is safe until the user calls
612 * dev_put to indicate they have finished with it.
615 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
617 struct net_device *dev;
619 read_lock(&dev_base_lock);
620 for (dev = dev_base; dev != NULL; dev = dev->next) {
621 if (((dev->flags ^ if_flags) & mask) == 0) {
626 read_unlock(&dev_base_lock);
631 * dev_valid_name - check if name is okay for network device
634 * Network device names need to be valid file names to
635 * to allow sysfs to work. We also disallow any kind of
638 int dev_valid_name(const char *name)
642 if (strlen(name) >= IFNAMSIZ)
644 if (!strcmp(name, ".") || !strcmp(name, ".."))
648 if (*name == '/' || isspace(*name))
656 * dev_alloc_name - allocate a name for a device
658 * @name: name format string
660 * Passed a format string - eg "lt%d" it will try and find a suitable
661 * id. It scans list of devices to build up a free map, then chooses
662 * the first empty slot. The caller must hold the dev_base or rtnl lock
663 * while allocating the name and adding the device in order to avoid
665 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
666 * Returns the number of the unit assigned or a negative errno code.
669 int dev_alloc_name(struct net_device *dev, const char *name)
674 const int max_netdevices = 8*PAGE_SIZE;
676 struct net_device *d;
678 p = strnchr(name, IFNAMSIZ-1, '%');
681 * Verify the string as this thing may have come from
682 * the user. There must be either one "%d" and no other "%"
685 if (p[1] != 'd' || strchr(p + 2, '%'))
688 /* Use one page as a bit array of possible slots */
689 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
693 for (d = dev_base; d; d = d->next) {
694 if (!sscanf(d->name, name, &i))
696 if (i < 0 || i >= max_netdevices)
699 /* avoid cases where sscanf is not exact inverse of printf */
700 snprintf(buf, sizeof(buf), name, i);
701 if (!strncmp(buf, d->name, IFNAMSIZ))
705 i = find_first_zero_bit(inuse, max_netdevices);
706 free_page((unsigned long) inuse);
709 snprintf(buf, sizeof(buf), name, i);
710 if (!__dev_get_by_name(buf)) {
711 strlcpy(dev->name, buf, IFNAMSIZ);
715 /* It is possible to run out of possible slots
716 * when the name is long and there isn't enough space left
717 * for the digits, or if all bits are used.
724 * dev_change_name - change name of a device
726 * @newname: name (or format string) must be at least IFNAMSIZ
728 * Change name of a device, can pass format strings "eth%d".
731 int dev_change_name(struct net_device *dev, char *newname)
737 if (dev->flags & IFF_UP)
740 if (!dev_valid_name(newname))
743 if (strchr(newname, '%')) {
744 err = dev_alloc_name(dev, newname);
747 strcpy(newname, dev->name);
749 else if (__dev_get_by_name(newname))
752 strlcpy(dev->name, newname, IFNAMSIZ);
754 device_rename(&dev->dev, dev->name);
755 hlist_del(&dev->name_hlist);
756 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
757 raw_notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
763 * netdev_features_change - device changes features
764 * @dev: device to cause notification
766 * Called to indicate a device has changed features.
768 void netdev_features_change(struct net_device *dev)
770 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
772 EXPORT_SYMBOL(netdev_features_change);
775 * netdev_state_change - device changes state
776 * @dev: device to cause notification
778 * Called to indicate a device has changed state. This function calls
779 * the notifier chains for netdev_chain and sends a NEWLINK message
780 * to the routing socket.
782 void netdev_state_change(struct net_device *dev)
784 if (dev->flags & IFF_UP) {
785 raw_notifier_call_chain(&netdev_chain,
787 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
792 * dev_load - load a network module
793 * @name: name of interface
795 * If a network interface is not present and the process has suitable
796 * privileges this function loads the module. If module loading is not
797 * available in this kernel then it becomes a nop.
800 void dev_load(const char *name)
802 struct net_device *dev;
804 read_lock(&dev_base_lock);
805 dev = __dev_get_by_name(name);
806 read_unlock(&dev_base_lock);
808 if (!dev && capable(CAP_SYS_MODULE))
809 request_module("%s", name);
812 static int default_rebuild_header(struct sk_buff *skb)
814 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
815 skb->dev ? skb->dev->name : "NULL!!!");
821 * dev_open - prepare an interface for use.
822 * @dev: device to open
824 * Takes a device from down to up state. The device's private open
825 * function is invoked and then the multicast lists are loaded. Finally
826 * the device is moved into the up state and a %NETDEV_UP message is
827 * sent to the netdev notifier chain.
829 * Calling this function on an active interface is a nop. On a failure
830 * a negative errno code is returned.
832 int dev_open(struct net_device *dev)
840 if (dev->flags & IFF_UP)
844 * Is it even present?
846 if (!netif_device_present(dev))
850 * Call device private open method
852 set_bit(__LINK_STATE_START, &dev->state);
854 ret = dev->open(dev);
856 clear_bit(__LINK_STATE_START, &dev->state);
860 * If it went open OK then:
867 dev->flags |= IFF_UP;
870 * Initialize multicasting status
875 * Wakeup transmit queue engine
880 * ... and announce new interface.
882 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
888 * dev_close - shutdown an interface.
889 * @dev: device to shutdown
891 * This function moves an active device into down state. A
892 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
893 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
896 int dev_close(struct net_device *dev)
898 if (!(dev->flags & IFF_UP))
902 * Tell people we are going down, so that they can
903 * prepare to death, when device is still operating.
905 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
909 clear_bit(__LINK_STATE_START, &dev->state);
911 /* Synchronize to scheduled poll. We cannot touch poll list,
912 * it can be even on different cpu. So just clear netif_running(),
913 * and wait when poll really will happen. Actually, the best place
914 * for this is inside dev->stop() after device stopped its irq
915 * engine, but this requires more changes in devices. */
917 smp_mb__after_clear_bit(); /* Commit netif_running(). */
918 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
924 * Call the device specific close. This cannot fail.
925 * Only if device is UP
927 * We allow it to be called even after a DETACH hot-plug
934 * Device is now down.
937 dev->flags &= ~IFF_UP;
940 * Tell people we are down
942 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
949 * Device change register/unregister. These are not inline or static
950 * as we export them to the world.
954 * register_netdevice_notifier - register a network notifier block
957 * Register a notifier to be called when network device events occur.
958 * The notifier passed is linked into the kernel structures and must
959 * not be reused until it has been unregistered. A negative errno code
960 * is returned on a failure.
962 * When registered all registration and up events are replayed
963 * to the new notifier to allow device to have a race free
964 * view of the network device list.
967 int register_netdevice_notifier(struct notifier_block *nb)
969 struct net_device *dev;
973 err = raw_notifier_chain_register(&netdev_chain, nb);
975 for (dev = dev_base; dev; dev = dev->next) {
976 nb->notifier_call(nb, NETDEV_REGISTER, dev);
978 if (dev->flags & IFF_UP)
979 nb->notifier_call(nb, NETDEV_UP, dev);
987 * unregister_netdevice_notifier - unregister a network notifier block
990 * Unregister a notifier previously registered by
991 * register_netdevice_notifier(). The notifier is unlinked into the
992 * kernel structures and may then be reused. A negative errno code
993 * is returned on a failure.
996 int unregister_netdevice_notifier(struct notifier_block *nb)
1001 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1007 * call_netdevice_notifiers - call all network notifier blocks
1008 * @val: value passed unmodified to notifier function
1009 * @v: pointer passed unmodified to notifier function
1011 * Call all network notifier blocks. Parameters and return value
1012 * are as for raw_notifier_call_chain().
1015 int call_netdevice_notifiers(unsigned long val, void *v)
1017 return raw_notifier_call_chain(&netdev_chain, val, v);
1020 /* When > 0 there are consumers of rx skb time stamps */
1021 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1023 void net_enable_timestamp(void)
1025 atomic_inc(&netstamp_needed);
1028 void net_disable_timestamp(void)
1030 atomic_dec(&netstamp_needed);
1033 static inline void net_timestamp(struct sk_buff *skb)
1035 if (atomic_read(&netstamp_needed))
1036 __net_timestamp(skb);
1038 skb->tstamp.tv64 = 0;
1042 * Support routine. Sends outgoing frames to any network
1043 * taps currently in use.
1046 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1048 struct packet_type *ptype;
1053 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1054 /* Never send packets back to the socket
1055 * they originated from - MvS (miquels@drinkel.ow.org)
1057 if ((ptype->dev == dev || !ptype->dev) &&
1058 (ptype->af_packet_priv == NULL ||
1059 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1060 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1064 /* skb->nh should be correctly
1065 set by sender, so that the second statement is
1066 just protection against buggy protocols.
1068 skb_reset_mac_header(skb2);
1070 if (skb_network_header(skb2) < skb2->data ||
1071 skb2->network_header > skb2->tail) {
1072 if (net_ratelimit())
1073 printk(KERN_CRIT "protocol %04x is "
1075 skb2->protocol, dev->name);
1076 skb_reset_network_header(skb2);
1079 skb2->transport_header = skb2->network_header;
1080 skb2->pkt_type = PACKET_OUTGOING;
1081 ptype->func(skb2, skb->dev, ptype, skb->dev);
1088 void __netif_schedule(struct net_device *dev)
1090 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1091 unsigned long flags;
1092 struct softnet_data *sd;
1094 local_irq_save(flags);
1095 sd = &__get_cpu_var(softnet_data);
1096 dev->next_sched = sd->output_queue;
1097 sd->output_queue = dev;
1098 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1099 local_irq_restore(flags);
1102 EXPORT_SYMBOL(__netif_schedule);
1104 void __netif_rx_schedule(struct net_device *dev)
1106 unsigned long flags;
1108 local_irq_save(flags);
1110 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1112 dev->quota += dev->weight;
1114 dev->quota = dev->weight;
1115 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1116 local_irq_restore(flags);
1118 EXPORT_SYMBOL(__netif_rx_schedule);
1120 void dev_kfree_skb_any(struct sk_buff *skb)
1122 if (in_irq() || irqs_disabled())
1123 dev_kfree_skb_irq(skb);
1127 EXPORT_SYMBOL(dev_kfree_skb_any);
1131 void netif_device_detach(struct net_device *dev)
1133 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1134 netif_running(dev)) {
1135 netif_stop_queue(dev);
1138 EXPORT_SYMBOL(netif_device_detach);
1140 void netif_device_attach(struct net_device *dev)
1142 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1143 netif_running(dev)) {
1144 netif_wake_queue(dev);
1145 __netdev_watchdog_up(dev);
1148 EXPORT_SYMBOL(netif_device_attach);
1152 * Invalidate hardware checksum when packet is to be mangled, and
1153 * complete checksum manually on outgoing path.
1155 int skb_checksum_help(struct sk_buff *skb)
1158 int ret = 0, offset;
1160 if (skb->ip_summed == CHECKSUM_COMPLETE)
1161 goto out_set_summed;
1163 if (unlikely(skb_shinfo(skb)->gso_size)) {
1164 /* Let GSO fix up the checksum. */
1165 goto out_set_summed;
1168 if (skb_cloned(skb)) {
1169 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1174 offset = skb->csum_start - skb_headroom(skb);
1175 BUG_ON(offset > (int)skb->len);
1176 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1178 offset = skb_headlen(skb) - offset;
1179 BUG_ON(offset <= 0);
1180 BUG_ON(skb->csum_offset + 2 > offset);
1182 *(__sum16 *)(skb->head + skb->csum_start + skb->csum_offset) =
1185 skb->ip_summed = CHECKSUM_NONE;
1191 * skb_gso_segment - Perform segmentation on skb.
1192 * @skb: buffer to segment
1193 * @features: features for the output path (see dev->features)
1195 * This function segments the given skb and returns a list of segments.
1197 * It may return NULL if the skb requires no segmentation. This is
1198 * only possible when GSO is used for verifying header integrity.
1200 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1202 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1203 struct packet_type *ptype;
1204 __be16 type = skb->protocol;
1207 BUG_ON(skb_shinfo(skb)->frag_list);
1209 skb_reset_mac_header(skb);
1210 skb->mac_len = skb->network_header - skb->mac_header;
1211 __skb_pull(skb, skb->mac_len);
1213 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1214 if (skb_header_cloned(skb) &&
1215 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1216 return ERR_PTR(err);
1220 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1221 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1222 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1223 err = ptype->gso_send_check(skb);
1224 segs = ERR_PTR(err);
1225 if (err || skb_gso_ok(skb, features))
1227 __skb_push(skb, (skb->data -
1228 skb_network_header(skb)));
1230 segs = ptype->gso_segment(skb, features);
1236 __skb_push(skb, skb->data - skb_mac_header(skb));
1241 EXPORT_SYMBOL(skb_gso_segment);
1243 /* Take action when hardware reception checksum errors are detected. */
1245 void netdev_rx_csum_fault(struct net_device *dev)
1247 if (net_ratelimit()) {
1248 printk(KERN_ERR "%s: hw csum failure.\n",
1249 dev ? dev->name : "<unknown>");
1253 EXPORT_SYMBOL(netdev_rx_csum_fault);
1256 /* Actually, we should eliminate this check as soon as we know, that:
1257 * 1. IOMMU is present and allows to map all the memory.
1258 * 2. No high memory really exists on this machine.
1261 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1263 #ifdef CONFIG_HIGHMEM
1266 if (dev->features & NETIF_F_HIGHDMA)
1269 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1270 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1278 void (*destructor)(struct sk_buff *skb);
1281 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1283 static void dev_gso_skb_destructor(struct sk_buff *skb)
1285 struct dev_gso_cb *cb;
1288 struct sk_buff *nskb = skb->next;
1290 skb->next = nskb->next;
1293 } while (skb->next);
1295 cb = DEV_GSO_CB(skb);
1297 cb->destructor(skb);
1301 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1302 * @skb: buffer to segment
1304 * This function segments the given skb and stores the list of segments
1307 static int dev_gso_segment(struct sk_buff *skb)
1309 struct net_device *dev = skb->dev;
1310 struct sk_buff *segs;
1311 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1314 segs = skb_gso_segment(skb, features);
1316 /* Verifying header integrity only. */
1320 if (unlikely(IS_ERR(segs)))
1321 return PTR_ERR(segs);
1324 DEV_GSO_CB(skb)->destructor = skb->destructor;
1325 skb->destructor = dev_gso_skb_destructor;
1330 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1332 if (likely(!skb->next)) {
1334 dev_queue_xmit_nit(skb, dev);
1336 if (netif_needs_gso(dev, skb)) {
1337 if (unlikely(dev_gso_segment(skb)))
1343 return dev->hard_start_xmit(skb, dev);
1348 struct sk_buff *nskb = skb->next;
1351 skb->next = nskb->next;
1353 rc = dev->hard_start_xmit(nskb, dev);
1355 nskb->next = skb->next;
1359 if (unlikely(netif_queue_stopped(dev) && skb->next))
1360 return NETDEV_TX_BUSY;
1361 } while (skb->next);
1363 skb->destructor = DEV_GSO_CB(skb)->destructor;
1370 #define HARD_TX_LOCK(dev, cpu) { \
1371 if ((dev->features & NETIF_F_LLTX) == 0) { \
1372 netif_tx_lock(dev); \
1376 #define HARD_TX_UNLOCK(dev) { \
1377 if ((dev->features & NETIF_F_LLTX) == 0) { \
1378 netif_tx_unlock(dev); \
1383 * dev_queue_xmit - transmit a buffer
1384 * @skb: buffer to transmit
1386 * Queue a buffer for transmission to a network device. The caller must
1387 * have set the device and priority and built the buffer before calling
1388 * this function. The function can be called from an interrupt.
1390 * A negative errno code is returned on a failure. A success does not
1391 * guarantee the frame will be transmitted as it may be dropped due
1392 * to congestion or traffic shaping.
1394 * -----------------------------------------------------------------------------------
1395 * I notice this method can also return errors from the queue disciplines,
1396 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1399 * Regardless of the return value, the skb is consumed, so it is currently
1400 * difficult to retry a send to this method. (You can bump the ref count
1401 * before sending to hold a reference for retry if you are careful.)
1403 * When calling this method, interrupts MUST be enabled. This is because
1404 * the BH enable code must have IRQs enabled so that it will not deadlock.
1408 int dev_queue_xmit(struct sk_buff *skb)
1410 struct net_device *dev = skb->dev;
1414 /* GSO will handle the following emulations directly. */
1415 if (netif_needs_gso(dev, skb))
1418 if (skb_shinfo(skb)->frag_list &&
1419 !(dev->features & NETIF_F_FRAGLIST) &&
1420 __skb_linearize(skb))
1423 /* Fragmented skb is linearized if device does not support SG,
1424 * or if at least one of fragments is in highmem and device
1425 * does not support DMA from it.
1427 if (skb_shinfo(skb)->nr_frags &&
1428 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1429 __skb_linearize(skb))
1432 /* If packet is not checksummed and device does not support
1433 * checksumming for this protocol, complete checksumming here.
1435 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1436 skb_set_transport_header(skb, skb->csum_start -
1439 if (!(dev->features & NETIF_F_GEN_CSUM) &&
1440 (!(dev->features & NETIF_F_IP_CSUM) ||
1441 skb->protocol != htons(ETH_P_IP)))
1442 if (skb_checksum_help(skb))
1447 spin_lock_prefetch(&dev->queue_lock);
1449 /* Disable soft irqs for various locks below. Also
1450 * stops preemption for RCU.
1454 /* Updates of qdisc are serialized by queue_lock.
1455 * The struct Qdisc which is pointed to by qdisc is now a
1456 * rcu structure - it may be accessed without acquiring
1457 * a lock (but the structure may be stale.) The freeing of the
1458 * qdisc will be deferred until it's known that there are no
1459 * more references to it.
1461 * If the qdisc has an enqueue function, we still need to
1462 * hold the queue_lock before calling it, since queue_lock
1463 * also serializes access to the device queue.
1466 q = rcu_dereference(dev->qdisc);
1467 #ifdef CONFIG_NET_CLS_ACT
1468 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1471 /* Grab device queue */
1472 spin_lock(&dev->queue_lock);
1475 rc = q->enqueue(skb, q);
1477 spin_unlock(&dev->queue_lock);
1479 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1482 spin_unlock(&dev->queue_lock);
1485 /* The device has no queue. Common case for software devices:
1486 loopback, all the sorts of tunnels...
1488 Really, it is unlikely that netif_tx_lock protection is necessary
1489 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1491 However, it is possible, that they rely on protection
1494 Check this and shot the lock. It is not prone from deadlocks.
1495 Either shot noqueue qdisc, it is even simpler 8)
1497 if (dev->flags & IFF_UP) {
1498 int cpu = smp_processor_id(); /* ok because BHs are off */
1500 if (dev->xmit_lock_owner != cpu) {
1502 HARD_TX_LOCK(dev, cpu);
1504 if (!netif_queue_stopped(dev)) {
1506 if (!dev_hard_start_xmit(skb, dev)) {
1507 HARD_TX_UNLOCK(dev);
1511 HARD_TX_UNLOCK(dev);
1512 if (net_ratelimit())
1513 printk(KERN_CRIT "Virtual device %s asks to "
1514 "queue packet!\n", dev->name);
1516 /* Recursion is detected! It is possible,
1518 if (net_ratelimit())
1519 printk(KERN_CRIT "Dead loop on virtual device "
1520 "%s, fix it urgently!\n", dev->name);
1525 rcu_read_unlock_bh();
1531 rcu_read_unlock_bh();
1536 /*=======================================================================
1538 =======================================================================*/
1540 int netdev_max_backlog __read_mostly = 1000;
1541 int netdev_budget __read_mostly = 300;
1542 int weight_p __read_mostly = 64; /* old backlog weight */
1544 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1548 * netif_rx - post buffer to the network code
1549 * @skb: buffer to post
1551 * This function receives a packet from a device driver and queues it for
1552 * the upper (protocol) levels to process. It always succeeds. The buffer
1553 * may be dropped during processing for congestion control or by the
1557 * NET_RX_SUCCESS (no congestion)
1558 * NET_RX_CN_LOW (low congestion)
1559 * NET_RX_CN_MOD (moderate congestion)
1560 * NET_RX_CN_HIGH (high congestion)
1561 * NET_RX_DROP (packet was dropped)
1565 int netif_rx(struct sk_buff *skb)
1567 struct softnet_data *queue;
1568 unsigned long flags;
1570 /* if netpoll wants it, pretend we never saw it */
1571 if (netpoll_rx(skb))
1574 if (!skb->tstamp.tv64)
1578 * The code is rearranged so that the path is the most
1579 * short when CPU is congested, but is still operating.
1581 local_irq_save(flags);
1582 queue = &__get_cpu_var(softnet_data);
1584 __get_cpu_var(netdev_rx_stat).total++;
1585 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1586 if (queue->input_pkt_queue.qlen) {
1589 __skb_queue_tail(&queue->input_pkt_queue, skb);
1590 local_irq_restore(flags);
1591 return NET_RX_SUCCESS;
1594 netif_rx_schedule(&queue->backlog_dev);
1598 __get_cpu_var(netdev_rx_stat).dropped++;
1599 local_irq_restore(flags);
1605 int netif_rx_ni(struct sk_buff *skb)
1610 err = netif_rx(skb);
1611 if (local_softirq_pending())
1618 EXPORT_SYMBOL(netif_rx_ni);
1620 static inline struct net_device *skb_bond(struct sk_buff *skb)
1622 struct net_device *dev = skb->dev;
1625 if (skb_bond_should_drop(skb)) {
1629 skb->dev = dev->master;
1635 static void net_tx_action(struct softirq_action *h)
1637 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1639 if (sd->completion_queue) {
1640 struct sk_buff *clist;
1642 local_irq_disable();
1643 clist = sd->completion_queue;
1644 sd->completion_queue = NULL;
1648 struct sk_buff *skb = clist;
1649 clist = clist->next;
1651 BUG_TRAP(!atomic_read(&skb->users));
1656 if (sd->output_queue) {
1657 struct net_device *head;
1659 local_irq_disable();
1660 head = sd->output_queue;
1661 sd->output_queue = NULL;
1665 struct net_device *dev = head;
1666 head = head->next_sched;
1668 smp_mb__before_clear_bit();
1669 clear_bit(__LINK_STATE_SCHED, &dev->state);
1671 if (spin_trylock(&dev->queue_lock)) {
1673 spin_unlock(&dev->queue_lock);
1675 netif_schedule(dev);
1681 static inline int deliver_skb(struct sk_buff *skb,
1682 struct packet_type *pt_prev,
1683 struct net_device *orig_dev)
1685 atomic_inc(&skb->users);
1686 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1689 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1690 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1692 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1693 unsigned char *addr);
1694 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1696 static __inline__ int handle_bridge(struct sk_buff **pskb,
1697 struct packet_type **pt_prev, int *ret,
1698 struct net_device *orig_dev)
1700 struct net_bridge_port *port;
1702 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1703 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1707 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1711 return br_handle_frame_hook(port, pskb);
1714 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1717 #ifdef CONFIG_NET_CLS_ACT
1718 /* TODO: Maybe we should just force sch_ingress to be compiled in
1719 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1720 * a compare and 2 stores extra right now if we dont have it on
1721 * but have CONFIG_NET_CLS_ACT
1722 * NOTE: This doesnt stop any functionality; if you dont have
1723 * the ingress scheduler, you just cant add policies on ingress.
1726 static int ing_filter(struct sk_buff *skb)
1729 struct net_device *dev = skb->dev;
1730 int result = TC_ACT_OK;
1732 if (dev->qdisc_ingress) {
1733 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1734 if (MAX_RED_LOOP < ttl++) {
1735 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1736 skb->iif, skb->dev->ifindex);
1740 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1742 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1744 spin_lock(&dev->queue_lock);
1745 if ((q = dev->qdisc_ingress) != NULL)
1746 result = q->enqueue(skb, q);
1747 spin_unlock(&dev->queue_lock);
1755 int netif_receive_skb(struct sk_buff *skb)
1757 struct packet_type *ptype, *pt_prev;
1758 struct net_device *orig_dev;
1759 int ret = NET_RX_DROP;
1762 /* if we've gotten here through NAPI, check netpoll */
1763 if (skb->dev->poll && netpoll_rx(skb))
1766 if (!skb->tstamp.tv64)
1770 skb->iif = skb->dev->ifindex;
1772 orig_dev = skb_bond(skb);
1777 __get_cpu_var(netdev_rx_stat).total++;
1779 skb_reset_network_header(skb);
1780 skb_reset_transport_header(skb);
1781 skb->mac_len = skb->network_header - skb->mac_header;
1787 #ifdef CONFIG_NET_CLS_ACT
1788 if (skb->tc_verd & TC_NCLS) {
1789 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1794 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1795 if (!ptype->dev || ptype->dev == skb->dev) {
1797 ret = deliver_skb(skb, pt_prev, orig_dev);
1802 #ifdef CONFIG_NET_CLS_ACT
1804 ret = deliver_skb(skb, pt_prev, orig_dev);
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 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1824 type = skb->protocol;
1825 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1826 if (ptype->type == type &&
1827 (!ptype->dev || ptype->dev == skb->dev)) {
1829 ret = deliver_skb(skb, pt_prev, orig_dev);
1835 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1838 /* Jamal, now you will not able to escape explaining
1839 * me how you were going to use this. :-)
1849 static int process_backlog(struct net_device *backlog_dev, int *budget)
1852 int quota = min(backlog_dev->quota, *budget);
1853 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1854 unsigned long start_time = jiffies;
1856 backlog_dev->weight = weight_p;
1858 struct sk_buff *skb;
1859 struct net_device *dev;
1861 local_irq_disable();
1862 skb = __skb_dequeue(&queue->input_pkt_queue);
1869 netif_receive_skb(skb);
1875 if (work >= quota || jiffies - start_time > 1)
1880 backlog_dev->quota -= work;
1885 backlog_dev->quota -= work;
1888 list_del(&backlog_dev->poll_list);
1889 smp_mb__before_clear_bit();
1890 netif_poll_enable(backlog_dev);
1896 static void net_rx_action(struct softirq_action *h)
1898 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1899 unsigned long start_time = jiffies;
1900 int budget = netdev_budget;
1903 local_irq_disable();
1905 while (!list_empty(&queue->poll_list)) {
1906 struct net_device *dev;
1908 if (budget <= 0 || jiffies - start_time > 1)
1913 dev = list_entry(queue->poll_list.next,
1914 struct net_device, poll_list);
1915 have = netpoll_poll_lock(dev);
1917 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1918 netpoll_poll_unlock(have);
1919 local_irq_disable();
1920 list_move_tail(&dev->poll_list, &queue->poll_list);
1922 dev->quota += dev->weight;
1924 dev->quota = dev->weight;
1926 netpoll_poll_unlock(have);
1928 local_irq_disable();
1932 #ifdef CONFIG_NET_DMA
1934 * There may not be any more sk_buffs coming right now, so push
1935 * any pending DMA copies to hardware
1937 if (net_dma_client) {
1938 struct dma_chan *chan;
1940 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1941 dma_async_memcpy_issue_pending(chan);
1949 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1950 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1954 static gifconf_func_t * gifconf_list [NPROTO];
1957 * register_gifconf - register a SIOCGIF handler
1958 * @family: Address family
1959 * @gifconf: Function handler
1961 * Register protocol dependent address dumping routines. The handler
1962 * that is passed must not be freed or reused until it has been replaced
1963 * by another handler.
1965 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1967 if (family >= NPROTO)
1969 gifconf_list[family] = gifconf;
1975 * Map an interface index to its name (SIOCGIFNAME)
1979 * We need this ioctl for efficient implementation of the
1980 * if_indextoname() function required by the IPv6 API. Without
1981 * it, we would have to search all the interfaces to find a
1985 static int dev_ifname(struct ifreq __user *arg)
1987 struct net_device *dev;
1991 * Fetch the caller's info block.
1994 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1997 read_lock(&dev_base_lock);
1998 dev = __dev_get_by_index(ifr.ifr_ifindex);
2000 read_unlock(&dev_base_lock);
2004 strcpy(ifr.ifr_name, dev->name);
2005 read_unlock(&dev_base_lock);
2007 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2013 * Perform a SIOCGIFCONF call. This structure will change
2014 * size eventually, and there is nothing I can do about it.
2015 * Thus we will need a 'compatibility mode'.
2018 static int dev_ifconf(char __user *arg)
2021 struct net_device *dev;
2028 * Fetch the caller's info block.
2031 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2038 * Loop over the interfaces, and write an info block for each.
2042 for (dev = dev_base; dev; dev = dev->next) {
2043 for (i = 0; i < NPROTO; i++) {
2044 if (gifconf_list[i]) {
2047 done = gifconf_list[i](dev, NULL, 0);
2049 done = gifconf_list[i](dev, pos + total,
2059 * All done. Write the updated control block back to the caller.
2061 ifc.ifc_len = total;
2064 * Both BSD and Solaris return 0 here, so we do too.
2066 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2069 #ifdef CONFIG_PROC_FS
2071 * This is invoked by the /proc filesystem handler to display a device
2074 static struct net_device *dev_get_idx(loff_t pos)
2076 struct net_device *dev;
2079 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2081 return i == pos ? dev : NULL;
2084 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2086 read_lock(&dev_base_lock);
2087 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2090 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2093 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2096 void dev_seq_stop(struct seq_file *seq, void *v)
2098 read_unlock(&dev_base_lock);
2101 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2103 struct net_device_stats *stats = dev->get_stats(dev);
2106 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2107 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2108 dev->name, stats->rx_bytes, stats->rx_packets,
2110 stats->rx_dropped + stats->rx_missed_errors,
2111 stats->rx_fifo_errors,
2112 stats->rx_length_errors + stats->rx_over_errors +
2113 stats->rx_crc_errors + stats->rx_frame_errors,
2114 stats->rx_compressed, stats->multicast,
2115 stats->tx_bytes, stats->tx_packets,
2116 stats->tx_errors, stats->tx_dropped,
2117 stats->tx_fifo_errors, stats->collisions,
2118 stats->tx_carrier_errors +
2119 stats->tx_aborted_errors +
2120 stats->tx_window_errors +
2121 stats->tx_heartbeat_errors,
2122 stats->tx_compressed);
2124 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2128 * Called from the PROCfs module. This now uses the new arbitrary sized
2129 * /proc/net interface to create /proc/net/dev
2131 static int dev_seq_show(struct seq_file *seq, void *v)
2133 if (v == SEQ_START_TOKEN)
2134 seq_puts(seq, "Inter-| Receive "
2136 " face |bytes packets errs drop fifo frame "
2137 "compressed multicast|bytes packets errs "
2138 "drop fifo colls carrier compressed\n");
2140 dev_seq_printf_stats(seq, v);
2144 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2146 struct netif_rx_stats *rc = NULL;
2148 while (*pos < NR_CPUS)
2149 if (cpu_online(*pos)) {
2150 rc = &per_cpu(netdev_rx_stat, *pos);
2157 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2159 return softnet_get_online(pos);
2162 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2165 return softnet_get_online(pos);
2168 static void softnet_seq_stop(struct seq_file *seq, void *v)
2172 static int softnet_seq_show(struct seq_file *seq, void *v)
2174 struct netif_rx_stats *s = v;
2176 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2177 s->total, s->dropped, s->time_squeeze, 0,
2178 0, 0, 0, 0, /* was fastroute */
2183 static const struct seq_operations dev_seq_ops = {
2184 .start = dev_seq_start,
2185 .next = dev_seq_next,
2186 .stop = dev_seq_stop,
2187 .show = dev_seq_show,
2190 static int dev_seq_open(struct inode *inode, struct file *file)
2192 return seq_open(file, &dev_seq_ops);
2195 static const struct file_operations dev_seq_fops = {
2196 .owner = THIS_MODULE,
2197 .open = dev_seq_open,
2199 .llseek = seq_lseek,
2200 .release = seq_release,
2203 static const struct seq_operations softnet_seq_ops = {
2204 .start = softnet_seq_start,
2205 .next = softnet_seq_next,
2206 .stop = softnet_seq_stop,
2207 .show = softnet_seq_show,
2210 static int softnet_seq_open(struct inode *inode, struct file *file)
2212 return seq_open(file, &softnet_seq_ops);
2215 static const struct file_operations softnet_seq_fops = {
2216 .owner = THIS_MODULE,
2217 .open = softnet_seq_open,
2219 .llseek = seq_lseek,
2220 .release = seq_release,
2223 static void *ptype_get_idx(loff_t pos)
2225 struct packet_type *pt = NULL;
2229 list_for_each_entry_rcu(pt, &ptype_all, list) {
2235 for (t = 0; t < 16; t++) {
2236 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2245 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2248 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2251 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2253 struct packet_type *pt;
2254 struct list_head *nxt;
2258 if (v == SEQ_START_TOKEN)
2259 return ptype_get_idx(0);
2262 nxt = pt->list.next;
2263 if (pt->type == htons(ETH_P_ALL)) {
2264 if (nxt != &ptype_all)
2267 nxt = ptype_base[0].next;
2269 hash = ntohs(pt->type) & 15;
2271 while (nxt == &ptype_base[hash]) {
2274 nxt = ptype_base[hash].next;
2277 return list_entry(nxt, struct packet_type, list);
2280 static void ptype_seq_stop(struct seq_file *seq, void *v)
2285 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2287 #ifdef CONFIG_KALLSYMS
2288 unsigned long offset = 0, symsize;
2289 const char *symname;
2293 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2300 modname = delim = "";
2301 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2307 seq_printf(seq, "[%p]", sym);
2310 static int ptype_seq_show(struct seq_file *seq, void *v)
2312 struct packet_type *pt = v;
2314 if (v == SEQ_START_TOKEN)
2315 seq_puts(seq, "Type Device Function\n");
2317 if (pt->type == htons(ETH_P_ALL))
2318 seq_puts(seq, "ALL ");
2320 seq_printf(seq, "%04x", ntohs(pt->type));
2322 seq_printf(seq, " %-8s ",
2323 pt->dev ? pt->dev->name : "");
2324 ptype_seq_decode(seq, pt->func);
2325 seq_putc(seq, '\n');
2331 static const struct seq_operations ptype_seq_ops = {
2332 .start = ptype_seq_start,
2333 .next = ptype_seq_next,
2334 .stop = ptype_seq_stop,
2335 .show = ptype_seq_show,
2338 static int ptype_seq_open(struct inode *inode, struct file *file)
2340 return seq_open(file, &ptype_seq_ops);
2343 static const struct file_operations ptype_seq_fops = {
2344 .owner = THIS_MODULE,
2345 .open = ptype_seq_open,
2347 .llseek = seq_lseek,
2348 .release = seq_release,
2352 #ifdef CONFIG_WIRELESS_EXT
2353 extern int wireless_proc_init(void);
2355 #define wireless_proc_init() 0
2358 static int __init dev_proc_init(void)
2362 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2364 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2366 if (!proc_net_fops_create("ptype", S_IRUGO, &ptype_seq_fops))
2369 if (wireless_proc_init())
2375 proc_net_remove("softnet_stat");
2377 proc_net_remove("ptype");
2379 proc_net_remove("dev");
2383 #define dev_proc_init() 0
2384 #endif /* CONFIG_PROC_FS */
2388 * netdev_set_master - set up master/slave pair
2389 * @slave: slave device
2390 * @master: new master device
2392 * Changes the master device of the slave. Pass %NULL to break the
2393 * bonding. The caller must hold the RTNL semaphore. On a failure
2394 * a negative errno code is returned. On success the reference counts
2395 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2396 * function returns zero.
2398 int netdev_set_master(struct net_device *slave, struct net_device *master)
2400 struct net_device *old = slave->master;
2410 slave->master = master;
2418 slave->flags |= IFF_SLAVE;
2420 slave->flags &= ~IFF_SLAVE;
2422 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2427 * dev_set_promiscuity - update promiscuity count on a device
2431 * Add or remove promiscuity from a device. While the count in the device
2432 * remains above zero the interface remains promiscuous. Once it hits zero
2433 * the device reverts back to normal filtering operation. A negative inc
2434 * value is used to drop promiscuity on the device.
2436 void dev_set_promiscuity(struct net_device *dev, int inc)
2438 unsigned short old_flags = dev->flags;
2440 if ((dev->promiscuity += inc) == 0)
2441 dev->flags &= ~IFF_PROMISC;
2443 dev->flags |= IFF_PROMISC;
2444 if (dev->flags != old_flags) {
2446 printk(KERN_INFO "device %s %s promiscuous mode\n",
2447 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2449 audit_log(current->audit_context, GFP_ATOMIC,
2450 AUDIT_ANOM_PROMISCUOUS,
2451 "dev=%s prom=%d old_prom=%d auid=%u",
2452 dev->name, (dev->flags & IFF_PROMISC),
2453 (old_flags & IFF_PROMISC),
2454 audit_get_loginuid(current->audit_context));
2459 * dev_set_allmulti - update allmulti count on a device
2463 * Add or remove reception of all multicast frames to a device. While the
2464 * count in the device remains above zero the interface remains listening
2465 * to all interfaces. Once it hits zero the device reverts back to normal
2466 * filtering operation. A negative @inc value is used to drop the counter
2467 * when releasing a resource needing all multicasts.
2470 void dev_set_allmulti(struct net_device *dev, int inc)
2472 unsigned short old_flags = dev->flags;
2474 dev->flags |= IFF_ALLMULTI;
2475 if ((dev->allmulti += inc) == 0)
2476 dev->flags &= ~IFF_ALLMULTI;
2477 if (dev->flags ^ old_flags)
2481 unsigned dev_get_flags(const struct net_device *dev)
2485 flags = (dev->flags & ~(IFF_PROMISC |
2490 (dev->gflags & (IFF_PROMISC |
2493 if (netif_running(dev)) {
2494 if (netif_oper_up(dev))
2495 flags |= IFF_RUNNING;
2496 if (netif_carrier_ok(dev))
2497 flags |= IFF_LOWER_UP;
2498 if (netif_dormant(dev))
2499 flags |= IFF_DORMANT;
2505 int dev_change_flags(struct net_device *dev, unsigned flags)
2508 int old_flags = dev->flags;
2511 * Set the flags on our device.
2514 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2515 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2517 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2521 * Load in the correct multicast list now the flags have changed.
2527 * Have we downed the interface. We handle IFF_UP ourselves
2528 * according to user attempts to set it, rather than blindly
2533 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2534 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2540 if (dev->flags & IFF_UP &&
2541 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2543 raw_notifier_call_chain(&netdev_chain,
2544 NETDEV_CHANGE, dev);
2546 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2547 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2548 dev->gflags ^= IFF_PROMISC;
2549 dev_set_promiscuity(dev, inc);
2552 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2553 is important. Some (broken) drivers set IFF_PROMISC, when
2554 IFF_ALLMULTI is requested not asking us and not reporting.
2556 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2557 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2558 dev->gflags ^= IFF_ALLMULTI;
2559 dev_set_allmulti(dev, inc);
2562 if (old_flags ^ dev->flags)
2563 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2568 int dev_set_mtu(struct net_device *dev, int new_mtu)
2572 if (new_mtu == dev->mtu)
2575 /* MTU must be positive. */
2579 if (!netif_device_present(dev))
2583 if (dev->change_mtu)
2584 err = dev->change_mtu(dev, new_mtu);
2587 if (!err && dev->flags & IFF_UP)
2588 raw_notifier_call_chain(&netdev_chain,
2589 NETDEV_CHANGEMTU, dev);
2593 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2597 if (!dev->set_mac_address)
2599 if (sa->sa_family != dev->type)
2601 if (!netif_device_present(dev))
2603 err = dev->set_mac_address(dev, sa);
2605 raw_notifier_call_chain(&netdev_chain,
2606 NETDEV_CHANGEADDR, dev);
2611 * Perform the SIOCxIFxxx calls.
2613 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2616 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2622 case SIOCGIFFLAGS: /* Get interface flags */
2623 ifr->ifr_flags = dev_get_flags(dev);
2626 case SIOCSIFFLAGS: /* Set interface flags */
2627 return dev_change_flags(dev, ifr->ifr_flags);
2629 case SIOCGIFMETRIC: /* Get the metric on the interface
2630 (currently unused) */
2631 ifr->ifr_metric = 0;
2634 case SIOCSIFMETRIC: /* Set the metric on the interface
2635 (currently unused) */
2638 case SIOCGIFMTU: /* Get the MTU of a device */
2639 ifr->ifr_mtu = dev->mtu;
2642 case SIOCSIFMTU: /* Set the MTU of a device */
2643 return dev_set_mtu(dev, ifr->ifr_mtu);
2647 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2649 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2650 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2651 ifr->ifr_hwaddr.sa_family = dev->type;
2655 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2657 case SIOCSIFHWBROADCAST:
2658 if (ifr->ifr_hwaddr.sa_family != dev->type)
2660 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2661 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2662 raw_notifier_call_chain(&netdev_chain,
2663 NETDEV_CHANGEADDR, dev);
2667 ifr->ifr_map.mem_start = dev->mem_start;
2668 ifr->ifr_map.mem_end = dev->mem_end;
2669 ifr->ifr_map.base_addr = dev->base_addr;
2670 ifr->ifr_map.irq = dev->irq;
2671 ifr->ifr_map.dma = dev->dma;
2672 ifr->ifr_map.port = dev->if_port;
2676 if (dev->set_config) {
2677 if (!netif_device_present(dev))
2679 return dev->set_config(dev, &ifr->ifr_map);
2684 if (!dev->set_multicast_list ||
2685 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2687 if (!netif_device_present(dev))
2689 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2693 if (!dev->set_multicast_list ||
2694 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2696 if (!netif_device_present(dev))
2698 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2702 ifr->ifr_ifindex = dev->ifindex;
2706 ifr->ifr_qlen = dev->tx_queue_len;
2710 if (ifr->ifr_qlen < 0)
2712 dev->tx_queue_len = ifr->ifr_qlen;
2716 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2717 return dev_change_name(dev, ifr->ifr_newname);
2720 * Unknown or private ioctl
2724 if ((cmd >= SIOCDEVPRIVATE &&
2725 cmd <= SIOCDEVPRIVATE + 15) ||
2726 cmd == SIOCBONDENSLAVE ||
2727 cmd == SIOCBONDRELEASE ||
2728 cmd == SIOCBONDSETHWADDR ||
2729 cmd == SIOCBONDSLAVEINFOQUERY ||
2730 cmd == SIOCBONDINFOQUERY ||
2731 cmd == SIOCBONDCHANGEACTIVE ||
2732 cmd == SIOCGMIIPHY ||
2733 cmd == SIOCGMIIREG ||
2734 cmd == SIOCSMIIREG ||
2735 cmd == SIOCBRADDIF ||
2736 cmd == SIOCBRDELIF ||
2737 cmd == SIOCWANDEV) {
2739 if (dev->do_ioctl) {
2740 if (netif_device_present(dev))
2741 err = dev->do_ioctl(dev, ifr,
2754 * This function handles all "interface"-type I/O control requests. The actual
2755 * 'doing' part of this is dev_ifsioc above.
2759 * dev_ioctl - network device ioctl
2760 * @cmd: command to issue
2761 * @arg: pointer to a struct ifreq in user space
2763 * Issue ioctl functions to devices. This is normally called by the
2764 * user space syscall interfaces but can sometimes be useful for
2765 * other purposes. The return value is the return from the syscall if
2766 * positive or a negative errno code on error.
2769 int dev_ioctl(unsigned int cmd, void __user *arg)
2775 /* One special case: SIOCGIFCONF takes ifconf argument
2776 and requires shared lock, because it sleeps writing
2780 if (cmd == SIOCGIFCONF) {
2782 ret = dev_ifconf((char __user *) arg);
2786 if (cmd == SIOCGIFNAME)
2787 return dev_ifname((struct ifreq __user *)arg);
2789 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2792 ifr.ifr_name[IFNAMSIZ-1] = 0;
2794 colon = strchr(ifr.ifr_name, ':');
2799 * See which interface the caller is talking about.
2804 * These ioctl calls:
2805 * - can be done by all.
2806 * - atomic and do not require locking.
2817 dev_load(ifr.ifr_name);
2818 read_lock(&dev_base_lock);
2819 ret = dev_ifsioc(&ifr, cmd);
2820 read_unlock(&dev_base_lock);
2824 if (copy_to_user(arg, &ifr,
2825 sizeof(struct ifreq)))
2831 dev_load(ifr.ifr_name);
2833 ret = dev_ethtool(&ifr);
2838 if (copy_to_user(arg, &ifr,
2839 sizeof(struct ifreq)))
2845 * These ioctl calls:
2846 * - require superuser power.
2847 * - require strict serialization.
2853 if (!capable(CAP_NET_ADMIN))
2855 dev_load(ifr.ifr_name);
2857 ret = dev_ifsioc(&ifr, cmd);
2862 if (copy_to_user(arg, &ifr,
2863 sizeof(struct ifreq)))
2869 * These ioctl calls:
2870 * - require superuser power.
2871 * - require strict serialization.
2872 * - do not return a value
2882 case SIOCSIFHWBROADCAST:
2885 case SIOCBONDENSLAVE:
2886 case SIOCBONDRELEASE:
2887 case SIOCBONDSETHWADDR:
2888 case SIOCBONDCHANGEACTIVE:
2891 if (!capable(CAP_NET_ADMIN))
2894 case SIOCBONDSLAVEINFOQUERY:
2895 case SIOCBONDINFOQUERY:
2896 dev_load(ifr.ifr_name);
2898 ret = dev_ifsioc(&ifr, cmd);
2903 /* Get the per device memory space. We can add this but
2904 * currently do not support it */
2906 /* Set the per device memory buffer space.
2907 * Not applicable in our case */
2912 * Unknown or private ioctl.
2915 if (cmd == SIOCWANDEV ||
2916 (cmd >= SIOCDEVPRIVATE &&
2917 cmd <= SIOCDEVPRIVATE + 15)) {
2918 dev_load(ifr.ifr_name);
2920 ret = dev_ifsioc(&ifr, cmd);
2922 if (!ret && copy_to_user(arg, &ifr,
2923 sizeof(struct ifreq)))
2927 #ifdef CONFIG_WIRELESS_EXT
2928 /* Take care of Wireless Extensions */
2929 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2930 /* If command is `set a parameter', or
2931 * `get the encoding parameters', check if
2932 * the user has the right to do it */
2933 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2934 || cmd == SIOCGIWENCODEEXT) {
2935 if (!capable(CAP_NET_ADMIN))
2938 dev_load(ifr.ifr_name);
2940 /* Follow me in net/core/wireless.c */
2941 ret = wireless_process_ioctl(&ifr, cmd);
2943 if (IW_IS_GET(cmd) &&
2944 copy_to_user(arg, &ifr,
2945 sizeof(struct ifreq)))
2949 #endif /* CONFIG_WIRELESS_EXT */
2956 * dev_new_index - allocate an ifindex
2958 * Returns a suitable unique value for a new device interface
2959 * number. The caller must hold the rtnl semaphore or the
2960 * dev_base_lock to be sure it remains unique.
2962 static int dev_new_index(void)
2968 if (!__dev_get_by_index(ifindex))
2973 static int dev_boot_phase = 1;
2975 /* Delayed registration/unregisteration */
2976 static DEFINE_SPINLOCK(net_todo_list_lock);
2977 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2979 static void net_set_todo(struct net_device *dev)
2981 spin_lock(&net_todo_list_lock);
2982 list_add_tail(&dev->todo_list, &net_todo_list);
2983 spin_unlock(&net_todo_list_lock);
2987 * register_netdevice - register a network device
2988 * @dev: device to register
2990 * Take a completed network device structure and add it to the kernel
2991 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2992 * chain. 0 is returned on success. A negative errno code is returned
2993 * on a failure to set up the device, or if the name is a duplicate.
2995 * Callers must hold the rtnl semaphore. You may want
2996 * register_netdev() instead of this.
2999 * The locking appears insufficient to guarantee two parallel registers
3000 * will not get the same name.
3003 int register_netdevice(struct net_device *dev)
3005 struct hlist_head *head;
3006 struct hlist_node *p;
3009 BUG_ON(dev_boot_phase);
3014 /* When net_device's are persistent, this will be fatal. */
3015 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3017 spin_lock_init(&dev->queue_lock);
3018 spin_lock_init(&dev->_xmit_lock);
3019 dev->xmit_lock_owner = -1;
3020 #ifdef CONFIG_NET_CLS_ACT
3021 spin_lock_init(&dev->ingress_lock);
3026 /* Init, if this function is available */
3028 ret = dev->init(dev);
3036 if (!dev_valid_name(dev->name)) {
3041 dev->ifindex = dev_new_index();
3042 if (dev->iflink == -1)
3043 dev->iflink = dev->ifindex;
3045 /* Check for existence of name */
3046 head = dev_name_hash(dev->name);
3047 hlist_for_each(p, head) {
3048 struct net_device *d
3049 = hlist_entry(p, struct net_device, name_hlist);
3050 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3056 /* Fix illegal SG+CSUM combinations. */
3057 if ((dev->features & NETIF_F_SG) &&
3058 !(dev->features & NETIF_F_ALL_CSUM)) {
3059 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3061 dev->features &= ~NETIF_F_SG;
3064 /* TSO requires that SG is present as well. */
3065 if ((dev->features & NETIF_F_TSO) &&
3066 !(dev->features & NETIF_F_SG)) {
3067 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3069 dev->features &= ~NETIF_F_TSO;
3071 if (dev->features & NETIF_F_UFO) {
3072 if (!(dev->features & NETIF_F_HW_CSUM)) {
3073 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3074 "NETIF_F_HW_CSUM feature.\n",
3076 dev->features &= ~NETIF_F_UFO;
3078 if (!(dev->features & NETIF_F_SG)) {
3079 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3080 "NETIF_F_SG feature.\n",
3082 dev->features &= ~NETIF_F_UFO;
3087 * nil rebuild_header routine,
3088 * that should be never called and used as just bug trap.
3091 if (!dev->rebuild_header)
3092 dev->rebuild_header = default_rebuild_header;
3094 ret = netdev_register_sysfs(dev);
3097 dev->reg_state = NETREG_REGISTERED;
3100 * Default initial state at registry is that the
3101 * device is present.
3104 set_bit(__LINK_STATE_PRESENT, &dev->state);
3107 dev_init_scheduler(dev);
3108 write_lock_bh(&dev_base_lock);
3110 dev_tail = &dev->next;
3111 hlist_add_head(&dev->name_hlist, head);
3112 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3114 write_unlock_bh(&dev_base_lock);
3116 /* Notify protocols, that a new device appeared. */
3117 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3126 * register_netdev - register a network device
3127 * @dev: device to register
3129 * Take a completed network device structure and add it to the kernel
3130 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3131 * chain. 0 is returned on success. A negative errno code is returned
3132 * on a failure to set up the device, or if the name is a duplicate.
3134 * This is a wrapper around register_netdev that takes the rtnl semaphore
3135 * and expands the device name if you passed a format string to
3138 int register_netdev(struct net_device *dev)
3145 * If the name is a format string the caller wants us to do a
3148 if (strchr(dev->name, '%')) {
3149 err = dev_alloc_name(dev, dev->name);
3154 err = register_netdevice(dev);
3159 EXPORT_SYMBOL(register_netdev);
3162 * netdev_wait_allrefs - wait until all references are gone.
3164 * This is called when unregistering network devices.
3166 * Any protocol or device that holds a reference should register
3167 * for netdevice notification, and cleanup and put back the
3168 * reference if they receive an UNREGISTER event.
3169 * We can get stuck here if buggy protocols don't correctly
3172 static void netdev_wait_allrefs(struct net_device *dev)
3174 unsigned long rebroadcast_time, warning_time;
3176 rebroadcast_time = warning_time = jiffies;
3177 while (atomic_read(&dev->refcnt) != 0) {
3178 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3181 /* Rebroadcast unregister notification */
3182 raw_notifier_call_chain(&netdev_chain,
3183 NETDEV_UNREGISTER, dev);
3185 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3187 /* We must not have linkwatch events
3188 * pending on unregister. If this
3189 * happens, we simply run the queue
3190 * unscheduled, resulting in a noop
3193 linkwatch_run_queue();
3198 rebroadcast_time = jiffies;
3203 if (time_after(jiffies, warning_time + 10 * HZ)) {
3204 printk(KERN_EMERG "unregister_netdevice: "
3205 "waiting for %s to become free. Usage "
3207 dev->name, atomic_read(&dev->refcnt));
3208 warning_time = jiffies;
3217 * register_netdevice(x1);
3218 * register_netdevice(x2);
3220 * unregister_netdevice(y1);
3221 * unregister_netdevice(y2);
3227 * We are invoked by rtnl_unlock() after it drops the semaphore.
3228 * This allows us to deal with problems:
3229 * 1) We can delete sysfs objects which invoke hotplug
3230 * without deadlocking with linkwatch via keventd.
3231 * 2) Since we run with the RTNL semaphore not held, we can sleep
3232 * safely in order to wait for the netdev refcnt to drop to zero.
3234 static DEFINE_MUTEX(net_todo_run_mutex);
3235 void netdev_run_todo(void)
3237 struct list_head list;
3239 /* Need to guard against multiple cpu's getting out of order. */
3240 mutex_lock(&net_todo_run_mutex);
3242 /* Not safe to do outside the semaphore. We must not return
3243 * until all unregister events invoked by the local processor
3244 * have been completed (either by this todo run, or one on
3247 if (list_empty(&net_todo_list))
3250 /* Snapshot list, allow later requests */
3251 spin_lock(&net_todo_list_lock);
3252 list_replace_init(&net_todo_list, &list);
3253 spin_unlock(&net_todo_list_lock);
3255 while (!list_empty(&list)) {
3256 struct net_device *dev
3257 = list_entry(list.next, struct net_device, todo_list);
3258 list_del(&dev->todo_list);
3260 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3261 printk(KERN_ERR "network todo '%s' but state %d\n",
3262 dev->name, dev->reg_state);
3267 netdev_unregister_sysfs(dev);
3268 dev->reg_state = NETREG_UNREGISTERED;
3270 netdev_wait_allrefs(dev);
3273 BUG_ON(atomic_read(&dev->refcnt));
3274 BUG_TRAP(!dev->ip_ptr);
3275 BUG_TRAP(!dev->ip6_ptr);
3276 BUG_TRAP(!dev->dn_ptr);
3278 /* It must be the very last action,
3279 * after this 'dev' may point to freed up memory.
3281 if (dev->destructor)
3282 dev->destructor(dev);
3286 mutex_unlock(&net_todo_run_mutex);
3289 static struct net_device_stats *maybe_internal_stats(struct net_device *dev)
3291 if (dev->features & NETIF_F_INTERNAL_STATS)
3297 * alloc_netdev - allocate network device
3298 * @sizeof_priv: size of private data to allocate space for
3299 * @name: device name format string
3300 * @setup: callback to initialize device
3302 * Allocates a struct net_device with private data area for driver use
3303 * and performs basic initialization.
3305 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3306 void (*setup)(struct net_device *))
3309 struct net_device *dev;
3312 BUG_ON(strlen(name) >= sizeof(dev->name));
3314 /* ensure 32-byte alignment of both the device and private area */
3315 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3316 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3318 p = kzalloc(alloc_size, GFP_KERNEL);
3320 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3324 dev = (struct net_device *)
3325 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3326 dev->padded = (char *)dev - (char *)p;
3329 dev->priv = netdev_priv(dev);
3331 dev->get_stats = maybe_internal_stats;
3333 strcpy(dev->name, name);
3336 EXPORT_SYMBOL(alloc_netdev);
3339 * free_netdev - free network device
3342 * This function does the last stage of destroying an allocated device
3343 * interface. The reference to the device object is released.
3344 * If this is the last reference then it will be freed.
3346 void free_netdev(struct net_device *dev)
3349 /* Compatibility with error handling in drivers */
3350 if (dev->reg_state == NETREG_UNINITIALIZED) {
3351 kfree((char *)dev - dev->padded);
3355 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3356 dev->reg_state = NETREG_RELEASED;
3358 /* will free via device release */
3359 put_device(&dev->dev);
3361 kfree((char *)dev - dev->padded);
3365 /* Synchronize with packet receive processing. */
3366 void synchronize_net(void)
3373 * unregister_netdevice - remove device from the kernel
3376 * This function shuts down a device interface and removes it
3377 * from the kernel tables. On success 0 is returned, on a failure
3378 * a negative errno code is returned.
3380 * Callers must hold the rtnl semaphore. You may want
3381 * unregister_netdev() instead of this.
3384 void unregister_netdevice(struct net_device *dev)
3386 struct net_device *d, **dp;
3388 BUG_ON(dev_boot_phase);
3391 /* Some devices call without registering for initialization unwind. */
3392 if (dev->reg_state == NETREG_UNINITIALIZED) {
3393 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3394 "was registered\n", dev->name, dev);
3400 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3402 /* If device is running, close it first. */
3403 if (dev->flags & IFF_UP)
3406 /* And unlink it from device chain. */
3407 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3409 write_lock_bh(&dev_base_lock);
3410 hlist_del(&dev->name_hlist);
3411 hlist_del(&dev->index_hlist);
3412 if (dev_tail == &dev->next)
3415 write_unlock_bh(&dev_base_lock);
3421 dev->reg_state = NETREG_UNREGISTERING;
3425 /* Shutdown queueing discipline. */
3429 /* Notify protocols, that we are about to destroy
3430 this device. They should clean all the things.
3432 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3435 * Flush the multicast chain
3437 dev_mc_discard(dev);
3442 /* Notifier chain MUST detach us from master device. */
3443 BUG_TRAP(!dev->master);
3445 /* Finish processing unregister after unlock */
3454 * unregister_netdev - remove device from the kernel
3457 * This function shuts down a device interface and removes it
3458 * from the kernel tables. On success 0 is returned, on a failure
3459 * a negative errno code is returned.
3461 * This is just a wrapper for unregister_netdevice that takes
3462 * the rtnl semaphore. In general you want to use this and not
3463 * unregister_netdevice.
3465 void unregister_netdev(struct net_device *dev)
3468 unregister_netdevice(dev);
3472 EXPORT_SYMBOL(unregister_netdev);
3474 static int dev_cpu_callback(struct notifier_block *nfb,
3475 unsigned long action,
3478 struct sk_buff **list_skb;
3479 struct net_device **list_net;
3480 struct sk_buff *skb;
3481 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3482 struct softnet_data *sd, *oldsd;
3484 if (action != CPU_DEAD)
3487 local_irq_disable();
3488 cpu = smp_processor_id();
3489 sd = &per_cpu(softnet_data, cpu);
3490 oldsd = &per_cpu(softnet_data, oldcpu);
3492 /* Find end of our completion_queue. */
3493 list_skb = &sd->completion_queue;
3495 list_skb = &(*list_skb)->next;
3496 /* Append completion queue from offline CPU. */
3497 *list_skb = oldsd->completion_queue;
3498 oldsd->completion_queue = NULL;
3500 /* Find end of our output_queue. */
3501 list_net = &sd->output_queue;
3503 list_net = &(*list_net)->next_sched;
3504 /* Append output queue from offline CPU. */
3505 *list_net = oldsd->output_queue;
3506 oldsd->output_queue = NULL;
3508 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3511 /* Process offline CPU's input_pkt_queue */
3512 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3518 #ifdef CONFIG_NET_DMA
3520 * net_dma_rebalance -
3521 * This is called when the number of channels allocated to the net_dma_client
3522 * changes. The net_dma_client tries to have one DMA channel per CPU.
3524 static void net_dma_rebalance(void)
3526 unsigned int cpu, i, n;
3527 struct dma_chan *chan;
3529 if (net_dma_count == 0) {
3530 for_each_online_cpu(cpu)
3531 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
3536 cpu = first_cpu(cpu_online_map);
3539 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3540 n = ((num_online_cpus() / net_dma_count)
3541 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3544 per_cpu(softnet_data, cpu).net_dma = chan;
3545 cpu = next_cpu(cpu, cpu_online_map);
3554 * netdev_dma_event - event callback for the net_dma_client
3555 * @client: should always be net_dma_client
3556 * @chan: DMA channel for the event
3557 * @event: event type
3559 static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3560 enum dma_event event)
3562 spin_lock(&net_dma_event_lock);
3564 case DMA_RESOURCE_ADDED:
3566 net_dma_rebalance();
3568 case DMA_RESOURCE_REMOVED:
3570 net_dma_rebalance();
3575 spin_unlock(&net_dma_event_lock);
3579 * netdev_dma_regiser - register the networking subsystem as a DMA client
3581 static int __init netdev_dma_register(void)
3583 spin_lock_init(&net_dma_event_lock);
3584 net_dma_client = dma_async_client_register(netdev_dma_event);
3585 if (net_dma_client == NULL)
3588 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3593 static int __init netdev_dma_register(void) { return -ENODEV; }
3594 #endif /* CONFIG_NET_DMA */
3597 * Initialize the DEV module. At boot time this walks the device list and
3598 * unhooks any devices that fail to initialise (normally hardware not
3599 * present) and leaves us with a valid list of present and active devices.
3604 * This is called single threaded during boot, so no need
3605 * to take the rtnl semaphore.
3607 static int __init net_dev_init(void)
3609 int i, rc = -ENOMEM;
3611 BUG_ON(!dev_boot_phase);
3613 if (dev_proc_init())
3616 if (netdev_sysfs_init())
3619 INIT_LIST_HEAD(&ptype_all);
3620 for (i = 0; i < 16; i++)
3621 INIT_LIST_HEAD(&ptype_base[i]);
3623 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3624 INIT_HLIST_HEAD(&dev_name_head[i]);
3626 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3627 INIT_HLIST_HEAD(&dev_index_head[i]);
3630 * Initialise the packet receive queues.
3633 for_each_possible_cpu(i) {
3634 struct softnet_data *queue;
3636 queue = &per_cpu(softnet_data, i);
3637 skb_queue_head_init(&queue->input_pkt_queue);
3638 queue->completion_queue = NULL;
3639 INIT_LIST_HEAD(&queue->poll_list);
3640 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3641 queue->backlog_dev.weight = weight_p;
3642 queue->backlog_dev.poll = process_backlog;
3643 atomic_set(&queue->backlog_dev.refcnt, 1);
3646 netdev_dma_register();
3650 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3651 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3653 hotcpu_notifier(dev_cpu_callback, 0);
3661 subsys_initcall(net_dev_init);
3663 EXPORT_SYMBOL(__dev_get_by_index);
3664 EXPORT_SYMBOL(__dev_get_by_name);
3665 EXPORT_SYMBOL(__dev_remove_pack);
3666 EXPORT_SYMBOL(dev_valid_name);
3667 EXPORT_SYMBOL(dev_add_pack);
3668 EXPORT_SYMBOL(dev_alloc_name);
3669 EXPORT_SYMBOL(dev_close);
3670 EXPORT_SYMBOL(dev_get_by_flags);
3671 EXPORT_SYMBOL(dev_get_by_index);
3672 EXPORT_SYMBOL(dev_get_by_name);
3673 EXPORT_SYMBOL(dev_open);
3674 EXPORT_SYMBOL(dev_queue_xmit);
3675 EXPORT_SYMBOL(dev_remove_pack);
3676 EXPORT_SYMBOL(dev_set_allmulti);
3677 EXPORT_SYMBOL(dev_set_promiscuity);
3678 EXPORT_SYMBOL(dev_change_flags);
3679 EXPORT_SYMBOL(dev_set_mtu);
3680 EXPORT_SYMBOL(dev_set_mac_address);
3681 EXPORT_SYMBOL(free_netdev);
3682 EXPORT_SYMBOL(netdev_boot_setup_check);
3683 EXPORT_SYMBOL(netdev_set_master);
3684 EXPORT_SYMBOL(netdev_state_change);
3685 EXPORT_SYMBOL(netif_receive_skb);
3686 EXPORT_SYMBOL(netif_rx);
3687 EXPORT_SYMBOL(register_gifconf);
3688 EXPORT_SYMBOL(register_netdevice);
3689 EXPORT_SYMBOL(register_netdevice_notifier);
3690 EXPORT_SYMBOL(skb_checksum_help);
3691 EXPORT_SYMBOL(synchronize_net);
3692 EXPORT_SYMBOL(unregister_netdevice);
3693 EXPORT_SYMBOL(unregister_netdevice_notifier);
3694 EXPORT_SYMBOL(net_enable_timestamp);
3695 EXPORT_SYMBOL(net_disable_timestamp);
3696 EXPORT_SYMBOL(dev_get_flags);
3698 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3699 EXPORT_SYMBOL(br_handle_frame_hook);
3700 EXPORT_SYMBOL(br_fdb_get_hook);
3701 EXPORT_SYMBOL(br_fdb_put_hook);
3705 EXPORT_SYMBOL(dev_load);
3708 EXPORT_PER_CPU_SYMBOL(softnet_data);