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/ethtool.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/net_namespace.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/stat.h>
102 #include <linux/if_bridge.h>
103 #include <linux/if_macvlan.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.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>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
130 #include "net-sysfs.h"
132 /* Instead of increasing this, you should create a hash table. */
133 #define MAX_GRO_SKBS 8
135 /* This should be increased if a protocol with a bigger head is added. */
136 #define GRO_MAX_HEAD (MAX_HEADER + 128)
147 * The list of packet types we will receive (as opposed to discard)
148 * and the routines to invoke.
150 * Why 16. Because with 16 the only overlap we get on a hash of the
151 * low nibble of the protocol value is RARP/SNAP/X.25.
153 * NOTE: That is no longer true with the addition of VLAN tags. Not
154 * sure which should go first, but I bet it won't make much
155 * difference if we are running VLANs. The good news is that
156 * this protocol won't be in the list unless compiled in, so
157 * the average user (w/out VLANs) will not be adversely affected.
174 #define PTYPE_HASH_SIZE (16)
175 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
177 static DEFINE_SPINLOCK(ptype_lock
);
178 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
179 static struct list_head ptype_all __read_mostly
; /* Taps */
182 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
185 * Pure readers hold dev_base_lock for reading.
187 * Writers must hold the rtnl semaphore while they loop through the
188 * dev_base_head list, and hold dev_base_lock for writing when they do the
189 * actual updates. This allows pure readers to access the list even
190 * while a writer is preparing to update it.
192 * To put it another way, dev_base_lock is held for writing only to
193 * protect against pure readers; the rtnl semaphore provides the
194 * protection against other writers.
196 * See, for example usages, register_netdevice() and
197 * unregister_netdevice(), which must be called with the rtnl
200 DEFINE_RWLOCK(dev_base_lock
);
202 EXPORT_SYMBOL(dev_base_lock
);
204 #define NETDEV_HASHBITS 8
205 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
207 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
209 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
210 return &net
->dev_name_head
[hash
& ((1 << NETDEV_HASHBITS
) - 1)];
213 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
215 return &net
->dev_index_head
[ifindex
& ((1 << NETDEV_HASHBITS
) - 1)];
218 static inline void *skb_gro_mac_header(struct sk_buff
*skb
)
220 return skb_mac_header(skb
) < skb
->data
? skb_mac_header(skb
) :
221 page_address(skb_shinfo(skb
)->frags
[0].page
) +
222 skb_shinfo(skb
)->frags
[0].page_offset
;
225 /* Device list insertion */
226 static int list_netdevice(struct net_device
*dev
)
228 struct net
*net
= dev_net(dev
);
232 write_lock_bh(&dev_base_lock
);
233 list_add_tail(&dev
->dev_list
, &net
->dev_base_head
);
234 hlist_add_head(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
235 hlist_add_head(&dev
->index_hlist
, dev_index_hash(net
, dev
->ifindex
));
236 write_unlock_bh(&dev_base_lock
);
240 /* Device list removal */
241 static void unlist_netdevice(struct net_device
*dev
)
245 /* Unlink dev from the device chain */
246 write_lock_bh(&dev_base_lock
);
247 list_del(&dev
->dev_list
);
248 hlist_del(&dev
->name_hlist
);
249 hlist_del(&dev
->index_hlist
);
250 write_unlock_bh(&dev_base_lock
);
257 static RAW_NOTIFIER_HEAD(netdev_chain
);
260 * Device drivers call our routines to queue packets here. We empty the
261 * queue in the local softnet handler.
264 DEFINE_PER_CPU(struct softnet_data
, softnet_data
);
266 #ifdef CONFIG_LOCKDEP
268 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
269 * according to dev->type
271 static const unsigned short netdev_lock_type
[] =
272 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
273 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
274 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
275 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
276 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
277 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
278 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
279 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
280 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
281 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
282 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
283 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
284 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
285 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
286 ARPHRD_PHONET_PIPE
, ARPHRD_VOID
, ARPHRD_NONE
};
288 static const char *netdev_lock_name
[] =
289 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
290 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
291 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
292 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
293 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
294 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
295 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
296 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
297 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
298 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
299 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
300 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
301 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
302 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
303 "_xmit_PHONET_PIPE", "_xmit_VOID", "_xmit_NONE"};
305 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
306 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
308 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
312 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
313 if (netdev_lock_type
[i
] == dev_type
)
315 /* the last key is used by default */
316 return ARRAY_SIZE(netdev_lock_type
) - 1;
319 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
320 unsigned short dev_type
)
324 i
= netdev_lock_pos(dev_type
);
325 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
326 netdev_lock_name
[i
]);
329 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
333 i
= netdev_lock_pos(dev
->type
);
334 lockdep_set_class_and_name(&dev
->addr_list_lock
,
335 &netdev_addr_lock_key
[i
],
336 netdev_lock_name
[i
]);
339 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
340 unsigned short dev_type
)
343 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
348 /*******************************************************************************
350 Protocol management and registration routines
352 *******************************************************************************/
355 * Add a protocol ID to the list. Now that the input handler is
356 * smarter we can dispense with all the messy stuff that used to be
359 * BEWARE!!! Protocol handlers, mangling input packets,
360 * MUST BE last in hash buckets and checking protocol handlers
361 * MUST start from promiscuous ptype_all chain in net_bh.
362 * It is true now, do not change it.
363 * Explanation follows: if protocol handler, mangling packet, will
364 * be the first on list, it is not able to sense, that packet
365 * is cloned and should be copied-on-write, so that it will
366 * change it and subsequent readers will get broken packet.
371 * dev_add_pack - add packet handler
372 * @pt: packet type declaration
374 * Add a protocol handler to the networking stack. The passed &packet_type
375 * is linked into kernel lists and may not be freed until it has been
376 * removed from the kernel lists.
378 * This call does not sleep therefore it can not
379 * guarantee all CPU's that are in middle of receiving packets
380 * will see the new packet type (until the next received packet).
383 void dev_add_pack(struct packet_type
*pt
)
387 spin_lock_bh(&ptype_lock
);
388 if (pt
->type
== htons(ETH_P_ALL
))
389 list_add_rcu(&pt
->list
, &ptype_all
);
391 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
392 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
394 spin_unlock_bh(&ptype_lock
);
398 * __dev_remove_pack - remove packet handler
399 * @pt: packet type declaration
401 * Remove a protocol handler that was previously added to the kernel
402 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
403 * from the kernel lists and can be freed or reused once this function
406 * The packet type might still be in use by receivers
407 * and must not be freed until after all the CPU's have gone
408 * through a quiescent state.
410 void __dev_remove_pack(struct packet_type
*pt
)
412 struct list_head
*head
;
413 struct packet_type
*pt1
;
415 spin_lock_bh(&ptype_lock
);
417 if (pt
->type
== htons(ETH_P_ALL
))
420 head
= &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
422 list_for_each_entry(pt1
, head
, list
) {
424 list_del_rcu(&pt
->list
);
429 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
431 spin_unlock_bh(&ptype_lock
);
434 * dev_remove_pack - remove packet handler
435 * @pt: packet type declaration
437 * Remove a protocol handler that was previously added to the kernel
438 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
439 * from the kernel lists and can be freed or reused once this function
442 * This call sleeps to guarantee that no CPU is looking at the packet
445 void dev_remove_pack(struct packet_type
*pt
)
447 __dev_remove_pack(pt
);
452 /******************************************************************************
454 Device Boot-time Settings Routines
456 *******************************************************************************/
458 /* Boot time configuration table */
459 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
462 * netdev_boot_setup_add - add new setup entry
463 * @name: name of the device
464 * @map: configured settings for the device
466 * Adds new setup entry to the dev_boot_setup list. The function
467 * returns 0 on error and 1 on success. This is a generic routine to
470 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
472 struct netdev_boot_setup
*s
;
476 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
477 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
478 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
479 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
480 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
485 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
489 * netdev_boot_setup_check - check boot time settings
490 * @dev: the netdevice
492 * Check boot time settings for the device.
493 * The found settings are set for the device to be used
494 * later in the device probing.
495 * Returns 0 if no settings found, 1 if they are.
497 int netdev_boot_setup_check(struct net_device
*dev
)
499 struct netdev_boot_setup
*s
= dev_boot_setup
;
502 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
503 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
504 !strcmp(dev
->name
, s
[i
].name
)) {
505 dev
->irq
= s
[i
].map
.irq
;
506 dev
->base_addr
= s
[i
].map
.base_addr
;
507 dev
->mem_start
= s
[i
].map
.mem_start
;
508 dev
->mem_end
= s
[i
].map
.mem_end
;
517 * netdev_boot_base - get address from boot time settings
518 * @prefix: prefix for network device
519 * @unit: id for network device
521 * Check boot time settings for the base address of device.
522 * The found settings are set for the device to be used
523 * later in the device probing.
524 * Returns 0 if no settings found.
526 unsigned long netdev_boot_base(const char *prefix
, int unit
)
528 const struct netdev_boot_setup
*s
= dev_boot_setup
;
532 sprintf(name
, "%s%d", prefix
, unit
);
535 * If device already registered then return base of 1
536 * to indicate not to probe for this interface
538 if (__dev_get_by_name(&init_net
, name
))
541 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
542 if (!strcmp(name
, s
[i
].name
))
543 return s
[i
].map
.base_addr
;
548 * Saves at boot time configured settings for any netdevice.
550 int __init
netdev_boot_setup(char *str
)
555 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
560 memset(&map
, 0, sizeof(map
));
564 map
.base_addr
= ints
[2];
566 map
.mem_start
= ints
[3];
568 map
.mem_end
= ints
[4];
570 /* Add new entry to the list */
571 return netdev_boot_setup_add(str
, &map
);
574 __setup("netdev=", netdev_boot_setup
);
576 /*******************************************************************************
578 Device Interface Subroutines
580 *******************************************************************************/
583 * __dev_get_by_name - find a device by its name
584 * @net: the applicable net namespace
585 * @name: name to find
587 * Find an interface by name. Must be called under RTNL semaphore
588 * or @dev_base_lock. If the name is found a pointer to the device
589 * is returned. If the name is not found then %NULL is returned. The
590 * reference counters are not incremented so the caller must be
591 * careful with locks.
594 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
596 struct hlist_node
*p
;
598 hlist_for_each(p
, dev_name_hash(net
, name
)) {
599 struct net_device
*dev
600 = hlist_entry(p
, struct net_device
, name_hlist
);
601 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
608 * dev_get_by_name - find a device by its name
609 * @net: the applicable net namespace
610 * @name: name to find
612 * Find an interface by name. This can be called from any
613 * context and does its own locking. The returned handle has
614 * the usage count incremented and the caller must use dev_put() to
615 * release it when it is no longer needed. %NULL is returned if no
616 * matching device is found.
619 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
621 struct net_device
*dev
;
623 read_lock(&dev_base_lock
);
624 dev
= __dev_get_by_name(net
, name
);
627 read_unlock(&dev_base_lock
);
632 * __dev_get_by_index - find a device by its ifindex
633 * @net: the applicable net namespace
634 * @ifindex: index of device
636 * Search for an interface by index. Returns %NULL if the device
637 * is not found or a pointer to the device. The device has not
638 * had its reference counter increased so the caller must be careful
639 * about locking. The caller must hold either the RTNL semaphore
643 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
645 struct hlist_node
*p
;
647 hlist_for_each(p
, dev_index_hash(net
, ifindex
)) {
648 struct net_device
*dev
649 = hlist_entry(p
, struct net_device
, index_hlist
);
650 if (dev
->ifindex
== ifindex
)
658 * dev_get_by_index - find a device by its ifindex
659 * @net: the applicable net namespace
660 * @ifindex: index of device
662 * Search for an interface by index. Returns NULL if the device
663 * is not found or a pointer to the device. The device returned has
664 * had a reference added and the pointer is safe until the user calls
665 * dev_put to indicate they have finished with it.
668 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
670 struct net_device
*dev
;
672 read_lock(&dev_base_lock
);
673 dev
= __dev_get_by_index(net
, ifindex
);
676 read_unlock(&dev_base_lock
);
681 * dev_getbyhwaddr - find a device by its hardware address
682 * @net: the applicable net namespace
683 * @type: media type of device
684 * @ha: hardware address
686 * Search for an interface by MAC address. Returns NULL if the device
687 * is not found or a pointer to the device. The caller must hold the
688 * rtnl semaphore. The returned device has not had its ref count increased
689 * and the caller must therefore be careful about locking
692 * If the API was consistent this would be __dev_get_by_hwaddr
695 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
697 struct net_device
*dev
;
701 for_each_netdev(net
, dev
)
702 if (dev
->type
== type
&&
703 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
709 EXPORT_SYMBOL(dev_getbyhwaddr
);
711 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
713 struct net_device
*dev
;
716 for_each_netdev(net
, dev
)
717 if (dev
->type
== type
)
723 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
725 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
727 struct net_device
*dev
;
730 dev
= __dev_getfirstbyhwtype(net
, type
);
737 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
740 * dev_get_by_flags - find any device with given flags
741 * @net: the applicable net namespace
742 * @if_flags: IFF_* values
743 * @mask: bitmask of bits in if_flags to check
745 * Search for any interface with the given flags. Returns NULL if a device
746 * is not found or a pointer to the device. The device returned has
747 * had a reference added and the pointer is safe until the user calls
748 * dev_put to indicate they have finished with it.
751 struct net_device
* dev_get_by_flags(struct net
*net
, unsigned short if_flags
, unsigned short mask
)
753 struct net_device
*dev
, *ret
;
756 read_lock(&dev_base_lock
);
757 for_each_netdev(net
, dev
) {
758 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
764 read_unlock(&dev_base_lock
);
769 * dev_valid_name - check if name is okay for network device
772 * Network device names need to be valid file names to
773 * to allow sysfs to work. We also disallow any kind of
776 int dev_valid_name(const char *name
)
780 if (strlen(name
) >= IFNAMSIZ
)
782 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
786 if (*name
== '/' || isspace(*name
))
794 * __dev_alloc_name - allocate a name for a device
795 * @net: network namespace to allocate the device name in
796 * @name: name format string
797 * @buf: scratch buffer and result name string
799 * Passed a format string - eg "lt%d" it will try and find a suitable
800 * id. It scans list of devices to build up a free map, then chooses
801 * the first empty slot. The caller must hold the dev_base or rtnl lock
802 * while allocating the name and adding the device in order to avoid
804 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
805 * Returns the number of the unit assigned or a negative errno code.
808 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
812 const int max_netdevices
= 8*PAGE_SIZE
;
813 unsigned long *inuse
;
814 struct net_device
*d
;
816 p
= strnchr(name
, IFNAMSIZ
-1, '%');
819 * Verify the string as this thing may have come from
820 * the user. There must be either one "%d" and no other "%"
823 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
826 /* Use one page as a bit array of possible slots */
827 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
831 for_each_netdev(net
, d
) {
832 if (!sscanf(d
->name
, name
, &i
))
834 if (i
< 0 || i
>= max_netdevices
)
837 /* avoid cases where sscanf is not exact inverse of printf */
838 snprintf(buf
, IFNAMSIZ
, name
, i
);
839 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
843 i
= find_first_zero_bit(inuse
, max_netdevices
);
844 free_page((unsigned long) inuse
);
847 snprintf(buf
, IFNAMSIZ
, name
, i
);
848 if (!__dev_get_by_name(net
, buf
))
851 /* It is possible to run out of possible slots
852 * when the name is long and there isn't enough space left
853 * for the digits, or if all bits are used.
859 * dev_alloc_name - allocate a name for a device
861 * @name: name format string
863 * Passed a format string - eg "lt%d" it will try and find a suitable
864 * id. It scans list of devices to build up a free map, then chooses
865 * the first empty slot. The caller must hold the dev_base or rtnl lock
866 * while allocating the name and adding the device in order to avoid
868 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
869 * Returns the number of the unit assigned or a negative errno code.
872 int dev_alloc_name(struct net_device
*dev
, const char *name
)
878 BUG_ON(!dev_net(dev
));
880 ret
= __dev_alloc_name(net
, name
, buf
);
882 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
888 * dev_change_name - change name of a device
890 * @newname: name (or format string) must be at least IFNAMSIZ
892 * Change name of a device, can pass format strings "eth%d".
895 int dev_change_name(struct net_device
*dev
, const char *newname
)
897 char oldname
[IFNAMSIZ
];
903 BUG_ON(!dev_net(dev
));
906 if (dev
->flags
& IFF_UP
)
909 if (!dev_valid_name(newname
))
912 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
915 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
917 if (strchr(newname
, '%')) {
918 err
= dev_alloc_name(dev
, newname
);
922 else if (__dev_get_by_name(net
, newname
))
925 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
928 /* For now only devices in the initial network namespace
931 if (net
== &init_net
) {
932 ret
= device_rename(&dev
->dev
, dev
->name
);
934 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
939 write_lock_bh(&dev_base_lock
);
940 hlist_del(&dev
->name_hlist
);
941 hlist_add_head(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
942 write_unlock_bh(&dev_base_lock
);
944 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
945 ret
= notifier_to_errno(ret
);
950 "%s: name change rollback failed: %d.\n",
954 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
963 * dev_set_alias - change ifalias of a device
965 * @alias: name up to IFALIASZ
966 * @len: limit of bytes to copy from info
968 * Set ifalias for a device,
970 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
985 dev
->ifalias
= krealloc(dev
->ifalias
, len
+1, GFP_KERNEL
);
989 strlcpy(dev
->ifalias
, alias
, len
+1);
995 * netdev_features_change - device changes features
996 * @dev: device to cause notification
998 * Called to indicate a device has changed features.
1000 void netdev_features_change(struct net_device
*dev
)
1002 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1004 EXPORT_SYMBOL(netdev_features_change
);
1007 * netdev_state_change - device changes state
1008 * @dev: device to cause notification
1010 * Called to indicate a device has changed state. This function calls
1011 * the notifier chains for netdev_chain and sends a NEWLINK message
1012 * to the routing socket.
1014 void netdev_state_change(struct net_device
*dev
)
1016 if (dev
->flags
& IFF_UP
) {
1017 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1018 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1022 void netdev_bonding_change(struct net_device
*dev
)
1024 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, dev
);
1026 EXPORT_SYMBOL(netdev_bonding_change
);
1029 * dev_load - load a network module
1030 * @net: the applicable net namespace
1031 * @name: name of interface
1033 * If a network interface is not present and the process has suitable
1034 * privileges this function loads the module. If module loading is not
1035 * available in this kernel then it becomes a nop.
1038 void dev_load(struct net
*net
, const char *name
)
1040 struct net_device
*dev
;
1042 read_lock(&dev_base_lock
);
1043 dev
= __dev_get_by_name(net
, name
);
1044 read_unlock(&dev_base_lock
);
1046 if (!dev
&& capable(CAP_SYS_MODULE
))
1047 request_module("%s", name
);
1051 * dev_open - prepare an interface for use.
1052 * @dev: device to open
1054 * Takes a device from down to up state. The device's private open
1055 * function is invoked and then the multicast lists are loaded. Finally
1056 * the device is moved into the up state and a %NETDEV_UP message is
1057 * sent to the netdev notifier chain.
1059 * Calling this function on an active interface is a nop. On a failure
1060 * a negative errno code is returned.
1062 int dev_open(struct net_device
*dev
)
1064 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1073 if (dev
->flags
& IFF_UP
)
1077 * Is it even present?
1079 if (!netif_device_present(dev
))
1083 * Call device private open method
1085 set_bit(__LINK_STATE_START
, &dev
->state
);
1087 if (ops
->ndo_validate_addr
)
1088 ret
= ops
->ndo_validate_addr(dev
);
1090 if (!ret
&& ops
->ndo_open
)
1091 ret
= ops
->ndo_open(dev
);
1094 * If it went open OK then:
1098 clear_bit(__LINK_STATE_START
, &dev
->state
);
1103 dev
->flags
|= IFF_UP
;
1111 * Initialize multicasting status
1113 dev_set_rx_mode(dev
);
1116 * Wakeup transmit queue engine
1121 * ... and announce new interface.
1123 call_netdevice_notifiers(NETDEV_UP
, dev
);
1130 * dev_close - shutdown an interface.
1131 * @dev: device to shutdown
1133 * This function moves an active device into down state. A
1134 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1135 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1138 int dev_close(struct net_device
*dev
)
1140 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1145 if (!(dev
->flags
& IFF_UP
))
1149 * Tell people we are going down, so that they can
1150 * prepare to death, when device is still operating.
1152 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1154 clear_bit(__LINK_STATE_START
, &dev
->state
);
1156 /* Synchronize to scheduled poll. We cannot touch poll list,
1157 * it can be even on different cpu. So just clear netif_running().
1159 * dev->stop() will invoke napi_disable() on all of it's
1160 * napi_struct instances on this device.
1162 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1164 dev_deactivate(dev
);
1167 * Call the device specific close. This cannot fail.
1168 * Only if device is UP
1170 * We allow it to be called even after a DETACH hot-plug
1177 * Device is now down.
1180 dev
->flags
&= ~IFF_UP
;
1183 * Tell people we are down
1185 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1197 * dev_disable_lro - disable Large Receive Offload on a device
1200 * Disable Large Receive Offload (LRO) on a net device. Must be
1201 * called under RTNL. This is needed if received packets may be
1202 * forwarded to another interface.
1204 void dev_disable_lro(struct net_device
*dev
)
1206 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1207 dev
->ethtool_ops
->set_flags
) {
1208 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1209 if (flags
& ETH_FLAG_LRO
) {
1210 flags
&= ~ETH_FLAG_LRO
;
1211 dev
->ethtool_ops
->set_flags(dev
, flags
);
1214 WARN_ON(dev
->features
& NETIF_F_LRO
);
1216 EXPORT_SYMBOL(dev_disable_lro
);
1219 static int dev_boot_phase
= 1;
1222 * Device change register/unregister. These are not inline or static
1223 * as we export them to the world.
1227 * register_netdevice_notifier - register a network notifier block
1230 * Register a notifier to be called when network device events occur.
1231 * The notifier passed is linked into the kernel structures and must
1232 * not be reused until it has been unregistered. A negative errno code
1233 * is returned on a failure.
1235 * When registered all registration and up events are replayed
1236 * to the new notifier to allow device to have a race free
1237 * view of the network device list.
1240 int register_netdevice_notifier(struct notifier_block
*nb
)
1242 struct net_device
*dev
;
1243 struct net_device
*last
;
1248 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1254 for_each_netdev(net
, dev
) {
1255 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1256 err
= notifier_to_errno(err
);
1260 if (!(dev
->flags
& IFF_UP
))
1263 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1274 for_each_netdev(net
, dev
) {
1278 if (dev
->flags
& IFF_UP
) {
1279 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1280 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1282 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1286 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1291 * unregister_netdevice_notifier - unregister a network notifier block
1294 * Unregister a notifier previously registered by
1295 * register_netdevice_notifier(). The notifier is unlinked into the
1296 * kernel structures and may then be reused. A negative errno code
1297 * is returned on a failure.
1300 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1305 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1311 * call_netdevice_notifiers - call all network notifier blocks
1312 * @val: value passed unmodified to notifier function
1313 * @dev: net_device pointer passed unmodified to notifier function
1315 * Call all network notifier blocks. Parameters and return value
1316 * are as for raw_notifier_call_chain().
1319 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1321 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1324 /* When > 0 there are consumers of rx skb time stamps */
1325 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1327 void net_enable_timestamp(void)
1329 atomic_inc(&netstamp_needed
);
1332 void net_disable_timestamp(void)
1334 atomic_dec(&netstamp_needed
);
1337 static inline void net_timestamp(struct sk_buff
*skb
)
1339 if (atomic_read(&netstamp_needed
))
1340 __net_timestamp(skb
);
1342 skb
->tstamp
.tv64
= 0;
1346 * Support routine. Sends outgoing frames to any network
1347 * taps currently in use.
1350 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1352 struct packet_type
*ptype
;
1357 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1358 /* Never send packets back to the socket
1359 * they originated from - MvS (miquels@drinkel.ow.org)
1361 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1362 (ptype
->af_packet_priv
== NULL
||
1363 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1364 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1368 /* skb->nh should be correctly
1369 set by sender, so that the second statement is
1370 just protection against buggy protocols.
1372 skb_reset_mac_header(skb2
);
1374 if (skb_network_header(skb2
) < skb2
->data
||
1375 skb2
->network_header
> skb2
->tail
) {
1376 if (net_ratelimit())
1377 printk(KERN_CRIT
"protocol %04x is "
1379 skb2
->protocol
, dev
->name
);
1380 skb_reset_network_header(skb2
);
1383 skb2
->transport_header
= skb2
->network_header
;
1384 skb2
->pkt_type
= PACKET_OUTGOING
;
1385 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1392 static inline void __netif_reschedule(struct Qdisc
*q
)
1394 struct softnet_data
*sd
;
1395 unsigned long flags
;
1397 local_irq_save(flags
);
1398 sd
= &__get_cpu_var(softnet_data
);
1399 q
->next_sched
= sd
->output_queue
;
1400 sd
->output_queue
= q
;
1401 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1402 local_irq_restore(flags
);
1405 void __netif_schedule(struct Qdisc
*q
)
1407 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1408 __netif_reschedule(q
);
1410 EXPORT_SYMBOL(__netif_schedule
);
1412 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1414 if (atomic_dec_and_test(&skb
->users
)) {
1415 struct softnet_data
*sd
;
1416 unsigned long flags
;
1418 local_irq_save(flags
);
1419 sd
= &__get_cpu_var(softnet_data
);
1420 skb
->next
= sd
->completion_queue
;
1421 sd
->completion_queue
= skb
;
1422 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1423 local_irq_restore(flags
);
1426 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1428 void dev_kfree_skb_any(struct sk_buff
*skb
)
1430 if (in_irq() || irqs_disabled())
1431 dev_kfree_skb_irq(skb
);
1435 EXPORT_SYMBOL(dev_kfree_skb_any
);
1439 * netif_device_detach - mark device as removed
1440 * @dev: network device
1442 * Mark device as removed from system and therefore no longer available.
1444 void netif_device_detach(struct net_device
*dev
)
1446 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1447 netif_running(dev
)) {
1448 netif_stop_queue(dev
);
1451 EXPORT_SYMBOL(netif_device_detach
);
1454 * netif_device_attach - mark device as attached
1455 * @dev: network device
1457 * Mark device as attached from system and restart if needed.
1459 void netif_device_attach(struct net_device
*dev
)
1461 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1462 netif_running(dev
)) {
1463 netif_wake_queue(dev
);
1464 __netdev_watchdog_up(dev
);
1467 EXPORT_SYMBOL(netif_device_attach
);
1469 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1471 return ((features
& NETIF_F_GEN_CSUM
) ||
1472 ((features
& NETIF_F_IP_CSUM
) &&
1473 protocol
== htons(ETH_P_IP
)) ||
1474 ((features
& NETIF_F_IPV6_CSUM
) &&
1475 protocol
== htons(ETH_P_IPV6
)));
1478 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1480 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1483 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1484 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1485 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1486 veh
->h_vlan_encapsulated_proto
))
1494 * Invalidate hardware checksum when packet is to be mangled, and
1495 * complete checksum manually on outgoing path.
1497 int skb_checksum_help(struct sk_buff
*skb
)
1500 int ret
= 0, offset
;
1502 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1503 goto out_set_summed
;
1505 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1506 /* Let GSO fix up the checksum. */
1507 goto out_set_summed
;
1510 offset
= skb
->csum_start
- skb_headroom(skb
);
1511 BUG_ON(offset
>= skb_headlen(skb
));
1512 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1514 offset
+= skb
->csum_offset
;
1515 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1517 if (skb_cloned(skb
) &&
1518 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1519 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1524 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1526 skb
->ip_summed
= CHECKSUM_NONE
;
1532 * skb_gso_segment - Perform segmentation on skb.
1533 * @skb: buffer to segment
1534 * @features: features for the output path (see dev->features)
1536 * This function segments the given skb and returns a list of segments.
1538 * It may return NULL if the skb requires no segmentation. This is
1539 * only possible when GSO is used for verifying header integrity.
1541 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1543 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1544 struct packet_type
*ptype
;
1545 __be16 type
= skb
->protocol
;
1548 skb_reset_mac_header(skb
);
1549 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1550 __skb_pull(skb
, skb
->mac_len
);
1552 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1553 struct net_device
*dev
= skb
->dev
;
1554 struct ethtool_drvinfo info
= {};
1556 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1557 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1559 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1561 info
.driver
, dev
? dev
->features
: 0L,
1562 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1563 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1565 if (skb_header_cloned(skb
) &&
1566 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1567 return ERR_PTR(err
);
1571 list_for_each_entry_rcu(ptype
,
1572 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1573 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1574 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1575 err
= ptype
->gso_send_check(skb
);
1576 segs
= ERR_PTR(err
);
1577 if (err
|| skb_gso_ok(skb
, features
))
1579 __skb_push(skb
, (skb
->data
-
1580 skb_network_header(skb
)));
1582 segs
= ptype
->gso_segment(skb
, features
);
1588 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1593 EXPORT_SYMBOL(skb_gso_segment
);
1595 /* Take action when hardware reception checksum errors are detected. */
1597 void netdev_rx_csum_fault(struct net_device
*dev
)
1599 if (net_ratelimit()) {
1600 printk(KERN_ERR
"%s: hw csum failure.\n",
1601 dev
? dev
->name
: "<unknown>");
1605 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1608 /* Actually, we should eliminate this check as soon as we know, that:
1609 * 1. IOMMU is present and allows to map all the memory.
1610 * 2. No high memory really exists on this machine.
1613 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1615 #ifdef CONFIG_HIGHMEM
1618 if (dev
->features
& NETIF_F_HIGHDMA
)
1621 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1622 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1630 void (*destructor
)(struct sk_buff
*skb
);
1633 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1635 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1637 struct dev_gso_cb
*cb
;
1640 struct sk_buff
*nskb
= skb
->next
;
1642 skb
->next
= nskb
->next
;
1645 } while (skb
->next
);
1647 cb
= DEV_GSO_CB(skb
);
1649 cb
->destructor(skb
);
1653 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1654 * @skb: buffer to segment
1656 * This function segments the given skb and stores the list of segments
1659 static int dev_gso_segment(struct sk_buff
*skb
)
1661 struct net_device
*dev
= skb
->dev
;
1662 struct sk_buff
*segs
;
1663 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1666 segs
= skb_gso_segment(skb
, features
);
1668 /* Verifying header integrity only. */
1673 return PTR_ERR(segs
);
1676 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1677 skb
->destructor
= dev_gso_skb_destructor
;
1682 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1683 struct netdev_queue
*txq
)
1685 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1687 prefetch(&dev
->netdev_ops
->ndo_start_xmit
);
1688 if (likely(!skb
->next
)) {
1689 if (!list_empty(&ptype_all
))
1690 dev_queue_xmit_nit(skb
, dev
);
1692 if (netif_needs_gso(dev
, skb
)) {
1693 if (unlikely(dev_gso_segment(skb
)))
1699 return ops
->ndo_start_xmit(skb
, dev
);
1704 struct sk_buff
*nskb
= skb
->next
;
1707 skb
->next
= nskb
->next
;
1709 rc
= ops
->ndo_start_xmit(nskb
, dev
);
1711 nskb
->next
= skb
->next
;
1715 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
1716 return NETDEV_TX_BUSY
;
1717 } while (skb
->next
);
1719 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1726 static u32 skb_tx_hashrnd
;
1727 static int skb_tx_hashrnd_initialized
= 0;
1729 static u16
skb_tx_hash(struct net_device
*dev
, struct sk_buff
*skb
)
1733 if (unlikely(!skb_tx_hashrnd_initialized
)) {
1734 get_random_bytes(&skb_tx_hashrnd
, 4);
1735 skb_tx_hashrnd_initialized
= 1;
1738 if (skb_rx_queue_recorded(skb
)) {
1739 hash
= skb_get_rx_queue(skb
);
1740 } else if (skb
->sk
&& skb
->sk
->sk_hash
) {
1741 hash
= skb
->sk
->sk_hash
;
1743 hash
= skb
->protocol
;
1745 hash
= jhash_1word(hash
, skb_tx_hashrnd
);
1747 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
1750 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
1751 struct sk_buff
*skb
)
1753 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1754 u16 queue_index
= 0;
1756 if (ops
->ndo_select_queue
)
1757 queue_index
= ops
->ndo_select_queue(dev
, skb
);
1758 else if (dev
->real_num_tx_queues
> 1)
1759 queue_index
= skb_tx_hash(dev
, skb
);
1761 skb_set_queue_mapping(skb
, queue_index
);
1762 return netdev_get_tx_queue(dev
, queue_index
);
1766 * dev_queue_xmit - transmit a buffer
1767 * @skb: buffer to transmit
1769 * Queue a buffer for transmission to a network device. The caller must
1770 * have set the device and priority and built the buffer before calling
1771 * this function. The function can be called from an interrupt.
1773 * A negative errno code is returned on a failure. A success does not
1774 * guarantee the frame will be transmitted as it may be dropped due
1775 * to congestion or traffic shaping.
1777 * -----------------------------------------------------------------------------------
1778 * I notice this method can also return errors from the queue disciplines,
1779 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1782 * Regardless of the return value, the skb is consumed, so it is currently
1783 * difficult to retry a send to this method. (You can bump the ref count
1784 * before sending to hold a reference for retry if you are careful.)
1786 * When calling this method, interrupts MUST be enabled. This is because
1787 * the BH enable code must have IRQs enabled so that it will not deadlock.
1790 int dev_queue_xmit(struct sk_buff
*skb
)
1792 struct net_device
*dev
= skb
->dev
;
1793 struct netdev_queue
*txq
;
1797 /* GSO will handle the following emulations directly. */
1798 if (netif_needs_gso(dev
, skb
))
1801 if (skb_shinfo(skb
)->frag_list
&&
1802 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1803 __skb_linearize(skb
))
1806 /* Fragmented skb is linearized if device does not support SG,
1807 * or if at least one of fragments is in highmem and device
1808 * does not support DMA from it.
1810 if (skb_shinfo(skb
)->nr_frags
&&
1811 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1812 __skb_linearize(skb
))
1815 /* If packet is not checksummed and device does not support
1816 * checksumming for this protocol, complete checksumming here.
1818 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1819 skb_set_transport_header(skb
, skb
->csum_start
-
1821 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
1826 /* Disable soft irqs for various locks below. Also
1827 * stops preemption for RCU.
1831 txq
= dev_pick_tx(dev
, skb
);
1832 q
= rcu_dereference(txq
->qdisc
);
1834 #ifdef CONFIG_NET_CLS_ACT
1835 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1838 spinlock_t
*root_lock
= qdisc_lock(q
);
1840 spin_lock(root_lock
);
1842 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
1846 rc
= qdisc_enqueue_root(skb
, q
);
1849 spin_unlock(root_lock
);
1854 /* The device has no queue. Common case for software devices:
1855 loopback, all the sorts of tunnels...
1857 Really, it is unlikely that netif_tx_lock protection is necessary
1858 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1860 However, it is possible, that they rely on protection
1863 Check this and shot the lock. It is not prone from deadlocks.
1864 Either shot noqueue qdisc, it is even simpler 8)
1866 if (dev
->flags
& IFF_UP
) {
1867 int cpu
= smp_processor_id(); /* ok because BHs are off */
1869 if (txq
->xmit_lock_owner
!= cpu
) {
1871 HARD_TX_LOCK(dev
, txq
, cpu
);
1873 if (!netif_tx_queue_stopped(txq
)) {
1875 if (!dev_hard_start_xmit(skb
, dev
, txq
)) {
1876 HARD_TX_UNLOCK(dev
, txq
);
1880 HARD_TX_UNLOCK(dev
, txq
);
1881 if (net_ratelimit())
1882 printk(KERN_CRIT
"Virtual device %s asks to "
1883 "queue packet!\n", dev
->name
);
1885 /* Recursion is detected! It is possible,
1887 if (net_ratelimit())
1888 printk(KERN_CRIT
"Dead loop on virtual device "
1889 "%s, fix it urgently!\n", dev
->name
);
1894 rcu_read_unlock_bh();
1900 rcu_read_unlock_bh();
1905 /*=======================================================================
1907 =======================================================================*/
1909 int netdev_max_backlog __read_mostly
= 1000;
1910 int netdev_budget __read_mostly
= 300;
1911 int weight_p __read_mostly
= 64; /* old backlog weight */
1913 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1917 * netif_rx - post buffer to the network code
1918 * @skb: buffer to post
1920 * This function receives a packet from a device driver and queues it for
1921 * the upper (protocol) levels to process. It always succeeds. The buffer
1922 * may be dropped during processing for congestion control or by the
1926 * NET_RX_SUCCESS (no congestion)
1927 * NET_RX_DROP (packet was dropped)
1931 int netif_rx(struct sk_buff
*skb
)
1933 struct softnet_data
*queue
;
1934 unsigned long flags
;
1936 /* if netpoll wants it, pretend we never saw it */
1937 if (netpoll_rx(skb
))
1940 if (!skb
->tstamp
.tv64
)
1944 * The code is rearranged so that the path is the most
1945 * short when CPU is congested, but is still operating.
1947 local_irq_save(flags
);
1948 queue
= &__get_cpu_var(softnet_data
);
1950 __get_cpu_var(netdev_rx_stat
).total
++;
1951 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1952 if (queue
->input_pkt_queue
.qlen
) {
1954 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1955 local_irq_restore(flags
);
1956 return NET_RX_SUCCESS
;
1959 napi_schedule(&queue
->backlog
);
1963 __get_cpu_var(netdev_rx_stat
).dropped
++;
1964 local_irq_restore(flags
);
1970 int netif_rx_ni(struct sk_buff
*skb
)
1975 err
= netif_rx(skb
);
1976 if (local_softirq_pending())
1983 EXPORT_SYMBOL(netif_rx_ni
);
1985 static void net_tx_action(struct softirq_action
*h
)
1987 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1989 if (sd
->completion_queue
) {
1990 struct sk_buff
*clist
;
1992 local_irq_disable();
1993 clist
= sd
->completion_queue
;
1994 sd
->completion_queue
= NULL
;
1998 struct sk_buff
*skb
= clist
;
1999 clist
= clist
->next
;
2001 WARN_ON(atomic_read(&skb
->users
));
2006 if (sd
->output_queue
) {
2009 local_irq_disable();
2010 head
= sd
->output_queue
;
2011 sd
->output_queue
= NULL
;
2015 struct Qdisc
*q
= head
;
2016 spinlock_t
*root_lock
;
2018 head
= head
->next_sched
;
2020 root_lock
= qdisc_lock(q
);
2021 if (spin_trylock(root_lock
)) {
2022 smp_mb__before_clear_bit();
2023 clear_bit(__QDISC_STATE_SCHED
,
2026 spin_unlock(root_lock
);
2028 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2030 __netif_reschedule(q
);
2032 smp_mb__before_clear_bit();
2033 clear_bit(__QDISC_STATE_SCHED
,
2041 static inline int deliver_skb(struct sk_buff
*skb
,
2042 struct packet_type
*pt_prev
,
2043 struct net_device
*orig_dev
)
2045 atomic_inc(&skb
->users
);
2046 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2049 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2050 /* These hooks defined here for ATM */
2052 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
2053 unsigned char *addr
);
2054 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
2057 * If bridge module is loaded call bridging hook.
2058 * returns NULL if packet was consumed.
2060 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
2061 struct sk_buff
*skb
) __read_mostly
;
2062 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
2063 struct packet_type
**pt_prev
, int *ret
,
2064 struct net_device
*orig_dev
)
2066 struct net_bridge_port
*port
;
2068 if (skb
->pkt_type
== PACKET_LOOPBACK
||
2069 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
2073 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2077 return br_handle_frame_hook(port
, skb
);
2080 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2083 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2084 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
2085 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
2087 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
2088 struct packet_type
**pt_prev
,
2090 struct net_device
*orig_dev
)
2092 if (skb
->dev
->macvlan_port
== NULL
)
2096 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2099 return macvlan_handle_frame_hook(skb
);
2102 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2105 #ifdef CONFIG_NET_CLS_ACT
2106 /* TODO: Maybe we should just force sch_ingress to be compiled in
2107 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2108 * a compare and 2 stores extra right now if we dont have it on
2109 * but have CONFIG_NET_CLS_ACT
2110 * NOTE: This doesnt stop any functionality; if you dont have
2111 * the ingress scheduler, you just cant add policies on ingress.
2114 static int ing_filter(struct sk_buff
*skb
)
2116 struct net_device
*dev
= skb
->dev
;
2117 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2118 struct netdev_queue
*rxq
;
2119 int result
= TC_ACT_OK
;
2122 if (MAX_RED_LOOP
< ttl
++) {
2124 "Redir loop detected Dropping packet (%d->%d)\n",
2125 skb
->iif
, dev
->ifindex
);
2129 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2130 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2132 rxq
= &dev
->rx_queue
;
2135 if (q
!= &noop_qdisc
) {
2136 spin_lock(qdisc_lock(q
));
2137 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2138 result
= qdisc_enqueue_root(skb
, q
);
2139 spin_unlock(qdisc_lock(q
));
2145 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2146 struct packet_type
**pt_prev
,
2147 int *ret
, struct net_device
*orig_dev
)
2149 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2153 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2156 /* Huh? Why does turning on AF_PACKET affect this? */
2157 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2160 switch (ing_filter(skb
)) {
2174 * netif_nit_deliver - deliver received packets to network taps
2177 * This function is used to deliver incoming packets to network
2178 * taps. It should be used when the normal netif_receive_skb path
2179 * is bypassed, for example because of VLAN acceleration.
2181 void netif_nit_deliver(struct sk_buff
*skb
)
2183 struct packet_type
*ptype
;
2185 if (list_empty(&ptype_all
))
2188 skb_reset_network_header(skb
);
2189 skb_reset_transport_header(skb
);
2190 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2193 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2194 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2195 deliver_skb(skb
, ptype
, skb
->dev
);
2201 * netif_receive_skb - process receive buffer from network
2202 * @skb: buffer to process
2204 * netif_receive_skb() is the main receive data processing function.
2205 * It always succeeds. The buffer may be dropped during processing
2206 * for congestion control or by the protocol layers.
2208 * This function may only be called from softirq context and interrupts
2209 * should be enabled.
2211 * Return values (usually ignored):
2212 * NET_RX_SUCCESS: no congestion
2213 * NET_RX_DROP: packet was dropped
2215 int netif_receive_skb(struct sk_buff
*skb
)
2217 struct packet_type
*ptype
, *pt_prev
;
2218 struct net_device
*orig_dev
;
2219 struct net_device
*null_or_orig
;
2220 int ret
= NET_RX_DROP
;
2223 if (skb
->vlan_tci
&& vlan_hwaccel_do_receive(skb
))
2224 return NET_RX_SUCCESS
;
2226 /* if we've gotten here through NAPI, check netpoll */
2227 if (netpoll_receive_skb(skb
))
2230 if (!skb
->tstamp
.tv64
)
2234 skb
->iif
= skb
->dev
->ifindex
;
2236 null_or_orig
= NULL
;
2237 orig_dev
= skb
->dev
;
2238 if (orig_dev
->master
) {
2239 if (skb_bond_should_drop(skb
))
2240 null_or_orig
= orig_dev
; /* deliver only exact match */
2242 skb
->dev
= orig_dev
->master
;
2245 __get_cpu_var(netdev_rx_stat
).total
++;
2247 skb_reset_network_header(skb
);
2248 skb_reset_transport_header(skb
);
2249 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2255 /* Don't receive packets in an exiting network namespace */
2256 if (!net_alive(dev_net(skb
->dev
))) {
2261 #ifdef CONFIG_NET_CLS_ACT
2262 if (skb
->tc_verd
& TC_NCLS
) {
2263 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2268 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2269 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2270 ptype
->dev
== orig_dev
) {
2272 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2277 #ifdef CONFIG_NET_CLS_ACT
2278 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2284 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
2287 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
2291 type
= skb
->protocol
;
2292 list_for_each_entry_rcu(ptype
,
2293 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2294 if (ptype
->type
== type
&&
2295 (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2296 ptype
->dev
== orig_dev
)) {
2298 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2304 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2307 /* Jamal, now you will not able to escape explaining
2308 * me how you were going to use this. :-)
2318 /* Network device is going away, flush any packets still pending */
2319 static void flush_backlog(void *arg
)
2321 struct net_device
*dev
= arg
;
2322 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2323 struct sk_buff
*skb
, *tmp
;
2325 skb_queue_walk_safe(&queue
->input_pkt_queue
, skb
, tmp
)
2326 if (skb
->dev
== dev
) {
2327 __skb_unlink(skb
, &queue
->input_pkt_queue
);
2332 static int napi_gro_complete(struct sk_buff
*skb
)
2334 struct packet_type
*ptype
;
2335 __be16 type
= skb
->protocol
;
2336 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2339 if (NAPI_GRO_CB(skb
)->count
== 1)
2343 list_for_each_entry_rcu(ptype
, head
, list
) {
2344 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
2347 err
= ptype
->gro_complete(skb
);
2353 WARN_ON(&ptype
->list
== head
);
2355 return NET_RX_SUCCESS
;
2359 skb_shinfo(skb
)->gso_size
= 0;
2360 return netif_receive_skb(skb
);
2363 void napi_gro_flush(struct napi_struct
*napi
)
2365 struct sk_buff
*skb
, *next
;
2367 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
2370 napi_gro_complete(skb
);
2373 napi
->gro_list
= NULL
;
2375 EXPORT_SYMBOL(napi_gro_flush
);
2377 void *skb_gro_header(struct sk_buff
*skb
, unsigned int hlen
)
2379 unsigned int offset
= skb_gro_offset(skb
);
2382 if (hlen
<= skb_headlen(skb
))
2383 return skb
->data
+ offset
;
2385 if (unlikely(!skb_shinfo(skb
)->nr_frags
||
2386 skb_shinfo(skb
)->frags
[0].size
<=
2387 hlen
- skb_headlen(skb
) ||
2388 PageHighMem(skb_shinfo(skb
)->frags
[0].page
)))
2389 return pskb_may_pull(skb
, hlen
) ? skb
->data
+ offset
: NULL
;
2391 return page_address(skb_shinfo(skb
)->frags
[0].page
) +
2392 skb_shinfo(skb
)->frags
[0].page_offset
+ offset
;
2394 EXPORT_SYMBOL(skb_gro_header
);
2396 int dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2398 struct sk_buff
**pp
= NULL
;
2399 struct packet_type
*ptype
;
2400 __be16 type
= skb
->protocol
;
2401 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2407 if (!(skb
->dev
->features
& NETIF_F_GRO
))
2410 if (skb_is_gso(skb
) || skb_shinfo(skb
)->frag_list
)
2414 list_for_each_entry_rcu(ptype
, head
, list
) {
2418 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
2421 skb_set_network_header(skb
, skb_gro_offset(skb
));
2422 mac
= skb_gro_mac_header(skb
);
2423 mac_len
= skb
->network_header
- skb
->mac_header
;
2424 skb
->mac_len
= mac_len
;
2425 NAPI_GRO_CB(skb
)->same_flow
= 0;
2426 NAPI_GRO_CB(skb
)->flush
= 0;
2427 NAPI_GRO_CB(skb
)->free
= 0;
2429 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
2432 if (!NAPI_GRO_CB(p
)->same_flow
)
2435 if (p
->mac_len
!= mac_len
||
2436 memcmp(skb_mac_header(p
), mac
, mac_len
))
2437 NAPI_GRO_CB(p
)->same_flow
= 0;
2440 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
2445 if (&ptype
->list
== head
)
2448 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
2449 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
2452 struct sk_buff
*nskb
= *pp
;
2456 napi_gro_complete(nskb
);
2463 if (NAPI_GRO_CB(skb
)->flush
|| count
>= MAX_GRO_SKBS
)
2466 NAPI_GRO_CB(skb
)->count
= 1;
2467 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
2468 skb
->next
= napi
->gro_list
;
2469 napi
->gro_list
= skb
;
2473 if (unlikely(!pskb_may_pull(skb
, skb_gro_offset(skb
)))) {
2474 if (napi
->gro_list
== skb
)
2475 napi
->gro_list
= skb
->next
;
2486 EXPORT_SYMBOL(dev_gro_receive
);
2488 static int __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2492 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
2493 NAPI_GRO_CB(p
)->same_flow
= 1;
2494 NAPI_GRO_CB(p
)->flush
= 0;
2497 return dev_gro_receive(napi
, skb
);
2500 int napi_skb_finish(int ret
, struct sk_buff
*skb
)
2502 int err
= NET_RX_SUCCESS
;
2506 return netif_receive_skb(skb
);
2512 case GRO_MERGED_FREE
:
2519 EXPORT_SYMBOL(napi_skb_finish
);
2521 int napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2523 skb_gro_reset_offset(skb
);
2525 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
2527 EXPORT_SYMBOL(napi_gro_receive
);
2529 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
2531 __skb_pull(skb
, skb_headlen(skb
));
2532 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
2536 EXPORT_SYMBOL(napi_reuse_skb
);
2538 struct sk_buff
*napi_fraginfo_skb(struct napi_struct
*napi
,
2539 struct napi_gro_fraginfo
*info
)
2541 struct net_device
*dev
= napi
->dev
;
2542 struct sk_buff
*skb
= napi
->skb
;
2550 skb
= netdev_alloc_skb(dev
, GRO_MAX_HEAD
+ NET_IP_ALIGN
);
2554 skb_reserve(skb
, NET_IP_ALIGN
);
2557 BUG_ON(info
->nr_frags
> MAX_SKB_FRAGS
);
2558 frag
= &info
->frags
[info
->nr_frags
- 1];
2560 for (i
= skb_shinfo(skb
)->nr_frags
; i
< info
->nr_frags
; i
++) {
2561 skb_fill_page_desc(skb
, i
, frag
->page
, frag
->page_offset
,
2565 skb_shinfo(skb
)->nr_frags
= info
->nr_frags
;
2567 skb
->data_len
= info
->len
;
2568 skb
->len
+= info
->len
;
2569 skb
->truesize
+= info
->len
;
2571 skb_reset_mac_header(skb
);
2572 skb_gro_reset_offset(skb
);
2574 eth
= skb_gro_header(skb
, sizeof(*eth
));
2576 napi_reuse_skb(napi
, skb
);
2581 skb_gro_pull(skb
, sizeof(*eth
));
2584 * This works because the only protocols we care about don't require
2585 * special handling. We'll fix it up properly at the end.
2587 skb
->protocol
= eth
->h_proto
;
2589 skb
->ip_summed
= info
->ip_summed
;
2590 skb
->csum
= info
->csum
;
2595 EXPORT_SYMBOL(napi_fraginfo_skb
);
2597 int napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
, int ret
)
2599 int err
= NET_RX_SUCCESS
;
2604 skb
->protocol
= eth_type_trans(skb
, napi
->dev
);
2606 if (ret
== GRO_NORMAL
)
2607 return netif_receive_skb(skb
);
2609 skb_gro_pull(skb
, -ETH_HLEN
);
2616 case GRO_MERGED_FREE
:
2617 napi_reuse_skb(napi
, skb
);
2623 EXPORT_SYMBOL(napi_frags_finish
);
2625 int napi_gro_frags(struct napi_struct
*napi
, struct napi_gro_fraginfo
*info
)
2627 struct sk_buff
*skb
= napi_fraginfo_skb(napi
, info
);
2632 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
2634 EXPORT_SYMBOL(napi_gro_frags
);
2636 static int process_backlog(struct napi_struct
*napi
, int quota
)
2639 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2640 unsigned long start_time
= jiffies
;
2642 napi
->weight
= weight_p
;
2644 struct sk_buff
*skb
;
2646 local_irq_disable();
2647 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
2649 __napi_complete(napi
);
2655 napi_gro_receive(napi
, skb
);
2656 } while (++work
< quota
&& jiffies
== start_time
);
2658 napi_gro_flush(napi
);
2664 * __napi_schedule - schedule for receive
2665 * @n: entry to schedule
2667 * The entry's receive function will be scheduled to run
2669 void __napi_schedule(struct napi_struct
*n
)
2671 unsigned long flags
;
2673 local_irq_save(flags
);
2674 list_add_tail(&n
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
2675 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2676 local_irq_restore(flags
);
2678 EXPORT_SYMBOL(__napi_schedule
);
2680 void __napi_complete(struct napi_struct
*n
)
2682 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
2683 BUG_ON(n
->gro_list
);
2685 list_del(&n
->poll_list
);
2686 smp_mb__before_clear_bit();
2687 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
2689 EXPORT_SYMBOL(__napi_complete
);
2691 void napi_complete(struct napi_struct
*n
)
2693 unsigned long flags
;
2696 * don't let napi dequeue from the cpu poll list
2697 * just in case its running on a different cpu
2699 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
2703 local_irq_save(flags
);
2705 local_irq_restore(flags
);
2707 EXPORT_SYMBOL(napi_complete
);
2709 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
2710 int (*poll
)(struct napi_struct
*, int), int weight
)
2712 INIT_LIST_HEAD(&napi
->poll_list
);
2713 napi
->gro_list
= NULL
;
2716 napi
->weight
= weight
;
2717 list_add(&napi
->dev_list
, &dev
->napi_list
);
2719 #ifdef CONFIG_NETPOLL
2720 spin_lock_init(&napi
->poll_lock
);
2721 napi
->poll_owner
= -1;
2723 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
2725 EXPORT_SYMBOL(netif_napi_add
);
2727 void netif_napi_del(struct napi_struct
*napi
)
2729 struct sk_buff
*skb
, *next
;
2731 list_del_init(&napi
->dev_list
);
2734 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
2740 napi
->gro_list
= NULL
;
2742 EXPORT_SYMBOL(netif_napi_del
);
2745 static void net_rx_action(struct softirq_action
*h
)
2747 struct list_head
*list
= &__get_cpu_var(softnet_data
).poll_list
;
2748 unsigned long time_limit
= jiffies
+ 2;
2749 int budget
= netdev_budget
;
2752 local_irq_disable();
2754 while (!list_empty(list
)) {
2755 struct napi_struct
*n
;
2758 /* If softirq window is exhuasted then punt.
2759 * Allow this to run for 2 jiffies since which will allow
2760 * an average latency of 1.5/HZ.
2762 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
2767 /* Even though interrupts have been re-enabled, this
2768 * access is safe because interrupts can only add new
2769 * entries to the tail of this list, and only ->poll()
2770 * calls can remove this head entry from the list.
2772 n
= list_entry(list
->next
, struct napi_struct
, poll_list
);
2774 have
= netpoll_poll_lock(n
);
2778 /* This NAPI_STATE_SCHED test is for avoiding a race
2779 * with netpoll's poll_napi(). Only the entity which
2780 * obtains the lock and sees NAPI_STATE_SCHED set will
2781 * actually make the ->poll() call. Therefore we avoid
2782 * accidently calling ->poll() when NAPI is not scheduled.
2785 if (test_bit(NAPI_STATE_SCHED
, &n
->state
))
2786 work
= n
->poll(n
, weight
);
2788 WARN_ON_ONCE(work
> weight
);
2792 local_irq_disable();
2794 /* Drivers must not modify the NAPI state if they
2795 * consume the entire weight. In such cases this code
2796 * still "owns" the NAPI instance and therefore can
2797 * move the instance around on the list at-will.
2799 if (unlikely(work
== weight
)) {
2800 if (unlikely(napi_disable_pending(n
)))
2803 list_move_tail(&n
->poll_list
, list
);
2806 netpoll_poll_unlock(have
);
2811 #ifdef CONFIG_NET_DMA
2813 * There may not be any more sk_buffs coming right now, so push
2814 * any pending DMA copies to hardware
2816 dma_issue_pending_all();
2822 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2823 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2827 static gifconf_func_t
* gifconf_list
[NPROTO
];
2830 * register_gifconf - register a SIOCGIF handler
2831 * @family: Address family
2832 * @gifconf: Function handler
2834 * Register protocol dependent address dumping routines. The handler
2835 * that is passed must not be freed or reused until it has been replaced
2836 * by another handler.
2838 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2840 if (family
>= NPROTO
)
2842 gifconf_list
[family
] = gifconf
;
2848 * Map an interface index to its name (SIOCGIFNAME)
2852 * We need this ioctl for efficient implementation of the
2853 * if_indextoname() function required by the IPv6 API. Without
2854 * it, we would have to search all the interfaces to find a
2858 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
2860 struct net_device
*dev
;
2864 * Fetch the caller's info block.
2867 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2870 read_lock(&dev_base_lock
);
2871 dev
= __dev_get_by_index(net
, ifr
.ifr_ifindex
);
2873 read_unlock(&dev_base_lock
);
2877 strcpy(ifr
.ifr_name
, dev
->name
);
2878 read_unlock(&dev_base_lock
);
2880 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2886 * Perform a SIOCGIFCONF call. This structure will change
2887 * size eventually, and there is nothing I can do about it.
2888 * Thus we will need a 'compatibility mode'.
2891 static int dev_ifconf(struct net
*net
, char __user
*arg
)
2894 struct net_device
*dev
;
2901 * Fetch the caller's info block.
2904 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2911 * Loop over the interfaces, and write an info block for each.
2915 for_each_netdev(net
, dev
) {
2916 for (i
= 0; i
< NPROTO
; i
++) {
2917 if (gifconf_list
[i
]) {
2920 done
= gifconf_list
[i
](dev
, NULL
, 0);
2922 done
= gifconf_list
[i
](dev
, pos
+ total
,
2932 * All done. Write the updated control block back to the caller.
2934 ifc
.ifc_len
= total
;
2937 * Both BSD and Solaris return 0 here, so we do too.
2939 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2942 #ifdef CONFIG_PROC_FS
2944 * This is invoked by the /proc filesystem handler to display a device
2947 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2948 __acquires(dev_base_lock
)
2950 struct net
*net
= seq_file_net(seq
);
2952 struct net_device
*dev
;
2954 read_lock(&dev_base_lock
);
2956 return SEQ_START_TOKEN
;
2959 for_each_netdev(net
, dev
)
2966 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2968 struct net
*net
= seq_file_net(seq
);
2970 return v
== SEQ_START_TOKEN
?
2971 first_net_device(net
) : next_net_device((struct net_device
*)v
);
2974 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2975 __releases(dev_base_lock
)
2977 read_unlock(&dev_base_lock
);
2980 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2982 const struct net_device_stats
*stats
= dev_get_stats(dev
);
2984 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2985 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2986 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2988 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2989 stats
->rx_fifo_errors
,
2990 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2991 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2992 stats
->rx_compressed
, stats
->multicast
,
2993 stats
->tx_bytes
, stats
->tx_packets
,
2994 stats
->tx_errors
, stats
->tx_dropped
,
2995 stats
->tx_fifo_errors
, stats
->collisions
,
2996 stats
->tx_carrier_errors
+
2997 stats
->tx_aborted_errors
+
2998 stats
->tx_window_errors
+
2999 stats
->tx_heartbeat_errors
,
3000 stats
->tx_compressed
);
3004 * Called from the PROCfs module. This now uses the new arbitrary sized
3005 * /proc/net interface to create /proc/net/dev
3007 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3009 if (v
== SEQ_START_TOKEN
)
3010 seq_puts(seq
, "Inter-| Receive "
3012 " face |bytes packets errs drop fifo frame "
3013 "compressed multicast|bytes packets errs "
3014 "drop fifo colls carrier compressed\n");
3016 dev_seq_printf_stats(seq
, v
);
3020 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
3022 struct netif_rx_stats
*rc
= NULL
;
3024 while (*pos
< nr_cpu_ids
)
3025 if (cpu_online(*pos
)) {
3026 rc
= &per_cpu(netdev_rx_stat
, *pos
);
3033 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3035 return softnet_get_online(pos
);
3038 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3041 return softnet_get_online(pos
);
3044 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3048 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3050 struct netif_rx_stats
*s
= v
;
3052 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3053 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
3054 0, 0, 0, 0, /* was fastroute */
3059 static const struct seq_operations dev_seq_ops
= {
3060 .start
= dev_seq_start
,
3061 .next
= dev_seq_next
,
3062 .stop
= dev_seq_stop
,
3063 .show
= dev_seq_show
,
3066 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3068 return seq_open_net(inode
, file
, &dev_seq_ops
,
3069 sizeof(struct seq_net_private
));
3072 static const struct file_operations dev_seq_fops
= {
3073 .owner
= THIS_MODULE
,
3074 .open
= dev_seq_open
,
3076 .llseek
= seq_lseek
,
3077 .release
= seq_release_net
,
3080 static const struct seq_operations softnet_seq_ops
= {
3081 .start
= softnet_seq_start
,
3082 .next
= softnet_seq_next
,
3083 .stop
= softnet_seq_stop
,
3084 .show
= softnet_seq_show
,
3087 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3089 return seq_open(file
, &softnet_seq_ops
);
3092 static const struct file_operations softnet_seq_fops
= {
3093 .owner
= THIS_MODULE
,
3094 .open
= softnet_seq_open
,
3096 .llseek
= seq_lseek
,
3097 .release
= seq_release
,
3100 static void *ptype_get_idx(loff_t pos
)
3102 struct packet_type
*pt
= NULL
;
3106 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3112 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3113 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3122 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3126 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3129 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3131 struct packet_type
*pt
;
3132 struct list_head
*nxt
;
3136 if (v
== SEQ_START_TOKEN
)
3137 return ptype_get_idx(0);
3140 nxt
= pt
->list
.next
;
3141 if (pt
->type
== htons(ETH_P_ALL
)) {
3142 if (nxt
!= &ptype_all
)
3145 nxt
= ptype_base
[0].next
;
3147 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3149 while (nxt
== &ptype_base
[hash
]) {
3150 if (++hash
>= PTYPE_HASH_SIZE
)
3152 nxt
= ptype_base
[hash
].next
;
3155 return list_entry(nxt
, struct packet_type
, list
);
3158 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3164 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3166 struct packet_type
*pt
= v
;
3168 if (v
== SEQ_START_TOKEN
)
3169 seq_puts(seq
, "Type Device Function\n");
3170 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3171 if (pt
->type
== htons(ETH_P_ALL
))
3172 seq_puts(seq
, "ALL ");
3174 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3176 seq_printf(seq
, " %-8s %pF\n",
3177 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
3183 static const struct seq_operations ptype_seq_ops
= {
3184 .start
= ptype_seq_start
,
3185 .next
= ptype_seq_next
,
3186 .stop
= ptype_seq_stop
,
3187 .show
= ptype_seq_show
,
3190 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
3192 return seq_open_net(inode
, file
, &ptype_seq_ops
,
3193 sizeof(struct seq_net_private
));
3196 static const struct file_operations ptype_seq_fops
= {
3197 .owner
= THIS_MODULE
,
3198 .open
= ptype_seq_open
,
3200 .llseek
= seq_lseek
,
3201 .release
= seq_release_net
,
3205 static int __net_init
dev_proc_net_init(struct net
*net
)
3209 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
3211 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
3213 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
3216 if (wext_proc_init(net
))
3222 proc_net_remove(net
, "ptype");
3224 proc_net_remove(net
, "softnet_stat");
3226 proc_net_remove(net
, "dev");
3230 static void __net_exit
dev_proc_net_exit(struct net
*net
)
3232 wext_proc_exit(net
);
3234 proc_net_remove(net
, "ptype");
3235 proc_net_remove(net
, "softnet_stat");
3236 proc_net_remove(net
, "dev");
3239 static struct pernet_operations __net_initdata dev_proc_ops
= {
3240 .init
= dev_proc_net_init
,
3241 .exit
= dev_proc_net_exit
,
3244 static int __init
dev_proc_init(void)
3246 return register_pernet_subsys(&dev_proc_ops
);
3249 #define dev_proc_init() 0
3250 #endif /* CONFIG_PROC_FS */
3254 * netdev_set_master - set up master/slave pair
3255 * @slave: slave device
3256 * @master: new master device
3258 * Changes the master device of the slave. Pass %NULL to break the
3259 * bonding. The caller must hold the RTNL semaphore. On a failure
3260 * a negative errno code is returned. On success the reference counts
3261 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3262 * function returns zero.
3264 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
3266 struct net_device
*old
= slave
->master
;
3276 slave
->master
= master
;
3284 slave
->flags
|= IFF_SLAVE
;
3286 slave
->flags
&= ~IFF_SLAVE
;
3288 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
3292 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
3294 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3296 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
3297 ops
->ndo_change_rx_flags(dev
, flags
);
3300 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
3302 unsigned short old_flags
= dev
->flags
;
3308 dev
->flags
|= IFF_PROMISC
;
3309 dev
->promiscuity
+= inc
;
3310 if (dev
->promiscuity
== 0) {
3313 * If inc causes overflow, untouch promisc and return error.
3316 dev
->flags
&= ~IFF_PROMISC
;
3318 dev
->promiscuity
-= inc
;
3319 printk(KERN_WARNING
"%s: promiscuity touches roof, "
3320 "set promiscuity failed, promiscuity feature "
3321 "of device might be broken.\n", dev
->name
);
3325 if (dev
->flags
!= old_flags
) {
3326 printk(KERN_INFO
"device %s %s promiscuous mode\n",
3327 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
3329 if (audit_enabled
) {
3330 current_uid_gid(&uid
, &gid
);
3331 audit_log(current
->audit_context
, GFP_ATOMIC
,
3332 AUDIT_ANOM_PROMISCUOUS
,
3333 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3334 dev
->name
, (dev
->flags
& IFF_PROMISC
),
3335 (old_flags
& IFF_PROMISC
),
3336 audit_get_loginuid(current
),
3338 audit_get_sessionid(current
));
3341 dev_change_rx_flags(dev
, IFF_PROMISC
);
3347 * dev_set_promiscuity - update promiscuity count on a device
3351 * Add or remove promiscuity from a device. While the count in the device
3352 * remains above zero the interface remains promiscuous. Once it hits zero
3353 * the device reverts back to normal filtering operation. A negative inc
3354 * value is used to drop promiscuity on the device.
3355 * Return 0 if successful or a negative errno code on error.
3357 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
3359 unsigned short old_flags
= dev
->flags
;
3362 err
= __dev_set_promiscuity(dev
, inc
);
3365 if (dev
->flags
!= old_flags
)
3366 dev_set_rx_mode(dev
);
3371 * dev_set_allmulti - update allmulti count on a device
3375 * Add or remove reception of all multicast frames to a device. While the
3376 * count in the device remains above zero the interface remains listening
3377 * to all interfaces. Once it hits zero the device reverts back to normal
3378 * filtering operation. A negative @inc value is used to drop the counter
3379 * when releasing a resource needing all multicasts.
3380 * Return 0 if successful or a negative errno code on error.
3383 int dev_set_allmulti(struct net_device
*dev
, int inc
)
3385 unsigned short old_flags
= dev
->flags
;
3389 dev
->flags
|= IFF_ALLMULTI
;
3390 dev
->allmulti
+= inc
;
3391 if (dev
->allmulti
== 0) {
3394 * If inc causes overflow, untouch allmulti and return error.
3397 dev
->flags
&= ~IFF_ALLMULTI
;
3399 dev
->allmulti
-= inc
;
3400 printk(KERN_WARNING
"%s: allmulti touches roof, "
3401 "set allmulti failed, allmulti feature of "
3402 "device might be broken.\n", dev
->name
);
3406 if (dev
->flags
^ old_flags
) {
3407 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
3408 dev_set_rx_mode(dev
);
3414 * Upload unicast and multicast address lists to device and
3415 * configure RX filtering. When the device doesn't support unicast
3416 * filtering it is put in promiscuous mode while unicast addresses
3419 void __dev_set_rx_mode(struct net_device
*dev
)
3421 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3423 /* dev_open will call this function so the list will stay sane. */
3424 if (!(dev
->flags
&IFF_UP
))
3427 if (!netif_device_present(dev
))
3430 if (ops
->ndo_set_rx_mode
)
3431 ops
->ndo_set_rx_mode(dev
);
3433 /* Unicast addresses changes may only happen under the rtnl,
3434 * therefore calling __dev_set_promiscuity here is safe.
3436 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
3437 __dev_set_promiscuity(dev
, 1);
3438 dev
->uc_promisc
= 1;
3439 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
3440 __dev_set_promiscuity(dev
, -1);
3441 dev
->uc_promisc
= 0;
3444 if (ops
->ndo_set_multicast_list
)
3445 ops
->ndo_set_multicast_list(dev
);
3449 void dev_set_rx_mode(struct net_device
*dev
)
3451 netif_addr_lock_bh(dev
);
3452 __dev_set_rx_mode(dev
);
3453 netif_addr_unlock_bh(dev
);
3456 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
3457 void *addr
, int alen
, int glbl
)
3459 struct dev_addr_list
*da
;
3461 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
3462 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
3463 alen
== da
->da_addrlen
) {
3465 int old_glbl
= da
->da_gusers
;
3482 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
3483 void *addr
, int alen
, int glbl
)
3485 struct dev_addr_list
*da
;
3487 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
3488 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
3489 da
->da_addrlen
== alen
) {
3491 int old_glbl
= da
->da_gusers
;
3501 da
= kzalloc(sizeof(*da
), GFP_ATOMIC
);
3504 memcpy(da
->da_addr
, addr
, alen
);
3505 da
->da_addrlen
= alen
;
3507 da
->da_gusers
= glbl
? 1 : 0;
3515 * dev_unicast_delete - Release secondary unicast address.
3517 * @addr: address to delete
3518 * @alen: length of @addr
3520 * Release reference to a secondary unicast address and remove it
3521 * from the device if the reference count drops to zero.
3523 * The caller must hold the rtnl_mutex.
3525 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
3531 netif_addr_lock_bh(dev
);
3532 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3534 __dev_set_rx_mode(dev
);
3535 netif_addr_unlock_bh(dev
);
3538 EXPORT_SYMBOL(dev_unicast_delete
);
3541 * dev_unicast_add - add a secondary unicast address
3543 * @addr: address to add
3544 * @alen: length of @addr
3546 * Add a secondary unicast address to the device or increase
3547 * the reference count if it already exists.
3549 * The caller must hold the rtnl_mutex.
3551 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
3557 netif_addr_lock_bh(dev
);
3558 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3560 __dev_set_rx_mode(dev
);
3561 netif_addr_unlock_bh(dev
);
3564 EXPORT_SYMBOL(dev_unicast_add
);
3566 int __dev_addr_sync(struct dev_addr_list
**to
, int *to_count
,
3567 struct dev_addr_list
**from
, int *from_count
)
3569 struct dev_addr_list
*da
, *next
;
3573 while (da
!= NULL
) {
3575 if (!da
->da_synced
) {
3576 err
= __dev_addr_add(to
, to_count
,
3577 da
->da_addr
, da
->da_addrlen
, 0);
3582 } else if (da
->da_users
== 1) {
3583 __dev_addr_delete(to
, to_count
,
3584 da
->da_addr
, da
->da_addrlen
, 0);
3585 __dev_addr_delete(from
, from_count
,
3586 da
->da_addr
, da
->da_addrlen
, 0);
3593 void __dev_addr_unsync(struct dev_addr_list
**to
, int *to_count
,
3594 struct dev_addr_list
**from
, int *from_count
)
3596 struct dev_addr_list
*da
, *next
;
3599 while (da
!= NULL
) {
3601 if (da
->da_synced
) {
3602 __dev_addr_delete(to
, to_count
,
3603 da
->da_addr
, da
->da_addrlen
, 0);
3605 __dev_addr_delete(from
, from_count
,
3606 da
->da_addr
, da
->da_addrlen
, 0);
3613 * dev_unicast_sync - Synchronize device's unicast list to another device
3614 * @to: destination device
3615 * @from: source device
3617 * Add newly added addresses to the destination device and release
3618 * addresses that have no users left. The source device must be
3619 * locked by netif_tx_lock_bh.
3621 * This function is intended to be called from the dev->set_rx_mode
3622 * function of layered software devices.
3624 int dev_unicast_sync(struct net_device
*to
, struct net_device
*from
)
3628 netif_addr_lock_bh(to
);
3629 err
= __dev_addr_sync(&to
->uc_list
, &to
->uc_count
,
3630 &from
->uc_list
, &from
->uc_count
);
3632 __dev_set_rx_mode(to
);
3633 netif_addr_unlock_bh(to
);
3636 EXPORT_SYMBOL(dev_unicast_sync
);
3639 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3640 * @to: destination device
3641 * @from: source device
3643 * Remove all addresses that were added to the destination device by
3644 * dev_unicast_sync(). This function is intended to be called from the
3645 * dev->stop function of layered software devices.
3647 void dev_unicast_unsync(struct net_device
*to
, struct net_device
*from
)
3649 netif_addr_lock_bh(from
);
3650 netif_addr_lock(to
);
3652 __dev_addr_unsync(&to
->uc_list
, &to
->uc_count
,
3653 &from
->uc_list
, &from
->uc_count
);
3654 __dev_set_rx_mode(to
);
3656 netif_addr_unlock(to
);
3657 netif_addr_unlock_bh(from
);
3659 EXPORT_SYMBOL(dev_unicast_unsync
);
3661 static void __dev_addr_discard(struct dev_addr_list
**list
)
3663 struct dev_addr_list
*tmp
;
3665 while (*list
!= NULL
) {
3668 if (tmp
->da_users
> tmp
->da_gusers
)
3669 printk("__dev_addr_discard: address leakage! "
3670 "da_users=%d\n", tmp
->da_users
);
3675 static void dev_addr_discard(struct net_device
*dev
)
3677 netif_addr_lock_bh(dev
);
3679 __dev_addr_discard(&dev
->uc_list
);
3682 __dev_addr_discard(&dev
->mc_list
);
3685 netif_addr_unlock_bh(dev
);
3689 * dev_get_flags - get flags reported to userspace
3692 * Get the combination of flag bits exported through APIs to userspace.
3694 unsigned dev_get_flags(const struct net_device
*dev
)
3698 flags
= (dev
->flags
& ~(IFF_PROMISC
|
3703 (dev
->gflags
& (IFF_PROMISC
|
3706 if (netif_running(dev
)) {
3707 if (netif_oper_up(dev
))
3708 flags
|= IFF_RUNNING
;
3709 if (netif_carrier_ok(dev
))
3710 flags
|= IFF_LOWER_UP
;
3711 if (netif_dormant(dev
))
3712 flags
|= IFF_DORMANT
;
3719 * dev_change_flags - change device settings
3721 * @flags: device state flags
3723 * Change settings on device based state flags. The flags are
3724 * in the userspace exported format.
3726 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
3729 int old_flags
= dev
->flags
;
3734 * Set the flags on our device.
3737 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
3738 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
3740 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
3744 * Load in the correct multicast list now the flags have changed.
3747 if ((old_flags
^ flags
) & IFF_MULTICAST
)
3748 dev_change_rx_flags(dev
, IFF_MULTICAST
);
3750 dev_set_rx_mode(dev
);
3753 * Have we downed the interface. We handle IFF_UP ourselves
3754 * according to user attempts to set it, rather than blindly
3759 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
3760 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
3763 dev_set_rx_mode(dev
);
3766 if (dev
->flags
& IFF_UP
&&
3767 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
3769 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
3771 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
3772 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
3773 dev
->gflags
^= IFF_PROMISC
;
3774 dev_set_promiscuity(dev
, inc
);
3777 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3778 is important. Some (broken) drivers set IFF_PROMISC, when
3779 IFF_ALLMULTI is requested not asking us and not reporting.
3781 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
3782 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
3783 dev
->gflags
^= IFF_ALLMULTI
;
3784 dev_set_allmulti(dev
, inc
);
3787 /* Exclude state transition flags, already notified */
3788 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
3790 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
3796 * dev_set_mtu - Change maximum transfer unit
3798 * @new_mtu: new transfer unit
3800 * Change the maximum transfer size of the network device.
3802 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
3804 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3807 if (new_mtu
== dev
->mtu
)
3810 /* MTU must be positive. */
3814 if (!netif_device_present(dev
))
3818 if (ops
->ndo_change_mtu
)
3819 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
3823 if (!err
&& dev
->flags
& IFF_UP
)
3824 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
3829 * dev_set_mac_address - Change Media Access Control Address
3833 * Change the hardware (MAC) address of the device
3835 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
3837 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3840 if (!ops
->ndo_set_mac_address
)
3842 if (sa
->sa_family
!= dev
->type
)
3844 if (!netif_device_present(dev
))
3846 err
= ops
->ndo_set_mac_address(dev
, sa
);
3848 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3853 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3855 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3858 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3864 case SIOCGIFFLAGS
: /* Get interface flags */
3865 ifr
->ifr_flags
= dev_get_flags(dev
);
3868 case SIOCGIFMETRIC
: /* Get the metric on the interface
3869 (currently unused) */
3870 ifr
->ifr_metric
= 0;
3873 case SIOCGIFMTU
: /* Get the MTU of a device */
3874 ifr
->ifr_mtu
= dev
->mtu
;
3879 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
3881 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
3882 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3883 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
3891 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
3892 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
3893 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
3894 ifr
->ifr_map
.irq
= dev
->irq
;
3895 ifr
->ifr_map
.dma
= dev
->dma
;
3896 ifr
->ifr_map
.port
= dev
->if_port
;
3900 ifr
->ifr_ifindex
= dev
->ifindex
;
3904 ifr
->ifr_qlen
= dev
->tx_queue_len
;
3908 /* dev_ioctl() should ensure this case
3920 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3922 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3925 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3926 const struct net_device_ops
*ops
;
3931 ops
= dev
->netdev_ops
;
3934 case SIOCSIFFLAGS
: /* Set interface flags */
3935 return dev_change_flags(dev
, ifr
->ifr_flags
);
3937 case SIOCSIFMETRIC
: /* Set the metric on the interface
3938 (currently unused) */
3941 case SIOCSIFMTU
: /* Set the MTU of a device */
3942 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
3945 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
3947 case SIOCSIFHWBROADCAST
:
3948 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
3950 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
3951 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3952 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3956 if (ops
->ndo_set_config
) {
3957 if (!netif_device_present(dev
))
3959 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
3964 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
3965 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3967 if (!netif_device_present(dev
))
3969 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
3973 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
3974 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3976 if (!netif_device_present(dev
))
3978 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
3982 if (ifr
->ifr_qlen
< 0)
3984 dev
->tx_queue_len
= ifr
->ifr_qlen
;
3988 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
3989 return dev_change_name(dev
, ifr
->ifr_newname
);
3992 * Unknown or private ioctl
3996 if ((cmd
>= SIOCDEVPRIVATE
&&
3997 cmd
<= SIOCDEVPRIVATE
+ 15) ||
3998 cmd
== SIOCBONDENSLAVE
||
3999 cmd
== SIOCBONDRELEASE
||
4000 cmd
== SIOCBONDSETHWADDR
||
4001 cmd
== SIOCBONDSLAVEINFOQUERY
||
4002 cmd
== SIOCBONDINFOQUERY
||
4003 cmd
== SIOCBONDCHANGEACTIVE
||
4004 cmd
== SIOCGMIIPHY
||
4005 cmd
== SIOCGMIIREG
||
4006 cmd
== SIOCSMIIREG
||
4007 cmd
== SIOCBRADDIF
||
4008 cmd
== SIOCBRDELIF
||
4009 cmd
== SIOCWANDEV
) {
4011 if (ops
->ndo_do_ioctl
) {
4012 if (netif_device_present(dev
))
4013 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4025 * This function handles all "interface"-type I/O control requests. The actual
4026 * 'doing' part of this is dev_ifsioc above.
4030 * dev_ioctl - network device ioctl
4031 * @net: the applicable net namespace
4032 * @cmd: command to issue
4033 * @arg: pointer to a struct ifreq in user space
4035 * Issue ioctl functions to devices. This is normally called by the
4036 * user space syscall interfaces but can sometimes be useful for
4037 * other purposes. The return value is the return from the syscall if
4038 * positive or a negative errno code on error.
4041 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4047 /* One special case: SIOCGIFCONF takes ifconf argument
4048 and requires shared lock, because it sleeps writing
4052 if (cmd
== SIOCGIFCONF
) {
4054 ret
= dev_ifconf(net
, (char __user
*) arg
);
4058 if (cmd
== SIOCGIFNAME
)
4059 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4061 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4064 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4066 colon
= strchr(ifr
.ifr_name
, ':');
4071 * See which interface the caller is talking about.
4076 * These ioctl calls:
4077 * - can be done by all.
4078 * - atomic and do not require locking.
4089 dev_load(net
, ifr
.ifr_name
);
4090 read_lock(&dev_base_lock
);
4091 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4092 read_unlock(&dev_base_lock
);
4096 if (copy_to_user(arg
, &ifr
,
4097 sizeof(struct ifreq
)))
4103 dev_load(net
, ifr
.ifr_name
);
4105 ret
= dev_ethtool(net
, &ifr
);
4110 if (copy_to_user(arg
, &ifr
,
4111 sizeof(struct ifreq
)))
4117 * These ioctl calls:
4118 * - require superuser power.
4119 * - require strict serialization.
4125 if (!capable(CAP_NET_ADMIN
))
4127 dev_load(net
, ifr
.ifr_name
);
4129 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4134 if (copy_to_user(arg
, &ifr
,
4135 sizeof(struct ifreq
)))
4141 * These ioctl calls:
4142 * - require superuser power.
4143 * - require strict serialization.
4144 * - do not return a value
4154 case SIOCSIFHWBROADCAST
:
4157 case SIOCBONDENSLAVE
:
4158 case SIOCBONDRELEASE
:
4159 case SIOCBONDSETHWADDR
:
4160 case SIOCBONDCHANGEACTIVE
:
4163 if (!capable(CAP_NET_ADMIN
))
4166 case SIOCBONDSLAVEINFOQUERY
:
4167 case SIOCBONDINFOQUERY
:
4168 dev_load(net
, ifr
.ifr_name
);
4170 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4175 /* Get the per device memory space. We can add this but
4176 * currently do not support it */
4178 /* Set the per device memory buffer space.
4179 * Not applicable in our case */
4184 * Unknown or private ioctl.
4187 if (cmd
== SIOCWANDEV
||
4188 (cmd
>= SIOCDEVPRIVATE
&&
4189 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4190 dev_load(net
, ifr
.ifr_name
);
4192 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4194 if (!ret
&& copy_to_user(arg
, &ifr
,
4195 sizeof(struct ifreq
)))
4199 /* Take care of Wireless Extensions */
4200 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4201 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4208 * dev_new_index - allocate an ifindex
4209 * @net: the applicable net namespace
4211 * Returns a suitable unique value for a new device interface
4212 * number. The caller must hold the rtnl semaphore or the
4213 * dev_base_lock to be sure it remains unique.
4215 static int dev_new_index(struct net
*net
)
4221 if (!__dev_get_by_index(net
, ifindex
))
4226 /* Delayed registration/unregisteration */
4227 static LIST_HEAD(net_todo_list
);
4229 static void net_set_todo(struct net_device
*dev
)
4231 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4234 static void rollback_registered(struct net_device
*dev
)
4236 BUG_ON(dev_boot_phase
);
4239 /* Some devices call without registering for initialization unwind. */
4240 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4241 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
4242 "was registered\n", dev
->name
, dev
);
4248 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4250 /* If device is running, close it first. */
4253 /* And unlink it from device chain. */
4254 unlist_netdevice(dev
);
4256 dev
->reg_state
= NETREG_UNREGISTERING
;
4260 /* Shutdown queueing discipline. */
4264 /* Notify protocols, that we are about to destroy
4265 this device. They should clean all the things.
4267 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4270 * Flush the unicast and multicast chains
4272 dev_addr_discard(dev
);
4274 if (dev
->netdev_ops
->ndo_uninit
)
4275 dev
->netdev_ops
->ndo_uninit(dev
);
4277 /* Notifier chain MUST detach us from master device. */
4278 WARN_ON(dev
->master
);
4280 /* Remove entries from kobject tree */
4281 netdev_unregister_kobject(dev
);
4288 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
4289 struct netdev_queue
*dev_queue
,
4292 spin_lock_init(&dev_queue
->_xmit_lock
);
4293 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
4294 dev_queue
->xmit_lock_owner
= -1;
4297 static void netdev_init_queue_locks(struct net_device
*dev
)
4299 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
4300 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
4303 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
4305 /* Fix illegal SG+CSUM combinations. */
4306 if ((features
& NETIF_F_SG
) &&
4307 !(features
& NETIF_F_ALL_CSUM
)) {
4309 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
4310 "checksum feature.\n", name
);
4311 features
&= ~NETIF_F_SG
;
4314 /* TSO requires that SG is present as well. */
4315 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
4317 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
4318 "SG feature.\n", name
);
4319 features
&= ~NETIF_F_TSO
;
4322 if (features
& NETIF_F_UFO
) {
4323 if (!(features
& NETIF_F_GEN_CSUM
)) {
4325 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4326 "since no NETIF_F_HW_CSUM feature.\n",
4328 features
&= ~NETIF_F_UFO
;
4331 if (!(features
& NETIF_F_SG
)) {
4333 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4334 "since no NETIF_F_SG feature.\n", name
);
4335 features
&= ~NETIF_F_UFO
;
4341 EXPORT_SYMBOL(netdev_fix_features
);
4344 * register_netdevice - register a network device
4345 * @dev: device to register
4347 * Take a completed network device structure and add it to the kernel
4348 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4349 * chain. 0 is returned on success. A negative errno code is returned
4350 * on a failure to set up the device, or if the name is a duplicate.
4352 * Callers must hold the rtnl semaphore. You may want
4353 * register_netdev() instead of this.
4356 * The locking appears insufficient to guarantee two parallel registers
4357 * will not get the same name.
4360 int register_netdevice(struct net_device
*dev
)
4362 struct hlist_head
*head
;
4363 struct hlist_node
*p
;
4365 struct net
*net
= dev_net(dev
);
4367 BUG_ON(dev_boot_phase
);
4372 /* When net_device's are persistent, this will be fatal. */
4373 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
4376 spin_lock_init(&dev
->addr_list_lock
);
4377 netdev_set_addr_lockdep_class(dev
);
4378 netdev_init_queue_locks(dev
);
4382 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4383 /* Netdevice_ops API compatiability support.
4384 * This is temporary until all network devices are converted.
4386 if (dev
->netdev_ops
) {
4387 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4389 dev
->init
= ops
->ndo_init
;
4390 dev
->uninit
= ops
->ndo_uninit
;
4391 dev
->open
= ops
->ndo_open
;
4392 dev
->change_rx_flags
= ops
->ndo_change_rx_flags
;
4393 dev
->set_rx_mode
= ops
->ndo_set_rx_mode
;
4394 dev
->set_multicast_list
= ops
->ndo_set_multicast_list
;
4395 dev
->set_mac_address
= ops
->ndo_set_mac_address
;
4396 dev
->validate_addr
= ops
->ndo_validate_addr
;
4397 dev
->do_ioctl
= ops
->ndo_do_ioctl
;
4398 dev
->set_config
= ops
->ndo_set_config
;
4399 dev
->change_mtu
= ops
->ndo_change_mtu
;
4400 dev
->tx_timeout
= ops
->ndo_tx_timeout
;
4401 dev
->get_stats
= ops
->ndo_get_stats
;
4402 dev
->vlan_rx_register
= ops
->ndo_vlan_rx_register
;
4403 dev
->vlan_rx_add_vid
= ops
->ndo_vlan_rx_add_vid
;
4404 dev
->vlan_rx_kill_vid
= ops
->ndo_vlan_rx_kill_vid
;
4405 #ifdef CONFIG_NET_POLL_CONTROLLER
4406 dev
->poll_controller
= ops
->ndo_poll_controller
;
4409 char drivername
[64];
4410 pr_info("%s (%s): not using net_device_ops yet\n",
4411 dev
->name
, netdev_drivername(dev
, drivername
, 64));
4413 /* This works only because net_device_ops and the
4414 compatiablity structure are the same. */
4415 dev
->netdev_ops
= (void *) &(dev
->init
);
4419 /* Init, if this function is available */
4420 if (dev
->netdev_ops
->ndo_init
) {
4421 ret
= dev
->netdev_ops
->ndo_init(dev
);
4429 if (!dev_valid_name(dev
->name
)) {
4434 dev
->ifindex
= dev_new_index(net
);
4435 if (dev
->iflink
== -1)
4436 dev
->iflink
= dev
->ifindex
;
4438 /* Check for existence of name */
4439 head
= dev_name_hash(net
, dev
->name
);
4440 hlist_for_each(p
, head
) {
4441 struct net_device
*d
4442 = hlist_entry(p
, struct net_device
, name_hlist
);
4443 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
4449 /* Fix illegal checksum combinations */
4450 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
4451 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4452 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
4454 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
4457 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
4458 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4459 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
4461 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
4464 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
4466 /* Enable software GSO if SG is supported. */
4467 if (dev
->features
& NETIF_F_SG
)
4468 dev
->features
|= NETIF_F_GSO
;
4470 netdev_initialize_kobject(dev
);
4471 ret
= netdev_register_kobject(dev
);
4474 dev
->reg_state
= NETREG_REGISTERED
;
4477 * Default initial state at registry is that the
4478 * device is present.
4481 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
4483 dev_init_scheduler(dev
);
4485 list_netdevice(dev
);
4487 /* Notify protocols, that a new device appeared. */
4488 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
4489 ret
= notifier_to_errno(ret
);
4491 rollback_registered(dev
);
4492 dev
->reg_state
= NETREG_UNREGISTERED
;
4499 if (dev
->netdev_ops
->ndo_uninit
)
4500 dev
->netdev_ops
->ndo_uninit(dev
);
4505 * init_dummy_netdev - init a dummy network device for NAPI
4506 * @dev: device to init
4508 * This takes a network device structure and initialize the minimum
4509 * amount of fields so it can be used to schedule NAPI polls without
4510 * registering a full blown interface. This is to be used by drivers
4511 * that need to tie several hardware interfaces to a single NAPI
4512 * poll scheduler due to HW limitations.
4514 int init_dummy_netdev(struct net_device
*dev
)
4516 /* Clear everything. Note we don't initialize spinlocks
4517 * are they aren't supposed to be taken by any of the
4518 * NAPI code and this dummy netdev is supposed to be
4519 * only ever used for NAPI polls
4521 memset(dev
, 0, sizeof(struct net_device
));
4523 /* make sure we BUG if trying to hit standard
4524 * register/unregister code path
4526 dev
->reg_state
= NETREG_DUMMY
;
4528 /* initialize the ref count */
4529 atomic_set(&dev
->refcnt
, 1);
4531 /* NAPI wants this */
4532 INIT_LIST_HEAD(&dev
->napi_list
);
4534 /* a dummy interface is started by default */
4535 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
4536 set_bit(__LINK_STATE_START
, &dev
->state
);
4540 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
4544 * register_netdev - register a network device
4545 * @dev: device to register
4547 * Take a completed network device structure and add it to the kernel
4548 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4549 * chain. 0 is returned on success. A negative errno code is returned
4550 * on a failure to set up the device, or if the name is a duplicate.
4552 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4553 * and expands the device name if you passed a format string to
4556 int register_netdev(struct net_device
*dev
)
4563 * If the name is a format string the caller wants us to do a
4566 if (strchr(dev
->name
, '%')) {
4567 err
= dev_alloc_name(dev
, dev
->name
);
4572 err
= register_netdevice(dev
);
4577 EXPORT_SYMBOL(register_netdev
);
4580 * netdev_wait_allrefs - wait until all references are gone.
4582 * This is called when unregistering network devices.
4584 * Any protocol or device that holds a reference should register
4585 * for netdevice notification, and cleanup and put back the
4586 * reference if they receive an UNREGISTER event.
4587 * We can get stuck here if buggy protocols don't correctly
4590 static void netdev_wait_allrefs(struct net_device
*dev
)
4592 unsigned long rebroadcast_time
, warning_time
;
4594 rebroadcast_time
= warning_time
= jiffies
;
4595 while (atomic_read(&dev
->refcnt
) != 0) {
4596 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
4599 /* Rebroadcast unregister notification */
4600 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4602 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
4604 /* We must not have linkwatch events
4605 * pending on unregister. If this
4606 * happens, we simply run the queue
4607 * unscheduled, resulting in a noop
4610 linkwatch_run_queue();
4615 rebroadcast_time
= jiffies
;
4620 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
4621 printk(KERN_EMERG
"unregister_netdevice: "
4622 "waiting for %s to become free. Usage "
4624 dev
->name
, atomic_read(&dev
->refcnt
));
4625 warning_time
= jiffies
;
4634 * register_netdevice(x1);
4635 * register_netdevice(x2);
4637 * unregister_netdevice(y1);
4638 * unregister_netdevice(y2);
4644 * We are invoked by rtnl_unlock().
4645 * This allows us to deal with problems:
4646 * 1) We can delete sysfs objects which invoke hotplug
4647 * without deadlocking with linkwatch via keventd.
4648 * 2) Since we run with the RTNL semaphore not held, we can sleep
4649 * safely in order to wait for the netdev refcnt to drop to zero.
4651 * We must not return until all unregister events added during
4652 * the interval the lock was held have been completed.
4654 void netdev_run_todo(void)
4656 struct list_head list
;
4658 /* Snapshot list, allow later requests */
4659 list_replace_init(&net_todo_list
, &list
);
4663 while (!list_empty(&list
)) {
4664 struct net_device
*dev
4665 = list_entry(list
.next
, struct net_device
, todo_list
);
4666 list_del(&dev
->todo_list
);
4668 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
4669 printk(KERN_ERR
"network todo '%s' but state %d\n",
4670 dev
->name
, dev
->reg_state
);
4675 dev
->reg_state
= NETREG_UNREGISTERED
;
4677 on_each_cpu(flush_backlog
, dev
, 1);
4679 netdev_wait_allrefs(dev
);
4682 BUG_ON(atomic_read(&dev
->refcnt
));
4683 WARN_ON(dev
->ip_ptr
);
4684 WARN_ON(dev
->ip6_ptr
);
4685 WARN_ON(dev
->dn_ptr
);
4687 if (dev
->destructor
)
4688 dev
->destructor(dev
);
4690 /* Free network device */
4691 kobject_put(&dev
->dev
.kobj
);
4696 * dev_get_stats - get network device statistics
4697 * @dev: device to get statistics from
4699 * Get network statistics from device. The device driver may provide
4700 * its own method by setting dev->netdev_ops->get_stats; otherwise
4701 * the internal statistics structure is used.
4703 const struct net_device_stats
*dev_get_stats(struct net_device
*dev
)
4705 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4707 if (ops
->ndo_get_stats
)
4708 return ops
->ndo_get_stats(dev
);
4712 EXPORT_SYMBOL(dev_get_stats
);
4714 static void netdev_init_one_queue(struct net_device
*dev
,
4715 struct netdev_queue
*queue
,
4721 static void netdev_init_queues(struct net_device
*dev
)
4723 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
4724 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
4725 spin_lock_init(&dev
->tx_global_lock
);
4729 * alloc_netdev_mq - allocate network device
4730 * @sizeof_priv: size of private data to allocate space for
4731 * @name: device name format string
4732 * @setup: callback to initialize device
4733 * @queue_count: the number of subqueues to allocate
4735 * Allocates a struct net_device with private data area for driver use
4736 * and performs basic initialization. Also allocates subquue structs
4737 * for each queue on the device at the end of the netdevice.
4739 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
4740 void (*setup
)(struct net_device
*), unsigned int queue_count
)
4742 struct netdev_queue
*tx
;
4743 struct net_device
*dev
;
4747 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
4749 alloc_size
= sizeof(struct net_device
);
4751 /* ensure 32-byte alignment of private area */
4752 alloc_size
= (alloc_size
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
4753 alloc_size
+= sizeof_priv
;
4755 /* ensure 32-byte alignment of whole construct */
4756 alloc_size
+= NETDEV_ALIGN_CONST
;
4758 p
= kzalloc(alloc_size
, GFP_KERNEL
);
4760 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
4764 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
4766 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
4772 dev
= (struct net_device
*)
4773 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
4774 dev
->padded
= (char *)dev
- (char *)p
;
4775 dev_net_set(dev
, &init_net
);
4778 dev
->num_tx_queues
= queue_count
;
4779 dev
->real_num_tx_queues
= queue_count
;
4781 dev
->gso_max_size
= GSO_MAX_SIZE
;
4783 netdev_init_queues(dev
);
4785 INIT_LIST_HEAD(&dev
->napi_list
);
4787 strcpy(dev
->name
, name
);
4790 EXPORT_SYMBOL(alloc_netdev_mq
);
4793 * free_netdev - free network device
4796 * This function does the last stage of destroying an allocated device
4797 * interface. The reference to the device object is released.
4798 * If this is the last reference then it will be freed.
4800 void free_netdev(struct net_device
*dev
)
4802 struct napi_struct
*p
, *n
;
4804 release_net(dev_net(dev
));
4808 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
4811 /* Compatibility with error handling in drivers */
4812 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4813 kfree((char *)dev
- dev
->padded
);
4817 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
4818 dev
->reg_state
= NETREG_RELEASED
;
4820 /* will free via device release */
4821 put_device(&dev
->dev
);
4825 * synchronize_net - Synchronize with packet receive processing
4827 * Wait for packets currently being received to be done.
4828 * Does not block later packets from starting.
4830 void synchronize_net(void)
4837 * unregister_netdevice - remove device from the kernel
4840 * This function shuts down a device interface and removes it
4841 * from the kernel tables.
4843 * Callers must hold the rtnl semaphore. You may want
4844 * unregister_netdev() instead of this.
4847 void unregister_netdevice(struct net_device
*dev
)
4851 rollback_registered(dev
);
4852 /* Finish processing unregister after unlock */
4857 * unregister_netdev - remove device from the kernel
4860 * This function shuts down a device interface and removes it
4861 * from the kernel tables.
4863 * This is just a wrapper for unregister_netdevice that takes
4864 * the rtnl semaphore. In general you want to use this and not
4865 * unregister_netdevice.
4867 void unregister_netdev(struct net_device
*dev
)
4870 unregister_netdevice(dev
);
4874 EXPORT_SYMBOL(unregister_netdev
);
4877 * dev_change_net_namespace - move device to different nethost namespace
4879 * @net: network namespace
4880 * @pat: If not NULL name pattern to try if the current device name
4881 * is already taken in the destination network namespace.
4883 * This function shuts down a device interface and moves it
4884 * to a new network namespace. On success 0 is returned, on
4885 * a failure a netagive errno code is returned.
4887 * Callers must hold the rtnl semaphore.
4890 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
4893 const char *destname
;
4898 /* Don't allow namespace local devices to be moved. */
4900 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
4904 /* Don't allow real devices to be moved when sysfs
4908 if (dev
->dev
.parent
)
4912 /* Ensure the device has been registrered */
4914 if (dev
->reg_state
!= NETREG_REGISTERED
)
4917 /* Get out if there is nothing todo */
4919 if (net_eq(dev_net(dev
), net
))
4922 /* Pick the destination device name, and ensure
4923 * we can use it in the destination network namespace.
4926 destname
= dev
->name
;
4927 if (__dev_get_by_name(net
, destname
)) {
4928 /* We get here if we can't use the current device name */
4931 if (!dev_valid_name(pat
))
4933 if (strchr(pat
, '%')) {
4934 if (__dev_alloc_name(net
, pat
, buf
) < 0)
4939 if (__dev_get_by_name(net
, destname
))
4944 * And now a mini version of register_netdevice unregister_netdevice.
4947 /* If device is running close it first. */
4950 /* And unlink it from device chain */
4952 unlist_netdevice(dev
);
4956 /* Shutdown queueing discipline. */
4959 /* Notify protocols, that we are about to destroy
4960 this device. They should clean all the things.
4962 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4965 * Flush the unicast and multicast chains
4967 dev_addr_discard(dev
);
4969 netdev_unregister_kobject(dev
);
4971 /* Actually switch the network namespace */
4972 dev_net_set(dev
, net
);
4974 /* Assign the new device name */
4975 if (destname
!= dev
->name
)
4976 strcpy(dev
->name
, destname
);
4978 /* If there is an ifindex conflict assign a new one */
4979 if (__dev_get_by_index(net
, dev
->ifindex
)) {
4980 int iflink
= (dev
->iflink
== dev
->ifindex
);
4981 dev
->ifindex
= dev_new_index(net
);
4983 dev
->iflink
= dev
->ifindex
;
4986 /* Fixup kobjects */
4987 err
= netdev_register_kobject(dev
);
4990 /* Add the device back in the hashes */
4991 list_netdevice(dev
);
4993 /* Notify protocols, that a new device appeared. */
4994 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5002 static int dev_cpu_callback(struct notifier_block
*nfb
,
5003 unsigned long action
,
5006 struct sk_buff
**list_skb
;
5007 struct Qdisc
**list_net
;
5008 struct sk_buff
*skb
;
5009 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5010 struct softnet_data
*sd
, *oldsd
;
5012 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5015 local_irq_disable();
5016 cpu
= smp_processor_id();
5017 sd
= &per_cpu(softnet_data
, cpu
);
5018 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5020 /* Find end of our completion_queue. */
5021 list_skb
= &sd
->completion_queue
;
5023 list_skb
= &(*list_skb
)->next
;
5024 /* Append completion queue from offline CPU. */
5025 *list_skb
= oldsd
->completion_queue
;
5026 oldsd
->completion_queue
= NULL
;
5028 /* Find end of our output_queue. */
5029 list_net
= &sd
->output_queue
;
5031 list_net
= &(*list_net
)->next_sched
;
5032 /* Append output queue from offline CPU. */
5033 *list_net
= oldsd
->output_queue
;
5034 oldsd
->output_queue
= NULL
;
5036 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5039 /* Process offline CPU's input_pkt_queue */
5040 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
5048 * netdev_increment_features - increment feature set by one
5049 * @all: current feature set
5050 * @one: new feature set
5051 * @mask: mask feature set
5053 * Computes a new feature set after adding a device with feature set
5054 * @one to the master device with current feature set @all. Will not
5055 * enable anything that is off in @mask. Returns the new feature set.
5057 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
5060 /* If device needs checksumming, downgrade to it. */
5061 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
5062 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
5063 else if (mask
& NETIF_F_ALL_CSUM
) {
5064 /* If one device supports v4/v6 checksumming, set for all. */
5065 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
5066 !(all
& NETIF_F_GEN_CSUM
)) {
5067 all
&= ~NETIF_F_ALL_CSUM
;
5068 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
5071 /* If one device supports hw checksumming, set for all. */
5072 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
5073 all
&= ~NETIF_F_ALL_CSUM
;
5074 all
|= NETIF_F_HW_CSUM
;
5078 one
|= NETIF_F_ALL_CSUM
;
5080 one
|= all
& NETIF_F_ONE_FOR_ALL
;
5081 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
;
5082 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
5086 EXPORT_SYMBOL(netdev_increment_features
);
5088 static struct hlist_head
*netdev_create_hash(void)
5091 struct hlist_head
*hash
;
5093 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5095 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5096 INIT_HLIST_HEAD(&hash
[i
]);
5101 /* Initialize per network namespace state */
5102 static int __net_init
netdev_init(struct net
*net
)
5104 INIT_LIST_HEAD(&net
->dev_base_head
);
5106 net
->dev_name_head
= netdev_create_hash();
5107 if (net
->dev_name_head
== NULL
)
5110 net
->dev_index_head
= netdev_create_hash();
5111 if (net
->dev_index_head
== NULL
)
5117 kfree(net
->dev_name_head
);
5123 * netdev_drivername - network driver for the device
5124 * @dev: network device
5125 * @buffer: buffer for resulting name
5126 * @len: size of buffer
5128 * Determine network driver for device.
5130 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
5132 const struct device_driver
*driver
;
5133 const struct device
*parent
;
5135 if (len
<= 0 || !buffer
)
5139 parent
= dev
->dev
.parent
;
5144 driver
= parent
->driver
;
5145 if (driver
&& driver
->name
)
5146 strlcpy(buffer
, driver
->name
, len
);
5150 static void __net_exit
netdev_exit(struct net
*net
)
5152 kfree(net
->dev_name_head
);
5153 kfree(net
->dev_index_head
);
5156 static struct pernet_operations __net_initdata netdev_net_ops
= {
5157 .init
= netdev_init
,
5158 .exit
= netdev_exit
,
5161 static void __net_exit
default_device_exit(struct net
*net
)
5163 struct net_device
*dev
;
5165 * Push all migratable of the network devices back to the
5166 * initial network namespace
5170 for_each_netdev(net
, dev
) {
5172 char fb_name
[IFNAMSIZ
];
5174 /* Ignore unmoveable devices (i.e. loopback) */
5175 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5178 /* Delete virtual devices */
5179 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
) {
5180 dev
->rtnl_link_ops
->dellink(dev
);
5184 /* Push remaing network devices to init_net */
5185 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
5186 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
5188 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
5189 __func__
, dev
->name
, err
);
5197 static struct pernet_operations __net_initdata default_device_ops
= {
5198 .exit
= default_device_exit
,
5202 * Initialize the DEV module. At boot time this walks the device list and
5203 * unhooks any devices that fail to initialise (normally hardware not
5204 * present) and leaves us with a valid list of present and active devices.
5209 * This is called single threaded during boot, so no need
5210 * to take the rtnl semaphore.
5212 static int __init
net_dev_init(void)
5214 int i
, rc
= -ENOMEM
;
5216 BUG_ON(!dev_boot_phase
);
5218 if (dev_proc_init())
5221 if (netdev_kobject_init())
5224 INIT_LIST_HEAD(&ptype_all
);
5225 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
5226 INIT_LIST_HEAD(&ptype_base
[i
]);
5228 if (register_pernet_subsys(&netdev_net_ops
))
5232 * Initialise the packet receive queues.
5235 for_each_possible_cpu(i
) {
5236 struct softnet_data
*queue
;
5238 queue
= &per_cpu(softnet_data
, i
);
5239 skb_queue_head_init(&queue
->input_pkt_queue
);
5240 queue
->completion_queue
= NULL
;
5241 INIT_LIST_HEAD(&queue
->poll_list
);
5243 queue
->backlog
.poll
= process_backlog
;
5244 queue
->backlog
.weight
= weight_p
;
5245 queue
->backlog
.gro_list
= NULL
;
5250 /* The loopback device is special if any other network devices
5251 * is present in a network namespace the loopback device must
5252 * be present. Since we now dynamically allocate and free the
5253 * loopback device ensure this invariant is maintained by
5254 * keeping the loopback device as the first device on the
5255 * list of network devices. Ensuring the loopback devices
5256 * is the first device that appears and the last network device
5259 if (register_pernet_device(&loopback_net_ops
))
5262 if (register_pernet_device(&default_device_ops
))
5265 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
5266 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
5268 hotcpu_notifier(dev_cpu_callback
, 0);
5276 subsys_initcall(net_dev_init
);
5278 EXPORT_SYMBOL(__dev_get_by_index
);
5279 EXPORT_SYMBOL(__dev_get_by_name
);
5280 EXPORT_SYMBOL(__dev_remove_pack
);
5281 EXPORT_SYMBOL(dev_valid_name
);
5282 EXPORT_SYMBOL(dev_add_pack
);
5283 EXPORT_SYMBOL(dev_alloc_name
);
5284 EXPORT_SYMBOL(dev_close
);
5285 EXPORT_SYMBOL(dev_get_by_flags
);
5286 EXPORT_SYMBOL(dev_get_by_index
);
5287 EXPORT_SYMBOL(dev_get_by_name
);
5288 EXPORT_SYMBOL(dev_open
);
5289 EXPORT_SYMBOL(dev_queue_xmit
);
5290 EXPORT_SYMBOL(dev_remove_pack
);
5291 EXPORT_SYMBOL(dev_set_allmulti
);
5292 EXPORT_SYMBOL(dev_set_promiscuity
);
5293 EXPORT_SYMBOL(dev_change_flags
);
5294 EXPORT_SYMBOL(dev_set_mtu
);
5295 EXPORT_SYMBOL(dev_set_mac_address
);
5296 EXPORT_SYMBOL(free_netdev
);
5297 EXPORT_SYMBOL(netdev_boot_setup_check
);
5298 EXPORT_SYMBOL(netdev_set_master
);
5299 EXPORT_SYMBOL(netdev_state_change
);
5300 EXPORT_SYMBOL(netif_receive_skb
);
5301 EXPORT_SYMBOL(netif_rx
);
5302 EXPORT_SYMBOL(register_gifconf
);
5303 EXPORT_SYMBOL(register_netdevice
);
5304 EXPORT_SYMBOL(register_netdevice_notifier
);
5305 EXPORT_SYMBOL(skb_checksum_help
);
5306 EXPORT_SYMBOL(synchronize_net
);
5307 EXPORT_SYMBOL(unregister_netdevice
);
5308 EXPORT_SYMBOL(unregister_netdevice_notifier
);
5309 EXPORT_SYMBOL(net_enable_timestamp
);
5310 EXPORT_SYMBOL(net_disable_timestamp
);
5311 EXPORT_SYMBOL(dev_get_flags
);
5313 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
5314 EXPORT_SYMBOL(br_handle_frame_hook
);
5315 EXPORT_SYMBOL(br_fdb_get_hook
);
5316 EXPORT_SYMBOL(br_fdb_put_hook
);
5319 EXPORT_SYMBOL(dev_load
);
5321 EXPORT_PER_CPU_SYMBOL(softnet_data
);