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>
119 #include <linux/ctype.h>
120 #include <linux/vs_network.h>
124 #include <linux/tcp.h>
125 #include <linux/udp.h>
129 * The list of packet types we will receive (as opposed to discard)
130 * and the routines to invoke.
132 * Why 16. Because with 16 the only overlap we get on a hash of the
133 * low nibble of the protocol value is RARP/SNAP/X.25.
135 * NOTE: That is no longer true with the addition of VLAN tags. Not
136 * sure which should go first, but I bet it won't make much
137 * difference if we are running VLANs. The good news is that
138 * this protocol won't be in the list unless compiled in, so
139 * the average user (w/out VLANs) will not be adversely affected.
156 static DEFINE_SPINLOCK(ptype_lock);
157 static struct list_head ptype_base[16]; /* 16 way hashed list */
158 static struct list_head ptype_all; /* Taps */
160 #ifdef CONFIG_NET_DMA
161 static struct dma_client *net_dma_client;
162 static unsigned int net_dma_count;
163 static spinlock_t net_dma_event_lock;
167 * The @dev_base list is protected by @dev_base_lock and the rtnl
170 * Pure readers hold dev_base_lock for reading.
172 * Writers must hold the rtnl semaphore while they loop through the
173 * dev_base list, and hold dev_base_lock for writing when they do the
174 * actual updates. This allows pure readers to access the list even
175 * while a writer is preparing to update it.
177 * To put it another way, dev_base_lock is held for writing only to
178 * protect against pure readers; the rtnl semaphore provides the
179 * protection against other writers.
181 * See, for example usages, register_netdevice() and
182 * unregister_netdevice(), which must be called with the rtnl
185 struct net_device *dev_base;
186 static struct net_device **dev_tail = &dev_base;
187 DEFINE_RWLOCK(dev_base_lock);
189 EXPORT_SYMBOL(dev_base);
190 EXPORT_SYMBOL(dev_base_lock);
192 #define NETDEV_HASHBITS 8
193 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
194 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
196 static inline struct hlist_head *dev_name_hash(const char *name)
198 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
199 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
202 static inline struct hlist_head *dev_index_hash(int ifindex)
204 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
211 static RAW_NOTIFIER_HEAD(netdev_chain);
214 * Device drivers call our routines to queue packets here. We empty the
215 * queue in the local softnet handler.
217 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
220 extern int netdev_sysfs_init(void);
221 extern int netdev_register_sysfs(struct net_device *);
222 extern void netdev_unregister_sysfs(struct net_device *);
224 #define netdev_sysfs_init() (0)
225 #define netdev_register_sysfs(dev) (0)
226 #define netdev_unregister_sysfs(dev) do { } while(0)
230 /*******************************************************************************
232 Protocol management and registration routines
234 *******************************************************************************/
240 static int netdev_nit;
243 * Add a protocol ID to the list. Now that the input handler is
244 * smarter we can dispense with all the messy stuff that used to be
247 * BEWARE!!! Protocol handlers, mangling input packets,
248 * MUST BE last in hash buckets and checking protocol handlers
249 * MUST start from promiscuous ptype_all chain in net_bh.
250 * It is true now, do not change it.
251 * Explanation follows: if protocol handler, mangling packet, will
252 * be the first on list, it is not able to sense, that packet
253 * is cloned and should be copied-on-write, so that it will
254 * change it and subsequent readers will get broken packet.
259 * dev_add_pack - add packet handler
260 * @pt: packet type declaration
262 * Add a protocol handler to the networking stack. The passed &packet_type
263 * is linked into kernel lists and may not be freed until it has been
264 * removed from the kernel lists.
266 * This call does not sleep therefore it can not
267 * guarantee all CPU's that are in middle of receiving packets
268 * will see the new packet type (until the next received packet).
271 void dev_add_pack(struct packet_type *pt)
275 spin_lock_bh(&ptype_lock);
276 if (pt->type == htons(ETH_P_ALL)) {
278 list_add_rcu(&pt->list, &ptype_all);
280 hash = ntohs(pt->type) & 15;
281 list_add_rcu(&pt->list, &ptype_base[hash]);
283 spin_unlock_bh(&ptype_lock);
287 * __dev_remove_pack - remove packet handler
288 * @pt: packet type declaration
290 * Remove a protocol handler that was previously added to the kernel
291 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
292 * from the kernel lists and can be freed or reused once this function
295 * The packet type might still be in use by receivers
296 * and must not be freed until after all the CPU's have gone
297 * through a quiescent state.
299 void __dev_remove_pack(struct packet_type *pt)
301 struct list_head *head;
302 struct packet_type *pt1;
304 spin_lock_bh(&ptype_lock);
306 if (pt->type == htons(ETH_P_ALL)) {
310 head = &ptype_base[ntohs(pt->type) & 15];
312 list_for_each_entry(pt1, head, list) {
314 list_del_rcu(&pt->list);
319 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
321 spin_unlock_bh(&ptype_lock);
324 * dev_remove_pack - remove packet handler
325 * @pt: packet type declaration
327 * Remove a protocol handler that was previously added to the kernel
328 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
329 * from the kernel lists and can be freed or reused once this function
332 * This call sleeps to guarantee that no CPU is looking at the packet
335 void dev_remove_pack(struct packet_type *pt)
337 __dev_remove_pack(pt);
342 /******************************************************************************
344 Device Boot-time Settings Routines
346 *******************************************************************************/
348 /* Boot time configuration table */
349 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
352 * netdev_boot_setup_add - add new setup entry
353 * @name: name of the device
354 * @map: configured settings for the device
356 * Adds new setup entry to the dev_boot_setup list. The function
357 * returns 0 on error and 1 on success. This is a generic routine to
360 static int netdev_boot_setup_add(char *name, struct ifmap *map)
362 struct netdev_boot_setup *s;
366 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
367 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
368 memset(s[i].name, 0, sizeof(s[i].name));
369 strcpy(s[i].name, name);
370 memcpy(&s[i].map, map, sizeof(s[i].map));
375 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
379 * netdev_boot_setup_check - check boot time settings
380 * @dev: the netdevice
382 * Check boot time settings for the device.
383 * The found settings are set for the device to be used
384 * later in the device probing.
385 * Returns 0 if no settings found, 1 if they are.
387 int netdev_boot_setup_check(struct net_device *dev)
389 struct netdev_boot_setup *s = dev_boot_setup;
392 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
393 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
394 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
395 dev->irq = s[i].map.irq;
396 dev->base_addr = s[i].map.base_addr;
397 dev->mem_start = s[i].map.mem_start;
398 dev->mem_end = s[i].map.mem_end;
407 * netdev_boot_base - get address from boot time settings
408 * @prefix: prefix for network device
409 * @unit: id for network device
411 * Check boot time settings for the base address of device.
412 * The found settings are set for the device to be used
413 * later in the device probing.
414 * Returns 0 if no settings found.
416 unsigned long netdev_boot_base(const char *prefix, int unit)
418 const struct netdev_boot_setup *s = dev_boot_setup;
422 sprintf(name, "%s%d", prefix, unit);
425 * If device already registered then return base of 1
426 * to indicate not to probe for this interface
428 if (__dev_get_by_name(name))
431 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
432 if (!strcmp(name, s[i].name))
433 return s[i].map.base_addr;
438 * Saves at boot time configured settings for any netdevice.
440 int __init netdev_boot_setup(char *str)
445 str = get_options(str, ARRAY_SIZE(ints), ints);
450 memset(&map, 0, sizeof(map));
454 map.base_addr = ints[2];
456 map.mem_start = ints[3];
458 map.mem_end = ints[4];
460 /* Add new entry to the list */
461 return netdev_boot_setup_add(str, &map);
464 __setup("netdev=", netdev_boot_setup);
466 /*******************************************************************************
468 Device Interface Subroutines
470 *******************************************************************************/
473 * __dev_get_by_name - find a device by its name
474 * @name: name to find
476 * Find an interface by name. Must be called under RTNL semaphore
477 * or @dev_base_lock. If the name is found a pointer to the device
478 * is returned. If the name is not found then %NULL is returned. The
479 * reference counters are not incremented so the caller must be
480 * careful with locks.
483 struct net_device *__dev_get_by_name(const char *name)
485 struct hlist_node *p;
487 hlist_for_each(p, dev_name_hash(name)) {
488 struct net_device *dev
489 = hlist_entry(p, struct net_device, name_hlist);
490 if (!strncmp(dev->name, name, IFNAMSIZ))
497 * dev_get_by_name - find a device by its name
498 * @name: name to find
500 * Find an interface by name. This can be called from any
501 * context and does its own locking. The returned handle has
502 * the usage count incremented and the caller must use dev_put() to
503 * release it when it is no longer needed. %NULL is returned if no
504 * matching device is found.
507 struct net_device *dev_get_by_name(const char *name)
509 struct net_device *dev;
511 read_lock(&dev_base_lock);
512 dev = __dev_get_by_name(name);
515 read_unlock(&dev_base_lock);
520 * __dev_get_by_index - find a device by its ifindex
521 * @ifindex: index of device
523 * Search for an interface by index. Returns %NULL if the device
524 * is not found or a pointer to the device. The device has not
525 * had its reference counter increased so the caller must be careful
526 * about locking. The caller must hold either the RTNL semaphore
530 struct net_device *__dev_get_by_index(int ifindex)
532 struct hlist_node *p;
534 hlist_for_each(p, dev_index_hash(ifindex)) {
535 struct net_device *dev
536 = hlist_entry(p, struct net_device, index_hlist);
537 if (dev->ifindex == ifindex)
545 * dev_get_by_index - find a device by its ifindex
546 * @ifindex: index of device
548 * Search for an interface by index. Returns NULL if the device
549 * is not found or a pointer to the device. The device returned has
550 * had a reference added and the pointer is safe until the user calls
551 * dev_put to indicate they have finished with it.
554 struct net_device *dev_get_by_index(int ifindex)
556 struct net_device *dev;
558 read_lock(&dev_base_lock);
559 dev = __dev_get_by_index(ifindex);
562 read_unlock(&dev_base_lock);
567 * dev_getbyhwaddr - find a device by its hardware address
568 * @type: media type of device
569 * @ha: hardware address
571 * Search for an interface by MAC address. Returns NULL if the device
572 * is not found or a pointer to the device. The caller must hold the
573 * rtnl semaphore. The returned device has not had its ref count increased
574 * and the caller must therefore be careful about locking
577 * If the API was consistent this would be __dev_get_by_hwaddr
580 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
582 struct net_device *dev;
586 for (dev = dev_base; dev; dev = dev->next)
587 if (dev->type == type &&
588 !memcmp(dev->dev_addr, ha, dev->addr_len))
593 EXPORT_SYMBOL(dev_getbyhwaddr);
595 struct net_device *dev_getfirstbyhwtype(unsigned short type)
597 struct net_device *dev;
600 for (dev = dev_base; dev; dev = dev->next) {
601 if (dev->type == type) {
610 EXPORT_SYMBOL(dev_getfirstbyhwtype);
613 * dev_get_by_flags - find any device with given flags
614 * @if_flags: IFF_* values
615 * @mask: bitmask of bits in if_flags to check
617 * Search for any interface with the given flags. Returns NULL if a device
618 * is not found or a pointer to the device. The device returned has
619 * had a reference added and the pointer is safe until the user calls
620 * dev_put to indicate they have finished with it.
623 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
625 struct net_device *dev;
627 read_lock(&dev_base_lock);
628 for (dev = dev_base; dev != NULL; dev = dev->next) {
629 if (((dev->flags ^ if_flags) & mask) == 0) {
634 read_unlock(&dev_base_lock);
639 * dev_valid_name - check if name is okay for network device
642 * Network device names need to be valid file names to
643 * to allow sysfs to work. We also disallow any kind of
646 int dev_valid_name(const char *name)
650 if (!strcmp(name, ".") || !strcmp(name, ".."))
654 if (*name == '/' || isspace(*name))
662 * dev_alloc_name - allocate a name for a device
664 * @name: name format string
666 * Passed a format string - eg "lt%d" it will try and find a suitable
667 * id. It scans list of devices to build up a free map, then chooses
668 * the first empty slot. The caller must hold the dev_base or rtnl lock
669 * while allocating the name and adding the device in order to avoid
671 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
672 * Returns the number of the unit assigned or a negative errno code.
675 int dev_alloc_name(struct net_device *dev, const char *name)
680 const int max_netdevices = 8*PAGE_SIZE;
682 struct net_device *d;
684 p = strnchr(name, IFNAMSIZ-1, '%');
687 * Verify the string as this thing may have come from
688 * the user. There must be either one "%d" and no other "%"
691 if (p[1] != 'd' || strchr(p + 2, '%'))
694 /* Use one page as a bit array of possible slots */
695 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
699 for (d = dev_base; d; d = d->next) {
700 if (!sscanf(d->name, name, &i))
702 if (i < 0 || i >= max_netdevices)
705 /* avoid cases where sscanf is not exact inverse of printf */
706 snprintf(buf, sizeof(buf), name, i);
707 if (!strncmp(buf, d->name, IFNAMSIZ))
711 i = find_first_zero_bit(inuse, max_netdevices);
712 free_page((unsigned long) inuse);
715 snprintf(buf, sizeof(buf), name, i);
716 if (!__dev_get_by_name(buf)) {
717 strlcpy(dev->name, buf, IFNAMSIZ);
721 /* It is possible to run out of possible slots
722 * when the name is long and there isn't enough space left
723 * for the digits, or if all bits are used.
730 * dev_change_name - change name of a device
732 * @newname: name (or format string) must be at least IFNAMSIZ
734 * Change name of a device, can pass format strings "eth%d".
737 int dev_change_name(struct net_device *dev, char *newname)
743 if (dev->flags & IFF_UP)
746 if (!dev_valid_name(newname))
749 if (strchr(newname, '%')) {
750 err = dev_alloc_name(dev, newname);
753 strcpy(newname, dev->name);
755 else if (__dev_get_by_name(newname))
758 strlcpy(dev->name, newname, IFNAMSIZ);
760 err = class_device_rename(&dev->class_dev, dev->name);
762 hlist_del(&dev->name_hlist);
763 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
764 raw_notifier_call_chain(&netdev_chain,
765 NETDEV_CHANGENAME, dev);
772 * netdev_features_change - device changes features
773 * @dev: device to cause notification
775 * Called to indicate a device has changed features.
777 void netdev_features_change(struct net_device *dev)
779 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
781 EXPORT_SYMBOL(netdev_features_change);
784 * netdev_state_change - device changes state
785 * @dev: device to cause notification
787 * Called to indicate a device has changed state. This function calls
788 * the notifier chains for netdev_chain and sends a NEWLINK message
789 * to the routing socket.
791 void netdev_state_change(struct net_device *dev)
793 if (dev->flags & IFF_UP) {
794 raw_notifier_call_chain(&netdev_chain,
796 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
801 * dev_load - load a network module
802 * @name: name of interface
804 * If a network interface is not present and the process has suitable
805 * privileges this function loads the module. If module loading is not
806 * available in this kernel then it becomes a nop.
809 void dev_load(const char *name)
811 struct net_device *dev;
813 read_lock(&dev_base_lock);
814 dev = __dev_get_by_name(name);
815 read_unlock(&dev_base_lock);
817 if (!dev && capable(CAP_SYS_MODULE))
818 request_module("%s", name);
821 static int default_rebuild_header(struct sk_buff *skb)
823 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
824 skb->dev ? skb->dev->name : "NULL!!!");
831 * dev_open - prepare an interface for use.
832 * @dev: device to open
834 * Takes a device from down to up state. The device's private open
835 * function is invoked and then the multicast lists are loaded. Finally
836 * the device is moved into the up state and a %NETDEV_UP message is
837 * sent to the netdev notifier chain.
839 * Calling this function on an active interface is a nop. On a failure
840 * a negative errno code is returned.
842 int dev_open(struct net_device *dev)
850 if (dev->flags & IFF_UP)
854 * Is it even present?
856 if (!netif_device_present(dev))
860 * Call device private open method
862 set_bit(__LINK_STATE_START, &dev->state);
864 ret = dev->open(dev);
866 clear_bit(__LINK_STATE_START, &dev->state);
870 * If it went open OK then:
877 dev->flags |= IFF_UP;
880 * Initialize multicasting status
885 * Wakeup transmit queue engine
890 * ... and announce new interface.
892 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
898 * dev_close - shutdown an interface.
899 * @dev: device to shutdown
901 * This function moves an active device into down state. A
902 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
903 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
906 int dev_close(struct net_device *dev)
908 if (!(dev->flags & IFF_UP))
912 * Tell people we are going down, so that they can
913 * prepare to death, when device is still operating.
915 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
919 clear_bit(__LINK_STATE_START, &dev->state);
921 /* Synchronize to scheduled poll. We cannot touch poll list,
922 * it can be even on different cpu. So just clear netif_running(),
923 * and wait when poll really will happen. Actually, the best place
924 * for this is inside dev->stop() after device stopped its irq
925 * engine, but this requires more changes in devices. */
927 smp_mb__after_clear_bit(); /* Commit netif_running(). */
928 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
934 * Call the device specific close. This cannot fail.
935 * Only if device is UP
937 * We allow it to be called even after a DETACH hot-plug
944 * Device is now down.
947 dev->flags &= ~IFF_UP;
950 * Tell people we are down
952 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
959 * Device change register/unregister. These are not inline or static
960 * as we export them to the world.
964 * register_netdevice_notifier - register a network notifier block
967 * Register a notifier to be called when network device events occur.
968 * The notifier passed is linked into the kernel structures and must
969 * not be reused until it has been unregistered. A negative errno code
970 * is returned on a failure.
972 * When registered all registration and up events are replayed
973 * to the new notifier to allow device to have a race free
974 * view of the network device list.
977 int register_netdevice_notifier(struct notifier_block *nb)
979 struct net_device *dev;
983 err = raw_notifier_chain_register(&netdev_chain, nb);
985 for (dev = dev_base; dev; dev = dev->next) {
986 nb->notifier_call(nb, NETDEV_REGISTER, dev);
988 if (dev->flags & IFF_UP)
989 nb->notifier_call(nb, NETDEV_UP, dev);
997 * unregister_netdevice_notifier - unregister a network notifier block
1000 * Unregister a notifier previously registered by
1001 * register_netdevice_notifier(). The notifier is unlinked into the
1002 * kernel structures and may then be reused. A negative errno code
1003 * is returned on a failure.
1006 int unregister_netdevice_notifier(struct notifier_block *nb)
1011 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1017 * call_netdevice_notifiers - call all network notifier blocks
1018 * @val: value passed unmodified to notifier function
1019 * @v: pointer passed unmodified to notifier function
1021 * Call all network notifier blocks. Parameters and return value
1022 * are as for raw_notifier_call_chain().
1025 int call_netdevice_notifiers(unsigned long val, void *v)
1027 return raw_notifier_call_chain(&netdev_chain, val, v);
1030 /* When > 0 there are consumers of rx skb time stamps */
1031 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1033 void net_enable_timestamp(void)
1035 atomic_inc(&netstamp_needed);
1038 void net_disable_timestamp(void)
1040 atomic_dec(&netstamp_needed);
1043 void __net_timestamp(struct sk_buff *skb)
1047 do_gettimeofday(&tv);
1048 skb_set_timestamp(skb, &tv);
1050 EXPORT_SYMBOL(__net_timestamp);
1052 static inline void net_timestamp(struct sk_buff *skb)
1054 if (atomic_read(&netstamp_needed))
1055 __net_timestamp(skb);
1057 skb->tstamp.off_sec = 0;
1058 skb->tstamp.off_usec = 0;
1063 * Support routine. Sends outgoing frames to any network
1064 * taps currently in use.
1067 #if !((defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)))
1070 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1072 struct packet_type *ptype;
1077 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1078 /* Never send packets back to the socket
1079 * they originated from - MvS (miquels@drinkel.ow.org)
1081 if ((ptype->dev == dev || !ptype->dev) &&
1082 (ptype->af_packet_priv == NULL ||
1083 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1084 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1088 /* skb->nh should be correctly
1089 set by sender, so that the second statement is
1090 just protection against buggy protocols.
1092 skb2->mac.raw = skb2->data;
1094 if (skb2->nh.raw < skb2->data ||
1095 skb2->nh.raw > skb2->tail) {
1096 if (net_ratelimit())
1097 printk(KERN_CRIT "protocol %04x is "
1099 skb2->protocol, dev->name);
1100 skb2->nh.raw = skb2->data;
1103 skb2->h.raw = skb2->nh.raw;
1104 skb2->pkt_type = PACKET_OUTGOING;
1105 ptype->func(skb2, skb->dev, ptype, skb->dev);
1112 void __netif_schedule(struct net_device *dev)
1114 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1115 unsigned long flags;
1116 struct softnet_data *sd;
1118 local_irq_save(flags);
1119 sd = &__get_cpu_var(softnet_data);
1120 dev->next_sched = sd->output_queue;
1121 sd->output_queue = dev;
1122 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1123 local_irq_restore(flags);
1126 EXPORT_SYMBOL(__netif_schedule);
1128 void __netif_rx_schedule(struct net_device *dev)
1130 unsigned long flags;
1132 local_irq_save(flags);
1134 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1136 dev->quota += dev->weight;
1138 dev->quota = dev->weight;
1139 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1140 local_irq_restore(flags);
1142 EXPORT_SYMBOL(__netif_rx_schedule);
1144 void dev_kfree_skb_any(struct sk_buff *skb)
1146 if (in_irq() || irqs_disabled())
1147 dev_kfree_skb_irq(skb);
1151 EXPORT_SYMBOL(dev_kfree_skb_any);
1155 void netif_device_detach(struct net_device *dev)
1157 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1158 netif_running(dev)) {
1159 netif_stop_queue(dev);
1162 EXPORT_SYMBOL(netif_device_detach);
1164 void netif_device_attach(struct net_device *dev)
1166 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1167 netif_running(dev)) {
1168 netif_wake_queue(dev);
1169 __netdev_watchdog_up(dev);
1172 EXPORT_SYMBOL(netif_device_attach);
1176 * Invalidate hardware checksum when packet is to be mangled, and
1177 * complete checksum manually on outgoing path.
1179 int skb_checksum_help(struct sk_buff *skb, int inward)
1182 int ret = 0, offset = skb->h.raw - skb->data;
1185 goto out_set_summed;
1187 if (unlikely(skb_shinfo(skb)->gso_size)) {
1188 /* Let GSO fix up the checksum. */
1189 goto out_set_summed;
1192 if (skb_cloned(skb)) {
1193 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1198 BUG_ON(offset > (int)skb->len);
1199 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1201 offset = skb->tail - skb->h.raw;
1202 BUG_ON(offset <= 0);
1203 BUG_ON(skb->csum + 2 > offset);
1205 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1208 skb->ip_summed = CHECKSUM_NONE;
1214 * skb_gso_segment - Perform segmentation on skb.
1215 * @skb: buffer to segment
1216 * @features: features for the output path (see dev->features)
1218 * This function segments the given skb and returns a list of segments.
1220 * It may return NULL if the skb requires no segmentation. This is
1221 * only possible when GSO is used for verifying header integrity.
1223 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1225 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1226 struct packet_type *ptype;
1227 int type = skb->protocol;
1230 BUG_ON(skb_shinfo(skb)->frag_list);
1232 skb->mac.raw = skb->data;
1233 skb->mac_len = skb->nh.raw - skb->data;
1234 __skb_pull(skb, skb->mac_len);
1236 if (unlikely(skb->ip_summed != CHECKSUM_HW)) {
1237 if (skb_header_cloned(skb) &&
1238 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1239 return ERR_PTR(err);
1243 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1244 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1245 if (unlikely(skb->ip_summed != CHECKSUM_HW)) {
1246 err = ptype->gso_send_check(skb);
1247 segs = ERR_PTR(err);
1248 if (err || skb_gso_ok(skb, features))
1250 __skb_push(skb, skb->data - skb->nh.raw);
1252 segs = ptype->gso_segment(skb, features);
1258 __skb_push(skb, skb->data - skb->mac.raw);
1263 EXPORT_SYMBOL(skb_gso_segment);
1265 /* Take action when hardware reception checksum errors are detected. */
1267 void netdev_rx_csum_fault(struct net_device *dev)
1269 if (net_ratelimit()) {
1270 printk(KERN_ERR "%s: hw csum failure.\n",
1271 dev ? dev->name : "<unknown>");
1275 EXPORT_SYMBOL(netdev_rx_csum_fault);
1278 /* Actually, we should eliminate this check as soon as we know, that:
1279 * 1. IOMMU is present and allows to map all the memory.
1280 * 2. No high memory really exists on this machine.
1283 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1285 #ifdef CONFIG_HIGHMEM
1288 if (dev->features & NETIF_F_HIGHDMA)
1291 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1292 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1300 void (*destructor)(struct sk_buff *skb);
1303 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1305 static void dev_gso_skb_destructor(struct sk_buff *skb)
1307 struct dev_gso_cb *cb;
1310 struct sk_buff *nskb = skb->next;
1312 skb->next = nskb->next;
1315 } while (skb->next);
1317 cb = DEV_GSO_CB(skb);
1319 cb->destructor(skb);
1323 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1324 * @skb: buffer to segment
1326 * This function segments the given skb and stores the list of segments
1329 static int dev_gso_segment(struct sk_buff *skb)
1331 struct net_device *dev = skb->dev;
1332 struct sk_buff *segs;
1333 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1336 segs = skb_gso_segment(skb, features);
1338 /* Verifying header integrity only. */
1342 if (unlikely(IS_ERR(segs)))
1343 return PTR_ERR(segs);
1346 DEV_GSO_CB(skb)->destructor = skb->destructor;
1347 skb->destructor = dev_gso_skb_destructor;
1352 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1354 if (likely(!skb->next)) {
1356 dev_queue_xmit_nit(skb, dev);
1358 if (netif_needs_gso(dev, skb)) {
1359 if (unlikely(dev_gso_segment(skb)))
1365 return dev->hard_start_xmit(skb, dev);
1370 struct sk_buff *nskb = skb->next;
1373 skb->next = nskb->next;
1375 rc = dev->hard_start_xmit(nskb, dev);
1377 nskb->next = skb->next;
1381 if (unlikely(netif_queue_stopped(dev) && skb->next))
1382 return NETDEV_TX_BUSY;
1383 } while (skb->next);
1385 skb->destructor = DEV_GSO_CB(skb)->destructor;
1392 #define HARD_TX_LOCK(dev, cpu) { \
1393 if ((dev->features & NETIF_F_LLTX) == 0) { \
1394 netif_tx_lock(dev); \
1398 #define HARD_TX_UNLOCK(dev) { \
1399 if ((dev->features & NETIF_F_LLTX) == 0) { \
1400 netif_tx_unlock(dev); \
1405 inline int skb_checksum_setup(struct sk_buff *skb)
1407 if (skb->proto_csum_blank) {
1408 if (skb->protocol != htons(ETH_P_IP))
1410 skb->h.raw = (unsigned char *)skb->nh.iph + 4*skb->nh.iph->ihl;
1411 if (skb->h.raw >= skb->tail)
1413 switch (skb->nh.iph->protocol) {
1415 skb->csum = offsetof(struct tcphdr, check);
1418 skb->csum = offsetof(struct udphdr, check);
1421 if (net_ratelimit())
1422 printk(KERN_ERR "Attempting to checksum a non-"
1423 "TCP/UDP packet, dropping a protocol"
1424 " %d packet", skb->nh.iph->protocol);
1427 if ((skb->h.raw + skb->csum + 2) > skb->tail)
1429 skb->ip_summed = CHECKSUM_HW;
1430 skb->proto_csum_blank = 0;
1437 inline int skb_checksum_setup(struct sk_buff *skb) { return 0; }
1442 * dev_queue_xmit - transmit a buffer
1443 * @skb: buffer to transmit
1445 * Queue a buffer for transmission to a network device. The caller must
1446 * have set the device and priority and built the buffer before calling
1447 * this function. The function can be called from an interrupt.
1449 * A negative errno code is returned on a failure. A success does not
1450 * guarantee the frame will be transmitted as it may be dropped due
1451 * to congestion or traffic shaping.
1453 * -----------------------------------------------------------------------------------
1454 * I notice this method can also return errors from the queue disciplines,
1455 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1458 * Regardless of the return value, the skb is consumed, so it is currently
1459 * difficult to retry a send to this method. (You can bump the ref count
1460 * before sending to hold a reference for retry if you are careful.)
1462 * When calling this method, interrupts MUST be enabled. This is because
1463 * the BH enable code must have IRQs enabled so that it will not deadlock.
1467 int dev_queue_xmit(struct sk_buff *skb)
1469 struct net_device *dev = skb->dev;
1473 /* If a checksum-deferred packet is forwarded to a device that needs a
1474 * checksum, correct the pointers and force checksumming.
1476 if (skb_checksum_setup(skb))
1479 /* GSO will handle the following emulations directly. */
1480 if (netif_needs_gso(dev, skb))
1483 if (skb_shinfo(skb)->frag_list &&
1484 !(dev->features & NETIF_F_FRAGLIST) &&
1485 __skb_linearize(skb))
1488 /* Fragmented skb is linearized if device does not support SG,
1489 * or if at least one of fragments is in highmem and device
1490 * does not support DMA from it.
1492 if (skb_shinfo(skb)->nr_frags &&
1493 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1494 __skb_linearize(skb))
1497 /* If packet is not checksummed and device does not support
1498 * checksumming for this protocol, complete checksumming here.
1500 if (skb->ip_summed == CHECKSUM_HW &&
1501 (!(dev->features & NETIF_F_GEN_CSUM) &&
1502 (!(dev->features & NETIF_F_IP_CSUM) ||
1503 skb->protocol != htons(ETH_P_IP))))
1504 if (skb_checksum_help(skb, 0))
1508 spin_lock_prefetch(&dev->queue_lock);
1510 /* Disable soft irqs for various locks below. Also
1511 * stops preemption for RCU.
1515 /* Updates of qdisc are serialized by queue_lock.
1516 * The struct Qdisc which is pointed to by qdisc is now a
1517 * rcu structure - it may be accessed without acquiring
1518 * a lock (but the structure may be stale.) The freeing of the
1519 * qdisc will be deferred until it's known that there are no
1520 * more references to it.
1522 * If the qdisc has an enqueue function, we still need to
1523 * hold the queue_lock before calling it, since queue_lock
1524 * also serializes access to the device queue.
1527 q = rcu_dereference(dev->qdisc);
1528 #ifdef CONFIG_NET_CLS_ACT
1529 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1532 /* Grab device queue */
1533 spin_lock(&dev->queue_lock);
1536 rc = q->enqueue(skb, q);
1538 spin_unlock(&dev->queue_lock);
1540 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1543 spin_unlock(&dev->queue_lock);
1546 /* The device has no queue. Common case for software devices:
1547 loopback, all the sorts of tunnels...
1549 Really, it is unlikely that netif_tx_lock protection is necessary
1550 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1552 However, it is possible, that they rely on protection
1555 Check this and shot the lock. It is not prone from deadlocks.
1556 Either shot noqueue qdisc, it is even simpler 8)
1558 if (dev->flags & IFF_UP) {
1559 int cpu = smp_processor_id(); /* ok because BHs are off */
1561 if (dev->xmit_lock_owner != cpu) {
1563 HARD_TX_LOCK(dev, cpu);
1565 if (!netif_queue_stopped(dev)) {
1567 if (!dev_hard_start_xmit(skb, dev)) {
1568 HARD_TX_UNLOCK(dev);
1572 HARD_TX_UNLOCK(dev);
1573 if (net_ratelimit())
1574 printk(KERN_CRIT "Virtual device %s asks to "
1575 "queue packet!\n", dev->name);
1577 /* Recursion is detected! It is possible,
1579 if (net_ratelimit())
1580 printk(KERN_CRIT "Dead loop on virtual device "
1581 "%s, fix it urgently!\n", dev->name);
1586 rcu_read_unlock_bh();
1592 rcu_read_unlock_bh();
1597 /*=======================================================================
1599 =======================================================================*/
1601 int netdev_max_backlog = 1000;
1602 int netdev_budget = 300;
1603 int weight_p = 64; /* old backlog weight */
1605 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1609 * netif_rx - post buffer to the network code
1610 * @skb: buffer to post
1612 * This function receives a packet from a device driver and queues it for
1613 * the upper (protocol) levels to process. It always succeeds. The buffer
1614 * may be dropped during processing for congestion control or by the
1618 * NET_RX_SUCCESS (no congestion)
1619 * NET_RX_CN_LOW (low congestion)
1620 * NET_RX_CN_MOD (moderate congestion)
1621 * NET_RX_CN_HIGH (high congestion)
1622 * NET_RX_DROP (packet was dropped)
1626 int netif_rx(struct sk_buff *skb)
1628 struct softnet_data *queue;
1629 unsigned long flags;
1631 /* if netpoll wants it, pretend we never saw it */
1632 if (netpoll_rx(skb))
1635 if (!skb->tstamp.off_sec)
1639 * The code is rearranged so that the path is the most
1640 * short when CPU is congested, but is still operating.
1642 local_irq_save(flags);
1643 queue = &__get_cpu_var(softnet_data);
1645 __get_cpu_var(netdev_rx_stat).total++;
1646 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1647 if (queue->input_pkt_queue.qlen) {
1650 __skb_queue_tail(&queue->input_pkt_queue, skb);
1651 local_irq_restore(flags);
1652 return NET_RX_SUCCESS;
1655 netif_rx_schedule(&queue->backlog_dev);
1659 __get_cpu_var(netdev_rx_stat).dropped++;
1660 local_irq_restore(flags);
1666 int netif_rx_ni(struct sk_buff *skb)
1671 err = netif_rx(skb);
1672 if (local_softirq_pending())
1679 EXPORT_SYMBOL(netif_rx_ni);
1681 static inline struct net_device *skb_bond(struct sk_buff *skb)
1683 struct net_device *dev = skb->dev;
1686 if (skb_bond_should_drop(skb)) {
1690 skb->dev = dev->master;
1696 static void net_tx_action(struct softirq_action *h)
1698 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1700 if (sd->completion_queue) {
1701 struct sk_buff *clist;
1703 local_irq_disable();
1704 clist = sd->completion_queue;
1705 sd->completion_queue = NULL;
1709 struct sk_buff *skb = clist;
1710 clist = clist->next;
1712 BUG_TRAP(!atomic_read(&skb->users));
1717 if (sd->output_queue) {
1718 struct net_device *head;
1720 local_irq_disable();
1721 head = sd->output_queue;
1722 sd->output_queue = NULL;
1726 struct net_device *dev = head;
1727 head = head->next_sched;
1729 smp_mb__before_clear_bit();
1730 clear_bit(__LINK_STATE_SCHED, &dev->state);
1732 if (spin_trylock(&dev->queue_lock)) {
1734 spin_unlock(&dev->queue_lock);
1736 netif_schedule(dev);
1742 static __inline__ int deliver_skb(struct sk_buff *skb,
1743 struct packet_type *pt_prev,
1744 struct net_device *orig_dev)
1746 atomic_inc(&skb->users);
1747 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1750 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1751 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1753 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1754 unsigned char *addr);
1755 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1757 static __inline__ int handle_bridge(struct sk_buff **pskb,
1758 struct packet_type **pt_prev, int *ret,
1759 struct net_device *orig_dev)
1761 struct net_bridge_port *port;
1763 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1764 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1768 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1772 return br_handle_frame_hook(port, pskb);
1775 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1778 #ifdef CONFIG_NET_CLS_ACT
1779 /* TODO: Maybe we should just force sch_ingress to be compiled in
1780 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1781 * a compare and 2 stores extra right now if we dont have it on
1782 * but have CONFIG_NET_CLS_ACT
1783 * NOTE: This doesnt stop any functionality; if you dont have
1784 * the ingress scheduler, you just cant add policies on ingress.
1787 static int ing_filter(struct sk_buff *skb)
1790 struct net_device *dev = skb->dev;
1791 int result = TC_ACT_OK;
1793 if (dev->qdisc_ingress) {
1794 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1795 if (MAX_RED_LOOP < ttl++) {
1796 printk(KERN_WARNING "Redir loop detected Dropping packet (%s->%s)\n",
1797 skb->input_dev->name, skb->dev->name);
1801 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1803 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1805 spin_lock(&dev->ingress_lock);
1806 if ((q = dev->qdisc_ingress) != NULL)
1807 result = q->enqueue(skb, q);
1808 spin_unlock(&dev->ingress_lock);
1816 int netif_receive_skb(struct sk_buff *skb)
1818 struct packet_type *ptype, *pt_prev;
1819 struct net_device *orig_dev;
1820 int ret = NET_RX_DROP;
1821 unsigned short type;
1823 /* if we've gotten here through NAPI, check netpoll */
1824 if (skb->dev->poll && netpoll_rx(skb))
1827 if (!skb->tstamp.off_sec)
1830 if (!skb->input_dev)
1831 skb->input_dev = skb->dev;
1833 orig_dev = skb_bond(skb);
1838 __get_cpu_var(netdev_rx_stat).total++;
1840 skb->h.raw = skb->nh.raw = skb->data;
1841 skb->mac_len = skb->nh.raw - skb->mac.raw;
1847 #ifdef CONFIG_NET_CLS_ACT
1848 if (skb->tc_verd & TC_NCLS) {
1849 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1855 switch (skb->ip_summed) {
1856 case CHECKSUM_UNNECESSARY:
1857 skb->proto_data_valid = 1;
1860 /* XXX Implement me. */
1862 skb->proto_data_valid = 0;
1867 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1868 if (!ptype->dev || ptype->dev == skb->dev) {
1870 ret = deliver_skb(skb, pt_prev, orig_dev);
1875 #ifdef CONFIG_NET_CLS_ACT
1877 ret = deliver_skb(skb, pt_prev, orig_dev);
1878 pt_prev = NULL; /* noone else should process this after*/
1880 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1883 ret = ing_filter(skb);
1885 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1894 handle_diverter(skb);
1896 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1899 type = skb->protocol;
1900 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1901 if (ptype->type == type &&
1902 (!ptype->dev || ptype->dev == skb->dev)) {
1904 ret = deliver_skb(skb, pt_prev, orig_dev);
1910 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1913 /* Jamal, now you will not able to escape explaining
1914 * me how you were going to use this. :-)
1924 static int process_backlog(struct net_device *backlog_dev, int *budget)
1927 int quota = min(backlog_dev->quota, *budget);
1928 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1929 unsigned long start_time = jiffies;
1931 backlog_dev->weight = weight_p;
1933 struct sk_buff *skb;
1934 struct net_device *dev;
1936 local_irq_disable();
1937 skb = __skb_dequeue(&queue->input_pkt_queue);
1944 netif_receive_skb(skb);
1950 if (work >= quota || jiffies - start_time > 1)
1955 backlog_dev->quota -= work;
1960 backlog_dev->quota -= work;
1963 list_del(&backlog_dev->poll_list);
1964 smp_mb__before_clear_bit();
1965 netif_poll_enable(backlog_dev);
1971 static void net_rx_action(struct softirq_action *h)
1973 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1974 unsigned long start_time = jiffies;
1975 int budget = netdev_budget;
1978 local_irq_disable();
1980 while (!list_empty(&queue->poll_list)) {
1981 struct net_device *dev;
1983 if (budget <= 0 || jiffies - start_time > 1)
1988 dev = list_entry(queue->poll_list.next,
1989 struct net_device, poll_list);
1990 have = netpoll_poll_lock(dev);
1992 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1993 netpoll_poll_unlock(have);
1994 local_irq_disable();
1995 list_move_tail(&dev->poll_list, &queue->poll_list);
1997 dev->quota += dev->weight;
1999 dev->quota = dev->weight;
2001 netpoll_poll_unlock(have);
2003 local_irq_disable();
2007 #ifdef CONFIG_NET_DMA
2009 * There may not be any more sk_buffs coming right now, so push
2010 * any pending DMA copies to hardware
2012 if (net_dma_client) {
2013 struct dma_chan *chan;
2015 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
2016 dma_async_memcpy_issue_pending(chan);
2024 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2025 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2029 static gifconf_func_t * gifconf_list [NPROTO];
2032 * register_gifconf - register a SIOCGIF handler
2033 * @family: Address family
2034 * @gifconf: Function handler
2036 * Register protocol dependent address dumping routines. The handler
2037 * that is passed must not be freed or reused until it has been replaced
2038 * by another handler.
2040 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2042 if (family >= NPROTO)
2044 gifconf_list[family] = gifconf;
2050 * Map an interface index to its name (SIOCGIFNAME)
2054 * We need this ioctl for efficient implementation of the
2055 * if_indextoname() function required by the IPv6 API. Without
2056 * it, we would have to search all the interfaces to find a
2060 static int dev_ifname(struct ifreq __user *arg)
2062 struct net_device *dev;
2066 * Fetch the caller's info block.
2069 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2072 read_lock(&dev_base_lock);
2073 dev = __dev_get_by_index(ifr.ifr_ifindex);
2075 read_unlock(&dev_base_lock);
2079 strcpy(ifr.ifr_name, dev->name);
2080 read_unlock(&dev_base_lock);
2082 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2088 * Perform a SIOCGIFCONF call. This structure will change
2089 * size eventually, and there is nothing I can do about it.
2090 * Thus we will need a 'compatibility mode'.
2093 static int dev_ifconf(char __user *arg)
2096 struct net_device *dev;
2103 * Fetch the caller's info block.
2106 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2113 * Loop over the interfaces, and write an info block for each.
2117 for (dev = dev_base; dev; dev = dev->next) {
2118 if (vx_flags(VXF_HIDE_NETIF, 0) &&
2119 !dev_in_nx_info(dev, current->nx_info))
2121 for (i = 0; i < NPROTO; i++) {
2122 if (gifconf_list[i]) {
2125 done = gifconf_list[i](dev, NULL, 0);
2127 done = gifconf_list[i](dev, pos + total,
2137 * All done. Write the updated control block back to the caller.
2139 ifc.ifc_len = total;
2142 * Both BSD and Solaris return 0 here, so we do too.
2144 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2147 #ifdef CONFIG_PROC_FS
2149 * This is invoked by the /proc filesystem handler to display a device
2152 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2154 struct net_device *dev;
2157 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2159 return i == pos ? dev : NULL;
2162 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2164 read_lock(&dev_base_lock);
2165 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2168 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2171 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2174 void dev_seq_stop(struct seq_file *seq, void *v)
2176 read_unlock(&dev_base_lock);
2179 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2181 struct nx_info *nxi = current->nx_info;
2183 if (vx_flags(VXF_HIDE_NETIF, 0) && !dev_in_nx_info(dev, nxi))
2185 if (dev->get_stats) {
2186 struct net_device_stats *stats = dev->get_stats(dev);
2188 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2189 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2190 dev->name, stats->rx_bytes, stats->rx_packets,
2192 stats->rx_dropped + stats->rx_missed_errors,
2193 stats->rx_fifo_errors,
2194 stats->rx_length_errors + stats->rx_over_errors +
2195 stats->rx_crc_errors + stats->rx_frame_errors,
2196 stats->rx_compressed, stats->multicast,
2197 stats->tx_bytes, stats->tx_packets,
2198 stats->tx_errors, stats->tx_dropped,
2199 stats->tx_fifo_errors, stats->collisions,
2200 stats->tx_carrier_errors +
2201 stats->tx_aborted_errors +
2202 stats->tx_window_errors +
2203 stats->tx_heartbeat_errors,
2204 stats->tx_compressed);
2206 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2210 * Called from the PROCfs module. This now uses the new arbitrary sized
2211 * /proc/net interface to create /proc/net/dev
2213 static int dev_seq_show(struct seq_file *seq, void *v)
2215 if (v == SEQ_START_TOKEN)
2216 seq_puts(seq, "Inter-| Receive "
2218 " face |bytes packets errs drop fifo frame "
2219 "compressed multicast|bytes packets errs "
2220 "drop fifo colls carrier compressed\n");
2222 dev_seq_printf_stats(seq, v);
2226 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2228 struct netif_rx_stats *rc = NULL;
2230 while (*pos < NR_CPUS)
2231 if (cpu_online(*pos)) {
2232 rc = &per_cpu(netdev_rx_stat, *pos);
2239 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2241 return softnet_get_online(pos);
2244 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2247 return softnet_get_online(pos);
2250 static void softnet_seq_stop(struct seq_file *seq, void *v)
2254 static int softnet_seq_show(struct seq_file *seq, void *v)
2256 struct netif_rx_stats *s = v;
2258 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2259 s->total, s->dropped, s->time_squeeze, 0,
2260 0, 0, 0, 0, /* was fastroute */
2265 static struct seq_operations dev_seq_ops = {
2266 .start = dev_seq_start,
2267 .next = dev_seq_next,
2268 .stop = dev_seq_stop,
2269 .show = dev_seq_show,
2272 static int dev_seq_open(struct inode *inode, struct file *file)
2274 return seq_open(file, &dev_seq_ops);
2277 static struct file_operations dev_seq_fops = {
2278 .owner = THIS_MODULE,
2279 .open = dev_seq_open,
2281 .llseek = seq_lseek,
2282 .release = seq_release,
2285 static struct seq_operations softnet_seq_ops = {
2286 .start = softnet_seq_start,
2287 .next = softnet_seq_next,
2288 .stop = softnet_seq_stop,
2289 .show = softnet_seq_show,
2292 static int softnet_seq_open(struct inode *inode, struct file *file)
2294 return seq_open(file, &softnet_seq_ops);
2297 static struct file_operations softnet_seq_fops = {
2298 .owner = THIS_MODULE,
2299 .open = softnet_seq_open,
2301 .llseek = seq_lseek,
2302 .release = seq_release,
2305 #ifdef CONFIG_WIRELESS_EXT
2306 extern int wireless_proc_init(void);
2308 #define wireless_proc_init() 0
2311 static int __init dev_proc_init(void)
2315 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2317 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2319 if (wireless_proc_init())
2325 proc_net_remove("softnet_stat");
2327 proc_net_remove("dev");
2331 #define dev_proc_init() 0
2332 #endif /* CONFIG_PROC_FS */
2336 * netdev_set_master - set up master/slave pair
2337 * @slave: slave device
2338 * @master: new master device
2340 * Changes the master device of the slave. Pass %NULL to break the
2341 * bonding. The caller must hold the RTNL semaphore. On a failure
2342 * a negative errno code is returned. On success the reference counts
2343 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2344 * function returns zero.
2346 int netdev_set_master(struct net_device *slave, struct net_device *master)
2348 struct net_device *old = slave->master;
2358 slave->master = master;
2366 slave->flags |= IFF_SLAVE;
2368 slave->flags &= ~IFF_SLAVE;
2370 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2375 * dev_set_promiscuity - update promiscuity count on a device
2379 * Add or remove promiscuity from a device. While the count in the device
2380 * remains above zero the interface remains promiscuous. Once it hits zero
2381 * the device reverts back to normal filtering operation. A negative inc
2382 * value is used to drop promiscuity on the device.
2384 void dev_set_promiscuity(struct net_device *dev, int inc)
2386 unsigned short old_flags = dev->flags;
2388 if ((dev->promiscuity += inc) == 0)
2389 dev->flags &= ~IFF_PROMISC;
2391 dev->flags |= IFF_PROMISC;
2392 if (dev->flags != old_flags) {
2394 printk(KERN_INFO "device %s %s promiscuous mode\n",
2395 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2397 audit_log(current->audit_context, GFP_ATOMIC,
2398 AUDIT_ANOM_PROMISCUOUS,
2399 "dev=%s prom=%d old_prom=%d auid=%u",
2400 dev->name, (dev->flags & IFF_PROMISC),
2401 (old_flags & IFF_PROMISC),
2402 audit_get_loginuid(current->audit_context));
2407 * dev_set_allmulti - update allmulti count on a device
2411 * Add or remove reception of all multicast frames to a device. While the
2412 * count in the device remains above zero the interface remains listening
2413 * to all interfaces. Once it hits zero the device reverts back to normal
2414 * filtering operation. A negative @inc value is used to drop the counter
2415 * when releasing a resource needing all multicasts.
2418 void dev_set_allmulti(struct net_device *dev, int inc)
2420 unsigned short old_flags = dev->flags;
2422 dev->flags |= IFF_ALLMULTI;
2423 if ((dev->allmulti += inc) == 0)
2424 dev->flags &= ~IFF_ALLMULTI;
2425 if (dev->flags ^ old_flags)
2429 unsigned dev_get_flags(const struct net_device *dev)
2433 flags = (dev->flags & ~(IFF_PROMISC |
2438 (dev->gflags & (IFF_PROMISC |
2441 if (netif_running(dev)) {
2442 if (netif_oper_up(dev))
2443 flags |= IFF_RUNNING;
2444 if (netif_carrier_ok(dev))
2445 flags |= IFF_LOWER_UP;
2446 if (netif_dormant(dev))
2447 flags |= IFF_DORMANT;
2453 int dev_change_flags(struct net_device *dev, unsigned flags)
2456 int old_flags = dev->flags;
2459 * Set the flags on our device.
2462 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2463 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2465 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2469 * Load in the correct multicast list now the flags have changed.
2475 * Have we downed the interface. We handle IFF_UP ourselves
2476 * according to user attempts to set it, rather than blindly
2481 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2482 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2488 if (dev->flags & IFF_UP &&
2489 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2491 raw_notifier_call_chain(&netdev_chain,
2492 NETDEV_CHANGE, dev);
2494 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2495 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2496 dev->gflags ^= IFF_PROMISC;
2497 dev_set_promiscuity(dev, inc);
2500 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2501 is important. Some (broken) drivers set IFF_PROMISC, when
2502 IFF_ALLMULTI is requested not asking us and not reporting.
2504 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2505 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2506 dev->gflags ^= IFF_ALLMULTI;
2507 dev_set_allmulti(dev, inc);
2510 if (old_flags ^ dev->flags)
2511 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2516 int dev_set_mtu(struct net_device *dev, int new_mtu)
2520 if (new_mtu == dev->mtu)
2523 /* MTU must be positive. */
2527 if (!netif_device_present(dev))
2531 if (dev->change_mtu)
2532 err = dev->change_mtu(dev, new_mtu);
2535 if (!err && dev->flags & IFF_UP)
2536 raw_notifier_call_chain(&netdev_chain,
2537 NETDEV_CHANGEMTU, dev);
2541 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2545 if (!dev->set_mac_address)
2547 if (sa->sa_family != dev->type)
2549 if (!netif_device_present(dev))
2551 err = dev->set_mac_address(dev, sa);
2553 raw_notifier_call_chain(&netdev_chain,
2554 NETDEV_CHANGEADDR, dev);
2559 * Perform the SIOCxIFxxx calls.
2561 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2564 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2570 case SIOCGIFFLAGS: /* Get interface flags */
2571 ifr->ifr_flags = dev_get_flags(dev);
2574 case SIOCSIFFLAGS: /* Set interface flags */
2575 return dev_change_flags(dev, ifr->ifr_flags);
2577 case SIOCGIFMETRIC: /* Get the metric on the interface
2578 (currently unused) */
2579 ifr->ifr_metric = 0;
2582 case SIOCSIFMETRIC: /* Set the metric on the interface
2583 (currently unused) */
2586 case SIOCGIFMTU: /* Get the MTU of a device */
2587 ifr->ifr_mtu = dev->mtu;
2590 case SIOCSIFMTU: /* Set the MTU of a device */
2591 return dev_set_mtu(dev, ifr->ifr_mtu);
2595 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2597 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2598 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2599 ifr->ifr_hwaddr.sa_family = dev->type;
2603 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2605 case SIOCSIFHWBROADCAST:
2606 if (ifr->ifr_hwaddr.sa_family != dev->type)
2608 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2609 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2610 raw_notifier_call_chain(&netdev_chain,
2611 NETDEV_CHANGEADDR, dev);
2615 ifr->ifr_map.mem_start = dev->mem_start;
2616 ifr->ifr_map.mem_end = dev->mem_end;
2617 ifr->ifr_map.base_addr = dev->base_addr;
2618 ifr->ifr_map.irq = dev->irq;
2619 ifr->ifr_map.dma = dev->dma;
2620 ifr->ifr_map.port = dev->if_port;
2624 if (dev->set_config) {
2625 if (!netif_device_present(dev))
2627 return dev->set_config(dev, &ifr->ifr_map);
2632 if (!dev->set_multicast_list ||
2633 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2635 if (!netif_device_present(dev))
2637 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2641 if (!dev->set_multicast_list ||
2642 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2644 if (!netif_device_present(dev))
2646 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2650 ifr->ifr_ifindex = dev->ifindex;
2654 ifr->ifr_qlen = dev->tx_queue_len;
2658 if (ifr->ifr_qlen < 0)
2660 dev->tx_queue_len = ifr->ifr_qlen;
2664 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2665 return dev_change_name(dev, ifr->ifr_newname);
2668 * Unknown or private ioctl
2672 if ((cmd >= SIOCDEVPRIVATE &&
2673 cmd <= SIOCDEVPRIVATE + 15) ||
2674 cmd == SIOCBONDENSLAVE ||
2675 cmd == SIOCBONDRELEASE ||
2676 cmd == SIOCBONDSETHWADDR ||
2677 cmd == SIOCBONDSLAVEINFOQUERY ||
2678 cmd == SIOCBONDINFOQUERY ||
2679 cmd == SIOCBONDCHANGEACTIVE ||
2680 cmd == SIOCGMIIPHY ||
2681 cmd == SIOCGMIIREG ||
2682 cmd == SIOCSMIIREG ||
2683 cmd == SIOCBRADDIF ||
2684 cmd == SIOCBRDELIF ||
2685 cmd == SIOCWANDEV) {
2687 if (dev->do_ioctl) {
2688 if (netif_device_present(dev))
2689 err = dev->do_ioctl(dev, ifr,
2702 * This function handles all "interface"-type I/O control requests. The actual
2703 * 'doing' part of this is dev_ifsioc above.
2707 * dev_ioctl - network device ioctl
2708 * @cmd: command to issue
2709 * @arg: pointer to a struct ifreq in user space
2711 * Issue ioctl functions to devices. This is normally called by the
2712 * user space syscall interfaces but can sometimes be useful for
2713 * other purposes. The return value is the return from the syscall if
2714 * positive or a negative errno code on error.
2717 int dev_ioctl(unsigned int cmd, void __user *arg)
2723 /* One special case: SIOCGIFCONF takes ifconf argument
2724 and requires shared lock, because it sleeps writing
2728 if (cmd == SIOCGIFCONF) {
2730 ret = dev_ifconf((char __user *) arg);
2734 if (cmd == SIOCGIFNAME)
2735 return dev_ifname((struct ifreq __user *)arg);
2737 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2740 ifr.ifr_name[IFNAMSIZ-1] = 0;
2742 colon = strchr(ifr.ifr_name, ':');
2747 * See which interface the caller is talking about.
2752 * These ioctl calls:
2753 * - can be done by all.
2754 * - atomic and do not require locking.
2765 dev_load(ifr.ifr_name);
2766 read_lock(&dev_base_lock);
2767 ret = dev_ifsioc(&ifr, cmd);
2768 read_unlock(&dev_base_lock);
2772 if (copy_to_user(arg, &ifr,
2773 sizeof(struct ifreq)))
2779 dev_load(ifr.ifr_name);
2781 ret = dev_ethtool(&ifr);
2786 if (copy_to_user(arg, &ifr,
2787 sizeof(struct ifreq)))
2793 * These ioctl calls:
2794 * - require superuser power.
2795 * - require strict serialization.
2801 if (!capable(CAP_NET_ADMIN))
2803 dev_load(ifr.ifr_name);
2805 ret = dev_ifsioc(&ifr, cmd);
2810 if (copy_to_user(arg, &ifr,
2811 sizeof(struct ifreq)))
2817 * These ioctl calls:
2818 * - require superuser power.
2819 * - require strict serialization.
2820 * - do not return a value
2830 case SIOCSIFHWBROADCAST:
2833 case SIOCBONDENSLAVE:
2834 case SIOCBONDRELEASE:
2835 case SIOCBONDSETHWADDR:
2836 case SIOCBONDCHANGEACTIVE:
2839 if (!capable(CAP_NET_ADMIN))
2842 case SIOCBONDSLAVEINFOQUERY:
2843 case SIOCBONDINFOQUERY:
2844 dev_load(ifr.ifr_name);
2846 ret = dev_ifsioc(&ifr, cmd);
2851 /* Get the per device memory space. We can add this but
2852 * currently do not support it */
2854 /* Set the per device memory buffer space.
2855 * Not applicable in our case */
2860 * Unknown or private ioctl.
2863 if (cmd == SIOCWANDEV ||
2864 (cmd >= SIOCDEVPRIVATE &&
2865 cmd <= SIOCDEVPRIVATE + 15)) {
2866 dev_load(ifr.ifr_name);
2868 ret = dev_ifsioc(&ifr, cmd);
2870 if (!ret && copy_to_user(arg, &ifr,
2871 sizeof(struct ifreq)))
2875 #ifdef CONFIG_WIRELESS_EXT
2876 /* Take care of Wireless Extensions */
2877 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2878 /* If command is `set a parameter', or
2879 * `get the encoding parameters', check if
2880 * the user has the right to do it */
2881 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2882 || cmd == SIOCGIWENCODEEXT) {
2883 if (!capable(CAP_NET_ADMIN))
2886 dev_load(ifr.ifr_name);
2888 /* Follow me in net/core/wireless.c */
2889 ret = wireless_process_ioctl(&ifr, cmd);
2891 if (IW_IS_GET(cmd) &&
2892 copy_to_user(arg, &ifr,
2893 sizeof(struct ifreq)))
2897 #endif /* CONFIG_WIRELESS_EXT */
2904 * dev_new_index - allocate an ifindex
2906 * Returns a suitable unique value for a new device interface
2907 * number. The caller must hold the rtnl semaphore or the
2908 * dev_base_lock to be sure it remains unique.
2910 static int dev_new_index(void)
2916 if (!__dev_get_by_index(ifindex))
2921 static int dev_boot_phase = 1;
2923 /* Delayed registration/unregisteration */
2924 static DEFINE_SPINLOCK(net_todo_list_lock);
2925 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2927 static inline void net_set_todo(struct net_device *dev)
2929 spin_lock(&net_todo_list_lock);
2930 list_add_tail(&dev->todo_list, &net_todo_list);
2931 spin_unlock(&net_todo_list_lock);
2935 * register_netdevice - register a network device
2936 * @dev: device to register
2938 * Take a completed network device structure and add it to the kernel
2939 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2940 * chain. 0 is returned on success. A negative errno code is returned
2941 * on a failure to set up the device, or if the name is a duplicate.
2943 * Callers must hold the rtnl semaphore. You may want
2944 * register_netdev() instead of this.
2947 * The locking appears insufficient to guarantee two parallel registers
2948 * will not get the same name.
2951 int register_netdevice(struct net_device *dev)
2953 struct hlist_head *head;
2954 struct hlist_node *p;
2957 BUG_ON(dev_boot_phase);
2962 /* When net_device's are persistent, this will be fatal. */
2963 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2965 spin_lock_init(&dev->queue_lock);
2966 spin_lock_init(&dev->_xmit_lock);
2967 dev->xmit_lock_owner = -1;
2968 #ifdef CONFIG_NET_CLS_ACT
2969 spin_lock_init(&dev->ingress_lock);
2972 ret = alloc_divert_blk(dev);
2978 /* Init, if this function is available */
2980 ret = dev->init(dev);
2988 if (!dev_valid_name(dev->name)) {
2993 dev->ifindex = dev_new_index();
2994 if (dev->iflink == -1)
2995 dev->iflink = dev->ifindex;
2997 /* Check for existence of name */
2998 head = dev_name_hash(dev->name);
2999 hlist_for_each(p, head) {
3000 struct net_device *d
3001 = hlist_entry(p, struct net_device, name_hlist);
3002 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3008 /* Fix illegal SG+CSUM combinations. */
3009 if ((dev->features & NETIF_F_SG) &&
3010 !(dev->features & NETIF_F_ALL_CSUM)) {
3011 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3013 dev->features &= ~NETIF_F_SG;
3016 /* TSO requires that SG is present as well. */
3017 if ((dev->features & NETIF_F_TSO) &&
3018 !(dev->features & NETIF_F_SG)) {
3019 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3021 dev->features &= ~NETIF_F_TSO;
3023 if (dev->features & NETIF_F_UFO) {
3024 if (!(dev->features & NETIF_F_HW_CSUM)) {
3025 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3026 "NETIF_F_HW_CSUM feature.\n",
3028 dev->features &= ~NETIF_F_UFO;
3030 if (!(dev->features & NETIF_F_SG)) {
3031 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3032 "NETIF_F_SG feature.\n",
3034 dev->features &= ~NETIF_F_UFO;
3039 * nil rebuild_header routine,
3040 * that should be never called and used as just bug trap.
3043 if (!dev->rebuild_header)
3044 dev->rebuild_header = default_rebuild_header;
3046 ret = netdev_register_sysfs(dev);
3049 dev->reg_state = NETREG_REGISTERED;
3052 * Default initial state at registry is that the
3053 * device is present.
3056 set_bit(__LINK_STATE_PRESENT, &dev->state);
3059 dev_init_scheduler(dev);
3060 write_lock_bh(&dev_base_lock);
3062 dev_tail = &dev->next;
3063 hlist_add_head(&dev->name_hlist, head);
3064 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3066 write_unlock_bh(&dev_base_lock);
3068 /* Notify protocols, that a new device appeared. */
3069 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3076 free_divert_blk(dev);
3081 * register_netdev - register a network device
3082 * @dev: device to register
3084 * Take a completed network device structure and add it to the kernel
3085 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3086 * chain. 0 is returned on success. A negative errno code is returned
3087 * on a failure to set up the device, or if the name is a duplicate.
3089 * This is a wrapper around register_netdev that takes the rtnl semaphore
3090 * and expands the device name if you passed a format string to
3093 int register_netdev(struct net_device *dev)
3100 * If the name is a format string the caller wants us to do a
3103 if (strchr(dev->name, '%')) {
3104 err = dev_alloc_name(dev, dev->name);
3110 * Back compatibility hook. Kill this one in 2.5
3112 if (dev->name[0] == 0 || dev->name[0] == ' ') {
3113 err = dev_alloc_name(dev, "eth%d");
3118 err = register_netdevice(dev);
3123 EXPORT_SYMBOL(register_netdev);
3126 * netdev_wait_allrefs - wait until all references are gone.
3128 * This is called when unregistering network devices.
3130 * Any protocol or device that holds a reference should register
3131 * for netdevice notification, and cleanup and put back the
3132 * reference if they receive an UNREGISTER event.
3133 * We can get stuck here if buggy protocols don't correctly
3136 static void netdev_wait_allrefs(struct net_device *dev)
3138 unsigned long rebroadcast_time, warning_time;
3140 rebroadcast_time = warning_time = jiffies;
3141 while (atomic_read(&dev->refcnt) != 0) {
3142 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3145 /* Rebroadcast unregister notification */
3146 raw_notifier_call_chain(&netdev_chain,
3147 NETDEV_UNREGISTER, dev);
3149 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3151 /* We must not have linkwatch events
3152 * pending on unregister. If this
3153 * happens, we simply run the queue
3154 * unscheduled, resulting in a noop
3157 linkwatch_run_queue();
3162 rebroadcast_time = jiffies;
3167 if (time_after(jiffies, warning_time + 10 * HZ)) {
3168 printk(KERN_EMERG "unregister_netdevice: "
3169 "waiting for %s to become free. Usage "
3171 dev->name, atomic_read(&dev->refcnt));
3172 warning_time = jiffies;
3181 * register_netdevice(x1);
3182 * register_netdevice(x2);
3184 * unregister_netdevice(y1);
3185 * unregister_netdevice(y2);
3191 * We are invoked by rtnl_unlock() after it drops the semaphore.
3192 * This allows us to deal with problems:
3193 * 1) We can delete sysfs objects which invoke hotplug
3194 * without deadlocking with linkwatch via keventd.
3195 * 2) Since we run with the RTNL semaphore not held, we can sleep
3196 * safely in order to wait for the netdev refcnt to drop to zero.
3198 static DEFINE_MUTEX(net_todo_run_mutex);
3199 void netdev_run_todo(void)
3201 struct list_head list;
3203 /* Need to guard against multiple cpu's getting out of order. */
3204 mutex_lock(&net_todo_run_mutex);
3206 /* Not safe to do outside the semaphore. We must not return
3207 * until all unregister events invoked by the local processor
3208 * have been completed (either by this todo run, or one on
3211 if (list_empty(&net_todo_list))
3214 /* Snapshot list, allow later requests */
3215 spin_lock(&net_todo_list_lock);
3216 list_replace_init(&net_todo_list, &list);
3217 spin_unlock(&net_todo_list_lock);
3219 while (!list_empty(&list)) {
3220 struct net_device *dev
3221 = list_entry(list.next, struct net_device, todo_list);
3222 list_del(&dev->todo_list);
3224 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3225 printk(KERN_ERR "network todo '%s' but state %d\n",
3226 dev->name, dev->reg_state);
3231 netdev_unregister_sysfs(dev);
3232 dev->reg_state = NETREG_UNREGISTERED;
3234 netdev_wait_allrefs(dev);
3237 BUG_ON(atomic_read(&dev->refcnt));
3238 BUG_TRAP(!dev->ip_ptr);
3239 BUG_TRAP(!dev->ip6_ptr);
3240 BUG_TRAP(!dev->dn_ptr);
3242 /* It must be the very last action,
3243 * after this 'dev' may point to freed up memory.
3245 if (dev->destructor)
3246 dev->destructor(dev);
3250 mutex_unlock(&net_todo_run_mutex);
3254 * alloc_netdev - allocate network device
3255 * @sizeof_priv: size of private data to allocate space for
3256 * @name: device name format string
3257 * @setup: callback to initialize device
3259 * Allocates a struct net_device with private data area for driver use
3260 * and performs basic initialization.
3262 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3263 void (*setup)(struct net_device *))
3266 struct net_device *dev;
3269 /* ensure 32-byte alignment of both the device and private area */
3270 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3271 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3273 p = kzalloc(alloc_size, GFP_KERNEL);
3275 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3279 dev = (struct net_device *)
3280 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3281 dev->padded = (char *)dev - (char *)p;
3284 dev->priv = netdev_priv(dev);
3287 strcpy(dev->name, name);
3290 EXPORT_SYMBOL(alloc_netdev);
3293 * free_netdev - free network device
3296 * This function does the last stage of destroying an allocated device
3297 * interface. The reference to the device object is released.
3298 * If this is the last reference then it will be freed.
3300 void free_netdev(struct net_device *dev)
3303 /* Compatibility with error handling in drivers */
3304 if (dev->reg_state == NETREG_UNINITIALIZED) {
3305 kfree((char *)dev - dev->padded);
3309 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3310 dev->reg_state = NETREG_RELEASED;
3312 /* will free via class release */
3313 class_device_put(&dev->class_dev);
3315 kfree((char *)dev - dev->padded);
3319 /* Synchronize with packet receive processing. */
3320 void synchronize_net(void)
3327 * unregister_netdevice - remove device from the kernel
3330 * This function shuts down a device interface and removes it
3331 * from the kernel tables. On success 0 is returned, on a failure
3332 * a negative errno code is returned.
3334 * Callers must hold the rtnl semaphore. You may want
3335 * unregister_netdev() instead of this.
3338 int unregister_netdevice(struct net_device *dev)
3340 struct net_device *d, **dp;
3342 BUG_ON(dev_boot_phase);
3345 /* Some devices call without registering for initialization unwind. */
3346 if (dev->reg_state == NETREG_UNINITIALIZED) {
3347 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3348 "was registered\n", dev->name, dev);
3352 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3354 /* If device is running, close it first. */
3355 if (dev->flags & IFF_UP)
3358 /* And unlink it from device chain. */
3359 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3361 write_lock_bh(&dev_base_lock);
3362 hlist_del(&dev->name_hlist);
3363 hlist_del(&dev->index_hlist);
3364 if (dev_tail == &dev->next)
3367 write_unlock_bh(&dev_base_lock);
3372 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3377 dev->reg_state = NETREG_UNREGISTERING;
3381 /* Shutdown queueing discipline. */
3385 /* Notify protocols, that we are about to destroy
3386 this device. They should clean all the things.
3388 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3391 * Flush the multicast chain
3393 dev_mc_discard(dev);
3398 /* Notifier chain MUST detach us from master device. */
3399 BUG_TRAP(!dev->master);
3401 free_divert_blk(dev);
3403 /* Finish processing unregister after unlock */
3413 * unregister_netdev - remove device from the kernel
3416 * This function shuts down a device interface and removes it
3417 * from the kernel tables. On success 0 is returned, on a failure
3418 * a negative errno code is returned.
3420 * This is just a wrapper for unregister_netdevice that takes
3421 * the rtnl semaphore. In general you want to use this and not
3422 * unregister_netdevice.
3424 void unregister_netdev(struct net_device *dev)
3427 unregister_netdevice(dev);
3431 EXPORT_SYMBOL(unregister_netdev);
3433 #ifdef CONFIG_HOTPLUG_CPU
3434 static int dev_cpu_callback(struct notifier_block *nfb,
3435 unsigned long action,
3438 struct sk_buff **list_skb;
3439 struct net_device **list_net;
3440 struct sk_buff *skb;
3441 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3442 struct softnet_data *sd, *oldsd;
3444 if (action != CPU_DEAD)
3447 local_irq_disable();
3448 cpu = smp_processor_id();
3449 sd = &per_cpu(softnet_data, cpu);
3450 oldsd = &per_cpu(softnet_data, oldcpu);
3452 /* Find end of our completion_queue. */
3453 list_skb = &sd->completion_queue;
3455 list_skb = &(*list_skb)->next;
3456 /* Append completion queue from offline CPU. */
3457 *list_skb = oldsd->completion_queue;
3458 oldsd->completion_queue = NULL;
3460 /* Find end of our output_queue. */
3461 list_net = &sd->output_queue;
3463 list_net = &(*list_net)->next_sched;
3464 /* Append output queue from offline CPU. */
3465 *list_net = oldsd->output_queue;
3466 oldsd->output_queue = NULL;
3468 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3471 /* Process offline CPU's input_pkt_queue */
3472 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3477 #endif /* CONFIG_HOTPLUG_CPU */
3479 #ifdef CONFIG_NET_DMA
3481 * net_dma_rebalance -
3482 * This is called when the number of channels allocated to the net_dma_client
3483 * changes. The net_dma_client tries to have one DMA channel per CPU.
3485 static void net_dma_rebalance(void)
3487 unsigned int cpu, i, n;
3488 struct dma_chan *chan;
3490 if (net_dma_count == 0) {
3491 for_each_online_cpu(cpu)
3492 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
3497 cpu = first_cpu(cpu_online_map);
3500 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3501 n = ((num_online_cpus() / net_dma_count)
3502 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3505 per_cpu(softnet_data, cpu).net_dma = chan;
3506 cpu = next_cpu(cpu, cpu_online_map);
3515 * netdev_dma_event - event callback for the net_dma_client
3516 * @client: should always be net_dma_client
3517 * @chan: DMA channel for the event
3518 * @event: event type
3520 static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3521 enum dma_event event)
3523 spin_lock(&net_dma_event_lock);
3525 case DMA_RESOURCE_ADDED:
3527 net_dma_rebalance();
3529 case DMA_RESOURCE_REMOVED:
3531 net_dma_rebalance();
3536 spin_unlock(&net_dma_event_lock);
3540 * netdev_dma_regiser - register the networking subsystem as a DMA client
3542 static int __init netdev_dma_register(void)
3544 spin_lock_init(&net_dma_event_lock);
3545 net_dma_client = dma_async_client_register(netdev_dma_event);
3546 if (net_dma_client == NULL)
3549 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3554 static int __init netdev_dma_register(void) { return -ENODEV; }
3555 #endif /* CONFIG_NET_DMA */
3558 * Initialize the DEV module. At boot time this walks the device list and
3559 * unhooks any devices that fail to initialise (normally hardware not
3560 * present) and leaves us with a valid list of present and active devices.
3565 * This is called single threaded during boot, so no need
3566 * to take the rtnl semaphore.
3568 static int __init net_dev_init(void)
3570 int i, rc = -ENOMEM;
3572 BUG_ON(!dev_boot_phase);
3576 if (dev_proc_init())
3579 if (netdev_sysfs_init())
3582 INIT_LIST_HEAD(&ptype_all);
3583 for (i = 0; i < 16; i++)
3584 INIT_LIST_HEAD(&ptype_base[i]);
3586 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3587 INIT_HLIST_HEAD(&dev_name_head[i]);
3589 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3590 INIT_HLIST_HEAD(&dev_index_head[i]);
3593 * Initialise the packet receive queues.
3596 for_each_possible_cpu(i) {
3597 struct softnet_data *queue;
3599 queue = &per_cpu(softnet_data, i);
3600 skb_queue_head_init(&queue->input_pkt_queue);
3601 queue->completion_queue = NULL;
3602 INIT_LIST_HEAD(&queue->poll_list);
3603 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3604 queue->backlog_dev.weight = weight_p;
3605 queue->backlog_dev.poll = process_backlog;
3606 atomic_set(&queue->backlog_dev.refcnt, 1);
3609 netdev_dma_register();
3613 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3614 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3616 hotcpu_notifier(dev_cpu_callback, 0);
3624 subsys_initcall(net_dev_init);
3626 EXPORT_SYMBOL(__dev_get_by_index);
3627 EXPORT_SYMBOL(__dev_get_by_name);
3628 EXPORT_SYMBOL(__dev_remove_pack);
3629 EXPORT_SYMBOL(dev_valid_name);
3630 EXPORT_SYMBOL(dev_add_pack);
3631 EXPORT_SYMBOL(dev_alloc_name);
3632 EXPORT_SYMBOL(dev_close);
3633 EXPORT_SYMBOL(dev_get_by_flags);
3634 EXPORT_SYMBOL(dev_get_by_index);
3635 EXPORT_SYMBOL(dev_get_by_name);
3636 EXPORT_SYMBOL(dev_open);
3637 EXPORT_SYMBOL(dev_queue_xmit);
3638 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
3639 EXPORT_SYMBOL(dev_queue_xmit_nit);
3641 EXPORT_SYMBOL(dev_remove_pack);
3642 EXPORT_SYMBOL(dev_set_allmulti);
3643 EXPORT_SYMBOL(dev_set_promiscuity);
3644 EXPORT_SYMBOL(dev_change_flags);
3645 EXPORT_SYMBOL(dev_set_mtu);
3646 EXPORT_SYMBOL(dev_set_mac_address);
3647 EXPORT_SYMBOL(free_netdev);
3648 EXPORT_SYMBOL(netdev_boot_setup_check);
3649 EXPORT_SYMBOL(netdev_set_master);
3650 EXPORT_SYMBOL(netdev_state_change);
3651 EXPORT_SYMBOL(netif_receive_skb);
3652 EXPORT_SYMBOL(netif_rx);
3653 EXPORT_SYMBOL(register_gifconf);
3654 EXPORT_SYMBOL(register_netdevice);
3655 EXPORT_SYMBOL(register_netdevice_notifier);
3656 EXPORT_SYMBOL(skb_checksum_help);
3657 EXPORT_SYMBOL(synchronize_net);
3658 EXPORT_SYMBOL(unregister_netdevice);
3659 EXPORT_SYMBOL(unregister_netdevice_notifier);
3660 EXPORT_SYMBOL(net_enable_timestamp);
3661 EXPORT_SYMBOL(net_disable_timestamp);
3662 EXPORT_SYMBOL(dev_get_flags);
3663 EXPORT_SYMBOL(skb_checksum_setup);
3665 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3666 EXPORT_SYMBOL(br_handle_frame_hook);
3667 EXPORT_SYMBOL(br_fdb_get_hook);
3668 EXPORT_SYMBOL(br_fdb_put_hook);
3672 EXPORT_SYMBOL(dev_load);
3675 EXPORT_PER_CPU_SYMBOL(softnet_data);