2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/config.h>
80 #include <linux/cpu.h>
81 #include <linux/types.h>
82 #include <linux/kernel.h>
83 #include <linux/sched.h>
84 #include <linux/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/notifier.h>
94 #include <linux/skbuff.h>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
101 #include <linux/divert.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/kmod.h>
108 #include <linux/module.h>
109 #include <linux/kallsyms.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #include <linux/wireless.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
118 * The list of packet types we will receive (as opposed to discard)
119 * and the routines to invoke.
121 * Why 16. Because with 16 the only overlap we get on a hash of the
122 * low nibble of the protocol value is RARP/SNAP/X.25.
124 * NOTE: That is no longer true with the addition of VLAN tags. Not
125 * sure which should go first, but I bet it won't make much
126 * difference if we are running VLANs. The good news is that
127 * this protocol won't be in the list unless compiled in, so
128 * the average user (w/out VLANs) will not be adversly affected.
145 static DEFINE_SPINLOCK(ptype_lock
);
146 static struct list_head ptype_base
[16]; /* 16 way hashed list */
147 static struct list_head ptype_all
; /* Taps */
150 * The @dev_base list is protected by @dev_base_lock and the rtln
153 * Pure readers hold dev_base_lock for reading.
155 * Writers must hold the rtnl semaphore while they loop through the
156 * dev_base list, and hold dev_base_lock for writing when they do the
157 * actual updates. This allows pure readers to access the list even
158 * while a writer is preparing to update it.
160 * To put it another way, dev_base_lock is held for writing only to
161 * protect against pure readers; the rtnl semaphore provides the
162 * protection against other writers.
164 * See, for example usages, register_netdevice() and
165 * unregister_netdevice(), which must be called with the rtnl
168 struct net_device
*dev_base
;
169 static struct net_device
**dev_tail
= &dev_base
;
170 DEFINE_RWLOCK(dev_base_lock
);
172 EXPORT_SYMBOL(dev_base
);
173 EXPORT_SYMBOL(dev_base_lock
);
175 #define NETDEV_HASHBITS 8
176 static struct hlist_head dev_name_head
[1<<NETDEV_HASHBITS
];
177 static struct hlist_head dev_index_head
[1<<NETDEV_HASHBITS
];
179 static inline struct hlist_head
*dev_name_hash(const char *name
)
181 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
182 return &dev_name_head
[hash
& ((1<<NETDEV_HASHBITS
)-1)];
185 static inline struct hlist_head
*dev_index_hash(int ifindex
)
187 return &dev_index_head
[ifindex
& ((1<<NETDEV_HASHBITS
)-1)];
194 static struct notifier_block
*netdev_chain
;
197 * Device drivers call our routines to queue packets here. We empty the
198 * queue in the local softnet handler.
200 DEFINE_PER_CPU(struct softnet_data
, softnet_data
) = { NULL
};
203 extern int netdev_sysfs_init(void);
204 extern int netdev_register_sysfs(struct net_device
*);
205 extern void netdev_unregister_sysfs(struct net_device
*);
207 #define netdev_sysfs_init() (0)
208 #define netdev_register_sysfs(dev) (0)
209 #define netdev_unregister_sysfs(dev) do { } while(0)
213 /*******************************************************************************
215 Protocol management and registration routines
217 *******************************************************************************/
226 * Add a protocol ID to the list. Now that the input handler is
227 * smarter we can dispense with all the messy stuff that used to be
230 * BEWARE!!! Protocol handlers, mangling input packets,
231 * MUST BE last in hash buckets and checking protocol handlers
232 * MUST start from promiscuous ptype_all chain in net_bh.
233 * It is true now, do not change it.
234 * Explanation follows: if protocol handler, mangling packet, will
235 * be the first on list, it is not able to sense, that packet
236 * is cloned and should be copied-on-write, so that it will
237 * change it and subsequent readers will get broken packet.
242 * dev_add_pack - add packet handler
243 * @pt: packet type declaration
245 * Add a protocol handler to the networking stack. The passed &packet_type
246 * is linked into kernel lists and may not be freed until it has been
247 * removed from the kernel lists.
249 * This call does not sleep therefore it can not
250 * guarantee all CPU's that are in middle of receiving packets
251 * will see the new packet type (until the next received packet).
254 void dev_add_pack(struct packet_type
*pt
)
258 spin_lock_bh(&ptype_lock
);
259 if (pt
->type
== htons(ETH_P_ALL
)) {
261 list_add_rcu(&pt
->list
, &ptype_all
);
263 hash
= ntohs(pt
->type
) & 15;
264 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
266 spin_unlock_bh(&ptype_lock
);
270 * __dev_remove_pack - remove packet handler
271 * @pt: packet type declaration
273 * Remove a protocol handler that was previously added to the kernel
274 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
275 * from the kernel lists and can be freed or reused once this function
278 * The packet type might still be in use by receivers
279 * and must not be freed until after all the CPU's have gone
280 * through a quiescent state.
282 void __dev_remove_pack(struct packet_type
*pt
)
284 struct list_head
*head
;
285 struct packet_type
*pt1
;
287 spin_lock_bh(&ptype_lock
);
289 if (pt
->type
== htons(ETH_P_ALL
)) {
293 head
= &ptype_base
[ntohs(pt
->type
) & 15];
295 list_for_each_entry(pt1
, head
, list
) {
297 list_del_rcu(&pt
->list
);
302 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
304 spin_unlock_bh(&ptype_lock
);
307 * dev_remove_pack - remove packet handler
308 * @pt: packet type declaration
310 * Remove a protocol handler that was previously added to the kernel
311 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
312 * from the kernel lists and can be freed or reused once this function
315 * This call sleeps to guarantee that no CPU is looking at the packet
318 void dev_remove_pack(struct packet_type
*pt
)
320 __dev_remove_pack(pt
);
325 /******************************************************************************
327 Device Boot-time Settings Routines
329 *******************************************************************************/
331 /* Boot time configuration table */
332 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
335 * netdev_boot_setup_add - add new setup entry
336 * @name: name of the device
337 * @map: configured settings for the device
339 * Adds new setup entry to the dev_boot_setup list. The function
340 * returns 0 on error and 1 on success. This is a generic routine to
343 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
345 struct netdev_boot_setup
*s
;
349 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
350 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
351 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
352 strcpy(s
[i
].name
, name
);
353 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
358 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
362 * netdev_boot_setup_check - check boot time settings
363 * @dev: the netdevice
365 * Check boot time settings for the device.
366 * The found settings are set for the device to be used
367 * later in the device probing.
368 * Returns 0 if no settings found, 1 if they are.
370 int netdev_boot_setup_check(struct net_device
*dev
)
372 struct netdev_boot_setup
*s
= dev_boot_setup
;
375 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
376 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
377 !strncmp(dev
->name
, s
[i
].name
, strlen(s
[i
].name
))) {
378 dev
->irq
= s
[i
].map
.irq
;
379 dev
->base_addr
= s
[i
].map
.base_addr
;
380 dev
->mem_start
= s
[i
].map
.mem_start
;
381 dev
->mem_end
= s
[i
].map
.mem_end
;
390 * netdev_boot_base - get address from boot time settings
391 * @prefix: prefix for network device
392 * @unit: id for network device
394 * Check boot time settings for the base address of device.
395 * The found settings are set for the device to be used
396 * later in the device probing.
397 * Returns 0 if no settings found.
399 unsigned long netdev_boot_base(const char *prefix
, int unit
)
401 const struct netdev_boot_setup
*s
= dev_boot_setup
;
405 sprintf(name
, "%s%d", prefix
, unit
);
408 * If device already registered then return base of 1
409 * to indicate not to probe for this interface
411 if (__dev_get_by_name(name
))
414 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
415 if (!strcmp(name
, s
[i
].name
))
416 return s
[i
].map
.base_addr
;
421 * Saves at boot time configured settings for any netdevice.
423 int __init
netdev_boot_setup(char *str
)
428 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
433 memset(&map
, 0, sizeof(map
));
437 map
.base_addr
= ints
[2];
439 map
.mem_start
= ints
[3];
441 map
.mem_end
= ints
[4];
443 /* Add new entry to the list */
444 return netdev_boot_setup_add(str
, &map
);
447 __setup("netdev=", netdev_boot_setup
);
449 /*******************************************************************************
451 Device Interface Subroutines
453 *******************************************************************************/
456 * __dev_get_by_name - find a device by its name
457 * @name: name to find
459 * Find an interface by name. Must be called under RTNL semaphore
460 * or @dev_base_lock. If the name is found a pointer to the device
461 * is returned. If the name is not found then %NULL is returned. The
462 * reference counters are not incremented so the caller must be
463 * careful with locks.
466 struct net_device
*__dev_get_by_name(const char *name
)
468 struct hlist_node
*p
;
470 hlist_for_each(p
, dev_name_hash(name
)) {
471 struct net_device
*dev
472 = hlist_entry(p
, struct net_device
, name_hlist
);
473 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
480 * dev_get_by_name - find a device by its name
481 * @name: name to find
483 * Find an interface by name. This can be called from any
484 * context and does its own locking. The returned handle has
485 * the usage count incremented and the caller must use dev_put() to
486 * release it when it is no longer needed. %NULL is returned if no
487 * matching device is found.
490 struct net_device
*dev_get_by_name(const char *name
)
492 struct net_device
*dev
;
494 read_lock(&dev_base_lock
);
495 dev
= __dev_get_by_name(name
);
498 read_unlock(&dev_base_lock
);
503 * __dev_get_by_index - find a device by its ifindex
504 * @ifindex: index of device
506 * Search for an interface by index. Returns %NULL if the device
507 * is not found or a pointer to the device. The device has not
508 * had its reference counter increased so the caller must be careful
509 * about locking. The caller must hold either the RTNL semaphore
513 struct net_device
*__dev_get_by_index(int ifindex
)
515 struct hlist_node
*p
;
517 hlist_for_each(p
, dev_index_hash(ifindex
)) {
518 struct net_device
*dev
519 = hlist_entry(p
, struct net_device
, index_hlist
);
520 if (dev
->ifindex
== ifindex
)
528 * dev_get_by_index - find a device by its ifindex
529 * @ifindex: index of device
531 * Search for an interface by index. Returns NULL if the device
532 * is not found or a pointer to the device. The device returned has
533 * had a reference added and the pointer is safe until the user calls
534 * dev_put to indicate they have finished with it.
537 struct net_device
*dev_get_by_index(int ifindex
)
539 struct net_device
*dev
;
541 read_lock(&dev_base_lock
);
542 dev
= __dev_get_by_index(ifindex
);
545 read_unlock(&dev_base_lock
);
550 * dev_getbyhwaddr - find a device by its hardware address
551 * @type: media type of device
552 * @ha: hardware address
554 * Search for an interface by MAC address. Returns NULL if the device
555 * is not found or a pointer to the device. The caller must hold the
556 * rtnl semaphore. The returned device has not had its ref count increased
557 * and the caller must therefore be careful about locking
560 * If the API was consistent this would be __dev_get_by_hwaddr
563 struct net_device
*dev_getbyhwaddr(unsigned short type
, char *ha
)
565 struct net_device
*dev
;
569 for (dev
= dev_base
; dev
; dev
= dev
->next
)
570 if (dev
->type
== type
&&
571 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
576 EXPORT_SYMBOL(dev_getbyhwaddr
);
578 struct net_device
*dev_getfirstbyhwtype(unsigned short type
)
580 struct net_device
*dev
;
583 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
584 if (dev
->type
== type
) {
593 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
596 * dev_get_by_flags - find any device with given flags
597 * @if_flags: IFF_* values
598 * @mask: bitmask of bits in if_flags to check
600 * Search for any interface with the given flags. Returns NULL if a device
601 * is not found or a pointer to the device. The device returned has
602 * had a reference added and the pointer is safe until the user calls
603 * dev_put to indicate they have finished with it.
606 struct net_device
* dev_get_by_flags(unsigned short if_flags
, unsigned short mask
)
608 struct net_device
*dev
;
610 read_lock(&dev_base_lock
);
611 for (dev
= dev_base
; dev
!= NULL
; dev
= dev
->next
) {
612 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
617 read_unlock(&dev_base_lock
);
622 * dev_valid_name - check if name is okay for network device
625 * Network device names need to be valid file names to
626 * to allow sysfs to work
628 int dev_valid_name(const char *name
)
630 return !(*name
== '\0'
631 || !strcmp(name
, ".")
632 || !strcmp(name
, "..")
633 || strchr(name
, '/'));
637 * dev_alloc_name - allocate a name for a device
639 * @name: name format string
641 * Passed a format string - eg "lt%d" it will try and find a suitable
642 * id. Not efficient for many devices, not called a lot. The caller
643 * must hold the dev_base or rtnl lock while allocating the name and
644 * adding the device in order to avoid duplicates. Returns the number
645 * of the unit assigned or a negative errno code.
648 int dev_alloc_name(struct net_device
*dev
, const char *name
)
653 const int max_netdevices
= 8*PAGE_SIZE
;
655 struct net_device
*d
;
657 p
= strnchr(name
, IFNAMSIZ
-1, '%');
660 * Verify the string as this thing may have come from
661 * the user. There must be either one "%d" and no other "%"
664 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
667 /* Use one page as a bit array of possible slots */
668 inuse
= (long *) get_zeroed_page(GFP_ATOMIC
);
672 for (d
= dev_base
; d
; d
= d
->next
) {
673 if (!sscanf(d
->name
, name
, &i
))
675 if (i
< 0 || i
>= max_netdevices
)
678 /* avoid cases where sscanf is not exact inverse of printf */
679 snprintf(buf
, sizeof(buf
), name
, i
);
680 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
684 i
= find_first_zero_bit(inuse
, max_netdevices
);
685 free_page((unsigned long) inuse
);
688 snprintf(buf
, sizeof(buf
), name
, i
);
689 if (!__dev_get_by_name(buf
)) {
690 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
694 /* It is possible to run out of possible slots
695 * when the name is long and there isn't enough space left
696 * for the digits, or if all bits are used.
703 * dev_change_name - change name of a device
705 * @newname: name (or format string) must be at least IFNAMSIZ
707 * Change name of a device, can pass format strings "eth%d".
710 int dev_change_name(struct net_device
*dev
, char *newname
)
716 if (dev
->flags
& IFF_UP
)
719 if (!dev_valid_name(newname
))
722 if (strchr(newname
, '%')) {
723 err
= dev_alloc_name(dev
, newname
);
726 strcpy(newname
, dev
->name
);
728 else if (__dev_get_by_name(newname
))
731 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
733 err
= class_device_rename(&dev
->class_dev
, dev
->name
);
735 hlist_del(&dev
->name_hlist
);
736 hlist_add_head(&dev
->name_hlist
, dev_name_hash(dev
->name
));
737 notifier_call_chain(&netdev_chain
, NETDEV_CHANGENAME
, dev
);
744 * netdev_features_change - device changes fatures
745 * @dev: device to cause notification
747 * Called to indicate a device has changed features.
749 void netdev_features_change(struct net_device
*dev
)
751 notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
753 EXPORT_SYMBOL(netdev_features_change
);
756 * netdev_state_change - device changes state
757 * @dev: device to cause notification
759 * Called to indicate a device has changed state. This function calls
760 * the notifier chains for netdev_chain and sends a NEWLINK message
761 * to the routing socket.
763 void netdev_state_change(struct net_device
*dev
)
765 if (dev
->flags
& IFF_UP
) {
766 notifier_call_chain(&netdev_chain
, NETDEV_CHANGE
, dev
);
767 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
772 * dev_load - load a network module
773 * @name: name of interface
775 * If a network interface is not present and the process has suitable
776 * privileges this function loads the module. If module loading is not
777 * available in this kernel then it becomes a nop.
780 void dev_load(const char *name
)
782 struct net_device
*dev
;
784 read_lock(&dev_base_lock
);
785 dev
= __dev_get_by_name(name
);
786 read_unlock(&dev_base_lock
);
788 if (!dev
&& capable(CAP_SYS_MODULE
))
789 request_module("%s", name
);
792 static int default_rebuild_header(struct sk_buff
*skb
)
794 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
795 skb
->dev
? skb
->dev
->name
: "NULL!!!");
802 * dev_open - prepare an interface for use.
803 * @dev: device to open
805 * Takes a device from down to up state. The device's private open
806 * function is invoked and then the multicast lists are loaded. Finally
807 * the device is moved into the up state and a %NETDEV_UP message is
808 * sent to the netdev notifier chain.
810 * Calling this function on an active interface is a nop. On a failure
811 * a negative errno code is returned.
813 int dev_open(struct net_device
*dev
)
821 if (dev
->flags
& IFF_UP
)
825 * Is it even present?
827 if (!netif_device_present(dev
))
831 * Call device private open method
833 set_bit(__LINK_STATE_START
, &dev
->state
);
835 ret
= dev
->open(dev
);
837 clear_bit(__LINK_STATE_START
, &dev
->state
);
841 * If it went open OK then:
848 dev
->flags
|= IFF_UP
;
851 * Initialize multicasting status
856 * Wakeup transmit queue engine
861 * ... and announce new interface.
863 notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
869 * dev_close - shutdown an interface.
870 * @dev: device to shutdown
872 * This function moves an active device into down state. A
873 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
874 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
877 int dev_close(struct net_device
*dev
)
879 if (!(dev
->flags
& IFF_UP
))
883 * Tell people we are going down, so that they can
884 * prepare to death, when device is still operating.
886 notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
890 clear_bit(__LINK_STATE_START
, &dev
->state
);
892 /* Synchronize to scheduled poll. We cannot touch poll list,
893 * it can be even on different cpu. So just clear netif_running(),
894 * and wait when poll really will happen. Actually, the best place
895 * for this is inside dev->stop() after device stopped its irq
896 * engine, but this requires more changes in devices. */
898 smp_mb__after_clear_bit(); /* Commit netif_running(). */
899 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
905 * Call the device specific close. This cannot fail.
906 * Only if device is UP
908 * We allow it to be called even after a DETACH hot-plug
915 * Device is now down.
918 dev
->flags
&= ~IFF_UP
;
921 * Tell people we are down
923 notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
930 * Device change register/unregister. These are not inline or static
931 * as we export them to the world.
935 * register_netdevice_notifier - register a network notifier block
938 * Register a notifier to be called when network device events occur.
939 * The notifier passed is linked into the kernel structures and must
940 * not be reused until it has been unregistered. A negative errno code
941 * is returned on a failure.
943 * When registered all registration and up events are replayed
944 * to the new notifier to allow device to have a race free
945 * view of the network device list.
948 int register_netdevice_notifier(struct notifier_block
*nb
)
950 struct net_device
*dev
;
954 err
= notifier_chain_register(&netdev_chain
, nb
);
956 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
957 nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
959 if (dev
->flags
& IFF_UP
)
960 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
968 * unregister_netdevice_notifier - unregister a network notifier block
971 * Unregister a notifier previously registered by
972 * register_netdevice_notifier(). The notifier is unlinked into the
973 * kernel structures and may then be reused. A negative errno code
974 * is returned on a failure.
977 int unregister_netdevice_notifier(struct notifier_block
*nb
)
979 return notifier_chain_unregister(&netdev_chain
, nb
);
983 * call_netdevice_notifiers - call all network notifier blocks
984 * @val: value passed unmodified to notifier function
985 * @v: pointer passed unmodified to notifier function
987 * Call all network notifier blocks. Parameters and return value
988 * are as for notifier_call_chain().
991 int call_netdevice_notifiers(unsigned long val
, void *v
)
993 return notifier_call_chain(&netdev_chain
, val
, v
);
996 /* When > 0 there are consumers of rx skb time stamps */
997 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
999 void net_enable_timestamp(void)
1001 atomic_inc(&netstamp_needed
);
1004 void net_disable_timestamp(void)
1006 atomic_dec(&netstamp_needed
);
1009 void __net_timestamp(struct sk_buff
*skb
)
1013 do_gettimeofday(&tv
);
1014 skb_set_timestamp(skb
, &tv
);
1016 EXPORT_SYMBOL(__net_timestamp
);
1018 static inline void net_timestamp(struct sk_buff
*skb
)
1020 if (atomic_read(&netstamp_needed
))
1021 __net_timestamp(skb
);
1023 skb
->tstamp
.off_sec
= 0;
1024 skb
->tstamp
.off_usec
= 0;
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
;
1040 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1041 /* Never send packets back to the socket
1042 * they originated from - MvS (miquels@drinkel.ow.org)
1044 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1045 (ptype
->af_packet_priv
== NULL
||
1046 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1047 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1051 /* skb->nh should be correctly
1052 set by sender, so that the second statement is
1053 just protection against buggy protocols.
1055 skb2
->mac
.raw
= skb2
->data
;
1057 if (skb2
->nh
.raw
< skb2
->data
||
1058 skb2
->nh
.raw
> skb2
->tail
) {
1059 if (net_ratelimit())
1060 printk(KERN_CRIT
"protocol %04x is "
1062 skb2
->protocol
, dev
->name
);
1063 skb2
->nh
.raw
= skb2
->data
;
1066 skb2
->h
.raw
= skb2
->nh
.raw
;
1067 skb2
->pkt_type
= PACKET_OUTGOING
;
1068 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1075 * Invalidate hardware checksum when packet is to be mangled, and
1076 * complete checksum manually on outgoing path.
1078 int skb_checksum_help(struct sk_buff
*skb
, int inward
)
1081 int ret
= 0, offset
= skb
->h
.raw
- skb
->data
;
1084 skb
->ip_summed
= CHECKSUM_NONE
;
1088 if (skb_cloned(skb
)) {
1089 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1094 BUG_ON(offset
> (int)skb
->len
);
1095 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1097 offset
= skb
->tail
- skb
->h
.raw
;
1098 BUG_ON(offset
<= 0);
1099 BUG_ON(skb
->csum
+ 2 > offset
);
1101 *(u16
*)(skb
->h
.raw
+ skb
->csum
) = csum_fold(csum
);
1102 skb
->ip_summed
= CHECKSUM_NONE
;
1107 /* Take action when hardware reception checksum errors are detected. */
1109 void netdev_rx_csum_fault(struct net_device
*dev
)
1111 if (net_ratelimit()) {
1112 printk(KERN_ERR
"%s: hw csum failure.\n",
1113 dev
? dev
->name
: "<unknown>");
1117 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1120 #ifdef CONFIG_HIGHMEM
1121 /* Actually, we should eliminate this check as soon as we know, that:
1122 * 1. IOMMU is present and allows to map all the memory.
1123 * 2. No high memory really exists on this machine.
1126 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1130 if (dev
->features
& NETIF_F_HIGHDMA
)
1133 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1134 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1140 #define illegal_highdma(dev, skb) (0)
1143 /* Keep head the same: replace data */
1144 int __skb_linearize(struct sk_buff
*skb
, gfp_t gfp_mask
)
1149 struct skb_shared_info
*ninfo
;
1150 int headerlen
= skb
->data
- skb
->head
;
1151 int expand
= (skb
->tail
+ skb
->data_len
) - skb
->end
;
1153 if (skb_shared(skb
))
1159 size
= skb
->end
- skb
->head
+ expand
;
1160 size
= SKB_DATA_ALIGN(size
);
1161 data
= kmalloc(size
+ sizeof(struct skb_shared_info
), gfp_mask
);
1165 /* Copy entire thing */
1166 if (skb_copy_bits(skb
, -headerlen
, data
, headerlen
+ skb
->len
))
1170 ninfo
= (struct skb_shared_info
*)(data
+ size
);
1171 atomic_set(&ninfo
->dataref
, 1);
1172 ninfo
->tso_size
= skb_shinfo(skb
)->tso_size
;
1173 ninfo
->tso_segs
= skb_shinfo(skb
)->tso_segs
;
1174 ninfo
->nr_frags
= 0;
1175 ninfo
->frag_list
= NULL
;
1177 /* Offset between the two in bytes */
1178 offset
= data
- skb
->head
;
1180 /* Free old data. */
1181 skb_release_data(skb
);
1184 skb
->end
= data
+ size
;
1186 /* Set up new pointers */
1187 skb
->h
.raw
+= offset
;
1188 skb
->nh
.raw
+= offset
;
1189 skb
->mac
.raw
+= offset
;
1190 skb
->tail
+= offset
;
1191 skb
->data
+= offset
;
1193 /* We are no longer a clone, even if we were. */
1196 skb
->tail
+= skb
->data_len
;
1201 #define HARD_TX_LOCK(dev, cpu) { \
1202 if ((dev->features & NETIF_F_LLTX) == 0) { \
1203 spin_lock(&dev->xmit_lock); \
1204 dev->xmit_lock_owner = cpu; \
1208 #define HARD_TX_UNLOCK(dev) { \
1209 if ((dev->features & NETIF_F_LLTX) == 0) { \
1210 dev->xmit_lock_owner = -1; \
1211 spin_unlock(&dev->xmit_lock); \
1216 * dev_queue_xmit - transmit a buffer
1217 * @skb: buffer to transmit
1219 * Queue a buffer for transmission to a network device. The caller must
1220 * have set the device and priority and built the buffer before calling
1221 * this function. The function can be called from an interrupt.
1223 * A negative errno code is returned on a failure. A success does not
1224 * guarantee the frame will be transmitted as it may be dropped due
1225 * to congestion or traffic shaping.
1227 * -----------------------------------------------------------------------------------
1228 * I notice this method can also return errors from the queue disciplines,
1229 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1232 * Regardless of the return value, the skb is consumed, so it is currently
1233 * difficult to retry a send to this method. (You can bump the ref count
1234 * before sending to hold a reference for retry if you are careful.)
1236 * When calling this method, interrupts MUST be enabled. This is because
1237 * the BH enable code must have IRQs enabled so that it will not deadlock.
1241 int dev_queue_xmit(struct sk_buff
*skb
)
1243 struct net_device
*dev
= skb
->dev
;
1247 if (skb_shinfo(skb
)->frag_list
&&
1248 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1249 __skb_linearize(skb
, GFP_ATOMIC
))
1252 /* Fragmented skb is linearized if device does not support SG,
1253 * or if at least one of fragments is in highmem and device
1254 * does not support DMA from it.
1256 if (skb_shinfo(skb
)->nr_frags
&&
1257 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1258 __skb_linearize(skb
, GFP_ATOMIC
))
1261 /* If packet is not checksummed and device does not support
1262 * checksumming for this protocol, complete checksumming here.
1264 if (skb
->ip_summed
== CHECKSUM_HW
&&
1265 (!(dev
->features
& (NETIF_F_HW_CSUM
| NETIF_F_NO_CSUM
)) &&
1266 (!(dev
->features
& NETIF_F_IP_CSUM
) ||
1267 skb
->protocol
!= htons(ETH_P_IP
))))
1268 if (skb_checksum_help(skb
, 0))
1271 spin_lock_prefetch(&dev
->queue_lock
);
1273 /* Disable soft irqs for various locks below. Also
1274 * stops preemption for RCU.
1278 /* Updates of qdisc are serialized by queue_lock.
1279 * The struct Qdisc which is pointed to by qdisc is now a
1280 * rcu structure - it may be accessed without acquiring
1281 * a lock (but the structure may be stale.) The freeing of the
1282 * qdisc will be deferred until it's known that there are no
1283 * more references to it.
1285 * If the qdisc has an enqueue function, we still need to
1286 * hold the queue_lock before calling it, since queue_lock
1287 * also serializes access to the device queue.
1290 q
= rcu_dereference(dev
->qdisc
);
1291 #ifdef CONFIG_NET_CLS_ACT
1292 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1295 /* Grab device queue */
1296 spin_lock(&dev
->queue_lock
);
1298 rc
= q
->enqueue(skb
, q
);
1302 spin_unlock(&dev
->queue_lock
);
1303 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1307 /* The device has no queue. Common case for software devices:
1308 loopback, all the sorts of tunnels...
1310 Really, it is unlikely that xmit_lock protection is necessary here.
1311 (f.e. loopback and IP tunnels are clean ignoring statistics
1313 However, it is possible, that they rely on protection
1316 Check this and shot the lock. It is not prone from deadlocks.
1317 Either shot noqueue qdisc, it is even simpler 8)
1319 if (dev
->flags
& IFF_UP
) {
1320 int cpu
= smp_processor_id(); /* ok because BHs are off */
1322 if (dev
->xmit_lock_owner
!= cpu
) {
1324 HARD_TX_LOCK(dev
, cpu
);
1326 if (!netif_queue_stopped(dev
)) {
1328 dev_queue_xmit_nit(skb
, dev
);
1331 if (!dev
->hard_start_xmit(skb
, dev
)) {
1332 HARD_TX_UNLOCK(dev
);
1336 HARD_TX_UNLOCK(dev
);
1337 if (net_ratelimit())
1338 printk(KERN_CRIT
"Virtual device %s asks to "
1339 "queue packet!\n", dev
->name
);
1341 /* Recursion is detected! It is possible,
1343 if (net_ratelimit())
1344 printk(KERN_CRIT
"Dead loop on virtual device "
1345 "%s, fix it urgently!\n", dev
->name
);
1361 /*=======================================================================
1363 =======================================================================*/
1365 int netdev_max_backlog
= 1000;
1366 int netdev_budget
= 300;
1367 int weight_p
= 64; /* old backlog weight */
1369 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1373 * netif_rx - post buffer to the network code
1374 * @skb: buffer to post
1376 * This function receives a packet from a device driver and queues it for
1377 * the upper (protocol) levels to process. It always succeeds. The buffer
1378 * may be dropped during processing for congestion control or by the
1382 * NET_RX_SUCCESS (no congestion)
1383 * NET_RX_CN_LOW (low congestion)
1384 * NET_RX_CN_MOD (moderate congestion)
1385 * NET_RX_CN_HIGH (high congestion)
1386 * NET_RX_DROP (packet was dropped)
1390 int netif_rx(struct sk_buff
*skb
)
1392 struct softnet_data
*queue
;
1393 unsigned long flags
;
1395 /* if netpoll wants it, pretend we never saw it */
1396 if (netpoll_rx(skb
))
1399 if (!skb
->tstamp
.off_sec
)
1403 * The code is rearranged so that the path is the most
1404 * short when CPU is congested, but is still operating.
1406 local_irq_save(flags
);
1407 queue
= &__get_cpu_var(softnet_data
);
1409 __get_cpu_var(netdev_rx_stat
).total
++;
1410 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1411 if (queue
->input_pkt_queue
.qlen
) {
1414 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1415 local_irq_restore(flags
);
1416 return NET_RX_SUCCESS
;
1419 netif_rx_schedule(&queue
->backlog_dev
);
1423 __get_cpu_var(netdev_rx_stat
).dropped
++;
1424 local_irq_restore(flags
);
1430 int netif_rx_ni(struct sk_buff
*skb
)
1435 err
= netif_rx(skb
);
1436 if (local_softirq_pending())
1443 EXPORT_SYMBOL(netif_rx_ni
);
1445 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1447 struct net_device
*dev
= skb
->dev
;
1450 skb
->dev
= dev
->master
;
1455 static void net_tx_action(struct softirq_action
*h
)
1457 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1459 if (sd
->completion_queue
) {
1460 struct sk_buff
*clist
;
1462 local_irq_disable();
1463 clist
= sd
->completion_queue
;
1464 sd
->completion_queue
= NULL
;
1468 struct sk_buff
*skb
= clist
;
1469 clist
= clist
->next
;
1471 BUG_TRAP(!atomic_read(&skb
->users
));
1476 if (sd
->output_queue
) {
1477 struct net_device
*head
;
1479 local_irq_disable();
1480 head
= sd
->output_queue
;
1481 sd
->output_queue
= NULL
;
1485 struct net_device
*dev
= head
;
1486 head
= head
->next_sched
;
1488 smp_mb__before_clear_bit();
1489 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1491 if (spin_trylock(&dev
->queue_lock
)) {
1493 spin_unlock(&dev
->queue_lock
);
1495 netif_schedule(dev
);
1501 static __inline__
int deliver_skb(struct sk_buff
*skb
,
1502 struct packet_type
*pt_prev
,
1503 struct net_device
*orig_dev
)
1505 atomic_inc(&skb
->users
);
1506 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1509 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1510 int (*br_handle_frame_hook
)(struct net_bridge_port
*p
, struct sk_buff
**pskb
);
1512 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1513 unsigned char *addr
);
1514 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
);
1516 static __inline__
int handle_bridge(struct sk_buff
**pskb
,
1517 struct packet_type
**pt_prev
, int *ret
,
1518 struct net_device
*orig_dev
)
1520 struct net_bridge_port
*port
;
1522 if ((*pskb
)->pkt_type
== PACKET_LOOPBACK
||
1523 (port
= rcu_dereference((*pskb
)->dev
->br_port
)) == NULL
)
1527 *ret
= deliver_skb(*pskb
, *pt_prev
, orig_dev
);
1531 return br_handle_frame_hook(port
, pskb
);
1534 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1537 #ifdef CONFIG_NET_CLS_ACT
1538 /* TODO: Maybe we should just force sch_ingress to be compiled in
1539 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1540 * a compare and 2 stores extra right now if we dont have it on
1541 * but have CONFIG_NET_CLS_ACT
1542 * NOTE: This doesnt stop any functionality; if you dont have
1543 * the ingress scheduler, you just cant add policies on ingress.
1546 static int ing_filter(struct sk_buff
*skb
)
1549 struct net_device
*dev
= skb
->dev
;
1550 int result
= TC_ACT_OK
;
1552 if (dev
->qdisc_ingress
) {
1553 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1554 if (MAX_RED_LOOP
< ttl
++) {
1555 printk("Redir loop detected Dropping packet (%s->%s)\n",
1556 skb
->input_dev
->name
, skb
->dev
->name
);
1560 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1562 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1564 spin_lock(&dev
->ingress_lock
);
1565 if ((q
= dev
->qdisc_ingress
) != NULL
)
1566 result
= q
->enqueue(skb
, q
);
1567 spin_unlock(&dev
->ingress_lock
);
1575 int netif_receive_skb(struct sk_buff
*skb
)
1577 struct packet_type
*ptype
, *pt_prev
;
1578 struct net_device
*orig_dev
;
1579 int ret
= NET_RX_DROP
;
1580 unsigned short type
;
1582 /* if we've gotten here through NAPI, check netpoll */
1583 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1586 if (!skb
->tstamp
.off_sec
)
1589 if (!skb
->input_dev
)
1590 skb
->input_dev
= skb
->dev
;
1592 orig_dev
= skb_bond(skb
);
1594 __get_cpu_var(netdev_rx_stat
).total
++;
1596 skb
->h
.raw
= skb
->nh
.raw
= skb
->data
;
1597 skb
->mac_len
= skb
->nh
.raw
- skb
->mac
.raw
;
1603 #ifdef CONFIG_NET_CLS_ACT
1604 if (skb
->tc_verd
& TC_NCLS
) {
1605 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1610 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1611 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1613 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1618 #ifdef CONFIG_NET_CLS_ACT
1620 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1621 pt_prev
= NULL
; /* noone else should process this after*/
1623 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1626 ret
= ing_filter(skb
);
1628 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
1637 handle_diverter(skb
);
1639 if (handle_bridge(&skb
, &pt_prev
, &ret
, orig_dev
))
1642 type
= skb
->protocol
;
1643 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
1644 if (ptype
->type
== type
&&
1645 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
1647 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1653 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1656 /* Jamal, now you will not able to escape explaining
1657 * me how you were going to use this. :-)
1667 static int process_backlog(struct net_device
*backlog_dev
, int *budget
)
1670 int quota
= min(backlog_dev
->quota
, *budget
);
1671 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1672 unsigned long start_time
= jiffies
;
1674 backlog_dev
->weight
= weight_p
;
1676 struct sk_buff
*skb
;
1677 struct net_device
*dev
;
1679 local_irq_disable();
1680 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
1687 netif_receive_skb(skb
);
1693 if (work
>= quota
|| jiffies
- start_time
> 1)
1698 backlog_dev
->quota
-= work
;
1703 backlog_dev
->quota
-= work
;
1706 list_del(&backlog_dev
->poll_list
);
1707 smp_mb__before_clear_bit();
1708 netif_poll_enable(backlog_dev
);
1714 static void net_rx_action(struct softirq_action
*h
)
1716 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1717 unsigned long start_time
= jiffies
;
1718 int budget
= netdev_budget
;
1721 local_irq_disable();
1723 while (!list_empty(&queue
->poll_list
)) {
1724 struct net_device
*dev
;
1726 if (budget
<= 0 || jiffies
- start_time
> 1)
1731 dev
= list_entry(queue
->poll_list
.next
,
1732 struct net_device
, poll_list
);
1733 have
= netpoll_poll_lock(dev
);
1735 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
1736 netpoll_poll_unlock(have
);
1737 local_irq_disable();
1738 list_del(&dev
->poll_list
);
1739 list_add_tail(&dev
->poll_list
, &queue
->poll_list
);
1741 dev
->quota
+= dev
->weight
;
1743 dev
->quota
= dev
->weight
;
1745 netpoll_poll_unlock(have
);
1747 local_irq_disable();
1755 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
1756 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1760 static gifconf_func_t
* gifconf_list
[NPROTO
];
1763 * register_gifconf - register a SIOCGIF handler
1764 * @family: Address family
1765 * @gifconf: Function handler
1767 * Register protocol dependent address dumping routines. The handler
1768 * that is passed must not be freed or reused until it has been replaced
1769 * by another handler.
1771 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
1773 if (family
>= NPROTO
)
1775 gifconf_list
[family
] = gifconf
;
1781 * Map an interface index to its name (SIOCGIFNAME)
1785 * We need this ioctl for efficient implementation of the
1786 * if_indextoname() function required by the IPv6 API. Without
1787 * it, we would have to search all the interfaces to find a
1791 static int dev_ifname(struct ifreq __user
*arg
)
1793 struct net_device
*dev
;
1797 * Fetch the caller's info block.
1800 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
1803 read_lock(&dev_base_lock
);
1804 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
1806 read_unlock(&dev_base_lock
);
1810 strcpy(ifr
.ifr_name
, dev
->name
);
1811 read_unlock(&dev_base_lock
);
1813 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
1819 * Perform a SIOCGIFCONF call. This structure will change
1820 * size eventually, and there is nothing I can do about it.
1821 * Thus we will need a 'compatibility mode'.
1824 static int dev_ifconf(char __user
*arg
)
1827 struct net_device
*dev
;
1834 * Fetch the caller's info block.
1837 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
1844 * Loop over the interfaces, and write an info block for each.
1848 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
1849 for (i
= 0; i
< NPROTO
; i
++) {
1850 if (gifconf_list
[i
]) {
1853 done
= gifconf_list
[i
](dev
, NULL
, 0);
1855 done
= gifconf_list
[i
](dev
, pos
+ total
,
1865 * All done. Write the updated control block back to the caller.
1867 ifc
.ifc_len
= total
;
1870 * Both BSD and Solaris return 0 here, so we do too.
1872 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
1875 #ifdef CONFIG_PROC_FS
1877 * This is invoked by the /proc filesystem handler to display a device
1880 static __inline__
struct net_device
*dev_get_idx(loff_t pos
)
1882 struct net_device
*dev
;
1885 for (i
= 0, dev
= dev_base
; dev
&& i
< pos
; ++i
, dev
= dev
->next
);
1887 return i
== pos
? dev
: NULL
;
1890 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1892 read_lock(&dev_base_lock
);
1893 return *pos
? dev_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
1896 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1899 return v
== SEQ_START_TOKEN
? dev_base
: ((struct net_device
*)v
)->next
;
1902 void dev_seq_stop(struct seq_file
*seq
, void *v
)
1904 read_unlock(&dev_base_lock
);
1907 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
1909 if (dev
->get_stats
) {
1910 struct net_device_stats
*stats
= dev
->get_stats(dev
);
1912 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1913 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1914 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
1916 stats
->rx_dropped
+ stats
->rx_missed_errors
,
1917 stats
->rx_fifo_errors
,
1918 stats
->rx_length_errors
+ stats
->rx_over_errors
+
1919 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
1920 stats
->rx_compressed
, stats
->multicast
,
1921 stats
->tx_bytes
, stats
->tx_packets
,
1922 stats
->tx_errors
, stats
->tx_dropped
,
1923 stats
->tx_fifo_errors
, stats
->collisions
,
1924 stats
->tx_carrier_errors
+
1925 stats
->tx_aborted_errors
+
1926 stats
->tx_window_errors
+
1927 stats
->tx_heartbeat_errors
,
1928 stats
->tx_compressed
);
1930 seq_printf(seq
, "%6s: No statistics available.\n", dev
->name
);
1934 * Called from the PROCfs module. This now uses the new arbitrary sized
1935 * /proc/net interface to create /proc/net/dev
1937 static int dev_seq_show(struct seq_file
*seq
, void *v
)
1939 if (v
== SEQ_START_TOKEN
)
1940 seq_puts(seq
, "Inter-| Receive "
1942 " face |bytes packets errs drop fifo frame "
1943 "compressed multicast|bytes packets errs "
1944 "drop fifo colls carrier compressed\n");
1946 dev_seq_printf_stats(seq
, v
);
1950 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
1952 struct netif_rx_stats
*rc
= NULL
;
1954 while (*pos
< NR_CPUS
)
1955 if (cpu_online(*pos
)) {
1956 rc
= &per_cpu(netdev_rx_stat
, *pos
);
1963 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1965 return softnet_get_online(pos
);
1968 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1971 return softnet_get_online(pos
);
1974 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
1978 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
1980 struct netif_rx_stats
*s
= v
;
1982 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
1983 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
1984 0, 0, 0, 0, /* was fastroute */
1989 static struct seq_operations dev_seq_ops
= {
1990 .start
= dev_seq_start
,
1991 .next
= dev_seq_next
,
1992 .stop
= dev_seq_stop
,
1993 .show
= dev_seq_show
,
1996 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
1998 return seq_open(file
, &dev_seq_ops
);
2001 static struct file_operations dev_seq_fops
= {
2002 .owner
= THIS_MODULE
,
2003 .open
= dev_seq_open
,
2005 .llseek
= seq_lseek
,
2006 .release
= seq_release
,
2009 static struct seq_operations softnet_seq_ops
= {
2010 .start
= softnet_seq_start
,
2011 .next
= softnet_seq_next
,
2012 .stop
= softnet_seq_stop
,
2013 .show
= softnet_seq_show
,
2016 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2018 return seq_open(file
, &softnet_seq_ops
);
2021 static struct file_operations softnet_seq_fops
= {
2022 .owner
= THIS_MODULE
,
2023 .open
= softnet_seq_open
,
2025 .llseek
= seq_lseek
,
2026 .release
= seq_release
,
2029 #ifdef CONFIG_WIRELESS_EXT
2030 extern int wireless_proc_init(void);
2032 #define wireless_proc_init() 0
2035 static int __init
dev_proc_init(void)
2039 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2041 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2043 if (wireless_proc_init())
2049 proc_net_remove("softnet_stat");
2051 proc_net_remove("dev");
2055 #define dev_proc_init() 0
2056 #endif /* CONFIG_PROC_FS */
2060 * netdev_set_master - set up master/slave pair
2061 * @slave: slave device
2062 * @master: new master device
2064 * Changes the master device of the slave. Pass %NULL to break the
2065 * bonding. The caller must hold the RTNL semaphore. On a failure
2066 * a negative errno code is returned. On success the reference counts
2067 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2068 * function returns zero.
2070 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2072 struct net_device
*old
= slave
->master
;
2082 slave
->master
= master
;
2090 slave
->flags
|= IFF_SLAVE
;
2092 slave
->flags
&= ~IFF_SLAVE
;
2094 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2099 * dev_set_promiscuity - update promiscuity count on a device
2103 * Add or remove promsicuity from a device. While the count in the device
2104 * remains above zero the interface remains promiscuous. Once it hits zero
2105 * the device reverts back to normal filtering operation. A negative inc
2106 * value is used to drop promiscuity on the device.
2108 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2110 unsigned short old_flags
= dev
->flags
;
2112 if ((dev
->promiscuity
+= inc
) == 0)
2113 dev
->flags
&= ~IFF_PROMISC
;
2115 dev
->flags
|= IFF_PROMISC
;
2116 if (dev
->flags
!= old_flags
) {
2118 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2119 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2125 * dev_set_allmulti - update allmulti count on a device
2129 * Add or remove reception of all multicast frames to a device. While the
2130 * count in the device remains above zero the interface remains listening
2131 * to all interfaces. Once it hits zero the device reverts back to normal
2132 * filtering operation. A negative @inc value is used to drop the counter
2133 * when releasing a resource needing all multicasts.
2136 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2138 unsigned short old_flags
= dev
->flags
;
2140 dev
->flags
|= IFF_ALLMULTI
;
2141 if ((dev
->allmulti
+= inc
) == 0)
2142 dev
->flags
&= ~IFF_ALLMULTI
;
2143 if (dev
->flags
^ old_flags
)
2147 unsigned dev_get_flags(const struct net_device
*dev
)
2151 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2154 (dev
->gflags
& (IFF_PROMISC
|
2157 if (netif_running(dev
) && netif_carrier_ok(dev
))
2158 flags
|= IFF_RUNNING
;
2163 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2166 int old_flags
= dev
->flags
;
2169 * Set the flags on our device.
2172 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2173 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2175 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2179 * Load in the correct multicast list now the flags have changed.
2185 * Have we downed the interface. We handle IFF_UP ourselves
2186 * according to user attempts to set it, rather than blindly
2191 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2192 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2198 if (dev
->flags
& IFF_UP
&&
2199 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2201 notifier_call_chain(&netdev_chain
, NETDEV_CHANGE
, dev
);
2203 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2204 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2205 dev
->gflags
^= IFF_PROMISC
;
2206 dev_set_promiscuity(dev
, inc
);
2209 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2210 is important. Some (broken) drivers set IFF_PROMISC, when
2211 IFF_ALLMULTI is requested not asking us and not reporting.
2213 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2214 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2215 dev
->gflags
^= IFF_ALLMULTI
;
2216 dev_set_allmulti(dev
, inc
);
2219 if (old_flags
^ dev
->flags
)
2220 rtmsg_ifinfo(RTM_NEWLINK
, dev
, old_flags
^ dev
->flags
);
2225 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2229 if (new_mtu
== dev
->mtu
)
2232 /* MTU must be positive. */
2236 if (!netif_device_present(dev
))
2240 if (dev
->change_mtu
)
2241 err
= dev
->change_mtu(dev
, new_mtu
);
2244 if (!err
&& dev
->flags
& IFF_UP
)
2245 notifier_call_chain(&netdev_chain
,
2246 NETDEV_CHANGEMTU
, dev
);
2250 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2254 if (!dev
->set_mac_address
)
2256 if (sa
->sa_family
!= dev
->type
)
2258 if (!netif_device_present(dev
))
2260 err
= dev
->set_mac_address(dev
, sa
);
2262 notifier_call_chain(&netdev_chain
, NETDEV_CHANGEADDR
, dev
);
2267 * Perform the SIOCxIFxxx calls.
2269 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2272 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2278 case SIOCGIFFLAGS
: /* Get interface flags */
2279 ifr
->ifr_flags
= dev_get_flags(dev
);
2282 case SIOCSIFFLAGS
: /* Set interface flags */
2283 return dev_change_flags(dev
, ifr
->ifr_flags
);
2285 case SIOCGIFMETRIC
: /* Get the metric on the interface
2286 (currently unused) */
2287 ifr
->ifr_metric
= 0;
2290 case SIOCSIFMETRIC
: /* Set the metric on the interface
2291 (currently unused) */
2294 case SIOCGIFMTU
: /* Get the MTU of a device */
2295 ifr
->ifr_mtu
= dev
->mtu
;
2298 case SIOCSIFMTU
: /* Set the MTU of a device */
2299 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
2303 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
2305 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
2306 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2307 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
2311 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
2313 case SIOCSIFHWBROADCAST
:
2314 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
2316 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
2317 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2318 notifier_call_chain(&netdev_chain
,
2319 NETDEV_CHANGEADDR
, dev
);
2323 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
2324 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
2325 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
2326 ifr
->ifr_map
.irq
= dev
->irq
;
2327 ifr
->ifr_map
.dma
= dev
->dma
;
2328 ifr
->ifr_map
.port
= dev
->if_port
;
2332 if (dev
->set_config
) {
2333 if (!netif_device_present(dev
))
2335 return dev
->set_config(dev
, &ifr
->ifr_map
);
2340 if (!dev
->set_multicast_list
||
2341 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2343 if (!netif_device_present(dev
))
2345 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
2349 if (!dev
->set_multicast_list
||
2350 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2352 if (!netif_device_present(dev
))
2354 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
2358 ifr
->ifr_ifindex
= dev
->ifindex
;
2362 ifr
->ifr_qlen
= dev
->tx_queue_len
;
2366 if (ifr
->ifr_qlen
< 0)
2368 dev
->tx_queue_len
= ifr
->ifr_qlen
;
2372 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
2373 return dev_change_name(dev
, ifr
->ifr_newname
);
2376 * Unknown or private ioctl
2380 if ((cmd
>= SIOCDEVPRIVATE
&&
2381 cmd
<= SIOCDEVPRIVATE
+ 15) ||
2382 cmd
== SIOCBONDENSLAVE
||
2383 cmd
== SIOCBONDRELEASE
||
2384 cmd
== SIOCBONDSETHWADDR
||
2385 cmd
== SIOCBONDSLAVEINFOQUERY
||
2386 cmd
== SIOCBONDINFOQUERY
||
2387 cmd
== SIOCBONDCHANGEACTIVE
||
2388 cmd
== SIOCGMIIPHY
||
2389 cmd
== SIOCGMIIREG
||
2390 cmd
== SIOCSMIIREG
||
2391 cmd
== SIOCBRADDIF
||
2392 cmd
== SIOCBRDELIF
||
2393 cmd
== SIOCWANDEV
) {
2395 if (dev
->do_ioctl
) {
2396 if (netif_device_present(dev
))
2397 err
= dev
->do_ioctl(dev
, ifr
,
2410 * This function handles all "interface"-type I/O control requests. The actual
2411 * 'doing' part of this is dev_ifsioc above.
2415 * dev_ioctl - network device ioctl
2416 * @cmd: command to issue
2417 * @arg: pointer to a struct ifreq in user space
2419 * Issue ioctl functions to devices. This is normally called by the
2420 * user space syscall interfaces but can sometimes be useful for
2421 * other purposes. The return value is the return from the syscall if
2422 * positive or a negative errno code on error.
2425 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
2431 /* One special case: SIOCGIFCONF takes ifconf argument
2432 and requires shared lock, because it sleeps writing
2436 if (cmd
== SIOCGIFCONF
) {
2438 ret
= dev_ifconf((char __user
*) arg
);
2442 if (cmd
== SIOCGIFNAME
)
2443 return dev_ifname((struct ifreq __user
*)arg
);
2445 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2448 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
2450 colon
= strchr(ifr
.ifr_name
, ':');
2455 * See which interface the caller is talking about.
2460 * These ioctl calls:
2461 * - can be done by all.
2462 * - atomic and do not require locking.
2473 dev_load(ifr
.ifr_name
);
2474 read_lock(&dev_base_lock
);
2475 ret
= dev_ifsioc(&ifr
, cmd
);
2476 read_unlock(&dev_base_lock
);
2480 if (copy_to_user(arg
, &ifr
,
2481 sizeof(struct ifreq
)))
2487 dev_load(ifr
.ifr_name
);
2489 ret
= dev_ethtool(&ifr
);
2494 if (copy_to_user(arg
, &ifr
,
2495 sizeof(struct ifreq
)))
2501 * These ioctl calls:
2502 * - require superuser power.
2503 * - require strict serialization.
2509 if (!capable(CAP_NET_ADMIN
))
2511 dev_load(ifr
.ifr_name
);
2513 ret
= dev_ifsioc(&ifr
, cmd
);
2518 if (copy_to_user(arg
, &ifr
,
2519 sizeof(struct ifreq
)))
2525 * These ioctl calls:
2526 * - require superuser power.
2527 * - require strict serialization.
2528 * - do not return a value
2538 case SIOCSIFHWBROADCAST
:
2541 case SIOCBONDENSLAVE
:
2542 case SIOCBONDRELEASE
:
2543 case SIOCBONDSETHWADDR
:
2544 case SIOCBONDCHANGEACTIVE
:
2547 if (!capable(CAP_NET_ADMIN
))
2550 case SIOCBONDSLAVEINFOQUERY
:
2551 case SIOCBONDINFOQUERY
:
2552 dev_load(ifr
.ifr_name
);
2554 ret
= dev_ifsioc(&ifr
, cmd
);
2559 /* Get the per device memory space. We can add this but
2560 * currently do not support it */
2562 /* Set the per device memory buffer space.
2563 * Not applicable in our case */
2568 * Unknown or private ioctl.
2571 if (cmd
== SIOCWANDEV
||
2572 (cmd
>= SIOCDEVPRIVATE
&&
2573 cmd
<= SIOCDEVPRIVATE
+ 15)) {
2574 dev_load(ifr
.ifr_name
);
2576 ret
= dev_ifsioc(&ifr
, cmd
);
2578 if (!ret
&& copy_to_user(arg
, &ifr
,
2579 sizeof(struct ifreq
)))
2583 #ifdef CONFIG_WIRELESS_EXT
2584 /* Take care of Wireless Extensions */
2585 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
) {
2586 /* If command is `set a parameter', or
2587 * `get the encoding parameters', check if
2588 * the user has the right to do it */
2589 if (IW_IS_SET(cmd
) || cmd
== SIOCGIWENCODE
) {
2590 if (!capable(CAP_NET_ADMIN
))
2593 dev_load(ifr
.ifr_name
);
2595 /* Follow me in net/core/wireless.c */
2596 ret
= wireless_process_ioctl(&ifr
, cmd
);
2598 if (IW_IS_GET(cmd
) &&
2599 copy_to_user(arg
, &ifr
,
2600 sizeof(struct ifreq
)))
2604 #endif /* CONFIG_WIRELESS_EXT */
2611 * dev_new_index - allocate an ifindex
2613 * Returns a suitable unique value for a new device interface
2614 * number. The caller must hold the rtnl semaphore or the
2615 * dev_base_lock to be sure it remains unique.
2617 static int dev_new_index(void)
2623 if (!__dev_get_by_index(ifindex
))
2628 static int dev_boot_phase
= 1;
2630 /* Delayed registration/unregisteration */
2631 static DEFINE_SPINLOCK(net_todo_list_lock
);
2632 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
2634 static inline void net_set_todo(struct net_device
*dev
)
2636 spin_lock(&net_todo_list_lock
);
2637 list_add_tail(&dev
->todo_list
, &net_todo_list
);
2638 spin_unlock(&net_todo_list_lock
);
2642 * register_netdevice - register a network device
2643 * @dev: device to register
2645 * Take a completed network device structure and add it to the kernel
2646 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2647 * chain. 0 is returned on success. A negative errno code is returned
2648 * on a failure to set up the device, or if the name is a duplicate.
2650 * Callers must hold the rtnl semaphore. You may want
2651 * register_netdev() instead of this.
2654 * The locking appears insufficient to guarantee two parallel registers
2655 * will not get the same name.
2658 int register_netdevice(struct net_device
*dev
)
2660 struct hlist_head
*head
;
2661 struct hlist_node
*p
;
2664 BUG_ON(dev_boot_phase
);
2667 /* When net_device's are persistent, this will be fatal. */
2668 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
2670 spin_lock_init(&dev
->queue_lock
);
2671 spin_lock_init(&dev
->xmit_lock
);
2672 dev
->xmit_lock_owner
= -1;
2673 #ifdef CONFIG_NET_CLS_ACT
2674 spin_lock_init(&dev
->ingress_lock
);
2677 ret
= alloc_divert_blk(dev
);
2683 /* Init, if this function is available */
2685 ret
= dev
->init(dev
);
2693 if (!dev_valid_name(dev
->name
)) {
2698 dev
->ifindex
= dev_new_index();
2699 if (dev
->iflink
== -1)
2700 dev
->iflink
= dev
->ifindex
;
2702 /* Check for existence of name */
2703 head
= dev_name_hash(dev
->name
);
2704 hlist_for_each(p
, head
) {
2705 struct net_device
*d
2706 = hlist_entry(p
, struct net_device
, name_hlist
);
2707 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
2713 /* Fix illegal SG+CSUM combinations. */
2714 if ((dev
->features
& NETIF_F_SG
) &&
2715 !(dev
->features
& (NETIF_F_IP_CSUM
|
2717 NETIF_F_HW_CSUM
))) {
2718 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2720 dev
->features
&= ~NETIF_F_SG
;
2723 /* TSO requires that SG is present as well. */
2724 if ((dev
->features
& NETIF_F_TSO
) &&
2725 !(dev
->features
& NETIF_F_SG
)) {
2726 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2728 dev
->features
&= ~NETIF_F_TSO
;
2730 if (dev
->features
& NETIF_F_UFO
) {
2731 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
2732 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
2733 "NETIF_F_HW_CSUM feature.\n",
2735 dev
->features
&= ~NETIF_F_UFO
;
2737 if (!(dev
->features
& NETIF_F_SG
)) {
2738 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
2739 "NETIF_F_SG feature.\n",
2741 dev
->features
&= ~NETIF_F_UFO
;
2746 * nil rebuild_header routine,
2747 * that should be never called and used as just bug trap.
2750 if (!dev
->rebuild_header
)
2751 dev
->rebuild_header
= default_rebuild_header
;
2754 * Default initial state at registry is that the
2755 * device is present.
2758 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
2761 dev_init_scheduler(dev
);
2762 write_lock_bh(&dev_base_lock
);
2764 dev_tail
= &dev
->next
;
2765 hlist_add_head(&dev
->name_hlist
, head
);
2766 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
2768 dev
->reg_state
= NETREG_REGISTERING
;
2769 write_unlock_bh(&dev_base_lock
);
2771 /* Notify protocols, that a new device appeared. */
2772 notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
2774 /* Finish registration after unlock */
2781 free_divert_blk(dev
);
2786 * register_netdev - register a network device
2787 * @dev: device to register
2789 * Take a completed network device structure and add it to the kernel
2790 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2791 * chain. 0 is returned on success. A negative errno code is returned
2792 * on a failure to set up the device, or if the name is a duplicate.
2794 * This is a wrapper around register_netdev that takes the rtnl semaphore
2795 * and expands the device name if you passed a format string to
2798 int register_netdev(struct net_device
*dev
)
2805 * If the name is a format string the caller wants us to do a
2808 if (strchr(dev
->name
, '%')) {
2809 err
= dev_alloc_name(dev
, dev
->name
);
2815 * Back compatibility hook. Kill this one in 2.5
2817 if (dev
->name
[0] == 0 || dev
->name
[0] == ' ') {
2818 err
= dev_alloc_name(dev
, "eth%d");
2823 err
= register_netdevice(dev
);
2828 EXPORT_SYMBOL(register_netdev
);
2831 * netdev_wait_allrefs - wait until all references are gone.
2833 * This is called when unregistering network devices.
2835 * Any protocol or device that holds a reference should register
2836 * for netdevice notification, and cleanup and put back the
2837 * reference if they receive an UNREGISTER event.
2838 * We can get stuck here if buggy protocols don't correctly
2841 static void netdev_wait_allrefs(struct net_device
*dev
)
2843 unsigned long rebroadcast_time
, warning_time
;
2845 rebroadcast_time
= warning_time
= jiffies
;
2846 while (atomic_read(&dev
->refcnt
) != 0) {
2847 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
2850 /* Rebroadcast unregister notification */
2851 notifier_call_chain(&netdev_chain
,
2852 NETDEV_UNREGISTER
, dev
);
2854 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
2856 /* We must not have linkwatch events
2857 * pending on unregister. If this
2858 * happens, we simply run the queue
2859 * unscheduled, resulting in a noop
2862 linkwatch_run_queue();
2867 rebroadcast_time
= jiffies
;
2872 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
2873 printk(KERN_EMERG
"unregister_netdevice: "
2874 "waiting for %s to become free. Usage "
2876 dev
->name
, atomic_read(&dev
->refcnt
));
2877 warning_time
= jiffies
;
2886 * register_netdevice(x1);
2887 * register_netdevice(x2);
2889 * unregister_netdevice(y1);
2890 * unregister_netdevice(y2);
2896 * We are invoked by rtnl_unlock() after it drops the semaphore.
2897 * This allows us to deal with problems:
2898 * 1) We can create/delete sysfs objects which invoke hotplug
2899 * without deadlocking with linkwatch via keventd.
2900 * 2) Since we run with the RTNL semaphore not held, we can sleep
2901 * safely in order to wait for the netdev refcnt to drop to zero.
2903 static DECLARE_MUTEX(net_todo_run_mutex
);
2904 void netdev_run_todo(void)
2906 struct list_head list
= LIST_HEAD_INIT(list
);
2910 /* Need to guard against multiple cpu's getting out of order. */
2911 down(&net_todo_run_mutex
);
2913 /* Not safe to do outside the semaphore. We must not return
2914 * until all unregister events invoked by the local processor
2915 * have been completed (either by this todo run, or one on
2918 if (list_empty(&net_todo_list
))
2921 /* Snapshot list, allow later requests */
2922 spin_lock(&net_todo_list_lock
);
2923 list_splice_init(&net_todo_list
, &list
);
2924 spin_unlock(&net_todo_list_lock
);
2926 while (!list_empty(&list
)) {
2927 struct net_device
*dev
2928 = list_entry(list
.next
, struct net_device
, todo_list
);
2929 list_del(&dev
->todo_list
);
2931 switch(dev
->reg_state
) {
2932 case NETREG_REGISTERING
:
2933 err
= netdev_register_sysfs(dev
);
2935 printk(KERN_ERR
"%s: failed sysfs registration (%d)\n",
2937 dev
->reg_state
= NETREG_REGISTERED
;
2940 case NETREG_UNREGISTERING
:
2941 netdev_unregister_sysfs(dev
);
2942 dev
->reg_state
= NETREG_UNREGISTERED
;
2944 netdev_wait_allrefs(dev
);
2947 BUG_ON(atomic_read(&dev
->refcnt
));
2948 BUG_TRAP(!dev
->ip_ptr
);
2949 BUG_TRAP(!dev
->ip6_ptr
);
2950 BUG_TRAP(!dev
->dn_ptr
);
2953 /* It must be the very last action,
2954 * after this 'dev' may point to freed up memory.
2956 if (dev
->destructor
)
2957 dev
->destructor(dev
);
2961 printk(KERN_ERR
"network todo '%s' but state %d\n",
2962 dev
->name
, dev
->reg_state
);
2968 up(&net_todo_run_mutex
);
2972 * alloc_netdev - allocate network device
2973 * @sizeof_priv: size of private data to allocate space for
2974 * @name: device name format string
2975 * @setup: callback to initialize device
2977 * Allocates a struct net_device with private data area for driver use
2978 * and performs basic initialization.
2980 struct net_device
*alloc_netdev(int sizeof_priv
, const char *name
,
2981 void (*setup
)(struct net_device
*))
2984 struct net_device
*dev
;
2987 /* ensure 32-byte alignment of both the device and private area */
2988 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
2989 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
2991 p
= kmalloc(alloc_size
, GFP_KERNEL
);
2993 printk(KERN_ERR
"alloc_dev: Unable to allocate device.\n");
2996 memset(p
, 0, alloc_size
);
2998 dev
= (struct net_device
*)
2999 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3000 dev
->padded
= (char *)dev
- (char *)p
;
3003 dev
->priv
= netdev_priv(dev
);
3006 strcpy(dev
->name
, name
);
3009 EXPORT_SYMBOL(alloc_netdev
);
3012 * free_netdev - free network device
3015 * This function does the last stage of destroying an allocated device
3016 * interface. The reference to the device object is released.
3017 * If this is the last reference then it will be freed.
3019 void free_netdev(struct net_device
*dev
)
3022 /* Compatiablity with error handling in drivers */
3023 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3024 kfree((char *)dev
- dev
->padded
);
3028 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3029 dev
->reg_state
= NETREG_RELEASED
;
3031 /* will free via class release */
3032 class_device_put(&dev
->class_dev
);
3034 kfree((char *)dev
- dev
->padded
);
3038 /* Synchronize with packet receive processing. */
3039 void synchronize_net(void)
3046 * unregister_netdevice - remove device from the kernel
3049 * This function shuts down a device interface and removes it
3050 * from the kernel tables. On success 0 is returned, on a failure
3051 * a negative errno code is returned.
3053 * Callers must hold the rtnl semaphore. You may want
3054 * unregister_netdev() instead of this.
3057 int unregister_netdevice(struct net_device
*dev
)
3059 struct net_device
*d
, **dp
;
3061 BUG_ON(dev_boot_phase
);
3064 /* Some devices call without registering for initialization unwind. */
3065 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3066 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3067 "was registered\n", dev
->name
, dev
);
3071 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3073 /* If device is running, close it first. */
3074 if (dev
->flags
& IFF_UP
)
3077 /* And unlink it from device chain. */
3078 for (dp
= &dev_base
; (d
= *dp
) != NULL
; dp
= &d
->next
) {
3080 write_lock_bh(&dev_base_lock
);
3081 hlist_del(&dev
->name_hlist
);
3082 hlist_del(&dev
->index_hlist
);
3083 if (dev_tail
== &dev
->next
)
3086 write_unlock_bh(&dev_base_lock
);
3091 printk(KERN_ERR
"unregister net_device: '%s' not found\n",
3096 dev
->reg_state
= NETREG_UNREGISTERING
;
3100 /* Shutdown queueing discipline. */
3104 /* Notify protocols, that we are about to destroy
3105 this device. They should clean all the things.
3107 notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3110 * Flush the multicast chain
3112 dev_mc_discard(dev
);
3117 /* Notifier chain MUST detach us from master device. */
3118 BUG_TRAP(!dev
->master
);
3120 free_divert_blk(dev
);
3122 /* Finish processing unregister after unlock */
3132 * unregister_netdev - remove device from the kernel
3135 * This function shuts down a device interface and removes it
3136 * from the kernel tables. On success 0 is returned, on a failure
3137 * a negative errno code is returned.
3139 * This is just a wrapper for unregister_netdevice that takes
3140 * the rtnl semaphore. In general you want to use this and not
3141 * unregister_netdevice.
3143 void unregister_netdev(struct net_device
*dev
)
3146 unregister_netdevice(dev
);
3150 EXPORT_SYMBOL(unregister_netdev
);
3152 #ifdef CONFIG_HOTPLUG_CPU
3153 static int dev_cpu_callback(struct notifier_block
*nfb
,
3154 unsigned long action
,
3157 struct sk_buff
**list_skb
;
3158 struct net_device
**list_net
;
3159 struct sk_buff
*skb
;
3160 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3161 struct softnet_data
*sd
, *oldsd
;
3163 if (action
!= CPU_DEAD
)
3166 local_irq_disable();
3167 cpu
= smp_processor_id();
3168 sd
= &per_cpu(softnet_data
, cpu
);
3169 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3171 /* Find end of our completion_queue. */
3172 list_skb
= &sd
->completion_queue
;
3174 list_skb
= &(*list_skb
)->next
;
3175 /* Append completion queue from offline CPU. */
3176 *list_skb
= oldsd
->completion_queue
;
3177 oldsd
->completion_queue
= NULL
;
3179 /* Find end of our output_queue. */
3180 list_net
= &sd
->output_queue
;
3182 list_net
= &(*list_net
)->next_sched
;
3183 /* Append output queue from offline CPU. */
3184 *list_net
= oldsd
->output_queue
;
3185 oldsd
->output_queue
= NULL
;
3187 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3190 /* Process offline CPU's input_pkt_queue */
3191 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3196 #endif /* CONFIG_HOTPLUG_CPU */
3200 * Initialize the DEV module. At boot time this walks the device list and
3201 * unhooks any devices that fail to initialise (normally hardware not
3202 * present) and leaves us with a valid list of present and active devices.
3207 * This is called single threaded during boot, so no need
3208 * to take the rtnl semaphore.
3210 static int __init
net_dev_init(void)
3212 int i
, rc
= -ENOMEM
;
3214 BUG_ON(!dev_boot_phase
);
3218 if (dev_proc_init())
3221 if (netdev_sysfs_init())
3224 INIT_LIST_HEAD(&ptype_all
);
3225 for (i
= 0; i
< 16; i
++)
3226 INIT_LIST_HEAD(&ptype_base
[i
]);
3228 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
3229 INIT_HLIST_HEAD(&dev_name_head
[i
]);
3231 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
3232 INIT_HLIST_HEAD(&dev_index_head
[i
]);
3235 * Initialise the packet receive queues.
3239 struct softnet_data
*queue
;
3241 queue
= &per_cpu(softnet_data
, i
);
3242 skb_queue_head_init(&queue
->input_pkt_queue
);
3243 queue
->completion_queue
= NULL
;
3244 INIT_LIST_HEAD(&queue
->poll_list
);
3245 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
3246 queue
->backlog_dev
.weight
= weight_p
;
3247 queue
->backlog_dev
.poll
= process_backlog
;
3248 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
3253 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
3254 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
3256 hotcpu_notifier(dev_cpu_callback
, 0);
3264 subsys_initcall(net_dev_init
);
3266 EXPORT_SYMBOL(__dev_get_by_index
);
3267 EXPORT_SYMBOL(__dev_get_by_name
);
3268 EXPORT_SYMBOL(__dev_remove_pack
);
3269 EXPORT_SYMBOL(__skb_linearize
);
3270 EXPORT_SYMBOL(dev_valid_name
);
3271 EXPORT_SYMBOL(dev_add_pack
);
3272 EXPORT_SYMBOL(dev_alloc_name
);
3273 EXPORT_SYMBOL(dev_close
);
3274 EXPORT_SYMBOL(dev_get_by_flags
);
3275 EXPORT_SYMBOL(dev_get_by_index
);
3276 EXPORT_SYMBOL(dev_get_by_name
);
3277 EXPORT_SYMBOL(dev_open
);
3278 EXPORT_SYMBOL(dev_queue_xmit
);
3279 EXPORT_SYMBOL(dev_remove_pack
);
3280 EXPORT_SYMBOL(dev_set_allmulti
);
3281 EXPORT_SYMBOL(dev_set_promiscuity
);
3282 EXPORT_SYMBOL(dev_change_flags
);
3283 EXPORT_SYMBOL(dev_set_mtu
);
3284 EXPORT_SYMBOL(dev_set_mac_address
);
3285 EXPORT_SYMBOL(free_netdev
);
3286 EXPORT_SYMBOL(netdev_boot_setup_check
);
3287 EXPORT_SYMBOL(netdev_set_master
);
3288 EXPORT_SYMBOL(netdev_state_change
);
3289 EXPORT_SYMBOL(netif_receive_skb
);
3290 EXPORT_SYMBOL(netif_rx
);
3291 EXPORT_SYMBOL(register_gifconf
);
3292 EXPORT_SYMBOL(register_netdevice
);
3293 EXPORT_SYMBOL(register_netdevice_notifier
);
3294 EXPORT_SYMBOL(skb_checksum_help
);
3295 EXPORT_SYMBOL(synchronize_net
);
3296 EXPORT_SYMBOL(unregister_netdevice
);
3297 EXPORT_SYMBOL(unregister_netdevice_notifier
);
3298 EXPORT_SYMBOL(net_enable_timestamp
);
3299 EXPORT_SYMBOL(net_disable_timestamp
);
3300 EXPORT_SYMBOL(dev_get_flags
);
3302 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3303 EXPORT_SYMBOL(br_handle_frame_hook
);
3304 EXPORT_SYMBOL(br_fdb_get_hook
);
3305 EXPORT_SYMBOL(br_fdb_put_hook
);
3309 EXPORT_SYMBOL(dev_load
);
3312 EXPORT_PER_CPU_SYMBOL(softnet_data
);