[PATCH] sparc: register_cpu() build fix
[linux-2.6/x86.git] / net / core / dev.c
blobf1c52cbd6ef7bfa8ae13ff04913b585b0c9be286
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/config.h>
80 #include <linux/cpu.h>
81 #include <linux/types.h>
82 #include <linux/kernel.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
86 #include <linux/mm.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/sock.h>
97 #include <linux/rtnetlink.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <linux/stat.h>
101 #include <linux/if_bridge.h>
102 #include <linux/divert.h>
103 #include <net/dst.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/kmod.h>
109 #include <linux/module.h>
110 #include <linux/kallsyms.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <linux/wireless.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
122 * The list of packet types we will receive (as opposed to discard)
123 * and the routines to invoke.
125 * Why 16. Because with 16 the only overlap we get on a hash of the
126 * low nibble of the protocol value is RARP/SNAP/X.25.
128 * NOTE: That is no longer true with the addition of VLAN tags. Not
129 * sure which should go first, but I bet it won't make much
130 * difference if we are running VLANs. The good news is that
131 * this protocol won't be in the list unless compiled in, so
132 * the average user (w/out VLANs) will not be adversely affected.
133 * --BLG
135 * 0800 IP
136 * 8100 802.1Q VLAN
137 * 0001 802.3
138 * 0002 AX.25
139 * 0004 802.2
140 * 8035 RARP
141 * 0005 SNAP
142 * 0805 X.25
143 * 0806 ARP
144 * 8137 IPX
145 * 0009 Localtalk
146 * 86DD IPv6
149 static DEFINE_SPINLOCK(ptype_lock);
150 static struct list_head ptype_base[16]; /* 16 way hashed list */
151 static struct list_head ptype_all; /* Taps */
153 #ifdef CONFIG_NET_DMA
154 static struct dma_client *net_dma_client;
155 static unsigned int net_dma_count;
156 static spinlock_t net_dma_event_lock;
157 #endif
160 * The @dev_base list is protected by @dev_base_lock and the rtnl
161 * semaphore.
163 * Pure readers hold dev_base_lock for reading.
165 * Writers must hold the rtnl semaphore while they loop through the
166 * dev_base list, and hold dev_base_lock for writing when they do the
167 * actual updates. This allows pure readers to access the list even
168 * while a writer is preparing to update it.
170 * To put it another way, dev_base_lock is held for writing only to
171 * protect against pure readers; the rtnl semaphore provides the
172 * protection against other writers.
174 * See, for example usages, register_netdevice() and
175 * unregister_netdevice(), which must be called with the rtnl
176 * semaphore held.
178 struct net_device *dev_base;
179 static struct net_device **dev_tail = &dev_base;
180 DEFINE_RWLOCK(dev_base_lock);
182 EXPORT_SYMBOL(dev_base);
183 EXPORT_SYMBOL(dev_base_lock);
185 #define NETDEV_HASHBITS 8
186 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
187 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
189 static inline struct hlist_head *dev_name_hash(const char *name)
191 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
192 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
195 static inline struct hlist_head *dev_index_hash(int ifindex)
197 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
201 * Our notifier list
204 static RAW_NOTIFIER_HEAD(netdev_chain);
207 * Device drivers call our routines to queue packets here. We empty the
208 * queue in the local softnet handler.
210 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
212 #ifdef CONFIG_SYSFS
213 extern int netdev_sysfs_init(void);
214 extern int netdev_register_sysfs(struct net_device *);
215 extern void netdev_unregister_sysfs(struct net_device *);
216 #else
217 #define netdev_sysfs_init() (0)
218 #define netdev_register_sysfs(dev) (0)
219 #define netdev_unregister_sysfs(dev) do { } while(0)
220 #endif
223 /*******************************************************************************
225 Protocol management and registration routines
227 *******************************************************************************/
230 * For efficiency
233 static int netdev_nit;
236 * Add a protocol ID to the list. Now that the input handler is
237 * smarter we can dispense with all the messy stuff that used to be
238 * here.
240 * BEWARE!!! Protocol handlers, mangling input packets,
241 * MUST BE last in hash buckets and checking protocol handlers
242 * MUST start from promiscuous ptype_all chain in net_bh.
243 * It is true now, do not change it.
244 * Explanation follows: if protocol handler, mangling packet, will
245 * be the first on list, it is not able to sense, that packet
246 * is cloned and should be copied-on-write, so that it will
247 * change it and subsequent readers will get broken packet.
248 * --ANK (980803)
252 * dev_add_pack - add packet handler
253 * @pt: packet type declaration
255 * Add a protocol handler to the networking stack. The passed &packet_type
256 * is linked into kernel lists and may not be freed until it has been
257 * removed from the kernel lists.
259 * This call does not sleep therefore it can not
260 * guarantee all CPU's that are in middle of receiving packets
261 * will see the new packet type (until the next received packet).
264 void dev_add_pack(struct packet_type *pt)
266 int hash;
268 spin_lock_bh(&ptype_lock);
269 if (pt->type == htons(ETH_P_ALL)) {
270 netdev_nit++;
271 list_add_rcu(&pt->list, &ptype_all);
272 } else {
273 hash = ntohs(pt->type) & 15;
274 list_add_rcu(&pt->list, &ptype_base[hash]);
276 spin_unlock_bh(&ptype_lock);
280 * __dev_remove_pack - remove packet handler
281 * @pt: packet type declaration
283 * Remove a protocol handler that was previously added to the kernel
284 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
285 * from the kernel lists and can be freed or reused once this function
286 * returns.
288 * The packet type might still be in use by receivers
289 * and must not be freed until after all the CPU's have gone
290 * through a quiescent state.
292 void __dev_remove_pack(struct packet_type *pt)
294 struct list_head *head;
295 struct packet_type *pt1;
297 spin_lock_bh(&ptype_lock);
299 if (pt->type == htons(ETH_P_ALL)) {
300 netdev_nit--;
301 head = &ptype_all;
302 } else
303 head = &ptype_base[ntohs(pt->type) & 15];
305 list_for_each_entry(pt1, head, list) {
306 if (pt == pt1) {
307 list_del_rcu(&pt->list);
308 goto out;
312 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
313 out:
314 spin_unlock_bh(&ptype_lock);
317 * dev_remove_pack - remove packet handler
318 * @pt: packet type declaration
320 * Remove a protocol handler that was previously added to the kernel
321 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
322 * from the kernel lists and can be freed or reused once this function
323 * returns.
325 * This call sleeps to guarantee that no CPU is looking at the packet
326 * type after return.
328 void dev_remove_pack(struct packet_type *pt)
330 __dev_remove_pack(pt);
332 synchronize_net();
335 /******************************************************************************
337 Device Boot-time Settings Routines
339 *******************************************************************************/
341 /* Boot time configuration table */
342 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
345 * netdev_boot_setup_add - add new setup entry
346 * @name: name of the device
347 * @map: configured settings for the device
349 * Adds new setup entry to the dev_boot_setup list. The function
350 * returns 0 on error and 1 on success. This is a generic routine to
351 * all netdevices.
353 static int netdev_boot_setup_add(char *name, struct ifmap *map)
355 struct netdev_boot_setup *s;
356 int i;
358 s = dev_boot_setup;
359 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
360 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
361 memset(s[i].name, 0, sizeof(s[i].name));
362 strcpy(s[i].name, name);
363 memcpy(&s[i].map, map, sizeof(s[i].map));
364 break;
368 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
372 * netdev_boot_setup_check - check boot time settings
373 * @dev: the netdevice
375 * Check boot time settings for the device.
376 * The found settings are set for the device to be used
377 * later in the device probing.
378 * Returns 0 if no settings found, 1 if they are.
380 int netdev_boot_setup_check(struct net_device *dev)
382 struct netdev_boot_setup *s = dev_boot_setup;
383 int i;
385 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
386 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
387 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
388 dev->irq = s[i].map.irq;
389 dev->base_addr = s[i].map.base_addr;
390 dev->mem_start = s[i].map.mem_start;
391 dev->mem_end = s[i].map.mem_end;
392 return 1;
395 return 0;
400 * netdev_boot_base - get address from boot time settings
401 * @prefix: prefix for network device
402 * @unit: id for network device
404 * Check boot time settings for the base address of device.
405 * The found settings are set for the device to be used
406 * later in the device probing.
407 * Returns 0 if no settings found.
409 unsigned long netdev_boot_base(const char *prefix, int unit)
411 const struct netdev_boot_setup *s = dev_boot_setup;
412 char name[IFNAMSIZ];
413 int i;
415 sprintf(name, "%s%d", prefix, unit);
418 * If device already registered then return base of 1
419 * to indicate not to probe for this interface
421 if (__dev_get_by_name(name))
422 return 1;
424 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
425 if (!strcmp(name, s[i].name))
426 return s[i].map.base_addr;
427 return 0;
431 * Saves at boot time configured settings for any netdevice.
433 int __init netdev_boot_setup(char *str)
435 int ints[5];
436 struct ifmap map;
438 str = get_options(str, ARRAY_SIZE(ints), ints);
439 if (!str || !*str)
440 return 0;
442 /* Save settings */
443 memset(&map, 0, sizeof(map));
444 if (ints[0] > 0)
445 map.irq = ints[1];
446 if (ints[0] > 1)
447 map.base_addr = ints[2];
448 if (ints[0] > 2)
449 map.mem_start = ints[3];
450 if (ints[0] > 3)
451 map.mem_end = ints[4];
453 /* Add new entry to the list */
454 return netdev_boot_setup_add(str, &map);
457 __setup("netdev=", netdev_boot_setup);
459 /*******************************************************************************
461 Device Interface Subroutines
463 *******************************************************************************/
466 * __dev_get_by_name - find a device by its name
467 * @name: name to find
469 * Find an interface by name. Must be called under RTNL semaphore
470 * or @dev_base_lock. If the name is found a pointer to the device
471 * is returned. If the name is not found then %NULL is returned. The
472 * reference counters are not incremented so the caller must be
473 * careful with locks.
476 struct net_device *__dev_get_by_name(const char *name)
478 struct hlist_node *p;
480 hlist_for_each(p, dev_name_hash(name)) {
481 struct net_device *dev
482 = hlist_entry(p, struct net_device, name_hlist);
483 if (!strncmp(dev->name, name, IFNAMSIZ))
484 return dev;
486 return NULL;
490 * dev_get_by_name - find a device by its name
491 * @name: name to find
493 * Find an interface by name. This can be called from any
494 * context and does its own locking. The returned handle has
495 * the usage count incremented and the caller must use dev_put() to
496 * release it when it is no longer needed. %NULL is returned if no
497 * matching device is found.
500 struct net_device *dev_get_by_name(const char *name)
502 struct net_device *dev;
504 read_lock(&dev_base_lock);
505 dev = __dev_get_by_name(name);
506 if (dev)
507 dev_hold(dev);
508 read_unlock(&dev_base_lock);
509 return dev;
513 * __dev_get_by_index - find a device by its ifindex
514 * @ifindex: index of device
516 * Search for an interface by index. Returns %NULL if the device
517 * is not found or a pointer to the device. The device has not
518 * had its reference counter increased so the caller must be careful
519 * about locking. The caller must hold either the RTNL semaphore
520 * or @dev_base_lock.
523 struct net_device *__dev_get_by_index(int ifindex)
525 struct hlist_node *p;
527 hlist_for_each(p, dev_index_hash(ifindex)) {
528 struct net_device *dev
529 = hlist_entry(p, struct net_device, index_hlist);
530 if (dev->ifindex == ifindex)
531 return dev;
533 return NULL;
538 * dev_get_by_index - find a device by its ifindex
539 * @ifindex: index of device
541 * Search for an interface by index. Returns NULL if the device
542 * is not found or a pointer to the device. The device returned has
543 * had a reference added and the pointer is safe until the user calls
544 * dev_put to indicate they have finished with it.
547 struct net_device *dev_get_by_index(int ifindex)
549 struct net_device *dev;
551 read_lock(&dev_base_lock);
552 dev = __dev_get_by_index(ifindex);
553 if (dev)
554 dev_hold(dev);
555 read_unlock(&dev_base_lock);
556 return dev;
560 * dev_getbyhwaddr - find a device by its hardware address
561 * @type: media type of device
562 * @ha: hardware address
564 * Search for an interface by MAC address. Returns NULL if the device
565 * is not found or a pointer to the device. The caller must hold the
566 * rtnl semaphore. The returned device has not had its ref count increased
567 * and the caller must therefore be careful about locking
569 * BUGS:
570 * If the API was consistent this would be __dev_get_by_hwaddr
573 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
575 struct net_device *dev;
577 ASSERT_RTNL();
579 for (dev = dev_base; dev; dev = dev->next)
580 if (dev->type == type &&
581 !memcmp(dev->dev_addr, ha, dev->addr_len))
582 break;
583 return dev;
586 EXPORT_SYMBOL(dev_getbyhwaddr);
588 struct net_device *dev_getfirstbyhwtype(unsigned short type)
590 struct net_device *dev;
592 rtnl_lock();
593 for (dev = dev_base; dev; dev = dev->next) {
594 if (dev->type == type) {
595 dev_hold(dev);
596 break;
599 rtnl_unlock();
600 return dev;
603 EXPORT_SYMBOL(dev_getfirstbyhwtype);
606 * dev_get_by_flags - find any device with given flags
607 * @if_flags: IFF_* values
608 * @mask: bitmask of bits in if_flags to check
610 * Search for any interface with the given flags. Returns NULL if a device
611 * is not found or a pointer to the device. The device returned has
612 * had a reference added and the pointer is safe until the user calls
613 * dev_put to indicate they have finished with it.
616 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
618 struct net_device *dev;
620 read_lock(&dev_base_lock);
621 for (dev = dev_base; dev != NULL; dev = dev->next) {
622 if (((dev->flags ^ if_flags) & mask) == 0) {
623 dev_hold(dev);
624 break;
627 read_unlock(&dev_base_lock);
628 return dev;
632 * dev_valid_name - check if name is okay for network device
633 * @name: name string
635 * Network device names need to be valid file names to
636 * to allow sysfs to work
638 int dev_valid_name(const char *name)
640 return !(*name == '\0'
641 || !strcmp(name, ".")
642 || !strcmp(name, "..")
643 || strchr(name, '/'));
647 * dev_alloc_name - allocate a name for a device
648 * @dev: device
649 * @name: name format string
651 * Passed a format string - eg "lt%d" it will try and find a suitable
652 * id. It scans list of devices to build up a free map, then chooses
653 * the first empty slot. The caller must hold the dev_base or rtnl lock
654 * while allocating the name and adding the device in order to avoid
655 * duplicates.
656 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
657 * Returns the number of the unit assigned or a negative errno code.
660 int dev_alloc_name(struct net_device *dev, const char *name)
662 int i = 0;
663 char buf[IFNAMSIZ];
664 const char *p;
665 const int max_netdevices = 8*PAGE_SIZE;
666 long *inuse;
667 struct net_device *d;
669 p = strnchr(name, IFNAMSIZ-1, '%');
670 if (p) {
672 * Verify the string as this thing may have come from
673 * the user. There must be either one "%d" and no other "%"
674 * characters.
676 if (p[1] != 'd' || strchr(p + 2, '%'))
677 return -EINVAL;
679 /* Use one page as a bit array of possible slots */
680 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
681 if (!inuse)
682 return -ENOMEM;
684 for (d = dev_base; d; d = d->next) {
685 if (!sscanf(d->name, name, &i))
686 continue;
687 if (i < 0 || i >= max_netdevices)
688 continue;
690 /* avoid cases where sscanf is not exact inverse of printf */
691 snprintf(buf, sizeof(buf), name, i);
692 if (!strncmp(buf, d->name, IFNAMSIZ))
693 set_bit(i, inuse);
696 i = find_first_zero_bit(inuse, max_netdevices);
697 free_page((unsigned long) inuse);
700 snprintf(buf, sizeof(buf), name, i);
701 if (!__dev_get_by_name(buf)) {
702 strlcpy(dev->name, buf, IFNAMSIZ);
703 return i;
706 /* It is possible to run out of possible slots
707 * when the name is long and there isn't enough space left
708 * for the digits, or if all bits are used.
710 return -ENFILE;
715 * dev_change_name - change name of a device
716 * @dev: device
717 * @newname: name (or format string) must be at least IFNAMSIZ
719 * Change name of a device, can pass format strings "eth%d".
720 * for wildcarding.
722 int dev_change_name(struct net_device *dev, char *newname)
724 int err = 0;
726 ASSERT_RTNL();
728 if (dev->flags & IFF_UP)
729 return -EBUSY;
731 if (!dev_valid_name(newname))
732 return -EINVAL;
734 if (strchr(newname, '%')) {
735 err = dev_alloc_name(dev, newname);
736 if (err < 0)
737 return err;
738 strcpy(newname, dev->name);
740 else if (__dev_get_by_name(newname))
741 return -EEXIST;
742 else
743 strlcpy(dev->name, newname, IFNAMSIZ);
745 err = class_device_rename(&dev->class_dev, dev->name);
746 if (!err) {
747 hlist_del(&dev->name_hlist);
748 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
749 raw_notifier_call_chain(&netdev_chain,
750 NETDEV_CHANGENAME, dev);
753 return err;
757 * netdev_features_change - device changes features
758 * @dev: device to cause notification
760 * Called to indicate a device has changed features.
762 void netdev_features_change(struct net_device *dev)
764 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
766 EXPORT_SYMBOL(netdev_features_change);
769 * netdev_state_change - device changes state
770 * @dev: device to cause notification
772 * Called to indicate a device has changed state. This function calls
773 * the notifier chains for netdev_chain and sends a NEWLINK message
774 * to the routing socket.
776 void netdev_state_change(struct net_device *dev)
778 if (dev->flags & IFF_UP) {
779 raw_notifier_call_chain(&netdev_chain,
780 NETDEV_CHANGE, dev);
781 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
786 * dev_load - load a network module
787 * @name: name of interface
789 * If a network interface is not present and the process has suitable
790 * privileges this function loads the module. If module loading is not
791 * available in this kernel then it becomes a nop.
794 void dev_load(const char *name)
796 struct net_device *dev;
798 read_lock(&dev_base_lock);
799 dev = __dev_get_by_name(name);
800 read_unlock(&dev_base_lock);
802 if (!dev && capable(CAP_SYS_MODULE))
803 request_module("%s", name);
806 static int default_rebuild_header(struct sk_buff *skb)
808 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
809 skb->dev ? skb->dev->name : "NULL!!!");
810 kfree_skb(skb);
811 return 1;
816 * dev_open - prepare an interface for use.
817 * @dev: device to open
819 * Takes a device from down to up state. The device's private open
820 * function is invoked and then the multicast lists are loaded. Finally
821 * the device is moved into the up state and a %NETDEV_UP message is
822 * sent to the netdev notifier chain.
824 * Calling this function on an active interface is a nop. On a failure
825 * a negative errno code is returned.
827 int dev_open(struct net_device *dev)
829 int ret = 0;
832 * Is it already up?
835 if (dev->flags & IFF_UP)
836 return 0;
839 * Is it even present?
841 if (!netif_device_present(dev))
842 return -ENODEV;
845 * Call device private open method
847 set_bit(__LINK_STATE_START, &dev->state);
848 if (dev->open) {
849 ret = dev->open(dev);
850 if (ret)
851 clear_bit(__LINK_STATE_START, &dev->state);
855 * If it went open OK then:
858 if (!ret) {
860 * Set the flags.
862 dev->flags |= IFF_UP;
865 * Initialize multicasting status
867 dev_mc_upload(dev);
870 * Wakeup transmit queue engine
872 dev_activate(dev);
875 * ... and announce new interface.
877 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
879 return ret;
883 * dev_close - shutdown an interface.
884 * @dev: device to shutdown
886 * This function moves an active device into down state. A
887 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
888 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
889 * chain.
891 int dev_close(struct net_device *dev)
893 if (!(dev->flags & IFF_UP))
894 return 0;
897 * Tell people we are going down, so that they can
898 * prepare to death, when device is still operating.
900 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
902 dev_deactivate(dev);
904 clear_bit(__LINK_STATE_START, &dev->state);
906 /* Synchronize to scheduled poll. We cannot touch poll list,
907 * it can be even on different cpu. So just clear netif_running(),
908 * and wait when poll really will happen. Actually, the best place
909 * for this is inside dev->stop() after device stopped its irq
910 * engine, but this requires more changes in devices. */
912 smp_mb__after_clear_bit(); /* Commit netif_running(). */
913 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
914 /* No hurry. */
915 msleep(1);
919 * Call the device specific close. This cannot fail.
920 * Only if device is UP
922 * We allow it to be called even after a DETACH hot-plug
923 * event.
925 if (dev->stop)
926 dev->stop(dev);
929 * Device is now down.
932 dev->flags &= ~IFF_UP;
935 * Tell people we are down
937 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
939 return 0;
944 * Device change register/unregister. These are not inline or static
945 * as we export them to the world.
949 * register_netdevice_notifier - register a network notifier block
950 * @nb: notifier
952 * Register a notifier to be called when network device events occur.
953 * The notifier passed is linked into the kernel structures and must
954 * not be reused until it has been unregistered. A negative errno code
955 * is returned on a failure.
957 * When registered all registration and up events are replayed
958 * to the new notifier to allow device to have a race free
959 * view of the network device list.
962 int register_netdevice_notifier(struct notifier_block *nb)
964 struct net_device *dev;
965 int err;
967 rtnl_lock();
968 err = raw_notifier_chain_register(&netdev_chain, nb);
969 if (!err) {
970 for (dev = dev_base; dev; dev = dev->next) {
971 nb->notifier_call(nb, NETDEV_REGISTER, dev);
973 if (dev->flags & IFF_UP)
974 nb->notifier_call(nb, NETDEV_UP, dev);
977 rtnl_unlock();
978 return err;
982 * unregister_netdevice_notifier - unregister a network notifier block
983 * @nb: notifier
985 * Unregister a notifier previously registered by
986 * register_netdevice_notifier(). The notifier is unlinked into the
987 * kernel structures and may then be reused. A negative errno code
988 * is returned on a failure.
991 int unregister_netdevice_notifier(struct notifier_block *nb)
993 int err;
995 rtnl_lock();
996 err = raw_notifier_chain_unregister(&netdev_chain, nb);
997 rtnl_unlock();
998 return err;
1002 * call_netdevice_notifiers - call all network notifier blocks
1003 * @val: value passed unmodified to notifier function
1004 * @v: pointer passed unmodified to notifier function
1006 * Call all network notifier blocks. Parameters and return value
1007 * are as for raw_notifier_call_chain().
1010 int call_netdevice_notifiers(unsigned long val, void *v)
1012 return raw_notifier_call_chain(&netdev_chain, val, v);
1015 /* When > 0 there are consumers of rx skb time stamps */
1016 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1018 void net_enable_timestamp(void)
1020 atomic_inc(&netstamp_needed);
1023 void net_disable_timestamp(void)
1025 atomic_dec(&netstamp_needed);
1028 void __net_timestamp(struct sk_buff *skb)
1030 struct timeval tv;
1032 do_gettimeofday(&tv);
1033 skb_set_timestamp(skb, &tv);
1035 EXPORT_SYMBOL(__net_timestamp);
1037 static inline void net_timestamp(struct sk_buff *skb)
1039 if (atomic_read(&netstamp_needed))
1040 __net_timestamp(skb);
1041 else {
1042 skb->tstamp.off_sec = 0;
1043 skb->tstamp.off_usec = 0;
1048 * Support routine. Sends outgoing frames to any network
1049 * taps currently in use.
1052 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1054 struct packet_type *ptype;
1056 net_timestamp(skb);
1058 rcu_read_lock();
1059 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1060 /* Never send packets back to the socket
1061 * they originated from - MvS (miquels@drinkel.ow.org)
1063 if ((ptype->dev == dev || !ptype->dev) &&
1064 (ptype->af_packet_priv == NULL ||
1065 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1066 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1067 if (!skb2)
1068 break;
1070 /* skb->nh should be correctly
1071 set by sender, so that the second statement is
1072 just protection against buggy protocols.
1074 skb2->mac.raw = skb2->data;
1076 if (skb2->nh.raw < skb2->data ||
1077 skb2->nh.raw > skb2->tail) {
1078 if (net_ratelimit())
1079 printk(KERN_CRIT "protocol %04x is "
1080 "buggy, dev %s\n",
1081 skb2->protocol, dev->name);
1082 skb2->nh.raw = skb2->data;
1085 skb2->h.raw = skb2->nh.raw;
1086 skb2->pkt_type = PACKET_OUTGOING;
1087 ptype->func(skb2, skb->dev, ptype, skb->dev);
1090 rcu_read_unlock();
1094 void __netif_schedule(struct net_device *dev)
1096 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1097 unsigned long flags;
1098 struct softnet_data *sd;
1100 local_irq_save(flags);
1101 sd = &__get_cpu_var(softnet_data);
1102 dev->next_sched = sd->output_queue;
1103 sd->output_queue = dev;
1104 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1105 local_irq_restore(flags);
1108 EXPORT_SYMBOL(__netif_schedule);
1110 void __netif_rx_schedule(struct net_device *dev)
1112 unsigned long flags;
1114 local_irq_save(flags);
1115 dev_hold(dev);
1116 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1117 if (dev->quota < 0)
1118 dev->quota += dev->weight;
1119 else
1120 dev->quota = dev->weight;
1121 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1122 local_irq_restore(flags);
1124 EXPORT_SYMBOL(__netif_rx_schedule);
1126 void dev_kfree_skb_any(struct sk_buff *skb)
1128 if (in_irq() || irqs_disabled())
1129 dev_kfree_skb_irq(skb);
1130 else
1131 dev_kfree_skb(skb);
1133 EXPORT_SYMBOL(dev_kfree_skb_any);
1136 /* Hot-plugging. */
1137 void netif_device_detach(struct net_device *dev)
1139 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1140 netif_running(dev)) {
1141 netif_stop_queue(dev);
1144 EXPORT_SYMBOL(netif_device_detach);
1146 void netif_device_attach(struct net_device *dev)
1148 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1149 netif_running(dev)) {
1150 netif_wake_queue(dev);
1151 __netdev_watchdog_up(dev);
1154 EXPORT_SYMBOL(netif_device_attach);
1158 * Invalidate hardware checksum when packet is to be mangled, and
1159 * complete checksum manually on outgoing path.
1161 int skb_checksum_help(struct sk_buff *skb, int inward)
1163 unsigned int csum;
1164 int ret = 0, offset = skb->h.raw - skb->data;
1166 if (inward) {
1167 skb->ip_summed = CHECKSUM_NONE;
1168 goto out;
1171 if (skb_cloned(skb)) {
1172 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1173 if (ret)
1174 goto out;
1177 BUG_ON(offset > (int)skb->len);
1178 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1180 offset = skb->tail - skb->h.raw;
1181 BUG_ON(offset <= 0);
1182 BUG_ON(skb->csum + 2 > offset);
1184 *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
1185 skb->ip_summed = CHECKSUM_NONE;
1186 out:
1187 return ret;
1191 * skb_gso_segment - Perform segmentation on skb.
1192 * @skb: buffer to segment
1193 * @sg: whether scatter-gather is supported on the target.
1195 * This function segments the given skb and returns a list of segments.
1197 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int sg)
1199 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1200 struct packet_type *ptype;
1201 int type = skb->protocol;
1203 BUG_ON(skb_shinfo(skb)->frag_list);
1204 BUG_ON(skb->ip_summed != CHECKSUM_HW);
1206 skb->mac.raw = skb->data;
1207 skb->mac_len = skb->nh.raw - skb->data;
1208 __skb_pull(skb, skb->mac_len);
1210 rcu_read_lock();
1211 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1212 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1213 segs = ptype->gso_segment(skb, sg);
1214 break;
1217 rcu_read_unlock();
1219 return segs;
1222 EXPORT_SYMBOL(skb_gso_segment);
1224 /* Take action when hardware reception checksum errors are detected. */
1225 #ifdef CONFIG_BUG
1226 void netdev_rx_csum_fault(struct net_device *dev)
1228 if (net_ratelimit()) {
1229 printk(KERN_ERR "%s: hw csum failure.\n",
1230 dev ? dev->name : "<unknown>");
1231 dump_stack();
1234 EXPORT_SYMBOL(netdev_rx_csum_fault);
1235 #endif
1237 #ifdef CONFIG_HIGHMEM
1238 /* Actually, we should eliminate this check as soon as we know, that:
1239 * 1. IOMMU is present and allows to map all the memory.
1240 * 2. No high memory really exists on this machine.
1243 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1245 int i;
1247 if (dev->features & NETIF_F_HIGHDMA)
1248 return 0;
1250 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1251 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1252 return 1;
1254 return 0;
1256 #else
1257 #define illegal_highdma(dev, skb) (0)
1258 #endif
1260 struct dev_gso_cb {
1261 void (*destructor)(struct sk_buff *skb);
1264 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1266 static void dev_gso_skb_destructor(struct sk_buff *skb)
1268 struct dev_gso_cb *cb;
1270 do {
1271 struct sk_buff *nskb = skb->next;
1273 skb->next = nskb->next;
1274 nskb->next = NULL;
1275 kfree_skb(nskb);
1276 } while (skb->next);
1278 cb = DEV_GSO_CB(skb);
1279 if (cb->destructor)
1280 cb->destructor(skb);
1284 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1285 * @skb: buffer to segment
1287 * This function segments the given skb and stores the list of segments
1288 * in skb->next.
1290 static int dev_gso_segment(struct sk_buff *skb)
1292 struct net_device *dev = skb->dev;
1293 struct sk_buff *segs;
1295 segs = skb_gso_segment(skb, dev->features & NETIF_F_SG &&
1296 !illegal_highdma(dev, skb));
1297 if (unlikely(IS_ERR(segs)))
1298 return PTR_ERR(segs);
1300 skb->next = segs;
1301 DEV_GSO_CB(skb)->destructor = skb->destructor;
1302 skb->destructor = dev_gso_skb_destructor;
1304 return 0;
1307 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1309 if (likely(!skb->next)) {
1310 if (netdev_nit)
1311 dev_queue_xmit_nit(skb, dev);
1313 if (!netif_needs_gso(dev, skb))
1314 return dev->hard_start_xmit(skb, dev);
1316 if (unlikely(dev_gso_segment(skb)))
1317 goto out_kfree_skb;
1320 do {
1321 struct sk_buff *nskb = skb->next;
1322 int rc;
1324 skb->next = nskb->next;
1325 nskb->next = NULL;
1326 rc = dev->hard_start_xmit(nskb, dev);
1327 if (unlikely(rc)) {
1328 nskb->next = skb->next;
1329 skb->next = nskb;
1330 return rc;
1332 if (unlikely(netif_queue_stopped(dev) && skb->next))
1333 return NETDEV_TX_BUSY;
1334 } while (skb->next);
1336 skb->destructor = DEV_GSO_CB(skb)->destructor;
1338 out_kfree_skb:
1339 kfree_skb(skb);
1340 return 0;
1343 #define HARD_TX_LOCK(dev, cpu) { \
1344 if ((dev->features & NETIF_F_LLTX) == 0) { \
1345 netif_tx_lock(dev); \
1349 #define HARD_TX_UNLOCK(dev) { \
1350 if ((dev->features & NETIF_F_LLTX) == 0) { \
1351 netif_tx_unlock(dev); \
1356 * dev_queue_xmit - transmit a buffer
1357 * @skb: buffer to transmit
1359 * Queue a buffer for transmission to a network device. The caller must
1360 * have set the device and priority and built the buffer before calling
1361 * this function. The function can be called from an interrupt.
1363 * A negative errno code is returned on a failure. A success does not
1364 * guarantee the frame will be transmitted as it may be dropped due
1365 * to congestion or traffic shaping.
1367 * -----------------------------------------------------------------------------------
1368 * I notice this method can also return errors from the queue disciplines,
1369 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1370 * be positive.
1372 * Regardless of the return value, the skb is consumed, so it is currently
1373 * difficult to retry a send to this method. (You can bump the ref count
1374 * before sending to hold a reference for retry if you are careful.)
1376 * When calling this method, interrupts MUST be enabled. This is because
1377 * the BH enable code must have IRQs enabled so that it will not deadlock.
1378 * --BLG
1381 int dev_queue_xmit(struct sk_buff *skb)
1383 struct net_device *dev = skb->dev;
1384 struct Qdisc *q;
1385 int rc = -ENOMEM;
1387 /* GSO will handle the following emulations directly. */
1388 if (netif_needs_gso(dev, skb))
1389 goto gso;
1391 if (skb_shinfo(skb)->frag_list &&
1392 !(dev->features & NETIF_F_FRAGLIST) &&
1393 __skb_linearize(skb))
1394 goto out_kfree_skb;
1396 /* Fragmented skb is linearized if device does not support SG,
1397 * or if at least one of fragments is in highmem and device
1398 * does not support DMA from it.
1400 if (skb_shinfo(skb)->nr_frags &&
1401 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1402 __skb_linearize(skb))
1403 goto out_kfree_skb;
1405 /* If packet is not checksummed and device does not support
1406 * checksumming for this protocol, complete checksumming here.
1408 if (skb->ip_summed == CHECKSUM_HW &&
1409 (!(dev->features & NETIF_F_GEN_CSUM) &&
1410 (!(dev->features & NETIF_F_IP_CSUM) ||
1411 skb->protocol != htons(ETH_P_IP))))
1412 if (skb_checksum_help(skb, 0))
1413 goto out_kfree_skb;
1415 gso:
1416 spin_lock_prefetch(&dev->queue_lock);
1418 /* Disable soft irqs for various locks below. Also
1419 * stops preemption for RCU.
1421 rcu_read_lock_bh();
1423 /* Updates of qdisc are serialized by queue_lock.
1424 * The struct Qdisc which is pointed to by qdisc is now a
1425 * rcu structure - it may be accessed without acquiring
1426 * a lock (but the structure may be stale.) The freeing of the
1427 * qdisc will be deferred until it's known that there are no
1428 * more references to it.
1430 * If the qdisc has an enqueue function, we still need to
1431 * hold the queue_lock before calling it, since queue_lock
1432 * also serializes access to the device queue.
1435 q = rcu_dereference(dev->qdisc);
1436 #ifdef CONFIG_NET_CLS_ACT
1437 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1438 #endif
1439 if (q->enqueue) {
1440 /* Grab device queue */
1441 spin_lock(&dev->queue_lock);
1443 rc = q->enqueue(skb, q);
1445 qdisc_run(dev);
1447 spin_unlock(&dev->queue_lock);
1448 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1449 goto out;
1452 /* The device has no queue. Common case for software devices:
1453 loopback, all the sorts of tunnels...
1455 Really, it is unlikely that netif_tx_lock protection is necessary
1456 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1457 counters.)
1458 However, it is possible, that they rely on protection
1459 made by us here.
1461 Check this and shot the lock. It is not prone from deadlocks.
1462 Either shot noqueue qdisc, it is even simpler 8)
1464 if (dev->flags & IFF_UP) {
1465 int cpu = smp_processor_id(); /* ok because BHs are off */
1467 if (dev->xmit_lock_owner != cpu) {
1469 HARD_TX_LOCK(dev, cpu);
1471 if (!netif_queue_stopped(dev)) {
1472 rc = 0;
1473 if (!dev_hard_start_xmit(skb, dev)) {
1474 HARD_TX_UNLOCK(dev);
1475 goto out;
1478 HARD_TX_UNLOCK(dev);
1479 if (net_ratelimit())
1480 printk(KERN_CRIT "Virtual device %s asks to "
1481 "queue packet!\n", dev->name);
1482 } else {
1483 /* Recursion is detected! It is possible,
1484 * unfortunately */
1485 if (net_ratelimit())
1486 printk(KERN_CRIT "Dead loop on virtual device "
1487 "%s, fix it urgently!\n", dev->name);
1491 rc = -ENETDOWN;
1492 rcu_read_unlock_bh();
1494 out_kfree_skb:
1495 kfree_skb(skb);
1496 return rc;
1497 out:
1498 rcu_read_unlock_bh();
1499 return rc;
1503 /*=======================================================================
1504 Receiver routines
1505 =======================================================================*/
1507 int netdev_max_backlog = 1000;
1508 int netdev_budget = 300;
1509 int weight_p = 64; /* old backlog weight */
1511 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1515 * netif_rx - post buffer to the network code
1516 * @skb: buffer to post
1518 * This function receives a packet from a device driver and queues it for
1519 * the upper (protocol) levels to process. It always succeeds. The buffer
1520 * may be dropped during processing for congestion control or by the
1521 * protocol layers.
1523 * return values:
1524 * NET_RX_SUCCESS (no congestion)
1525 * NET_RX_CN_LOW (low congestion)
1526 * NET_RX_CN_MOD (moderate congestion)
1527 * NET_RX_CN_HIGH (high congestion)
1528 * NET_RX_DROP (packet was dropped)
1532 int netif_rx(struct sk_buff *skb)
1534 struct softnet_data *queue;
1535 unsigned long flags;
1537 /* if netpoll wants it, pretend we never saw it */
1538 if (netpoll_rx(skb))
1539 return NET_RX_DROP;
1541 if (!skb->tstamp.off_sec)
1542 net_timestamp(skb);
1545 * The code is rearranged so that the path is the most
1546 * short when CPU is congested, but is still operating.
1548 local_irq_save(flags);
1549 queue = &__get_cpu_var(softnet_data);
1551 __get_cpu_var(netdev_rx_stat).total++;
1552 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1553 if (queue->input_pkt_queue.qlen) {
1554 enqueue:
1555 dev_hold(skb->dev);
1556 __skb_queue_tail(&queue->input_pkt_queue, skb);
1557 local_irq_restore(flags);
1558 return NET_RX_SUCCESS;
1561 netif_rx_schedule(&queue->backlog_dev);
1562 goto enqueue;
1565 __get_cpu_var(netdev_rx_stat).dropped++;
1566 local_irq_restore(flags);
1568 kfree_skb(skb);
1569 return NET_RX_DROP;
1572 int netif_rx_ni(struct sk_buff *skb)
1574 int err;
1576 preempt_disable();
1577 err = netif_rx(skb);
1578 if (local_softirq_pending())
1579 do_softirq();
1580 preempt_enable();
1582 return err;
1585 EXPORT_SYMBOL(netif_rx_ni);
1587 static inline struct net_device *skb_bond(struct sk_buff *skb)
1589 struct net_device *dev = skb->dev;
1591 if (dev->master) {
1593 * On bonding slaves other than the currently active
1594 * slave, suppress duplicates except for 802.3ad
1595 * ETH_P_SLOW and alb non-mcast/bcast.
1597 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
1598 if (dev->master->priv_flags & IFF_MASTER_ALB) {
1599 if (skb->pkt_type != PACKET_BROADCAST &&
1600 skb->pkt_type != PACKET_MULTICAST)
1601 goto keep;
1604 if (dev->master->priv_flags & IFF_MASTER_8023AD &&
1605 skb->protocol == __constant_htons(ETH_P_SLOW))
1606 goto keep;
1608 kfree_skb(skb);
1609 return NULL;
1611 keep:
1612 skb->dev = dev->master;
1615 return dev;
1618 static void net_tx_action(struct softirq_action *h)
1620 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1622 if (sd->completion_queue) {
1623 struct sk_buff *clist;
1625 local_irq_disable();
1626 clist = sd->completion_queue;
1627 sd->completion_queue = NULL;
1628 local_irq_enable();
1630 while (clist) {
1631 struct sk_buff *skb = clist;
1632 clist = clist->next;
1634 BUG_TRAP(!atomic_read(&skb->users));
1635 __kfree_skb(skb);
1639 if (sd->output_queue) {
1640 struct net_device *head;
1642 local_irq_disable();
1643 head = sd->output_queue;
1644 sd->output_queue = NULL;
1645 local_irq_enable();
1647 while (head) {
1648 struct net_device *dev = head;
1649 head = head->next_sched;
1651 smp_mb__before_clear_bit();
1652 clear_bit(__LINK_STATE_SCHED, &dev->state);
1654 if (spin_trylock(&dev->queue_lock)) {
1655 qdisc_run(dev);
1656 spin_unlock(&dev->queue_lock);
1657 } else {
1658 netif_schedule(dev);
1664 static __inline__ int deliver_skb(struct sk_buff *skb,
1665 struct packet_type *pt_prev,
1666 struct net_device *orig_dev)
1668 atomic_inc(&skb->users);
1669 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1672 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1673 int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
1674 struct net_bridge;
1675 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1676 unsigned char *addr);
1677 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
1679 static __inline__ int handle_bridge(struct sk_buff **pskb,
1680 struct packet_type **pt_prev, int *ret,
1681 struct net_device *orig_dev)
1683 struct net_bridge_port *port;
1685 if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
1686 (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
1687 return 0;
1689 if (*pt_prev) {
1690 *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
1691 *pt_prev = NULL;
1694 return br_handle_frame_hook(port, pskb);
1696 #else
1697 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1698 #endif
1700 #ifdef CONFIG_NET_CLS_ACT
1701 /* TODO: Maybe we should just force sch_ingress to be compiled in
1702 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1703 * a compare and 2 stores extra right now if we dont have it on
1704 * but have CONFIG_NET_CLS_ACT
1705 * NOTE: This doesnt stop any functionality; if you dont have
1706 * the ingress scheduler, you just cant add policies on ingress.
1709 static int ing_filter(struct sk_buff *skb)
1711 struct Qdisc *q;
1712 struct net_device *dev = skb->dev;
1713 int result = TC_ACT_OK;
1715 if (dev->qdisc_ingress) {
1716 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1717 if (MAX_RED_LOOP < ttl++) {
1718 printk("Redir loop detected Dropping packet (%s->%s)\n",
1719 skb->input_dev->name, skb->dev->name);
1720 return TC_ACT_SHOT;
1723 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1725 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1727 spin_lock(&dev->ingress_lock);
1728 if ((q = dev->qdisc_ingress) != NULL)
1729 result = q->enqueue(skb, q);
1730 spin_unlock(&dev->ingress_lock);
1734 return result;
1736 #endif
1738 int netif_receive_skb(struct sk_buff *skb)
1740 struct packet_type *ptype, *pt_prev;
1741 struct net_device *orig_dev;
1742 int ret = NET_RX_DROP;
1743 unsigned short type;
1745 /* if we've gotten here through NAPI, check netpoll */
1746 if (skb->dev->poll && netpoll_rx(skb))
1747 return NET_RX_DROP;
1749 if (!skb->tstamp.off_sec)
1750 net_timestamp(skb);
1752 if (!skb->input_dev)
1753 skb->input_dev = skb->dev;
1755 orig_dev = skb_bond(skb);
1757 if (!orig_dev)
1758 return NET_RX_DROP;
1760 __get_cpu_var(netdev_rx_stat).total++;
1762 skb->h.raw = skb->nh.raw = skb->data;
1763 skb->mac_len = skb->nh.raw - skb->mac.raw;
1765 pt_prev = NULL;
1767 rcu_read_lock();
1769 #ifdef CONFIG_NET_CLS_ACT
1770 if (skb->tc_verd & TC_NCLS) {
1771 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1772 goto ncls;
1774 #endif
1776 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1777 if (!ptype->dev || ptype->dev == skb->dev) {
1778 if (pt_prev)
1779 ret = deliver_skb(skb, pt_prev, orig_dev);
1780 pt_prev = ptype;
1784 #ifdef CONFIG_NET_CLS_ACT
1785 if (pt_prev) {
1786 ret = deliver_skb(skb, pt_prev, orig_dev);
1787 pt_prev = NULL; /* noone else should process this after*/
1788 } else {
1789 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1792 ret = ing_filter(skb);
1794 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1795 kfree_skb(skb);
1796 goto out;
1799 skb->tc_verd = 0;
1800 ncls:
1801 #endif
1803 handle_diverter(skb);
1805 if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
1806 goto out;
1808 type = skb->protocol;
1809 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1810 if (ptype->type == type &&
1811 (!ptype->dev || ptype->dev == skb->dev)) {
1812 if (pt_prev)
1813 ret = deliver_skb(skb, pt_prev, orig_dev);
1814 pt_prev = ptype;
1818 if (pt_prev) {
1819 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1820 } else {
1821 kfree_skb(skb);
1822 /* Jamal, now you will not able to escape explaining
1823 * me how you were going to use this. :-)
1825 ret = NET_RX_DROP;
1828 out:
1829 rcu_read_unlock();
1830 return ret;
1833 static int process_backlog(struct net_device *backlog_dev, int *budget)
1835 int work = 0;
1836 int quota = min(backlog_dev->quota, *budget);
1837 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1838 unsigned long start_time = jiffies;
1840 backlog_dev->weight = weight_p;
1841 for (;;) {
1842 struct sk_buff *skb;
1843 struct net_device *dev;
1845 local_irq_disable();
1846 skb = __skb_dequeue(&queue->input_pkt_queue);
1847 if (!skb)
1848 goto job_done;
1849 local_irq_enable();
1851 dev = skb->dev;
1853 netif_receive_skb(skb);
1855 dev_put(dev);
1857 work++;
1859 if (work >= quota || jiffies - start_time > 1)
1860 break;
1864 backlog_dev->quota -= work;
1865 *budget -= work;
1866 return -1;
1868 job_done:
1869 backlog_dev->quota -= work;
1870 *budget -= work;
1872 list_del(&backlog_dev->poll_list);
1873 smp_mb__before_clear_bit();
1874 netif_poll_enable(backlog_dev);
1876 local_irq_enable();
1877 return 0;
1880 static void net_rx_action(struct softirq_action *h)
1882 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1883 unsigned long start_time = jiffies;
1884 int budget = netdev_budget;
1885 void *have;
1887 local_irq_disable();
1889 while (!list_empty(&queue->poll_list)) {
1890 struct net_device *dev;
1892 if (budget <= 0 || jiffies - start_time > 1)
1893 goto softnet_break;
1895 local_irq_enable();
1897 dev = list_entry(queue->poll_list.next,
1898 struct net_device, poll_list);
1899 have = netpoll_poll_lock(dev);
1901 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
1902 netpoll_poll_unlock(have);
1903 local_irq_disable();
1904 list_move_tail(&dev->poll_list, &queue->poll_list);
1905 if (dev->quota < 0)
1906 dev->quota += dev->weight;
1907 else
1908 dev->quota = dev->weight;
1909 } else {
1910 netpoll_poll_unlock(have);
1911 dev_put(dev);
1912 local_irq_disable();
1915 out:
1916 #ifdef CONFIG_NET_DMA
1918 * There may not be any more sk_buffs coming right now, so push
1919 * any pending DMA copies to hardware
1921 if (net_dma_client) {
1922 struct dma_chan *chan;
1923 rcu_read_lock();
1924 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
1925 dma_async_memcpy_issue_pending(chan);
1926 rcu_read_unlock();
1928 #endif
1929 local_irq_enable();
1930 return;
1932 softnet_break:
1933 __get_cpu_var(netdev_rx_stat).time_squeeze++;
1934 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1935 goto out;
1938 static gifconf_func_t * gifconf_list [NPROTO];
1941 * register_gifconf - register a SIOCGIF handler
1942 * @family: Address family
1943 * @gifconf: Function handler
1945 * Register protocol dependent address dumping routines. The handler
1946 * that is passed must not be freed or reused until it has been replaced
1947 * by another handler.
1949 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
1951 if (family >= NPROTO)
1952 return -EINVAL;
1953 gifconf_list[family] = gifconf;
1954 return 0;
1959 * Map an interface index to its name (SIOCGIFNAME)
1963 * We need this ioctl for efficient implementation of the
1964 * if_indextoname() function required by the IPv6 API. Without
1965 * it, we would have to search all the interfaces to find a
1966 * match. --pb
1969 static int dev_ifname(struct ifreq __user *arg)
1971 struct net_device *dev;
1972 struct ifreq ifr;
1975 * Fetch the caller's info block.
1978 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
1979 return -EFAULT;
1981 read_lock(&dev_base_lock);
1982 dev = __dev_get_by_index(ifr.ifr_ifindex);
1983 if (!dev) {
1984 read_unlock(&dev_base_lock);
1985 return -ENODEV;
1988 strcpy(ifr.ifr_name, dev->name);
1989 read_unlock(&dev_base_lock);
1991 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
1992 return -EFAULT;
1993 return 0;
1997 * Perform a SIOCGIFCONF call. This structure will change
1998 * size eventually, and there is nothing I can do about it.
1999 * Thus we will need a 'compatibility mode'.
2002 static int dev_ifconf(char __user *arg)
2004 struct ifconf ifc;
2005 struct net_device *dev;
2006 char __user *pos;
2007 int len;
2008 int total;
2009 int i;
2012 * Fetch the caller's info block.
2015 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2016 return -EFAULT;
2018 pos = ifc.ifc_buf;
2019 len = ifc.ifc_len;
2022 * Loop over the interfaces, and write an info block for each.
2025 total = 0;
2026 for (dev = dev_base; dev; dev = dev->next) {
2027 for (i = 0; i < NPROTO; i++) {
2028 if (gifconf_list[i]) {
2029 int done;
2030 if (!pos)
2031 done = gifconf_list[i](dev, NULL, 0);
2032 else
2033 done = gifconf_list[i](dev, pos + total,
2034 len - total);
2035 if (done < 0)
2036 return -EFAULT;
2037 total += done;
2043 * All done. Write the updated control block back to the caller.
2045 ifc.ifc_len = total;
2048 * Both BSD and Solaris return 0 here, so we do too.
2050 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2053 #ifdef CONFIG_PROC_FS
2055 * This is invoked by the /proc filesystem handler to display a device
2056 * in detail.
2058 static __inline__ struct net_device *dev_get_idx(loff_t pos)
2060 struct net_device *dev;
2061 loff_t i;
2063 for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
2065 return i == pos ? dev : NULL;
2068 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2070 read_lock(&dev_base_lock);
2071 return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
2074 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2076 ++*pos;
2077 return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
2080 void dev_seq_stop(struct seq_file *seq, void *v)
2082 read_unlock(&dev_base_lock);
2085 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2087 if (dev->get_stats) {
2088 struct net_device_stats *stats = dev->get_stats(dev);
2090 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2091 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2092 dev->name, stats->rx_bytes, stats->rx_packets,
2093 stats->rx_errors,
2094 stats->rx_dropped + stats->rx_missed_errors,
2095 stats->rx_fifo_errors,
2096 stats->rx_length_errors + stats->rx_over_errors +
2097 stats->rx_crc_errors + stats->rx_frame_errors,
2098 stats->rx_compressed, stats->multicast,
2099 stats->tx_bytes, stats->tx_packets,
2100 stats->tx_errors, stats->tx_dropped,
2101 stats->tx_fifo_errors, stats->collisions,
2102 stats->tx_carrier_errors +
2103 stats->tx_aborted_errors +
2104 stats->tx_window_errors +
2105 stats->tx_heartbeat_errors,
2106 stats->tx_compressed);
2107 } else
2108 seq_printf(seq, "%6s: No statistics available.\n", dev->name);
2112 * Called from the PROCfs module. This now uses the new arbitrary sized
2113 * /proc/net interface to create /proc/net/dev
2115 static int dev_seq_show(struct seq_file *seq, void *v)
2117 if (v == SEQ_START_TOKEN)
2118 seq_puts(seq, "Inter-| Receive "
2119 " | Transmit\n"
2120 " face |bytes packets errs drop fifo frame "
2121 "compressed multicast|bytes packets errs "
2122 "drop fifo colls carrier compressed\n");
2123 else
2124 dev_seq_printf_stats(seq, v);
2125 return 0;
2128 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2130 struct netif_rx_stats *rc = NULL;
2132 while (*pos < NR_CPUS)
2133 if (cpu_online(*pos)) {
2134 rc = &per_cpu(netdev_rx_stat, *pos);
2135 break;
2136 } else
2137 ++*pos;
2138 return rc;
2141 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2143 return softnet_get_online(pos);
2146 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2148 ++*pos;
2149 return softnet_get_online(pos);
2152 static void softnet_seq_stop(struct seq_file *seq, void *v)
2156 static int softnet_seq_show(struct seq_file *seq, void *v)
2158 struct netif_rx_stats *s = v;
2160 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2161 s->total, s->dropped, s->time_squeeze, 0,
2162 0, 0, 0, 0, /* was fastroute */
2163 s->cpu_collision );
2164 return 0;
2167 static struct seq_operations dev_seq_ops = {
2168 .start = dev_seq_start,
2169 .next = dev_seq_next,
2170 .stop = dev_seq_stop,
2171 .show = dev_seq_show,
2174 static int dev_seq_open(struct inode *inode, struct file *file)
2176 return seq_open(file, &dev_seq_ops);
2179 static struct file_operations dev_seq_fops = {
2180 .owner = THIS_MODULE,
2181 .open = dev_seq_open,
2182 .read = seq_read,
2183 .llseek = seq_lseek,
2184 .release = seq_release,
2187 static struct seq_operations softnet_seq_ops = {
2188 .start = softnet_seq_start,
2189 .next = softnet_seq_next,
2190 .stop = softnet_seq_stop,
2191 .show = softnet_seq_show,
2194 static int softnet_seq_open(struct inode *inode, struct file *file)
2196 return seq_open(file, &softnet_seq_ops);
2199 static struct file_operations softnet_seq_fops = {
2200 .owner = THIS_MODULE,
2201 .open = softnet_seq_open,
2202 .read = seq_read,
2203 .llseek = seq_lseek,
2204 .release = seq_release,
2207 #ifdef CONFIG_WIRELESS_EXT
2208 extern int wireless_proc_init(void);
2209 #else
2210 #define wireless_proc_init() 0
2211 #endif
2213 static int __init dev_proc_init(void)
2215 int rc = -ENOMEM;
2217 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2218 goto out;
2219 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2220 goto out_dev;
2221 if (wireless_proc_init())
2222 goto out_softnet;
2223 rc = 0;
2224 out:
2225 return rc;
2226 out_softnet:
2227 proc_net_remove("softnet_stat");
2228 out_dev:
2229 proc_net_remove("dev");
2230 goto out;
2232 #else
2233 #define dev_proc_init() 0
2234 #endif /* CONFIG_PROC_FS */
2238 * netdev_set_master - set up master/slave pair
2239 * @slave: slave device
2240 * @master: new master device
2242 * Changes the master device of the slave. Pass %NULL to break the
2243 * bonding. The caller must hold the RTNL semaphore. On a failure
2244 * a negative errno code is returned. On success the reference counts
2245 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2246 * function returns zero.
2248 int netdev_set_master(struct net_device *slave, struct net_device *master)
2250 struct net_device *old = slave->master;
2252 ASSERT_RTNL();
2254 if (master) {
2255 if (old)
2256 return -EBUSY;
2257 dev_hold(master);
2260 slave->master = master;
2262 synchronize_net();
2264 if (old)
2265 dev_put(old);
2267 if (master)
2268 slave->flags |= IFF_SLAVE;
2269 else
2270 slave->flags &= ~IFF_SLAVE;
2272 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2273 return 0;
2277 * dev_set_promiscuity - update promiscuity count on a device
2278 * @dev: device
2279 * @inc: modifier
2281 * Add or remove promiscuity from a device. While the count in the device
2282 * remains above zero the interface remains promiscuous. Once it hits zero
2283 * the device reverts back to normal filtering operation. A negative inc
2284 * value is used to drop promiscuity on the device.
2286 void dev_set_promiscuity(struct net_device *dev, int inc)
2288 unsigned short old_flags = dev->flags;
2290 if ((dev->promiscuity += inc) == 0)
2291 dev->flags &= ~IFF_PROMISC;
2292 else
2293 dev->flags |= IFF_PROMISC;
2294 if (dev->flags != old_flags) {
2295 dev_mc_upload(dev);
2296 printk(KERN_INFO "device %s %s promiscuous mode\n",
2297 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2298 "left");
2299 audit_log(current->audit_context, GFP_ATOMIC,
2300 AUDIT_ANOM_PROMISCUOUS,
2301 "dev=%s prom=%d old_prom=%d auid=%u",
2302 dev->name, (dev->flags & IFF_PROMISC),
2303 (old_flags & IFF_PROMISC),
2304 audit_get_loginuid(current->audit_context));
2309 * dev_set_allmulti - update allmulti count on a device
2310 * @dev: device
2311 * @inc: modifier
2313 * Add or remove reception of all multicast frames to a device. While the
2314 * count in the device remains above zero the interface remains listening
2315 * to all interfaces. Once it hits zero the device reverts back to normal
2316 * filtering operation. A negative @inc value is used to drop the counter
2317 * when releasing a resource needing all multicasts.
2320 void dev_set_allmulti(struct net_device *dev, int inc)
2322 unsigned short old_flags = dev->flags;
2324 dev->flags |= IFF_ALLMULTI;
2325 if ((dev->allmulti += inc) == 0)
2326 dev->flags &= ~IFF_ALLMULTI;
2327 if (dev->flags ^ old_flags)
2328 dev_mc_upload(dev);
2331 unsigned dev_get_flags(const struct net_device *dev)
2333 unsigned flags;
2335 flags = (dev->flags & ~(IFF_PROMISC |
2336 IFF_ALLMULTI |
2337 IFF_RUNNING |
2338 IFF_LOWER_UP |
2339 IFF_DORMANT)) |
2340 (dev->gflags & (IFF_PROMISC |
2341 IFF_ALLMULTI));
2343 if (netif_running(dev)) {
2344 if (netif_oper_up(dev))
2345 flags |= IFF_RUNNING;
2346 if (netif_carrier_ok(dev))
2347 flags |= IFF_LOWER_UP;
2348 if (netif_dormant(dev))
2349 flags |= IFF_DORMANT;
2352 return flags;
2355 int dev_change_flags(struct net_device *dev, unsigned flags)
2357 int ret;
2358 int old_flags = dev->flags;
2361 * Set the flags on our device.
2364 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2365 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2366 IFF_AUTOMEDIA)) |
2367 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2368 IFF_ALLMULTI));
2371 * Load in the correct multicast list now the flags have changed.
2374 dev_mc_upload(dev);
2377 * Have we downed the interface. We handle IFF_UP ourselves
2378 * according to user attempts to set it, rather than blindly
2379 * setting it.
2382 ret = 0;
2383 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2384 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2386 if (!ret)
2387 dev_mc_upload(dev);
2390 if (dev->flags & IFF_UP &&
2391 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2392 IFF_VOLATILE)))
2393 raw_notifier_call_chain(&netdev_chain,
2394 NETDEV_CHANGE, dev);
2396 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2397 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2398 dev->gflags ^= IFF_PROMISC;
2399 dev_set_promiscuity(dev, inc);
2402 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2403 is important. Some (broken) drivers set IFF_PROMISC, when
2404 IFF_ALLMULTI is requested not asking us and not reporting.
2406 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2407 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2408 dev->gflags ^= IFF_ALLMULTI;
2409 dev_set_allmulti(dev, inc);
2412 if (old_flags ^ dev->flags)
2413 rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
2415 return ret;
2418 int dev_set_mtu(struct net_device *dev, int new_mtu)
2420 int err;
2422 if (new_mtu == dev->mtu)
2423 return 0;
2425 /* MTU must be positive. */
2426 if (new_mtu < 0)
2427 return -EINVAL;
2429 if (!netif_device_present(dev))
2430 return -ENODEV;
2432 err = 0;
2433 if (dev->change_mtu)
2434 err = dev->change_mtu(dev, new_mtu);
2435 else
2436 dev->mtu = new_mtu;
2437 if (!err && dev->flags & IFF_UP)
2438 raw_notifier_call_chain(&netdev_chain,
2439 NETDEV_CHANGEMTU, dev);
2440 return err;
2443 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2445 int err;
2447 if (!dev->set_mac_address)
2448 return -EOPNOTSUPP;
2449 if (sa->sa_family != dev->type)
2450 return -EINVAL;
2451 if (!netif_device_present(dev))
2452 return -ENODEV;
2453 err = dev->set_mac_address(dev, sa);
2454 if (!err)
2455 raw_notifier_call_chain(&netdev_chain,
2456 NETDEV_CHANGEADDR, dev);
2457 return err;
2461 * Perform the SIOCxIFxxx calls.
2463 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2465 int err;
2466 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2468 if (!dev)
2469 return -ENODEV;
2471 switch (cmd) {
2472 case SIOCGIFFLAGS: /* Get interface flags */
2473 ifr->ifr_flags = dev_get_flags(dev);
2474 return 0;
2476 case SIOCSIFFLAGS: /* Set interface flags */
2477 return dev_change_flags(dev, ifr->ifr_flags);
2479 case SIOCGIFMETRIC: /* Get the metric on the interface
2480 (currently unused) */
2481 ifr->ifr_metric = 0;
2482 return 0;
2484 case SIOCSIFMETRIC: /* Set the metric on the interface
2485 (currently unused) */
2486 return -EOPNOTSUPP;
2488 case SIOCGIFMTU: /* Get the MTU of a device */
2489 ifr->ifr_mtu = dev->mtu;
2490 return 0;
2492 case SIOCSIFMTU: /* Set the MTU of a device */
2493 return dev_set_mtu(dev, ifr->ifr_mtu);
2495 case SIOCGIFHWADDR:
2496 if (!dev->addr_len)
2497 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2498 else
2499 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2500 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2501 ifr->ifr_hwaddr.sa_family = dev->type;
2502 return 0;
2504 case SIOCSIFHWADDR:
2505 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2507 case SIOCSIFHWBROADCAST:
2508 if (ifr->ifr_hwaddr.sa_family != dev->type)
2509 return -EINVAL;
2510 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2511 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2512 raw_notifier_call_chain(&netdev_chain,
2513 NETDEV_CHANGEADDR, dev);
2514 return 0;
2516 case SIOCGIFMAP:
2517 ifr->ifr_map.mem_start = dev->mem_start;
2518 ifr->ifr_map.mem_end = dev->mem_end;
2519 ifr->ifr_map.base_addr = dev->base_addr;
2520 ifr->ifr_map.irq = dev->irq;
2521 ifr->ifr_map.dma = dev->dma;
2522 ifr->ifr_map.port = dev->if_port;
2523 return 0;
2525 case SIOCSIFMAP:
2526 if (dev->set_config) {
2527 if (!netif_device_present(dev))
2528 return -ENODEV;
2529 return dev->set_config(dev, &ifr->ifr_map);
2531 return -EOPNOTSUPP;
2533 case SIOCADDMULTI:
2534 if (!dev->set_multicast_list ||
2535 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2536 return -EINVAL;
2537 if (!netif_device_present(dev))
2538 return -ENODEV;
2539 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2540 dev->addr_len, 1);
2542 case SIOCDELMULTI:
2543 if (!dev->set_multicast_list ||
2544 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2545 return -EINVAL;
2546 if (!netif_device_present(dev))
2547 return -ENODEV;
2548 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2549 dev->addr_len, 1);
2551 case SIOCGIFINDEX:
2552 ifr->ifr_ifindex = dev->ifindex;
2553 return 0;
2555 case SIOCGIFTXQLEN:
2556 ifr->ifr_qlen = dev->tx_queue_len;
2557 return 0;
2559 case SIOCSIFTXQLEN:
2560 if (ifr->ifr_qlen < 0)
2561 return -EINVAL;
2562 dev->tx_queue_len = ifr->ifr_qlen;
2563 return 0;
2565 case SIOCSIFNAME:
2566 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2567 return dev_change_name(dev, ifr->ifr_newname);
2570 * Unknown or private ioctl
2573 default:
2574 if ((cmd >= SIOCDEVPRIVATE &&
2575 cmd <= SIOCDEVPRIVATE + 15) ||
2576 cmd == SIOCBONDENSLAVE ||
2577 cmd == SIOCBONDRELEASE ||
2578 cmd == SIOCBONDSETHWADDR ||
2579 cmd == SIOCBONDSLAVEINFOQUERY ||
2580 cmd == SIOCBONDINFOQUERY ||
2581 cmd == SIOCBONDCHANGEACTIVE ||
2582 cmd == SIOCGMIIPHY ||
2583 cmd == SIOCGMIIREG ||
2584 cmd == SIOCSMIIREG ||
2585 cmd == SIOCBRADDIF ||
2586 cmd == SIOCBRDELIF ||
2587 cmd == SIOCWANDEV) {
2588 err = -EOPNOTSUPP;
2589 if (dev->do_ioctl) {
2590 if (netif_device_present(dev))
2591 err = dev->do_ioctl(dev, ifr,
2592 cmd);
2593 else
2594 err = -ENODEV;
2596 } else
2597 err = -EINVAL;
2600 return err;
2604 * This function handles all "interface"-type I/O control requests. The actual
2605 * 'doing' part of this is dev_ifsioc above.
2609 * dev_ioctl - network device ioctl
2610 * @cmd: command to issue
2611 * @arg: pointer to a struct ifreq in user space
2613 * Issue ioctl functions to devices. This is normally called by the
2614 * user space syscall interfaces but can sometimes be useful for
2615 * other purposes. The return value is the return from the syscall if
2616 * positive or a negative errno code on error.
2619 int dev_ioctl(unsigned int cmd, void __user *arg)
2621 struct ifreq ifr;
2622 int ret;
2623 char *colon;
2625 /* One special case: SIOCGIFCONF takes ifconf argument
2626 and requires shared lock, because it sleeps writing
2627 to user space.
2630 if (cmd == SIOCGIFCONF) {
2631 rtnl_lock();
2632 ret = dev_ifconf((char __user *) arg);
2633 rtnl_unlock();
2634 return ret;
2636 if (cmd == SIOCGIFNAME)
2637 return dev_ifname((struct ifreq __user *)arg);
2639 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2640 return -EFAULT;
2642 ifr.ifr_name[IFNAMSIZ-1] = 0;
2644 colon = strchr(ifr.ifr_name, ':');
2645 if (colon)
2646 *colon = 0;
2649 * See which interface the caller is talking about.
2652 switch (cmd) {
2654 * These ioctl calls:
2655 * - can be done by all.
2656 * - atomic and do not require locking.
2657 * - return a value
2659 case SIOCGIFFLAGS:
2660 case SIOCGIFMETRIC:
2661 case SIOCGIFMTU:
2662 case SIOCGIFHWADDR:
2663 case SIOCGIFSLAVE:
2664 case SIOCGIFMAP:
2665 case SIOCGIFINDEX:
2666 case SIOCGIFTXQLEN:
2667 dev_load(ifr.ifr_name);
2668 read_lock(&dev_base_lock);
2669 ret = dev_ifsioc(&ifr, cmd);
2670 read_unlock(&dev_base_lock);
2671 if (!ret) {
2672 if (colon)
2673 *colon = ':';
2674 if (copy_to_user(arg, &ifr,
2675 sizeof(struct ifreq)))
2676 ret = -EFAULT;
2678 return ret;
2680 case SIOCETHTOOL:
2681 dev_load(ifr.ifr_name);
2682 rtnl_lock();
2683 ret = dev_ethtool(&ifr);
2684 rtnl_unlock();
2685 if (!ret) {
2686 if (colon)
2687 *colon = ':';
2688 if (copy_to_user(arg, &ifr,
2689 sizeof(struct ifreq)))
2690 ret = -EFAULT;
2692 return ret;
2695 * These ioctl calls:
2696 * - require superuser power.
2697 * - require strict serialization.
2698 * - return a value
2700 case SIOCGMIIPHY:
2701 case SIOCGMIIREG:
2702 case SIOCSIFNAME:
2703 if (!capable(CAP_NET_ADMIN))
2704 return -EPERM;
2705 dev_load(ifr.ifr_name);
2706 rtnl_lock();
2707 ret = dev_ifsioc(&ifr, cmd);
2708 rtnl_unlock();
2709 if (!ret) {
2710 if (colon)
2711 *colon = ':';
2712 if (copy_to_user(arg, &ifr,
2713 sizeof(struct ifreq)))
2714 ret = -EFAULT;
2716 return ret;
2719 * These ioctl calls:
2720 * - require superuser power.
2721 * - require strict serialization.
2722 * - do not return a value
2724 case SIOCSIFFLAGS:
2725 case SIOCSIFMETRIC:
2726 case SIOCSIFMTU:
2727 case SIOCSIFMAP:
2728 case SIOCSIFHWADDR:
2729 case SIOCSIFSLAVE:
2730 case SIOCADDMULTI:
2731 case SIOCDELMULTI:
2732 case SIOCSIFHWBROADCAST:
2733 case SIOCSIFTXQLEN:
2734 case SIOCSMIIREG:
2735 case SIOCBONDENSLAVE:
2736 case SIOCBONDRELEASE:
2737 case SIOCBONDSETHWADDR:
2738 case SIOCBONDCHANGEACTIVE:
2739 case SIOCBRADDIF:
2740 case SIOCBRDELIF:
2741 if (!capable(CAP_NET_ADMIN))
2742 return -EPERM;
2743 /* fall through */
2744 case SIOCBONDSLAVEINFOQUERY:
2745 case SIOCBONDINFOQUERY:
2746 dev_load(ifr.ifr_name);
2747 rtnl_lock();
2748 ret = dev_ifsioc(&ifr, cmd);
2749 rtnl_unlock();
2750 return ret;
2752 case SIOCGIFMEM:
2753 /* Get the per device memory space. We can add this but
2754 * currently do not support it */
2755 case SIOCSIFMEM:
2756 /* Set the per device memory buffer space.
2757 * Not applicable in our case */
2758 case SIOCSIFLINK:
2759 return -EINVAL;
2762 * Unknown or private ioctl.
2764 default:
2765 if (cmd == SIOCWANDEV ||
2766 (cmd >= SIOCDEVPRIVATE &&
2767 cmd <= SIOCDEVPRIVATE + 15)) {
2768 dev_load(ifr.ifr_name);
2769 rtnl_lock();
2770 ret = dev_ifsioc(&ifr, cmd);
2771 rtnl_unlock();
2772 if (!ret && copy_to_user(arg, &ifr,
2773 sizeof(struct ifreq)))
2774 ret = -EFAULT;
2775 return ret;
2777 #ifdef CONFIG_WIRELESS_EXT
2778 /* Take care of Wireless Extensions */
2779 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
2780 /* If command is `set a parameter', or
2781 * `get the encoding parameters', check if
2782 * the user has the right to do it */
2783 if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE
2784 || cmd == SIOCGIWENCODEEXT) {
2785 if (!capable(CAP_NET_ADMIN))
2786 return -EPERM;
2788 dev_load(ifr.ifr_name);
2789 rtnl_lock();
2790 /* Follow me in net/core/wireless.c */
2791 ret = wireless_process_ioctl(&ifr, cmd);
2792 rtnl_unlock();
2793 if (IW_IS_GET(cmd) &&
2794 copy_to_user(arg, &ifr,
2795 sizeof(struct ifreq)))
2796 ret = -EFAULT;
2797 return ret;
2799 #endif /* CONFIG_WIRELESS_EXT */
2800 return -EINVAL;
2806 * dev_new_index - allocate an ifindex
2808 * Returns a suitable unique value for a new device interface
2809 * number. The caller must hold the rtnl semaphore or the
2810 * dev_base_lock to be sure it remains unique.
2812 static int dev_new_index(void)
2814 static int ifindex;
2815 for (;;) {
2816 if (++ifindex <= 0)
2817 ifindex = 1;
2818 if (!__dev_get_by_index(ifindex))
2819 return ifindex;
2823 static int dev_boot_phase = 1;
2825 /* Delayed registration/unregisteration */
2826 static DEFINE_SPINLOCK(net_todo_list_lock);
2827 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
2829 static inline void net_set_todo(struct net_device *dev)
2831 spin_lock(&net_todo_list_lock);
2832 list_add_tail(&dev->todo_list, &net_todo_list);
2833 spin_unlock(&net_todo_list_lock);
2837 * register_netdevice - register a network device
2838 * @dev: device to register
2840 * Take a completed network device structure and add it to the kernel
2841 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2842 * chain. 0 is returned on success. A negative errno code is returned
2843 * on a failure to set up the device, or if the name is a duplicate.
2845 * Callers must hold the rtnl semaphore. You may want
2846 * register_netdev() instead of this.
2848 * BUGS:
2849 * The locking appears insufficient to guarantee two parallel registers
2850 * will not get the same name.
2853 int register_netdevice(struct net_device *dev)
2855 struct hlist_head *head;
2856 struct hlist_node *p;
2857 int ret;
2859 BUG_ON(dev_boot_phase);
2860 ASSERT_RTNL();
2862 might_sleep();
2864 /* When net_device's are persistent, this will be fatal. */
2865 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
2867 spin_lock_init(&dev->queue_lock);
2868 spin_lock_init(&dev->_xmit_lock);
2869 dev->xmit_lock_owner = -1;
2870 #ifdef CONFIG_NET_CLS_ACT
2871 spin_lock_init(&dev->ingress_lock);
2872 #endif
2874 ret = alloc_divert_blk(dev);
2875 if (ret)
2876 goto out;
2878 dev->iflink = -1;
2880 /* Init, if this function is available */
2881 if (dev->init) {
2882 ret = dev->init(dev);
2883 if (ret) {
2884 if (ret > 0)
2885 ret = -EIO;
2886 goto out_err;
2890 if (!dev_valid_name(dev->name)) {
2891 ret = -EINVAL;
2892 goto out_err;
2895 dev->ifindex = dev_new_index();
2896 if (dev->iflink == -1)
2897 dev->iflink = dev->ifindex;
2899 /* Check for existence of name */
2900 head = dev_name_hash(dev->name);
2901 hlist_for_each(p, head) {
2902 struct net_device *d
2903 = hlist_entry(p, struct net_device, name_hlist);
2904 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
2905 ret = -EEXIST;
2906 goto out_err;
2910 /* Fix illegal SG+CSUM combinations. */
2911 if ((dev->features & NETIF_F_SG) &&
2912 !(dev->features & NETIF_F_ALL_CSUM)) {
2913 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2914 dev->name);
2915 dev->features &= ~NETIF_F_SG;
2918 /* TSO requires that SG is present as well. */
2919 if ((dev->features & NETIF_F_TSO) &&
2920 !(dev->features & NETIF_F_SG)) {
2921 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2922 dev->name);
2923 dev->features &= ~NETIF_F_TSO;
2925 if (dev->features & NETIF_F_UFO) {
2926 if (!(dev->features & NETIF_F_HW_CSUM)) {
2927 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2928 "NETIF_F_HW_CSUM feature.\n",
2929 dev->name);
2930 dev->features &= ~NETIF_F_UFO;
2932 if (!(dev->features & NETIF_F_SG)) {
2933 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
2934 "NETIF_F_SG feature.\n",
2935 dev->name);
2936 dev->features &= ~NETIF_F_UFO;
2941 * nil rebuild_header routine,
2942 * that should be never called and used as just bug trap.
2945 if (!dev->rebuild_header)
2946 dev->rebuild_header = default_rebuild_header;
2948 ret = netdev_register_sysfs(dev);
2949 if (ret)
2950 goto out_err;
2951 dev->reg_state = NETREG_REGISTERED;
2954 * Default initial state at registry is that the
2955 * device is present.
2958 set_bit(__LINK_STATE_PRESENT, &dev->state);
2960 dev->next = NULL;
2961 dev_init_scheduler(dev);
2962 write_lock_bh(&dev_base_lock);
2963 *dev_tail = dev;
2964 dev_tail = &dev->next;
2965 hlist_add_head(&dev->name_hlist, head);
2966 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
2967 dev_hold(dev);
2968 write_unlock_bh(&dev_base_lock);
2970 /* Notify protocols, that a new device appeared. */
2971 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
2973 ret = 0;
2975 out:
2976 return ret;
2977 out_err:
2978 free_divert_blk(dev);
2979 goto out;
2983 * register_netdev - 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 * This is a wrapper around register_netdev that takes the rtnl semaphore
2992 * and expands the device name if you passed a format string to
2993 * alloc_netdev.
2995 int register_netdev(struct net_device *dev)
2997 int err;
2999 rtnl_lock();
3002 * If the name is a format string the caller wants us to do a
3003 * name allocation.
3005 if (strchr(dev->name, '%')) {
3006 err = dev_alloc_name(dev, dev->name);
3007 if (err < 0)
3008 goto out;
3012 * Back compatibility hook. Kill this one in 2.5
3014 if (dev->name[0] == 0 || dev->name[0] == ' ') {
3015 err = dev_alloc_name(dev, "eth%d");
3016 if (err < 0)
3017 goto out;
3020 err = register_netdevice(dev);
3021 out:
3022 rtnl_unlock();
3023 return err;
3025 EXPORT_SYMBOL(register_netdev);
3028 * netdev_wait_allrefs - wait until all references are gone.
3030 * This is called when unregistering network devices.
3032 * Any protocol or device that holds a reference should register
3033 * for netdevice notification, and cleanup and put back the
3034 * reference if they receive an UNREGISTER event.
3035 * We can get stuck here if buggy protocols don't correctly
3036 * call dev_put.
3038 static void netdev_wait_allrefs(struct net_device *dev)
3040 unsigned long rebroadcast_time, warning_time;
3042 rebroadcast_time = warning_time = jiffies;
3043 while (atomic_read(&dev->refcnt) != 0) {
3044 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3045 rtnl_lock();
3047 /* Rebroadcast unregister notification */
3048 raw_notifier_call_chain(&netdev_chain,
3049 NETDEV_UNREGISTER, dev);
3051 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3052 &dev->state)) {
3053 /* We must not have linkwatch events
3054 * pending on unregister. If this
3055 * happens, we simply run the queue
3056 * unscheduled, resulting in a noop
3057 * for this device.
3059 linkwatch_run_queue();
3062 __rtnl_unlock();
3064 rebroadcast_time = jiffies;
3067 msleep(250);
3069 if (time_after(jiffies, warning_time + 10 * HZ)) {
3070 printk(KERN_EMERG "unregister_netdevice: "
3071 "waiting for %s to become free. Usage "
3072 "count = %d\n",
3073 dev->name, atomic_read(&dev->refcnt));
3074 warning_time = jiffies;
3079 /* The sequence is:
3081 * rtnl_lock();
3082 * ...
3083 * register_netdevice(x1);
3084 * register_netdevice(x2);
3085 * ...
3086 * unregister_netdevice(y1);
3087 * unregister_netdevice(y2);
3088 * ...
3089 * rtnl_unlock();
3090 * free_netdev(y1);
3091 * free_netdev(y2);
3093 * We are invoked by rtnl_unlock() after it drops the semaphore.
3094 * This allows us to deal with problems:
3095 * 1) We can delete sysfs objects which invoke hotplug
3096 * without deadlocking with linkwatch via keventd.
3097 * 2) Since we run with the RTNL semaphore not held, we can sleep
3098 * safely in order to wait for the netdev refcnt to drop to zero.
3100 static DEFINE_MUTEX(net_todo_run_mutex);
3101 void netdev_run_todo(void)
3103 struct list_head list;
3105 /* Need to guard against multiple cpu's getting out of order. */
3106 mutex_lock(&net_todo_run_mutex);
3108 /* Not safe to do outside the semaphore. We must not return
3109 * until all unregister events invoked by the local processor
3110 * have been completed (either by this todo run, or one on
3111 * another cpu).
3113 if (list_empty(&net_todo_list))
3114 goto out;
3116 /* Snapshot list, allow later requests */
3117 spin_lock(&net_todo_list_lock);
3118 list_replace_init(&net_todo_list, &list);
3119 spin_unlock(&net_todo_list_lock);
3121 while (!list_empty(&list)) {
3122 struct net_device *dev
3123 = list_entry(list.next, struct net_device, todo_list);
3124 list_del(&dev->todo_list);
3126 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3127 printk(KERN_ERR "network todo '%s' but state %d\n",
3128 dev->name, dev->reg_state);
3129 dump_stack();
3130 continue;
3133 netdev_unregister_sysfs(dev);
3134 dev->reg_state = NETREG_UNREGISTERED;
3136 netdev_wait_allrefs(dev);
3138 /* paranoia */
3139 BUG_ON(atomic_read(&dev->refcnt));
3140 BUG_TRAP(!dev->ip_ptr);
3141 BUG_TRAP(!dev->ip6_ptr);
3142 BUG_TRAP(!dev->dn_ptr);
3144 /* It must be the very last action,
3145 * after this 'dev' may point to freed up memory.
3147 if (dev->destructor)
3148 dev->destructor(dev);
3151 out:
3152 mutex_unlock(&net_todo_run_mutex);
3156 * alloc_netdev - allocate network device
3157 * @sizeof_priv: size of private data to allocate space for
3158 * @name: device name format string
3159 * @setup: callback to initialize device
3161 * Allocates a struct net_device with private data area for driver use
3162 * and performs basic initialization.
3164 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3165 void (*setup)(struct net_device *))
3167 void *p;
3168 struct net_device *dev;
3169 int alloc_size;
3171 /* ensure 32-byte alignment of both the device and private area */
3172 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3173 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3175 p = kzalloc(alloc_size, GFP_KERNEL);
3176 if (!p) {
3177 printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
3178 return NULL;
3181 dev = (struct net_device *)
3182 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3183 dev->padded = (char *)dev - (char *)p;
3185 if (sizeof_priv)
3186 dev->priv = netdev_priv(dev);
3188 setup(dev);
3189 strcpy(dev->name, name);
3190 return dev;
3192 EXPORT_SYMBOL(alloc_netdev);
3195 * free_netdev - free network device
3196 * @dev: device
3198 * This function does the last stage of destroying an allocated device
3199 * interface. The reference to the device object is released.
3200 * If this is the last reference then it will be freed.
3202 void free_netdev(struct net_device *dev)
3204 #ifdef CONFIG_SYSFS
3205 /* Compatibility with error handling in drivers */
3206 if (dev->reg_state == NETREG_UNINITIALIZED) {
3207 kfree((char *)dev - dev->padded);
3208 return;
3211 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3212 dev->reg_state = NETREG_RELEASED;
3214 /* will free via class release */
3215 class_device_put(&dev->class_dev);
3216 #else
3217 kfree((char *)dev - dev->padded);
3218 #endif
3221 /* Synchronize with packet receive processing. */
3222 void synchronize_net(void)
3224 might_sleep();
3225 synchronize_rcu();
3229 * unregister_netdevice - remove device from the kernel
3230 * @dev: device
3232 * This function shuts down a device interface and removes it
3233 * from the kernel tables. On success 0 is returned, on a failure
3234 * a negative errno code is returned.
3236 * Callers must hold the rtnl semaphore. You may want
3237 * unregister_netdev() instead of this.
3240 int unregister_netdevice(struct net_device *dev)
3242 struct net_device *d, **dp;
3244 BUG_ON(dev_boot_phase);
3245 ASSERT_RTNL();
3247 /* Some devices call without registering for initialization unwind. */
3248 if (dev->reg_state == NETREG_UNINITIALIZED) {
3249 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3250 "was registered\n", dev->name, dev);
3251 return -ENODEV;
3254 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3256 /* If device is running, close it first. */
3257 if (dev->flags & IFF_UP)
3258 dev_close(dev);
3260 /* And unlink it from device chain. */
3261 for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
3262 if (d == dev) {
3263 write_lock_bh(&dev_base_lock);
3264 hlist_del(&dev->name_hlist);
3265 hlist_del(&dev->index_hlist);
3266 if (dev_tail == &dev->next)
3267 dev_tail = dp;
3268 *dp = d->next;
3269 write_unlock_bh(&dev_base_lock);
3270 break;
3273 if (!d) {
3274 printk(KERN_ERR "unregister net_device: '%s' not found\n",
3275 dev->name);
3276 return -ENODEV;
3279 dev->reg_state = NETREG_UNREGISTERING;
3281 synchronize_net();
3283 /* Shutdown queueing discipline. */
3284 dev_shutdown(dev);
3287 /* Notify protocols, that we are about to destroy
3288 this device. They should clean all the things.
3290 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3293 * Flush the multicast chain
3295 dev_mc_discard(dev);
3297 if (dev->uninit)
3298 dev->uninit(dev);
3300 /* Notifier chain MUST detach us from master device. */
3301 BUG_TRAP(!dev->master);
3303 free_divert_blk(dev);
3305 /* Finish processing unregister after unlock */
3306 net_set_todo(dev);
3308 synchronize_net();
3310 dev_put(dev);
3311 return 0;
3315 * unregister_netdev - remove device from the kernel
3316 * @dev: device
3318 * This function shuts down a device interface and removes it
3319 * from the kernel tables. On success 0 is returned, on a failure
3320 * a negative errno code is returned.
3322 * This is just a wrapper for unregister_netdevice that takes
3323 * the rtnl semaphore. In general you want to use this and not
3324 * unregister_netdevice.
3326 void unregister_netdev(struct net_device *dev)
3328 rtnl_lock();
3329 unregister_netdevice(dev);
3330 rtnl_unlock();
3333 EXPORT_SYMBOL(unregister_netdev);
3335 #ifdef CONFIG_HOTPLUG_CPU
3336 static int dev_cpu_callback(struct notifier_block *nfb,
3337 unsigned long action,
3338 void *ocpu)
3340 struct sk_buff **list_skb;
3341 struct net_device **list_net;
3342 struct sk_buff *skb;
3343 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3344 struct softnet_data *sd, *oldsd;
3346 if (action != CPU_DEAD)
3347 return NOTIFY_OK;
3349 local_irq_disable();
3350 cpu = smp_processor_id();
3351 sd = &per_cpu(softnet_data, cpu);
3352 oldsd = &per_cpu(softnet_data, oldcpu);
3354 /* Find end of our completion_queue. */
3355 list_skb = &sd->completion_queue;
3356 while (*list_skb)
3357 list_skb = &(*list_skb)->next;
3358 /* Append completion queue from offline CPU. */
3359 *list_skb = oldsd->completion_queue;
3360 oldsd->completion_queue = NULL;
3362 /* Find end of our output_queue. */
3363 list_net = &sd->output_queue;
3364 while (*list_net)
3365 list_net = &(*list_net)->next_sched;
3366 /* Append output queue from offline CPU. */
3367 *list_net = oldsd->output_queue;
3368 oldsd->output_queue = NULL;
3370 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3371 local_irq_enable();
3373 /* Process offline CPU's input_pkt_queue */
3374 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3375 netif_rx(skb);
3377 return NOTIFY_OK;
3379 #endif /* CONFIG_HOTPLUG_CPU */
3381 #ifdef CONFIG_NET_DMA
3383 * net_dma_rebalance -
3384 * This is called when the number of channels allocated to the net_dma_client
3385 * changes. The net_dma_client tries to have one DMA channel per CPU.
3387 static void net_dma_rebalance(void)
3389 unsigned int cpu, i, n;
3390 struct dma_chan *chan;
3392 lock_cpu_hotplug();
3394 if (net_dma_count == 0) {
3395 for_each_online_cpu(cpu)
3396 rcu_assign_pointer(per_cpu(softnet_data.net_dma, cpu), NULL);
3397 unlock_cpu_hotplug();
3398 return;
3401 i = 0;
3402 cpu = first_cpu(cpu_online_map);
3404 rcu_read_lock();
3405 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3406 n = ((num_online_cpus() / net_dma_count)
3407 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3409 while(n) {
3410 per_cpu(softnet_data.net_dma, cpu) = chan;
3411 cpu = next_cpu(cpu, cpu_online_map);
3412 n--;
3414 i++;
3416 rcu_read_unlock();
3418 unlock_cpu_hotplug();
3422 * netdev_dma_event - event callback for the net_dma_client
3423 * @client: should always be net_dma_client
3424 * @chan: DMA channel for the event
3425 * @event: event type
3427 static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3428 enum dma_event event)
3430 spin_lock(&net_dma_event_lock);
3431 switch (event) {
3432 case DMA_RESOURCE_ADDED:
3433 net_dma_count++;
3434 net_dma_rebalance();
3435 break;
3436 case DMA_RESOURCE_REMOVED:
3437 net_dma_count--;
3438 net_dma_rebalance();
3439 break;
3440 default:
3441 break;
3443 spin_unlock(&net_dma_event_lock);
3447 * netdev_dma_regiser - register the networking subsystem as a DMA client
3449 static int __init netdev_dma_register(void)
3451 spin_lock_init(&net_dma_event_lock);
3452 net_dma_client = dma_async_client_register(netdev_dma_event);
3453 if (net_dma_client == NULL)
3454 return -ENOMEM;
3456 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3457 return 0;
3460 #else
3461 static int __init netdev_dma_register(void) { return -ENODEV; }
3462 #endif /* CONFIG_NET_DMA */
3465 * Initialize the DEV module. At boot time this walks the device list and
3466 * unhooks any devices that fail to initialise (normally hardware not
3467 * present) and leaves us with a valid list of present and active devices.
3472 * This is called single threaded during boot, so no need
3473 * to take the rtnl semaphore.
3475 static int __init net_dev_init(void)
3477 int i, rc = -ENOMEM;
3479 BUG_ON(!dev_boot_phase);
3481 net_random_init();
3483 if (dev_proc_init())
3484 goto out;
3486 if (netdev_sysfs_init())
3487 goto out;
3489 INIT_LIST_HEAD(&ptype_all);
3490 for (i = 0; i < 16; i++)
3491 INIT_LIST_HEAD(&ptype_base[i]);
3493 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3494 INIT_HLIST_HEAD(&dev_name_head[i]);
3496 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3497 INIT_HLIST_HEAD(&dev_index_head[i]);
3500 * Initialise the packet receive queues.
3503 for_each_possible_cpu(i) {
3504 struct softnet_data *queue;
3506 queue = &per_cpu(softnet_data, i);
3507 skb_queue_head_init(&queue->input_pkt_queue);
3508 queue->completion_queue = NULL;
3509 INIT_LIST_HEAD(&queue->poll_list);
3510 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3511 queue->backlog_dev.weight = weight_p;
3512 queue->backlog_dev.poll = process_backlog;
3513 atomic_set(&queue->backlog_dev.refcnt, 1);
3516 netdev_dma_register();
3518 dev_boot_phase = 0;
3520 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3521 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3523 hotcpu_notifier(dev_cpu_callback, 0);
3524 dst_init();
3525 dev_mcast_init();
3526 rc = 0;
3527 out:
3528 return rc;
3531 subsys_initcall(net_dev_init);
3533 EXPORT_SYMBOL(__dev_get_by_index);
3534 EXPORT_SYMBOL(__dev_get_by_name);
3535 EXPORT_SYMBOL(__dev_remove_pack);
3536 EXPORT_SYMBOL(dev_valid_name);
3537 EXPORT_SYMBOL(dev_add_pack);
3538 EXPORT_SYMBOL(dev_alloc_name);
3539 EXPORT_SYMBOL(dev_close);
3540 EXPORT_SYMBOL(dev_get_by_flags);
3541 EXPORT_SYMBOL(dev_get_by_index);
3542 EXPORT_SYMBOL(dev_get_by_name);
3543 EXPORT_SYMBOL(dev_open);
3544 EXPORT_SYMBOL(dev_queue_xmit);
3545 EXPORT_SYMBOL(dev_remove_pack);
3546 EXPORT_SYMBOL(dev_set_allmulti);
3547 EXPORT_SYMBOL(dev_set_promiscuity);
3548 EXPORT_SYMBOL(dev_change_flags);
3549 EXPORT_SYMBOL(dev_set_mtu);
3550 EXPORT_SYMBOL(dev_set_mac_address);
3551 EXPORT_SYMBOL(free_netdev);
3552 EXPORT_SYMBOL(netdev_boot_setup_check);
3553 EXPORT_SYMBOL(netdev_set_master);
3554 EXPORT_SYMBOL(netdev_state_change);
3555 EXPORT_SYMBOL(netif_receive_skb);
3556 EXPORT_SYMBOL(netif_rx);
3557 EXPORT_SYMBOL(register_gifconf);
3558 EXPORT_SYMBOL(register_netdevice);
3559 EXPORT_SYMBOL(register_netdevice_notifier);
3560 EXPORT_SYMBOL(skb_checksum_help);
3561 EXPORT_SYMBOL(synchronize_net);
3562 EXPORT_SYMBOL(unregister_netdevice);
3563 EXPORT_SYMBOL(unregister_netdevice_notifier);
3564 EXPORT_SYMBOL(net_enable_timestamp);
3565 EXPORT_SYMBOL(net_disable_timestamp);
3566 EXPORT_SYMBOL(dev_get_flags);
3568 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3569 EXPORT_SYMBOL(br_handle_frame_hook);
3570 EXPORT_SYMBOL(br_fdb_get_hook);
3571 EXPORT_SYMBOL(br_fdb_put_hook);
3572 #endif
3574 #ifdef CONFIG_KMOD
3575 EXPORT_SYMBOL(dev_load);
3576 #endif
3578 EXPORT_PER_CPU_SYMBOL(softnet_data);