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>
119 * The list of packet types we will receive (as opposed to discard)
120 * and the routines to invoke.
122 * Why 16. Because with 16 the only overlap we get on a hash of the
123 * low nibble of the protocol value is RARP/SNAP/X.25.
125 * NOTE: That is no longer true with the addition of VLAN tags. Not
126 * sure which should go first, but I bet it won't make much
127 * difference if we are running VLANs. The good news is that
128 * this protocol won't be in the list unless compiled in, so
129 * the average user (w/out VLANs) will not be adversly affected.
146 static DEFINE_SPINLOCK(ptype_lock
);
147 static struct list_head ptype_base
[16]; /* 16 way hashed list */
148 static struct list_head ptype_all
; /* Taps */
151 * The @dev_base list is protected by @dev_base_lock and the rtln
154 * Pure readers hold dev_base_lock for reading.
156 * Writers must hold the rtnl semaphore while they loop through the
157 * dev_base list, and hold dev_base_lock for writing when they do the
158 * actual updates. This allows pure readers to access the list even
159 * while a writer is preparing to update it.
161 * To put it another way, dev_base_lock is held for writing only to
162 * protect against pure readers; the rtnl semaphore provides the
163 * protection against other writers.
165 * See, for example usages, register_netdevice() and
166 * unregister_netdevice(), which must be called with the rtnl
169 struct net_device
*dev_base
;
170 static struct net_device
**dev_tail
= &dev_base
;
171 DEFINE_RWLOCK(dev_base_lock
);
173 EXPORT_SYMBOL(dev_base
);
174 EXPORT_SYMBOL(dev_base_lock
);
176 #define NETDEV_HASHBITS 8
177 static struct hlist_head dev_name_head
[1<<NETDEV_HASHBITS
];
178 static struct hlist_head dev_index_head
[1<<NETDEV_HASHBITS
];
180 static inline struct hlist_head
*dev_name_hash(const char *name
)
182 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
183 return &dev_name_head
[hash
& ((1<<NETDEV_HASHBITS
)-1)];
186 static inline struct hlist_head
*dev_index_hash(int ifindex
)
188 return &dev_index_head
[ifindex
& ((1<<NETDEV_HASHBITS
)-1)];
195 static struct notifier_block
*netdev_chain
;
198 * Device drivers call our routines to queue packets here. We empty the
199 * queue in the local softnet handler.
201 DEFINE_PER_CPU(struct softnet_data
, softnet_data
) = { NULL
};
204 extern int netdev_sysfs_init(void);
205 extern int netdev_register_sysfs(struct net_device
*);
206 extern void netdev_unregister_sysfs(struct net_device
*);
208 #define netdev_sysfs_init() (0)
209 #define netdev_register_sysfs(dev) (0)
210 #define netdev_unregister_sysfs(dev) do { } while(0)
214 /*******************************************************************************
216 Protocol management and registration routines
218 *******************************************************************************/
227 * Add a protocol ID to the list. Now that the input handler is
228 * smarter we can dispense with all the messy stuff that used to be
231 * BEWARE!!! Protocol handlers, mangling input packets,
232 * MUST BE last in hash buckets and checking protocol handlers
233 * MUST start from promiscuous ptype_all chain in net_bh.
234 * It is true now, do not change it.
235 * Explanation follows: if protocol handler, mangling packet, will
236 * be the first on list, it is not able to sense, that packet
237 * is cloned and should be copied-on-write, so that it will
238 * change it and subsequent readers will get broken packet.
243 * dev_add_pack - add packet handler
244 * @pt: packet type declaration
246 * Add a protocol handler to the networking stack. The passed &packet_type
247 * is linked into kernel lists and may not be freed until it has been
248 * removed from the kernel lists.
250 * This call does not sleep therefore it can not
251 * guarantee all CPU's that are in middle of receiving packets
252 * will see the new packet type (until the next received packet).
255 void dev_add_pack(struct packet_type
*pt
)
259 spin_lock_bh(&ptype_lock
);
260 if (pt
->type
== htons(ETH_P_ALL
)) {
262 list_add_rcu(&pt
->list
, &ptype_all
);
264 hash
= ntohs(pt
->type
) & 15;
265 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
267 spin_unlock_bh(&ptype_lock
);
271 * __dev_remove_pack - remove packet handler
272 * @pt: packet type declaration
274 * Remove a protocol handler that was previously added to the kernel
275 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
276 * from the kernel lists and can be freed or reused once this function
279 * The packet type might still be in use by receivers
280 * and must not be freed until after all the CPU's have gone
281 * through a quiescent state.
283 void __dev_remove_pack(struct packet_type
*pt
)
285 struct list_head
*head
;
286 struct packet_type
*pt1
;
288 spin_lock_bh(&ptype_lock
);
290 if (pt
->type
== htons(ETH_P_ALL
)) {
294 head
= &ptype_base
[ntohs(pt
->type
) & 15];
296 list_for_each_entry(pt1
, head
, list
) {
298 list_del_rcu(&pt
->list
);
303 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
305 spin_unlock_bh(&ptype_lock
);
308 * dev_remove_pack - remove packet handler
309 * @pt: packet type declaration
311 * Remove a protocol handler that was previously added to the kernel
312 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
313 * from the kernel lists and can be freed or reused once this function
316 * This call sleeps to guarantee that no CPU is looking at the packet
319 void dev_remove_pack(struct packet_type
*pt
)
321 __dev_remove_pack(pt
);
326 /******************************************************************************
328 Device Boot-time Settings Routines
330 *******************************************************************************/
332 /* Boot time configuration table */
333 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
336 * netdev_boot_setup_add - add new setup entry
337 * @name: name of the device
338 * @map: configured settings for the device
340 * Adds new setup entry to the dev_boot_setup list. The function
341 * returns 0 on error and 1 on success. This is a generic routine to
344 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
346 struct netdev_boot_setup
*s
;
350 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
351 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
352 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
353 strcpy(s
[i
].name
, name
);
354 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
359 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
363 * netdev_boot_setup_check - check boot time settings
364 * @dev: the netdevice
366 * Check boot time settings for the device.
367 * The found settings are set for the device to be used
368 * later in the device probing.
369 * Returns 0 if no settings found, 1 if they are.
371 int netdev_boot_setup_check(struct net_device
*dev
)
373 struct netdev_boot_setup
*s
= dev_boot_setup
;
376 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
377 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
378 !strncmp(dev
->name
, s
[i
].name
, strlen(s
[i
].name
))) {
379 dev
->irq
= s
[i
].map
.irq
;
380 dev
->base_addr
= s
[i
].map
.base_addr
;
381 dev
->mem_start
= s
[i
].map
.mem_start
;
382 dev
->mem_end
= s
[i
].map
.mem_end
;
391 * netdev_boot_base - get address from boot time settings
392 * @prefix: prefix for network device
393 * @unit: id for network device
395 * Check boot time settings for the base address of device.
396 * The found settings are set for the device to be used
397 * later in the device probing.
398 * Returns 0 if no settings found.
400 unsigned long netdev_boot_base(const char *prefix
, int unit
)
402 const struct netdev_boot_setup
*s
= dev_boot_setup
;
406 sprintf(name
, "%s%d", prefix
, unit
);
409 * If device already registered then return base of 1
410 * to indicate not to probe for this interface
412 if (__dev_get_by_name(name
))
415 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
416 if (!strcmp(name
, s
[i
].name
))
417 return s
[i
].map
.base_addr
;
422 * Saves at boot time configured settings for any netdevice.
424 int __init
netdev_boot_setup(char *str
)
429 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
434 memset(&map
, 0, sizeof(map
));
438 map
.base_addr
= ints
[2];
440 map
.mem_start
= ints
[3];
442 map
.mem_end
= ints
[4];
444 /* Add new entry to the list */
445 return netdev_boot_setup_add(str
, &map
);
448 __setup("netdev=", netdev_boot_setup
);
450 /*******************************************************************************
452 Device Interface Subroutines
454 *******************************************************************************/
457 * __dev_get_by_name - find a device by its name
458 * @name: name to find
460 * Find an interface by name. Must be called under RTNL semaphore
461 * or @dev_base_lock. If the name is found a pointer to the device
462 * is returned. If the name is not found then %NULL is returned. The
463 * reference counters are not incremented so the caller must be
464 * careful with locks.
467 struct net_device
*__dev_get_by_name(const char *name
)
469 struct hlist_node
*p
;
471 hlist_for_each(p
, dev_name_hash(name
)) {
472 struct net_device
*dev
473 = hlist_entry(p
, struct net_device
, name_hlist
);
474 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
481 * dev_get_by_name - find a device by its name
482 * @name: name to find
484 * Find an interface by name. This can be called from any
485 * context and does its own locking. The returned handle has
486 * the usage count incremented and the caller must use dev_put() to
487 * release it when it is no longer needed. %NULL is returned if no
488 * matching device is found.
491 struct net_device
*dev_get_by_name(const char *name
)
493 struct net_device
*dev
;
495 read_lock(&dev_base_lock
);
496 dev
= __dev_get_by_name(name
);
499 read_unlock(&dev_base_lock
);
504 * __dev_get_by_index - find a device by its ifindex
505 * @ifindex: index of device
507 * Search for an interface by index. Returns %NULL if the device
508 * is not found or a pointer to the device. The device has not
509 * had its reference counter increased so the caller must be careful
510 * about locking. The caller must hold either the RTNL semaphore
514 struct net_device
*__dev_get_by_index(int ifindex
)
516 struct hlist_node
*p
;
518 hlist_for_each(p
, dev_index_hash(ifindex
)) {
519 struct net_device
*dev
520 = hlist_entry(p
, struct net_device
, index_hlist
);
521 if (dev
->ifindex
== ifindex
)
529 * dev_get_by_index - find a device by its ifindex
530 * @ifindex: index of device
532 * Search for an interface by index. Returns NULL if the device
533 * is not found or a pointer to the device. The device returned has
534 * had a reference added and the pointer is safe until the user calls
535 * dev_put to indicate they have finished with it.
538 struct net_device
*dev_get_by_index(int ifindex
)
540 struct net_device
*dev
;
542 read_lock(&dev_base_lock
);
543 dev
= __dev_get_by_index(ifindex
);
546 read_unlock(&dev_base_lock
);
551 * dev_getbyhwaddr - find a device by its hardware address
552 * @type: media type of device
553 * @ha: hardware address
555 * Search for an interface by MAC address. Returns NULL if the device
556 * is not found or a pointer to the device. The caller must hold the
557 * rtnl semaphore. The returned device has not had its ref count increased
558 * and the caller must therefore be careful about locking
561 * If the API was consistent this would be __dev_get_by_hwaddr
564 struct net_device
*dev_getbyhwaddr(unsigned short type
, char *ha
)
566 struct net_device
*dev
;
570 for (dev
= dev_base
; dev
; dev
= dev
->next
)
571 if (dev
->type
== type
&&
572 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
577 EXPORT_SYMBOL(dev_getbyhwaddr
);
579 struct net_device
*dev_getfirstbyhwtype(unsigned short type
)
581 struct net_device
*dev
;
584 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
585 if (dev
->type
== type
) {
594 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
597 * dev_get_by_flags - find any device with given flags
598 * @if_flags: IFF_* values
599 * @mask: bitmask of bits in if_flags to check
601 * Search for any interface with the given flags. Returns NULL if a device
602 * is not found or a pointer to the device. The device returned has
603 * had a reference added and the pointer is safe until the user calls
604 * dev_put to indicate they have finished with it.
607 struct net_device
* dev_get_by_flags(unsigned short if_flags
, unsigned short mask
)
609 struct net_device
*dev
;
611 read_lock(&dev_base_lock
);
612 for (dev
= dev_base
; dev
!= NULL
; dev
= dev
->next
) {
613 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
618 read_unlock(&dev_base_lock
);
623 * dev_valid_name - check if name is okay for network device
626 * Network device names need to be valid file names to
627 * to allow sysfs to work
629 int dev_valid_name(const char *name
)
631 return !(*name
== '\0'
632 || !strcmp(name
, ".")
633 || !strcmp(name
, "..")
634 || strchr(name
, '/'));
638 * dev_alloc_name - allocate a name for a device
640 * @name: name format string
642 * Passed a format string - eg "lt%d" it will try and find a suitable
643 * id. Not efficient for many devices, not called a lot. The caller
644 * must hold the dev_base or rtnl lock while allocating the name and
645 * adding the device in order to avoid duplicates. Returns the number
646 * of the unit assigned or a negative errno code.
649 int dev_alloc_name(struct net_device
*dev
, const char *name
)
654 const int max_netdevices
= 8*PAGE_SIZE
;
656 struct net_device
*d
;
658 p
= strnchr(name
, IFNAMSIZ
-1, '%');
661 * Verify the string as this thing may have come from
662 * the user. There must be either one "%d" and no other "%"
665 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
668 /* Use one page as a bit array of possible slots */
669 inuse
= (long *) get_zeroed_page(GFP_ATOMIC
);
673 for (d
= dev_base
; d
; d
= d
->next
) {
674 if (!sscanf(d
->name
, name
, &i
))
676 if (i
< 0 || i
>= max_netdevices
)
679 /* avoid cases where sscanf is not exact inverse of printf */
680 snprintf(buf
, sizeof(buf
), name
, i
);
681 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
685 i
= find_first_zero_bit(inuse
, max_netdevices
);
686 free_page((unsigned long) inuse
);
689 snprintf(buf
, sizeof(buf
), name
, i
);
690 if (!__dev_get_by_name(buf
)) {
691 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
695 /* It is possible to run out of possible slots
696 * when the name is long and there isn't enough space left
697 * for the digits, or if all bits are used.
704 * dev_change_name - change name of a device
706 * @newname: name (or format string) must be at least IFNAMSIZ
708 * Change name of a device, can pass format strings "eth%d".
711 int dev_change_name(struct net_device
*dev
, char *newname
)
717 if (dev
->flags
& IFF_UP
)
720 if (!dev_valid_name(newname
))
723 if (strchr(newname
, '%')) {
724 err
= dev_alloc_name(dev
, newname
);
727 strcpy(newname
, dev
->name
);
729 else if (__dev_get_by_name(newname
))
732 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
734 err
= class_device_rename(&dev
->class_dev
, dev
->name
);
736 hlist_del(&dev
->name_hlist
);
737 hlist_add_head(&dev
->name_hlist
, dev_name_hash(dev
->name
));
738 notifier_call_chain(&netdev_chain
, NETDEV_CHANGENAME
, dev
);
745 * netdev_features_change - device changes fatures
746 * @dev: device to cause notification
748 * Called to indicate a device has changed features.
750 void netdev_features_change(struct net_device
*dev
)
752 notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
754 EXPORT_SYMBOL(netdev_features_change
);
757 * netdev_state_change - device changes state
758 * @dev: device to cause notification
760 * Called to indicate a device has changed state. This function calls
761 * the notifier chains for netdev_chain and sends a NEWLINK message
762 * to the routing socket.
764 void netdev_state_change(struct net_device
*dev
)
766 if (dev
->flags
& IFF_UP
) {
767 notifier_call_chain(&netdev_chain
, NETDEV_CHANGE
, dev
);
768 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
773 * dev_load - load a network module
774 * @name: name of interface
776 * If a network interface is not present and the process has suitable
777 * privileges this function loads the module. If module loading is not
778 * available in this kernel then it becomes a nop.
781 void dev_load(const char *name
)
783 struct net_device
*dev
;
785 read_lock(&dev_base_lock
);
786 dev
= __dev_get_by_name(name
);
787 read_unlock(&dev_base_lock
);
789 if (!dev
&& capable(CAP_SYS_MODULE
))
790 request_module("%s", name
);
793 static int default_rebuild_header(struct sk_buff
*skb
)
795 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
796 skb
->dev
? skb
->dev
->name
: "NULL!!!");
803 * dev_open - prepare an interface for use.
804 * @dev: device to open
806 * Takes a device from down to up state. The device's private open
807 * function is invoked and then the multicast lists are loaded. Finally
808 * the device is moved into the up state and a %NETDEV_UP message is
809 * sent to the netdev notifier chain.
811 * Calling this function on an active interface is a nop. On a failure
812 * a negative errno code is returned.
814 int dev_open(struct net_device
*dev
)
822 if (dev
->flags
& IFF_UP
)
826 * Is it even present?
828 if (!netif_device_present(dev
))
832 * Call device private open method
834 set_bit(__LINK_STATE_START
, &dev
->state
);
836 ret
= dev
->open(dev
);
838 clear_bit(__LINK_STATE_START
, &dev
->state
);
842 * If it went open OK then:
849 dev
->flags
|= IFF_UP
;
852 * Initialize multicasting status
857 * Wakeup transmit queue engine
862 * ... and announce new interface.
864 notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
870 * dev_close - shutdown an interface.
871 * @dev: device to shutdown
873 * This function moves an active device into down state. A
874 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
875 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
878 int dev_close(struct net_device
*dev
)
880 if (!(dev
->flags
& IFF_UP
))
884 * Tell people we are going down, so that they can
885 * prepare to death, when device is still operating.
887 notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
891 clear_bit(__LINK_STATE_START
, &dev
->state
);
893 /* Synchronize to scheduled poll. We cannot touch poll list,
894 * it can be even on different cpu. So just clear netif_running(),
895 * and wait when poll really will happen. Actually, the best place
896 * for this is inside dev->stop() after device stopped its irq
897 * engine, but this requires more changes in devices. */
899 smp_mb__after_clear_bit(); /* Commit netif_running(). */
900 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
906 * Call the device specific close. This cannot fail.
907 * Only if device is UP
909 * We allow it to be called even after a DETACH hot-plug
916 * Device is now down.
919 dev
->flags
&= ~IFF_UP
;
922 * Tell people we are down
924 notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
931 * Device change register/unregister. These are not inline or static
932 * as we export them to the world.
936 * register_netdevice_notifier - register a network notifier block
939 * Register a notifier to be called when network device events occur.
940 * The notifier passed is linked into the kernel structures and must
941 * not be reused until it has been unregistered. A negative errno code
942 * is returned on a failure.
944 * When registered all registration and up events are replayed
945 * to the new notifier to allow device to have a race free
946 * view of the network device list.
949 int register_netdevice_notifier(struct notifier_block
*nb
)
951 struct net_device
*dev
;
955 err
= notifier_chain_register(&netdev_chain
, nb
);
957 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
958 nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
960 if (dev
->flags
& IFF_UP
)
961 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
969 * unregister_netdevice_notifier - unregister a network notifier block
972 * Unregister a notifier previously registered by
973 * register_netdevice_notifier(). The notifier is unlinked into the
974 * kernel structures and may then be reused. A negative errno code
975 * is returned on a failure.
978 int unregister_netdevice_notifier(struct notifier_block
*nb
)
980 return notifier_chain_unregister(&netdev_chain
, nb
);
984 * call_netdevice_notifiers - call all network notifier blocks
985 * @val: value passed unmodified to notifier function
986 * @v: pointer passed unmodified to notifier function
988 * Call all network notifier blocks. Parameters and return value
989 * are as for notifier_call_chain().
992 int call_netdevice_notifiers(unsigned long val
, void *v
)
994 return notifier_call_chain(&netdev_chain
, val
, v
);
997 /* When > 0 there are consumers of rx skb time stamps */
998 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1000 void net_enable_timestamp(void)
1002 atomic_inc(&netstamp_needed
);
1005 void net_disable_timestamp(void)
1007 atomic_dec(&netstamp_needed
);
1010 void __net_timestamp(struct sk_buff
*skb
)
1014 do_gettimeofday(&tv
);
1015 skb_set_timestamp(skb
, &tv
);
1017 EXPORT_SYMBOL(__net_timestamp
);
1019 static inline void net_timestamp(struct sk_buff
*skb
)
1021 if (atomic_read(&netstamp_needed
))
1022 __net_timestamp(skb
);
1024 skb
->tstamp
.off_sec
= 0;
1025 skb
->tstamp
.off_usec
= 0;
1030 * Support routine. Sends outgoing frames to any network
1031 * taps currently in use.
1034 void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1036 struct packet_type
*ptype
;
1041 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1042 /* Never send packets back to the socket
1043 * they originated from - MvS (miquels@drinkel.ow.org)
1045 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1046 (ptype
->af_packet_priv
== NULL
||
1047 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1048 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1052 /* skb->nh should be correctly
1053 set by sender, so that the second statement is
1054 just protection against buggy protocols.
1056 skb2
->mac
.raw
= skb2
->data
;
1058 if (skb2
->nh
.raw
< skb2
->data
||
1059 skb2
->nh
.raw
> skb2
->tail
) {
1060 if (net_ratelimit())
1061 printk(KERN_CRIT
"protocol %04x is "
1063 skb2
->protocol
, dev
->name
);
1064 skb2
->nh
.raw
= skb2
->data
;
1067 skb2
->h
.raw
= skb2
->nh
.raw
;
1068 skb2
->pkt_type
= PACKET_OUTGOING
;
1069 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1076 * Invalidate hardware checksum when packet is to be mangled, and
1077 * complete checksum manually on outgoing path.
1079 int skb_checksum_help(struct sk_buff
*skb
, int inward
)
1082 int ret
= 0, offset
= skb
->h
.raw
- skb
->data
;
1085 skb
->ip_summed
= CHECKSUM_NONE
;
1089 if (skb_cloned(skb
)) {
1090 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1095 if (offset
> (int)skb
->len
)
1097 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1099 offset
= skb
->tail
- skb
->h
.raw
;
1102 if (skb
->csum
+ 2 > offset
)
1105 *(u16
*)(skb
->h
.raw
+ skb
->csum
) = csum_fold(csum
);
1106 skb
->ip_summed
= CHECKSUM_NONE
;
1111 /* Take action when hardware reception checksum errors are detected. */
1113 void netdev_rx_csum_fault(struct net_device
*dev
)
1115 if (net_ratelimit()) {
1116 printk(KERN_ERR
"%s: hw csum failure.\n",
1117 dev
? dev
->name
: "<unknown>");
1121 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1124 #ifdef CONFIG_HIGHMEM
1125 /* Actually, we should eliminate this check as soon as we know, that:
1126 * 1. IOMMU is present and allows to map all the memory.
1127 * 2. No high memory really exists on this machine.
1130 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1134 if (dev
->features
& NETIF_F_HIGHDMA
)
1137 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1138 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1144 #define illegal_highdma(dev, skb) (0)
1147 /* Keep head the same: replace data */
1148 int __skb_linearize(struct sk_buff
*skb
, gfp_t gfp_mask
)
1153 struct skb_shared_info
*ninfo
;
1154 int headerlen
= skb
->data
- skb
->head
;
1155 int expand
= (skb
->tail
+ skb
->data_len
) - skb
->end
;
1157 if (skb_shared(skb
))
1163 size
= skb
->end
- skb
->head
+ expand
;
1164 size
= SKB_DATA_ALIGN(size
);
1165 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
1169 /* Copy entire thing */
1170 if (skb_copy_bits(skb
, -headerlen
, data
, headerlen
+ skb
->len
))
1174 ninfo
= (struct skb_shared_info
*)(data
+ size
);
1175 atomic_set(&ninfo
->dataref
, 1);
1176 ninfo
->tso_size
= skb_shinfo(skb
)->tso_size
;
1177 ninfo
->tso_segs
= skb_shinfo(skb
)->tso_segs
;
1178 ninfo
->nr_frags
= 0;
1179 ninfo
->frag_list
= NULL
;
1181 /* Offset between the two in bytes */
1182 offset
= data
- skb
->head
;
1184 /* Free old data. */
1185 skb_release_data(skb
);
1188 skb
->end
= data
+ size
;
1190 /* Set up new pointers */
1191 skb
->h
.raw
+= offset
;
1192 skb
->nh
.raw
+= offset
;
1193 skb
->mac
.raw
+= offset
;
1194 skb
->tail
+= offset
;
1195 skb
->data
+= offset
;
1197 /* We are no longer a clone, even if we were. */
1200 skb
->tail
+= skb
->data_len
;
1205 #define HARD_TX_LOCK(dev, cpu) { \
1206 if ((dev->features & NETIF_F_LLTX) == 0) { \
1207 spin_lock(&dev->xmit_lock); \
1208 dev->xmit_lock_owner = cpu; \
1212 #define HARD_TX_UNLOCK(dev) { \
1213 if ((dev->features & NETIF_F_LLTX) == 0) { \
1214 dev->xmit_lock_owner = -1; \
1215 spin_unlock(&dev->xmit_lock); \
1220 * dev_queue_xmit - transmit a buffer
1221 * @skb: buffer to transmit
1223 * Queue a buffer for transmission to a network device. The caller must
1224 * have set the device and priority and built the buffer before calling
1225 * this function. The function can be called from an interrupt.
1227 * A negative errno code is returned on a failure. A success does not
1228 * guarantee the frame will be transmitted as it may be dropped due
1229 * to congestion or traffic shaping.
1231 * -----------------------------------------------------------------------------------
1232 * I notice this method can also return errors from the queue disciplines,
1233 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1236 * Regardless of the return value, the skb is consumed, so it is currently
1237 * difficult to retry a send to this method. (You can bump the ref count
1238 * before sending to hold a reference for retry if you are careful.)
1240 * When calling this method, interrupts MUST be enabled. This is because
1241 * the BH enable code must have IRQs enabled so that it will not deadlock.
1245 int dev_queue_xmit(struct sk_buff
*skb
)
1247 struct net_device
*dev
= skb
->dev
;
1251 if (skb_shinfo(skb
)->frag_list
&&
1252 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1253 __skb_linearize(skb
, GFP_ATOMIC
))
1256 /* Fragmented skb is linearized if device does not support SG,
1257 * or if at least one of fragments is in highmem and device
1258 * does not support DMA from it.
1260 if (skb_shinfo(skb
)->nr_frags
&&
1261 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1262 __skb_linearize(skb
, GFP_ATOMIC
))
1265 /* If packet is not checksummed and device does not support
1266 * checksumming for this protocol, complete checksumming here.
1268 if (skb
->ip_summed
== CHECKSUM_HW
&&
1269 (!(dev
->features
& (NETIF_F_HW_CSUM
| NETIF_F_NO_CSUM
)) &&
1270 (!(dev
->features
& NETIF_F_IP_CSUM
) ||
1271 skb
->protocol
!= htons(ETH_P_IP
))))
1272 if (skb_checksum_help(skb
, 0))
1275 spin_lock_prefetch(&dev
->queue_lock
);
1277 /* Disable soft irqs for various locks below. Also
1278 * stops preemption for RCU.
1282 /* Updates of qdisc are serialized by queue_lock.
1283 * The struct Qdisc which is pointed to by qdisc is now a
1284 * rcu structure - it may be accessed without acquiring
1285 * a lock (but the structure may be stale.) The freeing of the
1286 * qdisc will be deferred until it's known that there are no
1287 * more references to it.
1289 * If the qdisc has an enqueue function, we still need to
1290 * hold the queue_lock before calling it, since queue_lock
1291 * also serializes access to the device queue.
1294 q
= rcu_dereference(dev
->qdisc
);
1295 #ifdef CONFIG_NET_CLS_ACT
1296 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1299 /* Grab device queue */
1300 spin_lock(&dev
->queue_lock
);
1302 rc
= q
->enqueue(skb
, q
);
1306 spin_unlock(&dev
->queue_lock
);
1307 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1311 /* The device has no queue. Common case for software devices:
1312 loopback, all the sorts of tunnels...
1314 Really, it is unlikely that xmit_lock protection is necessary here.
1315 (f.e. loopback and IP tunnels are clean ignoring statistics
1317 However, it is possible, that they rely on protection
1320 Check this and shot the lock. It is not prone from deadlocks.
1321 Either shot noqueue qdisc, it is even simpler 8)
1323 if (dev
->flags
& IFF_UP
) {
1324 int cpu
= smp_processor_id(); /* ok because BHs are off */
1326 if (dev
->xmit_lock_owner
!= cpu
) {
1328 HARD_TX_LOCK(dev
, cpu
);
1330 if (!netif_queue_stopped(dev
)) {
1332 dev_queue_xmit_nit(skb
, dev
);
1335 if (!dev
->hard_start_xmit(skb
, dev
)) {
1336 HARD_TX_UNLOCK(dev
);
1340 HARD_TX_UNLOCK(dev
);
1341 if (net_ratelimit())
1342 printk(KERN_CRIT
"Virtual device %s asks to "
1343 "queue packet!\n", dev
->name
);
1345 /* Recursion is detected! It is possible,
1347 if (net_ratelimit())
1348 printk(KERN_CRIT
"Dead loop on virtual device "
1349 "%s, fix it urgently!\n", dev
->name
);
1365 /*=======================================================================
1367 =======================================================================*/
1369 int netdev_max_backlog
= 1000;
1370 int netdev_budget
= 300;
1371 int weight_p
= 64; /* old backlog weight */
1373 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1377 * netif_rx - post buffer to the network code
1378 * @skb: buffer to post
1380 * This function receives a packet from a device driver and queues it for
1381 * the upper (protocol) levels to process. It always succeeds. The buffer
1382 * may be dropped during processing for congestion control or by the
1386 * NET_RX_SUCCESS (no congestion)
1387 * NET_RX_CN_LOW (low congestion)
1388 * NET_RX_CN_MOD (moderate congestion)
1389 * NET_RX_CN_HIGH (high congestion)
1390 * NET_RX_DROP (packet was dropped)
1394 int netif_rx(struct sk_buff
*skb
)
1396 struct softnet_data
*queue
;
1397 unsigned long flags
;
1399 /* if netpoll wants it, pretend we never saw it */
1400 if (netpoll_rx(skb
))
1403 if (!skb
->tstamp
.off_sec
)
1407 * The code is rearranged so that the path is the most
1408 * short when CPU is congested, but is still operating.
1410 local_irq_save(flags
);
1411 queue
= &__get_cpu_var(softnet_data
);
1413 __get_cpu_var(netdev_rx_stat
).total
++;
1414 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1415 if (queue
->input_pkt_queue
.qlen
) {
1418 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1419 local_irq_restore(flags
);
1420 return NET_RX_SUCCESS
;
1423 netif_rx_schedule(&queue
->backlog_dev
);
1427 __get_cpu_var(netdev_rx_stat
).dropped
++;
1428 local_irq_restore(flags
);
1434 int netif_rx_ni(struct sk_buff
*skb
)
1439 err
= netif_rx(skb
);
1440 if (local_softirq_pending())
1447 EXPORT_SYMBOL(netif_rx_ni
);
1449 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1451 struct net_device
*dev
= skb
->dev
;
1454 skb
->dev
= dev
->master
;
1459 static void net_tx_action(struct softirq_action
*h
)
1461 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1463 if (sd
->completion_queue
) {
1464 struct sk_buff
*clist
;
1466 local_irq_disable();
1467 clist
= sd
->completion_queue
;
1468 sd
->completion_queue
= NULL
;
1472 struct sk_buff
*skb
= clist
;
1473 clist
= clist
->next
;
1475 BUG_TRAP(!atomic_read(&skb
->users
));
1480 if (sd
->output_queue
) {
1481 struct net_device
*head
;
1483 local_irq_disable();
1484 head
= sd
->output_queue
;
1485 sd
->output_queue
= NULL
;
1489 struct net_device
*dev
= head
;
1490 head
= head
->next_sched
;
1492 smp_mb__before_clear_bit();
1493 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1495 if (spin_trylock(&dev
->queue_lock
)) {
1497 spin_unlock(&dev
->queue_lock
);
1499 netif_schedule(dev
);
1505 static __inline__
int deliver_skb(struct sk_buff
*skb
,
1506 struct packet_type
*pt_prev
,
1507 struct net_device
*orig_dev
)
1509 atomic_inc(&skb
->users
);
1510 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1513 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1514 int (*br_handle_frame_hook
)(struct net_bridge_port
*p
, struct sk_buff
**pskb
);
1516 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1517 unsigned char *addr
);
1518 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
);
1520 static __inline__
int handle_bridge(struct sk_buff
**pskb
,
1521 struct packet_type
**pt_prev
, int *ret
,
1522 struct net_device
*orig_dev
)
1524 struct net_bridge_port
*port
;
1526 if ((*pskb
)->pkt_type
== PACKET_LOOPBACK
||
1527 (port
= rcu_dereference((*pskb
)->dev
->br_port
)) == NULL
)
1531 *ret
= deliver_skb(*pskb
, *pt_prev
, orig_dev
);
1535 return br_handle_frame_hook(port
, pskb
);
1538 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1541 #ifdef CONFIG_NET_CLS_ACT
1542 /* TODO: Maybe we should just force sch_ingress to be compiled in
1543 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1544 * a compare and 2 stores extra right now if we dont have it on
1545 * but have CONFIG_NET_CLS_ACT
1546 * NOTE: This doesnt stop any functionality; if you dont have
1547 * the ingress scheduler, you just cant add policies on ingress.
1550 static int ing_filter(struct sk_buff
*skb
)
1553 struct net_device
*dev
= skb
->dev
;
1554 int result
= TC_ACT_OK
;
1556 if (dev
->qdisc_ingress
) {
1557 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1558 if (MAX_RED_LOOP
< ttl
++) {
1559 printk("Redir loop detected Dropping packet (%s->%s)\n",
1560 skb
->input_dev
->name
, skb
->dev
->name
);
1564 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1566 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1568 spin_lock(&dev
->ingress_lock
);
1569 if ((q
= dev
->qdisc_ingress
) != NULL
)
1570 result
= q
->enqueue(skb
, q
);
1571 spin_unlock(&dev
->ingress_lock
);
1579 int netif_receive_skb(struct sk_buff
*skb
)
1581 struct packet_type
*ptype
, *pt_prev
;
1582 struct net_device
*orig_dev
;
1583 int ret
= NET_RX_DROP
;
1584 unsigned short type
;
1586 /* if we've gotten here through NAPI, check netpoll */
1587 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1590 if (!skb
->tstamp
.off_sec
)
1593 if (!skb
->input_dev
)
1594 skb
->input_dev
= skb
->dev
;
1596 orig_dev
= skb_bond(skb
);
1598 __get_cpu_var(netdev_rx_stat
).total
++;
1600 skb
->h
.raw
= skb
->nh
.raw
= skb
->data
;
1601 skb
->mac_len
= skb
->nh
.raw
- skb
->mac
.raw
;
1607 #ifdef CONFIG_NET_CLS_ACT
1608 if (skb
->tc_verd
& TC_NCLS
) {
1609 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1614 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1615 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1617 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1622 #ifdef CONFIG_NET_CLS_ACT
1624 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1625 pt_prev
= NULL
; /* noone else should process this after*/
1627 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1630 ret
= ing_filter(skb
);
1632 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
1641 handle_diverter(skb
);
1643 if (handle_bridge(&skb
, &pt_prev
, &ret
, orig_dev
))
1646 type
= skb
->protocol
;
1647 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
1648 if (ptype
->type
== type
&&
1649 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
1651 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1657 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1660 /* Jamal, now you will not able to escape explaining
1661 * me how you were going to use this. :-)
1671 static int process_backlog(struct net_device
*backlog_dev
, int *budget
)
1674 int quota
= min(backlog_dev
->quota
, *budget
);
1675 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1676 unsigned long start_time
= jiffies
;
1678 backlog_dev
->weight
= weight_p
;
1680 struct sk_buff
*skb
;
1681 struct net_device
*dev
;
1683 local_irq_disable();
1684 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
1691 netif_receive_skb(skb
);
1697 if (work
>= quota
|| jiffies
- start_time
> 1)
1702 backlog_dev
->quota
-= work
;
1707 backlog_dev
->quota
-= work
;
1710 list_del(&backlog_dev
->poll_list
);
1711 smp_mb__before_clear_bit();
1712 netif_poll_enable(backlog_dev
);
1718 static void net_rx_action(struct softirq_action
*h
)
1720 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1721 unsigned long start_time
= jiffies
;
1722 int budget
= netdev_budget
;
1725 local_irq_disable();
1727 while (!list_empty(&queue
->poll_list
)) {
1728 struct net_device
*dev
;
1730 if (budget
<= 0 || jiffies
- start_time
> 1)
1735 dev
= list_entry(queue
->poll_list
.next
,
1736 struct net_device
, poll_list
);
1737 have
= netpoll_poll_lock(dev
);
1739 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
1740 netpoll_poll_unlock(have
);
1741 local_irq_disable();
1742 list_del(&dev
->poll_list
);
1743 list_add_tail(&dev
->poll_list
, &queue
->poll_list
);
1745 dev
->quota
+= dev
->weight
;
1747 dev
->quota
= dev
->weight
;
1749 netpoll_poll_unlock(have
);
1751 local_irq_disable();
1759 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
1760 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1764 static gifconf_func_t
* gifconf_list
[NPROTO
];
1767 * register_gifconf - register a SIOCGIF handler
1768 * @family: Address family
1769 * @gifconf: Function handler
1771 * Register protocol dependent address dumping routines. The handler
1772 * that is passed must not be freed or reused until it has been replaced
1773 * by another handler.
1775 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
1777 if (family
>= NPROTO
)
1779 gifconf_list
[family
] = gifconf
;
1785 * Map an interface index to its name (SIOCGIFNAME)
1789 * We need this ioctl for efficient implementation of the
1790 * if_indextoname() function required by the IPv6 API. Without
1791 * it, we would have to search all the interfaces to find a
1795 static int dev_ifname(struct ifreq __user
*arg
)
1797 struct net_device
*dev
;
1801 * Fetch the caller's info block.
1804 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
1807 read_lock(&dev_base_lock
);
1808 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
1810 read_unlock(&dev_base_lock
);
1814 strcpy(ifr
.ifr_name
, dev
->name
);
1815 read_unlock(&dev_base_lock
);
1817 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
1823 * Perform a SIOCGIFCONF call. This structure will change
1824 * size eventually, and there is nothing I can do about it.
1825 * Thus we will need a 'compatibility mode'.
1828 static int dev_ifconf(char __user
*arg
)
1831 struct net_device
*dev
;
1838 * Fetch the caller's info block.
1841 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
1848 * Loop over the interfaces, and write an info block for each.
1852 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
1853 for (i
= 0; i
< NPROTO
; i
++) {
1854 if (gifconf_list
[i
]) {
1857 done
= gifconf_list
[i
](dev
, NULL
, 0);
1859 done
= gifconf_list
[i
](dev
, pos
+ total
,
1869 * All done. Write the updated control block back to the caller.
1871 ifc
.ifc_len
= total
;
1874 * Both BSD and Solaris return 0 here, so we do too.
1876 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
1879 #ifdef CONFIG_PROC_FS
1881 * This is invoked by the /proc filesystem handler to display a device
1884 static __inline__
struct net_device
*dev_get_idx(loff_t pos
)
1886 struct net_device
*dev
;
1889 for (i
= 0, dev
= dev_base
; dev
&& i
< pos
; ++i
, dev
= dev
->next
);
1891 return i
== pos
? dev
: NULL
;
1894 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1896 read_lock(&dev_base_lock
);
1897 return *pos
? dev_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
1900 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1903 return v
== SEQ_START_TOKEN
? dev_base
: ((struct net_device
*)v
)->next
;
1906 void dev_seq_stop(struct seq_file
*seq
, void *v
)
1908 read_unlock(&dev_base_lock
);
1911 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
1913 if (dev
->get_stats
) {
1914 struct net_device_stats
*stats
= dev
->get_stats(dev
);
1916 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1917 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1918 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
1920 stats
->rx_dropped
+ stats
->rx_missed_errors
,
1921 stats
->rx_fifo_errors
,
1922 stats
->rx_length_errors
+ stats
->rx_over_errors
+
1923 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
1924 stats
->rx_compressed
, stats
->multicast
,
1925 stats
->tx_bytes
, stats
->tx_packets
,
1926 stats
->tx_errors
, stats
->tx_dropped
,
1927 stats
->tx_fifo_errors
, stats
->collisions
,
1928 stats
->tx_carrier_errors
+
1929 stats
->tx_aborted_errors
+
1930 stats
->tx_window_errors
+
1931 stats
->tx_heartbeat_errors
,
1932 stats
->tx_compressed
);
1934 seq_printf(seq
, "%6s: No statistics available.\n", dev
->name
);
1938 * Called from the PROCfs module. This now uses the new arbitrary sized
1939 * /proc/net interface to create /proc/net/dev
1941 static int dev_seq_show(struct seq_file
*seq
, void *v
)
1943 if (v
== SEQ_START_TOKEN
)
1944 seq_puts(seq
, "Inter-| Receive "
1946 " face |bytes packets errs drop fifo frame "
1947 "compressed multicast|bytes packets errs "
1948 "drop fifo colls carrier compressed\n");
1950 dev_seq_printf_stats(seq
, v
);
1954 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
1956 struct netif_rx_stats
*rc
= NULL
;
1958 while (*pos
< NR_CPUS
)
1959 if (cpu_online(*pos
)) {
1960 rc
= &per_cpu(netdev_rx_stat
, *pos
);
1967 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1969 return softnet_get_online(pos
);
1972 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1975 return softnet_get_online(pos
);
1978 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
1982 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
1984 struct netif_rx_stats
*s
= v
;
1986 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
1987 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
1988 0, 0, 0, 0, /* was fastroute */
1993 static struct seq_operations dev_seq_ops
= {
1994 .start
= dev_seq_start
,
1995 .next
= dev_seq_next
,
1996 .stop
= dev_seq_stop
,
1997 .show
= dev_seq_show
,
2000 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2002 return seq_open(file
, &dev_seq_ops
);
2005 static struct file_operations dev_seq_fops
= {
2006 .owner
= THIS_MODULE
,
2007 .open
= dev_seq_open
,
2009 .llseek
= seq_lseek
,
2010 .release
= seq_release
,
2013 static struct seq_operations softnet_seq_ops
= {
2014 .start
= softnet_seq_start
,
2015 .next
= softnet_seq_next
,
2016 .stop
= softnet_seq_stop
,
2017 .show
= softnet_seq_show
,
2020 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2022 return seq_open(file
, &softnet_seq_ops
);
2025 static struct file_operations softnet_seq_fops
= {
2026 .owner
= THIS_MODULE
,
2027 .open
= softnet_seq_open
,
2029 .llseek
= seq_lseek
,
2030 .release
= seq_release
,
2034 extern int wireless_proc_init(void);
2036 #define wireless_proc_init() 0
2039 static int __init
dev_proc_init(void)
2043 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2045 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2047 if (wireless_proc_init())
2053 proc_net_remove("softnet_stat");
2055 proc_net_remove("dev");
2059 #define dev_proc_init() 0
2060 #endif /* CONFIG_PROC_FS */
2064 * netdev_set_master - set up master/slave pair
2065 * @slave: slave device
2066 * @master: new master device
2068 * Changes the master device of the slave. Pass %NULL to break the
2069 * bonding. The caller must hold the RTNL semaphore. On a failure
2070 * a negative errno code is returned. On success the reference counts
2071 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2072 * function returns zero.
2074 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2076 struct net_device
*old
= slave
->master
;
2086 slave
->master
= master
;
2094 slave
->flags
|= IFF_SLAVE
;
2096 slave
->flags
&= ~IFF_SLAVE
;
2098 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2103 * dev_set_promiscuity - update promiscuity count on a device
2107 * Add or remove promsicuity from a device. While the count in the device
2108 * remains above zero the interface remains promiscuous. Once it hits zero
2109 * the device reverts back to normal filtering operation. A negative inc
2110 * value is used to drop promiscuity on the device.
2112 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2114 unsigned short old_flags
= dev
->flags
;
2116 if ((dev
->promiscuity
+= inc
) == 0)
2117 dev
->flags
&= ~IFF_PROMISC
;
2119 dev
->flags
|= IFF_PROMISC
;
2120 if (dev
->flags
!= old_flags
) {
2122 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2123 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2129 * dev_set_allmulti - update allmulti count on a device
2133 * Add or remove reception of all multicast frames to a device. While the
2134 * count in the device remains above zero the interface remains listening
2135 * to all interfaces. Once it hits zero the device reverts back to normal
2136 * filtering operation. A negative @inc value is used to drop the counter
2137 * when releasing a resource needing all multicasts.
2140 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2142 unsigned short old_flags
= dev
->flags
;
2144 dev
->flags
|= IFF_ALLMULTI
;
2145 if ((dev
->allmulti
+= inc
) == 0)
2146 dev
->flags
&= ~IFF_ALLMULTI
;
2147 if (dev
->flags
^ old_flags
)
2151 unsigned dev_get_flags(const struct net_device
*dev
)
2155 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2158 (dev
->gflags
& (IFF_PROMISC
|
2161 if (netif_running(dev
) && netif_carrier_ok(dev
))
2162 flags
|= IFF_RUNNING
;
2167 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2170 int old_flags
= dev
->flags
;
2173 * Set the flags on our device.
2176 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2177 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2179 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2183 * Load in the correct multicast list now the flags have changed.
2189 * Have we downed the interface. We handle IFF_UP ourselves
2190 * according to user attempts to set it, rather than blindly
2195 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2196 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2202 if (dev
->flags
& IFF_UP
&&
2203 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2205 notifier_call_chain(&netdev_chain
, NETDEV_CHANGE
, dev
);
2207 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2208 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2209 dev
->gflags
^= IFF_PROMISC
;
2210 dev_set_promiscuity(dev
, inc
);
2213 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2214 is important. Some (broken) drivers set IFF_PROMISC, when
2215 IFF_ALLMULTI is requested not asking us and not reporting.
2217 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2218 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2219 dev
->gflags
^= IFF_ALLMULTI
;
2220 dev_set_allmulti(dev
, inc
);
2223 if (old_flags
^ dev
->flags
)
2224 rtmsg_ifinfo(RTM_NEWLINK
, dev
, old_flags
^ dev
->flags
);
2229 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2233 if (new_mtu
== dev
->mtu
)
2236 /* MTU must be positive. */
2240 if (!netif_device_present(dev
))
2244 if (dev
->change_mtu
)
2245 err
= dev
->change_mtu(dev
, new_mtu
);
2248 if (!err
&& dev
->flags
& IFF_UP
)
2249 notifier_call_chain(&netdev_chain
,
2250 NETDEV_CHANGEMTU
, dev
);
2254 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2258 if (!dev
->set_mac_address
)
2260 if (sa
->sa_family
!= dev
->type
)
2262 if (!netif_device_present(dev
))
2264 err
= dev
->set_mac_address(dev
, sa
);
2266 notifier_call_chain(&netdev_chain
, NETDEV_CHANGEADDR
, dev
);
2271 * Perform the SIOCxIFxxx calls.
2273 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2276 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2282 case SIOCGIFFLAGS
: /* Get interface flags */
2283 ifr
->ifr_flags
= dev_get_flags(dev
);
2286 case SIOCSIFFLAGS
: /* Set interface flags */
2287 return dev_change_flags(dev
, ifr
->ifr_flags
);
2289 case SIOCGIFMETRIC
: /* Get the metric on the interface
2290 (currently unused) */
2291 ifr
->ifr_metric
= 0;
2294 case SIOCSIFMETRIC
: /* Set the metric on the interface
2295 (currently unused) */
2298 case SIOCGIFMTU
: /* Get the MTU of a device */
2299 ifr
->ifr_mtu
= dev
->mtu
;
2302 case SIOCSIFMTU
: /* Set the MTU of a device */
2303 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
2307 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
2309 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
2310 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2311 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
2315 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
2317 case SIOCSIFHWBROADCAST
:
2318 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
2320 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
2321 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2322 notifier_call_chain(&netdev_chain
,
2323 NETDEV_CHANGEADDR
, dev
);
2327 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
2328 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
2329 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
2330 ifr
->ifr_map
.irq
= dev
->irq
;
2331 ifr
->ifr_map
.dma
= dev
->dma
;
2332 ifr
->ifr_map
.port
= dev
->if_port
;
2336 if (dev
->set_config
) {
2337 if (!netif_device_present(dev
))
2339 return dev
->set_config(dev
, &ifr
->ifr_map
);
2344 if (!dev
->set_multicast_list
||
2345 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2347 if (!netif_device_present(dev
))
2349 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
2353 if (!dev
->set_multicast_list
||
2354 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2356 if (!netif_device_present(dev
))
2358 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
2362 ifr
->ifr_ifindex
= dev
->ifindex
;
2366 ifr
->ifr_qlen
= dev
->tx_queue_len
;
2370 if (ifr
->ifr_qlen
< 0)
2372 dev
->tx_queue_len
= ifr
->ifr_qlen
;
2376 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
2377 return dev_change_name(dev
, ifr
->ifr_newname
);
2380 * Unknown or private ioctl
2384 if ((cmd
>= SIOCDEVPRIVATE
&&
2385 cmd
<= SIOCDEVPRIVATE
+ 15) ||
2386 cmd
== SIOCBONDENSLAVE
||
2387 cmd
== SIOCBONDRELEASE
||
2388 cmd
== SIOCBONDSETHWADDR
||
2389 cmd
== SIOCBONDSLAVEINFOQUERY
||
2390 cmd
== SIOCBONDINFOQUERY
||
2391 cmd
== SIOCBONDCHANGEACTIVE
||
2392 cmd
== SIOCGMIIPHY
||
2393 cmd
== SIOCGMIIREG
||
2394 cmd
== SIOCSMIIREG
||
2395 cmd
== SIOCBRADDIF
||
2396 cmd
== SIOCBRDELIF
||
2397 cmd
== SIOCWANDEV
) {
2399 if (dev
->do_ioctl
) {
2400 if (netif_device_present(dev
))
2401 err
= dev
->do_ioctl(dev
, ifr
,
2414 * This function handles all "interface"-type I/O control requests. The actual
2415 * 'doing' part of this is dev_ifsioc above.
2419 * dev_ioctl - network device ioctl
2420 * @cmd: command to issue
2421 * @arg: pointer to a struct ifreq in user space
2423 * Issue ioctl functions to devices. This is normally called by the
2424 * user space syscall interfaces but can sometimes be useful for
2425 * other purposes. The return value is the return from the syscall if
2426 * positive or a negative errno code on error.
2429 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
2435 /* One special case: SIOCGIFCONF takes ifconf argument
2436 and requires shared lock, because it sleeps writing
2440 if (cmd
== SIOCGIFCONF
) {
2442 ret
= dev_ifconf((char __user
*) arg
);
2446 if (cmd
== SIOCGIFNAME
)
2447 return dev_ifname((struct ifreq __user
*)arg
);
2449 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2452 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
2454 colon
= strchr(ifr
.ifr_name
, ':');
2459 * See which interface the caller is talking about.
2464 * These ioctl calls:
2465 * - can be done by all.
2466 * - atomic and do not require locking.
2477 dev_load(ifr
.ifr_name
);
2478 read_lock(&dev_base_lock
);
2479 ret
= dev_ifsioc(&ifr
, cmd
);
2480 read_unlock(&dev_base_lock
);
2484 if (copy_to_user(arg
, &ifr
,
2485 sizeof(struct ifreq
)))
2491 dev_load(ifr
.ifr_name
);
2493 ret
= dev_ethtool(&ifr
);
2498 if (copy_to_user(arg
, &ifr
,
2499 sizeof(struct ifreq
)))
2505 * These ioctl calls:
2506 * - require superuser power.
2507 * - require strict serialization.
2513 if (!capable(CAP_NET_ADMIN
))
2515 dev_load(ifr
.ifr_name
);
2517 ret
= dev_ifsioc(&ifr
, cmd
);
2522 if (copy_to_user(arg
, &ifr
,
2523 sizeof(struct ifreq
)))
2529 * These ioctl calls:
2530 * - require superuser power.
2531 * - require strict serialization.
2532 * - do not return a value
2542 case SIOCSIFHWBROADCAST
:
2545 case SIOCBONDENSLAVE
:
2546 case SIOCBONDRELEASE
:
2547 case SIOCBONDSETHWADDR
:
2548 case SIOCBONDSLAVEINFOQUERY
:
2549 case SIOCBONDINFOQUERY
:
2550 case SIOCBONDCHANGEACTIVE
:
2553 if (!capable(CAP_NET_ADMIN
))
2555 dev_load(ifr
.ifr_name
);
2557 ret
= dev_ifsioc(&ifr
, cmd
);
2562 /* Get the per device memory space. We can add this but
2563 * currently do not support it */
2565 /* Set the per device memory buffer space.
2566 * Not applicable in our case */
2571 * Unknown or private ioctl.
2574 if (cmd
== SIOCWANDEV
||
2575 (cmd
>= SIOCDEVPRIVATE
&&
2576 cmd
<= SIOCDEVPRIVATE
+ 15)) {
2577 dev_load(ifr
.ifr_name
);
2579 ret
= dev_ifsioc(&ifr
, cmd
);
2581 if (!ret
&& copy_to_user(arg
, &ifr
,
2582 sizeof(struct ifreq
)))
2587 /* Take care of Wireless Extensions */
2588 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
) {
2589 /* If command is `set a parameter', or
2590 * `get the encoding parameters', check if
2591 * the user has the right to do it */
2592 if (IW_IS_SET(cmd
) || cmd
== SIOCGIWENCODE
) {
2593 if (!capable(CAP_NET_ADMIN
))
2596 dev_load(ifr
.ifr_name
);
2598 /* Follow me in net/core/wireless.c */
2599 ret
= wireless_process_ioctl(&ifr
, cmd
);
2601 if (IW_IS_GET(cmd
) &&
2602 copy_to_user(arg
, &ifr
,
2603 sizeof(struct ifreq
)))
2607 #endif /* WIRELESS_EXT */
2614 * dev_new_index - allocate an ifindex
2616 * Returns a suitable unique value for a new device interface
2617 * number. The caller must hold the rtnl semaphore or the
2618 * dev_base_lock to be sure it remains unique.
2620 static int dev_new_index(void)
2626 if (!__dev_get_by_index(ifindex
))
2631 static int dev_boot_phase
= 1;
2633 /* Delayed registration/unregisteration */
2634 static DEFINE_SPINLOCK(net_todo_list_lock
);
2635 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
2637 static inline void net_set_todo(struct net_device
*dev
)
2639 spin_lock(&net_todo_list_lock
);
2640 list_add_tail(&dev
->todo_list
, &net_todo_list
);
2641 spin_unlock(&net_todo_list_lock
);
2645 * register_netdevice - register a network device
2646 * @dev: device to register
2648 * Take a completed network device structure and add it to the kernel
2649 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2650 * chain. 0 is returned on success. A negative errno code is returned
2651 * on a failure to set up the device, or if the name is a duplicate.
2653 * Callers must hold the rtnl semaphore. You may want
2654 * register_netdev() instead of this.
2657 * The locking appears insufficient to guarantee two parallel registers
2658 * will not get the same name.
2661 int register_netdevice(struct net_device
*dev
)
2663 struct hlist_head
*head
;
2664 struct hlist_node
*p
;
2667 BUG_ON(dev_boot_phase
);
2670 /* When net_device's are persistent, this will be fatal. */
2671 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
2673 spin_lock_init(&dev
->queue_lock
);
2674 spin_lock_init(&dev
->xmit_lock
);
2675 dev
->xmit_lock_owner
= -1;
2676 #ifdef CONFIG_NET_CLS_ACT
2677 spin_lock_init(&dev
->ingress_lock
);
2680 ret
= alloc_divert_blk(dev
);
2686 /* Init, if this function is available */
2688 ret
= dev
->init(dev
);
2696 if (!dev_valid_name(dev
->name
)) {
2701 dev
->ifindex
= dev_new_index();
2702 if (dev
->iflink
== -1)
2703 dev
->iflink
= dev
->ifindex
;
2705 /* Check for existence of name */
2706 head
= dev_name_hash(dev
->name
);
2707 hlist_for_each(p
, head
) {
2708 struct net_device
*d
2709 = hlist_entry(p
, struct net_device
, name_hlist
);
2710 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
2716 /* Fix illegal SG+CSUM combinations. */
2717 if ((dev
->features
& NETIF_F_SG
) &&
2718 !(dev
->features
& (NETIF_F_IP_CSUM
|
2720 NETIF_F_HW_CSUM
))) {
2721 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2723 dev
->features
&= ~NETIF_F_SG
;
2726 /* TSO requires that SG is present as well. */
2727 if ((dev
->features
& NETIF_F_TSO
) &&
2728 !(dev
->features
& NETIF_F_SG
)) {
2729 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2731 dev
->features
&= ~NETIF_F_TSO
;
2733 if (dev
->features
& NETIF_F_UFO
) {
2734 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
2735 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
2736 "NETIF_F_HW_CSUM feature.\n",
2738 dev
->features
&= ~NETIF_F_UFO
;
2740 if (!(dev
->features
& NETIF_F_SG
)) {
2741 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
2742 "NETIF_F_SG feature.\n",
2744 dev
->features
&= ~NETIF_F_UFO
;
2749 * nil rebuild_header routine,
2750 * that should be never called and used as just bug trap.
2753 if (!dev
->rebuild_header
)
2754 dev
->rebuild_header
= default_rebuild_header
;
2757 * Default initial state at registry is that the
2758 * device is present.
2761 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
2764 dev_init_scheduler(dev
);
2765 write_lock_bh(&dev_base_lock
);
2767 dev_tail
= &dev
->next
;
2768 hlist_add_head(&dev
->name_hlist
, head
);
2769 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
2771 dev
->reg_state
= NETREG_REGISTERING
;
2772 write_unlock_bh(&dev_base_lock
);
2774 /* Notify protocols, that a new device appeared. */
2775 notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
2777 /* Finish registration after unlock */
2784 free_divert_blk(dev
);
2789 * register_netdev - register a network device
2790 * @dev: device to register
2792 * Take a completed network device structure and add it to the kernel
2793 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2794 * chain. 0 is returned on success. A negative errno code is returned
2795 * on a failure to set up the device, or if the name is a duplicate.
2797 * This is a wrapper around register_netdev that takes the rtnl semaphore
2798 * and expands the device name if you passed a format string to
2801 int register_netdev(struct net_device
*dev
)
2808 * If the name is a format string the caller wants us to do a
2811 if (strchr(dev
->name
, '%')) {
2812 err
= dev_alloc_name(dev
, dev
->name
);
2818 * Back compatibility hook. Kill this one in 2.5
2820 if (dev
->name
[0] == 0 || dev
->name
[0] == ' ') {
2821 err
= dev_alloc_name(dev
, "eth%d");
2826 err
= register_netdevice(dev
);
2831 EXPORT_SYMBOL(register_netdev
);
2834 * netdev_wait_allrefs - wait until all references are gone.
2836 * This is called when unregistering network devices.
2838 * Any protocol or device that holds a reference should register
2839 * for netdevice notification, and cleanup and put back the
2840 * reference if they receive an UNREGISTER event.
2841 * We can get stuck here if buggy protocols don't correctly
2844 static void netdev_wait_allrefs(struct net_device
*dev
)
2846 unsigned long rebroadcast_time
, warning_time
;
2848 rebroadcast_time
= warning_time
= jiffies
;
2849 while (atomic_read(&dev
->refcnt
) != 0) {
2850 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
2853 /* Rebroadcast unregister notification */
2854 notifier_call_chain(&netdev_chain
,
2855 NETDEV_UNREGISTER
, dev
);
2857 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
2859 /* We must not have linkwatch events
2860 * pending on unregister. If this
2861 * happens, we simply run the queue
2862 * unscheduled, resulting in a noop
2865 linkwatch_run_queue();
2870 rebroadcast_time
= jiffies
;
2875 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
2876 printk(KERN_EMERG
"unregister_netdevice: "
2877 "waiting for %s to become free. Usage "
2879 dev
->name
, atomic_read(&dev
->refcnt
));
2880 warning_time
= jiffies
;
2889 * register_netdevice(x1);
2890 * register_netdevice(x2);
2892 * unregister_netdevice(y1);
2893 * unregister_netdevice(y2);
2899 * We are invoked by rtnl_unlock() after it drops the semaphore.
2900 * This allows us to deal with problems:
2901 * 1) We can create/delete sysfs objects which invoke hotplug
2902 * without deadlocking with linkwatch via keventd.
2903 * 2) Since we run with the RTNL semaphore not held, we can sleep
2904 * safely in order to wait for the netdev refcnt to drop to zero.
2906 static DECLARE_MUTEX(net_todo_run_mutex
);
2907 void netdev_run_todo(void)
2909 struct list_head list
= LIST_HEAD_INIT(list
);
2913 /* Need to guard against multiple cpu's getting out of order. */
2914 down(&net_todo_run_mutex
);
2916 /* Not safe to do outside the semaphore. We must not return
2917 * until all unregister events invoked by the local processor
2918 * have been completed (either by this todo run, or one on
2921 if (list_empty(&net_todo_list
))
2924 /* Snapshot list, allow later requests */
2925 spin_lock(&net_todo_list_lock
);
2926 list_splice_init(&net_todo_list
, &list
);
2927 spin_unlock(&net_todo_list_lock
);
2929 while (!list_empty(&list
)) {
2930 struct net_device
*dev
2931 = list_entry(list
.next
, struct net_device
, todo_list
);
2932 list_del(&dev
->todo_list
);
2934 switch(dev
->reg_state
) {
2935 case NETREG_REGISTERING
:
2936 err
= netdev_register_sysfs(dev
);
2938 printk(KERN_ERR
"%s: failed sysfs registration (%d)\n",
2940 dev
->reg_state
= NETREG_REGISTERED
;
2943 case NETREG_UNREGISTERING
:
2944 netdev_unregister_sysfs(dev
);
2945 dev
->reg_state
= NETREG_UNREGISTERED
;
2947 netdev_wait_allrefs(dev
);
2950 BUG_ON(atomic_read(&dev
->refcnt
));
2951 BUG_TRAP(!dev
->ip_ptr
);
2952 BUG_TRAP(!dev
->ip6_ptr
);
2953 BUG_TRAP(!dev
->dn_ptr
);
2956 /* It must be the very last action,
2957 * after this 'dev' may point to freed up memory.
2959 if (dev
->destructor
)
2960 dev
->destructor(dev
);
2964 printk(KERN_ERR
"network todo '%s' but state %d\n",
2965 dev
->name
, dev
->reg_state
);
2971 up(&net_todo_run_mutex
);
2975 * alloc_netdev - allocate network device
2976 * @sizeof_priv: size of private data to allocate space for
2977 * @name: device name format string
2978 * @setup: callback to initialize device
2980 * Allocates a struct net_device with private data area for driver use
2981 * and performs basic initialization.
2983 struct net_device
*alloc_netdev(int sizeof_priv
, const char *name
,
2984 void (*setup
)(struct net_device
*))
2987 struct net_device
*dev
;
2990 /* ensure 32-byte alignment of both the device and private area */
2991 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
2992 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
2994 p
= kmalloc(alloc_size
, GFP_KERNEL
);
2996 printk(KERN_ERR
"alloc_dev: Unable to allocate device.\n");
2999 memset(p
, 0, alloc_size
);
3001 dev
= (struct net_device
*)
3002 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3003 dev
->padded
= (char *)dev
- (char *)p
;
3006 dev
->priv
= netdev_priv(dev
);
3009 strcpy(dev
->name
, name
);
3012 EXPORT_SYMBOL(alloc_netdev
);
3015 * free_netdev - free network device
3018 * This function does the last stage of destroying an allocated device
3019 * interface. The reference to the device object is released.
3020 * If this is the last reference then it will be freed.
3022 void free_netdev(struct net_device
*dev
)
3025 /* Compatiablity with error handling in drivers */
3026 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3027 kfree((char *)dev
- dev
->padded
);
3031 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3032 dev
->reg_state
= NETREG_RELEASED
;
3034 /* will free via class release */
3035 class_device_put(&dev
->class_dev
);
3037 kfree((char *)dev
- dev
->padded
);
3041 /* Synchronize with packet receive processing. */
3042 void synchronize_net(void)
3049 * unregister_netdevice - remove device from the kernel
3052 * This function shuts down a device interface and removes it
3053 * from the kernel tables. On success 0 is returned, on a failure
3054 * a negative errno code is returned.
3056 * Callers must hold the rtnl semaphore. You may want
3057 * unregister_netdev() instead of this.
3060 int unregister_netdevice(struct net_device
*dev
)
3062 struct net_device
*d
, **dp
;
3064 BUG_ON(dev_boot_phase
);
3067 /* Some devices call without registering for initialization unwind. */
3068 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3069 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3070 "was registered\n", dev
->name
, dev
);
3074 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3076 /* If device is running, close it first. */
3077 if (dev
->flags
& IFF_UP
)
3080 /* And unlink it from device chain. */
3081 for (dp
= &dev_base
; (d
= *dp
) != NULL
; dp
= &d
->next
) {
3083 write_lock_bh(&dev_base_lock
);
3084 hlist_del(&dev
->name_hlist
);
3085 hlist_del(&dev
->index_hlist
);
3086 if (dev_tail
== &dev
->next
)
3089 write_unlock_bh(&dev_base_lock
);
3094 printk(KERN_ERR
"unregister net_device: '%s' not found\n",
3099 dev
->reg_state
= NETREG_UNREGISTERING
;
3103 /* Shutdown queueing discipline. */
3107 /* Notify protocols, that we are about to destroy
3108 this device. They should clean all the things.
3110 notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3113 * Flush the multicast chain
3115 dev_mc_discard(dev
);
3120 /* Notifier chain MUST detach us from master device. */
3121 BUG_TRAP(!dev
->master
);
3123 free_divert_blk(dev
);
3125 /* Finish processing unregister after unlock */
3135 * unregister_netdev - remove device from the kernel
3138 * This function shuts down a device interface and removes it
3139 * from the kernel tables. On success 0 is returned, on a failure
3140 * a negative errno code is returned.
3142 * This is just a wrapper for unregister_netdevice that takes
3143 * the rtnl semaphore. In general you want to use this and not
3144 * unregister_netdevice.
3146 void unregister_netdev(struct net_device
*dev
)
3149 unregister_netdevice(dev
);
3153 EXPORT_SYMBOL(unregister_netdev
);
3155 #ifdef CONFIG_HOTPLUG_CPU
3156 static int dev_cpu_callback(struct notifier_block
*nfb
,
3157 unsigned long action
,
3160 struct sk_buff
**list_skb
;
3161 struct net_device
**list_net
;
3162 struct sk_buff
*skb
;
3163 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3164 struct softnet_data
*sd
, *oldsd
;
3166 if (action
!= CPU_DEAD
)
3169 local_irq_disable();
3170 cpu
= smp_processor_id();
3171 sd
= &per_cpu(softnet_data
, cpu
);
3172 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3174 /* Find end of our completion_queue. */
3175 list_skb
= &sd
->completion_queue
;
3177 list_skb
= &(*list_skb
)->next
;
3178 /* Append completion queue from offline CPU. */
3179 *list_skb
= oldsd
->completion_queue
;
3180 oldsd
->completion_queue
= NULL
;
3182 /* Find end of our output_queue. */
3183 list_net
= &sd
->output_queue
;
3185 list_net
= &(*list_net
)->next_sched
;
3186 /* Append output queue from offline CPU. */
3187 *list_net
= oldsd
->output_queue
;
3188 oldsd
->output_queue
= NULL
;
3190 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3193 /* Process offline CPU's input_pkt_queue */
3194 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3199 #endif /* CONFIG_HOTPLUG_CPU */
3203 * Initialize the DEV module. At boot time this walks the device list and
3204 * unhooks any devices that fail to initialise (normally hardware not
3205 * present) and leaves us with a valid list of present and active devices.
3210 * This is called single threaded during boot, so no need
3211 * to take the rtnl semaphore.
3213 static int __init
net_dev_init(void)
3215 int i
, rc
= -ENOMEM
;
3217 BUG_ON(!dev_boot_phase
);
3221 if (dev_proc_init())
3224 if (netdev_sysfs_init())
3227 INIT_LIST_HEAD(&ptype_all
);
3228 for (i
= 0; i
< 16; i
++)
3229 INIT_LIST_HEAD(&ptype_base
[i
]);
3231 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
3232 INIT_HLIST_HEAD(&dev_name_head
[i
]);
3234 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
3235 INIT_HLIST_HEAD(&dev_index_head
[i
]);
3238 * Initialise the packet receive queues.
3241 for (i
= 0; i
< NR_CPUS
; i
++) {
3242 struct softnet_data
*queue
;
3244 queue
= &per_cpu(softnet_data
, i
);
3245 skb_queue_head_init(&queue
->input_pkt_queue
);
3246 queue
->completion_queue
= NULL
;
3247 INIT_LIST_HEAD(&queue
->poll_list
);
3248 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
3249 queue
->backlog_dev
.weight
= weight_p
;
3250 queue
->backlog_dev
.poll
= process_backlog
;
3251 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
3256 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
3257 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
3259 hotcpu_notifier(dev_cpu_callback
, 0);
3267 subsys_initcall(net_dev_init
);
3269 EXPORT_SYMBOL(__dev_get_by_index
);
3270 EXPORT_SYMBOL(__dev_get_by_name
);
3271 EXPORT_SYMBOL(__dev_remove_pack
);
3272 EXPORT_SYMBOL(__skb_linearize
);
3273 EXPORT_SYMBOL(dev_valid_name
);
3274 EXPORT_SYMBOL(dev_add_pack
);
3275 EXPORT_SYMBOL(dev_alloc_name
);
3276 EXPORT_SYMBOL(dev_close
);
3277 EXPORT_SYMBOL(dev_get_by_flags
);
3278 EXPORT_SYMBOL(dev_get_by_index
);
3279 EXPORT_SYMBOL(dev_get_by_name
);
3280 EXPORT_SYMBOL(dev_open
);
3281 EXPORT_SYMBOL(dev_queue_xmit
);
3282 EXPORT_SYMBOL(dev_remove_pack
);
3283 EXPORT_SYMBOL(dev_set_allmulti
);
3284 EXPORT_SYMBOL(dev_set_promiscuity
);
3285 EXPORT_SYMBOL(dev_change_flags
);
3286 EXPORT_SYMBOL(dev_set_mtu
);
3287 EXPORT_SYMBOL(dev_set_mac_address
);
3288 EXPORT_SYMBOL(free_netdev
);
3289 EXPORT_SYMBOL(netdev_boot_setup_check
);
3290 EXPORT_SYMBOL(netdev_set_master
);
3291 EXPORT_SYMBOL(netdev_state_change
);
3292 EXPORT_SYMBOL(netif_receive_skb
);
3293 EXPORT_SYMBOL(netif_rx
);
3294 EXPORT_SYMBOL(register_gifconf
);
3295 EXPORT_SYMBOL(register_netdevice
);
3296 EXPORT_SYMBOL(register_netdevice_notifier
);
3297 EXPORT_SYMBOL(skb_checksum_help
);
3298 EXPORT_SYMBOL(synchronize_net
);
3299 EXPORT_SYMBOL(unregister_netdevice
);
3300 EXPORT_SYMBOL(unregister_netdevice_notifier
);
3301 EXPORT_SYMBOL(net_enable_timestamp
);
3302 EXPORT_SYMBOL(net_disable_timestamp
);
3303 EXPORT_SYMBOL(dev_get_flags
);
3305 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3306 EXPORT_SYMBOL(br_handle_frame_hook
);
3307 EXPORT_SYMBOL(br_fdb_get_hook
);
3308 EXPORT_SYMBOL(br_fdb_put_hook
);
3312 EXPORT_SYMBOL(dev_load
);
3315 EXPORT_PER_CPU_SYMBOL(softnet_data
);