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/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/notifier.h>
94 #include <linux/skbuff.h>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
101 #include <linux/divert.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/kmod.h>
108 #include <linux/module.h>
109 #include <linux/kallsyms.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #ifdef CONFIG_NET_RADIO
114 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
115 #include <net/iw_handler.h>
116 #endif /* CONFIG_NET_RADIO */
117 #include <asm/current.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 struct notifier_block
*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 notifier_call_chain(&netdev_chain
, NETDEV_CHANGENAME
, dev
);
746 * netdev_features_change - device changes fatures
747 * @dev: device to cause notification
749 * Called to indicate a device has changed features.
751 void netdev_features_change(struct net_device
*dev
)
753 notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
755 EXPORT_SYMBOL(netdev_features_change
);
758 * netdev_state_change - device changes state
759 * @dev: device to cause notification
761 * Called to indicate a device has changed state. This function calls
762 * the notifier chains for netdev_chain and sends a NEWLINK message
763 * to the routing socket.
765 void netdev_state_change(struct net_device
*dev
)
767 if (dev
->flags
& IFF_UP
) {
768 notifier_call_chain(&netdev_chain
, NETDEV_CHANGE
, dev
);
769 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
774 * dev_load - load a network module
775 * @name: name of interface
777 * If a network interface is not present and the process has suitable
778 * privileges this function loads the module. If module loading is not
779 * available in this kernel then it becomes a nop.
782 void dev_load(const char *name
)
784 struct net_device
*dev
;
786 read_lock(&dev_base_lock
);
787 dev
= __dev_get_by_name(name
);
788 read_unlock(&dev_base_lock
);
790 if (!dev
&& capable(CAP_SYS_MODULE
))
791 request_module("%s", name
);
794 static int default_rebuild_header(struct sk_buff
*skb
)
796 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
797 skb
->dev
? skb
->dev
->name
: "NULL!!!");
804 * dev_open - prepare an interface for use.
805 * @dev: device to open
807 * Takes a device from down to up state. The device's private open
808 * function is invoked and then the multicast lists are loaded. Finally
809 * the device is moved into the up state and a %NETDEV_UP message is
810 * sent to the netdev notifier chain.
812 * Calling this function on an active interface is a nop. On a failure
813 * a negative errno code is returned.
815 int dev_open(struct net_device
*dev
)
823 if (dev
->flags
& IFF_UP
)
827 * Is it even present?
829 if (!netif_device_present(dev
))
833 * Call device private open method
835 set_bit(__LINK_STATE_START
, &dev
->state
);
837 ret
= dev
->open(dev
);
839 clear_bit(__LINK_STATE_START
, &dev
->state
);
843 * If it went open OK then:
850 dev
->flags
|= IFF_UP
;
853 * Initialize multicasting status
858 * Wakeup transmit queue engine
863 * ... and announce new interface.
865 notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
871 * dev_close - shutdown an interface.
872 * @dev: device to shutdown
874 * This function moves an active device into down state. A
875 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
876 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
879 int dev_close(struct net_device
*dev
)
881 if (!(dev
->flags
& IFF_UP
))
885 * Tell people we are going down, so that they can
886 * prepare to death, when device is still operating.
888 notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
892 clear_bit(__LINK_STATE_START
, &dev
->state
);
894 /* Synchronize to scheduled poll. We cannot touch poll list,
895 * it can be even on different cpu. So just clear netif_running(),
896 * and wait when poll really will happen. Actually, the best place
897 * for this is inside dev->stop() after device stopped its irq
898 * engine, but this requires more changes in devices. */
900 smp_mb__after_clear_bit(); /* Commit netif_running(). */
901 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
907 * Call the device specific close. This cannot fail.
908 * Only if device is UP
910 * We allow it to be called even after a DETACH hot-plug
917 * Device is now down.
920 dev
->flags
&= ~IFF_UP
;
923 * Tell people we are down
925 notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
932 * Device change register/unregister. These are not inline or static
933 * as we export them to the world.
937 * register_netdevice_notifier - register a network notifier block
940 * Register a notifier to be called when network device events occur.
941 * The notifier passed is linked into the kernel structures and must
942 * not be reused until it has been unregistered. A negative errno code
943 * is returned on a failure.
945 * When registered all registration and up events are replayed
946 * to the new notifier to allow device to have a race free
947 * view of the network device list.
950 int register_netdevice_notifier(struct notifier_block
*nb
)
952 struct net_device
*dev
;
956 err
= notifier_chain_register(&netdev_chain
, nb
);
958 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
959 nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
961 if (dev
->flags
& IFF_UP
)
962 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
970 * unregister_netdevice_notifier - unregister a network notifier block
973 * Unregister a notifier previously registered by
974 * register_netdevice_notifier(). The notifier is unlinked into the
975 * kernel structures and may then be reused. A negative errno code
976 * is returned on a failure.
979 int unregister_netdevice_notifier(struct notifier_block
*nb
)
981 return notifier_chain_unregister(&netdev_chain
, nb
);
985 * call_netdevice_notifiers - call all network notifier blocks
986 * @val: value passed unmodified to notifier function
987 * @v: pointer passed unmodified to notifier function
989 * Call all network notifier blocks. Parameters and return value
990 * are as for notifier_call_chain().
993 int call_netdevice_notifiers(unsigned long val
, void *v
)
995 return notifier_call_chain(&netdev_chain
, val
, v
);
998 /* When > 0 there are consumers of rx skb time stamps */
999 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1001 void net_enable_timestamp(void)
1003 atomic_inc(&netstamp_needed
);
1006 void net_disable_timestamp(void)
1008 atomic_dec(&netstamp_needed
);
1011 void __net_timestamp(struct sk_buff
*skb
)
1015 do_gettimeofday(&tv
);
1016 skb_set_timestamp(skb
, &tv
);
1018 EXPORT_SYMBOL(__net_timestamp
);
1020 static inline void net_timestamp(struct sk_buff
*skb
)
1022 if (atomic_read(&netstamp_needed
))
1023 __net_timestamp(skb
);
1025 skb
->tstamp
.off_sec
= 0;
1026 skb
->tstamp
.off_usec
= 0;
1031 * Support routine. Sends outgoing frames to any network
1032 * taps currently in use.
1035 void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1037 struct packet_type
*ptype
;
1042 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1043 /* Never send packets back to the socket
1044 * they originated from - MvS (miquels@drinkel.ow.org)
1046 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1047 (ptype
->af_packet_priv
== NULL
||
1048 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1049 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1053 /* skb->nh should be correctly
1054 set by sender, so that the second statement is
1055 just protection against buggy protocols.
1057 skb2
->mac
.raw
= skb2
->data
;
1059 if (skb2
->nh
.raw
< skb2
->data
||
1060 skb2
->nh
.raw
> skb2
->tail
) {
1061 if (net_ratelimit())
1062 printk(KERN_CRIT
"protocol %04x is "
1064 skb2
->protocol
, dev
->name
);
1065 skb2
->nh
.raw
= skb2
->data
;
1068 skb2
->h
.raw
= skb2
->nh
.raw
;
1069 skb2
->pkt_type
= PACKET_OUTGOING
;
1070 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1077 * Invalidate hardware checksum when packet is to be mangled, and
1078 * complete checksum manually on outgoing path.
1080 int skb_checksum_help(struct sk_buff
*skb
, int inward
)
1083 int ret
= 0, offset
= skb
->h
.raw
- skb
->data
;
1086 skb
->ip_summed
= CHECKSUM_NONE
;
1090 if (skb_cloned(skb
)) {
1091 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1096 BUG_ON(offset
> (int)skb
->len
);
1097 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1099 offset
= skb
->tail
- skb
->h
.raw
;
1100 BUG_ON(offset
<= 0);
1101 BUG_ON(skb
->csum
+ 2 > offset
);
1103 *(u16
*)(skb
->h
.raw
+ skb
->csum
) = csum_fold(csum
);
1104 skb
->ip_summed
= CHECKSUM_NONE
;
1109 /* Take action when hardware reception checksum errors are detected. */
1111 void netdev_rx_csum_fault(struct net_device
*dev
)
1113 if (net_ratelimit()) {
1114 printk(KERN_ERR
"%s: hw csum failure.\n",
1115 dev
? dev
->name
: "<unknown>");
1119 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1122 #ifdef CONFIG_HIGHMEM
1123 /* Actually, we should eliminate this check as soon as we know, that:
1124 * 1. IOMMU is present and allows to map all the memory.
1125 * 2. No high memory really exists on this machine.
1128 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1132 if (dev
->features
& NETIF_F_HIGHDMA
)
1135 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1136 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1142 #define illegal_highdma(dev, skb) (0)
1145 /* Keep head the same: replace data */
1146 int __skb_linearize(struct sk_buff
*skb
, gfp_t gfp_mask
)
1151 struct skb_shared_info
*ninfo
;
1152 int headerlen
= skb
->data
- skb
->head
;
1153 int expand
= (skb
->tail
+ skb
->data_len
) - skb
->end
;
1155 if (skb_shared(skb
))
1161 size
= skb
->end
- skb
->head
+ expand
;
1162 size
= SKB_DATA_ALIGN(size
);
1163 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
1167 /* Copy entire thing */
1168 if (skb_copy_bits(skb
, -headerlen
, data
, headerlen
+ skb
->len
))
1172 ninfo
= (struct skb_shared_info
*)(data
+ size
);
1173 atomic_set(&ninfo
->dataref
, 1);
1174 ninfo
->tso_size
= skb_shinfo(skb
)->tso_size
;
1175 ninfo
->tso_segs
= skb_shinfo(skb
)->tso_segs
;
1176 ninfo
->ufo_size
= skb_shinfo(skb
)->ufo_size
;
1177 ninfo
->nr_frags
= 0;
1178 ninfo
->frag_list
= NULL
;
1180 /* Offset between the two in bytes */
1181 offset
= data
- skb
->head
;
1183 /* Free old data. */
1184 skb_release_data(skb
);
1187 skb
->end
= data
+ size
;
1189 /* Set up new pointers */
1190 skb
->h
.raw
+= offset
;
1191 skb
->nh
.raw
+= offset
;
1192 skb
->mac
.raw
+= offset
;
1193 skb
->tail
+= offset
;
1194 skb
->data
+= offset
;
1196 /* We are no longer a clone, even if we were. */
1199 skb
->tail
+= skb
->data_len
;
1204 #define HARD_TX_LOCK(dev, cpu) { \
1205 if ((dev->features & NETIF_F_LLTX) == 0) { \
1206 spin_lock(&dev->xmit_lock); \
1207 dev->xmit_lock_owner = cpu; \
1211 #define HARD_TX_UNLOCK(dev) { \
1212 if ((dev->features & NETIF_F_LLTX) == 0) { \
1213 dev->xmit_lock_owner = -1; \
1214 spin_unlock(&dev->xmit_lock); \
1219 * dev_queue_xmit - transmit a buffer
1220 * @skb: buffer to transmit
1222 * Queue a buffer for transmission to a network device. The caller must
1223 * have set the device and priority and built the buffer before calling
1224 * this function. The function can be called from an interrupt.
1226 * A negative errno code is returned on a failure. A success does not
1227 * guarantee the frame will be transmitted as it may be dropped due
1228 * to congestion or traffic shaping.
1230 * -----------------------------------------------------------------------------------
1231 * I notice this method can also return errors from the queue disciplines,
1232 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1235 * Regardless of the return value, the skb is consumed, so it is currently
1236 * difficult to retry a send to this method. (You can bump the ref count
1237 * before sending to hold a reference for retry if you are careful.)
1239 * When calling this method, interrupts MUST be enabled. This is because
1240 * the BH enable code must have IRQs enabled so that it will not deadlock.
1244 int dev_queue_xmit(struct sk_buff
*skb
)
1246 struct net_device
*dev
= skb
->dev
;
1250 if (skb_shinfo(skb
)->frag_list
&&
1251 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1252 __skb_linearize(skb
, GFP_ATOMIC
))
1255 /* Fragmented skb is linearized if device does not support SG,
1256 * or if at least one of fragments is in highmem and device
1257 * does not support DMA from it.
1259 if (skb_shinfo(skb
)->nr_frags
&&
1260 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1261 __skb_linearize(skb
, GFP_ATOMIC
))
1264 /* If packet is not checksummed and device does not support
1265 * checksumming for this protocol, complete checksumming here.
1267 if (skb
->ip_summed
== CHECKSUM_HW
&&
1268 (!(dev
->features
& (NETIF_F_HW_CSUM
| NETIF_F_NO_CSUM
)) &&
1269 (!(dev
->features
& NETIF_F_IP_CSUM
) ||
1270 skb
->protocol
!= htons(ETH_P_IP
))))
1271 if (skb_checksum_help(skb
, 0))
1274 spin_lock_prefetch(&dev
->queue_lock
);
1276 /* Disable soft irqs for various locks below. Also
1277 * stops preemption for RCU.
1281 /* Updates of qdisc are serialized by queue_lock.
1282 * The struct Qdisc which is pointed to by qdisc is now a
1283 * rcu structure - it may be accessed without acquiring
1284 * a lock (but the structure may be stale.) The freeing of the
1285 * qdisc will be deferred until it's known that there are no
1286 * more references to it.
1288 * If the qdisc has an enqueue function, we still need to
1289 * hold the queue_lock before calling it, since queue_lock
1290 * also serializes access to the device queue.
1293 q
= rcu_dereference(dev
->qdisc
);
1294 #ifdef CONFIG_NET_CLS_ACT
1295 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1298 /* Grab device queue */
1299 spin_lock(&dev
->queue_lock
);
1302 rc
= q
->enqueue(skb
, q
);
1304 spin_unlock(&dev
->queue_lock
);
1306 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1309 spin_unlock(&dev
->queue_lock
);
1312 /* The device has no queue. Common case for software devices:
1313 loopback, all the sorts of tunnels...
1315 Really, it is unlikely that xmit_lock protection is necessary here.
1316 (f.e. loopback and IP tunnels are clean ignoring statistics
1318 However, it is possible, that they rely on protection
1321 Check this and shot the lock. It is not prone from deadlocks.
1322 Either shot noqueue qdisc, it is even simpler 8)
1324 if (dev
->flags
& IFF_UP
) {
1325 int cpu
= smp_processor_id(); /* ok because BHs are off */
1327 if (dev
->xmit_lock_owner
!= cpu
) {
1329 HARD_TX_LOCK(dev
, cpu
);
1331 if (!netif_queue_stopped(dev
)) {
1333 dev_queue_xmit_nit(skb
, dev
);
1336 if (!dev
->hard_start_xmit(skb
, dev
)) {
1337 HARD_TX_UNLOCK(dev
);
1341 HARD_TX_UNLOCK(dev
);
1342 if (net_ratelimit())
1343 printk(KERN_CRIT
"Virtual device %s asks to "
1344 "queue packet!\n", dev
->name
);
1346 /* Recursion is detected! It is possible,
1348 if (net_ratelimit())
1349 printk(KERN_CRIT
"Dead loop on virtual device "
1350 "%s, fix it urgently!\n", dev
->name
);
1366 /*=======================================================================
1368 =======================================================================*/
1370 int netdev_max_backlog
= 1000;
1371 int netdev_budget
= 300;
1372 int weight_p
= 64; /* old backlog weight */
1374 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1378 * netif_rx - post buffer to the network code
1379 * @skb: buffer to post
1381 * This function receives a packet from a device driver and queues it for
1382 * the upper (protocol) levels to process. It always succeeds. The buffer
1383 * may be dropped during processing for congestion control or by the
1387 * NET_RX_SUCCESS (no congestion)
1388 * NET_RX_CN_LOW (low congestion)
1389 * NET_RX_CN_MOD (moderate congestion)
1390 * NET_RX_CN_HIGH (high congestion)
1391 * NET_RX_DROP (packet was dropped)
1395 int netif_rx(struct sk_buff
*skb
)
1397 struct softnet_data
*queue
;
1398 unsigned long flags
;
1400 /* if netpoll wants it, pretend we never saw it */
1401 if (netpoll_rx(skb
))
1404 if (!skb
->tstamp
.off_sec
)
1408 * The code is rearranged so that the path is the most
1409 * short when CPU is congested, but is still operating.
1411 local_irq_save(flags
);
1412 queue
= &__get_cpu_var(softnet_data
);
1414 __get_cpu_var(netdev_rx_stat
).total
++;
1415 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1416 if (queue
->input_pkt_queue
.qlen
) {
1419 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1420 local_irq_restore(flags
);
1421 return NET_RX_SUCCESS
;
1424 netif_rx_schedule(&queue
->backlog_dev
);
1428 __get_cpu_var(netdev_rx_stat
).dropped
++;
1429 local_irq_restore(flags
);
1435 int netif_rx_ni(struct sk_buff
*skb
)
1440 err
= netif_rx(skb
);
1441 if (local_softirq_pending())
1448 EXPORT_SYMBOL(netif_rx_ni
);
1450 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1452 struct net_device
*dev
= skb
->dev
;
1455 skb
->dev
= dev
->master
;
1460 static void net_tx_action(struct softirq_action
*h
)
1462 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1464 if (sd
->completion_queue
) {
1465 struct sk_buff
*clist
;
1467 local_irq_disable();
1468 clist
= sd
->completion_queue
;
1469 sd
->completion_queue
= NULL
;
1473 struct sk_buff
*skb
= clist
;
1474 clist
= clist
->next
;
1476 BUG_TRAP(!atomic_read(&skb
->users
));
1481 if (sd
->output_queue
) {
1482 struct net_device
*head
;
1484 local_irq_disable();
1485 head
= sd
->output_queue
;
1486 sd
->output_queue
= NULL
;
1490 struct net_device
*dev
= head
;
1491 head
= head
->next_sched
;
1493 smp_mb__before_clear_bit();
1494 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1496 if (spin_trylock(&dev
->queue_lock
)) {
1498 spin_unlock(&dev
->queue_lock
);
1500 netif_schedule(dev
);
1506 static __inline__
int deliver_skb(struct sk_buff
*skb
,
1507 struct packet_type
*pt_prev
,
1508 struct net_device
*orig_dev
)
1510 atomic_inc(&skb
->users
);
1511 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1514 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1515 int (*br_handle_frame_hook
)(struct net_bridge_port
*p
, struct sk_buff
**pskb
);
1517 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1518 unsigned char *addr
);
1519 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
);
1521 static __inline__
int handle_bridge(struct sk_buff
**pskb
,
1522 struct packet_type
**pt_prev
, int *ret
,
1523 struct net_device
*orig_dev
)
1525 struct net_bridge_port
*port
;
1527 if ((*pskb
)->pkt_type
== PACKET_LOOPBACK
||
1528 (port
= rcu_dereference((*pskb
)->dev
->br_port
)) == NULL
)
1532 *ret
= deliver_skb(*pskb
, *pt_prev
, orig_dev
);
1536 return br_handle_frame_hook(port
, pskb
);
1539 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1542 #ifdef CONFIG_NET_CLS_ACT
1543 /* TODO: Maybe we should just force sch_ingress to be compiled in
1544 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1545 * a compare and 2 stores extra right now if we dont have it on
1546 * but have CONFIG_NET_CLS_ACT
1547 * NOTE: This doesnt stop any functionality; if you dont have
1548 * the ingress scheduler, you just cant add policies on ingress.
1551 static int ing_filter(struct sk_buff
*skb
)
1554 struct net_device
*dev
= skb
->dev
;
1555 int result
= TC_ACT_OK
;
1557 if (dev
->qdisc_ingress
) {
1558 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1559 if (MAX_RED_LOOP
< ttl
++) {
1560 printk("Redir loop detected Dropping packet (%s->%s)\n",
1561 skb
->input_dev
->name
, skb
->dev
->name
);
1565 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1567 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1569 spin_lock(&dev
->queue_lock
);
1570 if ((q
= dev
->qdisc_ingress
) != NULL
)
1571 result
= q
->enqueue(skb
, q
);
1572 spin_unlock(&dev
->queue_lock
);
1580 int netif_receive_skb(struct sk_buff
*skb
)
1582 struct packet_type
*ptype
, *pt_prev
;
1583 struct net_device
*orig_dev
;
1584 int ret
= NET_RX_DROP
;
1585 unsigned short type
;
1587 /* if we've gotten here through NAPI, check netpoll */
1588 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1591 if (!skb
->tstamp
.off_sec
)
1594 if (!skb
->input_dev
)
1595 skb
->input_dev
= skb
->dev
;
1597 orig_dev
= skb_bond(skb
);
1599 __get_cpu_var(netdev_rx_stat
).total
++;
1601 skb
->h
.raw
= skb
->nh
.raw
= skb
->data
;
1602 skb
->mac_len
= skb
->nh
.raw
- skb
->mac
.raw
;
1608 #ifdef CONFIG_NET_CLS_ACT
1609 if (skb
->tc_verd
& TC_NCLS
) {
1610 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1615 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1616 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1618 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1623 #ifdef CONFIG_NET_CLS_ACT
1625 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1626 pt_prev
= NULL
; /* noone else should process this after*/
1628 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1631 ret
= ing_filter(skb
);
1633 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
1642 handle_diverter(skb
);
1644 if (handle_bridge(&skb
, &pt_prev
, &ret
, orig_dev
))
1647 type
= skb
->protocol
;
1648 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
1649 if (ptype
->type
== type
&&
1650 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
1652 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1658 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1661 /* Jamal, now you will not able to escape explaining
1662 * me how you were going to use this. :-)
1672 static int process_backlog(struct net_device
*backlog_dev
, int *budget
)
1675 int quota
= min(backlog_dev
->quota
, *budget
);
1676 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1677 unsigned long start_time
= jiffies
;
1679 backlog_dev
->weight
= weight_p
;
1681 struct sk_buff
*skb
;
1682 struct net_device
*dev
;
1684 local_irq_disable();
1685 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
1692 netif_receive_skb(skb
);
1698 if (work
>= quota
|| jiffies
- start_time
> 1)
1703 backlog_dev
->quota
-= work
;
1708 backlog_dev
->quota
-= work
;
1711 list_del(&backlog_dev
->poll_list
);
1712 smp_mb__before_clear_bit();
1713 netif_poll_enable(backlog_dev
);
1719 static void net_rx_action(struct softirq_action
*h
)
1721 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1722 unsigned long start_time
= jiffies
;
1723 int budget
= netdev_budget
;
1726 local_irq_disable();
1728 while (!list_empty(&queue
->poll_list
)) {
1729 struct net_device
*dev
;
1731 if (budget
<= 0 || jiffies
- start_time
> 1)
1736 dev
= list_entry(queue
->poll_list
.next
,
1737 struct net_device
, poll_list
);
1738 have
= netpoll_poll_lock(dev
);
1740 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
1741 netpoll_poll_unlock(have
);
1742 local_irq_disable();
1743 list_del(&dev
->poll_list
);
1744 list_add_tail(&dev
->poll_list
, &queue
->poll_list
);
1746 dev
->quota
+= dev
->weight
;
1748 dev
->quota
= dev
->weight
;
1750 netpoll_poll_unlock(have
);
1752 local_irq_disable();
1760 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
1761 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1765 static gifconf_func_t
* gifconf_list
[NPROTO
];
1768 * register_gifconf - register a SIOCGIF handler
1769 * @family: Address family
1770 * @gifconf: Function handler
1772 * Register protocol dependent address dumping routines. The handler
1773 * that is passed must not be freed or reused until it has been replaced
1774 * by another handler.
1776 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
1778 if (family
>= NPROTO
)
1780 gifconf_list
[family
] = gifconf
;
1786 * Map an interface index to its name (SIOCGIFNAME)
1790 * We need this ioctl for efficient implementation of the
1791 * if_indextoname() function required by the IPv6 API. Without
1792 * it, we would have to search all the interfaces to find a
1796 static int dev_ifname(struct ifreq __user
*arg
)
1798 struct net_device
*dev
;
1802 * Fetch the caller's info block.
1805 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
1808 read_lock(&dev_base_lock
);
1809 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
1811 read_unlock(&dev_base_lock
);
1815 strcpy(ifr
.ifr_name
, dev
->name
);
1816 read_unlock(&dev_base_lock
);
1818 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
1824 * Perform a SIOCGIFCONF call. This structure will change
1825 * size eventually, and there is nothing I can do about it.
1826 * Thus we will need a 'compatibility mode'.
1829 static int dev_ifconf(char __user
*arg
)
1832 struct net_device
*dev
;
1839 * Fetch the caller's info block.
1842 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
1849 * Loop over the interfaces, and write an info block for each.
1853 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
1854 for (i
= 0; i
< NPROTO
; i
++) {
1855 if (gifconf_list
[i
]) {
1858 done
= gifconf_list
[i
](dev
, NULL
, 0);
1860 done
= gifconf_list
[i
](dev
, pos
+ total
,
1870 * All done. Write the updated control block back to the caller.
1872 ifc
.ifc_len
= total
;
1875 * Both BSD and Solaris return 0 here, so we do too.
1877 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
1880 #ifdef CONFIG_PROC_FS
1882 * This is invoked by the /proc filesystem handler to display a device
1885 static __inline__
struct net_device
*dev_get_idx(loff_t pos
)
1887 struct net_device
*dev
;
1890 for (i
= 0, dev
= dev_base
; dev
&& i
< pos
; ++i
, dev
= dev
->next
);
1892 return i
== pos
? dev
: NULL
;
1895 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1897 read_lock(&dev_base_lock
);
1898 return *pos
? dev_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
1901 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1904 return v
== SEQ_START_TOKEN
? dev_base
: ((struct net_device
*)v
)->next
;
1907 void dev_seq_stop(struct seq_file
*seq
, void *v
)
1909 read_unlock(&dev_base_lock
);
1912 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
1914 if (dev
->get_stats
) {
1915 struct net_device_stats
*stats
= dev
->get_stats(dev
);
1917 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1918 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1919 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
1921 stats
->rx_dropped
+ stats
->rx_missed_errors
,
1922 stats
->rx_fifo_errors
,
1923 stats
->rx_length_errors
+ stats
->rx_over_errors
+
1924 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
1925 stats
->rx_compressed
, stats
->multicast
,
1926 stats
->tx_bytes
, stats
->tx_packets
,
1927 stats
->tx_errors
, stats
->tx_dropped
,
1928 stats
->tx_fifo_errors
, stats
->collisions
,
1929 stats
->tx_carrier_errors
+
1930 stats
->tx_aborted_errors
+
1931 stats
->tx_window_errors
+
1932 stats
->tx_heartbeat_errors
,
1933 stats
->tx_compressed
);
1935 seq_printf(seq
, "%6s: No statistics available.\n", dev
->name
);
1939 * Called from the PROCfs module. This now uses the new arbitrary sized
1940 * /proc/net interface to create /proc/net/dev
1942 static int dev_seq_show(struct seq_file
*seq
, void *v
)
1944 if (v
== SEQ_START_TOKEN
)
1945 seq_puts(seq
, "Inter-| Receive "
1947 " face |bytes packets errs drop fifo frame "
1948 "compressed multicast|bytes packets errs "
1949 "drop fifo colls carrier compressed\n");
1951 dev_seq_printf_stats(seq
, v
);
1955 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
1957 struct netif_rx_stats
*rc
= NULL
;
1959 while (*pos
< NR_CPUS
)
1960 if (cpu_online(*pos
)) {
1961 rc
= &per_cpu(netdev_rx_stat
, *pos
);
1968 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1970 return softnet_get_online(pos
);
1973 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1976 return softnet_get_online(pos
);
1979 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
1983 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
1985 struct netif_rx_stats
*s
= v
;
1987 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
1988 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
1989 0, 0, 0, 0, /* was fastroute */
1994 static struct seq_operations dev_seq_ops
= {
1995 .start
= dev_seq_start
,
1996 .next
= dev_seq_next
,
1997 .stop
= dev_seq_stop
,
1998 .show
= dev_seq_show
,
2001 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2003 return seq_open(file
, &dev_seq_ops
);
2006 static struct file_operations dev_seq_fops
= {
2007 .owner
= THIS_MODULE
,
2008 .open
= dev_seq_open
,
2010 .llseek
= seq_lseek
,
2011 .release
= seq_release
,
2014 static struct seq_operations softnet_seq_ops
= {
2015 .start
= softnet_seq_start
,
2016 .next
= softnet_seq_next
,
2017 .stop
= softnet_seq_stop
,
2018 .show
= softnet_seq_show
,
2021 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2023 return seq_open(file
, &softnet_seq_ops
);
2026 static struct file_operations softnet_seq_fops
= {
2027 .owner
= THIS_MODULE
,
2028 .open
= softnet_seq_open
,
2030 .llseek
= seq_lseek
,
2031 .release
= seq_release
,
2035 extern int wireless_proc_init(void);
2037 #define wireless_proc_init() 0
2040 static int __init
dev_proc_init(void)
2044 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2046 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2048 if (wireless_proc_init())
2054 proc_net_remove("softnet_stat");
2056 proc_net_remove("dev");
2060 #define dev_proc_init() 0
2061 #endif /* CONFIG_PROC_FS */
2065 * netdev_set_master - set up master/slave pair
2066 * @slave: slave device
2067 * @master: new master device
2069 * Changes the master device of the slave. Pass %NULL to break the
2070 * bonding. The caller must hold the RTNL semaphore. On a failure
2071 * a negative errno code is returned. On success the reference counts
2072 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2073 * function returns zero.
2075 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2077 struct net_device
*old
= slave
->master
;
2087 slave
->master
= master
;
2095 slave
->flags
|= IFF_SLAVE
;
2097 slave
->flags
&= ~IFF_SLAVE
;
2099 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2104 * dev_set_promiscuity - update promiscuity count on a device
2108 * Add or remove promsicuity from a device. While the count in the device
2109 * remains above zero the interface remains promiscuous. Once it hits zero
2110 * the device reverts back to normal filtering operation. A negative inc
2111 * value is used to drop promiscuity on the device.
2113 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2115 unsigned short old_flags
= dev
->flags
;
2117 if ((dev
->promiscuity
+= inc
) == 0)
2118 dev
->flags
&= ~IFF_PROMISC
;
2120 dev
->flags
|= IFF_PROMISC
;
2121 if (dev
->flags
!= old_flags
) {
2123 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2124 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2130 * dev_set_allmulti - update allmulti count on a device
2134 * Add or remove reception of all multicast frames to a device. While the
2135 * count in the device remains above zero the interface remains listening
2136 * to all interfaces. Once it hits zero the device reverts back to normal
2137 * filtering operation. A negative @inc value is used to drop the counter
2138 * when releasing a resource needing all multicasts.
2141 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2143 unsigned short old_flags
= dev
->flags
;
2145 dev
->flags
|= IFF_ALLMULTI
;
2146 if ((dev
->allmulti
+= inc
) == 0)
2147 dev
->flags
&= ~IFF_ALLMULTI
;
2148 if (dev
->flags
^ old_flags
)
2152 unsigned dev_get_flags(const struct net_device
*dev
)
2156 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2159 (dev
->gflags
& (IFF_PROMISC
|
2162 if (netif_running(dev
) && netif_carrier_ok(dev
))
2163 flags
|= IFF_RUNNING
;
2168 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2171 int old_flags
= dev
->flags
;
2174 * Set the flags on our device.
2177 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2178 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2180 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2184 * Load in the correct multicast list now the flags have changed.
2190 * Have we downed the interface. We handle IFF_UP ourselves
2191 * according to user attempts to set it, rather than blindly
2196 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2197 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2203 if (dev
->flags
& IFF_UP
&&
2204 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2206 notifier_call_chain(&netdev_chain
, NETDEV_CHANGE
, dev
);
2208 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2209 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2210 dev
->gflags
^= IFF_PROMISC
;
2211 dev_set_promiscuity(dev
, inc
);
2214 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2215 is important. Some (broken) drivers set IFF_PROMISC, when
2216 IFF_ALLMULTI is requested not asking us and not reporting.
2218 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2219 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2220 dev
->gflags
^= IFF_ALLMULTI
;
2221 dev_set_allmulti(dev
, inc
);
2224 if (old_flags
^ dev
->flags
)
2225 rtmsg_ifinfo(RTM_NEWLINK
, dev
, old_flags
^ dev
->flags
);
2230 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2234 if (new_mtu
== dev
->mtu
)
2237 /* MTU must be positive. */
2241 if (!netif_device_present(dev
))
2245 if (dev
->change_mtu
)
2246 err
= dev
->change_mtu(dev
, new_mtu
);
2249 if (!err
&& dev
->flags
& IFF_UP
)
2250 notifier_call_chain(&netdev_chain
,
2251 NETDEV_CHANGEMTU
, dev
);
2255 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2259 if (!dev
->set_mac_address
)
2261 if (sa
->sa_family
!= dev
->type
)
2263 if (!netif_device_present(dev
))
2265 err
= dev
->set_mac_address(dev
, sa
);
2267 notifier_call_chain(&netdev_chain
, NETDEV_CHANGEADDR
, dev
);
2272 * Perform the SIOCxIFxxx calls.
2274 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2277 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2283 case SIOCGIFFLAGS
: /* Get interface flags */
2284 ifr
->ifr_flags
= dev_get_flags(dev
);
2287 case SIOCSIFFLAGS
: /* Set interface flags */
2288 return dev_change_flags(dev
, ifr
->ifr_flags
);
2290 case SIOCGIFMETRIC
: /* Get the metric on the interface
2291 (currently unused) */
2292 ifr
->ifr_metric
= 0;
2295 case SIOCSIFMETRIC
: /* Set the metric on the interface
2296 (currently unused) */
2299 case SIOCGIFMTU
: /* Get the MTU of a device */
2300 ifr
->ifr_mtu
= dev
->mtu
;
2303 case SIOCSIFMTU
: /* Set the MTU of a device */
2304 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
2308 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
2310 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
2311 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2312 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
2316 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
2318 case SIOCSIFHWBROADCAST
:
2319 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
2321 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
2322 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2323 notifier_call_chain(&netdev_chain
,
2324 NETDEV_CHANGEADDR
, dev
);
2328 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
2329 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
2330 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
2331 ifr
->ifr_map
.irq
= dev
->irq
;
2332 ifr
->ifr_map
.dma
= dev
->dma
;
2333 ifr
->ifr_map
.port
= dev
->if_port
;
2337 if (dev
->set_config
) {
2338 if (!netif_device_present(dev
))
2340 return dev
->set_config(dev
, &ifr
->ifr_map
);
2345 if (!dev
->set_multicast_list
||
2346 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2348 if (!netif_device_present(dev
))
2350 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
2354 if (!dev
->set_multicast_list
||
2355 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2357 if (!netif_device_present(dev
))
2359 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
2363 ifr
->ifr_ifindex
= dev
->ifindex
;
2367 ifr
->ifr_qlen
= dev
->tx_queue_len
;
2371 if (ifr
->ifr_qlen
< 0)
2373 dev
->tx_queue_len
= ifr
->ifr_qlen
;
2377 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
2378 return dev_change_name(dev
, ifr
->ifr_newname
);
2381 * Unknown or private ioctl
2385 if ((cmd
>= SIOCDEVPRIVATE
&&
2386 cmd
<= SIOCDEVPRIVATE
+ 15) ||
2387 cmd
== SIOCBONDENSLAVE
||
2388 cmd
== SIOCBONDRELEASE
||
2389 cmd
== SIOCBONDSETHWADDR
||
2390 cmd
== SIOCBONDSLAVEINFOQUERY
||
2391 cmd
== SIOCBONDINFOQUERY
||
2392 cmd
== SIOCBONDCHANGEACTIVE
||
2393 cmd
== SIOCGMIIPHY
||
2394 cmd
== SIOCGMIIREG
||
2395 cmd
== SIOCSMIIREG
||
2396 cmd
== SIOCBRADDIF
||
2397 cmd
== SIOCBRDELIF
||
2398 cmd
== SIOCWANDEV
) {
2400 if (dev
->do_ioctl
) {
2401 if (netif_device_present(dev
))
2402 err
= dev
->do_ioctl(dev
, ifr
,
2415 * This function handles all "interface"-type I/O control requests. The actual
2416 * 'doing' part of this is dev_ifsioc above.
2420 * dev_ioctl - network device ioctl
2421 * @cmd: command to issue
2422 * @arg: pointer to a struct ifreq in user space
2424 * Issue ioctl functions to devices. This is normally called by the
2425 * user space syscall interfaces but can sometimes be useful for
2426 * other purposes. The return value is the return from the syscall if
2427 * positive or a negative errno code on error.
2430 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
2436 /* One special case: SIOCGIFCONF takes ifconf argument
2437 and requires shared lock, because it sleeps writing
2441 if (cmd
== SIOCGIFCONF
) {
2443 ret
= dev_ifconf((char __user
*) arg
);
2447 if (cmd
== SIOCGIFNAME
)
2448 return dev_ifname((struct ifreq __user
*)arg
);
2450 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2453 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
2455 colon
= strchr(ifr
.ifr_name
, ':');
2460 * See which interface the caller is talking about.
2465 * These ioctl calls:
2466 * - can be done by all.
2467 * - atomic and do not require locking.
2478 dev_load(ifr
.ifr_name
);
2479 read_lock(&dev_base_lock
);
2480 ret
= dev_ifsioc(&ifr
, cmd
);
2481 read_unlock(&dev_base_lock
);
2485 if (copy_to_user(arg
, &ifr
,
2486 sizeof(struct ifreq
)))
2492 dev_load(ifr
.ifr_name
);
2494 ret
= dev_ethtool(&ifr
);
2499 if (copy_to_user(arg
, &ifr
,
2500 sizeof(struct ifreq
)))
2506 * These ioctl calls:
2507 * - require superuser power.
2508 * - require strict serialization.
2514 if (!capable(CAP_NET_ADMIN
))
2516 dev_load(ifr
.ifr_name
);
2518 ret
= dev_ifsioc(&ifr
, cmd
);
2523 if (copy_to_user(arg
, &ifr
,
2524 sizeof(struct ifreq
)))
2530 * These ioctl calls:
2531 * - require superuser power.
2532 * - require strict serialization.
2533 * - do not return a value
2543 case SIOCSIFHWBROADCAST
:
2546 case SIOCBONDENSLAVE
:
2547 case SIOCBONDRELEASE
:
2548 case SIOCBONDSETHWADDR
:
2549 case SIOCBONDCHANGEACTIVE
:
2552 if (!capable(CAP_NET_ADMIN
))
2555 case SIOCBONDSLAVEINFOQUERY
:
2556 case SIOCBONDINFOQUERY
:
2557 dev_load(ifr
.ifr_name
);
2559 ret
= dev_ifsioc(&ifr
, cmd
);
2564 /* Get the per device memory space. We can add this but
2565 * currently do not support it */
2567 /* Set the per device memory buffer space.
2568 * Not applicable in our case */
2573 * Unknown or private ioctl.
2576 if (cmd
== SIOCWANDEV
||
2577 (cmd
>= SIOCDEVPRIVATE
&&
2578 cmd
<= SIOCDEVPRIVATE
+ 15)) {
2579 dev_load(ifr
.ifr_name
);
2581 ret
= dev_ifsioc(&ifr
, cmd
);
2583 if (!ret
&& copy_to_user(arg
, &ifr
,
2584 sizeof(struct ifreq
)))
2589 /* Take care of Wireless Extensions */
2590 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
) {
2591 /* If command is `set a parameter', or
2592 * `get the encoding parameters', check if
2593 * the user has the right to do it */
2594 if (IW_IS_SET(cmd
) || cmd
== SIOCGIWENCODE
) {
2595 if (!capable(CAP_NET_ADMIN
))
2598 dev_load(ifr
.ifr_name
);
2600 /* Follow me in net/core/wireless.c */
2601 ret
= wireless_process_ioctl(&ifr
, cmd
);
2603 if (IW_IS_GET(cmd
) &&
2604 copy_to_user(arg
, &ifr
,
2605 sizeof(struct ifreq
)))
2609 #endif /* WIRELESS_EXT */
2616 * dev_new_index - allocate an ifindex
2618 * Returns a suitable unique value for a new device interface
2619 * number. The caller must hold the rtnl semaphore or the
2620 * dev_base_lock to be sure it remains unique.
2622 static int dev_new_index(void)
2628 if (!__dev_get_by_index(ifindex
))
2633 static int dev_boot_phase
= 1;
2635 /* Delayed registration/unregisteration */
2636 static DEFINE_SPINLOCK(net_todo_list_lock
);
2637 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
2639 static inline void net_set_todo(struct net_device
*dev
)
2641 spin_lock(&net_todo_list_lock
);
2642 list_add_tail(&dev
->todo_list
, &net_todo_list
);
2643 spin_unlock(&net_todo_list_lock
);
2647 * register_netdevice - register a network device
2648 * @dev: device to register
2650 * Take a completed network device structure and add it to the kernel
2651 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2652 * chain. 0 is returned on success. A negative errno code is returned
2653 * on a failure to set up the device, or if the name is a duplicate.
2655 * Callers must hold the rtnl semaphore. You may want
2656 * register_netdev() instead of this.
2659 * The locking appears insufficient to guarantee two parallel registers
2660 * will not get the same name.
2663 int register_netdevice(struct net_device
*dev
)
2665 struct hlist_head
*head
;
2666 struct hlist_node
*p
;
2669 BUG_ON(dev_boot_phase
);
2672 /* When net_device's are persistent, this will be fatal. */
2673 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
2675 spin_lock_init(&dev
->queue_lock
);
2676 spin_lock_init(&dev
->xmit_lock
);
2677 dev
->xmit_lock_owner
= -1;
2678 #ifdef CONFIG_NET_CLS_ACT
2679 spin_lock_init(&dev
->ingress_lock
);
2682 ret
= alloc_divert_blk(dev
);
2688 /* Init, if this function is available */
2690 ret
= dev
->init(dev
);
2698 if (!dev_valid_name(dev
->name
)) {
2703 dev
->ifindex
= dev_new_index();
2704 if (dev
->iflink
== -1)
2705 dev
->iflink
= dev
->ifindex
;
2707 /* Check for existence of name */
2708 head
= dev_name_hash(dev
->name
);
2709 hlist_for_each(p
, head
) {
2710 struct net_device
*d
2711 = hlist_entry(p
, struct net_device
, name_hlist
);
2712 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
2718 /* Fix illegal SG+CSUM combinations. */
2719 if ((dev
->features
& NETIF_F_SG
) &&
2720 !(dev
->features
& (NETIF_F_IP_CSUM
|
2722 NETIF_F_HW_CSUM
))) {
2723 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2725 dev
->features
&= ~NETIF_F_SG
;
2728 /* TSO requires that SG is present as well. */
2729 if ((dev
->features
& NETIF_F_TSO
) &&
2730 !(dev
->features
& NETIF_F_SG
)) {
2731 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2733 dev
->features
&= ~NETIF_F_TSO
;
2735 if (dev
->features
& NETIF_F_UFO
) {
2736 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
2737 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
2738 "NETIF_F_HW_CSUM feature.\n",
2740 dev
->features
&= ~NETIF_F_UFO
;
2742 if (!(dev
->features
& NETIF_F_SG
)) {
2743 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
2744 "NETIF_F_SG feature.\n",
2746 dev
->features
&= ~NETIF_F_UFO
;
2751 * nil rebuild_header routine,
2752 * that should be never called and used as just bug trap.
2755 if (!dev
->rebuild_header
)
2756 dev
->rebuild_header
= default_rebuild_header
;
2759 * Default initial state at registry is that the
2760 * device is present.
2763 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
2766 dev_init_scheduler(dev
);
2767 write_lock_bh(&dev_base_lock
);
2769 dev_tail
= &dev
->next
;
2770 hlist_add_head(&dev
->name_hlist
, head
);
2771 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
2773 dev
->reg_state
= NETREG_REGISTERING
;
2774 write_unlock_bh(&dev_base_lock
);
2776 /* Notify protocols, that a new device appeared. */
2777 notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
2779 /* Finish registration after unlock */
2786 free_divert_blk(dev
);
2791 * register_netdev - register a network device
2792 * @dev: device to register
2794 * Take a completed network device structure and add it to the kernel
2795 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2796 * chain. 0 is returned on success. A negative errno code is returned
2797 * on a failure to set up the device, or if the name is a duplicate.
2799 * This is a wrapper around register_netdev that takes the rtnl semaphore
2800 * and expands the device name if you passed a format string to
2803 int register_netdev(struct net_device
*dev
)
2810 * If the name is a format string the caller wants us to do a
2813 if (strchr(dev
->name
, '%')) {
2814 err
= dev_alloc_name(dev
, dev
->name
);
2820 * Back compatibility hook. Kill this one in 2.5
2822 if (dev
->name
[0] == 0 || dev
->name
[0] == ' ') {
2823 err
= dev_alloc_name(dev
, "eth%d");
2828 err
= register_netdevice(dev
);
2833 EXPORT_SYMBOL(register_netdev
);
2836 * netdev_wait_allrefs - wait until all references are gone.
2838 * This is called when unregistering network devices.
2840 * Any protocol or device that holds a reference should register
2841 * for netdevice notification, and cleanup and put back the
2842 * reference if they receive an UNREGISTER event.
2843 * We can get stuck here if buggy protocols don't correctly
2846 static void netdev_wait_allrefs(struct net_device
*dev
)
2848 unsigned long rebroadcast_time
, warning_time
;
2850 rebroadcast_time
= warning_time
= jiffies
;
2851 while (atomic_read(&dev
->refcnt
) != 0) {
2852 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
2855 /* Rebroadcast unregister notification */
2856 notifier_call_chain(&netdev_chain
,
2857 NETDEV_UNREGISTER
, dev
);
2859 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
2861 /* We must not have linkwatch events
2862 * pending on unregister. If this
2863 * happens, we simply run the queue
2864 * unscheduled, resulting in a noop
2867 linkwatch_run_queue();
2872 rebroadcast_time
= jiffies
;
2877 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
2878 printk(KERN_EMERG
"unregister_netdevice: "
2879 "waiting for %s to become free. Usage "
2881 dev
->name
, atomic_read(&dev
->refcnt
));
2882 warning_time
= jiffies
;
2891 * register_netdevice(x1);
2892 * register_netdevice(x2);
2894 * unregister_netdevice(y1);
2895 * unregister_netdevice(y2);
2901 * We are invoked by rtnl_unlock() after it drops the semaphore.
2902 * This allows us to deal with problems:
2903 * 1) We can create/delete sysfs objects which invoke hotplug
2904 * without deadlocking with linkwatch via keventd.
2905 * 2) Since we run with the RTNL semaphore not held, we can sleep
2906 * safely in order to wait for the netdev refcnt to drop to zero.
2908 static DECLARE_MUTEX(net_todo_run_mutex
);
2909 void netdev_run_todo(void)
2911 struct list_head list
= LIST_HEAD_INIT(list
);
2915 /* Need to guard against multiple cpu's getting out of order. */
2916 down(&net_todo_run_mutex
);
2918 /* Not safe to do outside the semaphore. We must not return
2919 * until all unregister events invoked by the local processor
2920 * have been completed (either by this todo run, or one on
2923 if (list_empty(&net_todo_list
))
2926 /* Snapshot list, allow later requests */
2927 spin_lock(&net_todo_list_lock
);
2928 list_splice_init(&net_todo_list
, &list
);
2929 spin_unlock(&net_todo_list_lock
);
2931 while (!list_empty(&list
)) {
2932 struct net_device
*dev
2933 = list_entry(list
.next
, struct net_device
, todo_list
);
2934 list_del(&dev
->todo_list
);
2936 switch(dev
->reg_state
) {
2937 case NETREG_REGISTERING
:
2938 dev
->reg_state
= NETREG_REGISTERED
;
2939 err
= netdev_register_sysfs(dev
);
2941 printk(KERN_ERR
"%s: failed sysfs registration (%d)\n",
2945 case NETREG_UNREGISTERING
:
2946 netdev_unregister_sysfs(dev
);
2947 dev
->reg_state
= NETREG_UNREGISTERED
;
2949 netdev_wait_allrefs(dev
);
2952 BUG_ON(atomic_read(&dev
->refcnt
));
2953 BUG_TRAP(!dev
->ip_ptr
);
2954 BUG_TRAP(!dev
->ip6_ptr
);
2955 BUG_TRAP(!dev
->dn_ptr
);
2958 /* It must be the very last action,
2959 * after this 'dev' may point to freed up memory.
2961 if (dev
->destructor
)
2962 dev
->destructor(dev
);
2966 printk(KERN_ERR
"network todo '%s' but state %d\n",
2967 dev
->name
, dev
->reg_state
);
2973 up(&net_todo_run_mutex
);
2977 * alloc_netdev - allocate network device
2978 * @sizeof_priv: size of private data to allocate space for
2979 * @name: device name format string
2980 * @setup: callback to initialize device
2982 * Allocates a struct net_device with private data area for driver use
2983 * and performs basic initialization.
2985 struct net_device
*alloc_netdev(int sizeof_priv
, const char *name
,
2986 void (*setup
)(struct net_device
*))
2989 struct net_device
*dev
;
2992 /* ensure 32-byte alignment of both the device and private area */
2993 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
2994 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
2996 p
= kmalloc(alloc_size
, GFP_KERNEL
);
2998 printk(KERN_ERR
"alloc_dev: Unable to allocate device.\n");
3001 memset(p
, 0, alloc_size
);
3003 dev
= (struct net_device
*)
3004 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3005 dev
->padded
= (char *)dev
- (char *)p
;
3008 dev
->priv
= netdev_priv(dev
);
3011 strcpy(dev
->name
, name
);
3014 EXPORT_SYMBOL(alloc_netdev
);
3017 * free_netdev - free network device
3020 * This function does the last stage of destroying an allocated device
3021 * interface. The reference to the device object is released.
3022 * If this is the last reference then it will be freed.
3024 void free_netdev(struct net_device
*dev
)
3027 /* Compatiablity with error handling in drivers */
3028 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3029 kfree((char *)dev
- dev
->padded
);
3033 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3034 dev
->reg_state
= NETREG_RELEASED
;
3036 /* will free via class release */
3037 class_device_put(&dev
->class_dev
);
3039 kfree((char *)dev
- dev
->padded
);
3043 /* Synchronize with packet receive processing. */
3044 void synchronize_net(void)
3051 * unregister_netdevice - remove device from the kernel
3054 * This function shuts down a device interface and removes it
3055 * from the kernel tables. On success 0 is returned, on a failure
3056 * a negative errno code is returned.
3058 * Callers must hold the rtnl semaphore. You may want
3059 * unregister_netdev() instead of this.
3062 int unregister_netdevice(struct net_device
*dev
)
3064 struct net_device
*d
, **dp
;
3066 BUG_ON(dev_boot_phase
);
3069 /* Some devices call without registering for initialization unwind. */
3070 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3071 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3072 "was registered\n", dev
->name
, dev
);
3076 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3078 /* If device is running, close it first. */
3079 if (dev
->flags
& IFF_UP
)
3082 /* And unlink it from device chain. */
3083 for (dp
= &dev_base
; (d
= *dp
) != NULL
; dp
= &d
->next
) {
3085 write_lock_bh(&dev_base_lock
);
3086 hlist_del(&dev
->name_hlist
);
3087 hlist_del(&dev
->index_hlist
);
3088 if (dev_tail
== &dev
->next
)
3091 write_unlock_bh(&dev_base_lock
);
3096 printk(KERN_ERR
"unregister net_device: '%s' not found\n",
3101 dev
->reg_state
= NETREG_UNREGISTERING
;
3105 /* Shutdown queueing discipline. */
3109 /* Notify protocols, that we are about to destroy
3110 this device. They should clean all the things.
3112 notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3115 * Flush the multicast chain
3117 dev_mc_discard(dev
);
3122 /* Notifier chain MUST detach us from master device. */
3123 BUG_TRAP(!dev
->master
);
3125 free_divert_blk(dev
);
3127 /* Finish processing unregister after unlock */
3137 * unregister_netdev - remove device from the kernel
3140 * This function shuts down a device interface and removes it
3141 * from the kernel tables. On success 0 is returned, on a failure
3142 * a negative errno code is returned.
3144 * This is just a wrapper for unregister_netdevice that takes
3145 * the rtnl semaphore. In general you want to use this and not
3146 * unregister_netdevice.
3148 void unregister_netdev(struct net_device
*dev
)
3151 unregister_netdevice(dev
);
3155 EXPORT_SYMBOL(unregister_netdev
);
3157 #ifdef CONFIG_HOTPLUG_CPU
3158 static int dev_cpu_callback(struct notifier_block
*nfb
,
3159 unsigned long action
,
3162 struct sk_buff
**list_skb
;
3163 struct net_device
**list_net
;
3164 struct sk_buff
*skb
;
3165 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3166 struct softnet_data
*sd
, *oldsd
;
3168 if (action
!= CPU_DEAD
)
3171 local_irq_disable();
3172 cpu
= smp_processor_id();
3173 sd
= &per_cpu(softnet_data
, cpu
);
3174 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3176 /* Find end of our completion_queue. */
3177 list_skb
= &sd
->completion_queue
;
3179 list_skb
= &(*list_skb
)->next
;
3180 /* Append completion queue from offline CPU. */
3181 *list_skb
= oldsd
->completion_queue
;
3182 oldsd
->completion_queue
= NULL
;
3184 /* Find end of our output_queue. */
3185 list_net
= &sd
->output_queue
;
3187 list_net
= &(*list_net
)->next_sched
;
3188 /* Append output queue from offline CPU. */
3189 *list_net
= oldsd
->output_queue
;
3190 oldsd
->output_queue
= NULL
;
3192 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3195 /* Process offline CPU's input_pkt_queue */
3196 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3201 #endif /* CONFIG_HOTPLUG_CPU */
3205 * Initialize the DEV module. At boot time this walks the device list and
3206 * unhooks any devices that fail to initialise (normally hardware not
3207 * present) and leaves us with a valid list of present and active devices.
3212 * This is called single threaded during boot, so no need
3213 * to take the rtnl semaphore.
3215 static int __init
net_dev_init(void)
3217 int i
, rc
= -ENOMEM
;
3219 BUG_ON(!dev_boot_phase
);
3223 if (dev_proc_init())
3226 if (netdev_sysfs_init())
3229 INIT_LIST_HEAD(&ptype_all
);
3230 for (i
= 0; i
< 16; i
++)
3231 INIT_LIST_HEAD(&ptype_base
[i
]);
3233 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
3234 INIT_HLIST_HEAD(&dev_name_head
[i
]);
3236 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
3237 INIT_HLIST_HEAD(&dev_index_head
[i
]);
3240 * Initialise the packet receive queues.
3244 struct softnet_data
*queue
;
3246 queue
= &per_cpu(softnet_data
, i
);
3247 skb_queue_head_init(&queue
->input_pkt_queue
);
3248 queue
->completion_queue
= NULL
;
3249 INIT_LIST_HEAD(&queue
->poll_list
);
3250 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
3251 queue
->backlog_dev
.weight
= weight_p
;
3252 queue
->backlog_dev
.poll
= process_backlog
;
3253 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
3258 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
3259 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
3261 hotcpu_notifier(dev_cpu_callback
, 0);
3269 subsys_initcall(net_dev_init
);
3271 EXPORT_SYMBOL(__dev_get_by_index
);
3272 EXPORT_SYMBOL(__dev_get_by_name
);
3273 EXPORT_SYMBOL(__dev_remove_pack
);
3274 EXPORT_SYMBOL(__skb_linearize
);
3275 EXPORT_SYMBOL(dev_valid_name
);
3276 EXPORT_SYMBOL(dev_add_pack
);
3277 EXPORT_SYMBOL(dev_alloc_name
);
3278 EXPORT_SYMBOL(dev_close
);
3279 EXPORT_SYMBOL(dev_get_by_flags
);
3280 EXPORT_SYMBOL(dev_get_by_index
);
3281 EXPORT_SYMBOL(dev_get_by_name
);
3282 EXPORT_SYMBOL(dev_open
);
3283 EXPORT_SYMBOL(dev_queue_xmit
);
3284 EXPORT_SYMBOL(dev_remove_pack
);
3285 EXPORT_SYMBOL(dev_set_allmulti
);
3286 EXPORT_SYMBOL(dev_set_promiscuity
);
3287 EXPORT_SYMBOL(dev_change_flags
);
3288 EXPORT_SYMBOL(dev_set_mtu
);
3289 EXPORT_SYMBOL(dev_set_mac_address
);
3290 EXPORT_SYMBOL(free_netdev
);
3291 EXPORT_SYMBOL(netdev_boot_setup_check
);
3292 EXPORT_SYMBOL(netdev_set_master
);
3293 EXPORT_SYMBOL(netdev_state_change
);
3294 EXPORT_SYMBOL(netif_receive_skb
);
3295 EXPORT_SYMBOL(netif_rx
);
3296 EXPORT_SYMBOL(register_gifconf
);
3297 EXPORT_SYMBOL(register_netdevice
);
3298 EXPORT_SYMBOL(register_netdevice_notifier
);
3299 EXPORT_SYMBOL(skb_checksum_help
);
3300 EXPORT_SYMBOL(synchronize_net
);
3301 EXPORT_SYMBOL(unregister_netdevice
);
3302 EXPORT_SYMBOL(unregister_netdevice_notifier
);
3303 EXPORT_SYMBOL(net_enable_timestamp
);
3304 EXPORT_SYMBOL(net_disable_timestamp
);
3305 EXPORT_SYMBOL(dev_get_flags
);
3307 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3308 EXPORT_SYMBOL(br_handle_frame_hook
);
3309 EXPORT_SYMBOL(br_fdb_get_hook
);
3310 EXPORT_SYMBOL(br_fdb_put_hook
);
3314 EXPORT_SYMBOL(dev_load
);
3317 EXPORT_PER_CPU_SYMBOL(softnet_data
);