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>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 * The list of packet types we will receive (as opposed to discard)
145 * and the routines to invoke.
147 * Why 16. Because with 16 the only overlap we get on a hash of the
148 * low nibble of the protocol value is RARP/SNAP/X.25.
150 * NOTE: That is no longer true with the addition of VLAN tags. Not
151 * sure which should go first, but I bet it won't make much
152 * difference if we are running VLANs. The good news is that
153 * this protocol won't be in the list unless compiled in, so
154 * the average user (w/out VLANs) will not be adversely affected.
171 #define PTYPE_HASH_SIZE (16)
172 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
174 static DEFINE_SPINLOCK(ptype_lock
);
175 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
176 static struct list_head ptype_all __read_mostly
; /* Taps */
179 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
182 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
184 * Writers must hold the rtnl semaphore while they loop through the
185 * dev_base_head list, and hold dev_base_lock for writing when they do the
186 * actual updates. This allows pure readers to access the list even
187 * while a writer is preparing to update it.
189 * To put it another way, dev_base_lock is held for writing only to
190 * protect against pure readers; the rtnl semaphore provides the
191 * protection against other writers.
193 * See, for example usages, register_netdevice() and
194 * unregister_netdevice(), which must be called with the rtnl
197 DEFINE_RWLOCK(dev_base_lock
);
198 EXPORT_SYMBOL(dev_base_lock
);
200 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
202 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
203 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
206 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
208 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
211 static inline void rps_lock(struct softnet_data
*sd
)
214 spin_lock(&sd
->input_pkt_queue
.lock
);
218 static inline void rps_unlock(struct softnet_data
*sd
)
221 spin_unlock(&sd
->input_pkt_queue
.lock
);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device
*dev
)
228 struct net
*net
= dev_net(dev
);
232 write_lock_bh(&dev_base_lock
);
233 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
234 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
235 hlist_add_head_rcu(&dev
->index_hlist
,
236 dev_index_hash(net
, dev
->ifindex
));
237 write_unlock_bh(&dev_base_lock
);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device
*dev
)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock
);
250 list_del_rcu(&dev
->dev_list
);
251 hlist_del_rcu(&dev
->name_hlist
);
252 hlist_del_rcu(&dev
->index_hlist
);
253 write_unlock_bh(&dev_base_lock
);
260 static RAW_NOTIFIER_HEAD(netdev_chain
);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
268 EXPORT_PER_CPU_SYMBOL(softnet_data
);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type
[] =
276 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
277 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
278 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
279 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
280 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
281 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
282 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
283 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
284 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
285 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
286 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
287 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
288 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
289 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
290 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
291 ARPHRD_VOID
, ARPHRD_NONE
};
293 static const char *const netdev_lock_name
[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
312 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
318 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
319 if (netdev_lock_type
[i
] == dev_type
)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type
) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
326 unsigned short dev_type
)
330 i
= netdev_lock_pos(dev_type
);
331 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
332 netdev_lock_name
[i
]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
339 i
= netdev_lock_pos(dev
->type
);
340 lockdep_set_class_and_name(&dev
->addr_list_lock
,
341 &netdev_addr_lock_key
[i
],
342 netdev_lock_name
[i
]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
346 unsigned short dev_type
)
349 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type
*pt
)
393 spin_lock_bh(&ptype_lock
);
394 if (pt
->type
== htons(ETH_P_ALL
))
395 list_add_rcu(&pt
->list
, &ptype_all
);
397 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
398 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
400 spin_unlock_bh(&ptype_lock
);
402 EXPORT_SYMBOL(dev_add_pack
);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type
*pt
)
419 struct list_head
*head
;
420 struct packet_type
*pt1
;
422 spin_lock_bh(&ptype_lock
);
424 if (pt
->type
== htons(ETH_P_ALL
))
427 head
= &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
429 list_for_each_entry(pt1
, head
, list
) {
431 list_del_rcu(&pt
->list
);
436 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
438 spin_unlock_bh(&ptype_lock
);
440 EXPORT_SYMBOL(__dev_remove_pack
);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type
*pt
)
456 __dev_remove_pack(pt
);
460 EXPORT_SYMBOL(dev_remove_pack
);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
482 struct netdev_boot_setup
*s
;
486 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
487 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
488 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
489 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
490 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
495 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device
*dev
)
509 struct netdev_boot_setup
*s
= dev_boot_setup
;
512 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
513 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
514 !strcmp(dev
->name
, s
[i
].name
)) {
515 dev
->irq
= s
[i
].map
.irq
;
516 dev
->base_addr
= s
[i
].map
.base_addr
;
517 dev
->mem_start
= s
[i
].map
.mem_start
;
518 dev
->mem_end
= s
[i
].map
.mem_end
;
524 EXPORT_SYMBOL(netdev_boot_setup_check
);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix
, int unit
)
539 const struct netdev_boot_setup
*s
= dev_boot_setup
;
543 sprintf(name
, "%s%d", prefix
, unit
);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net
, name
))
552 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
553 if (!strcmp(name
, s
[i
].name
))
554 return s
[i
].map
.base_addr
;
559 * Saves at boot time configured settings for any netdevice.
561 int __init
netdev_boot_setup(char *str
)
566 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
571 memset(&map
, 0, sizeof(map
));
575 map
.base_addr
= ints
[2];
577 map
.mem_start
= ints
[3];
579 map
.mem_end
= ints
[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str
, &map
);
585 __setup("netdev=", netdev_boot_setup
);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
607 struct hlist_node
*p
;
608 struct net_device
*dev
;
609 struct hlist_head
*head
= dev_name_hash(net
, name
);
611 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
612 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
617 EXPORT_SYMBOL(__dev_get_by_name
);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
633 struct hlist_node
*p
;
634 struct net_device
*dev
;
635 struct hlist_head
*head
= dev_name_hash(net
, name
);
637 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
638 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
643 EXPORT_SYMBOL(dev_get_by_name_rcu
);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
659 struct net_device
*dev
;
662 dev
= dev_get_by_name_rcu(net
, name
);
668 EXPORT_SYMBOL(dev_get_by_name
);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
684 struct hlist_node
*p
;
685 struct net_device
*dev
;
686 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
688 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
689 if (dev
->ifindex
== ifindex
)
694 EXPORT_SYMBOL(__dev_get_by_index
);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
709 struct hlist_node
*p
;
710 struct net_device
*dev
;
711 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
713 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
714 if (dev
->ifindex
== ifindex
)
719 EXPORT_SYMBOL(dev_get_by_index_rcu
);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
735 struct net_device
*dev
;
738 dev
= dev_get_by_index_rcu(net
, ifindex
);
744 EXPORT_SYMBOL(dev_get_by_index
);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
763 struct net_device
*dev
;
767 for_each_netdev(net
, dev
)
768 if (dev
->type
== type
&&
769 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
774 EXPORT_SYMBOL(dev_getbyhwaddr
);
776 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
778 struct net_device
*dev
;
781 for_each_netdev(net
, dev
)
782 if (dev
->type
== type
)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
789 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
791 struct net_device
*dev
, *ret
= NULL
;
794 for_each_netdev_rcu(net
, dev
)
795 if (dev
->type
== type
) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
806 * dev_get_by_flags_rcu - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. Must be called inside
813 * rcu_read_lock(), and result refcount is unchanged.
816 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
819 struct net_device
*dev
, *ret
;
822 for_each_netdev_rcu(net
, dev
) {
823 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
830 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
833 * dev_valid_name - check if name is okay for network device
836 * Network device names need to be valid file names to
837 * to allow sysfs to work. We also disallow any kind of
840 int dev_valid_name(const char *name
)
844 if (strlen(name
) >= IFNAMSIZ
)
846 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
850 if (*name
== '/' || isspace(*name
))
856 EXPORT_SYMBOL(dev_valid_name
);
859 * __dev_alloc_name - allocate a name for a device
860 * @net: network namespace to allocate the device name in
861 * @name: name format string
862 * @buf: scratch buffer and result name string
864 * Passed a format string - eg "lt%d" it will try and find a suitable
865 * id. It scans list of devices to build up a free map, then chooses
866 * the first empty slot. The caller must hold the dev_base or rtnl lock
867 * while allocating the name and adding the device in order to avoid
869 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
870 * Returns the number of the unit assigned or a negative errno code.
873 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
877 const int max_netdevices
= 8*PAGE_SIZE
;
878 unsigned long *inuse
;
879 struct net_device
*d
;
881 p
= strnchr(name
, IFNAMSIZ
-1, '%');
884 * Verify the string as this thing may have come from
885 * the user. There must be either one "%d" and no other "%"
888 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
891 /* Use one page as a bit array of possible slots */
892 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
896 for_each_netdev(net
, d
) {
897 if (!sscanf(d
->name
, name
, &i
))
899 if (i
< 0 || i
>= max_netdevices
)
902 /* avoid cases where sscanf is not exact inverse of printf */
903 snprintf(buf
, IFNAMSIZ
, name
, i
);
904 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
908 i
= find_first_zero_bit(inuse
, max_netdevices
);
909 free_page((unsigned long) inuse
);
913 snprintf(buf
, IFNAMSIZ
, name
, i
);
914 if (!__dev_get_by_name(net
, buf
))
917 /* It is possible to run out of possible slots
918 * when the name is long and there isn't enough space left
919 * for the digits, or if all bits are used.
925 * dev_alloc_name - allocate a name for a device
927 * @name: name format string
929 * Passed a format string - eg "lt%d" it will try and find a suitable
930 * id. It scans list of devices to build up a free map, then chooses
931 * the first empty slot. The caller must hold the dev_base or rtnl lock
932 * while allocating the name and adding the device in order to avoid
934 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
935 * Returns the number of the unit assigned or a negative errno code.
938 int dev_alloc_name(struct net_device
*dev
, const char *name
)
944 BUG_ON(!dev_net(dev
));
946 ret
= __dev_alloc_name(net
, name
, buf
);
948 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
951 EXPORT_SYMBOL(dev_alloc_name
);
953 static int dev_get_valid_name(struct net_device
*dev
, const char *name
, bool fmt
)
957 BUG_ON(!dev_net(dev
));
960 if (!dev_valid_name(name
))
963 if (fmt
&& strchr(name
, '%'))
964 return dev_alloc_name(dev
, name
);
965 else if (__dev_get_by_name(net
, name
))
967 else if (dev
->name
!= name
)
968 strlcpy(dev
->name
, name
, IFNAMSIZ
);
974 * dev_change_name - change name of a device
976 * @newname: name (or format string) must be at least IFNAMSIZ
978 * Change name of a device, can pass format strings "eth%d".
981 int dev_change_name(struct net_device
*dev
, const char *newname
)
983 char oldname
[IFNAMSIZ
];
989 BUG_ON(!dev_net(dev
));
992 if (dev
->flags
& IFF_UP
)
995 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
998 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1000 err
= dev_get_valid_name(dev
, newname
, 1);
1005 ret
= device_rename(&dev
->dev
, dev
->name
);
1007 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1011 write_lock_bh(&dev_base_lock
);
1012 hlist_del(&dev
->name_hlist
);
1013 write_unlock_bh(&dev_base_lock
);
1017 write_lock_bh(&dev_base_lock
);
1018 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1019 write_unlock_bh(&dev_base_lock
);
1021 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1022 ret
= notifier_to_errno(ret
);
1025 /* err >= 0 after dev_alloc_name() or stores the first errno */
1028 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1032 "%s: name change rollback failed: %d.\n",
1041 * dev_set_alias - change ifalias of a device
1043 * @alias: name up to IFALIASZ
1044 * @len: limit of bytes to copy from info
1046 * Set ifalias for a device,
1048 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1052 if (len
>= IFALIASZ
)
1057 kfree(dev
->ifalias
);
1058 dev
->ifalias
= NULL
;
1063 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1067 strlcpy(dev
->ifalias
, alias
, len
+1);
1073 * netdev_features_change - device changes features
1074 * @dev: device to cause notification
1076 * Called to indicate a device has changed features.
1078 void netdev_features_change(struct net_device
*dev
)
1080 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1082 EXPORT_SYMBOL(netdev_features_change
);
1085 * netdev_state_change - device changes state
1086 * @dev: device to cause notification
1088 * Called to indicate a device has changed state. This function calls
1089 * the notifier chains for netdev_chain and sends a NEWLINK message
1090 * to the routing socket.
1092 void netdev_state_change(struct net_device
*dev
)
1094 if (dev
->flags
& IFF_UP
) {
1095 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1096 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1099 EXPORT_SYMBOL(netdev_state_change
);
1101 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1103 return call_netdevice_notifiers(event
, dev
);
1105 EXPORT_SYMBOL(netdev_bonding_change
);
1108 * dev_load - load a network module
1109 * @net: the applicable net namespace
1110 * @name: name of interface
1112 * If a network interface is not present and the process has suitable
1113 * privileges this function loads the module. If module loading is not
1114 * available in this kernel then it becomes a nop.
1117 void dev_load(struct net
*net
, const char *name
)
1119 struct net_device
*dev
;
1122 dev
= dev_get_by_name_rcu(net
, name
);
1125 if (!dev
&& capable(CAP_NET_ADMIN
))
1126 request_module("%s", name
);
1128 EXPORT_SYMBOL(dev_load
);
1130 static int __dev_open(struct net_device
*dev
)
1132 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1138 * Is it even present?
1140 if (!netif_device_present(dev
))
1143 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1144 ret
= notifier_to_errno(ret
);
1149 * Call device private open method
1151 set_bit(__LINK_STATE_START
, &dev
->state
);
1153 if (ops
->ndo_validate_addr
)
1154 ret
= ops
->ndo_validate_addr(dev
);
1156 if (!ret
&& ops
->ndo_open
)
1157 ret
= ops
->ndo_open(dev
);
1160 * If it went open OK then:
1164 clear_bit(__LINK_STATE_START
, &dev
->state
);
1169 dev
->flags
|= IFF_UP
;
1174 net_dmaengine_get();
1177 * Initialize multicasting status
1179 dev_set_rx_mode(dev
);
1182 * Wakeup transmit queue engine
1191 * dev_open - prepare an interface for use.
1192 * @dev: device to open
1194 * Takes a device from down to up state. The device's private open
1195 * function is invoked and then the multicast lists are loaded. Finally
1196 * the device is moved into the up state and a %NETDEV_UP message is
1197 * sent to the netdev notifier chain.
1199 * Calling this function on an active interface is a nop. On a failure
1200 * a negative errno code is returned.
1202 int dev_open(struct net_device
*dev
)
1209 if (dev
->flags
& IFF_UP
)
1215 ret
= __dev_open(dev
);
1220 * ... and announce new interface.
1222 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1223 call_netdevice_notifiers(NETDEV_UP
, dev
);
1227 EXPORT_SYMBOL(dev_open
);
1229 static int __dev_close(struct net_device
*dev
)
1231 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1237 * Tell people we are going down, so that they can
1238 * prepare to death, when device is still operating.
1240 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1242 clear_bit(__LINK_STATE_START
, &dev
->state
);
1244 /* Synchronize to scheduled poll. We cannot touch poll list,
1245 * it can be even on different cpu. So just clear netif_running().
1247 * dev->stop() will invoke napi_disable() on all of it's
1248 * napi_struct instances on this device.
1250 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1252 dev_deactivate(dev
);
1255 * Call the device specific close. This cannot fail.
1256 * Only if device is UP
1258 * We allow it to be called even after a DETACH hot-plug
1265 * Device is now down.
1268 dev
->flags
&= ~IFF_UP
;
1273 net_dmaengine_put();
1279 * dev_close - shutdown an interface.
1280 * @dev: device to shutdown
1282 * This function moves an active device into down state. A
1283 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1284 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1287 int dev_close(struct net_device
*dev
)
1289 if (!(dev
->flags
& IFF_UP
))
1295 * Tell people we are down
1297 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1298 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1302 EXPORT_SYMBOL(dev_close
);
1306 * dev_disable_lro - disable Large Receive Offload on a device
1309 * Disable Large Receive Offload (LRO) on a net device. Must be
1310 * called under RTNL. This is needed if received packets may be
1311 * forwarded to another interface.
1313 void dev_disable_lro(struct net_device
*dev
)
1315 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1316 dev
->ethtool_ops
->set_flags
) {
1317 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1318 if (flags
& ETH_FLAG_LRO
) {
1319 flags
&= ~ETH_FLAG_LRO
;
1320 dev
->ethtool_ops
->set_flags(dev
, flags
);
1323 WARN_ON(dev
->features
& NETIF_F_LRO
);
1325 EXPORT_SYMBOL(dev_disable_lro
);
1328 static int dev_boot_phase
= 1;
1331 * Device change register/unregister. These are not inline or static
1332 * as we export them to the world.
1336 * register_netdevice_notifier - register a network notifier block
1339 * Register a notifier to be called when network device events occur.
1340 * The notifier passed is linked into the kernel structures and must
1341 * not be reused until it has been unregistered. A negative errno code
1342 * is returned on a failure.
1344 * When registered all registration and up events are replayed
1345 * to the new notifier to allow device to have a race free
1346 * view of the network device list.
1349 int register_netdevice_notifier(struct notifier_block
*nb
)
1351 struct net_device
*dev
;
1352 struct net_device
*last
;
1357 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1363 for_each_netdev(net
, dev
) {
1364 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1365 err
= notifier_to_errno(err
);
1369 if (!(dev
->flags
& IFF_UP
))
1372 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1383 for_each_netdev(net
, dev
) {
1387 if (dev
->flags
& IFF_UP
) {
1388 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1389 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1391 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1392 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1396 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1399 EXPORT_SYMBOL(register_netdevice_notifier
);
1402 * unregister_netdevice_notifier - unregister a network notifier block
1405 * Unregister a notifier previously registered by
1406 * register_netdevice_notifier(). The notifier is unlinked into the
1407 * kernel structures and may then be reused. A negative errno code
1408 * is returned on a failure.
1411 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1416 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1420 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1423 * call_netdevice_notifiers - call all network notifier blocks
1424 * @val: value passed unmodified to notifier function
1425 * @dev: net_device pointer passed unmodified to notifier function
1427 * Call all network notifier blocks. Parameters and return value
1428 * are as for raw_notifier_call_chain().
1431 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1434 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1437 /* When > 0 there are consumers of rx skb time stamps */
1438 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1440 void net_enable_timestamp(void)
1442 atomic_inc(&netstamp_needed
);
1444 EXPORT_SYMBOL(net_enable_timestamp
);
1446 void net_disable_timestamp(void)
1448 atomic_dec(&netstamp_needed
);
1450 EXPORT_SYMBOL(net_disable_timestamp
);
1452 static inline void net_timestamp_set(struct sk_buff
*skb
)
1454 if (atomic_read(&netstamp_needed
))
1455 __net_timestamp(skb
);
1457 skb
->tstamp
.tv64
= 0;
1460 static inline void net_timestamp_check(struct sk_buff
*skb
)
1462 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1463 __net_timestamp(skb
);
1467 * dev_forward_skb - loopback an skb to another netif
1469 * @dev: destination network device
1470 * @skb: buffer to forward
1473 * NET_RX_SUCCESS (no congestion)
1474 * NET_RX_DROP (packet was dropped, but freed)
1476 * dev_forward_skb can be used for injecting an skb from the
1477 * start_xmit function of one device into the receive queue
1478 * of another device.
1480 * The receiving device may be in another namespace, so
1481 * we have to clear all information in the skb that could
1482 * impact namespace isolation.
1484 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1489 if (!(dev
->flags
& IFF_UP
) ||
1490 (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
))) {
1494 skb_set_dev(skb
, dev
);
1495 skb
->tstamp
.tv64
= 0;
1496 skb
->pkt_type
= PACKET_HOST
;
1497 skb
->protocol
= eth_type_trans(skb
, dev
);
1498 return netif_rx(skb
);
1500 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1503 * Support routine. Sends outgoing frames to any network
1504 * taps currently in use.
1507 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1509 struct packet_type
*ptype
;
1511 #ifdef CONFIG_NET_CLS_ACT
1512 if (!(skb
->tstamp
.tv64
&& (G_TC_FROM(skb
->tc_verd
) & AT_INGRESS
)))
1513 net_timestamp_set(skb
);
1515 net_timestamp_set(skb
);
1519 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1520 /* Never send packets back to the socket
1521 * they originated from - MvS (miquels@drinkel.ow.org)
1523 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1524 (ptype
->af_packet_priv
== NULL
||
1525 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1526 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1530 /* skb->nh should be correctly
1531 set by sender, so that the second statement is
1532 just protection against buggy protocols.
1534 skb_reset_mac_header(skb2
);
1536 if (skb_network_header(skb2
) < skb2
->data
||
1537 skb2
->network_header
> skb2
->tail
) {
1538 if (net_ratelimit())
1539 printk(KERN_CRIT
"protocol %04x is "
1541 ntohs(skb2
->protocol
),
1543 skb_reset_network_header(skb2
);
1546 skb2
->transport_header
= skb2
->network_header
;
1547 skb2
->pkt_type
= PACKET_OUTGOING
;
1548 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1555 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1556 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1558 void netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1560 unsigned int real_num
= dev
->real_num_tx_queues
;
1562 if (unlikely(txq
> dev
->num_tx_queues
))
1564 else if (txq
> real_num
)
1565 dev
->real_num_tx_queues
= txq
;
1566 else if (txq
< real_num
) {
1567 dev
->real_num_tx_queues
= txq
;
1568 qdisc_reset_all_tx_gt(dev
, txq
);
1571 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1573 static inline void __netif_reschedule(struct Qdisc
*q
)
1575 struct softnet_data
*sd
;
1576 unsigned long flags
;
1578 local_irq_save(flags
);
1579 sd
= &__get_cpu_var(softnet_data
);
1580 q
->next_sched
= NULL
;
1581 *sd
->output_queue_tailp
= q
;
1582 sd
->output_queue_tailp
= &q
->next_sched
;
1583 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1584 local_irq_restore(flags
);
1587 void __netif_schedule(struct Qdisc
*q
)
1589 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1590 __netif_reschedule(q
);
1592 EXPORT_SYMBOL(__netif_schedule
);
1594 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1596 if (atomic_dec_and_test(&skb
->users
)) {
1597 struct softnet_data
*sd
;
1598 unsigned long flags
;
1600 local_irq_save(flags
);
1601 sd
= &__get_cpu_var(softnet_data
);
1602 skb
->next
= sd
->completion_queue
;
1603 sd
->completion_queue
= skb
;
1604 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1605 local_irq_restore(flags
);
1608 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1610 void dev_kfree_skb_any(struct sk_buff
*skb
)
1612 if (in_irq() || irqs_disabled())
1613 dev_kfree_skb_irq(skb
);
1617 EXPORT_SYMBOL(dev_kfree_skb_any
);
1621 * netif_device_detach - mark device as removed
1622 * @dev: network device
1624 * Mark device as removed from system and therefore no longer available.
1626 void netif_device_detach(struct net_device
*dev
)
1628 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1629 netif_running(dev
)) {
1630 netif_tx_stop_all_queues(dev
);
1633 EXPORT_SYMBOL(netif_device_detach
);
1636 * netif_device_attach - mark device as attached
1637 * @dev: network device
1639 * Mark device as attached from system and restart if needed.
1641 void netif_device_attach(struct net_device
*dev
)
1643 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1644 netif_running(dev
)) {
1645 netif_tx_wake_all_queues(dev
);
1646 __netdev_watchdog_up(dev
);
1649 EXPORT_SYMBOL(netif_device_attach
);
1651 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1653 return ((features
& NETIF_F_GEN_CSUM
) ||
1654 ((features
& NETIF_F_IP_CSUM
) &&
1655 protocol
== htons(ETH_P_IP
)) ||
1656 ((features
& NETIF_F_IPV6_CSUM
) &&
1657 protocol
== htons(ETH_P_IPV6
)) ||
1658 ((features
& NETIF_F_FCOE_CRC
) &&
1659 protocol
== htons(ETH_P_FCOE
)));
1662 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1664 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1667 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1668 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1669 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1670 veh
->h_vlan_encapsulated_proto
))
1678 * skb_dev_set -- assign a new device to a buffer
1679 * @skb: buffer for the new device
1680 * @dev: network device
1682 * If an skb is owned by a device already, we have to reset
1683 * all data private to the namespace a device belongs to
1684 * before assigning it a new device.
1686 #ifdef CONFIG_NET_NS
1687 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1690 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1693 skb_init_secmark(skb
);
1697 skb
->ipvs_property
= 0;
1698 #ifdef CONFIG_NET_SCHED
1704 EXPORT_SYMBOL(skb_set_dev
);
1705 #endif /* CONFIG_NET_NS */
1708 * Invalidate hardware checksum when packet is to be mangled, and
1709 * complete checksum manually on outgoing path.
1711 int skb_checksum_help(struct sk_buff
*skb
)
1714 int ret
= 0, offset
;
1716 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1717 goto out_set_summed
;
1719 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1720 /* Let GSO fix up the checksum. */
1721 goto out_set_summed
;
1724 offset
= skb
->csum_start
- skb_headroom(skb
);
1725 BUG_ON(offset
>= skb_headlen(skb
));
1726 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1728 offset
+= skb
->csum_offset
;
1729 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1731 if (skb_cloned(skb
) &&
1732 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1733 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1738 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1740 skb
->ip_summed
= CHECKSUM_NONE
;
1744 EXPORT_SYMBOL(skb_checksum_help
);
1747 * skb_gso_segment - Perform segmentation on skb.
1748 * @skb: buffer to segment
1749 * @features: features for the output path (see dev->features)
1751 * This function segments the given skb and returns a list of segments.
1753 * It may return NULL if the skb requires no segmentation. This is
1754 * only possible when GSO is used for verifying header integrity.
1756 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1758 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1759 struct packet_type
*ptype
;
1760 __be16 type
= skb
->protocol
;
1763 skb_reset_mac_header(skb
);
1764 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1765 __skb_pull(skb
, skb
->mac_len
);
1767 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1768 struct net_device
*dev
= skb
->dev
;
1769 struct ethtool_drvinfo info
= {};
1771 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1772 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1774 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1776 info
.driver
, dev
? dev
->features
: 0L,
1777 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1778 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1780 if (skb_header_cloned(skb
) &&
1781 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1782 return ERR_PTR(err
);
1786 list_for_each_entry_rcu(ptype
,
1787 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1788 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1789 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1790 err
= ptype
->gso_send_check(skb
);
1791 segs
= ERR_PTR(err
);
1792 if (err
|| skb_gso_ok(skb
, features
))
1794 __skb_push(skb
, (skb
->data
-
1795 skb_network_header(skb
)));
1797 segs
= ptype
->gso_segment(skb
, features
);
1803 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1807 EXPORT_SYMBOL(skb_gso_segment
);
1809 /* Take action when hardware reception checksum errors are detected. */
1811 void netdev_rx_csum_fault(struct net_device
*dev
)
1813 if (net_ratelimit()) {
1814 printk(KERN_ERR
"%s: hw csum failure.\n",
1815 dev
? dev
->name
: "<unknown>");
1819 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1822 /* Actually, we should eliminate this check as soon as we know, that:
1823 * 1. IOMMU is present and allows to map all the memory.
1824 * 2. No high memory really exists on this machine.
1827 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1829 #ifdef CONFIG_HIGHMEM
1831 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1832 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1833 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1837 if (PCI_DMA_BUS_IS_PHYS
) {
1838 struct device
*pdev
= dev
->dev
.parent
;
1842 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1843 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1844 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1853 void (*destructor
)(struct sk_buff
*skb
);
1856 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1858 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1860 struct dev_gso_cb
*cb
;
1863 struct sk_buff
*nskb
= skb
->next
;
1865 skb
->next
= nskb
->next
;
1868 } while (skb
->next
);
1870 cb
= DEV_GSO_CB(skb
);
1872 cb
->destructor(skb
);
1876 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1877 * @skb: buffer to segment
1879 * This function segments the given skb and stores the list of segments
1882 static int dev_gso_segment(struct sk_buff
*skb
)
1884 struct net_device
*dev
= skb
->dev
;
1885 struct sk_buff
*segs
;
1886 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1889 segs
= skb_gso_segment(skb
, features
);
1891 /* Verifying header integrity only. */
1896 return PTR_ERR(segs
);
1899 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1900 skb
->destructor
= dev_gso_skb_destructor
;
1906 * Try to orphan skb early, right before transmission by the device.
1907 * We cannot orphan skb if tx timestamp is requested, since
1908 * drivers need to call skb_tstamp_tx() to send the timestamp.
1910 static inline void skb_orphan_try(struct sk_buff
*skb
)
1912 struct sock
*sk
= skb
->sk
;
1914 if (sk
&& !skb_tx(skb
)->flags
) {
1915 /* skb_tx_hash() wont be able to get sk.
1916 * We copy sk_hash into skb->rxhash
1919 skb
->rxhash
= sk
->sk_hash
;
1925 * Returns true if either:
1926 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1927 * 2. skb is fragmented and the device does not support SG, or if
1928 * at least one of fragments is in highmem and device does not
1929 * support DMA from it.
1931 static inline int skb_needs_linearize(struct sk_buff
*skb
,
1932 struct net_device
*dev
)
1934 return skb_is_nonlinear(skb
) &&
1935 ((skb_has_frags(skb
) && !(dev
->features
& NETIF_F_FRAGLIST
)) ||
1936 (skb_shinfo(skb
)->nr_frags
&& (!(dev
->features
& NETIF_F_SG
) ||
1937 illegal_highdma(dev
, skb
))));
1940 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1941 struct netdev_queue
*txq
)
1943 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1944 int rc
= NETDEV_TX_OK
;
1946 if (likely(!skb
->next
)) {
1947 if (!list_empty(&ptype_all
))
1948 dev_queue_xmit_nit(skb
, dev
);
1951 * If device doesnt need skb->dst, release it right now while
1952 * its hot in this cpu cache
1954 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1957 skb_orphan_try(skb
);
1959 if (netif_needs_gso(dev
, skb
)) {
1960 if (unlikely(dev_gso_segment(skb
)))
1965 if (skb_needs_linearize(skb
, dev
) &&
1966 __skb_linearize(skb
))
1969 /* If packet is not checksummed and device does not
1970 * support checksumming for this protocol, complete
1971 * checksumming here.
1973 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1974 skb_set_transport_header(skb
, skb
->csum_start
-
1976 if (!dev_can_checksum(dev
, skb
) &&
1977 skb_checksum_help(skb
))
1982 rc
= ops
->ndo_start_xmit(skb
, dev
);
1983 trace_net_dev_xmit(skb
, rc
);
1984 if (rc
== NETDEV_TX_OK
)
1985 txq_trans_update(txq
);
1991 struct sk_buff
*nskb
= skb
->next
;
1993 skb
->next
= nskb
->next
;
1997 * If device doesnt need nskb->dst, release it right now while
1998 * its hot in this cpu cache
2000 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2003 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2004 trace_net_dev_xmit(nskb
, rc
);
2005 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2006 if (rc
& ~NETDEV_TX_MASK
)
2007 goto out_kfree_gso_skb
;
2008 nskb
->next
= skb
->next
;
2012 txq_trans_update(txq
);
2013 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2014 return NETDEV_TX_BUSY
;
2015 } while (skb
->next
);
2018 if (likely(skb
->next
== NULL
))
2019 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2025 static u32 hashrnd __read_mostly
;
2027 u16
skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
)
2031 if (skb_rx_queue_recorded(skb
)) {
2032 hash
= skb_get_rx_queue(skb
);
2033 while (unlikely(hash
>= dev
->real_num_tx_queues
))
2034 hash
-= dev
->real_num_tx_queues
;
2038 if (skb
->sk
&& skb
->sk
->sk_hash
)
2039 hash
= skb
->sk
->sk_hash
;
2041 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2042 hash
= jhash_1word(hash
, hashrnd
);
2044 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
2046 EXPORT_SYMBOL(skb_tx_hash
);
2048 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2050 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2051 if (net_ratelimit()) {
2052 pr_warning("%s selects TX queue %d, but "
2053 "real number of TX queues is %d\n",
2054 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2061 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2062 struct sk_buff
*skb
)
2065 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2067 if (ops
->ndo_select_queue
) {
2068 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2069 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2071 struct sock
*sk
= skb
->sk
;
2072 queue_index
= sk_tx_queue_get(sk
);
2073 if (queue_index
< 0) {
2076 if (dev
->real_num_tx_queues
> 1)
2077 queue_index
= skb_tx_hash(dev
, skb
);
2080 struct dst_entry
*dst
= rcu_dereference_check(sk
->sk_dst_cache
, 1);
2082 if (dst
&& skb_dst(skb
) == dst
)
2083 sk_tx_queue_set(sk
, queue_index
);
2088 skb_set_queue_mapping(skb
, queue_index
);
2089 return netdev_get_tx_queue(dev
, queue_index
);
2092 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2093 struct net_device
*dev
,
2094 struct netdev_queue
*txq
)
2096 spinlock_t
*root_lock
= qdisc_lock(q
);
2097 bool contended
= qdisc_is_running(q
);
2101 * Heuristic to force contended enqueues to serialize on a
2102 * separate lock before trying to get qdisc main lock.
2103 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2104 * and dequeue packets faster.
2106 if (unlikely(contended
))
2107 spin_lock(&q
->busylock
);
2109 spin_lock(root_lock
);
2110 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2113 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2114 qdisc_run_begin(q
)) {
2116 * This is a work-conserving queue; there are no old skbs
2117 * waiting to be sent out; and the qdisc is not running -
2118 * xmit the skb directly.
2120 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2122 __qdisc_update_bstats(q
, skb
->len
);
2123 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2124 if (unlikely(contended
)) {
2125 spin_unlock(&q
->busylock
);
2132 rc
= NET_XMIT_SUCCESS
;
2135 rc
= qdisc_enqueue_root(skb
, q
);
2136 if (qdisc_run_begin(q
)) {
2137 if (unlikely(contended
)) {
2138 spin_unlock(&q
->busylock
);
2144 spin_unlock(root_lock
);
2145 if (unlikely(contended
))
2146 spin_unlock(&q
->busylock
);
2151 * dev_queue_xmit - transmit a buffer
2152 * @skb: buffer to transmit
2154 * Queue a buffer for transmission to a network device. The caller must
2155 * have set the device and priority and built the buffer before calling
2156 * this function. The function can be called from an interrupt.
2158 * A negative errno code is returned on a failure. A success does not
2159 * guarantee the frame will be transmitted as it may be dropped due
2160 * to congestion or traffic shaping.
2162 * -----------------------------------------------------------------------------------
2163 * I notice this method can also return errors from the queue disciplines,
2164 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2167 * Regardless of the return value, the skb is consumed, so it is currently
2168 * difficult to retry a send to this method. (You can bump the ref count
2169 * before sending to hold a reference for retry if you are careful.)
2171 * When calling this method, interrupts MUST be enabled. This is because
2172 * the BH enable code must have IRQs enabled so that it will not deadlock.
2175 int dev_queue_xmit(struct sk_buff
*skb
)
2177 struct net_device
*dev
= skb
->dev
;
2178 struct netdev_queue
*txq
;
2182 /* Disable soft irqs for various locks below. Also
2183 * stops preemption for RCU.
2187 txq
= dev_pick_tx(dev
, skb
);
2188 q
= rcu_dereference_bh(txq
->qdisc
);
2190 #ifdef CONFIG_NET_CLS_ACT
2191 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2193 trace_net_dev_queue(skb
);
2195 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2199 /* The device has no queue. Common case for software devices:
2200 loopback, all the sorts of tunnels...
2202 Really, it is unlikely that netif_tx_lock protection is necessary
2203 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2205 However, it is possible, that they rely on protection
2208 Check this and shot the lock. It is not prone from deadlocks.
2209 Either shot noqueue qdisc, it is even simpler 8)
2211 if (dev
->flags
& IFF_UP
) {
2212 int cpu
= smp_processor_id(); /* ok because BHs are off */
2214 if (txq
->xmit_lock_owner
!= cpu
) {
2216 HARD_TX_LOCK(dev
, txq
, cpu
);
2218 if (!netif_tx_queue_stopped(txq
)) {
2219 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2220 if (dev_xmit_complete(rc
)) {
2221 HARD_TX_UNLOCK(dev
, txq
);
2225 HARD_TX_UNLOCK(dev
, txq
);
2226 if (net_ratelimit())
2227 printk(KERN_CRIT
"Virtual device %s asks to "
2228 "queue packet!\n", dev
->name
);
2230 /* Recursion is detected! It is possible,
2232 if (net_ratelimit())
2233 printk(KERN_CRIT
"Dead loop on virtual device "
2234 "%s, fix it urgently!\n", dev
->name
);
2239 rcu_read_unlock_bh();
2244 rcu_read_unlock_bh();
2247 EXPORT_SYMBOL(dev_queue_xmit
);
2250 /*=======================================================================
2252 =======================================================================*/
2254 int netdev_max_backlog __read_mostly
= 1000;
2255 int netdev_tstamp_prequeue __read_mostly
= 1;
2256 int netdev_budget __read_mostly
= 300;
2257 int weight_p __read_mostly
= 64; /* old backlog weight */
2259 /* Called with irq disabled */
2260 static inline void ____napi_schedule(struct softnet_data
*sd
,
2261 struct napi_struct
*napi
)
2263 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2264 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2269 /* One global table that all flow-based protocols share. */
2270 struct rps_sock_flow_table
*rps_sock_flow_table __read_mostly
;
2271 EXPORT_SYMBOL(rps_sock_flow_table
);
2274 * get_rps_cpu is called from netif_receive_skb and returns the target
2275 * CPU from the RPS map of the receiving queue for a given skb.
2276 * rcu_read_lock must be held on entry.
2278 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2279 struct rps_dev_flow
**rflowp
)
2281 struct ipv6hdr
*ip6
;
2283 struct netdev_rx_queue
*rxqueue
;
2284 struct rps_map
*map
;
2285 struct rps_dev_flow_table
*flow_table
;
2286 struct rps_sock_flow_table
*sock_flow_table
;
2290 u32 addr1
, addr2
, ihl
;
2296 if (skb_rx_queue_recorded(skb
)) {
2297 u16 index
= skb_get_rx_queue(skb
);
2298 if (unlikely(index
>= dev
->num_rx_queues
)) {
2299 WARN_ONCE(dev
->num_rx_queues
> 1, "%s received packet "
2300 "on queue %u, but number of RX queues is %u\n",
2301 dev
->name
, index
, dev
->num_rx_queues
);
2304 rxqueue
= dev
->_rx
+ index
;
2308 if (!rxqueue
->rps_map
&& !rxqueue
->rps_flow_table
)
2312 goto got_hash
; /* Skip hash computation on packet header */
2314 switch (skb
->protocol
) {
2315 case __constant_htons(ETH_P_IP
):
2316 if (!pskb_may_pull(skb
, sizeof(*ip
)))
2319 ip
= (struct iphdr
*) skb
->data
;
2320 ip_proto
= ip
->protocol
;
2321 addr1
= (__force u32
) ip
->saddr
;
2322 addr2
= (__force u32
) ip
->daddr
;
2325 case __constant_htons(ETH_P_IPV6
):
2326 if (!pskb_may_pull(skb
, sizeof(*ip6
)))
2329 ip6
= (struct ipv6hdr
*) skb
->data
;
2330 ip_proto
= ip6
->nexthdr
;
2331 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2332 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2345 case IPPROTO_UDPLITE
:
2346 if (pskb_may_pull(skb
, (ihl
* 4) + 4)) {
2347 ports
.v32
= * (__force u32
*) (skb
->data
+ (ihl
* 4));
2348 if (ports
.v16
[1] < ports
.v16
[0])
2349 swap(ports
.v16
[0], ports
.v16
[1]);
2357 /* get a consistent hash (same value on both flow directions) */
2360 skb
->rxhash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2365 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2366 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2367 if (flow_table
&& sock_flow_table
) {
2369 struct rps_dev_flow
*rflow
;
2371 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2374 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2375 sock_flow_table
->mask
];
2378 * If the desired CPU (where last recvmsg was done) is
2379 * different from current CPU (one in the rx-queue flow
2380 * table entry), switch if one of the following holds:
2381 * - Current CPU is unset (equal to RPS_NO_CPU).
2382 * - Current CPU is offline.
2383 * - The current CPU's queue tail has advanced beyond the
2384 * last packet that was enqueued using this table entry.
2385 * This guarantees that all previous packets for the flow
2386 * have been dequeued, thus preserving in order delivery.
2388 if (unlikely(tcpu
!= next_cpu
) &&
2389 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2390 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2391 rflow
->last_qtail
)) >= 0)) {
2392 tcpu
= rflow
->cpu
= next_cpu
;
2393 if (tcpu
!= RPS_NO_CPU
)
2394 rflow
->last_qtail
= per_cpu(softnet_data
,
2395 tcpu
).input_queue_head
;
2397 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2404 map
= rcu_dereference(rxqueue
->rps_map
);
2406 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2408 if (cpu_online(tcpu
)) {
2418 /* Called from hardirq (IPI) context */
2419 static void rps_trigger_softirq(void *data
)
2421 struct softnet_data
*sd
= data
;
2423 ____napi_schedule(sd
, &sd
->backlog
);
2427 #endif /* CONFIG_RPS */
2430 * Check if this softnet_data structure is another cpu one
2431 * If yes, queue it to our IPI list and return 1
2434 static int rps_ipi_queued(struct softnet_data
*sd
)
2437 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2440 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2441 mysd
->rps_ipi_list
= sd
;
2443 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2446 #endif /* CONFIG_RPS */
2451 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2452 * queue (may be a remote CPU queue).
2454 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2455 unsigned int *qtail
)
2457 struct softnet_data
*sd
;
2458 unsigned long flags
;
2460 sd
= &per_cpu(softnet_data
, cpu
);
2462 local_irq_save(flags
);
2465 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2466 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2468 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2469 input_queue_tail_incr_save(sd
, qtail
);
2471 local_irq_restore(flags
);
2472 return NET_RX_SUCCESS
;
2475 /* Schedule NAPI for backlog device
2476 * We can use non atomic operation since we own the queue lock
2478 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2479 if (!rps_ipi_queued(sd
))
2480 ____napi_schedule(sd
, &sd
->backlog
);
2488 local_irq_restore(flags
);
2495 * netif_rx - post buffer to the network code
2496 * @skb: buffer to post
2498 * This function receives a packet from a device driver and queues it for
2499 * the upper (protocol) levels to process. It always succeeds. The buffer
2500 * may be dropped during processing for congestion control or by the
2504 * NET_RX_SUCCESS (no congestion)
2505 * NET_RX_DROP (packet was dropped)
2509 int netif_rx(struct sk_buff
*skb
)
2513 /* if netpoll wants it, pretend we never saw it */
2514 if (netpoll_rx(skb
))
2517 if (netdev_tstamp_prequeue
)
2518 net_timestamp_check(skb
);
2520 trace_netif_rx(skb
);
2523 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2529 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2531 cpu
= smp_processor_id();
2533 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2541 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2547 EXPORT_SYMBOL(netif_rx
);
2549 int netif_rx_ni(struct sk_buff
*skb
)
2554 err
= netif_rx(skb
);
2555 if (local_softirq_pending())
2561 EXPORT_SYMBOL(netif_rx_ni
);
2563 static void net_tx_action(struct softirq_action
*h
)
2565 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2567 if (sd
->completion_queue
) {
2568 struct sk_buff
*clist
;
2570 local_irq_disable();
2571 clist
= sd
->completion_queue
;
2572 sd
->completion_queue
= NULL
;
2576 struct sk_buff
*skb
= clist
;
2577 clist
= clist
->next
;
2579 WARN_ON(atomic_read(&skb
->users
));
2580 trace_kfree_skb(skb
, net_tx_action
);
2585 if (sd
->output_queue
) {
2588 local_irq_disable();
2589 head
= sd
->output_queue
;
2590 sd
->output_queue
= NULL
;
2591 sd
->output_queue_tailp
= &sd
->output_queue
;
2595 struct Qdisc
*q
= head
;
2596 spinlock_t
*root_lock
;
2598 head
= head
->next_sched
;
2600 root_lock
= qdisc_lock(q
);
2601 if (spin_trylock(root_lock
)) {
2602 smp_mb__before_clear_bit();
2603 clear_bit(__QDISC_STATE_SCHED
,
2606 spin_unlock(root_lock
);
2608 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2610 __netif_reschedule(q
);
2612 smp_mb__before_clear_bit();
2613 clear_bit(__QDISC_STATE_SCHED
,
2621 static inline int deliver_skb(struct sk_buff
*skb
,
2622 struct packet_type
*pt_prev
,
2623 struct net_device
*orig_dev
)
2625 atomic_inc(&skb
->users
);
2626 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2629 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2630 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2631 /* This hook is defined here for ATM LANE */
2632 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2633 unsigned char *addr
) __read_mostly
;
2634 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2637 #ifdef CONFIG_NET_CLS_ACT
2638 /* TODO: Maybe we should just force sch_ingress to be compiled in
2639 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2640 * a compare and 2 stores extra right now if we dont have it on
2641 * but have CONFIG_NET_CLS_ACT
2642 * NOTE: This doesnt stop any functionality; if you dont have
2643 * the ingress scheduler, you just cant add policies on ingress.
2646 static int ing_filter(struct sk_buff
*skb
)
2648 struct net_device
*dev
= skb
->dev
;
2649 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2650 struct netdev_queue
*rxq
;
2651 int result
= TC_ACT_OK
;
2654 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
2655 if (net_ratelimit())
2656 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2657 skb
->skb_iif
, dev
->ifindex
);
2661 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2662 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2664 rxq
= &dev
->rx_queue
;
2667 if (q
!= &noop_qdisc
) {
2668 spin_lock(qdisc_lock(q
));
2669 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2670 result
= qdisc_enqueue_root(skb
, q
);
2671 spin_unlock(qdisc_lock(q
));
2677 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2678 struct packet_type
**pt_prev
,
2679 int *ret
, struct net_device
*orig_dev
)
2681 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2685 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2689 switch (ing_filter(skb
)) {
2703 * netif_nit_deliver - deliver received packets to network taps
2706 * This function is used to deliver incoming packets to network
2707 * taps. It should be used when the normal netif_receive_skb path
2708 * is bypassed, for example because of VLAN acceleration.
2710 void netif_nit_deliver(struct sk_buff
*skb
)
2712 struct packet_type
*ptype
;
2714 if (list_empty(&ptype_all
))
2717 skb_reset_network_header(skb
);
2718 skb_reset_transport_header(skb
);
2719 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2722 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2723 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2724 deliver_skb(skb
, ptype
, skb
->dev
);
2730 * netdev_rx_handler_register - register receive handler
2731 * @dev: device to register a handler for
2732 * @rx_handler: receive handler to register
2733 * @rx_handler_data: data pointer that is used by rx handler
2735 * Register a receive hander for a device. This handler will then be
2736 * called from __netif_receive_skb. A negative errno code is returned
2739 * The caller must hold the rtnl_mutex.
2741 int netdev_rx_handler_register(struct net_device
*dev
,
2742 rx_handler_func_t
*rx_handler
,
2743 void *rx_handler_data
)
2747 if (dev
->rx_handler
)
2750 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
2751 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
2755 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
2758 * netdev_rx_handler_unregister - unregister receive handler
2759 * @dev: device to unregister a handler from
2761 * Unregister a receive hander from a device.
2763 * The caller must hold the rtnl_mutex.
2765 void netdev_rx_handler_unregister(struct net_device
*dev
)
2769 rcu_assign_pointer(dev
->rx_handler
, NULL
);
2770 rcu_assign_pointer(dev
->rx_handler_data
, NULL
);
2772 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
2774 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2775 struct net_device
*master
)
2777 if (skb
->pkt_type
== PACKET_HOST
) {
2778 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2780 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2784 /* On bonding slaves other than the currently active slave, suppress
2785 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2786 * ARP on active-backup slaves with arp_validate enabled.
2788 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2790 struct net_device
*dev
= skb
->dev
;
2792 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2793 dev
->last_rx
= jiffies
;
2795 if ((master
->priv_flags
& IFF_MASTER_ALB
) &&
2796 (master
->priv_flags
& IFF_BRIDGE_PORT
)) {
2797 /* Do address unmangle. The local destination address
2798 * will be always the one master has. Provides the right
2799 * functionality in a bridge.
2801 skb_bond_set_mac_by_master(skb
, master
);
2804 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
2805 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
2806 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
2809 if (master
->priv_flags
& IFF_MASTER_ALB
) {
2810 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
2811 skb
->pkt_type
!= PACKET_MULTICAST
)
2814 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
2815 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
2822 EXPORT_SYMBOL(__skb_bond_should_drop
);
2824 static int __netif_receive_skb(struct sk_buff
*skb
)
2826 struct packet_type
*ptype
, *pt_prev
;
2827 rx_handler_func_t
*rx_handler
;
2828 struct net_device
*orig_dev
;
2829 struct net_device
*master
;
2830 struct net_device
*null_or_orig
;
2831 struct net_device
*orig_or_bond
;
2832 int ret
= NET_RX_DROP
;
2835 if (!netdev_tstamp_prequeue
)
2836 net_timestamp_check(skb
);
2838 trace_netif_receive_skb(skb
);
2839 if (vlan_tx_tag_present(skb
) && vlan_hwaccel_do_receive(skb
))
2840 return NET_RX_SUCCESS
;
2842 /* if we've gotten here through NAPI, check netpoll */
2843 if (netpoll_receive_skb(skb
))
2847 skb
->skb_iif
= skb
->dev
->ifindex
;
2850 * bonding note: skbs received on inactive slaves should only
2851 * be delivered to pkt handlers that are exact matches. Also
2852 * the deliver_no_wcard flag will be set. If packet handlers
2853 * are sensitive to duplicate packets these skbs will need to
2854 * be dropped at the handler. The vlan accel path may have
2855 * already set the deliver_no_wcard flag.
2857 null_or_orig
= NULL
;
2858 orig_dev
= skb
->dev
;
2859 master
= ACCESS_ONCE(orig_dev
->master
);
2860 if (skb
->deliver_no_wcard
)
2861 null_or_orig
= orig_dev
;
2863 if (skb_bond_should_drop(skb
, master
)) {
2864 skb
->deliver_no_wcard
= 1;
2865 null_or_orig
= orig_dev
; /* deliver only exact match */
2870 __this_cpu_inc(softnet_data
.processed
);
2871 skb_reset_network_header(skb
);
2872 skb_reset_transport_header(skb
);
2873 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2879 #ifdef CONFIG_NET_CLS_ACT
2880 if (skb
->tc_verd
& TC_NCLS
) {
2881 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2886 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2887 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2888 ptype
->dev
== orig_dev
) {
2890 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2895 #ifdef CONFIG_NET_CLS_ACT
2896 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2902 /* Handle special case of bridge or macvlan */
2903 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
2906 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2909 skb
= rx_handler(skb
);
2915 * Make sure frames received on VLAN interfaces stacked on
2916 * bonding interfaces still make their way to any base bonding
2917 * device that may have registered for a specific ptype. The
2918 * handler may have to adjust skb->dev and orig_dev.
2920 orig_or_bond
= orig_dev
;
2921 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
2922 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
2923 orig_or_bond
= vlan_dev_real_dev(skb
->dev
);
2926 type
= skb
->protocol
;
2927 list_for_each_entry_rcu(ptype
,
2928 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2929 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
2930 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
2931 ptype
->dev
== orig_or_bond
)) {
2933 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2939 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2942 /* Jamal, now you will not able to escape explaining
2943 * me how you were going to use this. :-)
2954 * netif_receive_skb - process receive buffer from network
2955 * @skb: buffer to process
2957 * netif_receive_skb() is the main receive data processing function.
2958 * It always succeeds. The buffer may be dropped during processing
2959 * for congestion control or by the protocol layers.
2961 * This function may only be called from softirq context and interrupts
2962 * should be enabled.
2964 * Return values (usually ignored):
2965 * NET_RX_SUCCESS: no congestion
2966 * NET_RX_DROP: packet was dropped
2968 int netif_receive_skb(struct sk_buff
*skb
)
2970 if (netdev_tstamp_prequeue
)
2971 net_timestamp_check(skb
);
2973 if (skb_defer_rx_timestamp(skb
))
2974 return NET_RX_SUCCESS
;
2978 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2983 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2986 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2990 ret
= __netif_receive_skb(skb
);
2996 return __netif_receive_skb(skb
);
2999 EXPORT_SYMBOL(netif_receive_skb
);
3001 /* Network device is going away, flush any packets still pending
3002 * Called with irqs disabled.
3004 static void flush_backlog(void *arg
)
3006 struct net_device
*dev
= arg
;
3007 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3008 struct sk_buff
*skb
, *tmp
;
3011 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3012 if (skb
->dev
== dev
) {
3013 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3015 input_queue_head_incr(sd
);
3020 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3021 if (skb
->dev
== dev
) {
3022 __skb_unlink(skb
, &sd
->process_queue
);
3024 input_queue_head_incr(sd
);
3029 static int napi_gro_complete(struct sk_buff
*skb
)
3031 struct packet_type
*ptype
;
3032 __be16 type
= skb
->protocol
;
3033 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3036 if (NAPI_GRO_CB(skb
)->count
== 1) {
3037 skb_shinfo(skb
)->gso_size
= 0;
3042 list_for_each_entry_rcu(ptype
, head
, list
) {
3043 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3046 err
= ptype
->gro_complete(skb
);
3052 WARN_ON(&ptype
->list
== head
);
3054 return NET_RX_SUCCESS
;
3058 return netif_receive_skb(skb
);
3061 static void napi_gro_flush(struct napi_struct
*napi
)
3063 struct sk_buff
*skb
, *next
;
3065 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3068 napi_gro_complete(skb
);
3071 napi
->gro_count
= 0;
3072 napi
->gro_list
= NULL
;
3075 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3077 struct sk_buff
**pp
= NULL
;
3078 struct packet_type
*ptype
;
3079 __be16 type
= skb
->protocol
;
3080 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3083 enum gro_result ret
;
3085 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3088 if (skb_is_gso(skb
) || skb_has_frags(skb
))
3092 list_for_each_entry_rcu(ptype
, head
, list
) {
3093 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3096 skb_set_network_header(skb
, skb_gro_offset(skb
));
3097 mac_len
= skb
->network_header
- skb
->mac_header
;
3098 skb
->mac_len
= mac_len
;
3099 NAPI_GRO_CB(skb
)->same_flow
= 0;
3100 NAPI_GRO_CB(skb
)->flush
= 0;
3101 NAPI_GRO_CB(skb
)->free
= 0;
3103 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3108 if (&ptype
->list
== head
)
3111 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3112 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3115 struct sk_buff
*nskb
= *pp
;
3119 napi_gro_complete(nskb
);
3126 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3130 NAPI_GRO_CB(skb
)->count
= 1;
3131 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3132 skb
->next
= napi
->gro_list
;
3133 napi
->gro_list
= skb
;
3137 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3138 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3140 BUG_ON(skb
->end
- skb
->tail
< grow
);
3142 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3145 skb
->data_len
-= grow
;
3147 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3148 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3150 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3151 put_page(skb_shinfo(skb
)->frags
[0].page
);
3152 memmove(skb_shinfo(skb
)->frags
,
3153 skb_shinfo(skb
)->frags
+ 1,
3154 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3165 EXPORT_SYMBOL(dev_gro_receive
);
3168 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3172 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3173 NAPI_GRO_CB(p
)->same_flow
=
3174 (p
->dev
== skb
->dev
) &&
3175 !compare_ether_header(skb_mac_header(p
),
3176 skb_gro_mac_header(skb
));
3177 NAPI_GRO_CB(p
)->flush
= 0;
3180 return dev_gro_receive(napi
, skb
);
3183 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3187 if (netif_receive_skb(skb
))
3192 case GRO_MERGED_FREE
:
3203 EXPORT_SYMBOL(napi_skb_finish
);
3205 void skb_gro_reset_offset(struct sk_buff
*skb
)
3207 NAPI_GRO_CB(skb
)->data_offset
= 0;
3208 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3209 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3211 if (skb
->mac_header
== skb
->tail
&&
3212 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3213 NAPI_GRO_CB(skb
)->frag0
=
3214 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3215 skb_shinfo(skb
)->frags
[0].page_offset
;
3216 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3219 EXPORT_SYMBOL(skb_gro_reset_offset
);
3221 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3223 skb_gro_reset_offset(skb
);
3225 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3227 EXPORT_SYMBOL(napi_gro_receive
);
3229 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3231 __skb_pull(skb
, skb_headlen(skb
));
3232 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3236 EXPORT_SYMBOL(napi_reuse_skb
);
3238 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3240 struct sk_buff
*skb
= napi
->skb
;
3243 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3249 EXPORT_SYMBOL(napi_get_frags
);
3251 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3257 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3259 if (ret
== GRO_HELD
)
3260 skb_gro_pull(skb
, -ETH_HLEN
);
3261 else if (netif_receive_skb(skb
))
3266 case GRO_MERGED_FREE
:
3267 napi_reuse_skb(napi
, skb
);
3276 EXPORT_SYMBOL(napi_frags_finish
);
3278 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3280 struct sk_buff
*skb
= napi
->skb
;
3287 skb_reset_mac_header(skb
);
3288 skb_gro_reset_offset(skb
);
3290 off
= skb_gro_offset(skb
);
3291 hlen
= off
+ sizeof(*eth
);
3292 eth
= skb_gro_header_fast(skb
, off
);
3293 if (skb_gro_header_hard(skb
, hlen
)) {
3294 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3295 if (unlikely(!eth
)) {
3296 napi_reuse_skb(napi
, skb
);
3302 skb_gro_pull(skb
, sizeof(*eth
));
3305 * This works because the only protocols we care about don't require
3306 * special handling. We'll fix it up properly at the end.
3308 skb
->protocol
= eth
->h_proto
;
3313 EXPORT_SYMBOL(napi_frags_skb
);
3315 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3317 struct sk_buff
*skb
= napi_frags_skb(napi
);
3322 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3324 EXPORT_SYMBOL(napi_gro_frags
);
3327 * net_rps_action sends any pending IPI's for rps.
3328 * Note: called with local irq disabled, but exits with local irq enabled.
3330 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3333 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3336 sd
->rps_ipi_list
= NULL
;
3340 /* Send pending IPI's to kick RPS processing on remote cpus. */
3342 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3344 if (cpu_online(remsd
->cpu
))
3345 __smp_call_function_single(remsd
->cpu
,
3354 static int process_backlog(struct napi_struct
*napi
, int quota
)
3357 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3360 /* Check if we have pending ipi, its better to send them now,
3361 * not waiting net_rx_action() end.
3363 if (sd
->rps_ipi_list
) {
3364 local_irq_disable();
3365 net_rps_action_and_irq_enable(sd
);
3368 napi
->weight
= weight_p
;
3369 local_irq_disable();
3370 while (work
< quota
) {
3371 struct sk_buff
*skb
;
3374 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3376 __netif_receive_skb(skb
);
3377 local_irq_disable();
3378 input_queue_head_incr(sd
);
3379 if (++work
>= quota
) {
3386 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3388 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3389 &sd
->process_queue
);
3391 if (qlen
< quota
- work
) {
3393 * Inline a custom version of __napi_complete().
3394 * only current cpu owns and manipulates this napi,
3395 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3396 * we can use a plain write instead of clear_bit(),
3397 * and we dont need an smp_mb() memory barrier.
3399 list_del(&napi
->poll_list
);
3402 quota
= work
+ qlen
;
3412 * __napi_schedule - schedule for receive
3413 * @n: entry to schedule
3415 * The entry's receive function will be scheduled to run
3417 void __napi_schedule(struct napi_struct
*n
)
3419 unsigned long flags
;
3421 local_irq_save(flags
);
3422 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3423 local_irq_restore(flags
);
3425 EXPORT_SYMBOL(__napi_schedule
);
3427 void __napi_complete(struct napi_struct
*n
)
3429 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3430 BUG_ON(n
->gro_list
);
3432 list_del(&n
->poll_list
);
3433 smp_mb__before_clear_bit();
3434 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3436 EXPORT_SYMBOL(__napi_complete
);
3438 void napi_complete(struct napi_struct
*n
)
3440 unsigned long flags
;
3443 * don't let napi dequeue from the cpu poll list
3444 * just in case its running on a different cpu
3446 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3450 local_irq_save(flags
);
3452 local_irq_restore(flags
);
3454 EXPORT_SYMBOL(napi_complete
);
3456 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3457 int (*poll
)(struct napi_struct
*, int), int weight
)
3459 INIT_LIST_HEAD(&napi
->poll_list
);
3460 napi
->gro_count
= 0;
3461 napi
->gro_list
= NULL
;
3464 napi
->weight
= weight
;
3465 list_add(&napi
->dev_list
, &dev
->napi_list
);
3467 #ifdef CONFIG_NETPOLL
3468 spin_lock_init(&napi
->poll_lock
);
3469 napi
->poll_owner
= -1;
3471 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3473 EXPORT_SYMBOL(netif_napi_add
);
3475 void netif_napi_del(struct napi_struct
*napi
)
3477 struct sk_buff
*skb
, *next
;
3479 list_del_init(&napi
->dev_list
);
3480 napi_free_frags(napi
);
3482 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3488 napi
->gro_list
= NULL
;
3489 napi
->gro_count
= 0;
3491 EXPORT_SYMBOL(netif_napi_del
);
3493 static void net_rx_action(struct softirq_action
*h
)
3495 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3496 unsigned long time_limit
= jiffies
+ 2;
3497 int budget
= netdev_budget
;
3500 local_irq_disable();
3502 while (!list_empty(&sd
->poll_list
)) {
3503 struct napi_struct
*n
;
3506 /* If softirq window is exhuasted then punt.
3507 * Allow this to run for 2 jiffies since which will allow
3508 * an average latency of 1.5/HZ.
3510 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3515 /* Even though interrupts have been re-enabled, this
3516 * access is safe because interrupts can only add new
3517 * entries to the tail of this list, and only ->poll()
3518 * calls can remove this head entry from the list.
3520 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3522 have
= netpoll_poll_lock(n
);
3526 /* This NAPI_STATE_SCHED test is for avoiding a race
3527 * with netpoll's poll_napi(). Only the entity which
3528 * obtains the lock and sees NAPI_STATE_SCHED set will
3529 * actually make the ->poll() call. Therefore we avoid
3530 * accidently calling ->poll() when NAPI is not scheduled.
3533 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3534 work
= n
->poll(n
, weight
);
3538 WARN_ON_ONCE(work
> weight
);
3542 local_irq_disable();
3544 /* Drivers must not modify the NAPI state if they
3545 * consume the entire weight. In such cases this code
3546 * still "owns" the NAPI instance and therefore can
3547 * move the instance around on the list at-will.
3549 if (unlikely(work
== weight
)) {
3550 if (unlikely(napi_disable_pending(n
))) {
3553 local_irq_disable();
3555 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3558 netpoll_poll_unlock(have
);
3561 net_rps_action_and_irq_enable(sd
);
3563 #ifdef CONFIG_NET_DMA
3565 * There may not be any more sk_buffs coming right now, so push
3566 * any pending DMA copies to hardware
3568 dma_issue_pending_all();
3575 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3579 static gifconf_func_t
*gifconf_list
[NPROTO
];
3582 * register_gifconf - register a SIOCGIF handler
3583 * @family: Address family
3584 * @gifconf: Function handler
3586 * Register protocol dependent address dumping routines. The handler
3587 * that is passed must not be freed or reused until it has been replaced
3588 * by another handler.
3590 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3592 if (family
>= NPROTO
)
3594 gifconf_list
[family
] = gifconf
;
3597 EXPORT_SYMBOL(register_gifconf
);
3601 * Map an interface index to its name (SIOCGIFNAME)
3605 * We need this ioctl for efficient implementation of the
3606 * if_indextoname() function required by the IPv6 API. Without
3607 * it, we would have to search all the interfaces to find a
3611 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3613 struct net_device
*dev
;
3617 * Fetch the caller's info block.
3620 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3624 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3630 strcpy(ifr
.ifr_name
, dev
->name
);
3633 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3639 * Perform a SIOCGIFCONF call. This structure will change
3640 * size eventually, and there is nothing I can do about it.
3641 * Thus we will need a 'compatibility mode'.
3644 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3647 struct net_device
*dev
;
3654 * Fetch the caller's info block.
3657 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3664 * Loop over the interfaces, and write an info block for each.
3668 for_each_netdev(net
, dev
) {
3669 for (i
= 0; i
< NPROTO
; i
++) {
3670 if (gifconf_list
[i
]) {
3673 done
= gifconf_list
[i
](dev
, NULL
, 0);
3675 done
= gifconf_list
[i
](dev
, pos
+ total
,
3685 * All done. Write the updated control block back to the caller.
3687 ifc
.ifc_len
= total
;
3690 * Both BSD and Solaris return 0 here, so we do too.
3692 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3695 #ifdef CONFIG_PROC_FS
3697 * This is invoked by the /proc filesystem handler to display a device
3700 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3703 struct net
*net
= seq_file_net(seq
);
3705 struct net_device
*dev
;
3709 return SEQ_START_TOKEN
;
3712 for_each_netdev_rcu(net
, dev
)
3719 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3721 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3722 first_net_device(seq_file_net(seq
)) :
3723 next_net_device((struct net_device
*)v
);
3726 return rcu_dereference(dev
);
3729 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3735 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3737 struct rtnl_link_stats64 temp
;
3738 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
3740 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3741 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3742 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3744 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3745 stats
->rx_fifo_errors
,
3746 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3747 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3748 stats
->rx_compressed
, stats
->multicast
,
3749 stats
->tx_bytes
, stats
->tx_packets
,
3750 stats
->tx_errors
, stats
->tx_dropped
,
3751 stats
->tx_fifo_errors
, stats
->collisions
,
3752 stats
->tx_carrier_errors
+
3753 stats
->tx_aborted_errors
+
3754 stats
->tx_window_errors
+
3755 stats
->tx_heartbeat_errors
,
3756 stats
->tx_compressed
);
3760 * Called from the PROCfs module. This now uses the new arbitrary sized
3761 * /proc/net interface to create /proc/net/dev
3763 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3765 if (v
== SEQ_START_TOKEN
)
3766 seq_puts(seq
, "Inter-| Receive "
3768 " face |bytes packets errs drop fifo frame "
3769 "compressed multicast|bytes packets errs "
3770 "drop fifo colls carrier compressed\n");
3772 dev_seq_printf_stats(seq
, v
);
3776 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
3778 struct softnet_data
*sd
= NULL
;
3780 while (*pos
< nr_cpu_ids
)
3781 if (cpu_online(*pos
)) {
3782 sd
= &per_cpu(softnet_data
, *pos
);
3789 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3791 return softnet_get_online(pos
);
3794 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3797 return softnet_get_online(pos
);
3800 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3804 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3806 struct softnet_data
*sd
= v
;
3808 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3809 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
3810 0, 0, 0, 0, /* was fastroute */
3811 sd
->cpu_collision
, sd
->received_rps
);
3815 static const struct seq_operations dev_seq_ops
= {
3816 .start
= dev_seq_start
,
3817 .next
= dev_seq_next
,
3818 .stop
= dev_seq_stop
,
3819 .show
= dev_seq_show
,
3822 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3824 return seq_open_net(inode
, file
, &dev_seq_ops
,
3825 sizeof(struct seq_net_private
));
3828 static const struct file_operations dev_seq_fops
= {
3829 .owner
= THIS_MODULE
,
3830 .open
= dev_seq_open
,
3832 .llseek
= seq_lseek
,
3833 .release
= seq_release_net
,
3836 static const struct seq_operations softnet_seq_ops
= {
3837 .start
= softnet_seq_start
,
3838 .next
= softnet_seq_next
,
3839 .stop
= softnet_seq_stop
,
3840 .show
= softnet_seq_show
,
3843 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3845 return seq_open(file
, &softnet_seq_ops
);
3848 static const struct file_operations softnet_seq_fops
= {
3849 .owner
= THIS_MODULE
,
3850 .open
= softnet_seq_open
,
3852 .llseek
= seq_lseek
,
3853 .release
= seq_release
,
3856 static void *ptype_get_idx(loff_t pos
)
3858 struct packet_type
*pt
= NULL
;
3862 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3868 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3869 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3878 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3882 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3885 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3887 struct packet_type
*pt
;
3888 struct list_head
*nxt
;
3892 if (v
== SEQ_START_TOKEN
)
3893 return ptype_get_idx(0);
3896 nxt
= pt
->list
.next
;
3897 if (pt
->type
== htons(ETH_P_ALL
)) {
3898 if (nxt
!= &ptype_all
)
3901 nxt
= ptype_base
[0].next
;
3903 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3905 while (nxt
== &ptype_base
[hash
]) {
3906 if (++hash
>= PTYPE_HASH_SIZE
)
3908 nxt
= ptype_base
[hash
].next
;
3911 return list_entry(nxt
, struct packet_type
, list
);
3914 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3920 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3922 struct packet_type
*pt
= v
;
3924 if (v
== SEQ_START_TOKEN
)
3925 seq_puts(seq
, "Type Device Function\n");
3926 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3927 if (pt
->type
== htons(ETH_P_ALL
))
3928 seq_puts(seq
, "ALL ");
3930 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3932 seq_printf(seq
, " %-8s %pF\n",
3933 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
3939 static const struct seq_operations ptype_seq_ops
= {
3940 .start
= ptype_seq_start
,
3941 .next
= ptype_seq_next
,
3942 .stop
= ptype_seq_stop
,
3943 .show
= ptype_seq_show
,
3946 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
3948 return seq_open_net(inode
, file
, &ptype_seq_ops
,
3949 sizeof(struct seq_net_private
));
3952 static const struct file_operations ptype_seq_fops
= {
3953 .owner
= THIS_MODULE
,
3954 .open
= ptype_seq_open
,
3956 .llseek
= seq_lseek
,
3957 .release
= seq_release_net
,
3961 static int __net_init
dev_proc_net_init(struct net
*net
)
3965 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
3967 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
3969 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
3972 if (wext_proc_init(net
))
3978 proc_net_remove(net
, "ptype");
3980 proc_net_remove(net
, "softnet_stat");
3982 proc_net_remove(net
, "dev");
3986 static void __net_exit
dev_proc_net_exit(struct net
*net
)
3988 wext_proc_exit(net
);
3990 proc_net_remove(net
, "ptype");
3991 proc_net_remove(net
, "softnet_stat");
3992 proc_net_remove(net
, "dev");
3995 static struct pernet_operations __net_initdata dev_proc_ops
= {
3996 .init
= dev_proc_net_init
,
3997 .exit
= dev_proc_net_exit
,
4000 static int __init
dev_proc_init(void)
4002 return register_pernet_subsys(&dev_proc_ops
);
4005 #define dev_proc_init() 0
4006 #endif /* CONFIG_PROC_FS */
4010 * netdev_set_master - set up master/slave pair
4011 * @slave: slave device
4012 * @master: new master device
4014 * Changes the master device of the slave. Pass %NULL to break the
4015 * bonding. The caller must hold the RTNL semaphore. On a failure
4016 * a negative errno code is returned. On success the reference counts
4017 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4018 * function returns zero.
4020 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4022 struct net_device
*old
= slave
->master
;
4032 slave
->master
= master
;
4039 slave
->flags
|= IFF_SLAVE
;
4041 slave
->flags
&= ~IFF_SLAVE
;
4043 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4046 EXPORT_SYMBOL(netdev_set_master
);
4048 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4050 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4052 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4053 ops
->ndo_change_rx_flags(dev
, flags
);
4056 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4058 unsigned short old_flags
= dev
->flags
;
4064 dev
->flags
|= IFF_PROMISC
;
4065 dev
->promiscuity
+= inc
;
4066 if (dev
->promiscuity
== 0) {
4069 * If inc causes overflow, untouch promisc and return error.
4072 dev
->flags
&= ~IFF_PROMISC
;
4074 dev
->promiscuity
-= inc
;
4075 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4076 "set promiscuity failed, promiscuity feature "
4077 "of device might be broken.\n", dev
->name
);
4081 if (dev
->flags
!= old_flags
) {
4082 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4083 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4085 if (audit_enabled
) {
4086 current_uid_gid(&uid
, &gid
);
4087 audit_log(current
->audit_context
, GFP_ATOMIC
,
4088 AUDIT_ANOM_PROMISCUOUS
,
4089 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4090 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4091 (old_flags
& IFF_PROMISC
),
4092 audit_get_loginuid(current
),
4094 audit_get_sessionid(current
));
4097 dev_change_rx_flags(dev
, IFF_PROMISC
);
4103 * dev_set_promiscuity - update promiscuity count on a device
4107 * Add or remove promiscuity from a device. While the count in the device
4108 * remains above zero the interface remains promiscuous. Once it hits zero
4109 * the device reverts back to normal filtering operation. A negative inc
4110 * value is used to drop promiscuity on the device.
4111 * Return 0 if successful or a negative errno code on error.
4113 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4115 unsigned short old_flags
= dev
->flags
;
4118 err
= __dev_set_promiscuity(dev
, inc
);
4121 if (dev
->flags
!= old_flags
)
4122 dev_set_rx_mode(dev
);
4125 EXPORT_SYMBOL(dev_set_promiscuity
);
4128 * dev_set_allmulti - update allmulti count on a device
4132 * Add or remove reception of all multicast frames to a device. While the
4133 * count in the device remains above zero the interface remains listening
4134 * to all interfaces. Once it hits zero the device reverts back to normal
4135 * filtering operation. A negative @inc value is used to drop the counter
4136 * when releasing a resource needing all multicasts.
4137 * Return 0 if successful or a negative errno code on error.
4140 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4142 unsigned short old_flags
= dev
->flags
;
4146 dev
->flags
|= IFF_ALLMULTI
;
4147 dev
->allmulti
+= inc
;
4148 if (dev
->allmulti
== 0) {
4151 * If inc causes overflow, untouch allmulti and return error.
4154 dev
->flags
&= ~IFF_ALLMULTI
;
4156 dev
->allmulti
-= inc
;
4157 printk(KERN_WARNING
"%s: allmulti touches roof, "
4158 "set allmulti failed, allmulti feature of "
4159 "device might be broken.\n", dev
->name
);
4163 if (dev
->flags
^ old_flags
) {
4164 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4165 dev_set_rx_mode(dev
);
4169 EXPORT_SYMBOL(dev_set_allmulti
);
4172 * Upload unicast and multicast address lists to device and
4173 * configure RX filtering. When the device doesn't support unicast
4174 * filtering it is put in promiscuous mode while unicast addresses
4177 void __dev_set_rx_mode(struct net_device
*dev
)
4179 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4181 /* dev_open will call this function so the list will stay sane. */
4182 if (!(dev
->flags
&IFF_UP
))
4185 if (!netif_device_present(dev
))
4188 if (ops
->ndo_set_rx_mode
)
4189 ops
->ndo_set_rx_mode(dev
);
4191 /* Unicast addresses changes may only happen under the rtnl,
4192 * therefore calling __dev_set_promiscuity here is safe.
4194 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4195 __dev_set_promiscuity(dev
, 1);
4196 dev
->uc_promisc
= 1;
4197 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4198 __dev_set_promiscuity(dev
, -1);
4199 dev
->uc_promisc
= 0;
4202 if (ops
->ndo_set_multicast_list
)
4203 ops
->ndo_set_multicast_list(dev
);
4207 void dev_set_rx_mode(struct net_device
*dev
)
4209 netif_addr_lock_bh(dev
);
4210 __dev_set_rx_mode(dev
);
4211 netif_addr_unlock_bh(dev
);
4215 * dev_get_flags - get flags reported to userspace
4218 * Get the combination of flag bits exported through APIs to userspace.
4220 unsigned dev_get_flags(const struct net_device
*dev
)
4224 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4229 (dev
->gflags
& (IFF_PROMISC
|
4232 if (netif_running(dev
)) {
4233 if (netif_oper_up(dev
))
4234 flags
|= IFF_RUNNING
;
4235 if (netif_carrier_ok(dev
))
4236 flags
|= IFF_LOWER_UP
;
4237 if (netif_dormant(dev
))
4238 flags
|= IFF_DORMANT
;
4243 EXPORT_SYMBOL(dev_get_flags
);
4245 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4247 int old_flags
= dev
->flags
;
4253 * Set the flags on our device.
4256 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4257 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4259 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4263 * Load in the correct multicast list now the flags have changed.
4266 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4267 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4269 dev_set_rx_mode(dev
);
4272 * Have we downed the interface. We handle IFF_UP ourselves
4273 * according to user attempts to set it, rather than blindly
4278 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4279 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4282 dev_set_rx_mode(dev
);
4285 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4286 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4288 dev
->gflags
^= IFF_PROMISC
;
4289 dev_set_promiscuity(dev
, inc
);
4292 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4293 is important. Some (broken) drivers set IFF_PROMISC, when
4294 IFF_ALLMULTI is requested not asking us and not reporting.
4296 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4297 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4299 dev
->gflags
^= IFF_ALLMULTI
;
4300 dev_set_allmulti(dev
, inc
);
4306 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4308 unsigned int changes
= dev
->flags
^ old_flags
;
4310 if (changes
& IFF_UP
) {
4311 if (dev
->flags
& IFF_UP
)
4312 call_netdevice_notifiers(NETDEV_UP
, dev
);
4314 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4317 if (dev
->flags
& IFF_UP
&&
4318 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4319 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4323 * dev_change_flags - change device settings
4325 * @flags: device state flags
4327 * Change settings on device based state flags. The flags are
4328 * in the userspace exported format.
4330 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4333 int old_flags
= dev
->flags
;
4335 ret
= __dev_change_flags(dev
, flags
);
4339 changes
= old_flags
^ dev
->flags
;
4341 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4343 __dev_notify_flags(dev
, old_flags
);
4346 EXPORT_SYMBOL(dev_change_flags
);
4349 * dev_set_mtu - Change maximum transfer unit
4351 * @new_mtu: new transfer unit
4353 * Change the maximum transfer size of the network device.
4355 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4357 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4360 if (new_mtu
== dev
->mtu
)
4363 /* MTU must be positive. */
4367 if (!netif_device_present(dev
))
4371 if (ops
->ndo_change_mtu
)
4372 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4376 if (!err
&& dev
->flags
& IFF_UP
)
4377 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4380 EXPORT_SYMBOL(dev_set_mtu
);
4383 * dev_set_mac_address - Change Media Access Control Address
4387 * Change the hardware (MAC) address of the device
4389 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4391 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4394 if (!ops
->ndo_set_mac_address
)
4396 if (sa
->sa_family
!= dev
->type
)
4398 if (!netif_device_present(dev
))
4400 err
= ops
->ndo_set_mac_address(dev
, sa
);
4402 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4405 EXPORT_SYMBOL(dev_set_mac_address
);
4408 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4410 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4413 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4419 case SIOCGIFFLAGS
: /* Get interface flags */
4420 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4423 case SIOCGIFMETRIC
: /* Get the metric on the interface
4424 (currently unused) */
4425 ifr
->ifr_metric
= 0;
4428 case SIOCGIFMTU
: /* Get the MTU of a device */
4429 ifr
->ifr_mtu
= dev
->mtu
;
4434 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4436 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4437 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4438 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4446 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4447 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4448 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4449 ifr
->ifr_map
.irq
= dev
->irq
;
4450 ifr
->ifr_map
.dma
= dev
->dma
;
4451 ifr
->ifr_map
.port
= dev
->if_port
;
4455 ifr
->ifr_ifindex
= dev
->ifindex
;
4459 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4463 /* dev_ioctl() should ensure this case
4475 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4477 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4480 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4481 const struct net_device_ops
*ops
;
4486 ops
= dev
->netdev_ops
;
4489 case SIOCSIFFLAGS
: /* Set interface flags */
4490 return dev_change_flags(dev
, ifr
->ifr_flags
);
4492 case SIOCSIFMETRIC
: /* Set the metric on the interface
4493 (currently unused) */
4496 case SIOCSIFMTU
: /* Set the MTU of a device */
4497 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4500 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4502 case SIOCSIFHWBROADCAST
:
4503 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4505 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4506 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4507 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4511 if (ops
->ndo_set_config
) {
4512 if (!netif_device_present(dev
))
4514 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4519 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4520 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4522 if (!netif_device_present(dev
))
4524 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4527 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4528 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4530 if (!netif_device_present(dev
))
4532 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4535 if (ifr
->ifr_qlen
< 0)
4537 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4541 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4542 return dev_change_name(dev
, ifr
->ifr_newname
);
4545 * Unknown or private ioctl
4548 if ((cmd
>= SIOCDEVPRIVATE
&&
4549 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4550 cmd
== SIOCBONDENSLAVE
||
4551 cmd
== SIOCBONDRELEASE
||
4552 cmd
== SIOCBONDSETHWADDR
||
4553 cmd
== SIOCBONDSLAVEINFOQUERY
||
4554 cmd
== SIOCBONDINFOQUERY
||
4555 cmd
== SIOCBONDCHANGEACTIVE
||
4556 cmd
== SIOCGMIIPHY
||
4557 cmd
== SIOCGMIIREG
||
4558 cmd
== SIOCSMIIREG
||
4559 cmd
== SIOCBRADDIF
||
4560 cmd
== SIOCBRDELIF
||
4561 cmd
== SIOCSHWTSTAMP
||
4562 cmd
== SIOCWANDEV
) {
4564 if (ops
->ndo_do_ioctl
) {
4565 if (netif_device_present(dev
))
4566 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4578 * This function handles all "interface"-type I/O control requests. The actual
4579 * 'doing' part of this is dev_ifsioc above.
4583 * dev_ioctl - network device ioctl
4584 * @net: the applicable net namespace
4585 * @cmd: command to issue
4586 * @arg: pointer to a struct ifreq in user space
4588 * Issue ioctl functions to devices. This is normally called by the
4589 * user space syscall interfaces but can sometimes be useful for
4590 * other purposes. The return value is the return from the syscall if
4591 * positive or a negative errno code on error.
4594 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4600 /* One special case: SIOCGIFCONF takes ifconf argument
4601 and requires shared lock, because it sleeps writing
4605 if (cmd
== SIOCGIFCONF
) {
4607 ret
= dev_ifconf(net
, (char __user
*) arg
);
4611 if (cmd
== SIOCGIFNAME
)
4612 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4614 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4617 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4619 colon
= strchr(ifr
.ifr_name
, ':');
4624 * See which interface the caller is talking about.
4629 * These ioctl calls:
4630 * - can be done by all.
4631 * - atomic and do not require locking.
4642 dev_load(net
, ifr
.ifr_name
);
4644 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4649 if (copy_to_user(arg
, &ifr
,
4650 sizeof(struct ifreq
)))
4656 dev_load(net
, ifr
.ifr_name
);
4658 ret
= dev_ethtool(net
, &ifr
);
4663 if (copy_to_user(arg
, &ifr
,
4664 sizeof(struct ifreq
)))
4670 * These ioctl calls:
4671 * - require superuser power.
4672 * - require strict serialization.
4678 if (!capable(CAP_NET_ADMIN
))
4680 dev_load(net
, ifr
.ifr_name
);
4682 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4687 if (copy_to_user(arg
, &ifr
,
4688 sizeof(struct ifreq
)))
4694 * These ioctl calls:
4695 * - require superuser power.
4696 * - require strict serialization.
4697 * - do not return a value
4707 case SIOCSIFHWBROADCAST
:
4710 case SIOCBONDENSLAVE
:
4711 case SIOCBONDRELEASE
:
4712 case SIOCBONDSETHWADDR
:
4713 case SIOCBONDCHANGEACTIVE
:
4717 if (!capable(CAP_NET_ADMIN
))
4720 case SIOCBONDSLAVEINFOQUERY
:
4721 case SIOCBONDINFOQUERY
:
4722 dev_load(net
, ifr
.ifr_name
);
4724 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4729 /* Get the per device memory space. We can add this but
4730 * currently do not support it */
4732 /* Set the per device memory buffer space.
4733 * Not applicable in our case */
4738 * Unknown or private ioctl.
4741 if (cmd
== SIOCWANDEV
||
4742 (cmd
>= SIOCDEVPRIVATE
&&
4743 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4744 dev_load(net
, ifr
.ifr_name
);
4746 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4748 if (!ret
&& copy_to_user(arg
, &ifr
,
4749 sizeof(struct ifreq
)))
4753 /* Take care of Wireless Extensions */
4754 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4755 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4762 * dev_new_index - allocate an ifindex
4763 * @net: the applicable net namespace
4765 * Returns a suitable unique value for a new device interface
4766 * number. The caller must hold the rtnl semaphore or the
4767 * dev_base_lock to be sure it remains unique.
4769 static int dev_new_index(struct net
*net
)
4775 if (!__dev_get_by_index(net
, ifindex
))
4780 /* Delayed registration/unregisteration */
4781 static LIST_HEAD(net_todo_list
);
4783 static void net_set_todo(struct net_device
*dev
)
4785 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4788 static void rollback_registered_many(struct list_head
*head
)
4790 struct net_device
*dev
, *tmp
;
4792 BUG_ON(dev_boot_phase
);
4795 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4796 /* Some devices call without registering
4797 * for initialization unwind. Remove those
4798 * devices and proceed with the remaining.
4800 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4801 pr_debug("unregister_netdevice: device %s/%p never "
4802 "was registered\n", dev
->name
, dev
);
4805 list_del(&dev
->unreg_list
);
4809 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4811 /* If device is running, close it first. */
4814 /* And unlink it from device chain. */
4815 unlist_netdevice(dev
);
4817 dev
->reg_state
= NETREG_UNREGISTERING
;
4822 list_for_each_entry(dev
, head
, unreg_list
) {
4823 /* Shutdown queueing discipline. */
4827 /* Notify protocols, that we are about to destroy
4828 this device. They should clean all the things.
4830 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4832 if (!dev
->rtnl_link_ops
||
4833 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
4834 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
4837 * Flush the unicast and multicast chains
4842 if (dev
->netdev_ops
->ndo_uninit
)
4843 dev
->netdev_ops
->ndo_uninit(dev
);
4845 /* Notifier chain MUST detach us from master device. */
4846 WARN_ON(dev
->master
);
4848 /* Remove entries from kobject tree */
4849 netdev_unregister_kobject(dev
);
4852 /* Process any work delayed until the end of the batch */
4853 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
4854 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
4858 list_for_each_entry(dev
, head
, unreg_list
)
4862 static void rollback_registered(struct net_device
*dev
)
4866 list_add(&dev
->unreg_list
, &single
);
4867 rollback_registered_many(&single
);
4870 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
4871 struct netdev_queue
*dev_queue
,
4874 spin_lock_init(&dev_queue
->_xmit_lock
);
4875 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
4876 dev_queue
->xmit_lock_owner
= -1;
4879 static void netdev_init_queue_locks(struct net_device
*dev
)
4881 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
4882 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
4885 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
4887 /* Fix illegal SG+CSUM combinations. */
4888 if ((features
& NETIF_F_SG
) &&
4889 !(features
& NETIF_F_ALL_CSUM
)) {
4891 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
4892 "checksum feature.\n", name
);
4893 features
&= ~NETIF_F_SG
;
4896 /* TSO requires that SG is present as well. */
4897 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
4899 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
4900 "SG feature.\n", name
);
4901 features
&= ~NETIF_F_TSO
;
4904 if (features
& NETIF_F_UFO
) {
4905 if (!(features
& NETIF_F_GEN_CSUM
)) {
4907 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4908 "since no NETIF_F_HW_CSUM feature.\n",
4910 features
&= ~NETIF_F_UFO
;
4913 if (!(features
& NETIF_F_SG
)) {
4915 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4916 "since no NETIF_F_SG feature.\n", name
);
4917 features
&= ~NETIF_F_UFO
;
4923 EXPORT_SYMBOL(netdev_fix_features
);
4926 * netif_stacked_transfer_operstate - transfer operstate
4927 * @rootdev: the root or lower level device to transfer state from
4928 * @dev: the device to transfer operstate to
4930 * Transfer operational state from root to device. This is normally
4931 * called when a stacking relationship exists between the root
4932 * device and the device(a leaf device).
4934 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
4935 struct net_device
*dev
)
4937 if (rootdev
->operstate
== IF_OPER_DORMANT
)
4938 netif_dormant_on(dev
);
4940 netif_dormant_off(dev
);
4942 if (netif_carrier_ok(rootdev
)) {
4943 if (!netif_carrier_ok(dev
))
4944 netif_carrier_on(dev
);
4946 if (netif_carrier_ok(dev
))
4947 netif_carrier_off(dev
);
4950 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
4953 * register_netdevice - register a network device
4954 * @dev: device to register
4956 * Take a completed network device structure and add it to the kernel
4957 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4958 * chain. 0 is returned on success. A negative errno code is returned
4959 * on a failure to set up the device, or if the name is a duplicate.
4961 * Callers must hold the rtnl semaphore. You may want
4962 * register_netdev() instead of this.
4965 * The locking appears insufficient to guarantee two parallel registers
4966 * will not get the same name.
4969 int register_netdevice(struct net_device
*dev
)
4972 struct net
*net
= dev_net(dev
);
4974 BUG_ON(dev_boot_phase
);
4979 /* When net_device's are persistent, this will be fatal. */
4980 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
4983 spin_lock_init(&dev
->addr_list_lock
);
4984 netdev_set_addr_lockdep_class(dev
);
4985 netdev_init_queue_locks(dev
);
4990 if (!dev
->num_rx_queues
) {
4992 * Allocate a single RX queue if driver never called
4996 dev
->_rx
= kzalloc(sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5002 dev
->_rx
->first
= dev
->_rx
;
5003 atomic_set(&dev
->_rx
->count
, 1);
5004 dev
->num_rx_queues
= 1;
5007 /* Init, if this function is available */
5008 if (dev
->netdev_ops
->ndo_init
) {
5009 ret
= dev
->netdev_ops
->ndo_init(dev
);
5017 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
5021 dev
->ifindex
= dev_new_index(net
);
5022 if (dev
->iflink
== -1)
5023 dev
->iflink
= dev
->ifindex
;
5025 /* Fix illegal checksum combinations */
5026 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
5027 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5028 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5030 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5033 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5034 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5035 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5037 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5040 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5042 /* Enable software GSO if SG is supported. */
5043 if (dev
->features
& NETIF_F_SG
)
5044 dev
->features
|= NETIF_F_GSO
;
5046 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5047 ret
= notifier_to_errno(ret
);
5051 ret
= netdev_register_kobject(dev
);
5054 dev
->reg_state
= NETREG_REGISTERED
;
5057 * Default initial state at registry is that the
5058 * device is present.
5061 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5063 dev_init_scheduler(dev
);
5065 list_netdevice(dev
);
5067 /* Notify protocols, that a new device appeared. */
5068 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5069 ret
= notifier_to_errno(ret
);
5071 rollback_registered(dev
);
5072 dev
->reg_state
= NETREG_UNREGISTERED
;
5075 * Prevent userspace races by waiting until the network
5076 * device is fully setup before sending notifications.
5078 if (!dev
->rtnl_link_ops
||
5079 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5080 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5086 if (dev
->netdev_ops
->ndo_uninit
)
5087 dev
->netdev_ops
->ndo_uninit(dev
);
5090 EXPORT_SYMBOL(register_netdevice
);
5093 * init_dummy_netdev - init a dummy network device for NAPI
5094 * @dev: device to init
5096 * This takes a network device structure and initialize the minimum
5097 * amount of fields so it can be used to schedule NAPI polls without
5098 * registering a full blown interface. This is to be used by drivers
5099 * that need to tie several hardware interfaces to a single NAPI
5100 * poll scheduler due to HW limitations.
5102 int init_dummy_netdev(struct net_device
*dev
)
5104 /* Clear everything. Note we don't initialize spinlocks
5105 * are they aren't supposed to be taken by any of the
5106 * NAPI code and this dummy netdev is supposed to be
5107 * only ever used for NAPI polls
5109 memset(dev
, 0, sizeof(struct net_device
));
5111 /* make sure we BUG if trying to hit standard
5112 * register/unregister code path
5114 dev
->reg_state
= NETREG_DUMMY
;
5116 /* initialize the ref count */
5117 atomic_set(&dev
->refcnt
, 1);
5119 /* NAPI wants this */
5120 INIT_LIST_HEAD(&dev
->napi_list
);
5122 /* a dummy interface is started by default */
5123 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5124 set_bit(__LINK_STATE_START
, &dev
->state
);
5128 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5132 * register_netdev - register a network device
5133 * @dev: device to register
5135 * Take a completed network device structure and add it to the kernel
5136 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5137 * chain. 0 is returned on success. A negative errno code is returned
5138 * on a failure to set up the device, or if the name is a duplicate.
5140 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5141 * and expands the device name if you passed a format string to
5144 int register_netdev(struct net_device
*dev
)
5151 * If the name is a format string the caller wants us to do a
5154 if (strchr(dev
->name
, '%')) {
5155 err
= dev_alloc_name(dev
, dev
->name
);
5160 err
= register_netdevice(dev
);
5165 EXPORT_SYMBOL(register_netdev
);
5168 * netdev_wait_allrefs - wait until all references are gone.
5170 * This is called when unregistering network devices.
5172 * Any protocol or device that holds a reference should register
5173 * for netdevice notification, and cleanup and put back the
5174 * reference if they receive an UNREGISTER event.
5175 * We can get stuck here if buggy protocols don't correctly
5178 static void netdev_wait_allrefs(struct net_device
*dev
)
5180 unsigned long rebroadcast_time
, warning_time
;
5182 linkwatch_forget_dev(dev
);
5184 rebroadcast_time
= warning_time
= jiffies
;
5185 while (atomic_read(&dev
->refcnt
) != 0) {
5186 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5189 /* Rebroadcast unregister notification */
5190 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5191 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5192 * should have already handle it the first time */
5194 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5196 /* We must not have linkwatch events
5197 * pending on unregister. If this
5198 * happens, we simply run the queue
5199 * unscheduled, resulting in a noop
5202 linkwatch_run_queue();
5207 rebroadcast_time
= jiffies
;
5212 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5213 printk(KERN_EMERG
"unregister_netdevice: "
5214 "waiting for %s to become free. Usage "
5216 dev
->name
, atomic_read(&dev
->refcnt
));
5217 warning_time
= jiffies
;
5226 * register_netdevice(x1);
5227 * register_netdevice(x2);
5229 * unregister_netdevice(y1);
5230 * unregister_netdevice(y2);
5236 * We are invoked by rtnl_unlock().
5237 * This allows us to deal with problems:
5238 * 1) We can delete sysfs objects which invoke hotplug
5239 * without deadlocking with linkwatch via keventd.
5240 * 2) Since we run with the RTNL semaphore not held, we can sleep
5241 * safely in order to wait for the netdev refcnt to drop to zero.
5243 * We must not return until all unregister events added during
5244 * the interval the lock was held have been completed.
5246 void netdev_run_todo(void)
5248 struct list_head list
;
5250 /* Snapshot list, allow later requests */
5251 list_replace_init(&net_todo_list
, &list
);
5255 while (!list_empty(&list
)) {
5256 struct net_device
*dev
5257 = list_first_entry(&list
, struct net_device
, todo_list
);
5258 list_del(&dev
->todo_list
);
5260 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5261 printk(KERN_ERR
"network todo '%s' but state %d\n",
5262 dev
->name
, dev
->reg_state
);
5267 dev
->reg_state
= NETREG_UNREGISTERED
;
5269 on_each_cpu(flush_backlog
, dev
, 1);
5271 netdev_wait_allrefs(dev
);
5274 BUG_ON(atomic_read(&dev
->refcnt
));
5275 WARN_ON(dev
->ip_ptr
);
5276 WARN_ON(dev
->ip6_ptr
);
5277 WARN_ON(dev
->dn_ptr
);
5279 if (dev
->destructor
)
5280 dev
->destructor(dev
);
5282 /* Free network device */
5283 kobject_put(&dev
->dev
.kobj
);
5288 * dev_txq_stats_fold - fold tx_queues stats
5289 * @dev: device to get statistics from
5290 * @stats: struct rtnl_link_stats64 to hold results
5292 void dev_txq_stats_fold(const struct net_device
*dev
,
5293 struct rtnl_link_stats64
*stats
)
5295 u64 tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5297 struct netdev_queue
*txq
;
5299 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5300 txq
= netdev_get_tx_queue(dev
, i
);
5301 spin_lock_bh(&txq
->_xmit_lock
);
5302 tx_bytes
+= txq
->tx_bytes
;
5303 tx_packets
+= txq
->tx_packets
;
5304 tx_dropped
+= txq
->tx_dropped
;
5305 spin_unlock_bh(&txq
->_xmit_lock
);
5307 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5308 stats
->tx_bytes
= tx_bytes
;
5309 stats
->tx_packets
= tx_packets
;
5310 stats
->tx_dropped
= tx_dropped
;
5313 EXPORT_SYMBOL(dev_txq_stats_fold
);
5315 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5316 * fields in the same order, with only the type differing.
5318 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5319 const struct net_device_stats
*netdev_stats
)
5321 #if BITS_PER_LONG == 64
5322 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5323 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5325 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5326 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5327 u64
*dst
= (u64
*)stats64
;
5329 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5330 sizeof(*stats64
) / sizeof(u64
));
5331 for (i
= 0; i
< n
; i
++)
5337 * dev_get_stats - get network device statistics
5338 * @dev: device to get statistics from
5339 * @storage: place to store stats
5341 * Get network statistics from device. Return @storage.
5342 * The device driver may provide its own method by setting
5343 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5344 * otherwise the internal statistics structure is used.
5346 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5347 struct rtnl_link_stats64
*storage
)
5349 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5351 if (ops
->ndo_get_stats64
) {
5352 memset(storage
, 0, sizeof(*storage
));
5353 return ops
->ndo_get_stats64(dev
, storage
);
5355 if (ops
->ndo_get_stats
) {
5356 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5359 netdev_stats_to_stats64(storage
, &dev
->stats
);
5360 dev_txq_stats_fold(dev
, storage
);
5363 EXPORT_SYMBOL(dev_get_stats
);
5365 static void netdev_init_one_queue(struct net_device
*dev
,
5366 struct netdev_queue
*queue
,
5372 static void netdev_init_queues(struct net_device
*dev
)
5374 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
5375 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5376 spin_lock_init(&dev
->tx_global_lock
);
5380 * alloc_netdev_mq - allocate network device
5381 * @sizeof_priv: size of private data to allocate space for
5382 * @name: device name format string
5383 * @setup: callback to initialize device
5384 * @queue_count: the number of subqueues to allocate
5386 * Allocates a struct net_device with private data area for driver use
5387 * and performs basic initialization. Also allocates subquue structs
5388 * for each queue on the device at the end of the netdevice.
5390 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
5391 void (*setup
)(struct net_device
*), unsigned int queue_count
)
5393 struct netdev_queue
*tx
;
5394 struct net_device
*dev
;
5396 struct net_device
*p
;
5398 struct netdev_rx_queue
*rx
;
5402 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5404 alloc_size
= sizeof(struct net_device
);
5406 /* ensure 32-byte alignment of private area */
5407 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5408 alloc_size
+= sizeof_priv
;
5410 /* ensure 32-byte alignment of whole construct */
5411 alloc_size
+= NETDEV_ALIGN
- 1;
5413 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5415 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5419 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5421 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5427 rx
= kcalloc(queue_count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5429 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5434 atomic_set(&rx
->count
, queue_count
);
5437 * Set a pointer to first element in the array which holds the
5440 for (i
= 0; i
< queue_count
; i
++)
5444 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5445 dev
->padded
= (char *)dev
- (char *)p
;
5447 if (dev_addr_init(dev
))
5453 dev_net_set(dev
, &init_net
);
5456 dev
->num_tx_queues
= queue_count
;
5457 dev
->real_num_tx_queues
= queue_count
;
5461 dev
->num_rx_queues
= queue_count
;
5464 dev
->gso_max_size
= GSO_MAX_SIZE
;
5466 netdev_init_queues(dev
);
5468 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5469 dev
->ethtool_ntuple_list
.count
= 0;
5470 INIT_LIST_HEAD(&dev
->napi_list
);
5471 INIT_LIST_HEAD(&dev
->unreg_list
);
5472 INIT_LIST_HEAD(&dev
->link_watch_list
);
5473 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5475 strcpy(dev
->name
, name
);
5488 EXPORT_SYMBOL(alloc_netdev_mq
);
5491 * free_netdev - free network device
5494 * This function does the last stage of destroying an allocated device
5495 * interface. The reference to the device object is released.
5496 * If this is the last reference then it will be freed.
5498 void free_netdev(struct net_device
*dev
)
5500 struct napi_struct
*p
, *n
;
5502 release_net(dev_net(dev
));
5506 /* Flush device addresses */
5507 dev_addr_flush(dev
);
5509 /* Clear ethtool n-tuple list */
5510 ethtool_ntuple_flush(dev
);
5512 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5515 /* Compatibility with error handling in drivers */
5516 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5517 kfree((char *)dev
- dev
->padded
);
5521 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5522 dev
->reg_state
= NETREG_RELEASED
;
5524 /* will free via device release */
5525 put_device(&dev
->dev
);
5527 EXPORT_SYMBOL(free_netdev
);
5530 * synchronize_net - Synchronize with packet receive processing
5532 * Wait for packets currently being received to be done.
5533 * Does not block later packets from starting.
5535 void synchronize_net(void)
5540 EXPORT_SYMBOL(synchronize_net
);
5543 * unregister_netdevice_queue - remove device from the kernel
5547 * This function shuts down a device interface and removes it
5548 * from the kernel tables.
5549 * If head not NULL, device is queued to be unregistered later.
5551 * Callers must hold the rtnl semaphore. You may want
5552 * unregister_netdev() instead of this.
5555 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5560 list_move_tail(&dev
->unreg_list
, head
);
5562 rollback_registered(dev
);
5563 /* Finish processing unregister after unlock */
5567 EXPORT_SYMBOL(unregister_netdevice_queue
);
5570 * unregister_netdevice_many - unregister many devices
5571 * @head: list of devices
5573 void unregister_netdevice_many(struct list_head
*head
)
5575 struct net_device
*dev
;
5577 if (!list_empty(head
)) {
5578 rollback_registered_many(head
);
5579 list_for_each_entry(dev
, head
, unreg_list
)
5583 EXPORT_SYMBOL(unregister_netdevice_many
);
5586 * unregister_netdev - remove device from the kernel
5589 * This function shuts down a device interface and removes it
5590 * from the kernel tables.
5592 * This is just a wrapper for unregister_netdevice that takes
5593 * the rtnl semaphore. In general you want to use this and not
5594 * unregister_netdevice.
5596 void unregister_netdev(struct net_device
*dev
)
5599 unregister_netdevice(dev
);
5602 EXPORT_SYMBOL(unregister_netdev
);
5605 * dev_change_net_namespace - move device to different nethost namespace
5607 * @net: network namespace
5608 * @pat: If not NULL name pattern to try if the current device name
5609 * is already taken in the destination network namespace.
5611 * This function shuts down a device interface and moves it
5612 * to a new network namespace. On success 0 is returned, on
5613 * a failure a netagive errno code is returned.
5615 * Callers must hold the rtnl semaphore.
5618 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5624 /* Don't allow namespace local devices to be moved. */
5626 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5629 /* Ensure the device has been registrered */
5631 if (dev
->reg_state
!= NETREG_REGISTERED
)
5634 /* Get out if there is nothing todo */
5636 if (net_eq(dev_net(dev
), net
))
5639 /* Pick the destination device name, and ensure
5640 * we can use it in the destination network namespace.
5643 if (__dev_get_by_name(net
, dev
->name
)) {
5644 /* We get here if we can't use the current device name */
5647 if (dev_get_valid_name(dev
, pat
, 1))
5652 * And now a mini version of register_netdevice unregister_netdevice.
5655 /* If device is running close it first. */
5658 /* And unlink it from device chain */
5660 unlist_netdevice(dev
);
5664 /* Shutdown queueing discipline. */
5667 /* Notify protocols, that we are about to destroy
5668 this device. They should clean all the things.
5670 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5671 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5674 * Flush the unicast and multicast chains
5679 /* Actually switch the network namespace */
5680 dev_net_set(dev
, net
);
5682 /* If there is an ifindex conflict assign a new one */
5683 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5684 int iflink
= (dev
->iflink
== dev
->ifindex
);
5685 dev
->ifindex
= dev_new_index(net
);
5687 dev
->iflink
= dev
->ifindex
;
5690 /* Fixup kobjects */
5691 err
= device_rename(&dev
->dev
, dev
->name
);
5694 /* Add the device back in the hashes */
5695 list_netdevice(dev
);
5697 /* Notify protocols, that a new device appeared. */
5698 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5701 * Prevent userspace races by waiting until the network
5702 * device is fully setup before sending notifications.
5704 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5711 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5713 static int dev_cpu_callback(struct notifier_block
*nfb
,
5714 unsigned long action
,
5717 struct sk_buff
**list_skb
;
5718 struct sk_buff
*skb
;
5719 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5720 struct softnet_data
*sd
, *oldsd
;
5722 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5725 local_irq_disable();
5726 cpu
= smp_processor_id();
5727 sd
= &per_cpu(softnet_data
, cpu
);
5728 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5730 /* Find end of our completion_queue. */
5731 list_skb
= &sd
->completion_queue
;
5733 list_skb
= &(*list_skb
)->next
;
5734 /* Append completion queue from offline CPU. */
5735 *list_skb
= oldsd
->completion_queue
;
5736 oldsd
->completion_queue
= NULL
;
5738 /* Append output queue from offline CPU. */
5739 if (oldsd
->output_queue
) {
5740 *sd
->output_queue_tailp
= oldsd
->output_queue
;
5741 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
5742 oldsd
->output_queue
= NULL
;
5743 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
5746 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5749 /* Process offline CPU's input_pkt_queue */
5750 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
5752 input_queue_head_incr(oldsd
);
5754 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
5756 input_queue_head_incr(oldsd
);
5764 * netdev_increment_features - increment feature set by one
5765 * @all: current feature set
5766 * @one: new feature set
5767 * @mask: mask feature set
5769 * Computes a new feature set after adding a device with feature set
5770 * @one to the master device with current feature set @all. Will not
5771 * enable anything that is off in @mask. Returns the new feature set.
5773 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
5776 /* If device needs checksumming, downgrade to it. */
5777 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
5778 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
5779 else if (mask
& NETIF_F_ALL_CSUM
) {
5780 /* If one device supports v4/v6 checksumming, set for all. */
5781 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
5782 !(all
& NETIF_F_GEN_CSUM
)) {
5783 all
&= ~NETIF_F_ALL_CSUM
;
5784 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
5787 /* If one device supports hw checksumming, set for all. */
5788 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
5789 all
&= ~NETIF_F_ALL_CSUM
;
5790 all
|= NETIF_F_HW_CSUM
;
5794 one
|= NETIF_F_ALL_CSUM
;
5796 one
|= all
& NETIF_F_ONE_FOR_ALL
;
5797 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
5798 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
5802 EXPORT_SYMBOL(netdev_increment_features
);
5804 static struct hlist_head
*netdev_create_hash(void)
5807 struct hlist_head
*hash
;
5809 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5811 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5812 INIT_HLIST_HEAD(&hash
[i
]);
5817 /* Initialize per network namespace state */
5818 static int __net_init
netdev_init(struct net
*net
)
5820 INIT_LIST_HEAD(&net
->dev_base_head
);
5822 net
->dev_name_head
= netdev_create_hash();
5823 if (net
->dev_name_head
== NULL
)
5826 net
->dev_index_head
= netdev_create_hash();
5827 if (net
->dev_index_head
== NULL
)
5833 kfree(net
->dev_name_head
);
5839 * netdev_drivername - network driver for the device
5840 * @dev: network device
5841 * @buffer: buffer for resulting name
5842 * @len: size of buffer
5844 * Determine network driver for device.
5846 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
5848 const struct device_driver
*driver
;
5849 const struct device
*parent
;
5851 if (len
<= 0 || !buffer
)
5855 parent
= dev
->dev
.parent
;
5860 driver
= parent
->driver
;
5861 if (driver
&& driver
->name
)
5862 strlcpy(buffer
, driver
->name
, len
);
5866 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
5867 struct va_format
*vaf
)
5871 if (dev
&& dev
->dev
.parent
)
5872 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
5873 netdev_name(dev
), vaf
);
5875 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
5877 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
5882 int netdev_printk(const char *level
, const struct net_device
*dev
,
5883 const char *format
, ...)
5885 struct va_format vaf
;
5889 va_start(args
, format
);
5894 r
= __netdev_printk(level
, dev
, &vaf
);
5899 EXPORT_SYMBOL(netdev_printk
);
5901 #define define_netdev_printk_level(func, level) \
5902 int func(const struct net_device *dev, const char *fmt, ...) \
5905 struct va_format vaf; \
5908 va_start(args, fmt); \
5913 r = __netdev_printk(level, dev, &vaf); \
5918 EXPORT_SYMBOL(func);
5920 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
5921 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
5922 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
5923 define_netdev_printk_level(netdev_err
, KERN_ERR
);
5924 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
5925 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
5926 define_netdev_printk_level(netdev_info
, KERN_INFO
);
5928 static void __net_exit
netdev_exit(struct net
*net
)
5930 kfree(net
->dev_name_head
);
5931 kfree(net
->dev_index_head
);
5934 static struct pernet_operations __net_initdata netdev_net_ops
= {
5935 .init
= netdev_init
,
5936 .exit
= netdev_exit
,
5939 static void __net_exit
default_device_exit(struct net
*net
)
5941 struct net_device
*dev
, *aux
;
5943 * Push all migratable network devices back to the
5944 * initial network namespace
5947 for_each_netdev_safe(net
, dev
, aux
) {
5949 char fb_name
[IFNAMSIZ
];
5951 /* Ignore unmoveable devices (i.e. loopback) */
5952 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5955 /* Leave virtual devices for the generic cleanup */
5956 if (dev
->rtnl_link_ops
)
5959 /* Push remaing network devices to init_net */
5960 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
5961 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
5963 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
5964 __func__
, dev
->name
, err
);
5971 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
5973 /* At exit all network devices most be removed from a network
5974 * namespace. Do this in the reverse order of registeration.
5975 * Do this across as many network namespaces as possible to
5976 * improve batching efficiency.
5978 struct net_device
*dev
;
5980 LIST_HEAD(dev_kill_list
);
5983 list_for_each_entry(net
, net_list
, exit_list
) {
5984 for_each_netdev_reverse(net
, dev
) {
5985 if (dev
->rtnl_link_ops
)
5986 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
5988 unregister_netdevice_queue(dev
, &dev_kill_list
);
5991 unregister_netdevice_many(&dev_kill_list
);
5995 static struct pernet_operations __net_initdata default_device_ops
= {
5996 .exit
= default_device_exit
,
5997 .exit_batch
= default_device_exit_batch
,
6001 * Initialize the DEV module. At boot time this walks the device list and
6002 * unhooks any devices that fail to initialise (normally hardware not
6003 * present) and leaves us with a valid list of present and active devices.
6008 * This is called single threaded during boot, so no need
6009 * to take the rtnl semaphore.
6011 static int __init
net_dev_init(void)
6013 int i
, rc
= -ENOMEM
;
6015 BUG_ON(!dev_boot_phase
);
6017 if (dev_proc_init())
6020 if (netdev_kobject_init())
6023 INIT_LIST_HEAD(&ptype_all
);
6024 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6025 INIT_LIST_HEAD(&ptype_base
[i
]);
6027 if (register_pernet_subsys(&netdev_net_ops
))
6031 * Initialise the packet receive queues.
6034 for_each_possible_cpu(i
) {
6035 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6037 memset(sd
, 0, sizeof(*sd
));
6038 skb_queue_head_init(&sd
->input_pkt_queue
);
6039 skb_queue_head_init(&sd
->process_queue
);
6040 sd
->completion_queue
= NULL
;
6041 INIT_LIST_HEAD(&sd
->poll_list
);
6042 sd
->output_queue
= NULL
;
6043 sd
->output_queue_tailp
= &sd
->output_queue
;
6045 sd
->csd
.func
= rps_trigger_softirq
;
6051 sd
->backlog
.poll
= process_backlog
;
6052 sd
->backlog
.weight
= weight_p
;
6053 sd
->backlog
.gro_list
= NULL
;
6054 sd
->backlog
.gro_count
= 0;
6059 /* The loopback device is special if any other network devices
6060 * is present in a network namespace the loopback device must
6061 * be present. Since we now dynamically allocate and free the
6062 * loopback device ensure this invariant is maintained by
6063 * keeping the loopback device as the first device on the
6064 * list of network devices. Ensuring the loopback devices
6065 * is the first device that appears and the last network device
6068 if (register_pernet_device(&loopback_net_ops
))
6071 if (register_pernet_device(&default_device_ops
))
6074 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6075 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6077 hotcpu_notifier(dev_cpu_callback
, 0);
6085 subsys_initcall(net_dev_init
);
6087 static int __init
initialize_hashrnd(void)
6089 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6093 late_initcall_sync(initialize_hashrnd
);