2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/mutex.h>
84 #include <linux/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/notifier.h>
94 #include <linux/skbuff.h>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <linux/wireless.h>
113 #include <net/iw_handler.h>
114 #include <asm/current.h>
115 #include <linux/audit.h>
116 #include <linux/dmaengine.h>
117 #include <linux/err.h>
118 #include <linux/ctype.h>
121 * The list of packet types we will receive (as opposed to discard)
122 * and the routines to invoke.
124 * Why 16. Because with 16 the only overlap we get on a hash of the
125 * low nibble of the protocol value is RARP/SNAP/X.25.
127 * NOTE: That is no longer true with the addition of VLAN tags. Not
128 * sure which should go first, but I bet it won't make much
129 * difference if we are running VLANs. The good news is that
130 * this protocol won't be in the list unless compiled in, so
131 * the average user (w/out VLANs) will not be adversely affected.
148 static DEFINE_SPINLOCK(ptype_lock);
149 static struct list_head ptype_base[16]; /* 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. We also disallow any kind of
638 int dev_valid_name(const char *name)
642 if (strlen(name) >= IFNAMSIZ)
644 if (!strcmp(name, ".") || !strcmp(name, ".."))
648 if (*name == '/' || isspace(*name))
656 * dev_alloc_name - allocate a name for a device
658 * @name: name format string
660 * Passed a format string - eg "lt%d" it will try and find a suitable
661 * id. It scans list of devices to build up a free map, then chooses
662 * the first empty slot. The caller must hold the dev_base or rtnl lock
663 * while allocating the name and adding the device in order to avoid
665 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
666 * Returns the number of the unit assigned or a negative errno code.
669 int dev_alloc_name(struct net_device *dev, const char *name)
674 const int max_netdevices = 8*PAGE_SIZE;
676 struct net_device *d;
678 p = strnchr(name, IFNAMSIZ-1, '%');
681 * Verify the string as this thing may have come from
682 * the user. There must be either one "%d" and no other "%"
685 if (p[1] != 'd' || strchr(p + 2, '%'))
688 /* Use one page as a bit array of possible slots */
689 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
693 for (d = dev_base; d; d = d->next) {
694 if (!sscanf(d->name, name, &i))
696 if (i < 0 || i >= max_netdevices)
699 /* avoid cases where sscanf is not exact inverse of printf */
700 snprintf(buf, sizeof(buf), name, i);
701 if (!strncmp(buf, d->name, IFNAMSIZ))
705 i = find_first_zero_bit(inuse, max_netdevices);
706 free_page((unsigned long) inuse);
709 snprintf(buf, sizeof(buf), name, i);
710 if (!__dev_get_by_name(buf)) {
711 strlcpy(dev->name, buf, IFNAMSIZ);
715 /* It is possible to run out of possible slots
716 * when the name is long and there isn't enough space left
717 * for the digits, or if all bits are used.
724 * dev_change_name - change name of a device
726 * @newname: name (or format string) must be at least IFNAMSIZ
728 * Change name of a device, can pass format strings "eth%d".
731 int dev_change_name(struct net_device *dev, char *newname)
737 if (dev->flags & IFF_UP)
740 if (!dev_valid_name(newname))
743 if (strchr(newname, '%')) {
744 err = dev_alloc_name(dev, newname);
747 strcpy(newname, dev->name);
749 else if (__dev_get_by_name(newname))
752 strlcpy(dev->name, newname, IFNAMSIZ);
754 device_rename(&dev->dev, dev->name);
755 hlist_del(&dev->name_hlist);
756 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
757 raw_notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
763 * netdev_features_change - device changes features
764 * @dev: device to cause notification
766 * Called to indicate a device has changed features.
768 void netdev_features_change(struct net_device *dev)
770 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
772 EXPORT_SYMBOL(netdev_features_change);
775 * netdev_state_change - device changes state
776 * @dev: device to cause notification
778 * Called to indicate a device has changed state. This function calls
779 * the notifier chains for netdev_chain and sends a NEWLINK message
780 * to the routing socket.
782 void netdev_state_change(struct net_device *dev)
784 if (dev->flags & IFF_UP) {
785 raw_notifier_call_chain(&netdev_chain,
787 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
792 * dev_load - load a network module
793 * @name: name of interface
795 * If a network interface is not present and the process has suitable
796 * privileges this function loads the module. If module loading is not
797 * available in this kernel then it becomes a nop.
800 void dev_load(const char *name)
802 struct net_device *dev;
804 read_lock(&dev_base_lock);
805 dev = __dev_get_by_name(name);
806 read_unlock(&dev_base_lock);
808 if (!dev && capable(CAP_SYS_MODULE))
809 request_module("%s", name);
812 static int default_rebuild_header(struct sk_buff *skb)
814 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
815 skb->dev ? skb->dev->name : "NULL!!!");
822 * dev_open - prepare an interface for use.
823 * @dev: device to open
825 * Takes a device from down to up state. The device's private open
826 * function is invoked and then the multicast lists are loaded. Finally
827 * the device is moved into the up state and a %NETDEV_UP message is
828 * sent to the netdev notifier chain.
830 * Calling this function on an active interface is a nop. On a failure
831 * a negative errno code is returned.
833 int dev_open(struct net_device *dev)
841 if (dev->flags & IFF_UP)
845 * Is it even present?
847 if (!netif_device_present(dev))
851 * Call device private open method
853 set_bit(__LINK_STATE_START, &dev->state);
855 ret = dev->open(dev);
857 clear_bit(__LINK_STATE_START, &dev->state);
861 * If it went open OK then:
868 dev->flags |= IFF_UP;
871 * Initialize multicasting status
876 * Wakeup transmit queue engine
881 * ... and announce new interface.
883 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
889 * dev_close - shutdown an interface.
890 * @dev: device to shutdown
892 * This function moves an active device into down state. A
893 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
894 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
897 int dev_close(struct net_device *dev)
899 if (!(dev->flags & IFF_UP))
903 * Tell people we are going down, so that they can
904 * prepare to death, when device is still operating.
906 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
910 clear_bit(__LINK_STATE_START, &dev->state);
912 /* Synchronize to scheduled poll. We cannot touch poll list,
913 * it can be even on different cpu. So just clear netif_running(),
914 * and wait when poll really will happen. Actually, the best place
915 * for this is inside dev->stop() after device stopped its irq
916 * engine, but this requires more changes in devices. */
918 smp_mb__after_clear_bit(); /* Commit netif_running(). */
919 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
925 * Call the device specific close. This cannot fail.
926 * Only if device is UP
928 * We allow it to be called even after a DETACH hot-plug
935 * Device is now down.
938 dev->flags &= ~IFF_UP;
941 * Tell people we are down
943 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
950 * Device change register/unregister. These are not inline or static
951 * as we export them to the world.
955 * register_netdevice_notifier - register a network notifier block
958 * Register a notifier to be called when network device events occur.
959 * The notifier passed is linked into the kernel structures and must
960 * not be reused until it has been unregistered. A negative errno code
961 * is returned on a failure.
963 * When registered all registration and up events are replayed
964 * to the new notifier to allow device to have a race free
965 * view of the network device list.
968 int register_netdevice_notifier(struct notifier_block *nb)
970 struct net_device *dev;
974 err = raw_notifier_chain_register(&netdev_chain, nb);
976 for (dev = dev_base; dev; dev = dev->next) {
977 nb->notifier_call(nb, NETDEV_REGISTER, dev);
979 if (dev->flags & IFF_UP)
980 nb->notifier_call(nb, NETDEV_UP, dev);
988 * unregister_netdevice_notifier - unregister a network notifier block
991 * Unregister a notifier previously registered by
992 * register_netdevice_notifier(). The notifier is unlinked into the
993 * kernel structures and may then be reused. A negative errno code
994 * is returned on a failure.
997 int unregister_netdevice_notifier(struct notifier_block *nb)
1002 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1008 * call_netdevice_notifiers - call all network notifier blocks
1009 * @val: value passed unmodified to notifier function
1010 * @v: pointer passed unmodified to notifier function
1012 * Call all network notifier blocks. Parameters and return value
1013 * are as for raw_notifier_call_chain().
1016 int call_netdevice_notifiers(unsigned long val, void *v)
1018 return raw_notifier_call_chain(&netdev_chain, val, v);
1021 /* When > 0 there are consumers of rx skb time stamps */
1022 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1024 void net_enable_timestamp(void)
1026 atomic_inc(&netstamp_needed);
1029 void net_disable_timestamp(void)
1031 atomic_dec(&netstamp_needed);
1034 static inline void net_timestamp(struct sk_buff *skb)
1036 if (atomic_read(&netstamp_needed))
1037 __net_timestamp(skb);
1039 skb->tstamp.tv64 = 0;
1043 * Support routine. Sends outgoing frames to any network
1044 * taps currently in use.
1047 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1049 struct packet_type *ptype;
1054 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1055 /* Never send packets back to the socket
1056 * they originated from - MvS (miquels@drinkel.ow.org)
1058 if ((ptype->dev == dev || !ptype->dev) &&
1059 (ptype->af_packet_priv == NULL ||
1060 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1061 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1065 /* skb->nh should be correctly
1066 set by sender, so that the second statement is
1067 just protection against buggy protocols.
1069 skb_reset_mac_header(skb2);
1071 if (skb2->nh.raw < skb2->data ||
1072 skb2->nh.raw > skb2->tail) {
1073 if (net_ratelimit())
1074 printk(KERN_CRIT "protocol %04x is "
1076 skb2->protocol, dev->name);
1077 skb_reset_network_header(skb2);
1080 skb2->h.raw = skb2->nh.raw;
1081 skb2->pkt_type = PACKET_OUTGOING;
1082 ptype->func(skb2, skb->dev, ptype, skb->dev);
1089 void __netif_schedule(struct net_device *dev)
1091 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1092 unsigned long flags;
1093 struct softnet_data *sd;
1095 local_irq_save(flags);
1096 sd = &__get_cpu_var(softnet_data);
1097 dev->next_sched = sd->output_queue;
1098 sd->output_queue = dev;
1099 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1100 local_irq_restore(flags);
1103 EXPORT_SYMBOL(__netif_schedule);
1105 void __netif_rx_schedule(struct net_device *dev)
1107 unsigned long flags;
1109 local_irq_save(flags);
1111 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1113 dev->quota += dev->weight;
1115 dev->quota = dev->weight;
1116 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1117 local_irq_restore(flags);
1119 EXPORT_SYMBOL(__netif_rx_schedule);
1121 void dev_kfree_skb_any(struct sk_buff *skb)
1123 if (in_irq() || irqs_disabled())
1124 dev_kfree_skb_irq(skb);
1128 EXPORT_SYMBOL(dev_kfree_skb_any);
1132 void netif_device_detach(struct net_device *dev)
1134 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1135 netif_running(dev)) {
1136 netif_stop_queue(dev);
1139 EXPORT_SYMBOL(netif_device_detach);
1141 void netif_device_attach(struct net_device *dev)
1143 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1144 netif_running(dev)) {
1145 netif_wake_queue(dev);
1146 __netdev_watchdog_up(dev);
1149 EXPORT_SYMBOL(netif_device_attach);
1153 * Invalidate hardware checksum when packet is to be mangled, and
1154 * complete checksum manually on outgoing path.
1156 int skb_checksum_help(struct sk_buff *skb)
1159 int ret = 0, offset = skb->h.raw - skb->data;
1161 if (skb->ip_summed == CHECKSUM_COMPLETE)
1162 goto out_set_summed;
1164 if (unlikely(skb_shinfo(skb)->gso_size)) {
1165 /* Let GSO fix up the checksum. */
1166 goto out_set_summed;
1169 if (skb_cloned(skb)) {
1170 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1175 BUG_ON(offset > (int)skb->len);
1176 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1178 offset = skb->tail - skb->h.raw;
1179 BUG_ON(offset <= 0);
1180 BUG_ON(skb->csum_offset + 2 > offset);
1182 *(__sum16*)(skb->h.raw + skb->csum_offset) = csum_fold(csum);
1185 skb->ip_summed = CHECKSUM_NONE;
1191 * skb_gso_segment - Perform segmentation on skb.
1192 * @skb: buffer to segment
1193 * @features: features for the output path (see dev->features)
1195 * This function segments the given skb and returns a list of segments.
1197 * It may return NULL if the skb requires no segmentation. This is
1198 * only possible when GSO is used for verifying header integrity.
1200 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1202 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1203 struct packet_type *ptype;
1204 __be16 type = skb->protocol;
1207 BUG_ON(skb_shinfo(skb)->frag_list);
1209 skb_reset_mac_header(skb);
1210 skb->mac_len = skb->nh.raw - skb->data;
1211 __skb_pull(skb, skb->mac_len);
1213 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1214 if (skb_header_cloned(skb) &&
1215 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1216 return ERR_PTR(err);
1220 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1221 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1222 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1223 err = ptype->gso_send_check(skb);
1224 segs = ERR_PTR(err);
1225 if (err || skb_gso_ok(skb, features))
1227 __skb_push(skb, skb->data - skb->nh.raw);
1229 segs = ptype->gso_segment(skb, features);
1235 __skb_push(skb, skb->data - skb_mac_header(skb));
1240 EXPORT_SYMBOL(skb_gso_segment);
1242 /* Take action when hardware reception checksum errors are detected. */
1244 void netdev_rx_csum_fault(struct net_device *dev)
1246 if (net_ratelimit()) {
1247 printk(KERN_ERR "%s: hw csum failure.\n",
1248 dev ? dev->name : "<unknown>");
1252 EXPORT_SYMBOL(netdev_rx_csum_fault);
1255 /* Actually, we should eliminate this check as soon as we know, that:
1256 * 1. IOMMU is present and allows to map all the memory.
1257 * 2. No high memory really exists on this machine.
1260 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1262 #ifdef CONFIG_HIGHMEM
1265 if (dev->features & NETIF_F_HIGHDMA)
1268 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1269 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1277 void (*destructor)(struct sk_buff *skb);
1280 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1282 static void dev_gso_skb_destructor(struct sk_buff *skb)
1284 struct dev_gso_cb *cb;
1287 struct sk_buff *nskb = skb->next;
1289 skb->next = nskb->next;
1292 } while (skb->next);
1294 cb = DEV_GSO_CB(skb);
1296 cb->destructor(skb);
1300 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1301 * @skb: buffer to segment
1303 * This function segments the given skb and stores the list of segments
1306 static int dev_gso_segment(struct sk_buff *skb)
1308 struct net_device *dev = skb->dev;
1309 struct sk_buff *segs;
1310 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1313 segs = skb_gso_segment(skb, features);
1315 /* Verifying header integrity only. */
1319 if (unlikely(IS_ERR(segs)))
1320 return PTR_ERR(segs);
1323 DEV_GSO_CB(skb)->destructor = skb->destructor;
1324 skb->destructor = dev_gso_skb_destructor;
1329 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1331 if (likely(!skb->next)) {
1333 dev_queue_xmit_nit(skb, dev);
1335 if (netif_needs_gso(dev, skb)) {
1336 if (unlikely(dev_gso_segment(skb)))
1342 return dev->hard_start_xmit(skb, dev);
1347 struct sk_buff *nskb = skb->next;
1350 skb->next = nskb->next;
1352 rc = dev->hard_start_xmit(nskb, dev);
1354 nskb->next = skb->next;
1358 if (unlikely(netif_queue_stopped(dev) && skb->next))
1359 return NETDEV_TX_BUSY;
1360 } while (skb->next);
1362 skb->destructor = DEV_GSO_CB(skb)->destructor;
1369 #define HARD_TX_LOCK(dev, cpu) { \
1370 if ((dev->features & NETIF_F_LLTX) == 0) { \
1371 netif_tx_lock(dev); \
1375 #define HARD_TX_UNLOCK(dev) { \
1376 if ((dev->features & NETIF_F_LLTX) == 0) { \
1377 netif_tx_unlock(dev); \
1382 * dev_queue_xmit - transmit a buffer
1383 * @skb: buffer to transmit
1385 * Queue a buffer for transmission to a network device. The caller must
1386 * have set the device and priority and built the buffer before calling
1387 * this function. The function can be called from an interrupt.
1389 * A negative errno code is returned on a failure. A success does not
1390 * guarantee the frame will be transmitted as it may be dropped due
1391 * to congestion or traffic shaping.
1393 * -----------------------------------------------------------------------------------
1394 * I notice this method can also return errors from the queue disciplines,
1395 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1398 * Regardless of the return value, the skb is consumed, so it is currently
1399 * difficult to retry a send to this method. (You can bump the ref count
1400 * before sending to hold a reference for retry if you are careful.)
1402 * When calling this method, interrupts MUST be enabled. This is because
1403 * the BH enable code must have IRQs enabled so that it will not deadlock.
1407 int dev_queue_xmit(struct sk_buff *skb)
1409 struct net_device *dev = skb->dev;
1413 /* GSO will handle the following emulations directly. */
1414 if (netif_needs_gso(dev, skb))
1417 if (skb_shinfo(skb)->frag_list &&
1418 !(dev->features & NETIF_F_FRAGLIST) &&
1419 __skb_linearize(skb))
1422 /* Fragmented skb is linearized if device does not support SG,
1423 * or if at least one of fragments is in highmem and device
1424 * does not support DMA from it.
1426 if (skb_shinfo(skb)->nr_frags &&
1427 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1428 __skb_linearize(skb))
1431 /* If packet is not checksummed and device does not support
1432 * checksumming for this protocol, complete checksumming here.
1434 if (skb->ip_summed == CHECKSUM_PARTIAL &&
1435 (!(dev->features & NETIF_F_GEN_CSUM) &&
1436 (!(dev->features & NETIF_F_IP_CSUM) ||
1437 skb->protocol != htons(ETH_P_IP))))
1438 if (skb_checksum_help(skb))
1442 spin_lock_prefetch(&dev->queue_lock);
1444 /* Disable soft irqs for various locks below. Also
1445 * stops preemption for RCU.
1449 /* Updates of qdisc are serialized by queue_lock.
1450 * The struct Qdisc which is pointed to by qdisc is now a
1451 * rcu structure - it may be accessed without acquiring
1452 * a lock (but the structure may be stale.) The freeing of the
1453 * qdisc will be deferred until it's known that there are no
1454 * more references to it.
1456 * If the qdisc has an enqueue function, we still need to
1457 * hold the queue_lock before calling it, since queue_lock
1458 * also serializes access to the device queue.
1461 q = rcu_dereference(dev->qdisc);
1462 #ifdef CONFIG_NET_CLS_ACT
1463 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1466 /* Grab device queue */
1467 spin_lock(&dev->queue_lock);
1470 rc = q->enqueue(skb, q);
1472 spin_unlock(&dev->queue_lock);
1474 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1477 spin_unlock(&dev->queue_lock);
1480 /* The device has no queue. Common case for software devices:
1481 loopback, all the sorts of tunnels...
1483 Really, it is unlikely that netif_tx_lock protection is necessary
1484 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1486 However, it is possible, that they rely on protection
1489 Check this and shot the lock. It is not prone from deadlocks.
1490 Either shot noqueue qdisc, it is even simpler 8)
1492 if (dev->flags & IFF_UP) {
1493 int cpu = smp_processor_id(); /* ok because BHs are off */
1495 if (dev->xmit_lock_owner != cpu) {
1497 HARD_TX_LOCK(dev, cpu);
1499 if (!netif_queue_stopped(dev)) {
1501 if (!dev_hard_start_xmit(skb, dev)) {
1502 HARD_TX_UNLOCK(dev);
1506 HARD_TX_UNLOCK(dev);
1507 if (net_ratelimit())
1508 printk(KERN_CRIT "Virtual device %s asks to "
1509 "queue packet!\n", dev->name);
1511 /* Recursion is detected! It is possible,
1513 if (net_ratelimit())
1514 printk(KERN_CRIT "Dead loop on virtual device "
1515 "%s, fix it urgently!\n", dev->name);
1520 rcu_read_unlock_bh();
1526 rcu_read_unlock_bh();
1531 /*=======================================================================
1533 =======================================================================*/
1535 int netdev_max_backlog = 1000;
1536 int netdev_budget = 300;
1537 int weight_p = 64; /* old backlog weight */
1539 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1543 * netif_rx - post buffer to the network code
1544 * @skb: buffer to post
1546 * This function receives a packet from a device driver and queues it for
1547 * the upper (protocol) levels to process. It always succeeds. The buffer
1548 * may be dropped during processing for congestion control or by the
1552 * NET_RX_SUCCESS (no congestion)
1553 * NET_RX_CN_LOW (low congestion)
1554 * NET_RX_CN_MOD (moderate congestion)
1555 * NET_RX_CN_HIGH (high congestion)
1556 * NET_RX_DROP (packet was dropped)
1560 int netif_rx(struct sk_buff *skb)
1562 struct softnet_data *queue;
1563 unsigned long flags;
1565 /* if netpoll wants it, pretend we never saw it */
1566 if (netpoll_rx(skb))
1569 if (!skb->tstamp.tv64)
1573 * The code is rearranged so that the path is the most
1574 * short when CPU is congested, but is still operating.
1576 local_irq_save(flags);
1577 queue = &__get_cpu_var(softnet_data);
1579 __get_cpu_var(netdev_rx_stat).total++;
1580 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1581 if (queue->input_pkt_queue.qlen) {
1584 __skb_queue_tail(&queue->input_pkt_queue, skb);
1585 local_irq_restore(flags);
1586 return NET_RX_SUCCESS;
1589 netif_rx_schedule(&queue->backlog_dev);
1593 __get_cpu_var(netdev_rx_stat).dropped++;
1594 local_irq_restore(flags);
1600 int netif_rx_ni(struct sk_buff *skb)
1605 err = netif_rx(skb);
1606 if (local_softirq_pending())
1613 EXPORT_SYMBOL(netif_rx_ni);
1615 static inline struct net_device *skb_bond(struct sk_buff *skb)
1617 struct net_device *dev = skb->dev;
1620 if (skb_bond_should_drop(skb)) {
1624 skb->dev = dev->master;
1630 static void net_tx_action(struct softirq_action *h)
1632 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1634 if (sd->completion_queue) {
1635 struct sk_buff *clist;
1637 local_irq_disable();
1638 clist = sd->completion_queue;
1639 sd->completion_queue = NULL;
1643 struct sk_buff *skb = clist;
1644 clist = clist->next;
1646 BUG_TRAP(!atomic_read(&skb->users));
1651 if (sd->output_queue) {
1652 struct net_device *head;
1654 local_irq_disable();
1655 head = sd->output_queue;
1656 sd->output_queue = NULL;
1660 struct net_device *dev = head;
1661 head = head->next_sched;
1663 smp_mb__before_clear_bit();
1664 clear_bit(__LINK_STATE_SCHED, &dev->state);
1666 if (spin_trylock(&dev->queue_lock)) {
1668 spin_unlock(&dev->queue_lock);
1670 netif_schedule(dev);
1676 static inline int deliver_skb(struct sk_buff *skb,
1677 struct packet_type *pt_prev,
1678 struct net_device *orig_dev)
1680 atomic_inc(&skb->users);
1681 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1684 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1685 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1687 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1688 unsigned char *addr);
1689 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1691 static __inline__ int handle_bridge(struct sk_buff **pskb,
1692 struct packet_type **pt_prev, int *ret,
1693 struct net_device *orig_dev)
1695 struct net_bridge_port *port;
1697 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1698 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1702 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1706 return br_handle_frame_hook(port, pskb);
1709 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1712 #ifdef CONFIG_NET_CLS_ACT
1713 /* TODO: Maybe we should just force sch_ingress to be compiled in
1714 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1715 * a compare and 2 stores extra right now if we dont have it on
1716 * but have CONFIG_NET_CLS_ACT
1717 * NOTE: This doesnt stop any functionality; if you dont have
1718 * the ingress scheduler, you just cant add policies on ingress.
1721 static int ing_filter(struct sk_buff *skb)
1724 struct net_device *dev = skb->dev;
1725 int result = TC_ACT_OK;
1727 if (dev->qdisc_ingress) {
1728 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1729 if (MAX_RED_LOOP < ttl++) {
1730 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1731 skb->iif, skb->dev->ifindex);
1735 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1737 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1739 spin_lock(&dev->queue_lock);
1740 if ((q = dev->qdisc_ingress) != NULL)
1741 result = q->enqueue(skb, q);
1742 spin_unlock(&dev->queue_lock);
1750 int netif_receive_skb(struct sk_buff *skb)
1752 struct packet_type *ptype, *pt_prev;
1753 struct net_device *orig_dev;
1754 int ret = NET_RX_DROP;
1757 /* if we've gotten here through NAPI, check netpoll */
1758 if (skb->dev->poll && netpoll_rx(skb))
1761 if (!skb->tstamp.tv64)
1765 skb->iif = skb->dev->ifindex;
1767 orig_dev = skb_bond(skb);
1772 __get_cpu_var(netdev_rx_stat).total++;
1774 skb_reset_network_header(skb);
1775 skb->h.raw = skb->data;
1776 skb->mac_len = skb->nh.raw - skb->mac.raw;
1782 #ifdef CONFIG_NET_CLS_ACT
1783 if (skb->tc_verd & TC_NCLS) {
1784 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1789 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1790 if (!ptype->dev || ptype->dev == skb->dev) {
1792 ret = deliver_skb(skb, pt_prev, orig_dev);
1797 #ifdef CONFIG_NET_CLS_ACT
1799 ret = deliver_skb(skb, pt_prev, orig_dev);
1800 pt_prev = NULL; /* noone else should process this after*/
1802 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1805 ret = ing_filter(skb);
1807 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1816 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1819 type = skb->protocol;
1820 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1821 if (ptype->type == type &&
1822 (!ptype->dev || ptype->dev == skb->dev)) {
1824 ret = deliver_skb(skb, pt_prev, orig_dev);
1830 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1833 /* Jamal, now you will not able to escape explaining
1834 * me how you were going to use this. :-)
1844 static int process_backlog(struct net_device *backlog_dev, int *budget)
1847 int quota = min(backlog_dev->quota, *budget);
1848 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1849 unsigned long start_time = jiffies;
1851 backlog_dev->weight = weight_p;
1853 struct sk_buff *skb;
1854 struct net_device *dev;
1856 local_irq_disable();
1857 skb = __skb_dequeue(&queue->input_pkt_queue);
1864 netif_receive_skb(skb);
1870 if (work >= quota || jiffies - start_time > 1)
1875 backlog_dev->quota -= work;
1880 backlog_dev->quota -= work;
1883 list_del(&backlog_dev->poll_list);
1884 smp_mb__before_clear_bit();
1885 netif_poll_enable(backlog_dev);
1891 static void net_rx_action(struct softirq_action *h)
1893 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1894 unsigned long start_time = jiffies;
1895 int budget = netdev_budget;
1898 local_irq_disable();
1900 while (!list_empty(&queue->poll_list)) {
1901 struct net_device *dev;
1903 if (budget <= 0 || jiffies - start_time > 1)
1908 dev = list_entry(queue->poll_list.next,
1909 struct net_device, poll_list);
1910 have = netpoll_poll_lock(dev);
1912 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1913 netpoll_poll_unlock(have);
1914 local_irq_disable();
1915 list_move_tail(&dev->poll_list, &queue->poll_list);
1917 dev->quota += dev->weight;
1919 dev->quota = dev->weight;
1921 netpoll_poll_unlock(have);
1923 local_irq_disable();
1927 #ifdef CONFIG_NET_DMA
1929 * There may not be any more sk_buffs coming right now, so push
1930 * any pending DMA copies to hardware
1932 if (net_dma_client) {
1933 struct dma_chan *chan;
1935 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1936 dma_async_memcpy_issue_pending(chan);
1944 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1945 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1949 static gifconf_func_t * gifconf_list [NPROTO];
1952 * register_gifconf - register a SIOCGIF handler
1953 * @family: Address family
1954 * @gifconf: Function handler
1956 * Register protocol dependent address dumping routines. The handler
1957 * that is passed must not be freed or reused until it has been replaced
1958 * by another handler.
1960 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1962 if (family >= NPROTO)
1964 gifconf_list[family] = gifconf;
1970 * Map an interface index to its name (SIOCGIFNAME)
1974 * We need this ioctl for efficient implementation of the
1975 * if_indextoname() function required by the IPv6 API. Without
1976 * it, we would have to search all the interfaces to find a
1980 static int dev_ifname(struct ifreq __user *arg)
1982 struct net_device *dev;
1986 * Fetch the caller's info block.
1989 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1992 read_lock(&dev_base_lock);
1993 dev = __dev_get_by_index(ifr.ifr_ifindex);
1995 read_unlock(&dev_base_lock);
1999 strcpy(ifr.ifr_name, dev->name);
2000 read_unlock(&dev_base_lock);
2002 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2008 * Perform a SIOCGIFCONF call. This structure will change
2009 * size eventually, and there is nothing I can do about it.
2010 * Thus we will need a 'compatibility mode'.
2013 static int dev_ifconf(char __user *arg)
2016 struct net_device *dev;
2023 * Fetch the caller's info block.
2026 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2033 * Loop over the interfaces, and write an info block for each.
2037 for (dev = dev_base; dev; dev = dev->next) {
2038 for (i = 0; i < NPROTO; i++) {
2039 if (gifconf_list[i]) {
2042 done = gifconf_list[i](dev, NULL, 0);
2044 done = gifconf_list[i](dev, pos + total,
2054 * All done. Write the updated control block back to the caller.
2056 ifc.ifc_len = total;
2059 * Both BSD and Solaris return 0 here, so we do too.
2061 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2064 #ifdef CONFIG_PROC_FS
2066 * This is invoked by the /proc filesystem handler to display a device
2069 static struct net_device *dev_get_idx(loff_t pos)
2071 struct net_device *dev;
2074 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2076 return i == pos ? dev : NULL;
2079 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2081 read_lock(&dev_base_lock);
2082 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2085 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2088 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2091 void dev_seq_stop(struct seq_file *seq, void *v)
2093 read_unlock(&dev_base_lock);
2096 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2098 if (dev->get_stats) {
2099 struct net_device_stats *stats = dev->get_stats(dev);
2101 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2102 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2103 dev->name, stats->rx_bytes, stats->rx_packets,
2105 stats->rx_dropped + stats->rx_missed_errors,
2106 stats->rx_fifo_errors,
2107 stats->rx_length_errors + stats->rx_over_errors +
2108 stats->rx_crc_errors + stats->rx_frame_errors,
2109 stats->rx_compressed, stats->multicast,
2110 stats->tx_bytes, stats->tx_packets,
2111 stats->tx_errors, stats->tx_dropped,
2112 stats->tx_fifo_errors, stats->collisions,
2113 stats->tx_carrier_errors +
2114 stats->tx_aborted_errors +
2115 stats->tx_window_errors +
2116 stats->tx_heartbeat_errors,
2117 stats->tx_compressed);
2119 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2123 * Called from the PROCfs module. This now uses the new arbitrary sized
2124 * /proc/net interface to create /proc/net/dev
2126 static int dev_seq_show(struct seq_file *seq, void *v)
2128 if (v == SEQ_START_TOKEN)
2129 seq_puts(seq, "Inter-| Receive "
2131 " face |bytes packets errs drop fifo frame "
2132 "compressed multicast|bytes packets errs "
2133 "drop fifo colls carrier compressed\n");
2135 dev_seq_printf_stats(seq, v);
2139 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2141 struct netif_rx_stats *rc = NULL;
2143 while (*pos < NR_CPUS)
2144 if (cpu_online(*pos)) {
2145 rc = &per_cpu(netdev_rx_stat, *pos);
2152 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2154 return softnet_get_online(pos);
2157 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2160 return softnet_get_online(pos);
2163 static void softnet_seq_stop(struct seq_file *seq, void *v)
2167 static int softnet_seq_show(struct seq_file *seq, void *v)
2169 struct netif_rx_stats *s = v;
2171 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2172 s->total, s->dropped, s->time_squeeze, 0,
2173 0, 0, 0, 0, /* was fastroute */
2178 static struct seq_operations dev_seq_ops = {
2179 .start = dev_seq_start,
2180 .next = dev_seq_next,
2181 .stop = dev_seq_stop,
2182 .show = dev_seq_show,
2185 static int dev_seq_open(struct inode *inode, struct file *file)
2187 return seq_open(file, &dev_seq_ops);
2190 static const struct file_operations dev_seq_fops = {
2191 .owner = THIS_MODULE,
2192 .open = dev_seq_open,
2194 .llseek = seq_lseek,
2195 .release = seq_release,
2198 static struct seq_operations softnet_seq_ops = {
2199 .start = softnet_seq_start,
2200 .next = softnet_seq_next,
2201 .stop = softnet_seq_stop,
2202 .show = softnet_seq_show,
2205 static int softnet_seq_open(struct inode *inode, struct file *file)
2207 return seq_open(file, &softnet_seq_ops);
2210 static const struct file_operations softnet_seq_fops = {
2211 .owner = THIS_MODULE,
2212 .open = softnet_seq_open,
2214 .llseek = seq_lseek,
2215 .release = seq_release,
2218 #ifdef CONFIG_WIRELESS_EXT
2219 extern int wireless_proc_init(void);
2221 #define wireless_proc_init() 0
2224 static int __init dev_proc_init(void)
2228 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2230 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2232 if (wireless_proc_init())
2238 proc_net_remove("softnet_stat");
2240 proc_net_remove("dev");
2244 #define dev_proc_init() 0
2245 #endif /* CONFIG_PROC_FS */
2249 * netdev_set_master - set up master/slave pair
2250 * @slave: slave device
2251 * @master: new master device
2253 * Changes the master device of the slave. Pass %NULL to break the
2254 * bonding. The caller must hold the RTNL semaphore. On a failure
2255 * a negative errno code is returned. On success the reference counts
2256 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2257 * function returns zero.
2259 int netdev_set_master(struct net_device *slave, struct net_device *master)
2261 struct net_device *old = slave->master;
2271 slave->master = master;
2279 slave->flags |= IFF_SLAVE;
2281 slave->flags &= ~IFF_SLAVE;
2283 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2288 * dev_set_promiscuity - update promiscuity count on a device
2292 * Add or remove promiscuity from a device. While the count in the device
2293 * remains above zero the interface remains promiscuous. Once it hits zero
2294 * the device reverts back to normal filtering operation. A negative inc
2295 * value is used to drop promiscuity on the device.
2297 void dev_set_promiscuity(struct net_device *dev, int inc)
2299 unsigned short old_flags = dev->flags;
2301 if ((dev->promiscuity += inc) == 0)
2302 dev->flags &= ~IFF_PROMISC;
2304 dev->flags |= IFF_PROMISC;
2305 if (dev->flags != old_flags) {
2307 printk(KERN_INFO "device %s %s promiscuous mode\n",
2308 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2310 audit_log(current->audit_context, GFP_ATOMIC,
2311 AUDIT_ANOM_PROMISCUOUS,
2312 "dev=%s prom=%d old_prom=%d auid=%u",
2313 dev->name, (dev->flags & IFF_PROMISC),
2314 (old_flags & IFF_PROMISC),
2315 audit_get_loginuid(current->audit_context));
2320 * dev_set_allmulti - update allmulti count on a device
2324 * Add or remove reception of all multicast frames to a device. While the
2325 * count in the device remains above zero the interface remains listening
2326 * to all interfaces. Once it hits zero the device reverts back to normal
2327 * filtering operation. A negative @inc value is used to drop the counter
2328 * when releasing a resource needing all multicasts.
2331 void dev_set_allmulti(struct net_device *dev, int inc)
2333 unsigned short old_flags = dev->flags;
2335 dev->flags |= IFF_ALLMULTI;
2336 if ((dev->allmulti += inc) == 0)
2337 dev->flags &= ~IFF_ALLMULTI;
2338 if (dev->flags ^ old_flags)
2342 unsigned dev_get_flags(const struct net_device *dev)
2346 flags = (dev->flags & ~(IFF_PROMISC |
2351 (dev->gflags & (IFF_PROMISC |
2354 if (netif_running(dev)) {
2355 if (netif_oper_up(dev))
2356 flags |= IFF_RUNNING;
2357 if (netif_carrier_ok(dev))
2358 flags |= IFF_LOWER_UP;
2359 if (netif_dormant(dev))
2360 flags |= IFF_DORMANT;
2366 int dev_change_flags(struct net_device *dev, unsigned flags)
2369 int old_flags = dev->flags;
2372 * Set the flags on our device.
2375 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2376 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2378 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2382 * Load in the correct multicast list now the flags have changed.
2388 * Have we downed the interface. We handle IFF_UP ourselves
2389 * according to user attempts to set it, rather than blindly
2394 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2395 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2401 if (dev->flags & IFF_UP &&
2402 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2404 raw_notifier_call_chain(&netdev_chain,
2405 NETDEV_CHANGE, dev);
2407 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2408 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2409 dev->gflags ^= IFF_PROMISC;
2410 dev_set_promiscuity(dev, inc);
2413 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2414 is important. Some (broken) drivers set IFF_PROMISC, when
2415 IFF_ALLMULTI is requested not asking us and not reporting.
2417 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2418 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2419 dev->gflags ^= IFF_ALLMULTI;
2420 dev_set_allmulti(dev, inc);
2423 if (old_flags ^ dev->flags)
2424 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2429 int dev_set_mtu(struct net_device *dev, int new_mtu)
2433 if (new_mtu == dev->mtu)
2436 /* MTU must be positive. */
2440 if (!netif_device_present(dev))
2444 if (dev->change_mtu)
2445 err = dev->change_mtu(dev, new_mtu);
2448 if (!err && dev->flags & IFF_UP)
2449 raw_notifier_call_chain(&netdev_chain,
2450 NETDEV_CHANGEMTU, dev);
2454 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2458 if (!dev->set_mac_address)
2460 if (sa->sa_family != dev->type)
2462 if (!netif_device_present(dev))
2464 err = dev->set_mac_address(dev, sa);
2466 raw_notifier_call_chain(&netdev_chain,
2467 NETDEV_CHANGEADDR, dev);
2472 * Perform the SIOCxIFxxx calls.
2474 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2477 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2483 case SIOCGIFFLAGS: /* Get interface flags */
2484 ifr->ifr_flags = dev_get_flags(dev);
2487 case SIOCSIFFLAGS: /* Set interface flags */
2488 return dev_change_flags(dev, ifr->ifr_flags);
2490 case SIOCGIFMETRIC: /* Get the metric on the interface
2491 (currently unused) */
2492 ifr->ifr_metric = 0;
2495 case SIOCSIFMETRIC: /* Set the metric on the interface
2496 (currently unused) */
2499 case SIOCGIFMTU: /* Get the MTU of a device */
2500 ifr->ifr_mtu = dev->mtu;
2503 case SIOCSIFMTU: /* Set the MTU of a device */
2504 return dev_set_mtu(dev, ifr->ifr_mtu);
2508 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2510 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2511 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2512 ifr->ifr_hwaddr.sa_family = dev->type;
2516 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2518 case SIOCSIFHWBROADCAST:
2519 if (ifr->ifr_hwaddr.sa_family != dev->type)
2521 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2522 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2523 raw_notifier_call_chain(&netdev_chain,
2524 NETDEV_CHANGEADDR, dev);
2528 ifr->ifr_map.mem_start = dev->mem_start;
2529 ifr->ifr_map.mem_end = dev->mem_end;
2530 ifr->ifr_map.base_addr = dev->base_addr;
2531 ifr->ifr_map.irq = dev->irq;
2532 ifr->ifr_map.dma = dev->dma;
2533 ifr->ifr_map.port = dev->if_port;
2537 if (dev->set_config) {
2538 if (!netif_device_present(dev))
2540 return dev->set_config(dev, &ifr->ifr_map);
2545 if (!dev->set_multicast_list ||
2546 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2548 if (!netif_device_present(dev))
2550 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2554 if (!dev->set_multicast_list ||
2555 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2557 if (!netif_device_present(dev))
2559 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2563 ifr->ifr_ifindex = dev->ifindex;
2567 ifr->ifr_qlen = dev->tx_queue_len;
2571 if (ifr->ifr_qlen < 0)
2573 dev->tx_queue_len = ifr->ifr_qlen;
2577 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2578 return dev_change_name(dev, ifr->ifr_newname);
2581 * Unknown or private ioctl
2585 if ((cmd >= SIOCDEVPRIVATE &&
2586 cmd <= SIOCDEVPRIVATE + 15) ||
2587 cmd == SIOCBONDENSLAVE ||
2588 cmd == SIOCBONDRELEASE ||
2589 cmd == SIOCBONDSETHWADDR ||
2590 cmd == SIOCBONDSLAVEINFOQUERY ||
2591 cmd == SIOCBONDINFOQUERY ||
2592 cmd == SIOCBONDCHANGEACTIVE ||
2593 cmd == SIOCGMIIPHY ||
2594 cmd == SIOCGMIIREG ||
2595 cmd == SIOCSMIIREG ||
2596 cmd == SIOCBRADDIF ||
2597 cmd == SIOCBRDELIF ||
2598 cmd == SIOCWANDEV) {
2600 if (dev->do_ioctl) {
2601 if (netif_device_present(dev))
2602 err = dev->do_ioctl(dev, ifr,
2615 * This function handles all "interface"-type I/O control requests. The actual
2616 * 'doing' part of this is dev_ifsioc above.
2620 * dev_ioctl - network device ioctl
2621 * @cmd: command to issue
2622 * @arg: pointer to a struct ifreq in user space
2624 * Issue ioctl functions to devices. This is normally called by the
2625 * user space syscall interfaces but can sometimes be useful for
2626 * other purposes. The return value is the return from the syscall if
2627 * positive or a negative errno code on error.
2630 int dev_ioctl(unsigned int cmd, void __user *arg)
2636 /* One special case: SIOCGIFCONF takes ifconf argument
2637 and requires shared lock, because it sleeps writing
2641 if (cmd == SIOCGIFCONF) {
2643 ret = dev_ifconf((char __user *) arg);
2647 if (cmd == SIOCGIFNAME)
2648 return dev_ifname((struct ifreq __user *)arg);
2650 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2653 ifr.ifr_name[IFNAMSIZ-1] = 0;
2655 colon = strchr(ifr.ifr_name, ':');
2660 * See which interface the caller is talking about.
2665 * These ioctl calls:
2666 * - can be done by all.
2667 * - atomic and do not require locking.
2678 dev_load(ifr.ifr_name);
2679 read_lock(&dev_base_lock);
2680 ret = dev_ifsioc(&ifr, cmd);
2681 read_unlock(&dev_base_lock);
2685 if (copy_to_user(arg, &ifr,
2686 sizeof(struct ifreq)))
2692 dev_load(ifr.ifr_name);
2694 ret = dev_ethtool(&ifr);
2699 if (copy_to_user(arg, &ifr,
2700 sizeof(struct ifreq)))
2706 * These ioctl calls:
2707 * - require superuser power.
2708 * - require strict serialization.
2714 if (!capable(CAP_NET_ADMIN))
2716 dev_load(ifr.ifr_name);
2718 ret = dev_ifsioc(&ifr, cmd);
2723 if (copy_to_user(arg, &ifr,
2724 sizeof(struct ifreq)))
2730 * These ioctl calls:
2731 * - require superuser power.
2732 * - require strict serialization.
2733 * - do not return a value
2743 case SIOCSIFHWBROADCAST:
2746 case SIOCBONDENSLAVE:
2747 case SIOCBONDRELEASE:
2748 case SIOCBONDSETHWADDR:
2749 case SIOCBONDCHANGEACTIVE:
2752 if (!capable(CAP_NET_ADMIN))
2755 case SIOCBONDSLAVEINFOQUERY:
2756 case SIOCBONDINFOQUERY:
2757 dev_load(ifr.ifr_name);
2759 ret = dev_ifsioc(&ifr, cmd);
2764 /* Get the per device memory space. We can add this but
2765 * currently do not support it */
2767 /* Set the per device memory buffer space.
2768 * Not applicable in our case */
2773 * Unknown or private ioctl.
2776 if (cmd == SIOCWANDEV ||
2777 (cmd >= SIOCDEVPRIVATE &&
2778 cmd <= SIOCDEVPRIVATE + 15)) {
2779 dev_load(ifr.ifr_name);
2781 ret = dev_ifsioc(&ifr, cmd);
2783 if (!ret && copy_to_user(arg, &ifr,
2784 sizeof(struct ifreq)))
2788 #ifdef CONFIG_WIRELESS_EXT
2789 /* Take care of Wireless Extensions */
2790 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2791 /* If command is `set a parameter', or
2792 * `get the encoding parameters', check if
2793 * the user has the right to do it */
2794 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2795 || cmd == SIOCGIWENCODEEXT) {
2796 if (!capable(CAP_NET_ADMIN))
2799 dev_load(ifr.ifr_name);
2801 /* Follow me in net/core/wireless.c */
2802 ret = wireless_process_ioctl(&ifr, cmd);
2804 if (IW_IS_GET(cmd) &&
2805 copy_to_user(arg, &ifr,
2806 sizeof(struct ifreq)))
2810 #endif /* CONFIG_WIRELESS_EXT */
2817 * dev_new_index - allocate an ifindex
2819 * Returns a suitable unique value for a new device interface
2820 * number. The caller must hold the rtnl semaphore or the
2821 * dev_base_lock to be sure it remains unique.
2823 static int dev_new_index(void)
2829 if (!__dev_get_by_index(ifindex))
2834 static int dev_boot_phase = 1;
2836 /* Delayed registration/unregisteration */
2837 static DEFINE_SPINLOCK(net_todo_list_lock);
2838 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2840 static void net_set_todo(struct net_device *dev)
2842 spin_lock(&net_todo_list_lock);
2843 list_add_tail(&dev->todo_list, &net_todo_list);
2844 spin_unlock(&net_todo_list_lock);
2848 * register_netdevice - register a network device
2849 * @dev: device to register
2851 * Take a completed network device structure and add it to the kernel
2852 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2853 * chain. 0 is returned on success. A negative errno code is returned
2854 * on a failure to set up the device, or if the name is a duplicate.
2856 * Callers must hold the rtnl semaphore. You may want
2857 * register_netdev() instead of this.
2860 * The locking appears insufficient to guarantee two parallel registers
2861 * will not get the same name.
2864 int register_netdevice(struct net_device *dev)
2866 struct hlist_head *head;
2867 struct hlist_node *p;
2870 BUG_ON(dev_boot_phase);
2875 /* When net_device's are persistent, this will be fatal. */
2876 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2878 spin_lock_init(&dev->queue_lock);
2879 spin_lock_init(&dev->_xmit_lock);
2880 dev->xmit_lock_owner = -1;
2881 #ifdef CONFIG_NET_CLS_ACT
2882 spin_lock_init(&dev->ingress_lock);
2887 /* Init, if this function is available */
2889 ret = dev->init(dev);
2897 if (!dev_valid_name(dev->name)) {
2902 dev->ifindex = dev_new_index();
2903 if (dev->iflink == -1)
2904 dev->iflink = dev->ifindex;
2906 /* Check for existence of name */
2907 head = dev_name_hash(dev->name);
2908 hlist_for_each(p, head) {
2909 struct net_device *d
2910 = hlist_entry(p, struct net_device, name_hlist);
2911 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2917 /* Fix illegal SG+CSUM combinations. */
2918 if ((dev->features & NETIF_F_SG) &&
2919 !(dev->features & NETIF_F_ALL_CSUM)) {
2920 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
2922 dev->features &= ~NETIF_F_SG;
2925 /* TSO requires that SG is present as well. */
2926 if ((dev->features & NETIF_F_TSO) &&
2927 !(dev->features & NETIF_F_SG)) {
2928 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
2930 dev->features &= ~NETIF_F_TSO;
2932 if (dev->features & NETIF_F_UFO) {
2933 if (!(dev->features & NETIF_F_HW_CSUM)) {
2934 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2935 "NETIF_F_HW_CSUM feature.\n",
2937 dev->features &= ~NETIF_F_UFO;
2939 if (!(dev->features & NETIF_F_SG)) {
2940 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2941 "NETIF_F_SG feature.\n",
2943 dev->features &= ~NETIF_F_UFO;
2948 * nil rebuild_header routine,
2949 * that should be never called and used as just bug trap.
2952 if (!dev->rebuild_header)
2953 dev->rebuild_header = default_rebuild_header;
2955 ret = netdev_register_sysfs(dev);
2958 dev->reg_state = NETREG_REGISTERED;
2961 * Default initial state at registry is that the
2962 * device is present.
2965 set_bit(__LINK_STATE_PRESENT, &dev->state);
2968 dev_init_scheduler(dev);
2969 write_lock_bh(&dev_base_lock);
2971 dev_tail = &dev->next;
2972 hlist_add_head(&dev->name_hlist, head);
2973 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2975 write_unlock_bh(&dev_base_lock);
2977 /* Notify protocols, that a new device appeared. */
2978 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2987 * register_netdev - register a network device
2988 * @dev: device to register
2990 * Take a completed network device structure and add it to the kernel
2991 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2992 * chain. 0 is returned on success. A negative errno code is returned
2993 * on a failure to set up the device, or if the name is a duplicate.
2995 * This is a wrapper around register_netdev that takes the rtnl semaphore
2996 * and expands the device name if you passed a format string to
2999 int register_netdev(struct net_device *dev)
3006 * If the name is a format string the caller wants us to do a
3009 if (strchr(dev->name, '%')) {
3010 err = dev_alloc_name(dev, dev->name);
3015 err = register_netdevice(dev);
3020 EXPORT_SYMBOL(register_netdev);
3023 * netdev_wait_allrefs - wait until all references are gone.
3025 * This is called when unregistering network devices.
3027 * Any protocol or device that holds a reference should register
3028 * for netdevice notification, and cleanup and put back the
3029 * reference if they receive an UNREGISTER event.
3030 * We can get stuck here if buggy protocols don't correctly
3033 static void netdev_wait_allrefs(struct net_device *dev)
3035 unsigned long rebroadcast_time, warning_time;
3037 rebroadcast_time = warning_time = jiffies;
3038 while (atomic_read(&dev->refcnt) != 0) {
3039 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3042 /* Rebroadcast unregister notification */
3043 raw_notifier_call_chain(&netdev_chain,
3044 NETDEV_UNREGISTER, dev);
3046 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3048 /* We must not have linkwatch events
3049 * pending on unregister. If this
3050 * happens, we simply run the queue
3051 * unscheduled, resulting in a noop
3054 linkwatch_run_queue();
3059 rebroadcast_time = jiffies;
3064 if (time_after(jiffies, warning_time + 10 * HZ)) {
3065 printk(KERN_EMERG "unregister_netdevice: "
3066 "waiting for %s to become free. Usage "
3068 dev->name, atomic_read(&dev->refcnt));
3069 warning_time = jiffies;
3078 * register_netdevice(x1);
3079 * register_netdevice(x2);
3081 * unregister_netdevice(y1);
3082 * unregister_netdevice(y2);
3088 * We are invoked by rtnl_unlock() after it drops the semaphore.
3089 * This allows us to deal with problems:
3090 * 1) We can delete sysfs objects which invoke hotplug
3091 * without deadlocking with linkwatch via keventd.
3092 * 2) Since we run with the RTNL semaphore not held, we can sleep
3093 * safely in order to wait for the netdev refcnt to drop to zero.
3095 static DEFINE_MUTEX(net_todo_run_mutex);
3096 void netdev_run_todo(void)
3098 struct list_head list;
3100 /* Need to guard against multiple cpu's getting out of order. */
3101 mutex_lock(&net_todo_run_mutex);
3103 /* Not safe to do outside the semaphore. We must not return
3104 * until all unregister events invoked by the local processor
3105 * have been completed (either by this todo run, or one on
3108 if (list_empty(&net_todo_list))
3111 /* Snapshot list, allow later requests */
3112 spin_lock(&net_todo_list_lock);
3113 list_replace_init(&net_todo_list, &list);
3114 spin_unlock(&net_todo_list_lock);
3116 while (!list_empty(&list)) {
3117 struct net_device *dev
3118 = list_entry(list.next, struct net_device, todo_list);
3119 list_del(&dev->todo_list);
3121 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3122 printk(KERN_ERR "network todo '%s' but state %d\n",
3123 dev->name, dev->reg_state);
3128 netdev_unregister_sysfs(dev);
3129 dev->reg_state = NETREG_UNREGISTERED;
3131 netdev_wait_allrefs(dev);
3134 BUG_ON(atomic_read(&dev->refcnt));
3135 BUG_TRAP(!dev->ip_ptr);
3136 BUG_TRAP(!dev->ip6_ptr);
3137 BUG_TRAP(!dev->dn_ptr);
3139 /* It must be the very last action,
3140 * after this 'dev' may point to freed up memory.
3142 if (dev->destructor)
3143 dev->destructor(dev);
3147 mutex_unlock(&net_todo_run_mutex);
3151 * alloc_netdev - allocate network device
3152 * @sizeof_priv: size of private data to allocate space for
3153 * @name: device name format string
3154 * @setup: callback to initialize device
3156 * Allocates a struct net_device with private data area for driver use
3157 * and performs basic initialization.
3159 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3160 void (*setup)(struct net_device *))
3163 struct net_device *dev;
3166 BUG_ON(strlen(name) >= sizeof(dev->name));
3168 /* ensure 32-byte alignment of both the device and private area */
3169 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3170 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3172 p = kzalloc(alloc_size, GFP_KERNEL);
3174 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3178 dev = (struct net_device *)
3179 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3180 dev->padded = (char *)dev - (char *)p;
3183 dev->priv = netdev_priv(dev);
3186 strcpy(dev->name, name);
3189 EXPORT_SYMBOL(alloc_netdev);
3192 * free_netdev - free network device
3195 * This function does the last stage of destroying an allocated device
3196 * interface. The reference to the device object is released.
3197 * If this is the last reference then it will be freed.
3199 void free_netdev(struct net_device *dev)
3202 /* Compatibility with error handling in drivers */
3203 if (dev->reg_state == NETREG_UNINITIALIZED) {
3204 kfree((char *)dev - dev->padded);
3208 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3209 dev->reg_state = NETREG_RELEASED;
3211 /* will free via device release */
3212 put_device(&dev->dev);
3214 kfree((char *)dev - dev->padded);
3218 /* Synchronize with packet receive processing. */
3219 void synchronize_net(void)
3226 * unregister_netdevice - remove device from the kernel
3229 * This function shuts down a device interface and removes it
3230 * from the kernel tables. On success 0 is returned, on a failure
3231 * a negative errno code is returned.
3233 * Callers must hold the rtnl semaphore. You may want
3234 * unregister_netdev() instead of this.
3237 void unregister_netdevice(struct net_device *dev)
3239 struct net_device *d, **dp;
3241 BUG_ON(dev_boot_phase);
3244 /* Some devices call without registering for initialization unwind. */
3245 if (dev->reg_state == NETREG_UNINITIALIZED) {
3246 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3247 "was registered\n", dev->name, dev);
3253 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3255 /* If device is running, close it first. */
3256 if (dev->flags & IFF_UP)
3259 /* And unlink it from device chain. */
3260 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3262 write_lock_bh(&dev_base_lock);
3263 hlist_del(&dev->name_hlist);
3264 hlist_del(&dev->index_hlist);
3265 if (dev_tail == &dev->next)
3268 write_unlock_bh(&dev_base_lock);
3274 dev->reg_state = NETREG_UNREGISTERING;
3278 /* Shutdown queueing discipline. */
3282 /* Notify protocols, that we are about to destroy
3283 this device. They should clean all the things.
3285 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3288 * Flush the multicast chain
3290 dev_mc_discard(dev);
3295 /* Notifier chain MUST detach us from master device. */
3296 BUG_TRAP(!dev->master);
3298 /* Finish processing unregister after unlock */
3307 * unregister_netdev - remove device from the kernel
3310 * This function shuts down a device interface and removes it
3311 * from the kernel tables. On success 0 is returned, on a failure
3312 * a negative errno code is returned.
3314 * This is just a wrapper for unregister_netdevice that takes
3315 * the rtnl semaphore. In general you want to use this and not
3316 * unregister_netdevice.
3318 void unregister_netdev(struct net_device *dev)
3321 unregister_netdevice(dev);
3325 EXPORT_SYMBOL(unregister_netdev);
3327 static int dev_cpu_callback(struct notifier_block *nfb,
3328 unsigned long action,
3331 struct sk_buff **list_skb;
3332 struct net_device **list_net;
3333 struct sk_buff *skb;
3334 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3335 struct softnet_data *sd, *oldsd;
3337 if (action != CPU_DEAD)
3340 local_irq_disable();
3341 cpu = smp_processor_id();
3342 sd = &per_cpu(softnet_data, cpu);
3343 oldsd = &per_cpu(softnet_data, oldcpu);
3345 /* Find end of our completion_queue. */
3346 list_skb = &sd->completion_queue;
3348 list_skb = &(*list_skb)->next;
3349 /* Append completion queue from offline CPU. */
3350 *list_skb = oldsd->completion_queue;
3351 oldsd->completion_queue = NULL;
3353 /* Find end of our output_queue. */
3354 list_net = &sd->output_queue;
3356 list_net = &(*list_net)->next_sched;
3357 /* Append output queue from offline CPU. */
3358 *list_net = oldsd->output_queue;
3359 oldsd->output_queue = NULL;
3361 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3364 /* Process offline CPU's input_pkt_queue */
3365 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3371 #ifdef CONFIG_NET_DMA
3373 * net_dma_rebalance -
3374 * This is called when the number of channels allocated to the net_dma_client
3375 * changes. The net_dma_client tries to have one DMA channel per CPU.
3377 static void net_dma_rebalance(void)
3379 unsigned int cpu, i, n;
3380 struct dma_chan *chan;
3382 if (net_dma_count == 0) {
3383 for_each_online_cpu(cpu)
3384 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
3389 cpu = first_cpu(cpu_online_map);
3392 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3393 n = ((num_online_cpus() / net_dma_count)
3394 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3397 per_cpu(softnet_data, cpu).net_dma = chan;
3398 cpu = next_cpu(cpu, cpu_online_map);
3407 * netdev_dma_event - event callback for the net_dma_client
3408 * @client: should always be net_dma_client
3409 * @chan: DMA channel for the event
3410 * @event: event type
3412 static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3413 enum dma_event event)
3415 spin_lock(&net_dma_event_lock);
3417 case DMA_RESOURCE_ADDED:
3419 net_dma_rebalance();
3421 case DMA_RESOURCE_REMOVED:
3423 net_dma_rebalance();
3428 spin_unlock(&net_dma_event_lock);
3432 * netdev_dma_regiser - register the networking subsystem as a DMA client
3434 static int __init netdev_dma_register(void)
3436 spin_lock_init(&net_dma_event_lock);
3437 net_dma_client = dma_async_client_register(netdev_dma_event);
3438 if (net_dma_client == NULL)
3441 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3446 static int __init netdev_dma_register(void) { return -ENODEV; }
3447 #endif /* CONFIG_NET_DMA */
3450 * Initialize the DEV module. At boot time this walks the device list and
3451 * unhooks any devices that fail to initialise (normally hardware not
3452 * present) and leaves us with a valid list of present and active devices.
3457 * This is called single threaded during boot, so no need
3458 * to take the rtnl semaphore.
3460 static int __init net_dev_init(void)
3462 int i, rc = -ENOMEM;
3464 BUG_ON(!dev_boot_phase);
3466 if (dev_proc_init())
3469 if (netdev_sysfs_init())
3472 INIT_LIST_HEAD(&ptype_all);
3473 for (i = 0; i < 16; i++)
3474 INIT_LIST_HEAD(&ptype_base[i]);
3476 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3477 INIT_HLIST_HEAD(&dev_name_head[i]);
3479 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3480 INIT_HLIST_HEAD(&dev_index_head[i]);
3483 * Initialise the packet receive queues.
3486 for_each_possible_cpu(i) {
3487 struct softnet_data *queue;
3489 queue = &per_cpu(softnet_data, i);
3490 skb_queue_head_init(&queue->input_pkt_queue);
3491 queue->completion_queue = NULL;
3492 INIT_LIST_HEAD(&queue->poll_list);
3493 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3494 queue->backlog_dev.weight = weight_p;
3495 queue->backlog_dev.poll = process_backlog;
3496 atomic_set(&queue->backlog_dev.refcnt, 1);
3499 netdev_dma_register();
3503 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3504 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3506 hotcpu_notifier(dev_cpu_callback, 0);
3514 subsys_initcall(net_dev_init);
3516 EXPORT_SYMBOL(__dev_get_by_index);
3517 EXPORT_SYMBOL(__dev_get_by_name);
3518 EXPORT_SYMBOL(__dev_remove_pack);
3519 EXPORT_SYMBOL(dev_valid_name);
3520 EXPORT_SYMBOL(dev_add_pack);
3521 EXPORT_SYMBOL(dev_alloc_name);
3522 EXPORT_SYMBOL(dev_close);
3523 EXPORT_SYMBOL(dev_get_by_flags);
3524 EXPORT_SYMBOL(dev_get_by_index);
3525 EXPORT_SYMBOL(dev_get_by_name);
3526 EXPORT_SYMBOL(dev_open);
3527 EXPORT_SYMBOL(dev_queue_xmit);
3528 EXPORT_SYMBOL(dev_remove_pack);
3529 EXPORT_SYMBOL(dev_set_allmulti);
3530 EXPORT_SYMBOL(dev_set_promiscuity);
3531 EXPORT_SYMBOL(dev_change_flags);
3532 EXPORT_SYMBOL(dev_set_mtu);
3533 EXPORT_SYMBOL(dev_set_mac_address);
3534 EXPORT_SYMBOL(free_netdev);
3535 EXPORT_SYMBOL(netdev_boot_setup_check);
3536 EXPORT_SYMBOL(netdev_set_master);
3537 EXPORT_SYMBOL(netdev_state_change);
3538 EXPORT_SYMBOL(netif_receive_skb);
3539 EXPORT_SYMBOL(netif_rx);
3540 EXPORT_SYMBOL(register_gifconf);
3541 EXPORT_SYMBOL(register_netdevice);
3542 EXPORT_SYMBOL(register_netdevice_notifier);
3543 EXPORT_SYMBOL(skb_checksum_help);
3544 EXPORT_SYMBOL(synchronize_net);
3545 EXPORT_SYMBOL(unregister_netdevice);
3546 EXPORT_SYMBOL(unregister_netdevice_notifier);
3547 EXPORT_SYMBOL(net_enable_timestamp);
3548 EXPORT_SYMBOL(net_disable_timestamp);
3549 EXPORT_SYMBOL(dev_get_flags);
3551 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3552 EXPORT_SYMBOL(br_handle_frame_hook);
3553 EXPORT_SYMBOL(br_fdb_get_hook);
3554 EXPORT_SYMBOL(br_fdb_put_hook);
3558 EXPORT_SYMBOL(dev_load);
3561 EXPORT_PER_CPU_SYMBOL(softnet_data);