[SK_BUFF]: Introduce skb_reset_transport_header(skb)
[linux-2.6/mini2440.git] / net / core / dev.c
blob99f15728d9cb2a6c14488832720b44b0760f708a
1 /*
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
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
14 * Additional Authors:
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>
22 * Changes:
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
34 * drivers
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
44 * call a packet.
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
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
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
58 * 1 device.
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
66 * the backlog queue.
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/mutex.h>
84 #include <linux/string.h>
85 #include <linux/mm.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/notifier.h>
94 #include <linux/skbuff.h>
95 #include <net/sock.h>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
101 #include <net/dst.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <linux/wireless.h>
113 #include <net/iw_handler.h>
114 #include <asm/current.h>
115 #include <linux/audit.h>
116 #include <linux/dmaengine.h>
117 #include <linux/err.h>
118 #include <linux/ctype.h>
121 * The list of packet types we will receive (as opposed to discard)
122 * and the routines to invoke.
124 * Why 16. Because with 16 the only overlap we get on a hash of the
125 * low nibble of the protocol value is RARP/SNAP/X.25.
127 * NOTE: That is no longer true with the addition of VLAN tags. Not
128 * sure which should go first, but I bet it won't make much
129 * difference if we are running VLANs. The good news is that
130 * this protocol won't be in the list unless compiled in, so
131 * the average user (w/out VLANs) will not be adversely affected.
132 * --BLG
134 * 0800 IP
135 * 8100 802.1Q VLAN
136 * 0001 802.3
137 * 0002 AX.25
138 * 0004 802.2
139 * 8035 RARP
140 * 0005 SNAP
141 * 0805 X.25
142 * 0806 ARP
143 * 8137 IPX
144 * 0009 Localtalk
145 * 86DD IPv6
148 static DEFINE_SPINLOCK(ptype_lock);
149 static struct list_head ptype_base[16] __read_mostly; /* 16 way hashed list */
150 static struct list_head ptype_all __read_mostly; /* Taps */
152 #ifdef CONFIG_NET_DMA
153 static struct dma_client *net_dma_client;
154 static unsigned int net_dma_count;
155 static spinlock_t net_dma_event_lock;
156 #endif
159 * The @dev_base list is protected by @dev_base_lock and the rtnl
160 * semaphore.
162 * Pure readers hold dev_base_lock for reading.
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
175 * semaphore held.
177 struct net_device *dev_base;
178 static struct net_device **dev_tail = &dev_base;
179 DEFINE_RWLOCK(dev_base_lock);
181 EXPORT_SYMBOL(dev_base);
182 EXPORT_SYMBOL(dev_base_lock);
184 #define NETDEV_HASHBITS 8
185 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
186 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
188 static inline struct hlist_head *dev_name_hash(const char *name)
190 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
194 static inline struct hlist_head *dev_index_hash(int ifindex)
196 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
200 * Our notifier list
203 static RAW_NOTIFIER_HEAD(netdev_chain);
206 * Device drivers call our routines to queue packets here. We empty the
207 * queue in the local softnet handler.
209 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
211 #ifdef CONFIG_SYSFS
212 extern int netdev_sysfs_init(void);
213 extern int netdev_register_sysfs(struct net_device *);
214 extern void netdev_unregister_sysfs(struct net_device *);
215 #else
216 #define netdev_sysfs_init() (0)
217 #define netdev_register_sysfs(dev) (0)
218 #define netdev_unregister_sysfs(dev) do { } while(0)
219 #endif
222 /*******************************************************************************
224 Protocol management and registration routines
226 *******************************************************************************/
229 * For efficiency
232 static int netdev_nit;
235 * Add a protocol ID to the list. Now that the input handler is
236 * smarter we can dispense with all the messy stuff that used to be
237 * here.
239 * BEWARE!!! Protocol handlers, mangling input packets,
240 * MUST BE last in hash buckets and checking protocol handlers
241 * MUST start from promiscuous ptype_all chain in net_bh.
242 * It is true now, do not change it.
243 * Explanation follows: if protocol handler, mangling packet, will
244 * be the first on list, it is not able to sense, that packet
245 * is cloned and should be copied-on-write, so that it will
246 * change it and subsequent readers will get broken packet.
247 * --ANK (980803)
251 * dev_add_pack - add packet handler
252 * @pt: packet type declaration
254 * Add a protocol handler to the networking stack. The passed &packet_type
255 * is linked into kernel lists and may not be freed until it has been
256 * removed from the kernel lists.
258 * This call does not sleep therefore it can not
259 * guarantee all CPU's that are in middle of receiving packets
260 * will see the new packet type (until the next received packet).
263 void dev_add_pack(struct packet_type *pt)
265 int hash;
267 spin_lock_bh(&ptype_lock);
268 if (pt->type == htons(ETH_P_ALL)) {
269 netdev_nit++;
270 list_add_rcu(&pt->list, &ptype_all);
271 } else {
272 hash = ntohs(pt->type) & 15;
273 list_add_rcu(&pt->list, &ptype_base[hash]);
275 spin_unlock_bh(&ptype_lock);
279 * __dev_remove_pack - remove packet handler
280 * @pt: packet type declaration
282 * Remove a protocol handler that was previously added to the kernel
283 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
284 * from the kernel lists and can be freed or reused once this function
285 * returns.
287 * The packet type might still be in use by receivers
288 * and must not be freed until after all the CPU's have gone
289 * through a quiescent state.
291 void __dev_remove_pack(struct packet_type *pt)
293 struct list_head *head;
294 struct packet_type *pt1;
296 spin_lock_bh(&ptype_lock);
298 if (pt->type == htons(ETH_P_ALL)) {
299 netdev_nit--;
300 head = &ptype_all;
301 } else
302 head = &ptype_base[ntohs(pt->type) & 15];
304 list_for_each_entry(pt1, head, list) {
305 if (pt == pt1) {
306 list_del_rcu(&pt->list);
307 goto out;
311 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
312 out:
313 spin_unlock_bh(&ptype_lock);
316 * dev_remove_pack - remove packet handler
317 * @pt: packet type declaration
319 * Remove a protocol handler that was previously added to the kernel
320 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
321 * from the kernel lists and can be freed or reused once this function
322 * returns.
324 * This call sleeps to guarantee that no CPU is looking at the packet
325 * type after return.
327 void dev_remove_pack(struct packet_type *pt)
329 __dev_remove_pack(pt);
331 synchronize_net();
334 /******************************************************************************
336 Device Boot-time Settings Routines
338 *******************************************************************************/
340 /* Boot time configuration table */
341 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
344 * netdev_boot_setup_add - add new setup entry
345 * @name: name of the device
346 * @map: configured settings for the device
348 * Adds new setup entry to the dev_boot_setup list. The function
349 * returns 0 on error and 1 on success. This is a generic routine to
350 * all netdevices.
352 static int netdev_boot_setup_add(char *name, struct ifmap *map)
354 struct netdev_boot_setup *s;
355 int i;
357 s = dev_boot_setup;
358 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
359 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
360 memset(s[i].name, 0, sizeof(s[i].name));
361 strcpy(s[i].name, name);
362 memcpy(&s[i].map, map, sizeof(s[i].map));
363 break;
367 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
371 * netdev_boot_setup_check - check boot time settings
372 * @dev: the netdevice
374 * Check boot time settings for the device.
375 * The found settings are set for the device to be used
376 * later in the device probing.
377 * Returns 0 if no settings found, 1 if they are.
379 int netdev_boot_setup_check(struct net_device *dev)
381 struct netdev_boot_setup *s = dev_boot_setup;
382 int i;
384 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
385 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
386 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
387 dev->irq = s[i].map.irq;
388 dev->base_addr = s[i].map.base_addr;
389 dev->mem_start = s[i].map.mem_start;
390 dev->mem_end = s[i].map.mem_end;
391 return 1;
394 return 0;
399 * netdev_boot_base - get address from boot time settings
400 * @prefix: prefix for network device
401 * @unit: id for network device
403 * Check boot time settings for the base address of device.
404 * The found settings are set for the device to be used
405 * later in the device probing.
406 * Returns 0 if no settings found.
408 unsigned long netdev_boot_base(const char *prefix, int unit)
410 const struct netdev_boot_setup *s = dev_boot_setup;
411 char name[IFNAMSIZ];
412 int i;
414 sprintf(name, "%s%d", prefix, unit);
417 * If device already registered then return base of 1
418 * to indicate not to probe for this interface
420 if (__dev_get_by_name(name))
421 return 1;
423 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
424 if (!strcmp(name, s[i].name))
425 return s[i].map.base_addr;
426 return 0;
430 * Saves at boot time configured settings for any netdevice.
432 int __init netdev_boot_setup(char *str)
434 int ints[5];
435 struct ifmap map;
437 str = get_options(str, ARRAY_SIZE(ints), ints);
438 if (!str || !*str)
439 return 0;
441 /* Save settings */
442 memset(&map, 0, sizeof(map));
443 if (ints[0] > 0)
444 map.irq = ints[1];
445 if (ints[0] > 1)
446 map.base_addr = ints[2];
447 if (ints[0] > 2)
448 map.mem_start = ints[3];
449 if (ints[0] > 3)
450 map.mem_end = ints[4];
452 /* Add new entry to the list */
453 return netdev_boot_setup_add(str, &map);
456 __setup("netdev=", netdev_boot_setup);
458 /*******************************************************************************
460 Device Interface Subroutines
462 *******************************************************************************/
465 * __dev_get_by_name - find a device by its name
466 * @name: name to find
468 * Find an interface by name. Must be called under RTNL semaphore
469 * or @dev_base_lock. If the name is found a pointer to the device
470 * is returned. If the name is not found then %NULL is returned. The
471 * reference counters are not incremented so the caller must be
472 * careful with locks.
475 struct net_device *__dev_get_by_name(const char *name)
477 struct hlist_node *p;
479 hlist_for_each(p, dev_name_hash(name)) {
480 struct net_device *dev
481 = hlist_entry(p, struct net_device, name_hlist);
482 if (!strncmp(dev->name, name, IFNAMSIZ))
483 return dev;
485 return NULL;
489 * dev_get_by_name - find a device by its name
490 * @name: name to find
492 * Find an interface by name. This can be called from any
493 * context and does its own locking. The returned handle has
494 * the usage count incremented and the caller must use dev_put() to
495 * release it when it is no longer needed. %NULL is returned if no
496 * matching device is found.
499 struct net_device *dev_get_by_name(const char *name)
501 struct net_device *dev;
503 read_lock(&dev_base_lock);
504 dev = __dev_get_by_name(name);
505 if (dev)
506 dev_hold(dev);
507 read_unlock(&dev_base_lock);
508 return dev;
512 * __dev_get_by_index - find a device by its ifindex
513 * @ifindex: index of device
515 * Search for an interface by index. Returns %NULL if the device
516 * is not found or a pointer to the device. The device has not
517 * had its reference counter increased so the caller must be careful
518 * about locking. The caller must hold either the RTNL semaphore
519 * or @dev_base_lock.
522 struct net_device *__dev_get_by_index(int ifindex)
524 struct hlist_node *p;
526 hlist_for_each(p, dev_index_hash(ifindex)) {
527 struct net_device *dev
528 = hlist_entry(p, struct net_device, index_hlist);
529 if (dev->ifindex == ifindex)
530 return dev;
532 return NULL;
537 * dev_get_by_index - find a device by its ifindex
538 * @ifindex: index of device
540 * Search for an interface by index. Returns NULL if the device
541 * is not found or a pointer to the device. The device returned has
542 * had a reference added and the pointer is safe until the user calls
543 * dev_put to indicate they have finished with it.
546 struct net_device *dev_get_by_index(int ifindex)
548 struct net_device *dev;
550 read_lock(&dev_base_lock);
551 dev = __dev_get_by_index(ifindex);
552 if (dev)
553 dev_hold(dev);
554 read_unlock(&dev_base_lock);
555 return dev;
559 * dev_getbyhwaddr - find a device by its hardware address
560 * @type: media type of device
561 * @ha: hardware address
563 * Search for an interface by MAC address. Returns NULL if the device
564 * is not found or a pointer to the device. The caller must hold the
565 * rtnl semaphore. The returned device has not had its ref count increased
566 * and the caller must therefore be careful about locking
568 * BUGS:
569 * If the API was consistent this would be __dev_get_by_hwaddr
572 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
574 struct net_device *dev;
576 ASSERT_RTNL();
578 for (dev = dev_base; dev; dev = dev->next)
579 if (dev->type == type &&
580 !memcmp(dev->dev_addr, ha, dev->addr_len))
581 break;
582 return dev;
585 EXPORT_SYMBOL(dev_getbyhwaddr);
587 struct net_device *dev_getfirstbyhwtype(unsigned short type)
589 struct net_device *dev;
591 rtnl_lock();
592 for (dev = dev_base; dev; dev = dev->next) {
593 if (dev->type == type) {
594 dev_hold(dev);
595 break;
598 rtnl_unlock();
599 return dev;
602 EXPORT_SYMBOL(dev_getfirstbyhwtype);
605 * dev_get_by_flags - find any device with given flags
606 * @if_flags: IFF_* values
607 * @mask: bitmask of bits in if_flags to check
609 * Search for any interface with the given flags. Returns NULL if a device
610 * is not found or a pointer to the device. The device returned has
611 * had a reference added and the pointer is safe until the user calls
612 * dev_put to indicate they have finished with it.
615 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
617 struct net_device *dev;
619 read_lock(&dev_base_lock);
620 for (dev = dev_base; dev != NULL; dev = dev->next) {
621 if (((dev->flags ^ if_flags) & mask) == 0) {
622 dev_hold(dev);
623 break;
626 read_unlock(&dev_base_lock);
627 return dev;
631 * dev_valid_name - check if name is okay for network device
632 * @name: name string
634 * Network device names need to be valid file names to
635 * to allow sysfs to work. We also disallow any kind of
636 * whitespace.
638 int dev_valid_name(const char *name)
640 if (*name == '\0')
641 return 0;
642 if (strlen(name) >= IFNAMSIZ)
643 return 0;
644 if (!strcmp(name, ".") || !strcmp(name, ".."))
645 return 0;
647 while (*name) {
648 if (*name == '/' || isspace(*name))
649 return 0;
650 name++;
652 return 1;
656 * dev_alloc_name - allocate a name for a device
657 * @dev: device
658 * @name: name format string
660 * Passed a format string - eg "lt%d" it will try and find a suitable
661 * id. It scans list of devices to build up a free map, then chooses
662 * the first empty slot. The caller must hold the dev_base or rtnl lock
663 * while allocating the name and adding the device in order to avoid
664 * duplicates.
665 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
666 * Returns the number of the unit assigned or a negative errno code.
669 int dev_alloc_name(struct net_device *dev, const char *name)
671 int i = 0;
672 char buf[IFNAMSIZ];
673 const char *p;
674 const int max_netdevices = 8*PAGE_SIZE;
675 long *inuse;
676 struct net_device *d;
678 p = strnchr(name, IFNAMSIZ-1, '%');
679 if (p) {
681 * Verify the string as this thing may have come from
682 * the user. There must be either one "%d" and no other "%"
683 * characters.
685 if (p[1] != 'd' || strchr(p + 2, '%'))
686 return -EINVAL;
688 /* Use one page as a bit array of possible slots */
689 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
690 if (!inuse)
691 return -ENOMEM;
693 for (d = dev_base; d; d = d->next) {
694 if (!sscanf(d->name, name, &i))
695 continue;
696 if (i < 0 || i >= max_netdevices)
697 continue;
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))
702 set_bit(i, inuse);
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);
712 return i;
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.
719 return -ENFILE;
724 * dev_change_name - change name of a device
725 * @dev: device
726 * @newname: name (or format string) must be at least IFNAMSIZ
728 * Change name of a device, can pass format strings "eth%d".
729 * for wildcarding.
731 int dev_change_name(struct net_device *dev, char *newname)
733 int err = 0;
735 ASSERT_RTNL();
737 if (dev->flags & IFF_UP)
738 return -EBUSY;
740 if (!dev_valid_name(newname))
741 return -EINVAL;
743 if (strchr(newname, '%')) {
744 err = dev_alloc_name(dev, newname);
745 if (err < 0)
746 return err;
747 strcpy(newname, dev->name);
749 else if (__dev_get_by_name(newname))
750 return -EEXIST;
751 else
752 strlcpy(dev->name, newname, IFNAMSIZ);
754 device_rename(&dev->dev, dev->name);
755 hlist_del(&dev->name_hlist);
756 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
757 raw_notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
759 return err;
763 * netdev_features_change - device changes features
764 * @dev: device to cause notification
766 * Called to indicate a device has changed features.
768 void netdev_features_change(struct net_device *dev)
770 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
772 EXPORT_SYMBOL(netdev_features_change);
775 * netdev_state_change - device changes state
776 * @dev: device to cause notification
778 * Called to indicate a device has changed state. This function calls
779 * the notifier chains for netdev_chain and sends a NEWLINK message
780 * to the routing socket.
782 void netdev_state_change(struct net_device *dev)
784 if (dev->flags & IFF_UP) {
785 raw_notifier_call_chain(&netdev_chain,
786 NETDEV_CHANGE, dev);
787 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
792 * dev_load - load a network module
793 * @name: name of interface
795 * If a network interface is not present and the process has suitable
796 * privileges this function loads the module. If module loading is not
797 * available in this kernel then it becomes a nop.
800 void dev_load(const char *name)
802 struct net_device *dev;
804 read_lock(&dev_base_lock);
805 dev = __dev_get_by_name(name);
806 read_unlock(&dev_base_lock);
808 if (!dev && capable(CAP_SYS_MODULE))
809 request_module("%s", name);
812 static int default_rebuild_header(struct sk_buff *skb)
814 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
815 skb->dev ? skb->dev->name : "NULL!!!");
816 kfree_skb(skb);
817 return 1;
822 * dev_open - prepare an interface for use.
823 * @dev: device to open
825 * Takes a device from down to up state. The device's private open
826 * function is invoked and then the multicast lists are loaded. Finally
827 * the device is moved into the up state and a %NETDEV_UP message is
828 * sent to the netdev notifier chain.
830 * Calling this function on an active interface is a nop. On a failure
831 * a negative errno code is returned.
833 int dev_open(struct net_device *dev)
835 int ret = 0;
838 * Is it already up?
841 if (dev->flags & IFF_UP)
842 return 0;
845 * Is it even present?
847 if (!netif_device_present(dev))
848 return -ENODEV;
851 * Call device private open method
853 set_bit(__LINK_STATE_START, &dev->state);
854 if (dev->open) {
855 ret = dev->open(dev);
856 if (ret)
857 clear_bit(__LINK_STATE_START, &dev->state);
861 * If it went open OK then:
864 if (!ret) {
866 * Set the flags.
868 dev->flags |= IFF_UP;
871 * Initialize multicasting status
873 dev_mc_upload(dev);
876 * Wakeup transmit queue engine
878 dev_activate(dev);
881 * ... and announce new interface.
883 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
885 return ret;
889 * dev_close - shutdown an interface.
890 * @dev: device to shutdown
892 * This function moves an active device into down state. A
893 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
894 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
895 * chain.
897 int dev_close(struct net_device *dev)
899 if (!(dev->flags & IFF_UP))
900 return 0;
903 * Tell people we are going down, so that they can
904 * prepare to death, when device is still operating.
906 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
908 dev_deactivate(dev);
910 clear_bit(__LINK_STATE_START, &dev->state);
912 /* Synchronize to scheduled poll. We cannot touch poll list,
913 * it can be even on different cpu. So just clear netif_running(),
914 * and wait when poll really will happen. Actually, the best place
915 * for this is inside dev->stop() after device stopped its irq
916 * engine, but this requires more changes in devices. */
918 smp_mb__after_clear_bit(); /* Commit netif_running(). */
919 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
920 /* No hurry. */
921 msleep(1);
925 * Call the device specific close. This cannot fail.
926 * Only if device is UP
928 * We allow it to be called even after a DETACH hot-plug
929 * event.
931 if (dev->stop)
932 dev->stop(dev);
935 * Device is now down.
938 dev->flags &= ~IFF_UP;
941 * Tell people we are down
943 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
945 return 0;
950 * Device change register/unregister. These are not inline or static
951 * as we export them to the world.
955 * register_netdevice_notifier - register a network notifier block
956 * @nb: notifier
958 * Register a notifier to be called when network device events occur.
959 * The notifier passed is linked into the kernel structures and must
960 * not be reused until it has been unregistered. A negative errno code
961 * is returned on a failure.
963 * When registered all registration and up events are replayed
964 * to the new notifier to allow device to have a race free
965 * view of the network device list.
968 int register_netdevice_notifier(struct notifier_block *nb)
970 struct net_device *dev;
971 int err;
973 rtnl_lock();
974 err = raw_notifier_chain_register(&netdev_chain, nb);
975 if (!err) {
976 for (dev = dev_base; dev; dev = dev->next) {
977 nb->notifier_call(nb, NETDEV_REGISTER, dev);
979 if (dev->flags & IFF_UP)
980 nb->notifier_call(nb, NETDEV_UP, dev);
983 rtnl_unlock();
984 return err;
988 * unregister_netdevice_notifier - unregister a network notifier block
989 * @nb: notifier
991 * Unregister a notifier previously registered by
992 * register_netdevice_notifier(). The notifier is unlinked into the
993 * kernel structures and may then be reused. A negative errno code
994 * is returned on a failure.
997 int unregister_netdevice_notifier(struct notifier_block *nb)
999 int err;
1001 rtnl_lock();
1002 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1003 rtnl_unlock();
1004 return err;
1008 * call_netdevice_notifiers - call all network notifier blocks
1009 * @val: value passed unmodified to notifier function
1010 * @v: pointer passed unmodified to notifier function
1012 * Call all network notifier blocks. Parameters and return value
1013 * are as for raw_notifier_call_chain().
1016 int call_netdevice_notifiers(unsigned long val, void *v)
1018 return raw_notifier_call_chain(&netdev_chain, val, v);
1021 /* When > 0 there are consumers of rx skb time stamps */
1022 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1024 void net_enable_timestamp(void)
1026 atomic_inc(&netstamp_needed);
1029 void net_disable_timestamp(void)
1031 atomic_dec(&netstamp_needed);
1034 static inline void net_timestamp(struct sk_buff *skb)
1036 if (atomic_read(&netstamp_needed))
1037 __net_timestamp(skb);
1038 else
1039 skb->tstamp.tv64 = 0;
1043 * Support routine. Sends outgoing frames to any network
1044 * taps currently in use.
1047 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1049 struct packet_type *ptype;
1051 net_timestamp(skb);
1053 rcu_read_lock();
1054 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1055 /* Never send packets back to the socket
1056 * they originated from - MvS (miquels@drinkel.ow.org)
1058 if ((ptype->dev == dev || !ptype->dev) &&
1059 (ptype->af_packet_priv == NULL ||
1060 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1061 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1062 if (!skb2)
1063 break;
1065 /* skb->nh should be correctly
1066 set by sender, so that the second statement is
1067 just protection against buggy protocols.
1069 skb_reset_mac_header(skb2);
1071 if (skb_network_header(skb2) < skb2->data ||
1072 skb_network_header(skb2) > skb2->tail) {
1073 if (net_ratelimit())
1074 printk(KERN_CRIT "protocol %04x is "
1075 "buggy, dev %s\n",
1076 skb2->protocol, dev->name);
1077 skb_reset_network_header(skb2);
1080 skb2->h.raw = skb2->nh.raw;
1081 skb2->pkt_type = PACKET_OUTGOING;
1082 ptype->func(skb2, skb->dev, ptype, skb->dev);
1085 rcu_read_unlock();
1089 void __netif_schedule(struct net_device *dev)
1091 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1092 unsigned long flags;
1093 struct softnet_data *sd;
1095 local_irq_save(flags);
1096 sd = &__get_cpu_var(softnet_data);
1097 dev->next_sched = sd->output_queue;
1098 sd->output_queue = dev;
1099 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1100 local_irq_restore(flags);
1103 EXPORT_SYMBOL(__netif_schedule);
1105 void __netif_rx_schedule(struct net_device *dev)
1107 unsigned long flags;
1109 local_irq_save(flags);
1110 dev_hold(dev);
1111 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1112 if (dev->quota < 0)
1113 dev->quota += dev->weight;
1114 else
1115 dev->quota = dev->weight;
1116 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1117 local_irq_restore(flags);
1119 EXPORT_SYMBOL(__netif_rx_schedule);
1121 void dev_kfree_skb_any(struct sk_buff *skb)
1123 if (in_irq() || irqs_disabled())
1124 dev_kfree_skb_irq(skb);
1125 else
1126 dev_kfree_skb(skb);
1128 EXPORT_SYMBOL(dev_kfree_skb_any);
1131 /* Hot-plugging. */
1132 void netif_device_detach(struct net_device *dev)
1134 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1135 netif_running(dev)) {
1136 netif_stop_queue(dev);
1139 EXPORT_SYMBOL(netif_device_detach);
1141 void netif_device_attach(struct net_device *dev)
1143 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1144 netif_running(dev)) {
1145 netif_wake_queue(dev);
1146 __netdev_watchdog_up(dev);
1149 EXPORT_SYMBOL(netif_device_attach);
1153 * Invalidate hardware checksum when packet is to be mangled, and
1154 * complete checksum manually on outgoing path.
1156 int skb_checksum_help(struct sk_buff *skb)
1158 __wsum csum;
1159 int ret = 0, offset = skb->h.raw - skb->data;
1161 if (skb->ip_summed == CHECKSUM_COMPLETE)
1162 goto out_set_summed;
1164 if (unlikely(skb_shinfo(skb)->gso_size)) {
1165 /* Let GSO fix up the checksum. */
1166 goto out_set_summed;
1169 if (skb_cloned(skb)) {
1170 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1171 if (ret)
1172 goto out;
1175 BUG_ON(offset > (int)skb->len);
1176 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1178 offset = skb->tail - skb->h.raw;
1179 BUG_ON(offset <= 0);
1180 BUG_ON(skb->csum_offset + 2 > offset);
1182 *(__sum16*)(skb->h.raw + skb->csum_offset) = csum_fold(csum);
1184 out_set_summed:
1185 skb->ip_summed = CHECKSUM_NONE;
1186 out:
1187 return ret;
1191 * skb_gso_segment - Perform segmentation on skb.
1192 * @skb: buffer to segment
1193 * @features: features for the output path (see dev->features)
1195 * This function segments the given skb and returns a list of segments.
1197 * It may return NULL if the skb requires no segmentation. This is
1198 * only possible when GSO is used for verifying header integrity.
1200 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1202 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1203 struct packet_type *ptype;
1204 __be16 type = skb->protocol;
1205 int err;
1207 BUG_ON(skb_shinfo(skb)->frag_list);
1209 skb_reset_mac_header(skb);
1210 skb->mac_len = skb->nh.raw - skb->mac.raw;
1211 __skb_pull(skb, skb->mac_len);
1213 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1214 if (skb_header_cloned(skb) &&
1215 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1216 return ERR_PTR(err);
1219 rcu_read_lock();
1220 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1221 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1222 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1223 err = ptype->gso_send_check(skb);
1224 segs = ERR_PTR(err);
1225 if (err || skb_gso_ok(skb, features))
1226 break;
1227 __skb_push(skb, (skb->data -
1228 skb_network_header(skb)));
1230 segs = ptype->gso_segment(skb, features);
1231 break;
1234 rcu_read_unlock();
1236 __skb_push(skb, skb->data - skb_mac_header(skb));
1238 return segs;
1241 EXPORT_SYMBOL(skb_gso_segment);
1243 /* Take action when hardware reception checksum errors are detected. */
1244 #ifdef CONFIG_BUG
1245 void netdev_rx_csum_fault(struct net_device *dev)
1247 if (net_ratelimit()) {
1248 printk(KERN_ERR "%s: hw csum failure.\n",
1249 dev ? dev->name : "<unknown>");
1250 dump_stack();
1253 EXPORT_SYMBOL(netdev_rx_csum_fault);
1254 #endif
1256 /* Actually, we should eliminate this check as soon as we know, that:
1257 * 1. IOMMU is present and allows to map all the memory.
1258 * 2. No high memory really exists on this machine.
1261 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1263 #ifdef CONFIG_HIGHMEM
1264 int i;
1266 if (dev->features & NETIF_F_HIGHDMA)
1267 return 0;
1269 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1270 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1271 return 1;
1273 #endif
1274 return 0;
1277 struct dev_gso_cb {
1278 void (*destructor)(struct sk_buff *skb);
1281 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1283 static void dev_gso_skb_destructor(struct sk_buff *skb)
1285 struct dev_gso_cb *cb;
1287 do {
1288 struct sk_buff *nskb = skb->next;
1290 skb->next = nskb->next;
1291 nskb->next = NULL;
1292 kfree_skb(nskb);
1293 } while (skb->next);
1295 cb = DEV_GSO_CB(skb);
1296 if (cb->destructor)
1297 cb->destructor(skb);
1301 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1302 * @skb: buffer to segment
1304 * This function segments the given skb and stores the list of segments
1305 * in skb->next.
1307 static int dev_gso_segment(struct sk_buff *skb)
1309 struct net_device *dev = skb->dev;
1310 struct sk_buff *segs;
1311 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1312 NETIF_F_SG : 0);
1314 segs = skb_gso_segment(skb, features);
1316 /* Verifying header integrity only. */
1317 if (!segs)
1318 return 0;
1320 if (unlikely(IS_ERR(segs)))
1321 return PTR_ERR(segs);
1323 skb->next = segs;
1324 DEV_GSO_CB(skb)->destructor = skb->destructor;
1325 skb->destructor = dev_gso_skb_destructor;
1327 return 0;
1330 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1332 if (likely(!skb->next)) {
1333 if (netdev_nit)
1334 dev_queue_xmit_nit(skb, dev);
1336 if (netif_needs_gso(dev, skb)) {
1337 if (unlikely(dev_gso_segment(skb)))
1338 goto out_kfree_skb;
1339 if (skb->next)
1340 goto gso;
1343 return dev->hard_start_xmit(skb, dev);
1346 gso:
1347 do {
1348 struct sk_buff *nskb = skb->next;
1349 int rc;
1351 skb->next = nskb->next;
1352 nskb->next = NULL;
1353 rc = dev->hard_start_xmit(nskb, dev);
1354 if (unlikely(rc)) {
1355 nskb->next = skb->next;
1356 skb->next = nskb;
1357 return rc;
1359 if (unlikely(netif_queue_stopped(dev) && skb->next))
1360 return NETDEV_TX_BUSY;
1361 } while (skb->next);
1363 skb->destructor = DEV_GSO_CB(skb)->destructor;
1365 out_kfree_skb:
1366 kfree_skb(skb);
1367 return 0;
1370 #define HARD_TX_LOCK(dev, cpu) { \
1371 if ((dev->features & NETIF_F_LLTX) == 0) { \
1372 netif_tx_lock(dev); \
1376 #define HARD_TX_UNLOCK(dev) { \
1377 if ((dev->features & NETIF_F_LLTX) == 0) { \
1378 netif_tx_unlock(dev); \
1383 * dev_queue_xmit - transmit a buffer
1384 * @skb: buffer to transmit
1386 * Queue a buffer for transmission to a network device. The caller must
1387 * have set the device and priority and built the buffer before calling
1388 * this function. The function can be called from an interrupt.
1390 * A negative errno code is returned on a failure. A success does not
1391 * guarantee the frame will be transmitted as it may be dropped due
1392 * to congestion or traffic shaping.
1394 * -----------------------------------------------------------------------------------
1395 * I notice this method can also return errors from the queue disciplines,
1396 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1397 * be positive.
1399 * Regardless of the return value, the skb is consumed, so it is currently
1400 * difficult to retry a send to this method. (You can bump the ref count
1401 * before sending to hold a reference for retry if you are careful.)
1403 * When calling this method, interrupts MUST be enabled. This is because
1404 * the BH enable code must have IRQs enabled so that it will not deadlock.
1405 * --BLG
1408 int dev_queue_xmit(struct sk_buff *skb)
1410 struct net_device *dev = skb->dev;
1411 struct Qdisc *q;
1412 int rc = -ENOMEM;
1414 /* GSO will handle the following emulations directly. */
1415 if (netif_needs_gso(dev, skb))
1416 goto gso;
1418 if (skb_shinfo(skb)->frag_list &&
1419 !(dev->features & NETIF_F_FRAGLIST) &&
1420 __skb_linearize(skb))
1421 goto out_kfree_skb;
1423 /* Fragmented skb is linearized if device does not support SG,
1424 * or if at least one of fragments is in highmem and device
1425 * does not support DMA from it.
1427 if (skb_shinfo(skb)->nr_frags &&
1428 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1429 __skb_linearize(skb))
1430 goto out_kfree_skb;
1432 /* If packet is not checksummed and device does not support
1433 * checksumming for this protocol, complete checksumming here.
1435 if (skb->ip_summed == CHECKSUM_PARTIAL &&
1436 (!(dev->features & NETIF_F_GEN_CSUM) &&
1437 (!(dev->features & NETIF_F_IP_CSUM) ||
1438 skb->protocol != htons(ETH_P_IP))))
1439 if (skb_checksum_help(skb))
1440 goto out_kfree_skb;
1442 gso:
1443 spin_lock_prefetch(&dev->queue_lock);
1445 /* Disable soft irqs for various locks below. Also
1446 * stops preemption for RCU.
1448 rcu_read_lock_bh();
1450 /* Updates of qdisc are serialized by queue_lock.
1451 * The struct Qdisc which is pointed to by qdisc is now a
1452 * rcu structure - it may be accessed without acquiring
1453 * a lock (but the structure may be stale.) The freeing of the
1454 * qdisc will be deferred until it's known that there are no
1455 * more references to it.
1457 * If the qdisc has an enqueue function, we still need to
1458 * hold the queue_lock before calling it, since queue_lock
1459 * also serializes access to the device queue.
1462 q = rcu_dereference(dev->qdisc);
1463 #ifdef CONFIG_NET_CLS_ACT
1464 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1465 #endif
1466 if (q->enqueue) {
1467 /* Grab device queue */
1468 spin_lock(&dev->queue_lock);
1469 q = dev->qdisc;
1470 if (q->enqueue) {
1471 rc = q->enqueue(skb, q);
1472 qdisc_run(dev);
1473 spin_unlock(&dev->queue_lock);
1475 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1476 goto out;
1478 spin_unlock(&dev->queue_lock);
1481 /* The device has no queue. Common case for software devices:
1482 loopback, all the sorts of tunnels...
1484 Really, it is unlikely that netif_tx_lock protection is necessary
1485 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1486 counters.)
1487 However, it is possible, that they rely on protection
1488 made by us here.
1490 Check this and shot the lock. It is not prone from deadlocks.
1491 Either shot noqueue qdisc, it is even simpler 8)
1493 if (dev->flags & IFF_UP) {
1494 int cpu = smp_processor_id(); /* ok because BHs are off */
1496 if (dev->xmit_lock_owner != cpu) {
1498 HARD_TX_LOCK(dev, cpu);
1500 if (!netif_queue_stopped(dev)) {
1501 rc = 0;
1502 if (!dev_hard_start_xmit(skb, dev)) {
1503 HARD_TX_UNLOCK(dev);
1504 goto out;
1507 HARD_TX_UNLOCK(dev);
1508 if (net_ratelimit())
1509 printk(KERN_CRIT "Virtual device %s asks to "
1510 "queue packet!\n", dev->name);
1511 } else {
1512 /* Recursion is detected! It is possible,
1513 * unfortunately */
1514 if (net_ratelimit())
1515 printk(KERN_CRIT "Dead loop on virtual device "
1516 "%s, fix it urgently!\n", dev->name);
1520 rc = -ENETDOWN;
1521 rcu_read_unlock_bh();
1523 out_kfree_skb:
1524 kfree_skb(skb);
1525 return rc;
1526 out:
1527 rcu_read_unlock_bh();
1528 return rc;
1532 /*=======================================================================
1533 Receiver routines
1534 =======================================================================*/
1536 int netdev_max_backlog __read_mostly = 1000;
1537 int netdev_budget __read_mostly = 300;
1538 int weight_p __read_mostly = 64; /* old backlog weight */
1540 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1544 * netif_rx - post buffer to the network code
1545 * @skb: buffer to post
1547 * This function receives a packet from a device driver and queues it for
1548 * the upper (protocol) levels to process. It always succeeds. The buffer
1549 * may be dropped during processing for congestion control or by the
1550 * protocol layers.
1552 * return values:
1553 * NET_RX_SUCCESS (no congestion)
1554 * NET_RX_CN_LOW (low congestion)
1555 * NET_RX_CN_MOD (moderate congestion)
1556 * NET_RX_CN_HIGH (high congestion)
1557 * NET_RX_DROP (packet was dropped)
1561 int netif_rx(struct sk_buff *skb)
1563 struct softnet_data *queue;
1564 unsigned long flags;
1566 /* if netpoll wants it, pretend we never saw it */
1567 if (netpoll_rx(skb))
1568 return NET_RX_DROP;
1570 if (!skb->tstamp.tv64)
1571 net_timestamp(skb);
1574 * The code is rearranged so that the path is the most
1575 * short when CPU is congested, but is still operating.
1577 local_irq_save(flags);
1578 queue = &__get_cpu_var(softnet_data);
1580 __get_cpu_var(netdev_rx_stat).total++;
1581 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1582 if (queue->input_pkt_queue.qlen) {
1583 enqueue:
1584 dev_hold(skb->dev);
1585 __skb_queue_tail(&queue->input_pkt_queue, skb);
1586 local_irq_restore(flags);
1587 return NET_RX_SUCCESS;
1590 netif_rx_schedule(&queue->backlog_dev);
1591 goto enqueue;
1594 __get_cpu_var(netdev_rx_stat).dropped++;
1595 local_irq_restore(flags);
1597 kfree_skb(skb);
1598 return NET_RX_DROP;
1601 int netif_rx_ni(struct sk_buff *skb)
1603 int err;
1605 preempt_disable();
1606 err = netif_rx(skb);
1607 if (local_softirq_pending())
1608 do_softirq();
1609 preempt_enable();
1611 return err;
1614 EXPORT_SYMBOL(netif_rx_ni);
1616 static inline struct net_device *skb_bond(struct sk_buff *skb)
1618 struct net_device *dev = skb->dev;
1620 if (dev->master) {
1621 if (skb_bond_should_drop(skb)) {
1622 kfree_skb(skb);
1623 return NULL;
1625 skb->dev = dev->master;
1628 return dev;
1631 static void net_tx_action(struct softirq_action *h)
1633 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1635 if (sd->completion_queue) {
1636 struct sk_buff *clist;
1638 local_irq_disable();
1639 clist = sd->completion_queue;
1640 sd->completion_queue = NULL;
1641 local_irq_enable();
1643 while (clist) {
1644 struct sk_buff *skb = clist;
1645 clist = clist->next;
1647 BUG_TRAP(!atomic_read(&skb->users));
1648 __kfree_skb(skb);
1652 if (sd->output_queue) {
1653 struct net_device *head;
1655 local_irq_disable();
1656 head = sd->output_queue;
1657 sd->output_queue = NULL;
1658 local_irq_enable();
1660 while (head) {
1661 struct net_device *dev = head;
1662 head = head->next_sched;
1664 smp_mb__before_clear_bit();
1665 clear_bit(__LINK_STATE_SCHED, &dev->state);
1667 if (spin_trylock(&dev->queue_lock)) {
1668 qdisc_run(dev);
1669 spin_unlock(&dev->queue_lock);
1670 } else {
1671 netif_schedule(dev);
1677 static inline int deliver_skb(struct sk_buff *skb,
1678 struct packet_type *pt_prev,
1679 struct net_device *orig_dev)
1681 atomic_inc(&skb->users);
1682 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1685 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1686 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1687 struct net_bridge;
1688 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1689 unsigned char *addr);
1690 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1692 static __inline__ int handle_bridge(struct sk_buff **pskb,
1693 struct packet_type **pt_prev, int *ret,
1694 struct net_device *orig_dev)
1696 struct net_bridge_port *port;
1698 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1699 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1700 return 0;
1702 if (*pt_prev) {
1703 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1704 *pt_prev = NULL;
1707 return br_handle_frame_hook(port, pskb);
1709 #else
1710 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1711 #endif
1713 #ifdef CONFIG_NET_CLS_ACT
1714 /* TODO: Maybe we should just force sch_ingress to be compiled in
1715 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1716 * a compare and 2 stores extra right now if we dont have it on
1717 * but have CONFIG_NET_CLS_ACT
1718 * NOTE: This doesnt stop any functionality; if you dont have
1719 * the ingress scheduler, you just cant add policies on ingress.
1722 static int ing_filter(struct sk_buff *skb)
1724 struct Qdisc *q;
1725 struct net_device *dev = skb->dev;
1726 int result = TC_ACT_OK;
1728 if (dev->qdisc_ingress) {
1729 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1730 if (MAX_RED_LOOP < ttl++) {
1731 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1732 skb->iif, skb->dev->ifindex);
1733 return TC_ACT_SHOT;
1736 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1738 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1740 spin_lock(&dev->queue_lock);
1741 if ((q = dev->qdisc_ingress) != NULL)
1742 result = q->enqueue(skb, q);
1743 spin_unlock(&dev->queue_lock);
1747 return result;
1749 #endif
1751 int netif_receive_skb(struct sk_buff *skb)
1753 struct packet_type *ptype, *pt_prev;
1754 struct net_device *orig_dev;
1755 int ret = NET_RX_DROP;
1756 __be16 type;
1758 /* if we've gotten here through NAPI, check netpoll */
1759 if (skb->dev->poll && netpoll_rx(skb))
1760 return NET_RX_DROP;
1762 if (!skb->tstamp.tv64)
1763 net_timestamp(skb);
1765 if (!skb->iif)
1766 skb->iif = skb->dev->ifindex;
1768 orig_dev = skb_bond(skb);
1770 if (!orig_dev)
1771 return NET_RX_DROP;
1773 __get_cpu_var(netdev_rx_stat).total++;
1775 skb_reset_network_header(skb);
1776 skb_reset_transport_header(skb);
1777 skb->mac_len = skb->nh.raw - skb->mac.raw;
1779 pt_prev = NULL;
1781 rcu_read_lock();
1783 #ifdef CONFIG_NET_CLS_ACT
1784 if (skb->tc_verd & TC_NCLS) {
1785 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1786 goto ncls;
1788 #endif
1790 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1791 if (!ptype->dev || ptype->dev == skb->dev) {
1792 if (pt_prev)
1793 ret = deliver_skb(skb, pt_prev, orig_dev);
1794 pt_prev = ptype;
1798 #ifdef CONFIG_NET_CLS_ACT
1799 if (pt_prev) {
1800 ret = deliver_skb(skb, pt_prev, orig_dev);
1801 pt_prev = NULL; /* noone else should process this after*/
1802 } else {
1803 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1806 ret = ing_filter(skb);
1808 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1809 kfree_skb(skb);
1810 goto out;
1813 skb->tc_verd = 0;
1814 ncls:
1815 #endif
1817 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1818 goto out;
1820 type = skb->protocol;
1821 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1822 if (ptype->type == type &&
1823 (!ptype->dev || ptype->dev == skb->dev)) {
1824 if (pt_prev)
1825 ret = deliver_skb(skb, pt_prev, orig_dev);
1826 pt_prev = ptype;
1830 if (pt_prev) {
1831 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1832 } else {
1833 kfree_skb(skb);
1834 /* Jamal, now you will not able to escape explaining
1835 * me how you were going to use this. :-)
1837 ret = NET_RX_DROP;
1840 out:
1841 rcu_read_unlock();
1842 return ret;
1845 static int process_backlog(struct net_device *backlog_dev, int *budget)
1847 int work = 0;
1848 int quota = min(backlog_dev->quota, *budget);
1849 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1850 unsigned long start_time = jiffies;
1852 backlog_dev->weight = weight_p;
1853 for (;;) {
1854 struct sk_buff *skb;
1855 struct net_device *dev;
1857 local_irq_disable();
1858 skb = __skb_dequeue(&queue->input_pkt_queue);
1859 if (!skb)
1860 goto job_done;
1861 local_irq_enable();
1863 dev = skb->dev;
1865 netif_receive_skb(skb);
1867 dev_put(dev);
1869 work++;
1871 if (work >= quota || jiffies - start_time > 1)
1872 break;
1876 backlog_dev->quota -= work;
1877 *budget -= work;
1878 return -1;
1880 job_done:
1881 backlog_dev->quota -= work;
1882 *budget -= work;
1884 list_del(&backlog_dev->poll_list);
1885 smp_mb__before_clear_bit();
1886 netif_poll_enable(backlog_dev);
1888 local_irq_enable();
1889 return 0;
1892 static void net_rx_action(struct softirq_action *h)
1894 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1895 unsigned long start_time = jiffies;
1896 int budget = netdev_budget;
1897 void *have;
1899 local_irq_disable();
1901 while (!list_empty(&queue->poll_list)) {
1902 struct net_device *dev;
1904 if (budget <= 0 || jiffies - start_time > 1)
1905 goto softnet_break;
1907 local_irq_enable();
1909 dev = list_entry(queue->poll_list.next,
1910 struct net_device, poll_list);
1911 have = netpoll_poll_lock(dev);
1913 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1914 netpoll_poll_unlock(have);
1915 local_irq_disable();
1916 list_move_tail(&dev->poll_list, &queue->poll_list);
1917 if (dev->quota < 0)
1918 dev->quota += dev->weight;
1919 else
1920 dev->quota = dev->weight;
1921 } else {
1922 netpoll_poll_unlock(have);
1923 dev_put(dev);
1924 local_irq_disable();
1927 out:
1928 #ifdef CONFIG_NET_DMA
1930 * There may not be any more sk_buffs coming right now, so push
1931 * any pending DMA copies to hardware
1933 if (net_dma_client) {
1934 struct dma_chan *chan;
1935 rcu_read_lock();
1936 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1937 dma_async_memcpy_issue_pending(chan);
1938 rcu_read_unlock();
1940 #endif
1941 local_irq_enable();
1942 return;
1944 softnet_break:
1945 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1946 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1947 goto out;
1950 static gifconf_func_t * gifconf_list [NPROTO];
1953 * register_gifconf - register a SIOCGIF handler
1954 * @family: Address family
1955 * @gifconf: Function handler
1957 * Register protocol dependent address dumping routines. The handler
1958 * that is passed must not be freed or reused until it has been replaced
1959 * by another handler.
1961 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1963 if (family >= NPROTO)
1964 return -EINVAL;
1965 gifconf_list[family] = gifconf;
1966 return 0;
1971 * Map an interface index to its name (SIOCGIFNAME)
1975 * We need this ioctl for efficient implementation of the
1976 * if_indextoname() function required by the IPv6 API. Without
1977 * it, we would have to search all the interfaces to find a
1978 * match. --pb
1981 static int dev_ifname(struct ifreq __user *arg)
1983 struct net_device *dev;
1984 struct ifreq ifr;
1987 * Fetch the caller's info block.
1990 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1991 return -EFAULT;
1993 read_lock(&dev_base_lock);
1994 dev = __dev_get_by_index(ifr.ifr_ifindex);
1995 if (!dev) {
1996 read_unlock(&dev_base_lock);
1997 return -ENODEV;
2000 strcpy(ifr.ifr_name, dev->name);
2001 read_unlock(&dev_base_lock);
2003 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2004 return -EFAULT;
2005 return 0;
2009 * Perform a SIOCGIFCONF call. This structure will change
2010 * size eventually, and there is nothing I can do about it.
2011 * Thus we will need a 'compatibility mode'.
2014 static int dev_ifconf(char __user *arg)
2016 struct ifconf ifc;
2017 struct net_device *dev;
2018 char __user *pos;
2019 int len;
2020 int total;
2021 int i;
2024 * Fetch the caller's info block.
2027 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2028 return -EFAULT;
2030 pos = ifc.ifc_buf;
2031 len = ifc.ifc_len;
2034 * Loop over the interfaces, and write an info block for each.
2037 total = 0;
2038 for (dev = dev_base; dev; dev = dev->next) {
2039 for (i = 0; i < NPROTO; i++) {
2040 if (gifconf_list[i]) {
2041 int done;
2042 if (!pos)
2043 done = gifconf_list[i](dev, NULL, 0);
2044 else
2045 done = gifconf_list[i](dev, pos + total,
2046 len - total);
2047 if (done < 0)
2048 return -EFAULT;
2049 total += done;
2055 * All done. Write the updated control block back to the caller.
2057 ifc.ifc_len = total;
2060 * Both BSD and Solaris return 0 here, so we do too.
2062 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2065 #ifdef CONFIG_PROC_FS
2067 * This is invoked by the /proc filesystem handler to display a device
2068 * in detail.
2070 static struct net_device *dev_get_idx(loff_t pos)
2072 struct net_device *dev;
2073 loff_t i;
2075 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2077 return i == pos ? dev : NULL;
2080 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2082 read_lock(&dev_base_lock);
2083 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2086 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2088 ++*pos;
2089 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2092 void dev_seq_stop(struct seq_file *seq, void *v)
2094 read_unlock(&dev_base_lock);
2097 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2099 if (dev->get_stats) {
2100 struct net_device_stats *stats = dev->get_stats(dev);
2102 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2103 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2104 dev->name, stats->rx_bytes, stats->rx_packets,
2105 stats->rx_errors,
2106 stats->rx_dropped + stats->rx_missed_errors,
2107 stats->rx_fifo_errors,
2108 stats->rx_length_errors + stats->rx_over_errors +
2109 stats->rx_crc_errors + stats->rx_frame_errors,
2110 stats->rx_compressed, stats->multicast,
2111 stats->tx_bytes, stats->tx_packets,
2112 stats->tx_errors, stats->tx_dropped,
2113 stats->tx_fifo_errors, stats->collisions,
2114 stats->tx_carrier_errors +
2115 stats->tx_aborted_errors +
2116 stats->tx_window_errors +
2117 stats->tx_heartbeat_errors,
2118 stats->tx_compressed);
2119 } else
2120 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2124 * Called from the PROCfs module. This now uses the new arbitrary sized
2125 * /proc/net interface to create /proc/net/dev
2127 static int dev_seq_show(struct seq_file *seq, void *v)
2129 if (v == SEQ_START_TOKEN)
2130 seq_puts(seq, "Inter-| Receive "
2131 " | Transmit\n"
2132 " face |bytes packets errs drop fifo frame "
2133 "compressed multicast|bytes packets errs "
2134 "drop fifo colls carrier compressed\n");
2135 else
2136 dev_seq_printf_stats(seq, v);
2137 return 0;
2140 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2142 struct netif_rx_stats *rc = NULL;
2144 while (*pos < NR_CPUS)
2145 if (cpu_online(*pos)) {
2146 rc = &per_cpu(netdev_rx_stat, *pos);
2147 break;
2148 } else
2149 ++*pos;
2150 return rc;
2153 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2155 return softnet_get_online(pos);
2158 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2160 ++*pos;
2161 return softnet_get_online(pos);
2164 static void softnet_seq_stop(struct seq_file *seq, void *v)
2168 static int softnet_seq_show(struct seq_file *seq, void *v)
2170 struct netif_rx_stats *s = v;
2172 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2173 s->total, s->dropped, s->time_squeeze, 0,
2174 0, 0, 0, 0, /* was fastroute */
2175 s->cpu_collision );
2176 return 0;
2179 static const struct seq_operations dev_seq_ops = {
2180 .start = dev_seq_start,
2181 .next = dev_seq_next,
2182 .stop = dev_seq_stop,
2183 .show = dev_seq_show,
2186 static int dev_seq_open(struct inode *inode, struct file *file)
2188 return seq_open(file, &dev_seq_ops);
2191 static const struct file_operations dev_seq_fops = {
2192 .owner = THIS_MODULE,
2193 .open = dev_seq_open,
2194 .read = seq_read,
2195 .llseek = seq_lseek,
2196 .release = seq_release,
2199 static const struct seq_operations softnet_seq_ops = {
2200 .start = softnet_seq_start,
2201 .next = softnet_seq_next,
2202 .stop = softnet_seq_stop,
2203 .show = softnet_seq_show,
2206 static int softnet_seq_open(struct inode *inode, struct file *file)
2208 return seq_open(file, &softnet_seq_ops);
2211 static const struct file_operations softnet_seq_fops = {
2212 .owner = THIS_MODULE,
2213 .open = softnet_seq_open,
2214 .read = seq_read,
2215 .llseek = seq_lseek,
2216 .release = seq_release,
2219 static void *ptype_get_idx(loff_t pos)
2221 struct packet_type *pt = NULL;
2222 loff_t i = 0;
2223 int t;
2225 list_for_each_entry_rcu(pt, &ptype_all, list) {
2226 if (i == pos)
2227 return pt;
2228 ++i;
2231 for (t = 0; t < 16; t++) {
2232 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2233 if (i == pos)
2234 return pt;
2235 ++i;
2238 return NULL;
2241 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2243 rcu_read_lock();
2244 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2247 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2249 struct packet_type *pt;
2250 struct list_head *nxt;
2251 int hash;
2253 ++*pos;
2254 if (v == SEQ_START_TOKEN)
2255 return ptype_get_idx(0);
2257 pt = v;
2258 nxt = pt->list.next;
2259 if (pt->type == htons(ETH_P_ALL)) {
2260 if (nxt != &ptype_all)
2261 goto found;
2262 hash = 0;
2263 nxt = ptype_base[0].next;
2264 } else
2265 hash = ntohs(pt->type) & 15;
2267 while (nxt == &ptype_base[hash]) {
2268 if (++hash >= 16)
2269 return NULL;
2270 nxt = ptype_base[hash].next;
2272 found:
2273 return list_entry(nxt, struct packet_type, list);
2276 static void ptype_seq_stop(struct seq_file *seq, void *v)
2278 rcu_read_unlock();
2281 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2283 #ifdef CONFIG_KALLSYMS
2284 unsigned long offset = 0, symsize;
2285 const char *symname;
2286 char *modname;
2287 char namebuf[128];
2289 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2290 &modname, namebuf);
2292 if (symname) {
2293 char *delim = ":";
2295 if (!modname)
2296 modname = delim = "";
2297 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2298 symname, offset);
2299 return;
2301 #endif
2303 seq_printf(seq, "[%p]", sym);
2306 static int ptype_seq_show(struct seq_file *seq, void *v)
2308 struct packet_type *pt = v;
2310 if (v == SEQ_START_TOKEN)
2311 seq_puts(seq, "Type Device Function\n");
2312 else {
2313 if (pt->type == htons(ETH_P_ALL))
2314 seq_puts(seq, "ALL ");
2315 else
2316 seq_printf(seq, "%04x", ntohs(pt->type));
2318 seq_printf(seq, " %-8s ",
2319 pt->dev ? pt->dev->name : "");
2320 ptype_seq_decode(seq, pt->func);
2321 seq_putc(seq, '\n');
2324 return 0;
2327 static const struct seq_operations ptype_seq_ops = {
2328 .start = ptype_seq_start,
2329 .next = ptype_seq_next,
2330 .stop = ptype_seq_stop,
2331 .show = ptype_seq_show,
2334 static int ptype_seq_open(struct inode *inode, struct file *file)
2336 return seq_open(file, &ptype_seq_ops);
2339 static const struct file_operations ptype_seq_fops = {
2340 .owner = THIS_MODULE,
2341 .open = ptype_seq_open,
2342 .read = seq_read,
2343 .llseek = seq_lseek,
2344 .release = seq_release,
2348 #ifdef CONFIG_WIRELESS_EXT
2349 extern int wireless_proc_init(void);
2350 #else
2351 #define wireless_proc_init() 0
2352 #endif
2354 static int __init dev_proc_init(void)
2356 int rc = -ENOMEM;
2358 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2359 goto out;
2360 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2361 goto out_dev;
2362 if (!proc_net_fops_create("ptype", S_IRUGO, &ptype_seq_fops))
2363 goto out_dev2;
2365 if (wireless_proc_init())
2366 goto out_softnet;
2367 rc = 0;
2368 out:
2369 return rc;
2370 out_softnet:
2371 proc_net_remove("softnet_stat");
2372 out_dev2:
2373 proc_net_remove("ptype");
2374 out_dev:
2375 proc_net_remove("dev");
2376 goto out;
2378 #else
2379 #define dev_proc_init() 0
2380 #endif /* CONFIG_PROC_FS */
2384 * netdev_set_master - set up master/slave pair
2385 * @slave: slave device
2386 * @master: new master device
2388 * Changes the master device of the slave. Pass %NULL to break the
2389 * bonding. The caller must hold the RTNL semaphore. On a failure
2390 * a negative errno code is returned. On success the reference counts
2391 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2392 * function returns zero.
2394 int netdev_set_master(struct net_device *slave, struct net_device *master)
2396 struct net_device *old = slave->master;
2398 ASSERT_RTNL();
2400 if (master) {
2401 if (old)
2402 return -EBUSY;
2403 dev_hold(master);
2406 slave->master = master;
2408 synchronize_net();
2410 if (old)
2411 dev_put(old);
2413 if (master)
2414 slave->flags |= IFF_SLAVE;
2415 else
2416 slave->flags &= ~IFF_SLAVE;
2418 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2419 return 0;
2423 * dev_set_promiscuity - update promiscuity count on a device
2424 * @dev: device
2425 * @inc: modifier
2427 * Add or remove promiscuity from a device. While the count in the device
2428 * remains above zero the interface remains promiscuous. Once it hits zero
2429 * the device reverts back to normal filtering operation. A negative inc
2430 * value is used to drop promiscuity on the device.
2432 void dev_set_promiscuity(struct net_device *dev, int inc)
2434 unsigned short old_flags = dev->flags;
2436 if ((dev->promiscuity += inc) == 0)
2437 dev->flags &= ~IFF_PROMISC;
2438 else
2439 dev->flags |= IFF_PROMISC;
2440 if (dev->flags != old_flags) {
2441 dev_mc_upload(dev);
2442 printk(KERN_INFO "device %s %s promiscuous mode\n",
2443 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2444 "left");
2445 audit_log(current->audit_context, GFP_ATOMIC,
2446 AUDIT_ANOM_PROMISCUOUS,
2447 "dev=%s prom=%d old_prom=%d auid=%u",
2448 dev->name, (dev->flags & IFF_PROMISC),
2449 (old_flags & IFF_PROMISC),
2450 audit_get_loginuid(current->audit_context));
2455 * dev_set_allmulti - update allmulti count on a device
2456 * @dev: device
2457 * @inc: modifier
2459 * Add or remove reception of all multicast frames to a device. While the
2460 * count in the device remains above zero the interface remains listening
2461 * to all interfaces. Once it hits zero the device reverts back to normal
2462 * filtering operation. A negative @inc value is used to drop the counter
2463 * when releasing a resource needing all multicasts.
2466 void dev_set_allmulti(struct net_device *dev, int inc)
2468 unsigned short old_flags = dev->flags;
2470 dev->flags |= IFF_ALLMULTI;
2471 if ((dev->allmulti += inc) == 0)
2472 dev->flags &= ~IFF_ALLMULTI;
2473 if (dev->flags ^ old_flags)
2474 dev_mc_upload(dev);
2477 unsigned dev_get_flags(const struct net_device *dev)
2479 unsigned flags;
2481 flags = (dev->flags & ~(IFF_PROMISC |
2482 IFF_ALLMULTI |
2483 IFF_RUNNING |
2484 IFF_LOWER_UP |
2485 IFF_DORMANT)) |
2486 (dev->gflags & (IFF_PROMISC |
2487 IFF_ALLMULTI));
2489 if (netif_running(dev)) {
2490 if (netif_oper_up(dev))
2491 flags |= IFF_RUNNING;
2492 if (netif_carrier_ok(dev))
2493 flags |= IFF_LOWER_UP;
2494 if (netif_dormant(dev))
2495 flags |= IFF_DORMANT;
2498 return flags;
2501 int dev_change_flags(struct net_device *dev, unsigned flags)
2503 int ret;
2504 int old_flags = dev->flags;
2507 * Set the flags on our device.
2510 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2511 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2512 IFF_AUTOMEDIA)) |
2513 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2514 IFF_ALLMULTI));
2517 * Load in the correct multicast list now the flags have changed.
2520 dev_mc_upload(dev);
2523 * Have we downed the interface. We handle IFF_UP ourselves
2524 * according to user attempts to set it, rather than blindly
2525 * setting it.
2528 ret = 0;
2529 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2530 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2532 if (!ret)
2533 dev_mc_upload(dev);
2536 if (dev->flags & IFF_UP &&
2537 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2538 IFF_VOLATILE)))
2539 raw_notifier_call_chain(&netdev_chain,
2540 NETDEV_CHANGE, dev);
2542 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2543 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2544 dev->gflags ^= IFF_PROMISC;
2545 dev_set_promiscuity(dev, inc);
2548 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2549 is important. Some (broken) drivers set IFF_PROMISC, when
2550 IFF_ALLMULTI is requested not asking us and not reporting.
2552 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2553 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2554 dev->gflags ^= IFF_ALLMULTI;
2555 dev_set_allmulti(dev, inc);
2558 if (old_flags ^ dev->flags)
2559 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2561 return ret;
2564 int dev_set_mtu(struct net_device *dev, int new_mtu)
2566 int err;
2568 if (new_mtu == dev->mtu)
2569 return 0;
2571 /* MTU must be positive. */
2572 if (new_mtu < 0)
2573 return -EINVAL;
2575 if (!netif_device_present(dev))
2576 return -ENODEV;
2578 err = 0;
2579 if (dev->change_mtu)
2580 err = dev->change_mtu(dev, new_mtu);
2581 else
2582 dev->mtu = new_mtu;
2583 if (!err && dev->flags & IFF_UP)
2584 raw_notifier_call_chain(&netdev_chain,
2585 NETDEV_CHANGEMTU, dev);
2586 return err;
2589 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2591 int err;
2593 if (!dev->set_mac_address)
2594 return -EOPNOTSUPP;
2595 if (sa->sa_family != dev->type)
2596 return -EINVAL;
2597 if (!netif_device_present(dev))
2598 return -ENODEV;
2599 err = dev->set_mac_address(dev, sa);
2600 if (!err)
2601 raw_notifier_call_chain(&netdev_chain,
2602 NETDEV_CHANGEADDR, dev);
2603 return err;
2607 * Perform the SIOCxIFxxx calls.
2609 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2611 int err;
2612 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2614 if (!dev)
2615 return -ENODEV;
2617 switch (cmd) {
2618 case SIOCGIFFLAGS: /* Get interface flags */
2619 ifr->ifr_flags = dev_get_flags(dev);
2620 return 0;
2622 case SIOCSIFFLAGS: /* Set interface flags */
2623 return dev_change_flags(dev, ifr->ifr_flags);
2625 case SIOCGIFMETRIC: /* Get the metric on the interface
2626 (currently unused) */
2627 ifr->ifr_metric = 0;
2628 return 0;
2630 case SIOCSIFMETRIC: /* Set the metric on the interface
2631 (currently unused) */
2632 return -EOPNOTSUPP;
2634 case SIOCGIFMTU: /* Get the MTU of a device */
2635 ifr->ifr_mtu = dev->mtu;
2636 return 0;
2638 case SIOCSIFMTU: /* Set the MTU of a device */
2639 return dev_set_mtu(dev, ifr->ifr_mtu);
2641 case SIOCGIFHWADDR:
2642 if (!dev->addr_len)
2643 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2644 else
2645 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2646 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2647 ifr->ifr_hwaddr.sa_family = dev->type;
2648 return 0;
2650 case SIOCSIFHWADDR:
2651 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2653 case SIOCSIFHWBROADCAST:
2654 if (ifr->ifr_hwaddr.sa_family != dev->type)
2655 return -EINVAL;
2656 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2657 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2658 raw_notifier_call_chain(&netdev_chain,
2659 NETDEV_CHANGEADDR, dev);
2660 return 0;
2662 case SIOCGIFMAP:
2663 ifr->ifr_map.mem_start = dev->mem_start;
2664 ifr->ifr_map.mem_end = dev->mem_end;
2665 ifr->ifr_map.base_addr = dev->base_addr;
2666 ifr->ifr_map.irq = dev->irq;
2667 ifr->ifr_map.dma = dev->dma;
2668 ifr->ifr_map.port = dev->if_port;
2669 return 0;
2671 case SIOCSIFMAP:
2672 if (dev->set_config) {
2673 if (!netif_device_present(dev))
2674 return -ENODEV;
2675 return dev->set_config(dev, &ifr->ifr_map);
2677 return -EOPNOTSUPP;
2679 case SIOCADDMULTI:
2680 if (!dev->set_multicast_list ||
2681 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2682 return -EINVAL;
2683 if (!netif_device_present(dev))
2684 return -ENODEV;
2685 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2686 dev->addr_len, 1);
2688 case SIOCDELMULTI:
2689 if (!dev->set_multicast_list ||
2690 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2691 return -EINVAL;
2692 if (!netif_device_present(dev))
2693 return -ENODEV;
2694 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2695 dev->addr_len, 1);
2697 case SIOCGIFINDEX:
2698 ifr->ifr_ifindex = dev->ifindex;
2699 return 0;
2701 case SIOCGIFTXQLEN:
2702 ifr->ifr_qlen = dev->tx_queue_len;
2703 return 0;
2705 case SIOCSIFTXQLEN:
2706 if (ifr->ifr_qlen < 0)
2707 return -EINVAL;
2708 dev->tx_queue_len = ifr->ifr_qlen;
2709 return 0;
2711 case SIOCSIFNAME:
2712 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2713 return dev_change_name(dev, ifr->ifr_newname);
2716 * Unknown or private ioctl
2719 default:
2720 if ((cmd >= SIOCDEVPRIVATE &&
2721 cmd <= SIOCDEVPRIVATE + 15) ||
2722 cmd == SIOCBONDENSLAVE ||
2723 cmd == SIOCBONDRELEASE ||
2724 cmd == SIOCBONDSETHWADDR ||
2725 cmd == SIOCBONDSLAVEINFOQUERY ||
2726 cmd == SIOCBONDINFOQUERY ||
2727 cmd == SIOCBONDCHANGEACTIVE ||
2728 cmd == SIOCGMIIPHY ||
2729 cmd == SIOCGMIIREG ||
2730 cmd == SIOCSMIIREG ||
2731 cmd == SIOCBRADDIF ||
2732 cmd == SIOCBRDELIF ||
2733 cmd == SIOCWANDEV) {
2734 err = -EOPNOTSUPP;
2735 if (dev->do_ioctl) {
2736 if (netif_device_present(dev))
2737 err = dev->do_ioctl(dev, ifr,
2738 cmd);
2739 else
2740 err = -ENODEV;
2742 } else
2743 err = -EINVAL;
2746 return err;
2750 * This function handles all "interface"-type I/O control requests. The actual
2751 * 'doing' part of this is dev_ifsioc above.
2755 * dev_ioctl - network device ioctl
2756 * @cmd: command to issue
2757 * @arg: pointer to a struct ifreq in user space
2759 * Issue ioctl functions to devices. This is normally called by the
2760 * user space syscall interfaces but can sometimes be useful for
2761 * other purposes. The return value is the return from the syscall if
2762 * positive or a negative errno code on error.
2765 int dev_ioctl(unsigned int cmd, void __user *arg)
2767 struct ifreq ifr;
2768 int ret;
2769 char *colon;
2771 /* One special case: SIOCGIFCONF takes ifconf argument
2772 and requires shared lock, because it sleeps writing
2773 to user space.
2776 if (cmd == SIOCGIFCONF) {
2777 rtnl_lock();
2778 ret = dev_ifconf((char __user *) arg);
2779 rtnl_unlock();
2780 return ret;
2782 if (cmd == SIOCGIFNAME)
2783 return dev_ifname((struct ifreq __user *)arg);
2785 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2786 return -EFAULT;
2788 ifr.ifr_name[IFNAMSIZ-1] = 0;
2790 colon = strchr(ifr.ifr_name, ':');
2791 if (colon)
2792 *colon = 0;
2795 * See which interface the caller is talking about.
2798 switch (cmd) {
2800 * These ioctl calls:
2801 * - can be done by all.
2802 * - atomic and do not require locking.
2803 * - return a value
2805 case SIOCGIFFLAGS:
2806 case SIOCGIFMETRIC:
2807 case SIOCGIFMTU:
2808 case SIOCGIFHWADDR:
2809 case SIOCGIFSLAVE:
2810 case SIOCGIFMAP:
2811 case SIOCGIFINDEX:
2812 case SIOCGIFTXQLEN:
2813 dev_load(ifr.ifr_name);
2814 read_lock(&dev_base_lock);
2815 ret = dev_ifsioc(&ifr, cmd);
2816 read_unlock(&dev_base_lock);
2817 if (!ret) {
2818 if (colon)
2819 *colon = ':';
2820 if (copy_to_user(arg, &ifr,
2821 sizeof(struct ifreq)))
2822 ret = -EFAULT;
2824 return ret;
2826 case SIOCETHTOOL:
2827 dev_load(ifr.ifr_name);
2828 rtnl_lock();
2829 ret = dev_ethtool(&ifr);
2830 rtnl_unlock();
2831 if (!ret) {
2832 if (colon)
2833 *colon = ':';
2834 if (copy_to_user(arg, &ifr,
2835 sizeof(struct ifreq)))
2836 ret = -EFAULT;
2838 return ret;
2841 * These ioctl calls:
2842 * - require superuser power.
2843 * - require strict serialization.
2844 * - return a value
2846 case SIOCGMIIPHY:
2847 case SIOCGMIIREG:
2848 case SIOCSIFNAME:
2849 if (!capable(CAP_NET_ADMIN))
2850 return -EPERM;
2851 dev_load(ifr.ifr_name);
2852 rtnl_lock();
2853 ret = dev_ifsioc(&ifr, cmd);
2854 rtnl_unlock();
2855 if (!ret) {
2856 if (colon)
2857 *colon = ':';
2858 if (copy_to_user(arg, &ifr,
2859 sizeof(struct ifreq)))
2860 ret = -EFAULT;
2862 return ret;
2865 * These ioctl calls:
2866 * - require superuser power.
2867 * - require strict serialization.
2868 * - do not return a value
2870 case SIOCSIFFLAGS:
2871 case SIOCSIFMETRIC:
2872 case SIOCSIFMTU:
2873 case SIOCSIFMAP:
2874 case SIOCSIFHWADDR:
2875 case SIOCSIFSLAVE:
2876 case SIOCADDMULTI:
2877 case SIOCDELMULTI:
2878 case SIOCSIFHWBROADCAST:
2879 case SIOCSIFTXQLEN:
2880 case SIOCSMIIREG:
2881 case SIOCBONDENSLAVE:
2882 case SIOCBONDRELEASE:
2883 case SIOCBONDSETHWADDR:
2884 case SIOCBONDCHANGEACTIVE:
2885 case SIOCBRADDIF:
2886 case SIOCBRDELIF:
2887 if (!capable(CAP_NET_ADMIN))
2888 return -EPERM;
2889 /* fall through */
2890 case SIOCBONDSLAVEINFOQUERY:
2891 case SIOCBONDINFOQUERY:
2892 dev_load(ifr.ifr_name);
2893 rtnl_lock();
2894 ret = dev_ifsioc(&ifr, cmd);
2895 rtnl_unlock();
2896 return ret;
2898 case SIOCGIFMEM:
2899 /* Get the per device memory space. We can add this but
2900 * currently do not support it */
2901 case SIOCSIFMEM:
2902 /* Set the per device memory buffer space.
2903 * Not applicable in our case */
2904 case SIOCSIFLINK:
2905 return -EINVAL;
2908 * Unknown or private ioctl.
2910 default:
2911 if (cmd == SIOCWANDEV ||
2912 (cmd >= SIOCDEVPRIVATE &&
2913 cmd <= SIOCDEVPRIVATE + 15)) {
2914 dev_load(ifr.ifr_name);
2915 rtnl_lock();
2916 ret = dev_ifsioc(&ifr, cmd);
2917 rtnl_unlock();
2918 if (!ret && copy_to_user(arg, &ifr,
2919 sizeof(struct ifreq)))
2920 ret = -EFAULT;
2921 return ret;
2923 #ifdef CONFIG_WIRELESS_EXT
2924 /* Take care of Wireless Extensions */
2925 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2926 /* If command is `set a parameter', or
2927 * `get the encoding parameters', check if
2928 * the user has the right to do it */
2929 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2930 || cmd == SIOCGIWENCODEEXT) {
2931 if (!capable(CAP_NET_ADMIN))
2932 return -EPERM;
2934 dev_load(ifr.ifr_name);
2935 rtnl_lock();
2936 /* Follow me in net/core/wireless.c */
2937 ret = wireless_process_ioctl(&ifr, cmd);
2938 rtnl_unlock();
2939 if (IW_IS_GET(cmd) &&
2940 copy_to_user(arg, &ifr,
2941 sizeof(struct ifreq)))
2942 ret = -EFAULT;
2943 return ret;
2945 #endif /* CONFIG_WIRELESS_EXT */
2946 return -EINVAL;
2952 * dev_new_index - allocate an ifindex
2954 * Returns a suitable unique value for a new device interface
2955 * number. The caller must hold the rtnl semaphore or the
2956 * dev_base_lock to be sure it remains unique.
2958 static int dev_new_index(void)
2960 static int ifindex;
2961 for (;;) {
2962 if (++ifindex <= 0)
2963 ifindex = 1;
2964 if (!__dev_get_by_index(ifindex))
2965 return ifindex;
2969 static int dev_boot_phase = 1;
2971 /* Delayed registration/unregisteration */
2972 static DEFINE_SPINLOCK(net_todo_list_lock);
2973 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2975 static void net_set_todo(struct net_device *dev)
2977 spin_lock(&net_todo_list_lock);
2978 list_add_tail(&dev->todo_list, &net_todo_list);
2979 spin_unlock(&net_todo_list_lock);
2983 * register_netdevice - register a network device
2984 * @dev: device to register
2986 * Take a completed network device structure and add it to the kernel
2987 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2988 * chain. 0 is returned on success. A negative errno code is returned
2989 * on a failure to set up the device, or if the name is a duplicate.
2991 * Callers must hold the rtnl semaphore. You may want
2992 * register_netdev() instead of this.
2994 * BUGS:
2995 * The locking appears insufficient to guarantee two parallel registers
2996 * will not get the same name.
2999 int register_netdevice(struct net_device *dev)
3001 struct hlist_head *head;
3002 struct hlist_node *p;
3003 int ret;
3005 BUG_ON(dev_boot_phase);
3006 ASSERT_RTNL();
3008 might_sleep();
3010 /* When net_device's are persistent, this will be fatal. */
3011 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3013 spin_lock_init(&dev->queue_lock);
3014 spin_lock_init(&dev->_xmit_lock);
3015 dev->xmit_lock_owner = -1;
3016 #ifdef CONFIG_NET_CLS_ACT
3017 spin_lock_init(&dev->ingress_lock);
3018 #endif
3020 dev->iflink = -1;
3022 /* Init, if this function is available */
3023 if (dev->init) {
3024 ret = dev->init(dev);
3025 if (ret) {
3026 if (ret > 0)
3027 ret = -EIO;
3028 goto out;
3032 if (!dev_valid_name(dev->name)) {
3033 ret = -EINVAL;
3034 goto out;
3037 dev->ifindex = dev_new_index();
3038 if (dev->iflink == -1)
3039 dev->iflink = dev->ifindex;
3041 /* Check for existence of name */
3042 head = dev_name_hash(dev->name);
3043 hlist_for_each(p, head) {
3044 struct net_device *d
3045 = hlist_entry(p, struct net_device, name_hlist);
3046 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3047 ret = -EEXIST;
3048 goto out;
3052 /* Fix illegal SG+CSUM combinations. */
3053 if ((dev->features & NETIF_F_SG) &&
3054 !(dev->features & NETIF_F_ALL_CSUM)) {
3055 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3056 dev->name);
3057 dev->features &= ~NETIF_F_SG;
3060 /* TSO requires that SG is present as well. */
3061 if ((dev->features & NETIF_F_TSO) &&
3062 !(dev->features & NETIF_F_SG)) {
3063 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3064 dev->name);
3065 dev->features &= ~NETIF_F_TSO;
3067 if (dev->features & NETIF_F_UFO) {
3068 if (!(dev->features & NETIF_F_HW_CSUM)) {
3069 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3070 "NETIF_F_HW_CSUM feature.\n",
3071 dev->name);
3072 dev->features &= ~NETIF_F_UFO;
3074 if (!(dev->features & NETIF_F_SG)) {
3075 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3076 "NETIF_F_SG feature.\n",
3077 dev->name);
3078 dev->features &= ~NETIF_F_UFO;
3083 * nil rebuild_header routine,
3084 * that should be never called and used as just bug trap.
3087 if (!dev->rebuild_header)
3088 dev->rebuild_header = default_rebuild_header;
3090 ret = netdev_register_sysfs(dev);
3091 if (ret)
3092 goto out;
3093 dev->reg_state = NETREG_REGISTERED;
3096 * Default initial state at registry is that the
3097 * device is present.
3100 set_bit(__LINK_STATE_PRESENT, &dev->state);
3102 dev->next = NULL;
3103 dev_init_scheduler(dev);
3104 write_lock_bh(&dev_base_lock);
3105 *dev_tail = dev;
3106 dev_tail = &dev->next;
3107 hlist_add_head(&dev->name_hlist, head);
3108 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3109 dev_hold(dev);
3110 write_unlock_bh(&dev_base_lock);
3112 /* Notify protocols, that a new device appeared. */
3113 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3115 ret = 0;
3117 out:
3118 return ret;
3122 * register_netdev - register a network device
3123 * @dev: device to register
3125 * Take a completed network device structure and add it to the kernel
3126 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3127 * chain. 0 is returned on success. A negative errno code is returned
3128 * on a failure to set up the device, or if the name is a duplicate.
3130 * This is a wrapper around register_netdev that takes the rtnl semaphore
3131 * and expands the device name if you passed a format string to
3132 * alloc_netdev.
3134 int register_netdev(struct net_device *dev)
3136 int err;
3138 rtnl_lock();
3141 * If the name is a format string the caller wants us to do a
3142 * name allocation.
3144 if (strchr(dev->name, '%')) {
3145 err = dev_alloc_name(dev, dev->name);
3146 if (err < 0)
3147 goto out;
3150 err = register_netdevice(dev);
3151 out:
3152 rtnl_unlock();
3153 return err;
3155 EXPORT_SYMBOL(register_netdev);
3158 * netdev_wait_allrefs - wait until all references are gone.
3160 * This is called when unregistering network devices.
3162 * Any protocol or device that holds a reference should register
3163 * for netdevice notification, and cleanup and put back the
3164 * reference if they receive an UNREGISTER event.
3165 * We can get stuck here if buggy protocols don't correctly
3166 * call dev_put.
3168 static void netdev_wait_allrefs(struct net_device *dev)
3170 unsigned long rebroadcast_time, warning_time;
3172 rebroadcast_time = warning_time = jiffies;
3173 while (atomic_read(&dev->refcnt) != 0) {
3174 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3175 rtnl_lock();
3177 /* Rebroadcast unregister notification */
3178 raw_notifier_call_chain(&netdev_chain,
3179 NETDEV_UNREGISTER, dev);
3181 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3182 &dev->state)) {
3183 /* We must not have linkwatch events
3184 * pending on unregister. If this
3185 * happens, we simply run the queue
3186 * unscheduled, resulting in a noop
3187 * for this device.
3189 linkwatch_run_queue();
3192 __rtnl_unlock();
3194 rebroadcast_time = jiffies;
3197 msleep(250);
3199 if (time_after(jiffies, warning_time + 10 * HZ)) {
3200 printk(KERN_EMERG "unregister_netdevice: "
3201 "waiting for %s to become free. Usage "
3202 "count = %d\n",
3203 dev->name, atomic_read(&dev->refcnt));
3204 warning_time = jiffies;
3209 /* The sequence is:
3211 * rtnl_lock();
3212 * ...
3213 * register_netdevice(x1);
3214 * register_netdevice(x2);
3215 * ...
3216 * unregister_netdevice(y1);
3217 * unregister_netdevice(y2);
3218 * ...
3219 * rtnl_unlock();
3220 * free_netdev(y1);
3221 * free_netdev(y2);
3223 * We are invoked by rtnl_unlock() after it drops the semaphore.
3224 * This allows us to deal with problems:
3225 * 1) We can delete sysfs objects which invoke hotplug
3226 * without deadlocking with linkwatch via keventd.
3227 * 2) Since we run with the RTNL semaphore not held, we can sleep
3228 * safely in order to wait for the netdev refcnt to drop to zero.
3230 static DEFINE_MUTEX(net_todo_run_mutex);
3231 void netdev_run_todo(void)
3233 struct list_head list;
3235 /* Need to guard against multiple cpu's getting out of order. */
3236 mutex_lock(&net_todo_run_mutex);
3238 /* Not safe to do outside the semaphore. We must not return
3239 * until all unregister events invoked by the local processor
3240 * have been completed (either by this todo run, or one on
3241 * another cpu).
3243 if (list_empty(&net_todo_list))
3244 goto out;
3246 /* Snapshot list, allow later requests */
3247 spin_lock(&net_todo_list_lock);
3248 list_replace_init(&net_todo_list, &list);
3249 spin_unlock(&net_todo_list_lock);
3251 while (!list_empty(&list)) {
3252 struct net_device *dev
3253 = list_entry(list.next, struct net_device, todo_list);
3254 list_del(&dev->todo_list);
3256 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3257 printk(KERN_ERR "network todo '%s' but state %d\n",
3258 dev->name, dev->reg_state);
3259 dump_stack();
3260 continue;
3263 netdev_unregister_sysfs(dev);
3264 dev->reg_state = NETREG_UNREGISTERED;
3266 netdev_wait_allrefs(dev);
3268 /* paranoia */
3269 BUG_ON(atomic_read(&dev->refcnt));
3270 BUG_TRAP(!dev->ip_ptr);
3271 BUG_TRAP(!dev->ip6_ptr);
3272 BUG_TRAP(!dev->dn_ptr);
3274 /* It must be the very last action,
3275 * after this 'dev' may point to freed up memory.
3277 if (dev->destructor)
3278 dev->destructor(dev);
3281 out:
3282 mutex_unlock(&net_todo_run_mutex);
3286 * alloc_netdev - allocate network device
3287 * @sizeof_priv: size of private data to allocate space for
3288 * @name: device name format string
3289 * @setup: callback to initialize device
3291 * Allocates a struct net_device with private data area for driver use
3292 * and performs basic initialization.
3294 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3295 void (*setup)(struct net_device *))
3297 void *p;
3298 struct net_device *dev;
3299 int alloc_size;
3301 BUG_ON(strlen(name) >= sizeof(dev->name));
3303 /* ensure 32-byte alignment of both the device and private area */
3304 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3305 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3307 p = kzalloc(alloc_size, GFP_KERNEL);
3308 if (!p) {
3309 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3310 return NULL;
3313 dev = (struct net_device *)
3314 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3315 dev->padded = (char *)dev - (char *)p;
3317 if (sizeof_priv)
3318 dev->priv = netdev_priv(dev);
3320 setup(dev);
3321 strcpy(dev->name, name);
3322 return dev;
3324 EXPORT_SYMBOL(alloc_netdev);
3327 * free_netdev - free network device
3328 * @dev: device
3330 * This function does the last stage of destroying an allocated device
3331 * interface. The reference to the device object is released.
3332 * If this is the last reference then it will be freed.
3334 void free_netdev(struct net_device *dev)
3336 #ifdef CONFIG_SYSFS
3337 /* Compatibility with error handling in drivers */
3338 if (dev->reg_state == NETREG_UNINITIALIZED) {
3339 kfree((char *)dev - dev->padded);
3340 return;
3343 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3344 dev->reg_state = NETREG_RELEASED;
3346 /* will free via device release */
3347 put_device(&dev->dev);
3348 #else
3349 kfree((char *)dev - dev->padded);
3350 #endif
3353 /* Synchronize with packet receive processing. */
3354 void synchronize_net(void)
3356 might_sleep();
3357 synchronize_rcu();
3361 * unregister_netdevice - remove device from the kernel
3362 * @dev: device
3364 * This function shuts down a device interface and removes it
3365 * from the kernel tables. On success 0 is returned, on a failure
3366 * a negative errno code is returned.
3368 * Callers must hold the rtnl semaphore. You may want
3369 * unregister_netdev() instead of this.
3372 void unregister_netdevice(struct net_device *dev)
3374 struct net_device *d, **dp;
3376 BUG_ON(dev_boot_phase);
3377 ASSERT_RTNL();
3379 /* Some devices call without registering for initialization unwind. */
3380 if (dev->reg_state == NETREG_UNINITIALIZED) {
3381 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3382 "was registered\n", dev->name, dev);
3384 WARN_ON(1);
3385 return;
3388 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3390 /* If device is running, close it first. */
3391 if (dev->flags & IFF_UP)
3392 dev_close(dev);
3394 /* And unlink it from device chain. */
3395 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3396 if (d == dev) {
3397 write_lock_bh(&dev_base_lock);
3398 hlist_del(&dev->name_hlist);
3399 hlist_del(&dev->index_hlist);
3400 if (dev_tail == &dev->next)
3401 dev_tail = dp;
3402 *dp = d->next;
3403 write_unlock_bh(&dev_base_lock);
3404 break;
3407 BUG_ON(!d);
3409 dev->reg_state = NETREG_UNREGISTERING;
3411 synchronize_net();
3413 /* Shutdown queueing discipline. */
3414 dev_shutdown(dev);
3417 /* Notify protocols, that we are about to destroy
3418 this device. They should clean all the things.
3420 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3423 * Flush the multicast chain
3425 dev_mc_discard(dev);
3427 if (dev->uninit)
3428 dev->uninit(dev);
3430 /* Notifier chain MUST detach us from master device. */
3431 BUG_TRAP(!dev->master);
3433 /* Finish processing unregister after unlock */
3434 net_set_todo(dev);
3436 synchronize_net();
3438 dev_put(dev);
3442 * unregister_netdev - remove device from the kernel
3443 * @dev: device
3445 * This function shuts down a device interface and removes it
3446 * from the kernel tables. On success 0 is returned, on a failure
3447 * a negative errno code is returned.
3449 * This is just a wrapper for unregister_netdevice that takes
3450 * the rtnl semaphore. In general you want to use this and not
3451 * unregister_netdevice.
3453 void unregister_netdev(struct net_device *dev)
3455 rtnl_lock();
3456 unregister_netdevice(dev);
3457 rtnl_unlock();
3460 EXPORT_SYMBOL(unregister_netdev);
3462 static int dev_cpu_callback(struct notifier_block *nfb,
3463 unsigned long action,
3464 void *ocpu)
3466 struct sk_buff **list_skb;
3467 struct net_device **list_net;
3468 struct sk_buff *skb;
3469 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3470 struct softnet_data *sd, *oldsd;
3472 if (action != CPU_DEAD)
3473 return NOTIFY_OK;
3475 local_irq_disable();
3476 cpu = smp_processor_id();
3477 sd = &per_cpu(softnet_data, cpu);
3478 oldsd = &per_cpu(softnet_data, oldcpu);
3480 /* Find end of our completion_queue. */
3481 list_skb = &sd->completion_queue;
3482 while (*list_skb)
3483 list_skb = &(*list_skb)->next;
3484 /* Append completion queue from offline CPU. */
3485 *list_skb = oldsd->completion_queue;
3486 oldsd->completion_queue = NULL;
3488 /* Find end of our output_queue. */
3489 list_net = &sd->output_queue;
3490 while (*list_net)
3491 list_net = &(*list_net)->next_sched;
3492 /* Append output queue from offline CPU. */
3493 *list_net = oldsd->output_queue;
3494 oldsd->output_queue = NULL;
3496 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3497 local_irq_enable();
3499 /* Process offline CPU's input_pkt_queue */
3500 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3501 netif_rx(skb);
3503 return NOTIFY_OK;
3506 #ifdef CONFIG_NET_DMA
3508 * net_dma_rebalance -
3509 * This is called when the number of channels allocated to the net_dma_client
3510 * changes. The net_dma_client tries to have one DMA channel per CPU.
3512 static void net_dma_rebalance(void)
3514 unsigned int cpu, i, n;
3515 struct dma_chan *chan;
3517 if (net_dma_count == 0) {
3518 for_each_online_cpu(cpu)
3519 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
3520 return;
3523 i = 0;
3524 cpu = first_cpu(cpu_online_map);
3526 rcu_read_lock();
3527 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3528 n = ((num_online_cpus() / net_dma_count)
3529 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3531 while(n) {
3532 per_cpu(softnet_data, cpu).net_dma = chan;
3533 cpu = next_cpu(cpu, cpu_online_map);
3534 n--;
3536 i++;
3538 rcu_read_unlock();
3542 * netdev_dma_event - event callback for the net_dma_client
3543 * @client: should always be net_dma_client
3544 * @chan: DMA channel for the event
3545 * @event: event type
3547 static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3548 enum dma_event event)
3550 spin_lock(&net_dma_event_lock);
3551 switch (event) {
3552 case DMA_RESOURCE_ADDED:
3553 net_dma_count++;
3554 net_dma_rebalance();
3555 break;
3556 case DMA_RESOURCE_REMOVED:
3557 net_dma_count--;
3558 net_dma_rebalance();
3559 break;
3560 default:
3561 break;
3563 spin_unlock(&net_dma_event_lock);
3567 * netdev_dma_regiser - register the networking subsystem as a DMA client
3569 static int __init netdev_dma_register(void)
3571 spin_lock_init(&net_dma_event_lock);
3572 net_dma_client = dma_async_client_register(netdev_dma_event);
3573 if (net_dma_client == NULL)
3574 return -ENOMEM;
3576 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3577 return 0;
3580 #else
3581 static int __init netdev_dma_register(void) { return -ENODEV; }
3582 #endif /* CONFIG_NET_DMA */
3585 * Initialize the DEV module. At boot time this walks the device list and
3586 * unhooks any devices that fail to initialise (normally hardware not
3587 * present) and leaves us with a valid list of present and active devices.
3592 * This is called single threaded during boot, so no need
3593 * to take the rtnl semaphore.
3595 static int __init net_dev_init(void)
3597 int i, rc = -ENOMEM;
3599 BUG_ON(!dev_boot_phase);
3601 if (dev_proc_init())
3602 goto out;
3604 if (netdev_sysfs_init())
3605 goto out;
3607 INIT_LIST_HEAD(&ptype_all);
3608 for (i = 0; i < 16; i++)
3609 INIT_LIST_HEAD(&ptype_base[i]);
3611 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3612 INIT_HLIST_HEAD(&dev_name_head[i]);
3614 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3615 INIT_HLIST_HEAD(&dev_index_head[i]);
3618 * Initialise the packet receive queues.
3621 for_each_possible_cpu(i) {
3622 struct softnet_data *queue;
3624 queue = &per_cpu(softnet_data, i);
3625 skb_queue_head_init(&queue->input_pkt_queue);
3626 queue->completion_queue = NULL;
3627 INIT_LIST_HEAD(&queue->poll_list);
3628 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3629 queue->backlog_dev.weight = weight_p;
3630 queue->backlog_dev.poll = process_backlog;
3631 atomic_set(&queue->backlog_dev.refcnt, 1);
3634 netdev_dma_register();
3636 dev_boot_phase = 0;
3638 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3639 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3641 hotcpu_notifier(dev_cpu_callback, 0);
3642 dst_init();
3643 dev_mcast_init();
3644 rc = 0;
3645 out:
3646 return rc;
3649 subsys_initcall(net_dev_init);
3651 EXPORT_SYMBOL(__dev_get_by_index);
3652 EXPORT_SYMBOL(__dev_get_by_name);
3653 EXPORT_SYMBOL(__dev_remove_pack);
3654 EXPORT_SYMBOL(dev_valid_name);
3655 EXPORT_SYMBOL(dev_add_pack);
3656 EXPORT_SYMBOL(dev_alloc_name);
3657 EXPORT_SYMBOL(dev_close);
3658 EXPORT_SYMBOL(dev_get_by_flags);
3659 EXPORT_SYMBOL(dev_get_by_index);
3660 EXPORT_SYMBOL(dev_get_by_name);
3661 EXPORT_SYMBOL(dev_open);
3662 EXPORT_SYMBOL(dev_queue_xmit);
3663 EXPORT_SYMBOL(dev_remove_pack);
3664 EXPORT_SYMBOL(dev_set_allmulti);
3665 EXPORT_SYMBOL(dev_set_promiscuity);
3666 EXPORT_SYMBOL(dev_change_flags);
3667 EXPORT_SYMBOL(dev_set_mtu);
3668 EXPORT_SYMBOL(dev_set_mac_address);
3669 EXPORT_SYMBOL(free_netdev);
3670 EXPORT_SYMBOL(netdev_boot_setup_check);
3671 EXPORT_SYMBOL(netdev_set_master);
3672 EXPORT_SYMBOL(netdev_state_change);
3673 EXPORT_SYMBOL(netif_receive_skb);
3674 EXPORT_SYMBOL(netif_rx);
3675 EXPORT_SYMBOL(register_gifconf);
3676 EXPORT_SYMBOL(register_netdevice);
3677 EXPORT_SYMBOL(register_netdevice_notifier);
3678 EXPORT_SYMBOL(skb_checksum_help);
3679 EXPORT_SYMBOL(synchronize_net);
3680 EXPORT_SYMBOL(unregister_netdevice);
3681 EXPORT_SYMBOL(unregister_netdevice_notifier);
3682 EXPORT_SYMBOL(net_enable_timestamp);
3683 EXPORT_SYMBOL(net_disable_timestamp);
3684 EXPORT_SYMBOL(dev_get_flags);
3686 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3687 EXPORT_SYMBOL(br_handle_frame_hook);
3688 EXPORT_SYMBOL(br_fdb_get_hook);
3689 EXPORT_SYMBOL(br_fdb_put_hook);
3690 #endif
3692 #ifdef CONFIG_KMOD
3693 EXPORT_SYMBOL(dev_load);
3694 #endif
3696 EXPORT_PER_CPU_SYMBOL(softnet_data);