2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/config.h>
80 #include <linux/cpu.h>
81 #include <linux/types.h>
82 #include <linux/kernel.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
97 #include <linux/rtnetlink.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <linux/stat.h>
101 #include <linux/if_bridge.h>
102 #include <linux/divert.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/kmod.h>
109 #include <linux/module.h>
110 #include <linux/kallsyms.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <linux/wireless.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
121 * The list of packet types we will receive (as opposed to discard)
122 * and the routines to invoke.
124 * Why 16. Because with 16 the only overlap we get on a hash of the
125 * low nibble of the protocol value is RARP/SNAP/X.25.
127 * NOTE: That is no longer true with the addition of VLAN tags. Not
128 * sure which should go first, but I bet it won't make much
129 * difference if we are running VLANs. The good news is that
130 * this protocol won't be in the list unless compiled in, so
131 * the average user (w/out VLANs) will not be adversely affected.
148 static DEFINE_SPINLOCK(ptype_lock
);
149 static struct list_head ptype_base
[16]; /* 16 way hashed list */
150 static struct list_head ptype_all
; /* Taps */
152 #ifdef CONFIG_NET_DMA
153 static struct dma_client
*net_dma_client
;
154 static unsigned int net_dma_count
;
155 static spinlock_t net_dma_event_lock
;
159 * The @dev_base list is protected by @dev_base_lock and the rtnl
162 * Pure readers hold dev_base_lock for reading.
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
177 struct net_device
*dev_base
;
178 static struct net_device
**dev_tail
= &dev_base
;
179 DEFINE_RWLOCK(dev_base_lock
);
181 EXPORT_SYMBOL(dev_base
);
182 EXPORT_SYMBOL(dev_base_lock
);
184 #define NETDEV_HASHBITS 8
185 static struct hlist_head dev_name_head
[1<<NETDEV_HASHBITS
];
186 static struct hlist_head dev_index_head
[1<<NETDEV_HASHBITS
];
188 static inline struct hlist_head
*dev_name_hash(const char *name
)
190 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
191 return &dev_name_head
[hash
& ((1<<NETDEV_HASHBITS
)-1)];
194 static inline struct hlist_head
*dev_index_hash(int ifindex
)
196 return &dev_index_head
[ifindex
& ((1<<NETDEV_HASHBITS
)-1)];
203 static RAW_NOTIFIER_HEAD(netdev_chain
);
206 * Device drivers call our routines to queue packets here. We empty the
207 * queue in the local softnet handler.
209 DEFINE_PER_CPU(struct softnet_data
, softnet_data
) = { NULL
};
212 extern int netdev_sysfs_init(void);
213 extern int netdev_register_sysfs(struct net_device
*);
214 extern void netdev_unregister_sysfs(struct net_device
*);
216 #define netdev_sysfs_init() (0)
217 #define netdev_register_sysfs(dev) (0)
218 #define netdev_unregister_sysfs(dev) do { } while(0)
222 /*******************************************************************************
224 Protocol management and registration routines
226 *******************************************************************************/
235 * Add a protocol ID to the list. Now that the input handler is
236 * smarter we can dispense with all the messy stuff that used to be
239 * BEWARE!!! Protocol handlers, mangling input packets,
240 * MUST BE last in hash buckets and checking protocol handlers
241 * MUST start from promiscuous ptype_all chain in net_bh.
242 * It is true now, do not change it.
243 * Explanation follows: if protocol handler, mangling packet, will
244 * be the first on list, it is not able to sense, that packet
245 * is cloned and should be copied-on-write, so that it will
246 * change it and subsequent readers will get broken packet.
251 * dev_add_pack - add packet handler
252 * @pt: packet type declaration
254 * Add a protocol handler to the networking stack. The passed &packet_type
255 * is linked into kernel lists and may not be freed until it has been
256 * removed from the kernel lists.
258 * This call does not sleep therefore it can not
259 * guarantee all CPU's that are in middle of receiving packets
260 * will see the new packet type (until the next received packet).
263 void dev_add_pack(struct packet_type
*pt
)
267 spin_lock_bh(&ptype_lock
);
268 if (pt
->type
== htons(ETH_P_ALL
)) {
270 list_add_rcu(&pt
->list
, &ptype_all
);
272 hash
= ntohs(pt
->type
) & 15;
273 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
275 spin_unlock_bh(&ptype_lock
);
279 * __dev_remove_pack - remove packet handler
280 * @pt: packet type declaration
282 * Remove a protocol handler that was previously added to the kernel
283 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
284 * from the kernel lists and can be freed or reused once this function
287 * The packet type might still be in use by receivers
288 * and must not be freed until after all the CPU's have gone
289 * through a quiescent state.
291 void __dev_remove_pack(struct packet_type
*pt
)
293 struct list_head
*head
;
294 struct packet_type
*pt1
;
296 spin_lock_bh(&ptype_lock
);
298 if (pt
->type
== htons(ETH_P_ALL
)) {
302 head
= &ptype_base
[ntohs(pt
->type
) & 15];
304 list_for_each_entry(pt1
, head
, list
) {
306 list_del_rcu(&pt
->list
);
311 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
313 spin_unlock_bh(&ptype_lock
);
316 * dev_remove_pack - remove packet handler
317 * @pt: packet type declaration
319 * Remove a protocol handler that was previously added to the kernel
320 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
321 * from the kernel lists and can be freed or reused once this function
324 * This call sleeps to guarantee that no CPU is looking at the packet
327 void dev_remove_pack(struct packet_type
*pt
)
329 __dev_remove_pack(pt
);
334 /******************************************************************************
336 Device Boot-time Settings Routines
338 *******************************************************************************/
340 /* Boot time configuration table */
341 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
344 * netdev_boot_setup_add - add new setup entry
345 * @name: name of the device
346 * @map: configured settings for the device
348 * Adds new setup entry to the dev_boot_setup list. The function
349 * returns 0 on error and 1 on success. This is a generic routine to
352 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
354 struct netdev_boot_setup
*s
;
358 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
359 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
360 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
361 strcpy(s
[i
].name
, name
);
362 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
367 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
371 * netdev_boot_setup_check - check boot time settings
372 * @dev: the netdevice
374 * Check boot time settings for the device.
375 * The found settings are set for the device to be used
376 * later in the device probing.
377 * Returns 0 if no settings found, 1 if they are.
379 int netdev_boot_setup_check(struct net_device
*dev
)
381 struct netdev_boot_setup
*s
= dev_boot_setup
;
384 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
385 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
386 !strncmp(dev
->name
, s
[i
].name
, strlen(s
[i
].name
))) {
387 dev
->irq
= s
[i
].map
.irq
;
388 dev
->base_addr
= s
[i
].map
.base_addr
;
389 dev
->mem_start
= s
[i
].map
.mem_start
;
390 dev
->mem_end
= s
[i
].map
.mem_end
;
399 * netdev_boot_base - get address from boot time settings
400 * @prefix: prefix for network device
401 * @unit: id for network device
403 * Check boot time settings for the base address of device.
404 * The found settings are set for the device to be used
405 * later in the device probing.
406 * Returns 0 if no settings found.
408 unsigned long netdev_boot_base(const char *prefix
, int unit
)
410 const struct netdev_boot_setup
*s
= dev_boot_setup
;
414 sprintf(name
, "%s%d", prefix
, unit
);
417 * If device already registered then return base of 1
418 * to indicate not to probe for this interface
420 if (__dev_get_by_name(name
))
423 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
424 if (!strcmp(name
, s
[i
].name
))
425 return s
[i
].map
.base_addr
;
430 * Saves at boot time configured settings for any netdevice.
432 int __init
netdev_boot_setup(char *str
)
437 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
442 memset(&map
, 0, sizeof(map
));
446 map
.base_addr
= ints
[2];
448 map
.mem_start
= ints
[3];
450 map
.mem_end
= ints
[4];
452 /* Add new entry to the list */
453 return netdev_boot_setup_add(str
, &map
);
456 __setup("netdev=", netdev_boot_setup
);
458 /*******************************************************************************
460 Device Interface Subroutines
462 *******************************************************************************/
465 * __dev_get_by_name - find a device by its name
466 * @name: name to find
468 * Find an interface by name. Must be called under RTNL semaphore
469 * or @dev_base_lock. If the name is found a pointer to the device
470 * is returned. If the name is not found then %NULL is returned. The
471 * reference counters are not incremented so the caller must be
472 * careful with locks.
475 struct net_device
*__dev_get_by_name(const char *name
)
477 struct hlist_node
*p
;
479 hlist_for_each(p
, dev_name_hash(name
)) {
480 struct net_device
*dev
481 = hlist_entry(p
, struct net_device
, name_hlist
);
482 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
489 * dev_get_by_name - find a device by its name
490 * @name: name to find
492 * Find an interface by name. This can be called from any
493 * context and does its own locking. The returned handle has
494 * the usage count incremented and the caller must use dev_put() to
495 * release it when it is no longer needed. %NULL is returned if no
496 * matching device is found.
499 struct net_device
*dev_get_by_name(const char *name
)
501 struct net_device
*dev
;
503 read_lock(&dev_base_lock
);
504 dev
= __dev_get_by_name(name
);
507 read_unlock(&dev_base_lock
);
512 * __dev_get_by_index - find a device by its ifindex
513 * @ifindex: index of device
515 * Search for an interface by index. Returns %NULL if the device
516 * is not found or a pointer to the device. The device has not
517 * had its reference counter increased so the caller must be careful
518 * about locking. The caller must hold either the RTNL semaphore
522 struct net_device
*__dev_get_by_index(int ifindex
)
524 struct hlist_node
*p
;
526 hlist_for_each(p
, dev_index_hash(ifindex
)) {
527 struct net_device
*dev
528 = hlist_entry(p
, struct net_device
, index_hlist
);
529 if (dev
->ifindex
== ifindex
)
537 * dev_get_by_index - find a device by its ifindex
538 * @ifindex: index of device
540 * Search for an interface by index. Returns NULL if the device
541 * is not found or a pointer to the device. The device returned has
542 * had a reference added and the pointer is safe until the user calls
543 * dev_put to indicate they have finished with it.
546 struct net_device
*dev_get_by_index(int ifindex
)
548 struct net_device
*dev
;
550 read_lock(&dev_base_lock
);
551 dev
= __dev_get_by_index(ifindex
);
554 read_unlock(&dev_base_lock
);
559 * dev_getbyhwaddr - find a device by its hardware address
560 * @type: media type of device
561 * @ha: hardware address
563 * Search for an interface by MAC address. Returns NULL if the device
564 * is not found or a pointer to the device. The caller must hold the
565 * rtnl semaphore. The returned device has not had its ref count increased
566 * and the caller must therefore be careful about locking
569 * If the API was consistent this would be __dev_get_by_hwaddr
572 struct net_device
*dev_getbyhwaddr(unsigned short type
, char *ha
)
574 struct net_device
*dev
;
578 for (dev
= dev_base
; dev
; dev
= dev
->next
)
579 if (dev
->type
== type
&&
580 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
585 EXPORT_SYMBOL(dev_getbyhwaddr
);
587 struct net_device
*dev_getfirstbyhwtype(unsigned short type
)
589 struct net_device
*dev
;
592 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
593 if (dev
->type
== type
) {
602 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
605 * dev_get_by_flags - find any device with given flags
606 * @if_flags: IFF_* values
607 * @mask: bitmask of bits in if_flags to check
609 * Search for any interface with the given flags. Returns NULL if a device
610 * is not found or a pointer to the device. The device returned has
611 * had a reference added and the pointer is safe until the user calls
612 * dev_put to indicate they have finished with it.
615 struct net_device
* dev_get_by_flags(unsigned short if_flags
, unsigned short mask
)
617 struct net_device
*dev
;
619 read_lock(&dev_base_lock
);
620 for (dev
= dev_base
; dev
!= NULL
; dev
= dev
->next
) {
621 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
626 read_unlock(&dev_base_lock
);
631 * dev_valid_name - check if name is okay for network device
634 * Network device names need to be valid file names to
635 * to allow sysfs to work
637 int dev_valid_name(const char *name
)
639 return !(*name
== '\0'
640 || !strcmp(name
, ".")
641 || !strcmp(name
, "..")
642 || strchr(name
, '/'));
646 * dev_alloc_name - allocate a name for a device
648 * @name: name format string
650 * Passed a format string - eg "lt%d" it will try and find a suitable
651 * id. It scans list of devices to build up a free map, then chooses
652 * the first empty slot. The caller must hold the dev_base or rtnl lock
653 * while allocating the name and adding the device in order to avoid
655 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
656 * Returns the number of the unit assigned or a negative errno code.
659 int dev_alloc_name(struct net_device
*dev
, const char *name
)
664 const int max_netdevices
= 8*PAGE_SIZE
;
666 struct net_device
*d
;
668 p
= strnchr(name
, IFNAMSIZ
-1, '%');
671 * Verify the string as this thing may have come from
672 * the user. There must be either one "%d" and no other "%"
675 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
678 /* Use one page as a bit array of possible slots */
679 inuse
= (long *) get_zeroed_page(GFP_ATOMIC
);
683 for (d
= dev_base
; d
; d
= d
->next
) {
684 if (!sscanf(d
->name
, name
, &i
))
686 if (i
< 0 || i
>= max_netdevices
)
689 /* avoid cases where sscanf is not exact inverse of printf */
690 snprintf(buf
, sizeof(buf
), name
, i
);
691 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
695 i
= find_first_zero_bit(inuse
, max_netdevices
);
696 free_page((unsigned long) inuse
);
699 snprintf(buf
, sizeof(buf
), name
, i
);
700 if (!__dev_get_by_name(buf
)) {
701 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
705 /* It is possible to run out of possible slots
706 * when the name is long and there isn't enough space left
707 * for the digits, or if all bits are used.
714 * dev_change_name - change name of a device
716 * @newname: name (or format string) must be at least IFNAMSIZ
718 * Change name of a device, can pass format strings "eth%d".
721 int dev_change_name(struct net_device
*dev
, char *newname
)
727 if (dev
->flags
& IFF_UP
)
730 if (!dev_valid_name(newname
))
733 if (strchr(newname
, '%')) {
734 err
= dev_alloc_name(dev
, newname
);
737 strcpy(newname
, dev
->name
);
739 else if (__dev_get_by_name(newname
))
742 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
744 err
= class_device_rename(&dev
->class_dev
, dev
->name
);
746 hlist_del(&dev
->name_hlist
);
747 hlist_add_head(&dev
->name_hlist
, dev_name_hash(dev
->name
));
748 raw_notifier_call_chain(&netdev_chain
,
749 NETDEV_CHANGENAME
, dev
);
756 * netdev_features_change - device changes features
757 * @dev: device to cause notification
759 * Called to indicate a device has changed features.
761 void netdev_features_change(struct net_device
*dev
)
763 raw_notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
765 EXPORT_SYMBOL(netdev_features_change
);
768 * netdev_state_change - device changes state
769 * @dev: device to cause notification
771 * Called to indicate a device has changed state. This function calls
772 * the notifier chains for netdev_chain and sends a NEWLINK message
773 * to the routing socket.
775 void netdev_state_change(struct net_device
*dev
)
777 if (dev
->flags
& IFF_UP
) {
778 raw_notifier_call_chain(&netdev_chain
,
780 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
785 * dev_load - load a network module
786 * @name: name of interface
788 * If a network interface is not present and the process has suitable
789 * privileges this function loads the module. If module loading is not
790 * available in this kernel then it becomes a nop.
793 void dev_load(const char *name
)
795 struct net_device
*dev
;
797 read_lock(&dev_base_lock
);
798 dev
= __dev_get_by_name(name
);
799 read_unlock(&dev_base_lock
);
801 if (!dev
&& capable(CAP_SYS_MODULE
))
802 request_module("%s", name
);
805 static int default_rebuild_header(struct sk_buff
*skb
)
807 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
808 skb
->dev
? skb
->dev
->name
: "NULL!!!");
815 * dev_open - prepare an interface for use.
816 * @dev: device to open
818 * Takes a device from down to up state. The device's private open
819 * function is invoked and then the multicast lists are loaded. Finally
820 * the device is moved into the up state and a %NETDEV_UP message is
821 * sent to the netdev notifier chain.
823 * Calling this function on an active interface is a nop. On a failure
824 * a negative errno code is returned.
826 int dev_open(struct net_device
*dev
)
834 if (dev
->flags
& IFF_UP
)
838 * Is it even present?
840 if (!netif_device_present(dev
))
844 * Call device private open method
846 set_bit(__LINK_STATE_START
, &dev
->state
);
848 ret
= dev
->open(dev
);
850 clear_bit(__LINK_STATE_START
, &dev
->state
);
854 * If it went open OK then:
861 dev
->flags
|= IFF_UP
;
864 * Initialize multicasting status
869 * Wakeup transmit queue engine
874 * ... and announce new interface.
876 raw_notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
882 * dev_close - shutdown an interface.
883 * @dev: device to shutdown
885 * This function moves an active device into down state. A
886 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
887 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
890 int dev_close(struct net_device
*dev
)
892 if (!(dev
->flags
& IFF_UP
))
896 * Tell people we are going down, so that they can
897 * prepare to death, when device is still operating.
899 raw_notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
903 clear_bit(__LINK_STATE_START
, &dev
->state
);
905 /* Synchronize to scheduled poll. We cannot touch poll list,
906 * it can be even on different cpu. So just clear netif_running(),
907 * and wait when poll really will happen. Actually, the best place
908 * for this is inside dev->stop() after device stopped its irq
909 * engine, but this requires more changes in devices. */
911 smp_mb__after_clear_bit(); /* Commit netif_running(). */
912 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
918 * Call the device specific close. This cannot fail.
919 * Only if device is UP
921 * We allow it to be called even after a DETACH hot-plug
928 * Device is now down.
931 dev
->flags
&= ~IFF_UP
;
934 * Tell people we are down
936 raw_notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
943 * Device change register/unregister. These are not inline or static
944 * as we export them to the world.
948 * register_netdevice_notifier - register a network notifier block
951 * Register a notifier to be called when network device events occur.
952 * The notifier passed is linked into the kernel structures and must
953 * not be reused until it has been unregistered. A negative errno code
954 * is returned on a failure.
956 * When registered all registration and up events are replayed
957 * to the new notifier to allow device to have a race free
958 * view of the network device list.
961 int register_netdevice_notifier(struct notifier_block
*nb
)
963 struct net_device
*dev
;
967 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
969 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
970 nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
972 if (dev
->flags
& IFF_UP
)
973 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
981 * unregister_netdevice_notifier - unregister a network notifier block
984 * Unregister a notifier previously registered by
985 * register_netdevice_notifier(). The notifier is unlinked into the
986 * kernel structures and may then be reused. A negative errno code
987 * is returned on a failure.
990 int unregister_netdevice_notifier(struct notifier_block
*nb
)
995 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1001 * call_netdevice_notifiers - call all network notifier blocks
1002 * @val: value passed unmodified to notifier function
1003 * @v: pointer passed unmodified to notifier function
1005 * Call all network notifier blocks. Parameters and return value
1006 * are as for raw_notifier_call_chain().
1009 int call_netdevice_notifiers(unsigned long val
, void *v
)
1011 return raw_notifier_call_chain(&netdev_chain
, val
, v
);
1014 /* When > 0 there are consumers of rx skb time stamps */
1015 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1017 void net_enable_timestamp(void)
1019 atomic_inc(&netstamp_needed
);
1022 void net_disable_timestamp(void)
1024 atomic_dec(&netstamp_needed
);
1027 void __net_timestamp(struct sk_buff
*skb
)
1031 do_gettimeofday(&tv
);
1032 skb_set_timestamp(skb
, &tv
);
1034 EXPORT_SYMBOL(__net_timestamp
);
1036 static inline void net_timestamp(struct sk_buff
*skb
)
1038 if (atomic_read(&netstamp_needed
))
1039 __net_timestamp(skb
);
1041 skb
->tstamp
.off_sec
= 0;
1042 skb
->tstamp
.off_usec
= 0;
1047 * Support routine. Sends outgoing frames to any network
1048 * taps currently in use.
1051 void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1053 struct packet_type
*ptype
;
1058 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1059 /* Never send packets back to the socket
1060 * they originated from - MvS (miquels@drinkel.ow.org)
1062 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1063 (ptype
->af_packet_priv
== NULL
||
1064 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1065 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1069 /* skb->nh should be correctly
1070 set by sender, so that the second statement is
1071 just protection against buggy protocols.
1073 skb2
->mac
.raw
= skb2
->data
;
1075 if (skb2
->nh
.raw
< skb2
->data
||
1076 skb2
->nh
.raw
> skb2
->tail
) {
1077 if (net_ratelimit())
1078 printk(KERN_CRIT
"protocol %04x is "
1080 skb2
->protocol
, dev
->name
);
1081 skb2
->nh
.raw
= skb2
->data
;
1084 skb2
->h
.raw
= skb2
->nh
.raw
;
1085 skb2
->pkt_type
= PACKET_OUTGOING
;
1086 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1093 void __netif_schedule(struct net_device
*dev
)
1095 if (!test_and_set_bit(__LINK_STATE_SCHED
, &dev
->state
)) {
1096 unsigned long flags
;
1097 struct softnet_data
*sd
;
1099 local_irq_save(flags
);
1100 sd
= &__get_cpu_var(softnet_data
);
1101 dev
->next_sched
= sd
->output_queue
;
1102 sd
->output_queue
= dev
;
1103 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1104 local_irq_restore(flags
);
1107 EXPORT_SYMBOL(__netif_schedule
);
1109 void __netif_rx_schedule(struct net_device
*dev
)
1111 unsigned long flags
;
1113 local_irq_save(flags
);
1115 list_add_tail(&dev
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
1117 dev
->quota
+= dev
->weight
;
1119 dev
->quota
= dev
->weight
;
1120 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1121 local_irq_restore(flags
);
1123 EXPORT_SYMBOL(__netif_rx_schedule
);
1125 void dev_kfree_skb_any(struct sk_buff
*skb
)
1127 if (in_irq() || irqs_disabled())
1128 dev_kfree_skb_irq(skb
);
1132 EXPORT_SYMBOL(dev_kfree_skb_any
);
1136 void netif_device_detach(struct net_device
*dev
)
1138 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1139 netif_running(dev
)) {
1140 netif_stop_queue(dev
);
1143 EXPORT_SYMBOL(netif_device_detach
);
1145 void netif_device_attach(struct net_device
*dev
)
1147 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1148 netif_running(dev
)) {
1149 netif_wake_queue(dev
);
1150 __netdev_watchdog_up(dev
);
1153 EXPORT_SYMBOL(netif_device_attach
);
1157 * Invalidate hardware checksum when packet is to be mangled, and
1158 * complete checksum manually on outgoing path.
1160 int skb_checksum_help(struct sk_buff
*skb
, int inward
)
1163 int ret
= 0, offset
= skb
->h
.raw
- skb
->data
;
1166 skb
->ip_summed
= CHECKSUM_NONE
;
1170 if (skb_cloned(skb
)) {
1171 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1176 BUG_ON(offset
> (int)skb
->len
);
1177 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1179 offset
= skb
->tail
- skb
->h
.raw
;
1180 BUG_ON(offset
<= 0);
1181 BUG_ON(skb
->csum
+ 2 > offset
);
1183 *(u16
*)(skb
->h
.raw
+ skb
->csum
) = csum_fold(csum
);
1184 skb
->ip_summed
= CHECKSUM_NONE
;
1189 /* Take action when hardware reception checksum errors are detected. */
1191 void netdev_rx_csum_fault(struct net_device
*dev
)
1193 if (net_ratelimit()) {
1194 printk(KERN_ERR
"%s: hw csum failure.\n",
1195 dev
? dev
->name
: "<unknown>");
1199 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1202 #ifdef CONFIG_HIGHMEM
1203 /* Actually, we should eliminate this check as soon as we know, that:
1204 * 1. IOMMU is present and allows to map all the memory.
1205 * 2. No high memory really exists on this machine.
1208 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1212 if (dev
->features
& NETIF_F_HIGHDMA
)
1215 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1216 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1222 #define illegal_highdma(dev, skb) (0)
1225 #define HARD_TX_LOCK(dev, cpu) { \
1226 if ((dev->features & NETIF_F_LLTX) == 0) { \
1227 netif_tx_lock(dev); \
1231 #define HARD_TX_UNLOCK(dev) { \
1232 if ((dev->features & NETIF_F_LLTX) == 0) { \
1233 netif_tx_unlock(dev); \
1238 * dev_queue_xmit - transmit a buffer
1239 * @skb: buffer to transmit
1241 * Queue a buffer for transmission to a network device. The caller must
1242 * have set the device and priority and built the buffer before calling
1243 * this function. The function can be called from an interrupt.
1245 * A negative errno code is returned on a failure. A success does not
1246 * guarantee the frame will be transmitted as it may be dropped due
1247 * to congestion or traffic shaping.
1249 * -----------------------------------------------------------------------------------
1250 * I notice this method can also return errors from the queue disciplines,
1251 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1254 * Regardless of the return value, the skb is consumed, so it is currently
1255 * difficult to retry a send to this method. (You can bump the ref count
1256 * before sending to hold a reference for retry if you are careful.)
1258 * When calling this method, interrupts MUST be enabled. This is because
1259 * the BH enable code must have IRQs enabled so that it will not deadlock.
1263 int dev_queue_xmit(struct sk_buff
*skb
)
1265 struct net_device
*dev
= skb
->dev
;
1269 if (skb_shinfo(skb
)->frag_list
&&
1270 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1271 __skb_linearize(skb
))
1274 /* Fragmented skb is linearized if device does not support SG,
1275 * or if at least one of fragments is in highmem and device
1276 * does not support DMA from it.
1278 if (skb_shinfo(skb
)->nr_frags
&&
1279 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1280 __skb_linearize(skb
))
1283 /* If packet is not checksummed and device does not support
1284 * checksumming for this protocol, complete checksumming here.
1286 if (skb
->ip_summed
== CHECKSUM_HW
&&
1287 (!(dev
->features
& NETIF_F_GEN_CSUM
) &&
1288 (!(dev
->features
& NETIF_F_IP_CSUM
) ||
1289 skb
->protocol
!= htons(ETH_P_IP
))))
1290 if (skb_checksum_help(skb
, 0))
1293 spin_lock_prefetch(&dev
->queue_lock
);
1295 /* Disable soft irqs for various locks below. Also
1296 * stops preemption for RCU.
1300 /* Updates of qdisc are serialized by queue_lock.
1301 * The struct Qdisc which is pointed to by qdisc is now a
1302 * rcu structure - it may be accessed without acquiring
1303 * a lock (but the structure may be stale.) The freeing of the
1304 * qdisc will be deferred until it's known that there are no
1305 * more references to it.
1307 * If the qdisc has an enqueue function, we still need to
1308 * hold the queue_lock before calling it, since queue_lock
1309 * also serializes access to the device queue.
1312 q
= rcu_dereference(dev
->qdisc
);
1313 #ifdef CONFIG_NET_CLS_ACT
1314 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1317 /* Grab device queue */
1318 spin_lock(&dev
->queue_lock
);
1320 rc
= q
->enqueue(skb
, q
);
1324 spin_unlock(&dev
->queue_lock
);
1325 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1329 /* The device has no queue. Common case for software devices:
1330 loopback, all the sorts of tunnels...
1332 Really, it is unlikely that netif_tx_lock protection is necessary
1333 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1335 However, it is possible, that they rely on protection
1338 Check this and shot the lock. It is not prone from deadlocks.
1339 Either shot noqueue qdisc, it is even simpler 8)
1341 if (dev
->flags
& IFF_UP
) {
1342 int cpu
= smp_processor_id(); /* ok because BHs are off */
1344 if (dev
->xmit_lock_owner
!= cpu
) {
1346 HARD_TX_LOCK(dev
, cpu
);
1348 if (!netif_queue_stopped(dev
)) {
1350 dev_queue_xmit_nit(skb
, dev
);
1353 if (!dev
->hard_start_xmit(skb
, dev
)) {
1354 HARD_TX_UNLOCK(dev
);
1358 HARD_TX_UNLOCK(dev
);
1359 if (net_ratelimit())
1360 printk(KERN_CRIT
"Virtual device %s asks to "
1361 "queue packet!\n", dev
->name
);
1363 /* Recursion is detected! It is possible,
1365 if (net_ratelimit())
1366 printk(KERN_CRIT
"Dead loop on virtual device "
1367 "%s, fix it urgently!\n", dev
->name
);
1383 /*=======================================================================
1385 =======================================================================*/
1387 int netdev_max_backlog
= 1000;
1388 int netdev_budget
= 300;
1389 int weight_p
= 64; /* old backlog weight */
1391 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1395 * netif_rx - post buffer to the network code
1396 * @skb: buffer to post
1398 * This function receives a packet from a device driver and queues it for
1399 * the upper (protocol) levels to process. It always succeeds. The buffer
1400 * may be dropped during processing for congestion control or by the
1404 * NET_RX_SUCCESS (no congestion)
1405 * NET_RX_CN_LOW (low congestion)
1406 * NET_RX_CN_MOD (moderate congestion)
1407 * NET_RX_CN_HIGH (high congestion)
1408 * NET_RX_DROP (packet was dropped)
1412 int netif_rx(struct sk_buff
*skb
)
1414 struct softnet_data
*queue
;
1415 unsigned long flags
;
1417 /* if netpoll wants it, pretend we never saw it */
1418 if (netpoll_rx(skb
))
1421 if (!skb
->tstamp
.off_sec
)
1425 * The code is rearranged so that the path is the most
1426 * short when CPU is congested, but is still operating.
1428 local_irq_save(flags
);
1429 queue
= &__get_cpu_var(softnet_data
);
1431 __get_cpu_var(netdev_rx_stat
).total
++;
1432 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1433 if (queue
->input_pkt_queue
.qlen
) {
1436 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1437 local_irq_restore(flags
);
1438 return NET_RX_SUCCESS
;
1441 netif_rx_schedule(&queue
->backlog_dev
);
1445 __get_cpu_var(netdev_rx_stat
).dropped
++;
1446 local_irq_restore(flags
);
1452 int netif_rx_ni(struct sk_buff
*skb
)
1457 err
= netif_rx(skb
);
1458 if (local_softirq_pending())
1465 EXPORT_SYMBOL(netif_rx_ni
);
1467 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1469 struct net_device
*dev
= skb
->dev
;
1473 * On bonding slaves other than the currently active
1474 * slave, suppress duplicates except for 802.3ad
1475 * ETH_P_SLOW and alb non-mcast/bcast.
1477 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
1478 if (dev
->master
->priv_flags
& IFF_MASTER_ALB
) {
1479 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
1480 skb
->pkt_type
!= PACKET_MULTICAST
)
1484 if (dev
->master
->priv_flags
& IFF_MASTER_8023AD
&&
1485 skb
->protocol
== __constant_htons(ETH_P_SLOW
))
1492 skb
->dev
= dev
->master
;
1498 static void net_tx_action(struct softirq_action
*h
)
1500 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1502 if (sd
->completion_queue
) {
1503 struct sk_buff
*clist
;
1505 local_irq_disable();
1506 clist
= sd
->completion_queue
;
1507 sd
->completion_queue
= NULL
;
1511 struct sk_buff
*skb
= clist
;
1512 clist
= clist
->next
;
1514 BUG_TRAP(!atomic_read(&skb
->users
));
1519 if (sd
->output_queue
) {
1520 struct net_device
*head
;
1522 local_irq_disable();
1523 head
= sd
->output_queue
;
1524 sd
->output_queue
= NULL
;
1528 struct net_device
*dev
= head
;
1529 head
= head
->next_sched
;
1531 smp_mb__before_clear_bit();
1532 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1534 if (spin_trylock(&dev
->queue_lock
)) {
1536 spin_unlock(&dev
->queue_lock
);
1538 netif_schedule(dev
);
1544 static __inline__
int deliver_skb(struct sk_buff
*skb
,
1545 struct packet_type
*pt_prev
,
1546 struct net_device
*orig_dev
)
1548 atomic_inc(&skb
->users
);
1549 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1552 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1553 int (*br_handle_frame_hook
)(struct net_bridge_port
*p
, struct sk_buff
**pskb
);
1555 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1556 unsigned char *addr
);
1557 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
);
1559 static __inline__
int handle_bridge(struct sk_buff
**pskb
,
1560 struct packet_type
**pt_prev
, int *ret
,
1561 struct net_device
*orig_dev
)
1563 struct net_bridge_port
*port
;
1565 if ((*pskb
)->pkt_type
== PACKET_LOOPBACK
||
1566 (port
= rcu_dereference((*pskb
)->dev
->br_port
)) == NULL
)
1570 *ret
= deliver_skb(*pskb
, *pt_prev
, orig_dev
);
1574 return br_handle_frame_hook(port
, pskb
);
1577 #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
1580 #ifdef CONFIG_NET_CLS_ACT
1581 /* TODO: Maybe we should just force sch_ingress to be compiled in
1582 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1583 * a compare and 2 stores extra right now if we dont have it on
1584 * but have CONFIG_NET_CLS_ACT
1585 * NOTE: This doesnt stop any functionality; if you dont have
1586 * the ingress scheduler, you just cant add policies on ingress.
1589 static int ing_filter(struct sk_buff
*skb
)
1592 struct net_device
*dev
= skb
->dev
;
1593 int result
= TC_ACT_OK
;
1595 if (dev
->qdisc_ingress
) {
1596 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1597 if (MAX_RED_LOOP
< ttl
++) {
1598 printk("Redir loop detected Dropping packet (%s->%s)\n",
1599 skb
->input_dev
->name
, skb
->dev
->name
);
1603 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1605 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1607 spin_lock(&dev
->ingress_lock
);
1608 if ((q
= dev
->qdisc_ingress
) != NULL
)
1609 result
= q
->enqueue(skb
, q
);
1610 spin_unlock(&dev
->ingress_lock
);
1618 int netif_receive_skb(struct sk_buff
*skb
)
1620 struct packet_type
*ptype
, *pt_prev
;
1621 struct net_device
*orig_dev
;
1622 int ret
= NET_RX_DROP
;
1623 unsigned short type
;
1625 /* if we've gotten here through NAPI, check netpoll */
1626 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1629 if (!skb
->tstamp
.off_sec
)
1632 if (!skb
->input_dev
)
1633 skb
->input_dev
= skb
->dev
;
1635 orig_dev
= skb_bond(skb
);
1640 __get_cpu_var(netdev_rx_stat
).total
++;
1642 skb
->h
.raw
= skb
->nh
.raw
= skb
->data
;
1643 skb
->mac_len
= skb
->nh
.raw
- skb
->mac
.raw
;
1649 #ifdef CONFIG_NET_CLS_ACT
1650 if (skb
->tc_verd
& TC_NCLS
) {
1651 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1656 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1657 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1659 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1664 #ifdef CONFIG_NET_CLS_ACT
1666 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1667 pt_prev
= NULL
; /* noone else should process this after*/
1669 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1672 ret
= ing_filter(skb
);
1674 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
1683 handle_diverter(skb
);
1685 if (handle_bridge(&skb
, &pt_prev
, &ret
, orig_dev
))
1688 type
= skb
->protocol
;
1689 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
1690 if (ptype
->type
== type
&&
1691 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
1693 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1699 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1702 /* Jamal, now you will not able to escape explaining
1703 * me how you were going to use this. :-)
1713 static int process_backlog(struct net_device
*backlog_dev
, int *budget
)
1716 int quota
= min(backlog_dev
->quota
, *budget
);
1717 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1718 unsigned long start_time
= jiffies
;
1720 backlog_dev
->weight
= weight_p
;
1722 struct sk_buff
*skb
;
1723 struct net_device
*dev
;
1725 local_irq_disable();
1726 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
1733 netif_receive_skb(skb
);
1739 if (work
>= quota
|| jiffies
- start_time
> 1)
1744 backlog_dev
->quota
-= work
;
1749 backlog_dev
->quota
-= work
;
1752 list_del(&backlog_dev
->poll_list
);
1753 smp_mb__before_clear_bit();
1754 netif_poll_enable(backlog_dev
);
1760 static void net_rx_action(struct softirq_action
*h
)
1762 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1763 unsigned long start_time
= jiffies
;
1764 int budget
= netdev_budget
;
1767 local_irq_disable();
1769 while (!list_empty(&queue
->poll_list
)) {
1770 struct net_device
*dev
;
1772 if (budget
<= 0 || jiffies
- start_time
> 1)
1777 dev
= list_entry(queue
->poll_list
.next
,
1778 struct net_device
, poll_list
);
1779 have
= netpoll_poll_lock(dev
);
1781 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
1782 netpoll_poll_unlock(have
);
1783 local_irq_disable();
1784 list_move_tail(&dev
->poll_list
, &queue
->poll_list
);
1786 dev
->quota
+= dev
->weight
;
1788 dev
->quota
= dev
->weight
;
1790 netpoll_poll_unlock(have
);
1792 local_irq_disable();
1796 #ifdef CONFIG_NET_DMA
1798 * There may not be any more sk_buffs coming right now, so push
1799 * any pending DMA copies to hardware
1801 if (net_dma_client
) {
1802 struct dma_chan
*chan
;
1804 list_for_each_entry_rcu(chan
, &net_dma_client
->channels
, client_node
)
1805 dma_async_memcpy_issue_pending(chan
);
1813 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
1814 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1818 static gifconf_func_t
* gifconf_list
[NPROTO
];
1821 * register_gifconf - register a SIOCGIF handler
1822 * @family: Address family
1823 * @gifconf: Function handler
1825 * Register protocol dependent address dumping routines. The handler
1826 * that is passed must not be freed or reused until it has been replaced
1827 * by another handler.
1829 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
1831 if (family
>= NPROTO
)
1833 gifconf_list
[family
] = gifconf
;
1839 * Map an interface index to its name (SIOCGIFNAME)
1843 * We need this ioctl for efficient implementation of the
1844 * if_indextoname() function required by the IPv6 API. Without
1845 * it, we would have to search all the interfaces to find a
1849 static int dev_ifname(struct ifreq __user
*arg
)
1851 struct net_device
*dev
;
1855 * Fetch the caller's info block.
1858 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
1861 read_lock(&dev_base_lock
);
1862 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
1864 read_unlock(&dev_base_lock
);
1868 strcpy(ifr
.ifr_name
, dev
->name
);
1869 read_unlock(&dev_base_lock
);
1871 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
1877 * Perform a SIOCGIFCONF call. This structure will change
1878 * size eventually, and there is nothing I can do about it.
1879 * Thus we will need a 'compatibility mode'.
1882 static int dev_ifconf(char __user
*arg
)
1885 struct net_device
*dev
;
1892 * Fetch the caller's info block.
1895 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
1902 * Loop over the interfaces, and write an info block for each.
1906 for (dev
= dev_base
; dev
; dev
= dev
->next
) {
1907 for (i
= 0; i
< NPROTO
; i
++) {
1908 if (gifconf_list
[i
]) {
1911 done
= gifconf_list
[i
](dev
, NULL
, 0);
1913 done
= gifconf_list
[i
](dev
, pos
+ total
,
1923 * All done. Write the updated control block back to the caller.
1925 ifc
.ifc_len
= total
;
1928 * Both BSD and Solaris return 0 here, so we do too.
1930 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
1933 #ifdef CONFIG_PROC_FS
1935 * This is invoked by the /proc filesystem handler to display a device
1938 static __inline__
struct net_device
*dev_get_idx(loff_t pos
)
1940 struct net_device
*dev
;
1943 for (i
= 0, dev
= dev_base
; dev
&& i
< pos
; ++i
, dev
= dev
->next
);
1945 return i
== pos
? dev
: NULL
;
1948 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1950 read_lock(&dev_base_lock
);
1951 return *pos
? dev_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
1954 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1957 return v
== SEQ_START_TOKEN
? dev_base
: ((struct net_device
*)v
)->next
;
1960 void dev_seq_stop(struct seq_file
*seq
, void *v
)
1962 read_unlock(&dev_base_lock
);
1965 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
1967 if (dev
->get_stats
) {
1968 struct net_device_stats
*stats
= dev
->get_stats(dev
);
1970 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
1971 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
1972 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
1974 stats
->rx_dropped
+ stats
->rx_missed_errors
,
1975 stats
->rx_fifo_errors
,
1976 stats
->rx_length_errors
+ stats
->rx_over_errors
+
1977 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
1978 stats
->rx_compressed
, stats
->multicast
,
1979 stats
->tx_bytes
, stats
->tx_packets
,
1980 stats
->tx_errors
, stats
->tx_dropped
,
1981 stats
->tx_fifo_errors
, stats
->collisions
,
1982 stats
->tx_carrier_errors
+
1983 stats
->tx_aborted_errors
+
1984 stats
->tx_window_errors
+
1985 stats
->tx_heartbeat_errors
,
1986 stats
->tx_compressed
);
1988 seq_printf(seq
, "%6s: No statistics available.\n", dev
->name
);
1992 * Called from the PROCfs module. This now uses the new arbitrary sized
1993 * /proc/net interface to create /proc/net/dev
1995 static int dev_seq_show(struct seq_file
*seq
, void *v
)
1997 if (v
== SEQ_START_TOKEN
)
1998 seq_puts(seq
, "Inter-| Receive "
2000 " face |bytes packets errs drop fifo frame "
2001 "compressed multicast|bytes packets errs "
2002 "drop fifo colls carrier compressed\n");
2004 dev_seq_printf_stats(seq
, v
);
2008 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2010 struct netif_rx_stats
*rc
= NULL
;
2012 while (*pos
< NR_CPUS
)
2013 if (cpu_online(*pos
)) {
2014 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2021 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2023 return softnet_get_online(pos
);
2026 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2029 return softnet_get_online(pos
);
2032 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2036 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2038 struct netif_rx_stats
*s
= v
;
2040 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2041 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2042 0, 0, 0, 0, /* was fastroute */
2047 static struct seq_operations dev_seq_ops
= {
2048 .start
= dev_seq_start
,
2049 .next
= dev_seq_next
,
2050 .stop
= dev_seq_stop
,
2051 .show
= dev_seq_show
,
2054 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2056 return seq_open(file
, &dev_seq_ops
);
2059 static struct file_operations dev_seq_fops
= {
2060 .owner
= THIS_MODULE
,
2061 .open
= dev_seq_open
,
2063 .llseek
= seq_lseek
,
2064 .release
= seq_release
,
2067 static struct seq_operations softnet_seq_ops
= {
2068 .start
= softnet_seq_start
,
2069 .next
= softnet_seq_next
,
2070 .stop
= softnet_seq_stop
,
2071 .show
= softnet_seq_show
,
2074 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2076 return seq_open(file
, &softnet_seq_ops
);
2079 static struct file_operations softnet_seq_fops
= {
2080 .owner
= THIS_MODULE
,
2081 .open
= softnet_seq_open
,
2083 .llseek
= seq_lseek
,
2084 .release
= seq_release
,
2087 #ifdef CONFIG_WIRELESS_EXT
2088 extern int wireless_proc_init(void);
2090 #define wireless_proc_init() 0
2093 static int __init
dev_proc_init(void)
2097 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2099 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2101 if (wireless_proc_init())
2107 proc_net_remove("softnet_stat");
2109 proc_net_remove("dev");
2113 #define dev_proc_init() 0
2114 #endif /* CONFIG_PROC_FS */
2118 * netdev_set_master - set up master/slave pair
2119 * @slave: slave device
2120 * @master: new master device
2122 * Changes the master device of the slave. Pass %NULL to break the
2123 * bonding. The caller must hold the RTNL semaphore. On a failure
2124 * a negative errno code is returned. On success the reference counts
2125 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2126 * function returns zero.
2128 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2130 struct net_device
*old
= slave
->master
;
2140 slave
->master
= master
;
2148 slave
->flags
|= IFF_SLAVE
;
2150 slave
->flags
&= ~IFF_SLAVE
;
2152 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2157 * dev_set_promiscuity - update promiscuity count on a device
2161 * Add or remove promiscuity from a device. While the count in the device
2162 * remains above zero the interface remains promiscuous. Once it hits zero
2163 * the device reverts back to normal filtering operation. A negative inc
2164 * value is used to drop promiscuity on the device.
2166 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2168 unsigned short old_flags
= dev
->flags
;
2170 if ((dev
->promiscuity
+= inc
) == 0)
2171 dev
->flags
&= ~IFF_PROMISC
;
2173 dev
->flags
|= IFF_PROMISC
;
2174 if (dev
->flags
!= old_flags
) {
2176 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2177 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2179 audit_log(current
->audit_context
, GFP_ATOMIC
,
2180 AUDIT_ANOM_PROMISCUOUS
,
2181 "dev=%s prom=%d old_prom=%d auid=%u",
2182 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2183 (old_flags
& IFF_PROMISC
),
2184 audit_get_loginuid(current
->audit_context
));
2189 * dev_set_allmulti - update allmulti count on a device
2193 * Add or remove reception of all multicast frames to a device. While the
2194 * count in the device remains above zero the interface remains listening
2195 * to all interfaces. Once it hits zero the device reverts back to normal
2196 * filtering operation. A negative @inc value is used to drop the counter
2197 * when releasing a resource needing all multicasts.
2200 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2202 unsigned short old_flags
= dev
->flags
;
2204 dev
->flags
|= IFF_ALLMULTI
;
2205 if ((dev
->allmulti
+= inc
) == 0)
2206 dev
->flags
&= ~IFF_ALLMULTI
;
2207 if (dev
->flags
^ old_flags
)
2211 unsigned dev_get_flags(const struct net_device
*dev
)
2215 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2220 (dev
->gflags
& (IFF_PROMISC
|
2223 if (netif_running(dev
)) {
2224 if (netif_oper_up(dev
))
2225 flags
|= IFF_RUNNING
;
2226 if (netif_carrier_ok(dev
))
2227 flags
|= IFF_LOWER_UP
;
2228 if (netif_dormant(dev
))
2229 flags
|= IFF_DORMANT
;
2235 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2238 int old_flags
= dev
->flags
;
2241 * Set the flags on our device.
2244 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2245 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2247 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2251 * Load in the correct multicast list now the flags have changed.
2257 * Have we downed the interface. We handle IFF_UP ourselves
2258 * according to user attempts to set it, rather than blindly
2263 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2264 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2270 if (dev
->flags
& IFF_UP
&&
2271 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2273 raw_notifier_call_chain(&netdev_chain
,
2274 NETDEV_CHANGE
, dev
);
2276 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2277 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2278 dev
->gflags
^= IFF_PROMISC
;
2279 dev_set_promiscuity(dev
, inc
);
2282 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2283 is important. Some (broken) drivers set IFF_PROMISC, when
2284 IFF_ALLMULTI is requested not asking us and not reporting.
2286 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2287 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2288 dev
->gflags
^= IFF_ALLMULTI
;
2289 dev_set_allmulti(dev
, inc
);
2292 if (old_flags
^ dev
->flags
)
2293 rtmsg_ifinfo(RTM_NEWLINK
, dev
, old_flags
^ dev
->flags
);
2298 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2302 if (new_mtu
== dev
->mtu
)
2305 /* MTU must be positive. */
2309 if (!netif_device_present(dev
))
2313 if (dev
->change_mtu
)
2314 err
= dev
->change_mtu(dev
, new_mtu
);
2317 if (!err
&& dev
->flags
& IFF_UP
)
2318 raw_notifier_call_chain(&netdev_chain
,
2319 NETDEV_CHANGEMTU
, dev
);
2323 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2327 if (!dev
->set_mac_address
)
2329 if (sa
->sa_family
!= dev
->type
)
2331 if (!netif_device_present(dev
))
2333 err
= dev
->set_mac_address(dev
, sa
);
2335 raw_notifier_call_chain(&netdev_chain
,
2336 NETDEV_CHANGEADDR
, dev
);
2341 * Perform the SIOCxIFxxx calls.
2343 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2346 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2352 case SIOCGIFFLAGS
: /* Get interface flags */
2353 ifr
->ifr_flags
= dev_get_flags(dev
);
2356 case SIOCSIFFLAGS
: /* Set interface flags */
2357 return dev_change_flags(dev
, ifr
->ifr_flags
);
2359 case SIOCGIFMETRIC
: /* Get the metric on the interface
2360 (currently unused) */
2361 ifr
->ifr_metric
= 0;
2364 case SIOCSIFMETRIC
: /* Set the metric on the interface
2365 (currently unused) */
2368 case SIOCGIFMTU
: /* Get the MTU of a device */
2369 ifr
->ifr_mtu
= dev
->mtu
;
2372 case SIOCSIFMTU
: /* Set the MTU of a device */
2373 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
2377 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
2379 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
2380 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2381 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
2385 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
2387 case SIOCSIFHWBROADCAST
:
2388 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
2390 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
2391 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2392 raw_notifier_call_chain(&netdev_chain
,
2393 NETDEV_CHANGEADDR
, dev
);
2397 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
2398 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
2399 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
2400 ifr
->ifr_map
.irq
= dev
->irq
;
2401 ifr
->ifr_map
.dma
= dev
->dma
;
2402 ifr
->ifr_map
.port
= dev
->if_port
;
2406 if (dev
->set_config
) {
2407 if (!netif_device_present(dev
))
2409 return dev
->set_config(dev
, &ifr
->ifr_map
);
2414 if (!dev
->set_multicast_list
||
2415 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2417 if (!netif_device_present(dev
))
2419 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
2423 if (!dev
->set_multicast_list
||
2424 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2426 if (!netif_device_present(dev
))
2428 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
2432 ifr
->ifr_ifindex
= dev
->ifindex
;
2436 ifr
->ifr_qlen
= dev
->tx_queue_len
;
2440 if (ifr
->ifr_qlen
< 0)
2442 dev
->tx_queue_len
= ifr
->ifr_qlen
;
2446 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
2447 return dev_change_name(dev
, ifr
->ifr_newname
);
2450 * Unknown or private ioctl
2454 if ((cmd
>= SIOCDEVPRIVATE
&&
2455 cmd
<= SIOCDEVPRIVATE
+ 15) ||
2456 cmd
== SIOCBONDENSLAVE
||
2457 cmd
== SIOCBONDRELEASE
||
2458 cmd
== SIOCBONDSETHWADDR
||
2459 cmd
== SIOCBONDSLAVEINFOQUERY
||
2460 cmd
== SIOCBONDINFOQUERY
||
2461 cmd
== SIOCBONDCHANGEACTIVE
||
2462 cmd
== SIOCGMIIPHY
||
2463 cmd
== SIOCGMIIREG
||
2464 cmd
== SIOCSMIIREG
||
2465 cmd
== SIOCBRADDIF
||
2466 cmd
== SIOCBRDELIF
||
2467 cmd
== SIOCWANDEV
) {
2469 if (dev
->do_ioctl
) {
2470 if (netif_device_present(dev
))
2471 err
= dev
->do_ioctl(dev
, ifr
,
2484 * This function handles all "interface"-type I/O control requests. The actual
2485 * 'doing' part of this is dev_ifsioc above.
2489 * dev_ioctl - network device ioctl
2490 * @cmd: command to issue
2491 * @arg: pointer to a struct ifreq in user space
2493 * Issue ioctl functions to devices. This is normally called by the
2494 * user space syscall interfaces but can sometimes be useful for
2495 * other purposes. The return value is the return from the syscall if
2496 * positive or a negative errno code on error.
2499 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
2505 /* One special case: SIOCGIFCONF takes ifconf argument
2506 and requires shared lock, because it sleeps writing
2510 if (cmd
== SIOCGIFCONF
) {
2512 ret
= dev_ifconf((char __user
*) arg
);
2516 if (cmd
== SIOCGIFNAME
)
2517 return dev_ifname((struct ifreq __user
*)arg
);
2519 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2522 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
2524 colon
= strchr(ifr
.ifr_name
, ':');
2529 * See which interface the caller is talking about.
2534 * These ioctl calls:
2535 * - can be done by all.
2536 * - atomic and do not require locking.
2547 dev_load(ifr
.ifr_name
);
2548 read_lock(&dev_base_lock
);
2549 ret
= dev_ifsioc(&ifr
, cmd
);
2550 read_unlock(&dev_base_lock
);
2554 if (copy_to_user(arg
, &ifr
,
2555 sizeof(struct ifreq
)))
2561 dev_load(ifr
.ifr_name
);
2563 ret
= dev_ethtool(&ifr
);
2568 if (copy_to_user(arg
, &ifr
,
2569 sizeof(struct ifreq
)))
2575 * These ioctl calls:
2576 * - require superuser power.
2577 * - require strict serialization.
2583 if (!capable(CAP_NET_ADMIN
))
2585 dev_load(ifr
.ifr_name
);
2587 ret
= dev_ifsioc(&ifr
, cmd
);
2592 if (copy_to_user(arg
, &ifr
,
2593 sizeof(struct ifreq
)))
2599 * These ioctl calls:
2600 * - require superuser power.
2601 * - require strict serialization.
2602 * - do not return a value
2612 case SIOCSIFHWBROADCAST
:
2615 case SIOCBONDENSLAVE
:
2616 case SIOCBONDRELEASE
:
2617 case SIOCBONDSETHWADDR
:
2618 case SIOCBONDCHANGEACTIVE
:
2621 if (!capable(CAP_NET_ADMIN
))
2624 case SIOCBONDSLAVEINFOQUERY
:
2625 case SIOCBONDINFOQUERY
:
2626 dev_load(ifr
.ifr_name
);
2628 ret
= dev_ifsioc(&ifr
, cmd
);
2633 /* Get the per device memory space. We can add this but
2634 * currently do not support it */
2636 /* Set the per device memory buffer space.
2637 * Not applicable in our case */
2642 * Unknown or private ioctl.
2645 if (cmd
== SIOCWANDEV
||
2646 (cmd
>= SIOCDEVPRIVATE
&&
2647 cmd
<= SIOCDEVPRIVATE
+ 15)) {
2648 dev_load(ifr
.ifr_name
);
2650 ret
= dev_ifsioc(&ifr
, cmd
);
2652 if (!ret
&& copy_to_user(arg
, &ifr
,
2653 sizeof(struct ifreq
)))
2657 #ifdef CONFIG_WIRELESS_EXT
2658 /* Take care of Wireless Extensions */
2659 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
) {
2660 /* If command is `set a parameter', or
2661 * `get the encoding parameters', check if
2662 * the user has the right to do it */
2663 if (IW_IS_SET(cmd
) || cmd
== SIOCGIWENCODE
2664 || cmd
== SIOCGIWENCODEEXT
) {
2665 if (!capable(CAP_NET_ADMIN
))
2668 dev_load(ifr
.ifr_name
);
2670 /* Follow me in net/core/wireless.c */
2671 ret
= wireless_process_ioctl(&ifr
, cmd
);
2673 if (IW_IS_GET(cmd
) &&
2674 copy_to_user(arg
, &ifr
,
2675 sizeof(struct ifreq
)))
2679 #endif /* CONFIG_WIRELESS_EXT */
2686 * dev_new_index - allocate an ifindex
2688 * Returns a suitable unique value for a new device interface
2689 * number. The caller must hold the rtnl semaphore or the
2690 * dev_base_lock to be sure it remains unique.
2692 static int dev_new_index(void)
2698 if (!__dev_get_by_index(ifindex
))
2703 static int dev_boot_phase
= 1;
2705 /* Delayed registration/unregisteration */
2706 static DEFINE_SPINLOCK(net_todo_list_lock
);
2707 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
2709 static inline void net_set_todo(struct net_device
*dev
)
2711 spin_lock(&net_todo_list_lock
);
2712 list_add_tail(&dev
->todo_list
, &net_todo_list
);
2713 spin_unlock(&net_todo_list_lock
);
2717 * register_netdevice - register a network device
2718 * @dev: device to register
2720 * Take a completed network device structure and add it to the kernel
2721 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2722 * chain. 0 is returned on success. A negative errno code is returned
2723 * on a failure to set up the device, or if the name is a duplicate.
2725 * Callers must hold the rtnl semaphore. You may want
2726 * register_netdev() instead of this.
2729 * The locking appears insufficient to guarantee two parallel registers
2730 * will not get the same name.
2733 int register_netdevice(struct net_device
*dev
)
2735 struct hlist_head
*head
;
2736 struct hlist_node
*p
;
2739 BUG_ON(dev_boot_phase
);
2744 /* When net_device's are persistent, this will be fatal. */
2745 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
2747 spin_lock_init(&dev
->queue_lock
);
2748 spin_lock_init(&dev
->_xmit_lock
);
2749 dev
->xmit_lock_owner
= -1;
2750 #ifdef CONFIG_NET_CLS_ACT
2751 spin_lock_init(&dev
->ingress_lock
);
2754 ret
= alloc_divert_blk(dev
);
2760 /* Init, if this function is available */
2762 ret
= dev
->init(dev
);
2770 if (!dev_valid_name(dev
->name
)) {
2775 dev
->ifindex
= dev_new_index();
2776 if (dev
->iflink
== -1)
2777 dev
->iflink
= dev
->ifindex
;
2779 /* Check for existence of name */
2780 head
= dev_name_hash(dev
->name
);
2781 hlist_for_each(p
, head
) {
2782 struct net_device
*d
2783 = hlist_entry(p
, struct net_device
, name_hlist
);
2784 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
2790 /* Fix illegal SG+CSUM combinations. */
2791 if ((dev
->features
& NETIF_F_SG
) &&
2792 !(dev
->features
& NETIF_F_ALL_CSUM
)) {
2793 printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
2795 dev
->features
&= ~NETIF_F_SG
;
2798 /* TSO requires that SG is present as well. */
2799 if ((dev
->features
& NETIF_F_TSO
) &&
2800 !(dev
->features
& NETIF_F_SG
)) {
2801 printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
2803 dev
->features
&= ~NETIF_F_TSO
;
2805 if (dev
->features
& NETIF_F_UFO
) {
2806 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
2807 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
2808 "NETIF_F_HW_CSUM feature.\n",
2810 dev
->features
&= ~NETIF_F_UFO
;
2812 if (!(dev
->features
& NETIF_F_SG
)) {
2813 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
2814 "NETIF_F_SG feature.\n",
2816 dev
->features
&= ~NETIF_F_UFO
;
2821 * nil rebuild_header routine,
2822 * that should be never called and used as just bug trap.
2825 if (!dev
->rebuild_header
)
2826 dev
->rebuild_header
= default_rebuild_header
;
2828 ret
= netdev_register_sysfs(dev
);
2831 dev
->reg_state
= NETREG_REGISTERED
;
2834 * Default initial state at registry is that the
2835 * device is present.
2838 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
2841 dev_init_scheduler(dev
);
2842 write_lock_bh(&dev_base_lock
);
2844 dev_tail
= &dev
->next
;
2845 hlist_add_head(&dev
->name_hlist
, head
);
2846 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
2848 write_unlock_bh(&dev_base_lock
);
2850 /* Notify protocols, that a new device appeared. */
2851 raw_notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
2858 free_divert_blk(dev
);
2863 * register_netdev - register a network device
2864 * @dev: device to register
2866 * Take a completed network device structure and add it to the kernel
2867 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
2868 * chain. 0 is returned on success. A negative errno code is returned
2869 * on a failure to set up the device, or if the name is a duplicate.
2871 * This is a wrapper around register_netdev that takes the rtnl semaphore
2872 * and expands the device name if you passed a format string to
2875 int register_netdev(struct net_device
*dev
)
2882 * If the name is a format string the caller wants us to do a
2885 if (strchr(dev
->name
, '%')) {
2886 err
= dev_alloc_name(dev
, dev
->name
);
2892 * Back compatibility hook. Kill this one in 2.5
2894 if (dev
->name
[0] == 0 || dev
->name
[0] == ' ') {
2895 err
= dev_alloc_name(dev
, "eth%d");
2900 err
= register_netdevice(dev
);
2905 EXPORT_SYMBOL(register_netdev
);
2908 * netdev_wait_allrefs - wait until all references are gone.
2910 * This is called when unregistering network devices.
2912 * Any protocol or device that holds a reference should register
2913 * for netdevice notification, and cleanup and put back the
2914 * reference if they receive an UNREGISTER event.
2915 * We can get stuck here if buggy protocols don't correctly
2918 static void netdev_wait_allrefs(struct net_device
*dev
)
2920 unsigned long rebroadcast_time
, warning_time
;
2922 rebroadcast_time
= warning_time
= jiffies
;
2923 while (atomic_read(&dev
->refcnt
) != 0) {
2924 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
2927 /* Rebroadcast unregister notification */
2928 raw_notifier_call_chain(&netdev_chain
,
2929 NETDEV_UNREGISTER
, dev
);
2931 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
2933 /* We must not have linkwatch events
2934 * pending on unregister. If this
2935 * happens, we simply run the queue
2936 * unscheduled, resulting in a noop
2939 linkwatch_run_queue();
2944 rebroadcast_time
= jiffies
;
2949 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
2950 printk(KERN_EMERG
"unregister_netdevice: "
2951 "waiting for %s to become free. Usage "
2953 dev
->name
, atomic_read(&dev
->refcnt
));
2954 warning_time
= jiffies
;
2963 * register_netdevice(x1);
2964 * register_netdevice(x2);
2966 * unregister_netdevice(y1);
2967 * unregister_netdevice(y2);
2973 * We are invoked by rtnl_unlock() after it drops the semaphore.
2974 * This allows us to deal with problems:
2975 * 1) We can delete sysfs objects which invoke hotplug
2976 * without deadlocking with linkwatch via keventd.
2977 * 2) Since we run with the RTNL semaphore not held, we can sleep
2978 * safely in order to wait for the netdev refcnt to drop to zero.
2980 static DEFINE_MUTEX(net_todo_run_mutex
);
2981 void netdev_run_todo(void)
2983 struct list_head list
= LIST_HEAD_INIT(list
);
2985 /* Need to guard against multiple cpu's getting out of order. */
2986 mutex_lock(&net_todo_run_mutex
);
2988 /* Not safe to do outside the semaphore. We must not return
2989 * until all unregister events invoked by the local processor
2990 * have been completed (either by this todo run, or one on
2993 if (list_empty(&net_todo_list
))
2996 /* Snapshot list, allow later requests */
2997 spin_lock(&net_todo_list_lock
);
2998 list_splice_init(&net_todo_list
, &list
);
2999 spin_unlock(&net_todo_list_lock
);
3001 while (!list_empty(&list
)) {
3002 struct net_device
*dev
3003 = list_entry(list
.next
, struct net_device
, todo_list
);
3004 list_del(&dev
->todo_list
);
3006 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
3007 printk(KERN_ERR
"network todo '%s' but state %d\n",
3008 dev
->name
, dev
->reg_state
);
3013 netdev_unregister_sysfs(dev
);
3014 dev
->reg_state
= NETREG_UNREGISTERED
;
3016 netdev_wait_allrefs(dev
);
3019 BUG_ON(atomic_read(&dev
->refcnt
));
3020 BUG_TRAP(!dev
->ip_ptr
);
3021 BUG_TRAP(!dev
->ip6_ptr
);
3022 BUG_TRAP(!dev
->dn_ptr
);
3024 /* It must be the very last action,
3025 * after this 'dev' may point to freed up memory.
3027 if (dev
->destructor
)
3028 dev
->destructor(dev
);
3032 mutex_unlock(&net_todo_run_mutex
);
3036 * alloc_netdev - allocate network device
3037 * @sizeof_priv: size of private data to allocate space for
3038 * @name: device name format string
3039 * @setup: callback to initialize device
3041 * Allocates a struct net_device with private data area for driver use
3042 * and performs basic initialization.
3044 struct net_device
*alloc_netdev(int sizeof_priv
, const char *name
,
3045 void (*setup
)(struct net_device
*))
3048 struct net_device
*dev
;
3051 /* ensure 32-byte alignment of both the device and private area */
3052 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
3053 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
3055 p
= kzalloc(alloc_size
, GFP_KERNEL
);
3057 printk(KERN_ERR
"alloc_dev: Unable to allocate device.\n");
3061 dev
= (struct net_device
*)
3062 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3063 dev
->padded
= (char *)dev
- (char *)p
;
3066 dev
->priv
= netdev_priv(dev
);
3069 strcpy(dev
->name
, name
);
3072 EXPORT_SYMBOL(alloc_netdev
);
3075 * free_netdev - free network device
3078 * This function does the last stage of destroying an allocated device
3079 * interface. The reference to the device object is released.
3080 * If this is the last reference then it will be freed.
3082 void free_netdev(struct net_device
*dev
)
3085 /* Compatibility with error handling in drivers */
3086 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3087 kfree((char *)dev
- dev
->padded
);
3091 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3092 dev
->reg_state
= NETREG_RELEASED
;
3094 /* will free via class release */
3095 class_device_put(&dev
->class_dev
);
3097 kfree((char *)dev
- dev
->padded
);
3101 /* Synchronize with packet receive processing. */
3102 void synchronize_net(void)
3109 * unregister_netdevice - remove device from the kernel
3112 * This function shuts down a device interface and removes it
3113 * from the kernel tables. On success 0 is returned, on a failure
3114 * a negative errno code is returned.
3116 * Callers must hold the rtnl semaphore. You may want
3117 * unregister_netdev() instead of this.
3120 int unregister_netdevice(struct net_device
*dev
)
3122 struct net_device
*d
, **dp
;
3124 BUG_ON(dev_boot_phase
);
3127 /* Some devices call without registering for initialization unwind. */
3128 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3129 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3130 "was registered\n", dev
->name
, dev
);
3134 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3136 /* If device is running, close it first. */
3137 if (dev
->flags
& IFF_UP
)
3140 /* And unlink it from device chain. */
3141 for (dp
= &dev_base
; (d
= *dp
) != NULL
; dp
= &d
->next
) {
3143 write_lock_bh(&dev_base_lock
);
3144 hlist_del(&dev
->name_hlist
);
3145 hlist_del(&dev
->index_hlist
);
3146 if (dev_tail
== &dev
->next
)
3149 write_unlock_bh(&dev_base_lock
);
3154 printk(KERN_ERR
"unregister net_device: '%s' not found\n",
3159 dev
->reg_state
= NETREG_UNREGISTERING
;
3163 /* Shutdown queueing discipline. */
3167 /* Notify protocols, that we are about to destroy
3168 this device. They should clean all the things.
3170 raw_notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3173 * Flush the multicast chain
3175 dev_mc_discard(dev
);
3180 /* Notifier chain MUST detach us from master device. */
3181 BUG_TRAP(!dev
->master
);
3183 free_divert_blk(dev
);
3185 /* Finish processing unregister after unlock */
3195 * unregister_netdev - remove device from the kernel
3198 * This function shuts down a device interface and removes it
3199 * from the kernel tables. On success 0 is returned, on a failure
3200 * a negative errno code is returned.
3202 * This is just a wrapper for unregister_netdevice that takes
3203 * the rtnl semaphore. In general you want to use this and not
3204 * unregister_netdevice.
3206 void unregister_netdev(struct net_device
*dev
)
3209 unregister_netdevice(dev
);
3213 EXPORT_SYMBOL(unregister_netdev
);
3215 #ifdef CONFIG_HOTPLUG_CPU
3216 static int dev_cpu_callback(struct notifier_block
*nfb
,
3217 unsigned long action
,
3220 struct sk_buff
**list_skb
;
3221 struct net_device
**list_net
;
3222 struct sk_buff
*skb
;
3223 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3224 struct softnet_data
*sd
, *oldsd
;
3226 if (action
!= CPU_DEAD
)
3229 local_irq_disable();
3230 cpu
= smp_processor_id();
3231 sd
= &per_cpu(softnet_data
, cpu
);
3232 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3234 /* Find end of our completion_queue. */
3235 list_skb
= &sd
->completion_queue
;
3237 list_skb
= &(*list_skb
)->next
;
3238 /* Append completion queue from offline CPU. */
3239 *list_skb
= oldsd
->completion_queue
;
3240 oldsd
->completion_queue
= NULL
;
3242 /* Find end of our output_queue. */
3243 list_net
= &sd
->output_queue
;
3245 list_net
= &(*list_net
)->next_sched
;
3246 /* Append output queue from offline CPU. */
3247 *list_net
= oldsd
->output_queue
;
3248 oldsd
->output_queue
= NULL
;
3250 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3253 /* Process offline CPU's input_pkt_queue */
3254 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3259 #endif /* CONFIG_HOTPLUG_CPU */
3261 #ifdef CONFIG_NET_DMA
3263 * net_dma_rebalance -
3264 * This is called when the number of channels allocated to the net_dma_client
3265 * changes. The net_dma_client tries to have one DMA channel per CPU.
3267 static void net_dma_rebalance(void)
3269 unsigned int cpu
, i
, n
;
3270 struct dma_chan
*chan
;
3274 if (net_dma_count
== 0) {
3275 for_each_online_cpu(cpu
)
3276 rcu_assign_pointer(per_cpu(softnet_data
.net_dma
, cpu
), NULL
);
3277 unlock_cpu_hotplug();
3282 cpu
= first_cpu(cpu_online_map
);
3285 list_for_each_entry(chan
, &net_dma_client
->channels
, client_node
) {
3286 n
= ((num_online_cpus() / net_dma_count
)
3287 + (i
< (num_online_cpus() % net_dma_count
) ? 1 : 0));
3290 per_cpu(softnet_data
.net_dma
, cpu
) = chan
;
3291 cpu
= next_cpu(cpu
, cpu_online_map
);
3298 unlock_cpu_hotplug();
3302 * netdev_dma_event - event callback for the net_dma_client
3303 * @client: should always be net_dma_client
3307 static void netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
3308 enum dma_event event
)
3310 spin_lock(&net_dma_event_lock
);
3312 case DMA_RESOURCE_ADDED
:
3314 net_dma_rebalance();
3316 case DMA_RESOURCE_REMOVED
:
3318 net_dma_rebalance();
3323 spin_unlock(&net_dma_event_lock
);
3327 * netdev_dma_regiser - register the networking subsystem as a DMA client
3329 static int __init
netdev_dma_register(void)
3331 spin_lock_init(&net_dma_event_lock
);
3332 net_dma_client
= dma_async_client_register(netdev_dma_event
);
3333 if (net_dma_client
== NULL
)
3336 dma_async_client_chan_request(net_dma_client
, num_online_cpus());
3341 static int __init
netdev_dma_register(void) { return -ENODEV
; }
3342 #endif /* CONFIG_NET_DMA */
3345 * Initialize the DEV module. At boot time this walks the device list and
3346 * unhooks any devices that fail to initialise (normally hardware not
3347 * present) and leaves us with a valid list of present and active devices.
3352 * This is called single threaded during boot, so no need
3353 * to take the rtnl semaphore.
3355 static int __init
net_dev_init(void)
3357 int i
, rc
= -ENOMEM
;
3359 BUG_ON(!dev_boot_phase
);
3363 if (dev_proc_init())
3366 if (netdev_sysfs_init())
3369 INIT_LIST_HEAD(&ptype_all
);
3370 for (i
= 0; i
< 16; i
++)
3371 INIT_LIST_HEAD(&ptype_base
[i
]);
3373 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
3374 INIT_HLIST_HEAD(&dev_name_head
[i
]);
3376 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
3377 INIT_HLIST_HEAD(&dev_index_head
[i
]);
3380 * Initialise the packet receive queues.
3383 for_each_possible_cpu(i
) {
3384 struct softnet_data
*queue
;
3386 queue
= &per_cpu(softnet_data
, i
);
3387 skb_queue_head_init(&queue
->input_pkt_queue
);
3388 queue
->completion_queue
= NULL
;
3389 INIT_LIST_HEAD(&queue
->poll_list
);
3390 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
3391 queue
->backlog_dev
.weight
= weight_p
;
3392 queue
->backlog_dev
.poll
= process_backlog
;
3393 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
3396 netdev_dma_register();
3400 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
3401 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
3403 hotcpu_notifier(dev_cpu_callback
, 0);
3411 subsys_initcall(net_dev_init
);
3413 EXPORT_SYMBOL(__dev_get_by_index
);
3414 EXPORT_SYMBOL(__dev_get_by_name
);
3415 EXPORT_SYMBOL(__dev_remove_pack
);
3416 EXPORT_SYMBOL(dev_valid_name
);
3417 EXPORT_SYMBOL(dev_add_pack
);
3418 EXPORT_SYMBOL(dev_alloc_name
);
3419 EXPORT_SYMBOL(dev_close
);
3420 EXPORT_SYMBOL(dev_get_by_flags
);
3421 EXPORT_SYMBOL(dev_get_by_index
);
3422 EXPORT_SYMBOL(dev_get_by_name
);
3423 EXPORT_SYMBOL(dev_open
);
3424 EXPORT_SYMBOL(dev_queue_xmit
);
3425 EXPORT_SYMBOL(dev_remove_pack
);
3426 EXPORT_SYMBOL(dev_set_allmulti
);
3427 EXPORT_SYMBOL(dev_set_promiscuity
);
3428 EXPORT_SYMBOL(dev_change_flags
);
3429 EXPORT_SYMBOL(dev_set_mtu
);
3430 EXPORT_SYMBOL(dev_set_mac_address
);
3431 EXPORT_SYMBOL(free_netdev
);
3432 EXPORT_SYMBOL(netdev_boot_setup_check
);
3433 EXPORT_SYMBOL(netdev_set_master
);
3434 EXPORT_SYMBOL(netdev_state_change
);
3435 EXPORT_SYMBOL(netif_receive_skb
);
3436 EXPORT_SYMBOL(netif_rx
);
3437 EXPORT_SYMBOL(register_gifconf
);
3438 EXPORT_SYMBOL(register_netdevice
);
3439 EXPORT_SYMBOL(register_netdevice_notifier
);
3440 EXPORT_SYMBOL(skb_checksum_help
);
3441 EXPORT_SYMBOL(synchronize_net
);
3442 EXPORT_SYMBOL(unregister_netdevice
);
3443 EXPORT_SYMBOL(unregister_netdevice_notifier
);
3444 EXPORT_SYMBOL(net_enable_timestamp
);
3445 EXPORT_SYMBOL(net_disable_timestamp
);
3446 EXPORT_SYMBOL(dev_get_flags
);
3448 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3449 EXPORT_SYMBOL(br_handle_frame_hook
);
3450 EXPORT_SYMBOL(br_fdb_get_hook
);
3451 EXPORT_SYMBOL(br_fdb_put_hook
);
3455 EXPORT_SYMBOL(dev_load
);
3458 EXPORT_PER_CPU_SYMBOL(softnet_data
);