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
)
2618 poff
= proto_ports_offset(ip_proto
);
2621 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2622 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2623 if (ports
.v16
[1] < ports
.v16
[0])
2624 swap(ports
.v16
[0], ports
.v16
[1]);
2629 /* get a consistent hash (same value on both flow directions) */
2633 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2640 EXPORT_SYMBOL(__skb_get_rxhash
);
2644 /* One global table that all flow-based protocols share. */
2645 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2646 EXPORT_SYMBOL(rps_sock_flow_table
);
2648 static struct rps_dev_flow
*
2649 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2650 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2654 tcpu
= rflow
->cpu
= next_cpu
;
2655 if (tcpu
!= RPS_NO_CPU
) {
2656 #ifdef CONFIG_RFS_ACCEL
2657 struct netdev_rx_queue
*rxqueue
;
2658 struct rps_dev_flow_table
*flow_table
;
2659 struct rps_dev_flow
*old_rflow
;
2664 /* Should we steer this flow to a different hardware queue? */
2665 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2666 !(dev
->features
& NETIF_F_NTUPLE
))
2668 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2669 if (rxq_index
== skb_get_rx_queue(skb
))
2672 rxqueue
= dev
->_rx
+ rxq_index
;
2673 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2676 flow_id
= skb
->rxhash
& flow_table
->mask
;
2677 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2678 rxq_index
, flow_id
);
2682 rflow
= &flow_table
->flows
[flow_id
];
2683 rflow
->cpu
= next_cpu
;
2685 if (old_rflow
->filter
== rflow
->filter
)
2686 old_rflow
->filter
= RPS_NO_FILTER
;
2690 per_cpu(softnet_data
, tcpu
).input_queue_head
;
2697 * get_rps_cpu is called from netif_receive_skb and returns the target
2698 * CPU from the RPS map of the receiving queue for a given skb.
2699 * rcu_read_lock must be held on entry.
2701 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2702 struct rps_dev_flow
**rflowp
)
2704 struct netdev_rx_queue
*rxqueue
;
2705 struct rps_map
*map
;
2706 struct rps_dev_flow_table
*flow_table
;
2707 struct rps_sock_flow_table
*sock_flow_table
;
2711 if (skb_rx_queue_recorded(skb
)) {
2712 u16 index
= skb_get_rx_queue(skb
);
2713 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2714 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2715 "%s received packet on queue %u, but number "
2716 "of RX queues is %u\n",
2717 dev
->name
, index
, dev
->real_num_rx_queues
);
2720 rxqueue
= dev
->_rx
+ index
;
2724 map
= rcu_dereference(rxqueue
->rps_map
);
2726 if (map
->len
== 1 &&
2727 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2728 tcpu
= map
->cpus
[0];
2729 if (cpu_online(tcpu
))
2733 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2737 skb_reset_network_header(skb
);
2738 if (!skb_get_rxhash(skb
))
2741 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2742 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2743 if (flow_table
&& sock_flow_table
) {
2745 struct rps_dev_flow
*rflow
;
2747 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2750 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2751 sock_flow_table
->mask
];
2754 * If the desired CPU (where last recvmsg was done) is
2755 * different from current CPU (one in the rx-queue flow
2756 * table entry), switch if one of the following holds:
2757 * - Current CPU is unset (equal to RPS_NO_CPU).
2758 * - Current CPU is offline.
2759 * - The current CPU's queue tail has advanced beyond the
2760 * last packet that was enqueued using this table entry.
2761 * This guarantees that all previous packets for the flow
2762 * have been dequeued, thus preserving in order delivery.
2764 if (unlikely(tcpu
!= next_cpu
) &&
2765 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2766 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2767 rflow
->last_qtail
)) >= 0))
2768 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2770 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2778 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2780 if (cpu_online(tcpu
)) {
2790 #ifdef CONFIG_RFS_ACCEL
2793 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2794 * @dev: Device on which the filter was set
2795 * @rxq_index: RX queue index
2796 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2797 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2799 * Drivers that implement ndo_rx_flow_steer() should periodically call
2800 * this function for each installed filter and remove the filters for
2801 * which it returns %true.
2803 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2804 u32 flow_id
, u16 filter_id
)
2806 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2807 struct rps_dev_flow_table
*flow_table
;
2808 struct rps_dev_flow
*rflow
;
2813 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2814 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2815 rflow
= &flow_table
->flows
[flow_id
];
2816 cpu
= ACCESS_ONCE(rflow
->cpu
);
2817 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2818 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2819 rflow
->last_qtail
) <
2820 (int)(10 * flow_table
->mask
)))
2826 EXPORT_SYMBOL(rps_may_expire_flow
);
2828 #endif /* CONFIG_RFS_ACCEL */
2830 /* Called from hardirq (IPI) context */
2831 static void rps_trigger_softirq(void *data
)
2833 struct softnet_data
*sd
= data
;
2835 ____napi_schedule(sd
, &sd
->backlog
);
2839 #endif /* CONFIG_RPS */
2842 * Check if this softnet_data structure is another cpu one
2843 * If yes, queue it to our IPI list and return 1
2846 static int rps_ipi_queued(struct softnet_data
*sd
)
2849 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2852 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2853 mysd
->rps_ipi_list
= sd
;
2855 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2858 #endif /* CONFIG_RPS */
2863 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2864 * queue (may be a remote CPU queue).
2866 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2867 unsigned int *qtail
)
2869 struct softnet_data
*sd
;
2870 unsigned long flags
;
2872 sd
= &per_cpu(softnet_data
, cpu
);
2874 local_irq_save(flags
);
2877 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2878 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2880 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2881 input_queue_tail_incr_save(sd
, qtail
);
2883 local_irq_restore(flags
);
2884 return NET_RX_SUCCESS
;
2887 /* Schedule NAPI for backlog device
2888 * We can use non atomic operation since we own the queue lock
2890 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2891 if (!rps_ipi_queued(sd
))
2892 ____napi_schedule(sd
, &sd
->backlog
);
2900 local_irq_restore(flags
);
2902 atomic_long_inc(&skb
->dev
->rx_dropped
);
2908 * netif_rx - post buffer to the network code
2909 * @skb: buffer to post
2911 * This function receives a packet from a device driver and queues it for
2912 * the upper (protocol) levels to process. It always succeeds. The buffer
2913 * may be dropped during processing for congestion control or by the
2917 * NET_RX_SUCCESS (no congestion)
2918 * NET_RX_DROP (packet was dropped)
2922 int netif_rx(struct sk_buff
*skb
)
2926 /* if netpoll wants it, pretend we never saw it */
2927 if (netpoll_rx(skb
))
2930 if (netdev_tstamp_prequeue
)
2931 net_timestamp_check(skb
);
2933 trace_netif_rx(skb
);
2936 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2942 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2944 cpu
= smp_processor_id();
2946 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2954 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2960 EXPORT_SYMBOL(netif_rx
);
2962 int netif_rx_ni(struct sk_buff
*skb
)
2967 err
= netif_rx(skb
);
2968 if (local_softirq_pending())
2974 EXPORT_SYMBOL(netif_rx_ni
);
2976 static void net_tx_action(struct softirq_action
*h
)
2978 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2980 if (sd
->completion_queue
) {
2981 struct sk_buff
*clist
;
2983 local_irq_disable();
2984 clist
= sd
->completion_queue
;
2985 sd
->completion_queue
= NULL
;
2989 struct sk_buff
*skb
= clist
;
2990 clist
= clist
->next
;
2992 WARN_ON(atomic_read(&skb
->users
));
2993 trace_kfree_skb(skb
, net_tx_action
);
2998 if (sd
->output_queue
) {
3001 local_irq_disable();
3002 head
= sd
->output_queue
;
3003 sd
->output_queue
= NULL
;
3004 sd
->output_queue_tailp
= &sd
->output_queue
;
3008 struct Qdisc
*q
= head
;
3009 spinlock_t
*root_lock
;
3011 head
= head
->next_sched
;
3013 root_lock
= qdisc_lock(q
);
3014 if (spin_trylock(root_lock
)) {
3015 smp_mb__before_clear_bit();
3016 clear_bit(__QDISC_STATE_SCHED
,
3019 spin_unlock(root_lock
);
3021 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
3023 __netif_reschedule(q
);
3025 smp_mb__before_clear_bit();
3026 clear_bit(__QDISC_STATE_SCHED
,
3034 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3035 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3036 /* This hook is defined here for ATM LANE */
3037 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3038 unsigned char *addr
) __read_mostly
;
3039 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3042 #ifdef CONFIG_NET_CLS_ACT
3043 /* TODO: Maybe we should just force sch_ingress to be compiled in
3044 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3045 * a compare and 2 stores extra right now if we dont have it on
3046 * but have CONFIG_NET_CLS_ACT
3047 * NOTE: This doesn't stop any functionality; if you dont have
3048 * the ingress scheduler, you just can't add policies on ingress.
3051 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3053 struct net_device
*dev
= skb
->dev
;
3054 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3055 int result
= TC_ACT_OK
;
3058 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3059 if (net_ratelimit())
3060 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3061 skb
->skb_iif
, dev
->ifindex
);
3065 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3066 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3069 if (q
!= &noop_qdisc
) {
3070 spin_lock(qdisc_lock(q
));
3071 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3072 result
= qdisc_enqueue_root(skb
, q
);
3073 spin_unlock(qdisc_lock(q
));
3079 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3080 struct packet_type
**pt_prev
,
3081 int *ret
, struct net_device
*orig_dev
)
3083 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3085 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3089 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3093 switch (ing_filter(skb
, rxq
)) {
3107 * netdev_rx_handler_register - register receive handler
3108 * @dev: device to register a handler for
3109 * @rx_handler: receive handler to register
3110 * @rx_handler_data: data pointer that is used by rx handler
3112 * Register a receive hander for a device. This handler will then be
3113 * called from __netif_receive_skb. A negative errno code is returned
3116 * The caller must hold the rtnl_mutex.
3118 * For a general description of rx_handler, see enum rx_handler_result.
3120 int netdev_rx_handler_register(struct net_device
*dev
,
3121 rx_handler_func_t
*rx_handler
,
3122 void *rx_handler_data
)
3126 if (dev
->rx_handler
)
3129 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3130 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3134 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3137 * netdev_rx_handler_unregister - unregister receive handler
3138 * @dev: device to unregister a handler from
3140 * Unregister a receive hander from a device.
3142 * The caller must hold the rtnl_mutex.
3144 void netdev_rx_handler_unregister(struct net_device
*dev
)
3148 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3149 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3151 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3153 static int __netif_receive_skb(struct sk_buff
*skb
)
3155 struct packet_type
*ptype
, *pt_prev
;
3156 rx_handler_func_t
*rx_handler
;
3157 struct net_device
*orig_dev
;
3158 struct net_device
*null_or_dev
;
3159 bool deliver_exact
= false;
3160 int ret
= NET_RX_DROP
;
3163 if (!netdev_tstamp_prequeue
)
3164 net_timestamp_check(skb
);
3166 trace_netif_receive_skb(skb
);
3168 /* if we've gotten here through NAPI, check netpoll */
3169 if (netpoll_receive_skb(skb
))
3173 skb
->skb_iif
= skb
->dev
->ifindex
;
3174 orig_dev
= skb
->dev
;
3176 skb_reset_network_header(skb
);
3177 skb_reset_transport_header(skb
);
3178 skb_reset_mac_len(skb
);
3186 __this_cpu_inc(softnet_data
.processed
);
3188 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3189 skb
= vlan_untag(skb
);
3194 #ifdef CONFIG_NET_CLS_ACT
3195 if (skb
->tc_verd
& TC_NCLS
) {
3196 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3201 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3202 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3204 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3209 #ifdef CONFIG_NET_CLS_ACT
3210 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3216 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3219 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3222 switch (rx_handler(&skb
)) {
3223 case RX_HANDLER_CONSUMED
:
3225 case RX_HANDLER_ANOTHER
:
3227 case RX_HANDLER_EXACT
:
3228 deliver_exact
= true;
3229 case RX_HANDLER_PASS
:
3236 if (vlan_tx_tag_present(skb
)) {
3238 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3241 if (vlan_do_receive(&skb
))
3243 else if (unlikely(!skb
))
3247 /* deliver only exact match when indicated */
3248 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3250 type
= skb
->protocol
;
3251 list_for_each_entry_rcu(ptype
,
3252 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3253 if (ptype
->type
== type
&&
3254 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3255 ptype
->dev
== orig_dev
)) {
3257 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3263 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3265 atomic_long_inc(&skb
->dev
->rx_dropped
);
3267 /* Jamal, now you will not able to escape explaining
3268 * me how you were going to use this. :-)
3279 * netif_receive_skb - process receive buffer from network
3280 * @skb: buffer to process
3282 * netif_receive_skb() is the main receive data processing function.
3283 * It always succeeds. The buffer may be dropped during processing
3284 * for congestion control or by the protocol layers.
3286 * This function may only be called from softirq context and interrupts
3287 * should be enabled.
3289 * Return values (usually ignored):
3290 * NET_RX_SUCCESS: no congestion
3291 * NET_RX_DROP: packet was dropped
3293 int netif_receive_skb(struct sk_buff
*skb
)
3295 if (netdev_tstamp_prequeue
)
3296 net_timestamp_check(skb
);
3298 if (skb_defer_rx_timestamp(skb
))
3299 return NET_RX_SUCCESS
;
3303 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3308 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3311 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3315 ret
= __netif_receive_skb(skb
);
3321 return __netif_receive_skb(skb
);
3324 EXPORT_SYMBOL(netif_receive_skb
);
3326 /* Network device is going away, flush any packets still pending
3327 * Called with irqs disabled.
3329 static void flush_backlog(void *arg
)
3331 struct net_device
*dev
= arg
;
3332 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3333 struct sk_buff
*skb
, *tmp
;
3336 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3337 if (skb
->dev
== dev
) {
3338 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3340 input_queue_head_incr(sd
);
3345 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3346 if (skb
->dev
== dev
) {
3347 __skb_unlink(skb
, &sd
->process_queue
);
3349 input_queue_head_incr(sd
);
3354 static int napi_gro_complete(struct sk_buff
*skb
)
3356 struct packet_type
*ptype
;
3357 __be16 type
= skb
->protocol
;
3358 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3361 if (NAPI_GRO_CB(skb
)->count
== 1) {
3362 skb_shinfo(skb
)->gso_size
= 0;
3367 list_for_each_entry_rcu(ptype
, head
, list
) {
3368 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3371 err
= ptype
->gro_complete(skb
);
3377 WARN_ON(&ptype
->list
== head
);
3379 return NET_RX_SUCCESS
;
3383 return netif_receive_skb(skb
);
3386 inline void napi_gro_flush(struct napi_struct
*napi
)
3388 struct sk_buff
*skb
, *next
;
3390 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3393 napi_gro_complete(skb
);
3396 napi
->gro_count
= 0;
3397 napi
->gro_list
= NULL
;
3399 EXPORT_SYMBOL(napi_gro_flush
);
3401 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3403 struct sk_buff
**pp
= NULL
;
3404 struct packet_type
*ptype
;
3405 __be16 type
= skb
->protocol
;
3406 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3409 enum gro_result ret
;
3411 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3414 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3418 list_for_each_entry_rcu(ptype
, head
, list
) {
3419 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3422 skb_set_network_header(skb
, skb_gro_offset(skb
));
3423 mac_len
= skb
->network_header
- skb
->mac_header
;
3424 skb
->mac_len
= mac_len
;
3425 NAPI_GRO_CB(skb
)->same_flow
= 0;
3426 NAPI_GRO_CB(skb
)->flush
= 0;
3427 NAPI_GRO_CB(skb
)->free
= 0;
3429 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3434 if (&ptype
->list
== head
)
3437 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3438 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3441 struct sk_buff
*nskb
= *pp
;
3445 napi_gro_complete(nskb
);
3452 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3456 NAPI_GRO_CB(skb
)->count
= 1;
3457 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3458 skb
->next
= napi
->gro_list
;
3459 napi
->gro_list
= skb
;
3463 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3464 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3466 BUG_ON(skb
->end
- skb
->tail
< grow
);
3468 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3471 skb
->data_len
-= grow
;
3473 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3474 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3476 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3477 put_page(skb_shinfo(skb
)->frags
[0].page
);
3478 memmove(skb_shinfo(skb
)->frags
,
3479 skb_shinfo(skb
)->frags
+ 1,
3480 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3491 EXPORT_SYMBOL(dev_gro_receive
);
3493 static inline gro_result_t
3494 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3498 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3499 unsigned long diffs
;
3501 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3502 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3503 diffs
|= compare_ether_header(skb_mac_header(p
),
3504 skb_gro_mac_header(skb
));
3505 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3506 NAPI_GRO_CB(p
)->flush
= 0;
3509 return dev_gro_receive(napi
, skb
);
3512 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3516 if (netif_receive_skb(skb
))
3521 case GRO_MERGED_FREE
:
3532 EXPORT_SYMBOL(napi_skb_finish
);
3534 void skb_gro_reset_offset(struct sk_buff
*skb
)
3536 NAPI_GRO_CB(skb
)->data_offset
= 0;
3537 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3538 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3540 if (skb
->mac_header
== skb
->tail
&&
3541 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3542 NAPI_GRO_CB(skb
)->frag0
=
3543 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3544 skb_shinfo(skb
)->frags
[0].page_offset
;
3545 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3548 EXPORT_SYMBOL(skb_gro_reset_offset
);
3550 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3552 skb_gro_reset_offset(skb
);
3554 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3556 EXPORT_SYMBOL(napi_gro_receive
);
3558 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3560 __skb_pull(skb
, skb_headlen(skb
));
3561 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3563 skb
->dev
= napi
->dev
;
3569 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3571 struct sk_buff
*skb
= napi
->skb
;
3574 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3580 EXPORT_SYMBOL(napi_get_frags
);
3582 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3588 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3590 if (ret
== GRO_HELD
)
3591 skb_gro_pull(skb
, -ETH_HLEN
);
3592 else if (netif_receive_skb(skb
))
3597 case GRO_MERGED_FREE
:
3598 napi_reuse_skb(napi
, skb
);
3607 EXPORT_SYMBOL(napi_frags_finish
);
3609 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3611 struct sk_buff
*skb
= napi
->skb
;
3618 skb_reset_mac_header(skb
);
3619 skb_gro_reset_offset(skb
);
3621 off
= skb_gro_offset(skb
);
3622 hlen
= off
+ sizeof(*eth
);
3623 eth
= skb_gro_header_fast(skb
, off
);
3624 if (skb_gro_header_hard(skb
, hlen
)) {
3625 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3626 if (unlikely(!eth
)) {
3627 napi_reuse_skb(napi
, skb
);
3633 skb_gro_pull(skb
, sizeof(*eth
));
3636 * This works because the only protocols we care about don't require
3637 * special handling. We'll fix it up properly at the end.
3639 skb
->protocol
= eth
->h_proto
;
3644 EXPORT_SYMBOL(napi_frags_skb
);
3646 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3648 struct sk_buff
*skb
= napi_frags_skb(napi
);
3653 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3655 EXPORT_SYMBOL(napi_gro_frags
);
3658 * net_rps_action sends any pending IPI's for rps.
3659 * Note: called with local irq disabled, but exits with local irq enabled.
3661 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3664 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3667 sd
->rps_ipi_list
= NULL
;
3671 /* Send pending IPI's to kick RPS processing on remote cpus. */
3673 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3675 if (cpu_online(remsd
->cpu
))
3676 __smp_call_function_single(remsd
->cpu
,
3685 static int process_backlog(struct napi_struct
*napi
, int quota
)
3688 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3691 /* Check if we have pending ipi, its better to send them now,
3692 * not waiting net_rx_action() end.
3694 if (sd
->rps_ipi_list
) {
3695 local_irq_disable();
3696 net_rps_action_and_irq_enable(sd
);
3699 napi
->weight
= weight_p
;
3700 local_irq_disable();
3701 while (work
< quota
) {
3702 struct sk_buff
*skb
;
3705 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3707 __netif_receive_skb(skb
);
3708 local_irq_disable();
3709 input_queue_head_incr(sd
);
3710 if (++work
>= quota
) {
3717 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3719 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3720 &sd
->process_queue
);
3722 if (qlen
< quota
- work
) {
3724 * Inline a custom version of __napi_complete().
3725 * only current cpu owns and manipulates this napi,
3726 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3727 * we can use a plain write instead of clear_bit(),
3728 * and we dont need an smp_mb() memory barrier.
3730 list_del(&napi
->poll_list
);
3733 quota
= work
+ qlen
;
3743 * __napi_schedule - schedule for receive
3744 * @n: entry to schedule
3746 * The entry's receive function will be scheduled to run
3748 void __napi_schedule(struct napi_struct
*n
)
3750 unsigned long flags
;
3752 local_irq_save(flags
);
3753 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3754 local_irq_restore(flags
);
3756 EXPORT_SYMBOL(__napi_schedule
);
3758 void __napi_complete(struct napi_struct
*n
)
3760 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3761 BUG_ON(n
->gro_list
);
3763 list_del(&n
->poll_list
);
3764 smp_mb__before_clear_bit();
3765 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3767 EXPORT_SYMBOL(__napi_complete
);
3769 void napi_complete(struct napi_struct
*n
)
3771 unsigned long flags
;
3774 * don't let napi dequeue from the cpu poll list
3775 * just in case its running on a different cpu
3777 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3781 local_irq_save(flags
);
3783 local_irq_restore(flags
);
3785 EXPORT_SYMBOL(napi_complete
);
3787 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3788 int (*poll
)(struct napi_struct
*, int), int weight
)
3790 INIT_LIST_HEAD(&napi
->poll_list
);
3791 napi
->gro_count
= 0;
3792 napi
->gro_list
= NULL
;
3795 napi
->weight
= weight
;
3796 list_add(&napi
->dev_list
, &dev
->napi_list
);
3798 #ifdef CONFIG_NETPOLL
3799 spin_lock_init(&napi
->poll_lock
);
3800 napi
->poll_owner
= -1;
3802 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3804 EXPORT_SYMBOL(netif_napi_add
);
3806 void netif_napi_del(struct napi_struct
*napi
)
3808 struct sk_buff
*skb
, *next
;
3810 list_del_init(&napi
->dev_list
);
3811 napi_free_frags(napi
);
3813 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3819 napi
->gro_list
= NULL
;
3820 napi
->gro_count
= 0;
3822 EXPORT_SYMBOL(netif_napi_del
);
3824 static void net_rx_action(struct softirq_action
*h
)
3826 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3827 unsigned long time_limit
= jiffies
+ 2;
3828 int budget
= netdev_budget
;
3831 local_irq_disable();
3833 while (!list_empty(&sd
->poll_list
)) {
3834 struct napi_struct
*n
;
3837 /* If softirq window is exhuasted then punt.
3838 * Allow this to run for 2 jiffies since which will allow
3839 * an average latency of 1.5/HZ.
3841 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3846 /* Even though interrupts have been re-enabled, this
3847 * access is safe because interrupts can only add new
3848 * entries to the tail of this list, and only ->poll()
3849 * calls can remove this head entry from the list.
3851 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3853 have
= netpoll_poll_lock(n
);
3857 /* This NAPI_STATE_SCHED test is for avoiding a race
3858 * with netpoll's poll_napi(). Only the entity which
3859 * obtains the lock and sees NAPI_STATE_SCHED set will
3860 * actually make the ->poll() call. Therefore we avoid
3861 * accidentally calling ->poll() when NAPI is not scheduled.
3864 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3865 work
= n
->poll(n
, weight
);
3869 WARN_ON_ONCE(work
> weight
);
3873 local_irq_disable();
3875 /* Drivers must not modify the NAPI state if they
3876 * consume the entire weight. In such cases this code
3877 * still "owns" the NAPI instance and therefore can
3878 * move the instance around on the list at-will.
3880 if (unlikely(work
== weight
)) {
3881 if (unlikely(napi_disable_pending(n
))) {
3884 local_irq_disable();
3886 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3889 netpoll_poll_unlock(have
);
3892 net_rps_action_and_irq_enable(sd
);
3894 #ifdef CONFIG_NET_DMA
3896 * There may not be any more sk_buffs coming right now, so push
3897 * any pending DMA copies to hardware
3899 dma_issue_pending_all();
3906 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3910 static gifconf_func_t
*gifconf_list
[NPROTO
];
3913 * register_gifconf - register a SIOCGIF handler
3914 * @family: Address family
3915 * @gifconf: Function handler
3917 * Register protocol dependent address dumping routines. The handler
3918 * that is passed must not be freed or reused until it has been replaced
3919 * by another handler.
3921 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3923 if (family
>= NPROTO
)
3925 gifconf_list
[family
] = gifconf
;
3928 EXPORT_SYMBOL(register_gifconf
);
3932 * Map an interface index to its name (SIOCGIFNAME)
3936 * We need this ioctl for efficient implementation of the
3937 * if_indextoname() function required by the IPv6 API. Without
3938 * it, we would have to search all the interfaces to find a
3942 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3944 struct net_device
*dev
;
3948 * Fetch the caller's info block.
3951 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3955 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3961 strcpy(ifr
.ifr_name
, dev
->name
);
3964 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3970 * Perform a SIOCGIFCONF call. This structure will change
3971 * size eventually, and there is nothing I can do about it.
3972 * Thus we will need a 'compatibility mode'.
3975 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3978 struct net_device
*dev
;
3985 * Fetch the caller's info block.
3988 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3995 * Loop over the interfaces, and write an info block for each.
3999 for_each_netdev(net
, dev
) {
4000 for (i
= 0; i
< NPROTO
; i
++) {
4001 if (gifconf_list
[i
]) {
4004 done
= gifconf_list
[i
](dev
, NULL
, 0);
4006 done
= gifconf_list
[i
](dev
, pos
+ total
,
4016 * All done. Write the updated control block back to the caller.
4018 ifc
.ifc_len
= total
;
4021 * Both BSD and Solaris return 0 here, so we do too.
4023 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4026 #ifdef CONFIG_PROC_FS
4028 * This is invoked by the /proc filesystem handler to display a device
4031 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4034 struct net
*net
= seq_file_net(seq
);
4036 struct net_device
*dev
;
4040 return SEQ_START_TOKEN
;
4043 for_each_netdev_rcu(net
, dev
)
4050 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4052 struct net_device
*dev
= v
;
4054 if (v
== SEQ_START_TOKEN
)
4055 dev
= first_net_device_rcu(seq_file_net(seq
));
4057 dev
= next_net_device_rcu(dev
);
4063 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4069 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4071 struct rtnl_link_stats64 temp
;
4072 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4074 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4075 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4076 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4078 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4079 stats
->rx_fifo_errors
,
4080 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4081 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4082 stats
->rx_compressed
, stats
->multicast
,
4083 stats
->tx_bytes
, stats
->tx_packets
,
4084 stats
->tx_errors
, stats
->tx_dropped
,
4085 stats
->tx_fifo_errors
, stats
->collisions
,
4086 stats
->tx_carrier_errors
+
4087 stats
->tx_aborted_errors
+
4088 stats
->tx_window_errors
+
4089 stats
->tx_heartbeat_errors
,
4090 stats
->tx_compressed
);
4094 * Called from the PROCfs module. This now uses the new arbitrary sized
4095 * /proc/net interface to create /proc/net/dev
4097 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4099 if (v
== SEQ_START_TOKEN
)
4100 seq_puts(seq
, "Inter-| Receive "
4102 " face |bytes packets errs drop fifo frame "
4103 "compressed multicast|bytes packets errs "
4104 "drop fifo colls carrier compressed\n");
4106 dev_seq_printf_stats(seq
, v
);
4110 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4112 struct softnet_data
*sd
= NULL
;
4114 while (*pos
< nr_cpu_ids
)
4115 if (cpu_online(*pos
)) {
4116 sd
= &per_cpu(softnet_data
, *pos
);
4123 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4125 return softnet_get_online(pos
);
4128 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4131 return softnet_get_online(pos
);
4134 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4138 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4140 struct softnet_data
*sd
= v
;
4142 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4143 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4144 0, 0, 0, 0, /* was fastroute */
4145 sd
->cpu_collision
, sd
->received_rps
);
4149 static const struct seq_operations dev_seq_ops
= {
4150 .start
= dev_seq_start
,
4151 .next
= dev_seq_next
,
4152 .stop
= dev_seq_stop
,
4153 .show
= dev_seq_show
,
4156 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4158 return seq_open_net(inode
, file
, &dev_seq_ops
,
4159 sizeof(struct seq_net_private
));
4162 static const struct file_operations dev_seq_fops
= {
4163 .owner
= THIS_MODULE
,
4164 .open
= dev_seq_open
,
4166 .llseek
= seq_lseek
,
4167 .release
= seq_release_net
,
4170 static const struct seq_operations softnet_seq_ops
= {
4171 .start
= softnet_seq_start
,
4172 .next
= softnet_seq_next
,
4173 .stop
= softnet_seq_stop
,
4174 .show
= softnet_seq_show
,
4177 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4179 return seq_open(file
, &softnet_seq_ops
);
4182 static const struct file_operations softnet_seq_fops
= {
4183 .owner
= THIS_MODULE
,
4184 .open
= softnet_seq_open
,
4186 .llseek
= seq_lseek
,
4187 .release
= seq_release
,
4190 static void *ptype_get_idx(loff_t pos
)
4192 struct packet_type
*pt
= NULL
;
4196 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4202 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4203 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4212 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4216 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4219 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4221 struct packet_type
*pt
;
4222 struct list_head
*nxt
;
4226 if (v
== SEQ_START_TOKEN
)
4227 return ptype_get_idx(0);
4230 nxt
= pt
->list
.next
;
4231 if (pt
->type
== htons(ETH_P_ALL
)) {
4232 if (nxt
!= &ptype_all
)
4235 nxt
= ptype_base
[0].next
;
4237 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4239 while (nxt
== &ptype_base
[hash
]) {
4240 if (++hash
>= PTYPE_HASH_SIZE
)
4242 nxt
= ptype_base
[hash
].next
;
4245 return list_entry(nxt
, struct packet_type
, list
);
4248 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4254 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4256 struct packet_type
*pt
= v
;
4258 if (v
== SEQ_START_TOKEN
)
4259 seq_puts(seq
, "Type Device Function\n");
4260 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4261 if (pt
->type
== htons(ETH_P_ALL
))
4262 seq_puts(seq
, "ALL ");
4264 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4266 seq_printf(seq
, " %-8s %pF\n",
4267 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4273 static const struct seq_operations ptype_seq_ops
= {
4274 .start
= ptype_seq_start
,
4275 .next
= ptype_seq_next
,
4276 .stop
= ptype_seq_stop
,
4277 .show
= ptype_seq_show
,
4280 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4282 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4283 sizeof(struct seq_net_private
));
4286 static const struct file_operations ptype_seq_fops
= {
4287 .owner
= THIS_MODULE
,
4288 .open
= ptype_seq_open
,
4290 .llseek
= seq_lseek
,
4291 .release
= seq_release_net
,
4295 static int __net_init
dev_proc_net_init(struct net
*net
)
4299 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4301 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4303 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4306 if (wext_proc_init(net
))
4312 proc_net_remove(net
, "ptype");
4314 proc_net_remove(net
, "softnet_stat");
4316 proc_net_remove(net
, "dev");
4320 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4322 wext_proc_exit(net
);
4324 proc_net_remove(net
, "ptype");
4325 proc_net_remove(net
, "softnet_stat");
4326 proc_net_remove(net
, "dev");
4329 static struct pernet_operations __net_initdata dev_proc_ops
= {
4330 .init
= dev_proc_net_init
,
4331 .exit
= dev_proc_net_exit
,
4334 static int __init
dev_proc_init(void)
4336 return register_pernet_subsys(&dev_proc_ops
);
4339 #define dev_proc_init() 0
4340 #endif /* CONFIG_PROC_FS */
4344 * netdev_set_master - set up master pointer
4345 * @slave: slave device
4346 * @master: new master device
4348 * Changes the master device of the slave. Pass %NULL to break the
4349 * bonding. The caller must hold the RTNL semaphore. On a failure
4350 * a negative errno code is returned. On success the reference counts
4351 * are adjusted and the function returns zero.
4353 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4355 struct net_device
*old
= slave
->master
;
4365 slave
->master
= master
;
4371 EXPORT_SYMBOL(netdev_set_master
);
4374 * netdev_set_bond_master - set up bonding master/slave pair
4375 * @slave: slave device
4376 * @master: new master device
4378 * Changes the master device of the slave. Pass %NULL to break the
4379 * bonding. The caller must hold the RTNL semaphore. On a failure
4380 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4381 * to the routing socket and the function returns zero.
4383 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4389 err
= netdev_set_master(slave
, master
);
4393 slave
->flags
|= IFF_SLAVE
;
4395 slave
->flags
&= ~IFF_SLAVE
;
4397 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4400 EXPORT_SYMBOL(netdev_set_bond_master
);
4402 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4404 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4406 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4407 ops
->ndo_change_rx_flags(dev
, flags
);
4410 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4412 unsigned short old_flags
= dev
->flags
;
4418 dev
->flags
|= IFF_PROMISC
;
4419 dev
->promiscuity
+= inc
;
4420 if (dev
->promiscuity
== 0) {
4423 * If inc causes overflow, untouch promisc and return error.
4426 dev
->flags
&= ~IFF_PROMISC
;
4428 dev
->promiscuity
-= inc
;
4429 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4430 "set promiscuity failed, promiscuity feature "
4431 "of device might be broken.\n", dev
->name
);
4435 if (dev
->flags
!= old_flags
) {
4436 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4437 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4439 if (audit_enabled
) {
4440 current_uid_gid(&uid
, &gid
);
4441 audit_log(current
->audit_context
, GFP_ATOMIC
,
4442 AUDIT_ANOM_PROMISCUOUS
,
4443 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4444 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4445 (old_flags
& IFF_PROMISC
),
4446 audit_get_loginuid(current
),
4448 audit_get_sessionid(current
));
4451 dev_change_rx_flags(dev
, IFF_PROMISC
);
4457 * dev_set_promiscuity - update promiscuity count on a device
4461 * Add or remove promiscuity from a device. While the count in the device
4462 * remains above zero the interface remains promiscuous. Once it hits zero
4463 * the device reverts back to normal filtering operation. A negative inc
4464 * value is used to drop promiscuity on the device.
4465 * Return 0 if successful or a negative errno code on error.
4467 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4469 unsigned short old_flags
= dev
->flags
;
4472 err
= __dev_set_promiscuity(dev
, inc
);
4475 if (dev
->flags
!= old_flags
)
4476 dev_set_rx_mode(dev
);
4479 EXPORT_SYMBOL(dev_set_promiscuity
);
4482 * dev_set_allmulti - update allmulti count on a device
4486 * Add or remove reception of all multicast frames to a device. While the
4487 * count in the device remains above zero the interface remains listening
4488 * to all interfaces. Once it hits zero the device reverts back to normal
4489 * filtering operation. A negative @inc value is used to drop the counter
4490 * when releasing a resource needing all multicasts.
4491 * Return 0 if successful or a negative errno code on error.
4494 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4496 unsigned short old_flags
= dev
->flags
;
4500 dev
->flags
|= IFF_ALLMULTI
;
4501 dev
->allmulti
+= inc
;
4502 if (dev
->allmulti
== 0) {
4505 * If inc causes overflow, untouch allmulti and return error.
4508 dev
->flags
&= ~IFF_ALLMULTI
;
4510 dev
->allmulti
-= inc
;
4511 printk(KERN_WARNING
"%s: allmulti touches roof, "
4512 "set allmulti failed, allmulti feature of "
4513 "device might be broken.\n", dev
->name
);
4517 if (dev
->flags
^ old_flags
) {
4518 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4519 dev_set_rx_mode(dev
);
4523 EXPORT_SYMBOL(dev_set_allmulti
);
4526 * Upload unicast and multicast address lists to device and
4527 * configure RX filtering. When the device doesn't support unicast
4528 * filtering it is put in promiscuous mode while unicast addresses
4531 void __dev_set_rx_mode(struct net_device
*dev
)
4533 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4535 /* dev_open will call this function so the list will stay sane. */
4536 if (!(dev
->flags
&IFF_UP
))
4539 if (!netif_device_present(dev
))
4542 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4543 /* Unicast addresses changes may only happen under the rtnl,
4544 * therefore calling __dev_set_promiscuity here is safe.
4546 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4547 __dev_set_promiscuity(dev
, 1);
4548 dev
->uc_promisc
= true;
4549 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4550 __dev_set_promiscuity(dev
, -1);
4551 dev
->uc_promisc
= false;
4555 if (ops
->ndo_set_rx_mode
)
4556 ops
->ndo_set_rx_mode(dev
);
4559 void dev_set_rx_mode(struct net_device
*dev
)
4561 netif_addr_lock_bh(dev
);
4562 __dev_set_rx_mode(dev
);
4563 netif_addr_unlock_bh(dev
);
4567 * dev_ethtool_get_settings - call device's ethtool_ops::get_settings()
4569 * @cmd: memory area for ethtool_ops::get_settings() result
4571 * The cmd arg is initialized properly (cleared and
4572 * ethtool_cmd::cmd field set to ETHTOOL_GSET).
4574 * Return device's ethtool_ops::get_settings() result value or
4575 * -EOPNOTSUPP when device doesn't expose
4576 * ethtool_ops::get_settings() operation.
4578 int dev_ethtool_get_settings(struct net_device
*dev
,
4579 struct ethtool_cmd
*cmd
)
4581 if (!dev
->ethtool_ops
|| !dev
->ethtool_ops
->get_settings
)
4584 memset(cmd
, 0, sizeof(struct ethtool_cmd
));
4585 cmd
->cmd
= ETHTOOL_GSET
;
4586 return dev
->ethtool_ops
->get_settings(dev
, cmd
);
4588 EXPORT_SYMBOL(dev_ethtool_get_settings
);
4591 * dev_get_flags - get flags reported to userspace
4594 * Get the combination of flag bits exported through APIs to userspace.
4596 unsigned dev_get_flags(const struct net_device
*dev
)
4600 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4605 (dev
->gflags
& (IFF_PROMISC
|
4608 if (netif_running(dev
)) {
4609 if (netif_oper_up(dev
))
4610 flags
|= IFF_RUNNING
;
4611 if (netif_carrier_ok(dev
))
4612 flags
|= IFF_LOWER_UP
;
4613 if (netif_dormant(dev
))
4614 flags
|= IFF_DORMANT
;
4619 EXPORT_SYMBOL(dev_get_flags
);
4621 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4623 int old_flags
= dev
->flags
;
4629 * Set the flags on our device.
4632 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4633 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4635 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4639 * Load in the correct multicast list now the flags have changed.
4642 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4643 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4645 dev_set_rx_mode(dev
);
4648 * Have we downed the interface. We handle IFF_UP ourselves
4649 * according to user attempts to set it, rather than blindly
4654 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4655 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4658 dev_set_rx_mode(dev
);
4661 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4662 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4664 dev
->gflags
^= IFF_PROMISC
;
4665 dev_set_promiscuity(dev
, inc
);
4668 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4669 is important. Some (broken) drivers set IFF_PROMISC, when
4670 IFF_ALLMULTI is requested not asking us and not reporting.
4672 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4673 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4675 dev
->gflags
^= IFF_ALLMULTI
;
4676 dev_set_allmulti(dev
, inc
);
4682 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4684 unsigned int changes
= dev
->flags
^ old_flags
;
4686 if (changes
& IFF_UP
) {
4687 if (dev
->flags
& IFF_UP
)
4688 call_netdevice_notifiers(NETDEV_UP
, dev
);
4690 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4693 if (dev
->flags
& IFF_UP
&&
4694 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4695 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4699 * dev_change_flags - change device settings
4701 * @flags: device state flags
4703 * Change settings on device based state flags. The flags are
4704 * in the userspace exported format.
4706 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4709 int old_flags
= dev
->flags
;
4711 ret
= __dev_change_flags(dev
, flags
);
4715 changes
= old_flags
^ dev
->flags
;
4717 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4719 __dev_notify_flags(dev
, old_flags
);
4722 EXPORT_SYMBOL(dev_change_flags
);
4725 * dev_set_mtu - Change maximum transfer unit
4727 * @new_mtu: new transfer unit
4729 * Change the maximum transfer size of the network device.
4731 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4733 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4736 if (new_mtu
== dev
->mtu
)
4739 /* MTU must be positive. */
4743 if (!netif_device_present(dev
))
4747 if (ops
->ndo_change_mtu
)
4748 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4752 if (!err
&& dev
->flags
& IFF_UP
)
4753 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4756 EXPORT_SYMBOL(dev_set_mtu
);
4759 * dev_set_group - Change group this device belongs to
4761 * @new_group: group this device should belong to
4763 void dev_set_group(struct net_device
*dev
, int new_group
)
4765 dev
->group
= new_group
;
4767 EXPORT_SYMBOL(dev_set_group
);
4770 * dev_set_mac_address - Change Media Access Control Address
4774 * Change the hardware (MAC) address of the device
4776 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4778 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4781 if (!ops
->ndo_set_mac_address
)
4783 if (sa
->sa_family
!= dev
->type
)
4785 if (!netif_device_present(dev
))
4787 err
= ops
->ndo_set_mac_address(dev
, sa
);
4789 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4792 EXPORT_SYMBOL(dev_set_mac_address
);
4795 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4797 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4800 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4806 case SIOCGIFFLAGS
: /* Get interface flags */
4807 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4810 case SIOCGIFMETRIC
: /* Get the metric on the interface
4811 (currently unused) */
4812 ifr
->ifr_metric
= 0;
4815 case SIOCGIFMTU
: /* Get the MTU of a device */
4816 ifr
->ifr_mtu
= dev
->mtu
;
4821 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4823 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4824 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4825 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4833 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4834 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4835 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4836 ifr
->ifr_map
.irq
= dev
->irq
;
4837 ifr
->ifr_map
.dma
= dev
->dma
;
4838 ifr
->ifr_map
.port
= dev
->if_port
;
4842 ifr
->ifr_ifindex
= dev
->ifindex
;
4846 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4850 /* dev_ioctl() should ensure this case
4862 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4864 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4867 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4868 const struct net_device_ops
*ops
;
4873 ops
= dev
->netdev_ops
;
4876 case SIOCSIFFLAGS
: /* Set interface flags */
4877 return dev_change_flags(dev
, ifr
->ifr_flags
);
4879 case SIOCSIFMETRIC
: /* Set the metric on the interface
4880 (currently unused) */
4883 case SIOCSIFMTU
: /* Set the MTU of a device */
4884 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4887 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4889 case SIOCSIFHWBROADCAST
:
4890 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4892 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4893 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4894 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4898 if (ops
->ndo_set_config
) {
4899 if (!netif_device_present(dev
))
4901 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4906 if (!ops
->ndo_set_rx_mode
||
4907 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4909 if (!netif_device_present(dev
))
4911 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4914 if (!ops
->ndo_set_rx_mode
||
4915 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4917 if (!netif_device_present(dev
))
4919 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4922 if (ifr
->ifr_qlen
< 0)
4924 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4928 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4929 return dev_change_name(dev
, ifr
->ifr_newname
);
4932 * Unknown or private ioctl
4935 if ((cmd
>= SIOCDEVPRIVATE
&&
4936 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4937 cmd
== SIOCBONDENSLAVE
||
4938 cmd
== SIOCBONDRELEASE
||
4939 cmd
== SIOCBONDSETHWADDR
||
4940 cmd
== SIOCBONDSLAVEINFOQUERY
||
4941 cmd
== SIOCBONDINFOQUERY
||
4942 cmd
== SIOCBONDCHANGEACTIVE
||
4943 cmd
== SIOCGMIIPHY
||
4944 cmd
== SIOCGMIIREG
||
4945 cmd
== SIOCSMIIREG
||
4946 cmd
== SIOCBRADDIF
||
4947 cmd
== SIOCBRDELIF
||
4948 cmd
== SIOCSHWTSTAMP
||
4949 cmd
== SIOCWANDEV
) {
4951 if (ops
->ndo_do_ioctl
) {
4952 if (netif_device_present(dev
))
4953 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4965 * This function handles all "interface"-type I/O control requests. The actual
4966 * 'doing' part of this is dev_ifsioc above.
4970 * dev_ioctl - network device ioctl
4971 * @net: the applicable net namespace
4972 * @cmd: command to issue
4973 * @arg: pointer to a struct ifreq in user space
4975 * Issue ioctl functions to devices. This is normally called by the
4976 * user space syscall interfaces but can sometimes be useful for
4977 * other purposes. The return value is the return from the syscall if
4978 * positive or a negative errno code on error.
4981 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4987 /* One special case: SIOCGIFCONF takes ifconf argument
4988 and requires shared lock, because it sleeps writing
4992 if (cmd
== SIOCGIFCONF
) {
4994 ret
= dev_ifconf(net
, (char __user
*) arg
);
4998 if (cmd
== SIOCGIFNAME
)
4999 return dev_ifname(net
, (struct ifreq __user
*)arg
);
5001 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
5004 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
5006 colon
= strchr(ifr
.ifr_name
, ':');
5011 * See which interface the caller is talking about.
5016 * These ioctl calls:
5017 * - can be done by all.
5018 * - atomic and do not require locking.
5029 dev_load(net
, ifr
.ifr_name
);
5031 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5036 if (copy_to_user(arg
, &ifr
,
5037 sizeof(struct ifreq
)))
5043 dev_load(net
, ifr
.ifr_name
);
5045 ret
= dev_ethtool(net
, &ifr
);
5050 if (copy_to_user(arg
, &ifr
,
5051 sizeof(struct ifreq
)))
5057 * These ioctl calls:
5058 * - require superuser power.
5059 * - require strict serialization.
5065 if (!capable(CAP_NET_ADMIN
))
5067 dev_load(net
, ifr
.ifr_name
);
5069 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5074 if (copy_to_user(arg
, &ifr
,
5075 sizeof(struct ifreq
)))
5081 * These ioctl calls:
5082 * - require superuser power.
5083 * - require strict serialization.
5084 * - do not return a value
5094 case SIOCSIFHWBROADCAST
:
5097 case SIOCBONDENSLAVE
:
5098 case SIOCBONDRELEASE
:
5099 case SIOCBONDSETHWADDR
:
5100 case SIOCBONDCHANGEACTIVE
:
5104 if (!capable(CAP_NET_ADMIN
))
5107 case SIOCBONDSLAVEINFOQUERY
:
5108 case SIOCBONDINFOQUERY
:
5109 dev_load(net
, ifr
.ifr_name
);
5111 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5116 /* Get the per device memory space. We can add this but
5117 * currently do not support it */
5119 /* Set the per device memory buffer space.
5120 * Not applicable in our case */
5125 * Unknown or private ioctl.
5128 if (cmd
== SIOCWANDEV
||
5129 (cmd
>= SIOCDEVPRIVATE
&&
5130 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5131 dev_load(net
, ifr
.ifr_name
);
5133 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5135 if (!ret
&& copy_to_user(arg
, &ifr
,
5136 sizeof(struct ifreq
)))
5140 /* Take care of Wireless Extensions */
5141 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5142 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5149 * dev_new_index - allocate an ifindex
5150 * @net: the applicable net namespace
5152 * Returns a suitable unique value for a new device interface
5153 * number. The caller must hold the rtnl semaphore or the
5154 * dev_base_lock to be sure it remains unique.
5156 static int dev_new_index(struct net
*net
)
5162 if (!__dev_get_by_index(net
, ifindex
))
5167 /* Delayed registration/unregisteration */
5168 static LIST_HEAD(net_todo_list
);
5170 static void net_set_todo(struct net_device
*dev
)
5172 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5175 static void rollback_registered_many(struct list_head
*head
)
5177 struct net_device
*dev
, *tmp
;
5179 BUG_ON(dev_boot_phase
);
5182 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5183 /* Some devices call without registering
5184 * for initialization unwind. Remove those
5185 * devices and proceed with the remaining.
5187 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5188 pr_debug("unregister_netdevice: device %s/%p never "
5189 "was registered\n", dev
->name
, dev
);
5192 list_del(&dev
->unreg_list
);
5195 dev
->dismantle
= true;
5196 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5199 /* If device is running, close it first. */
5200 dev_close_many(head
);
5202 list_for_each_entry(dev
, head
, unreg_list
) {
5203 /* And unlink it from device chain. */
5204 unlist_netdevice(dev
);
5206 dev
->reg_state
= NETREG_UNREGISTERING
;
5211 list_for_each_entry(dev
, head
, unreg_list
) {
5212 /* Shutdown queueing discipline. */
5216 /* Notify protocols, that we are about to destroy
5217 this device. They should clean all the things.
5219 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5221 if (!dev
->rtnl_link_ops
||
5222 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5223 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5226 * Flush the unicast and multicast chains
5231 if (dev
->netdev_ops
->ndo_uninit
)
5232 dev
->netdev_ops
->ndo_uninit(dev
);
5234 /* Notifier chain MUST detach us from master device. */
5235 WARN_ON(dev
->master
);
5237 /* Remove entries from kobject tree */
5238 netdev_unregister_kobject(dev
);
5241 /* Process any work delayed until the end of the batch */
5242 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5243 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5247 list_for_each_entry(dev
, head
, unreg_list
)
5251 static void rollback_registered(struct net_device
*dev
)
5255 list_add(&dev
->unreg_list
, &single
);
5256 rollback_registered_many(&single
);
5260 static u32
netdev_fix_features(struct net_device
*dev
, u32 features
)
5262 /* Fix illegal checksum combinations */
5263 if ((features
& NETIF_F_HW_CSUM
) &&
5264 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5265 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5266 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5269 if ((features
& NETIF_F_NO_CSUM
) &&
5270 (features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5271 netdev_warn(dev
, "mixed no checksumming and other settings.\n");
5272 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5275 /* Fix illegal SG+CSUM combinations. */
5276 if ((features
& NETIF_F_SG
) &&
5277 !(features
& NETIF_F_ALL_CSUM
)) {
5279 "Dropping NETIF_F_SG since no checksum feature.\n");
5280 features
&= ~NETIF_F_SG
;
5283 /* TSO requires that SG is present as well. */
5284 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5285 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5286 features
&= ~NETIF_F_ALL_TSO
;
5289 /* TSO ECN requires that TSO is present as well. */
5290 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5291 features
&= ~NETIF_F_TSO_ECN
;
5293 /* Software GSO depends on SG. */
5294 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5295 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5296 features
&= ~NETIF_F_GSO
;
5299 /* UFO needs SG and checksumming */
5300 if (features
& NETIF_F_UFO
) {
5301 /* maybe split UFO into V4 and V6? */
5302 if (!((features
& NETIF_F_GEN_CSUM
) ||
5303 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5304 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5306 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5307 features
&= ~NETIF_F_UFO
;
5310 if (!(features
& NETIF_F_SG
)) {
5312 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5313 features
&= ~NETIF_F_UFO
;
5320 int __netdev_update_features(struct net_device
*dev
)
5327 features
= netdev_get_wanted_features(dev
);
5329 if (dev
->netdev_ops
->ndo_fix_features
)
5330 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5332 /* driver might be less strict about feature dependencies */
5333 features
= netdev_fix_features(dev
, features
);
5335 if (dev
->features
== features
)
5338 netdev_dbg(dev
, "Features changed: 0x%08x -> 0x%08x\n",
5339 dev
->features
, features
);
5341 if (dev
->netdev_ops
->ndo_set_features
)
5342 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5344 if (unlikely(err
< 0)) {
5346 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5347 err
, features
, dev
->features
);
5352 dev
->features
= features
;
5358 * netdev_update_features - recalculate device features
5359 * @dev: the device to check
5361 * Recalculate dev->features set and send notifications if it
5362 * has changed. Should be called after driver or hardware dependent
5363 * conditions might have changed that influence the features.
5365 void netdev_update_features(struct net_device
*dev
)
5367 if (__netdev_update_features(dev
))
5368 netdev_features_change(dev
);
5370 EXPORT_SYMBOL(netdev_update_features
);
5373 * netdev_change_features - recalculate device features
5374 * @dev: the device to check
5376 * Recalculate dev->features set and send notifications even
5377 * if they have not changed. Should be called instead of
5378 * netdev_update_features() if also dev->vlan_features might
5379 * have changed to allow the changes to be propagated to stacked
5382 void netdev_change_features(struct net_device
*dev
)
5384 __netdev_update_features(dev
);
5385 netdev_features_change(dev
);
5387 EXPORT_SYMBOL(netdev_change_features
);
5390 * netif_stacked_transfer_operstate - transfer operstate
5391 * @rootdev: the root or lower level device to transfer state from
5392 * @dev: the device to transfer operstate to
5394 * Transfer operational state from root to device. This is normally
5395 * called when a stacking relationship exists between the root
5396 * device and the device(a leaf device).
5398 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5399 struct net_device
*dev
)
5401 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5402 netif_dormant_on(dev
);
5404 netif_dormant_off(dev
);
5406 if (netif_carrier_ok(rootdev
)) {
5407 if (!netif_carrier_ok(dev
))
5408 netif_carrier_on(dev
);
5410 if (netif_carrier_ok(dev
))
5411 netif_carrier_off(dev
);
5414 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5417 static int netif_alloc_rx_queues(struct net_device
*dev
)
5419 unsigned int i
, count
= dev
->num_rx_queues
;
5420 struct netdev_rx_queue
*rx
;
5424 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5426 pr_err("netdev: Unable to allocate %u rx queues.\n", count
);
5431 for (i
= 0; i
< count
; i
++)
5437 static void netdev_init_one_queue(struct net_device
*dev
,
5438 struct netdev_queue
*queue
, void *_unused
)
5440 /* Initialize queue lock */
5441 spin_lock_init(&queue
->_xmit_lock
);
5442 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5443 queue
->xmit_lock_owner
= -1;
5444 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5448 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5450 unsigned int count
= dev
->num_tx_queues
;
5451 struct netdev_queue
*tx
;
5455 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5457 pr_err("netdev: Unable to allocate %u tx queues.\n",
5463 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5464 spin_lock_init(&dev
->tx_global_lock
);
5470 * register_netdevice - register a network device
5471 * @dev: device to register
5473 * Take a completed network device structure and add it to the kernel
5474 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5475 * chain. 0 is returned on success. A negative errno code is returned
5476 * on a failure to set up the device, or if the name is a duplicate.
5478 * Callers must hold the rtnl semaphore. You may want
5479 * register_netdev() instead of this.
5482 * The locking appears insufficient to guarantee two parallel registers
5483 * will not get the same name.
5486 int register_netdevice(struct net_device
*dev
)
5489 struct net
*net
= dev_net(dev
);
5491 BUG_ON(dev_boot_phase
);
5496 /* When net_device's are persistent, this will be fatal. */
5497 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5500 spin_lock_init(&dev
->addr_list_lock
);
5501 netdev_set_addr_lockdep_class(dev
);
5505 ret
= dev_get_valid_name(dev
, dev
->name
);
5509 /* Init, if this function is available */
5510 if (dev
->netdev_ops
->ndo_init
) {
5511 ret
= dev
->netdev_ops
->ndo_init(dev
);
5519 dev
->ifindex
= dev_new_index(net
);
5520 if (dev
->iflink
== -1)
5521 dev
->iflink
= dev
->ifindex
;
5523 /* Transfer changeable features to wanted_features and enable
5524 * software offloads (GSO and GRO).
5526 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5527 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5528 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5530 /* Turn on no cache copy if HW is doing checksum */
5531 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5532 if ((dev
->features
& NETIF_F_ALL_CSUM
) &&
5533 !(dev
->features
& NETIF_F_NO_CSUM
)) {
5534 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5535 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5538 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5540 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5542 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5543 ret
= notifier_to_errno(ret
);
5547 ret
= netdev_register_kobject(dev
);
5550 dev
->reg_state
= NETREG_REGISTERED
;
5552 __netdev_update_features(dev
);
5555 * Default initial state at registry is that the
5556 * device is present.
5559 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5561 dev_init_scheduler(dev
);
5563 list_netdevice(dev
);
5565 /* Notify protocols, that a new device appeared. */
5566 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5567 ret
= notifier_to_errno(ret
);
5569 rollback_registered(dev
);
5570 dev
->reg_state
= NETREG_UNREGISTERED
;
5573 * Prevent userspace races by waiting until the network
5574 * device is fully setup before sending notifications.
5576 if (!dev
->rtnl_link_ops
||
5577 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5578 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5584 if (dev
->netdev_ops
->ndo_uninit
)
5585 dev
->netdev_ops
->ndo_uninit(dev
);
5588 EXPORT_SYMBOL(register_netdevice
);
5591 * init_dummy_netdev - init a dummy network device for NAPI
5592 * @dev: device to init
5594 * This takes a network device structure and initialize the minimum
5595 * amount of fields so it can be used to schedule NAPI polls without
5596 * registering a full blown interface. This is to be used by drivers
5597 * that need to tie several hardware interfaces to a single NAPI
5598 * poll scheduler due to HW limitations.
5600 int init_dummy_netdev(struct net_device
*dev
)
5602 /* Clear everything. Note we don't initialize spinlocks
5603 * are they aren't supposed to be taken by any of the
5604 * NAPI code and this dummy netdev is supposed to be
5605 * only ever used for NAPI polls
5607 memset(dev
, 0, sizeof(struct net_device
));
5609 /* make sure we BUG if trying to hit standard
5610 * register/unregister code path
5612 dev
->reg_state
= NETREG_DUMMY
;
5614 /* NAPI wants this */
5615 INIT_LIST_HEAD(&dev
->napi_list
);
5617 /* a dummy interface is started by default */
5618 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5619 set_bit(__LINK_STATE_START
, &dev
->state
);
5621 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5622 * because users of this 'device' dont need to change
5628 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5632 * register_netdev - register a network device
5633 * @dev: device to register
5635 * Take a completed network device structure and add it to the kernel
5636 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5637 * chain. 0 is returned on success. A negative errno code is returned
5638 * on a failure to set up the device, or if the name is a duplicate.
5640 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5641 * and expands the device name if you passed a format string to
5644 int register_netdev(struct net_device
*dev
)
5649 err
= register_netdevice(dev
);
5653 EXPORT_SYMBOL(register_netdev
);
5655 int netdev_refcnt_read(const struct net_device
*dev
)
5659 for_each_possible_cpu(i
)
5660 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5663 EXPORT_SYMBOL(netdev_refcnt_read
);
5666 * netdev_wait_allrefs - wait until all references are gone.
5668 * This is called when unregistering network devices.
5670 * Any protocol or device that holds a reference should register
5671 * for netdevice notification, and cleanup and put back the
5672 * reference if they receive an UNREGISTER event.
5673 * We can get stuck here if buggy protocols don't correctly
5676 static void netdev_wait_allrefs(struct net_device
*dev
)
5678 unsigned long rebroadcast_time
, warning_time
;
5681 linkwatch_forget_dev(dev
);
5683 rebroadcast_time
= warning_time
= jiffies
;
5684 refcnt
= netdev_refcnt_read(dev
);
5686 while (refcnt
!= 0) {
5687 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5690 /* Rebroadcast unregister notification */
5691 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5692 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5693 * should have already handle it the first time */
5695 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5697 /* We must not have linkwatch events
5698 * pending on unregister. If this
5699 * happens, we simply run the queue
5700 * unscheduled, resulting in a noop
5703 linkwatch_run_queue();
5708 rebroadcast_time
= jiffies
;
5713 refcnt
= netdev_refcnt_read(dev
);
5715 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5716 printk(KERN_EMERG
"unregister_netdevice: "
5717 "waiting for %s to become free. Usage "
5720 warning_time
= jiffies
;
5729 * register_netdevice(x1);
5730 * register_netdevice(x2);
5732 * unregister_netdevice(y1);
5733 * unregister_netdevice(y2);
5739 * We are invoked by rtnl_unlock().
5740 * This allows us to deal with problems:
5741 * 1) We can delete sysfs objects which invoke hotplug
5742 * without deadlocking with linkwatch via keventd.
5743 * 2) Since we run with the RTNL semaphore not held, we can sleep
5744 * safely in order to wait for the netdev refcnt to drop to zero.
5746 * We must not return until all unregister events added during
5747 * the interval the lock was held have been completed.
5749 void netdev_run_todo(void)
5751 struct list_head list
;
5753 /* Snapshot list, allow later requests */
5754 list_replace_init(&net_todo_list
, &list
);
5758 while (!list_empty(&list
)) {
5759 struct net_device
*dev
5760 = list_first_entry(&list
, struct net_device
, todo_list
);
5761 list_del(&dev
->todo_list
);
5763 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5764 printk(KERN_ERR
"network todo '%s' but state %d\n",
5765 dev
->name
, dev
->reg_state
);
5770 dev
->reg_state
= NETREG_UNREGISTERED
;
5772 on_each_cpu(flush_backlog
, dev
, 1);
5774 netdev_wait_allrefs(dev
);
5777 BUG_ON(netdev_refcnt_read(dev
));
5778 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5779 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5780 WARN_ON(dev
->dn_ptr
);
5782 if (dev
->destructor
)
5783 dev
->destructor(dev
);
5785 /* Free network device */
5786 kobject_put(&dev
->dev
.kobj
);
5790 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5791 * fields in the same order, with only the type differing.
5793 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5794 const struct net_device_stats
*netdev_stats
)
5796 #if BITS_PER_LONG == 64
5797 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5798 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5800 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5801 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5802 u64
*dst
= (u64
*)stats64
;
5804 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5805 sizeof(*stats64
) / sizeof(u64
));
5806 for (i
= 0; i
< n
; i
++)
5812 * dev_get_stats - get network device statistics
5813 * @dev: device to get statistics from
5814 * @storage: place to store stats
5816 * Get network statistics from device. Return @storage.
5817 * The device driver may provide its own method by setting
5818 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5819 * otherwise the internal statistics structure is used.
5821 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5822 struct rtnl_link_stats64
*storage
)
5824 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5826 if (ops
->ndo_get_stats64
) {
5827 memset(storage
, 0, sizeof(*storage
));
5828 ops
->ndo_get_stats64(dev
, storage
);
5829 } else if (ops
->ndo_get_stats
) {
5830 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5832 netdev_stats_to_stats64(storage
, &dev
->stats
);
5834 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5837 EXPORT_SYMBOL(dev_get_stats
);
5839 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5841 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5843 #ifdef CONFIG_NET_CLS_ACT
5846 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5849 netdev_init_one_queue(dev
, queue
, NULL
);
5850 queue
->qdisc
= &noop_qdisc
;
5851 queue
->qdisc_sleeping
= &noop_qdisc
;
5852 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5858 * alloc_netdev_mqs - allocate network device
5859 * @sizeof_priv: size of private data to allocate space for
5860 * @name: device name format string
5861 * @setup: callback to initialize device
5862 * @txqs: the number of TX subqueues to allocate
5863 * @rxqs: the number of RX subqueues to allocate
5865 * Allocates a struct net_device with private data area for driver use
5866 * and performs basic initialization. Also allocates subquue structs
5867 * for each queue on the device.
5869 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5870 void (*setup
)(struct net_device
*),
5871 unsigned int txqs
, unsigned int rxqs
)
5873 struct net_device
*dev
;
5875 struct net_device
*p
;
5877 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5880 pr_err("alloc_netdev: Unable to allocate device "
5881 "with zero queues.\n");
5887 pr_err("alloc_netdev: Unable to allocate device "
5888 "with zero RX queues.\n");
5893 alloc_size
= sizeof(struct net_device
);
5895 /* ensure 32-byte alignment of private area */
5896 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5897 alloc_size
+= sizeof_priv
;
5899 /* ensure 32-byte alignment of whole construct */
5900 alloc_size
+= NETDEV_ALIGN
- 1;
5902 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5904 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5908 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5909 dev
->padded
= (char *)dev
- (char *)p
;
5911 dev
->pcpu_refcnt
= alloc_percpu(int);
5912 if (!dev
->pcpu_refcnt
)
5915 if (dev_addr_init(dev
))
5921 dev_net_set(dev
, &init_net
);
5923 dev
->gso_max_size
= GSO_MAX_SIZE
;
5925 INIT_LIST_HEAD(&dev
->napi_list
);
5926 INIT_LIST_HEAD(&dev
->unreg_list
);
5927 INIT_LIST_HEAD(&dev
->link_watch_list
);
5928 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5931 dev
->num_tx_queues
= txqs
;
5932 dev
->real_num_tx_queues
= txqs
;
5933 if (netif_alloc_netdev_queues(dev
))
5937 dev
->num_rx_queues
= rxqs
;
5938 dev
->real_num_rx_queues
= rxqs
;
5939 if (netif_alloc_rx_queues(dev
))
5943 strcpy(dev
->name
, name
);
5944 dev
->group
= INIT_NETDEV_GROUP
;
5952 free_percpu(dev
->pcpu_refcnt
);
5962 EXPORT_SYMBOL(alloc_netdev_mqs
);
5965 * free_netdev - free network device
5968 * This function does the last stage of destroying an allocated device
5969 * interface. The reference to the device object is released.
5970 * If this is the last reference then it will be freed.
5972 void free_netdev(struct net_device
*dev
)
5974 struct napi_struct
*p
, *n
;
5976 release_net(dev_net(dev
));
5983 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
5985 /* Flush device addresses */
5986 dev_addr_flush(dev
);
5988 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5991 free_percpu(dev
->pcpu_refcnt
);
5992 dev
->pcpu_refcnt
= NULL
;
5994 /* Compatibility with error handling in drivers */
5995 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5996 kfree((char *)dev
- dev
->padded
);
6000 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
6001 dev
->reg_state
= NETREG_RELEASED
;
6003 /* will free via device release */
6004 put_device(&dev
->dev
);
6006 EXPORT_SYMBOL(free_netdev
);
6009 * synchronize_net - Synchronize with packet receive processing
6011 * Wait for packets currently being received to be done.
6012 * Does not block later packets from starting.
6014 void synchronize_net(void)
6017 if (rtnl_is_locked())
6018 synchronize_rcu_expedited();
6022 EXPORT_SYMBOL(synchronize_net
);
6025 * unregister_netdevice_queue - remove device from the kernel
6029 * This function shuts down a device interface and removes it
6030 * from the kernel tables.
6031 * If head not NULL, device is queued to be unregistered later.
6033 * Callers must hold the rtnl semaphore. You may want
6034 * unregister_netdev() instead of this.
6037 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6042 list_move_tail(&dev
->unreg_list
, head
);
6044 rollback_registered(dev
);
6045 /* Finish processing unregister after unlock */
6049 EXPORT_SYMBOL(unregister_netdevice_queue
);
6052 * unregister_netdevice_many - unregister many devices
6053 * @head: list of devices
6055 void unregister_netdevice_many(struct list_head
*head
)
6057 struct net_device
*dev
;
6059 if (!list_empty(head
)) {
6060 rollback_registered_many(head
);
6061 list_for_each_entry(dev
, head
, unreg_list
)
6065 EXPORT_SYMBOL(unregister_netdevice_many
);
6068 * unregister_netdev - remove device from the kernel
6071 * This function shuts down a device interface and removes it
6072 * from the kernel tables.
6074 * This is just a wrapper for unregister_netdevice that takes
6075 * the rtnl semaphore. In general you want to use this and not
6076 * unregister_netdevice.
6078 void unregister_netdev(struct net_device
*dev
)
6081 unregister_netdevice(dev
);
6084 EXPORT_SYMBOL(unregister_netdev
);
6087 * dev_change_net_namespace - move device to different nethost namespace
6089 * @net: network namespace
6090 * @pat: If not NULL name pattern to try if the current device name
6091 * is already taken in the destination network namespace.
6093 * This function shuts down a device interface and moves it
6094 * to a new network namespace. On success 0 is returned, on
6095 * a failure a netagive errno code is returned.
6097 * Callers must hold the rtnl semaphore.
6100 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6106 /* Don't allow namespace local devices to be moved. */
6108 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6111 /* Ensure the device has been registrered */
6113 if (dev
->reg_state
!= NETREG_REGISTERED
)
6116 /* Get out if there is nothing todo */
6118 if (net_eq(dev_net(dev
), net
))
6121 /* Pick the destination device name, and ensure
6122 * we can use it in the destination network namespace.
6125 if (__dev_get_by_name(net
, dev
->name
)) {
6126 /* We get here if we can't use the current device name */
6129 if (dev_get_valid_name(dev
, pat
) < 0)
6134 * And now a mini version of register_netdevice unregister_netdevice.
6137 /* If device is running close it first. */
6140 /* And unlink it from device chain */
6142 unlist_netdevice(dev
);
6146 /* Shutdown queueing discipline. */
6149 /* Notify protocols, that we are about to destroy
6150 this device. They should clean all the things.
6152 Note that dev->reg_state stays at NETREG_REGISTERED.
6153 This is wanted because this way 8021q and macvlan know
6154 the device is just moving and can keep their slaves up.
6156 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6157 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6160 * Flush the unicast and multicast chains
6165 /* Actually switch the network namespace */
6166 dev_net_set(dev
, net
);
6168 /* If there is an ifindex conflict assign a new one */
6169 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6170 int iflink
= (dev
->iflink
== dev
->ifindex
);
6171 dev
->ifindex
= dev_new_index(net
);
6173 dev
->iflink
= dev
->ifindex
;
6176 /* Fixup kobjects */
6177 err
= device_rename(&dev
->dev
, dev
->name
);
6180 /* Add the device back in the hashes */
6181 list_netdevice(dev
);
6183 /* Notify protocols, that a new device appeared. */
6184 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6187 * Prevent userspace races by waiting until the network
6188 * device is fully setup before sending notifications.
6190 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6197 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6199 static int dev_cpu_callback(struct notifier_block
*nfb
,
6200 unsigned long action
,
6203 struct sk_buff
**list_skb
;
6204 struct sk_buff
*skb
;
6205 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6206 struct softnet_data
*sd
, *oldsd
;
6208 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6211 local_irq_disable();
6212 cpu
= smp_processor_id();
6213 sd
= &per_cpu(softnet_data
, cpu
);
6214 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6216 /* Find end of our completion_queue. */
6217 list_skb
= &sd
->completion_queue
;
6219 list_skb
= &(*list_skb
)->next
;
6220 /* Append completion queue from offline CPU. */
6221 *list_skb
= oldsd
->completion_queue
;
6222 oldsd
->completion_queue
= NULL
;
6224 /* Append output queue from offline CPU. */
6225 if (oldsd
->output_queue
) {
6226 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6227 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6228 oldsd
->output_queue
= NULL
;
6229 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6231 /* Append NAPI poll list from offline CPU. */
6232 if (!list_empty(&oldsd
->poll_list
)) {
6233 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6234 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6237 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6240 /* Process offline CPU's input_pkt_queue */
6241 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6243 input_queue_head_incr(oldsd
);
6245 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6247 input_queue_head_incr(oldsd
);
6255 * netdev_increment_features - increment feature set by one
6256 * @all: current feature set
6257 * @one: new feature set
6258 * @mask: mask feature set
6260 * Computes a new feature set after adding a device with feature set
6261 * @one to the master device with current feature set @all. Will not
6262 * enable anything that is off in @mask. Returns the new feature set.
6264 u32
netdev_increment_features(u32 all
, u32 one
, u32 mask
)
6266 if (mask
& NETIF_F_GEN_CSUM
)
6267 mask
|= NETIF_F_ALL_CSUM
;
6268 mask
|= NETIF_F_VLAN_CHALLENGED
;
6270 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6271 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6273 /* If device needs checksumming, downgrade to it. */
6274 if (all
& (NETIF_F_ALL_CSUM
& ~NETIF_F_NO_CSUM
))
6275 all
&= ~NETIF_F_NO_CSUM
;
6277 /* If one device supports hw checksumming, set for all. */
6278 if (all
& NETIF_F_GEN_CSUM
)
6279 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6283 EXPORT_SYMBOL(netdev_increment_features
);
6285 static struct hlist_head
*netdev_create_hash(void)
6288 struct hlist_head
*hash
;
6290 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6292 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6293 INIT_HLIST_HEAD(&hash
[i
]);
6298 /* Initialize per network namespace state */
6299 static int __net_init
netdev_init(struct net
*net
)
6301 INIT_LIST_HEAD(&net
->dev_base_head
);
6303 net
->dev_name_head
= netdev_create_hash();
6304 if (net
->dev_name_head
== NULL
)
6307 net
->dev_index_head
= netdev_create_hash();
6308 if (net
->dev_index_head
== NULL
)
6314 kfree(net
->dev_name_head
);
6320 * netdev_drivername - network driver for the device
6321 * @dev: network device
6323 * Determine network driver for device.
6325 const char *netdev_drivername(const struct net_device
*dev
)
6327 const struct device_driver
*driver
;
6328 const struct device
*parent
;
6329 const char *empty
= "";
6331 parent
= dev
->dev
.parent
;
6335 driver
= parent
->driver
;
6336 if (driver
&& driver
->name
)
6337 return driver
->name
;
6341 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6342 struct va_format
*vaf
)
6346 if (dev
&& dev
->dev
.parent
)
6347 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6348 netdev_name(dev
), vaf
);
6350 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6352 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6357 int netdev_printk(const char *level
, const struct net_device
*dev
,
6358 const char *format
, ...)
6360 struct va_format vaf
;
6364 va_start(args
, format
);
6369 r
= __netdev_printk(level
, dev
, &vaf
);
6374 EXPORT_SYMBOL(netdev_printk
);
6376 #define define_netdev_printk_level(func, level) \
6377 int func(const struct net_device *dev, const char *fmt, ...) \
6380 struct va_format vaf; \
6383 va_start(args, fmt); \
6388 r = __netdev_printk(level, dev, &vaf); \
6393 EXPORT_SYMBOL(func);
6395 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6396 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6397 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6398 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6399 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6400 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6401 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6403 static void __net_exit
netdev_exit(struct net
*net
)
6405 kfree(net
->dev_name_head
);
6406 kfree(net
->dev_index_head
);
6409 static struct pernet_operations __net_initdata netdev_net_ops
= {
6410 .init
= netdev_init
,
6411 .exit
= netdev_exit
,
6414 static void __net_exit
default_device_exit(struct net
*net
)
6416 struct net_device
*dev
, *aux
;
6418 * Push all migratable network devices back to the
6419 * initial network namespace
6422 for_each_netdev_safe(net
, dev
, aux
) {
6424 char fb_name
[IFNAMSIZ
];
6426 /* Ignore unmoveable devices (i.e. loopback) */
6427 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6430 /* Leave virtual devices for the generic cleanup */
6431 if (dev
->rtnl_link_ops
)
6434 /* Push remaining network devices to init_net */
6435 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6436 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6438 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
6439 __func__
, dev
->name
, err
);
6446 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6448 /* At exit all network devices most be removed from a network
6449 * namespace. Do this in the reverse order of registration.
6450 * Do this across as many network namespaces as possible to
6451 * improve batching efficiency.
6453 struct net_device
*dev
;
6455 LIST_HEAD(dev_kill_list
);
6458 list_for_each_entry(net
, net_list
, exit_list
) {
6459 for_each_netdev_reverse(net
, dev
) {
6460 if (dev
->rtnl_link_ops
)
6461 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6463 unregister_netdevice_queue(dev
, &dev_kill_list
);
6466 unregister_netdevice_many(&dev_kill_list
);
6467 list_del(&dev_kill_list
);
6471 static struct pernet_operations __net_initdata default_device_ops
= {
6472 .exit
= default_device_exit
,
6473 .exit_batch
= default_device_exit_batch
,
6477 * Initialize the DEV module. At boot time this walks the device list and
6478 * unhooks any devices that fail to initialise (normally hardware not
6479 * present) and leaves us with a valid list of present and active devices.
6484 * This is called single threaded during boot, so no need
6485 * to take the rtnl semaphore.
6487 static int __init
net_dev_init(void)
6489 int i
, rc
= -ENOMEM
;
6491 BUG_ON(!dev_boot_phase
);
6493 if (dev_proc_init())
6496 if (netdev_kobject_init())
6499 INIT_LIST_HEAD(&ptype_all
);
6500 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6501 INIT_LIST_HEAD(&ptype_base
[i
]);
6503 if (register_pernet_subsys(&netdev_net_ops
))
6507 * Initialise the packet receive queues.
6510 for_each_possible_cpu(i
) {
6511 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6513 memset(sd
, 0, sizeof(*sd
));
6514 skb_queue_head_init(&sd
->input_pkt_queue
);
6515 skb_queue_head_init(&sd
->process_queue
);
6516 sd
->completion_queue
= NULL
;
6517 INIT_LIST_HEAD(&sd
->poll_list
);
6518 sd
->output_queue
= NULL
;
6519 sd
->output_queue_tailp
= &sd
->output_queue
;
6521 sd
->csd
.func
= rps_trigger_softirq
;
6527 sd
->backlog
.poll
= process_backlog
;
6528 sd
->backlog
.weight
= weight_p
;
6529 sd
->backlog
.gro_list
= NULL
;
6530 sd
->backlog
.gro_count
= 0;
6535 /* The loopback device is special if any other network devices
6536 * is present in a network namespace the loopback device must
6537 * be present. Since we now dynamically allocate and free the
6538 * loopback device ensure this invariant is maintained by
6539 * keeping the loopback device as the first device on the
6540 * list of network devices. Ensuring the loopback devices
6541 * is the first device that appears and the last network device
6544 if (register_pernet_device(&loopback_net_ops
))
6547 if (register_pernet_device(&default_device_ops
))
6550 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6551 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6553 hotcpu_notifier(dev_cpu_callback
, 0);
6561 subsys_initcall(net_dev_init
);
6563 static int __init
initialize_hashrnd(void)
6565 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6569 late_initcall_sync(initialize_hashrnd
);