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/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <linux/if_bridge.h>
105 #include <linux/if_macvlan.h>
107 #include <net/pkt_sched.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/kmod.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/wext.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <linux/pci.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 * The list of packet types we will receive (as opposed to discard)
145 * and the routines to invoke.
147 * Why 16. Because with 16 the only overlap we get on a hash of the
148 * low nibble of the protocol value is RARP/SNAP/X.25.
150 * NOTE: That is no longer true with the addition of VLAN tags. Not
151 * sure which should go first, but I bet it won't make much
152 * difference if we are running VLANs. The good news is that
153 * this protocol won't be in the list unless compiled in, so
154 * the average user (w/out VLANs) will not be adversely affected.
171 #define PTYPE_HASH_SIZE (16)
172 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
174 static DEFINE_SPINLOCK(ptype_lock
);
175 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
176 static struct list_head ptype_all __read_mostly
; /* Taps */
179 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
182 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
184 * Writers must hold the rtnl semaphore while they loop through the
185 * dev_base_head list, and hold dev_base_lock for writing when they do the
186 * actual updates. This allows pure readers to access the list even
187 * while a writer is preparing to update it.
189 * To put it another way, dev_base_lock is held for writing only to
190 * protect against pure readers; the rtnl semaphore provides the
191 * protection against other writers.
193 * See, for example usages, register_netdevice() and
194 * unregister_netdevice(), which must be called with the rtnl
197 DEFINE_RWLOCK(dev_base_lock
);
198 EXPORT_SYMBOL(dev_base_lock
);
200 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
202 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
203 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
206 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
208 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
211 static inline void rps_lock(struct softnet_data
*sd
)
214 spin_lock(&sd
->input_pkt_queue
.lock
);
218 static inline void rps_unlock(struct softnet_data
*sd
)
221 spin_unlock(&sd
->input_pkt_queue
.lock
);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device
*dev
)
228 struct net
*net
= dev_net(dev
);
232 write_lock_bh(&dev_base_lock
);
233 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
234 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
235 hlist_add_head_rcu(&dev
->index_hlist
,
236 dev_index_hash(net
, dev
->ifindex
));
237 write_unlock_bh(&dev_base_lock
);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device
*dev
)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock
);
250 list_del_rcu(&dev
->dev_list
);
251 hlist_del_rcu(&dev
->name_hlist
);
252 hlist_del_rcu(&dev
->index_hlist
);
253 write_unlock_bh(&dev_base_lock
);
260 static RAW_NOTIFIER_HEAD(netdev_chain
);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
268 EXPORT_PER_CPU_SYMBOL(softnet_data
);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type
[] =
276 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
277 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
278 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
279 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
280 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
281 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
282 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
283 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
284 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
285 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
286 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
287 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
288 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
289 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
290 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
291 ARPHRD_VOID
, ARPHRD_NONE
};
293 static const char *const netdev_lock_name
[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
312 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
318 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
319 if (netdev_lock_type
[i
] == dev_type
)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type
) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
326 unsigned short dev_type
)
330 i
= netdev_lock_pos(dev_type
);
331 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
332 netdev_lock_name
[i
]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
339 i
= netdev_lock_pos(dev
->type
);
340 lockdep_set_class_and_name(&dev
->addr_list_lock
,
341 &netdev_addr_lock_key
[i
],
342 netdev_lock_name
[i
]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
346 unsigned short dev_type
)
349 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type
*pt
)
393 spin_lock_bh(&ptype_lock
);
394 if (pt
->type
== htons(ETH_P_ALL
))
395 list_add_rcu(&pt
->list
, &ptype_all
);
397 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
398 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
400 spin_unlock_bh(&ptype_lock
);
402 EXPORT_SYMBOL(dev_add_pack
);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type
*pt
)
419 struct list_head
*head
;
420 struct packet_type
*pt1
;
422 spin_lock_bh(&ptype_lock
);
424 if (pt
->type
== htons(ETH_P_ALL
))
427 head
= &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
429 list_for_each_entry(pt1
, head
, list
) {
431 list_del_rcu(&pt
->list
);
436 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
438 spin_unlock_bh(&ptype_lock
);
440 EXPORT_SYMBOL(__dev_remove_pack
);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type
*pt
)
456 __dev_remove_pack(pt
);
460 EXPORT_SYMBOL(dev_remove_pack
);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
482 struct netdev_boot_setup
*s
;
486 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
487 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
488 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
489 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
490 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
495 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device
*dev
)
509 struct netdev_boot_setup
*s
= dev_boot_setup
;
512 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
513 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
514 !strcmp(dev
->name
, s
[i
].name
)) {
515 dev
->irq
= s
[i
].map
.irq
;
516 dev
->base_addr
= s
[i
].map
.base_addr
;
517 dev
->mem_start
= s
[i
].map
.mem_start
;
518 dev
->mem_end
= s
[i
].map
.mem_end
;
524 EXPORT_SYMBOL(netdev_boot_setup_check
);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix
, int unit
)
539 const struct netdev_boot_setup
*s
= dev_boot_setup
;
543 sprintf(name
, "%s%d", prefix
, unit
);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net
, name
))
552 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
553 if (!strcmp(name
, s
[i
].name
))
554 return s
[i
].map
.base_addr
;
559 * Saves at boot time configured settings for any netdevice.
561 int __init
netdev_boot_setup(char *str
)
566 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
571 memset(&map
, 0, sizeof(map
));
575 map
.base_addr
= ints
[2];
577 map
.mem_start
= ints
[3];
579 map
.mem_end
= ints
[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str
, &map
);
585 __setup("netdev=", netdev_boot_setup
);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
607 struct hlist_node
*p
;
608 struct net_device
*dev
;
609 struct hlist_head
*head
= dev_name_hash(net
, name
);
611 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
612 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
617 EXPORT_SYMBOL(__dev_get_by_name
);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
633 struct hlist_node
*p
;
634 struct net_device
*dev
;
635 struct hlist_head
*head
= dev_name_hash(net
, name
);
637 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
638 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
643 EXPORT_SYMBOL(dev_get_by_name_rcu
);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
659 struct net_device
*dev
;
662 dev
= dev_get_by_name_rcu(net
, name
);
668 EXPORT_SYMBOL(dev_get_by_name
);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
684 struct hlist_node
*p
;
685 struct net_device
*dev
;
686 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
688 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
689 if (dev
->ifindex
== ifindex
)
694 EXPORT_SYMBOL(__dev_get_by_index
);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
709 struct hlist_node
*p
;
710 struct net_device
*dev
;
711 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
713 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
714 if (dev
->ifindex
== ifindex
)
719 EXPORT_SYMBOL(dev_get_by_index_rcu
);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
735 struct net_device
*dev
;
738 dev
= dev_get_by_index_rcu(net
, ifindex
);
744 EXPORT_SYMBOL(dev_get_by_index
);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
763 struct net_device
*dev
;
767 for_each_netdev(net
, dev
)
768 if (dev
->type
== type
&&
769 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
774 EXPORT_SYMBOL(dev_getbyhwaddr
);
776 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
778 struct net_device
*dev
;
781 for_each_netdev(net
, dev
)
782 if (dev
->type
== type
)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
789 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
791 struct net_device
*dev
, *ret
= NULL
;
794 for_each_netdev_rcu(net
, dev
)
795 if (dev
->type
== type
) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
806 * dev_get_by_flags - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. The device returned has
813 * had a reference added and the pointer is safe until the user calls
814 * dev_put to indicate they have finished with it.
817 struct net_device
*dev_get_by_flags(struct net
*net
, unsigned short if_flags
,
820 struct net_device
*dev
, *ret
;
824 for_each_netdev_rcu(net
, dev
) {
825 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
834 EXPORT_SYMBOL(dev_get_by_flags
);
837 * dev_valid_name - check if name is okay for network device
840 * Network device names need to be valid file names to
841 * to allow sysfs to work. We also disallow any kind of
844 int dev_valid_name(const char *name
)
848 if (strlen(name
) >= IFNAMSIZ
)
850 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
854 if (*name
== '/' || isspace(*name
))
860 EXPORT_SYMBOL(dev_valid_name
);
863 * __dev_alloc_name - allocate a name for a device
864 * @net: network namespace to allocate the device name in
865 * @name: name format string
866 * @buf: scratch buffer and result name string
868 * Passed a format string - eg "lt%d" it will try and find a suitable
869 * id. It scans list of devices to build up a free map, then chooses
870 * the first empty slot. The caller must hold the dev_base or rtnl lock
871 * while allocating the name and adding the device in order to avoid
873 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
874 * Returns the number of the unit assigned or a negative errno code.
877 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
881 const int max_netdevices
= 8*PAGE_SIZE
;
882 unsigned long *inuse
;
883 struct net_device
*d
;
885 p
= strnchr(name
, IFNAMSIZ
-1, '%');
888 * Verify the string as this thing may have come from
889 * the user. There must be either one "%d" and no other "%"
892 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
895 /* Use one page as a bit array of possible slots */
896 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
900 for_each_netdev(net
, d
) {
901 if (!sscanf(d
->name
, name
, &i
))
903 if (i
< 0 || i
>= max_netdevices
)
906 /* avoid cases where sscanf is not exact inverse of printf */
907 snprintf(buf
, IFNAMSIZ
, name
, i
);
908 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
912 i
= find_first_zero_bit(inuse
, max_netdevices
);
913 free_page((unsigned long) inuse
);
917 snprintf(buf
, IFNAMSIZ
, name
, i
);
918 if (!__dev_get_by_name(net
, buf
))
921 /* It is possible to run out of possible slots
922 * when the name is long and there isn't enough space left
923 * for the digits, or if all bits are used.
929 * dev_alloc_name - allocate a name for a device
931 * @name: name format string
933 * Passed a format string - eg "lt%d" it will try and find a suitable
934 * id. It scans list of devices to build up a free map, then chooses
935 * the first empty slot. The caller must hold the dev_base or rtnl lock
936 * while allocating the name and adding the device in order to avoid
938 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
939 * Returns the number of the unit assigned or a negative errno code.
942 int dev_alloc_name(struct net_device
*dev
, const char *name
)
948 BUG_ON(!dev_net(dev
));
950 ret
= __dev_alloc_name(net
, name
, buf
);
952 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
955 EXPORT_SYMBOL(dev_alloc_name
);
957 static int dev_get_valid_name(struct net
*net
, const char *name
, char *buf
,
960 if (!dev_valid_name(name
))
963 if (fmt
&& strchr(name
, '%'))
964 return __dev_alloc_name(net
, name
, buf
);
965 else if (__dev_get_by_name(net
, name
))
967 else if (buf
!= name
)
968 strlcpy(buf
, name
, IFNAMSIZ
);
974 * dev_change_name - change name of a device
976 * @newname: name (or format string) must be at least IFNAMSIZ
978 * Change name of a device, can pass format strings "eth%d".
981 int dev_change_name(struct net_device
*dev
, const char *newname
)
983 char oldname
[IFNAMSIZ
];
989 BUG_ON(!dev_net(dev
));
992 if (dev
->flags
& IFF_UP
)
995 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
998 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1000 err
= dev_get_valid_name(net
, newname
, dev
->name
, 1);
1005 /* For now only devices in the initial network namespace
1008 if (net_eq(net
, &init_net
)) {
1009 ret
= device_rename(&dev
->dev
, dev
->name
);
1011 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1016 write_lock_bh(&dev_base_lock
);
1017 hlist_del(&dev
->name_hlist
);
1018 write_unlock_bh(&dev_base_lock
);
1022 write_lock_bh(&dev_base_lock
);
1023 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1024 write_unlock_bh(&dev_base_lock
);
1026 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1027 ret
= notifier_to_errno(ret
);
1030 /* err >= 0 after dev_alloc_name() or stores the first errno */
1033 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1037 "%s: name change rollback failed: %d.\n",
1046 * dev_set_alias - change ifalias of a device
1048 * @alias: name up to IFALIASZ
1049 * @len: limit of bytes to copy from info
1051 * Set ifalias for a device,
1053 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1057 if (len
>= IFALIASZ
)
1062 kfree(dev
->ifalias
);
1063 dev
->ifalias
= NULL
;
1068 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1072 strlcpy(dev
->ifalias
, alias
, len
+1);
1078 * netdev_features_change - device changes features
1079 * @dev: device to cause notification
1081 * Called to indicate a device has changed features.
1083 void netdev_features_change(struct net_device
*dev
)
1085 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1087 EXPORT_SYMBOL(netdev_features_change
);
1090 * netdev_state_change - device changes state
1091 * @dev: device to cause notification
1093 * Called to indicate a device has changed state. This function calls
1094 * the notifier chains for netdev_chain and sends a NEWLINK message
1095 * to the routing socket.
1097 void netdev_state_change(struct net_device
*dev
)
1099 if (dev
->flags
& IFF_UP
) {
1100 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1101 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1104 EXPORT_SYMBOL(netdev_state_change
);
1106 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1108 return call_netdevice_notifiers(event
, dev
);
1110 EXPORT_SYMBOL(netdev_bonding_change
);
1113 * dev_load - load a network module
1114 * @net: the applicable net namespace
1115 * @name: name of interface
1117 * If a network interface is not present and the process has suitable
1118 * privileges this function loads the module. If module loading is not
1119 * available in this kernel then it becomes a nop.
1122 void dev_load(struct net
*net
, const char *name
)
1124 struct net_device
*dev
;
1127 dev
= dev_get_by_name_rcu(net
, name
);
1130 if (!dev
&& capable(CAP_NET_ADMIN
))
1131 request_module("%s", name
);
1133 EXPORT_SYMBOL(dev_load
);
1135 static int __dev_open(struct net_device
*dev
)
1137 const struct net_device_ops
*ops
= dev
->netdev_ops
;
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
1196 * dev_open - prepare an interface for use.
1197 * @dev: device to open
1199 * Takes a device from down to up state. The device's private open
1200 * function is invoked and then the multicast lists are loaded. Finally
1201 * the device is moved into the up state and a %NETDEV_UP message is
1202 * sent to the netdev notifier chain.
1204 * Calling this function on an active interface is a nop. On a failure
1205 * a negative errno code is returned.
1207 int dev_open(struct net_device
*dev
)
1214 if (dev
->flags
& IFF_UP
)
1220 ret
= __dev_open(dev
);
1225 * ... and announce new interface.
1227 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1228 call_netdevice_notifiers(NETDEV_UP
, dev
);
1232 EXPORT_SYMBOL(dev_open
);
1234 static int __dev_close(struct net_device
*dev
)
1236 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1242 * Tell people we are going down, so that they can
1243 * prepare to death, when device is still operating.
1245 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1247 clear_bit(__LINK_STATE_START
, &dev
->state
);
1249 /* Synchronize to scheduled poll. We cannot touch poll list,
1250 * it can be even on different cpu. So just clear netif_running().
1252 * dev->stop() will invoke napi_disable() on all of it's
1253 * napi_struct instances on this device.
1255 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1257 dev_deactivate(dev
);
1260 * Call the device specific close. This cannot fail.
1261 * Only if device is UP
1263 * We allow it to be called even after a DETACH hot-plug
1270 * Device is now down.
1273 dev
->flags
&= ~IFF_UP
;
1278 net_dmaengine_put();
1284 * dev_close - shutdown an interface.
1285 * @dev: device to shutdown
1287 * This function moves an active device into down state. A
1288 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1289 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1292 int dev_close(struct net_device
*dev
)
1294 if (!(dev
->flags
& IFF_UP
))
1300 * Tell people we are down
1302 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1303 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1307 EXPORT_SYMBOL(dev_close
);
1311 * dev_disable_lro - disable Large Receive Offload on a device
1314 * Disable Large Receive Offload (LRO) on a net device. Must be
1315 * called under RTNL. This is needed if received packets may be
1316 * forwarded to another interface.
1318 void dev_disable_lro(struct net_device
*dev
)
1320 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1321 dev
->ethtool_ops
->set_flags
) {
1322 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1323 if (flags
& ETH_FLAG_LRO
) {
1324 flags
&= ~ETH_FLAG_LRO
;
1325 dev
->ethtool_ops
->set_flags(dev
, flags
);
1328 WARN_ON(dev
->features
& NETIF_F_LRO
);
1330 EXPORT_SYMBOL(dev_disable_lro
);
1333 static int dev_boot_phase
= 1;
1336 * Device change register/unregister. These are not inline or static
1337 * as we export them to the world.
1341 * register_netdevice_notifier - register a network notifier block
1344 * Register a notifier to be called when network device events occur.
1345 * The notifier passed is linked into the kernel structures and must
1346 * not be reused until it has been unregistered. A negative errno code
1347 * is returned on a failure.
1349 * When registered all registration and up events are replayed
1350 * to the new notifier to allow device to have a race free
1351 * view of the network device list.
1354 int register_netdevice_notifier(struct notifier_block
*nb
)
1356 struct net_device
*dev
;
1357 struct net_device
*last
;
1362 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1368 for_each_netdev(net
, dev
) {
1369 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1370 err
= notifier_to_errno(err
);
1374 if (!(dev
->flags
& IFF_UP
))
1377 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1388 for_each_netdev(net
, dev
) {
1392 if (dev
->flags
& IFF_UP
) {
1393 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1394 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1396 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1397 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1401 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1404 EXPORT_SYMBOL(register_netdevice_notifier
);
1407 * unregister_netdevice_notifier - unregister a network notifier block
1410 * Unregister a notifier previously registered by
1411 * register_netdevice_notifier(). The notifier is unlinked into the
1412 * kernel structures and may then be reused. A negative errno code
1413 * is returned on a failure.
1416 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1421 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1425 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1428 * call_netdevice_notifiers - call all network notifier blocks
1429 * @val: value passed unmodified to notifier function
1430 * @dev: net_device pointer passed unmodified to notifier function
1432 * Call all network notifier blocks. Parameters and return value
1433 * are as for raw_notifier_call_chain().
1436 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1439 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1442 /* When > 0 there are consumers of rx skb time stamps */
1443 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1445 void net_enable_timestamp(void)
1447 atomic_inc(&netstamp_needed
);
1449 EXPORT_SYMBOL(net_enable_timestamp
);
1451 void net_disable_timestamp(void)
1453 atomic_dec(&netstamp_needed
);
1455 EXPORT_SYMBOL(net_disable_timestamp
);
1457 static inline void net_timestamp(struct sk_buff
*skb
)
1459 if (atomic_read(&netstamp_needed
))
1460 __net_timestamp(skb
);
1462 skb
->tstamp
.tv64
= 0;
1466 * dev_forward_skb - loopback an skb to another netif
1468 * @dev: destination network device
1469 * @skb: buffer to forward
1472 * NET_RX_SUCCESS (no congestion)
1473 * NET_RX_DROP (packet was dropped)
1475 * dev_forward_skb can be used for injecting an skb from the
1476 * start_xmit function of one device into the receive queue
1477 * of another device.
1479 * The receiving device may be in another namespace, so
1480 * we have to clear all information in the skb that could
1481 * impact namespace isolation.
1483 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1487 if (!(dev
->flags
& IFF_UP
))
1490 if (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
))
1493 skb_set_dev(skb
, dev
);
1494 skb
->tstamp
.tv64
= 0;
1495 skb
->pkt_type
= PACKET_HOST
;
1496 skb
->protocol
= eth_type_trans(skb
, dev
);
1497 return netif_rx(skb
);
1499 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1502 * Support routine. Sends outgoing frames to any network
1503 * taps currently in use.
1506 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1508 struct packet_type
*ptype
;
1510 #ifdef CONFIG_NET_CLS_ACT
1511 if (!(skb
->tstamp
.tv64
&& (G_TC_FROM(skb
->tc_verd
) & AT_INGRESS
)))
1518 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1519 /* Never send packets back to the socket
1520 * they originated from - MvS (miquels@drinkel.ow.org)
1522 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1523 (ptype
->af_packet_priv
== NULL
||
1524 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1525 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1529 /* skb->nh should be correctly
1530 set by sender, so that the second statement is
1531 just protection against buggy protocols.
1533 skb_reset_mac_header(skb2
);
1535 if (skb_network_header(skb2
) < skb2
->data
||
1536 skb2
->network_header
> skb2
->tail
) {
1537 if (net_ratelimit())
1538 printk(KERN_CRIT
"protocol %04x is "
1540 skb2
->protocol
, dev
->name
);
1541 skb_reset_network_header(skb2
);
1544 skb2
->transport_header
= skb2
->network_header
;
1545 skb2
->pkt_type
= PACKET_OUTGOING
;
1546 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1553 static inline void __netif_reschedule(struct Qdisc
*q
)
1555 struct softnet_data
*sd
;
1556 unsigned long flags
;
1558 local_irq_save(flags
);
1559 sd
= &__get_cpu_var(softnet_data
);
1560 q
->next_sched
= sd
->output_queue
;
1561 sd
->output_queue
= q
;
1562 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1563 local_irq_restore(flags
);
1566 void __netif_schedule(struct Qdisc
*q
)
1568 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1569 __netif_reschedule(q
);
1571 EXPORT_SYMBOL(__netif_schedule
);
1573 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1575 if (atomic_dec_and_test(&skb
->users
)) {
1576 struct softnet_data
*sd
;
1577 unsigned long flags
;
1579 local_irq_save(flags
);
1580 sd
= &__get_cpu_var(softnet_data
);
1581 skb
->next
= sd
->completion_queue
;
1582 sd
->completion_queue
= skb
;
1583 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1584 local_irq_restore(flags
);
1587 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1589 void dev_kfree_skb_any(struct sk_buff
*skb
)
1591 if (in_irq() || irqs_disabled())
1592 dev_kfree_skb_irq(skb
);
1596 EXPORT_SYMBOL(dev_kfree_skb_any
);
1600 * netif_device_detach - mark device as removed
1601 * @dev: network device
1603 * Mark device as removed from system and therefore no longer available.
1605 void netif_device_detach(struct net_device
*dev
)
1607 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1608 netif_running(dev
)) {
1609 netif_tx_stop_all_queues(dev
);
1612 EXPORT_SYMBOL(netif_device_detach
);
1615 * netif_device_attach - mark device as attached
1616 * @dev: network device
1618 * Mark device as attached from system and restart if needed.
1620 void netif_device_attach(struct net_device
*dev
)
1622 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1623 netif_running(dev
)) {
1624 netif_tx_wake_all_queues(dev
);
1625 __netdev_watchdog_up(dev
);
1628 EXPORT_SYMBOL(netif_device_attach
);
1630 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1632 return ((features
& NETIF_F_GEN_CSUM
) ||
1633 ((features
& NETIF_F_IP_CSUM
) &&
1634 protocol
== htons(ETH_P_IP
)) ||
1635 ((features
& NETIF_F_IPV6_CSUM
) &&
1636 protocol
== htons(ETH_P_IPV6
)) ||
1637 ((features
& NETIF_F_FCOE_CRC
) &&
1638 protocol
== htons(ETH_P_FCOE
)));
1641 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1643 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1646 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1647 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1648 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1649 veh
->h_vlan_encapsulated_proto
))
1657 * skb_dev_set -- assign a new device to a buffer
1658 * @skb: buffer for the new device
1659 * @dev: network device
1661 * If an skb is owned by a device already, we have to reset
1662 * all data private to the namespace a device belongs to
1663 * before assigning it a new device.
1665 #ifdef CONFIG_NET_NS
1666 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1669 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1672 skb_init_secmark(skb
);
1676 skb
->ipvs_property
= 0;
1677 #ifdef CONFIG_NET_SCHED
1683 EXPORT_SYMBOL(skb_set_dev
);
1684 #endif /* CONFIG_NET_NS */
1687 * Invalidate hardware checksum when packet is to be mangled, and
1688 * complete checksum manually on outgoing path.
1690 int skb_checksum_help(struct sk_buff
*skb
)
1693 int ret
= 0, offset
;
1695 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1696 goto out_set_summed
;
1698 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1699 /* Let GSO fix up the checksum. */
1700 goto out_set_summed
;
1703 offset
= skb
->csum_start
- skb_headroom(skb
);
1704 BUG_ON(offset
>= skb_headlen(skb
));
1705 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1707 offset
+= skb
->csum_offset
;
1708 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1710 if (skb_cloned(skb
) &&
1711 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1712 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1717 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1719 skb
->ip_summed
= CHECKSUM_NONE
;
1723 EXPORT_SYMBOL(skb_checksum_help
);
1726 * skb_gso_segment - Perform segmentation on skb.
1727 * @skb: buffer to segment
1728 * @features: features for the output path (see dev->features)
1730 * This function segments the given skb and returns a list of segments.
1732 * It may return NULL if the skb requires no segmentation. This is
1733 * only possible when GSO is used for verifying header integrity.
1735 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1737 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1738 struct packet_type
*ptype
;
1739 __be16 type
= skb
->protocol
;
1742 skb_reset_mac_header(skb
);
1743 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1744 __skb_pull(skb
, skb
->mac_len
);
1746 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1747 struct net_device
*dev
= skb
->dev
;
1748 struct ethtool_drvinfo info
= {};
1750 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1751 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1753 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1755 info
.driver
, dev
? dev
->features
: 0L,
1756 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1757 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1759 if (skb_header_cloned(skb
) &&
1760 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1761 return ERR_PTR(err
);
1765 list_for_each_entry_rcu(ptype
,
1766 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1767 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1768 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1769 err
= ptype
->gso_send_check(skb
);
1770 segs
= ERR_PTR(err
);
1771 if (err
|| skb_gso_ok(skb
, features
))
1773 __skb_push(skb
, (skb
->data
-
1774 skb_network_header(skb
)));
1776 segs
= ptype
->gso_segment(skb
, features
);
1782 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1786 EXPORT_SYMBOL(skb_gso_segment
);
1788 /* Take action when hardware reception checksum errors are detected. */
1790 void netdev_rx_csum_fault(struct net_device
*dev
)
1792 if (net_ratelimit()) {
1793 printk(KERN_ERR
"%s: hw csum failure.\n",
1794 dev
? dev
->name
: "<unknown>");
1798 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1801 /* Actually, we should eliminate this check as soon as we know, that:
1802 * 1. IOMMU is present and allows to map all the memory.
1803 * 2. No high memory really exists on this machine.
1806 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1808 #ifdef CONFIG_HIGHMEM
1810 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1811 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1812 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1816 if (PCI_DMA_BUS_IS_PHYS
) {
1817 struct device
*pdev
= dev
->dev
.parent
;
1821 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1822 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1823 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1832 void (*destructor
)(struct sk_buff
*skb
);
1835 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1837 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1839 struct dev_gso_cb
*cb
;
1842 struct sk_buff
*nskb
= skb
->next
;
1844 skb
->next
= nskb
->next
;
1847 } while (skb
->next
);
1849 cb
= DEV_GSO_CB(skb
);
1851 cb
->destructor(skb
);
1855 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1856 * @skb: buffer to segment
1858 * This function segments the given skb and stores the list of segments
1861 static int dev_gso_segment(struct sk_buff
*skb
)
1863 struct net_device
*dev
= skb
->dev
;
1864 struct sk_buff
*segs
;
1865 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1868 segs
= skb_gso_segment(skb
, features
);
1870 /* Verifying header integrity only. */
1875 return PTR_ERR(segs
);
1878 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1879 skb
->destructor
= dev_gso_skb_destructor
;
1885 * Try to orphan skb early, right before transmission by the device.
1886 * We cannot orphan skb if tx timestamp is requested, since
1887 * drivers need to call skb_tstamp_tx() to send the timestamp.
1889 static inline void skb_orphan_try(struct sk_buff
*skb
)
1891 if (!skb_tx(skb
)->flags
)
1895 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1896 struct netdev_queue
*txq
)
1898 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1899 int rc
= NETDEV_TX_OK
;
1901 if (likely(!skb
->next
)) {
1902 if (!list_empty(&ptype_all
))
1903 dev_queue_xmit_nit(skb
, dev
);
1905 if (netif_needs_gso(dev
, skb
)) {
1906 if (unlikely(dev_gso_segment(skb
)))
1913 * If device doesnt need skb->dst, release it right now while
1914 * its hot in this cpu cache
1916 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1919 skb_orphan_try(skb
);
1920 rc
= ops
->ndo_start_xmit(skb
, dev
);
1921 if (rc
== NETDEV_TX_OK
)
1922 txq_trans_update(txq
);
1928 struct sk_buff
*nskb
= skb
->next
;
1930 skb
->next
= nskb
->next
;
1934 * If device doesnt need nskb->dst, release it right now while
1935 * its hot in this cpu cache
1937 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1940 rc
= ops
->ndo_start_xmit(nskb
, dev
);
1941 if (unlikely(rc
!= NETDEV_TX_OK
)) {
1942 if (rc
& ~NETDEV_TX_MASK
)
1943 goto out_kfree_gso_skb
;
1944 nskb
->next
= skb
->next
;
1948 txq_trans_update(txq
);
1949 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
1950 return NETDEV_TX_BUSY
;
1951 } while (skb
->next
);
1954 if (likely(skb
->next
== NULL
))
1955 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1961 static u32 hashrnd __read_mostly
;
1963 u16
skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
)
1967 if (skb_rx_queue_recorded(skb
)) {
1968 hash
= skb_get_rx_queue(skb
);
1969 while (unlikely(hash
>= dev
->real_num_tx_queues
))
1970 hash
-= dev
->real_num_tx_queues
;
1974 if (skb
->sk
&& skb
->sk
->sk_hash
)
1975 hash
= skb
->sk
->sk_hash
;
1977 hash
= (__force u16
) skb
->protocol
;
1979 hash
= jhash_1word(hash
, hashrnd
);
1981 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
1983 EXPORT_SYMBOL(skb_tx_hash
);
1985 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
1987 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
1988 if (net_ratelimit()) {
1989 pr_warning("%s selects TX queue %d, but "
1990 "real number of TX queues is %d\n",
1991 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
1998 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
1999 struct sk_buff
*skb
)
2002 struct sock
*sk
= skb
->sk
;
2004 if (sk_tx_queue_recorded(sk
)) {
2005 queue_index
= sk_tx_queue_get(sk
);
2007 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2009 if (ops
->ndo_select_queue
) {
2010 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2011 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2014 if (dev
->real_num_tx_queues
> 1)
2015 queue_index
= skb_tx_hash(dev
, skb
);
2018 struct dst_entry
*dst
= rcu_dereference_check(sk
->sk_dst_cache
, 1);
2020 if (dst
&& skb_dst(skb
) == dst
)
2021 sk_tx_queue_set(sk
, queue_index
);
2026 skb_set_queue_mapping(skb
, queue_index
);
2027 return netdev_get_tx_queue(dev
, queue_index
);
2030 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2031 struct net_device
*dev
,
2032 struct netdev_queue
*txq
)
2034 spinlock_t
*root_lock
= qdisc_lock(q
);
2037 spin_lock(root_lock
);
2038 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2041 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2042 !test_and_set_bit(__QDISC_STATE_RUNNING
, &q
->state
)) {
2044 * This is a work-conserving queue; there are no old skbs
2045 * waiting to be sent out; and the qdisc is not running -
2046 * xmit the skb directly.
2048 __qdisc_update_bstats(q
, skb
->len
);
2049 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
))
2052 clear_bit(__QDISC_STATE_RUNNING
, &q
->state
);
2054 rc
= NET_XMIT_SUCCESS
;
2056 rc
= qdisc_enqueue_root(skb
, q
);
2059 spin_unlock(root_lock
);
2065 * Returns true if either:
2066 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2067 * 2. skb is fragmented and the device does not support SG, or if
2068 * at least one of fragments is in highmem and device does not
2069 * support DMA from it.
2071 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2072 struct net_device
*dev
)
2074 return (skb_has_frags(skb
) && !(dev
->features
& NETIF_F_FRAGLIST
)) ||
2075 (skb_shinfo(skb
)->nr_frags
&& (!(dev
->features
& NETIF_F_SG
) ||
2076 illegal_highdma(dev
, skb
)));
2080 * dev_queue_xmit - transmit a buffer
2081 * @skb: buffer to transmit
2083 * Queue a buffer for transmission to a network device. The caller must
2084 * have set the device and priority and built the buffer before calling
2085 * this function. The function can be called from an interrupt.
2087 * A negative errno code is returned on a failure. A success does not
2088 * guarantee the frame will be transmitted as it may be dropped due
2089 * to congestion or traffic shaping.
2091 * -----------------------------------------------------------------------------------
2092 * I notice this method can also return errors from the queue disciplines,
2093 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2096 * Regardless of the return value, the skb is consumed, so it is currently
2097 * difficult to retry a send to this method. (You can bump the ref count
2098 * before sending to hold a reference for retry if you are careful.)
2100 * When calling this method, interrupts MUST be enabled. This is because
2101 * the BH enable code must have IRQs enabled so that it will not deadlock.
2104 int dev_queue_xmit(struct sk_buff
*skb
)
2106 struct net_device
*dev
= skb
->dev
;
2107 struct netdev_queue
*txq
;
2111 /* GSO will handle the following emulations directly. */
2112 if (netif_needs_gso(dev
, skb
))
2115 /* Convert a paged skb to linear, if required */
2116 if (skb_needs_linearize(skb
, dev
) && __skb_linearize(skb
))
2119 /* If packet is not checksummed and device does not support
2120 * checksumming for this protocol, complete checksumming here.
2122 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2123 skb_set_transport_header(skb
, skb
->csum_start
-
2125 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
2130 /* Disable soft irqs for various locks below. Also
2131 * stops preemption for RCU.
2135 txq
= dev_pick_tx(dev
, skb
);
2136 q
= rcu_dereference_bh(txq
->qdisc
);
2138 #ifdef CONFIG_NET_CLS_ACT
2139 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2142 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2146 /* The device has no queue. Common case for software devices:
2147 loopback, all the sorts of tunnels...
2149 Really, it is unlikely that netif_tx_lock protection is necessary
2150 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2152 However, it is possible, that they rely on protection
2155 Check this and shot the lock. It is not prone from deadlocks.
2156 Either shot noqueue qdisc, it is even simpler 8)
2158 if (dev
->flags
& IFF_UP
) {
2159 int cpu
= smp_processor_id(); /* ok because BHs are off */
2161 if (txq
->xmit_lock_owner
!= cpu
) {
2163 HARD_TX_LOCK(dev
, txq
, cpu
);
2165 if (!netif_tx_queue_stopped(txq
)) {
2166 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2167 if (dev_xmit_complete(rc
)) {
2168 HARD_TX_UNLOCK(dev
, txq
);
2172 HARD_TX_UNLOCK(dev
, txq
);
2173 if (net_ratelimit())
2174 printk(KERN_CRIT
"Virtual device %s asks to "
2175 "queue packet!\n", dev
->name
);
2177 /* Recursion is detected! It is possible,
2179 if (net_ratelimit())
2180 printk(KERN_CRIT
"Dead loop on virtual device "
2181 "%s, fix it urgently!\n", dev
->name
);
2186 rcu_read_unlock_bh();
2192 rcu_read_unlock_bh();
2195 EXPORT_SYMBOL(dev_queue_xmit
);
2198 /*=======================================================================
2200 =======================================================================*/
2202 int netdev_max_backlog __read_mostly
= 1000;
2203 int netdev_budget __read_mostly
= 300;
2204 int weight_p __read_mostly
= 64; /* old backlog weight */
2206 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
2210 /* One global table that all flow-based protocols share. */
2211 struct rps_sock_flow_table
*rps_sock_flow_table __read_mostly
;
2212 EXPORT_SYMBOL(rps_sock_flow_table
);
2215 * get_rps_cpu is called from netif_receive_skb and returns the target
2216 * CPU from the RPS map of the receiving queue for a given skb.
2217 * rcu_read_lock must be held on entry.
2219 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2220 struct rps_dev_flow
**rflowp
)
2222 struct ipv6hdr
*ip6
;
2224 struct netdev_rx_queue
*rxqueue
;
2225 struct rps_map
*map
;
2226 struct rps_dev_flow_table
*flow_table
;
2227 struct rps_sock_flow_table
*sock_flow_table
;
2231 u32 addr1
, addr2
, ports
, ihl
;
2233 if (skb_rx_queue_recorded(skb
)) {
2234 u16 index
= skb_get_rx_queue(skb
);
2235 if (unlikely(index
>= dev
->num_rx_queues
)) {
2236 if (net_ratelimit()) {
2237 pr_warning("%s received packet on queue "
2238 "%u, but number of RX queues is %u\n",
2239 dev
->name
, index
, dev
->num_rx_queues
);
2243 rxqueue
= dev
->_rx
+ index
;
2247 if (!rxqueue
->rps_map
&& !rxqueue
->rps_flow_table
)
2251 goto got_hash
; /* Skip hash computation on packet header */
2253 switch (skb
->protocol
) {
2254 case __constant_htons(ETH_P_IP
):
2255 if (!pskb_may_pull(skb
, sizeof(*ip
)))
2258 ip
= (struct iphdr
*) skb
->data
;
2259 ip_proto
= ip
->protocol
;
2260 addr1
= (__force u32
) ip
->saddr
;
2261 addr2
= (__force u32
) ip
->daddr
;
2264 case __constant_htons(ETH_P_IPV6
):
2265 if (!pskb_may_pull(skb
, sizeof(*ip6
)))
2268 ip6
= (struct ipv6hdr
*) skb
->data
;
2269 ip_proto
= ip6
->nexthdr
;
2270 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2271 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2285 case IPPROTO_UDPLITE
:
2286 if (pskb_may_pull(skb
, (ihl
* 4) + 4)) {
2287 __be16
*hports
= (__be16
*) (skb
->data
+ (ihl
* 4));
2290 sport
= (__force u16
) hports
[0];
2291 dport
= (__force u16
) hports
[1];
2294 ports
= (sport
<< 16) + dport
;
2302 /* get a consistent hash (same value on both flow directions) */
2305 skb
->rxhash
= jhash_3words(addr1
, addr2
, ports
, hashrnd
);
2310 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2311 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2312 if (flow_table
&& sock_flow_table
) {
2314 struct rps_dev_flow
*rflow
;
2316 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2319 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2320 sock_flow_table
->mask
];
2323 * If the desired CPU (where last recvmsg was done) is
2324 * different from current CPU (one in the rx-queue flow
2325 * table entry), switch if one of the following holds:
2326 * - Current CPU is unset (equal to RPS_NO_CPU).
2327 * - Current CPU is offline.
2328 * - The current CPU's queue tail has advanced beyond the
2329 * last packet that was enqueued using this table entry.
2330 * This guarantees that all previous packets for the flow
2331 * have been dequeued, thus preserving in order delivery.
2333 if (unlikely(tcpu
!= next_cpu
) &&
2334 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2335 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2336 rflow
->last_qtail
)) >= 0)) {
2337 tcpu
= rflow
->cpu
= next_cpu
;
2338 if (tcpu
!= RPS_NO_CPU
)
2339 rflow
->last_qtail
= per_cpu(softnet_data
,
2340 tcpu
).input_queue_head
;
2342 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2349 map
= rcu_dereference(rxqueue
->rps_map
);
2351 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2353 if (cpu_online(tcpu
)) {
2363 /* Called from hardirq (IPI) context */
2364 static void rps_trigger_softirq(void *data
)
2366 struct softnet_data
*sd
= data
;
2368 __napi_schedule(&sd
->backlog
);
2369 __get_cpu_var(netdev_rx_stat
).received_rps
++;
2372 #endif /* CONFIG_RPS */
2375 * Check if this softnet_data structure is another cpu one
2376 * If yes, queue it to our IPI list and return 1
2379 static int rps_ipi_queued(struct softnet_data
*sd
)
2382 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2385 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2386 mysd
->rps_ipi_list
= sd
;
2388 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2391 #endif /* CONFIG_RPS */
2396 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2397 * queue (may be a remote CPU queue).
2399 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2400 unsigned int *qtail
)
2402 struct softnet_data
*sd
;
2403 unsigned long flags
;
2405 sd
= &per_cpu(softnet_data
, cpu
);
2407 local_irq_save(flags
);
2408 __get_cpu_var(netdev_rx_stat
).total
++;
2411 if (sd
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
2412 if (sd
->input_pkt_queue
.qlen
) {
2414 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2416 *qtail
= sd
->input_queue_head
+ sd
->input_pkt_queue
.qlen
;
2419 local_irq_restore(flags
);
2420 return NET_RX_SUCCESS
;
2423 /* Schedule NAPI for backlog device */
2424 if (napi_schedule_prep(&sd
->backlog
)) {
2425 if (!rps_ipi_queued(sd
))
2426 __napi_schedule(&sd
->backlog
);
2433 __get_cpu_var(netdev_rx_stat
).dropped
++;
2434 local_irq_restore(flags
);
2441 * netif_rx - post buffer to the network code
2442 * @skb: buffer to post
2444 * This function receives a packet from a device driver and queues it for
2445 * the upper (protocol) levels to process. It always succeeds. The buffer
2446 * may be dropped during processing for congestion control or by the
2450 * NET_RX_SUCCESS (no congestion)
2451 * NET_RX_DROP (packet was dropped)
2455 int netif_rx(struct sk_buff
*skb
)
2459 /* if netpoll wants it, pretend we never saw it */
2460 if (netpoll_rx(skb
))
2463 if (!skb
->tstamp
.tv64
)
2468 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2473 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2475 cpu
= smp_processor_id();
2477 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2484 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2490 EXPORT_SYMBOL(netif_rx
);
2492 int netif_rx_ni(struct sk_buff
*skb
)
2497 err
= netif_rx(skb
);
2498 if (local_softirq_pending())
2504 EXPORT_SYMBOL(netif_rx_ni
);
2506 static void net_tx_action(struct softirq_action
*h
)
2508 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2510 if (sd
->completion_queue
) {
2511 struct sk_buff
*clist
;
2513 local_irq_disable();
2514 clist
= sd
->completion_queue
;
2515 sd
->completion_queue
= NULL
;
2519 struct sk_buff
*skb
= clist
;
2520 clist
= clist
->next
;
2522 WARN_ON(atomic_read(&skb
->users
));
2527 if (sd
->output_queue
) {
2530 local_irq_disable();
2531 head
= sd
->output_queue
;
2532 sd
->output_queue
= NULL
;
2536 struct Qdisc
*q
= head
;
2537 spinlock_t
*root_lock
;
2539 head
= head
->next_sched
;
2541 root_lock
= qdisc_lock(q
);
2542 if (spin_trylock(root_lock
)) {
2543 smp_mb__before_clear_bit();
2544 clear_bit(__QDISC_STATE_SCHED
,
2547 spin_unlock(root_lock
);
2549 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2551 __netif_reschedule(q
);
2553 smp_mb__before_clear_bit();
2554 clear_bit(__QDISC_STATE_SCHED
,
2562 static inline int deliver_skb(struct sk_buff
*skb
,
2563 struct packet_type
*pt_prev
,
2564 struct net_device
*orig_dev
)
2566 atomic_inc(&skb
->users
);
2567 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2570 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2572 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2573 /* This hook is defined here for ATM LANE */
2574 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2575 unsigned char *addr
) __read_mostly
;
2576 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2580 * If bridge module is loaded call bridging hook.
2581 * returns NULL if packet was consumed.
2583 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
2584 struct sk_buff
*skb
) __read_mostly
;
2585 EXPORT_SYMBOL_GPL(br_handle_frame_hook
);
2587 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
2588 struct packet_type
**pt_prev
, int *ret
,
2589 struct net_device
*orig_dev
)
2591 struct net_bridge_port
*port
;
2593 if (skb
->pkt_type
== PACKET_LOOPBACK
||
2594 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
2598 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2602 return br_handle_frame_hook(port
, skb
);
2605 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2608 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2609 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
2610 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
2612 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
2613 struct packet_type
**pt_prev
,
2615 struct net_device
*orig_dev
)
2617 if (skb
->dev
->macvlan_port
== NULL
)
2621 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2624 return macvlan_handle_frame_hook(skb
);
2627 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2630 #ifdef CONFIG_NET_CLS_ACT
2631 /* TODO: Maybe we should just force sch_ingress to be compiled in
2632 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2633 * a compare and 2 stores extra right now if we dont have it on
2634 * but have CONFIG_NET_CLS_ACT
2635 * NOTE: This doesnt stop any functionality; if you dont have
2636 * the ingress scheduler, you just cant add policies on ingress.
2639 static int ing_filter(struct sk_buff
*skb
)
2641 struct net_device
*dev
= skb
->dev
;
2642 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2643 struct netdev_queue
*rxq
;
2644 int result
= TC_ACT_OK
;
2647 if (MAX_RED_LOOP
< ttl
++) {
2649 "Redir loop detected Dropping packet (%d->%d)\n",
2650 skb
->skb_iif
, dev
->ifindex
);
2654 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2655 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2657 rxq
= &dev
->rx_queue
;
2660 if (q
!= &noop_qdisc
) {
2661 spin_lock(qdisc_lock(q
));
2662 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2663 result
= qdisc_enqueue_root(skb
, q
);
2664 spin_unlock(qdisc_lock(q
));
2670 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2671 struct packet_type
**pt_prev
,
2672 int *ret
, struct net_device
*orig_dev
)
2674 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2678 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2681 /* Huh? Why does turning on AF_PACKET affect this? */
2682 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2685 switch (ing_filter(skb
)) {
2699 * netif_nit_deliver - deliver received packets to network taps
2702 * This function is used to deliver incoming packets to network
2703 * taps. It should be used when the normal netif_receive_skb path
2704 * is bypassed, for example because of VLAN acceleration.
2706 void netif_nit_deliver(struct sk_buff
*skb
)
2708 struct packet_type
*ptype
;
2710 if (list_empty(&ptype_all
))
2713 skb_reset_network_header(skb
);
2714 skb_reset_transport_header(skb
);
2715 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2718 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2719 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2720 deliver_skb(skb
, ptype
, skb
->dev
);
2725 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2726 struct net_device
*master
)
2728 if (skb
->pkt_type
== PACKET_HOST
) {
2729 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2731 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2735 /* On bonding slaves other than the currently active slave, suppress
2736 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2737 * ARP on active-backup slaves with arp_validate enabled.
2739 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2741 struct net_device
*dev
= skb
->dev
;
2743 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2744 dev
->last_rx
= jiffies
;
2746 if ((master
->priv_flags
& IFF_MASTER_ALB
) && master
->br_port
) {
2747 /* Do address unmangle. The local destination address
2748 * will be always the one master has. Provides the right
2749 * functionality in a bridge.
2751 skb_bond_set_mac_by_master(skb
, master
);
2754 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
2755 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
2756 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
2759 if (master
->priv_flags
& IFF_MASTER_ALB
) {
2760 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
2761 skb
->pkt_type
!= PACKET_MULTICAST
)
2764 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
2765 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
2772 EXPORT_SYMBOL(__skb_bond_should_drop
);
2774 static int __netif_receive_skb(struct sk_buff
*skb
)
2776 struct packet_type
*ptype
, *pt_prev
;
2777 struct net_device
*orig_dev
;
2778 struct net_device
*master
;
2779 struct net_device
*null_or_orig
;
2780 struct net_device
*null_or_bond
;
2781 int ret
= NET_RX_DROP
;
2784 if (!skb
->tstamp
.tv64
)
2787 if (vlan_tx_tag_present(skb
) && vlan_hwaccel_do_receive(skb
))
2788 return NET_RX_SUCCESS
;
2790 /* if we've gotten here through NAPI, check netpoll */
2791 if (netpoll_receive_skb(skb
))
2795 skb
->skb_iif
= skb
->dev
->ifindex
;
2797 null_or_orig
= NULL
;
2798 orig_dev
= skb
->dev
;
2799 master
= ACCESS_ONCE(orig_dev
->master
);
2801 if (skb_bond_should_drop(skb
, master
))
2802 null_or_orig
= orig_dev
; /* deliver only exact match */
2807 __get_cpu_var(netdev_rx_stat
).total
++;
2809 skb_reset_network_header(skb
);
2810 skb_reset_transport_header(skb
);
2811 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2817 #ifdef CONFIG_NET_CLS_ACT
2818 if (skb
->tc_verd
& TC_NCLS
) {
2819 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2824 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2825 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2826 ptype
->dev
== orig_dev
) {
2828 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2833 #ifdef CONFIG_NET_CLS_ACT
2834 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2840 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
2843 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
2848 * Make sure frames received on VLAN interfaces stacked on
2849 * bonding interfaces still make their way to any base bonding
2850 * device that may have registered for a specific ptype. The
2851 * handler may have to adjust skb->dev and orig_dev.
2853 null_or_bond
= NULL
;
2854 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
2855 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
2856 null_or_bond
= vlan_dev_real_dev(skb
->dev
);
2859 type
= skb
->protocol
;
2860 list_for_each_entry_rcu(ptype
,
2861 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2862 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
2863 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
2864 ptype
->dev
== null_or_bond
)) {
2866 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2872 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2875 /* Jamal, now you will not able to escape explaining
2876 * me how you were going to use this. :-)
2887 * netif_receive_skb - process receive buffer from network
2888 * @skb: buffer to process
2890 * netif_receive_skb() is the main receive data processing function.
2891 * It always succeeds. The buffer may be dropped during processing
2892 * for congestion control or by the protocol layers.
2894 * This function may only be called from softirq context and interrupts
2895 * should be enabled.
2897 * Return values (usually ignored):
2898 * NET_RX_SUCCESS: no congestion
2899 * NET_RX_DROP: packet was dropped
2901 int netif_receive_skb(struct sk_buff
*skb
)
2904 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2909 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2912 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2916 ret
= __netif_receive_skb(skb
);
2921 return __netif_receive_skb(skb
);
2924 EXPORT_SYMBOL(netif_receive_skb
);
2926 /* Network device is going away, flush any packets still pending
2927 * Called with irqs disabled.
2929 static void flush_backlog(void *arg
)
2931 struct net_device
*dev
= arg
;
2932 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2933 struct sk_buff
*skb
, *tmp
;
2936 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
)
2937 if (skb
->dev
== dev
) {
2938 __skb_unlink(skb
, &sd
->input_pkt_queue
);
2940 input_queue_head_incr(sd
);
2945 static int napi_gro_complete(struct sk_buff
*skb
)
2947 struct packet_type
*ptype
;
2948 __be16 type
= skb
->protocol
;
2949 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2952 if (NAPI_GRO_CB(skb
)->count
== 1) {
2953 skb_shinfo(skb
)->gso_size
= 0;
2958 list_for_each_entry_rcu(ptype
, head
, list
) {
2959 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
2962 err
= ptype
->gro_complete(skb
);
2968 WARN_ON(&ptype
->list
== head
);
2970 return NET_RX_SUCCESS
;
2974 return netif_receive_skb(skb
);
2977 static void napi_gro_flush(struct napi_struct
*napi
)
2979 struct sk_buff
*skb
, *next
;
2981 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
2984 napi_gro_complete(skb
);
2987 napi
->gro_count
= 0;
2988 napi
->gro_list
= NULL
;
2991 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
2993 struct sk_buff
**pp
= NULL
;
2994 struct packet_type
*ptype
;
2995 __be16 type
= skb
->protocol
;
2996 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2999 enum gro_result ret
;
3001 if (!(skb
->dev
->features
& NETIF_F_GRO
))
3004 if (skb_is_gso(skb
) || skb_has_frags(skb
))
3008 list_for_each_entry_rcu(ptype
, head
, list
) {
3009 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3012 skb_set_network_header(skb
, skb_gro_offset(skb
));
3013 mac_len
= skb
->network_header
- skb
->mac_header
;
3014 skb
->mac_len
= mac_len
;
3015 NAPI_GRO_CB(skb
)->same_flow
= 0;
3016 NAPI_GRO_CB(skb
)->flush
= 0;
3017 NAPI_GRO_CB(skb
)->free
= 0;
3019 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3024 if (&ptype
->list
== head
)
3027 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3028 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3031 struct sk_buff
*nskb
= *pp
;
3035 napi_gro_complete(nskb
);
3042 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3046 NAPI_GRO_CB(skb
)->count
= 1;
3047 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3048 skb
->next
= napi
->gro_list
;
3049 napi
->gro_list
= skb
;
3053 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3054 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3056 BUG_ON(skb
->end
- skb
->tail
< grow
);
3058 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3061 skb
->data_len
-= grow
;
3063 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3064 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3066 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3067 put_page(skb_shinfo(skb
)->frags
[0].page
);
3068 memmove(skb_shinfo(skb
)->frags
,
3069 skb_shinfo(skb
)->frags
+ 1,
3070 --skb_shinfo(skb
)->nr_frags
);
3081 EXPORT_SYMBOL(dev_gro_receive
);
3084 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3088 if (netpoll_rx_on(skb
))
3091 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3092 NAPI_GRO_CB(p
)->same_flow
=
3093 (p
->dev
== skb
->dev
) &&
3094 !compare_ether_header(skb_mac_header(p
),
3095 skb_gro_mac_header(skb
));
3096 NAPI_GRO_CB(p
)->flush
= 0;
3099 return dev_gro_receive(napi
, skb
);
3102 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3106 if (netif_receive_skb(skb
))
3111 case GRO_MERGED_FREE
:
3122 EXPORT_SYMBOL(napi_skb_finish
);
3124 void skb_gro_reset_offset(struct sk_buff
*skb
)
3126 NAPI_GRO_CB(skb
)->data_offset
= 0;
3127 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3128 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3130 if (skb
->mac_header
== skb
->tail
&&
3131 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3132 NAPI_GRO_CB(skb
)->frag0
=
3133 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3134 skb_shinfo(skb
)->frags
[0].page_offset
;
3135 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3138 EXPORT_SYMBOL(skb_gro_reset_offset
);
3140 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3142 skb_gro_reset_offset(skb
);
3144 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3146 EXPORT_SYMBOL(napi_gro_receive
);
3148 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3150 __skb_pull(skb
, skb_headlen(skb
));
3151 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3155 EXPORT_SYMBOL(napi_reuse_skb
);
3157 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3159 struct sk_buff
*skb
= napi
->skb
;
3162 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3168 EXPORT_SYMBOL(napi_get_frags
);
3170 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3176 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3178 if (ret
== GRO_HELD
)
3179 skb_gro_pull(skb
, -ETH_HLEN
);
3180 else if (netif_receive_skb(skb
))
3185 case GRO_MERGED_FREE
:
3186 napi_reuse_skb(napi
, skb
);
3195 EXPORT_SYMBOL(napi_frags_finish
);
3197 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3199 struct sk_buff
*skb
= napi
->skb
;
3206 skb_reset_mac_header(skb
);
3207 skb_gro_reset_offset(skb
);
3209 off
= skb_gro_offset(skb
);
3210 hlen
= off
+ sizeof(*eth
);
3211 eth
= skb_gro_header_fast(skb
, off
);
3212 if (skb_gro_header_hard(skb
, hlen
)) {
3213 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3214 if (unlikely(!eth
)) {
3215 napi_reuse_skb(napi
, skb
);
3221 skb_gro_pull(skb
, sizeof(*eth
));
3224 * This works because the only protocols we care about don't require
3225 * special handling. We'll fix it up properly at the end.
3227 skb
->protocol
= eth
->h_proto
;
3232 EXPORT_SYMBOL(napi_frags_skb
);
3234 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3236 struct sk_buff
*skb
= napi_frags_skb(napi
);
3241 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3243 EXPORT_SYMBOL(napi_gro_frags
);
3245 static int process_backlog(struct napi_struct
*napi
, int quota
)
3248 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3250 napi
->weight
= weight_p
;
3252 struct sk_buff
*skb
;
3254 local_irq_disable();
3256 skb
= __skb_dequeue(&sd
->input_pkt_queue
);
3258 __napi_complete(napi
);
3263 input_queue_head_incr(sd
);
3267 __netif_receive_skb(skb
);
3268 } while (++work
< quota
);
3274 * __napi_schedule - schedule for receive
3275 * @n: entry to schedule
3277 * The entry's receive function will be scheduled to run
3279 void __napi_schedule(struct napi_struct
*n
)
3281 unsigned long flags
;
3283 local_irq_save(flags
);
3284 list_add_tail(&n
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
3285 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3286 local_irq_restore(flags
);
3288 EXPORT_SYMBOL(__napi_schedule
);
3290 void __napi_complete(struct napi_struct
*n
)
3292 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3293 BUG_ON(n
->gro_list
);
3295 list_del(&n
->poll_list
);
3296 smp_mb__before_clear_bit();
3297 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3299 EXPORT_SYMBOL(__napi_complete
);
3301 void napi_complete(struct napi_struct
*n
)
3303 unsigned long flags
;
3306 * don't let napi dequeue from the cpu poll list
3307 * just in case its running on a different cpu
3309 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3313 local_irq_save(flags
);
3315 local_irq_restore(flags
);
3317 EXPORT_SYMBOL(napi_complete
);
3319 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3320 int (*poll
)(struct napi_struct
*, int), int weight
)
3322 INIT_LIST_HEAD(&napi
->poll_list
);
3323 napi
->gro_count
= 0;
3324 napi
->gro_list
= NULL
;
3327 napi
->weight
= weight
;
3328 list_add(&napi
->dev_list
, &dev
->napi_list
);
3330 #ifdef CONFIG_NETPOLL
3331 spin_lock_init(&napi
->poll_lock
);
3332 napi
->poll_owner
= -1;
3334 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3336 EXPORT_SYMBOL(netif_napi_add
);
3338 void netif_napi_del(struct napi_struct
*napi
)
3340 struct sk_buff
*skb
, *next
;
3342 list_del_init(&napi
->dev_list
);
3343 napi_free_frags(napi
);
3345 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3351 napi
->gro_list
= NULL
;
3352 napi
->gro_count
= 0;
3354 EXPORT_SYMBOL(netif_napi_del
);
3357 * net_rps_action sends any pending IPI's for rps.
3358 * Note: called with local irq disabled, but exits with local irq enabled.
3360 static void net_rps_action_and_irq_disable(void)
3363 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3364 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3367 sd
->rps_ipi_list
= NULL
;
3371 /* Send pending IPI's to kick RPS processing on remote cpus. */
3373 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3375 if (cpu_online(remsd
->cpu
))
3376 __smp_call_function_single(remsd
->cpu
,
3385 static void net_rx_action(struct softirq_action
*h
)
3387 struct list_head
*list
= &__get_cpu_var(softnet_data
).poll_list
;
3388 unsigned long time_limit
= jiffies
+ 2;
3389 int budget
= netdev_budget
;
3392 local_irq_disable();
3394 while (!list_empty(list
)) {
3395 struct napi_struct
*n
;
3398 /* If softirq window is exhuasted then punt.
3399 * Allow this to run for 2 jiffies since which will allow
3400 * an average latency of 1.5/HZ.
3402 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3407 /* Even though interrupts have been re-enabled, this
3408 * access is safe because interrupts can only add new
3409 * entries to the tail of this list, and only ->poll()
3410 * calls can remove this head entry from the list.
3412 n
= list_first_entry(list
, struct napi_struct
, poll_list
);
3414 have
= netpoll_poll_lock(n
);
3418 /* This NAPI_STATE_SCHED test is for avoiding a race
3419 * with netpoll's poll_napi(). Only the entity which
3420 * obtains the lock and sees NAPI_STATE_SCHED set will
3421 * actually make the ->poll() call. Therefore we avoid
3422 * accidently calling ->poll() when NAPI is not scheduled.
3425 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3426 work
= n
->poll(n
, weight
);
3430 WARN_ON_ONCE(work
> weight
);
3434 local_irq_disable();
3436 /* Drivers must not modify the NAPI state if they
3437 * consume the entire weight. In such cases this code
3438 * still "owns" the NAPI instance and therefore can
3439 * move the instance around on the list at-will.
3441 if (unlikely(work
== weight
)) {
3442 if (unlikely(napi_disable_pending(n
))) {
3445 local_irq_disable();
3447 list_move_tail(&n
->poll_list
, list
);
3450 netpoll_poll_unlock(have
);
3453 net_rps_action_and_irq_disable();
3455 #ifdef CONFIG_NET_DMA
3457 * There may not be any more sk_buffs coming right now, so push
3458 * any pending DMA copies to hardware
3460 dma_issue_pending_all();
3466 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
3467 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3471 static gifconf_func_t
*gifconf_list
[NPROTO
];
3474 * register_gifconf - register a SIOCGIF handler
3475 * @family: Address family
3476 * @gifconf: Function handler
3478 * Register protocol dependent address dumping routines. The handler
3479 * that is passed must not be freed or reused until it has been replaced
3480 * by another handler.
3482 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3484 if (family
>= NPROTO
)
3486 gifconf_list
[family
] = gifconf
;
3489 EXPORT_SYMBOL(register_gifconf
);
3493 * Map an interface index to its name (SIOCGIFNAME)
3497 * We need this ioctl for efficient implementation of the
3498 * if_indextoname() function required by the IPv6 API. Without
3499 * it, we would have to search all the interfaces to find a
3503 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3505 struct net_device
*dev
;
3509 * Fetch the caller's info block.
3512 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3516 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3522 strcpy(ifr
.ifr_name
, dev
->name
);
3525 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3531 * Perform a SIOCGIFCONF call. This structure will change
3532 * size eventually, and there is nothing I can do about it.
3533 * Thus we will need a 'compatibility mode'.
3536 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3539 struct net_device
*dev
;
3546 * Fetch the caller's info block.
3549 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3556 * Loop over the interfaces, and write an info block for each.
3560 for_each_netdev(net
, dev
) {
3561 for (i
= 0; i
< NPROTO
; i
++) {
3562 if (gifconf_list
[i
]) {
3565 done
= gifconf_list
[i
](dev
, NULL
, 0);
3567 done
= gifconf_list
[i
](dev
, pos
+ total
,
3577 * All done. Write the updated control block back to the caller.
3579 ifc
.ifc_len
= total
;
3582 * Both BSD and Solaris return 0 here, so we do too.
3584 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3587 #ifdef CONFIG_PROC_FS
3589 * This is invoked by the /proc filesystem handler to display a device
3592 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3595 struct net
*net
= seq_file_net(seq
);
3597 struct net_device
*dev
;
3601 return SEQ_START_TOKEN
;
3604 for_each_netdev_rcu(net
, dev
)
3611 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3613 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3614 first_net_device(seq_file_net(seq
)) :
3615 next_net_device((struct net_device
*)v
);
3618 return rcu_dereference(dev
);
3621 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3627 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3629 const struct net_device_stats
*stats
= dev_get_stats(dev
);
3631 seq_printf(seq
, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3632 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3633 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3635 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3636 stats
->rx_fifo_errors
,
3637 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3638 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3639 stats
->rx_compressed
, stats
->multicast
,
3640 stats
->tx_bytes
, stats
->tx_packets
,
3641 stats
->tx_errors
, stats
->tx_dropped
,
3642 stats
->tx_fifo_errors
, stats
->collisions
,
3643 stats
->tx_carrier_errors
+
3644 stats
->tx_aborted_errors
+
3645 stats
->tx_window_errors
+
3646 stats
->tx_heartbeat_errors
,
3647 stats
->tx_compressed
);
3651 * Called from the PROCfs module. This now uses the new arbitrary sized
3652 * /proc/net interface to create /proc/net/dev
3654 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3656 if (v
== SEQ_START_TOKEN
)
3657 seq_puts(seq
, "Inter-| Receive "
3659 " face |bytes packets errs drop fifo frame "
3660 "compressed multicast|bytes packets errs "
3661 "drop fifo colls carrier compressed\n");
3663 dev_seq_printf_stats(seq
, v
);
3667 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
3669 struct netif_rx_stats
*rc
= NULL
;
3671 while (*pos
< nr_cpu_ids
)
3672 if (cpu_online(*pos
)) {
3673 rc
= &per_cpu(netdev_rx_stat
, *pos
);
3680 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3682 return softnet_get_online(pos
);
3685 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3688 return softnet_get_online(pos
);
3691 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3695 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3697 struct netif_rx_stats
*s
= v
;
3699 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3700 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
3701 0, 0, 0, 0, /* was fastroute */
3702 s
->cpu_collision
, s
->received_rps
);
3706 static const struct seq_operations dev_seq_ops
= {
3707 .start
= dev_seq_start
,
3708 .next
= dev_seq_next
,
3709 .stop
= dev_seq_stop
,
3710 .show
= dev_seq_show
,
3713 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3715 return seq_open_net(inode
, file
, &dev_seq_ops
,
3716 sizeof(struct seq_net_private
));
3719 static const struct file_operations dev_seq_fops
= {
3720 .owner
= THIS_MODULE
,
3721 .open
= dev_seq_open
,
3723 .llseek
= seq_lseek
,
3724 .release
= seq_release_net
,
3727 static const struct seq_operations softnet_seq_ops
= {
3728 .start
= softnet_seq_start
,
3729 .next
= softnet_seq_next
,
3730 .stop
= softnet_seq_stop
,
3731 .show
= softnet_seq_show
,
3734 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3736 return seq_open(file
, &softnet_seq_ops
);
3739 static const struct file_operations softnet_seq_fops
= {
3740 .owner
= THIS_MODULE
,
3741 .open
= softnet_seq_open
,
3743 .llseek
= seq_lseek
,
3744 .release
= seq_release
,
3747 static void *ptype_get_idx(loff_t pos
)
3749 struct packet_type
*pt
= NULL
;
3753 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3759 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3760 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3769 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3773 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3776 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3778 struct packet_type
*pt
;
3779 struct list_head
*nxt
;
3783 if (v
== SEQ_START_TOKEN
)
3784 return ptype_get_idx(0);
3787 nxt
= pt
->list
.next
;
3788 if (pt
->type
== htons(ETH_P_ALL
)) {
3789 if (nxt
!= &ptype_all
)
3792 nxt
= ptype_base
[0].next
;
3794 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3796 while (nxt
== &ptype_base
[hash
]) {
3797 if (++hash
>= PTYPE_HASH_SIZE
)
3799 nxt
= ptype_base
[hash
].next
;
3802 return list_entry(nxt
, struct packet_type
, list
);
3805 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3811 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3813 struct packet_type
*pt
= v
;
3815 if (v
== SEQ_START_TOKEN
)
3816 seq_puts(seq
, "Type Device Function\n");
3817 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3818 if (pt
->type
== htons(ETH_P_ALL
))
3819 seq_puts(seq
, "ALL ");
3821 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3823 seq_printf(seq
, " %-8s %pF\n",
3824 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
3830 static const struct seq_operations ptype_seq_ops
= {
3831 .start
= ptype_seq_start
,
3832 .next
= ptype_seq_next
,
3833 .stop
= ptype_seq_stop
,
3834 .show
= ptype_seq_show
,
3837 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
3839 return seq_open_net(inode
, file
, &ptype_seq_ops
,
3840 sizeof(struct seq_net_private
));
3843 static const struct file_operations ptype_seq_fops
= {
3844 .owner
= THIS_MODULE
,
3845 .open
= ptype_seq_open
,
3847 .llseek
= seq_lseek
,
3848 .release
= seq_release_net
,
3852 static int __net_init
dev_proc_net_init(struct net
*net
)
3856 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
3858 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
3860 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
3863 if (wext_proc_init(net
))
3869 proc_net_remove(net
, "ptype");
3871 proc_net_remove(net
, "softnet_stat");
3873 proc_net_remove(net
, "dev");
3877 static void __net_exit
dev_proc_net_exit(struct net
*net
)
3879 wext_proc_exit(net
);
3881 proc_net_remove(net
, "ptype");
3882 proc_net_remove(net
, "softnet_stat");
3883 proc_net_remove(net
, "dev");
3886 static struct pernet_operations __net_initdata dev_proc_ops
= {
3887 .init
= dev_proc_net_init
,
3888 .exit
= dev_proc_net_exit
,
3891 static int __init
dev_proc_init(void)
3893 return register_pernet_subsys(&dev_proc_ops
);
3896 #define dev_proc_init() 0
3897 #endif /* CONFIG_PROC_FS */
3901 * netdev_set_master - set up master/slave pair
3902 * @slave: slave device
3903 * @master: new master device
3905 * Changes the master device of the slave. Pass %NULL to break the
3906 * bonding. The caller must hold the RTNL semaphore. On a failure
3907 * a negative errno code is returned. On success the reference counts
3908 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3909 * function returns zero.
3911 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
3913 struct net_device
*old
= slave
->master
;
3923 slave
->master
= master
;
3930 slave
->flags
|= IFF_SLAVE
;
3932 slave
->flags
&= ~IFF_SLAVE
;
3934 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
3937 EXPORT_SYMBOL(netdev_set_master
);
3939 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
3941 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3943 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
3944 ops
->ndo_change_rx_flags(dev
, flags
);
3947 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
3949 unsigned short old_flags
= dev
->flags
;
3955 dev
->flags
|= IFF_PROMISC
;
3956 dev
->promiscuity
+= inc
;
3957 if (dev
->promiscuity
== 0) {
3960 * If inc causes overflow, untouch promisc and return error.
3963 dev
->flags
&= ~IFF_PROMISC
;
3965 dev
->promiscuity
-= inc
;
3966 printk(KERN_WARNING
"%s: promiscuity touches roof, "
3967 "set promiscuity failed, promiscuity feature "
3968 "of device might be broken.\n", dev
->name
);
3972 if (dev
->flags
!= old_flags
) {
3973 printk(KERN_INFO
"device %s %s promiscuous mode\n",
3974 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
3976 if (audit_enabled
) {
3977 current_uid_gid(&uid
, &gid
);
3978 audit_log(current
->audit_context
, GFP_ATOMIC
,
3979 AUDIT_ANOM_PROMISCUOUS
,
3980 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3981 dev
->name
, (dev
->flags
& IFF_PROMISC
),
3982 (old_flags
& IFF_PROMISC
),
3983 audit_get_loginuid(current
),
3985 audit_get_sessionid(current
));
3988 dev_change_rx_flags(dev
, IFF_PROMISC
);
3994 * dev_set_promiscuity - update promiscuity count on a device
3998 * Add or remove promiscuity from a device. While the count in the device
3999 * remains above zero the interface remains promiscuous. Once it hits zero
4000 * the device reverts back to normal filtering operation. A negative inc
4001 * value is used to drop promiscuity on the device.
4002 * Return 0 if successful or a negative errno code on error.
4004 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4006 unsigned short old_flags
= dev
->flags
;
4009 err
= __dev_set_promiscuity(dev
, inc
);
4012 if (dev
->flags
!= old_flags
)
4013 dev_set_rx_mode(dev
);
4016 EXPORT_SYMBOL(dev_set_promiscuity
);
4019 * dev_set_allmulti - update allmulti count on a device
4023 * Add or remove reception of all multicast frames to a device. While the
4024 * count in the device remains above zero the interface remains listening
4025 * to all interfaces. Once it hits zero the device reverts back to normal
4026 * filtering operation. A negative @inc value is used to drop the counter
4027 * when releasing a resource needing all multicasts.
4028 * Return 0 if successful or a negative errno code on error.
4031 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4033 unsigned short old_flags
= dev
->flags
;
4037 dev
->flags
|= IFF_ALLMULTI
;
4038 dev
->allmulti
+= inc
;
4039 if (dev
->allmulti
== 0) {
4042 * If inc causes overflow, untouch allmulti and return error.
4045 dev
->flags
&= ~IFF_ALLMULTI
;
4047 dev
->allmulti
-= inc
;
4048 printk(KERN_WARNING
"%s: allmulti touches roof, "
4049 "set allmulti failed, allmulti feature of "
4050 "device might be broken.\n", dev
->name
);
4054 if (dev
->flags
^ old_flags
) {
4055 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4056 dev_set_rx_mode(dev
);
4060 EXPORT_SYMBOL(dev_set_allmulti
);
4063 * Upload unicast and multicast address lists to device and
4064 * configure RX filtering. When the device doesn't support unicast
4065 * filtering it is put in promiscuous mode while unicast addresses
4068 void __dev_set_rx_mode(struct net_device
*dev
)
4070 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4072 /* dev_open will call this function so the list will stay sane. */
4073 if (!(dev
->flags
&IFF_UP
))
4076 if (!netif_device_present(dev
))
4079 if (ops
->ndo_set_rx_mode
)
4080 ops
->ndo_set_rx_mode(dev
);
4082 /* Unicast addresses changes may only happen under the rtnl,
4083 * therefore calling __dev_set_promiscuity here is safe.
4085 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4086 __dev_set_promiscuity(dev
, 1);
4087 dev
->uc_promisc
= 1;
4088 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4089 __dev_set_promiscuity(dev
, -1);
4090 dev
->uc_promisc
= 0;
4093 if (ops
->ndo_set_multicast_list
)
4094 ops
->ndo_set_multicast_list(dev
);
4098 void dev_set_rx_mode(struct net_device
*dev
)
4100 netif_addr_lock_bh(dev
);
4101 __dev_set_rx_mode(dev
);
4102 netif_addr_unlock_bh(dev
);
4106 * dev_get_flags - get flags reported to userspace
4109 * Get the combination of flag bits exported through APIs to userspace.
4111 unsigned dev_get_flags(const struct net_device
*dev
)
4115 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4120 (dev
->gflags
& (IFF_PROMISC
|
4123 if (netif_running(dev
)) {
4124 if (netif_oper_up(dev
))
4125 flags
|= IFF_RUNNING
;
4126 if (netif_carrier_ok(dev
))
4127 flags
|= IFF_LOWER_UP
;
4128 if (netif_dormant(dev
))
4129 flags
|= IFF_DORMANT
;
4134 EXPORT_SYMBOL(dev_get_flags
);
4136 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4138 int old_flags
= dev
->flags
;
4144 * Set the flags on our device.
4147 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4148 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4150 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4154 * Load in the correct multicast list now the flags have changed.
4157 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4158 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4160 dev_set_rx_mode(dev
);
4163 * Have we downed the interface. We handle IFF_UP ourselves
4164 * according to user attempts to set it, rather than blindly
4169 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4170 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4173 dev_set_rx_mode(dev
);
4176 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4177 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4179 dev
->gflags
^= IFF_PROMISC
;
4180 dev_set_promiscuity(dev
, inc
);
4183 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4184 is important. Some (broken) drivers set IFF_PROMISC, when
4185 IFF_ALLMULTI is requested not asking us and not reporting.
4187 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4188 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4190 dev
->gflags
^= IFF_ALLMULTI
;
4191 dev_set_allmulti(dev
, inc
);
4197 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4199 unsigned int changes
= dev
->flags
^ old_flags
;
4201 if (changes
& IFF_UP
) {
4202 if (dev
->flags
& IFF_UP
)
4203 call_netdevice_notifiers(NETDEV_UP
, dev
);
4205 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4208 if (dev
->flags
& IFF_UP
&&
4209 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4210 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4214 * dev_change_flags - change device settings
4216 * @flags: device state flags
4218 * Change settings on device based state flags. The flags are
4219 * in the userspace exported format.
4221 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4224 int old_flags
= dev
->flags
;
4226 ret
= __dev_change_flags(dev
, flags
);
4230 changes
= old_flags
^ dev
->flags
;
4232 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4234 __dev_notify_flags(dev
, old_flags
);
4237 EXPORT_SYMBOL(dev_change_flags
);
4240 * dev_set_mtu - Change maximum transfer unit
4242 * @new_mtu: new transfer unit
4244 * Change the maximum transfer size of the network device.
4246 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4248 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4251 if (new_mtu
== dev
->mtu
)
4254 /* MTU must be positive. */
4258 if (!netif_device_present(dev
))
4262 if (ops
->ndo_change_mtu
)
4263 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4267 if (!err
&& dev
->flags
& IFF_UP
)
4268 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4271 EXPORT_SYMBOL(dev_set_mtu
);
4274 * dev_set_mac_address - Change Media Access Control Address
4278 * Change the hardware (MAC) address of the device
4280 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4282 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4285 if (!ops
->ndo_set_mac_address
)
4287 if (sa
->sa_family
!= dev
->type
)
4289 if (!netif_device_present(dev
))
4291 err
= ops
->ndo_set_mac_address(dev
, sa
);
4293 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4296 EXPORT_SYMBOL(dev_set_mac_address
);
4299 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4301 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4304 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4310 case SIOCGIFFLAGS
: /* Get interface flags */
4311 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4314 case SIOCGIFMETRIC
: /* Get the metric on the interface
4315 (currently unused) */
4316 ifr
->ifr_metric
= 0;
4319 case SIOCGIFMTU
: /* Get the MTU of a device */
4320 ifr
->ifr_mtu
= dev
->mtu
;
4325 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4327 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4328 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4329 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4337 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4338 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4339 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4340 ifr
->ifr_map
.irq
= dev
->irq
;
4341 ifr
->ifr_map
.dma
= dev
->dma
;
4342 ifr
->ifr_map
.port
= dev
->if_port
;
4346 ifr
->ifr_ifindex
= dev
->ifindex
;
4350 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4354 /* dev_ioctl() should ensure this case
4366 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4368 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4371 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4372 const struct net_device_ops
*ops
;
4377 ops
= dev
->netdev_ops
;
4380 case SIOCSIFFLAGS
: /* Set interface flags */
4381 return dev_change_flags(dev
, ifr
->ifr_flags
);
4383 case SIOCSIFMETRIC
: /* Set the metric on the interface
4384 (currently unused) */
4387 case SIOCSIFMTU
: /* Set the MTU of a device */
4388 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4391 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4393 case SIOCSIFHWBROADCAST
:
4394 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4396 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4397 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4398 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4402 if (ops
->ndo_set_config
) {
4403 if (!netif_device_present(dev
))
4405 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4410 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4411 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4413 if (!netif_device_present(dev
))
4415 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4418 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4419 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4421 if (!netif_device_present(dev
))
4423 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4426 if (ifr
->ifr_qlen
< 0)
4428 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4432 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4433 return dev_change_name(dev
, ifr
->ifr_newname
);
4436 * Unknown or private ioctl
4439 if ((cmd
>= SIOCDEVPRIVATE
&&
4440 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4441 cmd
== SIOCBONDENSLAVE
||
4442 cmd
== SIOCBONDRELEASE
||
4443 cmd
== SIOCBONDSETHWADDR
||
4444 cmd
== SIOCBONDSLAVEINFOQUERY
||
4445 cmd
== SIOCBONDINFOQUERY
||
4446 cmd
== SIOCBONDCHANGEACTIVE
||
4447 cmd
== SIOCGMIIPHY
||
4448 cmd
== SIOCGMIIREG
||
4449 cmd
== SIOCSMIIREG
||
4450 cmd
== SIOCBRADDIF
||
4451 cmd
== SIOCBRDELIF
||
4452 cmd
== SIOCSHWTSTAMP
||
4453 cmd
== SIOCWANDEV
) {
4455 if (ops
->ndo_do_ioctl
) {
4456 if (netif_device_present(dev
))
4457 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4469 * This function handles all "interface"-type I/O control requests. The actual
4470 * 'doing' part of this is dev_ifsioc above.
4474 * dev_ioctl - network device ioctl
4475 * @net: the applicable net namespace
4476 * @cmd: command to issue
4477 * @arg: pointer to a struct ifreq in user space
4479 * Issue ioctl functions to devices. This is normally called by the
4480 * user space syscall interfaces but can sometimes be useful for
4481 * other purposes. The return value is the return from the syscall if
4482 * positive or a negative errno code on error.
4485 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4491 /* One special case: SIOCGIFCONF takes ifconf argument
4492 and requires shared lock, because it sleeps writing
4496 if (cmd
== SIOCGIFCONF
) {
4498 ret
= dev_ifconf(net
, (char __user
*) arg
);
4502 if (cmd
== SIOCGIFNAME
)
4503 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4505 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4508 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4510 colon
= strchr(ifr
.ifr_name
, ':');
4515 * See which interface the caller is talking about.
4520 * These ioctl calls:
4521 * - can be done by all.
4522 * - atomic and do not require locking.
4533 dev_load(net
, ifr
.ifr_name
);
4535 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4540 if (copy_to_user(arg
, &ifr
,
4541 sizeof(struct ifreq
)))
4547 dev_load(net
, ifr
.ifr_name
);
4549 ret
= dev_ethtool(net
, &ifr
);
4554 if (copy_to_user(arg
, &ifr
,
4555 sizeof(struct ifreq
)))
4561 * These ioctl calls:
4562 * - require superuser power.
4563 * - require strict serialization.
4569 if (!capable(CAP_NET_ADMIN
))
4571 dev_load(net
, ifr
.ifr_name
);
4573 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4578 if (copy_to_user(arg
, &ifr
,
4579 sizeof(struct ifreq
)))
4585 * These ioctl calls:
4586 * - require superuser power.
4587 * - require strict serialization.
4588 * - do not return a value
4598 case SIOCSIFHWBROADCAST
:
4601 case SIOCBONDENSLAVE
:
4602 case SIOCBONDRELEASE
:
4603 case SIOCBONDSETHWADDR
:
4604 case SIOCBONDCHANGEACTIVE
:
4608 if (!capable(CAP_NET_ADMIN
))
4611 case SIOCBONDSLAVEINFOQUERY
:
4612 case SIOCBONDINFOQUERY
:
4613 dev_load(net
, ifr
.ifr_name
);
4615 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4620 /* Get the per device memory space. We can add this but
4621 * currently do not support it */
4623 /* Set the per device memory buffer space.
4624 * Not applicable in our case */
4629 * Unknown or private ioctl.
4632 if (cmd
== SIOCWANDEV
||
4633 (cmd
>= SIOCDEVPRIVATE
&&
4634 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4635 dev_load(net
, ifr
.ifr_name
);
4637 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4639 if (!ret
&& copy_to_user(arg
, &ifr
,
4640 sizeof(struct ifreq
)))
4644 /* Take care of Wireless Extensions */
4645 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4646 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4653 * dev_new_index - allocate an ifindex
4654 * @net: the applicable net namespace
4656 * Returns a suitable unique value for a new device interface
4657 * number. The caller must hold the rtnl semaphore or the
4658 * dev_base_lock to be sure it remains unique.
4660 static int dev_new_index(struct net
*net
)
4666 if (!__dev_get_by_index(net
, ifindex
))
4671 /* Delayed registration/unregisteration */
4672 static LIST_HEAD(net_todo_list
);
4674 static void net_set_todo(struct net_device
*dev
)
4676 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4679 static void rollback_registered_many(struct list_head
*head
)
4681 struct net_device
*dev
, *tmp
;
4683 BUG_ON(dev_boot_phase
);
4686 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4687 /* Some devices call without registering
4688 * for initialization unwind. Remove those
4689 * devices and proceed with the remaining.
4691 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4692 pr_debug("unregister_netdevice: device %s/%p never "
4693 "was registered\n", dev
->name
, dev
);
4696 list_del(&dev
->unreg_list
);
4700 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4702 /* If device is running, close it first. */
4705 /* And unlink it from device chain. */
4706 unlist_netdevice(dev
);
4708 dev
->reg_state
= NETREG_UNREGISTERING
;
4713 list_for_each_entry(dev
, head
, unreg_list
) {
4714 /* Shutdown queueing discipline. */
4718 /* Notify protocols, that we are about to destroy
4719 this device. They should clean all the things.
4721 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4723 if (!dev
->rtnl_link_ops
||
4724 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
4725 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
4728 * Flush the unicast and multicast chains
4733 if (dev
->netdev_ops
->ndo_uninit
)
4734 dev
->netdev_ops
->ndo_uninit(dev
);
4736 /* Notifier chain MUST detach us from master device. */
4737 WARN_ON(dev
->master
);
4739 /* Remove entries from kobject tree */
4740 netdev_unregister_kobject(dev
);
4743 /* Process any work delayed until the end of the batch */
4744 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
4745 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
4749 list_for_each_entry(dev
, head
, unreg_list
)
4753 static void rollback_registered(struct net_device
*dev
)
4757 list_add(&dev
->unreg_list
, &single
);
4758 rollback_registered_many(&single
);
4761 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
4762 struct netdev_queue
*dev_queue
,
4765 spin_lock_init(&dev_queue
->_xmit_lock
);
4766 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
4767 dev_queue
->xmit_lock_owner
= -1;
4770 static void netdev_init_queue_locks(struct net_device
*dev
)
4772 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
4773 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
4776 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
4778 /* Fix illegal SG+CSUM combinations. */
4779 if ((features
& NETIF_F_SG
) &&
4780 !(features
& NETIF_F_ALL_CSUM
)) {
4782 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
4783 "checksum feature.\n", name
);
4784 features
&= ~NETIF_F_SG
;
4787 /* TSO requires that SG is present as well. */
4788 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
4790 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
4791 "SG feature.\n", name
);
4792 features
&= ~NETIF_F_TSO
;
4795 if (features
& NETIF_F_UFO
) {
4796 if (!(features
& NETIF_F_GEN_CSUM
)) {
4798 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4799 "since no NETIF_F_HW_CSUM feature.\n",
4801 features
&= ~NETIF_F_UFO
;
4804 if (!(features
& NETIF_F_SG
)) {
4806 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4807 "since no NETIF_F_SG feature.\n", name
);
4808 features
&= ~NETIF_F_UFO
;
4814 EXPORT_SYMBOL(netdev_fix_features
);
4817 * netif_stacked_transfer_operstate - transfer operstate
4818 * @rootdev: the root or lower level device to transfer state from
4819 * @dev: the device to transfer operstate to
4821 * Transfer operational state from root to device. This is normally
4822 * called when a stacking relationship exists between the root
4823 * device and the device(a leaf device).
4825 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
4826 struct net_device
*dev
)
4828 if (rootdev
->operstate
== IF_OPER_DORMANT
)
4829 netif_dormant_on(dev
);
4831 netif_dormant_off(dev
);
4833 if (netif_carrier_ok(rootdev
)) {
4834 if (!netif_carrier_ok(dev
))
4835 netif_carrier_on(dev
);
4837 if (netif_carrier_ok(dev
))
4838 netif_carrier_off(dev
);
4841 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
4844 * register_netdevice - register a network device
4845 * @dev: device to register
4847 * Take a completed network device structure and add it to the kernel
4848 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4849 * chain. 0 is returned on success. A negative errno code is returned
4850 * on a failure to set up the device, or if the name is a duplicate.
4852 * Callers must hold the rtnl semaphore. You may want
4853 * register_netdev() instead of this.
4856 * The locking appears insufficient to guarantee two parallel registers
4857 * will not get the same name.
4860 int register_netdevice(struct net_device
*dev
)
4863 struct net
*net
= dev_net(dev
);
4865 BUG_ON(dev_boot_phase
);
4870 /* When net_device's are persistent, this will be fatal. */
4871 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
4874 spin_lock_init(&dev
->addr_list_lock
);
4875 netdev_set_addr_lockdep_class(dev
);
4876 netdev_init_queue_locks(dev
);
4881 if (!dev
->num_rx_queues
) {
4883 * Allocate a single RX queue if driver never called
4887 dev
->_rx
= kzalloc(sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
4893 dev
->_rx
->first
= dev
->_rx
;
4894 atomic_set(&dev
->_rx
->count
, 1);
4895 dev
->num_rx_queues
= 1;
4898 /* Init, if this function is available */
4899 if (dev
->netdev_ops
->ndo_init
) {
4900 ret
= dev
->netdev_ops
->ndo_init(dev
);
4908 ret
= dev_get_valid_name(net
, dev
->name
, dev
->name
, 0);
4912 dev
->ifindex
= dev_new_index(net
);
4913 if (dev
->iflink
== -1)
4914 dev
->iflink
= dev
->ifindex
;
4916 /* Fix illegal checksum combinations */
4917 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
4918 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4919 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
4921 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
4924 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
4925 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4926 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
4928 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
4931 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
4933 /* Enable software GSO if SG is supported. */
4934 if (dev
->features
& NETIF_F_SG
)
4935 dev
->features
|= NETIF_F_GSO
;
4937 netdev_initialize_kobject(dev
);
4939 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
4940 ret
= notifier_to_errno(ret
);
4944 ret
= netdev_register_kobject(dev
);
4947 dev
->reg_state
= NETREG_REGISTERED
;
4950 * Default initial state at registry is that the
4951 * device is present.
4954 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
4956 dev_init_scheduler(dev
);
4958 list_netdevice(dev
);
4960 /* Notify protocols, that a new device appeared. */
4961 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
4962 ret
= notifier_to_errno(ret
);
4964 rollback_registered(dev
);
4965 dev
->reg_state
= NETREG_UNREGISTERED
;
4968 * Prevent userspace races by waiting until the network
4969 * device is fully setup before sending notifications.
4971 if (!dev
->rtnl_link_ops
||
4972 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
4973 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
4979 if (dev
->netdev_ops
->ndo_uninit
)
4980 dev
->netdev_ops
->ndo_uninit(dev
);
4983 EXPORT_SYMBOL(register_netdevice
);
4986 * init_dummy_netdev - init a dummy network device for NAPI
4987 * @dev: device to init
4989 * This takes a network device structure and initialize the minimum
4990 * amount of fields so it can be used to schedule NAPI polls without
4991 * registering a full blown interface. This is to be used by drivers
4992 * that need to tie several hardware interfaces to a single NAPI
4993 * poll scheduler due to HW limitations.
4995 int init_dummy_netdev(struct net_device
*dev
)
4997 /* Clear everything. Note we don't initialize spinlocks
4998 * are they aren't supposed to be taken by any of the
4999 * NAPI code and this dummy netdev is supposed to be
5000 * only ever used for NAPI polls
5002 memset(dev
, 0, sizeof(struct net_device
));
5004 /* make sure we BUG if trying to hit standard
5005 * register/unregister code path
5007 dev
->reg_state
= NETREG_DUMMY
;
5009 /* initialize the ref count */
5010 atomic_set(&dev
->refcnt
, 1);
5012 /* NAPI wants this */
5013 INIT_LIST_HEAD(&dev
->napi_list
);
5015 /* a dummy interface is started by default */
5016 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5017 set_bit(__LINK_STATE_START
, &dev
->state
);
5021 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5025 * register_netdev - register a network device
5026 * @dev: device to register
5028 * Take a completed network device structure and add it to the kernel
5029 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5030 * chain. 0 is returned on success. A negative errno code is returned
5031 * on a failure to set up the device, or if the name is a duplicate.
5033 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5034 * and expands the device name if you passed a format string to
5037 int register_netdev(struct net_device
*dev
)
5044 * If the name is a format string the caller wants us to do a
5047 if (strchr(dev
->name
, '%')) {
5048 err
= dev_alloc_name(dev
, dev
->name
);
5053 err
= register_netdevice(dev
);
5058 EXPORT_SYMBOL(register_netdev
);
5061 * netdev_wait_allrefs - wait until all references are gone.
5063 * This is called when unregistering network devices.
5065 * Any protocol or device that holds a reference should register
5066 * for netdevice notification, and cleanup and put back the
5067 * reference if they receive an UNREGISTER event.
5068 * We can get stuck here if buggy protocols don't correctly
5071 static void netdev_wait_allrefs(struct net_device
*dev
)
5073 unsigned long rebroadcast_time
, warning_time
;
5075 linkwatch_forget_dev(dev
);
5077 rebroadcast_time
= warning_time
= jiffies
;
5078 while (atomic_read(&dev
->refcnt
) != 0) {
5079 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5082 /* Rebroadcast unregister notification */
5083 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5084 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5085 * should have already handle it the first time */
5087 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5089 /* We must not have linkwatch events
5090 * pending on unregister. If this
5091 * happens, we simply run the queue
5092 * unscheduled, resulting in a noop
5095 linkwatch_run_queue();
5100 rebroadcast_time
= jiffies
;
5105 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5106 printk(KERN_EMERG
"unregister_netdevice: "
5107 "waiting for %s to become free. Usage "
5109 dev
->name
, atomic_read(&dev
->refcnt
));
5110 warning_time
= jiffies
;
5119 * register_netdevice(x1);
5120 * register_netdevice(x2);
5122 * unregister_netdevice(y1);
5123 * unregister_netdevice(y2);
5129 * We are invoked by rtnl_unlock().
5130 * This allows us to deal with problems:
5131 * 1) We can delete sysfs objects which invoke hotplug
5132 * without deadlocking with linkwatch via keventd.
5133 * 2) Since we run with the RTNL semaphore not held, we can sleep
5134 * safely in order to wait for the netdev refcnt to drop to zero.
5136 * We must not return until all unregister events added during
5137 * the interval the lock was held have been completed.
5139 void netdev_run_todo(void)
5141 struct list_head list
;
5143 /* Snapshot list, allow later requests */
5144 list_replace_init(&net_todo_list
, &list
);
5148 while (!list_empty(&list
)) {
5149 struct net_device
*dev
5150 = list_first_entry(&list
, struct net_device
, todo_list
);
5151 list_del(&dev
->todo_list
);
5153 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5154 printk(KERN_ERR
"network todo '%s' but state %d\n",
5155 dev
->name
, dev
->reg_state
);
5160 dev
->reg_state
= NETREG_UNREGISTERED
;
5162 on_each_cpu(flush_backlog
, dev
, 1);
5164 netdev_wait_allrefs(dev
);
5167 BUG_ON(atomic_read(&dev
->refcnt
));
5168 WARN_ON(dev
->ip_ptr
);
5169 WARN_ON(dev
->ip6_ptr
);
5170 WARN_ON(dev
->dn_ptr
);
5172 if (dev
->destructor
)
5173 dev
->destructor(dev
);
5175 /* Free network device */
5176 kobject_put(&dev
->dev
.kobj
);
5181 * dev_txq_stats_fold - fold tx_queues stats
5182 * @dev: device to get statistics from
5183 * @stats: struct net_device_stats to hold results
5185 void dev_txq_stats_fold(const struct net_device
*dev
,
5186 struct net_device_stats
*stats
)
5188 unsigned long tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5190 struct netdev_queue
*txq
;
5192 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5193 txq
= netdev_get_tx_queue(dev
, i
);
5194 tx_bytes
+= txq
->tx_bytes
;
5195 tx_packets
+= txq
->tx_packets
;
5196 tx_dropped
+= txq
->tx_dropped
;
5198 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5199 stats
->tx_bytes
= tx_bytes
;
5200 stats
->tx_packets
= tx_packets
;
5201 stats
->tx_dropped
= tx_dropped
;
5204 EXPORT_SYMBOL(dev_txq_stats_fold
);
5207 * dev_get_stats - get network device statistics
5208 * @dev: device to get statistics from
5210 * Get network statistics from device. The device driver may provide
5211 * its own method by setting dev->netdev_ops->get_stats; otherwise
5212 * the internal statistics structure is used.
5214 const struct net_device_stats
*dev_get_stats(struct net_device
*dev
)
5216 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5218 if (ops
->ndo_get_stats
)
5219 return ops
->ndo_get_stats(dev
);
5221 dev_txq_stats_fold(dev
, &dev
->stats
);
5224 EXPORT_SYMBOL(dev_get_stats
);
5226 static void netdev_init_one_queue(struct net_device
*dev
,
5227 struct netdev_queue
*queue
,
5233 static void netdev_init_queues(struct net_device
*dev
)
5235 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
5236 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5237 spin_lock_init(&dev
->tx_global_lock
);
5241 * alloc_netdev_mq - allocate network device
5242 * @sizeof_priv: size of private data to allocate space for
5243 * @name: device name format string
5244 * @setup: callback to initialize device
5245 * @queue_count: the number of subqueues to allocate
5247 * Allocates a struct net_device with private data area for driver use
5248 * and performs basic initialization. Also allocates subquue structs
5249 * for each queue on the device at the end of the netdevice.
5251 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
5252 void (*setup
)(struct net_device
*), unsigned int queue_count
)
5254 struct netdev_queue
*tx
;
5255 struct net_device
*dev
;
5257 struct net_device
*p
;
5259 struct netdev_rx_queue
*rx
;
5263 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5265 alloc_size
= sizeof(struct net_device
);
5267 /* ensure 32-byte alignment of private area */
5268 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5269 alloc_size
+= sizeof_priv
;
5271 /* ensure 32-byte alignment of whole construct */
5272 alloc_size
+= NETDEV_ALIGN
- 1;
5274 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5276 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5280 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5282 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5288 rx
= kcalloc(queue_count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5290 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5295 atomic_set(&rx
->count
, queue_count
);
5298 * Set a pointer to first element in the array which holds the
5301 for (i
= 0; i
< queue_count
; i
++)
5305 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5306 dev
->padded
= (char *)dev
- (char *)p
;
5308 if (dev_addr_init(dev
))
5314 dev_net_set(dev
, &init_net
);
5317 dev
->num_tx_queues
= queue_count
;
5318 dev
->real_num_tx_queues
= queue_count
;
5322 dev
->num_rx_queues
= queue_count
;
5325 dev
->gso_max_size
= GSO_MAX_SIZE
;
5327 netdev_init_queues(dev
);
5329 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5330 dev
->ethtool_ntuple_list
.count
= 0;
5331 INIT_LIST_HEAD(&dev
->napi_list
);
5332 INIT_LIST_HEAD(&dev
->unreg_list
);
5333 INIT_LIST_HEAD(&dev
->link_watch_list
);
5334 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5336 strcpy(dev
->name
, name
);
5349 EXPORT_SYMBOL(alloc_netdev_mq
);
5352 * free_netdev - free network device
5355 * This function does the last stage of destroying an allocated device
5356 * interface. The reference to the device object is released.
5357 * If this is the last reference then it will be freed.
5359 void free_netdev(struct net_device
*dev
)
5361 struct napi_struct
*p
, *n
;
5363 release_net(dev_net(dev
));
5367 /* Flush device addresses */
5368 dev_addr_flush(dev
);
5370 /* Clear ethtool n-tuple list */
5371 ethtool_ntuple_flush(dev
);
5373 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5376 /* Compatibility with error handling in drivers */
5377 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5378 kfree((char *)dev
- dev
->padded
);
5382 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5383 dev
->reg_state
= NETREG_RELEASED
;
5385 /* will free via device release */
5386 put_device(&dev
->dev
);
5388 EXPORT_SYMBOL(free_netdev
);
5391 * synchronize_net - Synchronize with packet receive processing
5393 * Wait for packets currently being received to be done.
5394 * Does not block later packets from starting.
5396 void synchronize_net(void)
5401 EXPORT_SYMBOL(synchronize_net
);
5404 * unregister_netdevice_queue - remove device from the kernel
5408 * This function shuts down a device interface and removes it
5409 * from the kernel tables.
5410 * If head not NULL, device is queued to be unregistered later.
5412 * Callers must hold the rtnl semaphore. You may want
5413 * unregister_netdev() instead of this.
5416 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5421 list_move_tail(&dev
->unreg_list
, head
);
5423 rollback_registered(dev
);
5424 /* Finish processing unregister after unlock */
5428 EXPORT_SYMBOL(unregister_netdevice_queue
);
5431 * unregister_netdevice_many - unregister many devices
5432 * @head: list of devices
5434 void unregister_netdevice_many(struct list_head
*head
)
5436 struct net_device
*dev
;
5438 if (!list_empty(head
)) {
5439 rollback_registered_many(head
);
5440 list_for_each_entry(dev
, head
, unreg_list
)
5444 EXPORT_SYMBOL(unregister_netdevice_many
);
5447 * unregister_netdev - remove device from the kernel
5450 * This function shuts down a device interface and removes it
5451 * from the kernel tables.
5453 * This is just a wrapper for unregister_netdevice that takes
5454 * the rtnl semaphore. In general you want to use this and not
5455 * unregister_netdevice.
5457 void unregister_netdev(struct net_device
*dev
)
5460 unregister_netdevice(dev
);
5463 EXPORT_SYMBOL(unregister_netdev
);
5466 * dev_change_net_namespace - move device to different nethost namespace
5468 * @net: network namespace
5469 * @pat: If not NULL name pattern to try if the current device name
5470 * is already taken in the destination network namespace.
5472 * This function shuts down a device interface and moves it
5473 * to a new network namespace. On success 0 is returned, on
5474 * a failure a netagive errno code is returned.
5476 * Callers must hold the rtnl semaphore.
5479 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5485 /* Don't allow namespace local devices to be moved. */
5487 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5491 /* Don't allow real devices to be moved when sysfs
5495 if (dev
->dev
.parent
)
5499 /* Ensure the device has been registrered */
5501 if (dev
->reg_state
!= NETREG_REGISTERED
)
5504 /* Get out if there is nothing todo */
5506 if (net_eq(dev_net(dev
), net
))
5509 /* Pick the destination device name, and ensure
5510 * we can use it in the destination network namespace.
5513 if (__dev_get_by_name(net
, dev
->name
)) {
5514 /* We get here if we can't use the current device name */
5517 if (dev_get_valid_name(net
, pat
, dev
->name
, 1))
5522 * And now a mini version of register_netdevice unregister_netdevice.
5525 /* If device is running close it first. */
5528 /* And unlink it from device chain */
5530 unlist_netdevice(dev
);
5534 /* Shutdown queueing discipline. */
5537 /* Notify protocols, that we are about to destroy
5538 this device. They should clean all the things.
5540 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5541 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5544 * Flush the unicast and multicast chains
5549 netdev_unregister_kobject(dev
);
5551 /* Actually switch the network namespace */
5552 dev_net_set(dev
, net
);
5554 /* If there is an ifindex conflict assign a new one */
5555 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5556 int iflink
= (dev
->iflink
== dev
->ifindex
);
5557 dev
->ifindex
= dev_new_index(net
);
5559 dev
->iflink
= dev
->ifindex
;
5562 /* Fixup kobjects */
5563 err
= netdev_register_kobject(dev
);
5566 /* Add the device back in the hashes */
5567 list_netdevice(dev
);
5569 /* Notify protocols, that a new device appeared. */
5570 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5573 * Prevent userspace races by waiting until the network
5574 * device is fully setup before sending notifications.
5576 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5583 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5585 static int dev_cpu_callback(struct notifier_block
*nfb
,
5586 unsigned long action
,
5589 struct sk_buff
**list_skb
;
5590 struct Qdisc
**list_net
;
5591 struct sk_buff
*skb
;
5592 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5593 struct softnet_data
*sd
, *oldsd
;
5595 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5598 local_irq_disable();
5599 cpu
= smp_processor_id();
5600 sd
= &per_cpu(softnet_data
, cpu
);
5601 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5603 /* Find end of our completion_queue. */
5604 list_skb
= &sd
->completion_queue
;
5606 list_skb
= &(*list_skb
)->next
;
5607 /* Append completion queue from offline CPU. */
5608 *list_skb
= oldsd
->completion_queue
;
5609 oldsd
->completion_queue
= NULL
;
5611 /* Find end of our output_queue. */
5612 list_net
= &sd
->output_queue
;
5614 list_net
= &(*list_net
)->next_sched
;
5615 /* Append output queue from offline CPU. */
5616 *list_net
= oldsd
->output_queue
;
5617 oldsd
->output_queue
= NULL
;
5619 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5622 /* Process offline CPU's input_pkt_queue */
5623 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
5625 input_queue_head_incr(oldsd
);
5633 * netdev_increment_features - increment feature set by one
5634 * @all: current feature set
5635 * @one: new feature set
5636 * @mask: mask feature set
5638 * Computes a new feature set after adding a device with feature set
5639 * @one to the master device with current feature set @all. Will not
5640 * enable anything that is off in @mask. Returns the new feature set.
5642 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
5645 /* If device needs checksumming, downgrade to it. */
5646 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
5647 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
5648 else if (mask
& NETIF_F_ALL_CSUM
) {
5649 /* If one device supports v4/v6 checksumming, set for all. */
5650 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
5651 !(all
& NETIF_F_GEN_CSUM
)) {
5652 all
&= ~NETIF_F_ALL_CSUM
;
5653 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
5656 /* If one device supports hw checksumming, set for all. */
5657 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
5658 all
&= ~NETIF_F_ALL_CSUM
;
5659 all
|= NETIF_F_HW_CSUM
;
5663 one
|= NETIF_F_ALL_CSUM
;
5665 one
|= all
& NETIF_F_ONE_FOR_ALL
;
5666 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
5667 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
5671 EXPORT_SYMBOL(netdev_increment_features
);
5673 static struct hlist_head
*netdev_create_hash(void)
5676 struct hlist_head
*hash
;
5678 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5680 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5681 INIT_HLIST_HEAD(&hash
[i
]);
5686 /* Initialize per network namespace state */
5687 static int __net_init
netdev_init(struct net
*net
)
5689 INIT_LIST_HEAD(&net
->dev_base_head
);
5691 net
->dev_name_head
= netdev_create_hash();
5692 if (net
->dev_name_head
== NULL
)
5695 net
->dev_index_head
= netdev_create_hash();
5696 if (net
->dev_index_head
== NULL
)
5702 kfree(net
->dev_name_head
);
5708 * netdev_drivername - network driver for the device
5709 * @dev: network device
5710 * @buffer: buffer for resulting name
5711 * @len: size of buffer
5713 * Determine network driver for device.
5715 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
5717 const struct device_driver
*driver
;
5718 const struct device
*parent
;
5720 if (len
<= 0 || !buffer
)
5724 parent
= dev
->dev
.parent
;
5729 driver
= parent
->driver
;
5730 if (driver
&& driver
->name
)
5731 strlcpy(buffer
, driver
->name
, len
);
5735 static void __net_exit
netdev_exit(struct net
*net
)
5737 kfree(net
->dev_name_head
);
5738 kfree(net
->dev_index_head
);
5741 static struct pernet_operations __net_initdata netdev_net_ops
= {
5742 .init
= netdev_init
,
5743 .exit
= netdev_exit
,
5746 static void __net_exit
default_device_exit(struct net
*net
)
5748 struct net_device
*dev
, *aux
;
5750 * Push all migratable network devices back to the
5751 * initial network namespace
5754 for_each_netdev_safe(net
, dev
, aux
) {
5756 char fb_name
[IFNAMSIZ
];
5758 /* Ignore unmoveable devices (i.e. loopback) */
5759 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5762 /* Leave virtual devices for the generic cleanup */
5763 if (dev
->rtnl_link_ops
)
5766 /* Push remaing network devices to init_net */
5767 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
5768 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
5770 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
5771 __func__
, dev
->name
, err
);
5778 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
5780 /* At exit all network devices most be removed from a network
5781 * namespace. Do this in the reverse order of registeration.
5782 * Do this across as many network namespaces as possible to
5783 * improve batching efficiency.
5785 struct net_device
*dev
;
5787 LIST_HEAD(dev_kill_list
);
5790 list_for_each_entry(net
, net_list
, exit_list
) {
5791 for_each_netdev_reverse(net
, dev
) {
5792 if (dev
->rtnl_link_ops
)
5793 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
5795 unregister_netdevice_queue(dev
, &dev_kill_list
);
5798 unregister_netdevice_many(&dev_kill_list
);
5802 static struct pernet_operations __net_initdata default_device_ops
= {
5803 .exit
= default_device_exit
,
5804 .exit_batch
= default_device_exit_batch
,
5808 * Initialize the DEV module. At boot time this walks the device list and
5809 * unhooks any devices that fail to initialise (normally hardware not
5810 * present) and leaves us with a valid list of present and active devices.
5815 * This is called single threaded during boot, so no need
5816 * to take the rtnl semaphore.
5818 static int __init
net_dev_init(void)
5820 int i
, rc
= -ENOMEM
;
5822 BUG_ON(!dev_boot_phase
);
5824 if (dev_proc_init())
5827 if (netdev_kobject_init())
5830 INIT_LIST_HEAD(&ptype_all
);
5831 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
5832 INIT_LIST_HEAD(&ptype_base
[i
]);
5834 if (register_pernet_subsys(&netdev_net_ops
))
5838 * Initialise the packet receive queues.
5841 for_each_possible_cpu(i
) {
5842 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
5844 skb_queue_head_init(&sd
->input_pkt_queue
);
5845 sd
->completion_queue
= NULL
;
5846 INIT_LIST_HEAD(&sd
->poll_list
);
5849 sd
->csd
.func
= rps_trigger_softirq
;
5855 sd
->backlog
.poll
= process_backlog
;
5856 sd
->backlog
.weight
= weight_p
;
5857 sd
->backlog
.gro_list
= NULL
;
5858 sd
->backlog
.gro_count
= 0;
5863 /* The loopback device is special if any other network devices
5864 * is present in a network namespace the loopback device must
5865 * be present. Since we now dynamically allocate and free the
5866 * loopback device ensure this invariant is maintained by
5867 * keeping the loopback device as the first device on the
5868 * list of network devices. Ensuring the loopback devices
5869 * is the first device that appears and the last network device
5872 if (register_pernet_device(&loopback_net_ops
))
5875 if (register_pernet_device(&default_device_ops
))
5878 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
5879 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
5881 hotcpu_notifier(dev_cpu_callback
, 0);
5889 subsys_initcall(net_dev_init
);
5891 static int __init
initialize_hashrnd(void)
5893 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
5897 late_initcall_sync(initialize_hashrnd
);