2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/config.h>
80 #include <linux/cpu.h>
81 #include <linux/types.h>
82 #include <linux/kernel.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
97 #include <linux/rtnetlink.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <linux/stat.h>
101 #include <linux/if_bridge.h>
102 #include <linux/divert.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/kmod.h>
109 #include <linux/module.h>
110 #include <linux/kallsyms.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <linux/wireless.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
120 * The list of packet types we will receive (as opposed to discard)
121 * and the routines to invoke.
123 * Why 16. Because with 16 the only overlap we get on a hash of the
124 * low nibble of the protocol value is RARP/SNAP/X.25.
126 * NOTE: That is no longer true with the addition of VLAN tags. Not
127 * sure which should go first, but I bet it won't make much
128 * difference if we are running VLANs. The good news is that
129 * this protocol won't be in the list unless compiled in, so
130 * the average user (w/out VLANs) will not be adversly affected.
147 static DEFINE_SPINLOCK(ptype_lock
);
148 static struct list_head ptype_base
[16]; /* 16 way hashed list */
149 static struct list_head ptype_all
; /* Taps */
152 * The @dev_base list is protected by @dev_base_lock and the rtln
155 * Pure readers hold dev_base_lock for reading.
157 * Writers must hold the rtnl semaphore while they loop through the
158 * dev_base list, and hold dev_base_lock for writing when they do the
159 * actual updates. This allows pure readers to access the list even
160 * while a writer is preparing to update it.
162 * To put it another way, dev_base_lock is held for writing only to
163 * protect against pure readers; the rtnl semaphore provides the
164 * protection against other writers.
166 * See, for example usages, register_netdevice() and
167 * unregister_netdevice(), which must be called with the rtnl
170 struct net_device
*dev_base
;
171 static struct net_device
**dev_tail
= &dev_base
;
172 DEFINE_RWLOCK(dev_base_lock
);
174 EXPORT_SYMBOL(dev_base
);
175 EXPORT_SYMBOL(dev_base_lock
);
177 #define NETDEV_HASHBITS 8
178 static struct hlist_head dev_name_head
[1<<NETDEV_HASHBITS
];
179 static struct hlist_head dev_index_head
[1<<NETDEV_HASHBITS
];
181 static inline struct hlist_head
*dev_name_hash(const char *name
)
183 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
184 return &dev_name_head
[hash
& ((1<<NETDEV_HASHBITS
)-1)];
187 static inline struct hlist_head
*dev_index_hash(int ifindex
)
189 return &dev_index_head
[ifindex
& ((1<<NETDEV_HASHBITS
)-1)];
196 static BLOCKING_NOTIFIER_HEAD(netdev_chain
);
199 * Device drivers call our routines to queue packets here. We empty the
200 * queue in the local softnet handler.
202 DEFINE_PER_CPU(struct softnet_data
, softnet_data
) = { NULL
};
205 extern int netdev_sysfs_init(void);
206 extern int netdev_register_sysfs(struct net_device
*);
207 extern void netdev_unregister_sysfs(struct net_device
*);
209 #define netdev_sysfs_init() (0)
210 #define netdev_register_sysfs(dev) (0)
211 #define netdev_unregister_sysfs(dev) do { } while(0)
215 /*******************************************************************************
217 Protocol management and registration routines
219 *******************************************************************************/
228 * Add a protocol ID to the list. Now that the input handler is
229 * smarter we can dispense with all the messy stuff that used to be
232 * BEWARE!!! Protocol handlers, mangling input packets,
233 * MUST BE last in hash buckets and checking protocol handlers
234 * MUST start from promiscuous ptype_all chain in net_bh.
235 * It is true now, do not change it.
236 * Explanation follows: if protocol handler, mangling packet, will
237 * be the first on list, it is not able to sense, that packet
238 * is cloned and should be copied-on-write, so that it will
239 * change it and subsequent readers will get broken packet.
244 * dev_add_pack - add packet handler
245 * @pt: packet type declaration
247 * Add a protocol handler to the networking stack. The passed &packet_type
248 * is linked into kernel lists and may not be freed until it has been
249 * removed from the kernel lists.
251 * This call does not sleep therefore it can not
252 * guarantee all CPU's that are in middle of receiving packets
253 * will see the new packet type (until the next received packet).
256 void dev_add_pack(struct packet_type
*pt
)
260 spin_lock_bh(&ptype_lock
);
261 if (pt
->type
== htons(ETH_P_ALL
)) {
263 list_add_rcu(&pt
->list
, &ptype_all
);
265 hash
= ntohs(pt
->type
) & 15;
266 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
268 spin_unlock_bh(&ptype_lock
);
272 * __dev_remove_pack - remove packet handler
273 * @pt: packet type declaration
275 * Remove a protocol handler that was previously added to the kernel
276 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
277 * from the kernel lists and can be freed or reused once this function
280 * The packet type might still be in use by receivers
281 * and must not be freed until after all the CPU's have gone
282 * through a quiescent state.
284 void __dev_remove_pack(struct packet_type
*pt
)
286 struct list_head
*head
;
287 struct packet_type
*pt1
;
289 spin_lock_bh(&ptype_lock
);
291 if (pt
->type
== htons(ETH_P_ALL
)) {
295 head
= &ptype_base
[ntohs(pt
->type
) & 15];
297 list_for_each_entry(pt1
, head
, list
) {
299 list_del_rcu(&pt
->list
);
304 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
306 spin_unlock_bh(&ptype_lock
);
309 * dev_remove_pack - remove packet handler
310 * @pt: packet type declaration
312 * Remove a protocol handler that was previously added to the kernel
313 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
314 * from the kernel lists and can be freed or reused once this function
317 * This call sleeps to guarantee that no CPU is looking at the packet
320 void dev_remove_pack(struct packet_type
*pt
)
322 __dev_remove_pack(pt
);
327 /******************************************************************************
329 Device Boot-time Settings Routines
331 *******************************************************************************/
333 /* Boot time configuration table */
334 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
337 * netdev_boot_setup_add - add new setup entry
338 * @name: name of the device
339 * @map: configured settings for the device
341 * Adds new setup entry to the dev_boot_setup list. The function
342 * returns 0 on error and 1 on success. This is a generic routine to
345 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
347 struct netdev_boot_setup
*s
;
351 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
352 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
353 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
354 strcpy(s
[i
].name
, name
);
355 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
360 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
364 * netdev_boot_setup_check - check boot time settings
365 * @dev: the netdevice
367 * Check boot time settings for the device.
368 * The found settings are set for the device to be used
369 * later in the device probing.
370 * Returns 0 if no settings found, 1 if they are.
372 int netdev_boot_setup_check(struct net_device
*dev
)
374 struct netdev_boot_setup
*s
= dev_boot_setup
;
377 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
378 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
379 !strncmp(dev
->name
, s
[i
].name
, strlen(s
[i
].name
))) {
380 dev
->irq
= s
[i
].map
.irq
;
381 dev
->base_addr
= s
[i
].map
.base_addr
;
382 dev
->mem_start
= s
[i
].map
.mem_start
;
383 dev
->mem_end
= s
[i
].map
.mem_end
;
392 * netdev_boot_base - get address from boot time settings
393 * @prefix: prefix for network device
394 * @unit: id for network device
396 * Check boot time settings for the base address of device.
397 * The found settings are set for the device to be used
398 * later in the device probing.
399 * Returns 0 if no settings found.
401 unsigned long netdev_boot_base(const char *prefix
, int unit
)
403 const struct netdev_boot_setup
*s
= dev_boot_setup
;
407 sprintf(name
, "%s%d", prefix
, unit
);
410 * If device already registered then return base of 1
411 * to indicate not to probe for this interface
413 if (__dev_get_by_name(name
))
416 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
417 if (!strcmp(name
, s
[i
].name
))
418 return s
[i
].map
.base_addr
;
423 * Saves at boot time configured settings for any netdevice.
425 int __init
netdev_boot_setup(char *str
)
430 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
435 memset(&map
, 0, sizeof(map
));
439 map
.base_addr
= ints
[2];
441 map
.mem_start
= ints
[3];
443 map
.mem_end
= ints
[4];
445 /* Add new entry to the list */
446 return netdev_boot_setup_add(str
, &map
);
449 __setup("netdev=", netdev_boot_setup
);
451 /*******************************************************************************
453 Device Interface Subroutines
455 *******************************************************************************/
458 * __dev_get_by_name - find a device by its name
459 * @name: name to find
461 * Find an interface by name. Must be called under RTNL semaphore
462 * or @dev_base_lock. If the name is found a pointer to the device
463 * is returned. If the name is not found then %NULL is returned. The
464 * reference counters are not incremented so the caller must be
465 * careful with locks.
468 struct net_device
*__dev_get_by_name(const char *name
)
470 struct hlist_node
*p
;
472 hlist_for_each(p
, dev_name_hash(name
)) {
473 struct net_device
*dev
474 = hlist_entry(p
, struct net_device
, name_hlist
);
475 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
482 * dev_get_by_name - find a device by its name
483 * @name: name to find
485 * Find an interface by name. This can be called from any
486 * context and does its own locking. The returned handle has
487 * the usage count incremented and the caller must use dev_put() to
488 * release it when it is no longer needed. %NULL is returned if no
489 * matching device is found.
492 struct net_device
*dev_get_by_name(const char *name
)
494 struct net_device
*dev
;
496 read_lock(&dev_base_lock
);
497 dev
= __dev_get_by_name(name
);
500 read_unlock(&dev_base_lock
);
505 * __dev_get_by_index - find a device by its ifindex
506 * @ifindex: index of device
508 * Search for an interface by index. Returns %NULL if the device
509 * is not found or a pointer to the device. The device has not
510 * had its reference counter increased so the caller must be careful
511 * about locking. The caller must hold either the RTNL semaphore
515 struct net_device
*__dev_get_by_index(int ifindex
)
517 struct hlist_node
*p
;
519 hlist_for_each(p
, dev_index_hash(ifindex
)) {
520 struct net_device
*dev
521 = hlist_entry(p
, struct net_device
, index_hlist
);
522 if (dev
->ifindex
== ifindex
)
530 * dev_get_by_index - find a device by its ifindex
531 * @ifindex: index of device
533 * Search for an interface by index. Returns NULL if the device
534 * is not found or a pointer to the device. The device returned has
535 * had a reference added and the pointer is safe until the user calls
536 * dev_put to indicate they have finished with it.
539 struct net_device
*dev_get_by_index(int ifindex
)
541 struct net_device
*dev
;
543 read_lock(&dev_base_lock
);
544 dev
= __dev_get_by_index(ifindex
);
547 read_unlock(&dev_base_lock
);
552 * dev_getbyhwaddr - find a device by its hardware address
553 * @type: media type of device
554 * @ha: hardware address
556 * Search for an interface by MAC address. Returns NULL if the device
557 * is not found or a pointer to the device. The caller must hold the
558 * rtnl semaphore. The returned device has not had its ref count increased
559 * and the caller must therefore be careful about locking
562 * If the API was consistent this would be __dev_get_by_hwaddr
565 struct net_device
*dev_getbyhwaddr(unsigned short type
, char *ha
)
567 struct net_device
*dev
;
571 for (dev
= dev_base
; dev
; dev
= dev
->next
)
572 if (dev
->type
== type
&&
573 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
578 EXPORT_SYMBOL(dev_getbyhwaddr
);
580 struct net_device
*dev_getfirstbyhwtype(unsigned short type
)
582 struct net_device
*dev
;
585 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
586 if (dev
->type
== type
) {
595 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
598 * dev_get_by_flags - find any device with given flags
599 * @if_flags: IFF_* values
600 * @mask: bitmask of bits in if_flags to check
602 * Search for any interface with the given flags. Returns NULL if a device
603 * is not found or a pointer to the device. The device returned has
604 * had a reference added and the pointer is safe until the user calls
605 * dev_put to indicate they have finished with it.
608 struct net_device
* dev_get_by_flags(unsigned short if_flags
, unsigned short mask
)
610 struct net_device
*dev
;
612 read_lock(&dev_base_lock
);
613 for (dev
= dev_base
; dev
!= NULL
; dev
= dev
->next
) {
614 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
619 read_unlock(&dev_base_lock
);
624 * dev_valid_name - check if name is okay for network device
627 * Network device names need to be valid file names to
628 * to allow sysfs to work
630 int dev_valid_name(const char *name
)
632 return !(*name
== '\0'
633 || !strcmp(name
, ".")
634 || !strcmp(name
, "..")
635 || strchr(name
, '/'));
639 * dev_alloc_name - allocate a name for a device
641 * @name: name format string
643 * Passed a format string - eg "lt%d" it will try and find a suitable
644 * id. Not efficient for many devices, not called a lot. The caller
645 * must hold the dev_base or rtnl lock while allocating the name and
646 * adding the device in order to avoid duplicates. Returns the number
647 * of the unit assigned or a negative errno code.
650 int dev_alloc_name(struct net_device
*dev
, const char *name
)
655 const int max_netdevices
= 8*PAGE_SIZE
;
657 struct net_device
*d
;
659 p
= strnchr(name
, IFNAMSIZ
-1, '%');
662 * Verify the string as this thing may have come from
663 * the user. There must be either one "%d" and no other "%"
666 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
669 /* Use one page as a bit array of possible slots */
670 inuse
= (long *) get_zeroed_page(GFP_ATOMIC
);
674 for (d
= dev_base
; d
; d
= d
->next
) {
675 if (!sscanf(d
->name
, name
, &i
))
677 if (i
< 0 || i
>= max_netdevices
)
680 /* avoid cases where sscanf is not exact inverse of printf */
681 snprintf(buf
, sizeof(buf
), name
, i
);
682 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
686 i
= find_first_zero_bit(inuse
, max_netdevices
);
687 free_page((unsigned long) inuse
);
690 snprintf(buf
, sizeof(buf
), name
, i
);
691 if (!__dev_get_by_name(buf
)) {
692 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
696 /* It is possible to run out of possible slots
697 * when the name is long and there isn't enough space left
698 * for the digits, or if all bits are used.
705 * dev_change_name - change name of a device
707 * @newname: name (or format string) must be at least IFNAMSIZ
709 * Change name of a device, can pass format strings "eth%d".
712 int dev_change_name(struct net_device
*dev
, char *newname
)
718 if (dev
->flags
& IFF_UP
)
721 if (!dev_valid_name(newname
))
724 if (strchr(newname
, '%')) {
725 err
= dev_alloc_name(dev
, newname
);
728 strcpy(newname
, dev
->name
);
730 else if (__dev_get_by_name(newname
))
733 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
735 err
= class_device_rename(&dev
->class_dev
, dev
->name
);
737 hlist_del(&dev
->name_hlist
);
738 hlist_add_head(&dev
->name_hlist
, dev_name_hash(dev
->name
));
739 blocking_notifier_call_chain(&netdev_chain
,
740 NETDEV_CHANGENAME
, dev
);
747 * netdev_features_change - device changes fatures
748 * @dev: device to cause notification
750 * Called to indicate a device has changed features.
752 void netdev_features_change(struct net_device
*dev
)
754 blocking_notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
756 EXPORT_SYMBOL(netdev_features_change
);
759 * netdev_state_change - device changes state
760 * @dev: device to cause notification
762 * Called to indicate a device has changed state. This function calls
763 * the notifier chains for netdev_chain and sends a NEWLINK message
764 * to the routing socket.
766 void netdev_state_change(struct net_device
*dev
)
768 if (dev
->flags
& IFF_UP
) {
769 blocking_notifier_call_chain(&netdev_chain
,
771 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
776 * dev_load - load a network module
777 * @name: name of interface
779 * If a network interface is not present and the process has suitable
780 * privileges this function loads the module. If module loading is not
781 * available in this kernel then it becomes a nop.
784 void dev_load(const char *name
)
786 struct net_device
*dev
;
788 read_lock(&dev_base_lock
);
789 dev
= __dev_get_by_name(name
);
790 read_unlock(&dev_base_lock
);
792 if (!dev
&& capable(CAP_SYS_MODULE
))
793 request_module("%s", name
);
796 static int default_rebuild_header(struct sk_buff
*skb
)
798 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
799 skb
->dev
? skb
->dev
->name
: "NULL!!!");
806 * dev_open - prepare an interface for use.
807 * @dev: device to open
809 * Takes a device from down to up state. The device's private open
810 * function is invoked and then the multicast lists are loaded. Finally
811 * the device is moved into the up state and a %NETDEV_UP message is
812 * sent to the netdev notifier chain.
814 * Calling this function on an active interface is a nop. On a failure
815 * a negative errno code is returned.
817 int dev_open(struct net_device
*dev
)
825 if (dev
->flags
& IFF_UP
)
829 * Is it even present?
831 if (!netif_device_present(dev
))
835 * Call device private open method
837 set_bit(__LINK_STATE_START
, &dev
->state
);
839 ret
= dev
->open(dev
);
841 clear_bit(__LINK_STATE_START
, &dev
->state
);
845 * If it went open OK then:
852 dev
->flags
|= IFF_UP
;
855 * Initialize multicasting status
860 * Wakeup transmit queue engine
865 * ... and announce new interface.
867 blocking_notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
873 * dev_close - shutdown an interface.
874 * @dev: device to shutdown
876 * This function moves an active device into down state. A
877 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
878 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
881 int dev_close(struct net_device
*dev
)
883 if (!(dev
->flags
& IFF_UP
))
887 * Tell people we are going down, so that they can
888 * prepare to death, when device is still operating.
890 blocking_notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
894 clear_bit(__LINK_STATE_START
, &dev
->state
);
896 /* Synchronize to scheduled poll. We cannot touch poll list,
897 * it can be even on different cpu. So just clear netif_running(),
898 * and wait when poll really will happen. Actually, the best place
899 * for this is inside dev->stop() after device stopped its irq
900 * engine, but this requires more changes in devices. */
902 smp_mb__after_clear_bit(); /* Commit netif_running(). */
903 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
909 * Call the device specific close. This cannot fail.
910 * Only if device is UP
912 * We allow it to be called even after a DETACH hot-plug
919 * Device is now down.
922 dev
->flags
&= ~IFF_UP
;
925 * Tell people we are down
927 blocking_notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
934 * Device change register/unregister. These are not inline or static
935 * as we export them to the world.
939 * register_netdevice_notifier - register a network notifier block
942 * Register a notifier to be called when network device events occur.
943 * The notifier passed is linked into the kernel structures and must
944 * not be reused until it has been unregistered. A negative errno code
945 * is returned on a failure.
947 * When registered all registration and up events are replayed
948 * to the new notifier to allow device to have a race free
949 * view of the network device list.
952 int register_netdevice_notifier(struct notifier_block
*nb
)
954 struct net_device
*dev
;
958 err
= blocking_notifier_chain_register(&netdev_chain
, nb
);
960 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
961 nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
963 if (dev
->flags
& IFF_UP
)
964 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
972 * unregister_netdevice_notifier - unregister a network notifier block
975 * Unregister a notifier previously registered by
976 * register_netdevice_notifier(). The notifier is unlinked into the
977 * kernel structures and may then be reused. A negative errno code
978 * is returned on a failure.
981 int unregister_netdevice_notifier(struct notifier_block
*nb
)
986 err
= blocking_notifier_chain_unregister(&netdev_chain
, nb
);
992 * call_netdevice_notifiers - call all network notifier blocks
993 * @val: value passed unmodified to notifier function
994 * @v: pointer passed unmodified to notifier function
996 * Call all network notifier blocks. Parameters and return value
997 * are as for blocking_notifier_call_chain().
1000 int call_netdevice_notifiers(unsigned long val
, void *v
)
1002 return blocking_notifier_call_chain(&netdev_chain
, val
, v
);
1005 /* When > 0 there are consumers of rx skb time stamps */
1006 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1008 void net_enable_timestamp(void)
1010 atomic_inc(&netstamp_needed
);
1013 void net_disable_timestamp(void)
1015 atomic_dec(&netstamp_needed
);
1018 void __net_timestamp(struct sk_buff
*skb
)
1022 do_gettimeofday(&tv
);
1023 skb_set_timestamp(skb
, &tv
);
1025 EXPORT_SYMBOL(__net_timestamp
);
1027 static inline void net_timestamp(struct sk_buff
*skb
)
1029 if (atomic_read(&netstamp_needed
))
1030 __net_timestamp(skb
);
1032 skb
->tstamp
.off_sec
= 0;
1033 skb
->tstamp
.off_usec
= 0;
1038 * Support routine. Sends outgoing frames to any network
1039 * taps currently in use.
1042 void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1044 struct packet_type
*ptype
;
1049 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1050 /* Never send packets back to the socket
1051 * they originated from - MvS (miquels@drinkel.ow.org)
1053 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1054 (ptype
->af_packet_priv
== NULL
||
1055 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1056 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1060 /* skb->nh should be correctly
1061 set by sender, so that the second statement is
1062 just protection against buggy protocols.
1064 skb2
->mac
.raw
= skb2
->data
;
1066 if (skb2
->nh
.raw
< skb2
->data
||
1067 skb2
->nh
.raw
> skb2
->tail
) {
1068 if (net_ratelimit())
1069 printk(KERN_CRIT
"protocol %04x is "
1071 skb2
->protocol
, dev
->name
);
1072 skb2
->nh
.raw
= skb2
->data
;
1075 skb2
->h
.raw
= skb2
->nh
.raw
;
1076 skb2
->pkt_type
= PACKET_OUTGOING
;
1077 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1084 * Invalidate hardware checksum when packet is to be mangled, and
1085 * complete checksum manually on outgoing path.
1087 int skb_checksum_help(struct sk_buff
*skb
, int inward
)
1090 int ret
= 0, offset
= skb
->h
.raw
- skb
->data
;
1093 skb
->ip_summed
= CHECKSUM_NONE
;
1097 if (skb_cloned(skb
)) {
1098 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1103 BUG_ON(offset
> (int)skb
->len
);
1104 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1106 offset
= skb
->tail
- skb
->h
.raw
;
1107 BUG_ON(offset
<= 0);
1108 BUG_ON(skb
->csum
+ 2 > offset
);
1110 *(u16
*)(skb
->h
.raw
+ skb
->csum
) = csum_fold(csum
);
1111 skb
->ip_summed
= CHECKSUM_NONE
;
1116 /* Take action when hardware reception checksum errors are detected. */
1118 void netdev_rx_csum_fault(struct net_device
*dev
)
1120 if (net_ratelimit()) {
1121 printk(KERN_ERR
"%s: hw csum failure.\n",
1122 dev
? dev
->name
: "<unknown>");
1126 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1129 #ifdef CONFIG_HIGHMEM
1130 /* Actually, we should eliminate this check as soon as we know, that:
1131 * 1. IOMMU is present and allows to map all the memory.
1132 * 2. No high memory really exists on this machine.
1135 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1139 if (dev
->features
& NETIF_F_HIGHDMA
)
1142 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1143 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1149 #define illegal_highdma(dev, skb) (0)
1152 /* Keep head the same: replace data */
1153 int __skb_linearize(struct sk_buff
*skb
, gfp_t gfp_mask
)
1158 struct skb_shared_info
*ninfo
;
1159 int headerlen
= skb
->data
- skb
->head
;
1160 int expand
= (skb
->tail
+ skb
->data_len
) - skb
->end
;
1162 if (skb_shared(skb
))
1168 size
= skb
->end
- skb
->head
+ expand
;
1169 size
= SKB_DATA_ALIGN(size
);
1170 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
1174 /* Copy entire thing */
1175 if (skb_copy_bits(skb
, -headerlen
, data
, headerlen
+ skb
->len
))
1179 ninfo
= (struct skb_shared_info
*)(data
+ size
);
1180 atomic_set(&ninfo
->dataref
, 1);
1181 ninfo
->tso_size
= skb_shinfo(skb
)->tso_size
;
1182 ninfo
->tso_segs
= skb_shinfo(skb
)->tso_segs
;
1183 ninfo
->nr_frags
= 0;
1184 ninfo
->frag_list
= NULL
;
1186 /* Offset between the two in bytes */
1187 offset
= data
- skb
->head
;
1189 /* Free old data. */
1190 skb_release_data(skb
);
1193 skb
->end
= data
+ size
;
1195 /* Set up new pointers */
1196 skb
->h
.raw
+= offset
;
1197 skb
->nh
.raw
+= offset
;
1198 skb
->mac
.raw
+= offset
;
1199 skb
->tail
+= offset
;
1200 skb
->data
+= offset
;
1202 /* We are no longer a clone, even if we were. */
1205 skb
->tail
+= skb
->data_len
;
1210 #define HARD_TX_LOCK(dev, cpu) { \
1211 if ((dev->features & NETIF_F_LLTX) == 0) { \
1212 spin_lock(&dev->xmit_lock); \
1213 dev->xmit_lock_owner = cpu; \
1217 #define HARD_TX_UNLOCK(dev) { \
1218 if ((dev->features & NETIF_F_LLTX) == 0) { \
1219 dev->xmit_lock_owner = -1; \
1220 spin_unlock(&dev->xmit_lock); \
1225 * dev_queue_xmit - transmit a buffer
1226 * @skb: buffer to transmit
1228 * Queue a buffer for transmission to a network device. The caller must
1229 * have set the device and priority and built the buffer before calling
1230 * this function. The function can be called from an interrupt.
1232 * A negative errno code is returned on a failure. A success does not
1233 * guarantee the frame will be transmitted as it may be dropped due
1234 * to congestion or traffic shaping.
1236 * -----------------------------------------------------------------------------------
1237 * I notice this method can also return errors from the queue disciplines,
1238 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1241 * Regardless of the return value, the skb is consumed, so it is currently
1242 * difficult to retry a send to this method. (You can bump the ref count
1243 * before sending to hold a reference for retry if you are careful.)
1245 * When calling this method, interrupts MUST be enabled. This is because
1246 * the BH enable code must have IRQs enabled so that it will not deadlock.
1250 int dev_queue_xmit(struct sk_buff
*skb
)
1252 struct net_device
*dev
= skb
->dev
;
1256 if (skb_shinfo(skb
)->frag_list
&&
1257 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1258 __skb_linearize(skb
, GFP_ATOMIC
))
1261 /* Fragmented skb is linearized if device does not support SG,
1262 * or if at least one of fragments is in highmem and device
1263 * does not support DMA from it.
1265 if (skb_shinfo(skb
)->nr_frags
&&
1266 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1267 __skb_linearize(skb
, GFP_ATOMIC
))
1270 /* If packet is not checksummed and device does not support
1271 * checksumming for this protocol, complete checksumming here.
1273 if (skb
->ip_summed
== CHECKSUM_HW
&&
1274 (!(dev
->features
& (NETIF_F_HW_CSUM
| NETIF_F_NO_CSUM
)) &&
1275 (!(dev
->features
& NETIF_F_IP_CSUM
) ||
1276 skb
->protocol
!= htons(ETH_P_IP
))))
1277 if (skb_checksum_help(skb
, 0))
1280 spin_lock_prefetch(&dev
->queue_lock
);
1282 /* Disable soft irqs for various locks below. Also
1283 * stops preemption for RCU.
1287 /* Updates of qdisc are serialized by queue_lock.
1288 * The struct Qdisc which is pointed to by qdisc is now a
1289 * rcu structure - it may be accessed without acquiring
1290 * a lock (but the structure may be stale.) The freeing of the
1291 * qdisc will be deferred until it's known that there are no
1292 * more references to it.
1294 * If the qdisc has an enqueue function, we still need to
1295 * hold the queue_lock before calling it, since queue_lock
1296 * also serializes access to the device queue.
1299 q
= rcu_dereference(dev
->qdisc
);
1300 #ifdef CONFIG_NET_CLS_ACT
1301 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1304 /* Grab device queue */
1305 spin_lock(&dev
->queue_lock
);
1307 rc
= q
->enqueue(skb
, q
);
1311 spin_unlock(&dev
->queue_lock
);
1312 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1316 /* The device has no queue. Common case for software devices:
1317 loopback, all the sorts of tunnels...
1319 Really, it is unlikely that xmit_lock protection is necessary here.
1320 (f.e. loopback and IP tunnels are clean ignoring statistics
1322 However, it is possible, that they rely on protection
1325 Check this and shot the lock. It is not prone from deadlocks.
1326 Either shot noqueue qdisc, it is even simpler 8)
1328 if (dev
->flags
& IFF_UP
) {
1329 int cpu
= smp_processor_id(); /* ok because BHs are off */
1331 if (dev
->xmit_lock_owner
!= cpu
) {
1333 HARD_TX_LOCK(dev
, cpu
);
1335 if (!netif_queue_stopped(dev
)) {
1337 dev_queue_xmit_nit(skb
, dev
);
1340 if (!dev
->hard_start_xmit(skb
, dev
)) {
1341 HARD_TX_UNLOCK(dev
);
1345 HARD_TX_UNLOCK(dev
);
1346 if (net_ratelimit())
1347 printk(KERN_CRIT
"Virtual device %s asks to "
1348 "queue packet!\n", dev
->name
);
1350 /* Recursion is detected! It is possible,
1352 if (net_ratelimit())
1353 printk(KERN_CRIT
"Dead loop on virtual device "
1354 "%s, fix it urgently!\n", dev
->name
);
1370 /*=======================================================================
1372 =======================================================================*/
1374 int netdev_max_backlog
= 1000;
1375 int netdev_budget
= 300;
1376 int weight_p
= 64; /* old backlog weight */
1378 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1382 * netif_rx - post buffer to the network code
1383 * @skb: buffer to post
1385 * This function receives a packet from a device driver and queues it for
1386 * the upper (protocol) levels to process. It always succeeds. The buffer
1387 * may be dropped during processing for congestion control or by the
1391 * NET_RX_SUCCESS (no congestion)
1392 * NET_RX_CN_LOW (low congestion)
1393 * NET_RX_CN_MOD (moderate congestion)
1394 * NET_RX_CN_HIGH (high congestion)
1395 * NET_RX_DROP (packet was dropped)
1399 int netif_rx(struct sk_buff
*skb
)
1401 struct softnet_data
*queue
;
1402 unsigned long flags
;
1404 /* if netpoll wants it, pretend we never saw it */
1405 if (netpoll_rx(skb
))
1408 if (!skb
->tstamp
.off_sec
)
1412 * The code is rearranged so that the path is the most
1413 * short when CPU is congested, but is still operating.
1415 local_irq_save(flags
);
1416 queue
= &__get_cpu_var(softnet_data
);
1418 __get_cpu_var(netdev_rx_stat
).total
++;
1419 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1420 if (queue
->input_pkt_queue
.qlen
) {
1423 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1424 local_irq_restore(flags
);
1425 return NET_RX_SUCCESS
;
1428 netif_rx_schedule(&queue
->backlog_dev
);
1432 __get_cpu_var(netdev_rx_stat
).dropped
++;
1433 local_irq_restore(flags
);
1439 int netif_rx_ni(struct sk_buff
*skb
)
1444 err
= netif_rx(skb
);
1445 if (local_softirq_pending())
1452 EXPORT_SYMBOL(netif_rx_ni
);
1454 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1456 struct net_device
*dev
= skb
->dev
;
1460 * On bonding slaves other than the currently active
1461 * slave, suppress duplicates except for 802.3ad
1462 * ETH_P_SLOW and alb non-mcast/bcast.
1464 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
1465 if (dev
->master
->priv_flags
& IFF_MASTER_ALB
) {
1466 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
1467 skb
->pkt_type
!= PACKET_MULTICAST
)
1471 if (dev
->master
->priv_flags
& IFF_MASTER_8023AD
&&
1472 skb
->protocol
== __constant_htons(ETH_P_SLOW
))
1479 skb
->dev
= dev
->master
;
1485 static void net_tx_action(struct softirq_action
*h
)
1487 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1489 if (sd
->completion_queue
) {
1490 struct sk_buff
*clist
;
1492 local_irq_disable();
1493 clist
= sd
->completion_queue
;
1494 sd
->completion_queue
= NULL
;
1498 struct sk_buff
*skb
= clist
;
1499 clist
= clist
->next
;
1501 BUG_TRAP(!atomic_read(&skb
->users
));
1506 if (sd
->output_queue
) {
1507 struct net_device
*head
;
1509 local_irq_disable();
1510 head
= sd
->output_queue
;
1511 sd
->output_queue
= NULL
;
1515 struct net_device
*dev
= head
;
1516 head
= head
->next_sched
;
1518 smp_mb__before_clear_bit();
1519 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1521 if (spin_trylock(&dev
->queue_lock
)) {
1523 spin_unlock(&dev
->queue_lock
);
1525 netif_schedule(dev
);
1531 static __inline__
int deliver_skb(struct sk_buff
*skb
,
1532 struct packet_type
*pt_prev
,
1533 struct net_device
*orig_dev
)
1535 atomic_inc(&skb
->users
);
1536 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1539 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1540 int (*br_handle_frame_hook
)(struct net_bridge_port
*p
, struct sk_buff
**pskb
);
1542 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1543 unsigned char *addr
);
1544 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
);
1546 static __inline__
int handle_bridge(struct sk_buff
**pskb
,
1547 struct packet_type
**pt_prev
, int *ret
,
1548 struct net_device
*orig_dev
)
1550 struct net_bridge_port
*port
;
1552 if ((*pskb
)->pkt_type
== PACKET_LOOPBACK
||
1553 (port
= rcu_dereference((*pskb
)->dev
->br_port
)) == NULL
)
1557 *ret
= deliver_skb(*pskb
, *pt_prev
, orig_dev
);
1561 return br_handle_frame_hook(port
, pskb
);
1564 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1567 #ifdef CONFIG_NET_CLS_ACT
1568 /* TODO: Maybe we should just force sch_ingress to be compiled in
1569 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1570 * a compare and 2 stores extra right now if we dont have it on
1571 * but have CONFIG_NET_CLS_ACT
1572 * NOTE: This doesnt stop any functionality; if you dont have
1573 * the ingress scheduler, you just cant add policies on ingress.
1576 static int ing_filter(struct sk_buff
*skb
)
1579 struct net_device
*dev
= skb
->dev
;
1580 int result
= TC_ACT_OK
;
1582 if (dev
->qdisc_ingress
) {
1583 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1584 if (MAX_RED_LOOP
< ttl
++) {
1585 printk("Redir loop detected Dropping packet (%s->%s)\n",
1586 skb
->input_dev
->name
, skb
->dev
->name
);
1590 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1592 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1594 spin_lock(&dev
->ingress_lock
);
1595 if ((q
= dev
->qdisc_ingress
) != NULL
)
1596 result
= q
->enqueue(skb
, q
);
1597 spin_unlock(&dev
->ingress_lock
);
1605 int netif_receive_skb(struct sk_buff
*skb
)
1607 struct packet_type
*ptype
, *pt_prev
;
1608 struct net_device
*orig_dev
;
1609 int ret
= NET_RX_DROP
;
1610 unsigned short type
;
1612 /* if we've gotten here through NAPI, check netpoll */
1613 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1616 if (!skb
->tstamp
.off_sec
)
1619 if (!skb
->input_dev
)
1620 skb
->input_dev
= skb
->dev
;
1622 orig_dev
= skb_bond(skb
);
1627 __get_cpu_var(netdev_rx_stat
).total
++;
1629 skb
->h
.raw
= skb
->nh
.raw
= skb
->data
;
1630 skb
->mac_len
= skb
->nh
.raw
- skb
->mac
.raw
;
1636 #ifdef CONFIG_NET_CLS_ACT
1637 if (skb
->tc_verd
& TC_NCLS
) {
1638 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1643 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1644 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1646 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1651 #ifdef CONFIG_NET_CLS_ACT
1653 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1654 pt_prev
= NULL
; /* noone else should process this after*/
1656 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1659 ret
= ing_filter(skb
);
1661 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
1670 handle_diverter(skb
);
1672 if (handle_bridge(&skb
, &pt_prev
, &ret
, orig_dev
))
1675 type
= skb
->protocol
;
1676 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
1677 if (ptype
->type
== type
&&
1678 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
1680 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1686 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1689 /* Jamal, now you will not able to escape explaining
1690 * me how you were going to use this. :-)
1700 static int process_backlog(struct net_device
*backlog_dev
, int *budget
)
1703 int quota
= min(backlog_dev
->quota
, *budget
);
1704 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1705 unsigned long start_time
= jiffies
;
1707 backlog_dev
->weight
= weight_p
;
1709 struct sk_buff
*skb
;
1710 struct net_device
*dev
;
1712 local_irq_disable();
1713 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
1720 netif_receive_skb(skb
);
1726 if (work
>= quota
|| jiffies
- start_time
> 1)
1731 backlog_dev
->quota
-= work
;
1736 backlog_dev
->quota
-= work
;
1739 list_del(&backlog_dev
->poll_list
);
1740 smp_mb__before_clear_bit();
1741 netif_poll_enable(backlog_dev
);
1747 static void net_rx_action(struct softirq_action
*h
)
1749 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1750 unsigned long start_time
= jiffies
;
1751 int budget
= netdev_budget
;
1754 local_irq_disable();
1756 while (!list_empty(&queue
->poll_list
)) {
1757 struct net_device
*dev
;
1759 if (budget
<= 0 || jiffies
- start_time
> 1)
1764 dev
= list_entry(queue
->poll_list
.next
,
1765 struct net_device
, poll_list
);
1766 have
= netpoll_poll_lock(dev
);
1768 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
1769 netpoll_poll_unlock(have
);
1770 local_irq_disable();
1771 list_move_tail(&dev
->poll_list
, &queue
->poll_list
);
1773 dev
->quota
+= dev
->weight
;
1775 dev
->quota
= dev
->weight
;
1777 netpoll_poll_unlock(have
);
1779 local_irq_disable();
1787 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
1788 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1792 static gifconf_func_t
* gifconf_list
[NPROTO
];
1795 * register_gifconf - register a SIOCGIF handler
1796 * @family: Address family
1797 * @gifconf: Function handler
1799 * Register protocol dependent address dumping routines. The handler
1800 * that is passed must not be freed or reused until it has been replaced
1801 * by another handler.
1803 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
1805 if (family
>= NPROTO
)
1807 gifconf_list
[family
] = gifconf
;
1813 * Map an interface index to its name (SIOCGIFNAME)
1817 * We need this ioctl for efficient implementation of the
1818 * if_indextoname() function required by the IPv6 API. Without
1819 * it, we would have to search all the interfaces to find a
1823 static int dev_ifname(struct ifreq __user
*arg
)
1825 struct net_device
*dev
;
1829 * Fetch the caller's info block.
1832 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
1835 read_lock(&dev_base_lock
);
1836 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
1838 read_unlock(&dev_base_lock
);
1842 strcpy(ifr
.ifr_name
, dev
->name
);
1843 read_unlock(&dev_base_lock
);
1845 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
1851 * Perform a SIOCGIFCONF call. This structure will change
1852 * size eventually, and there is nothing I can do about it.
1853 * Thus we will need a 'compatibility mode'.
1856 static int dev_ifconf(char __user
*arg
)
1859 struct net_device
*dev
;
1866 * Fetch the caller's info block.
1869 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
1876 * Loop over the interfaces, and write an info block for each.
1880 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
1881 for (i
= 0; i
< NPROTO
; i
++) {
1882 if (gifconf_list
[i
]) {
1885 done
= gifconf_list
[i
](dev
, NULL
, 0);
1887 done
= gifconf_list
[i
](dev
, pos
+ total
,
1897 * All done. Write the updated control block back to the caller.
1899 ifc
.ifc_len
= total
;
1902 * Both BSD and Solaris return 0 here, so we do too.
1904 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
1907 #ifdef CONFIG_PROC_FS
1909 * This is invoked by the /proc filesystem handler to display a device
1912 static __inline__
struct net_device
*dev_get_idx(loff_t pos
)
1914 struct net_device
*dev
;
1917 for (i
= 0, dev
= dev_base
; dev
&& i
< pos
; ++i
, dev
= dev
->next
);
1919 return i
== pos
? dev
: NULL
;
1922 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1924 read_lock(&dev_base_lock
);
1925 return *pos
? dev_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
1928 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1931 return v
== SEQ_START_TOKEN
? dev_base
: ((struct net_device
*)v
)->next
;
1934 void dev_seq_stop(struct seq_file
*seq
, void *v
)
1936 read_unlock(&dev_base_lock
);
1939 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
1941 if (dev
->get_stats
) {
1942 struct net_device_stats
*stats
= dev
->get_stats(dev
);
1944 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1945 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1946 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
1948 stats
->rx_dropped
+ stats
->rx_missed_errors
,
1949 stats
->rx_fifo_errors
,
1950 stats
->rx_length_errors
+ stats
->rx_over_errors
+
1951 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
1952 stats
->rx_compressed
, stats
->multicast
,
1953 stats
->tx_bytes
, stats
->tx_packets
,
1954 stats
->tx_errors
, stats
->tx_dropped
,
1955 stats
->tx_fifo_errors
, stats
->collisions
,
1956 stats
->tx_carrier_errors
+
1957 stats
->tx_aborted_errors
+
1958 stats
->tx_window_errors
+
1959 stats
->tx_heartbeat_errors
,
1960 stats
->tx_compressed
);
1962 seq_printf(seq
, "%6s: No statistics available.\n", dev
->name
);
1966 * Called from the PROCfs module. This now uses the new arbitrary sized
1967 * /proc/net interface to create /proc/net/dev
1969 static int dev_seq_show(struct seq_file
*seq
, void *v
)
1971 if (v
== SEQ_START_TOKEN
)
1972 seq_puts(seq
, "Inter-| Receive "
1974 " face |bytes packets errs drop fifo frame "
1975 "compressed multicast|bytes packets errs "
1976 "drop fifo colls carrier compressed\n");
1978 dev_seq_printf_stats(seq
, v
);
1982 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
1984 struct netif_rx_stats
*rc
= NULL
;
1986 while (*pos
< NR_CPUS
)
1987 if (cpu_online(*pos
)) {
1988 rc
= &per_cpu(netdev_rx_stat
, *pos
);
1995 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1997 return softnet_get_online(pos
);
2000 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2003 return softnet_get_online(pos
);
2006 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2010 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2012 struct netif_rx_stats
*s
= v
;
2014 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2015 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2016 0, 0, 0, 0, /* was fastroute */
2021 static struct seq_operations dev_seq_ops
= {
2022 .start
= dev_seq_start
,
2023 .next
= dev_seq_next
,
2024 .stop
= dev_seq_stop
,
2025 .show
= dev_seq_show
,
2028 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2030 return seq_open(file
, &dev_seq_ops
);
2033 static struct file_operations dev_seq_fops
= {
2034 .owner
= THIS_MODULE
,
2035 .open
= dev_seq_open
,
2037 .llseek
= seq_lseek
,
2038 .release
= seq_release
,
2041 static struct seq_operations softnet_seq_ops
= {
2042 .start
= softnet_seq_start
,
2043 .next
= softnet_seq_next
,
2044 .stop
= softnet_seq_stop
,
2045 .show
= softnet_seq_show
,
2048 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2050 return seq_open(file
, &softnet_seq_ops
);
2053 static struct file_operations softnet_seq_fops
= {
2054 .owner
= THIS_MODULE
,
2055 .open
= softnet_seq_open
,
2057 .llseek
= seq_lseek
,
2058 .release
= seq_release
,
2061 #ifdef CONFIG_WIRELESS_EXT
2062 extern int wireless_proc_init(void);
2064 #define wireless_proc_init() 0
2067 static int __init
dev_proc_init(void)
2071 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2073 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2075 if (wireless_proc_init())
2081 proc_net_remove("softnet_stat");
2083 proc_net_remove("dev");
2087 #define dev_proc_init() 0
2088 #endif /* CONFIG_PROC_FS */
2092 * netdev_set_master - set up master/slave pair
2093 * @slave: slave device
2094 * @master: new master device
2096 * Changes the master device of the slave. Pass %NULL to break the
2097 * bonding. The caller must hold the RTNL semaphore. On a failure
2098 * a negative errno code is returned. On success the reference counts
2099 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2100 * function returns zero.
2102 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2104 struct net_device
*old
= slave
->master
;
2114 slave
->master
= master
;
2122 slave
->flags
|= IFF_SLAVE
;
2124 slave
->flags
&= ~IFF_SLAVE
;
2126 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2131 * dev_set_promiscuity - update promiscuity count on a device
2135 * Add or remove promsicuity from a device. While the count in the device
2136 * remains above zero the interface remains promiscuous. Once it hits zero
2137 * the device reverts back to normal filtering operation. A negative inc
2138 * value is used to drop promiscuity on the device.
2140 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2142 unsigned short old_flags
= dev
->flags
;
2144 if ((dev
->promiscuity
+= inc
) == 0)
2145 dev
->flags
&= ~IFF_PROMISC
;
2147 dev
->flags
|= IFF_PROMISC
;
2148 if (dev
->flags
!= old_flags
) {
2150 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2151 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2153 audit_log(current
->audit_context
, GFP_ATOMIC
,
2154 AUDIT_ANOM_PROMISCUOUS
,
2155 "dev=%s prom=%d old_prom=%d auid=%u",
2156 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2157 (old_flags
& IFF_PROMISC
),
2158 audit_get_loginuid(current
->audit_context
));
2163 * dev_set_allmulti - update allmulti count on a device
2167 * Add or remove reception of all multicast frames to a device. While the
2168 * count in the device remains above zero the interface remains listening
2169 * to all interfaces. Once it hits zero the device reverts back to normal
2170 * filtering operation. A negative @inc value is used to drop the counter
2171 * when releasing a resource needing all multicasts.
2174 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2176 unsigned short old_flags
= dev
->flags
;
2178 dev
->flags
|= IFF_ALLMULTI
;
2179 if ((dev
->allmulti
+= inc
) == 0)
2180 dev
->flags
&= ~IFF_ALLMULTI
;
2181 if (dev
->flags
^ old_flags
)
2185 unsigned dev_get_flags(const struct net_device
*dev
)
2189 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2194 (dev
->gflags
& (IFF_PROMISC
|
2197 if (netif_running(dev
)) {
2198 if (netif_oper_up(dev
))
2199 flags
|= IFF_RUNNING
;
2200 if (netif_carrier_ok(dev
))
2201 flags
|= IFF_LOWER_UP
;
2202 if (netif_dormant(dev
))
2203 flags
|= IFF_DORMANT
;
2209 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2212 int old_flags
= dev
->flags
;
2215 * Set the flags on our device.
2218 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2219 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2221 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2225 * Load in the correct multicast list now the flags have changed.
2231 * Have we downed the interface. We handle IFF_UP ourselves
2232 * according to user attempts to set it, rather than blindly
2237 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2238 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2244 if (dev
->flags
& IFF_UP
&&
2245 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2247 blocking_notifier_call_chain(&netdev_chain
,
2248 NETDEV_CHANGE
, dev
);
2250 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2251 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2252 dev
->gflags
^= IFF_PROMISC
;
2253 dev_set_promiscuity(dev
, inc
);
2256 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2257 is important. Some (broken) drivers set IFF_PROMISC, when
2258 IFF_ALLMULTI is requested not asking us and not reporting.
2260 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2261 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2262 dev
->gflags
^= IFF_ALLMULTI
;
2263 dev_set_allmulti(dev
, inc
);
2266 if (old_flags
^ dev
->flags
)
2267 rtmsg_ifinfo(RTM_NEWLINK
, dev
, old_flags
^ dev
->flags
);
2272 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2276 if (new_mtu
== dev
->mtu
)
2279 /* MTU must be positive. */
2283 if (!netif_device_present(dev
))
2287 if (dev
->change_mtu
)
2288 err
= dev
->change_mtu(dev
, new_mtu
);
2291 if (!err
&& dev
->flags
& IFF_UP
)
2292 blocking_notifier_call_chain(&netdev_chain
,
2293 NETDEV_CHANGEMTU
, dev
);
2297 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2301 if (!dev
->set_mac_address
)
2303 if (sa
->sa_family
!= dev
->type
)
2305 if (!netif_device_present(dev
))
2307 err
= dev
->set_mac_address(dev
, sa
);
2309 blocking_notifier_call_chain(&netdev_chain
,
2310 NETDEV_CHANGEADDR
, dev
);
2315 * Perform the SIOCxIFxxx calls.
2317 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2320 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2326 case SIOCGIFFLAGS
: /* Get interface flags */
2327 ifr
->ifr_flags
= dev_get_flags(dev
);
2330 case SIOCSIFFLAGS
: /* Set interface flags */
2331 return dev_change_flags(dev
, ifr
->ifr_flags
);
2333 case SIOCGIFMETRIC
: /* Get the metric on the interface
2334 (currently unused) */
2335 ifr
->ifr_metric
= 0;
2338 case SIOCSIFMETRIC
: /* Set the metric on the interface
2339 (currently unused) */
2342 case SIOCGIFMTU
: /* Get the MTU of a device */
2343 ifr
->ifr_mtu
= dev
->mtu
;
2346 case SIOCSIFMTU
: /* Set the MTU of a device */
2347 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
2351 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
2353 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
2354 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2355 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
2359 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
2361 case SIOCSIFHWBROADCAST
:
2362 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
2364 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
2365 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2366 blocking_notifier_call_chain(&netdev_chain
,
2367 NETDEV_CHANGEADDR
, dev
);
2371 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
2372 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
2373 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
2374 ifr
->ifr_map
.irq
= dev
->irq
;
2375 ifr
->ifr_map
.dma
= dev
->dma
;
2376 ifr
->ifr_map
.port
= dev
->if_port
;
2380 if (dev
->set_config
) {
2381 if (!netif_device_present(dev
))
2383 return dev
->set_config(dev
, &ifr
->ifr_map
);
2388 if (!dev
->set_multicast_list
||
2389 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2391 if (!netif_device_present(dev
))
2393 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
2397 if (!dev
->set_multicast_list
||
2398 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2400 if (!netif_device_present(dev
))
2402 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
2406 ifr
->ifr_ifindex
= dev
->ifindex
;
2410 ifr
->ifr_qlen
= dev
->tx_queue_len
;
2414 if (ifr
->ifr_qlen
< 0)
2416 dev
->tx_queue_len
= ifr
->ifr_qlen
;
2420 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
2421 return dev_change_name(dev
, ifr
->ifr_newname
);
2424 * Unknown or private ioctl
2428 if ((cmd
>= SIOCDEVPRIVATE
&&
2429 cmd
<= SIOCDEVPRIVATE
+ 15) ||
2430 cmd
== SIOCBONDENSLAVE
||
2431 cmd
== SIOCBONDRELEASE
||
2432 cmd
== SIOCBONDSETHWADDR
||
2433 cmd
== SIOCBONDSLAVEINFOQUERY
||
2434 cmd
== SIOCBONDINFOQUERY
||
2435 cmd
== SIOCBONDCHANGEACTIVE
||
2436 cmd
== SIOCGMIIPHY
||
2437 cmd
== SIOCGMIIREG
||
2438 cmd
== SIOCSMIIREG
||
2439 cmd
== SIOCBRADDIF
||
2440 cmd
== SIOCBRDELIF
||
2441 cmd
== SIOCWANDEV
) {
2443 if (dev
->do_ioctl
) {
2444 if (netif_device_present(dev
))
2445 err
= dev
->do_ioctl(dev
, ifr
,
2458 * This function handles all "interface"-type I/O control requests. The actual
2459 * 'doing' part of this is dev_ifsioc above.
2463 * dev_ioctl - network device ioctl
2464 * @cmd: command to issue
2465 * @arg: pointer to a struct ifreq in user space
2467 * Issue ioctl functions to devices. This is normally called by the
2468 * user space syscall interfaces but can sometimes be useful for
2469 * other purposes. The return value is the return from the syscall if
2470 * positive or a negative errno code on error.
2473 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
2479 /* One special case: SIOCGIFCONF takes ifconf argument
2480 and requires shared lock, because it sleeps writing
2484 if (cmd
== SIOCGIFCONF
) {
2486 ret
= dev_ifconf((char __user
*) arg
);
2490 if (cmd
== SIOCGIFNAME
)
2491 return dev_ifname((struct ifreq __user
*)arg
);
2493 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2496 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
2498 colon
= strchr(ifr
.ifr_name
, ':');
2503 * See which interface the caller is talking about.
2508 * These ioctl calls:
2509 * - can be done by all.
2510 * - atomic and do not require locking.
2521 dev_load(ifr
.ifr_name
);
2522 read_lock(&dev_base_lock
);
2523 ret
= dev_ifsioc(&ifr
, cmd
);
2524 read_unlock(&dev_base_lock
);
2528 if (copy_to_user(arg
, &ifr
,
2529 sizeof(struct ifreq
)))
2535 dev_load(ifr
.ifr_name
);
2537 ret
= dev_ethtool(&ifr
);
2542 if (copy_to_user(arg
, &ifr
,
2543 sizeof(struct ifreq
)))
2549 * These ioctl calls:
2550 * - require superuser power.
2551 * - require strict serialization.
2557 if (!capable(CAP_NET_ADMIN
))
2559 dev_load(ifr
.ifr_name
);
2561 ret
= dev_ifsioc(&ifr
, cmd
);
2566 if (copy_to_user(arg
, &ifr
,
2567 sizeof(struct ifreq
)))
2573 * These ioctl calls:
2574 * - require superuser power.
2575 * - require strict serialization.
2576 * - do not return a value
2586 case SIOCSIFHWBROADCAST
:
2589 case SIOCBONDENSLAVE
:
2590 case SIOCBONDRELEASE
:
2591 case SIOCBONDSETHWADDR
:
2592 case SIOCBONDCHANGEACTIVE
:
2595 if (!capable(CAP_NET_ADMIN
))
2598 case SIOCBONDSLAVEINFOQUERY
:
2599 case SIOCBONDINFOQUERY
:
2600 dev_load(ifr
.ifr_name
);
2602 ret
= dev_ifsioc(&ifr
, cmd
);
2607 /* Get the per device memory space. We can add this but
2608 * currently do not support it */
2610 /* Set the per device memory buffer space.
2611 * Not applicable in our case */
2616 * Unknown or private ioctl.
2619 if (cmd
== SIOCWANDEV
||
2620 (cmd
>= SIOCDEVPRIVATE
&&
2621 cmd
<= SIOCDEVPRIVATE
+ 15)) {
2622 dev_load(ifr
.ifr_name
);
2624 ret
= dev_ifsioc(&ifr
, cmd
);
2626 if (!ret
&& copy_to_user(arg
, &ifr
,
2627 sizeof(struct ifreq
)))
2631 #ifdef CONFIG_WIRELESS_EXT
2632 /* Take care of Wireless Extensions */
2633 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
) {
2634 /* If command is `set a parameter', or
2635 * `get the encoding parameters', check if
2636 * the user has the right to do it */
2637 if (IW_IS_SET(cmd
) || cmd
== SIOCGIWENCODE
) {
2638 if (!capable(CAP_NET_ADMIN
))
2641 dev_load(ifr
.ifr_name
);
2643 /* Follow me in net/core/wireless.c */
2644 ret
= wireless_process_ioctl(&ifr
, cmd
);
2646 if (IW_IS_GET(cmd
) &&
2647 copy_to_user(arg
, &ifr
,
2648 sizeof(struct ifreq
)))
2652 #endif /* CONFIG_WIRELESS_EXT */
2659 * dev_new_index - allocate an ifindex
2661 * Returns a suitable unique value for a new device interface
2662 * number. The caller must hold the rtnl semaphore or the
2663 * dev_base_lock to be sure it remains unique.
2665 static int dev_new_index(void)
2671 if (!__dev_get_by_index(ifindex
))
2676 static int dev_boot_phase
= 1;
2678 /* Delayed registration/unregisteration */
2679 static DEFINE_SPINLOCK(net_todo_list_lock
);
2680 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
2682 static inline void net_set_todo(struct net_device
*dev
)
2684 spin_lock(&net_todo_list_lock
);
2685 list_add_tail(&dev
->todo_list
, &net_todo_list
);
2686 spin_unlock(&net_todo_list_lock
);
2690 * register_netdevice - register a network device
2691 * @dev: device to register
2693 * Take a completed network device structure and add it to the kernel
2694 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2695 * chain. 0 is returned on success. A negative errno code is returned
2696 * on a failure to set up the device, or if the name is a duplicate.
2698 * Callers must hold the rtnl semaphore. You may want
2699 * register_netdev() instead of this.
2702 * The locking appears insufficient to guarantee two parallel registers
2703 * will not get the same name.
2706 int register_netdevice(struct net_device
*dev
)
2708 struct hlist_head
*head
;
2709 struct hlist_node
*p
;
2712 BUG_ON(dev_boot_phase
);
2715 /* When net_device's are persistent, this will be fatal. */
2716 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
2718 spin_lock_init(&dev
->queue_lock
);
2719 spin_lock_init(&dev
->xmit_lock
);
2720 dev
->xmit_lock_owner
= -1;
2721 #ifdef CONFIG_NET_CLS_ACT
2722 spin_lock_init(&dev
->ingress_lock
);
2725 ret
= alloc_divert_blk(dev
);
2731 /* Init, if this function is available */
2733 ret
= dev
->init(dev
);
2741 if (!dev_valid_name(dev
->name
)) {
2746 dev
->ifindex
= dev_new_index();
2747 if (dev
->iflink
== -1)
2748 dev
->iflink
= dev
->ifindex
;
2750 /* Check for existence of name */
2751 head
= dev_name_hash(dev
->name
);
2752 hlist_for_each(p
, head
) {
2753 struct net_device
*d
2754 = hlist_entry(p
, struct net_device
, name_hlist
);
2755 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
2761 /* Fix illegal SG+CSUM combinations. */
2762 if ((dev
->features
& NETIF_F_SG
) &&
2763 !(dev
->features
& (NETIF_F_IP_CSUM
|
2765 NETIF_F_HW_CSUM
))) {
2766 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2768 dev
->features
&= ~NETIF_F_SG
;
2771 /* TSO requires that SG is present as well. */
2772 if ((dev
->features
& NETIF_F_TSO
) &&
2773 !(dev
->features
& NETIF_F_SG
)) {
2774 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2776 dev
->features
&= ~NETIF_F_TSO
;
2778 if (dev
->features
& NETIF_F_UFO
) {
2779 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
2780 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
2781 "NETIF_F_HW_CSUM feature.\n",
2783 dev
->features
&= ~NETIF_F_UFO
;
2785 if (!(dev
->features
& NETIF_F_SG
)) {
2786 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
2787 "NETIF_F_SG feature.\n",
2789 dev
->features
&= ~NETIF_F_UFO
;
2794 * nil rebuild_header routine,
2795 * that should be never called and used as just bug trap.
2798 if (!dev
->rebuild_header
)
2799 dev
->rebuild_header
= default_rebuild_header
;
2802 * Default initial state at registry is that the
2803 * device is present.
2806 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
2809 dev_init_scheduler(dev
);
2810 write_lock_bh(&dev_base_lock
);
2812 dev_tail
= &dev
->next
;
2813 hlist_add_head(&dev
->name_hlist
, head
);
2814 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
2816 dev
->reg_state
= NETREG_REGISTERING
;
2817 write_unlock_bh(&dev_base_lock
);
2819 /* Notify protocols, that a new device appeared. */
2820 blocking_notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
2822 /* Finish registration after unlock */
2829 free_divert_blk(dev
);
2834 * register_netdev - register a network device
2835 * @dev: device to register
2837 * Take a completed network device structure and add it to the kernel
2838 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2839 * chain. 0 is returned on success. A negative errno code is returned
2840 * on a failure to set up the device, or if the name is a duplicate.
2842 * This is a wrapper around register_netdev that takes the rtnl semaphore
2843 * and expands the device name if you passed a format string to
2846 int register_netdev(struct net_device
*dev
)
2853 * If the name is a format string the caller wants us to do a
2856 if (strchr(dev
->name
, '%')) {
2857 err
= dev_alloc_name(dev
, dev
->name
);
2863 * Back compatibility hook. Kill this one in 2.5
2865 if (dev
->name
[0] == 0 || dev
->name
[0] == ' ') {
2866 err
= dev_alloc_name(dev
, "eth%d");
2871 err
= register_netdevice(dev
);
2876 EXPORT_SYMBOL(register_netdev
);
2879 * netdev_wait_allrefs - wait until all references are gone.
2881 * This is called when unregistering network devices.
2883 * Any protocol or device that holds a reference should register
2884 * for netdevice notification, and cleanup and put back the
2885 * reference if they receive an UNREGISTER event.
2886 * We can get stuck here if buggy protocols don't correctly
2889 static void netdev_wait_allrefs(struct net_device
*dev
)
2891 unsigned long rebroadcast_time
, warning_time
;
2893 rebroadcast_time
= warning_time
= jiffies
;
2894 while (atomic_read(&dev
->refcnt
) != 0) {
2895 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
2898 /* Rebroadcast unregister notification */
2899 blocking_notifier_call_chain(&netdev_chain
,
2900 NETDEV_UNREGISTER
, dev
);
2902 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
2904 /* We must not have linkwatch events
2905 * pending on unregister. If this
2906 * happens, we simply run the queue
2907 * unscheduled, resulting in a noop
2910 linkwatch_run_queue();
2915 rebroadcast_time
= jiffies
;
2920 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
2921 printk(KERN_EMERG
"unregister_netdevice: "
2922 "waiting for %s to become free. Usage "
2924 dev
->name
, atomic_read(&dev
->refcnt
));
2925 warning_time
= jiffies
;
2934 * register_netdevice(x1);
2935 * register_netdevice(x2);
2937 * unregister_netdevice(y1);
2938 * unregister_netdevice(y2);
2944 * We are invoked by rtnl_unlock() after it drops the semaphore.
2945 * This allows us to deal with problems:
2946 * 1) We can create/delete sysfs objects which invoke hotplug
2947 * without deadlocking with linkwatch via keventd.
2948 * 2) Since we run with the RTNL semaphore not held, we can sleep
2949 * safely in order to wait for the netdev refcnt to drop to zero.
2951 static DEFINE_MUTEX(net_todo_run_mutex
);
2952 void netdev_run_todo(void)
2954 struct list_head list
= LIST_HEAD_INIT(list
);
2958 /* Need to guard against multiple cpu's getting out of order. */
2959 mutex_lock(&net_todo_run_mutex
);
2961 /* Not safe to do outside the semaphore. We must not return
2962 * until all unregister events invoked by the local processor
2963 * have been completed (either by this todo run, or one on
2966 if (list_empty(&net_todo_list
))
2969 /* Snapshot list, allow later requests */
2970 spin_lock(&net_todo_list_lock
);
2971 list_splice_init(&net_todo_list
, &list
);
2972 spin_unlock(&net_todo_list_lock
);
2974 while (!list_empty(&list
)) {
2975 struct net_device
*dev
2976 = list_entry(list
.next
, struct net_device
, todo_list
);
2977 list_del(&dev
->todo_list
);
2979 switch(dev
->reg_state
) {
2980 case NETREG_REGISTERING
:
2981 err
= netdev_register_sysfs(dev
);
2983 printk(KERN_ERR
"%s: failed sysfs registration (%d)\n",
2985 dev
->reg_state
= NETREG_REGISTERED
;
2988 case NETREG_UNREGISTERING
:
2989 netdev_unregister_sysfs(dev
);
2990 dev
->reg_state
= NETREG_UNREGISTERED
;
2992 netdev_wait_allrefs(dev
);
2995 BUG_ON(atomic_read(&dev
->refcnt
));
2996 BUG_TRAP(!dev
->ip_ptr
);
2997 BUG_TRAP(!dev
->ip6_ptr
);
2998 BUG_TRAP(!dev
->dn_ptr
);
3001 /* It must be the very last action,
3002 * after this 'dev' may point to freed up memory.
3004 if (dev
->destructor
)
3005 dev
->destructor(dev
);
3009 printk(KERN_ERR
"network todo '%s' but state %d\n",
3010 dev
->name
, dev
->reg_state
);
3016 mutex_unlock(&net_todo_run_mutex
);
3020 * alloc_netdev - allocate network device
3021 * @sizeof_priv: size of private data to allocate space for
3022 * @name: device name format string
3023 * @setup: callback to initialize device
3025 * Allocates a struct net_device with private data area for driver use
3026 * and performs basic initialization.
3028 struct net_device
*alloc_netdev(int sizeof_priv
, const char *name
,
3029 void (*setup
)(struct net_device
*))
3032 struct net_device
*dev
;
3035 /* ensure 32-byte alignment of both the device and private area */
3036 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
3037 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
3039 p
= kmalloc(alloc_size
, GFP_KERNEL
);
3041 printk(KERN_ERR
"alloc_dev: Unable to allocate device.\n");
3044 memset(p
, 0, alloc_size
);
3046 dev
= (struct net_device
*)
3047 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3048 dev
->padded
= (char *)dev
- (char *)p
;
3051 dev
->priv
= netdev_priv(dev
);
3054 strcpy(dev
->name
, name
);
3057 EXPORT_SYMBOL(alloc_netdev
);
3060 * free_netdev - free network device
3063 * This function does the last stage of destroying an allocated device
3064 * interface. The reference to the device object is released.
3065 * If this is the last reference then it will be freed.
3067 void free_netdev(struct net_device
*dev
)
3070 /* Compatiablity with error handling in drivers */
3071 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3072 kfree((char *)dev
- dev
->padded
);
3076 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3077 dev
->reg_state
= NETREG_RELEASED
;
3079 /* will free via class release */
3080 class_device_put(&dev
->class_dev
);
3082 kfree((char *)dev
- dev
->padded
);
3086 /* Synchronize with packet receive processing. */
3087 void synchronize_net(void)
3094 * unregister_netdevice - remove device from the kernel
3097 * This function shuts down a device interface and removes it
3098 * from the kernel tables. On success 0 is returned, on a failure
3099 * a negative errno code is returned.
3101 * Callers must hold the rtnl semaphore. You may want
3102 * unregister_netdev() instead of this.
3105 int unregister_netdevice(struct net_device
*dev
)
3107 struct net_device
*d
, **dp
;
3109 BUG_ON(dev_boot_phase
);
3112 /* Some devices call without registering for initialization unwind. */
3113 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3114 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3115 "was registered\n", dev
->name
, dev
);
3119 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3121 /* If device is running, close it first. */
3122 if (dev
->flags
& IFF_UP
)
3125 /* And unlink it from device chain. */
3126 for (dp
= &dev_base
; (d
= *dp
) != NULL
; dp
= &d
->next
) {
3128 write_lock_bh(&dev_base_lock
);
3129 hlist_del(&dev
->name_hlist
);
3130 hlist_del(&dev
->index_hlist
);
3131 if (dev_tail
== &dev
->next
)
3134 write_unlock_bh(&dev_base_lock
);
3139 printk(KERN_ERR
"unregister net_device: '%s' not found\n",
3144 dev
->reg_state
= NETREG_UNREGISTERING
;
3148 /* Shutdown queueing discipline. */
3152 /* Notify protocols, that we are about to destroy
3153 this device. They should clean all the things.
3155 blocking_notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3158 * Flush the multicast chain
3160 dev_mc_discard(dev
);
3165 /* Notifier chain MUST detach us from master device. */
3166 BUG_TRAP(!dev
->master
);
3168 free_divert_blk(dev
);
3170 /* Finish processing unregister after unlock */
3180 * unregister_netdev - remove device from the kernel
3183 * This function shuts down a device interface and removes it
3184 * from the kernel tables. On success 0 is returned, on a failure
3185 * a negative errno code is returned.
3187 * This is just a wrapper for unregister_netdevice that takes
3188 * the rtnl semaphore. In general you want to use this and not
3189 * unregister_netdevice.
3191 void unregister_netdev(struct net_device
*dev
)
3194 unregister_netdevice(dev
);
3198 EXPORT_SYMBOL(unregister_netdev
);
3200 #ifdef CONFIG_HOTPLUG_CPU
3201 static int dev_cpu_callback(struct notifier_block
*nfb
,
3202 unsigned long action
,
3205 struct sk_buff
**list_skb
;
3206 struct net_device
**list_net
;
3207 struct sk_buff
*skb
;
3208 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3209 struct softnet_data
*sd
, *oldsd
;
3211 if (action
!= CPU_DEAD
)
3214 local_irq_disable();
3215 cpu
= smp_processor_id();
3216 sd
= &per_cpu(softnet_data
, cpu
);
3217 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3219 /* Find end of our completion_queue. */
3220 list_skb
= &sd
->completion_queue
;
3222 list_skb
= &(*list_skb
)->next
;
3223 /* Append completion queue from offline CPU. */
3224 *list_skb
= oldsd
->completion_queue
;
3225 oldsd
->completion_queue
= NULL
;
3227 /* Find end of our output_queue. */
3228 list_net
= &sd
->output_queue
;
3230 list_net
= &(*list_net
)->next_sched
;
3231 /* Append output queue from offline CPU. */
3232 *list_net
= oldsd
->output_queue
;
3233 oldsd
->output_queue
= NULL
;
3235 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3238 /* Process offline CPU's input_pkt_queue */
3239 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3244 #endif /* CONFIG_HOTPLUG_CPU */
3248 * Initialize the DEV module. At boot time this walks the device list and
3249 * unhooks any devices that fail to initialise (normally hardware not
3250 * present) and leaves us with a valid list of present and active devices.
3255 * This is called single threaded during boot, so no need
3256 * to take the rtnl semaphore.
3258 static int __init
net_dev_init(void)
3260 int i
, rc
= -ENOMEM
;
3262 BUG_ON(!dev_boot_phase
);
3266 if (dev_proc_init())
3269 if (netdev_sysfs_init())
3272 INIT_LIST_HEAD(&ptype_all
);
3273 for (i
= 0; i
< 16; i
++)
3274 INIT_LIST_HEAD(&ptype_base
[i
]);
3276 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
3277 INIT_HLIST_HEAD(&dev_name_head
[i
]);
3279 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
3280 INIT_HLIST_HEAD(&dev_index_head
[i
]);
3283 * Initialise the packet receive queues.
3287 struct softnet_data
*queue
;
3289 queue
= &per_cpu(softnet_data
, i
);
3290 skb_queue_head_init(&queue
->input_pkt_queue
);
3291 queue
->completion_queue
= NULL
;
3292 INIT_LIST_HEAD(&queue
->poll_list
);
3293 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
3294 queue
->backlog_dev
.weight
= weight_p
;
3295 queue
->backlog_dev
.poll
= process_backlog
;
3296 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
3301 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
3302 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
3304 hotcpu_notifier(dev_cpu_callback
, 0);
3312 subsys_initcall(net_dev_init
);
3314 EXPORT_SYMBOL(__dev_get_by_index
);
3315 EXPORT_SYMBOL(__dev_get_by_name
);
3316 EXPORT_SYMBOL(__dev_remove_pack
);
3317 EXPORT_SYMBOL(__skb_linearize
);
3318 EXPORT_SYMBOL(dev_valid_name
);
3319 EXPORT_SYMBOL(dev_add_pack
);
3320 EXPORT_SYMBOL(dev_alloc_name
);
3321 EXPORT_SYMBOL(dev_close
);
3322 EXPORT_SYMBOL(dev_get_by_flags
);
3323 EXPORT_SYMBOL(dev_get_by_index
);
3324 EXPORT_SYMBOL(dev_get_by_name
);
3325 EXPORT_SYMBOL(dev_open
);
3326 EXPORT_SYMBOL(dev_queue_xmit
);
3327 EXPORT_SYMBOL(dev_remove_pack
);
3328 EXPORT_SYMBOL(dev_set_allmulti
);
3329 EXPORT_SYMBOL(dev_set_promiscuity
);
3330 EXPORT_SYMBOL(dev_change_flags
);
3331 EXPORT_SYMBOL(dev_set_mtu
);
3332 EXPORT_SYMBOL(dev_set_mac_address
);
3333 EXPORT_SYMBOL(free_netdev
);
3334 EXPORT_SYMBOL(netdev_boot_setup_check
);
3335 EXPORT_SYMBOL(netdev_set_master
);
3336 EXPORT_SYMBOL(netdev_state_change
);
3337 EXPORT_SYMBOL(netif_receive_skb
);
3338 EXPORT_SYMBOL(netif_rx
);
3339 EXPORT_SYMBOL(register_gifconf
);
3340 EXPORT_SYMBOL(register_netdevice
);
3341 EXPORT_SYMBOL(register_netdevice_notifier
);
3342 EXPORT_SYMBOL(skb_checksum_help
);
3343 EXPORT_SYMBOL(synchronize_net
);
3344 EXPORT_SYMBOL(unregister_netdevice
);
3345 EXPORT_SYMBOL(unregister_netdevice_notifier
);
3346 EXPORT_SYMBOL(net_enable_timestamp
);
3347 EXPORT_SYMBOL(net_disable_timestamp
);
3348 EXPORT_SYMBOL(dev_get_flags
);
3350 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3351 EXPORT_SYMBOL(br_handle_frame_hook
);
3352 EXPORT_SYMBOL(br_fdb_get_hook
);
3353 EXPORT_SYMBOL(br_fdb_put_hook
);
3357 EXPORT_SYMBOL(dev_load
);
3360 EXPORT_PER_CPU_SYMBOL(softnet_data
);