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>
101 #include <linux/divert.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/kmod.h>
108 #include <linux/module.h>
109 #include <linux/kallsyms.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #include <linux/wireless.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.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]; /* 16 way hashed list */
150 static struct list_head ptype_all; /* 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
637 int dev_valid_name(const char *name)
639 return !(*name == '\0'
640 || !strcmp(name, ".")
641 || !strcmp(name, "..")
642 || strchr(name, '/'));
646 * dev_alloc_name - allocate a name for a device
648 * @name: name format string
650 * Passed a format string - eg "lt%d" it will try and find a suitable
651 * id. It scans list of devices to build up a free map, then chooses
652 * the first empty slot. The caller must hold the dev_base or rtnl lock
653 * while allocating the name and adding the device in order to avoid
655 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
656 * Returns the number of the unit assigned or a negative errno code.
659 int dev_alloc_name(struct net_device *dev, const char *name)
664 const int max_netdevices = 8*PAGE_SIZE;
666 struct net_device *d;
668 p = strnchr(name, IFNAMSIZ-1, '%');
671 * Verify the string as this thing may have come from
672 * the user. There must be either one "%d" and no other "%"
675 if (p[1] != 'd' || strchr(p + 2, '%'))
678 /* Use one page as a bit array of possible slots */
679 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
683 for (d = dev_base; d; d = d->next) {
684 if (!sscanf(d->name, name, &i))
686 if (i < 0 || i >= max_netdevices)
689 /* avoid cases where sscanf is not exact inverse of printf */
690 snprintf(buf, sizeof(buf), name, i);
691 if (!strncmp(buf, d->name, IFNAMSIZ))
695 i = find_first_zero_bit(inuse, max_netdevices);
696 free_page((unsigned long) inuse);
699 snprintf(buf, sizeof(buf), name, i);
700 if (!__dev_get_by_name(buf)) {
701 strlcpy(dev->name, buf, IFNAMSIZ);
705 /* It is possible to run out of possible slots
706 * when the name is long and there isn't enough space left
707 * for the digits, or if all bits are used.
714 * dev_change_name - change name of a device
716 * @newname: name (or format string) must be at least IFNAMSIZ
718 * Change name of a device, can pass format strings "eth%d".
721 int dev_change_name(struct net_device *dev, char *newname)
727 if (dev->flags & IFF_UP)
730 if (!dev_valid_name(newname))
733 if (strchr(newname, '%')) {
734 err = dev_alloc_name(dev, newname);
737 strcpy(newname, dev->name);
739 else if (__dev_get_by_name(newname))
742 strlcpy(dev->name, newname, IFNAMSIZ);
744 err = class_device_rename(&dev->class_dev, dev->name);
746 hlist_del(&dev->name_hlist);
747 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
748 raw_notifier_call_chain(&netdev_chain,
749 NETDEV_CHANGENAME, dev);
756 * netdev_features_change - device changes features
757 * @dev: device to cause notification
759 * Called to indicate a device has changed features.
761 void netdev_features_change(struct net_device *dev)
763 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
765 EXPORT_SYMBOL(netdev_features_change);
768 * netdev_state_change - device changes state
769 * @dev: device to cause notification
771 * Called to indicate a device has changed state. This function calls
772 * the notifier chains for netdev_chain and sends a NEWLINK message
773 * to the routing socket.
775 void netdev_state_change(struct net_device *dev)
777 if (dev->flags & IFF_UP) {
778 raw_notifier_call_chain(&netdev_chain,
780 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
785 * dev_load - load a network module
786 * @name: name of interface
788 * If a network interface is not present and the process has suitable
789 * privileges this function loads the module. If module loading is not
790 * available in this kernel then it becomes a nop.
793 void dev_load(const char *name)
795 struct net_device *dev;
797 read_lock(&dev_base_lock);
798 dev = __dev_get_by_name(name);
799 read_unlock(&dev_base_lock);
801 if (!dev && capable(CAP_SYS_MODULE))
802 request_module("%s", name);
805 static int default_rebuild_header(struct sk_buff *skb)
807 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
808 skb->dev ? skb->dev->name : "NULL!!!");
815 * dev_open - prepare an interface for use.
816 * @dev: device to open
818 * Takes a device from down to up state. The device's private open
819 * function is invoked and then the multicast lists are loaded. Finally
820 * the device is moved into the up state and a %NETDEV_UP message is
821 * sent to the netdev notifier chain.
823 * Calling this function on an active interface is a nop. On a failure
824 * a negative errno code is returned.
826 int dev_open(struct net_device *dev)
834 if (dev->flags & IFF_UP)
838 * Is it even present?
840 if (!netif_device_present(dev))
844 * Call device private open method
846 set_bit(__LINK_STATE_START, &dev->state);
848 ret = dev->open(dev);
850 clear_bit(__LINK_STATE_START, &dev->state);
854 * If it went open OK then:
861 dev->flags |= IFF_UP;
864 * Initialize multicasting status
869 * Wakeup transmit queue engine
874 * ... and announce new interface.
876 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
882 * dev_close - shutdown an interface.
883 * @dev: device to shutdown
885 * This function moves an active device into down state. A
886 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
887 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
890 int dev_close(struct net_device *dev)
892 if (!(dev->flags & IFF_UP))
896 * Tell people we are going down, so that they can
897 * prepare to death, when device is still operating.
899 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
903 clear_bit(__LINK_STATE_START, &dev->state);
905 /* Synchronize to scheduled poll. We cannot touch poll list,
906 * it can be even on different cpu. So just clear netif_running(),
907 * and wait when poll really will happen. Actually, the best place
908 * for this is inside dev->stop() after device stopped its irq
909 * engine, but this requires more changes in devices. */
911 smp_mb__after_clear_bit(); /* Commit netif_running(). */
912 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
918 * Call the device specific close. This cannot fail.
919 * Only if device is UP
921 * We allow it to be called even after a DETACH hot-plug
928 * Device is now down.
931 dev->flags &= ~IFF_UP;
934 * Tell people we are down
936 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
943 * Device change register/unregister. These are not inline or static
944 * as we export them to the world.
948 * register_netdevice_notifier - register a network notifier block
951 * Register a notifier to be called when network device events occur.
952 * The notifier passed is linked into the kernel structures and must
953 * not be reused until it has been unregistered. A negative errno code
954 * is returned on a failure.
956 * When registered all registration and up events are replayed
957 * to the new notifier to allow device to have a race free
958 * view of the network device list.
961 int register_netdevice_notifier(struct notifier_block *nb)
963 struct net_device *dev;
967 err = raw_notifier_chain_register(&netdev_chain, nb);
969 for (dev = dev_base; dev; dev = dev->next) {
970 nb->notifier_call(nb, NETDEV_REGISTER, dev);
972 if (dev->flags & IFF_UP)
973 nb->notifier_call(nb, NETDEV_UP, dev);
981 * unregister_netdevice_notifier - unregister a network notifier block
984 * Unregister a notifier previously registered by
985 * register_netdevice_notifier(). The notifier is unlinked into the
986 * kernel structures and may then be reused. A negative errno code
987 * is returned on a failure.
990 int unregister_netdevice_notifier(struct notifier_block *nb)
995 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1001 * call_netdevice_notifiers - call all network notifier blocks
1002 * @val: value passed unmodified to notifier function
1003 * @v: pointer passed unmodified to notifier function
1005 * Call all network notifier blocks. Parameters and return value
1006 * are as for raw_notifier_call_chain().
1009 int call_netdevice_notifiers(unsigned long val, void *v)
1011 return raw_notifier_call_chain(&netdev_chain, val, v);
1014 /* When > 0 there are consumers of rx skb time stamps */
1015 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1017 void net_enable_timestamp(void)
1019 atomic_inc(&netstamp_needed);
1022 void net_disable_timestamp(void)
1024 atomic_dec(&netstamp_needed);
1027 void __net_timestamp(struct sk_buff *skb)
1031 do_gettimeofday(&tv);
1032 skb_set_timestamp(skb, &tv);
1034 EXPORT_SYMBOL(__net_timestamp);
1036 static inline void net_timestamp(struct sk_buff *skb)
1038 if (atomic_read(&netstamp_needed))
1039 __net_timestamp(skb);
1041 skb->tstamp.off_sec = 0;
1042 skb->tstamp.off_usec = 0;
1047 * Support routine. Sends outgoing frames to any network
1048 * taps currently in use.
1051 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1053 struct packet_type *ptype;
1058 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1059 /* Never send packets back to the socket
1060 * they originated from - MvS (miquels@drinkel.ow.org)
1062 if ((ptype->dev == dev || !ptype->dev) &&
1063 (ptype->af_packet_priv == NULL ||
1064 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1065 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1069 /* skb->nh should be correctly
1070 set by sender, so that the second statement is
1071 just protection against buggy protocols.
1073 skb2->mac.raw = skb2->data;
1075 if (skb2->nh.raw < skb2->data ||
1076 skb2->nh.raw > skb2->tail) {
1077 if (net_ratelimit())
1078 printk(KERN_CRIT "protocol %04x is "
1080 skb2->protocol, dev->name);
1081 skb2->nh.raw = skb2->data;
1084 skb2->h.raw = skb2->nh.raw;
1085 skb2->pkt_type = PACKET_OUTGOING;
1086 ptype->func(skb2, skb->dev, ptype, skb->dev);
1093 void __netif_schedule(struct net_device *dev)
1095 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1096 unsigned long flags;
1097 struct softnet_data *sd;
1099 local_irq_save(flags);
1100 sd = &__get_cpu_var(softnet_data);
1101 dev->next_sched = sd->output_queue;
1102 sd->output_queue = dev;
1103 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1104 local_irq_restore(flags);
1107 EXPORT_SYMBOL(__netif_schedule);
1109 void __netif_rx_schedule(struct net_device *dev)
1111 unsigned long flags;
1113 local_irq_save(flags);
1115 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1117 dev->quota += dev->weight;
1119 dev->quota = dev->weight;
1120 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1121 local_irq_restore(flags);
1123 EXPORT_SYMBOL(__netif_rx_schedule);
1125 void dev_kfree_skb_any(struct sk_buff *skb)
1127 if (in_irq() || irqs_disabled())
1128 dev_kfree_skb_irq(skb);
1132 EXPORT_SYMBOL(dev_kfree_skb_any);
1136 void netif_device_detach(struct net_device *dev)
1138 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1139 netif_running(dev)) {
1140 netif_stop_queue(dev);
1143 EXPORT_SYMBOL(netif_device_detach);
1145 void netif_device_attach(struct net_device *dev)
1147 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1148 netif_running(dev)) {
1149 netif_wake_queue(dev);
1150 __netdev_watchdog_up(dev);
1153 EXPORT_SYMBOL(netif_device_attach);
1157 * Invalidate hardware checksum when packet is to be mangled, and
1158 * complete checksum manually on outgoing path.
1160 int skb_checksum_help(struct sk_buff *skb, int inward)
1163 int ret = 0, offset = skb->h.raw - skb->data;
1166 goto out_set_summed;
1168 if (unlikely(skb_shinfo(skb)->gso_size)) {
1169 /* Let GSO fix up the checksum. */
1170 goto out_set_summed;
1173 if (skb_cloned(skb)) {
1174 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1179 BUG_ON(offset > (int)skb->len);
1180 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1182 offset = skb->tail - skb->h.raw;
1183 BUG_ON(offset <= 0);
1184 BUG_ON(skb->csum + 2 > offset);
1186 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1189 skb->ip_summed = CHECKSUM_NONE;
1195 * skb_gso_segment - Perform segmentation on skb.
1196 * @skb: buffer to segment
1197 * @features: features for the output path (see dev->features)
1199 * This function segments the given skb and returns a list of segments.
1201 * It may return NULL if the skb requires no segmentation. This is
1202 * only possible when GSO is used for verifying header integrity.
1204 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1206 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1207 struct packet_type *ptype;
1208 int type = skb->protocol;
1211 BUG_ON(skb_shinfo(skb)->frag_list);
1213 skb->mac.raw = skb->data;
1214 skb->mac_len = skb->nh.raw - skb->data;
1215 __skb_pull(skb, skb->mac_len);
1217 if (unlikely(skb->ip_summed != CHECKSUM_HW)) {
1218 if (skb_header_cloned(skb) &&
1219 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1220 return ERR_PTR(err);
1224 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1225 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1226 if (unlikely(skb->ip_summed != CHECKSUM_HW)) {
1227 err = ptype->gso_send_check(skb);
1228 segs = ERR_PTR(err);
1229 if (err || skb_gso_ok(skb, features))
1231 __skb_push(skb, skb->data - skb->nh.raw);
1233 segs = ptype->gso_segment(skb, features);
1239 __skb_push(skb, skb->data - skb->mac.raw);
1244 EXPORT_SYMBOL(skb_gso_segment);
1246 /* Take action when hardware reception checksum errors are detected. */
1248 void netdev_rx_csum_fault(struct net_device *dev)
1250 if (net_ratelimit()) {
1251 printk(KERN_ERR "%s: hw csum failure.\n",
1252 dev ? dev->name : "<unknown>");
1256 EXPORT_SYMBOL(netdev_rx_csum_fault);
1259 /* Actually, we should eliminate this check as soon as we know, that:
1260 * 1. IOMMU is present and allows to map all the memory.
1261 * 2. No high memory really exists on this machine.
1264 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1266 #ifdef CONFIG_HIGHMEM
1269 if (dev->features & NETIF_F_HIGHDMA)
1272 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1273 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1281 void (*destructor)(struct sk_buff *skb);
1284 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1286 static void dev_gso_skb_destructor(struct sk_buff *skb)
1288 struct dev_gso_cb *cb;
1291 struct sk_buff *nskb = skb->next;
1293 skb->next = nskb->next;
1296 } while (skb->next);
1298 cb = DEV_GSO_CB(skb);
1300 cb->destructor(skb);
1304 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1305 * @skb: buffer to segment
1307 * This function segments the given skb and stores the list of segments
1310 static int dev_gso_segment(struct sk_buff *skb)
1312 struct net_device *dev = skb->dev;
1313 struct sk_buff *segs;
1314 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1317 segs = skb_gso_segment(skb, features);
1319 /* Verifying header integrity only. */
1323 if (unlikely(IS_ERR(segs)))
1324 return PTR_ERR(segs);
1327 DEV_GSO_CB(skb)->destructor = skb->destructor;
1328 skb->destructor = dev_gso_skb_destructor;
1333 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1335 if (likely(!skb->next)) {
1337 dev_queue_xmit_nit(skb, dev);
1339 if (netif_needs_gso(dev, skb)) {
1340 if (unlikely(dev_gso_segment(skb)))
1346 return dev->hard_start_xmit(skb, dev);
1351 struct sk_buff *nskb = skb->next;
1354 skb->next = nskb->next;
1356 rc = dev->hard_start_xmit(nskb, dev);
1358 nskb->next = skb->next;
1362 if (unlikely(netif_queue_stopped(dev) && skb->next))
1363 return NETDEV_TX_BUSY;
1364 } while (skb->next);
1366 skb->destructor = DEV_GSO_CB(skb)->destructor;
1373 #define HARD_TX_LOCK(dev, cpu) { \
1374 if ((dev->features & NETIF_F_LLTX) == 0) { \
1375 netif_tx_lock(dev); \
1379 #define HARD_TX_UNLOCK(dev) { \
1380 if ((dev->features & NETIF_F_LLTX) == 0) { \
1381 netif_tx_unlock(dev); \
1386 * dev_queue_xmit - transmit a buffer
1387 * @skb: buffer to transmit
1389 * Queue a buffer for transmission to a network device. The caller must
1390 * have set the device and priority and built the buffer before calling
1391 * this function. The function can be called from an interrupt.
1393 * A negative errno code is returned on a failure. A success does not
1394 * guarantee the frame will be transmitted as it may be dropped due
1395 * to congestion or traffic shaping.
1397 * -----------------------------------------------------------------------------------
1398 * I notice this method can also return errors from the queue disciplines,
1399 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1402 * Regardless of the return value, the skb is consumed, so it is currently
1403 * difficult to retry a send to this method. (You can bump the ref count
1404 * before sending to hold a reference for retry if you are careful.)
1406 * When calling this method, interrupts MUST be enabled. This is because
1407 * the BH enable code must have IRQs enabled so that it will not deadlock.
1411 int dev_queue_xmit(struct sk_buff *skb)
1413 struct net_device *dev = skb->dev;
1417 /* GSO will handle the following emulations directly. */
1418 if (netif_needs_gso(dev, skb))
1421 if (skb_shinfo(skb)->frag_list &&
1422 !(dev->features & NETIF_F_FRAGLIST) &&
1423 __skb_linearize(skb))
1426 /* Fragmented skb is linearized if device does not support SG,
1427 * or if at least one of fragments is in highmem and device
1428 * does not support DMA from it.
1430 if (skb_shinfo(skb)->nr_frags &&
1431 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1432 __skb_linearize(skb))
1435 /* If packet is not checksummed and device does not support
1436 * checksumming for this protocol, complete checksumming here.
1438 if (skb->ip_summed == CHECKSUM_HW &&
1439 (!(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, 0))
1446 spin_lock_prefetch(&dev->queue_lock);
1448 /* Disable soft irqs for various locks below. Also
1449 * stops preemption for RCU.
1453 /* Updates of qdisc are serialized by queue_lock.
1454 * The struct Qdisc which is pointed to by qdisc is now a
1455 * rcu structure - it may be accessed without acquiring
1456 * a lock (but the structure may be stale.) The freeing of the
1457 * qdisc will be deferred until it's known that there are no
1458 * more references to it.
1460 * If the qdisc has an enqueue function, we still need to
1461 * hold the queue_lock before calling it, since queue_lock
1462 * also serializes access to the device queue.
1465 q = rcu_dereference(dev->qdisc);
1466 #ifdef CONFIG_NET_CLS_ACT
1467 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1470 /* Grab device queue */
1471 spin_lock(&dev->queue_lock);
1473 rc = q->enqueue(skb, q);
1477 spin_unlock(&dev->queue_lock);
1478 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1482 /* The device has no queue. Common case for software devices:
1483 loopback, all the sorts of tunnels...
1485 Really, it is unlikely that netif_tx_lock protection is necessary
1486 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1488 However, it is possible, that they rely on protection
1491 Check this and shot the lock. It is not prone from deadlocks.
1492 Either shot noqueue qdisc, it is even simpler 8)
1494 if (dev->flags & IFF_UP) {
1495 int cpu = smp_processor_id(); /* ok because BHs are off */
1497 if (dev->xmit_lock_owner != cpu) {
1499 HARD_TX_LOCK(dev, cpu);
1501 if (!netif_queue_stopped(dev)) {
1503 if (!dev_hard_start_xmit(skb, dev)) {
1504 HARD_TX_UNLOCK(dev);
1508 HARD_TX_UNLOCK(dev);
1509 if (net_ratelimit())
1510 printk(KERN_CRIT "Virtual device %s asks to "
1511 "queue packet!\n", dev->name);
1513 /* Recursion is detected! It is possible,
1515 if (net_ratelimit())
1516 printk(KERN_CRIT "Dead loop on virtual device "
1517 "%s, fix it urgently!\n", dev->name);
1522 rcu_read_unlock_bh();
1528 rcu_read_unlock_bh();
1533 /*=======================================================================
1535 =======================================================================*/
1537 int netdev_max_backlog = 1000;
1538 int netdev_budget = 300;
1539 int weight_p = 64; /* old backlog weight */
1541 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1545 * netif_rx - post buffer to the network code
1546 * @skb: buffer to post
1548 * This function receives a packet from a device driver and queues it for
1549 * the upper (protocol) levels to process. It always succeeds. The buffer
1550 * may be dropped during processing for congestion control or by the
1554 * NET_RX_SUCCESS (no congestion)
1555 * NET_RX_CN_LOW (low congestion)
1556 * NET_RX_CN_MOD (moderate congestion)
1557 * NET_RX_CN_HIGH (high congestion)
1558 * NET_RX_DROP (packet was dropped)
1562 int netif_rx(struct sk_buff *skb)
1564 struct softnet_data *queue;
1565 unsigned long flags;
1567 /* if netpoll wants it, pretend we never saw it */
1568 if (netpoll_rx(skb))
1571 if (!skb->tstamp.off_sec)
1575 * The code is rearranged so that the path is the most
1576 * short when CPU is congested, but is still operating.
1578 local_irq_save(flags);
1579 queue = &__get_cpu_var(softnet_data);
1581 __get_cpu_var(netdev_rx_stat).total++;
1582 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1583 if (queue->input_pkt_queue.qlen) {
1586 __skb_queue_tail(&queue->input_pkt_queue, skb);
1587 local_irq_restore(flags);
1588 return NET_RX_SUCCESS;
1591 netif_rx_schedule(&queue->backlog_dev);
1595 __get_cpu_var(netdev_rx_stat).dropped++;
1596 local_irq_restore(flags);
1602 int netif_rx_ni(struct sk_buff *skb)
1607 err = netif_rx(skb);
1608 if (local_softirq_pending())
1615 EXPORT_SYMBOL(netif_rx_ni);
1617 static inline struct net_device *skb_bond(struct sk_buff *skb)
1619 struct net_device *dev = skb->dev;
1622 if (skb_bond_should_drop(skb)) {
1626 skb->dev = dev->master;
1632 static void net_tx_action(struct softirq_action *h)
1634 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1636 if (sd->completion_queue) {
1637 struct sk_buff *clist;
1639 local_irq_disable();
1640 clist = sd->completion_queue;
1641 sd->completion_queue = NULL;
1645 struct sk_buff *skb = clist;
1646 clist = clist->next;
1648 BUG_TRAP(!atomic_read(&skb->users));
1653 if (sd->output_queue) {
1654 struct net_device *head;
1656 local_irq_disable();
1657 head = sd->output_queue;
1658 sd->output_queue = NULL;
1662 struct net_device *dev = head;
1663 head = head->next_sched;
1665 smp_mb__before_clear_bit();
1666 clear_bit(__LINK_STATE_SCHED, &dev->state);
1668 if (spin_trylock(&dev->queue_lock)) {
1670 spin_unlock(&dev->queue_lock);
1672 netif_schedule(dev);
1678 static __inline__ int deliver_skb(struct sk_buff *skb,
1679 struct packet_type *pt_prev,
1680 struct net_device *orig_dev)
1682 atomic_inc(&skb->users);
1683 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1686 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1687 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1689 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1690 unsigned char *addr);
1691 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1693 static __inline__ int handle_bridge(struct sk_buff **pskb,
1694 struct packet_type **pt_prev, int *ret,
1695 struct net_device *orig_dev)
1697 struct net_bridge_port *port;
1699 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1700 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1704 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1708 return br_handle_frame_hook(port, pskb);
1711 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1714 #ifdef CONFIG_NET_CLS_ACT
1715 /* TODO: Maybe we should just force sch_ingress to be compiled in
1716 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1717 * a compare and 2 stores extra right now if we dont have it on
1718 * but have CONFIG_NET_CLS_ACT
1719 * NOTE: This doesnt stop any functionality; if you dont have
1720 * the ingress scheduler, you just cant add policies on ingress.
1723 static int ing_filter(struct sk_buff *skb)
1726 struct net_device *dev = skb->dev;
1727 int result = TC_ACT_OK;
1729 if (dev->qdisc_ingress) {
1730 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1731 if (MAX_RED_LOOP < ttl++) {
1732 printk(KERN_WARNING "Redir loop detected Dropping packet (%s->%s)\n",
1733 skb->input_dev->name, skb->dev->name);
1737 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1739 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1741 spin_lock(&dev->ingress_lock);
1742 if ((q = dev->qdisc_ingress) != NULL)
1743 result = q->enqueue(skb, q);
1744 spin_unlock(&dev->ingress_lock);
1752 int netif_receive_skb(struct sk_buff *skb)
1754 struct packet_type *ptype, *pt_prev;
1755 struct net_device *orig_dev;
1756 int ret = NET_RX_DROP;
1757 unsigned short type;
1759 /* if we've gotten here through NAPI, check netpoll */
1760 if (skb->dev->poll && netpoll_rx(skb))
1763 if (!skb->tstamp.off_sec)
1766 if (!skb->input_dev)
1767 skb->input_dev = skb->dev;
1769 orig_dev = skb_bond(skb);
1774 __get_cpu_var(netdev_rx_stat).total++;
1776 skb->h.raw = skb->nh.raw = skb->data;
1777 skb->mac_len = skb->nh.raw - skb->mac.raw;
1783 #ifdef CONFIG_NET_CLS_ACT
1784 if (skb->tc_verd & TC_NCLS) {
1785 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1790 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1791 if (!ptype->dev || ptype->dev == skb->dev) {
1793 ret = deliver_skb(skb, pt_prev, orig_dev);
1798 #ifdef CONFIG_NET_CLS_ACT
1800 ret = deliver_skb(skb, pt_prev, orig_dev);
1801 pt_prev = NULL; /* noone else should process this after*/
1803 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1806 ret = ing_filter(skb);
1808 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1817 handle_diverter(skb);
1819 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1822 type = skb->protocol;
1823 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1824 if (ptype->type == type &&
1825 (!ptype->dev || ptype->dev == skb->dev)) {
1827 ret = deliver_skb(skb, pt_prev, orig_dev);
1833 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1836 /* Jamal, now you will not able to escape explaining
1837 * me how you were going to use this. :-)
1847 static int process_backlog(struct net_device *backlog_dev, int *budget)
1850 int quota = min(backlog_dev->quota, *budget);
1851 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1852 unsigned long start_time = jiffies;
1854 backlog_dev->weight = weight_p;
1856 struct sk_buff *skb;
1857 struct net_device *dev;
1859 local_irq_disable();
1860 skb = __skb_dequeue(&queue->input_pkt_queue);
1867 netif_receive_skb(skb);
1873 if (work >= quota || jiffies - start_time > 1)
1878 backlog_dev->quota -= work;
1883 backlog_dev->quota -= work;
1886 list_del(&backlog_dev->poll_list);
1887 smp_mb__before_clear_bit();
1888 netif_poll_enable(backlog_dev);
1894 static void net_rx_action(struct softirq_action *h)
1896 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1897 unsigned long start_time = jiffies;
1898 int budget = netdev_budget;
1901 local_irq_disable();
1903 while (!list_empty(&queue->poll_list)) {
1904 struct net_device *dev;
1906 if (budget <= 0 || jiffies - start_time > 1)
1911 dev = list_entry(queue->poll_list.next,
1912 struct net_device, poll_list);
1913 have = netpoll_poll_lock(dev);
1915 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1916 netpoll_poll_unlock(have);
1917 local_irq_disable();
1918 list_move_tail(&dev->poll_list, &queue->poll_list);
1920 dev->quota += dev->weight;
1922 dev->quota = dev->weight;
1924 netpoll_poll_unlock(have);
1926 local_irq_disable();
1930 #ifdef CONFIG_NET_DMA
1932 * There may not be any more sk_buffs coming right now, so push
1933 * any pending DMA copies to hardware
1935 if (net_dma_client) {
1936 struct dma_chan *chan;
1938 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1939 dma_async_memcpy_issue_pending(chan);
1947 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1948 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1952 static gifconf_func_t * gifconf_list [NPROTO];
1955 * register_gifconf - register a SIOCGIF handler
1956 * @family: Address family
1957 * @gifconf: Function handler
1959 * Register protocol dependent address dumping routines. The handler
1960 * that is passed must not be freed or reused until it has been replaced
1961 * by another handler.
1963 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1965 if (family >= NPROTO)
1967 gifconf_list[family] = gifconf;
1973 * Map an interface index to its name (SIOCGIFNAME)
1977 * We need this ioctl for efficient implementation of the
1978 * if_indextoname() function required by the IPv6 API. Without
1979 * it, we would have to search all the interfaces to find a
1983 static int dev_ifname(struct ifreq __user *arg)
1985 struct net_device *dev;
1989 * Fetch the caller's info block.
1992 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1995 read_lock(&dev_base_lock);
1996 dev = __dev_get_by_index(ifr.ifr_ifindex);
1998 read_unlock(&dev_base_lock);
2002 strcpy(ifr.ifr_name, dev->name);
2003 read_unlock(&dev_base_lock);
2005 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2011 * Perform a SIOCGIFCONF call. This structure will change
2012 * size eventually, and there is nothing I can do about it.
2013 * Thus we will need a 'compatibility mode'.
2016 static int dev_ifconf(char __user *arg)
2019 struct net_device *dev;
2026 * Fetch the caller's info block.
2029 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2036 * Loop over the interfaces, and write an info block for each.
2040 for (dev = dev_base; dev; dev = dev->next) {
2041 for (i = 0; i < NPROTO; i++) {
2042 if (gifconf_list[i]) {
2045 done = gifconf_list[i](dev, NULL, 0);
2047 done = gifconf_list[i](dev, pos + total,
2057 * All done. Write the updated control block back to the caller.
2059 ifc.ifc_len = total;
2062 * Both BSD and Solaris return 0 here, so we do too.
2064 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2067 #ifdef CONFIG_PROC_FS
2069 * This is invoked by the /proc filesystem handler to display a device
2072 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2074 struct net_device *dev;
2077 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2079 return i == pos ? dev : NULL;
2082 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2084 read_lock(&dev_base_lock);
2085 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2088 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2091 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2094 void dev_seq_stop(struct seq_file *seq, void *v)
2096 read_unlock(&dev_base_lock);
2099 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2101 if (dev->get_stats) {
2102 struct net_device_stats *stats = dev->get_stats(dev);
2104 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2105 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2106 dev->name, stats->rx_bytes, stats->rx_packets,
2108 stats->rx_dropped + stats->rx_missed_errors,
2109 stats->rx_fifo_errors,
2110 stats->rx_length_errors + stats->rx_over_errors +
2111 stats->rx_crc_errors + stats->rx_frame_errors,
2112 stats->rx_compressed, stats->multicast,
2113 stats->tx_bytes, stats->tx_packets,
2114 stats->tx_errors, stats->tx_dropped,
2115 stats->tx_fifo_errors, stats->collisions,
2116 stats->tx_carrier_errors +
2117 stats->tx_aborted_errors +
2118 stats->tx_window_errors +
2119 stats->tx_heartbeat_errors,
2120 stats->tx_compressed);
2122 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2126 * Called from the PROCfs module. This now uses the new arbitrary sized
2127 * /proc/net interface to create /proc/net/dev
2129 static int dev_seq_show(struct seq_file *seq, void *v)
2131 if (v == SEQ_START_TOKEN)
2132 seq_puts(seq, "Inter-| Receive "
2134 " face |bytes packets errs drop fifo frame "
2135 "compressed multicast|bytes packets errs "
2136 "drop fifo colls carrier compressed\n");
2138 dev_seq_printf_stats(seq, v);
2142 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2144 struct netif_rx_stats *rc = NULL;
2146 while (*pos < NR_CPUS)
2147 if (cpu_online(*pos)) {
2148 rc = &per_cpu(netdev_rx_stat, *pos);
2155 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2157 return softnet_get_online(pos);
2160 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2163 return softnet_get_online(pos);
2166 static void softnet_seq_stop(struct seq_file *seq, void *v)
2170 static int softnet_seq_show(struct seq_file *seq, void *v)
2172 struct netif_rx_stats *s = v;
2174 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2175 s->total, s->dropped, s->time_squeeze, 0,
2176 0, 0, 0, 0, /* was fastroute */
2181 static struct seq_operations dev_seq_ops = {
2182 .start = dev_seq_start,
2183 .next = dev_seq_next,
2184 .stop = dev_seq_stop,
2185 .show = dev_seq_show,
2188 static int dev_seq_open(struct inode *inode, struct file *file)
2190 return seq_open(file, &dev_seq_ops);
2193 static struct file_operations dev_seq_fops = {
2194 .owner = THIS_MODULE,
2195 .open = dev_seq_open,
2197 .llseek = seq_lseek,
2198 .release = seq_release,
2201 static struct seq_operations softnet_seq_ops = {
2202 .start = softnet_seq_start,
2203 .next = softnet_seq_next,
2204 .stop = softnet_seq_stop,
2205 .show = softnet_seq_show,
2208 static int softnet_seq_open(struct inode *inode, struct file *file)
2210 return seq_open(file, &softnet_seq_ops);
2213 static struct file_operations softnet_seq_fops = {
2214 .owner = THIS_MODULE,
2215 .open = softnet_seq_open,
2217 .llseek = seq_lseek,
2218 .release = seq_release,
2221 #ifdef CONFIG_WIRELESS_EXT
2222 extern int wireless_proc_init(void);
2224 #define wireless_proc_init() 0
2227 static int __init dev_proc_init(void)
2231 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2233 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2235 if (wireless_proc_init())
2241 proc_net_remove("softnet_stat");
2243 proc_net_remove("dev");
2247 #define dev_proc_init() 0
2248 #endif /* CONFIG_PROC_FS */
2252 * netdev_set_master - set up master/slave pair
2253 * @slave: slave device
2254 * @master: new master device
2256 * Changes the master device of the slave. Pass %NULL to break the
2257 * bonding. The caller must hold the RTNL semaphore. On a failure
2258 * a negative errno code is returned. On success the reference counts
2259 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2260 * function returns zero.
2262 int netdev_set_master(struct net_device *slave, struct net_device *master)
2264 struct net_device *old = slave->master;
2274 slave->master = master;
2282 slave->flags |= IFF_SLAVE;
2284 slave->flags &= ~IFF_SLAVE;
2286 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2291 * dev_set_promiscuity - update promiscuity count on a device
2295 * Add or remove promiscuity from a device. While the count in the device
2296 * remains above zero the interface remains promiscuous. Once it hits zero
2297 * the device reverts back to normal filtering operation. A negative inc
2298 * value is used to drop promiscuity on the device.
2300 void dev_set_promiscuity(struct net_device *dev, int inc)
2302 unsigned short old_flags = dev->flags;
2304 if ((dev->promiscuity += inc) == 0)
2305 dev->flags &= ~IFF_PROMISC;
2307 dev->flags |= IFF_PROMISC;
2308 if (dev->flags != old_flags) {
2310 printk(KERN_INFO "device %s %s promiscuous mode\n",
2311 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2313 audit_log(current->audit_context, GFP_ATOMIC,
2314 AUDIT_ANOM_PROMISCUOUS,
2315 "dev=%s prom=%d old_prom=%d auid=%u",
2316 dev->name, (dev->flags & IFF_PROMISC),
2317 (old_flags & IFF_PROMISC),
2318 audit_get_loginuid(current->audit_context));
2323 * dev_set_allmulti - update allmulti count on a device
2327 * Add or remove reception of all multicast frames to a device. While the
2328 * count in the device remains above zero the interface remains listening
2329 * to all interfaces. Once it hits zero the device reverts back to normal
2330 * filtering operation. A negative @inc value is used to drop the counter
2331 * when releasing a resource needing all multicasts.
2334 void dev_set_allmulti(struct net_device *dev, int inc)
2336 unsigned short old_flags = dev->flags;
2338 dev->flags |= IFF_ALLMULTI;
2339 if ((dev->allmulti += inc) == 0)
2340 dev->flags &= ~IFF_ALLMULTI;
2341 if (dev->flags ^ old_flags)
2345 unsigned dev_get_flags(const struct net_device *dev)
2349 flags = (dev->flags & ~(IFF_PROMISC |
2354 (dev->gflags & (IFF_PROMISC |
2357 if (netif_running(dev)) {
2358 if (netif_oper_up(dev))
2359 flags |= IFF_RUNNING;
2360 if (netif_carrier_ok(dev))
2361 flags |= IFF_LOWER_UP;
2362 if (netif_dormant(dev))
2363 flags |= IFF_DORMANT;
2369 int dev_change_flags(struct net_device *dev, unsigned flags)
2372 int old_flags = dev->flags;
2375 * Set the flags on our device.
2378 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2379 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2381 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2385 * Load in the correct multicast list now the flags have changed.
2391 * Have we downed the interface. We handle IFF_UP ourselves
2392 * according to user attempts to set it, rather than blindly
2397 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2398 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2404 if (dev->flags & IFF_UP &&
2405 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2407 raw_notifier_call_chain(&netdev_chain,
2408 NETDEV_CHANGE, dev);
2410 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2411 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2412 dev->gflags ^= IFF_PROMISC;
2413 dev_set_promiscuity(dev, inc);
2416 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2417 is important. Some (broken) drivers set IFF_PROMISC, when
2418 IFF_ALLMULTI is requested not asking us and not reporting.
2420 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2421 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2422 dev->gflags ^= IFF_ALLMULTI;
2423 dev_set_allmulti(dev, inc);
2426 if (old_flags ^ dev->flags)
2427 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2432 int dev_set_mtu(struct net_device *dev, int new_mtu)
2436 if (new_mtu == dev->mtu)
2439 /* MTU must be positive. */
2443 if (!netif_device_present(dev))
2447 if (dev->change_mtu)
2448 err = dev->change_mtu(dev, new_mtu);
2451 if (!err && dev->flags & IFF_UP)
2452 raw_notifier_call_chain(&netdev_chain,
2453 NETDEV_CHANGEMTU, dev);
2457 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2461 if (!dev->set_mac_address)
2463 if (sa->sa_family != dev->type)
2465 if (!netif_device_present(dev))
2467 err = dev->set_mac_address(dev, sa);
2469 raw_notifier_call_chain(&netdev_chain,
2470 NETDEV_CHANGEADDR, dev);
2475 * Perform the SIOCxIFxxx calls.
2477 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2480 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2486 case SIOCGIFFLAGS: /* Get interface flags */
2487 ifr->ifr_flags = dev_get_flags(dev);
2490 case SIOCSIFFLAGS: /* Set interface flags */
2491 return dev_change_flags(dev, ifr->ifr_flags);
2493 case SIOCGIFMETRIC: /* Get the metric on the interface
2494 (currently unused) */
2495 ifr->ifr_metric = 0;
2498 case SIOCSIFMETRIC: /* Set the metric on the interface
2499 (currently unused) */
2502 case SIOCGIFMTU: /* Get the MTU of a device */
2503 ifr->ifr_mtu = dev->mtu;
2506 case SIOCSIFMTU: /* Set the MTU of a device */
2507 return dev_set_mtu(dev, ifr->ifr_mtu);
2511 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2513 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2514 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2515 ifr->ifr_hwaddr.sa_family = dev->type;
2519 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2521 case SIOCSIFHWBROADCAST:
2522 if (ifr->ifr_hwaddr.sa_family != dev->type)
2524 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2525 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2526 raw_notifier_call_chain(&netdev_chain,
2527 NETDEV_CHANGEADDR, dev);
2531 ifr->ifr_map.mem_start = dev->mem_start;
2532 ifr->ifr_map.mem_end = dev->mem_end;
2533 ifr->ifr_map.base_addr = dev->base_addr;
2534 ifr->ifr_map.irq = dev->irq;
2535 ifr->ifr_map.dma = dev->dma;
2536 ifr->ifr_map.port = dev->if_port;
2540 if (dev->set_config) {
2541 if (!netif_device_present(dev))
2543 return dev->set_config(dev, &ifr->ifr_map);
2548 if (!dev->set_multicast_list ||
2549 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2551 if (!netif_device_present(dev))
2553 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2557 if (!dev->set_multicast_list ||
2558 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2560 if (!netif_device_present(dev))
2562 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2566 ifr->ifr_ifindex = dev->ifindex;
2570 ifr->ifr_qlen = dev->tx_queue_len;
2574 if (ifr->ifr_qlen < 0)
2576 dev->tx_queue_len = ifr->ifr_qlen;
2580 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2581 return dev_change_name(dev, ifr->ifr_newname);
2584 * Unknown or private ioctl
2588 if ((cmd >= SIOCDEVPRIVATE &&
2589 cmd <= SIOCDEVPRIVATE + 15) ||
2590 cmd == SIOCBONDENSLAVE ||
2591 cmd == SIOCBONDRELEASE ||
2592 cmd == SIOCBONDSETHWADDR ||
2593 cmd == SIOCBONDSLAVEINFOQUERY ||
2594 cmd == SIOCBONDINFOQUERY ||
2595 cmd == SIOCBONDCHANGEACTIVE ||
2596 cmd == SIOCGMIIPHY ||
2597 cmd == SIOCGMIIREG ||
2598 cmd == SIOCSMIIREG ||
2599 cmd == SIOCBRADDIF ||
2600 cmd == SIOCBRDELIF ||
2601 cmd == SIOCWANDEV) {
2603 if (dev->do_ioctl) {
2604 if (netif_device_present(dev))
2605 err = dev->do_ioctl(dev, ifr,
2618 * This function handles all "interface"-type I/O control requests. The actual
2619 * 'doing' part of this is dev_ifsioc above.
2623 * dev_ioctl - network device ioctl
2624 * @cmd: command to issue
2625 * @arg: pointer to a struct ifreq in user space
2627 * Issue ioctl functions to devices. This is normally called by the
2628 * user space syscall interfaces but can sometimes be useful for
2629 * other purposes. The return value is the return from the syscall if
2630 * positive or a negative errno code on error.
2633 int dev_ioctl(unsigned int cmd, void __user *arg)
2639 /* One special case: SIOCGIFCONF takes ifconf argument
2640 and requires shared lock, because it sleeps writing
2644 if (cmd == SIOCGIFCONF) {
2646 ret = dev_ifconf((char __user *) arg);
2650 if (cmd == SIOCGIFNAME)
2651 return dev_ifname((struct ifreq __user *)arg);
2653 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2656 ifr.ifr_name[IFNAMSIZ-1] = 0;
2658 colon = strchr(ifr.ifr_name, ':');
2663 * See which interface the caller is talking about.
2668 * These ioctl calls:
2669 * - can be done by all.
2670 * - atomic and do not require locking.
2681 dev_load(ifr.ifr_name);
2682 read_lock(&dev_base_lock);
2683 ret = dev_ifsioc(&ifr, cmd);
2684 read_unlock(&dev_base_lock);
2688 if (copy_to_user(arg, &ifr,
2689 sizeof(struct ifreq)))
2695 dev_load(ifr.ifr_name);
2697 ret = dev_ethtool(&ifr);
2702 if (copy_to_user(arg, &ifr,
2703 sizeof(struct ifreq)))
2709 * These ioctl calls:
2710 * - require superuser power.
2711 * - require strict serialization.
2717 if (!capable(CAP_NET_ADMIN))
2719 dev_load(ifr.ifr_name);
2721 ret = dev_ifsioc(&ifr, cmd);
2726 if (copy_to_user(arg, &ifr,
2727 sizeof(struct ifreq)))
2733 * These ioctl calls:
2734 * - require superuser power.
2735 * - require strict serialization.
2736 * - do not return a value
2746 case SIOCSIFHWBROADCAST:
2749 case SIOCBONDENSLAVE:
2750 case SIOCBONDRELEASE:
2751 case SIOCBONDSETHWADDR:
2752 case SIOCBONDCHANGEACTIVE:
2755 if (!capable(CAP_NET_ADMIN))
2758 case SIOCBONDSLAVEINFOQUERY:
2759 case SIOCBONDINFOQUERY:
2760 dev_load(ifr.ifr_name);
2762 ret = dev_ifsioc(&ifr, cmd);
2767 /* Get the per device memory space. We can add this but
2768 * currently do not support it */
2770 /* Set the per device memory buffer space.
2771 * Not applicable in our case */
2776 * Unknown or private ioctl.
2779 if (cmd == SIOCWANDEV ||
2780 (cmd >= SIOCDEVPRIVATE &&
2781 cmd <= SIOCDEVPRIVATE + 15)) {
2782 dev_load(ifr.ifr_name);
2784 ret = dev_ifsioc(&ifr, cmd);
2786 if (!ret && copy_to_user(arg, &ifr,
2787 sizeof(struct ifreq)))
2791 #ifdef CONFIG_WIRELESS_EXT
2792 /* Take care of Wireless Extensions */
2793 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2794 /* If command is `set a parameter', or
2795 * `get the encoding parameters', check if
2796 * the user has the right to do it */
2797 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2798 || cmd == SIOCGIWENCODEEXT) {
2799 if (!capable(CAP_NET_ADMIN))
2802 dev_load(ifr.ifr_name);
2804 /* Follow me in net/core/wireless.c */
2805 ret = wireless_process_ioctl(&ifr, cmd);
2807 if (IW_IS_GET(cmd) &&
2808 copy_to_user(arg, &ifr,
2809 sizeof(struct ifreq)))
2813 #endif /* CONFIG_WIRELESS_EXT */
2820 * dev_new_index - allocate an ifindex
2822 * Returns a suitable unique value for a new device interface
2823 * number. The caller must hold the rtnl semaphore or the
2824 * dev_base_lock to be sure it remains unique.
2826 static int dev_new_index(void)
2832 if (!__dev_get_by_index(ifindex))
2837 static int dev_boot_phase = 1;
2839 /* Delayed registration/unregisteration */
2840 static DEFINE_SPINLOCK(net_todo_list_lock);
2841 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2843 static inline void net_set_todo(struct net_device *dev)
2845 spin_lock(&net_todo_list_lock);
2846 list_add_tail(&dev->todo_list, &net_todo_list);
2847 spin_unlock(&net_todo_list_lock);
2851 * register_netdevice - register a network device
2852 * @dev: device to register
2854 * Take a completed network device structure and add it to the kernel
2855 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2856 * chain. 0 is returned on success. A negative errno code is returned
2857 * on a failure to set up the device, or if the name is a duplicate.
2859 * Callers must hold the rtnl semaphore. You may want
2860 * register_netdev() instead of this.
2863 * The locking appears insufficient to guarantee two parallel registers
2864 * will not get the same name.
2867 int register_netdevice(struct net_device *dev)
2869 struct hlist_head *head;
2870 struct hlist_node *p;
2873 BUG_ON(dev_boot_phase);
2878 /* When net_device's are persistent, this will be fatal. */
2879 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2881 spin_lock_init(&dev->queue_lock);
2882 spin_lock_init(&dev->_xmit_lock);
2883 dev->xmit_lock_owner = -1;
2884 #ifdef CONFIG_NET_CLS_ACT
2885 spin_lock_init(&dev->ingress_lock);
2888 ret = alloc_divert_blk(dev);
2894 /* Init, if this function is available */
2896 ret = dev->init(dev);
2904 if (!dev_valid_name(dev->name)) {
2909 dev->ifindex = dev_new_index();
2910 if (dev->iflink == -1)
2911 dev->iflink = dev->ifindex;
2913 /* Check for existence of name */
2914 head = dev_name_hash(dev->name);
2915 hlist_for_each(p, head) {
2916 struct net_device *d
2917 = hlist_entry(p, struct net_device, name_hlist);
2918 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2924 /* Fix illegal SG+CSUM combinations. */
2925 if ((dev->features & NETIF_F_SG) &&
2926 !(dev->features & NETIF_F_ALL_CSUM)) {
2927 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
2929 dev->features &= ~NETIF_F_SG;
2932 /* TSO requires that SG is present as well. */
2933 if ((dev->features & NETIF_F_TSO) &&
2934 !(dev->features & NETIF_F_SG)) {
2935 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
2937 dev->features &= ~NETIF_F_TSO;
2939 if (dev->features & NETIF_F_UFO) {
2940 if (!(dev->features & NETIF_F_HW_CSUM)) {
2941 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2942 "NETIF_F_HW_CSUM feature.\n",
2944 dev->features &= ~NETIF_F_UFO;
2946 if (!(dev->features & NETIF_F_SG)) {
2947 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2948 "NETIF_F_SG feature.\n",
2950 dev->features &= ~NETIF_F_UFO;
2955 * nil rebuild_header routine,
2956 * that should be never called and used as just bug trap.
2959 if (!dev->rebuild_header)
2960 dev->rebuild_header = default_rebuild_header;
2962 ret = netdev_register_sysfs(dev);
2965 dev->reg_state = NETREG_REGISTERED;
2968 * Default initial state at registry is that the
2969 * device is present.
2972 set_bit(__LINK_STATE_PRESENT, &dev->state);
2975 dev_init_scheduler(dev);
2976 write_lock_bh(&dev_base_lock);
2978 dev_tail = &dev->next;
2979 hlist_add_head(&dev->name_hlist, head);
2980 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2982 write_unlock_bh(&dev_base_lock);
2984 /* Notify protocols, that a new device appeared. */
2985 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2992 free_divert_blk(dev);
2997 * register_netdev - register a network device
2998 * @dev: device to register
3000 * Take a completed network device structure and add it to the kernel
3001 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3002 * chain. 0 is returned on success. A negative errno code is returned
3003 * on a failure to set up the device, or if the name is a duplicate.
3005 * This is a wrapper around register_netdev that takes the rtnl semaphore
3006 * and expands the device name if you passed a format string to
3009 int register_netdev(struct net_device *dev)
3016 * If the name is a format string the caller wants us to do a
3019 if (strchr(dev->name, '%')) {
3020 err = dev_alloc_name(dev, dev->name);
3026 * Back compatibility hook. Kill this one in 2.5
3028 if (dev->name[0] == 0 || dev->name[0] == ' ') {
3029 err = dev_alloc_name(dev, "eth%d");
3034 err = register_netdevice(dev);
3039 EXPORT_SYMBOL(register_netdev);
3042 * netdev_wait_allrefs - wait until all references are gone.
3044 * This is called when unregistering network devices.
3046 * Any protocol or device that holds a reference should register
3047 * for netdevice notification, and cleanup and put back the
3048 * reference if they receive an UNREGISTER event.
3049 * We can get stuck here if buggy protocols don't correctly
3052 static void netdev_wait_allrefs(struct net_device *dev)
3054 unsigned long rebroadcast_time, warning_time;
3056 rebroadcast_time = warning_time = jiffies;
3057 while (atomic_read(&dev->refcnt) != 0) {
3058 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3061 /* Rebroadcast unregister notification */
3062 raw_notifier_call_chain(&netdev_chain,
3063 NETDEV_UNREGISTER, dev);
3065 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3067 /* We must not have linkwatch events
3068 * pending on unregister. If this
3069 * happens, we simply run the queue
3070 * unscheduled, resulting in a noop
3073 linkwatch_run_queue();
3078 rebroadcast_time = jiffies;
3083 if (time_after(jiffies, warning_time + 10 * HZ)) {
3084 printk(KERN_EMERG "unregister_netdevice: "
3085 "waiting for %s to become free. Usage "
3087 dev->name, atomic_read(&dev->refcnt));
3088 warning_time = jiffies;
3097 * register_netdevice(x1);
3098 * register_netdevice(x2);
3100 * unregister_netdevice(y1);
3101 * unregister_netdevice(y2);
3107 * We are invoked by rtnl_unlock() after it drops the semaphore.
3108 * This allows us to deal with problems:
3109 * 1) We can delete sysfs objects which invoke hotplug
3110 * without deadlocking with linkwatch via keventd.
3111 * 2) Since we run with the RTNL semaphore not held, we can sleep
3112 * safely in order to wait for the netdev refcnt to drop to zero.
3114 static DEFINE_MUTEX(net_todo_run_mutex);
3115 void netdev_run_todo(void)
3117 struct list_head list;
3119 /* Need to guard against multiple cpu's getting out of order. */
3120 mutex_lock(&net_todo_run_mutex);
3122 /* Not safe to do outside the semaphore. We must not return
3123 * until all unregister events invoked by the local processor
3124 * have been completed (either by this todo run, or one on
3127 if (list_empty(&net_todo_list))
3130 /* Snapshot list, allow later requests */
3131 spin_lock(&net_todo_list_lock);
3132 list_replace_init(&net_todo_list, &list);
3133 spin_unlock(&net_todo_list_lock);
3135 while (!list_empty(&list)) {
3136 struct net_device *dev
3137 = list_entry(list.next, struct net_device, todo_list);
3138 list_del(&dev->todo_list);
3140 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3141 printk(KERN_ERR "network todo '%s' but state %d\n",
3142 dev->name, dev->reg_state);
3147 netdev_unregister_sysfs(dev);
3148 dev->reg_state = NETREG_UNREGISTERED;
3150 netdev_wait_allrefs(dev);
3153 BUG_ON(atomic_read(&dev->refcnt));
3154 BUG_TRAP(!dev->ip_ptr);
3155 BUG_TRAP(!dev->ip6_ptr);
3156 BUG_TRAP(!dev->dn_ptr);
3158 /* It must be the very last action,
3159 * after this 'dev' may point to freed up memory.
3161 if (dev->destructor)
3162 dev->destructor(dev);
3166 mutex_unlock(&net_todo_run_mutex);
3170 * alloc_netdev - allocate network device
3171 * @sizeof_priv: size of private data to allocate space for
3172 * @name: device name format string
3173 * @setup: callback to initialize device
3175 * Allocates a struct net_device with private data area for driver use
3176 * and performs basic initialization.
3178 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3179 void (*setup)(struct net_device *))
3182 struct net_device *dev;
3185 /* ensure 32-byte alignment of both the device and private area */
3186 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3187 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3189 p = kzalloc(alloc_size, GFP_KERNEL);
3191 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3195 dev = (struct net_device *)
3196 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3197 dev->padded = (char *)dev - (char *)p;
3200 dev->priv = netdev_priv(dev);
3203 strcpy(dev->name, name);
3206 EXPORT_SYMBOL(alloc_netdev);
3209 * free_netdev - free network device
3212 * This function does the last stage of destroying an allocated device
3213 * interface. The reference to the device object is released.
3214 * If this is the last reference then it will be freed.
3216 void free_netdev(struct net_device *dev)
3219 /* Compatibility with error handling in drivers */
3220 if (dev->reg_state == NETREG_UNINITIALIZED) {
3221 kfree((char *)dev - dev->padded);
3225 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3226 dev->reg_state = NETREG_RELEASED;
3228 /* will free via class release */
3229 class_device_put(&dev->class_dev);
3231 kfree((char *)dev - dev->padded);
3235 /* Synchronize with packet receive processing. */
3236 void synchronize_net(void)
3243 * unregister_netdevice - remove device from the kernel
3246 * This function shuts down a device interface and removes it
3247 * from the kernel tables. On success 0 is returned, on a failure
3248 * a negative errno code is returned.
3250 * Callers must hold the rtnl semaphore. You may want
3251 * unregister_netdev() instead of this.
3254 int unregister_netdevice(struct net_device *dev)
3256 struct net_device *d, **dp;
3258 BUG_ON(dev_boot_phase);
3261 /* Some devices call without registering for initialization unwind. */
3262 if (dev->reg_state == NETREG_UNINITIALIZED) {
3263 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3264 "was registered\n", dev->name, dev);
3268 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3270 /* If device is running, close it first. */
3271 if (dev->flags & IFF_UP)
3274 /* And unlink it from device chain. */
3275 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3277 write_lock_bh(&dev_base_lock);
3278 hlist_del(&dev->name_hlist);
3279 hlist_del(&dev->index_hlist);
3280 if (dev_tail == &dev->next)
3283 write_unlock_bh(&dev_base_lock);
3288 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3293 dev->reg_state = NETREG_UNREGISTERING;
3297 /* Shutdown queueing discipline. */
3301 /* Notify protocols, that we are about to destroy
3302 this device. They should clean all the things.
3304 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3307 * Flush the multicast chain
3309 dev_mc_discard(dev);
3314 /* Notifier chain MUST detach us from master device. */
3315 BUG_TRAP(!dev->master);
3317 free_divert_blk(dev);
3319 /* Finish processing unregister after unlock */
3329 * unregister_netdev - remove device from the kernel
3332 * This function shuts down a device interface and removes it
3333 * from the kernel tables. On success 0 is returned, on a failure
3334 * a negative errno code is returned.
3336 * This is just a wrapper for unregister_netdevice that takes
3337 * the rtnl semaphore. In general you want to use this and not
3338 * unregister_netdevice.
3340 void unregister_netdev(struct net_device *dev)
3343 unregister_netdevice(dev);
3347 EXPORT_SYMBOL(unregister_netdev);
3349 #ifdef CONFIG_HOTPLUG_CPU
3350 static int dev_cpu_callback(struct notifier_block *nfb,
3351 unsigned long action,
3354 struct sk_buff **list_skb;
3355 struct net_device **list_net;
3356 struct sk_buff *skb;
3357 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3358 struct softnet_data *sd, *oldsd;
3360 if (action != CPU_DEAD)
3363 local_irq_disable();
3364 cpu = smp_processor_id();
3365 sd = &per_cpu(softnet_data, cpu);
3366 oldsd = &per_cpu(softnet_data, oldcpu);
3368 /* Find end of our completion_queue. */
3369 list_skb = &sd->completion_queue;
3371 list_skb = &(*list_skb)->next;
3372 /* Append completion queue from offline CPU. */
3373 *list_skb = oldsd->completion_queue;
3374 oldsd->completion_queue = NULL;
3376 /* Find end of our output_queue. */
3377 list_net = &sd->output_queue;
3379 list_net = &(*list_net)->next_sched;
3380 /* Append output queue from offline CPU. */
3381 *list_net = oldsd->output_queue;
3382 oldsd->output_queue = NULL;
3384 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3387 /* Process offline CPU's input_pkt_queue */
3388 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3393 #endif /* CONFIG_HOTPLUG_CPU */
3395 #ifdef CONFIG_NET_DMA
3397 * net_dma_rebalance -
3398 * This is called when the number of channels allocated to the net_dma_client
3399 * changes. The net_dma_client tries to have one DMA channel per CPU.
3401 static void net_dma_rebalance(void)
3403 unsigned int cpu, i, n;
3404 struct dma_chan *chan;
3406 if (net_dma_count == 0) {
3407 for_each_online_cpu(cpu)
3408 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
3413 cpu = first_cpu(cpu_online_map);
3416 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3417 n = ((num_online_cpus() / net_dma_count)
3418 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3421 per_cpu(softnet_data, cpu).net_dma = chan;
3422 cpu = next_cpu(cpu, cpu_online_map);
3431 * netdev_dma_event - event callback for the net_dma_client
3432 * @client: should always be net_dma_client
3433 * @chan: DMA channel for the event
3434 * @event: event type
3436 static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3437 enum dma_event event)
3439 spin_lock(&net_dma_event_lock);
3441 case DMA_RESOURCE_ADDED:
3443 net_dma_rebalance();
3445 case DMA_RESOURCE_REMOVED:
3447 net_dma_rebalance();
3452 spin_unlock(&net_dma_event_lock);
3456 * netdev_dma_regiser - register the networking subsystem as a DMA client
3458 static int __init netdev_dma_register(void)
3460 spin_lock_init(&net_dma_event_lock);
3461 net_dma_client = dma_async_client_register(netdev_dma_event);
3462 if (net_dma_client == NULL)
3465 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3470 static int __init netdev_dma_register(void) { return -ENODEV; }
3471 #endif /* CONFIG_NET_DMA */
3474 * Initialize the DEV module. At boot time this walks the device list and
3475 * unhooks any devices that fail to initialise (normally hardware not
3476 * present) and leaves us with a valid list of present and active devices.
3481 * This is called single threaded during boot, so no need
3482 * to take the rtnl semaphore.
3484 static int __init net_dev_init(void)
3486 int i, rc = -ENOMEM;
3488 BUG_ON(!dev_boot_phase);
3492 if (dev_proc_init())
3495 if (netdev_sysfs_init())
3498 INIT_LIST_HEAD(&ptype_all);
3499 for (i = 0; i < 16; i++)
3500 INIT_LIST_HEAD(&ptype_base[i]);
3502 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3503 INIT_HLIST_HEAD(&dev_name_head[i]);
3505 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3506 INIT_HLIST_HEAD(&dev_index_head[i]);
3509 * Initialise the packet receive queues.
3512 for_each_possible_cpu(i) {
3513 struct softnet_data *queue;
3515 queue = &per_cpu(softnet_data, i);
3516 skb_queue_head_init(&queue->input_pkt_queue);
3517 queue->completion_queue = NULL;
3518 INIT_LIST_HEAD(&queue->poll_list);
3519 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3520 queue->backlog_dev.weight = weight_p;
3521 queue->backlog_dev.poll = process_backlog;
3522 atomic_set(&queue->backlog_dev.refcnt, 1);
3525 netdev_dma_register();
3529 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3530 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3532 hotcpu_notifier(dev_cpu_callback, 0);
3540 subsys_initcall(net_dev_init);
3542 EXPORT_SYMBOL(__dev_get_by_index);
3543 EXPORT_SYMBOL(__dev_get_by_name);
3544 EXPORT_SYMBOL(__dev_remove_pack);
3545 EXPORT_SYMBOL(dev_valid_name);
3546 EXPORT_SYMBOL(dev_add_pack);
3547 EXPORT_SYMBOL(dev_alloc_name);
3548 EXPORT_SYMBOL(dev_close);
3549 EXPORT_SYMBOL(dev_get_by_flags);
3550 EXPORT_SYMBOL(dev_get_by_index);
3551 EXPORT_SYMBOL(dev_get_by_name);
3552 EXPORT_SYMBOL(dev_open);
3553 EXPORT_SYMBOL(dev_queue_xmit);
3554 EXPORT_SYMBOL(dev_remove_pack);
3555 EXPORT_SYMBOL(dev_set_allmulti);
3556 EXPORT_SYMBOL(dev_set_promiscuity);
3557 EXPORT_SYMBOL(dev_change_flags);
3558 EXPORT_SYMBOL(dev_set_mtu);
3559 EXPORT_SYMBOL(dev_set_mac_address);
3560 EXPORT_SYMBOL(free_netdev);
3561 EXPORT_SYMBOL(netdev_boot_setup_check);
3562 EXPORT_SYMBOL(netdev_set_master);
3563 EXPORT_SYMBOL(netdev_state_change);
3564 EXPORT_SYMBOL(netif_receive_skb);
3565 EXPORT_SYMBOL(netif_rx);
3566 EXPORT_SYMBOL(register_gifconf);
3567 EXPORT_SYMBOL(register_netdevice);
3568 EXPORT_SYMBOL(register_netdevice_notifier);
3569 EXPORT_SYMBOL(skb_checksum_help);
3570 EXPORT_SYMBOL(synchronize_net);
3571 EXPORT_SYMBOL(unregister_netdevice);
3572 EXPORT_SYMBOL(unregister_netdevice_notifier);
3573 EXPORT_SYMBOL(net_enable_timestamp);
3574 EXPORT_SYMBOL(net_disable_timestamp);
3575 EXPORT_SYMBOL(dev_get_flags);
3577 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3578 EXPORT_SYMBOL(br_handle_frame_hook);
3579 EXPORT_SYMBOL(br_fdb_get_hook);
3580 EXPORT_SYMBOL(br_fdb_put_hook);
3584 EXPORT_SYMBOL(dev_load);
3587 EXPORT_PER_CPU_SYMBOL(softnet_data);