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_rcu - 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. Must be called inside
813 * rcu_read_lock(), and result refcount is unchanged.
816 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
819 struct net_device
*dev
, *ret
;
822 for_each_netdev_rcu(net
, dev
) {
823 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
830 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
833 * dev_valid_name - check if name is okay for network device
836 * Network device names need to be valid file names to
837 * to allow sysfs to work. We also disallow any kind of
840 int dev_valid_name(const char *name
)
844 if (strlen(name
) >= IFNAMSIZ
)
846 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
850 if (*name
== '/' || isspace(*name
))
856 EXPORT_SYMBOL(dev_valid_name
);
859 * __dev_alloc_name - allocate a name for a device
860 * @net: network namespace to allocate the device name in
861 * @name: name format string
862 * @buf: scratch buffer and result name string
864 * Passed a format string - eg "lt%d" it will try and find a suitable
865 * id. It scans list of devices to build up a free map, then chooses
866 * the first empty slot. The caller must hold the dev_base or rtnl lock
867 * while allocating the name and adding the device in order to avoid
869 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
870 * Returns the number of the unit assigned or a negative errno code.
873 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
877 const int max_netdevices
= 8*PAGE_SIZE
;
878 unsigned long *inuse
;
879 struct net_device
*d
;
881 p
= strnchr(name
, IFNAMSIZ
-1, '%');
884 * Verify the string as this thing may have come from
885 * the user. There must be either one "%d" and no other "%"
888 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
891 /* Use one page as a bit array of possible slots */
892 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
896 for_each_netdev(net
, d
) {
897 if (!sscanf(d
->name
, name
, &i
))
899 if (i
< 0 || i
>= max_netdevices
)
902 /* avoid cases where sscanf is not exact inverse of printf */
903 snprintf(buf
, IFNAMSIZ
, name
, i
);
904 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
908 i
= find_first_zero_bit(inuse
, max_netdevices
);
909 free_page((unsigned long) inuse
);
913 snprintf(buf
, IFNAMSIZ
, name
, i
);
914 if (!__dev_get_by_name(net
, buf
))
917 /* It is possible to run out of possible slots
918 * when the name is long and there isn't enough space left
919 * for the digits, or if all bits are used.
925 * dev_alloc_name - allocate a name for a device
927 * @name: name format string
929 * Passed a format string - eg "lt%d" it will try and find a suitable
930 * id. It scans list of devices to build up a free map, then chooses
931 * the first empty slot. The caller must hold the dev_base or rtnl lock
932 * while allocating the name and adding the device in order to avoid
934 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
935 * Returns the number of the unit assigned or a negative errno code.
938 int dev_alloc_name(struct net_device
*dev
, const char *name
)
944 BUG_ON(!dev_net(dev
));
946 ret
= __dev_alloc_name(net
, name
, buf
);
948 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
951 EXPORT_SYMBOL(dev_alloc_name
);
953 static int dev_get_valid_name(struct net_device
*dev
, const char *name
, bool fmt
)
957 BUG_ON(!dev_net(dev
));
960 if (!dev_valid_name(name
))
963 if (fmt
&& strchr(name
, '%'))
964 return dev_alloc_name(dev
, name
);
965 else if (__dev_get_by_name(net
, name
))
967 else if (dev
->name
!= name
)
968 strlcpy(dev
->name
, 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(dev
, newname
, 1);
1005 ret
= device_rename(&dev
->dev
, dev
->name
);
1007 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1011 write_lock_bh(&dev_base_lock
);
1012 hlist_del(&dev
->name_hlist
);
1013 write_unlock_bh(&dev_base_lock
);
1017 write_lock_bh(&dev_base_lock
);
1018 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1019 write_unlock_bh(&dev_base_lock
);
1021 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1022 ret
= notifier_to_errno(ret
);
1025 /* err >= 0 after dev_alloc_name() or stores the first errno */
1028 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1032 "%s: name change rollback failed: %d.\n",
1041 * dev_set_alias - change ifalias of a device
1043 * @alias: name up to IFALIASZ
1044 * @len: limit of bytes to copy from info
1046 * Set ifalias for a device,
1048 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1052 if (len
>= IFALIASZ
)
1057 kfree(dev
->ifalias
);
1058 dev
->ifalias
= NULL
;
1063 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1067 strlcpy(dev
->ifalias
, alias
, len
+1);
1073 * netdev_features_change - device changes features
1074 * @dev: device to cause notification
1076 * Called to indicate a device has changed features.
1078 void netdev_features_change(struct net_device
*dev
)
1080 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1082 EXPORT_SYMBOL(netdev_features_change
);
1085 * netdev_state_change - device changes state
1086 * @dev: device to cause notification
1088 * Called to indicate a device has changed state. This function calls
1089 * the notifier chains for netdev_chain and sends a NEWLINK message
1090 * to the routing socket.
1092 void netdev_state_change(struct net_device
*dev
)
1094 if (dev
->flags
& IFF_UP
) {
1095 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1096 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1099 EXPORT_SYMBOL(netdev_state_change
);
1101 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1103 return call_netdevice_notifiers(event
, dev
);
1105 EXPORT_SYMBOL(netdev_bonding_change
);
1108 * dev_load - load a network module
1109 * @net: the applicable net namespace
1110 * @name: name of interface
1112 * If a network interface is not present and the process has suitable
1113 * privileges this function loads the module. If module loading is not
1114 * available in this kernel then it becomes a nop.
1117 void dev_load(struct net
*net
, const char *name
)
1119 struct net_device
*dev
;
1122 dev
= dev_get_by_name_rcu(net
, name
);
1125 if (!dev
&& capable(CAP_NET_ADMIN
))
1126 request_module("%s", name
);
1128 EXPORT_SYMBOL(dev_load
);
1130 static int __dev_open(struct net_device
*dev
)
1132 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1138 * Is it even present?
1140 if (!netif_device_present(dev
))
1143 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1144 ret
= notifier_to_errno(ret
);
1149 * Call device private open method
1151 set_bit(__LINK_STATE_START
, &dev
->state
);
1153 if (ops
->ndo_validate_addr
)
1154 ret
= ops
->ndo_validate_addr(dev
);
1156 if (!ret
&& ops
->ndo_open
)
1157 ret
= ops
->ndo_open(dev
);
1160 * If it went open OK then:
1164 clear_bit(__LINK_STATE_START
, &dev
->state
);
1169 dev
->flags
|= IFF_UP
;
1174 net_dmaengine_get();
1177 * Initialize multicasting status
1179 dev_set_rx_mode(dev
);
1182 * Wakeup transmit queue engine
1191 * dev_open - prepare an interface for use.
1192 * @dev: device to open
1194 * Takes a device from down to up state. The device's private open
1195 * function is invoked and then the multicast lists are loaded. Finally
1196 * the device is moved into the up state and a %NETDEV_UP message is
1197 * sent to the netdev notifier chain.
1199 * Calling this function on an active interface is a nop. On a failure
1200 * a negative errno code is returned.
1202 int dev_open(struct net_device
*dev
)
1209 if (dev
->flags
& IFF_UP
)
1215 ret
= __dev_open(dev
);
1220 * ... and announce new interface.
1222 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1223 call_netdevice_notifiers(NETDEV_UP
, dev
);
1227 EXPORT_SYMBOL(dev_open
);
1229 static int __dev_close(struct net_device
*dev
)
1231 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1237 * Tell people we are going down, so that they can
1238 * prepare to death, when device is still operating.
1240 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1242 clear_bit(__LINK_STATE_START
, &dev
->state
);
1244 /* Synchronize to scheduled poll. We cannot touch poll list,
1245 * it can be even on different cpu. So just clear netif_running().
1247 * dev->stop() will invoke napi_disable() on all of it's
1248 * napi_struct instances on this device.
1250 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1252 dev_deactivate(dev
);
1255 * Call the device specific close. This cannot fail.
1256 * Only if device is UP
1258 * We allow it to be called even after a DETACH hot-plug
1265 * Device is now down.
1268 dev
->flags
&= ~IFF_UP
;
1273 net_dmaengine_put();
1279 * dev_close - shutdown an interface.
1280 * @dev: device to shutdown
1282 * This function moves an active device into down state. A
1283 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1284 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1287 int dev_close(struct net_device
*dev
)
1289 if (!(dev
->flags
& IFF_UP
))
1295 * Tell people we are down
1297 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1298 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1302 EXPORT_SYMBOL(dev_close
);
1306 * dev_disable_lro - disable Large Receive Offload on a device
1309 * Disable Large Receive Offload (LRO) on a net device. Must be
1310 * called under RTNL. This is needed if received packets may be
1311 * forwarded to another interface.
1313 void dev_disable_lro(struct net_device
*dev
)
1315 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1316 dev
->ethtool_ops
->set_flags
) {
1317 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1318 if (flags
& ETH_FLAG_LRO
) {
1319 flags
&= ~ETH_FLAG_LRO
;
1320 dev
->ethtool_ops
->set_flags(dev
, flags
);
1323 WARN_ON(dev
->features
& NETIF_F_LRO
);
1325 EXPORT_SYMBOL(dev_disable_lro
);
1328 static int dev_boot_phase
= 1;
1331 * Device change register/unregister. These are not inline or static
1332 * as we export them to the world.
1336 * register_netdevice_notifier - register a network notifier block
1339 * Register a notifier to be called when network device events occur.
1340 * The notifier passed is linked into the kernel structures and must
1341 * not be reused until it has been unregistered. A negative errno code
1342 * is returned on a failure.
1344 * When registered all registration and up events are replayed
1345 * to the new notifier to allow device to have a race free
1346 * view of the network device list.
1349 int register_netdevice_notifier(struct notifier_block
*nb
)
1351 struct net_device
*dev
;
1352 struct net_device
*last
;
1357 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1363 for_each_netdev(net
, dev
) {
1364 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1365 err
= notifier_to_errno(err
);
1369 if (!(dev
->flags
& IFF_UP
))
1372 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1383 for_each_netdev(net
, dev
) {
1387 if (dev
->flags
& IFF_UP
) {
1388 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1389 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1391 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1392 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1396 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1399 EXPORT_SYMBOL(register_netdevice_notifier
);
1402 * unregister_netdevice_notifier - unregister a network notifier block
1405 * Unregister a notifier previously registered by
1406 * register_netdevice_notifier(). The notifier is unlinked into the
1407 * kernel structures and may then be reused. A negative errno code
1408 * is returned on a failure.
1411 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1416 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1420 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1423 * call_netdevice_notifiers - call all network notifier blocks
1424 * @val: value passed unmodified to notifier function
1425 * @dev: net_device pointer passed unmodified to notifier function
1427 * Call all network notifier blocks. Parameters and return value
1428 * are as for raw_notifier_call_chain().
1431 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1434 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1437 /* When > 0 there are consumers of rx skb time stamps */
1438 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1440 void net_enable_timestamp(void)
1442 atomic_inc(&netstamp_needed
);
1444 EXPORT_SYMBOL(net_enable_timestamp
);
1446 void net_disable_timestamp(void)
1448 atomic_dec(&netstamp_needed
);
1450 EXPORT_SYMBOL(net_disable_timestamp
);
1452 static inline void net_timestamp_set(struct sk_buff
*skb
)
1454 if (atomic_read(&netstamp_needed
))
1455 __net_timestamp(skb
);
1457 skb
->tstamp
.tv64
= 0;
1460 static inline void net_timestamp_check(struct sk_buff
*skb
)
1462 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1463 __net_timestamp(skb
);
1467 * dev_forward_skb - loopback an skb to another netif
1469 * @dev: destination network device
1470 * @skb: buffer to forward
1473 * NET_RX_SUCCESS (no congestion)
1474 * NET_RX_DROP (packet was dropped, but freed)
1476 * dev_forward_skb can be used for injecting an skb from the
1477 * start_xmit function of one device into the receive queue
1478 * of another device.
1480 * The receiving device may be in another namespace, so
1481 * we have to clear all information in the skb that could
1482 * impact namespace isolation.
1484 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1488 if (!(dev
->flags
& IFF_UP
) ||
1489 (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
)))
1512 net_timestamp_set(skb
);
1514 net_timestamp_set(skb
);
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
= NULL
;
1561 *sd
->output_queue_tailp
= q
;
1562 sd
->output_queue_tailp
= &q
->next_sched
;
1563 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1564 local_irq_restore(flags
);
1567 void __netif_schedule(struct Qdisc
*q
)
1569 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1570 __netif_reschedule(q
);
1572 EXPORT_SYMBOL(__netif_schedule
);
1574 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1576 if (!skb
->destructor
)
1578 else if (atomic_dec_and_test(&skb
->users
)) {
1579 struct softnet_data
*sd
;
1580 unsigned long flags
;
1582 local_irq_save(flags
);
1583 sd
= &__get_cpu_var(softnet_data
);
1584 skb
->next
= sd
->completion_queue
;
1585 sd
->completion_queue
= skb
;
1586 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1587 local_irq_restore(flags
);
1590 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1592 void dev_kfree_skb_any(struct sk_buff
*skb
)
1594 if (in_irq() || irqs_disabled())
1595 dev_kfree_skb_irq(skb
);
1599 EXPORT_SYMBOL(dev_kfree_skb_any
);
1603 * netif_device_detach - mark device as removed
1604 * @dev: network device
1606 * Mark device as removed from system and therefore no longer available.
1608 void netif_device_detach(struct net_device
*dev
)
1610 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1611 netif_running(dev
)) {
1612 netif_tx_stop_all_queues(dev
);
1615 EXPORT_SYMBOL(netif_device_detach
);
1618 * netif_device_attach - mark device as attached
1619 * @dev: network device
1621 * Mark device as attached from system and restart if needed.
1623 void netif_device_attach(struct net_device
*dev
)
1625 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1626 netif_running(dev
)) {
1627 netif_tx_wake_all_queues(dev
);
1628 __netdev_watchdog_up(dev
);
1631 EXPORT_SYMBOL(netif_device_attach
);
1633 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1635 return ((features
& NETIF_F_GEN_CSUM
) ||
1636 ((features
& NETIF_F_IP_CSUM
) &&
1637 protocol
== htons(ETH_P_IP
)) ||
1638 ((features
& NETIF_F_IPV6_CSUM
) &&
1639 protocol
== htons(ETH_P_IPV6
)) ||
1640 ((features
& NETIF_F_FCOE_CRC
) &&
1641 protocol
== htons(ETH_P_FCOE
)));
1644 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1646 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1649 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1650 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1651 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1652 veh
->h_vlan_encapsulated_proto
))
1660 * skb_dev_set -- assign a new device to a buffer
1661 * @skb: buffer for the new device
1662 * @dev: network device
1664 * If an skb is owned by a device already, we have to reset
1665 * all data private to the namespace a device belongs to
1666 * before assigning it a new device.
1668 #ifdef CONFIG_NET_NS
1669 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1672 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1675 skb_init_secmark(skb
);
1679 skb
->ipvs_property
= 0;
1680 #ifdef CONFIG_NET_SCHED
1686 EXPORT_SYMBOL(skb_set_dev
);
1687 #endif /* CONFIG_NET_NS */
1690 * Invalidate hardware checksum when packet is to be mangled, and
1691 * complete checksum manually on outgoing path.
1693 int skb_checksum_help(struct sk_buff
*skb
)
1696 int ret
= 0, offset
;
1698 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1699 goto out_set_summed
;
1701 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1702 /* Let GSO fix up the checksum. */
1703 goto out_set_summed
;
1706 offset
= skb
->csum_start
- skb_headroom(skb
);
1707 BUG_ON(offset
>= skb_headlen(skb
));
1708 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1710 offset
+= skb
->csum_offset
;
1711 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1713 if (skb_cloned(skb
) &&
1714 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1715 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1720 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1722 skb
->ip_summed
= CHECKSUM_NONE
;
1726 EXPORT_SYMBOL(skb_checksum_help
);
1729 * skb_gso_segment - Perform segmentation on skb.
1730 * @skb: buffer to segment
1731 * @features: features for the output path (see dev->features)
1733 * This function segments the given skb and returns a list of segments.
1735 * It may return NULL if the skb requires no segmentation. This is
1736 * only possible when GSO is used for verifying header integrity.
1738 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1740 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1741 struct packet_type
*ptype
;
1742 __be16 type
= skb
->protocol
;
1745 skb_reset_mac_header(skb
);
1746 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1747 __skb_pull(skb
, skb
->mac_len
);
1749 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1750 struct net_device
*dev
= skb
->dev
;
1751 struct ethtool_drvinfo info
= {};
1753 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1754 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1756 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1758 info
.driver
, dev
? dev
->features
: 0L,
1759 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1760 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1762 if (skb_header_cloned(skb
) &&
1763 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1764 return ERR_PTR(err
);
1768 list_for_each_entry_rcu(ptype
,
1769 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1770 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1771 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1772 err
= ptype
->gso_send_check(skb
);
1773 segs
= ERR_PTR(err
);
1774 if (err
|| skb_gso_ok(skb
, features
))
1776 __skb_push(skb
, (skb
->data
-
1777 skb_network_header(skb
)));
1779 segs
= ptype
->gso_segment(skb
, features
);
1785 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1789 EXPORT_SYMBOL(skb_gso_segment
);
1791 /* Take action when hardware reception checksum errors are detected. */
1793 void netdev_rx_csum_fault(struct net_device
*dev
)
1795 if (net_ratelimit()) {
1796 printk(KERN_ERR
"%s: hw csum failure.\n",
1797 dev
? dev
->name
: "<unknown>");
1801 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1804 /* Actually, we should eliminate this check as soon as we know, that:
1805 * 1. IOMMU is present and allows to map all the memory.
1806 * 2. No high memory really exists on this machine.
1809 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1811 #ifdef CONFIG_HIGHMEM
1813 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1814 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1815 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1819 if (PCI_DMA_BUS_IS_PHYS
) {
1820 struct device
*pdev
= dev
->dev
.parent
;
1824 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1825 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1826 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1835 void (*destructor
)(struct sk_buff
*skb
);
1838 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1840 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1842 struct dev_gso_cb
*cb
;
1845 struct sk_buff
*nskb
= skb
->next
;
1847 skb
->next
= nskb
->next
;
1850 } while (skb
->next
);
1852 cb
= DEV_GSO_CB(skb
);
1854 cb
->destructor(skb
);
1858 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1859 * @skb: buffer to segment
1861 * This function segments the given skb and stores the list of segments
1864 static int dev_gso_segment(struct sk_buff
*skb
)
1866 struct net_device
*dev
= skb
->dev
;
1867 struct sk_buff
*segs
;
1868 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1871 segs
= skb_gso_segment(skb
, features
);
1873 /* Verifying header integrity only. */
1878 return PTR_ERR(segs
);
1881 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1882 skb
->destructor
= dev_gso_skb_destructor
;
1888 * Try to orphan skb early, right before transmission by the device.
1889 * We cannot orphan skb if tx timestamp is requested, since
1890 * drivers need to call skb_tstamp_tx() to send the timestamp.
1892 static inline void skb_orphan_try(struct sk_buff
*skb
)
1894 if (!skb_tx(skb
)->flags
)
1898 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1899 struct netdev_queue
*txq
)
1901 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1902 int rc
= NETDEV_TX_OK
;
1904 if (likely(!skb
->next
)) {
1905 if (!list_empty(&ptype_all
))
1906 dev_queue_xmit_nit(skb
, dev
);
1909 * If device doesnt need skb->dst, release it right now while
1910 * its hot in this cpu cache
1912 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1915 skb_orphan_try(skb
);
1917 if (netif_needs_gso(dev
, skb
)) {
1918 if (unlikely(dev_gso_segment(skb
)))
1924 rc
= ops
->ndo_start_xmit(skb
, dev
);
1925 if (rc
== NETDEV_TX_OK
)
1926 txq_trans_update(txq
);
1932 struct sk_buff
*nskb
= skb
->next
;
1934 skb
->next
= nskb
->next
;
1938 * If device doesnt need nskb->dst, release it right now while
1939 * its hot in this cpu cache
1941 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1944 rc
= ops
->ndo_start_xmit(nskb
, dev
);
1945 if (unlikely(rc
!= NETDEV_TX_OK
)) {
1946 if (rc
& ~NETDEV_TX_MASK
)
1947 goto out_kfree_gso_skb
;
1948 nskb
->next
= skb
->next
;
1952 txq_trans_update(txq
);
1953 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
1954 return NETDEV_TX_BUSY
;
1955 } while (skb
->next
);
1958 if (likely(skb
->next
== NULL
))
1959 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1965 static u32 hashrnd __read_mostly
;
1967 u16
skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
)
1971 if (skb_rx_queue_recorded(skb
)) {
1972 hash
= skb_get_rx_queue(skb
);
1973 while (unlikely(hash
>= dev
->real_num_tx_queues
))
1974 hash
-= dev
->real_num_tx_queues
;
1978 if (skb
->sk
&& skb
->sk
->sk_hash
)
1979 hash
= skb
->sk
->sk_hash
;
1981 hash
= (__force u16
) skb
->protocol
;
1983 hash
= jhash_1word(hash
, hashrnd
);
1985 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
1987 EXPORT_SYMBOL(skb_tx_hash
);
1989 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
1991 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
1992 if (net_ratelimit()) {
1993 pr_warning("%s selects TX queue %d, but "
1994 "real number of TX queues is %d\n",
1995 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2002 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2003 struct sk_buff
*skb
)
2006 struct sock
*sk
= skb
->sk
;
2008 if (sk_tx_queue_recorded(sk
)) {
2009 queue_index
= sk_tx_queue_get(sk
);
2011 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2013 if (ops
->ndo_select_queue
) {
2014 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2015 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2018 if (dev
->real_num_tx_queues
> 1)
2019 queue_index
= skb_tx_hash(dev
, skb
);
2022 struct dst_entry
*dst
= rcu_dereference_check(sk
->sk_dst_cache
, 1);
2024 if (dst
&& skb_dst(skb
) == dst
)
2025 sk_tx_queue_set(sk
, queue_index
);
2030 skb_set_queue_mapping(skb
, queue_index
);
2031 return netdev_get_tx_queue(dev
, queue_index
);
2034 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2035 struct net_device
*dev
,
2036 struct netdev_queue
*txq
)
2038 spinlock_t
*root_lock
= qdisc_lock(q
);
2039 bool contended
= qdisc_is_running(q
);
2043 * Heuristic to force contended enqueues to serialize on a
2044 * separate lock before trying to get qdisc main lock.
2045 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2046 * and dequeue packets faster.
2048 if (unlikely(contended
))
2049 spin_lock(&q
->busylock
);
2051 spin_lock(root_lock
);
2052 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2055 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2056 qdisc_run_begin(q
)) {
2058 * This is a work-conserving queue; there are no old skbs
2059 * waiting to be sent out; and the qdisc is not running -
2060 * xmit the skb directly.
2062 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2064 __qdisc_update_bstats(q
, skb
->len
);
2065 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2066 if (unlikely(contended
)) {
2067 spin_unlock(&q
->busylock
);
2074 rc
= NET_XMIT_SUCCESS
;
2077 rc
= qdisc_enqueue_root(skb
, q
);
2078 if (qdisc_run_begin(q
)) {
2079 if (unlikely(contended
)) {
2080 spin_unlock(&q
->busylock
);
2086 spin_unlock(root_lock
);
2087 if (unlikely(contended
))
2088 spin_unlock(&q
->busylock
);
2093 * Returns true if either:
2094 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2095 * 2. skb is fragmented and the device does not support SG, or if
2096 * at least one of fragments is in highmem and device does not
2097 * support DMA from it.
2099 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2100 struct net_device
*dev
)
2102 return skb_is_nonlinear(skb
) &&
2103 ((skb_has_frags(skb
) && !(dev
->features
& NETIF_F_FRAGLIST
)) ||
2104 (skb_shinfo(skb
)->nr_frags
&& (!(dev
->features
& NETIF_F_SG
) ||
2105 illegal_highdma(dev
, skb
))));
2109 * dev_queue_xmit - transmit a buffer
2110 * @skb: buffer to transmit
2112 * Queue a buffer for transmission to a network device. The caller must
2113 * have set the device and priority and built the buffer before calling
2114 * this function. The function can be called from an interrupt.
2116 * A negative errno code is returned on a failure. A success does not
2117 * guarantee the frame will be transmitted as it may be dropped due
2118 * to congestion or traffic shaping.
2120 * -----------------------------------------------------------------------------------
2121 * I notice this method can also return errors from the queue disciplines,
2122 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2125 * Regardless of the return value, the skb is consumed, so it is currently
2126 * difficult to retry a send to this method. (You can bump the ref count
2127 * before sending to hold a reference for retry if you are careful.)
2129 * When calling this method, interrupts MUST be enabled. This is because
2130 * the BH enable code must have IRQs enabled so that it will not deadlock.
2133 int dev_queue_xmit(struct sk_buff
*skb
)
2135 struct net_device
*dev
= skb
->dev
;
2136 struct netdev_queue
*txq
;
2140 /* GSO will handle the following emulations directly. */
2141 if (netif_needs_gso(dev
, skb
))
2144 /* Convert a paged skb to linear, if required */
2145 if (skb_needs_linearize(skb
, dev
) && __skb_linearize(skb
))
2148 /* If packet is not checksummed and device does not support
2149 * checksumming for this protocol, complete checksumming here.
2151 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2152 skb_set_transport_header(skb
, skb
->csum_start
-
2154 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
2159 /* Disable soft irqs for various locks below. Also
2160 * stops preemption for RCU.
2164 txq
= dev_pick_tx(dev
, skb
);
2165 q
= rcu_dereference_bh(txq
->qdisc
);
2167 #ifdef CONFIG_NET_CLS_ACT
2168 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2171 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2175 /* The device has no queue. Common case for software devices:
2176 loopback, all the sorts of tunnels...
2178 Really, it is unlikely that netif_tx_lock protection is necessary
2179 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2181 However, it is possible, that they rely on protection
2184 Check this and shot the lock. It is not prone from deadlocks.
2185 Either shot noqueue qdisc, it is even simpler 8)
2187 if (dev
->flags
& IFF_UP
) {
2188 int cpu
= smp_processor_id(); /* ok because BHs are off */
2190 if (txq
->xmit_lock_owner
!= cpu
) {
2192 HARD_TX_LOCK(dev
, txq
, cpu
);
2194 if (!netif_tx_queue_stopped(txq
)) {
2195 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2196 if (dev_xmit_complete(rc
)) {
2197 HARD_TX_UNLOCK(dev
, txq
);
2201 HARD_TX_UNLOCK(dev
, txq
);
2202 if (net_ratelimit())
2203 printk(KERN_CRIT
"Virtual device %s asks to "
2204 "queue packet!\n", dev
->name
);
2206 /* Recursion is detected! It is possible,
2208 if (net_ratelimit())
2209 printk(KERN_CRIT
"Dead loop on virtual device "
2210 "%s, fix it urgently!\n", dev
->name
);
2215 rcu_read_unlock_bh();
2221 rcu_read_unlock_bh();
2224 EXPORT_SYMBOL(dev_queue_xmit
);
2227 /*=======================================================================
2229 =======================================================================*/
2231 int netdev_max_backlog __read_mostly
= 1000;
2232 int netdev_tstamp_prequeue __read_mostly
= 1;
2233 int netdev_budget __read_mostly
= 300;
2234 int weight_p __read_mostly
= 64; /* old backlog weight */
2236 /* Called with irq disabled */
2237 static inline void ____napi_schedule(struct softnet_data
*sd
,
2238 struct napi_struct
*napi
)
2240 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2241 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2246 /* One global table that all flow-based protocols share. */
2247 struct rps_sock_flow_table
*rps_sock_flow_table __read_mostly
;
2248 EXPORT_SYMBOL(rps_sock_flow_table
);
2251 * get_rps_cpu is called from netif_receive_skb and returns the target
2252 * CPU from the RPS map of the receiving queue for a given skb.
2253 * rcu_read_lock must be held on entry.
2255 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2256 struct rps_dev_flow
**rflowp
)
2258 struct ipv6hdr
*ip6
;
2260 struct netdev_rx_queue
*rxqueue
;
2261 struct rps_map
*map
;
2262 struct rps_dev_flow_table
*flow_table
;
2263 struct rps_sock_flow_table
*sock_flow_table
;
2267 u32 addr1
, addr2
, ihl
;
2273 if (skb_rx_queue_recorded(skb
)) {
2274 u16 index
= skb_get_rx_queue(skb
);
2275 if (unlikely(index
>= dev
->num_rx_queues
)) {
2276 WARN_ONCE(dev
->num_rx_queues
> 1, "%s received packet "
2277 "on queue %u, but number of RX queues is %u\n",
2278 dev
->name
, index
, dev
->num_rx_queues
);
2281 rxqueue
= dev
->_rx
+ index
;
2285 if (!rxqueue
->rps_map
&& !rxqueue
->rps_flow_table
)
2289 goto got_hash
; /* Skip hash computation on packet header */
2291 switch (skb
->protocol
) {
2292 case __constant_htons(ETH_P_IP
):
2293 if (!pskb_may_pull(skb
, sizeof(*ip
)))
2296 ip
= (struct iphdr
*) skb
->data
;
2297 ip_proto
= ip
->protocol
;
2298 addr1
= (__force u32
) ip
->saddr
;
2299 addr2
= (__force u32
) ip
->daddr
;
2302 case __constant_htons(ETH_P_IPV6
):
2303 if (!pskb_may_pull(skb
, sizeof(*ip6
)))
2306 ip6
= (struct ipv6hdr
*) skb
->data
;
2307 ip_proto
= ip6
->nexthdr
;
2308 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2309 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2322 case IPPROTO_UDPLITE
:
2323 if (pskb_may_pull(skb
, (ihl
* 4) + 4)) {
2324 ports
.v32
= * (__force u32
*) (skb
->data
+ (ihl
* 4));
2325 if (ports
.v16
[1] < ports
.v16
[0])
2326 swap(ports
.v16
[0], ports
.v16
[1]);
2334 /* get a consistent hash (same value on both flow directions) */
2337 skb
->rxhash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2342 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2343 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2344 if (flow_table
&& sock_flow_table
) {
2346 struct rps_dev_flow
*rflow
;
2348 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2351 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2352 sock_flow_table
->mask
];
2355 * If the desired CPU (where last recvmsg was done) is
2356 * different from current CPU (one in the rx-queue flow
2357 * table entry), switch if one of the following holds:
2358 * - Current CPU is unset (equal to RPS_NO_CPU).
2359 * - Current CPU is offline.
2360 * - The current CPU's queue tail has advanced beyond the
2361 * last packet that was enqueued using this table entry.
2362 * This guarantees that all previous packets for the flow
2363 * have been dequeued, thus preserving in order delivery.
2365 if (unlikely(tcpu
!= next_cpu
) &&
2366 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2367 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2368 rflow
->last_qtail
)) >= 0)) {
2369 tcpu
= rflow
->cpu
= next_cpu
;
2370 if (tcpu
!= RPS_NO_CPU
)
2371 rflow
->last_qtail
= per_cpu(softnet_data
,
2372 tcpu
).input_queue_head
;
2374 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2381 map
= rcu_dereference(rxqueue
->rps_map
);
2383 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2385 if (cpu_online(tcpu
)) {
2395 /* Called from hardirq (IPI) context */
2396 static void rps_trigger_softirq(void *data
)
2398 struct softnet_data
*sd
= data
;
2400 ____napi_schedule(sd
, &sd
->backlog
);
2404 #endif /* CONFIG_RPS */
2407 * Check if this softnet_data structure is another cpu one
2408 * If yes, queue it to our IPI list and return 1
2411 static int rps_ipi_queued(struct softnet_data
*sd
)
2414 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2417 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2418 mysd
->rps_ipi_list
= sd
;
2420 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2423 #endif /* CONFIG_RPS */
2428 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2429 * queue (may be a remote CPU queue).
2431 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2432 unsigned int *qtail
)
2434 struct softnet_data
*sd
;
2435 unsigned long flags
;
2437 sd
= &per_cpu(softnet_data
, cpu
);
2439 local_irq_save(flags
);
2442 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2443 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2445 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2446 input_queue_tail_incr_save(sd
, qtail
);
2448 local_irq_restore(flags
);
2449 return NET_RX_SUCCESS
;
2452 /* Schedule NAPI for backlog device
2453 * We can use non atomic operation since we own the queue lock
2455 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2456 if (!rps_ipi_queued(sd
))
2457 ____napi_schedule(sd
, &sd
->backlog
);
2465 local_irq_restore(flags
);
2472 * netif_rx - post buffer to the network code
2473 * @skb: buffer to post
2475 * This function receives a packet from a device driver and queues it for
2476 * the upper (protocol) levels to process. It always succeeds. The buffer
2477 * may be dropped during processing for congestion control or by the
2481 * NET_RX_SUCCESS (no congestion)
2482 * NET_RX_DROP (packet was dropped)
2486 int netif_rx(struct sk_buff
*skb
)
2490 /* if netpoll wants it, pretend we never saw it */
2491 if (netpoll_rx(skb
))
2494 if (netdev_tstamp_prequeue
)
2495 net_timestamp_check(skb
);
2499 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2504 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2506 cpu
= smp_processor_id();
2508 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2515 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2521 EXPORT_SYMBOL(netif_rx
);
2523 int netif_rx_ni(struct sk_buff
*skb
)
2528 err
= netif_rx(skb
);
2529 if (local_softirq_pending())
2535 EXPORT_SYMBOL(netif_rx_ni
);
2537 static void net_tx_action(struct softirq_action
*h
)
2539 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2541 if (sd
->completion_queue
) {
2542 struct sk_buff
*clist
;
2544 local_irq_disable();
2545 clist
= sd
->completion_queue
;
2546 sd
->completion_queue
= NULL
;
2550 struct sk_buff
*skb
= clist
;
2551 clist
= clist
->next
;
2553 WARN_ON(atomic_read(&skb
->users
));
2558 if (sd
->output_queue
) {
2561 local_irq_disable();
2562 head
= sd
->output_queue
;
2563 sd
->output_queue
= NULL
;
2564 sd
->output_queue_tailp
= &sd
->output_queue
;
2568 struct Qdisc
*q
= head
;
2569 spinlock_t
*root_lock
;
2571 head
= head
->next_sched
;
2573 root_lock
= qdisc_lock(q
);
2574 if (spin_trylock(root_lock
)) {
2575 smp_mb__before_clear_bit();
2576 clear_bit(__QDISC_STATE_SCHED
,
2579 spin_unlock(root_lock
);
2581 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2583 __netif_reschedule(q
);
2585 smp_mb__before_clear_bit();
2586 clear_bit(__QDISC_STATE_SCHED
,
2594 static inline int deliver_skb(struct sk_buff
*skb
,
2595 struct packet_type
*pt_prev
,
2596 struct net_device
*orig_dev
)
2598 atomic_inc(&skb
->users
);
2599 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2602 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2603 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2604 /* This hook is defined here for ATM LANE */
2605 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2606 unsigned char *addr
) __read_mostly
;
2607 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2610 #ifdef CONFIG_NET_CLS_ACT
2611 /* TODO: Maybe we should just force sch_ingress to be compiled in
2612 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2613 * a compare and 2 stores extra right now if we dont have it on
2614 * but have CONFIG_NET_CLS_ACT
2615 * NOTE: This doesnt stop any functionality; if you dont have
2616 * the ingress scheduler, you just cant add policies on ingress.
2619 static int ing_filter(struct sk_buff
*skb
)
2621 struct net_device
*dev
= skb
->dev
;
2622 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2623 struct netdev_queue
*rxq
;
2624 int result
= TC_ACT_OK
;
2627 if (MAX_RED_LOOP
< ttl
++) {
2629 "Redir loop detected Dropping packet (%d->%d)\n",
2630 skb
->skb_iif
, dev
->ifindex
);
2634 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2635 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2637 rxq
= &dev
->rx_queue
;
2640 if (q
!= &noop_qdisc
) {
2641 spin_lock(qdisc_lock(q
));
2642 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2643 result
= qdisc_enqueue_root(skb
, q
);
2644 spin_unlock(qdisc_lock(q
));
2650 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2651 struct packet_type
**pt_prev
,
2652 int *ret
, struct net_device
*orig_dev
)
2654 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2658 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2662 switch (ing_filter(skb
)) {
2676 * netif_nit_deliver - deliver received packets to network taps
2679 * This function is used to deliver incoming packets to network
2680 * taps. It should be used when the normal netif_receive_skb path
2681 * is bypassed, for example because of VLAN acceleration.
2683 void netif_nit_deliver(struct sk_buff
*skb
)
2685 struct packet_type
*ptype
;
2687 if (list_empty(&ptype_all
))
2690 skb_reset_network_header(skb
);
2691 skb_reset_transport_header(skb
);
2692 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2695 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2696 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2697 deliver_skb(skb
, ptype
, skb
->dev
);
2703 * netdev_rx_handler_register - register receive handler
2704 * @dev: device to register a handler for
2705 * @rx_handler: receive handler to register
2706 * @rx_handler_data: data pointer that is used by rx handler
2708 * Register a receive hander for a device. This handler will then be
2709 * called from __netif_receive_skb. A negative errno code is returned
2712 * The caller must hold the rtnl_mutex.
2714 int netdev_rx_handler_register(struct net_device
*dev
,
2715 rx_handler_func_t
*rx_handler
,
2716 void *rx_handler_data
)
2720 if (dev
->rx_handler
)
2723 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
2724 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
2728 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
2731 * netdev_rx_handler_unregister - unregister receive handler
2732 * @dev: device to unregister a handler from
2734 * Unregister a receive hander from a device.
2736 * The caller must hold the rtnl_mutex.
2738 void netdev_rx_handler_unregister(struct net_device
*dev
)
2742 rcu_assign_pointer(dev
->rx_handler
, NULL
);
2743 rcu_assign_pointer(dev
->rx_handler_data
, NULL
);
2745 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
2747 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2748 struct net_device
*master
)
2750 if (skb
->pkt_type
== PACKET_HOST
) {
2751 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2753 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2757 /* On bonding slaves other than the currently active slave, suppress
2758 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2759 * ARP on active-backup slaves with arp_validate enabled.
2761 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2763 struct net_device
*dev
= skb
->dev
;
2765 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2766 dev
->last_rx
= jiffies
;
2768 if ((master
->priv_flags
& IFF_MASTER_ALB
) &&
2769 (master
->priv_flags
& IFF_BRIDGE_PORT
)) {
2770 /* Do address unmangle. The local destination address
2771 * will be always the one master has. Provides the right
2772 * functionality in a bridge.
2774 skb_bond_set_mac_by_master(skb
, master
);
2777 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
2778 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
2779 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
2782 if (master
->priv_flags
& IFF_MASTER_ALB
) {
2783 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
2784 skb
->pkt_type
!= PACKET_MULTICAST
)
2787 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
2788 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
2795 EXPORT_SYMBOL(__skb_bond_should_drop
);
2797 static int __netif_receive_skb(struct sk_buff
*skb
)
2799 struct packet_type
*ptype
, *pt_prev
;
2800 rx_handler_func_t
*rx_handler
;
2801 struct net_device
*orig_dev
;
2802 struct net_device
*master
;
2803 struct net_device
*null_or_orig
;
2804 struct net_device
*orig_or_bond
;
2805 int ret
= NET_RX_DROP
;
2808 if (!netdev_tstamp_prequeue
)
2809 net_timestamp_check(skb
);
2811 if (vlan_tx_tag_present(skb
) && vlan_hwaccel_do_receive(skb
))
2812 return NET_RX_SUCCESS
;
2814 /* if we've gotten here through NAPI, check netpoll */
2815 if (netpoll_receive_skb(skb
))
2819 skb
->skb_iif
= skb
->dev
->ifindex
;
2822 * bonding note: skbs received on inactive slaves should only
2823 * be delivered to pkt handlers that are exact matches. Also
2824 * the deliver_no_wcard flag will be set. If packet handlers
2825 * are sensitive to duplicate packets these skbs will need to
2826 * be dropped at the handler. The vlan accel path may have
2827 * already set the deliver_no_wcard flag.
2829 null_or_orig
= NULL
;
2830 orig_dev
= skb
->dev
;
2831 master
= ACCESS_ONCE(orig_dev
->master
);
2832 if (skb
->deliver_no_wcard
)
2833 null_or_orig
= orig_dev
;
2835 if (skb_bond_should_drop(skb
, master
)) {
2836 skb
->deliver_no_wcard
= 1;
2837 null_or_orig
= orig_dev
; /* deliver only exact match */
2842 __this_cpu_inc(softnet_data
.processed
);
2843 skb_reset_network_header(skb
);
2844 skb_reset_transport_header(skb
);
2845 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2851 #ifdef CONFIG_NET_CLS_ACT
2852 if (skb
->tc_verd
& TC_NCLS
) {
2853 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2858 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2859 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2860 ptype
->dev
== orig_dev
) {
2862 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2867 #ifdef CONFIG_NET_CLS_ACT
2868 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2874 /* Handle special case of bridge or macvlan */
2875 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
2878 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2881 skb
= rx_handler(skb
);
2887 * Make sure frames received on VLAN interfaces stacked on
2888 * bonding interfaces still make their way to any base bonding
2889 * device that may have registered for a specific ptype. The
2890 * handler may have to adjust skb->dev and orig_dev.
2892 orig_or_bond
= orig_dev
;
2893 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
2894 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
2895 orig_or_bond
= vlan_dev_real_dev(skb
->dev
);
2898 type
= skb
->protocol
;
2899 list_for_each_entry_rcu(ptype
,
2900 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2901 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
2902 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
2903 ptype
->dev
== orig_or_bond
)) {
2905 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2911 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2914 /* Jamal, now you will not able to escape explaining
2915 * me how you were going to use this. :-)
2926 * netif_receive_skb - process receive buffer from network
2927 * @skb: buffer to process
2929 * netif_receive_skb() is the main receive data processing function.
2930 * It always succeeds. The buffer may be dropped during processing
2931 * for congestion control or by the protocol layers.
2933 * This function may only be called from softirq context and interrupts
2934 * should be enabled.
2936 * Return values (usually ignored):
2937 * NET_RX_SUCCESS: no congestion
2938 * NET_RX_DROP: packet was dropped
2940 int netif_receive_skb(struct sk_buff
*skb
)
2942 if (netdev_tstamp_prequeue
)
2943 net_timestamp_check(skb
);
2947 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2952 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2955 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2959 ret
= __netif_receive_skb(skb
);
2965 return __netif_receive_skb(skb
);
2968 EXPORT_SYMBOL(netif_receive_skb
);
2970 /* Network device is going away, flush any packets still pending
2971 * Called with irqs disabled.
2973 static void flush_backlog(void *arg
)
2975 struct net_device
*dev
= arg
;
2976 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2977 struct sk_buff
*skb
, *tmp
;
2980 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
2981 if (skb
->dev
== dev
) {
2982 __skb_unlink(skb
, &sd
->input_pkt_queue
);
2984 input_queue_head_incr(sd
);
2989 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
2990 if (skb
->dev
== dev
) {
2991 __skb_unlink(skb
, &sd
->process_queue
);
2993 input_queue_head_incr(sd
);
2998 static int napi_gro_complete(struct sk_buff
*skb
)
3000 struct packet_type
*ptype
;
3001 __be16 type
= skb
->protocol
;
3002 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3005 if (NAPI_GRO_CB(skb
)->count
== 1) {
3006 skb_shinfo(skb
)->gso_size
= 0;
3011 list_for_each_entry_rcu(ptype
, head
, list
) {
3012 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3015 err
= ptype
->gro_complete(skb
);
3021 WARN_ON(&ptype
->list
== head
);
3023 return NET_RX_SUCCESS
;
3027 return netif_receive_skb(skb
);
3030 static void napi_gro_flush(struct napi_struct
*napi
)
3032 struct sk_buff
*skb
, *next
;
3034 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3037 napi_gro_complete(skb
);
3040 napi
->gro_count
= 0;
3041 napi
->gro_list
= NULL
;
3044 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3046 struct sk_buff
**pp
= NULL
;
3047 struct packet_type
*ptype
;
3048 __be16 type
= skb
->protocol
;
3049 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3052 enum gro_result ret
;
3054 if (!(skb
->dev
->features
& NETIF_F_GRO
))
3057 if (skb_is_gso(skb
) || skb_has_frags(skb
))
3061 list_for_each_entry_rcu(ptype
, head
, list
) {
3062 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3065 skb_set_network_header(skb
, skb_gro_offset(skb
));
3066 mac_len
= skb
->network_header
- skb
->mac_header
;
3067 skb
->mac_len
= mac_len
;
3068 NAPI_GRO_CB(skb
)->same_flow
= 0;
3069 NAPI_GRO_CB(skb
)->flush
= 0;
3070 NAPI_GRO_CB(skb
)->free
= 0;
3072 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3077 if (&ptype
->list
== head
)
3080 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3081 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3084 struct sk_buff
*nskb
= *pp
;
3088 napi_gro_complete(nskb
);
3095 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3099 NAPI_GRO_CB(skb
)->count
= 1;
3100 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3101 skb
->next
= napi
->gro_list
;
3102 napi
->gro_list
= skb
;
3106 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3107 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3109 BUG_ON(skb
->end
- skb
->tail
< grow
);
3111 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3114 skb
->data_len
-= grow
;
3116 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3117 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3119 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3120 put_page(skb_shinfo(skb
)->frags
[0].page
);
3121 memmove(skb_shinfo(skb
)->frags
,
3122 skb_shinfo(skb
)->frags
+ 1,
3123 --skb_shinfo(skb
)->nr_frags
);
3134 EXPORT_SYMBOL(dev_gro_receive
);
3137 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3141 if (netpoll_rx_on(skb
))
3144 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3145 NAPI_GRO_CB(p
)->same_flow
=
3146 (p
->dev
== skb
->dev
) &&
3147 !compare_ether_header(skb_mac_header(p
),
3148 skb_gro_mac_header(skb
));
3149 NAPI_GRO_CB(p
)->flush
= 0;
3152 return dev_gro_receive(napi
, skb
);
3155 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3159 if (netif_receive_skb(skb
))
3164 case GRO_MERGED_FREE
:
3175 EXPORT_SYMBOL(napi_skb_finish
);
3177 void skb_gro_reset_offset(struct sk_buff
*skb
)
3179 NAPI_GRO_CB(skb
)->data_offset
= 0;
3180 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3181 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3183 if (skb
->mac_header
== skb
->tail
&&
3184 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3185 NAPI_GRO_CB(skb
)->frag0
=
3186 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3187 skb_shinfo(skb
)->frags
[0].page_offset
;
3188 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3191 EXPORT_SYMBOL(skb_gro_reset_offset
);
3193 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3195 skb_gro_reset_offset(skb
);
3197 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3199 EXPORT_SYMBOL(napi_gro_receive
);
3201 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3203 __skb_pull(skb
, skb_headlen(skb
));
3204 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3208 EXPORT_SYMBOL(napi_reuse_skb
);
3210 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3212 struct sk_buff
*skb
= napi
->skb
;
3215 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3221 EXPORT_SYMBOL(napi_get_frags
);
3223 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3229 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3231 if (ret
== GRO_HELD
)
3232 skb_gro_pull(skb
, -ETH_HLEN
);
3233 else if (netif_receive_skb(skb
))
3238 case GRO_MERGED_FREE
:
3239 napi_reuse_skb(napi
, skb
);
3248 EXPORT_SYMBOL(napi_frags_finish
);
3250 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3252 struct sk_buff
*skb
= napi
->skb
;
3259 skb_reset_mac_header(skb
);
3260 skb_gro_reset_offset(skb
);
3262 off
= skb_gro_offset(skb
);
3263 hlen
= off
+ sizeof(*eth
);
3264 eth
= skb_gro_header_fast(skb
, off
);
3265 if (skb_gro_header_hard(skb
, hlen
)) {
3266 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3267 if (unlikely(!eth
)) {
3268 napi_reuse_skb(napi
, skb
);
3274 skb_gro_pull(skb
, sizeof(*eth
));
3277 * This works because the only protocols we care about don't require
3278 * special handling. We'll fix it up properly at the end.
3280 skb
->protocol
= eth
->h_proto
;
3285 EXPORT_SYMBOL(napi_frags_skb
);
3287 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3289 struct sk_buff
*skb
= napi_frags_skb(napi
);
3294 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3296 EXPORT_SYMBOL(napi_gro_frags
);
3299 * net_rps_action sends any pending IPI's for rps.
3300 * Note: called with local irq disabled, but exits with local irq enabled.
3302 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3305 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3308 sd
->rps_ipi_list
= NULL
;
3312 /* Send pending IPI's to kick RPS processing on remote cpus. */
3314 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3316 if (cpu_online(remsd
->cpu
))
3317 __smp_call_function_single(remsd
->cpu
,
3326 static int process_backlog(struct napi_struct
*napi
, int quota
)
3329 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3332 /* Check if we have pending ipi, its better to send them now,
3333 * not waiting net_rx_action() end.
3335 if (sd
->rps_ipi_list
) {
3336 local_irq_disable();
3337 net_rps_action_and_irq_enable(sd
);
3340 napi
->weight
= weight_p
;
3341 local_irq_disable();
3342 while (work
< quota
) {
3343 struct sk_buff
*skb
;
3346 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3348 __netif_receive_skb(skb
);
3349 local_irq_disable();
3350 input_queue_head_incr(sd
);
3351 if (++work
>= quota
) {
3358 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3360 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3361 &sd
->process_queue
);
3363 if (qlen
< quota
- work
) {
3365 * Inline a custom version of __napi_complete().
3366 * only current cpu owns and manipulates this napi,
3367 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3368 * we can use a plain write instead of clear_bit(),
3369 * and we dont need an smp_mb() memory barrier.
3371 list_del(&napi
->poll_list
);
3374 quota
= work
+ qlen
;
3384 * __napi_schedule - schedule for receive
3385 * @n: entry to schedule
3387 * The entry's receive function will be scheduled to run
3389 void __napi_schedule(struct napi_struct
*n
)
3391 unsigned long flags
;
3393 local_irq_save(flags
);
3394 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3395 local_irq_restore(flags
);
3397 EXPORT_SYMBOL(__napi_schedule
);
3399 void __napi_complete(struct napi_struct
*n
)
3401 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3402 BUG_ON(n
->gro_list
);
3404 list_del(&n
->poll_list
);
3405 smp_mb__before_clear_bit();
3406 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3408 EXPORT_SYMBOL(__napi_complete
);
3410 void napi_complete(struct napi_struct
*n
)
3412 unsigned long flags
;
3415 * don't let napi dequeue from the cpu poll list
3416 * just in case its running on a different cpu
3418 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3422 local_irq_save(flags
);
3424 local_irq_restore(flags
);
3426 EXPORT_SYMBOL(napi_complete
);
3428 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3429 int (*poll
)(struct napi_struct
*, int), int weight
)
3431 INIT_LIST_HEAD(&napi
->poll_list
);
3432 napi
->gro_count
= 0;
3433 napi
->gro_list
= NULL
;
3436 napi
->weight
= weight
;
3437 list_add(&napi
->dev_list
, &dev
->napi_list
);
3439 #ifdef CONFIG_NETPOLL
3440 spin_lock_init(&napi
->poll_lock
);
3441 napi
->poll_owner
= -1;
3443 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3445 EXPORT_SYMBOL(netif_napi_add
);
3447 void netif_napi_del(struct napi_struct
*napi
)
3449 struct sk_buff
*skb
, *next
;
3451 list_del_init(&napi
->dev_list
);
3452 napi_free_frags(napi
);
3454 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3460 napi
->gro_list
= NULL
;
3461 napi
->gro_count
= 0;
3463 EXPORT_SYMBOL(netif_napi_del
);
3465 static void net_rx_action(struct softirq_action
*h
)
3467 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3468 unsigned long time_limit
= jiffies
+ 2;
3469 int budget
= netdev_budget
;
3472 local_irq_disable();
3474 while (!list_empty(&sd
->poll_list
)) {
3475 struct napi_struct
*n
;
3478 /* If softirq window is exhuasted then punt.
3479 * Allow this to run for 2 jiffies since which will allow
3480 * an average latency of 1.5/HZ.
3482 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3487 /* Even though interrupts have been re-enabled, this
3488 * access is safe because interrupts can only add new
3489 * entries to the tail of this list, and only ->poll()
3490 * calls can remove this head entry from the list.
3492 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3494 have
= netpoll_poll_lock(n
);
3498 /* This NAPI_STATE_SCHED test is for avoiding a race
3499 * with netpoll's poll_napi(). Only the entity which
3500 * obtains the lock and sees NAPI_STATE_SCHED set will
3501 * actually make the ->poll() call. Therefore we avoid
3502 * accidently calling ->poll() when NAPI is not scheduled.
3505 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3506 work
= n
->poll(n
, weight
);
3510 WARN_ON_ONCE(work
> weight
);
3514 local_irq_disable();
3516 /* Drivers must not modify the NAPI state if they
3517 * consume the entire weight. In such cases this code
3518 * still "owns" the NAPI instance and therefore can
3519 * move the instance around on the list at-will.
3521 if (unlikely(work
== weight
)) {
3522 if (unlikely(napi_disable_pending(n
))) {
3525 local_irq_disable();
3527 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3530 netpoll_poll_unlock(have
);
3533 net_rps_action_and_irq_enable(sd
);
3535 #ifdef CONFIG_NET_DMA
3537 * There may not be any more sk_buffs coming right now, so push
3538 * any pending DMA copies to hardware
3540 dma_issue_pending_all();
3547 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3551 static gifconf_func_t
*gifconf_list
[NPROTO
];
3554 * register_gifconf - register a SIOCGIF handler
3555 * @family: Address family
3556 * @gifconf: Function handler
3558 * Register protocol dependent address dumping routines. The handler
3559 * that is passed must not be freed or reused until it has been replaced
3560 * by another handler.
3562 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3564 if (family
>= NPROTO
)
3566 gifconf_list
[family
] = gifconf
;
3569 EXPORT_SYMBOL(register_gifconf
);
3573 * Map an interface index to its name (SIOCGIFNAME)
3577 * We need this ioctl for efficient implementation of the
3578 * if_indextoname() function required by the IPv6 API. Without
3579 * it, we would have to search all the interfaces to find a
3583 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3585 struct net_device
*dev
;
3589 * Fetch the caller's info block.
3592 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3596 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3602 strcpy(ifr
.ifr_name
, dev
->name
);
3605 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3611 * Perform a SIOCGIFCONF call. This structure will change
3612 * size eventually, and there is nothing I can do about it.
3613 * Thus we will need a 'compatibility mode'.
3616 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3619 struct net_device
*dev
;
3626 * Fetch the caller's info block.
3629 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3636 * Loop over the interfaces, and write an info block for each.
3640 for_each_netdev(net
, dev
) {
3641 for (i
= 0; i
< NPROTO
; i
++) {
3642 if (gifconf_list
[i
]) {
3645 done
= gifconf_list
[i
](dev
, NULL
, 0);
3647 done
= gifconf_list
[i
](dev
, pos
+ total
,
3657 * All done. Write the updated control block back to the caller.
3659 ifc
.ifc_len
= total
;
3662 * Both BSD and Solaris return 0 here, so we do too.
3664 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3667 #ifdef CONFIG_PROC_FS
3669 * This is invoked by the /proc filesystem handler to display a device
3672 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3675 struct net
*net
= seq_file_net(seq
);
3677 struct net_device
*dev
;
3681 return SEQ_START_TOKEN
;
3684 for_each_netdev_rcu(net
, dev
)
3691 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3693 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3694 first_net_device(seq_file_net(seq
)) :
3695 next_net_device((struct net_device
*)v
);
3698 return rcu_dereference(dev
);
3701 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3707 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3709 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
);
3711 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3712 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3713 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3715 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3716 stats
->rx_fifo_errors
,
3717 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3718 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3719 stats
->rx_compressed
, stats
->multicast
,
3720 stats
->tx_bytes
, stats
->tx_packets
,
3721 stats
->tx_errors
, stats
->tx_dropped
,
3722 stats
->tx_fifo_errors
, stats
->collisions
,
3723 stats
->tx_carrier_errors
+
3724 stats
->tx_aborted_errors
+
3725 stats
->tx_window_errors
+
3726 stats
->tx_heartbeat_errors
,
3727 stats
->tx_compressed
);
3731 * Called from the PROCfs module. This now uses the new arbitrary sized
3732 * /proc/net interface to create /proc/net/dev
3734 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3736 if (v
== SEQ_START_TOKEN
)
3737 seq_puts(seq
, "Inter-| Receive "
3739 " face |bytes packets errs drop fifo frame "
3740 "compressed multicast|bytes packets errs "
3741 "drop fifo colls carrier compressed\n");
3743 dev_seq_printf_stats(seq
, v
);
3747 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
3749 struct softnet_data
*sd
= NULL
;
3751 while (*pos
< nr_cpu_ids
)
3752 if (cpu_online(*pos
)) {
3753 sd
= &per_cpu(softnet_data
, *pos
);
3760 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3762 return softnet_get_online(pos
);
3765 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3768 return softnet_get_online(pos
);
3771 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3775 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3777 struct softnet_data
*sd
= v
;
3779 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3780 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
3781 0, 0, 0, 0, /* was fastroute */
3782 sd
->cpu_collision
, sd
->received_rps
);
3786 static const struct seq_operations dev_seq_ops
= {
3787 .start
= dev_seq_start
,
3788 .next
= dev_seq_next
,
3789 .stop
= dev_seq_stop
,
3790 .show
= dev_seq_show
,
3793 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3795 return seq_open_net(inode
, file
, &dev_seq_ops
,
3796 sizeof(struct seq_net_private
));
3799 static const struct file_operations dev_seq_fops
= {
3800 .owner
= THIS_MODULE
,
3801 .open
= dev_seq_open
,
3803 .llseek
= seq_lseek
,
3804 .release
= seq_release_net
,
3807 static const struct seq_operations softnet_seq_ops
= {
3808 .start
= softnet_seq_start
,
3809 .next
= softnet_seq_next
,
3810 .stop
= softnet_seq_stop
,
3811 .show
= softnet_seq_show
,
3814 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3816 return seq_open(file
, &softnet_seq_ops
);
3819 static const struct file_operations softnet_seq_fops
= {
3820 .owner
= THIS_MODULE
,
3821 .open
= softnet_seq_open
,
3823 .llseek
= seq_lseek
,
3824 .release
= seq_release
,
3827 static void *ptype_get_idx(loff_t pos
)
3829 struct packet_type
*pt
= NULL
;
3833 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3839 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3840 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3849 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3853 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3856 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3858 struct packet_type
*pt
;
3859 struct list_head
*nxt
;
3863 if (v
== SEQ_START_TOKEN
)
3864 return ptype_get_idx(0);
3867 nxt
= pt
->list
.next
;
3868 if (pt
->type
== htons(ETH_P_ALL
)) {
3869 if (nxt
!= &ptype_all
)
3872 nxt
= ptype_base
[0].next
;
3874 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3876 while (nxt
== &ptype_base
[hash
]) {
3877 if (++hash
>= PTYPE_HASH_SIZE
)
3879 nxt
= ptype_base
[hash
].next
;
3882 return list_entry(nxt
, struct packet_type
, list
);
3885 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3891 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3893 struct packet_type
*pt
= v
;
3895 if (v
== SEQ_START_TOKEN
)
3896 seq_puts(seq
, "Type Device Function\n");
3897 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3898 if (pt
->type
== htons(ETH_P_ALL
))
3899 seq_puts(seq
, "ALL ");
3901 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3903 seq_printf(seq
, " %-8s %pF\n",
3904 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
3910 static const struct seq_operations ptype_seq_ops
= {
3911 .start
= ptype_seq_start
,
3912 .next
= ptype_seq_next
,
3913 .stop
= ptype_seq_stop
,
3914 .show
= ptype_seq_show
,
3917 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
3919 return seq_open_net(inode
, file
, &ptype_seq_ops
,
3920 sizeof(struct seq_net_private
));
3923 static const struct file_operations ptype_seq_fops
= {
3924 .owner
= THIS_MODULE
,
3925 .open
= ptype_seq_open
,
3927 .llseek
= seq_lseek
,
3928 .release
= seq_release_net
,
3932 static int __net_init
dev_proc_net_init(struct net
*net
)
3936 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
3938 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
3940 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
3943 if (wext_proc_init(net
))
3949 proc_net_remove(net
, "ptype");
3951 proc_net_remove(net
, "softnet_stat");
3953 proc_net_remove(net
, "dev");
3957 static void __net_exit
dev_proc_net_exit(struct net
*net
)
3959 wext_proc_exit(net
);
3961 proc_net_remove(net
, "ptype");
3962 proc_net_remove(net
, "softnet_stat");
3963 proc_net_remove(net
, "dev");
3966 static struct pernet_operations __net_initdata dev_proc_ops
= {
3967 .init
= dev_proc_net_init
,
3968 .exit
= dev_proc_net_exit
,
3971 static int __init
dev_proc_init(void)
3973 return register_pernet_subsys(&dev_proc_ops
);
3976 #define dev_proc_init() 0
3977 #endif /* CONFIG_PROC_FS */
3981 * netdev_set_master - set up master/slave pair
3982 * @slave: slave device
3983 * @master: new master device
3985 * Changes the master device of the slave. Pass %NULL to break the
3986 * bonding. The caller must hold the RTNL semaphore. On a failure
3987 * a negative errno code is returned. On success the reference counts
3988 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3989 * function returns zero.
3991 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
3993 struct net_device
*old
= slave
->master
;
4003 slave
->master
= master
;
4010 slave
->flags
|= IFF_SLAVE
;
4012 slave
->flags
&= ~IFF_SLAVE
;
4014 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4017 EXPORT_SYMBOL(netdev_set_master
);
4019 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4021 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4023 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4024 ops
->ndo_change_rx_flags(dev
, flags
);
4027 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4029 unsigned short old_flags
= dev
->flags
;
4035 dev
->flags
|= IFF_PROMISC
;
4036 dev
->promiscuity
+= inc
;
4037 if (dev
->promiscuity
== 0) {
4040 * If inc causes overflow, untouch promisc and return error.
4043 dev
->flags
&= ~IFF_PROMISC
;
4045 dev
->promiscuity
-= inc
;
4046 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4047 "set promiscuity failed, promiscuity feature "
4048 "of device might be broken.\n", dev
->name
);
4052 if (dev
->flags
!= old_flags
) {
4053 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4054 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4056 if (audit_enabled
) {
4057 current_uid_gid(&uid
, &gid
);
4058 audit_log(current
->audit_context
, GFP_ATOMIC
,
4059 AUDIT_ANOM_PROMISCUOUS
,
4060 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4061 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4062 (old_flags
& IFF_PROMISC
),
4063 audit_get_loginuid(current
),
4065 audit_get_sessionid(current
));
4068 dev_change_rx_flags(dev
, IFF_PROMISC
);
4074 * dev_set_promiscuity - update promiscuity count on a device
4078 * Add or remove promiscuity from a device. While the count in the device
4079 * remains above zero the interface remains promiscuous. Once it hits zero
4080 * the device reverts back to normal filtering operation. A negative inc
4081 * value is used to drop promiscuity on the device.
4082 * Return 0 if successful or a negative errno code on error.
4084 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4086 unsigned short old_flags
= dev
->flags
;
4089 err
= __dev_set_promiscuity(dev
, inc
);
4092 if (dev
->flags
!= old_flags
)
4093 dev_set_rx_mode(dev
);
4096 EXPORT_SYMBOL(dev_set_promiscuity
);
4099 * dev_set_allmulti - update allmulti count on a device
4103 * Add or remove reception of all multicast frames to a device. While the
4104 * count in the device remains above zero the interface remains listening
4105 * to all interfaces. Once it hits zero the device reverts back to normal
4106 * filtering operation. A negative @inc value is used to drop the counter
4107 * when releasing a resource needing all multicasts.
4108 * Return 0 if successful or a negative errno code on error.
4111 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4113 unsigned short old_flags
= dev
->flags
;
4117 dev
->flags
|= IFF_ALLMULTI
;
4118 dev
->allmulti
+= inc
;
4119 if (dev
->allmulti
== 0) {
4122 * If inc causes overflow, untouch allmulti and return error.
4125 dev
->flags
&= ~IFF_ALLMULTI
;
4127 dev
->allmulti
-= inc
;
4128 printk(KERN_WARNING
"%s: allmulti touches roof, "
4129 "set allmulti failed, allmulti feature of "
4130 "device might be broken.\n", dev
->name
);
4134 if (dev
->flags
^ old_flags
) {
4135 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4136 dev_set_rx_mode(dev
);
4140 EXPORT_SYMBOL(dev_set_allmulti
);
4143 * Upload unicast and multicast address lists to device and
4144 * configure RX filtering. When the device doesn't support unicast
4145 * filtering it is put in promiscuous mode while unicast addresses
4148 void __dev_set_rx_mode(struct net_device
*dev
)
4150 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4152 /* dev_open will call this function so the list will stay sane. */
4153 if (!(dev
->flags
&IFF_UP
))
4156 if (!netif_device_present(dev
))
4159 if (ops
->ndo_set_rx_mode
)
4160 ops
->ndo_set_rx_mode(dev
);
4162 /* Unicast addresses changes may only happen under the rtnl,
4163 * therefore calling __dev_set_promiscuity here is safe.
4165 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4166 __dev_set_promiscuity(dev
, 1);
4167 dev
->uc_promisc
= 1;
4168 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4169 __dev_set_promiscuity(dev
, -1);
4170 dev
->uc_promisc
= 0;
4173 if (ops
->ndo_set_multicast_list
)
4174 ops
->ndo_set_multicast_list(dev
);
4178 void dev_set_rx_mode(struct net_device
*dev
)
4180 netif_addr_lock_bh(dev
);
4181 __dev_set_rx_mode(dev
);
4182 netif_addr_unlock_bh(dev
);
4186 * dev_get_flags - get flags reported to userspace
4189 * Get the combination of flag bits exported through APIs to userspace.
4191 unsigned dev_get_flags(const struct net_device
*dev
)
4195 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4200 (dev
->gflags
& (IFF_PROMISC
|
4203 if (netif_running(dev
)) {
4204 if (netif_oper_up(dev
))
4205 flags
|= IFF_RUNNING
;
4206 if (netif_carrier_ok(dev
))
4207 flags
|= IFF_LOWER_UP
;
4208 if (netif_dormant(dev
))
4209 flags
|= IFF_DORMANT
;
4214 EXPORT_SYMBOL(dev_get_flags
);
4216 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4218 int old_flags
= dev
->flags
;
4224 * Set the flags on our device.
4227 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4228 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4230 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4234 * Load in the correct multicast list now the flags have changed.
4237 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4238 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4240 dev_set_rx_mode(dev
);
4243 * Have we downed the interface. We handle IFF_UP ourselves
4244 * according to user attempts to set it, rather than blindly
4249 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4250 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4253 dev_set_rx_mode(dev
);
4256 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4257 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4259 dev
->gflags
^= IFF_PROMISC
;
4260 dev_set_promiscuity(dev
, inc
);
4263 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4264 is important. Some (broken) drivers set IFF_PROMISC, when
4265 IFF_ALLMULTI is requested not asking us and not reporting.
4267 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4268 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4270 dev
->gflags
^= IFF_ALLMULTI
;
4271 dev_set_allmulti(dev
, inc
);
4277 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4279 unsigned int changes
= dev
->flags
^ old_flags
;
4281 if (changes
& IFF_UP
) {
4282 if (dev
->flags
& IFF_UP
)
4283 call_netdevice_notifiers(NETDEV_UP
, dev
);
4285 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4288 if (dev
->flags
& IFF_UP
&&
4289 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4290 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4294 * dev_change_flags - change device settings
4296 * @flags: device state flags
4298 * Change settings on device based state flags. The flags are
4299 * in the userspace exported format.
4301 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4304 int old_flags
= dev
->flags
;
4306 ret
= __dev_change_flags(dev
, flags
);
4310 changes
= old_flags
^ dev
->flags
;
4312 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4314 __dev_notify_flags(dev
, old_flags
);
4317 EXPORT_SYMBOL(dev_change_flags
);
4320 * dev_set_mtu - Change maximum transfer unit
4322 * @new_mtu: new transfer unit
4324 * Change the maximum transfer size of the network device.
4326 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4328 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4331 if (new_mtu
== dev
->mtu
)
4334 /* MTU must be positive. */
4338 if (!netif_device_present(dev
))
4342 if (ops
->ndo_change_mtu
)
4343 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4347 if (!err
&& dev
->flags
& IFF_UP
)
4348 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4351 EXPORT_SYMBOL(dev_set_mtu
);
4354 * dev_set_mac_address - Change Media Access Control Address
4358 * Change the hardware (MAC) address of the device
4360 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4362 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4365 if (!ops
->ndo_set_mac_address
)
4367 if (sa
->sa_family
!= dev
->type
)
4369 if (!netif_device_present(dev
))
4371 err
= ops
->ndo_set_mac_address(dev
, sa
);
4373 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4376 EXPORT_SYMBOL(dev_set_mac_address
);
4379 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4381 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4384 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4390 case SIOCGIFFLAGS
: /* Get interface flags */
4391 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4394 case SIOCGIFMETRIC
: /* Get the metric on the interface
4395 (currently unused) */
4396 ifr
->ifr_metric
= 0;
4399 case SIOCGIFMTU
: /* Get the MTU of a device */
4400 ifr
->ifr_mtu
= dev
->mtu
;
4405 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4407 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4408 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4409 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4417 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4418 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4419 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4420 ifr
->ifr_map
.irq
= dev
->irq
;
4421 ifr
->ifr_map
.dma
= dev
->dma
;
4422 ifr
->ifr_map
.port
= dev
->if_port
;
4426 ifr
->ifr_ifindex
= dev
->ifindex
;
4430 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4434 /* dev_ioctl() should ensure this case
4446 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4448 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4451 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4452 const struct net_device_ops
*ops
;
4457 ops
= dev
->netdev_ops
;
4460 case SIOCSIFFLAGS
: /* Set interface flags */
4461 return dev_change_flags(dev
, ifr
->ifr_flags
);
4463 case SIOCSIFMETRIC
: /* Set the metric on the interface
4464 (currently unused) */
4467 case SIOCSIFMTU
: /* Set the MTU of a device */
4468 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4471 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4473 case SIOCSIFHWBROADCAST
:
4474 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4476 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4477 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4478 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4482 if (ops
->ndo_set_config
) {
4483 if (!netif_device_present(dev
))
4485 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4490 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4491 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4493 if (!netif_device_present(dev
))
4495 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4498 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4499 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4501 if (!netif_device_present(dev
))
4503 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4506 if (ifr
->ifr_qlen
< 0)
4508 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4512 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4513 return dev_change_name(dev
, ifr
->ifr_newname
);
4516 * Unknown or private ioctl
4519 if ((cmd
>= SIOCDEVPRIVATE
&&
4520 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4521 cmd
== SIOCBONDENSLAVE
||
4522 cmd
== SIOCBONDRELEASE
||
4523 cmd
== SIOCBONDSETHWADDR
||
4524 cmd
== SIOCBONDSLAVEINFOQUERY
||
4525 cmd
== SIOCBONDINFOQUERY
||
4526 cmd
== SIOCBONDCHANGEACTIVE
||
4527 cmd
== SIOCGMIIPHY
||
4528 cmd
== SIOCGMIIREG
||
4529 cmd
== SIOCSMIIREG
||
4530 cmd
== SIOCBRADDIF
||
4531 cmd
== SIOCBRDELIF
||
4532 cmd
== SIOCSHWTSTAMP
||
4533 cmd
== SIOCWANDEV
) {
4535 if (ops
->ndo_do_ioctl
) {
4536 if (netif_device_present(dev
))
4537 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4549 * This function handles all "interface"-type I/O control requests. The actual
4550 * 'doing' part of this is dev_ifsioc above.
4554 * dev_ioctl - network device ioctl
4555 * @net: the applicable net namespace
4556 * @cmd: command to issue
4557 * @arg: pointer to a struct ifreq in user space
4559 * Issue ioctl functions to devices. This is normally called by the
4560 * user space syscall interfaces but can sometimes be useful for
4561 * other purposes. The return value is the return from the syscall if
4562 * positive or a negative errno code on error.
4565 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4571 /* One special case: SIOCGIFCONF takes ifconf argument
4572 and requires shared lock, because it sleeps writing
4576 if (cmd
== SIOCGIFCONF
) {
4578 ret
= dev_ifconf(net
, (char __user
*) arg
);
4582 if (cmd
== SIOCGIFNAME
)
4583 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4585 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4588 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4590 colon
= strchr(ifr
.ifr_name
, ':');
4595 * See which interface the caller is talking about.
4600 * These ioctl calls:
4601 * - can be done by all.
4602 * - atomic and do not require locking.
4613 dev_load(net
, ifr
.ifr_name
);
4615 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4620 if (copy_to_user(arg
, &ifr
,
4621 sizeof(struct ifreq
)))
4627 dev_load(net
, ifr
.ifr_name
);
4629 ret
= dev_ethtool(net
, &ifr
);
4634 if (copy_to_user(arg
, &ifr
,
4635 sizeof(struct ifreq
)))
4641 * These ioctl calls:
4642 * - require superuser power.
4643 * - require strict serialization.
4649 if (!capable(CAP_NET_ADMIN
))
4651 dev_load(net
, ifr
.ifr_name
);
4653 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4658 if (copy_to_user(arg
, &ifr
,
4659 sizeof(struct ifreq
)))
4665 * These ioctl calls:
4666 * - require superuser power.
4667 * - require strict serialization.
4668 * - do not return a value
4678 case SIOCSIFHWBROADCAST
:
4681 case SIOCBONDENSLAVE
:
4682 case SIOCBONDRELEASE
:
4683 case SIOCBONDSETHWADDR
:
4684 case SIOCBONDCHANGEACTIVE
:
4688 if (!capable(CAP_NET_ADMIN
))
4691 case SIOCBONDSLAVEINFOQUERY
:
4692 case SIOCBONDINFOQUERY
:
4693 dev_load(net
, ifr
.ifr_name
);
4695 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4700 /* Get the per device memory space. We can add this but
4701 * currently do not support it */
4703 /* Set the per device memory buffer space.
4704 * Not applicable in our case */
4709 * Unknown or private ioctl.
4712 if (cmd
== SIOCWANDEV
||
4713 (cmd
>= SIOCDEVPRIVATE
&&
4714 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4715 dev_load(net
, ifr
.ifr_name
);
4717 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4719 if (!ret
&& copy_to_user(arg
, &ifr
,
4720 sizeof(struct ifreq
)))
4724 /* Take care of Wireless Extensions */
4725 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4726 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4733 * dev_new_index - allocate an ifindex
4734 * @net: the applicable net namespace
4736 * Returns a suitable unique value for a new device interface
4737 * number. The caller must hold the rtnl semaphore or the
4738 * dev_base_lock to be sure it remains unique.
4740 static int dev_new_index(struct net
*net
)
4746 if (!__dev_get_by_index(net
, ifindex
))
4751 /* Delayed registration/unregisteration */
4752 static LIST_HEAD(net_todo_list
);
4754 static void net_set_todo(struct net_device
*dev
)
4756 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4759 static void rollback_registered_many(struct list_head
*head
)
4761 struct net_device
*dev
, *tmp
;
4763 BUG_ON(dev_boot_phase
);
4766 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4767 /* Some devices call without registering
4768 * for initialization unwind. Remove those
4769 * devices and proceed with the remaining.
4771 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4772 pr_debug("unregister_netdevice: device %s/%p never "
4773 "was registered\n", dev
->name
, dev
);
4776 list_del(&dev
->unreg_list
);
4780 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4782 /* If device is running, close it first. */
4785 /* And unlink it from device chain. */
4786 unlist_netdevice(dev
);
4788 dev
->reg_state
= NETREG_UNREGISTERING
;
4793 list_for_each_entry(dev
, head
, unreg_list
) {
4794 /* Shutdown queueing discipline. */
4798 /* Notify protocols, that we are about to destroy
4799 this device. They should clean all the things.
4801 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4803 if (!dev
->rtnl_link_ops
||
4804 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
4805 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
4808 * Flush the unicast and multicast chains
4813 if (dev
->netdev_ops
->ndo_uninit
)
4814 dev
->netdev_ops
->ndo_uninit(dev
);
4816 /* Notifier chain MUST detach us from master device. */
4817 WARN_ON(dev
->master
);
4819 /* Remove entries from kobject tree */
4820 netdev_unregister_kobject(dev
);
4823 /* Process any work delayed until the end of the batch */
4824 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
4825 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
4829 list_for_each_entry(dev
, head
, unreg_list
)
4833 static void rollback_registered(struct net_device
*dev
)
4837 list_add(&dev
->unreg_list
, &single
);
4838 rollback_registered_many(&single
);
4841 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
4842 struct netdev_queue
*dev_queue
,
4845 spin_lock_init(&dev_queue
->_xmit_lock
);
4846 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
4847 dev_queue
->xmit_lock_owner
= -1;
4850 static void netdev_init_queue_locks(struct net_device
*dev
)
4852 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
4853 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
4856 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
4858 /* Fix illegal SG+CSUM combinations. */
4859 if ((features
& NETIF_F_SG
) &&
4860 !(features
& NETIF_F_ALL_CSUM
)) {
4862 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
4863 "checksum feature.\n", name
);
4864 features
&= ~NETIF_F_SG
;
4867 /* TSO requires that SG is present as well. */
4868 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
4870 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
4871 "SG feature.\n", name
);
4872 features
&= ~NETIF_F_TSO
;
4875 if (features
& NETIF_F_UFO
) {
4876 if (!(features
& NETIF_F_GEN_CSUM
)) {
4878 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4879 "since no NETIF_F_HW_CSUM feature.\n",
4881 features
&= ~NETIF_F_UFO
;
4884 if (!(features
& NETIF_F_SG
)) {
4886 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4887 "since no NETIF_F_SG feature.\n", name
);
4888 features
&= ~NETIF_F_UFO
;
4894 EXPORT_SYMBOL(netdev_fix_features
);
4897 * netif_stacked_transfer_operstate - transfer operstate
4898 * @rootdev: the root or lower level device to transfer state from
4899 * @dev: the device to transfer operstate to
4901 * Transfer operational state from root to device. This is normally
4902 * called when a stacking relationship exists between the root
4903 * device and the device(a leaf device).
4905 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
4906 struct net_device
*dev
)
4908 if (rootdev
->operstate
== IF_OPER_DORMANT
)
4909 netif_dormant_on(dev
);
4911 netif_dormant_off(dev
);
4913 if (netif_carrier_ok(rootdev
)) {
4914 if (!netif_carrier_ok(dev
))
4915 netif_carrier_on(dev
);
4917 if (netif_carrier_ok(dev
))
4918 netif_carrier_off(dev
);
4921 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
4924 * register_netdevice - register a network device
4925 * @dev: device to register
4927 * Take a completed network device structure and add it to the kernel
4928 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4929 * chain. 0 is returned on success. A negative errno code is returned
4930 * on a failure to set up the device, or if the name is a duplicate.
4932 * Callers must hold the rtnl semaphore. You may want
4933 * register_netdev() instead of this.
4936 * The locking appears insufficient to guarantee two parallel registers
4937 * will not get the same name.
4940 int register_netdevice(struct net_device
*dev
)
4943 struct net
*net
= dev_net(dev
);
4945 BUG_ON(dev_boot_phase
);
4950 /* When net_device's are persistent, this will be fatal. */
4951 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
4954 spin_lock_init(&dev
->addr_list_lock
);
4955 netdev_set_addr_lockdep_class(dev
);
4956 netdev_init_queue_locks(dev
);
4961 if (!dev
->num_rx_queues
) {
4963 * Allocate a single RX queue if driver never called
4967 dev
->_rx
= kzalloc(sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
4973 dev
->_rx
->first
= dev
->_rx
;
4974 atomic_set(&dev
->_rx
->count
, 1);
4975 dev
->num_rx_queues
= 1;
4978 /* Init, if this function is available */
4979 if (dev
->netdev_ops
->ndo_init
) {
4980 ret
= dev
->netdev_ops
->ndo_init(dev
);
4988 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
4992 dev
->ifindex
= dev_new_index(net
);
4993 if (dev
->iflink
== -1)
4994 dev
->iflink
= dev
->ifindex
;
4996 /* Fix illegal checksum combinations */
4997 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
4998 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4999 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5001 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5004 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5005 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5006 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5008 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5011 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5013 /* Enable software GSO if SG is supported. */
5014 if (dev
->features
& NETIF_F_SG
)
5015 dev
->features
|= NETIF_F_GSO
;
5017 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5018 ret
= notifier_to_errno(ret
);
5022 ret
= netdev_register_kobject(dev
);
5025 dev
->reg_state
= NETREG_REGISTERED
;
5028 * Default initial state at registry is that the
5029 * device is present.
5032 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5034 dev_init_scheduler(dev
);
5036 list_netdevice(dev
);
5038 /* Notify protocols, that a new device appeared. */
5039 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5040 ret
= notifier_to_errno(ret
);
5042 rollback_registered(dev
);
5043 dev
->reg_state
= NETREG_UNREGISTERED
;
5046 * Prevent userspace races by waiting until the network
5047 * device is fully setup before sending notifications.
5049 if (!dev
->rtnl_link_ops
||
5050 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5051 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5057 if (dev
->netdev_ops
->ndo_uninit
)
5058 dev
->netdev_ops
->ndo_uninit(dev
);
5061 EXPORT_SYMBOL(register_netdevice
);
5064 * init_dummy_netdev - init a dummy network device for NAPI
5065 * @dev: device to init
5067 * This takes a network device structure and initialize the minimum
5068 * amount of fields so it can be used to schedule NAPI polls without
5069 * registering a full blown interface. This is to be used by drivers
5070 * that need to tie several hardware interfaces to a single NAPI
5071 * poll scheduler due to HW limitations.
5073 int init_dummy_netdev(struct net_device
*dev
)
5075 /* Clear everything. Note we don't initialize spinlocks
5076 * are they aren't supposed to be taken by any of the
5077 * NAPI code and this dummy netdev is supposed to be
5078 * only ever used for NAPI polls
5080 memset(dev
, 0, sizeof(struct net_device
));
5082 /* make sure we BUG if trying to hit standard
5083 * register/unregister code path
5085 dev
->reg_state
= NETREG_DUMMY
;
5087 /* initialize the ref count */
5088 atomic_set(&dev
->refcnt
, 1);
5090 /* NAPI wants this */
5091 INIT_LIST_HEAD(&dev
->napi_list
);
5093 /* a dummy interface is started by default */
5094 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5095 set_bit(__LINK_STATE_START
, &dev
->state
);
5099 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5103 * register_netdev - register a network device
5104 * @dev: device to register
5106 * Take a completed network device structure and add it to the kernel
5107 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5108 * chain. 0 is returned on success. A negative errno code is returned
5109 * on a failure to set up the device, or if the name is a duplicate.
5111 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5112 * and expands the device name if you passed a format string to
5115 int register_netdev(struct net_device
*dev
)
5122 * If the name is a format string the caller wants us to do a
5125 if (strchr(dev
->name
, '%')) {
5126 err
= dev_alloc_name(dev
, dev
->name
);
5131 err
= register_netdevice(dev
);
5136 EXPORT_SYMBOL(register_netdev
);
5139 * netdev_wait_allrefs - wait until all references are gone.
5141 * This is called when unregistering network devices.
5143 * Any protocol or device that holds a reference should register
5144 * for netdevice notification, and cleanup and put back the
5145 * reference if they receive an UNREGISTER event.
5146 * We can get stuck here if buggy protocols don't correctly
5149 static void netdev_wait_allrefs(struct net_device
*dev
)
5151 unsigned long rebroadcast_time
, warning_time
;
5153 linkwatch_forget_dev(dev
);
5155 rebroadcast_time
= warning_time
= jiffies
;
5156 while (atomic_read(&dev
->refcnt
) != 0) {
5157 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5160 /* Rebroadcast unregister notification */
5161 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5162 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5163 * should have already handle it the first time */
5165 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5167 /* We must not have linkwatch events
5168 * pending on unregister. If this
5169 * happens, we simply run the queue
5170 * unscheduled, resulting in a noop
5173 linkwatch_run_queue();
5178 rebroadcast_time
= jiffies
;
5183 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5184 printk(KERN_EMERG
"unregister_netdevice: "
5185 "waiting for %s to become free. Usage "
5187 dev
->name
, atomic_read(&dev
->refcnt
));
5188 warning_time
= jiffies
;
5197 * register_netdevice(x1);
5198 * register_netdevice(x2);
5200 * unregister_netdevice(y1);
5201 * unregister_netdevice(y2);
5207 * We are invoked by rtnl_unlock().
5208 * This allows us to deal with problems:
5209 * 1) We can delete sysfs objects which invoke hotplug
5210 * without deadlocking with linkwatch via keventd.
5211 * 2) Since we run with the RTNL semaphore not held, we can sleep
5212 * safely in order to wait for the netdev refcnt to drop to zero.
5214 * We must not return until all unregister events added during
5215 * the interval the lock was held have been completed.
5217 void netdev_run_todo(void)
5219 struct list_head list
;
5221 /* Snapshot list, allow later requests */
5222 list_replace_init(&net_todo_list
, &list
);
5226 while (!list_empty(&list
)) {
5227 struct net_device
*dev
5228 = list_first_entry(&list
, struct net_device
, todo_list
);
5229 list_del(&dev
->todo_list
);
5231 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5232 printk(KERN_ERR
"network todo '%s' but state %d\n",
5233 dev
->name
, dev
->reg_state
);
5238 dev
->reg_state
= NETREG_UNREGISTERED
;
5240 on_each_cpu(flush_backlog
, dev
, 1);
5242 netdev_wait_allrefs(dev
);
5245 BUG_ON(atomic_read(&dev
->refcnt
));
5246 WARN_ON(dev
->ip_ptr
);
5247 WARN_ON(dev
->ip6_ptr
);
5248 WARN_ON(dev
->dn_ptr
);
5250 if (dev
->destructor
)
5251 dev
->destructor(dev
);
5253 /* Free network device */
5254 kobject_put(&dev
->dev
.kobj
);
5259 * dev_txq_stats_fold - fold tx_queues stats
5260 * @dev: device to get statistics from
5261 * @stats: struct net_device_stats to hold results
5263 void dev_txq_stats_fold(const struct net_device
*dev
,
5264 struct net_device_stats
*stats
)
5266 unsigned long tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5268 struct netdev_queue
*txq
;
5270 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5271 txq
= netdev_get_tx_queue(dev
, i
);
5272 tx_bytes
+= txq
->tx_bytes
;
5273 tx_packets
+= txq
->tx_packets
;
5274 tx_dropped
+= txq
->tx_dropped
;
5276 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5277 stats
->tx_bytes
= tx_bytes
;
5278 stats
->tx_packets
= tx_packets
;
5279 stats
->tx_dropped
= tx_dropped
;
5282 EXPORT_SYMBOL(dev_txq_stats_fold
);
5285 * dev_get_stats - get network device statistics
5286 * @dev: device to get statistics from
5288 * Get network statistics from device. The device driver may provide
5289 * its own method by setting dev->netdev_ops->get_stats64 or
5290 * dev->netdev_ops->get_stats; otherwise the internal statistics
5291 * structure is used.
5293 const struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
)
5295 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5297 if (ops
->ndo_get_stats64
)
5298 return ops
->ndo_get_stats64(dev
);
5299 if (ops
->ndo_get_stats
)
5300 return (struct rtnl_link_stats64
*)ops
->ndo_get_stats(dev
);
5302 dev_txq_stats_fold(dev
, &dev
->stats
);
5303 return &dev
->stats64
;
5305 EXPORT_SYMBOL(dev_get_stats
);
5307 static void netdev_init_one_queue(struct net_device
*dev
,
5308 struct netdev_queue
*queue
,
5314 static void netdev_init_queues(struct net_device
*dev
)
5316 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
5317 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5318 spin_lock_init(&dev
->tx_global_lock
);
5322 * alloc_netdev_mq - allocate network device
5323 * @sizeof_priv: size of private data to allocate space for
5324 * @name: device name format string
5325 * @setup: callback to initialize device
5326 * @queue_count: the number of subqueues to allocate
5328 * Allocates a struct net_device with private data area for driver use
5329 * and performs basic initialization. Also allocates subquue structs
5330 * for each queue on the device at the end of the netdevice.
5332 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
5333 void (*setup
)(struct net_device
*), unsigned int queue_count
)
5335 struct netdev_queue
*tx
;
5336 struct net_device
*dev
;
5338 struct net_device
*p
;
5340 struct netdev_rx_queue
*rx
;
5344 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5346 alloc_size
= sizeof(struct net_device
);
5348 /* ensure 32-byte alignment of private area */
5349 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5350 alloc_size
+= sizeof_priv
;
5352 /* ensure 32-byte alignment of whole construct */
5353 alloc_size
+= NETDEV_ALIGN
- 1;
5355 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5357 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5361 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5363 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5369 rx
= kcalloc(queue_count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5371 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5376 atomic_set(&rx
->count
, queue_count
);
5379 * Set a pointer to first element in the array which holds the
5382 for (i
= 0; i
< queue_count
; i
++)
5386 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5387 dev
->padded
= (char *)dev
- (char *)p
;
5389 if (dev_addr_init(dev
))
5395 dev_net_set(dev
, &init_net
);
5398 dev
->num_tx_queues
= queue_count
;
5399 dev
->real_num_tx_queues
= queue_count
;
5403 dev
->num_rx_queues
= queue_count
;
5406 dev
->gso_max_size
= GSO_MAX_SIZE
;
5408 netdev_init_queues(dev
);
5410 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5411 dev
->ethtool_ntuple_list
.count
= 0;
5412 INIT_LIST_HEAD(&dev
->napi_list
);
5413 INIT_LIST_HEAD(&dev
->unreg_list
);
5414 INIT_LIST_HEAD(&dev
->link_watch_list
);
5415 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5417 strcpy(dev
->name
, name
);
5430 EXPORT_SYMBOL(alloc_netdev_mq
);
5433 * free_netdev - free network device
5436 * This function does the last stage of destroying an allocated device
5437 * interface. The reference to the device object is released.
5438 * If this is the last reference then it will be freed.
5440 void free_netdev(struct net_device
*dev
)
5442 struct napi_struct
*p
, *n
;
5444 release_net(dev_net(dev
));
5448 /* Flush device addresses */
5449 dev_addr_flush(dev
);
5451 /* Clear ethtool n-tuple list */
5452 ethtool_ntuple_flush(dev
);
5454 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5457 /* Compatibility with error handling in drivers */
5458 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5459 kfree((char *)dev
- dev
->padded
);
5463 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5464 dev
->reg_state
= NETREG_RELEASED
;
5466 /* will free via device release */
5467 put_device(&dev
->dev
);
5469 EXPORT_SYMBOL(free_netdev
);
5472 * synchronize_net - Synchronize with packet receive processing
5474 * Wait for packets currently being received to be done.
5475 * Does not block later packets from starting.
5477 void synchronize_net(void)
5482 EXPORT_SYMBOL(synchronize_net
);
5485 * unregister_netdevice_queue - remove device from the kernel
5489 * This function shuts down a device interface and removes it
5490 * from the kernel tables.
5491 * If head not NULL, device is queued to be unregistered later.
5493 * Callers must hold the rtnl semaphore. You may want
5494 * unregister_netdev() instead of this.
5497 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5502 list_move_tail(&dev
->unreg_list
, head
);
5504 rollback_registered(dev
);
5505 /* Finish processing unregister after unlock */
5509 EXPORT_SYMBOL(unregister_netdevice_queue
);
5512 * unregister_netdevice_many - unregister many devices
5513 * @head: list of devices
5515 void unregister_netdevice_many(struct list_head
*head
)
5517 struct net_device
*dev
;
5519 if (!list_empty(head
)) {
5520 rollback_registered_many(head
);
5521 list_for_each_entry(dev
, head
, unreg_list
)
5525 EXPORT_SYMBOL(unregister_netdevice_many
);
5528 * unregister_netdev - remove device from the kernel
5531 * This function shuts down a device interface and removes it
5532 * from the kernel tables.
5534 * This is just a wrapper for unregister_netdevice that takes
5535 * the rtnl semaphore. In general you want to use this and not
5536 * unregister_netdevice.
5538 void unregister_netdev(struct net_device
*dev
)
5541 unregister_netdevice(dev
);
5544 EXPORT_SYMBOL(unregister_netdev
);
5547 * dev_change_net_namespace - move device to different nethost namespace
5549 * @net: network namespace
5550 * @pat: If not NULL name pattern to try if the current device name
5551 * is already taken in the destination network namespace.
5553 * This function shuts down a device interface and moves it
5554 * to a new network namespace. On success 0 is returned, on
5555 * a failure a netagive errno code is returned.
5557 * Callers must hold the rtnl semaphore.
5560 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5566 /* Don't allow namespace local devices to be moved. */
5568 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5571 /* Ensure the device has been registrered */
5573 if (dev
->reg_state
!= NETREG_REGISTERED
)
5576 /* Get out if there is nothing todo */
5578 if (net_eq(dev_net(dev
), net
))
5581 /* Pick the destination device name, and ensure
5582 * we can use it in the destination network namespace.
5585 if (__dev_get_by_name(net
, dev
->name
)) {
5586 /* We get here if we can't use the current device name */
5589 if (dev_get_valid_name(dev
, pat
, 1))
5594 * And now a mini version of register_netdevice unregister_netdevice.
5597 /* If device is running close it first. */
5600 /* And unlink it from device chain */
5602 unlist_netdevice(dev
);
5606 /* Shutdown queueing discipline. */
5609 /* Notify protocols, that we are about to destroy
5610 this device. They should clean all the things.
5612 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5613 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5616 * Flush the unicast and multicast chains
5621 /* Actually switch the network namespace */
5622 dev_net_set(dev
, net
);
5624 /* If there is an ifindex conflict assign a new one */
5625 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5626 int iflink
= (dev
->iflink
== dev
->ifindex
);
5627 dev
->ifindex
= dev_new_index(net
);
5629 dev
->iflink
= dev
->ifindex
;
5632 /* Fixup kobjects */
5633 err
= device_rename(&dev
->dev
, dev
->name
);
5636 /* Add the device back in the hashes */
5637 list_netdevice(dev
);
5639 /* Notify protocols, that a new device appeared. */
5640 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5643 * Prevent userspace races by waiting until the network
5644 * device is fully setup before sending notifications.
5646 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5653 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5655 static int dev_cpu_callback(struct notifier_block
*nfb
,
5656 unsigned long action
,
5659 struct sk_buff
**list_skb
;
5660 struct sk_buff
*skb
;
5661 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5662 struct softnet_data
*sd
, *oldsd
;
5664 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5667 local_irq_disable();
5668 cpu
= smp_processor_id();
5669 sd
= &per_cpu(softnet_data
, cpu
);
5670 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5672 /* Find end of our completion_queue. */
5673 list_skb
= &sd
->completion_queue
;
5675 list_skb
= &(*list_skb
)->next
;
5676 /* Append completion queue from offline CPU. */
5677 *list_skb
= oldsd
->completion_queue
;
5678 oldsd
->completion_queue
= NULL
;
5680 /* Append output queue from offline CPU. */
5681 if (oldsd
->output_queue
) {
5682 *sd
->output_queue_tailp
= oldsd
->output_queue
;
5683 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
5684 oldsd
->output_queue
= NULL
;
5685 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
5688 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5691 /* Process offline CPU's input_pkt_queue */
5692 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
5694 input_queue_head_incr(oldsd
);
5696 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
5698 input_queue_head_incr(oldsd
);
5706 * netdev_increment_features - increment feature set by one
5707 * @all: current feature set
5708 * @one: new feature set
5709 * @mask: mask feature set
5711 * Computes a new feature set after adding a device with feature set
5712 * @one to the master device with current feature set @all. Will not
5713 * enable anything that is off in @mask. Returns the new feature set.
5715 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
5718 /* If device needs checksumming, downgrade to it. */
5719 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
5720 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
5721 else if (mask
& NETIF_F_ALL_CSUM
) {
5722 /* If one device supports v4/v6 checksumming, set for all. */
5723 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
5724 !(all
& NETIF_F_GEN_CSUM
)) {
5725 all
&= ~NETIF_F_ALL_CSUM
;
5726 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
5729 /* If one device supports hw checksumming, set for all. */
5730 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
5731 all
&= ~NETIF_F_ALL_CSUM
;
5732 all
|= NETIF_F_HW_CSUM
;
5736 one
|= NETIF_F_ALL_CSUM
;
5738 one
|= all
& NETIF_F_ONE_FOR_ALL
;
5739 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
5740 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
5744 EXPORT_SYMBOL(netdev_increment_features
);
5746 static struct hlist_head
*netdev_create_hash(void)
5749 struct hlist_head
*hash
;
5751 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5753 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5754 INIT_HLIST_HEAD(&hash
[i
]);
5759 /* Initialize per network namespace state */
5760 static int __net_init
netdev_init(struct net
*net
)
5762 INIT_LIST_HEAD(&net
->dev_base_head
);
5764 net
->dev_name_head
= netdev_create_hash();
5765 if (net
->dev_name_head
== NULL
)
5768 net
->dev_index_head
= netdev_create_hash();
5769 if (net
->dev_index_head
== NULL
)
5775 kfree(net
->dev_name_head
);
5781 * netdev_drivername - network driver for the device
5782 * @dev: network device
5783 * @buffer: buffer for resulting name
5784 * @len: size of buffer
5786 * Determine network driver for device.
5788 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
5790 const struct device_driver
*driver
;
5791 const struct device
*parent
;
5793 if (len
<= 0 || !buffer
)
5797 parent
= dev
->dev
.parent
;
5802 driver
= parent
->driver
;
5803 if (driver
&& driver
->name
)
5804 strlcpy(buffer
, driver
->name
, len
);
5808 static void __net_exit
netdev_exit(struct net
*net
)
5810 kfree(net
->dev_name_head
);
5811 kfree(net
->dev_index_head
);
5814 static struct pernet_operations __net_initdata netdev_net_ops
= {
5815 .init
= netdev_init
,
5816 .exit
= netdev_exit
,
5819 static void __net_exit
default_device_exit(struct net
*net
)
5821 struct net_device
*dev
, *aux
;
5823 * Push all migratable network devices back to the
5824 * initial network namespace
5827 for_each_netdev_safe(net
, dev
, aux
) {
5829 char fb_name
[IFNAMSIZ
];
5831 /* Ignore unmoveable devices (i.e. loopback) */
5832 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5835 /* Leave virtual devices for the generic cleanup */
5836 if (dev
->rtnl_link_ops
)
5839 /* Push remaing network devices to init_net */
5840 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
5841 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
5843 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
5844 __func__
, dev
->name
, err
);
5851 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
5853 /* At exit all network devices most be removed from a network
5854 * namespace. Do this in the reverse order of registeration.
5855 * Do this across as many network namespaces as possible to
5856 * improve batching efficiency.
5858 struct net_device
*dev
;
5860 LIST_HEAD(dev_kill_list
);
5863 list_for_each_entry(net
, net_list
, exit_list
) {
5864 for_each_netdev_reverse(net
, dev
) {
5865 if (dev
->rtnl_link_ops
)
5866 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
5868 unregister_netdevice_queue(dev
, &dev_kill_list
);
5871 unregister_netdevice_many(&dev_kill_list
);
5875 static struct pernet_operations __net_initdata default_device_ops
= {
5876 .exit
= default_device_exit
,
5877 .exit_batch
= default_device_exit_batch
,
5881 * Initialize the DEV module. At boot time this walks the device list and
5882 * unhooks any devices that fail to initialise (normally hardware not
5883 * present) and leaves us with a valid list of present and active devices.
5888 * This is called single threaded during boot, so no need
5889 * to take the rtnl semaphore.
5891 static int __init
net_dev_init(void)
5893 int i
, rc
= -ENOMEM
;
5895 BUG_ON(!dev_boot_phase
);
5897 if (dev_proc_init())
5900 if (netdev_kobject_init())
5903 INIT_LIST_HEAD(&ptype_all
);
5904 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
5905 INIT_LIST_HEAD(&ptype_base
[i
]);
5907 if (register_pernet_subsys(&netdev_net_ops
))
5911 * Initialise the packet receive queues.
5914 for_each_possible_cpu(i
) {
5915 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
5917 memset(sd
, 0, sizeof(*sd
));
5918 skb_queue_head_init(&sd
->input_pkt_queue
);
5919 skb_queue_head_init(&sd
->process_queue
);
5920 sd
->completion_queue
= NULL
;
5921 INIT_LIST_HEAD(&sd
->poll_list
);
5922 sd
->output_queue
= NULL
;
5923 sd
->output_queue_tailp
= &sd
->output_queue
;
5925 sd
->csd
.func
= rps_trigger_softirq
;
5931 sd
->backlog
.poll
= process_backlog
;
5932 sd
->backlog
.weight
= weight_p
;
5933 sd
->backlog
.gro_list
= NULL
;
5934 sd
->backlog
.gro_count
= 0;
5939 /* The loopback device is special if any other network devices
5940 * is present in a network namespace the loopback device must
5941 * be present. Since we now dynamically allocate and free the
5942 * loopback device ensure this invariant is maintained by
5943 * keeping the loopback device as the first device on the
5944 * list of network devices. Ensuring the loopback devices
5945 * is the first device that appears and the last network device
5948 if (register_pernet_device(&loopback_net_ops
))
5951 if (register_pernet_device(&default_device_ops
))
5954 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
5955 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
5957 hotcpu_notifier(dev_cpu_callback
, 0);
5965 subsys_initcall(net_dev_init
);
5967 static int __init
initialize_hashrnd(void)
5969 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
5973 late_initcall_sync(initialize_hashrnd
);