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
)
1285 list_add(&dev
->unreg_list
, &single
);
1286 return __dev_close_many(&single
);
1289 int dev_close_many(struct list_head
*head
)
1291 struct net_device
*dev
, *tmp
;
1292 LIST_HEAD(tmp_list
);
1294 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1295 if (!(dev
->flags
& IFF_UP
))
1296 list_move(&dev
->unreg_list
, &tmp_list
);
1298 __dev_close_many(head
);
1301 * Tell people we are down
1303 list_for_each_entry(dev
, head
, unreg_list
) {
1304 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1305 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1308 /* rollback_registered_many needs the complete original list */
1309 list_splice(&tmp_list
, head
);
1314 * dev_close - shutdown an interface.
1315 * @dev: device to shutdown
1317 * This function moves an active device into down state. A
1318 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1319 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1322 int dev_close(struct net_device
*dev
)
1326 list_add(&dev
->unreg_list
, &single
);
1327 dev_close_many(&single
);
1331 EXPORT_SYMBOL(dev_close
);
1335 * dev_disable_lro - disable Large Receive Offload on a device
1338 * Disable Large Receive Offload (LRO) on a net device. Must be
1339 * called under RTNL. This is needed if received packets may be
1340 * forwarded to another interface.
1342 void dev_disable_lro(struct net_device
*dev
)
1344 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1345 dev
->ethtool_ops
->set_flags
) {
1346 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1347 if (flags
& ETH_FLAG_LRO
) {
1348 flags
&= ~ETH_FLAG_LRO
;
1349 dev
->ethtool_ops
->set_flags(dev
, flags
);
1352 WARN_ON(dev
->features
& NETIF_F_LRO
);
1354 EXPORT_SYMBOL(dev_disable_lro
);
1357 static int dev_boot_phase
= 1;
1360 * Device change register/unregister. These are not inline or static
1361 * as we export them to the world.
1365 * register_netdevice_notifier - register a network notifier block
1368 * Register a notifier to be called when network device events occur.
1369 * The notifier passed is linked into the kernel structures and must
1370 * not be reused until it has been unregistered. A negative errno code
1371 * is returned on a failure.
1373 * When registered all registration and up events are replayed
1374 * to the new notifier to allow device to have a race free
1375 * view of the network device list.
1378 int register_netdevice_notifier(struct notifier_block
*nb
)
1380 struct net_device
*dev
;
1381 struct net_device
*last
;
1386 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1392 for_each_netdev(net
, dev
) {
1393 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1394 err
= notifier_to_errno(err
);
1398 if (!(dev
->flags
& IFF_UP
))
1401 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1412 for_each_netdev(net
, dev
) {
1416 if (dev
->flags
& IFF_UP
) {
1417 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1418 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1420 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1421 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1425 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1428 EXPORT_SYMBOL(register_netdevice_notifier
);
1431 * unregister_netdevice_notifier - unregister a network notifier block
1434 * Unregister a notifier previously registered by
1435 * register_netdevice_notifier(). The notifier is unlinked into the
1436 * kernel structures and may then be reused. A negative errno code
1437 * is returned on a failure.
1440 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1445 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1449 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1452 * call_netdevice_notifiers - call all network notifier blocks
1453 * @val: value passed unmodified to notifier function
1454 * @dev: net_device pointer passed unmodified to notifier function
1456 * Call all network notifier blocks. Parameters and return value
1457 * are as for raw_notifier_call_chain().
1460 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1463 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1466 /* When > 0 there are consumers of rx skb time stamps */
1467 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1469 void net_enable_timestamp(void)
1471 atomic_inc(&netstamp_needed
);
1473 EXPORT_SYMBOL(net_enable_timestamp
);
1475 void net_disable_timestamp(void)
1477 atomic_dec(&netstamp_needed
);
1479 EXPORT_SYMBOL(net_disable_timestamp
);
1481 static inline void net_timestamp_set(struct sk_buff
*skb
)
1483 if (atomic_read(&netstamp_needed
))
1484 __net_timestamp(skb
);
1486 skb
->tstamp
.tv64
= 0;
1489 static inline void net_timestamp_check(struct sk_buff
*skb
)
1491 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1492 __net_timestamp(skb
);
1496 * dev_forward_skb - loopback an skb to another netif
1498 * @dev: destination network device
1499 * @skb: buffer to forward
1502 * NET_RX_SUCCESS (no congestion)
1503 * NET_RX_DROP (packet was dropped, but freed)
1505 * dev_forward_skb can be used for injecting an skb from the
1506 * start_xmit function of one device into the receive queue
1507 * of another device.
1509 * The receiving device may be in another namespace, so
1510 * we have to clear all information in the skb that could
1511 * impact namespace isolation.
1513 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1518 if (unlikely(!(dev
->flags
& IFF_UP
) ||
1519 (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
)))) {
1520 atomic_long_inc(&dev
->rx_dropped
);
1524 skb_set_dev(skb
, dev
);
1525 skb
->tstamp
.tv64
= 0;
1526 skb
->pkt_type
= PACKET_HOST
;
1527 skb
->protocol
= eth_type_trans(skb
, dev
);
1528 return netif_rx(skb
);
1530 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1532 static inline int deliver_skb(struct sk_buff
*skb
,
1533 struct packet_type
*pt_prev
,
1534 struct net_device
*orig_dev
)
1536 atomic_inc(&skb
->users
);
1537 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1541 * Support routine. Sends outgoing frames to any network
1542 * taps currently in use.
1545 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1547 struct packet_type
*ptype
;
1548 struct sk_buff
*skb2
= NULL
;
1549 struct packet_type
*pt_prev
= NULL
;
1552 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1553 /* Never send packets back to the socket
1554 * they originated from - MvS (miquels@drinkel.ow.org)
1556 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1557 (ptype
->af_packet_priv
== NULL
||
1558 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1560 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1565 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1569 net_timestamp_set(skb2
);
1571 /* skb->nh should be correctly
1572 set by sender, so that the second statement is
1573 just protection against buggy protocols.
1575 skb_reset_mac_header(skb2
);
1577 if (skb_network_header(skb2
) < skb2
->data
||
1578 skb2
->network_header
> skb2
->tail
) {
1579 if (net_ratelimit())
1580 printk(KERN_CRIT
"protocol %04x is "
1582 ntohs(skb2
->protocol
),
1584 skb_reset_network_header(skb2
);
1587 skb2
->transport_header
= skb2
->network_header
;
1588 skb2
->pkt_type
= PACKET_OUTGOING
;
1593 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1598 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1599 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1601 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1605 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1608 if (dev
->reg_state
== NETREG_REGISTERED
) {
1611 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1616 if (txq
< dev
->real_num_tx_queues
)
1617 qdisc_reset_all_tx_gt(dev
, txq
);
1620 dev
->real_num_tx_queues
= txq
;
1623 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1627 * netif_set_real_num_rx_queues - set actual number of RX queues used
1628 * @dev: Network device
1629 * @rxq: Actual number of RX queues
1631 * This must be called either with the rtnl_lock held or before
1632 * registration of the net device. Returns 0 on success, or a
1633 * negative error code. If called before registration, it always
1636 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1640 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1643 if (dev
->reg_state
== NETREG_REGISTERED
) {
1646 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1652 dev
->real_num_rx_queues
= rxq
;
1655 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1658 static inline void __netif_reschedule(struct Qdisc
*q
)
1660 struct softnet_data
*sd
;
1661 unsigned long flags
;
1663 local_irq_save(flags
);
1664 sd
= &__get_cpu_var(softnet_data
);
1665 q
->next_sched
= NULL
;
1666 *sd
->output_queue_tailp
= q
;
1667 sd
->output_queue_tailp
= &q
->next_sched
;
1668 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1669 local_irq_restore(flags
);
1672 void __netif_schedule(struct Qdisc
*q
)
1674 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1675 __netif_reschedule(q
);
1677 EXPORT_SYMBOL(__netif_schedule
);
1679 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1681 if (atomic_dec_and_test(&skb
->users
)) {
1682 struct softnet_data
*sd
;
1683 unsigned long flags
;
1685 local_irq_save(flags
);
1686 sd
= &__get_cpu_var(softnet_data
);
1687 skb
->next
= sd
->completion_queue
;
1688 sd
->completion_queue
= skb
;
1689 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1690 local_irq_restore(flags
);
1693 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1695 void dev_kfree_skb_any(struct sk_buff
*skb
)
1697 if (in_irq() || irqs_disabled())
1698 dev_kfree_skb_irq(skb
);
1702 EXPORT_SYMBOL(dev_kfree_skb_any
);
1706 * netif_device_detach - mark device as removed
1707 * @dev: network device
1709 * Mark device as removed from system and therefore no longer available.
1711 void netif_device_detach(struct net_device
*dev
)
1713 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1714 netif_running(dev
)) {
1715 netif_tx_stop_all_queues(dev
);
1718 EXPORT_SYMBOL(netif_device_detach
);
1721 * netif_device_attach - mark device as attached
1722 * @dev: network device
1724 * Mark device as attached from system and restart if needed.
1726 void netif_device_attach(struct net_device
*dev
)
1728 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1729 netif_running(dev
)) {
1730 netif_tx_wake_all_queues(dev
);
1731 __netdev_watchdog_up(dev
);
1734 EXPORT_SYMBOL(netif_device_attach
);
1737 * skb_dev_set -- assign a new device to a buffer
1738 * @skb: buffer for the new device
1739 * @dev: network device
1741 * If an skb is owned by a device already, we have to reset
1742 * all data private to the namespace a device belongs to
1743 * before assigning it a new device.
1745 #ifdef CONFIG_NET_NS
1746 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1749 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1752 skb_init_secmark(skb
);
1756 skb
->ipvs_property
= 0;
1757 #ifdef CONFIG_NET_SCHED
1763 EXPORT_SYMBOL(skb_set_dev
);
1764 #endif /* CONFIG_NET_NS */
1767 * Invalidate hardware checksum when packet is to be mangled, and
1768 * complete checksum manually on outgoing path.
1770 int skb_checksum_help(struct sk_buff
*skb
)
1773 int ret
= 0, offset
;
1775 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1776 goto out_set_summed
;
1778 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1779 /* Let GSO fix up the checksum. */
1780 goto out_set_summed
;
1783 offset
= skb_checksum_start_offset(skb
);
1784 BUG_ON(offset
>= skb_headlen(skb
));
1785 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1787 offset
+= skb
->csum_offset
;
1788 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1790 if (skb_cloned(skb
) &&
1791 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1792 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1797 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1799 skb
->ip_summed
= CHECKSUM_NONE
;
1803 EXPORT_SYMBOL(skb_checksum_help
);
1806 * skb_gso_segment - Perform segmentation on skb.
1807 * @skb: buffer to segment
1808 * @features: features for the output path (see dev->features)
1810 * This function segments the given skb and returns a list of segments.
1812 * It may return NULL if the skb requires no segmentation. This is
1813 * only possible when GSO is used for verifying header integrity.
1815 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1817 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1818 struct packet_type
*ptype
;
1819 __be16 type
= skb
->protocol
;
1820 int vlan_depth
= ETH_HLEN
;
1823 while (type
== htons(ETH_P_8021Q
)) {
1824 struct vlan_hdr
*vh
;
1826 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1827 return ERR_PTR(-EINVAL
);
1829 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1830 type
= vh
->h_vlan_encapsulated_proto
;
1831 vlan_depth
+= VLAN_HLEN
;
1834 skb_reset_mac_header(skb
);
1835 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1836 __skb_pull(skb
, skb
->mac_len
);
1838 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1839 struct net_device
*dev
= skb
->dev
;
1840 struct ethtool_drvinfo info
= {};
1842 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1843 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1845 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1846 info
.driver
, dev
? dev
->features
: 0L,
1847 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1848 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1850 if (skb_header_cloned(skb
) &&
1851 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1852 return ERR_PTR(err
);
1856 list_for_each_entry_rcu(ptype
,
1857 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1858 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1859 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1860 err
= ptype
->gso_send_check(skb
);
1861 segs
= ERR_PTR(err
);
1862 if (err
|| skb_gso_ok(skb
, features
))
1864 __skb_push(skb
, (skb
->data
-
1865 skb_network_header(skb
)));
1867 segs
= ptype
->gso_segment(skb
, features
);
1873 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1877 EXPORT_SYMBOL(skb_gso_segment
);
1879 /* Take action when hardware reception checksum errors are detected. */
1881 void netdev_rx_csum_fault(struct net_device
*dev
)
1883 if (net_ratelimit()) {
1884 printk(KERN_ERR
"%s: hw csum failure.\n",
1885 dev
? dev
->name
: "<unknown>");
1889 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1892 /* Actually, we should eliminate this check as soon as we know, that:
1893 * 1. IOMMU is present and allows to map all the memory.
1894 * 2. No high memory really exists on this machine.
1897 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1899 #ifdef CONFIG_HIGHMEM
1901 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1902 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1903 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1907 if (PCI_DMA_BUS_IS_PHYS
) {
1908 struct device
*pdev
= dev
->dev
.parent
;
1912 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1913 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1914 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1923 void (*destructor
)(struct sk_buff
*skb
);
1926 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1928 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1930 struct dev_gso_cb
*cb
;
1933 struct sk_buff
*nskb
= skb
->next
;
1935 skb
->next
= nskb
->next
;
1938 } while (skb
->next
);
1940 cb
= DEV_GSO_CB(skb
);
1942 cb
->destructor(skb
);
1946 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1947 * @skb: buffer to segment
1948 * @features: device features as applicable to this skb
1950 * This function segments the given skb and stores the list of segments
1953 static int dev_gso_segment(struct sk_buff
*skb
, int features
)
1955 struct sk_buff
*segs
;
1957 segs
= skb_gso_segment(skb
, features
);
1959 /* Verifying header integrity only. */
1964 return PTR_ERR(segs
);
1967 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1968 skb
->destructor
= dev_gso_skb_destructor
;
1974 * Try to orphan skb early, right before transmission by the device.
1975 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1976 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1978 static inline void skb_orphan_try(struct sk_buff
*skb
)
1980 struct sock
*sk
= skb
->sk
;
1982 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
1983 /* skb_tx_hash() wont be able to get sk.
1984 * We copy sk_hash into skb->rxhash
1987 skb
->rxhash
= sk
->sk_hash
;
1992 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1994 return ((features
& NETIF_F_GEN_CSUM
) ||
1995 ((features
& NETIF_F_V4_CSUM
) &&
1996 protocol
== htons(ETH_P_IP
)) ||
1997 ((features
& NETIF_F_V6_CSUM
) &&
1998 protocol
== htons(ETH_P_IPV6
)) ||
1999 ((features
& NETIF_F_FCOE_CRC
) &&
2000 protocol
== htons(ETH_P_FCOE
)));
2003 static int harmonize_features(struct sk_buff
*skb
, __be16 protocol
, int features
)
2005 if (!can_checksum_protocol(features
, protocol
)) {
2006 features
&= ~NETIF_F_ALL_CSUM
;
2007 features
&= ~NETIF_F_SG
;
2008 } else if (illegal_highdma(skb
->dev
, skb
)) {
2009 features
&= ~NETIF_F_SG
;
2015 int netif_skb_features(struct sk_buff
*skb
)
2017 __be16 protocol
= skb
->protocol
;
2018 int features
= skb
->dev
->features
;
2020 if (protocol
== htons(ETH_P_8021Q
)) {
2021 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2022 protocol
= veh
->h_vlan_encapsulated_proto
;
2023 } else if (!vlan_tx_tag_present(skb
)) {
2024 return harmonize_features(skb
, protocol
, features
);
2027 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2029 if (protocol
!= htons(ETH_P_8021Q
)) {
2030 return harmonize_features(skb
, protocol
, features
);
2032 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2033 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2034 return harmonize_features(skb
, protocol
, features
);
2037 EXPORT_SYMBOL(netif_skb_features
);
2040 * Returns true if either:
2041 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2042 * 2. skb is fragmented and the device does not support SG, or if
2043 * at least one of fragments is in highmem and device does not
2044 * support DMA from it.
2046 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2049 return skb_is_nonlinear(skb
) &&
2050 ((skb_has_frag_list(skb
) &&
2051 !(features
& NETIF_F_FRAGLIST
)) ||
2052 (skb_shinfo(skb
)->nr_frags
&&
2053 !(features
& NETIF_F_SG
)));
2056 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2057 struct netdev_queue
*txq
)
2059 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2060 int rc
= NETDEV_TX_OK
;
2062 if (likely(!skb
->next
)) {
2066 * If device doesnt need skb->dst, release it right now while
2067 * its hot in this cpu cache
2069 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2072 if (!list_empty(&ptype_all
))
2073 dev_queue_xmit_nit(skb
, dev
);
2075 skb_orphan_try(skb
);
2077 features
= netif_skb_features(skb
);
2079 if (vlan_tx_tag_present(skb
) &&
2080 !(features
& NETIF_F_HW_VLAN_TX
)) {
2081 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2088 if (netif_needs_gso(skb
, features
)) {
2089 if (unlikely(dev_gso_segment(skb
, features
)))
2094 if (skb_needs_linearize(skb
, features
) &&
2095 __skb_linearize(skb
))
2098 /* If packet is not checksummed and device does not
2099 * support checksumming for this protocol, complete
2100 * checksumming here.
2102 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2103 skb_set_transport_header(skb
,
2104 skb_checksum_start_offset(skb
));
2105 if (!(features
& NETIF_F_ALL_CSUM
) &&
2106 skb_checksum_help(skb
))
2111 rc
= ops
->ndo_start_xmit(skb
, dev
);
2112 trace_net_dev_xmit(skb
, rc
);
2113 if (rc
== NETDEV_TX_OK
)
2114 txq_trans_update(txq
);
2120 struct sk_buff
*nskb
= skb
->next
;
2122 skb
->next
= nskb
->next
;
2126 * If device doesnt need nskb->dst, release it right now while
2127 * its hot in this cpu cache
2129 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2132 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2133 trace_net_dev_xmit(nskb
, rc
);
2134 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2135 if (rc
& ~NETDEV_TX_MASK
)
2136 goto out_kfree_gso_skb
;
2137 nskb
->next
= skb
->next
;
2141 txq_trans_update(txq
);
2142 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2143 return NETDEV_TX_BUSY
;
2144 } while (skb
->next
);
2147 if (likely(skb
->next
== NULL
))
2148 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2155 static u32 hashrnd __read_mostly
;
2158 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2159 * to be used as a distribution range.
2161 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2162 unsigned int num_tx_queues
)
2166 if (skb_rx_queue_recorded(skb
)) {
2167 hash
= skb_get_rx_queue(skb
);
2168 while (unlikely(hash
>= num_tx_queues
))
2169 hash
-= num_tx_queues
;
2173 if (skb
->sk
&& skb
->sk
->sk_hash
)
2174 hash
= skb
->sk
->sk_hash
;
2176 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2177 hash
= jhash_1word(hash
, hashrnd
);
2179 return (u16
) (((u64
) hash
* num_tx_queues
) >> 32);
2181 EXPORT_SYMBOL(__skb_tx_hash
);
2183 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2185 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2186 if (net_ratelimit()) {
2187 pr_warning("%s selects TX queue %d, but "
2188 "real number of TX queues is %d\n",
2189 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2196 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2199 struct xps_dev_maps
*dev_maps
;
2200 struct xps_map
*map
;
2201 int queue_index
= -1;
2204 dev_maps
= rcu_dereference(dev
->xps_maps
);
2206 map
= rcu_dereference(
2207 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2210 queue_index
= map
->queues
[0];
2213 if (skb
->sk
&& skb
->sk
->sk_hash
)
2214 hash
= skb
->sk
->sk_hash
;
2216 hash
= (__force u16
) skb
->protocol
^
2218 hash
= jhash_1word(hash
, hashrnd
);
2219 queue_index
= map
->queues
[
2220 ((u64
)hash
* map
->len
) >> 32];
2222 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2234 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2235 struct sk_buff
*skb
)
2238 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2240 if (dev
->real_num_tx_queues
== 1)
2242 else if (ops
->ndo_select_queue
) {
2243 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2244 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2246 struct sock
*sk
= skb
->sk
;
2247 queue_index
= sk_tx_queue_get(sk
);
2249 if (queue_index
< 0 || skb
->ooo_okay
||
2250 queue_index
>= dev
->real_num_tx_queues
) {
2251 int old_index
= queue_index
;
2253 queue_index
= get_xps_queue(dev
, skb
);
2254 if (queue_index
< 0)
2255 queue_index
= skb_tx_hash(dev
, skb
);
2257 if (queue_index
!= old_index
&& sk
) {
2258 struct dst_entry
*dst
=
2259 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2261 if (dst
&& skb_dst(skb
) == dst
)
2262 sk_tx_queue_set(sk
, queue_index
);
2267 skb_set_queue_mapping(skb
, queue_index
);
2268 return netdev_get_tx_queue(dev
, queue_index
);
2271 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2272 struct net_device
*dev
,
2273 struct netdev_queue
*txq
)
2275 spinlock_t
*root_lock
= qdisc_lock(q
);
2276 bool contended
= qdisc_is_running(q
);
2280 * Heuristic to force contended enqueues to serialize on a
2281 * separate lock before trying to get qdisc main lock.
2282 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2283 * and dequeue packets faster.
2285 if (unlikely(contended
))
2286 spin_lock(&q
->busylock
);
2288 spin_lock(root_lock
);
2289 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2292 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2293 qdisc_run_begin(q
)) {
2295 * This is a work-conserving queue; there are no old skbs
2296 * waiting to be sent out; and the qdisc is not running -
2297 * xmit the skb directly.
2299 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2302 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2303 qdisc_bstats_update(q
, skb
);
2305 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2306 if (unlikely(contended
)) {
2307 spin_unlock(&q
->busylock
);
2314 rc
= NET_XMIT_SUCCESS
;
2317 rc
= qdisc_enqueue_root(skb
, q
);
2318 if (qdisc_run_begin(q
)) {
2319 if (unlikely(contended
)) {
2320 spin_unlock(&q
->busylock
);
2326 spin_unlock(root_lock
);
2327 if (unlikely(contended
))
2328 spin_unlock(&q
->busylock
);
2332 static DEFINE_PER_CPU(int, xmit_recursion
);
2333 #define RECURSION_LIMIT 10
2336 * dev_queue_xmit - transmit a buffer
2337 * @skb: buffer to transmit
2339 * Queue a buffer for transmission to a network device. The caller must
2340 * have set the device and priority and built the buffer before calling
2341 * this function. The function can be called from an interrupt.
2343 * A negative errno code is returned on a failure. A success does not
2344 * guarantee the frame will be transmitted as it may be dropped due
2345 * to congestion or traffic shaping.
2347 * -----------------------------------------------------------------------------------
2348 * I notice this method can also return errors from the queue disciplines,
2349 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2352 * Regardless of the return value, the skb is consumed, so it is currently
2353 * difficult to retry a send to this method. (You can bump the ref count
2354 * before sending to hold a reference for retry if you are careful.)
2356 * When calling this method, interrupts MUST be enabled. This is because
2357 * the BH enable code must have IRQs enabled so that it will not deadlock.
2360 int dev_queue_xmit(struct sk_buff
*skb
)
2362 struct net_device
*dev
= skb
->dev
;
2363 struct netdev_queue
*txq
;
2367 /* Disable soft irqs for various locks below. Also
2368 * stops preemption for RCU.
2372 txq
= dev_pick_tx(dev
, skb
);
2373 q
= rcu_dereference_bh(txq
->qdisc
);
2375 #ifdef CONFIG_NET_CLS_ACT
2376 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2378 trace_net_dev_queue(skb
);
2380 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2384 /* The device has no queue. Common case for software devices:
2385 loopback, all the sorts of tunnels...
2387 Really, it is unlikely that netif_tx_lock protection is necessary
2388 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2390 However, it is possible, that they rely on protection
2393 Check this and shot the lock. It is not prone from deadlocks.
2394 Either shot noqueue qdisc, it is even simpler 8)
2396 if (dev
->flags
& IFF_UP
) {
2397 int cpu
= smp_processor_id(); /* ok because BHs are off */
2399 if (txq
->xmit_lock_owner
!= cpu
) {
2401 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2402 goto recursion_alert
;
2404 HARD_TX_LOCK(dev
, txq
, cpu
);
2406 if (!netif_tx_queue_stopped(txq
)) {
2407 __this_cpu_inc(xmit_recursion
);
2408 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2409 __this_cpu_dec(xmit_recursion
);
2410 if (dev_xmit_complete(rc
)) {
2411 HARD_TX_UNLOCK(dev
, txq
);
2415 HARD_TX_UNLOCK(dev
, txq
);
2416 if (net_ratelimit())
2417 printk(KERN_CRIT
"Virtual device %s asks to "
2418 "queue packet!\n", dev
->name
);
2420 /* Recursion is detected! It is possible,
2424 if (net_ratelimit())
2425 printk(KERN_CRIT
"Dead loop on virtual device "
2426 "%s, fix it urgently!\n", dev
->name
);
2431 rcu_read_unlock_bh();
2436 rcu_read_unlock_bh();
2439 EXPORT_SYMBOL(dev_queue_xmit
);
2442 /*=======================================================================
2444 =======================================================================*/
2446 int netdev_max_backlog __read_mostly
= 1000;
2447 int netdev_tstamp_prequeue __read_mostly
= 1;
2448 int netdev_budget __read_mostly
= 300;
2449 int weight_p __read_mostly
= 64; /* old backlog weight */
2451 /* Called with irq disabled */
2452 static inline void ____napi_schedule(struct softnet_data
*sd
,
2453 struct napi_struct
*napi
)
2455 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2456 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2460 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2461 * and src/dst port numbers. Returns a non-zero hash number on success
2464 __u32
__skb_get_rxhash(struct sk_buff
*skb
)
2466 int nhoff
, hash
= 0, poff
;
2467 struct ipv6hdr
*ip6
;
2470 u32 addr1
, addr2
, ihl
;
2476 nhoff
= skb_network_offset(skb
);
2478 switch (skb
->protocol
) {
2479 case __constant_htons(ETH_P_IP
):
2480 if (!pskb_may_pull(skb
, sizeof(*ip
) + nhoff
))
2483 ip
= (struct iphdr
*) (skb
->data
+ nhoff
);
2484 if (ip
->frag_off
& htons(IP_MF
| IP_OFFSET
))
2487 ip_proto
= ip
->protocol
;
2488 addr1
= (__force u32
) ip
->saddr
;
2489 addr2
= (__force u32
) ip
->daddr
;
2492 case __constant_htons(ETH_P_IPV6
):
2493 if (!pskb_may_pull(skb
, sizeof(*ip6
) + nhoff
))
2496 ip6
= (struct ipv6hdr
*) (skb
->data
+ nhoff
);
2497 ip_proto
= ip6
->nexthdr
;
2498 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2499 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2507 poff
= proto_ports_offset(ip_proto
);
2509 nhoff
+= ihl
* 4 + poff
;
2510 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2511 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2512 if (ports
.v16
[1] < ports
.v16
[0])
2513 swap(ports
.v16
[0], ports
.v16
[1]);
2517 /* get a consistent hash (same value on both flow directions) */
2521 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2528 EXPORT_SYMBOL(__skb_get_rxhash
);
2532 /* One global table that all flow-based protocols share. */
2533 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2534 EXPORT_SYMBOL(rps_sock_flow_table
);
2537 * get_rps_cpu is called from netif_receive_skb and returns the target
2538 * CPU from the RPS map of the receiving queue for a given skb.
2539 * rcu_read_lock must be held on entry.
2541 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2542 struct rps_dev_flow
**rflowp
)
2544 struct netdev_rx_queue
*rxqueue
;
2545 struct rps_map
*map
;
2546 struct rps_dev_flow_table
*flow_table
;
2547 struct rps_sock_flow_table
*sock_flow_table
;
2551 if (skb_rx_queue_recorded(skb
)) {
2552 u16 index
= skb_get_rx_queue(skb
);
2553 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2554 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2555 "%s received packet on queue %u, but number "
2556 "of RX queues is %u\n",
2557 dev
->name
, index
, dev
->real_num_rx_queues
);
2560 rxqueue
= dev
->_rx
+ index
;
2564 map
= rcu_dereference(rxqueue
->rps_map
);
2566 if (map
->len
== 1 &&
2567 !rcu_dereference_raw(rxqueue
->rps_flow_table
)) {
2568 tcpu
= map
->cpus
[0];
2569 if (cpu_online(tcpu
))
2573 } else if (!rcu_dereference_raw(rxqueue
->rps_flow_table
)) {
2577 skb_reset_network_header(skb
);
2578 if (!skb_get_rxhash(skb
))
2581 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2582 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2583 if (flow_table
&& sock_flow_table
) {
2585 struct rps_dev_flow
*rflow
;
2587 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2590 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2591 sock_flow_table
->mask
];
2594 * If the desired CPU (where last recvmsg was done) is
2595 * different from current CPU (one in the rx-queue flow
2596 * table entry), switch if one of the following holds:
2597 * - Current CPU is unset (equal to RPS_NO_CPU).
2598 * - Current CPU is offline.
2599 * - The current CPU's queue tail has advanced beyond the
2600 * last packet that was enqueued using this table entry.
2601 * This guarantees that all previous packets for the flow
2602 * have been dequeued, thus preserving in order delivery.
2604 if (unlikely(tcpu
!= next_cpu
) &&
2605 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2606 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2607 rflow
->last_qtail
)) >= 0)) {
2608 tcpu
= rflow
->cpu
= next_cpu
;
2609 if (tcpu
!= RPS_NO_CPU
)
2610 rflow
->last_qtail
= per_cpu(softnet_data
,
2611 tcpu
).input_queue_head
;
2613 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2621 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2623 if (cpu_online(tcpu
)) {
2633 /* Called from hardirq (IPI) context */
2634 static void rps_trigger_softirq(void *data
)
2636 struct softnet_data
*sd
= data
;
2638 ____napi_schedule(sd
, &sd
->backlog
);
2642 #endif /* CONFIG_RPS */
2645 * Check if this softnet_data structure is another cpu one
2646 * If yes, queue it to our IPI list and return 1
2649 static int rps_ipi_queued(struct softnet_data
*sd
)
2652 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2655 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2656 mysd
->rps_ipi_list
= sd
;
2658 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2661 #endif /* CONFIG_RPS */
2666 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2667 * queue (may be a remote CPU queue).
2669 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2670 unsigned int *qtail
)
2672 struct softnet_data
*sd
;
2673 unsigned long flags
;
2675 sd
= &per_cpu(softnet_data
, cpu
);
2677 local_irq_save(flags
);
2680 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2681 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2683 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2684 input_queue_tail_incr_save(sd
, qtail
);
2686 local_irq_restore(flags
);
2687 return NET_RX_SUCCESS
;
2690 /* Schedule NAPI for backlog device
2691 * We can use non atomic operation since we own the queue lock
2693 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2694 if (!rps_ipi_queued(sd
))
2695 ____napi_schedule(sd
, &sd
->backlog
);
2703 local_irq_restore(flags
);
2705 atomic_long_inc(&skb
->dev
->rx_dropped
);
2711 * netif_rx - post buffer to the network code
2712 * @skb: buffer to post
2714 * This function receives a packet from a device driver and queues it for
2715 * the upper (protocol) levels to process. It always succeeds. The buffer
2716 * may be dropped during processing for congestion control or by the
2720 * NET_RX_SUCCESS (no congestion)
2721 * NET_RX_DROP (packet was dropped)
2725 int netif_rx(struct sk_buff
*skb
)
2729 /* if netpoll wants it, pretend we never saw it */
2730 if (netpoll_rx(skb
))
2733 if (netdev_tstamp_prequeue
)
2734 net_timestamp_check(skb
);
2736 trace_netif_rx(skb
);
2739 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2745 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2747 cpu
= smp_processor_id();
2749 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2757 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2763 EXPORT_SYMBOL(netif_rx
);
2765 int netif_rx_ni(struct sk_buff
*skb
)
2770 err
= netif_rx(skb
);
2771 if (local_softirq_pending())
2777 EXPORT_SYMBOL(netif_rx_ni
);
2779 static void net_tx_action(struct softirq_action
*h
)
2781 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2783 if (sd
->completion_queue
) {
2784 struct sk_buff
*clist
;
2786 local_irq_disable();
2787 clist
= sd
->completion_queue
;
2788 sd
->completion_queue
= NULL
;
2792 struct sk_buff
*skb
= clist
;
2793 clist
= clist
->next
;
2795 WARN_ON(atomic_read(&skb
->users
));
2796 trace_kfree_skb(skb
, net_tx_action
);
2801 if (sd
->output_queue
) {
2804 local_irq_disable();
2805 head
= sd
->output_queue
;
2806 sd
->output_queue
= NULL
;
2807 sd
->output_queue_tailp
= &sd
->output_queue
;
2811 struct Qdisc
*q
= head
;
2812 spinlock_t
*root_lock
;
2814 head
= head
->next_sched
;
2816 root_lock
= qdisc_lock(q
);
2817 if (spin_trylock(root_lock
)) {
2818 smp_mb__before_clear_bit();
2819 clear_bit(__QDISC_STATE_SCHED
,
2822 spin_unlock(root_lock
);
2824 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2826 __netif_reschedule(q
);
2828 smp_mb__before_clear_bit();
2829 clear_bit(__QDISC_STATE_SCHED
,
2837 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2838 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2839 /* This hook is defined here for ATM LANE */
2840 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2841 unsigned char *addr
) __read_mostly
;
2842 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2845 #ifdef CONFIG_NET_CLS_ACT
2846 /* TODO: Maybe we should just force sch_ingress to be compiled in
2847 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2848 * a compare and 2 stores extra right now if we dont have it on
2849 * but have CONFIG_NET_CLS_ACT
2850 * NOTE: This doesnt stop any functionality; if you dont have
2851 * the ingress scheduler, you just cant add policies on ingress.
2854 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
2856 struct net_device
*dev
= skb
->dev
;
2857 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2858 int result
= TC_ACT_OK
;
2861 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
2862 if (net_ratelimit())
2863 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2864 skb
->skb_iif
, dev
->ifindex
);
2868 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2869 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2872 if (q
!= &noop_qdisc
) {
2873 spin_lock(qdisc_lock(q
));
2874 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2875 result
= qdisc_enqueue_root(skb
, q
);
2876 spin_unlock(qdisc_lock(q
));
2882 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2883 struct packet_type
**pt_prev
,
2884 int *ret
, struct net_device
*orig_dev
)
2886 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
2888 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
2892 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2896 switch (ing_filter(skb
, rxq
)) {
2910 * netdev_rx_handler_register - register receive handler
2911 * @dev: device to register a handler for
2912 * @rx_handler: receive handler to register
2913 * @rx_handler_data: data pointer that is used by rx handler
2915 * Register a receive hander for a device. This handler will then be
2916 * called from __netif_receive_skb. A negative errno code is returned
2919 * The caller must hold the rtnl_mutex.
2921 int netdev_rx_handler_register(struct net_device
*dev
,
2922 rx_handler_func_t
*rx_handler
,
2923 void *rx_handler_data
)
2927 if (dev
->rx_handler
)
2930 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
2931 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
2935 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
2938 * netdev_rx_handler_unregister - unregister receive handler
2939 * @dev: device to unregister a handler from
2941 * Unregister a receive hander from a device.
2943 * The caller must hold the rtnl_mutex.
2945 void netdev_rx_handler_unregister(struct net_device
*dev
)
2949 rcu_assign_pointer(dev
->rx_handler
, NULL
);
2950 rcu_assign_pointer(dev
->rx_handler_data
, NULL
);
2952 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
2954 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2955 struct net_device
*master
)
2957 if (skb
->pkt_type
== PACKET_HOST
) {
2958 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2960 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2964 /* On bonding slaves other than the currently active slave, suppress
2965 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2966 * ARP on active-backup slaves with arp_validate enabled.
2968 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2970 struct net_device
*dev
= skb
->dev
;
2972 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2973 dev
->last_rx
= jiffies
;
2975 if ((master
->priv_flags
& IFF_MASTER_ALB
) &&
2976 (master
->priv_flags
& IFF_BRIDGE_PORT
)) {
2977 /* Do address unmangle. The local destination address
2978 * will be always the one master has. Provides the right
2979 * functionality in a bridge.
2981 skb_bond_set_mac_by_master(skb
, master
);
2984 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
2985 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
2986 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
2989 if (master
->priv_flags
& IFF_MASTER_ALB
) {
2990 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
2991 skb
->pkt_type
!= PACKET_MULTICAST
)
2994 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
2995 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
3002 EXPORT_SYMBOL(__skb_bond_should_drop
);
3004 static int __netif_receive_skb(struct sk_buff
*skb
)
3006 struct packet_type
*ptype
, *pt_prev
;
3007 rx_handler_func_t
*rx_handler
;
3008 struct net_device
*orig_dev
;
3009 struct net_device
*master
;
3010 struct net_device
*null_or_orig
;
3011 struct net_device
*orig_or_bond
;
3012 int ret
= NET_RX_DROP
;
3015 if (!netdev_tstamp_prequeue
)
3016 net_timestamp_check(skb
);
3018 trace_netif_receive_skb(skb
);
3020 /* if we've gotten here through NAPI, check netpoll */
3021 if (netpoll_receive_skb(skb
))
3025 skb
->skb_iif
= skb
->dev
->ifindex
;
3028 * bonding note: skbs received on inactive slaves should only
3029 * be delivered to pkt handlers that are exact matches. Also
3030 * the deliver_no_wcard flag will be set. If packet handlers
3031 * are sensitive to duplicate packets these skbs will need to
3032 * be dropped at the handler.
3034 null_or_orig
= NULL
;
3035 orig_dev
= skb
->dev
;
3036 master
= ACCESS_ONCE(orig_dev
->master
);
3037 if (skb
->deliver_no_wcard
)
3038 null_or_orig
= orig_dev
;
3040 if (skb_bond_should_drop(skb
, master
)) {
3041 skb
->deliver_no_wcard
= 1;
3042 null_or_orig
= orig_dev
; /* deliver only exact match */
3047 __this_cpu_inc(softnet_data
.processed
);
3048 skb_reset_network_header(skb
);
3049 skb_reset_transport_header(skb
);
3050 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
3056 #ifdef CONFIG_NET_CLS_ACT
3057 if (skb
->tc_verd
& TC_NCLS
) {
3058 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3063 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3064 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
3065 ptype
->dev
== orig_dev
) {
3067 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3072 #ifdef CONFIG_NET_CLS_ACT
3073 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3079 /* Handle special case of bridge or macvlan */
3080 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3083 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3086 skb
= rx_handler(skb
);
3091 if (vlan_tx_tag_present(skb
)) {
3093 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3096 if (vlan_hwaccel_do_receive(&skb
)) {
3097 ret
= __netif_receive_skb(skb
);
3099 } else if (unlikely(!skb
))
3104 * Make sure frames received on VLAN interfaces stacked on
3105 * bonding interfaces still make their way to any base bonding
3106 * device that may have registered for a specific ptype. The
3107 * handler may have to adjust skb->dev and orig_dev.
3109 orig_or_bond
= orig_dev
;
3110 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
3111 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
3112 orig_or_bond
= vlan_dev_real_dev(skb
->dev
);
3115 type
= skb
->protocol
;
3116 list_for_each_entry_rcu(ptype
,
3117 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3118 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
3119 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
3120 ptype
->dev
== orig_or_bond
)) {
3122 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3128 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3130 atomic_long_inc(&skb
->dev
->rx_dropped
);
3132 /* Jamal, now you will not able to escape explaining
3133 * me how you were going to use this. :-)
3144 * netif_receive_skb - process receive buffer from network
3145 * @skb: buffer to process
3147 * netif_receive_skb() is the main receive data processing function.
3148 * It always succeeds. The buffer may be dropped during processing
3149 * for congestion control or by the protocol layers.
3151 * This function may only be called from softirq context and interrupts
3152 * should be enabled.
3154 * Return values (usually ignored):
3155 * NET_RX_SUCCESS: no congestion
3156 * NET_RX_DROP: packet was dropped
3158 int netif_receive_skb(struct sk_buff
*skb
)
3160 if (netdev_tstamp_prequeue
)
3161 net_timestamp_check(skb
);
3163 if (skb_defer_rx_timestamp(skb
))
3164 return NET_RX_SUCCESS
;
3168 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3173 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3176 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3180 ret
= __netif_receive_skb(skb
);
3186 return __netif_receive_skb(skb
);
3189 EXPORT_SYMBOL(netif_receive_skb
);
3191 /* Network device is going away, flush any packets still pending
3192 * Called with irqs disabled.
3194 static void flush_backlog(void *arg
)
3196 struct net_device
*dev
= arg
;
3197 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3198 struct sk_buff
*skb
, *tmp
;
3201 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3202 if (skb
->dev
== dev
) {
3203 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3205 input_queue_head_incr(sd
);
3210 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3211 if (skb
->dev
== dev
) {
3212 __skb_unlink(skb
, &sd
->process_queue
);
3214 input_queue_head_incr(sd
);
3219 static int napi_gro_complete(struct sk_buff
*skb
)
3221 struct packet_type
*ptype
;
3222 __be16 type
= skb
->protocol
;
3223 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3226 if (NAPI_GRO_CB(skb
)->count
== 1) {
3227 skb_shinfo(skb
)->gso_size
= 0;
3232 list_for_each_entry_rcu(ptype
, head
, list
) {
3233 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3236 err
= ptype
->gro_complete(skb
);
3242 WARN_ON(&ptype
->list
== head
);
3244 return NET_RX_SUCCESS
;
3248 return netif_receive_skb(skb
);
3251 inline void napi_gro_flush(struct napi_struct
*napi
)
3253 struct sk_buff
*skb
, *next
;
3255 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3258 napi_gro_complete(skb
);
3261 napi
->gro_count
= 0;
3262 napi
->gro_list
= NULL
;
3264 EXPORT_SYMBOL(napi_gro_flush
);
3266 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3268 struct sk_buff
**pp
= NULL
;
3269 struct packet_type
*ptype
;
3270 __be16 type
= skb
->protocol
;
3271 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3274 enum gro_result ret
;
3276 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3279 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3283 list_for_each_entry_rcu(ptype
, head
, list
) {
3284 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3287 skb_set_network_header(skb
, skb_gro_offset(skb
));
3288 mac_len
= skb
->network_header
- skb
->mac_header
;
3289 skb
->mac_len
= mac_len
;
3290 NAPI_GRO_CB(skb
)->same_flow
= 0;
3291 NAPI_GRO_CB(skb
)->flush
= 0;
3292 NAPI_GRO_CB(skb
)->free
= 0;
3294 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3299 if (&ptype
->list
== head
)
3302 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3303 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3306 struct sk_buff
*nskb
= *pp
;
3310 napi_gro_complete(nskb
);
3317 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3321 NAPI_GRO_CB(skb
)->count
= 1;
3322 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3323 skb
->next
= napi
->gro_list
;
3324 napi
->gro_list
= skb
;
3328 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3329 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3331 BUG_ON(skb
->end
- skb
->tail
< grow
);
3333 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3336 skb
->data_len
-= grow
;
3338 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3339 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3341 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3342 put_page(skb_shinfo(skb
)->frags
[0].page
);
3343 memmove(skb_shinfo(skb
)->frags
,
3344 skb_shinfo(skb
)->frags
+ 1,
3345 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3356 EXPORT_SYMBOL(dev_gro_receive
);
3358 static inline gro_result_t
3359 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3363 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3364 unsigned long diffs
;
3366 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3367 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3368 diffs
|= compare_ether_header(skb_mac_header(p
),
3369 skb_gro_mac_header(skb
));
3370 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3371 NAPI_GRO_CB(p
)->flush
= 0;
3374 return dev_gro_receive(napi
, skb
);
3377 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3381 if (netif_receive_skb(skb
))
3386 case GRO_MERGED_FREE
:
3397 EXPORT_SYMBOL(napi_skb_finish
);
3399 void skb_gro_reset_offset(struct sk_buff
*skb
)
3401 NAPI_GRO_CB(skb
)->data_offset
= 0;
3402 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3403 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3405 if (skb
->mac_header
== skb
->tail
&&
3406 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3407 NAPI_GRO_CB(skb
)->frag0
=
3408 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3409 skb_shinfo(skb
)->frags
[0].page_offset
;
3410 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3413 EXPORT_SYMBOL(skb_gro_reset_offset
);
3415 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3417 skb_gro_reset_offset(skb
);
3419 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3421 EXPORT_SYMBOL(napi_gro_receive
);
3423 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3425 __skb_pull(skb
, skb_headlen(skb
));
3426 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3428 skb
->dev
= napi
->dev
;
3434 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3436 struct sk_buff
*skb
= napi
->skb
;
3439 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3445 EXPORT_SYMBOL(napi_get_frags
);
3447 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3453 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3455 if (ret
== GRO_HELD
)
3456 skb_gro_pull(skb
, -ETH_HLEN
);
3457 else if (netif_receive_skb(skb
))
3462 case GRO_MERGED_FREE
:
3463 napi_reuse_skb(napi
, skb
);
3472 EXPORT_SYMBOL(napi_frags_finish
);
3474 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3476 struct sk_buff
*skb
= napi
->skb
;
3483 skb_reset_mac_header(skb
);
3484 skb_gro_reset_offset(skb
);
3486 off
= skb_gro_offset(skb
);
3487 hlen
= off
+ sizeof(*eth
);
3488 eth
= skb_gro_header_fast(skb
, off
);
3489 if (skb_gro_header_hard(skb
, hlen
)) {
3490 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3491 if (unlikely(!eth
)) {
3492 napi_reuse_skb(napi
, skb
);
3498 skb_gro_pull(skb
, sizeof(*eth
));
3501 * This works because the only protocols we care about don't require
3502 * special handling. We'll fix it up properly at the end.
3504 skb
->protocol
= eth
->h_proto
;
3509 EXPORT_SYMBOL(napi_frags_skb
);
3511 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3513 struct sk_buff
*skb
= napi_frags_skb(napi
);
3518 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3520 EXPORT_SYMBOL(napi_gro_frags
);
3523 * net_rps_action sends any pending IPI's for rps.
3524 * Note: called with local irq disabled, but exits with local irq enabled.
3526 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3529 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3532 sd
->rps_ipi_list
= NULL
;
3536 /* Send pending IPI's to kick RPS processing on remote cpus. */
3538 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3540 if (cpu_online(remsd
->cpu
))
3541 __smp_call_function_single(remsd
->cpu
,
3550 static int process_backlog(struct napi_struct
*napi
, int quota
)
3553 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3556 /* Check if we have pending ipi, its better to send them now,
3557 * not waiting net_rx_action() end.
3559 if (sd
->rps_ipi_list
) {
3560 local_irq_disable();
3561 net_rps_action_and_irq_enable(sd
);
3564 napi
->weight
= weight_p
;
3565 local_irq_disable();
3566 while (work
< quota
) {
3567 struct sk_buff
*skb
;
3570 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3572 __netif_receive_skb(skb
);
3573 local_irq_disable();
3574 input_queue_head_incr(sd
);
3575 if (++work
>= quota
) {
3582 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3584 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3585 &sd
->process_queue
);
3587 if (qlen
< quota
- work
) {
3589 * Inline a custom version of __napi_complete().
3590 * only current cpu owns and manipulates this napi,
3591 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3592 * we can use a plain write instead of clear_bit(),
3593 * and we dont need an smp_mb() memory barrier.
3595 list_del(&napi
->poll_list
);
3598 quota
= work
+ qlen
;
3608 * __napi_schedule - schedule for receive
3609 * @n: entry to schedule
3611 * The entry's receive function will be scheduled to run
3613 void __napi_schedule(struct napi_struct
*n
)
3615 unsigned long flags
;
3617 local_irq_save(flags
);
3618 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3619 local_irq_restore(flags
);
3621 EXPORT_SYMBOL(__napi_schedule
);
3623 void __napi_complete(struct napi_struct
*n
)
3625 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3626 BUG_ON(n
->gro_list
);
3628 list_del(&n
->poll_list
);
3629 smp_mb__before_clear_bit();
3630 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3632 EXPORT_SYMBOL(__napi_complete
);
3634 void napi_complete(struct napi_struct
*n
)
3636 unsigned long flags
;
3639 * don't let napi dequeue from the cpu poll list
3640 * just in case its running on a different cpu
3642 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3646 local_irq_save(flags
);
3648 local_irq_restore(flags
);
3650 EXPORT_SYMBOL(napi_complete
);
3652 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3653 int (*poll
)(struct napi_struct
*, int), int weight
)
3655 INIT_LIST_HEAD(&napi
->poll_list
);
3656 napi
->gro_count
= 0;
3657 napi
->gro_list
= NULL
;
3660 napi
->weight
= weight
;
3661 list_add(&napi
->dev_list
, &dev
->napi_list
);
3663 #ifdef CONFIG_NETPOLL
3664 spin_lock_init(&napi
->poll_lock
);
3665 napi
->poll_owner
= -1;
3667 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3669 EXPORT_SYMBOL(netif_napi_add
);
3671 void netif_napi_del(struct napi_struct
*napi
)
3673 struct sk_buff
*skb
, *next
;
3675 list_del_init(&napi
->dev_list
);
3676 napi_free_frags(napi
);
3678 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3684 napi
->gro_list
= NULL
;
3685 napi
->gro_count
= 0;
3687 EXPORT_SYMBOL(netif_napi_del
);
3689 static void net_rx_action(struct softirq_action
*h
)
3691 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3692 unsigned long time_limit
= jiffies
+ 2;
3693 int budget
= netdev_budget
;
3696 local_irq_disable();
3698 while (!list_empty(&sd
->poll_list
)) {
3699 struct napi_struct
*n
;
3702 /* If softirq window is exhuasted then punt.
3703 * Allow this to run for 2 jiffies since which will allow
3704 * an average latency of 1.5/HZ.
3706 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3711 /* Even though interrupts have been re-enabled, this
3712 * access is safe because interrupts can only add new
3713 * entries to the tail of this list, and only ->poll()
3714 * calls can remove this head entry from the list.
3716 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3718 have
= netpoll_poll_lock(n
);
3722 /* This NAPI_STATE_SCHED test is for avoiding a race
3723 * with netpoll's poll_napi(). Only the entity which
3724 * obtains the lock and sees NAPI_STATE_SCHED set will
3725 * actually make the ->poll() call. Therefore we avoid
3726 * accidently calling ->poll() when NAPI is not scheduled.
3729 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3730 work
= n
->poll(n
, weight
);
3734 WARN_ON_ONCE(work
> weight
);
3738 local_irq_disable();
3740 /* Drivers must not modify the NAPI state if they
3741 * consume the entire weight. In such cases this code
3742 * still "owns" the NAPI instance and therefore can
3743 * move the instance around on the list at-will.
3745 if (unlikely(work
== weight
)) {
3746 if (unlikely(napi_disable_pending(n
))) {
3749 local_irq_disable();
3751 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3754 netpoll_poll_unlock(have
);
3757 net_rps_action_and_irq_enable(sd
);
3759 #ifdef CONFIG_NET_DMA
3761 * There may not be any more sk_buffs coming right now, so push
3762 * any pending DMA copies to hardware
3764 dma_issue_pending_all();
3771 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3775 static gifconf_func_t
*gifconf_list
[NPROTO
];
3778 * register_gifconf - register a SIOCGIF handler
3779 * @family: Address family
3780 * @gifconf: Function handler
3782 * Register protocol dependent address dumping routines. The handler
3783 * that is passed must not be freed or reused until it has been replaced
3784 * by another handler.
3786 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3788 if (family
>= NPROTO
)
3790 gifconf_list
[family
] = gifconf
;
3793 EXPORT_SYMBOL(register_gifconf
);
3797 * Map an interface index to its name (SIOCGIFNAME)
3801 * We need this ioctl for efficient implementation of the
3802 * if_indextoname() function required by the IPv6 API. Without
3803 * it, we would have to search all the interfaces to find a
3807 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3809 struct net_device
*dev
;
3813 * Fetch the caller's info block.
3816 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3820 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3826 strcpy(ifr
.ifr_name
, dev
->name
);
3829 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3835 * Perform a SIOCGIFCONF call. This structure will change
3836 * size eventually, and there is nothing I can do about it.
3837 * Thus we will need a 'compatibility mode'.
3840 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3843 struct net_device
*dev
;
3850 * Fetch the caller's info block.
3853 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3860 * Loop over the interfaces, and write an info block for each.
3864 for_each_netdev(net
, dev
) {
3865 for (i
= 0; i
< NPROTO
; i
++) {
3866 if (gifconf_list
[i
]) {
3869 done
= gifconf_list
[i
](dev
, NULL
, 0);
3871 done
= gifconf_list
[i
](dev
, pos
+ total
,
3881 * All done. Write the updated control block back to the caller.
3883 ifc
.ifc_len
= total
;
3886 * Both BSD and Solaris return 0 here, so we do too.
3888 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3891 #ifdef CONFIG_PROC_FS
3893 * This is invoked by the /proc filesystem handler to display a device
3896 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3899 struct net
*net
= seq_file_net(seq
);
3901 struct net_device
*dev
;
3905 return SEQ_START_TOKEN
;
3908 for_each_netdev_rcu(net
, dev
)
3915 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3917 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3918 first_net_device(seq_file_net(seq
)) :
3919 next_net_device((struct net_device
*)v
);
3922 return rcu_dereference(dev
);
3925 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3931 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3933 struct rtnl_link_stats64 temp
;
3934 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
3936 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3937 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3938 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3940 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3941 stats
->rx_fifo_errors
,
3942 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3943 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3944 stats
->rx_compressed
, stats
->multicast
,
3945 stats
->tx_bytes
, stats
->tx_packets
,
3946 stats
->tx_errors
, stats
->tx_dropped
,
3947 stats
->tx_fifo_errors
, stats
->collisions
,
3948 stats
->tx_carrier_errors
+
3949 stats
->tx_aborted_errors
+
3950 stats
->tx_window_errors
+
3951 stats
->tx_heartbeat_errors
,
3952 stats
->tx_compressed
);
3956 * Called from the PROCfs module. This now uses the new arbitrary sized
3957 * /proc/net interface to create /proc/net/dev
3959 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3961 if (v
== SEQ_START_TOKEN
)
3962 seq_puts(seq
, "Inter-| Receive "
3964 " face |bytes packets errs drop fifo frame "
3965 "compressed multicast|bytes packets errs "
3966 "drop fifo colls carrier compressed\n");
3968 dev_seq_printf_stats(seq
, v
);
3972 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
3974 struct softnet_data
*sd
= NULL
;
3976 while (*pos
< nr_cpu_ids
)
3977 if (cpu_online(*pos
)) {
3978 sd
= &per_cpu(softnet_data
, *pos
);
3985 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3987 return softnet_get_online(pos
);
3990 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3993 return softnet_get_online(pos
);
3996 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4000 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4002 struct softnet_data
*sd
= v
;
4004 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4005 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4006 0, 0, 0, 0, /* was fastroute */
4007 sd
->cpu_collision
, sd
->received_rps
);
4011 static const struct seq_operations dev_seq_ops
= {
4012 .start
= dev_seq_start
,
4013 .next
= dev_seq_next
,
4014 .stop
= dev_seq_stop
,
4015 .show
= dev_seq_show
,
4018 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4020 return seq_open_net(inode
, file
, &dev_seq_ops
,
4021 sizeof(struct seq_net_private
));
4024 static const struct file_operations dev_seq_fops
= {
4025 .owner
= THIS_MODULE
,
4026 .open
= dev_seq_open
,
4028 .llseek
= seq_lseek
,
4029 .release
= seq_release_net
,
4032 static const struct seq_operations softnet_seq_ops
= {
4033 .start
= softnet_seq_start
,
4034 .next
= softnet_seq_next
,
4035 .stop
= softnet_seq_stop
,
4036 .show
= softnet_seq_show
,
4039 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4041 return seq_open(file
, &softnet_seq_ops
);
4044 static const struct file_operations softnet_seq_fops
= {
4045 .owner
= THIS_MODULE
,
4046 .open
= softnet_seq_open
,
4048 .llseek
= seq_lseek
,
4049 .release
= seq_release
,
4052 static void *ptype_get_idx(loff_t pos
)
4054 struct packet_type
*pt
= NULL
;
4058 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4064 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4065 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4074 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4078 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4081 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4083 struct packet_type
*pt
;
4084 struct list_head
*nxt
;
4088 if (v
== SEQ_START_TOKEN
)
4089 return ptype_get_idx(0);
4092 nxt
= pt
->list
.next
;
4093 if (pt
->type
== htons(ETH_P_ALL
)) {
4094 if (nxt
!= &ptype_all
)
4097 nxt
= ptype_base
[0].next
;
4099 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4101 while (nxt
== &ptype_base
[hash
]) {
4102 if (++hash
>= PTYPE_HASH_SIZE
)
4104 nxt
= ptype_base
[hash
].next
;
4107 return list_entry(nxt
, struct packet_type
, list
);
4110 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4116 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4118 struct packet_type
*pt
= v
;
4120 if (v
== SEQ_START_TOKEN
)
4121 seq_puts(seq
, "Type Device Function\n");
4122 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4123 if (pt
->type
== htons(ETH_P_ALL
))
4124 seq_puts(seq
, "ALL ");
4126 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4128 seq_printf(seq
, " %-8s %pF\n",
4129 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4135 static const struct seq_operations ptype_seq_ops
= {
4136 .start
= ptype_seq_start
,
4137 .next
= ptype_seq_next
,
4138 .stop
= ptype_seq_stop
,
4139 .show
= ptype_seq_show
,
4142 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4144 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4145 sizeof(struct seq_net_private
));
4148 static const struct file_operations ptype_seq_fops
= {
4149 .owner
= THIS_MODULE
,
4150 .open
= ptype_seq_open
,
4152 .llseek
= seq_lseek
,
4153 .release
= seq_release_net
,
4157 static int __net_init
dev_proc_net_init(struct net
*net
)
4161 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4163 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4165 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4168 if (wext_proc_init(net
))
4174 proc_net_remove(net
, "ptype");
4176 proc_net_remove(net
, "softnet_stat");
4178 proc_net_remove(net
, "dev");
4182 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4184 wext_proc_exit(net
);
4186 proc_net_remove(net
, "ptype");
4187 proc_net_remove(net
, "softnet_stat");
4188 proc_net_remove(net
, "dev");
4191 static struct pernet_operations __net_initdata dev_proc_ops
= {
4192 .init
= dev_proc_net_init
,
4193 .exit
= dev_proc_net_exit
,
4196 static int __init
dev_proc_init(void)
4198 return register_pernet_subsys(&dev_proc_ops
);
4201 #define dev_proc_init() 0
4202 #endif /* CONFIG_PROC_FS */
4206 * netdev_set_master - set up master/slave pair
4207 * @slave: slave device
4208 * @master: new master device
4210 * Changes the master device of the slave. Pass %NULL to break the
4211 * bonding. The caller must hold the RTNL semaphore. On a failure
4212 * a negative errno code is returned. On success the reference counts
4213 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4214 * function returns zero.
4216 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4218 struct net_device
*old
= slave
->master
;
4228 slave
->master
= master
;
4235 slave
->flags
|= IFF_SLAVE
;
4237 slave
->flags
&= ~IFF_SLAVE
;
4239 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4242 EXPORT_SYMBOL(netdev_set_master
);
4244 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4246 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4248 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4249 ops
->ndo_change_rx_flags(dev
, flags
);
4252 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4254 unsigned short old_flags
= dev
->flags
;
4260 dev
->flags
|= IFF_PROMISC
;
4261 dev
->promiscuity
+= inc
;
4262 if (dev
->promiscuity
== 0) {
4265 * If inc causes overflow, untouch promisc and return error.
4268 dev
->flags
&= ~IFF_PROMISC
;
4270 dev
->promiscuity
-= inc
;
4271 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4272 "set promiscuity failed, promiscuity feature "
4273 "of device might be broken.\n", dev
->name
);
4277 if (dev
->flags
!= old_flags
) {
4278 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4279 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4281 if (audit_enabled
) {
4282 current_uid_gid(&uid
, &gid
);
4283 audit_log(current
->audit_context
, GFP_ATOMIC
,
4284 AUDIT_ANOM_PROMISCUOUS
,
4285 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4286 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4287 (old_flags
& IFF_PROMISC
),
4288 audit_get_loginuid(current
),
4290 audit_get_sessionid(current
));
4293 dev_change_rx_flags(dev
, IFF_PROMISC
);
4299 * dev_set_promiscuity - update promiscuity count on a device
4303 * Add or remove promiscuity from a device. While the count in the device
4304 * remains above zero the interface remains promiscuous. Once it hits zero
4305 * the device reverts back to normal filtering operation. A negative inc
4306 * value is used to drop promiscuity on the device.
4307 * Return 0 if successful or a negative errno code on error.
4309 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4311 unsigned short old_flags
= dev
->flags
;
4314 err
= __dev_set_promiscuity(dev
, inc
);
4317 if (dev
->flags
!= old_flags
)
4318 dev_set_rx_mode(dev
);
4321 EXPORT_SYMBOL(dev_set_promiscuity
);
4324 * dev_set_allmulti - update allmulti count on a device
4328 * Add or remove reception of all multicast frames to a device. While the
4329 * count in the device remains above zero the interface remains listening
4330 * to all interfaces. Once it hits zero the device reverts back to normal
4331 * filtering operation. A negative @inc value is used to drop the counter
4332 * when releasing a resource needing all multicasts.
4333 * Return 0 if successful or a negative errno code on error.
4336 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4338 unsigned short old_flags
= dev
->flags
;
4342 dev
->flags
|= IFF_ALLMULTI
;
4343 dev
->allmulti
+= inc
;
4344 if (dev
->allmulti
== 0) {
4347 * If inc causes overflow, untouch allmulti and return error.
4350 dev
->flags
&= ~IFF_ALLMULTI
;
4352 dev
->allmulti
-= inc
;
4353 printk(KERN_WARNING
"%s: allmulti touches roof, "
4354 "set allmulti failed, allmulti feature of "
4355 "device might be broken.\n", dev
->name
);
4359 if (dev
->flags
^ old_flags
) {
4360 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4361 dev_set_rx_mode(dev
);
4365 EXPORT_SYMBOL(dev_set_allmulti
);
4368 * Upload unicast and multicast address lists to device and
4369 * configure RX filtering. When the device doesn't support unicast
4370 * filtering it is put in promiscuous mode while unicast addresses
4373 void __dev_set_rx_mode(struct net_device
*dev
)
4375 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4377 /* dev_open will call this function so the list will stay sane. */
4378 if (!(dev
->flags
&IFF_UP
))
4381 if (!netif_device_present(dev
))
4384 if (ops
->ndo_set_rx_mode
)
4385 ops
->ndo_set_rx_mode(dev
);
4387 /* Unicast addresses changes may only happen under the rtnl,
4388 * therefore calling __dev_set_promiscuity here is safe.
4390 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4391 __dev_set_promiscuity(dev
, 1);
4392 dev
->uc_promisc
= 1;
4393 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4394 __dev_set_promiscuity(dev
, -1);
4395 dev
->uc_promisc
= 0;
4398 if (ops
->ndo_set_multicast_list
)
4399 ops
->ndo_set_multicast_list(dev
);
4403 void dev_set_rx_mode(struct net_device
*dev
)
4405 netif_addr_lock_bh(dev
);
4406 __dev_set_rx_mode(dev
);
4407 netif_addr_unlock_bh(dev
);
4411 * dev_get_flags - get flags reported to userspace
4414 * Get the combination of flag bits exported through APIs to userspace.
4416 unsigned dev_get_flags(const struct net_device
*dev
)
4420 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4425 (dev
->gflags
& (IFF_PROMISC
|
4428 if (netif_running(dev
)) {
4429 if (netif_oper_up(dev
))
4430 flags
|= IFF_RUNNING
;
4431 if (netif_carrier_ok(dev
))
4432 flags
|= IFF_LOWER_UP
;
4433 if (netif_dormant(dev
))
4434 flags
|= IFF_DORMANT
;
4439 EXPORT_SYMBOL(dev_get_flags
);
4441 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4443 int old_flags
= dev
->flags
;
4449 * Set the flags on our device.
4452 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4453 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4455 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4459 * Load in the correct multicast list now the flags have changed.
4462 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4463 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4465 dev_set_rx_mode(dev
);
4468 * Have we downed the interface. We handle IFF_UP ourselves
4469 * according to user attempts to set it, rather than blindly
4474 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4475 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4478 dev_set_rx_mode(dev
);
4481 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4482 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4484 dev
->gflags
^= IFF_PROMISC
;
4485 dev_set_promiscuity(dev
, inc
);
4488 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4489 is important. Some (broken) drivers set IFF_PROMISC, when
4490 IFF_ALLMULTI is requested not asking us and not reporting.
4492 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4493 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4495 dev
->gflags
^= IFF_ALLMULTI
;
4496 dev_set_allmulti(dev
, inc
);
4502 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4504 unsigned int changes
= dev
->flags
^ old_flags
;
4506 if (changes
& IFF_UP
) {
4507 if (dev
->flags
& IFF_UP
)
4508 call_netdevice_notifiers(NETDEV_UP
, dev
);
4510 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4513 if (dev
->flags
& IFF_UP
&&
4514 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4515 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4519 * dev_change_flags - change device settings
4521 * @flags: device state flags
4523 * Change settings on device based state flags. The flags are
4524 * in the userspace exported format.
4526 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4529 int old_flags
= dev
->flags
;
4531 ret
= __dev_change_flags(dev
, flags
);
4535 changes
= old_flags
^ dev
->flags
;
4537 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4539 __dev_notify_flags(dev
, old_flags
);
4542 EXPORT_SYMBOL(dev_change_flags
);
4545 * dev_set_mtu - Change maximum transfer unit
4547 * @new_mtu: new transfer unit
4549 * Change the maximum transfer size of the network device.
4551 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4553 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4556 if (new_mtu
== dev
->mtu
)
4559 /* MTU must be positive. */
4563 if (!netif_device_present(dev
))
4567 if (ops
->ndo_change_mtu
)
4568 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4572 if (!err
&& dev
->flags
& IFF_UP
)
4573 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4576 EXPORT_SYMBOL(dev_set_mtu
);
4579 * dev_set_mac_address - Change Media Access Control Address
4583 * Change the hardware (MAC) address of the device
4585 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4587 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4590 if (!ops
->ndo_set_mac_address
)
4592 if (sa
->sa_family
!= dev
->type
)
4594 if (!netif_device_present(dev
))
4596 err
= ops
->ndo_set_mac_address(dev
, sa
);
4598 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4601 EXPORT_SYMBOL(dev_set_mac_address
);
4604 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4606 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4609 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4615 case SIOCGIFFLAGS
: /* Get interface flags */
4616 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4619 case SIOCGIFMETRIC
: /* Get the metric on the interface
4620 (currently unused) */
4621 ifr
->ifr_metric
= 0;
4624 case SIOCGIFMTU
: /* Get the MTU of a device */
4625 ifr
->ifr_mtu
= dev
->mtu
;
4630 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4632 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4633 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4634 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4642 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4643 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4644 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4645 ifr
->ifr_map
.irq
= dev
->irq
;
4646 ifr
->ifr_map
.dma
= dev
->dma
;
4647 ifr
->ifr_map
.port
= dev
->if_port
;
4651 ifr
->ifr_ifindex
= dev
->ifindex
;
4655 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4659 /* dev_ioctl() should ensure this case
4671 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4673 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4676 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4677 const struct net_device_ops
*ops
;
4682 ops
= dev
->netdev_ops
;
4685 case SIOCSIFFLAGS
: /* Set interface flags */
4686 return dev_change_flags(dev
, ifr
->ifr_flags
);
4688 case SIOCSIFMETRIC
: /* Set the metric on the interface
4689 (currently unused) */
4692 case SIOCSIFMTU
: /* Set the MTU of a device */
4693 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4696 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4698 case SIOCSIFHWBROADCAST
:
4699 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4701 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4702 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4703 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4707 if (ops
->ndo_set_config
) {
4708 if (!netif_device_present(dev
))
4710 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4715 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4716 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4718 if (!netif_device_present(dev
))
4720 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4723 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4724 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4726 if (!netif_device_present(dev
))
4728 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4731 if (ifr
->ifr_qlen
< 0)
4733 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4737 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4738 return dev_change_name(dev
, ifr
->ifr_newname
);
4741 * Unknown or private ioctl
4744 if ((cmd
>= SIOCDEVPRIVATE
&&
4745 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4746 cmd
== SIOCBONDENSLAVE
||
4747 cmd
== SIOCBONDRELEASE
||
4748 cmd
== SIOCBONDSETHWADDR
||
4749 cmd
== SIOCBONDSLAVEINFOQUERY
||
4750 cmd
== SIOCBONDINFOQUERY
||
4751 cmd
== SIOCBONDCHANGEACTIVE
||
4752 cmd
== SIOCGMIIPHY
||
4753 cmd
== SIOCGMIIREG
||
4754 cmd
== SIOCSMIIREG
||
4755 cmd
== SIOCBRADDIF
||
4756 cmd
== SIOCBRDELIF
||
4757 cmd
== SIOCSHWTSTAMP
||
4758 cmd
== SIOCWANDEV
) {
4760 if (ops
->ndo_do_ioctl
) {
4761 if (netif_device_present(dev
))
4762 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4774 * This function handles all "interface"-type I/O control requests. The actual
4775 * 'doing' part of this is dev_ifsioc above.
4779 * dev_ioctl - network device ioctl
4780 * @net: the applicable net namespace
4781 * @cmd: command to issue
4782 * @arg: pointer to a struct ifreq in user space
4784 * Issue ioctl functions to devices. This is normally called by the
4785 * user space syscall interfaces but can sometimes be useful for
4786 * other purposes. The return value is the return from the syscall if
4787 * positive or a negative errno code on error.
4790 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4796 /* One special case: SIOCGIFCONF takes ifconf argument
4797 and requires shared lock, because it sleeps writing
4801 if (cmd
== SIOCGIFCONF
) {
4803 ret
= dev_ifconf(net
, (char __user
*) arg
);
4807 if (cmd
== SIOCGIFNAME
)
4808 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4810 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4813 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4815 colon
= strchr(ifr
.ifr_name
, ':');
4820 * See which interface the caller is talking about.
4825 * These ioctl calls:
4826 * - can be done by all.
4827 * - atomic and do not require locking.
4838 dev_load(net
, ifr
.ifr_name
);
4840 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4845 if (copy_to_user(arg
, &ifr
,
4846 sizeof(struct ifreq
)))
4852 dev_load(net
, ifr
.ifr_name
);
4854 ret
= dev_ethtool(net
, &ifr
);
4859 if (copy_to_user(arg
, &ifr
,
4860 sizeof(struct ifreq
)))
4866 * These ioctl calls:
4867 * - require superuser power.
4868 * - require strict serialization.
4874 if (!capable(CAP_NET_ADMIN
))
4876 dev_load(net
, ifr
.ifr_name
);
4878 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4883 if (copy_to_user(arg
, &ifr
,
4884 sizeof(struct ifreq
)))
4890 * These ioctl calls:
4891 * - require superuser power.
4892 * - require strict serialization.
4893 * - do not return a value
4903 case SIOCSIFHWBROADCAST
:
4906 case SIOCBONDENSLAVE
:
4907 case SIOCBONDRELEASE
:
4908 case SIOCBONDSETHWADDR
:
4909 case SIOCBONDCHANGEACTIVE
:
4913 if (!capable(CAP_NET_ADMIN
))
4916 case SIOCBONDSLAVEINFOQUERY
:
4917 case SIOCBONDINFOQUERY
:
4918 dev_load(net
, ifr
.ifr_name
);
4920 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4925 /* Get the per device memory space. We can add this but
4926 * currently do not support it */
4928 /* Set the per device memory buffer space.
4929 * Not applicable in our case */
4934 * Unknown or private ioctl.
4937 if (cmd
== SIOCWANDEV
||
4938 (cmd
>= SIOCDEVPRIVATE
&&
4939 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4940 dev_load(net
, ifr
.ifr_name
);
4942 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4944 if (!ret
&& copy_to_user(arg
, &ifr
,
4945 sizeof(struct ifreq
)))
4949 /* Take care of Wireless Extensions */
4950 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4951 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4958 * dev_new_index - allocate an ifindex
4959 * @net: the applicable net namespace
4961 * Returns a suitable unique value for a new device interface
4962 * number. The caller must hold the rtnl semaphore or the
4963 * dev_base_lock to be sure it remains unique.
4965 static int dev_new_index(struct net
*net
)
4971 if (!__dev_get_by_index(net
, ifindex
))
4976 /* Delayed registration/unregisteration */
4977 static LIST_HEAD(net_todo_list
);
4979 static void net_set_todo(struct net_device
*dev
)
4981 list_add_tail(&dev
->todo_list
, &net_todo_list
);
4984 static void rollback_registered_many(struct list_head
*head
)
4986 struct net_device
*dev
, *tmp
;
4988 BUG_ON(dev_boot_phase
);
4991 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
4992 /* Some devices call without registering
4993 * for initialization unwind. Remove those
4994 * devices and proceed with the remaining.
4996 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4997 pr_debug("unregister_netdevice: device %s/%p never "
4998 "was registered\n", dev
->name
, dev
);
5001 list_del(&dev
->unreg_list
);
5005 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5008 /* If device is running, close it first. */
5009 dev_close_many(head
);
5011 list_for_each_entry(dev
, head
, unreg_list
) {
5012 /* And unlink it from device chain. */
5013 unlist_netdevice(dev
);
5015 dev
->reg_state
= NETREG_UNREGISTERING
;
5020 list_for_each_entry(dev
, head
, unreg_list
) {
5021 /* Shutdown queueing discipline. */
5025 /* Notify protocols, that we are about to destroy
5026 this device. They should clean all the things.
5028 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5030 if (!dev
->rtnl_link_ops
||
5031 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5032 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5035 * Flush the unicast and multicast chains
5040 if (dev
->netdev_ops
->ndo_uninit
)
5041 dev
->netdev_ops
->ndo_uninit(dev
);
5043 /* Notifier chain MUST detach us from master device. */
5044 WARN_ON(dev
->master
);
5046 /* Remove entries from kobject tree */
5047 netdev_unregister_kobject(dev
);
5050 /* Process any work delayed until the end of the batch */
5051 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5052 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5056 list_for_each_entry(dev
, head
, unreg_list
)
5060 static void rollback_registered(struct net_device
*dev
)
5064 list_add(&dev
->unreg_list
, &single
);
5065 rollback_registered_many(&single
);
5068 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
5070 /* Fix illegal SG+CSUM combinations. */
5071 if ((features
& NETIF_F_SG
) &&
5072 !(features
& NETIF_F_ALL_CSUM
)) {
5074 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
5075 "checksum feature.\n", name
);
5076 features
&= ~NETIF_F_SG
;
5079 /* TSO requires that SG is present as well. */
5080 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
5082 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
5083 "SG feature.\n", name
);
5084 features
&= ~NETIF_F_TSO
;
5087 if (features
& NETIF_F_UFO
) {
5088 /* maybe split UFO into V4 and V6? */
5089 if (!((features
& NETIF_F_GEN_CSUM
) ||
5090 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5091 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5093 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
5094 "since no checksum offload features.\n",
5096 features
&= ~NETIF_F_UFO
;
5099 if (!(features
& NETIF_F_SG
)) {
5101 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
5102 "since no NETIF_F_SG feature.\n", name
);
5103 features
&= ~NETIF_F_UFO
;
5109 EXPORT_SYMBOL(netdev_fix_features
);
5112 * netif_stacked_transfer_operstate - transfer operstate
5113 * @rootdev: the root or lower level device to transfer state from
5114 * @dev: the device to transfer operstate to
5116 * Transfer operational state from root to device. This is normally
5117 * called when a stacking relationship exists between the root
5118 * device and the device(a leaf device).
5120 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5121 struct net_device
*dev
)
5123 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5124 netif_dormant_on(dev
);
5126 netif_dormant_off(dev
);
5128 if (netif_carrier_ok(rootdev
)) {
5129 if (!netif_carrier_ok(dev
))
5130 netif_carrier_on(dev
);
5132 if (netif_carrier_ok(dev
))
5133 netif_carrier_off(dev
);
5136 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5139 static int netif_alloc_rx_queues(struct net_device
*dev
)
5141 unsigned int i
, count
= dev
->num_rx_queues
;
5142 struct netdev_rx_queue
*rx
;
5146 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5148 pr_err("netdev: Unable to allocate %u rx queues.\n", count
);
5153 for (i
= 0; i
< count
; i
++)
5159 static void netdev_init_one_queue(struct net_device
*dev
,
5160 struct netdev_queue
*queue
, void *_unused
)
5162 /* Initialize queue lock */
5163 spin_lock_init(&queue
->_xmit_lock
);
5164 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5165 queue
->xmit_lock_owner
= -1;
5166 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5170 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5172 unsigned int count
= dev
->num_tx_queues
;
5173 struct netdev_queue
*tx
;
5177 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5179 pr_err("netdev: Unable to allocate %u tx queues.\n",
5185 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5186 spin_lock_init(&dev
->tx_global_lock
);
5192 * register_netdevice - register a network device
5193 * @dev: device to register
5195 * Take a completed network device structure and add it to the kernel
5196 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5197 * chain. 0 is returned on success. A negative errno code is returned
5198 * on a failure to set up the device, or if the name is a duplicate.
5200 * Callers must hold the rtnl semaphore. You may want
5201 * register_netdev() instead of this.
5204 * The locking appears insufficient to guarantee two parallel registers
5205 * will not get the same name.
5208 int register_netdevice(struct net_device
*dev
)
5211 struct net
*net
= dev_net(dev
);
5213 BUG_ON(dev_boot_phase
);
5218 /* When net_device's are persistent, this will be fatal. */
5219 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5222 spin_lock_init(&dev
->addr_list_lock
);
5223 netdev_set_addr_lockdep_class(dev
);
5227 /* Init, if this function is available */
5228 if (dev
->netdev_ops
->ndo_init
) {
5229 ret
= dev
->netdev_ops
->ndo_init(dev
);
5237 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
5241 dev
->ifindex
= dev_new_index(net
);
5242 if (dev
->iflink
== -1)
5243 dev
->iflink
= dev
->ifindex
;
5245 /* Fix illegal checksum combinations */
5246 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
5247 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5248 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5250 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5253 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5254 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5255 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5257 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5260 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5262 /* Enable software GSO if SG is supported. */
5263 if (dev
->features
& NETIF_F_SG
)
5264 dev
->features
|= NETIF_F_GSO
;
5266 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5267 * vlan_dev_init() will do the dev->features check, so these features
5268 * are enabled only if supported by underlying device.
5270 dev
->vlan_features
|= (NETIF_F_GRO
| NETIF_F_HIGHDMA
);
5272 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5273 ret
= notifier_to_errno(ret
);
5277 ret
= netdev_register_kobject(dev
);
5280 dev
->reg_state
= NETREG_REGISTERED
;
5283 * Default initial state at registry is that the
5284 * device is present.
5287 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5289 dev_init_scheduler(dev
);
5291 list_netdevice(dev
);
5293 /* Notify protocols, that a new device appeared. */
5294 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5295 ret
= notifier_to_errno(ret
);
5297 rollback_registered(dev
);
5298 dev
->reg_state
= NETREG_UNREGISTERED
;
5301 * Prevent userspace races by waiting until the network
5302 * device is fully setup before sending notifications.
5304 if (!dev
->rtnl_link_ops
||
5305 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5306 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5312 if (dev
->netdev_ops
->ndo_uninit
)
5313 dev
->netdev_ops
->ndo_uninit(dev
);
5316 EXPORT_SYMBOL(register_netdevice
);
5319 * init_dummy_netdev - init a dummy network device for NAPI
5320 * @dev: device to init
5322 * This takes a network device structure and initialize the minimum
5323 * amount of fields so it can be used to schedule NAPI polls without
5324 * registering a full blown interface. This is to be used by drivers
5325 * that need to tie several hardware interfaces to a single NAPI
5326 * poll scheduler due to HW limitations.
5328 int init_dummy_netdev(struct net_device
*dev
)
5330 /* Clear everything. Note we don't initialize spinlocks
5331 * are they aren't supposed to be taken by any of the
5332 * NAPI code and this dummy netdev is supposed to be
5333 * only ever used for NAPI polls
5335 memset(dev
, 0, sizeof(struct net_device
));
5337 /* make sure we BUG if trying to hit standard
5338 * register/unregister code path
5340 dev
->reg_state
= NETREG_DUMMY
;
5342 /* NAPI wants this */
5343 INIT_LIST_HEAD(&dev
->napi_list
);
5345 /* a dummy interface is started by default */
5346 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5347 set_bit(__LINK_STATE_START
, &dev
->state
);
5349 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5350 * because users of this 'device' dont need to change
5356 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5360 * register_netdev - register a network device
5361 * @dev: device to register
5363 * Take a completed network device structure and add it to the kernel
5364 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5365 * chain. 0 is returned on success. A negative errno code is returned
5366 * on a failure to set up the device, or if the name is a duplicate.
5368 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5369 * and expands the device name if you passed a format string to
5372 int register_netdev(struct net_device
*dev
)
5379 * If the name is a format string the caller wants us to do a
5382 if (strchr(dev
->name
, '%')) {
5383 err
= dev_alloc_name(dev
, dev
->name
);
5388 err
= register_netdevice(dev
);
5393 EXPORT_SYMBOL(register_netdev
);
5395 int netdev_refcnt_read(const struct net_device
*dev
)
5399 for_each_possible_cpu(i
)
5400 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5403 EXPORT_SYMBOL(netdev_refcnt_read
);
5406 * netdev_wait_allrefs - wait until all references are gone.
5408 * This is called when unregistering network devices.
5410 * Any protocol or device that holds a reference should register
5411 * for netdevice notification, and cleanup and put back the
5412 * reference if they receive an UNREGISTER event.
5413 * We can get stuck here if buggy protocols don't correctly
5416 static void netdev_wait_allrefs(struct net_device
*dev
)
5418 unsigned long rebroadcast_time
, warning_time
;
5421 linkwatch_forget_dev(dev
);
5423 rebroadcast_time
= warning_time
= jiffies
;
5424 refcnt
= netdev_refcnt_read(dev
);
5426 while (refcnt
!= 0) {
5427 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5430 /* Rebroadcast unregister notification */
5431 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5432 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5433 * should have already handle it the first time */
5435 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5437 /* We must not have linkwatch events
5438 * pending on unregister. If this
5439 * happens, we simply run the queue
5440 * unscheduled, resulting in a noop
5443 linkwatch_run_queue();
5448 rebroadcast_time
= jiffies
;
5453 refcnt
= netdev_refcnt_read(dev
);
5455 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5456 printk(KERN_EMERG
"unregister_netdevice: "
5457 "waiting for %s to become free. Usage "
5460 warning_time
= jiffies
;
5469 * register_netdevice(x1);
5470 * register_netdevice(x2);
5472 * unregister_netdevice(y1);
5473 * unregister_netdevice(y2);
5479 * We are invoked by rtnl_unlock().
5480 * This allows us to deal with problems:
5481 * 1) We can delete sysfs objects which invoke hotplug
5482 * without deadlocking with linkwatch via keventd.
5483 * 2) Since we run with the RTNL semaphore not held, we can sleep
5484 * safely in order to wait for the netdev refcnt to drop to zero.
5486 * We must not return until all unregister events added during
5487 * the interval the lock was held have been completed.
5489 void netdev_run_todo(void)
5491 struct list_head list
;
5493 /* Snapshot list, allow later requests */
5494 list_replace_init(&net_todo_list
, &list
);
5498 while (!list_empty(&list
)) {
5499 struct net_device
*dev
5500 = list_first_entry(&list
, struct net_device
, todo_list
);
5501 list_del(&dev
->todo_list
);
5503 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5504 printk(KERN_ERR
"network todo '%s' but state %d\n",
5505 dev
->name
, dev
->reg_state
);
5510 dev
->reg_state
= NETREG_UNREGISTERED
;
5512 on_each_cpu(flush_backlog
, dev
, 1);
5514 netdev_wait_allrefs(dev
);
5517 BUG_ON(netdev_refcnt_read(dev
));
5518 WARN_ON(rcu_dereference_raw(dev
->ip_ptr
));
5519 WARN_ON(rcu_dereference_raw(dev
->ip6_ptr
));
5520 WARN_ON(dev
->dn_ptr
);
5522 if (dev
->destructor
)
5523 dev
->destructor(dev
);
5525 /* Free network device */
5526 kobject_put(&dev
->dev
.kobj
);
5530 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5531 * fields in the same order, with only the type differing.
5533 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5534 const struct net_device_stats
*netdev_stats
)
5536 #if BITS_PER_LONG == 64
5537 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5538 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5540 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5541 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5542 u64
*dst
= (u64
*)stats64
;
5544 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5545 sizeof(*stats64
) / sizeof(u64
));
5546 for (i
= 0; i
< n
; i
++)
5552 * dev_get_stats - get network device statistics
5553 * @dev: device to get statistics from
5554 * @storage: place to store stats
5556 * Get network statistics from device. Return @storage.
5557 * The device driver may provide its own method by setting
5558 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5559 * otherwise the internal statistics structure is used.
5561 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5562 struct rtnl_link_stats64
*storage
)
5564 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5566 if (ops
->ndo_get_stats64
) {
5567 memset(storage
, 0, sizeof(*storage
));
5568 ops
->ndo_get_stats64(dev
, storage
);
5569 } else if (ops
->ndo_get_stats
) {
5570 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5572 netdev_stats_to_stats64(storage
, &dev
->stats
);
5574 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5577 EXPORT_SYMBOL(dev_get_stats
);
5579 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5581 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5583 #ifdef CONFIG_NET_CLS_ACT
5586 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5589 netdev_init_one_queue(dev
, queue
, NULL
);
5590 queue
->qdisc
= &noop_qdisc
;
5591 queue
->qdisc_sleeping
= &noop_qdisc
;
5592 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5598 * alloc_netdev_mqs - allocate network device
5599 * @sizeof_priv: size of private data to allocate space for
5600 * @name: device name format string
5601 * @setup: callback to initialize device
5602 * @txqs: the number of TX subqueues to allocate
5603 * @rxqs: the number of RX subqueues to allocate
5605 * Allocates a struct net_device with private data area for driver use
5606 * and performs basic initialization. Also allocates subquue structs
5607 * for each queue on the device.
5609 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5610 void (*setup
)(struct net_device
*),
5611 unsigned int txqs
, unsigned int rxqs
)
5613 struct net_device
*dev
;
5615 struct net_device
*p
;
5617 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5620 pr_err("alloc_netdev: Unable to allocate device "
5621 "with zero queues.\n");
5627 pr_err("alloc_netdev: Unable to allocate device "
5628 "with zero RX queues.\n");
5633 alloc_size
= sizeof(struct net_device
);
5635 /* ensure 32-byte alignment of private area */
5636 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5637 alloc_size
+= sizeof_priv
;
5639 /* ensure 32-byte alignment of whole construct */
5640 alloc_size
+= NETDEV_ALIGN
- 1;
5642 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5644 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5648 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5649 dev
->padded
= (char *)dev
- (char *)p
;
5651 dev
->pcpu_refcnt
= alloc_percpu(int);
5652 if (!dev
->pcpu_refcnt
)
5655 if (dev_addr_init(dev
))
5661 dev_net_set(dev
, &init_net
);
5663 dev
->gso_max_size
= GSO_MAX_SIZE
;
5665 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5666 dev
->ethtool_ntuple_list
.count
= 0;
5667 INIT_LIST_HEAD(&dev
->napi_list
);
5668 INIT_LIST_HEAD(&dev
->unreg_list
);
5669 INIT_LIST_HEAD(&dev
->link_watch_list
);
5670 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5673 dev
->num_tx_queues
= txqs
;
5674 dev
->real_num_tx_queues
= txqs
;
5675 if (netif_alloc_netdev_queues(dev
))
5679 dev
->num_rx_queues
= rxqs
;
5680 dev
->real_num_rx_queues
= rxqs
;
5681 if (netif_alloc_rx_queues(dev
))
5685 strcpy(dev
->name
, name
);
5693 free_percpu(dev
->pcpu_refcnt
);
5703 EXPORT_SYMBOL(alloc_netdev_mqs
);
5706 * free_netdev - free network device
5709 * This function does the last stage of destroying an allocated device
5710 * interface. The reference to the device object is released.
5711 * If this is the last reference then it will be freed.
5713 void free_netdev(struct net_device
*dev
)
5715 struct napi_struct
*p
, *n
;
5717 release_net(dev_net(dev
));
5724 kfree(rcu_dereference_raw(dev
->ingress_queue
));
5726 /* Flush device addresses */
5727 dev_addr_flush(dev
);
5729 /* Clear ethtool n-tuple list */
5730 ethtool_ntuple_flush(dev
);
5732 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5735 free_percpu(dev
->pcpu_refcnt
);
5736 dev
->pcpu_refcnt
= NULL
;
5738 /* Compatibility with error handling in drivers */
5739 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5740 kfree((char *)dev
- dev
->padded
);
5744 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5745 dev
->reg_state
= NETREG_RELEASED
;
5747 /* will free via device release */
5748 put_device(&dev
->dev
);
5750 EXPORT_SYMBOL(free_netdev
);
5753 * synchronize_net - Synchronize with packet receive processing
5755 * Wait for packets currently being received to be done.
5756 * Does not block later packets from starting.
5758 void synchronize_net(void)
5763 EXPORT_SYMBOL(synchronize_net
);
5766 * unregister_netdevice_queue - remove device from the kernel
5770 * This function shuts down a device interface and removes it
5771 * from the kernel tables.
5772 * If head not NULL, device is queued to be unregistered later.
5774 * Callers must hold the rtnl semaphore. You may want
5775 * unregister_netdev() instead of this.
5778 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5783 list_move_tail(&dev
->unreg_list
, head
);
5785 rollback_registered(dev
);
5786 /* Finish processing unregister after unlock */
5790 EXPORT_SYMBOL(unregister_netdevice_queue
);
5793 * unregister_netdevice_many - unregister many devices
5794 * @head: list of devices
5796 void unregister_netdevice_many(struct list_head
*head
)
5798 struct net_device
*dev
;
5800 if (!list_empty(head
)) {
5801 rollback_registered_many(head
);
5802 list_for_each_entry(dev
, head
, unreg_list
)
5806 EXPORT_SYMBOL(unregister_netdevice_many
);
5809 * unregister_netdev - remove device from the kernel
5812 * This function shuts down a device interface and removes it
5813 * from the kernel tables.
5815 * This is just a wrapper for unregister_netdevice that takes
5816 * the rtnl semaphore. In general you want to use this and not
5817 * unregister_netdevice.
5819 void unregister_netdev(struct net_device
*dev
)
5822 unregister_netdevice(dev
);
5825 EXPORT_SYMBOL(unregister_netdev
);
5828 * dev_change_net_namespace - move device to different nethost namespace
5830 * @net: network namespace
5831 * @pat: If not NULL name pattern to try if the current device name
5832 * is already taken in the destination network namespace.
5834 * This function shuts down a device interface and moves it
5835 * to a new network namespace. On success 0 is returned, on
5836 * a failure a netagive errno code is returned.
5838 * Callers must hold the rtnl semaphore.
5841 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5847 /* Don't allow namespace local devices to be moved. */
5849 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5852 /* Ensure the device has been registrered */
5854 if (dev
->reg_state
!= NETREG_REGISTERED
)
5857 /* Get out if there is nothing todo */
5859 if (net_eq(dev_net(dev
), net
))
5862 /* Pick the destination device name, and ensure
5863 * we can use it in the destination network namespace.
5866 if (__dev_get_by_name(net
, dev
->name
)) {
5867 /* We get here if we can't use the current device name */
5870 if (dev_get_valid_name(dev
, pat
, 1))
5875 * And now a mini version of register_netdevice unregister_netdevice.
5878 /* If device is running close it first. */
5881 /* And unlink it from device chain */
5883 unlist_netdevice(dev
);
5887 /* Shutdown queueing discipline. */
5890 /* Notify protocols, that we are about to destroy
5891 this device. They should clean all the things.
5893 Note that dev->reg_state stays at NETREG_REGISTERED.
5894 This is wanted because this way 8021q and macvlan know
5895 the device is just moving and can keep their slaves up.
5897 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5898 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5901 * Flush the unicast and multicast chains
5906 /* Actually switch the network namespace */
5907 dev_net_set(dev
, net
);
5909 /* If there is an ifindex conflict assign a new one */
5910 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5911 int iflink
= (dev
->iflink
== dev
->ifindex
);
5912 dev
->ifindex
= dev_new_index(net
);
5914 dev
->iflink
= dev
->ifindex
;
5917 /* Fixup kobjects */
5918 err
= device_rename(&dev
->dev
, dev
->name
);
5921 /* Add the device back in the hashes */
5922 list_netdevice(dev
);
5924 /* Notify protocols, that a new device appeared. */
5925 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5928 * Prevent userspace races by waiting until the network
5929 * device is fully setup before sending notifications.
5931 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5938 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5940 static int dev_cpu_callback(struct notifier_block
*nfb
,
5941 unsigned long action
,
5944 struct sk_buff
**list_skb
;
5945 struct sk_buff
*skb
;
5946 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5947 struct softnet_data
*sd
, *oldsd
;
5949 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5952 local_irq_disable();
5953 cpu
= smp_processor_id();
5954 sd
= &per_cpu(softnet_data
, cpu
);
5955 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5957 /* Find end of our completion_queue. */
5958 list_skb
= &sd
->completion_queue
;
5960 list_skb
= &(*list_skb
)->next
;
5961 /* Append completion queue from offline CPU. */
5962 *list_skb
= oldsd
->completion_queue
;
5963 oldsd
->completion_queue
= NULL
;
5965 /* Append output queue from offline CPU. */
5966 if (oldsd
->output_queue
) {
5967 *sd
->output_queue_tailp
= oldsd
->output_queue
;
5968 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
5969 oldsd
->output_queue
= NULL
;
5970 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
5973 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
5976 /* Process offline CPU's input_pkt_queue */
5977 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
5979 input_queue_head_incr(oldsd
);
5981 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
5983 input_queue_head_incr(oldsd
);
5991 * netdev_increment_features - increment feature set by one
5992 * @all: current feature set
5993 * @one: new feature set
5994 * @mask: mask feature set
5996 * Computes a new feature set after adding a device with feature set
5997 * @one to the master device with current feature set @all. Will not
5998 * enable anything that is off in @mask. Returns the new feature set.
6000 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
6003 /* If device needs checksumming, downgrade to it. */
6004 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
6005 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
6006 else if (mask
& NETIF_F_ALL_CSUM
) {
6007 /* If one device supports v4/v6 checksumming, set for all. */
6008 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
6009 !(all
& NETIF_F_GEN_CSUM
)) {
6010 all
&= ~NETIF_F_ALL_CSUM
;
6011 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
6014 /* If one device supports hw checksumming, set for all. */
6015 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
6016 all
&= ~NETIF_F_ALL_CSUM
;
6017 all
|= NETIF_F_HW_CSUM
;
6021 one
|= NETIF_F_ALL_CSUM
;
6023 one
|= all
& NETIF_F_ONE_FOR_ALL
;
6024 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
6025 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
6029 EXPORT_SYMBOL(netdev_increment_features
);
6031 static struct hlist_head
*netdev_create_hash(void)
6034 struct hlist_head
*hash
;
6036 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6038 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6039 INIT_HLIST_HEAD(&hash
[i
]);
6044 /* Initialize per network namespace state */
6045 static int __net_init
netdev_init(struct net
*net
)
6047 INIT_LIST_HEAD(&net
->dev_base_head
);
6049 net
->dev_name_head
= netdev_create_hash();
6050 if (net
->dev_name_head
== NULL
)
6053 net
->dev_index_head
= netdev_create_hash();
6054 if (net
->dev_index_head
== NULL
)
6060 kfree(net
->dev_name_head
);
6066 * netdev_drivername - network driver for the device
6067 * @dev: network device
6068 * @buffer: buffer for resulting name
6069 * @len: size of buffer
6071 * Determine network driver for device.
6073 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
6075 const struct device_driver
*driver
;
6076 const struct device
*parent
;
6078 if (len
<= 0 || !buffer
)
6082 parent
= dev
->dev
.parent
;
6087 driver
= parent
->driver
;
6088 if (driver
&& driver
->name
)
6089 strlcpy(buffer
, driver
->name
, len
);
6093 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6094 struct va_format
*vaf
)
6098 if (dev
&& dev
->dev
.parent
)
6099 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6100 netdev_name(dev
), vaf
);
6102 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6104 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6109 int netdev_printk(const char *level
, const struct net_device
*dev
,
6110 const char *format
, ...)
6112 struct va_format vaf
;
6116 va_start(args
, format
);
6121 r
= __netdev_printk(level
, dev
, &vaf
);
6126 EXPORT_SYMBOL(netdev_printk
);
6128 #define define_netdev_printk_level(func, level) \
6129 int func(const struct net_device *dev, const char *fmt, ...) \
6132 struct va_format vaf; \
6135 va_start(args, fmt); \
6140 r = __netdev_printk(level, dev, &vaf); \
6145 EXPORT_SYMBOL(func);
6147 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6148 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6149 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6150 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6151 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6152 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6153 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6155 static void __net_exit
netdev_exit(struct net
*net
)
6157 kfree(net
->dev_name_head
);
6158 kfree(net
->dev_index_head
);
6161 static struct pernet_operations __net_initdata netdev_net_ops
= {
6162 .init
= netdev_init
,
6163 .exit
= netdev_exit
,
6166 static void __net_exit
default_device_exit(struct net
*net
)
6168 struct net_device
*dev
, *aux
;
6170 * Push all migratable network devices back to the
6171 * initial network namespace
6174 for_each_netdev_safe(net
, dev
, aux
) {
6176 char fb_name
[IFNAMSIZ
];
6178 /* Ignore unmoveable devices (i.e. loopback) */
6179 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6182 /* Leave virtual devices for the generic cleanup */
6183 if (dev
->rtnl_link_ops
)
6186 /* Push remaing network devices to init_net */
6187 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6188 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6190 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
6191 __func__
, dev
->name
, err
);
6198 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6200 /* At exit all network devices most be removed from a network
6201 * namespace. Do this in the reverse order of registration.
6202 * Do this across as many network namespaces as possible to
6203 * improve batching efficiency.
6205 struct net_device
*dev
;
6207 LIST_HEAD(dev_kill_list
);
6210 list_for_each_entry(net
, net_list
, exit_list
) {
6211 for_each_netdev_reverse(net
, dev
) {
6212 if (dev
->rtnl_link_ops
)
6213 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6215 unregister_netdevice_queue(dev
, &dev_kill_list
);
6218 unregister_netdevice_many(&dev_kill_list
);
6222 static struct pernet_operations __net_initdata default_device_ops
= {
6223 .exit
= default_device_exit
,
6224 .exit_batch
= default_device_exit_batch
,
6228 * Initialize the DEV module. At boot time this walks the device list and
6229 * unhooks any devices that fail to initialise (normally hardware not
6230 * present) and leaves us with a valid list of present and active devices.
6235 * This is called single threaded during boot, so no need
6236 * to take the rtnl semaphore.
6238 static int __init
net_dev_init(void)
6240 int i
, rc
= -ENOMEM
;
6242 BUG_ON(!dev_boot_phase
);
6244 if (dev_proc_init())
6247 if (netdev_kobject_init())
6250 INIT_LIST_HEAD(&ptype_all
);
6251 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6252 INIT_LIST_HEAD(&ptype_base
[i
]);
6254 if (register_pernet_subsys(&netdev_net_ops
))
6258 * Initialise the packet receive queues.
6261 for_each_possible_cpu(i
) {
6262 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6264 memset(sd
, 0, sizeof(*sd
));
6265 skb_queue_head_init(&sd
->input_pkt_queue
);
6266 skb_queue_head_init(&sd
->process_queue
);
6267 sd
->completion_queue
= NULL
;
6268 INIT_LIST_HEAD(&sd
->poll_list
);
6269 sd
->output_queue
= NULL
;
6270 sd
->output_queue_tailp
= &sd
->output_queue
;
6272 sd
->csd
.func
= rps_trigger_softirq
;
6278 sd
->backlog
.poll
= process_backlog
;
6279 sd
->backlog
.weight
= weight_p
;
6280 sd
->backlog
.gro_list
= NULL
;
6281 sd
->backlog
.gro_count
= 0;
6286 /* The loopback device is special if any other network devices
6287 * is present in a network namespace the loopback device must
6288 * be present. Since we now dynamically allocate and free the
6289 * loopback device ensure this invariant is maintained by
6290 * keeping the loopback device as the first device on the
6291 * list of network devices. Ensuring the loopback devices
6292 * is the first device that appears and the last network device
6295 if (register_pernet_device(&loopback_net_ops
))
6298 if (register_pernet_device(&default_device_ops
))
6301 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6302 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6304 hotcpu_notifier(dev_cpu_callback
, 0);
6312 subsys_initcall(net_dev_init
);
6314 static int __init
initialize_hashrnd(void)
6316 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6320 late_initcall_sync(initialize_hashrnd
);