2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/config.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/notifier.h>
93 #include <linux/skbuff.h>
95 #include <linux/rtnetlink.h>
96 #include <linux/proc_fs.h>
97 #include <linux/seq_file.h>
98 #include <linux/stat.h>
99 #include <linux/if_bridge.h>
100 #include <linux/divert.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #ifdef CONFIG_NET_RADIO
113 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
114 #include <net/iw_handler.h>
115 #endif /* CONFIG_NET_RADIO */
116 #include <asm/current.h>
118 /* This define, if set, will randomly drop a packet when congestion
119 * is more than moderate. It helps fairness in the multi-interface
120 * case when one of them is a hog, but it kills performance for the
121 * single interface case so it is off now by default.
125 /* Setting this will sample the queue lengths and thus congestion
126 * via a timer instead of as each packet is received.
128 #undef OFFLINE_SAMPLE
131 * The list of packet types we will receive (as opposed to discard)
132 * and the routines to invoke.
134 * Why 16. Because with 16 the only overlap we get on a hash of the
135 * low nibble of the protocol value is RARP/SNAP/X.25.
137 * NOTE: That is no longer true with the addition of VLAN tags. Not
138 * sure which should go first, but I bet it won't make much
139 * difference if we are running VLANs. The good news is that
140 * this protocol won't be in the list unless compiled in, so
141 * the average user (w/out VLANs) will not be adversly affected.
158 static DEFINE_SPINLOCK(ptype_lock
);
159 static struct list_head ptype_base
[16]; /* 16 way hashed list */
160 static struct list_head ptype_all
; /* Taps */
162 #ifdef OFFLINE_SAMPLE
163 static void sample_queue(unsigned long dummy
);
164 static struct timer_list samp_timer
= TIMER_INITIALIZER(sample_queue
, 0, 0);
168 * The @dev_base list is protected by @dev_base_lock and the rtln
171 * Pure readers hold dev_base_lock for reading.
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
186 struct net_device
*dev_base
;
187 static struct net_device
**dev_tail
= &dev_base
;
188 DEFINE_RWLOCK(dev_base_lock
);
190 EXPORT_SYMBOL(dev_base
);
191 EXPORT_SYMBOL(dev_base_lock
);
193 #define NETDEV_HASHBITS 8
194 static struct hlist_head dev_name_head
[1<<NETDEV_HASHBITS
];
195 static struct hlist_head dev_index_head
[1<<NETDEV_HASHBITS
];
197 static inline struct hlist_head
*dev_name_hash(const char *name
)
199 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
200 return &dev_name_head
[hash
& ((1<<NETDEV_HASHBITS
)-1)];
203 static inline struct hlist_head
*dev_index_hash(int ifindex
)
205 return &dev_index_head
[ifindex
& ((1<<NETDEV_HASHBITS
)-1)];
212 static struct notifier_block
*netdev_chain
;
215 * Device drivers call our routines to queue packets here. We empty the
216 * queue in the local softnet handler.
218 DEFINE_PER_CPU(struct softnet_data
, softnet_data
) = { 0, };
221 extern int netdev_sysfs_init(void);
222 extern int netdev_register_sysfs(struct net_device
*);
223 extern void netdev_unregister_sysfs(struct net_device
*);
225 #define netdev_sysfs_init() (0)
226 #define netdev_register_sysfs(dev) (0)
227 #define netdev_unregister_sysfs(dev) do { } while(0)
231 /*******************************************************************************
233 Protocol management and registration routines
235 *******************************************************************************/
244 * Add a protocol ID to the list. Now that the input handler is
245 * smarter we can dispense with all the messy stuff that used to be
248 * BEWARE!!! Protocol handlers, mangling input packets,
249 * MUST BE last in hash buckets and checking protocol handlers
250 * MUST start from promiscuous ptype_all chain in net_bh.
251 * It is true now, do not change it.
252 * Explanation follows: if protocol handler, mangling packet, will
253 * be the first on list, it is not able to sense, that packet
254 * is cloned and should be copied-on-write, so that it will
255 * change it and subsequent readers will get broken packet.
260 * dev_add_pack - add packet handler
261 * @pt: packet type declaration
263 * Add a protocol handler to the networking stack. The passed &packet_type
264 * is linked into kernel lists and may not be freed until it has been
265 * removed from the kernel lists.
267 * This call does not sleep therefore it can not
268 * guarantee all CPU's that are in middle of receiving packets
269 * will see the new packet type (until the next received packet).
272 void dev_add_pack(struct packet_type
*pt
)
276 spin_lock_bh(&ptype_lock
);
277 if (pt
->type
== htons(ETH_P_ALL
)) {
279 list_add_rcu(&pt
->list
, &ptype_all
);
281 hash
= ntohs(pt
->type
) & 15;
282 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
284 spin_unlock_bh(&ptype_lock
);
287 extern void linkwatch_run_queue(void);
292 * __dev_remove_pack - remove packet handler
293 * @pt: packet type declaration
295 * Remove a protocol handler that was previously added to the kernel
296 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
297 * from the kernel lists and can be freed or reused once this function
300 * The packet type might still be in use by receivers
301 * and must not be freed until after all the CPU's have gone
302 * through a quiescent state.
304 void __dev_remove_pack(struct packet_type
*pt
)
306 struct list_head
*head
;
307 struct packet_type
*pt1
;
309 spin_lock_bh(&ptype_lock
);
311 if (pt
->type
== htons(ETH_P_ALL
)) {
315 head
= &ptype_base
[ntohs(pt
->type
) & 15];
317 list_for_each_entry(pt1
, head
, list
) {
319 list_del_rcu(&pt
->list
);
324 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
326 spin_unlock_bh(&ptype_lock
);
329 * dev_remove_pack - remove packet handler
330 * @pt: packet type declaration
332 * Remove a protocol handler that was previously added to the kernel
333 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
334 * from the kernel lists and can be freed or reused once this function
337 * This call sleeps to guarantee that no CPU is looking at the packet
340 void dev_remove_pack(struct packet_type
*pt
)
342 __dev_remove_pack(pt
);
347 /******************************************************************************
349 Device Boot-time Settings Routines
351 *******************************************************************************/
353 /* Boot time configuration table */
354 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
357 * netdev_boot_setup_add - add new setup entry
358 * @name: name of the device
359 * @map: configured settings for the device
361 * Adds new setup entry to the dev_boot_setup list. The function
362 * returns 0 on error and 1 on success. This is a generic routine to
365 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
367 struct netdev_boot_setup
*s
;
371 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
372 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
373 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
374 strcpy(s
[i
].name
, name
);
375 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
380 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
384 * netdev_boot_setup_check - check boot time settings
385 * @dev: the netdevice
387 * Check boot time settings for the device.
388 * The found settings are set for the device to be used
389 * later in the device probing.
390 * Returns 0 if no settings found, 1 if they are.
392 int netdev_boot_setup_check(struct net_device
*dev
)
394 struct netdev_boot_setup
*s
= dev_boot_setup
;
397 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
398 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
399 !strncmp(dev
->name
, s
[i
].name
, strlen(s
[i
].name
))) {
400 dev
->irq
= s
[i
].map
.irq
;
401 dev
->base_addr
= s
[i
].map
.base_addr
;
402 dev
->mem_start
= s
[i
].map
.mem_start
;
403 dev
->mem_end
= s
[i
].map
.mem_end
;
412 * netdev_boot_base - get address from boot time settings
413 * @prefix: prefix for network device
414 * @unit: id for network device
416 * Check boot time settings for the base address of device.
417 * The found settings are set for the device to be used
418 * later in the device probing.
419 * Returns 0 if no settings found.
421 unsigned long netdev_boot_base(const char *prefix
, int unit
)
423 const struct netdev_boot_setup
*s
= dev_boot_setup
;
427 sprintf(name
, "%s%d", prefix
, unit
);
430 * If device already registered then return base of 1
431 * to indicate not to probe for this interface
433 if (__dev_get_by_name(name
))
436 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
437 if (!strcmp(name
, s
[i
].name
))
438 return s
[i
].map
.base_addr
;
443 * Saves at boot time configured settings for any netdevice.
445 int __init
netdev_boot_setup(char *str
)
450 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
455 memset(&map
, 0, sizeof(map
));
459 map
.base_addr
= ints
[2];
461 map
.mem_start
= ints
[3];
463 map
.mem_end
= ints
[4];
465 /* Add new entry to the list */
466 return netdev_boot_setup_add(str
, &map
);
469 __setup("netdev=", netdev_boot_setup
);
471 /*******************************************************************************
473 Device Interface Subroutines
475 *******************************************************************************/
478 * __dev_get_by_name - find a device by its name
479 * @name: name to find
481 * Find an interface by name. Must be called under RTNL semaphore
482 * or @dev_base_lock. If the name is found a pointer to the device
483 * is returned. If the name is not found then %NULL is returned. The
484 * reference counters are not incremented so the caller must be
485 * careful with locks.
488 struct net_device
*__dev_get_by_name(const char *name
)
490 struct hlist_node
*p
;
492 hlist_for_each(p
, dev_name_hash(name
)) {
493 struct net_device
*dev
494 = hlist_entry(p
, struct net_device
, name_hlist
);
495 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
502 * dev_get_by_name - find a device by its name
503 * @name: name to find
505 * Find an interface by name. This can be called from any
506 * context and does its own locking. The returned handle has
507 * the usage count incremented and the caller must use dev_put() to
508 * release it when it is no longer needed. %NULL is returned if no
509 * matching device is found.
512 struct net_device
*dev_get_by_name(const char *name
)
514 struct net_device
*dev
;
516 read_lock(&dev_base_lock
);
517 dev
= __dev_get_by_name(name
);
520 read_unlock(&dev_base_lock
);
525 * __dev_get_by_index - find a device by its ifindex
526 * @ifindex: index of device
528 * Search for an interface by index. Returns %NULL if the device
529 * is not found or a pointer to the device. The device has not
530 * had its reference counter increased so the caller must be careful
531 * about locking. The caller must hold either the RTNL semaphore
535 struct net_device
*__dev_get_by_index(int ifindex
)
537 struct hlist_node
*p
;
539 hlist_for_each(p
, dev_index_hash(ifindex
)) {
540 struct net_device
*dev
541 = hlist_entry(p
, struct net_device
, index_hlist
);
542 if (dev
->ifindex
== ifindex
)
550 * dev_get_by_index - find a device by its ifindex
551 * @ifindex: index of device
553 * Search for an interface by index. Returns NULL if the device
554 * is not found or a pointer to the device. The device returned has
555 * had a reference added and the pointer is safe until the user calls
556 * dev_put to indicate they have finished with it.
559 struct net_device
*dev_get_by_index(int ifindex
)
561 struct net_device
*dev
;
563 read_lock(&dev_base_lock
);
564 dev
= __dev_get_by_index(ifindex
);
567 read_unlock(&dev_base_lock
);
572 * dev_getbyhwaddr - find a device by its hardware address
573 * @type: media type of device
574 * @ha: hardware address
576 * Search for an interface by MAC address. Returns NULL if the device
577 * is not found or a pointer to the device. The caller must hold the
578 * rtnl semaphore. The returned device has not had its ref count increased
579 * and the caller must therefore be careful about locking
582 * If the API was consistent this would be __dev_get_by_hwaddr
585 struct net_device
*dev_getbyhwaddr(unsigned short type
, char *ha
)
587 struct net_device
*dev
;
591 for (dev
= dev_base
; dev
; dev
= dev
->next
)
592 if (dev
->type
== type
&&
593 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
598 struct net_device
*dev_getfirstbyhwtype(unsigned short type
)
600 struct net_device
*dev
;
603 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
604 if (dev
->type
== type
) {
613 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
616 * dev_get_by_flags - find any device with given flags
617 * @if_flags: IFF_* values
618 * @mask: bitmask of bits in if_flags to check
620 * Search for any interface with the given flags. Returns NULL if a device
621 * is not found or a pointer to the device. The device returned has
622 * had a reference added and the pointer is safe until the user calls
623 * dev_put to indicate they have finished with it.
626 struct net_device
* dev_get_by_flags(unsigned short if_flags
, unsigned short mask
)
628 struct net_device
*dev
;
630 read_lock(&dev_base_lock
);
631 for (dev
= dev_base
; dev
!= NULL
; dev
= dev
->next
) {
632 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
637 read_unlock(&dev_base_lock
);
642 * dev_valid_name - check if name is okay for network device
645 * Network device names need to be valid file names to
646 * to allow sysfs to work
648 static int dev_valid_name(const char *name
)
650 return !(*name
== '\0'
651 || !strcmp(name
, ".")
652 || !strcmp(name
, "..")
653 || strchr(name
, '/'));
657 * dev_alloc_name - allocate a name for a device
659 * @name: name format string
661 * Passed a format string - eg "lt%d" it will try and find a suitable
662 * id. Not efficient for many devices, not called a lot. The caller
663 * must hold the dev_base or rtnl lock while allocating the name and
664 * adding the device in order to avoid duplicates. Returns the number
665 * of the unit assigned or a negative errno code.
668 int dev_alloc_name(struct net_device
*dev
, const char *name
)
673 const int max_netdevices
= 8*PAGE_SIZE
;
675 struct net_device
*d
;
677 p
= strnchr(name
, IFNAMSIZ
-1, '%');
680 * Verify the string as this thing may have come from
681 * the user. There must be either one "%d" and no other "%"
684 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
687 /* Use one page as a bit array of possible slots */
688 inuse
= (long *) get_zeroed_page(GFP_ATOMIC
);
692 for (d
= dev_base
; d
; d
= d
->next
) {
693 if (!sscanf(d
->name
, name
, &i
))
695 if (i
< 0 || i
>= max_netdevices
)
698 /* avoid cases where sscanf is not exact inverse of printf */
699 snprintf(buf
, sizeof(buf
), name
, i
);
700 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
704 i
= find_first_zero_bit(inuse
, max_netdevices
);
705 free_page((unsigned long) inuse
);
708 snprintf(buf
, sizeof(buf
), name
, i
);
709 if (!__dev_get_by_name(buf
)) {
710 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
714 /* It is possible to run out of possible slots
715 * when the name is long and there isn't enough space left
716 * for the digits, or if all bits are used.
723 * dev_change_name - change name of a device
725 * @newname: name (or format string) must be at least IFNAMSIZ
727 * Change name of a device, can pass format strings "eth%d".
730 int dev_change_name(struct net_device
*dev
, char *newname
)
736 if (dev
->flags
& IFF_UP
)
739 if (!dev_valid_name(newname
))
742 if (strchr(newname
, '%')) {
743 err
= dev_alloc_name(dev
, newname
);
746 strcpy(newname
, dev
->name
);
748 else if (__dev_get_by_name(newname
))
751 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
753 err
= class_device_rename(&dev
->class_dev
, dev
->name
);
755 hlist_del(&dev
->name_hlist
);
756 hlist_add_head(&dev
->name_hlist
, dev_name_hash(dev
->name
));
757 notifier_call_chain(&netdev_chain
, NETDEV_CHANGENAME
, dev
);
764 * netdev_state_change - device changes state
765 * @dev: device to cause notification
767 * Called to indicate a device has changed state. This function calls
768 * the notifier chains for netdev_chain and sends a NEWLINK message
769 * to the routing socket.
771 void netdev_state_change(struct net_device
*dev
)
773 if (dev
->flags
& IFF_UP
) {
774 notifier_call_chain(&netdev_chain
, NETDEV_CHANGE
, dev
);
775 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
780 * dev_load - load a network module
781 * @name: name of interface
783 * If a network interface is not present and the process has suitable
784 * privileges this function loads the module. If module loading is not
785 * available in this kernel then it becomes a nop.
788 void dev_load(const char *name
)
790 struct net_device
*dev
;
792 read_lock(&dev_base_lock
);
793 dev
= __dev_get_by_name(name
);
794 read_unlock(&dev_base_lock
);
796 if (!dev
&& capable(CAP_SYS_MODULE
))
797 request_module("%s", name
);
800 static int default_rebuild_header(struct sk_buff
*skb
)
802 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
803 skb
->dev
? skb
->dev
->name
: "NULL!!!");
810 * dev_open - prepare an interface for use.
811 * @dev: device to open
813 * Takes a device from down to up state. The device's private open
814 * function is invoked and then the multicast lists are loaded. Finally
815 * the device is moved into the up state and a %NETDEV_UP message is
816 * sent to the netdev notifier chain.
818 * Calling this function on an active interface is a nop. On a failure
819 * a negative errno code is returned.
821 int dev_open(struct net_device
*dev
)
829 if (dev
->flags
& IFF_UP
)
833 * Is it even present?
835 if (!netif_device_present(dev
))
839 * Call device private open method
841 set_bit(__LINK_STATE_START
, &dev
->state
);
843 ret
= dev
->open(dev
);
845 clear_bit(__LINK_STATE_START
, &dev
->state
);
849 * If it went open OK then:
856 dev
->flags
|= IFF_UP
;
859 * Initialize multicasting status
864 * Wakeup transmit queue engine
869 * ... and announce new interface.
871 notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
877 * dev_close - shutdown an interface.
878 * @dev: device to shutdown
880 * This function moves an active device into down state. A
881 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
882 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
885 int dev_close(struct net_device
*dev
)
887 if (!(dev
->flags
& IFF_UP
))
891 * Tell people we are going down, so that they can
892 * prepare to death, when device is still operating.
894 notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
898 clear_bit(__LINK_STATE_START
, &dev
->state
);
900 /* Synchronize to scheduled poll. We cannot touch poll list,
901 * it can be even on different cpu. So just clear netif_running(),
902 * and wait when poll really will happen. Actually, the best place
903 * for this is inside dev->stop() after device stopped its irq
904 * engine, but this requires more changes in devices. */
906 smp_mb__after_clear_bit(); /* Commit netif_running(). */
907 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
909 current
->state
= TASK_INTERRUPTIBLE
;
914 * Call the device specific close. This cannot fail.
915 * Only if device is UP
917 * We allow it to be called even after a DETACH hot-plug
924 * Device is now down.
927 dev
->flags
&= ~IFF_UP
;
930 * Tell people we are down
932 notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
939 * Device change register/unregister. These are not inline or static
940 * as we export them to the world.
944 * register_netdevice_notifier - register a network notifier block
947 * Register a notifier to be called when network device events occur.
948 * The notifier passed is linked into the kernel structures and must
949 * not be reused until it has been unregistered. A negative errno code
950 * is returned on a failure.
952 * When registered all registration and up events are replayed
953 * to the new notifier to allow device to have a race free
954 * view of the network device list.
957 int register_netdevice_notifier(struct notifier_block
*nb
)
959 struct net_device
*dev
;
963 err
= notifier_chain_register(&netdev_chain
, nb
);
965 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
966 nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
968 if (dev
->flags
& IFF_UP
)
969 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
977 * unregister_netdevice_notifier - unregister a network notifier block
980 * Unregister a notifier previously registered by
981 * register_netdevice_notifier(). The notifier is unlinked into the
982 * kernel structures and may then be reused. A negative errno code
983 * is returned on a failure.
986 int unregister_netdevice_notifier(struct notifier_block
*nb
)
988 return 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 notifier_call_chain().
1000 int call_netdevice_notifiers(unsigned long val
, void *v
)
1002 return 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 static inline void net_timestamp(struct timeval
*stamp
)
1020 if (atomic_read(&netstamp_needed
))
1021 do_gettimeofday(stamp
);
1029 * Support routine. Sends outgoing frames to any network
1030 * taps currently in use.
1033 void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1035 struct packet_type
*ptype
;
1036 net_timestamp(&skb
->stamp
);
1039 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1040 /* Never send packets back to the socket
1041 * they originated from - MvS (miquels@drinkel.ow.org)
1043 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1044 (ptype
->af_packet_priv
== NULL
||
1045 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1046 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1050 /* skb->nh should be correctly
1051 set by sender, so that the second statement is
1052 just protection against buggy protocols.
1054 skb2
->mac
.raw
= skb2
->data
;
1056 if (skb2
->nh
.raw
< skb2
->data
||
1057 skb2
->nh
.raw
> skb2
->tail
) {
1058 if (net_ratelimit())
1059 printk(KERN_CRIT
"protocol %04x is "
1061 skb2
->protocol
, dev
->name
);
1062 skb2
->nh
.raw
= skb2
->data
;
1065 skb2
->h
.raw
= skb2
->nh
.raw
;
1066 skb2
->pkt_type
= PACKET_OUTGOING
;
1067 ptype
->func(skb2
, skb
->dev
, ptype
);
1074 * Invalidate hardware checksum when packet is to be mangled, and
1075 * complete checksum manually on outgoing path.
1077 int skb_checksum_help(struct sk_buff
*skb
, int inward
)
1080 int ret
= 0, offset
= skb
->h
.raw
- skb
->data
;
1083 skb
->ip_summed
= CHECKSUM_NONE
;
1087 if (skb_cloned(skb
)) {
1088 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1093 if (offset
> (int)skb
->len
)
1095 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1097 offset
= skb
->tail
- skb
->h
.raw
;
1100 if (skb
->csum
+ 2 > offset
)
1103 *(u16
*)(skb
->h
.raw
+ skb
->csum
) = csum_fold(csum
);
1104 skb
->ip_summed
= CHECKSUM_NONE
;
1109 #ifdef CONFIG_HIGHMEM
1110 /* Actually, we should eliminate this check as soon as we know, that:
1111 * 1. IOMMU is present and allows to map all the memory.
1112 * 2. No high memory really exists on this machine.
1115 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1119 if (dev
->features
& NETIF_F_HIGHDMA
)
1122 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1123 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1129 #define illegal_highdma(dev, skb) (0)
1132 extern void skb_release_data(struct sk_buff
*);
1134 /* Keep head the same: replace data */
1135 int __skb_linearize(struct sk_buff
*skb
, int gfp_mask
)
1140 struct skb_shared_info
*ninfo
;
1141 int headerlen
= skb
->data
- skb
->head
;
1142 int expand
= (skb
->tail
+ skb
->data_len
) - skb
->end
;
1144 if (skb_shared(skb
))
1150 size
= skb
->end
- skb
->head
+ expand
;
1151 size
= SKB_DATA_ALIGN(size
);
1152 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
1156 /* Copy entire thing */
1157 if (skb_copy_bits(skb
, -headerlen
, data
, headerlen
+ skb
->len
))
1161 ninfo
= (struct skb_shared_info
*)(data
+ size
);
1162 atomic_set(&ninfo
->dataref
, 1);
1163 ninfo
->tso_size
= skb_shinfo(skb
)->tso_size
;
1164 ninfo
->tso_segs
= skb_shinfo(skb
)->tso_segs
;
1165 ninfo
->nr_frags
= 0;
1166 ninfo
->frag_list
= NULL
;
1168 /* Offset between the two in bytes */
1169 offset
= data
- skb
->head
;
1171 /* Free old data. */
1172 skb_release_data(skb
);
1175 skb
->end
= data
+ size
;
1177 /* Set up new pointers */
1178 skb
->h
.raw
+= offset
;
1179 skb
->nh
.raw
+= offset
;
1180 skb
->mac
.raw
+= offset
;
1181 skb
->tail
+= offset
;
1182 skb
->data
+= offset
;
1184 /* We are no longer a clone, even if we were. */
1187 skb
->tail
+= skb
->data_len
;
1192 #define HARD_TX_LOCK(dev, cpu) { \
1193 if ((dev->features & NETIF_F_LLTX) == 0) { \
1194 spin_lock(&dev->xmit_lock); \
1195 dev->xmit_lock_owner = cpu; \
1199 #define HARD_TX_UNLOCK(dev) { \
1200 if ((dev->features & NETIF_F_LLTX) == 0) { \
1201 dev->xmit_lock_owner = -1; \
1202 spin_unlock(&dev->xmit_lock); \
1207 * dev_queue_xmit - transmit a buffer
1208 * @skb: buffer to transmit
1210 * Queue a buffer for transmission to a network device. The caller must
1211 * have set the device and priority and built the buffer before calling
1212 * this function. The function can be called from an interrupt.
1214 * A negative errno code is returned on a failure. A success does not
1215 * guarantee the frame will be transmitted as it may be dropped due
1216 * to congestion or traffic shaping.
1218 * -----------------------------------------------------------------------------------
1219 * I notice this method can also return errors from the queue disciplines,
1220 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1223 * Regardless of the return value, the skb is consumed, so it is currently
1224 * difficult to retry a send to this method. (You can bump the ref count
1225 * before sending to hold a reference for retry if you are careful.)
1227 * When calling this method, interrupts MUST be enabled. This is because
1228 * the BH enable code must have IRQs enabled so that it will not deadlock.
1232 int dev_queue_xmit(struct sk_buff
*skb
)
1234 struct net_device
*dev
= skb
->dev
;
1238 if (skb_shinfo(skb
)->frag_list
&&
1239 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1240 __skb_linearize(skb
, GFP_ATOMIC
))
1243 /* Fragmented skb is linearized if device does not support SG,
1244 * or if at least one of fragments is in highmem and device
1245 * does not support DMA from it.
1247 if (skb_shinfo(skb
)->nr_frags
&&
1248 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1249 __skb_linearize(skb
, GFP_ATOMIC
))
1252 /* If packet is not checksummed and device does not support
1253 * checksumming for this protocol, complete checksumming here.
1255 if (skb
->ip_summed
== CHECKSUM_HW
&&
1256 (!(dev
->features
& (NETIF_F_HW_CSUM
| NETIF_F_NO_CSUM
)) &&
1257 (!(dev
->features
& NETIF_F_IP_CSUM
) ||
1258 skb
->protocol
!= htons(ETH_P_IP
))))
1259 if (skb_checksum_help(skb
, 0))
1262 /* Disable soft irqs for various locks below. Also
1263 * stops preemption for RCU.
1267 /* Updates of qdisc are serialized by queue_lock.
1268 * The struct Qdisc which is pointed to by qdisc is now a
1269 * rcu structure - it may be accessed without acquiring
1270 * a lock (but the structure may be stale.) The freeing of the
1271 * qdisc will be deferred until it's known that there are no
1272 * more references to it.
1274 * If the qdisc has an enqueue function, we still need to
1275 * hold the queue_lock before calling it, since queue_lock
1276 * also serializes access to the device queue.
1279 q
= rcu_dereference(dev
->qdisc
);
1280 #ifdef CONFIG_NET_CLS_ACT
1281 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1284 /* Grab device queue */
1285 spin_lock(&dev
->queue_lock
);
1287 rc
= q
->enqueue(skb
, q
);
1291 spin_unlock(&dev
->queue_lock
);
1292 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1296 /* The device has no queue. Common case for software devices:
1297 loopback, all the sorts of tunnels...
1299 Really, it is unlikely that xmit_lock protection is necessary here.
1300 (f.e. loopback and IP tunnels are clean ignoring statistics
1302 However, it is possible, that they rely on protection
1305 Check this and shot the lock. It is not prone from deadlocks.
1306 Either shot noqueue qdisc, it is even simpler 8)
1308 if (dev
->flags
& IFF_UP
) {
1309 int cpu
= smp_processor_id(); /* ok because BHs are off */
1311 if (dev
->xmit_lock_owner
!= cpu
) {
1313 HARD_TX_LOCK(dev
, cpu
);
1315 if (!netif_queue_stopped(dev
)) {
1317 dev_queue_xmit_nit(skb
, dev
);
1320 if (!dev
->hard_start_xmit(skb
, dev
)) {
1321 HARD_TX_UNLOCK(dev
);
1325 HARD_TX_UNLOCK(dev
);
1326 if (net_ratelimit())
1327 printk(KERN_CRIT
"Virtual device %s asks to "
1328 "queue packet!\n", dev
->name
);
1330 /* Recursion is detected! It is possible,
1332 if (net_ratelimit())
1333 printk(KERN_CRIT
"Dead loop on virtual device "
1334 "%s, fix it urgently!\n", dev
->name
);
1350 /*=======================================================================
1352 =======================================================================*/
1354 int netdev_max_backlog
= 300;
1355 int weight_p
= 64; /* old backlog weight */
1356 /* These numbers are selected based on intuition and some
1357 * experimentatiom, if you have more scientific way of doing this
1358 * please go ahead and fix things.
1360 int no_cong_thresh
= 10;
1365 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1368 static void get_sample_stats(int cpu
)
1374 struct softnet_data
*sd
= &per_cpu(softnet_data
, cpu
);
1375 int blog
= sd
->input_pkt_queue
.qlen
;
1376 int avg_blog
= sd
->avg_blog
;
1378 avg_blog
= (avg_blog
>> 1) + (blog
>> 1);
1380 if (avg_blog
> mod_cong
) {
1381 /* Above moderate congestion levels. */
1382 sd
->cng_level
= NET_RX_CN_HIGH
;
1385 rq
= rd
% netdev_max_backlog
;
1386 if (rq
< avg_blog
) /* unlucky bastard */
1387 sd
->cng_level
= NET_RX_DROP
;
1389 } else if (avg_blog
> lo_cong
) {
1390 sd
->cng_level
= NET_RX_CN_MOD
;
1393 rq
= rd
% netdev_max_backlog
;
1394 if (rq
< avg_blog
) /* unlucky bastard */
1395 sd
->cng_level
= NET_RX_CN_HIGH
;
1397 } else if (avg_blog
> no_cong
)
1398 sd
->cng_level
= NET_RX_CN_LOW
;
1399 else /* no congestion */
1400 sd
->cng_level
= NET_RX_SUCCESS
;
1402 sd
->avg_blog
= avg_blog
;
1405 #ifdef OFFLINE_SAMPLE
1406 static void sample_queue(unsigned long dummy
)
1408 /* 10 ms 0r 1ms -- i don't care -- JHS */
1410 int cpu
= smp_processor_id();
1412 get_sample_stats(cpu
);
1413 next_tick
+= jiffies
;
1414 mod_timer(&samp_timer
, next_tick
);
1420 * netif_rx - post buffer to the network code
1421 * @skb: buffer to post
1423 * This function receives a packet from a device driver and queues it for
1424 * the upper (protocol) levels to process. It always succeeds. The buffer
1425 * may be dropped during processing for congestion control or by the
1429 * NET_RX_SUCCESS (no congestion)
1430 * NET_RX_CN_LOW (low congestion)
1431 * NET_RX_CN_MOD (moderate congestion)
1432 * NET_RX_CN_HIGH (high congestion)
1433 * NET_RX_DROP (packet was dropped)
1437 int netif_rx(struct sk_buff
*skb
)
1440 struct softnet_data
*queue
;
1441 unsigned long flags
;
1443 /* if netpoll wants it, pretend we never saw it */
1444 if (netpoll_rx(skb
))
1447 if (!skb
->stamp
.tv_sec
)
1448 net_timestamp(&skb
->stamp
);
1451 * The code is rearranged so that the path is the most
1452 * short when CPU is congested, but is still operating.
1454 local_irq_save(flags
);
1455 this_cpu
= smp_processor_id();
1456 queue
= &__get_cpu_var(softnet_data
);
1458 __get_cpu_var(netdev_rx_stat
).total
++;
1459 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1460 if (queue
->input_pkt_queue
.qlen
) {
1461 if (queue
->throttle
)
1466 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1467 #ifndef OFFLINE_SAMPLE
1468 get_sample_stats(this_cpu
);
1470 local_irq_restore(flags
);
1471 return queue
->cng_level
;
1474 if (queue
->throttle
)
1475 queue
->throttle
= 0;
1477 netif_rx_schedule(&queue
->backlog_dev
);
1481 if (!queue
->throttle
) {
1482 queue
->throttle
= 1;
1483 __get_cpu_var(netdev_rx_stat
).throttled
++;
1487 __get_cpu_var(netdev_rx_stat
).dropped
++;
1488 local_irq_restore(flags
);
1494 int netif_rx_ni(struct sk_buff
*skb
)
1499 err
= netif_rx(skb
);
1500 if (local_softirq_pending())
1507 EXPORT_SYMBOL(netif_rx_ni
);
1509 static __inline__
void skb_bond(struct sk_buff
*skb
)
1511 struct net_device
*dev
= skb
->dev
;
1514 skb
->real_dev
= skb
->dev
;
1515 skb
->dev
= dev
->master
;
1519 static void net_tx_action(struct softirq_action
*h
)
1521 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1523 if (sd
->completion_queue
) {
1524 struct sk_buff
*clist
;
1526 local_irq_disable();
1527 clist
= sd
->completion_queue
;
1528 sd
->completion_queue
= NULL
;
1532 struct sk_buff
*skb
= clist
;
1533 clist
= clist
->next
;
1535 BUG_TRAP(!atomic_read(&skb
->users
));
1540 if (sd
->output_queue
) {
1541 struct net_device
*head
;
1543 local_irq_disable();
1544 head
= sd
->output_queue
;
1545 sd
->output_queue
= NULL
;
1549 struct net_device
*dev
= head
;
1550 head
= head
->next_sched
;
1552 smp_mb__before_clear_bit();
1553 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1555 if (spin_trylock(&dev
->queue_lock
)) {
1557 spin_unlock(&dev
->queue_lock
);
1559 netif_schedule(dev
);
1565 static __inline__
int deliver_skb(struct sk_buff
*skb
,
1566 struct packet_type
*pt_prev
)
1568 atomic_inc(&skb
->users
);
1569 return pt_prev
->func(skb
, skb
->dev
, pt_prev
);
1572 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1573 int (*br_handle_frame_hook
)(struct net_bridge_port
*p
, struct sk_buff
**pskb
);
1575 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1576 unsigned char *addr
);
1577 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
);
1579 static __inline__
int handle_bridge(struct sk_buff
**pskb
,
1580 struct packet_type
**pt_prev
, int *ret
)
1582 struct net_bridge_port
*port
;
1584 if ((*pskb
)->pkt_type
== PACKET_LOOPBACK
||
1585 (port
= rcu_dereference((*pskb
)->dev
->br_port
)) == NULL
)
1589 *ret
= deliver_skb(*pskb
, *pt_prev
);
1593 return br_handle_frame_hook(port
, pskb
);
1596 #define handle_bridge(skb, pt_prev, ret) (0)
1599 #ifdef CONFIG_NET_CLS_ACT
1600 /* TODO: Maybe we should just force sch_ingress to be compiled in
1601 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1602 * a compare and 2 stores extra right now if we dont have it on
1603 * but have CONFIG_NET_CLS_ACT
1604 * NOTE: This doesnt stop any functionality; if you dont have
1605 * the ingress scheduler, you just cant add policies on ingress.
1608 static int ing_filter(struct sk_buff
*skb
)
1611 struct net_device
*dev
= skb
->dev
;
1612 int result
= TC_ACT_OK
;
1614 if (dev
->qdisc_ingress
) {
1615 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1616 if (MAX_RED_LOOP
< ttl
++) {
1617 printk("Redir loop detected Dropping packet (%s->%s)\n",
1618 skb
->input_dev
?skb
->input_dev
->name
:"??",skb
->dev
->name
);
1622 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1624 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1625 if (NULL
== skb
->input_dev
) {
1626 skb
->input_dev
= skb
->dev
;
1627 printk("ing_filter: fixed %s out %s\n",skb
->input_dev
->name
,skb
->dev
->name
);
1629 spin_lock(&dev
->ingress_lock
);
1630 if ((q
= dev
->qdisc_ingress
) != NULL
)
1631 result
= q
->enqueue(skb
, q
);
1632 spin_unlock(&dev
->ingress_lock
);
1640 int netif_receive_skb(struct sk_buff
*skb
)
1642 struct packet_type
*ptype
, *pt_prev
;
1643 int ret
= NET_RX_DROP
;
1644 unsigned short type
;
1646 /* if we've gotten here through NAPI, check netpoll */
1647 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1650 if (!skb
->stamp
.tv_sec
)
1651 net_timestamp(&skb
->stamp
);
1655 __get_cpu_var(netdev_rx_stat
).total
++;
1657 skb
->h
.raw
= skb
->nh
.raw
= skb
->data
;
1658 skb
->mac_len
= skb
->nh
.raw
- skb
->mac
.raw
;
1664 #ifdef CONFIG_NET_CLS_ACT
1665 if (skb
->tc_verd
& TC_NCLS
) {
1666 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1671 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1672 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1674 ret
= deliver_skb(skb
, pt_prev
);
1679 #ifdef CONFIG_NET_CLS_ACT
1681 ret
= deliver_skb(skb
, pt_prev
);
1682 pt_prev
= NULL
; /* noone else should process this after*/
1684 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1687 ret
= ing_filter(skb
);
1689 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
1698 handle_diverter(skb
);
1700 if (handle_bridge(&skb
, &pt_prev
, &ret
))
1703 type
= skb
->protocol
;
1704 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
1705 if (ptype
->type
== type
&&
1706 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
1708 ret
= deliver_skb(skb
, pt_prev
);
1714 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
);
1717 /* Jamal, now you will not able to escape explaining
1718 * me how you were going to use this. :-)
1728 static int process_backlog(struct net_device
*backlog_dev
, int *budget
)
1731 int quota
= min(backlog_dev
->quota
, *budget
);
1732 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1733 unsigned long start_time
= jiffies
;
1736 struct sk_buff
*skb
;
1737 struct net_device
*dev
;
1739 local_irq_disable();
1740 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
1747 netif_receive_skb(skb
);
1753 if (work
>= quota
|| jiffies
- start_time
> 1)
1758 backlog_dev
->quota
-= work
;
1763 backlog_dev
->quota
-= work
;
1766 list_del(&backlog_dev
->poll_list
);
1767 smp_mb__before_clear_bit();
1768 netif_poll_enable(backlog_dev
);
1770 if (queue
->throttle
)
1771 queue
->throttle
= 0;
1776 static void net_rx_action(struct softirq_action
*h
)
1778 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1779 unsigned long start_time
= jiffies
;
1780 int budget
= netdev_max_backlog
;
1783 local_irq_disable();
1785 while (!list_empty(&queue
->poll_list
)) {
1786 struct net_device
*dev
;
1788 if (budget
<= 0 || jiffies
- start_time
> 1)
1793 dev
= list_entry(queue
->poll_list
.next
,
1794 struct net_device
, poll_list
);
1795 netpoll_poll_lock(dev
);
1797 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
1798 netpoll_poll_unlock(dev
);
1799 local_irq_disable();
1800 list_del(&dev
->poll_list
);
1801 list_add_tail(&dev
->poll_list
, &queue
->poll_list
);
1803 dev
->quota
+= dev
->weight
;
1805 dev
->quota
= dev
->weight
;
1807 netpoll_poll_unlock(dev
);
1809 local_irq_disable();
1817 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
1818 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1822 static gifconf_func_t
* gifconf_list
[NPROTO
];
1825 * register_gifconf - register a SIOCGIF handler
1826 * @family: Address family
1827 * @gifconf: Function handler
1829 * Register protocol dependent address dumping routines. The handler
1830 * that is passed must not be freed or reused until it has been replaced
1831 * by another handler.
1833 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
1835 if (family
>= NPROTO
)
1837 gifconf_list
[family
] = gifconf
;
1843 * Map an interface index to its name (SIOCGIFNAME)
1847 * We need this ioctl for efficient implementation of the
1848 * if_indextoname() function required by the IPv6 API. Without
1849 * it, we would have to search all the interfaces to find a
1853 static int dev_ifname(struct ifreq __user
*arg
)
1855 struct net_device
*dev
;
1859 * Fetch the caller's info block.
1862 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
1865 read_lock(&dev_base_lock
);
1866 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
1868 read_unlock(&dev_base_lock
);
1872 strcpy(ifr
.ifr_name
, dev
->name
);
1873 read_unlock(&dev_base_lock
);
1875 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
1881 * Perform a SIOCGIFCONF call. This structure will change
1882 * size eventually, and there is nothing I can do about it.
1883 * Thus we will need a 'compatibility mode'.
1886 static int dev_ifconf(char __user
*arg
)
1889 struct net_device
*dev
;
1896 * Fetch the caller's info block.
1899 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
1906 * Loop over the interfaces, and write an info block for each.
1910 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
1911 for (i
= 0; i
< NPROTO
; i
++) {
1912 if (gifconf_list
[i
]) {
1915 done
= gifconf_list
[i
](dev
, NULL
, 0);
1917 done
= gifconf_list
[i
](dev
, pos
+ total
,
1927 * All done. Write the updated control block back to the caller.
1929 ifc
.ifc_len
= total
;
1932 * Both BSD and Solaris return 0 here, so we do too.
1934 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
1937 #ifdef CONFIG_PROC_FS
1939 * This is invoked by the /proc filesystem handler to display a device
1942 static __inline__
struct net_device
*dev_get_idx(loff_t pos
)
1944 struct net_device
*dev
;
1947 for (i
= 0, dev
= dev_base
; dev
&& i
< pos
; ++i
, dev
= dev
->next
);
1949 return i
== pos
? dev
: NULL
;
1952 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1954 read_lock(&dev_base_lock
);
1955 return *pos
? dev_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
1958 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1961 return v
== SEQ_START_TOKEN
? dev_base
: ((struct net_device
*)v
)->next
;
1964 void dev_seq_stop(struct seq_file
*seq
, void *v
)
1966 read_unlock(&dev_base_lock
);
1969 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
1971 if (dev
->get_stats
) {
1972 struct net_device_stats
*stats
= dev
->get_stats(dev
);
1974 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1975 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1976 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
1978 stats
->rx_dropped
+ stats
->rx_missed_errors
,
1979 stats
->rx_fifo_errors
,
1980 stats
->rx_length_errors
+ stats
->rx_over_errors
+
1981 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
1982 stats
->rx_compressed
, stats
->multicast
,
1983 stats
->tx_bytes
, stats
->tx_packets
,
1984 stats
->tx_errors
, stats
->tx_dropped
,
1985 stats
->tx_fifo_errors
, stats
->collisions
,
1986 stats
->tx_carrier_errors
+
1987 stats
->tx_aborted_errors
+
1988 stats
->tx_window_errors
+
1989 stats
->tx_heartbeat_errors
,
1990 stats
->tx_compressed
);
1992 seq_printf(seq
, "%6s: No statistics available.\n", dev
->name
);
1996 * Called from the PROCfs module. This now uses the new arbitrary sized
1997 * /proc/net interface to create /proc/net/dev
1999 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2001 if (v
== SEQ_START_TOKEN
)
2002 seq_puts(seq
, "Inter-| Receive "
2004 " face |bytes packets errs drop fifo frame "
2005 "compressed multicast|bytes packets errs "
2006 "drop fifo colls carrier compressed\n");
2008 dev_seq_printf_stats(seq
, v
);
2012 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2014 struct netif_rx_stats
*rc
= NULL
;
2016 while (*pos
< NR_CPUS
)
2017 if (cpu_online(*pos
)) {
2018 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2025 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2027 return softnet_get_online(pos
);
2030 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2033 return softnet_get_online(pos
);
2036 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2040 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2042 struct netif_rx_stats
*s
= v
;
2044 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2045 s
->total
, s
->dropped
, s
->time_squeeze
, s
->throttled
,
2046 s
->fastroute_hit
, s
->fastroute_success
, s
->fastroute_defer
,
2047 s
->fastroute_deferred_out
,
2049 s
->fastroute_latency_reduction
2057 static struct seq_operations dev_seq_ops
= {
2058 .start
= dev_seq_start
,
2059 .next
= dev_seq_next
,
2060 .stop
= dev_seq_stop
,
2061 .show
= dev_seq_show
,
2064 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2066 return seq_open(file
, &dev_seq_ops
);
2069 static struct file_operations dev_seq_fops
= {
2070 .owner
= THIS_MODULE
,
2071 .open
= dev_seq_open
,
2073 .llseek
= seq_lseek
,
2074 .release
= seq_release
,
2077 static struct seq_operations softnet_seq_ops
= {
2078 .start
= softnet_seq_start
,
2079 .next
= softnet_seq_next
,
2080 .stop
= softnet_seq_stop
,
2081 .show
= softnet_seq_show
,
2084 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2086 return seq_open(file
, &softnet_seq_ops
);
2089 static struct file_operations softnet_seq_fops
= {
2090 .owner
= THIS_MODULE
,
2091 .open
= softnet_seq_open
,
2093 .llseek
= seq_lseek
,
2094 .release
= seq_release
,
2098 extern int wireless_proc_init(void);
2100 #define wireless_proc_init() 0
2103 static int __init
dev_proc_init(void)
2107 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2109 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2111 if (wireless_proc_init())
2117 proc_net_remove("softnet_stat");
2119 proc_net_remove("dev");
2123 #define dev_proc_init() 0
2124 #endif /* CONFIG_PROC_FS */
2128 * netdev_set_master - set up master/slave pair
2129 * @slave: slave device
2130 * @master: new master device
2132 * Changes the master device of the slave. Pass %NULL to break the
2133 * bonding. The caller must hold the RTNL semaphore. On a failure
2134 * a negative errno code is returned. On success the reference counts
2135 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2136 * function returns zero.
2138 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2140 struct net_device
*old
= slave
->master
;
2150 slave
->master
= master
;
2158 slave
->flags
|= IFF_SLAVE
;
2160 slave
->flags
&= ~IFF_SLAVE
;
2162 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2167 * dev_set_promiscuity - update promiscuity count on a device
2171 * Add or remove promsicuity from a device. While the count in the device
2172 * remains above zero the interface remains promiscuous. Once it hits zero
2173 * the device reverts back to normal filtering operation. A negative inc
2174 * value is used to drop promiscuity on the device.
2176 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2178 unsigned short old_flags
= dev
->flags
;
2180 dev
->flags
|= IFF_PROMISC
;
2181 if ((dev
->promiscuity
+= inc
) == 0)
2182 dev
->flags
&= ~IFF_PROMISC
;
2183 if (dev
->flags
^ old_flags
) {
2185 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2186 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2192 * dev_set_allmulti - update allmulti count on a device
2196 * Add or remove reception of all multicast frames to a device. While the
2197 * count in the device remains above zero the interface remains listening
2198 * to all interfaces. Once it hits zero the device reverts back to normal
2199 * filtering operation. A negative @inc value is used to drop the counter
2200 * when releasing a resource needing all multicasts.
2203 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2205 unsigned short old_flags
= dev
->flags
;
2207 dev
->flags
|= IFF_ALLMULTI
;
2208 if ((dev
->allmulti
+= inc
) == 0)
2209 dev
->flags
&= ~IFF_ALLMULTI
;
2210 if (dev
->flags
^ old_flags
)
2214 unsigned dev_get_flags(const struct net_device
*dev
)
2218 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2221 (dev
->gflags
& (IFF_PROMISC
|
2224 if (netif_running(dev
) && netif_carrier_ok(dev
))
2225 flags
|= IFF_RUNNING
;
2230 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2233 int old_flags
= dev
->flags
;
2236 * Set the flags on our device.
2239 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2240 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2242 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2246 * Load in the correct multicast list now the flags have changed.
2252 * Have we downed the interface. We handle IFF_UP ourselves
2253 * according to user attempts to set it, rather than blindly
2258 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2259 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2265 if (dev
->flags
& IFF_UP
&&
2266 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2268 notifier_call_chain(&netdev_chain
, NETDEV_CHANGE
, dev
);
2270 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2271 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2272 dev
->gflags
^= IFF_PROMISC
;
2273 dev_set_promiscuity(dev
, inc
);
2276 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2277 is important. Some (broken) drivers set IFF_PROMISC, when
2278 IFF_ALLMULTI is requested not asking us and not reporting.
2280 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2281 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2282 dev
->gflags
^= IFF_ALLMULTI
;
2283 dev_set_allmulti(dev
, inc
);
2286 if (old_flags
^ dev
->flags
)
2287 rtmsg_ifinfo(RTM_NEWLINK
, dev
, old_flags
^ dev
->flags
);
2292 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2296 if (new_mtu
== dev
->mtu
)
2299 /* MTU must be positive. */
2303 if (!netif_device_present(dev
))
2307 if (dev
->change_mtu
)
2308 err
= dev
->change_mtu(dev
, new_mtu
);
2311 if (!err
&& dev
->flags
& IFF_UP
)
2312 notifier_call_chain(&netdev_chain
,
2313 NETDEV_CHANGEMTU
, dev
);
2317 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2321 if (!dev
->set_mac_address
)
2323 if (sa
->sa_family
!= dev
->type
)
2325 if (!netif_device_present(dev
))
2327 err
= dev
->set_mac_address(dev
, sa
);
2329 notifier_call_chain(&netdev_chain
, NETDEV_CHANGEADDR
, dev
);
2334 * Perform the SIOCxIFxxx calls.
2336 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2339 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2345 case SIOCGIFFLAGS
: /* Get interface flags */
2346 ifr
->ifr_flags
= dev_get_flags(dev
);
2349 case SIOCSIFFLAGS
: /* Set interface flags */
2350 return dev_change_flags(dev
, ifr
->ifr_flags
);
2352 case SIOCGIFMETRIC
: /* Get the metric on the interface
2353 (currently unused) */
2354 ifr
->ifr_metric
= 0;
2357 case SIOCSIFMETRIC
: /* Set the metric on the interface
2358 (currently unused) */
2361 case SIOCGIFMTU
: /* Get the MTU of a device */
2362 ifr
->ifr_mtu
= dev
->mtu
;
2365 case SIOCSIFMTU
: /* Set the MTU of a device */
2366 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
2370 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
2372 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
2373 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2374 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
2378 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
2380 case SIOCSIFHWBROADCAST
:
2381 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
2383 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
2384 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2385 notifier_call_chain(&netdev_chain
,
2386 NETDEV_CHANGEADDR
, dev
);
2390 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
2391 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
2392 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
2393 ifr
->ifr_map
.irq
= dev
->irq
;
2394 ifr
->ifr_map
.dma
= dev
->dma
;
2395 ifr
->ifr_map
.port
= dev
->if_port
;
2399 if (dev
->set_config
) {
2400 if (!netif_device_present(dev
))
2402 return dev
->set_config(dev
, &ifr
->ifr_map
);
2407 if (!dev
->set_multicast_list
||
2408 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2410 if (!netif_device_present(dev
))
2412 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
2416 if (!dev
->set_multicast_list
||
2417 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2419 if (!netif_device_present(dev
))
2421 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
2425 ifr
->ifr_ifindex
= dev
->ifindex
;
2429 ifr
->ifr_qlen
= dev
->tx_queue_len
;
2433 if (ifr
->ifr_qlen
< 0)
2435 dev
->tx_queue_len
= ifr
->ifr_qlen
;
2439 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
2440 return dev_change_name(dev
, ifr
->ifr_newname
);
2443 * Unknown or private ioctl
2447 if ((cmd
>= SIOCDEVPRIVATE
&&
2448 cmd
<= SIOCDEVPRIVATE
+ 15) ||
2449 cmd
== SIOCBONDENSLAVE
||
2450 cmd
== SIOCBONDRELEASE
||
2451 cmd
== SIOCBONDSETHWADDR
||
2452 cmd
== SIOCBONDSLAVEINFOQUERY
||
2453 cmd
== SIOCBONDINFOQUERY
||
2454 cmd
== SIOCBONDCHANGEACTIVE
||
2455 cmd
== SIOCGMIIPHY
||
2456 cmd
== SIOCGMIIREG
||
2457 cmd
== SIOCSMIIREG
||
2458 cmd
== SIOCBRADDIF
||
2459 cmd
== SIOCBRDELIF
||
2460 cmd
== SIOCWANDEV
) {
2462 if (dev
->do_ioctl
) {
2463 if (netif_device_present(dev
))
2464 err
= dev
->do_ioctl(dev
, ifr
,
2477 * This function handles all "interface"-type I/O control requests. The actual
2478 * 'doing' part of this is dev_ifsioc above.
2482 * dev_ioctl - network device ioctl
2483 * @cmd: command to issue
2484 * @arg: pointer to a struct ifreq in user space
2486 * Issue ioctl functions to devices. This is normally called by the
2487 * user space syscall interfaces but can sometimes be useful for
2488 * other purposes. The return value is the return from the syscall if
2489 * positive or a negative errno code on error.
2492 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
2498 /* One special case: SIOCGIFCONF takes ifconf argument
2499 and requires shared lock, because it sleeps writing
2503 if (cmd
== SIOCGIFCONF
) {
2505 ret
= dev_ifconf((char __user
*) arg
);
2509 if (cmd
== SIOCGIFNAME
)
2510 return dev_ifname((struct ifreq __user
*)arg
);
2512 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2515 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
2517 colon
= strchr(ifr
.ifr_name
, ':');
2522 * See which interface the caller is talking about.
2527 * These ioctl calls:
2528 * - can be done by all.
2529 * - atomic and do not require locking.
2540 dev_load(ifr
.ifr_name
);
2541 read_lock(&dev_base_lock
);
2542 ret
= dev_ifsioc(&ifr
, cmd
);
2543 read_unlock(&dev_base_lock
);
2547 if (copy_to_user(arg
, &ifr
,
2548 sizeof(struct ifreq
)))
2554 dev_load(ifr
.ifr_name
);
2556 ret
= dev_ethtool(&ifr
);
2561 if (copy_to_user(arg
, &ifr
,
2562 sizeof(struct ifreq
)))
2568 * These ioctl calls:
2569 * - require superuser power.
2570 * - require strict serialization.
2576 if (!capable(CAP_NET_ADMIN
))
2578 dev_load(ifr
.ifr_name
);
2580 ret
= dev_ifsioc(&ifr
, cmd
);
2585 if (copy_to_user(arg
, &ifr
,
2586 sizeof(struct ifreq
)))
2592 * These ioctl calls:
2593 * - require superuser power.
2594 * - require strict serialization.
2595 * - do not return a value
2605 case SIOCSIFHWBROADCAST
:
2608 case SIOCBONDENSLAVE
:
2609 case SIOCBONDRELEASE
:
2610 case SIOCBONDSETHWADDR
:
2611 case SIOCBONDSLAVEINFOQUERY
:
2612 case SIOCBONDINFOQUERY
:
2613 case SIOCBONDCHANGEACTIVE
:
2616 if (!capable(CAP_NET_ADMIN
))
2618 dev_load(ifr
.ifr_name
);
2620 ret
= dev_ifsioc(&ifr
, cmd
);
2625 /* Get the per device memory space. We can add this but
2626 * currently do not support it */
2628 /* Set the per device memory buffer space.
2629 * Not applicable in our case */
2634 * Unknown or private ioctl.
2637 if (cmd
== SIOCWANDEV
||
2638 (cmd
>= SIOCDEVPRIVATE
&&
2639 cmd
<= SIOCDEVPRIVATE
+ 15)) {
2640 dev_load(ifr
.ifr_name
);
2642 ret
= dev_ifsioc(&ifr
, cmd
);
2644 if (!ret
&& copy_to_user(arg
, &ifr
,
2645 sizeof(struct ifreq
)))
2650 /* Take care of Wireless Extensions */
2651 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
) {
2652 /* If command is `set a parameter', or
2653 * `get the encoding parameters', check if
2654 * the user has the right to do it */
2655 if (IW_IS_SET(cmd
) || cmd
== SIOCGIWENCODE
) {
2656 if (!capable(CAP_NET_ADMIN
))
2659 dev_load(ifr
.ifr_name
);
2661 /* Follow me in net/core/wireless.c */
2662 ret
= wireless_process_ioctl(&ifr
, cmd
);
2664 if (IW_IS_GET(cmd
) &&
2665 copy_to_user(arg
, &ifr
,
2666 sizeof(struct ifreq
)))
2670 #endif /* WIRELESS_EXT */
2677 * dev_new_index - allocate an ifindex
2679 * Returns a suitable unique value for a new device interface
2680 * number. The caller must hold the rtnl semaphore or the
2681 * dev_base_lock to be sure it remains unique.
2683 static int dev_new_index(void)
2689 if (!__dev_get_by_index(ifindex
))
2694 static int dev_boot_phase
= 1;
2696 /* Delayed registration/unregisteration */
2697 static DEFINE_SPINLOCK(net_todo_list_lock
);
2698 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
2700 static inline void net_set_todo(struct net_device
*dev
)
2702 spin_lock(&net_todo_list_lock
);
2703 list_add_tail(&dev
->todo_list
, &net_todo_list
);
2704 spin_unlock(&net_todo_list_lock
);
2708 * register_netdevice - register a network device
2709 * @dev: device to register
2711 * Take a completed network device structure and add it to the kernel
2712 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2713 * chain. 0 is returned on success. A negative errno code is returned
2714 * on a failure to set up the device, or if the name is a duplicate.
2716 * Callers must hold the rtnl semaphore. You may want
2717 * register_netdev() instead of this.
2720 * The locking appears insufficient to guarantee two parallel registers
2721 * will not get the same name.
2724 int register_netdevice(struct net_device
*dev
)
2726 struct hlist_head
*head
;
2727 struct hlist_node
*p
;
2730 BUG_ON(dev_boot_phase
);
2733 /* When net_device's are persistent, this will be fatal. */
2734 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
2736 spin_lock_init(&dev
->queue_lock
);
2737 spin_lock_init(&dev
->xmit_lock
);
2738 dev
->xmit_lock_owner
= -1;
2739 #ifdef CONFIG_NET_CLS_ACT
2740 spin_lock_init(&dev
->ingress_lock
);
2743 ret
= alloc_divert_blk(dev
);
2749 /* Init, if this function is available */
2751 ret
= dev
->init(dev
);
2759 if (!dev_valid_name(dev
->name
)) {
2764 dev
->ifindex
= dev_new_index();
2765 if (dev
->iflink
== -1)
2766 dev
->iflink
= dev
->ifindex
;
2768 /* Check for existence of name */
2769 head
= dev_name_hash(dev
->name
);
2770 hlist_for_each(p
, head
) {
2771 struct net_device
*d
2772 = hlist_entry(p
, struct net_device
, name_hlist
);
2773 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
2779 /* Fix illegal SG+CSUM combinations. */
2780 if ((dev
->features
& NETIF_F_SG
) &&
2781 !(dev
->features
& (NETIF_F_IP_CSUM
|
2783 NETIF_F_HW_CSUM
))) {
2784 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2786 dev
->features
&= ~NETIF_F_SG
;
2789 /* TSO requires that SG is present as well. */
2790 if ((dev
->features
& NETIF_F_TSO
) &&
2791 !(dev
->features
& NETIF_F_SG
)) {
2792 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2794 dev
->features
&= ~NETIF_F_TSO
;
2798 * nil rebuild_header routine,
2799 * that should be never called and used as just bug trap.
2802 if (!dev
->rebuild_header
)
2803 dev
->rebuild_header
= default_rebuild_header
;
2806 * Default initial state at registry is that the
2807 * device is present.
2810 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
2813 dev_init_scheduler(dev
);
2814 write_lock_bh(&dev_base_lock
);
2816 dev_tail
= &dev
->next
;
2817 hlist_add_head(&dev
->name_hlist
, head
);
2818 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
2820 dev
->reg_state
= NETREG_REGISTERING
;
2821 write_unlock_bh(&dev_base_lock
);
2823 /* Notify protocols, that a new device appeared. */
2824 notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
2826 /* Finish registration after unlock */
2833 free_divert_blk(dev
);
2838 * register_netdev - register a network device
2839 * @dev: device to register
2841 * Take a completed network device structure and add it to the kernel
2842 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2843 * chain. 0 is returned on success. A negative errno code is returned
2844 * on a failure to set up the device, or if the name is a duplicate.
2846 * This is a wrapper around register_netdev that takes the rtnl semaphore
2847 * and expands the device name if you passed a format string to
2850 int register_netdev(struct net_device
*dev
)
2857 * If the name is a format string the caller wants us to do a
2860 if (strchr(dev
->name
, '%')) {
2861 err
= dev_alloc_name(dev
, dev
->name
);
2867 * Back compatibility hook. Kill this one in 2.5
2869 if (dev
->name
[0] == 0 || dev
->name
[0] == ' ') {
2870 err
= dev_alloc_name(dev
, "eth%d");
2875 err
= register_netdevice(dev
);
2880 EXPORT_SYMBOL(register_netdev
);
2883 * netdev_wait_allrefs - wait until all references are gone.
2885 * This is called when unregistering network devices.
2887 * Any protocol or device that holds a reference should register
2888 * for netdevice notification, and cleanup and put back the
2889 * reference if they receive an UNREGISTER event.
2890 * We can get stuck here if buggy protocols don't correctly
2893 static void netdev_wait_allrefs(struct net_device
*dev
)
2895 unsigned long rebroadcast_time
, warning_time
;
2897 rebroadcast_time
= warning_time
= jiffies
;
2898 while (atomic_read(&dev
->refcnt
) != 0) {
2899 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
2902 /* Rebroadcast unregister notification */
2903 notifier_call_chain(&netdev_chain
,
2904 NETDEV_UNREGISTER
, dev
);
2906 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
2908 /* We must not have linkwatch events
2909 * pending on unregister. If this
2910 * happens, we simply run the queue
2911 * unscheduled, resulting in a noop
2914 linkwatch_run_queue();
2919 rebroadcast_time
= jiffies
;
2924 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
2925 printk(KERN_EMERG
"unregister_netdevice: "
2926 "waiting for %s to become free. Usage "
2928 dev
->name
, atomic_read(&dev
->refcnt
));
2929 warning_time
= jiffies
;
2938 * register_netdevice(x1);
2939 * register_netdevice(x2);
2941 * unregister_netdevice(y1);
2942 * unregister_netdevice(y2);
2948 * We are invoked by rtnl_unlock() after it drops the semaphore.
2949 * This allows us to deal with problems:
2950 * 1) We can create/delete sysfs objects which invoke hotplug
2951 * without deadlocking with linkwatch via keventd.
2952 * 2) Since we run with the RTNL semaphore not held, we can sleep
2953 * safely in order to wait for the netdev refcnt to drop to zero.
2955 static DECLARE_MUTEX(net_todo_run_mutex
);
2956 void netdev_run_todo(void)
2958 struct list_head list
= LIST_HEAD_INIT(list
);
2962 /* Need to guard against multiple cpu's getting out of order. */
2963 down(&net_todo_run_mutex
);
2965 /* Not safe to do outside the semaphore. We must not return
2966 * until all unregister events invoked by the local processor
2967 * have been completed (either by this todo run, or one on
2970 if (list_empty(&net_todo_list
))
2973 /* Snapshot list, allow later requests */
2974 spin_lock(&net_todo_list_lock
);
2975 list_splice_init(&net_todo_list
, &list
);
2976 spin_unlock(&net_todo_list_lock
);
2978 while (!list_empty(&list
)) {
2979 struct net_device
*dev
2980 = list_entry(list
.next
, struct net_device
, todo_list
);
2981 list_del(&dev
->todo_list
);
2983 switch(dev
->reg_state
) {
2984 case NETREG_REGISTERING
:
2985 err
= netdev_register_sysfs(dev
);
2987 printk(KERN_ERR
"%s: failed sysfs registration (%d)\n",
2989 dev
->reg_state
= NETREG_REGISTERED
;
2992 case NETREG_UNREGISTERING
:
2993 netdev_unregister_sysfs(dev
);
2994 dev
->reg_state
= NETREG_UNREGISTERED
;
2996 netdev_wait_allrefs(dev
);
2999 BUG_ON(atomic_read(&dev
->refcnt
));
3000 BUG_TRAP(!dev
->ip_ptr
);
3001 BUG_TRAP(!dev
->ip6_ptr
);
3002 BUG_TRAP(!dev
->dn_ptr
);
3005 /* It must be the very last action,
3006 * after this 'dev' may point to freed up memory.
3008 if (dev
->destructor
)
3009 dev
->destructor(dev
);
3013 printk(KERN_ERR
"network todo '%s' but state %d\n",
3014 dev
->name
, dev
->reg_state
);
3020 up(&net_todo_run_mutex
);
3024 * alloc_netdev - allocate network device
3025 * @sizeof_priv: size of private data to allocate space for
3026 * @name: device name format string
3027 * @setup: callback to initialize device
3029 * Allocates a struct net_device with private data area for driver use
3030 * and performs basic initialization.
3032 struct net_device
*alloc_netdev(int sizeof_priv
, const char *name
,
3033 void (*setup
)(struct net_device
*))
3036 struct net_device
*dev
;
3039 /* ensure 32-byte alignment of both the device and private area */
3040 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
3041 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
3043 p
= kmalloc(alloc_size
, GFP_KERNEL
);
3045 printk(KERN_ERR
"alloc_dev: Unable to allocate device.\n");
3048 memset(p
, 0, alloc_size
);
3050 dev
= (struct net_device
*)
3051 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3052 dev
->padded
= (char *)dev
- (char *)p
;
3055 dev
->priv
= netdev_priv(dev
);
3058 strcpy(dev
->name
, name
);
3061 EXPORT_SYMBOL(alloc_netdev
);
3064 * free_netdev - free network device
3067 * This function does the last stage of destroying an allocated device
3068 * interface. The reference to the device object is released.
3069 * If this is the last reference then it will be freed.
3071 void free_netdev(struct net_device
*dev
)
3074 /* Compatiablity with error handling in drivers */
3075 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3076 kfree((char *)dev
- dev
->padded
);
3080 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3081 dev
->reg_state
= NETREG_RELEASED
;
3083 /* will free via class release */
3084 class_device_put(&dev
->class_dev
);
3086 kfree((char *)dev
- dev
->padded
);
3090 /* Synchronize with packet receive processing. */
3091 void synchronize_net(void)
3098 * unregister_netdevice - remove device from the kernel
3101 * This function shuts down a device interface and removes it
3102 * from the kernel tables. On success 0 is returned, on a failure
3103 * a negative errno code is returned.
3105 * Callers must hold the rtnl semaphore. You may want
3106 * unregister_netdev() instead of this.
3109 int unregister_netdevice(struct net_device
*dev
)
3111 struct net_device
*d
, **dp
;
3113 BUG_ON(dev_boot_phase
);
3116 /* Some devices call without registering for initialization unwind. */
3117 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3118 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3119 "was registered\n", dev
->name
, dev
);
3123 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3125 /* If device is running, close it first. */
3126 if (dev
->flags
& IFF_UP
)
3129 /* And unlink it from device chain. */
3130 for (dp
= &dev_base
; (d
= *dp
) != NULL
; dp
= &d
->next
) {
3132 write_lock_bh(&dev_base_lock
);
3133 hlist_del(&dev
->name_hlist
);
3134 hlist_del(&dev
->index_hlist
);
3135 if (dev_tail
== &dev
->next
)
3138 write_unlock_bh(&dev_base_lock
);
3143 printk(KERN_ERR
"unregister net_device: '%s' not found\n",
3148 dev
->reg_state
= NETREG_UNREGISTERING
;
3152 /* Shutdown queueing discipline. */
3156 /* Notify protocols, that we are about to destroy
3157 this device. They should clean all the things.
3159 notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3162 * Flush the multicast chain
3164 dev_mc_discard(dev
);
3169 /* Notifier chain MUST detach us from master device. */
3170 BUG_TRAP(!dev
->master
);
3172 free_divert_blk(dev
);
3174 /* Finish processing unregister after unlock */
3184 * unregister_netdev - remove device from the kernel
3187 * This function shuts down a device interface and removes it
3188 * from the kernel tables. On success 0 is returned, on a failure
3189 * a negative errno code is returned.
3191 * This is just a wrapper for unregister_netdevice that takes
3192 * the rtnl semaphore. In general you want to use this and not
3193 * unregister_netdevice.
3195 void unregister_netdev(struct net_device
*dev
)
3198 unregister_netdevice(dev
);
3202 EXPORT_SYMBOL(unregister_netdev
);
3204 #ifdef CONFIG_HOTPLUG_CPU
3205 static int dev_cpu_callback(struct notifier_block
*nfb
,
3206 unsigned long action
,
3209 struct sk_buff
**list_skb
;
3210 struct net_device
**list_net
;
3211 struct sk_buff
*skb
;
3212 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3213 struct softnet_data
*sd
, *oldsd
;
3215 if (action
!= CPU_DEAD
)
3218 local_irq_disable();
3219 cpu
= smp_processor_id();
3220 sd
= &per_cpu(softnet_data
, cpu
);
3221 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3223 /* Find end of our completion_queue. */
3224 list_skb
= &sd
->completion_queue
;
3226 list_skb
= &(*list_skb
)->next
;
3227 /* Append completion queue from offline CPU. */
3228 *list_skb
= oldsd
->completion_queue
;
3229 oldsd
->completion_queue
= NULL
;
3231 /* Find end of our output_queue. */
3232 list_net
= &sd
->output_queue
;
3234 list_net
= &(*list_net
)->next_sched
;
3235 /* Append output queue from offline CPU. */
3236 *list_net
= oldsd
->output_queue
;
3237 oldsd
->output_queue
= NULL
;
3239 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3242 /* Process offline CPU's input_pkt_queue */
3243 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3248 #endif /* CONFIG_HOTPLUG_CPU */
3252 * Initialize the DEV module. At boot time this walks the device list and
3253 * unhooks any devices that fail to initialise (normally hardware not
3254 * present) and leaves us with a valid list of present and active devices.
3259 * This is called single threaded during boot, so no need
3260 * to take the rtnl semaphore.
3262 static int __init
net_dev_init(void)
3264 int i
, rc
= -ENOMEM
;
3266 BUG_ON(!dev_boot_phase
);
3270 if (dev_proc_init())
3273 if (netdev_sysfs_init())
3276 INIT_LIST_HEAD(&ptype_all
);
3277 for (i
= 0; i
< 16; i
++)
3278 INIT_LIST_HEAD(&ptype_base
[i
]);
3280 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
3281 INIT_HLIST_HEAD(&dev_name_head
[i
]);
3283 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
3284 INIT_HLIST_HEAD(&dev_index_head
[i
]);
3287 * Initialise the packet receive queues.
3290 for (i
= 0; i
< NR_CPUS
; i
++) {
3291 struct softnet_data
*queue
;
3293 queue
= &per_cpu(softnet_data
, i
);
3294 skb_queue_head_init(&queue
->input_pkt_queue
);
3295 queue
->throttle
= 0;
3296 queue
->cng_level
= 0;
3297 queue
->avg_blog
= 10; /* arbitrary non-zero */
3298 queue
->completion_queue
= NULL
;
3299 INIT_LIST_HEAD(&queue
->poll_list
);
3300 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
3301 queue
->backlog_dev
.weight
= weight_p
;
3302 queue
->backlog_dev
.poll
= process_backlog
;
3303 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
3306 #ifdef OFFLINE_SAMPLE
3307 samp_timer
.expires
= jiffies
+ (10 * HZ
);
3308 add_timer(&samp_timer
);
3313 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
3314 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
3316 hotcpu_notifier(dev_cpu_callback
, 0);
3324 subsys_initcall(net_dev_init
);
3326 EXPORT_SYMBOL(__dev_get_by_index
);
3327 EXPORT_SYMBOL(__dev_get_by_name
);
3328 EXPORT_SYMBOL(__dev_remove_pack
);
3329 EXPORT_SYMBOL(__skb_linearize
);
3330 EXPORT_SYMBOL(dev_add_pack
);
3331 EXPORT_SYMBOL(dev_alloc_name
);
3332 EXPORT_SYMBOL(dev_close
);
3333 EXPORT_SYMBOL(dev_get_by_flags
);
3334 EXPORT_SYMBOL(dev_get_by_index
);
3335 EXPORT_SYMBOL(dev_get_by_name
);
3336 EXPORT_SYMBOL(dev_ioctl
);
3337 EXPORT_SYMBOL(dev_open
);
3338 EXPORT_SYMBOL(dev_queue_xmit
);
3339 EXPORT_SYMBOL(dev_remove_pack
);
3340 EXPORT_SYMBOL(dev_set_allmulti
);
3341 EXPORT_SYMBOL(dev_set_promiscuity
);
3342 EXPORT_SYMBOL(dev_change_flags
);
3343 EXPORT_SYMBOL(dev_set_mtu
);
3344 EXPORT_SYMBOL(dev_set_mac_address
);
3345 EXPORT_SYMBOL(free_netdev
);
3346 EXPORT_SYMBOL(netdev_boot_setup_check
);
3347 EXPORT_SYMBOL(netdev_set_master
);
3348 EXPORT_SYMBOL(netdev_state_change
);
3349 EXPORT_SYMBOL(netif_receive_skb
);
3350 EXPORT_SYMBOL(netif_rx
);
3351 EXPORT_SYMBOL(register_gifconf
);
3352 EXPORT_SYMBOL(register_netdevice
);
3353 EXPORT_SYMBOL(register_netdevice_notifier
);
3354 EXPORT_SYMBOL(skb_checksum_help
);
3355 EXPORT_SYMBOL(synchronize_net
);
3356 EXPORT_SYMBOL(unregister_netdevice
);
3357 EXPORT_SYMBOL(unregister_netdevice_notifier
);
3358 EXPORT_SYMBOL(net_enable_timestamp
);
3359 EXPORT_SYMBOL(net_disable_timestamp
);
3360 EXPORT_SYMBOL(dev_get_flags
);
3362 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3363 EXPORT_SYMBOL(br_handle_frame_hook
);
3364 EXPORT_SYMBOL(br_fdb_get_hook
);
3365 EXPORT_SYMBOL(br_fdb_put_hook
);
3369 EXPORT_SYMBOL(dev_load
);
3372 EXPORT_PER_CPU_SYMBOL(softnet_data
);