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/hash.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
103 #include <linux/if_bridge.h>
104 #include <linux/if_macvlan.h>
106 #include <net/pkt_sched.h>
107 #include <net/checksum.h>
108 #include <net/xfrm.h>
109 #include <linux/highmem.h>
110 #include <linux/init.h>
111 #include <linux/kmod.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/wext.h>
117 #include <net/iw_handler.h>
118 #include <asm/current.h>
119 #include <linux/audit.h>
120 #include <linux/dmaengine.h>
121 #include <linux/err.h>
122 #include <linux/ctype.h>
123 #include <linux/if_arp.h>
124 #include <linux/if_vlan.h>
125 #include <linux/ip.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
142 * The list of packet types we will receive (as opposed to discard)
143 * and the routines to invoke.
145 * Why 16. Because with 16 the only overlap we get on a hash of the
146 * low nibble of the protocol value is RARP/SNAP/X.25.
148 * NOTE: That is no longer true with the addition of VLAN tags. Not
149 * sure which should go first, but I bet it won't make much
150 * difference if we are running VLANs. The good news is that
151 * this protocol won't be in the list unless compiled in, so
152 * the average user (w/out VLANs) will not be adversely affected.
169 #define PTYPE_HASH_SIZE (16)
170 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
172 static DEFINE_SPINLOCK(ptype_lock
);
173 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
174 static struct list_head ptype_all __read_mostly
; /* Taps */
177 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
180 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
182 * Writers must hold the rtnl semaphore while they loop through the
183 * dev_base_head list, and hold dev_base_lock for writing when they do the
184 * actual updates. This allows pure readers to access the list even
185 * while a writer is preparing to update it.
187 * To put it another way, dev_base_lock is held for writing only to
188 * protect against pure readers; the rtnl semaphore provides the
189 * protection against other writers.
191 * See, for example usages, register_netdevice() and
192 * unregister_netdevice(), which must be called with the rtnl
195 DEFINE_RWLOCK(dev_base_lock
);
196 EXPORT_SYMBOL(dev_base_lock
);
198 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
200 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
201 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
204 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
206 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
209 static inline void rps_lock(struct softnet_data
*queue
)
212 spin_lock(&queue
->input_pkt_queue
.lock
);
216 static inline void rps_unlock(struct softnet_data
*queue
)
219 spin_unlock(&queue
->input_pkt_queue
.lock
);
223 /* Device list insertion */
224 static int list_netdevice(struct net_device
*dev
)
226 struct net
*net
= dev_net(dev
);
230 write_lock_bh(&dev_base_lock
);
231 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
232 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
233 hlist_add_head_rcu(&dev
->index_hlist
,
234 dev_index_hash(net
, dev
->ifindex
));
235 write_unlock_bh(&dev_base_lock
);
239 /* Device list removal
240 * caller must respect a RCU grace period before freeing/reusing dev
242 static void unlist_netdevice(struct net_device
*dev
)
246 /* Unlink dev from the device chain */
247 write_lock_bh(&dev_base_lock
);
248 list_del_rcu(&dev
->dev_list
);
249 hlist_del_rcu(&dev
->name_hlist
);
250 hlist_del_rcu(&dev
->index_hlist
);
251 write_unlock_bh(&dev_base_lock
);
258 static RAW_NOTIFIER_HEAD(netdev_chain
);
261 * Device drivers call our routines to queue packets here. We empty the
262 * queue in the local softnet handler.
265 DEFINE_PER_CPU(struct softnet_data
, softnet_data
);
266 EXPORT_PER_CPU_SYMBOL(softnet_data
);
268 #ifdef CONFIG_LOCKDEP
270 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
271 * according to dev->type
273 static const unsigned short netdev_lock_type
[] =
274 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
275 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
276 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
277 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
278 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
279 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
280 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
281 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
282 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
283 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
284 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
285 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
286 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
287 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
288 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
289 ARPHRD_VOID
, ARPHRD_NONE
};
291 static const char *const netdev_lock_name
[] =
292 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
293 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
294 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
295 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
296 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
297 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
298 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
299 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
300 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
301 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
302 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
303 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
304 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
305 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
306 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
307 "_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
);
400 EXPORT_SYMBOL(dev_add_pack
);
403 * __dev_remove_pack - remove packet handler
404 * @pt: packet type declaration
406 * Remove a protocol handler that was previously added to the kernel
407 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
408 * from the kernel lists and can be freed or reused once this function
411 * The packet type might still be in use by receivers
412 * and must not be freed until after all the CPU's have gone
413 * through a quiescent state.
415 void __dev_remove_pack(struct packet_type
*pt
)
417 struct list_head
*head
;
418 struct packet_type
*pt1
;
420 spin_lock_bh(&ptype_lock
);
422 if (pt
->type
== htons(ETH_P_ALL
))
425 head
= &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
427 list_for_each_entry(pt1
, head
, list
) {
429 list_del_rcu(&pt
->list
);
434 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
436 spin_unlock_bh(&ptype_lock
);
438 EXPORT_SYMBOL(__dev_remove_pack
);
441 * dev_remove_pack - remove packet handler
442 * @pt: packet type declaration
444 * Remove a protocol handler that was previously added to the kernel
445 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
446 * from the kernel lists and can be freed or reused once this function
449 * This call sleeps to guarantee that no CPU is looking at the packet
452 void dev_remove_pack(struct packet_type
*pt
)
454 __dev_remove_pack(pt
);
458 EXPORT_SYMBOL(dev_remove_pack
);
460 /******************************************************************************
462 Device Boot-time Settings Routines
464 *******************************************************************************/
466 /* Boot time configuration table */
467 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
470 * netdev_boot_setup_add - add new setup entry
471 * @name: name of the device
472 * @map: configured settings for the device
474 * Adds new setup entry to the dev_boot_setup list. The function
475 * returns 0 on error and 1 on success. This is a generic routine to
478 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
480 struct netdev_boot_setup
*s
;
484 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
485 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
486 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
487 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
488 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
493 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
497 * netdev_boot_setup_check - check boot time settings
498 * @dev: the netdevice
500 * Check boot time settings for the device.
501 * The found settings are set for the device to be used
502 * later in the device probing.
503 * Returns 0 if no settings found, 1 if they are.
505 int netdev_boot_setup_check(struct net_device
*dev
)
507 struct netdev_boot_setup
*s
= dev_boot_setup
;
510 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
511 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
512 !strcmp(dev
->name
, s
[i
].name
)) {
513 dev
->irq
= s
[i
].map
.irq
;
514 dev
->base_addr
= s
[i
].map
.base_addr
;
515 dev
->mem_start
= s
[i
].map
.mem_start
;
516 dev
->mem_end
= s
[i
].map
.mem_end
;
522 EXPORT_SYMBOL(netdev_boot_setup_check
);
526 * netdev_boot_base - get address from boot time settings
527 * @prefix: prefix for network device
528 * @unit: id for network device
530 * Check boot time settings for the base address of device.
531 * The found settings are set for the device to be used
532 * later in the device probing.
533 * Returns 0 if no settings found.
535 unsigned long netdev_boot_base(const char *prefix
, int unit
)
537 const struct netdev_boot_setup
*s
= dev_boot_setup
;
541 sprintf(name
, "%s%d", prefix
, unit
);
544 * If device already registered then return base of 1
545 * to indicate not to probe for this interface
547 if (__dev_get_by_name(&init_net
, name
))
550 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
551 if (!strcmp(name
, s
[i
].name
))
552 return s
[i
].map
.base_addr
;
557 * Saves at boot time configured settings for any netdevice.
559 int __init
netdev_boot_setup(char *str
)
564 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
569 memset(&map
, 0, sizeof(map
));
573 map
.base_addr
= ints
[2];
575 map
.mem_start
= ints
[3];
577 map
.mem_end
= ints
[4];
579 /* Add new entry to the list */
580 return netdev_boot_setup_add(str
, &map
);
583 __setup("netdev=", netdev_boot_setup
);
585 /*******************************************************************************
587 Device Interface Subroutines
589 *******************************************************************************/
592 * __dev_get_by_name - find a device by its name
593 * @net: the applicable net namespace
594 * @name: name to find
596 * Find an interface by name. Must be called under RTNL semaphore
597 * or @dev_base_lock. If the name is found a pointer to the device
598 * is returned. If the name is not found then %NULL is returned. The
599 * reference counters are not incremented so the caller must be
600 * careful with locks.
603 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
605 struct hlist_node
*p
;
606 struct net_device
*dev
;
607 struct hlist_head
*head
= dev_name_hash(net
, name
);
609 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
610 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
615 EXPORT_SYMBOL(__dev_get_by_name
);
618 * dev_get_by_name_rcu - find a device by its name
619 * @net: the applicable net namespace
620 * @name: name to find
622 * Find an interface by name.
623 * If the name is found a pointer to the device is returned.
624 * If the name is not found then %NULL is returned.
625 * The reference counters are not incremented so the caller must be
626 * careful with locks. The caller must hold RCU lock.
629 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
631 struct hlist_node
*p
;
632 struct net_device
*dev
;
633 struct hlist_head
*head
= dev_name_hash(net
, name
);
635 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
636 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
641 EXPORT_SYMBOL(dev_get_by_name_rcu
);
644 * dev_get_by_name - find a device by its name
645 * @net: the applicable net namespace
646 * @name: name to find
648 * Find an interface by name. This can be called from any
649 * context and does its own locking. The returned handle has
650 * the usage count incremented and the caller must use dev_put() to
651 * release it when it is no longer needed. %NULL is returned if no
652 * matching device is found.
655 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
657 struct net_device
*dev
;
660 dev
= dev_get_by_name_rcu(net
, name
);
666 EXPORT_SYMBOL(dev_get_by_name
);
669 * __dev_get_by_index - find a device by its ifindex
670 * @net: the applicable net namespace
671 * @ifindex: index of device
673 * Search for an interface by index. Returns %NULL if the device
674 * is not found or a pointer to the device. The device has not
675 * had its reference counter increased so the caller must be careful
676 * about locking. The caller must hold either the RTNL semaphore
680 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
682 struct hlist_node
*p
;
683 struct net_device
*dev
;
684 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
686 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
687 if (dev
->ifindex
== ifindex
)
692 EXPORT_SYMBOL(__dev_get_by_index
);
695 * dev_get_by_index_rcu - find a device by its ifindex
696 * @net: the applicable net namespace
697 * @ifindex: index of device
699 * Search for an interface by index. Returns %NULL if the device
700 * is not found or a pointer to the device. The device has not
701 * had its reference counter increased so the caller must be careful
702 * about locking. The caller must hold RCU lock.
705 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
707 struct hlist_node
*p
;
708 struct net_device
*dev
;
709 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
711 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
712 if (dev
->ifindex
== ifindex
)
717 EXPORT_SYMBOL(dev_get_by_index_rcu
);
721 * dev_get_by_index - find a device by its ifindex
722 * @net: the applicable net namespace
723 * @ifindex: index of device
725 * Search for an interface by index. Returns NULL if the device
726 * is not found or a pointer to the device. The device returned has
727 * had a reference added and the pointer is safe until the user calls
728 * dev_put to indicate they have finished with it.
731 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
733 struct net_device
*dev
;
736 dev
= dev_get_by_index_rcu(net
, ifindex
);
742 EXPORT_SYMBOL(dev_get_by_index
);
745 * dev_getbyhwaddr - find a device by its hardware address
746 * @net: the applicable net namespace
747 * @type: media type of device
748 * @ha: hardware address
750 * Search for an interface by MAC address. Returns NULL if the device
751 * is not found or a pointer to the device. The caller must hold the
752 * rtnl semaphore. The returned device has not had its ref count increased
753 * and the caller must therefore be careful about locking
756 * If the API was consistent this would be __dev_get_by_hwaddr
759 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
761 struct net_device
*dev
;
765 for_each_netdev(net
, dev
)
766 if (dev
->type
== type
&&
767 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
772 EXPORT_SYMBOL(dev_getbyhwaddr
);
774 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
776 struct net_device
*dev
;
779 for_each_netdev(net
, dev
)
780 if (dev
->type
== type
)
785 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
787 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
789 struct net_device
*dev
, *ret
= NULL
;
792 for_each_netdev_rcu(net
, dev
)
793 if (dev
->type
== type
) {
801 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
804 * dev_get_by_flags - find any device with given flags
805 * @net: the applicable net namespace
806 * @if_flags: IFF_* values
807 * @mask: bitmask of bits in if_flags to check
809 * Search for any interface with the given flags. Returns NULL if a device
810 * is not found or a pointer to the device. The device returned has
811 * had a reference added and the pointer is safe until the user calls
812 * dev_put to indicate they have finished with it.
815 struct net_device
*dev_get_by_flags(struct net
*net
, unsigned short if_flags
,
818 struct net_device
*dev
, *ret
;
822 for_each_netdev_rcu(net
, dev
) {
823 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
832 EXPORT_SYMBOL(dev_get_by_flags
);
835 * dev_valid_name - check if name is okay for network device
838 * Network device names need to be valid file names to
839 * to allow sysfs to work. We also disallow any kind of
842 int dev_valid_name(const char *name
)
846 if (strlen(name
) >= IFNAMSIZ
)
848 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
852 if (*name
== '/' || isspace(*name
))
858 EXPORT_SYMBOL(dev_valid_name
);
861 * __dev_alloc_name - allocate a name for a device
862 * @net: network namespace to allocate the device name in
863 * @name: name format string
864 * @buf: scratch buffer and result name string
866 * Passed a format string - eg "lt%d" it will try and find a suitable
867 * id. It scans list of devices to build up a free map, then chooses
868 * the first empty slot. The caller must hold the dev_base or rtnl lock
869 * while allocating the name and adding the device in order to avoid
871 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
872 * Returns the number of the unit assigned or a negative errno code.
875 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
879 const int max_netdevices
= 8*PAGE_SIZE
;
880 unsigned long *inuse
;
881 struct net_device
*d
;
883 p
= strnchr(name
, IFNAMSIZ
-1, '%');
886 * Verify the string as this thing may have come from
887 * the user. There must be either one "%d" and no other "%"
890 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
893 /* Use one page as a bit array of possible slots */
894 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
898 for_each_netdev(net
, d
) {
899 if (!sscanf(d
->name
, name
, &i
))
901 if (i
< 0 || i
>= max_netdevices
)
904 /* avoid cases where sscanf is not exact inverse of printf */
905 snprintf(buf
, IFNAMSIZ
, name
, i
);
906 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
910 i
= find_first_zero_bit(inuse
, max_netdevices
);
911 free_page((unsigned long) inuse
);
915 snprintf(buf
, IFNAMSIZ
, name
, i
);
916 if (!__dev_get_by_name(net
, buf
))
919 /* It is possible to run out of possible slots
920 * when the name is long and there isn't enough space left
921 * for the digits, or if all bits are used.
927 * dev_alloc_name - allocate a name for a device
929 * @name: name format string
931 * Passed a format string - eg "lt%d" it will try and find a suitable
932 * id. It scans list of devices to build up a free map, then chooses
933 * the first empty slot. The caller must hold the dev_base or rtnl lock
934 * while allocating the name and adding the device in order to avoid
936 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
937 * Returns the number of the unit assigned or a negative errno code.
940 int dev_alloc_name(struct net_device
*dev
, const char *name
)
946 BUG_ON(!dev_net(dev
));
948 ret
= __dev_alloc_name(net
, name
, buf
);
950 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
953 EXPORT_SYMBOL(dev_alloc_name
);
955 static int dev_get_valid_name(struct net
*net
, const char *name
, char *buf
,
958 if (!dev_valid_name(name
))
961 if (fmt
&& strchr(name
, '%'))
962 return __dev_alloc_name(net
, name
, buf
);
963 else if (__dev_get_by_name(net
, name
))
965 else if (buf
!= name
)
966 strlcpy(buf
, name
, IFNAMSIZ
);
972 * dev_change_name - change name of a device
974 * @newname: name (or format string) must be at least IFNAMSIZ
976 * Change name of a device, can pass format strings "eth%d".
979 int dev_change_name(struct net_device
*dev
, const char *newname
)
981 char oldname
[IFNAMSIZ
];
987 BUG_ON(!dev_net(dev
));
990 if (dev
->flags
& IFF_UP
)
993 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
996 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
998 err
= dev_get_valid_name(net
, newname
, dev
->name
, 1);
1003 /* For now only devices in the initial network namespace
1006 if (net_eq(net
, &init_net
)) {
1007 ret
= device_rename(&dev
->dev
, dev
->name
);
1009 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1014 write_lock_bh(&dev_base_lock
);
1015 hlist_del(&dev
->name_hlist
);
1016 write_unlock_bh(&dev_base_lock
);
1020 write_lock_bh(&dev_base_lock
);
1021 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1022 write_unlock_bh(&dev_base_lock
);
1024 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1025 ret
= notifier_to_errno(ret
);
1028 /* err >= 0 after dev_alloc_name() or stores the first errno */
1031 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1035 "%s: name change rollback failed: %d.\n",
1044 * dev_set_alias - change ifalias of a device
1046 * @alias: name up to IFALIASZ
1047 * @len: limit of bytes to copy from info
1049 * Set ifalias for a device,
1051 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1055 if (len
>= IFALIASZ
)
1060 kfree(dev
->ifalias
);
1061 dev
->ifalias
= NULL
;
1066 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1070 strlcpy(dev
->ifalias
, alias
, len
+1);
1076 * netdev_features_change - device changes features
1077 * @dev: device to cause notification
1079 * Called to indicate a device has changed features.
1081 void netdev_features_change(struct net_device
*dev
)
1083 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1085 EXPORT_SYMBOL(netdev_features_change
);
1088 * netdev_state_change - device changes state
1089 * @dev: device to cause notification
1091 * Called to indicate a device has changed state. This function calls
1092 * the notifier chains for netdev_chain and sends a NEWLINK message
1093 * to the routing socket.
1095 void netdev_state_change(struct net_device
*dev
)
1097 if (dev
->flags
& IFF_UP
) {
1098 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1099 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1102 EXPORT_SYMBOL(netdev_state_change
);
1104 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1106 return call_netdevice_notifiers(event
, dev
);
1108 EXPORT_SYMBOL(netdev_bonding_change
);
1111 * dev_load - load a network module
1112 * @net: the applicable net namespace
1113 * @name: name of interface
1115 * If a network interface is not present and the process has suitable
1116 * privileges this function loads the module. If module loading is not
1117 * available in this kernel then it becomes a nop.
1120 void dev_load(struct net
*net
, const char *name
)
1122 struct net_device
*dev
;
1125 dev
= dev_get_by_name_rcu(net
, name
);
1128 if (!dev
&& capable(CAP_NET_ADMIN
))
1129 request_module("%s", name
);
1131 EXPORT_SYMBOL(dev_load
);
1133 static int __dev_open(struct net_device
*dev
)
1135 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1141 * Is it even present?
1143 if (!netif_device_present(dev
))
1146 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1147 ret
= notifier_to_errno(ret
);
1152 * Call device private open method
1154 set_bit(__LINK_STATE_START
, &dev
->state
);
1156 if (ops
->ndo_validate_addr
)
1157 ret
= ops
->ndo_validate_addr(dev
);
1159 if (!ret
&& ops
->ndo_open
)
1160 ret
= ops
->ndo_open(dev
);
1163 * If it went open OK then:
1167 clear_bit(__LINK_STATE_START
, &dev
->state
);
1172 dev
->flags
|= IFF_UP
;
1177 net_dmaengine_get();
1180 * Initialize multicasting status
1182 dev_set_rx_mode(dev
);
1185 * Wakeup transmit queue engine
1194 * dev_open - prepare an interface for use.
1195 * @dev: device to open
1197 * Takes a device from down to up state. The device's private open
1198 * function is invoked and then the multicast lists are loaded. Finally
1199 * the device is moved into the up state and a %NETDEV_UP message is
1200 * sent to the netdev notifier chain.
1202 * Calling this function on an active interface is a nop. On a failure
1203 * a negative errno code is returned.
1205 int dev_open(struct net_device
*dev
)
1212 if (dev
->flags
& IFF_UP
)
1218 ret
= __dev_open(dev
);
1223 * ... and announce new interface.
1225 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1226 call_netdevice_notifiers(NETDEV_UP
, dev
);
1230 EXPORT_SYMBOL(dev_open
);
1232 static int __dev_close(struct net_device
*dev
)
1234 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1240 * Tell people we are going down, so that they can
1241 * prepare to death, when device is still operating.
1243 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1245 clear_bit(__LINK_STATE_START
, &dev
->state
);
1247 /* Synchronize to scheduled poll. We cannot touch poll list,
1248 * it can be even on different cpu. So just clear netif_running().
1250 * dev->stop() will invoke napi_disable() on all of it's
1251 * napi_struct instances on this device.
1253 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1255 dev_deactivate(dev
);
1258 * Call the device specific close. This cannot fail.
1259 * Only if device is UP
1261 * We allow it to be called even after a DETACH hot-plug
1268 * Device is now down.
1271 dev
->flags
&= ~IFF_UP
;
1276 net_dmaengine_put();
1282 * dev_close - shutdown an interface.
1283 * @dev: device to shutdown
1285 * This function moves an active device into down state. A
1286 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1287 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1290 int dev_close(struct net_device
*dev
)
1292 if (!(dev
->flags
& IFF_UP
))
1298 * Tell people we are down
1300 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1301 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1305 EXPORT_SYMBOL(dev_close
);
1309 * dev_disable_lro - disable Large Receive Offload on a device
1312 * Disable Large Receive Offload (LRO) on a net device. Must be
1313 * called under RTNL. This is needed if received packets may be
1314 * forwarded to another interface.
1316 void dev_disable_lro(struct net_device
*dev
)
1318 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1319 dev
->ethtool_ops
->set_flags
) {
1320 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1321 if (flags
& ETH_FLAG_LRO
) {
1322 flags
&= ~ETH_FLAG_LRO
;
1323 dev
->ethtool_ops
->set_flags(dev
, flags
);
1326 WARN_ON(dev
->features
& NETIF_F_LRO
);
1328 EXPORT_SYMBOL(dev_disable_lro
);
1331 static int dev_boot_phase
= 1;
1334 * Device change register/unregister. These are not inline or static
1335 * as we export them to the world.
1339 * register_netdevice_notifier - register a network notifier block
1342 * Register a notifier to be called when network device events occur.
1343 * The notifier passed is linked into the kernel structures and must
1344 * not be reused until it has been unregistered. A negative errno code
1345 * is returned on a failure.
1347 * When registered all registration and up events are replayed
1348 * to the new notifier to allow device to have a race free
1349 * view of the network device list.
1352 int register_netdevice_notifier(struct notifier_block
*nb
)
1354 struct net_device
*dev
;
1355 struct net_device
*last
;
1360 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1366 for_each_netdev(net
, dev
) {
1367 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1368 err
= notifier_to_errno(err
);
1372 if (!(dev
->flags
& IFF_UP
))
1375 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1386 for_each_netdev(net
, dev
) {
1390 if (dev
->flags
& IFF_UP
) {
1391 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1392 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1394 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1395 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1399 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1402 EXPORT_SYMBOL(register_netdevice_notifier
);
1405 * unregister_netdevice_notifier - unregister a network notifier block
1408 * Unregister a notifier previously registered by
1409 * register_netdevice_notifier(). The notifier is unlinked into the
1410 * kernel structures and may then be reused. A negative errno code
1411 * is returned on a failure.
1414 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1419 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1423 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1426 * call_netdevice_notifiers - call all network notifier blocks
1427 * @val: value passed unmodified to notifier function
1428 * @dev: net_device pointer passed unmodified to notifier function
1430 * Call all network notifier blocks. Parameters and return value
1431 * are as for raw_notifier_call_chain().
1434 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1436 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1439 /* When > 0 there are consumers of rx skb time stamps */
1440 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1442 void net_enable_timestamp(void)
1444 atomic_inc(&netstamp_needed
);
1446 EXPORT_SYMBOL(net_enable_timestamp
);
1448 void net_disable_timestamp(void)
1450 atomic_dec(&netstamp_needed
);
1452 EXPORT_SYMBOL(net_disable_timestamp
);
1454 static inline void net_timestamp(struct sk_buff
*skb
)
1456 if (atomic_read(&netstamp_needed
))
1457 __net_timestamp(skb
);
1459 skb
->tstamp
.tv64
= 0;
1463 * dev_forward_skb - loopback an skb to another netif
1465 * @dev: destination network device
1466 * @skb: buffer to forward
1469 * NET_RX_SUCCESS (no congestion)
1470 * NET_RX_DROP (packet was dropped)
1472 * dev_forward_skb can be used for injecting an skb from the
1473 * start_xmit function of one device into the receive queue
1474 * of another device.
1476 * The receiving device may be in another namespace, so
1477 * we have to clear all information in the skb that could
1478 * impact namespace isolation.
1480 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1484 if (!(dev
->flags
& IFF_UP
))
1487 if (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
))
1490 skb_set_dev(skb
, dev
);
1491 skb
->tstamp
.tv64
= 0;
1492 skb
->pkt_type
= PACKET_HOST
;
1493 skb
->protocol
= eth_type_trans(skb
, dev
);
1494 return netif_rx(skb
);
1496 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1499 * Support routine. Sends outgoing frames to any network
1500 * taps currently in use.
1503 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1505 struct packet_type
*ptype
;
1507 #ifdef CONFIG_NET_CLS_ACT
1508 if (!(skb
->tstamp
.tv64
&& (G_TC_FROM(skb
->tc_verd
) & AT_INGRESS
)))
1515 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1516 /* Never send packets back to the socket
1517 * they originated from - MvS (miquels@drinkel.ow.org)
1519 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1520 (ptype
->af_packet_priv
== NULL
||
1521 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1522 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1526 /* skb->nh should be correctly
1527 set by sender, so that the second statement is
1528 just protection against buggy protocols.
1530 skb_reset_mac_header(skb2
);
1532 if (skb_network_header(skb2
) < skb2
->data
||
1533 skb2
->network_header
> skb2
->tail
) {
1534 if (net_ratelimit())
1535 printk(KERN_CRIT
"protocol %04x is "
1537 skb2
->protocol
, dev
->name
);
1538 skb_reset_network_header(skb2
);
1541 skb2
->transport_header
= skb2
->network_header
;
1542 skb2
->pkt_type
= PACKET_OUTGOING
;
1543 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1550 static inline void __netif_reschedule(struct Qdisc
*q
)
1552 struct softnet_data
*sd
;
1553 unsigned long flags
;
1555 local_irq_save(flags
);
1556 sd
= &__get_cpu_var(softnet_data
);
1557 q
->next_sched
= sd
->output_queue
;
1558 sd
->output_queue
= q
;
1559 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1560 local_irq_restore(flags
);
1563 void __netif_schedule(struct Qdisc
*q
)
1565 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1566 __netif_reschedule(q
);
1568 EXPORT_SYMBOL(__netif_schedule
);
1570 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1572 if (atomic_dec_and_test(&skb
->users
)) {
1573 struct softnet_data
*sd
;
1574 unsigned long flags
;
1576 local_irq_save(flags
);
1577 sd
= &__get_cpu_var(softnet_data
);
1578 skb
->next
= sd
->completion_queue
;
1579 sd
->completion_queue
= skb
;
1580 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1581 local_irq_restore(flags
);
1584 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1586 void dev_kfree_skb_any(struct sk_buff
*skb
)
1588 if (in_irq() || irqs_disabled())
1589 dev_kfree_skb_irq(skb
);
1593 EXPORT_SYMBOL(dev_kfree_skb_any
);
1597 * netif_device_detach - mark device as removed
1598 * @dev: network device
1600 * Mark device as removed from system and therefore no longer available.
1602 void netif_device_detach(struct net_device
*dev
)
1604 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1605 netif_running(dev
)) {
1606 netif_tx_stop_all_queues(dev
);
1609 EXPORT_SYMBOL(netif_device_detach
);
1612 * netif_device_attach - mark device as attached
1613 * @dev: network device
1615 * Mark device as attached from system and restart if needed.
1617 void netif_device_attach(struct net_device
*dev
)
1619 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1620 netif_running(dev
)) {
1621 netif_tx_wake_all_queues(dev
);
1622 __netdev_watchdog_up(dev
);
1625 EXPORT_SYMBOL(netif_device_attach
);
1627 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1629 return ((features
& NETIF_F_GEN_CSUM
) ||
1630 ((features
& NETIF_F_IP_CSUM
) &&
1631 protocol
== htons(ETH_P_IP
)) ||
1632 ((features
& NETIF_F_IPV6_CSUM
) &&
1633 protocol
== htons(ETH_P_IPV6
)) ||
1634 ((features
& NETIF_F_FCOE_CRC
) &&
1635 protocol
== htons(ETH_P_FCOE
)));
1638 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1640 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1643 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1644 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1645 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1646 veh
->h_vlan_encapsulated_proto
))
1654 * skb_dev_set -- assign a new device to a buffer
1655 * @skb: buffer for the new device
1656 * @dev: network device
1658 * If an skb is owned by a device already, we have to reset
1659 * all data private to the namespace a device belongs to
1660 * before assigning it a new device.
1662 #ifdef CONFIG_NET_NS
1663 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1666 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1669 skb_init_secmark(skb
);
1673 skb
->ipvs_property
= 0;
1674 #ifdef CONFIG_NET_SCHED
1680 EXPORT_SYMBOL(skb_set_dev
);
1681 #endif /* CONFIG_NET_NS */
1684 * Invalidate hardware checksum when packet is to be mangled, and
1685 * complete checksum manually on outgoing path.
1687 int skb_checksum_help(struct sk_buff
*skb
)
1690 int ret
= 0, offset
;
1692 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1693 goto out_set_summed
;
1695 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1696 /* Let GSO fix up the checksum. */
1697 goto out_set_summed
;
1700 offset
= skb
->csum_start
- skb_headroom(skb
);
1701 BUG_ON(offset
>= skb_headlen(skb
));
1702 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1704 offset
+= skb
->csum_offset
;
1705 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1707 if (skb_cloned(skb
) &&
1708 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1709 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1714 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1716 skb
->ip_summed
= CHECKSUM_NONE
;
1720 EXPORT_SYMBOL(skb_checksum_help
);
1723 * skb_gso_segment - Perform segmentation on skb.
1724 * @skb: buffer to segment
1725 * @features: features for the output path (see dev->features)
1727 * This function segments the given skb and returns a list of segments.
1729 * It may return NULL if the skb requires no segmentation. This is
1730 * only possible when GSO is used for verifying header integrity.
1732 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1734 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1735 struct packet_type
*ptype
;
1736 __be16 type
= skb
->protocol
;
1739 skb_reset_mac_header(skb
);
1740 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1741 __skb_pull(skb
, skb
->mac_len
);
1743 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1744 struct net_device
*dev
= skb
->dev
;
1745 struct ethtool_drvinfo info
= {};
1747 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1748 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1750 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1752 info
.driver
, dev
? dev
->features
: 0L,
1753 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1754 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1756 if (skb_header_cloned(skb
) &&
1757 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1758 return ERR_PTR(err
);
1762 list_for_each_entry_rcu(ptype
,
1763 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1764 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1765 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1766 err
= ptype
->gso_send_check(skb
);
1767 segs
= ERR_PTR(err
);
1768 if (err
|| skb_gso_ok(skb
, features
))
1770 __skb_push(skb
, (skb
->data
-
1771 skb_network_header(skb
)));
1773 segs
= ptype
->gso_segment(skb
, features
);
1779 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1783 EXPORT_SYMBOL(skb_gso_segment
);
1785 /* Take action when hardware reception checksum errors are detected. */
1787 void netdev_rx_csum_fault(struct net_device
*dev
)
1789 if (net_ratelimit()) {
1790 printk(KERN_ERR
"%s: hw csum failure.\n",
1791 dev
? dev
->name
: "<unknown>");
1795 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1798 /* Actually, we should eliminate this check as soon as we know, that:
1799 * 1. IOMMU is present and allows to map all the memory.
1800 * 2. No high memory really exists on this machine.
1803 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1805 #ifdef CONFIG_HIGHMEM
1808 if (dev
->features
& NETIF_F_HIGHDMA
)
1811 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1812 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1820 void (*destructor
)(struct sk_buff
*skb
);
1823 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1825 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1827 struct dev_gso_cb
*cb
;
1830 struct sk_buff
*nskb
= skb
->next
;
1832 skb
->next
= nskb
->next
;
1835 } while (skb
->next
);
1837 cb
= DEV_GSO_CB(skb
);
1839 cb
->destructor(skb
);
1843 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1844 * @skb: buffer to segment
1846 * This function segments the given skb and stores the list of segments
1849 static int dev_gso_segment(struct sk_buff
*skb
)
1851 struct net_device
*dev
= skb
->dev
;
1852 struct sk_buff
*segs
;
1853 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1856 segs
= skb_gso_segment(skb
, features
);
1858 /* Verifying header integrity only. */
1863 return PTR_ERR(segs
);
1866 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1867 skb
->destructor
= dev_gso_skb_destructor
;
1872 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1873 struct netdev_queue
*txq
)
1875 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1876 int rc
= NETDEV_TX_OK
;
1878 if (likely(!skb
->next
)) {
1879 if (!list_empty(&ptype_all
))
1880 dev_queue_xmit_nit(skb
, dev
);
1882 if (netif_needs_gso(dev
, skb
)) {
1883 if (unlikely(dev_gso_segment(skb
)))
1890 * If device doesnt need skb->dst, release it right now while
1891 * its hot in this cpu cache
1893 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1896 rc
= ops
->ndo_start_xmit(skb
, dev
);
1897 if (rc
== NETDEV_TX_OK
)
1898 txq_trans_update(txq
);
1900 * TODO: if skb_orphan() was called by
1901 * dev->hard_start_xmit() (for example, the unmodified
1902 * igb driver does that; bnx2 doesn't), then
1903 * skb_tx_software_timestamp() will be unable to send
1904 * back the time stamp.
1906 * How can this be prevented? Always create another
1907 * reference to the socket before calling
1908 * dev->hard_start_xmit()? Prevent that skb_orphan()
1909 * does anything in dev->hard_start_xmit() by clearing
1910 * the skb destructor before the call and restoring it
1911 * afterwards, then doing the skb_orphan() ourselves?
1918 struct sk_buff
*nskb
= skb
->next
;
1920 skb
->next
= nskb
->next
;
1924 * If device doesnt need nskb->dst, release it right now while
1925 * its hot in this cpu cache
1927 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1930 rc
= ops
->ndo_start_xmit(nskb
, dev
);
1931 if (unlikely(rc
!= NETDEV_TX_OK
)) {
1932 if (rc
& ~NETDEV_TX_MASK
)
1933 goto out_kfree_gso_skb
;
1934 nskb
->next
= skb
->next
;
1938 txq_trans_update(txq
);
1939 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
1940 return NETDEV_TX_BUSY
;
1941 } while (skb
->next
);
1944 if (likely(skb
->next
== NULL
))
1945 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1951 static u32 hashrnd __read_mostly
;
1953 u16
skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
)
1957 if (skb_rx_queue_recorded(skb
)) {
1958 hash
= skb_get_rx_queue(skb
);
1959 while (unlikely(hash
>= dev
->real_num_tx_queues
))
1960 hash
-= dev
->real_num_tx_queues
;
1964 if (skb
->sk
&& skb
->sk
->sk_hash
)
1965 hash
= skb
->sk
->sk_hash
;
1967 hash
= skb
->protocol
;
1969 hash
= jhash_1word(hash
, hashrnd
);
1971 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
1973 EXPORT_SYMBOL(skb_tx_hash
);
1975 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
1977 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
1978 if (net_ratelimit()) {
1979 netdev_warn(dev
, "selects TX queue %d, but "
1980 "real number of TX queues is %d\n",
1981 queue_index
, dev
->real_num_tx_queues
);
1988 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
1989 struct sk_buff
*skb
)
1992 struct sock
*sk
= skb
->sk
;
1994 if (sk_tx_queue_recorded(sk
)) {
1995 queue_index
= sk_tx_queue_get(sk
);
1997 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1999 if (ops
->ndo_select_queue
) {
2000 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2001 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2004 if (dev
->real_num_tx_queues
> 1)
2005 queue_index
= skb_tx_hash(dev
, skb
);
2007 if (sk
&& sk
->sk_dst_cache
)
2008 sk_tx_queue_set(sk
, queue_index
);
2012 skb_set_queue_mapping(skb
, queue_index
);
2013 return netdev_get_tx_queue(dev
, queue_index
);
2016 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2017 struct net_device
*dev
,
2018 struct netdev_queue
*txq
)
2020 spinlock_t
*root_lock
= qdisc_lock(q
);
2023 spin_lock(root_lock
);
2024 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2027 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2028 !test_and_set_bit(__QDISC_STATE_RUNNING
, &q
->state
)) {
2030 * This is a work-conserving queue; there are no old skbs
2031 * waiting to be sent out; and the qdisc is not running -
2032 * xmit the skb directly.
2034 __qdisc_update_bstats(q
, skb
->len
);
2035 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
))
2038 clear_bit(__QDISC_STATE_RUNNING
, &q
->state
);
2040 rc
= NET_XMIT_SUCCESS
;
2042 rc
= qdisc_enqueue_root(skb
, q
);
2045 spin_unlock(root_lock
);
2051 * Returns true if either:
2052 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2053 * 2. skb is fragmented and the device does not support SG, or if
2054 * at least one of fragments is in highmem and device does not
2055 * support DMA from it.
2057 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2058 struct net_device
*dev
)
2060 return (skb_has_frags(skb
) && !(dev
->features
& NETIF_F_FRAGLIST
)) ||
2061 (skb_shinfo(skb
)->nr_frags
&& (!(dev
->features
& NETIF_F_SG
) ||
2062 illegal_highdma(dev
, skb
)));
2066 * dev_queue_xmit - transmit a buffer
2067 * @skb: buffer to transmit
2069 * Queue a buffer for transmission to a network device. The caller must
2070 * have set the device and priority and built the buffer before calling
2071 * this function. The function can be called from an interrupt.
2073 * A negative errno code is returned on a failure. A success does not
2074 * guarantee the frame will be transmitted as it may be dropped due
2075 * to congestion or traffic shaping.
2077 * -----------------------------------------------------------------------------------
2078 * I notice this method can also return errors from the queue disciplines,
2079 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2082 * Regardless of the return value, the skb is consumed, so it is currently
2083 * difficult to retry a send to this method. (You can bump the ref count
2084 * before sending to hold a reference for retry if you are careful.)
2086 * When calling this method, interrupts MUST be enabled. This is because
2087 * the BH enable code must have IRQs enabled so that it will not deadlock.
2090 int dev_queue_xmit(struct sk_buff
*skb
)
2092 struct net_device
*dev
= skb
->dev
;
2093 struct netdev_queue
*txq
;
2097 /* GSO will handle the following emulations directly. */
2098 if (netif_needs_gso(dev
, skb
))
2101 /* Convert a paged skb to linear, if required */
2102 if (skb_needs_linearize(skb
, dev
) && __skb_linearize(skb
))
2105 /* If packet is not checksummed and device does not support
2106 * checksumming for this protocol, complete checksumming here.
2108 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2109 skb_set_transport_header(skb
, skb
->csum_start
-
2111 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
2116 /* Disable soft irqs for various locks below. Also
2117 * stops preemption for RCU.
2121 txq
= dev_pick_tx(dev
, skb
);
2122 q
= rcu_dereference_bh(txq
->qdisc
);
2124 #ifdef CONFIG_NET_CLS_ACT
2125 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2128 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2132 /* The device has no queue. Common case for software devices:
2133 loopback, all the sorts of tunnels...
2135 Really, it is unlikely that netif_tx_lock protection is necessary
2136 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2138 However, it is possible, that they rely on protection
2141 Check this and shot the lock. It is not prone from deadlocks.
2142 Either shot noqueue qdisc, it is even simpler 8)
2144 if (dev
->flags
& IFF_UP
) {
2145 int cpu
= smp_processor_id(); /* ok because BHs are off */
2147 if (txq
->xmit_lock_owner
!= cpu
) {
2149 HARD_TX_LOCK(dev
, txq
, cpu
);
2151 if (!netif_tx_queue_stopped(txq
)) {
2152 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2153 if (dev_xmit_complete(rc
)) {
2154 HARD_TX_UNLOCK(dev
, txq
);
2158 HARD_TX_UNLOCK(dev
, txq
);
2159 if (net_ratelimit())
2160 printk(KERN_CRIT
"Virtual device %s asks to "
2161 "queue packet!\n", dev
->name
);
2163 /* Recursion is detected! It is possible,
2165 if (net_ratelimit())
2166 printk(KERN_CRIT
"Dead loop on virtual device "
2167 "%s, fix it urgently!\n", dev
->name
);
2172 rcu_read_unlock_bh();
2178 rcu_read_unlock_bh();
2181 EXPORT_SYMBOL(dev_queue_xmit
);
2184 /*=======================================================================
2186 =======================================================================*/
2188 int netdev_max_backlog __read_mostly
= 1000;
2189 int netdev_budget __read_mostly
= 300;
2190 int weight_p __read_mostly
= 64; /* old backlog weight */
2192 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
2196 * get_rps_cpu is called from netif_receive_skb and returns the target
2197 * CPU from the RPS map of the receiving queue for a given skb.
2199 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
)
2201 struct ipv6hdr
*ip6
;
2203 struct netdev_rx_queue
*rxqueue
;
2204 struct rps_map
*map
;
2207 u32 addr1
, addr2
, ports
, ihl
;
2211 if (skb_rx_queue_recorded(skb
)) {
2212 u16 index
= skb_get_rx_queue(skb
);
2213 if (unlikely(index
>= dev
->num_rx_queues
)) {
2214 if (net_ratelimit()) {
2215 netdev_warn(dev
, "received packet on queue "
2216 "%u, but number of RX queues is %u\n",
2217 index
, dev
->num_rx_queues
);
2221 rxqueue
= dev
->_rx
+ index
;
2225 if (!rxqueue
->rps_map
)
2229 goto got_hash
; /* Skip hash computation on packet header */
2231 switch (skb
->protocol
) {
2232 case __constant_htons(ETH_P_IP
):
2233 if (!pskb_may_pull(skb
, sizeof(*ip
)))
2236 ip
= (struct iphdr
*) skb
->data
;
2237 ip_proto
= ip
->protocol
;
2242 case __constant_htons(ETH_P_IPV6
):
2243 if (!pskb_may_pull(skb
, sizeof(*ip6
)))
2246 ip6
= (struct ipv6hdr
*) skb
->data
;
2247 ip_proto
= ip6
->nexthdr
;
2248 addr1
= ip6
->saddr
.s6_addr32
[3];
2249 addr2
= ip6
->daddr
.s6_addr32
[3];
2263 case IPPROTO_UDPLITE
:
2264 if (pskb_may_pull(skb
, (ihl
* 4) + 4))
2265 ports
= *((u32
*) (skb
->data
+ (ihl
* 4)));
2272 skb
->rxhash
= jhash_3words(addr1
, addr2
, ports
, hashrnd
);
2277 map
= rcu_dereference(rxqueue
->rps_map
);
2279 u16 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2281 if (cpu_online(tcpu
)) {
2293 * This structure holds the per-CPU mask of CPUs for which IPIs are scheduled
2294 * to be sent to kick remote softirq processing. There are two masks since
2295 * the sending of IPIs must be done with interrupts enabled. The select field
2296 * indicates the current mask that enqueue_backlog uses to schedule IPIs.
2297 * select is flipped before net_rps_action is called while still under lock,
2298 * net_rps_action then uses the non-selected mask to send the IPIs and clears
2299 * it without conflicting with enqueue_backlog operation.
2301 struct rps_remote_softirq_cpus
{
2305 static DEFINE_PER_CPU(struct rps_remote_softirq_cpus
, rps_remote_softirq_cpus
);
2307 /* Called from hardirq (IPI) context */
2308 static void trigger_softirq(void *data
)
2310 struct softnet_data
*queue
= data
;
2311 __napi_schedule(&queue
->backlog
);
2312 __get_cpu_var(netdev_rx_stat
).received_rps
++;
2314 #endif /* CONFIG_SMP */
2317 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2318 * queue (may be a remote CPU queue).
2320 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
)
2322 struct softnet_data
*queue
;
2323 unsigned long flags
;
2325 queue
= &per_cpu(softnet_data
, cpu
);
2327 local_irq_save(flags
);
2328 __get_cpu_var(netdev_rx_stat
).total
++;
2331 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
2332 if (queue
->input_pkt_queue
.qlen
) {
2334 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
2336 local_irq_restore(flags
);
2337 return NET_RX_SUCCESS
;
2340 /* Schedule NAPI for backlog device */
2341 if (napi_schedule_prep(&queue
->backlog
)) {
2343 if (cpu
!= smp_processor_id()) {
2344 struct rps_remote_softirq_cpus
*rcpus
=
2345 &__get_cpu_var(rps_remote_softirq_cpus
);
2347 cpu_set(cpu
, rcpus
->mask
[rcpus
->select
]);
2348 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2350 __napi_schedule(&queue
->backlog
);
2352 __napi_schedule(&queue
->backlog
);
2360 __get_cpu_var(netdev_rx_stat
).dropped
++;
2361 local_irq_restore(flags
);
2368 * netif_rx - post buffer to the network code
2369 * @skb: buffer to post
2371 * This function receives a packet from a device driver and queues it for
2372 * the upper (protocol) levels to process. It always succeeds. The buffer
2373 * may be dropped during processing for congestion control or by the
2377 * NET_RX_SUCCESS (no congestion)
2378 * NET_RX_DROP (packet was dropped)
2382 int netif_rx(struct sk_buff
*skb
)
2386 /* if netpoll wants it, pretend we never saw it */
2387 if (netpoll_rx(skb
))
2390 if (!skb
->tstamp
.tv64
)
2394 cpu
= get_rps_cpu(skb
->dev
, skb
);
2396 cpu
= smp_processor_id();
2398 cpu
= smp_processor_id();
2401 return enqueue_to_backlog(skb
, cpu
);
2403 EXPORT_SYMBOL(netif_rx
);
2405 int netif_rx_ni(struct sk_buff
*skb
)
2410 err
= netif_rx(skb
);
2411 if (local_softirq_pending())
2417 EXPORT_SYMBOL(netif_rx_ni
);
2419 static void net_tx_action(struct softirq_action
*h
)
2421 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2423 if (sd
->completion_queue
) {
2424 struct sk_buff
*clist
;
2426 local_irq_disable();
2427 clist
= sd
->completion_queue
;
2428 sd
->completion_queue
= NULL
;
2432 struct sk_buff
*skb
= clist
;
2433 clist
= clist
->next
;
2435 WARN_ON(atomic_read(&skb
->users
));
2440 if (sd
->output_queue
) {
2443 local_irq_disable();
2444 head
= sd
->output_queue
;
2445 sd
->output_queue
= NULL
;
2449 struct Qdisc
*q
= head
;
2450 spinlock_t
*root_lock
;
2452 head
= head
->next_sched
;
2454 root_lock
= qdisc_lock(q
);
2455 if (spin_trylock(root_lock
)) {
2456 smp_mb__before_clear_bit();
2457 clear_bit(__QDISC_STATE_SCHED
,
2460 spin_unlock(root_lock
);
2462 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2464 __netif_reschedule(q
);
2466 smp_mb__before_clear_bit();
2467 clear_bit(__QDISC_STATE_SCHED
,
2475 static inline int deliver_skb(struct sk_buff
*skb
,
2476 struct packet_type
*pt_prev
,
2477 struct net_device
*orig_dev
)
2479 atomic_inc(&skb
->users
);
2480 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2483 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2485 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2486 /* This hook is defined here for ATM LANE */
2487 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2488 unsigned char *addr
) __read_mostly
;
2489 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2493 * If bridge module is loaded call bridging hook.
2494 * returns NULL if packet was consumed.
2496 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
2497 struct sk_buff
*skb
) __read_mostly
;
2498 EXPORT_SYMBOL_GPL(br_handle_frame_hook
);
2500 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
2501 struct packet_type
**pt_prev
, int *ret
,
2502 struct net_device
*orig_dev
)
2504 struct net_bridge_port
*port
;
2506 if (skb
->pkt_type
== PACKET_LOOPBACK
||
2507 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
2511 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2515 return br_handle_frame_hook(port
, skb
);
2518 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2521 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2522 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
2523 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
2525 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
2526 struct packet_type
**pt_prev
,
2528 struct net_device
*orig_dev
)
2530 if (skb
->dev
->macvlan_port
== NULL
)
2534 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2537 return macvlan_handle_frame_hook(skb
);
2540 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2543 #ifdef CONFIG_NET_CLS_ACT
2544 /* TODO: Maybe we should just force sch_ingress to be compiled in
2545 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2546 * a compare and 2 stores extra right now if we dont have it on
2547 * but have CONFIG_NET_CLS_ACT
2548 * NOTE: This doesnt stop any functionality; if you dont have
2549 * the ingress scheduler, you just cant add policies on ingress.
2552 static int ing_filter(struct sk_buff
*skb
)
2554 struct net_device
*dev
= skb
->dev
;
2555 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2556 struct netdev_queue
*rxq
;
2557 int result
= TC_ACT_OK
;
2560 if (MAX_RED_LOOP
< ttl
++) {
2562 "Redir loop detected Dropping packet (%d->%d)\n",
2563 skb
->skb_iif
, dev
->ifindex
);
2567 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2568 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2570 rxq
= &dev
->rx_queue
;
2573 if (q
!= &noop_qdisc
) {
2574 spin_lock(qdisc_lock(q
));
2575 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2576 result
= qdisc_enqueue_root(skb
, q
);
2577 spin_unlock(qdisc_lock(q
));
2583 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2584 struct packet_type
**pt_prev
,
2585 int *ret
, struct net_device
*orig_dev
)
2587 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2591 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2594 /* Huh? Why does turning on AF_PACKET affect this? */
2595 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2598 switch (ing_filter(skb
)) {
2612 * netif_nit_deliver - deliver received packets to network taps
2615 * This function is used to deliver incoming packets to network
2616 * taps. It should be used when the normal netif_receive_skb path
2617 * is bypassed, for example because of VLAN acceleration.
2619 void netif_nit_deliver(struct sk_buff
*skb
)
2621 struct packet_type
*ptype
;
2623 if (list_empty(&ptype_all
))
2626 skb_reset_network_header(skb
);
2627 skb_reset_transport_header(skb
);
2628 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2631 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2632 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2633 deliver_skb(skb
, ptype
, skb
->dev
);
2638 static int __netif_receive_skb(struct sk_buff
*skb
)
2640 struct packet_type
*ptype
, *pt_prev
;
2641 struct net_device
*orig_dev
;
2642 struct net_device
*master
;
2643 struct net_device
*null_or_orig
;
2644 struct net_device
*null_or_bond
;
2645 int ret
= NET_RX_DROP
;
2648 if (!skb
->tstamp
.tv64
)
2651 if (vlan_tx_tag_present(skb
) && vlan_hwaccel_do_receive(skb
))
2652 return NET_RX_SUCCESS
;
2654 /* if we've gotten here through NAPI, check netpoll */
2655 if (netpoll_receive_skb(skb
))
2659 skb
->skb_iif
= skb
->dev
->ifindex
;
2661 null_or_orig
= NULL
;
2662 orig_dev
= skb
->dev
;
2663 master
= ACCESS_ONCE(orig_dev
->master
);
2665 if (skb_bond_should_drop(skb
, master
))
2666 null_or_orig
= orig_dev
; /* deliver only exact match */
2671 __get_cpu_var(netdev_rx_stat
).total
++;
2673 skb_reset_network_header(skb
);
2674 skb_reset_transport_header(skb
);
2675 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2681 #ifdef CONFIG_NET_CLS_ACT
2682 if (skb
->tc_verd
& TC_NCLS
) {
2683 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2688 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2689 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2690 ptype
->dev
== orig_dev
) {
2692 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2697 #ifdef CONFIG_NET_CLS_ACT
2698 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2704 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
2707 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
2712 * Make sure frames received on VLAN interfaces stacked on
2713 * bonding interfaces still make their way to any base bonding
2714 * device that may have registered for a specific ptype. The
2715 * handler may have to adjust skb->dev and orig_dev.
2717 null_or_bond
= NULL
;
2718 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
2719 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
2720 null_or_bond
= vlan_dev_real_dev(skb
->dev
);
2723 type
= skb
->protocol
;
2724 list_for_each_entry_rcu(ptype
,
2725 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2726 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
2727 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
2728 ptype
->dev
== null_or_bond
)) {
2730 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2736 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2739 /* Jamal, now you will not able to escape explaining
2740 * me how you were going to use this. :-)
2751 * netif_receive_skb - process receive buffer from network
2752 * @skb: buffer to process
2754 * netif_receive_skb() is the main receive data processing function.
2755 * It always succeeds. The buffer may be dropped during processing
2756 * for congestion control or by the protocol layers.
2758 * This function may only be called from softirq context and interrupts
2759 * should be enabled.
2761 * Return values (usually ignored):
2762 * NET_RX_SUCCESS: no congestion
2763 * NET_RX_DROP: packet was dropped
2765 int netif_receive_skb(struct sk_buff
*skb
)
2770 cpu
= get_rps_cpu(skb
->dev
, skb
);
2773 return __netif_receive_skb(skb
);
2775 return enqueue_to_backlog(skb
, cpu
);
2777 return __netif_receive_skb(skb
);
2780 EXPORT_SYMBOL(netif_receive_skb
);
2782 /* Network device is going away, flush any packets still pending */
2783 static void flush_backlog(void *arg
)
2785 struct net_device
*dev
= arg
;
2786 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2787 struct sk_buff
*skb
, *tmp
;
2790 skb_queue_walk_safe(&queue
->input_pkt_queue
, skb
, tmp
)
2791 if (skb
->dev
== dev
) {
2792 __skb_unlink(skb
, &queue
->input_pkt_queue
);
2798 static int napi_gro_complete(struct sk_buff
*skb
)
2800 struct packet_type
*ptype
;
2801 __be16 type
= skb
->protocol
;
2802 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2805 if (NAPI_GRO_CB(skb
)->count
== 1) {
2806 skb_shinfo(skb
)->gso_size
= 0;
2811 list_for_each_entry_rcu(ptype
, head
, list
) {
2812 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
2815 err
= ptype
->gro_complete(skb
);
2821 WARN_ON(&ptype
->list
== head
);
2823 return NET_RX_SUCCESS
;
2827 return netif_receive_skb(skb
);
2830 static void napi_gro_flush(struct napi_struct
*napi
)
2832 struct sk_buff
*skb
, *next
;
2834 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
2837 napi_gro_complete(skb
);
2840 napi
->gro_count
= 0;
2841 napi
->gro_list
= NULL
;
2844 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2846 struct sk_buff
**pp
= NULL
;
2847 struct packet_type
*ptype
;
2848 __be16 type
= skb
->protocol
;
2849 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2852 enum gro_result ret
;
2854 if (!(skb
->dev
->features
& NETIF_F_GRO
))
2857 if (skb_is_gso(skb
) || skb_has_frags(skb
))
2861 list_for_each_entry_rcu(ptype
, head
, list
) {
2862 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
2865 skb_set_network_header(skb
, skb_gro_offset(skb
));
2866 mac_len
= skb
->network_header
- skb
->mac_header
;
2867 skb
->mac_len
= mac_len
;
2868 NAPI_GRO_CB(skb
)->same_flow
= 0;
2869 NAPI_GRO_CB(skb
)->flush
= 0;
2870 NAPI_GRO_CB(skb
)->free
= 0;
2872 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
2877 if (&ptype
->list
== head
)
2880 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
2881 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
2884 struct sk_buff
*nskb
= *pp
;
2888 napi_gro_complete(nskb
);
2895 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
2899 NAPI_GRO_CB(skb
)->count
= 1;
2900 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
2901 skb
->next
= napi
->gro_list
;
2902 napi
->gro_list
= skb
;
2906 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
2907 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
2909 BUG_ON(skb
->end
- skb
->tail
< grow
);
2911 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
2914 skb
->data_len
-= grow
;
2916 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
2917 skb_shinfo(skb
)->frags
[0].size
-= grow
;
2919 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
2920 put_page(skb_shinfo(skb
)->frags
[0].page
);
2921 memmove(skb_shinfo(skb
)->frags
,
2922 skb_shinfo(skb
)->frags
+ 1,
2923 --skb_shinfo(skb
)->nr_frags
);
2934 EXPORT_SYMBOL(dev_gro_receive
);
2937 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2941 if (netpoll_rx_on(skb
))
2944 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
2945 NAPI_GRO_CB(p
)->same_flow
=
2946 (p
->dev
== skb
->dev
) &&
2947 !compare_ether_header(skb_mac_header(p
),
2948 skb_gro_mac_header(skb
));
2949 NAPI_GRO_CB(p
)->flush
= 0;
2952 return dev_gro_receive(napi
, skb
);
2955 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
2959 if (netif_receive_skb(skb
))
2964 case GRO_MERGED_FREE
:
2975 EXPORT_SYMBOL(napi_skb_finish
);
2977 void skb_gro_reset_offset(struct sk_buff
*skb
)
2979 NAPI_GRO_CB(skb
)->data_offset
= 0;
2980 NAPI_GRO_CB(skb
)->frag0
= NULL
;
2981 NAPI_GRO_CB(skb
)->frag0_len
= 0;
2983 if (skb
->mac_header
== skb
->tail
&&
2984 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
2985 NAPI_GRO_CB(skb
)->frag0
=
2986 page_address(skb_shinfo(skb
)->frags
[0].page
) +
2987 skb_shinfo(skb
)->frags
[0].page_offset
;
2988 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
2991 EXPORT_SYMBOL(skb_gro_reset_offset
);
2993 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2995 skb_gro_reset_offset(skb
);
2997 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
2999 EXPORT_SYMBOL(napi_gro_receive
);
3001 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3003 __skb_pull(skb
, skb_headlen(skb
));
3004 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3008 EXPORT_SYMBOL(napi_reuse_skb
);
3010 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3012 struct sk_buff
*skb
= napi
->skb
;
3015 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3021 EXPORT_SYMBOL(napi_get_frags
);
3023 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3029 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3031 if (ret
== GRO_HELD
)
3032 skb_gro_pull(skb
, -ETH_HLEN
);
3033 else if (netif_receive_skb(skb
))
3038 case GRO_MERGED_FREE
:
3039 napi_reuse_skb(napi
, skb
);
3048 EXPORT_SYMBOL(napi_frags_finish
);
3050 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3052 struct sk_buff
*skb
= napi
->skb
;
3059 skb_reset_mac_header(skb
);
3060 skb_gro_reset_offset(skb
);
3062 off
= skb_gro_offset(skb
);
3063 hlen
= off
+ sizeof(*eth
);
3064 eth
= skb_gro_header_fast(skb
, off
);
3065 if (skb_gro_header_hard(skb
, hlen
)) {
3066 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3067 if (unlikely(!eth
)) {
3068 napi_reuse_skb(napi
, skb
);
3074 skb_gro_pull(skb
, sizeof(*eth
));
3077 * This works because the only protocols we care about don't require
3078 * special handling. We'll fix it up properly at the end.
3080 skb
->protocol
= eth
->h_proto
;
3085 EXPORT_SYMBOL(napi_frags_skb
);
3087 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3089 struct sk_buff
*skb
= napi_frags_skb(napi
);
3094 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3096 EXPORT_SYMBOL(napi_gro_frags
);
3098 static int process_backlog(struct napi_struct
*napi
, int quota
)
3101 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
3102 unsigned long start_time
= jiffies
;
3104 napi
->weight
= weight_p
;
3106 struct sk_buff
*skb
;
3108 local_irq_disable();
3110 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
3112 __napi_complete(napi
);
3113 spin_unlock_irq(&queue
->input_pkt_queue
.lock
);
3119 __netif_receive_skb(skb
);
3120 } while (++work
< quota
&& jiffies
== start_time
);
3126 * __napi_schedule - schedule for receive
3127 * @n: entry to schedule
3129 * The entry's receive function will be scheduled to run
3131 void __napi_schedule(struct napi_struct
*n
)
3133 unsigned long flags
;
3135 local_irq_save(flags
);
3136 list_add_tail(&n
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
3137 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3138 local_irq_restore(flags
);
3140 EXPORT_SYMBOL(__napi_schedule
);
3142 void __napi_complete(struct napi_struct
*n
)
3144 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3145 BUG_ON(n
->gro_list
);
3147 list_del(&n
->poll_list
);
3148 smp_mb__before_clear_bit();
3149 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3151 EXPORT_SYMBOL(__napi_complete
);
3153 void napi_complete(struct napi_struct
*n
)
3155 unsigned long flags
;
3158 * don't let napi dequeue from the cpu poll list
3159 * just in case its running on a different cpu
3161 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3165 local_irq_save(flags
);
3167 local_irq_restore(flags
);
3169 EXPORT_SYMBOL(napi_complete
);
3171 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3172 int (*poll
)(struct napi_struct
*, int), int weight
)
3174 INIT_LIST_HEAD(&napi
->poll_list
);
3175 napi
->gro_count
= 0;
3176 napi
->gro_list
= NULL
;
3179 napi
->weight
= weight
;
3180 list_add(&napi
->dev_list
, &dev
->napi_list
);
3182 #ifdef CONFIG_NETPOLL
3183 spin_lock_init(&napi
->poll_lock
);
3184 napi
->poll_owner
= -1;
3186 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3188 EXPORT_SYMBOL(netif_napi_add
);
3190 void netif_napi_del(struct napi_struct
*napi
)
3192 struct sk_buff
*skb
, *next
;
3194 list_del_init(&napi
->dev_list
);
3195 napi_free_frags(napi
);
3197 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3203 napi
->gro_list
= NULL
;
3204 napi
->gro_count
= 0;
3206 EXPORT_SYMBOL(netif_napi_del
);
3210 * net_rps_action sends any pending IPI's for rps. This is only called from
3211 * softirq and interrupts must be enabled.
3213 static void net_rps_action(cpumask_t
*mask
)
3217 /* Send pending IPI's to kick RPS processing on remote cpus. */
3218 for_each_cpu_mask_nr(cpu
, *mask
) {
3219 struct softnet_data
*queue
= &per_cpu(softnet_data
, cpu
);
3220 if (cpu_online(cpu
))
3221 __smp_call_function_single(cpu
, &queue
->csd
, 0);
3227 static void net_rx_action(struct softirq_action
*h
)
3229 struct list_head
*list
= &__get_cpu_var(softnet_data
).poll_list
;
3230 unsigned long time_limit
= jiffies
+ 2;
3231 int budget
= netdev_budget
;
3235 struct rps_remote_softirq_cpus
*rcpus
;
3238 local_irq_disable();
3240 while (!list_empty(list
)) {
3241 struct napi_struct
*n
;
3244 /* If softirq window is exhuasted then punt.
3245 * Allow this to run for 2 jiffies since which will allow
3246 * an average latency of 1.5/HZ.
3248 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3253 /* Even though interrupts have been re-enabled, this
3254 * access is safe because interrupts can only add new
3255 * entries to the tail of this list, and only ->poll()
3256 * calls can remove this head entry from the list.
3258 n
= list_first_entry(list
, struct napi_struct
, poll_list
);
3260 have
= netpoll_poll_lock(n
);
3264 /* This NAPI_STATE_SCHED test is for avoiding a race
3265 * with netpoll's poll_napi(). Only the entity which
3266 * obtains the lock and sees NAPI_STATE_SCHED set will
3267 * actually make the ->poll() call. Therefore we avoid
3268 * accidently calling ->poll() when NAPI is not scheduled.
3271 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3272 work
= n
->poll(n
, weight
);
3276 WARN_ON_ONCE(work
> weight
);
3280 local_irq_disable();
3282 /* Drivers must not modify the NAPI state if they
3283 * consume the entire weight. In such cases this code
3284 * still "owns" the NAPI instance and therefore can
3285 * move the instance around on the list at-will.
3287 if (unlikely(work
== weight
)) {
3288 if (unlikely(napi_disable_pending(n
))) {
3291 local_irq_disable();
3293 list_move_tail(&n
->poll_list
, list
);
3296 netpoll_poll_unlock(have
);
3300 rcpus
= &__get_cpu_var(rps_remote_softirq_cpus
);
3301 select
= rcpus
->select
;
3306 net_rps_action(&rcpus
->mask
[select
]);
3311 #ifdef CONFIG_NET_DMA
3313 * There may not be any more sk_buffs coming right now, so push
3314 * any pending DMA copies to hardware
3316 dma_issue_pending_all();
3322 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
3323 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3327 static gifconf_func_t
*gifconf_list
[NPROTO
];
3330 * register_gifconf - register a SIOCGIF handler
3331 * @family: Address family
3332 * @gifconf: Function handler
3334 * Register protocol dependent address dumping routines. The handler
3335 * that is passed must not be freed or reused until it has been replaced
3336 * by another handler.
3338 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3340 if (family
>= NPROTO
)
3342 gifconf_list
[family
] = gifconf
;
3345 EXPORT_SYMBOL(register_gifconf
);
3349 * Map an interface index to its name (SIOCGIFNAME)
3353 * We need this ioctl for efficient implementation of the
3354 * if_indextoname() function required by the IPv6 API. Without
3355 * it, we would have to search all the interfaces to find a
3359 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3361 struct net_device
*dev
;
3365 * Fetch the caller's info block.
3368 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3372 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3378 strcpy(ifr
.ifr_name
, dev
->name
);
3381 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3387 * Perform a SIOCGIFCONF call. This structure will change
3388 * size eventually, and there is nothing I can do about it.
3389 * Thus we will need a 'compatibility mode'.
3392 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3395 struct net_device
*dev
;
3402 * Fetch the caller's info block.
3405 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3412 * Loop over the interfaces, and write an info block for each.
3416 for_each_netdev(net
, dev
) {
3417 for (i
= 0; i
< NPROTO
; i
++) {
3418 if (gifconf_list
[i
]) {
3421 done
= gifconf_list
[i
](dev
, NULL
, 0);
3423 done
= gifconf_list
[i
](dev
, pos
+ total
,
3433 * All done. Write the updated control block back to the caller.
3435 ifc
.ifc_len
= total
;
3438 * Both BSD and Solaris return 0 here, so we do too.
3440 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3443 #ifdef CONFIG_PROC_FS
3445 * This is invoked by the /proc filesystem handler to display a device
3448 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3451 struct net
*net
= seq_file_net(seq
);
3453 struct net_device
*dev
;
3457 return SEQ_START_TOKEN
;
3460 for_each_netdev_rcu(net
, dev
)
3467 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3469 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3470 first_net_device(seq_file_net(seq
)) :
3471 next_net_device((struct net_device
*)v
);
3474 return rcu_dereference(dev
);
3477 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3483 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3485 const struct net_device_stats
*stats
= dev_get_stats(dev
);
3487 seq_printf(seq
, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3488 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3489 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3491 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3492 stats
->rx_fifo_errors
,
3493 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3494 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3495 stats
->rx_compressed
, stats
->multicast
,
3496 stats
->tx_bytes
, stats
->tx_packets
,
3497 stats
->tx_errors
, stats
->tx_dropped
,
3498 stats
->tx_fifo_errors
, stats
->collisions
,
3499 stats
->tx_carrier_errors
+
3500 stats
->tx_aborted_errors
+
3501 stats
->tx_window_errors
+
3502 stats
->tx_heartbeat_errors
,
3503 stats
->tx_compressed
);
3507 * Called from the PROCfs module. This now uses the new arbitrary sized
3508 * /proc/net interface to create /proc/net/dev
3510 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3512 if (v
== SEQ_START_TOKEN
)
3513 seq_puts(seq
, "Inter-| Receive "
3515 " face |bytes packets errs drop fifo frame "
3516 "compressed multicast|bytes packets errs "
3517 "drop fifo colls carrier compressed\n");
3519 dev_seq_printf_stats(seq
, v
);
3523 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
3525 struct netif_rx_stats
*rc
= NULL
;
3527 while (*pos
< nr_cpu_ids
)
3528 if (cpu_online(*pos
)) {
3529 rc
= &per_cpu(netdev_rx_stat
, *pos
);
3536 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3538 return softnet_get_online(pos
);
3541 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3544 return softnet_get_online(pos
);
3547 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3551 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3553 struct netif_rx_stats
*s
= v
;
3555 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3556 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
3557 0, 0, 0, 0, /* was fastroute */
3558 s
->cpu_collision
, s
->received_rps
);
3562 static const struct seq_operations dev_seq_ops
= {
3563 .start
= dev_seq_start
,
3564 .next
= dev_seq_next
,
3565 .stop
= dev_seq_stop
,
3566 .show
= dev_seq_show
,
3569 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3571 return seq_open_net(inode
, file
, &dev_seq_ops
,
3572 sizeof(struct seq_net_private
));
3575 static const struct file_operations dev_seq_fops
= {
3576 .owner
= THIS_MODULE
,
3577 .open
= dev_seq_open
,
3579 .llseek
= seq_lseek
,
3580 .release
= seq_release_net
,
3583 static const struct seq_operations softnet_seq_ops
= {
3584 .start
= softnet_seq_start
,
3585 .next
= softnet_seq_next
,
3586 .stop
= softnet_seq_stop
,
3587 .show
= softnet_seq_show
,
3590 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3592 return seq_open(file
, &softnet_seq_ops
);
3595 static const struct file_operations softnet_seq_fops
= {
3596 .owner
= THIS_MODULE
,
3597 .open
= softnet_seq_open
,
3599 .llseek
= seq_lseek
,
3600 .release
= seq_release
,
3603 static void *ptype_get_idx(loff_t pos
)
3605 struct packet_type
*pt
= NULL
;
3609 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3615 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3616 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3625 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3629 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3632 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3634 struct packet_type
*pt
;
3635 struct list_head
*nxt
;
3639 if (v
== SEQ_START_TOKEN
)
3640 return ptype_get_idx(0);
3643 nxt
= pt
->list
.next
;
3644 if (pt
->type
== htons(ETH_P_ALL
)) {
3645 if (nxt
!= &ptype_all
)
3648 nxt
= ptype_base
[0].next
;
3650 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3652 while (nxt
== &ptype_base
[hash
]) {
3653 if (++hash
>= PTYPE_HASH_SIZE
)
3655 nxt
= ptype_base
[hash
].next
;
3658 return list_entry(nxt
, struct packet_type
, list
);
3661 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3667 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3669 struct packet_type
*pt
= v
;
3671 if (v
== SEQ_START_TOKEN
)
3672 seq_puts(seq
, "Type Device Function\n");
3673 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3674 if (pt
->type
== htons(ETH_P_ALL
))
3675 seq_puts(seq
, "ALL ");
3677 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3679 seq_printf(seq
, " %-8s %pF\n",
3680 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
3686 static const struct seq_operations ptype_seq_ops
= {
3687 .start
= ptype_seq_start
,
3688 .next
= ptype_seq_next
,
3689 .stop
= ptype_seq_stop
,
3690 .show
= ptype_seq_show
,
3693 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
3695 return seq_open_net(inode
, file
, &ptype_seq_ops
,
3696 sizeof(struct seq_net_private
));
3699 static const struct file_operations ptype_seq_fops
= {
3700 .owner
= THIS_MODULE
,
3701 .open
= ptype_seq_open
,
3703 .llseek
= seq_lseek
,
3704 .release
= seq_release_net
,
3708 static int __net_init
dev_proc_net_init(struct net
*net
)
3712 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
3714 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
3716 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
3719 if (wext_proc_init(net
))
3725 proc_net_remove(net
, "ptype");
3727 proc_net_remove(net
, "softnet_stat");
3729 proc_net_remove(net
, "dev");
3733 static void __net_exit
dev_proc_net_exit(struct net
*net
)
3735 wext_proc_exit(net
);
3737 proc_net_remove(net
, "ptype");
3738 proc_net_remove(net
, "softnet_stat");
3739 proc_net_remove(net
, "dev");
3742 static struct pernet_operations __net_initdata dev_proc_ops
= {
3743 .init
= dev_proc_net_init
,
3744 .exit
= dev_proc_net_exit
,
3747 static int __init
dev_proc_init(void)
3749 return register_pernet_subsys(&dev_proc_ops
);
3752 #define dev_proc_init() 0
3753 #endif /* CONFIG_PROC_FS */
3757 * netdev_set_master - set up master/slave pair
3758 * @slave: slave device
3759 * @master: new master device
3761 * Changes the master device of the slave. Pass %NULL to break the
3762 * bonding. The caller must hold the RTNL semaphore. On a failure
3763 * a negative errno code is returned. On success the reference counts
3764 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3765 * function returns zero.
3767 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
3769 struct net_device
*old
= slave
->master
;
3779 slave
->master
= master
;
3786 slave
->flags
|= IFF_SLAVE
;
3788 slave
->flags
&= ~IFF_SLAVE
;
3790 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
3793 EXPORT_SYMBOL(netdev_set_master
);
3795 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
3797 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3799 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
3800 ops
->ndo_change_rx_flags(dev
, flags
);
3803 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
3805 unsigned short old_flags
= dev
->flags
;
3811 dev
->flags
|= IFF_PROMISC
;
3812 dev
->promiscuity
+= inc
;
3813 if (dev
->promiscuity
== 0) {
3816 * If inc causes overflow, untouch promisc and return error.
3819 dev
->flags
&= ~IFF_PROMISC
;
3821 dev
->promiscuity
-= inc
;
3822 printk(KERN_WARNING
"%s: promiscuity touches roof, "
3823 "set promiscuity failed, promiscuity feature "
3824 "of device might be broken.\n", dev
->name
);
3828 if (dev
->flags
!= old_flags
) {
3829 printk(KERN_INFO
"device %s %s promiscuous mode\n",
3830 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
3832 if (audit_enabled
) {
3833 current_uid_gid(&uid
, &gid
);
3834 audit_log(current
->audit_context
, GFP_ATOMIC
,
3835 AUDIT_ANOM_PROMISCUOUS
,
3836 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3837 dev
->name
, (dev
->flags
& IFF_PROMISC
),
3838 (old_flags
& IFF_PROMISC
),
3839 audit_get_loginuid(current
),
3841 audit_get_sessionid(current
));
3844 dev_change_rx_flags(dev
, IFF_PROMISC
);
3850 * dev_set_promiscuity - update promiscuity count on a device
3854 * Add or remove promiscuity from a device. While the count in the device
3855 * remains above zero the interface remains promiscuous. Once it hits zero
3856 * the device reverts back to normal filtering operation. A negative inc
3857 * value is used to drop promiscuity on the device.
3858 * Return 0 if successful or a negative errno code on error.
3860 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
3862 unsigned short old_flags
= dev
->flags
;
3865 err
= __dev_set_promiscuity(dev
, inc
);
3868 if (dev
->flags
!= old_flags
)
3869 dev_set_rx_mode(dev
);
3872 EXPORT_SYMBOL(dev_set_promiscuity
);
3875 * dev_set_allmulti - update allmulti count on a device
3879 * Add or remove reception of all multicast frames to a device. While the
3880 * count in the device remains above zero the interface remains listening
3881 * to all interfaces. Once it hits zero the device reverts back to normal
3882 * filtering operation. A negative @inc value is used to drop the counter
3883 * when releasing a resource needing all multicasts.
3884 * Return 0 if successful or a negative errno code on error.
3887 int dev_set_allmulti(struct net_device
*dev
, int inc
)
3889 unsigned short old_flags
= dev
->flags
;
3893 dev
->flags
|= IFF_ALLMULTI
;
3894 dev
->allmulti
+= inc
;
3895 if (dev
->allmulti
== 0) {
3898 * If inc causes overflow, untouch allmulti and return error.
3901 dev
->flags
&= ~IFF_ALLMULTI
;
3903 dev
->allmulti
-= inc
;
3904 printk(KERN_WARNING
"%s: allmulti touches roof, "
3905 "set allmulti failed, allmulti feature of "
3906 "device might be broken.\n", dev
->name
);
3910 if (dev
->flags
^ old_flags
) {
3911 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
3912 dev_set_rx_mode(dev
);
3916 EXPORT_SYMBOL(dev_set_allmulti
);
3919 * Upload unicast and multicast address lists to device and
3920 * configure RX filtering. When the device doesn't support unicast
3921 * filtering it is put in promiscuous mode while unicast addresses
3924 void __dev_set_rx_mode(struct net_device
*dev
)
3926 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3928 /* dev_open will call this function so the list will stay sane. */
3929 if (!(dev
->flags
&IFF_UP
))
3932 if (!netif_device_present(dev
))
3935 if (ops
->ndo_set_rx_mode
)
3936 ops
->ndo_set_rx_mode(dev
);
3938 /* Unicast addresses changes may only happen under the rtnl,
3939 * therefore calling __dev_set_promiscuity here is safe.
3941 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
3942 __dev_set_promiscuity(dev
, 1);
3943 dev
->uc_promisc
= 1;
3944 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
3945 __dev_set_promiscuity(dev
, -1);
3946 dev
->uc_promisc
= 0;
3949 if (ops
->ndo_set_multicast_list
)
3950 ops
->ndo_set_multicast_list(dev
);
3954 void dev_set_rx_mode(struct net_device
*dev
)
3956 netif_addr_lock_bh(dev
);
3957 __dev_set_rx_mode(dev
);
3958 netif_addr_unlock_bh(dev
);
3961 /* hw addresses list handling functions */
3963 static int __hw_addr_add(struct netdev_hw_addr_list
*list
, unsigned char *addr
,
3964 int addr_len
, unsigned char addr_type
)
3966 struct netdev_hw_addr
*ha
;
3969 if (addr_len
> MAX_ADDR_LEN
)
3972 list_for_each_entry(ha
, &list
->list
, list
) {
3973 if (!memcmp(ha
->addr
, addr
, addr_len
) &&
3974 ha
->type
== addr_type
) {
3981 alloc_size
= sizeof(*ha
);
3982 if (alloc_size
< L1_CACHE_BYTES
)
3983 alloc_size
= L1_CACHE_BYTES
;
3984 ha
= kmalloc(alloc_size
, GFP_ATOMIC
);
3987 memcpy(ha
->addr
, addr
, addr_len
);
3988 ha
->type
= addr_type
;
3991 list_add_tail_rcu(&ha
->list
, &list
->list
);
3996 static void ha_rcu_free(struct rcu_head
*head
)
3998 struct netdev_hw_addr
*ha
;
4000 ha
= container_of(head
, struct netdev_hw_addr
, rcu_head
);
4004 static int __hw_addr_del(struct netdev_hw_addr_list
*list
, unsigned char *addr
,
4005 int addr_len
, unsigned char addr_type
)
4007 struct netdev_hw_addr
*ha
;
4009 list_for_each_entry(ha
, &list
->list
, list
) {
4010 if (!memcmp(ha
->addr
, addr
, addr_len
) &&
4011 (ha
->type
== addr_type
|| !addr_type
)) {
4014 list_del_rcu(&ha
->list
);
4015 call_rcu(&ha
->rcu_head
, ha_rcu_free
);
4023 static int __hw_addr_add_multiple(struct netdev_hw_addr_list
*to_list
,
4024 struct netdev_hw_addr_list
*from_list
,
4026 unsigned char addr_type
)
4029 struct netdev_hw_addr
*ha
, *ha2
;
4032 list_for_each_entry(ha
, &from_list
->list
, list
) {
4033 type
= addr_type
? addr_type
: ha
->type
;
4034 err
= __hw_addr_add(to_list
, ha
->addr
, addr_len
, type
);
4041 list_for_each_entry(ha2
, &from_list
->list
, list
) {
4044 type
= addr_type
? addr_type
: ha2
->type
;
4045 __hw_addr_del(to_list
, ha2
->addr
, addr_len
, type
);
4050 static void __hw_addr_del_multiple(struct netdev_hw_addr_list
*to_list
,
4051 struct netdev_hw_addr_list
*from_list
,
4053 unsigned char addr_type
)
4055 struct netdev_hw_addr
*ha
;
4058 list_for_each_entry(ha
, &from_list
->list
, list
) {
4059 type
= addr_type
? addr_type
: ha
->type
;
4060 __hw_addr_del(to_list
, ha
->addr
, addr_len
, addr_type
);
4064 static int __hw_addr_sync(struct netdev_hw_addr_list
*to_list
,
4065 struct netdev_hw_addr_list
*from_list
,
4069 struct netdev_hw_addr
*ha
, *tmp
;
4071 list_for_each_entry_safe(ha
, tmp
, &from_list
->list
, list
) {
4073 err
= __hw_addr_add(to_list
, ha
->addr
,
4074 addr_len
, ha
->type
);
4079 } else if (ha
->refcount
== 1) {
4080 __hw_addr_del(to_list
, ha
->addr
, addr_len
, ha
->type
);
4081 __hw_addr_del(from_list
, ha
->addr
, addr_len
, ha
->type
);
4087 static void __hw_addr_unsync(struct netdev_hw_addr_list
*to_list
,
4088 struct netdev_hw_addr_list
*from_list
,
4091 struct netdev_hw_addr
*ha
, *tmp
;
4093 list_for_each_entry_safe(ha
, tmp
, &from_list
->list
, list
) {
4095 __hw_addr_del(to_list
, ha
->addr
,
4096 addr_len
, ha
->type
);
4098 __hw_addr_del(from_list
, ha
->addr
,
4099 addr_len
, ha
->type
);
4104 static void __hw_addr_flush(struct netdev_hw_addr_list
*list
)
4106 struct netdev_hw_addr
*ha
, *tmp
;
4108 list_for_each_entry_safe(ha
, tmp
, &list
->list
, list
) {
4109 list_del_rcu(&ha
->list
);
4110 call_rcu(&ha
->rcu_head
, ha_rcu_free
);
4115 static void __hw_addr_init(struct netdev_hw_addr_list
*list
)
4117 INIT_LIST_HEAD(&list
->list
);
4121 /* Device addresses handling functions */
4123 static void dev_addr_flush(struct net_device
*dev
)
4125 /* rtnl_mutex must be held here */
4127 __hw_addr_flush(&dev
->dev_addrs
);
4128 dev
->dev_addr
= NULL
;
4131 static int dev_addr_init(struct net_device
*dev
)
4133 unsigned char addr
[MAX_ADDR_LEN
];
4134 struct netdev_hw_addr
*ha
;
4137 /* rtnl_mutex must be held here */
4139 __hw_addr_init(&dev
->dev_addrs
);
4140 memset(addr
, 0, sizeof(addr
));
4141 err
= __hw_addr_add(&dev
->dev_addrs
, addr
, sizeof(addr
),
4142 NETDEV_HW_ADDR_T_LAN
);
4145 * Get the first (previously created) address from the list
4146 * and set dev_addr pointer to this location.
4148 ha
= list_first_entry(&dev
->dev_addrs
.list
,
4149 struct netdev_hw_addr
, list
);
4150 dev
->dev_addr
= ha
->addr
;
4156 * dev_addr_add - Add a device address
4158 * @addr: address to add
4159 * @addr_type: address type
4161 * Add a device address to the device or increase the reference count if
4162 * it already exists.
4164 * The caller must hold the rtnl_mutex.
4166 int dev_addr_add(struct net_device
*dev
, unsigned char *addr
,
4167 unsigned char addr_type
)
4173 err
= __hw_addr_add(&dev
->dev_addrs
, addr
, dev
->addr_len
, addr_type
);
4175 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4178 EXPORT_SYMBOL(dev_addr_add
);
4181 * dev_addr_del - Release a device address.
4183 * @addr: address to delete
4184 * @addr_type: address type
4186 * Release reference to a device address and remove it from the device
4187 * if the reference count drops to zero.
4189 * The caller must hold the rtnl_mutex.
4191 int dev_addr_del(struct net_device
*dev
, unsigned char *addr
,
4192 unsigned char addr_type
)
4195 struct netdev_hw_addr
*ha
;
4200 * We can not remove the first address from the list because
4201 * dev->dev_addr points to that.
4203 ha
= list_first_entry(&dev
->dev_addrs
.list
,
4204 struct netdev_hw_addr
, list
);
4205 if (ha
->addr
== dev
->dev_addr
&& ha
->refcount
== 1)
4208 err
= __hw_addr_del(&dev
->dev_addrs
, addr
, dev
->addr_len
,
4211 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4214 EXPORT_SYMBOL(dev_addr_del
);
4217 * dev_addr_add_multiple - Add device addresses from another device
4218 * @to_dev: device to which addresses will be added
4219 * @from_dev: device from which addresses will be added
4220 * @addr_type: address type - 0 means type will be used from from_dev
4222 * Add device addresses of the one device to another.
4224 * The caller must hold the rtnl_mutex.
4226 int dev_addr_add_multiple(struct net_device
*to_dev
,
4227 struct net_device
*from_dev
,
4228 unsigned char addr_type
)
4234 if (from_dev
->addr_len
!= to_dev
->addr_len
)
4236 err
= __hw_addr_add_multiple(&to_dev
->dev_addrs
, &from_dev
->dev_addrs
,
4237 to_dev
->addr_len
, addr_type
);
4239 call_netdevice_notifiers(NETDEV_CHANGEADDR
, to_dev
);
4242 EXPORT_SYMBOL(dev_addr_add_multiple
);
4245 * dev_addr_del_multiple - Delete device addresses by another device
4246 * @to_dev: device where the addresses will be deleted
4247 * @from_dev: device by which addresses the addresses will be deleted
4248 * @addr_type: address type - 0 means type will used from from_dev
4250 * Deletes addresses in to device by the list of addresses in from device.
4252 * The caller must hold the rtnl_mutex.
4254 int dev_addr_del_multiple(struct net_device
*to_dev
,
4255 struct net_device
*from_dev
,
4256 unsigned char addr_type
)
4260 if (from_dev
->addr_len
!= to_dev
->addr_len
)
4262 __hw_addr_del_multiple(&to_dev
->dev_addrs
, &from_dev
->dev_addrs
,
4263 to_dev
->addr_len
, addr_type
);
4264 call_netdevice_notifiers(NETDEV_CHANGEADDR
, to_dev
);
4267 EXPORT_SYMBOL(dev_addr_del_multiple
);
4269 /* multicast addresses handling functions */
4271 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
4272 void *addr
, int alen
, int glbl
)
4274 struct dev_addr_list
*da
;
4276 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
4277 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
4278 alen
== da
->da_addrlen
) {
4280 int old_glbl
= da
->da_gusers
;
4297 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
4298 void *addr
, int alen
, int glbl
)
4300 struct dev_addr_list
*da
;
4302 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
4303 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
4304 da
->da_addrlen
== alen
) {
4306 int old_glbl
= da
->da_gusers
;
4316 da
= kzalloc(sizeof(*da
), GFP_ATOMIC
);
4319 memcpy(da
->da_addr
, addr
, alen
);
4320 da
->da_addrlen
= alen
;
4322 da
->da_gusers
= glbl
? 1 : 0;
4330 * dev_unicast_delete - Release secondary unicast address.
4332 * @addr: address to delete
4334 * Release reference to a secondary unicast address and remove it
4335 * from the device if the reference count drops to zero.
4337 * The caller must hold the rtnl_mutex.
4339 int dev_unicast_delete(struct net_device
*dev
, void *addr
)
4345 netif_addr_lock_bh(dev
);
4346 err
= __hw_addr_del(&dev
->uc
, addr
, dev
->addr_len
,
4347 NETDEV_HW_ADDR_T_UNICAST
);
4349 __dev_set_rx_mode(dev
);
4350 netif_addr_unlock_bh(dev
);
4353 EXPORT_SYMBOL(dev_unicast_delete
);
4356 * dev_unicast_add - add a secondary unicast address
4358 * @addr: address to add
4360 * Add a secondary unicast address to the device or increase
4361 * the reference count if it already exists.
4363 * The caller must hold the rtnl_mutex.
4365 int dev_unicast_add(struct net_device
*dev
, void *addr
)
4371 netif_addr_lock_bh(dev
);
4372 err
= __hw_addr_add(&dev
->uc
, addr
, dev
->addr_len
,
4373 NETDEV_HW_ADDR_T_UNICAST
);
4375 __dev_set_rx_mode(dev
);
4376 netif_addr_unlock_bh(dev
);
4379 EXPORT_SYMBOL(dev_unicast_add
);
4381 int __dev_addr_sync(struct dev_addr_list
**to
, int *to_count
,
4382 struct dev_addr_list
**from
, int *from_count
)
4384 struct dev_addr_list
*da
, *next
;
4388 while (da
!= NULL
) {
4390 if (!da
->da_synced
) {
4391 err
= __dev_addr_add(to
, to_count
,
4392 da
->da_addr
, da
->da_addrlen
, 0);
4397 } else if (da
->da_users
== 1) {
4398 __dev_addr_delete(to
, to_count
,
4399 da
->da_addr
, da
->da_addrlen
, 0);
4400 __dev_addr_delete(from
, from_count
,
4401 da
->da_addr
, da
->da_addrlen
, 0);
4407 EXPORT_SYMBOL_GPL(__dev_addr_sync
);
4409 void __dev_addr_unsync(struct dev_addr_list
**to
, int *to_count
,
4410 struct dev_addr_list
**from
, int *from_count
)
4412 struct dev_addr_list
*da
, *next
;
4415 while (da
!= NULL
) {
4417 if (da
->da_synced
) {
4418 __dev_addr_delete(to
, to_count
,
4419 da
->da_addr
, da
->da_addrlen
, 0);
4421 __dev_addr_delete(from
, from_count
,
4422 da
->da_addr
, da
->da_addrlen
, 0);
4427 EXPORT_SYMBOL_GPL(__dev_addr_unsync
);
4430 * dev_unicast_sync - Synchronize device's unicast list to another device
4431 * @to: destination device
4432 * @from: source device
4434 * Add newly added addresses to the destination device and release
4435 * addresses that have no users left. The source device must be
4436 * locked by netif_tx_lock_bh.
4438 * This function is intended to be called from the dev->set_rx_mode
4439 * function of layered software devices.
4441 int dev_unicast_sync(struct net_device
*to
, struct net_device
*from
)
4445 if (to
->addr_len
!= from
->addr_len
)
4448 netif_addr_lock_bh(to
);
4449 err
= __hw_addr_sync(&to
->uc
, &from
->uc
, to
->addr_len
);
4451 __dev_set_rx_mode(to
);
4452 netif_addr_unlock_bh(to
);
4455 EXPORT_SYMBOL(dev_unicast_sync
);
4458 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4459 * @to: destination device
4460 * @from: source device
4462 * Remove all addresses that were added to the destination device by
4463 * dev_unicast_sync(). This function is intended to be called from the
4464 * dev->stop function of layered software devices.
4466 void dev_unicast_unsync(struct net_device
*to
, struct net_device
*from
)
4468 if (to
->addr_len
!= from
->addr_len
)
4471 netif_addr_lock_bh(from
);
4472 netif_addr_lock(to
);
4473 __hw_addr_unsync(&to
->uc
, &from
->uc
, to
->addr_len
);
4474 __dev_set_rx_mode(to
);
4475 netif_addr_unlock(to
);
4476 netif_addr_unlock_bh(from
);
4478 EXPORT_SYMBOL(dev_unicast_unsync
);
4480 void dev_unicast_flush(struct net_device
*dev
)
4482 netif_addr_lock_bh(dev
);
4483 __hw_addr_flush(&dev
->uc
);
4484 netif_addr_unlock_bh(dev
);
4486 EXPORT_SYMBOL(dev_unicast_flush
);
4488 static void dev_unicast_init(struct net_device
*dev
)
4490 __hw_addr_init(&dev
->uc
);
4494 static void __dev_addr_discard(struct dev_addr_list
**list
)
4496 struct dev_addr_list
*tmp
;
4498 while (*list
!= NULL
) {
4501 if (tmp
->da_users
> tmp
->da_gusers
)
4502 printk("__dev_addr_discard: address leakage! "
4503 "da_users=%d\n", tmp
->da_users
);
4508 void dev_addr_discard(struct net_device
*dev
)
4510 netif_addr_lock_bh(dev
);
4512 __dev_addr_discard(&dev
->mc_list
);
4513 netdev_mc_count(dev
) = 0;
4515 netif_addr_unlock_bh(dev
);
4517 EXPORT_SYMBOL(dev_addr_discard
);
4520 * dev_get_flags - get flags reported to userspace
4523 * Get the combination of flag bits exported through APIs to userspace.
4525 unsigned dev_get_flags(const struct net_device
*dev
)
4529 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4534 (dev
->gflags
& (IFF_PROMISC
|
4537 if (netif_running(dev
)) {
4538 if (netif_oper_up(dev
))
4539 flags
|= IFF_RUNNING
;
4540 if (netif_carrier_ok(dev
))
4541 flags
|= IFF_LOWER_UP
;
4542 if (netif_dormant(dev
))
4543 flags
|= IFF_DORMANT
;
4548 EXPORT_SYMBOL(dev_get_flags
);
4550 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4552 int old_flags
= dev
->flags
;
4558 * Set the flags on our device.
4561 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4562 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4564 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4568 * Load in the correct multicast list now the flags have changed.
4571 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4572 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4574 dev_set_rx_mode(dev
);
4577 * Have we downed the interface. We handle IFF_UP ourselves
4578 * according to user attempts to set it, rather than blindly
4583 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4584 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4587 dev_set_rx_mode(dev
);
4590 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4591 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4593 dev
->gflags
^= IFF_PROMISC
;
4594 dev_set_promiscuity(dev
, inc
);
4597 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4598 is important. Some (broken) drivers set IFF_PROMISC, when
4599 IFF_ALLMULTI is requested not asking us and not reporting.
4601 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4602 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4604 dev
->gflags
^= IFF_ALLMULTI
;
4605 dev_set_allmulti(dev
, inc
);
4611 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4613 unsigned int changes
= dev
->flags
^ old_flags
;
4615 if (changes
& IFF_UP
) {
4616 if (dev
->flags
& IFF_UP
)
4617 call_netdevice_notifiers(NETDEV_UP
, dev
);
4619 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4622 if (dev
->flags
& IFF_UP
&&
4623 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4624 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4628 * dev_change_flags - change device settings
4630 * @flags: device state flags
4632 * Change settings on device based state flags. The flags are
4633 * in the userspace exported format.
4635 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4638 int old_flags
= dev
->flags
;
4640 ret
= __dev_change_flags(dev
, flags
);
4644 changes
= old_flags
^ dev
->flags
;
4646 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4648 __dev_notify_flags(dev
, old_flags
);
4651 EXPORT_SYMBOL(dev_change_flags
);
4654 * dev_set_mtu - Change maximum transfer unit
4656 * @new_mtu: new transfer unit
4658 * Change the maximum transfer size of the network device.
4660 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4662 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4665 if (new_mtu
== dev
->mtu
)
4668 /* MTU must be positive. */
4672 if (!netif_device_present(dev
))
4676 if (ops
->ndo_change_mtu
)
4677 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4681 if (!err
&& dev
->flags
& IFF_UP
)
4682 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4685 EXPORT_SYMBOL(dev_set_mtu
);
4688 * dev_set_mac_address - Change Media Access Control Address
4692 * Change the hardware (MAC) address of the device
4694 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4696 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4699 if (!ops
->ndo_set_mac_address
)
4701 if (sa
->sa_family
!= dev
->type
)
4703 if (!netif_device_present(dev
))
4705 err
= ops
->ndo_set_mac_address(dev
, sa
);
4707 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4710 EXPORT_SYMBOL(dev_set_mac_address
);
4713 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4715 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4718 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4724 case SIOCGIFFLAGS
: /* Get interface flags */
4725 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4728 case SIOCGIFMETRIC
: /* Get the metric on the interface
4729 (currently unused) */
4730 ifr
->ifr_metric
= 0;
4733 case SIOCGIFMTU
: /* Get the MTU of a device */
4734 ifr
->ifr_mtu
= dev
->mtu
;
4739 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4741 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4742 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4743 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4751 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4752 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4753 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4754 ifr
->ifr_map
.irq
= dev
->irq
;
4755 ifr
->ifr_map
.dma
= dev
->dma
;
4756 ifr
->ifr_map
.port
= dev
->if_port
;
4760 ifr
->ifr_ifindex
= dev
->ifindex
;
4764 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4768 /* dev_ioctl() should ensure this case
4780 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4782 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4785 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4786 const struct net_device_ops
*ops
;
4791 ops
= dev
->netdev_ops
;
4794 case SIOCSIFFLAGS
: /* Set interface flags */
4795 return dev_change_flags(dev
, ifr
->ifr_flags
);
4797 case SIOCSIFMETRIC
: /* Set the metric on the interface
4798 (currently unused) */
4801 case SIOCSIFMTU
: /* Set the MTU of a device */
4802 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4805 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4807 case SIOCSIFHWBROADCAST
:
4808 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4810 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4811 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4812 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4816 if (ops
->ndo_set_config
) {
4817 if (!netif_device_present(dev
))
4819 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4824 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4825 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4827 if (!netif_device_present(dev
))
4829 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
4833 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4834 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4836 if (!netif_device_present(dev
))
4838 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
4842 if (ifr
->ifr_qlen
< 0)
4844 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4848 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4849 return dev_change_name(dev
, ifr
->ifr_newname
);
4852 * Unknown or private ioctl
4855 if ((cmd
>= SIOCDEVPRIVATE
&&
4856 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4857 cmd
== SIOCBONDENSLAVE
||
4858 cmd
== SIOCBONDRELEASE
||
4859 cmd
== SIOCBONDSETHWADDR
||
4860 cmd
== SIOCBONDSLAVEINFOQUERY
||
4861 cmd
== SIOCBONDINFOQUERY
||
4862 cmd
== SIOCBONDCHANGEACTIVE
||
4863 cmd
== SIOCGMIIPHY
||
4864 cmd
== SIOCGMIIREG
||
4865 cmd
== SIOCSMIIREG
||
4866 cmd
== SIOCBRADDIF
||
4867 cmd
== SIOCBRDELIF
||
4868 cmd
== SIOCSHWTSTAMP
||
4869 cmd
== SIOCWANDEV
) {
4871 if (ops
->ndo_do_ioctl
) {
4872 if (netif_device_present(dev
))
4873 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4885 * This function handles all "interface"-type I/O control requests. The actual
4886 * 'doing' part of this is dev_ifsioc above.
4890 * dev_ioctl - network device ioctl
4891 * @net: the applicable net namespace
4892 * @cmd: command to issue
4893 * @arg: pointer to a struct ifreq in user space
4895 * Issue ioctl functions to devices. This is normally called by the
4896 * user space syscall interfaces but can sometimes be useful for
4897 * other purposes. The return value is the return from the syscall if
4898 * positive or a negative errno code on error.
4901 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4907 /* One special case: SIOCGIFCONF takes ifconf argument
4908 and requires shared lock, because it sleeps writing
4912 if (cmd
== SIOCGIFCONF
) {
4914 ret
= dev_ifconf(net
, (char __user
*) arg
);
4918 if (cmd
== SIOCGIFNAME
)
4919 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4921 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4924 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4926 colon
= strchr(ifr
.ifr_name
, ':');
4931 * See which interface the caller is talking about.
4936 * These ioctl calls:
4937 * - can be done by all.
4938 * - atomic and do not require locking.
4949 dev_load(net
, ifr
.ifr_name
);
4951 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4956 if (copy_to_user(arg
, &ifr
,
4957 sizeof(struct ifreq
)))
4963 dev_load(net
, ifr
.ifr_name
);
4965 ret
= dev_ethtool(net
, &ifr
);
4970 if (copy_to_user(arg
, &ifr
,
4971 sizeof(struct ifreq
)))
4977 * These ioctl calls:
4978 * - require superuser power.
4979 * - require strict serialization.
4985 if (!capable(CAP_NET_ADMIN
))
4987 dev_load(net
, ifr
.ifr_name
);
4989 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4994 if (copy_to_user(arg
, &ifr
,
4995 sizeof(struct ifreq
)))
5001 * These ioctl calls:
5002 * - require superuser power.
5003 * - require strict serialization.
5004 * - do not return a value
5014 case SIOCSIFHWBROADCAST
:
5017 case SIOCBONDENSLAVE
:
5018 case SIOCBONDRELEASE
:
5019 case SIOCBONDSETHWADDR
:
5020 case SIOCBONDCHANGEACTIVE
:
5024 if (!capable(CAP_NET_ADMIN
))
5027 case SIOCBONDSLAVEINFOQUERY
:
5028 case SIOCBONDINFOQUERY
:
5029 dev_load(net
, ifr
.ifr_name
);
5031 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5036 /* Get the per device memory space. We can add this but
5037 * currently do not support it */
5039 /* Set the per device memory buffer space.
5040 * Not applicable in our case */
5045 * Unknown or private ioctl.
5048 if (cmd
== SIOCWANDEV
||
5049 (cmd
>= SIOCDEVPRIVATE
&&
5050 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5051 dev_load(net
, ifr
.ifr_name
);
5053 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5055 if (!ret
&& copy_to_user(arg
, &ifr
,
5056 sizeof(struct ifreq
)))
5060 /* Take care of Wireless Extensions */
5061 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5062 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5069 * dev_new_index - allocate an ifindex
5070 * @net: the applicable net namespace
5072 * Returns a suitable unique value for a new device interface
5073 * number. The caller must hold the rtnl semaphore or the
5074 * dev_base_lock to be sure it remains unique.
5076 static int dev_new_index(struct net
*net
)
5082 if (!__dev_get_by_index(net
, ifindex
))
5087 /* Delayed registration/unregisteration */
5088 static LIST_HEAD(net_todo_list
);
5090 static void net_set_todo(struct net_device
*dev
)
5092 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5095 static void rollback_registered_many(struct list_head
*head
)
5097 struct net_device
*dev
, *tmp
;
5099 BUG_ON(dev_boot_phase
);
5102 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5103 /* Some devices call without registering
5104 * for initialization unwind. Remove those
5105 * devices and proceed with the remaining.
5107 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5108 pr_debug("unregister_netdevice: device %s/%p never "
5109 "was registered\n", dev
->name
, dev
);
5112 list_del(&dev
->unreg_list
);
5116 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5118 /* If device is running, close it first. */
5121 /* And unlink it from device chain. */
5122 unlist_netdevice(dev
);
5124 dev
->reg_state
= NETREG_UNREGISTERING
;
5129 list_for_each_entry(dev
, head
, unreg_list
) {
5130 /* Shutdown queueing discipline. */
5134 /* Notify protocols, that we are about to destroy
5135 this device. They should clean all the things.
5137 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5139 if (!dev
->rtnl_link_ops
||
5140 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5141 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5144 * Flush the unicast and multicast chains
5146 dev_unicast_flush(dev
);
5147 dev_addr_discard(dev
);
5149 if (dev
->netdev_ops
->ndo_uninit
)
5150 dev
->netdev_ops
->ndo_uninit(dev
);
5152 /* Notifier chain MUST detach us from master device. */
5153 WARN_ON(dev
->master
);
5155 /* Remove entries from kobject tree */
5156 netdev_unregister_kobject(dev
);
5159 /* Process any work delayed until the end of the batch */
5160 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5161 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5165 list_for_each_entry(dev
, head
, unreg_list
)
5169 static void rollback_registered(struct net_device
*dev
)
5173 list_add(&dev
->unreg_list
, &single
);
5174 rollback_registered_many(&single
);
5177 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
5178 struct netdev_queue
*dev_queue
,
5181 spin_lock_init(&dev_queue
->_xmit_lock
);
5182 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
5183 dev_queue
->xmit_lock_owner
= -1;
5186 static void netdev_init_queue_locks(struct net_device
*dev
)
5188 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
5189 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
5192 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
5194 /* Fix illegal SG+CSUM combinations. */
5195 if ((features
& NETIF_F_SG
) &&
5196 !(features
& NETIF_F_ALL_CSUM
)) {
5198 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
5199 "checksum feature.\n", name
);
5200 features
&= ~NETIF_F_SG
;
5203 /* TSO requires that SG is present as well. */
5204 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
5206 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
5207 "SG feature.\n", name
);
5208 features
&= ~NETIF_F_TSO
;
5211 if (features
& NETIF_F_UFO
) {
5212 if (!(features
& NETIF_F_GEN_CSUM
)) {
5214 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
5215 "since no NETIF_F_HW_CSUM feature.\n",
5217 features
&= ~NETIF_F_UFO
;
5220 if (!(features
& NETIF_F_SG
)) {
5222 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
5223 "since no NETIF_F_SG feature.\n", name
);
5224 features
&= ~NETIF_F_UFO
;
5230 EXPORT_SYMBOL(netdev_fix_features
);
5233 * netif_stacked_transfer_operstate - transfer operstate
5234 * @rootdev: the root or lower level device to transfer state from
5235 * @dev: the device to transfer operstate to
5237 * Transfer operational state from root to device. This is normally
5238 * called when a stacking relationship exists between the root
5239 * device and the device(a leaf device).
5241 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5242 struct net_device
*dev
)
5244 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5245 netif_dormant_on(dev
);
5247 netif_dormant_off(dev
);
5249 if (netif_carrier_ok(rootdev
)) {
5250 if (!netif_carrier_ok(dev
))
5251 netif_carrier_on(dev
);
5253 if (netif_carrier_ok(dev
))
5254 netif_carrier_off(dev
);
5257 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5260 * register_netdevice - register a network device
5261 * @dev: device to register
5263 * Take a completed network device structure and add it to the kernel
5264 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5265 * chain. 0 is returned on success. A negative errno code is returned
5266 * on a failure to set up the device, or if the name is a duplicate.
5268 * Callers must hold the rtnl semaphore. You may want
5269 * register_netdev() instead of this.
5272 * The locking appears insufficient to guarantee two parallel registers
5273 * will not get the same name.
5276 int register_netdevice(struct net_device
*dev
)
5279 struct net
*net
= dev_net(dev
);
5281 BUG_ON(dev_boot_phase
);
5286 /* When net_device's are persistent, this will be fatal. */
5287 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5290 spin_lock_init(&dev
->addr_list_lock
);
5291 netdev_set_addr_lockdep_class(dev
);
5292 netdev_init_queue_locks(dev
);
5297 if (!dev
->num_rx_queues
) {
5299 * Allocate a single RX queue if driver never called
5303 dev
->_rx
= kzalloc(sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5309 dev
->_rx
->first
= dev
->_rx
;
5310 atomic_set(&dev
->_rx
->count
, 1);
5311 dev
->num_rx_queues
= 1;
5314 /* Init, if this function is available */
5315 if (dev
->netdev_ops
->ndo_init
) {
5316 ret
= dev
->netdev_ops
->ndo_init(dev
);
5324 ret
= dev_get_valid_name(net
, dev
->name
, dev
->name
, 0);
5328 dev
->ifindex
= dev_new_index(net
);
5329 if (dev
->iflink
== -1)
5330 dev
->iflink
= dev
->ifindex
;
5332 /* Fix illegal checksum combinations */
5333 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
5334 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5335 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5337 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5340 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5341 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5342 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5344 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5347 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5349 /* Enable software GSO if SG is supported. */
5350 if (dev
->features
& NETIF_F_SG
)
5351 dev
->features
|= NETIF_F_GSO
;
5353 netdev_initialize_kobject(dev
);
5355 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5356 ret
= notifier_to_errno(ret
);
5360 ret
= netdev_register_kobject(dev
);
5363 dev
->reg_state
= NETREG_REGISTERED
;
5366 * Default initial state at registry is that the
5367 * device is present.
5370 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5372 dev_init_scheduler(dev
);
5374 list_netdevice(dev
);
5376 /* Notify protocols, that a new device appeared. */
5377 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5378 ret
= notifier_to_errno(ret
);
5380 rollback_registered(dev
);
5381 dev
->reg_state
= NETREG_UNREGISTERED
;
5384 * Prevent userspace races by waiting until the network
5385 * device is fully setup before sending notifications.
5387 if (!dev
->rtnl_link_ops
||
5388 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5389 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5395 if (dev
->netdev_ops
->ndo_uninit
)
5396 dev
->netdev_ops
->ndo_uninit(dev
);
5399 EXPORT_SYMBOL(register_netdevice
);
5402 * init_dummy_netdev - init a dummy network device for NAPI
5403 * @dev: device to init
5405 * This takes a network device structure and initialize the minimum
5406 * amount of fields so it can be used to schedule NAPI polls without
5407 * registering a full blown interface. This is to be used by drivers
5408 * that need to tie several hardware interfaces to a single NAPI
5409 * poll scheduler due to HW limitations.
5411 int init_dummy_netdev(struct net_device
*dev
)
5413 /* Clear everything. Note we don't initialize spinlocks
5414 * are they aren't supposed to be taken by any of the
5415 * NAPI code and this dummy netdev is supposed to be
5416 * only ever used for NAPI polls
5418 memset(dev
, 0, sizeof(struct net_device
));
5420 /* make sure we BUG if trying to hit standard
5421 * register/unregister code path
5423 dev
->reg_state
= NETREG_DUMMY
;
5425 /* initialize the ref count */
5426 atomic_set(&dev
->refcnt
, 1);
5428 /* NAPI wants this */
5429 INIT_LIST_HEAD(&dev
->napi_list
);
5431 /* a dummy interface is started by default */
5432 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5433 set_bit(__LINK_STATE_START
, &dev
->state
);
5437 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5441 * register_netdev - register a network device
5442 * @dev: device to register
5444 * Take a completed network device structure and add it to the kernel
5445 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5446 * chain. 0 is returned on success. A negative errno code is returned
5447 * on a failure to set up the device, or if the name is a duplicate.
5449 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5450 * and expands the device name if you passed a format string to
5453 int register_netdev(struct net_device
*dev
)
5460 * If the name is a format string the caller wants us to do a
5463 if (strchr(dev
->name
, '%')) {
5464 err
= dev_alloc_name(dev
, dev
->name
);
5469 err
= register_netdevice(dev
);
5474 EXPORT_SYMBOL(register_netdev
);
5477 * netdev_wait_allrefs - wait until all references are gone.
5479 * This is called when unregistering network devices.
5481 * Any protocol or device that holds a reference should register
5482 * for netdevice notification, and cleanup and put back the
5483 * reference if they receive an UNREGISTER event.
5484 * We can get stuck here if buggy protocols don't correctly
5487 static void netdev_wait_allrefs(struct net_device
*dev
)
5489 unsigned long rebroadcast_time
, warning_time
;
5491 linkwatch_forget_dev(dev
);
5493 rebroadcast_time
= warning_time
= jiffies
;
5494 while (atomic_read(&dev
->refcnt
) != 0) {
5495 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5498 /* Rebroadcast unregister notification */
5499 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5500 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5501 * should have already handle it the first time */
5503 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5505 /* We must not have linkwatch events
5506 * pending on unregister. If this
5507 * happens, we simply run the queue
5508 * unscheduled, resulting in a noop
5511 linkwatch_run_queue();
5516 rebroadcast_time
= jiffies
;
5521 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5522 printk(KERN_EMERG
"unregister_netdevice: "
5523 "waiting for %s to become free. Usage "
5525 dev
->name
, atomic_read(&dev
->refcnt
));
5526 warning_time
= jiffies
;
5535 * register_netdevice(x1);
5536 * register_netdevice(x2);
5538 * unregister_netdevice(y1);
5539 * unregister_netdevice(y2);
5545 * We are invoked by rtnl_unlock().
5546 * This allows us to deal with problems:
5547 * 1) We can delete sysfs objects which invoke hotplug
5548 * without deadlocking with linkwatch via keventd.
5549 * 2) Since we run with the RTNL semaphore not held, we can sleep
5550 * safely in order to wait for the netdev refcnt to drop to zero.
5552 * We must not return until all unregister events added during
5553 * the interval the lock was held have been completed.
5555 void netdev_run_todo(void)
5557 struct list_head list
;
5559 /* Snapshot list, allow later requests */
5560 list_replace_init(&net_todo_list
, &list
);
5564 while (!list_empty(&list
)) {
5565 struct net_device
*dev
5566 = list_first_entry(&list
, struct net_device
, todo_list
);
5567 list_del(&dev
->todo_list
);
5569 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5570 printk(KERN_ERR
"network todo '%s' but state %d\n",
5571 dev
->name
, dev
->reg_state
);
5576 dev
->reg_state
= NETREG_UNREGISTERED
;
5578 on_each_cpu(flush_backlog
, dev
, 1);
5580 netdev_wait_allrefs(dev
);
5583 BUG_ON(atomic_read(&dev
->refcnt
));
5584 WARN_ON(dev
->ip_ptr
);
5585 WARN_ON(dev
->ip6_ptr
);
5586 WARN_ON(dev
->dn_ptr
);
5588 if (dev
->destructor
)
5589 dev
->destructor(dev
);
5591 /* Free network device */
5592 kobject_put(&dev
->dev
.kobj
);
5597 * dev_txq_stats_fold - fold tx_queues stats
5598 * @dev: device to get statistics from
5599 * @stats: struct net_device_stats to hold results
5601 void dev_txq_stats_fold(const struct net_device
*dev
,
5602 struct net_device_stats
*stats
)
5604 unsigned long tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5606 struct netdev_queue
*txq
;
5608 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5609 txq
= netdev_get_tx_queue(dev
, i
);
5610 tx_bytes
+= txq
->tx_bytes
;
5611 tx_packets
+= txq
->tx_packets
;
5612 tx_dropped
+= txq
->tx_dropped
;
5614 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5615 stats
->tx_bytes
= tx_bytes
;
5616 stats
->tx_packets
= tx_packets
;
5617 stats
->tx_dropped
= tx_dropped
;
5620 EXPORT_SYMBOL(dev_txq_stats_fold
);
5623 * dev_get_stats - get network device statistics
5624 * @dev: device to get statistics from
5626 * Get network statistics from device. The device driver may provide
5627 * its own method by setting dev->netdev_ops->get_stats; otherwise
5628 * the internal statistics structure is used.
5630 const struct net_device_stats
*dev_get_stats(struct net_device
*dev
)
5632 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5634 if (ops
->ndo_get_stats
)
5635 return ops
->ndo_get_stats(dev
);
5637 dev_txq_stats_fold(dev
, &dev
->stats
);
5640 EXPORT_SYMBOL(dev_get_stats
);
5642 static void netdev_init_one_queue(struct net_device
*dev
,
5643 struct netdev_queue
*queue
,
5649 static void netdev_init_queues(struct net_device
*dev
)
5651 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
5652 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5653 spin_lock_init(&dev
->tx_global_lock
);
5657 * alloc_netdev_mq - allocate network device
5658 * @sizeof_priv: size of private data to allocate space for
5659 * @name: device name format string
5660 * @setup: callback to initialize device
5661 * @queue_count: the number of subqueues to allocate
5663 * Allocates a struct net_device with private data area for driver use
5664 * and performs basic initialization. Also allocates subquue structs
5665 * for each queue on the device at the end of the netdevice.
5667 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
5668 void (*setup
)(struct net_device
*), unsigned int queue_count
)
5670 struct netdev_queue
*tx
;
5671 struct net_device
*dev
;
5673 struct net_device
*p
;
5675 struct netdev_rx_queue
*rx
;
5679 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5681 alloc_size
= sizeof(struct net_device
);
5683 /* ensure 32-byte alignment of private area */
5684 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5685 alloc_size
+= sizeof_priv
;
5687 /* ensure 32-byte alignment of whole construct */
5688 alloc_size
+= NETDEV_ALIGN
- 1;
5690 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5692 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5696 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5698 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5704 rx
= kcalloc(queue_count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5706 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5711 atomic_set(&rx
->count
, queue_count
);
5714 * Set a pointer to first element in the array which holds the
5717 for (i
= 0; i
< queue_count
; i
++)
5721 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5722 dev
->padded
= (char *)dev
- (char *)p
;
5724 if (dev_addr_init(dev
))
5727 dev_unicast_init(dev
);
5729 dev_net_set(dev
, &init_net
);
5732 dev
->num_tx_queues
= queue_count
;
5733 dev
->real_num_tx_queues
= queue_count
;
5737 dev
->num_rx_queues
= queue_count
;
5740 dev
->gso_max_size
= GSO_MAX_SIZE
;
5742 netdev_init_queues(dev
);
5744 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5745 dev
->ethtool_ntuple_list
.count
= 0;
5746 INIT_LIST_HEAD(&dev
->napi_list
);
5747 INIT_LIST_HEAD(&dev
->unreg_list
);
5748 INIT_LIST_HEAD(&dev
->link_watch_list
);
5749 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5751 strcpy(dev
->name
, name
);
5764 EXPORT_SYMBOL(alloc_netdev_mq
);
5767 * free_netdev - free network device
5770 * This function does the last stage of destroying an allocated device
5771 * interface. The reference to the device object is released.
5772 * If this is the last reference then it will be freed.
5774 void free_netdev(struct net_device
*dev
)
5776 struct napi_struct
*p
, *n
;
5778 release_net(dev_net(dev
));
5782 /* Flush device addresses */
5783 dev_addr_flush(dev
);
5785 /* Clear ethtool n-tuple list */
5786 ethtool_ntuple_flush(dev
);
5788 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5791 /* Compatibility with error handling in drivers */
5792 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5793 kfree((char *)dev
- dev
->padded
);
5797 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5798 dev
->reg_state
= NETREG_RELEASED
;
5800 /* will free via device release */
5801 put_device(&dev
->dev
);
5803 EXPORT_SYMBOL(free_netdev
);
5806 * synchronize_net - Synchronize with packet receive processing
5808 * Wait for packets currently being received to be done.
5809 * Does not block later packets from starting.
5811 void synchronize_net(void)
5816 EXPORT_SYMBOL(synchronize_net
);
5819 * unregister_netdevice_queue - remove device from the kernel
5823 * This function shuts down a device interface and removes it
5824 * from the kernel tables.
5825 * If head not NULL, device is queued to be unregistered later.
5827 * Callers must hold the rtnl semaphore. You may want
5828 * unregister_netdev() instead of this.
5831 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5836 list_move_tail(&dev
->unreg_list
, head
);
5838 rollback_registered(dev
);
5839 /* Finish processing unregister after unlock */
5843 EXPORT_SYMBOL(unregister_netdevice_queue
);
5846 * unregister_netdevice_many - unregister many devices
5847 * @head: list of devices
5849 void unregister_netdevice_many(struct list_head
*head
)
5851 struct net_device
*dev
;
5853 if (!list_empty(head
)) {
5854 rollback_registered_many(head
);
5855 list_for_each_entry(dev
, head
, unreg_list
)
5859 EXPORT_SYMBOL(unregister_netdevice_many
);
5862 * unregister_netdev - remove device from the kernel
5865 * This function shuts down a device interface and removes it
5866 * from the kernel tables.
5868 * This is just a wrapper for unregister_netdevice that takes
5869 * the rtnl semaphore. In general you want to use this and not
5870 * unregister_netdevice.
5872 void unregister_netdev(struct net_device
*dev
)
5875 unregister_netdevice(dev
);
5878 EXPORT_SYMBOL(unregister_netdev
);
5881 * dev_change_net_namespace - move device to different nethost namespace
5883 * @net: network namespace
5884 * @pat: If not NULL name pattern to try if the current device name
5885 * is already taken in the destination network namespace.
5887 * This function shuts down a device interface and moves it
5888 * to a new network namespace. On success 0 is returned, on
5889 * a failure a netagive errno code is returned.
5891 * Callers must hold the rtnl semaphore.
5894 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5900 /* Don't allow namespace local devices to be moved. */
5902 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5906 /* Don't allow real devices to be moved when sysfs
5910 if (dev
->dev
.parent
)
5914 /* Ensure the device has been registrered */
5916 if (dev
->reg_state
!= NETREG_REGISTERED
)
5919 /* Get out if there is nothing todo */
5921 if (net_eq(dev_net(dev
), net
))
5924 /* Pick the destination device name, and ensure
5925 * we can use it in the destination network namespace.
5928 if (__dev_get_by_name(net
, dev
->name
)) {
5929 /* We get here if we can't use the current device name */
5932 if (dev_get_valid_name(net
, pat
, dev
->name
, 1))
5937 * And now a mini version of register_netdevice unregister_netdevice.
5940 /* If device is running close it first. */
5943 /* And unlink it from device chain */
5945 unlist_netdevice(dev
);
5949 /* Shutdown queueing discipline. */
5952 /* Notify protocols, that we are about to destroy
5953 this device. They should clean all the things.
5955 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5956 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5959 * Flush the unicast and multicast chains
5961 dev_unicast_flush(dev
);
5962 dev_addr_discard(dev
);
5964 netdev_unregister_kobject(dev
);
5966 /* Actually switch the network namespace */
5967 dev_net_set(dev
, net
);
5969 /* If there is an ifindex conflict assign a new one */
5970 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5971 int iflink
= (dev
->iflink
== dev
->ifindex
);
5972 dev
->ifindex
= dev_new_index(net
);
5974 dev
->iflink
= dev
->ifindex
;
5977 /* Fixup kobjects */
5978 err
= netdev_register_kobject(dev
);
5981 /* Add the device back in the hashes */
5982 list_netdevice(dev
);
5984 /* Notify protocols, that a new device appeared. */
5985 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5988 * Prevent userspace races by waiting until the network
5989 * device is fully setup before sending notifications.
5991 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5998 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6000 static int dev_cpu_callback(struct notifier_block
*nfb
,
6001 unsigned long action
,
6004 struct sk_buff
**list_skb
;
6005 struct Qdisc
**list_net
;
6006 struct sk_buff
*skb
;
6007 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6008 struct softnet_data
*sd
, *oldsd
;
6010 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6013 local_irq_disable();
6014 cpu
= smp_processor_id();
6015 sd
= &per_cpu(softnet_data
, cpu
);
6016 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6018 /* Find end of our completion_queue. */
6019 list_skb
= &sd
->completion_queue
;
6021 list_skb
= &(*list_skb
)->next
;
6022 /* Append completion queue from offline CPU. */
6023 *list_skb
= oldsd
->completion_queue
;
6024 oldsd
->completion_queue
= NULL
;
6026 /* Find end of our output_queue. */
6027 list_net
= &sd
->output_queue
;
6029 list_net
= &(*list_net
)->next_sched
;
6030 /* Append output queue from offline CPU. */
6031 *list_net
= oldsd
->output_queue
;
6032 oldsd
->output_queue
= NULL
;
6034 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6037 /* Process offline CPU's input_pkt_queue */
6038 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
6046 * netdev_increment_features - increment feature set by one
6047 * @all: current feature set
6048 * @one: new feature set
6049 * @mask: mask feature set
6051 * Computes a new feature set after adding a device with feature set
6052 * @one to the master device with current feature set @all. Will not
6053 * enable anything that is off in @mask. Returns the new feature set.
6055 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
6058 /* If device needs checksumming, downgrade to it. */
6059 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
6060 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
6061 else if (mask
& NETIF_F_ALL_CSUM
) {
6062 /* If one device supports v4/v6 checksumming, set for all. */
6063 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
6064 !(all
& NETIF_F_GEN_CSUM
)) {
6065 all
&= ~NETIF_F_ALL_CSUM
;
6066 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
6069 /* If one device supports hw checksumming, set for all. */
6070 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
6071 all
&= ~NETIF_F_ALL_CSUM
;
6072 all
|= NETIF_F_HW_CSUM
;
6076 one
|= NETIF_F_ALL_CSUM
;
6078 one
|= all
& NETIF_F_ONE_FOR_ALL
;
6079 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
6080 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
6084 EXPORT_SYMBOL(netdev_increment_features
);
6086 static struct hlist_head
*netdev_create_hash(void)
6089 struct hlist_head
*hash
;
6091 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6093 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6094 INIT_HLIST_HEAD(&hash
[i
]);
6099 /* Initialize per network namespace state */
6100 static int __net_init
netdev_init(struct net
*net
)
6102 INIT_LIST_HEAD(&net
->dev_base_head
);
6104 net
->dev_name_head
= netdev_create_hash();
6105 if (net
->dev_name_head
== NULL
)
6108 net
->dev_index_head
= netdev_create_hash();
6109 if (net
->dev_index_head
== NULL
)
6115 kfree(net
->dev_name_head
);
6121 * netdev_drivername - network driver for the device
6122 * @dev: network device
6123 * @buffer: buffer for resulting name
6124 * @len: size of buffer
6126 * Determine network driver for device.
6128 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
6130 const struct device_driver
*driver
;
6131 const struct device
*parent
;
6133 if (len
<= 0 || !buffer
)
6137 parent
= dev
->dev
.parent
;
6142 driver
= parent
->driver
;
6143 if (driver
&& driver
->name
)
6144 strlcpy(buffer
, driver
->name
, len
);
6148 static void __net_exit
netdev_exit(struct net
*net
)
6150 kfree(net
->dev_name_head
);
6151 kfree(net
->dev_index_head
);
6154 static struct pernet_operations __net_initdata netdev_net_ops
= {
6155 .init
= netdev_init
,
6156 .exit
= netdev_exit
,
6159 static void __net_exit
default_device_exit(struct net
*net
)
6161 struct net_device
*dev
, *aux
;
6163 * Push all migratable network devices back to the
6164 * initial network namespace
6167 for_each_netdev_safe(net
, dev
, aux
) {
6169 char fb_name
[IFNAMSIZ
];
6171 /* Ignore unmoveable devices (i.e. loopback) */
6172 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6175 /* Leave virtual devices for the generic cleanup */
6176 if (dev
->rtnl_link_ops
)
6179 /* Push remaing network devices to init_net */
6180 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6181 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6183 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
6184 __func__
, dev
->name
, err
);
6191 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6193 /* At exit all network devices most be removed from a network
6194 * namespace. Do this in the reverse order of registeration.
6195 * Do this across as many network namespaces as possible to
6196 * improve batching efficiency.
6198 struct net_device
*dev
;
6200 LIST_HEAD(dev_kill_list
);
6203 list_for_each_entry(net
, net_list
, exit_list
) {
6204 for_each_netdev_reverse(net
, dev
) {
6205 if (dev
->rtnl_link_ops
)
6206 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6208 unregister_netdevice_queue(dev
, &dev_kill_list
);
6211 unregister_netdevice_many(&dev_kill_list
);
6215 static struct pernet_operations __net_initdata default_device_ops
= {
6216 .exit
= default_device_exit
,
6217 .exit_batch
= default_device_exit_batch
,
6221 * Initialize the DEV module. At boot time this walks the device list and
6222 * unhooks any devices that fail to initialise (normally hardware not
6223 * present) and leaves us with a valid list of present and active devices.
6228 * This is called single threaded during boot, so no need
6229 * to take the rtnl semaphore.
6231 static int __init
net_dev_init(void)
6233 int i
, rc
= -ENOMEM
;
6235 BUG_ON(!dev_boot_phase
);
6237 if (dev_proc_init())
6240 if (netdev_kobject_init())
6243 INIT_LIST_HEAD(&ptype_all
);
6244 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6245 INIT_LIST_HEAD(&ptype_base
[i
]);
6247 if (register_pernet_subsys(&netdev_net_ops
))
6251 * Initialise the packet receive queues.
6254 for_each_possible_cpu(i
) {
6255 struct softnet_data
*queue
;
6257 queue
= &per_cpu(softnet_data
, i
);
6258 skb_queue_head_init(&queue
->input_pkt_queue
);
6259 queue
->completion_queue
= NULL
;
6260 INIT_LIST_HEAD(&queue
->poll_list
);
6263 queue
->csd
.func
= trigger_softirq
;
6264 queue
->csd
.info
= queue
;
6265 queue
->csd
.flags
= 0;
6268 queue
->backlog
.poll
= process_backlog
;
6269 queue
->backlog
.weight
= weight_p
;
6270 queue
->backlog
.gro_list
= NULL
;
6271 queue
->backlog
.gro_count
= 0;
6276 /* The loopback device is special if any other network devices
6277 * is present in a network namespace the loopback device must
6278 * be present. Since we now dynamically allocate and free the
6279 * loopback device ensure this invariant is maintained by
6280 * keeping the loopback device as the first device on the
6281 * list of network devices. Ensuring the loopback devices
6282 * is the first device that appears and the last network device
6285 if (register_pernet_device(&loopback_net_ops
))
6288 if (register_pernet_device(&default_device_ops
))
6291 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6292 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6294 hotcpu_notifier(dev_cpu_callback
, 0);
6302 subsys_initcall(net_dev_init
);
6304 static int __init
initialize_hashrnd(void)
6306 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6310 late_initcall_sync(initialize_hashrnd
);