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 = skb_transport_offset(skb);
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 BUG_ON(offset > (int)skb->len);
1175 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1177 offset = skb->tail - skb->transport_header;
1178 BUG_ON(offset <= 0);
1179 BUG_ON(skb->csum_offset + 2 > offset);
1181 *(__sum16 *)(skb_transport_header(skb) +
1182 skb->csum_offset) = csum_fold(csum);
1184 skb->ip_summed = CHECKSUM_NONE;
1190 * skb_gso_segment - Perform segmentation on skb.
1191 * @skb: buffer to segment
1192 * @features: features for the output path (see dev->features)
1194 * This function segments the given skb and returns a list of segments.
1196 * It may return NULL if the skb requires no segmentation. This is
1197 * only possible when GSO is used for verifying header integrity.
1199 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1201 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1202 struct packet_type *ptype;
1203 __be16 type = skb->protocol;
1206 BUG_ON(skb_shinfo(skb)->frag_list);
1208 skb_reset_mac_header(skb);
1209 skb->mac_len = skb->network_header - skb->mac_header;
1210 __skb_pull(skb, skb->mac_len);
1212 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1213 if (skb_header_cloned(skb) &&
1214 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1215 return ERR_PTR(err);
1219 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1220 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1221 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1222 err = ptype->gso_send_check(skb);
1223 segs = ERR_PTR(err);
1224 if (err || skb_gso_ok(skb, features))
1226 __skb_push(skb, (skb->data -
1227 skb_network_header(skb)));
1229 segs = ptype->gso_segment(skb, features);
1235 __skb_push(skb, skb->data - skb_mac_header(skb));
1240 EXPORT_SYMBOL(skb_gso_segment);
1242 /* Take action when hardware reception checksum errors are detected. */
1244 void netdev_rx_csum_fault(struct net_device *dev)
1246 if (net_ratelimit()) {
1247 printk(KERN_ERR "%s: hw csum failure.\n",
1248 dev ? dev->name : "<unknown>");
1252 EXPORT_SYMBOL(netdev_rx_csum_fault);
1255 /* Actually, we should eliminate this check as soon as we know, that:
1256 * 1. IOMMU is present and allows to map all the memory.
1257 * 2. No high memory really exists on this machine.
1260 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1262 #ifdef CONFIG_HIGHMEM
1265 if (dev->features & NETIF_F_HIGHDMA)
1268 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1269 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1277 void (*destructor)(struct sk_buff *skb);
1280 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1282 static void dev_gso_skb_destructor(struct sk_buff *skb)
1284 struct dev_gso_cb *cb;
1287 struct sk_buff *nskb = skb->next;
1289 skb->next = nskb->next;
1292 } while (skb->next);
1294 cb = DEV_GSO_CB(skb);
1296 cb->destructor(skb);
1300 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1301 * @skb: buffer to segment
1303 * This function segments the given skb and stores the list of segments
1306 static int dev_gso_segment(struct sk_buff *skb)
1308 struct net_device *dev = skb->dev;
1309 struct sk_buff *segs;
1310 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1313 segs = skb_gso_segment(skb, features);
1315 /* Verifying header integrity only. */
1319 if (unlikely(IS_ERR(segs)))
1320 return PTR_ERR(segs);
1323 DEV_GSO_CB(skb)->destructor = skb->destructor;
1324 skb->destructor = dev_gso_skb_destructor;
1329 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1331 if (likely(!skb->next)) {
1333 dev_queue_xmit_nit(skb, dev);
1335 if (netif_needs_gso(dev, skb)) {
1336 if (unlikely(dev_gso_segment(skb)))
1342 return dev->hard_start_xmit(skb, dev);
1347 struct sk_buff *nskb = skb->next;
1350 skb->next = nskb->next;
1352 rc = dev->hard_start_xmit(nskb, dev);
1354 nskb->next = skb->next;
1358 if (unlikely(netif_queue_stopped(dev) && skb->next))
1359 return NETDEV_TX_BUSY;
1360 } while (skb->next);
1362 skb->destructor = DEV_GSO_CB(skb)->destructor;
1369 #define HARD_TX_LOCK(dev, cpu) { \
1370 if ((dev->features & NETIF_F_LLTX) == 0) { \
1371 netif_tx_lock(dev); \
1375 #define HARD_TX_UNLOCK(dev) { \
1376 if ((dev->features & NETIF_F_LLTX) == 0) { \
1377 netif_tx_unlock(dev); \
1382 * dev_queue_xmit - transmit a buffer
1383 * @skb: buffer to transmit
1385 * Queue a buffer for transmission to a network device. The caller must
1386 * have set the device and priority and built the buffer before calling
1387 * this function. The function can be called from an interrupt.
1389 * A negative errno code is returned on a failure. A success does not
1390 * guarantee the frame will be transmitted as it may be dropped due
1391 * to congestion or traffic shaping.
1393 * -----------------------------------------------------------------------------------
1394 * I notice this method can also return errors from the queue disciplines,
1395 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1398 * Regardless of the return value, the skb is consumed, so it is currently
1399 * difficult to retry a send to this method. (You can bump the ref count
1400 * before sending to hold a reference for retry if you are careful.)
1402 * When calling this method, interrupts MUST be enabled. This is because
1403 * the BH enable code must have IRQs enabled so that it will not deadlock.
1407 int dev_queue_xmit(struct sk_buff *skb)
1409 struct net_device *dev = skb->dev;
1413 /* GSO will handle the following emulations directly. */
1414 if (netif_needs_gso(dev, skb))
1417 if (skb_shinfo(skb)->frag_list &&
1418 !(dev->features & NETIF_F_FRAGLIST) &&
1419 __skb_linearize(skb))
1422 /* Fragmented skb is linearized if device does not support SG,
1423 * or if at least one of fragments is in highmem and device
1424 * does not support DMA from it.
1426 if (skb_shinfo(skb)->nr_frags &&
1427 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1428 __skb_linearize(skb))
1431 /* If packet is not checksummed and device does not support
1432 * checksumming for this protocol, complete checksumming here.
1434 if (skb->ip_summed == CHECKSUM_PARTIAL &&
1435 (!(dev->features & NETIF_F_GEN_CSUM) &&
1436 (!(dev->features & NETIF_F_IP_CSUM) ||
1437 skb->protocol != htons(ETH_P_IP))))
1438 if (skb_checksum_help(skb))
1442 spin_lock_prefetch(&dev->queue_lock);
1444 /* Disable soft irqs for various locks below. Also
1445 * stops preemption for RCU.
1449 /* Updates of qdisc are serialized by queue_lock.
1450 * The struct Qdisc which is pointed to by qdisc is now a
1451 * rcu structure - it may be accessed without acquiring
1452 * a lock (but the structure may be stale.) The freeing of the
1453 * qdisc will be deferred until it's known that there are no
1454 * more references to it.
1456 * If the qdisc has an enqueue function, we still need to
1457 * hold the queue_lock before calling it, since queue_lock
1458 * also serializes access to the device queue.
1461 q = rcu_dereference(dev->qdisc);
1462 #ifdef CONFIG_NET_CLS_ACT
1463 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1466 /* Grab device queue */
1467 spin_lock(&dev->queue_lock);
1470 rc = q->enqueue(skb, q);
1472 spin_unlock(&dev->queue_lock);
1474 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1477 spin_unlock(&dev->queue_lock);
1480 /* The device has no queue. Common case for software devices:
1481 loopback, all the sorts of tunnels...
1483 Really, it is unlikely that netif_tx_lock protection is necessary
1484 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1486 However, it is possible, that they rely on protection
1489 Check this and shot the lock. It is not prone from deadlocks.
1490 Either shot noqueue qdisc, it is even simpler 8)
1492 if (dev->flags & IFF_UP) {
1493 int cpu = smp_processor_id(); /* ok because BHs are off */
1495 if (dev->xmit_lock_owner != cpu) {
1497 HARD_TX_LOCK(dev, cpu);
1499 if (!netif_queue_stopped(dev)) {
1501 if (!dev_hard_start_xmit(skb, dev)) {
1502 HARD_TX_UNLOCK(dev);
1506 HARD_TX_UNLOCK(dev);
1507 if (net_ratelimit())
1508 printk(KERN_CRIT "Virtual device %s asks to "
1509 "queue packet!\n", dev->name);
1511 /* Recursion is detected! It is possible,
1513 if (net_ratelimit())
1514 printk(KERN_CRIT "Dead loop on virtual device "
1515 "%s, fix it urgently!\n", dev->name);
1520 rcu_read_unlock_bh();
1526 rcu_read_unlock_bh();
1531 /*=======================================================================
1533 =======================================================================*/
1535 int netdev_max_backlog __read_mostly = 1000;
1536 int netdev_budget __read_mostly = 300;
1537 int weight_p __read_mostly = 64; /* old backlog weight */
1539 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1543 * netif_rx - post buffer to the network code
1544 * @skb: buffer to post
1546 * This function receives a packet from a device driver and queues it for
1547 * the upper (protocol) levels to process. It always succeeds. The buffer
1548 * may be dropped during processing for congestion control or by the
1552 * NET_RX_SUCCESS (no congestion)
1553 * NET_RX_CN_LOW (low congestion)
1554 * NET_RX_CN_MOD (moderate congestion)
1555 * NET_RX_CN_HIGH (high congestion)
1556 * NET_RX_DROP (packet was dropped)
1560 int netif_rx(struct sk_buff *skb)
1562 struct softnet_data *queue;
1563 unsigned long flags;
1565 /* if netpoll wants it, pretend we never saw it */
1566 if (netpoll_rx(skb))
1569 if (!skb->tstamp.tv64)
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 queue = &__get_cpu_var(softnet_data);
1579 __get_cpu_var(netdev_rx_stat).total++;
1580 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1581 if (queue->input_pkt_queue.qlen) {
1584 __skb_queue_tail(&queue->input_pkt_queue, skb);
1585 local_irq_restore(flags);
1586 return NET_RX_SUCCESS;
1589 netif_rx_schedule(&queue->backlog_dev);
1593 __get_cpu_var(netdev_rx_stat).dropped++;
1594 local_irq_restore(flags);
1600 int netif_rx_ni(struct sk_buff *skb)
1605 err = netif_rx(skb);
1606 if (local_softirq_pending())
1613 EXPORT_SYMBOL(netif_rx_ni);
1615 static inline struct net_device *skb_bond(struct sk_buff *skb)
1617 struct net_device *dev = skb->dev;
1620 if (skb_bond_should_drop(skb)) {
1624 skb->dev = dev->master;
1630 static void net_tx_action(struct softirq_action *h)
1632 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1634 if (sd->completion_queue) {
1635 struct sk_buff *clist;
1637 local_irq_disable();
1638 clist = sd->completion_queue;
1639 sd->completion_queue = NULL;
1643 struct sk_buff *skb = clist;
1644 clist = clist->next;
1646 BUG_TRAP(!atomic_read(&skb->users));
1651 if (sd->output_queue) {
1652 struct net_device *head;
1654 local_irq_disable();
1655 head = sd->output_queue;
1656 sd->output_queue = NULL;
1660 struct net_device *dev = head;
1661 head = head->next_sched;
1663 smp_mb__before_clear_bit();
1664 clear_bit(__LINK_STATE_SCHED, &dev->state);
1666 if (spin_trylock(&dev->queue_lock)) {
1668 spin_unlock(&dev->queue_lock);
1670 netif_schedule(dev);
1676 static inline int deliver_skb(struct sk_buff *skb,
1677 struct packet_type *pt_prev,
1678 struct net_device *orig_dev)
1680 atomic_inc(&skb->users);
1681 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1684 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1685 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1687 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1688 unsigned char *addr);
1689 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1691 static __inline__ int handle_bridge(struct sk_buff **pskb,
1692 struct packet_type **pt_prev, int *ret,
1693 struct net_device *orig_dev)
1695 struct net_bridge_port *port;
1697 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1698 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1702 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1706 return br_handle_frame_hook(port, pskb);
1709 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1712 #ifdef CONFIG_NET_CLS_ACT
1713 /* TODO: Maybe we should just force sch_ingress to be compiled in
1714 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1715 * a compare and 2 stores extra right now if we dont have it on
1716 * but have CONFIG_NET_CLS_ACT
1717 * NOTE: This doesnt stop any functionality; if you dont have
1718 * the ingress scheduler, you just cant add policies on ingress.
1721 static int ing_filter(struct sk_buff *skb)
1724 struct net_device *dev = skb->dev;
1725 int result = TC_ACT_OK;
1727 if (dev->qdisc_ingress) {
1728 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1729 if (MAX_RED_LOOP < ttl++) {
1730 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1731 skb->iif, skb->dev->ifindex);
1735 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1737 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1739 spin_lock(&dev->queue_lock);
1740 if ((q = dev->qdisc_ingress) != NULL)
1741 result = q->enqueue(skb, q);
1742 spin_unlock(&dev->queue_lock);
1750 int netif_receive_skb(struct sk_buff *skb)
1752 struct packet_type *ptype, *pt_prev;
1753 struct net_device *orig_dev;
1754 int ret = NET_RX_DROP;
1757 /* if we've gotten here through NAPI, check netpoll */
1758 if (skb->dev->poll && netpoll_rx(skb))
1761 if (!skb->tstamp.tv64)
1765 skb->iif = skb->dev->ifindex;
1767 orig_dev = skb_bond(skb);
1772 __get_cpu_var(netdev_rx_stat).total++;
1774 skb_reset_network_header(skb);
1775 skb_reset_transport_header(skb);
1776 skb->mac_len = skb->network_header - skb->mac_header;
1782 #ifdef CONFIG_NET_CLS_ACT
1783 if (skb->tc_verd & TC_NCLS) {
1784 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1789 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1790 if (!ptype->dev || ptype->dev == skb->dev) {
1792 ret = deliver_skb(skb, pt_prev, orig_dev);
1797 #ifdef CONFIG_NET_CLS_ACT
1799 ret = deliver_skb(skb, pt_prev, orig_dev);
1800 pt_prev = NULL; /* noone else should process this after*/
1802 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1805 ret = ing_filter(skb);
1807 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1816 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1819 type = skb->protocol;
1820 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1821 if (ptype->type == type &&
1822 (!ptype->dev || ptype->dev == skb->dev)) {
1824 ret = deliver_skb(skb, pt_prev, orig_dev);
1830 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1833 /* Jamal, now you will not able to escape explaining
1834 * me how you were going to use this. :-)
1844 static int process_backlog(struct net_device *backlog_dev, int *budget)
1847 int quota = min(backlog_dev->quota, *budget);
1848 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1849 unsigned long start_time = jiffies;
1851 backlog_dev->weight = weight_p;
1853 struct sk_buff *skb;
1854 struct net_device *dev;
1856 local_irq_disable();
1857 skb = __skb_dequeue(&queue->input_pkt_queue);
1864 netif_receive_skb(skb);
1870 if (work >= quota || jiffies - start_time > 1)
1875 backlog_dev->quota -= work;
1880 backlog_dev->quota -= work;
1883 list_del(&backlog_dev->poll_list);
1884 smp_mb__before_clear_bit();
1885 netif_poll_enable(backlog_dev);
1891 static void net_rx_action(struct softirq_action *h)
1893 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1894 unsigned long start_time = jiffies;
1895 int budget = netdev_budget;
1898 local_irq_disable();
1900 while (!list_empty(&queue->poll_list)) {
1901 struct net_device *dev;
1903 if (budget <= 0 || jiffies - start_time > 1)
1908 dev = list_entry(queue->poll_list.next,
1909 struct net_device, poll_list);
1910 have = netpoll_poll_lock(dev);
1912 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1913 netpoll_poll_unlock(have);
1914 local_irq_disable();
1915 list_move_tail(&dev->poll_list, &queue->poll_list);
1917 dev->quota += dev->weight;
1919 dev->quota = dev->weight;
1921 netpoll_poll_unlock(have);
1923 local_irq_disable();
1927 #ifdef CONFIG_NET_DMA
1929 * There may not be any more sk_buffs coming right now, so push
1930 * any pending DMA copies to hardware
1932 if (net_dma_client) {
1933 struct dma_chan *chan;
1935 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1936 dma_async_memcpy_issue_pending(chan);
1944 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1945 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1949 static gifconf_func_t * gifconf_list [NPROTO];
1952 * register_gifconf - register a SIOCGIF handler
1953 * @family: Address family
1954 * @gifconf: Function handler
1956 * Register protocol dependent address dumping routines. The handler
1957 * that is passed must not be freed or reused until it has been replaced
1958 * by another handler.
1960 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1962 if (family >= NPROTO)
1964 gifconf_list[family] = gifconf;
1970 * Map an interface index to its name (SIOCGIFNAME)
1974 * We need this ioctl for efficient implementation of the
1975 * if_indextoname() function required by the IPv6 API. Without
1976 * it, we would have to search all the interfaces to find a
1980 static int dev_ifname(struct ifreq __user *arg)
1982 struct net_device *dev;
1986 * Fetch the caller's info block.
1989 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1992 read_lock(&dev_base_lock);
1993 dev = __dev_get_by_index(ifr.ifr_ifindex);
1995 read_unlock(&dev_base_lock);
1999 strcpy(ifr.ifr_name, dev->name);
2000 read_unlock(&dev_base_lock);
2002 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2008 * Perform a SIOCGIFCONF call. This structure will change
2009 * size eventually, and there is nothing I can do about it.
2010 * Thus we will need a 'compatibility mode'.
2013 static int dev_ifconf(char __user *arg)
2016 struct net_device *dev;
2023 * Fetch the caller's info block.
2026 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2033 * Loop over the interfaces, and write an info block for each.
2037 for (dev = dev_base; dev; dev = dev->next) {
2038 for (i = 0; i < NPROTO; i++) {
2039 if (gifconf_list[i]) {
2042 done = gifconf_list[i](dev, NULL, 0);
2044 done = gifconf_list[i](dev, pos + total,
2054 * All done. Write the updated control block back to the caller.
2056 ifc.ifc_len = total;
2059 * Both BSD and Solaris return 0 here, so we do too.
2061 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2064 #ifdef CONFIG_PROC_FS
2066 * This is invoked by the /proc filesystem handler to display a device
2069 static struct net_device *dev_get_idx(loff_t pos)
2071 struct net_device *dev;
2074 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2076 return i == pos ? dev : NULL;
2079 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2081 read_lock(&dev_base_lock);
2082 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2085 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2088 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2091 void dev_seq_stop(struct seq_file *seq, void *v)
2093 read_unlock(&dev_base_lock);
2096 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2098 struct net_device_stats *stats = dev->get_stats(dev);
2101 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2102 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2103 dev->name, stats->rx_bytes, stats->rx_packets,
2105 stats->rx_dropped + stats->rx_missed_errors,
2106 stats->rx_fifo_errors,
2107 stats->rx_length_errors + stats->rx_over_errors +
2108 stats->rx_crc_errors + stats->rx_frame_errors,
2109 stats->rx_compressed, stats->multicast,
2110 stats->tx_bytes, stats->tx_packets,
2111 stats->tx_errors, stats->tx_dropped,
2112 stats->tx_fifo_errors, stats->collisions,
2113 stats->tx_carrier_errors +
2114 stats->tx_aborted_errors +
2115 stats->tx_window_errors +
2116 stats->tx_heartbeat_errors,
2117 stats->tx_compressed);
2119 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2123 * Called from the PROCfs module. This now uses the new arbitrary sized
2124 * /proc/net interface to create /proc/net/dev
2126 static int dev_seq_show(struct seq_file *seq, void *v)
2128 if (v == SEQ_START_TOKEN)
2129 seq_puts(seq, "Inter-| Receive "
2131 " face |bytes packets errs drop fifo frame "
2132 "compressed multicast|bytes packets errs "
2133 "drop fifo colls carrier compressed\n");
2135 dev_seq_printf_stats(seq, v);
2139 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2141 struct netif_rx_stats *rc = NULL;
2143 while (*pos < NR_CPUS)
2144 if (cpu_online(*pos)) {
2145 rc = &per_cpu(netdev_rx_stat, *pos);
2152 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2154 return softnet_get_online(pos);
2157 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2160 return softnet_get_online(pos);
2163 static void softnet_seq_stop(struct seq_file *seq, void *v)
2167 static int softnet_seq_show(struct seq_file *seq, void *v)
2169 struct netif_rx_stats *s = v;
2171 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2172 s->total, s->dropped, s->time_squeeze, 0,
2173 0, 0, 0, 0, /* was fastroute */
2178 static const struct seq_operations dev_seq_ops = {
2179 .start = dev_seq_start,
2180 .next = dev_seq_next,
2181 .stop = dev_seq_stop,
2182 .show = dev_seq_show,
2185 static int dev_seq_open(struct inode *inode, struct file *file)
2187 return seq_open(file, &dev_seq_ops);
2190 static const struct file_operations dev_seq_fops = {
2191 .owner = THIS_MODULE,
2192 .open = dev_seq_open,
2194 .llseek = seq_lseek,
2195 .release = seq_release,
2198 static const struct seq_operations softnet_seq_ops = {
2199 .start = softnet_seq_start,
2200 .next = softnet_seq_next,
2201 .stop = softnet_seq_stop,
2202 .show = softnet_seq_show,
2205 static int softnet_seq_open(struct inode *inode, struct file *file)
2207 return seq_open(file, &softnet_seq_ops);
2210 static const struct file_operations softnet_seq_fops = {
2211 .owner = THIS_MODULE,
2212 .open = softnet_seq_open,
2214 .llseek = seq_lseek,
2215 .release = seq_release,
2218 static void *ptype_get_idx(loff_t pos)
2220 struct packet_type *pt = NULL;
2224 list_for_each_entry_rcu(pt, &ptype_all, list) {
2230 for (t = 0; t < 16; t++) {
2231 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2240 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2243 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2246 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2248 struct packet_type *pt;
2249 struct list_head *nxt;
2253 if (v == SEQ_START_TOKEN)
2254 return ptype_get_idx(0);
2257 nxt = pt->list.next;
2258 if (pt->type == htons(ETH_P_ALL)) {
2259 if (nxt != &ptype_all)
2262 nxt = ptype_base[0].next;
2264 hash = ntohs(pt->type) & 15;
2266 while (nxt == &ptype_base[hash]) {
2269 nxt = ptype_base[hash].next;
2272 return list_entry(nxt, struct packet_type, list);
2275 static void ptype_seq_stop(struct seq_file *seq, void *v)
2280 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2282 #ifdef CONFIG_KALLSYMS
2283 unsigned long offset = 0, symsize;
2284 const char *symname;
2288 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2295 modname = delim = "";
2296 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2302 seq_printf(seq, "[%p]", sym);
2305 static int ptype_seq_show(struct seq_file *seq, void *v)
2307 struct packet_type *pt = v;
2309 if (v == SEQ_START_TOKEN)
2310 seq_puts(seq, "Type Device Function\n");
2312 if (pt->type == htons(ETH_P_ALL))
2313 seq_puts(seq, "ALL ");
2315 seq_printf(seq, "%04x", ntohs(pt->type));
2317 seq_printf(seq, " %-8s ",
2318 pt->dev ? pt->dev->name : "");
2319 ptype_seq_decode(seq, pt->func);
2320 seq_putc(seq, '\n');
2326 static const struct seq_operations ptype_seq_ops = {
2327 .start = ptype_seq_start,
2328 .next = ptype_seq_next,
2329 .stop = ptype_seq_stop,
2330 .show = ptype_seq_show,
2333 static int ptype_seq_open(struct inode *inode, struct file *file)
2335 return seq_open(file, &ptype_seq_ops);
2338 static const struct file_operations ptype_seq_fops = {
2339 .owner = THIS_MODULE,
2340 .open = ptype_seq_open,
2342 .llseek = seq_lseek,
2343 .release = seq_release,
2347 #ifdef CONFIG_WIRELESS_EXT
2348 extern int wireless_proc_init(void);
2350 #define wireless_proc_init() 0
2353 static int __init dev_proc_init(void)
2357 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2359 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2361 if (!proc_net_fops_create("ptype", S_IRUGO, &ptype_seq_fops))
2364 if (wireless_proc_init())
2370 proc_net_remove("softnet_stat");
2372 proc_net_remove("ptype");
2374 proc_net_remove("dev");
2378 #define dev_proc_init() 0
2379 #endif /* CONFIG_PROC_FS */
2383 * netdev_set_master - set up master/slave pair
2384 * @slave: slave device
2385 * @master: new master device
2387 * Changes the master device of the slave. Pass %NULL to break the
2388 * bonding. The caller must hold the RTNL semaphore. On a failure
2389 * a negative errno code is returned. On success the reference counts
2390 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2391 * function returns zero.
2393 int netdev_set_master(struct net_device *slave, struct net_device *master)
2395 struct net_device *old = slave->master;
2405 slave->master = master;
2413 slave->flags |= IFF_SLAVE;
2415 slave->flags &= ~IFF_SLAVE;
2417 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2422 * dev_set_promiscuity - update promiscuity count on a device
2426 * Add or remove promiscuity from a device. While the count in the device
2427 * remains above zero the interface remains promiscuous. Once it hits zero
2428 * the device reverts back to normal filtering operation. A negative inc
2429 * value is used to drop promiscuity on the device.
2431 void dev_set_promiscuity(struct net_device *dev, int inc)
2433 unsigned short old_flags = dev->flags;
2435 if ((dev->promiscuity += inc) == 0)
2436 dev->flags &= ~IFF_PROMISC;
2438 dev->flags |= IFF_PROMISC;
2439 if (dev->flags != old_flags) {
2441 printk(KERN_INFO "device %s %s promiscuous mode\n",
2442 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2444 audit_log(current->audit_context, GFP_ATOMIC,
2445 AUDIT_ANOM_PROMISCUOUS,
2446 "dev=%s prom=%d old_prom=%d auid=%u",
2447 dev->name, (dev->flags & IFF_PROMISC),
2448 (old_flags & IFF_PROMISC),
2449 audit_get_loginuid(current->audit_context));
2454 * dev_set_allmulti - update allmulti count on a device
2458 * Add or remove reception of all multicast frames to a device. While the
2459 * count in the device remains above zero the interface remains listening
2460 * to all interfaces. Once it hits zero the device reverts back to normal
2461 * filtering operation. A negative @inc value is used to drop the counter
2462 * when releasing a resource needing all multicasts.
2465 void dev_set_allmulti(struct net_device *dev, int inc)
2467 unsigned short old_flags = dev->flags;
2469 dev->flags |= IFF_ALLMULTI;
2470 if ((dev->allmulti += inc) == 0)
2471 dev->flags &= ~IFF_ALLMULTI;
2472 if (dev->flags ^ old_flags)
2476 unsigned dev_get_flags(const struct net_device *dev)
2480 flags = (dev->flags & ~(IFF_PROMISC |
2485 (dev->gflags & (IFF_PROMISC |
2488 if (netif_running(dev)) {
2489 if (netif_oper_up(dev))
2490 flags |= IFF_RUNNING;
2491 if (netif_carrier_ok(dev))
2492 flags |= IFF_LOWER_UP;
2493 if (netif_dormant(dev))
2494 flags |= IFF_DORMANT;
2500 int dev_change_flags(struct net_device *dev, unsigned flags)
2503 int old_flags = dev->flags;
2506 * Set the flags on our device.
2509 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2510 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2512 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2516 * Load in the correct multicast list now the flags have changed.
2522 * Have we downed the interface. We handle IFF_UP ourselves
2523 * according to user attempts to set it, rather than blindly
2528 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2529 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2535 if (dev->flags & IFF_UP &&
2536 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2538 raw_notifier_call_chain(&netdev_chain,
2539 NETDEV_CHANGE, dev);
2541 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2542 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2543 dev->gflags ^= IFF_PROMISC;
2544 dev_set_promiscuity(dev, inc);
2547 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2548 is important. Some (broken) drivers set IFF_PROMISC, when
2549 IFF_ALLMULTI is requested not asking us and not reporting.
2551 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2552 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2553 dev->gflags ^= IFF_ALLMULTI;
2554 dev_set_allmulti(dev, inc);
2557 if (old_flags ^ dev->flags)
2558 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2563 int dev_set_mtu(struct net_device *dev, int new_mtu)
2567 if (new_mtu == dev->mtu)
2570 /* MTU must be positive. */
2574 if (!netif_device_present(dev))
2578 if (dev->change_mtu)
2579 err = dev->change_mtu(dev, new_mtu);
2582 if (!err && dev->flags & IFF_UP)
2583 raw_notifier_call_chain(&netdev_chain,
2584 NETDEV_CHANGEMTU, dev);
2588 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2592 if (!dev->set_mac_address)
2594 if (sa->sa_family != dev->type)
2596 if (!netif_device_present(dev))
2598 err = dev->set_mac_address(dev, sa);
2600 raw_notifier_call_chain(&netdev_chain,
2601 NETDEV_CHANGEADDR, dev);
2606 * Perform the SIOCxIFxxx calls.
2608 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2611 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2617 case SIOCGIFFLAGS: /* Get interface flags */
2618 ifr->ifr_flags = dev_get_flags(dev);
2621 case SIOCSIFFLAGS: /* Set interface flags */
2622 return dev_change_flags(dev, ifr->ifr_flags);
2624 case SIOCGIFMETRIC: /* Get the metric on the interface
2625 (currently unused) */
2626 ifr->ifr_metric = 0;
2629 case SIOCSIFMETRIC: /* Set the metric on the interface
2630 (currently unused) */
2633 case SIOCGIFMTU: /* Get the MTU of a device */
2634 ifr->ifr_mtu = dev->mtu;
2637 case SIOCSIFMTU: /* Set the MTU of a device */
2638 return dev_set_mtu(dev, ifr->ifr_mtu);
2642 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2644 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2645 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2646 ifr->ifr_hwaddr.sa_family = dev->type;
2650 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2652 case SIOCSIFHWBROADCAST:
2653 if (ifr->ifr_hwaddr.sa_family != dev->type)
2655 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2656 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2657 raw_notifier_call_chain(&netdev_chain,
2658 NETDEV_CHANGEADDR, dev);
2662 ifr->ifr_map.mem_start = dev->mem_start;
2663 ifr->ifr_map.mem_end = dev->mem_end;
2664 ifr->ifr_map.base_addr = dev->base_addr;
2665 ifr->ifr_map.irq = dev->irq;
2666 ifr->ifr_map.dma = dev->dma;
2667 ifr->ifr_map.port = dev->if_port;
2671 if (dev->set_config) {
2672 if (!netif_device_present(dev))
2674 return dev->set_config(dev, &ifr->ifr_map);
2679 if (!dev->set_multicast_list ||
2680 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2682 if (!netif_device_present(dev))
2684 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2688 if (!dev->set_multicast_list ||
2689 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2691 if (!netif_device_present(dev))
2693 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2697 ifr->ifr_ifindex = dev->ifindex;
2701 ifr->ifr_qlen = dev->tx_queue_len;
2705 if (ifr->ifr_qlen < 0)
2707 dev->tx_queue_len = ifr->ifr_qlen;
2711 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2712 return dev_change_name(dev, ifr->ifr_newname);
2715 * Unknown or private ioctl
2719 if ((cmd >= SIOCDEVPRIVATE &&
2720 cmd <= SIOCDEVPRIVATE + 15) ||
2721 cmd == SIOCBONDENSLAVE ||
2722 cmd == SIOCBONDRELEASE ||
2723 cmd == SIOCBONDSETHWADDR ||
2724 cmd == SIOCBONDSLAVEINFOQUERY ||
2725 cmd == SIOCBONDINFOQUERY ||
2726 cmd == SIOCBONDCHANGEACTIVE ||
2727 cmd == SIOCGMIIPHY ||
2728 cmd == SIOCGMIIREG ||
2729 cmd == SIOCSMIIREG ||
2730 cmd == SIOCBRADDIF ||
2731 cmd == SIOCBRDELIF ||
2732 cmd == SIOCWANDEV) {
2734 if (dev->do_ioctl) {
2735 if (netif_device_present(dev))
2736 err = dev->do_ioctl(dev, ifr,
2749 * This function handles all "interface"-type I/O control requests. The actual
2750 * 'doing' part of this is dev_ifsioc above.
2754 * dev_ioctl - network device ioctl
2755 * @cmd: command to issue
2756 * @arg: pointer to a struct ifreq in user space
2758 * Issue ioctl functions to devices. This is normally called by the
2759 * user space syscall interfaces but can sometimes be useful for
2760 * other purposes. The return value is the return from the syscall if
2761 * positive or a negative errno code on error.
2764 int dev_ioctl(unsigned int cmd, void __user *arg)
2770 /* One special case: SIOCGIFCONF takes ifconf argument
2771 and requires shared lock, because it sleeps writing
2775 if (cmd == SIOCGIFCONF) {
2777 ret = dev_ifconf((char __user *) arg);
2781 if (cmd == SIOCGIFNAME)
2782 return dev_ifname((struct ifreq __user *)arg);
2784 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2787 ifr.ifr_name[IFNAMSIZ-1] = 0;
2789 colon = strchr(ifr.ifr_name, ':');
2794 * See which interface the caller is talking about.
2799 * These ioctl calls:
2800 * - can be done by all.
2801 * - atomic and do not require locking.
2812 dev_load(ifr.ifr_name);
2813 read_lock(&dev_base_lock);
2814 ret = dev_ifsioc(&ifr, cmd);
2815 read_unlock(&dev_base_lock);
2819 if (copy_to_user(arg, &ifr,
2820 sizeof(struct ifreq)))
2826 dev_load(ifr.ifr_name);
2828 ret = dev_ethtool(&ifr);
2833 if (copy_to_user(arg, &ifr,
2834 sizeof(struct ifreq)))
2840 * These ioctl calls:
2841 * - require superuser power.
2842 * - require strict serialization.
2848 if (!capable(CAP_NET_ADMIN))
2850 dev_load(ifr.ifr_name);
2852 ret = dev_ifsioc(&ifr, cmd);
2857 if (copy_to_user(arg, &ifr,
2858 sizeof(struct ifreq)))
2864 * These ioctl calls:
2865 * - require superuser power.
2866 * - require strict serialization.
2867 * - do not return a value
2877 case SIOCSIFHWBROADCAST:
2880 case SIOCBONDENSLAVE:
2881 case SIOCBONDRELEASE:
2882 case SIOCBONDSETHWADDR:
2883 case SIOCBONDCHANGEACTIVE:
2886 if (!capable(CAP_NET_ADMIN))
2889 case SIOCBONDSLAVEINFOQUERY:
2890 case SIOCBONDINFOQUERY:
2891 dev_load(ifr.ifr_name);
2893 ret = dev_ifsioc(&ifr, cmd);
2898 /* Get the per device memory space. We can add this but
2899 * currently do not support it */
2901 /* Set the per device memory buffer space.
2902 * Not applicable in our case */
2907 * Unknown or private ioctl.
2910 if (cmd == SIOCWANDEV ||
2911 (cmd >= SIOCDEVPRIVATE &&
2912 cmd <= SIOCDEVPRIVATE + 15)) {
2913 dev_load(ifr.ifr_name);
2915 ret = dev_ifsioc(&ifr, cmd);
2917 if (!ret && copy_to_user(arg, &ifr,
2918 sizeof(struct ifreq)))
2922 #ifdef CONFIG_WIRELESS_EXT
2923 /* Take care of Wireless Extensions */
2924 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2925 /* If command is `set a parameter', or
2926 * `get the encoding parameters', check if
2927 * the user has the right to do it */
2928 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2929 || cmd == SIOCGIWENCODEEXT) {
2930 if (!capable(CAP_NET_ADMIN))
2933 dev_load(ifr.ifr_name);
2935 /* Follow me in net/core/wireless.c */
2936 ret = wireless_process_ioctl(&ifr, cmd);
2938 if (IW_IS_GET(cmd) &&
2939 copy_to_user(arg, &ifr,
2940 sizeof(struct ifreq)))
2944 #endif /* CONFIG_WIRELESS_EXT */
2951 * dev_new_index - allocate an ifindex
2953 * Returns a suitable unique value for a new device interface
2954 * number. The caller must hold the rtnl semaphore or the
2955 * dev_base_lock to be sure it remains unique.
2957 static int dev_new_index(void)
2963 if (!__dev_get_by_index(ifindex))
2968 static int dev_boot_phase = 1;
2970 /* Delayed registration/unregisteration */
2971 static DEFINE_SPINLOCK(net_todo_list_lock);
2972 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2974 static void net_set_todo(struct net_device *dev)
2976 spin_lock(&net_todo_list_lock);
2977 list_add_tail(&dev->todo_list, &net_todo_list);
2978 spin_unlock(&net_todo_list_lock);
2982 * register_netdevice - register a network device
2983 * @dev: device to register
2985 * Take a completed network device structure and add it to the kernel
2986 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2987 * chain. 0 is returned on success. A negative errno code is returned
2988 * on a failure to set up the device, or if the name is a duplicate.
2990 * Callers must hold the rtnl semaphore. You may want
2991 * register_netdev() instead of this.
2994 * The locking appears insufficient to guarantee two parallel registers
2995 * will not get the same name.
2998 int register_netdevice(struct net_device *dev)
3000 struct hlist_head *head;
3001 struct hlist_node *p;
3004 BUG_ON(dev_boot_phase);
3009 /* When net_device's are persistent, this will be fatal. */
3010 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3012 spin_lock_init(&dev->queue_lock);
3013 spin_lock_init(&dev->_xmit_lock);
3014 dev->xmit_lock_owner = -1;
3015 #ifdef CONFIG_NET_CLS_ACT
3016 spin_lock_init(&dev->ingress_lock);
3021 /* Init, if this function is available */
3023 ret = dev->init(dev);
3031 if (!dev_valid_name(dev->name)) {
3036 dev->ifindex = dev_new_index();
3037 if (dev->iflink == -1)
3038 dev->iflink = dev->ifindex;
3040 /* Check for existence of name */
3041 head = dev_name_hash(dev->name);
3042 hlist_for_each(p, head) {
3043 struct net_device *d
3044 = hlist_entry(p, struct net_device, name_hlist);
3045 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3051 /* Fix illegal SG+CSUM combinations. */
3052 if ((dev->features & NETIF_F_SG) &&
3053 !(dev->features & NETIF_F_ALL_CSUM)) {
3054 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3056 dev->features &= ~NETIF_F_SG;
3059 /* TSO requires that SG is present as well. */
3060 if ((dev->features & NETIF_F_TSO) &&
3061 !(dev->features & NETIF_F_SG)) {
3062 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3064 dev->features &= ~NETIF_F_TSO;
3066 if (dev->features & NETIF_F_UFO) {
3067 if (!(dev->features & NETIF_F_HW_CSUM)) {
3068 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3069 "NETIF_F_HW_CSUM feature.\n",
3071 dev->features &= ~NETIF_F_UFO;
3073 if (!(dev->features & NETIF_F_SG)) {
3074 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3075 "NETIF_F_SG feature.\n",
3077 dev->features &= ~NETIF_F_UFO;
3082 * nil rebuild_header routine,
3083 * that should be never called and used as just bug trap.
3086 if (!dev->rebuild_header)
3087 dev->rebuild_header = default_rebuild_header;
3089 ret = netdev_register_sysfs(dev);
3092 dev->reg_state = NETREG_REGISTERED;
3095 * Default initial state at registry is that the
3096 * device is present.
3099 set_bit(__LINK_STATE_PRESENT, &dev->state);
3102 dev_init_scheduler(dev);
3103 write_lock_bh(&dev_base_lock);
3105 dev_tail = &dev->next;
3106 hlist_add_head(&dev->name_hlist, head);
3107 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3109 write_unlock_bh(&dev_base_lock);
3111 /* Notify protocols, that a new device appeared. */
3112 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3121 * register_netdev - register a network device
3122 * @dev: device to register
3124 * Take a completed network device structure and add it to the kernel
3125 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3126 * chain. 0 is returned on success. A negative errno code is returned
3127 * on a failure to set up the device, or if the name is a duplicate.
3129 * This is a wrapper around register_netdev that takes the rtnl semaphore
3130 * and expands the device name if you passed a format string to
3133 int register_netdev(struct net_device *dev)
3140 * If the name is a format string the caller wants us to do a
3143 if (strchr(dev->name, '%')) {
3144 err = dev_alloc_name(dev, dev->name);
3149 err = register_netdevice(dev);
3154 EXPORT_SYMBOL(register_netdev);
3157 * netdev_wait_allrefs - wait until all references are gone.
3159 * This is called when unregistering network devices.
3161 * Any protocol or device that holds a reference should register
3162 * for netdevice notification, and cleanup and put back the
3163 * reference if they receive an UNREGISTER event.
3164 * We can get stuck here if buggy protocols don't correctly
3167 static void netdev_wait_allrefs(struct net_device *dev)
3169 unsigned long rebroadcast_time, warning_time;
3171 rebroadcast_time = warning_time = jiffies;
3172 while (atomic_read(&dev->refcnt) != 0) {
3173 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3176 /* Rebroadcast unregister notification */
3177 raw_notifier_call_chain(&netdev_chain,
3178 NETDEV_UNREGISTER, dev);
3180 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3182 /* We must not have linkwatch events
3183 * pending on unregister. If this
3184 * happens, we simply run the queue
3185 * unscheduled, resulting in a noop
3188 linkwatch_run_queue();
3193 rebroadcast_time = jiffies;
3198 if (time_after(jiffies, warning_time + 10 * HZ)) {
3199 printk(KERN_EMERG "unregister_netdevice: "
3200 "waiting for %s to become free. Usage "
3202 dev->name, atomic_read(&dev->refcnt));
3203 warning_time = jiffies;
3212 * register_netdevice(x1);
3213 * register_netdevice(x2);
3215 * unregister_netdevice(y1);
3216 * unregister_netdevice(y2);
3222 * We are invoked by rtnl_unlock() after it drops the semaphore.
3223 * This allows us to deal with problems:
3224 * 1) We can delete sysfs objects which invoke hotplug
3225 * without deadlocking with linkwatch via keventd.
3226 * 2) Since we run with the RTNL semaphore not held, we can sleep
3227 * safely in order to wait for the netdev refcnt to drop to zero.
3229 static DEFINE_MUTEX(net_todo_run_mutex);
3230 void netdev_run_todo(void)
3232 struct list_head list;
3234 /* Need to guard against multiple cpu's getting out of order. */
3235 mutex_lock(&net_todo_run_mutex);
3237 /* Not safe to do outside the semaphore. We must not return
3238 * until all unregister events invoked by the local processor
3239 * have been completed (either by this todo run, or one on
3242 if (list_empty(&net_todo_list))
3245 /* Snapshot list, allow later requests */
3246 spin_lock(&net_todo_list_lock);
3247 list_replace_init(&net_todo_list, &list);
3248 spin_unlock(&net_todo_list_lock);
3250 while (!list_empty(&list)) {
3251 struct net_device *dev
3252 = list_entry(list.next, struct net_device, todo_list);
3253 list_del(&dev->todo_list);
3255 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3256 printk(KERN_ERR "network todo '%s' but state %d\n",
3257 dev->name, dev->reg_state);
3262 netdev_unregister_sysfs(dev);
3263 dev->reg_state = NETREG_UNREGISTERED;
3265 netdev_wait_allrefs(dev);
3268 BUG_ON(atomic_read(&dev->refcnt));
3269 BUG_TRAP(!dev->ip_ptr);
3270 BUG_TRAP(!dev->ip6_ptr);
3271 BUG_TRAP(!dev->dn_ptr);
3273 /* It must be the very last action,
3274 * after this 'dev' may point to freed up memory.
3276 if (dev->destructor)
3277 dev->destructor(dev);
3281 mutex_unlock(&net_todo_run_mutex);
3284 static struct net_device_stats *maybe_internal_stats(struct net_device *dev)
3286 if (dev->features & NETIF_F_INTERNAL_STATS)
3292 * alloc_netdev - allocate network device
3293 * @sizeof_priv: size of private data to allocate space for
3294 * @name: device name format string
3295 * @setup: callback to initialize device
3297 * Allocates a struct net_device with private data area for driver use
3298 * and performs basic initialization.
3300 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3301 void (*setup)(struct net_device *))
3304 struct net_device *dev;
3307 BUG_ON(strlen(name) >= sizeof(dev->name));
3309 /* ensure 32-byte alignment of both the device and private area */
3310 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3311 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3313 p = kzalloc(alloc_size, GFP_KERNEL);
3315 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3319 dev = (struct net_device *)
3320 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3321 dev->padded = (char *)dev - (char *)p;
3324 dev->priv = netdev_priv(dev);
3326 dev->get_stats = maybe_internal_stats;
3328 strcpy(dev->name, name);
3331 EXPORT_SYMBOL(alloc_netdev);
3334 * free_netdev - free network device
3337 * This function does the last stage of destroying an allocated device
3338 * interface. The reference to the device object is released.
3339 * If this is the last reference then it will be freed.
3341 void free_netdev(struct net_device *dev)
3344 /* Compatibility with error handling in drivers */
3345 if (dev->reg_state == NETREG_UNINITIALIZED) {
3346 kfree((char *)dev - dev->padded);
3350 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3351 dev->reg_state = NETREG_RELEASED;
3353 /* will free via device release */
3354 put_device(&dev->dev);
3356 kfree((char *)dev - dev->padded);
3360 /* Synchronize with packet receive processing. */
3361 void synchronize_net(void)
3368 * unregister_netdevice - remove device from the kernel
3371 * This function shuts down a device interface and removes it
3372 * from the kernel tables. On success 0 is returned, on a failure
3373 * a negative errno code is returned.
3375 * Callers must hold the rtnl semaphore. You may want
3376 * unregister_netdev() instead of this.
3379 void unregister_netdevice(struct net_device *dev)
3381 struct net_device *d, **dp;
3383 BUG_ON(dev_boot_phase);
3386 /* Some devices call without registering for initialization unwind. */
3387 if (dev->reg_state == NETREG_UNINITIALIZED) {
3388 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3389 "was registered\n", dev->name, dev);
3395 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3397 /* If device is running, close it first. */
3398 if (dev->flags & IFF_UP)
3401 /* And unlink it from device chain. */
3402 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3404 write_lock_bh(&dev_base_lock);
3405 hlist_del(&dev->name_hlist);
3406 hlist_del(&dev->index_hlist);
3407 if (dev_tail == &dev->next)
3410 write_unlock_bh(&dev_base_lock);
3416 dev->reg_state = NETREG_UNREGISTERING;
3420 /* Shutdown queueing discipline. */
3424 /* Notify protocols, that we are about to destroy
3425 this device. They should clean all the things.
3427 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3430 * Flush the multicast chain
3432 dev_mc_discard(dev);
3437 /* Notifier chain MUST detach us from master device. */
3438 BUG_TRAP(!dev->master);
3440 /* Finish processing unregister after unlock */
3449 * unregister_netdev - remove device from the kernel
3452 * This function shuts down a device interface and removes it
3453 * from the kernel tables. On success 0 is returned, on a failure
3454 * a negative errno code is returned.
3456 * This is just a wrapper for unregister_netdevice that takes
3457 * the rtnl semaphore. In general you want to use this and not
3458 * unregister_netdevice.
3460 void unregister_netdev(struct net_device *dev)
3463 unregister_netdevice(dev);
3467 EXPORT_SYMBOL(unregister_netdev);
3469 static int dev_cpu_callback(struct notifier_block *nfb,
3470 unsigned long action,
3473 struct sk_buff **list_skb;
3474 struct net_device **list_net;
3475 struct sk_buff *skb;
3476 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3477 struct softnet_data *sd, *oldsd;
3479 if (action != CPU_DEAD)
3482 local_irq_disable();
3483 cpu = smp_processor_id();
3484 sd = &per_cpu(softnet_data, cpu);
3485 oldsd = &per_cpu(softnet_data, oldcpu);
3487 /* Find end of our completion_queue. */
3488 list_skb = &sd->completion_queue;
3490 list_skb = &(*list_skb)->next;
3491 /* Append completion queue from offline CPU. */
3492 *list_skb = oldsd->completion_queue;
3493 oldsd->completion_queue = NULL;
3495 /* Find end of our output_queue. */
3496 list_net = &sd->output_queue;
3498 list_net = &(*list_net)->next_sched;
3499 /* Append output queue from offline CPU. */
3500 *list_net = oldsd->output_queue;
3501 oldsd->output_queue = NULL;
3503 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3506 /* Process offline CPU's input_pkt_queue */
3507 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3513 #ifdef CONFIG_NET_DMA
3515 * net_dma_rebalance -
3516 * This is called when the number of channels allocated to the net_dma_client
3517 * changes. The net_dma_client tries to have one DMA channel per CPU.
3519 static void net_dma_rebalance(void)
3521 unsigned int cpu, i, n;
3522 struct dma_chan *chan;
3524 if (net_dma_count == 0) {
3525 for_each_online_cpu(cpu)
3526 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
3531 cpu = first_cpu(cpu_online_map);
3534 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3535 n = ((num_online_cpus() / net_dma_count)
3536 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3539 per_cpu(softnet_data, cpu).net_dma = chan;
3540 cpu = next_cpu(cpu, cpu_online_map);
3549 * netdev_dma_event - event callback for the net_dma_client
3550 * @client: should always be net_dma_client
3551 * @chan: DMA channel for the event
3552 * @event: event type
3554 static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3555 enum dma_event event)
3557 spin_lock(&net_dma_event_lock);
3559 case DMA_RESOURCE_ADDED:
3561 net_dma_rebalance();
3563 case DMA_RESOURCE_REMOVED:
3565 net_dma_rebalance();
3570 spin_unlock(&net_dma_event_lock);
3574 * netdev_dma_regiser - register the networking subsystem as a DMA client
3576 static int __init netdev_dma_register(void)
3578 spin_lock_init(&net_dma_event_lock);
3579 net_dma_client = dma_async_client_register(netdev_dma_event);
3580 if (net_dma_client == NULL)
3583 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3588 static int __init netdev_dma_register(void) { return -ENODEV; }
3589 #endif /* CONFIG_NET_DMA */
3592 * Initialize the DEV module. At boot time this walks the device list and
3593 * unhooks any devices that fail to initialise (normally hardware not
3594 * present) and leaves us with a valid list of present and active devices.
3599 * This is called single threaded during boot, so no need
3600 * to take the rtnl semaphore.
3602 static int __init net_dev_init(void)
3604 int i, rc = -ENOMEM;
3606 BUG_ON(!dev_boot_phase);
3608 if (dev_proc_init())
3611 if (netdev_sysfs_init())
3614 INIT_LIST_HEAD(&ptype_all);
3615 for (i = 0; i < 16; i++)
3616 INIT_LIST_HEAD(&ptype_base[i]);
3618 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3619 INIT_HLIST_HEAD(&dev_name_head[i]);
3621 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3622 INIT_HLIST_HEAD(&dev_index_head[i]);
3625 * Initialise the packet receive queues.
3628 for_each_possible_cpu(i) {
3629 struct softnet_data *queue;
3631 queue = &per_cpu(softnet_data, i);
3632 skb_queue_head_init(&queue->input_pkt_queue);
3633 queue->completion_queue = NULL;
3634 INIT_LIST_HEAD(&queue->poll_list);
3635 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3636 queue->backlog_dev.weight = weight_p;
3637 queue->backlog_dev.poll = process_backlog;
3638 atomic_set(&queue->backlog_dev.refcnt, 1);
3641 netdev_dma_register();
3645 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3646 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3648 hotcpu_notifier(dev_cpu_callback, 0);
3656 subsys_initcall(net_dev_init);
3658 EXPORT_SYMBOL(__dev_get_by_index);
3659 EXPORT_SYMBOL(__dev_get_by_name);
3660 EXPORT_SYMBOL(__dev_remove_pack);
3661 EXPORT_SYMBOL(dev_valid_name);
3662 EXPORT_SYMBOL(dev_add_pack);
3663 EXPORT_SYMBOL(dev_alloc_name);
3664 EXPORT_SYMBOL(dev_close);
3665 EXPORT_SYMBOL(dev_get_by_flags);
3666 EXPORT_SYMBOL(dev_get_by_index);
3667 EXPORT_SYMBOL(dev_get_by_name);
3668 EXPORT_SYMBOL(dev_open);
3669 EXPORT_SYMBOL(dev_queue_xmit);
3670 EXPORT_SYMBOL(dev_remove_pack);
3671 EXPORT_SYMBOL(dev_set_allmulti);
3672 EXPORT_SYMBOL(dev_set_promiscuity);
3673 EXPORT_SYMBOL(dev_change_flags);
3674 EXPORT_SYMBOL(dev_set_mtu);
3675 EXPORT_SYMBOL(dev_set_mac_address);
3676 EXPORT_SYMBOL(free_netdev);
3677 EXPORT_SYMBOL(netdev_boot_setup_check);
3678 EXPORT_SYMBOL(netdev_set_master);
3679 EXPORT_SYMBOL(netdev_state_change);
3680 EXPORT_SYMBOL(netif_receive_skb);
3681 EXPORT_SYMBOL(netif_rx);
3682 EXPORT_SYMBOL(register_gifconf);
3683 EXPORT_SYMBOL(register_netdevice);
3684 EXPORT_SYMBOL(register_netdevice_notifier);
3685 EXPORT_SYMBOL(skb_checksum_help);
3686 EXPORT_SYMBOL(synchronize_net);
3687 EXPORT_SYMBOL(unregister_netdevice);
3688 EXPORT_SYMBOL(unregister_netdevice_notifier);
3689 EXPORT_SYMBOL(net_enable_timestamp);
3690 EXPORT_SYMBOL(net_disable_timestamp);
3691 EXPORT_SYMBOL(dev_get_flags);
3693 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3694 EXPORT_SYMBOL(br_handle_frame_hook);
3695 EXPORT_SYMBOL(br_fdb_get_hook);
3696 EXPORT_SYMBOL(br_fdb_put_hook);
3700 EXPORT_SYMBOL(dev_load);
3703 EXPORT_PER_CPU_SYMBOL(softnet_data);