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 (unlikely(!(dev
->flags
& IFF_UP
) ||
1487 (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
)))) {
1488 atomic_long_inc(&dev
->rx_dropped
);
1492 skb_set_dev(skb
, dev
);
1493 skb
->tstamp
.tv64
= 0;
1494 skb
->pkt_type
= PACKET_HOST
;
1495 skb
->protocol
= eth_type_trans(skb
, dev
);
1496 return netif_rx(skb
);
1498 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1501 * Support routine. Sends outgoing frames to any network
1502 * taps currently in use.
1505 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1507 struct packet_type
*ptype
;
1509 #ifdef CONFIG_NET_CLS_ACT
1510 if (!(skb
->tstamp
.tv64
&& (G_TC_FROM(skb
->tc_verd
) & AT_INGRESS
)))
1511 net_timestamp_set(skb
);
1513 net_timestamp_set(skb
);
1517 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1518 /* Never send packets back to the socket
1519 * they originated from - MvS (miquels@drinkel.ow.org)
1521 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1522 (ptype
->af_packet_priv
== NULL
||
1523 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1524 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1528 /* skb->nh should be correctly
1529 set by sender, so that the second statement is
1530 just protection against buggy protocols.
1532 skb_reset_mac_header(skb2
);
1534 if (skb_network_header(skb2
) < skb2
->data
||
1535 skb2
->network_header
> skb2
->tail
) {
1536 if (net_ratelimit())
1537 printk(KERN_CRIT
"protocol %04x is "
1539 ntohs(skb2
->protocol
),
1541 skb_reset_network_header(skb2
);
1544 skb2
->transport_header
= skb2
->network_header
;
1545 skb2
->pkt_type
= PACKET_OUTGOING
;
1546 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1553 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1554 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1556 void netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1558 unsigned int real_num
= dev
->real_num_tx_queues
;
1560 if (unlikely(txq
> dev
->num_tx_queues
))
1562 else if (txq
> real_num
)
1563 dev
->real_num_tx_queues
= txq
;
1564 else if (txq
< real_num
) {
1565 dev
->real_num_tx_queues
= txq
;
1566 qdisc_reset_all_tx_gt(dev
, txq
);
1569 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1573 * netif_set_real_num_rx_queues - set actual number of RX queues used
1574 * @dev: Network device
1575 * @rxq: Actual number of RX queues
1577 * This must be called either with the rtnl_lock held or before
1578 * registration of the net device. Returns 0 on success, or a
1579 * negative error code. If called before registration, it always
1582 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1586 if (dev
->reg_state
== NETREG_REGISTERED
) {
1589 if (rxq
> dev
->num_rx_queues
)
1592 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1598 dev
->real_num_rx_queues
= rxq
;
1601 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1604 static inline void __netif_reschedule(struct Qdisc
*q
)
1606 struct softnet_data
*sd
;
1607 unsigned long flags
;
1609 local_irq_save(flags
);
1610 sd
= &__get_cpu_var(softnet_data
);
1611 q
->next_sched
= NULL
;
1612 *sd
->output_queue_tailp
= q
;
1613 sd
->output_queue_tailp
= &q
->next_sched
;
1614 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1615 local_irq_restore(flags
);
1618 void __netif_schedule(struct Qdisc
*q
)
1620 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1621 __netif_reschedule(q
);
1623 EXPORT_SYMBOL(__netif_schedule
);
1625 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1627 if (atomic_dec_and_test(&skb
->users
)) {
1628 struct softnet_data
*sd
;
1629 unsigned long flags
;
1631 local_irq_save(flags
);
1632 sd
= &__get_cpu_var(softnet_data
);
1633 skb
->next
= sd
->completion_queue
;
1634 sd
->completion_queue
= skb
;
1635 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1636 local_irq_restore(flags
);
1639 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1641 void dev_kfree_skb_any(struct sk_buff
*skb
)
1643 if (in_irq() || irqs_disabled())
1644 dev_kfree_skb_irq(skb
);
1648 EXPORT_SYMBOL(dev_kfree_skb_any
);
1652 * netif_device_detach - mark device as removed
1653 * @dev: network device
1655 * Mark device as removed from system and therefore no longer available.
1657 void netif_device_detach(struct net_device
*dev
)
1659 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1660 netif_running(dev
)) {
1661 netif_tx_stop_all_queues(dev
);
1664 EXPORT_SYMBOL(netif_device_detach
);
1667 * netif_device_attach - mark device as attached
1668 * @dev: network device
1670 * Mark device as attached from system and restart if needed.
1672 void netif_device_attach(struct net_device
*dev
)
1674 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1675 netif_running(dev
)) {
1676 netif_tx_wake_all_queues(dev
);
1677 __netdev_watchdog_up(dev
);
1680 EXPORT_SYMBOL(netif_device_attach
);
1682 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1684 return ((features
& NETIF_F_GEN_CSUM
) ||
1685 ((features
& NETIF_F_IP_CSUM
) &&
1686 protocol
== htons(ETH_P_IP
)) ||
1687 ((features
& NETIF_F_IPV6_CSUM
) &&
1688 protocol
== htons(ETH_P_IPV6
)) ||
1689 ((features
& NETIF_F_FCOE_CRC
) &&
1690 protocol
== htons(ETH_P_FCOE
)));
1693 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1695 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1698 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1699 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1700 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1701 veh
->h_vlan_encapsulated_proto
))
1709 * skb_dev_set -- assign a new device to a buffer
1710 * @skb: buffer for the new device
1711 * @dev: network device
1713 * If an skb is owned by a device already, we have to reset
1714 * all data private to the namespace a device belongs to
1715 * before assigning it a new device.
1717 #ifdef CONFIG_NET_NS
1718 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1721 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1724 skb_init_secmark(skb
);
1728 skb
->ipvs_property
= 0;
1729 #ifdef CONFIG_NET_SCHED
1735 EXPORT_SYMBOL(skb_set_dev
);
1736 #endif /* CONFIG_NET_NS */
1739 * Invalidate hardware checksum when packet is to be mangled, and
1740 * complete checksum manually on outgoing path.
1742 int skb_checksum_help(struct sk_buff
*skb
)
1745 int ret
= 0, offset
;
1747 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1748 goto out_set_summed
;
1750 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1751 /* Let GSO fix up the checksum. */
1752 goto out_set_summed
;
1755 offset
= skb
->csum_start
- skb_headroom(skb
);
1756 BUG_ON(offset
>= skb_headlen(skb
));
1757 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1759 offset
+= skb
->csum_offset
;
1760 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1762 if (skb_cloned(skb
) &&
1763 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1764 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1769 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1771 skb
->ip_summed
= CHECKSUM_NONE
;
1775 EXPORT_SYMBOL(skb_checksum_help
);
1778 * skb_gso_segment - Perform segmentation on skb.
1779 * @skb: buffer to segment
1780 * @features: features for the output path (see dev->features)
1782 * This function segments the given skb and returns a list of segments.
1784 * It may return NULL if the skb requires no segmentation. This is
1785 * only possible when GSO is used for verifying header integrity.
1787 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1789 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1790 struct packet_type
*ptype
;
1791 __be16 type
= skb
->protocol
;
1794 skb_reset_mac_header(skb
);
1795 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1796 __skb_pull(skb
, skb
->mac_len
);
1798 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1799 struct net_device
*dev
= skb
->dev
;
1800 struct ethtool_drvinfo info
= {};
1802 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1803 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1805 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1807 info
.driver
, dev
? dev
->features
: 0L,
1808 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1809 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1811 if (skb_header_cloned(skb
) &&
1812 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1813 return ERR_PTR(err
);
1817 list_for_each_entry_rcu(ptype
,
1818 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1819 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1820 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1821 err
= ptype
->gso_send_check(skb
);
1822 segs
= ERR_PTR(err
);
1823 if (err
|| skb_gso_ok(skb
, features
))
1825 __skb_push(skb
, (skb
->data
-
1826 skb_network_header(skb
)));
1828 segs
= ptype
->gso_segment(skb
, features
);
1834 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1838 EXPORT_SYMBOL(skb_gso_segment
);
1840 /* Take action when hardware reception checksum errors are detected. */
1842 void netdev_rx_csum_fault(struct net_device
*dev
)
1844 if (net_ratelimit()) {
1845 printk(KERN_ERR
"%s: hw csum failure.\n",
1846 dev
? dev
->name
: "<unknown>");
1850 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1853 /* Actually, we should eliminate this check as soon as we know, that:
1854 * 1. IOMMU is present and allows to map all the memory.
1855 * 2. No high memory really exists on this machine.
1858 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1860 #ifdef CONFIG_HIGHMEM
1862 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1863 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1864 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1868 if (PCI_DMA_BUS_IS_PHYS
) {
1869 struct device
*pdev
= dev
->dev
.parent
;
1873 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1874 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1875 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1884 void (*destructor
)(struct sk_buff
*skb
);
1887 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1889 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1891 struct dev_gso_cb
*cb
;
1894 struct sk_buff
*nskb
= skb
->next
;
1896 skb
->next
= nskb
->next
;
1899 } while (skb
->next
);
1901 cb
= DEV_GSO_CB(skb
);
1903 cb
->destructor(skb
);
1907 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1908 * @skb: buffer to segment
1910 * This function segments the given skb and stores the list of segments
1913 static int dev_gso_segment(struct sk_buff
*skb
)
1915 struct net_device
*dev
= skb
->dev
;
1916 struct sk_buff
*segs
;
1917 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1920 segs
= skb_gso_segment(skb
, features
);
1922 /* Verifying header integrity only. */
1927 return PTR_ERR(segs
);
1930 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1931 skb
->destructor
= dev_gso_skb_destructor
;
1937 * Try to orphan skb early, right before transmission by the device.
1938 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1939 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1941 static inline void skb_orphan_try(struct sk_buff
*skb
)
1943 struct sock
*sk
= skb
->sk
;
1945 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
1946 /* skb_tx_hash() wont be able to get sk.
1947 * We copy sk_hash into skb->rxhash
1950 skb
->rxhash
= sk
->sk_hash
;
1956 * Returns true if either:
1957 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1958 * 2. skb is fragmented and the device does not support SG, or if
1959 * at least one of fragments is in highmem and device does not
1960 * support DMA from it.
1962 static inline int skb_needs_linearize(struct sk_buff
*skb
,
1963 struct net_device
*dev
)
1965 return skb_is_nonlinear(skb
) &&
1966 ((skb_has_frag_list(skb
) && !(dev
->features
& NETIF_F_FRAGLIST
)) ||
1967 (skb_shinfo(skb
)->nr_frags
&& (!(dev
->features
& NETIF_F_SG
) ||
1968 illegal_highdma(dev
, skb
))));
1971 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1972 struct netdev_queue
*txq
)
1974 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1975 int rc
= NETDEV_TX_OK
;
1977 if (likely(!skb
->next
)) {
1978 if (!list_empty(&ptype_all
))
1979 dev_queue_xmit_nit(skb
, dev
);
1982 * If device doesnt need skb->dst, release it right now while
1983 * its hot in this cpu cache
1985 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
1988 skb_orphan_try(skb
);
1990 if (netif_needs_gso(dev
, skb
)) {
1991 if (unlikely(dev_gso_segment(skb
)))
1996 if (skb_needs_linearize(skb
, dev
) &&
1997 __skb_linearize(skb
))
2000 /* If packet is not checksummed and device does not
2001 * support checksumming for this protocol, complete
2002 * checksumming here.
2004 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2005 skb_set_transport_header(skb
, skb
->csum_start
-
2007 if (!dev_can_checksum(dev
, skb
) &&
2008 skb_checksum_help(skb
))
2013 rc
= ops
->ndo_start_xmit(skb
, dev
);
2014 if (rc
== NETDEV_TX_OK
)
2015 txq_trans_update(txq
);
2021 struct sk_buff
*nskb
= skb
->next
;
2023 skb
->next
= nskb
->next
;
2027 * If device doesnt need nskb->dst, release it right now while
2028 * its hot in this cpu cache
2030 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2033 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2034 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2035 if (rc
& ~NETDEV_TX_MASK
)
2036 goto out_kfree_gso_skb
;
2037 nskb
->next
= skb
->next
;
2041 txq_trans_update(txq
);
2042 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2043 return NETDEV_TX_BUSY
;
2044 } while (skb
->next
);
2047 if (likely(skb
->next
== NULL
))
2048 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2054 static u32 hashrnd __read_mostly
;
2056 u16
skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
)
2060 if (skb_rx_queue_recorded(skb
)) {
2061 hash
= skb_get_rx_queue(skb
);
2062 while (unlikely(hash
>= dev
->real_num_tx_queues
))
2063 hash
-= dev
->real_num_tx_queues
;
2067 if (skb
->sk
&& skb
->sk
->sk_hash
)
2068 hash
= skb
->sk
->sk_hash
;
2070 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2071 hash
= jhash_1word(hash
, hashrnd
);
2073 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
2075 EXPORT_SYMBOL(skb_tx_hash
);
2077 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2079 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2080 if (net_ratelimit()) {
2081 pr_warning("%s selects TX queue %d, but "
2082 "real number of TX queues is %d\n",
2083 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2090 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2091 struct sk_buff
*skb
)
2094 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2096 if (ops
->ndo_select_queue
) {
2097 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2098 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2100 struct sock
*sk
= skb
->sk
;
2101 queue_index
= sk_tx_queue_get(sk
);
2102 if (queue_index
< 0) {
2105 if (dev
->real_num_tx_queues
> 1)
2106 queue_index
= skb_tx_hash(dev
, skb
);
2109 struct dst_entry
*dst
= rcu_dereference_check(sk
->sk_dst_cache
, 1);
2111 if (dst
&& skb_dst(skb
) == dst
)
2112 sk_tx_queue_set(sk
, queue_index
);
2117 skb_set_queue_mapping(skb
, queue_index
);
2118 return netdev_get_tx_queue(dev
, queue_index
);
2121 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2122 struct net_device
*dev
,
2123 struct netdev_queue
*txq
)
2125 spinlock_t
*root_lock
= qdisc_lock(q
);
2126 bool contended
= qdisc_is_running(q
);
2130 * Heuristic to force contended enqueues to serialize on a
2131 * separate lock before trying to get qdisc main lock.
2132 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2133 * and dequeue packets faster.
2135 if (unlikely(contended
))
2136 spin_lock(&q
->busylock
);
2138 spin_lock(root_lock
);
2139 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2142 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2143 qdisc_run_begin(q
)) {
2145 * This is a work-conserving queue; there are no old skbs
2146 * waiting to be sent out; and the qdisc is not running -
2147 * xmit the skb directly.
2149 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2151 __qdisc_update_bstats(q
, skb
->len
);
2152 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2153 if (unlikely(contended
)) {
2154 spin_unlock(&q
->busylock
);
2161 rc
= NET_XMIT_SUCCESS
;
2164 rc
= qdisc_enqueue_root(skb
, q
);
2165 if (qdisc_run_begin(q
)) {
2166 if (unlikely(contended
)) {
2167 spin_unlock(&q
->busylock
);
2173 spin_unlock(root_lock
);
2174 if (unlikely(contended
))
2175 spin_unlock(&q
->busylock
);
2179 static DEFINE_PER_CPU(int, xmit_recursion
);
2180 #define RECURSION_LIMIT 3
2183 * dev_queue_xmit - transmit a buffer
2184 * @skb: buffer to transmit
2186 * Queue a buffer for transmission to a network device. The caller must
2187 * have set the device and priority and built the buffer before calling
2188 * this function. The function can be called from an interrupt.
2190 * A negative errno code is returned on a failure. A success does not
2191 * guarantee the frame will be transmitted as it may be dropped due
2192 * to congestion or traffic shaping.
2194 * -----------------------------------------------------------------------------------
2195 * I notice this method can also return errors from the queue disciplines,
2196 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2199 * Regardless of the return value, the skb is consumed, so it is currently
2200 * difficult to retry a send to this method. (You can bump the ref count
2201 * before sending to hold a reference for retry if you are careful.)
2203 * When calling this method, interrupts MUST be enabled. This is because
2204 * the BH enable code must have IRQs enabled so that it will not deadlock.
2207 int dev_queue_xmit(struct sk_buff
*skb
)
2209 struct net_device
*dev
= skb
->dev
;
2210 struct netdev_queue
*txq
;
2214 /* Disable soft irqs for various locks below. Also
2215 * stops preemption for RCU.
2219 txq
= dev_pick_tx(dev
, skb
);
2220 q
= rcu_dereference_bh(txq
->qdisc
);
2222 #ifdef CONFIG_NET_CLS_ACT
2223 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2226 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2230 /* The device has no queue. Common case for software devices:
2231 loopback, all the sorts of tunnels...
2233 Really, it is unlikely that netif_tx_lock protection is necessary
2234 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2236 However, it is possible, that they rely on protection
2239 Check this and shot the lock. It is not prone from deadlocks.
2240 Either shot noqueue qdisc, it is even simpler 8)
2242 if (dev
->flags
& IFF_UP
) {
2243 int cpu
= smp_processor_id(); /* ok because BHs are off */
2245 if (txq
->xmit_lock_owner
!= cpu
) {
2247 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2248 goto recursion_alert
;
2250 HARD_TX_LOCK(dev
, txq
, cpu
);
2252 if (!netif_tx_queue_stopped(txq
)) {
2253 __this_cpu_inc(xmit_recursion
);
2254 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2255 __this_cpu_dec(xmit_recursion
);
2256 if (dev_xmit_complete(rc
)) {
2257 HARD_TX_UNLOCK(dev
, txq
);
2261 HARD_TX_UNLOCK(dev
, txq
);
2262 if (net_ratelimit())
2263 printk(KERN_CRIT
"Virtual device %s asks to "
2264 "queue packet!\n", dev
->name
);
2266 /* Recursion is detected! It is possible,
2270 if (net_ratelimit())
2271 printk(KERN_CRIT
"Dead loop on virtual device "
2272 "%s, fix it urgently!\n", dev
->name
);
2277 rcu_read_unlock_bh();
2282 rcu_read_unlock_bh();
2285 EXPORT_SYMBOL(dev_queue_xmit
);
2288 /*=======================================================================
2290 =======================================================================*/
2292 int netdev_max_backlog __read_mostly
= 1000;
2293 int netdev_tstamp_prequeue __read_mostly
= 1;
2294 int netdev_budget __read_mostly
= 300;
2295 int weight_p __read_mostly
= 64; /* old backlog weight */
2297 /* Called with irq disabled */
2298 static inline void ____napi_schedule(struct softnet_data
*sd
,
2299 struct napi_struct
*napi
)
2301 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2302 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2306 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2307 * and src/dst port numbers. Returns a non-zero hash number on success
2310 __u32
__skb_get_rxhash(struct sk_buff
*skb
)
2312 int nhoff
, hash
= 0, poff
;
2313 struct ipv6hdr
*ip6
;
2316 u32 addr1
, addr2
, ihl
;
2322 nhoff
= skb_network_offset(skb
);
2324 switch (skb
->protocol
) {
2325 case __constant_htons(ETH_P_IP
):
2326 if (!pskb_may_pull(skb
, sizeof(*ip
) + nhoff
))
2329 ip
= (struct iphdr
*) (skb
->data
+ nhoff
);
2330 if (ip
->frag_off
& htons(IP_MF
| IP_OFFSET
))
2333 ip_proto
= ip
->protocol
;
2334 addr1
= (__force u32
) ip
->saddr
;
2335 addr2
= (__force u32
) ip
->daddr
;
2338 case __constant_htons(ETH_P_IPV6
):
2339 if (!pskb_may_pull(skb
, sizeof(*ip6
) + nhoff
))
2342 ip6
= (struct ipv6hdr
*) (skb
->data
+ nhoff
);
2343 ip_proto
= ip6
->nexthdr
;
2344 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2345 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2353 poff
= proto_ports_offset(ip_proto
);
2355 nhoff
+= ihl
* 4 + poff
;
2356 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2357 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2358 if (ports
.v16
[1] < ports
.v16
[0])
2359 swap(ports
.v16
[0], ports
.v16
[1]);
2363 /* get a consistent hash (same value on both flow directions) */
2367 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2374 EXPORT_SYMBOL(__skb_get_rxhash
);
2378 /* One global table that all flow-based protocols share. */
2379 struct rps_sock_flow_table
*rps_sock_flow_table __read_mostly
;
2380 EXPORT_SYMBOL(rps_sock_flow_table
);
2383 * get_rps_cpu is called from netif_receive_skb and returns the target
2384 * CPU from the RPS map of the receiving queue for a given skb.
2385 * rcu_read_lock must be held on entry.
2387 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2388 struct rps_dev_flow
**rflowp
)
2390 struct netdev_rx_queue
*rxqueue
;
2391 struct rps_map
*map
= NULL
;
2392 struct rps_dev_flow_table
*flow_table
;
2393 struct rps_sock_flow_table
*sock_flow_table
;
2397 if (skb_rx_queue_recorded(skb
)) {
2398 u16 index
= skb_get_rx_queue(skb
);
2399 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2400 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2401 "%s received packet on queue %u, but number "
2402 "of RX queues is %u\n",
2403 dev
->name
, index
, dev
->real_num_rx_queues
);
2406 rxqueue
= dev
->_rx
+ index
;
2410 if (rxqueue
->rps_map
) {
2411 map
= rcu_dereference(rxqueue
->rps_map
);
2412 if (map
&& map
->len
== 1) {
2413 tcpu
= map
->cpus
[0];
2414 if (cpu_online(tcpu
))
2418 } else if (!rxqueue
->rps_flow_table
) {
2422 skb_reset_network_header(skb
);
2423 if (!skb_get_rxhash(skb
))
2426 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2427 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2428 if (flow_table
&& sock_flow_table
) {
2430 struct rps_dev_flow
*rflow
;
2432 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2435 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2436 sock_flow_table
->mask
];
2439 * If the desired CPU (where last recvmsg was done) is
2440 * different from current CPU (one in the rx-queue flow
2441 * table entry), switch if one of the following holds:
2442 * - Current CPU is unset (equal to RPS_NO_CPU).
2443 * - Current CPU is offline.
2444 * - The current CPU's queue tail has advanced beyond the
2445 * last packet that was enqueued using this table entry.
2446 * This guarantees that all previous packets for the flow
2447 * have been dequeued, thus preserving in order delivery.
2449 if (unlikely(tcpu
!= next_cpu
) &&
2450 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2451 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2452 rflow
->last_qtail
)) >= 0)) {
2453 tcpu
= rflow
->cpu
= next_cpu
;
2454 if (tcpu
!= RPS_NO_CPU
)
2455 rflow
->last_qtail
= per_cpu(softnet_data
,
2456 tcpu
).input_queue_head
;
2458 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2466 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2468 if (cpu_online(tcpu
)) {
2478 /* Called from hardirq (IPI) context */
2479 static void rps_trigger_softirq(void *data
)
2481 struct softnet_data
*sd
= data
;
2483 ____napi_schedule(sd
, &sd
->backlog
);
2487 #endif /* CONFIG_RPS */
2490 * Check if this softnet_data structure is another cpu one
2491 * If yes, queue it to our IPI list and return 1
2494 static int rps_ipi_queued(struct softnet_data
*sd
)
2497 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2500 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2501 mysd
->rps_ipi_list
= sd
;
2503 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2506 #endif /* CONFIG_RPS */
2511 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2512 * queue (may be a remote CPU queue).
2514 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2515 unsigned int *qtail
)
2517 struct softnet_data
*sd
;
2518 unsigned long flags
;
2520 sd
= &per_cpu(softnet_data
, cpu
);
2522 local_irq_save(flags
);
2525 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2526 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2528 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2529 input_queue_tail_incr_save(sd
, qtail
);
2531 local_irq_restore(flags
);
2532 return NET_RX_SUCCESS
;
2535 /* Schedule NAPI for backlog device
2536 * We can use non atomic operation since we own the queue lock
2538 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2539 if (!rps_ipi_queued(sd
))
2540 ____napi_schedule(sd
, &sd
->backlog
);
2548 local_irq_restore(flags
);
2550 atomic_long_inc(&skb
->dev
->rx_dropped
);
2556 * netif_rx - post buffer to the network code
2557 * @skb: buffer to post
2559 * This function receives a packet from a device driver and queues it for
2560 * the upper (protocol) levels to process. It always succeeds. The buffer
2561 * may be dropped during processing for congestion control or by the
2565 * NET_RX_SUCCESS (no congestion)
2566 * NET_RX_DROP (packet was dropped)
2570 int netif_rx(struct sk_buff
*skb
)
2574 /* if netpoll wants it, pretend we never saw it */
2575 if (netpoll_rx(skb
))
2578 if (netdev_tstamp_prequeue
)
2579 net_timestamp_check(skb
);
2583 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2589 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2591 cpu
= smp_processor_id();
2593 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2601 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2607 EXPORT_SYMBOL(netif_rx
);
2609 int netif_rx_ni(struct sk_buff
*skb
)
2614 err
= netif_rx(skb
);
2615 if (local_softirq_pending())
2621 EXPORT_SYMBOL(netif_rx_ni
);
2623 static void net_tx_action(struct softirq_action
*h
)
2625 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2627 if (sd
->completion_queue
) {
2628 struct sk_buff
*clist
;
2630 local_irq_disable();
2631 clist
= sd
->completion_queue
;
2632 sd
->completion_queue
= NULL
;
2636 struct sk_buff
*skb
= clist
;
2637 clist
= clist
->next
;
2639 WARN_ON(atomic_read(&skb
->users
));
2644 if (sd
->output_queue
) {
2647 local_irq_disable();
2648 head
= sd
->output_queue
;
2649 sd
->output_queue
= NULL
;
2650 sd
->output_queue_tailp
= &sd
->output_queue
;
2654 struct Qdisc
*q
= head
;
2655 spinlock_t
*root_lock
;
2657 head
= head
->next_sched
;
2659 root_lock
= qdisc_lock(q
);
2660 if (spin_trylock(root_lock
)) {
2661 smp_mb__before_clear_bit();
2662 clear_bit(__QDISC_STATE_SCHED
,
2665 spin_unlock(root_lock
);
2667 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2669 __netif_reschedule(q
);
2671 smp_mb__before_clear_bit();
2672 clear_bit(__QDISC_STATE_SCHED
,
2680 static inline int deliver_skb(struct sk_buff
*skb
,
2681 struct packet_type
*pt_prev
,
2682 struct net_device
*orig_dev
)
2684 atomic_inc(&skb
->users
);
2685 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2688 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2689 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2690 /* This hook is defined here for ATM LANE */
2691 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2692 unsigned char *addr
) __read_mostly
;
2693 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2696 #ifdef CONFIG_NET_CLS_ACT
2697 /* TODO: Maybe we should just force sch_ingress to be compiled in
2698 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2699 * a compare and 2 stores extra right now if we dont have it on
2700 * but have CONFIG_NET_CLS_ACT
2701 * NOTE: This doesnt stop any functionality; if you dont have
2702 * the ingress scheduler, you just cant add policies on ingress.
2705 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
2707 struct net_device
*dev
= skb
->dev
;
2708 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2709 int result
= TC_ACT_OK
;
2712 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
2713 if (net_ratelimit())
2714 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2715 skb
->skb_iif
, dev
->ifindex
);
2719 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2720 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2723 if (q
!= &noop_qdisc
) {
2724 spin_lock(qdisc_lock(q
));
2725 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2726 result
= qdisc_enqueue_root(skb
, q
);
2727 spin_unlock(qdisc_lock(q
));
2733 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2734 struct packet_type
**pt_prev
,
2735 int *ret
, struct net_device
*orig_dev
)
2737 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
2739 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
2743 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2747 switch (ing_filter(skb
, rxq
)) {
2761 * netif_nit_deliver - deliver received packets to network taps
2764 * This function is used to deliver incoming packets to network
2765 * taps. It should be used when the normal netif_receive_skb path
2766 * is bypassed, for example because of VLAN acceleration.
2768 void netif_nit_deliver(struct sk_buff
*skb
)
2770 struct packet_type
*ptype
;
2772 if (list_empty(&ptype_all
))
2775 skb_reset_network_header(skb
);
2776 skb_reset_transport_header(skb
);
2777 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2780 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2781 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2782 deliver_skb(skb
, ptype
, skb
->dev
);
2788 * netdev_rx_handler_register - register receive handler
2789 * @dev: device to register a handler for
2790 * @rx_handler: receive handler to register
2791 * @rx_handler_data: data pointer that is used by rx handler
2793 * Register a receive hander for a device. This handler will then be
2794 * called from __netif_receive_skb. A negative errno code is returned
2797 * The caller must hold the rtnl_mutex.
2799 int netdev_rx_handler_register(struct net_device
*dev
,
2800 rx_handler_func_t
*rx_handler
,
2801 void *rx_handler_data
)
2805 if (dev
->rx_handler
)
2808 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
2809 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
2813 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
2816 * netdev_rx_handler_unregister - unregister receive handler
2817 * @dev: device to unregister a handler from
2819 * Unregister a receive hander from a device.
2821 * The caller must hold the rtnl_mutex.
2823 void netdev_rx_handler_unregister(struct net_device
*dev
)
2827 rcu_assign_pointer(dev
->rx_handler
, NULL
);
2828 rcu_assign_pointer(dev
->rx_handler_data
, NULL
);
2830 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
2832 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2833 struct net_device
*master
)
2835 if (skb
->pkt_type
== PACKET_HOST
) {
2836 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2838 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2842 /* On bonding slaves other than the currently active slave, suppress
2843 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2844 * ARP on active-backup slaves with arp_validate enabled.
2846 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2848 struct net_device
*dev
= skb
->dev
;
2850 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2851 dev
->last_rx
= jiffies
;
2853 if ((master
->priv_flags
& IFF_MASTER_ALB
) &&
2854 (master
->priv_flags
& IFF_BRIDGE_PORT
)) {
2855 /* Do address unmangle. The local destination address
2856 * will be always the one master has. Provides the right
2857 * functionality in a bridge.
2859 skb_bond_set_mac_by_master(skb
, master
);
2862 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
2863 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
2864 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
2867 if (master
->priv_flags
& IFF_MASTER_ALB
) {
2868 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
2869 skb
->pkt_type
!= PACKET_MULTICAST
)
2872 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
2873 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
2880 EXPORT_SYMBOL(__skb_bond_should_drop
);
2882 static int __netif_receive_skb(struct sk_buff
*skb
)
2884 struct packet_type
*ptype
, *pt_prev
;
2885 rx_handler_func_t
*rx_handler
;
2886 struct net_device
*orig_dev
;
2887 struct net_device
*master
;
2888 struct net_device
*null_or_orig
;
2889 struct net_device
*orig_or_bond
;
2890 int ret
= NET_RX_DROP
;
2893 if (!netdev_tstamp_prequeue
)
2894 net_timestamp_check(skb
);
2896 if (vlan_tx_tag_present(skb
))
2897 vlan_hwaccel_do_receive(skb
);
2899 /* if we've gotten here through NAPI, check netpoll */
2900 if (netpoll_receive_skb(skb
))
2904 skb
->skb_iif
= skb
->dev
->ifindex
;
2907 * bonding note: skbs received on inactive slaves should only
2908 * be delivered to pkt handlers that are exact matches. Also
2909 * the deliver_no_wcard flag will be set. If packet handlers
2910 * are sensitive to duplicate packets these skbs will need to
2911 * be dropped at the handler. The vlan accel path may have
2912 * already set the deliver_no_wcard flag.
2914 null_or_orig
= NULL
;
2915 orig_dev
= skb
->dev
;
2916 master
= ACCESS_ONCE(orig_dev
->master
);
2917 if (skb
->deliver_no_wcard
)
2918 null_or_orig
= orig_dev
;
2920 if (skb_bond_should_drop(skb
, master
)) {
2921 skb
->deliver_no_wcard
= 1;
2922 null_or_orig
= orig_dev
; /* deliver only exact match */
2927 __this_cpu_inc(softnet_data
.processed
);
2928 skb_reset_network_header(skb
);
2929 skb_reset_transport_header(skb
);
2930 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2936 #ifdef CONFIG_NET_CLS_ACT
2937 if (skb
->tc_verd
& TC_NCLS
) {
2938 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2943 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2944 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2945 ptype
->dev
== orig_dev
) {
2947 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2952 #ifdef CONFIG_NET_CLS_ACT
2953 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2959 /* Handle special case of bridge or macvlan */
2960 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
2963 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2966 skb
= rx_handler(skb
);
2972 * Make sure frames received on VLAN interfaces stacked on
2973 * bonding interfaces still make their way to any base bonding
2974 * device that may have registered for a specific ptype. The
2975 * handler may have to adjust skb->dev and orig_dev.
2977 orig_or_bond
= orig_dev
;
2978 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
2979 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
2980 orig_or_bond
= vlan_dev_real_dev(skb
->dev
);
2983 type
= skb
->protocol
;
2984 list_for_each_entry_rcu(ptype
,
2985 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2986 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
2987 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
2988 ptype
->dev
== orig_or_bond
)) {
2990 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2996 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2998 atomic_long_inc(&skb
->dev
->rx_dropped
);
3000 /* Jamal, now you will not able to escape explaining
3001 * me how you were going to use this. :-)
3012 * netif_receive_skb - process receive buffer from network
3013 * @skb: buffer to process
3015 * netif_receive_skb() is the main receive data processing function.
3016 * It always succeeds. The buffer may be dropped during processing
3017 * for congestion control or by the protocol layers.
3019 * This function may only be called from softirq context and interrupts
3020 * should be enabled.
3022 * Return values (usually ignored):
3023 * NET_RX_SUCCESS: no congestion
3024 * NET_RX_DROP: packet was dropped
3026 int netif_receive_skb(struct sk_buff
*skb
)
3028 if (netdev_tstamp_prequeue
)
3029 net_timestamp_check(skb
);
3031 if (skb_defer_rx_timestamp(skb
))
3032 return NET_RX_SUCCESS
;
3036 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3041 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3044 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3048 ret
= __netif_receive_skb(skb
);
3054 return __netif_receive_skb(skb
);
3057 EXPORT_SYMBOL(netif_receive_skb
);
3059 /* Network device is going away, flush any packets still pending
3060 * Called with irqs disabled.
3062 static void flush_backlog(void *arg
)
3064 struct net_device
*dev
= arg
;
3065 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3066 struct sk_buff
*skb
, *tmp
;
3069 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3070 if (skb
->dev
== dev
) {
3071 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3073 input_queue_head_incr(sd
);
3078 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3079 if (skb
->dev
== dev
) {
3080 __skb_unlink(skb
, &sd
->process_queue
);
3082 input_queue_head_incr(sd
);
3087 static int napi_gro_complete(struct sk_buff
*skb
)
3089 struct packet_type
*ptype
;
3090 __be16 type
= skb
->protocol
;
3091 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3094 if (NAPI_GRO_CB(skb
)->count
== 1) {
3095 skb_shinfo(skb
)->gso_size
= 0;
3100 list_for_each_entry_rcu(ptype
, head
, list
) {
3101 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3104 err
= ptype
->gro_complete(skb
);
3110 WARN_ON(&ptype
->list
== head
);
3112 return NET_RX_SUCCESS
;
3116 return netif_receive_skb(skb
);
3119 inline void napi_gro_flush(struct napi_struct
*napi
)
3121 struct sk_buff
*skb
, *next
;
3123 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3126 napi_gro_complete(skb
);
3129 napi
->gro_count
= 0;
3130 napi
->gro_list
= NULL
;
3132 EXPORT_SYMBOL(napi_gro_flush
);
3134 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3136 struct sk_buff
**pp
= NULL
;
3137 struct packet_type
*ptype
;
3138 __be16 type
= skb
->protocol
;
3139 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3142 enum gro_result ret
;
3144 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3147 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3151 list_for_each_entry_rcu(ptype
, head
, list
) {
3152 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3155 skb_set_network_header(skb
, skb_gro_offset(skb
));
3156 mac_len
= skb
->network_header
- skb
->mac_header
;
3157 skb
->mac_len
= mac_len
;
3158 NAPI_GRO_CB(skb
)->same_flow
= 0;
3159 NAPI_GRO_CB(skb
)->flush
= 0;
3160 NAPI_GRO_CB(skb
)->free
= 0;
3162 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3167 if (&ptype
->list
== head
)
3170 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3171 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3174 struct sk_buff
*nskb
= *pp
;
3178 napi_gro_complete(nskb
);
3185 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3189 NAPI_GRO_CB(skb
)->count
= 1;
3190 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3191 skb
->next
= napi
->gro_list
;
3192 napi
->gro_list
= skb
;
3196 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3197 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3199 BUG_ON(skb
->end
- skb
->tail
< grow
);
3201 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3204 skb
->data_len
-= grow
;
3206 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3207 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3209 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3210 put_page(skb_shinfo(skb
)->frags
[0].page
);
3211 memmove(skb_shinfo(skb
)->frags
,
3212 skb_shinfo(skb
)->frags
+ 1,
3213 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3224 EXPORT_SYMBOL(dev_gro_receive
);
3226 static inline gro_result_t
3227 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3231 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3232 unsigned long diffs
;
3234 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3235 diffs
|= compare_ether_header(skb_mac_header(p
),
3236 skb_gro_mac_header(skb
));
3237 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3238 NAPI_GRO_CB(p
)->flush
= 0;
3241 return dev_gro_receive(napi
, skb
);
3244 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3248 if (netif_receive_skb(skb
))
3253 case GRO_MERGED_FREE
:
3264 EXPORT_SYMBOL(napi_skb_finish
);
3266 void skb_gro_reset_offset(struct sk_buff
*skb
)
3268 NAPI_GRO_CB(skb
)->data_offset
= 0;
3269 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3270 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3272 if (skb
->mac_header
== skb
->tail
&&
3273 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3274 NAPI_GRO_CB(skb
)->frag0
=
3275 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3276 skb_shinfo(skb
)->frags
[0].page_offset
;
3277 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3280 EXPORT_SYMBOL(skb_gro_reset_offset
);
3282 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3284 skb_gro_reset_offset(skb
);
3286 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3288 EXPORT_SYMBOL(napi_gro_receive
);
3290 void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3292 __skb_pull(skb
, skb_headlen(skb
));
3293 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3297 EXPORT_SYMBOL(napi_reuse_skb
);
3299 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3301 struct sk_buff
*skb
= napi
->skb
;
3304 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3310 EXPORT_SYMBOL(napi_get_frags
);
3312 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3318 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3320 if (ret
== GRO_HELD
)
3321 skb_gro_pull(skb
, -ETH_HLEN
);
3322 else if (netif_receive_skb(skb
))
3327 case GRO_MERGED_FREE
:
3328 napi_reuse_skb(napi
, skb
);
3337 EXPORT_SYMBOL(napi_frags_finish
);
3339 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3341 struct sk_buff
*skb
= napi
->skb
;
3348 skb_reset_mac_header(skb
);
3349 skb_gro_reset_offset(skb
);
3351 off
= skb_gro_offset(skb
);
3352 hlen
= off
+ sizeof(*eth
);
3353 eth
= skb_gro_header_fast(skb
, off
);
3354 if (skb_gro_header_hard(skb
, hlen
)) {
3355 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3356 if (unlikely(!eth
)) {
3357 napi_reuse_skb(napi
, skb
);
3363 skb_gro_pull(skb
, sizeof(*eth
));
3366 * This works because the only protocols we care about don't require
3367 * special handling. We'll fix it up properly at the end.
3369 skb
->protocol
= eth
->h_proto
;
3374 EXPORT_SYMBOL(napi_frags_skb
);
3376 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3378 struct sk_buff
*skb
= napi_frags_skb(napi
);
3383 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3385 EXPORT_SYMBOL(napi_gro_frags
);
3388 * net_rps_action sends any pending IPI's for rps.
3389 * Note: called with local irq disabled, but exits with local irq enabled.
3391 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3394 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3397 sd
->rps_ipi_list
= NULL
;
3401 /* Send pending IPI's to kick RPS processing on remote cpus. */
3403 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3405 if (cpu_online(remsd
->cpu
))
3406 __smp_call_function_single(remsd
->cpu
,
3415 static int process_backlog(struct napi_struct
*napi
, int quota
)
3418 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3421 /* Check if we have pending ipi, its better to send them now,
3422 * not waiting net_rx_action() end.
3424 if (sd
->rps_ipi_list
) {
3425 local_irq_disable();
3426 net_rps_action_and_irq_enable(sd
);
3429 napi
->weight
= weight_p
;
3430 local_irq_disable();
3431 while (work
< quota
) {
3432 struct sk_buff
*skb
;
3435 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3437 __netif_receive_skb(skb
);
3438 local_irq_disable();
3439 input_queue_head_incr(sd
);
3440 if (++work
>= quota
) {
3447 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3449 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3450 &sd
->process_queue
);
3452 if (qlen
< quota
- work
) {
3454 * Inline a custom version of __napi_complete().
3455 * only current cpu owns and manipulates this napi,
3456 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3457 * we can use a plain write instead of clear_bit(),
3458 * and we dont need an smp_mb() memory barrier.
3460 list_del(&napi
->poll_list
);
3463 quota
= work
+ qlen
;
3473 * __napi_schedule - schedule for receive
3474 * @n: entry to schedule
3476 * The entry's receive function will be scheduled to run
3478 void __napi_schedule(struct napi_struct
*n
)
3480 unsigned long flags
;
3482 local_irq_save(flags
);
3483 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3484 local_irq_restore(flags
);
3486 EXPORT_SYMBOL(__napi_schedule
);
3488 void __napi_complete(struct napi_struct
*n
)
3490 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3491 BUG_ON(n
->gro_list
);
3493 list_del(&n
->poll_list
);
3494 smp_mb__before_clear_bit();
3495 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3497 EXPORT_SYMBOL(__napi_complete
);
3499 void napi_complete(struct napi_struct
*n
)
3501 unsigned long flags
;
3504 * don't let napi dequeue from the cpu poll list
3505 * just in case its running on a different cpu
3507 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3511 local_irq_save(flags
);
3513 local_irq_restore(flags
);
3515 EXPORT_SYMBOL(napi_complete
);
3517 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3518 int (*poll
)(struct napi_struct
*, int), int weight
)
3520 INIT_LIST_HEAD(&napi
->poll_list
);
3521 napi
->gro_count
= 0;
3522 napi
->gro_list
= NULL
;
3525 napi
->weight
= weight
;
3526 list_add(&napi
->dev_list
, &dev
->napi_list
);
3528 #ifdef CONFIG_NETPOLL
3529 spin_lock_init(&napi
->poll_lock
);
3530 napi
->poll_owner
= -1;
3532 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3534 EXPORT_SYMBOL(netif_napi_add
);
3536 void netif_napi_del(struct napi_struct
*napi
)
3538 struct sk_buff
*skb
, *next
;
3540 list_del_init(&napi
->dev_list
);
3541 napi_free_frags(napi
);
3543 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3549 napi
->gro_list
= NULL
;
3550 napi
->gro_count
= 0;
3552 EXPORT_SYMBOL(netif_napi_del
);
3554 static void net_rx_action(struct softirq_action
*h
)
3556 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3557 unsigned long time_limit
= jiffies
+ 2;
3558 int budget
= netdev_budget
;
3561 local_irq_disable();
3563 while (!list_empty(&sd
->poll_list
)) {
3564 struct napi_struct
*n
;
3567 /* If softirq window is exhuasted then punt.
3568 * Allow this to run for 2 jiffies since which will allow
3569 * an average latency of 1.5/HZ.
3571 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3576 /* Even though interrupts have been re-enabled, this
3577 * access is safe because interrupts can only add new
3578 * entries to the tail of this list, and only ->poll()
3579 * calls can remove this head entry from the list.
3581 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3583 have
= netpoll_poll_lock(n
);
3587 /* This NAPI_STATE_SCHED test is for avoiding a race
3588 * with netpoll's poll_napi(). Only the entity which
3589 * obtains the lock and sees NAPI_STATE_SCHED set will
3590 * actually make the ->poll() call. Therefore we avoid
3591 * accidently calling ->poll() when NAPI is not scheduled.
3594 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3595 work
= n
->poll(n
, weight
);
3599 WARN_ON_ONCE(work
> weight
);
3603 local_irq_disable();
3605 /* Drivers must not modify the NAPI state if they
3606 * consume the entire weight. In such cases this code
3607 * still "owns" the NAPI instance and therefore can
3608 * move the instance around on the list at-will.
3610 if (unlikely(work
== weight
)) {
3611 if (unlikely(napi_disable_pending(n
))) {
3614 local_irq_disable();
3616 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3619 netpoll_poll_unlock(have
);
3622 net_rps_action_and_irq_enable(sd
);
3624 #ifdef CONFIG_NET_DMA
3626 * There may not be any more sk_buffs coming right now, so push
3627 * any pending DMA copies to hardware
3629 dma_issue_pending_all();
3636 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3640 static gifconf_func_t
*gifconf_list
[NPROTO
];
3643 * register_gifconf - register a SIOCGIF handler
3644 * @family: Address family
3645 * @gifconf: Function handler
3647 * Register protocol dependent address dumping routines. The handler
3648 * that is passed must not be freed or reused until it has been replaced
3649 * by another handler.
3651 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3653 if (family
>= NPROTO
)
3655 gifconf_list
[family
] = gifconf
;
3658 EXPORT_SYMBOL(register_gifconf
);
3662 * Map an interface index to its name (SIOCGIFNAME)
3666 * We need this ioctl for efficient implementation of the
3667 * if_indextoname() function required by the IPv6 API. Without
3668 * it, we would have to search all the interfaces to find a
3672 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3674 struct net_device
*dev
;
3678 * Fetch the caller's info block.
3681 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3685 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3691 strcpy(ifr
.ifr_name
, dev
->name
);
3694 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3700 * Perform a SIOCGIFCONF call. This structure will change
3701 * size eventually, and there is nothing I can do about it.
3702 * Thus we will need a 'compatibility mode'.
3705 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3708 struct net_device
*dev
;
3715 * Fetch the caller's info block.
3718 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3725 * Loop over the interfaces, and write an info block for each.
3729 for_each_netdev(net
, dev
) {
3730 for (i
= 0; i
< NPROTO
; i
++) {
3731 if (gifconf_list
[i
]) {
3734 done
= gifconf_list
[i
](dev
, NULL
, 0);
3736 done
= gifconf_list
[i
](dev
, pos
+ total
,
3746 * All done. Write the updated control block back to the caller.
3748 ifc
.ifc_len
= total
;
3751 * Both BSD and Solaris return 0 here, so we do too.
3753 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3756 #ifdef CONFIG_PROC_FS
3758 * This is invoked by the /proc filesystem handler to display a device
3761 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3764 struct net
*net
= seq_file_net(seq
);
3766 struct net_device
*dev
;
3770 return SEQ_START_TOKEN
;
3773 for_each_netdev_rcu(net
, dev
)
3780 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3782 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3783 first_net_device(seq_file_net(seq
)) :
3784 next_net_device((struct net_device
*)v
);
3787 return rcu_dereference(dev
);
3790 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3796 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3798 struct rtnl_link_stats64 temp
;
3799 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
3801 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3802 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3803 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3805 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3806 stats
->rx_fifo_errors
,
3807 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3808 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3809 stats
->rx_compressed
, stats
->multicast
,
3810 stats
->tx_bytes
, stats
->tx_packets
,
3811 stats
->tx_errors
, stats
->tx_dropped
,
3812 stats
->tx_fifo_errors
, stats
->collisions
,
3813 stats
->tx_carrier_errors
+
3814 stats
->tx_aborted_errors
+
3815 stats
->tx_window_errors
+
3816 stats
->tx_heartbeat_errors
,
3817 stats
->tx_compressed
);
3821 * Called from the PROCfs module. This now uses the new arbitrary sized
3822 * /proc/net interface to create /proc/net/dev
3824 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3826 if (v
== SEQ_START_TOKEN
)
3827 seq_puts(seq
, "Inter-| Receive "
3829 " face |bytes packets errs drop fifo frame "
3830 "compressed multicast|bytes packets errs "
3831 "drop fifo colls carrier compressed\n");
3833 dev_seq_printf_stats(seq
, v
);
3837 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
3839 struct softnet_data
*sd
= NULL
;
3841 while (*pos
< nr_cpu_ids
)
3842 if (cpu_online(*pos
)) {
3843 sd
= &per_cpu(softnet_data
, *pos
);
3850 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3852 return softnet_get_online(pos
);
3855 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3858 return softnet_get_online(pos
);
3861 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3865 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3867 struct softnet_data
*sd
= v
;
3869 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3870 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
3871 0, 0, 0, 0, /* was fastroute */
3872 sd
->cpu_collision
, sd
->received_rps
);
3876 static const struct seq_operations dev_seq_ops
= {
3877 .start
= dev_seq_start
,
3878 .next
= dev_seq_next
,
3879 .stop
= dev_seq_stop
,
3880 .show
= dev_seq_show
,
3883 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
3885 return seq_open_net(inode
, file
, &dev_seq_ops
,
3886 sizeof(struct seq_net_private
));
3889 static const struct file_operations dev_seq_fops
= {
3890 .owner
= THIS_MODULE
,
3891 .open
= dev_seq_open
,
3893 .llseek
= seq_lseek
,
3894 .release
= seq_release_net
,
3897 static const struct seq_operations softnet_seq_ops
= {
3898 .start
= softnet_seq_start
,
3899 .next
= softnet_seq_next
,
3900 .stop
= softnet_seq_stop
,
3901 .show
= softnet_seq_show
,
3904 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
3906 return seq_open(file
, &softnet_seq_ops
);
3909 static const struct file_operations softnet_seq_fops
= {
3910 .owner
= THIS_MODULE
,
3911 .open
= softnet_seq_open
,
3913 .llseek
= seq_lseek
,
3914 .release
= seq_release
,
3917 static void *ptype_get_idx(loff_t pos
)
3919 struct packet_type
*pt
= NULL
;
3923 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
3929 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
3930 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
3939 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3943 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
3946 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3948 struct packet_type
*pt
;
3949 struct list_head
*nxt
;
3953 if (v
== SEQ_START_TOKEN
)
3954 return ptype_get_idx(0);
3957 nxt
= pt
->list
.next
;
3958 if (pt
->type
== htons(ETH_P_ALL
)) {
3959 if (nxt
!= &ptype_all
)
3962 nxt
= ptype_base
[0].next
;
3964 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
3966 while (nxt
== &ptype_base
[hash
]) {
3967 if (++hash
>= PTYPE_HASH_SIZE
)
3969 nxt
= ptype_base
[hash
].next
;
3972 return list_entry(nxt
, struct packet_type
, list
);
3975 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
3981 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
3983 struct packet_type
*pt
= v
;
3985 if (v
== SEQ_START_TOKEN
)
3986 seq_puts(seq
, "Type Device Function\n");
3987 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
3988 if (pt
->type
== htons(ETH_P_ALL
))
3989 seq_puts(seq
, "ALL ");
3991 seq_printf(seq
, "%04x", ntohs(pt
->type
));
3993 seq_printf(seq
, " %-8s %pF\n",
3994 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4000 static const struct seq_operations ptype_seq_ops
= {
4001 .start
= ptype_seq_start
,
4002 .next
= ptype_seq_next
,
4003 .stop
= ptype_seq_stop
,
4004 .show
= ptype_seq_show
,
4007 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4009 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4010 sizeof(struct seq_net_private
));
4013 static const struct file_operations ptype_seq_fops
= {
4014 .owner
= THIS_MODULE
,
4015 .open
= ptype_seq_open
,
4017 .llseek
= seq_lseek
,
4018 .release
= seq_release_net
,
4022 static int __net_init
dev_proc_net_init(struct net
*net
)
4026 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4028 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4030 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4033 if (wext_proc_init(net
))
4039 proc_net_remove(net
, "ptype");
4041 proc_net_remove(net
, "softnet_stat");
4043 proc_net_remove(net
, "dev");
4047 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4049 wext_proc_exit(net
);
4051 proc_net_remove(net
, "ptype");
4052 proc_net_remove(net
, "softnet_stat");
4053 proc_net_remove(net
, "dev");
4056 static struct pernet_operations __net_initdata dev_proc_ops
= {
4057 .init
= dev_proc_net_init
,
4058 .exit
= dev_proc_net_exit
,
4061 static int __init
dev_proc_init(void)
4063 return register_pernet_subsys(&dev_proc_ops
);
4066 #define dev_proc_init() 0
4067 #endif /* CONFIG_PROC_FS */
4071 * netdev_set_master - set up master/slave pair
4072 * @slave: slave device
4073 * @master: new master device
4075 * Changes the master device of the slave. Pass %NULL to break the
4076 * bonding. The caller must hold the RTNL semaphore. On a failure
4077 * a negative errno code is returned. On success the reference counts
4078 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4079 * function returns zero.
4081 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4083 struct net_device
*old
= slave
->master
;
4093 slave
->master
= master
;
4100 slave
->flags
|= IFF_SLAVE
;
4102 slave
->flags
&= ~IFF_SLAVE
;
4104 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4107 EXPORT_SYMBOL(netdev_set_master
);
4109 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4111 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4113 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4114 ops
->ndo_change_rx_flags(dev
, flags
);
4117 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4119 unsigned short old_flags
= dev
->flags
;
4125 dev
->flags
|= IFF_PROMISC
;
4126 dev
->promiscuity
+= inc
;
4127 if (dev
->promiscuity
== 0) {
4130 * If inc causes overflow, untouch promisc and return error.
4133 dev
->flags
&= ~IFF_PROMISC
;
4135 dev
->promiscuity
-= inc
;
4136 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4137 "set promiscuity failed, promiscuity feature "
4138 "of device might be broken.\n", dev
->name
);
4142 if (dev
->flags
!= old_flags
) {
4143 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4144 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4146 if (audit_enabled
) {
4147 current_uid_gid(&uid
, &gid
);
4148 audit_log(current
->audit_context
, GFP_ATOMIC
,
4149 AUDIT_ANOM_PROMISCUOUS
,
4150 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4151 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4152 (old_flags
& IFF_PROMISC
),
4153 audit_get_loginuid(current
),
4155 audit_get_sessionid(current
));
4158 dev_change_rx_flags(dev
, IFF_PROMISC
);
4164 * dev_set_promiscuity - update promiscuity count on a device
4168 * Add or remove promiscuity from a device. While the count in the device
4169 * remains above zero the interface remains promiscuous. Once it hits zero
4170 * the device reverts back to normal filtering operation. A negative inc
4171 * value is used to drop promiscuity on the device.
4172 * Return 0 if successful or a negative errno code on error.
4174 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4176 unsigned short old_flags
= dev
->flags
;
4179 err
= __dev_set_promiscuity(dev
, inc
);
4182 if (dev
->flags
!= old_flags
)
4183 dev_set_rx_mode(dev
);
4186 EXPORT_SYMBOL(dev_set_promiscuity
);
4189 * dev_set_allmulti - update allmulti count on a device
4193 * Add or remove reception of all multicast frames to a device. While the
4194 * count in the device remains above zero the interface remains listening
4195 * to all interfaces. Once it hits zero the device reverts back to normal
4196 * filtering operation. A negative @inc value is used to drop the counter
4197 * when releasing a resource needing all multicasts.
4198 * Return 0 if successful or a negative errno code on error.
4201 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4203 unsigned short old_flags
= dev
->flags
;
4207 dev
->flags
|= IFF_ALLMULTI
;
4208 dev
->allmulti
+= inc
;
4209 if (dev
->allmulti
== 0) {
4212 * If inc causes overflow, untouch allmulti and return error.
4215 dev
->flags
&= ~IFF_ALLMULTI
;
4217 dev
->allmulti
-= inc
;
4218 printk(KERN_WARNING
"%s: allmulti touches roof, "
4219 "set allmulti failed, allmulti feature of "
4220 "device might be broken.\n", dev
->name
);
4224 if (dev
->flags
^ old_flags
) {
4225 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4226 dev_set_rx_mode(dev
);
4230 EXPORT_SYMBOL(dev_set_allmulti
);
4233 * Upload unicast and multicast address lists to device and
4234 * configure RX filtering. When the device doesn't support unicast
4235 * filtering it is put in promiscuous mode while unicast addresses
4238 void __dev_set_rx_mode(struct net_device
*dev
)
4240 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4242 /* dev_open will call this function so the list will stay sane. */
4243 if (!(dev
->flags
&IFF_UP
))
4246 if (!netif_device_present(dev
))
4249 if (ops
->ndo_set_rx_mode
)
4250 ops
->ndo_set_rx_mode(dev
);
4252 /* Unicast addresses changes may only happen under the rtnl,
4253 * therefore calling __dev_set_promiscuity here is safe.
4255 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4256 __dev_set_promiscuity(dev
, 1);
4257 dev
->uc_promisc
= 1;
4258 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4259 __dev_set_promiscuity(dev
, -1);
4260 dev
->uc_promisc
= 0;
4263 if (ops
->ndo_set_multicast_list
)
4264 ops
->ndo_set_multicast_list(dev
);
4268 void dev_set_rx_mode(struct net_device
*dev
)
4270 netif_addr_lock_bh(dev
);
4271 __dev_set_rx_mode(dev
);
4272 netif_addr_unlock_bh(dev
);
4276 * dev_get_flags - get flags reported to userspace
4279 * Get the combination of flag bits exported through APIs to userspace.
4281 unsigned dev_get_flags(const struct net_device
*dev
)
4285 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4290 (dev
->gflags
& (IFF_PROMISC
|
4293 if (netif_running(dev
)) {
4294 if (netif_oper_up(dev
))
4295 flags
|= IFF_RUNNING
;
4296 if (netif_carrier_ok(dev
))
4297 flags
|= IFF_LOWER_UP
;
4298 if (netif_dormant(dev
))
4299 flags
|= IFF_DORMANT
;
4304 EXPORT_SYMBOL(dev_get_flags
);
4306 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4308 int old_flags
= dev
->flags
;
4314 * Set the flags on our device.
4317 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4318 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4320 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4324 * Load in the correct multicast list now the flags have changed.
4327 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4328 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4330 dev_set_rx_mode(dev
);
4333 * Have we downed the interface. We handle IFF_UP ourselves
4334 * according to user attempts to set it, rather than blindly
4339 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4340 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4343 dev_set_rx_mode(dev
);
4346 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4347 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4349 dev
->gflags
^= IFF_PROMISC
;
4350 dev_set_promiscuity(dev
, inc
);
4353 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4354 is important. Some (broken) drivers set IFF_PROMISC, when
4355 IFF_ALLMULTI is requested not asking us and not reporting.
4357 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4358 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4360 dev
->gflags
^= IFF_ALLMULTI
;
4361 dev_set_allmulti(dev
, inc
);
4367 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4369 unsigned int changes
= dev
->flags
^ old_flags
;
4371 if (changes
& IFF_UP
) {
4372 if (dev
->flags
& IFF_UP
)
4373 call_netdevice_notifiers(NETDEV_UP
, dev
);
4375 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4378 if (dev
->flags
& IFF_UP
&&
4379 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4380 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4384 * dev_change_flags - change device settings
4386 * @flags: device state flags
4388 * Change settings on device based state flags. The flags are
4389 * in the userspace exported format.
4391 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4394 int old_flags
= dev
->flags
;
4396 ret
= __dev_change_flags(dev
, flags
);
4400 changes
= old_flags
^ dev
->flags
;
4402 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4404 __dev_notify_flags(dev
, old_flags
);
4407 EXPORT_SYMBOL(dev_change_flags
);
4410 * dev_set_mtu - Change maximum transfer unit
4412 * @new_mtu: new transfer unit
4414 * Change the maximum transfer size of the network device.
4416 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4418 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4421 if (new_mtu
== dev
->mtu
)
4424 /* MTU must be positive. */
4428 if (!netif_device_present(dev
))
4432 if (ops
->ndo_change_mtu
)
4433 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4437 if (!err
&& dev
->flags
& IFF_UP
)
4438 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4441 EXPORT_SYMBOL(dev_set_mtu
);
4444 * dev_set_mac_address - Change Media Access Control Address
4448 * Change the hardware (MAC) address of the device
4450 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4452 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4455 if (!ops
->ndo_set_mac_address
)
4457 if (sa
->sa_family
!= dev
->type
)
4459 if (!netif_device_present(dev
))
4461 err
= ops
->ndo_set_mac_address(dev
, sa
);
4463 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4466 EXPORT_SYMBOL(dev_set_mac_address
);
4469 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4471 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4474 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4480 case SIOCGIFFLAGS
: /* Get interface flags */
4481 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4484 case SIOCGIFMETRIC
: /* Get the metric on the interface
4485 (currently unused) */
4486 ifr
->ifr_metric
= 0;
4489 case SIOCGIFMTU
: /* Get the MTU of a device */
4490 ifr
->ifr_mtu
= dev
->mtu
;
4495 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4497 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4498 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4499 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4507 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4508 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4509 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4510 ifr
->ifr_map
.irq
= dev
->irq
;
4511 ifr
->ifr_map
.dma
= dev
->dma
;
4512 ifr
->ifr_map
.port
= dev
->if_port
;
4516 ifr
->ifr_ifindex
= dev
->ifindex
;
4520 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4524 /* dev_ioctl() should ensure this case
4536 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4538 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4541 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4542 const struct net_device_ops
*ops
;
4547 ops
= dev
->netdev_ops
;
4550 case SIOCSIFFLAGS
: /* Set interface flags */
4551 return dev_change_flags(dev
, ifr
->ifr_flags
);
4553 case SIOCSIFMETRIC
: /* Set the metric on the interface
4554 (currently unused) */
4557 case SIOCSIFMTU
: /* Set the MTU of a device */
4558 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4561 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4563 case SIOCSIFHWBROADCAST
:
4564 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4566 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4567 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4568 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4572 if (ops
->ndo_set_config
) {
4573 if (!netif_device_present(dev
))
4575 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4580 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4581 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4583 if (!netif_device_present(dev
))
4585 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4588 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4589 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4591 if (!netif_device_present(dev
))
4593 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4596 if (ifr
->ifr_qlen
< 0)
4598 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4602 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4603 return dev_change_name(dev
, ifr
->ifr_newname
);
4606 * Unknown or private ioctl
4609 if ((cmd
>= SIOCDEVPRIVATE
&&
4610 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4611 cmd
== SIOCBONDENSLAVE
||
4612 cmd
== SIOCBONDRELEASE
||
4613 cmd
== SIOCBONDSETHWADDR
||
4614 cmd
== SIOCBONDSLAVEINFOQUERY
||
4615 cmd
== SIOCBONDINFOQUERY
||
4616 cmd
== SIOCBONDCHANGEACTIVE
||
4617 cmd
== SIOCGMIIPHY
||
4618 cmd
== SIOCGMIIREG
||
4619 cmd
== SIOCSMIIREG
||
4620 cmd
== SIOCBRADDIF
||
4621 cmd
== SIOCBRDELIF
||
4622 cmd
== SIOCSHWTSTAMP
||
4623 cmd
== SIOCWANDEV
) {
4625 if (ops
->ndo_do_ioctl
) {
4626 if (netif_device_present(dev
))
4627 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4639 * This function handles all "interface"-type I/O control requests. The actual
4640 * 'doing' part of this is dev_ifsioc above.
4644 * dev_ioctl - network device ioctl
4645 * @net: the applicable net namespace
4646 * @cmd: command to issue
4647 * @arg: pointer to a struct ifreq in user space
4649 * Issue ioctl functions to devices. This is normally called by the
4650 * user space syscall interfaces but can sometimes be useful for
4651 * other purposes. The return value is the return from the syscall if
4652 * positive or a negative errno code on error.
4655 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4661 /* One special case: SIOCGIFCONF takes ifconf argument
4662 and requires shared lock, because it sleeps writing
4666 if (cmd
== SIOCGIFCONF
) {
4668 ret
= dev_ifconf(net
, (char __user
*) arg
);
4672 if (cmd
== SIOCGIFNAME
)
4673 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4675 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4678 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4680 colon
= strchr(ifr
.ifr_name
, ':');
4685 * See which interface the caller is talking about.
4690 * These ioctl calls:
4691 * - can be done by all.
4692 * - atomic and do not require locking.
4703 dev_load(net
, ifr
.ifr_name
);
4705 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4710 if (copy_to_user(arg
, &ifr
,
4711 sizeof(struct ifreq
)))
4717 dev_load(net
, ifr
.ifr_name
);
4719 ret
= dev_ethtool(net
, &ifr
);
4724 if (copy_to_user(arg
, &ifr
,
4725 sizeof(struct ifreq
)))
4731 * These ioctl calls:
4732 * - require superuser power.
4733 * - require strict serialization.
4739 if (!capable(CAP_NET_ADMIN
))
4741 dev_load(net
, ifr
.ifr_name
);
4743 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4748 if (copy_to_user(arg
, &ifr
,
4749 sizeof(struct ifreq
)))
4755 * These ioctl calls:
4756 * - require superuser power.
4757 * - require strict serialization.
4758 * - do not return a value
4768 case SIOCSIFHWBROADCAST
:
4771 case SIOCBONDENSLAVE
:
4772 case SIOCBONDRELEASE
:
4773 case SIOCBONDSETHWADDR
:
4774 case SIOCBONDCHANGEACTIVE
:
4778 if (!capable(CAP_NET_ADMIN
))
4781 case SIOCBONDSLAVEINFOQUERY
:
4782 case SIOCBONDINFOQUERY
:
4783 dev_load(net
, ifr
.ifr_name
);
4785 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4790 /* Get the per device memory space. We can add this but
4791 * currently do not support it */
4793 /* Set the per device memory buffer space.
4794 * Not applicable in our case */
4799 * Unknown or private ioctl.
4802 if (cmd
== SIOCWANDEV
||
4803 (cmd
>= SIOCDEVPRIVATE
&&
4804 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4805 dev_load(net
, ifr
.ifr_name
);
4807 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4809 if (!ret
&& copy_to_user(arg
, &ifr
,
4810 sizeof(struct ifreq
)))
4814 /* Take care of Wireless Extensions */
4815 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4816 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4823 * dev_new_index - allocate an ifindex
4824 * @net: the applicable net namespace
4826 * Returns a suitable unique value for a new device interface
4827 * number. The caller must hold the rtnl semaphore or the
4828 * dev_base_lock to be sure it remains unique.
4830 static int dev_new_index(struct net
*net
)
4836 if (!__dev_get_by_index(net
, ifindex
))
4841 /* Delayed registration/unregisteration */
4842 static LIST_HEAD(net_todo_list
);
4844 static void net_set_todo(struct net_device
*dev
)
4846 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4849 static void rollback_registered_many(struct list_head
*head
)
4851 struct net_device
*dev
, *tmp
;
4853 BUG_ON(dev_boot_phase
);
4856 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4857 /* Some devices call without registering
4858 * for initialization unwind. Remove those
4859 * devices and proceed with the remaining.
4861 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4862 pr_debug("unregister_netdevice: device %s/%p never "
4863 "was registered\n", dev
->name
, dev
);
4866 list_del(&dev
->unreg_list
);
4870 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
4872 /* If device is running, close it first. */
4875 /* And unlink it from device chain. */
4876 unlist_netdevice(dev
);
4878 dev
->reg_state
= NETREG_UNREGISTERING
;
4883 list_for_each_entry(dev
, head
, unreg_list
) {
4884 /* Shutdown queueing discipline. */
4888 /* Notify protocols, that we are about to destroy
4889 this device. They should clean all the things.
4891 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4893 if (!dev
->rtnl_link_ops
||
4894 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
4895 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
4898 * Flush the unicast and multicast chains
4903 if (dev
->netdev_ops
->ndo_uninit
)
4904 dev
->netdev_ops
->ndo_uninit(dev
);
4906 /* Notifier chain MUST detach us from master device. */
4907 WARN_ON(dev
->master
);
4909 /* Remove entries from kobject tree */
4910 netdev_unregister_kobject(dev
);
4913 /* Process any work delayed until the end of the batch */
4914 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
4915 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
4919 list_for_each_entry(dev
, head
, unreg_list
)
4923 static void rollback_registered(struct net_device
*dev
)
4927 list_add(&dev
->unreg_list
, &single
);
4928 rollback_registered_many(&single
);
4931 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
4932 struct netdev_queue
*dev_queue
,
4935 spin_lock_init(&dev_queue
->_xmit_lock
);
4936 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
4937 dev_queue
->xmit_lock_owner
= -1;
4940 static void netdev_init_queue_locks(struct net_device
*dev
)
4942 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
4945 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
4947 /* Fix illegal SG+CSUM combinations. */
4948 if ((features
& NETIF_F_SG
) &&
4949 !(features
& NETIF_F_ALL_CSUM
)) {
4951 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
4952 "checksum feature.\n", name
);
4953 features
&= ~NETIF_F_SG
;
4956 /* TSO requires that SG is present as well. */
4957 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
4959 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
4960 "SG feature.\n", name
);
4961 features
&= ~NETIF_F_TSO
;
4964 if (features
& NETIF_F_UFO
) {
4965 if (!(features
& NETIF_F_GEN_CSUM
)) {
4967 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4968 "since no NETIF_F_HW_CSUM feature.\n",
4970 features
&= ~NETIF_F_UFO
;
4973 if (!(features
& NETIF_F_SG
)) {
4975 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
4976 "since no NETIF_F_SG feature.\n", name
);
4977 features
&= ~NETIF_F_UFO
;
4983 EXPORT_SYMBOL(netdev_fix_features
);
4986 * netif_stacked_transfer_operstate - transfer operstate
4987 * @rootdev: the root or lower level device to transfer state from
4988 * @dev: the device to transfer operstate to
4990 * Transfer operational state from root to device. This is normally
4991 * called when a stacking relationship exists between the root
4992 * device and the device(a leaf device).
4994 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
4995 struct net_device
*dev
)
4997 if (rootdev
->operstate
== IF_OPER_DORMANT
)
4998 netif_dormant_on(dev
);
5000 netif_dormant_off(dev
);
5002 if (netif_carrier_ok(rootdev
)) {
5003 if (!netif_carrier_ok(dev
))
5004 netif_carrier_on(dev
);
5006 if (netif_carrier_ok(dev
))
5007 netif_carrier_off(dev
);
5010 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5012 static int netif_alloc_rx_queues(struct net_device
*dev
)
5015 unsigned int i
, count
= dev
->num_rx_queues
;
5018 struct netdev_rx_queue
*rx
;
5020 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5022 pr_err("netdev: Unable to allocate %u rx queues.\n",
5029 * Set a pointer to first element in the array which holds the
5032 for (i
= 0; i
< count
; i
++)
5040 * register_netdevice - register a network device
5041 * @dev: device to register
5043 * Take a completed network device structure and add it to the kernel
5044 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5045 * chain. 0 is returned on success. A negative errno code is returned
5046 * on a failure to set up the device, or if the name is a duplicate.
5048 * Callers must hold the rtnl semaphore. You may want
5049 * register_netdev() instead of this.
5052 * The locking appears insufficient to guarantee two parallel registers
5053 * will not get the same name.
5056 int register_netdevice(struct net_device
*dev
)
5059 struct net
*net
= dev_net(dev
);
5061 BUG_ON(dev_boot_phase
);
5066 /* When net_device's are persistent, this will be fatal. */
5067 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5070 spin_lock_init(&dev
->addr_list_lock
);
5071 netdev_set_addr_lockdep_class(dev
);
5072 netdev_init_queue_locks(dev
);
5076 ret
= netif_alloc_rx_queues(dev
);
5080 /* Init, if this function is available */
5081 if (dev
->netdev_ops
->ndo_init
) {
5082 ret
= dev
->netdev_ops
->ndo_init(dev
);
5090 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
5094 dev
->ifindex
= dev_new_index(net
);
5095 if (dev
->iflink
== -1)
5096 dev
->iflink
= dev
->ifindex
;
5098 /* Fix illegal checksum combinations */
5099 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
5100 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5101 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5103 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5106 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5107 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5108 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5110 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5113 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5115 /* Enable software GSO if SG is supported. */
5116 if (dev
->features
& NETIF_F_SG
)
5117 dev
->features
|= NETIF_F_GSO
;
5119 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5120 * vlan_dev_init() will do the dev->features check, so these features
5121 * are enabled only if supported by underlying device.
5123 dev
->vlan_features
|= (NETIF_F_GRO
| NETIF_F_HIGHDMA
);
5125 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5126 ret
= notifier_to_errno(ret
);
5130 ret
= netdev_register_kobject(dev
);
5133 dev
->reg_state
= NETREG_REGISTERED
;
5136 * Default initial state at registry is that the
5137 * device is present.
5140 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5142 dev_init_scheduler(dev
);
5144 list_netdevice(dev
);
5146 /* Notify protocols, that a new device appeared. */
5147 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5148 ret
= notifier_to_errno(ret
);
5150 rollback_registered(dev
);
5151 dev
->reg_state
= NETREG_UNREGISTERED
;
5154 * Prevent userspace races by waiting until the network
5155 * device is fully setup before sending notifications.
5157 if (!dev
->rtnl_link_ops
||
5158 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5159 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5165 if (dev
->netdev_ops
->ndo_uninit
)
5166 dev
->netdev_ops
->ndo_uninit(dev
);
5169 EXPORT_SYMBOL(register_netdevice
);
5172 * init_dummy_netdev - init a dummy network device for NAPI
5173 * @dev: device to init
5175 * This takes a network device structure and initialize the minimum
5176 * amount of fields so it can be used to schedule NAPI polls without
5177 * registering a full blown interface. This is to be used by drivers
5178 * that need to tie several hardware interfaces to a single NAPI
5179 * poll scheduler due to HW limitations.
5181 int init_dummy_netdev(struct net_device
*dev
)
5183 /* Clear everything. Note we don't initialize spinlocks
5184 * are they aren't supposed to be taken by any of the
5185 * NAPI code and this dummy netdev is supposed to be
5186 * only ever used for NAPI polls
5188 memset(dev
, 0, sizeof(struct net_device
));
5190 /* make sure we BUG if trying to hit standard
5191 * register/unregister code path
5193 dev
->reg_state
= NETREG_DUMMY
;
5195 /* initialize the ref count */
5196 atomic_set(&dev
->refcnt
, 1);
5198 /* NAPI wants this */
5199 INIT_LIST_HEAD(&dev
->napi_list
);
5201 /* a dummy interface is started by default */
5202 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5203 set_bit(__LINK_STATE_START
, &dev
->state
);
5207 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5211 * register_netdev - register a network device
5212 * @dev: device to register
5214 * Take a completed network device structure and add it to the kernel
5215 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5216 * chain. 0 is returned on success. A negative errno code is returned
5217 * on a failure to set up the device, or if the name is a duplicate.
5219 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5220 * and expands the device name if you passed a format string to
5223 int register_netdev(struct net_device
*dev
)
5230 * If the name is a format string the caller wants us to do a
5233 if (strchr(dev
->name
, '%')) {
5234 err
= dev_alloc_name(dev
, dev
->name
);
5239 err
= register_netdevice(dev
);
5244 EXPORT_SYMBOL(register_netdev
);
5247 * netdev_wait_allrefs - wait until all references are gone.
5249 * This is called when unregistering network devices.
5251 * Any protocol or device that holds a reference should register
5252 * for netdevice notification, and cleanup and put back the
5253 * reference if they receive an UNREGISTER event.
5254 * We can get stuck here if buggy protocols don't correctly
5257 static void netdev_wait_allrefs(struct net_device
*dev
)
5259 unsigned long rebroadcast_time
, warning_time
;
5261 linkwatch_forget_dev(dev
);
5263 rebroadcast_time
= warning_time
= jiffies
;
5264 while (atomic_read(&dev
->refcnt
) != 0) {
5265 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5268 /* Rebroadcast unregister notification */
5269 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5270 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5271 * should have already handle it the first time */
5273 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5275 /* We must not have linkwatch events
5276 * pending on unregister. If this
5277 * happens, we simply run the queue
5278 * unscheduled, resulting in a noop
5281 linkwatch_run_queue();
5286 rebroadcast_time
= jiffies
;
5291 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5292 printk(KERN_EMERG
"unregister_netdevice: "
5293 "waiting for %s to become free. Usage "
5295 dev
->name
, atomic_read(&dev
->refcnt
));
5296 warning_time
= jiffies
;
5305 * register_netdevice(x1);
5306 * register_netdevice(x2);
5308 * unregister_netdevice(y1);
5309 * unregister_netdevice(y2);
5315 * We are invoked by rtnl_unlock().
5316 * This allows us to deal with problems:
5317 * 1) We can delete sysfs objects which invoke hotplug
5318 * without deadlocking with linkwatch via keventd.
5319 * 2) Since we run with the RTNL semaphore not held, we can sleep
5320 * safely in order to wait for the netdev refcnt to drop to zero.
5322 * We must not return until all unregister events added during
5323 * the interval the lock was held have been completed.
5325 void netdev_run_todo(void)
5327 struct list_head list
;
5329 /* Snapshot list, allow later requests */
5330 list_replace_init(&net_todo_list
, &list
);
5334 while (!list_empty(&list
)) {
5335 struct net_device
*dev
5336 = list_first_entry(&list
, struct net_device
, todo_list
);
5337 list_del(&dev
->todo_list
);
5339 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5340 printk(KERN_ERR
"network todo '%s' but state %d\n",
5341 dev
->name
, dev
->reg_state
);
5346 dev
->reg_state
= NETREG_UNREGISTERED
;
5348 on_each_cpu(flush_backlog
, dev
, 1);
5350 netdev_wait_allrefs(dev
);
5353 BUG_ON(atomic_read(&dev
->refcnt
));
5354 WARN_ON(rcu_dereference_raw(dev
->ip_ptr
));
5355 WARN_ON(dev
->ip6_ptr
);
5356 WARN_ON(dev
->dn_ptr
);
5358 if (dev
->destructor
)
5359 dev
->destructor(dev
);
5361 /* Free network device */
5362 kobject_put(&dev
->dev
.kobj
);
5367 * dev_txq_stats_fold - fold tx_queues stats
5368 * @dev: device to get statistics from
5369 * @stats: struct rtnl_link_stats64 to hold results
5371 void dev_txq_stats_fold(const struct net_device
*dev
,
5372 struct rtnl_link_stats64
*stats
)
5374 u64 tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5376 struct netdev_queue
*txq
;
5378 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5379 txq
= netdev_get_tx_queue(dev
, i
);
5380 spin_lock_bh(&txq
->_xmit_lock
);
5381 tx_bytes
+= txq
->tx_bytes
;
5382 tx_packets
+= txq
->tx_packets
;
5383 tx_dropped
+= txq
->tx_dropped
;
5384 spin_unlock_bh(&txq
->_xmit_lock
);
5386 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5387 stats
->tx_bytes
= tx_bytes
;
5388 stats
->tx_packets
= tx_packets
;
5389 stats
->tx_dropped
= tx_dropped
;
5392 EXPORT_SYMBOL(dev_txq_stats_fold
);
5394 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5395 * fields in the same order, with only the type differing.
5397 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5398 const struct net_device_stats
*netdev_stats
)
5400 #if BITS_PER_LONG == 64
5401 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5402 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5404 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5405 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5406 u64
*dst
= (u64
*)stats64
;
5408 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5409 sizeof(*stats64
) / sizeof(u64
));
5410 for (i
= 0; i
< n
; i
++)
5416 * dev_get_stats - get network device statistics
5417 * @dev: device to get statistics from
5418 * @storage: place to store stats
5420 * Get network statistics from device. Return @storage.
5421 * The device driver may provide its own method by setting
5422 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5423 * otherwise the internal statistics structure is used.
5425 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5426 struct rtnl_link_stats64
*storage
)
5428 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5430 if (ops
->ndo_get_stats64
) {
5431 memset(storage
, 0, sizeof(*storage
));
5432 ops
->ndo_get_stats64(dev
, storage
);
5433 } else if (ops
->ndo_get_stats
) {
5434 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5436 netdev_stats_to_stats64(storage
, &dev
->stats
);
5437 dev_txq_stats_fold(dev
, storage
);
5439 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5442 EXPORT_SYMBOL(dev_get_stats
);
5444 static void netdev_init_one_queue(struct net_device
*dev
,
5445 struct netdev_queue
*queue
,
5451 static void netdev_init_queues(struct net_device
*dev
)
5453 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5454 spin_lock_init(&dev
->tx_global_lock
);
5457 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5459 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5461 #ifdef CONFIG_NET_CLS_ACT
5464 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5467 netdev_init_one_queue(dev
, queue
, NULL
);
5468 __netdev_init_queue_locks_one(dev
, queue
, NULL
);
5469 queue
->qdisc
= &noop_qdisc
;
5470 queue
->qdisc_sleeping
= &noop_qdisc
;
5471 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5477 * alloc_netdev_mq - allocate network device
5478 * @sizeof_priv: size of private data to allocate space for
5479 * @name: device name format string
5480 * @setup: callback to initialize device
5481 * @queue_count: the number of subqueues to allocate
5483 * Allocates a struct net_device with private data area for driver use
5484 * and performs basic initialization. Also allocates subquue structs
5485 * for each queue on the device at the end of the netdevice.
5487 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
5488 void (*setup
)(struct net_device
*), unsigned int queue_count
)
5490 struct netdev_queue
*tx
;
5491 struct net_device
*dev
;
5493 struct net_device
*p
;
5495 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5497 alloc_size
= sizeof(struct net_device
);
5499 /* ensure 32-byte alignment of private area */
5500 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5501 alloc_size
+= sizeof_priv
;
5503 /* ensure 32-byte alignment of whole construct */
5504 alloc_size
+= NETDEV_ALIGN
- 1;
5506 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5508 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5512 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5514 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
5520 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5521 dev
->padded
= (char *)dev
- (char *)p
;
5523 if (dev_addr_init(dev
))
5529 dev_net_set(dev
, &init_net
);
5532 dev
->num_tx_queues
= queue_count
;
5533 dev
->real_num_tx_queues
= queue_count
;
5536 dev
->num_rx_queues
= queue_count
;
5537 dev
->real_num_rx_queues
= queue_count
;
5540 dev
->gso_max_size
= GSO_MAX_SIZE
;
5542 netdev_init_queues(dev
);
5544 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5545 dev
->ethtool_ntuple_list
.count
= 0;
5546 INIT_LIST_HEAD(&dev
->napi_list
);
5547 INIT_LIST_HEAD(&dev
->unreg_list
);
5548 INIT_LIST_HEAD(&dev
->link_watch_list
);
5549 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5551 strcpy(dev
->name
, name
);
5560 EXPORT_SYMBOL(alloc_netdev_mq
);
5563 * free_netdev - free network device
5566 * This function does the last stage of destroying an allocated device
5567 * interface. The reference to the device object is released.
5568 * If this is the last reference then it will be freed.
5570 void free_netdev(struct net_device
*dev
)
5572 struct napi_struct
*p
, *n
;
5574 release_net(dev_net(dev
));
5578 kfree(rcu_dereference_raw(dev
->ingress_queue
));
5580 /* Flush device addresses */
5581 dev_addr_flush(dev
);
5583 /* Clear ethtool n-tuple list */
5584 ethtool_ntuple_flush(dev
);
5586 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5589 /* Compatibility with error handling in drivers */
5590 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5591 kfree((char *)dev
- dev
->padded
);
5595 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5596 dev
->reg_state
= NETREG_RELEASED
;
5598 /* will free via device release */
5599 put_device(&dev
->dev
);
5601 EXPORT_SYMBOL(free_netdev
);
5604 * synchronize_net - Synchronize with packet receive processing
5606 * Wait for packets currently being received to be done.
5607 * Does not block later packets from starting.
5609 void synchronize_net(void)
5614 EXPORT_SYMBOL(synchronize_net
);
5617 * unregister_netdevice_queue - remove device from the kernel
5621 * This function shuts down a device interface and removes it
5622 * from the kernel tables.
5623 * If head not NULL, device is queued to be unregistered later.
5625 * Callers must hold the rtnl semaphore. You may want
5626 * unregister_netdev() instead of this.
5629 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5634 list_move_tail(&dev
->unreg_list
, head
);
5636 rollback_registered(dev
);
5637 /* Finish processing unregister after unlock */
5641 EXPORT_SYMBOL(unregister_netdevice_queue
);
5644 * unregister_netdevice_many - unregister many devices
5645 * @head: list of devices
5647 void unregister_netdevice_many(struct list_head
*head
)
5649 struct net_device
*dev
;
5651 if (!list_empty(head
)) {
5652 rollback_registered_many(head
);
5653 list_for_each_entry(dev
, head
, unreg_list
)
5657 EXPORT_SYMBOL(unregister_netdevice_many
);
5660 * unregister_netdev - remove device from the kernel
5663 * This function shuts down a device interface and removes it
5664 * from the kernel tables.
5666 * This is just a wrapper for unregister_netdevice that takes
5667 * the rtnl semaphore. In general you want to use this and not
5668 * unregister_netdevice.
5670 void unregister_netdev(struct net_device
*dev
)
5673 unregister_netdevice(dev
);
5676 EXPORT_SYMBOL(unregister_netdev
);
5679 * dev_change_net_namespace - move device to different nethost namespace
5681 * @net: network namespace
5682 * @pat: If not NULL name pattern to try if the current device name
5683 * is already taken in the destination network namespace.
5685 * This function shuts down a device interface and moves it
5686 * to a new network namespace. On success 0 is returned, on
5687 * a failure a netagive errno code is returned.
5689 * Callers must hold the rtnl semaphore.
5692 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5698 /* Don't allow namespace local devices to be moved. */
5700 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5703 /* Ensure the device has been registrered */
5705 if (dev
->reg_state
!= NETREG_REGISTERED
)
5708 /* Get out if there is nothing todo */
5710 if (net_eq(dev_net(dev
), net
))
5713 /* Pick the destination device name, and ensure
5714 * we can use it in the destination network namespace.
5717 if (__dev_get_by_name(net
, dev
->name
)) {
5718 /* We get here if we can't use the current device name */
5721 if (dev_get_valid_name(dev
, pat
, 1))
5726 * And now a mini version of register_netdevice unregister_netdevice.
5729 /* If device is running close it first. */
5732 /* And unlink it from device chain */
5734 unlist_netdevice(dev
);
5738 /* Shutdown queueing discipline. */
5741 /* Notify protocols, that we are about to destroy
5742 this device. They should clean all the things.
5744 Note that dev->reg_state stays at NETREG_REGISTERED.
5745 This is wanted because this way 8021q and macvlan know
5746 the device is just moving and can keep their slaves up.
5748 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5749 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5752 * Flush the unicast and multicast chains
5757 /* Actually switch the network namespace */
5758 dev_net_set(dev
, net
);
5760 /* If there is an ifindex conflict assign a new one */
5761 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5762 int iflink
= (dev
->iflink
== dev
->ifindex
);
5763 dev
->ifindex
= dev_new_index(net
);
5765 dev
->iflink
= dev
->ifindex
;
5768 /* Fixup kobjects */
5769 err
= device_rename(&dev
->dev
, dev
->name
);
5772 /* Add the device back in the hashes */
5773 list_netdevice(dev
);
5775 /* Notify protocols, that a new device appeared. */
5776 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5779 * Prevent userspace races by waiting until the network
5780 * device is fully setup before sending notifications.
5782 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5789 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5791 static int dev_cpu_callback(struct notifier_block
*nfb
,
5792 unsigned long action
,
5795 struct sk_buff
**list_skb
;
5796 struct sk_buff
*skb
;
5797 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5798 struct softnet_data
*sd
, *oldsd
;
5800 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5803 local_irq_disable();
5804 cpu
= smp_processor_id();
5805 sd
= &per_cpu(softnet_data
, cpu
);
5806 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5808 /* Find end of our completion_queue. */
5809 list_skb
= &sd
->completion_queue
;
5811 list_skb
= &(*list_skb
)->next
;
5812 /* Append completion queue from offline CPU. */
5813 *list_skb
= oldsd
->completion_queue
;
5814 oldsd
->completion_queue
= NULL
;
5816 /* Append output queue from offline CPU. */
5817 if (oldsd
->output_queue
) {
5818 *sd
->output_queue_tailp
= oldsd
->output_queue
;
5819 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
5820 oldsd
->output_queue
= NULL
;
5821 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
5824 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5827 /* Process offline CPU's input_pkt_queue */
5828 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
5830 input_queue_head_incr(oldsd
);
5832 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
5834 input_queue_head_incr(oldsd
);
5842 * netdev_increment_features - increment feature set by one
5843 * @all: current feature set
5844 * @one: new feature set
5845 * @mask: mask feature set
5847 * Computes a new feature set after adding a device with feature set
5848 * @one to the master device with current feature set @all. Will not
5849 * enable anything that is off in @mask. Returns the new feature set.
5851 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
5854 /* If device needs checksumming, downgrade to it. */
5855 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
5856 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
5857 else if (mask
& NETIF_F_ALL_CSUM
) {
5858 /* If one device supports v4/v6 checksumming, set for all. */
5859 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
5860 !(all
& NETIF_F_GEN_CSUM
)) {
5861 all
&= ~NETIF_F_ALL_CSUM
;
5862 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
5865 /* If one device supports hw checksumming, set for all. */
5866 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
5867 all
&= ~NETIF_F_ALL_CSUM
;
5868 all
|= NETIF_F_HW_CSUM
;
5872 one
|= NETIF_F_ALL_CSUM
;
5874 one
|= all
& NETIF_F_ONE_FOR_ALL
;
5875 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
5876 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
5880 EXPORT_SYMBOL(netdev_increment_features
);
5882 static struct hlist_head
*netdev_create_hash(void)
5885 struct hlist_head
*hash
;
5887 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
5889 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
5890 INIT_HLIST_HEAD(&hash
[i
]);
5895 /* Initialize per network namespace state */
5896 static int __net_init
netdev_init(struct net
*net
)
5898 INIT_LIST_HEAD(&net
->dev_base_head
);
5900 net
->dev_name_head
= netdev_create_hash();
5901 if (net
->dev_name_head
== NULL
)
5904 net
->dev_index_head
= netdev_create_hash();
5905 if (net
->dev_index_head
== NULL
)
5911 kfree(net
->dev_name_head
);
5917 * netdev_drivername - network driver for the device
5918 * @dev: network device
5919 * @buffer: buffer for resulting name
5920 * @len: size of buffer
5922 * Determine network driver for device.
5924 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
5926 const struct device_driver
*driver
;
5927 const struct device
*parent
;
5929 if (len
<= 0 || !buffer
)
5933 parent
= dev
->dev
.parent
;
5938 driver
= parent
->driver
;
5939 if (driver
&& driver
->name
)
5940 strlcpy(buffer
, driver
->name
, len
);
5944 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
5945 struct va_format
*vaf
)
5949 if (dev
&& dev
->dev
.parent
)
5950 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
5951 netdev_name(dev
), vaf
);
5953 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
5955 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
5960 int netdev_printk(const char *level
, const struct net_device
*dev
,
5961 const char *format
, ...)
5963 struct va_format vaf
;
5967 va_start(args
, format
);
5972 r
= __netdev_printk(level
, dev
, &vaf
);
5977 EXPORT_SYMBOL(netdev_printk
);
5979 #define define_netdev_printk_level(func, level) \
5980 int func(const struct net_device *dev, const char *fmt, ...) \
5983 struct va_format vaf; \
5986 va_start(args, fmt); \
5991 r = __netdev_printk(level, dev, &vaf); \
5996 EXPORT_SYMBOL(func);
5998 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
5999 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6000 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6001 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6002 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6003 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6004 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6006 static void __net_exit
netdev_exit(struct net
*net
)
6008 kfree(net
->dev_name_head
);
6009 kfree(net
->dev_index_head
);
6012 static struct pernet_operations __net_initdata netdev_net_ops
= {
6013 .init
= netdev_init
,
6014 .exit
= netdev_exit
,
6017 static void __net_exit
default_device_exit(struct net
*net
)
6019 struct net_device
*dev
, *aux
;
6021 * Push all migratable network devices back to the
6022 * initial network namespace
6025 for_each_netdev_safe(net
, dev
, aux
) {
6027 char fb_name
[IFNAMSIZ
];
6029 /* Ignore unmoveable devices (i.e. loopback) */
6030 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6033 /* Leave virtual devices for the generic cleanup */
6034 if (dev
->rtnl_link_ops
)
6037 /* Push remaing network devices to init_net */
6038 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6039 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6041 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
6042 __func__
, dev
->name
, err
);
6049 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6051 /* At exit all network devices most be removed from a network
6052 * namespace. Do this in the reverse order of registeration.
6053 * Do this across as many network namespaces as possible to
6054 * improve batching efficiency.
6056 struct net_device
*dev
;
6058 LIST_HEAD(dev_kill_list
);
6061 list_for_each_entry(net
, net_list
, exit_list
) {
6062 for_each_netdev_reverse(net
, dev
) {
6063 if (dev
->rtnl_link_ops
)
6064 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6066 unregister_netdevice_queue(dev
, &dev_kill_list
);
6069 unregister_netdevice_many(&dev_kill_list
);
6073 static struct pernet_operations __net_initdata default_device_ops
= {
6074 .exit
= default_device_exit
,
6075 .exit_batch
= default_device_exit_batch
,
6079 * Initialize the DEV module. At boot time this walks the device list and
6080 * unhooks any devices that fail to initialise (normally hardware not
6081 * present) and leaves us with a valid list of present and active devices.
6086 * This is called single threaded during boot, so no need
6087 * to take the rtnl semaphore.
6089 static int __init
net_dev_init(void)
6091 int i
, rc
= -ENOMEM
;
6093 BUG_ON(!dev_boot_phase
);
6095 if (dev_proc_init())
6098 if (netdev_kobject_init())
6101 INIT_LIST_HEAD(&ptype_all
);
6102 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6103 INIT_LIST_HEAD(&ptype_base
[i
]);
6105 if (register_pernet_subsys(&netdev_net_ops
))
6109 * Initialise the packet receive queues.
6112 for_each_possible_cpu(i
) {
6113 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6115 memset(sd
, 0, sizeof(*sd
));
6116 skb_queue_head_init(&sd
->input_pkt_queue
);
6117 skb_queue_head_init(&sd
->process_queue
);
6118 sd
->completion_queue
= NULL
;
6119 INIT_LIST_HEAD(&sd
->poll_list
);
6120 sd
->output_queue
= NULL
;
6121 sd
->output_queue_tailp
= &sd
->output_queue
;
6123 sd
->csd
.func
= rps_trigger_softirq
;
6129 sd
->backlog
.poll
= process_backlog
;
6130 sd
->backlog
.weight
= weight_p
;
6131 sd
->backlog
.gro_list
= NULL
;
6132 sd
->backlog
.gro_count
= 0;
6137 /* The loopback device is special if any other network devices
6138 * is present in a network namespace the loopback device must
6139 * be present. Since we now dynamically allocate and free the
6140 * loopback device ensure this invariant is maintained by
6141 * keeping the loopback device as the first device on the
6142 * list of network devices. Ensuring the loopback devices
6143 * is the first device that appears and the last network device
6146 if (register_pernet_device(&loopback_net_ops
))
6149 if (register_pernet_device(&default_device_ops
))
6152 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6153 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6155 hotcpu_notifier(dev_cpu_callback
, 0);
6163 subsys_initcall(net_dev_init
);
6165 static int __init
initialize_hashrnd(void)
6167 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6171 late_initcall_sync(initialize_hashrnd
);