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/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/mutex.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>
95 #include <net/net_namespace.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/if_macvlan.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 <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
124 * The list of packet types we will receive (as opposed to discard)
125 * and the routines to invoke.
127 * Why 16. Because with 16 the only overlap we get on a hash of the
128 * low nibble of the protocol value is RARP/SNAP/X.25.
130 * NOTE: That is no longer true with the addition of VLAN tags. Not
131 * sure which should go first, but I bet it won't make much
132 * difference if we are running VLANs. The good news is that
133 * this protocol won't be in the list unless compiled in, so
134 * the average user (w/out VLANs) will not be adversely affected.
151 static DEFINE_SPINLOCK(ptype_lock
);
152 static struct list_head ptype_base
[16] __read_mostly
; /* 16 way hashed list */
153 static struct list_head ptype_all __read_mostly
; /* Taps */
155 #ifdef CONFIG_NET_DMA
157 struct dma_client client
;
159 cpumask_t channel_mask
;
160 struct dma_chan
*channels
[NR_CPUS
];
163 static enum dma_state_client
164 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
165 enum dma_state state
);
167 static struct net_dma net_dma
= {
169 .event_callback
= netdev_dma_event
,
175 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
178 * Pure readers hold dev_base_lock for reading.
180 * Writers must hold the rtnl semaphore while they loop through the
181 * dev_base_head list, and hold dev_base_lock for writing when they do the
182 * actual updates. This allows pure readers to access the list even
183 * while a writer is preparing to update it.
185 * To put it another way, dev_base_lock is held for writing only to
186 * protect against pure readers; the rtnl semaphore provides the
187 * protection against other writers.
189 * See, for example usages, register_netdevice() and
190 * unregister_netdevice(), which must be called with the rtnl
193 DEFINE_RWLOCK(dev_base_lock
);
195 EXPORT_SYMBOL(dev_base_lock
);
197 #define NETDEV_HASHBITS 8
198 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
200 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
202 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
203 return &net
->dev_name_head
[hash
& ((1 << NETDEV_HASHBITS
) - 1)];
206 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
208 return &net
->dev_index_head
[ifindex
& ((1 << NETDEV_HASHBITS
) - 1)];
215 static RAW_NOTIFIER_HEAD(netdev_chain
);
218 * Device drivers call our routines to queue packets here. We empty the
219 * queue in the local softnet handler.
222 DEFINE_PER_CPU(struct softnet_data
, softnet_data
);
225 extern int netdev_sysfs_init(void);
226 extern int netdev_register_sysfs(struct net_device
*);
227 extern void netdev_unregister_sysfs(struct net_device
*);
229 #define netdev_sysfs_init() (0)
230 #define netdev_register_sysfs(dev) (0)
231 #define netdev_unregister_sysfs(dev) do { } while(0)
234 #ifdef CONFIG_DEBUG_LOCK_ALLOC
236 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
237 * according to dev->type
239 static const unsigned short netdev_lock_type
[] =
240 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
241 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
242 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
243 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
244 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
245 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
246 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
247 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
248 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
249 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
250 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
251 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
252 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
253 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_VOID
,
256 static const char *netdev_lock_name
[] =
257 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
258 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
259 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
260 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
261 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
262 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
263 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
264 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
265 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
266 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
267 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
268 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
269 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
270 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
273 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
275 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
279 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
280 if (netdev_lock_type
[i
] == dev_type
)
282 /* the last key is used by default */
283 return ARRAY_SIZE(netdev_lock_type
) - 1;
286 static inline void netdev_set_lockdep_class(spinlock_t
*lock
,
287 unsigned short dev_type
)
291 i
= netdev_lock_pos(dev_type
);
292 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
293 netdev_lock_name
[i
]);
296 static inline void netdev_set_lockdep_class(spinlock_t
*lock
,
297 unsigned short dev_type
)
302 /*******************************************************************************
304 Protocol management and registration routines
306 *******************************************************************************/
309 * Add a protocol ID to the list. Now that the input handler is
310 * smarter we can dispense with all the messy stuff that used to be
313 * BEWARE!!! Protocol handlers, mangling input packets,
314 * MUST BE last in hash buckets and checking protocol handlers
315 * MUST start from promiscuous ptype_all chain in net_bh.
316 * It is true now, do not change it.
317 * Explanation follows: if protocol handler, mangling packet, will
318 * be the first on list, it is not able to sense, that packet
319 * is cloned and should be copied-on-write, so that it will
320 * change it and subsequent readers will get broken packet.
325 * dev_add_pack - add packet handler
326 * @pt: packet type declaration
328 * Add a protocol handler to the networking stack. The passed &packet_type
329 * is linked into kernel lists and may not be freed until it has been
330 * removed from the kernel lists.
332 * This call does not sleep therefore it can not
333 * guarantee all CPU's that are in middle of receiving packets
334 * will see the new packet type (until the next received packet).
337 void dev_add_pack(struct packet_type
*pt
)
341 spin_lock_bh(&ptype_lock
);
342 if (pt
->type
== htons(ETH_P_ALL
))
343 list_add_rcu(&pt
->list
, &ptype_all
);
345 hash
= ntohs(pt
->type
) & 15;
346 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
348 spin_unlock_bh(&ptype_lock
);
352 * __dev_remove_pack - remove packet handler
353 * @pt: packet type declaration
355 * Remove a protocol handler that was previously added to the kernel
356 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
357 * from the kernel lists and can be freed or reused once this function
360 * The packet type might still be in use by receivers
361 * and must not be freed until after all the CPU's have gone
362 * through a quiescent state.
364 void __dev_remove_pack(struct packet_type
*pt
)
366 struct list_head
*head
;
367 struct packet_type
*pt1
;
369 spin_lock_bh(&ptype_lock
);
371 if (pt
->type
== htons(ETH_P_ALL
))
374 head
= &ptype_base
[ntohs(pt
->type
) & 15];
376 list_for_each_entry(pt1
, head
, list
) {
378 list_del_rcu(&pt
->list
);
383 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
385 spin_unlock_bh(&ptype_lock
);
388 * dev_remove_pack - remove packet handler
389 * @pt: packet type declaration
391 * Remove a protocol handler that was previously added to the kernel
392 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
393 * from the kernel lists and can be freed or reused once this function
396 * This call sleeps to guarantee that no CPU is looking at the packet
399 void dev_remove_pack(struct packet_type
*pt
)
401 __dev_remove_pack(pt
);
406 /******************************************************************************
408 Device Boot-time Settings Routines
410 *******************************************************************************/
412 /* Boot time configuration table */
413 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
416 * netdev_boot_setup_add - add new setup entry
417 * @name: name of the device
418 * @map: configured settings for the device
420 * Adds new setup entry to the dev_boot_setup list. The function
421 * returns 0 on error and 1 on success. This is a generic routine to
424 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
426 struct netdev_boot_setup
*s
;
430 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
431 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
432 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
433 strcpy(s
[i
].name
, name
);
434 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
439 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
443 * netdev_boot_setup_check - check boot time settings
444 * @dev: the netdevice
446 * Check boot time settings for the device.
447 * The found settings are set for the device to be used
448 * later in the device probing.
449 * Returns 0 if no settings found, 1 if they are.
451 int netdev_boot_setup_check(struct net_device
*dev
)
453 struct netdev_boot_setup
*s
= dev_boot_setup
;
456 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
457 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
458 !strncmp(dev
->name
, s
[i
].name
, strlen(s
[i
].name
))) {
459 dev
->irq
= s
[i
].map
.irq
;
460 dev
->base_addr
= s
[i
].map
.base_addr
;
461 dev
->mem_start
= s
[i
].map
.mem_start
;
462 dev
->mem_end
= s
[i
].map
.mem_end
;
471 * netdev_boot_base - get address from boot time settings
472 * @prefix: prefix for network device
473 * @unit: id for network device
475 * Check boot time settings for the base address of device.
476 * The found settings are set for the device to be used
477 * later in the device probing.
478 * Returns 0 if no settings found.
480 unsigned long netdev_boot_base(const char *prefix
, int unit
)
482 const struct netdev_boot_setup
*s
= dev_boot_setup
;
486 sprintf(name
, "%s%d", prefix
, unit
);
489 * If device already registered then return base of 1
490 * to indicate not to probe for this interface
492 if (__dev_get_by_name(&init_net
, name
))
495 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
496 if (!strcmp(name
, s
[i
].name
))
497 return s
[i
].map
.base_addr
;
502 * Saves at boot time configured settings for any netdevice.
504 int __init
netdev_boot_setup(char *str
)
509 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
514 memset(&map
, 0, sizeof(map
));
518 map
.base_addr
= ints
[2];
520 map
.mem_start
= ints
[3];
522 map
.mem_end
= ints
[4];
524 /* Add new entry to the list */
525 return netdev_boot_setup_add(str
, &map
);
528 __setup("netdev=", netdev_boot_setup
);
530 /*******************************************************************************
532 Device Interface Subroutines
534 *******************************************************************************/
537 * __dev_get_by_name - find a device by its name
538 * @name: name to find
540 * Find an interface by name. Must be called under RTNL semaphore
541 * or @dev_base_lock. If the name is found a pointer to the device
542 * is returned. If the name is not found then %NULL is returned. The
543 * reference counters are not incremented so the caller must be
544 * careful with locks.
547 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
549 struct hlist_node
*p
;
551 hlist_for_each(p
, dev_name_hash(net
, name
)) {
552 struct net_device
*dev
553 = hlist_entry(p
, struct net_device
, name_hlist
);
554 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
561 * dev_get_by_name - find a device by its name
562 * @name: name to find
564 * Find an interface by name. This can be called from any
565 * context and does its own locking. The returned handle has
566 * the usage count incremented and the caller must use dev_put() to
567 * release it when it is no longer needed. %NULL is returned if no
568 * matching device is found.
571 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
573 struct net_device
*dev
;
575 read_lock(&dev_base_lock
);
576 dev
= __dev_get_by_name(net
, name
);
579 read_unlock(&dev_base_lock
);
584 * __dev_get_by_index - find a device by its ifindex
585 * @ifindex: index of device
587 * Search for an interface by index. Returns %NULL if the device
588 * is not found or a pointer to the device. The device has not
589 * had its reference counter increased so the caller must be careful
590 * about locking. The caller must hold either the RTNL semaphore
594 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
596 struct hlist_node
*p
;
598 hlist_for_each(p
, dev_index_hash(net
, ifindex
)) {
599 struct net_device
*dev
600 = hlist_entry(p
, struct net_device
, index_hlist
);
601 if (dev
->ifindex
== ifindex
)
609 * dev_get_by_index - find a device by its ifindex
610 * @ifindex: index of device
612 * Search for an interface by index. Returns NULL if the device
613 * is not found or a pointer to the device. The device returned has
614 * had a reference added and the pointer is safe until the user calls
615 * dev_put to indicate they have finished with it.
618 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
620 struct net_device
*dev
;
622 read_lock(&dev_base_lock
);
623 dev
= __dev_get_by_index(net
, ifindex
);
626 read_unlock(&dev_base_lock
);
631 * dev_getbyhwaddr - find a device by its hardware address
632 * @type: media type of device
633 * @ha: hardware address
635 * Search for an interface by MAC address. Returns NULL if the device
636 * is not found or a pointer to the device. The caller must hold the
637 * rtnl semaphore. The returned device has not had its ref count increased
638 * and the caller must therefore be careful about locking
641 * If the API was consistent this would be __dev_get_by_hwaddr
644 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
646 struct net_device
*dev
;
650 for_each_netdev(&init_net
, dev
)
651 if (dev
->type
== type
&&
652 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
658 EXPORT_SYMBOL(dev_getbyhwaddr
);
660 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
662 struct net_device
*dev
;
665 for_each_netdev(net
, dev
)
666 if (dev
->type
== type
)
672 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
674 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
676 struct net_device
*dev
;
679 dev
= __dev_getfirstbyhwtype(net
, type
);
686 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
689 * dev_get_by_flags - find any device with given flags
690 * @if_flags: IFF_* values
691 * @mask: bitmask of bits in if_flags to check
693 * Search for any interface with the given flags. Returns NULL if a device
694 * is not found or a pointer to the device. The device returned has
695 * had a reference added and the pointer is safe until the user calls
696 * dev_put to indicate they have finished with it.
699 struct net_device
* dev_get_by_flags(struct net
*net
, unsigned short if_flags
, unsigned short mask
)
701 struct net_device
*dev
, *ret
;
704 read_lock(&dev_base_lock
);
705 for_each_netdev(net
, dev
) {
706 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
712 read_unlock(&dev_base_lock
);
717 * dev_valid_name - check if name is okay for network device
720 * Network device names need to be valid file names to
721 * to allow sysfs to work. We also disallow any kind of
724 int dev_valid_name(const char *name
)
728 if (strlen(name
) >= IFNAMSIZ
)
730 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
734 if (*name
== '/' || isspace(*name
))
742 * __dev_alloc_name - allocate a name for a device
743 * @net: network namespace to allocate the device name in
744 * @name: name format string
745 * @buf: scratch buffer and result name string
747 * Passed a format string - eg "lt%d" it will try and find a suitable
748 * id. It scans list of devices to build up a free map, then chooses
749 * the first empty slot. The caller must hold the dev_base or rtnl lock
750 * while allocating the name and adding the device in order to avoid
752 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
753 * Returns the number of the unit assigned or a negative errno code.
756 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
760 const int max_netdevices
= 8*PAGE_SIZE
;
762 struct net_device
*d
;
764 p
= strnchr(name
, IFNAMSIZ
-1, '%');
767 * Verify the string as this thing may have come from
768 * the user. There must be either one "%d" and no other "%"
771 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
774 /* Use one page as a bit array of possible slots */
775 inuse
= (long *) get_zeroed_page(GFP_ATOMIC
);
779 for_each_netdev(net
, d
) {
780 if (!sscanf(d
->name
, name
, &i
))
782 if (i
< 0 || i
>= max_netdevices
)
785 /* avoid cases where sscanf is not exact inverse of printf */
786 snprintf(buf
, IFNAMSIZ
, name
, i
);
787 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
791 i
= find_first_zero_bit(inuse
, max_netdevices
);
792 free_page((unsigned long) inuse
);
795 snprintf(buf
, IFNAMSIZ
, name
, i
);
796 if (!__dev_get_by_name(net
, buf
))
799 /* It is possible to run out of possible slots
800 * when the name is long and there isn't enough space left
801 * for the digits, or if all bits are used.
807 * dev_alloc_name - allocate a name for a device
809 * @name: name format string
811 * Passed a format string - eg "lt%d" it will try and find a suitable
812 * id. It scans list of devices to build up a free map, then chooses
813 * the first empty slot. The caller must hold the dev_base or rtnl lock
814 * while allocating the name and adding the device in order to avoid
816 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
817 * Returns the number of the unit assigned or a negative errno code.
820 int dev_alloc_name(struct net_device
*dev
, const char *name
)
826 BUG_ON(!dev
->nd_net
);
828 ret
= __dev_alloc_name(net
, name
, buf
);
830 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
836 * dev_change_name - change name of a device
838 * @newname: name (or format string) must be at least IFNAMSIZ
840 * Change name of a device, can pass format strings "eth%d".
843 int dev_change_name(struct net_device
*dev
, char *newname
)
845 char oldname
[IFNAMSIZ
];
851 BUG_ON(!dev
->nd_net
);
854 if (dev
->flags
& IFF_UP
)
857 if (!dev_valid_name(newname
))
860 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
862 if (strchr(newname
, '%')) {
863 err
= dev_alloc_name(dev
, newname
);
866 strcpy(newname
, dev
->name
);
868 else if (__dev_get_by_name(net
, newname
))
871 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
874 device_rename(&dev
->dev
, dev
->name
);
876 write_lock_bh(&dev_base_lock
);
877 hlist_del(&dev
->name_hlist
);
878 hlist_add_head(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
879 write_unlock_bh(&dev_base_lock
);
881 ret
= raw_notifier_call_chain(&netdev_chain
, NETDEV_CHANGENAME
, dev
);
882 ret
= notifier_to_errno(ret
);
887 "%s: name change rollback failed: %d.\n",
891 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
900 * netdev_features_change - device changes features
901 * @dev: device to cause notification
903 * Called to indicate a device has changed features.
905 void netdev_features_change(struct net_device
*dev
)
907 raw_notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
909 EXPORT_SYMBOL(netdev_features_change
);
912 * netdev_state_change - device changes state
913 * @dev: device to cause notification
915 * Called to indicate a device has changed state. This function calls
916 * the notifier chains for netdev_chain and sends a NEWLINK message
917 * to the routing socket.
919 void netdev_state_change(struct net_device
*dev
)
921 if (dev
->flags
& IFF_UP
) {
922 raw_notifier_call_chain(&netdev_chain
,
924 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
929 * dev_load - load a network module
930 * @name: name of interface
932 * If a network interface is not present and the process has suitable
933 * privileges this function loads the module. If module loading is not
934 * available in this kernel then it becomes a nop.
937 void dev_load(struct net
*net
, const char *name
)
939 struct net_device
*dev
;
941 read_lock(&dev_base_lock
);
942 dev
= __dev_get_by_name(net
, name
);
943 read_unlock(&dev_base_lock
);
945 if (!dev
&& capable(CAP_SYS_MODULE
))
946 request_module("%s", name
);
949 static int default_rebuild_header(struct sk_buff
*skb
)
951 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
952 skb
->dev
? skb
->dev
->name
: "NULL!!!");
958 * dev_open - prepare an interface for use.
959 * @dev: device to open
961 * Takes a device from down to up state. The device's private open
962 * function is invoked and then the multicast lists are loaded. Finally
963 * the device is moved into the up state and a %NETDEV_UP message is
964 * sent to the netdev notifier chain.
966 * Calling this function on an active interface is a nop. On a failure
967 * a negative errno code is returned.
969 int dev_open(struct net_device
*dev
)
977 if (dev
->flags
& IFF_UP
)
981 * Is it even present?
983 if (!netif_device_present(dev
))
987 * Call device private open method
989 set_bit(__LINK_STATE_START
, &dev
->state
);
991 ret
= dev
->open(dev
);
993 clear_bit(__LINK_STATE_START
, &dev
->state
);
997 * If it went open OK then:
1004 dev
->flags
|= IFF_UP
;
1007 * Initialize multicasting status
1009 dev_set_rx_mode(dev
);
1012 * Wakeup transmit queue engine
1017 * ... and announce new interface.
1019 raw_notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
1025 * dev_close - shutdown an interface.
1026 * @dev: device to shutdown
1028 * This function moves an active device into down state. A
1029 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1030 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1033 int dev_close(struct net_device
*dev
)
1035 if (!(dev
->flags
& IFF_UP
))
1039 * Tell people we are going down, so that they can
1040 * prepare to death, when device is still operating.
1042 raw_notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
1044 dev_deactivate(dev
);
1046 clear_bit(__LINK_STATE_START
, &dev
->state
);
1048 /* Synchronize to scheduled poll. We cannot touch poll list,
1049 * it can be even on different cpu. So just clear netif_running().
1051 * dev->stop() will invoke napi_disable() on all of it's
1052 * napi_struct instances on this device.
1054 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1057 * Call the device specific close. This cannot fail.
1058 * Only if device is UP
1060 * We allow it to be called even after a DETACH hot-plug
1067 * Device is now down.
1070 dev
->flags
&= ~IFF_UP
;
1073 * Tell people we are down
1075 raw_notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
1081 static int dev_boot_phase
= 1;
1084 * Device change register/unregister. These are not inline or static
1085 * as we export them to the world.
1089 * register_netdevice_notifier - register a network notifier block
1092 * Register a notifier to be called when network device events occur.
1093 * The notifier passed is linked into the kernel structures and must
1094 * not be reused until it has been unregistered. A negative errno code
1095 * is returned on a failure.
1097 * When registered all registration and up events are replayed
1098 * to the new notifier to allow device to have a race free
1099 * view of the network device list.
1102 int register_netdevice_notifier(struct notifier_block
*nb
)
1104 struct net_device
*dev
;
1105 struct net_device
*last
;
1110 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1116 for_each_netdev(net
, dev
) {
1117 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1118 err
= notifier_to_errno(err
);
1122 if (!(dev
->flags
& IFF_UP
))
1125 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1136 for_each_netdev(net
, dev
) {
1140 if (dev
->flags
& IFF_UP
) {
1141 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1142 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1144 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1151 * unregister_netdevice_notifier - unregister a network notifier block
1154 * Unregister a notifier previously registered by
1155 * register_netdevice_notifier(). The notifier is unlinked into the
1156 * kernel structures and may then be reused. A negative errno code
1157 * is returned on a failure.
1160 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1165 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1171 * call_netdevice_notifiers - call all network notifier blocks
1172 * @val: value passed unmodified to notifier function
1173 * @v: pointer passed unmodified to notifier function
1175 * Call all network notifier blocks. Parameters and return value
1176 * are as for raw_notifier_call_chain().
1179 int call_netdevice_notifiers(unsigned long val
, void *v
)
1181 return raw_notifier_call_chain(&netdev_chain
, val
, v
);
1184 /* When > 0 there are consumers of rx skb time stamps */
1185 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1187 void net_enable_timestamp(void)
1189 atomic_inc(&netstamp_needed
);
1192 void net_disable_timestamp(void)
1194 atomic_dec(&netstamp_needed
);
1197 static inline void net_timestamp(struct sk_buff
*skb
)
1199 if (atomic_read(&netstamp_needed
))
1200 __net_timestamp(skb
);
1202 skb
->tstamp
.tv64
= 0;
1206 * Support routine. Sends outgoing frames to any network
1207 * taps currently in use.
1210 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1212 struct packet_type
*ptype
;
1217 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1218 /* Never send packets back to the socket
1219 * they originated from - MvS (miquels@drinkel.ow.org)
1221 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1222 (ptype
->af_packet_priv
== NULL
||
1223 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1224 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1228 /* skb->nh should be correctly
1229 set by sender, so that the second statement is
1230 just protection against buggy protocols.
1232 skb_reset_mac_header(skb2
);
1234 if (skb_network_header(skb2
) < skb2
->data
||
1235 skb2
->network_header
> skb2
->tail
) {
1236 if (net_ratelimit())
1237 printk(KERN_CRIT
"protocol %04x is "
1239 skb2
->protocol
, dev
->name
);
1240 skb_reset_network_header(skb2
);
1243 skb2
->transport_header
= skb2
->network_header
;
1244 skb2
->pkt_type
= PACKET_OUTGOING
;
1245 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1252 void __netif_schedule(struct net_device
*dev
)
1254 if (!test_and_set_bit(__LINK_STATE_SCHED
, &dev
->state
)) {
1255 unsigned long flags
;
1256 struct softnet_data
*sd
;
1258 local_irq_save(flags
);
1259 sd
= &__get_cpu_var(softnet_data
);
1260 dev
->next_sched
= sd
->output_queue
;
1261 sd
->output_queue
= dev
;
1262 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1263 local_irq_restore(flags
);
1266 EXPORT_SYMBOL(__netif_schedule
);
1268 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1270 if (atomic_dec_and_test(&skb
->users
)) {
1271 struct softnet_data
*sd
;
1272 unsigned long flags
;
1274 local_irq_save(flags
);
1275 sd
= &__get_cpu_var(softnet_data
);
1276 skb
->next
= sd
->completion_queue
;
1277 sd
->completion_queue
= skb
;
1278 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1279 local_irq_restore(flags
);
1282 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1284 void dev_kfree_skb_any(struct sk_buff
*skb
)
1286 if (in_irq() || irqs_disabled())
1287 dev_kfree_skb_irq(skb
);
1291 EXPORT_SYMBOL(dev_kfree_skb_any
);
1295 * netif_device_detach - mark device as removed
1296 * @dev: network device
1298 * Mark device as removed from system and therefore no longer available.
1300 void netif_device_detach(struct net_device
*dev
)
1302 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1303 netif_running(dev
)) {
1304 netif_stop_queue(dev
);
1307 EXPORT_SYMBOL(netif_device_detach
);
1310 * netif_device_attach - mark device as attached
1311 * @dev: network device
1313 * Mark device as attached from system and restart if needed.
1315 void netif_device_attach(struct net_device
*dev
)
1317 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1318 netif_running(dev
)) {
1319 netif_wake_queue(dev
);
1320 __netdev_watchdog_up(dev
);
1323 EXPORT_SYMBOL(netif_device_attach
);
1327 * Invalidate hardware checksum when packet is to be mangled, and
1328 * complete checksum manually on outgoing path.
1330 int skb_checksum_help(struct sk_buff
*skb
)
1333 int ret
= 0, offset
;
1335 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1336 goto out_set_summed
;
1338 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1339 /* Let GSO fix up the checksum. */
1340 goto out_set_summed
;
1343 if (skb_cloned(skb
)) {
1344 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1349 offset
= skb
->csum_start
- skb_headroom(skb
);
1350 BUG_ON(offset
> (int)skb
->len
);
1351 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1353 offset
= skb_headlen(skb
) - offset
;
1354 BUG_ON(offset
<= 0);
1355 BUG_ON(skb
->csum_offset
+ 2 > offset
);
1357 *(__sum16
*)(skb
->head
+ skb
->csum_start
+ skb
->csum_offset
) =
1360 skb
->ip_summed
= CHECKSUM_NONE
;
1366 * skb_gso_segment - Perform segmentation on skb.
1367 * @skb: buffer to segment
1368 * @features: features for the output path (see dev->features)
1370 * This function segments the given skb and returns a list of segments.
1372 * It may return NULL if the skb requires no segmentation. This is
1373 * only possible when GSO is used for verifying header integrity.
1375 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1377 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1378 struct packet_type
*ptype
;
1379 __be16 type
= skb
->protocol
;
1382 BUG_ON(skb_shinfo(skb
)->frag_list
);
1384 skb_reset_mac_header(skb
);
1385 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1386 __skb_pull(skb
, skb
->mac_len
);
1388 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1389 if (skb_header_cloned(skb
) &&
1390 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1391 return ERR_PTR(err
);
1395 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
) & 15], list
) {
1396 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1397 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1398 err
= ptype
->gso_send_check(skb
);
1399 segs
= ERR_PTR(err
);
1400 if (err
|| skb_gso_ok(skb
, features
))
1402 __skb_push(skb
, (skb
->data
-
1403 skb_network_header(skb
)));
1405 segs
= ptype
->gso_segment(skb
, features
);
1411 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1416 EXPORT_SYMBOL(skb_gso_segment
);
1418 /* Take action when hardware reception checksum errors are detected. */
1420 void netdev_rx_csum_fault(struct net_device
*dev
)
1422 if (net_ratelimit()) {
1423 printk(KERN_ERR
"%s: hw csum failure.\n",
1424 dev
? dev
->name
: "<unknown>");
1428 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1431 /* Actually, we should eliminate this check as soon as we know, that:
1432 * 1. IOMMU is present and allows to map all the memory.
1433 * 2. No high memory really exists on this machine.
1436 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1438 #ifdef CONFIG_HIGHMEM
1441 if (dev
->features
& NETIF_F_HIGHDMA
)
1444 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1445 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1453 void (*destructor
)(struct sk_buff
*skb
);
1456 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1458 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1460 struct dev_gso_cb
*cb
;
1463 struct sk_buff
*nskb
= skb
->next
;
1465 skb
->next
= nskb
->next
;
1468 } while (skb
->next
);
1470 cb
= DEV_GSO_CB(skb
);
1472 cb
->destructor(skb
);
1476 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1477 * @skb: buffer to segment
1479 * This function segments the given skb and stores the list of segments
1482 static int dev_gso_segment(struct sk_buff
*skb
)
1484 struct net_device
*dev
= skb
->dev
;
1485 struct sk_buff
*segs
;
1486 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1489 segs
= skb_gso_segment(skb
, features
);
1491 /* Verifying header integrity only. */
1495 if (unlikely(IS_ERR(segs
)))
1496 return PTR_ERR(segs
);
1499 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1500 skb
->destructor
= dev_gso_skb_destructor
;
1505 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1507 if (likely(!skb
->next
)) {
1508 if (!list_empty(&ptype_all
))
1509 dev_queue_xmit_nit(skb
, dev
);
1511 if (netif_needs_gso(dev
, skb
)) {
1512 if (unlikely(dev_gso_segment(skb
)))
1518 return dev
->hard_start_xmit(skb
, dev
);
1523 struct sk_buff
*nskb
= skb
->next
;
1526 skb
->next
= nskb
->next
;
1528 rc
= dev
->hard_start_xmit(nskb
, dev
);
1530 nskb
->next
= skb
->next
;
1534 if (unlikely((netif_queue_stopped(dev
) ||
1535 netif_subqueue_stopped(dev
, skb
->queue_mapping
)) &&
1537 return NETDEV_TX_BUSY
;
1538 } while (skb
->next
);
1540 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1547 #define HARD_TX_LOCK(dev, cpu) { \
1548 if ((dev->features & NETIF_F_LLTX) == 0) { \
1549 netif_tx_lock(dev); \
1553 #define HARD_TX_UNLOCK(dev) { \
1554 if ((dev->features & NETIF_F_LLTX) == 0) { \
1555 netif_tx_unlock(dev); \
1560 * dev_queue_xmit - transmit a buffer
1561 * @skb: buffer to transmit
1563 * Queue a buffer for transmission to a network device. The caller must
1564 * have set the device and priority and built the buffer before calling
1565 * this function. The function can be called from an interrupt.
1567 * A negative errno code is returned on a failure. A success does not
1568 * guarantee the frame will be transmitted as it may be dropped due
1569 * to congestion or traffic shaping.
1571 * -----------------------------------------------------------------------------------
1572 * I notice this method can also return errors from the queue disciplines,
1573 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1576 * Regardless of the return value, the skb is consumed, so it is currently
1577 * difficult to retry a send to this method. (You can bump the ref count
1578 * before sending to hold a reference for retry if you are careful.)
1580 * When calling this method, interrupts MUST be enabled. This is because
1581 * the BH enable code must have IRQs enabled so that it will not deadlock.
1585 int dev_queue_xmit(struct sk_buff
*skb
)
1587 struct net_device
*dev
= skb
->dev
;
1591 /* GSO will handle the following emulations directly. */
1592 if (netif_needs_gso(dev
, skb
))
1595 if (skb_shinfo(skb
)->frag_list
&&
1596 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1597 __skb_linearize(skb
))
1600 /* Fragmented skb is linearized if device does not support SG,
1601 * or if at least one of fragments is in highmem and device
1602 * does not support DMA from it.
1604 if (skb_shinfo(skb
)->nr_frags
&&
1605 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1606 __skb_linearize(skb
))
1609 /* If packet is not checksummed and device does not support
1610 * checksumming for this protocol, complete checksumming here.
1612 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1613 skb_set_transport_header(skb
, skb
->csum_start
-
1616 if (!(dev
->features
& NETIF_F_GEN_CSUM
) &&
1617 !((dev
->features
& NETIF_F_IP_CSUM
) &&
1618 skb
->protocol
== htons(ETH_P_IP
)) &&
1619 !((dev
->features
& NETIF_F_IPV6_CSUM
) &&
1620 skb
->protocol
== htons(ETH_P_IPV6
)))
1621 if (skb_checksum_help(skb
))
1626 spin_lock_prefetch(&dev
->queue_lock
);
1628 /* Disable soft irqs for various locks below. Also
1629 * stops preemption for RCU.
1633 /* Updates of qdisc are serialized by queue_lock.
1634 * The struct Qdisc which is pointed to by qdisc is now a
1635 * rcu structure - it may be accessed without acquiring
1636 * a lock (but the structure may be stale.) The freeing of the
1637 * qdisc will be deferred until it's known that there are no
1638 * more references to it.
1640 * If the qdisc has an enqueue function, we still need to
1641 * hold the queue_lock before calling it, since queue_lock
1642 * also serializes access to the device queue.
1645 q
= rcu_dereference(dev
->qdisc
);
1646 #ifdef CONFIG_NET_CLS_ACT
1647 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1650 /* Grab device queue */
1651 spin_lock(&dev
->queue_lock
);
1654 /* reset queue_mapping to zero */
1655 skb
->queue_mapping
= 0;
1656 rc
= q
->enqueue(skb
, q
);
1658 spin_unlock(&dev
->queue_lock
);
1660 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1663 spin_unlock(&dev
->queue_lock
);
1666 /* The device has no queue. Common case for software devices:
1667 loopback, all the sorts of tunnels...
1669 Really, it is unlikely that netif_tx_lock protection is necessary
1670 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1672 However, it is possible, that they rely on protection
1675 Check this and shot the lock. It is not prone from deadlocks.
1676 Either shot noqueue qdisc, it is even simpler 8)
1678 if (dev
->flags
& IFF_UP
) {
1679 int cpu
= smp_processor_id(); /* ok because BHs are off */
1681 if (dev
->xmit_lock_owner
!= cpu
) {
1683 HARD_TX_LOCK(dev
, cpu
);
1685 if (!netif_queue_stopped(dev
) &&
1686 !netif_subqueue_stopped(dev
, skb
->queue_mapping
)) {
1688 if (!dev_hard_start_xmit(skb
, dev
)) {
1689 HARD_TX_UNLOCK(dev
);
1693 HARD_TX_UNLOCK(dev
);
1694 if (net_ratelimit())
1695 printk(KERN_CRIT
"Virtual device %s asks to "
1696 "queue packet!\n", dev
->name
);
1698 /* Recursion is detected! It is possible,
1700 if (net_ratelimit())
1701 printk(KERN_CRIT
"Dead loop on virtual device "
1702 "%s, fix it urgently!\n", dev
->name
);
1707 rcu_read_unlock_bh();
1713 rcu_read_unlock_bh();
1718 /*=======================================================================
1720 =======================================================================*/
1722 int netdev_max_backlog __read_mostly
= 1000;
1723 int netdev_budget __read_mostly
= 300;
1724 int weight_p __read_mostly
= 64; /* old backlog weight */
1726 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1730 * netif_rx - post buffer to the network code
1731 * @skb: buffer to post
1733 * This function receives a packet from a device driver and queues it for
1734 * the upper (protocol) levels to process. It always succeeds. The buffer
1735 * may be dropped during processing for congestion control or by the
1739 * NET_RX_SUCCESS (no congestion)
1740 * NET_RX_CN_LOW (low congestion)
1741 * NET_RX_CN_MOD (moderate congestion)
1742 * NET_RX_CN_HIGH (high congestion)
1743 * NET_RX_DROP (packet was dropped)
1747 int netif_rx(struct sk_buff
*skb
)
1749 struct softnet_data
*queue
;
1750 unsigned long flags
;
1752 /* if netpoll wants it, pretend we never saw it */
1753 if (netpoll_rx(skb
))
1756 if (!skb
->tstamp
.tv64
)
1760 * The code is rearranged so that the path is the most
1761 * short when CPU is congested, but is still operating.
1763 local_irq_save(flags
);
1764 queue
= &__get_cpu_var(softnet_data
);
1766 __get_cpu_var(netdev_rx_stat
).total
++;
1767 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1768 if (queue
->input_pkt_queue
.qlen
) {
1771 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1772 local_irq_restore(flags
);
1773 return NET_RX_SUCCESS
;
1776 napi_schedule(&queue
->backlog
);
1780 __get_cpu_var(netdev_rx_stat
).dropped
++;
1781 local_irq_restore(flags
);
1787 int netif_rx_ni(struct sk_buff
*skb
)
1792 err
= netif_rx(skb
);
1793 if (local_softirq_pending())
1800 EXPORT_SYMBOL(netif_rx_ni
);
1802 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1804 struct net_device
*dev
= skb
->dev
;
1807 if (skb_bond_should_drop(skb
)) {
1811 skb
->dev
= dev
->master
;
1818 static void net_tx_action(struct softirq_action
*h
)
1820 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1822 if (sd
->completion_queue
) {
1823 struct sk_buff
*clist
;
1825 local_irq_disable();
1826 clist
= sd
->completion_queue
;
1827 sd
->completion_queue
= NULL
;
1831 struct sk_buff
*skb
= clist
;
1832 clist
= clist
->next
;
1834 BUG_TRAP(!atomic_read(&skb
->users
));
1839 if (sd
->output_queue
) {
1840 struct net_device
*head
;
1842 local_irq_disable();
1843 head
= sd
->output_queue
;
1844 sd
->output_queue
= NULL
;
1848 struct net_device
*dev
= head
;
1849 head
= head
->next_sched
;
1851 smp_mb__before_clear_bit();
1852 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1854 if (spin_trylock(&dev
->queue_lock
)) {
1856 spin_unlock(&dev
->queue_lock
);
1858 netif_schedule(dev
);
1864 static inline int deliver_skb(struct sk_buff
*skb
,
1865 struct packet_type
*pt_prev
,
1866 struct net_device
*orig_dev
)
1868 atomic_inc(&skb
->users
);
1869 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1872 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1873 /* These hooks defined here for ATM */
1875 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1876 unsigned char *addr
);
1877 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
1880 * If bridge module is loaded call bridging hook.
1881 * returns NULL if packet was consumed.
1883 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
1884 struct sk_buff
*skb
) __read_mostly
;
1885 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
1886 struct packet_type
**pt_prev
, int *ret
,
1887 struct net_device
*orig_dev
)
1889 struct net_bridge_port
*port
;
1891 if (skb
->pkt_type
== PACKET_LOOPBACK
||
1892 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
1896 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1900 return br_handle_frame_hook(port
, skb
);
1903 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1906 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1907 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
1908 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
1910 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
1911 struct packet_type
**pt_prev
,
1913 struct net_device
*orig_dev
)
1915 if (skb
->dev
->macvlan_port
== NULL
)
1919 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1922 return macvlan_handle_frame_hook(skb
);
1925 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1928 #ifdef CONFIG_NET_CLS_ACT
1929 /* TODO: Maybe we should just force sch_ingress to be compiled in
1930 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1931 * a compare and 2 stores extra right now if we dont have it on
1932 * but have CONFIG_NET_CLS_ACT
1933 * NOTE: This doesnt stop any functionality; if you dont have
1934 * the ingress scheduler, you just cant add policies on ingress.
1937 static int ing_filter(struct sk_buff
*skb
)
1940 struct net_device
*dev
= skb
->dev
;
1941 int result
= TC_ACT_OK
;
1943 if (dev
->qdisc_ingress
) {
1944 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1945 if (MAX_RED_LOOP
< ttl
++) {
1946 printk(KERN_WARNING
"Redir loop detected Dropping packet (%d->%d)\n",
1947 skb
->iif
, skb
->dev
->ifindex
);
1951 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1953 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1955 spin_lock(&dev
->ingress_lock
);
1956 if ((q
= dev
->qdisc_ingress
) != NULL
)
1957 result
= q
->enqueue(skb
, q
);
1958 spin_unlock(&dev
->ingress_lock
);
1966 int netif_receive_skb(struct sk_buff
*skb
)
1968 struct packet_type
*ptype
, *pt_prev
;
1969 struct net_device
*orig_dev
;
1970 int ret
= NET_RX_DROP
;
1973 /* if we've gotten here through NAPI, check netpoll */
1974 if (netpoll_receive_skb(skb
))
1977 if (!skb
->tstamp
.tv64
)
1981 skb
->iif
= skb
->dev
->ifindex
;
1983 orig_dev
= skb_bond(skb
);
1988 __get_cpu_var(netdev_rx_stat
).total
++;
1990 skb_reset_network_header(skb
);
1991 skb_reset_transport_header(skb
);
1992 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1998 #ifdef CONFIG_NET_CLS_ACT
1999 if (skb
->tc_verd
& TC_NCLS
) {
2000 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2005 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2006 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
2008 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2013 #ifdef CONFIG_NET_CLS_ACT
2015 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2016 pt_prev
= NULL
; /* noone else should process this after*/
2018 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2021 ret
= ing_filter(skb
);
2023 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
2032 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
2035 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
2039 type
= skb
->protocol
;
2040 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
2041 if (ptype
->type
== type
&&
2042 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
2044 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2050 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2053 /* Jamal, now you will not able to escape explaining
2054 * me how you were going to use this. :-)
2064 static int process_backlog(struct napi_struct
*napi
, int quota
)
2067 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2068 unsigned long start_time
= jiffies
;
2070 napi
->weight
= weight_p
;
2072 struct sk_buff
*skb
;
2073 struct net_device
*dev
;
2075 local_irq_disable();
2076 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
2078 __napi_complete(napi
);
2087 netif_receive_skb(skb
);
2090 } while (++work
< quota
&& jiffies
== start_time
);
2096 * __napi_schedule - schedule for receive
2097 * @napi: entry to schedule
2099 * The entry's receive function will be scheduled to run
2101 void fastcall
__napi_schedule(struct napi_struct
*n
)
2103 unsigned long flags
;
2105 local_irq_save(flags
);
2106 list_add_tail(&n
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
2107 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2108 local_irq_restore(flags
);
2110 EXPORT_SYMBOL(__napi_schedule
);
2113 static void net_rx_action(struct softirq_action
*h
)
2115 struct list_head
*list
= &__get_cpu_var(softnet_data
).poll_list
;
2116 unsigned long start_time
= jiffies
;
2117 int budget
= netdev_budget
;
2120 local_irq_disable();
2122 while (!list_empty(list
)) {
2123 struct napi_struct
*n
;
2126 /* If softirq window is exhuasted then punt.
2128 * Note that this is a slight policy change from the
2129 * previous NAPI code, which would allow up to 2
2130 * jiffies to pass before breaking out. The test
2131 * used to be "jiffies - start_time > 1".
2133 if (unlikely(budget
<= 0 || jiffies
!= start_time
))
2138 /* Even though interrupts have been re-enabled, this
2139 * access is safe because interrupts can only add new
2140 * entries to the tail of this list, and only ->poll()
2141 * calls can remove this head entry from the list.
2143 n
= list_entry(list
->next
, struct napi_struct
, poll_list
);
2145 have
= netpoll_poll_lock(n
);
2149 work
= n
->poll(n
, weight
);
2151 WARN_ON_ONCE(work
> weight
);
2155 local_irq_disable();
2157 /* Drivers must not modify the NAPI state if they
2158 * consume the entire weight. In such cases this code
2159 * still "owns" the NAPI instance and therefore can
2160 * move the instance around on the list at-will.
2162 if (unlikely(work
== weight
))
2163 list_move_tail(&n
->poll_list
, list
);
2165 netpoll_poll_unlock(have
);
2170 #ifdef CONFIG_NET_DMA
2172 * There may not be any more sk_buffs coming right now, so push
2173 * any pending DMA copies to hardware
2175 if (!cpus_empty(net_dma
.channel_mask
)) {
2177 for_each_cpu_mask(chan_idx
, net_dma
.channel_mask
) {
2178 struct dma_chan
*chan
= net_dma
.channels
[chan_idx
];
2180 dma_async_memcpy_issue_pending(chan
);
2188 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2189 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2193 static gifconf_func_t
* gifconf_list
[NPROTO
];
2196 * register_gifconf - register a SIOCGIF handler
2197 * @family: Address family
2198 * @gifconf: Function handler
2200 * Register protocol dependent address dumping routines. The handler
2201 * that is passed must not be freed or reused until it has been replaced
2202 * by another handler.
2204 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2206 if (family
>= NPROTO
)
2208 gifconf_list
[family
] = gifconf
;
2214 * Map an interface index to its name (SIOCGIFNAME)
2218 * We need this ioctl for efficient implementation of the
2219 * if_indextoname() function required by the IPv6 API. Without
2220 * it, we would have to search all the interfaces to find a
2224 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
2226 struct net_device
*dev
;
2230 * Fetch the caller's info block.
2233 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2236 read_lock(&dev_base_lock
);
2237 dev
= __dev_get_by_index(net
, ifr
.ifr_ifindex
);
2239 read_unlock(&dev_base_lock
);
2243 strcpy(ifr
.ifr_name
, dev
->name
);
2244 read_unlock(&dev_base_lock
);
2246 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2252 * Perform a SIOCGIFCONF call. This structure will change
2253 * size eventually, and there is nothing I can do about it.
2254 * Thus we will need a 'compatibility mode'.
2257 static int dev_ifconf(struct net
*net
, char __user
*arg
)
2260 struct net_device
*dev
;
2267 * Fetch the caller's info block.
2270 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2277 * Loop over the interfaces, and write an info block for each.
2281 for_each_netdev(net
, dev
) {
2282 for (i
= 0; i
< NPROTO
; i
++) {
2283 if (gifconf_list
[i
]) {
2286 done
= gifconf_list
[i
](dev
, NULL
, 0);
2288 done
= gifconf_list
[i
](dev
, pos
+ total
,
2298 * All done. Write the updated control block back to the caller.
2300 ifc
.ifc_len
= total
;
2303 * Both BSD and Solaris return 0 here, so we do too.
2305 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2308 #ifdef CONFIG_PROC_FS
2310 * This is invoked by the /proc filesystem handler to display a device
2313 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2315 struct net
*net
= seq
->private;
2317 struct net_device
*dev
;
2319 read_lock(&dev_base_lock
);
2321 return SEQ_START_TOKEN
;
2324 for_each_netdev(net
, dev
)
2331 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2333 struct net
*net
= seq
->private;
2335 return v
== SEQ_START_TOKEN
?
2336 first_net_device(net
) : next_net_device((struct net_device
*)v
);
2339 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2341 read_unlock(&dev_base_lock
);
2344 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2346 struct net_device_stats
*stats
= dev
->get_stats(dev
);
2348 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2349 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2350 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2352 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2353 stats
->rx_fifo_errors
,
2354 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2355 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2356 stats
->rx_compressed
, stats
->multicast
,
2357 stats
->tx_bytes
, stats
->tx_packets
,
2358 stats
->tx_errors
, stats
->tx_dropped
,
2359 stats
->tx_fifo_errors
, stats
->collisions
,
2360 stats
->tx_carrier_errors
+
2361 stats
->tx_aborted_errors
+
2362 stats
->tx_window_errors
+
2363 stats
->tx_heartbeat_errors
,
2364 stats
->tx_compressed
);
2368 * Called from the PROCfs module. This now uses the new arbitrary sized
2369 * /proc/net interface to create /proc/net/dev
2371 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2373 if (v
== SEQ_START_TOKEN
)
2374 seq_puts(seq
, "Inter-| Receive "
2376 " face |bytes packets errs drop fifo frame "
2377 "compressed multicast|bytes packets errs "
2378 "drop fifo colls carrier compressed\n");
2380 dev_seq_printf_stats(seq
, v
);
2384 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2386 struct netif_rx_stats
*rc
= NULL
;
2388 while (*pos
< NR_CPUS
)
2389 if (cpu_online(*pos
)) {
2390 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2397 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2399 return softnet_get_online(pos
);
2402 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2405 return softnet_get_online(pos
);
2408 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2412 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2414 struct netif_rx_stats
*s
= v
;
2416 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2417 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2418 0, 0, 0, 0, /* was fastroute */
2423 static const struct seq_operations dev_seq_ops
= {
2424 .start
= dev_seq_start
,
2425 .next
= dev_seq_next
,
2426 .stop
= dev_seq_stop
,
2427 .show
= dev_seq_show
,
2430 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2432 struct seq_file
*seq
;
2434 res
= seq_open(file
, &dev_seq_ops
);
2436 seq
= file
->private_data
;
2437 seq
->private = get_net(PROC_NET(inode
));
2442 static int dev_seq_release(struct inode
*inode
, struct file
*file
)
2444 struct seq_file
*seq
= file
->private_data
;
2445 struct net
*net
= seq
->private;
2447 return seq_release(inode
, file
);
2450 static const struct file_operations dev_seq_fops
= {
2451 .owner
= THIS_MODULE
,
2452 .open
= dev_seq_open
,
2454 .llseek
= seq_lseek
,
2455 .release
= dev_seq_release
,
2458 static const struct seq_operations softnet_seq_ops
= {
2459 .start
= softnet_seq_start
,
2460 .next
= softnet_seq_next
,
2461 .stop
= softnet_seq_stop
,
2462 .show
= softnet_seq_show
,
2465 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2467 return seq_open(file
, &softnet_seq_ops
);
2470 static const struct file_operations softnet_seq_fops
= {
2471 .owner
= THIS_MODULE
,
2472 .open
= softnet_seq_open
,
2474 .llseek
= seq_lseek
,
2475 .release
= seq_release
,
2478 static void *ptype_get_idx(loff_t pos
)
2480 struct packet_type
*pt
= NULL
;
2484 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
2490 for (t
= 0; t
< 16; t
++) {
2491 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
2500 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2503 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
2506 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2508 struct packet_type
*pt
;
2509 struct list_head
*nxt
;
2513 if (v
== SEQ_START_TOKEN
)
2514 return ptype_get_idx(0);
2517 nxt
= pt
->list
.next
;
2518 if (pt
->type
== htons(ETH_P_ALL
)) {
2519 if (nxt
!= &ptype_all
)
2522 nxt
= ptype_base
[0].next
;
2524 hash
= ntohs(pt
->type
) & 15;
2526 while (nxt
== &ptype_base
[hash
]) {
2529 nxt
= ptype_base
[hash
].next
;
2532 return list_entry(nxt
, struct packet_type
, list
);
2535 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
2540 static void ptype_seq_decode(struct seq_file
*seq
, void *sym
)
2542 #ifdef CONFIG_KALLSYMS
2543 unsigned long offset
= 0, symsize
;
2544 const char *symname
;
2548 symname
= kallsyms_lookup((unsigned long)sym
, &symsize
, &offset
,
2555 modname
= delim
= "";
2556 seq_printf(seq
, "%s%s%s%s+0x%lx", delim
, modname
, delim
,
2562 seq_printf(seq
, "[%p]", sym
);
2565 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
2567 struct packet_type
*pt
= v
;
2569 if (v
== SEQ_START_TOKEN
)
2570 seq_puts(seq
, "Type Device Function\n");
2572 if (pt
->type
== htons(ETH_P_ALL
))
2573 seq_puts(seq
, "ALL ");
2575 seq_printf(seq
, "%04x", ntohs(pt
->type
));
2577 seq_printf(seq
, " %-8s ",
2578 pt
->dev
? pt
->dev
->name
: "");
2579 ptype_seq_decode(seq
, pt
->func
);
2580 seq_putc(seq
, '\n');
2586 static const struct seq_operations ptype_seq_ops
= {
2587 .start
= ptype_seq_start
,
2588 .next
= ptype_seq_next
,
2589 .stop
= ptype_seq_stop
,
2590 .show
= ptype_seq_show
,
2593 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
2595 return seq_open(file
, &ptype_seq_ops
);
2598 static const struct file_operations ptype_seq_fops
= {
2599 .owner
= THIS_MODULE
,
2600 .open
= ptype_seq_open
,
2602 .llseek
= seq_lseek
,
2603 .release
= seq_release
,
2607 static int dev_proc_net_init(struct net
*net
)
2611 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
2613 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2615 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
2618 if (wext_proc_init(net
))
2624 proc_net_remove(net
, "ptype");
2626 proc_net_remove(net
, "softnet_stat");
2628 proc_net_remove(net
, "dev");
2632 static void dev_proc_net_exit(struct net
*net
)
2634 wext_proc_exit(net
);
2636 proc_net_remove(net
, "ptype");
2637 proc_net_remove(net
, "softnet_stat");
2638 proc_net_remove(net
, "dev");
2641 static struct pernet_operations dev_proc_ops
= {
2642 .init
= dev_proc_net_init
,
2643 .exit
= dev_proc_net_exit
,
2646 static int __init
dev_proc_init(void)
2648 return register_pernet_subsys(&dev_proc_ops
);
2651 #define dev_proc_init() 0
2652 #endif /* CONFIG_PROC_FS */
2656 * netdev_set_master - set up master/slave pair
2657 * @slave: slave device
2658 * @master: new master device
2660 * Changes the master device of the slave. Pass %NULL to break the
2661 * bonding. The caller must hold the RTNL semaphore. On a failure
2662 * a negative errno code is returned. On success the reference counts
2663 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2664 * function returns zero.
2666 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2668 struct net_device
*old
= slave
->master
;
2678 slave
->master
= master
;
2686 slave
->flags
|= IFF_SLAVE
;
2688 slave
->flags
&= ~IFF_SLAVE
;
2690 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2694 static void __dev_set_promiscuity(struct net_device
*dev
, int inc
)
2696 unsigned short old_flags
= dev
->flags
;
2700 if ((dev
->promiscuity
+= inc
) == 0)
2701 dev
->flags
&= ~IFF_PROMISC
;
2703 dev
->flags
|= IFF_PROMISC
;
2704 if (dev
->flags
!= old_flags
) {
2705 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2706 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2708 audit_log(current
->audit_context
, GFP_ATOMIC
,
2709 AUDIT_ANOM_PROMISCUOUS
,
2710 "dev=%s prom=%d old_prom=%d auid=%u",
2711 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2712 (old_flags
& IFF_PROMISC
),
2713 audit_get_loginuid(current
->audit_context
));
2715 if (dev
->change_rx_flags
)
2716 dev
->change_rx_flags(dev
, IFF_PROMISC
);
2721 * dev_set_promiscuity - update promiscuity count on a device
2725 * Add or remove promiscuity from a device. While the count in the device
2726 * remains above zero the interface remains promiscuous. Once it hits zero
2727 * the device reverts back to normal filtering operation. A negative inc
2728 * value is used to drop promiscuity on the device.
2730 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2732 unsigned short old_flags
= dev
->flags
;
2734 __dev_set_promiscuity(dev
, inc
);
2735 if (dev
->flags
!= old_flags
)
2736 dev_set_rx_mode(dev
);
2740 * dev_set_allmulti - update allmulti count on a device
2744 * Add or remove reception of all multicast frames to a device. While the
2745 * count in the device remains above zero the interface remains listening
2746 * to all interfaces. Once it hits zero the device reverts back to normal
2747 * filtering operation. A negative @inc value is used to drop the counter
2748 * when releasing a resource needing all multicasts.
2751 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2753 unsigned short old_flags
= dev
->flags
;
2757 dev
->flags
|= IFF_ALLMULTI
;
2758 if ((dev
->allmulti
+= inc
) == 0)
2759 dev
->flags
&= ~IFF_ALLMULTI
;
2760 if (dev
->flags
^ old_flags
) {
2761 if (dev
->change_rx_flags
)
2762 dev
->change_rx_flags(dev
, IFF_ALLMULTI
);
2763 dev_set_rx_mode(dev
);
2768 * Upload unicast and multicast address lists to device and
2769 * configure RX filtering. When the device doesn't support unicast
2770 * filtering it is put in promiscous mode while unicast addresses
2773 void __dev_set_rx_mode(struct net_device
*dev
)
2775 /* dev_open will call this function so the list will stay sane. */
2776 if (!(dev
->flags
&IFF_UP
))
2779 if (!netif_device_present(dev
))
2782 if (dev
->set_rx_mode
)
2783 dev
->set_rx_mode(dev
);
2785 /* Unicast addresses changes may only happen under the rtnl,
2786 * therefore calling __dev_set_promiscuity here is safe.
2788 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
2789 __dev_set_promiscuity(dev
, 1);
2790 dev
->uc_promisc
= 1;
2791 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
2792 __dev_set_promiscuity(dev
, -1);
2793 dev
->uc_promisc
= 0;
2796 if (dev
->set_multicast_list
)
2797 dev
->set_multicast_list(dev
);
2801 void dev_set_rx_mode(struct net_device
*dev
)
2803 netif_tx_lock_bh(dev
);
2804 __dev_set_rx_mode(dev
);
2805 netif_tx_unlock_bh(dev
);
2808 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
2809 void *addr
, int alen
, int glbl
)
2811 struct dev_addr_list
*da
;
2813 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
2814 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2815 alen
== da
->da_addrlen
) {
2817 int old_glbl
= da
->da_gusers
;
2834 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
2835 void *addr
, int alen
, int glbl
)
2837 struct dev_addr_list
*da
;
2839 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
2840 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2841 da
->da_addrlen
== alen
) {
2843 int old_glbl
= da
->da_gusers
;
2853 da
= kmalloc(sizeof(*da
), GFP_ATOMIC
);
2856 memcpy(da
->da_addr
, addr
, alen
);
2857 da
->da_addrlen
= alen
;
2859 da
->da_gusers
= glbl
? 1 : 0;
2867 * dev_unicast_delete - Release secondary unicast address.
2869 * @addr: address to delete
2870 * @alen: length of @addr
2872 * Release reference to a secondary unicast address and remove it
2873 * from the device if the reference count drops to zero.
2875 * The caller must hold the rtnl_mutex.
2877 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
2883 netif_tx_lock_bh(dev
);
2884 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
2886 __dev_set_rx_mode(dev
);
2887 netif_tx_unlock_bh(dev
);
2890 EXPORT_SYMBOL(dev_unicast_delete
);
2893 * dev_unicast_add - add a secondary unicast address
2895 * @addr: address to delete
2896 * @alen: length of @addr
2898 * Add a secondary unicast address to the device or increase
2899 * the reference count if it already exists.
2901 * The caller must hold the rtnl_mutex.
2903 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
2909 netif_tx_lock_bh(dev
);
2910 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
2912 __dev_set_rx_mode(dev
);
2913 netif_tx_unlock_bh(dev
);
2916 EXPORT_SYMBOL(dev_unicast_add
);
2918 static void __dev_addr_discard(struct dev_addr_list
**list
)
2920 struct dev_addr_list
*tmp
;
2922 while (*list
!= NULL
) {
2925 if (tmp
->da_users
> tmp
->da_gusers
)
2926 printk("__dev_addr_discard: address leakage! "
2927 "da_users=%d\n", tmp
->da_users
);
2932 static void dev_addr_discard(struct net_device
*dev
)
2934 netif_tx_lock_bh(dev
);
2936 __dev_addr_discard(&dev
->uc_list
);
2939 __dev_addr_discard(&dev
->mc_list
);
2942 netif_tx_unlock_bh(dev
);
2945 unsigned dev_get_flags(const struct net_device
*dev
)
2949 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2954 (dev
->gflags
& (IFF_PROMISC
|
2957 if (netif_running(dev
)) {
2958 if (netif_oper_up(dev
))
2959 flags
|= IFF_RUNNING
;
2960 if (netif_carrier_ok(dev
))
2961 flags
|= IFF_LOWER_UP
;
2962 if (netif_dormant(dev
))
2963 flags
|= IFF_DORMANT
;
2969 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2972 int old_flags
= dev
->flags
;
2977 * Set the flags on our device.
2980 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2981 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2983 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2987 * Load in the correct multicast list now the flags have changed.
2990 if (dev
->change_rx_flags
&& (dev
->flags
^ flags
) & IFF_MULTICAST
)
2991 dev
->change_rx_flags(dev
, IFF_MULTICAST
);
2993 dev_set_rx_mode(dev
);
2996 * Have we downed the interface. We handle IFF_UP ourselves
2997 * according to user attempts to set it, rather than blindly
3002 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
3003 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
3006 dev_set_rx_mode(dev
);
3009 if (dev
->flags
& IFF_UP
&&
3010 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
3012 raw_notifier_call_chain(&netdev_chain
,
3013 NETDEV_CHANGE
, dev
);
3015 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
3016 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
3017 dev
->gflags
^= IFF_PROMISC
;
3018 dev_set_promiscuity(dev
, inc
);
3021 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3022 is important. Some (broken) drivers set IFF_PROMISC, when
3023 IFF_ALLMULTI is requested not asking us and not reporting.
3025 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
3026 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
3027 dev
->gflags
^= IFF_ALLMULTI
;
3028 dev_set_allmulti(dev
, inc
);
3031 /* Exclude state transition flags, already notified */
3032 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
3034 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
3039 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
3043 if (new_mtu
== dev
->mtu
)
3046 /* MTU must be positive. */
3050 if (!netif_device_present(dev
))
3054 if (dev
->change_mtu
)
3055 err
= dev
->change_mtu(dev
, new_mtu
);
3058 if (!err
&& dev
->flags
& IFF_UP
)
3059 raw_notifier_call_chain(&netdev_chain
,
3060 NETDEV_CHANGEMTU
, dev
);
3064 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
3068 if (!dev
->set_mac_address
)
3070 if (sa
->sa_family
!= dev
->type
)
3072 if (!netif_device_present(dev
))
3074 err
= dev
->set_mac_address(dev
, sa
);
3076 raw_notifier_call_chain(&netdev_chain
,
3077 NETDEV_CHANGEADDR
, dev
);
3082 * Perform the SIOCxIFxxx calls.
3084 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3087 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3093 case SIOCGIFFLAGS
: /* Get interface flags */
3094 ifr
->ifr_flags
= dev_get_flags(dev
);
3097 case SIOCSIFFLAGS
: /* Set interface flags */
3098 return dev_change_flags(dev
, ifr
->ifr_flags
);
3100 case SIOCGIFMETRIC
: /* Get the metric on the interface
3101 (currently unused) */
3102 ifr
->ifr_metric
= 0;
3105 case SIOCSIFMETRIC
: /* Set the metric on the interface
3106 (currently unused) */
3109 case SIOCGIFMTU
: /* Get the MTU of a device */
3110 ifr
->ifr_mtu
= dev
->mtu
;
3113 case SIOCSIFMTU
: /* Set the MTU of a device */
3114 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
3118 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
3120 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
3121 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3122 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
3126 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
3128 case SIOCSIFHWBROADCAST
:
3129 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
3131 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
3132 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3133 raw_notifier_call_chain(&netdev_chain
,
3134 NETDEV_CHANGEADDR
, dev
);
3138 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
3139 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
3140 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
3141 ifr
->ifr_map
.irq
= dev
->irq
;
3142 ifr
->ifr_map
.dma
= dev
->dma
;
3143 ifr
->ifr_map
.port
= dev
->if_port
;
3147 if (dev
->set_config
) {
3148 if (!netif_device_present(dev
))
3150 return dev
->set_config(dev
, &ifr
->ifr_map
);
3155 if (!dev
->set_multicast_list
||
3156 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3158 if (!netif_device_present(dev
))
3160 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
3164 if (!dev
->set_multicast_list
||
3165 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3167 if (!netif_device_present(dev
))
3169 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
3173 ifr
->ifr_ifindex
= dev
->ifindex
;
3177 ifr
->ifr_qlen
= dev
->tx_queue_len
;
3181 if (ifr
->ifr_qlen
< 0)
3183 dev
->tx_queue_len
= ifr
->ifr_qlen
;
3187 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
3188 return dev_change_name(dev
, ifr
->ifr_newname
);
3191 * Unknown or private ioctl
3195 if ((cmd
>= SIOCDEVPRIVATE
&&
3196 cmd
<= SIOCDEVPRIVATE
+ 15) ||
3197 cmd
== SIOCBONDENSLAVE
||
3198 cmd
== SIOCBONDRELEASE
||
3199 cmd
== SIOCBONDSETHWADDR
||
3200 cmd
== SIOCBONDSLAVEINFOQUERY
||
3201 cmd
== SIOCBONDINFOQUERY
||
3202 cmd
== SIOCBONDCHANGEACTIVE
||
3203 cmd
== SIOCGMIIPHY
||
3204 cmd
== SIOCGMIIREG
||
3205 cmd
== SIOCSMIIREG
||
3206 cmd
== SIOCBRADDIF
||
3207 cmd
== SIOCBRDELIF
||
3208 cmd
== SIOCWANDEV
) {
3210 if (dev
->do_ioctl
) {
3211 if (netif_device_present(dev
))
3212 err
= dev
->do_ioctl(dev
, ifr
,
3225 * This function handles all "interface"-type I/O control requests. The actual
3226 * 'doing' part of this is dev_ifsioc above.
3230 * dev_ioctl - network device ioctl
3231 * @cmd: command to issue
3232 * @arg: pointer to a struct ifreq in user space
3234 * Issue ioctl functions to devices. This is normally called by the
3235 * user space syscall interfaces but can sometimes be useful for
3236 * other purposes. The return value is the return from the syscall if
3237 * positive or a negative errno code on error.
3240 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
3246 /* One special case: SIOCGIFCONF takes ifconf argument
3247 and requires shared lock, because it sleeps writing
3251 if (cmd
== SIOCGIFCONF
) {
3253 ret
= dev_ifconf(net
, (char __user
*) arg
);
3257 if (cmd
== SIOCGIFNAME
)
3258 return dev_ifname(net
, (struct ifreq __user
*)arg
);
3260 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3263 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
3265 colon
= strchr(ifr
.ifr_name
, ':');
3270 * See which interface the caller is talking about.
3275 * These ioctl calls:
3276 * - can be done by all.
3277 * - atomic and do not require locking.
3288 dev_load(net
, ifr
.ifr_name
);
3289 read_lock(&dev_base_lock
);
3290 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3291 read_unlock(&dev_base_lock
);
3295 if (copy_to_user(arg
, &ifr
,
3296 sizeof(struct ifreq
)))
3302 dev_load(net
, ifr
.ifr_name
);
3304 ret
= dev_ethtool(net
, &ifr
);
3309 if (copy_to_user(arg
, &ifr
,
3310 sizeof(struct ifreq
)))
3316 * These ioctl calls:
3317 * - require superuser power.
3318 * - require strict serialization.
3324 if (!capable(CAP_NET_ADMIN
))
3326 dev_load(net
, ifr
.ifr_name
);
3328 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3333 if (copy_to_user(arg
, &ifr
,
3334 sizeof(struct ifreq
)))
3340 * These ioctl calls:
3341 * - require superuser power.
3342 * - require strict serialization.
3343 * - do not return a value
3353 case SIOCSIFHWBROADCAST
:
3356 case SIOCBONDENSLAVE
:
3357 case SIOCBONDRELEASE
:
3358 case SIOCBONDSETHWADDR
:
3359 case SIOCBONDCHANGEACTIVE
:
3362 if (!capable(CAP_NET_ADMIN
))
3365 case SIOCBONDSLAVEINFOQUERY
:
3366 case SIOCBONDINFOQUERY
:
3367 dev_load(net
, ifr
.ifr_name
);
3369 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3374 /* Get the per device memory space. We can add this but
3375 * currently do not support it */
3377 /* Set the per device memory buffer space.
3378 * Not applicable in our case */
3383 * Unknown or private ioctl.
3386 if (cmd
== SIOCWANDEV
||
3387 (cmd
>= SIOCDEVPRIVATE
&&
3388 cmd
<= SIOCDEVPRIVATE
+ 15)) {
3389 dev_load(net
, ifr
.ifr_name
);
3391 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3393 if (!ret
&& copy_to_user(arg
, &ifr
,
3394 sizeof(struct ifreq
)))
3398 /* Take care of Wireless Extensions */
3399 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
3400 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
3407 * dev_new_index - allocate an ifindex
3409 * Returns a suitable unique value for a new device interface
3410 * number. The caller must hold the rtnl semaphore or the
3411 * dev_base_lock to be sure it remains unique.
3413 static int dev_new_index(struct net
*net
)
3419 if (!__dev_get_by_index(net
, ifindex
))
3424 /* Delayed registration/unregisteration */
3425 static DEFINE_SPINLOCK(net_todo_list_lock
);
3426 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
3428 static void net_set_todo(struct net_device
*dev
)
3430 spin_lock(&net_todo_list_lock
);
3431 list_add_tail(&dev
->todo_list
, &net_todo_list
);
3432 spin_unlock(&net_todo_list_lock
);
3436 * register_netdevice - register a network device
3437 * @dev: device to register
3439 * Take a completed network device structure and add it to the kernel
3440 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3441 * chain. 0 is returned on success. A negative errno code is returned
3442 * on a failure to set up the device, or if the name is a duplicate.
3444 * Callers must hold the rtnl semaphore. You may want
3445 * register_netdev() instead of this.
3448 * The locking appears insufficient to guarantee two parallel registers
3449 * will not get the same name.
3452 int register_netdevice(struct net_device
*dev
)
3454 struct hlist_head
*head
;
3455 struct hlist_node
*p
;
3459 BUG_ON(dev_boot_phase
);
3464 /* When net_device's are persistent, this will be fatal. */
3465 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
3466 BUG_ON(!dev
->nd_net
);
3469 spin_lock_init(&dev
->queue_lock
);
3470 spin_lock_init(&dev
->_xmit_lock
);
3471 netdev_set_lockdep_class(&dev
->_xmit_lock
, dev
->type
);
3472 dev
->xmit_lock_owner
= -1;
3473 spin_lock_init(&dev
->ingress_lock
);
3477 /* Init, if this function is available */
3479 ret
= dev
->init(dev
);
3487 if (!dev_valid_name(dev
->name
)) {
3492 dev
->ifindex
= dev_new_index(net
);
3493 if (dev
->iflink
== -1)
3494 dev
->iflink
= dev
->ifindex
;
3496 /* Check for existence of name */
3497 head
= dev_name_hash(net
, dev
->name
);
3498 hlist_for_each(p
, head
) {
3499 struct net_device
*d
3500 = hlist_entry(p
, struct net_device
, name_hlist
);
3501 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
3507 /* Fix illegal checksum combinations */
3508 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
3509 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3510 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
3512 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
3515 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
3516 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3517 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
3519 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
3523 /* Fix illegal SG+CSUM combinations. */
3524 if ((dev
->features
& NETIF_F_SG
) &&
3525 !(dev
->features
& NETIF_F_ALL_CSUM
)) {
3526 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no checksum feature.\n",
3528 dev
->features
&= ~NETIF_F_SG
;
3531 /* TSO requires that SG is present as well. */
3532 if ((dev
->features
& NETIF_F_TSO
) &&
3533 !(dev
->features
& NETIF_F_SG
)) {
3534 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no SG feature.\n",
3536 dev
->features
&= ~NETIF_F_TSO
;
3538 if (dev
->features
& NETIF_F_UFO
) {
3539 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
3540 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3541 "NETIF_F_HW_CSUM feature.\n",
3543 dev
->features
&= ~NETIF_F_UFO
;
3545 if (!(dev
->features
& NETIF_F_SG
)) {
3546 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3547 "NETIF_F_SG feature.\n",
3549 dev
->features
&= ~NETIF_F_UFO
;
3554 * nil rebuild_header routine,
3555 * that should be never called and used as just bug trap.
3558 if (!dev
->rebuild_header
)
3559 dev
->rebuild_header
= default_rebuild_header
;
3561 ret
= netdev_register_sysfs(dev
);
3564 dev
->reg_state
= NETREG_REGISTERED
;
3567 * Default initial state at registry is that the
3568 * device is present.
3571 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
3573 dev_init_scheduler(dev
);
3574 write_lock_bh(&dev_base_lock
);
3575 list_add_tail(&dev
->dev_list
, &net
->dev_base_head
);
3576 hlist_add_head(&dev
->name_hlist
, head
);
3577 hlist_add_head(&dev
->index_hlist
, dev_index_hash(net
, dev
->ifindex
));
3579 write_unlock_bh(&dev_base_lock
);
3581 /* Notify protocols, that a new device appeared. */
3582 ret
= raw_notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
3583 ret
= notifier_to_errno(ret
);
3585 unregister_netdevice(dev
);
3597 * register_netdev - register a network device
3598 * @dev: device to register
3600 * Take a completed network device structure and add it to the kernel
3601 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3602 * chain. 0 is returned on success. A negative errno code is returned
3603 * on a failure to set up the device, or if the name is a duplicate.
3605 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3606 * and expands the device name if you passed a format string to
3609 int register_netdev(struct net_device
*dev
)
3616 * If the name is a format string the caller wants us to do a
3619 if (strchr(dev
->name
, '%')) {
3620 err
= dev_alloc_name(dev
, dev
->name
);
3625 err
= register_netdevice(dev
);
3630 EXPORT_SYMBOL(register_netdev
);
3633 * netdev_wait_allrefs - wait until all references are gone.
3635 * This is called when unregistering network devices.
3637 * Any protocol or device that holds a reference should register
3638 * for netdevice notification, and cleanup and put back the
3639 * reference if they receive an UNREGISTER event.
3640 * We can get stuck here if buggy protocols don't correctly
3643 static void netdev_wait_allrefs(struct net_device
*dev
)
3645 unsigned long rebroadcast_time
, warning_time
;
3647 rebroadcast_time
= warning_time
= jiffies
;
3648 while (atomic_read(&dev
->refcnt
) != 0) {
3649 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
3652 /* Rebroadcast unregister notification */
3653 raw_notifier_call_chain(&netdev_chain
,
3654 NETDEV_UNREGISTER
, dev
);
3656 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
3658 /* We must not have linkwatch events
3659 * pending on unregister. If this
3660 * happens, we simply run the queue
3661 * unscheduled, resulting in a noop
3664 linkwatch_run_queue();
3669 rebroadcast_time
= jiffies
;
3674 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
3675 printk(KERN_EMERG
"unregister_netdevice: "
3676 "waiting for %s to become free. Usage "
3678 dev
->name
, atomic_read(&dev
->refcnt
));
3679 warning_time
= jiffies
;
3688 * register_netdevice(x1);
3689 * register_netdevice(x2);
3691 * unregister_netdevice(y1);
3692 * unregister_netdevice(y2);
3698 * We are invoked by rtnl_unlock() after it drops the semaphore.
3699 * This allows us to deal with problems:
3700 * 1) We can delete sysfs objects which invoke hotplug
3701 * without deadlocking with linkwatch via keventd.
3702 * 2) Since we run with the RTNL semaphore not held, we can sleep
3703 * safely in order to wait for the netdev refcnt to drop to zero.
3705 static DEFINE_MUTEX(net_todo_run_mutex
);
3706 void netdev_run_todo(void)
3708 struct list_head list
;
3710 /* Need to guard against multiple cpu's getting out of order. */
3711 mutex_lock(&net_todo_run_mutex
);
3713 /* Not safe to do outside the semaphore. We must not return
3714 * until all unregister events invoked by the local processor
3715 * have been completed (either by this todo run, or one on
3718 if (list_empty(&net_todo_list
))
3721 /* Snapshot list, allow later requests */
3722 spin_lock(&net_todo_list_lock
);
3723 list_replace_init(&net_todo_list
, &list
);
3724 spin_unlock(&net_todo_list_lock
);
3726 while (!list_empty(&list
)) {
3727 struct net_device
*dev
3728 = list_entry(list
.next
, struct net_device
, todo_list
);
3729 list_del(&dev
->todo_list
);
3731 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
3732 printk(KERN_ERR
"network todo '%s' but state %d\n",
3733 dev
->name
, dev
->reg_state
);
3738 dev
->reg_state
= NETREG_UNREGISTERED
;
3740 netdev_wait_allrefs(dev
);
3743 BUG_ON(atomic_read(&dev
->refcnt
));
3744 BUG_TRAP(!dev
->ip_ptr
);
3745 BUG_TRAP(!dev
->ip6_ptr
);
3746 BUG_TRAP(!dev
->dn_ptr
);
3748 if (dev
->destructor
)
3749 dev
->destructor(dev
);
3751 /* Free network device */
3752 kobject_put(&dev
->dev
.kobj
);
3756 mutex_unlock(&net_todo_run_mutex
);
3759 static struct net_device_stats
*internal_stats(struct net_device
*dev
)
3765 * alloc_netdev_mq - allocate network device
3766 * @sizeof_priv: size of private data to allocate space for
3767 * @name: device name format string
3768 * @setup: callback to initialize device
3769 * @queue_count: the number of subqueues to allocate
3771 * Allocates a struct net_device with private data area for driver use
3772 * and performs basic initialization. Also allocates subquue structs
3773 * for each queue on the device at the end of the netdevice.
3775 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
3776 void (*setup
)(struct net_device
*), unsigned int queue_count
)
3779 struct net_device
*dev
;
3782 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
3784 /* ensure 32-byte alignment of both the device and private area */
3785 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
+
3786 (sizeof(struct net_device_subqueue
) * (queue_count
- 1))) &
3787 ~NETDEV_ALIGN_CONST
;
3788 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
3790 p
= kzalloc(alloc_size
, GFP_KERNEL
);
3792 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
3796 dev
= (struct net_device
*)
3797 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3798 dev
->padded
= (char *)dev
- (char *)p
;
3799 dev
->nd_net
= &init_net
;
3802 dev
->priv
= ((char *)dev
+
3803 ((sizeof(struct net_device
) +
3804 (sizeof(struct net_device_subqueue
) *
3805 (queue_count
- 1)) + NETDEV_ALIGN_CONST
)
3806 & ~NETDEV_ALIGN_CONST
));
3809 dev
->egress_subqueue_count
= queue_count
;
3811 dev
->get_stats
= internal_stats
;
3812 netpoll_netdev_init(dev
);
3814 strcpy(dev
->name
, name
);
3817 EXPORT_SYMBOL(alloc_netdev_mq
);
3820 * free_netdev - free network device
3823 * This function does the last stage of destroying an allocated device
3824 * interface. The reference to the device object is released.
3825 * If this is the last reference then it will be freed.
3827 void free_netdev(struct net_device
*dev
)
3830 /* Compatibility with error handling in drivers */
3831 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3832 kfree((char *)dev
- dev
->padded
);
3836 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3837 dev
->reg_state
= NETREG_RELEASED
;
3839 /* will free via device release */
3840 put_device(&dev
->dev
);
3842 kfree((char *)dev
- dev
->padded
);
3846 /* Synchronize with packet receive processing. */
3847 void synchronize_net(void)
3854 * unregister_netdevice - remove device from the kernel
3857 * This function shuts down a device interface and removes it
3858 * from the kernel tables. On success 0 is returned, on a failure
3859 * a negative errno code is returned.
3861 * Callers must hold the rtnl semaphore. You may want
3862 * unregister_netdev() instead of this.
3865 void unregister_netdevice(struct net_device
*dev
)
3867 BUG_ON(dev_boot_phase
);
3870 /* Some devices call without registering for initialization unwind. */
3871 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3872 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3873 "was registered\n", dev
->name
, dev
);
3879 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3881 /* If device is running, close it first. */
3882 if (dev
->flags
& IFF_UP
)
3885 /* And unlink it from device chain. */
3886 write_lock_bh(&dev_base_lock
);
3887 list_del(&dev
->dev_list
);
3888 hlist_del(&dev
->name_hlist
);
3889 hlist_del(&dev
->index_hlist
);
3890 write_unlock_bh(&dev_base_lock
);
3892 dev
->reg_state
= NETREG_UNREGISTERING
;
3896 /* Shutdown queueing discipline. */
3900 /* Notify protocols, that we are about to destroy
3901 this device. They should clean all the things.
3903 raw_notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3906 * Flush the unicast and multicast chains
3908 dev_addr_discard(dev
);
3913 /* Notifier chain MUST detach us from master device. */
3914 BUG_TRAP(!dev
->master
);
3916 /* Remove entries from sysfs */
3917 netdev_unregister_sysfs(dev
);
3919 /* Finish processing unregister after unlock */
3928 * unregister_netdev - remove device from the kernel
3931 * This function shuts down a device interface and removes it
3932 * from the kernel tables. On success 0 is returned, on a failure
3933 * a negative errno code is returned.
3935 * This is just a wrapper for unregister_netdevice that takes
3936 * the rtnl semaphore. In general you want to use this and not
3937 * unregister_netdevice.
3939 void unregister_netdev(struct net_device
*dev
)
3942 unregister_netdevice(dev
);
3946 EXPORT_SYMBOL(unregister_netdev
);
3948 static int dev_cpu_callback(struct notifier_block
*nfb
,
3949 unsigned long action
,
3952 struct sk_buff
**list_skb
;
3953 struct net_device
**list_net
;
3954 struct sk_buff
*skb
;
3955 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3956 struct softnet_data
*sd
, *oldsd
;
3958 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
3961 local_irq_disable();
3962 cpu
= smp_processor_id();
3963 sd
= &per_cpu(softnet_data
, cpu
);
3964 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3966 /* Find end of our completion_queue. */
3967 list_skb
= &sd
->completion_queue
;
3969 list_skb
= &(*list_skb
)->next
;
3970 /* Append completion queue from offline CPU. */
3971 *list_skb
= oldsd
->completion_queue
;
3972 oldsd
->completion_queue
= NULL
;
3974 /* Find end of our output_queue. */
3975 list_net
= &sd
->output_queue
;
3977 list_net
= &(*list_net
)->next_sched
;
3978 /* Append output queue from offline CPU. */
3979 *list_net
= oldsd
->output_queue
;
3980 oldsd
->output_queue
= NULL
;
3982 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3985 /* Process offline CPU's input_pkt_queue */
3986 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3992 #ifdef CONFIG_NET_DMA
3994 * net_dma_rebalance - try to maintain one DMA channel per CPU
3995 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
3997 * This is called when the number of channels allocated to the net_dma client
3998 * changes. The net_dma client tries to have one DMA channel per CPU.
4001 static void net_dma_rebalance(struct net_dma
*net_dma
)
4003 unsigned int cpu
, i
, n
, chan_idx
;
4004 struct dma_chan
*chan
;
4006 if (cpus_empty(net_dma
->channel_mask
)) {
4007 for_each_online_cpu(cpu
)
4008 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
4013 cpu
= first_cpu(cpu_online_map
);
4015 for_each_cpu_mask(chan_idx
, net_dma
->channel_mask
) {
4016 chan
= net_dma
->channels
[chan_idx
];
4018 n
= ((num_online_cpus() / cpus_weight(net_dma
->channel_mask
))
4019 + (i
< (num_online_cpus() %
4020 cpus_weight(net_dma
->channel_mask
)) ? 1 : 0));
4023 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
4024 cpu
= next_cpu(cpu
, cpu_online_map
);
4032 * netdev_dma_event - event callback for the net_dma_client
4033 * @client: should always be net_dma_client
4034 * @chan: DMA channel for the event
4035 * @state: DMA state to be handled
4037 static enum dma_state_client
4038 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
4039 enum dma_state state
)
4041 int i
, found
= 0, pos
= -1;
4042 struct net_dma
*net_dma
=
4043 container_of(client
, struct net_dma
, client
);
4044 enum dma_state_client ack
= DMA_DUP
; /* default: take no action */
4046 spin_lock(&net_dma
->lock
);
4048 case DMA_RESOURCE_AVAILABLE
:
4049 for (i
= 0; i
< NR_CPUS
; i
++)
4050 if (net_dma
->channels
[i
] == chan
) {
4053 } else if (net_dma
->channels
[i
] == NULL
&& pos
< 0)
4056 if (!found
&& pos
>= 0) {
4058 net_dma
->channels
[pos
] = chan
;
4059 cpu_set(pos
, net_dma
->channel_mask
);
4060 net_dma_rebalance(net_dma
);
4063 case DMA_RESOURCE_REMOVED
:
4064 for (i
= 0; i
< NR_CPUS
; i
++)
4065 if (net_dma
->channels
[i
] == chan
) {
4073 cpu_clear(pos
, net_dma
->channel_mask
);
4074 net_dma
->channels
[i
] = NULL
;
4075 net_dma_rebalance(net_dma
);
4081 spin_unlock(&net_dma
->lock
);
4087 * netdev_dma_regiser - register the networking subsystem as a DMA client
4089 static int __init
netdev_dma_register(void)
4091 spin_lock_init(&net_dma
.lock
);
4092 dma_cap_set(DMA_MEMCPY
, net_dma
.client
.cap_mask
);
4093 dma_async_client_register(&net_dma
.client
);
4094 dma_async_client_chan_request(&net_dma
.client
);
4099 static int __init
netdev_dma_register(void) { return -ENODEV
; }
4100 #endif /* CONFIG_NET_DMA */
4103 * netdev_compute_feature - compute conjunction of two feature sets
4104 * @all: first feature set
4105 * @one: second feature set
4107 * Computes a new feature set after adding a device with feature set
4108 * @one to the master device with current feature set @all. Returns
4109 * the new feature set.
4111 int netdev_compute_features(unsigned long all
, unsigned long one
)
4113 /* if device needs checksumming, downgrade to hw checksumming */
4114 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
4115 all
^= NETIF_F_NO_CSUM
| NETIF_F_HW_CSUM
;
4117 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4118 if (all
& NETIF_F_HW_CSUM
&& !(one
& NETIF_F_HW_CSUM
))
4119 all
^= NETIF_F_HW_CSUM
4120 | NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
;
4122 if (one
& NETIF_F_GSO
)
4123 one
|= NETIF_F_GSO_SOFTWARE
;
4126 /* If even one device supports robust GSO, enable it for all. */
4127 if (one
& NETIF_F_GSO_ROBUST
)
4128 all
|= NETIF_F_GSO_ROBUST
;
4130 all
&= one
| NETIF_F_LLTX
;
4132 if (!(all
& NETIF_F_ALL_CSUM
))
4134 if (!(all
& NETIF_F_SG
))
4135 all
&= ~NETIF_F_GSO_MASK
;
4139 EXPORT_SYMBOL(netdev_compute_features
);
4141 /* Initialize per network namespace state */
4142 static int netdev_init(struct net
*net
)
4145 INIT_LIST_HEAD(&net
->dev_base_head
);
4146 rwlock_init(&dev_base_lock
);
4148 net
->dev_name_head
= kmalloc(
4149 sizeof(*net
->dev_name_head
)*NETDEV_HASHENTRIES
, GFP_KERNEL
);
4150 if (!net
->dev_name_head
)
4153 net
->dev_index_head
= kmalloc(
4154 sizeof(*net
->dev_index_head
)*NETDEV_HASHENTRIES
, GFP_KERNEL
);
4155 if (!net
->dev_index_head
) {
4156 kfree(net
->dev_name_head
);
4160 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
4161 INIT_HLIST_HEAD(&net
->dev_name_head
[i
]);
4163 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
4164 INIT_HLIST_HEAD(&net
->dev_index_head
[i
]);
4169 static void netdev_exit(struct net
*net
)
4171 kfree(net
->dev_name_head
);
4172 kfree(net
->dev_index_head
);
4175 static struct pernet_operations netdev_net_ops
= {
4176 .init
= netdev_init
,
4177 .exit
= netdev_exit
,
4181 * Initialize the DEV module. At boot time this walks the device list and
4182 * unhooks any devices that fail to initialise (normally hardware not
4183 * present) and leaves us with a valid list of present and active devices.
4188 * This is called single threaded during boot, so no need
4189 * to take the rtnl semaphore.
4191 static int __init
net_dev_init(void)
4193 int i
, rc
= -ENOMEM
;
4195 BUG_ON(!dev_boot_phase
);
4197 if (dev_proc_init())
4200 if (netdev_sysfs_init())
4203 INIT_LIST_HEAD(&ptype_all
);
4204 for (i
= 0; i
< 16; i
++)
4205 INIT_LIST_HEAD(&ptype_base
[i
]);
4207 if (register_pernet_subsys(&netdev_net_ops
))
4211 * Initialise the packet receive queues.
4214 for_each_possible_cpu(i
) {
4215 struct softnet_data
*queue
;
4217 queue
= &per_cpu(softnet_data
, i
);
4218 skb_queue_head_init(&queue
->input_pkt_queue
);
4219 queue
->completion_queue
= NULL
;
4220 INIT_LIST_HEAD(&queue
->poll_list
);
4222 queue
->backlog
.poll
= process_backlog
;
4223 queue
->backlog
.weight
= weight_p
;
4226 netdev_dma_register();
4230 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
4231 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
4233 hotcpu_notifier(dev_cpu_callback
, 0);
4241 subsys_initcall(net_dev_init
);
4243 EXPORT_SYMBOL(__dev_get_by_index
);
4244 EXPORT_SYMBOL(__dev_get_by_name
);
4245 EXPORT_SYMBOL(__dev_remove_pack
);
4246 EXPORT_SYMBOL(dev_valid_name
);
4247 EXPORT_SYMBOL(dev_add_pack
);
4248 EXPORT_SYMBOL(dev_alloc_name
);
4249 EXPORT_SYMBOL(dev_close
);
4250 EXPORT_SYMBOL(dev_get_by_flags
);
4251 EXPORT_SYMBOL(dev_get_by_index
);
4252 EXPORT_SYMBOL(dev_get_by_name
);
4253 EXPORT_SYMBOL(dev_open
);
4254 EXPORT_SYMBOL(dev_queue_xmit
);
4255 EXPORT_SYMBOL(dev_remove_pack
);
4256 EXPORT_SYMBOL(dev_set_allmulti
);
4257 EXPORT_SYMBOL(dev_set_promiscuity
);
4258 EXPORT_SYMBOL(dev_change_flags
);
4259 EXPORT_SYMBOL(dev_set_mtu
);
4260 EXPORT_SYMBOL(dev_set_mac_address
);
4261 EXPORT_SYMBOL(free_netdev
);
4262 EXPORT_SYMBOL(netdev_boot_setup_check
);
4263 EXPORT_SYMBOL(netdev_set_master
);
4264 EXPORT_SYMBOL(netdev_state_change
);
4265 EXPORT_SYMBOL(netif_receive_skb
);
4266 EXPORT_SYMBOL(netif_rx
);
4267 EXPORT_SYMBOL(register_gifconf
);
4268 EXPORT_SYMBOL(register_netdevice
);
4269 EXPORT_SYMBOL(register_netdevice_notifier
);
4270 EXPORT_SYMBOL(skb_checksum_help
);
4271 EXPORT_SYMBOL(synchronize_net
);
4272 EXPORT_SYMBOL(unregister_netdevice
);
4273 EXPORT_SYMBOL(unregister_netdevice_notifier
);
4274 EXPORT_SYMBOL(net_enable_timestamp
);
4275 EXPORT_SYMBOL(net_disable_timestamp
);
4276 EXPORT_SYMBOL(dev_get_flags
);
4278 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4279 EXPORT_SYMBOL(br_handle_frame_hook
);
4280 EXPORT_SYMBOL(br_fdb_get_hook
);
4281 EXPORT_SYMBOL(br_fdb_put_hook
);
4285 EXPORT_SYMBOL(dev_load
);
4288 EXPORT_PER_CPU_SYMBOL(softnet_data
);