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>
136 #include <linux/if_tunnel.h>
137 #include <linux/if_pppox.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock
);
179 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
180 static struct list_head ptype_all __read_mostly
; /* Taps */
183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
188 * Writers must hold the rtnl semaphore while they loop through the
189 * dev_base_head list, and hold dev_base_lock for writing when they do the
190 * actual updates. This allows pure readers to access the list even
191 * while a writer is preparing to update it.
193 * To put it another way, dev_base_lock is held for writing only to
194 * protect against pure readers; the rtnl semaphore provides the
195 * protection against other writers.
197 * See, for example usages, register_netdevice() and
198 * unregister_netdevice(), which must be called with the rtnl
201 DEFINE_RWLOCK(dev_base_lock
);
202 EXPORT_SYMBOL(dev_base_lock
);
204 static inline void dev_base_seq_inc(struct net
*net
)
206 while (++net
->dev_base_seq
== 0);
209 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
211 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
212 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
215 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
217 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
220 static inline void rps_lock(struct softnet_data
*sd
)
223 spin_lock(&sd
->input_pkt_queue
.lock
);
227 static inline void rps_unlock(struct softnet_data
*sd
)
230 spin_unlock(&sd
->input_pkt_queue
.lock
);
234 /* Device list insertion */
235 static int list_netdevice(struct net_device
*dev
)
237 struct net
*net
= dev_net(dev
);
241 write_lock_bh(&dev_base_lock
);
242 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
243 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
244 hlist_add_head_rcu(&dev
->index_hlist
,
245 dev_index_hash(net
, dev
->ifindex
));
246 write_unlock_bh(&dev_base_lock
);
248 dev_base_seq_inc(net
);
253 /* Device list removal
254 * caller must respect a RCU grace period before freeing/reusing dev
256 static void unlist_netdevice(struct net_device
*dev
)
260 /* Unlink dev from the device chain */
261 write_lock_bh(&dev_base_lock
);
262 list_del_rcu(&dev
->dev_list
);
263 hlist_del_rcu(&dev
->name_hlist
);
264 hlist_del_rcu(&dev
->index_hlist
);
265 write_unlock_bh(&dev_base_lock
);
267 dev_base_seq_inc(dev_net(dev
));
274 static RAW_NOTIFIER_HEAD(netdev_chain
);
277 * Device drivers call our routines to queue packets here. We empty the
278 * queue in the local softnet handler.
281 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
282 EXPORT_PER_CPU_SYMBOL(softnet_data
);
284 #ifdef CONFIG_LOCKDEP
286 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
287 * according to dev->type
289 static const unsigned short netdev_lock_type
[] =
290 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
291 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
292 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
293 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
294 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
295 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
296 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
297 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
298 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
299 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
300 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
301 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
302 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
303 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
304 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
305 ARPHRD_VOID
, ARPHRD_NONE
};
307 static const char *const netdev_lock_name
[] =
308 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
309 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
310 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
311 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
312 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
313 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
314 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
315 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
316 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
317 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
318 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
319 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
320 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
321 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
322 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
323 "_xmit_VOID", "_xmit_NONE"};
325 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
326 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
328 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
332 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
333 if (netdev_lock_type
[i
] == dev_type
)
335 /* the last key is used by default */
336 return ARRAY_SIZE(netdev_lock_type
) - 1;
339 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
340 unsigned short dev_type
)
344 i
= netdev_lock_pos(dev_type
);
345 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
346 netdev_lock_name
[i
]);
349 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
353 i
= netdev_lock_pos(dev
->type
);
354 lockdep_set_class_and_name(&dev
->addr_list_lock
,
355 &netdev_addr_lock_key
[i
],
356 netdev_lock_name
[i
]);
359 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
360 unsigned short dev_type
)
363 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
368 /*******************************************************************************
370 Protocol management and registration routines
372 *******************************************************************************/
375 * Add a protocol ID to the list. Now that the input handler is
376 * smarter we can dispense with all the messy stuff that used to be
379 * BEWARE!!! Protocol handlers, mangling input packets,
380 * MUST BE last in hash buckets and checking protocol handlers
381 * MUST start from promiscuous ptype_all chain in net_bh.
382 * It is true now, do not change it.
383 * Explanation follows: if protocol handler, mangling packet, will
384 * be the first on list, it is not able to sense, that packet
385 * is cloned and should be copied-on-write, so that it will
386 * change it and subsequent readers will get broken packet.
390 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
392 if (pt
->type
== htons(ETH_P_ALL
))
395 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
399 * dev_add_pack - add packet handler
400 * @pt: packet type declaration
402 * Add a protocol handler to the networking stack. The passed &packet_type
403 * is linked into kernel lists and may not be freed until it has been
404 * removed from the kernel lists.
406 * This call does not sleep therefore it can not
407 * guarantee all CPU's that are in middle of receiving packets
408 * will see the new packet type (until the next received packet).
411 void dev_add_pack(struct packet_type
*pt
)
413 struct list_head
*head
= ptype_head(pt
);
415 spin_lock(&ptype_lock
);
416 list_add_rcu(&pt
->list
, head
);
417 spin_unlock(&ptype_lock
);
419 EXPORT_SYMBOL(dev_add_pack
);
422 * __dev_remove_pack - remove packet handler
423 * @pt: packet type declaration
425 * Remove a protocol handler that was previously added to the kernel
426 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
427 * from the kernel lists and can be freed or reused once this function
430 * The packet type might still be in use by receivers
431 * and must not be freed until after all the CPU's have gone
432 * through a quiescent state.
434 void __dev_remove_pack(struct packet_type
*pt
)
436 struct list_head
*head
= ptype_head(pt
);
437 struct packet_type
*pt1
;
439 spin_lock(&ptype_lock
);
441 list_for_each_entry(pt1
, head
, list
) {
443 list_del_rcu(&pt
->list
);
448 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
450 spin_unlock(&ptype_lock
);
452 EXPORT_SYMBOL(__dev_remove_pack
);
455 * dev_remove_pack - remove packet handler
456 * @pt: packet type declaration
458 * Remove a protocol handler that was previously added to the kernel
459 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
460 * from the kernel lists and can be freed or reused once this function
463 * This call sleeps to guarantee that no CPU is looking at the packet
466 void dev_remove_pack(struct packet_type
*pt
)
468 __dev_remove_pack(pt
);
472 EXPORT_SYMBOL(dev_remove_pack
);
474 /******************************************************************************
476 Device Boot-time Settings Routines
478 *******************************************************************************/
480 /* Boot time configuration table */
481 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
484 * netdev_boot_setup_add - add new setup entry
485 * @name: name of the device
486 * @map: configured settings for the device
488 * Adds new setup entry to the dev_boot_setup list. The function
489 * returns 0 on error and 1 on success. This is a generic routine to
492 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
494 struct netdev_boot_setup
*s
;
498 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
499 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
500 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
501 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
502 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
507 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
511 * netdev_boot_setup_check - check boot time settings
512 * @dev: the netdevice
514 * Check boot time settings for the device.
515 * The found settings are set for the device to be used
516 * later in the device probing.
517 * Returns 0 if no settings found, 1 if they are.
519 int netdev_boot_setup_check(struct net_device
*dev
)
521 struct netdev_boot_setup
*s
= dev_boot_setup
;
524 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
525 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
526 !strcmp(dev
->name
, s
[i
].name
)) {
527 dev
->irq
= s
[i
].map
.irq
;
528 dev
->base_addr
= s
[i
].map
.base_addr
;
529 dev
->mem_start
= s
[i
].map
.mem_start
;
530 dev
->mem_end
= s
[i
].map
.mem_end
;
536 EXPORT_SYMBOL(netdev_boot_setup_check
);
540 * netdev_boot_base - get address from boot time settings
541 * @prefix: prefix for network device
542 * @unit: id for network device
544 * Check boot time settings for the base address of device.
545 * The found settings are set for the device to be used
546 * later in the device probing.
547 * Returns 0 if no settings found.
549 unsigned long netdev_boot_base(const char *prefix
, int unit
)
551 const struct netdev_boot_setup
*s
= dev_boot_setup
;
555 sprintf(name
, "%s%d", prefix
, unit
);
558 * If device already registered then return base of 1
559 * to indicate not to probe for this interface
561 if (__dev_get_by_name(&init_net
, name
))
564 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
565 if (!strcmp(name
, s
[i
].name
))
566 return s
[i
].map
.base_addr
;
571 * Saves at boot time configured settings for any netdevice.
573 int __init
netdev_boot_setup(char *str
)
578 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
583 memset(&map
, 0, sizeof(map
));
587 map
.base_addr
= ints
[2];
589 map
.mem_start
= ints
[3];
591 map
.mem_end
= ints
[4];
593 /* Add new entry to the list */
594 return netdev_boot_setup_add(str
, &map
);
597 __setup("netdev=", netdev_boot_setup
);
599 /*******************************************************************************
601 Device Interface Subroutines
603 *******************************************************************************/
606 * __dev_get_by_name - find a device by its name
607 * @net: the applicable net namespace
608 * @name: name to find
610 * Find an interface by name. Must be called under RTNL semaphore
611 * or @dev_base_lock. If the name is found a pointer to the device
612 * is returned. If the name is not found then %NULL is returned. The
613 * reference counters are not incremented so the caller must be
614 * careful with locks.
617 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
619 struct hlist_node
*p
;
620 struct net_device
*dev
;
621 struct hlist_head
*head
= dev_name_hash(net
, name
);
623 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
624 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
629 EXPORT_SYMBOL(__dev_get_by_name
);
632 * dev_get_by_name_rcu - find a device by its name
633 * @net: the applicable net namespace
634 * @name: name to find
636 * Find an interface by name.
637 * If the name is found a pointer to the device is returned.
638 * If the name is not found then %NULL is returned.
639 * The reference counters are not incremented so the caller must be
640 * careful with locks. The caller must hold RCU lock.
643 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
645 struct hlist_node
*p
;
646 struct net_device
*dev
;
647 struct hlist_head
*head
= dev_name_hash(net
, name
);
649 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
650 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
655 EXPORT_SYMBOL(dev_get_by_name_rcu
);
658 * dev_get_by_name - find a device by its name
659 * @net: the applicable net namespace
660 * @name: name to find
662 * Find an interface by name. This can be called from any
663 * context and does its own locking. The returned handle has
664 * the usage count incremented and the caller must use dev_put() to
665 * release it when it is no longer needed. %NULL is returned if no
666 * matching device is found.
669 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
671 struct net_device
*dev
;
674 dev
= dev_get_by_name_rcu(net
, name
);
680 EXPORT_SYMBOL(dev_get_by_name
);
683 * __dev_get_by_index - find a device by its ifindex
684 * @net: the applicable net namespace
685 * @ifindex: index of device
687 * Search for an interface by index. Returns %NULL if the device
688 * is not found or a pointer to the device. The device has not
689 * had its reference counter increased so the caller must be careful
690 * about locking. The caller must hold either the RTNL semaphore
694 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
696 struct hlist_node
*p
;
697 struct net_device
*dev
;
698 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
700 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
701 if (dev
->ifindex
== ifindex
)
706 EXPORT_SYMBOL(__dev_get_by_index
);
709 * dev_get_by_index_rcu - find a device by its ifindex
710 * @net: the applicable net namespace
711 * @ifindex: index of device
713 * Search for an interface by index. Returns %NULL if the device
714 * is not found or a pointer to the device. The device has not
715 * had its reference counter increased so the caller must be careful
716 * about locking. The caller must hold RCU lock.
719 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
721 struct hlist_node
*p
;
722 struct net_device
*dev
;
723 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
725 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
726 if (dev
->ifindex
== ifindex
)
731 EXPORT_SYMBOL(dev_get_by_index_rcu
);
735 * dev_get_by_index - find a device by its ifindex
736 * @net: the applicable net namespace
737 * @ifindex: index of device
739 * Search for an interface by index. Returns NULL if the device
740 * is not found or a pointer to the device. The device returned has
741 * had a reference added and the pointer is safe until the user calls
742 * dev_put to indicate they have finished with it.
745 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
747 struct net_device
*dev
;
750 dev
= dev_get_by_index_rcu(net
, ifindex
);
756 EXPORT_SYMBOL(dev_get_by_index
);
759 * dev_getbyhwaddr_rcu - find a device by its hardware address
760 * @net: the applicable net namespace
761 * @type: media type of device
762 * @ha: hardware address
764 * Search for an interface by MAC address. Returns NULL if the device
765 * is not found or a pointer to the device.
766 * The caller must hold RCU or RTNL.
767 * The returned device has not had its ref count increased
768 * and the caller must therefore be careful about locking
772 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
775 struct net_device
*dev
;
777 for_each_netdev_rcu(net
, dev
)
778 if (dev
->type
== type
&&
779 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
784 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
786 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
788 struct net_device
*dev
;
791 for_each_netdev(net
, dev
)
792 if (dev
->type
== type
)
797 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
799 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
801 struct net_device
*dev
, *ret
= NULL
;
804 for_each_netdev_rcu(net
, dev
)
805 if (dev
->type
== type
) {
813 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
816 * dev_get_by_flags_rcu - find any device with given flags
817 * @net: the applicable net namespace
818 * @if_flags: IFF_* values
819 * @mask: bitmask of bits in if_flags to check
821 * Search for any interface with the given flags. Returns NULL if a device
822 * is not found or a pointer to the device. Must be called inside
823 * rcu_read_lock(), and result refcount is unchanged.
826 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
829 struct net_device
*dev
, *ret
;
832 for_each_netdev_rcu(net
, dev
) {
833 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
840 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
843 * dev_valid_name - check if name is okay for network device
846 * Network device names need to be valid file names to
847 * to allow sysfs to work. We also disallow any kind of
850 int dev_valid_name(const char *name
)
854 if (strlen(name
) >= IFNAMSIZ
)
856 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
860 if (*name
== '/' || isspace(*name
))
866 EXPORT_SYMBOL(dev_valid_name
);
869 * __dev_alloc_name - allocate a name for a device
870 * @net: network namespace to allocate the device name in
871 * @name: name format string
872 * @buf: scratch buffer and result name string
874 * Passed a format string - eg "lt%d" it will try and find a suitable
875 * id. It scans list of devices to build up a free map, then chooses
876 * the first empty slot. The caller must hold the dev_base or rtnl lock
877 * while allocating the name and adding the device in order to avoid
879 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
880 * Returns the number of the unit assigned or a negative errno code.
883 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
887 const int max_netdevices
= 8*PAGE_SIZE
;
888 unsigned long *inuse
;
889 struct net_device
*d
;
891 p
= strnchr(name
, IFNAMSIZ
-1, '%');
894 * Verify the string as this thing may have come from
895 * the user. There must be either one "%d" and no other "%"
898 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
901 /* Use one page as a bit array of possible slots */
902 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
906 for_each_netdev(net
, d
) {
907 if (!sscanf(d
->name
, name
, &i
))
909 if (i
< 0 || i
>= max_netdevices
)
912 /* avoid cases where sscanf is not exact inverse of printf */
913 snprintf(buf
, IFNAMSIZ
, name
, i
);
914 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
918 i
= find_first_zero_bit(inuse
, max_netdevices
);
919 free_page((unsigned long) inuse
);
923 snprintf(buf
, IFNAMSIZ
, name
, i
);
924 if (!__dev_get_by_name(net
, buf
))
927 /* It is possible to run out of possible slots
928 * when the name is long and there isn't enough space left
929 * for the digits, or if all bits are used.
935 * dev_alloc_name - allocate a name for a device
937 * @name: name format string
939 * Passed a format string - eg "lt%d" it will try and find a suitable
940 * id. It scans list of devices to build up a free map, then chooses
941 * the first empty slot. The caller must hold the dev_base or rtnl lock
942 * while allocating the name and adding the device in order to avoid
944 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
945 * Returns the number of the unit assigned or a negative errno code.
948 int dev_alloc_name(struct net_device
*dev
, const char *name
)
954 BUG_ON(!dev_net(dev
));
956 ret
= __dev_alloc_name(net
, name
, buf
);
958 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
961 EXPORT_SYMBOL(dev_alloc_name
);
963 static int dev_get_valid_name(struct net_device
*dev
, const char *name
)
967 BUG_ON(!dev_net(dev
));
970 if (!dev_valid_name(name
))
973 if (strchr(name
, '%'))
974 return dev_alloc_name(dev
, name
);
975 else if (__dev_get_by_name(net
, name
))
977 else if (dev
->name
!= name
)
978 strlcpy(dev
->name
, name
, IFNAMSIZ
);
984 * dev_change_name - change name of a device
986 * @newname: name (or format string) must be at least IFNAMSIZ
988 * Change name of a device, can pass format strings "eth%d".
991 int dev_change_name(struct net_device
*dev
, const char *newname
)
993 char oldname
[IFNAMSIZ
];
999 BUG_ON(!dev_net(dev
));
1002 if (dev
->flags
& IFF_UP
)
1005 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
1008 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1010 err
= dev_get_valid_name(dev
, newname
);
1015 ret
= device_rename(&dev
->dev
, dev
->name
);
1017 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1021 write_lock_bh(&dev_base_lock
);
1022 hlist_del_rcu(&dev
->name_hlist
);
1023 write_unlock_bh(&dev_base_lock
);
1027 write_lock_bh(&dev_base_lock
);
1028 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1029 write_unlock_bh(&dev_base_lock
);
1031 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1032 ret
= notifier_to_errno(ret
);
1035 /* err >= 0 after dev_alloc_name() or stores the first errno */
1038 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1042 "%s: name change rollback failed: %d.\n",
1051 * dev_set_alias - change ifalias of a device
1053 * @alias: name up to IFALIASZ
1054 * @len: limit of bytes to copy from info
1056 * Set ifalias for a device,
1058 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1062 if (len
>= IFALIASZ
)
1067 kfree(dev
->ifalias
);
1068 dev
->ifalias
= NULL
;
1073 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1077 strlcpy(dev
->ifalias
, alias
, len
+1);
1083 * netdev_features_change - device changes features
1084 * @dev: device to cause notification
1086 * Called to indicate a device has changed features.
1088 void netdev_features_change(struct net_device
*dev
)
1090 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1092 EXPORT_SYMBOL(netdev_features_change
);
1095 * netdev_state_change - device changes state
1096 * @dev: device to cause notification
1098 * Called to indicate a device has changed state. This function calls
1099 * the notifier chains for netdev_chain and sends a NEWLINK message
1100 * to the routing socket.
1102 void netdev_state_change(struct net_device
*dev
)
1104 if (dev
->flags
& IFF_UP
) {
1105 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1106 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1109 EXPORT_SYMBOL(netdev_state_change
);
1111 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1113 return call_netdevice_notifiers(event
, dev
);
1115 EXPORT_SYMBOL(netdev_bonding_change
);
1118 * dev_load - load a network module
1119 * @net: the applicable net namespace
1120 * @name: name of interface
1122 * If a network interface is not present and the process has suitable
1123 * privileges this function loads the module. If module loading is not
1124 * available in this kernel then it becomes a nop.
1127 void dev_load(struct net
*net
, const char *name
)
1129 struct net_device
*dev
;
1133 dev
= dev_get_by_name_rcu(net
, name
);
1137 if (no_module
&& capable(CAP_NET_ADMIN
))
1138 no_module
= request_module("netdev-%s", name
);
1139 if (no_module
&& capable(CAP_SYS_MODULE
)) {
1140 if (!request_module("%s", name
))
1141 pr_err("Loading kernel module for a network device "
1142 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1146 EXPORT_SYMBOL(dev_load
);
1148 static int __dev_open(struct net_device
*dev
)
1150 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1155 if (!netif_device_present(dev
))
1158 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1159 ret
= notifier_to_errno(ret
);
1163 set_bit(__LINK_STATE_START
, &dev
->state
);
1165 if (ops
->ndo_validate_addr
)
1166 ret
= ops
->ndo_validate_addr(dev
);
1168 if (!ret
&& ops
->ndo_open
)
1169 ret
= ops
->ndo_open(dev
);
1172 clear_bit(__LINK_STATE_START
, &dev
->state
);
1174 dev
->flags
|= IFF_UP
;
1175 net_dmaengine_get();
1176 dev_set_rx_mode(dev
);
1184 * dev_open - prepare an interface for use.
1185 * @dev: device to open
1187 * Takes a device from down to up state. The device's private open
1188 * function is invoked and then the multicast lists are loaded. Finally
1189 * the device is moved into the up state and a %NETDEV_UP message is
1190 * sent to the netdev notifier chain.
1192 * Calling this function on an active interface is a nop. On a failure
1193 * a negative errno code is returned.
1195 int dev_open(struct net_device
*dev
)
1199 if (dev
->flags
& IFF_UP
)
1202 ret
= __dev_open(dev
);
1206 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1207 call_netdevice_notifiers(NETDEV_UP
, dev
);
1211 EXPORT_SYMBOL(dev_open
);
1213 static int __dev_close_many(struct list_head
*head
)
1215 struct net_device
*dev
;
1220 list_for_each_entry(dev
, head
, unreg_list
) {
1221 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1223 clear_bit(__LINK_STATE_START
, &dev
->state
);
1225 /* Synchronize to scheduled poll. We cannot touch poll list, it
1226 * can be even on different cpu. So just clear netif_running().
1228 * dev->stop() will invoke napi_disable() on all of it's
1229 * napi_struct instances on this device.
1231 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1234 dev_deactivate_many(head
);
1236 list_for_each_entry(dev
, head
, unreg_list
) {
1237 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1240 * Call the device specific close. This cannot fail.
1241 * Only if device is UP
1243 * We allow it to be called even after a DETACH hot-plug
1249 dev
->flags
&= ~IFF_UP
;
1250 net_dmaengine_put();
1256 static int __dev_close(struct net_device
*dev
)
1261 list_add(&dev
->unreg_list
, &single
);
1262 retval
= __dev_close_many(&single
);
1267 static int dev_close_many(struct list_head
*head
)
1269 struct net_device
*dev
, *tmp
;
1270 LIST_HEAD(tmp_list
);
1272 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1273 if (!(dev
->flags
& IFF_UP
))
1274 list_move(&dev
->unreg_list
, &tmp_list
);
1276 __dev_close_many(head
);
1278 list_for_each_entry(dev
, head
, unreg_list
) {
1279 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1280 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1283 /* rollback_registered_many needs the complete original list */
1284 list_splice(&tmp_list
, head
);
1289 * dev_close - shutdown an interface.
1290 * @dev: device to shutdown
1292 * This function moves an active device into down state. A
1293 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1294 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1297 int dev_close(struct net_device
*dev
)
1299 if (dev
->flags
& IFF_UP
) {
1302 list_add(&dev
->unreg_list
, &single
);
1303 dev_close_many(&single
);
1308 EXPORT_SYMBOL(dev_close
);
1312 * dev_disable_lro - disable Large Receive Offload on a device
1315 * Disable Large Receive Offload (LRO) on a net device. Must be
1316 * called under RTNL. This is needed if received packets may be
1317 * forwarded to another interface.
1319 void dev_disable_lro(struct net_device
*dev
)
1324 * If we're trying to disable lro on a vlan device
1325 * use the underlying physical device instead
1327 if (is_vlan_dev(dev
))
1328 dev
= vlan_dev_real_dev(dev
);
1330 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
)
1331 flags
= dev
->ethtool_ops
->get_flags(dev
);
1333 flags
= ethtool_op_get_flags(dev
);
1335 if (!(flags
& ETH_FLAG_LRO
))
1338 __ethtool_set_flags(dev
, flags
& ~ETH_FLAG_LRO
);
1339 if (unlikely(dev
->features
& NETIF_F_LRO
))
1340 netdev_WARN(dev
, "failed to disable LRO!\n");
1342 EXPORT_SYMBOL(dev_disable_lro
);
1345 static int dev_boot_phase
= 1;
1348 * register_netdevice_notifier - register a network notifier block
1351 * Register a notifier to be called when network device events occur.
1352 * The notifier passed is linked into the kernel structures and must
1353 * not be reused until it has been unregistered. A negative errno code
1354 * is returned on a failure.
1356 * When registered all registration and up events are replayed
1357 * to the new notifier to allow device to have a race free
1358 * view of the network device list.
1361 int register_netdevice_notifier(struct notifier_block
*nb
)
1363 struct net_device
*dev
;
1364 struct net_device
*last
;
1369 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1375 for_each_netdev(net
, dev
) {
1376 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1377 err
= notifier_to_errno(err
);
1381 if (!(dev
->flags
& IFF_UP
))
1384 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1395 for_each_netdev(net
, dev
) {
1399 if (dev
->flags
& IFF_UP
) {
1400 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1401 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1403 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1404 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1408 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1411 EXPORT_SYMBOL(register_netdevice_notifier
);
1414 * unregister_netdevice_notifier - unregister a network notifier block
1417 * Unregister a notifier previously registered by
1418 * register_netdevice_notifier(). The notifier is unlinked into the
1419 * kernel structures and may then be reused. A negative errno code
1420 * is returned on a failure.
1423 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1428 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1432 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1435 * call_netdevice_notifiers - call all network notifier blocks
1436 * @val: value passed unmodified to notifier function
1437 * @dev: net_device pointer passed unmodified to notifier function
1439 * Call all network notifier blocks. Parameters and return value
1440 * are as for raw_notifier_call_chain().
1443 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1446 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1448 EXPORT_SYMBOL(call_netdevice_notifiers
);
1450 /* When > 0 there are consumers of rx skb time stamps */
1451 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1453 void net_enable_timestamp(void)
1455 atomic_inc(&netstamp_needed
);
1457 EXPORT_SYMBOL(net_enable_timestamp
);
1459 void net_disable_timestamp(void)
1461 atomic_dec(&netstamp_needed
);
1463 EXPORT_SYMBOL(net_disable_timestamp
);
1465 static inline void net_timestamp_set(struct sk_buff
*skb
)
1467 if (atomic_read(&netstamp_needed
))
1468 __net_timestamp(skb
);
1470 skb
->tstamp
.tv64
= 0;
1473 static inline void net_timestamp_check(struct sk_buff
*skb
)
1475 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1476 __net_timestamp(skb
);
1479 static inline bool is_skb_forwardable(struct net_device
*dev
,
1480 struct sk_buff
*skb
)
1484 if (!(dev
->flags
& IFF_UP
))
1487 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1488 if (skb
->len
<= len
)
1491 /* if TSO is enabled, we don't care about the length as the packet
1492 * could be forwarded without being segmented before
1494 if (skb_is_gso(skb
))
1501 * dev_forward_skb - loopback an skb to another netif
1503 * @dev: destination network device
1504 * @skb: buffer to forward
1507 * NET_RX_SUCCESS (no congestion)
1508 * NET_RX_DROP (packet was dropped, but freed)
1510 * dev_forward_skb can be used for injecting an skb from the
1511 * start_xmit function of one device into the receive queue
1512 * of another device.
1514 * The receiving device may be in another namespace, so
1515 * we have to clear all information in the skb that could
1516 * impact namespace isolation.
1518 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1523 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1524 atomic_long_inc(&dev
->rx_dropped
);
1528 skb_set_dev(skb
, dev
);
1529 skb
->tstamp
.tv64
= 0;
1530 skb
->pkt_type
= PACKET_HOST
;
1531 skb
->protocol
= eth_type_trans(skb
, dev
);
1532 return netif_rx(skb
);
1534 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1536 static inline int deliver_skb(struct sk_buff
*skb
,
1537 struct packet_type
*pt_prev
,
1538 struct net_device
*orig_dev
)
1540 atomic_inc(&skb
->users
);
1541 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1545 * Support routine. Sends outgoing frames to any network
1546 * taps currently in use.
1549 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1551 struct packet_type
*ptype
;
1552 struct sk_buff
*skb2
= NULL
;
1553 struct packet_type
*pt_prev
= NULL
;
1556 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1557 /* Never send packets back to the socket
1558 * they originated from - MvS (miquels@drinkel.ow.org)
1560 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1561 (ptype
->af_packet_priv
== NULL
||
1562 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1564 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1569 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1573 net_timestamp_set(skb2
);
1575 /* skb->nh should be correctly
1576 set by sender, so that the second statement is
1577 just protection against buggy protocols.
1579 skb_reset_mac_header(skb2
);
1581 if (skb_network_header(skb2
) < skb2
->data
||
1582 skb2
->network_header
> skb2
->tail
) {
1583 if (net_ratelimit())
1584 printk(KERN_CRIT
"protocol %04x is "
1586 ntohs(skb2
->protocol
),
1588 skb_reset_network_header(skb2
);
1591 skb2
->transport_header
= skb2
->network_header
;
1592 skb2
->pkt_type
= PACKET_OUTGOING
;
1597 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1601 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1602 * @dev: Network device
1603 * @txq: number of queues available
1605 * If real_num_tx_queues is changed the tc mappings may no longer be
1606 * valid. To resolve this verify the tc mapping remains valid and if
1607 * not NULL the mapping. With no priorities mapping to this
1608 * offset/count pair it will no longer be used. In the worst case TC0
1609 * is invalid nothing can be done so disable priority mappings. If is
1610 * expected that drivers will fix this mapping if they can before
1611 * calling netif_set_real_num_tx_queues.
1613 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1616 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1618 /* If TC0 is invalidated disable TC mapping */
1619 if (tc
->offset
+ tc
->count
> txq
) {
1620 pr_warning("Number of in use tx queues changed "
1621 "invalidating tc mappings. Priority "
1622 "traffic classification disabled!\n");
1627 /* Invalidated prio to tc mappings set to TC0 */
1628 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1629 int q
= netdev_get_prio_tc_map(dev
, i
);
1631 tc
= &dev
->tc_to_txq
[q
];
1632 if (tc
->offset
+ tc
->count
> txq
) {
1633 pr_warning("Number of in use tx queues "
1634 "changed. Priority %i to tc "
1635 "mapping %i is no longer valid "
1636 "setting map to 0\n",
1638 netdev_set_prio_tc_map(dev
, i
, 0);
1644 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1645 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1647 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1651 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1654 if (dev
->reg_state
== NETREG_REGISTERED
||
1655 dev
->reg_state
== NETREG_UNREGISTERING
) {
1658 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1664 netif_setup_tc(dev
, txq
);
1666 if (txq
< dev
->real_num_tx_queues
)
1667 qdisc_reset_all_tx_gt(dev
, txq
);
1670 dev
->real_num_tx_queues
= txq
;
1673 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1677 * netif_set_real_num_rx_queues - set actual number of RX queues used
1678 * @dev: Network device
1679 * @rxq: Actual number of RX queues
1681 * This must be called either with the rtnl_lock held or before
1682 * registration of the net device. Returns 0 on success, or a
1683 * negative error code. If called before registration, it always
1686 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1690 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1693 if (dev
->reg_state
== NETREG_REGISTERED
) {
1696 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1702 dev
->real_num_rx_queues
= rxq
;
1705 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1708 static inline void __netif_reschedule(struct Qdisc
*q
)
1710 struct softnet_data
*sd
;
1711 unsigned long flags
;
1713 local_irq_save(flags
);
1714 sd
= &__get_cpu_var(softnet_data
);
1715 q
->next_sched
= NULL
;
1716 *sd
->output_queue_tailp
= q
;
1717 sd
->output_queue_tailp
= &q
->next_sched
;
1718 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1719 local_irq_restore(flags
);
1722 void __netif_schedule(struct Qdisc
*q
)
1724 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1725 __netif_reschedule(q
);
1727 EXPORT_SYMBOL(__netif_schedule
);
1729 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1731 if (atomic_dec_and_test(&skb
->users
)) {
1732 struct softnet_data
*sd
;
1733 unsigned long flags
;
1735 local_irq_save(flags
);
1736 sd
= &__get_cpu_var(softnet_data
);
1737 skb
->next
= sd
->completion_queue
;
1738 sd
->completion_queue
= skb
;
1739 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1740 local_irq_restore(flags
);
1743 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1745 void dev_kfree_skb_any(struct sk_buff
*skb
)
1747 if (in_irq() || irqs_disabled())
1748 dev_kfree_skb_irq(skb
);
1752 EXPORT_SYMBOL(dev_kfree_skb_any
);
1756 * netif_device_detach - mark device as removed
1757 * @dev: network device
1759 * Mark device as removed from system and therefore no longer available.
1761 void netif_device_detach(struct net_device
*dev
)
1763 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1764 netif_running(dev
)) {
1765 netif_tx_stop_all_queues(dev
);
1768 EXPORT_SYMBOL(netif_device_detach
);
1771 * netif_device_attach - mark device as attached
1772 * @dev: network device
1774 * Mark device as attached from system and restart if needed.
1776 void netif_device_attach(struct net_device
*dev
)
1778 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1779 netif_running(dev
)) {
1780 netif_tx_wake_all_queues(dev
);
1781 __netdev_watchdog_up(dev
);
1784 EXPORT_SYMBOL(netif_device_attach
);
1787 * skb_dev_set -- assign a new device to a buffer
1788 * @skb: buffer for the new device
1789 * @dev: network device
1791 * If an skb is owned by a device already, we have to reset
1792 * all data private to the namespace a device belongs to
1793 * before assigning it a new device.
1795 #ifdef CONFIG_NET_NS
1796 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1799 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1802 skb_init_secmark(skb
);
1806 skb
->ipvs_property
= 0;
1807 #ifdef CONFIG_NET_SCHED
1813 EXPORT_SYMBOL(skb_set_dev
);
1814 #endif /* CONFIG_NET_NS */
1817 * Invalidate hardware checksum when packet is to be mangled, and
1818 * complete checksum manually on outgoing path.
1820 int skb_checksum_help(struct sk_buff
*skb
)
1823 int ret
= 0, offset
;
1825 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1826 goto out_set_summed
;
1828 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1829 /* Let GSO fix up the checksum. */
1830 goto out_set_summed
;
1833 offset
= skb_checksum_start_offset(skb
);
1834 BUG_ON(offset
>= skb_headlen(skb
));
1835 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1837 offset
+= skb
->csum_offset
;
1838 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1840 if (skb_cloned(skb
) &&
1841 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1842 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1847 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1849 skb
->ip_summed
= CHECKSUM_NONE
;
1853 EXPORT_SYMBOL(skb_checksum_help
);
1856 * skb_gso_segment - Perform segmentation on skb.
1857 * @skb: buffer to segment
1858 * @features: features for the output path (see dev->features)
1860 * This function segments the given skb and returns a list of segments.
1862 * It may return NULL if the skb requires no segmentation. This is
1863 * only possible when GSO is used for verifying header integrity.
1865 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, u32 features
)
1867 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1868 struct packet_type
*ptype
;
1869 __be16 type
= skb
->protocol
;
1870 int vlan_depth
= ETH_HLEN
;
1873 while (type
== htons(ETH_P_8021Q
)) {
1874 struct vlan_hdr
*vh
;
1876 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1877 return ERR_PTR(-EINVAL
);
1879 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1880 type
= vh
->h_vlan_encapsulated_proto
;
1881 vlan_depth
+= VLAN_HLEN
;
1884 skb_reset_mac_header(skb
);
1885 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1886 __skb_pull(skb
, skb
->mac_len
);
1888 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1889 struct net_device
*dev
= skb
->dev
;
1890 struct ethtool_drvinfo info
= {};
1892 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1893 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1895 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1896 info
.driver
, dev
? dev
->features
: 0L,
1897 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1898 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1900 if (skb_header_cloned(skb
) &&
1901 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1902 return ERR_PTR(err
);
1906 list_for_each_entry_rcu(ptype
,
1907 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1908 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1909 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1910 err
= ptype
->gso_send_check(skb
);
1911 segs
= ERR_PTR(err
);
1912 if (err
|| skb_gso_ok(skb
, features
))
1914 __skb_push(skb
, (skb
->data
-
1915 skb_network_header(skb
)));
1917 segs
= ptype
->gso_segment(skb
, features
);
1923 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1927 EXPORT_SYMBOL(skb_gso_segment
);
1929 /* Take action when hardware reception checksum errors are detected. */
1931 void netdev_rx_csum_fault(struct net_device
*dev
)
1933 if (net_ratelimit()) {
1934 printk(KERN_ERR
"%s: hw csum failure.\n",
1935 dev
? dev
->name
: "<unknown>");
1939 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1942 /* Actually, we should eliminate this check as soon as we know, that:
1943 * 1. IOMMU is present and allows to map all the memory.
1944 * 2. No high memory really exists on this machine.
1947 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1949 #ifdef CONFIG_HIGHMEM
1951 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1952 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1953 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1957 if (PCI_DMA_BUS_IS_PHYS
) {
1958 struct device
*pdev
= dev
->dev
.parent
;
1962 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1963 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1964 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1973 void (*destructor
)(struct sk_buff
*skb
);
1976 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1978 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1980 struct dev_gso_cb
*cb
;
1983 struct sk_buff
*nskb
= skb
->next
;
1985 skb
->next
= nskb
->next
;
1988 } while (skb
->next
);
1990 cb
= DEV_GSO_CB(skb
);
1992 cb
->destructor(skb
);
1996 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1997 * @skb: buffer to segment
1998 * @features: device features as applicable to this skb
2000 * This function segments the given skb and stores the list of segments
2003 static int dev_gso_segment(struct sk_buff
*skb
, int features
)
2005 struct sk_buff
*segs
;
2007 segs
= skb_gso_segment(skb
, features
);
2009 /* Verifying header integrity only. */
2014 return PTR_ERR(segs
);
2017 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2018 skb
->destructor
= dev_gso_skb_destructor
;
2024 * Try to orphan skb early, right before transmission by the device.
2025 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2026 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2028 static inline void skb_orphan_try(struct sk_buff
*skb
)
2030 struct sock
*sk
= skb
->sk
;
2032 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
2033 /* skb_tx_hash() wont be able to get sk.
2034 * We copy sk_hash into skb->rxhash
2037 skb
->rxhash
= sk
->sk_hash
;
2042 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
2044 return ((features
& NETIF_F_GEN_CSUM
) ||
2045 ((features
& NETIF_F_V4_CSUM
) &&
2046 protocol
== htons(ETH_P_IP
)) ||
2047 ((features
& NETIF_F_V6_CSUM
) &&
2048 protocol
== htons(ETH_P_IPV6
)) ||
2049 ((features
& NETIF_F_FCOE_CRC
) &&
2050 protocol
== htons(ETH_P_FCOE
)));
2053 static u32
harmonize_features(struct sk_buff
*skb
, __be16 protocol
, u32 features
)
2055 if (!can_checksum_protocol(features
, protocol
)) {
2056 features
&= ~NETIF_F_ALL_CSUM
;
2057 features
&= ~NETIF_F_SG
;
2058 } else if (illegal_highdma(skb
->dev
, skb
)) {
2059 features
&= ~NETIF_F_SG
;
2065 u32
netif_skb_features(struct sk_buff
*skb
)
2067 __be16 protocol
= skb
->protocol
;
2068 u32 features
= skb
->dev
->features
;
2070 if (protocol
== htons(ETH_P_8021Q
)) {
2071 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2072 protocol
= veh
->h_vlan_encapsulated_proto
;
2073 } else if (!vlan_tx_tag_present(skb
)) {
2074 return harmonize_features(skb
, protocol
, features
);
2077 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2079 if (protocol
!= htons(ETH_P_8021Q
)) {
2080 return harmonize_features(skb
, protocol
, features
);
2082 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2083 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2084 return harmonize_features(skb
, protocol
, features
);
2087 EXPORT_SYMBOL(netif_skb_features
);
2090 * Returns true if either:
2091 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2092 * 2. skb is fragmented and the device does not support SG, or if
2093 * at least one of fragments is in highmem and device does not
2094 * support DMA from it.
2096 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2099 return skb_is_nonlinear(skb
) &&
2100 ((skb_has_frag_list(skb
) &&
2101 !(features
& NETIF_F_FRAGLIST
)) ||
2102 (skb_shinfo(skb
)->nr_frags
&&
2103 !(features
& NETIF_F_SG
)));
2106 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2107 struct netdev_queue
*txq
)
2109 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2110 int rc
= NETDEV_TX_OK
;
2111 unsigned int skb_len
;
2113 if (likely(!skb
->next
)) {
2117 * If device doesn't need skb->dst, release it right now while
2118 * its hot in this cpu cache
2120 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2123 if (!list_empty(&ptype_all
))
2124 dev_queue_xmit_nit(skb
, dev
);
2126 skb_orphan_try(skb
);
2128 features
= netif_skb_features(skb
);
2130 if (vlan_tx_tag_present(skb
) &&
2131 !(features
& NETIF_F_HW_VLAN_TX
)) {
2132 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2139 if (netif_needs_gso(skb
, features
)) {
2140 if (unlikely(dev_gso_segment(skb
, features
)))
2145 if (skb_needs_linearize(skb
, features
) &&
2146 __skb_linearize(skb
))
2149 /* If packet is not checksummed and device does not
2150 * support checksumming for this protocol, complete
2151 * checksumming here.
2153 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2154 skb_set_transport_header(skb
,
2155 skb_checksum_start_offset(skb
));
2156 if (!(features
& NETIF_F_ALL_CSUM
) &&
2157 skb_checksum_help(skb
))
2163 rc
= ops
->ndo_start_xmit(skb
, dev
);
2164 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2165 if (rc
== NETDEV_TX_OK
)
2166 txq_trans_update(txq
);
2172 struct sk_buff
*nskb
= skb
->next
;
2174 skb
->next
= nskb
->next
;
2178 * If device doesn't need nskb->dst, release it right now while
2179 * its hot in this cpu cache
2181 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2184 skb_len
= nskb
->len
;
2185 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2186 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2187 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2188 if (rc
& ~NETDEV_TX_MASK
)
2189 goto out_kfree_gso_skb
;
2190 nskb
->next
= skb
->next
;
2194 txq_trans_update(txq
);
2195 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2196 return NETDEV_TX_BUSY
;
2197 } while (skb
->next
);
2200 if (likely(skb
->next
== NULL
))
2201 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2208 static u32 hashrnd __read_mostly
;
2211 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2212 * to be used as a distribution range.
2214 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2215 unsigned int num_tx_queues
)
2219 u16 qcount
= num_tx_queues
;
2221 if (skb_rx_queue_recorded(skb
)) {
2222 hash
= skb_get_rx_queue(skb
);
2223 while (unlikely(hash
>= num_tx_queues
))
2224 hash
-= num_tx_queues
;
2229 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2230 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2231 qcount
= dev
->tc_to_txq
[tc
].count
;
2234 if (skb
->sk
&& skb
->sk
->sk_hash
)
2235 hash
= skb
->sk
->sk_hash
;
2237 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2238 hash
= jhash_1word(hash
, hashrnd
);
2240 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2242 EXPORT_SYMBOL(__skb_tx_hash
);
2244 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2246 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2247 if (net_ratelimit()) {
2248 pr_warning("%s selects TX queue %d, but "
2249 "real number of TX queues is %d\n",
2250 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2257 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2260 struct xps_dev_maps
*dev_maps
;
2261 struct xps_map
*map
;
2262 int queue_index
= -1;
2265 dev_maps
= rcu_dereference(dev
->xps_maps
);
2267 map
= rcu_dereference(
2268 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2271 queue_index
= map
->queues
[0];
2274 if (skb
->sk
&& skb
->sk
->sk_hash
)
2275 hash
= skb
->sk
->sk_hash
;
2277 hash
= (__force u16
) skb
->protocol
^
2279 hash
= jhash_1word(hash
, hashrnd
);
2280 queue_index
= map
->queues
[
2281 ((u64
)hash
* map
->len
) >> 32];
2283 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2295 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2296 struct sk_buff
*skb
)
2299 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2301 if (dev
->real_num_tx_queues
== 1)
2303 else if (ops
->ndo_select_queue
) {
2304 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2305 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2307 struct sock
*sk
= skb
->sk
;
2308 queue_index
= sk_tx_queue_get(sk
);
2310 if (queue_index
< 0 || skb
->ooo_okay
||
2311 queue_index
>= dev
->real_num_tx_queues
) {
2312 int old_index
= queue_index
;
2314 queue_index
= get_xps_queue(dev
, skb
);
2315 if (queue_index
< 0)
2316 queue_index
= skb_tx_hash(dev
, skb
);
2318 if (queue_index
!= old_index
&& sk
) {
2319 struct dst_entry
*dst
=
2320 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2322 if (dst
&& skb_dst(skb
) == dst
)
2323 sk_tx_queue_set(sk
, queue_index
);
2328 skb_set_queue_mapping(skb
, queue_index
);
2329 return netdev_get_tx_queue(dev
, queue_index
);
2332 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2333 struct net_device
*dev
,
2334 struct netdev_queue
*txq
)
2336 spinlock_t
*root_lock
= qdisc_lock(q
);
2340 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2341 qdisc_calculate_pkt_len(skb
, q
);
2343 * Heuristic to force contended enqueues to serialize on a
2344 * separate lock before trying to get qdisc main lock.
2345 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2346 * and dequeue packets faster.
2348 contended
= qdisc_is_running(q
);
2349 if (unlikely(contended
))
2350 spin_lock(&q
->busylock
);
2352 spin_lock(root_lock
);
2353 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2356 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2357 qdisc_run_begin(q
)) {
2359 * This is a work-conserving queue; there are no old skbs
2360 * waiting to be sent out; and the qdisc is not running -
2361 * xmit the skb directly.
2363 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2366 qdisc_bstats_update(q
, skb
);
2368 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2369 if (unlikely(contended
)) {
2370 spin_unlock(&q
->busylock
);
2377 rc
= NET_XMIT_SUCCESS
;
2380 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2381 if (qdisc_run_begin(q
)) {
2382 if (unlikely(contended
)) {
2383 spin_unlock(&q
->busylock
);
2389 spin_unlock(root_lock
);
2390 if (unlikely(contended
))
2391 spin_unlock(&q
->busylock
);
2395 static DEFINE_PER_CPU(int, xmit_recursion
);
2396 #define RECURSION_LIMIT 10
2399 * dev_queue_xmit - transmit a buffer
2400 * @skb: buffer to transmit
2402 * Queue a buffer for transmission to a network device. The caller must
2403 * have set the device and priority and built the buffer before calling
2404 * this function. The function can be called from an interrupt.
2406 * A negative errno code is returned on a failure. A success does not
2407 * guarantee the frame will be transmitted as it may be dropped due
2408 * to congestion or traffic shaping.
2410 * -----------------------------------------------------------------------------------
2411 * I notice this method can also return errors from the queue disciplines,
2412 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2415 * Regardless of the return value, the skb is consumed, so it is currently
2416 * difficult to retry a send to this method. (You can bump the ref count
2417 * before sending to hold a reference for retry if you are careful.)
2419 * When calling this method, interrupts MUST be enabled. This is because
2420 * the BH enable code must have IRQs enabled so that it will not deadlock.
2423 int dev_queue_xmit(struct sk_buff
*skb
)
2425 struct net_device
*dev
= skb
->dev
;
2426 struct netdev_queue
*txq
;
2430 /* Disable soft irqs for various locks below. Also
2431 * stops preemption for RCU.
2435 txq
= dev_pick_tx(dev
, skb
);
2436 q
= rcu_dereference_bh(txq
->qdisc
);
2438 #ifdef CONFIG_NET_CLS_ACT
2439 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2441 trace_net_dev_queue(skb
);
2443 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2447 /* The device has no queue. Common case for software devices:
2448 loopback, all the sorts of tunnels...
2450 Really, it is unlikely that netif_tx_lock protection is necessary
2451 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2453 However, it is possible, that they rely on protection
2456 Check this and shot the lock. It is not prone from deadlocks.
2457 Either shot noqueue qdisc, it is even simpler 8)
2459 if (dev
->flags
& IFF_UP
) {
2460 int cpu
= smp_processor_id(); /* ok because BHs are off */
2462 if (txq
->xmit_lock_owner
!= cpu
) {
2464 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2465 goto recursion_alert
;
2467 HARD_TX_LOCK(dev
, txq
, cpu
);
2469 if (!netif_tx_queue_stopped(txq
)) {
2470 __this_cpu_inc(xmit_recursion
);
2471 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2472 __this_cpu_dec(xmit_recursion
);
2473 if (dev_xmit_complete(rc
)) {
2474 HARD_TX_UNLOCK(dev
, txq
);
2478 HARD_TX_UNLOCK(dev
, txq
);
2479 if (net_ratelimit())
2480 printk(KERN_CRIT
"Virtual device %s asks to "
2481 "queue packet!\n", dev
->name
);
2483 /* Recursion is detected! It is possible,
2487 if (net_ratelimit())
2488 printk(KERN_CRIT
"Dead loop on virtual device "
2489 "%s, fix it urgently!\n", dev
->name
);
2494 rcu_read_unlock_bh();
2499 rcu_read_unlock_bh();
2502 EXPORT_SYMBOL(dev_queue_xmit
);
2505 /*=======================================================================
2507 =======================================================================*/
2509 int netdev_max_backlog __read_mostly
= 1000;
2510 int netdev_tstamp_prequeue __read_mostly
= 1;
2511 int netdev_budget __read_mostly
= 300;
2512 int weight_p __read_mostly
= 64; /* old backlog weight */
2514 /* Called with irq disabled */
2515 static inline void ____napi_schedule(struct softnet_data
*sd
,
2516 struct napi_struct
*napi
)
2518 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2519 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2523 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2524 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2525 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2526 * if hash is a canonical 4-tuple hash over transport ports.
2528 void __skb_get_rxhash(struct sk_buff
*skb
)
2530 int nhoff
, hash
= 0, poff
;
2531 const struct ipv6hdr
*ip6
;
2532 const struct iphdr
*ip
;
2533 const struct vlan_hdr
*vlan
;
2542 nhoff
= skb_network_offset(skb
);
2543 proto
= skb
->protocol
;
2547 case __constant_htons(ETH_P_IP
):
2548 if (!pskb_may_pull(skb
, sizeof(*ip
) + nhoff
))
2551 ip
= (const struct iphdr
*) (skb
->data
+ nhoff
);
2552 if (ip_is_fragment(ip
))
2555 ip_proto
= ip
->protocol
;
2556 addr1
= (__force u32
) ip
->saddr
;
2557 addr2
= (__force u32
) ip
->daddr
;
2558 nhoff
+= ip
->ihl
* 4;
2560 case __constant_htons(ETH_P_IPV6
):
2561 if (!pskb_may_pull(skb
, sizeof(*ip6
) + nhoff
))
2564 ip6
= (const struct ipv6hdr
*) (skb
->data
+ nhoff
);
2565 ip_proto
= ip6
->nexthdr
;
2566 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2567 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2570 case __constant_htons(ETH_P_8021Q
):
2571 if (!pskb_may_pull(skb
, sizeof(*vlan
) + nhoff
))
2573 vlan
= (const struct vlan_hdr
*) (skb
->data
+ nhoff
);
2574 proto
= vlan
->h_vlan_encapsulated_proto
;
2575 nhoff
+= sizeof(*vlan
);
2577 case __constant_htons(ETH_P_PPP_SES
):
2578 if (!pskb_may_pull(skb
, PPPOE_SES_HLEN
+ nhoff
))
2580 proto
= *((__be16
*) (skb
->data
+ nhoff
+
2581 sizeof(struct pppoe_hdr
)));
2582 nhoff
+= PPPOE_SES_HLEN
;
2590 if (pskb_may_pull(skb
, nhoff
+ 16)) {
2591 u8
*h
= skb
->data
+ nhoff
;
2592 __be16 flags
= *(__be16
*)h
;
2595 * Only look inside GRE if version zero and no
2598 if (!(flags
& (GRE_VERSION
|GRE_ROUTING
))) {
2599 proto
= *(__be16
*)(h
+ 2);
2601 if (flags
& GRE_CSUM
)
2603 if (flags
& GRE_KEY
)
2605 if (flags
& GRE_SEQ
)
2616 poff
= proto_ports_offset(ip_proto
);
2619 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2620 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2621 if (ports
.v16
[1] < ports
.v16
[0])
2622 swap(ports
.v16
[0], ports
.v16
[1]);
2627 /* get a consistent hash (same value on both flow directions) */
2631 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2638 EXPORT_SYMBOL(__skb_get_rxhash
);
2642 /* One global table that all flow-based protocols share. */
2643 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2644 EXPORT_SYMBOL(rps_sock_flow_table
);
2646 static struct rps_dev_flow
*
2647 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2648 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2652 tcpu
= rflow
->cpu
= next_cpu
;
2653 if (tcpu
!= RPS_NO_CPU
) {
2654 #ifdef CONFIG_RFS_ACCEL
2655 struct netdev_rx_queue
*rxqueue
;
2656 struct rps_dev_flow_table
*flow_table
;
2657 struct rps_dev_flow
*old_rflow
;
2662 /* Should we steer this flow to a different hardware queue? */
2663 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2664 !(dev
->features
& NETIF_F_NTUPLE
))
2666 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2667 if (rxq_index
== skb_get_rx_queue(skb
))
2670 rxqueue
= dev
->_rx
+ rxq_index
;
2671 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2674 flow_id
= skb
->rxhash
& flow_table
->mask
;
2675 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2676 rxq_index
, flow_id
);
2680 rflow
= &flow_table
->flows
[flow_id
];
2681 rflow
->cpu
= next_cpu
;
2683 if (old_rflow
->filter
== rflow
->filter
)
2684 old_rflow
->filter
= RPS_NO_FILTER
;
2688 per_cpu(softnet_data
, tcpu
).input_queue_head
;
2695 * get_rps_cpu is called from netif_receive_skb and returns the target
2696 * CPU from the RPS map of the receiving queue for a given skb.
2697 * rcu_read_lock must be held on entry.
2699 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2700 struct rps_dev_flow
**rflowp
)
2702 struct netdev_rx_queue
*rxqueue
;
2703 struct rps_map
*map
;
2704 struct rps_dev_flow_table
*flow_table
;
2705 struct rps_sock_flow_table
*sock_flow_table
;
2709 if (skb_rx_queue_recorded(skb
)) {
2710 u16 index
= skb_get_rx_queue(skb
);
2711 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2712 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2713 "%s received packet on queue %u, but number "
2714 "of RX queues is %u\n",
2715 dev
->name
, index
, dev
->real_num_rx_queues
);
2718 rxqueue
= dev
->_rx
+ index
;
2722 map
= rcu_dereference(rxqueue
->rps_map
);
2724 if (map
->len
== 1 &&
2725 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2726 tcpu
= map
->cpus
[0];
2727 if (cpu_online(tcpu
))
2731 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2735 skb_reset_network_header(skb
);
2736 if (!skb_get_rxhash(skb
))
2739 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2740 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2741 if (flow_table
&& sock_flow_table
) {
2743 struct rps_dev_flow
*rflow
;
2745 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2748 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2749 sock_flow_table
->mask
];
2752 * If the desired CPU (where last recvmsg was done) is
2753 * different from current CPU (one in the rx-queue flow
2754 * table entry), switch if one of the following holds:
2755 * - Current CPU is unset (equal to RPS_NO_CPU).
2756 * - Current CPU is offline.
2757 * - The current CPU's queue tail has advanced beyond the
2758 * last packet that was enqueued using this table entry.
2759 * This guarantees that all previous packets for the flow
2760 * have been dequeued, thus preserving in order delivery.
2762 if (unlikely(tcpu
!= next_cpu
) &&
2763 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2764 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2765 rflow
->last_qtail
)) >= 0))
2766 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2768 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2776 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2778 if (cpu_online(tcpu
)) {
2788 #ifdef CONFIG_RFS_ACCEL
2791 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2792 * @dev: Device on which the filter was set
2793 * @rxq_index: RX queue index
2794 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2795 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2797 * Drivers that implement ndo_rx_flow_steer() should periodically call
2798 * this function for each installed filter and remove the filters for
2799 * which it returns %true.
2801 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2802 u32 flow_id
, u16 filter_id
)
2804 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2805 struct rps_dev_flow_table
*flow_table
;
2806 struct rps_dev_flow
*rflow
;
2811 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2812 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2813 rflow
= &flow_table
->flows
[flow_id
];
2814 cpu
= ACCESS_ONCE(rflow
->cpu
);
2815 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2816 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2817 rflow
->last_qtail
) <
2818 (int)(10 * flow_table
->mask
)))
2824 EXPORT_SYMBOL(rps_may_expire_flow
);
2826 #endif /* CONFIG_RFS_ACCEL */
2828 /* Called from hardirq (IPI) context */
2829 static void rps_trigger_softirq(void *data
)
2831 struct softnet_data
*sd
= data
;
2833 ____napi_schedule(sd
, &sd
->backlog
);
2837 #endif /* CONFIG_RPS */
2840 * Check if this softnet_data structure is another cpu one
2841 * If yes, queue it to our IPI list and return 1
2844 static int rps_ipi_queued(struct softnet_data
*sd
)
2847 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2850 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2851 mysd
->rps_ipi_list
= sd
;
2853 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2856 #endif /* CONFIG_RPS */
2861 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2862 * queue (may be a remote CPU queue).
2864 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2865 unsigned int *qtail
)
2867 struct softnet_data
*sd
;
2868 unsigned long flags
;
2870 sd
= &per_cpu(softnet_data
, cpu
);
2872 local_irq_save(flags
);
2875 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2876 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2878 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2879 input_queue_tail_incr_save(sd
, qtail
);
2881 local_irq_restore(flags
);
2882 return NET_RX_SUCCESS
;
2885 /* Schedule NAPI for backlog device
2886 * We can use non atomic operation since we own the queue lock
2888 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2889 if (!rps_ipi_queued(sd
))
2890 ____napi_schedule(sd
, &sd
->backlog
);
2898 local_irq_restore(flags
);
2900 atomic_long_inc(&skb
->dev
->rx_dropped
);
2906 * netif_rx - post buffer to the network code
2907 * @skb: buffer to post
2909 * This function receives a packet from a device driver and queues it for
2910 * the upper (protocol) levels to process. It always succeeds. The buffer
2911 * may be dropped during processing for congestion control or by the
2915 * NET_RX_SUCCESS (no congestion)
2916 * NET_RX_DROP (packet was dropped)
2920 int netif_rx(struct sk_buff
*skb
)
2924 /* if netpoll wants it, pretend we never saw it */
2925 if (netpoll_rx(skb
))
2928 if (netdev_tstamp_prequeue
)
2929 net_timestamp_check(skb
);
2931 trace_netif_rx(skb
);
2934 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2940 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2942 cpu
= smp_processor_id();
2944 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2952 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2958 EXPORT_SYMBOL(netif_rx
);
2960 int netif_rx_ni(struct sk_buff
*skb
)
2965 err
= netif_rx(skb
);
2966 if (local_softirq_pending())
2972 EXPORT_SYMBOL(netif_rx_ni
);
2974 static void net_tx_action(struct softirq_action
*h
)
2976 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2978 if (sd
->completion_queue
) {
2979 struct sk_buff
*clist
;
2981 local_irq_disable();
2982 clist
= sd
->completion_queue
;
2983 sd
->completion_queue
= NULL
;
2987 struct sk_buff
*skb
= clist
;
2988 clist
= clist
->next
;
2990 WARN_ON(atomic_read(&skb
->users
));
2991 trace_kfree_skb(skb
, net_tx_action
);
2996 if (sd
->output_queue
) {
2999 local_irq_disable();
3000 head
= sd
->output_queue
;
3001 sd
->output_queue
= NULL
;
3002 sd
->output_queue_tailp
= &sd
->output_queue
;
3006 struct Qdisc
*q
= head
;
3007 spinlock_t
*root_lock
;
3009 head
= head
->next_sched
;
3011 root_lock
= qdisc_lock(q
);
3012 if (spin_trylock(root_lock
)) {
3013 smp_mb__before_clear_bit();
3014 clear_bit(__QDISC_STATE_SCHED
,
3017 spin_unlock(root_lock
);
3019 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3021 __netif_reschedule(q
);
3023 smp_mb__before_clear_bit();
3024 clear_bit(__QDISC_STATE_SCHED
,
3032 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3033 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3034 /* This hook is defined here for ATM LANE */
3035 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3036 unsigned char *addr
) __read_mostly
;
3037 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3040 #ifdef CONFIG_NET_CLS_ACT
3041 /* TODO: Maybe we should just force sch_ingress to be compiled in
3042 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3043 * a compare and 2 stores extra right now if we dont have it on
3044 * but have CONFIG_NET_CLS_ACT
3045 * NOTE: This doesn't stop any functionality; if you dont have
3046 * the ingress scheduler, you just can't add policies on ingress.
3049 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3051 struct net_device
*dev
= skb
->dev
;
3052 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3053 int result
= TC_ACT_OK
;
3056 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3057 if (net_ratelimit())
3058 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3059 skb
->skb_iif
, dev
->ifindex
);
3063 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3064 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3067 if (q
!= &noop_qdisc
) {
3068 spin_lock(qdisc_lock(q
));
3069 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3070 result
= qdisc_enqueue_root(skb
, q
);
3071 spin_unlock(qdisc_lock(q
));
3077 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3078 struct packet_type
**pt_prev
,
3079 int *ret
, struct net_device
*orig_dev
)
3081 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3083 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3087 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3091 switch (ing_filter(skb
, rxq
)) {
3105 * netdev_rx_handler_register - register receive handler
3106 * @dev: device to register a handler for
3107 * @rx_handler: receive handler to register
3108 * @rx_handler_data: data pointer that is used by rx handler
3110 * Register a receive hander for a device. This handler will then be
3111 * called from __netif_receive_skb. A negative errno code is returned
3114 * The caller must hold the rtnl_mutex.
3116 * For a general description of rx_handler, see enum rx_handler_result.
3118 int netdev_rx_handler_register(struct net_device
*dev
,
3119 rx_handler_func_t
*rx_handler
,
3120 void *rx_handler_data
)
3124 if (dev
->rx_handler
)
3127 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3128 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3132 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3135 * netdev_rx_handler_unregister - unregister receive handler
3136 * @dev: device to unregister a handler from
3138 * Unregister a receive hander from a device.
3140 * The caller must hold the rtnl_mutex.
3142 void netdev_rx_handler_unregister(struct net_device
*dev
)
3146 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3147 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3149 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3151 static int __netif_receive_skb(struct sk_buff
*skb
)
3153 struct packet_type
*ptype
, *pt_prev
;
3154 rx_handler_func_t
*rx_handler
;
3155 struct net_device
*orig_dev
;
3156 struct net_device
*null_or_dev
;
3157 bool deliver_exact
= false;
3158 int ret
= NET_RX_DROP
;
3161 if (!netdev_tstamp_prequeue
)
3162 net_timestamp_check(skb
);
3164 trace_netif_receive_skb(skb
);
3166 /* if we've gotten here through NAPI, check netpoll */
3167 if (netpoll_receive_skb(skb
))
3171 skb
->skb_iif
= skb
->dev
->ifindex
;
3172 orig_dev
= skb
->dev
;
3174 skb_reset_network_header(skb
);
3175 skb_reset_transport_header(skb
);
3176 skb_reset_mac_len(skb
);
3184 __this_cpu_inc(softnet_data
.processed
);
3186 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3187 skb
= vlan_untag(skb
);
3192 #ifdef CONFIG_NET_CLS_ACT
3193 if (skb
->tc_verd
& TC_NCLS
) {
3194 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3199 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3200 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3202 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3207 #ifdef CONFIG_NET_CLS_ACT
3208 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3214 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3217 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3220 switch (rx_handler(&skb
)) {
3221 case RX_HANDLER_CONSUMED
:
3223 case RX_HANDLER_ANOTHER
:
3225 case RX_HANDLER_EXACT
:
3226 deliver_exact
= true;
3227 case RX_HANDLER_PASS
:
3234 if (vlan_tx_tag_present(skb
)) {
3236 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3239 if (vlan_do_receive(&skb
)) {
3240 ret
= __netif_receive_skb(skb
);
3242 } else if (unlikely(!skb
))
3246 /* deliver only exact match when indicated */
3247 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3249 type
= skb
->protocol
;
3250 list_for_each_entry_rcu(ptype
,
3251 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3252 if (ptype
->type
== type
&&
3253 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3254 ptype
->dev
== orig_dev
)) {
3256 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3262 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3264 atomic_long_inc(&skb
->dev
->rx_dropped
);
3266 /* Jamal, now you will not able to escape explaining
3267 * me how you were going to use this. :-)
3278 * netif_receive_skb - process receive buffer from network
3279 * @skb: buffer to process
3281 * netif_receive_skb() is the main receive data processing function.
3282 * It always succeeds. The buffer may be dropped during processing
3283 * for congestion control or by the protocol layers.
3285 * This function may only be called from softirq context and interrupts
3286 * should be enabled.
3288 * Return values (usually ignored):
3289 * NET_RX_SUCCESS: no congestion
3290 * NET_RX_DROP: packet was dropped
3292 int netif_receive_skb(struct sk_buff
*skb
)
3294 if (netdev_tstamp_prequeue
)
3295 net_timestamp_check(skb
);
3297 if (skb_defer_rx_timestamp(skb
))
3298 return NET_RX_SUCCESS
;
3302 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3307 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3310 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3314 ret
= __netif_receive_skb(skb
);
3320 return __netif_receive_skb(skb
);
3323 EXPORT_SYMBOL(netif_receive_skb
);
3325 /* Network device is going away, flush any packets still pending
3326 * Called with irqs disabled.
3328 static void flush_backlog(void *arg
)
3330 struct net_device
*dev
= arg
;
3331 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3332 struct sk_buff
*skb
, *tmp
;
3335 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3336 if (skb
->dev
== dev
) {
3337 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3339 input_queue_head_incr(sd
);
3344 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3345 if (skb
->dev
== dev
) {
3346 __skb_unlink(skb
, &sd
->process_queue
);
3348 input_queue_head_incr(sd
);
3353 static int napi_gro_complete(struct sk_buff
*skb
)
3355 struct packet_type
*ptype
;
3356 __be16 type
= skb
->protocol
;
3357 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3360 if (NAPI_GRO_CB(skb
)->count
== 1) {
3361 skb_shinfo(skb
)->gso_size
= 0;
3366 list_for_each_entry_rcu(ptype
, head
, list
) {
3367 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3370 err
= ptype
->gro_complete(skb
);
3376 WARN_ON(&ptype
->list
== head
);
3378 return NET_RX_SUCCESS
;
3382 return netif_receive_skb(skb
);
3385 inline void napi_gro_flush(struct napi_struct
*napi
)
3387 struct sk_buff
*skb
, *next
;
3389 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3392 napi_gro_complete(skb
);
3395 napi
->gro_count
= 0;
3396 napi
->gro_list
= NULL
;
3398 EXPORT_SYMBOL(napi_gro_flush
);
3400 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3402 struct sk_buff
**pp
= NULL
;
3403 struct packet_type
*ptype
;
3404 __be16 type
= skb
->protocol
;
3405 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3408 enum gro_result ret
;
3410 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3413 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3417 list_for_each_entry_rcu(ptype
, head
, list
) {
3418 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3421 skb_set_network_header(skb
, skb_gro_offset(skb
));
3422 mac_len
= skb
->network_header
- skb
->mac_header
;
3423 skb
->mac_len
= mac_len
;
3424 NAPI_GRO_CB(skb
)->same_flow
= 0;
3425 NAPI_GRO_CB(skb
)->flush
= 0;
3426 NAPI_GRO_CB(skb
)->free
= 0;
3428 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3433 if (&ptype
->list
== head
)
3436 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3437 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3440 struct sk_buff
*nskb
= *pp
;
3444 napi_gro_complete(nskb
);
3451 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3455 NAPI_GRO_CB(skb
)->count
= 1;
3456 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3457 skb
->next
= napi
->gro_list
;
3458 napi
->gro_list
= skb
;
3462 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3463 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3465 BUG_ON(skb
->end
- skb
->tail
< grow
);
3467 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3470 skb
->data_len
-= grow
;
3472 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3473 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3475 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3476 put_page(skb_shinfo(skb
)->frags
[0].page
);
3477 memmove(skb_shinfo(skb
)->frags
,
3478 skb_shinfo(skb
)->frags
+ 1,
3479 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3490 EXPORT_SYMBOL(dev_gro_receive
);
3492 static inline gro_result_t
3493 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3497 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3498 unsigned long diffs
;
3500 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3501 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3502 diffs
|= compare_ether_header(skb_mac_header(p
),
3503 skb_gro_mac_header(skb
));
3504 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3505 NAPI_GRO_CB(p
)->flush
= 0;
3508 return dev_gro_receive(napi
, skb
);
3511 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3515 if (netif_receive_skb(skb
))
3520 case GRO_MERGED_FREE
:
3531 EXPORT_SYMBOL(napi_skb_finish
);
3533 void skb_gro_reset_offset(struct sk_buff
*skb
)
3535 NAPI_GRO_CB(skb
)->data_offset
= 0;
3536 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3537 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3539 if (skb
->mac_header
== skb
->tail
&&
3540 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3541 NAPI_GRO_CB(skb
)->frag0
=
3542 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3543 skb_shinfo(skb
)->frags
[0].page_offset
;
3544 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3547 EXPORT_SYMBOL(skb_gro_reset_offset
);
3549 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3551 skb_gro_reset_offset(skb
);
3553 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3555 EXPORT_SYMBOL(napi_gro_receive
);
3557 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3559 __skb_pull(skb
, skb_headlen(skb
));
3560 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3562 skb
->dev
= napi
->dev
;
3568 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3570 struct sk_buff
*skb
= napi
->skb
;
3573 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3579 EXPORT_SYMBOL(napi_get_frags
);
3581 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3587 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3589 if (ret
== GRO_HELD
)
3590 skb_gro_pull(skb
, -ETH_HLEN
);
3591 else if (netif_receive_skb(skb
))
3596 case GRO_MERGED_FREE
:
3597 napi_reuse_skb(napi
, skb
);
3606 EXPORT_SYMBOL(napi_frags_finish
);
3608 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3610 struct sk_buff
*skb
= napi
->skb
;
3617 skb_reset_mac_header(skb
);
3618 skb_gro_reset_offset(skb
);
3620 off
= skb_gro_offset(skb
);
3621 hlen
= off
+ sizeof(*eth
);
3622 eth
= skb_gro_header_fast(skb
, off
);
3623 if (skb_gro_header_hard(skb
, hlen
)) {
3624 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3625 if (unlikely(!eth
)) {
3626 napi_reuse_skb(napi
, skb
);
3632 skb_gro_pull(skb
, sizeof(*eth
));
3635 * This works because the only protocols we care about don't require
3636 * special handling. We'll fix it up properly at the end.
3638 skb
->protocol
= eth
->h_proto
;
3643 EXPORT_SYMBOL(napi_frags_skb
);
3645 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3647 struct sk_buff
*skb
= napi_frags_skb(napi
);
3652 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3654 EXPORT_SYMBOL(napi_gro_frags
);
3657 * net_rps_action sends any pending IPI's for rps.
3658 * Note: called with local irq disabled, but exits with local irq enabled.
3660 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3663 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3666 sd
->rps_ipi_list
= NULL
;
3670 /* Send pending IPI's to kick RPS processing on remote cpus. */
3672 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3674 if (cpu_online(remsd
->cpu
))
3675 __smp_call_function_single(remsd
->cpu
,
3684 static int process_backlog(struct napi_struct
*napi
, int quota
)
3687 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3690 /* Check if we have pending ipi, its better to send them now,
3691 * not waiting net_rx_action() end.
3693 if (sd
->rps_ipi_list
) {
3694 local_irq_disable();
3695 net_rps_action_and_irq_enable(sd
);
3698 napi
->weight
= weight_p
;
3699 local_irq_disable();
3700 while (work
< quota
) {
3701 struct sk_buff
*skb
;
3704 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3706 __netif_receive_skb(skb
);
3707 local_irq_disable();
3708 input_queue_head_incr(sd
);
3709 if (++work
>= quota
) {
3716 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3718 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3719 &sd
->process_queue
);
3721 if (qlen
< quota
- work
) {
3723 * Inline a custom version of __napi_complete().
3724 * only current cpu owns and manipulates this napi,
3725 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3726 * we can use a plain write instead of clear_bit(),
3727 * and we dont need an smp_mb() memory barrier.
3729 list_del(&napi
->poll_list
);
3732 quota
= work
+ qlen
;
3742 * __napi_schedule - schedule for receive
3743 * @n: entry to schedule
3745 * The entry's receive function will be scheduled to run
3747 void __napi_schedule(struct napi_struct
*n
)
3749 unsigned long flags
;
3751 local_irq_save(flags
);
3752 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3753 local_irq_restore(flags
);
3755 EXPORT_SYMBOL(__napi_schedule
);
3757 void __napi_complete(struct napi_struct
*n
)
3759 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3760 BUG_ON(n
->gro_list
);
3762 list_del(&n
->poll_list
);
3763 smp_mb__before_clear_bit();
3764 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3766 EXPORT_SYMBOL(__napi_complete
);
3768 void napi_complete(struct napi_struct
*n
)
3770 unsigned long flags
;
3773 * don't let napi dequeue from the cpu poll list
3774 * just in case its running on a different cpu
3776 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3780 local_irq_save(flags
);
3782 local_irq_restore(flags
);
3784 EXPORT_SYMBOL(napi_complete
);
3786 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3787 int (*poll
)(struct napi_struct
*, int), int weight
)
3789 INIT_LIST_HEAD(&napi
->poll_list
);
3790 napi
->gro_count
= 0;
3791 napi
->gro_list
= NULL
;
3794 napi
->weight
= weight
;
3795 list_add(&napi
->dev_list
, &dev
->napi_list
);
3797 #ifdef CONFIG_NETPOLL
3798 spin_lock_init(&napi
->poll_lock
);
3799 napi
->poll_owner
= -1;
3801 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3803 EXPORT_SYMBOL(netif_napi_add
);
3805 void netif_napi_del(struct napi_struct
*napi
)
3807 struct sk_buff
*skb
, *next
;
3809 list_del_init(&napi
->dev_list
);
3810 napi_free_frags(napi
);
3812 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3818 napi
->gro_list
= NULL
;
3819 napi
->gro_count
= 0;
3821 EXPORT_SYMBOL(netif_napi_del
);
3823 static void net_rx_action(struct softirq_action
*h
)
3825 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3826 unsigned long time_limit
= jiffies
+ 2;
3827 int budget
= netdev_budget
;
3830 local_irq_disable();
3832 while (!list_empty(&sd
->poll_list
)) {
3833 struct napi_struct
*n
;
3836 /* If softirq window is exhuasted then punt.
3837 * Allow this to run for 2 jiffies since which will allow
3838 * an average latency of 1.5/HZ.
3840 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3845 /* Even though interrupts have been re-enabled, this
3846 * access is safe because interrupts can only add new
3847 * entries to the tail of this list, and only ->poll()
3848 * calls can remove this head entry from the list.
3850 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3852 have
= netpoll_poll_lock(n
);
3856 /* This NAPI_STATE_SCHED test is for avoiding a race
3857 * with netpoll's poll_napi(). Only the entity which
3858 * obtains the lock and sees NAPI_STATE_SCHED set will
3859 * actually make the ->poll() call. Therefore we avoid
3860 * accidentally calling ->poll() when NAPI is not scheduled.
3863 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3864 work
= n
->poll(n
, weight
);
3868 WARN_ON_ONCE(work
> weight
);
3872 local_irq_disable();
3874 /* Drivers must not modify the NAPI state if they
3875 * consume the entire weight. In such cases this code
3876 * still "owns" the NAPI instance and therefore can
3877 * move the instance around on the list at-will.
3879 if (unlikely(work
== weight
)) {
3880 if (unlikely(napi_disable_pending(n
))) {
3883 local_irq_disable();
3885 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3888 netpoll_poll_unlock(have
);
3891 net_rps_action_and_irq_enable(sd
);
3893 #ifdef CONFIG_NET_DMA
3895 * There may not be any more sk_buffs coming right now, so push
3896 * any pending DMA copies to hardware
3898 dma_issue_pending_all();
3905 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3909 static gifconf_func_t
*gifconf_list
[NPROTO
];
3912 * register_gifconf - register a SIOCGIF handler
3913 * @family: Address family
3914 * @gifconf: Function handler
3916 * Register protocol dependent address dumping routines. The handler
3917 * that is passed must not be freed or reused until it has been replaced
3918 * by another handler.
3920 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3922 if (family
>= NPROTO
)
3924 gifconf_list
[family
] = gifconf
;
3927 EXPORT_SYMBOL(register_gifconf
);
3931 * Map an interface index to its name (SIOCGIFNAME)
3935 * We need this ioctl for efficient implementation of the
3936 * if_indextoname() function required by the IPv6 API. Without
3937 * it, we would have to search all the interfaces to find a
3941 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3943 struct net_device
*dev
;
3947 * Fetch the caller's info block.
3950 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3954 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3960 strcpy(ifr
.ifr_name
, dev
->name
);
3963 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3969 * Perform a SIOCGIFCONF call. This structure will change
3970 * size eventually, and there is nothing I can do about it.
3971 * Thus we will need a 'compatibility mode'.
3974 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3977 struct net_device
*dev
;
3984 * Fetch the caller's info block.
3987 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3994 * Loop over the interfaces, and write an info block for each.
3998 for_each_netdev(net
, dev
) {
3999 for (i
= 0; i
< NPROTO
; i
++) {
4000 if (gifconf_list
[i
]) {
4003 done
= gifconf_list
[i
](dev
, NULL
, 0);
4005 done
= gifconf_list
[i
](dev
, pos
+ total
,
4015 * All done. Write the updated control block back to the caller.
4017 ifc
.ifc_len
= total
;
4020 * Both BSD and Solaris return 0 here, so we do too.
4022 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4025 #ifdef CONFIG_PROC_FS
4027 * This is invoked by the /proc filesystem handler to display a device
4030 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4033 struct net
*net
= seq_file_net(seq
);
4035 struct net_device
*dev
;
4039 return SEQ_START_TOKEN
;
4042 for_each_netdev_rcu(net
, dev
)
4049 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4051 struct net_device
*dev
= v
;
4053 if (v
== SEQ_START_TOKEN
)
4054 dev
= first_net_device_rcu(seq_file_net(seq
));
4056 dev
= next_net_device_rcu(dev
);
4062 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4068 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4070 struct rtnl_link_stats64 temp
;
4071 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4073 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4074 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4075 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4077 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4078 stats
->rx_fifo_errors
,
4079 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4080 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4081 stats
->rx_compressed
, stats
->multicast
,
4082 stats
->tx_bytes
, stats
->tx_packets
,
4083 stats
->tx_errors
, stats
->tx_dropped
,
4084 stats
->tx_fifo_errors
, stats
->collisions
,
4085 stats
->tx_carrier_errors
+
4086 stats
->tx_aborted_errors
+
4087 stats
->tx_window_errors
+
4088 stats
->tx_heartbeat_errors
,
4089 stats
->tx_compressed
);
4093 * Called from the PROCfs module. This now uses the new arbitrary sized
4094 * /proc/net interface to create /proc/net/dev
4096 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4098 if (v
== SEQ_START_TOKEN
)
4099 seq_puts(seq
, "Inter-| Receive "
4101 " face |bytes packets errs drop fifo frame "
4102 "compressed multicast|bytes packets errs "
4103 "drop fifo colls carrier compressed\n");
4105 dev_seq_printf_stats(seq
, v
);
4109 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4111 struct softnet_data
*sd
= NULL
;
4113 while (*pos
< nr_cpu_ids
)
4114 if (cpu_online(*pos
)) {
4115 sd
= &per_cpu(softnet_data
, *pos
);
4122 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4124 return softnet_get_online(pos
);
4127 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4130 return softnet_get_online(pos
);
4133 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4137 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4139 struct softnet_data
*sd
= v
;
4141 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4142 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4143 0, 0, 0, 0, /* was fastroute */
4144 sd
->cpu_collision
, sd
->received_rps
);
4148 static const struct seq_operations dev_seq_ops
= {
4149 .start
= dev_seq_start
,
4150 .next
= dev_seq_next
,
4151 .stop
= dev_seq_stop
,
4152 .show
= dev_seq_show
,
4155 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4157 return seq_open_net(inode
, file
, &dev_seq_ops
,
4158 sizeof(struct seq_net_private
));
4161 static const struct file_operations dev_seq_fops
= {
4162 .owner
= THIS_MODULE
,
4163 .open
= dev_seq_open
,
4165 .llseek
= seq_lseek
,
4166 .release
= seq_release_net
,
4169 static const struct seq_operations softnet_seq_ops
= {
4170 .start
= softnet_seq_start
,
4171 .next
= softnet_seq_next
,
4172 .stop
= softnet_seq_stop
,
4173 .show
= softnet_seq_show
,
4176 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4178 return seq_open(file
, &softnet_seq_ops
);
4181 static const struct file_operations softnet_seq_fops
= {
4182 .owner
= THIS_MODULE
,
4183 .open
= softnet_seq_open
,
4185 .llseek
= seq_lseek
,
4186 .release
= seq_release
,
4189 static void *ptype_get_idx(loff_t pos
)
4191 struct packet_type
*pt
= NULL
;
4195 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4201 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4202 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4211 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4215 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4218 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4220 struct packet_type
*pt
;
4221 struct list_head
*nxt
;
4225 if (v
== SEQ_START_TOKEN
)
4226 return ptype_get_idx(0);
4229 nxt
= pt
->list
.next
;
4230 if (pt
->type
== htons(ETH_P_ALL
)) {
4231 if (nxt
!= &ptype_all
)
4234 nxt
= ptype_base
[0].next
;
4236 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4238 while (nxt
== &ptype_base
[hash
]) {
4239 if (++hash
>= PTYPE_HASH_SIZE
)
4241 nxt
= ptype_base
[hash
].next
;
4244 return list_entry(nxt
, struct packet_type
, list
);
4247 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4253 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4255 struct packet_type
*pt
= v
;
4257 if (v
== SEQ_START_TOKEN
)
4258 seq_puts(seq
, "Type Device Function\n");
4259 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4260 if (pt
->type
== htons(ETH_P_ALL
))
4261 seq_puts(seq
, "ALL ");
4263 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4265 seq_printf(seq
, " %-8s %pF\n",
4266 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4272 static const struct seq_operations ptype_seq_ops
= {
4273 .start
= ptype_seq_start
,
4274 .next
= ptype_seq_next
,
4275 .stop
= ptype_seq_stop
,
4276 .show
= ptype_seq_show
,
4279 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4281 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4282 sizeof(struct seq_net_private
));
4285 static const struct file_operations ptype_seq_fops
= {
4286 .owner
= THIS_MODULE
,
4287 .open
= ptype_seq_open
,
4289 .llseek
= seq_lseek
,
4290 .release
= seq_release_net
,
4294 static int __net_init
dev_proc_net_init(struct net
*net
)
4298 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4300 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4302 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4305 if (wext_proc_init(net
))
4311 proc_net_remove(net
, "ptype");
4313 proc_net_remove(net
, "softnet_stat");
4315 proc_net_remove(net
, "dev");
4319 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4321 wext_proc_exit(net
);
4323 proc_net_remove(net
, "ptype");
4324 proc_net_remove(net
, "softnet_stat");
4325 proc_net_remove(net
, "dev");
4328 static struct pernet_operations __net_initdata dev_proc_ops
= {
4329 .init
= dev_proc_net_init
,
4330 .exit
= dev_proc_net_exit
,
4333 static int __init
dev_proc_init(void)
4335 return register_pernet_subsys(&dev_proc_ops
);
4338 #define dev_proc_init() 0
4339 #endif /* CONFIG_PROC_FS */
4343 * netdev_set_master - set up master pointer
4344 * @slave: slave device
4345 * @master: new master device
4347 * Changes the master device of the slave. Pass %NULL to break the
4348 * bonding. The caller must hold the RTNL semaphore. On a failure
4349 * a negative errno code is returned. On success the reference counts
4350 * are adjusted and the function returns zero.
4352 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4354 struct net_device
*old
= slave
->master
;
4364 slave
->master
= master
;
4370 EXPORT_SYMBOL(netdev_set_master
);
4373 * netdev_set_bond_master - set up bonding master/slave pair
4374 * @slave: slave device
4375 * @master: new master device
4377 * Changes the master device of the slave. Pass %NULL to break the
4378 * bonding. The caller must hold the RTNL semaphore. On a failure
4379 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4380 * to the routing socket and the function returns zero.
4382 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4388 err
= netdev_set_master(slave
, master
);
4392 slave
->flags
|= IFF_SLAVE
;
4394 slave
->flags
&= ~IFF_SLAVE
;
4396 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4399 EXPORT_SYMBOL(netdev_set_bond_master
);
4401 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4403 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4405 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4406 ops
->ndo_change_rx_flags(dev
, flags
);
4409 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4411 unsigned short old_flags
= dev
->flags
;
4417 dev
->flags
|= IFF_PROMISC
;
4418 dev
->promiscuity
+= inc
;
4419 if (dev
->promiscuity
== 0) {
4422 * If inc causes overflow, untouch promisc and return error.
4425 dev
->flags
&= ~IFF_PROMISC
;
4427 dev
->promiscuity
-= inc
;
4428 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4429 "set promiscuity failed, promiscuity feature "
4430 "of device might be broken.\n", dev
->name
);
4434 if (dev
->flags
!= old_flags
) {
4435 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4436 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4438 if (audit_enabled
) {
4439 current_uid_gid(&uid
, &gid
);
4440 audit_log(current
->audit_context
, GFP_ATOMIC
,
4441 AUDIT_ANOM_PROMISCUOUS
,
4442 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4443 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4444 (old_flags
& IFF_PROMISC
),
4445 audit_get_loginuid(current
),
4447 audit_get_sessionid(current
));
4450 dev_change_rx_flags(dev
, IFF_PROMISC
);
4456 * dev_set_promiscuity - update promiscuity count on a device
4460 * Add or remove promiscuity from a device. While the count in the device
4461 * remains above zero the interface remains promiscuous. Once it hits zero
4462 * the device reverts back to normal filtering operation. A negative inc
4463 * value is used to drop promiscuity on the device.
4464 * Return 0 if successful or a negative errno code on error.
4466 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4468 unsigned short old_flags
= dev
->flags
;
4471 err
= __dev_set_promiscuity(dev
, inc
);
4474 if (dev
->flags
!= old_flags
)
4475 dev_set_rx_mode(dev
);
4478 EXPORT_SYMBOL(dev_set_promiscuity
);
4481 * dev_set_allmulti - update allmulti count on a device
4485 * Add or remove reception of all multicast frames to a device. While the
4486 * count in the device remains above zero the interface remains listening
4487 * to all interfaces. Once it hits zero the device reverts back to normal
4488 * filtering operation. A negative @inc value is used to drop the counter
4489 * when releasing a resource needing all multicasts.
4490 * Return 0 if successful or a negative errno code on error.
4493 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4495 unsigned short old_flags
= dev
->flags
;
4499 dev
->flags
|= IFF_ALLMULTI
;
4500 dev
->allmulti
+= inc
;
4501 if (dev
->allmulti
== 0) {
4504 * If inc causes overflow, untouch allmulti and return error.
4507 dev
->flags
&= ~IFF_ALLMULTI
;
4509 dev
->allmulti
-= inc
;
4510 printk(KERN_WARNING
"%s: allmulti touches roof, "
4511 "set allmulti failed, allmulti feature of "
4512 "device might be broken.\n", dev
->name
);
4516 if (dev
->flags
^ old_flags
) {
4517 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4518 dev_set_rx_mode(dev
);
4522 EXPORT_SYMBOL(dev_set_allmulti
);
4525 * Upload unicast and multicast address lists to device and
4526 * configure RX filtering. When the device doesn't support unicast
4527 * filtering it is put in promiscuous mode while unicast addresses
4530 void __dev_set_rx_mode(struct net_device
*dev
)
4532 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4534 /* dev_open will call this function so the list will stay sane. */
4535 if (!(dev
->flags
&IFF_UP
))
4538 if (!netif_device_present(dev
))
4541 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4542 /* Unicast addresses changes may only happen under the rtnl,
4543 * therefore calling __dev_set_promiscuity here is safe.
4545 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4546 __dev_set_promiscuity(dev
, 1);
4547 dev
->uc_promisc
= true;
4548 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4549 __dev_set_promiscuity(dev
, -1);
4550 dev
->uc_promisc
= false;
4554 if (ops
->ndo_set_rx_mode
)
4555 ops
->ndo_set_rx_mode(dev
);
4558 void dev_set_rx_mode(struct net_device
*dev
)
4560 netif_addr_lock_bh(dev
);
4561 __dev_set_rx_mode(dev
);
4562 netif_addr_unlock_bh(dev
);
4566 * dev_ethtool_get_settings - call device's ethtool_ops::get_settings()
4568 * @cmd: memory area for ethtool_ops::get_settings() result
4570 * The cmd arg is initialized properly (cleared and
4571 * ethtool_cmd::cmd field set to ETHTOOL_GSET).
4573 * Return device's ethtool_ops::get_settings() result value or
4574 * -EOPNOTSUPP when device doesn't expose
4575 * ethtool_ops::get_settings() operation.
4577 int dev_ethtool_get_settings(struct net_device
*dev
,
4578 struct ethtool_cmd
*cmd
)
4580 if (!dev
->ethtool_ops
|| !dev
->ethtool_ops
->get_settings
)
4583 memset(cmd
, 0, sizeof(struct ethtool_cmd
));
4584 cmd
->cmd
= ETHTOOL_GSET
;
4585 return dev
->ethtool_ops
->get_settings(dev
, cmd
);
4587 EXPORT_SYMBOL(dev_ethtool_get_settings
);
4590 * dev_get_flags - get flags reported to userspace
4593 * Get the combination of flag bits exported through APIs to userspace.
4595 unsigned dev_get_flags(const struct net_device
*dev
)
4599 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4604 (dev
->gflags
& (IFF_PROMISC
|
4607 if (netif_running(dev
)) {
4608 if (netif_oper_up(dev
))
4609 flags
|= IFF_RUNNING
;
4610 if (netif_carrier_ok(dev
))
4611 flags
|= IFF_LOWER_UP
;
4612 if (netif_dormant(dev
))
4613 flags
|= IFF_DORMANT
;
4618 EXPORT_SYMBOL(dev_get_flags
);
4620 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4622 int old_flags
= dev
->flags
;
4628 * Set the flags on our device.
4631 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4632 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4634 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4638 * Load in the correct multicast list now the flags have changed.
4641 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4642 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4644 dev_set_rx_mode(dev
);
4647 * Have we downed the interface. We handle IFF_UP ourselves
4648 * according to user attempts to set it, rather than blindly
4653 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4654 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4657 dev_set_rx_mode(dev
);
4660 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4661 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4663 dev
->gflags
^= IFF_PROMISC
;
4664 dev_set_promiscuity(dev
, inc
);
4667 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4668 is important. Some (broken) drivers set IFF_PROMISC, when
4669 IFF_ALLMULTI is requested not asking us and not reporting.
4671 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4672 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4674 dev
->gflags
^= IFF_ALLMULTI
;
4675 dev_set_allmulti(dev
, inc
);
4681 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4683 unsigned int changes
= dev
->flags
^ old_flags
;
4685 if (changes
& IFF_UP
) {
4686 if (dev
->flags
& IFF_UP
)
4687 call_netdevice_notifiers(NETDEV_UP
, dev
);
4689 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4692 if (dev
->flags
& IFF_UP
&&
4693 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4694 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4698 * dev_change_flags - change device settings
4700 * @flags: device state flags
4702 * Change settings on device based state flags. The flags are
4703 * in the userspace exported format.
4705 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4708 int old_flags
= dev
->flags
;
4710 ret
= __dev_change_flags(dev
, flags
);
4714 changes
= old_flags
^ dev
->flags
;
4716 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4718 __dev_notify_flags(dev
, old_flags
);
4721 EXPORT_SYMBOL(dev_change_flags
);
4724 * dev_set_mtu - Change maximum transfer unit
4726 * @new_mtu: new transfer unit
4728 * Change the maximum transfer size of the network device.
4730 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4732 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4735 if (new_mtu
== dev
->mtu
)
4738 /* MTU must be positive. */
4742 if (!netif_device_present(dev
))
4746 if (ops
->ndo_change_mtu
)
4747 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4751 if (!err
&& dev
->flags
& IFF_UP
)
4752 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4755 EXPORT_SYMBOL(dev_set_mtu
);
4758 * dev_set_group - Change group this device belongs to
4760 * @new_group: group this device should belong to
4762 void dev_set_group(struct net_device
*dev
, int new_group
)
4764 dev
->group
= new_group
;
4766 EXPORT_SYMBOL(dev_set_group
);
4769 * dev_set_mac_address - Change Media Access Control Address
4773 * Change the hardware (MAC) address of the device
4775 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4777 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4780 if (!ops
->ndo_set_mac_address
)
4782 if (sa
->sa_family
!= dev
->type
)
4784 if (!netif_device_present(dev
))
4786 err
= ops
->ndo_set_mac_address(dev
, sa
);
4788 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4791 EXPORT_SYMBOL(dev_set_mac_address
);
4794 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4796 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4799 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4805 case SIOCGIFFLAGS
: /* Get interface flags */
4806 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4809 case SIOCGIFMETRIC
: /* Get the metric on the interface
4810 (currently unused) */
4811 ifr
->ifr_metric
= 0;
4814 case SIOCGIFMTU
: /* Get the MTU of a device */
4815 ifr
->ifr_mtu
= dev
->mtu
;
4820 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4822 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4823 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4824 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4832 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4833 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4834 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4835 ifr
->ifr_map
.irq
= dev
->irq
;
4836 ifr
->ifr_map
.dma
= dev
->dma
;
4837 ifr
->ifr_map
.port
= dev
->if_port
;
4841 ifr
->ifr_ifindex
= dev
->ifindex
;
4845 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4849 /* dev_ioctl() should ensure this case
4861 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4863 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4866 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4867 const struct net_device_ops
*ops
;
4872 ops
= dev
->netdev_ops
;
4875 case SIOCSIFFLAGS
: /* Set interface flags */
4876 return dev_change_flags(dev
, ifr
->ifr_flags
);
4878 case SIOCSIFMETRIC
: /* Set the metric on the interface
4879 (currently unused) */
4882 case SIOCSIFMTU
: /* Set the MTU of a device */
4883 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4886 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4888 case SIOCSIFHWBROADCAST
:
4889 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4891 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4892 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4893 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4897 if (ops
->ndo_set_config
) {
4898 if (!netif_device_present(dev
))
4900 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4905 if (!ops
->ndo_set_rx_mode
||
4906 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4908 if (!netif_device_present(dev
))
4910 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4913 if (!ops
->ndo_set_rx_mode
||
4914 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4916 if (!netif_device_present(dev
))
4918 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4921 if (ifr
->ifr_qlen
< 0)
4923 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4927 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4928 return dev_change_name(dev
, ifr
->ifr_newname
);
4931 * Unknown or private ioctl
4934 if ((cmd
>= SIOCDEVPRIVATE
&&
4935 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4936 cmd
== SIOCBONDENSLAVE
||
4937 cmd
== SIOCBONDRELEASE
||
4938 cmd
== SIOCBONDSETHWADDR
||
4939 cmd
== SIOCBONDSLAVEINFOQUERY
||
4940 cmd
== SIOCBONDINFOQUERY
||
4941 cmd
== SIOCBONDCHANGEACTIVE
||
4942 cmd
== SIOCGMIIPHY
||
4943 cmd
== SIOCGMIIREG
||
4944 cmd
== SIOCSMIIREG
||
4945 cmd
== SIOCBRADDIF
||
4946 cmd
== SIOCBRDELIF
||
4947 cmd
== SIOCSHWTSTAMP
||
4948 cmd
== SIOCWANDEV
) {
4950 if (ops
->ndo_do_ioctl
) {
4951 if (netif_device_present(dev
))
4952 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4964 * This function handles all "interface"-type I/O control requests. The actual
4965 * 'doing' part of this is dev_ifsioc above.
4969 * dev_ioctl - network device ioctl
4970 * @net: the applicable net namespace
4971 * @cmd: command to issue
4972 * @arg: pointer to a struct ifreq in user space
4974 * Issue ioctl functions to devices. This is normally called by the
4975 * user space syscall interfaces but can sometimes be useful for
4976 * other purposes. The return value is the return from the syscall if
4977 * positive or a negative errno code on error.
4980 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4986 /* One special case: SIOCGIFCONF takes ifconf argument
4987 and requires shared lock, because it sleeps writing
4991 if (cmd
== SIOCGIFCONF
) {
4993 ret
= dev_ifconf(net
, (char __user
*) arg
);
4997 if (cmd
== SIOCGIFNAME
)
4998 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5000 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5003 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5005 colon
= strchr(ifr
.ifr_name
, ':');
5010 * See which interface the caller is talking about.
5015 * These ioctl calls:
5016 * - can be done by all.
5017 * - atomic and do not require locking.
5028 dev_load(net
, ifr
.ifr_name
);
5030 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5035 if (copy_to_user(arg
, &ifr
,
5036 sizeof(struct ifreq
)))
5042 dev_load(net
, ifr
.ifr_name
);
5044 ret
= dev_ethtool(net
, &ifr
);
5049 if (copy_to_user(arg
, &ifr
,
5050 sizeof(struct ifreq
)))
5056 * These ioctl calls:
5057 * - require superuser power.
5058 * - require strict serialization.
5064 if (!capable(CAP_NET_ADMIN
))
5066 dev_load(net
, ifr
.ifr_name
);
5068 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5073 if (copy_to_user(arg
, &ifr
,
5074 sizeof(struct ifreq
)))
5080 * These ioctl calls:
5081 * - require superuser power.
5082 * - require strict serialization.
5083 * - do not return a value
5093 case SIOCSIFHWBROADCAST
:
5096 case SIOCBONDENSLAVE
:
5097 case SIOCBONDRELEASE
:
5098 case SIOCBONDSETHWADDR
:
5099 case SIOCBONDCHANGEACTIVE
:
5103 if (!capable(CAP_NET_ADMIN
))
5106 case SIOCBONDSLAVEINFOQUERY
:
5107 case SIOCBONDINFOQUERY
:
5108 dev_load(net
, ifr
.ifr_name
);
5110 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5115 /* Get the per device memory space. We can add this but
5116 * currently do not support it */
5118 /* Set the per device memory buffer space.
5119 * Not applicable in our case */
5124 * Unknown or private ioctl.
5127 if (cmd
== SIOCWANDEV
||
5128 (cmd
>= SIOCDEVPRIVATE
&&
5129 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5130 dev_load(net
, ifr
.ifr_name
);
5132 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5134 if (!ret
&& copy_to_user(arg
, &ifr
,
5135 sizeof(struct ifreq
)))
5139 /* Take care of Wireless Extensions */
5140 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5141 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5148 * dev_new_index - allocate an ifindex
5149 * @net: the applicable net namespace
5151 * Returns a suitable unique value for a new device interface
5152 * number. The caller must hold the rtnl semaphore or the
5153 * dev_base_lock to be sure it remains unique.
5155 static int dev_new_index(struct net
*net
)
5161 if (!__dev_get_by_index(net
, ifindex
))
5166 /* Delayed registration/unregisteration */
5167 static LIST_HEAD(net_todo_list
);
5169 static void net_set_todo(struct net_device
*dev
)
5171 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5174 static void rollback_registered_many(struct list_head
*head
)
5176 struct net_device
*dev
, *tmp
;
5178 BUG_ON(dev_boot_phase
);
5181 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5182 /* Some devices call without registering
5183 * for initialization unwind. Remove those
5184 * devices and proceed with the remaining.
5186 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5187 pr_debug("unregister_netdevice: device %s/%p never "
5188 "was registered\n", dev
->name
, dev
);
5191 list_del(&dev
->unreg_list
);
5194 dev
->dismantle
= true;
5195 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5198 /* If device is running, close it first. */
5199 dev_close_many(head
);
5201 list_for_each_entry(dev
, head
, unreg_list
) {
5202 /* And unlink it from device chain. */
5203 unlist_netdevice(dev
);
5205 dev
->reg_state
= NETREG_UNREGISTERING
;
5210 list_for_each_entry(dev
, head
, unreg_list
) {
5211 /* Shutdown queueing discipline. */
5215 /* Notify protocols, that we are about to destroy
5216 this device. They should clean all the things.
5218 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5220 if (!dev
->rtnl_link_ops
||
5221 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5222 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5225 * Flush the unicast and multicast chains
5230 if (dev
->netdev_ops
->ndo_uninit
)
5231 dev
->netdev_ops
->ndo_uninit(dev
);
5233 /* Notifier chain MUST detach us from master device. */
5234 WARN_ON(dev
->master
);
5236 /* Remove entries from kobject tree */
5237 netdev_unregister_kobject(dev
);
5240 /* Process any work delayed until the end of the batch */
5241 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5242 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5246 list_for_each_entry(dev
, head
, unreg_list
)
5250 static void rollback_registered(struct net_device
*dev
)
5254 list_add(&dev
->unreg_list
, &single
);
5255 rollback_registered_many(&single
);
5259 static u32
netdev_fix_features(struct net_device
*dev
, u32 features
)
5261 /* Fix illegal checksum combinations */
5262 if ((features
& NETIF_F_HW_CSUM
) &&
5263 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5264 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5265 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5268 if ((features
& NETIF_F_NO_CSUM
) &&
5269 (features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5270 netdev_warn(dev
, "mixed no checksumming and other settings.\n");
5271 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5274 /* Fix illegal SG+CSUM combinations. */
5275 if ((features
& NETIF_F_SG
) &&
5276 !(features
& NETIF_F_ALL_CSUM
)) {
5278 "Dropping NETIF_F_SG since no checksum feature.\n");
5279 features
&= ~NETIF_F_SG
;
5282 /* TSO requires that SG is present as well. */
5283 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5284 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5285 features
&= ~NETIF_F_ALL_TSO
;
5288 /* TSO ECN requires that TSO is present as well. */
5289 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5290 features
&= ~NETIF_F_TSO_ECN
;
5292 /* Software GSO depends on SG. */
5293 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5294 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5295 features
&= ~NETIF_F_GSO
;
5298 /* UFO needs SG and checksumming */
5299 if (features
& NETIF_F_UFO
) {
5300 /* maybe split UFO into V4 and V6? */
5301 if (!((features
& NETIF_F_GEN_CSUM
) ||
5302 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5303 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5305 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5306 features
&= ~NETIF_F_UFO
;
5309 if (!(features
& NETIF_F_SG
)) {
5311 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5312 features
&= ~NETIF_F_UFO
;
5319 int __netdev_update_features(struct net_device
*dev
)
5326 features
= netdev_get_wanted_features(dev
);
5328 if (dev
->netdev_ops
->ndo_fix_features
)
5329 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5331 /* driver might be less strict about feature dependencies */
5332 features
= netdev_fix_features(dev
, features
);
5334 if (dev
->features
== features
)
5337 netdev_dbg(dev
, "Features changed: 0x%08x -> 0x%08x\n",
5338 dev
->features
, features
);
5340 if (dev
->netdev_ops
->ndo_set_features
)
5341 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5343 if (unlikely(err
< 0)) {
5345 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5346 err
, features
, dev
->features
);
5351 dev
->features
= features
;
5357 * netdev_update_features - recalculate device features
5358 * @dev: the device to check
5360 * Recalculate dev->features set and send notifications if it
5361 * has changed. Should be called after driver or hardware dependent
5362 * conditions might have changed that influence the features.
5364 void netdev_update_features(struct net_device
*dev
)
5366 if (__netdev_update_features(dev
))
5367 netdev_features_change(dev
);
5369 EXPORT_SYMBOL(netdev_update_features
);
5372 * netdev_change_features - recalculate device features
5373 * @dev: the device to check
5375 * Recalculate dev->features set and send notifications even
5376 * if they have not changed. Should be called instead of
5377 * netdev_update_features() if also dev->vlan_features might
5378 * have changed to allow the changes to be propagated to stacked
5381 void netdev_change_features(struct net_device
*dev
)
5383 __netdev_update_features(dev
);
5384 netdev_features_change(dev
);
5386 EXPORT_SYMBOL(netdev_change_features
);
5389 * netif_stacked_transfer_operstate - transfer operstate
5390 * @rootdev: the root or lower level device to transfer state from
5391 * @dev: the device to transfer operstate to
5393 * Transfer operational state from root to device. This is normally
5394 * called when a stacking relationship exists between the root
5395 * device and the device(a leaf device).
5397 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5398 struct net_device
*dev
)
5400 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5401 netif_dormant_on(dev
);
5403 netif_dormant_off(dev
);
5405 if (netif_carrier_ok(rootdev
)) {
5406 if (!netif_carrier_ok(dev
))
5407 netif_carrier_on(dev
);
5409 if (netif_carrier_ok(dev
))
5410 netif_carrier_off(dev
);
5413 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5416 static int netif_alloc_rx_queues(struct net_device
*dev
)
5418 unsigned int i
, count
= dev
->num_rx_queues
;
5419 struct netdev_rx_queue
*rx
;
5423 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5425 pr_err("netdev: Unable to allocate %u rx queues.\n", count
);
5430 for (i
= 0; i
< count
; i
++)
5436 static void netdev_init_one_queue(struct net_device
*dev
,
5437 struct netdev_queue
*queue
, void *_unused
)
5439 /* Initialize queue lock */
5440 spin_lock_init(&queue
->_xmit_lock
);
5441 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5442 queue
->xmit_lock_owner
= -1;
5443 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5447 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5449 unsigned int count
= dev
->num_tx_queues
;
5450 struct netdev_queue
*tx
;
5454 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5456 pr_err("netdev: Unable to allocate %u tx queues.\n",
5462 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5463 spin_lock_init(&dev
->tx_global_lock
);
5469 * register_netdevice - register a network device
5470 * @dev: device to register
5472 * Take a completed network device structure and add it to the kernel
5473 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5474 * chain. 0 is returned on success. A negative errno code is returned
5475 * on a failure to set up the device, or if the name is a duplicate.
5477 * Callers must hold the rtnl semaphore. You may want
5478 * register_netdev() instead of this.
5481 * The locking appears insufficient to guarantee two parallel registers
5482 * will not get the same name.
5485 int register_netdevice(struct net_device
*dev
)
5488 struct net
*net
= dev_net(dev
);
5490 BUG_ON(dev_boot_phase
);
5495 /* When net_device's are persistent, this will be fatal. */
5496 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5499 spin_lock_init(&dev
->addr_list_lock
);
5500 netdev_set_addr_lockdep_class(dev
);
5504 ret
= dev_get_valid_name(dev
, dev
->name
);
5508 /* Init, if this function is available */
5509 if (dev
->netdev_ops
->ndo_init
) {
5510 ret
= dev
->netdev_ops
->ndo_init(dev
);
5518 dev
->ifindex
= dev_new_index(net
);
5519 if (dev
->iflink
== -1)
5520 dev
->iflink
= dev
->ifindex
;
5522 /* Transfer changeable features to wanted_features and enable
5523 * software offloads (GSO and GRO).
5525 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5526 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5527 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5529 /* Turn on no cache copy if HW is doing checksum */
5530 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5531 if ((dev
->features
& NETIF_F_ALL_CSUM
) &&
5532 !(dev
->features
& NETIF_F_NO_CSUM
)) {
5533 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5534 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5537 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5539 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5541 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5542 ret
= notifier_to_errno(ret
);
5546 ret
= netdev_register_kobject(dev
);
5549 dev
->reg_state
= NETREG_REGISTERED
;
5551 __netdev_update_features(dev
);
5554 * Default initial state at registry is that the
5555 * device is present.
5558 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5560 dev_init_scheduler(dev
);
5562 list_netdevice(dev
);
5564 /* Notify protocols, that a new device appeared. */
5565 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5566 ret
= notifier_to_errno(ret
);
5568 rollback_registered(dev
);
5569 dev
->reg_state
= NETREG_UNREGISTERED
;
5572 * Prevent userspace races by waiting until the network
5573 * device is fully setup before sending notifications.
5575 if (!dev
->rtnl_link_ops
||
5576 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5577 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5583 if (dev
->netdev_ops
->ndo_uninit
)
5584 dev
->netdev_ops
->ndo_uninit(dev
);
5587 EXPORT_SYMBOL(register_netdevice
);
5590 * init_dummy_netdev - init a dummy network device for NAPI
5591 * @dev: device to init
5593 * This takes a network device structure and initialize the minimum
5594 * amount of fields so it can be used to schedule NAPI polls without
5595 * registering a full blown interface. This is to be used by drivers
5596 * that need to tie several hardware interfaces to a single NAPI
5597 * poll scheduler due to HW limitations.
5599 int init_dummy_netdev(struct net_device
*dev
)
5601 /* Clear everything. Note we don't initialize spinlocks
5602 * are they aren't supposed to be taken by any of the
5603 * NAPI code and this dummy netdev is supposed to be
5604 * only ever used for NAPI polls
5606 memset(dev
, 0, sizeof(struct net_device
));
5608 /* make sure we BUG if trying to hit standard
5609 * register/unregister code path
5611 dev
->reg_state
= NETREG_DUMMY
;
5613 /* NAPI wants this */
5614 INIT_LIST_HEAD(&dev
->napi_list
);
5616 /* a dummy interface is started by default */
5617 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5618 set_bit(__LINK_STATE_START
, &dev
->state
);
5620 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5621 * because users of this 'device' dont need to change
5627 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5631 * register_netdev - register a network device
5632 * @dev: device to register
5634 * Take a completed network device structure and add it to the kernel
5635 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5636 * chain. 0 is returned on success. A negative errno code is returned
5637 * on a failure to set up the device, or if the name is a duplicate.
5639 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5640 * and expands the device name if you passed a format string to
5643 int register_netdev(struct net_device
*dev
)
5648 err
= register_netdevice(dev
);
5652 EXPORT_SYMBOL(register_netdev
);
5654 int netdev_refcnt_read(const struct net_device
*dev
)
5658 for_each_possible_cpu(i
)
5659 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5662 EXPORT_SYMBOL(netdev_refcnt_read
);
5665 * netdev_wait_allrefs - wait until all references are gone.
5667 * This is called when unregistering network devices.
5669 * Any protocol or device that holds a reference should register
5670 * for netdevice notification, and cleanup and put back the
5671 * reference if they receive an UNREGISTER event.
5672 * We can get stuck here if buggy protocols don't correctly
5675 static void netdev_wait_allrefs(struct net_device
*dev
)
5677 unsigned long rebroadcast_time
, warning_time
;
5680 linkwatch_forget_dev(dev
);
5682 rebroadcast_time
= warning_time
= jiffies
;
5683 refcnt
= netdev_refcnt_read(dev
);
5685 while (refcnt
!= 0) {
5686 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5689 /* Rebroadcast unregister notification */
5690 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5691 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5692 * should have already handle it the first time */
5694 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5696 /* We must not have linkwatch events
5697 * pending on unregister. If this
5698 * happens, we simply run the queue
5699 * unscheduled, resulting in a noop
5702 linkwatch_run_queue();
5707 rebroadcast_time
= jiffies
;
5712 refcnt
= netdev_refcnt_read(dev
);
5714 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5715 printk(KERN_EMERG
"unregister_netdevice: "
5716 "waiting for %s to become free. Usage "
5719 warning_time
= jiffies
;
5728 * register_netdevice(x1);
5729 * register_netdevice(x2);
5731 * unregister_netdevice(y1);
5732 * unregister_netdevice(y2);
5738 * We are invoked by rtnl_unlock().
5739 * This allows us to deal with problems:
5740 * 1) We can delete sysfs objects which invoke hotplug
5741 * without deadlocking with linkwatch via keventd.
5742 * 2) Since we run with the RTNL semaphore not held, we can sleep
5743 * safely in order to wait for the netdev refcnt to drop to zero.
5745 * We must not return until all unregister events added during
5746 * the interval the lock was held have been completed.
5748 void netdev_run_todo(void)
5750 struct list_head list
;
5752 /* Snapshot list, allow later requests */
5753 list_replace_init(&net_todo_list
, &list
);
5757 while (!list_empty(&list
)) {
5758 struct net_device
*dev
5759 = list_first_entry(&list
, struct net_device
, todo_list
);
5760 list_del(&dev
->todo_list
);
5762 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5763 printk(KERN_ERR
"network todo '%s' but state %d\n",
5764 dev
->name
, dev
->reg_state
);
5769 dev
->reg_state
= NETREG_UNREGISTERED
;
5771 on_each_cpu(flush_backlog
, dev
, 1);
5773 netdev_wait_allrefs(dev
);
5776 BUG_ON(netdev_refcnt_read(dev
));
5777 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5778 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5779 WARN_ON(dev
->dn_ptr
);
5781 if (dev
->destructor
)
5782 dev
->destructor(dev
);
5784 /* Free network device */
5785 kobject_put(&dev
->dev
.kobj
);
5789 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5790 * fields in the same order, with only the type differing.
5792 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5793 const struct net_device_stats
*netdev_stats
)
5795 #if BITS_PER_LONG == 64
5796 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5797 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5799 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5800 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5801 u64
*dst
= (u64
*)stats64
;
5803 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5804 sizeof(*stats64
) / sizeof(u64
));
5805 for (i
= 0; i
< n
; i
++)
5811 * dev_get_stats - get network device statistics
5812 * @dev: device to get statistics from
5813 * @storage: place to store stats
5815 * Get network statistics from device. Return @storage.
5816 * The device driver may provide its own method by setting
5817 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5818 * otherwise the internal statistics structure is used.
5820 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5821 struct rtnl_link_stats64
*storage
)
5823 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5825 if (ops
->ndo_get_stats64
) {
5826 memset(storage
, 0, sizeof(*storage
));
5827 ops
->ndo_get_stats64(dev
, storage
);
5828 } else if (ops
->ndo_get_stats
) {
5829 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5831 netdev_stats_to_stats64(storage
, &dev
->stats
);
5833 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5836 EXPORT_SYMBOL(dev_get_stats
);
5838 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5840 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5842 #ifdef CONFIG_NET_CLS_ACT
5845 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5848 netdev_init_one_queue(dev
, queue
, NULL
);
5849 queue
->qdisc
= &noop_qdisc
;
5850 queue
->qdisc_sleeping
= &noop_qdisc
;
5851 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5857 * alloc_netdev_mqs - allocate network device
5858 * @sizeof_priv: size of private data to allocate space for
5859 * @name: device name format string
5860 * @setup: callback to initialize device
5861 * @txqs: the number of TX subqueues to allocate
5862 * @rxqs: the number of RX subqueues to allocate
5864 * Allocates a struct net_device with private data area for driver use
5865 * and performs basic initialization. Also allocates subquue structs
5866 * for each queue on the device.
5868 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5869 void (*setup
)(struct net_device
*),
5870 unsigned int txqs
, unsigned int rxqs
)
5872 struct net_device
*dev
;
5874 struct net_device
*p
;
5876 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5879 pr_err("alloc_netdev: Unable to allocate device "
5880 "with zero queues.\n");
5886 pr_err("alloc_netdev: Unable to allocate device "
5887 "with zero RX queues.\n");
5892 alloc_size
= sizeof(struct net_device
);
5894 /* ensure 32-byte alignment of private area */
5895 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5896 alloc_size
+= sizeof_priv
;
5898 /* ensure 32-byte alignment of whole construct */
5899 alloc_size
+= NETDEV_ALIGN
- 1;
5901 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5903 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5907 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5908 dev
->padded
= (char *)dev
- (char *)p
;
5910 dev
->pcpu_refcnt
= alloc_percpu(int);
5911 if (!dev
->pcpu_refcnt
)
5914 if (dev_addr_init(dev
))
5920 dev_net_set(dev
, &init_net
);
5922 dev
->gso_max_size
= GSO_MAX_SIZE
;
5924 INIT_LIST_HEAD(&dev
->napi_list
);
5925 INIT_LIST_HEAD(&dev
->unreg_list
);
5926 INIT_LIST_HEAD(&dev
->link_watch_list
);
5927 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5930 dev
->num_tx_queues
= txqs
;
5931 dev
->real_num_tx_queues
= txqs
;
5932 if (netif_alloc_netdev_queues(dev
))
5936 dev
->num_rx_queues
= rxqs
;
5937 dev
->real_num_rx_queues
= rxqs
;
5938 if (netif_alloc_rx_queues(dev
))
5942 strcpy(dev
->name
, name
);
5943 dev
->group
= INIT_NETDEV_GROUP
;
5951 free_percpu(dev
->pcpu_refcnt
);
5961 EXPORT_SYMBOL(alloc_netdev_mqs
);
5964 * free_netdev - free network device
5967 * This function does the last stage of destroying an allocated device
5968 * interface. The reference to the device object is released.
5969 * If this is the last reference then it will be freed.
5971 void free_netdev(struct net_device
*dev
)
5973 struct napi_struct
*p
, *n
;
5975 release_net(dev_net(dev
));
5982 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
5984 /* Flush device addresses */
5985 dev_addr_flush(dev
);
5987 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5990 free_percpu(dev
->pcpu_refcnt
);
5991 dev
->pcpu_refcnt
= NULL
;
5993 /* Compatibility with error handling in drivers */
5994 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5995 kfree((char *)dev
- dev
->padded
);
5999 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6000 dev
->reg_state
= NETREG_RELEASED
;
6002 /* will free via device release */
6003 put_device(&dev
->dev
);
6005 EXPORT_SYMBOL(free_netdev
);
6008 * synchronize_net - Synchronize with packet receive processing
6010 * Wait for packets currently being received to be done.
6011 * Does not block later packets from starting.
6013 void synchronize_net(void)
6016 if (rtnl_is_locked())
6017 synchronize_rcu_expedited();
6021 EXPORT_SYMBOL(synchronize_net
);
6024 * unregister_netdevice_queue - remove device from the kernel
6028 * This function shuts down a device interface and removes it
6029 * from the kernel tables.
6030 * If head not NULL, device is queued to be unregistered later.
6032 * Callers must hold the rtnl semaphore. You may want
6033 * unregister_netdev() instead of this.
6036 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6041 list_move_tail(&dev
->unreg_list
, head
);
6043 rollback_registered(dev
);
6044 /* Finish processing unregister after unlock */
6048 EXPORT_SYMBOL(unregister_netdevice_queue
);
6051 * unregister_netdevice_many - unregister many devices
6052 * @head: list of devices
6054 void unregister_netdevice_many(struct list_head
*head
)
6056 struct net_device
*dev
;
6058 if (!list_empty(head
)) {
6059 rollback_registered_many(head
);
6060 list_for_each_entry(dev
, head
, unreg_list
)
6064 EXPORT_SYMBOL(unregister_netdevice_many
);
6067 * unregister_netdev - remove device from the kernel
6070 * This function shuts down a device interface and removes it
6071 * from the kernel tables.
6073 * This is just a wrapper for unregister_netdevice that takes
6074 * the rtnl semaphore. In general you want to use this and not
6075 * unregister_netdevice.
6077 void unregister_netdev(struct net_device
*dev
)
6080 unregister_netdevice(dev
);
6083 EXPORT_SYMBOL(unregister_netdev
);
6086 * dev_change_net_namespace - move device to different nethost namespace
6088 * @net: network namespace
6089 * @pat: If not NULL name pattern to try if the current device name
6090 * is already taken in the destination network namespace.
6092 * This function shuts down a device interface and moves it
6093 * to a new network namespace. On success 0 is returned, on
6094 * a failure a netagive errno code is returned.
6096 * Callers must hold the rtnl semaphore.
6099 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6105 /* Don't allow namespace local devices to be moved. */
6107 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6110 /* Ensure the device has been registrered */
6112 if (dev
->reg_state
!= NETREG_REGISTERED
)
6115 /* Get out if there is nothing todo */
6117 if (net_eq(dev_net(dev
), net
))
6120 /* Pick the destination device name, and ensure
6121 * we can use it in the destination network namespace.
6124 if (__dev_get_by_name(net
, dev
->name
)) {
6125 /* We get here if we can't use the current device name */
6128 if (dev_get_valid_name(dev
, pat
) < 0)
6133 * And now a mini version of register_netdevice unregister_netdevice.
6136 /* If device is running close it first. */
6139 /* And unlink it from device chain */
6141 unlist_netdevice(dev
);
6145 /* Shutdown queueing discipline. */
6148 /* Notify protocols, that we are about to destroy
6149 this device. They should clean all the things.
6151 Note that dev->reg_state stays at NETREG_REGISTERED.
6152 This is wanted because this way 8021q and macvlan know
6153 the device is just moving and can keep their slaves up.
6155 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6156 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6159 * Flush the unicast and multicast chains
6164 /* Actually switch the network namespace */
6165 dev_net_set(dev
, net
);
6167 /* If there is an ifindex conflict assign a new one */
6168 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6169 int iflink
= (dev
->iflink
== dev
->ifindex
);
6170 dev
->ifindex
= dev_new_index(net
);
6172 dev
->iflink
= dev
->ifindex
;
6175 /* Fixup kobjects */
6176 err
= device_rename(&dev
->dev
, dev
->name
);
6179 /* Add the device back in the hashes */
6180 list_netdevice(dev
);
6182 /* Notify protocols, that a new device appeared. */
6183 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6186 * Prevent userspace races by waiting until the network
6187 * device is fully setup before sending notifications.
6189 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6196 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6198 static int dev_cpu_callback(struct notifier_block
*nfb
,
6199 unsigned long action
,
6202 struct sk_buff
**list_skb
;
6203 struct sk_buff
*skb
;
6204 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6205 struct softnet_data
*sd
, *oldsd
;
6207 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6210 local_irq_disable();
6211 cpu
= smp_processor_id();
6212 sd
= &per_cpu(softnet_data
, cpu
);
6213 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6215 /* Find end of our completion_queue. */
6216 list_skb
= &sd
->completion_queue
;
6218 list_skb
= &(*list_skb
)->next
;
6219 /* Append completion queue from offline CPU. */
6220 *list_skb
= oldsd
->completion_queue
;
6221 oldsd
->completion_queue
= NULL
;
6223 /* Append output queue from offline CPU. */
6224 if (oldsd
->output_queue
) {
6225 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6226 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6227 oldsd
->output_queue
= NULL
;
6228 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6230 /* Append NAPI poll list from offline CPU. */
6231 if (!list_empty(&oldsd
->poll_list
)) {
6232 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6233 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6236 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6239 /* Process offline CPU's input_pkt_queue */
6240 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6242 input_queue_head_incr(oldsd
);
6244 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6246 input_queue_head_incr(oldsd
);
6254 * netdev_increment_features - increment feature set by one
6255 * @all: current feature set
6256 * @one: new feature set
6257 * @mask: mask feature set
6259 * Computes a new feature set after adding a device with feature set
6260 * @one to the master device with current feature set @all. Will not
6261 * enable anything that is off in @mask. Returns the new feature set.
6263 u32
netdev_increment_features(u32 all
, u32 one
, u32 mask
)
6265 if (mask
& NETIF_F_GEN_CSUM
)
6266 mask
|= NETIF_F_ALL_CSUM
;
6267 mask
|= NETIF_F_VLAN_CHALLENGED
;
6269 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6270 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6272 /* If device needs checksumming, downgrade to it. */
6273 if (all
& (NETIF_F_ALL_CSUM
& ~NETIF_F_NO_CSUM
))
6274 all
&= ~NETIF_F_NO_CSUM
;
6276 /* If one device supports hw checksumming, set for all. */
6277 if (all
& NETIF_F_GEN_CSUM
)
6278 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6282 EXPORT_SYMBOL(netdev_increment_features
);
6284 static struct hlist_head
*netdev_create_hash(void)
6287 struct hlist_head
*hash
;
6289 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6291 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6292 INIT_HLIST_HEAD(&hash
[i
]);
6297 /* Initialize per network namespace state */
6298 static int __net_init
netdev_init(struct net
*net
)
6300 INIT_LIST_HEAD(&net
->dev_base_head
);
6302 net
->dev_name_head
= netdev_create_hash();
6303 if (net
->dev_name_head
== NULL
)
6306 net
->dev_index_head
= netdev_create_hash();
6307 if (net
->dev_index_head
== NULL
)
6313 kfree(net
->dev_name_head
);
6319 * netdev_drivername - network driver for the device
6320 * @dev: network device
6322 * Determine network driver for device.
6324 const char *netdev_drivername(const struct net_device
*dev
)
6326 const struct device_driver
*driver
;
6327 const struct device
*parent
;
6328 const char *empty
= "";
6330 parent
= dev
->dev
.parent
;
6334 driver
= parent
->driver
;
6335 if (driver
&& driver
->name
)
6336 return driver
->name
;
6340 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6341 struct va_format
*vaf
)
6345 if (dev
&& dev
->dev
.parent
)
6346 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6347 netdev_name(dev
), vaf
);
6349 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6351 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6356 int netdev_printk(const char *level
, const struct net_device
*dev
,
6357 const char *format
, ...)
6359 struct va_format vaf
;
6363 va_start(args
, format
);
6368 r
= __netdev_printk(level
, dev
, &vaf
);
6373 EXPORT_SYMBOL(netdev_printk
);
6375 #define define_netdev_printk_level(func, level) \
6376 int func(const struct net_device *dev, const char *fmt, ...) \
6379 struct va_format vaf; \
6382 va_start(args, fmt); \
6387 r = __netdev_printk(level, dev, &vaf); \
6392 EXPORT_SYMBOL(func);
6394 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6395 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6396 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6397 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6398 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6399 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6400 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6402 static void __net_exit
netdev_exit(struct net
*net
)
6404 kfree(net
->dev_name_head
);
6405 kfree(net
->dev_index_head
);
6408 static struct pernet_operations __net_initdata netdev_net_ops
= {
6409 .init
= netdev_init
,
6410 .exit
= netdev_exit
,
6413 static void __net_exit
default_device_exit(struct net
*net
)
6415 struct net_device
*dev
, *aux
;
6417 * Push all migratable network devices back to the
6418 * initial network namespace
6421 for_each_netdev_safe(net
, dev
, aux
) {
6423 char fb_name
[IFNAMSIZ
];
6425 /* Ignore unmoveable devices (i.e. loopback) */
6426 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6429 /* Leave virtual devices for the generic cleanup */
6430 if (dev
->rtnl_link_ops
)
6433 /* Push remaining network devices to init_net */
6434 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6435 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6437 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
6438 __func__
, dev
->name
, err
);
6445 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6447 /* At exit all network devices most be removed from a network
6448 * namespace. Do this in the reverse order of registration.
6449 * Do this across as many network namespaces as possible to
6450 * improve batching efficiency.
6452 struct net_device
*dev
;
6454 LIST_HEAD(dev_kill_list
);
6457 list_for_each_entry(net
, net_list
, exit_list
) {
6458 for_each_netdev_reverse(net
, dev
) {
6459 if (dev
->rtnl_link_ops
)
6460 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6462 unregister_netdevice_queue(dev
, &dev_kill_list
);
6465 unregister_netdevice_many(&dev_kill_list
);
6466 list_del(&dev_kill_list
);
6470 static struct pernet_operations __net_initdata default_device_ops
= {
6471 .exit
= default_device_exit
,
6472 .exit_batch
= default_device_exit_batch
,
6476 * Initialize the DEV module. At boot time this walks the device list and
6477 * unhooks any devices that fail to initialise (normally hardware not
6478 * present) and leaves us with a valid list of present and active devices.
6483 * This is called single threaded during boot, so no need
6484 * to take the rtnl semaphore.
6486 static int __init
net_dev_init(void)
6488 int i
, rc
= -ENOMEM
;
6490 BUG_ON(!dev_boot_phase
);
6492 if (dev_proc_init())
6495 if (netdev_kobject_init())
6498 INIT_LIST_HEAD(&ptype_all
);
6499 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6500 INIT_LIST_HEAD(&ptype_base
[i
]);
6502 if (register_pernet_subsys(&netdev_net_ops
))
6506 * Initialise the packet receive queues.
6509 for_each_possible_cpu(i
) {
6510 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6512 memset(sd
, 0, sizeof(*sd
));
6513 skb_queue_head_init(&sd
->input_pkt_queue
);
6514 skb_queue_head_init(&sd
->process_queue
);
6515 sd
->completion_queue
= NULL
;
6516 INIT_LIST_HEAD(&sd
->poll_list
);
6517 sd
->output_queue
= NULL
;
6518 sd
->output_queue_tailp
= &sd
->output_queue
;
6520 sd
->csd
.func
= rps_trigger_softirq
;
6526 sd
->backlog
.poll
= process_backlog
;
6527 sd
->backlog
.weight
= weight_p
;
6528 sd
->backlog
.gro_list
= NULL
;
6529 sd
->backlog
.gro_count
= 0;
6534 /* The loopback device is special if any other network devices
6535 * is present in a network namespace the loopback device must
6536 * be present. Since we now dynamically allocate and free the
6537 * loopback device ensure this invariant is maintained by
6538 * keeping the loopback device as the first device on the
6539 * list of network devices. Ensuring the loopback devices
6540 * is the first device that appears and the last network device
6543 if (register_pernet_device(&loopback_net_ops
))
6546 if (register_pernet_device(&default_device_ops
))
6549 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6550 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6552 hotcpu_notifier(dev_cpu_callback
, 0);
6560 subsys_initcall(net_dev_init
);
6562 static int __init
initialize_hashrnd(void)
6564 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6568 late_initcall_sync(initialize_hashrnd
);