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/config.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/notifier.h>
93 #include <linux/skbuff.h>
95 #include <linux/rtnetlink.h>
96 #include <linux/proc_fs.h>
97 #include <linux/seq_file.h>
98 #include <linux/stat.h>
99 #include <linux/if_bridge.h>
100 #include <linux/divert.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 #ifdef CONFIG_NET_RADIO
113 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
114 #include <net/iw_handler.h>
115 #endif /* CONFIG_NET_RADIO */
116 #include <linux/vs_network.h>
117 #include <asm/current.h>
119 /* This define, if set, will randomly drop a packet when congestion
120 * is more than moderate. It helps fairness in the multi-interface
121 * case when one of them is a hog, but it kills performance for the
122 * single interface case so it is off now by default.
126 /* Setting this will sample the queue lengths and thus congestion
127 * via a timer instead of as each packet is received.
129 #undef OFFLINE_SAMPLE
132 * The list of packet types we will receive (as opposed to discard)
133 * and the routines to invoke.
135 * Why 16. Because with 16 the only overlap we get on a hash of the
136 * low nibble of the protocol value is RARP/SNAP/X.25.
138 * NOTE: That is no longer true with the addition of VLAN tags. Not
139 * sure which should go first, but I bet it won't make much
140 * difference if we are running VLANs. The good news is that
141 * this protocol won't be in the list unless compiled in, so
142 * the average user (w/out VLANs) will not be adversly affected.
159 static DEFINE_SPINLOCK(ptype_lock);
160 static struct list_head ptype_base[16]; /* 16 way hashed list */
161 static struct list_head ptype_all; /* Taps */
163 #ifdef OFFLINE_SAMPLE
164 static void sample_queue(unsigned long dummy);
165 static struct timer_list samp_timer = TIMER_INITIALIZER(sample_queue, 0, 0);
169 * The @dev_base list is protected by @dev_base_lock and the rtln
172 * Pure readers hold dev_base_lock for reading.
174 * Writers must hold the rtnl semaphore while they loop through the
175 * dev_base list, and hold dev_base_lock for writing when they do the
176 * actual updates. This allows pure readers to access the list even
177 * while a writer is preparing to update it.
179 * To put it another way, dev_base_lock is held for writing only to
180 * protect against pure readers; the rtnl semaphore provides the
181 * protection against other writers.
183 * See, for example usages, register_netdevice() and
184 * unregister_netdevice(), which must be called with the rtnl
187 struct net_device *dev_base;
188 static struct net_device **dev_tail = &dev_base;
189 DEFINE_RWLOCK(dev_base_lock);
191 EXPORT_SYMBOL(dev_base);
192 EXPORT_SYMBOL(dev_base_lock);
194 #define NETDEV_HASHBITS 8
195 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
196 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
198 static inline struct hlist_head *dev_name_hash(const char *name)
200 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
201 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
204 static inline struct hlist_head *dev_index_hash(int ifindex)
206 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
213 static struct notifier_block *netdev_chain;
216 * Device drivers call our routines to queue packets here. We empty the
217 * queue in the local softnet handler.
219 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { 0, };
222 extern int netdev_sysfs_init(void);
223 extern int netdev_register_sysfs(struct net_device *);
224 extern void netdev_unregister_sysfs(struct net_device *);
226 #define netdev_sysfs_init() (0)
227 #define netdev_register_sysfs(dev) (0)
228 #define netdev_unregister_sysfs(dev) do { } while(0)
232 /*******************************************************************************
234 Protocol management and registration routines
236 *******************************************************************************/
245 * Add a protocol ID to the list. Now that the input handler is
246 * smarter we can dispense with all the messy stuff that used to be
249 * BEWARE!!! Protocol handlers, mangling input packets,
250 * MUST BE last in hash buckets and checking protocol handlers
251 * MUST start from promiscuous ptype_all chain in net_bh.
252 * It is true now, do not change it.
253 * Explanation follows: if protocol handler, mangling packet, will
254 * be the first on list, it is not able to sense, that packet
255 * is cloned and should be copied-on-write, so that it will
256 * change it and subsequent readers will get broken packet.
261 * dev_add_pack - add packet handler
262 * @pt: packet type declaration
264 * Add a protocol handler to the networking stack. The passed &packet_type
265 * is linked into kernel lists and may not be freed until it has been
266 * removed from the kernel lists.
268 * This call does not sleep therefore it can not
269 * guarantee all CPU's that are in middle of receiving packets
270 * will see the new packet type (until the next received packet).
273 void dev_add_pack(struct packet_type *pt)
277 spin_lock_bh(&ptype_lock);
278 if (pt->type == htons(ETH_P_ALL)) {
280 list_add_rcu(&pt->list, &ptype_all);
282 hash = ntohs(pt->type) & 15;
283 list_add_rcu(&pt->list, &ptype_base[hash]);
285 spin_unlock_bh(&ptype_lock);
288 extern void linkwatch_run_queue(void);
293 * __dev_remove_pack - remove packet handler
294 * @pt: packet type declaration
296 * Remove a protocol handler that was previously added to the kernel
297 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
298 * from the kernel lists and can be freed or reused once this function
301 * The packet type might still be in use by receivers
302 * and must not be freed until after all the CPU's have gone
303 * through a quiescent state.
305 void __dev_remove_pack(struct packet_type *pt)
307 struct list_head *head;
308 struct packet_type *pt1;
310 spin_lock_bh(&ptype_lock);
312 if (pt->type == htons(ETH_P_ALL)) {
316 head = &ptype_base[ntohs(pt->type) & 15];
318 list_for_each_entry(pt1, head, list) {
320 list_del_rcu(&pt->list);
325 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
327 spin_unlock_bh(&ptype_lock);
330 * dev_remove_pack - remove packet handler
331 * @pt: packet type declaration
333 * Remove a protocol handler that was previously added to the kernel
334 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
335 * from the kernel lists and can be freed or reused once this function
338 * This call sleeps to guarantee that no CPU is looking at the packet
341 void dev_remove_pack(struct packet_type *pt)
343 __dev_remove_pack(pt);
348 /******************************************************************************
350 Device Boot-time Settings Routines
352 *******************************************************************************/
354 /* Boot time configuration table */
355 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
358 * netdev_boot_setup_add - add new setup entry
359 * @name: name of the device
360 * @map: configured settings for the device
362 * Adds new setup entry to the dev_boot_setup list. The function
363 * returns 0 on error and 1 on success. This is a generic routine to
366 static int netdev_boot_setup_add(char *name, struct ifmap *map)
368 struct netdev_boot_setup *s;
372 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
373 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
374 memset(s[i].name, 0, sizeof(s[i].name));
375 strcpy(s[i].name, name);
376 memcpy(&s[i].map, map, sizeof(s[i].map));
381 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
385 * netdev_boot_setup_check - check boot time settings
386 * @dev: the netdevice
388 * Check boot time settings for the device.
389 * The found settings are set for the device to be used
390 * later in the device probing.
391 * Returns 0 if no settings found, 1 if they are.
393 int netdev_boot_setup_check(struct net_device *dev)
395 struct netdev_boot_setup *s = dev_boot_setup;
398 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
399 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
400 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
401 dev->irq = s[i].map.irq;
402 dev->base_addr = s[i].map.base_addr;
403 dev->mem_start = s[i].map.mem_start;
404 dev->mem_end = s[i].map.mem_end;
413 * netdev_boot_base - get address from boot time settings
414 * @prefix: prefix for network device
415 * @unit: id for network device
417 * Check boot time settings for the base address of device.
418 * The found settings are set for the device to be used
419 * later in the device probing.
420 * Returns 0 if no settings found.
422 unsigned long netdev_boot_base(const char *prefix, int unit)
424 const struct netdev_boot_setup *s = dev_boot_setup;
428 sprintf(name, "%s%d", prefix, unit);
431 * If device already registered then return base of 1
432 * to indicate not to probe for this interface
434 if (__dev_get_by_name(name))
437 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
438 if (!strcmp(name, s[i].name))
439 return s[i].map.base_addr;
444 * Saves at boot time configured settings for any netdevice.
446 int __init netdev_boot_setup(char *str)
451 str = get_options(str, ARRAY_SIZE(ints), ints);
456 memset(&map, 0, sizeof(map));
460 map.base_addr = ints[2];
462 map.mem_start = ints[3];
464 map.mem_end = ints[4];
466 /* Add new entry to the list */
467 return netdev_boot_setup_add(str, &map);
470 __setup("netdev=", netdev_boot_setup);
472 /*******************************************************************************
474 Device Interface Subroutines
476 *******************************************************************************/
479 * __dev_get_by_name - find a device by its name
480 * @name: name to find
482 * Find an interface by name. Must be called under RTNL semaphore
483 * or @dev_base_lock. If the name is found a pointer to the device
484 * is returned. If the name is not found then %NULL is returned. The
485 * reference counters are not incremented so the caller must be
486 * careful with locks.
489 struct net_device *__dev_get_by_name(const char *name)
491 struct hlist_node *p;
493 hlist_for_each(p, dev_name_hash(name)) {
494 struct net_device *dev
495 = hlist_entry(p, struct net_device, name_hlist);
496 if (!strncmp(dev->name, name, IFNAMSIZ))
503 * dev_get_by_name - find a device by its name
504 * @name: name to find
506 * Find an interface by name. This can be called from any
507 * context and does its own locking. The returned handle has
508 * the usage count incremented and the caller must use dev_put() to
509 * release it when it is no longer needed. %NULL is returned if no
510 * matching device is found.
513 struct net_device *dev_get_by_name(const char *name)
515 struct net_device *dev;
517 read_lock(&dev_base_lock);
518 dev = __dev_get_by_name(name);
521 read_unlock(&dev_base_lock);
526 * __dev_get_by_index - find a device by its ifindex
527 * @ifindex: index of device
529 * Search for an interface by index. Returns %NULL if the device
530 * is not found or a pointer to the device. The device has not
531 * had its reference counter increased so the caller must be careful
532 * about locking. The caller must hold either the RTNL semaphore
536 struct net_device *__dev_get_by_index(int ifindex)
538 struct hlist_node *p;
540 hlist_for_each(p, dev_index_hash(ifindex)) {
541 struct net_device *dev
542 = hlist_entry(p, struct net_device, index_hlist);
543 if (dev->ifindex == ifindex)
551 * dev_get_by_index - find a device by its ifindex
552 * @ifindex: index of device
554 * Search for an interface by index. Returns NULL if the device
555 * is not found or a pointer to the device. The device returned has
556 * had a reference added and the pointer is safe until the user calls
557 * dev_put to indicate they have finished with it.
560 struct net_device *dev_get_by_index(int ifindex)
562 struct net_device *dev;
564 read_lock(&dev_base_lock);
565 dev = __dev_get_by_index(ifindex);
568 read_unlock(&dev_base_lock);
573 * dev_getbyhwaddr - find a device by its hardware address
574 * @type: media type of device
575 * @ha: hardware address
577 * Search for an interface by MAC address. Returns NULL if the device
578 * is not found or a pointer to the device. The caller must hold the
579 * rtnl semaphore. The returned device has not had its ref count increased
580 * and the caller must therefore be careful about locking
583 * If the API was consistent this would be __dev_get_by_hwaddr
586 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
588 struct net_device *dev;
592 for (dev = dev_base; dev; dev = dev->next)
593 if (dev->type == type &&
594 !memcmp(dev->dev_addr, ha, dev->addr_len))
599 struct net_device *dev_getfirstbyhwtype(unsigned short type)
601 struct net_device *dev;
604 for (dev = dev_base; dev; dev = dev->next) {
605 if (dev->type == type) {
614 EXPORT_SYMBOL(dev_getfirstbyhwtype);
617 * dev_get_by_flags - find any device with given flags
618 * @if_flags: IFF_* values
619 * @mask: bitmask of bits in if_flags to check
621 * Search for any interface with the given flags. Returns NULL if a device
622 * is not found or a pointer to the device. The device returned has
623 * had a reference added and the pointer is safe until the user calls
624 * dev_put to indicate they have finished with it.
627 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
629 struct net_device *dev;
631 read_lock(&dev_base_lock);
632 for (dev = dev_base; dev != NULL; dev = dev->next) {
633 if (((dev->flags ^ if_flags) & mask) == 0) {
638 read_unlock(&dev_base_lock);
643 * dev_valid_name - check if name is okay for network device
646 * Network device names need to be valid file names to
647 * to allow sysfs to work
649 static int dev_valid_name(const char *name)
651 return !(*name == '\0'
652 || !strcmp(name, ".")
653 || !strcmp(name, "..")
654 || strchr(name, '/'));
658 * dev_alloc_name - allocate a name for a device
660 * @name: name format string
662 * Passed a format string - eg "lt%d" it will try and find a suitable
663 * id. Not efficient for many devices, not called a lot. The caller
664 * must hold the dev_base or rtnl lock while allocating the name and
665 * adding the device in order to avoid duplicates. Returns the number
666 * 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 err = class_device_rename(&dev->class_dev, dev->name);
756 hlist_del(&dev->name_hlist);
757 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
758 notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
765 * netdev_features_change - device changes fatures
766 * @dev: device to cause notification
768 * Called to indicate a device has changed features.
770 void netdev_features_change(struct net_device *dev)
772 notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
774 EXPORT_SYMBOL(netdev_features_change);
777 * netdev_state_change - device changes state
778 * @dev: device to cause notification
780 * Called to indicate a device has changed state. This function calls
781 * the notifier chains for netdev_chain and sends a NEWLINK message
782 * to the routing socket.
784 void netdev_state_change(struct net_device *dev)
786 if (dev->flags & IFF_UP) {
787 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
788 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
793 * dev_load - load a network module
794 * @name: name of interface
796 * If a network interface is not present and the process has suitable
797 * privileges this function loads the module. If module loading is not
798 * available in this kernel then it becomes a nop.
801 void dev_load(const char *name)
803 struct net_device *dev;
805 read_lock(&dev_base_lock);
806 dev = __dev_get_by_name(name);
807 read_unlock(&dev_base_lock);
809 if (!dev && capable(CAP_SYS_MODULE))
810 request_module("%s", name);
813 static int default_rebuild_header(struct sk_buff *skb)
815 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
816 skb->dev ? skb->dev->name : "NULL!!!");
823 * dev_open - prepare an interface for use.
824 * @dev: device to open
826 * Takes a device from down to up state. The device's private open
827 * function is invoked and then the multicast lists are loaded. Finally
828 * the device is moved into the up state and a %NETDEV_UP message is
829 * sent to the netdev notifier chain.
831 * Calling this function on an active interface is a nop. On a failure
832 * a negative errno code is returned.
834 int dev_open(struct net_device *dev)
842 if (dev->flags & IFF_UP)
846 * Is it even present?
848 if (!netif_device_present(dev))
852 * Call device private open method
854 set_bit(__LINK_STATE_START, &dev->state);
856 ret = dev->open(dev);
858 clear_bit(__LINK_STATE_START, &dev->state);
862 * If it went open OK then:
869 dev->flags |= IFF_UP;
872 * Initialize multicasting status
877 * Wakeup transmit queue engine
882 * ... and announce new interface.
884 notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
890 * dev_close - shutdown an interface.
891 * @dev: device to shutdown
893 * This function moves an active device into down state. A
894 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
895 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
898 int dev_close(struct net_device *dev)
900 if (!(dev->flags & IFF_UP))
904 * Tell people we are going down, so that they can
905 * prepare to death, when device is still operating.
907 notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
911 clear_bit(__LINK_STATE_START, &dev->state);
913 /* Synchronize to scheduled poll. We cannot touch poll list,
914 * it can be even on different cpu. So just clear netif_running(),
915 * and wait when poll really will happen. Actually, the best place
916 * for this is inside dev->stop() after device stopped its irq
917 * engine, but this requires more changes in devices. */
919 smp_mb__after_clear_bit(); /* Commit netif_running(). */
920 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
922 current->state = TASK_INTERRUPTIBLE;
927 * Call the device specific close. This cannot fail.
928 * Only if device is UP
930 * We allow it to be called even after a DETACH hot-plug
937 * Device is now down.
940 dev->flags &= ~IFF_UP;
943 * Tell people we are down
945 notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
952 * Device change register/unregister. These are not inline or static
953 * as we export them to the world.
957 * register_netdevice_notifier - register a network notifier block
960 * Register a notifier to be called when network device events occur.
961 * The notifier passed is linked into the kernel structures and must
962 * not be reused until it has been unregistered. A negative errno code
963 * is returned on a failure.
965 * When registered all registration and up events are replayed
966 * to the new notifier to allow device to have a race free
967 * view of the network device list.
970 int register_netdevice_notifier(struct notifier_block *nb)
972 struct net_device *dev;
976 err = notifier_chain_register(&netdev_chain, nb);
978 for (dev = dev_base; dev; dev = dev->next) {
979 nb->notifier_call(nb, NETDEV_REGISTER, dev);
981 if (dev->flags & IFF_UP)
982 nb->notifier_call(nb, NETDEV_UP, dev);
990 * unregister_netdevice_notifier - unregister a network notifier block
993 * Unregister a notifier previously registered by
994 * register_netdevice_notifier(). The notifier is unlinked into the
995 * kernel structures and may then be reused. A negative errno code
996 * is returned on a failure.
999 int unregister_netdevice_notifier(struct notifier_block *nb)
1001 return notifier_chain_unregister(&netdev_chain, nb);
1005 * call_netdevice_notifiers - call all network notifier blocks
1006 * @val: value passed unmodified to notifier function
1007 * @v: pointer passed unmodified to notifier function
1009 * Call all network notifier blocks. Parameters and return value
1010 * are as for notifier_call_chain().
1013 int call_netdevice_notifiers(unsigned long val, void *v)
1015 return notifier_call_chain(&netdev_chain, val, v);
1018 /* When > 0 there are consumers of rx skb time stamps */
1019 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1021 void net_enable_timestamp(void)
1023 atomic_inc(&netstamp_needed);
1026 void net_disable_timestamp(void)
1028 atomic_dec(&netstamp_needed);
1031 static inline void net_timestamp(struct timeval *stamp)
1033 if (atomic_read(&netstamp_needed))
1034 do_gettimeofday(stamp);
1042 * Support routine. Sends outgoing frames to any network
1043 * taps currently in use.
1046 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1048 struct packet_type *ptype;
1049 net_timestamp(&skb->stamp);
1052 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1053 /* Never send packets back to the socket
1054 * they originated from - MvS (miquels@drinkel.ow.org)
1056 if ((ptype->dev == dev || !ptype->dev) &&
1057 (ptype->af_packet_priv == NULL ||
1058 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1059 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1063 /* skb->nh should be correctly
1064 set by sender, so that the second statement is
1065 just protection against buggy protocols.
1067 skb2->mac.raw = skb2->data;
1069 if (skb2->nh.raw < skb2->data ||
1070 skb2->nh.raw > skb2->tail) {
1071 if (net_ratelimit())
1072 printk(KERN_CRIT "protocol %04x is "
1074 skb2->protocol, dev->name);
1075 skb2->nh.raw = skb2->data;
1078 skb2->h.raw = skb2->nh.raw;
1079 skb2->pkt_type = PACKET_OUTGOING;
1080 ptype->func(skb2, skb->dev, ptype);
1087 * Invalidate hardware checksum when packet is to be mangled, and
1088 * complete checksum manually on outgoing path.
1090 int skb_checksum_help(struct sk_buff *skb, int inward)
1093 int ret = 0, offset = skb->h.raw - skb->data;
1096 skb->ip_summed = CHECKSUM_NONE;
1100 if (skb_cloned(skb)) {
1101 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1106 if (offset > (int)skb->len)
1108 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1110 offset = skb->tail - skb->h.raw;
1113 if (skb->csum + 2 > offset)
1116 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1117 skb->ip_summed = CHECKSUM_NONE;
1122 #ifdef CONFIG_HIGHMEM
1123 /* Actually, we should eliminate this check as soon as we know, that:
1124 * 1. IOMMU is present and allows to map all the memory.
1125 * 2. No high memory really exists on this machine.
1128 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1132 if (dev->features & NETIF_F_HIGHDMA)
1135 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1136 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1142 #define illegal_highdma(dev, skb) (0)
1145 extern void skb_release_data(struct sk_buff *);
1147 /* Keep head the same: replace data */
1148 int __skb_linearize(struct sk_buff *skb, int gfp_mask)
1153 struct skb_shared_info *ninfo;
1154 int headerlen = skb->data - skb->head;
1155 int expand = (skb->tail + skb->data_len) - skb->end;
1157 if (skb_shared(skb))
1163 size = skb->end - skb->head + expand;
1164 size = SKB_DATA_ALIGN(size);
1165 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
1169 /* Copy entire thing */
1170 if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
1174 ninfo = (struct skb_shared_info*)(data + size);
1175 atomic_set(&ninfo->dataref, 1);
1176 ninfo->tso_size = skb_shinfo(skb)->tso_size;
1177 ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
1178 ninfo->nr_frags = 0;
1179 ninfo->frag_list = NULL;
1181 /* Offset between the two in bytes */
1182 offset = data - skb->head;
1184 /* Free old data. */
1185 skb_release_data(skb);
1188 skb->end = data + size;
1190 /* Set up new pointers */
1191 skb->h.raw += offset;
1192 skb->nh.raw += offset;
1193 skb->mac.raw += offset;
1194 skb->tail += offset;
1195 skb->data += offset;
1197 /* We are no longer a clone, even if we were. */
1200 skb->tail += skb->data_len;
1205 #define HARD_TX_LOCK(dev, cpu) { \
1206 if ((dev->features & NETIF_F_LLTX) == 0) { \
1207 spin_lock(&dev->xmit_lock); \
1208 dev->xmit_lock_owner = cpu; \
1212 #define HARD_TX_UNLOCK(dev) { \
1213 if ((dev->features & NETIF_F_LLTX) == 0) { \
1214 dev->xmit_lock_owner = -1; \
1215 spin_unlock(&dev->xmit_lock); \
1220 * dev_queue_xmit - transmit a buffer
1221 * @skb: buffer to transmit
1223 * Queue a buffer for transmission to a network device. The caller must
1224 * have set the device and priority and built the buffer before calling
1225 * this function. The function can be called from an interrupt.
1227 * A negative errno code is returned on a failure. A success does not
1228 * guarantee the frame will be transmitted as it may be dropped due
1229 * to congestion or traffic shaping.
1231 * -----------------------------------------------------------------------------------
1232 * I notice this method can also return errors from the queue disciplines,
1233 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1236 * Regardless of the return value, the skb is consumed, so it is currently
1237 * difficult to retry a send to this method. (You can bump the ref count
1238 * before sending to hold a reference for retry if you are careful.)
1240 * When calling this method, interrupts MUST be enabled. This is because
1241 * the BH enable code must have IRQs enabled so that it will not deadlock.
1245 int dev_queue_xmit(struct sk_buff *skb)
1247 struct net_device *dev = skb->dev;
1251 if (skb_shinfo(skb)->frag_list &&
1252 !(dev->features & NETIF_F_FRAGLIST) &&
1253 __skb_linearize(skb, GFP_ATOMIC))
1256 /* Fragmented skb is linearized if device does not support SG,
1257 * or if at least one of fragments is in highmem and device
1258 * does not support DMA from it.
1260 if (skb_shinfo(skb)->nr_frags &&
1261 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1262 __skb_linearize(skb, GFP_ATOMIC))
1265 /* If packet is not checksummed and device does not support
1266 * checksumming for this protocol, complete checksumming here.
1268 if (skb->ip_summed == CHECKSUM_HW &&
1269 (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
1270 (!(dev->features & NETIF_F_IP_CSUM) ||
1271 skb->protocol != htons(ETH_P_IP))))
1272 if (skb_checksum_help(skb, 0))
1275 /* Disable soft irqs for various locks below. Also
1276 * stops preemption for RCU.
1280 /* Updates of qdisc are serialized by queue_lock.
1281 * The struct Qdisc which is pointed to by qdisc is now a
1282 * rcu structure - it may be accessed without acquiring
1283 * a lock (but the structure may be stale.) The freeing of the
1284 * qdisc will be deferred until it's known that there are no
1285 * more references to it.
1287 * If the qdisc has an enqueue function, we still need to
1288 * hold the queue_lock before calling it, since queue_lock
1289 * also serializes access to the device queue.
1292 q = rcu_dereference(dev->qdisc);
1293 #ifdef CONFIG_NET_CLS_ACT
1294 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1297 /* Grab device queue */
1298 spin_lock(&dev->queue_lock);
1300 rc = q->enqueue(skb, q);
1304 spin_unlock(&dev->queue_lock);
1305 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1309 /* The device has no queue. Common case for software devices:
1310 loopback, all the sorts of tunnels...
1312 Really, it is unlikely that xmit_lock protection is necessary here.
1313 (f.e. loopback and IP tunnels are clean ignoring statistics
1315 However, it is possible, that they rely on protection
1318 Check this and shot the lock. It is not prone from deadlocks.
1319 Either shot noqueue qdisc, it is even simpler 8)
1321 if (dev->flags & IFF_UP) {
1322 int cpu = smp_processor_id(); /* ok because BHs are off */
1324 if (dev->xmit_lock_owner != cpu) {
1326 HARD_TX_LOCK(dev, cpu);
1328 if (!netif_queue_stopped(dev)) {
1330 dev_queue_xmit_nit(skb, dev);
1333 if (!dev->hard_start_xmit(skb, dev)) {
1334 HARD_TX_UNLOCK(dev);
1338 HARD_TX_UNLOCK(dev);
1339 if (net_ratelimit())
1340 printk(KERN_CRIT "Virtual device %s asks to "
1341 "queue packet!\n", dev->name);
1343 /* Recursion is detected! It is possible,
1345 if (net_ratelimit())
1346 printk(KERN_CRIT "Dead loop on virtual device "
1347 "%s, fix it urgently!\n", dev->name);
1363 /*=======================================================================
1365 =======================================================================*/
1367 int netdev_max_backlog = 300;
1368 int weight_p = 64; /* old backlog weight */
1369 /* These numbers are selected based on intuition and some
1370 * experimentatiom, if you have more scientific way of doing this
1371 * please go ahead and fix things.
1373 int no_cong_thresh = 10;
1378 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1381 static void get_sample_stats(int cpu)
1387 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
1388 int blog = sd->input_pkt_queue.qlen;
1389 int avg_blog = sd->avg_blog;
1391 avg_blog = (avg_blog >> 1) + (blog >> 1);
1393 if (avg_blog > mod_cong) {
1394 /* Above moderate congestion levels. */
1395 sd->cng_level = NET_RX_CN_HIGH;
1398 rq = rd % netdev_max_backlog;
1399 if (rq < avg_blog) /* unlucky bastard */
1400 sd->cng_level = NET_RX_DROP;
1402 } else if (avg_blog > lo_cong) {
1403 sd->cng_level = NET_RX_CN_MOD;
1406 rq = rd % netdev_max_backlog;
1407 if (rq < avg_blog) /* unlucky bastard */
1408 sd->cng_level = NET_RX_CN_HIGH;
1410 } else if (avg_blog > no_cong)
1411 sd->cng_level = NET_RX_CN_LOW;
1412 else /* no congestion */
1413 sd->cng_level = NET_RX_SUCCESS;
1415 sd->avg_blog = avg_blog;
1418 #ifdef OFFLINE_SAMPLE
1419 static void sample_queue(unsigned long dummy)
1421 /* 10 ms 0r 1ms -- i don't care -- JHS */
1423 int cpu = smp_processor_id();
1425 get_sample_stats(cpu);
1426 next_tick += jiffies;
1427 mod_timer(&samp_timer, next_tick);
1433 * netif_rx - post buffer to the network code
1434 * @skb: buffer to post
1436 * This function receives a packet from a device driver and queues it for
1437 * the upper (protocol) levels to process. It always succeeds. The buffer
1438 * may be dropped during processing for congestion control or by the
1442 * NET_RX_SUCCESS (no congestion)
1443 * NET_RX_CN_LOW (low congestion)
1444 * NET_RX_CN_MOD (moderate congestion)
1445 * NET_RX_CN_HIGH (high congestion)
1446 * NET_RX_DROP (packet was dropped)
1450 int netif_rx(struct sk_buff *skb)
1453 struct softnet_data *queue;
1454 unsigned long flags;
1456 /* if netpoll wants it, pretend we never saw it */
1457 if (netpoll_rx(skb))
1460 if (!skb->stamp.tv_sec)
1461 net_timestamp(&skb->stamp);
1464 * The code is rearranged so that the path is the most
1465 * short when CPU is congested, but is still operating.
1467 local_irq_save(flags);
1468 this_cpu = smp_processor_id();
1469 queue = &__get_cpu_var(softnet_data);
1471 __get_cpu_var(netdev_rx_stat).total++;
1472 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1473 if (queue->input_pkt_queue.qlen) {
1474 if (queue->throttle)
1479 __skb_queue_tail(&queue->input_pkt_queue, skb);
1480 #ifndef OFFLINE_SAMPLE
1481 get_sample_stats(this_cpu);
1483 local_irq_restore(flags);
1484 return queue->cng_level;
1487 if (queue->throttle)
1488 queue->throttle = 0;
1490 netif_rx_schedule(&queue->backlog_dev);
1494 if (!queue->throttle) {
1495 queue->throttle = 1;
1496 __get_cpu_var(netdev_rx_stat).throttled++;
1500 __get_cpu_var(netdev_rx_stat).dropped++;
1501 local_irq_restore(flags);
1507 int netif_rx_ni(struct sk_buff *skb)
1512 err = netif_rx(skb);
1513 if (local_softirq_pending())
1520 EXPORT_SYMBOL(netif_rx_ni);
1522 static __inline__ void skb_bond(struct sk_buff *skb)
1524 struct net_device *dev = skb->dev;
1527 skb->real_dev = skb->dev;
1528 skb->dev = dev->master;
1532 static void net_tx_action(struct softirq_action *h)
1534 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1536 if (sd->completion_queue) {
1537 struct sk_buff *clist;
1539 local_irq_disable();
1540 clist = sd->completion_queue;
1541 sd->completion_queue = NULL;
1545 struct sk_buff *skb = clist;
1546 clist = clist->next;
1548 BUG_TRAP(!atomic_read(&skb->users));
1553 if (sd->output_queue) {
1554 struct net_device *head;
1556 local_irq_disable();
1557 head = sd->output_queue;
1558 sd->output_queue = NULL;
1562 struct net_device *dev = head;
1563 head = head->next_sched;
1565 smp_mb__before_clear_bit();
1566 clear_bit(__LINK_STATE_SCHED, &dev->state);
1568 if (spin_trylock(&dev->queue_lock)) {
1570 spin_unlock(&dev->queue_lock);
1572 netif_schedule(dev);
1578 static __inline__ int deliver_skb(struct sk_buff *skb,
1579 struct packet_type *pt_prev)
1581 atomic_inc(&skb->users);
1582 return pt_prev->func(skb, skb->dev, pt_prev);
1585 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1586 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1588 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1589 unsigned char *addr);
1590 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1592 static __inline__ int handle_bridge(struct sk_buff **pskb,
1593 struct packet_type **pt_prev, int *ret)
1595 struct net_bridge_port *port;
1597 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1598 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1602 *ret = deliver_skb(*pskb, *pt_prev);
1606 return br_handle_frame_hook(port, pskb);
1609 #define handle_bridge(skb, pt_prev, ret) (0)
1612 #ifdef CONFIG_NET_CLS_ACT
1613 /* TODO: Maybe we should just force sch_ingress to be compiled in
1614 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1615 * a compare and 2 stores extra right now if we dont have it on
1616 * but have CONFIG_NET_CLS_ACT
1617 * NOTE: This doesnt stop any functionality; if you dont have
1618 * the ingress scheduler, you just cant add policies on ingress.
1621 static int ing_filter(struct sk_buff *skb)
1624 struct net_device *dev = skb->dev;
1625 int result = TC_ACT_OK;
1627 if (dev->qdisc_ingress) {
1628 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1629 if (MAX_RED_LOOP < ttl++) {
1630 printk("Redir loop detected Dropping packet (%s->%s)\n",
1631 skb->input_dev?skb->input_dev->name:"??",skb->dev->name);
1635 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1637 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1638 if (NULL == skb->input_dev) {
1639 skb->input_dev = skb->dev;
1640 printk("ing_filter: fixed %s out %s\n",skb->input_dev->name,skb->dev->name);
1642 spin_lock(&dev->ingress_lock);
1643 if ((q = dev->qdisc_ingress) != NULL)
1644 result = q->enqueue(skb, q);
1645 spin_unlock(&dev->ingress_lock);
1653 int netif_receive_skb(struct sk_buff *skb)
1655 struct packet_type *ptype, *pt_prev;
1656 int ret = NET_RX_DROP;
1657 unsigned short type;
1659 /* if we've gotten here through NAPI, check netpoll */
1660 if (skb->dev->poll && netpoll_rx(skb))
1663 if (!skb->stamp.tv_sec)
1664 net_timestamp(&skb->stamp);
1668 __get_cpu_var(netdev_rx_stat).total++;
1670 skb->h.raw = skb->nh.raw = skb->data;
1671 skb->mac_len = skb->nh.raw - skb->mac.raw;
1677 #ifdef CONFIG_NET_CLS_ACT
1678 if (skb->tc_verd & TC_NCLS) {
1679 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1684 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1685 if (!ptype->dev || ptype->dev == skb->dev) {
1687 ret = deliver_skb(skb, pt_prev);
1692 #ifdef CONFIG_NET_CLS_ACT
1694 ret = deliver_skb(skb, pt_prev);
1695 pt_prev = NULL; /* noone else should process this after*/
1697 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1700 ret = ing_filter(skb);
1702 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1711 handle_diverter(skb);
1713 if (handle_bridge(&skb, &pt_prev, &ret))
1716 type = skb->protocol;
1717 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1718 if (ptype->type == type &&
1719 (!ptype->dev || ptype->dev == skb->dev)) {
1721 ret = deliver_skb(skb, pt_prev);
1727 ret = pt_prev->func(skb, skb->dev, pt_prev);
1730 /* Jamal, now you will not able to escape explaining
1731 * me how you were going to use this. :-)
1741 static int process_backlog(struct net_device *backlog_dev, int *budget)
1744 int quota = min(backlog_dev->quota, *budget);
1745 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1746 unsigned long start_time = jiffies;
1748 backlog_dev->weight = weight_p;
1750 struct sk_buff *skb;
1751 struct net_device *dev;
1753 local_irq_disable();
1754 skb = __skb_dequeue(&queue->input_pkt_queue);
1761 netif_receive_skb(skb);
1767 if (work >= quota || jiffies - start_time > 1)
1772 backlog_dev->quota -= work;
1777 backlog_dev->quota -= work;
1780 list_del(&backlog_dev->poll_list);
1781 smp_mb__before_clear_bit();
1782 netif_poll_enable(backlog_dev);
1784 if (queue->throttle)
1785 queue->throttle = 0;
1790 static void net_rx_action(struct softirq_action *h)
1792 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1793 unsigned long start_time = jiffies;
1794 int budget = netdev_max_backlog;
1797 local_irq_disable();
1799 while (!list_empty(&queue->poll_list)) {
1800 struct net_device *dev;
1802 if (budget <= 0 || jiffies - start_time > 1)
1807 dev = list_entry(queue->poll_list.next,
1808 struct net_device, poll_list);
1809 netpoll_poll_lock(dev);
1811 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1812 netpoll_poll_unlock(dev);
1813 local_irq_disable();
1814 list_del(&dev->poll_list);
1815 list_add_tail(&dev->poll_list, &queue->poll_list);
1817 dev->quota += dev->weight;
1819 dev->quota = dev->weight;
1821 netpoll_poll_unlock(dev);
1823 local_irq_disable();
1831 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1832 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1836 static gifconf_func_t * gifconf_list [NPROTO];
1839 * register_gifconf - register a SIOCGIF handler
1840 * @family: Address family
1841 * @gifconf: Function handler
1843 * Register protocol dependent address dumping routines. The handler
1844 * that is passed must not be freed or reused until it has been replaced
1845 * by another handler.
1847 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1849 if (family >= NPROTO)
1851 gifconf_list[family] = gifconf;
1857 * Map an interface index to its name (SIOCGIFNAME)
1861 * We need this ioctl for efficient implementation of the
1862 * if_indextoname() function required by the IPv6 API. Without
1863 * it, we would have to search all the interfaces to find a
1867 static int dev_ifname(struct ifreq __user *arg)
1869 struct net_device *dev;
1873 * Fetch the caller's info block.
1876 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1879 read_lock(&dev_base_lock);
1880 dev = __dev_get_by_index(ifr.ifr_ifindex);
1882 read_unlock(&dev_base_lock);
1886 strcpy(ifr.ifr_name, dev->name);
1887 read_unlock(&dev_base_lock);
1889 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1895 * Perform a SIOCGIFCONF call. This structure will change
1896 * size eventually, and there is nothing I can do about it.
1897 * Thus we will need a 'compatibility mode'.
1900 static int dev_ifconf(char __user *arg)
1903 struct net_device *dev;
1910 * Fetch the caller's info block.
1913 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
1920 * Loop over the interfaces, and write an info block for each.
1924 for (dev = dev_base; dev; dev = dev->next) {
1925 if (vx_flags(VXF_HIDE_NETIF, 0) &&
1926 !dev_in_nx_info(dev, current->nx_info))
1928 for (i = 0; i < NPROTO; i++) {
1929 if (gifconf_list[i]) {
1932 done = gifconf_list[i](dev, NULL, 0);
1934 done = gifconf_list[i](dev, pos + total,
1944 * All done. Write the updated control block back to the caller.
1946 ifc.ifc_len = total;
1949 * Both BSD and Solaris return 0 here, so we do too.
1951 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
1954 #ifdef CONFIG_PROC_FS
1956 * This is invoked by the /proc filesystem handler to display a device
1959 static __inline__ struct net_device *dev_get_idx(loff_t pos)
1961 struct net_device *dev;
1964 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
1966 return i == pos ? dev : NULL;
1969 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
1971 read_lock(&dev_base_lock);
1972 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
1975 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1978 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
1981 void dev_seq_stop(struct seq_file *seq, void *v)
1983 read_unlock(&dev_base_lock);
1986 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
1988 struct nx_info *nxi = current->nx_info;
1990 if (vx_flags(VXF_HIDE_NETIF, 0) && !dev_in_nx_info(dev, nxi))
1992 if (dev->get_stats) {
1993 struct net_device_stats *stats = dev->get_stats(dev);
1995 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1996 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1997 dev->name, stats->rx_bytes, stats->rx_packets,
1999 stats->rx_dropped + stats->rx_missed_errors,
2000 stats->rx_fifo_errors,
2001 stats->rx_length_errors + stats->rx_over_errors +
2002 stats->rx_crc_errors + stats->rx_frame_errors,
2003 stats->rx_compressed, stats->multicast,
2004 stats->tx_bytes, stats->tx_packets,
2005 stats->tx_errors, stats->tx_dropped,
2006 stats->tx_fifo_errors, stats->collisions,
2007 stats->tx_carrier_errors +
2008 stats->tx_aborted_errors +
2009 stats->tx_window_errors +
2010 stats->tx_heartbeat_errors,
2011 stats->tx_compressed);
2013 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2017 * Called from the PROCfs module. This now uses the new arbitrary sized
2018 * /proc/net interface to create /proc/net/dev
2020 static int dev_seq_show(struct seq_file *seq, void *v)
2022 if (v == SEQ_START_TOKEN)
2023 seq_puts(seq, "Inter-| Receive "
2025 " face |bytes packets errs drop fifo frame "
2026 "compressed multicast|bytes packets errs "
2027 "drop fifo colls carrier compressed\n");
2029 dev_seq_printf_stats(seq, v);
2033 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2035 struct netif_rx_stats *rc = NULL;
2037 while (*pos < NR_CPUS)
2038 if (cpu_online(*pos)) {
2039 rc = &per_cpu(netdev_rx_stat, *pos);
2046 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2048 return softnet_get_online(pos);
2051 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2054 return softnet_get_online(pos);
2057 static void softnet_seq_stop(struct seq_file *seq, void *v)
2061 static int softnet_seq_show(struct seq_file *seq, void *v)
2063 struct netif_rx_stats *s = v;
2065 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2066 s->total, s->dropped, s->time_squeeze, s->throttled,
2067 s->fastroute_hit, s->fastroute_success, s->fastroute_defer,
2068 s->fastroute_deferred_out,
2070 s->fastroute_latency_reduction
2078 static struct seq_operations dev_seq_ops = {
2079 .start = dev_seq_start,
2080 .next = dev_seq_next,
2081 .stop = dev_seq_stop,
2082 .show = dev_seq_show,
2085 static int dev_seq_open(struct inode *inode, struct file *file)
2087 return seq_open(file, &dev_seq_ops);
2090 static struct file_operations dev_seq_fops = {
2091 .owner = THIS_MODULE,
2092 .open = dev_seq_open,
2094 .llseek = seq_lseek,
2095 .release = seq_release,
2098 static struct seq_operations softnet_seq_ops = {
2099 .start = softnet_seq_start,
2100 .next = softnet_seq_next,
2101 .stop = softnet_seq_stop,
2102 .show = softnet_seq_show,
2105 static int softnet_seq_open(struct inode *inode, struct file *file)
2107 return seq_open(file, &softnet_seq_ops);
2110 static struct file_operations softnet_seq_fops = {
2111 .owner = THIS_MODULE,
2112 .open = softnet_seq_open,
2114 .llseek = seq_lseek,
2115 .release = seq_release,
2119 extern int wireless_proc_init(void);
2121 #define wireless_proc_init() 0
2124 static int __init dev_proc_init(void)
2128 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2130 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2132 if (wireless_proc_init())
2138 proc_net_remove("softnet_stat");
2140 proc_net_remove("dev");
2144 #define dev_proc_init() 0
2145 #endif /* CONFIG_PROC_FS */
2149 * netdev_set_master - set up master/slave pair
2150 * @slave: slave device
2151 * @master: new master device
2153 * Changes the master device of the slave. Pass %NULL to break the
2154 * bonding. The caller must hold the RTNL semaphore. On a failure
2155 * a negative errno code is returned. On success the reference counts
2156 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2157 * function returns zero.
2159 int netdev_set_master(struct net_device *slave, struct net_device *master)
2161 struct net_device *old = slave->master;
2171 slave->master = master;
2179 slave->flags |= IFF_SLAVE;
2181 slave->flags &= ~IFF_SLAVE;
2183 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2188 * dev_set_promiscuity - update promiscuity count on a device
2192 * Add or remove promsicuity from a device. While the count in the device
2193 * remains above zero the interface remains promiscuous. Once it hits zero
2194 * the device reverts back to normal filtering operation. A negative inc
2195 * value is used to drop promiscuity on the device.
2197 void dev_set_promiscuity(struct net_device *dev, int inc)
2199 unsigned short old_flags = dev->flags;
2201 dev->flags |= IFF_PROMISC;
2202 if ((dev->promiscuity += inc) == 0)
2203 dev->flags &= ~IFF_PROMISC;
2204 if (dev->flags ^ old_flags) {
2206 printk(KERN_INFO "device %s %s promiscuous mode\n",
2207 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2213 * dev_set_allmulti - update allmulti count on a device
2217 * Add or remove reception of all multicast frames to a device. While the
2218 * count in the device remains above zero the interface remains listening
2219 * to all interfaces. Once it hits zero the device reverts back to normal
2220 * filtering operation. A negative @inc value is used to drop the counter
2221 * when releasing a resource needing all multicasts.
2224 void dev_set_allmulti(struct net_device *dev, int inc)
2226 unsigned short old_flags = dev->flags;
2228 dev->flags |= IFF_ALLMULTI;
2229 if ((dev->allmulti += inc) == 0)
2230 dev->flags &= ~IFF_ALLMULTI;
2231 if (dev->flags ^ old_flags)
2235 unsigned dev_get_flags(const struct net_device *dev)
2239 flags = (dev->flags & ~(IFF_PROMISC |
2242 (dev->gflags & (IFF_PROMISC |
2245 if (netif_running(dev) && netif_carrier_ok(dev))
2246 flags |= IFF_RUNNING;
2251 int dev_change_flags(struct net_device *dev, unsigned flags)
2254 int old_flags = dev->flags;
2257 * Set the flags on our device.
2260 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2261 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2263 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2267 * Load in the correct multicast list now the flags have changed.
2273 * Have we downed the interface. We handle IFF_UP ourselves
2274 * according to user attempts to set it, rather than blindly
2279 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2280 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2286 if (dev->flags & IFF_UP &&
2287 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2289 notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
2291 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2292 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2293 dev->gflags ^= IFF_PROMISC;
2294 dev_set_promiscuity(dev, inc);
2297 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2298 is important. Some (broken) drivers set IFF_PROMISC, when
2299 IFF_ALLMULTI is requested not asking us and not reporting.
2301 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2302 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2303 dev->gflags ^= IFF_ALLMULTI;
2304 dev_set_allmulti(dev, inc);
2307 if (old_flags ^ dev->flags)
2308 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2313 int dev_set_mtu(struct net_device *dev, int new_mtu)
2317 if (new_mtu == dev->mtu)
2320 /* MTU must be positive. */
2324 if (!netif_device_present(dev))
2328 if (dev->change_mtu)
2329 err = dev->change_mtu(dev, new_mtu);
2332 if (!err && dev->flags & IFF_UP)
2333 notifier_call_chain(&netdev_chain,
2334 NETDEV_CHANGEMTU, dev);
2338 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2342 if (!dev->set_mac_address)
2344 if (sa->sa_family != dev->type)
2346 if (!netif_device_present(dev))
2348 err = dev->set_mac_address(dev, sa);
2350 notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
2355 * Perform the SIOCxIFxxx calls.
2357 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2360 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2366 case SIOCGIFFLAGS: /* Get interface flags */
2367 ifr->ifr_flags = dev_get_flags(dev);
2370 case SIOCSIFFLAGS: /* Set interface flags */
2371 return dev_change_flags(dev, ifr->ifr_flags);
2373 case SIOCGIFMETRIC: /* Get the metric on the interface
2374 (currently unused) */
2375 ifr->ifr_metric = 0;
2378 case SIOCSIFMETRIC: /* Set the metric on the interface
2379 (currently unused) */
2382 case SIOCGIFMTU: /* Get the MTU of a device */
2383 ifr->ifr_mtu = dev->mtu;
2386 case SIOCSIFMTU: /* Set the MTU of a device */
2387 return dev_set_mtu(dev, ifr->ifr_mtu);
2391 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2393 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2394 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2395 ifr->ifr_hwaddr.sa_family = dev->type;
2399 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2401 case SIOCSIFHWBROADCAST:
2402 if (ifr->ifr_hwaddr.sa_family != dev->type)
2404 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2405 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2406 notifier_call_chain(&netdev_chain,
2407 NETDEV_CHANGEADDR, dev);
2411 ifr->ifr_map.mem_start = dev->mem_start;
2412 ifr->ifr_map.mem_end = dev->mem_end;
2413 ifr->ifr_map.base_addr = dev->base_addr;
2414 ifr->ifr_map.irq = dev->irq;
2415 ifr->ifr_map.dma = dev->dma;
2416 ifr->ifr_map.port = dev->if_port;
2420 if (dev->set_config) {
2421 if (!netif_device_present(dev))
2423 return dev->set_config(dev, &ifr->ifr_map);
2428 if (!dev->set_multicast_list ||
2429 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2431 if (!netif_device_present(dev))
2433 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2437 if (!dev->set_multicast_list ||
2438 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2440 if (!netif_device_present(dev))
2442 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2446 ifr->ifr_ifindex = dev->ifindex;
2450 ifr->ifr_qlen = dev->tx_queue_len;
2454 if (ifr->ifr_qlen < 0)
2456 dev->tx_queue_len = ifr->ifr_qlen;
2460 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2461 return dev_change_name(dev, ifr->ifr_newname);
2464 * Unknown or private ioctl
2468 if ((cmd >= SIOCDEVPRIVATE &&
2469 cmd <= SIOCDEVPRIVATE + 15) ||
2470 cmd == SIOCBONDENSLAVE ||
2471 cmd == SIOCBONDRELEASE ||
2472 cmd == SIOCBONDSETHWADDR ||
2473 cmd == SIOCBONDSLAVEINFOQUERY ||
2474 cmd == SIOCBONDINFOQUERY ||
2475 cmd == SIOCBONDCHANGEACTIVE ||
2476 cmd == SIOCGMIIPHY ||
2477 cmd == SIOCGMIIREG ||
2478 cmd == SIOCSMIIREG ||
2479 cmd == SIOCBRADDIF ||
2480 cmd == SIOCBRDELIF ||
2481 cmd == SIOCWANDEV) {
2483 if (dev->do_ioctl) {
2484 if (netif_device_present(dev))
2485 err = dev->do_ioctl(dev, ifr,
2498 * This function handles all "interface"-type I/O control requests. The actual
2499 * 'doing' part of this is dev_ifsioc above.
2503 * dev_ioctl - network device ioctl
2504 * @cmd: command to issue
2505 * @arg: pointer to a struct ifreq in user space
2507 * Issue ioctl functions to devices. This is normally called by the
2508 * user space syscall interfaces but can sometimes be useful for
2509 * other purposes. The return value is the return from the syscall if
2510 * positive or a negative errno code on error.
2513 int dev_ioctl(unsigned int cmd, void __user *arg)
2519 /* One special case: SIOCGIFCONF takes ifconf argument
2520 and requires shared lock, because it sleeps writing
2524 if (cmd == SIOCGIFCONF) {
2526 ret = dev_ifconf((char __user *) arg);
2530 if (cmd == SIOCGIFNAME)
2531 return dev_ifname((struct ifreq __user *)arg);
2533 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2536 ifr.ifr_name[IFNAMSIZ-1] = 0;
2538 colon = strchr(ifr.ifr_name, ':');
2543 * See which interface the caller is talking about.
2548 * These ioctl calls:
2549 * - can be done by all.
2550 * - atomic and do not require locking.
2561 dev_load(ifr.ifr_name);
2562 read_lock(&dev_base_lock);
2563 ret = dev_ifsioc(&ifr, cmd);
2564 read_unlock(&dev_base_lock);
2568 if (copy_to_user(arg, &ifr,
2569 sizeof(struct ifreq)))
2575 dev_load(ifr.ifr_name);
2577 ret = dev_ethtool(&ifr);
2582 if (copy_to_user(arg, &ifr,
2583 sizeof(struct ifreq)))
2589 * These ioctl calls:
2590 * - require superuser power.
2591 * - require strict serialization.
2597 if (!capable(CAP_NET_ADMIN))
2599 dev_load(ifr.ifr_name);
2601 ret = dev_ifsioc(&ifr, cmd);
2606 if (copy_to_user(arg, &ifr,
2607 sizeof(struct ifreq)))
2613 * These ioctl calls:
2614 * - require superuser power.
2615 * - require strict serialization.
2616 * - do not return a value
2626 case SIOCSIFHWBROADCAST:
2629 case SIOCBONDENSLAVE:
2630 case SIOCBONDRELEASE:
2631 case SIOCBONDSETHWADDR:
2632 case SIOCBONDSLAVEINFOQUERY:
2633 case SIOCBONDINFOQUERY:
2634 case SIOCBONDCHANGEACTIVE:
2637 if (!capable(CAP_NET_ADMIN))
2639 dev_load(ifr.ifr_name);
2641 ret = dev_ifsioc(&ifr, cmd);
2646 /* Get the per device memory space. We can add this but
2647 * currently do not support it */
2649 /* Set the per device memory buffer space.
2650 * Not applicable in our case */
2655 * Unknown or private ioctl.
2658 if (cmd == SIOCWANDEV ||
2659 (cmd >= SIOCDEVPRIVATE &&
2660 cmd <= SIOCDEVPRIVATE + 15)) {
2661 dev_load(ifr.ifr_name);
2663 ret = dev_ifsioc(&ifr, cmd);
2665 if (!ret && copy_to_user(arg, &ifr,
2666 sizeof(struct ifreq)))
2671 /* Take care of Wireless Extensions */
2672 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2673 /* If command is `set a parameter', or
2674 * `get the encoding parameters', check if
2675 * the user has the right to do it */
2676 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
2677 if (!capable(CAP_NET_ADMIN))
2680 dev_load(ifr.ifr_name);
2682 /* Follow me in net/core/wireless.c */
2683 ret = wireless_process_ioctl(&ifr, cmd);
2685 if (IW_IS_GET(cmd) &&
2686 copy_to_user(arg, &ifr,
2687 sizeof(struct ifreq)))
2691 #endif /* WIRELESS_EXT */
2698 * dev_new_index - allocate an ifindex
2700 * Returns a suitable unique value for a new device interface
2701 * number. The caller must hold the rtnl semaphore or the
2702 * dev_base_lock to be sure it remains unique.
2704 static int dev_new_index(void)
2710 if (!__dev_get_by_index(ifindex))
2715 static int dev_boot_phase = 1;
2717 /* Delayed registration/unregisteration */
2718 static DEFINE_SPINLOCK(net_todo_list_lock);
2719 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2721 static inline void net_set_todo(struct net_device *dev)
2723 spin_lock(&net_todo_list_lock);
2724 list_add_tail(&dev->todo_list, &net_todo_list);
2725 spin_unlock(&net_todo_list_lock);
2729 * register_netdevice - register a network device
2730 * @dev: device to register
2732 * Take a completed network device structure and add it to the kernel
2733 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2734 * chain. 0 is returned on success. A negative errno code is returned
2735 * on a failure to set up the device, or if the name is a duplicate.
2737 * Callers must hold the rtnl semaphore. You may want
2738 * register_netdev() instead of this.
2741 * The locking appears insufficient to guarantee two parallel registers
2742 * will not get the same name.
2745 int register_netdevice(struct net_device *dev)
2747 struct hlist_head *head;
2748 struct hlist_node *p;
2751 BUG_ON(dev_boot_phase);
2754 /* When net_device's are persistent, this will be fatal. */
2755 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2757 spin_lock_init(&dev->queue_lock);
2758 spin_lock_init(&dev->xmit_lock);
2759 dev->xmit_lock_owner = -1;
2760 #ifdef CONFIG_NET_CLS_ACT
2761 spin_lock_init(&dev->ingress_lock);
2764 ret = alloc_divert_blk(dev);
2770 /* Init, if this function is available */
2772 ret = dev->init(dev);
2780 if (!dev_valid_name(dev->name)) {
2785 dev->ifindex = dev_new_index();
2786 if (dev->iflink == -1)
2787 dev->iflink = dev->ifindex;
2789 /* Check for existence of name */
2790 head = dev_name_hash(dev->name);
2791 hlist_for_each(p, head) {
2792 struct net_device *d
2793 = hlist_entry(p, struct net_device, name_hlist);
2794 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2800 /* Fix illegal SG+CSUM combinations. */
2801 if ((dev->features & NETIF_F_SG) &&
2802 !(dev->features & (NETIF_F_IP_CSUM |
2804 NETIF_F_HW_CSUM))) {
2805 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2807 dev->features &= ~NETIF_F_SG;
2810 /* TSO requires that SG is present as well. */
2811 if ((dev->features & NETIF_F_TSO) &&
2812 !(dev->features & NETIF_F_SG)) {
2813 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2815 dev->features &= ~NETIF_F_TSO;
2819 * nil rebuild_header routine,
2820 * that should be never called and used as just bug trap.
2823 if (!dev->rebuild_header)
2824 dev->rebuild_header = default_rebuild_header;
2827 * Default initial state at registry is that the
2828 * device is present.
2831 set_bit(__LINK_STATE_PRESENT, &dev->state);
2834 dev_init_scheduler(dev);
2835 write_lock_bh(&dev_base_lock);
2837 dev_tail = &dev->next;
2838 hlist_add_head(&dev->name_hlist, head);
2839 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2841 dev->reg_state = NETREG_REGISTERING;
2842 write_unlock_bh(&dev_base_lock);
2844 /* Notify protocols, that a new device appeared. */
2845 notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2847 /* Finish registration after unlock */
2854 free_divert_blk(dev);
2859 * register_netdev - register a network device
2860 * @dev: device to register
2862 * Take a completed network device structure and add it to the kernel
2863 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2864 * chain. 0 is returned on success. A negative errno code is returned
2865 * on a failure to set up the device, or if the name is a duplicate.
2867 * This is a wrapper around register_netdev that takes the rtnl semaphore
2868 * and expands the device name if you passed a format string to
2871 int register_netdev(struct net_device *dev)
2878 * If the name is a format string the caller wants us to do a
2881 if (strchr(dev->name, '%')) {
2882 err = dev_alloc_name(dev, dev->name);
2888 * Back compatibility hook. Kill this one in 2.5
2890 if (dev->name[0] == 0 || dev->name[0] == ' ') {
2891 err = dev_alloc_name(dev, "eth%d");
2896 err = register_netdevice(dev);
2901 EXPORT_SYMBOL(register_netdev);
2904 * netdev_wait_allrefs - wait until all references are gone.
2906 * This is called when unregistering network devices.
2908 * Any protocol or device that holds a reference should register
2909 * for netdevice notification, and cleanup and put back the
2910 * reference if they receive an UNREGISTER event.
2911 * We can get stuck here if buggy protocols don't correctly
2914 static void netdev_wait_allrefs(struct net_device *dev)
2916 unsigned long rebroadcast_time, warning_time;
2918 rebroadcast_time = warning_time = jiffies;
2919 while (atomic_read(&dev->refcnt) != 0) {
2920 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
2923 /* Rebroadcast unregister notification */
2924 notifier_call_chain(&netdev_chain,
2925 NETDEV_UNREGISTER, dev);
2927 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
2929 /* We must not have linkwatch events
2930 * pending on unregister. If this
2931 * happens, we simply run the queue
2932 * unscheduled, resulting in a noop
2935 linkwatch_run_queue();
2940 rebroadcast_time = jiffies;
2945 if (time_after(jiffies, warning_time + 10 * HZ)) {
2946 printk(KERN_EMERG "unregister_netdevice: "
2947 "waiting for %s to become free. Usage "
2949 dev->name, atomic_read(&dev->refcnt));
2950 warning_time = jiffies;
2959 * register_netdevice(x1);
2960 * register_netdevice(x2);
2962 * unregister_netdevice(y1);
2963 * unregister_netdevice(y2);
2969 * We are invoked by rtnl_unlock() after it drops the semaphore.
2970 * This allows us to deal with problems:
2971 * 1) We can create/delete sysfs objects which invoke hotplug
2972 * without deadlocking with linkwatch via keventd.
2973 * 2) Since we run with the RTNL semaphore not held, we can sleep
2974 * safely in order to wait for the netdev refcnt to drop to zero.
2976 static DECLARE_MUTEX(net_todo_run_mutex);
2977 void netdev_run_todo(void)
2979 struct list_head list = LIST_HEAD_INIT(list);
2983 /* Need to guard against multiple cpu's getting out of order. */
2984 down(&net_todo_run_mutex);
2986 /* Not safe to do outside the semaphore. We must not return
2987 * until all unregister events invoked by the local processor
2988 * have been completed (either by this todo run, or one on
2991 if (list_empty(&net_todo_list))
2994 /* Snapshot list, allow later requests */
2995 spin_lock(&net_todo_list_lock);
2996 list_splice_init(&net_todo_list, &list);
2997 spin_unlock(&net_todo_list_lock);
2999 while (!list_empty(&list)) {
3000 struct net_device *dev
3001 = list_entry(list.next, struct net_device, todo_list);
3002 list_del(&dev->todo_list);
3004 switch(dev->reg_state) {
3005 case NETREG_REGISTERING:
3006 err = netdev_register_sysfs(dev);
3008 printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
3010 dev->reg_state = NETREG_REGISTERED;
3013 case NETREG_UNREGISTERING:
3014 netdev_unregister_sysfs(dev);
3015 dev->reg_state = NETREG_UNREGISTERED;
3017 netdev_wait_allrefs(dev);
3020 BUG_ON(atomic_read(&dev->refcnt));
3021 BUG_TRAP(!dev->ip_ptr);
3022 BUG_TRAP(!dev->ip6_ptr);
3023 BUG_TRAP(!dev->dn_ptr);
3026 /* It must be the very last action,
3027 * after this 'dev' may point to freed up memory.
3029 if (dev->destructor)
3030 dev->destructor(dev);
3034 printk(KERN_ERR "network todo '%s' but state %d\n",
3035 dev->name, dev->reg_state);
3041 up(&net_todo_run_mutex);
3045 * alloc_netdev - allocate network device
3046 * @sizeof_priv: size of private data to allocate space for
3047 * @name: device name format string
3048 * @setup: callback to initialize device
3050 * Allocates a struct net_device with private data area for driver use
3051 * and performs basic initialization.
3053 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3054 void (*setup)(struct net_device *))
3057 struct net_device *dev;
3060 /* ensure 32-byte alignment of both the device and private area */
3061 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3062 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3064 p = kmalloc(alloc_size, GFP_KERNEL);
3066 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3069 memset(p, 0, alloc_size);
3071 dev = (struct net_device *)
3072 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3073 dev->padded = (char *)dev - (char *)p;
3076 dev->priv = netdev_priv(dev);
3079 strcpy(dev->name, name);
3082 EXPORT_SYMBOL(alloc_netdev);
3085 * free_netdev - free network device
3088 * This function does the last stage of destroying an allocated device
3089 * interface. The reference to the device object is released.
3090 * If this is the last reference then it will be freed.
3092 void free_netdev(struct net_device *dev)
3095 /* Compatiablity with error handling in drivers */
3096 if (dev->reg_state == NETREG_UNINITIALIZED) {
3097 kfree((char *)dev - dev->padded);
3101 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3102 dev->reg_state = NETREG_RELEASED;
3104 /* will free via class release */
3105 class_device_put(&dev->class_dev);
3107 kfree((char *)dev - dev->padded);
3111 /* Synchronize with packet receive processing. */
3112 void synchronize_net(void)
3119 * unregister_netdevice - remove device from the kernel
3122 * This function shuts down a device interface and removes it
3123 * from the kernel tables. On success 0 is returned, on a failure
3124 * a negative errno code is returned.
3126 * Callers must hold the rtnl semaphore. You may want
3127 * unregister_netdev() instead of this.
3130 int unregister_netdevice(struct net_device *dev)
3132 struct net_device *d, **dp;
3134 BUG_ON(dev_boot_phase);
3137 /* Some devices call without registering for initialization unwind. */
3138 if (dev->reg_state == NETREG_UNINITIALIZED) {
3139 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3140 "was registered\n", dev->name, dev);
3144 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3146 /* If device is running, close it first. */
3147 if (dev->flags & IFF_UP)
3150 /* And unlink it from device chain. */
3151 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3153 write_lock_bh(&dev_base_lock);
3154 hlist_del(&dev->name_hlist);
3155 hlist_del(&dev->index_hlist);
3156 if (dev_tail == &dev->next)
3159 write_unlock_bh(&dev_base_lock);
3164 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3169 dev->reg_state = NETREG_UNREGISTERING;
3173 /* Shutdown queueing discipline. */
3177 /* Notify protocols, that we are about to destroy
3178 this device. They should clean all the things.
3180 notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3183 * Flush the multicast chain
3185 dev_mc_discard(dev);
3190 /* Notifier chain MUST detach us from master device. */
3191 BUG_TRAP(!dev->master);
3193 free_divert_blk(dev);
3195 /* Finish processing unregister after unlock */
3205 * unregister_netdev - remove device from the kernel
3208 * This function shuts down a device interface and removes it
3209 * from the kernel tables. On success 0 is returned, on a failure
3210 * a negative errno code is returned.
3212 * This is just a wrapper for unregister_netdevice that takes
3213 * the rtnl semaphore. In general you want to use this and not
3214 * unregister_netdevice.
3216 void unregister_netdev(struct net_device *dev)
3219 unregister_netdevice(dev);
3223 EXPORT_SYMBOL(unregister_netdev);
3225 #ifdef CONFIG_HOTPLUG_CPU
3226 static int dev_cpu_callback(struct notifier_block *nfb,
3227 unsigned long action,
3230 struct sk_buff **list_skb;
3231 struct net_device **list_net;
3232 struct sk_buff *skb;
3233 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3234 struct softnet_data *sd, *oldsd;
3236 if (action != CPU_DEAD)
3239 local_irq_disable();
3240 cpu = smp_processor_id();
3241 sd = &per_cpu(softnet_data, cpu);
3242 oldsd = &per_cpu(softnet_data, oldcpu);
3244 /* Find end of our completion_queue. */
3245 list_skb = &sd->completion_queue;
3247 list_skb = &(*list_skb)->next;
3248 /* Append completion queue from offline CPU. */
3249 *list_skb = oldsd->completion_queue;
3250 oldsd->completion_queue = NULL;
3252 /* Find end of our output_queue. */
3253 list_net = &sd->output_queue;
3255 list_net = &(*list_net)->next_sched;
3256 /* Append output queue from offline CPU. */
3257 *list_net = oldsd->output_queue;
3258 oldsd->output_queue = NULL;
3260 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3263 /* Process offline CPU's input_pkt_queue */
3264 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3269 #endif /* CONFIG_HOTPLUG_CPU */
3273 * Initialize the DEV module. At boot time this walks the device list and
3274 * unhooks any devices that fail to initialise (normally hardware not
3275 * present) and leaves us with a valid list of present and active devices.
3280 * This is called single threaded during boot, so no need
3281 * to take the rtnl semaphore.
3283 static int __init net_dev_init(void)
3285 int i, rc = -ENOMEM;
3287 BUG_ON(!dev_boot_phase);
3291 if (dev_proc_init())
3294 if (netdev_sysfs_init())
3297 INIT_LIST_HEAD(&ptype_all);
3298 for (i = 0; i < 16; i++)
3299 INIT_LIST_HEAD(&ptype_base[i]);
3301 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3302 INIT_HLIST_HEAD(&dev_name_head[i]);
3304 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3305 INIT_HLIST_HEAD(&dev_index_head[i]);
3308 * Initialise the packet receive queues.
3311 for (i = 0; i < NR_CPUS; i++) {
3312 struct softnet_data *queue;
3314 queue = &per_cpu(softnet_data, i);
3315 skb_queue_head_init(&queue->input_pkt_queue);
3316 queue->throttle = 0;
3317 queue->cng_level = 0;
3318 queue->avg_blog = 10; /* arbitrary non-zero */
3319 queue->completion_queue = NULL;
3320 INIT_LIST_HEAD(&queue->poll_list);
3321 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3322 queue->backlog_dev.weight = weight_p;
3323 queue->backlog_dev.poll = process_backlog;
3324 atomic_set(&queue->backlog_dev.refcnt, 1);
3327 #ifdef OFFLINE_SAMPLE
3328 samp_timer.expires = jiffies + (10 * HZ);
3329 add_timer(&samp_timer);
3334 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3335 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3337 hotcpu_notifier(dev_cpu_callback, 0);
3345 subsys_initcall(net_dev_init);
3347 EXPORT_SYMBOL(__dev_get_by_index);
3348 EXPORT_SYMBOL(__dev_get_by_name);
3349 EXPORT_SYMBOL(__dev_remove_pack);
3350 EXPORT_SYMBOL(__skb_linearize);
3351 EXPORT_SYMBOL(dev_add_pack);
3352 EXPORT_SYMBOL(dev_alloc_name);
3353 EXPORT_SYMBOL(dev_close);
3354 EXPORT_SYMBOL(dev_get_by_flags);
3355 EXPORT_SYMBOL(dev_get_by_index);
3356 EXPORT_SYMBOL(dev_get_by_name);
3357 EXPORT_SYMBOL(dev_ioctl);
3358 EXPORT_SYMBOL(dev_open);
3359 EXPORT_SYMBOL(dev_queue_xmit);
3360 EXPORT_SYMBOL(dev_remove_pack);
3361 EXPORT_SYMBOL(dev_set_allmulti);
3362 EXPORT_SYMBOL(dev_set_promiscuity);
3363 EXPORT_SYMBOL(dev_change_flags);
3364 EXPORT_SYMBOL(dev_set_mtu);
3365 EXPORT_SYMBOL(dev_set_mac_address);
3366 EXPORT_SYMBOL(free_netdev);
3367 EXPORT_SYMBOL(netdev_boot_setup_check);
3368 EXPORT_SYMBOL(netdev_set_master);
3369 EXPORT_SYMBOL(netdev_state_change);
3370 EXPORT_SYMBOL(netif_receive_skb);
3371 EXPORT_SYMBOL(netif_rx);
3372 EXPORT_SYMBOL(register_gifconf);
3373 EXPORT_SYMBOL(register_netdevice);
3374 EXPORT_SYMBOL(register_netdevice_notifier);
3375 EXPORT_SYMBOL(skb_checksum_help);
3376 EXPORT_SYMBOL(synchronize_net);
3377 EXPORT_SYMBOL(unregister_netdevice);
3378 EXPORT_SYMBOL(unregister_netdevice_notifier);
3379 EXPORT_SYMBOL(net_enable_timestamp);
3380 EXPORT_SYMBOL(net_disable_timestamp);
3381 EXPORT_SYMBOL(dev_get_flags);
3383 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3384 EXPORT_SYMBOL(br_handle_frame_hook);
3385 EXPORT_SYMBOL(br_fdb_get_hook);
3386 EXPORT_SYMBOL(br_fdb_put_hook);
3390 EXPORT_SYMBOL(dev_load);
3393 EXPORT_PER_CPU_SYMBOL(softnet_data);