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/sched.h>
83 #include <linux/mutex.h>
84 #include <linux/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/ethtool.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/net_namespace.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/stat.h>
102 #include <linux/if_bridge.h>
103 #include <linux/if_macvlan.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
130 #include "net-sysfs.h"
133 * The list of packet types we will receive (as opposed to discard)
134 * and the routines to invoke.
136 * Why 16. Because with 16 the only overlap we get on a hash of the
137 * low nibble of the protocol value is RARP/SNAP/X.25.
139 * NOTE: That is no longer true with the addition of VLAN tags. Not
140 * sure which should go first, but I bet it won't make much
141 * difference if we are running VLANs. The good news is that
142 * this protocol won't be in the list unless compiled in, so
143 * the average user (w/out VLANs) will not be adversely affected.
160 #define PTYPE_HASH_SIZE (16)
161 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
163 static DEFINE_SPINLOCK(ptype_lock
);
164 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
165 static struct list_head ptype_all __read_mostly
; /* Taps */
167 #ifdef CONFIG_NET_DMA
169 struct dma_client client
;
171 cpumask_t channel_mask
;
172 struct dma_chan
**channels
;
175 static enum dma_state_client
176 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
177 enum dma_state state
);
179 static struct net_dma net_dma
= {
181 .event_callback
= netdev_dma_event
,
187 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
190 * Pure readers hold dev_base_lock for reading.
192 * Writers must hold the rtnl semaphore while they loop through the
193 * dev_base_head list, and hold dev_base_lock for writing when they do the
194 * actual updates. This allows pure readers to access the list even
195 * while a writer is preparing to update it.
197 * To put it another way, dev_base_lock is held for writing only to
198 * protect against pure readers; the rtnl semaphore provides the
199 * protection against other writers.
201 * See, for example usages, register_netdevice() and
202 * unregister_netdevice(), which must be called with the rtnl
205 DEFINE_RWLOCK(dev_base_lock
);
207 EXPORT_SYMBOL(dev_base_lock
);
209 #define NETDEV_HASHBITS 8
210 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
212 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
214 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
215 return &net
->dev_name_head
[hash
& ((1 << NETDEV_HASHBITS
) - 1)];
218 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
220 return &net
->dev_index_head
[ifindex
& ((1 << NETDEV_HASHBITS
) - 1)];
223 /* Device list insertion */
224 static int list_netdevice(struct net_device
*dev
)
226 struct net
*net
= dev_net(dev
);
230 write_lock_bh(&dev_base_lock
);
231 list_add_tail(&dev
->dev_list
, &net
->dev_base_head
);
232 hlist_add_head(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
233 hlist_add_head(&dev
->index_hlist
, dev_index_hash(net
, dev
->ifindex
));
234 write_unlock_bh(&dev_base_lock
);
238 /* Device list removal */
239 static void unlist_netdevice(struct net_device
*dev
)
243 /* Unlink dev from the device chain */
244 write_lock_bh(&dev_base_lock
);
245 list_del(&dev
->dev_list
);
246 hlist_del(&dev
->name_hlist
);
247 hlist_del(&dev
->index_hlist
);
248 write_unlock_bh(&dev_base_lock
);
255 static RAW_NOTIFIER_HEAD(netdev_chain
);
258 * Device drivers call our routines to queue packets here. We empty the
259 * queue in the local softnet handler.
262 DEFINE_PER_CPU(struct softnet_data
, softnet_data
);
264 #ifdef CONFIG_LOCKDEP
266 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
267 * according to dev->type
269 static const unsigned short netdev_lock_type
[] =
270 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
271 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
272 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
273 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
274 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
275 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
276 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
277 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
278 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
279 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
280 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
281 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
282 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
283 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_VOID
,
286 static const char *netdev_lock_name
[] =
287 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
288 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
289 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
290 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
291 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
292 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
293 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
294 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
295 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
296 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
297 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
298 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
299 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
300 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
303 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
304 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
306 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
310 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
311 if (netdev_lock_type
[i
] == dev_type
)
313 /* the last key is used by default */
314 return ARRAY_SIZE(netdev_lock_type
) - 1;
317 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
318 unsigned short dev_type
)
322 i
= netdev_lock_pos(dev_type
);
323 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
324 netdev_lock_name
[i
]);
327 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
331 i
= netdev_lock_pos(dev
->type
);
332 lockdep_set_class_and_name(&dev
->addr_list_lock
,
333 &netdev_addr_lock_key
[i
],
334 netdev_lock_name
[i
]);
337 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
338 unsigned short dev_type
)
341 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
346 /*******************************************************************************
348 Protocol management and registration routines
350 *******************************************************************************/
353 * Add a protocol ID to the list. Now that the input handler is
354 * smarter we can dispense with all the messy stuff that used to be
357 * BEWARE!!! Protocol handlers, mangling input packets,
358 * MUST BE last in hash buckets and checking protocol handlers
359 * MUST start from promiscuous ptype_all chain in net_bh.
360 * It is true now, do not change it.
361 * Explanation follows: if protocol handler, mangling packet, will
362 * be the first on list, it is not able to sense, that packet
363 * is cloned and should be copied-on-write, so that it will
364 * change it and subsequent readers will get broken packet.
369 * dev_add_pack - add packet handler
370 * @pt: packet type declaration
372 * Add a protocol handler to the networking stack. The passed &packet_type
373 * is linked into kernel lists and may not be freed until it has been
374 * removed from the kernel lists.
376 * This call does not sleep therefore it can not
377 * guarantee all CPU's that are in middle of receiving packets
378 * will see the new packet type (until the next received packet).
381 void dev_add_pack(struct packet_type
*pt
)
385 spin_lock_bh(&ptype_lock
);
386 if (pt
->type
== htons(ETH_P_ALL
))
387 list_add_rcu(&pt
->list
, &ptype_all
);
389 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
390 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
392 spin_unlock_bh(&ptype_lock
);
396 * __dev_remove_pack - remove packet handler
397 * @pt: packet type declaration
399 * Remove a protocol handler that was previously added to the kernel
400 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
401 * from the kernel lists and can be freed or reused once this function
404 * The packet type might still be in use by receivers
405 * and must not be freed until after all the CPU's have gone
406 * through a quiescent state.
408 void __dev_remove_pack(struct packet_type
*pt
)
410 struct list_head
*head
;
411 struct packet_type
*pt1
;
413 spin_lock_bh(&ptype_lock
);
415 if (pt
->type
== htons(ETH_P_ALL
))
418 head
= &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
420 list_for_each_entry(pt1
, head
, list
) {
422 list_del_rcu(&pt
->list
);
427 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
429 spin_unlock_bh(&ptype_lock
);
432 * dev_remove_pack - remove packet handler
433 * @pt: packet type declaration
435 * Remove a protocol handler that was previously added to the kernel
436 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
437 * from the kernel lists and can be freed or reused once this function
440 * This call sleeps to guarantee that no CPU is looking at the packet
443 void dev_remove_pack(struct packet_type
*pt
)
445 __dev_remove_pack(pt
);
450 /******************************************************************************
452 Device Boot-time Settings Routines
454 *******************************************************************************/
456 /* Boot time configuration table */
457 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
460 * netdev_boot_setup_add - add new setup entry
461 * @name: name of the device
462 * @map: configured settings for the device
464 * Adds new setup entry to the dev_boot_setup list. The function
465 * returns 0 on error and 1 on success. This is a generic routine to
468 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
470 struct netdev_boot_setup
*s
;
474 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
475 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
476 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
477 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
478 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
483 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
487 * netdev_boot_setup_check - check boot time settings
488 * @dev: the netdevice
490 * Check boot time settings for the device.
491 * The found settings are set for the device to be used
492 * later in the device probing.
493 * Returns 0 if no settings found, 1 if they are.
495 int netdev_boot_setup_check(struct net_device
*dev
)
497 struct netdev_boot_setup
*s
= dev_boot_setup
;
500 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
501 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
502 !strcmp(dev
->name
, s
[i
].name
)) {
503 dev
->irq
= s
[i
].map
.irq
;
504 dev
->base_addr
= s
[i
].map
.base_addr
;
505 dev
->mem_start
= s
[i
].map
.mem_start
;
506 dev
->mem_end
= s
[i
].map
.mem_end
;
515 * netdev_boot_base - get address from boot time settings
516 * @prefix: prefix for network device
517 * @unit: id for network device
519 * Check boot time settings for the base address of device.
520 * The found settings are set for the device to be used
521 * later in the device probing.
522 * Returns 0 if no settings found.
524 unsigned long netdev_boot_base(const char *prefix
, int unit
)
526 const struct netdev_boot_setup
*s
= dev_boot_setup
;
530 sprintf(name
, "%s%d", prefix
, unit
);
533 * If device already registered then return base of 1
534 * to indicate not to probe for this interface
536 if (__dev_get_by_name(&init_net
, name
))
539 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
540 if (!strcmp(name
, s
[i
].name
))
541 return s
[i
].map
.base_addr
;
546 * Saves at boot time configured settings for any netdevice.
548 int __init
netdev_boot_setup(char *str
)
553 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
558 memset(&map
, 0, sizeof(map
));
562 map
.base_addr
= ints
[2];
564 map
.mem_start
= ints
[3];
566 map
.mem_end
= ints
[4];
568 /* Add new entry to the list */
569 return netdev_boot_setup_add(str
, &map
);
572 __setup("netdev=", netdev_boot_setup
);
574 /*******************************************************************************
576 Device Interface Subroutines
578 *******************************************************************************/
581 * __dev_get_by_name - find a device by its name
582 * @net: the applicable net namespace
583 * @name: name to find
585 * Find an interface by name. Must be called under RTNL semaphore
586 * or @dev_base_lock. If the name is found a pointer to the device
587 * is returned. If the name is not found then %NULL is returned. The
588 * reference counters are not incremented so the caller must be
589 * careful with locks.
592 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
594 struct hlist_node
*p
;
596 hlist_for_each(p
, dev_name_hash(net
, name
)) {
597 struct net_device
*dev
598 = hlist_entry(p
, struct net_device
, name_hlist
);
599 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
606 * dev_get_by_name - find a device by its name
607 * @net: the applicable net namespace
608 * @name: name to find
610 * Find an interface by name. This can be called from any
611 * context and does its own locking. The returned handle has
612 * the usage count incremented and the caller must use dev_put() to
613 * release it when it is no longer needed. %NULL is returned if no
614 * matching device is found.
617 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
619 struct net_device
*dev
;
621 read_lock(&dev_base_lock
);
622 dev
= __dev_get_by_name(net
, name
);
625 read_unlock(&dev_base_lock
);
630 * __dev_get_by_index - find a device by its ifindex
631 * @net: the applicable net namespace
632 * @ifindex: index of device
634 * Search for an interface by index. Returns %NULL if the device
635 * is not found or a pointer to the device. The device has not
636 * had its reference counter increased so the caller must be careful
637 * about locking. The caller must hold either the RTNL semaphore
641 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
643 struct hlist_node
*p
;
645 hlist_for_each(p
, dev_index_hash(net
, ifindex
)) {
646 struct net_device
*dev
647 = hlist_entry(p
, struct net_device
, index_hlist
);
648 if (dev
->ifindex
== ifindex
)
656 * dev_get_by_index - find a device by its ifindex
657 * @net: the applicable net namespace
658 * @ifindex: index of device
660 * Search for an interface by index. Returns NULL if the device
661 * is not found or a pointer to the device. The device returned has
662 * had a reference added and the pointer is safe until the user calls
663 * dev_put to indicate they have finished with it.
666 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
668 struct net_device
*dev
;
670 read_lock(&dev_base_lock
);
671 dev
= __dev_get_by_index(net
, ifindex
);
674 read_unlock(&dev_base_lock
);
679 * dev_getbyhwaddr - find a device by its hardware address
680 * @net: the applicable net namespace
681 * @type: media type of device
682 * @ha: hardware address
684 * Search for an interface by MAC address. Returns NULL if the device
685 * is not found or a pointer to the device. The caller must hold the
686 * rtnl semaphore. The returned device has not had its ref count increased
687 * and the caller must therefore be careful about locking
690 * If the API was consistent this would be __dev_get_by_hwaddr
693 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
695 struct net_device
*dev
;
699 for_each_netdev(net
, dev
)
700 if (dev
->type
== type
&&
701 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
707 EXPORT_SYMBOL(dev_getbyhwaddr
);
709 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
711 struct net_device
*dev
;
714 for_each_netdev(net
, dev
)
715 if (dev
->type
== type
)
721 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
723 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
725 struct net_device
*dev
;
728 dev
= __dev_getfirstbyhwtype(net
, type
);
735 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
738 * dev_get_by_flags - find any device with given flags
739 * @net: the applicable net namespace
740 * @if_flags: IFF_* values
741 * @mask: bitmask of bits in if_flags to check
743 * Search for any interface with the given flags. Returns NULL if a device
744 * is not found or a pointer to the device. The device returned has
745 * had a reference added and the pointer is safe until the user calls
746 * dev_put to indicate they have finished with it.
749 struct net_device
* dev_get_by_flags(struct net
*net
, unsigned short if_flags
, unsigned short mask
)
751 struct net_device
*dev
, *ret
;
754 read_lock(&dev_base_lock
);
755 for_each_netdev(net
, dev
) {
756 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
762 read_unlock(&dev_base_lock
);
767 * dev_valid_name - check if name is okay for network device
770 * Network device names need to be valid file names to
771 * to allow sysfs to work. We also disallow any kind of
774 int dev_valid_name(const char *name
)
778 if (strlen(name
) >= IFNAMSIZ
)
780 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
784 if (*name
== '/' || isspace(*name
))
792 * __dev_alloc_name - allocate a name for a device
793 * @net: network namespace to allocate the device name in
794 * @name: name format string
795 * @buf: scratch buffer and result name string
797 * Passed a format string - eg "lt%d" it will try and find a suitable
798 * id. It scans list of devices to build up a free map, then chooses
799 * the first empty slot. The caller must hold the dev_base or rtnl lock
800 * while allocating the name and adding the device in order to avoid
802 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
803 * Returns the number of the unit assigned or a negative errno code.
806 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
810 const int max_netdevices
= 8*PAGE_SIZE
;
811 unsigned long *inuse
;
812 struct net_device
*d
;
814 p
= strnchr(name
, IFNAMSIZ
-1, '%');
817 * Verify the string as this thing may have come from
818 * the user. There must be either one "%d" and no other "%"
821 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
824 /* Use one page as a bit array of possible slots */
825 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
829 for_each_netdev(net
, d
) {
830 if (!sscanf(d
->name
, name
, &i
))
832 if (i
< 0 || i
>= max_netdevices
)
835 /* avoid cases where sscanf is not exact inverse of printf */
836 snprintf(buf
, IFNAMSIZ
, name
, i
);
837 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
841 i
= find_first_zero_bit(inuse
, max_netdevices
);
842 free_page((unsigned long) inuse
);
845 snprintf(buf
, IFNAMSIZ
, name
, i
);
846 if (!__dev_get_by_name(net
, buf
))
849 /* It is possible to run out of possible slots
850 * when the name is long and there isn't enough space left
851 * for the digits, or if all bits are used.
857 * dev_alloc_name - allocate a name for a device
859 * @name: name format string
861 * Passed a format string - eg "lt%d" it will try and find a suitable
862 * id. It scans list of devices to build up a free map, then chooses
863 * the first empty slot. The caller must hold the dev_base or rtnl lock
864 * while allocating the name and adding the device in order to avoid
866 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
867 * Returns the number of the unit assigned or a negative errno code.
870 int dev_alloc_name(struct net_device
*dev
, const char *name
)
876 BUG_ON(!dev_net(dev
));
878 ret
= __dev_alloc_name(net
, name
, buf
);
880 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
886 * dev_change_name - change name of a device
888 * @newname: name (or format string) must be at least IFNAMSIZ
890 * Change name of a device, can pass format strings "eth%d".
893 int dev_change_name(struct net_device
*dev
, const char *newname
)
895 char oldname
[IFNAMSIZ
];
901 BUG_ON(!dev_net(dev
));
904 if (dev
->flags
& IFF_UP
)
907 if (!dev_valid_name(newname
))
910 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
913 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
915 if (strchr(newname
, '%')) {
916 err
= dev_alloc_name(dev
, newname
);
920 else if (__dev_get_by_name(net
, newname
))
923 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
926 /* For now only devices in the initial network namespace
929 if (net
== &init_net
) {
930 ret
= device_rename(&dev
->dev
, dev
->name
);
932 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
937 write_lock_bh(&dev_base_lock
);
938 hlist_del(&dev
->name_hlist
);
939 hlist_add_head(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
940 write_unlock_bh(&dev_base_lock
);
942 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
943 ret
= notifier_to_errno(ret
);
948 "%s: name change rollback failed: %d.\n",
952 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
961 * dev_set_alias - change ifalias of a device
963 * @alias: name up to IFALIASZ
964 * @len: limit of bytes to copy from info
966 * Set ifalias for a device,
968 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
983 dev
->ifalias
= krealloc(dev
->ifalias
, len
+1, GFP_KERNEL
);
987 strlcpy(dev
->ifalias
, alias
, len
+1);
993 * netdev_features_change - device changes features
994 * @dev: device to cause notification
996 * Called to indicate a device has changed features.
998 void netdev_features_change(struct net_device
*dev
)
1000 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1002 EXPORT_SYMBOL(netdev_features_change
);
1005 * netdev_state_change - device changes state
1006 * @dev: device to cause notification
1008 * Called to indicate a device has changed state. This function calls
1009 * the notifier chains for netdev_chain and sends a NEWLINK message
1010 * to the routing socket.
1012 void netdev_state_change(struct net_device
*dev
)
1014 if (dev
->flags
& IFF_UP
) {
1015 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1016 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1020 void netdev_bonding_change(struct net_device
*dev
)
1022 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, dev
);
1024 EXPORT_SYMBOL(netdev_bonding_change
);
1027 * dev_load - load a network module
1028 * @net: the applicable net namespace
1029 * @name: name of interface
1031 * If a network interface is not present and the process has suitable
1032 * privileges this function loads the module. If module loading is not
1033 * available in this kernel then it becomes a nop.
1036 void dev_load(struct net
*net
, const char *name
)
1038 struct net_device
*dev
;
1040 read_lock(&dev_base_lock
);
1041 dev
= __dev_get_by_name(net
, name
);
1042 read_unlock(&dev_base_lock
);
1044 if (!dev
&& capable(CAP_SYS_MODULE
))
1045 request_module("%s", name
);
1049 * dev_open - prepare an interface for use.
1050 * @dev: device to open
1052 * Takes a device from down to up state. The device's private open
1053 * function is invoked and then the multicast lists are loaded. Finally
1054 * the device is moved into the up state and a %NETDEV_UP message is
1055 * sent to the netdev notifier chain.
1057 * Calling this function on an active interface is a nop. On a failure
1058 * a negative errno code is returned.
1060 int dev_open(struct net_device
*dev
)
1062 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1071 if (dev
->flags
& IFF_UP
)
1075 * Is it even present?
1077 if (!netif_device_present(dev
))
1081 * Call device private open method
1083 set_bit(__LINK_STATE_START
, &dev
->state
);
1085 if (ops
->ndo_validate_addr
)
1086 ret
= ops
->ndo_validate_addr(dev
);
1088 if (!ret
&& ops
->ndo_open
)
1089 ret
= ops
->ndo_open(dev
);
1092 * If it went open OK then:
1096 clear_bit(__LINK_STATE_START
, &dev
->state
);
1101 dev
->flags
|= IFF_UP
;
1104 * Initialize multicasting status
1106 dev_set_rx_mode(dev
);
1109 * Wakeup transmit queue engine
1114 * ... and announce new interface.
1116 call_netdevice_notifiers(NETDEV_UP
, dev
);
1123 * dev_close - shutdown an interface.
1124 * @dev: device to shutdown
1126 * This function moves an active device into down state. A
1127 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1128 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1131 int dev_close(struct net_device
*dev
)
1133 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1138 if (!(dev
->flags
& IFF_UP
))
1142 * Tell people we are going down, so that they can
1143 * prepare to death, when device is still operating.
1145 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1147 clear_bit(__LINK_STATE_START
, &dev
->state
);
1149 /* Synchronize to scheduled poll. We cannot touch poll list,
1150 * it can be even on different cpu. So just clear netif_running().
1152 * dev->stop() will invoke napi_disable() on all of it's
1153 * napi_struct instances on this device.
1155 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1157 dev_deactivate(dev
);
1160 * Call the device specific close. This cannot fail.
1161 * Only if device is UP
1163 * We allow it to be called even after a DETACH hot-plug
1170 * Device is now down.
1173 dev
->flags
&= ~IFF_UP
;
1176 * Tell people we are down
1178 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1185 * dev_disable_lro - disable Large Receive Offload on a device
1188 * Disable Large Receive Offload (LRO) on a net device. Must be
1189 * called under RTNL. This is needed if received packets may be
1190 * forwarded to another interface.
1192 void dev_disable_lro(struct net_device
*dev
)
1194 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1195 dev
->ethtool_ops
->set_flags
) {
1196 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1197 if (flags
& ETH_FLAG_LRO
) {
1198 flags
&= ~ETH_FLAG_LRO
;
1199 dev
->ethtool_ops
->set_flags(dev
, flags
);
1202 WARN_ON(dev
->features
& NETIF_F_LRO
);
1204 EXPORT_SYMBOL(dev_disable_lro
);
1207 static int dev_boot_phase
= 1;
1210 * Device change register/unregister. These are not inline or static
1211 * as we export them to the world.
1215 * register_netdevice_notifier - register a network notifier block
1218 * Register a notifier to be called when network device events occur.
1219 * The notifier passed is linked into the kernel structures and must
1220 * not be reused until it has been unregistered. A negative errno code
1221 * is returned on a failure.
1223 * When registered all registration and up events are replayed
1224 * to the new notifier to allow device to have a race free
1225 * view of the network device list.
1228 int register_netdevice_notifier(struct notifier_block
*nb
)
1230 struct net_device
*dev
;
1231 struct net_device
*last
;
1236 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1242 for_each_netdev(net
, dev
) {
1243 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1244 err
= notifier_to_errno(err
);
1248 if (!(dev
->flags
& IFF_UP
))
1251 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1262 for_each_netdev(net
, dev
) {
1266 if (dev
->flags
& IFF_UP
) {
1267 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1268 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1270 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1274 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1279 * unregister_netdevice_notifier - unregister a network notifier block
1282 * Unregister a notifier previously registered by
1283 * register_netdevice_notifier(). The notifier is unlinked into the
1284 * kernel structures and may then be reused. A negative errno code
1285 * is returned on a failure.
1288 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1293 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1299 * call_netdevice_notifiers - call all network notifier blocks
1300 * @val: value passed unmodified to notifier function
1301 * @dev: net_device pointer passed unmodified to notifier function
1303 * Call all network notifier blocks. Parameters and return value
1304 * are as for raw_notifier_call_chain().
1307 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1309 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1312 /* When > 0 there are consumers of rx skb time stamps */
1313 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1315 void net_enable_timestamp(void)
1317 atomic_inc(&netstamp_needed
);
1320 void net_disable_timestamp(void)
1322 atomic_dec(&netstamp_needed
);
1325 static inline void net_timestamp(struct sk_buff
*skb
)
1327 if (atomic_read(&netstamp_needed
))
1328 __net_timestamp(skb
);
1330 skb
->tstamp
.tv64
= 0;
1334 * Support routine. Sends outgoing frames to any network
1335 * taps currently in use.
1338 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1340 struct packet_type
*ptype
;
1345 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1346 /* Never send packets back to the socket
1347 * they originated from - MvS (miquels@drinkel.ow.org)
1349 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1350 (ptype
->af_packet_priv
== NULL
||
1351 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1352 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1356 /* skb->nh should be correctly
1357 set by sender, so that the second statement is
1358 just protection against buggy protocols.
1360 skb_reset_mac_header(skb2
);
1362 if (skb_network_header(skb2
) < skb2
->data
||
1363 skb2
->network_header
> skb2
->tail
) {
1364 if (net_ratelimit())
1365 printk(KERN_CRIT
"protocol %04x is "
1367 skb2
->protocol
, dev
->name
);
1368 skb_reset_network_header(skb2
);
1371 skb2
->transport_header
= skb2
->network_header
;
1372 skb2
->pkt_type
= PACKET_OUTGOING
;
1373 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1380 static inline void __netif_reschedule(struct Qdisc
*q
)
1382 struct softnet_data
*sd
;
1383 unsigned long flags
;
1385 local_irq_save(flags
);
1386 sd
= &__get_cpu_var(softnet_data
);
1387 q
->next_sched
= sd
->output_queue
;
1388 sd
->output_queue
= q
;
1389 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1390 local_irq_restore(flags
);
1393 void __netif_schedule(struct Qdisc
*q
)
1395 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1396 __netif_reschedule(q
);
1398 EXPORT_SYMBOL(__netif_schedule
);
1400 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1402 if (atomic_dec_and_test(&skb
->users
)) {
1403 struct softnet_data
*sd
;
1404 unsigned long flags
;
1406 local_irq_save(flags
);
1407 sd
= &__get_cpu_var(softnet_data
);
1408 skb
->next
= sd
->completion_queue
;
1409 sd
->completion_queue
= skb
;
1410 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1411 local_irq_restore(flags
);
1414 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1416 void dev_kfree_skb_any(struct sk_buff
*skb
)
1418 if (in_irq() || irqs_disabled())
1419 dev_kfree_skb_irq(skb
);
1423 EXPORT_SYMBOL(dev_kfree_skb_any
);
1427 * netif_device_detach - mark device as removed
1428 * @dev: network device
1430 * Mark device as removed from system and therefore no longer available.
1432 void netif_device_detach(struct net_device
*dev
)
1434 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1435 netif_running(dev
)) {
1436 netif_stop_queue(dev
);
1439 EXPORT_SYMBOL(netif_device_detach
);
1442 * netif_device_attach - mark device as attached
1443 * @dev: network device
1445 * Mark device as attached from system and restart if needed.
1447 void netif_device_attach(struct net_device
*dev
)
1449 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1450 netif_running(dev
)) {
1451 netif_wake_queue(dev
);
1452 __netdev_watchdog_up(dev
);
1455 EXPORT_SYMBOL(netif_device_attach
);
1457 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1459 return ((features
& NETIF_F_GEN_CSUM
) ||
1460 ((features
& NETIF_F_IP_CSUM
) &&
1461 protocol
== htons(ETH_P_IP
)) ||
1462 ((features
& NETIF_F_IPV6_CSUM
) &&
1463 protocol
== htons(ETH_P_IPV6
)));
1466 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1468 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1471 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1472 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1473 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1474 veh
->h_vlan_encapsulated_proto
))
1482 * Invalidate hardware checksum when packet is to be mangled, and
1483 * complete checksum manually on outgoing path.
1485 int skb_checksum_help(struct sk_buff
*skb
)
1488 int ret
= 0, offset
;
1490 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1491 goto out_set_summed
;
1493 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1494 /* Let GSO fix up the checksum. */
1495 goto out_set_summed
;
1498 offset
= skb
->csum_start
- skb_headroom(skb
);
1499 BUG_ON(offset
>= skb_headlen(skb
));
1500 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1502 offset
+= skb
->csum_offset
;
1503 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1505 if (skb_cloned(skb
) &&
1506 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1507 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1512 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1514 skb
->ip_summed
= CHECKSUM_NONE
;
1520 * skb_gso_segment - Perform segmentation on skb.
1521 * @skb: buffer to segment
1522 * @features: features for the output path (see dev->features)
1524 * This function segments the given skb and returns a list of segments.
1526 * It may return NULL if the skb requires no segmentation. This is
1527 * only possible when GSO is used for verifying header integrity.
1529 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1531 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1532 struct packet_type
*ptype
;
1533 __be16 type
= skb
->protocol
;
1536 BUG_ON(skb_shinfo(skb
)->frag_list
);
1538 skb_reset_mac_header(skb
);
1539 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1540 __skb_pull(skb
, skb
->mac_len
);
1542 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1543 if (skb_header_cloned(skb
) &&
1544 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1545 return ERR_PTR(err
);
1549 list_for_each_entry_rcu(ptype
,
1550 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1551 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1552 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1553 err
= ptype
->gso_send_check(skb
);
1554 segs
= ERR_PTR(err
);
1555 if (err
|| skb_gso_ok(skb
, features
))
1557 __skb_push(skb
, (skb
->data
-
1558 skb_network_header(skb
)));
1560 segs
= ptype
->gso_segment(skb
, features
);
1566 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1571 EXPORT_SYMBOL(skb_gso_segment
);
1573 /* Take action when hardware reception checksum errors are detected. */
1575 void netdev_rx_csum_fault(struct net_device
*dev
)
1577 if (net_ratelimit()) {
1578 printk(KERN_ERR
"%s: hw csum failure.\n",
1579 dev
? dev
->name
: "<unknown>");
1583 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1586 /* Actually, we should eliminate this check as soon as we know, that:
1587 * 1. IOMMU is present and allows to map all the memory.
1588 * 2. No high memory really exists on this machine.
1591 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1593 #ifdef CONFIG_HIGHMEM
1596 if (dev
->features
& NETIF_F_HIGHDMA
)
1599 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1600 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1608 void (*destructor
)(struct sk_buff
*skb
);
1611 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1613 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1615 struct dev_gso_cb
*cb
;
1618 struct sk_buff
*nskb
= skb
->next
;
1620 skb
->next
= nskb
->next
;
1623 } while (skb
->next
);
1625 cb
= DEV_GSO_CB(skb
);
1627 cb
->destructor(skb
);
1631 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1632 * @skb: buffer to segment
1634 * This function segments the given skb and stores the list of segments
1637 static int dev_gso_segment(struct sk_buff
*skb
)
1639 struct net_device
*dev
= skb
->dev
;
1640 struct sk_buff
*segs
;
1641 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1644 segs
= skb_gso_segment(skb
, features
);
1646 /* Verifying header integrity only. */
1651 return PTR_ERR(segs
);
1654 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1655 skb
->destructor
= dev_gso_skb_destructor
;
1660 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1661 struct netdev_queue
*txq
)
1663 if (likely(!skb
->next
)) {
1664 if (!list_empty(&ptype_all
))
1665 dev_queue_xmit_nit(skb
, dev
);
1667 if (netif_needs_gso(dev
, skb
)) {
1668 if (unlikely(dev_gso_segment(skb
)))
1674 return dev
->hard_start_xmit(skb
, dev
);
1679 struct sk_buff
*nskb
= skb
->next
;
1682 skb
->next
= nskb
->next
;
1684 rc
= dev
->hard_start_xmit(nskb
, dev
);
1686 nskb
->next
= skb
->next
;
1690 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
1691 return NETDEV_TX_BUSY
;
1692 } while (skb
->next
);
1694 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1701 static u32 simple_tx_hashrnd
;
1702 static int simple_tx_hashrnd_initialized
= 0;
1704 static u16
simple_tx_hash(struct net_device
*dev
, struct sk_buff
*skb
)
1706 u32 addr1
, addr2
, ports
;
1710 if (unlikely(!simple_tx_hashrnd_initialized
)) {
1711 get_random_bytes(&simple_tx_hashrnd
, 4);
1712 simple_tx_hashrnd_initialized
= 1;
1715 switch (skb
->protocol
) {
1716 case htons(ETH_P_IP
):
1717 if (!(ip_hdr(skb
)->frag_off
& htons(IP_MF
| IP_OFFSET
)))
1718 ip_proto
= ip_hdr(skb
)->protocol
;
1719 addr1
= ip_hdr(skb
)->saddr
;
1720 addr2
= ip_hdr(skb
)->daddr
;
1721 ihl
= ip_hdr(skb
)->ihl
;
1723 case htons(ETH_P_IPV6
):
1724 ip_proto
= ipv6_hdr(skb
)->nexthdr
;
1725 addr1
= ipv6_hdr(skb
)->saddr
.s6_addr32
[3];
1726 addr2
= ipv6_hdr(skb
)->daddr
.s6_addr32
[3];
1741 case IPPROTO_UDPLITE
:
1742 ports
= *((u32
*) (skb_network_header(skb
) + (ihl
* 4)));
1750 hash
= jhash_3words(addr1
, addr2
, ports
, simple_tx_hashrnd
);
1752 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
1755 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
1756 struct sk_buff
*skb
)
1758 u16 queue_index
= 0;
1760 if (dev
->select_queue
)
1761 queue_index
= dev
->select_queue(dev
, skb
);
1762 else if (dev
->real_num_tx_queues
> 1)
1763 queue_index
= simple_tx_hash(dev
, skb
);
1765 skb_set_queue_mapping(skb
, queue_index
);
1766 return netdev_get_tx_queue(dev
, queue_index
);
1770 * dev_queue_xmit - transmit a buffer
1771 * @skb: buffer to transmit
1773 * Queue a buffer for transmission to a network device. The caller must
1774 * have set the device and priority and built the buffer before calling
1775 * this function. The function can be called from an interrupt.
1777 * A negative errno code is returned on a failure. A success does not
1778 * guarantee the frame will be transmitted as it may be dropped due
1779 * to congestion or traffic shaping.
1781 * -----------------------------------------------------------------------------------
1782 * I notice this method can also return errors from the queue disciplines,
1783 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1786 * Regardless of the return value, the skb is consumed, so it is currently
1787 * difficult to retry a send to this method. (You can bump the ref count
1788 * before sending to hold a reference for retry if you are careful.)
1790 * When calling this method, interrupts MUST be enabled. This is because
1791 * the BH enable code must have IRQs enabled so that it will not deadlock.
1794 int dev_queue_xmit(struct sk_buff
*skb
)
1796 struct net_device
*dev
= skb
->dev
;
1797 struct netdev_queue
*txq
;
1801 /* GSO will handle the following emulations directly. */
1802 if (netif_needs_gso(dev
, skb
))
1805 if (skb_shinfo(skb
)->frag_list
&&
1806 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1807 __skb_linearize(skb
))
1810 /* Fragmented skb is linearized if device does not support SG,
1811 * or if at least one of fragments is in highmem and device
1812 * does not support DMA from it.
1814 if (skb_shinfo(skb
)->nr_frags
&&
1815 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1816 __skb_linearize(skb
))
1819 /* If packet is not checksummed and device does not support
1820 * checksumming for this protocol, complete checksumming here.
1822 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1823 skb_set_transport_header(skb
, skb
->csum_start
-
1825 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
1830 /* Disable soft irqs for various locks below. Also
1831 * stops preemption for RCU.
1835 txq
= dev_pick_tx(dev
, skb
);
1836 q
= rcu_dereference(txq
->qdisc
);
1838 #ifdef CONFIG_NET_CLS_ACT
1839 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1842 spinlock_t
*root_lock
= qdisc_lock(q
);
1844 spin_lock(root_lock
);
1846 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
1850 rc
= qdisc_enqueue_root(skb
, q
);
1853 spin_unlock(root_lock
);
1858 /* The device has no queue. Common case for software devices:
1859 loopback, all the sorts of tunnels...
1861 Really, it is unlikely that netif_tx_lock protection is necessary
1862 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1864 However, it is possible, that they rely on protection
1867 Check this and shot the lock. It is not prone from deadlocks.
1868 Either shot noqueue qdisc, it is even simpler 8)
1870 if (dev
->flags
& IFF_UP
) {
1871 int cpu
= smp_processor_id(); /* ok because BHs are off */
1873 if (txq
->xmit_lock_owner
!= cpu
) {
1875 HARD_TX_LOCK(dev
, txq
, cpu
);
1877 if (!netif_tx_queue_stopped(txq
)) {
1879 if (!dev_hard_start_xmit(skb
, dev
, txq
)) {
1880 HARD_TX_UNLOCK(dev
, txq
);
1884 HARD_TX_UNLOCK(dev
, txq
);
1885 if (net_ratelimit())
1886 printk(KERN_CRIT
"Virtual device %s asks to "
1887 "queue packet!\n", dev
->name
);
1889 /* Recursion is detected! It is possible,
1891 if (net_ratelimit())
1892 printk(KERN_CRIT
"Dead loop on virtual device "
1893 "%s, fix it urgently!\n", dev
->name
);
1898 rcu_read_unlock_bh();
1904 rcu_read_unlock_bh();
1909 /*=======================================================================
1911 =======================================================================*/
1913 int netdev_max_backlog __read_mostly
= 1000;
1914 int netdev_budget __read_mostly
= 300;
1915 int weight_p __read_mostly
= 64; /* old backlog weight */
1917 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1921 * netif_rx - post buffer to the network code
1922 * @skb: buffer to post
1924 * This function receives a packet from a device driver and queues it for
1925 * the upper (protocol) levels to process. It always succeeds. The buffer
1926 * may be dropped during processing for congestion control or by the
1930 * NET_RX_SUCCESS (no congestion)
1931 * NET_RX_DROP (packet was dropped)
1935 int netif_rx(struct sk_buff
*skb
)
1937 struct softnet_data
*queue
;
1938 unsigned long flags
;
1940 /* if netpoll wants it, pretend we never saw it */
1941 if (netpoll_rx(skb
))
1944 if (!skb
->tstamp
.tv64
)
1948 * The code is rearranged so that the path is the most
1949 * short when CPU is congested, but is still operating.
1951 local_irq_save(flags
);
1952 queue
= &__get_cpu_var(softnet_data
);
1954 __get_cpu_var(netdev_rx_stat
).total
++;
1955 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1956 if (queue
->input_pkt_queue
.qlen
) {
1958 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1959 local_irq_restore(flags
);
1960 return NET_RX_SUCCESS
;
1963 napi_schedule(&queue
->backlog
);
1967 __get_cpu_var(netdev_rx_stat
).dropped
++;
1968 local_irq_restore(flags
);
1974 int netif_rx_ni(struct sk_buff
*skb
)
1979 err
= netif_rx(skb
);
1980 if (local_softirq_pending())
1987 EXPORT_SYMBOL(netif_rx_ni
);
1989 static void net_tx_action(struct softirq_action
*h
)
1991 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1993 if (sd
->completion_queue
) {
1994 struct sk_buff
*clist
;
1996 local_irq_disable();
1997 clist
= sd
->completion_queue
;
1998 sd
->completion_queue
= NULL
;
2002 struct sk_buff
*skb
= clist
;
2003 clist
= clist
->next
;
2005 WARN_ON(atomic_read(&skb
->users
));
2010 if (sd
->output_queue
) {
2013 local_irq_disable();
2014 head
= sd
->output_queue
;
2015 sd
->output_queue
= NULL
;
2019 struct Qdisc
*q
= head
;
2020 spinlock_t
*root_lock
;
2022 head
= head
->next_sched
;
2024 root_lock
= qdisc_lock(q
);
2025 if (spin_trylock(root_lock
)) {
2026 smp_mb__before_clear_bit();
2027 clear_bit(__QDISC_STATE_SCHED
,
2030 spin_unlock(root_lock
);
2032 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2034 __netif_reschedule(q
);
2036 smp_mb__before_clear_bit();
2037 clear_bit(__QDISC_STATE_SCHED
,
2045 static inline int deliver_skb(struct sk_buff
*skb
,
2046 struct packet_type
*pt_prev
,
2047 struct net_device
*orig_dev
)
2049 atomic_inc(&skb
->users
);
2050 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2053 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2054 /* These hooks defined here for ATM */
2056 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
2057 unsigned char *addr
);
2058 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
2061 * If bridge module is loaded call bridging hook.
2062 * returns NULL if packet was consumed.
2064 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
2065 struct sk_buff
*skb
) __read_mostly
;
2066 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
2067 struct packet_type
**pt_prev
, int *ret
,
2068 struct net_device
*orig_dev
)
2070 struct net_bridge_port
*port
;
2072 if (skb
->pkt_type
== PACKET_LOOPBACK
||
2073 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
2077 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2081 return br_handle_frame_hook(port
, skb
);
2084 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2087 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2088 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
2089 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
2091 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
2092 struct packet_type
**pt_prev
,
2094 struct net_device
*orig_dev
)
2096 if (skb
->dev
->macvlan_port
== NULL
)
2100 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2103 return macvlan_handle_frame_hook(skb
);
2106 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2109 #ifdef CONFIG_NET_CLS_ACT
2110 /* TODO: Maybe we should just force sch_ingress to be compiled in
2111 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2112 * a compare and 2 stores extra right now if we dont have it on
2113 * but have CONFIG_NET_CLS_ACT
2114 * NOTE: This doesnt stop any functionality; if you dont have
2115 * the ingress scheduler, you just cant add policies on ingress.
2118 static int ing_filter(struct sk_buff
*skb
)
2120 struct net_device
*dev
= skb
->dev
;
2121 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2122 struct netdev_queue
*rxq
;
2123 int result
= TC_ACT_OK
;
2126 if (MAX_RED_LOOP
< ttl
++) {
2128 "Redir loop detected Dropping packet (%d->%d)\n",
2129 skb
->iif
, dev
->ifindex
);
2133 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2134 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2136 rxq
= &dev
->rx_queue
;
2139 if (q
!= &noop_qdisc
) {
2140 spin_lock(qdisc_lock(q
));
2141 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
2142 result
= qdisc_enqueue_root(skb
, q
);
2143 spin_unlock(qdisc_lock(q
));
2149 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2150 struct packet_type
**pt_prev
,
2151 int *ret
, struct net_device
*orig_dev
)
2153 if (skb
->dev
->rx_queue
.qdisc
== &noop_qdisc
)
2157 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2160 /* Huh? Why does turning on AF_PACKET affect this? */
2161 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2164 switch (ing_filter(skb
)) {
2178 * netif_nit_deliver - deliver received packets to network taps
2181 * This function is used to deliver incoming packets to network
2182 * taps. It should be used when the normal netif_receive_skb path
2183 * is bypassed, for example because of VLAN acceleration.
2185 void netif_nit_deliver(struct sk_buff
*skb
)
2187 struct packet_type
*ptype
;
2189 if (list_empty(&ptype_all
))
2192 skb_reset_network_header(skb
);
2193 skb_reset_transport_header(skb
);
2194 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2197 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2198 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2199 deliver_skb(skb
, ptype
, skb
->dev
);
2205 * netif_receive_skb - process receive buffer from network
2206 * @skb: buffer to process
2208 * netif_receive_skb() is the main receive data processing function.
2209 * It always succeeds. The buffer may be dropped during processing
2210 * for congestion control or by the protocol layers.
2212 * This function may only be called from softirq context and interrupts
2213 * should be enabled.
2215 * Return values (usually ignored):
2216 * NET_RX_SUCCESS: no congestion
2217 * NET_RX_DROP: packet was dropped
2219 int netif_receive_skb(struct sk_buff
*skb
)
2221 struct packet_type
*ptype
, *pt_prev
;
2222 struct net_device
*orig_dev
;
2223 struct net_device
*null_or_orig
;
2224 int ret
= NET_RX_DROP
;
2227 if (skb
->vlan_tci
&& vlan_hwaccel_do_receive(skb
))
2228 return NET_RX_SUCCESS
;
2230 /* if we've gotten here through NAPI, check netpoll */
2231 if (netpoll_receive_skb(skb
))
2234 if (!skb
->tstamp
.tv64
)
2238 skb
->iif
= skb
->dev
->ifindex
;
2240 null_or_orig
= NULL
;
2241 orig_dev
= skb
->dev
;
2242 if (orig_dev
->master
) {
2243 if (skb_bond_should_drop(skb
))
2244 null_or_orig
= orig_dev
; /* deliver only exact match */
2246 skb
->dev
= orig_dev
->master
;
2249 __get_cpu_var(netdev_rx_stat
).total
++;
2251 skb_reset_network_header(skb
);
2252 skb_reset_transport_header(skb
);
2253 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2259 /* Don't receive packets in an exiting network namespace */
2260 if (!net_alive(dev_net(skb
->dev
))) {
2265 #ifdef CONFIG_NET_CLS_ACT
2266 if (skb
->tc_verd
& TC_NCLS
) {
2267 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2272 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2273 if (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2274 ptype
->dev
== orig_dev
) {
2276 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2281 #ifdef CONFIG_NET_CLS_ACT
2282 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2288 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
2291 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
2295 type
= skb
->protocol
;
2296 list_for_each_entry_rcu(ptype
,
2297 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2298 if (ptype
->type
== type
&&
2299 (ptype
->dev
== null_or_orig
|| ptype
->dev
== skb
->dev
||
2300 ptype
->dev
== orig_dev
)) {
2302 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2308 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2311 /* Jamal, now you will not able to escape explaining
2312 * me how you were going to use this. :-)
2322 /* Network device is going away, flush any packets still pending */
2323 static void flush_backlog(void *arg
)
2325 struct net_device
*dev
= arg
;
2326 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2327 struct sk_buff
*skb
, *tmp
;
2329 skb_queue_walk_safe(&queue
->input_pkt_queue
, skb
, tmp
)
2330 if (skb
->dev
== dev
) {
2331 __skb_unlink(skb
, &queue
->input_pkt_queue
);
2336 static int process_backlog(struct napi_struct
*napi
, int quota
)
2339 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2340 unsigned long start_time
= jiffies
;
2342 napi
->weight
= weight_p
;
2344 struct sk_buff
*skb
;
2346 local_irq_disable();
2347 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
2349 __napi_complete(napi
);
2355 netif_receive_skb(skb
);
2356 } while (++work
< quota
&& jiffies
== start_time
);
2362 * __napi_schedule - schedule for receive
2363 * @n: entry to schedule
2365 * The entry's receive function will be scheduled to run
2367 void __napi_schedule(struct napi_struct
*n
)
2369 unsigned long flags
;
2371 local_irq_save(flags
);
2372 list_add_tail(&n
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
2373 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2374 local_irq_restore(flags
);
2376 EXPORT_SYMBOL(__napi_schedule
);
2379 static void net_rx_action(struct softirq_action
*h
)
2381 struct list_head
*list
= &__get_cpu_var(softnet_data
).poll_list
;
2382 unsigned long time_limit
= jiffies
+ 2;
2383 int budget
= netdev_budget
;
2386 local_irq_disable();
2388 while (!list_empty(list
)) {
2389 struct napi_struct
*n
;
2392 /* If softirq window is exhuasted then punt.
2393 * Allow this to run for 2 jiffies since which will allow
2394 * an average latency of 1.5/HZ.
2396 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
2401 /* Even though interrupts have been re-enabled, this
2402 * access is safe because interrupts can only add new
2403 * entries to the tail of this list, and only ->poll()
2404 * calls can remove this head entry from the list.
2406 n
= list_entry(list
->next
, struct napi_struct
, poll_list
);
2408 have
= netpoll_poll_lock(n
);
2412 /* This NAPI_STATE_SCHED test is for avoiding a race
2413 * with netpoll's poll_napi(). Only the entity which
2414 * obtains the lock and sees NAPI_STATE_SCHED set will
2415 * actually make the ->poll() call. Therefore we avoid
2416 * accidently calling ->poll() when NAPI is not scheduled.
2419 if (test_bit(NAPI_STATE_SCHED
, &n
->state
))
2420 work
= n
->poll(n
, weight
);
2422 WARN_ON_ONCE(work
> weight
);
2426 local_irq_disable();
2428 /* Drivers must not modify the NAPI state if they
2429 * consume the entire weight. In such cases this code
2430 * still "owns" the NAPI instance and therefore can
2431 * move the instance around on the list at-will.
2433 if (unlikely(work
== weight
)) {
2434 if (unlikely(napi_disable_pending(n
)))
2437 list_move_tail(&n
->poll_list
, list
);
2440 netpoll_poll_unlock(have
);
2445 #ifdef CONFIG_NET_DMA
2447 * There may not be any more sk_buffs coming right now, so push
2448 * any pending DMA copies to hardware
2450 if (!cpus_empty(net_dma
.channel_mask
)) {
2452 for_each_cpu_mask_nr(chan_idx
, net_dma
.channel_mask
) {
2453 struct dma_chan
*chan
= net_dma
.channels
[chan_idx
];
2455 dma_async_memcpy_issue_pending(chan
);
2463 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2464 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2468 static gifconf_func_t
* gifconf_list
[NPROTO
];
2471 * register_gifconf - register a SIOCGIF handler
2472 * @family: Address family
2473 * @gifconf: Function handler
2475 * Register protocol dependent address dumping routines. The handler
2476 * that is passed must not be freed or reused until it has been replaced
2477 * by another handler.
2479 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2481 if (family
>= NPROTO
)
2483 gifconf_list
[family
] = gifconf
;
2489 * Map an interface index to its name (SIOCGIFNAME)
2493 * We need this ioctl for efficient implementation of the
2494 * if_indextoname() function required by the IPv6 API. Without
2495 * it, we would have to search all the interfaces to find a
2499 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
2501 struct net_device
*dev
;
2505 * Fetch the caller's info block.
2508 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2511 read_lock(&dev_base_lock
);
2512 dev
= __dev_get_by_index(net
, ifr
.ifr_ifindex
);
2514 read_unlock(&dev_base_lock
);
2518 strcpy(ifr
.ifr_name
, dev
->name
);
2519 read_unlock(&dev_base_lock
);
2521 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2527 * Perform a SIOCGIFCONF call. This structure will change
2528 * size eventually, and there is nothing I can do about it.
2529 * Thus we will need a 'compatibility mode'.
2532 static int dev_ifconf(struct net
*net
, char __user
*arg
)
2535 struct net_device
*dev
;
2542 * Fetch the caller's info block.
2545 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2552 * Loop over the interfaces, and write an info block for each.
2556 for_each_netdev(net
, dev
) {
2557 for (i
= 0; i
< NPROTO
; i
++) {
2558 if (gifconf_list
[i
]) {
2561 done
= gifconf_list
[i
](dev
, NULL
, 0);
2563 done
= gifconf_list
[i
](dev
, pos
+ total
,
2573 * All done. Write the updated control block back to the caller.
2575 ifc
.ifc_len
= total
;
2578 * Both BSD and Solaris return 0 here, so we do too.
2580 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2583 #ifdef CONFIG_PROC_FS
2585 * This is invoked by the /proc filesystem handler to display a device
2588 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2589 __acquires(dev_base_lock
)
2591 struct net
*net
= seq_file_net(seq
);
2593 struct net_device
*dev
;
2595 read_lock(&dev_base_lock
);
2597 return SEQ_START_TOKEN
;
2600 for_each_netdev(net
, dev
)
2607 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2609 struct net
*net
= seq_file_net(seq
);
2611 return v
== SEQ_START_TOKEN
?
2612 first_net_device(net
) : next_net_device((struct net_device
*)v
);
2615 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2616 __releases(dev_base_lock
)
2618 read_unlock(&dev_base_lock
);
2621 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2623 struct net_device_stats
*stats
= dev
->get_stats(dev
);
2625 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2626 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2627 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2629 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2630 stats
->rx_fifo_errors
,
2631 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2632 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2633 stats
->rx_compressed
, stats
->multicast
,
2634 stats
->tx_bytes
, stats
->tx_packets
,
2635 stats
->tx_errors
, stats
->tx_dropped
,
2636 stats
->tx_fifo_errors
, stats
->collisions
,
2637 stats
->tx_carrier_errors
+
2638 stats
->tx_aborted_errors
+
2639 stats
->tx_window_errors
+
2640 stats
->tx_heartbeat_errors
,
2641 stats
->tx_compressed
);
2645 * Called from the PROCfs module. This now uses the new arbitrary sized
2646 * /proc/net interface to create /proc/net/dev
2648 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2650 if (v
== SEQ_START_TOKEN
)
2651 seq_puts(seq
, "Inter-| Receive "
2653 " face |bytes packets errs drop fifo frame "
2654 "compressed multicast|bytes packets errs "
2655 "drop fifo colls carrier compressed\n");
2657 dev_seq_printf_stats(seq
, v
);
2661 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2663 struct netif_rx_stats
*rc
= NULL
;
2665 while (*pos
< nr_cpu_ids
)
2666 if (cpu_online(*pos
)) {
2667 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2674 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2676 return softnet_get_online(pos
);
2679 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2682 return softnet_get_online(pos
);
2685 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2689 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2691 struct netif_rx_stats
*s
= v
;
2693 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2694 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2695 0, 0, 0, 0, /* was fastroute */
2700 static const struct seq_operations dev_seq_ops
= {
2701 .start
= dev_seq_start
,
2702 .next
= dev_seq_next
,
2703 .stop
= dev_seq_stop
,
2704 .show
= dev_seq_show
,
2707 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2709 return seq_open_net(inode
, file
, &dev_seq_ops
,
2710 sizeof(struct seq_net_private
));
2713 static const struct file_operations dev_seq_fops
= {
2714 .owner
= THIS_MODULE
,
2715 .open
= dev_seq_open
,
2717 .llseek
= seq_lseek
,
2718 .release
= seq_release_net
,
2721 static const struct seq_operations softnet_seq_ops
= {
2722 .start
= softnet_seq_start
,
2723 .next
= softnet_seq_next
,
2724 .stop
= softnet_seq_stop
,
2725 .show
= softnet_seq_show
,
2728 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2730 return seq_open(file
, &softnet_seq_ops
);
2733 static const struct file_operations softnet_seq_fops
= {
2734 .owner
= THIS_MODULE
,
2735 .open
= softnet_seq_open
,
2737 .llseek
= seq_lseek
,
2738 .release
= seq_release
,
2741 static void *ptype_get_idx(loff_t pos
)
2743 struct packet_type
*pt
= NULL
;
2747 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
2753 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
2754 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
2763 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2767 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
2770 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2772 struct packet_type
*pt
;
2773 struct list_head
*nxt
;
2777 if (v
== SEQ_START_TOKEN
)
2778 return ptype_get_idx(0);
2781 nxt
= pt
->list
.next
;
2782 if (pt
->type
== htons(ETH_P_ALL
)) {
2783 if (nxt
!= &ptype_all
)
2786 nxt
= ptype_base
[0].next
;
2788 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
2790 while (nxt
== &ptype_base
[hash
]) {
2791 if (++hash
>= PTYPE_HASH_SIZE
)
2793 nxt
= ptype_base
[hash
].next
;
2796 return list_entry(nxt
, struct packet_type
, list
);
2799 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
2805 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
2807 struct packet_type
*pt
= v
;
2809 if (v
== SEQ_START_TOKEN
)
2810 seq_puts(seq
, "Type Device Function\n");
2811 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
2812 if (pt
->type
== htons(ETH_P_ALL
))
2813 seq_puts(seq
, "ALL ");
2815 seq_printf(seq
, "%04x", ntohs(pt
->type
));
2817 seq_printf(seq
, " %-8s %pF\n",
2818 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
2824 static const struct seq_operations ptype_seq_ops
= {
2825 .start
= ptype_seq_start
,
2826 .next
= ptype_seq_next
,
2827 .stop
= ptype_seq_stop
,
2828 .show
= ptype_seq_show
,
2831 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
2833 return seq_open_net(inode
, file
, &ptype_seq_ops
,
2834 sizeof(struct seq_net_private
));
2837 static const struct file_operations ptype_seq_fops
= {
2838 .owner
= THIS_MODULE
,
2839 .open
= ptype_seq_open
,
2841 .llseek
= seq_lseek
,
2842 .release
= seq_release_net
,
2846 static int __net_init
dev_proc_net_init(struct net
*net
)
2850 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
2852 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2854 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
2857 if (wext_proc_init(net
))
2863 proc_net_remove(net
, "ptype");
2865 proc_net_remove(net
, "softnet_stat");
2867 proc_net_remove(net
, "dev");
2871 static void __net_exit
dev_proc_net_exit(struct net
*net
)
2873 wext_proc_exit(net
);
2875 proc_net_remove(net
, "ptype");
2876 proc_net_remove(net
, "softnet_stat");
2877 proc_net_remove(net
, "dev");
2880 static struct pernet_operations __net_initdata dev_proc_ops
= {
2881 .init
= dev_proc_net_init
,
2882 .exit
= dev_proc_net_exit
,
2885 static int __init
dev_proc_init(void)
2887 return register_pernet_subsys(&dev_proc_ops
);
2890 #define dev_proc_init() 0
2891 #endif /* CONFIG_PROC_FS */
2895 * netdev_set_master - set up master/slave pair
2896 * @slave: slave device
2897 * @master: new master device
2899 * Changes the master device of the slave. Pass %NULL to break the
2900 * bonding. The caller must hold the RTNL semaphore. On a failure
2901 * a negative errno code is returned. On success the reference counts
2902 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2903 * function returns zero.
2905 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2907 struct net_device
*old
= slave
->master
;
2917 slave
->master
= master
;
2925 slave
->flags
|= IFF_SLAVE
;
2927 slave
->flags
&= ~IFF_SLAVE
;
2929 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2933 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
2935 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2937 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
2938 ops
->ndo_change_rx_flags(dev
, flags
);
2941 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
2943 unsigned short old_flags
= dev
->flags
;
2947 dev
->flags
|= IFF_PROMISC
;
2948 dev
->promiscuity
+= inc
;
2949 if (dev
->promiscuity
== 0) {
2952 * If inc causes overflow, untouch promisc and return error.
2955 dev
->flags
&= ~IFF_PROMISC
;
2957 dev
->promiscuity
-= inc
;
2958 printk(KERN_WARNING
"%s: promiscuity touches roof, "
2959 "set promiscuity failed, promiscuity feature "
2960 "of device might be broken.\n", dev
->name
);
2964 if (dev
->flags
!= old_flags
) {
2965 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2966 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2969 audit_log(current
->audit_context
, GFP_ATOMIC
,
2970 AUDIT_ANOM_PROMISCUOUS
,
2971 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2972 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2973 (old_flags
& IFF_PROMISC
),
2974 audit_get_loginuid(current
),
2975 current
->uid
, current
->gid
,
2976 audit_get_sessionid(current
));
2978 dev_change_rx_flags(dev
, IFF_PROMISC
);
2984 * dev_set_promiscuity - update promiscuity count on a device
2988 * Add or remove promiscuity from a device. While the count in the device
2989 * remains above zero the interface remains promiscuous. Once it hits zero
2990 * the device reverts back to normal filtering operation. A negative inc
2991 * value is used to drop promiscuity on the device.
2992 * Return 0 if successful or a negative errno code on error.
2994 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
2996 unsigned short old_flags
= dev
->flags
;
2999 err
= __dev_set_promiscuity(dev
, inc
);
3002 if (dev
->flags
!= old_flags
)
3003 dev_set_rx_mode(dev
);
3008 * dev_set_allmulti - update allmulti count on a device
3012 * Add or remove reception of all multicast frames to a device. While the
3013 * count in the device remains above zero the interface remains listening
3014 * to all interfaces. Once it hits zero the device reverts back to normal
3015 * filtering operation. A negative @inc value is used to drop the counter
3016 * when releasing a resource needing all multicasts.
3017 * Return 0 if successful or a negative errno code on error.
3020 int dev_set_allmulti(struct net_device
*dev
, int inc
)
3022 unsigned short old_flags
= dev
->flags
;
3026 dev
->flags
|= IFF_ALLMULTI
;
3027 dev
->allmulti
+= inc
;
3028 if (dev
->allmulti
== 0) {
3031 * If inc causes overflow, untouch allmulti and return error.
3034 dev
->flags
&= ~IFF_ALLMULTI
;
3036 dev
->allmulti
-= inc
;
3037 printk(KERN_WARNING
"%s: allmulti touches roof, "
3038 "set allmulti failed, allmulti feature of "
3039 "device might be broken.\n", dev
->name
);
3043 if (dev
->flags
^ old_flags
) {
3044 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
3045 dev_set_rx_mode(dev
);
3051 * Upload unicast and multicast address lists to device and
3052 * configure RX filtering. When the device doesn't support unicast
3053 * filtering it is put in promiscuous mode while unicast addresses
3056 void __dev_set_rx_mode(struct net_device
*dev
)
3058 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3060 /* dev_open will call this function so the list will stay sane. */
3061 if (!(dev
->flags
&IFF_UP
))
3064 if (!netif_device_present(dev
))
3067 if (ops
->ndo_set_rx_mode
)
3068 ops
->ndo_set_rx_mode(dev
);
3070 /* Unicast addresses changes may only happen under the rtnl,
3071 * therefore calling __dev_set_promiscuity here is safe.
3073 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
3074 __dev_set_promiscuity(dev
, 1);
3075 dev
->uc_promisc
= 1;
3076 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
3077 __dev_set_promiscuity(dev
, -1);
3078 dev
->uc_promisc
= 0;
3081 if (ops
->ndo_set_multicast_list
)
3082 ops
->ndo_set_multicast_list(dev
);
3086 void dev_set_rx_mode(struct net_device
*dev
)
3088 netif_addr_lock_bh(dev
);
3089 __dev_set_rx_mode(dev
);
3090 netif_addr_unlock_bh(dev
);
3093 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
3094 void *addr
, int alen
, int glbl
)
3096 struct dev_addr_list
*da
;
3098 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
3099 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
3100 alen
== da
->da_addrlen
) {
3102 int old_glbl
= da
->da_gusers
;
3119 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
3120 void *addr
, int alen
, int glbl
)
3122 struct dev_addr_list
*da
;
3124 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
3125 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
3126 da
->da_addrlen
== alen
) {
3128 int old_glbl
= da
->da_gusers
;
3138 da
= kzalloc(sizeof(*da
), GFP_ATOMIC
);
3141 memcpy(da
->da_addr
, addr
, alen
);
3142 da
->da_addrlen
= alen
;
3144 da
->da_gusers
= glbl
? 1 : 0;
3152 * dev_unicast_delete - Release secondary unicast address.
3154 * @addr: address to delete
3155 * @alen: length of @addr
3157 * Release reference to a secondary unicast address and remove it
3158 * from the device if the reference count drops to zero.
3160 * The caller must hold the rtnl_mutex.
3162 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
3168 netif_addr_lock_bh(dev
);
3169 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3171 __dev_set_rx_mode(dev
);
3172 netif_addr_unlock_bh(dev
);
3175 EXPORT_SYMBOL(dev_unicast_delete
);
3178 * dev_unicast_add - add a secondary unicast address
3180 * @addr: address to add
3181 * @alen: length of @addr
3183 * Add a secondary unicast address to the device or increase
3184 * the reference count if it already exists.
3186 * The caller must hold the rtnl_mutex.
3188 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
3194 netif_addr_lock_bh(dev
);
3195 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3197 __dev_set_rx_mode(dev
);
3198 netif_addr_unlock_bh(dev
);
3201 EXPORT_SYMBOL(dev_unicast_add
);
3203 int __dev_addr_sync(struct dev_addr_list
**to
, int *to_count
,
3204 struct dev_addr_list
**from
, int *from_count
)
3206 struct dev_addr_list
*da
, *next
;
3210 while (da
!= NULL
) {
3212 if (!da
->da_synced
) {
3213 err
= __dev_addr_add(to
, to_count
,
3214 da
->da_addr
, da
->da_addrlen
, 0);
3219 } else if (da
->da_users
== 1) {
3220 __dev_addr_delete(to
, to_count
,
3221 da
->da_addr
, da
->da_addrlen
, 0);
3222 __dev_addr_delete(from
, from_count
,
3223 da
->da_addr
, da
->da_addrlen
, 0);
3230 void __dev_addr_unsync(struct dev_addr_list
**to
, int *to_count
,
3231 struct dev_addr_list
**from
, int *from_count
)
3233 struct dev_addr_list
*da
, *next
;
3236 while (da
!= NULL
) {
3238 if (da
->da_synced
) {
3239 __dev_addr_delete(to
, to_count
,
3240 da
->da_addr
, da
->da_addrlen
, 0);
3242 __dev_addr_delete(from
, from_count
,
3243 da
->da_addr
, da
->da_addrlen
, 0);
3250 * dev_unicast_sync - Synchronize device's unicast list to another device
3251 * @to: destination device
3252 * @from: source device
3254 * Add newly added addresses to the destination device and release
3255 * addresses that have no users left. The source device must be
3256 * locked by netif_tx_lock_bh.
3258 * This function is intended to be called from the dev->set_rx_mode
3259 * function of layered software devices.
3261 int dev_unicast_sync(struct net_device
*to
, struct net_device
*from
)
3265 netif_addr_lock_bh(to
);
3266 err
= __dev_addr_sync(&to
->uc_list
, &to
->uc_count
,
3267 &from
->uc_list
, &from
->uc_count
);
3269 __dev_set_rx_mode(to
);
3270 netif_addr_unlock_bh(to
);
3273 EXPORT_SYMBOL(dev_unicast_sync
);
3276 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3277 * @to: destination device
3278 * @from: source device
3280 * Remove all addresses that were added to the destination device by
3281 * dev_unicast_sync(). This function is intended to be called from the
3282 * dev->stop function of layered software devices.
3284 void dev_unicast_unsync(struct net_device
*to
, struct net_device
*from
)
3286 netif_addr_lock_bh(from
);
3287 netif_addr_lock(to
);
3289 __dev_addr_unsync(&to
->uc_list
, &to
->uc_count
,
3290 &from
->uc_list
, &from
->uc_count
);
3291 __dev_set_rx_mode(to
);
3293 netif_addr_unlock(to
);
3294 netif_addr_unlock_bh(from
);
3296 EXPORT_SYMBOL(dev_unicast_unsync
);
3298 static void __dev_addr_discard(struct dev_addr_list
**list
)
3300 struct dev_addr_list
*tmp
;
3302 while (*list
!= NULL
) {
3305 if (tmp
->da_users
> tmp
->da_gusers
)
3306 printk("__dev_addr_discard: address leakage! "
3307 "da_users=%d\n", tmp
->da_users
);
3312 static void dev_addr_discard(struct net_device
*dev
)
3314 netif_addr_lock_bh(dev
);
3316 __dev_addr_discard(&dev
->uc_list
);
3319 __dev_addr_discard(&dev
->mc_list
);
3322 netif_addr_unlock_bh(dev
);
3326 * dev_get_flags - get flags reported to userspace
3329 * Get the combination of flag bits exported through APIs to userspace.
3331 unsigned dev_get_flags(const struct net_device
*dev
)
3335 flags
= (dev
->flags
& ~(IFF_PROMISC
|
3340 (dev
->gflags
& (IFF_PROMISC
|
3343 if (netif_running(dev
)) {
3344 if (netif_oper_up(dev
))
3345 flags
|= IFF_RUNNING
;
3346 if (netif_carrier_ok(dev
))
3347 flags
|= IFF_LOWER_UP
;
3348 if (netif_dormant(dev
))
3349 flags
|= IFF_DORMANT
;
3356 * dev_change_flags - change device settings
3358 * @flags: device state flags
3360 * Change settings on device based state flags. The flags are
3361 * in the userspace exported format.
3363 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
3366 int old_flags
= dev
->flags
;
3371 * Set the flags on our device.
3374 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
3375 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
3377 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
3381 * Load in the correct multicast list now the flags have changed.
3384 if ((old_flags
^ flags
) & IFF_MULTICAST
)
3385 dev_change_rx_flags(dev
, IFF_MULTICAST
);
3387 dev_set_rx_mode(dev
);
3390 * Have we downed the interface. We handle IFF_UP ourselves
3391 * according to user attempts to set it, rather than blindly
3396 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
3397 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
3400 dev_set_rx_mode(dev
);
3403 if (dev
->flags
& IFF_UP
&&
3404 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
3406 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
3408 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
3409 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
3410 dev
->gflags
^= IFF_PROMISC
;
3411 dev_set_promiscuity(dev
, inc
);
3414 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3415 is important. Some (broken) drivers set IFF_PROMISC, when
3416 IFF_ALLMULTI is requested not asking us and not reporting.
3418 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
3419 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
3420 dev
->gflags
^= IFF_ALLMULTI
;
3421 dev_set_allmulti(dev
, inc
);
3424 /* Exclude state transition flags, already notified */
3425 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
3427 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
3433 * dev_set_mtu - Change maximum transfer unit
3435 * @new_mtu: new transfer unit
3437 * Change the maximum transfer size of the network device.
3439 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
3441 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3444 if (new_mtu
== dev
->mtu
)
3447 /* MTU must be positive. */
3451 if (!netif_device_present(dev
))
3455 if (ops
->ndo_change_mtu
)
3456 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
3460 if (!err
&& dev
->flags
& IFF_UP
)
3461 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
3466 * dev_set_mac_address - Change Media Access Control Address
3470 * Change the hardware (MAC) address of the device
3472 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
3474 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3477 if (!ops
->ndo_set_mac_address
)
3479 if (sa
->sa_family
!= dev
->type
)
3481 if (!netif_device_present(dev
))
3483 err
= ops
->ndo_set_mac_address(dev
, sa
);
3485 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3490 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3492 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3495 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3501 case SIOCGIFFLAGS
: /* Get interface flags */
3502 ifr
->ifr_flags
= dev_get_flags(dev
);
3505 case SIOCGIFMETRIC
: /* Get the metric on the interface
3506 (currently unused) */
3507 ifr
->ifr_metric
= 0;
3510 case SIOCGIFMTU
: /* Get the MTU of a device */
3511 ifr
->ifr_mtu
= dev
->mtu
;
3516 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
3518 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
3519 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3520 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
3528 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
3529 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
3530 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
3531 ifr
->ifr_map
.irq
= dev
->irq
;
3532 ifr
->ifr_map
.dma
= dev
->dma
;
3533 ifr
->ifr_map
.port
= dev
->if_port
;
3537 ifr
->ifr_ifindex
= dev
->ifindex
;
3541 ifr
->ifr_qlen
= dev
->tx_queue_len
;
3545 /* dev_ioctl() should ensure this case
3557 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3559 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3562 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3563 const struct net_device_ops
*ops
= dev
->netdev_ops
;
3569 case SIOCSIFFLAGS
: /* Set interface flags */
3570 return dev_change_flags(dev
, ifr
->ifr_flags
);
3572 case SIOCSIFMETRIC
: /* Set the metric on the interface
3573 (currently unused) */
3576 case SIOCSIFMTU
: /* Set the MTU of a device */
3577 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
3580 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
3582 case SIOCSIFHWBROADCAST
:
3583 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
3585 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
3586 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3587 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3591 if (ops
->ndo_set_config
) {
3592 if (!netif_device_present(dev
))
3594 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
3599 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
3600 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3602 if (!netif_device_present(dev
))
3604 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
3608 if ((!ops
->ndo_set_multicast_list
&& !ops
->ndo_set_rx_mode
) ||
3609 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3611 if (!netif_device_present(dev
))
3613 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
3617 if (ifr
->ifr_qlen
< 0)
3619 dev
->tx_queue_len
= ifr
->ifr_qlen
;
3623 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
3624 return dev_change_name(dev
, ifr
->ifr_newname
);
3627 * Unknown or private ioctl
3631 if ((cmd
>= SIOCDEVPRIVATE
&&
3632 cmd
<= SIOCDEVPRIVATE
+ 15) ||
3633 cmd
== SIOCBONDENSLAVE
||
3634 cmd
== SIOCBONDRELEASE
||
3635 cmd
== SIOCBONDSETHWADDR
||
3636 cmd
== SIOCBONDSLAVEINFOQUERY
||
3637 cmd
== SIOCBONDINFOQUERY
||
3638 cmd
== SIOCBONDCHANGEACTIVE
||
3639 cmd
== SIOCGMIIPHY
||
3640 cmd
== SIOCGMIIREG
||
3641 cmd
== SIOCSMIIREG
||
3642 cmd
== SIOCBRADDIF
||
3643 cmd
== SIOCBRDELIF
||
3644 cmd
== SIOCWANDEV
) {
3646 if (ops
->ndo_do_ioctl
) {
3647 if (netif_device_present(dev
))
3648 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
3660 * This function handles all "interface"-type I/O control requests. The actual
3661 * 'doing' part of this is dev_ifsioc above.
3665 * dev_ioctl - network device ioctl
3666 * @net: the applicable net namespace
3667 * @cmd: command to issue
3668 * @arg: pointer to a struct ifreq in user space
3670 * Issue ioctl functions to devices. This is normally called by the
3671 * user space syscall interfaces but can sometimes be useful for
3672 * other purposes. The return value is the return from the syscall if
3673 * positive or a negative errno code on error.
3676 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
3682 /* One special case: SIOCGIFCONF takes ifconf argument
3683 and requires shared lock, because it sleeps writing
3687 if (cmd
== SIOCGIFCONF
) {
3689 ret
= dev_ifconf(net
, (char __user
*) arg
);
3693 if (cmd
== SIOCGIFNAME
)
3694 return dev_ifname(net
, (struct ifreq __user
*)arg
);
3696 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3699 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
3701 colon
= strchr(ifr
.ifr_name
, ':');
3706 * See which interface the caller is talking about.
3711 * These ioctl calls:
3712 * - can be done by all.
3713 * - atomic and do not require locking.
3724 dev_load(net
, ifr
.ifr_name
);
3725 read_lock(&dev_base_lock
);
3726 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
3727 read_unlock(&dev_base_lock
);
3731 if (copy_to_user(arg
, &ifr
,
3732 sizeof(struct ifreq
)))
3738 dev_load(net
, ifr
.ifr_name
);
3740 ret
= dev_ethtool(net
, &ifr
);
3745 if (copy_to_user(arg
, &ifr
,
3746 sizeof(struct ifreq
)))
3752 * These ioctl calls:
3753 * - require superuser power.
3754 * - require strict serialization.
3760 if (!capable(CAP_NET_ADMIN
))
3762 dev_load(net
, ifr
.ifr_name
);
3764 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3769 if (copy_to_user(arg
, &ifr
,
3770 sizeof(struct ifreq
)))
3776 * These ioctl calls:
3777 * - require superuser power.
3778 * - require strict serialization.
3779 * - do not return a value
3789 case SIOCSIFHWBROADCAST
:
3792 case SIOCBONDENSLAVE
:
3793 case SIOCBONDRELEASE
:
3794 case SIOCBONDSETHWADDR
:
3795 case SIOCBONDCHANGEACTIVE
:
3798 if (!capable(CAP_NET_ADMIN
))
3801 case SIOCBONDSLAVEINFOQUERY
:
3802 case SIOCBONDINFOQUERY
:
3803 dev_load(net
, ifr
.ifr_name
);
3805 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3810 /* Get the per device memory space. We can add this but
3811 * currently do not support it */
3813 /* Set the per device memory buffer space.
3814 * Not applicable in our case */
3819 * Unknown or private ioctl.
3822 if (cmd
== SIOCWANDEV
||
3823 (cmd
>= SIOCDEVPRIVATE
&&
3824 cmd
<= SIOCDEVPRIVATE
+ 15)) {
3825 dev_load(net
, ifr
.ifr_name
);
3827 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3829 if (!ret
&& copy_to_user(arg
, &ifr
,
3830 sizeof(struct ifreq
)))
3834 /* Take care of Wireless Extensions */
3835 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
3836 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
3843 * dev_new_index - allocate an ifindex
3844 * @net: the applicable net namespace
3846 * Returns a suitable unique value for a new device interface
3847 * number. The caller must hold the rtnl semaphore or the
3848 * dev_base_lock to be sure it remains unique.
3850 static int dev_new_index(struct net
*net
)
3856 if (!__dev_get_by_index(net
, ifindex
))
3861 /* Delayed registration/unregisteration */
3862 static LIST_HEAD(net_todo_list
);
3864 static void net_set_todo(struct net_device
*dev
)
3866 list_add_tail(&dev
->todo_list
, &net_todo_list
);
3869 static void rollback_registered(struct net_device
*dev
)
3871 BUG_ON(dev_boot_phase
);
3874 /* Some devices call without registering for initialization unwind. */
3875 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3876 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3877 "was registered\n", dev
->name
, dev
);
3883 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3885 /* If device is running, close it first. */
3888 /* And unlink it from device chain. */
3889 unlist_netdevice(dev
);
3891 dev
->reg_state
= NETREG_UNREGISTERING
;
3895 /* Shutdown queueing discipline. */
3899 /* Notify protocols, that we are about to destroy
3900 this device. They should clean all the things.
3902 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
3905 * Flush the unicast and multicast chains
3907 dev_addr_discard(dev
);
3909 if (dev
->netdev_ops
->ndo_uninit
)
3910 dev
->netdev_ops
->ndo_uninit(dev
);
3912 /* Notifier chain MUST detach us from master device. */
3913 WARN_ON(dev
->master
);
3915 /* Remove entries from kobject tree */
3916 netdev_unregister_kobject(dev
);
3923 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
3924 struct netdev_queue
*dev_queue
,
3927 spin_lock_init(&dev_queue
->_xmit_lock
);
3928 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
3929 dev_queue
->xmit_lock_owner
= -1;
3932 static void netdev_init_queue_locks(struct net_device
*dev
)
3934 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
3935 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
3938 unsigned long netdev_fix_features(unsigned long features
, const char *name
)
3940 /* Fix illegal SG+CSUM combinations. */
3941 if ((features
& NETIF_F_SG
) &&
3942 !(features
& NETIF_F_ALL_CSUM
)) {
3944 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no "
3945 "checksum feature.\n", name
);
3946 features
&= ~NETIF_F_SG
;
3949 /* TSO requires that SG is present as well. */
3950 if ((features
& NETIF_F_TSO
) && !(features
& NETIF_F_SG
)) {
3952 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no "
3953 "SG feature.\n", name
);
3954 features
&= ~NETIF_F_TSO
;
3957 if (features
& NETIF_F_UFO
) {
3958 if (!(features
& NETIF_F_GEN_CSUM
)) {
3960 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
3961 "since no NETIF_F_HW_CSUM feature.\n",
3963 features
&= ~NETIF_F_UFO
;
3966 if (!(features
& NETIF_F_SG
)) {
3968 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO "
3969 "since no NETIF_F_SG feature.\n", name
);
3970 features
&= ~NETIF_F_UFO
;
3976 EXPORT_SYMBOL(netdev_fix_features
);
3979 * register_netdevice - register a network device
3980 * @dev: device to register
3982 * Take a completed network device structure and add it to the kernel
3983 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3984 * chain. 0 is returned on success. A negative errno code is returned
3985 * on a failure to set up the device, or if the name is a duplicate.
3987 * Callers must hold the rtnl semaphore. You may want
3988 * register_netdev() instead of this.
3991 * The locking appears insufficient to guarantee two parallel registers
3992 * will not get the same name.
3995 int register_netdevice(struct net_device
*dev
)
3997 struct hlist_head
*head
;
3998 struct hlist_node
*p
;
4000 struct net
*net
= dev_net(dev
);
4002 BUG_ON(dev_boot_phase
);
4007 /* When net_device's are persistent, this will be fatal. */
4008 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
4011 spin_lock_init(&dev
->addr_list_lock
);
4012 netdev_set_addr_lockdep_class(dev
);
4013 netdev_init_queue_locks(dev
);
4017 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4018 /* Netdevice_ops API compatiability support.
4019 * This is temporary until all network devices are converted.
4021 if (dev
->netdev_ops
) {
4022 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4024 dev
->init
= ops
->ndo_init
;
4025 dev
->uninit
= ops
->ndo_uninit
;
4026 dev
->open
= ops
->ndo_open
;
4027 dev
->change_rx_flags
= ops
->ndo_change_rx_flags
;
4028 dev
->set_rx_mode
= ops
->ndo_set_rx_mode
;
4029 dev
->set_multicast_list
= ops
->ndo_set_multicast_list
;
4030 dev
->set_mac_address
= ops
->ndo_set_mac_address
;
4031 dev
->validate_addr
= ops
->ndo_validate_addr
;
4032 dev
->do_ioctl
= ops
->ndo_do_ioctl
;
4033 dev
->set_config
= ops
->ndo_set_config
;
4034 dev
->change_mtu
= ops
->ndo_change_mtu
;
4035 dev
->tx_timeout
= ops
->ndo_tx_timeout
;
4036 dev
->get_stats
= ops
->ndo_get_stats
;
4037 dev
->vlan_rx_register
= ops
->ndo_vlan_rx_register
;
4038 dev
->vlan_rx_add_vid
= ops
->ndo_vlan_rx_add_vid
;
4039 dev
->vlan_rx_kill_vid
= ops
->ndo_vlan_rx_kill_vid
;
4040 #ifdef CONFIG_NET_POLL_CONTROLLER
4041 dev
->poll_controller
= ops
->ndo_poll_controller
;
4044 char drivername
[64];
4045 pr_info("%s (%s): not using net_device_ops yet\n",
4046 dev
->name
, netdev_drivername(dev
, drivername
, 64));
4048 /* This works only because net_device_ops and the
4049 compatiablity structure are the same. */
4050 dev
->netdev_ops
= (void *) &(dev
->init
);
4054 /* Init, if this function is available */
4055 if (dev
->netdev_ops
->ndo_init
) {
4056 ret
= dev
->netdev_ops
->ndo_init(dev
);
4064 if (!dev_valid_name(dev
->name
)) {
4069 dev
->ifindex
= dev_new_index(net
);
4070 if (dev
->iflink
== -1)
4071 dev
->iflink
= dev
->ifindex
;
4073 /* Check for existence of name */
4074 head
= dev_name_hash(net
, dev
->name
);
4075 hlist_for_each(p
, head
) {
4076 struct net_device
*d
4077 = hlist_entry(p
, struct net_device
, name_hlist
);
4078 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
4084 /* Fix illegal checksum combinations */
4085 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
4086 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4087 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
4089 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
4092 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
4093 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
4094 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
4096 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
4099 dev
->features
= netdev_fix_features(dev
->features
, dev
->name
);
4101 /* Enable software GSO if SG is supported. */
4102 if (dev
->features
& NETIF_F_SG
)
4103 dev
->features
|= NETIF_F_GSO
;
4105 netdev_initialize_kobject(dev
);
4106 ret
= netdev_register_kobject(dev
);
4109 dev
->reg_state
= NETREG_REGISTERED
;
4112 * Default initial state at registry is that the
4113 * device is present.
4116 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
4118 dev_init_scheduler(dev
);
4120 list_netdevice(dev
);
4122 /* Notify protocols, that a new device appeared. */
4123 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
4124 ret
= notifier_to_errno(ret
);
4126 rollback_registered(dev
);
4127 dev
->reg_state
= NETREG_UNREGISTERED
;
4134 if (dev
->netdev_ops
->ndo_uninit
)
4135 dev
->netdev_ops
->ndo_uninit(dev
);
4140 * register_netdev - register a network device
4141 * @dev: device to register
4143 * Take a completed network device structure and add it to the kernel
4144 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4145 * chain. 0 is returned on success. A negative errno code is returned
4146 * on a failure to set up the device, or if the name is a duplicate.
4148 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4149 * and expands the device name if you passed a format string to
4152 int register_netdev(struct net_device
*dev
)
4159 * If the name is a format string the caller wants us to do a
4162 if (strchr(dev
->name
, '%')) {
4163 err
= dev_alloc_name(dev
, dev
->name
);
4168 err
= register_netdevice(dev
);
4173 EXPORT_SYMBOL(register_netdev
);
4176 * netdev_wait_allrefs - wait until all references are gone.
4178 * This is called when unregistering network devices.
4180 * Any protocol or device that holds a reference should register
4181 * for netdevice notification, and cleanup and put back the
4182 * reference if they receive an UNREGISTER event.
4183 * We can get stuck here if buggy protocols don't correctly
4186 static void netdev_wait_allrefs(struct net_device
*dev
)
4188 unsigned long rebroadcast_time
, warning_time
;
4190 rebroadcast_time
= warning_time
= jiffies
;
4191 while (atomic_read(&dev
->refcnt
) != 0) {
4192 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
4195 /* Rebroadcast unregister notification */
4196 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4198 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
4200 /* We must not have linkwatch events
4201 * pending on unregister. If this
4202 * happens, we simply run the queue
4203 * unscheduled, resulting in a noop
4206 linkwatch_run_queue();
4211 rebroadcast_time
= jiffies
;
4216 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
4217 printk(KERN_EMERG
"unregister_netdevice: "
4218 "waiting for %s to become free. Usage "
4220 dev
->name
, atomic_read(&dev
->refcnt
));
4221 warning_time
= jiffies
;
4230 * register_netdevice(x1);
4231 * register_netdevice(x2);
4233 * unregister_netdevice(y1);
4234 * unregister_netdevice(y2);
4240 * We are invoked by rtnl_unlock().
4241 * This allows us to deal with problems:
4242 * 1) We can delete sysfs objects which invoke hotplug
4243 * without deadlocking with linkwatch via keventd.
4244 * 2) Since we run with the RTNL semaphore not held, we can sleep
4245 * safely in order to wait for the netdev refcnt to drop to zero.
4247 * We must not return until all unregister events added during
4248 * the interval the lock was held have been completed.
4250 void netdev_run_todo(void)
4252 struct list_head list
;
4254 /* Snapshot list, allow later requests */
4255 list_replace_init(&net_todo_list
, &list
);
4259 while (!list_empty(&list
)) {
4260 struct net_device
*dev
4261 = list_entry(list
.next
, struct net_device
, todo_list
);
4262 list_del(&dev
->todo_list
);
4264 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
4265 printk(KERN_ERR
"network todo '%s' but state %d\n",
4266 dev
->name
, dev
->reg_state
);
4271 dev
->reg_state
= NETREG_UNREGISTERED
;
4273 on_each_cpu(flush_backlog
, dev
, 1);
4275 netdev_wait_allrefs(dev
);
4278 BUG_ON(atomic_read(&dev
->refcnt
));
4279 WARN_ON(dev
->ip_ptr
);
4280 WARN_ON(dev
->ip6_ptr
);
4281 WARN_ON(dev
->dn_ptr
);
4283 if (dev
->destructor
)
4284 dev
->destructor(dev
);
4286 /* Free network device */
4287 kobject_put(&dev
->dev
.kobj
);
4291 static struct net_device_stats
*internal_stats(struct net_device
*dev
)
4296 static void netdev_init_one_queue(struct net_device
*dev
,
4297 struct netdev_queue
*queue
,
4303 static void netdev_init_queues(struct net_device
*dev
)
4305 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
4306 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
4307 spin_lock_init(&dev
->tx_global_lock
);
4311 * alloc_netdev_mq - allocate network device
4312 * @sizeof_priv: size of private data to allocate space for
4313 * @name: device name format string
4314 * @setup: callback to initialize device
4315 * @queue_count: the number of subqueues to allocate
4317 * Allocates a struct net_device with private data area for driver use
4318 * and performs basic initialization. Also allocates subquue structs
4319 * for each queue on the device at the end of the netdevice.
4321 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
4322 void (*setup
)(struct net_device
*), unsigned int queue_count
)
4324 struct netdev_queue
*tx
;
4325 struct net_device
*dev
;
4329 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
4331 alloc_size
= sizeof(struct net_device
);
4333 /* ensure 32-byte alignment of private area */
4334 alloc_size
= (alloc_size
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
4335 alloc_size
+= sizeof_priv
;
4337 /* ensure 32-byte alignment of whole construct */
4338 alloc_size
+= NETDEV_ALIGN_CONST
;
4340 p
= kzalloc(alloc_size
, GFP_KERNEL
);
4342 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
4346 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
4348 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
4354 dev
= (struct net_device
*)
4355 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
4356 dev
->padded
= (char *)dev
- (char *)p
;
4357 dev_net_set(dev
, &init_net
);
4360 dev
->num_tx_queues
= queue_count
;
4361 dev
->real_num_tx_queues
= queue_count
;
4364 dev
->priv
= ((char *)dev
+
4365 ((sizeof(struct net_device
) + NETDEV_ALIGN_CONST
)
4366 & ~NETDEV_ALIGN_CONST
));
4369 dev
->gso_max_size
= GSO_MAX_SIZE
;
4371 netdev_init_queues(dev
);
4373 dev
->get_stats
= internal_stats
;
4374 netpoll_netdev_init(dev
);
4376 strcpy(dev
->name
, name
);
4379 EXPORT_SYMBOL(alloc_netdev_mq
);
4382 * free_netdev - free network device
4385 * This function does the last stage of destroying an allocated device
4386 * interface. The reference to the device object is released.
4387 * If this is the last reference then it will be freed.
4389 void free_netdev(struct net_device
*dev
)
4391 release_net(dev_net(dev
));
4395 /* Compatibility with error handling in drivers */
4396 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4397 kfree((char *)dev
- dev
->padded
);
4401 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
4402 dev
->reg_state
= NETREG_RELEASED
;
4404 /* will free via device release */
4405 put_device(&dev
->dev
);
4409 * synchronize_net - Synchronize with packet receive processing
4411 * Wait for packets currently being received to be done.
4412 * Does not block later packets from starting.
4414 void synchronize_net(void)
4421 * unregister_netdevice - remove device from the kernel
4424 * This function shuts down a device interface and removes it
4425 * from the kernel tables.
4427 * Callers must hold the rtnl semaphore. You may want
4428 * unregister_netdev() instead of this.
4431 void unregister_netdevice(struct net_device
*dev
)
4435 rollback_registered(dev
);
4436 /* Finish processing unregister after unlock */
4441 * unregister_netdev - remove device from the kernel
4444 * This function shuts down a device interface and removes it
4445 * from the kernel tables.
4447 * This is just a wrapper for unregister_netdevice that takes
4448 * the rtnl semaphore. In general you want to use this and not
4449 * unregister_netdevice.
4451 void unregister_netdev(struct net_device
*dev
)
4454 unregister_netdevice(dev
);
4458 EXPORT_SYMBOL(unregister_netdev
);
4461 * dev_change_net_namespace - move device to different nethost namespace
4463 * @net: network namespace
4464 * @pat: If not NULL name pattern to try if the current device name
4465 * is already taken in the destination network namespace.
4467 * This function shuts down a device interface and moves it
4468 * to a new network namespace. On success 0 is returned, on
4469 * a failure a netagive errno code is returned.
4471 * Callers must hold the rtnl semaphore.
4474 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
4477 const char *destname
;
4482 /* Don't allow namespace local devices to be moved. */
4484 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
4488 /* Don't allow real devices to be moved when sysfs
4492 if (dev
->dev
.parent
)
4496 /* Ensure the device has been registrered */
4498 if (dev
->reg_state
!= NETREG_REGISTERED
)
4501 /* Get out if there is nothing todo */
4503 if (net_eq(dev_net(dev
), net
))
4506 /* Pick the destination device name, and ensure
4507 * we can use it in the destination network namespace.
4510 destname
= dev
->name
;
4511 if (__dev_get_by_name(net
, destname
)) {
4512 /* We get here if we can't use the current device name */
4515 if (!dev_valid_name(pat
))
4517 if (strchr(pat
, '%')) {
4518 if (__dev_alloc_name(net
, pat
, buf
) < 0)
4523 if (__dev_get_by_name(net
, destname
))
4528 * And now a mini version of register_netdevice unregister_netdevice.
4531 /* If device is running close it first. */
4534 /* And unlink it from device chain */
4536 unlist_netdevice(dev
);
4540 /* Shutdown queueing discipline. */
4543 /* Notify protocols, that we are about to destroy
4544 this device. They should clean all the things.
4546 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4549 * Flush the unicast and multicast chains
4551 dev_addr_discard(dev
);
4553 netdev_unregister_kobject(dev
);
4555 /* Actually switch the network namespace */
4556 dev_net_set(dev
, net
);
4558 /* Assign the new device name */
4559 if (destname
!= dev
->name
)
4560 strcpy(dev
->name
, destname
);
4562 /* If there is an ifindex conflict assign a new one */
4563 if (__dev_get_by_index(net
, dev
->ifindex
)) {
4564 int iflink
= (dev
->iflink
== dev
->ifindex
);
4565 dev
->ifindex
= dev_new_index(net
);
4567 dev
->iflink
= dev
->ifindex
;
4570 /* Fixup kobjects */
4571 err
= netdev_register_kobject(dev
);
4574 /* Add the device back in the hashes */
4575 list_netdevice(dev
);
4577 /* Notify protocols, that a new device appeared. */
4578 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
4586 static int dev_cpu_callback(struct notifier_block
*nfb
,
4587 unsigned long action
,
4590 struct sk_buff
**list_skb
;
4591 struct Qdisc
**list_net
;
4592 struct sk_buff
*skb
;
4593 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
4594 struct softnet_data
*sd
, *oldsd
;
4596 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
4599 local_irq_disable();
4600 cpu
= smp_processor_id();
4601 sd
= &per_cpu(softnet_data
, cpu
);
4602 oldsd
= &per_cpu(softnet_data
, oldcpu
);
4604 /* Find end of our completion_queue. */
4605 list_skb
= &sd
->completion_queue
;
4607 list_skb
= &(*list_skb
)->next
;
4608 /* Append completion queue from offline CPU. */
4609 *list_skb
= oldsd
->completion_queue
;
4610 oldsd
->completion_queue
= NULL
;
4612 /* Find end of our output_queue. */
4613 list_net
= &sd
->output_queue
;
4615 list_net
= &(*list_net
)->next_sched
;
4616 /* Append output queue from offline CPU. */
4617 *list_net
= oldsd
->output_queue
;
4618 oldsd
->output_queue
= NULL
;
4620 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
4623 /* Process offline CPU's input_pkt_queue */
4624 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
4630 #ifdef CONFIG_NET_DMA
4632 * net_dma_rebalance - try to maintain one DMA channel per CPU
4633 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4635 * This is called when the number of channels allocated to the net_dma client
4636 * changes. The net_dma client tries to have one DMA channel per CPU.
4639 static void net_dma_rebalance(struct net_dma
*net_dma
)
4641 unsigned int cpu
, i
, n
, chan_idx
;
4642 struct dma_chan
*chan
;
4644 if (cpus_empty(net_dma
->channel_mask
)) {
4645 for_each_online_cpu(cpu
)
4646 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
4651 cpu
= first_cpu(cpu_online_map
);
4653 for_each_cpu_mask_nr(chan_idx
, net_dma
->channel_mask
) {
4654 chan
= net_dma
->channels
[chan_idx
];
4656 n
= ((num_online_cpus() / cpus_weight(net_dma
->channel_mask
))
4657 + (i
< (num_online_cpus() %
4658 cpus_weight(net_dma
->channel_mask
)) ? 1 : 0));
4661 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
4662 cpu
= next_cpu(cpu
, cpu_online_map
);
4670 * netdev_dma_event - event callback for the net_dma_client
4671 * @client: should always be net_dma_client
4672 * @chan: DMA channel for the event
4673 * @state: DMA state to be handled
4675 static enum dma_state_client
4676 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
4677 enum dma_state state
)
4679 int i
, found
= 0, pos
= -1;
4680 struct net_dma
*net_dma
=
4681 container_of(client
, struct net_dma
, client
);
4682 enum dma_state_client ack
= DMA_DUP
; /* default: take no action */
4684 spin_lock(&net_dma
->lock
);
4686 case DMA_RESOURCE_AVAILABLE
:
4687 for (i
= 0; i
< nr_cpu_ids
; i
++)
4688 if (net_dma
->channels
[i
] == chan
) {
4691 } else if (net_dma
->channels
[i
] == NULL
&& pos
< 0)
4694 if (!found
&& pos
>= 0) {
4696 net_dma
->channels
[pos
] = chan
;
4697 cpu_set(pos
, net_dma
->channel_mask
);
4698 net_dma_rebalance(net_dma
);
4701 case DMA_RESOURCE_REMOVED
:
4702 for (i
= 0; i
< nr_cpu_ids
; i
++)
4703 if (net_dma
->channels
[i
] == chan
) {
4711 cpu_clear(pos
, net_dma
->channel_mask
);
4712 net_dma
->channels
[i
] = NULL
;
4713 net_dma_rebalance(net_dma
);
4719 spin_unlock(&net_dma
->lock
);
4725 * netdev_dma_register - register the networking subsystem as a DMA client
4727 static int __init
netdev_dma_register(void)
4729 net_dma
.channels
= kzalloc(nr_cpu_ids
* sizeof(struct net_dma
),
4731 if (unlikely(!net_dma
.channels
)) {
4733 "netdev_dma: no memory for net_dma.channels\n");
4736 spin_lock_init(&net_dma
.lock
);
4737 dma_cap_set(DMA_MEMCPY
, net_dma
.client
.cap_mask
);
4738 dma_async_client_register(&net_dma
.client
);
4739 dma_async_client_chan_request(&net_dma
.client
);
4744 static int __init
netdev_dma_register(void) { return -ENODEV
; }
4745 #endif /* CONFIG_NET_DMA */
4748 * netdev_increment_features - increment feature set by one
4749 * @all: current feature set
4750 * @one: new feature set
4751 * @mask: mask feature set
4753 * Computes a new feature set after adding a device with feature set
4754 * @one to the master device with current feature set @all. Will not
4755 * enable anything that is off in @mask. Returns the new feature set.
4757 unsigned long netdev_increment_features(unsigned long all
, unsigned long one
,
4760 /* If device needs checksumming, downgrade to it. */
4761 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
4762 all
^= NETIF_F_NO_CSUM
| (one
& NETIF_F_ALL_CSUM
);
4763 else if (mask
& NETIF_F_ALL_CSUM
) {
4764 /* If one device supports v4/v6 checksumming, set for all. */
4765 if (one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
) &&
4766 !(all
& NETIF_F_GEN_CSUM
)) {
4767 all
&= ~NETIF_F_ALL_CSUM
;
4768 all
|= one
& (NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
);
4771 /* If one device supports hw checksumming, set for all. */
4772 if (one
& NETIF_F_GEN_CSUM
&& !(all
& NETIF_F_GEN_CSUM
)) {
4773 all
&= ~NETIF_F_ALL_CSUM
;
4774 all
|= NETIF_F_HW_CSUM
;
4778 one
|= NETIF_F_ALL_CSUM
;
4780 one
|= all
& NETIF_F_ONE_FOR_ALL
;
4781 all
&= one
| NETIF_F_LLTX
| NETIF_F_GSO
;
4782 all
|= one
& mask
& NETIF_F_ONE_FOR_ALL
;
4786 EXPORT_SYMBOL(netdev_increment_features
);
4788 static struct hlist_head
*netdev_create_hash(void)
4791 struct hlist_head
*hash
;
4793 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
4795 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
4796 INIT_HLIST_HEAD(&hash
[i
]);
4801 /* Initialize per network namespace state */
4802 static int __net_init
netdev_init(struct net
*net
)
4804 INIT_LIST_HEAD(&net
->dev_base_head
);
4806 net
->dev_name_head
= netdev_create_hash();
4807 if (net
->dev_name_head
== NULL
)
4810 net
->dev_index_head
= netdev_create_hash();
4811 if (net
->dev_index_head
== NULL
)
4817 kfree(net
->dev_name_head
);
4823 * netdev_drivername - network driver for the device
4824 * @dev: network device
4825 * @buffer: buffer for resulting name
4826 * @len: size of buffer
4828 * Determine network driver for device.
4830 char *netdev_drivername(const struct net_device
*dev
, char *buffer
, int len
)
4832 const struct device_driver
*driver
;
4833 const struct device
*parent
;
4835 if (len
<= 0 || !buffer
)
4839 parent
= dev
->dev
.parent
;
4844 driver
= parent
->driver
;
4845 if (driver
&& driver
->name
)
4846 strlcpy(buffer
, driver
->name
, len
);
4850 static void __net_exit
netdev_exit(struct net
*net
)
4852 kfree(net
->dev_name_head
);
4853 kfree(net
->dev_index_head
);
4856 static struct pernet_operations __net_initdata netdev_net_ops
= {
4857 .init
= netdev_init
,
4858 .exit
= netdev_exit
,
4861 static void __net_exit
default_device_exit(struct net
*net
)
4863 struct net_device
*dev
, *next
;
4865 * Push all migratable of the network devices back to the
4866 * initial network namespace
4869 for_each_netdev_safe(net
, dev
, next
) {
4871 char fb_name
[IFNAMSIZ
];
4873 /* Ignore unmoveable devices (i.e. loopback) */
4874 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
4877 /* Delete virtual devices */
4878 if (dev
->rtnl_link_ops
&& dev
->rtnl_link_ops
->dellink
) {
4879 dev
->rtnl_link_ops
->dellink(dev
);
4883 /* Push remaing network devices to init_net */
4884 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
4885 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
4887 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
4888 __func__
, dev
->name
, err
);
4895 static struct pernet_operations __net_initdata default_device_ops
= {
4896 .exit
= default_device_exit
,
4900 * Initialize the DEV module. At boot time this walks the device list and
4901 * unhooks any devices that fail to initialise (normally hardware not
4902 * present) and leaves us with a valid list of present and active devices.
4907 * This is called single threaded during boot, so no need
4908 * to take the rtnl semaphore.
4910 static int __init
net_dev_init(void)
4912 int i
, rc
= -ENOMEM
;
4914 BUG_ON(!dev_boot_phase
);
4916 if (dev_proc_init())
4919 if (netdev_kobject_init())
4922 INIT_LIST_HEAD(&ptype_all
);
4923 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
4924 INIT_LIST_HEAD(&ptype_base
[i
]);
4926 if (register_pernet_subsys(&netdev_net_ops
))
4930 * Initialise the packet receive queues.
4933 for_each_possible_cpu(i
) {
4934 struct softnet_data
*queue
;
4936 queue
= &per_cpu(softnet_data
, i
);
4937 skb_queue_head_init(&queue
->input_pkt_queue
);
4938 queue
->completion_queue
= NULL
;
4939 INIT_LIST_HEAD(&queue
->poll_list
);
4941 queue
->backlog
.poll
= process_backlog
;
4942 queue
->backlog
.weight
= weight_p
;
4947 /* The loopback device is special if any other network devices
4948 * is present in a network namespace the loopback device must
4949 * be present. Since we now dynamically allocate and free the
4950 * loopback device ensure this invariant is maintained by
4951 * keeping the loopback device as the first device on the
4952 * list of network devices. Ensuring the loopback devices
4953 * is the first device that appears and the last network device
4956 if (register_pernet_device(&loopback_net_ops
))
4959 if (register_pernet_device(&default_device_ops
))
4962 netdev_dma_register();
4964 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
4965 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
4967 hotcpu_notifier(dev_cpu_callback
, 0);
4975 subsys_initcall(net_dev_init
);
4977 EXPORT_SYMBOL(__dev_get_by_index
);
4978 EXPORT_SYMBOL(__dev_get_by_name
);
4979 EXPORT_SYMBOL(__dev_remove_pack
);
4980 EXPORT_SYMBOL(dev_valid_name
);
4981 EXPORT_SYMBOL(dev_add_pack
);
4982 EXPORT_SYMBOL(dev_alloc_name
);
4983 EXPORT_SYMBOL(dev_close
);
4984 EXPORT_SYMBOL(dev_get_by_flags
);
4985 EXPORT_SYMBOL(dev_get_by_index
);
4986 EXPORT_SYMBOL(dev_get_by_name
);
4987 EXPORT_SYMBOL(dev_open
);
4988 EXPORT_SYMBOL(dev_queue_xmit
);
4989 EXPORT_SYMBOL(dev_remove_pack
);
4990 EXPORT_SYMBOL(dev_set_allmulti
);
4991 EXPORT_SYMBOL(dev_set_promiscuity
);
4992 EXPORT_SYMBOL(dev_change_flags
);
4993 EXPORT_SYMBOL(dev_set_mtu
);
4994 EXPORT_SYMBOL(dev_set_mac_address
);
4995 EXPORT_SYMBOL(free_netdev
);
4996 EXPORT_SYMBOL(netdev_boot_setup_check
);
4997 EXPORT_SYMBOL(netdev_set_master
);
4998 EXPORT_SYMBOL(netdev_state_change
);
4999 EXPORT_SYMBOL(netif_receive_skb
);
5000 EXPORT_SYMBOL(netif_rx
);
5001 EXPORT_SYMBOL(register_gifconf
);
5002 EXPORT_SYMBOL(register_netdevice
);
5003 EXPORT_SYMBOL(register_netdevice_notifier
);
5004 EXPORT_SYMBOL(skb_checksum_help
);
5005 EXPORT_SYMBOL(synchronize_net
);
5006 EXPORT_SYMBOL(unregister_netdevice
);
5007 EXPORT_SYMBOL(unregister_netdevice_notifier
);
5008 EXPORT_SYMBOL(net_enable_timestamp
);
5009 EXPORT_SYMBOL(net_disable_timestamp
);
5010 EXPORT_SYMBOL(dev_get_flags
);
5012 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
5013 EXPORT_SYMBOL(br_handle_frame_hook
);
5014 EXPORT_SYMBOL(br_fdb_get_hook
);
5015 EXPORT_SYMBOL(br_fdb_put_hook
);
5018 EXPORT_SYMBOL(dev_load
);
5020 EXPORT_PER_CPU_SYMBOL(softnet_data
);