2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 * The list of packet types we will receive (as opposed to discard)
146 * and the routines to invoke.
148 * Why 16. Because with 16 the only overlap we get on a hash of the
149 * low nibble of the protocol value is RARP/SNAP/X.25.
151 * NOTE: That is no longer true with the addition of VLAN tags. Not
152 * sure which should go first, but I bet it won't make much
153 * difference if we are running VLANs. The good news is that
154 * this protocol won't be in the list unless compiled in, so
155 * the average user (w/out VLANs) will not be adversely affected.
172 #define PTYPE_HASH_SIZE (16)
173 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
175 static DEFINE_SPINLOCK(ptype_lock
);
176 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
177 static struct list_head ptype_all __read_mostly
; /* Taps */
180 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
183 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
185 * Writers must hold the rtnl semaphore while they loop through the
186 * dev_base_head list, and hold dev_base_lock for writing when they do the
187 * actual updates. This allows pure readers to access the list even
188 * while a writer is preparing to update it.
190 * To put it another way, dev_base_lock is held for writing only to
191 * protect against pure readers; the rtnl semaphore provides the
192 * protection against other writers.
194 * See, for example usages, register_netdevice() and
195 * unregister_netdevice(), which must be called with the rtnl
198 DEFINE_RWLOCK(dev_base_lock
);
199 EXPORT_SYMBOL(dev_base_lock
);
201 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
203 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
204 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
207 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
209 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
212 static inline void rps_lock(struct softnet_data
*sd
)
215 spin_lock(&sd
->input_pkt_queue
.lock
);
219 static inline void rps_unlock(struct softnet_data
*sd
)
222 spin_unlock(&sd
->input_pkt_queue
.lock
);
226 /* Device list insertion */
227 static int list_netdevice(struct net_device
*dev
)
229 struct net
*net
= dev_net(dev
);
233 write_lock_bh(&dev_base_lock
);
234 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
235 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
236 hlist_add_head_rcu(&dev
->index_hlist
,
237 dev_index_hash(net
, dev
->ifindex
));
238 write_unlock_bh(&dev_base_lock
);
242 /* Device list removal
243 * caller must respect a RCU grace period before freeing/reusing dev
245 static void unlist_netdevice(struct net_device
*dev
)
249 /* Unlink dev from the device chain */
250 write_lock_bh(&dev_base_lock
);
251 list_del_rcu(&dev
->dev_list
);
252 hlist_del_rcu(&dev
->name_hlist
);
253 hlist_del_rcu(&dev
->index_hlist
);
254 write_unlock_bh(&dev_base_lock
);
261 static RAW_NOTIFIER_HEAD(netdev_chain
);
264 * Device drivers call our routines to queue packets here. We empty the
265 * queue in the local softnet handler.
268 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
269 EXPORT_PER_CPU_SYMBOL(softnet_data
);
271 #ifdef CONFIG_LOCKDEP
273 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
274 * according to dev->type
276 static const unsigned short netdev_lock_type
[] =
277 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
278 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
279 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
280 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
281 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
282 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
283 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
284 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
285 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
286 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
287 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
288 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
289 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
290 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
291 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
292 ARPHRD_VOID
, ARPHRD_NONE
};
294 static const char *const netdev_lock_name
[] =
295 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
296 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
297 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
298 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
299 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
300 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
301 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
302 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
303 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
304 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
305 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
306 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
307 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
308 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
309 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
310 "_xmit_VOID", "_xmit_NONE"};
312 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
313 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
315 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
319 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
320 if (netdev_lock_type
[i
] == dev_type
)
322 /* the last key is used by default */
323 return ARRAY_SIZE(netdev_lock_type
) - 1;
326 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
327 unsigned short dev_type
)
331 i
= netdev_lock_pos(dev_type
);
332 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
333 netdev_lock_name
[i
]);
336 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
340 i
= netdev_lock_pos(dev
->type
);
341 lockdep_set_class_and_name(&dev
->addr_list_lock
,
342 &netdev_addr_lock_key
[i
],
343 netdev_lock_name
[i
]);
346 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
347 unsigned short dev_type
)
350 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
355 /*******************************************************************************
357 Protocol management and registration routines
359 *******************************************************************************/
362 * Add a protocol ID to the list. Now that the input handler is
363 * smarter we can dispense with all the messy stuff that used to be
366 * BEWARE!!! Protocol handlers, mangling input packets,
367 * MUST BE last in hash buckets and checking protocol handlers
368 * MUST start from promiscuous ptype_all chain in net_bh.
369 * It is true now, do not change it.
370 * Explanation follows: if protocol handler, mangling packet, will
371 * be the first on list, it is not able to sense, that packet
372 * is cloned and should be copied-on-write, so that it will
373 * change it and subsequent readers will get broken packet.
377 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
379 if (pt
->type
== htons(ETH_P_ALL
))
382 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
386 * dev_add_pack - add packet handler
387 * @pt: packet type declaration
389 * Add a protocol handler to the networking stack. The passed &packet_type
390 * is linked into kernel lists and may not be freed until it has been
391 * removed from the kernel lists.
393 * This call does not sleep therefore it can not
394 * guarantee all CPU's that are in middle of receiving packets
395 * will see the new packet type (until the next received packet).
398 void dev_add_pack(struct packet_type
*pt
)
400 struct list_head
*head
= ptype_head(pt
);
402 spin_lock(&ptype_lock
);
403 list_add_rcu(&pt
->list
, head
);
404 spin_unlock(&ptype_lock
);
406 EXPORT_SYMBOL(dev_add_pack
);
409 * __dev_remove_pack - remove packet handler
410 * @pt: packet type declaration
412 * Remove a protocol handler that was previously added to the kernel
413 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
414 * from the kernel lists and can be freed or reused once this function
417 * The packet type might still be in use by receivers
418 * and must not be freed until after all the CPU's have gone
419 * through a quiescent state.
421 void __dev_remove_pack(struct packet_type
*pt
)
423 struct list_head
*head
= ptype_head(pt
);
424 struct packet_type
*pt1
;
426 spin_lock(&ptype_lock
);
428 list_for_each_entry(pt1
, head
, list
) {
430 list_del_rcu(&pt
->list
);
435 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
437 spin_unlock(&ptype_lock
);
439 EXPORT_SYMBOL(__dev_remove_pack
);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
450 * This call sleeps to guarantee that no CPU is looking at the packet
453 void dev_remove_pack(struct packet_type
*pt
)
455 __dev_remove_pack(pt
);
459 EXPORT_SYMBOL(dev_remove_pack
);
461 /******************************************************************************
463 Device Boot-time Settings Routines
465 *******************************************************************************/
467 /* Boot time configuration table */
468 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
471 * netdev_boot_setup_add - add new setup entry
472 * @name: name of the device
473 * @map: configured settings for the device
475 * Adds new setup entry to the dev_boot_setup list. The function
476 * returns 0 on error and 1 on success. This is a generic routine to
479 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
481 struct netdev_boot_setup
*s
;
485 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
486 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
487 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
488 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
489 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
494 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
498 * netdev_boot_setup_check - check boot time settings
499 * @dev: the netdevice
501 * Check boot time settings for the device.
502 * The found settings are set for the device to be used
503 * later in the device probing.
504 * Returns 0 if no settings found, 1 if they are.
506 int netdev_boot_setup_check(struct net_device
*dev
)
508 struct netdev_boot_setup
*s
= dev_boot_setup
;
511 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
512 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
513 !strcmp(dev
->name
, s
[i
].name
)) {
514 dev
->irq
= s
[i
].map
.irq
;
515 dev
->base_addr
= s
[i
].map
.base_addr
;
516 dev
->mem_start
= s
[i
].map
.mem_start
;
517 dev
->mem_end
= s
[i
].map
.mem_end
;
523 EXPORT_SYMBOL(netdev_boot_setup_check
);
527 * netdev_boot_base - get address from boot time settings
528 * @prefix: prefix for network device
529 * @unit: id for network device
531 * Check boot time settings for the base address of device.
532 * The found settings are set for the device to be used
533 * later in the device probing.
534 * Returns 0 if no settings found.
536 unsigned long netdev_boot_base(const char *prefix
, int unit
)
538 const struct netdev_boot_setup
*s
= dev_boot_setup
;
542 sprintf(name
, "%s%d", prefix
, unit
);
545 * If device already registered then return base of 1
546 * to indicate not to probe for this interface
548 if (__dev_get_by_name(&init_net
, name
))
551 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
552 if (!strcmp(name
, s
[i
].name
))
553 return s
[i
].map
.base_addr
;
558 * Saves at boot time configured settings for any netdevice.
560 int __init
netdev_boot_setup(char *str
)
565 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
570 memset(&map
, 0, sizeof(map
));
574 map
.base_addr
= ints
[2];
576 map
.mem_start
= ints
[3];
578 map
.mem_end
= ints
[4];
580 /* Add new entry to the list */
581 return netdev_boot_setup_add(str
, &map
);
584 __setup("netdev=", netdev_boot_setup
);
586 /*******************************************************************************
588 Device Interface Subroutines
590 *******************************************************************************/
593 * __dev_get_by_name - find a device by its name
594 * @net: the applicable net namespace
595 * @name: name to find
597 * Find an interface by name. Must be called under RTNL semaphore
598 * or @dev_base_lock. If the name is found a pointer to the device
599 * is returned. If the name is not found then %NULL is returned. The
600 * reference counters are not incremented so the caller must be
601 * careful with locks.
604 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
606 struct hlist_node
*p
;
607 struct net_device
*dev
;
608 struct hlist_head
*head
= dev_name_hash(net
, name
);
610 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
611 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
616 EXPORT_SYMBOL(__dev_get_by_name
);
619 * dev_get_by_name_rcu - find a device by its name
620 * @net: the applicable net namespace
621 * @name: name to find
623 * Find an interface by name.
624 * If the name is found a pointer to the device is returned.
625 * If the name is not found then %NULL is returned.
626 * The reference counters are not incremented so the caller must be
627 * careful with locks. The caller must hold RCU lock.
630 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
632 struct hlist_node
*p
;
633 struct net_device
*dev
;
634 struct hlist_head
*head
= dev_name_hash(net
, name
);
636 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
637 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
642 EXPORT_SYMBOL(dev_get_by_name_rcu
);
645 * dev_get_by_name - find a device by its name
646 * @net: the applicable net namespace
647 * @name: name to find
649 * Find an interface by name. This can be called from any
650 * context and does its own locking. The returned handle has
651 * the usage count incremented and the caller must use dev_put() to
652 * release it when it is no longer needed. %NULL is returned if no
653 * matching device is found.
656 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
658 struct net_device
*dev
;
661 dev
= dev_get_by_name_rcu(net
, name
);
667 EXPORT_SYMBOL(dev_get_by_name
);
670 * __dev_get_by_index - find a device by its ifindex
671 * @net: the applicable net namespace
672 * @ifindex: index of device
674 * Search for an interface by index. Returns %NULL if the device
675 * is not found or a pointer to the device. The device has not
676 * had its reference counter increased so the caller must be careful
677 * about locking. The caller must hold either the RTNL semaphore
681 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
683 struct hlist_node
*p
;
684 struct net_device
*dev
;
685 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
687 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
688 if (dev
->ifindex
== ifindex
)
693 EXPORT_SYMBOL(__dev_get_by_index
);
696 * dev_get_by_index_rcu - find a device by its ifindex
697 * @net: the applicable net namespace
698 * @ifindex: index of device
700 * Search for an interface by index. Returns %NULL if the device
701 * is not found or a pointer to the device. The device has not
702 * had its reference counter increased so the caller must be careful
703 * about locking. The caller must hold RCU lock.
706 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
708 struct hlist_node
*p
;
709 struct net_device
*dev
;
710 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
712 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
713 if (dev
->ifindex
== ifindex
)
718 EXPORT_SYMBOL(dev_get_by_index_rcu
);
722 * dev_get_by_index - find a device by its ifindex
723 * @net: the applicable net namespace
724 * @ifindex: index of device
726 * Search for an interface by index. Returns NULL if the device
727 * is not found or a pointer to the device. The device returned has
728 * had a reference added and the pointer is safe until the user calls
729 * dev_put to indicate they have finished with it.
732 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
734 struct net_device
*dev
;
737 dev
= dev_get_by_index_rcu(net
, ifindex
);
743 EXPORT_SYMBOL(dev_get_by_index
);
746 * dev_getbyhwaddr_rcu - find a device by its hardware address
747 * @net: the applicable net namespace
748 * @type: media type of device
749 * @ha: hardware address
751 * Search for an interface by MAC address. Returns NULL if the device
752 * is not found or a pointer to the device. The caller must hold RCU
753 * The returned device has not had its ref count increased
754 * and the caller must therefore be careful about locking
758 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
761 struct net_device
*dev
;
763 for_each_netdev_rcu(net
, dev
)
764 if (dev
->type
== type
&&
765 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
770 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
772 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
774 struct net_device
*dev
;
777 for_each_netdev(net
, dev
)
778 if (dev
->type
== type
)
783 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
785 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
787 struct net_device
*dev
, *ret
= NULL
;
790 for_each_netdev_rcu(net
, dev
)
791 if (dev
->type
== type
) {
799 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
802 * dev_get_by_flags_rcu - find any device with given flags
803 * @net: the applicable net namespace
804 * @if_flags: IFF_* values
805 * @mask: bitmask of bits in if_flags to check
807 * Search for any interface with the given flags. Returns NULL if a device
808 * is not found or a pointer to the device. Must be called inside
809 * rcu_read_lock(), and result refcount is unchanged.
812 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
815 struct net_device
*dev
, *ret
;
818 for_each_netdev_rcu(net
, dev
) {
819 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
826 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
829 * dev_valid_name - check if name is okay for network device
832 * Network device names need to be valid file names to
833 * to allow sysfs to work. We also disallow any kind of
836 int dev_valid_name(const char *name
)
840 if (strlen(name
) >= IFNAMSIZ
)
842 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
846 if (*name
== '/' || isspace(*name
))
852 EXPORT_SYMBOL(dev_valid_name
);
855 * __dev_alloc_name - allocate a name for a device
856 * @net: network namespace to allocate the device name in
857 * @name: name format string
858 * @buf: scratch buffer and result name string
860 * Passed a format string - eg "lt%d" it will try and find a suitable
861 * id. It scans list of devices to build up a free map, then chooses
862 * the first empty slot. The caller must hold the dev_base or rtnl lock
863 * while allocating the name and adding the device in order to avoid
865 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
866 * Returns the number of the unit assigned or a negative errno code.
869 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
873 const int max_netdevices
= 8*PAGE_SIZE
;
874 unsigned long *inuse
;
875 struct net_device
*d
;
877 p
= strnchr(name
, IFNAMSIZ
-1, '%');
880 * Verify the string as this thing may have come from
881 * the user. There must be either one "%d" and no other "%"
884 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
887 /* Use one page as a bit array of possible slots */
888 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
892 for_each_netdev(net
, d
) {
893 if (!sscanf(d
->name
, name
, &i
))
895 if (i
< 0 || i
>= max_netdevices
)
898 /* avoid cases where sscanf is not exact inverse of printf */
899 snprintf(buf
, IFNAMSIZ
, name
, i
);
900 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
904 i
= find_first_zero_bit(inuse
, max_netdevices
);
905 free_page((unsigned long) inuse
);
909 snprintf(buf
, IFNAMSIZ
, name
, i
);
910 if (!__dev_get_by_name(net
, buf
))
913 /* It is possible to run out of possible slots
914 * when the name is long and there isn't enough space left
915 * for the digits, or if all bits are used.
921 * dev_alloc_name - allocate a name for a device
923 * @name: name format string
925 * Passed a format string - eg "lt%d" it will try and find a suitable
926 * id. It scans list of devices to build up a free map, then chooses
927 * the first empty slot. The caller must hold the dev_base or rtnl lock
928 * while allocating the name and adding the device in order to avoid
930 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
931 * Returns the number of the unit assigned or a negative errno code.
934 int dev_alloc_name(struct net_device
*dev
, const char *name
)
940 BUG_ON(!dev_net(dev
));
942 ret
= __dev_alloc_name(net
, name
, buf
);
944 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
947 EXPORT_SYMBOL(dev_alloc_name
);
949 static int dev_get_valid_name(struct net_device
*dev
, const char *name
, bool fmt
)
953 BUG_ON(!dev_net(dev
));
956 if (!dev_valid_name(name
))
959 if (fmt
&& strchr(name
, '%'))
960 return dev_alloc_name(dev
, name
);
961 else if (__dev_get_by_name(net
, name
))
963 else if (dev
->name
!= name
)
964 strlcpy(dev
->name
, name
, IFNAMSIZ
);
970 * dev_change_name - change name of a device
972 * @newname: name (or format string) must be at least IFNAMSIZ
974 * Change name of a device, can pass format strings "eth%d".
977 int dev_change_name(struct net_device
*dev
, const char *newname
)
979 char oldname
[IFNAMSIZ
];
985 BUG_ON(!dev_net(dev
));
988 if (dev
->flags
& IFF_UP
)
991 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
994 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
996 err
= dev_get_valid_name(dev
, newname
, 1);
1001 ret
= device_rename(&dev
->dev
, dev
->name
);
1003 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1007 write_lock_bh(&dev_base_lock
);
1008 hlist_del(&dev
->name_hlist
);
1009 write_unlock_bh(&dev_base_lock
);
1013 write_lock_bh(&dev_base_lock
);
1014 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1015 write_unlock_bh(&dev_base_lock
);
1017 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1018 ret
= notifier_to_errno(ret
);
1021 /* err >= 0 after dev_alloc_name() or stores the first errno */
1024 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1028 "%s: name change rollback failed: %d.\n",
1037 * dev_set_alias - change ifalias of a device
1039 * @alias: name up to IFALIASZ
1040 * @len: limit of bytes to copy from info
1042 * Set ifalias for a device,
1044 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1048 if (len
>= IFALIASZ
)
1053 kfree(dev
->ifalias
);
1054 dev
->ifalias
= NULL
;
1059 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1063 strlcpy(dev
->ifalias
, alias
, len
+1);
1069 * netdev_features_change - device changes features
1070 * @dev: device to cause notification
1072 * Called to indicate a device has changed features.
1074 void netdev_features_change(struct net_device
*dev
)
1076 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1078 EXPORT_SYMBOL(netdev_features_change
);
1081 * netdev_state_change - device changes state
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed state. This function calls
1085 * the notifier chains for netdev_chain and sends a NEWLINK message
1086 * to the routing socket.
1088 void netdev_state_change(struct net_device
*dev
)
1090 if (dev
->flags
& IFF_UP
) {
1091 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1092 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1095 EXPORT_SYMBOL(netdev_state_change
);
1097 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1099 return call_netdevice_notifiers(event
, dev
);
1101 EXPORT_SYMBOL(netdev_bonding_change
);
1104 * dev_load - load a network module
1105 * @net: the applicable net namespace
1106 * @name: name of interface
1108 * If a network interface is not present and the process has suitable
1109 * privileges this function loads the module. If module loading is not
1110 * available in this kernel then it becomes a nop.
1113 void dev_load(struct net
*net
, const char *name
)
1115 struct net_device
*dev
;
1118 dev
= dev_get_by_name_rcu(net
, name
);
1121 if (!dev
&& capable(CAP_NET_ADMIN
))
1122 request_module("%s", name
);
1124 EXPORT_SYMBOL(dev_load
);
1126 static int __dev_open(struct net_device
*dev
)
1128 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1134 * Is it even present?
1136 if (!netif_device_present(dev
))
1139 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1140 ret
= notifier_to_errno(ret
);
1145 * Call device private open method
1147 set_bit(__LINK_STATE_START
, &dev
->state
);
1149 if (ops
->ndo_validate_addr
)
1150 ret
= ops
->ndo_validate_addr(dev
);
1152 if (!ret
&& ops
->ndo_open
)
1153 ret
= ops
->ndo_open(dev
);
1156 * If it went open OK then:
1160 clear_bit(__LINK_STATE_START
, &dev
->state
);
1165 dev
->flags
|= IFF_UP
;
1170 net_dmaengine_get();
1173 * Initialize multicasting status
1175 dev_set_rx_mode(dev
);
1178 * Wakeup transmit queue engine
1187 * dev_open - prepare an interface for use.
1188 * @dev: device to open
1190 * Takes a device from down to up state. The device's private open
1191 * function is invoked and then the multicast lists are loaded. Finally
1192 * the device is moved into the up state and a %NETDEV_UP message is
1193 * sent to the netdev notifier chain.
1195 * Calling this function on an active interface is a nop. On a failure
1196 * a negative errno code is returned.
1198 int dev_open(struct net_device
*dev
)
1205 if (dev
->flags
& IFF_UP
)
1211 ret
= __dev_open(dev
);
1216 * ... and announce new interface.
1218 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1219 call_netdevice_notifiers(NETDEV_UP
, dev
);
1223 EXPORT_SYMBOL(dev_open
);
1225 static int __dev_close_many(struct list_head
*head
)
1227 struct net_device
*dev
;
1232 list_for_each_entry(dev
, head
, unreg_list
) {
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, it
1242 * 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(). */
1250 dev_deactivate_many(head
);
1252 list_for_each_entry(dev
, head
, unreg_list
) {
1253 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1256 * Call the device specific close. This cannot fail.
1257 * Only if device is UP
1259 * We allow it to be called even after a DETACH hot-plug
1266 * Device is now down.
1269 dev
->flags
&= ~IFF_UP
;
1274 net_dmaengine_put();
1280 static int __dev_close(struct net_device
*dev
)
1284 list_add(&dev
->unreg_list
, &single
);
1285 return __dev_close_many(&single
);
1288 int dev_close_many(struct list_head
*head
)
1290 struct net_device
*dev
, *tmp
;
1291 LIST_HEAD(tmp_list
);
1293 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1294 if (!(dev
->flags
& IFF_UP
))
1295 list_move(&dev
->unreg_list
, &tmp_list
);
1297 __dev_close_many(head
);
1300 * Tell people we are down
1302 list_for_each_entry(dev
, head
, unreg_list
) {
1303 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1304 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1307 /* rollback_registered_many needs the complete original list */
1308 list_splice(&tmp_list
, head
);
1313 * dev_close - shutdown an interface.
1314 * @dev: device to shutdown
1316 * This function moves an active device into down state. A
1317 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1318 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1321 int dev_close(struct net_device
*dev
)
1325 list_add(&dev
->unreg_list
, &single
);
1326 dev_close_many(&single
);
1330 EXPORT_SYMBOL(dev_close
);
1334 * dev_disable_lro - disable Large Receive Offload on a device
1337 * Disable Large Receive Offload (LRO) on a net device. Must be
1338 * called under RTNL. This is needed if received packets may be
1339 * forwarded to another interface.
1341 void dev_disable_lro(struct net_device
*dev
)
1343 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1344 dev
->ethtool_ops
->set_flags
) {
1345 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1346 if (flags
& ETH_FLAG_LRO
) {
1347 flags
&= ~ETH_FLAG_LRO
;
1348 dev
->ethtool_ops
->set_flags(dev
, flags
);
1351 WARN_ON(dev
->features
& NETIF_F_LRO
);
1353 EXPORT_SYMBOL(dev_disable_lro
);
1356 static int dev_boot_phase
= 1;
1359 * Device change register/unregister. These are not inline or static
1360 * as we export them to the world.
1364 * register_netdevice_notifier - register a network notifier block
1367 * Register a notifier to be called when network device events occur.
1368 * The notifier passed is linked into the kernel structures and must
1369 * not be reused until it has been unregistered. A negative errno code
1370 * is returned on a failure.
1372 * When registered all registration and up events are replayed
1373 * to the new notifier to allow device to have a race free
1374 * view of the network device list.
1377 int register_netdevice_notifier(struct notifier_block
*nb
)
1379 struct net_device
*dev
;
1380 struct net_device
*last
;
1385 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1391 for_each_netdev(net
, dev
) {
1392 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1393 err
= notifier_to_errno(err
);
1397 if (!(dev
->flags
& IFF_UP
))
1400 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1411 for_each_netdev(net
, dev
) {
1415 if (dev
->flags
& IFF_UP
) {
1416 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1417 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1419 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1420 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1424 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1427 EXPORT_SYMBOL(register_netdevice_notifier
);
1430 * unregister_netdevice_notifier - unregister a network notifier block
1433 * Unregister a notifier previously registered by
1434 * register_netdevice_notifier(). The notifier is unlinked into the
1435 * kernel structures and may then be reused. A negative errno code
1436 * is returned on a failure.
1439 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1444 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1448 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1451 * call_netdevice_notifiers - call all network notifier blocks
1452 * @val: value passed unmodified to notifier function
1453 * @dev: net_device pointer passed unmodified to notifier function
1455 * Call all network notifier blocks. Parameters and return value
1456 * are as for raw_notifier_call_chain().
1459 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1462 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1465 /* When > 0 there are consumers of rx skb time stamps */
1466 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1468 void net_enable_timestamp(void)
1470 atomic_inc(&netstamp_needed
);
1472 EXPORT_SYMBOL(net_enable_timestamp
);
1474 void net_disable_timestamp(void)
1476 atomic_dec(&netstamp_needed
);
1478 EXPORT_SYMBOL(net_disable_timestamp
);
1480 static inline void net_timestamp_set(struct sk_buff
*skb
)
1482 if (atomic_read(&netstamp_needed
))
1483 __net_timestamp(skb
);
1485 skb
->tstamp
.tv64
= 0;
1488 static inline void net_timestamp_check(struct sk_buff
*skb
)
1490 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1491 __net_timestamp(skb
);
1495 * dev_forward_skb - loopback an skb to another netif
1497 * @dev: destination network device
1498 * @skb: buffer to forward
1501 * NET_RX_SUCCESS (no congestion)
1502 * NET_RX_DROP (packet was dropped, but freed)
1504 * dev_forward_skb can be used for injecting an skb from the
1505 * start_xmit function of one device into the receive queue
1506 * of another device.
1508 * The receiving device may be in another namespace, so
1509 * we have to clear all information in the skb that could
1510 * impact namespace isolation.
1512 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1517 if (unlikely(!(dev
->flags
& IFF_UP
) ||
1518 (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
)))) {
1519 atomic_long_inc(&dev
->rx_dropped
);
1523 skb_set_dev(skb
, dev
);
1524 skb
->tstamp
.tv64
= 0;
1525 skb
->pkt_type
= PACKET_HOST
;
1526 skb
->protocol
= eth_type_trans(skb
, dev
);
1527 return netif_rx(skb
);
1529 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1531 static inline int deliver_skb(struct sk_buff
*skb
,
1532 struct packet_type
*pt_prev
,
1533 struct net_device
*orig_dev
)
1535 atomic_inc(&skb
->users
);
1536 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1540 * Support routine. Sends outgoing frames to any network
1541 * taps currently in use.
1544 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1546 struct packet_type
*ptype
;
1547 struct sk_buff
*skb2
= NULL
;
1548 struct packet_type
*pt_prev
= NULL
;
1551 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1552 /* Never send packets back to the socket
1553 * they originated from - MvS (miquels@drinkel.ow.org)
1555 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1556 (ptype
->af_packet_priv
== NULL
||
1557 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1559 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1564 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1568 net_timestamp_set(skb2
);
1570 /* skb->nh should be correctly
1571 set by sender, so that the second statement is
1572 just protection against buggy protocols.
1574 skb_reset_mac_header(skb2
);
1576 if (skb_network_header(skb2
) < skb2
->data
||
1577 skb2
->network_header
> skb2
->tail
) {
1578 if (net_ratelimit())
1579 printk(KERN_CRIT
"protocol %04x is "
1581 ntohs(skb2
->protocol
),
1583 skb_reset_network_header(skb2
);
1586 skb2
->transport_header
= skb2
->network_header
;
1587 skb2
->pkt_type
= PACKET_OUTGOING
;
1592 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1597 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1598 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1600 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1604 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1607 if (dev
->reg_state
== NETREG_REGISTERED
) {
1610 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1615 if (txq
< dev
->real_num_tx_queues
)
1616 qdisc_reset_all_tx_gt(dev
, txq
);
1619 dev
->real_num_tx_queues
= txq
;
1622 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1626 * netif_set_real_num_rx_queues - set actual number of RX queues used
1627 * @dev: Network device
1628 * @rxq: Actual number of RX queues
1630 * This must be called either with the rtnl_lock held or before
1631 * registration of the net device. Returns 0 on success, or a
1632 * negative error code. If called before registration, it always
1635 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1639 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1642 if (dev
->reg_state
== NETREG_REGISTERED
) {
1645 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1651 dev
->real_num_rx_queues
= rxq
;
1654 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1657 static inline void __netif_reschedule(struct Qdisc
*q
)
1659 struct softnet_data
*sd
;
1660 unsigned long flags
;
1662 local_irq_save(flags
);
1663 sd
= &__get_cpu_var(softnet_data
);
1664 q
->next_sched
= NULL
;
1665 *sd
->output_queue_tailp
= q
;
1666 sd
->output_queue_tailp
= &q
->next_sched
;
1667 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1668 local_irq_restore(flags
);
1671 void __netif_schedule(struct Qdisc
*q
)
1673 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1674 __netif_reschedule(q
);
1676 EXPORT_SYMBOL(__netif_schedule
);
1678 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1680 if (atomic_dec_and_test(&skb
->users
)) {
1681 struct softnet_data
*sd
;
1682 unsigned long flags
;
1684 local_irq_save(flags
);
1685 sd
= &__get_cpu_var(softnet_data
);
1686 skb
->next
= sd
->completion_queue
;
1687 sd
->completion_queue
= skb
;
1688 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1689 local_irq_restore(flags
);
1692 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1694 void dev_kfree_skb_any(struct sk_buff
*skb
)
1696 if (in_irq() || irqs_disabled())
1697 dev_kfree_skb_irq(skb
);
1701 EXPORT_SYMBOL(dev_kfree_skb_any
);
1705 * netif_device_detach - mark device as removed
1706 * @dev: network device
1708 * Mark device as removed from system and therefore no longer available.
1710 void netif_device_detach(struct net_device
*dev
)
1712 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1713 netif_running(dev
)) {
1714 netif_tx_stop_all_queues(dev
);
1717 EXPORT_SYMBOL(netif_device_detach
);
1720 * netif_device_attach - mark device as attached
1721 * @dev: network device
1723 * Mark device as attached from system and restart if needed.
1725 void netif_device_attach(struct net_device
*dev
)
1727 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1728 netif_running(dev
)) {
1729 netif_tx_wake_all_queues(dev
);
1730 __netdev_watchdog_up(dev
);
1733 EXPORT_SYMBOL(netif_device_attach
);
1736 * skb_dev_set -- assign a new device to a buffer
1737 * @skb: buffer for the new device
1738 * @dev: network device
1740 * If an skb is owned by a device already, we have to reset
1741 * all data private to the namespace a device belongs to
1742 * before assigning it a new device.
1744 #ifdef CONFIG_NET_NS
1745 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1748 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1751 skb_init_secmark(skb
);
1755 skb
->ipvs_property
= 0;
1756 #ifdef CONFIG_NET_SCHED
1762 EXPORT_SYMBOL(skb_set_dev
);
1763 #endif /* CONFIG_NET_NS */
1766 * Invalidate hardware checksum when packet is to be mangled, and
1767 * complete checksum manually on outgoing path.
1769 int skb_checksum_help(struct sk_buff
*skb
)
1772 int ret
= 0, offset
;
1774 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1775 goto out_set_summed
;
1777 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1778 /* Let GSO fix up the checksum. */
1779 goto out_set_summed
;
1782 offset
= skb_checksum_start_offset(skb
);
1783 BUG_ON(offset
>= skb_headlen(skb
));
1784 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1786 offset
+= skb
->csum_offset
;
1787 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1789 if (skb_cloned(skb
) &&
1790 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1791 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1796 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1798 skb
->ip_summed
= CHECKSUM_NONE
;
1802 EXPORT_SYMBOL(skb_checksum_help
);
1805 * skb_gso_segment - Perform segmentation on skb.
1806 * @skb: buffer to segment
1807 * @features: features for the output path (see dev->features)
1809 * This function segments the given skb and returns a list of segments.
1811 * It may return NULL if the skb requires no segmentation. This is
1812 * only possible when GSO is used for verifying header integrity.
1814 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1816 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1817 struct packet_type
*ptype
;
1818 __be16 type
= skb
->protocol
;
1819 int vlan_depth
= ETH_HLEN
;
1822 while (type
== htons(ETH_P_8021Q
)) {
1823 struct vlan_hdr
*vh
;
1825 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1826 return ERR_PTR(-EINVAL
);
1828 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1829 type
= vh
->h_vlan_encapsulated_proto
;
1830 vlan_depth
+= VLAN_HLEN
;
1833 skb_reset_mac_header(skb
);
1834 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1835 __skb_pull(skb
, skb
->mac_len
);
1837 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1838 struct net_device
*dev
= skb
->dev
;
1839 struct ethtool_drvinfo info
= {};
1841 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1842 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1844 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1845 info
.driver
, dev
? dev
->features
: 0L,
1846 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1847 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1849 if (skb_header_cloned(skb
) &&
1850 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1851 return ERR_PTR(err
);
1855 list_for_each_entry_rcu(ptype
,
1856 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1857 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1858 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1859 err
= ptype
->gso_send_check(skb
);
1860 segs
= ERR_PTR(err
);
1861 if (err
|| skb_gso_ok(skb
, features
))
1863 __skb_push(skb
, (skb
->data
-
1864 skb_network_header(skb
)));
1866 segs
= ptype
->gso_segment(skb
, features
);
1872 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1876 EXPORT_SYMBOL(skb_gso_segment
);
1878 /* Take action when hardware reception checksum errors are detected. */
1880 void netdev_rx_csum_fault(struct net_device
*dev
)
1882 if (net_ratelimit()) {
1883 printk(KERN_ERR
"%s: hw csum failure.\n",
1884 dev
? dev
->name
: "<unknown>");
1888 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1891 /* Actually, we should eliminate this check as soon as we know, that:
1892 * 1. IOMMU is present and allows to map all the memory.
1893 * 2. No high memory really exists on this machine.
1896 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1898 #ifdef CONFIG_HIGHMEM
1900 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1901 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1902 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1906 if (PCI_DMA_BUS_IS_PHYS
) {
1907 struct device
*pdev
= dev
->dev
.parent
;
1911 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1912 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1913 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1922 void (*destructor
)(struct sk_buff
*skb
);
1925 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1927 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1929 struct dev_gso_cb
*cb
;
1932 struct sk_buff
*nskb
= skb
->next
;
1934 skb
->next
= nskb
->next
;
1937 } while (skb
->next
);
1939 cb
= DEV_GSO_CB(skb
);
1941 cb
->destructor(skb
);
1945 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1946 * @skb: buffer to segment
1947 * @features: device features as applicable to this skb
1949 * This function segments the given skb and stores the list of segments
1952 static int dev_gso_segment(struct sk_buff
*skb
, int features
)
1954 struct sk_buff
*segs
;
1956 segs
= skb_gso_segment(skb
, features
);
1958 /* Verifying header integrity only. */
1963 return PTR_ERR(segs
);
1966 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1967 skb
->destructor
= dev_gso_skb_destructor
;
1973 * Try to orphan skb early, right before transmission by the device.
1974 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1975 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1977 static inline void skb_orphan_try(struct sk_buff
*skb
)
1979 struct sock
*sk
= skb
->sk
;
1981 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
1982 /* skb_tx_hash() wont be able to get sk.
1983 * We copy sk_hash into skb->rxhash
1986 skb
->rxhash
= sk
->sk_hash
;
1991 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1993 return ((features
& NETIF_F_GEN_CSUM
) ||
1994 ((features
& NETIF_F_V4_CSUM
) &&
1995 protocol
== htons(ETH_P_IP
)) ||
1996 ((features
& NETIF_F_V6_CSUM
) &&
1997 protocol
== htons(ETH_P_IPV6
)) ||
1998 ((features
& NETIF_F_FCOE_CRC
) &&
1999 protocol
== htons(ETH_P_FCOE
)));
2002 static int harmonize_features(struct sk_buff
*skb
, __be16 protocol
, int features
)
2004 if (!can_checksum_protocol(protocol
, features
)) {
2005 features
&= ~NETIF_F_ALL_CSUM
;
2006 features
&= ~NETIF_F_SG
;
2007 } else if (illegal_highdma(skb
->dev
, skb
)) {
2008 features
&= ~NETIF_F_SG
;
2014 int netif_skb_features(struct sk_buff
*skb
)
2016 __be16 protocol
= skb
->protocol
;
2017 int features
= skb
->dev
->features
;
2019 if (protocol
== htons(ETH_P_8021Q
)) {
2020 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2021 protocol
= veh
->h_vlan_encapsulated_proto
;
2022 } else if (!vlan_tx_tag_present(skb
)) {
2023 return harmonize_features(skb
, protocol
, features
);
2026 features
&= skb
->dev
->vlan_features
;
2028 if (protocol
!= htons(ETH_P_8021Q
)) {
2029 return harmonize_features(skb
, protocol
, features
);
2031 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2033 return harmonize_features(skb
, protocol
, features
);
2036 EXPORT_SYMBOL(netif_skb_features
);
2039 * Returns true if either:
2040 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2041 * 2. skb is fragmented and the device does not support SG, or if
2042 * at least one of fragments is in highmem and device does not
2043 * support DMA from it.
2045 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2048 return skb_is_nonlinear(skb
) &&
2049 ((skb_has_frag_list(skb
) &&
2050 !(features
& NETIF_F_FRAGLIST
)) ||
2051 (skb_shinfo(skb
)->nr_frags
&&
2052 !(features
& NETIF_F_SG
)));
2055 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2056 struct netdev_queue
*txq
)
2058 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2059 int rc
= NETDEV_TX_OK
;
2061 if (likely(!skb
->next
)) {
2065 * If device doesnt need skb->dst, release it right now while
2066 * its hot in this cpu cache
2068 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2071 if (!list_empty(&ptype_all
))
2072 dev_queue_xmit_nit(skb
, dev
);
2074 skb_orphan_try(skb
);
2076 features
= netif_skb_features(skb
);
2078 if (vlan_tx_tag_present(skb
) &&
2079 !(features
& NETIF_F_HW_VLAN_TX
)) {
2080 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2087 if (netif_needs_gso(skb
, features
)) {
2088 if (unlikely(dev_gso_segment(skb
, features
)))
2093 if (skb_needs_linearize(skb
, features
) &&
2094 __skb_linearize(skb
))
2097 /* If packet is not checksummed and device does not
2098 * support checksumming for this protocol, complete
2099 * checksumming here.
2101 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2102 skb_set_transport_header(skb
,
2103 skb_checksum_start_offset(skb
));
2104 if (!(features
& NETIF_F_ALL_CSUM
) &&
2105 skb_checksum_help(skb
))
2110 rc
= ops
->ndo_start_xmit(skb
, dev
);
2111 trace_net_dev_xmit(skb
, rc
);
2112 if (rc
== NETDEV_TX_OK
)
2113 txq_trans_update(txq
);
2119 struct sk_buff
*nskb
= skb
->next
;
2121 skb
->next
= nskb
->next
;
2125 * If device doesnt need nskb->dst, release it right now while
2126 * its hot in this cpu cache
2128 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2131 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2132 trace_net_dev_xmit(nskb
, rc
);
2133 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2134 if (rc
& ~NETDEV_TX_MASK
)
2135 goto out_kfree_gso_skb
;
2136 nskb
->next
= skb
->next
;
2140 txq_trans_update(txq
);
2141 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2142 return NETDEV_TX_BUSY
;
2143 } while (skb
->next
);
2146 if (likely(skb
->next
== NULL
))
2147 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2154 static u32 hashrnd __read_mostly
;
2157 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2158 * to be used as a distribution range.
2160 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2161 unsigned int num_tx_queues
)
2165 if (skb_rx_queue_recorded(skb
)) {
2166 hash
= skb_get_rx_queue(skb
);
2167 while (unlikely(hash
>= num_tx_queues
))
2168 hash
-= num_tx_queues
;
2172 if (skb
->sk
&& skb
->sk
->sk_hash
)
2173 hash
= skb
->sk
->sk_hash
;
2175 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2176 hash
= jhash_1word(hash
, hashrnd
);
2178 return (u16
) (((u64
) hash
* num_tx_queues
) >> 32);
2180 EXPORT_SYMBOL(__skb_tx_hash
);
2182 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2184 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2185 if (net_ratelimit()) {
2186 pr_warning("%s selects TX queue %d, but "
2187 "real number of TX queues is %d\n",
2188 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2195 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2198 struct xps_dev_maps
*dev_maps
;
2199 struct xps_map
*map
;
2200 int queue_index
= -1;
2203 dev_maps
= rcu_dereference(dev
->xps_maps
);
2205 map
= rcu_dereference(
2206 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2209 queue_index
= map
->queues
[0];
2212 if (skb
->sk
&& skb
->sk
->sk_hash
)
2213 hash
= skb
->sk
->sk_hash
;
2215 hash
= (__force u16
) skb
->protocol
^
2217 hash
= jhash_1word(hash
, hashrnd
);
2218 queue_index
= map
->queues
[
2219 ((u64
)hash
* map
->len
) >> 32];
2221 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2233 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2234 struct sk_buff
*skb
)
2237 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2239 if (dev
->real_num_tx_queues
== 1)
2241 else if (ops
->ndo_select_queue
) {
2242 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2243 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2245 struct sock
*sk
= skb
->sk
;
2246 queue_index
= sk_tx_queue_get(sk
);
2248 if (queue_index
< 0 || skb
->ooo_okay
||
2249 queue_index
>= dev
->real_num_tx_queues
) {
2250 int old_index
= queue_index
;
2252 queue_index
= get_xps_queue(dev
, skb
);
2253 if (queue_index
< 0)
2254 queue_index
= skb_tx_hash(dev
, skb
);
2256 if (queue_index
!= old_index
&& sk
) {
2257 struct dst_entry
*dst
=
2258 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2260 if (dst
&& skb_dst(skb
) == dst
)
2261 sk_tx_queue_set(sk
, queue_index
);
2266 skb_set_queue_mapping(skb
, queue_index
);
2267 return netdev_get_tx_queue(dev
, queue_index
);
2270 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2271 struct net_device
*dev
,
2272 struct netdev_queue
*txq
)
2274 spinlock_t
*root_lock
= qdisc_lock(q
);
2275 bool contended
= qdisc_is_running(q
);
2279 * Heuristic to force contended enqueues to serialize on a
2280 * separate lock before trying to get qdisc main lock.
2281 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2282 * and dequeue packets faster.
2284 if (unlikely(contended
))
2285 spin_lock(&q
->busylock
);
2287 spin_lock(root_lock
);
2288 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2291 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2292 qdisc_run_begin(q
)) {
2294 * This is a work-conserving queue; there are no old skbs
2295 * waiting to be sent out; and the qdisc is not running -
2296 * xmit the skb directly.
2298 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2301 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2302 qdisc_bstats_update(q
, skb
);
2304 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2305 if (unlikely(contended
)) {
2306 spin_unlock(&q
->busylock
);
2313 rc
= NET_XMIT_SUCCESS
;
2316 rc
= qdisc_enqueue_root(skb
, q
);
2317 if (qdisc_run_begin(q
)) {
2318 if (unlikely(contended
)) {
2319 spin_unlock(&q
->busylock
);
2325 spin_unlock(root_lock
);
2326 if (unlikely(contended
))
2327 spin_unlock(&q
->busylock
);
2331 static DEFINE_PER_CPU(int, xmit_recursion
);
2332 #define RECURSION_LIMIT 10
2335 * dev_queue_xmit - transmit a buffer
2336 * @skb: buffer to transmit
2338 * Queue a buffer for transmission to a network device. The caller must
2339 * have set the device and priority and built the buffer before calling
2340 * this function. The function can be called from an interrupt.
2342 * A negative errno code is returned on a failure. A success does not
2343 * guarantee the frame will be transmitted as it may be dropped due
2344 * to congestion or traffic shaping.
2346 * -----------------------------------------------------------------------------------
2347 * I notice this method can also return errors from the queue disciplines,
2348 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2351 * Regardless of the return value, the skb is consumed, so it is currently
2352 * difficult to retry a send to this method. (You can bump the ref count
2353 * before sending to hold a reference for retry if you are careful.)
2355 * When calling this method, interrupts MUST be enabled. This is because
2356 * the BH enable code must have IRQs enabled so that it will not deadlock.
2359 int dev_queue_xmit(struct sk_buff
*skb
)
2361 struct net_device
*dev
= skb
->dev
;
2362 struct netdev_queue
*txq
;
2366 /* Disable soft irqs for various locks below. Also
2367 * stops preemption for RCU.
2371 txq
= dev_pick_tx(dev
, skb
);
2372 q
= rcu_dereference_bh(txq
->qdisc
);
2374 #ifdef CONFIG_NET_CLS_ACT
2375 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2377 trace_net_dev_queue(skb
);
2379 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2383 /* The device has no queue. Common case for software devices:
2384 loopback, all the sorts of tunnels...
2386 Really, it is unlikely that netif_tx_lock protection is necessary
2387 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2389 However, it is possible, that they rely on protection
2392 Check this and shot the lock. It is not prone from deadlocks.
2393 Either shot noqueue qdisc, it is even simpler 8)
2395 if (dev
->flags
& IFF_UP
) {
2396 int cpu
= smp_processor_id(); /* ok because BHs are off */
2398 if (txq
->xmit_lock_owner
!= cpu
) {
2400 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2401 goto recursion_alert
;
2403 HARD_TX_LOCK(dev
, txq
, cpu
);
2405 if (!netif_tx_queue_stopped(txq
)) {
2406 __this_cpu_inc(xmit_recursion
);
2407 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2408 __this_cpu_dec(xmit_recursion
);
2409 if (dev_xmit_complete(rc
)) {
2410 HARD_TX_UNLOCK(dev
, txq
);
2414 HARD_TX_UNLOCK(dev
, txq
);
2415 if (net_ratelimit())
2416 printk(KERN_CRIT
"Virtual device %s asks to "
2417 "queue packet!\n", dev
->name
);
2419 /* Recursion is detected! It is possible,
2423 if (net_ratelimit())
2424 printk(KERN_CRIT
"Dead loop on virtual device "
2425 "%s, fix it urgently!\n", dev
->name
);
2430 rcu_read_unlock_bh();
2435 rcu_read_unlock_bh();
2438 EXPORT_SYMBOL(dev_queue_xmit
);
2441 /*=======================================================================
2443 =======================================================================*/
2445 int netdev_max_backlog __read_mostly
= 1000;
2446 int netdev_tstamp_prequeue __read_mostly
= 1;
2447 int netdev_budget __read_mostly
= 300;
2448 int weight_p __read_mostly
= 64; /* old backlog weight */
2450 /* Called with irq disabled */
2451 static inline void ____napi_schedule(struct softnet_data
*sd
,
2452 struct napi_struct
*napi
)
2454 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2455 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2459 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2460 * and src/dst port numbers. Returns a non-zero hash number on success
2463 __u32
__skb_get_rxhash(struct sk_buff
*skb
)
2465 int nhoff
, hash
= 0, poff
;
2466 struct ipv6hdr
*ip6
;
2469 u32 addr1
, addr2
, ihl
;
2475 nhoff
= skb_network_offset(skb
);
2477 switch (skb
->protocol
) {
2478 case __constant_htons(ETH_P_IP
):
2479 if (!pskb_may_pull(skb
, sizeof(*ip
) + nhoff
))
2482 ip
= (struct iphdr
*) (skb
->data
+ nhoff
);
2483 if (ip
->frag_off
& htons(IP_MF
| IP_OFFSET
))
2486 ip_proto
= ip
->protocol
;
2487 addr1
= (__force u32
) ip
->saddr
;
2488 addr2
= (__force u32
) ip
->daddr
;
2491 case __constant_htons(ETH_P_IPV6
):
2492 if (!pskb_may_pull(skb
, sizeof(*ip6
) + nhoff
))
2495 ip6
= (struct ipv6hdr
*) (skb
->data
+ nhoff
);
2496 ip_proto
= ip6
->nexthdr
;
2497 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2498 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2506 poff
= proto_ports_offset(ip_proto
);
2508 nhoff
+= ihl
* 4 + poff
;
2509 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2510 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2511 if (ports
.v16
[1] < ports
.v16
[0])
2512 swap(ports
.v16
[0], ports
.v16
[1]);
2516 /* get a consistent hash (same value on both flow directions) */
2520 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2527 EXPORT_SYMBOL(__skb_get_rxhash
);
2531 /* One global table that all flow-based protocols share. */
2532 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2533 EXPORT_SYMBOL(rps_sock_flow_table
);
2536 * get_rps_cpu is called from netif_receive_skb and returns the target
2537 * CPU from the RPS map of the receiving queue for a given skb.
2538 * rcu_read_lock must be held on entry.
2540 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2541 struct rps_dev_flow
**rflowp
)
2543 struct netdev_rx_queue
*rxqueue
;
2544 struct rps_map
*map
;
2545 struct rps_dev_flow_table
*flow_table
;
2546 struct rps_sock_flow_table
*sock_flow_table
;
2550 if (skb_rx_queue_recorded(skb
)) {
2551 u16 index
= skb_get_rx_queue(skb
);
2552 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2553 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2554 "%s received packet on queue %u, but number "
2555 "of RX queues is %u\n",
2556 dev
->name
, index
, dev
->real_num_rx_queues
);
2559 rxqueue
= dev
->_rx
+ index
;
2563 map
= rcu_dereference(rxqueue
->rps_map
);
2565 if (map
->len
== 1) {
2566 tcpu
= map
->cpus
[0];
2567 if (cpu_online(tcpu
))
2571 } else if (!rcu_dereference_raw(rxqueue
->rps_flow_table
)) {
2575 skb_reset_network_header(skb
);
2576 if (!skb_get_rxhash(skb
))
2579 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2580 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2581 if (flow_table
&& sock_flow_table
) {
2583 struct rps_dev_flow
*rflow
;
2585 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2588 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2589 sock_flow_table
->mask
];
2592 * If the desired CPU (where last recvmsg was done) is
2593 * different from current CPU (one in the rx-queue flow
2594 * table entry), switch if one of the following holds:
2595 * - Current CPU is unset (equal to RPS_NO_CPU).
2596 * - Current CPU is offline.
2597 * - The current CPU's queue tail has advanced beyond the
2598 * last packet that was enqueued using this table entry.
2599 * This guarantees that all previous packets for the flow
2600 * have been dequeued, thus preserving in order delivery.
2602 if (unlikely(tcpu
!= next_cpu
) &&
2603 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2604 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2605 rflow
->last_qtail
)) >= 0)) {
2606 tcpu
= rflow
->cpu
= next_cpu
;
2607 if (tcpu
!= RPS_NO_CPU
)
2608 rflow
->last_qtail
= per_cpu(softnet_data
,
2609 tcpu
).input_queue_head
;
2611 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2619 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2621 if (cpu_online(tcpu
)) {
2631 /* Called from hardirq (IPI) context */
2632 static void rps_trigger_softirq(void *data
)
2634 struct softnet_data
*sd
= data
;
2636 ____napi_schedule(sd
, &sd
->backlog
);
2640 #endif /* CONFIG_RPS */
2643 * Check if this softnet_data structure is another cpu one
2644 * If yes, queue it to our IPI list and return 1
2647 static int rps_ipi_queued(struct softnet_data
*sd
)
2650 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2653 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2654 mysd
->rps_ipi_list
= sd
;
2656 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2659 #endif /* CONFIG_RPS */
2664 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2665 * queue (may be a remote CPU queue).
2667 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2668 unsigned int *qtail
)
2670 struct softnet_data
*sd
;
2671 unsigned long flags
;
2673 sd
= &per_cpu(softnet_data
, cpu
);
2675 local_irq_save(flags
);
2678 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2679 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2681 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2682 input_queue_tail_incr_save(sd
, qtail
);
2684 local_irq_restore(flags
);
2685 return NET_RX_SUCCESS
;
2688 /* Schedule NAPI for backlog device
2689 * We can use non atomic operation since we own the queue lock
2691 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2692 if (!rps_ipi_queued(sd
))
2693 ____napi_schedule(sd
, &sd
->backlog
);
2701 local_irq_restore(flags
);
2703 atomic_long_inc(&skb
->dev
->rx_dropped
);
2709 * netif_rx - post buffer to the network code
2710 * @skb: buffer to post
2712 * This function receives a packet from a device driver and queues it for
2713 * the upper (protocol) levels to process. It always succeeds. The buffer
2714 * may be dropped during processing for congestion control or by the
2718 * NET_RX_SUCCESS (no congestion)
2719 * NET_RX_DROP (packet was dropped)
2723 int netif_rx(struct sk_buff
*skb
)
2727 /* if netpoll wants it, pretend we never saw it */
2728 if (netpoll_rx(skb
))
2731 if (netdev_tstamp_prequeue
)
2732 net_timestamp_check(skb
);
2734 trace_netif_rx(skb
);
2737 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2743 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2745 cpu
= smp_processor_id();
2747 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2755 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2761 EXPORT_SYMBOL(netif_rx
);
2763 int netif_rx_ni(struct sk_buff
*skb
)
2768 err
= netif_rx(skb
);
2769 if (local_softirq_pending())
2775 EXPORT_SYMBOL(netif_rx_ni
);
2777 static void net_tx_action(struct softirq_action
*h
)
2779 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2781 if (sd
->completion_queue
) {
2782 struct sk_buff
*clist
;
2784 local_irq_disable();
2785 clist
= sd
->completion_queue
;
2786 sd
->completion_queue
= NULL
;
2790 struct sk_buff
*skb
= clist
;
2791 clist
= clist
->next
;
2793 WARN_ON(atomic_read(&skb
->users
));
2794 trace_kfree_skb(skb
, net_tx_action
);
2799 if (sd
->output_queue
) {
2802 local_irq_disable();
2803 head
= sd
->output_queue
;
2804 sd
->output_queue
= NULL
;
2805 sd
->output_queue_tailp
= &sd
->output_queue
;
2809 struct Qdisc
*q
= head
;
2810 spinlock_t
*root_lock
;
2812 head
= head
->next_sched
;
2814 root_lock
= qdisc_lock(q
);
2815 if (spin_trylock(root_lock
)) {
2816 smp_mb__before_clear_bit();
2817 clear_bit(__QDISC_STATE_SCHED
,
2820 spin_unlock(root_lock
);
2822 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2824 __netif_reschedule(q
);
2826 smp_mb__before_clear_bit();
2827 clear_bit(__QDISC_STATE_SCHED
,
2835 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2836 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2837 /* This hook is defined here for ATM LANE */
2838 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2839 unsigned char *addr
) __read_mostly
;
2840 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2843 #ifdef CONFIG_NET_CLS_ACT
2844 /* TODO: Maybe we should just force sch_ingress to be compiled in
2845 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2846 * a compare and 2 stores extra right now if we dont have it on
2847 * but have CONFIG_NET_CLS_ACT
2848 * NOTE: This doesnt stop any functionality; if you dont have
2849 * the ingress scheduler, you just cant add policies on ingress.
2852 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
2854 struct net_device
*dev
= skb
->dev
;
2855 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2856 int result
= TC_ACT_OK
;
2859 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
2860 if (net_ratelimit())
2861 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2862 skb
->skb_iif
, dev
->ifindex
);
2866 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2867 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2870 if (q
!= &noop_qdisc
) {
2871 spin_lock(qdisc_lock(q
));
2872 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2873 result
= qdisc_enqueue_root(skb
, q
);
2874 spin_unlock(qdisc_lock(q
));
2880 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2881 struct packet_type
**pt_prev
,
2882 int *ret
, struct net_device
*orig_dev
)
2884 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
2886 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
2890 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2894 switch (ing_filter(skb
, rxq
)) {
2908 * netdev_rx_handler_register - register receive handler
2909 * @dev: device to register a handler for
2910 * @rx_handler: receive handler to register
2911 * @rx_handler_data: data pointer that is used by rx handler
2913 * Register a receive hander for a device. This handler will then be
2914 * called from __netif_receive_skb. A negative errno code is returned
2917 * The caller must hold the rtnl_mutex.
2919 int netdev_rx_handler_register(struct net_device
*dev
,
2920 rx_handler_func_t
*rx_handler
,
2921 void *rx_handler_data
)
2925 if (dev
->rx_handler
)
2928 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
2929 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
2933 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
2936 * netdev_rx_handler_unregister - unregister receive handler
2937 * @dev: device to unregister a handler from
2939 * Unregister a receive hander from a device.
2941 * The caller must hold the rtnl_mutex.
2943 void netdev_rx_handler_unregister(struct net_device
*dev
)
2947 rcu_assign_pointer(dev
->rx_handler
, NULL
);
2948 rcu_assign_pointer(dev
->rx_handler_data
, NULL
);
2950 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
2952 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2953 struct net_device
*master
)
2955 if (skb
->pkt_type
== PACKET_HOST
) {
2956 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2958 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2962 /* On bonding slaves other than the currently active slave, suppress
2963 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2964 * ARP on active-backup slaves with arp_validate enabled.
2966 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2968 struct net_device
*dev
= skb
->dev
;
2970 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2971 dev
->last_rx
= jiffies
;
2973 if ((master
->priv_flags
& IFF_MASTER_ALB
) &&
2974 (master
->priv_flags
& IFF_BRIDGE_PORT
)) {
2975 /* Do address unmangle. The local destination address
2976 * will be always the one master has. Provides the right
2977 * functionality in a bridge.
2979 skb_bond_set_mac_by_master(skb
, master
);
2982 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
2983 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
2984 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
2987 if (master
->priv_flags
& IFF_MASTER_ALB
) {
2988 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
2989 skb
->pkt_type
!= PACKET_MULTICAST
)
2992 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
2993 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
3000 EXPORT_SYMBOL(__skb_bond_should_drop
);
3002 static int __netif_receive_skb(struct sk_buff
*skb
)
3004 struct packet_type
*ptype
, *pt_prev
;
3005 rx_handler_func_t
*rx_handler
;
3006 struct net_device
*orig_dev
;
3007 struct net_device
*master
;
3008 struct net_device
*null_or_orig
;
3009 struct net_device
*orig_or_bond
;
3010 int ret
= NET_RX_DROP
;
3013 if (!netdev_tstamp_prequeue
)
3014 net_timestamp_check(skb
);
3016 trace_netif_receive_skb(skb
);
3018 /* if we've gotten here through NAPI, check netpoll */
3019 if (netpoll_receive_skb(skb
))
3023 skb
->skb_iif
= skb
->dev
->ifindex
;
3026 * bonding note: skbs received on inactive slaves should only
3027 * be delivered to pkt handlers that are exact matches. Also
3028 * the deliver_no_wcard flag will be set. If packet handlers
3029 * are sensitive to duplicate packets these skbs will need to
3030 * be dropped at the handler.
3032 null_or_orig
= NULL
;
3033 orig_dev
= skb
->dev
;
3034 master
= ACCESS_ONCE(orig_dev
->master
);
3035 if (skb
->deliver_no_wcard
)
3036 null_or_orig
= orig_dev
;
3038 if (skb_bond_should_drop(skb
, master
)) {
3039 skb
->deliver_no_wcard
= 1;
3040 null_or_orig
= orig_dev
; /* deliver only exact match */
3045 __this_cpu_inc(softnet_data
.processed
);
3046 skb_reset_network_header(skb
);
3047 skb_reset_transport_header(skb
);
3048 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
3054 #ifdef CONFIG_NET_CLS_ACT
3055 if (skb
->tc_verd
& TC_NCLS
) {
3056 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3061 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3062 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
3063 ptype
->dev
== orig_dev
) {
3065 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3070 #ifdef CONFIG_NET_CLS_ACT
3071 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3077 /* Handle special case of bridge or macvlan */
3078 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3081 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3084 skb
= rx_handler(skb
);
3089 if (vlan_tx_tag_present(skb
)) {
3091 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3094 if (vlan_hwaccel_do_receive(&skb
)) {
3095 ret
= __netif_receive_skb(skb
);
3097 } else if (unlikely(!skb
))
3102 * Make sure frames received on VLAN interfaces stacked on
3103 * bonding interfaces still make their way to any base bonding
3104 * device that may have registered for a specific ptype. The
3105 * handler may have to adjust skb->dev and orig_dev.
3107 orig_or_bond
= orig_dev
;
3108 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
3109 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
3110 orig_or_bond
= vlan_dev_real_dev(skb
->dev
);
3113 type
= skb
->protocol
;
3114 list_for_each_entry_rcu(ptype
,
3115 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3116 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
3117 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
3118 ptype
->dev
== orig_or_bond
)) {
3120 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3126 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3128 atomic_long_inc(&skb
->dev
->rx_dropped
);
3130 /* Jamal, now you will not able to escape explaining
3131 * me how you were going to use this. :-)
3142 * netif_receive_skb - process receive buffer from network
3143 * @skb: buffer to process
3145 * netif_receive_skb() is the main receive data processing function.
3146 * It always succeeds. The buffer may be dropped during processing
3147 * for congestion control or by the protocol layers.
3149 * This function may only be called from softirq context and interrupts
3150 * should be enabled.
3152 * Return values (usually ignored):
3153 * NET_RX_SUCCESS: no congestion
3154 * NET_RX_DROP: packet was dropped
3156 int netif_receive_skb(struct sk_buff
*skb
)
3158 if (netdev_tstamp_prequeue
)
3159 net_timestamp_check(skb
);
3161 if (skb_defer_rx_timestamp(skb
))
3162 return NET_RX_SUCCESS
;
3166 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3171 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3174 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3178 ret
= __netif_receive_skb(skb
);
3184 return __netif_receive_skb(skb
);
3187 EXPORT_SYMBOL(netif_receive_skb
);
3189 /* Network device is going away, flush any packets still pending
3190 * Called with irqs disabled.
3192 static void flush_backlog(void *arg
)
3194 struct net_device
*dev
= arg
;
3195 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3196 struct sk_buff
*skb
, *tmp
;
3199 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3200 if (skb
->dev
== dev
) {
3201 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3203 input_queue_head_incr(sd
);
3208 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3209 if (skb
->dev
== dev
) {
3210 __skb_unlink(skb
, &sd
->process_queue
);
3212 input_queue_head_incr(sd
);
3217 static int napi_gro_complete(struct sk_buff
*skb
)
3219 struct packet_type
*ptype
;
3220 __be16 type
= skb
->protocol
;
3221 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3224 if (NAPI_GRO_CB(skb
)->count
== 1) {
3225 skb_shinfo(skb
)->gso_size
= 0;
3230 list_for_each_entry_rcu(ptype
, head
, list
) {
3231 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3234 err
= ptype
->gro_complete(skb
);
3240 WARN_ON(&ptype
->list
== head
);
3242 return NET_RX_SUCCESS
;
3246 return netif_receive_skb(skb
);
3249 inline void napi_gro_flush(struct napi_struct
*napi
)
3251 struct sk_buff
*skb
, *next
;
3253 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3256 napi_gro_complete(skb
);
3259 napi
->gro_count
= 0;
3260 napi
->gro_list
= NULL
;
3262 EXPORT_SYMBOL(napi_gro_flush
);
3264 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3266 struct sk_buff
**pp
= NULL
;
3267 struct packet_type
*ptype
;
3268 __be16 type
= skb
->protocol
;
3269 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3272 enum gro_result ret
;
3274 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3277 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3281 list_for_each_entry_rcu(ptype
, head
, list
) {
3282 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3285 skb_set_network_header(skb
, skb_gro_offset(skb
));
3286 mac_len
= skb
->network_header
- skb
->mac_header
;
3287 skb
->mac_len
= mac_len
;
3288 NAPI_GRO_CB(skb
)->same_flow
= 0;
3289 NAPI_GRO_CB(skb
)->flush
= 0;
3290 NAPI_GRO_CB(skb
)->free
= 0;
3292 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3297 if (&ptype
->list
== head
)
3300 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3301 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3304 struct sk_buff
*nskb
= *pp
;
3308 napi_gro_complete(nskb
);
3315 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3319 NAPI_GRO_CB(skb
)->count
= 1;
3320 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3321 skb
->next
= napi
->gro_list
;
3322 napi
->gro_list
= skb
;
3326 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3327 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3329 BUG_ON(skb
->end
- skb
->tail
< grow
);
3331 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3334 skb
->data_len
-= grow
;
3336 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3337 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3339 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3340 put_page(skb_shinfo(skb
)->frags
[0].page
);
3341 memmove(skb_shinfo(skb
)->frags
,
3342 skb_shinfo(skb
)->frags
+ 1,
3343 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3354 EXPORT_SYMBOL(dev_gro_receive
);
3356 static inline gro_result_t
3357 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3361 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3362 unsigned long diffs
;
3364 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3365 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3366 diffs
|= compare_ether_header(skb_mac_header(p
),
3367 skb_gro_mac_header(skb
));
3368 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3369 NAPI_GRO_CB(p
)->flush
= 0;
3372 return dev_gro_receive(napi
, skb
);
3375 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3379 if (netif_receive_skb(skb
))
3384 case GRO_MERGED_FREE
:
3395 EXPORT_SYMBOL(napi_skb_finish
);
3397 void skb_gro_reset_offset(struct sk_buff
*skb
)
3399 NAPI_GRO_CB(skb
)->data_offset
= 0;
3400 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3401 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3403 if (skb
->mac_header
== skb
->tail
&&
3404 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3405 NAPI_GRO_CB(skb
)->frag0
=
3406 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3407 skb_shinfo(skb
)->frags
[0].page_offset
;
3408 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3411 EXPORT_SYMBOL(skb_gro_reset_offset
);
3413 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3415 skb_gro_reset_offset(skb
);
3417 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3419 EXPORT_SYMBOL(napi_gro_receive
);
3421 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3423 __skb_pull(skb
, skb_headlen(skb
));
3424 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3430 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3432 struct sk_buff
*skb
= napi
->skb
;
3435 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3441 EXPORT_SYMBOL(napi_get_frags
);
3443 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3449 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3451 if (ret
== GRO_HELD
)
3452 skb_gro_pull(skb
, -ETH_HLEN
);
3453 else if (netif_receive_skb(skb
))
3458 case GRO_MERGED_FREE
:
3459 napi_reuse_skb(napi
, skb
);
3468 EXPORT_SYMBOL(napi_frags_finish
);
3470 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3472 struct sk_buff
*skb
= napi
->skb
;
3479 skb_reset_mac_header(skb
);
3480 skb_gro_reset_offset(skb
);
3482 off
= skb_gro_offset(skb
);
3483 hlen
= off
+ sizeof(*eth
);
3484 eth
= skb_gro_header_fast(skb
, off
);
3485 if (skb_gro_header_hard(skb
, hlen
)) {
3486 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3487 if (unlikely(!eth
)) {
3488 napi_reuse_skb(napi
, skb
);
3494 skb_gro_pull(skb
, sizeof(*eth
));
3497 * This works because the only protocols we care about don't require
3498 * special handling. We'll fix it up properly at the end.
3500 skb
->protocol
= eth
->h_proto
;
3505 EXPORT_SYMBOL(napi_frags_skb
);
3507 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3509 struct sk_buff
*skb
= napi_frags_skb(napi
);
3514 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3516 EXPORT_SYMBOL(napi_gro_frags
);
3519 * net_rps_action sends any pending IPI's for rps.
3520 * Note: called with local irq disabled, but exits with local irq enabled.
3522 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3525 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3528 sd
->rps_ipi_list
= NULL
;
3532 /* Send pending IPI's to kick RPS processing on remote cpus. */
3534 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3536 if (cpu_online(remsd
->cpu
))
3537 __smp_call_function_single(remsd
->cpu
,
3546 static int process_backlog(struct napi_struct
*napi
, int quota
)
3549 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3552 /* Check if we have pending ipi, its better to send them now,
3553 * not waiting net_rx_action() end.
3555 if (sd
->rps_ipi_list
) {
3556 local_irq_disable();
3557 net_rps_action_and_irq_enable(sd
);
3560 napi
->weight
= weight_p
;
3561 local_irq_disable();
3562 while (work
< quota
) {
3563 struct sk_buff
*skb
;
3566 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3568 __netif_receive_skb(skb
);
3569 local_irq_disable();
3570 input_queue_head_incr(sd
);
3571 if (++work
>= quota
) {
3578 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3580 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3581 &sd
->process_queue
);
3583 if (qlen
< quota
- work
) {
3585 * Inline a custom version of __napi_complete().
3586 * only current cpu owns and manipulates this napi,
3587 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3588 * we can use a plain write instead of clear_bit(),
3589 * and we dont need an smp_mb() memory barrier.
3591 list_del(&napi
->poll_list
);
3594 quota
= work
+ qlen
;
3604 * __napi_schedule - schedule for receive
3605 * @n: entry to schedule
3607 * The entry's receive function will be scheduled to run
3609 void __napi_schedule(struct napi_struct
*n
)
3611 unsigned long flags
;
3613 local_irq_save(flags
);
3614 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3615 local_irq_restore(flags
);
3617 EXPORT_SYMBOL(__napi_schedule
);
3619 void __napi_complete(struct napi_struct
*n
)
3621 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3622 BUG_ON(n
->gro_list
);
3624 list_del(&n
->poll_list
);
3625 smp_mb__before_clear_bit();
3626 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3628 EXPORT_SYMBOL(__napi_complete
);
3630 void napi_complete(struct napi_struct
*n
)
3632 unsigned long flags
;
3635 * don't let napi dequeue from the cpu poll list
3636 * just in case its running on a different cpu
3638 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3642 local_irq_save(flags
);
3644 local_irq_restore(flags
);
3646 EXPORT_SYMBOL(napi_complete
);
3648 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3649 int (*poll
)(struct napi_struct
*, int), int weight
)
3651 INIT_LIST_HEAD(&napi
->poll_list
);
3652 napi
->gro_count
= 0;
3653 napi
->gro_list
= NULL
;
3656 napi
->weight
= weight
;
3657 list_add(&napi
->dev_list
, &dev
->napi_list
);
3659 #ifdef CONFIG_NETPOLL
3660 spin_lock_init(&napi
->poll_lock
);
3661 napi
->poll_owner
= -1;
3663 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3665 EXPORT_SYMBOL(netif_napi_add
);
3667 void netif_napi_del(struct napi_struct
*napi
)
3669 struct sk_buff
*skb
, *next
;
3671 list_del_init(&napi
->dev_list
);
3672 napi_free_frags(napi
);
3674 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3680 napi
->gro_list
= NULL
;
3681 napi
->gro_count
= 0;
3683 EXPORT_SYMBOL(netif_napi_del
);
3685 static void net_rx_action(struct softirq_action
*h
)
3687 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3688 unsigned long time_limit
= jiffies
+ 2;
3689 int budget
= netdev_budget
;
3692 local_irq_disable();
3694 while (!list_empty(&sd
->poll_list
)) {
3695 struct napi_struct
*n
;
3698 /* If softirq window is exhuasted then punt.
3699 * Allow this to run for 2 jiffies since which will allow
3700 * an average latency of 1.5/HZ.
3702 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3707 /* Even though interrupts have been re-enabled, this
3708 * access is safe because interrupts can only add new
3709 * entries to the tail of this list, and only ->poll()
3710 * calls can remove this head entry from the list.
3712 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3714 have
= netpoll_poll_lock(n
);
3718 /* This NAPI_STATE_SCHED test is for avoiding a race
3719 * with netpoll's poll_napi(). Only the entity which
3720 * obtains the lock and sees NAPI_STATE_SCHED set will
3721 * actually make the ->poll() call. Therefore we avoid
3722 * accidently calling ->poll() when NAPI is not scheduled.
3725 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3726 work
= n
->poll(n
, weight
);
3730 WARN_ON_ONCE(work
> weight
);
3734 local_irq_disable();
3736 /* Drivers must not modify the NAPI state if they
3737 * consume the entire weight. In such cases this code
3738 * still "owns" the NAPI instance and therefore can
3739 * move the instance around on the list at-will.
3741 if (unlikely(work
== weight
)) {
3742 if (unlikely(napi_disable_pending(n
))) {
3745 local_irq_disable();
3747 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3750 netpoll_poll_unlock(have
);
3753 net_rps_action_and_irq_enable(sd
);
3755 #ifdef CONFIG_NET_DMA
3757 * There may not be any more sk_buffs coming right now, so push
3758 * any pending DMA copies to hardware
3760 dma_issue_pending_all();
3767 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3771 static gifconf_func_t
*gifconf_list
[NPROTO
];
3774 * register_gifconf - register a SIOCGIF handler
3775 * @family: Address family
3776 * @gifconf: Function handler
3778 * Register protocol dependent address dumping routines. The handler
3779 * that is passed must not be freed or reused until it has been replaced
3780 * by another handler.
3782 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3784 if (family
>= NPROTO
)
3786 gifconf_list
[family
] = gifconf
;
3789 EXPORT_SYMBOL(register_gifconf
);
3793 * Map an interface index to its name (SIOCGIFNAME)
3797 * We need this ioctl for efficient implementation of the
3798 * if_indextoname() function required by the IPv6 API. Without
3799 * it, we would have to search all the interfaces to find a
3803 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3805 struct net_device
*dev
;
3809 * Fetch the caller's info block.
3812 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3816 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3822 strcpy(ifr
.ifr_name
, dev
->name
);
3825 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3831 * Perform a SIOCGIFCONF call. This structure will change
3832 * size eventually, and there is nothing I can do about it.
3833 * Thus we will need a 'compatibility mode'.
3836 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3839 struct net_device
*dev
;
3846 * Fetch the caller's info block.
3849 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3856 * Loop over the interfaces, and write an info block for each.
3860 for_each_netdev(net
, dev
) {
3861 for (i
= 0; i
< NPROTO
; i
++) {
3862 if (gifconf_list
[i
]) {
3865 done
= gifconf_list
[i
](dev
, NULL
, 0);
3867 done
= gifconf_list
[i
](dev
, pos
+ total
,
3877 * All done. Write the updated control block back to the caller.
3879 ifc
.ifc_len
= total
;
3882 * Both BSD and Solaris return 0 here, so we do too.
3884 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3887 #ifdef CONFIG_PROC_FS
3889 * This is invoked by the /proc filesystem handler to display a device
3892 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3895 struct net
*net
= seq_file_net(seq
);
3897 struct net_device
*dev
;
3901 return SEQ_START_TOKEN
;
3904 for_each_netdev_rcu(net
, dev
)
3911 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3913 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3914 first_net_device(seq_file_net(seq
)) :
3915 next_net_device((struct net_device
*)v
);
3918 return rcu_dereference(dev
);
3921 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3927 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3929 struct rtnl_link_stats64 temp
;
3930 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
3932 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3933 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3934 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3936 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3937 stats
->rx_fifo_errors
,
3938 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3939 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3940 stats
->rx_compressed
, stats
->multicast
,
3941 stats
->tx_bytes
, stats
->tx_packets
,
3942 stats
->tx_errors
, stats
->tx_dropped
,
3943 stats
->tx_fifo_errors
, stats
->collisions
,
3944 stats
->tx_carrier_errors
+
3945 stats
->tx_aborted_errors
+
3946 stats
->tx_window_errors
+
3947 stats
->tx_heartbeat_errors
,
3948 stats
->tx_compressed
);
3952 * Called from the PROCfs module. This now uses the new arbitrary sized
3953 * /proc/net interface to create /proc/net/dev
3955 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3957 if (v
== SEQ_START_TOKEN
)
3958 seq_puts(seq
, "Inter-| Receive "
3960 " face |bytes packets errs drop fifo frame "
3961 "compressed multicast|bytes packets errs "
3962 "drop fifo colls carrier compressed\n");
3964 dev_seq_printf_stats(seq
, v
);
3968 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
3970 struct softnet_data
*sd
= NULL
;
3972 while (*pos
< nr_cpu_ids
)
3973 if (cpu_online(*pos
)) {
3974 sd
= &per_cpu(softnet_data
, *pos
);
3981 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3983 return softnet_get_online(pos
);
3986 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3989 return softnet_get_online(pos
);
3992 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
3996 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
3998 struct softnet_data
*sd
= v
;
4000 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4001 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4002 0, 0, 0, 0, /* was fastroute */
4003 sd
->cpu_collision
, sd
->received_rps
);
4007 static const struct seq_operations dev_seq_ops
= {
4008 .start
= dev_seq_start
,
4009 .next
= dev_seq_next
,
4010 .stop
= dev_seq_stop
,
4011 .show
= dev_seq_show
,
4014 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4016 return seq_open_net(inode
, file
, &dev_seq_ops
,
4017 sizeof(struct seq_net_private
));
4020 static const struct file_operations dev_seq_fops
= {
4021 .owner
= THIS_MODULE
,
4022 .open
= dev_seq_open
,
4024 .llseek
= seq_lseek
,
4025 .release
= seq_release_net
,
4028 static const struct seq_operations softnet_seq_ops
= {
4029 .start
= softnet_seq_start
,
4030 .next
= softnet_seq_next
,
4031 .stop
= softnet_seq_stop
,
4032 .show
= softnet_seq_show
,
4035 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4037 return seq_open(file
, &softnet_seq_ops
);
4040 static const struct file_operations softnet_seq_fops
= {
4041 .owner
= THIS_MODULE
,
4042 .open
= softnet_seq_open
,
4044 .llseek
= seq_lseek
,
4045 .release
= seq_release
,
4048 static void *ptype_get_idx(loff_t pos
)
4050 struct packet_type
*pt
= NULL
;
4054 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4060 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4061 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4070 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4074 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4077 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4079 struct packet_type
*pt
;
4080 struct list_head
*nxt
;
4084 if (v
== SEQ_START_TOKEN
)
4085 return ptype_get_idx(0);
4088 nxt
= pt
->list
.next
;
4089 if (pt
->type
== htons(ETH_P_ALL
)) {
4090 if (nxt
!= &ptype_all
)
4093 nxt
= ptype_base
[0].next
;
4095 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4097 while (nxt
== &ptype_base
[hash
]) {
4098 if (++hash
>= PTYPE_HASH_SIZE
)
4100 nxt
= ptype_base
[hash
].next
;
4103 return list_entry(nxt
, struct packet_type
, list
);
4106 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4112 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4114 struct packet_type
*pt
= v
;
4116 if (v
== SEQ_START_TOKEN
)
4117 seq_puts(seq
, "Type Device Function\n");
4118 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4119 if (pt
->type
== htons(ETH_P_ALL
))
4120 seq_puts(seq
, "ALL ");
4122 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4124 seq_printf(seq
, " %-8s %pF\n",
4125 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4131 static const struct seq_operations ptype_seq_ops
= {
4132 .start
= ptype_seq_start
,
4133 .next
= ptype_seq_next
,
4134 .stop
= ptype_seq_stop
,
4135 .show
= ptype_seq_show
,
4138 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4140 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4141 sizeof(struct seq_net_private
));
4144 static const struct file_operations ptype_seq_fops
= {
4145 .owner
= THIS_MODULE
,
4146 .open
= ptype_seq_open
,
4148 .llseek
= seq_lseek
,
4149 .release
= seq_release_net
,
4153 static int __net_init
dev_proc_net_init(struct net
*net
)
4157 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4159 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4161 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4164 if (wext_proc_init(net
))
4170 proc_net_remove(net
, "ptype");
4172 proc_net_remove(net
, "softnet_stat");
4174 proc_net_remove(net
, "dev");
4178 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4180 wext_proc_exit(net
);
4182 proc_net_remove(net
, "ptype");
4183 proc_net_remove(net
, "softnet_stat");
4184 proc_net_remove(net
, "dev");
4187 static struct pernet_operations __net_initdata dev_proc_ops
= {
4188 .init
= dev_proc_net_init
,
4189 .exit
= dev_proc_net_exit
,
4192 static int __init
dev_proc_init(void)
4194 return register_pernet_subsys(&dev_proc_ops
);
4197 #define dev_proc_init() 0
4198 #endif /* CONFIG_PROC_FS */
4202 * netdev_set_master - set up master/slave pair
4203 * @slave: slave device
4204 * @master: new master device
4206 * Changes the master device of the slave. Pass %NULL to break the
4207 * bonding. The caller must hold the RTNL semaphore. On a failure
4208 * a negative errno code is returned. On success the reference counts
4209 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4210 * function returns zero.
4212 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4214 struct net_device
*old
= slave
->master
;
4224 slave
->master
= master
;
4231 slave
->flags
|= IFF_SLAVE
;
4233 slave
->flags
&= ~IFF_SLAVE
;
4235 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4238 EXPORT_SYMBOL(netdev_set_master
);
4240 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4242 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4244 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4245 ops
->ndo_change_rx_flags(dev
, flags
);
4248 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4250 unsigned short old_flags
= dev
->flags
;
4256 dev
->flags
|= IFF_PROMISC
;
4257 dev
->promiscuity
+= inc
;
4258 if (dev
->promiscuity
== 0) {
4261 * If inc causes overflow, untouch promisc and return error.
4264 dev
->flags
&= ~IFF_PROMISC
;
4266 dev
->promiscuity
-= inc
;
4267 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4268 "set promiscuity failed, promiscuity feature "
4269 "of device might be broken.\n", dev
->name
);
4273 if (dev
->flags
!= old_flags
) {
4274 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4275 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4277 if (audit_enabled
) {
4278 current_uid_gid(&uid
, &gid
);
4279 audit_log(current
->audit_context
, GFP_ATOMIC
,
4280 AUDIT_ANOM_PROMISCUOUS
,
4281 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4282 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4283 (old_flags
& IFF_PROMISC
),
4284 audit_get_loginuid(current
),
4286 audit_get_sessionid(current
));
4289 dev_change_rx_flags(dev
, IFF_PROMISC
);
4295 * dev_set_promiscuity - update promiscuity count on a device
4299 * Add or remove promiscuity from a device. While the count in the device
4300 * remains above zero the interface remains promiscuous. Once it hits zero
4301 * the device reverts back to normal filtering operation. A negative inc
4302 * value is used to drop promiscuity on the device.
4303 * Return 0 if successful or a negative errno code on error.
4305 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4307 unsigned short old_flags
= dev
->flags
;
4310 err
= __dev_set_promiscuity(dev
, inc
);
4313 if (dev
->flags
!= old_flags
)
4314 dev_set_rx_mode(dev
);
4317 EXPORT_SYMBOL(dev_set_promiscuity
);
4320 * dev_set_allmulti - update allmulti count on a device
4324 * Add or remove reception of all multicast frames to a device. While the
4325 * count in the device remains above zero the interface remains listening
4326 * to all interfaces. Once it hits zero the device reverts back to normal
4327 * filtering operation. A negative @inc value is used to drop the counter
4328 * when releasing a resource needing all multicasts.
4329 * Return 0 if successful or a negative errno code on error.
4332 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4334 unsigned short old_flags
= dev
->flags
;
4338 dev
->flags
|= IFF_ALLMULTI
;
4339 dev
->allmulti
+= inc
;
4340 if (dev
->allmulti
== 0) {
4343 * If inc causes overflow, untouch allmulti and return error.
4346 dev
->flags
&= ~IFF_ALLMULTI
;
4348 dev
->allmulti
-= inc
;
4349 printk(KERN_WARNING
"%s: allmulti touches roof, "
4350 "set allmulti failed, allmulti feature of "
4351 "device might be broken.\n", dev
->name
);
4355 if (dev
->flags
^ old_flags
) {
4356 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4357 dev_set_rx_mode(dev
);
4361 EXPORT_SYMBOL(dev_set_allmulti
);
4364 * Upload unicast and multicast address lists to device and
4365 * configure RX filtering. When the device doesn't support unicast
4366 * filtering it is put in promiscuous mode while unicast addresses
4369 void __dev_set_rx_mode(struct net_device
*dev
)
4371 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4373 /* dev_open will call this function so the list will stay sane. */
4374 if (!(dev
->flags
&IFF_UP
))
4377 if (!netif_device_present(dev
))
4380 if (ops
->ndo_set_rx_mode
)
4381 ops
->ndo_set_rx_mode(dev
);
4383 /* Unicast addresses changes may only happen under the rtnl,
4384 * therefore calling __dev_set_promiscuity here is safe.
4386 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4387 __dev_set_promiscuity(dev
, 1);
4388 dev
->uc_promisc
= 1;
4389 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4390 __dev_set_promiscuity(dev
, -1);
4391 dev
->uc_promisc
= 0;
4394 if (ops
->ndo_set_multicast_list
)
4395 ops
->ndo_set_multicast_list(dev
);
4399 void dev_set_rx_mode(struct net_device
*dev
)
4401 netif_addr_lock_bh(dev
);
4402 __dev_set_rx_mode(dev
);
4403 netif_addr_unlock_bh(dev
);
4407 * dev_get_flags - get flags reported to userspace
4410 * Get the combination of flag bits exported through APIs to userspace.
4412 unsigned dev_get_flags(const struct net_device
*dev
)
4416 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4421 (dev
->gflags
& (IFF_PROMISC
|
4424 if (netif_running(dev
)) {
4425 if (netif_oper_up(dev
))
4426 flags
|= IFF_RUNNING
;
4427 if (netif_carrier_ok(dev
))
4428 flags
|= IFF_LOWER_UP
;
4429 if (netif_dormant(dev
))
4430 flags
|= IFF_DORMANT
;
4435 EXPORT_SYMBOL(dev_get_flags
);
4437 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4439 int old_flags
= dev
->flags
;
4445 * Set the flags on our device.
4448 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4449 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4451 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4455 * Load in the correct multicast list now the flags have changed.
4458 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4459 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4461 dev_set_rx_mode(dev
);
4464 * Have we downed the interface. We handle IFF_UP ourselves
4465 * according to user attempts to set it, rather than blindly
4470 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4471 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4474 dev_set_rx_mode(dev
);
4477 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4478 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4480 dev
->gflags
^= IFF_PROMISC
;
4481 dev_set_promiscuity(dev
, inc
);
4484 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4485 is important. Some (broken) drivers set IFF_PROMISC, when
4486 IFF_ALLMULTI is requested not asking us and not reporting.
4488 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4489 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4491 dev
->gflags
^= IFF_ALLMULTI
;
4492 dev_set_allmulti(dev
, inc
);
4498 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4500 unsigned int changes
= dev
->flags
^ old_flags
;
4502 if (changes
& IFF_UP
) {
4503 if (dev
->flags
& IFF_UP
)
4504 call_netdevice_notifiers(NETDEV_UP
, dev
);
4506 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4509 if (dev
->flags
& IFF_UP
&&
4510 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4511 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4515 * dev_change_flags - change device settings
4517 * @flags: device state flags
4519 * Change settings on device based state flags. The flags are
4520 * in the userspace exported format.
4522 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4525 int old_flags
= dev
->flags
;
4527 ret
= __dev_change_flags(dev
, flags
);
4531 changes
= old_flags
^ dev
->flags
;
4533 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4535 __dev_notify_flags(dev
, old_flags
);
4538 EXPORT_SYMBOL(dev_change_flags
);
4541 * dev_set_mtu - Change maximum transfer unit
4543 * @new_mtu: new transfer unit
4545 * Change the maximum transfer size of the network device.
4547 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4549 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4552 if (new_mtu
== dev
->mtu
)
4555 /* MTU must be positive. */
4559 if (!netif_device_present(dev
))
4563 if (ops
->ndo_change_mtu
)
4564 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4568 if (!err
&& dev
->flags
& IFF_UP
)
4569 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4572 EXPORT_SYMBOL(dev_set_mtu
);
4575 * dev_set_mac_address - Change Media Access Control Address
4579 * Change the hardware (MAC) address of the device
4581 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4583 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4586 if (!ops
->ndo_set_mac_address
)
4588 if (sa
->sa_family
!= dev
->type
)
4590 if (!netif_device_present(dev
))
4592 err
= ops
->ndo_set_mac_address(dev
, sa
);
4594 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4597 EXPORT_SYMBOL(dev_set_mac_address
);
4600 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4602 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4605 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4611 case SIOCGIFFLAGS
: /* Get interface flags */
4612 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4615 case SIOCGIFMETRIC
: /* Get the metric on the interface
4616 (currently unused) */
4617 ifr
->ifr_metric
= 0;
4620 case SIOCGIFMTU
: /* Get the MTU of a device */
4621 ifr
->ifr_mtu
= dev
->mtu
;
4626 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4628 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4629 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4630 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4638 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4639 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4640 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4641 ifr
->ifr_map
.irq
= dev
->irq
;
4642 ifr
->ifr_map
.dma
= dev
->dma
;
4643 ifr
->ifr_map
.port
= dev
->if_port
;
4647 ifr
->ifr_ifindex
= dev
->ifindex
;
4651 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4655 /* dev_ioctl() should ensure this case
4667 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4669 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4672 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4673 const struct net_device_ops
*ops
;
4678 ops
= dev
->netdev_ops
;
4681 case SIOCSIFFLAGS
: /* Set interface flags */
4682 return dev_change_flags(dev
, ifr
->ifr_flags
);
4684 case SIOCSIFMETRIC
: /* Set the metric on the interface
4685 (currently unused) */
4688 case SIOCSIFMTU
: /* Set the MTU of a device */
4689 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4692 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4694 case SIOCSIFHWBROADCAST
:
4695 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4697 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4698 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4699 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4703 if (ops
->ndo_set_config
) {
4704 if (!netif_device_present(dev
))
4706 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4711 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4712 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4714 if (!netif_device_present(dev
))
4716 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4719 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4720 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4722 if (!netif_device_present(dev
))
4724 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4727 if (ifr
->ifr_qlen
< 0)
4729 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4733 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4734 return dev_change_name(dev
, ifr
->ifr_newname
);
4737 * Unknown or private ioctl
4740 if ((cmd
>= SIOCDEVPRIVATE
&&
4741 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4742 cmd
== SIOCBONDENSLAVE
||
4743 cmd
== SIOCBONDRELEASE
||
4744 cmd
== SIOCBONDSETHWADDR
||
4745 cmd
== SIOCBONDSLAVEINFOQUERY
||
4746 cmd
== SIOCBONDINFOQUERY
||
4747 cmd
== SIOCBONDCHANGEACTIVE
||
4748 cmd
== SIOCGMIIPHY
||
4749 cmd
== SIOCGMIIREG
||
4750 cmd
== SIOCSMIIREG
||
4751 cmd
== SIOCBRADDIF
||
4752 cmd
== SIOCBRDELIF
||
4753 cmd
== SIOCSHWTSTAMP
||
4754 cmd
== SIOCWANDEV
) {
4756 if (ops
->ndo_do_ioctl
) {
4757 if (netif_device_present(dev
))
4758 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4770 * This function handles all "interface"-type I/O control requests. The actual
4771 * 'doing' part of this is dev_ifsioc above.
4775 * dev_ioctl - network device ioctl
4776 * @net: the applicable net namespace
4777 * @cmd: command to issue
4778 * @arg: pointer to a struct ifreq in user space
4780 * Issue ioctl functions to devices. This is normally called by the
4781 * user space syscall interfaces but can sometimes be useful for
4782 * other purposes. The return value is the return from the syscall if
4783 * positive or a negative errno code on error.
4786 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4792 /* One special case: SIOCGIFCONF takes ifconf argument
4793 and requires shared lock, because it sleeps writing
4797 if (cmd
== SIOCGIFCONF
) {
4799 ret
= dev_ifconf(net
, (char __user
*) arg
);
4803 if (cmd
== SIOCGIFNAME
)
4804 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4806 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4809 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4811 colon
= strchr(ifr
.ifr_name
, ':');
4816 * See which interface the caller is talking about.
4821 * These ioctl calls:
4822 * - can be done by all.
4823 * - atomic and do not require locking.
4834 dev_load(net
, ifr
.ifr_name
);
4836 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4841 if (copy_to_user(arg
, &ifr
,
4842 sizeof(struct ifreq
)))
4848 dev_load(net
, ifr
.ifr_name
);
4850 ret
= dev_ethtool(net
, &ifr
);
4855 if (copy_to_user(arg
, &ifr
,
4856 sizeof(struct ifreq
)))
4862 * These ioctl calls:
4863 * - require superuser power.
4864 * - require strict serialization.
4870 if (!capable(CAP_NET_ADMIN
))
4872 dev_load(net
, ifr
.ifr_name
);
4874 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4879 if (copy_to_user(arg
, &ifr
,
4880 sizeof(struct ifreq
)))
4886 * These ioctl calls:
4887 * - require superuser power.
4888 * - require strict serialization.
4889 * - do not return a value
4899 case SIOCSIFHWBROADCAST
:
4902 case SIOCBONDENSLAVE
:
4903 case SIOCBONDRELEASE
:
4904 case SIOCBONDSETHWADDR
:
4905 case SIOCBONDCHANGEACTIVE
:
4909 if (!capable(CAP_NET_ADMIN
))
4912 case SIOCBONDSLAVEINFOQUERY
:
4913 case SIOCBONDINFOQUERY
:
4914 dev_load(net
, ifr
.ifr_name
);
4916 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4921 /* Get the per device memory space. We can add this but
4922 * currently do not support it */
4924 /* Set the per device memory buffer space.
4925 * Not applicable in our case */
4930 * Unknown or private ioctl.
4933 if (cmd
== SIOCWANDEV
||
4934 (cmd
>= SIOCDEVPRIVATE
&&
4935 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4936 dev_load(net
, ifr
.ifr_name
);
4938 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4940 if (!ret
&& copy_to_user(arg
, &ifr
,
4941 sizeof(struct ifreq
)))
4945 /* Take care of Wireless Extensions */
4946 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4947 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4954 * dev_new_index - allocate an ifindex
4955 * @net: the applicable net namespace
4957 * Returns a suitable unique value for a new device interface
4958 * number. The caller must hold the rtnl semaphore or the
4959 * dev_base_lock to be sure it remains unique.
4961 static int dev_new_index(struct net
*net
)
4967 if (!__dev_get_by_index(net
, ifindex
))
4972 /* Delayed registration/unregisteration */
4973 static LIST_HEAD(net_todo_list
);
4975 static void net_set_todo(struct net_device
*dev
)
4977 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4980 static void rollback_registered_many(struct list_head
*head
)
4982 struct net_device
*dev
, *tmp
;
4984 BUG_ON(dev_boot_phase
);
4987 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4988 /* Some devices call without registering
4989 * for initialization unwind. Remove those
4990 * devices and proceed with the remaining.
4992 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4993 pr_debug("unregister_netdevice: device %s/%p never "
4994 "was registered\n", dev
->name
, dev
);
4997 list_del(&dev
->unreg_list
);
5001 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5004 /* If device is running, close it first. */
5005 dev_close_many(head
);
5007 list_for_each_entry(dev
, head
, unreg_list
) {
5008 /* And unlink it from device chain. */
5009 unlist_netdevice(dev
);
5011 dev
->reg_state
= NETREG_UNREGISTERING
;
5016 list_for_each_entry(dev
, head
, unreg_list
) {
5017 /* Shutdown queueing discipline. */
5021 /* Notify protocols, that we are about to destroy
5022 this device. They should clean all the things.
5024 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5026 if (!dev
->rtnl_link_ops
||
5027 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5028 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5031 * Flush the unicast and multicast chains
5036 if (dev
->netdev_ops
->ndo_uninit
)
5037 dev
->netdev_ops
->ndo_uninit(dev
);
5039 /* Notifier chain MUST detach us from master device. */
5040 WARN_ON(dev
->master
);
5042 /* Remove entries from kobject tree */
5043 netdev_unregister_kobject(dev
);
5046 /* Process any work delayed until the end of the batch */
5047 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5048 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5052 list_for_each_entry(dev
, head
, unreg_list
)
5056 static void rollback_registered(struct net_device
*dev
)
5060 list_add(&dev
->unreg_list
, &single
);
5061 rollback_registered_many(&single
);
5064 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
5066 /* Fix illegal SG+CSUM combinations. */
5067 if ((features
& NETIF_F_SG
) &&
5068 !(features
& NETIF_F_ALL_CSUM
)) {
5070 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
5071 "checksum feature.\n", name
);
5072 features
&= ~NETIF_F_SG
;
5075 /* TSO requires that SG is present as well. */
5076 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
5078 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
5079 "SG feature.\n", name
);
5080 features
&= ~NETIF_F_TSO
;
5083 if (features
& NETIF_F_UFO
) {
5084 /* maybe split UFO into V4 and V6? */
5085 if (!((features
& NETIF_F_GEN_CSUM
) ||
5086 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5087 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5089 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
5090 "since no checksum offload features.\n",
5092 features
&= ~NETIF_F_UFO
;
5095 if (!(features
& NETIF_F_SG
)) {
5097 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
5098 "since no NETIF_F_SG feature.\n", name
);
5099 features
&= ~NETIF_F_UFO
;
5105 EXPORT_SYMBOL(netdev_fix_features
);
5108 * netif_stacked_transfer_operstate - transfer operstate
5109 * @rootdev: the root or lower level device to transfer state from
5110 * @dev: the device to transfer operstate to
5112 * Transfer operational state from root to device. This is normally
5113 * called when a stacking relationship exists between the root
5114 * device and the device(a leaf device).
5116 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5117 struct net_device
*dev
)
5119 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5120 netif_dormant_on(dev
);
5122 netif_dormant_off(dev
);
5124 if (netif_carrier_ok(rootdev
)) {
5125 if (!netif_carrier_ok(dev
))
5126 netif_carrier_on(dev
);
5128 if (netif_carrier_ok(dev
))
5129 netif_carrier_off(dev
);
5132 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5135 static int netif_alloc_rx_queues(struct net_device
*dev
)
5137 unsigned int i
, count
= dev
->num_rx_queues
;
5138 struct netdev_rx_queue
*rx
;
5142 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5144 pr_err("netdev: Unable to allocate %u rx queues.\n", count
);
5149 for (i
= 0; i
< count
; i
++)
5155 static void netdev_init_one_queue(struct net_device
*dev
,
5156 struct netdev_queue
*queue
, void *_unused
)
5158 /* Initialize queue lock */
5159 spin_lock_init(&queue
->_xmit_lock
);
5160 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5161 queue
->xmit_lock_owner
= -1;
5162 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5166 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5168 unsigned int count
= dev
->num_tx_queues
;
5169 struct netdev_queue
*tx
;
5173 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5175 pr_err("netdev: Unable to allocate %u tx queues.\n",
5181 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5182 spin_lock_init(&dev
->tx_global_lock
);
5188 * register_netdevice - register a network device
5189 * @dev: device to register
5191 * Take a completed network device structure and add it to the kernel
5192 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5193 * chain. 0 is returned on success. A negative errno code is returned
5194 * on a failure to set up the device, or if the name is a duplicate.
5196 * Callers must hold the rtnl semaphore. You may want
5197 * register_netdev() instead of this.
5200 * The locking appears insufficient to guarantee two parallel registers
5201 * will not get the same name.
5204 int register_netdevice(struct net_device
*dev
)
5207 struct net
*net
= dev_net(dev
);
5209 BUG_ON(dev_boot_phase
);
5214 /* When net_device's are persistent, this will be fatal. */
5215 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5218 spin_lock_init(&dev
->addr_list_lock
);
5219 netdev_set_addr_lockdep_class(dev
);
5223 /* Init, if this function is available */
5224 if (dev
->netdev_ops
->ndo_init
) {
5225 ret
= dev
->netdev_ops
->ndo_init(dev
);
5233 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
5237 dev
->ifindex
= dev_new_index(net
);
5238 if (dev
->iflink
== -1)
5239 dev
->iflink
= dev
->ifindex
;
5241 /* Fix illegal checksum combinations */
5242 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
5243 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5244 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5246 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5249 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5250 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5251 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5253 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5256 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5258 /* Enable software GSO if SG is supported. */
5259 if (dev
->features
& NETIF_F_SG
)
5260 dev
->features
|= NETIF_F_GSO
;
5262 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5263 * vlan_dev_init() will do the dev->features check, so these features
5264 * are enabled only if supported by underlying device.
5266 dev
->vlan_features
|= (NETIF_F_GRO
| NETIF_F_HIGHDMA
);
5268 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5269 ret
= notifier_to_errno(ret
);
5273 ret
= netdev_register_kobject(dev
);
5276 dev
->reg_state
= NETREG_REGISTERED
;
5279 * Default initial state at registry is that the
5280 * device is present.
5283 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5285 dev_init_scheduler(dev
);
5287 list_netdevice(dev
);
5289 /* Notify protocols, that a new device appeared. */
5290 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5291 ret
= notifier_to_errno(ret
);
5293 rollback_registered(dev
);
5294 dev
->reg_state
= NETREG_UNREGISTERED
;
5297 * Prevent userspace races by waiting until the network
5298 * device is fully setup before sending notifications.
5300 if (!dev
->rtnl_link_ops
||
5301 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5302 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5308 if (dev
->netdev_ops
->ndo_uninit
)
5309 dev
->netdev_ops
->ndo_uninit(dev
);
5312 EXPORT_SYMBOL(register_netdevice
);
5315 * init_dummy_netdev - init a dummy network device for NAPI
5316 * @dev: device to init
5318 * This takes a network device structure and initialize the minimum
5319 * amount of fields so it can be used to schedule NAPI polls without
5320 * registering a full blown interface. This is to be used by drivers
5321 * that need to tie several hardware interfaces to a single NAPI
5322 * poll scheduler due to HW limitations.
5324 int init_dummy_netdev(struct net_device
*dev
)
5326 /* Clear everything. Note we don't initialize spinlocks
5327 * are they aren't supposed to be taken by any of the
5328 * NAPI code and this dummy netdev is supposed to be
5329 * only ever used for NAPI polls
5331 memset(dev
, 0, sizeof(struct net_device
));
5333 /* make sure we BUG if trying to hit standard
5334 * register/unregister code path
5336 dev
->reg_state
= NETREG_DUMMY
;
5338 /* NAPI wants this */
5339 INIT_LIST_HEAD(&dev
->napi_list
);
5341 /* a dummy interface is started by default */
5342 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5343 set_bit(__LINK_STATE_START
, &dev
->state
);
5345 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5346 * because users of this 'device' dont need to change
5352 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5356 * register_netdev - register a network device
5357 * @dev: device to register
5359 * Take a completed network device structure and add it to the kernel
5360 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5361 * chain. 0 is returned on success. A negative errno code is returned
5362 * on a failure to set up the device, or if the name is a duplicate.
5364 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5365 * and expands the device name if you passed a format string to
5368 int register_netdev(struct net_device
*dev
)
5375 * If the name is a format string the caller wants us to do a
5378 if (strchr(dev
->name
, '%')) {
5379 err
= dev_alloc_name(dev
, dev
->name
);
5384 err
= register_netdevice(dev
);
5389 EXPORT_SYMBOL(register_netdev
);
5391 int netdev_refcnt_read(const struct net_device
*dev
)
5395 for_each_possible_cpu(i
)
5396 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5399 EXPORT_SYMBOL(netdev_refcnt_read
);
5402 * netdev_wait_allrefs - wait until all references are gone.
5404 * This is called when unregistering network devices.
5406 * Any protocol or device that holds a reference should register
5407 * for netdevice notification, and cleanup and put back the
5408 * reference if they receive an UNREGISTER event.
5409 * We can get stuck here if buggy protocols don't correctly
5412 static void netdev_wait_allrefs(struct net_device
*dev
)
5414 unsigned long rebroadcast_time
, warning_time
;
5417 linkwatch_forget_dev(dev
);
5419 rebroadcast_time
= warning_time
= jiffies
;
5420 refcnt
= netdev_refcnt_read(dev
);
5422 while (refcnt
!= 0) {
5423 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5426 /* Rebroadcast unregister notification */
5427 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5428 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5429 * should have already handle it the first time */
5431 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5433 /* We must not have linkwatch events
5434 * pending on unregister. If this
5435 * happens, we simply run the queue
5436 * unscheduled, resulting in a noop
5439 linkwatch_run_queue();
5444 rebroadcast_time
= jiffies
;
5449 refcnt
= netdev_refcnt_read(dev
);
5451 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5452 printk(KERN_EMERG
"unregister_netdevice: "
5453 "waiting for %s to become free. Usage "
5456 warning_time
= jiffies
;
5465 * register_netdevice(x1);
5466 * register_netdevice(x2);
5468 * unregister_netdevice(y1);
5469 * unregister_netdevice(y2);
5475 * We are invoked by rtnl_unlock().
5476 * This allows us to deal with problems:
5477 * 1) We can delete sysfs objects which invoke hotplug
5478 * without deadlocking with linkwatch via keventd.
5479 * 2) Since we run with the RTNL semaphore not held, we can sleep
5480 * safely in order to wait for the netdev refcnt to drop to zero.
5482 * We must not return until all unregister events added during
5483 * the interval the lock was held have been completed.
5485 void netdev_run_todo(void)
5487 struct list_head list
;
5489 /* Snapshot list, allow later requests */
5490 list_replace_init(&net_todo_list
, &list
);
5494 while (!list_empty(&list
)) {
5495 struct net_device
*dev
5496 = list_first_entry(&list
, struct net_device
, todo_list
);
5497 list_del(&dev
->todo_list
);
5499 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5500 printk(KERN_ERR
"network todo '%s' but state %d\n",
5501 dev
->name
, dev
->reg_state
);
5506 dev
->reg_state
= NETREG_UNREGISTERED
;
5508 on_each_cpu(flush_backlog
, dev
, 1);
5510 netdev_wait_allrefs(dev
);
5513 BUG_ON(netdev_refcnt_read(dev
));
5514 WARN_ON(rcu_dereference_raw(dev
->ip_ptr
));
5515 WARN_ON(rcu_dereference_raw(dev
->ip6_ptr
));
5516 WARN_ON(dev
->dn_ptr
);
5518 if (dev
->destructor
)
5519 dev
->destructor(dev
);
5521 /* Free network device */
5522 kobject_put(&dev
->dev
.kobj
);
5527 * dev_txq_stats_fold - fold tx_queues stats
5528 * @dev: device to get statistics from
5529 * @stats: struct rtnl_link_stats64 to hold results
5531 void dev_txq_stats_fold(const struct net_device
*dev
,
5532 struct rtnl_link_stats64
*stats
)
5534 u64 tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5536 struct netdev_queue
*txq
;
5538 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5539 txq
= netdev_get_tx_queue(dev
, i
);
5540 spin_lock_bh(&txq
->_xmit_lock
);
5541 tx_bytes
+= txq
->tx_bytes
;
5542 tx_packets
+= txq
->tx_packets
;
5543 tx_dropped
+= txq
->tx_dropped
;
5544 spin_unlock_bh(&txq
->_xmit_lock
);
5546 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5547 stats
->tx_bytes
= tx_bytes
;
5548 stats
->tx_packets
= tx_packets
;
5549 stats
->tx_dropped
= tx_dropped
;
5552 EXPORT_SYMBOL(dev_txq_stats_fold
);
5554 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5555 * fields in the same order, with only the type differing.
5557 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5558 const struct net_device_stats
*netdev_stats
)
5560 #if BITS_PER_LONG == 64
5561 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5562 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5564 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5565 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5566 u64
*dst
= (u64
*)stats64
;
5568 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5569 sizeof(*stats64
) / sizeof(u64
));
5570 for (i
= 0; i
< n
; i
++)
5576 * dev_get_stats - get network device statistics
5577 * @dev: device to get statistics from
5578 * @storage: place to store stats
5580 * Get network statistics from device. Return @storage.
5581 * The device driver may provide its own method by setting
5582 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5583 * otherwise the internal statistics structure is used.
5585 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5586 struct rtnl_link_stats64
*storage
)
5588 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5590 if (ops
->ndo_get_stats64
) {
5591 memset(storage
, 0, sizeof(*storage
));
5592 ops
->ndo_get_stats64(dev
, storage
);
5593 } else if (ops
->ndo_get_stats
) {
5594 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5596 netdev_stats_to_stats64(storage
, &dev
->stats
);
5597 dev_txq_stats_fold(dev
, storage
);
5599 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5602 EXPORT_SYMBOL(dev_get_stats
);
5604 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5606 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5608 #ifdef CONFIG_NET_CLS_ACT
5611 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5614 netdev_init_one_queue(dev
, queue
, NULL
);
5615 queue
->qdisc
= &noop_qdisc
;
5616 queue
->qdisc_sleeping
= &noop_qdisc
;
5617 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5623 * alloc_netdev_mqs - allocate network device
5624 * @sizeof_priv: size of private data to allocate space for
5625 * @name: device name format string
5626 * @setup: callback to initialize device
5627 * @txqs: the number of TX subqueues to allocate
5628 * @rxqs: the number of RX subqueues to allocate
5630 * Allocates a struct net_device with private data area for driver use
5631 * and performs basic initialization. Also allocates subquue structs
5632 * for each queue on the device.
5634 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5635 void (*setup
)(struct net_device
*),
5636 unsigned int txqs
, unsigned int rxqs
)
5638 struct net_device
*dev
;
5640 struct net_device
*p
;
5642 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5645 pr_err("alloc_netdev: Unable to allocate device "
5646 "with zero queues.\n");
5652 pr_err("alloc_netdev: Unable to allocate device "
5653 "with zero RX queues.\n");
5658 alloc_size
= sizeof(struct net_device
);
5660 /* ensure 32-byte alignment of private area */
5661 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5662 alloc_size
+= sizeof_priv
;
5664 /* ensure 32-byte alignment of whole construct */
5665 alloc_size
+= NETDEV_ALIGN
- 1;
5667 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5669 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5673 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5674 dev
->padded
= (char *)dev
- (char *)p
;
5676 dev
->pcpu_refcnt
= alloc_percpu(int);
5677 if (!dev
->pcpu_refcnt
)
5680 if (dev_addr_init(dev
))
5686 dev_net_set(dev
, &init_net
);
5688 dev
->num_tx_queues
= txqs
;
5689 dev
->real_num_tx_queues
= txqs
;
5690 if (netif_alloc_netdev_queues(dev
))
5694 dev
->num_rx_queues
= rxqs
;
5695 dev
->real_num_rx_queues
= rxqs
;
5696 if (netif_alloc_rx_queues(dev
))
5700 dev
->gso_max_size
= GSO_MAX_SIZE
;
5702 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5703 dev
->ethtool_ntuple_list
.count
= 0;
5704 INIT_LIST_HEAD(&dev
->napi_list
);
5705 INIT_LIST_HEAD(&dev
->unreg_list
);
5706 INIT_LIST_HEAD(&dev
->link_watch_list
);
5707 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5709 strcpy(dev
->name
, name
);
5713 free_percpu(dev
->pcpu_refcnt
);
5723 EXPORT_SYMBOL(alloc_netdev_mqs
);
5726 * free_netdev - free network device
5729 * This function does the last stage of destroying an allocated device
5730 * interface. The reference to the device object is released.
5731 * If this is the last reference then it will be freed.
5733 void free_netdev(struct net_device
*dev
)
5735 struct napi_struct
*p
, *n
;
5737 release_net(dev_net(dev
));
5744 kfree(rcu_dereference_raw(dev
->ingress_queue
));
5746 /* Flush device addresses */
5747 dev_addr_flush(dev
);
5749 /* Clear ethtool n-tuple list */
5750 ethtool_ntuple_flush(dev
);
5752 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5755 free_percpu(dev
->pcpu_refcnt
);
5756 dev
->pcpu_refcnt
= NULL
;
5758 /* Compatibility with error handling in drivers */
5759 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5760 kfree((char *)dev
- dev
->padded
);
5764 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5765 dev
->reg_state
= NETREG_RELEASED
;
5767 /* will free via device release */
5768 put_device(&dev
->dev
);
5770 EXPORT_SYMBOL(free_netdev
);
5773 * synchronize_net - Synchronize with packet receive processing
5775 * Wait for packets currently being received to be done.
5776 * Does not block later packets from starting.
5778 void synchronize_net(void)
5783 EXPORT_SYMBOL(synchronize_net
);
5786 * unregister_netdevice_queue - remove device from the kernel
5790 * This function shuts down a device interface and removes it
5791 * from the kernel tables.
5792 * If head not NULL, device is queued to be unregistered later.
5794 * Callers must hold the rtnl semaphore. You may want
5795 * unregister_netdev() instead of this.
5798 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5803 list_move_tail(&dev
->unreg_list
, head
);
5805 rollback_registered(dev
);
5806 /* Finish processing unregister after unlock */
5810 EXPORT_SYMBOL(unregister_netdevice_queue
);
5813 * unregister_netdevice_many - unregister many devices
5814 * @head: list of devices
5816 void unregister_netdevice_many(struct list_head
*head
)
5818 struct net_device
*dev
;
5820 if (!list_empty(head
)) {
5821 rollback_registered_many(head
);
5822 list_for_each_entry(dev
, head
, unreg_list
)
5826 EXPORT_SYMBOL(unregister_netdevice_many
);
5829 * unregister_netdev - remove device from the kernel
5832 * This function shuts down a device interface and removes it
5833 * from the kernel tables.
5835 * This is just a wrapper for unregister_netdevice that takes
5836 * the rtnl semaphore. In general you want to use this and not
5837 * unregister_netdevice.
5839 void unregister_netdev(struct net_device
*dev
)
5842 unregister_netdevice(dev
);
5845 EXPORT_SYMBOL(unregister_netdev
);
5848 * dev_change_net_namespace - move device to different nethost namespace
5850 * @net: network namespace
5851 * @pat: If not NULL name pattern to try if the current device name
5852 * is already taken in the destination network namespace.
5854 * This function shuts down a device interface and moves it
5855 * to a new network namespace. On success 0 is returned, on
5856 * a failure a netagive errno code is returned.
5858 * Callers must hold the rtnl semaphore.
5861 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5867 /* Don't allow namespace local devices to be moved. */
5869 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5872 /* Ensure the device has been registrered */
5874 if (dev
->reg_state
!= NETREG_REGISTERED
)
5877 /* Get out if there is nothing todo */
5879 if (net_eq(dev_net(dev
), net
))
5882 /* Pick the destination device name, and ensure
5883 * we can use it in the destination network namespace.
5886 if (__dev_get_by_name(net
, dev
->name
)) {
5887 /* We get here if we can't use the current device name */
5890 if (dev_get_valid_name(dev
, pat
, 1))
5895 * And now a mini version of register_netdevice unregister_netdevice.
5898 /* If device is running close it first. */
5901 /* And unlink it from device chain */
5903 unlist_netdevice(dev
);
5907 /* Shutdown queueing discipline. */
5910 /* Notify protocols, that we are about to destroy
5911 this device. They should clean all the things.
5913 Note that dev->reg_state stays at NETREG_REGISTERED.
5914 This is wanted because this way 8021q and macvlan know
5915 the device is just moving and can keep their slaves up.
5917 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5918 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5921 * Flush the unicast and multicast chains
5926 /* Actually switch the network namespace */
5927 dev_net_set(dev
, net
);
5929 /* If there is an ifindex conflict assign a new one */
5930 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5931 int iflink
= (dev
->iflink
== dev
->ifindex
);
5932 dev
->ifindex
= dev_new_index(net
);
5934 dev
->iflink
= dev
->ifindex
;
5937 /* Fixup kobjects */
5938 err
= device_rename(&dev
->dev
, dev
->name
);
5941 /* Add the device back in the hashes */
5942 list_netdevice(dev
);
5944 /* Notify protocols, that a new device appeared. */
5945 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5948 * Prevent userspace races by waiting until the network
5949 * device is fully setup before sending notifications.
5951 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5958 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5960 static int dev_cpu_callback(struct notifier_block
*nfb
,
5961 unsigned long action
,
5964 struct sk_buff
**list_skb
;
5965 struct sk_buff
*skb
;
5966 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5967 struct softnet_data
*sd
, *oldsd
;
5969 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5972 local_irq_disable();
5973 cpu
= smp_processor_id();
5974 sd
= &per_cpu(softnet_data
, cpu
);
5975 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5977 /* Find end of our completion_queue. */
5978 list_skb
= &sd
->completion_queue
;
5980 list_skb
= &(*list_skb
)->next
;
5981 /* Append completion queue from offline CPU. */
5982 *list_skb
= oldsd
->completion_queue
;
5983 oldsd
->completion_queue
= NULL
;
5985 /* Append output queue from offline CPU. */
5986 if (oldsd
->output_queue
) {
5987 *sd
->output_queue_tailp
= oldsd
->output_queue
;
5988 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
5989 oldsd
->output_queue
= NULL
;
5990 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
5993 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5996 /* Process offline CPU's input_pkt_queue */
5997 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
5999 input_queue_head_incr(oldsd
);
6001 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6003 input_queue_head_incr(oldsd
);
6011 * netdev_increment_features - increment feature set by one
6012 * @all: current feature set
6013 * @one: new feature set
6014 * @mask: mask feature set
6016 * Computes a new feature set after adding a device with feature set
6017 * @one to the master device with current feature set @all. Will not
6018 * enable anything that is off in @mask. Returns the new feature set.
6020 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
6023 /* If device needs checksumming, downgrade to it. */
6024 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
6025 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
6026 else if (mask
& NETIF_F_ALL_CSUM
) {
6027 /* If one device supports v4/v6 checksumming, set for all. */
6028 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
6029 !(all
& NETIF_F_GEN_CSUM
)) {
6030 all
&= ~NETIF_F_ALL_CSUM
;
6031 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
6034 /* If one device supports hw checksumming, set for all. */
6035 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
6036 all
&= ~NETIF_F_ALL_CSUM
;
6037 all
|= NETIF_F_HW_CSUM
;
6041 one
|= NETIF_F_ALL_CSUM
;
6043 one
|= all
& NETIF_F_ONE_FOR_ALL
;
6044 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
6045 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
6049 EXPORT_SYMBOL(netdev_increment_features
);
6051 static struct hlist_head
*netdev_create_hash(void)
6054 struct hlist_head
*hash
;
6056 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6058 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6059 INIT_HLIST_HEAD(&hash
[i
]);
6064 /* Initialize per network namespace state */
6065 static int __net_init
netdev_init(struct net
*net
)
6067 INIT_LIST_HEAD(&net
->dev_base_head
);
6069 net
->dev_name_head
= netdev_create_hash();
6070 if (net
->dev_name_head
== NULL
)
6073 net
->dev_index_head
= netdev_create_hash();
6074 if (net
->dev_index_head
== NULL
)
6080 kfree(net
->dev_name_head
);
6086 * netdev_drivername - network driver for the device
6087 * @dev: network device
6088 * @buffer: buffer for resulting name
6089 * @len: size of buffer
6091 * Determine network driver for device.
6093 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
6095 const struct device_driver
*driver
;
6096 const struct device
*parent
;
6098 if (len
<= 0 || !buffer
)
6102 parent
= dev
->dev
.parent
;
6107 driver
= parent
->driver
;
6108 if (driver
&& driver
->name
)
6109 strlcpy(buffer
, driver
->name
, len
);
6113 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6114 struct va_format
*vaf
)
6118 if (dev
&& dev
->dev
.parent
)
6119 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6120 netdev_name(dev
), vaf
);
6122 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6124 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6129 int netdev_printk(const char *level
, const struct net_device
*dev
,
6130 const char *format
, ...)
6132 struct va_format vaf
;
6136 va_start(args
, format
);
6141 r
= __netdev_printk(level
, dev
, &vaf
);
6146 EXPORT_SYMBOL(netdev_printk
);
6148 #define define_netdev_printk_level(func, level) \
6149 int func(const struct net_device *dev, const char *fmt, ...) \
6152 struct va_format vaf; \
6155 va_start(args, fmt); \
6160 r = __netdev_printk(level, dev, &vaf); \
6165 EXPORT_SYMBOL(func);
6167 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6168 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6169 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6170 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6171 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6172 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6173 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6175 static void __net_exit
netdev_exit(struct net
*net
)
6177 kfree(net
->dev_name_head
);
6178 kfree(net
->dev_index_head
);
6181 static struct pernet_operations __net_initdata netdev_net_ops
= {
6182 .init
= netdev_init
,
6183 .exit
= netdev_exit
,
6186 static void __net_exit
default_device_exit(struct net
*net
)
6188 struct net_device
*dev
, *aux
;
6190 * Push all migratable network devices back to the
6191 * initial network namespace
6194 for_each_netdev_safe(net
, dev
, aux
) {
6196 char fb_name
[IFNAMSIZ
];
6198 /* Ignore unmoveable devices (i.e. loopback) */
6199 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6202 /* Leave virtual devices for the generic cleanup */
6203 if (dev
->rtnl_link_ops
)
6206 /* Push remaing network devices to init_net */
6207 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6208 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6210 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
6211 __func__
, dev
->name
, err
);
6218 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6220 /* At exit all network devices most be removed from a network
6221 * namespace. Do this in the reverse order of registeration.
6222 * Do this across as many network namespaces as possible to
6223 * improve batching efficiency.
6225 struct net_device
*dev
;
6227 LIST_HEAD(dev_kill_list
);
6230 list_for_each_entry(net
, net_list
, exit_list
) {
6231 for_each_netdev_reverse(net
, dev
) {
6232 if (dev
->rtnl_link_ops
)
6233 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6235 unregister_netdevice_queue(dev
, &dev_kill_list
);
6238 unregister_netdevice_many(&dev_kill_list
);
6242 static struct pernet_operations __net_initdata default_device_ops
= {
6243 .exit
= default_device_exit
,
6244 .exit_batch
= default_device_exit_batch
,
6248 * Initialize the DEV module. At boot time this walks the device list and
6249 * unhooks any devices that fail to initialise (normally hardware not
6250 * present) and leaves us with a valid list of present and active devices.
6255 * This is called single threaded during boot, so no need
6256 * to take the rtnl semaphore.
6258 static int __init
net_dev_init(void)
6260 int i
, rc
= -ENOMEM
;
6262 BUG_ON(!dev_boot_phase
);
6264 if (dev_proc_init())
6267 if (netdev_kobject_init())
6270 INIT_LIST_HEAD(&ptype_all
);
6271 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6272 INIT_LIST_HEAD(&ptype_base
[i
]);
6274 if (register_pernet_subsys(&netdev_net_ops
))
6278 * Initialise the packet receive queues.
6281 for_each_possible_cpu(i
) {
6282 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6284 memset(sd
, 0, sizeof(*sd
));
6285 skb_queue_head_init(&sd
->input_pkt_queue
);
6286 skb_queue_head_init(&sd
->process_queue
);
6287 sd
->completion_queue
= NULL
;
6288 INIT_LIST_HEAD(&sd
->poll_list
);
6289 sd
->output_queue
= NULL
;
6290 sd
->output_queue_tailp
= &sd
->output_queue
;
6292 sd
->csd
.func
= rps_trigger_softirq
;
6298 sd
->backlog
.poll
= process_backlog
;
6299 sd
->backlog
.weight
= weight_p
;
6300 sd
->backlog
.gro_list
= NULL
;
6301 sd
->backlog
.gro_count
= 0;
6306 /* The loopback device is special if any other network devices
6307 * is present in a network namespace the loopback device must
6308 * be present. Since we now dynamically allocate and free the
6309 * loopback device ensure this invariant is maintained by
6310 * keeping the loopback device as the first device on the
6311 * list of network devices. Ensuring the loopback devices
6312 * is the first device that appears and the last network device
6315 if (register_pernet_device(&loopback_net_ops
))
6318 if (register_pernet_device(&default_device_ops
))
6321 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6322 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6324 hotcpu_notifier(dev_cpu_callback
, 0);
6332 subsys_initcall(net_dev_init
);
6334 static int __init
initialize_hashrnd(void)
6336 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6340 late_initcall_sync(initialize_hashrnd
);