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>
135 #include <linux/cpu_rmap.h>
137 #include "net-sysfs.h"
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
146 * The list of packet types we will receive (as opposed to discard)
147 * and the routines to invoke.
149 * Why 16. Because with 16 the only overlap we get on a hash of the
150 * low nibble of the protocol value is RARP/SNAP/X.25.
152 * NOTE: That is no longer true with the addition of VLAN tags. Not
153 * sure which should go first, but I bet it won't make much
154 * difference if we are running VLANs. The good news is that
155 * this protocol won't be in the list unless compiled in, so
156 * the average user (w/out VLANs) will not be adversely affected.
173 #define PTYPE_HASH_SIZE (16)
174 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
176 static DEFINE_SPINLOCK(ptype_lock
);
177 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
178 static struct list_head ptype_all __read_mostly
; /* Taps */
181 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
184 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
186 * Writers must hold the rtnl semaphore while they loop through the
187 * dev_base_head list, and hold dev_base_lock for writing when they do the
188 * actual updates. This allows pure readers to access the list even
189 * while a writer is preparing to update it.
191 * To put it another way, dev_base_lock is held for writing only to
192 * protect against pure readers; the rtnl semaphore provides the
193 * protection against other writers.
195 * See, for example usages, register_netdevice() and
196 * unregister_netdevice(), which must be called with the rtnl
199 DEFINE_RWLOCK(dev_base_lock
);
200 EXPORT_SYMBOL(dev_base_lock
);
202 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
204 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
205 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
208 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
210 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
213 static inline void rps_lock(struct softnet_data
*sd
)
216 spin_lock(&sd
->input_pkt_queue
.lock
);
220 static inline void rps_unlock(struct softnet_data
*sd
)
223 spin_unlock(&sd
->input_pkt_queue
.lock
);
227 /* Device list insertion */
228 static int list_netdevice(struct net_device
*dev
)
230 struct net
*net
= dev_net(dev
);
234 write_lock_bh(&dev_base_lock
);
235 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
236 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
237 hlist_add_head_rcu(&dev
->index_hlist
,
238 dev_index_hash(net
, dev
->ifindex
));
239 write_unlock_bh(&dev_base_lock
);
243 /* Device list removal
244 * caller must respect a RCU grace period before freeing/reusing dev
246 static void unlist_netdevice(struct net_device
*dev
)
250 /* Unlink dev from the device chain */
251 write_lock_bh(&dev_base_lock
);
252 list_del_rcu(&dev
->dev_list
);
253 hlist_del_rcu(&dev
->name_hlist
);
254 hlist_del_rcu(&dev
->index_hlist
);
255 write_unlock_bh(&dev_base_lock
);
262 static RAW_NOTIFIER_HEAD(netdev_chain
);
265 * Device drivers call our routines to queue packets here. We empty the
266 * queue in the local softnet handler.
269 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
270 EXPORT_PER_CPU_SYMBOL(softnet_data
);
272 #ifdef CONFIG_LOCKDEP
274 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
275 * according to dev->type
277 static const unsigned short netdev_lock_type
[] =
278 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
279 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
280 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
281 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
282 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
283 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
284 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
285 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
286 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
287 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
288 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
289 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
290 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
291 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
292 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
293 ARPHRD_VOID
, ARPHRD_NONE
};
295 static const char *const netdev_lock_name
[] =
296 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
297 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
298 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
299 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
300 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
301 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
302 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
303 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
304 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
305 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
306 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
307 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
308 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
309 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
310 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
311 "_xmit_VOID", "_xmit_NONE"};
313 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
314 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
316 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
320 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
321 if (netdev_lock_type
[i
] == dev_type
)
323 /* the last key is used by default */
324 return ARRAY_SIZE(netdev_lock_type
) - 1;
327 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
328 unsigned short dev_type
)
332 i
= netdev_lock_pos(dev_type
);
333 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
334 netdev_lock_name
[i
]);
337 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
341 i
= netdev_lock_pos(dev
->type
);
342 lockdep_set_class_and_name(&dev
->addr_list_lock
,
343 &netdev_addr_lock_key
[i
],
344 netdev_lock_name
[i
]);
347 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
348 unsigned short dev_type
)
351 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
356 /*******************************************************************************
358 Protocol management and registration routines
360 *******************************************************************************/
363 * Add a protocol ID to the list. Now that the input handler is
364 * smarter we can dispense with all the messy stuff that used to be
367 * BEWARE!!! Protocol handlers, mangling input packets,
368 * MUST BE last in hash buckets and checking protocol handlers
369 * MUST start from promiscuous ptype_all chain in net_bh.
370 * It is true now, do not change it.
371 * Explanation follows: if protocol handler, mangling packet, will
372 * be the first on list, it is not able to sense, that packet
373 * is cloned and should be copied-on-write, so that it will
374 * change it and subsequent readers will get broken packet.
378 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
380 if (pt
->type
== htons(ETH_P_ALL
))
383 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
387 * dev_add_pack - add packet handler
388 * @pt: packet type declaration
390 * Add a protocol handler to the networking stack. The passed &packet_type
391 * is linked into kernel lists and may not be freed until it has been
392 * removed from the kernel lists.
394 * This call does not sleep therefore it can not
395 * guarantee all CPU's that are in middle of receiving packets
396 * will see the new packet type (until the next received packet).
399 void dev_add_pack(struct packet_type
*pt
)
401 struct list_head
*head
= ptype_head(pt
);
403 spin_lock(&ptype_lock
);
404 list_add_rcu(&pt
->list
, head
);
405 spin_unlock(&ptype_lock
);
407 EXPORT_SYMBOL(dev_add_pack
);
410 * __dev_remove_pack - remove packet handler
411 * @pt: packet type declaration
413 * Remove a protocol handler that was previously added to the kernel
414 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
415 * from the kernel lists and can be freed or reused once this function
418 * The packet type might still be in use by receivers
419 * and must not be freed until after all the CPU's have gone
420 * through a quiescent state.
422 void __dev_remove_pack(struct packet_type
*pt
)
424 struct list_head
*head
= ptype_head(pt
);
425 struct packet_type
*pt1
;
427 spin_lock(&ptype_lock
);
429 list_for_each_entry(pt1
, head
, list
) {
431 list_del_rcu(&pt
->list
);
436 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
438 spin_unlock(&ptype_lock
);
440 EXPORT_SYMBOL(__dev_remove_pack
);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type
*pt
)
456 __dev_remove_pack(pt
);
460 EXPORT_SYMBOL(dev_remove_pack
);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
482 struct netdev_boot_setup
*s
;
486 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
487 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
488 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
489 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
490 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
495 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device
*dev
)
509 struct netdev_boot_setup
*s
= dev_boot_setup
;
512 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
513 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
514 !strcmp(dev
->name
, s
[i
].name
)) {
515 dev
->irq
= s
[i
].map
.irq
;
516 dev
->base_addr
= s
[i
].map
.base_addr
;
517 dev
->mem_start
= s
[i
].map
.mem_start
;
518 dev
->mem_end
= s
[i
].map
.mem_end
;
524 EXPORT_SYMBOL(netdev_boot_setup_check
);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix
, int unit
)
539 const struct netdev_boot_setup
*s
= dev_boot_setup
;
543 sprintf(name
, "%s%d", prefix
, unit
);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net
, name
))
552 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
553 if (!strcmp(name
, s
[i
].name
))
554 return s
[i
].map
.base_addr
;
559 * Saves at boot time configured settings for any netdevice.
561 int __init
netdev_boot_setup(char *str
)
566 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
571 memset(&map
, 0, sizeof(map
));
575 map
.base_addr
= ints
[2];
577 map
.mem_start
= ints
[3];
579 map
.mem_end
= ints
[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str
, &map
);
585 __setup("netdev=", netdev_boot_setup
);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
607 struct hlist_node
*p
;
608 struct net_device
*dev
;
609 struct hlist_head
*head
= dev_name_hash(net
, name
);
611 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
612 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
617 EXPORT_SYMBOL(__dev_get_by_name
);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
633 struct hlist_node
*p
;
634 struct net_device
*dev
;
635 struct hlist_head
*head
= dev_name_hash(net
, name
);
637 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
638 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
643 EXPORT_SYMBOL(dev_get_by_name_rcu
);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
659 struct net_device
*dev
;
662 dev
= dev_get_by_name_rcu(net
, name
);
668 EXPORT_SYMBOL(dev_get_by_name
);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
684 struct hlist_node
*p
;
685 struct net_device
*dev
;
686 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
688 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
689 if (dev
->ifindex
== ifindex
)
694 EXPORT_SYMBOL(__dev_get_by_index
);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
709 struct hlist_node
*p
;
710 struct net_device
*dev
;
711 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
713 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
714 if (dev
->ifindex
== ifindex
)
719 EXPORT_SYMBOL(dev_get_by_index_rcu
);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
735 struct net_device
*dev
;
738 dev
= dev_get_by_index_rcu(net
, ifindex
);
744 EXPORT_SYMBOL(dev_get_by_index
);
747 * dev_getbyhwaddr_rcu - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device.
754 * The caller must hold RCU or RTNL.
755 * The returned device has not had its ref count increased
756 * and the caller must therefore be careful about locking
760 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
763 struct net_device
*dev
;
765 for_each_netdev_rcu(net
, dev
)
766 if (dev
->type
== type
&&
767 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
772 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
774 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
776 struct net_device
*dev
;
779 for_each_netdev(net
, dev
)
780 if (dev
->type
== type
)
785 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
787 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
789 struct net_device
*dev
, *ret
= NULL
;
792 for_each_netdev_rcu(net
, dev
)
793 if (dev
->type
== type
) {
801 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
804 * dev_get_by_flags_rcu - find any device with given flags
805 * @net: the applicable net namespace
806 * @if_flags: IFF_* values
807 * @mask: bitmask of bits in if_flags to check
809 * Search for any interface with the given flags. Returns NULL if a device
810 * is not found or a pointer to the device. Must be called inside
811 * rcu_read_lock(), and result refcount is unchanged.
814 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
817 struct net_device
*dev
, *ret
;
820 for_each_netdev_rcu(net
, dev
) {
821 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
828 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
831 * dev_valid_name - check if name is okay for network device
834 * Network device names need to be valid file names to
835 * to allow sysfs to work. We also disallow any kind of
838 int dev_valid_name(const char *name
)
842 if (strlen(name
) >= IFNAMSIZ
)
844 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
848 if (*name
== '/' || isspace(*name
))
854 EXPORT_SYMBOL(dev_valid_name
);
857 * __dev_alloc_name - allocate a name for a device
858 * @net: network namespace to allocate the device name in
859 * @name: name format string
860 * @buf: scratch buffer and result name string
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
871 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
875 const int max_netdevices
= 8*PAGE_SIZE
;
876 unsigned long *inuse
;
877 struct net_device
*d
;
879 p
= strnchr(name
, IFNAMSIZ
-1, '%');
882 * Verify the string as this thing may have come from
883 * the user. There must be either one "%d" and no other "%"
886 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
889 /* Use one page as a bit array of possible slots */
890 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
894 for_each_netdev(net
, d
) {
895 if (!sscanf(d
->name
, name
, &i
))
897 if (i
< 0 || i
>= max_netdevices
)
900 /* avoid cases where sscanf is not exact inverse of printf */
901 snprintf(buf
, IFNAMSIZ
, name
, i
);
902 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
906 i
= find_first_zero_bit(inuse
, max_netdevices
);
907 free_page((unsigned long) inuse
);
911 snprintf(buf
, IFNAMSIZ
, name
, i
);
912 if (!__dev_get_by_name(net
, buf
))
915 /* It is possible to run out of possible slots
916 * when the name is long and there isn't enough space left
917 * for the digits, or if all bits are used.
923 * dev_alloc_name - allocate a name for a device
925 * @name: name format string
927 * Passed a format string - eg "lt%d" it will try and find a suitable
928 * id. It scans list of devices to build up a free map, then chooses
929 * the first empty slot. The caller must hold the dev_base or rtnl lock
930 * while allocating the name and adding the device in order to avoid
932 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
933 * Returns the number of the unit assigned or a negative errno code.
936 int dev_alloc_name(struct net_device
*dev
, const char *name
)
942 BUG_ON(!dev_net(dev
));
944 ret
= __dev_alloc_name(net
, name
, buf
);
946 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
949 EXPORT_SYMBOL(dev_alloc_name
);
951 static int dev_get_valid_name(struct net_device
*dev
, const char *name
, bool fmt
)
955 BUG_ON(!dev_net(dev
));
958 if (!dev_valid_name(name
))
961 if (fmt
&& strchr(name
, '%'))
962 return dev_alloc_name(dev
, name
);
963 else if (__dev_get_by_name(net
, name
))
965 else if (dev
->name
!= name
)
966 strlcpy(dev
->name
, name
, IFNAMSIZ
);
972 * dev_change_name - change name of a device
974 * @newname: name (or format string) must be at least IFNAMSIZ
976 * Change name of a device, can pass format strings "eth%d".
979 int dev_change_name(struct net_device
*dev
, const char *newname
)
981 char oldname
[IFNAMSIZ
];
987 BUG_ON(!dev_net(dev
));
990 if (dev
->flags
& IFF_UP
)
993 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
996 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
998 err
= dev_get_valid_name(dev
, newname
, 1);
1003 ret
= device_rename(&dev
->dev
, dev
->name
);
1005 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1009 write_lock_bh(&dev_base_lock
);
1010 hlist_del(&dev
->name_hlist
);
1011 write_unlock_bh(&dev_base_lock
);
1015 write_lock_bh(&dev_base_lock
);
1016 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1017 write_unlock_bh(&dev_base_lock
);
1019 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1020 ret
= notifier_to_errno(ret
);
1023 /* err >= 0 after dev_alloc_name() or stores the first errno */
1026 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1030 "%s: name change rollback failed: %d.\n",
1039 * dev_set_alias - change ifalias of a device
1041 * @alias: name up to IFALIASZ
1042 * @len: limit of bytes to copy from info
1044 * Set ifalias for a device,
1046 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1050 if (len
>= IFALIASZ
)
1055 kfree(dev
->ifalias
);
1056 dev
->ifalias
= NULL
;
1061 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1065 strlcpy(dev
->ifalias
, alias
, len
+1);
1071 * netdev_features_change - device changes features
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed features.
1076 void netdev_features_change(struct net_device
*dev
)
1078 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1080 EXPORT_SYMBOL(netdev_features_change
);
1083 * netdev_state_change - device changes state
1084 * @dev: device to cause notification
1086 * Called to indicate a device has changed state. This function calls
1087 * the notifier chains for netdev_chain and sends a NEWLINK message
1088 * to the routing socket.
1090 void netdev_state_change(struct net_device
*dev
)
1092 if (dev
->flags
& IFF_UP
) {
1093 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1094 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1097 EXPORT_SYMBOL(netdev_state_change
);
1099 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1101 return call_netdevice_notifiers(event
, dev
);
1103 EXPORT_SYMBOL(netdev_bonding_change
);
1106 * dev_load - load a network module
1107 * @net: the applicable net namespace
1108 * @name: name of interface
1110 * If a network interface is not present and the process has suitable
1111 * privileges this function loads the module. If module loading is not
1112 * available in this kernel then it becomes a nop.
1115 void dev_load(struct net
*net
, const char *name
)
1117 struct net_device
*dev
;
1120 dev
= dev_get_by_name_rcu(net
, name
);
1123 if (!dev
&& capable(CAP_NET_ADMIN
))
1124 request_module("%s", name
);
1126 EXPORT_SYMBOL(dev_load
);
1128 static int __dev_open(struct net_device
*dev
)
1130 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1136 * Is it even present?
1138 if (!netif_device_present(dev
))
1141 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1142 ret
= notifier_to_errno(ret
);
1147 * Call device private open method
1149 set_bit(__LINK_STATE_START
, &dev
->state
);
1151 if (ops
->ndo_validate_addr
)
1152 ret
= ops
->ndo_validate_addr(dev
);
1154 if (!ret
&& ops
->ndo_open
)
1155 ret
= ops
->ndo_open(dev
);
1158 * If it went open OK then:
1162 clear_bit(__LINK_STATE_START
, &dev
->state
);
1167 dev
->flags
|= IFF_UP
;
1172 net_dmaengine_get();
1175 * Initialize multicasting status
1177 dev_set_rx_mode(dev
);
1180 * Wakeup transmit queue engine
1189 * dev_open - prepare an interface for use.
1190 * @dev: device to open
1192 * Takes a device from down to up state. The device's private open
1193 * function is invoked and then the multicast lists are loaded. Finally
1194 * the device is moved into the up state and a %NETDEV_UP message is
1195 * sent to the netdev notifier chain.
1197 * Calling this function on an active interface is a nop. On a failure
1198 * a negative errno code is returned.
1200 int dev_open(struct net_device
*dev
)
1207 if (dev
->flags
& IFF_UP
)
1213 ret
= __dev_open(dev
);
1218 * ... and announce new interface.
1220 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1221 call_netdevice_notifiers(NETDEV_UP
, dev
);
1225 EXPORT_SYMBOL(dev_open
);
1227 static int __dev_close_many(struct list_head
*head
)
1229 struct net_device
*dev
;
1234 list_for_each_entry(dev
, head
, unreg_list
) {
1236 * Tell people we are going down, so that they can
1237 * prepare to death, when device is still operating.
1239 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1241 clear_bit(__LINK_STATE_START
, &dev
->state
);
1243 /* Synchronize to scheduled poll. We cannot touch poll list, it
1244 * can be even on different cpu. So just clear netif_running().
1246 * dev->stop() will invoke napi_disable() on all of it's
1247 * napi_struct instances on this device.
1249 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1252 dev_deactivate_many(head
);
1254 list_for_each_entry(dev
, head
, unreg_list
) {
1255 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1258 * Call the device specific close. This cannot fail.
1259 * Only if device is UP
1261 * We allow it to be called even after a DETACH hot-plug
1268 * Device is now down.
1271 dev
->flags
&= ~IFF_UP
;
1276 net_dmaengine_put();
1282 static int __dev_close(struct net_device
*dev
)
1286 list_add(&dev
->unreg_list
, &single
);
1287 return __dev_close_many(&single
);
1290 static int dev_close_many(struct list_head
*head
)
1292 struct net_device
*dev
, *tmp
;
1293 LIST_HEAD(tmp_list
);
1295 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1296 if (!(dev
->flags
& IFF_UP
))
1297 list_move(&dev
->unreg_list
, &tmp_list
);
1299 __dev_close_many(head
);
1302 * Tell people we are down
1304 list_for_each_entry(dev
, head
, unreg_list
) {
1305 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1306 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1309 /* rollback_registered_many needs the complete original list */
1310 list_splice(&tmp_list
, head
);
1315 * dev_close - shutdown an interface.
1316 * @dev: device to shutdown
1318 * This function moves an active device into down state. A
1319 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1320 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1323 int dev_close(struct net_device
*dev
)
1327 list_add(&dev
->unreg_list
, &single
);
1328 dev_close_many(&single
);
1332 EXPORT_SYMBOL(dev_close
);
1336 * dev_disable_lro - disable Large Receive Offload on a device
1339 * Disable Large Receive Offload (LRO) on a net device. Must be
1340 * called under RTNL. This is needed if received packets may be
1341 * forwarded to another interface.
1343 void dev_disable_lro(struct net_device
*dev
)
1345 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1346 dev
->ethtool_ops
->set_flags
) {
1347 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1348 if (flags
& ETH_FLAG_LRO
) {
1349 flags
&= ~ETH_FLAG_LRO
;
1350 dev
->ethtool_ops
->set_flags(dev
, flags
);
1353 WARN_ON(dev
->features
& NETIF_F_LRO
);
1355 EXPORT_SYMBOL(dev_disable_lro
);
1358 static int dev_boot_phase
= 1;
1361 * Device change register/unregister. These are not inline or static
1362 * as we export them to the world.
1366 * register_netdevice_notifier - register a network notifier block
1369 * Register a notifier to be called when network device events occur.
1370 * The notifier passed is linked into the kernel structures and must
1371 * not be reused until it has been unregistered. A negative errno code
1372 * is returned on a failure.
1374 * When registered all registration and up events are replayed
1375 * to the new notifier to allow device to have a race free
1376 * view of the network device list.
1379 int register_netdevice_notifier(struct notifier_block
*nb
)
1381 struct net_device
*dev
;
1382 struct net_device
*last
;
1387 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1393 for_each_netdev(net
, dev
) {
1394 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1395 err
= notifier_to_errno(err
);
1399 if (!(dev
->flags
& IFF_UP
))
1402 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1413 for_each_netdev(net
, dev
) {
1417 if (dev
->flags
& IFF_UP
) {
1418 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1419 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1421 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1422 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1426 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1429 EXPORT_SYMBOL(register_netdevice_notifier
);
1432 * unregister_netdevice_notifier - unregister a network notifier block
1435 * Unregister a notifier previously registered by
1436 * register_netdevice_notifier(). The notifier is unlinked into the
1437 * kernel structures and may then be reused. A negative errno code
1438 * is returned on a failure.
1441 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1446 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1450 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1453 * call_netdevice_notifiers - call all network notifier blocks
1454 * @val: value passed unmodified to notifier function
1455 * @dev: net_device pointer passed unmodified to notifier function
1457 * Call all network notifier blocks. Parameters and return value
1458 * are as for raw_notifier_call_chain().
1461 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1464 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1467 /* When > 0 there are consumers of rx skb time stamps */
1468 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1470 void net_enable_timestamp(void)
1472 atomic_inc(&netstamp_needed
);
1474 EXPORT_SYMBOL(net_enable_timestamp
);
1476 void net_disable_timestamp(void)
1478 atomic_dec(&netstamp_needed
);
1480 EXPORT_SYMBOL(net_disable_timestamp
);
1482 static inline void net_timestamp_set(struct sk_buff
*skb
)
1484 if (atomic_read(&netstamp_needed
))
1485 __net_timestamp(skb
);
1487 skb
->tstamp
.tv64
= 0;
1490 static inline void net_timestamp_check(struct sk_buff
*skb
)
1492 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1493 __net_timestamp(skb
);
1497 * dev_forward_skb - loopback an skb to another netif
1499 * @dev: destination network device
1500 * @skb: buffer to forward
1503 * NET_RX_SUCCESS (no congestion)
1504 * NET_RX_DROP (packet was dropped, but freed)
1506 * dev_forward_skb can be used for injecting an skb from the
1507 * start_xmit function of one device into the receive queue
1508 * of another device.
1510 * The receiving device may be in another namespace, so
1511 * we have to clear all information in the skb that could
1512 * impact namespace isolation.
1514 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1519 if (unlikely(!(dev
->flags
& IFF_UP
) ||
1520 (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
)))) {
1521 atomic_long_inc(&dev
->rx_dropped
);
1525 skb_set_dev(skb
, dev
);
1526 skb
->tstamp
.tv64
= 0;
1527 skb
->pkt_type
= PACKET_HOST
;
1528 skb
->protocol
= eth_type_trans(skb
, dev
);
1529 return netif_rx(skb
);
1531 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1533 static inline int deliver_skb(struct sk_buff
*skb
,
1534 struct packet_type
*pt_prev
,
1535 struct net_device
*orig_dev
)
1537 atomic_inc(&skb
->users
);
1538 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1542 * Support routine. Sends outgoing frames to any network
1543 * taps currently in use.
1546 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1548 struct packet_type
*ptype
;
1549 struct sk_buff
*skb2
= NULL
;
1550 struct packet_type
*pt_prev
= NULL
;
1553 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1554 /* Never send packets back to the socket
1555 * they originated from - MvS (miquels@drinkel.ow.org)
1557 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1558 (ptype
->af_packet_priv
== NULL
||
1559 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1561 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1566 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1570 net_timestamp_set(skb2
);
1572 /* skb->nh should be correctly
1573 set by sender, so that the second statement is
1574 just protection against buggy protocols.
1576 skb_reset_mac_header(skb2
);
1578 if (skb_network_header(skb2
) < skb2
->data
||
1579 skb2
->network_header
> skb2
->tail
) {
1580 if (net_ratelimit())
1581 printk(KERN_CRIT
"protocol %04x is "
1583 ntohs(skb2
->protocol
),
1585 skb_reset_network_header(skb2
);
1588 skb2
->transport_header
= skb2
->network_header
;
1589 skb2
->pkt_type
= PACKET_OUTGOING
;
1594 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1598 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1599 * @dev: Network device
1600 * @txq: number of queues available
1602 * If real_num_tx_queues is changed the tc mappings may no longer be
1603 * valid. To resolve this verify the tc mapping remains valid and if
1604 * not NULL the mapping. With no priorities mapping to this
1605 * offset/count pair it will no longer be used. In the worst case TC0
1606 * is invalid nothing can be done so disable priority mappings. If is
1607 * expected that drivers will fix this mapping if they can before
1608 * calling netif_set_real_num_tx_queues.
1610 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1613 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1615 /* If TC0 is invalidated disable TC mapping */
1616 if (tc
->offset
+ tc
->count
> txq
) {
1617 pr_warning("Number of in use tx queues changed "
1618 "invalidating tc mappings. Priority "
1619 "traffic classification disabled!\n");
1624 /* Invalidated prio to tc mappings set to TC0 */
1625 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1626 int q
= netdev_get_prio_tc_map(dev
, i
);
1628 tc
= &dev
->tc_to_txq
[q
];
1629 if (tc
->offset
+ tc
->count
> txq
) {
1630 pr_warning("Number of in use tx queues "
1631 "changed. Priority %i to tc "
1632 "mapping %i is no longer valid "
1633 "setting map to 0\n",
1635 netdev_set_prio_tc_map(dev
, i
, 0);
1641 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1642 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1644 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1648 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1651 if (dev
->reg_state
== NETREG_REGISTERED
||
1652 dev
->reg_state
== NETREG_UNREGISTERING
) {
1655 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1661 netif_setup_tc(dev
, txq
);
1663 if (txq
< dev
->real_num_tx_queues
)
1664 qdisc_reset_all_tx_gt(dev
, txq
);
1667 dev
->real_num_tx_queues
= txq
;
1670 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1674 * netif_set_real_num_rx_queues - set actual number of RX queues used
1675 * @dev: Network device
1676 * @rxq: Actual number of RX queues
1678 * This must be called either with the rtnl_lock held or before
1679 * registration of the net device. Returns 0 on success, or a
1680 * negative error code. If called before registration, it always
1683 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1687 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1690 if (dev
->reg_state
== NETREG_REGISTERED
) {
1693 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1699 dev
->real_num_rx_queues
= rxq
;
1702 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1705 static inline void __netif_reschedule(struct Qdisc
*q
)
1707 struct softnet_data
*sd
;
1708 unsigned long flags
;
1710 local_irq_save(flags
);
1711 sd
= &__get_cpu_var(softnet_data
);
1712 q
->next_sched
= NULL
;
1713 *sd
->output_queue_tailp
= q
;
1714 sd
->output_queue_tailp
= &q
->next_sched
;
1715 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1716 local_irq_restore(flags
);
1719 void __netif_schedule(struct Qdisc
*q
)
1721 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1722 __netif_reschedule(q
);
1724 EXPORT_SYMBOL(__netif_schedule
);
1726 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1728 if (atomic_dec_and_test(&skb
->users
)) {
1729 struct softnet_data
*sd
;
1730 unsigned long flags
;
1732 local_irq_save(flags
);
1733 sd
= &__get_cpu_var(softnet_data
);
1734 skb
->next
= sd
->completion_queue
;
1735 sd
->completion_queue
= skb
;
1736 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1737 local_irq_restore(flags
);
1740 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1742 void dev_kfree_skb_any(struct sk_buff
*skb
)
1744 if (in_irq() || irqs_disabled())
1745 dev_kfree_skb_irq(skb
);
1749 EXPORT_SYMBOL(dev_kfree_skb_any
);
1753 * netif_device_detach - mark device as removed
1754 * @dev: network device
1756 * Mark device as removed from system and therefore no longer available.
1758 void netif_device_detach(struct net_device
*dev
)
1760 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1761 netif_running(dev
)) {
1762 netif_tx_stop_all_queues(dev
);
1765 EXPORT_SYMBOL(netif_device_detach
);
1768 * netif_device_attach - mark device as attached
1769 * @dev: network device
1771 * Mark device as attached from system and restart if needed.
1773 void netif_device_attach(struct net_device
*dev
)
1775 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1776 netif_running(dev
)) {
1777 netif_tx_wake_all_queues(dev
);
1778 __netdev_watchdog_up(dev
);
1781 EXPORT_SYMBOL(netif_device_attach
);
1784 * skb_dev_set -- assign a new device to a buffer
1785 * @skb: buffer for the new device
1786 * @dev: network device
1788 * If an skb is owned by a device already, we have to reset
1789 * all data private to the namespace a device belongs to
1790 * before assigning it a new device.
1792 #ifdef CONFIG_NET_NS
1793 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1796 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1799 skb_init_secmark(skb
);
1803 skb
->ipvs_property
= 0;
1804 #ifdef CONFIG_NET_SCHED
1810 EXPORT_SYMBOL(skb_set_dev
);
1811 #endif /* CONFIG_NET_NS */
1814 * Invalidate hardware checksum when packet is to be mangled, and
1815 * complete checksum manually on outgoing path.
1817 int skb_checksum_help(struct sk_buff
*skb
)
1820 int ret
= 0, offset
;
1822 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1823 goto out_set_summed
;
1825 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1826 /* Let GSO fix up the checksum. */
1827 goto out_set_summed
;
1830 offset
= skb_checksum_start_offset(skb
);
1831 BUG_ON(offset
>= skb_headlen(skb
));
1832 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1834 offset
+= skb
->csum_offset
;
1835 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1837 if (skb_cloned(skb
) &&
1838 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1839 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1844 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1846 skb
->ip_summed
= CHECKSUM_NONE
;
1850 EXPORT_SYMBOL(skb_checksum_help
);
1853 * skb_gso_segment - Perform segmentation on skb.
1854 * @skb: buffer to segment
1855 * @features: features for the output path (see dev->features)
1857 * This function segments the given skb and returns a list of segments.
1859 * It may return NULL if the skb requires no segmentation. This is
1860 * only possible when GSO is used for verifying header integrity.
1862 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, u32 features
)
1864 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1865 struct packet_type
*ptype
;
1866 __be16 type
= skb
->protocol
;
1867 int vlan_depth
= ETH_HLEN
;
1870 while (type
== htons(ETH_P_8021Q
)) {
1871 struct vlan_hdr
*vh
;
1873 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1874 return ERR_PTR(-EINVAL
);
1876 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1877 type
= vh
->h_vlan_encapsulated_proto
;
1878 vlan_depth
+= VLAN_HLEN
;
1881 skb_reset_mac_header(skb
);
1882 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1883 __skb_pull(skb
, skb
->mac_len
);
1885 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1886 struct net_device
*dev
= skb
->dev
;
1887 struct ethtool_drvinfo info
= {};
1889 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1890 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1892 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1893 info
.driver
, dev
? dev
->features
: 0L,
1894 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1895 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1897 if (skb_header_cloned(skb
) &&
1898 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1899 return ERR_PTR(err
);
1903 list_for_each_entry_rcu(ptype
,
1904 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1905 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1906 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1907 err
= ptype
->gso_send_check(skb
);
1908 segs
= ERR_PTR(err
);
1909 if (err
|| skb_gso_ok(skb
, features
))
1911 __skb_push(skb
, (skb
->data
-
1912 skb_network_header(skb
)));
1914 segs
= ptype
->gso_segment(skb
, features
);
1920 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1924 EXPORT_SYMBOL(skb_gso_segment
);
1926 /* Take action when hardware reception checksum errors are detected. */
1928 void netdev_rx_csum_fault(struct net_device
*dev
)
1930 if (net_ratelimit()) {
1931 printk(KERN_ERR
"%s: hw csum failure.\n",
1932 dev
? dev
->name
: "<unknown>");
1936 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1939 /* Actually, we should eliminate this check as soon as we know, that:
1940 * 1. IOMMU is present and allows to map all the memory.
1941 * 2. No high memory really exists on this machine.
1944 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1946 #ifdef CONFIG_HIGHMEM
1948 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1949 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1950 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1954 if (PCI_DMA_BUS_IS_PHYS
) {
1955 struct device
*pdev
= dev
->dev
.parent
;
1959 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1960 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1961 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1970 void (*destructor
)(struct sk_buff
*skb
);
1973 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1975 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1977 struct dev_gso_cb
*cb
;
1980 struct sk_buff
*nskb
= skb
->next
;
1982 skb
->next
= nskb
->next
;
1985 } while (skb
->next
);
1987 cb
= DEV_GSO_CB(skb
);
1989 cb
->destructor(skb
);
1993 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1994 * @skb: buffer to segment
1995 * @features: device features as applicable to this skb
1997 * This function segments the given skb and stores the list of segments
2000 static int dev_gso_segment(struct sk_buff
*skb
, int features
)
2002 struct sk_buff
*segs
;
2004 segs
= skb_gso_segment(skb
, features
);
2006 /* Verifying header integrity only. */
2011 return PTR_ERR(segs
);
2014 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2015 skb
->destructor
= dev_gso_skb_destructor
;
2021 * Try to orphan skb early, right before transmission by the device.
2022 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2023 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2025 static inline void skb_orphan_try(struct sk_buff
*skb
)
2027 struct sock
*sk
= skb
->sk
;
2029 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
2030 /* skb_tx_hash() wont be able to get sk.
2031 * We copy sk_hash into skb->rxhash
2034 skb
->rxhash
= sk
->sk_hash
;
2039 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
2041 return ((features
& NETIF_F_GEN_CSUM
) ||
2042 ((features
& NETIF_F_V4_CSUM
) &&
2043 protocol
== htons(ETH_P_IP
)) ||
2044 ((features
& NETIF_F_V6_CSUM
) &&
2045 protocol
== htons(ETH_P_IPV6
)) ||
2046 ((features
& NETIF_F_FCOE_CRC
) &&
2047 protocol
== htons(ETH_P_FCOE
)));
2050 static u32
harmonize_features(struct sk_buff
*skb
, __be16 protocol
, u32 features
)
2052 if (!can_checksum_protocol(features
, protocol
)) {
2053 features
&= ~NETIF_F_ALL_CSUM
;
2054 features
&= ~NETIF_F_SG
;
2055 } else if (illegal_highdma(skb
->dev
, skb
)) {
2056 features
&= ~NETIF_F_SG
;
2062 u32
netif_skb_features(struct sk_buff
*skb
)
2064 __be16 protocol
= skb
->protocol
;
2065 u32 features
= skb
->dev
->features
;
2067 if (protocol
== htons(ETH_P_8021Q
)) {
2068 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2069 protocol
= veh
->h_vlan_encapsulated_proto
;
2070 } else if (!vlan_tx_tag_present(skb
)) {
2071 return harmonize_features(skb
, protocol
, features
);
2074 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2076 if (protocol
!= htons(ETH_P_8021Q
)) {
2077 return harmonize_features(skb
, protocol
, features
);
2079 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2080 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2081 return harmonize_features(skb
, protocol
, features
);
2084 EXPORT_SYMBOL(netif_skb_features
);
2087 * Returns true if either:
2088 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2089 * 2. skb is fragmented and the device does not support SG, or if
2090 * at least one of fragments is in highmem and device does not
2091 * support DMA from it.
2093 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2096 return skb_is_nonlinear(skb
) &&
2097 ((skb_has_frag_list(skb
) &&
2098 !(features
& NETIF_F_FRAGLIST
)) ||
2099 (skb_shinfo(skb
)->nr_frags
&&
2100 !(features
& NETIF_F_SG
)));
2103 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2104 struct netdev_queue
*txq
)
2106 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2107 int rc
= NETDEV_TX_OK
;
2109 if (likely(!skb
->next
)) {
2113 * If device doesnt need skb->dst, release it right now while
2114 * its hot in this cpu cache
2116 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2119 if (!list_empty(&ptype_all
))
2120 dev_queue_xmit_nit(skb
, dev
);
2122 skb_orphan_try(skb
);
2124 features
= netif_skb_features(skb
);
2126 if (vlan_tx_tag_present(skb
) &&
2127 !(features
& NETIF_F_HW_VLAN_TX
)) {
2128 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2135 if (netif_needs_gso(skb
, features
)) {
2136 if (unlikely(dev_gso_segment(skb
, features
)))
2141 if (skb_needs_linearize(skb
, features
) &&
2142 __skb_linearize(skb
))
2145 /* If packet is not checksummed and device does not
2146 * support checksumming for this protocol, complete
2147 * checksumming here.
2149 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2150 skb_set_transport_header(skb
,
2151 skb_checksum_start_offset(skb
));
2152 if (!(features
& NETIF_F_ALL_CSUM
) &&
2153 skb_checksum_help(skb
))
2158 rc
= ops
->ndo_start_xmit(skb
, dev
);
2159 trace_net_dev_xmit(skb
, rc
);
2160 if (rc
== NETDEV_TX_OK
)
2161 txq_trans_update(txq
);
2167 struct sk_buff
*nskb
= skb
->next
;
2169 skb
->next
= nskb
->next
;
2173 * If device doesnt need nskb->dst, release it right now while
2174 * its hot in this cpu cache
2176 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2179 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2180 trace_net_dev_xmit(nskb
, rc
);
2181 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2182 if (rc
& ~NETDEV_TX_MASK
)
2183 goto out_kfree_gso_skb
;
2184 nskb
->next
= skb
->next
;
2188 txq_trans_update(txq
);
2189 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2190 return NETDEV_TX_BUSY
;
2191 } while (skb
->next
);
2194 if (likely(skb
->next
== NULL
))
2195 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2202 static u32 hashrnd __read_mostly
;
2205 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2206 * to be used as a distribution range.
2208 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2209 unsigned int num_tx_queues
)
2213 u16 qcount
= num_tx_queues
;
2215 if (skb_rx_queue_recorded(skb
)) {
2216 hash
= skb_get_rx_queue(skb
);
2217 while (unlikely(hash
>= num_tx_queues
))
2218 hash
-= num_tx_queues
;
2223 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2224 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2225 qcount
= dev
->tc_to_txq
[tc
].count
;
2228 if (skb
->sk
&& skb
->sk
->sk_hash
)
2229 hash
= skb
->sk
->sk_hash
;
2231 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2232 hash
= jhash_1word(hash
, hashrnd
);
2234 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2236 EXPORT_SYMBOL(__skb_tx_hash
);
2238 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2240 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2241 if (net_ratelimit()) {
2242 pr_warning("%s selects TX queue %d, but "
2243 "real number of TX queues is %d\n",
2244 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2251 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2254 struct xps_dev_maps
*dev_maps
;
2255 struct xps_map
*map
;
2256 int queue_index
= -1;
2259 dev_maps
= rcu_dereference(dev
->xps_maps
);
2261 map
= rcu_dereference(
2262 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2265 queue_index
= map
->queues
[0];
2268 if (skb
->sk
&& skb
->sk
->sk_hash
)
2269 hash
= skb
->sk
->sk_hash
;
2271 hash
= (__force u16
) skb
->protocol
^
2273 hash
= jhash_1word(hash
, hashrnd
);
2274 queue_index
= map
->queues
[
2275 ((u64
)hash
* map
->len
) >> 32];
2277 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2289 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2290 struct sk_buff
*skb
)
2293 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2295 if (dev
->real_num_tx_queues
== 1)
2297 else if (ops
->ndo_select_queue
) {
2298 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2299 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2301 struct sock
*sk
= skb
->sk
;
2302 queue_index
= sk_tx_queue_get(sk
);
2304 if (queue_index
< 0 || skb
->ooo_okay
||
2305 queue_index
>= dev
->real_num_tx_queues
) {
2306 int old_index
= queue_index
;
2308 queue_index
= get_xps_queue(dev
, skb
);
2309 if (queue_index
< 0)
2310 queue_index
= skb_tx_hash(dev
, skb
);
2312 if (queue_index
!= old_index
&& sk
) {
2313 struct dst_entry
*dst
=
2314 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2316 if (dst
&& skb_dst(skb
) == dst
)
2317 sk_tx_queue_set(sk
, queue_index
);
2322 skb_set_queue_mapping(skb
, queue_index
);
2323 return netdev_get_tx_queue(dev
, queue_index
);
2326 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2327 struct net_device
*dev
,
2328 struct netdev_queue
*txq
)
2330 spinlock_t
*root_lock
= qdisc_lock(q
);
2334 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2335 qdisc_calculate_pkt_len(skb
, q
);
2337 * Heuristic to force contended enqueues to serialize on a
2338 * separate lock before trying to get qdisc main lock.
2339 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2340 * and dequeue packets faster.
2342 contended
= qdisc_is_running(q
);
2343 if (unlikely(contended
))
2344 spin_lock(&q
->busylock
);
2346 spin_lock(root_lock
);
2347 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2350 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2351 qdisc_run_begin(q
)) {
2353 * This is a work-conserving queue; there are no old skbs
2354 * waiting to be sent out; and the qdisc is not running -
2355 * xmit the skb directly.
2357 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2360 qdisc_bstats_update(q
, skb
);
2362 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2363 if (unlikely(contended
)) {
2364 spin_unlock(&q
->busylock
);
2371 rc
= NET_XMIT_SUCCESS
;
2374 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2375 if (qdisc_run_begin(q
)) {
2376 if (unlikely(contended
)) {
2377 spin_unlock(&q
->busylock
);
2383 spin_unlock(root_lock
);
2384 if (unlikely(contended
))
2385 spin_unlock(&q
->busylock
);
2389 static DEFINE_PER_CPU(int, xmit_recursion
);
2390 #define RECURSION_LIMIT 10
2393 * dev_queue_xmit - transmit a buffer
2394 * @skb: buffer to transmit
2396 * Queue a buffer for transmission to a network device. The caller must
2397 * have set the device and priority and built the buffer before calling
2398 * this function. The function can be called from an interrupt.
2400 * A negative errno code is returned on a failure. A success does not
2401 * guarantee the frame will be transmitted as it may be dropped due
2402 * to congestion or traffic shaping.
2404 * -----------------------------------------------------------------------------------
2405 * I notice this method can also return errors from the queue disciplines,
2406 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2409 * Regardless of the return value, the skb is consumed, so it is currently
2410 * difficult to retry a send to this method. (You can bump the ref count
2411 * before sending to hold a reference for retry if you are careful.)
2413 * When calling this method, interrupts MUST be enabled. This is because
2414 * the BH enable code must have IRQs enabled so that it will not deadlock.
2417 int dev_queue_xmit(struct sk_buff
*skb
)
2419 struct net_device
*dev
= skb
->dev
;
2420 struct netdev_queue
*txq
;
2424 /* Disable soft irqs for various locks below. Also
2425 * stops preemption for RCU.
2429 txq
= dev_pick_tx(dev
, skb
);
2430 q
= rcu_dereference_bh(txq
->qdisc
);
2432 #ifdef CONFIG_NET_CLS_ACT
2433 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2435 trace_net_dev_queue(skb
);
2437 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2441 /* The device has no queue. Common case for software devices:
2442 loopback, all the sorts of tunnels...
2444 Really, it is unlikely that netif_tx_lock protection is necessary
2445 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2447 However, it is possible, that they rely on protection
2450 Check this and shot the lock. It is not prone from deadlocks.
2451 Either shot noqueue qdisc, it is even simpler 8)
2453 if (dev
->flags
& IFF_UP
) {
2454 int cpu
= smp_processor_id(); /* ok because BHs are off */
2456 if (txq
->xmit_lock_owner
!= cpu
) {
2458 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2459 goto recursion_alert
;
2461 HARD_TX_LOCK(dev
, txq
, cpu
);
2463 if (!netif_tx_queue_stopped(txq
)) {
2464 __this_cpu_inc(xmit_recursion
);
2465 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2466 __this_cpu_dec(xmit_recursion
);
2467 if (dev_xmit_complete(rc
)) {
2468 HARD_TX_UNLOCK(dev
, txq
);
2472 HARD_TX_UNLOCK(dev
, txq
);
2473 if (net_ratelimit())
2474 printk(KERN_CRIT
"Virtual device %s asks to "
2475 "queue packet!\n", dev
->name
);
2477 /* Recursion is detected! It is possible,
2481 if (net_ratelimit())
2482 printk(KERN_CRIT
"Dead loop on virtual device "
2483 "%s, fix it urgently!\n", dev
->name
);
2488 rcu_read_unlock_bh();
2493 rcu_read_unlock_bh();
2496 EXPORT_SYMBOL(dev_queue_xmit
);
2499 /*=======================================================================
2501 =======================================================================*/
2503 int netdev_max_backlog __read_mostly
= 1000;
2504 int netdev_tstamp_prequeue __read_mostly
= 1;
2505 int netdev_budget __read_mostly
= 300;
2506 int weight_p __read_mostly
= 64; /* old backlog weight */
2508 /* Called with irq disabled */
2509 static inline void ____napi_schedule(struct softnet_data
*sd
,
2510 struct napi_struct
*napi
)
2512 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2513 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2517 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2518 * and src/dst port numbers. Returns a non-zero hash number on success
2521 __u32
__skb_get_rxhash(struct sk_buff
*skb
)
2523 int nhoff
, hash
= 0, poff
;
2524 struct ipv6hdr
*ip6
;
2527 u32 addr1
, addr2
, ihl
;
2533 nhoff
= skb_network_offset(skb
);
2535 switch (skb
->protocol
) {
2536 case __constant_htons(ETH_P_IP
):
2537 if (!pskb_may_pull(skb
, sizeof(*ip
) + nhoff
))
2540 ip
= (struct iphdr
*) (skb
->data
+ nhoff
);
2541 if (ip
->frag_off
& htons(IP_MF
| IP_OFFSET
))
2544 ip_proto
= ip
->protocol
;
2545 addr1
= (__force u32
) ip
->saddr
;
2546 addr2
= (__force u32
) ip
->daddr
;
2549 case __constant_htons(ETH_P_IPV6
):
2550 if (!pskb_may_pull(skb
, sizeof(*ip6
) + nhoff
))
2553 ip6
= (struct ipv6hdr
*) (skb
->data
+ nhoff
);
2554 ip_proto
= ip6
->nexthdr
;
2555 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2556 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2564 poff
= proto_ports_offset(ip_proto
);
2566 nhoff
+= ihl
* 4 + poff
;
2567 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2568 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2569 if (ports
.v16
[1] < ports
.v16
[0])
2570 swap(ports
.v16
[0], ports
.v16
[1]);
2574 /* get a consistent hash (same value on both flow directions) */
2578 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2585 EXPORT_SYMBOL(__skb_get_rxhash
);
2589 /* One global table that all flow-based protocols share. */
2590 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2591 EXPORT_SYMBOL(rps_sock_flow_table
);
2593 static struct rps_dev_flow
*
2594 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2595 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2599 tcpu
= rflow
->cpu
= next_cpu
;
2600 if (tcpu
!= RPS_NO_CPU
) {
2601 #ifdef CONFIG_RFS_ACCEL
2602 struct netdev_rx_queue
*rxqueue
;
2603 struct rps_dev_flow_table
*flow_table
;
2604 struct rps_dev_flow
*old_rflow
;
2609 /* Should we steer this flow to a different hardware queue? */
2610 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
)
2612 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2613 if (rxq_index
== skb_get_rx_queue(skb
))
2616 rxqueue
= dev
->_rx
+ rxq_index
;
2617 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2620 flow_id
= skb
->rxhash
& flow_table
->mask
;
2621 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2622 rxq_index
, flow_id
);
2626 rflow
= &flow_table
->flows
[flow_id
];
2627 rflow
->cpu
= next_cpu
;
2629 if (old_rflow
->filter
== rflow
->filter
)
2630 old_rflow
->filter
= RPS_NO_FILTER
;
2634 per_cpu(softnet_data
, tcpu
).input_queue_head
;
2641 * get_rps_cpu is called from netif_receive_skb and returns the target
2642 * CPU from the RPS map of the receiving queue for a given skb.
2643 * rcu_read_lock must be held on entry.
2645 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2646 struct rps_dev_flow
**rflowp
)
2648 struct netdev_rx_queue
*rxqueue
;
2649 struct rps_map
*map
;
2650 struct rps_dev_flow_table
*flow_table
;
2651 struct rps_sock_flow_table
*sock_flow_table
;
2655 if (skb_rx_queue_recorded(skb
)) {
2656 u16 index
= skb_get_rx_queue(skb
);
2657 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2658 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2659 "%s received packet on queue %u, but number "
2660 "of RX queues is %u\n",
2661 dev
->name
, index
, dev
->real_num_rx_queues
);
2664 rxqueue
= dev
->_rx
+ index
;
2668 map
= rcu_dereference(rxqueue
->rps_map
);
2670 if (map
->len
== 1 &&
2671 !rcu_dereference_raw(rxqueue
->rps_flow_table
)) {
2672 tcpu
= map
->cpus
[0];
2673 if (cpu_online(tcpu
))
2677 } else if (!rcu_dereference_raw(rxqueue
->rps_flow_table
)) {
2681 skb_reset_network_header(skb
);
2682 if (!skb_get_rxhash(skb
))
2685 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2686 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2687 if (flow_table
&& sock_flow_table
) {
2689 struct rps_dev_flow
*rflow
;
2691 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2694 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2695 sock_flow_table
->mask
];
2698 * If the desired CPU (where last recvmsg was done) is
2699 * different from current CPU (one in the rx-queue flow
2700 * table entry), switch if one of the following holds:
2701 * - Current CPU is unset (equal to RPS_NO_CPU).
2702 * - Current CPU is offline.
2703 * - The current CPU's queue tail has advanced beyond the
2704 * last packet that was enqueued using this table entry.
2705 * This guarantees that all previous packets for the flow
2706 * have been dequeued, thus preserving in order delivery.
2708 if (unlikely(tcpu
!= next_cpu
) &&
2709 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2710 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2711 rflow
->last_qtail
)) >= 0))
2712 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2714 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2722 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2724 if (cpu_online(tcpu
)) {
2734 #ifdef CONFIG_RFS_ACCEL
2737 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2738 * @dev: Device on which the filter was set
2739 * @rxq_index: RX queue index
2740 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2741 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2743 * Drivers that implement ndo_rx_flow_steer() should periodically call
2744 * this function for each installed filter and remove the filters for
2745 * which it returns %true.
2747 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2748 u32 flow_id
, u16 filter_id
)
2750 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2751 struct rps_dev_flow_table
*flow_table
;
2752 struct rps_dev_flow
*rflow
;
2757 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2758 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2759 rflow
= &flow_table
->flows
[flow_id
];
2760 cpu
= ACCESS_ONCE(rflow
->cpu
);
2761 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2762 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2763 rflow
->last_qtail
) <
2764 (int)(10 * flow_table
->mask
)))
2770 EXPORT_SYMBOL(rps_may_expire_flow
);
2772 #endif /* CONFIG_RFS_ACCEL */
2774 /* Called from hardirq (IPI) context */
2775 static void rps_trigger_softirq(void *data
)
2777 struct softnet_data
*sd
= data
;
2779 ____napi_schedule(sd
, &sd
->backlog
);
2783 #endif /* CONFIG_RPS */
2786 * Check if this softnet_data structure is another cpu one
2787 * If yes, queue it to our IPI list and return 1
2790 static int rps_ipi_queued(struct softnet_data
*sd
)
2793 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2796 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2797 mysd
->rps_ipi_list
= sd
;
2799 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2802 #endif /* CONFIG_RPS */
2807 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2808 * queue (may be a remote CPU queue).
2810 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2811 unsigned int *qtail
)
2813 struct softnet_data
*sd
;
2814 unsigned long flags
;
2816 sd
= &per_cpu(softnet_data
, cpu
);
2818 local_irq_save(flags
);
2821 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2822 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2824 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2825 input_queue_tail_incr_save(sd
, qtail
);
2827 local_irq_restore(flags
);
2828 return NET_RX_SUCCESS
;
2831 /* Schedule NAPI for backlog device
2832 * We can use non atomic operation since we own the queue lock
2834 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2835 if (!rps_ipi_queued(sd
))
2836 ____napi_schedule(sd
, &sd
->backlog
);
2844 local_irq_restore(flags
);
2846 atomic_long_inc(&skb
->dev
->rx_dropped
);
2852 * netif_rx - post buffer to the network code
2853 * @skb: buffer to post
2855 * This function receives a packet from a device driver and queues it for
2856 * the upper (protocol) levels to process. It always succeeds. The buffer
2857 * may be dropped during processing for congestion control or by the
2861 * NET_RX_SUCCESS (no congestion)
2862 * NET_RX_DROP (packet was dropped)
2866 int netif_rx(struct sk_buff
*skb
)
2870 /* if netpoll wants it, pretend we never saw it */
2871 if (netpoll_rx(skb
))
2874 if (netdev_tstamp_prequeue
)
2875 net_timestamp_check(skb
);
2877 trace_netif_rx(skb
);
2880 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2886 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2888 cpu
= smp_processor_id();
2890 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2898 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2904 EXPORT_SYMBOL(netif_rx
);
2906 int netif_rx_ni(struct sk_buff
*skb
)
2911 err
= netif_rx(skb
);
2912 if (local_softirq_pending())
2918 EXPORT_SYMBOL(netif_rx_ni
);
2920 static void net_tx_action(struct softirq_action
*h
)
2922 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2924 if (sd
->completion_queue
) {
2925 struct sk_buff
*clist
;
2927 local_irq_disable();
2928 clist
= sd
->completion_queue
;
2929 sd
->completion_queue
= NULL
;
2933 struct sk_buff
*skb
= clist
;
2934 clist
= clist
->next
;
2936 WARN_ON(atomic_read(&skb
->users
));
2937 trace_kfree_skb(skb
, net_tx_action
);
2942 if (sd
->output_queue
) {
2945 local_irq_disable();
2946 head
= sd
->output_queue
;
2947 sd
->output_queue
= NULL
;
2948 sd
->output_queue_tailp
= &sd
->output_queue
;
2952 struct Qdisc
*q
= head
;
2953 spinlock_t
*root_lock
;
2955 head
= head
->next_sched
;
2957 root_lock
= qdisc_lock(q
);
2958 if (spin_trylock(root_lock
)) {
2959 smp_mb__before_clear_bit();
2960 clear_bit(__QDISC_STATE_SCHED
,
2963 spin_unlock(root_lock
);
2965 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2967 __netif_reschedule(q
);
2969 smp_mb__before_clear_bit();
2970 clear_bit(__QDISC_STATE_SCHED
,
2978 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2979 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2980 /* This hook is defined here for ATM LANE */
2981 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2982 unsigned char *addr
) __read_mostly
;
2983 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2986 #ifdef CONFIG_NET_CLS_ACT
2987 /* TODO: Maybe we should just force sch_ingress to be compiled in
2988 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2989 * a compare and 2 stores extra right now if we dont have it on
2990 * but have CONFIG_NET_CLS_ACT
2991 * NOTE: This doesnt stop any functionality; if you dont have
2992 * the ingress scheduler, you just cant add policies on ingress.
2995 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
2997 struct net_device
*dev
= skb
->dev
;
2998 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2999 int result
= TC_ACT_OK
;
3002 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3003 if (net_ratelimit())
3004 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3005 skb
->skb_iif
, dev
->ifindex
);
3009 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3010 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3013 if (q
!= &noop_qdisc
) {
3014 spin_lock(qdisc_lock(q
));
3015 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3016 result
= qdisc_enqueue_root(skb
, q
);
3017 spin_unlock(qdisc_lock(q
));
3023 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3024 struct packet_type
**pt_prev
,
3025 int *ret
, struct net_device
*orig_dev
)
3027 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3029 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3033 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3037 switch (ing_filter(skb
, rxq
)) {
3051 * netdev_rx_handler_register - register receive handler
3052 * @dev: device to register a handler for
3053 * @rx_handler: receive handler to register
3054 * @rx_handler_data: data pointer that is used by rx handler
3056 * Register a receive hander for a device. This handler will then be
3057 * called from __netif_receive_skb. A negative errno code is returned
3060 * The caller must hold the rtnl_mutex.
3062 int netdev_rx_handler_register(struct net_device
*dev
,
3063 rx_handler_func_t
*rx_handler
,
3064 void *rx_handler_data
)
3068 if (dev
->rx_handler
)
3071 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3072 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3076 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3079 * netdev_rx_handler_unregister - unregister receive handler
3080 * @dev: device to unregister a handler from
3082 * Unregister a receive hander from a device.
3084 * The caller must hold the rtnl_mutex.
3086 void netdev_rx_handler_unregister(struct net_device
*dev
)
3090 rcu_assign_pointer(dev
->rx_handler
, NULL
);
3091 rcu_assign_pointer(dev
->rx_handler_data
, NULL
);
3093 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3095 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
3096 struct net_device
*master
)
3098 if (skb
->pkt_type
== PACKET_HOST
) {
3099 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
3101 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
3105 /* On bonding slaves other than the currently active slave, suppress
3106 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
3107 * ARP on active-backup slaves with arp_validate enabled.
3109 static int __skb_bond_should_drop(struct sk_buff
*skb
,
3110 struct net_device
*master
)
3112 struct net_device
*dev
= skb
->dev
;
3114 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
3115 dev
->last_rx
= jiffies
;
3117 if ((master
->priv_flags
& IFF_MASTER_ALB
) &&
3118 (master
->priv_flags
& IFF_BRIDGE_PORT
)) {
3119 /* Do address unmangle. The local destination address
3120 * will be always the one master has. Provides the right
3121 * functionality in a bridge.
3123 skb_bond_set_mac_by_master(skb
, master
);
3126 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
3127 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
3128 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
3131 if (master
->priv_flags
& IFF_MASTER_ALB
) {
3132 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
3133 skb
->pkt_type
!= PACKET_MULTICAST
)
3136 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
3137 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
3145 static int __netif_receive_skb(struct sk_buff
*skb
)
3147 struct packet_type
*ptype
, *pt_prev
;
3148 rx_handler_func_t
*rx_handler
;
3149 struct net_device
*orig_dev
;
3150 struct net_device
*null_or_orig
;
3151 struct net_device
*orig_or_bond
;
3152 int ret
= NET_RX_DROP
;
3155 if (!netdev_tstamp_prequeue
)
3156 net_timestamp_check(skb
);
3158 trace_netif_receive_skb(skb
);
3160 /* if we've gotten here through NAPI, check netpoll */
3161 if (netpoll_receive_skb(skb
))
3165 skb
->skb_iif
= skb
->dev
->ifindex
;
3168 * bonding note: skbs received on inactive slaves should only
3169 * be delivered to pkt handlers that are exact matches. Also
3170 * the deliver_no_wcard flag will be set. If packet handlers
3171 * are sensitive to duplicate packets these skbs will need to
3172 * be dropped at the handler.
3174 null_or_orig
= NULL
;
3175 orig_dev
= skb
->dev
;
3176 if (skb
->deliver_no_wcard
)
3177 null_or_orig
= orig_dev
;
3178 else if (netif_is_bond_slave(orig_dev
)) {
3179 struct net_device
*bond_master
= ACCESS_ONCE(orig_dev
->master
);
3181 if (likely(bond_master
)) {
3182 if (__skb_bond_should_drop(skb
, bond_master
)) {
3183 skb
->deliver_no_wcard
= 1;
3184 /* deliver only exact match */
3185 null_or_orig
= orig_dev
;
3187 skb
->dev
= bond_master
;
3191 __this_cpu_inc(softnet_data
.processed
);
3192 skb_reset_network_header(skb
);
3193 skb_reset_transport_header(skb
);
3194 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
3200 #ifdef CONFIG_NET_CLS_ACT
3201 if (skb
->tc_verd
& TC_NCLS
) {
3202 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3207 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3208 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
3209 ptype
->dev
== orig_dev
) {
3211 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3216 #ifdef CONFIG_NET_CLS_ACT
3217 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3223 /* Handle special case of bridge or macvlan */
3224 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3227 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3230 skb
= rx_handler(skb
);
3235 if (vlan_tx_tag_present(skb
)) {
3237 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3240 if (vlan_hwaccel_do_receive(&skb
)) {
3241 ret
= __netif_receive_skb(skb
);
3243 } else if (unlikely(!skb
))
3248 * Make sure frames received on VLAN interfaces stacked on
3249 * bonding interfaces still make their way to any base bonding
3250 * device that may have registered for a specific ptype. The
3251 * handler may have to adjust skb->dev and orig_dev.
3253 orig_or_bond
= orig_dev
;
3254 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
3255 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
3256 orig_or_bond
= vlan_dev_real_dev(skb
->dev
);
3259 type
= skb
->protocol
;
3260 list_for_each_entry_rcu(ptype
,
3261 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3262 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
3263 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
3264 ptype
->dev
== orig_or_bond
)) {
3266 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3272 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3274 atomic_long_inc(&skb
->dev
->rx_dropped
);
3276 /* Jamal, now you will not able to escape explaining
3277 * me how you were going to use this. :-)
3288 * netif_receive_skb - process receive buffer from network
3289 * @skb: buffer to process
3291 * netif_receive_skb() is the main receive data processing function.
3292 * It always succeeds. The buffer may be dropped during processing
3293 * for congestion control or by the protocol layers.
3295 * This function may only be called from softirq context and interrupts
3296 * should be enabled.
3298 * Return values (usually ignored):
3299 * NET_RX_SUCCESS: no congestion
3300 * NET_RX_DROP: packet was dropped
3302 int netif_receive_skb(struct sk_buff
*skb
)
3304 if (netdev_tstamp_prequeue
)
3305 net_timestamp_check(skb
);
3307 if (skb_defer_rx_timestamp(skb
))
3308 return NET_RX_SUCCESS
;
3312 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3317 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3320 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3324 ret
= __netif_receive_skb(skb
);
3330 return __netif_receive_skb(skb
);
3333 EXPORT_SYMBOL(netif_receive_skb
);
3335 /* Network device is going away, flush any packets still pending
3336 * Called with irqs disabled.
3338 static void flush_backlog(void *arg
)
3340 struct net_device
*dev
= arg
;
3341 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3342 struct sk_buff
*skb
, *tmp
;
3345 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3346 if (skb
->dev
== dev
) {
3347 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3349 input_queue_head_incr(sd
);
3354 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3355 if (skb
->dev
== dev
) {
3356 __skb_unlink(skb
, &sd
->process_queue
);
3358 input_queue_head_incr(sd
);
3363 static int napi_gro_complete(struct sk_buff
*skb
)
3365 struct packet_type
*ptype
;
3366 __be16 type
= skb
->protocol
;
3367 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3370 if (NAPI_GRO_CB(skb
)->count
== 1) {
3371 skb_shinfo(skb
)->gso_size
= 0;
3376 list_for_each_entry_rcu(ptype
, head
, list
) {
3377 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3380 err
= ptype
->gro_complete(skb
);
3386 WARN_ON(&ptype
->list
== head
);
3388 return NET_RX_SUCCESS
;
3392 return netif_receive_skb(skb
);
3395 inline void napi_gro_flush(struct napi_struct
*napi
)
3397 struct sk_buff
*skb
, *next
;
3399 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3402 napi_gro_complete(skb
);
3405 napi
->gro_count
= 0;
3406 napi
->gro_list
= NULL
;
3408 EXPORT_SYMBOL(napi_gro_flush
);
3410 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3412 struct sk_buff
**pp
= NULL
;
3413 struct packet_type
*ptype
;
3414 __be16 type
= skb
->protocol
;
3415 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3418 enum gro_result ret
;
3420 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3423 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3427 list_for_each_entry_rcu(ptype
, head
, list
) {
3428 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3431 skb_set_network_header(skb
, skb_gro_offset(skb
));
3432 mac_len
= skb
->network_header
- skb
->mac_header
;
3433 skb
->mac_len
= mac_len
;
3434 NAPI_GRO_CB(skb
)->same_flow
= 0;
3435 NAPI_GRO_CB(skb
)->flush
= 0;
3436 NAPI_GRO_CB(skb
)->free
= 0;
3438 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3443 if (&ptype
->list
== head
)
3446 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3447 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3450 struct sk_buff
*nskb
= *pp
;
3454 napi_gro_complete(nskb
);
3461 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3465 NAPI_GRO_CB(skb
)->count
= 1;
3466 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3467 skb
->next
= napi
->gro_list
;
3468 napi
->gro_list
= skb
;
3472 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3473 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3475 BUG_ON(skb
->end
- skb
->tail
< grow
);
3477 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3480 skb
->data_len
-= grow
;
3482 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3483 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3485 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3486 put_page(skb_shinfo(skb
)->frags
[0].page
);
3487 memmove(skb_shinfo(skb
)->frags
,
3488 skb_shinfo(skb
)->frags
+ 1,
3489 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3500 EXPORT_SYMBOL(dev_gro_receive
);
3502 static inline gro_result_t
3503 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3507 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3508 unsigned long diffs
;
3510 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3511 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3512 diffs
|= compare_ether_header(skb_mac_header(p
),
3513 skb_gro_mac_header(skb
));
3514 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3515 NAPI_GRO_CB(p
)->flush
= 0;
3518 return dev_gro_receive(napi
, skb
);
3521 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3525 if (netif_receive_skb(skb
))
3530 case GRO_MERGED_FREE
:
3541 EXPORT_SYMBOL(napi_skb_finish
);
3543 void skb_gro_reset_offset(struct sk_buff
*skb
)
3545 NAPI_GRO_CB(skb
)->data_offset
= 0;
3546 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3547 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3549 if (skb
->mac_header
== skb
->tail
&&
3550 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3551 NAPI_GRO_CB(skb
)->frag0
=
3552 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3553 skb_shinfo(skb
)->frags
[0].page_offset
;
3554 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3557 EXPORT_SYMBOL(skb_gro_reset_offset
);
3559 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3561 skb_gro_reset_offset(skb
);
3563 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3565 EXPORT_SYMBOL(napi_gro_receive
);
3567 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3569 __skb_pull(skb
, skb_headlen(skb
));
3570 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3572 skb
->dev
= napi
->dev
;
3578 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3580 struct sk_buff
*skb
= napi
->skb
;
3583 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3589 EXPORT_SYMBOL(napi_get_frags
);
3591 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3597 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3599 if (ret
== GRO_HELD
)
3600 skb_gro_pull(skb
, -ETH_HLEN
);
3601 else if (netif_receive_skb(skb
))
3606 case GRO_MERGED_FREE
:
3607 napi_reuse_skb(napi
, skb
);
3616 EXPORT_SYMBOL(napi_frags_finish
);
3618 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3620 struct sk_buff
*skb
= napi
->skb
;
3627 skb_reset_mac_header(skb
);
3628 skb_gro_reset_offset(skb
);
3630 off
= skb_gro_offset(skb
);
3631 hlen
= off
+ sizeof(*eth
);
3632 eth
= skb_gro_header_fast(skb
, off
);
3633 if (skb_gro_header_hard(skb
, hlen
)) {
3634 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3635 if (unlikely(!eth
)) {
3636 napi_reuse_skb(napi
, skb
);
3642 skb_gro_pull(skb
, sizeof(*eth
));
3645 * This works because the only protocols we care about don't require
3646 * special handling. We'll fix it up properly at the end.
3648 skb
->protocol
= eth
->h_proto
;
3653 EXPORT_SYMBOL(napi_frags_skb
);
3655 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3657 struct sk_buff
*skb
= napi_frags_skb(napi
);
3662 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3664 EXPORT_SYMBOL(napi_gro_frags
);
3667 * net_rps_action sends any pending IPI's for rps.
3668 * Note: called with local irq disabled, but exits with local irq enabled.
3670 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3673 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3676 sd
->rps_ipi_list
= NULL
;
3680 /* Send pending IPI's to kick RPS processing on remote cpus. */
3682 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3684 if (cpu_online(remsd
->cpu
))
3685 __smp_call_function_single(remsd
->cpu
,
3694 static int process_backlog(struct napi_struct
*napi
, int quota
)
3697 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3700 /* Check if we have pending ipi, its better to send them now,
3701 * not waiting net_rx_action() end.
3703 if (sd
->rps_ipi_list
) {
3704 local_irq_disable();
3705 net_rps_action_and_irq_enable(sd
);
3708 napi
->weight
= weight_p
;
3709 local_irq_disable();
3710 while (work
< quota
) {
3711 struct sk_buff
*skb
;
3714 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3716 __netif_receive_skb(skb
);
3717 local_irq_disable();
3718 input_queue_head_incr(sd
);
3719 if (++work
>= quota
) {
3726 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3728 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3729 &sd
->process_queue
);
3731 if (qlen
< quota
- work
) {
3733 * Inline a custom version of __napi_complete().
3734 * only current cpu owns and manipulates this napi,
3735 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3736 * we can use a plain write instead of clear_bit(),
3737 * and we dont need an smp_mb() memory barrier.
3739 list_del(&napi
->poll_list
);
3742 quota
= work
+ qlen
;
3752 * __napi_schedule - schedule for receive
3753 * @n: entry to schedule
3755 * The entry's receive function will be scheduled to run
3757 void __napi_schedule(struct napi_struct
*n
)
3759 unsigned long flags
;
3761 local_irq_save(flags
);
3762 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3763 local_irq_restore(flags
);
3765 EXPORT_SYMBOL(__napi_schedule
);
3767 void __napi_complete(struct napi_struct
*n
)
3769 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3770 BUG_ON(n
->gro_list
);
3772 list_del(&n
->poll_list
);
3773 smp_mb__before_clear_bit();
3774 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3776 EXPORT_SYMBOL(__napi_complete
);
3778 void napi_complete(struct napi_struct
*n
)
3780 unsigned long flags
;
3783 * don't let napi dequeue from the cpu poll list
3784 * just in case its running on a different cpu
3786 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3790 local_irq_save(flags
);
3792 local_irq_restore(flags
);
3794 EXPORT_SYMBOL(napi_complete
);
3796 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3797 int (*poll
)(struct napi_struct
*, int), int weight
)
3799 INIT_LIST_HEAD(&napi
->poll_list
);
3800 napi
->gro_count
= 0;
3801 napi
->gro_list
= NULL
;
3804 napi
->weight
= weight
;
3805 list_add(&napi
->dev_list
, &dev
->napi_list
);
3807 #ifdef CONFIG_NETPOLL
3808 spin_lock_init(&napi
->poll_lock
);
3809 napi
->poll_owner
= -1;
3811 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3813 EXPORT_SYMBOL(netif_napi_add
);
3815 void netif_napi_del(struct napi_struct
*napi
)
3817 struct sk_buff
*skb
, *next
;
3819 list_del_init(&napi
->dev_list
);
3820 napi_free_frags(napi
);
3822 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3828 napi
->gro_list
= NULL
;
3829 napi
->gro_count
= 0;
3831 EXPORT_SYMBOL(netif_napi_del
);
3833 static void net_rx_action(struct softirq_action
*h
)
3835 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3836 unsigned long time_limit
= jiffies
+ 2;
3837 int budget
= netdev_budget
;
3840 local_irq_disable();
3842 while (!list_empty(&sd
->poll_list
)) {
3843 struct napi_struct
*n
;
3846 /* If softirq window is exhuasted then punt.
3847 * Allow this to run for 2 jiffies since which will allow
3848 * an average latency of 1.5/HZ.
3850 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3855 /* Even though interrupts have been re-enabled, this
3856 * access is safe because interrupts can only add new
3857 * entries to the tail of this list, and only ->poll()
3858 * calls can remove this head entry from the list.
3860 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3862 have
= netpoll_poll_lock(n
);
3866 /* This NAPI_STATE_SCHED test is for avoiding a race
3867 * with netpoll's poll_napi(). Only the entity which
3868 * obtains the lock and sees NAPI_STATE_SCHED set will
3869 * actually make the ->poll() call. Therefore we avoid
3870 * accidently calling ->poll() when NAPI is not scheduled.
3873 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3874 work
= n
->poll(n
, weight
);
3878 WARN_ON_ONCE(work
> weight
);
3882 local_irq_disable();
3884 /* Drivers must not modify the NAPI state if they
3885 * consume the entire weight. In such cases this code
3886 * still "owns" the NAPI instance and therefore can
3887 * move the instance around on the list at-will.
3889 if (unlikely(work
== weight
)) {
3890 if (unlikely(napi_disable_pending(n
))) {
3893 local_irq_disable();
3895 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3898 netpoll_poll_unlock(have
);
3901 net_rps_action_and_irq_enable(sd
);
3903 #ifdef CONFIG_NET_DMA
3905 * There may not be any more sk_buffs coming right now, so push
3906 * any pending DMA copies to hardware
3908 dma_issue_pending_all();
3915 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3919 static gifconf_func_t
*gifconf_list
[NPROTO
];
3922 * register_gifconf - register a SIOCGIF handler
3923 * @family: Address family
3924 * @gifconf: Function handler
3926 * Register protocol dependent address dumping routines. The handler
3927 * that is passed must not be freed or reused until it has been replaced
3928 * by another handler.
3930 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3932 if (family
>= NPROTO
)
3934 gifconf_list
[family
] = gifconf
;
3937 EXPORT_SYMBOL(register_gifconf
);
3941 * Map an interface index to its name (SIOCGIFNAME)
3945 * We need this ioctl for efficient implementation of the
3946 * if_indextoname() function required by the IPv6 API. Without
3947 * it, we would have to search all the interfaces to find a
3951 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3953 struct net_device
*dev
;
3957 * Fetch the caller's info block.
3960 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3964 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3970 strcpy(ifr
.ifr_name
, dev
->name
);
3973 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3979 * Perform a SIOCGIFCONF call. This structure will change
3980 * size eventually, and there is nothing I can do about it.
3981 * Thus we will need a 'compatibility mode'.
3984 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3987 struct net_device
*dev
;
3994 * Fetch the caller's info block.
3997 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
4004 * Loop over the interfaces, and write an info block for each.
4008 for_each_netdev(net
, dev
) {
4009 for (i
= 0; i
< NPROTO
; i
++) {
4010 if (gifconf_list
[i
]) {
4013 done
= gifconf_list
[i
](dev
, NULL
, 0);
4015 done
= gifconf_list
[i
](dev
, pos
+ total
,
4025 * All done. Write the updated control block back to the caller.
4027 ifc
.ifc_len
= total
;
4030 * Both BSD and Solaris return 0 here, so we do too.
4032 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4035 #ifdef CONFIG_PROC_FS
4037 * This is invoked by the /proc filesystem handler to display a device
4040 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4043 struct net
*net
= seq_file_net(seq
);
4045 struct net_device
*dev
;
4049 return SEQ_START_TOKEN
;
4052 for_each_netdev_rcu(net
, dev
)
4059 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4061 struct net_device
*dev
= v
;
4063 if (v
== SEQ_START_TOKEN
)
4064 dev
= first_net_device_rcu(seq_file_net(seq
));
4066 dev
= next_net_device_rcu(dev
);
4072 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4078 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4080 struct rtnl_link_stats64 temp
;
4081 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4083 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4084 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4085 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4087 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4088 stats
->rx_fifo_errors
,
4089 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4090 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4091 stats
->rx_compressed
, stats
->multicast
,
4092 stats
->tx_bytes
, stats
->tx_packets
,
4093 stats
->tx_errors
, stats
->tx_dropped
,
4094 stats
->tx_fifo_errors
, stats
->collisions
,
4095 stats
->tx_carrier_errors
+
4096 stats
->tx_aborted_errors
+
4097 stats
->tx_window_errors
+
4098 stats
->tx_heartbeat_errors
,
4099 stats
->tx_compressed
);
4103 * Called from the PROCfs module. This now uses the new arbitrary sized
4104 * /proc/net interface to create /proc/net/dev
4106 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4108 if (v
== SEQ_START_TOKEN
)
4109 seq_puts(seq
, "Inter-| Receive "
4111 " face |bytes packets errs drop fifo frame "
4112 "compressed multicast|bytes packets errs "
4113 "drop fifo colls carrier compressed\n");
4115 dev_seq_printf_stats(seq
, v
);
4119 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4121 struct softnet_data
*sd
= NULL
;
4123 while (*pos
< nr_cpu_ids
)
4124 if (cpu_online(*pos
)) {
4125 sd
= &per_cpu(softnet_data
, *pos
);
4132 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4134 return softnet_get_online(pos
);
4137 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4140 return softnet_get_online(pos
);
4143 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4147 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4149 struct softnet_data
*sd
= v
;
4151 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4152 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4153 0, 0, 0, 0, /* was fastroute */
4154 sd
->cpu_collision
, sd
->received_rps
);
4158 static const struct seq_operations dev_seq_ops
= {
4159 .start
= dev_seq_start
,
4160 .next
= dev_seq_next
,
4161 .stop
= dev_seq_stop
,
4162 .show
= dev_seq_show
,
4165 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4167 return seq_open_net(inode
, file
, &dev_seq_ops
,
4168 sizeof(struct seq_net_private
));
4171 static const struct file_operations dev_seq_fops
= {
4172 .owner
= THIS_MODULE
,
4173 .open
= dev_seq_open
,
4175 .llseek
= seq_lseek
,
4176 .release
= seq_release_net
,
4179 static const struct seq_operations softnet_seq_ops
= {
4180 .start
= softnet_seq_start
,
4181 .next
= softnet_seq_next
,
4182 .stop
= softnet_seq_stop
,
4183 .show
= softnet_seq_show
,
4186 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4188 return seq_open(file
, &softnet_seq_ops
);
4191 static const struct file_operations softnet_seq_fops
= {
4192 .owner
= THIS_MODULE
,
4193 .open
= softnet_seq_open
,
4195 .llseek
= seq_lseek
,
4196 .release
= seq_release
,
4199 static void *ptype_get_idx(loff_t pos
)
4201 struct packet_type
*pt
= NULL
;
4205 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4211 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4212 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4221 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4225 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4228 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4230 struct packet_type
*pt
;
4231 struct list_head
*nxt
;
4235 if (v
== SEQ_START_TOKEN
)
4236 return ptype_get_idx(0);
4239 nxt
= pt
->list
.next
;
4240 if (pt
->type
== htons(ETH_P_ALL
)) {
4241 if (nxt
!= &ptype_all
)
4244 nxt
= ptype_base
[0].next
;
4246 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4248 while (nxt
== &ptype_base
[hash
]) {
4249 if (++hash
>= PTYPE_HASH_SIZE
)
4251 nxt
= ptype_base
[hash
].next
;
4254 return list_entry(nxt
, struct packet_type
, list
);
4257 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4263 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4265 struct packet_type
*pt
= v
;
4267 if (v
== SEQ_START_TOKEN
)
4268 seq_puts(seq
, "Type Device Function\n");
4269 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4270 if (pt
->type
== htons(ETH_P_ALL
))
4271 seq_puts(seq
, "ALL ");
4273 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4275 seq_printf(seq
, " %-8s %pF\n",
4276 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4282 static const struct seq_operations ptype_seq_ops
= {
4283 .start
= ptype_seq_start
,
4284 .next
= ptype_seq_next
,
4285 .stop
= ptype_seq_stop
,
4286 .show
= ptype_seq_show
,
4289 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4291 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4292 sizeof(struct seq_net_private
));
4295 static const struct file_operations ptype_seq_fops
= {
4296 .owner
= THIS_MODULE
,
4297 .open
= ptype_seq_open
,
4299 .llseek
= seq_lseek
,
4300 .release
= seq_release_net
,
4304 static int __net_init
dev_proc_net_init(struct net
*net
)
4308 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4310 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4312 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4315 if (wext_proc_init(net
))
4321 proc_net_remove(net
, "ptype");
4323 proc_net_remove(net
, "softnet_stat");
4325 proc_net_remove(net
, "dev");
4329 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4331 wext_proc_exit(net
);
4333 proc_net_remove(net
, "ptype");
4334 proc_net_remove(net
, "softnet_stat");
4335 proc_net_remove(net
, "dev");
4338 static struct pernet_operations __net_initdata dev_proc_ops
= {
4339 .init
= dev_proc_net_init
,
4340 .exit
= dev_proc_net_exit
,
4343 static int __init
dev_proc_init(void)
4345 return register_pernet_subsys(&dev_proc_ops
);
4348 #define dev_proc_init() 0
4349 #endif /* CONFIG_PROC_FS */
4353 * netdev_set_master - set up master pointer
4354 * @slave: slave device
4355 * @master: new master device
4357 * Changes the master device of the slave. Pass %NULL to break the
4358 * bonding. The caller must hold the RTNL semaphore. On a failure
4359 * a negative errno code is returned. On success the reference counts
4360 * are adjusted and the function returns zero.
4362 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4364 struct net_device
*old
= slave
->master
;
4374 slave
->master
= master
;
4382 EXPORT_SYMBOL(netdev_set_master
);
4385 * netdev_set_bond_master - set up bonding master/slave pair
4386 * @slave: slave device
4387 * @master: new master device
4389 * Changes the master device of the slave. Pass %NULL to break the
4390 * bonding. The caller must hold the RTNL semaphore. On a failure
4391 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4392 * to the routing socket and the function returns zero.
4394 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4400 err
= netdev_set_master(slave
, master
);
4404 slave
->flags
|= IFF_SLAVE
;
4406 slave
->flags
&= ~IFF_SLAVE
;
4408 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4411 EXPORT_SYMBOL(netdev_set_bond_master
);
4413 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4415 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4417 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4418 ops
->ndo_change_rx_flags(dev
, flags
);
4421 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4423 unsigned short old_flags
= dev
->flags
;
4429 dev
->flags
|= IFF_PROMISC
;
4430 dev
->promiscuity
+= inc
;
4431 if (dev
->promiscuity
== 0) {
4434 * If inc causes overflow, untouch promisc and return error.
4437 dev
->flags
&= ~IFF_PROMISC
;
4439 dev
->promiscuity
-= inc
;
4440 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4441 "set promiscuity failed, promiscuity feature "
4442 "of device might be broken.\n", dev
->name
);
4446 if (dev
->flags
!= old_flags
) {
4447 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4448 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4450 if (audit_enabled
) {
4451 current_uid_gid(&uid
, &gid
);
4452 audit_log(current
->audit_context
, GFP_ATOMIC
,
4453 AUDIT_ANOM_PROMISCUOUS
,
4454 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4455 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4456 (old_flags
& IFF_PROMISC
),
4457 audit_get_loginuid(current
),
4459 audit_get_sessionid(current
));
4462 dev_change_rx_flags(dev
, IFF_PROMISC
);
4468 * dev_set_promiscuity - update promiscuity count on a device
4472 * Add or remove promiscuity from a device. While the count in the device
4473 * remains above zero the interface remains promiscuous. Once it hits zero
4474 * the device reverts back to normal filtering operation. A negative inc
4475 * value is used to drop promiscuity on the device.
4476 * Return 0 if successful or a negative errno code on error.
4478 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4480 unsigned short old_flags
= dev
->flags
;
4483 err
= __dev_set_promiscuity(dev
, inc
);
4486 if (dev
->flags
!= old_flags
)
4487 dev_set_rx_mode(dev
);
4490 EXPORT_SYMBOL(dev_set_promiscuity
);
4493 * dev_set_allmulti - update allmulti count on a device
4497 * Add or remove reception of all multicast frames to a device. While the
4498 * count in the device remains above zero the interface remains listening
4499 * to all interfaces. Once it hits zero the device reverts back to normal
4500 * filtering operation. A negative @inc value is used to drop the counter
4501 * when releasing a resource needing all multicasts.
4502 * Return 0 if successful or a negative errno code on error.
4505 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4507 unsigned short old_flags
= dev
->flags
;
4511 dev
->flags
|= IFF_ALLMULTI
;
4512 dev
->allmulti
+= inc
;
4513 if (dev
->allmulti
== 0) {
4516 * If inc causes overflow, untouch allmulti and return error.
4519 dev
->flags
&= ~IFF_ALLMULTI
;
4521 dev
->allmulti
-= inc
;
4522 printk(KERN_WARNING
"%s: allmulti touches roof, "
4523 "set allmulti failed, allmulti feature of "
4524 "device might be broken.\n", dev
->name
);
4528 if (dev
->flags
^ old_flags
) {
4529 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4530 dev_set_rx_mode(dev
);
4534 EXPORT_SYMBOL(dev_set_allmulti
);
4537 * Upload unicast and multicast address lists to device and
4538 * configure RX filtering. When the device doesn't support unicast
4539 * filtering it is put in promiscuous mode while unicast addresses
4542 void __dev_set_rx_mode(struct net_device
*dev
)
4544 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4546 /* dev_open will call this function so the list will stay sane. */
4547 if (!(dev
->flags
&IFF_UP
))
4550 if (!netif_device_present(dev
))
4553 if (ops
->ndo_set_rx_mode
)
4554 ops
->ndo_set_rx_mode(dev
);
4556 /* Unicast addresses changes may only happen under the rtnl,
4557 * therefore calling __dev_set_promiscuity here is safe.
4559 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4560 __dev_set_promiscuity(dev
, 1);
4561 dev
->uc_promisc
= 1;
4562 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4563 __dev_set_promiscuity(dev
, -1);
4564 dev
->uc_promisc
= 0;
4567 if (ops
->ndo_set_multicast_list
)
4568 ops
->ndo_set_multicast_list(dev
);
4572 void dev_set_rx_mode(struct net_device
*dev
)
4574 netif_addr_lock_bh(dev
);
4575 __dev_set_rx_mode(dev
);
4576 netif_addr_unlock_bh(dev
);
4580 * dev_get_flags - get flags reported to userspace
4583 * Get the combination of flag bits exported through APIs to userspace.
4585 unsigned dev_get_flags(const struct net_device
*dev
)
4589 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4594 (dev
->gflags
& (IFF_PROMISC
|
4597 if (netif_running(dev
)) {
4598 if (netif_oper_up(dev
))
4599 flags
|= IFF_RUNNING
;
4600 if (netif_carrier_ok(dev
))
4601 flags
|= IFF_LOWER_UP
;
4602 if (netif_dormant(dev
))
4603 flags
|= IFF_DORMANT
;
4608 EXPORT_SYMBOL(dev_get_flags
);
4610 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4612 int old_flags
= dev
->flags
;
4618 * Set the flags on our device.
4621 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4622 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4624 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4628 * Load in the correct multicast list now the flags have changed.
4631 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4632 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4634 dev_set_rx_mode(dev
);
4637 * Have we downed the interface. We handle IFF_UP ourselves
4638 * according to user attempts to set it, rather than blindly
4643 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4644 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4647 dev_set_rx_mode(dev
);
4650 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4651 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4653 dev
->gflags
^= IFF_PROMISC
;
4654 dev_set_promiscuity(dev
, inc
);
4657 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4658 is important. Some (broken) drivers set IFF_PROMISC, when
4659 IFF_ALLMULTI is requested not asking us and not reporting.
4661 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4662 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4664 dev
->gflags
^= IFF_ALLMULTI
;
4665 dev_set_allmulti(dev
, inc
);
4671 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4673 unsigned int changes
= dev
->flags
^ old_flags
;
4675 if (changes
& IFF_UP
) {
4676 if (dev
->flags
& IFF_UP
)
4677 call_netdevice_notifiers(NETDEV_UP
, dev
);
4679 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4682 if (dev
->flags
& IFF_UP
&&
4683 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4684 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4688 * dev_change_flags - change device settings
4690 * @flags: device state flags
4692 * Change settings on device based state flags. The flags are
4693 * in the userspace exported format.
4695 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4698 int old_flags
= dev
->flags
;
4700 ret
= __dev_change_flags(dev
, flags
);
4704 changes
= old_flags
^ dev
->flags
;
4706 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4708 __dev_notify_flags(dev
, old_flags
);
4711 EXPORT_SYMBOL(dev_change_flags
);
4714 * dev_set_mtu - Change maximum transfer unit
4716 * @new_mtu: new transfer unit
4718 * Change the maximum transfer size of the network device.
4720 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4722 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4725 if (new_mtu
== dev
->mtu
)
4728 /* MTU must be positive. */
4732 if (!netif_device_present(dev
))
4736 if (ops
->ndo_change_mtu
)
4737 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4741 if (!err
&& dev
->flags
& IFF_UP
)
4742 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4745 EXPORT_SYMBOL(dev_set_mtu
);
4748 * dev_set_group - Change group this device belongs to
4750 * @new_group: group this device should belong to
4752 void dev_set_group(struct net_device
*dev
, int new_group
)
4754 dev
->group
= new_group
;
4756 EXPORT_SYMBOL(dev_set_group
);
4759 * dev_set_mac_address - Change Media Access Control Address
4763 * Change the hardware (MAC) address of the device
4765 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4767 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4770 if (!ops
->ndo_set_mac_address
)
4772 if (sa
->sa_family
!= dev
->type
)
4774 if (!netif_device_present(dev
))
4776 err
= ops
->ndo_set_mac_address(dev
, sa
);
4778 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4781 EXPORT_SYMBOL(dev_set_mac_address
);
4784 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4786 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4789 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4795 case SIOCGIFFLAGS
: /* Get interface flags */
4796 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4799 case SIOCGIFMETRIC
: /* Get the metric on the interface
4800 (currently unused) */
4801 ifr
->ifr_metric
= 0;
4804 case SIOCGIFMTU
: /* Get the MTU of a device */
4805 ifr
->ifr_mtu
= dev
->mtu
;
4810 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4812 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4813 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4814 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4822 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4823 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4824 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4825 ifr
->ifr_map
.irq
= dev
->irq
;
4826 ifr
->ifr_map
.dma
= dev
->dma
;
4827 ifr
->ifr_map
.port
= dev
->if_port
;
4831 ifr
->ifr_ifindex
= dev
->ifindex
;
4835 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4839 /* dev_ioctl() should ensure this case
4851 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4853 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4856 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4857 const struct net_device_ops
*ops
;
4862 ops
= dev
->netdev_ops
;
4865 case SIOCSIFFLAGS
: /* Set interface flags */
4866 return dev_change_flags(dev
, ifr
->ifr_flags
);
4868 case SIOCSIFMETRIC
: /* Set the metric on the interface
4869 (currently unused) */
4872 case SIOCSIFMTU
: /* Set the MTU of a device */
4873 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4876 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4878 case SIOCSIFHWBROADCAST
:
4879 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4881 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4882 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4883 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4887 if (ops
->ndo_set_config
) {
4888 if (!netif_device_present(dev
))
4890 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4895 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4896 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4898 if (!netif_device_present(dev
))
4900 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4903 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4904 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4906 if (!netif_device_present(dev
))
4908 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4911 if (ifr
->ifr_qlen
< 0)
4913 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4917 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4918 return dev_change_name(dev
, ifr
->ifr_newname
);
4921 * Unknown or private ioctl
4924 if ((cmd
>= SIOCDEVPRIVATE
&&
4925 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4926 cmd
== SIOCBONDENSLAVE
||
4927 cmd
== SIOCBONDRELEASE
||
4928 cmd
== SIOCBONDSETHWADDR
||
4929 cmd
== SIOCBONDSLAVEINFOQUERY
||
4930 cmd
== SIOCBONDINFOQUERY
||
4931 cmd
== SIOCBONDCHANGEACTIVE
||
4932 cmd
== SIOCGMIIPHY
||
4933 cmd
== SIOCGMIIREG
||
4934 cmd
== SIOCSMIIREG
||
4935 cmd
== SIOCBRADDIF
||
4936 cmd
== SIOCBRDELIF
||
4937 cmd
== SIOCSHWTSTAMP
||
4938 cmd
== SIOCWANDEV
) {
4940 if (ops
->ndo_do_ioctl
) {
4941 if (netif_device_present(dev
))
4942 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4954 * This function handles all "interface"-type I/O control requests. The actual
4955 * 'doing' part of this is dev_ifsioc above.
4959 * dev_ioctl - network device ioctl
4960 * @net: the applicable net namespace
4961 * @cmd: command to issue
4962 * @arg: pointer to a struct ifreq in user space
4964 * Issue ioctl functions to devices. This is normally called by the
4965 * user space syscall interfaces but can sometimes be useful for
4966 * other purposes. The return value is the return from the syscall if
4967 * positive or a negative errno code on error.
4970 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4976 /* One special case: SIOCGIFCONF takes ifconf argument
4977 and requires shared lock, because it sleeps writing
4981 if (cmd
== SIOCGIFCONF
) {
4983 ret
= dev_ifconf(net
, (char __user
*) arg
);
4987 if (cmd
== SIOCGIFNAME
)
4988 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4990 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4993 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4995 colon
= strchr(ifr
.ifr_name
, ':');
5000 * See which interface the caller is talking about.
5005 * These ioctl calls:
5006 * - can be done by all.
5007 * - atomic and do not require locking.
5018 dev_load(net
, ifr
.ifr_name
);
5020 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5025 if (copy_to_user(arg
, &ifr
,
5026 sizeof(struct ifreq
)))
5032 dev_load(net
, ifr
.ifr_name
);
5034 ret
= dev_ethtool(net
, &ifr
);
5039 if (copy_to_user(arg
, &ifr
,
5040 sizeof(struct ifreq
)))
5046 * These ioctl calls:
5047 * - require superuser power.
5048 * - require strict serialization.
5054 if (!capable(CAP_NET_ADMIN
))
5056 dev_load(net
, ifr
.ifr_name
);
5058 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5063 if (copy_to_user(arg
, &ifr
,
5064 sizeof(struct ifreq
)))
5070 * These ioctl calls:
5071 * - require superuser power.
5072 * - require strict serialization.
5073 * - do not return a value
5083 case SIOCSIFHWBROADCAST
:
5086 case SIOCBONDENSLAVE
:
5087 case SIOCBONDRELEASE
:
5088 case SIOCBONDSETHWADDR
:
5089 case SIOCBONDCHANGEACTIVE
:
5093 if (!capable(CAP_NET_ADMIN
))
5096 case SIOCBONDSLAVEINFOQUERY
:
5097 case SIOCBONDINFOQUERY
:
5098 dev_load(net
, ifr
.ifr_name
);
5100 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5105 /* Get the per device memory space. We can add this but
5106 * currently do not support it */
5108 /* Set the per device memory buffer space.
5109 * Not applicable in our case */
5114 * Unknown or private ioctl.
5117 if (cmd
== SIOCWANDEV
||
5118 (cmd
>= SIOCDEVPRIVATE
&&
5119 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5120 dev_load(net
, ifr
.ifr_name
);
5122 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5124 if (!ret
&& copy_to_user(arg
, &ifr
,
5125 sizeof(struct ifreq
)))
5129 /* Take care of Wireless Extensions */
5130 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5131 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5138 * dev_new_index - allocate an ifindex
5139 * @net: the applicable net namespace
5141 * Returns a suitable unique value for a new device interface
5142 * number. The caller must hold the rtnl semaphore or the
5143 * dev_base_lock to be sure it remains unique.
5145 static int dev_new_index(struct net
*net
)
5151 if (!__dev_get_by_index(net
, ifindex
))
5156 /* Delayed registration/unregisteration */
5157 static LIST_HEAD(net_todo_list
);
5159 static void net_set_todo(struct net_device
*dev
)
5161 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5164 static void rollback_registered_many(struct list_head
*head
)
5166 struct net_device
*dev
, *tmp
;
5168 BUG_ON(dev_boot_phase
);
5171 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5172 /* Some devices call without registering
5173 * for initialization unwind. Remove those
5174 * devices and proceed with the remaining.
5176 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5177 pr_debug("unregister_netdevice: device %s/%p never "
5178 "was registered\n", dev
->name
, dev
);
5181 list_del(&dev
->unreg_list
);
5185 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5188 /* If device is running, close it first. */
5189 dev_close_many(head
);
5191 list_for_each_entry(dev
, head
, unreg_list
) {
5192 /* And unlink it from device chain. */
5193 unlist_netdevice(dev
);
5195 dev
->reg_state
= NETREG_UNREGISTERING
;
5200 list_for_each_entry(dev
, head
, unreg_list
) {
5201 /* Shutdown queueing discipline. */
5205 /* Notify protocols, that we are about to destroy
5206 this device. They should clean all the things.
5208 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5210 if (!dev
->rtnl_link_ops
||
5211 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5212 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5215 * Flush the unicast and multicast chains
5220 if (dev
->netdev_ops
->ndo_uninit
)
5221 dev
->netdev_ops
->ndo_uninit(dev
);
5223 /* Notifier chain MUST detach us from master device. */
5224 WARN_ON(dev
->master
);
5226 /* Remove entries from kobject tree */
5227 netdev_unregister_kobject(dev
);
5230 /* Process any work delayed until the end of the batch */
5231 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5232 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5236 list_for_each_entry(dev
, head
, unreg_list
)
5240 static void rollback_registered(struct net_device
*dev
)
5244 list_add(&dev
->unreg_list
, &single
);
5245 rollback_registered_many(&single
);
5248 u32
netdev_fix_features(struct net_device
*dev
, u32 features
)
5250 /* Fix illegal checksum combinations */
5251 if ((features
& NETIF_F_HW_CSUM
) &&
5252 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5253 netdev_info(dev
, "mixed HW and IP checksum settings.\n");
5254 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5257 if ((features
& NETIF_F_NO_CSUM
) &&
5258 (features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5259 netdev_info(dev
, "mixed no checksumming and other settings.\n");
5260 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5263 /* Fix illegal SG+CSUM combinations. */
5264 if ((features
& NETIF_F_SG
) &&
5265 !(features
& NETIF_F_ALL_CSUM
)) {
5267 "Dropping NETIF_F_SG since no checksum feature.\n");
5268 features
&= ~NETIF_F_SG
;
5271 /* TSO requires that SG is present as well. */
5272 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
5273 netdev_info(dev
, "Dropping NETIF_F_TSO since no SG feature.\n");
5274 features
&= ~NETIF_F_TSO
;
5277 /* Software GSO depends on SG. */
5278 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5279 netdev_info(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5280 features
&= ~NETIF_F_GSO
;
5283 /* UFO needs SG and checksumming */
5284 if (features
& NETIF_F_UFO
) {
5285 /* maybe split UFO into V4 and V6? */
5286 if (!((features
& NETIF_F_GEN_CSUM
) ||
5287 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5288 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5290 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5291 features
&= ~NETIF_F_UFO
;
5294 if (!(features
& NETIF_F_SG
)) {
5296 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5297 features
&= ~NETIF_F_UFO
;
5303 EXPORT_SYMBOL(netdev_fix_features
);
5306 * netif_stacked_transfer_operstate - transfer operstate
5307 * @rootdev: the root or lower level device to transfer state from
5308 * @dev: the device to transfer operstate to
5310 * Transfer operational state from root to device. This is normally
5311 * called when a stacking relationship exists between the root
5312 * device and the device(a leaf device).
5314 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5315 struct net_device
*dev
)
5317 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5318 netif_dormant_on(dev
);
5320 netif_dormant_off(dev
);
5322 if (netif_carrier_ok(rootdev
)) {
5323 if (!netif_carrier_ok(dev
))
5324 netif_carrier_on(dev
);
5326 if (netif_carrier_ok(dev
))
5327 netif_carrier_off(dev
);
5330 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5333 static int netif_alloc_rx_queues(struct net_device
*dev
)
5335 unsigned int i
, count
= dev
->num_rx_queues
;
5336 struct netdev_rx_queue
*rx
;
5340 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5342 pr_err("netdev: Unable to allocate %u rx queues.\n", count
);
5347 for (i
= 0; i
< count
; i
++)
5353 static void netdev_init_one_queue(struct net_device
*dev
,
5354 struct netdev_queue
*queue
, void *_unused
)
5356 /* Initialize queue lock */
5357 spin_lock_init(&queue
->_xmit_lock
);
5358 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5359 queue
->xmit_lock_owner
= -1;
5360 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5364 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5366 unsigned int count
= dev
->num_tx_queues
;
5367 struct netdev_queue
*tx
;
5371 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5373 pr_err("netdev: Unable to allocate %u tx queues.\n",
5379 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5380 spin_lock_init(&dev
->tx_global_lock
);
5386 * register_netdevice - register a network device
5387 * @dev: device to register
5389 * Take a completed network device structure and add it to the kernel
5390 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5391 * chain. 0 is returned on success. A negative errno code is returned
5392 * on a failure to set up the device, or if the name is a duplicate.
5394 * Callers must hold the rtnl semaphore. You may want
5395 * register_netdev() instead of this.
5398 * The locking appears insufficient to guarantee two parallel registers
5399 * will not get the same name.
5402 int register_netdevice(struct net_device
*dev
)
5405 struct net
*net
= dev_net(dev
);
5407 BUG_ON(dev_boot_phase
);
5412 /* When net_device's are persistent, this will be fatal. */
5413 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5416 spin_lock_init(&dev
->addr_list_lock
);
5417 netdev_set_addr_lockdep_class(dev
);
5421 /* Init, if this function is available */
5422 if (dev
->netdev_ops
->ndo_init
) {
5423 ret
= dev
->netdev_ops
->ndo_init(dev
);
5431 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
5435 dev
->ifindex
= dev_new_index(net
);
5436 if (dev
->iflink
== -1)
5437 dev
->iflink
= dev
->ifindex
;
5439 /* Enable software offloads by default - will be stripped in
5440 * netdev_fix_features() if not supported. */
5441 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5443 /* Avoid warning from netdev_fix_features() for GSO without SG */
5444 if (!(dev
->features
& NETIF_F_SG
))
5445 dev
->features
&= ~NETIF_F_GSO
;
5447 dev
->features
= netdev_fix_features(dev
, dev
->features
);
5449 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5450 * vlan_dev_init() will do the dev->features check, so these features
5451 * are enabled only if supported by underlying device.
5453 dev
->vlan_features
|= (NETIF_F_GRO
| NETIF_F_HIGHDMA
);
5455 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5456 ret
= notifier_to_errno(ret
);
5460 ret
= netdev_register_kobject(dev
);
5463 dev
->reg_state
= NETREG_REGISTERED
;
5466 * Default initial state at registry is that the
5467 * device is present.
5470 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5472 dev_init_scheduler(dev
);
5474 list_netdevice(dev
);
5476 /* Notify protocols, that a new device appeared. */
5477 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5478 ret
= notifier_to_errno(ret
);
5480 rollback_registered(dev
);
5481 dev
->reg_state
= NETREG_UNREGISTERED
;
5484 * Prevent userspace races by waiting until the network
5485 * device is fully setup before sending notifications.
5487 if (!dev
->rtnl_link_ops
||
5488 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5489 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5495 if (dev
->netdev_ops
->ndo_uninit
)
5496 dev
->netdev_ops
->ndo_uninit(dev
);
5499 EXPORT_SYMBOL(register_netdevice
);
5502 * init_dummy_netdev - init a dummy network device for NAPI
5503 * @dev: device to init
5505 * This takes a network device structure and initialize the minimum
5506 * amount of fields so it can be used to schedule NAPI polls without
5507 * registering a full blown interface. This is to be used by drivers
5508 * that need to tie several hardware interfaces to a single NAPI
5509 * poll scheduler due to HW limitations.
5511 int init_dummy_netdev(struct net_device
*dev
)
5513 /* Clear everything. Note we don't initialize spinlocks
5514 * are they aren't supposed to be taken by any of the
5515 * NAPI code and this dummy netdev is supposed to be
5516 * only ever used for NAPI polls
5518 memset(dev
, 0, sizeof(struct net_device
));
5520 /* make sure we BUG if trying to hit standard
5521 * register/unregister code path
5523 dev
->reg_state
= NETREG_DUMMY
;
5525 /* NAPI wants this */
5526 INIT_LIST_HEAD(&dev
->napi_list
);
5528 /* a dummy interface is started by default */
5529 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5530 set_bit(__LINK_STATE_START
, &dev
->state
);
5532 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5533 * because users of this 'device' dont need to change
5539 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5543 * register_netdev - register a network device
5544 * @dev: device to register
5546 * Take a completed network device structure and add it to the kernel
5547 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5548 * chain. 0 is returned on success. A negative errno code is returned
5549 * on a failure to set up the device, or if the name is a duplicate.
5551 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5552 * and expands the device name if you passed a format string to
5555 int register_netdev(struct net_device
*dev
)
5562 * If the name is a format string the caller wants us to do a
5565 if (strchr(dev
->name
, '%')) {
5566 err
= dev_alloc_name(dev
, dev
->name
);
5571 err
= register_netdevice(dev
);
5576 EXPORT_SYMBOL(register_netdev
);
5578 int netdev_refcnt_read(const struct net_device
*dev
)
5582 for_each_possible_cpu(i
)
5583 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5586 EXPORT_SYMBOL(netdev_refcnt_read
);
5589 * netdev_wait_allrefs - wait until all references are gone.
5591 * This is called when unregistering network devices.
5593 * Any protocol or device that holds a reference should register
5594 * for netdevice notification, and cleanup and put back the
5595 * reference if they receive an UNREGISTER event.
5596 * We can get stuck here if buggy protocols don't correctly
5599 static void netdev_wait_allrefs(struct net_device
*dev
)
5601 unsigned long rebroadcast_time
, warning_time
;
5604 linkwatch_forget_dev(dev
);
5606 rebroadcast_time
= warning_time
= jiffies
;
5607 refcnt
= netdev_refcnt_read(dev
);
5609 while (refcnt
!= 0) {
5610 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5613 /* Rebroadcast unregister notification */
5614 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5615 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5616 * should have already handle it the first time */
5618 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5620 /* We must not have linkwatch events
5621 * pending on unregister. If this
5622 * happens, we simply run the queue
5623 * unscheduled, resulting in a noop
5626 linkwatch_run_queue();
5631 rebroadcast_time
= jiffies
;
5636 refcnt
= netdev_refcnt_read(dev
);
5638 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5639 printk(KERN_EMERG
"unregister_netdevice: "
5640 "waiting for %s to become free. Usage "
5643 warning_time
= jiffies
;
5652 * register_netdevice(x1);
5653 * register_netdevice(x2);
5655 * unregister_netdevice(y1);
5656 * unregister_netdevice(y2);
5662 * We are invoked by rtnl_unlock().
5663 * This allows us to deal with problems:
5664 * 1) We can delete sysfs objects which invoke hotplug
5665 * without deadlocking with linkwatch via keventd.
5666 * 2) Since we run with the RTNL semaphore not held, we can sleep
5667 * safely in order to wait for the netdev refcnt to drop to zero.
5669 * We must not return until all unregister events added during
5670 * the interval the lock was held have been completed.
5672 void netdev_run_todo(void)
5674 struct list_head list
;
5676 /* Snapshot list, allow later requests */
5677 list_replace_init(&net_todo_list
, &list
);
5681 while (!list_empty(&list
)) {
5682 struct net_device
*dev
5683 = list_first_entry(&list
, struct net_device
, todo_list
);
5684 list_del(&dev
->todo_list
);
5686 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5687 printk(KERN_ERR
"network todo '%s' but state %d\n",
5688 dev
->name
, dev
->reg_state
);
5693 dev
->reg_state
= NETREG_UNREGISTERED
;
5695 on_each_cpu(flush_backlog
, dev
, 1);
5697 netdev_wait_allrefs(dev
);
5700 BUG_ON(netdev_refcnt_read(dev
));
5701 WARN_ON(rcu_dereference_raw(dev
->ip_ptr
));
5702 WARN_ON(rcu_dereference_raw(dev
->ip6_ptr
));
5703 WARN_ON(dev
->dn_ptr
);
5705 if (dev
->destructor
)
5706 dev
->destructor(dev
);
5708 /* Free network device */
5709 kobject_put(&dev
->dev
.kobj
);
5713 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5714 * fields in the same order, with only the type differing.
5716 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5717 const struct net_device_stats
*netdev_stats
)
5719 #if BITS_PER_LONG == 64
5720 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5721 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5723 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5724 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5725 u64
*dst
= (u64
*)stats64
;
5727 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5728 sizeof(*stats64
) / sizeof(u64
));
5729 for (i
= 0; i
< n
; i
++)
5735 * dev_get_stats - get network device statistics
5736 * @dev: device to get statistics from
5737 * @storage: place to store stats
5739 * Get network statistics from device. Return @storage.
5740 * The device driver may provide its own method by setting
5741 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5742 * otherwise the internal statistics structure is used.
5744 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5745 struct rtnl_link_stats64
*storage
)
5747 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5749 if (ops
->ndo_get_stats64
) {
5750 memset(storage
, 0, sizeof(*storage
));
5751 ops
->ndo_get_stats64(dev
, storage
);
5752 } else if (ops
->ndo_get_stats
) {
5753 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5755 netdev_stats_to_stats64(storage
, &dev
->stats
);
5757 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5760 EXPORT_SYMBOL(dev_get_stats
);
5762 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5764 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5766 #ifdef CONFIG_NET_CLS_ACT
5769 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5772 netdev_init_one_queue(dev
, queue
, NULL
);
5773 queue
->qdisc
= &noop_qdisc
;
5774 queue
->qdisc_sleeping
= &noop_qdisc
;
5775 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5781 * alloc_netdev_mqs - allocate network device
5782 * @sizeof_priv: size of private data to allocate space for
5783 * @name: device name format string
5784 * @setup: callback to initialize device
5785 * @txqs: the number of TX subqueues to allocate
5786 * @rxqs: the number of RX subqueues to allocate
5788 * Allocates a struct net_device with private data area for driver use
5789 * and performs basic initialization. Also allocates subquue structs
5790 * for each queue on the device.
5792 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5793 void (*setup
)(struct net_device
*),
5794 unsigned int txqs
, unsigned int rxqs
)
5796 struct net_device
*dev
;
5798 struct net_device
*p
;
5800 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5803 pr_err("alloc_netdev: Unable to allocate device "
5804 "with zero queues.\n");
5810 pr_err("alloc_netdev: Unable to allocate device "
5811 "with zero RX queues.\n");
5816 alloc_size
= sizeof(struct net_device
);
5818 /* ensure 32-byte alignment of private area */
5819 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5820 alloc_size
+= sizeof_priv
;
5822 /* ensure 32-byte alignment of whole construct */
5823 alloc_size
+= NETDEV_ALIGN
- 1;
5825 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5827 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5831 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5832 dev
->padded
= (char *)dev
- (char *)p
;
5834 dev
->pcpu_refcnt
= alloc_percpu(int);
5835 if (!dev
->pcpu_refcnt
)
5838 if (dev_addr_init(dev
))
5844 dev_net_set(dev
, &init_net
);
5846 dev
->gso_max_size
= GSO_MAX_SIZE
;
5848 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5849 dev
->ethtool_ntuple_list
.count
= 0;
5850 INIT_LIST_HEAD(&dev
->napi_list
);
5851 INIT_LIST_HEAD(&dev
->unreg_list
);
5852 INIT_LIST_HEAD(&dev
->link_watch_list
);
5853 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5856 dev
->num_tx_queues
= txqs
;
5857 dev
->real_num_tx_queues
= txqs
;
5858 if (netif_alloc_netdev_queues(dev
))
5862 dev
->num_rx_queues
= rxqs
;
5863 dev
->real_num_rx_queues
= rxqs
;
5864 if (netif_alloc_rx_queues(dev
))
5868 strcpy(dev
->name
, name
);
5869 dev
->group
= INIT_NETDEV_GROUP
;
5877 free_percpu(dev
->pcpu_refcnt
);
5887 EXPORT_SYMBOL(alloc_netdev_mqs
);
5890 * free_netdev - free network device
5893 * This function does the last stage of destroying an allocated device
5894 * interface. The reference to the device object is released.
5895 * If this is the last reference then it will be freed.
5897 void free_netdev(struct net_device
*dev
)
5899 struct napi_struct
*p
, *n
;
5901 release_net(dev_net(dev
));
5908 kfree(rcu_dereference_raw(dev
->ingress_queue
));
5910 /* Flush device addresses */
5911 dev_addr_flush(dev
);
5913 /* Clear ethtool n-tuple list */
5914 ethtool_ntuple_flush(dev
);
5916 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5919 free_percpu(dev
->pcpu_refcnt
);
5920 dev
->pcpu_refcnt
= NULL
;
5922 /* Compatibility with error handling in drivers */
5923 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5924 kfree((char *)dev
- dev
->padded
);
5928 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5929 dev
->reg_state
= NETREG_RELEASED
;
5931 /* will free via device release */
5932 put_device(&dev
->dev
);
5934 EXPORT_SYMBOL(free_netdev
);
5937 * synchronize_net - Synchronize with packet receive processing
5939 * Wait for packets currently being received to be done.
5940 * Does not block later packets from starting.
5942 void synchronize_net(void)
5947 EXPORT_SYMBOL(synchronize_net
);
5950 * unregister_netdevice_queue - remove device from the kernel
5954 * This function shuts down a device interface and removes it
5955 * from the kernel tables.
5956 * If head not NULL, device is queued to be unregistered later.
5958 * Callers must hold the rtnl semaphore. You may want
5959 * unregister_netdev() instead of this.
5962 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5967 list_move_tail(&dev
->unreg_list
, head
);
5969 rollback_registered(dev
);
5970 /* Finish processing unregister after unlock */
5974 EXPORT_SYMBOL(unregister_netdevice_queue
);
5977 * unregister_netdevice_many - unregister many devices
5978 * @head: list of devices
5980 void unregister_netdevice_many(struct list_head
*head
)
5982 struct net_device
*dev
;
5984 if (!list_empty(head
)) {
5985 rollback_registered_many(head
);
5986 list_for_each_entry(dev
, head
, unreg_list
)
5990 EXPORT_SYMBOL(unregister_netdevice_many
);
5993 * unregister_netdev - remove device from the kernel
5996 * This function shuts down a device interface and removes it
5997 * from the kernel tables.
5999 * This is just a wrapper for unregister_netdevice that takes
6000 * the rtnl semaphore. In general you want to use this and not
6001 * unregister_netdevice.
6003 void unregister_netdev(struct net_device
*dev
)
6006 unregister_netdevice(dev
);
6009 EXPORT_SYMBOL(unregister_netdev
);
6012 * dev_change_net_namespace - move device to different nethost namespace
6014 * @net: network namespace
6015 * @pat: If not NULL name pattern to try if the current device name
6016 * is already taken in the destination network namespace.
6018 * This function shuts down a device interface and moves it
6019 * to a new network namespace. On success 0 is returned, on
6020 * a failure a netagive errno code is returned.
6022 * Callers must hold the rtnl semaphore.
6025 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6031 /* Don't allow namespace local devices to be moved. */
6033 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6036 /* Ensure the device has been registrered */
6038 if (dev
->reg_state
!= NETREG_REGISTERED
)
6041 /* Get out if there is nothing todo */
6043 if (net_eq(dev_net(dev
), net
))
6046 /* Pick the destination device name, and ensure
6047 * we can use it in the destination network namespace.
6050 if (__dev_get_by_name(net
, dev
->name
)) {
6051 /* We get here if we can't use the current device name */
6054 if (dev_get_valid_name(dev
, pat
, 1))
6059 * And now a mini version of register_netdevice unregister_netdevice.
6062 /* If device is running close it first. */
6065 /* And unlink it from device chain */
6067 unlist_netdevice(dev
);
6071 /* Shutdown queueing discipline. */
6074 /* Notify protocols, that we are about to destroy
6075 this device. They should clean all the things.
6077 Note that dev->reg_state stays at NETREG_REGISTERED.
6078 This is wanted because this way 8021q and macvlan know
6079 the device is just moving and can keep their slaves up.
6081 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6082 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6085 * Flush the unicast and multicast chains
6090 /* Actually switch the network namespace */
6091 dev_net_set(dev
, net
);
6093 /* If there is an ifindex conflict assign a new one */
6094 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6095 int iflink
= (dev
->iflink
== dev
->ifindex
);
6096 dev
->ifindex
= dev_new_index(net
);
6098 dev
->iflink
= dev
->ifindex
;
6101 /* Fixup kobjects */
6102 err
= device_rename(&dev
->dev
, dev
->name
);
6105 /* Add the device back in the hashes */
6106 list_netdevice(dev
);
6108 /* Notify protocols, that a new device appeared. */
6109 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6112 * Prevent userspace races by waiting until the network
6113 * device is fully setup before sending notifications.
6115 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6122 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6124 static int dev_cpu_callback(struct notifier_block
*nfb
,
6125 unsigned long action
,
6128 struct sk_buff
**list_skb
;
6129 struct sk_buff
*skb
;
6130 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6131 struct softnet_data
*sd
, *oldsd
;
6133 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6136 local_irq_disable();
6137 cpu
= smp_processor_id();
6138 sd
= &per_cpu(softnet_data
, cpu
);
6139 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6141 /* Find end of our completion_queue. */
6142 list_skb
= &sd
->completion_queue
;
6144 list_skb
= &(*list_skb
)->next
;
6145 /* Append completion queue from offline CPU. */
6146 *list_skb
= oldsd
->completion_queue
;
6147 oldsd
->completion_queue
= NULL
;
6149 /* Append output queue from offline CPU. */
6150 if (oldsd
->output_queue
) {
6151 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6152 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6153 oldsd
->output_queue
= NULL
;
6154 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6157 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6160 /* Process offline CPU's input_pkt_queue */
6161 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6163 input_queue_head_incr(oldsd
);
6165 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6167 input_queue_head_incr(oldsd
);
6175 * netdev_increment_features - increment feature set by one
6176 * @all: current feature set
6177 * @one: new feature set
6178 * @mask: mask feature set
6180 * Computes a new feature set after adding a device with feature set
6181 * @one to the master device with current feature set @all. Will not
6182 * enable anything that is off in @mask. Returns the new feature set.
6184 u32
netdev_increment_features(u32 all
, u32 one
, u32 mask
)
6186 /* If device needs checksumming, downgrade to it. */
6187 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
6188 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
6189 else if (mask
& NETIF_F_ALL_CSUM
) {
6190 /* If one device supports v4/v6 checksumming, set for all. */
6191 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
6192 !(all
& NETIF_F_GEN_CSUM
)) {
6193 all
&= ~NETIF_F_ALL_CSUM
;
6194 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
6197 /* If one device supports hw checksumming, set for all. */
6198 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
6199 all
&= ~NETIF_F_ALL_CSUM
;
6200 all
|= NETIF_F_HW_CSUM
;
6204 one
|= NETIF_F_ALL_CSUM
;
6206 one
|= all
& NETIF_F_ONE_FOR_ALL
;
6207 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
6208 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
6212 EXPORT_SYMBOL(netdev_increment_features
);
6214 static struct hlist_head
*netdev_create_hash(void)
6217 struct hlist_head
*hash
;
6219 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6221 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6222 INIT_HLIST_HEAD(&hash
[i
]);
6227 /* Initialize per network namespace state */
6228 static int __net_init
netdev_init(struct net
*net
)
6230 INIT_LIST_HEAD(&net
->dev_base_head
);
6232 net
->dev_name_head
= netdev_create_hash();
6233 if (net
->dev_name_head
== NULL
)
6236 net
->dev_index_head
= netdev_create_hash();
6237 if (net
->dev_index_head
== NULL
)
6243 kfree(net
->dev_name_head
);
6249 * netdev_drivername - network driver for the device
6250 * @dev: network device
6251 * @buffer: buffer for resulting name
6252 * @len: size of buffer
6254 * Determine network driver for device.
6256 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
6258 const struct device_driver
*driver
;
6259 const struct device
*parent
;
6261 if (len
<= 0 || !buffer
)
6265 parent
= dev
->dev
.parent
;
6270 driver
= parent
->driver
;
6271 if (driver
&& driver
->name
)
6272 strlcpy(buffer
, driver
->name
, len
);
6276 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6277 struct va_format
*vaf
)
6281 if (dev
&& dev
->dev
.parent
)
6282 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6283 netdev_name(dev
), vaf
);
6285 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6287 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6292 int netdev_printk(const char *level
, const struct net_device
*dev
,
6293 const char *format
, ...)
6295 struct va_format vaf
;
6299 va_start(args
, format
);
6304 r
= __netdev_printk(level
, dev
, &vaf
);
6309 EXPORT_SYMBOL(netdev_printk
);
6311 #define define_netdev_printk_level(func, level) \
6312 int func(const struct net_device *dev, const char *fmt, ...) \
6315 struct va_format vaf; \
6318 va_start(args, fmt); \
6323 r = __netdev_printk(level, dev, &vaf); \
6328 EXPORT_SYMBOL(func);
6330 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6331 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6332 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6333 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6334 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6335 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6336 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6338 static void __net_exit
netdev_exit(struct net
*net
)
6340 kfree(net
->dev_name_head
);
6341 kfree(net
->dev_index_head
);
6344 static struct pernet_operations __net_initdata netdev_net_ops
= {
6345 .init
= netdev_init
,
6346 .exit
= netdev_exit
,
6349 static void __net_exit
default_device_exit(struct net
*net
)
6351 struct net_device
*dev
, *aux
;
6353 * Push all migratable network devices back to the
6354 * initial network namespace
6357 for_each_netdev_safe(net
, dev
, aux
) {
6359 char fb_name
[IFNAMSIZ
];
6361 /* Ignore unmoveable devices (i.e. loopback) */
6362 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6365 /* Leave virtual devices for the generic cleanup */
6366 if (dev
->rtnl_link_ops
)
6369 /* Push remaing network devices to init_net */
6370 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6371 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6373 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
6374 __func__
, dev
->name
, err
);
6381 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6383 /* At exit all network devices most be removed from a network
6384 * namespace. Do this in the reverse order of registration.
6385 * Do this across as many network namespaces as possible to
6386 * improve batching efficiency.
6388 struct net_device
*dev
;
6390 LIST_HEAD(dev_kill_list
);
6393 list_for_each_entry(net
, net_list
, exit_list
) {
6394 for_each_netdev_reverse(net
, dev
) {
6395 if (dev
->rtnl_link_ops
)
6396 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6398 unregister_netdevice_queue(dev
, &dev_kill_list
);
6401 unregister_netdevice_many(&dev_kill_list
);
6405 static struct pernet_operations __net_initdata default_device_ops
= {
6406 .exit
= default_device_exit
,
6407 .exit_batch
= default_device_exit_batch
,
6411 * Initialize the DEV module. At boot time this walks the device list and
6412 * unhooks any devices that fail to initialise (normally hardware not
6413 * present) and leaves us with a valid list of present and active devices.
6418 * This is called single threaded during boot, so no need
6419 * to take the rtnl semaphore.
6421 static int __init
net_dev_init(void)
6423 int i
, rc
= -ENOMEM
;
6425 BUG_ON(!dev_boot_phase
);
6427 if (dev_proc_init())
6430 if (netdev_kobject_init())
6433 INIT_LIST_HEAD(&ptype_all
);
6434 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6435 INIT_LIST_HEAD(&ptype_base
[i
]);
6437 if (register_pernet_subsys(&netdev_net_ops
))
6441 * Initialise the packet receive queues.
6444 for_each_possible_cpu(i
) {
6445 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6447 memset(sd
, 0, sizeof(*sd
));
6448 skb_queue_head_init(&sd
->input_pkt_queue
);
6449 skb_queue_head_init(&sd
->process_queue
);
6450 sd
->completion_queue
= NULL
;
6451 INIT_LIST_HEAD(&sd
->poll_list
);
6452 sd
->output_queue
= NULL
;
6453 sd
->output_queue_tailp
= &sd
->output_queue
;
6455 sd
->csd
.func
= rps_trigger_softirq
;
6461 sd
->backlog
.poll
= process_backlog
;
6462 sd
->backlog
.weight
= weight_p
;
6463 sd
->backlog
.gro_list
= NULL
;
6464 sd
->backlog
.gro_count
= 0;
6469 /* The loopback device is special if any other network devices
6470 * is present in a network namespace the loopback device must
6471 * be present. Since we now dynamically allocate and free the
6472 * loopback device ensure this invariant is maintained by
6473 * keeping the loopback device as the first device on the
6474 * list of network devices. Ensuring the loopback devices
6475 * is the first device that appears and the last network device
6478 if (register_pernet_device(&loopback_net_ops
))
6481 if (register_pernet_device(&default_device_ops
))
6484 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6485 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6487 hotcpu_notifier(dev_cpu_callback
, 0);
6495 subsys_initcall(net_dev_init
);
6497 static int __init
initialize_hashrnd(void)
6499 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6503 late_initcall_sync(initialize_hashrnd
);