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/config.h>
80 #include <linux/cpu.h>
81 #include <linux/types.h>
82 #include <linux/kernel.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
97 #include <linux/rtnetlink.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <linux/stat.h>
101 #include <linux/if_bridge.h>
102 #include <linux/divert.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/kmod.h>
109 #include <linux/module.h>
110 #include <linux/kallsyms.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <linux/wireless.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
122 * The list of packet types we will receive (as opposed to discard)
123 * and the routines to invoke.
125 * Why 16. Because with 16 the only overlap we get on a hash of the
126 * low nibble of the protocol value is RARP/SNAP/X.25.
128 * NOTE: That is no longer true with the addition of VLAN tags. Not
129 * sure which should go first, but I bet it won't make much
130 * difference if we are running VLANs. The good news is that
131 * this protocol won't be in the list unless compiled in, so
132 * the average user (w/out VLANs) will not be adversely affected.
149 static DEFINE_SPINLOCK(ptype_lock);
150 static struct list_head ptype_base[16]; /* 16 way hashed list */
151 static struct list_head ptype_all; /* Taps */
153 #ifdef CONFIG_NET_DMA
154 static struct dma_client *net_dma_client;
155 static unsigned int net_dma_count;
156 static spinlock_t net_dma_event_lock;
160 * The @dev_base list is protected by @dev_base_lock and the rtnl
163 * Pure readers hold dev_base_lock for reading.
165 * Writers must hold the rtnl semaphore while they loop through the
166 * dev_base list, and hold dev_base_lock for writing when they do the
167 * actual updates. This allows pure readers to access the list even
168 * while a writer is preparing to update it.
170 * To put it another way, dev_base_lock is held for writing only to
171 * protect against pure readers; the rtnl semaphore provides the
172 * protection against other writers.
174 * See, for example usages, register_netdevice() and
175 * unregister_netdevice(), which must be called with the rtnl
178 struct net_device *dev_base;
179 static struct net_device **dev_tail = &dev_base;
180 DEFINE_RWLOCK(dev_base_lock);
182 EXPORT_SYMBOL(dev_base);
183 EXPORT_SYMBOL(dev_base_lock);
185 #define NETDEV_HASHBITS 8
186 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
187 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
189 static inline struct hlist_head *dev_name_hash(const char *name)
191 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
192 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
195 static inline struct hlist_head *dev_index_hash(int ifindex)
197 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
204 static RAW_NOTIFIER_HEAD(netdev_chain);
207 * Device drivers call our routines to queue packets here. We empty the
208 * queue in the local softnet handler.
210 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
213 extern int netdev_sysfs_init(void);
214 extern int netdev_register_sysfs(struct net_device *);
215 extern void netdev_unregister_sysfs(struct net_device *);
217 #define netdev_sysfs_init() (0)
218 #define netdev_register_sysfs(dev) (0)
219 #define netdev_unregister_sysfs(dev) do { } while(0)
223 /*******************************************************************************
225 Protocol management and registration routines
227 *******************************************************************************/
233 static int netdev_nit;
236 * Add a protocol ID to the list. Now that the input handler is
237 * smarter we can dispense with all the messy stuff that used to be
240 * BEWARE!!! Protocol handlers, mangling input packets,
241 * MUST BE last in hash buckets and checking protocol handlers
242 * MUST start from promiscuous ptype_all chain in net_bh.
243 * It is true now, do not change it.
244 * Explanation follows: if protocol handler, mangling packet, will
245 * be the first on list, it is not able to sense, that packet
246 * is cloned and should be copied-on-write, so that it will
247 * change it and subsequent readers will get broken packet.
252 * dev_add_pack - add packet handler
253 * @pt: packet type declaration
255 * Add a protocol handler to the networking stack. The passed &packet_type
256 * is linked into kernel lists and may not be freed until it has been
257 * removed from the kernel lists.
259 * This call does not sleep therefore it can not
260 * guarantee all CPU's that are in middle of receiving packets
261 * will see the new packet type (until the next received packet).
264 void dev_add_pack(struct packet_type *pt)
268 spin_lock_bh(&ptype_lock);
269 if (pt->type == htons(ETH_P_ALL)) {
271 list_add_rcu(&pt->list, &ptype_all);
273 hash = ntohs(pt->type) & 15;
274 list_add_rcu(&pt->list, &ptype_base[hash]);
276 spin_unlock_bh(&ptype_lock);
280 * __dev_remove_pack - remove packet handler
281 * @pt: packet type declaration
283 * Remove a protocol handler that was previously added to the kernel
284 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
285 * from the kernel lists and can be freed or reused once this function
288 * The packet type might still be in use by receivers
289 * and must not be freed until after all the CPU's have gone
290 * through a quiescent state.
292 void __dev_remove_pack(struct packet_type *pt)
294 struct list_head *head;
295 struct packet_type *pt1;
297 spin_lock_bh(&ptype_lock);
299 if (pt->type == htons(ETH_P_ALL)) {
303 head = &ptype_base[ntohs(pt->type) & 15];
305 list_for_each_entry(pt1, head, list) {
307 list_del_rcu(&pt->list);
312 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
314 spin_unlock_bh(&ptype_lock);
317 * dev_remove_pack - remove packet handler
318 * @pt: packet type declaration
320 * Remove a protocol handler that was previously added to the kernel
321 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
322 * from the kernel lists and can be freed or reused once this function
325 * This call sleeps to guarantee that no CPU is looking at the packet
328 void dev_remove_pack(struct packet_type *pt)
330 __dev_remove_pack(pt);
335 /******************************************************************************
337 Device Boot-time Settings Routines
339 *******************************************************************************/
341 /* Boot time configuration table */
342 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
345 * netdev_boot_setup_add - add new setup entry
346 * @name: name of the device
347 * @map: configured settings for the device
349 * Adds new setup entry to the dev_boot_setup list. The function
350 * returns 0 on error and 1 on success. This is a generic routine to
353 static int netdev_boot_setup_add(char *name, struct ifmap *map)
355 struct netdev_boot_setup *s;
359 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
360 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
361 memset(s[i].name, 0, sizeof(s[i].name));
362 strcpy(s[i].name, name);
363 memcpy(&s[i].map, map, sizeof(s[i].map));
368 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
372 * netdev_boot_setup_check - check boot time settings
373 * @dev: the netdevice
375 * Check boot time settings for the device.
376 * The found settings are set for the device to be used
377 * later in the device probing.
378 * Returns 0 if no settings found, 1 if they are.
380 int netdev_boot_setup_check(struct net_device *dev)
382 struct netdev_boot_setup *s = dev_boot_setup;
385 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
386 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
387 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
388 dev->irq = s[i].map.irq;
389 dev->base_addr = s[i].map.base_addr;
390 dev->mem_start = s[i].map.mem_start;
391 dev->mem_end = s[i].map.mem_end;
400 * netdev_boot_base - get address from boot time settings
401 * @prefix: prefix for network device
402 * @unit: id for network device
404 * Check boot time settings for the base address of device.
405 * The found settings are set for the device to be used
406 * later in the device probing.
407 * Returns 0 if no settings found.
409 unsigned long netdev_boot_base(const char *prefix, int unit)
411 const struct netdev_boot_setup *s = dev_boot_setup;
415 sprintf(name, "%s%d", prefix, unit);
418 * If device already registered then return base of 1
419 * to indicate not to probe for this interface
421 if (__dev_get_by_name(name))
424 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
425 if (!strcmp(name, s[i].name))
426 return s[i].map.base_addr;
431 * Saves at boot time configured settings for any netdevice.
433 int __init netdev_boot_setup(char *str)
438 str = get_options(str, ARRAY_SIZE(ints), ints);
443 memset(&map, 0, sizeof(map));
447 map.base_addr = ints[2];
449 map.mem_start = ints[3];
451 map.mem_end = ints[4];
453 /* Add new entry to the list */
454 return netdev_boot_setup_add(str, &map);
457 __setup("netdev=", netdev_boot_setup);
459 /*******************************************************************************
461 Device Interface Subroutines
463 *******************************************************************************/
466 * __dev_get_by_name - find a device by its name
467 * @name: name to find
469 * Find an interface by name. Must be called under RTNL semaphore
470 * or @dev_base_lock. If the name is found a pointer to the device
471 * is returned. If the name is not found then %NULL is returned. The
472 * reference counters are not incremented so the caller must be
473 * careful with locks.
476 struct net_device *__dev_get_by_name(const char *name)
478 struct hlist_node *p;
480 hlist_for_each(p, dev_name_hash(name)) {
481 struct net_device *dev
482 = hlist_entry(p, struct net_device, name_hlist);
483 if (!strncmp(dev->name, name, IFNAMSIZ))
490 * dev_get_by_name - find a device by its name
491 * @name: name to find
493 * Find an interface by name. This can be called from any
494 * context and does its own locking. The returned handle has
495 * the usage count incremented and the caller must use dev_put() to
496 * release it when it is no longer needed. %NULL is returned if no
497 * matching device is found.
500 struct net_device *dev_get_by_name(const char *name)
502 struct net_device *dev;
504 read_lock(&dev_base_lock);
505 dev = __dev_get_by_name(name);
508 read_unlock(&dev_base_lock);
513 * __dev_get_by_index - find a device by its ifindex
514 * @ifindex: index of device
516 * Search for an interface by index. Returns %NULL if the device
517 * is not found or a pointer to the device. The device has not
518 * had its reference counter increased so the caller must be careful
519 * about locking. The caller must hold either the RTNL semaphore
523 struct net_device *__dev_get_by_index(int ifindex)
525 struct hlist_node *p;
527 hlist_for_each(p, dev_index_hash(ifindex)) {
528 struct net_device *dev
529 = hlist_entry(p, struct net_device, index_hlist);
530 if (dev->ifindex == ifindex)
538 * dev_get_by_index - find a device by its ifindex
539 * @ifindex: index of device
541 * Search for an interface by index. Returns NULL if the device
542 * is not found or a pointer to the device. The device returned has
543 * had a reference added and the pointer is safe until the user calls
544 * dev_put to indicate they have finished with it.
547 struct net_device *dev_get_by_index(int ifindex)
549 struct net_device *dev;
551 read_lock(&dev_base_lock);
552 dev = __dev_get_by_index(ifindex);
555 read_unlock(&dev_base_lock);
560 * dev_getbyhwaddr - find a device by its hardware address
561 * @type: media type of device
562 * @ha: hardware address
564 * Search for an interface by MAC address. Returns NULL if the device
565 * is not found or a pointer to the device. The caller must hold the
566 * rtnl semaphore. The returned device has not had its ref count increased
567 * and the caller must therefore be careful about locking
570 * If the API was consistent this would be __dev_get_by_hwaddr
573 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
575 struct net_device *dev;
579 for (dev = dev_base; dev; dev = dev->next)
580 if (dev->type == type &&
581 !memcmp(dev->dev_addr, ha, dev->addr_len))
586 EXPORT_SYMBOL(dev_getbyhwaddr);
588 struct net_device *dev_getfirstbyhwtype(unsigned short type)
590 struct net_device *dev;
593 for (dev = dev_base; dev; dev = dev->next) {
594 if (dev->type == type) {
603 EXPORT_SYMBOL(dev_getfirstbyhwtype);
606 * dev_get_by_flags - find any device with given flags
607 * @if_flags: IFF_* values
608 * @mask: bitmask of bits in if_flags to check
610 * Search for any interface with the given flags. Returns NULL if a device
611 * is not found or a pointer to the device. The device returned has
612 * had a reference added and the pointer is safe until the user calls
613 * dev_put to indicate they have finished with it.
616 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
618 struct net_device *dev;
620 read_lock(&dev_base_lock);
621 for (dev = dev_base; dev != NULL; dev = dev->next) {
622 if (((dev->flags ^ if_flags) & mask) == 0) {
627 read_unlock(&dev_base_lock);
632 * dev_valid_name - check if name is okay for network device
635 * Network device names need to be valid file names to
636 * to allow sysfs to work
638 int dev_valid_name(const char *name)
640 return !(*name == '\0'
641 || !strcmp(name, ".")
642 || !strcmp(name, "..")
643 || strchr(name, '/'));
647 * dev_alloc_name - allocate a name for a device
649 * @name: name format string
651 * Passed a format string - eg "lt%d" it will try and find a suitable
652 * id. It scans list of devices to build up a free map, then chooses
653 * the first empty slot. The caller must hold the dev_base or rtnl lock
654 * while allocating the name and adding the device in order to avoid
656 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
657 * Returns the number of the unit assigned or a negative errno code.
660 int dev_alloc_name(struct net_device *dev, const char *name)
665 const int max_netdevices = 8*PAGE_SIZE;
667 struct net_device *d;
669 p = strnchr(name, IFNAMSIZ-1, '%');
672 * Verify the string as this thing may have come from
673 * the user. There must be either one "%d" and no other "%"
676 if (p[1] != 'd' || strchr(p + 2, '%'))
679 /* Use one page as a bit array of possible slots */
680 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
684 for (d = dev_base; d; d = d->next) {
685 if (!sscanf(d->name, name, &i))
687 if (i < 0 || i >= max_netdevices)
690 /* avoid cases where sscanf is not exact inverse of printf */
691 snprintf(buf, sizeof(buf), name, i);
692 if (!strncmp(buf, d->name, IFNAMSIZ))
696 i = find_first_zero_bit(inuse, max_netdevices);
697 free_page((unsigned long) inuse);
700 snprintf(buf, sizeof(buf), name, i);
701 if (!__dev_get_by_name(buf)) {
702 strlcpy(dev->name, buf, IFNAMSIZ);
706 /* It is possible to run out of possible slots
707 * when the name is long and there isn't enough space left
708 * for the digits, or if all bits are used.
715 * dev_change_name - change name of a device
717 * @newname: name (or format string) must be at least IFNAMSIZ
719 * Change name of a device, can pass format strings "eth%d".
722 int dev_change_name(struct net_device *dev, char *newname)
728 if (dev->flags & IFF_UP)
731 if (!dev_valid_name(newname))
734 if (strchr(newname, '%')) {
735 err = dev_alloc_name(dev, newname);
738 strcpy(newname, dev->name);
740 else if (__dev_get_by_name(newname))
743 strlcpy(dev->name, newname, IFNAMSIZ);
745 err = class_device_rename(&dev->class_dev, dev->name);
747 hlist_del(&dev->name_hlist);
748 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
749 raw_notifier_call_chain(&netdev_chain,
750 NETDEV_CHANGENAME, dev);
757 * netdev_features_change - device changes features
758 * @dev: device to cause notification
760 * Called to indicate a device has changed features.
762 void netdev_features_change(struct net_device *dev)
764 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
766 EXPORT_SYMBOL(netdev_features_change);
769 * netdev_state_change - device changes state
770 * @dev: device to cause notification
772 * Called to indicate a device has changed state. This function calls
773 * the notifier chains for netdev_chain and sends a NEWLINK message
774 * to the routing socket.
776 void netdev_state_change(struct net_device *dev)
778 if (dev->flags & IFF_UP) {
779 raw_notifier_call_chain(&netdev_chain,
781 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
786 * dev_load - load a network module
787 * @name: name of interface
789 * If a network interface is not present and the process has suitable
790 * privileges this function loads the module. If module loading is not
791 * available in this kernel then it becomes a nop.
794 void dev_load(const char *name)
796 struct net_device *dev;
798 read_lock(&dev_base_lock);
799 dev = __dev_get_by_name(name);
800 read_unlock(&dev_base_lock);
802 if (!dev && capable(CAP_SYS_MODULE))
803 request_module("%s", name);
806 static int default_rebuild_header(struct sk_buff *skb)
808 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
809 skb->dev ? skb->dev->name : "NULL!!!");
816 * dev_open - prepare an interface for use.
817 * @dev: device to open
819 * Takes a device from down to up state. The device's private open
820 * function is invoked and then the multicast lists are loaded. Finally
821 * the device is moved into the up state and a %NETDEV_UP message is
822 * sent to the netdev notifier chain.
824 * Calling this function on an active interface is a nop. On a failure
825 * a negative errno code is returned.
827 int dev_open(struct net_device *dev)
835 if (dev->flags & IFF_UP)
839 * Is it even present?
841 if (!netif_device_present(dev))
845 * Call device private open method
847 set_bit(__LINK_STATE_START, &dev->state);
849 ret = dev->open(dev);
851 clear_bit(__LINK_STATE_START, &dev->state);
855 * If it went open OK then:
862 dev->flags |= IFF_UP;
865 * Initialize multicasting status
870 * Wakeup transmit queue engine
875 * ... and announce new interface.
877 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
883 * dev_close - shutdown an interface.
884 * @dev: device to shutdown
886 * This function moves an active device into down state. A
887 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
888 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
891 int dev_close(struct net_device *dev)
893 if (!(dev->flags & IFF_UP))
897 * Tell people we are going down, so that they can
898 * prepare to death, when device is still operating.
900 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
904 clear_bit(__LINK_STATE_START, &dev->state);
906 /* Synchronize to scheduled poll. We cannot touch poll list,
907 * it can be even on different cpu. So just clear netif_running(),
908 * and wait when poll really will happen. Actually, the best place
909 * for this is inside dev->stop() after device stopped its irq
910 * engine, but this requires more changes in devices. */
912 smp_mb__after_clear_bit(); /* Commit netif_running(). */
913 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
919 * Call the device specific close. This cannot fail.
920 * Only if device is UP
922 * We allow it to be called even after a DETACH hot-plug
929 * Device is now down.
932 dev->flags &= ~IFF_UP;
935 * Tell people we are down
937 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
944 * Device change register/unregister. These are not inline or static
945 * as we export them to the world.
949 * register_netdevice_notifier - register a network notifier block
952 * Register a notifier to be called when network device events occur.
953 * The notifier passed is linked into the kernel structures and must
954 * not be reused until it has been unregistered. A negative errno code
955 * is returned on a failure.
957 * When registered all registration and up events are replayed
958 * to the new notifier to allow device to have a race free
959 * view of the network device list.
962 int register_netdevice_notifier(struct notifier_block *nb)
964 struct net_device *dev;
968 err = raw_notifier_chain_register(&netdev_chain, nb);
970 for (dev = dev_base; dev; dev = dev->next) {
971 nb->notifier_call(nb, NETDEV_REGISTER, dev);
973 if (dev->flags & IFF_UP)
974 nb->notifier_call(nb, NETDEV_UP, dev);
982 * unregister_netdevice_notifier - unregister a network notifier block
985 * Unregister a notifier previously registered by
986 * register_netdevice_notifier(). The notifier is unlinked into the
987 * kernel structures and may then be reused. A negative errno code
988 * is returned on a failure.
991 int unregister_netdevice_notifier(struct notifier_block *nb)
996 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1002 * call_netdevice_notifiers - call all network notifier blocks
1003 * @val: value passed unmodified to notifier function
1004 * @v: pointer passed unmodified to notifier function
1006 * Call all network notifier blocks. Parameters and return value
1007 * are as for raw_notifier_call_chain().
1010 int call_netdevice_notifiers(unsigned long val, void *v)
1012 return raw_notifier_call_chain(&netdev_chain, val, v);
1015 /* When > 0 there are consumers of rx skb time stamps */
1016 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1018 void net_enable_timestamp(void)
1020 atomic_inc(&netstamp_needed);
1023 void net_disable_timestamp(void)
1025 atomic_dec(&netstamp_needed);
1028 void __net_timestamp(struct sk_buff *skb)
1032 do_gettimeofday(&tv);
1033 skb_set_timestamp(skb, &tv);
1035 EXPORT_SYMBOL(__net_timestamp);
1037 static inline void net_timestamp(struct sk_buff *skb)
1039 if (atomic_read(&netstamp_needed))
1040 __net_timestamp(skb);
1042 skb->tstamp.off_sec = 0;
1043 skb->tstamp.off_usec = 0;
1048 * Support routine. Sends outgoing frames to any network
1049 * taps currently in use.
1052 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1054 struct packet_type *ptype;
1059 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1060 /* Never send packets back to the socket
1061 * they originated from - MvS (miquels@drinkel.ow.org)
1063 if ((ptype->dev == dev || !ptype->dev) &&
1064 (ptype->af_packet_priv == NULL ||
1065 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1066 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1070 /* skb->nh should be correctly
1071 set by sender, so that the second statement is
1072 just protection against buggy protocols.
1074 skb2->mac.raw = skb2->data;
1076 if (skb2->nh.raw < skb2->data ||
1077 skb2->nh.raw > skb2->tail) {
1078 if (net_ratelimit())
1079 printk(KERN_CRIT "protocol %04x is "
1081 skb2->protocol, dev->name);
1082 skb2->nh.raw = skb2->data;
1085 skb2->h.raw = skb2->nh.raw;
1086 skb2->pkt_type = PACKET_OUTGOING;
1087 ptype->func(skb2, skb->dev, ptype, skb->dev);
1094 void __netif_schedule(struct net_device *dev)
1096 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1097 unsigned long flags;
1098 struct softnet_data *sd;
1100 local_irq_save(flags);
1101 sd = &__get_cpu_var(softnet_data);
1102 dev->next_sched = sd->output_queue;
1103 sd->output_queue = dev;
1104 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1105 local_irq_restore(flags);
1108 EXPORT_SYMBOL(__netif_schedule);
1110 void __netif_rx_schedule(struct net_device *dev)
1112 unsigned long flags;
1114 local_irq_save(flags);
1116 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1118 dev->quota += dev->weight;
1120 dev->quota = dev->weight;
1121 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1122 local_irq_restore(flags);
1124 EXPORT_SYMBOL(__netif_rx_schedule);
1126 void dev_kfree_skb_any(struct sk_buff *skb)
1128 if (in_irq() || irqs_disabled())
1129 dev_kfree_skb_irq(skb);
1133 EXPORT_SYMBOL(dev_kfree_skb_any);
1137 void netif_device_detach(struct net_device *dev)
1139 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1140 netif_running(dev)) {
1141 netif_stop_queue(dev);
1144 EXPORT_SYMBOL(netif_device_detach);
1146 void netif_device_attach(struct net_device *dev)
1148 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1149 netif_running(dev)) {
1150 netif_wake_queue(dev);
1151 __netdev_watchdog_up(dev);
1154 EXPORT_SYMBOL(netif_device_attach);
1158 * Invalidate hardware checksum when packet is to be mangled, and
1159 * complete checksum manually on outgoing path.
1161 int skb_checksum_help(struct sk_buff *skb, int inward)
1164 int ret = 0, offset = skb->h.raw - skb->data;
1167 skb->ip_summed = CHECKSUM_NONE;
1171 if (skb_cloned(skb)) {
1172 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1177 BUG_ON(offset > (int)skb->len);
1178 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1180 offset = skb->tail - skb->h.raw;
1181 BUG_ON(offset <= 0);
1182 BUG_ON(skb->csum + 2 > offset);
1184 *(u16*)(skb->h.raw + skb->csum) = 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 int type = skb->protocol;
1206 BUG_ON(skb_shinfo(skb)->frag_list);
1207 BUG_ON(skb->ip_summed != CHECKSUM_HW);
1209 skb->mac.raw = skb->data;
1210 skb->mac_len = skb->nh.raw - skb->data;
1211 __skb_pull(skb, skb->mac_len);
1214 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1215 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1216 segs = ptype->gso_segment(skb, features);
1222 __skb_push(skb, skb->data - skb->mac.raw);
1227 EXPORT_SYMBOL(skb_gso_segment);
1229 /* Take action when hardware reception checksum errors are detected. */
1231 void netdev_rx_csum_fault(struct net_device *dev)
1233 if (net_ratelimit()) {
1234 printk(KERN_ERR "%s: hw csum failure.\n",
1235 dev ? dev->name : "<unknown>");
1239 EXPORT_SYMBOL(netdev_rx_csum_fault);
1242 #ifdef CONFIG_HIGHMEM
1243 /* Actually, we should eliminate this check as soon as we know, that:
1244 * 1. IOMMU is present and allows to map all the memory.
1245 * 2. No high memory really exists on this machine.
1248 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1252 if (dev->features & NETIF_F_HIGHDMA)
1255 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1256 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1262 #define illegal_highdma(dev, skb) (0)
1266 void (*destructor)(struct sk_buff *skb);
1269 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1271 static void dev_gso_skb_destructor(struct sk_buff *skb)
1273 struct dev_gso_cb *cb;
1276 struct sk_buff *nskb = skb->next;
1278 skb->next = nskb->next;
1281 } while (skb->next);
1283 cb = DEV_GSO_CB(skb);
1285 cb->destructor(skb);
1289 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1290 * @skb: buffer to segment
1292 * This function segments the given skb and stores the list of segments
1295 static int dev_gso_segment(struct sk_buff *skb)
1297 struct net_device *dev = skb->dev;
1298 struct sk_buff *segs;
1299 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1302 segs = skb_gso_segment(skb, features);
1304 /* Verifying header integrity only. */
1308 if (unlikely(IS_ERR(segs)))
1309 return PTR_ERR(segs);
1312 DEV_GSO_CB(skb)->destructor = skb->destructor;
1313 skb->destructor = dev_gso_skb_destructor;
1318 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1320 if (likely(!skb->next)) {
1322 dev_queue_xmit_nit(skb, dev);
1324 if (netif_needs_gso(dev, skb)) {
1325 if (unlikely(dev_gso_segment(skb)))
1331 return dev->hard_start_xmit(skb, dev);
1336 struct sk_buff *nskb = skb->next;
1339 skb->next = nskb->next;
1341 rc = dev->hard_start_xmit(nskb, dev);
1343 nskb->next = skb->next;
1347 if (unlikely(netif_queue_stopped(dev) && skb->next))
1348 return NETDEV_TX_BUSY;
1349 } while (skb->next);
1351 skb->destructor = DEV_GSO_CB(skb)->destructor;
1358 #define HARD_TX_LOCK(dev, cpu) { \
1359 if ((dev->features & NETIF_F_LLTX) == 0) { \
1360 netif_tx_lock(dev); \
1364 #define HARD_TX_UNLOCK(dev) { \
1365 if ((dev->features & NETIF_F_LLTX) == 0) { \
1366 netif_tx_unlock(dev); \
1371 * dev_queue_xmit - transmit a buffer
1372 * @skb: buffer to transmit
1374 * Queue a buffer for transmission to a network device. The caller must
1375 * have set the device and priority and built the buffer before calling
1376 * this function. The function can be called from an interrupt.
1378 * A negative errno code is returned on a failure. A success does not
1379 * guarantee the frame will be transmitted as it may be dropped due
1380 * to congestion or traffic shaping.
1382 * -----------------------------------------------------------------------------------
1383 * I notice this method can also return errors from the queue disciplines,
1384 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1387 * Regardless of the return value, the skb is consumed, so it is currently
1388 * difficult to retry a send to this method. (You can bump the ref count
1389 * before sending to hold a reference for retry if you are careful.)
1391 * When calling this method, interrupts MUST be enabled. This is because
1392 * the BH enable code must have IRQs enabled so that it will not deadlock.
1396 int dev_queue_xmit(struct sk_buff *skb)
1398 struct net_device *dev = skb->dev;
1402 /* GSO will handle the following emulations directly. */
1403 if (netif_needs_gso(dev, skb))
1406 if (skb_shinfo(skb)->frag_list &&
1407 !(dev->features & NETIF_F_FRAGLIST) &&
1408 __skb_linearize(skb))
1411 /* Fragmented skb is linearized if device does not support SG,
1412 * or if at least one of fragments is in highmem and device
1413 * does not support DMA from it.
1415 if (skb_shinfo(skb)->nr_frags &&
1416 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1417 __skb_linearize(skb))
1420 /* If packet is not checksummed and device does not support
1421 * checksumming for this protocol, complete checksumming here.
1423 if (skb->ip_summed == CHECKSUM_HW &&
1424 (!(dev->features & NETIF_F_GEN_CSUM) &&
1425 (!(dev->features & NETIF_F_IP_CSUM) ||
1426 skb->protocol != htons(ETH_P_IP))))
1427 if (skb_checksum_help(skb, 0))
1431 spin_lock_prefetch(&dev->queue_lock);
1433 /* Disable soft irqs for various locks below. Also
1434 * stops preemption for RCU.
1438 /* Updates of qdisc are serialized by queue_lock.
1439 * The struct Qdisc which is pointed to by qdisc is now a
1440 * rcu structure - it may be accessed without acquiring
1441 * a lock (but the structure may be stale.) The freeing of the
1442 * qdisc will be deferred until it's known that there are no
1443 * more references to it.
1445 * If the qdisc has an enqueue function, we still need to
1446 * hold the queue_lock before calling it, since queue_lock
1447 * also serializes access to the device queue.
1450 q = rcu_dereference(dev->qdisc);
1451 #ifdef CONFIG_NET_CLS_ACT
1452 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1455 /* Grab device queue */
1456 spin_lock(&dev->queue_lock);
1458 rc = q->enqueue(skb, q);
1462 spin_unlock(&dev->queue_lock);
1463 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1467 /* The device has no queue. Common case for software devices:
1468 loopback, all the sorts of tunnels...
1470 Really, it is unlikely that netif_tx_lock protection is necessary
1471 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1473 However, it is possible, that they rely on protection
1476 Check this and shot the lock. It is not prone from deadlocks.
1477 Either shot noqueue qdisc, it is even simpler 8)
1479 if (dev->flags & IFF_UP) {
1480 int cpu = smp_processor_id(); /* ok because BHs are off */
1482 if (dev->xmit_lock_owner != cpu) {
1484 HARD_TX_LOCK(dev, cpu);
1486 if (!netif_queue_stopped(dev)) {
1488 if (!dev_hard_start_xmit(skb, dev)) {
1489 HARD_TX_UNLOCK(dev);
1493 HARD_TX_UNLOCK(dev);
1494 if (net_ratelimit())
1495 printk(KERN_CRIT "Virtual device %s asks to "
1496 "queue packet!\n", dev->name);
1498 /* Recursion is detected! It is possible,
1500 if (net_ratelimit())
1501 printk(KERN_CRIT "Dead loop on virtual device "
1502 "%s, fix it urgently!\n", dev->name);
1507 rcu_read_unlock_bh();
1513 rcu_read_unlock_bh();
1518 /*=======================================================================
1520 =======================================================================*/
1522 int netdev_max_backlog = 1000;
1523 int netdev_budget = 300;
1524 int weight_p = 64; /* old backlog weight */
1526 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1530 * netif_rx - post buffer to the network code
1531 * @skb: buffer to post
1533 * This function receives a packet from a device driver and queues it for
1534 * the upper (protocol) levels to process. It always succeeds. The buffer
1535 * may be dropped during processing for congestion control or by the
1539 * NET_RX_SUCCESS (no congestion)
1540 * NET_RX_CN_LOW (low congestion)
1541 * NET_RX_CN_MOD (moderate congestion)
1542 * NET_RX_CN_HIGH (high congestion)
1543 * NET_RX_DROP (packet was dropped)
1547 int netif_rx(struct sk_buff *skb)
1549 struct softnet_data *queue;
1550 unsigned long flags;
1552 /* if netpoll wants it, pretend we never saw it */
1553 if (netpoll_rx(skb))
1556 if (!skb->tstamp.off_sec)
1560 * The code is rearranged so that the path is the most
1561 * short when CPU is congested, but is still operating.
1563 local_irq_save(flags);
1564 queue = &__get_cpu_var(softnet_data);
1566 __get_cpu_var(netdev_rx_stat).total++;
1567 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1568 if (queue->input_pkt_queue.qlen) {
1571 __skb_queue_tail(&queue->input_pkt_queue, skb);
1572 local_irq_restore(flags);
1573 return NET_RX_SUCCESS;
1576 netif_rx_schedule(&queue->backlog_dev);
1580 __get_cpu_var(netdev_rx_stat).dropped++;
1581 local_irq_restore(flags);
1587 int netif_rx_ni(struct sk_buff *skb)
1592 err = netif_rx(skb);
1593 if (local_softirq_pending())
1600 EXPORT_SYMBOL(netif_rx_ni);
1602 static inline struct net_device *skb_bond(struct sk_buff *skb)
1604 struct net_device *dev = skb->dev;
1608 * On bonding slaves other than the currently active
1609 * slave, suppress duplicates except for 802.3ad
1610 * ETH_P_SLOW and alb non-mcast/bcast.
1612 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1613 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1614 if (skb->pkt_type != PACKET_BROADCAST &&
1615 skb->pkt_type != PACKET_MULTICAST)
1619 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1620 skb->protocol == __constant_htons(ETH_P_SLOW))
1627 skb->dev = dev->master;
1633 static void net_tx_action(struct softirq_action *h)
1635 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1637 if (sd->completion_queue) {
1638 struct sk_buff *clist;
1640 local_irq_disable();
1641 clist = sd->completion_queue;
1642 sd->completion_queue = NULL;
1646 struct sk_buff *skb = clist;
1647 clist = clist->next;
1649 BUG_TRAP(!atomic_read(&skb->users));
1654 if (sd->output_queue) {
1655 struct net_device *head;
1657 local_irq_disable();
1658 head = sd->output_queue;
1659 sd->output_queue = NULL;
1663 struct net_device *dev = head;
1664 head = head->next_sched;
1666 smp_mb__before_clear_bit();
1667 clear_bit(__LINK_STATE_SCHED, &dev->state);
1669 if (spin_trylock(&dev->queue_lock)) {
1671 spin_unlock(&dev->queue_lock);
1673 netif_schedule(dev);
1679 static __inline__ int deliver_skb(struct sk_buff *skb,
1680 struct packet_type *pt_prev,
1681 struct net_device *orig_dev)
1683 atomic_inc(&skb->users);
1684 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1687 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1688 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1690 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1691 unsigned char *addr);
1692 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1694 static __inline__ int handle_bridge(struct sk_buff **pskb,
1695 struct packet_type **pt_prev, int *ret,
1696 struct net_device *orig_dev)
1698 struct net_bridge_port *port;
1700 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1701 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1705 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1709 return br_handle_frame_hook(port, pskb);
1712 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1715 #ifdef CONFIG_NET_CLS_ACT
1716 /* TODO: Maybe we should just force sch_ingress to be compiled in
1717 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1718 * a compare and 2 stores extra right now if we dont have it on
1719 * but have CONFIG_NET_CLS_ACT
1720 * NOTE: This doesnt stop any functionality; if you dont have
1721 * the ingress scheduler, you just cant add policies on ingress.
1724 static int ing_filter(struct sk_buff *skb)
1727 struct net_device *dev = skb->dev;
1728 int result = TC_ACT_OK;
1730 if (dev->qdisc_ingress) {
1731 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1732 if (MAX_RED_LOOP < ttl++) {
1733 printk("Redir loop detected Dropping packet (%s->%s)\n",
1734 skb->input_dev->name, skb->dev->name);
1738 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1740 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1742 spin_lock(&dev->ingress_lock);
1743 if ((q = dev->qdisc_ingress) != NULL)
1744 result = q->enqueue(skb, q);
1745 spin_unlock(&dev->ingress_lock);
1753 int netif_receive_skb(struct sk_buff *skb)
1755 struct packet_type *ptype, *pt_prev;
1756 struct net_device *orig_dev;
1757 int ret = NET_RX_DROP;
1758 unsigned short type;
1760 /* if we've gotten here through NAPI, check netpoll */
1761 if (skb->dev->poll && netpoll_rx(skb))
1764 if (!skb->tstamp.off_sec)
1767 if (!skb->input_dev)
1768 skb->input_dev = skb->dev;
1770 orig_dev = skb_bond(skb);
1775 __get_cpu_var(netdev_rx_stat).total++;
1777 skb->h.raw = skb->nh.raw = skb->data;
1778 skb->mac_len = skb->nh.raw - skb->mac.raw;
1784 #ifdef CONFIG_NET_CLS_ACT
1785 if (skb->tc_verd & TC_NCLS) {
1786 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1791 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1792 if (!ptype->dev || ptype->dev == skb->dev) {
1794 ret = deliver_skb(skb, pt_prev, orig_dev);
1799 #ifdef CONFIG_NET_CLS_ACT
1801 ret = deliver_skb(skb, pt_prev, orig_dev);
1802 pt_prev = NULL; /* noone else should process this after*/
1804 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1807 ret = ing_filter(skb);
1809 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1818 handle_diverter(skb);
1820 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1823 type = skb->protocol;
1824 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1825 if (ptype->type == type &&
1826 (!ptype->dev || ptype->dev == skb->dev)) {
1828 ret = deliver_skb(skb, pt_prev, orig_dev);
1834 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1837 /* Jamal, now you will not able to escape explaining
1838 * me how you were going to use this. :-)
1848 static int process_backlog(struct net_device *backlog_dev, int *budget)
1851 int quota = min(backlog_dev->quota, *budget);
1852 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1853 unsigned long start_time = jiffies;
1855 backlog_dev->weight = weight_p;
1857 struct sk_buff *skb;
1858 struct net_device *dev;
1860 local_irq_disable();
1861 skb = __skb_dequeue(&queue->input_pkt_queue);
1868 netif_receive_skb(skb);
1874 if (work >= quota || jiffies - start_time > 1)
1879 backlog_dev->quota -= work;
1884 backlog_dev->quota -= work;
1887 list_del(&backlog_dev->poll_list);
1888 smp_mb__before_clear_bit();
1889 netif_poll_enable(backlog_dev);
1895 static void net_rx_action(struct softirq_action *h)
1897 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1898 unsigned long start_time = jiffies;
1899 int budget = netdev_budget;
1902 local_irq_disable();
1904 while (!list_empty(&queue->poll_list)) {
1905 struct net_device *dev;
1907 if (budget <= 0 || jiffies - start_time > 1)
1912 dev = list_entry(queue->poll_list.next,
1913 struct net_device, poll_list);
1914 have = netpoll_poll_lock(dev);
1916 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1917 netpoll_poll_unlock(have);
1918 local_irq_disable();
1919 list_move_tail(&dev->poll_list, &queue->poll_list);
1921 dev->quota += dev->weight;
1923 dev->quota = dev->weight;
1925 netpoll_poll_unlock(have);
1927 local_irq_disable();
1931 #ifdef CONFIG_NET_DMA
1933 * There may not be any more sk_buffs coming right now, so push
1934 * any pending DMA copies to hardware
1936 if (net_dma_client) {
1937 struct dma_chan *chan;
1939 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1940 dma_async_memcpy_issue_pending(chan);
1948 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1949 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1953 static gifconf_func_t * gifconf_list [NPROTO];
1956 * register_gifconf - register a SIOCGIF handler
1957 * @family: Address family
1958 * @gifconf: Function handler
1960 * Register protocol dependent address dumping routines. The handler
1961 * that is passed must not be freed or reused until it has been replaced
1962 * by another handler.
1964 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1966 if (family >= NPROTO)
1968 gifconf_list[family] = gifconf;
1974 * Map an interface index to its name (SIOCGIFNAME)
1978 * We need this ioctl for efficient implementation of the
1979 * if_indextoname() function required by the IPv6 API. Without
1980 * it, we would have to search all the interfaces to find a
1984 static int dev_ifname(struct ifreq __user *arg)
1986 struct net_device *dev;
1990 * Fetch the caller's info block.
1993 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1996 read_lock(&dev_base_lock);
1997 dev = __dev_get_by_index(ifr.ifr_ifindex);
1999 read_unlock(&dev_base_lock);
2003 strcpy(ifr.ifr_name, dev->name);
2004 read_unlock(&dev_base_lock);
2006 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2012 * Perform a SIOCGIFCONF call. This structure will change
2013 * size eventually, and there is nothing I can do about it.
2014 * Thus we will need a 'compatibility mode'.
2017 static int dev_ifconf(char __user *arg)
2020 struct net_device *dev;
2027 * Fetch the caller's info block.
2030 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2037 * Loop over the interfaces, and write an info block for each.
2041 for (dev = dev_base; dev; dev = dev->next) {
2042 for (i = 0; i < NPROTO; i++) {
2043 if (gifconf_list[i]) {
2046 done = gifconf_list[i](dev, NULL, 0);
2048 done = gifconf_list[i](dev, pos + total,
2058 * All done. Write the updated control block back to the caller.
2060 ifc.ifc_len = total;
2063 * Both BSD and Solaris return 0 here, so we do too.
2065 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2068 #ifdef CONFIG_PROC_FS
2070 * This is invoked by the /proc filesystem handler to display a device
2073 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2075 struct net_device *dev;
2078 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2080 return i == pos ? dev : NULL;
2083 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2085 read_lock(&dev_base_lock);
2086 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2089 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2092 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2095 void dev_seq_stop(struct seq_file *seq, void *v)
2097 read_unlock(&dev_base_lock);
2100 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2102 if (dev->get_stats) {
2103 struct net_device_stats *stats = dev->get_stats(dev);
2105 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2106 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2107 dev->name, stats->rx_bytes, stats->rx_packets,
2109 stats->rx_dropped + stats->rx_missed_errors,
2110 stats->rx_fifo_errors,
2111 stats->rx_length_errors + stats->rx_over_errors +
2112 stats->rx_crc_errors + stats->rx_frame_errors,
2113 stats->rx_compressed, stats->multicast,
2114 stats->tx_bytes, stats->tx_packets,
2115 stats->tx_errors, stats->tx_dropped,
2116 stats->tx_fifo_errors, stats->collisions,
2117 stats->tx_carrier_errors +
2118 stats->tx_aborted_errors +
2119 stats->tx_window_errors +
2120 stats->tx_heartbeat_errors,
2121 stats->tx_compressed);
2123 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2127 * Called from the PROCfs module. This now uses the new arbitrary sized
2128 * /proc/net interface to create /proc/net/dev
2130 static int dev_seq_show(struct seq_file *seq, void *v)
2132 if (v == SEQ_START_TOKEN)
2133 seq_puts(seq, "Inter-| Receive "
2135 " face |bytes packets errs drop fifo frame "
2136 "compressed multicast|bytes packets errs "
2137 "drop fifo colls carrier compressed\n");
2139 dev_seq_printf_stats(seq, v);
2143 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2145 struct netif_rx_stats *rc = NULL;
2147 while (*pos < NR_CPUS)
2148 if (cpu_online(*pos)) {
2149 rc = &per_cpu(netdev_rx_stat, *pos);
2156 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2158 return softnet_get_online(pos);
2161 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2164 return softnet_get_online(pos);
2167 static void softnet_seq_stop(struct seq_file *seq, void *v)
2171 static int softnet_seq_show(struct seq_file *seq, void *v)
2173 struct netif_rx_stats *s = v;
2175 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2176 s->total, s->dropped, s->time_squeeze, 0,
2177 0, 0, 0, 0, /* was fastroute */
2182 static struct seq_operations dev_seq_ops = {
2183 .start = dev_seq_start,
2184 .next = dev_seq_next,
2185 .stop = dev_seq_stop,
2186 .show = dev_seq_show,
2189 static int dev_seq_open(struct inode *inode, struct file *file)
2191 return seq_open(file, &dev_seq_ops);
2194 static struct file_operations dev_seq_fops = {
2195 .owner = THIS_MODULE,
2196 .open = dev_seq_open,
2198 .llseek = seq_lseek,
2199 .release = seq_release,
2202 static struct seq_operations softnet_seq_ops = {
2203 .start = softnet_seq_start,
2204 .next = softnet_seq_next,
2205 .stop = softnet_seq_stop,
2206 .show = softnet_seq_show,
2209 static int softnet_seq_open(struct inode *inode, struct file *file)
2211 return seq_open(file, &softnet_seq_ops);
2214 static struct file_operations softnet_seq_fops = {
2215 .owner = THIS_MODULE,
2216 .open = softnet_seq_open,
2218 .llseek = seq_lseek,
2219 .release = seq_release,
2222 #ifdef CONFIG_WIRELESS_EXT
2223 extern int wireless_proc_init(void);
2225 #define wireless_proc_init() 0
2228 static int __init dev_proc_init(void)
2232 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2234 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2236 if (wireless_proc_init())
2242 proc_net_remove("softnet_stat");
2244 proc_net_remove("dev");
2248 #define dev_proc_init() 0
2249 #endif /* CONFIG_PROC_FS */
2253 * netdev_set_master - set up master/slave pair
2254 * @slave: slave device
2255 * @master: new master device
2257 * Changes the master device of the slave. Pass %NULL to break the
2258 * bonding. The caller must hold the RTNL semaphore. On a failure
2259 * a negative errno code is returned. On success the reference counts
2260 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2261 * function returns zero.
2263 int netdev_set_master(struct net_device *slave, struct net_device *master)
2265 struct net_device *old = slave->master;
2275 slave->master = master;
2283 slave->flags |= IFF_SLAVE;
2285 slave->flags &= ~IFF_SLAVE;
2287 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2292 * dev_set_promiscuity - update promiscuity count on a device
2296 * Add or remove promiscuity from a device. While the count in the device
2297 * remains above zero the interface remains promiscuous. Once it hits zero
2298 * the device reverts back to normal filtering operation. A negative inc
2299 * value is used to drop promiscuity on the device.
2301 void dev_set_promiscuity(struct net_device *dev, int inc)
2303 unsigned short old_flags = dev->flags;
2305 if ((dev->promiscuity += inc) == 0)
2306 dev->flags &= ~IFF_PROMISC;
2308 dev->flags |= IFF_PROMISC;
2309 if (dev->flags != old_flags) {
2311 printk(KERN_INFO "device %s %s promiscuous mode\n",
2312 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2314 audit_log(current->audit_context, GFP_ATOMIC,
2315 AUDIT_ANOM_PROMISCUOUS,
2316 "dev=%s prom=%d old_prom=%d auid=%u",
2317 dev->name, (dev->flags & IFF_PROMISC),
2318 (old_flags & IFF_PROMISC),
2319 audit_get_loginuid(current->audit_context));
2324 * dev_set_allmulti - update allmulti count on a device
2328 * Add or remove reception of all multicast frames to a device. While the
2329 * count in the device remains above zero the interface remains listening
2330 * to all interfaces. Once it hits zero the device reverts back to normal
2331 * filtering operation. A negative @inc value is used to drop the counter
2332 * when releasing a resource needing all multicasts.
2335 void dev_set_allmulti(struct net_device *dev, int inc)
2337 unsigned short old_flags = dev->flags;
2339 dev->flags |= IFF_ALLMULTI;
2340 if ((dev->allmulti += inc) == 0)
2341 dev->flags &= ~IFF_ALLMULTI;
2342 if (dev->flags ^ old_flags)
2346 unsigned dev_get_flags(const struct net_device *dev)
2350 flags = (dev->flags & ~(IFF_PROMISC |
2355 (dev->gflags & (IFF_PROMISC |
2358 if (netif_running(dev)) {
2359 if (netif_oper_up(dev))
2360 flags |= IFF_RUNNING;
2361 if (netif_carrier_ok(dev))
2362 flags |= IFF_LOWER_UP;
2363 if (netif_dormant(dev))
2364 flags |= IFF_DORMANT;
2370 int dev_change_flags(struct net_device *dev, unsigned flags)
2373 int old_flags = dev->flags;
2376 * Set the flags on our device.
2379 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2380 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2382 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2386 * Load in the correct multicast list now the flags have changed.
2392 * Have we downed the interface. We handle IFF_UP ourselves
2393 * according to user attempts to set it, rather than blindly
2398 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2399 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2405 if (dev->flags & IFF_UP &&
2406 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2408 raw_notifier_call_chain(&netdev_chain,
2409 NETDEV_CHANGE, dev);
2411 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2412 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2413 dev->gflags ^= IFF_PROMISC;
2414 dev_set_promiscuity(dev, inc);
2417 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2418 is important. Some (broken) drivers set IFF_PROMISC, when
2419 IFF_ALLMULTI is requested not asking us and not reporting.
2421 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2422 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2423 dev->gflags ^= IFF_ALLMULTI;
2424 dev_set_allmulti(dev, inc);
2427 if (old_flags ^ dev->flags)
2428 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2433 int dev_set_mtu(struct net_device *dev, int new_mtu)
2437 if (new_mtu == dev->mtu)
2440 /* MTU must be positive. */
2444 if (!netif_device_present(dev))
2448 if (dev->change_mtu)
2449 err = dev->change_mtu(dev, new_mtu);
2452 if (!err && dev->flags & IFF_UP)
2453 raw_notifier_call_chain(&netdev_chain,
2454 NETDEV_CHANGEMTU, dev);
2458 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2462 if (!dev->set_mac_address)
2464 if (sa->sa_family != dev->type)
2466 if (!netif_device_present(dev))
2468 err = dev->set_mac_address(dev, sa);
2470 raw_notifier_call_chain(&netdev_chain,
2471 NETDEV_CHANGEADDR, dev);
2476 * Perform the SIOCxIFxxx calls.
2478 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2481 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2487 case SIOCGIFFLAGS: /* Get interface flags */
2488 ifr->ifr_flags = dev_get_flags(dev);
2491 case SIOCSIFFLAGS: /* Set interface flags */
2492 return dev_change_flags(dev, ifr->ifr_flags);
2494 case SIOCGIFMETRIC: /* Get the metric on the interface
2495 (currently unused) */
2496 ifr->ifr_metric = 0;
2499 case SIOCSIFMETRIC: /* Set the metric on the interface
2500 (currently unused) */
2503 case SIOCGIFMTU: /* Get the MTU of a device */
2504 ifr->ifr_mtu = dev->mtu;
2507 case SIOCSIFMTU: /* Set the MTU of a device */
2508 return dev_set_mtu(dev, ifr->ifr_mtu);
2512 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2514 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2515 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2516 ifr->ifr_hwaddr.sa_family = dev->type;
2520 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2522 case SIOCSIFHWBROADCAST:
2523 if (ifr->ifr_hwaddr.sa_family != dev->type)
2525 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2526 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2527 raw_notifier_call_chain(&netdev_chain,
2528 NETDEV_CHANGEADDR, dev);
2532 ifr->ifr_map.mem_start = dev->mem_start;
2533 ifr->ifr_map.mem_end = dev->mem_end;
2534 ifr->ifr_map.base_addr = dev->base_addr;
2535 ifr->ifr_map.irq = dev->irq;
2536 ifr->ifr_map.dma = dev->dma;
2537 ifr->ifr_map.port = dev->if_port;
2541 if (dev->set_config) {
2542 if (!netif_device_present(dev))
2544 return dev->set_config(dev, &ifr->ifr_map);
2549 if (!dev->set_multicast_list ||
2550 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2552 if (!netif_device_present(dev))
2554 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2558 if (!dev->set_multicast_list ||
2559 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2561 if (!netif_device_present(dev))
2563 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2567 ifr->ifr_ifindex = dev->ifindex;
2571 ifr->ifr_qlen = dev->tx_queue_len;
2575 if (ifr->ifr_qlen < 0)
2577 dev->tx_queue_len = ifr->ifr_qlen;
2581 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2582 return dev_change_name(dev, ifr->ifr_newname);
2585 * Unknown or private ioctl
2589 if ((cmd >= SIOCDEVPRIVATE &&
2590 cmd <= SIOCDEVPRIVATE + 15) ||
2591 cmd == SIOCBONDENSLAVE ||
2592 cmd == SIOCBONDRELEASE ||
2593 cmd == SIOCBONDSETHWADDR ||
2594 cmd == SIOCBONDSLAVEINFOQUERY ||
2595 cmd == SIOCBONDINFOQUERY ||
2596 cmd == SIOCBONDCHANGEACTIVE ||
2597 cmd == SIOCGMIIPHY ||
2598 cmd == SIOCGMIIREG ||
2599 cmd == SIOCSMIIREG ||
2600 cmd == SIOCBRADDIF ||
2601 cmd == SIOCBRDELIF ||
2602 cmd == SIOCWANDEV) {
2604 if (dev->do_ioctl) {
2605 if (netif_device_present(dev))
2606 err = dev->do_ioctl(dev, ifr,
2619 * This function handles all "interface"-type I/O control requests. The actual
2620 * 'doing' part of this is dev_ifsioc above.
2624 * dev_ioctl - network device ioctl
2625 * @cmd: command to issue
2626 * @arg: pointer to a struct ifreq in user space
2628 * Issue ioctl functions to devices. This is normally called by the
2629 * user space syscall interfaces but can sometimes be useful for
2630 * other purposes. The return value is the return from the syscall if
2631 * positive or a negative errno code on error.
2634 int dev_ioctl(unsigned int cmd, void __user *arg)
2640 /* One special case: SIOCGIFCONF takes ifconf argument
2641 and requires shared lock, because it sleeps writing
2645 if (cmd == SIOCGIFCONF) {
2647 ret = dev_ifconf((char __user *) arg);
2651 if (cmd == SIOCGIFNAME)
2652 return dev_ifname((struct ifreq __user *)arg);
2654 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2657 ifr.ifr_name[IFNAMSIZ-1] = 0;
2659 colon = strchr(ifr.ifr_name, ':');
2664 * See which interface the caller is talking about.
2669 * These ioctl calls:
2670 * - can be done by all.
2671 * - atomic and do not require locking.
2682 dev_load(ifr.ifr_name);
2683 read_lock(&dev_base_lock);
2684 ret = dev_ifsioc(&ifr, cmd);
2685 read_unlock(&dev_base_lock);
2689 if (copy_to_user(arg, &ifr,
2690 sizeof(struct ifreq)))
2696 dev_load(ifr.ifr_name);
2698 ret = dev_ethtool(&ifr);
2703 if (copy_to_user(arg, &ifr,
2704 sizeof(struct ifreq)))
2710 * These ioctl calls:
2711 * - require superuser power.
2712 * - require strict serialization.
2718 if (!capable(CAP_NET_ADMIN))
2720 dev_load(ifr.ifr_name);
2722 ret = dev_ifsioc(&ifr, cmd);
2727 if (copy_to_user(arg, &ifr,
2728 sizeof(struct ifreq)))
2734 * These ioctl calls:
2735 * - require superuser power.
2736 * - require strict serialization.
2737 * - do not return a value
2747 case SIOCSIFHWBROADCAST:
2750 case SIOCBONDENSLAVE:
2751 case SIOCBONDRELEASE:
2752 case SIOCBONDSETHWADDR:
2753 case SIOCBONDCHANGEACTIVE:
2756 if (!capable(CAP_NET_ADMIN))
2759 case SIOCBONDSLAVEINFOQUERY:
2760 case SIOCBONDINFOQUERY:
2761 dev_load(ifr.ifr_name);
2763 ret = dev_ifsioc(&ifr, cmd);
2768 /* Get the per device memory space. We can add this but
2769 * currently do not support it */
2771 /* Set the per device memory buffer space.
2772 * Not applicable in our case */
2777 * Unknown or private ioctl.
2780 if (cmd == SIOCWANDEV ||
2781 (cmd >= SIOCDEVPRIVATE &&
2782 cmd <= SIOCDEVPRIVATE + 15)) {
2783 dev_load(ifr.ifr_name);
2785 ret = dev_ifsioc(&ifr, cmd);
2787 if (!ret && copy_to_user(arg, &ifr,
2788 sizeof(struct ifreq)))
2792 #ifdef CONFIG_WIRELESS_EXT
2793 /* Take care of Wireless Extensions */
2794 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2795 /* If command is `set a parameter', or
2796 * `get the encoding parameters', check if
2797 * the user has the right to do it */
2798 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2799 || cmd == SIOCGIWENCODEEXT) {
2800 if (!capable(CAP_NET_ADMIN))
2803 dev_load(ifr.ifr_name);
2805 /* Follow me in net/core/wireless.c */
2806 ret = wireless_process_ioctl(&ifr, cmd);
2808 if (IW_IS_GET(cmd) &&
2809 copy_to_user(arg, &ifr,
2810 sizeof(struct ifreq)))
2814 #endif /* CONFIG_WIRELESS_EXT */
2821 * dev_new_index - allocate an ifindex
2823 * Returns a suitable unique value for a new device interface
2824 * number. The caller must hold the rtnl semaphore or the
2825 * dev_base_lock to be sure it remains unique.
2827 static int dev_new_index(void)
2833 if (!__dev_get_by_index(ifindex))
2838 static int dev_boot_phase = 1;
2840 /* Delayed registration/unregisteration */
2841 static DEFINE_SPINLOCK(net_todo_list_lock);
2842 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2844 static inline void net_set_todo(struct net_device *dev)
2846 spin_lock(&net_todo_list_lock);
2847 list_add_tail(&dev->todo_list, &net_todo_list);
2848 spin_unlock(&net_todo_list_lock);
2852 * register_netdevice - register a network device
2853 * @dev: device to register
2855 * Take a completed network device structure and add it to the kernel
2856 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2857 * chain. 0 is returned on success. A negative errno code is returned
2858 * on a failure to set up the device, or if the name is a duplicate.
2860 * Callers must hold the rtnl semaphore. You may want
2861 * register_netdev() instead of this.
2864 * The locking appears insufficient to guarantee two parallel registers
2865 * will not get the same name.
2868 int register_netdevice(struct net_device *dev)
2870 struct hlist_head *head;
2871 struct hlist_node *p;
2874 BUG_ON(dev_boot_phase);
2879 /* When net_device's are persistent, this will be fatal. */
2880 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2882 spin_lock_init(&dev->queue_lock);
2883 spin_lock_init(&dev->_xmit_lock);
2884 dev->xmit_lock_owner = -1;
2885 #ifdef CONFIG_NET_CLS_ACT
2886 spin_lock_init(&dev->ingress_lock);
2889 ret = alloc_divert_blk(dev);
2895 /* Init, if this function is available */
2897 ret = dev->init(dev);
2905 if (!dev_valid_name(dev->name)) {
2910 dev->ifindex = dev_new_index();
2911 if (dev->iflink == -1)
2912 dev->iflink = dev->ifindex;
2914 /* Check for existence of name */
2915 head = dev_name_hash(dev->name);
2916 hlist_for_each(p, head) {
2917 struct net_device *d
2918 = hlist_entry(p, struct net_device, name_hlist);
2919 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2925 /* Fix illegal SG+CSUM combinations. */
2926 if ((dev->features & NETIF_F_SG) &&
2927 !(dev->features & NETIF_F_ALL_CSUM)) {
2928 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2930 dev->features &= ~NETIF_F_SG;
2933 /* TSO requires that SG is present as well. */
2934 if ((dev->features & NETIF_F_TSO) &&
2935 !(dev->features & NETIF_F_SG)) {
2936 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2938 dev->features &= ~NETIF_F_TSO;
2940 if (dev->features & NETIF_F_UFO) {
2941 if (!(dev->features & NETIF_F_HW_CSUM)) {
2942 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2943 "NETIF_F_HW_CSUM feature.\n",
2945 dev->features &= ~NETIF_F_UFO;
2947 if (!(dev->features & NETIF_F_SG)) {
2948 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2949 "NETIF_F_SG feature.\n",
2951 dev->features &= ~NETIF_F_UFO;
2956 * nil rebuild_header routine,
2957 * that should be never called and used as just bug trap.
2960 if (!dev->rebuild_header)
2961 dev->rebuild_header = default_rebuild_header;
2963 ret = netdev_register_sysfs(dev);
2966 dev->reg_state = NETREG_REGISTERED;
2969 * Default initial state at registry is that the
2970 * device is present.
2973 set_bit(__LINK_STATE_PRESENT, &dev->state);
2976 dev_init_scheduler(dev);
2977 write_lock_bh(&dev_base_lock);
2979 dev_tail = &dev->next;
2980 hlist_add_head(&dev->name_hlist, head);
2981 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2983 write_unlock_bh(&dev_base_lock);
2985 /* Notify protocols, that a new device appeared. */
2986 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2993 free_divert_blk(dev);
2998 * register_netdev - register a network device
2999 * @dev: device to register
3001 * Take a completed network device structure and add it to the kernel
3002 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3003 * chain. 0 is returned on success. A negative errno code is returned
3004 * on a failure to set up the device, or if the name is a duplicate.
3006 * This is a wrapper around register_netdev that takes the rtnl semaphore
3007 * and expands the device name if you passed a format string to
3010 int register_netdev(struct net_device *dev)
3017 * If the name is a format string the caller wants us to do a
3020 if (strchr(dev->name, '%')) {
3021 err = dev_alloc_name(dev, dev->name);
3027 * Back compatibility hook. Kill this one in 2.5
3029 if (dev->name[0] == 0 || dev->name[0] == ' ') {
3030 err = dev_alloc_name(dev, "eth%d");
3035 err = register_netdevice(dev);
3040 EXPORT_SYMBOL(register_netdev);
3043 * netdev_wait_allrefs - wait until all references are gone.
3045 * This is called when unregistering network devices.
3047 * Any protocol or device that holds a reference should register
3048 * for netdevice notification, and cleanup and put back the
3049 * reference if they receive an UNREGISTER event.
3050 * We can get stuck here if buggy protocols don't correctly
3053 static void netdev_wait_allrefs(struct net_device *dev)
3055 unsigned long rebroadcast_time, warning_time;
3057 rebroadcast_time = warning_time = jiffies;
3058 while (atomic_read(&dev->refcnt) != 0) {
3059 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3062 /* Rebroadcast unregister notification */
3063 raw_notifier_call_chain(&netdev_chain,
3064 NETDEV_UNREGISTER, dev);
3066 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3068 /* We must not have linkwatch events
3069 * pending on unregister. If this
3070 * happens, we simply run the queue
3071 * unscheduled, resulting in a noop
3074 linkwatch_run_queue();
3079 rebroadcast_time = jiffies;
3084 if (time_after(jiffies, warning_time + 10 * HZ)) {
3085 printk(KERN_EMERG "unregister_netdevice: "
3086 "waiting for %s to become free. Usage "
3088 dev->name, atomic_read(&dev->refcnt));
3089 warning_time = jiffies;
3098 * register_netdevice(x1);
3099 * register_netdevice(x2);
3101 * unregister_netdevice(y1);
3102 * unregister_netdevice(y2);
3108 * We are invoked by rtnl_unlock() after it drops the semaphore.
3109 * This allows us to deal with problems:
3110 * 1) We can delete sysfs objects which invoke hotplug
3111 * without deadlocking with linkwatch via keventd.
3112 * 2) Since we run with the RTNL semaphore not held, we can sleep
3113 * safely in order to wait for the netdev refcnt to drop to zero.
3115 static DEFINE_MUTEX(net_todo_run_mutex);
3116 void netdev_run_todo(void)
3118 struct list_head list;
3120 /* Need to guard against multiple cpu's getting out of order. */
3121 mutex_lock(&net_todo_run_mutex);
3123 /* Not safe to do outside the semaphore. We must not return
3124 * until all unregister events invoked by the local processor
3125 * have been completed (either by this todo run, or one on
3128 if (list_empty(&net_todo_list))
3131 /* Snapshot list, allow later requests */
3132 spin_lock(&net_todo_list_lock);
3133 list_replace_init(&net_todo_list, &list);
3134 spin_unlock(&net_todo_list_lock);
3136 while (!list_empty(&list)) {
3137 struct net_device *dev
3138 = list_entry(list.next, struct net_device, todo_list);
3139 list_del(&dev->todo_list);
3141 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3142 printk(KERN_ERR "network todo '%s' but state %d\n",
3143 dev->name, dev->reg_state);
3148 netdev_unregister_sysfs(dev);
3149 dev->reg_state = NETREG_UNREGISTERED;
3151 netdev_wait_allrefs(dev);
3154 BUG_ON(atomic_read(&dev->refcnt));
3155 BUG_TRAP(!dev->ip_ptr);
3156 BUG_TRAP(!dev->ip6_ptr);
3157 BUG_TRAP(!dev->dn_ptr);
3159 /* It must be the very last action,
3160 * after this 'dev' may point to freed up memory.
3162 if (dev->destructor)
3163 dev->destructor(dev);
3167 mutex_unlock(&net_todo_run_mutex);
3171 * alloc_netdev - allocate network device
3172 * @sizeof_priv: size of private data to allocate space for
3173 * @name: device name format string
3174 * @setup: callback to initialize device
3176 * Allocates a struct net_device with private data area for driver use
3177 * and performs basic initialization.
3179 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3180 void (*setup)(struct net_device *))
3183 struct net_device *dev;
3186 /* ensure 32-byte alignment of both the device and private area */
3187 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3188 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3190 p = kzalloc(alloc_size, GFP_KERNEL);
3192 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3196 dev = (struct net_device *)
3197 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3198 dev->padded = (char *)dev - (char *)p;
3201 dev->priv = netdev_priv(dev);
3204 strcpy(dev->name, name);
3207 EXPORT_SYMBOL(alloc_netdev);
3210 * free_netdev - free network device
3213 * This function does the last stage of destroying an allocated device
3214 * interface. The reference to the device object is released.
3215 * If this is the last reference then it will be freed.
3217 void free_netdev(struct net_device *dev)
3220 /* Compatibility with error handling in drivers */
3221 if (dev->reg_state == NETREG_UNINITIALIZED) {
3222 kfree((char *)dev - dev->padded);
3226 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3227 dev->reg_state = NETREG_RELEASED;
3229 /* will free via class release */
3230 class_device_put(&dev->class_dev);
3232 kfree((char *)dev - dev->padded);
3236 /* Synchronize with packet receive processing. */
3237 void synchronize_net(void)
3244 * unregister_netdevice - remove device from the kernel
3247 * This function shuts down a device interface and removes it
3248 * from the kernel tables. On success 0 is returned, on a failure
3249 * a negative errno code is returned.
3251 * Callers must hold the rtnl semaphore. You may want
3252 * unregister_netdev() instead of this.
3255 int unregister_netdevice(struct net_device *dev)
3257 struct net_device *d, **dp;
3259 BUG_ON(dev_boot_phase);
3262 /* Some devices call without registering for initialization unwind. */
3263 if (dev->reg_state == NETREG_UNINITIALIZED) {
3264 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3265 "was registered\n", dev->name, dev);
3269 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3271 /* If device is running, close it first. */
3272 if (dev->flags & IFF_UP)
3275 /* And unlink it from device chain. */
3276 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3278 write_lock_bh(&dev_base_lock);
3279 hlist_del(&dev->name_hlist);
3280 hlist_del(&dev->index_hlist);
3281 if (dev_tail == &dev->next)
3284 write_unlock_bh(&dev_base_lock);
3289 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3294 dev->reg_state = NETREG_UNREGISTERING;
3298 /* Shutdown queueing discipline. */
3302 /* Notify protocols, that we are about to destroy
3303 this device. They should clean all the things.
3305 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3308 * Flush the multicast chain
3310 dev_mc_discard(dev);
3315 /* Notifier chain MUST detach us from master device. */
3316 BUG_TRAP(!dev->master);
3318 free_divert_blk(dev);
3320 /* Finish processing unregister after unlock */
3330 * unregister_netdev - remove device from the kernel
3333 * This function shuts down a device interface and removes it
3334 * from the kernel tables. On success 0 is returned, on a failure
3335 * a negative errno code is returned.
3337 * This is just a wrapper for unregister_netdevice that takes
3338 * the rtnl semaphore. In general you want to use this and not
3339 * unregister_netdevice.
3341 void unregister_netdev(struct net_device *dev)
3344 unregister_netdevice(dev);
3348 EXPORT_SYMBOL(unregister_netdev);
3350 #ifdef CONFIG_HOTPLUG_CPU
3351 static int dev_cpu_callback(struct notifier_block *nfb,
3352 unsigned long action,
3355 struct sk_buff **list_skb;
3356 struct net_device **list_net;
3357 struct sk_buff *skb;
3358 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3359 struct softnet_data *sd, *oldsd;
3361 if (action != CPU_DEAD)
3364 local_irq_disable();
3365 cpu = smp_processor_id();
3366 sd = &per_cpu(softnet_data, cpu);
3367 oldsd = &per_cpu(softnet_data, oldcpu);
3369 /* Find end of our completion_queue. */
3370 list_skb = &sd->completion_queue;
3372 list_skb = &(*list_skb)->next;
3373 /* Append completion queue from offline CPU. */
3374 *list_skb = oldsd->completion_queue;
3375 oldsd->completion_queue = NULL;
3377 /* Find end of our output_queue. */
3378 list_net = &sd->output_queue;
3380 list_net = &(*list_net)->next_sched;
3381 /* Append output queue from offline CPU. */
3382 *list_net = oldsd->output_queue;
3383 oldsd->output_queue = NULL;
3385 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3388 /* Process offline CPU's input_pkt_queue */
3389 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3394 #endif /* CONFIG_HOTPLUG_CPU */
3396 #ifdef CONFIG_NET_DMA
3398 * net_dma_rebalance -
3399 * This is called when the number of channels allocated to the net_dma_client
3400 * changes. The net_dma_client tries to have one DMA channel per CPU.
3402 static void net_dma_rebalance(void)
3404 unsigned int cpu, i, n;
3405 struct dma_chan *chan;
3409 if (net_dma_count == 0) {
3410 for_each_online_cpu(cpu)
3411 rcu_assign_pointer(per_cpu(softnet_data.net_dma, cpu), NULL);
3412 unlock_cpu_hotplug();
3417 cpu = first_cpu(cpu_online_map);
3420 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3421 n = ((num_online_cpus() / net_dma_count)
3422 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3425 per_cpu(softnet_data.net_dma, cpu) = chan;
3426 cpu = next_cpu(cpu, cpu_online_map);
3433 unlock_cpu_hotplug();
3437 * netdev_dma_event - event callback for the net_dma_client
3438 * @client: should always be net_dma_client
3439 * @chan: DMA channel for the event
3440 * @event: event type
3442 static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3443 enum dma_event event)
3445 spin_lock(&net_dma_event_lock);
3447 case DMA_RESOURCE_ADDED:
3449 net_dma_rebalance();
3451 case DMA_RESOURCE_REMOVED:
3453 net_dma_rebalance();
3458 spin_unlock(&net_dma_event_lock);
3462 * netdev_dma_regiser - register the networking subsystem as a DMA client
3464 static int __init netdev_dma_register(void)
3466 spin_lock_init(&net_dma_event_lock);
3467 net_dma_client = dma_async_client_register(netdev_dma_event);
3468 if (net_dma_client == NULL)
3471 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3476 static int __init netdev_dma_register(void) { return -ENODEV; }
3477 #endif /* CONFIG_NET_DMA */
3480 * Initialize the DEV module. At boot time this walks the device list and
3481 * unhooks any devices that fail to initialise (normally hardware not
3482 * present) and leaves us with a valid list of present and active devices.
3487 * This is called single threaded during boot, so no need
3488 * to take the rtnl semaphore.
3490 static int __init net_dev_init(void)
3492 int i, rc = -ENOMEM;
3494 BUG_ON(!dev_boot_phase);
3498 if (dev_proc_init())
3501 if (netdev_sysfs_init())
3504 INIT_LIST_HEAD(&ptype_all);
3505 for (i = 0; i < 16; i++)
3506 INIT_LIST_HEAD(&ptype_base[i]);
3508 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3509 INIT_HLIST_HEAD(&dev_name_head[i]);
3511 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3512 INIT_HLIST_HEAD(&dev_index_head[i]);
3515 * Initialise the packet receive queues.
3518 for_each_possible_cpu(i) {
3519 struct softnet_data *queue;
3521 queue = &per_cpu(softnet_data, i);
3522 skb_queue_head_init(&queue->input_pkt_queue);
3523 queue->completion_queue = NULL;
3524 INIT_LIST_HEAD(&queue->poll_list);
3525 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3526 queue->backlog_dev.weight = weight_p;
3527 queue->backlog_dev.poll = process_backlog;
3528 atomic_set(&queue->backlog_dev.refcnt, 1);
3531 netdev_dma_register();
3535 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3536 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3538 hotcpu_notifier(dev_cpu_callback, 0);
3546 subsys_initcall(net_dev_init);
3548 EXPORT_SYMBOL(__dev_get_by_index);
3549 EXPORT_SYMBOL(__dev_get_by_name);
3550 EXPORT_SYMBOL(__dev_remove_pack);
3551 EXPORT_SYMBOL(dev_valid_name);
3552 EXPORT_SYMBOL(dev_add_pack);
3553 EXPORT_SYMBOL(dev_alloc_name);
3554 EXPORT_SYMBOL(dev_close);
3555 EXPORT_SYMBOL(dev_get_by_flags);
3556 EXPORT_SYMBOL(dev_get_by_index);
3557 EXPORT_SYMBOL(dev_get_by_name);
3558 EXPORT_SYMBOL(dev_open);
3559 EXPORT_SYMBOL(dev_queue_xmit);
3560 EXPORT_SYMBOL(dev_remove_pack);
3561 EXPORT_SYMBOL(dev_set_allmulti);
3562 EXPORT_SYMBOL(dev_set_promiscuity);
3563 EXPORT_SYMBOL(dev_change_flags);
3564 EXPORT_SYMBOL(dev_set_mtu);
3565 EXPORT_SYMBOL(dev_set_mac_address);
3566 EXPORT_SYMBOL(free_netdev);
3567 EXPORT_SYMBOL(netdev_boot_setup_check);
3568 EXPORT_SYMBOL(netdev_set_master);
3569 EXPORT_SYMBOL(netdev_state_change);
3570 EXPORT_SYMBOL(netif_receive_skb);
3571 EXPORT_SYMBOL(netif_rx);
3572 EXPORT_SYMBOL(register_gifconf);
3573 EXPORT_SYMBOL(register_netdevice);
3574 EXPORT_SYMBOL(register_netdevice_notifier);
3575 EXPORT_SYMBOL(skb_checksum_help);
3576 EXPORT_SYMBOL(synchronize_net);
3577 EXPORT_SYMBOL(unregister_netdevice);
3578 EXPORT_SYMBOL(unregister_netdevice_notifier);
3579 EXPORT_SYMBOL(net_enable_timestamp);
3580 EXPORT_SYMBOL(net_disable_timestamp);
3581 EXPORT_SYMBOL(dev_get_flags);
3583 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3584 EXPORT_SYMBOL(br_handle_frame_hook);
3585 EXPORT_SYMBOL(br_fdb_get_hook);
3586 EXPORT_SYMBOL(br_fdb_put_hook);
3590 EXPORT_SYMBOL(dev_load);
3593 EXPORT_PER_CPU_SYMBOL(softnet_data);