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.
753 * The caller must hold RCU or RTNL.
754 * The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
759 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
762 struct net_device
*dev
;
764 for_each_netdev_rcu(net
, dev
)
765 if (dev
->type
== type
&&
766 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
771 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
773 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
775 struct net_device
*dev
;
778 for_each_netdev(net
, dev
)
779 if (dev
->type
== type
)
784 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
786 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
788 struct net_device
*dev
, *ret
= NULL
;
791 for_each_netdev_rcu(net
, dev
)
792 if (dev
->type
== type
) {
800 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
803 * dev_get_by_flags_rcu - find any device with given flags
804 * @net: the applicable net namespace
805 * @if_flags: IFF_* values
806 * @mask: bitmask of bits in if_flags to check
808 * Search for any interface with the given flags. Returns NULL if a device
809 * is not found or a pointer to the device. Must be called inside
810 * rcu_read_lock(), and result refcount is unchanged.
813 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
816 struct net_device
*dev
, *ret
;
819 for_each_netdev_rcu(net
, dev
) {
820 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
827 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
830 * dev_valid_name - check if name is okay for network device
833 * Network device names need to be valid file names to
834 * to allow sysfs to work. We also disallow any kind of
837 int dev_valid_name(const char *name
)
841 if (strlen(name
) >= IFNAMSIZ
)
843 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
847 if (*name
== '/' || isspace(*name
))
853 EXPORT_SYMBOL(dev_valid_name
);
856 * __dev_alloc_name - allocate a name for a device
857 * @net: network namespace to allocate the device name in
858 * @name: name format string
859 * @buf: scratch buffer and result name string
861 * Passed a format string - eg "lt%d" it will try and find a suitable
862 * id. It scans list of devices to build up a free map, then chooses
863 * the first empty slot. The caller must hold the dev_base or rtnl lock
864 * while allocating the name and adding the device in order to avoid
866 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
867 * Returns the number of the unit assigned or a negative errno code.
870 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
874 const int max_netdevices
= 8*PAGE_SIZE
;
875 unsigned long *inuse
;
876 struct net_device
*d
;
878 p
= strnchr(name
, IFNAMSIZ
-1, '%');
881 * Verify the string as this thing may have come from
882 * the user. There must be either one "%d" and no other "%"
885 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
888 /* Use one page as a bit array of possible slots */
889 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
893 for_each_netdev(net
, d
) {
894 if (!sscanf(d
->name
, name
, &i
))
896 if (i
< 0 || i
>= max_netdevices
)
899 /* avoid cases where sscanf is not exact inverse of printf */
900 snprintf(buf
, IFNAMSIZ
, name
, i
);
901 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
905 i
= find_first_zero_bit(inuse
, max_netdevices
);
906 free_page((unsigned long) inuse
);
910 snprintf(buf
, IFNAMSIZ
, name
, i
);
911 if (!__dev_get_by_name(net
, buf
))
914 /* It is possible to run out of possible slots
915 * when the name is long and there isn't enough space left
916 * for the digits, or if all bits are used.
922 * dev_alloc_name - allocate a name for a device
924 * @name: name format string
926 * Passed a format string - eg "lt%d" it will try and find a suitable
927 * id. It scans list of devices to build up a free map, then chooses
928 * the first empty slot. The caller must hold the dev_base or rtnl lock
929 * while allocating the name and adding the device in order to avoid
931 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
932 * Returns the number of the unit assigned or a negative errno code.
935 int dev_alloc_name(struct net_device
*dev
, const char *name
)
941 BUG_ON(!dev_net(dev
));
943 ret
= __dev_alloc_name(net
, name
, buf
);
945 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
948 EXPORT_SYMBOL(dev_alloc_name
);
950 static int dev_get_valid_name(struct net_device
*dev
, const char *name
, bool fmt
)
954 BUG_ON(!dev_net(dev
));
957 if (!dev_valid_name(name
))
960 if (fmt
&& strchr(name
, '%'))
961 return dev_alloc_name(dev
, name
);
962 else if (__dev_get_by_name(net
, name
))
964 else if (dev
->name
!= name
)
965 strlcpy(dev
->name
, name
, IFNAMSIZ
);
971 * dev_change_name - change name of a device
973 * @newname: name (or format string) must be at least IFNAMSIZ
975 * Change name of a device, can pass format strings "eth%d".
978 int dev_change_name(struct net_device
*dev
, const char *newname
)
980 char oldname
[IFNAMSIZ
];
986 BUG_ON(!dev_net(dev
));
989 if (dev
->flags
& IFF_UP
)
992 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
995 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
997 err
= dev_get_valid_name(dev
, newname
, 1);
1002 ret
= device_rename(&dev
->dev
, dev
->name
);
1004 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1008 write_lock_bh(&dev_base_lock
);
1009 hlist_del(&dev
->name_hlist
);
1010 write_unlock_bh(&dev_base_lock
);
1014 write_lock_bh(&dev_base_lock
);
1015 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1016 write_unlock_bh(&dev_base_lock
);
1018 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1019 ret
= notifier_to_errno(ret
);
1022 /* err >= 0 after dev_alloc_name() or stores the first errno */
1025 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1029 "%s: name change rollback failed: %d.\n",
1038 * dev_set_alias - change ifalias of a device
1040 * @alias: name up to IFALIASZ
1041 * @len: limit of bytes to copy from info
1043 * Set ifalias for a device,
1045 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1049 if (len
>= IFALIASZ
)
1054 kfree(dev
->ifalias
);
1055 dev
->ifalias
= NULL
;
1060 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1064 strlcpy(dev
->ifalias
, alias
, len
+1);
1070 * netdev_features_change - device changes features
1071 * @dev: device to cause notification
1073 * Called to indicate a device has changed features.
1075 void netdev_features_change(struct net_device
*dev
)
1077 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1079 EXPORT_SYMBOL(netdev_features_change
);
1082 * netdev_state_change - device changes state
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed state. This function calls
1086 * the notifier chains for netdev_chain and sends a NEWLINK message
1087 * to the routing socket.
1089 void netdev_state_change(struct net_device
*dev
)
1091 if (dev
->flags
& IFF_UP
) {
1092 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1093 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1096 EXPORT_SYMBOL(netdev_state_change
);
1098 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1100 return call_netdevice_notifiers(event
, dev
);
1102 EXPORT_SYMBOL(netdev_bonding_change
);
1105 * dev_load - load a network module
1106 * @net: the applicable net namespace
1107 * @name: name of interface
1109 * If a network interface is not present and the process has suitable
1110 * privileges this function loads the module. If module loading is not
1111 * available in this kernel then it becomes a nop.
1114 void dev_load(struct net
*net
, const char *name
)
1116 struct net_device
*dev
;
1119 dev
= dev_get_by_name_rcu(net
, name
);
1122 if (!dev
&& capable(CAP_NET_ADMIN
))
1123 request_module("%s", name
);
1125 EXPORT_SYMBOL(dev_load
);
1127 static int __dev_open(struct net_device
*dev
)
1129 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1135 * Is it even present?
1137 if (!netif_device_present(dev
))
1140 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1141 ret
= notifier_to_errno(ret
);
1146 * Call device private open method
1148 set_bit(__LINK_STATE_START
, &dev
->state
);
1150 if (ops
->ndo_validate_addr
)
1151 ret
= ops
->ndo_validate_addr(dev
);
1153 if (!ret
&& ops
->ndo_open
)
1154 ret
= ops
->ndo_open(dev
);
1157 * If it went open OK then:
1161 clear_bit(__LINK_STATE_START
, &dev
->state
);
1166 dev
->flags
|= IFF_UP
;
1171 net_dmaengine_get();
1174 * Initialize multicasting status
1176 dev_set_rx_mode(dev
);
1179 * Wakeup transmit queue engine
1188 * dev_open - prepare an interface for use.
1189 * @dev: device to open
1191 * Takes a device from down to up state. The device's private open
1192 * function is invoked and then the multicast lists are loaded. Finally
1193 * the device is moved into the up state and a %NETDEV_UP message is
1194 * sent to the netdev notifier chain.
1196 * Calling this function on an active interface is a nop. On a failure
1197 * a negative errno code is returned.
1199 int dev_open(struct net_device
*dev
)
1206 if (dev
->flags
& IFF_UP
)
1212 ret
= __dev_open(dev
);
1217 * ... and announce new interface.
1219 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1220 call_netdevice_notifiers(NETDEV_UP
, dev
);
1224 EXPORT_SYMBOL(dev_open
);
1226 static int __dev_close_many(struct list_head
*head
)
1228 struct net_device
*dev
;
1233 list_for_each_entry(dev
, head
, unreg_list
) {
1235 * Tell people we are going down, so that they can
1236 * prepare to death, when device is still operating.
1238 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1240 clear_bit(__LINK_STATE_START
, &dev
->state
);
1242 /* Synchronize to scheduled poll. We cannot touch poll list, it
1243 * can be even on different cpu. So just clear netif_running().
1245 * dev->stop() will invoke napi_disable() on all of it's
1246 * napi_struct instances on this device.
1248 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1251 dev_deactivate_many(head
);
1253 list_for_each_entry(dev
, head
, unreg_list
) {
1254 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1257 * Call the device specific close. This cannot fail.
1258 * Only if device is UP
1260 * We allow it to be called even after a DETACH hot-plug
1267 * Device is now down.
1270 dev
->flags
&= ~IFF_UP
;
1275 net_dmaengine_put();
1281 static int __dev_close(struct net_device
*dev
)
1286 list_add(&dev
->unreg_list
, &single
);
1287 retval
= __dev_close_many(&single
);
1292 int dev_close_many(struct list_head
*head
)
1294 struct net_device
*dev
, *tmp
;
1295 LIST_HEAD(tmp_list
);
1297 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1298 if (!(dev
->flags
& IFF_UP
))
1299 list_move(&dev
->unreg_list
, &tmp_list
);
1301 __dev_close_many(head
);
1304 * Tell people we are down
1306 list_for_each_entry(dev
, head
, unreg_list
) {
1307 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1308 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1311 /* rollback_registered_many needs the complete original list */
1312 list_splice(&tmp_list
, head
);
1317 * dev_close - shutdown an interface.
1318 * @dev: device to shutdown
1320 * This function moves an active device into down state. A
1321 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1322 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1325 int dev_close(struct net_device
*dev
)
1329 list_add(&dev
->unreg_list
, &single
);
1330 dev_close_many(&single
);
1334 EXPORT_SYMBOL(dev_close
);
1338 * dev_disable_lro - disable Large Receive Offload on a device
1341 * Disable Large Receive Offload (LRO) on a net device. Must be
1342 * called under RTNL. This is needed if received packets may be
1343 * forwarded to another interface.
1345 void dev_disable_lro(struct net_device
*dev
)
1347 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1348 dev
->ethtool_ops
->set_flags
) {
1349 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1350 if (flags
& ETH_FLAG_LRO
) {
1351 flags
&= ~ETH_FLAG_LRO
;
1352 dev
->ethtool_ops
->set_flags(dev
, flags
);
1355 WARN_ON(dev
->features
& NETIF_F_LRO
);
1357 EXPORT_SYMBOL(dev_disable_lro
);
1360 static int dev_boot_phase
= 1;
1363 * Device change register/unregister. These are not inline or static
1364 * as we export them to the world.
1368 * register_netdevice_notifier - register a network notifier block
1371 * Register a notifier to be called when network device events occur.
1372 * The notifier passed is linked into the kernel structures and must
1373 * not be reused until it has been unregistered. A negative errno code
1374 * is returned on a failure.
1376 * When registered all registration and up events are replayed
1377 * to the new notifier to allow device to have a race free
1378 * view of the network device list.
1381 int register_netdevice_notifier(struct notifier_block
*nb
)
1383 struct net_device
*dev
;
1384 struct net_device
*last
;
1389 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1395 for_each_netdev(net
, dev
) {
1396 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1397 err
= notifier_to_errno(err
);
1401 if (!(dev
->flags
& IFF_UP
))
1404 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1415 for_each_netdev(net
, dev
) {
1419 if (dev
->flags
& IFF_UP
) {
1420 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1421 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1423 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1424 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1428 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1431 EXPORT_SYMBOL(register_netdevice_notifier
);
1434 * unregister_netdevice_notifier - unregister a network notifier block
1437 * Unregister a notifier previously registered by
1438 * register_netdevice_notifier(). The notifier is unlinked into the
1439 * kernel structures and may then be reused. A negative errno code
1440 * is returned on a failure.
1443 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1448 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1452 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1455 * call_netdevice_notifiers - call all network notifier blocks
1456 * @val: value passed unmodified to notifier function
1457 * @dev: net_device pointer passed unmodified to notifier function
1459 * Call all network notifier blocks. Parameters and return value
1460 * are as for raw_notifier_call_chain().
1463 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1466 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1469 /* When > 0 there are consumers of rx skb time stamps */
1470 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1472 void net_enable_timestamp(void)
1474 atomic_inc(&netstamp_needed
);
1476 EXPORT_SYMBOL(net_enable_timestamp
);
1478 void net_disable_timestamp(void)
1480 atomic_dec(&netstamp_needed
);
1482 EXPORT_SYMBOL(net_disable_timestamp
);
1484 static inline void net_timestamp_set(struct sk_buff
*skb
)
1486 if (atomic_read(&netstamp_needed
))
1487 __net_timestamp(skb
);
1489 skb
->tstamp
.tv64
= 0;
1492 static inline void net_timestamp_check(struct sk_buff
*skb
)
1494 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1495 __net_timestamp(skb
);
1499 * dev_forward_skb - loopback an skb to another netif
1501 * @dev: destination network device
1502 * @skb: buffer to forward
1505 * NET_RX_SUCCESS (no congestion)
1506 * NET_RX_DROP (packet was dropped, but freed)
1508 * dev_forward_skb can be used for injecting an skb from the
1509 * start_xmit function of one device into the receive queue
1510 * of another device.
1512 * The receiving device may be in another namespace, so
1513 * we have to clear all information in the skb that could
1514 * impact namespace isolation.
1516 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1521 if (unlikely(!(dev
->flags
& IFF_UP
) ||
1522 (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
)))) {
1523 atomic_long_inc(&dev
->rx_dropped
);
1527 skb_set_dev(skb
, dev
);
1528 skb
->tstamp
.tv64
= 0;
1529 skb
->pkt_type
= PACKET_HOST
;
1530 skb
->protocol
= eth_type_trans(skb
, dev
);
1531 return netif_rx(skb
);
1533 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1535 static inline int deliver_skb(struct sk_buff
*skb
,
1536 struct packet_type
*pt_prev
,
1537 struct net_device
*orig_dev
)
1539 atomic_inc(&skb
->users
);
1540 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1544 * Support routine. Sends outgoing frames to any network
1545 * taps currently in use.
1548 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1550 struct packet_type
*ptype
;
1551 struct sk_buff
*skb2
= NULL
;
1552 struct packet_type
*pt_prev
= NULL
;
1555 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1556 /* Never send packets back to the socket
1557 * they originated from - MvS (miquels@drinkel.ow.org)
1559 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1560 (ptype
->af_packet_priv
== NULL
||
1561 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1563 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1568 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1572 net_timestamp_set(skb2
);
1574 /* skb->nh should be correctly
1575 set by sender, so that the second statement is
1576 just protection against buggy protocols.
1578 skb_reset_mac_header(skb2
);
1580 if (skb_network_header(skb2
) < skb2
->data
||
1581 skb2
->network_header
> skb2
->tail
) {
1582 if (net_ratelimit())
1583 printk(KERN_CRIT
"protocol %04x is "
1585 ntohs(skb2
->protocol
),
1587 skb_reset_network_header(skb2
);
1590 skb2
->transport_header
= skb2
->network_header
;
1591 skb2
->pkt_type
= PACKET_OUTGOING
;
1596 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1601 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1602 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1604 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1608 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1611 if (dev
->reg_state
== NETREG_REGISTERED
) {
1614 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1619 if (txq
< dev
->real_num_tx_queues
)
1620 qdisc_reset_all_tx_gt(dev
, txq
);
1623 dev
->real_num_tx_queues
= txq
;
1626 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1630 * netif_set_real_num_rx_queues - set actual number of RX queues used
1631 * @dev: Network device
1632 * @rxq: Actual number of RX queues
1634 * This must be called either with the rtnl_lock held or before
1635 * registration of the net device. Returns 0 on success, or a
1636 * negative error code. If called before registration, it always
1639 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1643 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1646 if (dev
->reg_state
== NETREG_REGISTERED
) {
1649 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1655 dev
->real_num_rx_queues
= rxq
;
1658 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1661 static inline void __netif_reschedule(struct Qdisc
*q
)
1663 struct softnet_data
*sd
;
1664 unsigned long flags
;
1666 local_irq_save(flags
);
1667 sd
= &__get_cpu_var(softnet_data
);
1668 q
->next_sched
= NULL
;
1669 *sd
->output_queue_tailp
= q
;
1670 sd
->output_queue_tailp
= &q
->next_sched
;
1671 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1672 local_irq_restore(flags
);
1675 void __netif_schedule(struct Qdisc
*q
)
1677 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1678 __netif_reschedule(q
);
1680 EXPORT_SYMBOL(__netif_schedule
);
1682 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1684 if (atomic_dec_and_test(&skb
->users
)) {
1685 struct softnet_data
*sd
;
1686 unsigned long flags
;
1688 local_irq_save(flags
);
1689 sd
= &__get_cpu_var(softnet_data
);
1690 skb
->next
= sd
->completion_queue
;
1691 sd
->completion_queue
= skb
;
1692 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1693 local_irq_restore(flags
);
1696 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1698 void dev_kfree_skb_any(struct sk_buff
*skb
)
1700 if (in_irq() || irqs_disabled())
1701 dev_kfree_skb_irq(skb
);
1705 EXPORT_SYMBOL(dev_kfree_skb_any
);
1709 * netif_device_detach - mark device as removed
1710 * @dev: network device
1712 * Mark device as removed from system and therefore no longer available.
1714 void netif_device_detach(struct net_device
*dev
)
1716 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1717 netif_running(dev
)) {
1718 netif_tx_stop_all_queues(dev
);
1721 EXPORT_SYMBOL(netif_device_detach
);
1724 * netif_device_attach - mark device as attached
1725 * @dev: network device
1727 * Mark device as attached from system and restart if needed.
1729 void netif_device_attach(struct net_device
*dev
)
1731 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1732 netif_running(dev
)) {
1733 netif_tx_wake_all_queues(dev
);
1734 __netdev_watchdog_up(dev
);
1737 EXPORT_SYMBOL(netif_device_attach
);
1740 * skb_dev_set -- assign a new device to a buffer
1741 * @skb: buffer for the new device
1742 * @dev: network device
1744 * If an skb is owned by a device already, we have to reset
1745 * all data private to the namespace a device belongs to
1746 * before assigning it a new device.
1748 #ifdef CONFIG_NET_NS
1749 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1752 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1755 skb_init_secmark(skb
);
1759 skb
->ipvs_property
= 0;
1760 #ifdef CONFIG_NET_SCHED
1766 EXPORT_SYMBOL(skb_set_dev
);
1767 #endif /* CONFIG_NET_NS */
1770 * Invalidate hardware checksum when packet is to be mangled, and
1771 * complete checksum manually on outgoing path.
1773 int skb_checksum_help(struct sk_buff
*skb
)
1776 int ret
= 0, offset
;
1778 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1779 goto out_set_summed
;
1781 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1782 /* Let GSO fix up the checksum. */
1783 goto out_set_summed
;
1786 offset
= skb_checksum_start_offset(skb
);
1787 BUG_ON(offset
>= skb_headlen(skb
));
1788 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1790 offset
+= skb
->csum_offset
;
1791 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1793 if (skb_cloned(skb
) &&
1794 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1795 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1800 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1802 skb
->ip_summed
= CHECKSUM_NONE
;
1806 EXPORT_SYMBOL(skb_checksum_help
);
1809 * skb_gso_segment - Perform segmentation on skb.
1810 * @skb: buffer to segment
1811 * @features: features for the output path (see dev->features)
1813 * This function segments the given skb and returns a list of segments.
1815 * It may return NULL if the skb requires no segmentation. This is
1816 * only possible when GSO is used for verifying header integrity.
1818 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1820 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1821 struct packet_type
*ptype
;
1822 __be16 type
= skb
->protocol
;
1823 int vlan_depth
= ETH_HLEN
;
1826 while (type
== htons(ETH_P_8021Q
)) {
1827 struct vlan_hdr
*vh
;
1829 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1830 return ERR_PTR(-EINVAL
);
1832 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1833 type
= vh
->h_vlan_encapsulated_proto
;
1834 vlan_depth
+= VLAN_HLEN
;
1837 skb_reset_mac_header(skb
);
1838 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1839 __skb_pull(skb
, skb
->mac_len
);
1841 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1842 struct net_device
*dev
= skb
->dev
;
1843 struct ethtool_drvinfo info
= {};
1845 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1846 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1848 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1849 info
.driver
, dev
? dev
->features
: 0L,
1850 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1851 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1853 if (skb_header_cloned(skb
) &&
1854 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1855 return ERR_PTR(err
);
1859 list_for_each_entry_rcu(ptype
,
1860 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1861 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1862 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1863 err
= ptype
->gso_send_check(skb
);
1864 segs
= ERR_PTR(err
);
1865 if (err
|| skb_gso_ok(skb
, features
))
1867 __skb_push(skb
, (skb
->data
-
1868 skb_network_header(skb
)));
1870 segs
= ptype
->gso_segment(skb
, features
);
1876 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1880 EXPORT_SYMBOL(skb_gso_segment
);
1882 /* Take action when hardware reception checksum errors are detected. */
1884 void netdev_rx_csum_fault(struct net_device
*dev
)
1886 if (net_ratelimit()) {
1887 printk(KERN_ERR
"%s: hw csum failure.\n",
1888 dev
? dev
->name
: "<unknown>");
1892 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1895 /* Actually, we should eliminate this check as soon as we know, that:
1896 * 1. IOMMU is present and allows to map all the memory.
1897 * 2. No high memory really exists on this machine.
1900 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1902 #ifdef CONFIG_HIGHMEM
1904 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1905 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1906 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1910 if (PCI_DMA_BUS_IS_PHYS
) {
1911 struct device
*pdev
= dev
->dev
.parent
;
1915 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1916 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1917 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1926 void (*destructor
)(struct sk_buff
*skb
);
1929 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1931 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1933 struct dev_gso_cb
*cb
;
1936 struct sk_buff
*nskb
= skb
->next
;
1938 skb
->next
= nskb
->next
;
1941 } while (skb
->next
);
1943 cb
= DEV_GSO_CB(skb
);
1945 cb
->destructor(skb
);
1949 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1950 * @skb: buffer to segment
1951 * @features: device features as applicable to this skb
1953 * This function segments the given skb and stores the list of segments
1956 static int dev_gso_segment(struct sk_buff
*skb
, int features
)
1958 struct sk_buff
*segs
;
1960 segs
= skb_gso_segment(skb
, features
);
1962 /* Verifying header integrity only. */
1967 return PTR_ERR(segs
);
1970 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1971 skb
->destructor
= dev_gso_skb_destructor
;
1977 * Try to orphan skb early, right before transmission by the device.
1978 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1979 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1981 static inline void skb_orphan_try(struct sk_buff
*skb
)
1983 struct sock
*sk
= skb
->sk
;
1985 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
1986 /* skb_tx_hash() wont be able to get sk.
1987 * We copy sk_hash into skb->rxhash
1990 skb
->rxhash
= sk
->sk_hash
;
1995 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1997 return ((features
& NETIF_F_GEN_CSUM
) ||
1998 ((features
& NETIF_F_V4_CSUM
) &&
1999 protocol
== htons(ETH_P_IP
)) ||
2000 ((features
& NETIF_F_V6_CSUM
) &&
2001 protocol
== htons(ETH_P_IPV6
)) ||
2002 ((features
& NETIF_F_FCOE_CRC
) &&
2003 protocol
== htons(ETH_P_FCOE
)));
2006 static int harmonize_features(struct sk_buff
*skb
, __be16 protocol
, int features
)
2008 if (!can_checksum_protocol(features
, protocol
)) {
2009 features
&= ~NETIF_F_ALL_CSUM
;
2010 features
&= ~NETIF_F_SG
;
2011 } else if (illegal_highdma(skb
->dev
, skb
)) {
2012 features
&= ~NETIF_F_SG
;
2018 int netif_skb_features(struct sk_buff
*skb
)
2020 __be16 protocol
= skb
->protocol
;
2021 int features
= skb
->dev
->features
;
2023 if (protocol
== htons(ETH_P_8021Q
)) {
2024 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2025 protocol
= veh
->h_vlan_encapsulated_proto
;
2026 } else if (!vlan_tx_tag_present(skb
)) {
2027 return harmonize_features(skb
, protocol
, features
);
2030 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2032 if (protocol
!= htons(ETH_P_8021Q
)) {
2033 return harmonize_features(skb
, protocol
, features
);
2035 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2036 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2037 return harmonize_features(skb
, protocol
, features
);
2040 EXPORT_SYMBOL(netif_skb_features
);
2043 * Returns true if either:
2044 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2045 * 2. skb is fragmented and the device does not support SG, or if
2046 * at least one of fragments is in highmem and device does not
2047 * support DMA from it.
2049 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2052 return skb_is_nonlinear(skb
) &&
2053 ((skb_has_frag_list(skb
) &&
2054 !(features
& NETIF_F_FRAGLIST
)) ||
2055 (skb_shinfo(skb
)->nr_frags
&&
2056 !(features
& NETIF_F_SG
)));
2059 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2060 struct netdev_queue
*txq
)
2062 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2063 int rc
= NETDEV_TX_OK
;
2065 if (likely(!skb
->next
)) {
2069 * If device doesnt need skb->dst, release it right now while
2070 * its hot in this cpu cache
2072 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2075 if (!list_empty(&ptype_all
))
2076 dev_queue_xmit_nit(skb
, dev
);
2078 skb_orphan_try(skb
);
2080 features
= netif_skb_features(skb
);
2082 if (vlan_tx_tag_present(skb
) &&
2083 !(features
& NETIF_F_HW_VLAN_TX
)) {
2084 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2091 if (netif_needs_gso(skb
, features
)) {
2092 if (unlikely(dev_gso_segment(skb
, features
)))
2097 if (skb_needs_linearize(skb
, features
) &&
2098 __skb_linearize(skb
))
2101 /* If packet is not checksummed and device does not
2102 * support checksumming for this protocol, complete
2103 * checksumming here.
2105 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2106 skb_set_transport_header(skb
,
2107 skb_checksum_start_offset(skb
));
2108 if (!(features
& NETIF_F_ALL_CSUM
) &&
2109 skb_checksum_help(skb
))
2114 rc
= ops
->ndo_start_xmit(skb
, dev
);
2115 trace_net_dev_xmit(skb
, rc
);
2116 if (rc
== NETDEV_TX_OK
)
2117 txq_trans_update(txq
);
2123 struct sk_buff
*nskb
= skb
->next
;
2125 skb
->next
= nskb
->next
;
2129 * If device doesnt need nskb->dst, release it right now while
2130 * its hot in this cpu cache
2132 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2135 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2136 trace_net_dev_xmit(nskb
, rc
);
2137 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2138 if (rc
& ~NETDEV_TX_MASK
)
2139 goto out_kfree_gso_skb
;
2140 nskb
->next
= skb
->next
;
2144 txq_trans_update(txq
);
2145 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2146 return NETDEV_TX_BUSY
;
2147 } while (skb
->next
);
2150 if (likely(skb
->next
== NULL
))
2151 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2158 static u32 hashrnd __read_mostly
;
2161 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2162 * to be used as a distribution range.
2164 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2165 unsigned int num_tx_queues
)
2169 if (skb_rx_queue_recorded(skb
)) {
2170 hash
= skb_get_rx_queue(skb
);
2171 while (unlikely(hash
>= num_tx_queues
))
2172 hash
-= num_tx_queues
;
2176 if (skb
->sk
&& skb
->sk
->sk_hash
)
2177 hash
= skb
->sk
->sk_hash
;
2179 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2180 hash
= jhash_1word(hash
, hashrnd
);
2182 return (u16
) (((u64
) hash
* num_tx_queues
) >> 32);
2184 EXPORT_SYMBOL(__skb_tx_hash
);
2186 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2188 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2189 if (net_ratelimit()) {
2190 pr_warning("%s selects TX queue %d, but "
2191 "real number of TX queues is %d\n",
2192 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2199 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2202 struct xps_dev_maps
*dev_maps
;
2203 struct xps_map
*map
;
2204 int queue_index
= -1;
2207 dev_maps
= rcu_dereference(dev
->xps_maps
);
2209 map
= rcu_dereference(
2210 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2213 queue_index
= map
->queues
[0];
2216 if (skb
->sk
&& skb
->sk
->sk_hash
)
2217 hash
= skb
->sk
->sk_hash
;
2219 hash
= (__force u16
) skb
->protocol
^
2221 hash
= jhash_1word(hash
, hashrnd
);
2222 queue_index
= map
->queues
[
2223 ((u64
)hash
* map
->len
) >> 32];
2225 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2237 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2238 struct sk_buff
*skb
)
2241 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2243 if (dev
->real_num_tx_queues
== 1)
2245 else if (ops
->ndo_select_queue
) {
2246 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2247 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2249 struct sock
*sk
= skb
->sk
;
2250 queue_index
= sk_tx_queue_get(sk
);
2252 if (queue_index
< 0 || skb
->ooo_okay
||
2253 queue_index
>= dev
->real_num_tx_queues
) {
2254 int old_index
= queue_index
;
2256 queue_index
= get_xps_queue(dev
, skb
);
2257 if (queue_index
< 0)
2258 queue_index
= skb_tx_hash(dev
, skb
);
2260 if (queue_index
!= old_index
&& sk
) {
2261 struct dst_entry
*dst
=
2262 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2264 if (dst
&& skb_dst(skb
) == dst
)
2265 sk_tx_queue_set(sk
, queue_index
);
2270 skb_set_queue_mapping(skb
, queue_index
);
2271 return netdev_get_tx_queue(dev
, queue_index
);
2274 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2275 struct net_device
*dev
,
2276 struct netdev_queue
*txq
)
2278 spinlock_t
*root_lock
= qdisc_lock(q
);
2279 bool contended
= qdisc_is_running(q
);
2283 * Heuristic to force contended enqueues to serialize on a
2284 * separate lock before trying to get qdisc main lock.
2285 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2286 * and dequeue packets faster.
2288 if (unlikely(contended
))
2289 spin_lock(&q
->busylock
);
2291 spin_lock(root_lock
);
2292 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2295 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2296 qdisc_run_begin(q
)) {
2298 * This is a work-conserving queue; there are no old skbs
2299 * waiting to be sent out; and the qdisc is not running -
2300 * xmit the skb directly.
2302 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2305 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2306 qdisc_bstats_update(q
, skb
);
2308 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2309 if (unlikely(contended
)) {
2310 spin_unlock(&q
->busylock
);
2317 rc
= NET_XMIT_SUCCESS
;
2320 rc
= qdisc_enqueue_root(skb
, q
);
2321 if (qdisc_run_begin(q
)) {
2322 if (unlikely(contended
)) {
2323 spin_unlock(&q
->busylock
);
2329 spin_unlock(root_lock
);
2330 if (unlikely(contended
))
2331 spin_unlock(&q
->busylock
);
2335 static DEFINE_PER_CPU(int, xmit_recursion
);
2336 #define RECURSION_LIMIT 10
2339 * dev_queue_xmit - transmit a buffer
2340 * @skb: buffer to transmit
2342 * Queue a buffer for transmission to a network device. The caller must
2343 * have set the device and priority and built the buffer before calling
2344 * this function. The function can be called from an interrupt.
2346 * A negative errno code is returned on a failure. A success does not
2347 * guarantee the frame will be transmitted as it may be dropped due
2348 * to congestion or traffic shaping.
2350 * -----------------------------------------------------------------------------------
2351 * I notice this method can also return errors from the queue disciplines,
2352 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2355 * Regardless of the return value, the skb is consumed, so it is currently
2356 * difficult to retry a send to this method. (You can bump the ref count
2357 * before sending to hold a reference for retry if you are careful.)
2359 * When calling this method, interrupts MUST be enabled. This is because
2360 * the BH enable code must have IRQs enabled so that it will not deadlock.
2363 int dev_queue_xmit(struct sk_buff
*skb
)
2365 struct net_device
*dev
= skb
->dev
;
2366 struct netdev_queue
*txq
;
2370 /* Disable soft irqs for various locks below. Also
2371 * stops preemption for RCU.
2375 txq
= dev_pick_tx(dev
, skb
);
2376 q
= rcu_dereference_bh(txq
->qdisc
);
2378 #ifdef CONFIG_NET_CLS_ACT
2379 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2381 trace_net_dev_queue(skb
);
2383 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2387 /* The device has no queue. Common case for software devices:
2388 loopback, all the sorts of tunnels...
2390 Really, it is unlikely that netif_tx_lock protection is necessary
2391 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2393 However, it is possible, that they rely on protection
2396 Check this and shot the lock. It is not prone from deadlocks.
2397 Either shot noqueue qdisc, it is even simpler 8)
2399 if (dev
->flags
& IFF_UP
) {
2400 int cpu
= smp_processor_id(); /* ok because BHs are off */
2402 if (txq
->xmit_lock_owner
!= cpu
) {
2404 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2405 goto recursion_alert
;
2407 HARD_TX_LOCK(dev
, txq
, cpu
);
2409 if (!netif_tx_queue_stopped(txq
)) {
2410 __this_cpu_inc(xmit_recursion
);
2411 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2412 __this_cpu_dec(xmit_recursion
);
2413 if (dev_xmit_complete(rc
)) {
2414 HARD_TX_UNLOCK(dev
, txq
);
2418 HARD_TX_UNLOCK(dev
, txq
);
2419 if (net_ratelimit())
2420 printk(KERN_CRIT
"Virtual device %s asks to "
2421 "queue packet!\n", dev
->name
);
2423 /* Recursion is detected! It is possible,
2427 if (net_ratelimit())
2428 printk(KERN_CRIT
"Dead loop on virtual device "
2429 "%s, fix it urgently!\n", dev
->name
);
2434 rcu_read_unlock_bh();
2439 rcu_read_unlock_bh();
2442 EXPORT_SYMBOL(dev_queue_xmit
);
2445 /*=======================================================================
2447 =======================================================================*/
2449 int netdev_max_backlog __read_mostly
= 1000;
2450 int netdev_tstamp_prequeue __read_mostly
= 1;
2451 int netdev_budget __read_mostly
= 300;
2452 int weight_p __read_mostly
= 64; /* old backlog weight */
2454 /* Called with irq disabled */
2455 static inline void ____napi_schedule(struct softnet_data
*sd
,
2456 struct napi_struct
*napi
)
2458 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2459 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2463 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2464 * and src/dst port numbers. Returns a non-zero hash number on success
2467 __u32
__skb_get_rxhash(struct sk_buff
*skb
)
2469 int nhoff
, hash
= 0, poff
;
2470 struct ipv6hdr
*ip6
;
2473 u32 addr1
, addr2
, ihl
;
2479 nhoff
= skb_network_offset(skb
);
2481 switch (skb
->protocol
) {
2482 case __constant_htons(ETH_P_IP
):
2483 if (!pskb_may_pull(skb
, sizeof(*ip
) + nhoff
))
2486 ip
= (struct iphdr
*) (skb
->data
+ nhoff
);
2487 if (ip
->frag_off
& htons(IP_MF
| IP_OFFSET
))
2490 ip_proto
= ip
->protocol
;
2491 addr1
= (__force u32
) ip
->saddr
;
2492 addr2
= (__force u32
) ip
->daddr
;
2495 case __constant_htons(ETH_P_IPV6
):
2496 if (!pskb_may_pull(skb
, sizeof(*ip6
) + nhoff
))
2499 ip6
= (struct ipv6hdr
*) (skb
->data
+ nhoff
);
2500 ip_proto
= ip6
->nexthdr
;
2501 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2502 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2510 poff
= proto_ports_offset(ip_proto
);
2512 nhoff
+= ihl
* 4 + poff
;
2513 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2514 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2515 if (ports
.v16
[1] < ports
.v16
[0])
2516 swap(ports
.v16
[0], ports
.v16
[1]);
2520 /* get a consistent hash (same value on both flow directions) */
2524 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2531 EXPORT_SYMBOL(__skb_get_rxhash
);
2535 /* One global table that all flow-based protocols share. */
2536 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2537 EXPORT_SYMBOL(rps_sock_flow_table
);
2540 * get_rps_cpu is called from netif_receive_skb and returns the target
2541 * CPU from the RPS map of the receiving queue for a given skb.
2542 * rcu_read_lock must be held on entry.
2544 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2545 struct rps_dev_flow
**rflowp
)
2547 struct netdev_rx_queue
*rxqueue
;
2548 struct rps_map
*map
;
2549 struct rps_dev_flow_table
*flow_table
;
2550 struct rps_sock_flow_table
*sock_flow_table
;
2554 if (skb_rx_queue_recorded(skb
)) {
2555 u16 index
= skb_get_rx_queue(skb
);
2556 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2557 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2558 "%s received packet on queue %u, but number "
2559 "of RX queues is %u\n",
2560 dev
->name
, index
, dev
->real_num_rx_queues
);
2563 rxqueue
= dev
->_rx
+ index
;
2567 map
= rcu_dereference(rxqueue
->rps_map
);
2569 if (map
->len
== 1 &&
2570 !rcu_dereference_raw(rxqueue
->rps_flow_table
)) {
2571 tcpu
= map
->cpus
[0];
2572 if (cpu_online(tcpu
))
2576 } else if (!rcu_dereference_raw(rxqueue
->rps_flow_table
)) {
2580 skb_reset_network_header(skb
);
2581 if (!skb_get_rxhash(skb
))
2584 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2585 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2586 if (flow_table
&& sock_flow_table
) {
2588 struct rps_dev_flow
*rflow
;
2590 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2593 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2594 sock_flow_table
->mask
];
2597 * If the desired CPU (where last recvmsg was done) is
2598 * different from current CPU (one in the rx-queue flow
2599 * table entry), switch if one of the following holds:
2600 * - Current CPU is unset (equal to RPS_NO_CPU).
2601 * - Current CPU is offline.
2602 * - The current CPU's queue tail has advanced beyond the
2603 * last packet that was enqueued using this table entry.
2604 * This guarantees that all previous packets for the flow
2605 * have been dequeued, thus preserving in order delivery.
2607 if (unlikely(tcpu
!= next_cpu
) &&
2608 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2609 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2610 rflow
->last_qtail
)) >= 0)) {
2611 tcpu
= rflow
->cpu
= next_cpu
;
2612 if (tcpu
!= RPS_NO_CPU
)
2613 rflow
->last_qtail
= per_cpu(softnet_data
,
2614 tcpu
).input_queue_head
;
2616 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2624 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2626 if (cpu_online(tcpu
)) {
2636 /* Called from hardirq (IPI) context */
2637 static void rps_trigger_softirq(void *data
)
2639 struct softnet_data
*sd
= data
;
2641 ____napi_schedule(sd
, &sd
->backlog
);
2645 #endif /* CONFIG_RPS */
2648 * Check if this softnet_data structure is another cpu one
2649 * If yes, queue it to our IPI list and return 1
2652 static int rps_ipi_queued(struct softnet_data
*sd
)
2655 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2658 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2659 mysd
->rps_ipi_list
= sd
;
2661 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2664 #endif /* CONFIG_RPS */
2669 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2670 * queue (may be a remote CPU queue).
2672 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2673 unsigned int *qtail
)
2675 struct softnet_data
*sd
;
2676 unsigned long flags
;
2678 sd
= &per_cpu(softnet_data
, cpu
);
2680 local_irq_save(flags
);
2683 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2684 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2686 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2687 input_queue_tail_incr_save(sd
, qtail
);
2689 local_irq_restore(flags
);
2690 return NET_RX_SUCCESS
;
2693 /* Schedule NAPI for backlog device
2694 * We can use non atomic operation since we own the queue lock
2696 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2697 if (!rps_ipi_queued(sd
))
2698 ____napi_schedule(sd
, &sd
->backlog
);
2706 local_irq_restore(flags
);
2708 atomic_long_inc(&skb
->dev
->rx_dropped
);
2714 * netif_rx - post buffer to the network code
2715 * @skb: buffer to post
2717 * This function receives a packet from a device driver and queues it for
2718 * the upper (protocol) levels to process. It always succeeds. The buffer
2719 * may be dropped during processing for congestion control or by the
2723 * NET_RX_SUCCESS (no congestion)
2724 * NET_RX_DROP (packet was dropped)
2728 int netif_rx(struct sk_buff
*skb
)
2732 /* if netpoll wants it, pretend we never saw it */
2733 if (netpoll_rx(skb
))
2736 if (netdev_tstamp_prequeue
)
2737 net_timestamp_check(skb
);
2739 trace_netif_rx(skb
);
2742 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2748 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2750 cpu
= smp_processor_id();
2752 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2760 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2766 EXPORT_SYMBOL(netif_rx
);
2768 int netif_rx_ni(struct sk_buff
*skb
)
2773 err
= netif_rx(skb
);
2774 if (local_softirq_pending())
2780 EXPORT_SYMBOL(netif_rx_ni
);
2782 static void net_tx_action(struct softirq_action
*h
)
2784 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2786 if (sd
->completion_queue
) {
2787 struct sk_buff
*clist
;
2789 local_irq_disable();
2790 clist
= sd
->completion_queue
;
2791 sd
->completion_queue
= NULL
;
2795 struct sk_buff
*skb
= clist
;
2796 clist
= clist
->next
;
2798 WARN_ON(atomic_read(&skb
->users
));
2799 trace_kfree_skb(skb
, net_tx_action
);
2804 if (sd
->output_queue
) {
2807 local_irq_disable();
2808 head
= sd
->output_queue
;
2809 sd
->output_queue
= NULL
;
2810 sd
->output_queue_tailp
= &sd
->output_queue
;
2814 struct Qdisc
*q
= head
;
2815 spinlock_t
*root_lock
;
2817 head
= head
->next_sched
;
2819 root_lock
= qdisc_lock(q
);
2820 if (spin_trylock(root_lock
)) {
2821 smp_mb__before_clear_bit();
2822 clear_bit(__QDISC_STATE_SCHED
,
2825 spin_unlock(root_lock
);
2827 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2829 __netif_reschedule(q
);
2831 smp_mb__before_clear_bit();
2832 clear_bit(__QDISC_STATE_SCHED
,
2840 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2841 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2842 /* This hook is defined here for ATM LANE */
2843 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2844 unsigned char *addr
) __read_mostly
;
2845 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2848 #ifdef CONFIG_NET_CLS_ACT
2849 /* TODO: Maybe we should just force sch_ingress to be compiled in
2850 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2851 * a compare and 2 stores extra right now if we dont have it on
2852 * but have CONFIG_NET_CLS_ACT
2853 * NOTE: This doesnt stop any functionality; if you dont have
2854 * the ingress scheduler, you just cant add policies on ingress.
2857 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
2859 struct net_device
*dev
= skb
->dev
;
2860 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2861 int result
= TC_ACT_OK
;
2864 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
2865 if (net_ratelimit())
2866 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2867 skb
->skb_iif
, dev
->ifindex
);
2871 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2872 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2875 if (q
!= &noop_qdisc
) {
2876 spin_lock(qdisc_lock(q
));
2877 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2878 result
= qdisc_enqueue_root(skb
, q
);
2879 spin_unlock(qdisc_lock(q
));
2885 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2886 struct packet_type
**pt_prev
,
2887 int *ret
, struct net_device
*orig_dev
)
2889 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
2891 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
2895 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2899 switch (ing_filter(skb
, rxq
)) {
2913 * netdev_rx_handler_register - register receive handler
2914 * @dev: device to register a handler for
2915 * @rx_handler: receive handler to register
2916 * @rx_handler_data: data pointer that is used by rx handler
2918 * Register a receive hander for a device. This handler will then be
2919 * called from __netif_receive_skb. A negative errno code is returned
2922 * The caller must hold the rtnl_mutex.
2924 int netdev_rx_handler_register(struct net_device
*dev
,
2925 rx_handler_func_t
*rx_handler
,
2926 void *rx_handler_data
)
2930 if (dev
->rx_handler
)
2933 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
2934 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
2938 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
2941 * netdev_rx_handler_unregister - unregister receive handler
2942 * @dev: device to unregister a handler from
2944 * Unregister a receive hander from a device.
2946 * The caller must hold the rtnl_mutex.
2948 void netdev_rx_handler_unregister(struct net_device
*dev
)
2952 rcu_assign_pointer(dev
->rx_handler
, NULL
);
2953 rcu_assign_pointer(dev
->rx_handler_data
, NULL
);
2955 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
2957 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2958 struct net_device
*master
)
2960 if (skb
->pkt_type
== PACKET_HOST
) {
2961 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2963 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2967 /* On bonding slaves other than the currently active slave, suppress
2968 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2969 * ARP on active-backup slaves with arp_validate enabled.
2971 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2973 struct net_device
*dev
= skb
->dev
;
2975 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2976 dev
->last_rx
= jiffies
;
2978 if ((master
->priv_flags
& IFF_MASTER_ALB
) &&
2979 (master
->priv_flags
& IFF_BRIDGE_PORT
)) {
2980 /* Do address unmangle. The local destination address
2981 * will be always the one master has. Provides the right
2982 * functionality in a bridge.
2984 skb_bond_set_mac_by_master(skb
, master
);
2987 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
2988 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
2989 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
2992 if (master
->priv_flags
& IFF_MASTER_ALB
) {
2993 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
2994 skb
->pkt_type
!= PACKET_MULTICAST
)
2997 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
2998 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
3005 EXPORT_SYMBOL(__skb_bond_should_drop
);
3007 static int __netif_receive_skb(struct sk_buff
*skb
)
3009 struct packet_type
*ptype
, *pt_prev
;
3010 rx_handler_func_t
*rx_handler
;
3011 struct net_device
*orig_dev
;
3012 struct net_device
*master
;
3013 struct net_device
*null_or_orig
;
3014 struct net_device
*orig_or_bond
;
3015 int ret
= NET_RX_DROP
;
3018 if (!netdev_tstamp_prequeue
)
3019 net_timestamp_check(skb
);
3021 trace_netif_receive_skb(skb
);
3023 /* if we've gotten here through NAPI, check netpoll */
3024 if (netpoll_receive_skb(skb
))
3028 skb
->skb_iif
= skb
->dev
->ifindex
;
3031 * bonding note: skbs received on inactive slaves should only
3032 * be delivered to pkt handlers that are exact matches. Also
3033 * the deliver_no_wcard flag will be set. If packet handlers
3034 * are sensitive to duplicate packets these skbs will need to
3035 * be dropped at the handler.
3037 null_or_orig
= NULL
;
3038 orig_dev
= skb
->dev
;
3039 master
= ACCESS_ONCE(orig_dev
->master
);
3040 if (skb
->deliver_no_wcard
)
3041 null_or_orig
= orig_dev
;
3043 if (skb_bond_should_drop(skb
, master
)) {
3044 skb
->deliver_no_wcard
= 1;
3045 null_or_orig
= orig_dev
; /* deliver only exact match */
3050 __this_cpu_inc(softnet_data
.processed
);
3051 skb_reset_network_header(skb
);
3052 skb_reset_transport_header(skb
);
3053 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
3059 #ifdef CONFIG_NET_CLS_ACT
3060 if (skb
->tc_verd
& TC_NCLS
) {
3061 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3066 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3067 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
3068 ptype
->dev
== orig_dev
) {
3070 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3075 #ifdef CONFIG_NET_CLS_ACT
3076 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3082 /* Handle special case of bridge or macvlan */
3083 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3086 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3089 skb
= rx_handler(skb
);
3094 if (vlan_tx_tag_present(skb
)) {
3096 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3099 if (vlan_hwaccel_do_receive(&skb
)) {
3100 ret
= __netif_receive_skb(skb
);
3102 } else if (unlikely(!skb
))
3107 * Make sure frames received on VLAN interfaces stacked on
3108 * bonding interfaces still make their way to any base bonding
3109 * device that may have registered for a specific ptype. The
3110 * handler may have to adjust skb->dev and orig_dev.
3112 orig_or_bond
= orig_dev
;
3113 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
3114 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
3115 orig_or_bond
= vlan_dev_real_dev(skb
->dev
);
3118 type
= skb
->protocol
;
3119 list_for_each_entry_rcu(ptype
,
3120 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3121 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
3122 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
3123 ptype
->dev
== orig_or_bond
)) {
3125 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3131 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3133 atomic_long_inc(&skb
->dev
->rx_dropped
);
3135 /* Jamal, now you will not able to escape explaining
3136 * me how you were going to use this. :-)
3147 * netif_receive_skb - process receive buffer from network
3148 * @skb: buffer to process
3150 * netif_receive_skb() is the main receive data processing function.
3151 * It always succeeds. The buffer may be dropped during processing
3152 * for congestion control or by the protocol layers.
3154 * This function may only be called from softirq context and interrupts
3155 * should be enabled.
3157 * Return values (usually ignored):
3158 * NET_RX_SUCCESS: no congestion
3159 * NET_RX_DROP: packet was dropped
3161 int netif_receive_skb(struct sk_buff
*skb
)
3163 if (netdev_tstamp_prequeue
)
3164 net_timestamp_check(skb
);
3166 if (skb_defer_rx_timestamp(skb
))
3167 return NET_RX_SUCCESS
;
3171 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3176 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3179 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3183 ret
= __netif_receive_skb(skb
);
3189 return __netif_receive_skb(skb
);
3192 EXPORT_SYMBOL(netif_receive_skb
);
3194 /* Network device is going away, flush any packets still pending
3195 * Called with irqs disabled.
3197 static void flush_backlog(void *arg
)
3199 struct net_device
*dev
= arg
;
3200 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3201 struct sk_buff
*skb
, *tmp
;
3204 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3205 if (skb
->dev
== dev
) {
3206 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3208 input_queue_head_incr(sd
);
3213 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3214 if (skb
->dev
== dev
) {
3215 __skb_unlink(skb
, &sd
->process_queue
);
3217 input_queue_head_incr(sd
);
3222 static int napi_gro_complete(struct sk_buff
*skb
)
3224 struct packet_type
*ptype
;
3225 __be16 type
= skb
->protocol
;
3226 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3229 if (NAPI_GRO_CB(skb
)->count
== 1) {
3230 skb_shinfo(skb
)->gso_size
= 0;
3235 list_for_each_entry_rcu(ptype
, head
, list
) {
3236 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3239 err
= ptype
->gro_complete(skb
);
3245 WARN_ON(&ptype
->list
== head
);
3247 return NET_RX_SUCCESS
;
3251 return netif_receive_skb(skb
);
3254 inline void napi_gro_flush(struct napi_struct
*napi
)
3256 struct sk_buff
*skb
, *next
;
3258 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3261 napi_gro_complete(skb
);
3264 napi
->gro_count
= 0;
3265 napi
->gro_list
= NULL
;
3267 EXPORT_SYMBOL(napi_gro_flush
);
3269 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3271 struct sk_buff
**pp
= NULL
;
3272 struct packet_type
*ptype
;
3273 __be16 type
= skb
->protocol
;
3274 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3277 enum gro_result ret
;
3279 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3282 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3286 list_for_each_entry_rcu(ptype
, head
, list
) {
3287 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3290 skb_set_network_header(skb
, skb_gro_offset(skb
));
3291 mac_len
= skb
->network_header
- skb
->mac_header
;
3292 skb
->mac_len
= mac_len
;
3293 NAPI_GRO_CB(skb
)->same_flow
= 0;
3294 NAPI_GRO_CB(skb
)->flush
= 0;
3295 NAPI_GRO_CB(skb
)->free
= 0;
3297 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3302 if (&ptype
->list
== head
)
3305 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3306 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3309 struct sk_buff
*nskb
= *pp
;
3313 napi_gro_complete(nskb
);
3320 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3324 NAPI_GRO_CB(skb
)->count
= 1;
3325 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3326 skb
->next
= napi
->gro_list
;
3327 napi
->gro_list
= skb
;
3331 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3332 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3334 BUG_ON(skb
->end
- skb
->tail
< grow
);
3336 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3339 skb
->data_len
-= grow
;
3341 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3342 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3344 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3345 put_page(skb_shinfo(skb
)->frags
[0].page
);
3346 memmove(skb_shinfo(skb
)->frags
,
3347 skb_shinfo(skb
)->frags
+ 1,
3348 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3359 EXPORT_SYMBOL(dev_gro_receive
);
3361 static inline gro_result_t
3362 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3366 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3367 unsigned long diffs
;
3369 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3370 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3371 diffs
|= compare_ether_header(skb_mac_header(p
),
3372 skb_gro_mac_header(skb
));
3373 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3374 NAPI_GRO_CB(p
)->flush
= 0;
3377 return dev_gro_receive(napi
, skb
);
3380 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3384 if (netif_receive_skb(skb
))
3389 case GRO_MERGED_FREE
:
3400 EXPORT_SYMBOL(napi_skb_finish
);
3402 void skb_gro_reset_offset(struct sk_buff
*skb
)
3404 NAPI_GRO_CB(skb
)->data_offset
= 0;
3405 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3406 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3408 if (skb
->mac_header
== skb
->tail
&&
3409 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3410 NAPI_GRO_CB(skb
)->frag0
=
3411 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3412 skb_shinfo(skb
)->frags
[0].page_offset
;
3413 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3416 EXPORT_SYMBOL(skb_gro_reset_offset
);
3418 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3420 skb_gro_reset_offset(skb
);
3422 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3424 EXPORT_SYMBOL(napi_gro_receive
);
3426 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3428 __skb_pull(skb
, skb_headlen(skb
));
3429 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3431 skb
->dev
= napi
->dev
;
3437 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3439 struct sk_buff
*skb
= napi
->skb
;
3442 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3448 EXPORT_SYMBOL(napi_get_frags
);
3450 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3456 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3458 if (ret
== GRO_HELD
)
3459 skb_gro_pull(skb
, -ETH_HLEN
);
3460 else if (netif_receive_skb(skb
))
3465 case GRO_MERGED_FREE
:
3466 napi_reuse_skb(napi
, skb
);
3475 EXPORT_SYMBOL(napi_frags_finish
);
3477 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3479 struct sk_buff
*skb
= napi
->skb
;
3486 skb_reset_mac_header(skb
);
3487 skb_gro_reset_offset(skb
);
3489 off
= skb_gro_offset(skb
);
3490 hlen
= off
+ sizeof(*eth
);
3491 eth
= skb_gro_header_fast(skb
, off
);
3492 if (skb_gro_header_hard(skb
, hlen
)) {
3493 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3494 if (unlikely(!eth
)) {
3495 napi_reuse_skb(napi
, skb
);
3501 skb_gro_pull(skb
, sizeof(*eth
));
3504 * This works because the only protocols we care about don't require
3505 * special handling. We'll fix it up properly at the end.
3507 skb
->protocol
= eth
->h_proto
;
3512 EXPORT_SYMBOL(napi_frags_skb
);
3514 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3516 struct sk_buff
*skb
= napi_frags_skb(napi
);
3521 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3523 EXPORT_SYMBOL(napi_gro_frags
);
3526 * net_rps_action sends any pending IPI's for rps.
3527 * Note: called with local irq disabled, but exits with local irq enabled.
3529 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3532 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3535 sd
->rps_ipi_list
= NULL
;
3539 /* Send pending IPI's to kick RPS processing on remote cpus. */
3541 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3543 if (cpu_online(remsd
->cpu
))
3544 __smp_call_function_single(remsd
->cpu
,
3553 static int process_backlog(struct napi_struct
*napi
, int quota
)
3556 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3559 /* Check if we have pending ipi, its better to send them now,
3560 * not waiting net_rx_action() end.
3562 if (sd
->rps_ipi_list
) {
3563 local_irq_disable();
3564 net_rps_action_and_irq_enable(sd
);
3567 napi
->weight
= weight_p
;
3568 local_irq_disable();
3569 while (work
< quota
) {
3570 struct sk_buff
*skb
;
3573 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3575 __netif_receive_skb(skb
);
3576 local_irq_disable();
3577 input_queue_head_incr(sd
);
3578 if (++work
>= quota
) {
3585 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3587 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3588 &sd
->process_queue
);
3590 if (qlen
< quota
- work
) {
3592 * Inline a custom version of __napi_complete().
3593 * only current cpu owns and manipulates this napi,
3594 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3595 * we can use a plain write instead of clear_bit(),
3596 * and we dont need an smp_mb() memory barrier.
3598 list_del(&napi
->poll_list
);
3601 quota
= work
+ qlen
;
3611 * __napi_schedule - schedule for receive
3612 * @n: entry to schedule
3614 * The entry's receive function will be scheduled to run
3616 void __napi_schedule(struct napi_struct
*n
)
3618 unsigned long flags
;
3620 local_irq_save(flags
);
3621 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3622 local_irq_restore(flags
);
3624 EXPORT_SYMBOL(__napi_schedule
);
3626 void __napi_complete(struct napi_struct
*n
)
3628 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3629 BUG_ON(n
->gro_list
);
3631 list_del(&n
->poll_list
);
3632 smp_mb__before_clear_bit();
3633 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3635 EXPORT_SYMBOL(__napi_complete
);
3637 void napi_complete(struct napi_struct
*n
)
3639 unsigned long flags
;
3642 * don't let napi dequeue from the cpu poll list
3643 * just in case its running on a different cpu
3645 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3649 local_irq_save(flags
);
3651 local_irq_restore(flags
);
3653 EXPORT_SYMBOL(napi_complete
);
3655 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3656 int (*poll
)(struct napi_struct
*, int), int weight
)
3658 INIT_LIST_HEAD(&napi
->poll_list
);
3659 napi
->gro_count
= 0;
3660 napi
->gro_list
= NULL
;
3663 napi
->weight
= weight
;
3664 list_add(&napi
->dev_list
, &dev
->napi_list
);
3666 #ifdef CONFIG_NETPOLL
3667 spin_lock_init(&napi
->poll_lock
);
3668 napi
->poll_owner
= -1;
3670 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3672 EXPORT_SYMBOL(netif_napi_add
);
3674 void netif_napi_del(struct napi_struct
*napi
)
3676 struct sk_buff
*skb
, *next
;
3678 list_del_init(&napi
->dev_list
);
3679 napi_free_frags(napi
);
3681 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3687 napi
->gro_list
= NULL
;
3688 napi
->gro_count
= 0;
3690 EXPORT_SYMBOL(netif_napi_del
);
3692 static void net_rx_action(struct softirq_action
*h
)
3694 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3695 unsigned long time_limit
= jiffies
+ 2;
3696 int budget
= netdev_budget
;
3699 local_irq_disable();
3701 while (!list_empty(&sd
->poll_list
)) {
3702 struct napi_struct
*n
;
3705 /* If softirq window is exhuasted then punt.
3706 * Allow this to run for 2 jiffies since which will allow
3707 * an average latency of 1.5/HZ.
3709 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3714 /* Even though interrupts have been re-enabled, this
3715 * access is safe because interrupts can only add new
3716 * entries to the tail of this list, and only ->poll()
3717 * calls can remove this head entry from the list.
3719 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3721 have
= netpoll_poll_lock(n
);
3725 /* This NAPI_STATE_SCHED test is for avoiding a race
3726 * with netpoll's poll_napi(). Only the entity which
3727 * obtains the lock and sees NAPI_STATE_SCHED set will
3728 * actually make the ->poll() call. Therefore we avoid
3729 * accidently calling ->poll() when NAPI is not scheduled.
3732 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3733 work
= n
->poll(n
, weight
);
3737 WARN_ON_ONCE(work
> weight
);
3741 local_irq_disable();
3743 /* Drivers must not modify the NAPI state if they
3744 * consume the entire weight. In such cases this code
3745 * still "owns" the NAPI instance and therefore can
3746 * move the instance around on the list at-will.
3748 if (unlikely(work
== weight
)) {
3749 if (unlikely(napi_disable_pending(n
))) {
3752 local_irq_disable();
3754 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3757 netpoll_poll_unlock(have
);
3760 net_rps_action_and_irq_enable(sd
);
3762 #ifdef CONFIG_NET_DMA
3764 * There may not be any more sk_buffs coming right now, so push
3765 * any pending DMA copies to hardware
3767 dma_issue_pending_all();
3774 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3778 static gifconf_func_t
*gifconf_list
[NPROTO
];
3781 * register_gifconf - register a SIOCGIF handler
3782 * @family: Address family
3783 * @gifconf: Function handler
3785 * Register protocol dependent address dumping routines. The handler
3786 * that is passed must not be freed or reused until it has been replaced
3787 * by another handler.
3789 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3791 if (family
>= NPROTO
)
3793 gifconf_list
[family
] = gifconf
;
3796 EXPORT_SYMBOL(register_gifconf
);
3800 * Map an interface index to its name (SIOCGIFNAME)
3804 * We need this ioctl for efficient implementation of the
3805 * if_indextoname() function required by the IPv6 API. Without
3806 * it, we would have to search all the interfaces to find a
3810 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3812 struct net_device
*dev
;
3816 * Fetch the caller's info block.
3819 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3823 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3829 strcpy(ifr
.ifr_name
, dev
->name
);
3832 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3838 * Perform a SIOCGIFCONF call. This structure will change
3839 * size eventually, and there is nothing I can do about it.
3840 * Thus we will need a 'compatibility mode'.
3843 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3846 struct net_device
*dev
;
3853 * Fetch the caller's info block.
3856 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3863 * Loop over the interfaces, and write an info block for each.
3867 for_each_netdev(net
, dev
) {
3868 for (i
= 0; i
< NPROTO
; i
++) {
3869 if (gifconf_list
[i
]) {
3872 done
= gifconf_list
[i
](dev
, NULL
, 0);
3874 done
= gifconf_list
[i
](dev
, pos
+ total
,
3884 * All done. Write the updated control block back to the caller.
3886 ifc
.ifc_len
= total
;
3889 * Both BSD and Solaris return 0 here, so we do too.
3891 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3894 #ifdef CONFIG_PROC_FS
3896 * This is invoked by the /proc filesystem handler to display a device
3899 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3902 struct net
*net
= seq_file_net(seq
);
3904 struct net_device
*dev
;
3908 return SEQ_START_TOKEN
;
3911 for_each_netdev_rcu(net
, dev
)
3918 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3920 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3921 first_net_device(seq_file_net(seq
)) :
3922 next_net_device((struct net_device
*)v
);
3925 return rcu_dereference(dev
);
3928 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3934 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3936 struct rtnl_link_stats64 temp
;
3937 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
3939 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3940 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3941 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3943 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3944 stats
->rx_fifo_errors
,
3945 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3946 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3947 stats
->rx_compressed
, stats
->multicast
,
3948 stats
->tx_bytes
, stats
->tx_packets
,
3949 stats
->tx_errors
, stats
->tx_dropped
,
3950 stats
->tx_fifo_errors
, stats
->collisions
,
3951 stats
->tx_carrier_errors
+
3952 stats
->tx_aborted_errors
+
3953 stats
->tx_window_errors
+
3954 stats
->tx_heartbeat_errors
,
3955 stats
->tx_compressed
);
3959 * Called from the PROCfs module. This now uses the new arbitrary sized
3960 * /proc/net interface to create /proc/net/dev
3962 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3964 if (v
== SEQ_START_TOKEN
)
3965 seq_puts(seq
, "Inter-| Receive "
3967 " face |bytes packets errs drop fifo frame "
3968 "compressed multicast|bytes packets errs "
3969 "drop fifo colls carrier compressed\n");
3971 dev_seq_printf_stats(seq
, v
);
3975 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
3977 struct softnet_data
*sd
= NULL
;
3979 while (*pos
< nr_cpu_ids
)
3980 if (cpu_online(*pos
)) {
3981 sd
= &per_cpu(softnet_data
, *pos
);
3988 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3990 return softnet_get_online(pos
);
3993 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3996 return softnet_get_online(pos
);
3999 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4003 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4005 struct softnet_data
*sd
= v
;
4007 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4008 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4009 0, 0, 0, 0, /* was fastroute */
4010 sd
->cpu_collision
, sd
->received_rps
);
4014 static const struct seq_operations dev_seq_ops
= {
4015 .start
= dev_seq_start
,
4016 .next
= dev_seq_next
,
4017 .stop
= dev_seq_stop
,
4018 .show
= dev_seq_show
,
4021 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4023 return seq_open_net(inode
, file
, &dev_seq_ops
,
4024 sizeof(struct seq_net_private
));
4027 static const struct file_operations dev_seq_fops
= {
4028 .owner
= THIS_MODULE
,
4029 .open
= dev_seq_open
,
4031 .llseek
= seq_lseek
,
4032 .release
= seq_release_net
,
4035 static const struct seq_operations softnet_seq_ops
= {
4036 .start
= softnet_seq_start
,
4037 .next
= softnet_seq_next
,
4038 .stop
= softnet_seq_stop
,
4039 .show
= softnet_seq_show
,
4042 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4044 return seq_open(file
, &softnet_seq_ops
);
4047 static const struct file_operations softnet_seq_fops
= {
4048 .owner
= THIS_MODULE
,
4049 .open
= softnet_seq_open
,
4051 .llseek
= seq_lseek
,
4052 .release
= seq_release
,
4055 static void *ptype_get_idx(loff_t pos
)
4057 struct packet_type
*pt
= NULL
;
4061 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4067 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4068 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4077 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4081 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4084 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4086 struct packet_type
*pt
;
4087 struct list_head
*nxt
;
4091 if (v
== SEQ_START_TOKEN
)
4092 return ptype_get_idx(0);
4095 nxt
= pt
->list
.next
;
4096 if (pt
->type
== htons(ETH_P_ALL
)) {
4097 if (nxt
!= &ptype_all
)
4100 nxt
= ptype_base
[0].next
;
4102 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4104 while (nxt
== &ptype_base
[hash
]) {
4105 if (++hash
>= PTYPE_HASH_SIZE
)
4107 nxt
= ptype_base
[hash
].next
;
4110 return list_entry(nxt
, struct packet_type
, list
);
4113 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4119 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4121 struct packet_type
*pt
= v
;
4123 if (v
== SEQ_START_TOKEN
)
4124 seq_puts(seq
, "Type Device Function\n");
4125 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4126 if (pt
->type
== htons(ETH_P_ALL
))
4127 seq_puts(seq
, "ALL ");
4129 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4131 seq_printf(seq
, " %-8s %pF\n",
4132 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4138 static const struct seq_operations ptype_seq_ops
= {
4139 .start
= ptype_seq_start
,
4140 .next
= ptype_seq_next
,
4141 .stop
= ptype_seq_stop
,
4142 .show
= ptype_seq_show
,
4145 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4147 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4148 sizeof(struct seq_net_private
));
4151 static const struct file_operations ptype_seq_fops
= {
4152 .owner
= THIS_MODULE
,
4153 .open
= ptype_seq_open
,
4155 .llseek
= seq_lseek
,
4156 .release
= seq_release_net
,
4160 static int __net_init
dev_proc_net_init(struct net
*net
)
4164 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4166 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4168 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4171 if (wext_proc_init(net
))
4177 proc_net_remove(net
, "ptype");
4179 proc_net_remove(net
, "softnet_stat");
4181 proc_net_remove(net
, "dev");
4185 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4187 wext_proc_exit(net
);
4189 proc_net_remove(net
, "ptype");
4190 proc_net_remove(net
, "softnet_stat");
4191 proc_net_remove(net
, "dev");
4194 static struct pernet_operations __net_initdata dev_proc_ops
= {
4195 .init
= dev_proc_net_init
,
4196 .exit
= dev_proc_net_exit
,
4199 static int __init
dev_proc_init(void)
4201 return register_pernet_subsys(&dev_proc_ops
);
4204 #define dev_proc_init() 0
4205 #endif /* CONFIG_PROC_FS */
4209 * netdev_set_master - set up master/slave pair
4210 * @slave: slave device
4211 * @master: new master device
4213 * Changes the master device of the slave. Pass %NULL to break the
4214 * bonding. The caller must hold the RTNL semaphore. On a failure
4215 * a negative errno code is returned. On success the reference counts
4216 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4217 * function returns zero.
4219 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4221 struct net_device
*old
= slave
->master
;
4231 slave
->master
= master
;
4238 slave
->flags
|= IFF_SLAVE
;
4240 slave
->flags
&= ~IFF_SLAVE
;
4242 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4245 EXPORT_SYMBOL(netdev_set_master
);
4247 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4249 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4251 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4252 ops
->ndo_change_rx_flags(dev
, flags
);
4255 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4257 unsigned short old_flags
= dev
->flags
;
4263 dev
->flags
|= IFF_PROMISC
;
4264 dev
->promiscuity
+= inc
;
4265 if (dev
->promiscuity
== 0) {
4268 * If inc causes overflow, untouch promisc and return error.
4271 dev
->flags
&= ~IFF_PROMISC
;
4273 dev
->promiscuity
-= inc
;
4274 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4275 "set promiscuity failed, promiscuity feature "
4276 "of device might be broken.\n", dev
->name
);
4280 if (dev
->flags
!= old_flags
) {
4281 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4282 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4284 if (audit_enabled
) {
4285 current_uid_gid(&uid
, &gid
);
4286 audit_log(current
->audit_context
, GFP_ATOMIC
,
4287 AUDIT_ANOM_PROMISCUOUS
,
4288 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4289 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4290 (old_flags
& IFF_PROMISC
),
4291 audit_get_loginuid(current
),
4293 audit_get_sessionid(current
));
4296 dev_change_rx_flags(dev
, IFF_PROMISC
);
4302 * dev_set_promiscuity - update promiscuity count on a device
4306 * Add or remove promiscuity from a device. While the count in the device
4307 * remains above zero the interface remains promiscuous. Once it hits zero
4308 * the device reverts back to normal filtering operation. A negative inc
4309 * value is used to drop promiscuity on the device.
4310 * Return 0 if successful or a negative errno code on error.
4312 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4314 unsigned short old_flags
= dev
->flags
;
4317 err
= __dev_set_promiscuity(dev
, inc
);
4320 if (dev
->flags
!= old_flags
)
4321 dev_set_rx_mode(dev
);
4324 EXPORT_SYMBOL(dev_set_promiscuity
);
4327 * dev_set_allmulti - update allmulti count on a device
4331 * Add or remove reception of all multicast frames to a device. While the
4332 * count in the device remains above zero the interface remains listening
4333 * to all interfaces. Once it hits zero the device reverts back to normal
4334 * filtering operation. A negative @inc value is used to drop the counter
4335 * when releasing a resource needing all multicasts.
4336 * Return 0 if successful or a negative errno code on error.
4339 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4341 unsigned short old_flags
= dev
->flags
;
4345 dev
->flags
|= IFF_ALLMULTI
;
4346 dev
->allmulti
+= inc
;
4347 if (dev
->allmulti
== 0) {
4350 * If inc causes overflow, untouch allmulti and return error.
4353 dev
->flags
&= ~IFF_ALLMULTI
;
4355 dev
->allmulti
-= inc
;
4356 printk(KERN_WARNING
"%s: allmulti touches roof, "
4357 "set allmulti failed, allmulti feature of "
4358 "device might be broken.\n", dev
->name
);
4362 if (dev
->flags
^ old_flags
) {
4363 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4364 dev_set_rx_mode(dev
);
4368 EXPORT_SYMBOL(dev_set_allmulti
);
4371 * Upload unicast and multicast address lists to device and
4372 * configure RX filtering. When the device doesn't support unicast
4373 * filtering it is put in promiscuous mode while unicast addresses
4376 void __dev_set_rx_mode(struct net_device
*dev
)
4378 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4380 /* dev_open will call this function so the list will stay sane. */
4381 if (!(dev
->flags
&IFF_UP
))
4384 if (!netif_device_present(dev
))
4387 if (ops
->ndo_set_rx_mode
)
4388 ops
->ndo_set_rx_mode(dev
);
4390 /* Unicast addresses changes may only happen under the rtnl,
4391 * therefore calling __dev_set_promiscuity here is safe.
4393 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4394 __dev_set_promiscuity(dev
, 1);
4395 dev
->uc_promisc
= 1;
4396 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4397 __dev_set_promiscuity(dev
, -1);
4398 dev
->uc_promisc
= 0;
4401 if (ops
->ndo_set_multicast_list
)
4402 ops
->ndo_set_multicast_list(dev
);
4406 void dev_set_rx_mode(struct net_device
*dev
)
4408 netif_addr_lock_bh(dev
);
4409 __dev_set_rx_mode(dev
);
4410 netif_addr_unlock_bh(dev
);
4414 * dev_get_flags - get flags reported to userspace
4417 * Get the combination of flag bits exported through APIs to userspace.
4419 unsigned dev_get_flags(const struct net_device
*dev
)
4423 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4428 (dev
->gflags
& (IFF_PROMISC
|
4431 if (netif_running(dev
)) {
4432 if (netif_oper_up(dev
))
4433 flags
|= IFF_RUNNING
;
4434 if (netif_carrier_ok(dev
))
4435 flags
|= IFF_LOWER_UP
;
4436 if (netif_dormant(dev
))
4437 flags
|= IFF_DORMANT
;
4442 EXPORT_SYMBOL(dev_get_flags
);
4444 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4446 int old_flags
= dev
->flags
;
4452 * Set the flags on our device.
4455 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4456 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4458 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4462 * Load in the correct multicast list now the flags have changed.
4465 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4466 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4468 dev_set_rx_mode(dev
);
4471 * Have we downed the interface. We handle IFF_UP ourselves
4472 * according to user attempts to set it, rather than blindly
4477 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4478 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4481 dev_set_rx_mode(dev
);
4484 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4485 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4487 dev
->gflags
^= IFF_PROMISC
;
4488 dev_set_promiscuity(dev
, inc
);
4491 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4492 is important. Some (broken) drivers set IFF_PROMISC, when
4493 IFF_ALLMULTI is requested not asking us and not reporting.
4495 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4496 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4498 dev
->gflags
^= IFF_ALLMULTI
;
4499 dev_set_allmulti(dev
, inc
);
4505 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4507 unsigned int changes
= dev
->flags
^ old_flags
;
4509 if (changes
& IFF_UP
) {
4510 if (dev
->flags
& IFF_UP
)
4511 call_netdevice_notifiers(NETDEV_UP
, dev
);
4513 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4516 if (dev
->flags
& IFF_UP
&&
4517 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4518 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4522 * dev_change_flags - change device settings
4524 * @flags: device state flags
4526 * Change settings on device based state flags. The flags are
4527 * in the userspace exported format.
4529 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4532 int old_flags
= dev
->flags
;
4534 ret
= __dev_change_flags(dev
, flags
);
4538 changes
= old_flags
^ dev
->flags
;
4540 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4542 __dev_notify_flags(dev
, old_flags
);
4545 EXPORT_SYMBOL(dev_change_flags
);
4548 * dev_set_mtu - Change maximum transfer unit
4550 * @new_mtu: new transfer unit
4552 * Change the maximum transfer size of the network device.
4554 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4556 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4559 if (new_mtu
== dev
->mtu
)
4562 /* MTU must be positive. */
4566 if (!netif_device_present(dev
))
4570 if (ops
->ndo_change_mtu
)
4571 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4575 if (!err
&& dev
->flags
& IFF_UP
)
4576 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4579 EXPORT_SYMBOL(dev_set_mtu
);
4582 * dev_set_mac_address - Change Media Access Control Address
4586 * Change the hardware (MAC) address of the device
4588 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4590 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4593 if (!ops
->ndo_set_mac_address
)
4595 if (sa
->sa_family
!= dev
->type
)
4597 if (!netif_device_present(dev
))
4599 err
= ops
->ndo_set_mac_address(dev
, sa
);
4601 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4604 EXPORT_SYMBOL(dev_set_mac_address
);
4607 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4609 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4612 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4618 case SIOCGIFFLAGS
: /* Get interface flags */
4619 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4622 case SIOCGIFMETRIC
: /* Get the metric on the interface
4623 (currently unused) */
4624 ifr
->ifr_metric
= 0;
4627 case SIOCGIFMTU
: /* Get the MTU of a device */
4628 ifr
->ifr_mtu
= dev
->mtu
;
4633 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4635 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4636 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4637 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4645 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4646 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4647 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4648 ifr
->ifr_map
.irq
= dev
->irq
;
4649 ifr
->ifr_map
.dma
= dev
->dma
;
4650 ifr
->ifr_map
.port
= dev
->if_port
;
4654 ifr
->ifr_ifindex
= dev
->ifindex
;
4658 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4662 /* dev_ioctl() should ensure this case
4674 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4676 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4679 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4680 const struct net_device_ops
*ops
;
4685 ops
= dev
->netdev_ops
;
4688 case SIOCSIFFLAGS
: /* Set interface flags */
4689 return dev_change_flags(dev
, ifr
->ifr_flags
);
4691 case SIOCSIFMETRIC
: /* Set the metric on the interface
4692 (currently unused) */
4695 case SIOCSIFMTU
: /* Set the MTU of a device */
4696 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4699 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4701 case SIOCSIFHWBROADCAST
:
4702 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4704 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4705 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4706 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4710 if (ops
->ndo_set_config
) {
4711 if (!netif_device_present(dev
))
4713 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4718 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4719 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4721 if (!netif_device_present(dev
))
4723 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4726 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4727 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4729 if (!netif_device_present(dev
))
4731 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4734 if (ifr
->ifr_qlen
< 0)
4736 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4740 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4741 return dev_change_name(dev
, ifr
->ifr_newname
);
4744 * Unknown or private ioctl
4747 if ((cmd
>= SIOCDEVPRIVATE
&&
4748 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4749 cmd
== SIOCBONDENSLAVE
||
4750 cmd
== SIOCBONDRELEASE
||
4751 cmd
== SIOCBONDSETHWADDR
||
4752 cmd
== SIOCBONDSLAVEINFOQUERY
||
4753 cmd
== SIOCBONDINFOQUERY
||
4754 cmd
== SIOCBONDCHANGEACTIVE
||
4755 cmd
== SIOCGMIIPHY
||
4756 cmd
== SIOCGMIIREG
||
4757 cmd
== SIOCSMIIREG
||
4758 cmd
== SIOCBRADDIF
||
4759 cmd
== SIOCBRDELIF
||
4760 cmd
== SIOCSHWTSTAMP
||
4761 cmd
== SIOCWANDEV
) {
4763 if (ops
->ndo_do_ioctl
) {
4764 if (netif_device_present(dev
))
4765 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4777 * This function handles all "interface"-type I/O control requests. The actual
4778 * 'doing' part of this is dev_ifsioc above.
4782 * dev_ioctl - network device ioctl
4783 * @net: the applicable net namespace
4784 * @cmd: command to issue
4785 * @arg: pointer to a struct ifreq in user space
4787 * Issue ioctl functions to devices. This is normally called by the
4788 * user space syscall interfaces but can sometimes be useful for
4789 * other purposes. The return value is the return from the syscall if
4790 * positive or a negative errno code on error.
4793 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4799 /* One special case: SIOCGIFCONF takes ifconf argument
4800 and requires shared lock, because it sleeps writing
4804 if (cmd
== SIOCGIFCONF
) {
4806 ret
= dev_ifconf(net
, (char __user
*) arg
);
4810 if (cmd
== SIOCGIFNAME
)
4811 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4813 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4816 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4818 colon
= strchr(ifr
.ifr_name
, ':');
4823 * See which interface the caller is talking about.
4828 * These ioctl calls:
4829 * - can be done by all.
4830 * - atomic and do not require locking.
4841 dev_load(net
, ifr
.ifr_name
);
4843 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4848 if (copy_to_user(arg
, &ifr
,
4849 sizeof(struct ifreq
)))
4855 dev_load(net
, ifr
.ifr_name
);
4857 ret
= dev_ethtool(net
, &ifr
);
4862 if (copy_to_user(arg
, &ifr
,
4863 sizeof(struct ifreq
)))
4869 * These ioctl calls:
4870 * - require superuser power.
4871 * - require strict serialization.
4877 if (!capable(CAP_NET_ADMIN
))
4879 dev_load(net
, ifr
.ifr_name
);
4881 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4886 if (copy_to_user(arg
, &ifr
,
4887 sizeof(struct ifreq
)))
4893 * These ioctl calls:
4894 * - require superuser power.
4895 * - require strict serialization.
4896 * - do not return a value
4906 case SIOCSIFHWBROADCAST
:
4909 case SIOCBONDENSLAVE
:
4910 case SIOCBONDRELEASE
:
4911 case SIOCBONDSETHWADDR
:
4912 case SIOCBONDCHANGEACTIVE
:
4916 if (!capable(CAP_NET_ADMIN
))
4919 case SIOCBONDSLAVEINFOQUERY
:
4920 case SIOCBONDINFOQUERY
:
4921 dev_load(net
, ifr
.ifr_name
);
4923 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4928 /* Get the per device memory space. We can add this but
4929 * currently do not support it */
4931 /* Set the per device memory buffer space.
4932 * Not applicable in our case */
4937 * Unknown or private ioctl.
4940 if (cmd
== SIOCWANDEV
||
4941 (cmd
>= SIOCDEVPRIVATE
&&
4942 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4943 dev_load(net
, ifr
.ifr_name
);
4945 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4947 if (!ret
&& copy_to_user(arg
, &ifr
,
4948 sizeof(struct ifreq
)))
4952 /* Take care of Wireless Extensions */
4953 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4954 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4961 * dev_new_index - allocate an ifindex
4962 * @net: the applicable net namespace
4964 * Returns a suitable unique value for a new device interface
4965 * number. The caller must hold the rtnl semaphore or the
4966 * dev_base_lock to be sure it remains unique.
4968 static int dev_new_index(struct net
*net
)
4974 if (!__dev_get_by_index(net
, ifindex
))
4979 /* Delayed registration/unregisteration */
4980 static LIST_HEAD(net_todo_list
);
4982 static void net_set_todo(struct net_device
*dev
)
4984 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4987 static void rollback_registered_many(struct list_head
*head
)
4989 struct net_device
*dev
, *tmp
;
4991 BUG_ON(dev_boot_phase
);
4994 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4995 /* Some devices call without registering
4996 * for initialization unwind. Remove those
4997 * devices and proceed with the remaining.
4999 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5000 pr_debug("unregister_netdevice: device %s/%p never "
5001 "was registered\n", dev
->name
, dev
);
5004 list_del(&dev
->unreg_list
);
5008 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5011 /* If device is running, close it first. */
5012 dev_close_many(head
);
5014 list_for_each_entry(dev
, head
, unreg_list
) {
5015 /* And unlink it from device chain. */
5016 unlist_netdevice(dev
);
5018 dev
->reg_state
= NETREG_UNREGISTERING
;
5023 list_for_each_entry(dev
, head
, unreg_list
) {
5024 /* Shutdown queueing discipline. */
5028 /* Notify protocols, that we are about to destroy
5029 this device. They should clean all the things.
5031 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5033 if (!dev
->rtnl_link_ops
||
5034 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5035 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5038 * Flush the unicast and multicast chains
5043 if (dev
->netdev_ops
->ndo_uninit
)
5044 dev
->netdev_ops
->ndo_uninit(dev
);
5046 /* Notifier chain MUST detach us from master device. */
5047 WARN_ON(dev
->master
);
5049 /* Remove entries from kobject tree */
5050 netdev_unregister_kobject(dev
);
5053 /* Process any work delayed until the end of the batch */
5054 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5055 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5059 list_for_each_entry(dev
, head
, unreg_list
)
5063 static void rollback_registered(struct net_device
*dev
)
5067 list_add(&dev
->unreg_list
, &single
);
5068 rollback_registered_many(&single
);
5071 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
5073 /* Fix illegal SG+CSUM combinations. */
5074 if ((features
& NETIF_F_SG
) &&
5075 !(features
& NETIF_F_ALL_CSUM
)) {
5077 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
5078 "checksum feature.\n", name
);
5079 features
&= ~NETIF_F_SG
;
5082 /* TSO requires that SG is present as well. */
5083 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
5085 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
5086 "SG feature.\n", name
);
5087 features
&= ~NETIF_F_TSO
;
5090 if (features
& NETIF_F_UFO
) {
5091 /* maybe split UFO into V4 and V6? */
5092 if (!((features
& NETIF_F_GEN_CSUM
) ||
5093 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5094 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5096 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
5097 "since no checksum offload features.\n",
5099 features
&= ~NETIF_F_UFO
;
5102 if (!(features
& NETIF_F_SG
)) {
5104 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
5105 "since no NETIF_F_SG feature.\n", name
);
5106 features
&= ~NETIF_F_UFO
;
5112 EXPORT_SYMBOL(netdev_fix_features
);
5115 * netif_stacked_transfer_operstate - transfer operstate
5116 * @rootdev: the root or lower level device to transfer state from
5117 * @dev: the device to transfer operstate to
5119 * Transfer operational state from root to device. This is normally
5120 * called when a stacking relationship exists between the root
5121 * device and the device(a leaf device).
5123 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5124 struct net_device
*dev
)
5126 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5127 netif_dormant_on(dev
);
5129 netif_dormant_off(dev
);
5131 if (netif_carrier_ok(rootdev
)) {
5132 if (!netif_carrier_ok(dev
))
5133 netif_carrier_on(dev
);
5135 if (netif_carrier_ok(dev
))
5136 netif_carrier_off(dev
);
5139 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5142 static int netif_alloc_rx_queues(struct net_device
*dev
)
5144 unsigned int i
, count
= dev
->num_rx_queues
;
5145 struct netdev_rx_queue
*rx
;
5149 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5151 pr_err("netdev: Unable to allocate %u rx queues.\n", count
);
5156 for (i
= 0; i
< count
; i
++)
5162 static void netdev_init_one_queue(struct net_device
*dev
,
5163 struct netdev_queue
*queue
, void *_unused
)
5165 /* Initialize queue lock */
5166 spin_lock_init(&queue
->_xmit_lock
);
5167 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5168 queue
->xmit_lock_owner
= -1;
5169 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5173 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5175 unsigned int count
= dev
->num_tx_queues
;
5176 struct netdev_queue
*tx
;
5180 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5182 pr_err("netdev: Unable to allocate %u tx queues.\n",
5188 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5189 spin_lock_init(&dev
->tx_global_lock
);
5195 * register_netdevice - register a network device
5196 * @dev: device to register
5198 * Take a completed network device structure and add it to the kernel
5199 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5200 * chain. 0 is returned on success. A negative errno code is returned
5201 * on a failure to set up the device, or if the name is a duplicate.
5203 * Callers must hold the rtnl semaphore. You may want
5204 * register_netdev() instead of this.
5207 * The locking appears insufficient to guarantee two parallel registers
5208 * will not get the same name.
5211 int register_netdevice(struct net_device
*dev
)
5214 struct net
*net
= dev_net(dev
);
5216 BUG_ON(dev_boot_phase
);
5221 /* When net_device's are persistent, this will be fatal. */
5222 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5225 spin_lock_init(&dev
->addr_list_lock
);
5226 netdev_set_addr_lockdep_class(dev
);
5230 /* Init, if this function is available */
5231 if (dev
->netdev_ops
->ndo_init
) {
5232 ret
= dev
->netdev_ops
->ndo_init(dev
);
5240 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
5244 dev
->ifindex
= dev_new_index(net
);
5245 if (dev
->iflink
== -1)
5246 dev
->iflink
= dev
->ifindex
;
5248 /* Fix illegal checksum combinations */
5249 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
5250 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5251 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5253 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5256 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5257 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5258 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5260 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5263 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5265 /* Enable software GSO if SG is supported. */
5266 if (dev
->features
& NETIF_F_SG
)
5267 dev
->features
|= NETIF_F_GSO
;
5269 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5270 * vlan_dev_init() will do the dev->features check, so these features
5271 * are enabled only if supported by underlying device.
5273 dev
->vlan_features
|= (NETIF_F_GRO
| NETIF_F_HIGHDMA
);
5275 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5276 ret
= notifier_to_errno(ret
);
5280 ret
= netdev_register_kobject(dev
);
5283 dev
->reg_state
= NETREG_REGISTERED
;
5286 * Default initial state at registry is that the
5287 * device is present.
5290 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5292 dev_init_scheduler(dev
);
5294 list_netdevice(dev
);
5296 /* Notify protocols, that a new device appeared. */
5297 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5298 ret
= notifier_to_errno(ret
);
5300 rollback_registered(dev
);
5301 dev
->reg_state
= NETREG_UNREGISTERED
;
5304 * Prevent userspace races by waiting until the network
5305 * device is fully setup before sending notifications.
5307 if (!dev
->rtnl_link_ops
||
5308 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5309 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5315 if (dev
->netdev_ops
->ndo_uninit
)
5316 dev
->netdev_ops
->ndo_uninit(dev
);
5319 EXPORT_SYMBOL(register_netdevice
);
5322 * init_dummy_netdev - init a dummy network device for NAPI
5323 * @dev: device to init
5325 * This takes a network device structure and initialize the minimum
5326 * amount of fields so it can be used to schedule NAPI polls without
5327 * registering a full blown interface. This is to be used by drivers
5328 * that need to tie several hardware interfaces to a single NAPI
5329 * poll scheduler due to HW limitations.
5331 int init_dummy_netdev(struct net_device
*dev
)
5333 /* Clear everything. Note we don't initialize spinlocks
5334 * are they aren't supposed to be taken by any of the
5335 * NAPI code and this dummy netdev is supposed to be
5336 * only ever used for NAPI polls
5338 memset(dev
, 0, sizeof(struct net_device
));
5340 /* make sure we BUG if trying to hit standard
5341 * register/unregister code path
5343 dev
->reg_state
= NETREG_DUMMY
;
5345 /* NAPI wants this */
5346 INIT_LIST_HEAD(&dev
->napi_list
);
5348 /* a dummy interface is started by default */
5349 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5350 set_bit(__LINK_STATE_START
, &dev
->state
);
5352 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5353 * because users of this 'device' dont need to change
5359 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5363 * register_netdev - register a network device
5364 * @dev: device to register
5366 * Take a completed network device structure and add it to the kernel
5367 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5368 * chain. 0 is returned on success. A negative errno code is returned
5369 * on a failure to set up the device, or if the name is a duplicate.
5371 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5372 * and expands the device name if you passed a format string to
5375 int register_netdev(struct net_device
*dev
)
5382 * If the name is a format string the caller wants us to do a
5385 if (strchr(dev
->name
, '%')) {
5386 err
= dev_alloc_name(dev
, dev
->name
);
5391 err
= register_netdevice(dev
);
5396 EXPORT_SYMBOL(register_netdev
);
5398 int netdev_refcnt_read(const struct net_device
*dev
)
5402 for_each_possible_cpu(i
)
5403 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5406 EXPORT_SYMBOL(netdev_refcnt_read
);
5409 * netdev_wait_allrefs - wait until all references are gone.
5411 * This is called when unregistering network devices.
5413 * Any protocol or device that holds a reference should register
5414 * for netdevice notification, and cleanup and put back the
5415 * reference if they receive an UNREGISTER event.
5416 * We can get stuck here if buggy protocols don't correctly
5419 static void netdev_wait_allrefs(struct net_device
*dev
)
5421 unsigned long rebroadcast_time
, warning_time
;
5424 linkwatch_forget_dev(dev
);
5426 rebroadcast_time
= warning_time
= jiffies
;
5427 refcnt
= netdev_refcnt_read(dev
);
5429 while (refcnt
!= 0) {
5430 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5433 /* Rebroadcast unregister notification */
5434 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5435 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5436 * should have already handle it the first time */
5438 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5440 /* We must not have linkwatch events
5441 * pending on unregister. If this
5442 * happens, we simply run the queue
5443 * unscheduled, resulting in a noop
5446 linkwatch_run_queue();
5451 rebroadcast_time
= jiffies
;
5456 refcnt
= netdev_refcnt_read(dev
);
5458 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5459 printk(KERN_EMERG
"unregister_netdevice: "
5460 "waiting for %s to become free. Usage "
5463 warning_time
= jiffies
;
5472 * register_netdevice(x1);
5473 * register_netdevice(x2);
5475 * unregister_netdevice(y1);
5476 * unregister_netdevice(y2);
5482 * We are invoked by rtnl_unlock().
5483 * This allows us to deal with problems:
5484 * 1) We can delete sysfs objects which invoke hotplug
5485 * without deadlocking with linkwatch via keventd.
5486 * 2) Since we run with the RTNL semaphore not held, we can sleep
5487 * safely in order to wait for the netdev refcnt to drop to zero.
5489 * We must not return until all unregister events added during
5490 * the interval the lock was held have been completed.
5492 void netdev_run_todo(void)
5494 struct list_head list
;
5496 /* Snapshot list, allow later requests */
5497 list_replace_init(&net_todo_list
, &list
);
5501 while (!list_empty(&list
)) {
5502 struct net_device
*dev
5503 = list_first_entry(&list
, struct net_device
, todo_list
);
5504 list_del(&dev
->todo_list
);
5506 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5507 printk(KERN_ERR
"network todo '%s' but state %d\n",
5508 dev
->name
, dev
->reg_state
);
5513 dev
->reg_state
= NETREG_UNREGISTERED
;
5515 on_each_cpu(flush_backlog
, dev
, 1);
5517 netdev_wait_allrefs(dev
);
5520 BUG_ON(netdev_refcnt_read(dev
));
5521 WARN_ON(rcu_dereference_raw(dev
->ip_ptr
));
5522 WARN_ON(rcu_dereference_raw(dev
->ip6_ptr
));
5523 WARN_ON(dev
->dn_ptr
);
5525 if (dev
->destructor
)
5526 dev
->destructor(dev
);
5528 /* Free network device */
5529 kobject_put(&dev
->dev
.kobj
);
5533 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5534 * fields in the same order, with only the type differing.
5536 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5537 const struct net_device_stats
*netdev_stats
)
5539 #if BITS_PER_LONG == 64
5540 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5541 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5543 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5544 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5545 u64
*dst
= (u64
*)stats64
;
5547 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5548 sizeof(*stats64
) / sizeof(u64
));
5549 for (i
= 0; i
< n
; i
++)
5555 * dev_get_stats - get network device statistics
5556 * @dev: device to get statistics from
5557 * @storage: place to store stats
5559 * Get network statistics from device. Return @storage.
5560 * The device driver may provide its own method by setting
5561 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5562 * otherwise the internal statistics structure is used.
5564 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5565 struct rtnl_link_stats64
*storage
)
5567 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5569 if (ops
->ndo_get_stats64
) {
5570 memset(storage
, 0, sizeof(*storage
));
5571 ops
->ndo_get_stats64(dev
, storage
);
5572 } else if (ops
->ndo_get_stats
) {
5573 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5575 netdev_stats_to_stats64(storage
, &dev
->stats
);
5577 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5580 EXPORT_SYMBOL(dev_get_stats
);
5582 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5584 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5586 #ifdef CONFIG_NET_CLS_ACT
5589 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5592 netdev_init_one_queue(dev
, queue
, NULL
);
5593 queue
->qdisc
= &noop_qdisc
;
5594 queue
->qdisc_sleeping
= &noop_qdisc
;
5595 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5601 * alloc_netdev_mqs - allocate network device
5602 * @sizeof_priv: size of private data to allocate space for
5603 * @name: device name format string
5604 * @setup: callback to initialize device
5605 * @txqs: the number of TX subqueues to allocate
5606 * @rxqs: the number of RX subqueues to allocate
5608 * Allocates a struct net_device with private data area for driver use
5609 * and performs basic initialization. Also allocates subquue structs
5610 * for each queue on the device.
5612 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5613 void (*setup
)(struct net_device
*),
5614 unsigned int txqs
, unsigned int rxqs
)
5616 struct net_device
*dev
;
5618 struct net_device
*p
;
5620 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5623 pr_err("alloc_netdev: Unable to allocate device "
5624 "with zero queues.\n");
5630 pr_err("alloc_netdev: Unable to allocate device "
5631 "with zero RX queues.\n");
5636 alloc_size
= sizeof(struct net_device
);
5638 /* ensure 32-byte alignment of private area */
5639 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5640 alloc_size
+= sizeof_priv
;
5642 /* ensure 32-byte alignment of whole construct */
5643 alloc_size
+= NETDEV_ALIGN
- 1;
5645 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5647 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5651 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5652 dev
->padded
= (char *)dev
- (char *)p
;
5654 dev
->pcpu_refcnt
= alloc_percpu(int);
5655 if (!dev
->pcpu_refcnt
)
5658 if (dev_addr_init(dev
))
5664 dev_net_set(dev
, &init_net
);
5666 dev
->gso_max_size
= GSO_MAX_SIZE
;
5668 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5669 dev
->ethtool_ntuple_list
.count
= 0;
5670 INIT_LIST_HEAD(&dev
->napi_list
);
5671 INIT_LIST_HEAD(&dev
->unreg_list
);
5672 INIT_LIST_HEAD(&dev
->link_watch_list
);
5673 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5676 dev
->num_tx_queues
= txqs
;
5677 dev
->real_num_tx_queues
= txqs
;
5678 if (netif_alloc_netdev_queues(dev
))
5682 dev
->num_rx_queues
= rxqs
;
5683 dev
->real_num_rx_queues
= rxqs
;
5684 if (netif_alloc_rx_queues(dev
))
5688 strcpy(dev
->name
, name
);
5696 free_percpu(dev
->pcpu_refcnt
);
5706 EXPORT_SYMBOL(alloc_netdev_mqs
);
5709 * free_netdev - free network device
5712 * This function does the last stage of destroying an allocated device
5713 * interface. The reference to the device object is released.
5714 * If this is the last reference then it will be freed.
5716 void free_netdev(struct net_device
*dev
)
5718 struct napi_struct
*p
, *n
;
5720 release_net(dev_net(dev
));
5727 kfree(rcu_dereference_raw(dev
->ingress_queue
));
5729 /* Flush device addresses */
5730 dev_addr_flush(dev
);
5732 /* Clear ethtool n-tuple list */
5733 ethtool_ntuple_flush(dev
);
5735 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5738 free_percpu(dev
->pcpu_refcnt
);
5739 dev
->pcpu_refcnt
= NULL
;
5741 /* Compatibility with error handling in drivers */
5742 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5743 kfree((char *)dev
- dev
->padded
);
5747 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5748 dev
->reg_state
= NETREG_RELEASED
;
5750 /* will free via device release */
5751 put_device(&dev
->dev
);
5753 EXPORT_SYMBOL(free_netdev
);
5756 * synchronize_net - Synchronize with packet receive processing
5758 * Wait for packets currently being received to be done.
5759 * Does not block later packets from starting.
5761 void synchronize_net(void)
5766 EXPORT_SYMBOL(synchronize_net
);
5769 * unregister_netdevice_queue - remove device from the kernel
5773 * This function shuts down a device interface and removes it
5774 * from the kernel tables.
5775 * If head not NULL, device is queued to be unregistered later.
5777 * Callers must hold the rtnl semaphore. You may want
5778 * unregister_netdev() instead of this.
5781 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5786 list_move_tail(&dev
->unreg_list
, head
);
5788 rollback_registered(dev
);
5789 /* Finish processing unregister after unlock */
5793 EXPORT_SYMBOL(unregister_netdevice_queue
);
5796 * unregister_netdevice_many - unregister many devices
5797 * @head: list of devices
5799 void unregister_netdevice_many(struct list_head
*head
)
5801 struct net_device
*dev
;
5803 if (!list_empty(head
)) {
5804 rollback_registered_many(head
);
5805 list_for_each_entry(dev
, head
, unreg_list
)
5809 EXPORT_SYMBOL(unregister_netdevice_many
);
5812 * unregister_netdev - remove device from the kernel
5815 * This function shuts down a device interface and removes it
5816 * from the kernel tables.
5818 * This is just a wrapper for unregister_netdevice that takes
5819 * the rtnl semaphore. In general you want to use this and not
5820 * unregister_netdevice.
5822 void unregister_netdev(struct net_device
*dev
)
5825 unregister_netdevice(dev
);
5828 EXPORT_SYMBOL(unregister_netdev
);
5831 * dev_change_net_namespace - move device to different nethost namespace
5833 * @net: network namespace
5834 * @pat: If not NULL name pattern to try if the current device name
5835 * is already taken in the destination network namespace.
5837 * This function shuts down a device interface and moves it
5838 * to a new network namespace. On success 0 is returned, on
5839 * a failure a netagive errno code is returned.
5841 * Callers must hold the rtnl semaphore.
5844 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5850 /* Don't allow namespace local devices to be moved. */
5852 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5855 /* Ensure the device has been registrered */
5857 if (dev
->reg_state
!= NETREG_REGISTERED
)
5860 /* Get out if there is nothing todo */
5862 if (net_eq(dev_net(dev
), net
))
5865 /* Pick the destination device name, and ensure
5866 * we can use it in the destination network namespace.
5869 if (__dev_get_by_name(net
, dev
->name
)) {
5870 /* We get here if we can't use the current device name */
5873 if (dev_get_valid_name(dev
, pat
, 1))
5878 * And now a mini version of register_netdevice unregister_netdevice.
5881 /* If device is running close it first. */
5884 /* And unlink it from device chain */
5886 unlist_netdevice(dev
);
5890 /* Shutdown queueing discipline. */
5893 /* Notify protocols, that we are about to destroy
5894 this device. They should clean all the things.
5896 Note that dev->reg_state stays at NETREG_REGISTERED.
5897 This is wanted because this way 8021q and macvlan know
5898 the device is just moving and can keep their slaves up.
5900 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5901 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5904 * Flush the unicast and multicast chains
5909 /* Actually switch the network namespace */
5910 dev_net_set(dev
, net
);
5912 /* If there is an ifindex conflict assign a new one */
5913 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5914 int iflink
= (dev
->iflink
== dev
->ifindex
);
5915 dev
->ifindex
= dev_new_index(net
);
5917 dev
->iflink
= dev
->ifindex
;
5920 /* Fixup kobjects */
5921 err
= device_rename(&dev
->dev
, dev
->name
);
5924 /* Add the device back in the hashes */
5925 list_netdevice(dev
);
5927 /* Notify protocols, that a new device appeared. */
5928 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5931 * Prevent userspace races by waiting until the network
5932 * device is fully setup before sending notifications.
5934 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5941 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5943 static int dev_cpu_callback(struct notifier_block
*nfb
,
5944 unsigned long action
,
5947 struct sk_buff
**list_skb
;
5948 struct sk_buff
*skb
;
5949 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5950 struct softnet_data
*sd
, *oldsd
;
5952 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5955 local_irq_disable();
5956 cpu
= smp_processor_id();
5957 sd
= &per_cpu(softnet_data
, cpu
);
5958 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5960 /* Find end of our completion_queue. */
5961 list_skb
= &sd
->completion_queue
;
5963 list_skb
= &(*list_skb
)->next
;
5964 /* Append completion queue from offline CPU. */
5965 *list_skb
= oldsd
->completion_queue
;
5966 oldsd
->completion_queue
= NULL
;
5968 /* Append output queue from offline CPU. */
5969 if (oldsd
->output_queue
) {
5970 *sd
->output_queue_tailp
= oldsd
->output_queue
;
5971 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
5972 oldsd
->output_queue
= NULL
;
5973 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
5976 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5979 /* Process offline CPU's input_pkt_queue */
5980 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
5982 input_queue_head_incr(oldsd
);
5984 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
5986 input_queue_head_incr(oldsd
);
5994 * netdev_increment_features - increment feature set by one
5995 * @all: current feature set
5996 * @one: new feature set
5997 * @mask: mask feature set
5999 * Computes a new feature set after adding a device with feature set
6000 * @one to the master device with current feature set @all. Will not
6001 * enable anything that is off in @mask. Returns the new feature set.
6003 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
6006 /* If device needs checksumming, downgrade to it. */
6007 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
6008 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
6009 else if (mask
& NETIF_F_ALL_CSUM
) {
6010 /* If one device supports v4/v6 checksumming, set for all. */
6011 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
6012 !(all
& NETIF_F_GEN_CSUM
)) {
6013 all
&= ~NETIF_F_ALL_CSUM
;
6014 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
6017 /* If one device supports hw checksumming, set for all. */
6018 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
6019 all
&= ~NETIF_F_ALL_CSUM
;
6020 all
|= NETIF_F_HW_CSUM
;
6024 one
|= NETIF_F_ALL_CSUM
;
6026 one
|= all
& NETIF_F_ONE_FOR_ALL
;
6027 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
6028 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
6032 EXPORT_SYMBOL(netdev_increment_features
);
6034 static struct hlist_head
*netdev_create_hash(void)
6037 struct hlist_head
*hash
;
6039 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6041 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6042 INIT_HLIST_HEAD(&hash
[i
]);
6047 /* Initialize per network namespace state */
6048 static int __net_init
netdev_init(struct net
*net
)
6050 INIT_LIST_HEAD(&net
->dev_base_head
);
6052 net
->dev_name_head
= netdev_create_hash();
6053 if (net
->dev_name_head
== NULL
)
6056 net
->dev_index_head
= netdev_create_hash();
6057 if (net
->dev_index_head
== NULL
)
6063 kfree(net
->dev_name_head
);
6069 * netdev_drivername - network driver for the device
6070 * @dev: network device
6071 * @buffer: buffer for resulting name
6072 * @len: size of buffer
6074 * Determine network driver for device.
6076 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
6078 const struct device_driver
*driver
;
6079 const struct device
*parent
;
6081 if (len
<= 0 || !buffer
)
6085 parent
= dev
->dev
.parent
;
6090 driver
= parent
->driver
;
6091 if (driver
&& driver
->name
)
6092 strlcpy(buffer
, driver
->name
, len
);
6096 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6097 struct va_format
*vaf
)
6101 if (dev
&& dev
->dev
.parent
)
6102 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6103 netdev_name(dev
), vaf
);
6105 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6107 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6112 int netdev_printk(const char *level
, const struct net_device
*dev
,
6113 const char *format
, ...)
6115 struct va_format vaf
;
6119 va_start(args
, format
);
6124 r
= __netdev_printk(level
, dev
, &vaf
);
6129 EXPORT_SYMBOL(netdev_printk
);
6131 #define define_netdev_printk_level(func, level) \
6132 int func(const struct net_device *dev, const char *fmt, ...) \
6135 struct va_format vaf; \
6138 va_start(args, fmt); \
6143 r = __netdev_printk(level, dev, &vaf); \
6148 EXPORT_SYMBOL(func);
6150 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6151 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6152 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6153 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6154 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6155 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6156 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6158 static void __net_exit
netdev_exit(struct net
*net
)
6160 kfree(net
->dev_name_head
);
6161 kfree(net
->dev_index_head
);
6164 static struct pernet_operations __net_initdata netdev_net_ops
= {
6165 .init
= netdev_init
,
6166 .exit
= netdev_exit
,
6169 static void __net_exit
default_device_exit(struct net
*net
)
6171 struct net_device
*dev
, *aux
;
6173 * Push all migratable network devices back to the
6174 * initial network namespace
6177 for_each_netdev_safe(net
, dev
, aux
) {
6179 char fb_name
[IFNAMSIZ
];
6181 /* Ignore unmoveable devices (i.e. loopback) */
6182 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6185 /* Leave virtual devices for the generic cleanup */
6186 if (dev
->rtnl_link_ops
)
6189 /* Push remaing network devices to init_net */
6190 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6191 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6193 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
6194 __func__
, dev
->name
, err
);
6201 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6203 /* At exit all network devices most be removed from a network
6204 * namespace. Do this in the reverse order of registration.
6205 * Do this across as many network namespaces as possible to
6206 * improve batching efficiency.
6208 struct net_device
*dev
;
6210 LIST_HEAD(dev_kill_list
);
6213 list_for_each_entry(net
, net_list
, exit_list
) {
6214 for_each_netdev_reverse(net
, dev
) {
6215 if (dev
->rtnl_link_ops
)
6216 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6218 unregister_netdevice_queue(dev
, &dev_kill_list
);
6221 unregister_netdevice_many(&dev_kill_list
);
6225 static struct pernet_operations __net_initdata default_device_ops
= {
6226 .exit
= default_device_exit
,
6227 .exit_batch
= default_device_exit_batch
,
6231 * Initialize the DEV module. At boot time this walks the device list and
6232 * unhooks any devices that fail to initialise (normally hardware not
6233 * present) and leaves us with a valid list of present and active devices.
6238 * This is called single threaded during boot, so no need
6239 * to take the rtnl semaphore.
6241 static int __init
net_dev_init(void)
6243 int i
, rc
= -ENOMEM
;
6245 BUG_ON(!dev_boot_phase
);
6247 if (dev_proc_init())
6250 if (netdev_kobject_init())
6253 INIT_LIST_HEAD(&ptype_all
);
6254 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6255 INIT_LIST_HEAD(&ptype_base
[i
]);
6257 if (register_pernet_subsys(&netdev_net_ops
))
6261 * Initialise the packet receive queues.
6264 for_each_possible_cpu(i
) {
6265 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6267 memset(sd
, 0, sizeof(*sd
));
6268 skb_queue_head_init(&sd
->input_pkt_queue
);
6269 skb_queue_head_init(&sd
->process_queue
);
6270 sd
->completion_queue
= NULL
;
6271 INIT_LIST_HEAD(&sd
->poll_list
);
6272 sd
->output_queue
= NULL
;
6273 sd
->output_queue_tailp
= &sd
->output_queue
;
6275 sd
->csd
.func
= rps_trigger_softirq
;
6281 sd
->backlog
.poll
= process_backlog
;
6282 sd
->backlog
.weight
= weight_p
;
6283 sd
->backlog
.gro_list
= NULL
;
6284 sd
->backlog
.gro_count
= 0;
6289 /* The loopback device is special if any other network devices
6290 * is present in a network namespace the loopback device must
6291 * be present. Since we now dynamically allocate and free the
6292 * loopback device ensure this invariant is maintained by
6293 * keeping the loopback device as the first device on the
6294 * list of network devices. Ensuring the loopback devices
6295 * is the first device that appears and the last network device
6298 if (register_pernet_device(&loopback_net_ops
))
6301 if (register_pernet_device(&default_device_ops
))
6304 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6305 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6307 hotcpu_notifier(dev_cpu_callback
, 0);
6315 subsys_initcall(net_dev_init
);
6317 static int __init
initialize_hashrnd(void)
6319 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6323 late_initcall_sync(initialize_hashrnd
);