2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 * The list of packet types we will receive (as opposed to discard)
146 * and the routines to invoke.
148 * Why 16. Because with 16 the only overlap we get on a hash of the
149 * low nibble of the protocol value is RARP/SNAP/X.25.
151 * NOTE: That is no longer true with the addition of VLAN tags. Not
152 * sure which should go first, but I bet it won't make much
153 * difference if we are running VLANs. The good news is that
154 * this protocol won't be in the list unless compiled in, so
155 * the average user (w/out VLANs) will not be adversely affected.
172 #define PTYPE_HASH_SIZE (16)
173 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
175 static DEFINE_SPINLOCK(ptype_lock
);
176 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
177 static struct list_head ptype_all __read_mostly
; /* Taps */
180 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
183 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
185 * Writers must hold the rtnl semaphore while they loop through the
186 * dev_base_head list, and hold dev_base_lock for writing when they do the
187 * actual updates. This allows pure readers to access the list even
188 * while a writer is preparing to update it.
190 * To put it another way, dev_base_lock is held for writing only to
191 * protect against pure readers; the rtnl semaphore provides the
192 * protection against other writers.
194 * See, for example usages, register_netdevice() and
195 * unregister_netdevice(), which must be called with the rtnl
198 DEFINE_RWLOCK(dev_base_lock
);
199 EXPORT_SYMBOL(dev_base_lock
);
201 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
203 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
204 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
207 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
209 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
212 static inline void rps_lock(struct softnet_data
*sd
)
215 spin_lock(&sd
->input_pkt_queue
.lock
);
219 static inline void rps_unlock(struct softnet_data
*sd
)
222 spin_unlock(&sd
->input_pkt_queue
.lock
);
226 /* Device list insertion */
227 static int list_netdevice(struct net_device
*dev
)
229 struct net
*net
= dev_net(dev
);
233 write_lock_bh(&dev_base_lock
);
234 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
235 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
236 hlist_add_head_rcu(&dev
->index_hlist
,
237 dev_index_hash(net
, dev
->ifindex
));
238 write_unlock_bh(&dev_base_lock
);
242 /* Device list removal
243 * caller must respect a RCU grace period before freeing/reusing dev
245 static void unlist_netdevice(struct net_device
*dev
)
249 /* Unlink dev from the device chain */
250 write_lock_bh(&dev_base_lock
);
251 list_del_rcu(&dev
->dev_list
);
252 hlist_del_rcu(&dev
->name_hlist
);
253 hlist_del_rcu(&dev
->index_hlist
);
254 write_unlock_bh(&dev_base_lock
);
261 static RAW_NOTIFIER_HEAD(netdev_chain
);
264 * Device drivers call our routines to queue packets here. We empty the
265 * queue in the local softnet handler.
268 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
269 EXPORT_PER_CPU_SYMBOL(softnet_data
);
271 #ifdef CONFIG_LOCKDEP
273 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
274 * according to dev->type
276 static const unsigned short netdev_lock_type
[] =
277 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
278 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
279 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
280 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
281 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
282 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
283 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
284 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
285 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
286 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
287 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
288 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
289 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
290 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
,
291 ARPHRD_PHONET_PIPE
, ARPHRD_IEEE802154
,
292 ARPHRD_VOID
, ARPHRD_NONE
};
294 static const char *const netdev_lock_name
[] =
295 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
296 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
297 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
298 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
299 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
300 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
301 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
302 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
303 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
304 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
305 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
306 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
307 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
308 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
309 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
310 "_xmit_VOID", "_xmit_NONE"};
312 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
313 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
315 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
319 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
320 if (netdev_lock_type
[i
] == dev_type
)
322 /* the last key is used by default */
323 return ARRAY_SIZE(netdev_lock_type
) - 1;
326 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
327 unsigned short dev_type
)
331 i
= netdev_lock_pos(dev_type
);
332 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
333 netdev_lock_name
[i
]);
336 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
340 i
= netdev_lock_pos(dev
->type
);
341 lockdep_set_class_and_name(&dev
->addr_list_lock
,
342 &netdev_addr_lock_key
[i
],
343 netdev_lock_name
[i
]);
346 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
347 unsigned short dev_type
)
350 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
355 /*******************************************************************************
357 Protocol management and registration routines
359 *******************************************************************************/
362 * Add a protocol ID to the list. Now that the input handler is
363 * smarter we can dispense with all the messy stuff that used to be
366 * BEWARE!!! Protocol handlers, mangling input packets,
367 * MUST BE last in hash buckets and checking protocol handlers
368 * MUST start from promiscuous ptype_all chain in net_bh.
369 * It is true now, do not change it.
370 * Explanation follows: if protocol handler, mangling packet, will
371 * be the first on list, it is not able to sense, that packet
372 * is cloned and should be copied-on-write, so that it will
373 * change it and subsequent readers will get broken packet.
377 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
379 if (pt
->type
== htons(ETH_P_ALL
))
382 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
386 * dev_add_pack - add packet handler
387 * @pt: packet type declaration
389 * Add a protocol handler to the networking stack. The passed &packet_type
390 * is linked into kernel lists and may not be freed until it has been
391 * removed from the kernel lists.
393 * This call does not sleep therefore it can not
394 * guarantee all CPU's that are in middle of receiving packets
395 * will see the new packet type (until the next received packet).
398 void dev_add_pack(struct packet_type
*pt
)
400 struct list_head
*head
= ptype_head(pt
);
402 spin_lock(&ptype_lock
);
403 list_add_rcu(&pt
->list
, head
);
404 spin_unlock(&ptype_lock
);
406 EXPORT_SYMBOL(dev_add_pack
);
409 * __dev_remove_pack - remove packet handler
410 * @pt: packet type declaration
412 * Remove a protocol handler that was previously added to the kernel
413 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
414 * from the kernel lists and can be freed or reused once this function
417 * The packet type might still be in use by receivers
418 * and must not be freed until after all the CPU's have gone
419 * through a quiescent state.
421 void __dev_remove_pack(struct packet_type
*pt
)
423 struct list_head
*head
= ptype_head(pt
);
424 struct packet_type
*pt1
;
426 spin_lock(&ptype_lock
);
428 list_for_each_entry(pt1
, head
, list
) {
430 list_del_rcu(&pt
->list
);
435 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
437 spin_unlock(&ptype_lock
);
439 EXPORT_SYMBOL(__dev_remove_pack
);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
450 * This call sleeps to guarantee that no CPU is looking at the packet
453 void dev_remove_pack(struct packet_type
*pt
)
455 __dev_remove_pack(pt
);
459 EXPORT_SYMBOL(dev_remove_pack
);
461 /******************************************************************************
463 Device Boot-time Settings Routines
465 *******************************************************************************/
467 /* Boot time configuration table */
468 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
471 * netdev_boot_setup_add - add new setup entry
472 * @name: name of the device
473 * @map: configured settings for the device
475 * Adds new setup entry to the dev_boot_setup list. The function
476 * returns 0 on error and 1 on success. This is a generic routine to
479 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
481 struct netdev_boot_setup
*s
;
485 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
486 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
487 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
488 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
489 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
494 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
498 * netdev_boot_setup_check - check boot time settings
499 * @dev: the netdevice
501 * Check boot time settings for the device.
502 * The found settings are set for the device to be used
503 * later in the device probing.
504 * Returns 0 if no settings found, 1 if they are.
506 int netdev_boot_setup_check(struct net_device
*dev
)
508 struct netdev_boot_setup
*s
= dev_boot_setup
;
511 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
512 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
513 !strcmp(dev
->name
, s
[i
].name
)) {
514 dev
->irq
= s
[i
].map
.irq
;
515 dev
->base_addr
= s
[i
].map
.base_addr
;
516 dev
->mem_start
= s
[i
].map
.mem_start
;
517 dev
->mem_end
= s
[i
].map
.mem_end
;
523 EXPORT_SYMBOL(netdev_boot_setup_check
);
527 * netdev_boot_base - get address from boot time settings
528 * @prefix: prefix for network device
529 * @unit: id for network device
531 * Check boot time settings for the base address of device.
532 * The found settings are set for the device to be used
533 * later in the device probing.
534 * Returns 0 if no settings found.
536 unsigned long netdev_boot_base(const char *prefix
, int unit
)
538 const struct netdev_boot_setup
*s
= dev_boot_setup
;
542 sprintf(name
, "%s%d", prefix
, unit
);
545 * If device already registered then return base of 1
546 * to indicate not to probe for this interface
548 if (__dev_get_by_name(&init_net
, name
))
551 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
552 if (!strcmp(name
, s
[i
].name
))
553 return s
[i
].map
.base_addr
;
558 * Saves at boot time configured settings for any netdevice.
560 int __init
netdev_boot_setup(char *str
)
565 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
570 memset(&map
, 0, sizeof(map
));
574 map
.base_addr
= ints
[2];
576 map
.mem_start
= ints
[3];
578 map
.mem_end
= ints
[4];
580 /* Add new entry to the list */
581 return netdev_boot_setup_add(str
, &map
);
584 __setup("netdev=", netdev_boot_setup
);
586 /*******************************************************************************
588 Device Interface Subroutines
590 *******************************************************************************/
593 * __dev_get_by_name - find a device by its name
594 * @net: the applicable net namespace
595 * @name: name to find
597 * Find an interface by name. Must be called under RTNL semaphore
598 * or @dev_base_lock. If the name is found a pointer to the device
599 * is returned. If the name is not found then %NULL is returned. The
600 * reference counters are not incremented so the caller must be
601 * careful with locks.
604 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
606 struct hlist_node
*p
;
607 struct net_device
*dev
;
608 struct hlist_head
*head
= dev_name_hash(net
, name
);
610 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
611 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
616 EXPORT_SYMBOL(__dev_get_by_name
);
619 * dev_get_by_name_rcu - find a device by its name
620 * @net: the applicable net namespace
621 * @name: name to find
623 * Find an interface by name.
624 * If the name is found a pointer to the device is returned.
625 * If the name is not found then %NULL is returned.
626 * The reference counters are not incremented so the caller must be
627 * careful with locks. The caller must hold RCU lock.
630 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
632 struct hlist_node
*p
;
633 struct net_device
*dev
;
634 struct hlist_head
*head
= dev_name_hash(net
, name
);
636 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
637 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
642 EXPORT_SYMBOL(dev_get_by_name_rcu
);
645 * dev_get_by_name - find a device by its name
646 * @net: the applicable net namespace
647 * @name: name to find
649 * Find an interface by name. This can be called from any
650 * context and does its own locking. The returned handle has
651 * the usage count incremented and the caller must use dev_put() to
652 * release it when it is no longer needed. %NULL is returned if no
653 * matching device is found.
656 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
658 struct net_device
*dev
;
661 dev
= dev_get_by_name_rcu(net
, name
);
667 EXPORT_SYMBOL(dev_get_by_name
);
670 * __dev_get_by_index - find a device by its ifindex
671 * @net: the applicable net namespace
672 * @ifindex: index of device
674 * Search for an interface by index. Returns %NULL if the device
675 * is not found or a pointer to the device. The device has not
676 * had its reference counter increased so the caller must be careful
677 * about locking. The caller must hold either the RTNL semaphore
681 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
683 struct hlist_node
*p
;
684 struct net_device
*dev
;
685 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
687 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
688 if (dev
->ifindex
== ifindex
)
693 EXPORT_SYMBOL(__dev_get_by_index
);
696 * dev_get_by_index_rcu - find a device by its ifindex
697 * @net: the applicable net namespace
698 * @ifindex: index of device
700 * Search for an interface by index. Returns %NULL if the device
701 * is not found or a pointer to the device. The device has not
702 * had its reference counter increased so the caller must be careful
703 * about locking. The caller must hold RCU lock.
706 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
708 struct hlist_node
*p
;
709 struct net_device
*dev
;
710 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
712 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
713 if (dev
->ifindex
== ifindex
)
718 EXPORT_SYMBOL(dev_get_by_index_rcu
);
722 * dev_get_by_index - find a device by its ifindex
723 * @net: the applicable net namespace
724 * @ifindex: index of device
726 * Search for an interface by index. Returns NULL if the device
727 * is not found or a pointer to the device. The device returned has
728 * had a reference added and the pointer is safe until the user calls
729 * dev_put to indicate they have finished with it.
732 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
734 struct net_device
*dev
;
737 dev
= dev_get_by_index_rcu(net
, ifindex
);
743 EXPORT_SYMBOL(dev_get_by_index
);
746 * dev_getbyhwaddr_rcu - find a device by its hardware address
747 * @net: the applicable net namespace
748 * @type: media type of device
749 * @ha: hardware address
751 * Search for an interface by MAC address. Returns NULL if the device
752 * is not found or a pointer to the device. The caller must hold RCU
753 * The returned device has not had its ref count increased
754 * and the caller must therefore be careful about locking
758 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
761 struct net_device
*dev
;
763 for_each_netdev_rcu(net
, dev
)
764 if (dev
->type
== type
&&
765 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
770 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
772 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
774 struct net_device
*dev
;
777 for_each_netdev(net
, dev
)
778 if (dev
->type
== type
)
783 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
785 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
787 struct net_device
*dev
, *ret
= NULL
;
790 for_each_netdev_rcu(net
, dev
)
791 if (dev
->type
== type
) {
799 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
802 * dev_get_by_flags_rcu - find any device with given flags
803 * @net: the applicable net namespace
804 * @if_flags: IFF_* values
805 * @mask: bitmask of bits in if_flags to check
807 * Search for any interface with the given flags. Returns NULL if a device
808 * is not found or a pointer to the device. Must be called inside
809 * rcu_read_lock(), and result refcount is unchanged.
812 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
815 struct net_device
*dev
, *ret
;
818 for_each_netdev_rcu(net
, dev
) {
819 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
826 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
829 * dev_valid_name - check if name is okay for network device
832 * Network device names need to be valid file names to
833 * to allow sysfs to work. We also disallow any kind of
836 int dev_valid_name(const char *name
)
840 if (strlen(name
) >= IFNAMSIZ
)
842 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
846 if (*name
== '/' || isspace(*name
))
852 EXPORT_SYMBOL(dev_valid_name
);
855 * __dev_alloc_name - allocate a name for a device
856 * @net: network namespace to allocate the device name in
857 * @name: name format string
858 * @buf: scratch buffer and result name string
860 * Passed a format string - eg "lt%d" it will try and find a suitable
861 * id. It scans list of devices to build up a free map, then chooses
862 * the first empty slot. The caller must hold the dev_base or rtnl lock
863 * while allocating the name and adding the device in order to avoid
865 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
866 * Returns the number of the unit assigned or a negative errno code.
869 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
873 const int max_netdevices
= 8*PAGE_SIZE
;
874 unsigned long *inuse
;
875 struct net_device
*d
;
877 p
= strnchr(name
, IFNAMSIZ
-1, '%');
880 * Verify the string as this thing may have come from
881 * the user. There must be either one "%d" and no other "%"
884 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
887 /* Use one page as a bit array of possible slots */
888 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
892 for_each_netdev(net
, d
) {
893 if (!sscanf(d
->name
, name
, &i
))
895 if (i
< 0 || i
>= max_netdevices
)
898 /* avoid cases where sscanf is not exact inverse of printf */
899 snprintf(buf
, IFNAMSIZ
, name
, i
);
900 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
904 i
= find_first_zero_bit(inuse
, max_netdevices
);
905 free_page((unsigned long) inuse
);
909 snprintf(buf
, IFNAMSIZ
, name
, i
);
910 if (!__dev_get_by_name(net
, buf
))
913 /* It is possible to run out of possible slots
914 * when the name is long and there isn't enough space left
915 * for the digits, or if all bits are used.
921 * dev_alloc_name - allocate a name for a device
923 * @name: name format string
925 * Passed a format string - eg "lt%d" it will try and find a suitable
926 * id. It scans list of devices to build up a free map, then chooses
927 * the first empty slot. The caller must hold the dev_base or rtnl lock
928 * while allocating the name and adding the device in order to avoid
930 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
931 * Returns the number of the unit assigned or a negative errno code.
934 int dev_alloc_name(struct net_device
*dev
, const char *name
)
940 BUG_ON(!dev_net(dev
));
942 ret
= __dev_alloc_name(net
, name
, buf
);
944 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
947 EXPORT_SYMBOL(dev_alloc_name
);
949 static int dev_get_valid_name(struct net_device
*dev
, const char *name
, bool fmt
)
953 BUG_ON(!dev_net(dev
));
956 if (!dev_valid_name(name
))
959 if (fmt
&& strchr(name
, '%'))
960 return dev_alloc_name(dev
, name
);
961 else if (__dev_get_by_name(net
, name
))
963 else if (dev
->name
!= name
)
964 strlcpy(dev
->name
, name
, IFNAMSIZ
);
970 * dev_change_name - change name of a device
972 * @newname: name (or format string) must be at least IFNAMSIZ
974 * Change name of a device, can pass format strings "eth%d".
977 int dev_change_name(struct net_device
*dev
, const char *newname
)
979 char oldname
[IFNAMSIZ
];
985 BUG_ON(!dev_net(dev
));
988 if (dev
->flags
& IFF_UP
)
991 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
994 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
996 err
= dev_get_valid_name(dev
, newname
, 1);
1001 ret
= device_rename(&dev
->dev
, dev
->name
);
1003 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1007 write_lock_bh(&dev_base_lock
);
1008 hlist_del(&dev
->name_hlist
);
1009 write_unlock_bh(&dev_base_lock
);
1013 write_lock_bh(&dev_base_lock
);
1014 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1015 write_unlock_bh(&dev_base_lock
);
1017 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1018 ret
= notifier_to_errno(ret
);
1021 /* err >= 0 after dev_alloc_name() or stores the first errno */
1024 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1028 "%s: name change rollback failed: %d.\n",
1037 * dev_set_alias - change ifalias of a device
1039 * @alias: name up to IFALIASZ
1040 * @len: limit of bytes to copy from info
1042 * Set ifalias for a device,
1044 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1048 if (len
>= IFALIASZ
)
1053 kfree(dev
->ifalias
);
1054 dev
->ifalias
= NULL
;
1059 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1063 strlcpy(dev
->ifalias
, alias
, len
+1);
1069 * netdev_features_change - device changes features
1070 * @dev: device to cause notification
1072 * Called to indicate a device has changed features.
1074 void netdev_features_change(struct net_device
*dev
)
1076 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1078 EXPORT_SYMBOL(netdev_features_change
);
1081 * netdev_state_change - device changes state
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed state. This function calls
1085 * the notifier chains for netdev_chain and sends a NEWLINK message
1086 * to the routing socket.
1088 void netdev_state_change(struct net_device
*dev
)
1090 if (dev
->flags
& IFF_UP
) {
1091 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1092 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1095 EXPORT_SYMBOL(netdev_state_change
);
1097 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1099 return call_netdevice_notifiers(event
, dev
);
1101 EXPORT_SYMBOL(netdev_bonding_change
);
1104 * dev_load - load a network module
1105 * @net: the applicable net namespace
1106 * @name: name of interface
1108 * If a network interface is not present and the process has suitable
1109 * privileges this function loads the module. If module loading is not
1110 * available in this kernel then it becomes a nop.
1113 void dev_load(struct net
*net
, const char *name
)
1115 struct net_device
*dev
;
1118 dev
= dev_get_by_name_rcu(net
, name
);
1121 if (!dev
&& capable(CAP_NET_ADMIN
))
1122 request_module("%s", name
);
1124 EXPORT_SYMBOL(dev_load
);
1126 static int __dev_open(struct net_device
*dev
)
1128 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1134 * Is it even present?
1136 if (!netif_device_present(dev
))
1139 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1140 ret
= notifier_to_errno(ret
);
1145 * Call device private open method
1147 set_bit(__LINK_STATE_START
, &dev
->state
);
1149 if (ops
->ndo_validate_addr
)
1150 ret
= ops
->ndo_validate_addr(dev
);
1152 if (!ret
&& ops
->ndo_open
)
1153 ret
= ops
->ndo_open(dev
);
1156 * If it went open OK then:
1160 clear_bit(__LINK_STATE_START
, &dev
->state
);
1165 dev
->flags
|= IFF_UP
;
1170 net_dmaengine_get();
1173 * Initialize multicasting status
1175 dev_set_rx_mode(dev
);
1178 * Wakeup transmit queue engine
1187 * dev_open - prepare an interface for use.
1188 * @dev: device to open
1190 * Takes a device from down to up state. The device's private open
1191 * function is invoked and then the multicast lists are loaded. Finally
1192 * the device is moved into the up state and a %NETDEV_UP message is
1193 * sent to the netdev notifier chain.
1195 * Calling this function on an active interface is a nop. On a failure
1196 * a negative errno code is returned.
1198 int dev_open(struct net_device
*dev
)
1205 if (dev
->flags
& IFF_UP
)
1211 ret
= __dev_open(dev
);
1216 * ... and announce new interface.
1218 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1219 call_netdevice_notifiers(NETDEV_UP
, dev
);
1223 EXPORT_SYMBOL(dev_open
);
1225 static int __dev_close_many(struct list_head
*head
)
1227 struct net_device
*dev
;
1232 list_for_each_entry(dev
, head
, unreg_list
) {
1234 * Tell people we are going down, so that they can
1235 * prepare to death, when device is still operating.
1237 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1239 clear_bit(__LINK_STATE_START
, &dev
->state
);
1241 /* Synchronize to scheduled poll. We cannot touch poll list, it
1242 * can be even on different cpu. So just clear netif_running().
1244 * dev->stop() will invoke napi_disable() on all of it's
1245 * napi_struct instances on this device.
1247 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1250 dev_deactivate_many(head
);
1252 list_for_each_entry(dev
, head
, unreg_list
) {
1253 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1256 * Call the device specific close. This cannot fail.
1257 * Only if device is UP
1259 * We allow it to be called even after a DETACH hot-plug
1266 * Device is now down.
1269 dev
->flags
&= ~IFF_UP
;
1274 net_dmaengine_put();
1280 static int __dev_close(struct net_device
*dev
)
1284 list_add(&dev
->unreg_list
, &single
);
1285 return __dev_close_many(&single
);
1288 int dev_close_many(struct list_head
*head
)
1290 struct net_device
*dev
, *tmp
;
1291 LIST_HEAD(tmp_list
);
1293 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1294 if (!(dev
->flags
& IFF_UP
))
1295 list_move(&dev
->unreg_list
, &tmp_list
);
1297 __dev_close_many(head
);
1300 * Tell people we are down
1302 list_for_each_entry(dev
, head
, unreg_list
) {
1303 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1304 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1307 /* rollback_registered_many needs the complete original list */
1308 list_splice(&tmp_list
, head
);
1313 * dev_close - shutdown an interface.
1314 * @dev: device to shutdown
1316 * This function moves an active device into down state. A
1317 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1318 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1321 int dev_close(struct net_device
*dev
)
1325 list_add(&dev
->unreg_list
, &single
);
1326 dev_close_many(&single
);
1330 EXPORT_SYMBOL(dev_close
);
1334 * dev_disable_lro - disable Large Receive Offload on a device
1337 * Disable Large Receive Offload (LRO) on a net device. Must be
1338 * called under RTNL. This is needed if received packets may be
1339 * forwarded to another interface.
1341 void dev_disable_lro(struct net_device
*dev
)
1343 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1344 dev
->ethtool_ops
->set_flags
) {
1345 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1346 if (flags
& ETH_FLAG_LRO
) {
1347 flags
&= ~ETH_FLAG_LRO
;
1348 dev
->ethtool_ops
->set_flags(dev
, flags
);
1351 WARN_ON(dev
->features
& NETIF_F_LRO
);
1353 EXPORT_SYMBOL(dev_disable_lro
);
1356 static int dev_boot_phase
= 1;
1359 * Device change register/unregister. These are not inline or static
1360 * as we export them to the world.
1364 * register_netdevice_notifier - register a network notifier block
1367 * Register a notifier to be called when network device events occur.
1368 * The notifier passed is linked into the kernel structures and must
1369 * not be reused until it has been unregistered. A negative errno code
1370 * is returned on a failure.
1372 * When registered all registration and up events are replayed
1373 * to the new notifier to allow device to have a race free
1374 * view of the network device list.
1377 int register_netdevice_notifier(struct notifier_block
*nb
)
1379 struct net_device
*dev
;
1380 struct net_device
*last
;
1385 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1391 for_each_netdev(net
, dev
) {
1392 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1393 err
= notifier_to_errno(err
);
1397 if (!(dev
->flags
& IFF_UP
))
1400 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1411 for_each_netdev(net
, dev
) {
1415 if (dev
->flags
& IFF_UP
) {
1416 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1417 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1419 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1420 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1424 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1427 EXPORT_SYMBOL(register_netdevice_notifier
);
1430 * unregister_netdevice_notifier - unregister a network notifier block
1433 * Unregister a notifier previously registered by
1434 * register_netdevice_notifier(). The notifier is unlinked into the
1435 * kernel structures and may then be reused. A negative errno code
1436 * is returned on a failure.
1439 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1444 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1448 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1451 * call_netdevice_notifiers - call all network notifier blocks
1452 * @val: value passed unmodified to notifier function
1453 * @dev: net_device pointer passed unmodified to notifier function
1455 * Call all network notifier blocks. Parameters and return value
1456 * are as for raw_notifier_call_chain().
1459 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1462 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1465 /* When > 0 there are consumers of rx skb time stamps */
1466 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1468 void net_enable_timestamp(void)
1470 atomic_inc(&netstamp_needed
);
1472 EXPORT_SYMBOL(net_enable_timestamp
);
1474 void net_disable_timestamp(void)
1476 atomic_dec(&netstamp_needed
);
1478 EXPORT_SYMBOL(net_disable_timestamp
);
1480 static inline void net_timestamp_set(struct sk_buff
*skb
)
1482 if (atomic_read(&netstamp_needed
))
1483 __net_timestamp(skb
);
1485 skb
->tstamp
.tv64
= 0;
1488 static inline void net_timestamp_check(struct sk_buff
*skb
)
1490 if (!skb
->tstamp
.tv64
&& atomic_read(&netstamp_needed
))
1491 __net_timestamp(skb
);
1495 * dev_forward_skb - loopback an skb to another netif
1497 * @dev: destination network device
1498 * @skb: buffer to forward
1501 * NET_RX_SUCCESS (no congestion)
1502 * NET_RX_DROP (packet was dropped, but freed)
1504 * dev_forward_skb can be used for injecting an skb from the
1505 * start_xmit function of one device into the receive queue
1506 * of another device.
1508 * The receiving device may be in another namespace, so
1509 * we have to clear all information in the skb that could
1510 * impact namespace isolation.
1512 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1517 if (unlikely(!(dev
->flags
& IFF_UP
) ||
1518 (skb
->len
> (dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
)))) {
1519 atomic_long_inc(&dev
->rx_dropped
);
1523 skb_set_dev(skb
, dev
);
1524 skb
->tstamp
.tv64
= 0;
1525 skb
->pkt_type
= PACKET_HOST
;
1526 skb
->protocol
= eth_type_trans(skb
, dev
);
1527 return netif_rx(skb
);
1529 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1531 static inline int deliver_skb(struct sk_buff
*skb
,
1532 struct packet_type
*pt_prev
,
1533 struct net_device
*orig_dev
)
1535 atomic_inc(&skb
->users
);
1536 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1540 * Support routine. Sends outgoing frames to any network
1541 * taps currently in use.
1544 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1546 struct packet_type
*ptype
;
1547 struct sk_buff
*skb2
= NULL
;
1548 struct packet_type
*pt_prev
= NULL
;
1551 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1552 /* Never send packets back to the socket
1553 * they originated from - MvS (miquels@drinkel.ow.org)
1555 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1556 (ptype
->af_packet_priv
== NULL
||
1557 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1559 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1564 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1568 net_timestamp_set(skb2
);
1570 /* skb->nh should be correctly
1571 set by sender, so that the second statement is
1572 just protection against buggy protocols.
1574 skb_reset_mac_header(skb2
);
1576 if (skb_network_header(skb2
) < skb2
->data
||
1577 skb2
->network_header
> skb2
->tail
) {
1578 if (net_ratelimit())
1579 printk(KERN_CRIT
"protocol %04x is "
1581 ntohs(skb2
->protocol
),
1583 skb_reset_network_header(skb2
);
1586 skb2
->transport_header
= skb2
->network_header
;
1587 skb2
->pkt_type
= PACKET_OUTGOING
;
1592 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1597 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1598 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1600 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1604 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1607 if (dev
->reg_state
== NETREG_REGISTERED
) {
1610 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1615 if (txq
< dev
->real_num_tx_queues
)
1616 qdisc_reset_all_tx_gt(dev
, txq
);
1619 dev
->real_num_tx_queues
= txq
;
1622 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1626 * netif_set_real_num_rx_queues - set actual number of RX queues used
1627 * @dev: Network device
1628 * @rxq: Actual number of RX queues
1630 * This must be called either with the rtnl_lock held or before
1631 * registration of the net device. Returns 0 on success, or a
1632 * negative error code. If called before registration, it always
1635 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1639 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1642 if (dev
->reg_state
== NETREG_REGISTERED
) {
1645 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1651 dev
->real_num_rx_queues
= rxq
;
1654 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1657 static inline void __netif_reschedule(struct Qdisc
*q
)
1659 struct softnet_data
*sd
;
1660 unsigned long flags
;
1662 local_irq_save(flags
);
1663 sd
= &__get_cpu_var(softnet_data
);
1664 q
->next_sched
= NULL
;
1665 *sd
->output_queue_tailp
= q
;
1666 sd
->output_queue_tailp
= &q
->next_sched
;
1667 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1668 local_irq_restore(flags
);
1671 void __netif_schedule(struct Qdisc
*q
)
1673 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1674 __netif_reschedule(q
);
1676 EXPORT_SYMBOL(__netif_schedule
);
1678 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1680 if (atomic_dec_and_test(&skb
->users
)) {
1681 struct softnet_data
*sd
;
1682 unsigned long flags
;
1684 local_irq_save(flags
);
1685 sd
= &__get_cpu_var(softnet_data
);
1686 skb
->next
= sd
->completion_queue
;
1687 sd
->completion_queue
= skb
;
1688 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1689 local_irq_restore(flags
);
1692 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1694 void dev_kfree_skb_any(struct sk_buff
*skb
)
1696 if (in_irq() || irqs_disabled())
1697 dev_kfree_skb_irq(skb
);
1701 EXPORT_SYMBOL(dev_kfree_skb_any
);
1705 * netif_device_detach - mark device as removed
1706 * @dev: network device
1708 * Mark device as removed from system and therefore no longer available.
1710 void netif_device_detach(struct net_device
*dev
)
1712 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1713 netif_running(dev
)) {
1714 netif_tx_stop_all_queues(dev
);
1717 EXPORT_SYMBOL(netif_device_detach
);
1720 * netif_device_attach - mark device as attached
1721 * @dev: network device
1723 * Mark device as attached from system and restart if needed.
1725 void netif_device_attach(struct net_device
*dev
)
1727 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1728 netif_running(dev
)) {
1729 netif_tx_wake_all_queues(dev
);
1730 __netdev_watchdog_up(dev
);
1733 EXPORT_SYMBOL(netif_device_attach
);
1735 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1737 return ((features
& NETIF_F_GEN_CSUM
) ||
1738 ((features
& NETIF_F_V4_CSUM
) &&
1739 protocol
== htons(ETH_P_IP
)) ||
1740 ((features
& NETIF_F_V6_CSUM
) &&
1741 protocol
== htons(ETH_P_IPV6
)) ||
1742 ((features
& NETIF_F_FCOE_CRC
) &&
1743 protocol
== htons(ETH_P_FCOE
)));
1746 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1748 __be16 protocol
= skb
->protocol
;
1749 int features
= dev
->features
;
1751 if (vlan_tx_tag_present(skb
)) {
1752 features
&= dev
->vlan_features
;
1753 } else if (protocol
== htons(ETH_P_8021Q
)) {
1754 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1755 protocol
= veh
->h_vlan_encapsulated_proto
;
1756 features
&= dev
->vlan_features
;
1759 return can_checksum_protocol(features
, protocol
);
1763 * skb_dev_set -- assign a new device to a buffer
1764 * @skb: buffer for the new device
1765 * @dev: network device
1767 * If an skb is owned by a device already, we have to reset
1768 * all data private to the namespace a device belongs to
1769 * before assigning it a new device.
1771 #ifdef CONFIG_NET_NS
1772 void skb_set_dev(struct sk_buff
*skb
, struct net_device
*dev
)
1775 if (skb
->dev
&& !net_eq(dev_net(skb
->dev
), dev_net(dev
))) {
1778 skb_init_secmark(skb
);
1782 skb
->ipvs_property
= 0;
1783 #ifdef CONFIG_NET_SCHED
1789 EXPORT_SYMBOL(skb_set_dev
);
1790 #endif /* CONFIG_NET_NS */
1793 * Invalidate hardware checksum when packet is to be mangled, and
1794 * complete checksum manually on outgoing path.
1796 int skb_checksum_help(struct sk_buff
*skb
)
1799 int ret
= 0, offset
;
1801 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1802 goto out_set_summed
;
1804 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1805 /* Let GSO fix up the checksum. */
1806 goto out_set_summed
;
1809 offset
= skb_checksum_start_offset(skb
);
1810 BUG_ON(offset
>= skb_headlen(skb
));
1811 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1813 offset
+= skb
->csum_offset
;
1814 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1816 if (skb_cloned(skb
) &&
1817 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1818 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1823 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1825 skb
->ip_summed
= CHECKSUM_NONE
;
1829 EXPORT_SYMBOL(skb_checksum_help
);
1832 * skb_gso_segment - Perform segmentation on skb.
1833 * @skb: buffer to segment
1834 * @features: features for the output path (see dev->features)
1836 * This function segments the given skb and returns a list of segments.
1838 * It may return NULL if the skb requires no segmentation. This is
1839 * only possible when GSO is used for verifying header integrity.
1841 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1843 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1844 struct packet_type
*ptype
;
1845 __be16 type
= skb
->protocol
;
1846 int vlan_depth
= ETH_HLEN
;
1849 while (type
== htons(ETH_P_8021Q
)) {
1850 struct vlan_hdr
*vh
;
1852 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1853 return ERR_PTR(-EINVAL
);
1855 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1856 type
= vh
->h_vlan_encapsulated_proto
;
1857 vlan_depth
+= VLAN_HLEN
;
1860 skb_reset_mac_header(skb
);
1861 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1862 __skb_pull(skb
, skb
->mac_len
);
1864 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1865 struct net_device
*dev
= skb
->dev
;
1866 struct ethtool_drvinfo info
= {};
1868 if (dev
&& dev
->ethtool_ops
&& dev
->ethtool_ops
->get_drvinfo
)
1869 dev
->ethtool_ops
->get_drvinfo(dev
, &info
);
1871 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1872 info
.driver
, dev
? dev
->features
: 0L,
1873 skb
->sk
? skb
->sk
->sk_route_caps
: 0L,
1874 skb
->len
, skb
->data_len
, skb
->ip_summed
);
1876 if (skb_header_cloned(skb
) &&
1877 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1878 return ERR_PTR(err
);
1882 list_for_each_entry_rcu(ptype
,
1883 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1884 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1885 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1886 err
= ptype
->gso_send_check(skb
);
1887 segs
= ERR_PTR(err
);
1888 if (err
|| skb_gso_ok(skb
, features
))
1890 __skb_push(skb
, (skb
->data
-
1891 skb_network_header(skb
)));
1893 segs
= ptype
->gso_segment(skb
, features
);
1899 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1903 EXPORT_SYMBOL(skb_gso_segment
);
1905 /* Take action when hardware reception checksum errors are detected. */
1907 void netdev_rx_csum_fault(struct net_device
*dev
)
1909 if (net_ratelimit()) {
1910 printk(KERN_ERR
"%s: hw csum failure.\n",
1911 dev
? dev
->name
: "<unknown>");
1915 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1918 /* Actually, we should eliminate this check as soon as we know, that:
1919 * 1. IOMMU is present and allows to map all the memory.
1920 * 2. No high memory really exists on this machine.
1923 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1925 #ifdef CONFIG_HIGHMEM
1927 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
1928 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1929 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1933 if (PCI_DMA_BUS_IS_PHYS
) {
1934 struct device
*pdev
= dev
->dev
.parent
;
1938 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1939 dma_addr_t addr
= page_to_phys(skb_shinfo(skb
)->frags
[i
].page
);
1940 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
1949 void (*destructor
)(struct sk_buff
*skb
);
1952 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1954 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1956 struct dev_gso_cb
*cb
;
1959 struct sk_buff
*nskb
= skb
->next
;
1961 skb
->next
= nskb
->next
;
1964 } while (skb
->next
);
1966 cb
= DEV_GSO_CB(skb
);
1968 cb
->destructor(skb
);
1972 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1973 * @skb: buffer to segment
1974 * @features: device features as applicable to this skb
1976 * This function segments the given skb and stores the list of segments
1979 static int dev_gso_segment(struct sk_buff
*skb
, int features
)
1981 struct sk_buff
*segs
;
1983 segs
= skb_gso_segment(skb
, features
);
1985 /* Verifying header integrity only. */
1990 return PTR_ERR(segs
);
1993 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1994 skb
->destructor
= dev_gso_skb_destructor
;
2000 * Try to orphan skb early, right before transmission by the device.
2001 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2002 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2004 static inline void skb_orphan_try(struct sk_buff
*skb
)
2006 struct sock
*sk
= skb
->sk
;
2008 if (sk
&& !skb_shinfo(skb
)->tx_flags
) {
2009 /* skb_tx_hash() wont be able to get sk.
2010 * We copy sk_hash into skb->rxhash
2013 skb
->rxhash
= sk
->sk_hash
;
2018 static int harmonize_features(struct sk_buff
*skb
, __be16 protocol
, int features
)
2020 if (!can_checksum_protocol(protocol
, features
)) {
2021 features
&= ~NETIF_F_ALL_CSUM
;
2022 features
&= ~NETIF_F_SG
;
2023 } else if (illegal_highdma(skb
->dev
, skb
)) {
2024 features
&= ~NETIF_F_SG
;
2030 int netif_skb_features(struct sk_buff
*skb
)
2032 __be16 protocol
= skb
->protocol
;
2033 int features
= skb
->dev
->features
;
2035 if (protocol
== htons(ETH_P_8021Q
)) {
2036 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2037 protocol
= veh
->h_vlan_encapsulated_proto
;
2038 } else if (!vlan_tx_tag_present(skb
)) {
2039 return harmonize_features(skb
, protocol
, features
);
2042 features
&= skb
->dev
->vlan_features
;
2044 if (protocol
!= htons(ETH_P_8021Q
)) {
2045 return harmonize_features(skb
, protocol
, features
);
2047 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2049 return harmonize_features(skb
, protocol
, features
);
2052 EXPORT_SYMBOL(netif_skb_features
);
2055 * Returns true if either:
2056 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2057 * 2. skb is fragmented and the device does not support SG, or if
2058 * at least one of fragments is in highmem and device does not
2059 * support DMA from it.
2061 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2062 struct net_device
*dev
)
2064 if (skb_is_nonlinear(skb
)) {
2065 int features
= dev
->features
;
2067 if (vlan_tx_tag_present(skb
))
2068 features
&= dev
->vlan_features
;
2070 return (skb_has_frag_list(skb
) &&
2071 !(features
& NETIF_F_FRAGLIST
)) ||
2072 (skb_shinfo(skb
)->nr_frags
&&
2073 (!(features
& NETIF_F_SG
) ||
2074 illegal_highdma(dev
, skb
)));
2080 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2081 struct netdev_queue
*txq
)
2083 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2084 int rc
= NETDEV_TX_OK
;
2086 if (likely(!skb
->next
)) {
2090 * If device doesnt need skb->dst, release it right now while
2091 * its hot in this cpu cache
2093 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2096 if (!list_empty(&ptype_all
))
2097 dev_queue_xmit_nit(skb
, dev
);
2099 skb_orphan_try(skb
);
2101 features
= netif_skb_features(skb
);
2103 if (vlan_tx_tag_present(skb
) &&
2104 !(features
& NETIF_F_HW_VLAN_TX
)) {
2105 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2112 if (netif_needs_gso(skb
, features
)) {
2113 if (unlikely(dev_gso_segment(skb
, features
)))
2118 if (skb_needs_linearize(skb
, dev
) &&
2119 __skb_linearize(skb
))
2122 /* If packet is not checksummed and device does not
2123 * support checksumming for this protocol, complete
2124 * checksumming here.
2126 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2127 skb_set_transport_header(skb
,
2128 skb_checksum_start_offset(skb
));
2129 if (!dev_can_checksum(dev
, skb
) &&
2130 skb_checksum_help(skb
))
2135 rc
= ops
->ndo_start_xmit(skb
, dev
);
2136 trace_net_dev_xmit(skb
, rc
);
2137 if (rc
== NETDEV_TX_OK
)
2138 txq_trans_update(txq
);
2144 struct sk_buff
*nskb
= skb
->next
;
2146 skb
->next
= nskb
->next
;
2150 * If device doesnt need nskb->dst, release it right now while
2151 * its hot in this cpu cache
2153 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2156 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2157 trace_net_dev_xmit(nskb
, rc
);
2158 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2159 if (rc
& ~NETDEV_TX_MASK
)
2160 goto out_kfree_gso_skb
;
2161 nskb
->next
= skb
->next
;
2165 txq_trans_update(txq
);
2166 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
2167 return NETDEV_TX_BUSY
;
2168 } while (skb
->next
);
2171 if (likely(skb
->next
== NULL
))
2172 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2179 static u32 hashrnd __read_mostly
;
2182 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2183 * to be used as a distribution range.
2185 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2186 unsigned int num_tx_queues
)
2190 if (skb_rx_queue_recorded(skb
)) {
2191 hash
= skb_get_rx_queue(skb
);
2192 while (unlikely(hash
>= num_tx_queues
))
2193 hash
-= num_tx_queues
;
2197 if (skb
->sk
&& skb
->sk
->sk_hash
)
2198 hash
= skb
->sk
->sk_hash
;
2200 hash
= (__force u16
) skb
->protocol
^ skb
->rxhash
;
2201 hash
= jhash_1word(hash
, hashrnd
);
2203 return (u16
) (((u64
) hash
* num_tx_queues
) >> 32);
2205 EXPORT_SYMBOL(__skb_tx_hash
);
2207 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2209 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2210 if (net_ratelimit()) {
2211 pr_warning("%s selects TX queue %d, but "
2212 "real number of TX queues is %d\n",
2213 dev
->name
, queue_index
, dev
->real_num_tx_queues
);
2220 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2223 struct xps_dev_maps
*dev_maps
;
2224 struct xps_map
*map
;
2225 int queue_index
= -1;
2228 dev_maps
= rcu_dereference(dev
->xps_maps
);
2230 map
= rcu_dereference(
2231 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2234 queue_index
= map
->queues
[0];
2237 if (skb
->sk
&& skb
->sk
->sk_hash
)
2238 hash
= skb
->sk
->sk_hash
;
2240 hash
= (__force u16
) skb
->protocol
^
2242 hash
= jhash_1word(hash
, hashrnd
);
2243 queue_index
= map
->queues
[
2244 ((u64
)hash
* map
->len
) >> 32];
2246 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2258 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2259 struct sk_buff
*skb
)
2262 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2264 if (dev
->real_num_tx_queues
== 1)
2266 else if (ops
->ndo_select_queue
) {
2267 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2268 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2270 struct sock
*sk
= skb
->sk
;
2271 queue_index
= sk_tx_queue_get(sk
);
2273 if (queue_index
< 0 || skb
->ooo_okay
||
2274 queue_index
>= dev
->real_num_tx_queues
) {
2275 int old_index
= queue_index
;
2277 queue_index
= get_xps_queue(dev
, skb
);
2278 if (queue_index
< 0)
2279 queue_index
= skb_tx_hash(dev
, skb
);
2281 if (queue_index
!= old_index
&& sk
) {
2282 struct dst_entry
*dst
=
2283 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2285 if (dst
&& skb_dst(skb
) == dst
)
2286 sk_tx_queue_set(sk
, queue_index
);
2291 skb_set_queue_mapping(skb
, queue_index
);
2292 return netdev_get_tx_queue(dev
, queue_index
);
2295 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2296 struct net_device
*dev
,
2297 struct netdev_queue
*txq
)
2299 spinlock_t
*root_lock
= qdisc_lock(q
);
2300 bool contended
= qdisc_is_running(q
);
2304 * Heuristic to force contended enqueues to serialize on a
2305 * separate lock before trying to get qdisc main lock.
2306 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2307 * and dequeue packets faster.
2309 if (unlikely(contended
))
2310 spin_lock(&q
->busylock
);
2312 spin_lock(root_lock
);
2313 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2316 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2317 qdisc_run_begin(q
)) {
2319 * This is a work-conserving queue; there are no old skbs
2320 * waiting to be sent out; and the qdisc is not running -
2321 * xmit the skb directly.
2323 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2325 __qdisc_update_bstats(q
, skb
->len
);
2326 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2327 if (unlikely(contended
)) {
2328 spin_unlock(&q
->busylock
);
2335 rc
= NET_XMIT_SUCCESS
;
2338 rc
= qdisc_enqueue_root(skb
, q
);
2339 if (qdisc_run_begin(q
)) {
2340 if (unlikely(contended
)) {
2341 spin_unlock(&q
->busylock
);
2347 spin_unlock(root_lock
);
2348 if (unlikely(contended
))
2349 spin_unlock(&q
->busylock
);
2353 static DEFINE_PER_CPU(int, xmit_recursion
);
2354 #define RECURSION_LIMIT 10
2357 * dev_queue_xmit - transmit a buffer
2358 * @skb: buffer to transmit
2360 * Queue a buffer for transmission to a network device. The caller must
2361 * have set the device and priority and built the buffer before calling
2362 * this function. The function can be called from an interrupt.
2364 * A negative errno code is returned on a failure. A success does not
2365 * guarantee the frame will be transmitted as it may be dropped due
2366 * to congestion or traffic shaping.
2368 * -----------------------------------------------------------------------------------
2369 * I notice this method can also return errors from the queue disciplines,
2370 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2373 * Regardless of the return value, the skb is consumed, so it is currently
2374 * difficult to retry a send to this method. (You can bump the ref count
2375 * before sending to hold a reference for retry if you are careful.)
2377 * When calling this method, interrupts MUST be enabled. This is because
2378 * the BH enable code must have IRQs enabled so that it will not deadlock.
2381 int dev_queue_xmit(struct sk_buff
*skb
)
2383 struct net_device
*dev
= skb
->dev
;
2384 struct netdev_queue
*txq
;
2388 /* Disable soft irqs for various locks below. Also
2389 * stops preemption for RCU.
2393 txq
= dev_pick_tx(dev
, skb
);
2394 q
= rcu_dereference_bh(txq
->qdisc
);
2396 #ifdef CONFIG_NET_CLS_ACT
2397 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2399 trace_net_dev_queue(skb
);
2401 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2405 /* The device has no queue. Common case for software devices:
2406 loopback, all the sorts of tunnels...
2408 Really, it is unlikely that netif_tx_lock protection is necessary
2409 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2411 However, it is possible, that they rely on protection
2414 Check this and shot the lock. It is not prone from deadlocks.
2415 Either shot noqueue qdisc, it is even simpler 8)
2417 if (dev
->flags
& IFF_UP
) {
2418 int cpu
= smp_processor_id(); /* ok because BHs are off */
2420 if (txq
->xmit_lock_owner
!= cpu
) {
2422 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2423 goto recursion_alert
;
2425 HARD_TX_LOCK(dev
, txq
, cpu
);
2427 if (!netif_tx_queue_stopped(txq
)) {
2428 __this_cpu_inc(xmit_recursion
);
2429 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2430 __this_cpu_dec(xmit_recursion
);
2431 if (dev_xmit_complete(rc
)) {
2432 HARD_TX_UNLOCK(dev
, txq
);
2436 HARD_TX_UNLOCK(dev
, txq
);
2437 if (net_ratelimit())
2438 printk(KERN_CRIT
"Virtual device %s asks to "
2439 "queue packet!\n", dev
->name
);
2441 /* Recursion is detected! It is possible,
2445 if (net_ratelimit())
2446 printk(KERN_CRIT
"Dead loop on virtual device "
2447 "%s, fix it urgently!\n", dev
->name
);
2452 rcu_read_unlock_bh();
2457 rcu_read_unlock_bh();
2460 EXPORT_SYMBOL(dev_queue_xmit
);
2463 /*=======================================================================
2465 =======================================================================*/
2467 int netdev_max_backlog __read_mostly
= 1000;
2468 int netdev_tstamp_prequeue __read_mostly
= 1;
2469 int netdev_budget __read_mostly
= 300;
2470 int weight_p __read_mostly
= 64; /* old backlog weight */
2472 /* Called with irq disabled */
2473 static inline void ____napi_schedule(struct softnet_data
*sd
,
2474 struct napi_struct
*napi
)
2476 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2477 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2481 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2482 * and src/dst port numbers. Returns a non-zero hash number on success
2485 __u32
__skb_get_rxhash(struct sk_buff
*skb
)
2487 int nhoff
, hash
= 0, poff
;
2488 struct ipv6hdr
*ip6
;
2491 u32 addr1
, addr2
, ihl
;
2497 nhoff
= skb_network_offset(skb
);
2499 switch (skb
->protocol
) {
2500 case __constant_htons(ETH_P_IP
):
2501 if (!pskb_may_pull(skb
, sizeof(*ip
) + nhoff
))
2504 ip
= (struct iphdr
*) (skb
->data
+ nhoff
);
2505 if (ip
->frag_off
& htons(IP_MF
| IP_OFFSET
))
2508 ip_proto
= ip
->protocol
;
2509 addr1
= (__force u32
) ip
->saddr
;
2510 addr2
= (__force u32
) ip
->daddr
;
2513 case __constant_htons(ETH_P_IPV6
):
2514 if (!pskb_may_pull(skb
, sizeof(*ip6
) + nhoff
))
2517 ip6
= (struct ipv6hdr
*) (skb
->data
+ nhoff
);
2518 ip_proto
= ip6
->nexthdr
;
2519 addr1
= (__force u32
) ip6
->saddr
.s6_addr32
[3];
2520 addr2
= (__force u32
) ip6
->daddr
.s6_addr32
[3];
2528 poff
= proto_ports_offset(ip_proto
);
2530 nhoff
+= ihl
* 4 + poff
;
2531 if (pskb_may_pull(skb
, nhoff
+ 4)) {
2532 ports
.v32
= * (__force u32
*) (skb
->data
+ nhoff
);
2533 if (ports
.v16
[1] < ports
.v16
[0])
2534 swap(ports
.v16
[0], ports
.v16
[1]);
2538 /* get a consistent hash (same value on both flow directions) */
2542 hash
= jhash_3words(addr1
, addr2
, ports
.v32
, hashrnd
);
2549 EXPORT_SYMBOL(__skb_get_rxhash
);
2553 /* One global table that all flow-based protocols share. */
2554 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2555 EXPORT_SYMBOL(rps_sock_flow_table
);
2558 * get_rps_cpu is called from netif_receive_skb and returns the target
2559 * CPU from the RPS map of the receiving queue for a given skb.
2560 * rcu_read_lock must be held on entry.
2562 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2563 struct rps_dev_flow
**rflowp
)
2565 struct netdev_rx_queue
*rxqueue
;
2566 struct rps_map
*map
;
2567 struct rps_dev_flow_table
*flow_table
;
2568 struct rps_sock_flow_table
*sock_flow_table
;
2572 if (skb_rx_queue_recorded(skb
)) {
2573 u16 index
= skb_get_rx_queue(skb
);
2574 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2575 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2576 "%s received packet on queue %u, but number "
2577 "of RX queues is %u\n",
2578 dev
->name
, index
, dev
->real_num_rx_queues
);
2581 rxqueue
= dev
->_rx
+ index
;
2585 map
= rcu_dereference(rxqueue
->rps_map
);
2587 if (map
->len
== 1) {
2588 tcpu
= map
->cpus
[0];
2589 if (cpu_online(tcpu
))
2593 } else if (!rcu_dereference_raw(rxqueue
->rps_flow_table
)) {
2597 skb_reset_network_header(skb
);
2598 if (!skb_get_rxhash(skb
))
2601 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2602 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2603 if (flow_table
&& sock_flow_table
) {
2605 struct rps_dev_flow
*rflow
;
2607 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2610 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2611 sock_flow_table
->mask
];
2614 * If the desired CPU (where last recvmsg was done) is
2615 * different from current CPU (one in the rx-queue flow
2616 * table entry), switch if one of the following holds:
2617 * - Current CPU is unset (equal to RPS_NO_CPU).
2618 * - Current CPU is offline.
2619 * - The current CPU's queue tail has advanced beyond the
2620 * last packet that was enqueued using this table entry.
2621 * This guarantees that all previous packets for the flow
2622 * have been dequeued, thus preserving in order delivery.
2624 if (unlikely(tcpu
!= next_cpu
) &&
2625 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2626 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2627 rflow
->last_qtail
)) >= 0)) {
2628 tcpu
= rflow
->cpu
= next_cpu
;
2629 if (tcpu
!= RPS_NO_CPU
)
2630 rflow
->last_qtail
= per_cpu(softnet_data
,
2631 tcpu
).input_queue_head
;
2633 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2641 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2643 if (cpu_online(tcpu
)) {
2653 /* Called from hardirq (IPI) context */
2654 static void rps_trigger_softirq(void *data
)
2656 struct softnet_data
*sd
= data
;
2658 ____napi_schedule(sd
, &sd
->backlog
);
2662 #endif /* CONFIG_RPS */
2665 * Check if this softnet_data structure is another cpu one
2666 * If yes, queue it to our IPI list and return 1
2669 static int rps_ipi_queued(struct softnet_data
*sd
)
2672 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2675 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2676 mysd
->rps_ipi_list
= sd
;
2678 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2681 #endif /* CONFIG_RPS */
2686 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2687 * queue (may be a remote CPU queue).
2689 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2690 unsigned int *qtail
)
2692 struct softnet_data
*sd
;
2693 unsigned long flags
;
2695 sd
= &per_cpu(softnet_data
, cpu
);
2697 local_irq_save(flags
);
2700 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2701 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2703 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2704 input_queue_tail_incr_save(sd
, qtail
);
2706 local_irq_restore(flags
);
2707 return NET_RX_SUCCESS
;
2710 /* Schedule NAPI for backlog device
2711 * We can use non atomic operation since we own the queue lock
2713 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2714 if (!rps_ipi_queued(sd
))
2715 ____napi_schedule(sd
, &sd
->backlog
);
2723 local_irq_restore(flags
);
2725 atomic_long_inc(&skb
->dev
->rx_dropped
);
2731 * netif_rx - post buffer to the network code
2732 * @skb: buffer to post
2734 * This function receives a packet from a device driver and queues it for
2735 * the upper (protocol) levels to process. It always succeeds. The buffer
2736 * may be dropped during processing for congestion control or by the
2740 * NET_RX_SUCCESS (no congestion)
2741 * NET_RX_DROP (packet was dropped)
2745 int netif_rx(struct sk_buff
*skb
)
2749 /* if netpoll wants it, pretend we never saw it */
2750 if (netpoll_rx(skb
))
2753 if (netdev_tstamp_prequeue
)
2754 net_timestamp_check(skb
);
2756 trace_netif_rx(skb
);
2759 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2765 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2767 cpu
= smp_processor_id();
2769 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2777 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2783 EXPORT_SYMBOL(netif_rx
);
2785 int netif_rx_ni(struct sk_buff
*skb
)
2790 err
= netif_rx(skb
);
2791 if (local_softirq_pending())
2797 EXPORT_SYMBOL(netif_rx_ni
);
2799 static void net_tx_action(struct softirq_action
*h
)
2801 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2803 if (sd
->completion_queue
) {
2804 struct sk_buff
*clist
;
2806 local_irq_disable();
2807 clist
= sd
->completion_queue
;
2808 sd
->completion_queue
= NULL
;
2812 struct sk_buff
*skb
= clist
;
2813 clist
= clist
->next
;
2815 WARN_ON(atomic_read(&skb
->users
));
2816 trace_kfree_skb(skb
, net_tx_action
);
2821 if (sd
->output_queue
) {
2824 local_irq_disable();
2825 head
= sd
->output_queue
;
2826 sd
->output_queue
= NULL
;
2827 sd
->output_queue_tailp
= &sd
->output_queue
;
2831 struct Qdisc
*q
= head
;
2832 spinlock_t
*root_lock
;
2834 head
= head
->next_sched
;
2836 root_lock
= qdisc_lock(q
);
2837 if (spin_trylock(root_lock
)) {
2838 smp_mb__before_clear_bit();
2839 clear_bit(__QDISC_STATE_SCHED
,
2842 spin_unlock(root_lock
);
2844 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2846 __netif_reschedule(q
);
2848 smp_mb__before_clear_bit();
2849 clear_bit(__QDISC_STATE_SCHED
,
2857 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2858 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2859 /* This hook is defined here for ATM LANE */
2860 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
2861 unsigned char *addr
) __read_mostly
;
2862 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
2865 #ifdef CONFIG_NET_CLS_ACT
2866 /* TODO: Maybe we should just force sch_ingress to be compiled in
2867 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2868 * a compare and 2 stores extra right now if we dont have it on
2869 * but have CONFIG_NET_CLS_ACT
2870 * NOTE: This doesnt stop any functionality; if you dont have
2871 * the ingress scheduler, you just cant add policies on ingress.
2874 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
2876 struct net_device
*dev
= skb
->dev
;
2877 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2878 int result
= TC_ACT_OK
;
2881 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
2882 if (net_ratelimit())
2883 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2884 skb
->skb_iif
, dev
->ifindex
);
2888 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2889 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2892 if (q
!= &noop_qdisc
) {
2893 spin_lock(qdisc_lock(q
));
2894 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2895 result
= qdisc_enqueue_root(skb
, q
);
2896 spin_unlock(qdisc_lock(q
));
2902 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2903 struct packet_type
**pt_prev
,
2904 int *ret
, struct net_device
*orig_dev
)
2906 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
2908 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
2912 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2916 switch (ing_filter(skb
, rxq
)) {
2930 * netdev_rx_handler_register - register receive handler
2931 * @dev: device to register a handler for
2932 * @rx_handler: receive handler to register
2933 * @rx_handler_data: data pointer that is used by rx handler
2935 * Register a receive hander for a device. This handler will then be
2936 * called from __netif_receive_skb. A negative errno code is returned
2939 * The caller must hold the rtnl_mutex.
2941 int netdev_rx_handler_register(struct net_device
*dev
,
2942 rx_handler_func_t
*rx_handler
,
2943 void *rx_handler_data
)
2947 if (dev
->rx_handler
)
2950 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
2951 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
2955 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
2958 * netdev_rx_handler_unregister - unregister receive handler
2959 * @dev: device to unregister a handler from
2961 * Unregister a receive hander from a device.
2963 * The caller must hold the rtnl_mutex.
2965 void netdev_rx_handler_unregister(struct net_device
*dev
)
2969 rcu_assign_pointer(dev
->rx_handler
, NULL
);
2970 rcu_assign_pointer(dev
->rx_handler_data
, NULL
);
2972 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
2974 static inline void skb_bond_set_mac_by_master(struct sk_buff
*skb
,
2975 struct net_device
*master
)
2977 if (skb
->pkt_type
== PACKET_HOST
) {
2978 u16
*dest
= (u16
*) eth_hdr(skb
)->h_dest
;
2980 memcpy(dest
, master
->dev_addr
, ETH_ALEN
);
2984 /* On bonding slaves other than the currently active slave, suppress
2985 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2986 * ARP on active-backup slaves with arp_validate enabled.
2988 int __skb_bond_should_drop(struct sk_buff
*skb
, struct net_device
*master
)
2990 struct net_device
*dev
= skb
->dev
;
2992 if (master
->priv_flags
& IFF_MASTER_ARPMON
)
2993 dev
->last_rx
= jiffies
;
2995 if ((master
->priv_flags
& IFF_MASTER_ALB
) &&
2996 (master
->priv_flags
& IFF_BRIDGE_PORT
)) {
2997 /* Do address unmangle. The local destination address
2998 * will be always the one master has. Provides the right
2999 * functionality in a bridge.
3001 skb_bond_set_mac_by_master(skb
, master
);
3004 if (dev
->priv_flags
& IFF_SLAVE_INACTIVE
) {
3005 if ((dev
->priv_flags
& IFF_SLAVE_NEEDARP
) &&
3006 skb
->protocol
== __cpu_to_be16(ETH_P_ARP
))
3009 if (master
->priv_flags
& IFF_MASTER_ALB
) {
3010 if (skb
->pkt_type
!= PACKET_BROADCAST
&&
3011 skb
->pkt_type
!= PACKET_MULTICAST
)
3014 if (master
->priv_flags
& IFF_MASTER_8023AD
&&
3015 skb
->protocol
== __cpu_to_be16(ETH_P_SLOW
))
3022 EXPORT_SYMBOL(__skb_bond_should_drop
);
3024 static int __netif_receive_skb(struct sk_buff
*skb
)
3026 struct packet_type
*ptype
, *pt_prev
;
3027 rx_handler_func_t
*rx_handler
;
3028 struct net_device
*orig_dev
;
3029 struct net_device
*master
;
3030 struct net_device
*null_or_orig
;
3031 struct net_device
*orig_or_bond
;
3032 int ret
= NET_RX_DROP
;
3035 if (!netdev_tstamp_prequeue
)
3036 net_timestamp_check(skb
);
3038 trace_netif_receive_skb(skb
);
3040 /* if we've gotten here through NAPI, check netpoll */
3041 if (netpoll_receive_skb(skb
))
3045 skb
->skb_iif
= skb
->dev
->ifindex
;
3048 * bonding note: skbs received on inactive slaves should only
3049 * be delivered to pkt handlers that are exact matches. Also
3050 * the deliver_no_wcard flag will be set. If packet handlers
3051 * are sensitive to duplicate packets these skbs will need to
3052 * be dropped at the handler.
3054 null_or_orig
= NULL
;
3055 orig_dev
= skb
->dev
;
3056 master
= ACCESS_ONCE(orig_dev
->master
);
3057 if (skb
->deliver_no_wcard
)
3058 null_or_orig
= orig_dev
;
3060 if (skb_bond_should_drop(skb
, master
)) {
3061 skb
->deliver_no_wcard
= 1;
3062 null_or_orig
= orig_dev
; /* deliver only exact match */
3067 __this_cpu_inc(softnet_data
.processed
);
3068 skb_reset_network_header(skb
);
3069 skb_reset_transport_header(skb
);
3070 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
3076 #ifdef CONFIG_NET_CLS_ACT
3077 if (skb
->tc_verd
& TC_NCLS
) {
3078 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3083 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3084 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
3085 ptype
->dev
== orig_dev
) {
3087 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3092 #ifdef CONFIG_NET_CLS_ACT
3093 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3099 /* Handle special case of bridge or macvlan */
3100 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3103 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3106 skb
= rx_handler(skb
);
3111 if (vlan_tx_tag_present(skb
)) {
3113 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3116 if (vlan_hwaccel_do_receive(&skb
)) {
3117 ret
= __netif_receive_skb(skb
);
3119 } else if (unlikely(!skb
))
3124 * Make sure frames received on VLAN interfaces stacked on
3125 * bonding interfaces still make their way to any base bonding
3126 * device that may have registered for a specific ptype. The
3127 * handler may have to adjust skb->dev and orig_dev.
3129 orig_or_bond
= orig_dev
;
3130 if ((skb
->dev
->priv_flags
& IFF_802_1Q_VLAN
) &&
3131 (vlan_dev_real_dev(skb
->dev
)->priv_flags
& IFF_BONDING
)) {
3132 orig_or_bond
= vlan_dev_real_dev(skb
->dev
);
3135 type
= skb
->protocol
;
3136 list_for_each_entry_rcu(ptype
,
3137 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3138 if (ptype
->type
== type
&& (ptype
->dev
== null_or_orig
||
3139 ptype
->dev
== skb
->dev
|| ptype
->dev
== orig_dev
||
3140 ptype
->dev
== orig_or_bond
)) {
3142 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3148 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3150 atomic_long_inc(&skb
->dev
->rx_dropped
);
3152 /* Jamal, now you will not able to escape explaining
3153 * me how you were going to use this. :-)
3164 * netif_receive_skb - process receive buffer from network
3165 * @skb: buffer to process
3167 * netif_receive_skb() is the main receive data processing function.
3168 * It always succeeds. The buffer may be dropped during processing
3169 * for congestion control or by the protocol layers.
3171 * This function may only be called from softirq context and interrupts
3172 * should be enabled.
3174 * Return values (usually ignored):
3175 * NET_RX_SUCCESS: no congestion
3176 * NET_RX_DROP: packet was dropped
3178 int netif_receive_skb(struct sk_buff
*skb
)
3180 if (netdev_tstamp_prequeue
)
3181 net_timestamp_check(skb
);
3183 if (skb_defer_rx_timestamp(skb
))
3184 return NET_RX_SUCCESS
;
3188 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3193 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3196 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3200 ret
= __netif_receive_skb(skb
);
3206 return __netif_receive_skb(skb
);
3209 EXPORT_SYMBOL(netif_receive_skb
);
3211 /* Network device is going away, flush any packets still pending
3212 * Called with irqs disabled.
3214 static void flush_backlog(void *arg
)
3216 struct net_device
*dev
= arg
;
3217 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3218 struct sk_buff
*skb
, *tmp
;
3221 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3222 if (skb
->dev
== dev
) {
3223 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3225 input_queue_head_incr(sd
);
3230 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3231 if (skb
->dev
== dev
) {
3232 __skb_unlink(skb
, &sd
->process_queue
);
3234 input_queue_head_incr(sd
);
3239 static int napi_gro_complete(struct sk_buff
*skb
)
3241 struct packet_type
*ptype
;
3242 __be16 type
= skb
->protocol
;
3243 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3246 if (NAPI_GRO_CB(skb
)->count
== 1) {
3247 skb_shinfo(skb
)->gso_size
= 0;
3252 list_for_each_entry_rcu(ptype
, head
, list
) {
3253 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3256 err
= ptype
->gro_complete(skb
);
3262 WARN_ON(&ptype
->list
== head
);
3264 return NET_RX_SUCCESS
;
3268 return netif_receive_skb(skb
);
3271 inline void napi_gro_flush(struct napi_struct
*napi
)
3273 struct sk_buff
*skb
, *next
;
3275 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3278 napi_gro_complete(skb
);
3281 napi
->gro_count
= 0;
3282 napi
->gro_list
= NULL
;
3284 EXPORT_SYMBOL(napi_gro_flush
);
3286 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3288 struct sk_buff
**pp
= NULL
;
3289 struct packet_type
*ptype
;
3290 __be16 type
= skb
->protocol
;
3291 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3294 enum gro_result ret
;
3296 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3299 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3303 list_for_each_entry_rcu(ptype
, head
, list
) {
3304 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3307 skb_set_network_header(skb
, skb_gro_offset(skb
));
3308 mac_len
= skb
->network_header
- skb
->mac_header
;
3309 skb
->mac_len
= mac_len
;
3310 NAPI_GRO_CB(skb
)->same_flow
= 0;
3311 NAPI_GRO_CB(skb
)->flush
= 0;
3312 NAPI_GRO_CB(skb
)->free
= 0;
3314 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3319 if (&ptype
->list
== head
)
3322 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3323 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3326 struct sk_buff
*nskb
= *pp
;
3330 napi_gro_complete(nskb
);
3337 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3341 NAPI_GRO_CB(skb
)->count
= 1;
3342 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3343 skb
->next
= napi
->gro_list
;
3344 napi
->gro_list
= skb
;
3348 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3349 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3351 BUG_ON(skb
->end
- skb
->tail
< grow
);
3353 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3356 skb
->data_len
-= grow
;
3358 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3359 skb_shinfo(skb
)->frags
[0].size
-= grow
;
3361 if (unlikely(!skb_shinfo(skb
)->frags
[0].size
)) {
3362 put_page(skb_shinfo(skb
)->frags
[0].page
);
3363 memmove(skb_shinfo(skb
)->frags
,
3364 skb_shinfo(skb
)->frags
+ 1,
3365 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3376 EXPORT_SYMBOL(dev_gro_receive
);
3378 static inline gro_result_t
3379 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3383 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3384 unsigned long diffs
;
3386 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3387 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3388 diffs
|= compare_ether_header(skb_mac_header(p
),
3389 skb_gro_mac_header(skb
));
3390 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3391 NAPI_GRO_CB(p
)->flush
= 0;
3394 return dev_gro_receive(napi
, skb
);
3397 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3401 if (netif_receive_skb(skb
))
3406 case GRO_MERGED_FREE
:
3417 EXPORT_SYMBOL(napi_skb_finish
);
3419 void skb_gro_reset_offset(struct sk_buff
*skb
)
3421 NAPI_GRO_CB(skb
)->data_offset
= 0;
3422 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3423 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3425 if (skb
->mac_header
== skb
->tail
&&
3426 !PageHighMem(skb_shinfo(skb
)->frags
[0].page
)) {
3427 NAPI_GRO_CB(skb
)->frag0
=
3428 page_address(skb_shinfo(skb
)->frags
[0].page
) +
3429 skb_shinfo(skb
)->frags
[0].page_offset
;
3430 NAPI_GRO_CB(skb
)->frag0_len
= skb_shinfo(skb
)->frags
[0].size
;
3433 EXPORT_SYMBOL(skb_gro_reset_offset
);
3435 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3437 skb_gro_reset_offset(skb
);
3439 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3441 EXPORT_SYMBOL(napi_gro_receive
);
3443 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3445 __skb_pull(skb
, skb_headlen(skb
));
3446 skb_reserve(skb
, NET_IP_ALIGN
- skb_headroom(skb
));
3452 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3454 struct sk_buff
*skb
= napi
->skb
;
3457 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3463 EXPORT_SYMBOL(napi_get_frags
);
3465 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3471 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3473 if (ret
== GRO_HELD
)
3474 skb_gro_pull(skb
, -ETH_HLEN
);
3475 else if (netif_receive_skb(skb
))
3480 case GRO_MERGED_FREE
:
3481 napi_reuse_skb(napi
, skb
);
3490 EXPORT_SYMBOL(napi_frags_finish
);
3492 struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3494 struct sk_buff
*skb
= napi
->skb
;
3501 skb_reset_mac_header(skb
);
3502 skb_gro_reset_offset(skb
);
3504 off
= skb_gro_offset(skb
);
3505 hlen
= off
+ sizeof(*eth
);
3506 eth
= skb_gro_header_fast(skb
, off
);
3507 if (skb_gro_header_hard(skb
, hlen
)) {
3508 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3509 if (unlikely(!eth
)) {
3510 napi_reuse_skb(napi
, skb
);
3516 skb_gro_pull(skb
, sizeof(*eth
));
3519 * This works because the only protocols we care about don't require
3520 * special handling. We'll fix it up properly at the end.
3522 skb
->protocol
= eth
->h_proto
;
3527 EXPORT_SYMBOL(napi_frags_skb
);
3529 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3531 struct sk_buff
*skb
= napi_frags_skb(napi
);
3536 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3538 EXPORT_SYMBOL(napi_gro_frags
);
3541 * net_rps_action sends any pending IPI's for rps.
3542 * Note: called with local irq disabled, but exits with local irq enabled.
3544 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3547 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3550 sd
->rps_ipi_list
= NULL
;
3554 /* Send pending IPI's to kick RPS processing on remote cpus. */
3556 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3558 if (cpu_online(remsd
->cpu
))
3559 __smp_call_function_single(remsd
->cpu
,
3568 static int process_backlog(struct napi_struct
*napi
, int quota
)
3571 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3574 /* Check if we have pending ipi, its better to send them now,
3575 * not waiting net_rx_action() end.
3577 if (sd
->rps_ipi_list
) {
3578 local_irq_disable();
3579 net_rps_action_and_irq_enable(sd
);
3582 napi
->weight
= weight_p
;
3583 local_irq_disable();
3584 while (work
< quota
) {
3585 struct sk_buff
*skb
;
3588 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3590 __netif_receive_skb(skb
);
3591 local_irq_disable();
3592 input_queue_head_incr(sd
);
3593 if (++work
>= quota
) {
3600 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3602 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3603 &sd
->process_queue
);
3605 if (qlen
< quota
- work
) {
3607 * Inline a custom version of __napi_complete().
3608 * only current cpu owns and manipulates this napi,
3609 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3610 * we can use a plain write instead of clear_bit(),
3611 * and we dont need an smp_mb() memory barrier.
3613 list_del(&napi
->poll_list
);
3616 quota
= work
+ qlen
;
3626 * __napi_schedule - schedule for receive
3627 * @n: entry to schedule
3629 * The entry's receive function will be scheduled to run
3631 void __napi_schedule(struct napi_struct
*n
)
3633 unsigned long flags
;
3635 local_irq_save(flags
);
3636 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3637 local_irq_restore(flags
);
3639 EXPORT_SYMBOL(__napi_schedule
);
3641 void __napi_complete(struct napi_struct
*n
)
3643 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3644 BUG_ON(n
->gro_list
);
3646 list_del(&n
->poll_list
);
3647 smp_mb__before_clear_bit();
3648 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3650 EXPORT_SYMBOL(__napi_complete
);
3652 void napi_complete(struct napi_struct
*n
)
3654 unsigned long flags
;
3657 * don't let napi dequeue from the cpu poll list
3658 * just in case its running on a different cpu
3660 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3664 local_irq_save(flags
);
3666 local_irq_restore(flags
);
3668 EXPORT_SYMBOL(napi_complete
);
3670 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3671 int (*poll
)(struct napi_struct
*, int), int weight
)
3673 INIT_LIST_HEAD(&napi
->poll_list
);
3674 napi
->gro_count
= 0;
3675 napi
->gro_list
= NULL
;
3678 napi
->weight
= weight
;
3679 list_add(&napi
->dev_list
, &dev
->napi_list
);
3681 #ifdef CONFIG_NETPOLL
3682 spin_lock_init(&napi
->poll_lock
);
3683 napi
->poll_owner
= -1;
3685 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3687 EXPORT_SYMBOL(netif_napi_add
);
3689 void netif_napi_del(struct napi_struct
*napi
)
3691 struct sk_buff
*skb
, *next
;
3693 list_del_init(&napi
->dev_list
);
3694 napi_free_frags(napi
);
3696 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3702 napi
->gro_list
= NULL
;
3703 napi
->gro_count
= 0;
3705 EXPORT_SYMBOL(netif_napi_del
);
3707 static void net_rx_action(struct softirq_action
*h
)
3709 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3710 unsigned long time_limit
= jiffies
+ 2;
3711 int budget
= netdev_budget
;
3714 local_irq_disable();
3716 while (!list_empty(&sd
->poll_list
)) {
3717 struct napi_struct
*n
;
3720 /* If softirq window is exhuasted then punt.
3721 * Allow this to run for 2 jiffies since which will allow
3722 * an average latency of 1.5/HZ.
3724 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3729 /* Even though interrupts have been re-enabled, this
3730 * access is safe because interrupts can only add new
3731 * entries to the tail of this list, and only ->poll()
3732 * calls can remove this head entry from the list.
3734 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3736 have
= netpoll_poll_lock(n
);
3740 /* This NAPI_STATE_SCHED test is for avoiding a race
3741 * with netpoll's poll_napi(). Only the entity which
3742 * obtains the lock and sees NAPI_STATE_SCHED set will
3743 * actually make the ->poll() call. Therefore we avoid
3744 * accidently calling ->poll() when NAPI is not scheduled.
3747 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3748 work
= n
->poll(n
, weight
);
3752 WARN_ON_ONCE(work
> weight
);
3756 local_irq_disable();
3758 /* Drivers must not modify the NAPI state if they
3759 * consume the entire weight. In such cases this code
3760 * still "owns" the NAPI instance and therefore can
3761 * move the instance around on the list at-will.
3763 if (unlikely(work
== weight
)) {
3764 if (unlikely(napi_disable_pending(n
))) {
3767 local_irq_disable();
3769 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3772 netpoll_poll_unlock(have
);
3775 net_rps_action_and_irq_enable(sd
);
3777 #ifdef CONFIG_NET_DMA
3779 * There may not be any more sk_buffs coming right now, so push
3780 * any pending DMA copies to hardware
3782 dma_issue_pending_all();
3789 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3793 static gifconf_func_t
*gifconf_list
[NPROTO
];
3796 * register_gifconf - register a SIOCGIF handler
3797 * @family: Address family
3798 * @gifconf: Function handler
3800 * Register protocol dependent address dumping routines. The handler
3801 * that is passed must not be freed or reused until it has been replaced
3802 * by another handler.
3804 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3806 if (family
>= NPROTO
)
3808 gifconf_list
[family
] = gifconf
;
3811 EXPORT_SYMBOL(register_gifconf
);
3815 * Map an interface index to its name (SIOCGIFNAME)
3819 * We need this ioctl for efficient implementation of the
3820 * if_indextoname() function required by the IPv6 API. Without
3821 * it, we would have to search all the interfaces to find a
3825 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3827 struct net_device
*dev
;
3831 * Fetch the caller's info block.
3834 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3838 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3844 strcpy(ifr
.ifr_name
, dev
->name
);
3847 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3853 * Perform a SIOCGIFCONF call. This structure will change
3854 * size eventually, and there is nothing I can do about it.
3855 * Thus we will need a 'compatibility mode'.
3858 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3861 struct net_device
*dev
;
3868 * Fetch the caller's info block.
3871 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3878 * Loop over the interfaces, and write an info block for each.
3882 for_each_netdev(net
, dev
) {
3883 for (i
= 0; i
< NPROTO
; i
++) {
3884 if (gifconf_list
[i
]) {
3887 done
= gifconf_list
[i
](dev
, NULL
, 0);
3889 done
= gifconf_list
[i
](dev
, pos
+ total
,
3899 * All done. Write the updated control block back to the caller.
3901 ifc
.ifc_len
= total
;
3904 * Both BSD and Solaris return 0 here, so we do too.
3906 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
3909 #ifdef CONFIG_PROC_FS
3911 * This is invoked by the /proc filesystem handler to display a device
3914 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3917 struct net
*net
= seq_file_net(seq
);
3919 struct net_device
*dev
;
3923 return SEQ_START_TOKEN
;
3926 for_each_netdev_rcu(net
, dev
)
3933 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3935 struct net_device
*dev
= (v
== SEQ_START_TOKEN
) ?
3936 first_net_device(seq_file_net(seq
)) :
3937 next_net_device((struct net_device
*)v
);
3940 return rcu_dereference(dev
);
3943 void dev_seq_stop(struct seq_file
*seq
, void *v
)
3949 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
3951 struct rtnl_link_stats64 temp
;
3952 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
3954 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3955 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3956 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
3958 stats
->rx_dropped
+ stats
->rx_missed_errors
,
3959 stats
->rx_fifo_errors
,
3960 stats
->rx_length_errors
+ stats
->rx_over_errors
+
3961 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
3962 stats
->rx_compressed
, stats
->multicast
,
3963 stats
->tx_bytes
, stats
->tx_packets
,
3964 stats
->tx_errors
, stats
->tx_dropped
,
3965 stats
->tx_fifo_errors
, stats
->collisions
,
3966 stats
->tx_carrier_errors
+
3967 stats
->tx_aborted_errors
+
3968 stats
->tx_window_errors
+
3969 stats
->tx_heartbeat_errors
,
3970 stats
->tx_compressed
);
3974 * Called from the PROCfs module. This now uses the new arbitrary sized
3975 * /proc/net interface to create /proc/net/dev
3977 static int dev_seq_show(struct seq_file
*seq
, void *v
)
3979 if (v
== SEQ_START_TOKEN
)
3980 seq_puts(seq
, "Inter-| Receive "
3982 " face |bytes packets errs drop fifo frame "
3983 "compressed multicast|bytes packets errs "
3984 "drop fifo colls carrier compressed\n");
3986 dev_seq_printf_stats(seq
, v
);
3990 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
3992 struct softnet_data
*sd
= NULL
;
3994 while (*pos
< nr_cpu_ids
)
3995 if (cpu_online(*pos
)) {
3996 sd
= &per_cpu(softnet_data
, *pos
);
4003 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4005 return softnet_get_online(pos
);
4008 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4011 return softnet_get_online(pos
);
4014 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4018 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4020 struct softnet_data
*sd
= v
;
4022 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4023 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4024 0, 0, 0, 0, /* was fastroute */
4025 sd
->cpu_collision
, sd
->received_rps
);
4029 static const struct seq_operations dev_seq_ops
= {
4030 .start
= dev_seq_start
,
4031 .next
= dev_seq_next
,
4032 .stop
= dev_seq_stop
,
4033 .show
= dev_seq_show
,
4036 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4038 return seq_open_net(inode
, file
, &dev_seq_ops
,
4039 sizeof(struct seq_net_private
));
4042 static const struct file_operations dev_seq_fops
= {
4043 .owner
= THIS_MODULE
,
4044 .open
= dev_seq_open
,
4046 .llseek
= seq_lseek
,
4047 .release
= seq_release_net
,
4050 static const struct seq_operations softnet_seq_ops
= {
4051 .start
= softnet_seq_start
,
4052 .next
= softnet_seq_next
,
4053 .stop
= softnet_seq_stop
,
4054 .show
= softnet_seq_show
,
4057 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4059 return seq_open(file
, &softnet_seq_ops
);
4062 static const struct file_operations softnet_seq_fops
= {
4063 .owner
= THIS_MODULE
,
4064 .open
= softnet_seq_open
,
4066 .llseek
= seq_lseek
,
4067 .release
= seq_release
,
4070 static void *ptype_get_idx(loff_t pos
)
4072 struct packet_type
*pt
= NULL
;
4076 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4082 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4083 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4092 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4096 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4099 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4101 struct packet_type
*pt
;
4102 struct list_head
*nxt
;
4106 if (v
== SEQ_START_TOKEN
)
4107 return ptype_get_idx(0);
4110 nxt
= pt
->list
.next
;
4111 if (pt
->type
== htons(ETH_P_ALL
)) {
4112 if (nxt
!= &ptype_all
)
4115 nxt
= ptype_base
[0].next
;
4117 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4119 while (nxt
== &ptype_base
[hash
]) {
4120 if (++hash
>= PTYPE_HASH_SIZE
)
4122 nxt
= ptype_base
[hash
].next
;
4125 return list_entry(nxt
, struct packet_type
, list
);
4128 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4134 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4136 struct packet_type
*pt
= v
;
4138 if (v
== SEQ_START_TOKEN
)
4139 seq_puts(seq
, "Type Device Function\n");
4140 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4141 if (pt
->type
== htons(ETH_P_ALL
))
4142 seq_puts(seq
, "ALL ");
4144 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4146 seq_printf(seq
, " %-8s %pF\n",
4147 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4153 static const struct seq_operations ptype_seq_ops
= {
4154 .start
= ptype_seq_start
,
4155 .next
= ptype_seq_next
,
4156 .stop
= ptype_seq_stop
,
4157 .show
= ptype_seq_show
,
4160 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4162 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4163 sizeof(struct seq_net_private
));
4166 static const struct file_operations ptype_seq_fops
= {
4167 .owner
= THIS_MODULE
,
4168 .open
= ptype_seq_open
,
4170 .llseek
= seq_lseek
,
4171 .release
= seq_release_net
,
4175 static int __net_init
dev_proc_net_init(struct net
*net
)
4179 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4181 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4183 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4186 if (wext_proc_init(net
))
4192 proc_net_remove(net
, "ptype");
4194 proc_net_remove(net
, "softnet_stat");
4196 proc_net_remove(net
, "dev");
4200 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4202 wext_proc_exit(net
);
4204 proc_net_remove(net
, "ptype");
4205 proc_net_remove(net
, "softnet_stat");
4206 proc_net_remove(net
, "dev");
4209 static struct pernet_operations __net_initdata dev_proc_ops
= {
4210 .init
= dev_proc_net_init
,
4211 .exit
= dev_proc_net_exit
,
4214 static int __init
dev_proc_init(void)
4216 return register_pernet_subsys(&dev_proc_ops
);
4219 #define dev_proc_init() 0
4220 #endif /* CONFIG_PROC_FS */
4224 * netdev_set_master - set up master/slave pair
4225 * @slave: slave device
4226 * @master: new master device
4228 * Changes the master device of the slave. Pass %NULL to break the
4229 * bonding. The caller must hold the RTNL semaphore. On a failure
4230 * a negative errno code is returned. On success the reference counts
4231 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4232 * function returns zero.
4234 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4236 struct net_device
*old
= slave
->master
;
4246 slave
->master
= master
;
4253 slave
->flags
|= IFF_SLAVE
;
4255 slave
->flags
&= ~IFF_SLAVE
;
4257 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4260 EXPORT_SYMBOL(netdev_set_master
);
4262 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4264 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4266 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4267 ops
->ndo_change_rx_flags(dev
, flags
);
4270 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4272 unsigned short old_flags
= dev
->flags
;
4278 dev
->flags
|= IFF_PROMISC
;
4279 dev
->promiscuity
+= inc
;
4280 if (dev
->promiscuity
== 0) {
4283 * If inc causes overflow, untouch promisc and return error.
4286 dev
->flags
&= ~IFF_PROMISC
;
4288 dev
->promiscuity
-= inc
;
4289 printk(KERN_WARNING
"%s: promiscuity touches roof, "
4290 "set promiscuity failed, promiscuity feature "
4291 "of device might be broken.\n", dev
->name
);
4295 if (dev
->flags
!= old_flags
) {
4296 printk(KERN_INFO
"device %s %s promiscuous mode\n",
4297 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
4299 if (audit_enabled
) {
4300 current_uid_gid(&uid
, &gid
);
4301 audit_log(current
->audit_context
, GFP_ATOMIC
,
4302 AUDIT_ANOM_PROMISCUOUS
,
4303 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4304 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4305 (old_flags
& IFF_PROMISC
),
4306 audit_get_loginuid(current
),
4308 audit_get_sessionid(current
));
4311 dev_change_rx_flags(dev
, IFF_PROMISC
);
4317 * dev_set_promiscuity - update promiscuity count on a device
4321 * Add or remove promiscuity from a device. While the count in the device
4322 * remains above zero the interface remains promiscuous. Once it hits zero
4323 * the device reverts back to normal filtering operation. A negative inc
4324 * value is used to drop promiscuity on the device.
4325 * Return 0 if successful or a negative errno code on error.
4327 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4329 unsigned short old_flags
= dev
->flags
;
4332 err
= __dev_set_promiscuity(dev
, inc
);
4335 if (dev
->flags
!= old_flags
)
4336 dev_set_rx_mode(dev
);
4339 EXPORT_SYMBOL(dev_set_promiscuity
);
4342 * dev_set_allmulti - update allmulti count on a device
4346 * Add or remove reception of all multicast frames to a device. While the
4347 * count in the device remains above zero the interface remains listening
4348 * to all interfaces. Once it hits zero the device reverts back to normal
4349 * filtering operation. A negative @inc value is used to drop the counter
4350 * when releasing a resource needing all multicasts.
4351 * Return 0 if successful or a negative errno code on error.
4354 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4356 unsigned short old_flags
= dev
->flags
;
4360 dev
->flags
|= IFF_ALLMULTI
;
4361 dev
->allmulti
+= inc
;
4362 if (dev
->allmulti
== 0) {
4365 * If inc causes overflow, untouch allmulti and return error.
4368 dev
->flags
&= ~IFF_ALLMULTI
;
4370 dev
->allmulti
-= inc
;
4371 printk(KERN_WARNING
"%s: allmulti touches roof, "
4372 "set allmulti failed, allmulti feature of "
4373 "device might be broken.\n", dev
->name
);
4377 if (dev
->flags
^ old_flags
) {
4378 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4379 dev_set_rx_mode(dev
);
4383 EXPORT_SYMBOL(dev_set_allmulti
);
4386 * Upload unicast and multicast address lists to device and
4387 * configure RX filtering. When the device doesn't support unicast
4388 * filtering it is put in promiscuous mode while unicast addresses
4391 void __dev_set_rx_mode(struct net_device
*dev
)
4393 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4395 /* dev_open will call this function so the list will stay sane. */
4396 if (!(dev
->flags
&IFF_UP
))
4399 if (!netif_device_present(dev
))
4402 if (ops
->ndo_set_rx_mode
)
4403 ops
->ndo_set_rx_mode(dev
);
4405 /* Unicast addresses changes may only happen under the rtnl,
4406 * therefore calling __dev_set_promiscuity here is safe.
4408 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4409 __dev_set_promiscuity(dev
, 1);
4410 dev
->uc_promisc
= 1;
4411 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4412 __dev_set_promiscuity(dev
, -1);
4413 dev
->uc_promisc
= 0;
4416 if (ops
->ndo_set_multicast_list
)
4417 ops
->ndo_set_multicast_list(dev
);
4421 void dev_set_rx_mode(struct net_device
*dev
)
4423 netif_addr_lock_bh(dev
);
4424 __dev_set_rx_mode(dev
);
4425 netif_addr_unlock_bh(dev
);
4429 * dev_get_flags - get flags reported to userspace
4432 * Get the combination of flag bits exported through APIs to userspace.
4434 unsigned dev_get_flags(const struct net_device
*dev
)
4438 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4443 (dev
->gflags
& (IFF_PROMISC
|
4446 if (netif_running(dev
)) {
4447 if (netif_oper_up(dev
))
4448 flags
|= IFF_RUNNING
;
4449 if (netif_carrier_ok(dev
))
4450 flags
|= IFF_LOWER_UP
;
4451 if (netif_dormant(dev
))
4452 flags
|= IFF_DORMANT
;
4457 EXPORT_SYMBOL(dev_get_flags
);
4459 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4461 int old_flags
= dev
->flags
;
4467 * Set the flags on our device.
4470 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4471 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4473 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4477 * Load in the correct multicast list now the flags have changed.
4480 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4481 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4483 dev_set_rx_mode(dev
);
4486 * Have we downed the interface. We handle IFF_UP ourselves
4487 * according to user attempts to set it, rather than blindly
4492 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4493 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4496 dev_set_rx_mode(dev
);
4499 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4500 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4502 dev
->gflags
^= IFF_PROMISC
;
4503 dev_set_promiscuity(dev
, inc
);
4506 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4507 is important. Some (broken) drivers set IFF_PROMISC, when
4508 IFF_ALLMULTI is requested not asking us and not reporting.
4510 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4511 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4513 dev
->gflags
^= IFF_ALLMULTI
;
4514 dev_set_allmulti(dev
, inc
);
4520 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4522 unsigned int changes
= dev
->flags
^ old_flags
;
4524 if (changes
& IFF_UP
) {
4525 if (dev
->flags
& IFF_UP
)
4526 call_netdevice_notifiers(NETDEV_UP
, dev
);
4528 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4531 if (dev
->flags
& IFF_UP
&&
4532 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4533 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4537 * dev_change_flags - change device settings
4539 * @flags: device state flags
4541 * Change settings on device based state flags. The flags are
4542 * in the userspace exported format.
4544 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
4547 int old_flags
= dev
->flags
;
4549 ret
= __dev_change_flags(dev
, flags
);
4553 changes
= old_flags
^ dev
->flags
;
4555 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4557 __dev_notify_flags(dev
, old_flags
);
4560 EXPORT_SYMBOL(dev_change_flags
);
4563 * dev_set_mtu - Change maximum transfer unit
4565 * @new_mtu: new transfer unit
4567 * Change the maximum transfer size of the network device.
4569 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4571 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4574 if (new_mtu
== dev
->mtu
)
4577 /* MTU must be positive. */
4581 if (!netif_device_present(dev
))
4585 if (ops
->ndo_change_mtu
)
4586 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4590 if (!err
&& dev
->flags
& IFF_UP
)
4591 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4594 EXPORT_SYMBOL(dev_set_mtu
);
4597 * dev_set_mac_address - Change Media Access Control Address
4601 * Change the hardware (MAC) address of the device
4603 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4605 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4608 if (!ops
->ndo_set_mac_address
)
4610 if (sa
->sa_family
!= dev
->type
)
4612 if (!netif_device_present(dev
))
4614 err
= ops
->ndo_set_mac_address(dev
, sa
);
4616 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4619 EXPORT_SYMBOL(dev_set_mac_address
);
4622 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4624 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4627 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4633 case SIOCGIFFLAGS
: /* Get interface flags */
4634 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4637 case SIOCGIFMETRIC
: /* Get the metric on the interface
4638 (currently unused) */
4639 ifr
->ifr_metric
= 0;
4642 case SIOCGIFMTU
: /* Get the MTU of a device */
4643 ifr
->ifr_mtu
= dev
->mtu
;
4648 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4650 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4651 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4652 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4660 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4661 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4662 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4663 ifr
->ifr_map
.irq
= dev
->irq
;
4664 ifr
->ifr_map
.dma
= dev
->dma
;
4665 ifr
->ifr_map
.port
= dev
->if_port
;
4669 ifr
->ifr_ifindex
= dev
->ifindex
;
4673 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4677 /* dev_ioctl() should ensure this case
4689 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4691 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4694 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4695 const struct net_device_ops
*ops
;
4700 ops
= dev
->netdev_ops
;
4703 case SIOCSIFFLAGS
: /* Set interface flags */
4704 return dev_change_flags(dev
, ifr
->ifr_flags
);
4706 case SIOCSIFMETRIC
: /* Set the metric on the interface
4707 (currently unused) */
4710 case SIOCSIFMTU
: /* Set the MTU of a device */
4711 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4714 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4716 case SIOCSIFHWBROADCAST
:
4717 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4719 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4720 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4721 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4725 if (ops
->ndo_set_config
) {
4726 if (!netif_device_present(dev
))
4728 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4733 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4734 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4736 if (!netif_device_present(dev
))
4738 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4741 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
4742 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4744 if (!netif_device_present(dev
))
4746 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4749 if (ifr
->ifr_qlen
< 0)
4751 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4755 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4756 return dev_change_name(dev
, ifr
->ifr_newname
);
4759 * Unknown or private ioctl
4762 if ((cmd
>= SIOCDEVPRIVATE
&&
4763 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4764 cmd
== SIOCBONDENSLAVE
||
4765 cmd
== SIOCBONDRELEASE
||
4766 cmd
== SIOCBONDSETHWADDR
||
4767 cmd
== SIOCBONDSLAVEINFOQUERY
||
4768 cmd
== SIOCBONDINFOQUERY
||
4769 cmd
== SIOCBONDCHANGEACTIVE
||
4770 cmd
== SIOCGMIIPHY
||
4771 cmd
== SIOCGMIIREG
||
4772 cmd
== SIOCSMIIREG
||
4773 cmd
== SIOCBRADDIF
||
4774 cmd
== SIOCBRDELIF
||
4775 cmd
== SIOCSHWTSTAMP
||
4776 cmd
== SIOCWANDEV
) {
4778 if (ops
->ndo_do_ioctl
) {
4779 if (netif_device_present(dev
))
4780 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4792 * This function handles all "interface"-type I/O control requests. The actual
4793 * 'doing' part of this is dev_ifsioc above.
4797 * dev_ioctl - network device ioctl
4798 * @net: the applicable net namespace
4799 * @cmd: command to issue
4800 * @arg: pointer to a struct ifreq in user space
4802 * Issue ioctl functions to devices. This is normally called by the
4803 * user space syscall interfaces but can sometimes be useful for
4804 * other purposes. The return value is the return from the syscall if
4805 * positive or a negative errno code on error.
4808 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4814 /* One special case: SIOCGIFCONF takes ifconf argument
4815 and requires shared lock, because it sleeps writing
4819 if (cmd
== SIOCGIFCONF
) {
4821 ret
= dev_ifconf(net
, (char __user
*) arg
);
4825 if (cmd
== SIOCGIFNAME
)
4826 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4828 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4831 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4833 colon
= strchr(ifr
.ifr_name
, ':');
4838 * See which interface the caller is talking about.
4843 * These ioctl calls:
4844 * - can be done by all.
4845 * - atomic and do not require locking.
4856 dev_load(net
, ifr
.ifr_name
);
4858 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
4863 if (copy_to_user(arg
, &ifr
,
4864 sizeof(struct ifreq
)))
4870 dev_load(net
, ifr
.ifr_name
);
4872 ret
= dev_ethtool(net
, &ifr
);
4877 if (copy_to_user(arg
, &ifr
,
4878 sizeof(struct ifreq
)))
4884 * These ioctl calls:
4885 * - require superuser power.
4886 * - require strict serialization.
4892 if (!capable(CAP_NET_ADMIN
))
4894 dev_load(net
, ifr
.ifr_name
);
4896 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4901 if (copy_to_user(arg
, &ifr
,
4902 sizeof(struct ifreq
)))
4908 * These ioctl calls:
4909 * - require superuser power.
4910 * - require strict serialization.
4911 * - do not return a value
4921 case SIOCSIFHWBROADCAST
:
4924 case SIOCBONDENSLAVE
:
4925 case SIOCBONDRELEASE
:
4926 case SIOCBONDSETHWADDR
:
4927 case SIOCBONDCHANGEACTIVE
:
4931 if (!capable(CAP_NET_ADMIN
))
4934 case SIOCBONDSLAVEINFOQUERY
:
4935 case SIOCBONDINFOQUERY
:
4936 dev_load(net
, ifr
.ifr_name
);
4938 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4943 /* Get the per device memory space. We can add this but
4944 * currently do not support it */
4946 /* Set the per device memory buffer space.
4947 * Not applicable in our case */
4952 * Unknown or private ioctl.
4955 if (cmd
== SIOCWANDEV
||
4956 (cmd
>= SIOCDEVPRIVATE
&&
4957 cmd
<= SIOCDEVPRIVATE
+ 15)) {
4958 dev_load(net
, ifr
.ifr_name
);
4960 ret
= dev_ifsioc(net
, &ifr
, cmd
);
4962 if (!ret
&& copy_to_user(arg
, &ifr
,
4963 sizeof(struct ifreq
)))
4967 /* Take care of Wireless Extensions */
4968 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
4969 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
4976 * dev_new_index - allocate an ifindex
4977 * @net: the applicable net namespace
4979 * Returns a suitable unique value for a new device interface
4980 * number. The caller must hold the rtnl semaphore or the
4981 * dev_base_lock to be sure it remains unique.
4983 static int dev_new_index(struct net
*net
)
4989 if (!__dev_get_by_index(net
, ifindex
))
4994 /* Delayed registration/unregisteration */
4995 static LIST_HEAD(net_todo_list
);
4997 static void net_set_todo(struct net_device
*dev
)
4999 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5002 static void rollback_registered_many(struct list_head
*head
)
5004 struct net_device
*dev
, *tmp
;
5006 BUG_ON(dev_boot_phase
);
5009 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5010 /* Some devices call without registering
5011 * for initialization unwind. Remove those
5012 * devices and proceed with the remaining.
5014 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5015 pr_debug("unregister_netdevice: device %s/%p never "
5016 "was registered\n", dev
->name
, dev
);
5019 list_del(&dev
->unreg_list
);
5023 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5026 /* If device is running, close it first. */
5027 dev_close_many(head
);
5029 list_for_each_entry(dev
, head
, unreg_list
) {
5030 /* And unlink it from device chain. */
5031 unlist_netdevice(dev
);
5033 dev
->reg_state
= NETREG_UNREGISTERING
;
5038 list_for_each_entry(dev
, head
, unreg_list
) {
5039 /* Shutdown queueing discipline. */
5043 /* Notify protocols, that we are about to destroy
5044 this device. They should clean all the things.
5046 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5048 if (!dev
->rtnl_link_ops
||
5049 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5050 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5053 * Flush the unicast and multicast chains
5058 if (dev
->netdev_ops
->ndo_uninit
)
5059 dev
->netdev_ops
->ndo_uninit(dev
);
5061 /* Notifier chain MUST detach us from master device. */
5062 WARN_ON(dev
->master
);
5064 /* Remove entries from kobject tree */
5065 netdev_unregister_kobject(dev
);
5068 /* Process any work delayed until the end of the batch */
5069 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5070 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5074 list_for_each_entry(dev
, head
, unreg_list
)
5078 static void rollback_registered(struct net_device
*dev
)
5082 list_add(&dev
->unreg_list
, &single
);
5083 rollback_registered_many(&single
);
5086 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
5088 /* Fix illegal SG+CSUM combinations. */
5089 if ((features
& NETIF_F_SG
) &&
5090 !(features
& NETIF_F_ALL_CSUM
)) {
5092 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
5093 "checksum feature.\n", name
);
5094 features
&= ~NETIF_F_SG
;
5097 /* TSO requires that SG is present as well. */
5098 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
5100 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
5101 "SG feature.\n", name
);
5102 features
&= ~NETIF_F_TSO
;
5105 if (features
& NETIF_F_UFO
) {
5106 /* maybe split UFO into V4 and V6? */
5107 if (!((features
& NETIF_F_GEN_CSUM
) ||
5108 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5109 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5111 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
5112 "since no checksum offload features.\n",
5114 features
&= ~NETIF_F_UFO
;
5117 if (!(features
& NETIF_F_SG
)) {
5119 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
5120 "since no NETIF_F_SG feature.\n", name
);
5121 features
&= ~NETIF_F_UFO
;
5127 EXPORT_SYMBOL(netdev_fix_features
);
5130 * netif_stacked_transfer_operstate - transfer operstate
5131 * @rootdev: the root or lower level device to transfer state from
5132 * @dev: the device to transfer operstate to
5134 * Transfer operational state from root to device. This is normally
5135 * called when a stacking relationship exists between the root
5136 * device and the device(a leaf device).
5138 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5139 struct net_device
*dev
)
5141 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5142 netif_dormant_on(dev
);
5144 netif_dormant_off(dev
);
5146 if (netif_carrier_ok(rootdev
)) {
5147 if (!netif_carrier_ok(dev
))
5148 netif_carrier_on(dev
);
5150 if (netif_carrier_ok(dev
))
5151 netif_carrier_off(dev
);
5154 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5157 static int netif_alloc_rx_queues(struct net_device
*dev
)
5159 unsigned int i
, count
= dev
->num_rx_queues
;
5160 struct netdev_rx_queue
*rx
;
5164 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5166 pr_err("netdev: Unable to allocate %u rx queues.\n", count
);
5171 for (i
= 0; i
< count
; i
++)
5177 static void netdev_init_one_queue(struct net_device
*dev
,
5178 struct netdev_queue
*queue
, void *_unused
)
5180 /* Initialize queue lock */
5181 spin_lock_init(&queue
->_xmit_lock
);
5182 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5183 queue
->xmit_lock_owner
= -1;
5184 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5188 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5190 unsigned int count
= dev
->num_tx_queues
;
5191 struct netdev_queue
*tx
;
5195 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5197 pr_err("netdev: Unable to allocate %u tx queues.\n",
5203 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5204 spin_lock_init(&dev
->tx_global_lock
);
5210 * register_netdevice - register a network device
5211 * @dev: device to register
5213 * Take a completed network device structure and add it to the kernel
5214 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5215 * chain. 0 is returned on success. A negative errno code is returned
5216 * on a failure to set up the device, or if the name is a duplicate.
5218 * Callers must hold the rtnl semaphore. You may want
5219 * register_netdev() instead of this.
5222 * The locking appears insufficient to guarantee two parallel registers
5223 * will not get the same name.
5226 int register_netdevice(struct net_device
*dev
)
5229 struct net
*net
= dev_net(dev
);
5231 BUG_ON(dev_boot_phase
);
5236 /* When net_device's are persistent, this will be fatal. */
5237 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5240 spin_lock_init(&dev
->addr_list_lock
);
5241 netdev_set_addr_lockdep_class(dev
);
5245 /* Init, if this function is available */
5246 if (dev
->netdev_ops
->ndo_init
) {
5247 ret
= dev
->netdev_ops
->ndo_init(dev
);
5255 ret
= dev_get_valid_name(dev
, dev
->name
, 0);
5259 dev
->ifindex
= dev_new_index(net
);
5260 if (dev
->iflink
== -1)
5261 dev
->iflink
= dev
->ifindex
;
5263 /* Fix illegal checksum combinations */
5264 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
5265 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5266 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
5268 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5271 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
5272 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5273 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
5275 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
5278 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
5280 /* Enable software GSO if SG is supported. */
5281 if (dev
->features
& NETIF_F_SG
)
5282 dev
->features
|= NETIF_F_GSO
;
5284 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5285 * vlan_dev_init() will do the dev->features check, so these features
5286 * are enabled only if supported by underlying device.
5288 dev
->vlan_features
|= (NETIF_F_GRO
| NETIF_F_HIGHDMA
);
5290 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5291 ret
= notifier_to_errno(ret
);
5295 ret
= netdev_register_kobject(dev
);
5298 dev
->reg_state
= NETREG_REGISTERED
;
5301 * Default initial state at registry is that the
5302 * device is present.
5305 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5307 dev_init_scheduler(dev
);
5309 list_netdevice(dev
);
5311 /* Notify protocols, that a new device appeared. */
5312 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5313 ret
= notifier_to_errno(ret
);
5315 rollback_registered(dev
);
5316 dev
->reg_state
= NETREG_UNREGISTERED
;
5319 * Prevent userspace races by waiting until the network
5320 * device is fully setup before sending notifications.
5322 if (!dev
->rtnl_link_ops
||
5323 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5324 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5330 if (dev
->netdev_ops
->ndo_uninit
)
5331 dev
->netdev_ops
->ndo_uninit(dev
);
5334 EXPORT_SYMBOL(register_netdevice
);
5337 * init_dummy_netdev - init a dummy network device for NAPI
5338 * @dev: device to init
5340 * This takes a network device structure and initialize the minimum
5341 * amount of fields so it can be used to schedule NAPI polls without
5342 * registering a full blown interface. This is to be used by drivers
5343 * that need to tie several hardware interfaces to a single NAPI
5344 * poll scheduler due to HW limitations.
5346 int init_dummy_netdev(struct net_device
*dev
)
5348 /* Clear everything. Note we don't initialize spinlocks
5349 * are they aren't supposed to be taken by any of the
5350 * NAPI code and this dummy netdev is supposed to be
5351 * only ever used for NAPI polls
5353 memset(dev
, 0, sizeof(struct net_device
));
5355 /* make sure we BUG if trying to hit standard
5356 * register/unregister code path
5358 dev
->reg_state
= NETREG_DUMMY
;
5360 /* NAPI wants this */
5361 INIT_LIST_HEAD(&dev
->napi_list
);
5363 /* a dummy interface is started by default */
5364 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5365 set_bit(__LINK_STATE_START
, &dev
->state
);
5367 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5368 * because users of this 'device' dont need to change
5374 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5378 * register_netdev - register a network device
5379 * @dev: device to register
5381 * Take a completed network device structure and add it to the kernel
5382 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5383 * chain. 0 is returned on success. A negative errno code is returned
5384 * on a failure to set up the device, or if the name is a duplicate.
5386 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5387 * and expands the device name if you passed a format string to
5390 int register_netdev(struct net_device
*dev
)
5397 * If the name is a format string the caller wants us to do a
5400 if (strchr(dev
->name
, '%')) {
5401 err
= dev_alloc_name(dev
, dev
->name
);
5406 err
= register_netdevice(dev
);
5411 EXPORT_SYMBOL(register_netdev
);
5413 int netdev_refcnt_read(const struct net_device
*dev
)
5417 for_each_possible_cpu(i
)
5418 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5421 EXPORT_SYMBOL(netdev_refcnt_read
);
5424 * netdev_wait_allrefs - wait until all references are gone.
5426 * This is called when unregistering network devices.
5428 * Any protocol or device that holds a reference should register
5429 * for netdevice notification, and cleanup and put back the
5430 * reference if they receive an UNREGISTER event.
5431 * We can get stuck here if buggy protocols don't correctly
5434 static void netdev_wait_allrefs(struct net_device
*dev
)
5436 unsigned long rebroadcast_time
, warning_time
;
5439 linkwatch_forget_dev(dev
);
5441 rebroadcast_time
= warning_time
= jiffies
;
5442 refcnt
= netdev_refcnt_read(dev
);
5444 while (refcnt
!= 0) {
5445 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5448 /* Rebroadcast unregister notification */
5449 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5450 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5451 * should have already handle it the first time */
5453 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5455 /* We must not have linkwatch events
5456 * pending on unregister. If this
5457 * happens, we simply run the queue
5458 * unscheduled, resulting in a noop
5461 linkwatch_run_queue();
5466 rebroadcast_time
= jiffies
;
5471 refcnt
= netdev_refcnt_read(dev
);
5473 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5474 printk(KERN_EMERG
"unregister_netdevice: "
5475 "waiting for %s to become free. Usage "
5478 warning_time
= jiffies
;
5487 * register_netdevice(x1);
5488 * register_netdevice(x2);
5490 * unregister_netdevice(y1);
5491 * unregister_netdevice(y2);
5497 * We are invoked by rtnl_unlock().
5498 * This allows us to deal with problems:
5499 * 1) We can delete sysfs objects which invoke hotplug
5500 * without deadlocking with linkwatch via keventd.
5501 * 2) Since we run with the RTNL semaphore not held, we can sleep
5502 * safely in order to wait for the netdev refcnt to drop to zero.
5504 * We must not return until all unregister events added during
5505 * the interval the lock was held have been completed.
5507 void netdev_run_todo(void)
5509 struct list_head list
;
5511 /* Snapshot list, allow later requests */
5512 list_replace_init(&net_todo_list
, &list
);
5516 while (!list_empty(&list
)) {
5517 struct net_device
*dev
5518 = list_first_entry(&list
, struct net_device
, todo_list
);
5519 list_del(&dev
->todo_list
);
5521 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5522 printk(KERN_ERR
"network todo '%s' but state %d\n",
5523 dev
->name
, dev
->reg_state
);
5528 dev
->reg_state
= NETREG_UNREGISTERED
;
5530 on_each_cpu(flush_backlog
, dev
, 1);
5532 netdev_wait_allrefs(dev
);
5535 BUG_ON(netdev_refcnt_read(dev
));
5536 WARN_ON(rcu_dereference_raw(dev
->ip_ptr
));
5537 WARN_ON(rcu_dereference_raw(dev
->ip6_ptr
));
5538 WARN_ON(dev
->dn_ptr
);
5540 if (dev
->destructor
)
5541 dev
->destructor(dev
);
5543 /* Free network device */
5544 kobject_put(&dev
->dev
.kobj
);
5549 * dev_txq_stats_fold - fold tx_queues stats
5550 * @dev: device to get statistics from
5551 * @stats: struct rtnl_link_stats64 to hold results
5553 void dev_txq_stats_fold(const struct net_device
*dev
,
5554 struct rtnl_link_stats64
*stats
)
5556 u64 tx_bytes
= 0, tx_packets
= 0, tx_dropped
= 0;
5558 struct netdev_queue
*txq
;
5560 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
5561 txq
= netdev_get_tx_queue(dev
, i
);
5562 spin_lock_bh(&txq
->_xmit_lock
);
5563 tx_bytes
+= txq
->tx_bytes
;
5564 tx_packets
+= txq
->tx_packets
;
5565 tx_dropped
+= txq
->tx_dropped
;
5566 spin_unlock_bh(&txq
->_xmit_lock
);
5568 if (tx_bytes
|| tx_packets
|| tx_dropped
) {
5569 stats
->tx_bytes
= tx_bytes
;
5570 stats
->tx_packets
= tx_packets
;
5571 stats
->tx_dropped
= tx_dropped
;
5574 EXPORT_SYMBOL(dev_txq_stats_fold
);
5576 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5577 * fields in the same order, with only the type differing.
5579 static void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5580 const struct net_device_stats
*netdev_stats
)
5582 #if BITS_PER_LONG == 64
5583 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5584 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5586 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5587 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5588 u64
*dst
= (u64
*)stats64
;
5590 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5591 sizeof(*stats64
) / sizeof(u64
));
5592 for (i
= 0; i
< n
; i
++)
5598 * dev_get_stats - get network device statistics
5599 * @dev: device to get statistics from
5600 * @storage: place to store stats
5602 * Get network statistics from device. Return @storage.
5603 * The device driver may provide its own method by setting
5604 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5605 * otherwise the internal statistics structure is used.
5607 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5608 struct rtnl_link_stats64
*storage
)
5610 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5612 if (ops
->ndo_get_stats64
) {
5613 memset(storage
, 0, sizeof(*storage
));
5614 ops
->ndo_get_stats64(dev
, storage
);
5615 } else if (ops
->ndo_get_stats
) {
5616 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5618 netdev_stats_to_stats64(storage
, &dev
->stats
);
5619 dev_txq_stats_fold(dev
, storage
);
5621 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5624 EXPORT_SYMBOL(dev_get_stats
);
5626 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5628 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5630 #ifdef CONFIG_NET_CLS_ACT
5633 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5636 netdev_init_one_queue(dev
, queue
, NULL
);
5637 queue
->qdisc
= &noop_qdisc
;
5638 queue
->qdisc_sleeping
= &noop_qdisc
;
5639 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5645 * alloc_netdev_mq - allocate network device
5646 * @sizeof_priv: size of private data to allocate space for
5647 * @name: device name format string
5648 * @setup: callback to initialize device
5649 * @queue_count: the number of subqueues to allocate
5651 * Allocates a struct net_device with private data area for driver use
5652 * and performs basic initialization. Also allocates subquue structs
5653 * for each queue on the device at the end of the netdevice.
5655 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
5656 void (*setup
)(struct net_device
*), unsigned int queue_count
)
5658 struct net_device
*dev
;
5660 struct net_device
*p
;
5662 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5664 if (queue_count
< 1) {
5665 pr_err("alloc_netdev: Unable to allocate device "
5666 "with zero queues.\n");
5670 alloc_size
= sizeof(struct net_device
);
5672 /* ensure 32-byte alignment of private area */
5673 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5674 alloc_size
+= sizeof_priv
;
5676 /* ensure 32-byte alignment of whole construct */
5677 alloc_size
+= NETDEV_ALIGN
- 1;
5679 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5681 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
5685 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5686 dev
->padded
= (char *)dev
- (char *)p
;
5688 dev
->pcpu_refcnt
= alloc_percpu(int);
5689 if (!dev
->pcpu_refcnt
)
5692 if (dev_addr_init(dev
))
5698 dev_net_set(dev
, &init_net
);
5700 dev
->num_tx_queues
= queue_count
;
5701 dev
->real_num_tx_queues
= queue_count
;
5702 if (netif_alloc_netdev_queues(dev
))
5706 dev
->num_rx_queues
= queue_count
;
5707 dev
->real_num_rx_queues
= queue_count
;
5708 if (netif_alloc_rx_queues(dev
))
5712 dev
->gso_max_size
= GSO_MAX_SIZE
;
5714 INIT_LIST_HEAD(&dev
->ethtool_ntuple_list
.list
);
5715 dev
->ethtool_ntuple_list
.count
= 0;
5716 INIT_LIST_HEAD(&dev
->napi_list
);
5717 INIT_LIST_HEAD(&dev
->unreg_list
);
5718 INIT_LIST_HEAD(&dev
->link_watch_list
);
5719 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5721 strcpy(dev
->name
, name
);
5725 free_percpu(dev
->pcpu_refcnt
);
5735 EXPORT_SYMBOL(alloc_netdev_mq
);
5738 * free_netdev - free network device
5741 * This function does the last stage of destroying an allocated device
5742 * interface. The reference to the device object is released.
5743 * If this is the last reference then it will be freed.
5745 void free_netdev(struct net_device
*dev
)
5747 struct napi_struct
*p
, *n
;
5749 release_net(dev_net(dev
));
5756 kfree(rcu_dereference_raw(dev
->ingress_queue
));
5758 /* Flush device addresses */
5759 dev_addr_flush(dev
);
5761 /* Clear ethtool n-tuple list */
5762 ethtool_ntuple_flush(dev
);
5764 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5767 free_percpu(dev
->pcpu_refcnt
);
5768 dev
->pcpu_refcnt
= NULL
;
5770 /* Compatibility with error handling in drivers */
5771 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5772 kfree((char *)dev
- dev
->padded
);
5776 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5777 dev
->reg_state
= NETREG_RELEASED
;
5779 /* will free via device release */
5780 put_device(&dev
->dev
);
5782 EXPORT_SYMBOL(free_netdev
);
5785 * synchronize_net - Synchronize with packet receive processing
5787 * Wait for packets currently being received to be done.
5788 * Does not block later packets from starting.
5790 void synchronize_net(void)
5795 EXPORT_SYMBOL(synchronize_net
);
5798 * unregister_netdevice_queue - remove device from the kernel
5802 * This function shuts down a device interface and removes it
5803 * from the kernel tables.
5804 * If head not NULL, device is queued to be unregistered later.
5806 * Callers must hold the rtnl semaphore. You may want
5807 * unregister_netdev() instead of this.
5810 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
5815 list_move_tail(&dev
->unreg_list
, head
);
5817 rollback_registered(dev
);
5818 /* Finish processing unregister after unlock */
5822 EXPORT_SYMBOL(unregister_netdevice_queue
);
5825 * unregister_netdevice_many - unregister many devices
5826 * @head: list of devices
5828 void unregister_netdevice_many(struct list_head
*head
)
5830 struct net_device
*dev
;
5832 if (!list_empty(head
)) {
5833 rollback_registered_many(head
);
5834 list_for_each_entry(dev
, head
, unreg_list
)
5838 EXPORT_SYMBOL(unregister_netdevice_many
);
5841 * unregister_netdev - remove device from the kernel
5844 * This function shuts down a device interface and removes it
5845 * from the kernel tables.
5847 * This is just a wrapper for unregister_netdevice that takes
5848 * the rtnl semaphore. In general you want to use this and not
5849 * unregister_netdevice.
5851 void unregister_netdev(struct net_device
*dev
)
5854 unregister_netdevice(dev
);
5857 EXPORT_SYMBOL(unregister_netdev
);
5860 * dev_change_net_namespace - move device to different nethost namespace
5862 * @net: network namespace
5863 * @pat: If not NULL name pattern to try if the current device name
5864 * is already taken in the destination network namespace.
5866 * This function shuts down a device interface and moves it
5867 * to a new network namespace. On success 0 is returned, on
5868 * a failure a netagive errno code is returned.
5870 * Callers must hold the rtnl semaphore.
5873 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
5879 /* Don't allow namespace local devices to be moved. */
5881 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
5884 /* Ensure the device has been registrered */
5886 if (dev
->reg_state
!= NETREG_REGISTERED
)
5889 /* Get out if there is nothing todo */
5891 if (net_eq(dev_net(dev
), net
))
5894 /* Pick the destination device name, and ensure
5895 * we can use it in the destination network namespace.
5898 if (__dev_get_by_name(net
, dev
->name
)) {
5899 /* We get here if we can't use the current device name */
5902 if (dev_get_valid_name(dev
, pat
, 1))
5907 * And now a mini version of register_netdevice unregister_netdevice.
5910 /* If device is running close it first. */
5913 /* And unlink it from device chain */
5915 unlist_netdevice(dev
);
5919 /* Shutdown queueing discipline. */
5922 /* Notify protocols, that we are about to destroy
5923 this device. They should clean all the things.
5925 Note that dev->reg_state stays at NETREG_REGISTERED.
5926 This is wanted because this way 8021q and macvlan know
5927 the device is just moving and can keep their slaves up.
5929 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5930 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5933 * Flush the unicast and multicast chains
5938 /* Actually switch the network namespace */
5939 dev_net_set(dev
, net
);
5941 /* If there is an ifindex conflict assign a new one */
5942 if (__dev_get_by_index(net
, dev
->ifindex
)) {
5943 int iflink
= (dev
->iflink
== dev
->ifindex
);
5944 dev
->ifindex
= dev_new_index(net
);
5946 dev
->iflink
= dev
->ifindex
;
5949 /* Fixup kobjects */
5950 err
= device_rename(&dev
->dev
, dev
->name
);
5953 /* Add the device back in the hashes */
5954 list_netdevice(dev
);
5956 /* Notify protocols, that a new device appeared. */
5957 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5960 * Prevent userspace races by waiting until the network
5961 * device is fully setup before sending notifications.
5963 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5970 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
5972 static int dev_cpu_callback(struct notifier_block
*nfb
,
5973 unsigned long action
,
5976 struct sk_buff
**list_skb
;
5977 struct sk_buff
*skb
;
5978 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
5979 struct softnet_data
*sd
, *oldsd
;
5981 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
5984 local_irq_disable();
5985 cpu
= smp_processor_id();
5986 sd
= &per_cpu(softnet_data
, cpu
);
5987 oldsd
= &per_cpu(softnet_data
, oldcpu
);
5989 /* Find end of our completion_queue. */
5990 list_skb
= &sd
->completion_queue
;
5992 list_skb
= &(*list_skb
)->next
;
5993 /* Append completion queue from offline CPU. */
5994 *list_skb
= oldsd
->completion_queue
;
5995 oldsd
->completion_queue
= NULL
;
5997 /* Append output queue from offline CPU. */
5998 if (oldsd
->output_queue
) {
5999 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6000 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6001 oldsd
->output_queue
= NULL
;
6002 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6005 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6008 /* Process offline CPU's input_pkt_queue */
6009 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6011 input_queue_head_incr(oldsd
);
6013 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6015 input_queue_head_incr(oldsd
);
6023 * netdev_increment_features - increment feature set by one
6024 * @all: current feature set
6025 * @one: new feature set
6026 * @mask: mask feature set
6028 * Computes a new feature set after adding a device with feature set
6029 * @one to the master device with current feature set @all. Will not
6030 * enable anything that is off in @mask. Returns the new feature set.
6032 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
6035 /* If device needs checksumming, downgrade to it. */
6036 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
6037 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
6038 else if (mask
& NETIF_F_ALL_CSUM
) {
6039 /* If one device supports v4/v6 checksumming, set for all. */
6040 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
6041 !(all
& NETIF_F_GEN_CSUM
)) {
6042 all
&= ~NETIF_F_ALL_CSUM
;
6043 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
6046 /* If one device supports hw checksumming, set for all. */
6047 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
6048 all
&= ~NETIF_F_ALL_CSUM
;
6049 all
|= NETIF_F_HW_CSUM
;
6053 one
|= NETIF_F_ALL_CSUM
;
6055 one
|= all
& NETIF_F_ONE_FOR_ALL
;
6056 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
| NETIF_F_UFO
;
6057 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
6061 EXPORT_SYMBOL(netdev_increment_features
);
6063 static struct hlist_head
*netdev_create_hash(void)
6066 struct hlist_head
*hash
;
6068 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6070 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6071 INIT_HLIST_HEAD(&hash
[i
]);
6076 /* Initialize per network namespace state */
6077 static int __net_init
netdev_init(struct net
*net
)
6079 INIT_LIST_HEAD(&net
->dev_base_head
);
6081 net
->dev_name_head
= netdev_create_hash();
6082 if (net
->dev_name_head
== NULL
)
6085 net
->dev_index_head
= netdev_create_hash();
6086 if (net
->dev_index_head
== NULL
)
6092 kfree(net
->dev_name_head
);
6098 * netdev_drivername - network driver for the device
6099 * @dev: network device
6100 * @buffer: buffer for resulting name
6101 * @len: size of buffer
6103 * Determine network driver for device.
6105 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
6107 const struct device_driver
*driver
;
6108 const struct device
*parent
;
6110 if (len
<= 0 || !buffer
)
6114 parent
= dev
->dev
.parent
;
6119 driver
= parent
->driver
;
6120 if (driver
&& driver
->name
)
6121 strlcpy(buffer
, driver
->name
, len
);
6125 static int __netdev_printk(const char *level
, const struct net_device
*dev
,
6126 struct va_format
*vaf
)
6130 if (dev
&& dev
->dev
.parent
)
6131 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6132 netdev_name(dev
), vaf
);
6134 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6136 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6141 int netdev_printk(const char *level
, const struct net_device
*dev
,
6142 const char *format
, ...)
6144 struct va_format vaf
;
6148 va_start(args
, format
);
6153 r
= __netdev_printk(level
, dev
, &vaf
);
6158 EXPORT_SYMBOL(netdev_printk
);
6160 #define define_netdev_printk_level(func, level) \
6161 int func(const struct net_device *dev, const char *fmt, ...) \
6164 struct va_format vaf; \
6167 va_start(args, fmt); \
6172 r = __netdev_printk(level, dev, &vaf); \
6177 EXPORT_SYMBOL(func);
6179 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6180 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6181 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6182 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6183 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6184 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6185 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6187 static void __net_exit
netdev_exit(struct net
*net
)
6189 kfree(net
->dev_name_head
);
6190 kfree(net
->dev_index_head
);
6193 static struct pernet_operations __net_initdata netdev_net_ops
= {
6194 .init
= netdev_init
,
6195 .exit
= netdev_exit
,
6198 static void __net_exit
default_device_exit(struct net
*net
)
6200 struct net_device
*dev
, *aux
;
6202 * Push all migratable network devices back to the
6203 * initial network namespace
6206 for_each_netdev_safe(net
, dev
, aux
) {
6208 char fb_name
[IFNAMSIZ
];
6210 /* Ignore unmoveable devices (i.e. loopback) */
6211 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6214 /* Leave virtual devices for the generic cleanup */
6215 if (dev
->rtnl_link_ops
)
6218 /* Push remaing network devices to init_net */
6219 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6220 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6222 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
6223 __func__
, dev
->name
, err
);
6230 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6232 /* At exit all network devices most be removed from a network
6233 * namespace. Do this in the reverse order of registeration.
6234 * Do this across as many network namespaces as possible to
6235 * improve batching efficiency.
6237 struct net_device
*dev
;
6239 LIST_HEAD(dev_kill_list
);
6242 list_for_each_entry(net
, net_list
, exit_list
) {
6243 for_each_netdev_reverse(net
, dev
) {
6244 if (dev
->rtnl_link_ops
)
6245 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6247 unregister_netdevice_queue(dev
, &dev_kill_list
);
6250 unregister_netdevice_many(&dev_kill_list
);
6254 static struct pernet_operations __net_initdata default_device_ops
= {
6255 .exit
= default_device_exit
,
6256 .exit_batch
= default_device_exit_batch
,
6260 * Initialize the DEV module. At boot time this walks the device list and
6261 * unhooks any devices that fail to initialise (normally hardware not
6262 * present) and leaves us with a valid list of present and active devices.
6267 * This is called single threaded during boot, so no need
6268 * to take the rtnl semaphore.
6270 static int __init
net_dev_init(void)
6272 int i
, rc
= -ENOMEM
;
6274 BUG_ON(!dev_boot_phase
);
6276 if (dev_proc_init())
6279 if (netdev_kobject_init())
6282 INIT_LIST_HEAD(&ptype_all
);
6283 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6284 INIT_LIST_HEAD(&ptype_base
[i
]);
6286 if (register_pernet_subsys(&netdev_net_ops
))
6290 * Initialise the packet receive queues.
6293 for_each_possible_cpu(i
) {
6294 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6296 memset(sd
, 0, sizeof(*sd
));
6297 skb_queue_head_init(&sd
->input_pkt_queue
);
6298 skb_queue_head_init(&sd
->process_queue
);
6299 sd
->completion_queue
= NULL
;
6300 INIT_LIST_HEAD(&sd
->poll_list
);
6301 sd
->output_queue
= NULL
;
6302 sd
->output_queue_tailp
= &sd
->output_queue
;
6304 sd
->csd
.func
= rps_trigger_softirq
;
6310 sd
->backlog
.poll
= process_backlog
;
6311 sd
->backlog
.weight
= weight_p
;
6312 sd
->backlog
.gro_list
= NULL
;
6313 sd
->backlog
.gro_count
= 0;
6318 /* The loopback device is special if any other network devices
6319 * is present in a network namespace the loopback device must
6320 * be present. Since we now dynamically allocate and free the
6321 * loopback device ensure this invariant is maintained by
6322 * keeping the loopback device as the first device on the
6323 * list of network devices. Ensuring the loopback devices
6324 * is the first device that appears and the last network device
6327 if (register_pernet_device(&loopback_net_ops
))
6330 if (register_pernet_device(&default_device_ops
))
6333 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6334 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6336 hotcpu_notifier(dev_cpu_callback
, 0);
6344 subsys_initcall(net_dev_init
);
6346 static int __init
initialize_hashrnd(void)
6348 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6352 late_initcall_sync(initialize_hashrnd
);