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 <linux/pci.h>
132 #include <linux/inetdevice.h>
134 #include "net-sysfs.h"
136 /* Instead of increasing this, you should create a hash table. */
137 #define MAX_GRO_SKBS 8
139 /* This should be increased if a protocol with a bigger head is added. */
140 #define GRO_MAX_HEAD (MAX_HEADER + 128)
143 * The list of packet types we will receive (as opposed to discard)
144 * and the routines to invoke.
146 * Why 16. Because with 16 the only overlap we get on a hash of the
147 * low nibble of the protocol value is RARP/SNAP/X.25.
149 * NOTE: That is no longer true with the addition of VLAN tags. Not
150 * sure which should go first, but I bet it won't make much
151 * difference if we are running VLANs. The good news is that
152 * this protocol won't be in the list unless compiled in, so
153 * the average user (w/out VLANs) will not be adversely affected.
170 #define PTYPE_HASH_SIZE (16)
171 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
173 static DEFINE_SPINLOCK(ptype_lock
);
174 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
175 static struct list_head ptype_all __read_mostly
; /* Taps */
178 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
181 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
183 * Writers must hold the rtnl semaphore while they loop through the
184 * dev_base_head list, and hold dev_base_lock for writing when they do the
185 * actual updates. This allows pure readers to access the list even
186 * while a writer is preparing to update it.
188 * To put it another way, dev_base_lock is held for writing only to
189 * protect against pure readers; the rtnl semaphore provides the
190 * protection against other writers.
192 * See, for example usages, register_netdevice() and
193 * unregister_netdevice(), which must be called with the rtnl
196 DEFINE_RWLOCK(dev_base_lock
);
197 EXPORT_SYMBOL(dev_base_lock
);
199 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
201 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
202 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
205 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
207 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
210 static inline void rps_lock(struct softnet_data
*sd
)
213 spin_lock(&sd
->input_pkt_queue
.lock
);
217 static inline void rps_unlock(struct softnet_data
*sd
)
220 spin_unlock(&sd
->input_pkt_queue
.lock
);
224 /* Device list insertion */
225 static int list_netdevice(struct net_device
*dev
)
227 struct net
*net
= dev_net(dev
);
231 write_lock_bh(&dev_base_lock
);
232 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
233 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
234 hlist_add_head_rcu(&dev
->index_hlist
,
235 dev_index_hash(net
, dev
->ifindex
));
236 write_unlock_bh(&dev_base_lock
);
240 /* Device list removal
241 * caller must respect a RCU grace period before freeing/reusing dev
243 static void unlist_netdevice(struct net_device
*dev
)
247 /* Unlink dev from the device chain */
248 write_lock_bh(&dev_base_lock
);
249 list_del_rcu(&dev
->dev_list
);
250 hlist_del_rcu(&dev
->name_hlist
);
251 hlist_del_rcu(&dev
->index_hlist
);
252 write_unlock_bh(&dev_base_lock
);
259 static RAW_NOTIFIER_HEAD(netdev_chain
);
262 * Device drivers call our routines to queue packets here. We empty the
263 * queue in the local softnet handler.
266 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
267 EXPORT_PER_CPU_SYMBOL(softnet_data
);
269 #ifdef CONFIG_LOCKDEP
271 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
272 * according to dev->type
274 static const unsigned short netdev_lock_type
[] =
275 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
276 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
277 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
278 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
279 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
280 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
281 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
282 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
283 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
284 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
285 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
286 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
287 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
288 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
289 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
290 ARPHRD_VOID
, ARPHRD_NONE
};
292 static const char *const netdev_lock_name
[] =
293 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
294 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
295 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
296 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
297 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
298 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
299 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
300 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
301 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
302 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
303 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
304 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
305 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
306 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
307 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
308 "_xmit_VOID", "_xmit_NONE"};
310 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
311 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
313 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
317 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
318 if (netdev_lock_type
[i
] == dev_type
)
320 /* the last key is used by default */
321 return ARRAY_SIZE(netdev_lock_type
) - 1;
324 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
325 unsigned short dev_type
)
329 i
= netdev_lock_pos(dev_type
);
330 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
331 netdev_lock_name
[i
]);
334 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
338 i
= netdev_lock_pos(dev
->type
);
339 lockdep_set_class_and_name(&dev
->addr_list_lock
,
340 &netdev_addr_lock_key
[i
],
341 netdev_lock_name
[i
]);
344 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
345 unsigned short dev_type
)
348 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
353 /*******************************************************************************
355 Protocol management and registration routines
357 *******************************************************************************/
360 * Add a protocol ID to the list. Now that the input handler is
361 * smarter we can dispense with all the messy stuff that used to be
364 * BEWARE!!! Protocol handlers, mangling input packets,
365 * MUST BE last in hash buckets and checking protocol handlers
366 * MUST start from promiscuous ptype_all chain in net_bh.
367 * It is true now, do not change it.
368 * Explanation follows: if protocol handler, mangling packet, will
369 * be the first on list, it is not able to sense, that packet
370 * is cloned and should be copied-on-write, so that it will
371 * change it and subsequent readers will get broken packet.
375 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
377 if (pt
->type
== htons(ETH_P_ALL
))
380 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
384 * dev_add_pack - add packet handler
385 * @pt: packet type declaration
387 * Add a protocol handler to the networking stack. The passed &packet_type
388 * is linked into kernel lists and may not be freed until it has been
389 * removed from the kernel lists.
391 * This call does not sleep therefore it can not
392 * guarantee all CPU's that are in middle of receiving packets
393 * will see the new packet type (until the next received packet).
396 void dev_add_pack(struct packet_type
*pt
)
398 struct list_head
*head
= ptype_head(pt
);
400 spin_lock(&ptype_lock
);
401 list_add_rcu(&pt
->list
, head
);
402 spin_unlock(&ptype_lock
);
404 EXPORT_SYMBOL(dev_add_pack
);
407 * __dev_remove_pack - remove packet handler
408 * @pt: packet type declaration
410 * Remove a protocol handler that was previously added to the kernel
411 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
412 * from the kernel lists and can be freed or reused once this function
415 * The packet type might still be in use by receivers
416 * and must not be freed until after all the CPU's have gone
417 * through a quiescent state.
419 void __dev_remove_pack(struct packet_type
*pt
)
421 struct list_head
*head
= ptype_head(pt
);
422 struct packet_type
*pt1
;
424 spin_lock(&ptype_lock
);
426 list_for_each_entry(pt1
, head
, list
) {
428 list_del_rcu(&pt
->list
);
433 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
435 spin_unlock(&ptype_lock
);
437 EXPORT_SYMBOL(__dev_remove_pack
);
440 * dev_remove_pack - remove packet handler
441 * @pt: packet type declaration
443 * Remove a protocol handler that was previously added to the kernel
444 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
445 * from the kernel lists and can be freed or reused once this function
448 * This call sleeps to guarantee that no CPU is looking at the packet
451 void dev_remove_pack(struct packet_type
*pt
)
453 __dev_remove_pack(pt
);
457 EXPORT_SYMBOL(dev_remove_pack
);
459 /******************************************************************************
461 Device Boot-time Settings Routines
463 *******************************************************************************/
465 /* Boot time configuration table */
466 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
469 * netdev_boot_setup_add - add new setup entry
470 * @name: name of the device
471 * @map: configured settings for the device
473 * Adds new setup entry to the dev_boot_setup list. The function
474 * returns 0 on error and 1 on success. This is a generic routine to
477 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
479 struct netdev_boot_setup
*s
;
483 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
484 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
485 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
486 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
487 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
492 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
496 * netdev_boot_setup_check - check boot time settings
497 * @dev: the netdevice
499 * Check boot time settings for the device.
500 * The found settings are set for the device to be used
501 * later in the device probing.
502 * Returns 0 if no settings found, 1 if they are.
504 int netdev_boot_setup_check(struct net_device
*dev
)
506 struct netdev_boot_setup
*s
= dev_boot_setup
;
509 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
510 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
511 !strcmp(dev
->name
, s
[i
].name
)) {
512 dev
->irq
= s
[i
].map
.irq
;
513 dev
->base_addr
= s
[i
].map
.base_addr
;
514 dev
->mem_start
= s
[i
].map
.mem_start
;
515 dev
->mem_end
= s
[i
].map
.mem_end
;
521 EXPORT_SYMBOL(netdev_boot_setup_check
);
525 * netdev_boot_base - get address from boot time settings
526 * @prefix: prefix for network device
527 * @unit: id for network device
529 * Check boot time settings for the base address of device.
530 * The found settings are set for the device to be used
531 * later in the device probing.
532 * Returns 0 if no settings found.
534 unsigned long netdev_boot_base(const char *prefix
, int unit
)
536 const struct netdev_boot_setup
*s
= dev_boot_setup
;
540 sprintf(name
, "%s%d", prefix
, unit
);
543 * If device already registered then return base of 1
544 * to indicate not to probe for this interface
546 if (__dev_get_by_name(&init_net
, name
))
549 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
550 if (!strcmp(name
, s
[i
].name
))
551 return s
[i
].map
.base_addr
;
556 * Saves at boot time configured settings for any netdevice.
558 int __init
netdev_boot_setup(char *str
)
563 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
568 memset(&map
, 0, sizeof(map
));
572 map
.base_addr
= ints
[2];
574 map
.mem_start
= ints
[3];
576 map
.mem_end
= ints
[4];
578 /* Add new entry to the list */
579 return netdev_boot_setup_add(str
, &map
);
582 __setup("netdev=", netdev_boot_setup
);
584 /*******************************************************************************
586 Device Interface Subroutines
588 *******************************************************************************/
591 * __dev_get_by_name - find a device by its name
592 * @net: the applicable net namespace
593 * @name: name to find
595 * Find an interface by name. Must be called under RTNL semaphore
596 * or @dev_base_lock. If the name is found a pointer to the device
597 * is returned. If the name is not found then %NULL is returned. The
598 * reference counters are not incremented so the caller must be
599 * careful with locks.
602 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
604 struct hlist_node
*p
;
605 struct net_device
*dev
;
606 struct hlist_head
*head
= dev_name_hash(net
, name
);
608 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
609 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
614 EXPORT_SYMBOL(__dev_get_by_name
);
617 * dev_get_by_name_rcu - find a device by its name
618 * @net: the applicable net namespace
619 * @name: name to find
621 * Find an interface by name.
622 * If the name is found a pointer to the device is returned.
623 * If the name is not found then %NULL is returned.
624 * The reference counters are not incremented so the caller must be
625 * careful with locks. The caller must hold RCU lock.
628 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
630 struct hlist_node
*p
;
631 struct net_device
*dev
;
632 struct hlist_head
*head
= dev_name_hash(net
, name
);
634 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
635 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
640 EXPORT_SYMBOL(dev_get_by_name_rcu
);
643 * dev_get_by_name - find a device by its name
644 * @net: the applicable net namespace
645 * @name: name to find
647 * Find an interface by name. This can be called from any
648 * context and does its own locking. The returned handle has
649 * the usage count incremented and the caller must use dev_put() to
650 * release it when it is no longer needed. %NULL is returned if no
651 * matching device is found.
654 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
656 struct net_device
*dev
;
659 dev
= dev_get_by_name_rcu(net
, name
);
665 EXPORT_SYMBOL(dev_get_by_name
);
668 * __dev_get_by_index - find a device by its ifindex
669 * @net: the applicable net namespace
670 * @ifindex: index of device
672 * Search for an interface by index. Returns %NULL if the device
673 * is not found or a pointer to the device. The device has not
674 * had its reference counter increased so the caller must be careful
675 * about locking. The caller must hold either the RTNL semaphore
679 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
681 struct hlist_node
*p
;
682 struct net_device
*dev
;
683 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
685 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
686 if (dev
->ifindex
== ifindex
)
691 EXPORT_SYMBOL(__dev_get_by_index
);
694 * dev_get_by_index_rcu - find a device by its ifindex
695 * @net: the applicable net namespace
696 * @ifindex: index of device
698 * Search for an interface by index. Returns %NULL if the device
699 * is not found or a pointer to the device. The device has not
700 * had its reference counter increased so the caller must be careful
701 * about locking. The caller must hold RCU lock.
704 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
706 struct hlist_node
*p
;
707 struct net_device
*dev
;
708 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
710 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
711 if (dev
->ifindex
== ifindex
)
716 EXPORT_SYMBOL(dev_get_by_index_rcu
);
720 * dev_get_by_index - find a device by its ifindex
721 * @net: the applicable net namespace
722 * @ifindex: index of device
724 * Search for an interface by index. Returns NULL if the device
725 * is not found or a pointer to the device. The device returned has
726 * had a reference added and the pointer is safe until the user calls
727 * dev_put to indicate they have finished with it.
730 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
732 struct net_device
*dev
;
735 dev
= dev_get_by_index_rcu(net
, ifindex
);
741 EXPORT_SYMBOL(dev_get_by_index
);
744 * dev_getbyhwaddr - find a device by its hardware address
745 * @net: the applicable net namespace
746 * @type: media type of device
747 * @ha: hardware address
749 * Search for an interface by MAC address. Returns NULL if the device
750 * is not found or a pointer to the device. The caller must hold the
751 * rtnl semaphore. The returned device has not had its ref count increased
752 * and the caller must therefore be careful about locking
755 * If the API was consistent this would be __dev_get_by_hwaddr
758 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
760 struct net_device
*dev
;
764 for_each_netdev(net
, dev
)
765 if (dev
->type
== type
&&
766 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
771 EXPORT_SYMBOL(dev_getbyhwaddr
);
773 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
775 struct net_device
*dev
;
778 for_each_netdev(net
, dev
)
779 if (dev
->type
== type
)
784 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
786 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
788 struct net_device
*dev
, *ret
= NULL
;
791 for_each_netdev_rcu(net
, dev
)
792 if (dev
->type
== type
) {
800 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
803 * dev_get_by_flags_rcu - find any device with given flags
804 * @net: the applicable net namespace
805 * @if_flags: IFF_* values
806 * @mask: bitmask of bits in if_flags to check
808 * Search for any interface with the given flags. Returns NULL if a device
809 * is not found or a pointer to the device. Must be called inside
810 * rcu_read_lock(), and result refcount is unchanged.
813 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
816 struct net_device
*dev
, *ret
;
819 for_each_netdev_rcu(net
, dev
) {
820 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
827 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
830 * dev_valid_name - check if name is okay for network device
833 * Network device names need to be valid file names to
834 * to allow sysfs to work. We also disallow any kind of
837 int dev_valid_name(const char *name
)
841 if (strlen(name
) >= IFNAMSIZ
)
843 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
847 if (*name
== '/' || isspace(*name
))
853 EXPORT_SYMBOL(dev_valid_name
);
856 * __dev_alloc_name - allocate a name for a device
857 * @net: network namespace to allocate the device name in
858 * @name: name format string
859 * @buf: scratch buffer and result name string
861 * Passed a format string - eg "lt%d" it will try and find a suitable
862 * id. It scans list of devices to build up a free map, then chooses
863 * the first empty slot. The caller must hold the dev_base or rtnl lock
864 * while allocating the name and adding the device in order to avoid
866 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
867 * Returns the number of the unit assigned or a negative errno code.
870 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
874 const int max_netdevices
= 8*PAGE_SIZE
;
875 unsigned long *inuse
;
876 struct net_device
*d
;
878 p
= strnchr(name
, IFNAMSIZ
-1, '%');
881 * Verify the string as this thing may have come from
882 * the user. There must be either one "%d" and no other "%"
885 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
888 /* Use one page as a bit array of possible slots */
889 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
893 for_each_netdev(net
, d
) {
894 if (!sscanf(d
->name
, name
, &i
))
896 if (i
< 0 || i
>= max_netdevices
)
899 /* avoid cases where sscanf is not exact inverse of printf */
900 snprintf(buf
, IFNAMSIZ
, name
, i
);
901 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
905 i
= find_first_zero_bit(inuse
, max_netdevices
);
906 free_page((unsigned long) inuse
);
910 snprintf(buf
, IFNAMSIZ
, name
, i
);
911 if (!__dev_get_by_name(net
, buf
))
914 /* It is possible to run out of possible slots
915 * when the name is long and there isn't enough space left
916 * for the digits, or if all bits are used.
922 * dev_alloc_name - allocate a name for a device
924 * @name: name format string
926 * Passed a format string - eg "lt%d" it will try and find a suitable
927 * id. It scans list of devices to build up a free map, then chooses
928 * the first empty slot. The caller must hold the dev_base or rtnl lock
929 * while allocating the name and adding the device in order to avoid
931 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
932 * Returns the number of the unit assigned or a negative errno code.
935 int dev_alloc_name(struct net_device
*dev
, const char *name
)
941 BUG_ON(!dev_net(dev
));
943 ret
= __dev_alloc_name(net
, name
, buf
);
945 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
948 EXPORT_SYMBOL(dev_alloc_name
);
950 static int dev_get_valid_name(struct net_device
*dev
, const char *name
, bool fmt
)
954 BUG_ON(!dev_net(dev
));
957 if (!dev_valid_name(name
))
960 if (fmt
&& strchr(name
, '%'))
961 return dev_alloc_name(dev
, name
);
962 else if (__dev_get_by_name(net
, name
))
964 else if (dev
->name
!= name
)
965 strlcpy(dev
->name
, name
, IFNAMSIZ
);
971 * dev_change_name - change name of a device
973 * @newname: name (or format string) must be at least IFNAMSIZ
975 * Change name of a device, can pass format strings "eth%d".
978 int dev_change_name(struct net_device
*dev
, const char *newname
)
980 char oldname
[IFNAMSIZ
];
986 BUG_ON(!dev_net(dev
));
989 if (dev
->flags
& IFF_UP
)
992 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
995 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
997 err
= dev_get_valid_name(dev
, newname
, 1);
1002 ret
= device_rename(&dev
->dev
, dev
->name
);
1004 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1008 write_lock_bh(&dev_base_lock
);
1009 hlist_del(&dev
->name_hlist
);
1010 write_unlock_bh(&dev_base_lock
);
1014 write_lock_bh(&dev_base_lock
);
1015 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1016 write_unlock_bh(&dev_base_lock
);
1018 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1019 ret
= notifier_to_errno(ret
);
1022 /* err >= 0 after dev_alloc_name() or stores the first errno */
1025 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1029 "%s: name change rollback failed: %d.\n",
1038 * dev_set_alias - change ifalias of a device
1040 * @alias: name up to IFALIASZ
1041 * @len: limit of bytes to copy from info
1043 * Set ifalias for a device,
1045 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1049 if (len
>= IFALIASZ
)
1054 kfree(dev
->ifalias
);
1055 dev
->ifalias
= NULL
;
1060 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1064 strlcpy(dev
->ifalias
, alias
, len
+1);
1070 * netdev_features_change - device changes features
1071 * @dev: device to cause notification
1073 * Called to indicate a device has changed features.
1075 void netdev_features_change(struct net_device
*dev
)
1077 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1079 EXPORT_SYMBOL(netdev_features_change
);
1082 * netdev_state_change - device changes state
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed state. This function calls
1086 * the notifier chains for netdev_chain and sends a NEWLINK message
1087 * to the routing socket.
1089 void netdev_state_change(struct net_device
*dev
)
1091 if (dev
->flags
& IFF_UP
) {
1092 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1093 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1096 EXPORT_SYMBOL(netdev_state_change
);
1098 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1100 return call_netdevice_notifiers(event
, dev
);
1102 EXPORT_SYMBOL(netdev_bonding_change
);
1105 * dev_load - load a network module
1106 * @net: the applicable net namespace
1107 * @name: name of interface
1109 * If a network interface is not present and the process has suitable
1110 * privileges this function loads the module. If module loading is not
1111 * available in this kernel then it becomes a nop.
1114 void dev_load(struct net
*net
, const char *name
)
1116 struct net_device
*dev
;
1119 dev
= dev_get_by_name_rcu(net
, name
);
1122 if (!dev
&& capable(CAP_NET_ADMIN
))
1123 request_module("%s", name
);
1125 EXPORT_SYMBOL(dev_load
);
1127 static int __dev_open(struct net_device
*dev
)
1129 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1135 * Is it even present?
1137 if (!netif_device_present(dev
))
1140 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1141 ret
= notifier_to_errno(ret
);
1146 * Call device private open method
1148 set_bit(__LINK_STATE_START
, &dev
->state
);
1150 if (ops
->ndo_validate_addr
)
1151 ret
= ops
->ndo_validate_addr(dev
);
1153 if (!ret
&& ops
->ndo_open
)
1154 ret
= ops
->ndo_open(dev
);
1157 * If it went open OK then:
1161 clear_bit(__LINK_STATE_START
, &dev
->state
);
1166 dev
->flags
|= IFF_UP
;
1171 net_dmaengine_get();
1174 * Initialize multicasting status
1176 dev_set_rx_mode(dev
);
1179 * Wakeup transmit queue engine
1188 * dev_open - prepare an interface for use.
1189 * @dev: device to open
1191 * Takes a device from down to up state. The device's private open
1192 * function is invoked and then the multicast lists are loaded. Finally
1193 * the device is moved into the up state and a %NETDEV_UP message is
1194 * sent to the netdev notifier chain.
1196 * Calling this function on an active interface is a nop. On a failure
1197 * a negative errno code is returned.
1199 int dev_open(struct net_device
*dev
)
1206 if (dev
->flags
& IFF_UP
)
1212 ret
= __dev_open(dev
);
1217 * ... and announce new interface.
1219 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1220 call_netdevice_notifiers(NETDEV_UP
, dev
);
1224 EXPORT_SYMBOL(dev_open
);
1226 static int __dev_close(struct net_device
*dev
)
1228 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1234 * Tell people we are going down, so that they can
1235 * prepare to death, when device is still operating.
1237 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1239 clear_bit(__LINK_STATE_START
, &dev
->state
);
1241 /* Synchronize to scheduled poll. We cannot touch poll list,
1242 * it can be even on different cpu. So just clear netif_running().
1244 * dev->stop() will invoke napi_disable() on all of it's
1245 * napi_struct instances on this device.
1247 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1249 dev_deactivate(dev
);
1252 * Call the device specific close. This cannot fail.
1253 * Only if device is UP
1255 * We allow it to be called even after a DETACH hot-plug
1262 * Device is now down.
1265 dev
->flags
&= ~IFF_UP
;
1270 net_dmaengine_put();
1276 * dev_close - shutdown an interface.
1277 * @dev: device to shutdown
1279 * This function moves an active device into down state. A
1280 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1281 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1284 int dev_close(struct net_device
*dev
)
1286 if (!(dev
->flags
& IFF_UP
))
1292 * Tell people we are down
1294 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1295 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1299 EXPORT_SYMBOL(dev_close
);
1303 * dev_disable_lro - disable Large Receive Offload on a device
1306 * Disable Large Receive Offload (LRO) on a net device. Must be
1307 * called under RTNL. This is needed if received packets may be
1308 * forwarded to another interface.
1310 void dev_disable_lro(struct net_device
*dev
)
1312 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1313 dev
->ethtool_ops
->set_flags
) {
1314 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1315 if (flags
& ETH_FLAG_LRO
) {
1316 flags
&= ~ETH_FLAG_LRO
;
1317 dev
->ethtool_ops
->set_flags(dev
, flags
);
1320 WARN_ON(dev
->features
& NETIF_F_LRO
);
1322 EXPORT_SYMBOL(dev_disable_lro
);
1325 static int dev_boot_phase
= 1;
1328 * Device change register/unregister. These are not inline or static
1329 * as we export them to the world.
1333 * register_netdevice_notifier - register a network notifier block
1336 * Register a notifier to be called when network device events occur.
1337 * The notifier passed is linked into the kernel structures and must
1338 * not be reused until it has been unregistered. A negative errno code
1339 * is returned on a failure.
1341 * When registered all registration and up events are replayed
1342 * to the new notifier to allow device to have a race free
1343 * view of the network device list.
1346 int register_netdevice_notifier(struct notifier_block
*nb
)
1348 struct net_device
*dev
;
1349 struct net_device
*last
;
1354 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1360 for_each_netdev(net
, dev
) {
1361 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1362 err
= notifier_to_errno(err
);
1366 if (!(dev
->flags
& IFF_UP
))
1369 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1380 for_each_netdev(net
, dev
) {
1384 if (dev
->flags
& IFF_UP
) {
1385 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1386 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1388 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1389 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1393 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1396 EXPORT_SYMBOL(register_netdevice_notifier
);
1399 * unregister_netdevice_notifier - unregister a network notifier block
1402 * Unregister a notifier previously registered by
1403 * register_netdevice_notifier(). The notifier is unlinked into the
1404 * kernel structures and may then be reused. A negative errno code
1405 * is returned on a failure.
1408 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1413 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1417 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1420 * call_netdevice_notifiers - call all network notifier blocks
1421 * @val: value passed unmodified to notifier function
1422 * @dev: net_device pointer passed unmodified to notifier function
1424 * Call all network notifier blocks. Parameters and return value
1425 * are as for raw_notifier_call_chain().
1428 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1431 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1434 /* When > 0 there are consumers of rx skb time stamps */
1435 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1437 void net_enable_timestamp(void)
1439 atomic_inc(&netstamp_needed
);
1441 EXPORT_SYMBOL(net_enable_timestamp
);
1443 void net_disable_timestamp(void)
1445 atomic_dec(&netstamp_needed
);
1447 EXPORT_SYMBOL(net_disable_timestamp
);
1449 static inline void net_timestamp_set(struct sk_buff
*skb
)
1451 if (atomic_read(&netstamp_needed
))
1452 __net_timestamp(skb
);
1454 skb
->tstamp
.tv64
= 0;
1457 static inline void net_timestamp_check(struct sk_buff
*skb
)
1459 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1460 __net_timestamp(skb
);
1464 * dev_forward_skb - loopback an skb to another netif
1466 * @dev: destination network device
1467 * @skb: buffer to forward
1470 * NET_RX_SUCCESS (no congestion)
1471 * NET_RX_DROP (packet was dropped, but freed)
1473 * dev_forward_skb can be used for injecting an skb from the
1474 * start_xmit function of one device into the receive queue
1475 * of another device.
1477 * The receiving device may be in another namespace, so
1478 * we have to clear all information in the skb that could
1479 * impact namespace isolation.
1481 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1486 if (!(dev
->flags
& IFF_UP
) ||
1487 (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
))) {
1491 skb_set_dev(skb
, dev
);
1492 skb
->tstamp
.tv64
= 0;
1493 skb
->pkt_type
= PACKET_HOST
;
1494 skb
->protocol
= eth_type_trans(skb
, dev
);
1495 return netif_rx(skb
);
1497 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1500 * Support routine. Sends outgoing frames to any network
1501 * taps currently in use.
1504 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1506 struct packet_type
*ptype
;
1508 #ifdef CONFIG_NET_CLS_ACT
1509 if (!(skb
->tstamp
.tv64
&& (G_TC_FROM(skb
->tc_verd
) & AT_INGRESS
)))
1510 net_timestamp_set(skb
);
1512 net_timestamp_set(skb
);
1516 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1517 /* Never send packets back to the socket
1518 * they originated from - MvS (miquels@drinkel.ow.org)
1520 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1521 (ptype
->af_packet_priv
== NULL
||
1522 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1523 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1527 /* skb->nh should be correctly
1528 set by sender, so that the second statement is
1529 just protection against buggy protocols.
1531 skb_reset_mac_header(skb2
);
1533 if (skb_network_header(skb2
) < skb2
->data
||
1534 skb2
->network_header
> skb2
->tail
) {
1535 if (net_ratelimit())
1536 printk(KERN_CRIT
"protocol %04x is "
1538 ntohs(skb2
->protocol
),
1540 skb_reset_network_header(skb2
);
1543 skb2
->transport_header
= skb2
->network_header
;
1544 skb2
->pkt_type
= PACKET_OUTGOING
;
1545 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1552 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1553 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1555 void netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1557 unsigned int real_num
= dev
->real_num_tx_queues
;
1559 if (unlikely(txq
> dev
->num_tx_queues
))
1561 else if (txq
> real_num
)
1562 dev
->real_num_tx_queues
= txq
;
1563 else if (txq
< real_num
) {
1564 dev
->real_num_tx_queues
= txq
;
1565 qdisc_reset_all_tx_gt(dev
, txq
);
1568 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1570 static inline void __netif_reschedule(struct Qdisc
*q
)
1572 struct softnet_data
*sd
;
1573 unsigned long flags
;
1575 local_irq_save(flags
);
1576 sd
= &__get_cpu_var(softnet_data
);
1577 q
->next_sched
= NULL
;
1578 *sd
->output_queue_tailp
= q
;
1579 sd
->output_queue_tailp
= &q
->next_sched
;
1580 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1581 local_irq_restore(flags
);
1584 void __netif_schedule(struct Qdisc
*q
)
1586 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1587 __netif_reschedule(q
);
1589 EXPORT_SYMBOL(__netif_schedule
);
1591 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1593 if (atomic_dec_and_test(&skb
->users
)) {
1594 struct softnet_data
*sd
;
1595 unsigned long flags
;
1597 local_irq_save(flags
);
1598 sd
= &__get_cpu_var(softnet_data
);
1599 skb
->next
= sd
->completion_queue
;
1600 sd
->completion_queue
= skb
;
1601 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1602 local_irq_restore(flags
);
1605 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1607 void dev_kfree_skb_any(struct sk_buff
*skb
)
1609 if (in_irq() || irqs_disabled())
1610 dev_kfree_skb_irq(skb
);
1614 EXPORT_SYMBOL(dev_kfree_skb_any
);
1618 * netif_device_detach - mark device as removed
1619 * @dev: network device
1621 * Mark device as removed from system and therefore no longer available.
1623 void netif_device_detach(struct net_device
*dev
)
1625 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1626 netif_running(dev
)) {
1627 netif_tx_stop_all_queues(dev
);
1630 EXPORT_SYMBOL(netif_device_detach
);
1633 * netif_device_attach - mark device as attached
1634 * @dev: network device
1636 * Mark device as attached from system and restart if needed.
1638 void netif_device_attach(struct net_device
*dev
)
1640 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1641 netif_running(dev
)) {
1642 netif_tx_wake_all_queues(dev
);
1643 __netdev_watchdog_up(dev
);
1646 EXPORT_SYMBOL(netif_device_attach
);
1648 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1650 return ((features
& NETIF_F_GEN_CSUM
) ||
1651 ((features
& NETIF_F_IP_CSUM
) &&
1652 protocol
== htons(ETH_P_IP
)) ||
1653 ((features
& NETIF_F_IPV6_CSUM
) &&
1654 protocol
== htons(ETH_P_IPV6
)) ||
1655 ((features
& NETIF_F_FCOE_CRC
) &&
1656 protocol
== htons(ETH_P_FCOE
)));
1659 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1661 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1664 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1665 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1666 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1667 veh
->h_vlan_encapsulated_proto
))
1675 * skb_dev_set -- assign a new device to a buffer
1676 * @skb: buffer for the new device
1677 * @dev: network device
1679 * If an skb is owned by a device already, we have to reset
1680 * all data private to the namespace a device belongs to
1681 * before assigning it a new device.
1683 #ifdef CONFIG_NET_NS
1684 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1687 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1690 skb_init_secmark(skb
);
1694 skb
->ipvs_property
= 0;
1695 #ifdef CONFIG_NET_SCHED
1701 EXPORT_SYMBOL(skb_set_dev
);
1702 #endif /* CONFIG_NET_NS */
1705 * Invalidate hardware checksum when packet is to be mangled, and
1706 * complete checksum manually on outgoing path.
1708 int skb_checksum_help(struct sk_buff
*skb
)
1711 int ret
= 0, offset
;
1713 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1714 goto out_set_summed
;
1716 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1717 /* Let GSO fix up the checksum. */
1718 goto out_set_summed
;
1721 offset
= skb
->csum_start
- skb_headroom(skb
);
1722 BUG_ON(offset
>= skb_headlen(skb
));
1723 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1725 offset
+= skb
->csum_offset
;
1726 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1728 if (skb_cloned(skb
) &&
1729 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1730 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1735 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1737 skb
->ip_summed
= CHECKSUM_NONE
;
1741 EXPORT_SYMBOL(skb_checksum_help
);
1744 * skb_gso_segment - Perform segmentation on skb.
1745 * @skb: buffer to segment
1746 * @features: features for the output path (see dev->features)
1748 * This function segments the given skb and returns a list of segments.
1750 * It may return NULL if the skb requires no segmentation. This is
1751 * only possible when GSO is used for verifying header integrity.
1753 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1755 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1756 struct packet_type
*ptype
;
1757 __be16 type
= skb
->protocol
;
1760 skb_reset_mac_header(skb
);
1761 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1762 __skb_pull(skb
, skb
->mac_len
);
1764 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1765 struct net_device
*dev
= skb
->dev
;
1766 struct ethtool_drvinfo info
= {};
1768 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1769 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1771 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1773 info
.driver
, dev
? dev
->features
: 0L,
1774 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1775 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1777 if (skb_header_cloned(skb
) &&
1778 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1779 return ERR_PTR(err
);
1783 list_for_each_entry_rcu(ptype
,
1784 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1785 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1786 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1787 err
= ptype
->gso_send_check(skb
);
1788 segs
= ERR_PTR(err
);
1789 if (err
|| skb_gso_ok(skb
, features
))
1791 __skb_push(skb
, (skb
->data
-
1792 skb_network_header(skb
)));
1794 segs
= ptype
->gso_segment(skb
, features
);
1800 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1804 EXPORT_SYMBOL(skb_gso_segment
);
1806 /* Take action when hardware reception checksum errors are detected. */
1808 void netdev_rx_csum_fault(struct net_device
*dev
)
1810 if (net_ratelimit()) {
1811 printk(KERN_ERR
"%s: hw csum failure.\n",
1812 dev
? dev
->name
: "<unknown>");
1816 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1819 /* Actually, we should eliminate this check as soon as we know, that:
1820 * 1. IOMMU is present and allows to map all the memory.
1821 * 2. No high memory really exists on this machine.
1824 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1826 #ifdef CONFIG_HIGHMEM
1828 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1829 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1830 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1834 if (PCI_DMA_BUS_IS_PHYS
) {
1835 struct device
*pdev
= dev
->dev
.parent
;
1839 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1840 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1841 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1850 void (*destructor
)(struct sk_buff
*skb
);
1853 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1855 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1857 struct dev_gso_cb
*cb
;
1860 struct sk_buff
*nskb
= skb
->next
;
1862 skb
->next
= nskb
->next
;
1865 } while (skb
->next
);
1867 cb
= DEV_GSO_CB(skb
);
1869 cb
->destructor(skb
);
1873 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1874 * @skb: buffer to segment
1876 * This function segments the given skb and stores the list of segments
1879 static int dev_gso_segment(struct sk_buff
*skb
)
1881 struct net_device
*dev
= skb
->dev
;
1882 struct sk_buff
*segs
;
1883 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1886 segs
= skb_gso_segment(skb
, features
);
1888 /* Verifying header integrity only. */
1893 return PTR_ERR(segs
);
1896 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1897 skb
->destructor
= dev_gso_skb_destructor
;
1903 * Try to orphan skb early, right before transmission by the device.
1904 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1905 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1907 static inline void skb_orphan_try(struct sk_buff
*skb
)
1909 struct sock
*sk
= skb
->sk
;
1911 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
1912 /* skb_tx_hash() wont be able to get sk.
1913 * We copy sk_hash into skb->rxhash
1916 skb
->rxhash
= sk
->sk_hash
;
1922 * Returns true if either:
1923 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1924 * 2. skb is fragmented and the device does not support SG, or if
1925 * at least one of fragments is in highmem and device does not
1926 * support DMA from it.
1928 static inline int skb_needs_linearize(struct sk_buff
*skb
,
1929 struct net_device
*dev
)
1931 return skb_is_nonlinear(skb
) &&
1932 ((skb_has_frag_list(skb
) && !(dev
->features
& NETIF_F_FRAGLIST
)) ||
1933 (skb_shinfo(skb
)->nr_frags
&& (!(dev
->features
& NETIF_F_SG
) ||
1934 illegal_highdma(dev
, skb
))));
1937 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1938 struct netdev_queue
*txq
)
1940 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1941 int rc
= NETDEV_TX_OK
;
1943 if (likely(!skb
->next
)) {
1944 if (!list_empty(&ptype_all
))
1945 dev_queue_xmit_nit(skb
, dev
);
1948 * If device doesnt need skb->dst, release it right now while
1949 * its hot in this cpu cache
1951 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1954 skb_orphan_try(skb
);
1956 if (netif_needs_gso(dev
, skb
)) {
1957 if (unlikely(dev_gso_segment(skb
)))
1962 if (skb_needs_linearize(skb
, dev
) &&
1963 __skb_linearize(skb
))
1966 /* If packet is not checksummed and device does not
1967 * support checksumming for this protocol, complete
1968 * checksumming here.
1970 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1971 skb_set_transport_header(skb
, skb
->csum_start
-
1973 if (!dev_can_checksum(dev
, skb
) &&
1974 skb_checksum_help(skb
))
1979 rc
= ops
->ndo_start_xmit(skb
, dev
);
1980 if (rc
== NETDEV_TX_OK
)
1981 txq_trans_update(txq
);
1987 struct sk_buff
*nskb
= skb
->next
;
1989 skb
->next
= nskb
->next
;
1993 * If device doesnt need nskb->dst, release it right now while
1994 * its hot in this cpu cache
1996 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1999 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2000 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2001 if (rc
& ~NETDEV_TX_MASK
)
2002 goto out_kfree_gso_skb
;
2003 nskb
->next
= skb
->next
;
2007 txq_trans_update(txq
);
2008 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2009 return NETDEV_TX_BUSY
;
2010 } while (skb
->next
);
2013 if (likely(skb
->next
== NULL
))
2014 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2020 static u32 hashrnd __read_mostly
;
2022 u16
skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
)
2026 if (skb_rx_queue_recorded(skb
)) {
2027 hash
= skb_get_rx_queue(skb
);
2028 while (unlikely(hash
>= dev
->real_num_tx_queues
))
2029 hash
-= dev
->real_num_tx_queues
;
2033 if (skb
->sk
&& skb
->sk
->sk_hash
)
2034 hash
= skb
->sk
->sk_hash
;
2036 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2037 hash
= jhash_1word(hash
, hashrnd
);
2039 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
2041 EXPORT_SYMBOL(skb_tx_hash
);
2043 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2045 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2046 if (net_ratelimit()) {
2047 pr_warning("%s selects TX queue %d, but "
2048 "real number of TX queues is %d\n",
2049 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2056 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2057 struct sk_buff
*skb
)
2060 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2062 if (ops
->ndo_select_queue
) {
2063 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2064 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2066 struct sock
*sk
= skb
->sk
;
2067 queue_index
= sk_tx_queue_get(sk
);
2068 if (queue_index
< 0) {
2071 if (dev
->real_num_tx_queues
> 1)
2072 queue_index
= skb_tx_hash(dev
, skb
);
2075 struct dst_entry
*dst
= rcu_dereference_check(sk
->sk_dst_cache
, 1);
2077 if (dst
&& skb_dst(skb
) == dst
)
2078 sk_tx_queue_set(sk
, queue_index
);
2083 skb_set_queue_mapping(skb
, queue_index
);
2084 return netdev_get_tx_queue(dev
, queue_index
);
2087 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2088 struct net_device
*dev
,
2089 struct netdev_queue
*txq
)
2091 spinlock_t
*root_lock
= qdisc_lock(q
);
2092 bool contended
= qdisc_is_running(q
);
2096 * Heuristic to force contended enqueues to serialize on a
2097 * separate lock before trying to get qdisc main lock.
2098 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2099 * and dequeue packets faster.
2101 if (unlikely(contended
))
2102 spin_lock(&q
->busylock
);
2104 spin_lock(root_lock
);
2105 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2108 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2109 qdisc_run_begin(q
)) {
2111 * This is a work-conserving queue; there are no old skbs
2112 * waiting to be sent out; and the qdisc is not running -
2113 * xmit the skb directly.
2115 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2117 __qdisc_update_bstats(q
, skb
->len
);
2118 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2119 if (unlikely(contended
)) {
2120 spin_unlock(&q
->busylock
);
2127 rc
= NET_XMIT_SUCCESS
;
2130 rc
= qdisc_enqueue_root(skb
, q
);
2131 if (qdisc_run_begin(q
)) {
2132 if (unlikely(contended
)) {
2133 spin_unlock(&q
->busylock
);
2139 spin_unlock(root_lock
);
2140 if (unlikely(contended
))
2141 spin_unlock(&q
->busylock
);
2146 * dev_queue_xmit - transmit a buffer
2147 * @skb: buffer to transmit
2149 * Queue a buffer for transmission to a network device. The caller must
2150 * have set the device and priority and built the buffer before calling
2151 * this function. The function can be called from an interrupt.
2153 * A negative errno code is returned on a failure. A success does not
2154 * guarantee the frame will be transmitted as it may be dropped due
2155 * to congestion or traffic shaping.
2157 * -----------------------------------------------------------------------------------
2158 * I notice this method can also return errors from the queue disciplines,
2159 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2162 * Regardless of the return value, the skb is consumed, so it is currently
2163 * difficult to retry a send to this method. (You can bump the ref count
2164 * before sending to hold a reference for retry if you are careful.)
2166 * When calling this method, interrupts MUST be enabled. This is because
2167 * the BH enable code must have IRQs enabled so that it will not deadlock.
2170 int dev_queue_xmit(struct sk_buff
*skb
)
2172 struct net_device
*dev
= skb
->dev
;
2173 struct netdev_queue
*txq
;
2177 /* Disable soft irqs for various locks below. Also
2178 * stops preemption for RCU.
2182 txq
= dev_pick_tx(dev
, skb
);
2183 q
= rcu_dereference_bh(txq
->qdisc
);
2185 #ifdef CONFIG_NET_CLS_ACT
2186 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2189 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2193 /* The device has no queue. Common case for software devices:
2194 loopback, all the sorts of tunnels...
2196 Really, it is unlikely that netif_tx_lock protection is necessary
2197 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2199 However, it is possible, that they rely on protection
2202 Check this and shot the lock. It is not prone from deadlocks.
2203 Either shot noqueue qdisc, it is even simpler 8)
2205 if (dev
->flags
& IFF_UP
) {
2206 int cpu
= smp_processor_id(); /* ok because BHs are off */
2208 if (txq
->xmit_lock_owner
!= cpu
) {
2210 HARD_TX_LOCK(dev
, txq
, cpu
);
2212 if (!netif_tx_queue_stopped(txq
)) {
2213 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2214 if (dev_xmit_complete(rc
)) {
2215 HARD_TX_UNLOCK(dev
, txq
);
2219 HARD_TX_UNLOCK(dev
, txq
);
2220 if (net_ratelimit())
2221 printk(KERN_CRIT
"Virtual device %s asks to "
2222 "queue packet!\n", dev
->name
);
2224 /* Recursion is detected! It is possible,
2226 if (net_ratelimit())
2227 printk(KERN_CRIT
"Dead loop on virtual device "
2228 "%s, fix it urgently!\n", dev
->name
);
2233 rcu_read_unlock_bh();
2238 rcu_read_unlock_bh();
2241 EXPORT_SYMBOL(dev_queue_xmit
);
2244 /*=======================================================================
2246 =======================================================================*/
2248 int netdev_max_backlog __read_mostly
= 1000;
2249 int netdev_tstamp_prequeue __read_mostly
= 1;
2250 int netdev_budget __read_mostly
= 300;
2251 int weight_p __read_mostly
= 64; /* old backlog weight */
2253 /* Called with irq disabled */
2254 static inline void ____napi_schedule(struct softnet_data
*sd
,
2255 struct napi_struct
*napi
)
2257 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2258 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2262 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2263 * and src/dst port numbers. Returns a non-zero hash number on success
2266 __u32
__skb_get_rxhash(struct sk_buff
*skb
)
2268 int nhoff
, hash
= 0, poff
;
2269 struct ipv6hdr
*ip6
;
2272 u32 addr1
, addr2
, ihl
;
2278 nhoff
= skb_network_offset(skb
);
2280 switch (skb
->protocol
) {
2281 case __constant_htons(ETH_P_IP
):
2282 if (!pskb_may_pull(skb
, sizeof(*ip
) + nhoff
))
2285 ip
= (struct iphdr
*) (skb
->data
+ nhoff
);
2286 if (ip
->frag_off
& htons(IP_MF
| IP_OFFSET
))
2289 ip_proto
= ip
->protocol
;
2290 addr1
= (__force u32
) ip
->saddr
;
2291 addr2
= (__force u32
) ip
->daddr
;
2294 case __constant_htons(ETH_P_IPV6
):
2295 if (!pskb_may_pull(skb
, sizeof(*ip6
) + nhoff
))
2298 ip6
= (struct ipv6hdr
*) (skb
->data
+ nhoff
);
2299 ip_proto
= ip6
->nexthdr
;
2300 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2301 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2309 poff
= proto_ports_offset(ip_proto
);
2311 nhoff
+= ihl
* 4 + poff
;
2312 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2313 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2314 if (ports
.v16
[1] < ports
.v16
[0])
2315 swap(ports
.v16
[0], ports
.v16
[1]);
2319 /* get a consistent hash (same value on both flow directions) */
2323 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2330 EXPORT_SYMBOL(__skb_get_rxhash
);
2334 /* One global table that all flow-based protocols share. */
2335 struct rps_sock_flow_table
*rps_sock_flow_table __read_mostly
;
2336 EXPORT_SYMBOL(rps_sock_flow_table
);
2339 * get_rps_cpu is called from netif_receive_skb and returns the target
2340 * CPU from the RPS map of the receiving queue for a given skb.
2341 * rcu_read_lock must be held on entry.
2343 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2344 struct rps_dev_flow
**rflowp
)
2346 struct netdev_rx_queue
*rxqueue
;
2347 struct rps_map
*map
= NULL
;
2348 struct rps_dev_flow_table
*flow_table
;
2349 struct rps_sock_flow_table
*sock_flow_table
;
2353 if (skb_rx_queue_recorded(skb
)) {
2354 u16 index
= skb_get_rx_queue(skb
);
2355 if (unlikely(index
>= dev
->num_rx_queues
)) {
2356 WARN_ONCE(dev
->num_rx_queues
> 1, "%s received packet "
2357 "on queue %u, but number of RX queues is %u\n",
2358 dev
->name
, index
, dev
->num_rx_queues
);
2361 rxqueue
= dev
->_rx
+ index
;
2365 if (rxqueue
->rps_map
) {
2366 map
= rcu_dereference(rxqueue
->rps_map
);
2367 if (map
&& map
->len
== 1) {
2368 tcpu
= map
->cpus
[0];
2369 if (cpu_online(tcpu
))
2373 } else if (!rxqueue
->rps_flow_table
) {
2377 skb_reset_network_header(skb
);
2378 if (!skb_get_rxhash(skb
))
2381 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2382 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2383 if (flow_table
&& sock_flow_table
) {
2385 struct rps_dev_flow
*rflow
;
2387 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2390 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2391 sock_flow_table
->mask
];
2394 * If the desired CPU (where last recvmsg was done) is
2395 * different from current CPU (one in the rx-queue flow
2396 * table entry), switch if one of the following holds:
2397 * - Current CPU is unset (equal to RPS_NO_CPU).
2398 * - Current CPU is offline.
2399 * - The current CPU's queue tail has advanced beyond the
2400 * last packet that was enqueued using this table entry.
2401 * This guarantees that all previous packets for the flow
2402 * have been dequeued, thus preserving in order delivery.
2404 if (unlikely(tcpu
!= next_cpu
) &&
2405 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2406 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2407 rflow
->last_qtail
)) >= 0)) {
2408 tcpu
= rflow
->cpu
= next_cpu
;
2409 if (tcpu
!= RPS_NO_CPU
)
2410 rflow
->last_qtail
= per_cpu(softnet_data
,
2411 tcpu
).input_queue_head
;
2413 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2421 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2423 if (cpu_online(tcpu
)) {
2433 /* Called from hardirq (IPI) context */
2434 static void rps_trigger_softirq(void *data
)
2436 struct softnet_data
*sd
= data
;
2438 ____napi_schedule(sd
, &sd
->backlog
);
2442 #endif /* CONFIG_RPS */
2445 * Check if this softnet_data structure is another cpu one
2446 * If yes, queue it to our IPI list and return 1
2449 static int rps_ipi_queued(struct softnet_data
*sd
)
2452 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2455 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2456 mysd
->rps_ipi_list
= sd
;
2458 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2461 #endif /* CONFIG_RPS */
2466 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2467 * queue (may be a remote CPU queue).
2469 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2470 unsigned int *qtail
)
2472 struct softnet_data
*sd
;
2473 unsigned long flags
;
2475 sd
= &per_cpu(softnet_data
, cpu
);
2477 local_irq_save(flags
);
2480 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2481 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2483 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2484 input_queue_tail_incr_save(sd
, qtail
);
2486 local_irq_restore(flags
);
2487 return NET_RX_SUCCESS
;
2490 /* Schedule NAPI for backlog device
2491 * We can use non atomic operation since we own the queue lock
2493 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2494 if (!rps_ipi_queued(sd
))
2495 ____napi_schedule(sd
, &sd
->backlog
);
2503 local_irq_restore(flags
);
2510 * netif_rx - post buffer to the network code
2511 * @skb: buffer to post
2513 * This function receives a packet from a device driver and queues it for
2514 * the upper (protocol) levels to process. It always succeeds. The buffer
2515 * may be dropped during processing for congestion control or by the
2519 * NET_RX_SUCCESS (no congestion)
2520 * NET_RX_DROP (packet was dropped)
2524 int netif_rx(struct sk_buff
*skb
)
2528 /* if netpoll wants it, pretend we never saw it */
2529 if (netpoll_rx(skb
))
2532 if (netdev_tstamp_prequeue
)
2533 net_timestamp_check(skb
);
2537 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2543 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2545 cpu
= smp_processor_id();
2547 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2555 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2561 EXPORT_SYMBOL(netif_rx
);
2563 int netif_rx_ni(struct sk_buff
*skb
)
2568 err
= netif_rx(skb
);
2569 if (local_softirq_pending())
2575 EXPORT_SYMBOL(netif_rx_ni
);
2577 static void net_tx_action(struct softirq_action
*h
)
2579 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2581 if (sd
->completion_queue
) {
2582 struct sk_buff
*clist
;
2584 local_irq_disable();
2585 clist
= sd
->completion_queue
;
2586 sd
->completion_queue
= NULL
;
2590 struct sk_buff
*skb
= clist
;
2591 clist
= clist
->next
;
2593 WARN_ON(atomic_read(&skb
->users
));
2598 if (sd
->output_queue
) {
2601 local_irq_disable();
2602 head
= sd
->output_queue
;
2603 sd
->output_queue
= NULL
;
2604 sd
->output_queue_tailp
= &sd
->output_queue
;
2608 struct Qdisc
*q
= head
;
2609 spinlock_t
*root_lock
;
2611 head
= head
->next_sched
;
2613 root_lock
= qdisc_lock(q
);
2614 if (spin_trylock(root_lock
)) {
2615 smp_mb__before_clear_bit();
2616 clear_bit(__QDISC_STATE_SCHED
,
2619 spin_unlock(root_lock
);
2621 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2623 __netif_reschedule(q
);
2625 smp_mb__before_clear_bit();
2626 clear_bit(__QDISC_STATE_SCHED
,
2634 static inline int deliver_skb(struct sk_buff
*skb
,
2635 struct packet_type
*pt_prev
,
2636 struct net_device
*orig_dev
)
2638 atomic_inc(&skb
->users
);
2639 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2642 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2643 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2644 /* This hook is defined here for ATM LANE */
2645 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2646 unsigned char *addr
) __read_mostly
;
2647 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2650 #ifdef CONFIG_NET_CLS_ACT
2651 /* TODO: Maybe we should just force sch_ingress to be compiled in
2652 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2653 * a compare and 2 stores extra right now if we dont have it on
2654 * but have CONFIG_NET_CLS_ACT
2655 * NOTE: This doesnt stop any functionality; if you dont have
2656 * the ingress scheduler, you just cant add policies on ingress.
2659 static int ing_filter(struct sk_buff
*skb
)
2661 struct net_device
*dev
= skb
->dev
;
2662 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2663 struct netdev_queue
*rxq
;
2664 int result
= TC_ACT_OK
;
2667 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
2668 if (net_ratelimit())
2669 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2670 skb
->skb_iif
, dev
->ifindex
);
2674 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2675 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2677 rxq
= &dev
->rx_queue
;
2680 if (q
!= &noop_qdisc
) {
2681 spin_lock(qdisc_lock(q
));
2682 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2683 result
= qdisc_enqueue_root(skb
, q
);
2684 spin_unlock(qdisc_lock(q
));
2690 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2691 struct packet_type
**pt_prev
,
2692 int *ret
, struct net_device
*orig_dev
)
2694 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2698 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2702 switch (ing_filter(skb
)) {
2716 * netif_nit_deliver - deliver received packets to network taps
2719 * This function is used to deliver incoming packets to network
2720 * taps. It should be used when the normal netif_receive_skb path
2721 * is bypassed, for example because of VLAN acceleration.
2723 void netif_nit_deliver(struct sk_buff
*skb
)
2725 struct packet_type
*ptype
;
2727 if (list_empty(&ptype_all
))
2730 skb_reset_network_header(skb
);
2731 skb_reset_transport_header(skb
);
2732 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2735 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2736 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2737 deliver_skb(skb
, ptype
, skb
->dev
);
2743 * netdev_rx_handler_register - register receive handler
2744 * @dev: device to register a handler for
2745 * @rx_handler: receive handler to register
2746 * @rx_handler_data: data pointer that is used by rx handler
2748 * Register a receive hander for a device. This handler will then be
2749 * called from __netif_receive_skb. A negative errno code is returned
2752 * The caller must hold the rtnl_mutex.
2754 int netdev_rx_handler_register(struct net_device
*dev
,
2755 rx_handler_func_t
*rx_handler
,
2756 void *rx_handler_data
)
2760 if (dev
->rx_handler
)
2763 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
2764 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
2768 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
2771 * netdev_rx_handler_unregister - unregister receive handler
2772 * @dev: device to unregister a handler from
2774 * Unregister a receive hander from a device.
2776 * The caller must hold the rtnl_mutex.
2778 void netdev_rx_handler_unregister(struct net_device
*dev
)
2782 rcu_assign_pointer(dev
->rx_handler
, NULL
);
2783 rcu_assign_pointer(dev
->rx_handler_data
, NULL
);
2785 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
2787 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2788 struct net_device
*master
)
2790 if (skb
->pkt_type
== PACKET_HOST
) {
2791 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2793 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2797 /* On bonding slaves other than the currently active slave, suppress
2798 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2799 * ARP on active-backup slaves with arp_validate enabled.
2801 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2803 struct net_device
*dev
= skb
->dev
;
2805 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2806 dev
->last_rx
= jiffies
;
2808 if ((master
->priv_flags
& IFF_MASTER_ALB
) &&
2809 (master
->priv_flags
& IFF_BRIDGE_PORT
)) {
2810 /* Do address unmangle. The local destination address
2811 * will be always the one master has. Provides the right
2812 * functionality in a bridge.
2814 skb_bond_set_mac_by_master(skb
, master
);
2817 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
2818 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
2819 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
2822 if (master
->priv_flags
& IFF_MASTER_ALB
) {
2823 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
2824 skb
->pkt_type
!= PACKET_MULTICAST
)
2827 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
2828 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
2835 EXPORT_SYMBOL(__skb_bond_should_drop
);
2837 static int __netif_receive_skb(struct sk_buff
*skb
)
2839 struct packet_type
*ptype
, *pt_prev
;
2840 rx_handler_func_t
*rx_handler
;
2841 struct net_device
*orig_dev
;
2842 struct net_device
*master
;
2843 struct net_device
*null_or_orig
;
2844 struct net_device
*orig_or_bond
;
2845 int ret
= NET_RX_DROP
;
2848 if (!netdev_tstamp_prequeue
)
2849 net_timestamp_check(skb
);
2851 if (vlan_tx_tag_present(skb
))
2852 vlan_hwaccel_do_receive(skb
);
2854 /* if we've gotten here through NAPI, check netpoll */
2855 if (netpoll_receive_skb(skb
))
2859 skb
->skb_iif
= skb
->dev
->ifindex
;
2862 * bonding note: skbs received on inactive slaves should only
2863 * be delivered to pkt handlers that are exact matches. Also
2864 * the deliver_no_wcard flag will be set. If packet handlers
2865 * are sensitive to duplicate packets these skbs will need to
2866 * be dropped at the handler. The vlan accel path may have
2867 * already set the deliver_no_wcard flag.
2869 null_or_orig
= NULL
;
2870 orig_dev
= skb
->dev
;
2871 master
= ACCESS_ONCE(orig_dev
->master
);
2872 if (skb
->deliver_no_wcard
)
2873 null_or_orig
= orig_dev
;
2875 if (skb_bond_should_drop(skb
, master
)) {
2876 skb
->deliver_no_wcard
= 1;
2877 null_or_orig
= orig_dev
; /* deliver only exact match */
2882 __this_cpu_inc(softnet_data
.processed
);
2883 skb_reset_network_header(skb
);
2884 skb_reset_transport_header(skb
);
2885 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2891 #ifdef CONFIG_NET_CLS_ACT
2892 if (skb
->tc_verd
& TC_NCLS
) {
2893 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2898 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2899 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2900 ptype
->dev
== orig_dev
) {
2902 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2907 #ifdef CONFIG_NET_CLS_ACT
2908 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2914 /* Handle special case of bridge or macvlan */
2915 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
2918 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2921 skb
= rx_handler(skb
);
2927 * Make sure frames received on VLAN interfaces stacked on
2928 * bonding interfaces still make their way to any base bonding
2929 * device that may have registered for a specific ptype. The
2930 * handler may have to adjust skb->dev and orig_dev.
2932 orig_or_bond
= orig_dev
;
2933 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
2934 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
2935 orig_or_bond
= vlan_dev_real_dev(skb
->dev
);
2938 type
= skb
->protocol
;
2939 list_for_each_entry_rcu(ptype
,
2940 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2941 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
2942 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
2943 ptype
->dev
== orig_or_bond
)) {
2945 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2951 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2954 /* Jamal, now you will not able to escape explaining
2955 * me how you were going to use this. :-)
2966 * netif_receive_skb - process receive buffer from network
2967 * @skb: buffer to process
2969 * netif_receive_skb() is the main receive data processing function.
2970 * It always succeeds. The buffer may be dropped during processing
2971 * for congestion control or by the protocol layers.
2973 * This function may only be called from softirq context and interrupts
2974 * should be enabled.
2976 * Return values (usually ignored):
2977 * NET_RX_SUCCESS: no congestion
2978 * NET_RX_DROP: packet was dropped
2980 int netif_receive_skb(struct sk_buff
*skb
)
2982 if (netdev_tstamp_prequeue
)
2983 net_timestamp_check(skb
);
2985 if (skb_defer_rx_timestamp(skb
))
2986 return NET_RX_SUCCESS
;
2990 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2995 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2998 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3002 ret
= __netif_receive_skb(skb
);
3008 return __netif_receive_skb(skb
);
3011 EXPORT_SYMBOL(netif_receive_skb
);
3013 /* Network device is going away, flush any packets still pending
3014 * Called with irqs disabled.
3016 static void flush_backlog(void *arg
)
3018 struct net_device
*dev
= arg
;
3019 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3020 struct sk_buff
*skb
, *tmp
;
3023 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3024 if (skb
->dev
== dev
) {
3025 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3027 input_queue_head_incr(sd
);
3032 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3033 if (skb
->dev
== dev
) {
3034 __skb_unlink(skb
, &sd
->process_queue
);
3036 input_queue_head_incr(sd
);
3041 static int napi_gro_complete(struct sk_buff
*skb
)
3043 struct packet_type
*ptype
;
3044 __be16 type
= skb
->protocol
;
3045 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3048 if (NAPI_GRO_CB(skb
)->count
== 1) {
3049 skb_shinfo(skb
)->gso_size
= 0;
3054 list_for_each_entry_rcu(ptype
, head
, list
) {
3055 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3058 err
= ptype
->gro_complete(skb
);
3064 WARN_ON(&ptype
->list
== head
);
3066 return NET_RX_SUCCESS
;
3070 return netif_receive_skb(skb
);
3073 inline void napi_gro_flush(struct napi_struct
*napi
)
3075 struct sk_buff
*skb
, *next
;
3077 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3080 napi_gro_complete(skb
);
3083 napi
->gro_count
= 0;
3084 napi
->gro_list
= NULL
;
3086 EXPORT_SYMBOL(napi_gro_flush
);
3088 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3090 struct sk_buff
**pp
= NULL
;
3091 struct packet_type
*ptype
;
3092 __be16 type
= skb
->protocol
;
3093 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3096 enum gro_result ret
;
3098 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3101 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3105 list_for_each_entry_rcu(ptype
, head
, list
) {
3106 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3109 skb_set_network_header(skb
, skb_gro_offset(skb
));
3110 mac_len
= skb
->network_header
- skb
->mac_header
;
3111 skb
->mac_len
= mac_len
;
3112 NAPI_GRO_CB(skb
)->same_flow
= 0;
3113 NAPI_GRO_CB(skb
)->flush
= 0;
3114 NAPI_GRO_CB(skb
)->free
= 0;
3116 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3121 if (&ptype
->list
== head
)
3124 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3125 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3128 struct sk_buff
*nskb
= *pp
;
3132 napi_gro_complete(nskb
);
3139 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3143 NAPI_GRO_CB(skb
)->count
= 1;
3144 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3145 skb
->next
= napi
->gro_list
;
3146 napi
->gro_list
= skb
;
3150 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3151 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3153 BUG_ON(skb
->end
- skb
->tail
< grow
);
3155 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3158 skb
->data_len
-= grow
;
3160 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3161 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3163 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3164 put_page(skb_shinfo(skb
)->frags
[0].page
);
3165 memmove(skb_shinfo(skb
)->frags
,
3166 skb_shinfo(skb
)->frags
+ 1,
3167 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3178 EXPORT_SYMBOL(dev_gro_receive
);
3180 static inline gro_result_t
3181 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3185 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3186 unsigned long diffs
;
3188 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3189 diffs
|= compare_ether_header(skb_mac_header(p
),
3190 skb_gro_mac_header(skb
));
3191 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3192 NAPI_GRO_CB(p
)->flush
= 0;
3195 return dev_gro_receive(napi
, skb
);
3198 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3202 if (netif_receive_skb(skb
))
3207 case GRO_MERGED_FREE
:
3218 EXPORT_SYMBOL(napi_skb_finish
);
3220 void skb_gro_reset_offset(struct sk_buff
*skb
)
3222 NAPI_GRO_CB(skb
)->data_offset
= 0;
3223 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3224 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3226 if (skb
->mac_header
== skb
->tail
&&
3227 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3228 NAPI_GRO_CB(skb
)->frag0
=
3229 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3230 skb_shinfo(skb
)->frags
[0].page_offset
;
3231 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3234 EXPORT_SYMBOL(skb_gro_reset_offset
);
3236 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3238 skb_gro_reset_offset(skb
);
3240 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3242 EXPORT_SYMBOL(napi_gro_receive
);
3244 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3246 __skb_pull(skb
, skb_headlen(skb
));
3247 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3251 EXPORT_SYMBOL(napi_reuse_skb
);
3253 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3255 struct sk_buff
*skb
= napi
->skb
;
3258 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3264 EXPORT_SYMBOL(napi_get_frags
);
3266 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3272 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3274 if (ret
== GRO_HELD
)
3275 skb_gro_pull(skb
, -ETH_HLEN
);
3276 else if (netif_receive_skb(skb
))
3281 case GRO_MERGED_FREE
:
3282 napi_reuse_skb(napi
, skb
);
3291 EXPORT_SYMBOL(napi_frags_finish
);
3293 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3295 struct sk_buff
*skb
= napi
->skb
;
3302 skb_reset_mac_header(skb
);
3303 skb_gro_reset_offset(skb
);
3305 off
= skb_gro_offset(skb
);
3306 hlen
= off
+ sizeof(*eth
);
3307 eth
= skb_gro_header_fast(skb
, off
);
3308 if (skb_gro_header_hard(skb
, hlen
)) {
3309 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3310 if (unlikely(!eth
)) {
3311 napi_reuse_skb(napi
, skb
);
3317 skb_gro_pull(skb
, sizeof(*eth
));
3320 * This works because the only protocols we care about don't require
3321 * special handling. We'll fix it up properly at the end.
3323 skb
->protocol
= eth
->h_proto
;
3328 EXPORT_SYMBOL(napi_frags_skb
);
3330 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3332 struct sk_buff
*skb
= napi_frags_skb(napi
);
3337 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3339 EXPORT_SYMBOL(napi_gro_frags
);
3342 * net_rps_action sends any pending IPI's for rps.
3343 * Note: called with local irq disabled, but exits with local irq enabled.
3345 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3348 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3351 sd
->rps_ipi_list
= NULL
;
3355 /* Send pending IPI's to kick RPS processing on remote cpus. */
3357 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3359 if (cpu_online(remsd
->cpu
))
3360 __smp_call_function_single(remsd
->cpu
,
3369 static int process_backlog(struct napi_struct
*napi
, int quota
)
3372 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3375 /* Check if we have pending ipi, its better to send them now,
3376 * not waiting net_rx_action() end.
3378 if (sd
->rps_ipi_list
) {
3379 local_irq_disable();
3380 net_rps_action_and_irq_enable(sd
);
3383 napi
->weight
= weight_p
;
3384 local_irq_disable();
3385 while (work
< quota
) {
3386 struct sk_buff
*skb
;
3389 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3391 __netif_receive_skb(skb
);
3392 local_irq_disable();
3393 input_queue_head_incr(sd
);
3394 if (++work
>= quota
) {
3401 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3403 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3404 &sd
->process_queue
);
3406 if (qlen
< quota
- work
) {
3408 * Inline a custom version of __napi_complete().
3409 * only current cpu owns and manipulates this napi,
3410 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3411 * we can use a plain write instead of clear_bit(),
3412 * and we dont need an smp_mb() memory barrier.
3414 list_del(&napi
->poll_list
);
3417 quota
= work
+ qlen
;
3427 * __napi_schedule - schedule for receive
3428 * @n: entry to schedule
3430 * The entry's receive function will be scheduled to run
3432 void __napi_schedule(struct napi_struct
*n
)
3434 unsigned long flags
;
3436 local_irq_save(flags
);
3437 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3438 local_irq_restore(flags
);
3440 EXPORT_SYMBOL(__napi_schedule
);
3442 void __napi_complete(struct napi_struct
*n
)
3444 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3445 BUG_ON(n
->gro_list
);
3447 list_del(&n
->poll_list
);
3448 smp_mb__before_clear_bit();
3449 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3451 EXPORT_SYMBOL(__napi_complete
);
3453 void napi_complete(struct napi_struct
*n
)
3455 unsigned long flags
;
3458 * don't let napi dequeue from the cpu poll list
3459 * just in case its running on a different cpu
3461 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3465 local_irq_save(flags
);
3467 local_irq_restore(flags
);
3469 EXPORT_SYMBOL(napi_complete
);
3471 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3472 int (*poll
)(struct napi_struct
*, int), int weight
)
3474 INIT_LIST_HEAD(&napi
->poll_list
);
3475 napi
->gro_count
= 0;
3476 napi
->gro_list
= NULL
;
3479 napi
->weight
= weight
;
3480 list_add(&napi
->dev_list
, &dev
->napi_list
);
3482 #ifdef CONFIG_NETPOLL
3483 spin_lock_init(&napi
->poll_lock
);
3484 napi
->poll_owner
= -1;
3486 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3488 EXPORT_SYMBOL(netif_napi_add
);
3490 void netif_napi_del(struct napi_struct
*napi
)
3492 struct sk_buff
*skb
, *next
;
3494 list_del_init(&napi
->dev_list
);
3495 napi_free_frags(napi
);
3497 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3503 napi
->gro_list
= NULL
;
3504 napi
->gro_count
= 0;
3506 EXPORT_SYMBOL(netif_napi_del
);
3508 static void net_rx_action(struct softirq_action
*h
)
3510 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3511 unsigned long time_limit
= jiffies
+ 2;
3512 int budget
= netdev_budget
;
3515 local_irq_disable();
3517 while (!list_empty(&sd
->poll_list
)) {
3518 struct napi_struct
*n
;
3521 /* If softirq window is exhuasted then punt.
3522 * Allow this to run for 2 jiffies since which will allow
3523 * an average latency of 1.5/HZ.
3525 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3530 /* Even though interrupts have been re-enabled, this
3531 * access is safe because interrupts can only add new
3532 * entries to the tail of this list, and only ->poll()
3533 * calls can remove this head entry from the list.
3535 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3537 have
= netpoll_poll_lock(n
);
3541 /* This NAPI_STATE_SCHED test is for avoiding a race
3542 * with netpoll's poll_napi(). Only the entity which
3543 * obtains the lock and sees NAPI_STATE_SCHED set will
3544 * actually make the ->poll() call. Therefore we avoid
3545 * accidently calling ->poll() when NAPI is not scheduled.
3548 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3549 work
= n
->poll(n
, weight
);
3553 WARN_ON_ONCE(work
> weight
);
3557 local_irq_disable();
3559 /* Drivers must not modify the NAPI state if they
3560 * consume the entire weight. In such cases this code
3561 * still "owns" the NAPI instance and therefore can
3562 * move the instance around on the list at-will.
3564 if (unlikely(work
== weight
)) {
3565 if (unlikely(napi_disable_pending(n
))) {
3568 local_irq_disable();
3570 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3573 netpoll_poll_unlock(have
);
3576 net_rps_action_and_irq_enable(sd
);
3578 #ifdef CONFIG_NET_DMA
3580 * There may not be any more sk_buffs coming right now, so push
3581 * any pending DMA copies to hardware
3583 dma_issue_pending_all();
3590 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3594 static gifconf_func_t
*gifconf_list
[NPROTO
];
3597 * register_gifconf - register a SIOCGIF handler
3598 * @family: Address family
3599 * @gifconf: Function handler
3601 * Register protocol dependent address dumping routines. The handler
3602 * that is passed must not be freed or reused until it has been replaced
3603 * by another handler.
3605 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3607 if (family
>= NPROTO
)
3609 gifconf_list
[family
] = gifconf
;
3612 EXPORT_SYMBOL(register_gifconf
);
3616 * Map an interface index to its name (SIOCGIFNAME)
3620 * We need this ioctl for efficient implementation of the
3621 * if_indextoname() function required by the IPv6 API. Without
3622 * it, we would have to search all the interfaces to find a
3626 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3628 struct net_device
*dev
;
3632 * Fetch the caller's info block.
3635 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3639 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3645 strcpy(ifr
.ifr_name
, dev
->name
);
3648 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3654 * Perform a SIOCGIFCONF call. This structure will change
3655 * size eventually, and there is nothing I can do about it.
3656 * Thus we will need a 'compatibility mode'.
3659 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3662 struct net_device
*dev
;
3669 * Fetch the caller's info block.
3672 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3679 * Loop over the interfaces, and write an info block for each.
3683 for_each_netdev(net
, dev
) {
3684 for (i
= 0; i
< NPROTO
; i
++) {
3685 if (gifconf_list
[i
]) {
3688 done
= gifconf_list
[i
](dev
, NULL
, 0);
3690 done
= gifconf_list
[i
](dev
, pos
+ total
,
3700 * All done. Write the updated control block back to the caller.
3702 ifc
.ifc_len
= total
;
3705 * Both BSD and Solaris return 0 here, so we do too.
3707 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3710 #ifdef CONFIG_PROC_FS
3712 * This is invoked by the /proc filesystem handler to display a device
3715 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3718 struct net
*net
= seq_file_net(seq
);
3720 struct net_device
*dev
;
3724 return SEQ_START_TOKEN
;
3727 for_each_netdev_rcu(net
, dev
)
3734 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3736 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3737 first_net_device(seq_file_net(seq
)) :
3738 next_net_device((struct net_device
*)v
);
3741 return rcu_dereference(dev
);
3744 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3750 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3752 struct rtnl_link_stats64 temp
;
3753 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
3755 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3756 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3757 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3759 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3760 stats
->rx_fifo_errors
,
3761 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3762 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3763 stats
->rx_compressed
, stats
->multicast
,
3764 stats
->tx_bytes
, stats
->tx_packets
,
3765 stats
->tx_errors
, stats
->tx_dropped
,
3766 stats
->tx_fifo_errors
, stats
->collisions
,
3767 stats
->tx_carrier_errors
+
3768 stats
->tx_aborted_errors
+
3769 stats
->tx_window_errors
+
3770 stats
->tx_heartbeat_errors
,
3771 stats
->tx_compressed
);
3775 * Called from the PROCfs module. This now uses the new arbitrary sized
3776 * /proc/net interface to create /proc/net/dev
3778 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3780 if (v
== SEQ_START_TOKEN
)
3781 seq_puts(seq
, "Inter-| Receive "
3783 " face |bytes packets errs drop fifo frame "
3784 "compressed multicast|bytes packets errs "
3785 "drop fifo colls carrier compressed\n");
3787 dev_seq_printf_stats(seq
, v
);
3791 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
3793 struct softnet_data
*sd
= NULL
;
3795 while (*pos
< nr_cpu_ids
)
3796 if (cpu_online(*pos
)) {
3797 sd
= &per_cpu(softnet_data
, *pos
);
3804 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3806 return softnet_get_online(pos
);
3809 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3812 return softnet_get_online(pos
);
3815 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3819 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3821 struct softnet_data
*sd
= v
;
3823 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3824 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
3825 0, 0, 0, 0, /* was fastroute */
3826 sd
->cpu_collision
, sd
->received_rps
);
3830 static const struct seq_operations dev_seq_ops
= {
3831 .start
= dev_seq_start
,
3832 .next
= dev_seq_next
,
3833 .stop
= dev_seq_stop
,
3834 .show
= dev_seq_show
,
3837 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3839 return seq_open_net(inode
, file
, &dev_seq_ops
,
3840 sizeof(struct seq_net_private
));
3843 static const struct file_operations dev_seq_fops
= {
3844 .owner
= THIS_MODULE
,
3845 .open
= dev_seq_open
,
3847 .llseek
= seq_lseek
,
3848 .release
= seq_release_net
,
3851 static const struct seq_operations softnet_seq_ops
= {
3852 .start
= softnet_seq_start
,
3853 .next
= softnet_seq_next
,
3854 .stop
= softnet_seq_stop
,
3855 .show
= softnet_seq_show
,
3858 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3860 return seq_open(file
, &softnet_seq_ops
);
3863 static const struct file_operations softnet_seq_fops
= {
3864 .owner
= THIS_MODULE
,
3865 .open
= softnet_seq_open
,
3867 .llseek
= seq_lseek
,
3868 .release
= seq_release
,
3871 static void *ptype_get_idx(loff_t pos
)
3873 struct packet_type
*pt
= NULL
;
3877 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3883 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3884 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3893 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3897 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3900 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3902 struct packet_type
*pt
;
3903 struct list_head
*nxt
;
3907 if (v
== SEQ_START_TOKEN
)
3908 return ptype_get_idx(0);
3911 nxt
= pt
->list
.next
;
3912 if (pt
->type
== htons(ETH_P_ALL
)) {
3913 if (nxt
!= &ptype_all
)
3916 nxt
= ptype_base
[0].next
;
3918 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3920 while (nxt
== &ptype_base
[hash
]) {
3921 if (++hash
>= PTYPE_HASH_SIZE
)
3923 nxt
= ptype_base
[hash
].next
;
3926 return list_entry(nxt
, struct packet_type
, list
);
3929 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3935 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3937 struct packet_type
*pt
= v
;
3939 if (v
== SEQ_START_TOKEN
)
3940 seq_puts(seq
, "Type Device Function\n");
3941 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3942 if (pt
->type
== htons(ETH_P_ALL
))
3943 seq_puts(seq
, "ALL ");
3945 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3947 seq_printf(seq
, " %-8s %pF\n",
3948 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
3954 static const struct seq_operations ptype_seq_ops
= {
3955 .start
= ptype_seq_start
,
3956 .next
= ptype_seq_next
,
3957 .stop
= ptype_seq_stop
,
3958 .show
= ptype_seq_show
,
3961 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
3963 return seq_open_net(inode
, file
, &ptype_seq_ops
,
3964 sizeof(struct seq_net_private
));
3967 static const struct file_operations ptype_seq_fops
= {
3968 .owner
= THIS_MODULE
,
3969 .open
= ptype_seq_open
,
3971 .llseek
= seq_lseek
,
3972 .release
= seq_release_net
,
3976 static int __net_init
dev_proc_net_init(struct net
*net
)
3980 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
3982 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
3984 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
3987 if (wext_proc_init(net
))
3993 proc_net_remove(net
, "ptype");
3995 proc_net_remove(net
, "softnet_stat");
3997 proc_net_remove(net
, "dev");
4001 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4003 wext_proc_exit(net
);
4005 proc_net_remove(net
, "ptype");
4006 proc_net_remove(net
, "softnet_stat");
4007 proc_net_remove(net
, "dev");
4010 static struct pernet_operations __net_initdata dev_proc_ops
= {
4011 .init
= dev_proc_net_init
,
4012 .exit
= dev_proc_net_exit
,
4015 static int __init
dev_proc_init(void)
4017 return register_pernet_subsys(&dev_proc_ops
);
4020 #define dev_proc_init() 0
4021 #endif /* CONFIG_PROC_FS */
4025 * netdev_set_master - set up master/slave pair
4026 * @slave: slave device
4027 * @master: new master device
4029 * Changes the master device of the slave. Pass %NULL to break the
4030 * bonding. The caller must hold the RTNL semaphore. On a failure
4031 * a negative errno code is returned. On success the reference counts
4032 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4033 * function returns zero.
4035 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4037 struct net_device
*old
= slave
->master
;
4047 slave
->master
= master
;
4054 slave
->flags
|= IFF_SLAVE
;
4056 slave
->flags
&= ~IFF_SLAVE
;
4058 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4061 EXPORT_SYMBOL(netdev_set_master
);
4063 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4065 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4067 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4068 ops
->ndo_change_rx_flags(dev
, flags
);
4071 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4073 unsigned short old_flags
= dev
->flags
;
4079 dev
->flags
|= IFF_PROMISC
;
4080 dev
->promiscuity
+= inc
;
4081 if (dev
->promiscuity
== 0) {
4084 * If inc causes overflow, untouch promisc and return error.
4087 dev
->flags
&= ~IFF_PROMISC
;
4089 dev
->promiscuity
-= inc
;
4090 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4091 "set promiscuity failed, promiscuity feature "
4092 "of device might be broken.\n", dev
->name
);
4096 if (dev
->flags
!= old_flags
) {
4097 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4098 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4100 if (audit_enabled
) {
4101 current_uid_gid(&uid
, &gid
);
4102 audit_log(current
->audit_context
, GFP_ATOMIC
,
4103 AUDIT_ANOM_PROMISCUOUS
,
4104 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4105 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4106 (old_flags
& IFF_PROMISC
),
4107 audit_get_loginuid(current
),
4109 audit_get_sessionid(current
));
4112 dev_change_rx_flags(dev
, IFF_PROMISC
);
4118 * dev_set_promiscuity - update promiscuity count on a device
4122 * Add or remove promiscuity from a device. While the count in the device
4123 * remains above zero the interface remains promiscuous. Once it hits zero
4124 * the device reverts back to normal filtering operation. A negative inc
4125 * value is used to drop promiscuity on the device.
4126 * Return 0 if successful or a negative errno code on error.
4128 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4130 unsigned short old_flags
= dev
->flags
;
4133 err
= __dev_set_promiscuity(dev
, inc
);
4136 if (dev
->flags
!= old_flags
)
4137 dev_set_rx_mode(dev
);
4140 EXPORT_SYMBOL(dev_set_promiscuity
);
4143 * dev_set_allmulti - update allmulti count on a device
4147 * Add or remove reception of all multicast frames to a device. While the
4148 * count in the device remains above zero the interface remains listening
4149 * to all interfaces. Once it hits zero the device reverts back to normal
4150 * filtering operation. A negative @inc value is used to drop the counter
4151 * when releasing a resource needing all multicasts.
4152 * Return 0 if successful or a negative errno code on error.
4155 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4157 unsigned short old_flags
= dev
->flags
;
4161 dev
->flags
|= IFF_ALLMULTI
;
4162 dev
->allmulti
+= inc
;
4163 if (dev
->allmulti
== 0) {
4166 * If inc causes overflow, untouch allmulti and return error.
4169 dev
->flags
&= ~IFF_ALLMULTI
;
4171 dev
->allmulti
-= inc
;
4172 printk(KERN_WARNING
"%s: allmulti touches roof, "
4173 "set allmulti failed, allmulti feature of "
4174 "device might be broken.\n", dev
->name
);
4178 if (dev
->flags
^ old_flags
) {
4179 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4180 dev_set_rx_mode(dev
);
4184 EXPORT_SYMBOL(dev_set_allmulti
);
4187 * Upload unicast and multicast address lists to device and
4188 * configure RX filtering. When the device doesn't support unicast
4189 * filtering it is put in promiscuous mode while unicast addresses
4192 void __dev_set_rx_mode(struct net_device
*dev
)
4194 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4196 /* dev_open will call this function so the list will stay sane. */
4197 if (!(dev
->flags
&IFF_UP
))
4200 if (!netif_device_present(dev
))
4203 if (ops
->ndo_set_rx_mode
)
4204 ops
->ndo_set_rx_mode(dev
);
4206 /* Unicast addresses changes may only happen under the rtnl,
4207 * therefore calling __dev_set_promiscuity here is safe.
4209 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4210 __dev_set_promiscuity(dev
, 1);
4211 dev
->uc_promisc
= 1;
4212 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4213 __dev_set_promiscuity(dev
, -1);
4214 dev
->uc_promisc
= 0;
4217 if (ops
->ndo_set_multicast_list
)
4218 ops
->ndo_set_multicast_list(dev
);
4222 void dev_set_rx_mode(struct net_device
*dev
)
4224 netif_addr_lock_bh(dev
);
4225 __dev_set_rx_mode(dev
);
4226 netif_addr_unlock_bh(dev
);
4230 * dev_get_flags - get flags reported to userspace
4233 * Get the combination of flag bits exported through APIs to userspace.
4235 unsigned dev_get_flags(const struct net_device
*dev
)
4239 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4244 (dev
->gflags
& (IFF_PROMISC
|
4247 if (netif_running(dev
)) {
4248 if (netif_oper_up(dev
))
4249 flags
|= IFF_RUNNING
;
4250 if (netif_carrier_ok(dev
))
4251 flags
|= IFF_LOWER_UP
;
4252 if (netif_dormant(dev
))
4253 flags
|= IFF_DORMANT
;
4258 EXPORT_SYMBOL(dev_get_flags
);
4260 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4262 int old_flags
= dev
->flags
;
4268 * Set the flags on our device.
4271 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4272 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4274 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4278 * Load in the correct multicast list now the flags have changed.
4281 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4282 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4284 dev_set_rx_mode(dev
);
4287 * Have we downed the interface. We handle IFF_UP ourselves
4288 * according to user attempts to set it, rather than blindly
4293 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4294 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4297 dev_set_rx_mode(dev
);
4300 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4301 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4303 dev
->gflags
^= IFF_PROMISC
;
4304 dev_set_promiscuity(dev
, inc
);
4307 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4308 is important. Some (broken) drivers set IFF_PROMISC, when
4309 IFF_ALLMULTI is requested not asking us and not reporting.
4311 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4312 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4314 dev
->gflags
^= IFF_ALLMULTI
;
4315 dev_set_allmulti(dev
, inc
);
4321 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4323 unsigned int changes
= dev
->flags
^ old_flags
;
4325 if (changes
& IFF_UP
) {
4326 if (dev
->flags
& IFF_UP
)
4327 call_netdevice_notifiers(NETDEV_UP
, dev
);
4329 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4332 if (dev
->flags
& IFF_UP
&&
4333 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4334 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4338 * dev_change_flags - change device settings
4340 * @flags: device state flags
4342 * Change settings on device based state flags. The flags are
4343 * in the userspace exported format.
4345 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4348 int old_flags
= dev
->flags
;
4350 ret
= __dev_change_flags(dev
, flags
);
4354 changes
= old_flags
^ dev
->flags
;
4356 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4358 __dev_notify_flags(dev
, old_flags
);
4361 EXPORT_SYMBOL(dev_change_flags
);
4364 * dev_set_mtu - Change maximum transfer unit
4366 * @new_mtu: new transfer unit
4368 * Change the maximum transfer size of the network device.
4370 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4372 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4375 if (new_mtu
== dev
->mtu
)
4378 /* MTU must be positive. */
4382 if (!netif_device_present(dev
))
4386 if (ops
->ndo_change_mtu
)
4387 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4391 if (!err
&& dev
->flags
& IFF_UP
)
4392 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4395 EXPORT_SYMBOL(dev_set_mtu
);
4398 * dev_set_mac_address - Change Media Access Control Address
4402 * Change the hardware (MAC) address of the device
4404 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4406 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4409 if (!ops
->ndo_set_mac_address
)
4411 if (sa
->sa_family
!= dev
->type
)
4413 if (!netif_device_present(dev
))
4415 err
= ops
->ndo_set_mac_address(dev
, sa
);
4417 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4420 EXPORT_SYMBOL(dev_set_mac_address
);
4423 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4425 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4428 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4434 case SIOCGIFFLAGS
: /* Get interface flags */
4435 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4438 case SIOCGIFMETRIC
: /* Get the metric on the interface
4439 (currently unused) */
4440 ifr
->ifr_metric
= 0;
4443 case SIOCGIFMTU
: /* Get the MTU of a device */
4444 ifr
->ifr_mtu
= dev
->mtu
;
4449 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4451 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4452 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4453 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4461 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4462 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4463 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4464 ifr
->ifr_map
.irq
= dev
->irq
;
4465 ifr
->ifr_map
.dma
= dev
->dma
;
4466 ifr
->ifr_map
.port
= dev
->if_port
;
4470 ifr
->ifr_ifindex
= dev
->ifindex
;
4474 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4478 /* dev_ioctl() should ensure this case
4490 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4492 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4495 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4496 const struct net_device_ops
*ops
;
4501 ops
= dev
->netdev_ops
;
4504 case SIOCSIFFLAGS
: /* Set interface flags */
4505 return dev_change_flags(dev
, ifr
->ifr_flags
);
4507 case SIOCSIFMETRIC
: /* Set the metric on the interface
4508 (currently unused) */
4511 case SIOCSIFMTU
: /* Set the MTU of a device */
4512 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4515 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4517 case SIOCSIFHWBROADCAST
:
4518 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4520 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4521 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4522 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4526 if (ops
->ndo_set_config
) {
4527 if (!netif_device_present(dev
))
4529 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4534 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4535 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4537 if (!netif_device_present(dev
))
4539 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4542 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4543 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4545 if (!netif_device_present(dev
))
4547 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4550 if (ifr
->ifr_qlen
< 0)
4552 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4556 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4557 return dev_change_name(dev
, ifr
->ifr_newname
);
4560 * Unknown or private ioctl
4563 if ((cmd
>= SIOCDEVPRIVATE
&&
4564 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4565 cmd
== SIOCBONDENSLAVE
||
4566 cmd
== SIOCBONDRELEASE
||
4567 cmd
== SIOCBONDSETHWADDR
||
4568 cmd
== SIOCBONDSLAVEINFOQUERY
||
4569 cmd
== SIOCBONDINFOQUERY
||
4570 cmd
== SIOCBONDCHANGEACTIVE
||
4571 cmd
== SIOCGMIIPHY
||
4572 cmd
== SIOCGMIIREG
||
4573 cmd
== SIOCSMIIREG
||
4574 cmd
== SIOCBRADDIF
||
4575 cmd
== SIOCBRDELIF
||
4576 cmd
== SIOCSHWTSTAMP
||
4577 cmd
== SIOCWANDEV
) {
4579 if (ops
->ndo_do_ioctl
) {
4580 if (netif_device_present(dev
))
4581 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4593 * This function handles all "interface"-type I/O control requests. The actual
4594 * 'doing' part of this is dev_ifsioc above.
4598 * dev_ioctl - network device ioctl
4599 * @net: the applicable net namespace
4600 * @cmd: command to issue
4601 * @arg: pointer to a struct ifreq in user space
4603 * Issue ioctl functions to devices. This is normally called by the
4604 * user space syscall interfaces but can sometimes be useful for
4605 * other purposes. The return value is the return from the syscall if
4606 * positive or a negative errno code on error.
4609 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4615 /* One special case: SIOCGIFCONF takes ifconf argument
4616 and requires shared lock, because it sleeps writing
4620 if (cmd
== SIOCGIFCONF
) {
4622 ret
= dev_ifconf(net
, (char __user
*) arg
);
4626 if (cmd
== SIOCGIFNAME
)
4627 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4629 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4632 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4634 colon
= strchr(ifr
.ifr_name
, ':');
4639 * See which interface the caller is talking about.
4644 * These ioctl calls:
4645 * - can be done by all.
4646 * - atomic and do not require locking.
4657 dev_load(net
, ifr
.ifr_name
);
4659 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4664 if (copy_to_user(arg
, &ifr
,
4665 sizeof(struct ifreq
)))
4671 dev_load(net
, ifr
.ifr_name
);
4673 ret
= dev_ethtool(net
, &ifr
);
4678 if (copy_to_user(arg
, &ifr
,
4679 sizeof(struct ifreq
)))
4685 * These ioctl calls:
4686 * - require superuser power.
4687 * - require strict serialization.
4693 if (!capable(CAP_NET_ADMIN
))
4695 dev_load(net
, ifr
.ifr_name
);
4697 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4702 if (copy_to_user(arg
, &ifr
,
4703 sizeof(struct ifreq
)))
4709 * These ioctl calls:
4710 * - require superuser power.
4711 * - require strict serialization.
4712 * - do not return a value
4722 case SIOCSIFHWBROADCAST
:
4725 case SIOCBONDENSLAVE
:
4726 case SIOCBONDRELEASE
:
4727 case SIOCBONDSETHWADDR
:
4728 case SIOCBONDCHANGEACTIVE
:
4732 if (!capable(CAP_NET_ADMIN
))
4735 case SIOCBONDSLAVEINFOQUERY
:
4736 case SIOCBONDINFOQUERY
:
4737 dev_load(net
, ifr
.ifr_name
);
4739 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4744 /* Get the per device memory space. We can add this but
4745 * currently do not support it */
4747 /* Set the per device memory buffer space.
4748 * Not applicable in our case */
4753 * Unknown or private ioctl.
4756 if (cmd
== SIOCWANDEV
||
4757 (cmd
>= SIOCDEVPRIVATE
&&
4758 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4759 dev_load(net
, ifr
.ifr_name
);
4761 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4763 if (!ret
&& copy_to_user(arg
, &ifr
,
4764 sizeof(struct ifreq
)))
4768 /* Take care of Wireless Extensions */
4769 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4770 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4777 * dev_new_index - allocate an ifindex
4778 * @net: the applicable net namespace
4780 * Returns a suitable unique value for a new device interface
4781 * number. The caller must hold the rtnl semaphore or the
4782 * dev_base_lock to be sure it remains unique.
4784 static int dev_new_index(struct net
*net
)
4790 if (!__dev_get_by_index(net
, ifindex
))
4795 /* Delayed registration/unregisteration */
4796 static LIST_HEAD(net_todo_list
);
4798 static void net_set_todo(struct net_device
*dev
)
4800 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4803 static void rollback_registered_many(struct list_head
*head
)
4805 struct net_device
*dev
, *tmp
;
4807 BUG_ON(dev_boot_phase
);
4810 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4811 /* Some devices call without registering
4812 * for initialization unwind. Remove those
4813 * devices and proceed with the remaining.
4815 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4816 pr_debug("unregister_netdevice: device %s/%p never "
4817 "was registered\n", dev
->name
, dev
);
4820 list_del(&dev
->unreg_list
);
4824 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4826 /* If device is running, close it first. */
4829 /* And unlink it from device chain. */
4830 unlist_netdevice(dev
);
4832 dev
->reg_state
= NETREG_UNREGISTERING
;
4837 list_for_each_entry(dev
, head
, unreg_list
) {
4838 /* Shutdown queueing discipline. */
4842 /* Notify protocols, that we are about to destroy
4843 this device. They should clean all the things.
4845 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4847 if (!dev
->rtnl_link_ops
||
4848 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
4849 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
4852 * Flush the unicast and multicast chains
4857 if (dev
->netdev_ops
->ndo_uninit
)
4858 dev
->netdev_ops
->ndo_uninit(dev
);
4860 /* Notifier chain MUST detach us from master device. */
4861 WARN_ON(dev
->master
);
4863 /* Remove entries from kobject tree */
4864 netdev_unregister_kobject(dev
);
4867 /* Process any work delayed until the end of the batch */
4868 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
4869 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
4873 list_for_each_entry(dev
, head
, unreg_list
)
4877 static void rollback_registered(struct net_device
*dev
)
4881 list_add(&dev
->unreg_list
, &single
);
4882 rollback_registered_many(&single
);
4885 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
4886 struct netdev_queue
*dev_queue
,
4889 spin_lock_init(&dev_queue
->_xmit_lock
);
4890 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
4891 dev_queue
->xmit_lock_owner
= -1;
4894 static void netdev_init_queue_locks(struct net_device
*dev
)
4896 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
4897 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
4900 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
4902 /* Fix illegal SG+CSUM combinations. */
4903 if ((features
& NETIF_F_SG
) &&
4904 !(features
& NETIF_F_ALL_CSUM
)) {
4906 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
4907 "checksum feature.\n", name
);
4908 features
&= ~NETIF_F_SG
;
4911 /* TSO requires that SG is present as well. */
4912 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
4914 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
4915 "SG feature.\n", name
);
4916 features
&= ~NETIF_F_TSO
;
4919 if (features
& NETIF_F_UFO
) {
4920 if (!(features
& NETIF_F_GEN_CSUM
)) {
4922 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4923 "since no NETIF_F_HW_CSUM feature.\n",
4925 features
&= ~NETIF_F_UFO
;
4928 if (!(features
& NETIF_F_SG
)) {
4930 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4931 "since no NETIF_F_SG feature.\n", name
);
4932 features
&= ~NETIF_F_UFO
;
4938 EXPORT_SYMBOL(netdev_fix_features
);
4941 * netif_stacked_transfer_operstate - transfer operstate
4942 * @rootdev: the root or lower level device to transfer state from
4943 * @dev: the device to transfer operstate to
4945 * Transfer operational state from root to device. This is normally
4946 * called when a stacking relationship exists between the root
4947 * device and the device(a leaf device).
4949 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
4950 struct net_device
*dev
)
4952 if (rootdev
->operstate
== IF_OPER_DORMANT
)
4953 netif_dormant_on(dev
);
4955 netif_dormant_off(dev
);
4957 if (netif_carrier_ok(rootdev
)) {
4958 if (!netif_carrier_ok(dev
))
4959 netif_carrier_on(dev
);
4961 if (netif_carrier_ok(dev
))
4962 netif_carrier_off(dev
);
4965 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
4968 * register_netdevice - register a network device
4969 * @dev: device to register
4971 * Take a completed network device structure and add it to the kernel
4972 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4973 * chain. 0 is returned on success. A negative errno code is returned
4974 * on a failure to set up the device, or if the name is a duplicate.
4976 * Callers must hold the rtnl semaphore. You may want
4977 * register_netdev() instead of this.
4980 * The locking appears insufficient to guarantee two parallel registers
4981 * will not get the same name.
4984 int register_netdevice(struct net_device
*dev
)
4987 struct net
*net
= dev_net(dev
);
4989 BUG_ON(dev_boot_phase
);
4994 /* When net_device's are persistent, this will be fatal. */
4995 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
4998 spin_lock_init(&dev
->addr_list_lock
);
4999 netdev_set_addr_lockdep_class(dev
);
5000 netdev_init_queue_locks(dev
);
5005 if (!dev
->num_rx_queues
) {
5007 * Allocate a single RX queue if driver never called
5011 dev
->_rx
= kzalloc(sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5017 dev
->_rx
->first
= dev
->_rx
;
5018 atomic_set(&dev
->_rx
->count
, 1);
5019 dev
->num_rx_queues
= 1;
5022 /* Init, if this function is available */
5023 if (dev
->netdev_ops
->ndo_init
) {
5024 ret
= dev
->netdev_ops
->ndo_init(dev
);
5032 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
5036 dev
->ifindex
= dev_new_index(net
);
5037 if (dev
->iflink
== -1)
5038 dev
->iflink
= dev
->ifindex
;
5040 /* Fix illegal checksum combinations */
5041 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
5042 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5043 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5045 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5048 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5049 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5050 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5052 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5055 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5057 /* Enable software GSO if SG is supported. */
5058 if (dev
->features
& NETIF_F_SG
)
5059 dev
->features
|= NETIF_F_GSO
;
5061 /* Enable GRO for vlans by default if dev->features has GRO also.
5062 * vlan_dev_init() will do the dev->features check.
5064 dev
->vlan_features
|= NETIF_F_GRO
;
5066 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5067 ret
= notifier_to_errno(ret
);
5071 ret
= netdev_register_kobject(dev
);
5074 dev
->reg_state
= NETREG_REGISTERED
;
5077 * Default initial state at registry is that the
5078 * device is present.
5081 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5083 dev_init_scheduler(dev
);
5085 list_netdevice(dev
);
5087 /* Notify protocols, that a new device appeared. */
5088 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5089 ret
= notifier_to_errno(ret
);
5091 rollback_registered(dev
);
5092 dev
->reg_state
= NETREG_UNREGISTERED
;
5095 * Prevent userspace races by waiting until the network
5096 * device is fully setup before sending notifications.
5098 if (!dev
->rtnl_link_ops
||
5099 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5100 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5106 if (dev
->netdev_ops
->ndo_uninit
)
5107 dev
->netdev_ops
->ndo_uninit(dev
);
5110 EXPORT_SYMBOL(register_netdevice
);
5113 * init_dummy_netdev - init a dummy network device for NAPI
5114 * @dev: device to init
5116 * This takes a network device structure and initialize the minimum
5117 * amount of fields so it can be used to schedule NAPI polls without
5118 * registering a full blown interface. This is to be used by drivers
5119 * that need to tie several hardware interfaces to a single NAPI
5120 * poll scheduler due to HW limitations.
5122 int init_dummy_netdev(struct net_device
*dev
)
5124 /* Clear everything. Note we don't initialize spinlocks
5125 * are they aren't supposed to be taken by any of the
5126 * NAPI code and this dummy netdev is supposed to be
5127 * only ever used for NAPI polls
5129 memset(dev
, 0, sizeof(struct net_device
));
5131 /* make sure we BUG if trying to hit standard
5132 * register/unregister code path
5134 dev
->reg_state
= NETREG_DUMMY
;
5136 /* initialize the ref count */
5137 atomic_set(&dev
->refcnt
, 1);
5139 /* NAPI wants this */
5140 INIT_LIST_HEAD(&dev
->napi_list
);
5142 /* a dummy interface is started by default */
5143 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5144 set_bit(__LINK_STATE_START
, &dev
->state
);
5148 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5152 * register_netdev - register a network device
5153 * @dev: device to register
5155 * Take a completed network device structure and add it to the kernel
5156 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5157 * chain. 0 is returned on success. A negative errno code is returned
5158 * on a failure to set up the device, or if the name is a duplicate.
5160 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5161 * and expands the device name if you passed a format string to
5164 int register_netdev(struct net_device
*dev
)
5171 * If the name is a format string the caller wants us to do a
5174 if (strchr(dev
->name
, '%')) {
5175 err
= dev_alloc_name(dev
, dev
->name
);
5180 err
= register_netdevice(dev
);
5185 EXPORT_SYMBOL(register_netdev
);
5188 * netdev_wait_allrefs - wait until all references are gone.
5190 * This is called when unregistering network devices.
5192 * Any protocol or device that holds a reference should register
5193 * for netdevice notification, and cleanup and put back the
5194 * reference if they receive an UNREGISTER event.
5195 * We can get stuck here if buggy protocols don't correctly
5198 static void netdev_wait_allrefs(struct net_device
*dev
)
5200 unsigned long rebroadcast_time
, warning_time
;
5202 linkwatch_forget_dev(dev
);
5204 rebroadcast_time
= warning_time
= jiffies
;
5205 while (atomic_read(&dev
->refcnt
) != 0) {
5206 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5209 /* Rebroadcast unregister notification */
5210 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5211 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5212 * should have already handle it the first time */
5214 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5216 /* We must not have linkwatch events
5217 * pending on unregister. If this
5218 * happens, we simply run the queue
5219 * unscheduled, resulting in a noop
5222 linkwatch_run_queue();
5227 rebroadcast_time
= jiffies
;
5232 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5233 printk(KERN_EMERG
"unregister_netdevice: "
5234 "waiting for %s to become free. Usage "
5236 dev
->name
, atomic_read(&dev
->refcnt
));
5237 warning_time
= jiffies
;
5246 * register_netdevice(x1);
5247 * register_netdevice(x2);
5249 * unregister_netdevice(y1);
5250 * unregister_netdevice(y2);
5256 * We are invoked by rtnl_unlock().
5257 * This allows us to deal with problems:
5258 * 1) We can delete sysfs objects which invoke hotplug
5259 * without deadlocking with linkwatch via keventd.
5260 * 2) Since we run with the RTNL semaphore not held, we can sleep
5261 * safely in order to wait for the netdev refcnt to drop to zero.
5263 * We must not return until all unregister events added during
5264 * the interval the lock was held have been completed.
5266 void netdev_run_todo(void)
5268 struct list_head list
;
5270 /* Snapshot list, allow later requests */
5271 list_replace_init(&net_todo_list
, &list
);
5275 while (!list_empty(&list
)) {
5276 struct net_device
*dev
5277 = list_first_entry(&list
, struct net_device
, todo_list
);
5278 list_del(&dev
->todo_list
);
5280 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5281 printk(KERN_ERR
"network todo '%s' but state %d\n",
5282 dev
->name
, dev
->reg_state
);
5287 dev
->reg_state
= NETREG_UNREGISTERED
;
5289 on_each_cpu(flush_backlog
, dev
, 1);
5291 netdev_wait_allrefs(dev
);
5294 BUG_ON(atomic_read(&dev
->refcnt
));
5295 WARN_ON(rcu_dereference_raw(dev
->ip_ptr
));
5296 WARN_ON(dev
->ip6_ptr
);
5297 WARN_ON(dev
->dn_ptr
);
5299 if (dev
->destructor
)
5300 dev
->destructor(dev
);
5302 /* Free network device */
5303 kobject_put(&dev
->dev
.kobj
);
5308 * dev_txq_stats_fold - fold tx_queues stats
5309 * @dev: device to get statistics from
5310 * @stats: struct rtnl_link_stats64 to hold results
5312 void dev_txq_stats_fold(const struct net_device
*dev
,
5313 struct rtnl_link_stats64
*stats
)
5315 u64 tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5317 struct netdev_queue
*txq
;
5319 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5320 txq
= netdev_get_tx_queue(dev
, i
);
5321 spin_lock_bh(&txq
->_xmit_lock
);
5322 tx_bytes
+= txq
->tx_bytes
;
5323 tx_packets
+= txq
->tx_packets
;
5324 tx_dropped
+= txq
->tx_dropped
;
5325 spin_unlock_bh(&txq
->_xmit_lock
);
5327 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5328 stats
->tx_bytes
= tx_bytes
;
5329 stats
->tx_packets
= tx_packets
;
5330 stats
->tx_dropped
= tx_dropped
;
5333 EXPORT_SYMBOL(dev_txq_stats_fold
);
5335 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5336 * fields in the same order, with only the type differing.
5338 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5339 const struct net_device_stats
*netdev_stats
)
5341 #if BITS_PER_LONG == 64
5342 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5343 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5345 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5346 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5347 u64
*dst
= (u64
*)stats64
;
5349 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5350 sizeof(*stats64
) / sizeof(u64
));
5351 for (i
= 0; i
< n
; i
++)
5357 * dev_get_stats - get network device statistics
5358 * @dev: device to get statistics from
5359 * @storage: place to store stats
5361 * Get network statistics from device. Return @storage.
5362 * The device driver may provide its own method by setting
5363 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5364 * otherwise the internal statistics structure is used.
5366 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5367 struct rtnl_link_stats64
*storage
)
5369 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5371 if (ops
->ndo_get_stats64
) {
5372 memset(storage
, 0, sizeof(*storage
));
5373 return ops
->ndo_get_stats64(dev
, storage
);
5375 if (ops
->ndo_get_stats
) {
5376 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5379 netdev_stats_to_stats64(storage
, &dev
->stats
);
5380 dev_txq_stats_fold(dev
, storage
);
5383 EXPORT_SYMBOL(dev_get_stats
);
5385 static void netdev_init_one_queue(struct net_device
*dev
,
5386 struct netdev_queue
*queue
,
5392 static void netdev_init_queues(struct net_device
*dev
)
5394 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
5395 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5396 spin_lock_init(&dev
->tx_global_lock
);
5400 * alloc_netdev_mq - allocate network device
5401 * @sizeof_priv: size of private data to allocate space for
5402 * @name: device name format string
5403 * @setup: callback to initialize device
5404 * @queue_count: the number of subqueues to allocate
5406 * Allocates a struct net_device with private data area for driver use
5407 * and performs basic initialization. Also allocates subquue structs
5408 * for each queue on the device at the end of the netdevice.
5410 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
5411 void (*setup
)(struct net_device
*), unsigned int queue_count
)
5413 struct netdev_queue
*tx
;
5414 struct net_device
*dev
;
5416 struct net_device
*p
;
5418 struct netdev_rx_queue
*rx
;
5422 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5424 alloc_size
= sizeof(struct net_device
);
5426 /* ensure 32-byte alignment of private area */
5427 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5428 alloc_size
+= sizeof_priv
;
5430 /* ensure 32-byte alignment of whole construct */
5431 alloc_size
+= NETDEV_ALIGN
- 1;
5433 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5435 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5439 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5441 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5447 rx
= kcalloc(queue_count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5449 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5454 atomic_set(&rx
->count
, queue_count
);
5457 * Set a pointer to first element in the array which holds the
5460 for (i
= 0; i
< queue_count
; i
++)
5464 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5465 dev
->padded
= (char *)dev
- (char *)p
;
5467 if (dev_addr_init(dev
))
5473 dev_net_set(dev
, &init_net
);
5476 dev
->num_tx_queues
= queue_count
;
5477 dev
->real_num_tx_queues
= queue_count
;
5481 dev
->num_rx_queues
= queue_count
;
5484 dev
->gso_max_size
= GSO_MAX_SIZE
;
5486 netdev_init_queues(dev
);
5488 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5489 dev
->ethtool_ntuple_list
.count
= 0;
5490 INIT_LIST_HEAD(&dev
->napi_list
);
5491 INIT_LIST_HEAD(&dev
->unreg_list
);
5492 INIT_LIST_HEAD(&dev
->link_watch_list
);
5493 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5495 strcpy(dev
->name
, name
);
5508 EXPORT_SYMBOL(alloc_netdev_mq
);
5511 * free_netdev - free network device
5514 * This function does the last stage of destroying an allocated device
5515 * interface. The reference to the device object is released.
5516 * If this is the last reference then it will be freed.
5518 void free_netdev(struct net_device
*dev
)
5520 struct napi_struct
*p
, *n
;
5522 release_net(dev_net(dev
));
5526 /* Flush device addresses */
5527 dev_addr_flush(dev
);
5529 /* Clear ethtool n-tuple list */
5530 ethtool_ntuple_flush(dev
);
5532 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5535 /* Compatibility with error handling in drivers */
5536 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5537 kfree((char *)dev
- dev
->padded
);
5541 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5542 dev
->reg_state
= NETREG_RELEASED
;
5544 /* will free via device release */
5545 put_device(&dev
->dev
);
5547 EXPORT_SYMBOL(free_netdev
);
5550 * synchronize_net - Synchronize with packet receive processing
5552 * Wait for packets currently being received to be done.
5553 * Does not block later packets from starting.
5555 void synchronize_net(void)
5560 EXPORT_SYMBOL(synchronize_net
);
5563 * unregister_netdevice_queue - remove device from the kernel
5567 * This function shuts down a device interface and removes it
5568 * from the kernel tables.
5569 * If head not NULL, device is queued to be unregistered later.
5571 * Callers must hold the rtnl semaphore. You may want
5572 * unregister_netdev() instead of this.
5575 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5580 list_move_tail(&dev
->unreg_list
, head
);
5582 rollback_registered(dev
);
5583 /* Finish processing unregister after unlock */
5587 EXPORT_SYMBOL(unregister_netdevice_queue
);
5590 * unregister_netdevice_many - unregister many devices
5591 * @head: list of devices
5593 void unregister_netdevice_many(struct list_head
*head
)
5595 struct net_device
*dev
;
5597 if (!list_empty(head
)) {
5598 rollback_registered_many(head
);
5599 list_for_each_entry(dev
, head
, unreg_list
)
5603 EXPORT_SYMBOL(unregister_netdevice_many
);
5606 * unregister_netdev - remove device from the kernel
5609 * This function shuts down a device interface and removes it
5610 * from the kernel tables.
5612 * This is just a wrapper for unregister_netdevice that takes
5613 * the rtnl semaphore. In general you want to use this and not
5614 * unregister_netdevice.
5616 void unregister_netdev(struct net_device
*dev
)
5619 unregister_netdevice(dev
);
5622 EXPORT_SYMBOL(unregister_netdev
);
5625 * dev_change_net_namespace - move device to different nethost namespace
5627 * @net: network namespace
5628 * @pat: If not NULL name pattern to try if the current device name
5629 * is already taken in the destination network namespace.
5631 * This function shuts down a device interface and moves it
5632 * to a new network namespace. On success 0 is returned, on
5633 * a failure a netagive errno code is returned.
5635 * Callers must hold the rtnl semaphore.
5638 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5644 /* Don't allow namespace local devices to be moved. */
5646 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5649 /* Ensure the device has been registrered */
5651 if (dev
->reg_state
!= NETREG_REGISTERED
)
5654 /* Get out if there is nothing todo */
5656 if (net_eq(dev_net(dev
), net
))
5659 /* Pick the destination device name, and ensure
5660 * we can use it in the destination network namespace.
5663 if (__dev_get_by_name(net
, dev
->name
)) {
5664 /* We get here if we can't use the current device name */
5667 if (dev_get_valid_name(dev
, pat
, 1))
5672 * And now a mini version of register_netdevice unregister_netdevice.
5675 /* If device is running close it first. */
5678 /* And unlink it from device chain */
5680 unlist_netdevice(dev
);
5684 /* Shutdown queueing discipline. */
5687 /* Notify protocols, that we are about to destroy
5688 this device. They should clean all the things.
5690 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5691 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5694 * Flush the unicast and multicast chains
5699 /* Actually switch the network namespace */
5700 dev_net_set(dev
, net
);
5702 /* If there is an ifindex conflict assign a new one */
5703 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5704 int iflink
= (dev
->iflink
== dev
->ifindex
);
5705 dev
->ifindex
= dev_new_index(net
);
5707 dev
->iflink
= dev
->ifindex
;
5710 /* Fixup kobjects */
5711 err
= device_rename(&dev
->dev
, dev
->name
);
5714 /* Add the device back in the hashes */
5715 list_netdevice(dev
);
5717 /* Notify protocols, that a new device appeared. */
5718 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5721 * Prevent userspace races by waiting until the network
5722 * device is fully setup before sending notifications.
5724 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5731 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5733 static int dev_cpu_callback(struct notifier_block
*nfb
,
5734 unsigned long action
,
5737 struct sk_buff
**list_skb
;
5738 struct sk_buff
*skb
;
5739 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5740 struct softnet_data
*sd
, *oldsd
;
5742 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5745 local_irq_disable();
5746 cpu
= smp_processor_id();
5747 sd
= &per_cpu(softnet_data
, cpu
);
5748 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5750 /* Find end of our completion_queue. */
5751 list_skb
= &sd
->completion_queue
;
5753 list_skb
= &(*list_skb
)->next
;
5754 /* Append completion queue from offline CPU. */
5755 *list_skb
= oldsd
->completion_queue
;
5756 oldsd
->completion_queue
= NULL
;
5758 /* Append output queue from offline CPU. */
5759 if (oldsd
->output_queue
) {
5760 *sd
->output_queue_tailp
= oldsd
->output_queue
;
5761 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
5762 oldsd
->output_queue
= NULL
;
5763 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
5766 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5769 /* Process offline CPU's input_pkt_queue */
5770 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
5772 input_queue_head_incr(oldsd
);
5774 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
5776 input_queue_head_incr(oldsd
);
5784 * netdev_increment_features - increment feature set by one
5785 * @all: current feature set
5786 * @one: new feature set
5787 * @mask: mask feature set
5789 * Computes a new feature set after adding a device with feature set
5790 * @one to the master device with current feature set @all. Will not
5791 * enable anything that is off in @mask. Returns the new feature set.
5793 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
5796 /* If device needs checksumming, downgrade to it. */
5797 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
5798 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
5799 else if (mask
& NETIF_F_ALL_CSUM
) {
5800 /* If one device supports v4/v6 checksumming, set for all. */
5801 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
5802 !(all
& NETIF_F_GEN_CSUM
)) {
5803 all
&= ~NETIF_F_ALL_CSUM
;
5804 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
5807 /* If one device supports hw checksumming, set for all. */
5808 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
5809 all
&= ~NETIF_F_ALL_CSUM
;
5810 all
|= NETIF_F_HW_CSUM
;
5814 one
|= NETIF_F_ALL_CSUM
;
5816 one
|= all
& NETIF_F_ONE_FOR_ALL
;
5817 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
5818 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
5822 EXPORT_SYMBOL(netdev_increment_features
);
5824 static struct hlist_head
*netdev_create_hash(void)
5827 struct hlist_head
*hash
;
5829 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5831 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5832 INIT_HLIST_HEAD(&hash
[i
]);
5837 /* Initialize per network namespace state */
5838 static int __net_init
netdev_init(struct net
*net
)
5840 INIT_LIST_HEAD(&net
->dev_base_head
);
5842 net
->dev_name_head
= netdev_create_hash();
5843 if (net
->dev_name_head
== NULL
)
5846 net
->dev_index_head
= netdev_create_hash();
5847 if (net
->dev_index_head
== NULL
)
5853 kfree(net
->dev_name_head
);
5859 * netdev_drivername - network driver for the device
5860 * @dev: network device
5861 * @buffer: buffer for resulting name
5862 * @len: size of buffer
5864 * Determine network driver for device.
5866 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
5868 const struct device_driver
*driver
;
5869 const struct device
*parent
;
5871 if (len
<= 0 || !buffer
)
5875 parent
= dev
->dev
.parent
;
5880 driver
= parent
->driver
;
5881 if (driver
&& driver
->name
)
5882 strlcpy(buffer
, driver
->name
, len
);
5886 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
5887 struct va_format
*vaf
)
5891 if (dev
&& dev
->dev
.parent
)
5892 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
5893 netdev_name(dev
), vaf
);
5895 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
5897 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
5902 int netdev_printk(const char *level
, const struct net_device
*dev
,
5903 const char *format
, ...)
5905 struct va_format vaf
;
5909 va_start(args
, format
);
5914 r
= __netdev_printk(level
, dev
, &vaf
);
5919 EXPORT_SYMBOL(netdev_printk
);
5921 #define define_netdev_printk_level(func, level) \
5922 int func(const struct net_device *dev, const char *fmt, ...) \
5925 struct va_format vaf; \
5928 va_start(args, fmt); \
5933 r = __netdev_printk(level, dev, &vaf); \
5938 EXPORT_SYMBOL(func);
5940 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
5941 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
5942 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
5943 define_netdev_printk_level(netdev_err
, KERN_ERR
);
5944 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
5945 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
5946 define_netdev_printk_level(netdev_info
, KERN_INFO
);
5948 static void __net_exit
netdev_exit(struct net
*net
)
5950 kfree(net
->dev_name_head
);
5951 kfree(net
->dev_index_head
);
5954 static struct pernet_operations __net_initdata netdev_net_ops
= {
5955 .init
= netdev_init
,
5956 .exit
= netdev_exit
,
5959 static void __net_exit
default_device_exit(struct net
*net
)
5961 struct net_device
*dev
, *aux
;
5963 * Push all migratable network devices back to the
5964 * initial network namespace
5967 for_each_netdev_safe(net
, dev
, aux
) {
5969 char fb_name
[IFNAMSIZ
];
5971 /* Ignore unmoveable devices (i.e. loopback) */
5972 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5975 /* Leave virtual devices for the generic cleanup */
5976 if (dev
->rtnl_link_ops
)
5979 /* Push remaing network devices to init_net */
5980 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
5981 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
5983 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
5984 __func__
, dev
->name
, err
);
5991 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
5993 /* At exit all network devices most be removed from a network
5994 * namespace. Do this in the reverse order of registeration.
5995 * Do this across as many network namespaces as possible to
5996 * improve batching efficiency.
5998 struct net_device
*dev
;
6000 LIST_HEAD(dev_kill_list
);
6003 list_for_each_entry(net
, net_list
, exit_list
) {
6004 for_each_netdev_reverse(net
, dev
) {
6005 if (dev
->rtnl_link_ops
)
6006 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6008 unregister_netdevice_queue(dev
, &dev_kill_list
);
6011 unregister_netdevice_many(&dev_kill_list
);
6015 static struct pernet_operations __net_initdata default_device_ops
= {
6016 .exit
= default_device_exit
,
6017 .exit_batch
= default_device_exit_batch
,
6021 * Initialize the DEV module. At boot time this walks the device list and
6022 * unhooks any devices that fail to initialise (normally hardware not
6023 * present) and leaves us with a valid list of present and active devices.
6028 * This is called single threaded during boot, so no need
6029 * to take the rtnl semaphore.
6031 static int __init
net_dev_init(void)
6033 int i
, rc
= -ENOMEM
;
6035 BUG_ON(!dev_boot_phase
);
6037 if (dev_proc_init())
6040 if (netdev_kobject_init())
6043 INIT_LIST_HEAD(&ptype_all
);
6044 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6045 INIT_LIST_HEAD(&ptype_base
[i
]);
6047 if (register_pernet_subsys(&netdev_net_ops
))
6051 * Initialise the packet receive queues.
6054 for_each_possible_cpu(i
) {
6055 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6057 memset(sd
, 0, sizeof(*sd
));
6058 skb_queue_head_init(&sd
->input_pkt_queue
);
6059 skb_queue_head_init(&sd
->process_queue
);
6060 sd
->completion_queue
= NULL
;
6061 INIT_LIST_HEAD(&sd
->poll_list
);
6062 sd
->output_queue
= NULL
;
6063 sd
->output_queue_tailp
= &sd
->output_queue
;
6065 sd
->csd
.func
= rps_trigger_softirq
;
6071 sd
->backlog
.poll
= process_backlog
;
6072 sd
->backlog
.weight
= weight_p
;
6073 sd
->backlog
.gro_list
= NULL
;
6074 sd
->backlog
.gro_count
= 0;
6079 /* The loopback device is special if any other network devices
6080 * is present in a network namespace the loopback device must
6081 * be present. Since we now dynamically allocate and free the
6082 * loopback device ensure this invariant is maintained by
6083 * keeping the loopback device as the first device on the
6084 * list of network devices. Ensuring the loopback devices
6085 * is the first device that appears and the last network device
6088 if (register_pernet_device(&loopback_net_ops
))
6091 if (register_pernet_device(&default_device_ops
))
6094 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6095 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6097 hotcpu_notifier(dev_cpu_callback
, 0);
6105 subsys_initcall(net_dev_init
);
6107 static int __init
initialize_hashrnd(void)
6109 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6113 late_initcall_sync(initialize_hashrnd
);