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 /* Device list insertion */
210 static int list_netdevice(struct net_device
*dev
)
212 struct net
*net
= dev_net(dev
);
216 write_lock_bh(&dev_base_lock
);
217 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
218 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
219 hlist_add_head_rcu(&dev
->index_hlist
,
220 dev_index_hash(net
, dev
->ifindex
));
221 write_unlock_bh(&dev_base_lock
);
225 /* Device list removal
226 * caller must respect a RCU grace period before freeing/reusing dev
228 static void unlist_netdevice(struct net_device
*dev
)
232 /* Unlink dev from the device chain */
233 write_lock_bh(&dev_base_lock
);
234 list_del_rcu(&dev
->dev_list
);
235 hlist_del_rcu(&dev
->name_hlist
);
236 hlist_del_rcu(&dev
->index_hlist
);
237 write_unlock_bh(&dev_base_lock
);
244 static RAW_NOTIFIER_HEAD(netdev_chain
);
247 * Device drivers call our routines to queue packets here. We empty the
248 * queue in the local softnet handler.
251 DEFINE_PER_CPU(struct softnet_data
, softnet_data
);
252 EXPORT_PER_CPU_SYMBOL(softnet_data
);
254 #ifdef CONFIG_LOCKDEP
256 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
257 * according to dev->type
259 static const unsigned short netdev_lock_type
[] =
260 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
261 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
262 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
263 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
264 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
265 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
266 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
267 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
268 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
269 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
270 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
271 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
272 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
273 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
274 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
275 ARPHRD_VOID
, ARPHRD_NONE
};
277 static const char *const netdev_lock_name
[] =
278 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
279 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
280 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
281 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
282 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
283 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
284 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
285 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
286 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
287 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
288 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
289 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
290 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
291 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
292 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
293 "_xmit_VOID", "_xmit_NONE"};
295 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
296 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
298 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
302 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
303 if (netdev_lock_type
[i
] == dev_type
)
305 /* the last key is used by default */
306 return ARRAY_SIZE(netdev_lock_type
) - 1;
309 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
310 unsigned short dev_type
)
314 i
= netdev_lock_pos(dev_type
);
315 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
316 netdev_lock_name
[i
]);
319 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
323 i
= netdev_lock_pos(dev
->type
);
324 lockdep_set_class_and_name(&dev
->addr_list_lock
,
325 &netdev_addr_lock_key
[i
],
326 netdev_lock_name
[i
]);
329 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
330 unsigned short dev_type
)
333 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
338 /*******************************************************************************
340 Protocol management and registration routines
342 *******************************************************************************/
345 * Add a protocol ID to the list. Now that the input handler is
346 * smarter we can dispense with all the messy stuff that used to be
349 * BEWARE!!! Protocol handlers, mangling input packets,
350 * MUST BE last in hash buckets and checking protocol handlers
351 * MUST start from promiscuous ptype_all chain in net_bh.
352 * It is true now, do not change it.
353 * Explanation follows: if protocol handler, mangling packet, will
354 * be the first on list, it is not able to sense, that packet
355 * is cloned and should be copied-on-write, so that it will
356 * change it and subsequent readers will get broken packet.
361 * dev_add_pack - add packet handler
362 * @pt: packet type declaration
364 * Add a protocol handler to the networking stack. The passed &packet_type
365 * is linked into kernel lists and may not be freed until it has been
366 * removed from the kernel lists.
368 * This call does not sleep therefore it can not
369 * guarantee all CPU's that are in middle of receiving packets
370 * will see the new packet type (until the next received packet).
373 void dev_add_pack(struct packet_type
*pt
)
377 spin_lock_bh(&ptype_lock
);
378 if (pt
->type
== htons(ETH_P_ALL
))
379 list_add_rcu(&pt
->list
, &ptype_all
);
381 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
382 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
384 spin_unlock_bh(&ptype_lock
);
386 EXPORT_SYMBOL(dev_add_pack
);
389 * __dev_remove_pack - remove packet handler
390 * @pt: packet type declaration
392 * Remove a protocol handler that was previously added to the kernel
393 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
394 * from the kernel lists and can be freed or reused once this function
397 * The packet type might still be in use by receivers
398 * and must not be freed until after all the CPU's have gone
399 * through a quiescent state.
401 void __dev_remove_pack(struct packet_type
*pt
)
403 struct list_head
*head
;
404 struct packet_type
*pt1
;
406 spin_lock_bh(&ptype_lock
);
408 if (pt
->type
== htons(ETH_P_ALL
))
411 head
= &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
413 list_for_each_entry(pt1
, head
, list
) {
415 list_del_rcu(&pt
->list
);
420 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
422 spin_unlock_bh(&ptype_lock
);
424 EXPORT_SYMBOL(__dev_remove_pack
);
427 * dev_remove_pack - remove packet handler
428 * @pt: packet type declaration
430 * Remove a protocol handler that was previously added to the kernel
431 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
432 * from the kernel lists and can be freed or reused once this function
435 * This call sleeps to guarantee that no CPU is looking at the packet
438 void dev_remove_pack(struct packet_type
*pt
)
440 __dev_remove_pack(pt
);
444 EXPORT_SYMBOL(dev_remove_pack
);
446 /******************************************************************************
448 Device Boot-time Settings Routines
450 *******************************************************************************/
452 /* Boot time configuration table */
453 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
456 * netdev_boot_setup_add - add new setup entry
457 * @name: name of the device
458 * @map: configured settings for the device
460 * Adds new setup entry to the dev_boot_setup list. The function
461 * returns 0 on error and 1 on success. This is a generic routine to
464 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
466 struct netdev_boot_setup
*s
;
470 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
471 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
472 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
473 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
474 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
479 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
483 * netdev_boot_setup_check - check boot time settings
484 * @dev: the netdevice
486 * Check boot time settings for the device.
487 * The found settings are set for the device to be used
488 * later in the device probing.
489 * Returns 0 if no settings found, 1 if they are.
491 int netdev_boot_setup_check(struct net_device
*dev
)
493 struct netdev_boot_setup
*s
= dev_boot_setup
;
496 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
497 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
498 !strcmp(dev
->name
, s
[i
].name
)) {
499 dev
->irq
= s
[i
].map
.irq
;
500 dev
->base_addr
= s
[i
].map
.base_addr
;
501 dev
->mem_start
= s
[i
].map
.mem_start
;
502 dev
->mem_end
= s
[i
].map
.mem_end
;
508 EXPORT_SYMBOL(netdev_boot_setup_check
);
512 * netdev_boot_base - get address from boot time settings
513 * @prefix: prefix for network device
514 * @unit: id for network device
516 * Check boot time settings for the base address of device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found.
521 unsigned long netdev_boot_base(const char *prefix
, int unit
)
523 const struct netdev_boot_setup
*s
= dev_boot_setup
;
527 sprintf(name
, "%s%d", prefix
, unit
);
530 * If device already registered then return base of 1
531 * to indicate not to probe for this interface
533 if (__dev_get_by_name(&init_net
, name
))
536 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
537 if (!strcmp(name
, s
[i
].name
))
538 return s
[i
].map
.base_addr
;
543 * Saves at boot time configured settings for any netdevice.
545 int __init
netdev_boot_setup(char *str
)
550 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
555 memset(&map
, 0, sizeof(map
));
559 map
.base_addr
= ints
[2];
561 map
.mem_start
= ints
[3];
563 map
.mem_end
= ints
[4];
565 /* Add new entry to the list */
566 return netdev_boot_setup_add(str
, &map
);
569 __setup("netdev=", netdev_boot_setup
);
571 /*******************************************************************************
573 Device Interface Subroutines
575 *******************************************************************************/
578 * __dev_get_by_name - find a device by its name
579 * @net: the applicable net namespace
580 * @name: name to find
582 * Find an interface by name. Must be called under RTNL semaphore
583 * or @dev_base_lock. If the name is found a pointer to the device
584 * is returned. If the name is not found then %NULL is returned. The
585 * reference counters are not incremented so the caller must be
586 * careful with locks.
589 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
591 struct hlist_node
*p
;
592 struct net_device
*dev
;
593 struct hlist_head
*head
= dev_name_hash(net
, name
);
595 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
596 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
601 EXPORT_SYMBOL(__dev_get_by_name
);
604 * dev_get_by_name_rcu - find a device by its name
605 * @net: the applicable net namespace
606 * @name: name to find
608 * Find an interface by name.
609 * If the name is found a pointer to the device is returned.
610 * If the name is not found then %NULL is returned.
611 * The reference counters are not incremented so the caller must be
612 * careful with locks. The caller must hold RCU lock.
615 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
617 struct hlist_node
*p
;
618 struct net_device
*dev
;
619 struct hlist_head
*head
= dev_name_hash(net
, name
);
621 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
622 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
627 EXPORT_SYMBOL(dev_get_by_name_rcu
);
630 * dev_get_by_name - find a device by its name
631 * @net: the applicable net namespace
632 * @name: name to find
634 * Find an interface by name. This can be called from any
635 * context and does its own locking. The returned handle has
636 * the usage count incremented and the caller must use dev_put() to
637 * release it when it is no longer needed. %NULL is returned if no
638 * matching device is found.
641 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
643 struct net_device
*dev
;
646 dev
= dev_get_by_name_rcu(net
, name
);
652 EXPORT_SYMBOL(dev_get_by_name
);
655 * __dev_get_by_index - find a device by its ifindex
656 * @net: the applicable net namespace
657 * @ifindex: index of device
659 * Search for an interface by index. Returns %NULL if the device
660 * is not found or a pointer to the device. The device has not
661 * had its reference counter increased so the caller must be careful
662 * about locking. The caller must hold either the RTNL semaphore
666 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
668 struct hlist_node
*p
;
669 struct net_device
*dev
;
670 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
672 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
673 if (dev
->ifindex
== ifindex
)
678 EXPORT_SYMBOL(__dev_get_by_index
);
681 * dev_get_by_index_rcu - find a device by its ifindex
682 * @net: the applicable net namespace
683 * @ifindex: index of device
685 * Search for an interface by index. Returns %NULL if the device
686 * is not found or a pointer to the device. The device has not
687 * had its reference counter increased so the caller must be careful
688 * about locking. The caller must hold RCU lock.
691 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
693 struct hlist_node
*p
;
694 struct net_device
*dev
;
695 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
697 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
698 if (dev
->ifindex
== ifindex
)
703 EXPORT_SYMBOL(dev_get_by_index_rcu
);
707 * dev_get_by_index - find a device by its ifindex
708 * @net: the applicable net namespace
709 * @ifindex: index of device
711 * Search for an interface by index. Returns NULL if the device
712 * is not found or a pointer to the device. The device returned has
713 * had a reference added and the pointer is safe until the user calls
714 * dev_put to indicate they have finished with it.
717 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
719 struct net_device
*dev
;
722 dev
= dev_get_by_index_rcu(net
, ifindex
);
728 EXPORT_SYMBOL(dev_get_by_index
);
731 * dev_getbyhwaddr - find a device by its hardware address
732 * @net: the applicable net namespace
733 * @type: media type of device
734 * @ha: hardware address
736 * Search for an interface by MAC address. Returns NULL if the device
737 * is not found or a pointer to the device. The caller must hold the
738 * rtnl semaphore. The returned device has not had its ref count increased
739 * and the caller must therefore be careful about locking
742 * If the API was consistent this would be __dev_get_by_hwaddr
745 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
747 struct net_device
*dev
;
751 for_each_netdev(net
, dev
)
752 if (dev
->type
== type
&&
753 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
758 EXPORT_SYMBOL(dev_getbyhwaddr
);
760 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
762 struct net_device
*dev
;
765 for_each_netdev(net
, dev
)
766 if (dev
->type
== type
)
771 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
773 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
775 struct net_device
*dev
;
778 dev
= __dev_getfirstbyhwtype(net
, type
);
784 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
787 * dev_get_by_flags - find any device with given flags
788 * @net: the applicable net namespace
789 * @if_flags: IFF_* values
790 * @mask: bitmask of bits in if_flags to check
792 * Search for any interface with the given flags. Returns NULL if a device
793 * is not found or a pointer to the device. The device returned has
794 * had a reference added and the pointer is safe until the user calls
795 * dev_put to indicate they have finished with it.
798 struct net_device
*dev_get_by_flags(struct net
*net
, unsigned short if_flags
,
801 struct net_device
*dev
, *ret
;
805 for_each_netdev_rcu(net
, dev
) {
806 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
815 EXPORT_SYMBOL(dev_get_by_flags
);
818 * dev_valid_name - check if name is okay for network device
821 * Network device names need to be valid file names to
822 * to allow sysfs to work. We also disallow any kind of
825 int dev_valid_name(const char *name
)
829 if (strlen(name
) >= IFNAMSIZ
)
831 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
835 if (*name
== '/' || isspace(*name
))
841 EXPORT_SYMBOL(dev_valid_name
);
844 * __dev_alloc_name - allocate a name for a device
845 * @net: network namespace to allocate the device name in
846 * @name: name format string
847 * @buf: scratch buffer and result name string
849 * Passed a format string - eg "lt%d" it will try and find a suitable
850 * id. It scans list of devices to build up a free map, then chooses
851 * the first empty slot. The caller must hold the dev_base or rtnl lock
852 * while allocating the name and adding the device in order to avoid
854 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
855 * Returns the number of the unit assigned or a negative errno code.
858 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
862 const int max_netdevices
= 8*PAGE_SIZE
;
863 unsigned long *inuse
;
864 struct net_device
*d
;
866 p
= strnchr(name
, IFNAMSIZ
-1, '%');
869 * Verify the string as this thing may have come from
870 * the user. There must be either one "%d" and no other "%"
873 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
876 /* Use one page as a bit array of possible slots */
877 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
881 for_each_netdev(net
, d
) {
882 if (!sscanf(d
->name
, name
, &i
))
884 if (i
< 0 || i
>= max_netdevices
)
887 /* avoid cases where sscanf is not exact inverse of printf */
888 snprintf(buf
, IFNAMSIZ
, name
, i
);
889 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
893 i
= find_first_zero_bit(inuse
, max_netdevices
);
894 free_page((unsigned long) inuse
);
898 snprintf(buf
, IFNAMSIZ
, name
, i
);
899 if (!__dev_get_by_name(net
, buf
))
902 /* It is possible to run out of possible slots
903 * when the name is long and there isn't enough space left
904 * for the digits, or if all bits are used.
910 * dev_alloc_name - allocate a name for a device
912 * @name: name format string
914 * Passed a format string - eg "lt%d" it will try and find a suitable
915 * id. It scans list of devices to build up a free map, then chooses
916 * the first empty slot. The caller must hold the dev_base or rtnl lock
917 * while allocating the name and adding the device in order to avoid
919 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
920 * Returns the number of the unit assigned or a negative errno code.
923 int dev_alloc_name(struct net_device
*dev
, const char *name
)
929 BUG_ON(!dev_net(dev
));
931 ret
= __dev_alloc_name(net
, name
, buf
);
933 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
936 EXPORT_SYMBOL(dev_alloc_name
);
938 static int dev_get_valid_name(struct net
*net
, const char *name
, char *buf
,
941 if (!dev_valid_name(name
))
944 if (fmt
&& strchr(name
, '%'))
945 return __dev_alloc_name(net
, name
, buf
);
946 else if (__dev_get_by_name(net
, name
))
948 else if (buf
!= name
)
949 strlcpy(buf
, name
, IFNAMSIZ
);
955 * dev_change_name - change name of a device
957 * @newname: name (or format string) must be at least IFNAMSIZ
959 * Change name of a device, can pass format strings "eth%d".
962 int dev_change_name(struct net_device
*dev
, const char *newname
)
964 char oldname
[IFNAMSIZ
];
970 BUG_ON(!dev_net(dev
));
973 if (dev
->flags
& IFF_UP
)
976 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
979 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
981 err
= dev_get_valid_name(net
, newname
, dev
->name
, 1);
986 /* For now only devices in the initial network namespace
989 if (net_eq(net
, &init_net
)) {
990 ret
= device_rename(&dev
->dev
, dev
->name
);
992 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
997 write_lock_bh(&dev_base_lock
);
998 hlist_del(&dev
->name_hlist
);
999 write_unlock_bh(&dev_base_lock
);
1003 write_lock_bh(&dev_base_lock
);
1004 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1005 write_unlock_bh(&dev_base_lock
);
1007 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1008 ret
= notifier_to_errno(ret
);
1011 /* err >= 0 after dev_alloc_name() or stores the first errno */
1014 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1018 "%s: name change rollback failed: %d.\n",
1027 * dev_set_alias - change ifalias of a device
1029 * @alias: name up to IFALIASZ
1030 * @len: limit of bytes to copy from info
1032 * Set ifalias for a device,
1034 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1038 if (len
>= IFALIASZ
)
1043 kfree(dev
->ifalias
);
1044 dev
->ifalias
= NULL
;
1049 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1053 strlcpy(dev
->ifalias
, alias
, len
+1);
1059 * netdev_features_change - device changes features
1060 * @dev: device to cause notification
1062 * Called to indicate a device has changed features.
1064 void netdev_features_change(struct net_device
*dev
)
1066 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1068 EXPORT_SYMBOL(netdev_features_change
);
1071 * netdev_state_change - device changes state
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed state. This function calls
1075 * the notifier chains for netdev_chain and sends a NEWLINK message
1076 * to the routing socket.
1078 void netdev_state_change(struct net_device
*dev
)
1080 if (dev
->flags
& IFF_UP
) {
1081 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1082 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1085 EXPORT_SYMBOL(netdev_state_change
);
1087 void netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1089 call_netdevice_notifiers(event
, dev
);
1091 EXPORT_SYMBOL(netdev_bonding_change
);
1094 * dev_load - load a network module
1095 * @net: the applicable net namespace
1096 * @name: name of interface
1098 * If a network interface is not present and the process has suitable
1099 * privileges this function loads the module. If module loading is not
1100 * available in this kernel then it becomes a nop.
1103 void dev_load(struct net
*net
, const char *name
)
1105 struct net_device
*dev
;
1108 dev
= dev_get_by_name_rcu(net
, name
);
1111 if (!dev
&& capable(CAP_NET_ADMIN
))
1112 request_module("%s", name
);
1114 EXPORT_SYMBOL(dev_load
);
1117 * dev_open - prepare an interface for use.
1118 * @dev: device to open
1120 * Takes a device from down to up state. The device's private open
1121 * function is invoked and then the multicast lists are loaded. Finally
1122 * the device is moved into the up state and a %NETDEV_UP message is
1123 * sent to the netdev notifier chain.
1125 * Calling this function on an active interface is a nop. On a failure
1126 * a negative errno code is returned.
1128 int dev_open(struct net_device
*dev
)
1130 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1139 if (dev
->flags
& IFF_UP
)
1143 * Is it even present?
1145 if (!netif_device_present(dev
))
1148 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1149 ret
= notifier_to_errno(ret
);
1154 * Call device private open method
1156 set_bit(__LINK_STATE_START
, &dev
->state
);
1158 if (ops
->ndo_validate_addr
)
1159 ret
= ops
->ndo_validate_addr(dev
);
1161 if (!ret
&& ops
->ndo_open
)
1162 ret
= ops
->ndo_open(dev
);
1165 * If it went open OK then:
1169 clear_bit(__LINK_STATE_START
, &dev
->state
);
1174 dev
->flags
|= IFF_UP
;
1179 net_dmaengine_get();
1182 * Initialize multicasting status
1184 dev_set_rx_mode(dev
);
1187 * Wakeup transmit queue engine
1192 * ... and announce new interface.
1194 call_netdevice_notifiers(NETDEV_UP
, dev
);
1199 EXPORT_SYMBOL(dev_open
);
1202 * dev_close - shutdown an interface.
1203 * @dev: device to shutdown
1205 * This function moves an active device into down state. A
1206 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1207 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1210 int dev_close(struct net_device
*dev
)
1212 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1217 if (!(dev
->flags
& IFF_UP
))
1221 * Tell people we are going down, so that they can
1222 * prepare to death, when device is still operating.
1224 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1226 clear_bit(__LINK_STATE_START
, &dev
->state
);
1228 /* Synchronize to scheduled poll. We cannot touch poll list,
1229 * it can be even on different cpu. So just clear netif_running().
1231 * dev->stop() will invoke napi_disable() on all of it's
1232 * napi_struct instances on this device.
1234 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1236 dev_deactivate(dev
);
1239 * Call the device specific close. This cannot fail.
1240 * Only if device is UP
1242 * We allow it to be called even after a DETACH hot-plug
1249 * Device is now down.
1252 dev
->flags
&= ~IFF_UP
;
1255 * Tell people we are down
1257 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1262 net_dmaengine_put();
1266 EXPORT_SYMBOL(dev_close
);
1270 * dev_disable_lro - disable Large Receive Offload on a device
1273 * Disable Large Receive Offload (LRO) on a net device. Must be
1274 * called under RTNL. This is needed if received packets may be
1275 * forwarded to another interface.
1277 void dev_disable_lro(struct net_device
*dev
)
1279 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1280 dev
->ethtool_ops
->set_flags
) {
1281 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1282 if (flags
& ETH_FLAG_LRO
) {
1283 flags
&= ~ETH_FLAG_LRO
;
1284 dev
->ethtool_ops
->set_flags(dev
, flags
);
1287 WARN_ON(dev
->features
& NETIF_F_LRO
);
1289 EXPORT_SYMBOL(dev_disable_lro
);
1292 static int dev_boot_phase
= 1;
1295 * Device change register/unregister. These are not inline or static
1296 * as we export them to the world.
1300 * register_netdevice_notifier - register a network notifier block
1303 * Register a notifier to be called when network device events occur.
1304 * The notifier passed is linked into the kernel structures and must
1305 * not be reused until it has been unregistered. A negative errno code
1306 * is returned on a failure.
1308 * When registered all registration and up events are replayed
1309 * to the new notifier to allow device to have a race free
1310 * view of the network device list.
1313 int register_netdevice_notifier(struct notifier_block
*nb
)
1315 struct net_device
*dev
;
1316 struct net_device
*last
;
1321 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1327 for_each_netdev(net
, dev
) {
1328 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1329 err
= notifier_to_errno(err
);
1333 if (!(dev
->flags
& IFF_UP
))
1336 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1347 for_each_netdev(net
, dev
) {
1351 if (dev
->flags
& IFF_UP
) {
1352 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1353 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1355 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1356 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1360 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1363 EXPORT_SYMBOL(register_netdevice_notifier
);
1366 * unregister_netdevice_notifier - unregister a network notifier block
1369 * Unregister a notifier previously registered by
1370 * register_netdevice_notifier(). The notifier is unlinked into the
1371 * kernel structures and may then be reused. A negative errno code
1372 * is returned on a failure.
1375 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1380 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1384 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1387 * call_netdevice_notifiers - call all network notifier blocks
1388 * @val: value passed unmodified to notifier function
1389 * @dev: net_device pointer passed unmodified to notifier function
1391 * Call all network notifier blocks. Parameters and return value
1392 * are as for raw_notifier_call_chain().
1395 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1397 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1400 /* When > 0 there are consumers of rx skb time stamps */
1401 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1403 void net_enable_timestamp(void)
1405 atomic_inc(&netstamp_needed
);
1407 EXPORT_SYMBOL(net_enable_timestamp
);
1409 void net_disable_timestamp(void)
1411 atomic_dec(&netstamp_needed
);
1413 EXPORT_SYMBOL(net_disable_timestamp
);
1415 static inline void net_timestamp(struct sk_buff
*skb
)
1417 if (atomic_read(&netstamp_needed
))
1418 __net_timestamp(skb
);
1420 skb
->tstamp
.tv64
= 0;
1424 * dev_forward_skb - loopback an skb to another netif
1426 * @dev: destination network device
1427 * @skb: buffer to forward
1430 * NET_RX_SUCCESS (no congestion)
1431 * NET_RX_DROP (packet was dropped)
1433 * dev_forward_skb can be used for injecting an skb from the
1434 * start_xmit function of one device into the receive queue
1435 * of another device.
1437 * The receiving device may be in another namespace, so
1438 * we have to clear all information in the skb that could
1439 * impact namespace isolation.
1441 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1445 if (!(dev
->flags
& IFF_UP
))
1448 if (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
))
1452 skb
->tstamp
.tv64
= 0;
1453 skb
->pkt_type
= PACKET_HOST
;
1454 skb
->protocol
= eth_type_trans(skb
, dev
);
1458 return netif_rx(skb
);
1460 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1463 * Support routine. Sends outgoing frames to any network
1464 * taps currently in use.
1467 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1469 struct packet_type
*ptype
;
1471 #ifdef CONFIG_NET_CLS_ACT
1472 if (!(skb
->tstamp
.tv64
&& (G_TC_FROM(skb
->tc_verd
) & AT_INGRESS
)))
1479 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1480 /* Never send packets back to the socket
1481 * they originated from - MvS (miquels@drinkel.ow.org)
1483 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1484 (ptype
->af_packet_priv
== NULL
||
1485 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1486 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1490 /* skb->nh should be correctly
1491 set by sender, so that the second statement is
1492 just protection against buggy protocols.
1494 skb_reset_mac_header(skb2
);
1496 if (skb_network_header(skb2
) < skb2
->data
||
1497 skb2
->network_header
> skb2
->tail
) {
1498 if (net_ratelimit())
1499 printk(KERN_CRIT
"protocol %04x is "
1501 skb2
->protocol
, dev
->name
);
1502 skb_reset_network_header(skb2
);
1505 skb2
->transport_header
= skb2
->network_header
;
1506 skb2
->pkt_type
= PACKET_OUTGOING
;
1507 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1514 static inline void __netif_reschedule(struct Qdisc
*q
)
1516 struct softnet_data
*sd
;
1517 unsigned long flags
;
1519 local_irq_save(flags
);
1520 sd
= &__get_cpu_var(softnet_data
);
1521 q
->next_sched
= sd
->output_queue
;
1522 sd
->output_queue
= q
;
1523 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1524 local_irq_restore(flags
);
1527 void __netif_schedule(struct Qdisc
*q
)
1529 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1530 __netif_reschedule(q
);
1532 EXPORT_SYMBOL(__netif_schedule
);
1534 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1536 if (atomic_dec_and_test(&skb
->users
)) {
1537 struct softnet_data
*sd
;
1538 unsigned long flags
;
1540 local_irq_save(flags
);
1541 sd
= &__get_cpu_var(softnet_data
);
1542 skb
->next
= sd
->completion_queue
;
1543 sd
->completion_queue
= skb
;
1544 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1545 local_irq_restore(flags
);
1548 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1550 void dev_kfree_skb_any(struct sk_buff
*skb
)
1552 if (in_irq() || irqs_disabled())
1553 dev_kfree_skb_irq(skb
);
1557 EXPORT_SYMBOL(dev_kfree_skb_any
);
1561 * netif_device_detach - mark device as removed
1562 * @dev: network device
1564 * Mark device as removed from system and therefore no longer available.
1566 void netif_device_detach(struct net_device
*dev
)
1568 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1569 netif_running(dev
)) {
1570 netif_tx_stop_all_queues(dev
);
1573 EXPORT_SYMBOL(netif_device_detach
);
1576 * netif_device_attach - mark device as attached
1577 * @dev: network device
1579 * Mark device as attached from system and restart if needed.
1581 void netif_device_attach(struct net_device
*dev
)
1583 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1584 netif_running(dev
)) {
1585 netif_tx_wake_all_queues(dev
);
1586 __netdev_watchdog_up(dev
);
1589 EXPORT_SYMBOL(netif_device_attach
);
1591 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1593 return ((features
& NETIF_F_GEN_CSUM
) ||
1594 ((features
& NETIF_F_IP_CSUM
) &&
1595 protocol
== htons(ETH_P_IP
)) ||
1596 ((features
& NETIF_F_IPV6_CSUM
) &&
1597 protocol
== htons(ETH_P_IPV6
)) ||
1598 ((features
& NETIF_F_FCOE_CRC
) &&
1599 protocol
== htons(ETH_P_FCOE
)));
1602 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1604 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1607 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1608 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1609 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1610 veh
->h_vlan_encapsulated_proto
))
1618 * Invalidate hardware checksum when packet is to be mangled, and
1619 * complete checksum manually on outgoing path.
1621 int skb_checksum_help(struct sk_buff
*skb
)
1624 int ret
= 0, offset
;
1626 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1627 goto out_set_summed
;
1629 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1630 /* Let GSO fix up the checksum. */
1631 goto out_set_summed
;
1634 offset
= skb
->csum_start
- skb_headroom(skb
);
1635 BUG_ON(offset
>= skb_headlen(skb
));
1636 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1638 offset
+= skb
->csum_offset
;
1639 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1641 if (skb_cloned(skb
) &&
1642 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1643 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1648 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1650 skb
->ip_summed
= CHECKSUM_NONE
;
1654 EXPORT_SYMBOL(skb_checksum_help
);
1657 * skb_gso_segment - Perform segmentation on skb.
1658 * @skb: buffer to segment
1659 * @features: features for the output path (see dev->features)
1661 * This function segments the given skb and returns a list of segments.
1663 * It may return NULL if the skb requires no segmentation. This is
1664 * only possible when GSO is used for verifying header integrity.
1666 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1668 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1669 struct packet_type
*ptype
;
1670 __be16 type
= skb
->protocol
;
1673 skb_reset_mac_header(skb
);
1674 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1675 __skb_pull(skb
, skb
->mac_len
);
1677 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1678 struct net_device
*dev
= skb
->dev
;
1679 struct ethtool_drvinfo info
= {};
1681 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1682 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1684 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1686 info
.driver
, dev
? dev
->features
: 0L,
1687 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1688 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1690 if (skb_header_cloned(skb
) &&
1691 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1692 return ERR_PTR(err
);
1696 list_for_each_entry_rcu(ptype
,
1697 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1698 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1699 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1700 err
= ptype
->gso_send_check(skb
);
1701 segs
= ERR_PTR(err
);
1702 if (err
|| skb_gso_ok(skb
, features
))
1704 __skb_push(skb
, (skb
->data
-
1705 skb_network_header(skb
)));
1707 segs
= ptype
->gso_segment(skb
, features
);
1713 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1717 EXPORT_SYMBOL(skb_gso_segment
);
1719 /* Take action when hardware reception checksum errors are detected. */
1721 void netdev_rx_csum_fault(struct net_device
*dev
)
1723 if (net_ratelimit()) {
1724 printk(KERN_ERR
"%s: hw csum failure.\n",
1725 dev
? dev
->name
: "<unknown>");
1729 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1732 /* Actually, we should eliminate this check as soon as we know, that:
1733 * 1. IOMMU is present and allows to map all the memory.
1734 * 2. No high memory really exists on this machine.
1737 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1739 #ifdef CONFIG_HIGHMEM
1742 if (dev
->features
& NETIF_F_HIGHDMA
)
1745 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1746 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1754 void (*destructor
)(struct sk_buff
*skb
);
1757 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1759 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1761 struct dev_gso_cb
*cb
;
1764 struct sk_buff
*nskb
= skb
->next
;
1766 skb
->next
= nskb
->next
;
1769 } while (skb
->next
);
1771 cb
= DEV_GSO_CB(skb
);
1773 cb
->destructor(skb
);
1777 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1778 * @skb: buffer to segment
1780 * This function segments the given skb and stores the list of segments
1783 static int dev_gso_segment(struct sk_buff
*skb
)
1785 struct net_device
*dev
= skb
->dev
;
1786 struct sk_buff
*segs
;
1787 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1790 segs
= skb_gso_segment(skb
, features
);
1792 /* Verifying header integrity only. */
1797 return PTR_ERR(segs
);
1800 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1801 skb
->destructor
= dev_gso_skb_destructor
;
1806 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1807 struct netdev_queue
*txq
)
1809 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1810 int rc
= NETDEV_TX_OK
;
1812 if (likely(!skb
->next
)) {
1813 if (!list_empty(&ptype_all
))
1814 dev_queue_xmit_nit(skb
, dev
);
1816 if (netif_needs_gso(dev
, skb
)) {
1817 if (unlikely(dev_gso_segment(skb
)))
1824 * If device doesnt need skb->dst, release it right now while
1825 * its hot in this cpu cache
1827 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1830 rc
= ops
->ndo_start_xmit(skb
, dev
);
1831 if (rc
== NETDEV_TX_OK
)
1832 txq_trans_update(txq
);
1834 * TODO: if skb_orphan() was called by
1835 * dev->hard_start_xmit() (for example, the unmodified
1836 * igb driver does that; bnx2 doesn't), then
1837 * skb_tx_software_timestamp() will be unable to send
1838 * back the time stamp.
1840 * How can this be prevented? Always create another
1841 * reference to the socket before calling
1842 * dev->hard_start_xmit()? Prevent that skb_orphan()
1843 * does anything in dev->hard_start_xmit() by clearing
1844 * the skb destructor before the call and restoring it
1845 * afterwards, then doing the skb_orphan() ourselves?
1852 struct sk_buff
*nskb
= skb
->next
;
1854 skb
->next
= nskb
->next
;
1858 * If device doesnt need nskb->dst, release it right now while
1859 * its hot in this cpu cache
1861 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1864 rc
= ops
->ndo_start_xmit(nskb
, dev
);
1865 if (unlikely(rc
!= NETDEV_TX_OK
)) {
1866 if (rc
& ~NETDEV_TX_MASK
)
1867 goto out_kfree_gso_skb
;
1868 nskb
->next
= skb
->next
;
1872 txq_trans_update(txq
);
1873 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
1874 return NETDEV_TX_BUSY
;
1875 } while (skb
->next
);
1878 if (likely(skb
->next
== NULL
))
1879 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1885 static u32 skb_tx_hashrnd
;
1887 u16
skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
)
1891 if (skb_rx_queue_recorded(skb
)) {
1892 hash
= skb_get_rx_queue(skb
);
1893 while (unlikely(hash
>= dev
->real_num_tx_queues
))
1894 hash
-= dev
->real_num_tx_queues
;
1898 if (skb
->sk
&& skb
->sk
->sk_hash
)
1899 hash
= skb
->sk
->sk_hash
;
1901 hash
= skb
->protocol
;
1903 hash
= jhash_1word(hash
, skb_tx_hashrnd
);
1905 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
1907 EXPORT_SYMBOL(skb_tx_hash
);
1909 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
1911 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
1912 if (net_ratelimit()) {
1913 WARN(1, "%s selects TX queue %d, but "
1914 "real number of TX queues is %d\n",
1915 dev
->name
, queue_index
,
1916 dev
->real_num_tx_queues
);
1923 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
1924 struct sk_buff
*skb
)
1927 struct sock
*sk
= skb
->sk
;
1929 if (sk_tx_queue_recorded(sk
)) {
1930 queue_index
= sk_tx_queue_get(sk
);
1932 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1934 if (ops
->ndo_select_queue
) {
1935 queue_index
= ops
->ndo_select_queue(dev
, skb
);
1936 queue_index
= dev_cap_txqueue(dev
, queue_index
);
1939 if (dev
->real_num_tx_queues
> 1)
1940 queue_index
= skb_tx_hash(dev
, skb
);
1942 if (sk
&& sk
->sk_dst_cache
)
1943 sk_tx_queue_set(sk
, queue_index
);
1947 skb_set_queue_mapping(skb
, queue_index
);
1948 return netdev_get_tx_queue(dev
, queue_index
);
1951 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
1952 struct net_device
*dev
,
1953 struct netdev_queue
*txq
)
1955 spinlock_t
*root_lock
= qdisc_lock(q
);
1958 spin_lock(root_lock
);
1959 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
1962 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
1963 !test_and_set_bit(__QDISC_STATE_RUNNING
, &q
->state
)) {
1965 * This is a work-conserving queue; there are no old skbs
1966 * waiting to be sent out; and the qdisc is not running -
1967 * xmit the skb directly.
1969 __qdisc_update_bstats(q
, skb
->len
);
1970 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
))
1973 clear_bit(__QDISC_STATE_RUNNING
, &q
->state
);
1975 rc
= NET_XMIT_SUCCESS
;
1977 rc
= qdisc_enqueue_root(skb
, q
);
1980 spin_unlock(root_lock
);
1986 * dev_queue_xmit - transmit a buffer
1987 * @skb: buffer to transmit
1989 * Queue a buffer for transmission to a network device. The caller must
1990 * have set the device and priority and built the buffer before calling
1991 * this function. The function can be called from an interrupt.
1993 * A negative errno code is returned on a failure. A success does not
1994 * guarantee the frame will be transmitted as it may be dropped due
1995 * to congestion or traffic shaping.
1997 * -----------------------------------------------------------------------------------
1998 * I notice this method can also return errors from the queue disciplines,
1999 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2002 * Regardless of the return value, the skb is consumed, so it is currently
2003 * difficult to retry a send to this method. (You can bump the ref count
2004 * before sending to hold a reference for retry if you are careful.)
2006 * When calling this method, interrupts MUST be enabled. This is because
2007 * the BH enable code must have IRQs enabled so that it will not deadlock.
2010 int dev_queue_xmit(struct sk_buff
*skb
)
2012 struct net_device
*dev
= skb
->dev
;
2013 struct netdev_queue
*txq
;
2017 /* GSO will handle the following emulations directly. */
2018 if (netif_needs_gso(dev
, skb
))
2021 if (skb_has_frags(skb
) &&
2022 !(dev
->features
& NETIF_F_FRAGLIST
) &&
2023 __skb_linearize(skb
))
2026 /* Fragmented skb is linearized if device does not support SG,
2027 * or if at least one of fragments is in highmem and device
2028 * does not support DMA from it.
2030 if (skb_shinfo(skb
)->nr_frags
&&
2031 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
2032 __skb_linearize(skb
))
2035 /* If packet is not checksummed and device does not support
2036 * checksumming for this protocol, complete checksumming here.
2038 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2039 skb_set_transport_header(skb
, skb
->csum_start
-
2041 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
2046 /* Disable soft irqs for various locks below. Also
2047 * stops preemption for RCU.
2051 txq
= dev_pick_tx(dev
, skb
);
2052 q
= rcu_dereference(txq
->qdisc
);
2054 #ifdef CONFIG_NET_CLS_ACT
2055 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2058 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2062 /* The device has no queue. Common case for software devices:
2063 loopback, all the sorts of tunnels...
2065 Really, it is unlikely that netif_tx_lock protection is necessary
2066 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2068 However, it is possible, that they rely on protection
2071 Check this and shot the lock. It is not prone from deadlocks.
2072 Either shot noqueue qdisc, it is even simpler 8)
2074 if (dev
->flags
& IFF_UP
) {
2075 int cpu
= smp_processor_id(); /* ok because BHs are off */
2077 if (txq
->xmit_lock_owner
!= cpu
) {
2079 HARD_TX_LOCK(dev
, txq
, cpu
);
2081 if (!netif_tx_queue_stopped(txq
)) {
2082 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2083 if (dev_xmit_complete(rc
)) {
2084 HARD_TX_UNLOCK(dev
, txq
);
2088 HARD_TX_UNLOCK(dev
, txq
);
2089 if (net_ratelimit())
2090 printk(KERN_CRIT
"Virtual device %s asks to "
2091 "queue packet!\n", dev
->name
);
2093 /* Recursion is detected! It is possible,
2095 if (net_ratelimit())
2096 printk(KERN_CRIT
"Dead loop on virtual device "
2097 "%s, fix it urgently!\n", dev
->name
);
2102 rcu_read_unlock_bh();
2108 rcu_read_unlock_bh();
2111 EXPORT_SYMBOL(dev_queue_xmit
);
2114 /*=======================================================================
2116 =======================================================================*/
2118 int netdev_max_backlog __read_mostly
= 1000;
2119 int netdev_budget __read_mostly
= 300;
2120 int weight_p __read_mostly
= 64; /* old backlog weight */
2122 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
2126 * netif_rx - post buffer to the network code
2127 * @skb: buffer to post
2129 * This function receives a packet from a device driver and queues it for
2130 * the upper (protocol) levels to process. It always succeeds. The buffer
2131 * may be dropped during processing for congestion control or by the
2135 * NET_RX_SUCCESS (no congestion)
2136 * NET_RX_DROP (packet was dropped)
2140 int netif_rx(struct sk_buff
*skb
)
2142 struct softnet_data
*queue
;
2143 unsigned long flags
;
2145 /* if netpoll wants it, pretend we never saw it */
2146 if (netpoll_rx(skb
))
2149 if (!skb
->tstamp
.tv64
)
2153 * The code is rearranged so that the path is the most
2154 * short when CPU is congested, but is still operating.
2156 local_irq_save(flags
);
2157 queue
= &__get_cpu_var(softnet_data
);
2159 __get_cpu_var(netdev_rx_stat
).total
++;
2160 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
2161 if (queue
->input_pkt_queue
.qlen
) {
2163 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
2164 local_irq_restore(flags
);
2165 return NET_RX_SUCCESS
;
2168 napi_schedule(&queue
->backlog
);
2172 __get_cpu_var(netdev_rx_stat
).dropped
++;
2173 local_irq_restore(flags
);
2178 EXPORT_SYMBOL(netif_rx
);
2180 int netif_rx_ni(struct sk_buff
*skb
)
2185 err
= netif_rx(skb
);
2186 if (local_softirq_pending())
2192 EXPORT_SYMBOL(netif_rx_ni
);
2194 static void net_tx_action(struct softirq_action
*h
)
2196 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2198 if (sd
->completion_queue
) {
2199 struct sk_buff
*clist
;
2201 local_irq_disable();
2202 clist
= sd
->completion_queue
;
2203 sd
->completion_queue
= NULL
;
2207 struct sk_buff
*skb
= clist
;
2208 clist
= clist
->next
;
2210 WARN_ON(atomic_read(&skb
->users
));
2215 if (sd
->output_queue
) {
2218 local_irq_disable();
2219 head
= sd
->output_queue
;
2220 sd
->output_queue
= NULL
;
2224 struct Qdisc
*q
= head
;
2225 spinlock_t
*root_lock
;
2227 head
= head
->next_sched
;
2229 root_lock
= qdisc_lock(q
);
2230 if (spin_trylock(root_lock
)) {
2231 smp_mb__before_clear_bit();
2232 clear_bit(__QDISC_STATE_SCHED
,
2235 spin_unlock(root_lock
);
2237 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2239 __netif_reschedule(q
);
2241 smp_mb__before_clear_bit();
2242 clear_bit(__QDISC_STATE_SCHED
,
2250 static inline int deliver_skb(struct sk_buff
*skb
,
2251 struct packet_type
*pt_prev
,
2252 struct net_device
*orig_dev
)
2254 atomic_inc(&skb
->users
);
2255 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2258 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2260 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2261 /* This hook is defined here for ATM LANE */
2262 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2263 unsigned char *addr
) __read_mostly
;
2264 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2268 * If bridge module is loaded call bridging hook.
2269 * returns NULL if packet was consumed.
2271 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
2272 struct sk_buff
*skb
) __read_mostly
;
2273 EXPORT_SYMBOL_GPL(br_handle_frame_hook
);
2275 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
2276 struct packet_type
**pt_prev
, int *ret
,
2277 struct net_device
*orig_dev
)
2279 struct net_bridge_port
*port
;
2281 if (skb
->pkt_type
== PACKET_LOOPBACK
||
2282 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
2286 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2290 return br_handle_frame_hook(port
, skb
);
2293 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2296 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2297 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
2298 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
2300 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
2301 struct packet_type
**pt_prev
,
2303 struct net_device
*orig_dev
)
2305 if (skb
->dev
->macvlan_port
== NULL
)
2309 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2312 return macvlan_handle_frame_hook(skb
);
2315 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2318 #ifdef CONFIG_NET_CLS_ACT
2319 /* TODO: Maybe we should just force sch_ingress to be compiled in
2320 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2321 * a compare and 2 stores extra right now if we dont have it on
2322 * but have CONFIG_NET_CLS_ACT
2323 * NOTE: This doesnt stop any functionality; if you dont have
2324 * the ingress scheduler, you just cant add policies on ingress.
2327 static int ing_filter(struct sk_buff
*skb
)
2329 struct net_device
*dev
= skb
->dev
;
2330 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2331 struct netdev_queue
*rxq
;
2332 int result
= TC_ACT_OK
;
2335 if (MAX_RED_LOOP
< ttl
++) {
2337 "Redir loop detected Dropping packet (%d->%d)\n",
2338 skb
->skb_iif
, dev
->ifindex
);
2342 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2343 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2345 rxq
= &dev
->rx_queue
;
2348 if (q
!= &noop_qdisc
) {
2349 spin_lock(qdisc_lock(q
));
2350 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2351 result
= qdisc_enqueue_root(skb
, q
);
2352 spin_unlock(qdisc_lock(q
));
2358 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2359 struct packet_type
**pt_prev
,
2360 int *ret
, struct net_device
*orig_dev
)
2362 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2366 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2369 /* Huh? Why does turning on AF_PACKET affect this? */
2370 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2373 switch (ing_filter(skb
)) {
2387 * netif_nit_deliver - deliver received packets to network taps
2390 * This function is used to deliver incoming packets to network
2391 * taps. It should be used when the normal netif_receive_skb path
2392 * is bypassed, for example because of VLAN acceleration.
2394 void netif_nit_deliver(struct sk_buff
*skb
)
2396 struct packet_type
*ptype
;
2398 if (list_empty(&ptype_all
))
2401 skb_reset_network_header(skb
);
2402 skb_reset_transport_header(skb
);
2403 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2406 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2407 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2408 deliver_skb(skb
, ptype
, skb
->dev
);
2414 * netif_receive_skb - process receive buffer from network
2415 * @skb: buffer to process
2417 * netif_receive_skb() is the main receive data processing function.
2418 * It always succeeds. The buffer may be dropped during processing
2419 * for congestion control or by the protocol layers.
2421 * This function may only be called from softirq context and interrupts
2422 * should be enabled.
2424 * Return values (usually ignored):
2425 * NET_RX_SUCCESS: no congestion
2426 * NET_RX_DROP: packet was dropped
2428 int netif_receive_skb(struct sk_buff
*skb
)
2430 struct packet_type
*ptype
, *pt_prev
;
2431 struct net_device
*orig_dev
;
2432 struct net_device
*null_or_orig
;
2433 int ret
= NET_RX_DROP
;
2436 if (!skb
->tstamp
.tv64
)
2439 if (vlan_tx_tag_present(skb
) && vlan_hwaccel_do_receive(skb
))
2440 return NET_RX_SUCCESS
;
2442 /* if we've gotten here through NAPI, check netpoll */
2443 if (netpoll_receive_skb(skb
))
2447 skb
->skb_iif
= skb
->dev
->ifindex
;
2449 null_or_orig
= NULL
;
2450 orig_dev
= skb
->dev
;
2451 if (orig_dev
->master
) {
2452 if (skb_bond_should_drop(skb
))
2453 null_or_orig
= orig_dev
; /* deliver only exact match */
2455 skb
->dev
= orig_dev
->master
;
2458 __get_cpu_var(netdev_rx_stat
).total
++;
2460 skb_reset_network_header(skb
);
2461 skb_reset_transport_header(skb
);
2462 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2468 #ifdef CONFIG_NET_CLS_ACT
2469 if (skb
->tc_verd
& TC_NCLS
) {
2470 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2475 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2476 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2477 ptype
->dev
== orig_dev
) {
2479 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2484 #ifdef CONFIG_NET_CLS_ACT
2485 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2491 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
2494 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
2499 * Make sure frames received on VLAN interfaces stacked on
2500 * bonding interfaces still make their way to any base bonding
2501 * device that may have registered for a specific ptype. The
2502 * handler may have to adjust skb->dev and orig_dev.
2504 * null_or_orig can be overloaded since it will not be set when
2505 * using VLANs on top of bonding. Putting it here prevents
2506 * disturbing the ptype_all handlers above.
2508 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
2509 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
2510 null_or_orig
= vlan_dev_real_dev(skb
->dev
);
2513 type
= skb
->protocol
;
2514 list_for_each_entry_rcu(ptype
,
2515 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2516 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
2517 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
)) {
2519 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2525 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2528 /* Jamal, now you will not able to escape explaining
2529 * me how you were going to use this. :-)
2538 EXPORT_SYMBOL(netif_receive_skb
);
2540 /* Network device is going away, flush any packets still pending */
2541 static void flush_backlog(void *arg
)
2543 struct net_device
*dev
= arg
;
2544 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2545 struct sk_buff
*skb
, *tmp
;
2547 skb_queue_walk_safe(&queue
->input_pkt_queue
, skb
, tmp
)
2548 if (skb
->dev
== dev
) {
2549 __skb_unlink(skb
, &queue
->input_pkt_queue
);
2554 static int napi_gro_complete(struct sk_buff
*skb
)
2556 struct packet_type
*ptype
;
2557 __be16 type
= skb
->protocol
;
2558 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2561 if (NAPI_GRO_CB(skb
)->count
== 1) {
2562 skb_shinfo(skb
)->gso_size
= 0;
2567 list_for_each_entry_rcu(ptype
, head
, list
) {
2568 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
2571 err
= ptype
->gro_complete(skb
);
2577 WARN_ON(&ptype
->list
== head
);
2579 return NET_RX_SUCCESS
;
2583 return netif_receive_skb(skb
);
2586 void napi_gro_flush(struct napi_struct
*napi
)
2588 struct sk_buff
*skb
, *next
;
2590 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
2593 napi_gro_complete(skb
);
2596 napi
->gro_count
= 0;
2597 napi
->gro_list
= NULL
;
2599 EXPORT_SYMBOL(napi_gro_flush
);
2601 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2603 struct sk_buff
**pp
= NULL
;
2604 struct packet_type
*ptype
;
2605 __be16 type
= skb
->protocol
;
2606 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2609 enum gro_result ret
;
2611 if (!(skb
->dev
->features
& NETIF_F_GRO
))
2614 if (skb_is_gso(skb
) || skb_has_frags(skb
))
2618 list_for_each_entry_rcu(ptype
, head
, list
) {
2619 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
2622 skb_set_network_header(skb
, skb_gro_offset(skb
));
2623 mac_len
= skb
->network_header
- skb
->mac_header
;
2624 skb
->mac_len
= mac_len
;
2625 NAPI_GRO_CB(skb
)->same_flow
= 0;
2626 NAPI_GRO_CB(skb
)->flush
= 0;
2627 NAPI_GRO_CB(skb
)->free
= 0;
2629 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
2634 if (&ptype
->list
== head
)
2637 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
2638 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
2641 struct sk_buff
*nskb
= *pp
;
2645 napi_gro_complete(nskb
);
2652 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
2656 NAPI_GRO_CB(skb
)->count
= 1;
2657 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
2658 skb
->next
= napi
->gro_list
;
2659 napi
->gro_list
= skb
;
2663 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
2664 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
2666 BUG_ON(skb
->end
- skb
->tail
< grow
);
2668 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
2671 skb
->data_len
-= grow
;
2673 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
2674 skb_shinfo(skb
)->frags
[0].size
-= grow
;
2676 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
2677 put_page(skb_shinfo(skb
)->frags
[0].page
);
2678 memmove(skb_shinfo(skb
)->frags
,
2679 skb_shinfo(skb
)->frags
+ 1,
2680 --skb_shinfo(skb
)->nr_frags
);
2691 EXPORT_SYMBOL(dev_gro_receive
);
2694 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2698 if (netpoll_rx_on(skb
))
2701 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
2702 NAPI_GRO_CB(p
)->same_flow
=
2703 (p
->dev
== skb
->dev
) &&
2704 !compare_ether_header(skb_mac_header(p
),
2705 skb_gro_mac_header(skb
));
2706 NAPI_GRO_CB(p
)->flush
= 0;
2709 return dev_gro_receive(napi
, skb
);
2712 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
2716 if (netif_receive_skb(skb
))
2721 case GRO_MERGED_FREE
:
2732 EXPORT_SYMBOL(napi_skb_finish
);
2734 void skb_gro_reset_offset(struct sk_buff
*skb
)
2736 NAPI_GRO_CB(skb
)->data_offset
= 0;
2737 NAPI_GRO_CB(skb
)->frag0
= NULL
;
2738 NAPI_GRO_CB(skb
)->frag0_len
= 0;
2740 if (skb
->mac_header
== skb
->tail
&&
2741 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
2742 NAPI_GRO_CB(skb
)->frag0
=
2743 page_address(skb_shinfo(skb
)->frags
[0].page
) +
2744 skb_shinfo(skb
)->frags
[0].page_offset
;
2745 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
2748 EXPORT_SYMBOL(skb_gro_reset_offset
);
2750 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2752 skb_gro_reset_offset(skb
);
2754 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
2756 EXPORT_SYMBOL(napi_gro_receive
);
2758 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
2760 __skb_pull(skb
, skb_headlen(skb
));
2761 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
2765 EXPORT_SYMBOL(napi_reuse_skb
);
2767 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
2769 struct sk_buff
*skb
= napi
->skb
;
2772 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
2778 EXPORT_SYMBOL(napi_get_frags
);
2780 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
2786 skb
->protocol
= eth_type_trans(skb
, napi
->dev
);
2788 if (ret
== GRO_HELD
)
2789 skb_gro_pull(skb
, -ETH_HLEN
);
2790 else if (netif_receive_skb(skb
))
2795 case GRO_MERGED_FREE
:
2796 napi_reuse_skb(napi
, skb
);
2805 EXPORT_SYMBOL(napi_frags_finish
);
2807 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
2809 struct sk_buff
*skb
= napi
->skb
;
2816 skb_reset_mac_header(skb
);
2817 skb_gro_reset_offset(skb
);
2819 off
= skb_gro_offset(skb
);
2820 hlen
= off
+ sizeof(*eth
);
2821 eth
= skb_gro_header_fast(skb
, off
);
2822 if (skb_gro_header_hard(skb
, hlen
)) {
2823 eth
= skb_gro_header_slow(skb
, hlen
, off
);
2824 if (unlikely(!eth
)) {
2825 napi_reuse_skb(napi
, skb
);
2831 skb_gro_pull(skb
, sizeof(*eth
));
2834 * This works because the only protocols we care about don't require
2835 * special handling. We'll fix it up properly at the end.
2837 skb
->protocol
= eth
->h_proto
;
2842 EXPORT_SYMBOL(napi_frags_skb
);
2844 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
2846 struct sk_buff
*skb
= napi_frags_skb(napi
);
2851 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
2853 EXPORT_SYMBOL(napi_gro_frags
);
2855 static int process_backlog(struct napi_struct
*napi
, int quota
)
2858 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2859 unsigned long start_time
= jiffies
;
2861 napi
->weight
= weight_p
;
2863 struct sk_buff
*skb
;
2865 local_irq_disable();
2866 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
2868 __napi_complete(napi
);
2874 netif_receive_skb(skb
);
2875 } while (++work
< quota
&& jiffies
== start_time
);
2881 * __napi_schedule - schedule for receive
2882 * @n: entry to schedule
2884 * The entry's receive function will be scheduled to run
2886 void __napi_schedule(struct napi_struct
*n
)
2888 unsigned long flags
;
2890 local_irq_save(flags
);
2891 list_add_tail(&n
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
2892 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2893 local_irq_restore(flags
);
2895 EXPORT_SYMBOL(__napi_schedule
);
2897 void __napi_complete(struct napi_struct
*n
)
2899 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
2900 BUG_ON(n
->gro_list
);
2902 list_del(&n
->poll_list
);
2903 smp_mb__before_clear_bit();
2904 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
2906 EXPORT_SYMBOL(__napi_complete
);
2908 void napi_complete(struct napi_struct
*n
)
2910 unsigned long flags
;
2913 * don't let napi dequeue from the cpu poll list
2914 * just in case its running on a different cpu
2916 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
2920 local_irq_save(flags
);
2922 local_irq_restore(flags
);
2924 EXPORT_SYMBOL(napi_complete
);
2926 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
2927 int (*poll
)(struct napi_struct
*, int), int weight
)
2929 INIT_LIST_HEAD(&napi
->poll_list
);
2930 napi
->gro_count
= 0;
2931 napi
->gro_list
= NULL
;
2934 napi
->weight
= weight
;
2935 list_add(&napi
->dev_list
, &dev
->napi_list
);
2937 #ifdef CONFIG_NETPOLL
2938 spin_lock_init(&napi
->poll_lock
);
2939 napi
->poll_owner
= -1;
2941 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
2943 EXPORT_SYMBOL(netif_napi_add
);
2945 void netif_napi_del(struct napi_struct
*napi
)
2947 struct sk_buff
*skb
, *next
;
2949 list_del_init(&napi
->dev_list
);
2950 napi_free_frags(napi
);
2952 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
2958 napi
->gro_list
= NULL
;
2959 napi
->gro_count
= 0;
2961 EXPORT_SYMBOL(netif_napi_del
);
2964 static void net_rx_action(struct softirq_action
*h
)
2966 struct list_head
*list
= &__get_cpu_var(softnet_data
).poll_list
;
2967 unsigned long time_limit
= jiffies
+ 2;
2968 int budget
= netdev_budget
;
2971 local_irq_disable();
2973 while (!list_empty(list
)) {
2974 struct napi_struct
*n
;
2977 /* If softirq window is exhuasted then punt.
2978 * Allow this to run for 2 jiffies since which will allow
2979 * an average latency of 1.5/HZ.
2981 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
2986 /* Even though interrupts have been re-enabled, this
2987 * access is safe because interrupts can only add new
2988 * entries to the tail of this list, and only ->poll()
2989 * calls can remove this head entry from the list.
2991 n
= list_entry(list
->next
, struct napi_struct
, poll_list
);
2993 have
= netpoll_poll_lock(n
);
2997 /* This NAPI_STATE_SCHED test is for avoiding a race
2998 * with netpoll's poll_napi(). Only the entity which
2999 * obtains the lock and sees NAPI_STATE_SCHED set will
3000 * actually make the ->poll() call. Therefore we avoid
3001 * accidently calling ->poll() when NAPI is not scheduled.
3004 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3005 work
= n
->poll(n
, weight
);
3009 WARN_ON_ONCE(work
> weight
);
3013 local_irq_disable();
3015 /* Drivers must not modify the NAPI state if they
3016 * consume the entire weight. In such cases this code
3017 * still "owns" the NAPI instance and therefore can
3018 * move the instance around on the list at-will.
3020 if (unlikely(work
== weight
)) {
3021 if (unlikely(napi_disable_pending(n
))) {
3024 local_irq_disable();
3026 list_move_tail(&n
->poll_list
, list
);
3029 netpoll_poll_unlock(have
);
3034 #ifdef CONFIG_NET_DMA
3036 * There may not be any more sk_buffs coming right now, so push
3037 * any pending DMA copies to hardware
3039 dma_issue_pending_all();
3045 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
3046 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3050 static gifconf_func_t
*gifconf_list
[NPROTO
];
3053 * register_gifconf - register a SIOCGIF handler
3054 * @family: Address family
3055 * @gifconf: Function handler
3057 * Register protocol dependent address dumping routines. The handler
3058 * that is passed must not be freed or reused until it has been replaced
3059 * by another handler.
3061 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3063 if (family
>= NPROTO
)
3065 gifconf_list
[family
] = gifconf
;
3068 EXPORT_SYMBOL(register_gifconf
);
3072 * Map an interface index to its name (SIOCGIFNAME)
3076 * We need this ioctl for efficient implementation of the
3077 * if_indextoname() function required by the IPv6 API. Without
3078 * it, we would have to search all the interfaces to find a
3082 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3084 struct net_device
*dev
;
3088 * Fetch the caller's info block.
3091 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3095 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3101 strcpy(ifr
.ifr_name
, dev
->name
);
3104 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3110 * Perform a SIOCGIFCONF call. This structure will change
3111 * size eventually, and there is nothing I can do about it.
3112 * Thus we will need a 'compatibility mode'.
3115 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3118 struct net_device
*dev
;
3125 * Fetch the caller's info block.
3128 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3135 * Loop over the interfaces, and write an info block for each.
3139 for_each_netdev(net
, dev
) {
3140 for (i
= 0; i
< NPROTO
; i
++) {
3141 if (gifconf_list
[i
]) {
3144 done
= gifconf_list
[i
](dev
, NULL
, 0);
3146 done
= gifconf_list
[i
](dev
, pos
+ total
,
3156 * All done. Write the updated control block back to the caller.
3158 ifc
.ifc_len
= total
;
3161 * Both BSD and Solaris return 0 here, so we do too.
3163 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3166 #ifdef CONFIG_PROC_FS
3168 * This is invoked by the /proc filesystem handler to display a device
3171 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3174 struct net
*net
= seq_file_net(seq
);
3176 struct net_device
*dev
;
3180 return SEQ_START_TOKEN
;
3183 for_each_netdev_rcu(net
, dev
)
3190 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3192 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3193 first_net_device(seq_file_net(seq
)) :
3194 next_net_device((struct net_device
*)v
);
3197 return rcu_dereference(dev
);
3200 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3206 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3208 const struct net_device_stats
*stats
= dev_get_stats(dev
);
3210 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3211 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3212 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3214 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3215 stats
->rx_fifo_errors
,
3216 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3217 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3218 stats
->rx_compressed
, stats
->multicast
,
3219 stats
->tx_bytes
, stats
->tx_packets
,
3220 stats
->tx_errors
, stats
->tx_dropped
,
3221 stats
->tx_fifo_errors
, stats
->collisions
,
3222 stats
->tx_carrier_errors
+
3223 stats
->tx_aborted_errors
+
3224 stats
->tx_window_errors
+
3225 stats
->tx_heartbeat_errors
,
3226 stats
->tx_compressed
);
3230 * Called from the PROCfs module. This now uses the new arbitrary sized
3231 * /proc/net interface to create /proc/net/dev
3233 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3235 if (v
== SEQ_START_TOKEN
)
3236 seq_puts(seq
, "Inter-| Receive "
3238 " face |bytes packets errs drop fifo frame "
3239 "compressed multicast|bytes packets errs "
3240 "drop fifo colls carrier compressed\n");
3242 dev_seq_printf_stats(seq
, v
);
3246 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
3248 struct netif_rx_stats
*rc
= NULL
;
3250 while (*pos
< nr_cpu_ids
)
3251 if (cpu_online(*pos
)) {
3252 rc
= &per_cpu(netdev_rx_stat
, *pos
);
3259 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3261 return softnet_get_online(pos
);
3264 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3267 return softnet_get_online(pos
);
3270 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3274 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3276 struct netif_rx_stats
*s
= v
;
3278 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3279 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
3280 0, 0, 0, 0, /* was fastroute */
3285 static const struct seq_operations dev_seq_ops
= {
3286 .start
= dev_seq_start
,
3287 .next
= dev_seq_next
,
3288 .stop
= dev_seq_stop
,
3289 .show
= dev_seq_show
,
3292 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3294 return seq_open_net(inode
, file
, &dev_seq_ops
,
3295 sizeof(struct seq_net_private
));
3298 static const struct file_operations dev_seq_fops
= {
3299 .owner
= THIS_MODULE
,
3300 .open
= dev_seq_open
,
3302 .llseek
= seq_lseek
,
3303 .release
= seq_release_net
,
3306 static const struct seq_operations softnet_seq_ops
= {
3307 .start
= softnet_seq_start
,
3308 .next
= softnet_seq_next
,
3309 .stop
= softnet_seq_stop
,
3310 .show
= softnet_seq_show
,
3313 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3315 return seq_open(file
, &softnet_seq_ops
);
3318 static const struct file_operations softnet_seq_fops
= {
3319 .owner
= THIS_MODULE
,
3320 .open
= softnet_seq_open
,
3322 .llseek
= seq_lseek
,
3323 .release
= seq_release
,
3326 static void *ptype_get_idx(loff_t pos
)
3328 struct packet_type
*pt
= NULL
;
3332 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3338 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3339 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3348 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3352 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3355 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3357 struct packet_type
*pt
;
3358 struct list_head
*nxt
;
3362 if (v
== SEQ_START_TOKEN
)
3363 return ptype_get_idx(0);
3366 nxt
= pt
->list
.next
;
3367 if (pt
->type
== htons(ETH_P_ALL
)) {
3368 if (nxt
!= &ptype_all
)
3371 nxt
= ptype_base
[0].next
;
3373 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3375 while (nxt
== &ptype_base
[hash
]) {
3376 if (++hash
>= PTYPE_HASH_SIZE
)
3378 nxt
= ptype_base
[hash
].next
;
3381 return list_entry(nxt
, struct packet_type
, list
);
3384 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3390 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3392 struct packet_type
*pt
= v
;
3394 if (v
== SEQ_START_TOKEN
)
3395 seq_puts(seq
, "Type Device Function\n");
3396 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3397 if (pt
->type
== htons(ETH_P_ALL
))
3398 seq_puts(seq
, "ALL ");
3400 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3402 seq_printf(seq
, " %-8s %pF\n",
3403 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
3409 static const struct seq_operations ptype_seq_ops
= {
3410 .start
= ptype_seq_start
,
3411 .next
= ptype_seq_next
,
3412 .stop
= ptype_seq_stop
,
3413 .show
= ptype_seq_show
,
3416 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
3418 return seq_open_net(inode
, file
, &ptype_seq_ops
,
3419 sizeof(struct seq_net_private
));
3422 static const struct file_operations ptype_seq_fops
= {
3423 .owner
= THIS_MODULE
,
3424 .open
= ptype_seq_open
,
3426 .llseek
= seq_lseek
,
3427 .release
= seq_release_net
,
3431 static int __net_init
dev_proc_net_init(struct net
*net
)
3435 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
3437 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
3439 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
3442 if (wext_proc_init(net
))
3448 proc_net_remove(net
, "ptype");
3450 proc_net_remove(net
, "softnet_stat");
3452 proc_net_remove(net
, "dev");
3456 static void __net_exit
dev_proc_net_exit(struct net
*net
)
3458 wext_proc_exit(net
);
3460 proc_net_remove(net
, "ptype");
3461 proc_net_remove(net
, "softnet_stat");
3462 proc_net_remove(net
, "dev");
3465 static struct pernet_operations __net_initdata dev_proc_ops
= {
3466 .init
= dev_proc_net_init
,
3467 .exit
= dev_proc_net_exit
,
3470 static int __init
dev_proc_init(void)
3472 return register_pernet_subsys(&dev_proc_ops
);
3475 #define dev_proc_init() 0
3476 #endif /* CONFIG_PROC_FS */
3480 * netdev_set_master - set up master/slave pair
3481 * @slave: slave device
3482 * @master: new master device
3484 * Changes the master device of the slave. Pass %NULL to break the
3485 * bonding. The caller must hold the RTNL semaphore. On a failure
3486 * a negative errno code is returned. On success the reference counts
3487 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3488 * function returns zero.
3490 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
3492 struct net_device
*old
= slave
->master
;
3502 slave
->master
= master
;
3510 slave
->flags
|= IFF_SLAVE
;
3512 slave
->flags
&= ~IFF_SLAVE
;
3514 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
3517 EXPORT_SYMBOL(netdev_set_master
);
3519 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
3521 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3523 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
3524 ops
->ndo_change_rx_flags(dev
, flags
);
3527 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
3529 unsigned short old_flags
= dev
->flags
;
3535 dev
->flags
|= IFF_PROMISC
;
3536 dev
->promiscuity
+= inc
;
3537 if (dev
->promiscuity
== 0) {
3540 * If inc causes overflow, untouch promisc and return error.
3543 dev
->flags
&= ~IFF_PROMISC
;
3545 dev
->promiscuity
-= inc
;
3546 printk(KERN_WARNING
"%s: promiscuity touches roof, "
3547 "set promiscuity failed, promiscuity feature "
3548 "of device might be broken.\n", dev
->name
);
3552 if (dev
->flags
!= old_flags
) {
3553 printk(KERN_INFO
"device %s %s promiscuous mode\n",
3554 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
3556 if (audit_enabled
) {
3557 current_uid_gid(&uid
, &gid
);
3558 audit_log(current
->audit_context
, GFP_ATOMIC
,
3559 AUDIT_ANOM_PROMISCUOUS
,
3560 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3561 dev
->name
, (dev
->flags
& IFF_PROMISC
),
3562 (old_flags
& IFF_PROMISC
),
3563 audit_get_loginuid(current
),
3565 audit_get_sessionid(current
));
3568 dev_change_rx_flags(dev
, IFF_PROMISC
);
3574 * dev_set_promiscuity - update promiscuity count on a device
3578 * Add or remove promiscuity from a device. While the count in the device
3579 * remains above zero the interface remains promiscuous. Once it hits zero
3580 * the device reverts back to normal filtering operation. A negative inc
3581 * value is used to drop promiscuity on the device.
3582 * Return 0 if successful or a negative errno code on error.
3584 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
3586 unsigned short old_flags
= dev
->flags
;
3589 err
= __dev_set_promiscuity(dev
, inc
);
3592 if (dev
->flags
!= old_flags
)
3593 dev_set_rx_mode(dev
);
3596 EXPORT_SYMBOL(dev_set_promiscuity
);
3599 * dev_set_allmulti - update allmulti count on a device
3603 * Add or remove reception of all multicast frames to a device. While the
3604 * count in the device remains above zero the interface remains listening
3605 * to all interfaces. Once it hits zero the device reverts back to normal
3606 * filtering operation. A negative @inc value is used to drop the counter
3607 * when releasing a resource needing all multicasts.
3608 * Return 0 if successful or a negative errno code on error.
3611 int dev_set_allmulti(struct net_device
*dev
, int inc
)
3613 unsigned short old_flags
= dev
->flags
;
3617 dev
->flags
|= IFF_ALLMULTI
;
3618 dev
->allmulti
+= inc
;
3619 if (dev
->allmulti
== 0) {
3622 * If inc causes overflow, untouch allmulti and return error.
3625 dev
->flags
&= ~IFF_ALLMULTI
;
3627 dev
->allmulti
-= inc
;
3628 printk(KERN_WARNING
"%s: allmulti touches roof, "
3629 "set allmulti failed, allmulti feature of "
3630 "device might be broken.\n", dev
->name
);
3634 if (dev
->flags
^ old_flags
) {
3635 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
3636 dev_set_rx_mode(dev
);
3640 EXPORT_SYMBOL(dev_set_allmulti
);
3643 * Upload unicast and multicast address lists to device and
3644 * configure RX filtering. When the device doesn't support unicast
3645 * filtering it is put in promiscuous mode while unicast addresses
3648 void __dev_set_rx_mode(struct net_device
*dev
)
3650 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3652 /* dev_open will call this function so the list will stay sane. */
3653 if (!(dev
->flags
&IFF_UP
))
3656 if (!netif_device_present(dev
))
3659 if (ops
->ndo_set_rx_mode
)
3660 ops
->ndo_set_rx_mode(dev
);
3662 /* Unicast addresses changes may only happen under the rtnl,
3663 * therefore calling __dev_set_promiscuity here is safe.
3665 if (dev
->uc
.count
> 0 && !dev
->uc_promisc
) {
3666 __dev_set_promiscuity(dev
, 1);
3667 dev
->uc_promisc
= 1;
3668 } else if (dev
->uc
.count
== 0 && dev
->uc_promisc
) {
3669 __dev_set_promiscuity(dev
, -1);
3670 dev
->uc_promisc
= 0;
3673 if (ops
->ndo_set_multicast_list
)
3674 ops
->ndo_set_multicast_list(dev
);
3678 void dev_set_rx_mode(struct net_device
*dev
)
3680 netif_addr_lock_bh(dev
);
3681 __dev_set_rx_mode(dev
);
3682 netif_addr_unlock_bh(dev
);
3685 /* hw addresses list handling functions */
3687 static int __hw_addr_add(struct netdev_hw_addr_list
*list
, unsigned char *addr
,
3688 int addr_len
, unsigned char addr_type
)
3690 struct netdev_hw_addr
*ha
;
3693 if (addr_len
> MAX_ADDR_LEN
)
3696 list_for_each_entry(ha
, &list
->list
, list
) {
3697 if (!memcmp(ha
->addr
, addr
, addr_len
) &&
3698 ha
->type
== addr_type
) {
3705 alloc_size
= sizeof(*ha
);
3706 if (alloc_size
< L1_CACHE_BYTES
)
3707 alloc_size
= L1_CACHE_BYTES
;
3708 ha
= kmalloc(alloc_size
, GFP_ATOMIC
);
3711 memcpy(ha
->addr
, addr
, addr_len
);
3712 ha
->type
= addr_type
;
3715 list_add_tail_rcu(&ha
->list
, &list
->list
);
3720 static void ha_rcu_free(struct rcu_head
*head
)
3722 struct netdev_hw_addr
*ha
;
3724 ha
= container_of(head
, struct netdev_hw_addr
, rcu_head
);
3728 static int __hw_addr_del(struct netdev_hw_addr_list
*list
, unsigned char *addr
,
3729 int addr_len
, unsigned char addr_type
)
3731 struct netdev_hw_addr
*ha
;
3733 list_for_each_entry(ha
, &list
->list
, list
) {
3734 if (!memcmp(ha
->addr
, addr
, addr_len
) &&
3735 (ha
->type
== addr_type
|| !addr_type
)) {
3738 list_del_rcu(&ha
->list
);
3739 call_rcu(&ha
->rcu_head
, ha_rcu_free
);
3747 static int __hw_addr_add_multiple(struct netdev_hw_addr_list
*to_list
,
3748 struct netdev_hw_addr_list
*from_list
,
3750 unsigned char addr_type
)
3753 struct netdev_hw_addr
*ha
, *ha2
;
3756 list_for_each_entry(ha
, &from_list
->list
, list
) {
3757 type
= addr_type
? addr_type
: ha
->type
;
3758 err
= __hw_addr_add(to_list
, ha
->addr
, addr_len
, type
);
3765 list_for_each_entry(ha2
, &from_list
->list
, list
) {
3768 type
= addr_type
? addr_type
: ha2
->type
;
3769 __hw_addr_del(to_list
, ha2
->addr
, addr_len
, type
);
3774 static void __hw_addr_del_multiple(struct netdev_hw_addr_list
*to_list
,
3775 struct netdev_hw_addr_list
*from_list
,
3777 unsigned char addr_type
)
3779 struct netdev_hw_addr
*ha
;
3782 list_for_each_entry(ha
, &from_list
->list
, list
) {
3783 type
= addr_type
? addr_type
: ha
->type
;
3784 __hw_addr_del(to_list
, ha
->addr
, addr_len
, addr_type
);
3788 static int __hw_addr_sync(struct netdev_hw_addr_list
*to_list
,
3789 struct netdev_hw_addr_list
*from_list
,
3793 struct netdev_hw_addr
*ha
, *tmp
;
3795 list_for_each_entry_safe(ha
, tmp
, &from_list
->list
, list
) {
3797 err
= __hw_addr_add(to_list
, ha
->addr
,
3798 addr_len
, ha
->type
);
3803 } else if (ha
->refcount
== 1) {
3804 __hw_addr_del(to_list
, ha
->addr
, addr_len
, ha
->type
);
3805 __hw_addr_del(from_list
, ha
->addr
, addr_len
, ha
->type
);
3811 static void __hw_addr_unsync(struct netdev_hw_addr_list
*to_list
,
3812 struct netdev_hw_addr_list
*from_list
,
3815 struct netdev_hw_addr
*ha
, *tmp
;
3817 list_for_each_entry_safe(ha
, tmp
, &from_list
->list
, list
) {
3819 __hw_addr_del(to_list
, ha
->addr
,
3820 addr_len
, ha
->type
);
3822 __hw_addr_del(from_list
, ha
->addr
,
3823 addr_len
, ha
->type
);
3828 static void __hw_addr_flush(struct netdev_hw_addr_list
*list
)
3830 struct netdev_hw_addr
*ha
, *tmp
;
3832 list_for_each_entry_safe(ha
, tmp
, &list
->list
, list
) {
3833 list_del_rcu(&ha
->list
);
3834 call_rcu(&ha
->rcu_head
, ha_rcu_free
);
3839 static void __hw_addr_init(struct netdev_hw_addr_list
*list
)
3841 INIT_LIST_HEAD(&list
->list
);
3845 /* Device addresses handling functions */
3847 static void dev_addr_flush(struct net_device
*dev
)
3849 /* rtnl_mutex must be held here */
3851 __hw_addr_flush(&dev
->dev_addrs
);
3852 dev
->dev_addr
= NULL
;
3855 static int dev_addr_init(struct net_device
*dev
)
3857 unsigned char addr
[MAX_ADDR_LEN
];
3858 struct netdev_hw_addr
*ha
;
3861 /* rtnl_mutex must be held here */
3863 __hw_addr_init(&dev
->dev_addrs
);
3864 memset(addr
, 0, sizeof(addr
));
3865 err
= __hw_addr_add(&dev
->dev_addrs
, addr
, sizeof(addr
),
3866 NETDEV_HW_ADDR_T_LAN
);
3869 * Get the first (previously created) address from the list
3870 * and set dev_addr pointer to this location.
3872 ha
= list_first_entry(&dev
->dev_addrs
.list
,
3873 struct netdev_hw_addr
, list
);
3874 dev
->dev_addr
= ha
->addr
;
3880 * dev_addr_add - Add a device address
3882 * @addr: address to add
3883 * @addr_type: address type
3885 * Add a device address to the device or increase the reference count if
3886 * it already exists.
3888 * The caller must hold the rtnl_mutex.
3890 int dev_addr_add(struct net_device
*dev
, unsigned char *addr
,
3891 unsigned char addr_type
)
3897 err
= __hw_addr_add(&dev
->dev_addrs
, addr
, dev
->addr_len
, addr_type
);
3899 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3902 EXPORT_SYMBOL(dev_addr_add
);
3905 * dev_addr_del - Release a device address.
3907 * @addr: address to delete
3908 * @addr_type: address type
3910 * Release reference to a device address and remove it from the device
3911 * if the reference count drops to zero.
3913 * The caller must hold the rtnl_mutex.
3915 int dev_addr_del(struct net_device
*dev
, unsigned char *addr
,
3916 unsigned char addr_type
)
3919 struct netdev_hw_addr
*ha
;
3924 * We can not remove the first address from the list because
3925 * dev->dev_addr points to that.
3927 ha
= list_first_entry(&dev
->dev_addrs
.list
,
3928 struct netdev_hw_addr
, list
);
3929 if (ha
->addr
== dev
->dev_addr
&& ha
->refcount
== 1)
3932 err
= __hw_addr_del(&dev
->dev_addrs
, addr
, dev
->addr_len
,
3935 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3938 EXPORT_SYMBOL(dev_addr_del
);
3941 * dev_addr_add_multiple - Add device addresses from another device
3942 * @to_dev: device to which addresses will be added
3943 * @from_dev: device from which addresses will be added
3944 * @addr_type: address type - 0 means type will be used from from_dev
3946 * Add device addresses of the one device to another.
3948 * The caller must hold the rtnl_mutex.
3950 int dev_addr_add_multiple(struct net_device
*to_dev
,
3951 struct net_device
*from_dev
,
3952 unsigned char addr_type
)
3958 if (from_dev
->addr_len
!= to_dev
->addr_len
)
3960 err
= __hw_addr_add_multiple(&to_dev
->dev_addrs
, &from_dev
->dev_addrs
,
3961 to_dev
->addr_len
, addr_type
);
3963 call_netdevice_notifiers(NETDEV_CHANGEADDR
, to_dev
);
3966 EXPORT_SYMBOL(dev_addr_add_multiple
);
3969 * dev_addr_del_multiple - Delete device addresses by another device
3970 * @to_dev: device where the addresses will be deleted
3971 * @from_dev: device by which addresses the addresses will be deleted
3972 * @addr_type: address type - 0 means type will used from from_dev
3974 * Deletes addresses in to device by the list of addresses in from device.
3976 * The caller must hold the rtnl_mutex.
3978 int dev_addr_del_multiple(struct net_device
*to_dev
,
3979 struct net_device
*from_dev
,
3980 unsigned char addr_type
)
3984 if (from_dev
->addr_len
!= to_dev
->addr_len
)
3986 __hw_addr_del_multiple(&to_dev
->dev_addrs
, &from_dev
->dev_addrs
,
3987 to_dev
->addr_len
, addr_type
);
3988 call_netdevice_notifiers(NETDEV_CHANGEADDR
, to_dev
);
3991 EXPORT_SYMBOL(dev_addr_del_multiple
);
3993 /* multicast addresses handling functions */
3995 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
3996 void *addr
, int alen
, int glbl
)
3998 struct dev_addr_list
*da
;
4000 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
4001 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
4002 alen
== da
->da_addrlen
) {
4004 int old_glbl
= da
->da_gusers
;
4021 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
4022 void *addr
, int alen
, int glbl
)
4024 struct dev_addr_list
*da
;
4026 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
4027 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
4028 da
->da_addrlen
== alen
) {
4030 int old_glbl
= da
->da_gusers
;
4040 da
= kzalloc(sizeof(*da
), GFP_ATOMIC
);
4043 memcpy(da
->da_addr
, addr
, alen
);
4044 da
->da_addrlen
= alen
;
4046 da
->da_gusers
= glbl
? 1 : 0;
4054 * dev_unicast_delete - Release secondary unicast address.
4056 * @addr: address to delete
4058 * Release reference to a secondary unicast address and remove it
4059 * from the device if the reference count drops to zero.
4061 * The caller must hold the rtnl_mutex.
4063 int dev_unicast_delete(struct net_device
*dev
, void *addr
)
4069 netif_addr_lock_bh(dev
);
4070 err
= __hw_addr_del(&dev
->uc
, addr
, dev
->addr_len
,
4071 NETDEV_HW_ADDR_T_UNICAST
);
4073 __dev_set_rx_mode(dev
);
4074 netif_addr_unlock_bh(dev
);
4077 EXPORT_SYMBOL(dev_unicast_delete
);
4080 * dev_unicast_add - add a secondary unicast address
4082 * @addr: address to add
4084 * Add a secondary unicast address to the device or increase
4085 * the reference count if it already exists.
4087 * The caller must hold the rtnl_mutex.
4089 int dev_unicast_add(struct net_device
*dev
, void *addr
)
4095 netif_addr_lock_bh(dev
);
4096 err
= __hw_addr_add(&dev
->uc
, addr
, dev
->addr_len
,
4097 NETDEV_HW_ADDR_T_UNICAST
);
4099 __dev_set_rx_mode(dev
);
4100 netif_addr_unlock_bh(dev
);
4103 EXPORT_SYMBOL(dev_unicast_add
);
4105 int __dev_addr_sync(struct dev_addr_list
**to
, int *to_count
,
4106 struct dev_addr_list
**from
, int *from_count
)
4108 struct dev_addr_list
*da
, *next
;
4112 while (da
!= NULL
) {
4114 if (!da
->da_synced
) {
4115 err
= __dev_addr_add(to
, to_count
,
4116 da
->da_addr
, da
->da_addrlen
, 0);
4121 } else if (da
->da_users
== 1) {
4122 __dev_addr_delete(to
, to_count
,
4123 da
->da_addr
, da
->da_addrlen
, 0);
4124 __dev_addr_delete(from
, from_count
,
4125 da
->da_addr
, da
->da_addrlen
, 0);
4131 EXPORT_SYMBOL_GPL(__dev_addr_sync
);
4133 void __dev_addr_unsync(struct dev_addr_list
**to
, int *to_count
,
4134 struct dev_addr_list
**from
, int *from_count
)
4136 struct dev_addr_list
*da
, *next
;
4139 while (da
!= NULL
) {
4141 if (da
->da_synced
) {
4142 __dev_addr_delete(to
, to_count
,
4143 da
->da_addr
, da
->da_addrlen
, 0);
4145 __dev_addr_delete(from
, from_count
,
4146 da
->da_addr
, da
->da_addrlen
, 0);
4151 EXPORT_SYMBOL_GPL(__dev_addr_unsync
);
4154 * dev_unicast_sync - Synchronize device's unicast list to another device
4155 * @to: destination device
4156 * @from: source device
4158 * Add newly added addresses to the destination device and release
4159 * addresses that have no users left. The source device must be
4160 * locked by netif_tx_lock_bh.
4162 * This function is intended to be called from the dev->set_rx_mode
4163 * function of layered software devices.
4165 int dev_unicast_sync(struct net_device
*to
, struct net_device
*from
)
4169 if (to
->addr_len
!= from
->addr_len
)
4172 netif_addr_lock_bh(to
);
4173 err
= __hw_addr_sync(&to
->uc
, &from
->uc
, to
->addr_len
);
4175 __dev_set_rx_mode(to
);
4176 netif_addr_unlock_bh(to
);
4179 EXPORT_SYMBOL(dev_unicast_sync
);
4182 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4183 * @to: destination device
4184 * @from: source device
4186 * Remove all addresses that were added to the destination device by
4187 * dev_unicast_sync(). This function is intended to be called from the
4188 * dev->stop function of layered software devices.
4190 void dev_unicast_unsync(struct net_device
*to
, struct net_device
*from
)
4192 if (to
->addr_len
!= from
->addr_len
)
4195 netif_addr_lock_bh(from
);
4196 netif_addr_lock(to
);
4197 __hw_addr_unsync(&to
->uc
, &from
->uc
, to
->addr_len
);
4198 __dev_set_rx_mode(to
);
4199 netif_addr_unlock(to
);
4200 netif_addr_unlock_bh(from
);
4202 EXPORT_SYMBOL(dev_unicast_unsync
);
4204 static void dev_unicast_flush(struct net_device
*dev
)
4206 netif_addr_lock_bh(dev
);
4207 __hw_addr_flush(&dev
->uc
);
4208 netif_addr_unlock_bh(dev
);
4211 static void dev_unicast_init(struct net_device
*dev
)
4213 __hw_addr_init(&dev
->uc
);
4217 static void __dev_addr_discard(struct dev_addr_list
**list
)
4219 struct dev_addr_list
*tmp
;
4221 while (*list
!= NULL
) {
4224 if (tmp
->da_users
> tmp
->da_gusers
)
4225 printk("__dev_addr_discard: address leakage! "
4226 "da_users=%d\n", tmp
->da_users
);
4231 static void dev_addr_discard(struct net_device
*dev
)
4233 netif_addr_lock_bh(dev
);
4235 __dev_addr_discard(&dev
->mc_list
);
4238 netif_addr_unlock_bh(dev
);
4242 * dev_get_flags - get flags reported to userspace
4245 * Get the combination of flag bits exported through APIs to userspace.
4247 unsigned dev_get_flags(const struct net_device
*dev
)
4251 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4256 (dev
->gflags
& (IFF_PROMISC
|
4259 if (netif_running(dev
)) {
4260 if (netif_oper_up(dev
))
4261 flags
|= IFF_RUNNING
;
4262 if (netif_carrier_ok(dev
))
4263 flags
|= IFF_LOWER_UP
;
4264 if (netif_dormant(dev
))
4265 flags
|= IFF_DORMANT
;
4270 EXPORT_SYMBOL(dev_get_flags
);
4273 * dev_change_flags - change device settings
4275 * @flags: device state flags
4277 * Change settings on device based state flags. The flags are
4278 * in the userspace exported format.
4280 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4283 int old_flags
= dev
->flags
;
4288 * Set the flags on our device.
4291 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4292 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4294 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4298 * Load in the correct multicast list now the flags have changed.
4301 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4302 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4304 dev_set_rx_mode(dev
);
4307 * Have we downed the interface. We handle IFF_UP ourselves
4308 * according to user attempts to set it, rather than blindly
4313 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4314 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
4317 dev_set_rx_mode(dev
);
4320 if (dev
->flags
& IFF_UP
&&
4321 ((old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
4323 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4325 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4326 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4328 dev
->gflags
^= IFF_PROMISC
;
4329 dev_set_promiscuity(dev
, inc
);
4332 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4333 is important. Some (broken) drivers set IFF_PROMISC, when
4334 IFF_ALLMULTI is requested not asking us and not reporting.
4336 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4337 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4339 dev
->gflags
^= IFF_ALLMULTI
;
4340 dev_set_allmulti(dev
, inc
);
4343 /* Exclude state transition flags, already notified */
4344 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
4346 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4350 EXPORT_SYMBOL(dev_change_flags
);
4353 * dev_set_mtu - Change maximum transfer unit
4355 * @new_mtu: new transfer unit
4357 * Change the maximum transfer size of the network device.
4359 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4361 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4364 if (new_mtu
== dev
->mtu
)
4367 /* MTU must be positive. */
4371 if (!netif_device_present(dev
))
4375 if (ops
->ndo_change_mtu
)
4376 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4380 if (!err
&& dev
->flags
& IFF_UP
)
4381 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4384 EXPORT_SYMBOL(dev_set_mtu
);
4387 * dev_set_mac_address - Change Media Access Control Address
4391 * Change the hardware (MAC) address of the device
4393 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4395 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4398 if (!ops
->ndo_set_mac_address
)
4400 if (sa
->sa_family
!= dev
->type
)
4402 if (!netif_device_present(dev
))
4404 err
= ops
->ndo_set_mac_address(dev
, sa
);
4406 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4409 EXPORT_SYMBOL(dev_set_mac_address
);
4412 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4414 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4417 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4423 case SIOCGIFFLAGS
: /* Get interface flags */
4424 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4427 case SIOCGIFMETRIC
: /* Get the metric on the interface
4428 (currently unused) */
4429 ifr
->ifr_metric
= 0;
4432 case SIOCGIFMTU
: /* Get the MTU of a device */
4433 ifr
->ifr_mtu
= dev
->mtu
;
4438 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4440 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4441 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4442 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4450 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4451 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4452 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4453 ifr
->ifr_map
.irq
= dev
->irq
;
4454 ifr
->ifr_map
.dma
= dev
->dma
;
4455 ifr
->ifr_map
.port
= dev
->if_port
;
4459 ifr
->ifr_ifindex
= dev
->ifindex
;
4463 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4467 /* dev_ioctl() should ensure this case
4479 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4481 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4484 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4485 const struct net_device_ops
*ops
;
4490 ops
= dev
->netdev_ops
;
4493 case SIOCSIFFLAGS
: /* Set interface flags */
4494 return dev_change_flags(dev
, ifr
->ifr_flags
);
4496 case SIOCSIFMETRIC
: /* Set the metric on the interface
4497 (currently unused) */
4500 case SIOCSIFMTU
: /* Set the MTU of a device */
4501 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4504 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4506 case SIOCSIFHWBROADCAST
:
4507 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4509 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4510 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4511 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4515 if (ops
->ndo_set_config
) {
4516 if (!netif_device_present(dev
))
4518 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4523 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4524 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4526 if (!netif_device_present(dev
))
4528 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
4532 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4533 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4535 if (!netif_device_present(dev
))
4537 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
4541 if (ifr
->ifr_qlen
< 0)
4543 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4547 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4548 return dev_change_name(dev
, ifr
->ifr_newname
);
4551 * Unknown or private ioctl
4554 if ((cmd
>= SIOCDEVPRIVATE
&&
4555 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4556 cmd
== SIOCBONDENSLAVE
||
4557 cmd
== SIOCBONDRELEASE
||
4558 cmd
== SIOCBONDSETHWADDR
||
4559 cmd
== SIOCBONDSLAVEINFOQUERY
||
4560 cmd
== SIOCBONDINFOQUERY
||
4561 cmd
== SIOCBONDCHANGEACTIVE
||
4562 cmd
== SIOCGMIIPHY
||
4563 cmd
== SIOCGMIIREG
||
4564 cmd
== SIOCSMIIREG
||
4565 cmd
== SIOCBRADDIF
||
4566 cmd
== SIOCBRDELIF
||
4567 cmd
== SIOCSHWTSTAMP
||
4568 cmd
== SIOCWANDEV
) {
4570 if (ops
->ndo_do_ioctl
) {
4571 if (netif_device_present(dev
))
4572 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4584 * This function handles all "interface"-type I/O control requests. The actual
4585 * 'doing' part of this is dev_ifsioc above.
4589 * dev_ioctl - network device ioctl
4590 * @net: the applicable net namespace
4591 * @cmd: command to issue
4592 * @arg: pointer to a struct ifreq in user space
4594 * Issue ioctl functions to devices. This is normally called by the
4595 * user space syscall interfaces but can sometimes be useful for
4596 * other purposes. The return value is the return from the syscall if
4597 * positive or a negative errno code on error.
4600 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4606 /* One special case: SIOCGIFCONF takes ifconf argument
4607 and requires shared lock, because it sleeps writing
4611 if (cmd
== SIOCGIFCONF
) {
4613 ret
= dev_ifconf(net
, (char __user
*) arg
);
4617 if (cmd
== SIOCGIFNAME
)
4618 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4620 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4623 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4625 colon
= strchr(ifr
.ifr_name
, ':');
4630 * See which interface the caller is talking about.
4635 * These ioctl calls:
4636 * - can be done by all.
4637 * - atomic and do not require locking.
4648 dev_load(net
, ifr
.ifr_name
);
4650 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4655 if (copy_to_user(arg
, &ifr
,
4656 sizeof(struct ifreq
)))
4662 dev_load(net
, ifr
.ifr_name
);
4664 ret
= dev_ethtool(net
, &ifr
);
4669 if (copy_to_user(arg
, &ifr
,
4670 sizeof(struct ifreq
)))
4676 * These ioctl calls:
4677 * - require superuser power.
4678 * - require strict serialization.
4684 if (!capable(CAP_NET_ADMIN
))
4686 dev_load(net
, ifr
.ifr_name
);
4688 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4693 if (copy_to_user(arg
, &ifr
,
4694 sizeof(struct ifreq
)))
4700 * These ioctl calls:
4701 * - require superuser power.
4702 * - require strict serialization.
4703 * - do not return a value
4713 case SIOCSIFHWBROADCAST
:
4716 case SIOCBONDENSLAVE
:
4717 case SIOCBONDRELEASE
:
4718 case SIOCBONDSETHWADDR
:
4719 case SIOCBONDCHANGEACTIVE
:
4723 if (!capable(CAP_NET_ADMIN
))
4726 case SIOCBONDSLAVEINFOQUERY
:
4727 case SIOCBONDINFOQUERY
:
4728 dev_load(net
, ifr
.ifr_name
);
4730 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4735 /* Get the per device memory space. We can add this but
4736 * currently do not support it */
4738 /* Set the per device memory buffer space.
4739 * Not applicable in our case */
4744 * Unknown or private ioctl.
4747 if (cmd
== SIOCWANDEV
||
4748 (cmd
>= SIOCDEVPRIVATE
&&
4749 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4750 dev_load(net
, ifr
.ifr_name
);
4752 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4754 if (!ret
&& copy_to_user(arg
, &ifr
,
4755 sizeof(struct ifreq
)))
4759 /* Take care of Wireless Extensions */
4760 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4761 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4768 * dev_new_index - allocate an ifindex
4769 * @net: the applicable net namespace
4771 * Returns a suitable unique value for a new device interface
4772 * number. The caller must hold the rtnl semaphore or the
4773 * dev_base_lock to be sure it remains unique.
4775 static int dev_new_index(struct net
*net
)
4781 if (!__dev_get_by_index(net
, ifindex
))
4786 /* Delayed registration/unregisteration */
4787 static LIST_HEAD(net_todo_list
);
4789 static void net_set_todo(struct net_device
*dev
)
4791 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4794 static void rollback_registered_many(struct list_head
*head
)
4796 struct net_device
*dev
, *tmp
;
4798 BUG_ON(dev_boot_phase
);
4801 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4802 /* Some devices call without registering
4803 * for initialization unwind. Remove those
4804 * devices and proceed with the remaining.
4806 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4807 pr_debug("unregister_netdevice: device %s/%p never "
4808 "was registered\n", dev
->name
, dev
);
4811 list_del(&dev
->unreg_list
);
4815 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4817 /* If device is running, close it first. */
4820 /* And unlink it from device chain. */
4821 unlist_netdevice(dev
);
4823 dev
->reg_state
= NETREG_UNREGISTERING
;
4828 list_for_each_entry(dev
, head
, unreg_list
) {
4829 /* Shutdown queueing discipline. */
4833 /* Notify protocols, that we are about to destroy
4834 this device. They should clean all the things.
4836 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4839 * Flush the unicast and multicast chains
4841 dev_unicast_flush(dev
);
4842 dev_addr_discard(dev
);
4844 if (dev
->netdev_ops
->ndo_uninit
)
4845 dev
->netdev_ops
->ndo_uninit(dev
);
4847 /* Notifier chain MUST detach us from master device. */
4848 WARN_ON(dev
->master
);
4850 /* Remove entries from kobject tree */
4851 netdev_unregister_kobject(dev
);
4854 /* Process any work delayed until the end of the batch */
4855 dev
= list_entry(head
->next
, struct net_device
, unreg_list
);
4856 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
4860 list_for_each_entry(dev
, head
, unreg_list
)
4864 static void rollback_registered(struct net_device
*dev
)
4868 list_add(&dev
->unreg_list
, &single
);
4869 rollback_registered_many(&single
);
4872 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
4873 struct netdev_queue
*dev_queue
,
4876 spin_lock_init(&dev_queue
->_xmit_lock
);
4877 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
4878 dev_queue
->xmit_lock_owner
= -1;
4881 static void netdev_init_queue_locks(struct net_device
*dev
)
4883 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
4884 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
4887 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
4889 /* Fix illegal SG+CSUM combinations. */
4890 if ((features
& NETIF_F_SG
) &&
4891 !(features
& NETIF_F_ALL_CSUM
)) {
4893 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
4894 "checksum feature.\n", name
);
4895 features
&= ~NETIF_F_SG
;
4898 /* TSO requires that SG is present as well. */
4899 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
4901 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
4902 "SG feature.\n", name
);
4903 features
&= ~NETIF_F_TSO
;
4906 if (features
& NETIF_F_UFO
) {
4907 if (!(features
& NETIF_F_GEN_CSUM
)) {
4909 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4910 "since no NETIF_F_HW_CSUM feature.\n",
4912 features
&= ~NETIF_F_UFO
;
4915 if (!(features
& NETIF_F_SG
)) {
4917 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4918 "since no NETIF_F_SG feature.\n", name
);
4919 features
&= ~NETIF_F_UFO
;
4925 EXPORT_SYMBOL(netdev_fix_features
);
4928 * netif_stacked_transfer_operstate - transfer operstate
4929 * @rootdev: the root or lower level device to transfer state from
4930 * @dev: the device to transfer operstate to
4932 * Transfer operational state from root to device. This is normally
4933 * called when a stacking relationship exists between the root
4934 * device and the device(a leaf device).
4936 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
4937 struct net_device
*dev
)
4939 if (rootdev
->operstate
== IF_OPER_DORMANT
)
4940 netif_dormant_on(dev
);
4942 netif_dormant_off(dev
);
4944 if (netif_carrier_ok(rootdev
)) {
4945 if (!netif_carrier_ok(dev
))
4946 netif_carrier_on(dev
);
4948 if (netif_carrier_ok(dev
))
4949 netif_carrier_off(dev
);
4952 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
4955 * register_netdevice - register a network device
4956 * @dev: device to register
4958 * Take a completed network device structure and add it to the kernel
4959 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4960 * chain. 0 is returned on success. A negative errno code is returned
4961 * on a failure to set up the device, or if the name is a duplicate.
4963 * Callers must hold the rtnl semaphore. You may want
4964 * register_netdev() instead of this.
4967 * The locking appears insufficient to guarantee two parallel registers
4968 * will not get the same name.
4971 int register_netdevice(struct net_device
*dev
)
4974 struct net
*net
= dev_net(dev
);
4976 BUG_ON(dev_boot_phase
);
4981 /* When net_device's are persistent, this will be fatal. */
4982 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
4985 spin_lock_init(&dev
->addr_list_lock
);
4986 netdev_set_addr_lockdep_class(dev
);
4987 netdev_init_queue_locks(dev
);
4991 /* Init, if this function is available */
4992 if (dev
->netdev_ops
->ndo_init
) {
4993 ret
= dev
->netdev_ops
->ndo_init(dev
);
5001 ret
= dev_get_valid_name(net
, dev
->name
, dev
->name
, 0);
5005 dev
->ifindex
= dev_new_index(net
);
5006 if (dev
->iflink
== -1)
5007 dev
->iflink
= dev
->ifindex
;
5009 /* Fix illegal checksum combinations */
5010 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
5011 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5012 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5014 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5017 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5018 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5019 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5021 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5024 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5026 /* Enable software GSO if SG is supported. */
5027 if (dev
->features
& NETIF_F_SG
)
5028 dev
->features
|= NETIF_F_GSO
;
5030 netdev_initialize_kobject(dev
);
5032 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5033 ret
= notifier_to_errno(ret
);
5037 ret
= netdev_register_kobject(dev
);
5040 dev
->reg_state
= NETREG_REGISTERED
;
5043 * Default initial state at registry is that the
5044 * device is present.
5047 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5049 dev_init_scheduler(dev
);
5051 list_netdevice(dev
);
5053 /* Notify protocols, that a new device appeared. */
5054 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5055 ret
= notifier_to_errno(ret
);
5057 rollback_registered(dev
);
5058 dev
->reg_state
= NETREG_UNREGISTERED
;
5061 * Prevent userspace races by waiting until the network
5062 * device is fully setup before sending notifications.
5064 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5070 if (dev
->netdev_ops
->ndo_uninit
)
5071 dev
->netdev_ops
->ndo_uninit(dev
);
5074 EXPORT_SYMBOL(register_netdevice
);
5077 * init_dummy_netdev - init a dummy network device for NAPI
5078 * @dev: device to init
5080 * This takes a network device structure and initialize the minimum
5081 * amount of fields so it can be used to schedule NAPI polls without
5082 * registering a full blown interface. This is to be used by drivers
5083 * that need to tie several hardware interfaces to a single NAPI
5084 * poll scheduler due to HW limitations.
5086 int init_dummy_netdev(struct net_device
*dev
)
5088 /* Clear everything. Note we don't initialize spinlocks
5089 * are they aren't supposed to be taken by any of the
5090 * NAPI code and this dummy netdev is supposed to be
5091 * only ever used for NAPI polls
5093 memset(dev
, 0, sizeof(struct net_device
));
5095 /* make sure we BUG if trying to hit standard
5096 * register/unregister code path
5098 dev
->reg_state
= NETREG_DUMMY
;
5100 /* initialize the ref count */
5101 atomic_set(&dev
->refcnt
, 1);
5103 /* NAPI wants this */
5104 INIT_LIST_HEAD(&dev
->napi_list
);
5106 /* a dummy interface is started by default */
5107 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5108 set_bit(__LINK_STATE_START
, &dev
->state
);
5112 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5116 * register_netdev - register a network device
5117 * @dev: device to register
5119 * Take a completed network device structure and add it to the kernel
5120 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5121 * chain. 0 is returned on success. A negative errno code is returned
5122 * on a failure to set up the device, or if the name is a duplicate.
5124 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5125 * and expands the device name if you passed a format string to
5128 int register_netdev(struct net_device
*dev
)
5135 * If the name is a format string the caller wants us to do a
5138 if (strchr(dev
->name
, '%')) {
5139 err
= dev_alloc_name(dev
, dev
->name
);
5144 err
= register_netdevice(dev
);
5149 EXPORT_SYMBOL(register_netdev
);
5152 * netdev_wait_allrefs - wait until all references are gone.
5154 * This is called when unregistering network devices.
5156 * Any protocol or device that holds a reference should register
5157 * for netdevice notification, and cleanup and put back the
5158 * reference if they receive an UNREGISTER event.
5159 * We can get stuck here if buggy protocols don't correctly
5162 static void netdev_wait_allrefs(struct net_device
*dev
)
5164 unsigned long rebroadcast_time
, warning_time
;
5166 linkwatch_forget_dev(dev
);
5168 rebroadcast_time
= warning_time
= jiffies
;
5169 while (atomic_read(&dev
->refcnt
) != 0) {
5170 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5173 /* Rebroadcast unregister notification */
5174 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5175 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5176 * should have already handle it the first time */
5178 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5180 /* We must not have linkwatch events
5181 * pending on unregister. If this
5182 * happens, we simply run the queue
5183 * unscheduled, resulting in a noop
5186 linkwatch_run_queue();
5191 rebroadcast_time
= jiffies
;
5196 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5197 printk(KERN_EMERG
"unregister_netdevice: "
5198 "waiting for %s to become free. Usage "
5200 dev
->name
, atomic_read(&dev
->refcnt
));
5201 warning_time
= jiffies
;
5210 * register_netdevice(x1);
5211 * register_netdevice(x2);
5213 * unregister_netdevice(y1);
5214 * unregister_netdevice(y2);
5220 * We are invoked by rtnl_unlock().
5221 * This allows us to deal with problems:
5222 * 1) We can delete sysfs objects which invoke hotplug
5223 * without deadlocking with linkwatch via keventd.
5224 * 2) Since we run with the RTNL semaphore not held, we can sleep
5225 * safely in order to wait for the netdev refcnt to drop to zero.
5227 * We must not return until all unregister events added during
5228 * the interval the lock was held have been completed.
5230 void netdev_run_todo(void)
5232 struct list_head list
;
5234 /* Snapshot list, allow later requests */
5235 list_replace_init(&net_todo_list
, &list
);
5239 while (!list_empty(&list
)) {
5240 struct net_device
*dev
5241 = list_entry(list
.next
, struct net_device
, todo_list
);
5242 list_del(&dev
->todo_list
);
5244 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5245 printk(KERN_ERR
"network todo '%s' but state %d\n",
5246 dev
->name
, dev
->reg_state
);
5251 dev
->reg_state
= NETREG_UNREGISTERED
;
5253 on_each_cpu(flush_backlog
, dev
, 1);
5255 netdev_wait_allrefs(dev
);
5258 BUG_ON(atomic_read(&dev
->refcnt
));
5259 WARN_ON(dev
->ip_ptr
);
5260 WARN_ON(dev
->ip6_ptr
);
5261 WARN_ON(dev
->dn_ptr
);
5263 if (dev
->destructor
)
5264 dev
->destructor(dev
);
5266 /* Free network device */
5267 kobject_put(&dev
->dev
.kobj
);
5272 * dev_txq_stats_fold - fold tx_queues stats
5273 * @dev: device to get statistics from
5274 * @stats: struct net_device_stats to hold results
5276 void dev_txq_stats_fold(const struct net_device
*dev
,
5277 struct net_device_stats
*stats
)
5279 unsigned long tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5281 struct netdev_queue
*txq
;
5283 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5284 txq
= netdev_get_tx_queue(dev
, i
);
5285 tx_bytes
+= txq
->tx_bytes
;
5286 tx_packets
+= txq
->tx_packets
;
5287 tx_dropped
+= txq
->tx_dropped
;
5289 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5290 stats
->tx_bytes
= tx_bytes
;
5291 stats
->tx_packets
= tx_packets
;
5292 stats
->tx_dropped
= tx_dropped
;
5295 EXPORT_SYMBOL(dev_txq_stats_fold
);
5298 * dev_get_stats - get network device statistics
5299 * @dev: device to get statistics from
5301 * Get network statistics from device. The device driver may provide
5302 * its own method by setting dev->netdev_ops->get_stats; otherwise
5303 * the internal statistics structure is used.
5305 const struct net_device_stats
*dev_get_stats(struct net_device
*dev
)
5307 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5309 if (ops
->ndo_get_stats
)
5310 return ops
->ndo_get_stats(dev
);
5312 dev_txq_stats_fold(dev
, &dev
->stats
);
5315 EXPORT_SYMBOL(dev_get_stats
);
5317 static void netdev_init_one_queue(struct net_device
*dev
,
5318 struct netdev_queue
*queue
,
5324 static void netdev_init_queues(struct net_device
*dev
)
5326 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
5327 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5328 spin_lock_init(&dev
->tx_global_lock
);
5332 * alloc_netdev_mq - allocate network device
5333 * @sizeof_priv: size of private data to allocate space for
5334 * @name: device name format string
5335 * @setup: callback to initialize device
5336 * @queue_count: the number of subqueues to allocate
5338 * Allocates a struct net_device with private data area for driver use
5339 * and performs basic initialization. Also allocates subquue structs
5340 * for each queue on the device at the end of the netdevice.
5342 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
5343 void (*setup
)(struct net_device
*), unsigned int queue_count
)
5345 struct netdev_queue
*tx
;
5346 struct net_device
*dev
;
5348 struct net_device
*p
;
5350 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5352 alloc_size
= sizeof(struct net_device
);
5354 /* ensure 32-byte alignment of private area */
5355 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5356 alloc_size
+= sizeof_priv
;
5358 /* ensure 32-byte alignment of whole construct */
5359 alloc_size
+= NETDEV_ALIGN
- 1;
5361 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5363 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5367 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5369 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5374 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5375 dev
->padded
= (char *)dev
- (char *)p
;
5377 if (dev_addr_init(dev
))
5380 dev_unicast_init(dev
);
5382 dev_net_set(dev
, &init_net
);
5385 dev
->num_tx_queues
= queue_count
;
5386 dev
->real_num_tx_queues
= queue_count
;
5388 dev
->gso_max_size
= GSO_MAX_SIZE
;
5390 netdev_init_queues(dev
);
5392 INIT_LIST_HEAD(&dev
->napi_list
);
5393 INIT_LIST_HEAD(&dev
->unreg_list
);
5394 INIT_LIST_HEAD(&dev
->link_watch_list
);
5395 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5397 strcpy(dev
->name
, name
);
5407 EXPORT_SYMBOL(alloc_netdev_mq
);
5410 * free_netdev - free network device
5413 * This function does the last stage of destroying an allocated device
5414 * interface. The reference to the device object is released.
5415 * If this is the last reference then it will be freed.
5417 void free_netdev(struct net_device
*dev
)
5419 struct napi_struct
*p
, *n
;
5421 release_net(dev_net(dev
));
5425 /* Flush device addresses */
5426 dev_addr_flush(dev
);
5428 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5431 /* Compatibility with error handling in drivers */
5432 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5433 kfree((char *)dev
- dev
->padded
);
5437 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5438 dev
->reg_state
= NETREG_RELEASED
;
5440 /* will free via device release */
5441 put_device(&dev
->dev
);
5443 EXPORT_SYMBOL(free_netdev
);
5446 * synchronize_net - Synchronize with packet receive processing
5448 * Wait for packets currently being received to be done.
5449 * Does not block later packets from starting.
5451 void synchronize_net(void)
5456 EXPORT_SYMBOL(synchronize_net
);
5459 * unregister_netdevice_queue - remove device from the kernel
5463 * This function shuts down a device interface and removes it
5464 * from the kernel tables.
5465 * If head not NULL, device is queued to be unregistered later.
5467 * Callers must hold the rtnl semaphore. You may want
5468 * unregister_netdev() instead of this.
5471 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5476 list_move_tail(&dev
->unreg_list
, head
);
5478 rollback_registered(dev
);
5479 /* Finish processing unregister after unlock */
5483 EXPORT_SYMBOL(unregister_netdevice_queue
);
5486 * unregister_netdevice_many - unregister many devices
5487 * @head: list of devices
5489 void unregister_netdevice_many(struct list_head
*head
)
5491 struct net_device
*dev
;
5493 if (!list_empty(head
)) {
5494 rollback_registered_many(head
);
5495 list_for_each_entry(dev
, head
, unreg_list
)
5499 EXPORT_SYMBOL(unregister_netdevice_many
);
5502 * unregister_netdev - remove device from the kernel
5505 * This function shuts down a device interface and removes it
5506 * from the kernel tables.
5508 * This is just a wrapper for unregister_netdevice that takes
5509 * the rtnl semaphore. In general you want to use this and not
5510 * unregister_netdevice.
5512 void unregister_netdev(struct net_device
*dev
)
5515 unregister_netdevice(dev
);
5518 EXPORT_SYMBOL(unregister_netdev
);
5521 * dev_change_net_namespace - move device to different nethost namespace
5523 * @net: network namespace
5524 * @pat: If not NULL name pattern to try if the current device name
5525 * is already taken in the destination network namespace.
5527 * This function shuts down a device interface and moves it
5528 * to a new network namespace. On success 0 is returned, on
5529 * a failure a netagive errno code is returned.
5531 * Callers must hold the rtnl semaphore.
5534 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5540 /* Don't allow namespace local devices to be moved. */
5542 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5546 /* Don't allow real devices to be moved when sysfs
5550 if (dev
->dev
.parent
)
5554 /* Ensure the device has been registrered */
5556 if (dev
->reg_state
!= NETREG_REGISTERED
)
5559 /* Get out if there is nothing todo */
5561 if (net_eq(dev_net(dev
), net
))
5564 /* Pick the destination device name, and ensure
5565 * we can use it in the destination network namespace.
5568 if (__dev_get_by_name(net
, dev
->name
)) {
5569 /* We get here if we can't use the current device name */
5572 if (dev_get_valid_name(net
, pat
, dev
->name
, 1))
5577 * And now a mini version of register_netdevice unregister_netdevice.
5580 /* If device is running close it first. */
5583 /* And unlink it from device chain */
5585 unlist_netdevice(dev
);
5589 /* Shutdown queueing discipline. */
5592 /* Notify protocols, that we are about to destroy
5593 this device. They should clean all the things.
5595 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5596 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5599 * Flush the unicast and multicast chains
5601 dev_unicast_flush(dev
);
5602 dev_addr_discard(dev
);
5604 netdev_unregister_kobject(dev
);
5606 /* Actually switch the network namespace */
5607 dev_net_set(dev
, net
);
5609 /* If there is an ifindex conflict assign a new one */
5610 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5611 int iflink
= (dev
->iflink
== dev
->ifindex
);
5612 dev
->ifindex
= dev_new_index(net
);
5614 dev
->iflink
= dev
->ifindex
;
5617 /* Fixup kobjects */
5618 err
= netdev_register_kobject(dev
);
5621 /* Add the device back in the hashes */
5622 list_netdevice(dev
);
5624 /* Notify protocols, that a new device appeared. */
5625 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5628 * Prevent userspace races by waiting until the network
5629 * device is fully setup before sending notifications.
5631 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5638 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5640 static int dev_cpu_callback(struct notifier_block
*nfb
,
5641 unsigned long action
,
5644 struct sk_buff
**list_skb
;
5645 struct Qdisc
**list_net
;
5646 struct sk_buff
*skb
;
5647 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5648 struct softnet_data
*sd
, *oldsd
;
5650 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5653 local_irq_disable();
5654 cpu
= smp_processor_id();
5655 sd
= &per_cpu(softnet_data
, cpu
);
5656 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5658 /* Find end of our completion_queue. */
5659 list_skb
= &sd
->completion_queue
;
5661 list_skb
= &(*list_skb
)->next
;
5662 /* Append completion queue from offline CPU. */
5663 *list_skb
= oldsd
->completion_queue
;
5664 oldsd
->completion_queue
= NULL
;
5666 /* Find end of our output_queue. */
5667 list_net
= &sd
->output_queue
;
5669 list_net
= &(*list_net
)->next_sched
;
5670 /* Append output queue from offline CPU. */
5671 *list_net
= oldsd
->output_queue
;
5672 oldsd
->output_queue
= NULL
;
5674 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5677 /* Process offline CPU's input_pkt_queue */
5678 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
5686 * netdev_increment_features - increment feature set by one
5687 * @all: current feature set
5688 * @one: new feature set
5689 * @mask: mask feature set
5691 * Computes a new feature set after adding a device with feature set
5692 * @one to the master device with current feature set @all. Will not
5693 * enable anything that is off in @mask. Returns the new feature set.
5695 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
5698 /* If device needs checksumming, downgrade to it. */
5699 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
5700 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
5701 else if (mask
& NETIF_F_ALL_CSUM
) {
5702 /* If one device supports v4/v6 checksumming, set for all. */
5703 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
5704 !(all
& NETIF_F_GEN_CSUM
)) {
5705 all
&= ~NETIF_F_ALL_CSUM
;
5706 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
5709 /* If one device supports hw checksumming, set for all. */
5710 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
5711 all
&= ~NETIF_F_ALL_CSUM
;
5712 all
|= NETIF_F_HW_CSUM
;
5716 one
|= NETIF_F_ALL_CSUM
;
5718 one
|= all
& NETIF_F_ONE_FOR_ALL
;
5719 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
5720 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
5724 EXPORT_SYMBOL(netdev_increment_features
);
5726 static struct hlist_head
*netdev_create_hash(void)
5729 struct hlist_head
*hash
;
5731 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5733 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5734 INIT_HLIST_HEAD(&hash
[i
]);
5739 /* Initialize per network namespace state */
5740 static int __net_init
netdev_init(struct net
*net
)
5742 INIT_LIST_HEAD(&net
->dev_base_head
);
5744 net
->dev_name_head
= netdev_create_hash();
5745 if (net
->dev_name_head
== NULL
)
5748 net
->dev_index_head
= netdev_create_hash();
5749 if (net
->dev_index_head
== NULL
)
5755 kfree(net
->dev_name_head
);
5761 * netdev_drivername - network driver for the device
5762 * @dev: network device
5763 * @buffer: buffer for resulting name
5764 * @len: size of buffer
5766 * Determine network driver for device.
5768 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
5770 const struct device_driver
*driver
;
5771 const struct device
*parent
;
5773 if (len
<= 0 || !buffer
)
5777 parent
= dev
->dev
.parent
;
5782 driver
= parent
->driver
;
5783 if (driver
&& driver
->name
)
5784 strlcpy(buffer
, driver
->name
, len
);
5788 static void __net_exit
netdev_exit(struct net
*net
)
5790 kfree(net
->dev_name_head
);
5791 kfree(net
->dev_index_head
);
5794 static struct pernet_operations __net_initdata netdev_net_ops
= {
5795 .init
= netdev_init
,
5796 .exit
= netdev_exit
,
5799 static void __net_exit
default_device_exit(struct net
*net
)
5801 struct net_device
*dev
, *aux
;
5803 * Push all migratable network devices back to the
5804 * initial network namespace
5807 for_each_netdev_safe(net
, dev
, aux
) {
5809 char fb_name
[IFNAMSIZ
];
5811 /* Ignore unmoveable devices (i.e. loopback) */
5812 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5815 /* Leave virtual devices for the generic cleanup */
5816 if (dev
->rtnl_link_ops
)
5819 /* Push remaing network devices to init_net */
5820 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
5821 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
5823 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
5824 __func__
, dev
->name
, err
);
5831 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
5833 /* At exit all network devices most be removed from a network
5834 * namespace. Do this in the reverse order of registeration.
5835 * Do this across as many network namespaces as possible to
5836 * improve batching efficiency.
5838 struct net_device
*dev
;
5840 LIST_HEAD(dev_kill_list
);
5843 list_for_each_entry(net
, net_list
, exit_list
) {
5844 for_each_netdev_reverse(net
, dev
) {
5845 if (dev
->rtnl_link_ops
)
5846 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
5848 unregister_netdevice_queue(dev
, &dev_kill_list
);
5851 unregister_netdevice_many(&dev_kill_list
);
5855 static struct pernet_operations __net_initdata default_device_ops
= {
5856 .exit
= default_device_exit
,
5857 .exit_batch
= default_device_exit_batch
,
5861 * Initialize the DEV module. At boot time this walks the device list and
5862 * unhooks any devices that fail to initialise (normally hardware not
5863 * present) and leaves us with a valid list of present and active devices.
5868 * This is called single threaded during boot, so no need
5869 * to take the rtnl semaphore.
5871 static int __init
net_dev_init(void)
5873 int i
, rc
= -ENOMEM
;
5875 BUG_ON(!dev_boot_phase
);
5877 if (dev_proc_init())
5880 if (netdev_kobject_init())
5883 INIT_LIST_HEAD(&ptype_all
);
5884 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
5885 INIT_LIST_HEAD(&ptype_base
[i
]);
5887 if (register_pernet_subsys(&netdev_net_ops
))
5891 * Initialise the packet receive queues.
5894 for_each_possible_cpu(i
) {
5895 struct softnet_data
*queue
;
5897 queue
= &per_cpu(softnet_data
, i
);
5898 skb_queue_head_init(&queue
->input_pkt_queue
);
5899 queue
->completion_queue
= NULL
;
5900 INIT_LIST_HEAD(&queue
->poll_list
);
5902 queue
->backlog
.poll
= process_backlog
;
5903 queue
->backlog
.weight
= weight_p
;
5904 queue
->backlog
.gro_list
= NULL
;
5905 queue
->backlog
.gro_count
= 0;
5910 /* The loopback device is special if any other network devices
5911 * is present in a network namespace the loopback device must
5912 * be present. Since we now dynamically allocate and free the
5913 * loopback device ensure this invariant is maintained by
5914 * keeping the loopback device as the first device on the
5915 * list of network devices. Ensuring the loopback devices
5916 * is the first device that appears and the last network device
5919 if (register_pernet_device(&loopback_net_ops
))
5922 if (register_pernet_device(&default_device_ops
))
5925 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
5926 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
5928 hotcpu_notifier(dev_cpu_callback
, 0);
5936 subsys_initcall(net_dev_init
);
5938 static int __init
initialize_hashrnd(void)
5940 get_random_bytes(&skb_tx_hashrnd
, sizeof(skb_tx_hashrnd
));
5944 late_initcall_sync(initialize_hashrnd
);