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/kallsyms.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
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
, 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
);
919 strcpy(newname
, dev
->name
);
921 else if (__dev_get_by_name(net
, newname
))
924 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
927 err
= device_rename(&dev
->dev
, dev
->name
);
929 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
933 write_lock_bh(&dev_base_lock
);
934 hlist_del(&dev
->name_hlist
);
935 hlist_add_head(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
936 write_unlock_bh(&dev_base_lock
);
938 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
939 ret
= notifier_to_errno(ret
);
944 "%s: name change rollback failed: %d.\n",
948 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
957 * netdev_features_change - device changes features
958 * @dev: device to cause notification
960 * Called to indicate a device has changed features.
962 void netdev_features_change(struct net_device
*dev
)
964 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
966 EXPORT_SYMBOL(netdev_features_change
);
969 * netdev_state_change - device changes state
970 * @dev: device to cause notification
972 * Called to indicate a device has changed state. This function calls
973 * the notifier chains for netdev_chain and sends a NEWLINK message
974 * to the routing socket.
976 void netdev_state_change(struct net_device
*dev
)
978 if (dev
->flags
& IFF_UP
) {
979 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
980 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
984 void netdev_bonding_change(struct net_device
*dev
)
986 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, dev
);
988 EXPORT_SYMBOL(netdev_bonding_change
);
991 * dev_load - load a network module
992 * @net: the applicable net namespace
993 * @name: name of interface
995 * If a network interface is not present and the process has suitable
996 * privileges this function loads the module. If module loading is not
997 * available in this kernel then it becomes a nop.
1000 void dev_load(struct net
*net
, const char *name
)
1002 struct net_device
*dev
;
1004 read_lock(&dev_base_lock
);
1005 dev
= __dev_get_by_name(net
, name
);
1006 read_unlock(&dev_base_lock
);
1008 if (!dev
&& capable(CAP_SYS_MODULE
))
1009 request_module("%s", name
);
1013 * dev_open - prepare an interface for use.
1014 * @dev: device to open
1016 * Takes a device from down to up state. The device's private open
1017 * function is invoked and then the multicast lists are loaded. Finally
1018 * the device is moved into the up state and a %NETDEV_UP message is
1019 * sent to the netdev notifier chain.
1021 * Calling this function on an active interface is a nop. On a failure
1022 * a negative errno code is returned.
1024 int dev_open(struct net_device
*dev
)
1034 if (dev
->flags
& IFF_UP
)
1038 * Is it even present?
1040 if (!netif_device_present(dev
))
1044 * Call device private open method
1046 set_bit(__LINK_STATE_START
, &dev
->state
);
1048 if (dev
->validate_addr
)
1049 ret
= dev
->validate_addr(dev
);
1051 if (!ret
&& dev
->open
)
1052 ret
= dev
->open(dev
);
1055 * If it went open OK then:
1059 clear_bit(__LINK_STATE_START
, &dev
->state
);
1064 dev
->flags
|= IFF_UP
;
1067 * Initialize multicasting status
1069 dev_set_rx_mode(dev
);
1072 * Wakeup transmit queue engine
1077 * ... and announce new interface.
1079 call_netdevice_notifiers(NETDEV_UP
, dev
);
1086 * dev_close - shutdown an interface.
1087 * @dev: device to shutdown
1089 * This function moves an active device into down state. A
1090 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1091 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1094 int dev_close(struct net_device
*dev
)
1100 if (!(dev
->flags
& IFF_UP
))
1104 * Tell people we are going down, so that they can
1105 * prepare to death, when device is still operating.
1107 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1109 clear_bit(__LINK_STATE_START
, &dev
->state
);
1111 /* Synchronize to scheduled poll. We cannot touch poll list,
1112 * it can be even on different cpu. So just clear netif_running().
1114 * dev->stop() will invoke napi_disable() on all of it's
1115 * napi_struct instances on this device.
1117 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1119 dev_deactivate(dev
);
1122 * Call the device specific close. This cannot fail.
1123 * Only if device is UP
1125 * We allow it to be called even after a DETACH hot-plug
1132 * Device is now down.
1135 dev
->flags
&= ~IFF_UP
;
1138 * Tell people we are down
1140 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1147 * dev_disable_lro - disable Large Receive Offload on a device
1150 * Disable Large Receive Offload (LRO) on a net device. Must be
1151 * called under RTNL. This is needed if received packets may be
1152 * forwarded to another interface.
1154 void dev_disable_lro(struct net_device
*dev
)
1156 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1157 dev
->ethtool_ops
->set_flags
) {
1158 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1159 if (flags
& ETH_FLAG_LRO
) {
1160 flags
&= ~ETH_FLAG_LRO
;
1161 dev
->ethtool_ops
->set_flags(dev
, flags
);
1164 WARN_ON(dev
->features
& NETIF_F_LRO
);
1166 EXPORT_SYMBOL(dev_disable_lro
);
1169 static int dev_boot_phase
= 1;
1172 * Device change register/unregister. These are not inline or static
1173 * as we export them to the world.
1177 * register_netdevice_notifier - register a network notifier block
1180 * Register a notifier to be called when network device events occur.
1181 * The notifier passed is linked into the kernel structures and must
1182 * not be reused until it has been unregistered. A negative errno code
1183 * is returned on a failure.
1185 * When registered all registration and up events are replayed
1186 * to the new notifier to allow device to have a race free
1187 * view of the network device list.
1190 int register_netdevice_notifier(struct notifier_block
*nb
)
1192 struct net_device
*dev
;
1193 struct net_device
*last
;
1198 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1204 for_each_netdev(net
, dev
) {
1205 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1206 err
= notifier_to_errno(err
);
1210 if (!(dev
->flags
& IFF_UP
))
1213 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1224 for_each_netdev(net
, dev
) {
1228 if (dev
->flags
& IFF_UP
) {
1229 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1230 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1232 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1236 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1241 * unregister_netdevice_notifier - unregister a network notifier block
1244 * Unregister a notifier previously registered by
1245 * register_netdevice_notifier(). The notifier is unlinked into the
1246 * kernel structures and may then be reused. A negative errno code
1247 * is returned on a failure.
1250 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1255 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1261 * call_netdevice_notifiers - call all network notifier blocks
1262 * @val: value passed unmodified to notifier function
1263 * @dev: net_device pointer passed unmodified to notifier function
1265 * Call all network notifier blocks. Parameters and return value
1266 * are as for raw_notifier_call_chain().
1269 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1271 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1274 /* When > 0 there are consumers of rx skb time stamps */
1275 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1277 void net_enable_timestamp(void)
1279 atomic_inc(&netstamp_needed
);
1282 void net_disable_timestamp(void)
1284 atomic_dec(&netstamp_needed
);
1287 static inline void net_timestamp(struct sk_buff
*skb
)
1289 if (atomic_read(&netstamp_needed
))
1290 __net_timestamp(skb
);
1292 skb
->tstamp
.tv64
= 0;
1296 * Support routine. Sends outgoing frames to any network
1297 * taps currently in use.
1300 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1302 struct packet_type
*ptype
;
1307 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1308 /* Never send packets back to the socket
1309 * they originated from - MvS (miquels@drinkel.ow.org)
1311 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1312 (ptype
->af_packet_priv
== NULL
||
1313 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1314 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1318 /* skb->nh should be correctly
1319 set by sender, so that the second statement is
1320 just protection against buggy protocols.
1322 skb_reset_mac_header(skb2
);
1324 if (skb_network_header(skb2
) < skb2
->data
||
1325 skb2
->network_header
> skb2
->tail
) {
1326 if (net_ratelimit())
1327 printk(KERN_CRIT
"protocol %04x is "
1329 skb2
->protocol
, dev
->name
);
1330 skb_reset_network_header(skb2
);
1333 skb2
->transport_header
= skb2
->network_header
;
1334 skb2
->pkt_type
= PACKET_OUTGOING
;
1335 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1342 void __netif_schedule(struct Qdisc
*q
)
1344 if (WARN_ON_ONCE(q
== &noop_qdisc
))
1347 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
)) {
1348 struct softnet_data
*sd
;
1349 unsigned long flags
;
1351 local_irq_save(flags
);
1352 sd
= &__get_cpu_var(softnet_data
);
1353 q
->next_sched
= sd
->output_queue
;
1354 sd
->output_queue
= q
;
1355 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1356 local_irq_restore(flags
);
1359 EXPORT_SYMBOL(__netif_schedule
);
1361 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1363 if (atomic_dec_and_test(&skb
->users
)) {
1364 struct softnet_data
*sd
;
1365 unsigned long flags
;
1367 local_irq_save(flags
);
1368 sd
= &__get_cpu_var(softnet_data
);
1369 skb
->next
= sd
->completion_queue
;
1370 sd
->completion_queue
= skb
;
1371 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1372 local_irq_restore(flags
);
1375 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1377 void dev_kfree_skb_any(struct sk_buff
*skb
)
1379 if (in_irq() || irqs_disabled())
1380 dev_kfree_skb_irq(skb
);
1384 EXPORT_SYMBOL(dev_kfree_skb_any
);
1388 * netif_device_detach - mark device as removed
1389 * @dev: network device
1391 * Mark device as removed from system and therefore no longer available.
1393 void netif_device_detach(struct net_device
*dev
)
1395 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1396 netif_running(dev
)) {
1397 netif_stop_queue(dev
);
1400 EXPORT_SYMBOL(netif_device_detach
);
1403 * netif_device_attach - mark device as attached
1404 * @dev: network device
1406 * Mark device as attached from system and restart if needed.
1408 void netif_device_attach(struct net_device
*dev
)
1410 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1411 netif_running(dev
)) {
1412 netif_wake_queue(dev
);
1413 __netdev_watchdog_up(dev
);
1416 EXPORT_SYMBOL(netif_device_attach
);
1418 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1420 return ((features
& NETIF_F_GEN_CSUM
) ||
1421 ((features
& NETIF_F_IP_CSUM
) &&
1422 protocol
== htons(ETH_P_IP
)) ||
1423 ((features
& NETIF_F_IPV6_CSUM
) &&
1424 protocol
== htons(ETH_P_IPV6
)));
1427 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1429 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1432 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1433 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1434 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1435 veh
->h_vlan_encapsulated_proto
))
1443 * Invalidate hardware checksum when packet is to be mangled, and
1444 * complete checksum manually on outgoing path.
1446 int skb_checksum_help(struct sk_buff
*skb
)
1449 int ret
= 0, offset
;
1451 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1452 goto out_set_summed
;
1454 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1455 /* Let GSO fix up the checksum. */
1456 goto out_set_summed
;
1459 offset
= skb
->csum_start
- skb_headroom(skb
);
1460 BUG_ON(offset
>= skb_headlen(skb
));
1461 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1463 offset
+= skb
->csum_offset
;
1464 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1466 if (skb_cloned(skb
) &&
1467 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1468 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1473 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1475 skb
->ip_summed
= CHECKSUM_NONE
;
1481 * skb_gso_segment - Perform segmentation on skb.
1482 * @skb: buffer to segment
1483 * @features: features for the output path (see dev->features)
1485 * This function segments the given skb and returns a list of segments.
1487 * It may return NULL if the skb requires no segmentation. This is
1488 * only possible when GSO is used for verifying header integrity.
1490 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1492 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1493 struct packet_type
*ptype
;
1494 __be16 type
= skb
->protocol
;
1497 BUG_ON(skb_shinfo(skb
)->frag_list
);
1499 skb_reset_mac_header(skb
);
1500 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1501 __skb_pull(skb
, skb
->mac_len
);
1503 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1504 if (skb_header_cloned(skb
) &&
1505 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1506 return ERR_PTR(err
);
1510 list_for_each_entry_rcu(ptype
,
1511 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1512 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1513 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1514 err
= ptype
->gso_send_check(skb
);
1515 segs
= ERR_PTR(err
);
1516 if (err
|| skb_gso_ok(skb
, features
))
1518 __skb_push(skb
, (skb
->data
-
1519 skb_network_header(skb
)));
1521 segs
= ptype
->gso_segment(skb
, features
);
1527 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1532 EXPORT_SYMBOL(skb_gso_segment
);
1534 /* Take action when hardware reception checksum errors are detected. */
1536 void netdev_rx_csum_fault(struct net_device
*dev
)
1538 if (net_ratelimit()) {
1539 printk(KERN_ERR
"%s: hw csum failure.\n",
1540 dev
? dev
->name
: "<unknown>");
1544 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1547 /* Actually, we should eliminate this check as soon as we know, that:
1548 * 1. IOMMU is present and allows to map all the memory.
1549 * 2. No high memory really exists on this machine.
1552 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1554 #ifdef CONFIG_HIGHMEM
1557 if (dev
->features
& NETIF_F_HIGHDMA
)
1560 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1561 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1569 void (*destructor
)(struct sk_buff
*skb
);
1572 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1574 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1576 struct dev_gso_cb
*cb
;
1579 struct sk_buff
*nskb
= skb
->next
;
1581 skb
->next
= nskb
->next
;
1584 } while (skb
->next
);
1586 cb
= DEV_GSO_CB(skb
);
1588 cb
->destructor(skb
);
1592 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1593 * @skb: buffer to segment
1595 * This function segments the given skb and stores the list of segments
1598 static int dev_gso_segment(struct sk_buff
*skb
)
1600 struct net_device
*dev
= skb
->dev
;
1601 struct sk_buff
*segs
;
1602 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1605 segs
= skb_gso_segment(skb
, features
);
1607 /* Verifying header integrity only. */
1612 return PTR_ERR(segs
);
1615 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1616 skb
->destructor
= dev_gso_skb_destructor
;
1621 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
1622 struct netdev_queue
*txq
)
1624 if (likely(!skb
->next
)) {
1625 if (!list_empty(&ptype_all
))
1626 dev_queue_xmit_nit(skb
, dev
);
1628 if (netif_needs_gso(dev
, skb
)) {
1629 if (unlikely(dev_gso_segment(skb
)))
1635 return dev
->hard_start_xmit(skb
, dev
);
1640 struct sk_buff
*nskb
= skb
->next
;
1643 skb
->next
= nskb
->next
;
1645 rc
= dev
->hard_start_xmit(nskb
, dev
);
1647 nskb
->next
= skb
->next
;
1651 if (unlikely(netif_tx_queue_stopped(txq
) && skb
->next
))
1652 return NETDEV_TX_BUSY
;
1653 } while (skb
->next
);
1655 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1662 static u32 simple_tx_hashrnd
;
1663 static int simple_tx_hashrnd_initialized
= 0;
1665 static u16
simple_tx_hash(struct net_device
*dev
, struct sk_buff
*skb
)
1667 u32 addr1
, addr2
, ports
;
1671 if (unlikely(!simple_tx_hashrnd_initialized
)) {
1672 get_random_bytes(&simple_tx_hashrnd
, 4);
1673 simple_tx_hashrnd_initialized
= 1;
1676 switch (skb
->protocol
) {
1677 case __constant_htons(ETH_P_IP
):
1678 ip_proto
= ip_hdr(skb
)->protocol
;
1679 addr1
= ip_hdr(skb
)->saddr
;
1680 addr2
= ip_hdr(skb
)->daddr
;
1681 ihl
= ip_hdr(skb
)->ihl
;
1683 case __constant_htons(ETH_P_IPV6
):
1684 ip_proto
= ipv6_hdr(skb
)->nexthdr
;
1685 addr1
= ipv6_hdr(skb
)->saddr
.s6_addr32
[3];
1686 addr2
= ipv6_hdr(skb
)->daddr
.s6_addr32
[3];
1701 case IPPROTO_UDPLITE
:
1702 ports
= *((u32
*) (skb_network_header(skb
) + (ihl
* 4)));
1710 hash
= jhash_3words(addr1
, addr2
, ports
, simple_tx_hashrnd
);
1712 return (u16
) (((u64
) hash
* dev
->real_num_tx_queues
) >> 32);
1715 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
1716 struct sk_buff
*skb
)
1718 u16 queue_index
= 0;
1720 if (dev
->select_queue
)
1721 queue_index
= dev
->select_queue(dev
, skb
);
1722 else if (dev
->real_num_tx_queues
> 1)
1723 queue_index
= simple_tx_hash(dev
, skb
);
1725 skb_set_queue_mapping(skb
, queue_index
);
1726 return netdev_get_tx_queue(dev
, queue_index
);
1730 * dev_queue_xmit - transmit a buffer
1731 * @skb: buffer to transmit
1733 * Queue a buffer for transmission to a network device. The caller must
1734 * have set the device and priority and built the buffer before calling
1735 * this function. The function can be called from an interrupt.
1737 * A negative errno code is returned on a failure. A success does not
1738 * guarantee the frame will be transmitted as it may be dropped due
1739 * to congestion or traffic shaping.
1741 * -----------------------------------------------------------------------------------
1742 * I notice this method can also return errors from the queue disciplines,
1743 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1746 * Regardless of the return value, the skb is consumed, so it is currently
1747 * difficult to retry a send to this method. (You can bump the ref count
1748 * before sending to hold a reference for retry if you are careful.)
1750 * When calling this method, interrupts MUST be enabled. This is because
1751 * the BH enable code must have IRQs enabled so that it will not deadlock.
1754 int dev_queue_xmit(struct sk_buff
*skb
)
1756 struct net_device
*dev
= skb
->dev
;
1757 struct netdev_queue
*txq
;
1761 /* GSO will handle the following emulations directly. */
1762 if (netif_needs_gso(dev
, skb
))
1765 if (skb_shinfo(skb
)->frag_list
&&
1766 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1767 __skb_linearize(skb
))
1770 /* Fragmented skb is linearized if device does not support SG,
1771 * or if at least one of fragments is in highmem and device
1772 * does not support DMA from it.
1774 if (skb_shinfo(skb
)->nr_frags
&&
1775 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1776 __skb_linearize(skb
))
1779 /* If packet is not checksummed and device does not support
1780 * checksumming for this protocol, complete checksumming here.
1782 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1783 skb_set_transport_header(skb
, skb
->csum_start
-
1785 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
1790 /* Disable soft irqs for various locks below. Also
1791 * stops preemption for RCU.
1795 txq
= dev_pick_tx(dev
, skb
);
1796 q
= rcu_dereference(txq
->qdisc
);
1798 #ifdef CONFIG_NET_CLS_ACT
1799 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1802 spinlock_t
*root_lock
= qdisc_root_lock(q
);
1804 spin_lock(root_lock
);
1806 rc
= qdisc_enqueue_root(skb
, q
);
1809 spin_unlock(root_lock
);
1811 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1815 /* The device has no queue. Common case for software devices:
1816 loopback, all the sorts of tunnels...
1818 Really, it is unlikely that netif_tx_lock protection is necessary
1819 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1821 However, it is possible, that they rely on protection
1824 Check this and shot the lock. It is not prone from deadlocks.
1825 Either shot noqueue qdisc, it is even simpler 8)
1827 if (dev
->flags
& IFF_UP
) {
1828 int cpu
= smp_processor_id(); /* ok because BHs are off */
1830 if (txq
->xmit_lock_owner
!= cpu
) {
1832 HARD_TX_LOCK(dev
, txq
, cpu
);
1834 if (!netif_tx_queue_stopped(txq
)) {
1836 if (!dev_hard_start_xmit(skb
, dev
, txq
)) {
1837 HARD_TX_UNLOCK(dev
, txq
);
1841 HARD_TX_UNLOCK(dev
, txq
);
1842 if (net_ratelimit())
1843 printk(KERN_CRIT
"Virtual device %s asks to "
1844 "queue packet!\n", dev
->name
);
1846 /* Recursion is detected! It is possible,
1848 if (net_ratelimit())
1849 printk(KERN_CRIT
"Dead loop on virtual device "
1850 "%s, fix it urgently!\n", dev
->name
);
1855 rcu_read_unlock_bh();
1861 rcu_read_unlock_bh();
1866 /*=======================================================================
1868 =======================================================================*/
1870 int netdev_max_backlog __read_mostly
= 1000;
1871 int netdev_budget __read_mostly
= 300;
1872 int weight_p __read_mostly
= 64; /* old backlog weight */
1874 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1878 * netif_rx - post buffer to the network code
1879 * @skb: buffer to post
1881 * This function receives a packet from a device driver and queues it for
1882 * the upper (protocol) levels to process. It always succeeds. The buffer
1883 * may be dropped during processing for congestion control or by the
1887 * NET_RX_SUCCESS (no congestion)
1888 * NET_RX_DROP (packet was dropped)
1892 int netif_rx(struct sk_buff
*skb
)
1894 struct softnet_data
*queue
;
1895 unsigned long flags
;
1897 /* if netpoll wants it, pretend we never saw it */
1898 if (netpoll_rx(skb
))
1901 if (!skb
->tstamp
.tv64
)
1905 * The code is rearranged so that the path is the most
1906 * short when CPU is congested, but is still operating.
1908 local_irq_save(flags
);
1909 queue
= &__get_cpu_var(softnet_data
);
1911 __get_cpu_var(netdev_rx_stat
).total
++;
1912 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1913 if (queue
->input_pkt_queue
.qlen
) {
1916 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1917 local_irq_restore(flags
);
1918 return NET_RX_SUCCESS
;
1921 napi_schedule(&queue
->backlog
);
1925 __get_cpu_var(netdev_rx_stat
).dropped
++;
1926 local_irq_restore(flags
);
1932 int netif_rx_ni(struct sk_buff
*skb
)
1937 err
= netif_rx(skb
);
1938 if (local_softirq_pending())
1945 EXPORT_SYMBOL(netif_rx_ni
);
1947 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1949 struct net_device
*dev
= skb
->dev
;
1952 if (skb_bond_should_drop(skb
)) {
1956 skb
->dev
= dev
->master
;
1963 static void net_tx_action(struct softirq_action
*h
)
1965 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1967 if (sd
->completion_queue
) {
1968 struct sk_buff
*clist
;
1970 local_irq_disable();
1971 clist
= sd
->completion_queue
;
1972 sd
->completion_queue
= NULL
;
1976 struct sk_buff
*skb
= clist
;
1977 clist
= clist
->next
;
1979 BUG_TRAP(!atomic_read(&skb
->users
));
1984 if (sd
->output_queue
) {
1987 local_irq_disable();
1988 head
= sd
->output_queue
;
1989 sd
->output_queue
= NULL
;
1993 struct Qdisc
*q
= head
;
1994 spinlock_t
*root_lock
;
1996 head
= head
->next_sched
;
1998 smp_mb__before_clear_bit();
1999 clear_bit(__QDISC_STATE_SCHED
, &q
->state
);
2001 root_lock
= qdisc_root_lock(q
);
2002 if (spin_trylock(root_lock
)) {
2004 spin_unlock(root_lock
);
2006 __netif_schedule(q
);
2012 static inline int deliver_skb(struct sk_buff
*skb
,
2013 struct packet_type
*pt_prev
,
2014 struct net_device
*orig_dev
)
2016 atomic_inc(&skb
->users
);
2017 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2020 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2021 /* These hooks defined here for ATM */
2023 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
2024 unsigned char *addr
);
2025 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
2028 * If bridge module is loaded call bridging hook.
2029 * returns NULL if packet was consumed.
2031 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
2032 struct sk_buff
*skb
) __read_mostly
;
2033 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
2034 struct packet_type
**pt_prev
, int *ret
,
2035 struct net_device
*orig_dev
)
2037 struct net_bridge_port
*port
;
2039 if (skb
->pkt_type
== PACKET_LOOPBACK
||
2040 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
2044 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2048 return br_handle_frame_hook(port
, skb
);
2051 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2054 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2055 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
2056 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
2058 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
2059 struct packet_type
**pt_prev
,
2061 struct net_device
*orig_dev
)
2063 if (skb
->dev
->macvlan_port
== NULL
)
2067 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2070 return macvlan_handle_frame_hook(skb
);
2073 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2076 #ifdef CONFIG_NET_CLS_ACT
2077 /* TODO: Maybe we should just force sch_ingress to be compiled in
2078 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2079 * a compare and 2 stores extra right now if we dont have it on
2080 * but have CONFIG_NET_CLS_ACT
2081 * NOTE: This doesnt stop any functionality; if you dont have
2082 * the ingress scheduler, you just cant add policies on ingress.
2085 static int ing_filter(struct sk_buff
*skb
)
2087 struct net_device
*dev
= skb
->dev
;
2088 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2089 struct netdev_queue
*rxq
;
2090 int result
= TC_ACT_OK
;
2093 if (MAX_RED_LOOP
< ttl
++) {
2095 "Redir loop detected Dropping packet (%d->%d)\n",
2096 skb
->iif
, dev
->ifindex
);
2100 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2101 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2103 rxq
= &dev
->rx_queue
;
2107 spin_lock(qdisc_lock(q
));
2108 result
= qdisc_enqueue_root(skb
, q
);
2109 spin_unlock(qdisc_lock(q
));
2115 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2116 struct packet_type
**pt_prev
,
2117 int *ret
, struct net_device
*orig_dev
)
2119 if (!skb
->dev
->rx_queue
.qdisc
)
2123 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2126 /* Huh? Why does turning on AF_PACKET affect this? */
2127 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2130 switch (ing_filter(skb
)) {
2144 * netif_nit_deliver - deliver received packets to network taps
2147 * This function is used to deliver incoming packets to network
2148 * taps. It should be used when the normal netif_receive_skb path
2149 * is bypassed, for example because of VLAN acceleration.
2151 void netif_nit_deliver(struct sk_buff
*skb
)
2153 struct packet_type
*ptype
;
2155 if (list_empty(&ptype_all
))
2158 skb_reset_network_header(skb
);
2159 skb_reset_transport_header(skb
);
2160 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2163 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2164 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2165 deliver_skb(skb
, ptype
, skb
->dev
);
2171 * netif_receive_skb - process receive buffer from network
2172 * @skb: buffer to process
2174 * netif_receive_skb() is the main receive data processing function.
2175 * It always succeeds. The buffer may be dropped during processing
2176 * for congestion control or by the protocol layers.
2178 * This function may only be called from softirq context and interrupts
2179 * should be enabled.
2181 * Return values (usually ignored):
2182 * NET_RX_SUCCESS: no congestion
2183 * NET_RX_DROP: packet was dropped
2185 int netif_receive_skb(struct sk_buff
*skb
)
2187 struct packet_type
*ptype
, *pt_prev
;
2188 struct net_device
*orig_dev
;
2189 int ret
= NET_RX_DROP
;
2192 /* if we've gotten here through NAPI, check netpoll */
2193 if (netpoll_receive_skb(skb
))
2196 if (!skb
->tstamp
.tv64
)
2200 skb
->iif
= skb
->dev
->ifindex
;
2202 orig_dev
= skb_bond(skb
);
2207 __get_cpu_var(netdev_rx_stat
).total
++;
2209 skb_reset_network_header(skb
);
2210 skb_reset_transport_header(skb
);
2211 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2217 /* Don't receive packets in an exiting network namespace */
2218 if (!net_alive(dev_net(skb
->dev
)))
2221 #ifdef CONFIG_NET_CLS_ACT
2222 if (skb
->tc_verd
& TC_NCLS
) {
2223 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2228 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2229 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
2231 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2236 #ifdef CONFIG_NET_CLS_ACT
2237 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2243 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
2246 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
2250 type
= skb
->protocol
;
2251 list_for_each_entry_rcu(ptype
,
2252 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2253 if (ptype
->type
== type
&&
2254 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
2256 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2262 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2265 /* Jamal, now you will not able to escape explaining
2266 * me how you were going to use this. :-)
2276 static int process_backlog(struct napi_struct
*napi
, int quota
)
2279 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2280 unsigned long start_time
= jiffies
;
2282 napi
->weight
= weight_p
;
2284 struct sk_buff
*skb
;
2285 struct net_device
*dev
;
2287 local_irq_disable();
2288 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
2290 __napi_complete(napi
);
2299 netif_receive_skb(skb
);
2302 } while (++work
< quota
&& jiffies
== start_time
);
2308 * __napi_schedule - schedule for receive
2309 * @n: entry to schedule
2311 * The entry's receive function will be scheduled to run
2313 void __napi_schedule(struct napi_struct
*n
)
2315 unsigned long flags
;
2317 local_irq_save(flags
);
2318 list_add_tail(&n
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
2319 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2320 local_irq_restore(flags
);
2322 EXPORT_SYMBOL(__napi_schedule
);
2325 static void net_rx_action(struct softirq_action
*h
)
2327 struct list_head
*list
= &__get_cpu_var(softnet_data
).poll_list
;
2328 unsigned long start_time
= jiffies
;
2329 int budget
= netdev_budget
;
2332 local_irq_disable();
2334 while (!list_empty(list
)) {
2335 struct napi_struct
*n
;
2338 /* If softirq window is exhuasted then punt.
2340 * Note that this is a slight policy change from the
2341 * previous NAPI code, which would allow up to 2
2342 * jiffies to pass before breaking out. The test
2343 * used to be "jiffies - start_time > 1".
2345 if (unlikely(budget
<= 0 || jiffies
!= start_time
))
2350 /* Even though interrupts have been re-enabled, this
2351 * access is safe because interrupts can only add new
2352 * entries to the tail of this list, and only ->poll()
2353 * calls can remove this head entry from the list.
2355 n
= list_entry(list
->next
, struct napi_struct
, poll_list
);
2357 have
= netpoll_poll_lock(n
);
2361 /* This NAPI_STATE_SCHED test is for avoiding a race
2362 * with netpoll's poll_napi(). Only the entity which
2363 * obtains the lock and sees NAPI_STATE_SCHED set will
2364 * actually make the ->poll() call. Therefore we avoid
2365 * accidently calling ->poll() when NAPI is not scheduled.
2368 if (test_bit(NAPI_STATE_SCHED
, &n
->state
))
2369 work
= n
->poll(n
, weight
);
2371 WARN_ON_ONCE(work
> weight
);
2375 local_irq_disable();
2377 /* Drivers must not modify the NAPI state if they
2378 * consume the entire weight. In such cases this code
2379 * still "owns" the NAPI instance and therefore can
2380 * move the instance around on the list at-will.
2382 if (unlikely(work
== weight
)) {
2383 if (unlikely(napi_disable_pending(n
)))
2386 list_move_tail(&n
->poll_list
, list
);
2389 netpoll_poll_unlock(have
);
2394 #ifdef CONFIG_NET_DMA
2396 * There may not be any more sk_buffs coming right now, so push
2397 * any pending DMA copies to hardware
2399 if (!cpus_empty(net_dma
.channel_mask
)) {
2401 for_each_cpu_mask_nr(chan_idx
, net_dma
.channel_mask
) {
2402 struct dma_chan
*chan
= net_dma
.channels
[chan_idx
];
2404 dma_async_memcpy_issue_pending(chan
);
2412 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2413 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2417 static gifconf_func_t
* gifconf_list
[NPROTO
];
2420 * register_gifconf - register a SIOCGIF handler
2421 * @family: Address family
2422 * @gifconf: Function handler
2424 * Register protocol dependent address dumping routines. The handler
2425 * that is passed must not be freed or reused until it has been replaced
2426 * by another handler.
2428 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2430 if (family
>= NPROTO
)
2432 gifconf_list
[family
] = gifconf
;
2438 * Map an interface index to its name (SIOCGIFNAME)
2442 * We need this ioctl for efficient implementation of the
2443 * if_indextoname() function required by the IPv6 API. Without
2444 * it, we would have to search all the interfaces to find a
2448 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
2450 struct net_device
*dev
;
2454 * Fetch the caller's info block.
2457 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2460 read_lock(&dev_base_lock
);
2461 dev
= __dev_get_by_index(net
, ifr
.ifr_ifindex
);
2463 read_unlock(&dev_base_lock
);
2467 strcpy(ifr
.ifr_name
, dev
->name
);
2468 read_unlock(&dev_base_lock
);
2470 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2476 * Perform a SIOCGIFCONF call. This structure will change
2477 * size eventually, and there is nothing I can do about it.
2478 * Thus we will need a 'compatibility mode'.
2481 static int dev_ifconf(struct net
*net
, char __user
*arg
)
2484 struct net_device
*dev
;
2491 * Fetch the caller's info block.
2494 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2501 * Loop over the interfaces, and write an info block for each.
2505 for_each_netdev(net
, dev
) {
2506 for (i
= 0; i
< NPROTO
; i
++) {
2507 if (gifconf_list
[i
]) {
2510 done
= gifconf_list
[i
](dev
, NULL
, 0);
2512 done
= gifconf_list
[i
](dev
, pos
+ total
,
2522 * All done. Write the updated control block back to the caller.
2524 ifc
.ifc_len
= total
;
2527 * Both BSD and Solaris return 0 here, so we do too.
2529 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2532 #ifdef CONFIG_PROC_FS
2534 * This is invoked by the /proc filesystem handler to display a device
2537 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2538 __acquires(dev_base_lock
)
2540 struct net
*net
= seq_file_net(seq
);
2542 struct net_device
*dev
;
2544 read_lock(&dev_base_lock
);
2546 return SEQ_START_TOKEN
;
2549 for_each_netdev(net
, dev
)
2556 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2558 struct net
*net
= seq_file_net(seq
);
2560 return v
== SEQ_START_TOKEN
?
2561 first_net_device(net
) : next_net_device((struct net_device
*)v
);
2564 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2565 __releases(dev_base_lock
)
2567 read_unlock(&dev_base_lock
);
2570 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2572 struct net_device_stats
*stats
= dev
->get_stats(dev
);
2574 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2575 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2576 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2578 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2579 stats
->rx_fifo_errors
,
2580 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2581 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2582 stats
->rx_compressed
, stats
->multicast
,
2583 stats
->tx_bytes
, stats
->tx_packets
,
2584 stats
->tx_errors
, stats
->tx_dropped
,
2585 stats
->tx_fifo_errors
, stats
->collisions
,
2586 stats
->tx_carrier_errors
+
2587 stats
->tx_aborted_errors
+
2588 stats
->tx_window_errors
+
2589 stats
->tx_heartbeat_errors
,
2590 stats
->tx_compressed
);
2594 * Called from the PROCfs module. This now uses the new arbitrary sized
2595 * /proc/net interface to create /proc/net/dev
2597 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2599 if (v
== SEQ_START_TOKEN
)
2600 seq_puts(seq
, "Inter-| Receive "
2602 " face |bytes packets errs drop fifo frame "
2603 "compressed multicast|bytes packets errs "
2604 "drop fifo colls carrier compressed\n");
2606 dev_seq_printf_stats(seq
, v
);
2610 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2612 struct netif_rx_stats
*rc
= NULL
;
2614 while (*pos
< nr_cpu_ids
)
2615 if (cpu_online(*pos
)) {
2616 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2623 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2625 return softnet_get_online(pos
);
2628 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2631 return softnet_get_online(pos
);
2634 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2638 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2640 struct netif_rx_stats
*s
= v
;
2642 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2643 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2644 0, 0, 0, 0, /* was fastroute */
2649 static const struct seq_operations dev_seq_ops
= {
2650 .start
= dev_seq_start
,
2651 .next
= dev_seq_next
,
2652 .stop
= dev_seq_stop
,
2653 .show
= dev_seq_show
,
2656 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2658 return seq_open_net(inode
, file
, &dev_seq_ops
,
2659 sizeof(struct seq_net_private
));
2662 static const struct file_operations dev_seq_fops
= {
2663 .owner
= THIS_MODULE
,
2664 .open
= dev_seq_open
,
2666 .llseek
= seq_lseek
,
2667 .release
= seq_release_net
,
2670 static const struct seq_operations softnet_seq_ops
= {
2671 .start
= softnet_seq_start
,
2672 .next
= softnet_seq_next
,
2673 .stop
= softnet_seq_stop
,
2674 .show
= softnet_seq_show
,
2677 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2679 return seq_open(file
, &softnet_seq_ops
);
2682 static const struct file_operations softnet_seq_fops
= {
2683 .owner
= THIS_MODULE
,
2684 .open
= softnet_seq_open
,
2686 .llseek
= seq_lseek
,
2687 .release
= seq_release
,
2690 static void *ptype_get_idx(loff_t pos
)
2692 struct packet_type
*pt
= NULL
;
2696 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
2702 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
2703 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
2712 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2716 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
2719 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2721 struct packet_type
*pt
;
2722 struct list_head
*nxt
;
2726 if (v
== SEQ_START_TOKEN
)
2727 return ptype_get_idx(0);
2730 nxt
= pt
->list
.next
;
2731 if (pt
->type
== htons(ETH_P_ALL
)) {
2732 if (nxt
!= &ptype_all
)
2735 nxt
= ptype_base
[0].next
;
2737 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
2739 while (nxt
== &ptype_base
[hash
]) {
2740 if (++hash
>= PTYPE_HASH_SIZE
)
2742 nxt
= ptype_base
[hash
].next
;
2745 return list_entry(nxt
, struct packet_type
, list
);
2748 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
2754 static void ptype_seq_decode(struct seq_file
*seq
, void *sym
)
2756 #ifdef CONFIG_KALLSYMS
2757 unsigned long offset
= 0, symsize
;
2758 const char *symname
;
2762 symname
= kallsyms_lookup((unsigned long)sym
, &symsize
, &offset
,
2769 modname
= delim
= "";
2770 seq_printf(seq
, "%s%s%s%s+0x%lx", delim
, modname
, delim
,
2776 seq_printf(seq
, "[%p]", sym
);
2779 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
2781 struct packet_type
*pt
= v
;
2783 if (v
== SEQ_START_TOKEN
)
2784 seq_puts(seq
, "Type Device Function\n");
2785 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
2786 if (pt
->type
== htons(ETH_P_ALL
))
2787 seq_puts(seq
, "ALL ");
2789 seq_printf(seq
, "%04x", ntohs(pt
->type
));
2791 seq_printf(seq
, " %-8s ",
2792 pt
->dev
? pt
->dev
->name
: "");
2793 ptype_seq_decode(seq
, pt
->func
);
2794 seq_putc(seq
, '\n');
2800 static const struct seq_operations ptype_seq_ops
= {
2801 .start
= ptype_seq_start
,
2802 .next
= ptype_seq_next
,
2803 .stop
= ptype_seq_stop
,
2804 .show
= ptype_seq_show
,
2807 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
2809 return seq_open_net(inode
, file
, &ptype_seq_ops
,
2810 sizeof(struct seq_net_private
));
2813 static const struct file_operations ptype_seq_fops
= {
2814 .owner
= THIS_MODULE
,
2815 .open
= ptype_seq_open
,
2817 .llseek
= seq_lseek
,
2818 .release
= seq_release_net
,
2822 static int __net_init
dev_proc_net_init(struct net
*net
)
2826 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
2828 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2830 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
2833 if (wext_proc_init(net
))
2839 proc_net_remove(net
, "ptype");
2841 proc_net_remove(net
, "softnet_stat");
2843 proc_net_remove(net
, "dev");
2847 static void __net_exit
dev_proc_net_exit(struct net
*net
)
2849 wext_proc_exit(net
);
2851 proc_net_remove(net
, "ptype");
2852 proc_net_remove(net
, "softnet_stat");
2853 proc_net_remove(net
, "dev");
2856 static struct pernet_operations __net_initdata dev_proc_ops
= {
2857 .init
= dev_proc_net_init
,
2858 .exit
= dev_proc_net_exit
,
2861 static int __init
dev_proc_init(void)
2863 return register_pernet_subsys(&dev_proc_ops
);
2866 #define dev_proc_init() 0
2867 #endif /* CONFIG_PROC_FS */
2871 * netdev_set_master - set up master/slave pair
2872 * @slave: slave device
2873 * @master: new master device
2875 * Changes the master device of the slave. Pass %NULL to break the
2876 * bonding. The caller must hold the RTNL semaphore. On a failure
2877 * a negative errno code is returned. On success the reference counts
2878 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2879 * function returns zero.
2881 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2883 struct net_device
*old
= slave
->master
;
2893 slave
->master
= master
;
2901 slave
->flags
|= IFF_SLAVE
;
2903 slave
->flags
&= ~IFF_SLAVE
;
2905 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2909 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
2911 unsigned short old_flags
= dev
->flags
;
2915 dev
->flags
|= IFF_PROMISC
;
2916 dev
->promiscuity
+= inc
;
2917 if (dev
->promiscuity
== 0) {
2920 * If inc causes overflow, untouch promisc and return error.
2923 dev
->flags
&= ~IFF_PROMISC
;
2925 dev
->promiscuity
-= inc
;
2926 printk(KERN_WARNING
"%s: promiscuity touches roof, "
2927 "set promiscuity failed, promiscuity feature "
2928 "of device might be broken.\n", dev
->name
);
2932 if (dev
->flags
!= old_flags
) {
2933 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2934 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2937 audit_log(current
->audit_context
, GFP_ATOMIC
,
2938 AUDIT_ANOM_PROMISCUOUS
,
2939 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2940 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2941 (old_flags
& IFF_PROMISC
),
2942 audit_get_loginuid(current
),
2943 current
->uid
, current
->gid
,
2944 audit_get_sessionid(current
));
2946 if (dev
->change_rx_flags
)
2947 dev
->change_rx_flags(dev
, IFF_PROMISC
);
2953 * dev_set_promiscuity - update promiscuity count on a device
2957 * Add or remove promiscuity from a device. While the count in the device
2958 * remains above zero the interface remains promiscuous. Once it hits zero
2959 * the device reverts back to normal filtering operation. A negative inc
2960 * value is used to drop promiscuity on the device.
2961 * Return 0 if successful or a negative errno code on error.
2963 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
2965 unsigned short old_flags
= dev
->flags
;
2968 err
= __dev_set_promiscuity(dev
, inc
);
2971 if (dev
->flags
!= old_flags
)
2972 dev_set_rx_mode(dev
);
2977 * dev_set_allmulti - update allmulti count on a device
2981 * Add or remove reception of all multicast frames to a device. While the
2982 * count in the device remains above zero the interface remains listening
2983 * to all interfaces. Once it hits zero the device reverts back to normal
2984 * filtering operation. A negative @inc value is used to drop the counter
2985 * when releasing a resource needing all multicasts.
2986 * Return 0 if successful or a negative errno code on error.
2989 int dev_set_allmulti(struct net_device
*dev
, int inc
)
2991 unsigned short old_flags
= dev
->flags
;
2995 dev
->flags
|= IFF_ALLMULTI
;
2996 dev
->allmulti
+= inc
;
2997 if (dev
->allmulti
== 0) {
3000 * If inc causes overflow, untouch allmulti and return error.
3003 dev
->flags
&= ~IFF_ALLMULTI
;
3005 dev
->allmulti
-= inc
;
3006 printk(KERN_WARNING
"%s: allmulti touches roof, "
3007 "set allmulti failed, allmulti feature of "
3008 "device might be broken.\n", dev
->name
);
3012 if (dev
->flags
^ old_flags
) {
3013 if (dev
->change_rx_flags
)
3014 dev
->change_rx_flags(dev
, IFF_ALLMULTI
);
3015 dev_set_rx_mode(dev
);
3021 * Upload unicast and multicast address lists to device and
3022 * configure RX filtering. When the device doesn't support unicast
3023 * filtering it is put in promiscuous mode while unicast addresses
3026 void __dev_set_rx_mode(struct net_device
*dev
)
3028 /* dev_open will call this function so the list will stay sane. */
3029 if (!(dev
->flags
&IFF_UP
))
3032 if (!netif_device_present(dev
))
3035 if (dev
->set_rx_mode
)
3036 dev
->set_rx_mode(dev
);
3038 /* Unicast addresses changes may only happen under the rtnl,
3039 * therefore calling __dev_set_promiscuity here is safe.
3041 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
3042 __dev_set_promiscuity(dev
, 1);
3043 dev
->uc_promisc
= 1;
3044 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
3045 __dev_set_promiscuity(dev
, -1);
3046 dev
->uc_promisc
= 0;
3049 if (dev
->set_multicast_list
)
3050 dev
->set_multicast_list(dev
);
3054 void dev_set_rx_mode(struct net_device
*dev
)
3056 netif_addr_lock_bh(dev
);
3057 __dev_set_rx_mode(dev
);
3058 netif_addr_unlock_bh(dev
);
3061 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
3062 void *addr
, int alen
, int glbl
)
3064 struct dev_addr_list
*da
;
3066 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
3067 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
3068 alen
== da
->da_addrlen
) {
3070 int old_glbl
= da
->da_gusers
;
3087 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
3088 void *addr
, int alen
, int glbl
)
3090 struct dev_addr_list
*da
;
3092 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
3093 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
3094 da
->da_addrlen
== alen
) {
3096 int old_glbl
= da
->da_gusers
;
3106 da
= kzalloc(sizeof(*da
), GFP_ATOMIC
);
3109 memcpy(da
->da_addr
, addr
, alen
);
3110 da
->da_addrlen
= alen
;
3112 da
->da_gusers
= glbl
? 1 : 0;
3120 * dev_unicast_delete - Release secondary unicast address.
3122 * @addr: address to delete
3123 * @alen: length of @addr
3125 * Release reference to a secondary unicast address and remove it
3126 * from the device if the reference count drops to zero.
3128 * The caller must hold the rtnl_mutex.
3130 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
3136 netif_addr_lock_bh(dev
);
3137 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3139 __dev_set_rx_mode(dev
);
3140 netif_addr_unlock_bh(dev
);
3143 EXPORT_SYMBOL(dev_unicast_delete
);
3146 * dev_unicast_add - add a secondary unicast address
3148 * @addr: address to add
3149 * @alen: length of @addr
3151 * Add a secondary unicast address to the device or increase
3152 * the reference count if it already exists.
3154 * The caller must hold the rtnl_mutex.
3156 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
3162 netif_addr_lock_bh(dev
);
3163 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3165 __dev_set_rx_mode(dev
);
3166 netif_addr_unlock_bh(dev
);
3169 EXPORT_SYMBOL(dev_unicast_add
);
3171 int __dev_addr_sync(struct dev_addr_list
**to
, int *to_count
,
3172 struct dev_addr_list
**from
, int *from_count
)
3174 struct dev_addr_list
*da
, *next
;
3178 while (da
!= NULL
) {
3180 if (!da
->da_synced
) {
3181 err
= __dev_addr_add(to
, to_count
,
3182 da
->da_addr
, da
->da_addrlen
, 0);
3187 } else if (da
->da_users
== 1) {
3188 __dev_addr_delete(to
, to_count
,
3189 da
->da_addr
, da
->da_addrlen
, 0);
3190 __dev_addr_delete(from
, from_count
,
3191 da
->da_addr
, da
->da_addrlen
, 0);
3198 void __dev_addr_unsync(struct dev_addr_list
**to
, int *to_count
,
3199 struct dev_addr_list
**from
, int *from_count
)
3201 struct dev_addr_list
*da
, *next
;
3204 while (da
!= NULL
) {
3206 if (da
->da_synced
) {
3207 __dev_addr_delete(to
, to_count
,
3208 da
->da_addr
, da
->da_addrlen
, 0);
3210 __dev_addr_delete(from
, from_count
,
3211 da
->da_addr
, da
->da_addrlen
, 0);
3218 * dev_unicast_sync - Synchronize device's unicast list to another device
3219 * @to: destination device
3220 * @from: source device
3222 * Add newly added addresses to the destination device and release
3223 * addresses that have no users left. The source device must be
3224 * locked by netif_tx_lock_bh.
3226 * This function is intended to be called from the dev->set_rx_mode
3227 * function of layered software devices.
3229 int dev_unicast_sync(struct net_device
*to
, struct net_device
*from
)
3233 netif_addr_lock_bh(to
);
3234 err
= __dev_addr_sync(&to
->uc_list
, &to
->uc_count
,
3235 &from
->uc_list
, &from
->uc_count
);
3237 __dev_set_rx_mode(to
);
3238 netif_addr_unlock_bh(to
);
3241 EXPORT_SYMBOL(dev_unicast_sync
);
3244 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3245 * @to: destination device
3246 * @from: source device
3248 * Remove all addresses that were added to the destination device by
3249 * dev_unicast_sync(). This function is intended to be called from the
3250 * dev->stop function of layered software devices.
3252 void dev_unicast_unsync(struct net_device
*to
, struct net_device
*from
)
3254 netif_addr_lock_bh(from
);
3255 netif_addr_lock(to
);
3257 __dev_addr_unsync(&to
->uc_list
, &to
->uc_count
,
3258 &from
->uc_list
, &from
->uc_count
);
3259 __dev_set_rx_mode(to
);
3261 netif_addr_unlock(to
);
3262 netif_addr_unlock_bh(from
);
3264 EXPORT_SYMBOL(dev_unicast_unsync
);
3266 static void __dev_addr_discard(struct dev_addr_list
**list
)
3268 struct dev_addr_list
*tmp
;
3270 while (*list
!= NULL
) {
3273 if (tmp
->da_users
> tmp
->da_gusers
)
3274 printk("__dev_addr_discard: address leakage! "
3275 "da_users=%d\n", tmp
->da_users
);
3280 static void dev_addr_discard(struct net_device
*dev
)
3282 netif_addr_lock_bh(dev
);
3284 __dev_addr_discard(&dev
->uc_list
);
3287 __dev_addr_discard(&dev
->mc_list
);
3290 netif_addr_unlock_bh(dev
);
3293 unsigned dev_get_flags(const struct net_device
*dev
)
3297 flags
= (dev
->flags
& ~(IFF_PROMISC
|
3302 (dev
->gflags
& (IFF_PROMISC
|
3305 if (netif_running(dev
)) {
3306 if (netif_oper_up(dev
))
3307 flags
|= IFF_RUNNING
;
3308 if (netif_carrier_ok(dev
))
3309 flags
|= IFF_LOWER_UP
;
3310 if (netif_dormant(dev
))
3311 flags
|= IFF_DORMANT
;
3317 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
3320 int old_flags
= dev
->flags
;
3325 * Set the flags on our device.
3328 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
3329 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
3331 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
3335 * Load in the correct multicast list now the flags have changed.
3338 if (dev
->change_rx_flags
&& (old_flags
^ flags
) & IFF_MULTICAST
)
3339 dev
->change_rx_flags(dev
, IFF_MULTICAST
);
3341 dev_set_rx_mode(dev
);
3344 * Have we downed the interface. We handle IFF_UP ourselves
3345 * according to user attempts to set it, rather than blindly
3350 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
3351 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
3354 dev_set_rx_mode(dev
);
3357 if (dev
->flags
& IFF_UP
&&
3358 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
3360 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
3362 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
3363 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
3364 dev
->gflags
^= IFF_PROMISC
;
3365 dev_set_promiscuity(dev
, inc
);
3368 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3369 is important. Some (broken) drivers set IFF_PROMISC, when
3370 IFF_ALLMULTI is requested not asking us and not reporting.
3372 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
3373 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
3374 dev
->gflags
^= IFF_ALLMULTI
;
3375 dev_set_allmulti(dev
, inc
);
3378 /* Exclude state transition flags, already notified */
3379 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
3381 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
3386 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
3390 if (new_mtu
== dev
->mtu
)
3393 /* MTU must be positive. */
3397 if (!netif_device_present(dev
))
3401 if (dev
->change_mtu
)
3402 err
= dev
->change_mtu(dev
, new_mtu
);
3405 if (!err
&& dev
->flags
& IFF_UP
)
3406 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
3410 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
3414 if (!dev
->set_mac_address
)
3416 if (sa
->sa_family
!= dev
->type
)
3418 if (!netif_device_present(dev
))
3420 err
= dev
->set_mac_address(dev
, sa
);
3422 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3427 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3429 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3432 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3438 case SIOCGIFFLAGS
: /* Get interface flags */
3439 ifr
->ifr_flags
= dev_get_flags(dev
);
3442 case SIOCGIFMETRIC
: /* Get the metric on the interface
3443 (currently unused) */
3444 ifr
->ifr_metric
= 0;
3447 case SIOCGIFMTU
: /* Get the MTU of a device */
3448 ifr
->ifr_mtu
= dev
->mtu
;
3453 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
3455 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
3456 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3457 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
3465 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
3466 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
3467 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
3468 ifr
->ifr_map
.irq
= dev
->irq
;
3469 ifr
->ifr_map
.dma
= dev
->dma
;
3470 ifr
->ifr_map
.port
= dev
->if_port
;
3474 ifr
->ifr_ifindex
= dev
->ifindex
;
3478 ifr
->ifr_qlen
= dev
->tx_queue_len
;
3482 /* dev_ioctl() should ensure this case
3494 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3496 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3499 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3505 case SIOCSIFFLAGS
: /* Set interface flags */
3506 return dev_change_flags(dev
, ifr
->ifr_flags
);
3508 case SIOCSIFMETRIC
: /* Set the metric on the interface
3509 (currently unused) */
3512 case SIOCSIFMTU
: /* Set the MTU of a device */
3513 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
3516 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
3518 case SIOCSIFHWBROADCAST
:
3519 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
3521 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
3522 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3523 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3527 if (dev
->set_config
) {
3528 if (!netif_device_present(dev
))
3530 return dev
->set_config(dev
, &ifr
->ifr_map
);
3535 if ((!dev
->set_multicast_list
&& !dev
->set_rx_mode
) ||
3536 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3538 if (!netif_device_present(dev
))
3540 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
3544 if ((!dev
->set_multicast_list
&& !dev
->set_rx_mode
) ||
3545 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3547 if (!netif_device_present(dev
))
3549 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
3553 if (ifr
->ifr_qlen
< 0)
3555 dev
->tx_queue_len
= ifr
->ifr_qlen
;
3559 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
3560 return dev_change_name(dev
, ifr
->ifr_newname
);
3563 * Unknown or private ioctl
3567 if ((cmd
>= SIOCDEVPRIVATE
&&
3568 cmd
<= SIOCDEVPRIVATE
+ 15) ||
3569 cmd
== SIOCBONDENSLAVE
||
3570 cmd
== SIOCBONDRELEASE
||
3571 cmd
== SIOCBONDSETHWADDR
||
3572 cmd
== SIOCBONDSLAVEINFOQUERY
||
3573 cmd
== SIOCBONDINFOQUERY
||
3574 cmd
== SIOCBONDCHANGEACTIVE
||
3575 cmd
== SIOCGMIIPHY
||
3576 cmd
== SIOCGMIIREG
||
3577 cmd
== SIOCSMIIREG
||
3578 cmd
== SIOCBRADDIF
||
3579 cmd
== SIOCBRDELIF
||
3580 cmd
== SIOCWANDEV
) {
3582 if (dev
->do_ioctl
) {
3583 if (netif_device_present(dev
))
3584 err
= dev
->do_ioctl(dev
, ifr
,
3597 * This function handles all "interface"-type I/O control requests. The actual
3598 * 'doing' part of this is dev_ifsioc above.
3602 * dev_ioctl - network device ioctl
3603 * @net: the applicable net namespace
3604 * @cmd: command to issue
3605 * @arg: pointer to a struct ifreq in user space
3607 * Issue ioctl functions to devices. This is normally called by the
3608 * user space syscall interfaces but can sometimes be useful for
3609 * other purposes. The return value is the return from the syscall if
3610 * positive or a negative errno code on error.
3613 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
3619 /* One special case: SIOCGIFCONF takes ifconf argument
3620 and requires shared lock, because it sleeps writing
3624 if (cmd
== SIOCGIFCONF
) {
3626 ret
= dev_ifconf(net
, (char __user
*) arg
);
3630 if (cmd
== SIOCGIFNAME
)
3631 return dev_ifname(net
, (struct ifreq __user
*)arg
);
3633 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3636 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
3638 colon
= strchr(ifr
.ifr_name
, ':');
3643 * See which interface the caller is talking about.
3648 * These ioctl calls:
3649 * - can be done by all.
3650 * - atomic and do not require locking.
3661 dev_load(net
, ifr
.ifr_name
);
3662 read_lock(&dev_base_lock
);
3663 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
3664 read_unlock(&dev_base_lock
);
3668 if (copy_to_user(arg
, &ifr
,
3669 sizeof(struct ifreq
)))
3675 dev_load(net
, ifr
.ifr_name
);
3677 ret
= dev_ethtool(net
, &ifr
);
3682 if (copy_to_user(arg
, &ifr
,
3683 sizeof(struct ifreq
)))
3689 * These ioctl calls:
3690 * - require superuser power.
3691 * - require strict serialization.
3697 if (!capable(CAP_NET_ADMIN
))
3699 dev_load(net
, ifr
.ifr_name
);
3701 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3706 if (copy_to_user(arg
, &ifr
,
3707 sizeof(struct ifreq
)))
3713 * These ioctl calls:
3714 * - require superuser power.
3715 * - require strict serialization.
3716 * - do not return a value
3726 case SIOCSIFHWBROADCAST
:
3729 case SIOCBONDENSLAVE
:
3730 case SIOCBONDRELEASE
:
3731 case SIOCBONDSETHWADDR
:
3732 case SIOCBONDCHANGEACTIVE
:
3735 if (!capable(CAP_NET_ADMIN
))
3738 case SIOCBONDSLAVEINFOQUERY
:
3739 case SIOCBONDINFOQUERY
:
3740 dev_load(net
, ifr
.ifr_name
);
3742 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3747 /* Get the per device memory space. We can add this but
3748 * currently do not support it */
3750 /* Set the per device memory buffer space.
3751 * Not applicable in our case */
3756 * Unknown or private ioctl.
3759 if (cmd
== SIOCWANDEV
||
3760 (cmd
>= SIOCDEVPRIVATE
&&
3761 cmd
<= SIOCDEVPRIVATE
+ 15)) {
3762 dev_load(net
, ifr
.ifr_name
);
3764 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3766 if (!ret
&& copy_to_user(arg
, &ifr
,
3767 sizeof(struct ifreq
)))
3771 /* Take care of Wireless Extensions */
3772 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
3773 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
3780 * dev_new_index - allocate an ifindex
3781 * @net: the applicable net namespace
3783 * Returns a suitable unique value for a new device interface
3784 * number. The caller must hold the rtnl semaphore or the
3785 * dev_base_lock to be sure it remains unique.
3787 static int dev_new_index(struct net
*net
)
3793 if (!__dev_get_by_index(net
, ifindex
))
3798 /* Delayed registration/unregisteration */
3799 static DEFINE_SPINLOCK(net_todo_list_lock
);
3800 static LIST_HEAD(net_todo_list
);
3802 static void net_set_todo(struct net_device
*dev
)
3804 spin_lock(&net_todo_list_lock
);
3805 list_add_tail(&dev
->todo_list
, &net_todo_list
);
3806 spin_unlock(&net_todo_list_lock
);
3809 static void rollback_registered(struct net_device
*dev
)
3811 BUG_ON(dev_boot_phase
);
3814 /* Some devices call without registering for initialization unwind. */
3815 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3816 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3817 "was registered\n", dev
->name
, dev
);
3823 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3825 /* If device is running, close it first. */
3828 /* And unlink it from device chain. */
3829 unlist_netdevice(dev
);
3831 dev
->reg_state
= NETREG_UNREGISTERING
;
3835 /* Shutdown queueing discipline. */
3839 /* Notify protocols, that we are about to destroy
3840 this device. They should clean all the things.
3842 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
3845 * Flush the unicast and multicast chains
3847 dev_addr_discard(dev
);
3852 /* Notifier chain MUST detach us from master device. */
3853 BUG_TRAP(!dev
->master
);
3855 /* Remove entries from kobject tree */
3856 netdev_unregister_kobject(dev
);
3863 static void __netdev_init_queue_locks_one(struct net_device
*dev
,
3864 struct netdev_queue
*dev_queue
,
3867 spin_lock_init(&dev_queue
->_xmit_lock
);
3868 netdev_set_xmit_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
3869 dev_queue
->xmit_lock_owner
= -1;
3872 static void netdev_init_queue_locks(struct net_device
*dev
)
3874 netdev_for_each_tx_queue(dev
, __netdev_init_queue_locks_one
, NULL
);
3875 __netdev_init_queue_locks_one(dev
, &dev
->rx_queue
, NULL
);
3879 * register_netdevice - register a network device
3880 * @dev: device to register
3882 * Take a completed network device structure and add it to the kernel
3883 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3884 * chain. 0 is returned on success. A negative errno code is returned
3885 * on a failure to set up the device, or if the name is a duplicate.
3887 * Callers must hold the rtnl semaphore. You may want
3888 * register_netdev() instead of this.
3891 * The locking appears insufficient to guarantee two parallel registers
3892 * will not get the same name.
3895 int register_netdevice(struct net_device
*dev
)
3897 struct hlist_head
*head
;
3898 struct hlist_node
*p
;
3902 BUG_ON(dev_boot_phase
);
3907 /* When net_device's are persistent, this will be fatal. */
3908 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
3909 BUG_ON(!dev_net(dev
));
3912 spin_lock_init(&dev
->addr_list_lock
);
3913 netdev_set_addr_lockdep_class(dev
);
3914 netdev_init_queue_locks(dev
);
3918 /* Init, if this function is available */
3920 ret
= dev
->init(dev
);
3928 if (!dev_valid_name(dev
->name
)) {
3933 dev
->ifindex
= dev_new_index(net
);
3934 if (dev
->iflink
== -1)
3935 dev
->iflink
= dev
->ifindex
;
3937 /* Check for existence of name */
3938 head
= dev_name_hash(net
, dev
->name
);
3939 hlist_for_each(p
, head
) {
3940 struct net_device
*d
3941 = hlist_entry(p
, struct net_device
, name_hlist
);
3942 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
3948 /* Fix illegal checksum combinations */
3949 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
3950 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3951 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
3953 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
3956 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
3957 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3958 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
3960 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
3964 /* Fix illegal SG+CSUM combinations. */
3965 if ((dev
->features
& NETIF_F_SG
) &&
3966 !(dev
->features
& NETIF_F_ALL_CSUM
)) {
3967 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no checksum feature.\n",
3969 dev
->features
&= ~NETIF_F_SG
;
3972 /* TSO requires that SG is present as well. */
3973 if ((dev
->features
& NETIF_F_TSO
) &&
3974 !(dev
->features
& NETIF_F_SG
)) {
3975 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no SG feature.\n",
3977 dev
->features
&= ~NETIF_F_TSO
;
3979 if (dev
->features
& NETIF_F_UFO
) {
3980 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
3981 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3982 "NETIF_F_HW_CSUM feature.\n",
3984 dev
->features
&= ~NETIF_F_UFO
;
3986 if (!(dev
->features
& NETIF_F_SG
)) {
3987 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3988 "NETIF_F_SG feature.\n",
3990 dev
->features
&= ~NETIF_F_UFO
;
3994 netdev_initialize_kobject(dev
);
3995 ret
= netdev_register_kobject(dev
);
3998 dev
->reg_state
= NETREG_REGISTERED
;
4001 * Default initial state at registry is that the
4002 * device is present.
4005 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
4007 dev_init_scheduler(dev
);
4009 list_netdevice(dev
);
4011 /* Notify protocols, that a new device appeared. */
4012 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
4013 ret
= notifier_to_errno(ret
);
4015 rollback_registered(dev
);
4016 dev
->reg_state
= NETREG_UNREGISTERED
;
4029 * register_netdev - register a network device
4030 * @dev: device to register
4032 * Take a completed network device structure and add it to the kernel
4033 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4034 * chain. 0 is returned on success. A negative errno code is returned
4035 * on a failure to set up the device, or if the name is a duplicate.
4037 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4038 * and expands the device name if you passed a format string to
4041 int register_netdev(struct net_device
*dev
)
4048 * If the name is a format string the caller wants us to do a
4051 if (strchr(dev
->name
, '%')) {
4052 err
= dev_alloc_name(dev
, dev
->name
);
4057 err
= register_netdevice(dev
);
4062 EXPORT_SYMBOL(register_netdev
);
4065 * netdev_wait_allrefs - wait until all references are gone.
4067 * This is called when unregistering network devices.
4069 * Any protocol or device that holds a reference should register
4070 * for netdevice notification, and cleanup and put back the
4071 * reference if they receive an UNREGISTER event.
4072 * We can get stuck here if buggy protocols don't correctly
4075 static void netdev_wait_allrefs(struct net_device
*dev
)
4077 unsigned long rebroadcast_time
, warning_time
;
4079 rebroadcast_time
= warning_time
= jiffies
;
4080 while (atomic_read(&dev
->refcnt
) != 0) {
4081 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
4084 /* Rebroadcast unregister notification */
4085 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4087 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
4089 /* We must not have linkwatch events
4090 * pending on unregister. If this
4091 * happens, we simply run the queue
4092 * unscheduled, resulting in a noop
4095 linkwatch_run_queue();
4100 rebroadcast_time
= jiffies
;
4105 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
4106 printk(KERN_EMERG
"unregister_netdevice: "
4107 "waiting for %s to become free. Usage "
4109 dev
->name
, atomic_read(&dev
->refcnt
));
4110 warning_time
= jiffies
;
4119 * register_netdevice(x1);
4120 * register_netdevice(x2);
4122 * unregister_netdevice(y1);
4123 * unregister_netdevice(y2);
4129 * We are invoked by rtnl_unlock() after it drops the semaphore.
4130 * This allows us to deal with problems:
4131 * 1) We can delete sysfs objects which invoke hotplug
4132 * without deadlocking with linkwatch via keventd.
4133 * 2) Since we run with the RTNL semaphore not held, we can sleep
4134 * safely in order to wait for the netdev refcnt to drop to zero.
4136 static DEFINE_MUTEX(net_todo_run_mutex
);
4137 void netdev_run_todo(void)
4139 struct list_head list
;
4141 /* Need to guard against multiple cpu's getting out of order. */
4142 mutex_lock(&net_todo_run_mutex
);
4144 /* Not safe to do outside the semaphore. We must not return
4145 * until all unregister events invoked by the local processor
4146 * have been completed (either by this todo run, or one on
4149 if (list_empty(&net_todo_list
))
4152 /* Snapshot list, allow later requests */
4153 spin_lock(&net_todo_list_lock
);
4154 list_replace_init(&net_todo_list
, &list
);
4155 spin_unlock(&net_todo_list_lock
);
4157 while (!list_empty(&list
)) {
4158 struct net_device
*dev
4159 = list_entry(list
.next
, struct net_device
, todo_list
);
4160 list_del(&dev
->todo_list
);
4162 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
4163 printk(KERN_ERR
"network todo '%s' but state %d\n",
4164 dev
->name
, dev
->reg_state
);
4169 dev
->reg_state
= NETREG_UNREGISTERED
;
4171 netdev_wait_allrefs(dev
);
4174 BUG_ON(atomic_read(&dev
->refcnt
));
4175 BUG_TRAP(!dev
->ip_ptr
);
4176 BUG_TRAP(!dev
->ip6_ptr
);
4177 BUG_TRAP(!dev
->dn_ptr
);
4179 if (dev
->destructor
)
4180 dev
->destructor(dev
);
4182 /* Free network device */
4183 kobject_put(&dev
->dev
.kobj
);
4187 mutex_unlock(&net_todo_run_mutex
);
4190 static struct net_device_stats
*internal_stats(struct net_device
*dev
)
4195 static void netdev_init_one_queue(struct net_device
*dev
,
4196 struct netdev_queue
*queue
,
4202 static void netdev_init_queues(struct net_device
*dev
)
4204 netdev_init_one_queue(dev
, &dev
->rx_queue
, NULL
);
4205 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
4209 * alloc_netdev_mq - allocate network device
4210 * @sizeof_priv: size of private data to allocate space for
4211 * @name: device name format string
4212 * @setup: callback to initialize device
4213 * @queue_count: the number of subqueues to allocate
4215 * Allocates a struct net_device with private data area for driver use
4216 * and performs basic initialization. Also allocates subquue structs
4217 * for each queue on the device at the end of the netdevice.
4219 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
4220 void (*setup
)(struct net_device
*), unsigned int queue_count
)
4222 struct netdev_queue
*tx
;
4223 struct net_device
*dev
;
4227 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
4229 alloc_size
= sizeof(struct net_device
);
4231 /* ensure 32-byte alignment of private area */
4232 alloc_size
= (alloc_size
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
4233 alloc_size
+= sizeof_priv
;
4235 /* ensure 32-byte alignment of whole construct */
4236 alloc_size
+= NETDEV_ALIGN_CONST
;
4238 p
= kzalloc(alloc_size
, GFP_KERNEL
);
4240 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
4244 tx
= kcalloc(queue_count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
4246 printk(KERN_ERR
"alloc_netdev: Unable to allocate "
4252 dev
= (struct net_device
*)
4253 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
4254 dev
->padded
= (char *)dev
- (char *)p
;
4255 dev_net_set(dev
, &init_net
);
4258 dev
->num_tx_queues
= queue_count
;
4259 dev
->real_num_tx_queues
= queue_count
;
4262 dev
->priv
= ((char *)dev
+
4263 ((sizeof(struct net_device
) + NETDEV_ALIGN_CONST
)
4264 & ~NETDEV_ALIGN_CONST
));
4267 dev
->gso_max_size
= GSO_MAX_SIZE
;
4269 netdev_init_queues(dev
);
4271 dev
->get_stats
= internal_stats
;
4272 netpoll_netdev_init(dev
);
4274 strcpy(dev
->name
, name
);
4277 EXPORT_SYMBOL(alloc_netdev_mq
);
4280 * free_netdev - free network device
4283 * This function does the last stage of destroying an allocated device
4284 * interface. The reference to the device object is released.
4285 * If this is the last reference then it will be freed.
4287 void free_netdev(struct net_device
*dev
)
4289 release_net(dev_net(dev
));
4293 /* Compatibility with error handling in drivers */
4294 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4295 kfree((char *)dev
- dev
->padded
);
4299 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
4300 dev
->reg_state
= NETREG_RELEASED
;
4302 /* will free via device release */
4303 put_device(&dev
->dev
);
4306 /* Synchronize with packet receive processing. */
4307 void synchronize_net(void)
4314 * unregister_netdevice - remove device from the kernel
4317 * This function shuts down a device interface and removes it
4318 * from the kernel tables.
4320 * Callers must hold the rtnl semaphore. You may want
4321 * unregister_netdev() instead of this.
4324 void unregister_netdevice(struct net_device
*dev
)
4328 rollback_registered(dev
);
4329 /* Finish processing unregister after unlock */
4334 * unregister_netdev - remove device from the kernel
4337 * This function shuts down a device interface and removes it
4338 * from the kernel tables.
4340 * This is just a wrapper for unregister_netdevice that takes
4341 * the rtnl semaphore. In general you want to use this and not
4342 * unregister_netdevice.
4344 void unregister_netdev(struct net_device
*dev
)
4347 unregister_netdevice(dev
);
4351 EXPORT_SYMBOL(unregister_netdev
);
4354 * dev_change_net_namespace - move device to different nethost namespace
4356 * @net: network namespace
4357 * @pat: If not NULL name pattern to try if the current device name
4358 * is already taken in the destination network namespace.
4360 * This function shuts down a device interface and moves it
4361 * to a new network namespace. On success 0 is returned, on
4362 * a failure a netagive errno code is returned.
4364 * Callers must hold the rtnl semaphore.
4367 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
4370 const char *destname
;
4375 /* Don't allow namespace local devices to be moved. */
4377 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
4380 /* Ensure the device has been registrered */
4382 if (dev
->reg_state
!= NETREG_REGISTERED
)
4385 /* Get out if there is nothing todo */
4387 if (net_eq(dev_net(dev
), net
))
4390 /* Pick the destination device name, and ensure
4391 * we can use it in the destination network namespace.
4394 destname
= dev
->name
;
4395 if (__dev_get_by_name(net
, destname
)) {
4396 /* We get here if we can't use the current device name */
4399 if (!dev_valid_name(pat
))
4401 if (strchr(pat
, '%')) {
4402 if (__dev_alloc_name(net
, pat
, buf
) < 0)
4407 if (__dev_get_by_name(net
, destname
))
4412 * And now a mini version of register_netdevice unregister_netdevice.
4415 /* If device is running close it first. */
4418 /* And unlink it from device chain */
4420 unlist_netdevice(dev
);
4424 /* Shutdown queueing discipline. */
4427 /* Notify protocols, that we are about to destroy
4428 this device. They should clean all the things.
4430 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4433 * Flush the unicast and multicast chains
4435 dev_addr_discard(dev
);
4437 /* Actually switch the network namespace */
4438 dev_net_set(dev
, net
);
4440 /* Assign the new device name */
4441 if (destname
!= dev
->name
)
4442 strcpy(dev
->name
, destname
);
4444 /* If there is an ifindex conflict assign a new one */
4445 if (__dev_get_by_index(net
, dev
->ifindex
)) {
4446 int iflink
= (dev
->iflink
== dev
->ifindex
);
4447 dev
->ifindex
= dev_new_index(net
);
4449 dev
->iflink
= dev
->ifindex
;
4452 /* Fixup kobjects */
4453 netdev_unregister_kobject(dev
);
4454 err
= netdev_register_kobject(dev
);
4457 /* Add the device back in the hashes */
4458 list_netdevice(dev
);
4460 /* Notify protocols, that a new device appeared. */
4461 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
4469 static int dev_cpu_callback(struct notifier_block
*nfb
,
4470 unsigned long action
,
4473 struct sk_buff
**list_skb
;
4474 struct Qdisc
**list_net
;
4475 struct sk_buff
*skb
;
4476 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
4477 struct softnet_data
*sd
, *oldsd
;
4479 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
4482 local_irq_disable();
4483 cpu
= smp_processor_id();
4484 sd
= &per_cpu(softnet_data
, cpu
);
4485 oldsd
= &per_cpu(softnet_data
, oldcpu
);
4487 /* Find end of our completion_queue. */
4488 list_skb
= &sd
->completion_queue
;
4490 list_skb
= &(*list_skb
)->next
;
4491 /* Append completion queue from offline CPU. */
4492 *list_skb
= oldsd
->completion_queue
;
4493 oldsd
->completion_queue
= NULL
;
4495 /* Find end of our output_queue. */
4496 list_net
= &sd
->output_queue
;
4498 list_net
= &(*list_net
)->next_sched
;
4499 /* Append output queue from offline CPU. */
4500 *list_net
= oldsd
->output_queue
;
4501 oldsd
->output_queue
= NULL
;
4503 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
4506 /* Process offline CPU's input_pkt_queue */
4507 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
4513 #ifdef CONFIG_NET_DMA
4515 * net_dma_rebalance - try to maintain one DMA channel per CPU
4516 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4518 * This is called when the number of channels allocated to the net_dma client
4519 * changes. The net_dma client tries to have one DMA channel per CPU.
4522 static void net_dma_rebalance(struct net_dma
*net_dma
)
4524 unsigned int cpu
, i
, n
, chan_idx
;
4525 struct dma_chan
*chan
;
4527 if (cpus_empty(net_dma
->channel_mask
)) {
4528 for_each_online_cpu(cpu
)
4529 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
4534 cpu
= first_cpu(cpu_online_map
);
4536 for_each_cpu_mask_nr(chan_idx
, net_dma
->channel_mask
) {
4537 chan
= net_dma
->channels
[chan_idx
];
4539 n
= ((num_online_cpus() / cpus_weight(net_dma
->channel_mask
))
4540 + (i
< (num_online_cpus() %
4541 cpus_weight(net_dma
->channel_mask
)) ? 1 : 0));
4544 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
4545 cpu
= next_cpu(cpu
, cpu_online_map
);
4553 * netdev_dma_event - event callback for the net_dma_client
4554 * @client: should always be net_dma_client
4555 * @chan: DMA channel for the event
4556 * @state: DMA state to be handled
4558 static enum dma_state_client
4559 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
4560 enum dma_state state
)
4562 int i
, found
= 0, pos
= -1;
4563 struct net_dma
*net_dma
=
4564 container_of(client
, struct net_dma
, client
);
4565 enum dma_state_client ack
= DMA_DUP
; /* default: take no action */
4567 spin_lock(&net_dma
->lock
);
4569 case DMA_RESOURCE_AVAILABLE
:
4570 for (i
= 0; i
< nr_cpu_ids
; i
++)
4571 if (net_dma
->channels
[i
] == chan
) {
4574 } else if (net_dma
->channels
[i
] == NULL
&& pos
< 0)
4577 if (!found
&& pos
>= 0) {
4579 net_dma
->channels
[pos
] = chan
;
4580 cpu_set(pos
, net_dma
->channel_mask
);
4581 net_dma_rebalance(net_dma
);
4584 case DMA_RESOURCE_REMOVED
:
4585 for (i
= 0; i
< nr_cpu_ids
; i
++)
4586 if (net_dma
->channels
[i
] == chan
) {
4594 cpu_clear(pos
, net_dma
->channel_mask
);
4595 net_dma
->channels
[i
] = NULL
;
4596 net_dma_rebalance(net_dma
);
4602 spin_unlock(&net_dma
->lock
);
4608 * netdev_dma_regiser - register the networking subsystem as a DMA client
4610 static int __init
netdev_dma_register(void)
4612 net_dma
.channels
= kzalloc(nr_cpu_ids
* sizeof(struct net_dma
),
4614 if (unlikely(!net_dma
.channels
)) {
4616 "netdev_dma: no memory for net_dma.channels\n");
4619 spin_lock_init(&net_dma
.lock
);
4620 dma_cap_set(DMA_MEMCPY
, net_dma
.client
.cap_mask
);
4621 dma_async_client_register(&net_dma
.client
);
4622 dma_async_client_chan_request(&net_dma
.client
);
4627 static int __init
netdev_dma_register(void) { return -ENODEV
; }
4628 #endif /* CONFIG_NET_DMA */
4631 * netdev_compute_feature - compute conjunction of two feature sets
4632 * @all: first feature set
4633 * @one: second feature set
4635 * Computes a new feature set after adding a device with feature set
4636 * @one to the master device with current feature set @all. Returns
4637 * the new feature set.
4639 int netdev_compute_features(unsigned long all
, unsigned long one
)
4641 /* if device needs checksumming, downgrade to hw checksumming */
4642 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
4643 all
^= NETIF_F_NO_CSUM
| NETIF_F_HW_CSUM
;
4645 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4646 if (all
& NETIF_F_HW_CSUM
&& !(one
& NETIF_F_HW_CSUM
))
4647 all
^= NETIF_F_HW_CSUM
4648 | NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
;
4650 if (one
& NETIF_F_GSO
)
4651 one
|= NETIF_F_GSO_SOFTWARE
;
4654 /* If even one device supports robust GSO, enable it for all. */
4655 if (one
& NETIF_F_GSO_ROBUST
)
4656 all
|= NETIF_F_GSO_ROBUST
;
4658 all
&= one
| NETIF_F_LLTX
;
4660 if (!(all
& NETIF_F_ALL_CSUM
))
4662 if (!(all
& NETIF_F_SG
))
4663 all
&= ~NETIF_F_GSO_MASK
;
4667 EXPORT_SYMBOL(netdev_compute_features
);
4669 static struct hlist_head
*netdev_create_hash(void)
4672 struct hlist_head
*hash
;
4674 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
4676 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
4677 INIT_HLIST_HEAD(&hash
[i
]);
4682 /* Initialize per network namespace state */
4683 static int __net_init
netdev_init(struct net
*net
)
4685 INIT_LIST_HEAD(&net
->dev_base_head
);
4687 net
->dev_name_head
= netdev_create_hash();
4688 if (net
->dev_name_head
== NULL
)
4691 net
->dev_index_head
= netdev_create_hash();
4692 if (net
->dev_index_head
== NULL
)
4698 kfree(net
->dev_name_head
);
4703 char *netdev_drivername(struct net_device
*dev
, char *buffer
, int len
)
4705 struct device_driver
*driver
;
4706 struct device
*parent
;
4708 if (len
<= 0 || !buffer
)
4712 parent
= dev
->dev
.parent
;
4717 driver
= parent
->driver
;
4718 if (driver
&& driver
->name
)
4719 strlcpy(buffer
, driver
->name
, len
);
4723 static void __net_exit
netdev_exit(struct net
*net
)
4725 kfree(net
->dev_name_head
);
4726 kfree(net
->dev_index_head
);
4729 static struct pernet_operations __net_initdata netdev_net_ops
= {
4730 .init
= netdev_init
,
4731 .exit
= netdev_exit
,
4734 static void __net_exit
default_device_exit(struct net
*net
)
4736 struct net_device
*dev
, *next
;
4738 * Push all migratable of the network devices back to the
4739 * initial network namespace
4742 for_each_netdev_safe(net
, dev
, next
) {
4744 char fb_name
[IFNAMSIZ
];
4746 /* Ignore unmoveable devices (i.e. loopback) */
4747 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
4750 /* Push remaing network devices to init_net */
4751 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
4752 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
4754 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
4755 __func__
, dev
->name
, err
);
4762 static struct pernet_operations __net_initdata default_device_ops
= {
4763 .exit
= default_device_exit
,
4767 * Initialize the DEV module. At boot time this walks the device list and
4768 * unhooks any devices that fail to initialise (normally hardware not
4769 * present) and leaves us with a valid list of present and active devices.
4774 * This is called single threaded during boot, so no need
4775 * to take the rtnl semaphore.
4777 static int __init
net_dev_init(void)
4779 int i
, rc
= -ENOMEM
;
4781 BUG_ON(!dev_boot_phase
);
4783 if (dev_proc_init())
4786 if (netdev_kobject_init())
4789 INIT_LIST_HEAD(&ptype_all
);
4790 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
4791 INIT_LIST_HEAD(&ptype_base
[i
]);
4793 if (register_pernet_subsys(&netdev_net_ops
))
4796 if (register_pernet_device(&default_device_ops
))
4800 * Initialise the packet receive queues.
4803 for_each_possible_cpu(i
) {
4804 struct softnet_data
*queue
;
4806 queue
= &per_cpu(softnet_data
, i
);
4807 skb_queue_head_init(&queue
->input_pkt_queue
);
4808 queue
->completion_queue
= NULL
;
4809 INIT_LIST_HEAD(&queue
->poll_list
);
4811 queue
->backlog
.poll
= process_backlog
;
4812 queue
->backlog
.weight
= weight_p
;
4815 netdev_dma_register();
4819 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
4820 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
4822 hotcpu_notifier(dev_cpu_callback
, 0);
4830 subsys_initcall(net_dev_init
);
4832 EXPORT_SYMBOL(__dev_get_by_index
);
4833 EXPORT_SYMBOL(__dev_get_by_name
);
4834 EXPORT_SYMBOL(__dev_remove_pack
);
4835 EXPORT_SYMBOL(dev_valid_name
);
4836 EXPORT_SYMBOL(dev_add_pack
);
4837 EXPORT_SYMBOL(dev_alloc_name
);
4838 EXPORT_SYMBOL(dev_close
);
4839 EXPORT_SYMBOL(dev_get_by_flags
);
4840 EXPORT_SYMBOL(dev_get_by_index
);
4841 EXPORT_SYMBOL(dev_get_by_name
);
4842 EXPORT_SYMBOL(dev_open
);
4843 EXPORT_SYMBOL(dev_queue_xmit
);
4844 EXPORT_SYMBOL(dev_remove_pack
);
4845 EXPORT_SYMBOL(dev_set_allmulti
);
4846 EXPORT_SYMBOL(dev_set_promiscuity
);
4847 EXPORT_SYMBOL(dev_change_flags
);
4848 EXPORT_SYMBOL(dev_set_mtu
);
4849 EXPORT_SYMBOL(dev_set_mac_address
);
4850 EXPORT_SYMBOL(free_netdev
);
4851 EXPORT_SYMBOL(netdev_boot_setup_check
);
4852 EXPORT_SYMBOL(netdev_set_master
);
4853 EXPORT_SYMBOL(netdev_state_change
);
4854 EXPORT_SYMBOL(netif_receive_skb
);
4855 EXPORT_SYMBOL(netif_rx
);
4856 EXPORT_SYMBOL(register_gifconf
);
4857 EXPORT_SYMBOL(register_netdevice
);
4858 EXPORT_SYMBOL(register_netdevice_notifier
);
4859 EXPORT_SYMBOL(skb_checksum_help
);
4860 EXPORT_SYMBOL(synchronize_net
);
4861 EXPORT_SYMBOL(unregister_netdevice
);
4862 EXPORT_SYMBOL(unregister_netdevice_notifier
);
4863 EXPORT_SYMBOL(net_enable_timestamp
);
4864 EXPORT_SYMBOL(net_disable_timestamp
);
4865 EXPORT_SYMBOL(dev_get_flags
);
4867 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4868 EXPORT_SYMBOL(br_handle_frame_hook
);
4869 EXPORT_SYMBOL(br_fdb_get_hook
);
4870 EXPORT_SYMBOL(br_fdb_put_hook
);
4874 EXPORT_SYMBOL(dev_load
);
4877 EXPORT_PER_CPU_SYMBOL(softnet_data
);