2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <linux/if_bridge.h>
105 #include <linux/if_macvlan.h>
107 #include <net/pkt_sched.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/kmod.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/wext.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <linux/pci.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 * The list of packet types we will receive (as opposed to discard)
145 * and the routines to invoke.
147 * Why 16. Because with 16 the only overlap we get on a hash of the
148 * low nibble of the protocol value is RARP/SNAP/X.25.
150 * NOTE: That is no longer true with the addition of VLAN tags. Not
151 * sure which should go first, but I bet it won't make much
152 * difference if we are running VLANs. The good news is that
153 * this protocol won't be in the list unless compiled in, so
154 * the average user (w/out VLANs) will not be adversely affected.
171 #define PTYPE_HASH_SIZE (16)
172 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
174 static DEFINE_SPINLOCK(ptype_lock
);
175 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
176 static struct list_head ptype_all __read_mostly
; /* Taps */
179 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
182 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
184 * Writers must hold the rtnl semaphore while they loop through the
185 * dev_base_head list, and hold dev_base_lock for writing when they do the
186 * actual updates. This allows pure readers to access the list even
187 * while a writer is preparing to update it.
189 * To put it another way, dev_base_lock is held for writing only to
190 * protect against pure readers; the rtnl semaphore provides the
191 * protection against other writers.
193 * See, for example usages, register_netdevice() and
194 * unregister_netdevice(), which must be called with the rtnl
197 DEFINE_RWLOCK(dev_base_lock
);
198 EXPORT_SYMBOL(dev_base_lock
);
200 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
202 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
203 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
206 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
208 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
211 static inline void rps_lock(struct softnet_data
*sd
)
214 spin_lock(&sd
->input_pkt_queue
.lock
);
218 static inline void rps_unlock(struct softnet_data
*sd
)
221 spin_unlock(&sd
->input_pkt_queue
.lock
);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device
*dev
)
228 struct net
*net
= dev_net(dev
);
232 write_lock_bh(&dev_base_lock
);
233 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
234 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
235 hlist_add_head_rcu(&dev
->index_hlist
,
236 dev_index_hash(net
, dev
->ifindex
));
237 write_unlock_bh(&dev_base_lock
);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device
*dev
)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock
);
250 list_del_rcu(&dev
->dev_list
);
251 hlist_del_rcu(&dev
->name_hlist
);
252 hlist_del_rcu(&dev
->index_hlist
);
253 write_unlock_bh(&dev_base_lock
);
260 static RAW_NOTIFIER_HEAD(netdev_chain
);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
268 EXPORT_PER_CPU_SYMBOL(softnet_data
);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type
[] =
276 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
277 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
278 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
279 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
280 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
281 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
282 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
283 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
284 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
285 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
286 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
287 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
288 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
289 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
290 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
291 ARPHRD_VOID
, ARPHRD_NONE
};
293 static const char *const netdev_lock_name
[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
312 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
318 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
319 if (netdev_lock_type
[i
] == dev_type
)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type
) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
326 unsigned short dev_type
)
330 i
= netdev_lock_pos(dev_type
);
331 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
332 netdev_lock_name
[i
]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
339 i
= netdev_lock_pos(dev
->type
);
340 lockdep_set_class_and_name(&dev
->addr_list_lock
,
341 &netdev_addr_lock_key
[i
],
342 netdev_lock_name
[i
]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
346 unsigned short dev_type
)
349 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type
*pt
)
393 spin_lock_bh(&ptype_lock
);
394 if (pt
->type
== htons(ETH_P_ALL
))
395 list_add_rcu(&pt
->list
, &ptype_all
);
397 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
398 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
400 spin_unlock_bh(&ptype_lock
);
402 EXPORT_SYMBOL(dev_add_pack
);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type
*pt
)
419 struct list_head
*head
;
420 struct packet_type
*pt1
;
422 spin_lock_bh(&ptype_lock
);
424 if (pt
->type
== htons(ETH_P_ALL
))
427 head
= &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
429 list_for_each_entry(pt1
, head
, list
) {
431 list_del_rcu(&pt
->list
);
436 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
438 spin_unlock_bh(&ptype_lock
);
440 EXPORT_SYMBOL(__dev_remove_pack
);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type
*pt
)
456 __dev_remove_pack(pt
);
460 EXPORT_SYMBOL(dev_remove_pack
);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
482 struct netdev_boot_setup
*s
;
486 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
487 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
488 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
489 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
490 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
495 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device
*dev
)
509 struct netdev_boot_setup
*s
= dev_boot_setup
;
512 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
513 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
514 !strcmp(dev
->name
, s
[i
].name
)) {
515 dev
->irq
= s
[i
].map
.irq
;
516 dev
->base_addr
= s
[i
].map
.base_addr
;
517 dev
->mem_start
= s
[i
].map
.mem_start
;
518 dev
->mem_end
= s
[i
].map
.mem_end
;
524 EXPORT_SYMBOL(netdev_boot_setup_check
);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix
, int unit
)
539 const struct netdev_boot_setup
*s
= dev_boot_setup
;
543 sprintf(name
, "%s%d", prefix
, unit
);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net
, name
))
552 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
553 if (!strcmp(name
, s
[i
].name
))
554 return s
[i
].map
.base_addr
;
559 * Saves at boot time configured settings for any netdevice.
561 int __init
netdev_boot_setup(char *str
)
566 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
571 memset(&map
, 0, sizeof(map
));
575 map
.base_addr
= ints
[2];
577 map
.mem_start
= ints
[3];
579 map
.mem_end
= ints
[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str
, &map
);
585 __setup("netdev=", netdev_boot_setup
);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
607 struct hlist_node
*p
;
608 struct net_device
*dev
;
609 struct hlist_head
*head
= dev_name_hash(net
, name
);
611 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
612 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
617 EXPORT_SYMBOL(__dev_get_by_name
);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
633 struct hlist_node
*p
;
634 struct net_device
*dev
;
635 struct hlist_head
*head
= dev_name_hash(net
, name
);
637 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
638 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
643 EXPORT_SYMBOL(dev_get_by_name_rcu
);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
659 struct net_device
*dev
;
662 dev
= dev_get_by_name_rcu(net
, name
);
668 EXPORT_SYMBOL(dev_get_by_name
);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
684 struct hlist_node
*p
;
685 struct net_device
*dev
;
686 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
688 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
689 if (dev
->ifindex
== ifindex
)
694 EXPORT_SYMBOL(__dev_get_by_index
);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
709 struct hlist_node
*p
;
710 struct net_device
*dev
;
711 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
713 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
714 if (dev
->ifindex
== ifindex
)
719 EXPORT_SYMBOL(dev_get_by_index_rcu
);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
735 struct net_device
*dev
;
738 dev
= dev_get_by_index_rcu(net
, ifindex
);
744 EXPORT_SYMBOL(dev_get_by_index
);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
763 struct net_device
*dev
;
767 for_each_netdev(net
, dev
)
768 if (dev
->type
== type
&&
769 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
774 EXPORT_SYMBOL(dev_getbyhwaddr
);
776 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
778 struct net_device
*dev
;
781 for_each_netdev(net
, dev
)
782 if (dev
->type
== type
)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
789 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
791 struct net_device
*dev
, *ret
= NULL
;
794 for_each_netdev_rcu(net
, dev
)
795 if (dev
->type
== type
) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
806 * dev_get_by_flags - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. The device returned has
813 * had a reference added and the pointer is safe until the user calls
814 * dev_put to indicate they have finished with it.
817 struct net_device
*dev_get_by_flags(struct net
*net
, unsigned short if_flags
,
820 struct net_device
*dev
, *ret
;
824 for_each_netdev_rcu(net
, dev
) {
825 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
834 EXPORT_SYMBOL(dev_get_by_flags
);
837 * dev_valid_name - check if name is okay for network device
840 * Network device names need to be valid file names to
841 * to allow sysfs to work. We also disallow any kind of
844 int dev_valid_name(const char *name
)
848 if (strlen(name
) >= IFNAMSIZ
)
850 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
854 if (*name
== '/' || isspace(*name
))
860 EXPORT_SYMBOL(dev_valid_name
);
863 * __dev_alloc_name - allocate a name for a device
864 * @net: network namespace to allocate the device name in
865 * @name: name format string
866 * @buf: scratch buffer and result name string
868 * Passed a format string - eg "lt%d" it will try and find a suitable
869 * id. It scans list of devices to build up a free map, then chooses
870 * the first empty slot. The caller must hold the dev_base or rtnl lock
871 * while allocating the name and adding the device in order to avoid
873 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
874 * Returns the number of the unit assigned or a negative errno code.
877 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
881 const int max_netdevices
= 8*PAGE_SIZE
;
882 unsigned long *inuse
;
883 struct net_device
*d
;
885 p
= strnchr(name
, IFNAMSIZ
-1, '%');
888 * Verify the string as this thing may have come from
889 * the user. There must be either one "%d" and no other "%"
892 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
895 /* Use one page as a bit array of possible slots */
896 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
900 for_each_netdev(net
, d
) {
901 if (!sscanf(d
->name
, name
, &i
))
903 if (i
< 0 || i
>= max_netdevices
)
906 /* avoid cases where sscanf is not exact inverse of printf */
907 snprintf(buf
, IFNAMSIZ
, name
, i
);
908 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
912 i
= find_first_zero_bit(inuse
, max_netdevices
);
913 free_page((unsigned long) inuse
);
917 snprintf(buf
, IFNAMSIZ
, name
, i
);
918 if (!__dev_get_by_name(net
, buf
))
921 /* It is possible to run out of possible slots
922 * when the name is long and there isn't enough space left
923 * for the digits, or if all bits are used.
929 * dev_alloc_name - allocate a name for a device
931 * @name: name format string
933 * Passed a format string - eg "lt%d" it will try and find a suitable
934 * id. It scans list of devices to build up a free map, then chooses
935 * the first empty slot. The caller must hold the dev_base or rtnl lock
936 * while allocating the name and adding the device in order to avoid
938 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
939 * Returns the number of the unit assigned or a negative errno code.
942 int dev_alloc_name(struct net_device
*dev
, const char *name
)
948 BUG_ON(!dev_net(dev
));
950 ret
= __dev_alloc_name(net
, name
, buf
);
952 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
955 EXPORT_SYMBOL(dev_alloc_name
);
957 static int dev_get_valid_name(struct net_device
*dev
, const char *name
, bool fmt
)
961 BUG_ON(!dev_net(dev
));
964 if (!dev_valid_name(name
))
967 if (fmt
&& strchr(name
, '%'))
968 return dev_alloc_name(dev
, name
);
969 else if (__dev_get_by_name(net
, name
))
971 else if (dev
->name
!= name
)
972 strlcpy(dev
->name
, name
, IFNAMSIZ
);
978 * dev_change_name - change name of a device
980 * @newname: name (or format string) must be at least IFNAMSIZ
982 * Change name of a device, can pass format strings "eth%d".
985 int dev_change_name(struct net_device
*dev
, const char *newname
)
987 char oldname
[IFNAMSIZ
];
993 BUG_ON(!dev_net(dev
));
996 if (dev
->flags
& IFF_UP
)
999 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
1002 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1004 err
= dev_get_valid_name(dev
, newname
, 1);
1009 ret
= device_rename(&dev
->dev
, dev
->name
);
1011 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1015 write_lock_bh(&dev_base_lock
);
1016 hlist_del(&dev
->name_hlist
);
1017 write_unlock_bh(&dev_base_lock
);
1021 write_lock_bh(&dev_base_lock
);
1022 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1023 write_unlock_bh(&dev_base_lock
);
1025 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1026 ret
= notifier_to_errno(ret
);
1029 /* err >= 0 after dev_alloc_name() or stores the first errno */
1032 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1036 "%s: name change rollback failed: %d.\n",
1045 * dev_set_alias - change ifalias of a device
1047 * @alias: name up to IFALIASZ
1048 * @len: limit of bytes to copy from info
1050 * Set ifalias for a device,
1052 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1056 if (len
>= IFALIASZ
)
1061 kfree(dev
->ifalias
);
1062 dev
->ifalias
= NULL
;
1067 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1071 strlcpy(dev
->ifalias
, alias
, len
+1);
1077 * netdev_features_change - device changes features
1078 * @dev: device to cause notification
1080 * Called to indicate a device has changed features.
1082 void netdev_features_change(struct net_device
*dev
)
1084 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1086 EXPORT_SYMBOL(netdev_features_change
);
1089 * netdev_state_change - device changes state
1090 * @dev: device to cause notification
1092 * Called to indicate a device has changed state. This function calls
1093 * the notifier chains for netdev_chain and sends a NEWLINK message
1094 * to the routing socket.
1096 void netdev_state_change(struct net_device
*dev
)
1098 if (dev
->flags
& IFF_UP
) {
1099 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1100 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1103 EXPORT_SYMBOL(netdev_state_change
);
1105 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1107 return call_netdevice_notifiers(event
, dev
);
1109 EXPORT_SYMBOL(netdev_bonding_change
);
1112 * dev_load - load a network module
1113 * @net: the applicable net namespace
1114 * @name: name of interface
1116 * If a network interface is not present and the process has suitable
1117 * privileges this function loads the module. If module loading is not
1118 * available in this kernel then it becomes a nop.
1121 void dev_load(struct net
*net
, const char *name
)
1123 struct net_device
*dev
;
1126 dev
= dev_get_by_name_rcu(net
, name
);
1129 if (!dev
&& capable(CAP_NET_ADMIN
))
1130 request_module("%s", name
);
1132 EXPORT_SYMBOL(dev_load
);
1134 static int __dev_open(struct net_device
*dev
)
1136 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1142 * Is it even present?
1144 if (!netif_device_present(dev
))
1147 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1148 ret
= notifier_to_errno(ret
);
1153 * Call device private open method
1155 set_bit(__LINK_STATE_START
, &dev
->state
);
1157 if (ops
->ndo_validate_addr
)
1158 ret
= ops
->ndo_validate_addr(dev
);
1160 if (!ret
&& ops
->ndo_open
)
1161 ret
= ops
->ndo_open(dev
);
1164 * If it went open OK then:
1168 clear_bit(__LINK_STATE_START
, &dev
->state
);
1173 dev
->flags
|= IFF_UP
;
1178 net_dmaengine_get();
1181 * Initialize multicasting status
1183 dev_set_rx_mode(dev
);
1186 * Wakeup transmit queue engine
1195 * dev_open - prepare an interface for use.
1196 * @dev: device to open
1198 * Takes a device from down to up state. The device's private open
1199 * function is invoked and then the multicast lists are loaded. Finally
1200 * the device is moved into the up state and a %NETDEV_UP message is
1201 * sent to the netdev notifier chain.
1203 * Calling this function on an active interface is a nop. On a failure
1204 * a negative errno code is returned.
1206 int dev_open(struct net_device
*dev
)
1213 if (dev
->flags
& IFF_UP
)
1219 ret
= __dev_open(dev
);
1224 * ... and announce new interface.
1226 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1227 call_netdevice_notifiers(NETDEV_UP
, dev
);
1231 EXPORT_SYMBOL(dev_open
);
1233 static int __dev_close(struct net_device
*dev
)
1235 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1241 * Tell people we are going down, so that they can
1242 * prepare to death, when device is still operating.
1244 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1246 clear_bit(__LINK_STATE_START
, &dev
->state
);
1248 /* Synchronize to scheduled poll. We cannot touch poll list,
1249 * it can be even on different cpu. So just clear netif_running().
1251 * dev->stop() will invoke napi_disable() on all of it's
1252 * napi_struct instances on this device.
1254 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1256 dev_deactivate(dev
);
1259 * Call the device specific close. This cannot fail.
1260 * Only if device is UP
1262 * We allow it to be called even after a DETACH hot-plug
1269 * Device is now down.
1272 dev
->flags
&= ~IFF_UP
;
1277 net_dmaengine_put();
1283 * dev_close - shutdown an interface.
1284 * @dev: device to shutdown
1286 * This function moves an active device into down state. A
1287 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1288 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1291 int dev_close(struct net_device
*dev
)
1293 if (!(dev
->flags
& IFF_UP
))
1299 * Tell people we are down
1301 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1302 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1306 EXPORT_SYMBOL(dev_close
);
1310 * dev_disable_lro - disable Large Receive Offload on a device
1313 * Disable Large Receive Offload (LRO) on a net device. Must be
1314 * called under RTNL. This is needed if received packets may be
1315 * forwarded to another interface.
1317 void dev_disable_lro(struct net_device
*dev
)
1319 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1320 dev
->ethtool_ops
->set_flags
) {
1321 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1322 if (flags
& ETH_FLAG_LRO
) {
1323 flags
&= ~ETH_FLAG_LRO
;
1324 dev
->ethtool_ops
->set_flags(dev
, flags
);
1327 WARN_ON(dev
->features
& NETIF_F_LRO
);
1329 EXPORT_SYMBOL(dev_disable_lro
);
1332 static int dev_boot_phase
= 1;
1335 * Device change register/unregister. These are not inline or static
1336 * as we export them to the world.
1340 * register_netdevice_notifier - register a network notifier block
1343 * Register a notifier to be called when network device events occur.
1344 * The notifier passed is linked into the kernel structures and must
1345 * not be reused until it has been unregistered. A negative errno code
1346 * is returned on a failure.
1348 * When registered all registration and up events are replayed
1349 * to the new notifier to allow device to have a race free
1350 * view of the network device list.
1353 int register_netdevice_notifier(struct notifier_block
*nb
)
1355 struct net_device
*dev
;
1356 struct net_device
*last
;
1361 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1367 for_each_netdev(net
, dev
) {
1368 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1369 err
= notifier_to_errno(err
);
1373 if (!(dev
->flags
& IFF_UP
))
1376 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1387 for_each_netdev(net
, dev
) {
1391 if (dev
->flags
& IFF_UP
) {
1392 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1393 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1395 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1396 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1400 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1403 EXPORT_SYMBOL(register_netdevice_notifier
);
1406 * unregister_netdevice_notifier - unregister a network notifier block
1409 * Unregister a notifier previously registered by
1410 * register_netdevice_notifier(). The notifier is unlinked into the
1411 * kernel structures and may then be reused. A negative errno code
1412 * is returned on a failure.
1415 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1420 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1424 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1427 * call_netdevice_notifiers - call all network notifier blocks
1428 * @val: value passed unmodified to notifier function
1429 * @dev: net_device pointer passed unmodified to notifier function
1431 * Call all network notifier blocks. Parameters and return value
1432 * are as for raw_notifier_call_chain().
1435 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1438 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1441 /* When > 0 there are consumers of rx skb time stamps */
1442 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1444 void net_enable_timestamp(void)
1446 atomic_inc(&netstamp_needed
);
1448 EXPORT_SYMBOL(net_enable_timestamp
);
1450 void net_disable_timestamp(void)
1452 atomic_dec(&netstamp_needed
);
1454 EXPORT_SYMBOL(net_disable_timestamp
);
1456 static inline void net_timestamp_set(struct sk_buff
*skb
)
1458 if (atomic_read(&netstamp_needed
))
1459 __net_timestamp(skb
);
1461 skb
->tstamp
.tv64
= 0;
1464 static inline void net_timestamp_check(struct sk_buff
*skb
)
1466 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1467 __net_timestamp(skb
);
1471 * dev_forward_skb - loopback an skb to another netif
1473 * @dev: destination network device
1474 * @skb: buffer to forward
1477 * NET_RX_SUCCESS (no congestion)
1478 * NET_RX_DROP (packet was dropped, but freed)
1480 * dev_forward_skb can be used for injecting an skb from the
1481 * start_xmit function of one device into the receive queue
1482 * of another device.
1484 * The receiving device may be in another namespace, so
1485 * we have to clear all information in the skb that could
1486 * impact namespace isolation.
1488 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1492 if (!(dev
->flags
& IFF_UP
) ||
1493 (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
))) {
1497 skb_set_dev(skb
, dev
);
1498 skb
->tstamp
.tv64
= 0;
1499 skb
->pkt_type
= PACKET_HOST
;
1500 skb
->protocol
= eth_type_trans(skb
, dev
);
1501 return netif_rx(skb
);
1503 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1506 * Support routine. Sends outgoing frames to any network
1507 * taps currently in use.
1510 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1512 struct packet_type
*ptype
;
1514 #ifdef CONFIG_NET_CLS_ACT
1515 if (!(skb
->tstamp
.tv64
&& (G_TC_FROM(skb
->tc_verd
) & AT_INGRESS
)))
1516 net_timestamp_set(skb
);
1518 net_timestamp_set(skb
);
1522 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1523 /* Never send packets back to the socket
1524 * they originated from - MvS (miquels@drinkel.ow.org)
1526 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1527 (ptype
->af_packet_priv
== NULL
||
1528 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1529 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1533 /* skb->nh should be correctly
1534 set by sender, so that the second statement is
1535 just protection against buggy protocols.
1537 skb_reset_mac_header(skb2
);
1539 if (skb_network_header(skb2
) < skb2
->data
||
1540 skb2
->network_header
> skb2
->tail
) {
1541 if (net_ratelimit())
1542 printk(KERN_CRIT
"protocol %04x is "
1544 skb2
->protocol
, dev
->name
);
1545 skb_reset_network_header(skb2
);
1548 skb2
->transport_header
= skb2
->network_header
;
1549 skb2
->pkt_type
= PACKET_OUTGOING
;
1550 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1557 static inline void __netif_reschedule(struct Qdisc
*q
)
1559 struct softnet_data
*sd
;
1560 unsigned long flags
;
1562 local_irq_save(flags
);
1563 sd
= &__get_cpu_var(softnet_data
);
1564 q
->next_sched
= NULL
;
1565 *sd
->output_queue_tailp
= q
;
1566 sd
->output_queue_tailp
= &q
->next_sched
;
1567 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1568 local_irq_restore(flags
);
1571 void __netif_schedule(struct Qdisc
*q
)
1573 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1574 __netif_reschedule(q
);
1576 EXPORT_SYMBOL(__netif_schedule
);
1578 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1580 if (!skb
->destructor
)
1582 else if (atomic_dec_and_test(&skb
->users
)) {
1583 struct softnet_data
*sd
;
1584 unsigned long flags
;
1586 local_irq_save(flags
);
1587 sd
= &__get_cpu_var(softnet_data
);
1588 skb
->next
= sd
->completion_queue
;
1589 sd
->completion_queue
= skb
;
1590 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1591 local_irq_restore(flags
);
1594 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1596 void dev_kfree_skb_any(struct sk_buff
*skb
)
1598 if (in_irq() || irqs_disabled())
1599 dev_kfree_skb_irq(skb
);
1603 EXPORT_SYMBOL(dev_kfree_skb_any
);
1607 * netif_device_detach - mark device as removed
1608 * @dev: network device
1610 * Mark device as removed from system and therefore no longer available.
1612 void netif_device_detach(struct net_device
*dev
)
1614 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1615 netif_running(dev
)) {
1616 netif_tx_stop_all_queues(dev
);
1619 EXPORT_SYMBOL(netif_device_detach
);
1622 * netif_device_attach - mark device as attached
1623 * @dev: network device
1625 * Mark device as attached from system and restart if needed.
1627 void netif_device_attach(struct net_device
*dev
)
1629 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1630 netif_running(dev
)) {
1631 netif_tx_wake_all_queues(dev
);
1632 __netdev_watchdog_up(dev
);
1635 EXPORT_SYMBOL(netif_device_attach
);
1637 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1639 return ((features
& NETIF_F_GEN_CSUM
) ||
1640 ((features
& NETIF_F_IP_CSUM
) &&
1641 protocol
== htons(ETH_P_IP
)) ||
1642 ((features
& NETIF_F_IPV6_CSUM
) &&
1643 protocol
== htons(ETH_P_IPV6
)) ||
1644 ((features
& NETIF_F_FCOE_CRC
) &&
1645 protocol
== htons(ETH_P_FCOE
)));
1648 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1650 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1653 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1654 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1655 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1656 veh
->h_vlan_encapsulated_proto
))
1664 * skb_dev_set -- assign a new device to a buffer
1665 * @skb: buffer for the new device
1666 * @dev: network device
1668 * If an skb is owned by a device already, we have to reset
1669 * all data private to the namespace a device belongs to
1670 * before assigning it a new device.
1672 #ifdef CONFIG_NET_NS
1673 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1676 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1679 skb_init_secmark(skb
);
1683 skb
->ipvs_property
= 0;
1684 #ifdef CONFIG_NET_SCHED
1690 EXPORT_SYMBOL(skb_set_dev
);
1691 #endif /* CONFIG_NET_NS */
1694 * Invalidate hardware checksum when packet is to be mangled, and
1695 * complete checksum manually on outgoing path.
1697 int skb_checksum_help(struct sk_buff
*skb
)
1700 int ret
= 0, offset
;
1702 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1703 goto out_set_summed
;
1705 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1706 /* Let GSO fix up the checksum. */
1707 goto out_set_summed
;
1710 offset
= skb
->csum_start
- skb_headroom(skb
);
1711 BUG_ON(offset
>= skb_headlen(skb
));
1712 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1714 offset
+= skb
->csum_offset
;
1715 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1717 if (skb_cloned(skb
) &&
1718 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1719 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1724 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1726 skb
->ip_summed
= CHECKSUM_NONE
;
1730 EXPORT_SYMBOL(skb_checksum_help
);
1733 * skb_gso_segment - Perform segmentation on skb.
1734 * @skb: buffer to segment
1735 * @features: features for the output path (see dev->features)
1737 * This function segments the given skb and returns a list of segments.
1739 * It may return NULL if the skb requires no segmentation. This is
1740 * only possible when GSO is used for verifying header integrity.
1742 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1744 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1745 struct packet_type
*ptype
;
1746 __be16 type
= skb
->protocol
;
1749 skb_reset_mac_header(skb
);
1750 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1751 __skb_pull(skb
, skb
->mac_len
);
1753 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1754 struct net_device
*dev
= skb
->dev
;
1755 struct ethtool_drvinfo info
= {};
1757 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1758 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1760 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1762 info
.driver
, dev
? dev
->features
: 0L,
1763 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1764 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1766 if (skb_header_cloned(skb
) &&
1767 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1768 return ERR_PTR(err
);
1772 list_for_each_entry_rcu(ptype
,
1773 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1774 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1775 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1776 err
= ptype
->gso_send_check(skb
);
1777 segs
= ERR_PTR(err
);
1778 if (err
|| skb_gso_ok(skb
, features
))
1780 __skb_push(skb
, (skb
->data
-
1781 skb_network_header(skb
)));
1783 segs
= ptype
->gso_segment(skb
, features
);
1789 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1793 EXPORT_SYMBOL(skb_gso_segment
);
1795 /* Take action when hardware reception checksum errors are detected. */
1797 void netdev_rx_csum_fault(struct net_device
*dev
)
1799 if (net_ratelimit()) {
1800 printk(KERN_ERR
"%s: hw csum failure.\n",
1801 dev
? dev
->name
: "<unknown>");
1805 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1808 /* Actually, we should eliminate this check as soon as we know, that:
1809 * 1. IOMMU is present and allows to map all the memory.
1810 * 2. No high memory really exists on this machine.
1813 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1815 #ifdef CONFIG_HIGHMEM
1817 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1818 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1819 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1823 if (PCI_DMA_BUS_IS_PHYS
) {
1824 struct device
*pdev
= dev
->dev
.parent
;
1828 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1829 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1830 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1839 void (*destructor
)(struct sk_buff
*skb
);
1842 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1844 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1846 struct dev_gso_cb
*cb
;
1849 struct sk_buff
*nskb
= skb
->next
;
1851 skb
->next
= nskb
->next
;
1854 } while (skb
->next
);
1856 cb
= DEV_GSO_CB(skb
);
1858 cb
->destructor(skb
);
1862 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1863 * @skb: buffer to segment
1865 * This function segments the given skb and stores the list of segments
1868 static int dev_gso_segment(struct sk_buff
*skb
)
1870 struct net_device
*dev
= skb
->dev
;
1871 struct sk_buff
*segs
;
1872 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1875 segs
= skb_gso_segment(skb
, features
);
1877 /* Verifying header integrity only. */
1882 return PTR_ERR(segs
);
1885 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1886 skb
->destructor
= dev_gso_skb_destructor
;
1892 * Try to orphan skb early, right before transmission by the device.
1893 * We cannot orphan skb if tx timestamp is requested, since
1894 * drivers need to call skb_tstamp_tx() to send the timestamp.
1896 static inline void skb_orphan_try(struct sk_buff
*skb
)
1898 if (!skb_tx(skb
)->flags
)
1902 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1903 struct netdev_queue
*txq
)
1905 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1906 int rc
= NETDEV_TX_OK
;
1908 if (likely(!skb
->next
)) {
1909 if (!list_empty(&ptype_all
))
1910 dev_queue_xmit_nit(skb
, dev
);
1913 * If device doesnt need skb->dst, release it right now while
1914 * its hot in this cpu cache
1916 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1919 skb_orphan_try(skb
);
1921 if (netif_needs_gso(dev
, skb
)) {
1922 if (unlikely(dev_gso_segment(skb
)))
1928 rc
= ops
->ndo_start_xmit(skb
, dev
);
1929 if (rc
== NETDEV_TX_OK
)
1930 txq_trans_update(txq
);
1936 struct sk_buff
*nskb
= skb
->next
;
1938 skb
->next
= nskb
->next
;
1942 * If device doesnt need nskb->dst, release it right now while
1943 * its hot in this cpu cache
1945 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1948 rc
= ops
->ndo_start_xmit(nskb
, dev
);
1949 if (unlikely(rc
!= NETDEV_TX_OK
)) {
1950 if (rc
& ~NETDEV_TX_MASK
)
1951 goto out_kfree_gso_skb
;
1952 nskb
->next
= skb
->next
;
1956 txq_trans_update(txq
);
1957 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
1958 return NETDEV_TX_BUSY
;
1959 } while (skb
->next
);
1962 if (likely(skb
->next
== NULL
))
1963 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1969 static u32 hashrnd __read_mostly
;
1971 u16
skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
)
1975 if (skb_rx_queue_recorded(skb
)) {
1976 hash
= skb_get_rx_queue(skb
);
1977 while (unlikely(hash
>= dev
->real_num_tx_queues
))
1978 hash
-= dev
->real_num_tx_queues
;
1982 if (skb
->sk
&& skb
->sk
->sk_hash
)
1983 hash
= skb
->sk
->sk_hash
;
1985 hash
= (__force u16
) skb
->protocol
;
1987 hash
= jhash_1word(hash
, hashrnd
);
1989 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
1991 EXPORT_SYMBOL(skb_tx_hash
);
1993 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
1995 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
1996 if (net_ratelimit()) {
1997 pr_warning("%s selects TX queue %d, but "
1998 "real number of TX queues is %d\n",
1999 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2006 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2007 struct sk_buff
*skb
)
2010 struct sock
*sk
= skb
->sk
;
2012 if (sk_tx_queue_recorded(sk
)) {
2013 queue_index
= sk_tx_queue_get(sk
);
2015 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2017 if (ops
->ndo_select_queue
) {
2018 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2019 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2022 if (dev
->real_num_tx_queues
> 1)
2023 queue_index
= skb_tx_hash(dev
, skb
);
2026 struct dst_entry
*dst
= rcu_dereference_check(sk
->sk_dst_cache
, 1);
2028 if (dst
&& skb_dst(skb
) == dst
)
2029 sk_tx_queue_set(sk
, queue_index
);
2034 skb_set_queue_mapping(skb
, queue_index
);
2035 return netdev_get_tx_queue(dev
, queue_index
);
2038 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2039 struct net_device
*dev
,
2040 struct netdev_queue
*txq
)
2042 spinlock_t
*root_lock
= qdisc_lock(q
);
2043 bool contended
= qdisc_is_running(q
);
2047 * Heuristic to force contended enqueues to serialize on a
2048 * separate lock before trying to get qdisc main lock.
2049 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2050 * and dequeue packets faster.
2052 if (unlikely(contended
))
2053 spin_lock(&q
->busylock
);
2055 spin_lock(root_lock
);
2056 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2059 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2060 qdisc_run_begin(q
)) {
2062 * This is a work-conserving queue; there are no old skbs
2063 * waiting to be sent out; and the qdisc is not running -
2064 * xmit the skb directly.
2066 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2068 __qdisc_update_bstats(q
, skb
->len
);
2069 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2070 if (unlikely(contended
)) {
2071 spin_unlock(&q
->busylock
);
2078 rc
= NET_XMIT_SUCCESS
;
2081 rc
= qdisc_enqueue_root(skb
, q
);
2082 if (qdisc_run_begin(q
)) {
2083 if (unlikely(contended
)) {
2084 spin_unlock(&q
->busylock
);
2090 spin_unlock(root_lock
);
2091 if (unlikely(contended
))
2092 spin_unlock(&q
->busylock
);
2097 * Returns true if either:
2098 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2099 * 2. skb is fragmented and the device does not support SG, or if
2100 * at least one of fragments is in highmem and device does not
2101 * support DMA from it.
2103 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2104 struct net_device
*dev
)
2106 return skb_is_nonlinear(skb
) &&
2107 ((skb_has_frags(skb
) && !(dev
->features
& NETIF_F_FRAGLIST
)) ||
2108 (skb_shinfo(skb
)->nr_frags
&& (!(dev
->features
& NETIF_F_SG
) ||
2109 illegal_highdma(dev
, skb
))));
2113 * dev_queue_xmit - transmit a buffer
2114 * @skb: buffer to transmit
2116 * Queue a buffer for transmission to a network device. The caller must
2117 * have set the device and priority and built the buffer before calling
2118 * this function. The function can be called from an interrupt.
2120 * A negative errno code is returned on a failure. A success does not
2121 * guarantee the frame will be transmitted as it may be dropped due
2122 * to congestion or traffic shaping.
2124 * -----------------------------------------------------------------------------------
2125 * I notice this method can also return errors from the queue disciplines,
2126 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2129 * Regardless of the return value, the skb is consumed, so it is currently
2130 * difficult to retry a send to this method. (You can bump the ref count
2131 * before sending to hold a reference for retry if you are careful.)
2133 * When calling this method, interrupts MUST be enabled. This is because
2134 * the BH enable code must have IRQs enabled so that it will not deadlock.
2137 int dev_queue_xmit(struct sk_buff
*skb
)
2139 struct net_device
*dev
= skb
->dev
;
2140 struct netdev_queue
*txq
;
2144 /* GSO will handle the following emulations directly. */
2145 if (netif_needs_gso(dev
, skb
))
2148 /* Convert a paged skb to linear, if required */
2149 if (skb_needs_linearize(skb
, dev
) && __skb_linearize(skb
))
2152 /* If packet is not checksummed and device does not support
2153 * checksumming for this protocol, complete checksumming here.
2155 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2156 skb_set_transport_header(skb
, skb
->csum_start
-
2158 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
2163 /* Disable soft irqs for various locks below. Also
2164 * stops preemption for RCU.
2168 txq
= dev_pick_tx(dev
, skb
);
2169 q
= rcu_dereference_bh(txq
->qdisc
);
2171 #ifdef CONFIG_NET_CLS_ACT
2172 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2175 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2179 /* The device has no queue. Common case for software devices:
2180 loopback, all the sorts of tunnels...
2182 Really, it is unlikely that netif_tx_lock protection is necessary
2183 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2185 However, it is possible, that they rely on protection
2188 Check this and shot the lock. It is not prone from deadlocks.
2189 Either shot noqueue qdisc, it is even simpler 8)
2191 if (dev
->flags
& IFF_UP
) {
2192 int cpu
= smp_processor_id(); /* ok because BHs are off */
2194 if (txq
->xmit_lock_owner
!= cpu
) {
2196 HARD_TX_LOCK(dev
, txq
, cpu
);
2198 if (!netif_tx_queue_stopped(txq
)) {
2199 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2200 if (dev_xmit_complete(rc
)) {
2201 HARD_TX_UNLOCK(dev
, txq
);
2205 HARD_TX_UNLOCK(dev
, txq
);
2206 if (net_ratelimit())
2207 printk(KERN_CRIT
"Virtual device %s asks to "
2208 "queue packet!\n", dev
->name
);
2210 /* Recursion is detected! It is possible,
2212 if (net_ratelimit())
2213 printk(KERN_CRIT
"Dead loop on virtual device "
2214 "%s, fix it urgently!\n", dev
->name
);
2219 rcu_read_unlock_bh();
2225 rcu_read_unlock_bh();
2228 EXPORT_SYMBOL(dev_queue_xmit
);
2231 /*=======================================================================
2233 =======================================================================*/
2235 int netdev_max_backlog __read_mostly
= 1000;
2236 int netdev_tstamp_prequeue __read_mostly
= 1;
2237 int netdev_budget __read_mostly
= 300;
2238 int weight_p __read_mostly
= 64; /* old backlog weight */
2240 /* Called with irq disabled */
2241 static inline void ____napi_schedule(struct softnet_data
*sd
,
2242 struct napi_struct
*napi
)
2244 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2245 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2250 /* One global table that all flow-based protocols share. */
2251 struct rps_sock_flow_table
*rps_sock_flow_table __read_mostly
;
2252 EXPORT_SYMBOL(rps_sock_flow_table
);
2255 * get_rps_cpu is called from netif_receive_skb and returns the target
2256 * CPU from the RPS map of the receiving queue for a given skb.
2257 * rcu_read_lock must be held on entry.
2259 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2260 struct rps_dev_flow
**rflowp
)
2262 struct ipv6hdr
*ip6
;
2264 struct netdev_rx_queue
*rxqueue
;
2265 struct rps_map
*map
;
2266 struct rps_dev_flow_table
*flow_table
;
2267 struct rps_sock_flow_table
*sock_flow_table
;
2271 u32 addr1
, addr2
, ihl
;
2277 if (skb_rx_queue_recorded(skb
)) {
2278 u16 index
= skb_get_rx_queue(skb
);
2279 if (unlikely(index
>= dev
->num_rx_queues
)) {
2280 if (net_ratelimit()) {
2281 pr_warning("%s received packet on queue "
2282 "%u, but number of RX queues is %u\n",
2283 dev
->name
, index
, dev
->num_rx_queues
);
2287 rxqueue
= dev
->_rx
+ index
;
2291 if (!rxqueue
->rps_map
&& !rxqueue
->rps_flow_table
)
2295 goto got_hash
; /* Skip hash computation on packet header */
2297 switch (skb
->protocol
) {
2298 case __constant_htons(ETH_P_IP
):
2299 if (!pskb_may_pull(skb
, sizeof(*ip
)))
2302 ip
= (struct iphdr
*) skb
->data
;
2303 ip_proto
= ip
->protocol
;
2304 addr1
= (__force u32
) ip
->saddr
;
2305 addr2
= (__force u32
) ip
->daddr
;
2308 case __constant_htons(ETH_P_IPV6
):
2309 if (!pskb_may_pull(skb
, sizeof(*ip6
)))
2312 ip6
= (struct ipv6hdr
*) skb
->data
;
2313 ip_proto
= ip6
->nexthdr
;
2314 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2315 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2328 case IPPROTO_UDPLITE
:
2329 if (pskb_may_pull(skb
, (ihl
* 4) + 4)) {
2330 ports
.v32
= * (__force u32
*) (skb
->data
+ (ihl
* 4));
2331 if (ports
.v16
[1] < ports
.v16
[0])
2332 swap(ports
.v16
[0], ports
.v16
[1]);
2340 /* get a consistent hash (same value on both flow directions) */
2343 skb
->rxhash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2348 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2349 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2350 if (flow_table
&& sock_flow_table
) {
2352 struct rps_dev_flow
*rflow
;
2354 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2357 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2358 sock_flow_table
->mask
];
2361 * If the desired CPU (where last recvmsg was done) is
2362 * different from current CPU (one in the rx-queue flow
2363 * table entry), switch if one of the following holds:
2364 * - Current CPU is unset (equal to RPS_NO_CPU).
2365 * - Current CPU is offline.
2366 * - The current CPU's queue tail has advanced beyond the
2367 * last packet that was enqueued using this table entry.
2368 * This guarantees that all previous packets for the flow
2369 * have been dequeued, thus preserving in order delivery.
2371 if (unlikely(tcpu
!= next_cpu
) &&
2372 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2373 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2374 rflow
->last_qtail
)) >= 0)) {
2375 tcpu
= rflow
->cpu
= next_cpu
;
2376 if (tcpu
!= RPS_NO_CPU
)
2377 rflow
->last_qtail
= per_cpu(softnet_data
,
2378 tcpu
).input_queue_head
;
2380 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2387 map
= rcu_dereference(rxqueue
->rps_map
);
2389 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2391 if (cpu_online(tcpu
)) {
2401 /* Called from hardirq (IPI) context */
2402 static void rps_trigger_softirq(void *data
)
2404 struct softnet_data
*sd
= data
;
2406 ____napi_schedule(sd
, &sd
->backlog
);
2410 #endif /* CONFIG_RPS */
2413 * Check if this softnet_data structure is another cpu one
2414 * If yes, queue it to our IPI list and return 1
2417 static int rps_ipi_queued(struct softnet_data
*sd
)
2420 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2423 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2424 mysd
->rps_ipi_list
= sd
;
2426 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2429 #endif /* CONFIG_RPS */
2434 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2435 * queue (may be a remote CPU queue).
2437 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2438 unsigned int *qtail
)
2440 struct softnet_data
*sd
;
2441 unsigned long flags
;
2443 sd
= &per_cpu(softnet_data
, cpu
);
2445 local_irq_save(flags
);
2448 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2449 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2451 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2452 input_queue_tail_incr_save(sd
, qtail
);
2454 local_irq_restore(flags
);
2455 return NET_RX_SUCCESS
;
2458 /* Schedule NAPI for backlog device
2459 * We can use non atomic operation since we own the queue lock
2461 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2462 if (!rps_ipi_queued(sd
))
2463 ____napi_schedule(sd
, &sd
->backlog
);
2471 local_irq_restore(flags
);
2478 * netif_rx - post buffer to the network code
2479 * @skb: buffer to post
2481 * This function receives a packet from a device driver and queues it for
2482 * the upper (protocol) levels to process. It always succeeds. The buffer
2483 * may be dropped during processing for congestion control or by the
2487 * NET_RX_SUCCESS (no congestion)
2488 * NET_RX_DROP (packet was dropped)
2492 int netif_rx(struct sk_buff
*skb
)
2496 /* if netpoll wants it, pretend we never saw it */
2497 if (netpoll_rx(skb
))
2500 if (netdev_tstamp_prequeue
)
2501 net_timestamp_check(skb
);
2505 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2510 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2512 cpu
= smp_processor_id();
2514 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2521 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2527 EXPORT_SYMBOL(netif_rx
);
2529 int netif_rx_ni(struct sk_buff
*skb
)
2534 err
= netif_rx(skb
);
2535 if (local_softirq_pending())
2541 EXPORT_SYMBOL(netif_rx_ni
);
2543 static void net_tx_action(struct softirq_action
*h
)
2545 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2547 if (sd
->completion_queue
) {
2548 struct sk_buff
*clist
;
2550 local_irq_disable();
2551 clist
= sd
->completion_queue
;
2552 sd
->completion_queue
= NULL
;
2556 struct sk_buff
*skb
= clist
;
2557 clist
= clist
->next
;
2559 WARN_ON(atomic_read(&skb
->users
));
2564 if (sd
->output_queue
) {
2567 local_irq_disable();
2568 head
= sd
->output_queue
;
2569 sd
->output_queue
= NULL
;
2570 sd
->output_queue_tailp
= &sd
->output_queue
;
2574 struct Qdisc
*q
= head
;
2575 spinlock_t
*root_lock
;
2577 head
= head
->next_sched
;
2579 root_lock
= qdisc_lock(q
);
2580 if (spin_trylock(root_lock
)) {
2581 smp_mb__before_clear_bit();
2582 clear_bit(__QDISC_STATE_SCHED
,
2585 spin_unlock(root_lock
);
2587 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2589 __netif_reschedule(q
);
2591 smp_mb__before_clear_bit();
2592 clear_bit(__QDISC_STATE_SCHED
,
2600 static inline int deliver_skb(struct sk_buff
*skb
,
2601 struct packet_type
*pt_prev
,
2602 struct net_device
*orig_dev
)
2604 atomic_inc(&skb
->users
);
2605 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2608 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2609 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2610 /* This hook is defined here for ATM LANE */
2611 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2612 unsigned char *addr
) __read_mostly
;
2613 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2616 #ifdef CONFIG_NET_CLS_ACT
2617 /* TODO: Maybe we should just force sch_ingress to be compiled in
2618 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2619 * a compare and 2 stores extra right now if we dont have it on
2620 * but have CONFIG_NET_CLS_ACT
2621 * NOTE: This doesnt stop any functionality; if you dont have
2622 * the ingress scheduler, you just cant add policies on ingress.
2625 static int ing_filter(struct sk_buff
*skb
)
2627 struct net_device
*dev
= skb
->dev
;
2628 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2629 struct netdev_queue
*rxq
;
2630 int result
= TC_ACT_OK
;
2633 if (MAX_RED_LOOP
< ttl
++) {
2635 "Redir loop detected Dropping packet (%d->%d)\n",
2636 skb
->skb_iif
, dev
->ifindex
);
2640 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2641 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2643 rxq
= &dev
->rx_queue
;
2646 if (q
!= &noop_qdisc
) {
2647 spin_lock(qdisc_lock(q
));
2648 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2649 result
= qdisc_enqueue_root(skb
, q
);
2650 spin_unlock(qdisc_lock(q
));
2656 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2657 struct packet_type
**pt_prev
,
2658 int *ret
, struct net_device
*orig_dev
)
2660 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2664 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2667 /* Huh? Why does turning on AF_PACKET affect this? */
2668 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2671 switch (ing_filter(skb
)) {
2685 * netif_nit_deliver - deliver received packets to network taps
2688 * This function is used to deliver incoming packets to network
2689 * taps. It should be used when the normal netif_receive_skb path
2690 * is bypassed, for example because of VLAN acceleration.
2692 void netif_nit_deliver(struct sk_buff
*skb
)
2694 struct packet_type
*ptype
;
2696 if (list_empty(&ptype_all
))
2699 skb_reset_network_header(skb
);
2700 skb_reset_transport_header(skb
);
2701 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2704 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2705 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2706 deliver_skb(skb
, ptype
, skb
->dev
);
2712 * netdev_rx_handler_register - register receive handler
2713 * @dev: device to register a handler for
2714 * @rx_handler: receive handler to register
2716 * Register a receive hander for a device. This handler will then be
2717 * called from __netif_receive_skb. A negative errno code is returned
2720 * The caller must hold the rtnl_mutex.
2722 int netdev_rx_handler_register(struct net_device
*dev
,
2723 rx_handler_func_t
*rx_handler
)
2727 if (dev
->rx_handler
)
2730 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
2734 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
2737 * netdev_rx_handler_unregister - unregister receive handler
2738 * @dev: device to unregister a handler from
2740 * Unregister a receive hander from a device.
2742 * The caller must hold the rtnl_mutex.
2744 void netdev_rx_handler_unregister(struct net_device
*dev
)
2748 rcu_assign_pointer(dev
->rx_handler
, NULL
);
2750 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
2752 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2753 struct net_device
*master
)
2755 if (skb
->pkt_type
== PACKET_HOST
) {
2756 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2758 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2762 /* On bonding slaves other than the currently active slave, suppress
2763 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2764 * ARP on active-backup slaves with arp_validate enabled.
2766 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2768 struct net_device
*dev
= skb
->dev
;
2770 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2771 dev
->last_rx
= jiffies
;
2773 if ((master
->priv_flags
& IFF_MASTER_ALB
) && master
->br_port
) {
2774 /* Do address unmangle. The local destination address
2775 * will be always the one master has. Provides the right
2776 * functionality in a bridge.
2778 skb_bond_set_mac_by_master(skb
, master
);
2781 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
2782 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
2783 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
2786 if (master
->priv_flags
& IFF_MASTER_ALB
) {
2787 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
2788 skb
->pkt_type
!= PACKET_MULTICAST
)
2791 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
2792 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
2799 EXPORT_SYMBOL(__skb_bond_should_drop
);
2801 static int __netif_receive_skb(struct sk_buff
*skb
)
2803 struct packet_type
*ptype
, *pt_prev
;
2804 rx_handler_func_t
*rx_handler
;
2805 struct net_device
*orig_dev
;
2806 struct net_device
*master
;
2807 struct net_device
*null_or_orig
;
2808 struct net_device
*null_or_bond
;
2809 int ret
= NET_RX_DROP
;
2812 if (!netdev_tstamp_prequeue
)
2813 net_timestamp_check(skb
);
2815 if (vlan_tx_tag_present(skb
) && vlan_hwaccel_do_receive(skb
))
2816 return NET_RX_SUCCESS
;
2818 /* if we've gotten here through NAPI, check netpoll */
2819 if (netpoll_receive_skb(skb
))
2823 skb
->skb_iif
= skb
->dev
->ifindex
;
2825 null_or_orig
= NULL
;
2826 orig_dev
= skb
->dev
;
2827 master
= ACCESS_ONCE(orig_dev
->master
);
2829 if (skb_bond_should_drop(skb
, master
))
2830 null_or_orig
= orig_dev
; /* deliver only exact match */
2835 __this_cpu_inc(softnet_data
.processed
);
2836 skb_reset_network_header(skb
);
2837 skb_reset_transport_header(skb
);
2838 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2844 #ifdef CONFIG_NET_CLS_ACT
2845 if (skb
->tc_verd
& TC_NCLS
) {
2846 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2851 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2852 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2853 ptype
->dev
== orig_dev
) {
2855 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2860 #ifdef CONFIG_NET_CLS_ACT
2861 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2867 /* Handle special case of bridge or macvlan */
2868 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
2871 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2874 skb
= rx_handler(skb
);
2880 * Make sure frames received on VLAN interfaces stacked on
2881 * bonding interfaces still make their way to any base bonding
2882 * device that may have registered for a specific ptype. The
2883 * handler may have to adjust skb->dev and orig_dev.
2885 null_or_bond
= NULL
;
2886 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
2887 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
2888 null_or_bond
= vlan_dev_real_dev(skb
->dev
);
2891 type
= skb
->protocol
;
2892 list_for_each_entry_rcu(ptype
,
2893 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2894 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
2895 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
2896 ptype
->dev
== null_or_bond
)) {
2898 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2904 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2907 /* Jamal, now you will not able to escape explaining
2908 * me how you were going to use this. :-)
2919 * netif_receive_skb - process receive buffer from network
2920 * @skb: buffer to process
2922 * netif_receive_skb() is the main receive data processing function.
2923 * It always succeeds. The buffer may be dropped during processing
2924 * for congestion control or by the protocol layers.
2926 * This function may only be called from softirq context and interrupts
2927 * should be enabled.
2929 * Return values (usually ignored):
2930 * NET_RX_SUCCESS: no congestion
2931 * NET_RX_DROP: packet was dropped
2933 int netif_receive_skb(struct sk_buff
*skb
)
2935 if (netdev_tstamp_prequeue
)
2936 net_timestamp_check(skb
);
2940 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2945 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2948 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2952 ret
= __netif_receive_skb(skb
);
2958 return __netif_receive_skb(skb
);
2961 EXPORT_SYMBOL(netif_receive_skb
);
2963 /* Network device is going away, flush any packets still pending
2964 * Called with irqs disabled.
2966 static void flush_backlog(void *arg
)
2968 struct net_device
*dev
= arg
;
2969 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2970 struct sk_buff
*skb
, *tmp
;
2973 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
2974 if (skb
->dev
== dev
) {
2975 __skb_unlink(skb
, &sd
->input_pkt_queue
);
2977 input_queue_head_incr(sd
);
2982 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
2983 if (skb
->dev
== dev
) {
2984 __skb_unlink(skb
, &sd
->process_queue
);
2986 input_queue_head_incr(sd
);
2991 static int napi_gro_complete(struct sk_buff
*skb
)
2993 struct packet_type
*ptype
;
2994 __be16 type
= skb
->protocol
;
2995 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
2998 if (NAPI_GRO_CB(skb
)->count
== 1) {
2999 skb_shinfo(skb
)->gso_size
= 0;
3004 list_for_each_entry_rcu(ptype
, head
, list
) {
3005 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3008 err
= ptype
->gro_complete(skb
);
3014 WARN_ON(&ptype
->list
== head
);
3016 return NET_RX_SUCCESS
;
3020 return netif_receive_skb(skb
);
3023 static void napi_gro_flush(struct napi_struct
*napi
)
3025 struct sk_buff
*skb
, *next
;
3027 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3030 napi_gro_complete(skb
);
3033 napi
->gro_count
= 0;
3034 napi
->gro_list
= NULL
;
3037 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3039 struct sk_buff
**pp
= NULL
;
3040 struct packet_type
*ptype
;
3041 __be16 type
= skb
->protocol
;
3042 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3045 enum gro_result ret
;
3047 if (!(skb
->dev
->features
& NETIF_F_GRO
))
3050 if (skb_is_gso(skb
) || skb_has_frags(skb
))
3054 list_for_each_entry_rcu(ptype
, head
, list
) {
3055 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3058 skb_set_network_header(skb
, skb_gro_offset(skb
));
3059 mac_len
= skb
->network_header
- skb
->mac_header
;
3060 skb
->mac_len
= mac_len
;
3061 NAPI_GRO_CB(skb
)->same_flow
= 0;
3062 NAPI_GRO_CB(skb
)->flush
= 0;
3063 NAPI_GRO_CB(skb
)->free
= 0;
3065 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3070 if (&ptype
->list
== head
)
3073 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3074 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3077 struct sk_buff
*nskb
= *pp
;
3081 napi_gro_complete(nskb
);
3088 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3092 NAPI_GRO_CB(skb
)->count
= 1;
3093 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3094 skb
->next
= napi
->gro_list
;
3095 napi
->gro_list
= skb
;
3099 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3100 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3102 BUG_ON(skb
->end
- skb
->tail
< grow
);
3104 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3107 skb
->data_len
-= grow
;
3109 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3110 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3112 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3113 put_page(skb_shinfo(skb
)->frags
[0].page
);
3114 memmove(skb_shinfo(skb
)->frags
,
3115 skb_shinfo(skb
)->frags
+ 1,
3116 --skb_shinfo(skb
)->nr_frags
);
3127 EXPORT_SYMBOL(dev_gro_receive
);
3130 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3134 if (netpoll_rx_on(skb
))
3137 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3138 NAPI_GRO_CB(p
)->same_flow
=
3139 (p
->dev
== skb
->dev
) &&
3140 !compare_ether_header(skb_mac_header(p
),
3141 skb_gro_mac_header(skb
));
3142 NAPI_GRO_CB(p
)->flush
= 0;
3145 return dev_gro_receive(napi
, skb
);
3148 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3152 if (netif_receive_skb(skb
))
3157 case GRO_MERGED_FREE
:
3168 EXPORT_SYMBOL(napi_skb_finish
);
3170 void skb_gro_reset_offset(struct sk_buff
*skb
)
3172 NAPI_GRO_CB(skb
)->data_offset
= 0;
3173 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3174 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3176 if (skb
->mac_header
== skb
->tail
&&
3177 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3178 NAPI_GRO_CB(skb
)->frag0
=
3179 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3180 skb_shinfo(skb
)->frags
[0].page_offset
;
3181 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3184 EXPORT_SYMBOL(skb_gro_reset_offset
);
3186 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3188 skb_gro_reset_offset(skb
);
3190 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3192 EXPORT_SYMBOL(napi_gro_receive
);
3194 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3196 __skb_pull(skb
, skb_headlen(skb
));
3197 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3201 EXPORT_SYMBOL(napi_reuse_skb
);
3203 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3205 struct sk_buff
*skb
= napi
->skb
;
3208 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3214 EXPORT_SYMBOL(napi_get_frags
);
3216 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3222 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3224 if (ret
== GRO_HELD
)
3225 skb_gro_pull(skb
, -ETH_HLEN
);
3226 else if (netif_receive_skb(skb
))
3231 case GRO_MERGED_FREE
:
3232 napi_reuse_skb(napi
, skb
);
3241 EXPORT_SYMBOL(napi_frags_finish
);
3243 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3245 struct sk_buff
*skb
= napi
->skb
;
3252 skb_reset_mac_header(skb
);
3253 skb_gro_reset_offset(skb
);
3255 off
= skb_gro_offset(skb
);
3256 hlen
= off
+ sizeof(*eth
);
3257 eth
= skb_gro_header_fast(skb
, off
);
3258 if (skb_gro_header_hard(skb
, hlen
)) {
3259 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3260 if (unlikely(!eth
)) {
3261 napi_reuse_skb(napi
, skb
);
3267 skb_gro_pull(skb
, sizeof(*eth
));
3270 * This works because the only protocols we care about don't require
3271 * special handling. We'll fix it up properly at the end.
3273 skb
->protocol
= eth
->h_proto
;
3278 EXPORT_SYMBOL(napi_frags_skb
);
3280 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3282 struct sk_buff
*skb
= napi_frags_skb(napi
);
3287 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3289 EXPORT_SYMBOL(napi_gro_frags
);
3292 * net_rps_action sends any pending IPI's for rps.
3293 * Note: called with local irq disabled, but exits with local irq enabled.
3295 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3298 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3301 sd
->rps_ipi_list
= NULL
;
3305 /* Send pending IPI's to kick RPS processing on remote cpus. */
3307 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3309 if (cpu_online(remsd
->cpu
))
3310 __smp_call_function_single(remsd
->cpu
,
3319 static int process_backlog(struct napi_struct
*napi
, int quota
)
3322 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3325 /* Check if we have pending ipi, its better to send them now,
3326 * not waiting net_rx_action() end.
3328 if (sd
->rps_ipi_list
) {
3329 local_irq_disable();
3330 net_rps_action_and_irq_enable(sd
);
3333 napi
->weight
= weight_p
;
3334 local_irq_disable();
3335 while (work
< quota
) {
3336 struct sk_buff
*skb
;
3339 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3341 __netif_receive_skb(skb
);
3342 local_irq_disable();
3343 input_queue_head_incr(sd
);
3344 if (++work
>= quota
) {
3351 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3353 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3354 &sd
->process_queue
);
3356 if (qlen
< quota
- work
) {
3358 * Inline a custom version of __napi_complete().
3359 * only current cpu owns and manipulates this napi,
3360 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3361 * we can use a plain write instead of clear_bit(),
3362 * and we dont need an smp_mb() memory barrier.
3364 list_del(&napi
->poll_list
);
3367 quota
= work
+ qlen
;
3377 * __napi_schedule - schedule for receive
3378 * @n: entry to schedule
3380 * The entry's receive function will be scheduled to run
3382 void __napi_schedule(struct napi_struct
*n
)
3384 unsigned long flags
;
3386 local_irq_save(flags
);
3387 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3388 local_irq_restore(flags
);
3390 EXPORT_SYMBOL(__napi_schedule
);
3392 void __napi_complete(struct napi_struct
*n
)
3394 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3395 BUG_ON(n
->gro_list
);
3397 list_del(&n
->poll_list
);
3398 smp_mb__before_clear_bit();
3399 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3401 EXPORT_SYMBOL(__napi_complete
);
3403 void napi_complete(struct napi_struct
*n
)
3405 unsigned long flags
;
3408 * don't let napi dequeue from the cpu poll list
3409 * just in case its running on a different cpu
3411 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3415 local_irq_save(flags
);
3417 local_irq_restore(flags
);
3419 EXPORT_SYMBOL(napi_complete
);
3421 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3422 int (*poll
)(struct napi_struct
*, int), int weight
)
3424 INIT_LIST_HEAD(&napi
->poll_list
);
3425 napi
->gro_count
= 0;
3426 napi
->gro_list
= NULL
;
3429 napi
->weight
= weight
;
3430 list_add(&napi
->dev_list
, &dev
->napi_list
);
3432 #ifdef CONFIG_NETPOLL
3433 spin_lock_init(&napi
->poll_lock
);
3434 napi
->poll_owner
= -1;
3436 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3438 EXPORT_SYMBOL(netif_napi_add
);
3440 void netif_napi_del(struct napi_struct
*napi
)
3442 struct sk_buff
*skb
, *next
;
3444 list_del_init(&napi
->dev_list
);
3445 napi_free_frags(napi
);
3447 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3453 napi
->gro_list
= NULL
;
3454 napi
->gro_count
= 0;
3456 EXPORT_SYMBOL(netif_napi_del
);
3458 static void net_rx_action(struct softirq_action
*h
)
3460 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3461 unsigned long time_limit
= jiffies
+ 2;
3462 int budget
= netdev_budget
;
3465 local_irq_disable();
3467 while (!list_empty(&sd
->poll_list
)) {
3468 struct napi_struct
*n
;
3471 /* If softirq window is exhuasted then punt.
3472 * Allow this to run for 2 jiffies since which will allow
3473 * an average latency of 1.5/HZ.
3475 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3480 /* Even though interrupts have been re-enabled, this
3481 * access is safe because interrupts can only add new
3482 * entries to the tail of this list, and only ->poll()
3483 * calls can remove this head entry from the list.
3485 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3487 have
= netpoll_poll_lock(n
);
3491 /* This NAPI_STATE_SCHED test is for avoiding a race
3492 * with netpoll's poll_napi(). Only the entity which
3493 * obtains the lock and sees NAPI_STATE_SCHED set will
3494 * actually make the ->poll() call. Therefore we avoid
3495 * accidently calling ->poll() when NAPI is not scheduled.
3498 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3499 work
= n
->poll(n
, weight
);
3503 WARN_ON_ONCE(work
> weight
);
3507 local_irq_disable();
3509 /* Drivers must not modify the NAPI state if they
3510 * consume the entire weight. In such cases this code
3511 * still "owns" the NAPI instance and therefore can
3512 * move the instance around on the list at-will.
3514 if (unlikely(work
== weight
)) {
3515 if (unlikely(napi_disable_pending(n
))) {
3518 local_irq_disable();
3520 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3523 netpoll_poll_unlock(have
);
3526 net_rps_action_and_irq_enable(sd
);
3528 #ifdef CONFIG_NET_DMA
3530 * There may not be any more sk_buffs coming right now, so push
3531 * any pending DMA copies to hardware
3533 dma_issue_pending_all();
3540 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3544 static gifconf_func_t
*gifconf_list
[NPROTO
];
3547 * register_gifconf - register a SIOCGIF handler
3548 * @family: Address family
3549 * @gifconf: Function handler
3551 * Register protocol dependent address dumping routines. The handler
3552 * that is passed must not be freed or reused until it has been replaced
3553 * by another handler.
3555 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3557 if (family
>= NPROTO
)
3559 gifconf_list
[family
] = gifconf
;
3562 EXPORT_SYMBOL(register_gifconf
);
3566 * Map an interface index to its name (SIOCGIFNAME)
3570 * We need this ioctl for efficient implementation of the
3571 * if_indextoname() function required by the IPv6 API. Without
3572 * it, we would have to search all the interfaces to find a
3576 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3578 struct net_device
*dev
;
3582 * Fetch the caller's info block.
3585 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3589 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3595 strcpy(ifr
.ifr_name
, dev
->name
);
3598 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3604 * Perform a SIOCGIFCONF call. This structure will change
3605 * size eventually, and there is nothing I can do about it.
3606 * Thus we will need a 'compatibility mode'.
3609 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3612 struct net_device
*dev
;
3619 * Fetch the caller's info block.
3622 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3629 * Loop over the interfaces, and write an info block for each.
3633 for_each_netdev(net
, dev
) {
3634 for (i
= 0; i
< NPROTO
; i
++) {
3635 if (gifconf_list
[i
]) {
3638 done
= gifconf_list
[i
](dev
, NULL
, 0);
3640 done
= gifconf_list
[i
](dev
, pos
+ total
,
3650 * All done. Write the updated control block back to the caller.
3652 ifc
.ifc_len
= total
;
3655 * Both BSD and Solaris return 0 here, so we do too.
3657 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3660 #ifdef CONFIG_PROC_FS
3662 * This is invoked by the /proc filesystem handler to display a device
3665 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3668 struct net
*net
= seq_file_net(seq
);
3670 struct net_device
*dev
;
3674 return SEQ_START_TOKEN
;
3677 for_each_netdev_rcu(net
, dev
)
3684 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3686 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3687 first_net_device(seq_file_net(seq
)) :
3688 next_net_device((struct net_device
*)v
);
3691 return rcu_dereference(dev
);
3694 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3700 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3702 const struct net_device_stats
*stats
= dev_get_stats(dev
);
3704 seq_printf(seq
, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3705 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3706 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3708 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3709 stats
->rx_fifo_errors
,
3710 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3711 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3712 stats
->rx_compressed
, stats
->multicast
,
3713 stats
->tx_bytes
, stats
->tx_packets
,
3714 stats
->tx_errors
, stats
->tx_dropped
,
3715 stats
->tx_fifo_errors
, stats
->collisions
,
3716 stats
->tx_carrier_errors
+
3717 stats
->tx_aborted_errors
+
3718 stats
->tx_window_errors
+
3719 stats
->tx_heartbeat_errors
,
3720 stats
->tx_compressed
);
3724 * Called from the PROCfs module. This now uses the new arbitrary sized
3725 * /proc/net interface to create /proc/net/dev
3727 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3729 if (v
== SEQ_START_TOKEN
)
3730 seq_puts(seq
, "Inter-| Receive "
3732 " face |bytes packets errs drop fifo frame "
3733 "compressed multicast|bytes packets errs "
3734 "drop fifo colls carrier compressed\n");
3736 dev_seq_printf_stats(seq
, v
);
3740 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
3742 struct softnet_data
*sd
= NULL
;
3744 while (*pos
< nr_cpu_ids
)
3745 if (cpu_online(*pos
)) {
3746 sd
= &per_cpu(softnet_data
, *pos
);
3753 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3755 return softnet_get_online(pos
);
3758 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3761 return softnet_get_online(pos
);
3764 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3768 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3770 struct softnet_data
*sd
= v
;
3772 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3773 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
3774 0, 0, 0, 0, /* was fastroute */
3775 sd
->cpu_collision
, sd
->received_rps
);
3779 static const struct seq_operations dev_seq_ops
= {
3780 .start
= dev_seq_start
,
3781 .next
= dev_seq_next
,
3782 .stop
= dev_seq_stop
,
3783 .show
= dev_seq_show
,
3786 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3788 return seq_open_net(inode
, file
, &dev_seq_ops
,
3789 sizeof(struct seq_net_private
));
3792 static const struct file_operations dev_seq_fops
= {
3793 .owner
= THIS_MODULE
,
3794 .open
= dev_seq_open
,
3796 .llseek
= seq_lseek
,
3797 .release
= seq_release_net
,
3800 static const struct seq_operations softnet_seq_ops
= {
3801 .start
= softnet_seq_start
,
3802 .next
= softnet_seq_next
,
3803 .stop
= softnet_seq_stop
,
3804 .show
= softnet_seq_show
,
3807 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3809 return seq_open(file
, &softnet_seq_ops
);
3812 static const struct file_operations softnet_seq_fops
= {
3813 .owner
= THIS_MODULE
,
3814 .open
= softnet_seq_open
,
3816 .llseek
= seq_lseek
,
3817 .release
= seq_release
,
3820 static void *ptype_get_idx(loff_t pos
)
3822 struct packet_type
*pt
= NULL
;
3826 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3832 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3833 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3842 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3846 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3849 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3851 struct packet_type
*pt
;
3852 struct list_head
*nxt
;
3856 if (v
== SEQ_START_TOKEN
)
3857 return ptype_get_idx(0);
3860 nxt
= pt
->list
.next
;
3861 if (pt
->type
== htons(ETH_P_ALL
)) {
3862 if (nxt
!= &ptype_all
)
3865 nxt
= ptype_base
[0].next
;
3867 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3869 while (nxt
== &ptype_base
[hash
]) {
3870 if (++hash
>= PTYPE_HASH_SIZE
)
3872 nxt
= ptype_base
[hash
].next
;
3875 return list_entry(nxt
, struct packet_type
, list
);
3878 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3884 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3886 struct packet_type
*pt
= v
;
3888 if (v
== SEQ_START_TOKEN
)
3889 seq_puts(seq
, "Type Device Function\n");
3890 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3891 if (pt
->type
== htons(ETH_P_ALL
))
3892 seq_puts(seq
, "ALL ");
3894 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3896 seq_printf(seq
, " %-8s %pF\n",
3897 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
3903 static const struct seq_operations ptype_seq_ops
= {
3904 .start
= ptype_seq_start
,
3905 .next
= ptype_seq_next
,
3906 .stop
= ptype_seq_stop
,
3907 .show
= ptype_seq_show
,
3910 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
3912 return seq_open_net(inode
, file
, &ptype_seq_ops
,
3913 sizeof(struct seq_net_private
));
3916 static const struct file_operations ptype_seq_fops
= {
3917 .owner
= THIS_MODULE
,
3918 .open
= ptype_seq_open
,
3920 .llseek
= seq_lseek
,
3921 .release
= seq_release_net
,
3925 static int __net_init
dev_proc_net_init(struct net
*net
)
3929 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
3931 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
3933 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
3936 if (wext_proc_init(net
))
3942 proc_net_remove(net
, "ptype");
3944 proc_net_remove(net
, "softnet_stat");
3946 proc_net_remove(net
, "dev");
3950 static void __net_exit
dev_proc_net_exit(struct net
*net
)
3952 wext_proc_exit(net
);
3954 proc_net_remove(net
, "ptype");
3955 proc_net_remove(net
, "softnet_stat");
3956 proc_net_remove(net
, "dev");
3959 static struct pernet_operations __net_initdata dev_proc_ops
= {
3960 .init
= dev_proc_net_init
,
3961 .exit
= dev_proc_net_exit
,
3964 static int __init
dev_proc_init(void)
3966 return register_pernet_subsys(&dev_proc_ops
);
3969 #define dev_proc_init() 0
3970 #endif /* CONFIG_PROC_FS */
3974 * netdev_set_master - set up master/slave pair
3975 * @slave: slave device
3976 * @master: new master device
3978 * Changes the master device of the slave. Pass %NULL to break the
3979 * bonding. The caller must hold the RTNL semaphore. On a failure
3980 * a negative errno code is returned. On success the reference counts
3981 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3982 * function returns zero.
3984 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
3986 struct net_device
*old
= slave
->master
;
3996 slave
->master
= master
;
4003 slave
->flags
|= IFF_SLAVE
;
4005 slave
->flags
&= ~IFF_SLAVE
;
4007 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4010 EXPORT_SYMBOL(netdev_set_master
);
4012 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4014 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4016 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4017 ops
->ndo_change_rx_flags(dev
, flags
);
4020 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4022 unsigned short old_flags
= dev
->flags
;
4028 dev
->flags
|= IFF_PROMISC
;
4029 dev
->promiscuity
+= inc
;
4030 if (dev
->promiscuity
== 0) {
4033 * If inc causes overflow, untouch promisc and return error.
4036 dev
->flags
&= ~IFF_PROMISC
;
4038 dev
->promiscuity
-= inc
;
4039 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4040 "set promiscuity failed, promiscuity feature "
4041 "of device might be broken.\n", dev
->name
);
4045 if (dev
->flags
!= old_flags
) {
4046 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4047 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4049 if (audit_enabled
) {
4050 current_uid_gid(&uid
, &gid
);
4051 audit_log(current
->audit_context
, GFP_ATOMIC
,
4052 AUDIT_ANOM_PROMISCUOUS
,
4053 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4054 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4055 (old_flags
& IFF_PROMISC
),
4056 audit_get_loginuid(current
),
4058 audit_get_sessionid(current
));
4061 dev_change_rx_flags(dev
, IFF_PROMISC
);
4067 * dev_set_promiscuity - update promiscuity count on a device
4071 * Add or remove promiscuity from a device. While the count in the device
4072 * remains above zero the interface remains promiscuous. Once it hits zero
4073 * the device reverts back to normal filtering operation. A negative inc
4074 * value is used to drop promiscuity on the device.
4075 * Return 0 if successful or a negative errno code on error.
4077 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4079 unsigned short old_flags
= dev
->flags
;
4082 err
= __dev_set_promiscuity(dev
, inc
);
4085 if (dev
->flags
!= old_flags
)
4086 dev_set_rx_mode(dev
);
4089 EXPORT_SYMBOL(dev_set_promiscuity
);
4092 * dev_set_allmulti - update allmulti count on a device
4096 * Add or remove reception of all multicast frames to a device. While the
4097 * count in the device remains above zero the interface remains listening
4098 * to all interfaces. Once it hits zero the device reverts back to normal
4099 * filtering operation. A negative @inc value is used to drop the counter
4100 * when releasing a resource needing all multicasts.
4101 * Return 0 if successful or a negative errno code on error.
4104 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4106 unsigned short old_flags
= dev
->flags
;
4110 dev
->flags
|= IFF_ALLMULTI
;
4111 dev
->allmulti
+= inc
;
4112 if (dev
->allmulti
== 0) {
4115 * If inc causes overflow, untouch allmulti and return error.
4118 dev
->flags
&= ~IFF_ALLMULTI
;
4120 dev
->allmulti
-= inc
;
4121 printk(KERN_WARNING
"%s: allmulti touches roof, "
4122 "set allmulti failed, allmulti feature of "
4123 "device might be broken.\n", dev
->name
);
4127 if (dev
->flags
^ old_flags
) {
4128 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4129 dev_set_rx_mode(dev
);
4133 EXPORT_SYMBOL(dev_set_allmulti
);
4136 * Upload unicast and multicast address lists to device and
4137 * configure RX filtering. When the device doesn't support unicast
4138 * filtering it is put in promiscuous mode while unicast addresses
4141 void __dev_set_rx_mode(struct net_device
*dev
)
4143 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4145 /* dev_open will call this function so the list will stay sane. */
4146 if (!(dev
->flags
&IFF_UP
))
4149 if (!netif_device_present(dev
))
4152 if (ops
->ndo_set_rx_mode
)
4153 ops
->ndo_set_rx_mode(dev
);
4155 /* Unicast addresses changes may only happen under the rtnl,
4156 * therefore calling __dev_set_promiscuity here is safe.
4158 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4159 __dev_set_promiscuity(dev
, 1);
4160 dev
->uc_promisc
= 1;
4161 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4162 __dev_set_promiscuity(dev
, -1);
4163 dev
->uc_promisc
= 0;
4166 if (ops
->ndo_set_multicast_list
)
4167 ops
->ndo_set_multicast_list(dev
);
4171 void dev_set_rx_mode(struct net_device
*dev
)
4173 netif_addr_lock_bh(dev
);
4174 __dev_set_rx_mode(dev
);
4175 netif_addr_unlock_bh(dev
);
4179 * dev_get_flags - get flags reported to userspace
4182 * Get the combination of flag bits exported through APIs to userspace.
4184 unsigned dev_get_flags(const struct net_device
*dev
)
4188 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4193 (dev
->gflags
& (IFF_PROMISC
|
4196 if (netif_running(dev
)) {
4197 if (netif_oper_up(dev
))
4198 flags
|= IFF_RUNNING
;
4199 if (netif_carrier_ok(dev
))
4200 flags
|= IFF_LOWER_UP
;
4201 if (netif_dormant(dev
))
4202 flags
|= IFF_DORMANT
;
4207 EXPORT_SYMBOL(dev_get_flags
);
4209 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4211 int old_flags
= dev
->flags
;
4217 * Set the flags on our device.
4220 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4221 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4223 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4227 * Load in the correct multicast list now the flags have changed.
4230 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4231 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4233 dev_set_rx_mode(dev
);
4236 * Have we downed the interface. We handle IFF_UP ourselves
4237 * according to user attempts to set it, rather than blindly
4242 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4243 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4246 dev_set_rx_mode(dev
);
4249 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4250 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4252 dev
->gflags
^= IFF_PROMISC
;
4253 dev_set_promiscuity(dev
, inc
);
4256 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4257 is important. Some (broken) drivers set IFF_PROMISC, when
4258 IFF_ALLMULTI is requested not asking us and not reporting.
4260 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4261 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4263 dev
->gflags
^= IFF_ALLMULTI
;
4264 dev_set_allmulti(dev
, inc
);
4270 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4272 unsigned int changes
= dev
->flags
^ old_flags
;
4274 if (changes
& IFF_UP
) {
4275 if (dev
->flags
& IFF_UP
)
4276 call_netdevice_notifiers(NETDEV_UP
, dev
);
4278 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4281 if (dev
->flags
& IFF_UP
&&
4282 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4283 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4287 * dev_change_flags - change device settings
4289 * @flags: device state flags
4291 * Change settings on device based state flags. The flags are
4292 * in the userspace exported format.
4294 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4297 int old_flags
= dev
->flags
;
4299 ret
= __dev_change_flags(dev
, flags
);
4303 changes
= old_flags
^ dev
->flags
;
4305 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4307 __dev_notify_flags(dev
, old_flags
);
4310 EXPORT_SYMBOL(dev_change_flags
);
4313 * dev_set_mtu - Change maximum transfer unit
4315 * @new_mtu: new transfer unit
4317 * Change the maximum transfer size of the network device.
4319 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4321 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4324 if (new_mtu
== dev
->mtu
)
4327 /* MTU must be positive. */
4331 if (!netif_device_present(dev
))
4335 if (ops
->ndo_change_mtu
)
4336 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4340 if (!err
&& dev
->flags
& IFF_UP
)
4341 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4344 EXPORT_SYMBOL(dev_set_mtu
);
4347 * dev_set_mac_address - Change Media Access Control Address
4351 * Change the hardware (MAC) address of the device
4353 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4355 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4358 if (!ops
->ndo_set_mac_address
)
4360 if (sa
->sa_family
!= dev
->type
)
4362 if (!netif_device_present(dev
))
4364 err
= ops
->ndo_set_mac_address(dev
, sa
);
4366 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4369 EXPORT_SYMBOL(dev_set_mac_address
);
4372 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4374 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4377 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4383 case SIOCGIFFLAGS
: /* Get interface flags */
4384 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4387 case SIOCGIFMETRIC
: /* Get the metric on the interface
4388 (currently unused) */
4389 ifr
->ifr_metric
= 0;
4392 case SIOCGIFMTU
: /* Get the MTU of a device */
4393 ifr
->ifr_mtu
= dev
->mtu
;
4398 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4400 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4401 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4402 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4410 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4411 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4412 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4413 ifr
->ifr_map
.irq
= dev
->irq
;
4414 ifr
->ifr_map
.dma
= dev
->dma
;
4415 ifr
->ifr_map
.port
= dev
->if_port
;
4419 ifr
->ifr_ifindex
= dev
->ifindex
;
4423 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4427 /* dev_ioctl() should ensure this case
4439 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4441 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4444 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4445 const struct net_device_ops
*ops
;
4450 ops
= dev
->netdev_ops
;
4453 case SIOCSIFFLAGS
: /* Set interface flags */
4454 return dev_change_flags(dev
, ifr
->ifr_flags
);
4456 case SIOCSIFMETRIC
: /* Set the metric on the interface
4457 (currently unused) */
4460 case SIOCSIFMTU
: /* Set the MTU of a device */
4461 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4464 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4466 case SIOCSIFHWBROADCAST
:
4467 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4469 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4470 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4471 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4475 if (ops
->ndo_set_config
) {
4476 if (!netif_device_present(dev
))
4478 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4483 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4484 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4486 if (!netif_device_present(dev
))
4488 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4491 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4492 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4494 if (!netif_device_present(dev
))
4496 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4499 if (ifr
->ifr_qlen
< 0)
4501 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4505 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4506 return dev_change_name(dev
, ifr
->ifr_newname
);
4509 * Unknown or private ioctl
4512 if ((cmd
>= SIOCDEVPRIVATE
&&
4513 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4514 cmd
== SIOCBONDENSLAVE
||
4515 cmd
== SIOCBONDRELEASE
||
4516 cmd
== SIOCBONDSETHWADDR
||
4517 cmd
== SIOCBONDSLAVEINFOQUERY
||
4518 cmd
== SIOCBONDINFOQUERY
||
4519 cmd
== SIOCBONDCHANGEACTIVE
||
4520 cmd
== SIOCGMIIPHY
||
4521 cmd
== SIOCGMIIREG
||
4522 cmd
== SIOCSMIIREG
||
4523 cmd
== SIOCBRADDIF
||
4524 cmd
== SIOCBRDELIF
||
4525 cmd
== SIOCSHWTSTAMP
||
4526 cmd
== SIOCWANDEV
) {
4528 if (ops
->ndo_do_ioctl
) {
4529 if (netif_device_present(dev
))
4530 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4542 * This function handles all "interface"-type I/O control requests. The actual
4543 * 'doing' part of this is dev_ifsioc above.
4547 * dev_ioctl - network device ioctl
4548 * @net: the applicable net namespace
4549 * @cmd: command to issue
4550 * @arg: pointer to a struct ifreq in user space
4552 * Issue ioctl functions to devices. This is normally called by the
4553 * user space syscall interfaces but can sometimes be useful for
4554 * other purposes. The return value is the return from the syscall if
4555 * positive or a negative errno code on error.
4558 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4564 /* One special case: SIOCGIFCONF takes ifconf argument
4565 and requires shared lock, because it sleeps writing
4569 if (cmd
== SIOCGIFCONF
) {
4571 ret
= dev_ifconf(net
, (char __user
*) arg
);
4575 if (cmd
== SIOCGIFNAME
)
4576 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4578 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4581 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4583 colon
= strchr(ifr
.ifr_name
, ':');
4588 * See which interface the caller is talking about.
4593 * These ioctl calls:
4594 * - can be done by all.
4595 * - atomic and do not require locking.
4606 dev_load(net
, ifr
.ifr_name
);
4608 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4613 if (copy_to_user(arg
, &ifr
,
4614 sizeof(struct ifreq
)))
4620 dev_load(net
, ifr
.ifr_name
);
4622 ret
= dev_ethtool(net
, &ifr
);
4627 if (copy_to_user(arg
, &ifr
,
4628 sizeof(struct ifreq
)))
4634 * These ioctl calls:
4635 * - require superuser power.
4636 * - require strict serialization.
4642 if (!capable(CAP_NET_ADMIN
))
4644 dev_load(net
, ifr
.ifr_name
);
4646 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4651 if (copy_to_user(arg
, &ifr
,
4652 sizeof(struct ifreq
)))
4658 * These ioctl calls:
4659 * - require superuser power.
4660 * - require strict serialization.
4661 * - do not return a value
4671 case SIOCSIFHWBROADCAST
:
4674 case SIOCBONDENSLAVE
:
4675 case SIOCBONDRELEASE
:
4676 case SIOCBONDSETHWADDR
:
4677 case SIOCBONDCHANGEACTIVE
:
4681 if (!capable(CAP_NET_ADMIN
))
4684 case SIOCBONDSLAVEINFOQUERY
:
4685 case SIOCBONDINFOQUERY
:
4686 dev_load(net
, ifr
.ifr_name
);
4688 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4693 /* Get the per device memory space. We can add this but
4694 * currently do not support it */
4696 /* Set the per device memory buffer space.
4697 * Not applicable in our case */
4702 * Unknown or private ioctl.
4705 if (cmd
== SIOCWANDEV
||
4706 (cmd
>= SIOCDEVPRIVATE
&&
4707 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4708 dev_load(net
, ifr
.ifr_name
);
4710 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4712 if (!ret
&& copy_to_user(arg
, &ifr
,
4713 sizeof(struct ifreq
)))
4717 /* Take care of Wireless Extensions */
4718 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4719 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4726 * dev_new_index - allocate an ifindex
4727 * @net: the applicable net namespace
4729 * Returns a suitable unique value for a new device interface
4730 * number. The caller must hold the rtnl semaphore or the
4731 * dev_base_lock to be sure it remains unique.
4733 static int dev_new_index(struct net
*net
)
4739 if (!__dev_get_by_index(net
, ifindex
))
4744 /* Delayed registration/unregisteration */
4745 static LIST_HEAD(net_todo_list
);
4747 static void net_set_todo(struct net_device
*dev
)
4749 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4752 static void rollback_registered_many(struct list_head
*head
)
4754 struct net_device
*dev
, *tmp
;
4756 BUG_ON(dev_boot_phase
);
4759 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4760 /* Some devices call without registering
4761 * for initialization unwind. Remove those
4762 * devices and proceed with the remaining.
4764 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4765 pr_debug("unregister_netdevice: device %s/%p never "
4766 "was registered\n", dev
->name
, dev
);
4769 list_del(&dev
->unreg_list
);
4773 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4775 /* If device is running, close it first. */
4778 /* And unlink it from device chain. */
4779 unlist_netdevice(dev
);
4781 dev
->reg_state
= NETREG_UNREGISTERING
;
4786 list_for_each_entry(dev
, head
, unreg_list
) {
4787 /* Shutdown queueing discipline. */
4791 /* Notify protocols, that we are about to destroy
4792 this device. They should clean all the things.
4794 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4796 if (!dev
->rtnl_link_ops
||
4797 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
4798 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
4801 * Flush the unicast and multicast chains
4806 if (dev
->netdev_ops
->ndo_uninit
)
4807 dev
->netdev_ops
->ndo_uninit(dev
);
4809 /* Notifier chain MUST detach us from master device. */
4810 WARN_ON(dev
->master
);
4812 /* Remove entries from kobject tree */
4813 netdev_unregister_kobject(dev
);
4816 /* Process any work delayed until the end of the batch */
4817 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
4818 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
4822 list_for_each_entry(dev
, head
, unreg_list
)
4826 static void rollback_registered(struct net_device
*dev
)
4830 list_add(&dev
->unreg_list
, &single
);
4831 rollback_registered_many(&single
);
4834 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
4835 struct netdev_queue
*dev_queue
,
4838 spin_lock_init(&dev_queue
->_xmit_lock
);
4839 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
4840 dev_queue
->xmit_lock_owner
= -1;
4843 static void netdev_init_queue_locks(struct net_device
*dev
)
4845 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
4846 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
4849 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
4851 /* Fix illegal SG+CSUM combinations. */
4852 if ((features
& NETIF_F_SG
) &&
4853 !(features
& NETIF_F_ALL_CSUM
)) {
4855 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
4856 "checksum feature.\n", name
);
4857 features
&= ~NETIF_F_SG
;
4860 /* TSO requires that SG is present as well. */
4861 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
4863 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
4864 "SG feature.\n", name
);
4865 features
&= ~NETIF_F_TSO
;
4868 if (features
& NETIF_F_UFO
) {
4869 if (!(features
& NETIF_F_GEN_CSUM
)) {
4871 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4872 "since no NETIF_F_HW_CSUM feature.\n",
4874 features
&= ~NETIF_F_UFO
;
4877 if (!(features
& NETIF_F_SG
)) {
4879 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4880 "since no NETIF_F_SG feature.\n", name
);
4881 features
&= ~NETIF_F_UFO
;
4887 EXPORT_SYMBOL(netdev_fix_features
);
4890 * netif_stacked_transfer_operstate - transfer operstate
4891 * @rootdev: the root or lower level device to transfer state from
4892 * @dev: the device to transfer operstate to
4894 * Transfer operational state from root to device. This is normally
4895 * called when a stacking relationship exists between the root
4896 * device and the device(a leaf device).
4898 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
4899 struct net_device
*dev
)
4901 if (rootdev
->operstate
== IF_OPER_DORMANT
)
4902 netif_dormant_on(dev
);
4904 netif_dormant_off(dev
);
4906 if (netif_carrier_ok(rootdev
)) {
4907 if (!netif_carrier_ok(dev
))
4908 netif_carrier_on(dev
);
4910 if (netif_carrier_ok(dev
))
4911 netif_carrier_off(dev
);
4914 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
4917 * register_netdevice - register a network device
4918 * @dev: device to register
4920 * Take a completed network device structure and add it to the kernel
4921 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4922 * chain. 0 is returned on success. A negative errno code is returned
4923 * on a failure to set up the device, or if the name is a duplicate.
4925 * Callers must hold the rtnl semaphore. You may want
4926 * register_netdev() instead of this.
4929 * The locking appears insufficient to guarantee two parallel registers
4930 * will not get the same name.
4933 int register_netdevice(struct net_device
*dev
)
4936 struct net
*net
= dev_net(dev
);
4938 BUG_ON(dev_boot_phase
);
4943 /* When net_device's are persistent, this will be fatal. */
4944 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
4947 spin_lock_init(&dev
->addr_list_lock
);
4948 netdev_set_addr_lockdep_class(dev
);
4949 netdev_init_queue_locks(dev
);
4954 if (!dev
->num_rx_queues
) {
4956 * Allocate a single RX queue if driver never called
4960 dev
->_rx
= kzalloc(sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
4966 dev
->_rx
->first
= dev
->_rx
;
4967 atomic_set(&dev
->_rx
->count
, 1);
4968 dev
->num_rx_queues
= 1;
4971 /* Init, if this function is available */
4972 if (dev
->netdev_ops
->ndo_init
) {
4973 ret
= dev
->netdev_ops
->ndo_init(dev
);
4981 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
4985 dev
->ifindex
= dev_new_index(net
);
4986 if (dev
->iflink
== -1)
4987 dev
->iflink
= dev
->ifindex
;
4989 /* Fix illegal checksum combinations */
4990 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
4991 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4992 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
4994 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
4997 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
4998 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4999 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5001 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5004 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5006 /* Enable software GSO if SG is supported. */
5007 if (dev
->features
& NETIF_F_SG
)
5008 dev
->features
|= NETIF_F_GSO
;
5010 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5011 ret
= notifier_to_errno(ret
);
5015 ret
= netdev_register_kobject(dev
);
5018 dev
->reg_state
= NETREG_REGISTERED
;
5021 * Default initial state at registry is that the
5022 * device is present.
5025 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5027 dev_init_scheduler(dev
);
5029 list_netdevice(dev
);
5031 /* Notify protocols, that a new device appeared. */
5032 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5033 ret
= notifier_to_errno(ret
);
5035 rollback_registered(dev
);
5036 dev
->reg_state
= NETREG_UNREGISTERED
;
5039 * Prevent userspace races by waiting until the network
5040 * device is fully setup before sending notifications.
5042 if (!dev
->rtnl_link_ops
||
5043 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5044 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5050 if (dev
->netdev_ops
->ndo_uninit
)
5051 dev
->netdev_ops
->ndo_uninit(dev
);
5054 EXPORT_SYMBOL(register_netdevice
);
5057 * init_dummy_netdev - init a dummy network device for NAPI
5058 * @dev: device to init
5060 * This takes a network device structure and initialize the minimum
5061 * amount of fields so it can be used to schedule NAPI polls without
5062 * registering a full blown interface. This is to be used by drivers
5063 * that need to tie several hardware interfaces to a single NAPI
5064 * poll scheduler due to HW limitations.
5066 int init_dummy_netdev(struct net_device
*dev
)
5068 /* Clear everything. Note we don't initialize spinlocks
5069 * are they aren't supposed to be taken by any of the
5070 * NAPI code and this dummy netdev is supposed to be
5071 * only ever used for NAPI polls
5073 memset(dev
, 0, sizeof(struct net_device
));
5075 /* make sure we BUG if trying to hit standard
5076 * register/unregister code path
5078 dev
->reg_state
= NETREG_DUMMY
;
5080 /* initialize the ref count */
5081 atomic_set(&dev
->refcnt
, 1);
5083 /* NAPI wants this */
5084 INIT_LIST_HEAD(&dev
->napi_list
);
5086 /* a dummy interface is started by default */
5087 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5088 set_bit(__LINK_STATE_START
, &dev
->state
);
5092 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5096 * register_netdev - register a network device
5097 * @dev: device to register
5099 * Take a completed network device structure and add it to the kernel
5100 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5101 * chain. 0 is returned on success. A negative errno code is returned
5102 * on a failure to set up the device, or if the name is a duplicate.
5104 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5105 * and expands the device name if you passed a format string to
5108 int register_netdev(struct net_device
*dev
)
5115 * If the name is a format string the caller wants us to do a
5118 if (strchr(dev
->name
, '%')) {
5119 err
= dev_alloc_name(dev
, dev
->name
);
5124 err
= register_netdevice(dev
);
5129 EXPORT_SYMBOL(register_netdev
);
5132 * netdev_wait_allrefs - wait until all references are gone.
5134 * This is called when unregistering network devices.
5136 * Any protocol or device that holds a reference should register
5137 * for netdevice notification, and cleanup and put back the
5138 * reference if they receive an UNREGISTER event.
5139 * We can get stuck here if buggy protocols don't correctly
5142 static void netdev_wait_allrefs(struct net_device
*dev
)
5144 unsigned long rebroadcast_time
, warning_time
;
5146 linkwatch_forget_dev(dev
);
5148 rebroadcast_time
= warning_time
= jiffies
;
5149 while (atomic_read(&dev
->refcnt
) != 0) {
5150 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5153 /* Rebroadcast unregister notification */
5154 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5155 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5156 * should have already handle it the first time */
5158 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5160 /* We must not have linkwatch events
5161 * pending on unregister. If this
5162 * happens, we simply run the queue
5163 * unscheduled, resulting in a noop
5166 linkwatch_run_queue();
5171 rebroadcast_time
= jiffies
;
5176 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5177 printk(KERN_EMERG
"unregister_netdevice: "
5178 "waiting for %s to become free. Usage "
5180 dev
->name
, atomic_read(&dev
->refcnt
));
5181 warning_time
= jiffies
;
5190 * register_netdevice(x1);
5191 * register_netdevice(x2);
5193 * unregister_netdevice(y1);
5194 * unregister_netdevice(y2);
5200 * We are invoked by rtnl_unlock().
5201 * This allows us to deal with problems:
5202 * 1) We can delete sysfs objects which invoke hotplug
5203 * without deadlocking with linkwatch via keventd.
5204 * 2) Since we run with the RTNL semaphore not held, we can sleep
5205 * safely in order to wait for the netdev refcnt to drop to zero.
5207 * We must not return until all unregister events added during
5208 * the interval the lock was held have been completed.
5210 void netdev_run_todo(void)
5212 struct list_head list
;
5214 /* Snapshot list, allow later requests */
5215 list_replace_init(&net_todo_list
, &list
);
5219 while (!list_empty(&list
)) {
5220 struct net_device
*dev
5221 = list_first_entry(&list
, struct net_device
, todo_list
);
5222 list_del(&dev
->todo_list
);
5224 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5225 printk(KERN_ERR
"network todo '%s' but state %d\n",
5226 dev
->name
, dev
->reg_state
);
5231 dev
->reg_state
= NETREG_UNREGISTERED
;
5233 on_each_cpu(flush_backlog
, dev
, 1);
5235 netdev_wait_allrefs(dev
);
5238 BUG_ON(atomic_read(&dev
->refcnt
));
5239 WARN_ON(dev
->ip_ptr
);
5240 WARN_ON(dev
->ip6_ptr
);
5241 WARN_ON(dev
->dn_ptr
);
5243 if (dev
->destructor
)
5244 dev
->destructor(dev
);
5246 /* Free network device */
5247 kobject_put(&dev
->dev
.kobj
);
5252 * dev_txq_stats_fold - fold tx_queues stats
5253 * @dev: device to get statistics from
5254 * @stats: struct net_device_stats to hold results
5256 void dev_txq_stats_fold(const struct net_device
*dev
,
5257 struct net_device_stats
*stats
)
5259 unsigned long tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5261 struct netdev_queue
*txq
;
5263 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5264 txq
= netdev_get_tx_queue(dev
, i
);
5265 tx_bytes
+= txq
->tx_bytes
;
5266 tx_packets
+= txq
->tx_packets
;
5267 tx_dropped
+= txq
->tx_dropped
;
5269 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5270 stats
->tx_bytes
= tx_bytes
;
5271 stats
->tx_packets
= tx_packets
;
5272 stats
->tx_dropped
= tx_dropped
;
5275 EXPORT_SYMBOL(dev_txq_stats_fold
);
5278 * dev_get_stats - get network device statistics
5279 * @dev: device to get statistics from
5281 * Get network statistics from device. The device driver may provide
5282 * its own method by setting dev->netdev_ops->get_stats; otherwise
5283 * the internal statistics structure is used.
5285 const struct net_device_stats
*dev_get_stats(struct net_device
*dev
)
5287 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5289 if (ops
->ndo_get_stats
)
5290 return ops
->ndo_get_stats(dev
);
5292 dev_txq_stats_fold(dev
, &dev
->stats
);
5295 EXPORT_SYMBOL(dev_get_stats
);
5297 static void netdev_init_one_queue(struct net_device
*dev
,
5298 struct netdev_queue
*queue
,
5304 static void netdev_init_queues(struct net_device
*dev
)
5306 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
5307 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5308 spin_lock_init(&dev
->tx_global_lock
);
5312 * alloc_netdev_mq - allocate network device
5313 * @sizeof_priv: size of private data to allocate space for
5314 * @name: device name format string
5315 * @setup: callback to initialize device
5316 * @queue_count: the number of subqueues to allocate
5318 * Allocates a struct net_device with private data area for driver use
5319 * and performs basic initialization. Also allocates subquue structs
5320 * for each queue on the device at the end of the netdevice.
5322 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
5323 void (*setup
)(struct net_device
*), unsigned int queue_count
)
5325 struct netdev_queue
*tx
;
5326 struct net_device
*dev
;
5328 struct net_device
*p
;
5330 struct netdev_rx_queue
*rx
;
5334 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5336 alloc_size
= sizeof(struct net_device
);
5338 /* ensure 32-byte alignment of private area */
5339 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5340 alloc_size
+= sizeof_priv
;
5342 /* ensure 32-byte alignment of whole construct */
5343 alloc_size
+= NETDEV_ALIGN
- 1;
5345 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5347 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5351 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5353 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5359 rx
= kcalloc(queue_count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5361 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5366 atomic_set(&rx
->count
, queue_count
);
5369 * Set a pointer to first element in the array which holds the
5372 for (i
= 0; i
< queue_count
; i
++)
5376 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5377 dev
->padded
= (char *)dev
- (char *)p
;
5379 if (dev_addr_init(dev
))
5385 dev_net_set(dev
, &init_net
);
5388 dev
->num_tx_queues
= queue_count
;
5389 dev
->real_num_tx_queues
= queue_count
;
5393 dev
->num_rx_queues
= queue_count
;
5396 dev
->gso_max_size
= GSO_MAX_SIZE
;
5398 netdev_init_queues(dev
);
5400 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5401 dev
->ethtool_ntuple_list
.count
= 0;
5402 INIT_LIST_HEAD(&dev
->napi_list
);
5403 INIT_LIST_HEAD(&dev
->unreg_list
);
5404 INIT_LIST_HEAD(&dev
->link_watch_list
);
5405 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5407 strcpy(dev
->name
, name
);
5420 EXPORT_SYMBOL(alloc_netdev_mq
);
5423 * free_netdev - free network device
5426 * This function does the last stage of destroying an allocated device
5427 * interface. The reference to the device object is released.
5428 * If this is the last reference then it will be freed.
5430 void free_netdev(struct net_device
*dev
)
5432 struct napi_struct
*p
, *n
;
5434 release_net(dev_net(dev
));
5438 /* Flush device addresses */
5439 dev_addr_flush(dev
);
5441 /* Clear ethtool n-tuple list */
5442 ethtool_ntuple_flush(dev
);
5444 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5447 /* Compatibility with error handling in drivers */
5448 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5449 kfree((char *)dev
- dev
->padded
);
5453 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5454 dev
->reg_state
= NETREG_RELEASED
;
5456 /* will free via device release */
5457 put_device(&dev
->dev
);
5459 EXPORT_SYMBOL(free_netdev
);
5462 * synchronize_net - Synchronize with packet receive processing
5464 * Wait for packets currently being received to be done.
5465 * Does not block later packets from starting.
5467 void synchronize_net(void)
5472 EXPORT_SYMBOL(synchronize_net
);
5475 * unregister_netdevice_queue - remove device from the kernel
5479 * This function shuts down a device interface and removes it
5480 * from the kernel tables.
5481 * If head not NULL, device is queued to be unregistered later.
5483 * Callers must hold the rtnl semaphore. You may want
5484 * unregister_netdev() instead of this.
5487 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5492 list_move_tail(&dev
->unreg_list
, head
);
5494 rollback_registered(dev
);
5495 /* Finish processing unregister after unlock */
5499 EXPORT_SYMBOL(unregister_netdevice_queue
);
5502 * unregister_netdevice_many - unregister many devices
5503 * @head: list of devices
5505 void unregister_netdevice_many(struct list_head
*head
)
5507 struct net_device
*dev
;
5509 if (!list_empty(head
)) {
5510 rollback_registered_many(head
);
5511 list_for_each_entry(dev
, head
, unreg_list
)
5515 EXPORT_SYMBOL(unregister_netdevice_many
);
5518 * unregister_netdev - remove device from the kernel
5521 * This function shuts down a device interface and removes it
5522 * from the kernel tables.
5524 * This is just a wrapper for unregister_netdevice that takes
5525 * the rtnl semaphore. In general you want to use this and not
5526 * unregister_netdevice.
5528 void unregister_netdev(struct net_device
*dev
)
5531 unregister_netdevice(dev
);
5534 EXPORT_SYMBOL(unregister_netdev
);
5537 * dev_change_net_namespace - move device to different nethost namespace
5539 * @net: network namespace
5540 * @pat: If not NULL name pattern to try if the current device name
5541 * is already taken in the destination network namespace.
5543 * This function shuts down a device interface and moves it
5544 * to a new network namespace. On success 0 is returned, on
5545 * a failure a netagive errno code is returned.
5547 * Callers must hold the rtnl semaphore.
5550 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5556 /* Don't allow namespace local devices to be moved. */
5558 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5561 /* Ensure the device has been registrered */
5563 if (dev
->reg_state
!= NETREG_REGISTERED
)
5566 /* Get out if there is nothing todo */
5568 if (net_eq(dev_net(dev
), net
))
5571 /* Pick the destination device name, and ensure
5572 * we can use it in the destination network namespace.
5575 if (__dev_get_by_name(net
, dev
->name
)) {
5576 /* We get here if we can't use the current device name */
5579 if (dev_get_valid_name(dev
, pat
, 1))
5584 * And now a mini version of register_netdevice unregister_netdevice.
5587 /* If device is running close it first. */
5590 /* And unlink it from device chain */
5592 unlist_netdevice(dev
);
5596 /* Shutdown queueing discipline. */
5599 /* Notify protocols, that we are about to destroy
5600 this device. They should clean all the things.
5602 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5603 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5606 * Flush the unicast and multicast chains
5611 /* Actually switch the network namespace */
5612 dev_net_set(dev
, net
);
5614 /* If there is an ifindex conflict assign a new one */
5615 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5616 int iflink
= (dev
->iflink
== dev
->ifindex
);
5617 dev
->ifindex
= dev_new_index(net
);
5619 dev
->iflink
= dev
->ifindex
;
5622 /* Fixup kobjects */
5623 err
= device_rename(&dev
->dev
, dev
->name
);
5626 /* Add the device back in the hashes */
5627 list_netdevice(dev
);
5629 /* Notify protocols, that a new device appeared. */
5630 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5633 * Prevent userspace races by waiting until the network
5634 * device is fully setup before sending notifications.
5636 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5643 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5645 static int dev_cpu_callback(struct notifier_block
*nfb
,
5646 unsigned long action
,
5649 struct sk_buff
**list_skb
;
5650 struct sk_buff
*skb
;
5651 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5652 struct softnet_data
*sd
, *oldsd
;
5654 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5657 local_irq_disable();
5658 cpu
= smp_processor_id();
5659 sd
= &per_cpu(softnet_data
, cpu
);
5660 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5662 /* Find end of our completion_queue. */
5663 list_skb
= &sd
->completion_queue
;
5665 list_skb
= &(*list_skb
)->next
;
5666 /* Append completion queue from offline CPU. */
5667 *list_skb
= oldsd
->completion_queue
;
5668 oldsd
->completion_queue
= NULL
;
5670 /* Append output queue from offline CPU. */
5671 if (oldsd
->output_queue
) {
5672 *sd
->output_queue_tailp
= oldsd
->output_queue
;
5673 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
5674 oldsd
->output_queue
= NULL
;
5675 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
5678 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5681 /* Process offline CPU's input_pkt_queue */
5682 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
5684 input_queue_head_incr(oldsd
);
5686 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
5688 input_queue_head_incr(oldsd
);
5696 * netdev_increment_features - increment feature set by one
5697 * @all: current feature set
5698 * @one: new feature set
5699 * @mask: mask feature set
5701 * Computes a new feature set after adding a device with feature set
5702 * @one to the master device with current feature set @all. Will not
5703 * enable anything that is off in @mask. Returns the new feature set.
5705 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
5708 /* If device needs checksumming, downgrade to it. */
5709 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
5710 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
5711 else if (mask
& NETIF_F_ALL_CSUM
) {
5712 /* If one device supports v4/v6 checksumming, set for all. */
5713 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
5714 !(all
& NETIF_F_GEN_CSUM
)) {
5715 all
&= ~NETIF_F_ALL_CSUM
;
5716 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
5719 /* If one device supports hw checksumming, set for all. */
5720 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
5721 all
&= ~NETIF_F_ALL_CSUM
;
5722 all
|= NETIF_F_HW_CSUM
;
5726 one
|= NETIF_F_ALL_CSUM
;
5728 one
|= all
& NETIF_F_ONE_FOR_ALL
;
5729 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
5730 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
5734 EXPORT_SYMBOL(netdev_increment_features
);
5736 static struct hlist_head
*netdev_create_hash(void)
5739 struct hlist_head
*hash
;
5741 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5743 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5744 INIT_HLIST_HEAD(&hash
[i
]);
5749 /* Initialize per network namespace state */
5750 static int __net_init
netdev_init(struct net
*net
)
5752 INIT_LIST_HEAD(&net
->dev_base_head
);
5754 net
->dev_name_head
= netdev_create_hash();
5755 if (net
->dev_name_head
== NULL
)
5758 net
->dev_index_head
= netdev_create_hash();
5759 if (net
->dev_index_head
== NULL
)
5765 kfree(net
->dev_name_head
);
5771 * netdev_drivername - network driver for the device
5772 * @dev: network device
5773 * @buffer: buffer for resulting name
5774 * @len: size of buffer
5776 * Determine network driver for device.
5778 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
5780 const struct device_driver
*driver
;
5781 const struct device
*parent
;
5783 if (len
<= 0 || !buffer
)
5787 parent
= dev
->dev
.parent
;
5792 driver
= parent
->driver
;
5793 if (driver
&& driver
->name
)
5794 strlcpy(buffer
, driver
->name
, len
);
5798 static void __net_exit
netdev_exit(struct net
*net
)
5800 kfree(net
->dev_name_head
);
5801 kfree(net
->dev_index_head
);
5804 static struct pernet_operations __net_initdata netdev_net_ops
= {
5805 .init
= netdev_init
,
5806 .exit
= netdev_exit
,
5809 static void __net_exit
default_device_exit(struct net
*net
)
5811 struct net_device
*dev
, *aux
;
5813 * Push all migratable network devices back to the
5814 * initial network namespace
5817 for_each_netdev_safe(net
, dev
, aux
) {
5819 char fb_name
[IFNAMSIZ
];
5821 /* Ignore unmoveable devices (i.e. loopback) */
5822 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5825 /* Leave virtual devices for the generic cleanup */
5826 if (dev
->rtnl_link_ops
)
5829 /* Push remaing network devices to init_net */
5830 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
5831 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
5833 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
5834 __func__
, dev
->name
, err
);
5841 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
5843 /* At exit all network devices most be removed from a network
5844 * namespace. Do this in the reverse order of registeration.
5845 * Do this across as many network namespaces as possible to
5846 * improve batching efficiency.
5848 struct net_device
*dev
;
5850 LIST_HEAD(dev_kill_list
);
5853 list_for_each_entry(net
, net_list
, exit_list
) {
5854 for_each_netdev_reverse(net
, dev
) {
5855 if (dev
->rtnl_link_ops
)
5856 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
5858 unregister_netdevice_queue(dev
, &dev_kill_list
);
5861 unregister_netdevice_many(&dev_kill_list
);
5865 static struct pernet_operations __net_initdata default_device_ops
= {
5866 .exit
= default_device_exit
,
5867 .exit_batch
= default_device_exit_batch
,
5871 * Initialize the DEV module. At boot time this walks the device list and
5872 * unhooks any devices that fail to initialise (normally hardware not
5873 * present) and leaves us with a valid list of present and active devices.
5878 * This is called single threaded during boot, so no need
5879 * to take the rtnl semaphore.
5881 static int __init
net_dev_init(void)
5883 int i
, rc
= -ENOMEM
;
5885 BUG_ON(!dev_boot_phase
);
5887 if (dev_proc_init())
5890 if (netdev_kobject_init())
5893 INIT_LIST_HEAD(&ptype_all
);
5894 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
5895 INIT_LIST_HEAD(&ptype_base
[i
]);
5897 if (register_pernet_subsys(&netdev_net_ops
))
5901 * Initialise the packet receive queues.
5904 for_each_possible_cpu(i
) {
5905 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
5907 memset(sd
, 0, sizeof(*sd
));
5908 skb_queue_head_init(&sd
->input_pkt_queue
);
5909 skb_queue_head_init(&sd
->process_queue
);
5910 sd
->completion_queue
= NULL
;
5911 INIT_LIST_HEAD(&sd
->poll_list
);
5912 sd
->output_queue
= NULL
;
5913 sd
->output_queue_tailp
= &sd
->output_queue
;
5915 sd
->csd
.func
= rps_trigger_softirq
;
5921 sd
->backlog
.poll
= process_backlog
;
5922 sd
->backlog
.weight
= weight_p
;
5923 sd
->backlog
.gro_list
= NULL
;
5924 sd
->backlog
.gro_count
= 0;
5929 /* The loopback device is special if any other network devices
5930 * is present in a network namespace the loopback device must
5931 * be present. Since we now dynamically allocate and free the
5932 * loopback device ensure this invariant is maintained by
5933 * keeping the loopback device as the first device on the
5934 * list of network devices. Ensuring the loopback devices
5935 * is the first device that appears and the last network device
5938 if (register_pernet_device(&loopback_net_ops
))
5941 if (register_pernet_device(&default_device_ops
))
5944 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
5945 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
5947 hotcpu_notifier(dev_cpu_callback
, 0);
5955 subsys_initcall(net_dev_init
);
5957 static int __init
initialize_hashrnd(void)
5959 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
5963 late_initcall_sync(initialize_hashrnd
);