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)
139 * The list of packet types we will receive (as opposed to discard)
140 * and the routines to invoke.
142 * Why 16. Because with 16 the only overlap we get on a hash of the
143 * low nibble of the protocol value is RARP/SNAP/X.25.
145 * NOTE: That is no longer true with the addition of VLAN tags. Not
146 * sure which should go first, but I bet it won't make much
147 * difference if we are running VLANs. The good news is that
148 * this protocol won't be in the list unless compiled in, so
149 * the average user (w/out VLANs) will not be adversely affected.
166 #define PTYPE_HASH_SIZE (16)
167 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
169 static DEFINE_SPINLOCK(ptype_lock
);
170 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
171 static struct list_head ptype_all __read_mostly
; /* Taps */
173 #ifdef CONFIG_NET_DMA
175 struct dma_client client
;
177 cpumask_t channel_mask
;
178 struct dma_chan
**channels
;
181 static enum dma_state_client
182 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
183 enum dma_state state
);
185 static struct net_dma net_dma
= {
187 .event_callback
= netdev_dma_event
,
193 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
196 * Pure readers hold dev_base_lock for reading.
198 * Writers must hold the rtnl semaphore while they loop through the
199 * dev_base_head list, and hold dev_base_lock for writing when they do the
200 * actual updates. This allows pure readers to access the list even
201 * while a writer is preparing to update it.
203 * To put it another way, dev_base_lock is held for writing only to
204 * protect against pure readers; the rtnl semaphore provides the
205 * protection against other writers.
207 * See, for example usages, register_netdevice() and
208 * unregister_netdevice(), which must be called with the rtnl
211 DEFINE_RWLOCK(dev_base_lock
);
213 EXPORT_SYMBOL(dev_base_lock
);
215 #define NETDEV_HASHBITS 8
216 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
218 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
220 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
221 return &net
->dev_name_head
[hash
& ((1 << NETDEV_HASHBITS
) - 1)];
224 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
226 return &net
->dev_index_head
[ifindex
& ((1 << NETDEV_HASHBITS
) - 1)];
229 /* Device list insertion */
230 static int list_netdevice(struct net_device
*dev
)
232 struct net
*net
= dev_net(dev
);
236 write_lock_bh(&dev_base_lock
);
237 list_add_tail(&dev
->dev_list
, &net
->dev_base_head
);
238 hlist_add_head(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
239 hlist_add_head(&dev
->index_hlist
, dev_index_hash(net
, dev
->ifindex
));
240 write_unlock_bh(&dev_base_lock
);
244 /* Device list removal */
245 static void unlist_netdevice(struct net_device
*dev
)
249 /* Unlink dev from the device chain */
250 write_lock_bh(&dev_base_lock
);
251 list_del(&dev
->dev_list
);
252 hlist_del(&dev
->name_hlist
);
253 hlist_del(&dev
->index_hlist
);
254 write_unlock_bh(&dev_base_lock
);
261 static RAW_NOTIFIER_HEAD(netdev_chain
);
264 * Device drivers call our routines to queue packets here. We empty the
265 * queue in the local softnet handler.
268 DEFINE_PER_CPU(struct softnet_data
, softnet_data
);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type
[] =
276 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
277 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
278 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
279 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
280 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
281 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
282 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
283 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
284 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
285 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
286 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
287 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
288 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
289 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
290 ARPHRD_PHONET_PIPE
, ARPHRD_VOID
, ARPHRD_NONE
};
292 static const char *netdev_lock_name
[] =
293 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
294 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
295 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
296 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
297 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
298 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
299 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
300 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
301 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
302 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
303 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
304 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
305 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
306 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
307 "_xmit_PHONET_PIPE", "_xmit_VOID", "_xmit_NONE"};
309 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
310 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
312 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
316 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
317 if (netdev_lock_type
[i
] == dev_type
)
319 /* the last key is used by default */
320 return ARRAY_SIZE(netdev_lock_type
) - 1;
323 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
324 unsigned short dev_type
)
328 i
= netdev_lock_pos(dev_type
);
329 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
330 netdev_lock_name
[i
]);
333 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
337 i
= netdev_lock_pos(dev
->type
);
338 lockdep_set_class_and_name(&dev
->addr_list_lock
,
339 &netdev_addr_lock_key
[i
],
340 netdev_lock_name
[i
]);
343 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
344 unsigned short dev_type
)
347 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
352 /*******************************************************************************
354 Protocol management and registration routines
356 *******************************************************************************/
359 * Add a protocol ID to the list. Now that the input handler is
360 * smarter we can dispense with all the messy stuff that used to be
363 * BEWARE!!! Protocol handlers, mangling input packets,
364 * MUST BE last in hash buckets and checking protocol handlers
365 * MUST start from promiscuous ptype_all chain in net_bh.
366 * It is true now, do not change it.
367 * Explanation follows: if protocol handler, mangling packet, will
368 * be the first on list, it is not able to sense, that packet
369 * is cloned and should be copied-on-write, so that it will
370 * change it and subsequent readers will get broken packet.
375 * dev_add_pack - add packet handler
376 * @pt: packet type declaration
378 * Add a protocol handler to the networking stack. The passed &packet_type
379 * is linked into kernel lists and may not be freed until it has been
380 * removed from the kernel lists.
382 * This call does not sleep therefore it can not
383 * guarantee all CPU's that are in middle of receiving packets
384 * will see the new packet type (until the next received packet).
387 void dev_add_pack(struct packet_type
*pt
)
391 spin_lock_bh(&ptype_lock
);
392 if (pt
->type
== htons(ETH_P_ALL
))
393 list_add_rcu(&pt
->list
, &ptype_all
);
395 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
396 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
398 spin_unlock_bh(&ptype_lock
);
402 * __dev_remove_pack - remove packet handler
403 * @pt: packet type declaration
405 * Remove a protocol handler that was previously added to the kernel
406 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
407 * from the kernel lists and can be freed or reused once this function
410 * The packet type might still be in use by receivers
411 * and must not be freed until after all the CPU's have gone
412 * through a quiescent state.
414 void __dev_remove_pack(struct packet_type
*pt
)
416 struct list_head
*head
;
417 struct packet_type
*pt1
;
419 spin_lock_bh(&ptype_lock
);
421 if (pt
->type
== htons(ETH_P_ALL
))
424 head
= &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
426 list_for_each_entry(pt1
, head
, list
) {
428 list_del_rcu(&pt
->list
);
433 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
435 spin_unlock_bh(&ptype_lock
);
438 * dev_remove_pack - remove packet handler
439 * @pt: packet type declaration
441 * Remove a protocol handler that was previously added to the kernel
442 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
443 * from the kernel lists and can be freed or reused once this function
446 * This call sleeps to guarantee that no CPU is looking at the packet
449 void dev_remove_pack(struct packet_type
*pt
)
451 __dev_remove_pack(pt
);
456 /******************************************************************************
458 Device Boot-time Settings Routines
460 *******************************************************************************/
462 /* Boot time configuration table */
463 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
466 * netdev_boot_setup_add - add new setup entry
467 * @name: name of the device
468 * @map: configured settings for the device
470 * Adds new setup entry to the dev_boot_setup list. The function
471 * returns 0 on error and 1 on success. This is a generic routine to
474 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
476 struct netdev_boot_setup
*s
;
480 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
481 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
482 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
483 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
484 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
489 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
493 * netdev_boot_setup_check - check boot time settings
494 * @dev: the netdevice
496 * Check boot time settings for the device.
497 * The found settings are set for the device to be used
498 * later in the device probing.
499 * Returns 0 if no settings found, 1 if they are.
501 int netdev_boot_setup_check(struct net_device
*dev
)
503 struct netdev_boot_setup
*s
= dev_boot_setup
;
506 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
507 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
508 !strcmp(dev
->name
, s
[i
].name
)) {
509 dev
->irq
= s
[i
].map
.irq
;
510 dev
->base_addr
= s
[i
].map
.base_addr
;
511 dev
->mem_start
= s
[i
].map
.mem_start
;
512 dev
->mem_end
= s
[i
].map
.mem_end
;
521 * netdev_boot_base - get address from boot time settings
522 * @prefix: prefix for network device
523 * @unit: id for network device
525 * Check boot time settings for the base address of device.
526 * The found settings are set for the device to be used
527 * later in the device probing.
528 * Returns 0 if no settings found.
530 unsigned long netdev_boot_base(const char *prefix
, int unit
)
532 const struct netdev_boot_setup
*s
= dev_boot_setup
;
536 sprintf(name
, "%s%d", prefix
, unit
);
539 * If device already registered then return base of 1
540 * to indicate not to probe for this interface
542 if (__dev_get_by_name(&init_net
, name
))
545 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
546 if (!strcmp(name
, s
[i
].name
))
547 return s
[i
].map
.base_addr
;
552 * Saves at boot time configured settings for any netdevice.
554 int __init
netdev_boot_setup(char *str
)
559 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
564 memset(&map
, 0, sizeof(map
));
568 map
.base_addr
= ints
[2];
570 map
.mem_start
= ints
[3];
572 map
.mem_end
= ints
[4];
574 /* Add new entry to the list */
575 return netdev_boot_setup_add(str
, &map
);
578 __setup("netdev=", netdev_boot_setup
);
580 /*******************************************************************************
582 Device Interface Subroutines
584 *******************************************************************************/
587 * __dev_get_by_name - find a device by its name
588 * @net: the applicable net namespace
589 * @name: name to find
591 * Find an interface by name. Must be called under RTNL semaphore
592 * or @dev_base_lock. If the name is found a pointer to the device
593 * is returned. If the name is not found then %NULL is returned. The
594 * reference counters are not incremented so the caller must be
595 * careful with locks.
598 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
600 struct hlist_node
*p
;
602 hlist_for_each(p
, dev_name_hash(net
, name
)) {
603 struct net_device
*dev
604 = hlist_entry(p
, struct net_device
, name_hlist
);
605 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
612 * dev_get_by_name - find a device by its name
613 * @net: the applicable net namespace
614 * @name: name to find
616 * Find an interface by name. This can be called from any
617 * context and does its own locking. The returned handle has
618 * the usage count incremented and the caller must use dev_put() to
619 * release it when it is no longer needed. %NULL is returned if no
620 * matching device is found.
623 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
625 struct net_device
*dev
;
627 read_lock(&dev_base_lock
);
628 dev
= __dev_get_by_name(net
, name
);
631 read_unlock(&dev_base_lock
);
636 * __dev_get_by_index - find a device by its ifindex
637 * @net: the applicable net namespace
638 * @ifindex: index of device
640 * Search for an interface by index. Returns %NULL if the device
641 * is not found or a pointer to the device. The device has not
642 * had its reference counter increased so the caller must be careful
643 * about locking. The caller must hold either the RTNL semaphore
647 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
649 struct hlist_node
*p
;
651 hlist_for_each(p
, dev_index_hash(net
, ifindex
)) {
652 struct net_device
*dev
653 = hlist_entry(p
, struct net_device
, index_hlist
);
654 if (dev
->ifindex
== ifindex
)
662 * dev_get_by_index - find a device by its ifindex
663 * @net: the applicable net namespace
664 * @ifindex: index of device
666 * Search for an interface by index. Returns NULL if the device
667 * is not found or a pointer to the device. The device returned has
668 * had a reference added and the pointer is safe until the user calls
669 * dev_put to indicate they have finished with it.
672 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
674 struct net_device
*dev
;
676 read_lock(&dev_base_lock
);
677 dev
= __dev_get_by_index(net
, ifindex
);
680 read_unlock(&dev_base_lock
);
685 * dev_getbyhwaddr - find a device by its hardware address
686 * @net: the applicable net namespace
687 * @type: media type of device
688 * @ha: hardware address
690 * Search for an interface by MAC address. Returns NULL if the device
691 * is not found or a pointer to the device. The caller must hold the
692 * rtnl semaphore. The returned device has not had its ref count increased
693 * and the caller must therefore be careful about locking
696 * If the API was consistent this would be __dev_get_by_hwaddr
699 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
701 struct net_device
*dev
;
705 for_each_netdev(net
, dev
)
706 if (dev
->type
== type
&&
707 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
713 EXPORT_SYMBOL(dev_getbyhwaddr
);
715 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
717 struct net_device
*dev
;
720 for_each_netdev(net
, dev
)
721 if (dev
->type
== type
)
727 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
729 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
731 struct net_device
*dev
;
734 dev
= __dev_getfirstbyhwtype(net
, type
);
741 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
744 * dev_get_by_flags - find any device with given flags
745 * @net: the applicable net namespace
746 * @if_flags: IFF_* values
747 * @mask: bitmask of bits in if_flags to check
749 * Search for any interface with the given flags. Returns NULL if a device
750 * is not found or a pointer to the device. The device returned has
751 * had a reference added and the pointer is safe until the user calls
752 * dev_put to indicate they have finished with it.
755 struct net_device
* dev_get_by_flags(struct net
*net
, unsigned short if_flags
, unsigned short mask
)
757 struct net_device
*dev
, *ret
;
760 read_lock(&dev_base_lock
);
761 for_each_netdev(net
, dev
) {
762 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
768 read_unlock(&dev_base_lock
);
773 * dev_valid_name - check if name is okay for network device
776 * Network device names need to be valid file names to
777 * to allow sysfs to work. We also disallow any kind of
780 int dev_valid_name(const char *name
)
784 if (strlen(name
) >= IFNAMSIZ
)
786 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
790 if (*name
== '/' || isspace(*name
))
798 * __dev_alloc_name - allocate a name for a device
799 * @net: network namespace to allocate the device name in
800 * @name: name format string
801 * @buf: scratch buffer and result name string
803 * Passed a format string - eg "lt%d" it will try and find a suitable
804 * id. It scans list of devices to build up a free map, then chooses
805 * the first empty slot. The caller must hold the dev_base or rtnl lock
806 * while allocating the name and adding the device in order to avoid
808 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
809 * Returns the number of the unit assigned or a negative errno code.
812 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
816 const int max_netdevices
= 8*PAGE_SIZE
;
817 unsigned long *inuse
;
818 struct net_device
*d
;
820 p
= strnchr(name
, IFNAMSIZ
-1, '%');
823 * Verify the string as this thing may have come from
824 * the user. There must be either one "%d" and no other "%"
827 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
830 /* Use one page as a bit array of possible slots */
831 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
835 for_each_netdev(net
, d
) {
836 if (!sscanf(d
->name
, name
, &i
))
838 if (i
< 0 || i
>= max_netdevices
)
841 /* avoid cases where sscanf is not exact inverse of printf */
842 snprintf(buf
, IFNAMSIZ
, name
, i
);
843 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
847 i
= find_first_zero_bit(inuse
, max_netdevices
);
848 free_page((unsigned long) inuse
);
851 snprintf(buf
, IFNAMSIZ
, name
, i
);
852 if (!__dev_get_by_name(net
, buf
))
855 /* It is possible to run out of possible slots
856 * when the name is long and there isn't enough space left
857 * for the digits, or if all bits are used.
863 * dev_alloc_name - allocate a name for a device
865 * @name: name format string
867 * Passed a format string - eg "lt%d" it will try and find a suitable
868 * id. It scans list of devices to build up a free map, then chooses
869 * the first empty slot. The caller must hold the dev_base or rtnl lock
870 * while allocating the name and adding the device in order to avoid
872 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
873 * Returns the number of the unit assigned or a negative errno code.
876 int dev_alloc_name(struct net_device
*dev
, const char *name
)
882 BUG_ON(!dev_net(dev
));
884 ret
= __dev_alloc_name(net
, name
, buf
);
886 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
892 * dev_change_name - change name of a device
894 * @newname: name (or format string) must be at least IFNAMSIZ
896 * Change name of a device, can pass format strings "eth%d".
899 int dev_change_name(struct net_device
*dev
, const char *newname
)
901 char oldname
[IFNAMSIZ
];
907 BUG_ON(!dev_net(dev
));
910 if (dev
->flags
& IFF_UP
)
913 if (!dev_valid_name(newname
))
916 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
919 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
921 if (strchr(newname
, '%')) {
922 err
= dev_alloc_name(dev
, newname
);
926 else if (__dev_get_by_name(net
, newname
))
929 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
932 /* For now only devices in the initial network namespace
935 if (net
== &init_net
) {
936 ret
= device_rename(&dev
->dev
, dev
->name
);
938 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
943 write_lock_bh(&dev_base_lock
);
944 hlist_del(&dev
->name_hlist
);
945 hlist_add_head(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
946 write_unlock_bh(&dev_base_lock
);
948 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
949 ret
= notifier_to_errno(ret
);
954 "%s: name change rollback failed: %d.\n",
958 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
967 * dev_set_alias - change ifalias of a device
969 * @alias: name up to IFALIASZ
970 * @len: limit of bytes to copy from info
972 * Set ifalias for a device,
974 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
989 dev
->ifalias
= krealloc(dev
->ifalias
, len
+1, GFP_KERNEL
);
993 strlcpy(dev
->ifalias
, alias
, len
+1);
999 * netdev_features_change - device changes features
1000 * @dev: device to cause notification
1002 * Called to indicate a device has changed features.
1004 void netdev_features_change(struct net_device
*dev
)
1006 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1008 EXPORT_SYMBOL(netdev_features_change
);
1011 * netdev_state_change - device changes state
1012 * @dev: device to cause notification
1014 * Called to indicate a device has changed state. This function calls
1015 * the notifier chains for netdev_chain and sends a NEWLINK message
1016 * to the routing socket.
1018 void netdev_state_change(struct net_device
*dev
)
1020 if (dev
->flags
& IFF_UP
) {
1021 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1022 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1026 void netdev_bonding_change(struct net_device
*dev
)
1028 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, dev
);
1030 EXPORT_SYMBOL(netdev_bonding_change
);
1033 * dev_load - load a network module
1034 * @net: the applicable net namespace
1035 * @name: name of interface
1037 * If a network interface is not present and the process has suitable
1038 * privileges this function loads the module. If module loading is not
1039 * available in this kernel then it becomes a nop.
1042 void dev_load(struct net
*net
, const char *name
)
1044 struct net_device
*dev
;
1046 read_lock(&dev_base_lock
);
1047 dev
= __dev_get_by_name(net
, name
);
1048 read_unlock(&dev_base_lock
);
1050 if (!dev
&& capable(CAP_SYS_MODULE
))
1051 request_module("%s", name
);
1055 * dev_open - prepare an interface for use.
1056 * @dev: device to open
1058 * Takes a device from down to up state. The device's private open
1059 * function is invoked and then the multicast lists are loaded. Finally
1060 * the device is moved into the up state and a %NETDEV_UP message is
1061 * sent to the netdev notifier chain.
1063 * Calling this function on an active interface is a nop. On a failure
1064 * a negative errno code is returned.
1066 int dev_open(struct net_device
*dev
)
1068 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1077 if (dev
->flags
& IFF_UP
)
1081 * Is it even present?
1083 if (!netif_device_present(dev
))
1087 * Call device private open method
1089 set_bit(__LINK_STATE_START
, &dev
->state
);
1091 if (ops
->ndo_validate_addr
)
1092 ret
= ops
->ndo_validate_addr(dev
);
1094 if (!ret
&& ops
->ndo_open
)
1095 ret
= ops
->ndo_open(dev
);
1098 * If it went open OK then:
1102 clear_bit(__LINK_STATE_START
, &dev
->state
);
1107 dev
->flags
|= IFF_UP
;
1110 * Initialize multicasting status
1112 dev_set_rx_mode(dev
);
1115 * Wakeup transmit queue engine
1120 * ... and announce new interface.
1122 call_netdevice_notifiers(NETDEV_UP
, dev
);
1129 * dev_close - shutdown an interface.
1130 * @dev: device to shutdown
1132 * This function moves an active device into down state. A
1133 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1134 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1137 int dev_close(struct net_device
*dev
)
1139 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1144 if (!(dev
->flags
& IFF_UP
))
1148 * Tell people we are going down, so that they can
1149 * prepare to death, when device is still operating.
1151 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1153 clear_bit(__LINK_STATE_START
, &dev
->state
);
1155 /* Synchronize to scheduled poll. We cannot touch poll list,
1156 * it can be even on different cpu. So just clear netif_running().
1158 * dev->stop() will invoke napi_disable() on all of it's
1159 * napi_struct instances on this device.
1161 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1163 dev_deactivate(dev
);
1166 * Call the device specific close. This cannot fail.
1167 * Only if device is UP
1169 * We allow it to be called even after a DETACH hot-plug
1176 * Device is now down.
1179 dev
->flags
&= ~IFF_UP
;
1182 * Tell people we are down
1184 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1191 * dev_disable_lro - disable Large Receive Offload on a device
1194 * Disable Large Receive Offload (LRO) on a net device. Must be
1195 * called under RTNL. This is needed if received packets may be
1196 * forwarded to another interface.
1198 void dev_disable_lro(struct net_device
*dev
)
1200 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1201 dev
->ethtool_ops
->set_flags
) {
1202 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1203 if (flags
& ETH_FLAG_LRO
) {
1204 flags
&= ~ETH_FLAG_LRO
;
1205 dev
->ethtool_ops
->set_flags(dev
, flags
);
1208 WARN_ON(dev
->features
& NETIF_F_LRO
);
1210 EXPORT_SYMBOL(dev_disable_lro
);
1213 static int dev_boot_phase
= 1;
1216 * Device change register/unregister. These are not inline or static
1217 * as we export them to the world.
1221 * register_netdevice_notifier - register a network notifier block
1224 * Register a notifier to be called when network device events occur.
1225 * The notifier passed is linked into the kernel structures and must
1226 * not be reused until it has been unregistered. A negative errno code
1227 * is returned on a failure.
1229 * When registered all registration and up events are replayed
1230 * to the new notifier to allow device to have a race free
1231 * view of the network device list.
1234 int register_netdevice_notifier(struct notifier_block
*nb
)
1236 struct net_device
*dev
;
1237 struct net_device
*last
;
1242 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1248 for_each_netdev(net
, dev
) {
1249 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1250 err
= notifier_to_errno(err
);
1254 if (!(dev
->flags
& IFF_UP
))
1257 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1268 for_each_netdev(net
, dev
) {
1272 if (dev
->flags
& IFF_UP
) {
1273 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1274 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1276 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1280 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1285 * unregister_netdevice_notifier - unregister a network notifier block
1288 * Unregister a notifier previously registered by
1289 * register_netdevice_notifier(). The notifier is unlinked into the
1290 * kernel structures and may then be reused. A negative errno code
1291 * is returned on a failure.
1294 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1299 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1305 * call_netdevice_notifiers - call all network notifier blocks
1306 * @val: value passed unmodified to notifier function
1307 * @dev: net_device pointer passed unmodified to notifier function
1309 * Call all network notifier blocks. Parameters and return value
1310 * are as for raw_notifier_call_chain().
1313 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1315 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1318 /* When > 0 there are consumers of rx skb time stamps */
1319 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1321 void net_enable_timestamp(void)
1323 atomic_inc(&netstamp_needed
);
1326 void net_disable_timestamp(void)
1328 atomic_dec(&netstamp_needed
);
1331 static inline void net_timestamp(struct sk_buff
*skb
)
1333 if (atomic_read(&netstamp_needed
))
1334 __net_timestamp(skb
);
1336 skb
->tstamp
.tv64
= 0;
1340 * Support routine. Sends outgoing frames to any network
1341 * taps currently in use.
1344 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1346 struct packet_type
*ptype
;
1351 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1352 /* Never send packets back to the socket
1353 * they originated from - MvS (miquels@drinkel.ow.org)
1355 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1356 (ptype
->af_packet_priv
== NULL
||
1357 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1358 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1362 /* skb->nh should be correctly
1363 set by sender, so that the second statement is
1364 just protection against buggy protocols.
1366 skb_reset_mac_header(skb2
);
1368 if (skb_network_header(skb2
) < skb2
->data
||
1369 skb2
->network_header
> skb2
->tail
) {
1370 if (net_ratelimit())
1371 printk(KERN_CRIT
"protocol %04x is "
1373 skb2
->protocol
, dev
->name
);
1374 skb_reset_network_header(skb2
);
1377 skb2
->transport_header
= skb2
->network_header
;
1378 skb2
->pkt_type
= PACKET_OUTGOING
;
1379 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1386 static inline void __netif_reschedule(struct Qdisc
*q
)
1388 struct softnet_data
*sd
;
1389 unsigned long flags
;
1391 local_irq_save(flags
);
1392 sd
= &__get_cpu_var(softnet_data
);
1393 q
->next_sched
= sd
->output_queue
;
1394 sd
->output_queue
= q
;
1395 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1396 local_irq_restore(flags
);
1399 void __netif_schedule(struct Qdisc
*q
)
1401 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1402 __netif_reschedule(q
);
1404 EXPORT_SYMBOL(__netif_schedule
);
1406 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1408 if (atomic_dec_and_test(&skb
->users
)) {
1409 struct softnet_data
*sd
;
1410 unsigned long flags
;
1412 local_irq_save(flags
);
1413 sd
= &__get_cpu_var(softnet_data
);
1414 skb
->next
= sd
->completion_queue
;
1415 sd
->completion_queue
= skb
;
1416 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1417 local_irq_restore(flags
);
1420 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1422 void dev_kfree_skb_any(struct sk_buff
*skb
)
1424 if (in_irq() || irqs_disabled())
1425 dev_kfree_skb_irq(skb
);
1429 EXPORT_SYMBOL(dev_kfree_skb_any
);
1433 * netif_device_detach - mark device as removed
1434 * @dev: network device
1436 * Mark device as removed from system and therefore no longer available.
1438 void netif_device_detach(struct net_device
*dev
)
1440 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1441 netif_running(dev
)) {
1442 netif_stop_queue(dev
);
1445 EXPORT_SYMBOL(netif_device_detach
);
1448 * netif_device_attach - mark device as attached
1449 * @dev: network device
1451 * Mark device as attached from system and restart if needed.
1453 void netif_device_attach(struct net_device
*dev
)
1455 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1456 netif_running(dev
)) {
1457 netif_wake_queue(dev
);
1458 __netdev_watchdog_up(dev
);
1461 EXPORT_SYMBOL(netif_device_attach
);
1463 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1465 return ((features
& NETIF_F_GEN_CSUM
) ||
1466 ((features
& NETIF_F_IP_CSUM
) &&
1467 protocol
== htons(ETH_P_IP
)) ||
1468 ((features
& NETIF_F_IPV6_CSUM
) &&
1469 protocol
== htons(ETH_P_IPV6
)));
1472 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1474 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1477 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1478 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1479 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1480 veh
->h_vlan_encapsulated_proto
))
1488 * Invalidate hardware checksum when packet is to be mangled, and
1489 * complete checksum manually on outgoing path.
1491 int skb_checksum_help(struct sk_buff
*skb
)
1494 int ret
= 0, offset
;
1496 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1497 goto out_set_summed
;
1499 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1500 /* Let GSO fix up the checksum. */
1501 goto out_set_summed
;
1504 offset
= skb
->csum_start
- skb_headroom(skb
);
1505 BUG_ON(offset
>= skb_headlen(skb
));
1506 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1508 offset
+= skb
->csum_offset
;
1509 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1511 if (skb_cloned(skb
) &&
1512 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1513 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1518 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1520 skb
->ip_summed
= CHECKSUM_NONE
;
1526 * skb_gso_segment - Perform segmentation on skb.
1527 * @skb: buffer to segment
1528 * @features: features for the output path (see dev->features)
1530 * This function segments the given skb and returns a list of segments.
1532 * It may return NULL if the skb requires no segmentation. This is
1533 * only possible when GSO is used for verifying header integrity.
1535 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1537 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1538 struct packet_type
*ptype
;
1539 __be16 type
= skb
->protocol
;
1542 skb_reset_mac_header(skb
);
1543 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1544 __skb_pull(skb
, skb
->mac_len
);
1546 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1547 if (skb_header_cloned(skb
) &&
1548 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1549 return ERR_PTR(err
);
1553 list_for_each_entry_rcu(ptype
,
1554 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1555 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1556 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1557 err
= ptype
->gso_send_check(skb
);
1558 segs
= ERR_PTR(err
);
1559 if (err
|| skb_gso_ok(skb
, features
))
1561 __skb_push(skb
, (skb
->data
-
1562 skb_network_header(skb
)));
1564 segs
= ptype
->gso_segment(skb
, features
);
1570 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1575 EXPORT_SYMBOL(skb_gso_segment
);
1577 /* Take action when hardware reception checksum errors are detected. */
1579 void netdev_rx_csum_fault(struct net_device
*dev
)
1581 if (net_ratelimit()) {
1582 printk(KERN_ERR
"%s: hw csum failure.\n",
1583 dev
? dev
->name
: "<unknown>");
1587 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1590 /* Actually, we should eliminate this check as soon as we know, that:
1591 * 1. IOMMU is present and allows to map all the memory.
1592 * 2. No high memory really exists on this machine.
1595 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1597 #ifdef CONFIG_HIGHMEM
1600 if (dev
->features
& NETIF_F_HIGHDMA
)
1603 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1604 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1612 void (*destructor
)(struct sk_buff
*skb
);
1615 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1617 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1619 struct dev_gso_cb
*cb
;
1622 struct sk_buff
*nskb
= skb
->next
;
1624 skb
->next
= nskb
->next
;
1627 } while (skb
->next
);
1629 cb
= DEV_GSO_CB(skb
);
1631 cb
->destructor(skb
);
1635 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1636 * @skb: buffer to segment
1638 * This function segments the given skb and stores the list of segments
1641 static int dev_gso_segment(struct sk_buff
*skb
)
1643 struct net_device
*dev
= skb
->dev
;
1644 struct sk_buff
*segs
;
1645 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1648 segs
= skb_gso_segment(skb
, features
);
1650 /* Verifying header integrity only. */
1655 return PTR_ERR(segs
);
1658 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1659 skb
->destructor
= dev_gso_skb_destructor
;
1664 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1665 struct netdev_queue
*txq
)
1667 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1669 prefetch(&dev
->netdev_ops
->ndo_start_xmit
);
1670 if (likely(!skb
->next
)) {
1671 if (!list_empty(&ptype_all
))
1672 dev_queue_xmit_nit(skb
, dev
);
1674 if (netif_needs_gso(dev
, skb
)) {
1675 if (unlikely(dev_gso_segment(skb
)))
1681 return ops
->ndo_start_xmit(skb
, dev
);
1686 struct sk_buff
*nskb
= skb
->next
;
1689 skb
->next
= nskb
->next
;
1691 rc
= ops
->ndo_start_xmit(nskb
, dev
);
1693 nskb
->next
= skb
->next
;
1697 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
1698 return NETDEV_TX_BUSY
;
1699 } while (skb
->next
);
1701 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1708 static u32 simple_tx_hashrnd
;
1709 static int simple_tx_hashrnd_initialized
= 0;
1711 static u16
simple_tx_hash(struct net_device
*dev
, struct sk_buff
*skb
)
1713 u32 addr1
, addr2
, ports
;
1717 if (unlikely(!simple_tx_hashrnd_initialized
)) {
1718 get_random_bytes(&simple_tx_hashrnd
, 4);
1719 simple_tx_hashrnd_initialized
= 1;
1722 switch (skb
->protocol
) {
1723 case htons(ETH_P_IP
):
1724 if (!(ip_hdr(skb
)->frag_off
& htons(IP_MF
| IP_OFFSET
)))
1725 ip_proto
= ip_hdr(skb
)->protocol
;
1726 addr1
= ip_hdr(skb
)->saddr
;
1727 addr2
= ip_hdr(skb
)->daddr
;
1728 ihl
= ip_hdr(skb
)->ihl
;
1730 case htons(ETH_P_IPV6
):
1731 ip_proto
= ipv6_hdr(skb
)->nexthdr
;
1732 addr1
= ipv6_hdr(skb
)->saddr
.s6_addr32
[3];
1733 addr2
= ipv6_hdr(skb
)->daddr
.s6_addr32
[3];
1748 case IPPROTO_UDPLITE
:
1749 ports
= *((u32
*) (skb_network_header(skb
) + (ihl
* 4)));
1757 hash
= jhash_3words(addr1
, addr2
, ports
, simple_tx_hashrnd
);
1759 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
1762 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
1763 struct sk_buff
*skb
)
1765 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1766 u16 queue_index
= 0;
1768 if (ops
->ndo_select_queue
)
1769 queue_index
= ops
->ndo_select_queue(dev
, skb
);
1770 else if (dev
->real_num_tx_queues
> 1)
1771 queue_index
= simple_tx_hash(dev
, skb
);
1773 skb_set_queue_mapping(skb
, queue_index
);
1774 return netdev_get_tx_queue(dev
, queue_index
);
1778 * dev_queue_xmit - transmit a buffer
1779 * @skb: buffer to transmit
1781 * Queue a buffer for transmission to a network device. The caller must
1782 * have set the device and priority and built the buffer before calling
1783 * this function. The function can be called from an interrupt.
1785 * A negative errno code is returned on a failure. A success does not
1786 * guarantee the frame will be transmitted as it may be dropped due
1787 * to congestion or traffic shaping.
1789 * -----------------------------------------------------------------------------------
1790 * I notice this method can also return errors from the queue disciplines,
1791 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1794 * Regardless of the return value, the skb is consumed, so it is currently
1795 * difficult to retry a send to this method. (You can bump the ref count
1796 * before sending to hold a reference for retry if you are careful.)
1798 * When calling this method, interrupts MUST be enabled. This is because
1799 * the BH enable code must have IRQs enabled so that it will not deadlock.
1802 int dev_queue_xmit(struct sk_buff
*skb
)
1804 struct net_device
*dev
= skb
->dev
;
1805 struct netdev_queue
*txq
;
1809 /* GSO will handle the following emulations directly. */
1810 if (netif_needs_gso(dev
, skb
))
1813 if (skb_shinfo(skb
)->frag_list
&&
1814 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1815 __skb_linearize(skb
))
1818 /* Fragmented skb is linearized if device does not support SG,
1819 * or if at least one of fragments is in highmem and device
1820 * does not support DMA from it.
1822 if (skb_shinfo(skb
)->nr_frags
&&
1823 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1824 __skb_linearize(skb
))
1827 /* If packet is not checksummed and device does not support
1828 * checksumming for this protocol, complete checksumming here.
1830 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1831 skb_set_transport_header(skb
, skb
->csum_start
-
1833 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
1838 /* Disable soft irqs for various locks below. Also
1839 * stops preemption for RCU.
1843 txq
= dev_pick_tx(dev
, skb
);
1844 q
= rcu_dereference(txq
->qdisc
);
1846 #ifdef CONFIG_NET_CLS_ACT
1847 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1850 spinlock_t
*root_lock
= qdisc_lock(q
);
1852 spin_lock(root_lock
);
1854 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
1858 rc
= qdisc_enqueue_root(skb
, q
);
1861 spin_unlock(root_lock
);
1866 /* The device has no queue. Common case for software devices:
1867 loopback, all the sorts of tunnels...
1869 Really, it is unlikely that netif_tx_lock protection is necessary
1870 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1872 However, it is possible, that they rely on protection
1875 Check this and shot the lock. It is not prone from deadlocks.
1876 Either shot noqueue qdisc, it is even simpler 8)
1878 if (dev
->flags
& IFF_UP
) {
1879 int cpu
= smp_processor_id(); /* ok because BHs are off */
1881 if (txq
->xmit_lock_owner
!= cpu
) {
1883 HARD_TX_LOCK(dev
, txq
, cpu
);
1885 if (!netif_tx_queue_stopped(txq
)) {
1887 if (!dev_hard_start_xmit(skb
, dev
, txq
)) {
1888 HARD_TX_UNLOCK(dev
, txq
);
1892 HARD_TX_UNLOCK(dev
, txq
);
1893 if (net_ratelimit())
1894 printk(KERN_CRIT
"Virtual device %s asks to "
1895 "queue packet!\n", dev
->name
);
1897 /* Recursion is detected! It is possible,
1899 if (net_ratelimit())
1900 printk(KERN_CRIT
"Dead loop on virtual device "
1901 "%s, fix it urgently!\n", dev
->name
);
1906 rcu_read_unlock_bh();
1912 rcu_read_unlock_bh();
1917 /*=======================================================================
1919 =======================================================================*/
1921 int netdev_max_backlog __read_mostly
= 1000;
1922 int netdev_budget __read_mostly
= 300;
1923 int weight_p __read_mostly
= 64; /* old backlog weight */
1925 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1929 * netif_rx - post buffer to the network code
1930 * @skb: buffer to post
1932 * This function receives a packet from a device driver and queues it for
1933 * the upper (protocol) levels to process. It always succeeds. The buffer
1934 * may be dropped during processing for congestion control or by the
1938 * NET_RX_SUCCESS (no congestion)
1939 * NET_RX_DROP (packet was dropped)
1943 int netif_rx(struct sk_buff
*skb
)
1945 struct softnet_data
*queue
;
1946 unsigned long flags
;
1948 /* if netpoll wants it, pretend we never saw it */
1949 if (netpoll_rx(skb
))
1952 if (!skb
->tstamp
.tv64
)
1956 * The code is rearranged so that the path is the most
1957 * short when CPU is congested, but is still operating.
1959 local_irq_save(flags
);
1960 queue
= &__get_cpu_var(softnet_data
);
1962 __get_cpu_var(netdev_rx_stat
).total
++;
1963 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1964 if (queue
->input_pkt_queue
.qlen
) {
1966 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1967 local_irq_restore(flags
);
1968 return NET_RX_SUCCESS
;
1971 napi_schedule(&queue
->backlog
);
1975 __get_cpu_var(netdev_rx_stat
).dropped
++;
1976 local_irq_restore(flags
);
1982 int netif_rx_ni(struct sk_buff
*skb
)
1987 err
= netif_rx(skb
);
1988 if (local_softirq_pending())
1995 EXPORT_SYMBOL(netif_rx_ni
);
1997 static void net_tx_action(struct softirq_action
*h
)
1999 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2001 if (sd
->completion_queue
) {
2002 struct sk_buff
*clist
;
2004 local_irq_disable();
2005 clist
= sd
->completion_queue
;
2006 sd
->completion_queue
= NULL
;
2010 struct sk_buff
*skb
= clist
;
2011 clist
= clist
->next
;
2013 WARN_ON(atomic_read(&skb
->users
));
2018 if (sd
->output_queue
) {
2021 local_irq_disable();
2022 head
= sd
->output_queue
;
2023 sd
->output_queue
= NULL
;
2027 struct Qdisc
*q
= head
;
2028 spinlock_t
*root_lock
;
2030 head
= head
->next_sched
;
2032 root_lock
= qdisc_lock(q
);
2033 if (spin_trylock(root_lock
)) {
2034 smp_mb__before_clear_bit();
2035 clear_bit(__QDISC_STATE_SCHED
,
2038 spin_unlock(root_lock
);
2040 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2042 __netif_reschedule(q
);
2044 smp_mb__before_clear_bit();
2045 clear_bit(__QDISC_STATE_SCHED
,
2053 static inline int deliver_skb(struct sk_buff
*skb
,
2054 struct packet_type
*pt_prev
,
2055 struct net_device
*orig_dev
)
2057 atomic_inc(&skb
->users
);
2058 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2061 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2062 /* These hooks defined here for ATM */
2064 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
2065 unsigned char *addr
);
2066 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
2069 * If bridge module is loaded call bridging hook.
2070 * returns NULL if packet was consumed.
2072 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
2073 struct sk_buff
*skb
) __read_mostly
;
2074 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
2075 struct packet_type
**pt_prev
, int *ret
,
2076 struct net_device
*orig_dev
)
2078 struct net_bridge_port
*port
;
2080 if (skb
->pkt_type
== PACKET_LOOPBACK
||
2081 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
2085 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2089 return br_handle_frame_hook(port
, skb
);
2092 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2095 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2096 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
2097 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
2099 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
2100 struct packet_type
**pt_prev
,
2102 struct net_device
*orig_dev
)
2104 if (skb
->dev
->macvlan_port
== NULL
)
2108 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2111 return macvlan_handle_frame_hook(skb
);
2114 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2117 #ifdef CONFIG_NET_CLS_ACT
2118 /* TODO: Maybe we should just force sch_ingress to be compiled in
2119 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2120 * a compare and 2 stores extra right now if we dont have it on
2121 * but have CONFIG_NET_CLS_ACT
2122 * NOTE: This doesnt stop any functionality; if you dont have
2123 * the ingress scheduler, you just cant add policies on ingress.
2126 static int ing_filter(struct sk_buff
*skb
)
2128 struct net_device
*dev
= skb
->dev
;
2129 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2130 struct netdev_queue
*rxq
;
2131 int result
= TC_ACT_OK
;
2134 if (MAX_RED_LOOP
< ttl
++) {
2136 "Redir loop detected Dropping packet (%d->%d)\n",
2137 skb
->iif
, dev
->ifindex
);
2141 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2142 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2144 rxq
= &dev
->rx_queue
;
2147 if (q
!= &noop_qdisc
) {
2148 spin_lock(qdisc_lock(q
));
2149 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2150 result
= qdisc_enqueue_root(skb
, q
);
2151 spin_unlock(qdisc_lock(q
));
2157 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2158 struct packet_type
**pt_prev
,
2159 int *ret
, struct net_device
*orig_dev
)
2161 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2165 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2168 /* Huh? Why does turning on AF_PACKET affect this? */
2169 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2172 switch (ing_filter(skb
)) {
2186 * netif_nit_deliver - deliver received packets to network taps
2189 * This function is used to deliver incoming packets to network
2190 * taps. It should be used when the normal netif_receive_skb path
2191 * is bypassed, for example because of VLAN acceleration.
2193 void netif_nit_deliver(struct sk_buff
*skb
)
2195 struct packet_type
*ptype
;
2197 if (list_empty(&ptype_all
))
2200 skb_reset_network_header(skb
);
2201 skb_reset_transport_header(skb
);
2202 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2205 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2206 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2207 deliver_skb(skb
, ptype
, skb
->dev
);
2213 * netif_receive_skb - process receive buffer from network
2214 * @skb: buffer to process
2216 * netif_receive_skb() is the main receive data processing function.
2217 * It always succeeds. The buffer may be dropped during processing
2218 * for congestion control or by the protocol layers.
2220 * This function may only be called from softirq context and interrupts
2221 * should be enabled.
2223 * Return values (usually ignored):
2224 * NET_RX_SUCCESS: no congestion
2225 * NET_RX_DROP: packet was dropped
2227 int netif_receive_skb(struct sk_buff
*skb
)
2229 struct packet_type
*ptype
, *pt_prev
;
2230 struct net_device
*orig_dev
;
2231 struct net_device
*null_or_orig
;
2232 int ret
= NET_RX_DROP
;
2235 if (skb
->vlan_tci
&& vlan_hwaccel_do_receive(skb
))
2236 return NET_RX_SUCCESS
;
2238 /* if we've gotten here through NAPI, check netpoll */
2239 if (netpoll_receive_skb(skb
))
2242 if (!skb
->tstamp
.tv64
)
2246 skb
->iif
= skb
->dev
->ifindex
;
2248 null_or_orig
= NULL
;
2249 orig_dev
= skb
->dev
;
2250 if (orig_dev
->master
) {
2251 if (skb_bond_should_drop(skb
))
2252 null_or_orig
= orig_dev
; /* deliver only exact match */
2254 skb
->dev
= orig_dev
->master
;
2257 __get_cpu_var(netdev_rx_stat
).total
++;
2259 skb_reset_network_header(skb
);
2260 skb_reset_transport_header(skb
);
2261 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2267 /* Don't receive packets in an exiting network namespace */
2268 if (!net_alive(dev_net(skb
->dev
))) {
2273 #ifdef CONFIG_NET_CLS_ACT
2274 if (skb
->tc_verd
& TC_NCLS
) {
2275 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2280 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2281 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2282 ptype
->dev
== orig_dev
) {
2284 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2289 #ifdef CONFIG_NET_CLS_ACT
2290 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2296 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
2299 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
2303 type
= skb
->protocol
;
2304 list_for_each_entry_rcu(ptype
,
2305 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2306 if (ptype
->type
== type
&&
2307 (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2308 ptype
->dev
== orig_dev
)) {
2310 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2316 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2319 /* Jamal, now you will not able to escape explaining
2320 * me how you were going to use this. :-)
2330 /* Network device is going away, flush any packets still pending */
2331 static void flush_backlog(void *arg
)
2333 struct net_device
*dev
= arg
;
2334 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2335 struct sk_buff
*skb
, *tmp
;
2337 skb_queue_walk_safe(&queue
->input_pkt_queue
, skb
, tmp
)
2338 if (skb
->dev
== dev
) {
2339 __skb_unlink(skb
, &queue
->input_pkt_queue
);
2344 static int napi_gro_complete(struct sk_buff
*skb
)
2346 struct packet_type
*ptype
;
2347 __be16 type
= skb
->protocol
;
2348 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2351 if (NAPI_GRO_CB(skb
)->count
== 1)
2355 list_for_each_entry_rcu(ptype
, head
, list
) {
2356 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
2359 err
= ptype
->gro_complete(skb
);
2365 WARN_ON(&ptype
->list
== head
);
2367 return NET_RX_SUCCESS
;
2371 skb_shinfo(skb
)->gso_size
= 0;
2372 __skb_push(skb
, -skb_network_offset(skb
));
2373 return netif_receive_skb(skb
);
2376 void napi_gro_flush(struct napi_struct
*napi
)
2378 struct sk_buff
*skb
, *next
;
2380 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
2383 napi_gro_complete(skb
);
2386 napi
->gro_list
= NULL
;
2388 EXPORT_SYMBOL(napi_gro_flush
);
2390 int dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2392 struct sk_buff
**pp
= NULL
;
2393 struct packet_type
*ptype
;
2394 __be16 type
= skb
->protocol
;
2395 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2401 if (!(skb
->dev
->features
& NETIF_F_GRO
))
2405 list_for_each_entry_rcu(ptype
, head
, list
) {
2408 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
2411 skb_reset_network_header(skb
);
2412 mac_len
= skb
->network_header
- skb
->mac_header
;
2413 skb
->mac_len
= mac_len
;
2414 NAPI_GRO_CB(skb
)->same_flow
= 0;
2415 NAPI_GRO_CB(skb
)->flush
= 0;
2416 NAPI_GRO_CB(skb
)->free
= 0;
2418 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
2421 if (!NAPI_GRO_CB(p
)->same_flow
)
2424 if (p
->mac_len
!= mac_len
||
2425 memcmp(skb_mac_header(p
), skb_mac_header(skb
),
2427 NAPI_GRO_CB(p
)->same_flow
= 0;
2430 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
2435 if (&ptype
->list
== head
)
2438 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
2439 free
= NAPI_GRO_CB(skb
)->free
;
2442 struct sk_buff
*nskb
= *pp
;
2446 napi_gro_complete(nskb
);
2453 if (NAPI_GRO_CB(skb
)->flush
|| count
>= MAX_GRO_SKBS
) {
2454 __skb_push(skb
, -skb_network_offset(skb
));
2458 NAPI_GRO_CB(skb
)->count
= 1;
2459 skb_shinfo(skb
)->gso_size
= skb
->len
;
2460 skb
->next
= napi
->gro_list
;
2461 napi
->gro_list
= skb
;
2469 EXPORT_SYMBOL(dev_gro_receive
);
2471 static int __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2475 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
2476 NAPI_GRO_CB(p
)->same_flow
= 1;
2477 NAPI_GRO_CB(p
)->flush
= 0;
2480 return dev_gro_receive(napi
, skb
);
2483 int napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2485 switch (__napi_gro_receive(napi
, skb
)) {
2487 return netif_receive_skb(skb
);
2494 return NET_RX_SUCCESS
;
2496 EXPORT_SYMBOL(napi_gro_receive
);
2498 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
2500 skb_shinfo(skb
)->nr_frags
= 0;
2502 skb
->len
-= skb
->data_len
;
2503 skb
->truesize
-= skb
->data_len
;
2506 __skb_pull(skb
, skb_headlen(skb
));
2507 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
2511 EXPORT_SYMBOL(napi_reuse_skb
);
2513 struct sk_buff
*napi_fraginfo_skb(struct napi_struct
*napi
,
2514 struct napi_gro_fraginfo
*info
)
2516 struct net_device
*dev
= napi
->dev
;
2517 struct sk_buff
*skb
= napi
->skb
;
2522 skb
= netdev_alloc_skb(dev
, GRO_MAX_HEAD
+ NET_IP_ALIGN
);
2526 skb_reserve(skb
, NET_IP_ALIGN
);
2529 BUG_ON(info
->nr_frags
> MAX_SKB_FRAGS
);
2530 skb_shinfo(skb
)->nr_frags
= info
->nr_frags
;
2531 memcpy(skb_shinfo(skb
)->frags
, info
->frags
, sizeof(info
->frags
));
2533 skb
->data_len
= info
->len
;
2534 skb
->len
+= info
->len
;
2535 skb
->truesize
+= info
->len
;
2537 if (!pskb_may_pull(skb
, ETH_HLEN
)) {
2538 napi_reuse_skb(napi
, skb
);
2542 skb
->protocol
= eth_type_trans(skb
, dev
);
2544 skb
->ip_summed
= info
->ip_summed
;
2545 skb
->csum
= info
->csum
;
2550 EXPORT_SYMBOL(napi_fraginfo_skb
);
2552 int napi_gro_frags(struct napi_struct
*napi
, struct napi_gro_fraginfo
*info
)
2554 struct sk_buff
*skb
= napi_fraginfo_skb(napi
, info
);
2555 int err
= NET_RX_DROP
;
2560 err
= NET_RX_SUCCESS
;
2562 switch (__napi_gro_receive(napi
, skb
)) {
2564 return netif_receive_skb(skb
);
2570 napi_reuse_skb(napi
, skb
);
2575 EXPORT_SYMBOL(napi_gro_frags
);
2577 static int process_backlog(struct napi_struct
*napi
, int quota
)
2580 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2581 unsigned long start_time
= jiffies
;
2583 napi
->weight
= weight_p
;
2585 struct sk_buff
*skb
;
2587 local_irq_disable();
2588 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
2590 __napi_complete(napi
);
2596 napi_gro_receive(napi
, skb
);
2597 } while (++work
< quota
&& jiffies
== start_time
);
2599 napi_gro_flush(napi
);
2605 * __napi_schedule - schedule for receive
2606 * @n: entry to schedule
2608 * The entry's receive function will be scheduled to run
2610 void __napi_schedule(struct napi_struct
*n
)
2612 unsigned long flags
;
2614 local_irq_save(flags
);
2615 list_add_tail(&n
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
2616 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2617 local_irq_restore(flags
);
2619 EXPORT_SYMBOL(__napi_schedule
);
2621 void __napi_complete(struct napi_struct
*n
)
2623 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
2624 BUG_ON(n
->gro_list
);
2626 list_del(&n
->poll_list
);
2627 smp_mb__before_clear_bit();
2628 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
2630 EXPORT_SYMBOL(__napi_complete
);
2632 void napi_complete(struct napi_struct
*n
)
2634 unsigned long flags
;
2637 * don't let napi dequeue from the cpu poll list
2638 * just in case its running on a different cpu
2640 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
2644 local_irq_save(flags
);
2646 local_irq_restore(flags
);
2648 EXPORT_SYMBOL(napi_complete
);
2650 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
2651 int (*poll
)(struct napi_struct
*, int), int weight
)
2653 INIT_LIST_HEAD(&napi
->poll_list
);
2654 napi
->gro_list
= NULL
;
2657 napi
->weight
= weight
;
2658 list_add(&napi
->dev_list
, &dev
->napi_list
);
2660 #ifdef CONFIG_NETPOLL
2661 spin_lock_init(&napi
->poll_lock
);
2662 napi
->poll_owner
= -1;
2664 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
2666 EXPORT_SYMBOL(netif_napi_add
);
2668 void netif_napi_del(struct napi_struct
*napi
)
2670 struct sk_buff
*skb
, *next
;
2672 list_del_init(&napi
->dev_list
);
2675 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
2681 napi
->gro_list
= NULL
;
2683 EXPORT_SYMBOL(netif_napi_del
);
2686 static void net_rx_action(struct softirq_action
*h
)
2688 struct list_head
*list
= &__get_cpu_var(softnet_data
).poll_list
;
2689 unsigned long time_limit
= jiffies
+ 2;
2690 int budget
= netdev_budget
;
2693 local_irq_disable();
2695 while (!list_empty(list
)) {
2696 struct napi_struct
*n
;
2699 /* If softirq window is exhuasted then punt.
2700 * Allow this to run for 2 jiffies since which will allow
2701 * an average latency of 1.5/HZ.
2703 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
2708 /* Even though interrupts have been re-enabled, this
2709 * access is safe because interrupts can only add new
2710 * entries to the tail of this list, and only ->poll()
2711 * calls can remove this head entry from the list.
2713 n
= list_entry(list
->next
, struct napi_struct
, poll_list
);
2715 have
= netpoll_poll_lock(n
);
2719 /* This NAPI_STATE_SCHED test is for avoiding a race
2720 * with netpoll's poll_napi(). Only the entity which
2721 * obtains the lock and sees NAPI_STATE_SCHED set will
2722 * actually make the ->poll() call. Therefore we avoid
2723 * accidently calling ->poll() when NAPI is not scheduled.
2726 if (test_bit(NAPI_STATE_SCHED
, &n
->state
))
2727 work
= n
->poll(n
, weight
);
2729 WARN_ON_ONCE(work
> weight
);
2733 local_irq_disable();
2735 /* Drivers must not modify the NAPI state if they
2736 * consume the entire weight. In such cases this code
2737 * still "owns" the NAPI instance and therefore can
2738 * move the instance around on the list at-will.
2740 if (unlikely(work
== weight
)) {
2741 if (unlikely(napi_disable_pending(n
)))
2744 list_move_tail(&n
->poll_list
, list
);
2747 netpoll_poll_unlock(have
);
2752 #ifdef CONFIG_NET_DMA
2754 * There may not be any more sk_buffs coming right now, so push
2755 * any pending DMA copies to hardware
2757 if (!cpus_empty(net_dma
.channel_mask
)) {
2759 for_each_cpu_mask_nr(chan_idx
, net_dma
.channel_mask
) {
2760 struct dma_chan
*chan
= net_dma
.channels
[chan_idx
];
2762 dma_async_memcpy_issue_pending(chan
);
2770 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2771 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2775 static gifconf_func_t
* gifconf_list
[NPROTO
];
2778 * register_gifconf - register a SIOCGIF handler
2779 * @family: Address family
2780 * @gifconf: Function handler
2782 * Register protocol dependent address dumping routines. The handler
2783 * that is passed must not be freed or reused until it has been replaced
2784 * by another handler.
2786 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2788 if (family
>= NPROTO
)
2790 gifconf_list
[family
] = gifconf
;
2796 * Map an interface index to its name (SIOCGIFNAME)
2800 * We need this ioctl for efficient implementation of the
2801 * if_indextoname() function required by the IPv6 API. Without
2802 * it, we would have to search all the interfaces to find a
2806 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
2808 struct net_device
*dev
;
2812 * Fetch the caller's info block.
2815 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2818 read_lock(&dev_base_lock
);
2819 dev
= __dev_get_by_index(net
, ifr
.ifr_ifindex
);
2821 read_unlock(&dev_base_lock
);
2825 strcpy(ifr
.ifr_name
, dev
->name
);
2826 read_unlock(&dev_base_lock
);
2828 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2834 * Perform a SIOCGIFCONF call. This structure will change
2835 * size eventually, and there is nothing I can do about it.
2836 * Thus we will need a 'compatibility mode'.
2839 static int dev_ifconf(struct net
*net
, char __user
*arg
)
2842 struct net_device
*dev
;
2849 * Fetch the caller's info block.
2852 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2859 * Loop over the interfaces, and write an info block for each.
2863 for_each_netdev(net
, dev
) {
2864 for (i
= 0; i
< NPROTO
; i
++) {
2865 if (gifconf_list
[i
]) {
2868 done
= gifconf_list
[i
](dev
, NULL
, 0);
2870 done
= gifconf_list
[i
](dev
, pos
+ total
,
2880 * All done. Write the updated control block back to the caller.
2882 ifc
.ifc_len
= total
;
2885 * Both BSD and Solaris return 0 here, so we do too.
2887 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2890 #ifdef CONFIG_PROC_FS
2892 * This is invoked by the /proc filesystem handler to display a device
2895 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2896 __acquires(dev_base_lock
)
2898 struct net
*net
= seq_file_net(seq
);
2900 struct net_device
*dev
;
2902 read_lock(&dev_base_lock
);
2904 return SEQ_START_TOKEN
;
2907 for_each_netdev(net
, dev
)
2914 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2916 struct net
*net
= seq_file_net(seq
);
2918 return v
== SEQ_START_TOKEN
?
2919 first_net_device(net
) : next_net_device((struct net_device
*)v
);
2922 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2923 __releases(dev_base_lock
)
2925 read_unlock(&dev_base_lock
);
2928 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2930 const struct net_device_stats
*stats
= dev_get_stats(dev
);
2932 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2933 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2934 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2936 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2937 stats
->rx_fifo_errors
,
2938 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2939 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2940 stats
->rx_compressed
, stats
->multicast
,
2941 stats
->tx_bytes
, stats
->tx_packets
,
2942 stats
->tx_errors
, stats
->tx_dropped
,
2943 stats
->tx_fifo_errors
, stats
->collisions
,
2944 stats
->tx_carrier_errors
+
2945 stats
->tx_aborted_errors
+
2946 stats
->tx_window_errors
+
2947 stats
->tx_heartbeat_errors
,
2948 stats
->tx_compressed
);
2952 * Called from the PROCfs module. This now uses the new arbitrary sized
2953 * /proc/net interface to create /proc/net/dev
2955 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2957 if (v
== SEQ_START_TOKEN
)
2958 seq_puts(seq
, "Inter-| Receive "
2960 " face |bytes packets errs drop fifo frame "
2961 "compressed multicast|bytes packets errs "
2962 "drop fifo colls carrier compressed\n");
2964 dev_seq_printf_stats(seq
, v
);
2968 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2970 struct netif_rx_stats
*rc
= NULL
;
2972 while (*pos
< nr_cpu_ids
)
2973 if (cpu_online(*pos
)) {
2974 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2981 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2983 return softnet_get_online(pos
);
2986 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2989 return softnet_get_online(pos
);
2992 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2996 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2998 struct netif_rx_stats
*s
= v
;
3000 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3001 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
3002 0, 0, 0, 0, /* was fastroute */
3007 static const struct seq_operations dev_seq_ops
= {
3008 .start
= dev_seq_start
,
3009 .next
= dev_seq_next
,
3010 .stop
= dev_seq_stop
,
3011 .show
= dev_seq_show
,
3014 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3016 return seq_open_net(inode
, file
, &dev_seq_ops
,
3017 sizeof(struct seq_net_private
));
3020 static const struct file_operations dev_seq_fops
= {
3021 .owner
= THIS_MODULE
,
3022 .open
= dev_seq_open
,
3024 .llseek
= seq_lseek
,
3025 .release
= seq_release_net
,
3028 static const struct seq_operations softnet_seq_ops
= {
3029 .start
= softnet_seq_start
,
3030 .next
= softnet_seq_next
,
3031 .stop
= softnet_seq_stop
,
3032 .show
= softnet_seq_show
,
3035 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3037 return seq_open(file
, &softnet_seq_ops
);
3040 static const struct file_operations softnet_seq_fops
= {
3041 .owner
= THIS_MODULE
,
3042 .open
= softnet_seq_open
,
3044 .llseek
= seq_lseek
,
3045 .release
= seq_release
,
3048 static void *ptype_get_idx(loff_t pos
)
3050 struct packet_type
*pt
= NULL
;
3054 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3060 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3061 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3070 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3074 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3077 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3079 struct packet_type
*pt
;
3080 struct list_head
*nxt
;
3084 if (v
== SEQ_START_TOKEN
)
3085 return ptype_get_idx(0);
3088 nxt
= pt
->list
.next
;
3089 if (pt
->type
== htons(ETH_P_ALL
)) {
3090 if (nxt
!= &ptype_all
)
3093 nxt
= ptype_base
[0].next
;
3095 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3097 while (nxt
== &ptype_base
[hash
]) {
3098 if (++hash
>= PTYPE_HASH_SIZE
)
3100 nxt
= ptype_base
[hash
].next
;
3103 return list_entry(nxt
, struct packet_type
, list
);
3106 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3112 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3114 struct packet_type
*pt
= v
;
3116 if (v
== SEQ_START_TOKEN
)
3117 seq_puts(seq
, "Type Device Function\n");
3118 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3119 if (pt
->type
== htons(ETH_P_ALL
))
3120 seq_puts(seq
, "ALL ");
3122 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3124 seq_printf(seq
, " %-8s %pF\n",
3125 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
3131 static const struct seq_operations ptype_seq_ops
= {
3132 .start
= ptype_seq_start
,
3133 .next
= ptype_seq_next
,
3134 .stop
= ptype_seq_stop
,
3135 .show
= ptype_seq_show
,
3138 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
3140 return seq_open_net(inode
, file
, &ptype_seq_ops
,
3141 sizeof(struct seq_net_private
));
3144 static const struct file_operations ptype_seq_fops
= {
3145 .owner
= THIS_MODULE
,
3146 .open
= ptype_seq_open
,
3148 .llseek
= seq_lseek
,
3149 .release
= seq_release_net
,
3153 static int __net_init
dev_proc_net_init(struct net
*net
)
3157 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
3159 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
3161 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
3164 if (wext_proc_init(net
))
3170 proc_net_remove(net
, "ptype");
3172 proc_net_remove(net
, "softnet_stat");
3174 proc_net_remove(net
, "dev");
3178 static void __net_exit
dev_proc_net_exit(struct net
*net
)
3180 wext_proc_exit(net
);
3182 proc_net_remove(net
, "ptype");
3183 proc_net_remove(net
, "softnet_stat");
3184 proc_net_remove(net
, "dev");
3187 static struct pernet_operations __net_initdata dev_proc_ops
= {
3188 .init
= dev_proc_net_init
,
3189 .exit
= dev_proc_net_exit
,
3192 static int __init
dev_proc_init(void)
3194 return register_pernet_subsys(&dev_proc_ops
);
3197 #define dev_proc_init() 0
3198 #endif /* CONFIG_PROC_FS */
3202 * netdev_set_master - set up master/slave pair
3203 * @slave: slave device
3204 * @master: new master device
3206 * Changes the master device of the slave. Pass %NULL to break the
3207 * bonding. The caller must hold the RTNL semaphore. On a failure
3208 * a negative errno code is returned. On success the reference counts
3209 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3210 * function returns zero.
3212 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
3214 struct net_device
*old
= slave
->master
;
3224 slave
->master
= master
;
3232 slave
->flags
|= IFF_SLAVE
;
3234 slave
->flags
&= ~IFF_SLAVE
;
3236 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
3240 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
3242 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3244 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
3245 ops
->ndo_change_rx_flags(dev
, flags
);
3248 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
3250 unsigned short old_flags
= dev
->flags
;
3256 dev
->flags
|= IFF_PROMISC
;
3257 dev
->promiscuity
+= inc
;
3258 if (dev
->promiscuity
== 0) {
3261 * If inc causes overflow, untouch promisc and return error.
3264 dev
->flags
&= ~IFF_PROMISC
;
3266 dev
->promiscuity
-= inc
;
3267 printk(KERN_WARNING
"%s: promiscuity touches roof, "
3268 "set promiscuity failed, promiscuity feature "
3269 "of device might be broken.\n", dev
->name
);
3273 if (dev
->flags
!= old_flags
) {
3274 printk(KERN_INFO
"device %s %s promiscuous mode\n",
3275 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
3277 if (audit_enabled
) {
3278 current_uid_gid(&uid
, &gid
);
3279 audit_log(current
->audit_context
, GFP_ATOMIC
,
3280 AUDIT_ANOM_PROMISCUOUS
,
3281 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3282 dev
->name
, (dev
->flags
& IFF_PROMISC
),
3283 (old_flags
& IFF_PROMISC
),
3284 audit_get_loginuid(current
),
3286 audit_get_sessionid(current
));
3289 dev_change_rx_flags(dev
, IFF_PROMISC
);
3295 * dev_set_promiscuity - update promiscuity count on a device
3299 * Add or remove promiscuity from a device. While the count in the device
3300 * remains above zero the interface remains promiscuous. Once it hits zero
3301 * the device reverts back to normal filtering operation. A negative inc
3302 * value is used to drop promiscuity on the device.
3303 * Return 0 if successful or a negative errno code on error.
3305 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
3307 unsigned short old_flags
= dev
->flags
;
3310 err
= __dev_set_promiscuity(dev
, inc
);
3313 if (dev
->flags
!= old_flags
)
3314 dev_set_rx_mode(dev
);
3319 * dev_set_allmulti - update allmulti count on a device
3323 * Add or remove reception of all multicast frames to a device. While the
3324 * count in the device remains above zero the interface remains listening
3325 * to all interfaces. Once it hits zero the device reverts back to normal
3326 * filtering operation. A negative @inc value is used to drop the counter
3327 * when releasing a resource needing all multicasts.
3328 * Return 0 if successful or a negative errno code on error.
3331 int dev_set_allmulti(struct net_device
*dev
, int inc
)
3333 unsigned short old_flags
= dev
->flags
;
3337 dev
->flags
|= IFF_ALLMULTI
;
3338 dev
->allmulti
+= inc
;
3339 if (dev
->allmulti
== 0) {
3342 * If inc causes overflow, untouch allmulti and return error.
3345 dev
->flags
&= ~IFF_ALLMULTI
;
3347 dev
->allmulti
-= inc
;
3348 printk(KERN_WARNING
"%s: allmulti touches roof, "
3349 "set allmulti failed, allmulti feature of "
3350 "device might be broken.\n", dev
->name
);
3354 if (dev
->flags
^ old_flags
) {
3355 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
3356 dev_set_rx_mode(dev
);
3362 * Upload unicast and multicast address lists to device and
3363 * configure RX filtering. When the device doesn't support unicast
3364 * filtering it is put in promiscuous mode while unicast addresses
3367 void __dev_set_rx_mode(struct net_device
*dev
)
3369 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3371 /* dev_open will call this function so the list will stay sane. */
3372 if (!(dev
->flags
&IFF_UP
))
3375 if (!netif_device_present(dev
))
3378 if (ops
->ndo_set_rx_mode
)
3379 ops
->ndo_set_rx_mode(dev
);
3381 /* Unicast addresses changes may only happen under the rtnl,
3382 * therefore calling __dev_set_promiscuity here is safe.
3384 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
3385 __dev_set_promiscuity(dev
, 1);
3386 dev
->uc_promisc
= 1;
3387 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
3388 __dev_set_promiscuity(dev
, -1);
3389 dev
->uc_promisc
= 0;
3392 if (ops
->ndo_set_multicast_list
)
3393 ops
->ndo_set_multicast_list(dev
);
3397 void dev_set_rx_mode(struct net_device
*dev
)
3399 netif_addr_lock_bh(dev
);
3400 __dev_set_rx_mode(dev
);
3401 netif_addr_unlock_bh(dev
);
3404 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
3405 void *addr
, int alen
, int glbl
)
3407 struct dev_addr_list
*da
;
3409 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
3410 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
3411 alen
== da
->da_addrlen
) {
3413 int old_glbl
= da
->da_gusers
;
3430 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
3431 void *addr
, int alen
, int glbl
)
3433 struct dev_addr_list
*da
;
3435 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
3436 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
3437 da
->da_addrlen
== alen
) {
3439 int old_glbl
= da
->da_gusers
;
3449 da
= kzalloc(sizeof(*da
), GFP_ATOMIC
);
3452 memcpy(da
->da_addr
, addr
, alen
);
3453 da
->da_addrlen
= alen
;
3455 da
->da_gusers
= glbl
? 1 : 0;
3463 * dev_unicast_delete - Release secondary unicast address.
3465 * @addr: address to delete
3466 * @alen: length of @addr
3468 * Release reference to a secondary unicast address and remove it
3469 * from the device if the reference count drops to zero.
3471 * The caller must hold the rtnl_mutex.
3473 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
3479 netif_addr_lock_bh(dev
);
3480 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3482 __dev_set_rx_mode(dev
);
3483 netif_addr_unlock_bh(dev
);
3486 EXPORT_SYMBOL(dev_unicast_delete
);
3489 * dev_unicast_add - add a secondary unicast address
3491 * @addr: address to add
3492 * @alen: length of @addr
3494 * Add a secondary unicast address to the device or increase
3495 * the reference count if it already exists.
3497 * The caller must hold the rtnl_mutex.
3499 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
3505 netif_addr_lock_bh(dev
);
3506 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3508 __dev_set_rx_mode(dev
);
3509 netif_addr_unlock_bh(dev
);
3512 EXPORT_SYMBOL(dev_unicast_add
);
3514 int __dev_addr_sync(struct dev_addr_list
**to
, int *to_count
,
3515 struct dev_addr_list
**from
, int *from_count
)
3517 struct dev_addr_list
*da
, *next
;
3521 while (da
!= NULL
) {
3523 if (!da
->da_synced
) {
3524 err
= __dev_addr_add(to
, to_count
,
3525 da
->da_addr
, da
->da_addrlen
, 0);
3530 } else if (da
->da_users
== 1) {
3531 __dev_addr_delete(to
, to_count
,
3532 da
->da_addr
, da
->da_addrlen
, 0);
3533 __dev_addr_delete(from
, from_count
,
3534 da
->da_addr
, da
->da_addrlen
, 0);
3541 void __dev_addr_unsync(struct dev_addr_list
**to
, int *to_count
,
3542 struct dev_addr_list
**from
, int *from_count
)
3544 struct dev_addr_list
*da
, *next
;
3547 while (da
!= NULL
) {
3549 if (da
->da_synced
) {
3550 __dev_addr_delete(to
, to_count
,
3551 da
->da_addr
, da
->da_addrlen
, 0);
3553 __dev_addr_delete(from
, from_count
,
3554 da
->da_addr
, da
->da_addrlen
, 0);
3561 * dev_unicast_sync - Synchronize device's unicast list to another device
3562 * @to: destination device
3563 * @from: source device
3565 * Add newly added addresses to the destination device and release
3566 * addresses that have no users left. The source device must be
3567 * locked by netif_tx_lock_bh.
3569 * This function is intended to be called from the dev->set_rx_mode
3570 * function of layered software devices.
3572 int dev_unicast_sync(struct net_device
*to
, struct net_device
*from
)
3576 netif_addr_lock_bh(to
);
3577 err
= __dev_addr_sync(&to
->uc_list
, &to
->uc_count
,
3578 &from
->uc_list
, &from
->uc_count
);
3580 __dev_set_rx_mode(to
);
3581 netif_addr_unlock_bh(to
);
3584 EXPORT_SYMBOL(dev_unicast_sync
);
3587 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3588 * @to: destination device
3589 * @from: source device
3591 * Remove all addresses that were added to the destination device by
3592 * dev_unicast_sync(). This function is intended to be called from the
3593 * dev->stop function of layered software devices.
3595 void dev_unicast_unsync(struct net_device
*to
, struct net_device
*from
)
3597 netif_addr_lock_bh(from
);
3598 netif_addr_lock(to
);
3600 __dev_addr_unsync(&to
->uc_list
, &to
->uc_count
,
3601 &from
->uc_list
, &from
->uc_count
);
3602 __dev_set_rx_mode(to
);
3604 netif_addr_unlock(to
);
3605 netif_addr_unlock_bh(from
);
3607 EXPORT_SYMBOL(dev_unicast_unsync
);
3609 static void __dev_addr_discard(struct dev_addr_list
**list
)
3611 struct dev_addr_list
*tmp
;
3613 while (*list
!= NULL
) {
3616 if (tmp
->da_users
> tmp
->da_gusers
)
3617 printk("__dev_addr_discard: address leakage! "
3618 "da_users=%d\n", tmp
->da_users
);
3623 static void dev_addr_discard(struct net_device
*dev
)
3625 netif_addr_lock_bh(dev
);
3627 __dev_addr_discard(&dev
->uc_list
);
3630 __dev_addr_discard(&dev
->mc_list
);
3633 netif_addr_unlock_bh(dev
);
3637 * dev_get_flags - get flags reported to userspace
3640 * Get the combination of flag bits exported through APIs to userspace.
3642 unsigned dev_get_flags(const struct net_device
*dev
)
3646 flags
= (dev
->flags
& ~(IFF_PROMISC
|
3651 (dev
->gflags
& (IFF_PROMISC
|
3654 if (netif_running(dev
)) {
3655 if (netif_oper_up(dev
))
3656 flags
|= IFF_RUNNING
;
3657 if (netif_carrier_ok(dev
))
3658 flags
|= IFF_LOWER_UP
;
3659 if (netif_dormant(dev
))
3660 flags
|= IFF_DORMANT
;
3667 * dev_change_flags - change device settings
3669 * @flags: device state flags
3671 * Change settings on device based state flags. The flags are
3672 * in the userspace exported format.
3674 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
3677 int old_flags
= dev
->flags
;
3682 * Set the flags on our device.
3685 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
3686 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
3688 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
3692 * Load in the correct multicast list now the flags have changed.
3695 if ((old_flags
^ flags
) & IFF_MULTICAST
)
3696 dev_change_rx_flags(dev
, IFF_MULTICAST
);
3698 dev_set_rx_mode(dev
);
3701 * Have we downed the interface. We handle IFF_UP ourselves
3702 * according to user attempts to set it, rather than blindly
3707 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
3708 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
3711 dev_set_rx_mode(dev
);
3714 if (dev
->flags
& IFF_UP
&&
3715 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
3717 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
3719 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
3720 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
3721 dev
->gflags
^= IFF_PROMISC
;
3722 dev_set_promiscuity(dev
, inc
);
3725 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3726 is important. Some (broken) drivers set IFF_PROMISC, when
3727 IFF_ALLMULTI is requested not asking us and not reporting.
3729 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
3730 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
3731 dev
->gflags
^= IFF_ALLMULTI
;
3732 dev_set_allmulti(dev
, inc
);
3735 /* Exclude state transition flags, already notified */
3736 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
3738 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
3744 * dev_set_mtu - Change maximum transfer unit
3746 * @new_mtu: new transfer unit
3748 * Change the maximum transfer size of the network device.
3750 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
3752 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3755 if (new_mtu
== dev
->mtu
)
3758 /* MTU must be positive. */
3762 if (!netif_device_present(dev
))
3766 if (ops
->ndo_change_mtu
)
3767 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
3771 if (!err
&& dev
->flags
& IFF_UP
)
3772 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
3777 * dev_set_mac_address - Change Media Access Control Address
3781 * Change the hardware (MAC) address of the device
3783 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
3785 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3788 if (!ops
->ndo_set_mac_address
)
3790 if (sa
->sa_family
!= dev
->type
)
3792 if (!netif_device_present(dev
))
3794 err
= ops
->ndo_set_mac_address(dev
, sa
);
3796 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3801 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3803 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3806 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3812 case SIOCGIFFLAGS
: /* Get interface flags */
3813 ifr
->ifr_flags
= dev_get_flags(dev
);
3816 case SIOCGIFMETRIC
: /* Get the metric on the interface
3817 (currently unused) */
3818 ifr
->ifr_metric
= 0;
3821 case SIOCGIFMTU
: /* Get the MTU of a device */
3822 ifr
->ifr_mtu
= dev
->mtu
;
3827 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
3829 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
3830 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3831 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
3839 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
3840 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
3841 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
3842 ifr
->ifr_map
.irq
= dev
->irq
;
3843 ifr
->ifr_map
.dma
= dev
->dma
;
3844 ifr
->ifr_map
.port
= dev
->if_port
;
3848 ifr
->ifr_ifindex
= dev
->ifindex
;
3852 ifr
->ifr_qlen
= dev
->tx_queue_len
;
3856 /* dev_ioctl() should ensure this case
3868 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3870 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3873 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3874 const struct net_device_ops
*ops
;
3879 ops
= dev
->netdev_ops
;
3882 case SIOCSIFFLAGS
: /* Set interface flags */
3883 return dev_change_flags(dev
, ifr
->ifr_flags
);
3885 case SIOCSIFMETRIC
: /* Set the metric on the interface
3886 (currently unused) */
3889 case SIOCSIFMTU
: /* Set the MTU of a device */
3890 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
3893 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
3895 case SIOCSIFHWBROADCAST
:
3896 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
3898 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
3899 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3900 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3904 if (ops
->ndo_set_config
) {
3905 if (!netif_device_present(dev
))
3907 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
3912 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
3913 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3915 if (!netif_device_present(dev
))
3917 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
3921 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
3922 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3924 if (!netif_device_present(dev
))
3926 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
3930 if (ifr
->ifr_qlen
< 0)
3932 dev
->tx_queue_len
= ifr
->ifr_qlen
;
3936 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
3937 return dev_change_name(dev
, ifr
->ifr_newname
);
3940 * Unknown or private ioctl
3944 if ((cmd
>= SIOCDEVPRIVATE
&&
3945 cmd
<= SIOCDEVPRIVATE
+ 15) ||
3946 cmd
== SIOCBONDENSLAVE
||
3947 cmd
== SIOCBONDRELEASE
||
3948 cmd
== SIOCBONDSETHWADDR
||
3949 cmd
== SIOCBONDSLAVEINFOQUERY
||
3950 cmd
== SIOCBONDINFOQUERY
||
3951 cmd
== SIOCBONDCHANGEACTIVE
||
3952 cmd
== SIOCGMIIPHY
||
3953 cmd
== SIOCGMIIREG
||
3954 cmd
== SIOCSMIIREG
||
3955 cmd
== SIOCBRADDIF
||
3956 cmd
== SIOCBRDELIF
||
3957 cmd
== SIOCWANDEV
) {
3959 if (ops
->ndo_do_ioctl
) {
3960 if (netif_device_present(dev
))
3961 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
3973 * This function handles all "interface"-type I/O control requests. The actual
3974 * 'doing' part of this is dev_ifsioc above.
3978 * dev_ioctl - network device ioctl
3979 * @net: the applicable net namespace
3980 * @cmd: command to issue
3981 * @arg: pointer to a struct ifreq in user space
3983 * Issue ioctl functions to devices. This is normally called by the
3984 * user space syscall interfaces but can sometimes be useful for
3985 * other purposes. The return value is the return from the syscall if
3986 * positive or a negative errno code on error.
3989 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
3995 /* One special case: SIOCGIFCONF takes ifconf argument
3996 and requires shared lock, because it sleeps writing
4000 if (cmd
== SIOCGIFCONF
) {
4002 ret
= dev_ifconf(net
, (char __user
*) arg
);
4006 if (cmd
== SIOCGIFNAME
)
4007 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4009 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4012 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4014 colon
= strchr(ifr
.ifr_name
, ':');
4019 * See which interface the caller is talking about.
4024 * These ioctl calls:
4025 * - can be done by all.
4026 * - atomic and do not require locking.
4037 dev_load(net
, ifr
.ifr_name
);
4038 read_lock(&dev_base_lock
);
4039 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4040 read_unlock(&dev_base_lock
);
4044 if (copy_to_user(arg
, &ifr
,
4045 sizeof(struct ifreq
)))
4051 dev_load(net
, ifr
.ifr_name
);
4053 ret
= dev_ethtool(net
, &ifr
);
4058 if (copy_to_user(arg
, &ifr
,
4059 sizeof(struct ifreq
)))
4065 * These ioctl calls:
4066 * - require superuser power.
4067 * - require strict serialization.
4073 if (!capable(CAP_NET_ADMIN
))
4075 dev_load(net
, ifr
.ifr_name
);
4077 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4082 if (copy_to_user(arg
, &ifr
,
4083 sizeof(struct ifreq
)))
4089 * These ioctl calls:
4090 * - require superuser power.
4091 * - require strict serialization.
4092 * - do not return a value
4102 case SIOCSIFHWBROADCAST
:
4105 case SIOCBONDENSLAVE
:
4106 case SIOCBONDRELEASE
:
4107 case SIOCBONDSETHWADDR
:
4108 case SIOCBONDCHANGEACTIVE
:
4111 if (!capable(CAP_NET_ADMIN
))
4114 case SIOCBONDSLAVEINFOQUERY
:
4115 case SIOCBONDINFOQUERY
:
4116 dev_load(net
, ifr
.ifr_name
);
4118 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4123 /* Get the per device memory space. We can add this but
4124 * currently do not support it */
4126 /* Set the per device memory buffer space.
4127 * Not applicable in our case */
4132 * Unknown or private ioctl.
4135 if (cmd
== SIOCWANDEV
||
4136 (cmd
>= SIOCDEVPRIVATE
&&
4137 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4138 dev_load(net
, ifr
.ifr_name
);
4140 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4142 if (!ret
&& copy_to_user(arg
, &ifr
,
4143 sizeof(struct ifreq
)))
4147 /* Take care of Wireless Extensions */
4148 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4149 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4156 * dev_new_index - allocate an ifindex
4157 * @net: the applicable net namespace
4159 * Returns a suitable unique value for a new device interface
4160 * number. The caller must hold the rtnl semaphore or the
4161 * dev_base_lock to be sure it remains unique.
4163 static int dev_new_index(struct net
*net
)
4169 if (!__dev_get_by_index(net
, ifindex
))
4174 /* Delayed registration/unregisteration */
4175 static LIST_HEAD(net_todo_list
);
4177 static void net_set_todo(struct net_device
*dev
)
4179 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4182 static void rollback_registered(struct net_device
*dev
)
4184 BUG_ON(dev_boot_phase
);
4187 /* Some devices call without registering for initialization unwind. */
4188 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4189 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
4190 "was registered\n", dev
->name
, dev
);
4196 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4198 /* If device is running, close it first. */
4201 /* And unlink it from device chain. */
4202 unlist_netdevice(dev
);
4204 dev
->reg_state
= NETREG_UNREGISTERING
;
4208 /* Shutdown queueing discipline. */
4212 /* Notify protocols, that we are about to destroy
4213 this device. They should clean all the things.
4215 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4218 * Flush the unicast and multicast chains
4220 dev_addr_discard(dev
);
4222 if (dev
->netdev_ops
->ndo_uninit
)
4223 dev
->netdev_ops
->ndo_uninit(dev
);
4225 /* Notifier chain MUST detach us from master device. */
4226 WARN_ON(dev
->master
);
4228 /* Remove entries from kobject tree */
4229 netdev_unregister_kobject(dev
);
4236 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
4237 struct netdev_queue
*dev_queue
,
4240 spin_lock_init(&dev_queue
->_xmit_lock
);
4241 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
4242 dev_queue
->xmit_lock_owner
= -1;
4245 static void netdev_init_queue_locks(struct net_device
*dev
)
4247 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
4248 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
4251 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
4253 /* Fix illegal SG+CSUM combinations. */
4254 if ((features
& NETIF_F_SG
) &&
4255 !(features
& NETIF_F_ALL_CSUM
)) {
4257 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
4258 "checksum feature.\n", name
);
4259 features
&= ~NETIF_F_SG
;
4262 /* TSO requires that SG is present as well. */
4263 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
4265 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
4266 "SG feature.\n", name
);
4267 features
&= ~NETIF_F_TSO
;
4270 if (features
& NETIF_F_UFO
) {
4271 if (!(features
& NETIF_F_GEN_CSUM
)) {
4273 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4274 "since no NETIF_F_HW_CSUM feature.\n",
4276 features
&= ~NETIF_F_UFO
;
4279 if (!(features
& NETIF_F_SG
)) {
4281 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4282 "since no NETIF_F_SG feature.\n", name
);
4283 features
&= ~NETIF_F_UFO
;
4289 EXPORT_SYMBOL(netdev_fix_features
);
4292 * register_netdevice - register a network device
4293 * @dev: device to register
4295 * Take a completed network device structure and add it to the kernel
4296 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4297 * chain. 0 is returned on success. A negative errno code is returned
4298 * on a failure to set up the device, or if the name is a duplicate.
4300 * Callers must hold the rtnl semaphore. You may want
4301 * register_netdev() instead of this.
4304 * The locking appears insufficient to guarantee two parallel registers
4305 * will not get the same name.
4308 int register_netdevice(struct net_device
*dev
)
4310 struct hlist_head
*head
;
4311 struct hlist_node
*p
;
4313 struct net
*net
= dev_net(dev
);
4315 BUG_ON(dev_boot_phase
);
4320 /* When net_device's are persistent, this will be fatal. */
4321 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
4324 spin_lock_init(&dev
->addr_list_lock
);
4325 netdev_set_addr_lockdep_class(dev
);
4326 netdev_init_queue_locks(dev
);
4330 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4331 /* Netdevice_ops API compatiability support.
4332 * This is temporary until all network devices are converted.
4334 if (dev
->netdev_ops
) {
4335 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4337 dev
->init
= ops
->ndo_init
;
4338 dev
->uninit
= ops
->ndo_uninit
;
4339 dev
->open
= ops
->ndo_open
;
4340 dev
->change_rx_flags
= ops
->ndo_change_rx_flags
;
4341 dev
->set_rx_mode
= ops
->ndo_set_rx_mode
;
4342 dev
->set_multicast_list
= ops
->ndo_set_multicast_list
;
4343 dev
->set_mac_address
= ops
->ndo_set_mac_address
;
4344 dev
->validate_addr
= ops
->ndo_validate_addr
;
4345 dev
->do_ioctl
= ops
->ndo_do_ioctl
;
4346 dev
->set_config
= ops
->ndo_set_config
;
4347 dev
->change_mtu
= ops
->ndo_change_mtu
;
4348 dev
->tx_timeout
= ops
->ndo_tx_timeout
;
4349 dev
->get_stats
= ops
->ndo_get_stats
;
4350 dev
->vlan_rx_register
= ops
->ndo_vlan_rx_register
;
4351 dev
->vlan_rx_add_vid
= ops
->ndo_vlan_rx_add_vid
;
4352 dev
->vlan_rx_kill_vid
= ops
->ndo_vlan_rx_kill_vid
;
4353 #ifdef CONFIG_NET_POLL_CONTROLLER
4354 dev
->poll_controller
= ops
->ndo_poll_controller
;
4357 char drivername
[64];
4358 pr_info("%s (%s): not using net_device_ops yet\n",
4359 dev
->name
, netdev_drivername(dev
, drivername
, 64));
4361 /* This works only because net_device_ops and the
4362 compatiablity structure are the same. */
4363 dev
->netdev_ops
= (void *) &(dev
->init
);
4367 /* Init, if this function is available */
4368 if (dev
->netdev_ops
->ndo_init
) {
4369 ret
= dev
->netdev_ops
->ndo_init(dev
);
4377 if (!dev_valid_name(dev
->name
)) {
4382 dev
->ifindex
= dev_new_index(net
);
4383 if (dev
->iflink
== -1)
4384 dev
->iflink
= dev
->ifindex
;
4386 /* Check for existence of name */
4387 head
= dev_name_hash(net
, dev
->name
);
4388 hlist_for_each(p
, head
) {
4389 struct net_device
*d
4390 = hlist_entry(p
, struct net_device
, name_hlist
);
4391 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
4397 /* Fix illegal checksum combinations */
4398 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
4399 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4400 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
4402 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
4405 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
4406 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4407 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
4409 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
4412 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
4414 /* Enable software GSO if SG is supported. */
4415 if (dev
->features
& NETIF_F_SG
)
4416 dev
->features
|= NETIF_F_GSO
;
4418 netdev_initialize_kobject(dev
);
4419 ret
= netdev_register_kobject(dev
);
4422 dev
->reg_state
= NETREG_REGISTERED
;
4425 * Default initial state at registry is that the
4426 * device is present.
4429 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
4431 dev_init_scheduler(dev
);
4433 list_netdevice(dev
);
4435 /* Notify protocols, that a new device appeared. */
4436 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
4437 ret
= notifier_to_errno(ret
);
4439 rollback_registered(dev
);
4440 dev
->reg_state
= NETREG_UNREGISTERED
;
4447 if (dev
->netdev_ops
->ndo_uninit
)
4448 dev
->netdev_ops
->ndo_uninit(dev
);
4453 * register_netdev - register a network device
4454 * @dev: device to register
4456 * Take a completed network device structure and add it to the kernel
4457 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4458 * chain. 0 is returned on success. A negative errno code is returned
4459 * on a failure to set up the device, or if the name is a duplicate.
4461 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4462 * and expands the device name if you passed a format string to
4465 int register_netdev(struct net_device
*dev
)
4472 * If the name is a format string the caller wants us to do a
4475 if (strchr(dev
->name
, '%')) {
4476 err
= dev_alloc_name(dev
, dev
->name
);
4481 err
= register_netdevice(dev
);
4486 EXPORT_SYMBOL(register_netdev
);
4489 * netdev_wait_allrefs - wait until all references are gone.
4491 * This is called when unregistering network devices.
4493 * Any protocol or device that holds a reference should register
4494 * for netdevice notification, and cleanup and put back the
4495 * reference if they receive an UNREGISTER event.
4496 * We can get stuck here if buggy protocols don't correctly
4499 static void netdev_wait_allrefs(struct net_device
*dev
)
4501 unsigned long rebroadcast_time
, warning_time
;
4503 rebroadcast_time
= warning_time
= jiffies
;
4504 while (atomic_read(&dev
->refcnt
) != 0) {
4505 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
4508 /* Rebroadcast unregister notification */
4509 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4511 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
4513 /* We must not have linkwatch events
4514 * pending on unregister. If this
4515 * happens, we simply run the queue
4516 * unscheduled, resulting in a noop
4519 linkwatch_run_queue();
4524 rebroadcast_time
= jiffies
;
4529 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
4530 printk(KERN_EMERG
"unregister_netdevice: "
4531 "waiting for %s to become free. Usage "
4533 dev
->name
, atomic_read(&dev
->refcnt
));
4534 warning_time
= jiffies
;
4543 * register_netdevice(x1);
4544 * register_netdevice(x2);
4546 * unregister_netdevice(y1);
4547 * unregister_netdevice(y2);
4553 * We are invoked by rtnl_unlock().
4554 * This allows us to deal with problems:
4555 * 1) We can delete sysfs objects which invoke hotplug
4556 * without deadlocking with linkwatch via keventd.
4557 * 2) Since we run with the RTNL semaphore not held, we can sleep
4558 * safely in order to wait for the netdev refcnt to drop to zero.
4560 * We must not return until all unregister events added during
4561 * the interval the lock was held have been completed.
4563 void netdev_run_todo(void)
4565 struct list_head list
;
4567 /* Snapshot list, allow later requests */
4568 list_replace_init(&net_todo_list
, &list
);
4572 while (!list_empty(&list
)) {
4573 struct net_device
*dev
4574 = list_entry(list
.next
, struct net_device
, todo_list
);
4575 list_del(&dev
->todo_list
);
4577 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
4578 printk(KERN_ERR
"network todo '%s' but state %d\n",
4579 dev
->name
, dev
->reg_state
);
4584 dev
->reg_state
= NETREG_UNREGISTERED
;
4586 on_each_cpu(flush_backlog
, dev
, 1);
4588 netdev_wait_allrefs(dev
);
4591 BUG_ON(atomic_read(&dev
->refcnt
));
4592 WARN_ON(dev
->ip_ptr
);
4593 WARN_ON(dev
->ip6_ptr
);
4594 WARN_ON(dev
->dn_ptr
);
4596 if (dev
->destructor
)
4597 dev
->destructor(dev
);
4599 /* Free network device */
4600 kobject_put(&dev
->dev
.kobj
);
4605 * dev_get_stats - get network device statistics
4606 * @dev: device to get statistics from
4608 * Get network statistics from device. The device driver may provide
4609 * its own method by setting dev->netdev_ops->get_stats; otherwise
4610 * the internal statistics structure is used.
4612 const struct net_device_stats
*dev_get_stats(struct net_device
*dev
)
4614 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4616 if (ops
->ndo_get_stats
)
4617 return ops
->ndo_get_stats(dev
);
4621 EXPORT_SYMBOL(dev_get_stats
);
4623 static void netdev_init_one_queue(struct net_device
*dev
,
4624 struct netdev_queue
*queue
,
4630 static void netdev_init_queues(struct net_device
*dev
)
4632 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
4633 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
4634 spin_lock_init(&dev
->tx_global_lock
);
4638 * alloc_netdev_mq - allocate network device
4639 * @sizeof_priv: size of private data to allocate space for
4640 * @name: device name format string
4641 * @setup: callback to initialize device
4642 * @queue_count: the number of subqueues to allocate
4644 * Allocates a struct net_device with private data area for driver use
4645 * and performs basic initialization. Also allocates subquue structs
4646 * for each queue on the device at the end of the netdevice.
4648 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
4649 void (*setup
)(struct net_device
*), unsigned int queue_count
)
4651 struct netdev_queue
*tx
;
4652 struct net_device
*dev
;
4656 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
4658 alloc_size
= sizeof(struct net_device
);
4660 /* ensure 32-byte alignment of private area */
4661 alloc_size
= (alloc_size
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
4662 alloc_size
+= sizeof_priv
;
4664 /* ensure 32-byte alignment of whole construct */
4665 alloc_size
+= NETDEV_ALIGN_CONST
;
4667 p
= kzalloc(alloc_size
, GFP_KERNEL
);
4669 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
4673 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
4675 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
4681 dev
= (struct net_device
*)
4682 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
4683 dev
->padded
= (char *)dev
- (char *)p
;
4684 dev_net_set(dev
, &init_net
);
4687 dev
->num_tx_queues
= queue_count
;
4688 dev
->real_num_tx_queues
= queue_count
;
4690 dev
->gso_max_size
= GSO_MAX_SIZE
;
4692 netdev_init_queues(dev
);
4694 INIT_LIST_HEAD(&dev
->napi_list
);
4696 strcpy(dev
->name
, name
);
4699 EXPORT_SYMBOL(alloc_netdev_mq
);
4702 * free_netdev - free network device
4705 * This function does the last stage of destroying an allocated device
4706 * interface. The reference to the device object is released.
4707 * If this is the last reference then it will be freed.
4709 void free_netdev(struct net_device
*dev
)
4711 struct napi_struct
*p
, *n
;
4713 release_net(dev_net(dev
));
4717 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
4720 /* Compatibility with error handling in drivers */
4721 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4722 kfree((char *)dev
- dev
->padded
);
4726 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
4727 dev
->reg_state
= NETREG_RELEASED
;
4729 /* will free via device release */
4730 put_device(&dev
->dev
);
4734 * synchronize_net - Synchronize with packet receive processing
4736 * Wait for packets currently being received to be done.
4737 * Does not block later packets from starting.
4739 void synchronize_net(void)
4746 * unregister_netdevice - remove device from the kernel
4749 * This function shuts down a device interface and removes it
4750 * from the kernel tables.
4752 * Callers must hold the rtnl semaphore. You may want
4753 * unregister_netdev() instead of this.
4756 void unregister_netdevice(struct net_device
*dev
)
4760 rollback_registered(dev
);
4761 /* Finish processing unregister after unlock */
4766 * unregister_netdev - remove device from the kernel
4769 * This function shuts down a device interface and removes it
4770 * from the kernel tables.
4772 * This is just a wrapper for unregister_netdevice that takes
4773 * the rtnl semaphore. In general you want to use this and not
4774 * unregister_netdevice.
4776 void unregister_netdev(struct net_device
*dev
)
4779 unregister_netdevice(dev
);
4783 EXPORT_SYMBOL(unregister_netdev
);
4786 * dev_change_net_namespace - move device to different nethost namespace
4788 * @net: network namespace
4789 * @pat: If not NULL name pattern to try if the current device name
4790 * is already taken in the destination network namespace.
4792 * This function shuts down a device interface and moves it
4793 * to a new network namespace. On success 0 is returned, on
4794 * a failure a netagive errno code is returned.
4796 * Callers must hold the rtnl semaphore.
4799 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
4802 const char *destname
;
4807 /* Don't allow namespace local devices to be moved. */
4809 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
4813 /* Don't allow real devices to be moved when sysfs
4817 if (dev
->dev
.parent
)
4821 /* Ensure the device has been registrered */
4823 if (dev
->reg_state
!= NETREG_REGISTERED
)
4826 /* Get out if there is nothing todo */
4828 if (net_eq(dev_net(dev
), net
))
4831 /* Pick the destination device name, and ensure
4832 * we can use it in the destination network namespace.
4835 destname
= dev
->name
;
4836 if (__dev_get_by_name(net
, destname
)) {
4837 /* We get here if we can't use the current device name */
4840 if (!dev_valid_name(pat
))
4842 if (strchr(pat
, '%')) {
4843 if (__dev_alloc_name(net
, pat
, buf
) < 0)
4848 if (__dev_get_by_name(net
, destname
))
4853 * And now a mini version of register_netdevice unregister_netdevice.
4856 /* If device is running close it first. */
4859 /* And unlink it from device chain */
4861 unlist_netdevice(dev
);
4865 /* Shutdown queueing discipline. */
4868 /* Notify protocols, that we are about to destroy
4869 this device. They should clean all the things.
4871 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4874 * Flush the unicast and multicast chains
4876 dev_addr_discard(dev
);
4878 netdev_unregister_kobject(dev
);
4880 /* Actually switch the network namespace */
4881 dev_net_set(dev
, net
);
4883 /* Assign the new device name */
4884 if (destname
!= dev
->name
)
4885 strcpy(dev
->name
, destname
);
4887 /* If there is an ifindex conflict assign a new one */
4888 if (__dev_get_by_index(net
, dev
->ifindex
)) {
4889 int iflink
= (dev
->iflink
== dev
->ifindex
);
4890 dev
->ifindex
= dev_new_index(net
);
4892 dev
->iflink
= dev
->ifindex
;
4895 /* Fixup kobjects */
4896 err
= netdev_register_kobject(dev
);
4899 /* Add the device back in the hashes */
4900 list_netdevice(dev
);
4902 /* Notify protocols, that a new device appeared. */
4903 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
4911 static int dev_cpu_callback(struct notifier_block
*nfb
,
4912 unsigned long action
,
4915 struct sk_buff
**list_skb
;
4916 struct Qdisc
**list_net
;
4917 struct sk_buff
*skb
;
4918 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
4919 struct softnet_data
*sd
, *oldsd
;
4921 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
4924 local_irq_disable();
4925 cpu
= smp_processor_id();
4926 sd
= &per_cpu(softnet_data
, cpu
);
4927 oldsd
= &per_cpu(softnet_data
, oldcpu
);
4929 /* Find end of our completion_queue. */
4930 list_skb
= &sd
->completion_queue
;
4932 list_skb
= &(*list_skb
)->next
;
4933 /* Append completion queue from offline CPU. */
4934 *list_skb
= oldsd
->completion_queue
;
4935 oldsd
->completion_queue
= NULL
;
4937 /* Find end of our output_queue. */
4938 list_net
= &sd
->output_queue
;
4940 list_net
= &(*list_net
)->next_sched
;
4941 /* Append output queue from offline CPU. */
4942 *list_net
= oldsd
->output_queue
;
4943 oldsd
->output_queue
= NULL
;
4945 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
4948 /* Process offline CPU's input_pkt_queue */
4949 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
4955 #ifdef CONFIG_NET_DMA
4957 * net_dma_rebalance - try to maintain one DMA channel per CPU
4958 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4960 * This is called when the number of channels allocated to the net_dma client
4961 * changes. The net_dma client tries to have one DMA channel per CPU.
4964 static void net_dma_rebalance(struct net_dma
*net_dma
)
4966 unsigned int cpu
, i
, n
, chan_idx
;
4967 struct dma_chan
*chan
;
4969 if (cpus_empty(net_dma
->channel_mask
)) {
4970 for_each_online_cpu(cpu
)
4971 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
4976 cpu
= first_cpu(cpu_online_map
);
4978 for_each_cpu_mask_nr(chan_idx
, net_dma
->channel_mask
) {
4979 chan
= net_dma
->channels
[chan_idx
];
4981 n
= ((num_online_cpus() / cpus_weight(net_dma
->channel_mask
))
4982 + (i
< (num_online_cpus() %
4983 cpus_weight(net_dma
->channel_mask
)) ? 1 : 0));
4986 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
4987 cpu
= next_cpu(cpu
, cpu_online_map
);
4995 * netdev_dma_event - event callback for the net_dma_client
4996 * @client: should always be net_dma_client
4997 * @chan: DMA channel for the event
4998 * @state: DMA state to be handled
5000 static enum dma_state_client
5001 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
5002 enum dma_state state
)
5004 int i
, found
= 0, pos
= -1;
5005 struct net_dma
*net_dma
=
5006 container_of(client
, struct net_dma
, client
);
5007 enum dma_state_client ack
= DMA_DUP
; /* default: take no action */
5009 spin_lock(&net_dma
->lock
);
5011 case DMA_RESOURCE_AVAILABLE
:
5012 for (i
= 0; i
< nr_cpu_ids
; i
++)
5013 if (net_dma
->channels
[i
] == chan
) {
5016 } else if (net_dma
->channels
[i
] == NULL
&& pos
< 0)
5019 if (!found
&& pos
>= 0) {
5021 net_dma
->channels
[pos
] = chan
;
5022 cpu_set(pos
, net_dma
->channel_mask
);
5023 net_dma_rebalance(net_dma
);
5026 case DMA_RESOURCE_REMOVED
:
5027 for (i
= 0; i
< nr_cpu_ids
; i
++)
5028 if (net_dma
->channels
[i
] == chan
) {
5036 cpu_clear(pos
, net_dma
->channel_mask
);
5037 net_dma
->channels
[i
] = NULL
;
5038 net_dma_rebalance(net_dma
);
5044 spin_unlock(&net_dma
->lock
);
5050 * netdev_dma_register - register the networking subsystem as a DMA client
5052 static int __init
netdev_dma_register(void)
5054 net_dma
.channels
= kzalloc(nr_cpu_ids
* sizeof(struct net_dma
),
5056 if (unlikely(!net_dma
.channels
)) {
5058 "netdev_dma: no memory for net_dma.channels\n");
5061 spin_lock_init(&net_dma
.lock
);
5062 dma_cap_set(DMA_MEMCPY
, net_dma
.client
.cap_mask
);
5063 dma_async_client_register(&net_dma
.client
);
5064 dma_async_client_chan_request(&net_dma
.client
);
5069 static int __init
netdev_dma_register(void) { return -ENODEV
; }
5070 #endif /* CONFIG_NET_DMA */
5073 * netdev_increment_features - increment feature set by one
5074 * @all: current feature set
5075 * @one: new feature set
5076 * @mask: mask feature set
5078 * Computes a new feature set after adding a device with feature set
5079 * @one to the master device with current feature set @all. Will not
5080 * enable anything that is off in @mask. Returns the new feature set.
5082 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
5085 /* If device needs checksumming, downgrade to it. */
5086 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
5087 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
5088 else if (mask
& NETIF_F_ALL_CSUM
) {
5089 /* If one device supports v4/v6 checksumming, set for all. */
5090 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
5091 !(all
& NETIF_F_GEN_CSUM
)) {
5092 all
&= ~NETIF_F_ALL_CSUM
;
5093 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
5096 /* If one device supports hw checksumming, set for all. */
5097 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
5098 all
&= ~NETIF_F_ALL_CSUM
;
5099 all
|= NETIF_F_HW_CSUM
;
5103 one
|= NETIF_F_ALL_CSUM
;
5105 one
|= all
& NETIF_F_ONE_FOR_ALL
;
5106 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
;
5107 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
5111 EXPORT_SYMBOL(netdev_increment_features
);
5113 static struct hlist_head
*netdev_create_hash(void)
5116 struct hlist_head
*hash
;
5118 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5120 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5121 INIT_HLIST_HEAD(&hash
[i
]);
5126 /* Initialize per network namespace state */
5127 static int __net_init
netdev_init(struct net
*net
)
5129 INIT_LIST_HEAD(&net
->dev_base_head
);
5131 net
->dev_name_head
= netdev_create_hash();
5132 if (net
->dev_name_head
== NULL
)
5135 net
->dev_index_head
= netdev_create_hash();
5136 if (net
->dev_index_head
== NULL
)
5142 kfree(net
->dev_name_head
);
5148 * netdev_drivername - network driver for the device
5149 * @dev: network device
5150 * @buffer: buffer for resulting name
5151 * @len: size of buffer
5153 * Determine network driver for device.
5155 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
5157 const struct device_driver
*driver
;
5158 const struct device
*parent
;
5160 if (len
<= 0 || !buffer
)
5164 parent
= dev
->dev
.parent
;
5169 driver
= parent
->driver
;
5170 if (driver
&& driver
->name
)
5171 strlcpy(buffer
, driver
->name
, len
);
5175 static void __net_exit
netdev_exit(struct net
*net
)
5177 kfree(net
->dev_name_head
);
5178 kfree(net
->dev_index_head
);
5181 static struct pernet_operations __net_initdata netdev_net_ops
= {
5182 .init
= netdev_init
,
5183 .exit
= netdev_exit
,
5186 static void __net_exit
default_device_exit(struct net
*net
)
5188 struct net_device
*dev
;
5190 * Push all migratable of the network devices back to the
5191 * initial network namespace
5195 for_each_netdev(net
, dev
) {
5197 char fb_name
[IFNAMSIZ
];
5199 /* Ignore unmoveable devices (i.e. loopback) */
5200 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5203 /* Delete virtual devices */
5204 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
) {
5205 dev
->rtnl_link_ops
->dellink(dev
);
5209 /* Push remaing network devices to init_net */
5210 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
5211 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
5213 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
5214 __func__
, dev
->name
, err
);
5222 static struct pernet_operations __net_initdata default_device_ops
= {
5223 .exit
= default_device_exit
,
5227 * Initialize the DEV module. At boot time this walks the device list and
5228 * unhooks any devices that fail to initialise (normally hardware not
5229 * present) and leaves us with a valid list of present and active devices.
5234 * This is called single threaded during boot, so no need
5235 * to take the rtnl semaphore.
5237 static int __init
net_dev_init(void)
5239 int i
, rc
= -ENOMEM
;
5241 BUG_ON(!dev_boot_phase
);
5243 if (dev_proc_init())
5246 if (netdev_kobject_init())
5249 INIT_LIST_HEAD(&ptype_all
);
5250 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
5251 INIT_LIST_HEAD(&ptype_base
[i
]);
5253 if (register_pernet_subsys(&netdev_net_ops
))
5257 * Initialise the packet receive queues.
5260 for_each_possible_cpu(i
) {
5261 struct softnet_data
*queue
;
5263 queue
= &per_cpu(softnet_data
, i
);
5264 skb_queue_head_init(&queue
->input_pkt_queue
);
5265 queue
->completion_queue
= NULL
;
5266 INIT_LIST_HEAD(&queue
->poll_list
);
5268 queue
->backlog
.poll
= process_backlog
;
5269 queue
->backlog
.weight
= weight_p
;
5270 queue
->backlog
.gro_list
= NULL
;
5275 /* The loopback device is special if any other network devices
5276 * is present in a network namespace the loopback device must
5277 * be present. Since we now dynamically allocate and free the
5278 * loopback device ensure this invariant is maintained by
5279 * keeping the loopback device as the first device on the
5280 * list of network devices. Ensuring the loopback devices
5281 * is the first device that appears and the last network device
5284 if (register_pernet_device(&loopback_net_ops
))
5287 if (register_pernet_device(&default_device_ops
))
5290 netdev_dma_register();
5292 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
5293 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
5295 hotcpu_notifier(dev_cpu_callback
, 0);
5303 subsys_initcall(net_dev_init
);
5305 EXPORT_SYMBOL(__dev_get_by_index
);
5306 EXPORT_SYMBOL(__dev_get_by_name
);
5307 EXPORT_SYMBOL(__dev_remove_pack
);
5308 EXPORT_SYMBOL(dev_valid_name
);
5309 EXPORT_SYMBOL(dev_add_pack
);
5310 EXPORT_SYMBOL(dev_alloc_name
);
5311 EXPORT_SYMBOL(dev_close
);
5312 EXPORT_SYMBOL(dev_get_by_flags
);
5313 EXPORT_SYMBOL(dev_get_by_index
);
5314 EXPORT_SYMBOL(dev_get_by_name
);
5315 EXPORT_SYMBOL(dev_open
);
5316 EXPORT_SYMBOL(dev_queue_xmit
);
5317 EXPORT_SYMBOL(dev_remove_pack
);
5318 EXPORT_SYMBOL(dev_set_allmulti
);
5319 EXPORT_SYMBOL(dev_set_promiscuity
);
5320 EXPORT_SYMBOL(dev_change_flags
);
5321 EXPORT_SYMBOL(dev_set_mtu
);
5322 EXPORT_SYMBOL(dev_set_mac_address
);
5323 EXPORT_SYMBOL(free_netdev
);
5324 EXPORT_SYMBOL(netdev_boot_setup_check
);
5325 EXPORT_SYMBOL(netdev_set_master
);
5326 EXPORT_SYMBOL(netdev_state_change
);
5327 EXPORT_SYMBOL(netif_receive_skb
);
5328 EXPORT_SYMBOL(netif_rx
);
5329 EXPORT_SYMBOL(register_gifconf
);
5330 EXPORT_SYMBOL(register_netdevice
);
5331 EXPORT_SYMBOL(register_netdevice_notifier
);
5332 EXPORT_SYMBOL(skb_checksum_help
);
5333 EXPORT_SYMBOL(synchronize_net
);
5334 EXPORT_SYMBOL(unregister_netdevice
);
5335 EXPORT_SYMBOL(unregister_netdevice_notifier
);
5336 EXPORT_SYMBOL(net_enable_timestamp
);
5337 EXPORT_SYMBOL(net_disable_timestamp
);
5338 EXPORT_SYMBOL(dev_get_flags
);
5340 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
5341 EXPORT_SYMBOL(br_handle_frame_hook
);
5342 EXPORT_SYMBOL(br_fdb_get_hook
);
5343 EXPORT_SYMBOL(br_fdb_put_hook
);
5346 EXPORT_SYMBOL(dev_load
);
5348 EXPORT_PER_CPU_SYMBOL(softnet_data
);