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>
125 #include "net-sysfs.h"
128 * The list of packet types we will receive (as opposed to discard)
129 * and the routines to invoke.
131 * Why 16. Because with 16 the only overlap we get on a hash of the
132 * low nibble of the protocol value is RARP/SNAP/X.25.
134 * NOTE: That is no longer true with the addition of VLAN tags. Not
135 * sure which should go first, but I bet it won't make much
136 * difference if we are running VLANs. The good news is that
137 * this protocol won't be in the list unless compiled in, so
138 * the average user (w/out VLANs) will not be adversely affected.
155 #define PTYPE_HASH_SIZE (16)
156 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
158 static DEFINE_SPINLOCK(ptype_lock
);
159 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
160 static struct list_head ptype_all __read_mostly
; /* Taps */
162 #ifdef CONFIG_NET_DMA
164 struct dma_client client
;
166 cpumask_t channel_mask
;
167 struct dma_chan
**channels
;
170 static enum dma_state_client
171 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
172 enum dma_state state
);
174 static struct net_dma net_dma
= {
176 .event_callback
= netdev_dma_event
,
182 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
185 * Pure readers hold dev_base_lock for reading.
187 * Writers must hold the rtnl semaphore while they loop through the
188 * dev_base_head list, and hold dev_base_lock for writing when they do the
189 * actual updates. This allows pure readers to access the list even
190 * while a writer is preparing to update it.
192 * To put it another way, dev_base_lock is held for writing only to
193 * protect against pure readers; the rtnl semaphore provides the
194 * protection against other writers.
196 * See, for example usages, register_netdevice() and
197 * unregister_netdevice(), which must be called with the rtnl
200 DEFINE_RWLOCK(dev_base_lock
);
202 EXPORT_SYMBOL(dev_base_lock
);
204 #define NETDEV_HASHBITS 8
205 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
207 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
209 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
210 return &net
->dev_name_head
[hash
& ((1 << NETDEV_HASHBITS
) - 1)];
213 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
215 return &net
->dev_index_head
[ifindex
& ((1 << NETDEV_HASHBITS
) - 1)];
218 /* Device list insertion */
219 static int list_netdevice(struct net_device
*dev
)
221 struct net
*net
= dev_net(dev
);
225 write_lock_bh(&dev_base_lock
);
226 list_add_tail(&dev
->dev_list
, &net
->dev_base_head
);
227 hlist_add_head(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
228 hlist_add_head(&dev
->index_hlist
, dev_index_hash(net
, dev
->ifindex
));
229 write_unlock_bh(&dev_base_lock
);
233 /* Device list removal */
234 static void unlist_netdevice(struct net_device
*dev
)
238 /* Unlink dev from the device chain */
239 write_lock_bh(&dev_base_lock
);
240 list_del(&dev
->dev_list
);
241 hlist_del(&dev
->name_hlist
);
242 hlist_del(&dev
->index_hlist
);
243 write_unlock_bh(&dev_base_lock
);
250 static RAW_NOTIFIER_HEAD(netdev_chain
);
253 * Device drivers call our routines to queue packets here. We empty the
254 * queue in the local softnet handler.
257 DEFINE_PER_CPU(struct softnet_data
, softnet_data
);
259 #ifdef CONFIG_DEBUG_LOCK_ALLOC
261 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
262 * according to dev->type
264 static const unsigned short netdev_lock_type
[] =
265 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
266 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
267 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
268 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
269 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
270 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
271 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
272 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
273 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
274 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
275 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
276 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
277 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
278 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_VOID
,
281 static const char *netdev_lock_name
[] =
282 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
283 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
284 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
285 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
286 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
287 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
288 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
289 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
290 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
291 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
292 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
293 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
294 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
295 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
298 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
300 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
304 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
305 if (netdev_lock_type
[i
] == dev_type
)
307 /* the last key is used by default */
308 return ARRAY_SIZE(netdev_lock_type
) - 1;
311 static inline void netdev_set_lockdep_class(spinlock_t
*lock
,
312 unsigned short dev_type
)
316 i
= netdev_lock_pos(dev_type
);
317 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
318 netdev_lock_name
[i
]);
321 static inline void netdev_set_lockdep_class(spinlock_t
*lock
,
322 unsigned short dev_type
)
327 /*******************************************************************************
329 Protocol management and registration routines
331 *******************************************************************************/
334 * Add a protocol ID to the list. Now that the input handler is
335 * smarter we can dispense with all the messy stuff that used to be
338 * BEWARE!!! Protocol handlers, mangling input packets,
339 * MUST BE last in hash buckets and checking protocol handlers
340 * MUST start from promiscuous ptype_all chain in net_bh.
341 * It is true now, do not change it.
342 * Explanation follows: if protocol handler, mangling packet, will
343 * be the first on list, it is not able to sense, that packet
344 * is cloned and should be copied-on-write, so that it will
345 * change it and subsequent readers will get broken packet.
350 * dev_add_pack - add packet handler
351 * @pt: packet type declaration
353 * Add a protocol handler to the networking stack. The passed &packet_type
354 * is linked into kernel lists and may not be freed until it has been
355 * removed from the kernel lists.
357 * This call does not sleep therefore it can not
358 * guarantee all CPU's that are in middle of receiving packets
359 * will see the new packet type (until the next received packet).
362 void dev_add_pack(struct packet_type
*pt
)
366 spin_lock_bh(&ptype_lock
);
367 if (pt
->type
== htons(ETH_P_ALL
))
368 list_add_rcu(&pt
->list
, &ptype_all
);
370 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
371 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
373 spin_unlock_bh(&ptype_lock
);
377 * __dev_remove_pack - remove packet handler
378 * @pt: packet type declaration
380 * Remove a protocol handler that was previously added to the kernel
381 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
382 * from the kernel lists and can be freed or reused once this function
385 * The packet type might still be in use by receivers
386 * and must not be freed until after all the CPU's have gone
387 * through a quiescent state.
389 void __dev_remove_pack(struct packet_type
*pt
)
391 struct list_head
*head
;
392 struct packet_type
*pt1
;
394 spin_lock_bh(&ptype_lock
);
396 if (pt
->type
== htons(ETH_P_ALL
))
399 head
= &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
401 list_for_each_entry(pt1
, head
, list
) {
403 list_del_rcu(&pt
->list
);
408 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
410 spin_unlock_bh(&ptype_lock
);
413 * dev_remove_pack - remove packet handler
414 * @pt: packet type declaration
416 * Remove a protocol handler that was previously added to the kernel
417 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
418 * from the kernel lists and can be freed or reused once this function
421 * This call sleeps to guarantee that no CPU is looking at the packet
424 void dev_remove_pack(struct packet_type
*pt
)
426 __dev_remove_pack(pt
);
431 /******************************************************************************
433 Device Boot-time Settings Routines
435 *******************************************************************************/
437 /* Boot time configuration table */
438 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
441 * netdev_boot_setup_add - add new setup entry
442 * @name: name of the device
443 * @map: configured settings for the device
445 * Adds new setup entry to the dev_boot_setup list. The function
446 * returns 0 on error and 1 on success. This is a generic routine to
449 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
451 struct netdev_boot_setup
*s
;
455 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
456 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
457 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
458 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
459 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
464 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
468 * netdev_boot_setup_check - check boot time settings
469 * @dev: the netdevice
471 * Check boot time settings for the device.
472 * The found settings are set for the device to be used
473 * later in the device probing.
474 * Returns 0 if no settings found, 1 if they are.
476 int netdev_boot_setup_check(struct net_device
*dev
)
478 struct netdev_boot_setup
*s
= dev_boot_setup
;
481 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
482 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
483 !strcmp(dev
->name
, s
[i
].name
)) {
484 dev
->irq
= s
[i
].map
.irq
;
485 dev
->base_addr
= s
[i
].map
.base_addr
;
486 dev
->mem_start
= s
[i
].map
.mem_start
;
487 dev
->mem_end
= s
[i
].map
.mem_end
;
496 * netdev_boot_base - get address from boot time settings
497 * @prefix: prefix for network device
498 * @unit: id for network device
500 * Check boot time settings for the base address of device.
501 * The found settings are set for the device to be used
502 * later in the device probing.
503 * Returns 0 if no settings found.
505 unsigned long netdev_boot_base(const char *prefix
, int unit
)
507 const struct netdev_boot_setup
*s
= dev_boot_setup
;
511 sprintf(name
, "%s%d", prefix
, unit
);
514 * If device already registered then return base of 1
515 * to indicate not to probe for this interface
517 if (__dev_get_by_name(&init_net
, name
))
520 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
521 if (!strcmp(name
, s
[i
].name
))
522 return s
[i
].map
.base_addr
;
527 * Saves at boot time configured settings for any netdevice.
529 int __init
netdev_boot_setup(char *str
)
534 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
539 memset(&map
, 0, sizeof(map
));
543 map
.base_addr
= ints
[2];
545 map
.mem_start
= ints
[3];
547 map
.mem_end
= ints
[4];
549 /* Add new entry to the list */
550 return netdev_boot_setup_add(str
, &map
);
553 __setup("netdev=", netdev_boot_setup
);
555 /*******************************************************************************
557 Device Interface Subroutines
559 *******************************************************************************/
562 * __dev_get_by_name - find a device by its name
563 * @net: the applicable net namespace
564 * @name: name to find
566 * Find an interface by name. Must be called under RTNL semaphore
567 * or @dev_base_lock. If the name is found a pointer to the device
568 * is returned. If the name is not found then %NULL is returned. The
569 * reference counters are not incremented so the caller must be
570 * careful with locks.
573 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
575 struct hlist_node
*p
;
577 hlist_for_each(p
, dev_name_hash(net
, name
)) {
578 struct net_device
*dev
579 = hlist_entry(p
, struct net_device
, name_hlist
);
580 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
587 * dev_get_by_name - find a device by its name
588 * @net: the applicable net namespace
589 * @name: name to find
591 * Find an interface by name. This can be called from any
592 * context and does its own locking. The returned handle has
593 * the usage count incremented and the caller must use dev_put() to
594 * release it when it is no longer needed. %NULL is returned if no
595 * matching device is found.
598 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
600 struct net_device
*dev
;
602 read_lock(&dev_base_lock
);
603 dev
= __dev_get_by_name(net
, name
);
606 read_unlock(&dev_base_lock
);
611 * __dev_get_by_index - find a device by its ifindex
612 * @net: the applicable net namespace
613 * @ifindex: index of device
615 * Search for an interface by index. Returns %NULL if the device
616 * is not found or a pointer to the device. The device has not
617 * had its reference counter increased so the caller must be careful
618 * about locking. The caller must hold either the RTNL semaphore
622 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
624 struct hlist_node
*p
;
626 hlist_for_each(p
, dev_index_hash(net
, ifindex
)) {
627 struct net_device
*dev
628 = hlist_entry(p
, struct net_device
, index_hlist
);
629 if (dev
->ifindex
== ifindex
)
637 * dev_get_by_index - find a device by its ifindex
638 * @net: the applicable net namespace
639 * @ifindex: index of device
641 * Search for an interface by index. Returns NULL if the device
642 * is not found or a pointer to the device. The device returned has
643 * had a reference added and the pointer is safe until the user calls
644 * dev_put to indicate they have finished with it.
647 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
649 struct net_device
*dev
;
651 read_lock(&dev_base_lock
);
652 dev
= __dev_get_by_index(net
, ifindex
);
655 read_unlock(&dev_base_lock
);
660 * dev_getbyhwaddr - find a device by its hardware address
661 * @net: the applicable net namespace
662 * @type: media type of device
663 * @ha: hardware address
665 * Search for an interface by MAC address. Returns NULL if the device
666 * is not found or a pointer to the device. The caller must hold the
667 * rtnl semaphore. The returned device has not had its ref count increased
668 * and the caller must therefore be careful about locking
671 * If the API was consistent this would be __dev_get_by_hwaddr
674 struct net_device
*dev_getbyhwaddr(struct net
*net
, unsigned short type
, char *ha
)
676 struct net_device
*dev
;
680 for_each_netdev(net
, dev
)
681 if (dev
->type
== type
&&
682 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
688 EXPORT_SYMBOL(dev_getbyhwaddr
);
690 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
692 struct net_device
*dev
;
695 for_each_netdev(net
, dev
)
696 if (dev
->type
== type
)
702 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
704 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
706 struct net_device
*dev
;
709 dev
= __dev_getfirstbyhwtype(net
, type
);
716 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
719 * dev_get_by_flags - find any device with given flags
720 * @net: the applicable net namespace
721 * @if_flags: IFF_* values
722 * @mask: bitmask of bits in if_flags to check
724 * Search for any interface with the given flags. Returns NULL if a device
725 * is not found or a pointer to the device. The device returned has
726 * had a reference added and the pointer is safe until the user calls
727 * dev_put to indicate they have finished with it.
730 struct net_device
* dev_get_by_flags(struct net
*net
, unsigned short if_flags
, unsigned short mask
)
732 struct net_device
*dev
, *ret
;
735 read_lock(&dev_base_lock
);
736 for_each_netdev(net
, dev
) {
737 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
743 read_unlock(&dev_base_lock
);
748 * dev_valid_name - check if name is okay for network device
751 * Network device names need to be valid file names to
752 * to allow sysfs to work. We also disallow any kind of
755 int dev_valid_name(const char *name
)
759 if (strlen(name
) >= IFNAMSIZ
)
761 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
765 if (*name
== '/' || isspace(*name
))
773 * __dev_alloc_name - allocate a name for a device
774 * @net: network namespace to allocate the device name in
775 * @name: name format string
776 * @buf: scratch buffer and result name string
778 * Passed a format string - eg "lt%d" it will try and find a suitable
779 * id. It scans list of devices to build up a free map, then chooses
780 * the first empty slot. The caller must hold the dev_base or rtnl lock
781 * while allocating the name and adding the device in order to avoid
783 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
784 * Returns the number of the unit assigned or a negative errno code.
787 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
791 const int max_netdevices
= 8*PAGE_SIZE
;
792 unsigned long *inuse
;
793 struct net_device
*d
;
795 p
= strnchr(name
, IFNAMSIZ
-1, '%');
798 * Verify the string as this thing may have come from
799 * the user. There must be either one "%d" and no other "%"
802 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
805 /* Use one page as a bit array of possible slots */
806 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
810 for_each_netdev(net
, d
) {
811 if (!sscanf(d
->name
, name
, &i
))
813 if (i
< 0 || i
>= max_netdevices
)
816 /* avoid cases where sscanf is not exact inverse of printf */
817 snprintf(buf
, IFNAMSIZ
, name
, i
);
818 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
822 i
= find_first_zero_bit(inuse
, max_netdevices
);
823 free_page((unsigned long) inuse
);
826 snprintf(buf
, IFNAMSIZ
, name
, i
);
827 if (!__dev_get_by_name(net
, buf
))
830 /* It is possible to run out of possible slots
831 * when the name is long and there isn't enough space left
832 * for the digits, or if all bits are used.
838 * dev_alloc_name - allocate a name for a device
840 * @name: name format string
842 * Passed a format string - eg "lt%d" it will try and find a suitable
843 * id. It scans list of devices to build up a free map, then chooses
844 * the first empty slot. The caller must hold the dev_base or rtnl lock
845 * while allocating the name and adding the device in order to avoid
847 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
848 * Returns the number of the unit assigned or a negative errno code.
851 int dev_alloc_name(struct net_device
*dev
, const char *name
)
857 BUG_ON(!dev_net(dev
));
859 ret
= __dev_alloc_name(net
, name
, buf
);
861 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
867 * dev_change_name - change name of a device
869 * @newname: name (or format string) must be at least IFNAMSIZ
871 * Change name of a device, can pass format strings "eth%d".
874 int dev_change_name(struct net_device
*dev
, char *newname
)
876 char oldname
[IFNAMSIZ
];
882 BUG_ON(!dev_net(dev
));
885 if (dev
->flags
& IFF_UP
)
888 if (!dev_valid_name(newname
))
891 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
894 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
896 if (strchr(newname
, '%')) {
897 err
= dev_alloc_name(dev
, newname
);
900 strcpy(newname
, dev
->name
);
902 else if (__dev_get_by_name(net
, newname
))
905 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
908 err
= device_rename(&dev
->dev
, dev
->name
);
910 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
914 write_lock_bh(&dev_base_lock
);
915 hlist_del(&dev
->name_hlist
);
916 hlist_add_head(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
917 write_unlock_bh(&dev_base_lock
);
919 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
920 ret
= notifier_to_errno(ret
);
925 "%s: name change rollback failed: %d.\n",
929 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
938 * netdev_features_change - device changes features
939 * @dev: device to cause notification
941 * Called to indicate a device has changed features.
943 void netdev_features_change(struct net_device
*dev
)
945 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
947 EXPORT_SYMBOL(netdev_features_change
);
950 * netdev_state_change - device changes state
951 * @dev: device to cause notification
953 * Called to indicate a device has changed state. This function calls
954 * the notifier chains for netdev_chain and sends a NEWLINK message
955 * to the routing socket.
957 void netdev_state_change(struct net_device
*dev
)
959 if (dev
->flags
& IFF_UP
) {
960 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
961 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
965 void netdev_bonding_change(struct net_device
*dev
)
967 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER
, dev
);
969 EXPORT_SYMBOL(netdev_bonding_change
);
972 * dev_load - load a network module
973 * @net: the applicable net namespace
974 * @name: name of interface
976 * If a network interface is not present and the process has suitable
977 * privileges this function loads the module. If module loading is not
978 * available in this kernel then it becomes a nop.
981 void dev_load(struct net
*net
, const char *name
)
983 struct net_device
*dev
;
985 read_lock(&dev_base_lock
);
986 dev
= __dev_get_by_name(net
, name
);
987 read_unlock(&dev_base_lock
);
989 if (!dev
&& capable(CAP_SYS_MODULE
))
990 request_module("%s", name
);
994 * dev_open - prepare an interface for use.
995 * @dev: device to open
997 * Takes a device from down to up state. The device's private open
998 * function is invoked and then the multicast lists are loaded. Finally
999 * the device is moved into the up state and a %NETDEV_UP message is
1000 * sent to the netdev notifier chain.
1002 * Calling this function on an active interface is a nop. On a failure
1003 * a negative errno code is returned.
1005 int dev_open(struct net_device
*dev
)
1015 if (dev
->flags
& IFF_UP
)
1019 * Is it even present?
1021 if (!netif_device_present(dev
))
1025 * Call device private open method
1027 set_bit(__LINK_STATE_START
, &dev
->state
);
1029 if (dev
->validate_addr
)
1030 ret
= dev
->validate_addr(dev
);
1032 if (!ret
&& dev
->open
)
1033 ret
= dev
->open(dev
);
1036 * If it went open OK then:
1040 clear_bit(__LINK_STATE_START
, &dev
->state
);
1045 dev
->flags
|= IFF_UP
;
1048 * Initialize multicasting status
1050 dev_set_rx_mode(dev
);
1053 * Wakeup transmit queue engine
1058 * ... and announce new interface.
1060 call_netdevice_notifiers(NETDEV_UP
, dev
);
1067 * dev_close - shutdown an interface.
1068 * @dev: device to shutdown
1070 * This function moves an active device into down state. A
1071 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1072 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1075 int dev_close(struct net_device
*dev
)
1081 if (!(dev
->flags
& IFF_UP
))
1085 * Tell people we are going down, so that they can
1086 * prepare to death, when device is still operating.
1088 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1090 clear_bit(__LINK_STATE_START
, &dev
->state
);
1092 /* Synchronize to scheduled poll. We cannot touch poll list,
1093 * it can be even on different cpu. So just clear netif_running().
1095 * dev->stop() will invoke napi_disable() on all of it's
1096 * napi_struct instances on this device.
1098 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1100 dev_deactivate(dev
);
1103 * Call the device specific close. This cannot fail.
1104 * Only if device is UP
1106 * We allow it to be called even after a DETACH hot-plug
1113 * Device is now down.
1116 dev
->flags
&= ~IFF_UP
;
1119 * Tell people we are down
1121 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1128 * dev_disable_lro - disable Large Receive Offload on a device
1131 * Disable Large Receive Offload (LRO) on a net device. Must be
1132 * called under RTNL. This is needed if received packets may be
1133 * forwarded to another interface.
1135 void dev_disable_lro(struct net_device
*dev
)
1137 if (dev
->ethtool_ops
&& dev
->ethtool_ops
->get_flags
&&
1138 dev
->ethtool_ops
->set_flags
) {
1139 u32 flags
= dev
->ethtool_ops
->get_flags(dev
);
1140 if (flags
& ETH_FLAG_LRO
) {
1141 flags
&= ~ETH_FLAG_LRO
;
1142 dev
->ethtool_ops
->set_flags(dev
, flags
);
1145 WARN_ON(dev
->features
& NETIF_F_LRO
);
1147 EXPORT_SYMBOL(dev_disable_lro
);
1150 static int dev_boot_phase
= 1;
1153 * Device change register/unregister. These are not inline or static
1154 * as we export them to the world.
1158 * register_netdevice_notifier - register a network notifier block
1161 * Register a notifier to be called when network device events occur.
1162 * The notifier passed is linked into the kernel structures and must
1163 * not be reused until it has been unregistered. A negative errno code
1164 * is returned on a failure.
1166 * When registered all registration and up events are replayed
1167 * to the new notifier to allow device to have a race free
1168 * view of the network device list.
1171 int register_netdevice_notifier(struct notifier_block
*nb
)
1173 struct net_device
*dev
;
1174 struct net_device
*last
;
1179 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1185 for_each_netdev(net
, dev
) {
1186 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1187 err
= notifier_to_errno(err
);
1191 if (!(dev
->flags
& IFF_UP
))
1194 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1205 for_each_netdev(net
, dev
) {
1209 if (dev
->flags
& IFF_UP
) {
1210 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1211 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1213 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1217 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1222 * unregister_netdevice_notifier - unregister a network notifier block
1225 * Unregister a notifier previously registered by
1226 * register_netdevice_notifier(). The notifier is unlinked into the
1227 * kernel structures and may then be reused. A negative errno code
1228 * is returned on a failure.
1231 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1236 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1242 * call_netdevice_notifiers - call all network notifier blocks
1243 * @val: value passed unmodified to notifier function
1244 * @dev: net_device pointer passed unmodified to notifier function
1246 * Call all network notifier blocks. Parameters and return value
1247 * are as for raw_notifier_call_chain().
1250 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1252 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1255 /* When > 0 there are consumers of rx skb time stamps */
1256 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1258 void net_enable_timestamp(void)
1260 atomic_inc(&netstamp_needed
);
1263 void net_disable_timestamp(void)
1265 atomic_dec(&netstamp_needed
);
1268 static inline void net_timestamp(struct sk_buff
*skb
)
1270 if (atomic_read(&netstamp_needed
))
1271 __net_timestamp(skb
);
1273 skb
->tstamp
.tv64
= 0;
1277 * Support routine. Sends outgoing frames to any network
1278 * taps currently in use.
1281 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1283 struct packet_type
*ptype
;
1288 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1289 /* Never send packets back to the socket
1290 * they originated from - MvS (miquels@drinkel.ow.org)
1292 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1293 (ptype
->af_packet_priv
== NULL
||
1294 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1295 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1299 /* skb->nh should be correctly
1300 set by sender, so that the second statement is
1301 just protection against buggy protocols.
1303 skb_reset_mac_header(skb2
);
1305 if (skb_network_header(skb2
) < skb2
->data
||
1306 skb2
->network_header
> skb2
->tail
) {
1307 if (net_ratelimit())
1308 printk(KERN_CRIT
"protocol %04x is "
1310 skb2
->protocol
, dev
->name
);
1311 skb_reset_network_header(skb2
);
1314 skb2
->transport_header
= skb2
->network_header
;
1315 skb2
->pkt_type
= PACKET_OUTGOING
;
1316 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1323 void __netif_schedule(struct netdev_queue
*txq
)
1325 struct net_device
*dev
= txq
->dev
;
1327 if (!test_and_set_bit(__LINK_STATE_SCHED
, &dev
->state
)) {
1328 struct softnet_data
*sd
;
1329 unsigned long flags
;
1331 local_irq_save(flags
);
1332 sd
= &__get_cpu_var(softnet_data
);
1333 txq
->next_sched
= sd
->output_queue
;
1334 sd
->output_queue
= txq
;
1335 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1336 local_irq_restore(flags
);
1339 EXPORT_SYMBOL(__netif_schedule
);
1341 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1343 if (atomic_dec_and_test(&skb
->users
)) {
1344 struct softnet_data
*sd
;
1345 unsigned long flags
;
1347 local_irq_save(flags
);
1348 sd
= &__get_cpu_var(softnet_data
);
1349 skb
->next
= sd
->completion_queue
;
1350 sd
->completion_queue
= skb
;
1351 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1352 local_irq_restore(flags
);
1355 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1357 void dev_kfree_skb_any(struct sk_buff
*skb
)
1359 if (in_irq() || irqs_disabled())
1360 dev_kfree_skb_irq(skb
);
1364 EXPORT_SYMBOL(dev_kfree_skb_any
);
1368 * netif_device_detach - mark device as removed
1369 * @dev: network device
1371 * Mark device as removed from system and therefore no longer available.
1373 void netif_device_detach(struct net_device
*dev
)
1375 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1376 netif_running(dev
)) {
1377 netif_stop_queue(dev
);
1380 EXPORT_SYMBOL(netif_device_detach
);
1383 * netif_device_attach - mark device as attached
1384 * @dev: network device
1386 * Mark device as attached from system and restart if needed.
1388 void netif_device_attach(struct net_device
*dev
)
1390 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1391 netif_running(dev
)) {
1392 netif_wake_queue(dev
);
1393 __netdev_watchdog_up(dev
);
1396 EXPORT_SYMBOL(netif_device_attach
);
1398 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1400 return ((features
& NETIF_F_GEN_CSUM
) ||
1401 ((features
& NETIF_F_IP_CSUM
) &&
1402 protocol
== htons(ETH_P_IP
)) ||
1403 ((features
& NETIF_F_IPV6_CSUM
) &&
1404 protocol
== htons(ETH_P_IPV6
)));
1407 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1409 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1412 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1413 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1414 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1415 veh
->h_vlan_encapsulated_proto
))
1423 * Invalidate hardware checksum when packet is to be mangled, and
1424 * complete checksum manually on outgoing path.
1426 int skb_checksum_help(struct sk_buff
*skb
)
1429 int ret
= 0, offset
;
1431 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1432 goto out_set_summed
;
1434 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1435 /* Let GSO fix up the checksum. */
1436 goto out_set_summed
;
1439 offset
= skb
->csum_start
- skb_headroom(skb
);
1440 BUG_ON(offset
>= skb_headlen(skb
));
1441 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1443 offset
+= skb
->csum_offset
;
1444 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1446 if (skb_cloned(skb
) &&
1447 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1448 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1453 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1455 skb
->ip_summed
= CHECKSUM_NONE
;
1461 * skb_gso_segment - Perform segmentation on skb.
1462 * @skb: buffer to segment
1463 * @features: features for the output path (see dev->features)
1465 * This function segments the given skb and returns a list of segments.
1467 * It may return NULL if the skb requires no segmentation. This is
1468 * only possible when GSO is used for verifying header integrity.
1470 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1472 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1473 struct packet_type
*ptype
;
1474 __be16 type
= skb
->protocol
;
1477 BUG_ON(skb_shinfo(skb
)->frag_list
);
1479 skb_reset_mac_header(skb
);
1480 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1481 __skb_pull(skb
, skb
->mac_len
);
1483 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1484 if (skb_header_cloned(skb
) &&
1485 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1486 return ERR_PTR(err
);
1490 list_for_each_entry_rcu(ptype
,
1491 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1492 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1493 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1494 err
= ptype
->gso_send_check(skb
);
1495 segs
= ERR_PTR(err
);
1496 if (err
|| skb_gso_ok(skb
, features
))
1498 __skb_push(skb
, (skb
->data
-
1499 skb_network_header(skb
)));
1501 segs
= ptype
->gso_segment(skb
, features
);
1507 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1512 EXPORT_SYMBOL(skb_gso_segment
);
1514 /* Take action when hardware reception checksum errors are detected. */
1516 void netdev_rx_csum_fault(struct net_device
*dev
)
1518 if (net_ratelimit()) {
1519 printk(KERN_ERR
"%s: hw csum failure.\n",
1520 dev
? dev
->name
: "<unknown>");
1524 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1527 /* Actually, we should eliminate this check as soon as we know, that:
1528 * 1. IOMMU is present and allows to map all the memory.
1529 * 2. No high memory really exists on this machine.
1532 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1534 #ifdef CONFIG_HIGHMEM
1537 if (dev
->features
& NETIF_F_HIGHDMA
)
1540 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1541 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1549 void (*destructor
)(struct sk_buff
*skb
);
1552 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1554 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1556 struct dev_gso_cb
*cb
;
1559 struct sk_buff
*nskb
= skb
->next
;
1561 skb
->next
= nskb
->next
;
1564 } while (skb
->next
);
1566 cb
= DEV_GSO_CB(skb
);
1568 cb
->destructor(skb
);
1572 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1573 * @skb: buffer to segment
1575 * This function segments the given skb and stores the list of segments
1578 static int dev_gso_segment(struct sk_buff
*skb
)
1580 struct net_device
*dev
= skb
->dev
;
1581 struct sk_buff
*segs
;
1582 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1585 segs
= skb_gso_segment(skb
, features
);
1587 /* Verifying header integrity only. */
1592 return PTR_ERR(segs
);
1595 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1596 skb
->destructor
= dev_gso_skb_destructor
;
1601 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1603 if (likely(!skb
->next
)) {
1604 if (!list_empty(&ptype_all
))
1605 dev_queue_xmit_nit(skb
, dev
);
1607 if (netif_needs_gso(dev
, skb
)) {
1608 if (unlikely(dev_gso_segment(skb
)))
1614 return dev
->hard_start_xmit(skb
, dev
);
1619 struct sk_buff
*nskb
= skb
->next
;
1622 skb
->next
= nskb
->next
;
1624 rc
= dev
->hard_start_xmit(nskb
, dev
);
1626 nskb
->next
= skb
->next
;
1630 if (unlikely((netif_queue_stopped(dev
) ||
1631 netif_subqueue_stopped(dev
, skb
)) &&
1633 return NETDEV_TX_BUSY
;
1634 } while (skb
->next
);
1636 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1644 * dev_queue_xmit - transmit a buffer
1645 * @skb: buffer to transmit
1647 * Queue a buffer for transmission to a network device. The caller must
1648 * have set the device and priority and built the buffer before calling
1649 * this function. The function can be called from an interrupt.
1651 * A negative errno code is returned on a failure. A success does not
1652 * guarantee the frame will be transmitted as it may be dropped due
1653 * to congestion or traffic shaping.
1655 * -----------------------------------------------------------------------------------
1656 * I notice this method can also return errors from the queue disciplines,
1657 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1660 * Regardless of the return value, the skb is consumed, so it is currently
1661 * difficult to retry a send to this method. (You can bump the ref count
1662 * before sending to hold a reference for retry if you are careful.)
1664 * When calling this method, interrupts MUST be enabled. This is because
1665 * the BH enable code must have IRQs enabled so that it will not deadlock.
1669 int dev_queue_xmit(struct sk_buff
*skb
)
1671 struct net_device
*dev
= skb
->dev
;
1672 struct netdev_queue
*txq
;
1676 /* GSO will handle the following emulations directly. */
1677 if (netif_needs_gso(dev
, skb
))
1680 if (skb_shinfo(skb
)->frag_list
&&
1681 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1682 __skb_linearize(skb
))
1685 /* Fragmented skb is linearized if device does not support SG,
1686 * or if at least one of fragments is in highmem and device
1687 * does not support DMA from it.
1689 if (skb_shinfo(skb
)->nr_frags
&&
1690 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1691 __skb_linearize(skb
))
1694 /* If packet is not checksummed and device does not support
1695 * checksumming for this protocol, complete checksumming here.
1697 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1698 skb_set_transport_header(skb
, skb
->csum_start
-
1700 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
1705 txq
= &dev
->tx_queue
;
1706 spin_lock_prefetch(&txq
->lock
);
1708 /* Disable soft irqs for various locks below. Also
1709 * stops preemption for RCU.
1713 /* Updates of qdisc are serialized by queue->lock.
1714 * The struct Qdisc which is pointed to by qdisc is now a
1715 * rcu structure - it may be accessed without acquiring
1716 * a lock (but the structure may be stale.) The freeing of the
1717 * qdisc will be deferred until it's known that there are no
1718 * more references to it.
1720 * If the qdisc has an enqueue function, we still need to
1721 * hold the queue->lock before calling it, since queue->lock
1722 * also serializes access to the device queue.
1725 q
= rcu_dereference(txq
->qdisc
);
1726 #ifdef CONFIG_NET_CLS_ACT
1727 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1730 /* Grab device queue */
1731 spin_lock(&txq
->lock
);
1734 /* reset queue_mapping to zero */
1735 skb_set_queue_mapping(skb
, 0);
1736 rc
= q
->enqueue(skb
, q
);
1738 spin_unlock(&txq
->lock
);
1740 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1743 spin_unlock(&txq
->lock
);
1746 /* The device has no queue. Common case for software devices:
1747 loopback, all the sorts of tunnels...
1749 Really, it is unlikely that netif_tx_lock protection is necessary
1750 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1752 However, it is possible, that they rely on protection
1755 Check this and shot the lock. It is not prone from deadlocks.
1756 Either shot noqueue qdisc, it is even simpler 8)
1758 if (dev
->flags
& IFF_UP
) {
1759 int cpu
= smp_processor_id(); /* ok because BHs are off */
1761 if (txq
->xmit_lock_owner
!= cpu
) {
1763 HARD_TX_LOCK(dev
, txq
, cpu
);
1765 if (!netif_queue_stopped(dev
) &&
1766 !netif_subqueue_stopped(dev
, skb
)) {
1768 if (!dev_hard_start_xmit(skb
, dev
)) {
1769 HARD_TX_UNLOCK(dev
, txq
);
1773 HARD_TX_UNLOCK(dev
, txq
);
1774 if (net_ratelimit())
1775 printk(KERN_CRIT
"Virtual device %s asks to "
1776 "queue packet!\n", dev
->name
);
1778 /* Recursion is detected! It is possible,
1780 if (net_ratelimit())
1781 printk(KERN_CRIT
"Dead loop on virtual device "
1782 "%s, fix it urgently!\n", dev
->name
);
1787 rcu_read_unlock_bh();
1793 rcu_read_unlock_bh();
1798 /*=======================================================================
1800 =======================================================================*/
1802 int netdev_max_backlog __read_mostly
= 1000;
1803 int netdev_budget __read_mostly
= 300;
1804 int weight_p __read_mostly
= 64; /* old backlog weight */
1806 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1810 * netif_rx - post buffer to the network code
1811 * @skb: buffer to post
1813 * This function receives a packet from a device driver and queues it for
1814 * the upper (protocol) levels to process. It always succeeds. The buffer
1815 * may be dropped during processing for congestion control or by the
1819 * NET_RX_SUCCESS (no congestion)
1820 * NET_RX_DROP (packet was dropped)
1824 int netif_rx(struct sk_buff
*skb
)
1826 struct softnet_data
*queue
;
1827 unsigned long flags
;
1829 /* if netpoll wants it, pretend we never saw it */
1830 if (netpoll_rx(skb
))
1833 if (!skb
->tstamp
.tv64
)
1837 * The code is rearranged so that the path is the most
1838 * short when CPU is congested, but is still operating.
1840 local_irq_save(flags
);
1841 queue
= &__get_cpu_var(softnet_data
);
1843 __get_cpu_var(netdev_rx_stat
).total
++;
1844 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1845 if (queue
->input_pkt_queue
.qlen
) {
1848 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1849 local_irq_restore(flags
);
1850 return NET_RX_SUCCESS
;
1853 napi_schedule(&queue
->backlog
);
1857 __get_cpu_var(netdev_rx_stat
).dropped
++;
1858 local_irq_restore(flags
);
1864 int netif_rx_ni(struct sk_buff
*skb
)
1869 err
= netif_rx(skb
);
1870 if (local_softirq_pending())
1877 EXPORT_SYMBOL(netif_rx_ni
);
1879 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1881 struct net_device
*dev
= skb
->dev
;
1884 if (skb_bond_should_drop(skb
)) {
1888 skb
->dev
= dev
->master
;
1895 static void net_tx_action(struct softirq_action
*h
)
1897 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1899 if (sd
->completion_queue
) {
1900 struct sk_buff
*clist
;
1902 local_irq_disable();
1903 clist
= sd
->completion_queue
;
1904 sd
->completion_queue
= NULL
;
1908 struct sk_buff
*skb
= clist
;
1909 clist
= clist
->next
;
1911 BUG_TRAP(!atomic_read(&skb
->users
));
1916 if (sd
->output_queue
) {
1917 struct netdev_queue
*head
;
1919 local_irq_disable();
1920 head
= sd
->output_queue
;
1921 sd
->output_queue
= NULL
;
1925 struct netdev_queue
*txq
= head
;
1926 struct net_device
*dev
= txq
->dev
;
1927 head
= head
->next_sched
;
1929 smp_mb__before_clear_bit();
1930 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1932 if (spin_trylock(&txq
->lock
)) {
1934 spin_unlock(&txq
->lock
);
1936 netif_schedule_queue(txq
);
1942 static inline int deliver_skb(struct sk_buff
*skb
,
1943 struct packet_type
*pt_prev
,
1944 struct net_device
*orig_dev
)
1946 atomic_inc(&skb
->users
);
1947 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1950 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1951 /* These hooks defined here for ATM */
1953 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1954 unsigned char *addr
);
1955 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
1958 * If bridge module is loaded call bridging hook.
1959 * returns NULL if packet was consumed.
1961 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
1962 struct sk_buff
*skb
) __read_mostly
;
1963 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
1964 struct packet_type
**pt_prev
, int *ret
,
1965 struct net_device
*orig_dev
)
1967 struct net_bridge_port
*port
;
1969 if (skb
->pkt_type
== PACKET_LOOPBACK
||
1970 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
1974 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1978 return br_handle_frame_hook(port
, skb
);
1981 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1984 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1985 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
1986 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
1988 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
1989 struct packet_type
**pt_prev
,
1991 struct net_device
*orig_dev
)
1993 if (skb
->dev
->macvlan_port
== NULL
)
1997 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2000 return macvlan_handle_frame_hook(skb
);
2003 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2006 #ifdef CONFIG_NET_CLS_ACT
2007 /* TODO: Maybe we should just force sch_ingress to be compiled in
2008 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2009 * a compare and 2 stores extra right now if we dont have it on
2010 * but have CONFIG_NET_CLS_ACT
2011 * NOTE: This doesnt stop any functionality; if you dont have
2012 * the ingress scheduler, you just cant add policies on ingress.
2015 static int ing_filter(struct sk_buff
*skb
)
2017 struct net_device
*dev
= skb
->dev
;
2018 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2019 struct netdev_queue
*rxq
;
2020 int result
= TC_ACT_OK
;
2023 if (MAX_RED_LOOP
< ttl
++) {
2025 "Redir loop detected Dropping packet (%d->%d)\n",
2026 skb
->iif
, dev
->ifindex
);
2030 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2031 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2033 rxq
= &dev
->rx_queue
;
2035 spin_lock(&rxq
->lock
);
2036 if ((q
= rxq
->qdisc
) != NULL
)
2037 result
= q
->enqueue(skb
, q
);
2038 spin_unlock(&rxq
->lock
);
2043 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2044 struct packet_type
**pt_prev
,
2045 int *ret
, struct net_device
*orig_dev
)
2047 if (!skb
->dev
->rx_queue
.qdisc
)
2051 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2054 /* Huh? Why does turning on AF_PACKET affect this? */
2055 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2058 switch (ing_filter(skb
)) {
2072 * netif_nit_deliver - deliver received packets to network taps
2075 * This function is used to deliver incoming packets to network
2076 * taps. It should be used when the normal netif_receive_skb path
2077 * is bypassed, for example because of VLAN acceleration.
2079 void netif_nit_deliver(struct sk_buff
*skb
)
2081 struct packet_type
*ptype
;
2083 if (list_empty(&ptype_all
))
2086 skb_reset_network_header(skb
);
2087 skb_reset_transport_header(skb
);
2088 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2091 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2092 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
)
2093 deliver_skb(skb
, ptype
, skb
->dev
);
2099 * netif_receive_skb - process receive buffer from network
2100 * @skb: buffer to process
2102 * netif_receive_skb() is the main receive data processing function.
2103 * It always succeeds. The buffer may be dropped during processing
2104 * for congestion control or by the protocol layers.
2106 * This function may only be called from softirq context and interrupts
2107 * should be enabled.
2109 * Return values (usually ignored):
2110 * NET_RX_SUCCESS: no congestion
2111 * NET_RX_DROP: packet was dropped
2113 int netif_receive_skb(struct sk_buff
*skb
)
2115 struct packet_type
*ptype
, *pt_prev
;
2116 struct net_device
*orig_dev
;
2117 int ret
= NET_RX_DROP
;
2120 /* if we've gotten here through NAPI, check netpoll */
2121 if (netpoll_receive_skb(skb
))
2124 if (!skb
->tstamp
.tv64
)
2128 skb
->iif
= skb
->dev
->ifindex
;
2130 orig_dev
= skb_bond(skb
);
2135 __get_cpu_var(netdev_rx_stat
).total
++;
2137 skb_reset_network_header(skb
);
2138 skb_reset_transport_header(skb
);
2139 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2145 /* Don't receive packets in an exiting network namespace */
2146 if (!net_alive(dev_net(skb
->dev
)))
2149 #ifdef CONFIG_NET_CLS_ACT
2150 if (skb
->tc_verd
& TC_NCLS
) {
2151 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2156 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2157 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
2159 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2164 #ifdef CONFIG_NET_CLS_ACT
2165 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2171 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
2174 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
2178 type
= skb
->protocol
;
2179 list_for_each_entry_rcu(ptype
,
2180 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2181 if (ptype
->type
== type
&&
2182 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
2184 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2190 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2193 /* Jamal, now you will not able to escape explaining
2194 * me how you were going to use this. :-)
2204 static int process_backlog(struct napi_struct
*napi
, int quota
)
2207 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2208 unsigned long start_time
= jiffies
;
2210 napi
->weight
= weight_p
;
2212 struct sk_buff
*skb
;
2213 struct net_device
*dev
;
2215 local_irq_disable();
2216 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
2218 __napi_complete(napi
);
2227 netif_receive_skb(skb
);
2230 } while (++work
< quota
&& jiffies
== start_time
);
2236 * __napi_schedule - schedule for receive
2237 * @n: entry to schedule
2239 * The entry's receive function will be scheduled to run
2241 void __napi_schedule(struct napi_struct
*n
)
2243 unsigned long flags
;
2245 local_irq_save(flags
);
2246 list_add_tail(&n
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
2247 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2248 local_irq_restore(flags
);
2250 EXPORT_SYMBOL(__napi_schedule
);
2253 static void net_rx_action(struct softirq_action
*h
)
2255 struct list_head
*list
= &__get_cpu_var(softnet_data
).poll_list
;
2256 unsigned long start_time
= jiffies
;
2257 int budget
= netdev_budget
;
2260 local_irq_disable();
2262 while (!list_empty(list
)) {
2263 struct napi_struct
*n
;
2266 /* If softirq window is exhuasted then punt.
2268 * Note that this is a slight policy change from the
2269 * previous NAPI code, which would allow up to 2
2270 * jiffies to pass before breaking out. The test
2271 * used to be "jiffies - start_time > 1".
2273 if (unlikely(budget
<= 0 || jiffies
!= start_time
))
2278 /* Even though interrupts have been re-enabled, this
2279 * access is safe because interrupts can only add new
2280 * entries to the tail of this list, and only ->poll()
2281 * calls can remove this head entry from the list.
2283 n
= list_entry(list
->next
, struct napi_struct
, poll_list
);
2285 have
= netpoll_poll_lock(n
);
2289 /* This NAPI_STATE_SCHED test is for avoiding a race
2290 * with netpoll's poll_napi(). Only the entity which
2291 * obtains the lock and sees NAPI_STATE_SCHED set will
2292 * actually make the ->poll() call. Therefore we avoid
2293 * accidently calling ->poll() when NAPI is not scheduled.
2296 if (test_bit(NAPI_STATE_SCHED
, &n
->state
))
2297 work
= n
->poll(n
, weight
);
2299 WARN_ON_ONCE(work
> weight
);
2303 local_irq_disable();
2305 /* Drivers must not modify the NAPI state if they
2306 * consume the entire weight. In such cases this code
2307 * still "owns" the NAPI instance and therefore can
2308 * move the instance around on the list at-will.
2310 if (unlikely(work
== weight
)) {
2311 if (unlikely(napi_disable_pending(n
)))
2314 list_move_tail(&n
->poll_list
, list
);
2317 netpoll_poll_unlock(have
);
2322 #ifdef CONFIG_NET_DMA
2324 * There may not be any more sk_buffs coming right now, so push
2325 * any pending DMA copies to hardware
2327 if (!cpus_empty(net_dma
.channel_mask
)) {
2329 for_each_cpu_mask(chan_idx
, net_dma
.channel_mask
) {
2330 struct dma_chan
*chan
= net_dma
.channels
[chan_idx
];
2332 dma_async_memcpy_issue_pending(chan
);
2340 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2341 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2345 static gifconf_func_t
* gifconf_list
[NPROTO
];
2348 * register_gifconf - register a SIOCGIF handler
2349 * @family: Address family
2350 * @gifconf: Function handler
2352 * Register protocol dependent address dumping routines. The handler
2353 * that is passed must not be freed or reused until it has been replaced
2354 * by another handler.
2356 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2358 if (family
>= NPROTO
)
2360 gifconf_list
[family
] = gifconf
;
2366 * Map an interface index to its name (SIOCGIFNAME)
2370 * We need this ioctl for efficient implementation of the
2371 * if_indextoname() function required by the IPv6 API. Without
2372 * it, we would have to search all the interfaces to find a
2376 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
2378 struct net_device
*dev
;
2382 * Fetch the caller's info block.
2385 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2388 read_lock(&dev_base_lock
);
2389 dev
= __dev_get_by_index(net
, ifr
.ifr_ifindex
);
2391 read_unlock(&dev_base_lock
);
2395 strcpy(ifr
.ifr_name
, dev
->name
);
2396 read_unlock(&dev_base_lock
);
2398 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2404 * Perform a SIOCGIFCONF call. This structure will change
2405 * size eventually, and there is nothing I can do about it.
2406 * Thus we will need a 'compatibility mode'.
2409 static int dev_ifconf(struct net
*net
, char __user
*arg
)
2412 struct net_device
*dev
;
2419 * Fetch the caller's info block.
2422 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2429 * Loop over the interfaces, and write an info block for each.
2433 for_each_netdev(net
, dev
) {
2434 for (i
= 0; i
< NPROTO
; i
++) {
2435 if (gifconf_list
[i
]) {
2438 done
= gifconf_list
[i
](dev
, NULL
, 0);
2440 done
= gifconf_list
[i
](dev
, pos
+ total
,
2450 * All done. Write the updated control block back to the caller.
2452 ifc
.ifc_len
= total
;
2455 * Both BSD and Solaris return 0 here, so we do too.
2457 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2460 #ifdef CONFIG_PROC_FS
2462 * This is invoked by the /proc filesystem handler to display a device
2465 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2466 __acquires(dev_base_lock
)
2468 struct net
*net
= seq_file_net(seq
);
2470 struct net_device
*dev
;
2472 read_lock(&dev_base_lock
);
2474 return SEQ_START_TOKEN
;
2477 for_each_netdev(net
, dev
)
2484 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2486 struct net
*net
= seq_file_net(seq
);
2488 return v
== SEQ_START_TOKEN
?
2489 first_net_device(net
) : next_net_device((struct net_device
*)v
);
2492 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2493 __releases(dev_base_lock
)
2495 read_unlock(&dev_base_lock
);
2498 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2500 struct net_device_stats
*stats
= dev
->get_stats(dev
);
2502 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2503 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2504 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2506 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2507 stats
->rx_fifo_errors
,
2508 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2509 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2510 stats
->rx_compressed
, stats
->multicast
,
2511 stats
->tx_bytes
, stats
->tx_packets
,
2512 stats
->tx_errors
, stats
->tx_dropped
,
2513 stats
->tx_fifo_errors
, stats
->collisions
,
2514 stats
->tx_carrier_errors
+
2515 stats
->tx_aborted_errors
+
2516 stats
->tx_window_errors
+
2517 stats
->tx_heartbeat_errors
,
2518 stats
->tx_compressed
);
2522 * Called from the PROCfs module. This now uses the new arbitrary sized
2523 * /proc/net interface to create /proc/net/dev
2525 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2527 if (v
== SEQ_START_TOKEN
)
2528 seq_puts(seq
, "Inter-| Receive "
2530 " face |bytes packets errs drop fifo frame "
2531 "compressed multicast|bytes packets errs "
2532 "drop fifo colls carrier compressed\n");
2534 dev_seq_printf_stats(seq
, v
);
2538 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2540 struct netif_rx_stats
*rc
= NULL
;
2542 while (*pos
< nr_cpu_ids
)
2543 if (cpu_online(*pos
)) {
2544 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2551 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2553 return softnet_get_online(pos
);
2556 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2559 return softnet_get_online(pos
);
2562 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2566 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2568 struct netif_rx_stats
*s
= v
;
2570 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2571 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2572 0, 0, 0, 0, /* was fastroute */
2577 static const struct seq_operations dev_seq_ops
= {
2578 .start
= dev_seq_start
,
2579 .next
= dev_seq_next
,
2580 .stop
= dev_seq_stop
,
2581 .show
= dev_seq_show
,
2584 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2586 return seq_open_net(inode
, file
, &dev_seq_ops
,
2587 sizeof(struct seq_net_private
));
2590 static const struct file_operations dev_seq_fops
= {
2591 .owner
= THIS_MODULE
,
2592 .open
= dev_seq_open
,
2594 .llseek
= seq_lseek
,
2595 .release
= seq_release_net
,
2598 static const struct seq_operations softnet_seq_ops
= {
2599 .start
= softnet_seq_start
,
2600 .next
= softnet_seq_next
,
2601 .stop
= softnet_seq_stop
,
2602 .show
= softnet_seq_show
,
2605 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2607 return seq_open(file
, &softnet_seq_ops
);
2610 static const struct file_operations softnet_seq_fops
= {
2611 .owner
= THIS_MODULE
,
2612 .open
= softnet_seq_open
,
2614 .llseek
= seq_lseek
,
2615 .release
= seq_release
,
2618 static void *ptype_get_idx(loff_t pos
)
2620 struct packet_type
*pt
= NULL
;
2624 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
2630 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
2631 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
2640 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2644 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
2647 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2649 struct packet_type
*pt
;
2650 struct list_head
*nxt
;
2654 if (v
== SEQ_START_TOKEN
)
2655 return ptype_get_idx(0);
2658 nxt
= pt
->list
.next
;
2659 if (pt
->type
== htons(ETH_P_ALL
)) {
2660 if (nxt
!= &ptype_all
)
2663 nxt
= ptype_base
[0].next
;
2665 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
2667 while (nxt
== &ptype_base
[hash
]) {
2668 if (++hash
>= PTYPE_HASH_SIZE
)
2670 nxt
= ptype_base
[hash
].next
;
2673 return list_entry(nxt
, struct packet_type
, list
);
2676 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
2682 static void ptype_seq_decode(struct seq_file
*seq
, void *sym
)
2684 #ifdef CONFIG_KALLSYMS
2685 unsigned long offset
= 0, symsize
;
2686 const char *symname
;
2690 symname
= kallsyms_lookup((unsigned long)sym
, &symsize
, &offset
,
2697 modname
= delim
= "";
2698 seq_printf(seq
, "%s%s%s%s+0x%lx", delim
, modname
, delim
,
2704 seq_printf(seq
, "[%p]", sym
);
2707 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
2709 struct packet_type
*pt
= v
;
2711 if (v
== SEQ_START_TOKEN
)
2712 seq_puts(seq
, "Type Device Function\n");
2713 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
2714 if (pt
->type
== htons(ETH_P_ALL
))
2715 seq_puts(seq
, "ALL ");
2717 seq_printf(seq
, "%04x", ntohs(pt
->type
));
2719 seq_printf(seq
, " %-8s ",
2720 pt
->dev
? pt
->dev
->name
: "");
2721 ptype_seq_decode(seq
, pt
->func
);
2722 seq_putc(seq
, '\n');
2728 static const struct seq_operations ptype_seq_ops
= {
2729 .start
= ptype_seq_start
,
2730 .next
= ptype_seq_next
,
2731 .stop
= ptype_seq_stop
,
2732 .show
= ptype_seq_show
,
2735 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
2737 return seq_open_net(inode
, file
, &ptype_seq_ops
,
2738 sizeof(struct seq_net_private
));
2741 static const struct file_operations ptype_seq_fops
= {
2742 .owner
= THIS_MODULE
,
2743 .open
= ptype_seq_open
,
2745 .llseek
= seq_lseek
,
2746 .release
= seq_release_net
,
2750 static int __net_init
dev_proc_net_init(struct net
*net
)
2754 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
2756 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2758 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
2761 if (wext_proc_init(net
))
2767 proc_net_remove(net
, "ptype");
2769 proc_net_remove(net
, "softnet_stat");
2771 proc_net_remove(net
, "dev");
2775 static void __net_exit
dev_proc_net_exit(struct net
*net
)
2777 wext_proc_exit(net
);
2779 proc_net_remove(net
, "ptype");
2780 proc_net_remove(net
, "softnet_stat");
2781 proc_net_remove(net
, "dev");
2784 static struct pernet_operations __net_initdata dev_proc_ops
= {
2785 .init
= dev_proc_net_init
,
2786 .exit
= dev_proc_net_exit
,
2789 static int __init
dev_proc_init(void)
2791 return register_pernet_subsys(&dev_proc_ops
);
2794 #define dev_proc_init() 0
2795 #endif /* CONFIG_PROC_FS */
2799 * netdev_set_master - set up master/slave pair
2800 * @slave: slave device
2801 * @master: new master device
2803 * Changes the master device of the slave. Pass %NULL to break the
2804 * bonding. The caller must hold the RTNL semaphore. On a failure
2805 * a negative errno code is returned. On success the reference counts
2806 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2807 * function returns zero.
2809 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2811 struct net_device
*old
= slave
->master
;
2821 slave
->master
= master
;
2829 slave
->flags
|= IFF_SLAVE
;
2831 slave
->flags
&= ~IFF_SLAVE
;
2833 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2837 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
2839 unsigned short old_flags
= dev
->flags
;
2843 dev
->flags
|= IFF_PROMISC
;
2844 dev
->promiscuity
+= inc
;
2845 if (dev
->promiscuity
== 0) {
2848 * If inc causes overflow, untouch promisc and return error.
2851 dev
->flags
&= ~IFF_PROMISC
;
2853 dev
->promiscuity
-= inc
;
2854 printk(KERN_WARNING
"%s: promiscuity touches roof, "
2855 "set promiscuity failed, promiscuity feature "
2856 "of device might be broken.\n", dev
->name
);
2860 if (dev
->flags
!= old_flags
) {
2861 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2862 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2865 audit_log(current
->audit_context
, GFP_ATOMIC
,
2866 AUDIT_ANOM_PROMISCUOUS
,
2867 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2868 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2869 (old_flags
& IFF_PROMISC
),
2870 audit_get_loginuid(current
),
2871 current
->uid
, current
->gid
,
2872 audit_get_sessionid(current
));
2874 if (dev
->change_rx_flags
)
2875 dev
->change_rx_flags(dev
, IFF_PROMISC
);
2881 * dev_set_promiscuity - update promiscuity count on a device
2885 * Add or remove promiscuity from a device. While the count in the device
2886 * remains above zero the interface remains promiscuous. Once it hits zero
2887 * the device reverts back to normal filtering operation. A negative inc
2888 * value is used to drop promiscuity on the device.
2889 * Return 0 if successful or a negative errno code on error.
2891 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
2893 unsigned short old_flags
= dev
->flags
;
2896 err
= __dev_set_promiscuity(dev
, inc
);
2899 if (dev
->flags
!= old_flags
)
2900 dev_set_rx_mode(dev
);
2905 * dev_set_allmulti - update allmulti count on a device
2909 * Add or remove reception of all multicast frames to a device. While the
2910 * count in the device remains above zero the interface remains listening
2911 * to all interfaces. Once it hits zero the device reverts back to normal
2912 * filtering operation. A negative @inc value is used to drop the counter
2913 * when releasing a resource needing all multicasts.
2914 * Return 0 if successful or a negative errno code on error.
2917 int dev_set_allmulti(struct net_device
*dev
, int inc
)
2919 unsigned short old_flags
= dev
->flags
;
2923 dev
->flags
|= IFF_ALLMULTI
;
2924 dev
->allmulti
+= inc
;
2925 if (dev
->allmulti
== 0) {
2928 * If inc causes overflow, untouch allmulti and return error.
2931 dev
->flags
&= ~IFF_ALLMULTI
;
2933 dev
->allmulti
-= inc
;
2934 printk(KERN_WARNING
"%s: allmulti touches roof, "
2935 "set allmulti failed, allmulti feature of "
2936 "device might be broken.\n", dev
->name
);
2940 if (dev
->flags
^ old_flags
) {
2941 if (dev
->change_rx_flags
)
2942 dev
->change_rx_flags(dev
, IFF_ALLMULTI
);
2943 dev_set_rx_mode(dev
);
2949 * Upload unicast and multicast address lists to device and
2950 * configure RX filtering. When the device doesn't support unicast
2951 * filtering it is put in promiscuous mode while unicast addresses
2954 void __dev_set_rx_mode(struct net_device
*dev
)
2956 /* dev_open will call this function so the list will stay sane. */
2957 if (!(dev
->flags
&IFF_UP
))
2960 if (!netif_device_present(dev
))
2963 if (dev
->set_rx_mode
)
2964 dev
->set_rx_mode(dev
);
2966 /* Unicast addresses changes may only happen under the rtnl,
2967 * therefore calling __dev_set_promiscuity here is safe.
2969 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
2970 __dev_set_promiscuity(dev
, 1);
2971 dev
->uc_promisc
= 1;
2972 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
2973 __dev_set_promiscuity(dev
, -1);
2974 dev
->uc_promisc
= 0;
2977 if (dev
->set_multicast_list
)
2978 dev
->set_multicast_list(dev
);
2982 void dev_set_rx_mode(struct net_device
*dev
)
2984 netif_tx_lock_bh(dev
);
2985 __dev_set_rx_mode(dev
);
2986 netif_tx_unlock_bh(dev
);
2989 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
2990 void *addr
, int alen
, int glbl
)
2992 struct dev_addr_list
*da
;
2994 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
2995 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2996 alen
== da
->da_addrlen
) {
2998 int old_glbl
= da
->da_gusers
;
3015 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
3016 void *addr
, int alen
, int glbl
)
3018 struct dev_addr_list
*da
;
3020 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
3021 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
3022 da
->da_addrlen
== alen
) {
3024 int old_glbl
= da
->da_gusers
;
3034 da
= kzalloc(sizeof(*da
), GFP_ATOMIC
);
3037 memcpy(da
->da_addr
, addr
, alen
);
3038 da
->da_addrlen
= alen
;
3040 da
->da_gusers
= glbl
? 1 : 0;
3048 * dev_unicast_delete - Release secondary unicast address.
3050 * @addr: address to delete
3051 * @alen: length of @addr
3053 * Release reference to a secondary unicast address and remove it
3054 * from the device if the reference count drops to zero.
3056 * The caller must hold the rtnl_mutex.
3058 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
3064 netif_tx_lock_bh(dev
);
3065 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3067 __dev_set_rx_mode(dev
);
3068 netif_tx_unlock_bh(dev
);
3071 EXPORT_SYMBOL(dev_unicast_delete
);
3074 * dev_unicast_add - add a secondary unicast address
3076 * @addr: address to add
3077 * @alen: length of @addr
3079 * Add a secondary unicast address to the device or increase
3080 * the reference count if it already exists.
3082 * The caller must hold the rtnl_mutex.
3084 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
3090 netif_tx_lock_bh(dev
);
3091 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3093 __dev_set_rx_mode(dev
);
3094 netif_tx_unlock_bh(dev
);
3097 EXPORT_SYMBOL(dev_unicast_add
);
3099 int __dev_addr_sync(struct dev_addr_list
**to
, int *to_count
,
3100 struct dev_addr_list
**from
, int *from_count
)
3102 struct dev_addr_list
*da
, *next
;
3106 while (da
!= NULL
) {
3108 if (!da
->da_synced
) {
3109 err
= __dev_addr_add(to
, to_count
,
3110 da
->da_addr
, da
->da_addrlen
, 0);
3115 } else if (da
->da_users
== 1) {
3116 __dev_addr_delete(to
, to_count
,
3117 da
->da_addr
, da
->da_addrlen
, 0);
3118 __dev_addr_delete(from
, from_count
,
3119 da
->da_addr
, da
->da_addrlen
, 0);
3126 void __dev_addr_unsync(struct dev_addr_list
**to
, int *to_count
,
3127 struct dev_addr_list
**from
, int *from_count
)
3129 struct dev_addr_list
*da
, *next
;
3132 while (da
!= NULL
) {
3134 if (da
->da_synced
) {
3135 __dev_addr_delete(to
, to_count
,
3136 da
->da_addr
, da
->da_addrlen
, 0);
3138 __dev_addr_delete(from
, from_count
,
3139 da
->da_addr
, da
->da_addrlen
, 0);
3146 * dev_unicast_sync - Synchronize device's unicast list to another device
3147 * @to: destination device
3148 * @from: source device
3150 * Add newly added addresses to the destination device and release
3151 * addresses that have no users left. The source device must be
3152 * locked by netif_tx_lock_bh.
3154 * This function is intended to be called from the dev->set_rx_mode
3155 * function of layered software devices.
3157 int dev_unicast_sync(struct net_device
*to
, struct net_device
*from
)
3161 netif_tx_lock_bh(to
);
3162 err
= __dev_addr_sync(&to
->uc_list
, &to
->uc_count
,
3163 &from
->uc_list
, &from
->uc_count
);
3165 __dev_set_rx_mode(to
);
3166 netif_tx_unlock_bh(to
);
3169 EXPORT_SYMBOL(dev_unicast_sync
);
3172 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3173 * @to: destination device
3174 * @from: source device
3176 * Remove all addresses that were added to the destination device by
3177 * dev_unicast_sync(). This function is intended to be called from the
3178 * dev->stop function of layered software devices.
3180 void dev_unicast_unsync(struct net_device
*to
, struct net_device
*from
)
3182 netif_tx_lock_bh(from
);
3183 netif_tx_lock_bh(to
);
3185 __dev_addr_unsync(&to
->uc_list
, &to
->uc_count
,
3186 &from
->uc_list
, &from
->uc_count
);
3187 __dev_set_rx_mode(to
);
3189 netif_tx_unlock_bh(to
);
3190 netif_tx_unlock_bh(from
);
3192 EXPORT_SYMBOL(dev_unicast_unsync
);
3194 static void __dev_addr_discard(struct dev_addr_list
**list
)
3196 struct dev_addr_list
*tmp
;
3198 while (*list
!= NULL
) {
3201 if (tmp
->da_users
> tmp
->da_gusers
)
3202 printk("__dev_addr_discard: address leakage! "
3203 "da_users=%d\n", tmp
->da_users
);
3208 static void dev_addr_discard(struct net_device
*dev
)
3210 netif_tx_lock_bh(dev
);
3212 __dev_addr_discard(&dev
->uc_list
);
3215 __dev_addr_discard(&dev
->mc_list
);
3218 netif_tx_unlock_bh(dev
);
3221 unsigned dev_get_flags(const struct net_device
*dev
)
3225 flags
= (dev
->flags
& ~(IFF_PROMISC
|
3230 (dev
->gflags
& (IFF_PROMISC
|
3233 if (netif_running(dev
)) {
3234 if (netif_oper_up(dev
))
3235 flags
|= IFF_RUNNING
;
3236 if (netif_carrier_ok(dev
))
3237 flags
|= IFF_LOWER_UP
;
3238 if (netif_dormant(dev
))
3239 flags
|= IFF_DORMANT
;
3245 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
3248 int old_flags
= dev
->flags
;
3253 * Set the flags on our device.
3256 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
3257 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
3259 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
3263 * Load in the correct multicast list now the flags have changed.
3266 if (dev
->change_rx_flags
&& (old_flags
^ flags
) & IFF_MULTICAST
)
3267 dev
->change_rx_flags(dev
, IFF_MULTICAST
);
3269 dev_set_rx_mode(dev
);
3272 * Have we downed the interface. We handle IFF_UP ourselves
3273 * according to user attempts to set it, rather than blindly
3278 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
3279 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
3282 dev_set_rx_mode(dev
);
3285 if (dev
->flags
& IFF_UP
&&
3286 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
3288 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
3290 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
3291 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
3292 dev
->gflags
^= IFF_PROMISC
;
3293 dev_set_promiscuity(dev
, inc
);
3296 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3297 is important. Some (broken) drivers set IFF_PROMISC, when
3298 IFF_ALLMULTI is requested not asking us and not reporting.
3300 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
3301 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
3302 dev
->gflags
^= IFF_ALLMULTI
;
3303 dev_set_allmulti(dev
, inc
);
3306 /* Exclude state transition flags, already notified */
3307 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
3309 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
3314 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
3318 if (new_mtu
== dev
->mtu
)
3321 /* MTU must be positive. */
3325 if (!netif_device_present(dev
))
3329 if (dev
->change_mtu
)
3330 err
= dev
->change_mtu(dev
, new_mtu
);
3333 if (!err
&& dev
->flags
& IFF_UP
)
3334 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
3338 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
3342 if (!dev
->set_mac_address
)
3344 if (sa
->sa_family
!= dev
->type
)
3346 if (!netif_device_present(dev
))
3348 err
= dev
->set_mac_address(dev
, sa
);
3350 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3355 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3357 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3360 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3366 case SIOCGIFFLAGS
: /* Get interface flags */
3367 ifr
->ifr_flags
= dev_get_flags(dev
);
3370 case SIOCGIFMETRIC
: /* Get the metric on the interface
3371 (currently unused) */
3372 ifr
->ifr_metric
= 0;
3375 case SIOCGIFMTU
: /* Get the MTU of a device */
3376 ifr
->ifr_mtu
= dev
->mtu
;
3381 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
3383 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
3384 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3385 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
3393 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
3394 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
3395 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
3396 ifr
->ifr_map
.irq
= dev
->irq
;
3397 ifr
->ifr_map
.dma
= dev
->dma
;
3398 ifr
->ifr_map
.port
= dev
->if_port
;
3402 ifr
->ifr_ifindex
= dev
->ifindex
;
3406 ifr
->ifr_qlen
= dev
->tx_queue_len
;
3410 /* dev_ioctl() should ensure this case
3422 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3424 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3427 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3433 case SIOCSIFFLAGS
: /* Set interface flags */
3434 return dev_change_flags(dev
, ifr
->ifr_flags
);
3436 case SIOCSIFMETRIC
: /* Set the metric on the interface
3437 (currently unused) */
3440 case SIOCSIFMTU
: /* Set the MTU of a device */
3441 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
3444 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
3446 case SIOCSIFHWBROADCAST
:
3447 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
3449 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
3450 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3451 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3455 if (dev
->set_config
) {
3456 if (!netif_device_present(dev
))
3458 return dev
->set_config(dev
, &ifr
->ifr_map
);
3463 if ((!dev
->set_multicast_list
&& !dev
->set_rx_mode
) ||
3464 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3466 if (!netif_device_present(dev
))
3468 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
3472 if ((!dev
->set_multicast_list
&& !dev
->set_rx_mode
) ||
3473 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3475 if (!netif_device_present(dev
))
3477 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
3481 if (ifr
->ifr_qlen
< 0)
3483 dev
->tx_queue_len
= ifr
->ifr_qlen
;
3487 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
3488 return dev_change_name(dev
, ifr
->ifr_newname
);
3491 * Unknown or private ioctl
3495 if ((cmd
>= SIOCDEVPRIVATE
&&
3496 cmd
<= SIOCDEVPRIVATE
+ 15) ||
3497 cmd
== SIOCBONDENSLAVE
||
3498 cmd
== SIOCBONDRELEASE
||
3499 cmd
== SIOCBONDSETHWADDR
||
3500 cmd
== SIOCBONDSLAVEINFOQUERY
||
3501 cmd
== SIOCBONDINFOQUERY
||
3502 cmd
== SIOCBONDCHANGEACTIVE
||
3503 cmd
== SIOCGMIIPHY
||
3504 cmd
== SIOCGMIIREG
||
3505 cmd
== SIOCSMIIREG
||
3506 cmd
== SIOCBRADDIF
||
3507 cmd
== SIOCBRDELIF
||
3508 cmd
== SIOCWANDEV
) {
3510 if (dev
->do_ioctl
) {
3511 if (netif_device_present(dev
))
3512 err
= dev
->do_ioctl(dev
, ifr
,
3525 * This function handles all "interface"-type I/O control requests. The actual
3526 * 'doing' part of this is dev_ifsioc above.
3530 * dev_ioctl - network device ioctl
3531 * @net: the applicable net namespace
3532 * @cmd: command to issue
3533 * @arg: pointer to a struct ifreq in user space
3535 * Issue ioctl functions to devices. This is normally called by the
3536 * user space syscall interfaces but can sometimes be useful for
3537 * other purposes. The return value is the return from the syscall if
3538 * positive or a negative errno code on error.
3541 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
3547 /* One special case: SIOCGIFCONF takes ifconf argument
3548 and requires shared lock, because it sleeps writing
3552 if (cmd
== SIOCGIFCONF
) {
3554 ret
= dev_ifconf(net
, (char __user
*) arg
);
3558 if (cmd
== SIOCGIFNAME
)
3559 return dev_ifname(net
, (struct ifreq __user
*)arg
);
3561 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3564 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
3566 colon
= strchr(ifr
.ifr_name
, ':');
3571 * See which interface the caller is talking about.
3576 * These ioctl calls:
3577 * - can be done by all.
3578 * - atomic and do not require locking.
3589 dev_load(net
, ifr
.ifr_name
);
3590 read_lock(&dev_base_lock
);
3591 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
3592 read_unlock(&dev_base_lock
);
3596 if (copy_to_user(arg
, &ifr
,
3597 sizeof(struct ifreq
)))
3603 dev_load(net
, ifr
.ifr_name
);
3605 ret
= dev_ethtool(net
, &ifr
);
3610 if (copy_to_user(arg
, &ifr
,
3611 sizeof(struct ifreq
)))
3617 * These ioctl calls:
3618 * - require superuser power.
3619 * - require strict serialization.
3625 if (!capable(CAP_NET_ADMIN
))
3627 dev_load(net
, ifr
.ifr_name
);
3629 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3634 if (copy_to_user(arg
, &ifr
,
3635 sizeof(struct ifreq
)))
3641 * These ioctl calls:
3642 * - require superuser power.
3643 * - require strict serialization.
3644 * - do not return a value
3654 case SIOCSIFHWBROADCAST
:
3657 case SIOCBONDENSLAVE
:
3658 case SIOCBONDRELEASE
:
3659 case SIOCBONDSETHWADDR
:
3660 case SIOCBONDCHANGEACTIVE
:
3663 if (!capable(CAP_NET_ADMIN
))
3666 case SIOCBONDSLAVEINFOQUERY
:
3667 case SIOCBONDINFOQUERY
:
3668 dev_load(net
, ifr
.ifr_name
);
3670 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3675 /* Get the per device memory space. We can add this but
3676 * currently do not support it */
3678 /* Set the per device memory buffer space.
3679 * Not applicable in our case */
3684 * Unknown or private ioctl.
3687 if (cmd
== SIOCWANDEV
||
3688 (cmd
>= SIOCDEVPRIVATE
&&
3689 cmd
<= SIOCDEVPRIVATE
+ 15)) {
3690 dev_load(net
, ifr
.ifr_name
);
3692 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3694 if (!ret
&& copy_to_user(arg
, &ifr
,
3695 sizeof(struct ifreq
)))
3699 /* Take care of Wireless Extensions */
3700 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
3701 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
3708 * dev_new_index - allocate an ifindex
3709 * @net: the applicable net namespace
3711 * Returns a suitable unique value for a new device interface
3712 * number. The caller must hold the rtnl semaphore or the
3713 * dev_base_lock to be sure it remains unique.
3715 static int dev_new_index(struct net
*net
)
3721 if (!__dev_get_by_index(net
, ifindex
))
3726 /* Delayed registration/unregisteration */
3727 static DEFINE_SPINLOCK(net_todo_list_lock
);
3728 static LIST_HEAD(net_todo_list
);
3730 static void net_set_todo(struct net_device
*dev
)
3732 spin_lock(&net_todo_list_lock
);
3733 list_add_tail(&dev
->todo_list
, &net_todo_list
);
3734 spin_unlock(&net_todo_list_lock
);
3737 static void rollback_registered(struct net_device
*dev
)
3739 BUG_ON(dev_boot_phase
);
3742 /* Some devices call without registering for initialization unwind. */
3743 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3744 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3745 "was registered\n", dev
->name
, dev
);
3751 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3753 /* If device is running, close it first. */
3756 /* And unlink it from device chain. */
3757 unlist_netdevice(dev
);
3759 dev
->reg_state
= NETREG_UNREGISTERING
;
3763 /* Shutdown queueing discipline. */
3767 /* Notify protocols, that we are about to destroy
3768 this device. They should clean all the things.
3770 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
3773 * Flush the unicast and multicast chains
3775 dev_addr_discard(dev
);
3780 /* Notifier chain MUST detach us from master device. */
3781 BUG_TRAP(!dev
->master
);
3783 /* Remove entries from kobject tree */
3784 netdev_unregister_kobject(dev
);
3791 static void __netdev_init_queue_locks_one(struct netdev_queue
*dev_queue
,
3792 struct net_device
*dev
)
3794 spin_lock_init(&dev_queue
->_xmit_lock
);
3795 netdev_set_lockdep_class(&dev_queue
->_xmit_lock
, dev
->type
);
3796 dev_queue
->xmit_lock_owner
= -1;
3799 static void netdev_init_queue_locks(struct net_device
*dev
)
3801 __netdev_init_queue_locks_one(&dev
->tx_queue
, dev
);
3802 __netdev_init_queue_locks_one(&dev
->rx_queue
, dev
);
3806 * register_netdevice - register a network device
3807 * @dev: device to register
3809 * Take a completed network device structure and add it to the kernel
3810 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3811 * chain. 0 is returned on success. A negative errno code is returned
3812 * on a failure to set up the device, or if the name is a duplicate.
3814 * Callers must hold the rtnl semaphore. You may want
3815 * register_netdev() instead of this.
3818 * The locking appears insufficient to guarantee two parallel registers
3819 * will not get the same name.
3822 int register_netdevice(struct net_device
*dev
)
3824 struct hlist_head
*head
;
3825 struct hlist_node
*p
;
3829 BUG_ON(dev_boot_phase
);
3834 /* When net_device's are persistent, this will be fatal. */
3835 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
3836 BUG_ON(!dev_net(dev
));
3839 spin_lock_init(&dev
->addr_list_lock
);
3840 netdev_init_queue_locks(dev
);
3844 /* Init, if this function is available */
3846 ret
= dev
->init(dev
);
3854 if (!dev_valid_name(dev
->name
)) {
3859 dev
->ifindex
= dev_new_index(net
);
3860 if (dev
->iflink
== -1)
3861 dev
->iflink
= dev
->ifindex
;
3863 /* Check for existence of name */
3864 head
= dev_name_hash(net
, dev
->name
);
3865 hlist_for_each(p
, head
) {
3866 struct net_device
*d
3867 = hlist_entry(p
, struct net_device
, name_hlist
);
3868 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
3874 /* Fix illegal checksum combinations */
3875 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
3876 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3877 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
3879 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
3882 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
3883 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3884 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
3886 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
3890 /* Fix illegal SG+CSUM combinations. */
3891 if ((dev
->features
& NETIF_F_SG
) &&
3892 !(dev
->features
& NETIF_F_ALL_CSUM
)) {
3893 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no checksum feature.\n",
3895 dev
->features
&= ~NETIF_F_SG
;
3898 /* TSO requires that SG is present as well. */
3899 if ((dev
->features
& NETIF_F_TSO
) &&
3900 !(dev
->features
& NETIF_F_SG
)) {
3901 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no SG feature.\n",
3903 dev
->features
&= ~NETIF_F_TSO
;
3905 if (dev
->features
& NETIF_F_UFO
) {
3906 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
3907 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3908 "NETIF_F_HW_CSUM feature.\n",
3910 dev
->features
&= ~NETIF_F_UFO
;
3912 if (!(dev
->features
& NETIF_F_SG
)) {
3913 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3914 "NETIF_F_SG feature.\n",
3916 dev
->features
&= ~NETIF_F_UFO
;
3920 netdev_initialize_kobject(dev
);
3921 ret
= netdev_register_kobject(dev
);
3924 dev
->reg_state
= NETREG_REGISTERED
;
3927 * Default initial state at registry is that the
3928 * device is present.
3931 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
3933 dev_init_scheduler(dev
);
3935 list_netdevice(dev
);
3937 /* Notify protocols, that a new device appeared. */
3938 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
3939 ret
= notifier_to_errno(ret
);
3941 rollback_registered(dev
);
3942 dev
->reg_state
= NETREG_UNREGISTERED
;
3955 * register_netdev - register a network device
3956 * @dev: device to register
3958 * Take a completed network device structure and add it to the kernel
3959 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3960 * chain. 0 is returned on success. A negative errno code is returned
3961 * on a failure to set up the device, or if the name is a duplicate.
3963 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3964 * and expands the device name if you passed a format string to
3967 int register_netdev(struct net_device
*dev
)
3974 * If the name is a format string the caller wants us to do a
3977 if (strchr(dev
->name
, '%')) {
3978 err
= dev_alloc_name(dev
, dev
->name
);
3983 err
= register_netdevice(dev
);
3988 EXPORT_SYMBOL(register_netdev
);
3991 * netdev_wait_allrefs - wait until all references are gone.
3993 * This is called when unregistering network devices.
3995 * Any protocol or device that holds a reference should register
3996 * for netdevice notification, and cleanup and put back the
3997 * reference if they receive an UNREGISTER event.
3998 * We can get stuck here if buggy protocols don't correctly
4001 static void netdev_wait_allrefs(struct net_device
*dev
)
4003 unsigned long rebroadcast_time
, warning_time
;
4005 rebroadcast_time
= warning_time
= jiffies
;
4006 while (atomic_read(&dev
->refcnt
) != 0) {
4007 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
4010 /* Rebroadcast unregister notification */
4011 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4013 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
4015 /* We must not have linkwatch events
4016 * pending on unregister. If this
4017 * happens, we simply run the queue
4018 * unscheduled, resulting in a noop
4021 linkwatch_run_queue();
4026 rebroadcast_time
= jiffies
;
4031 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
4032 printk(KERN_EMERG
"unregister_netdevice: "
4033 "waiting for %s to become free. Usage "
4035 dev
->name
, atomic_read(&dev
->refcnt
));
4036 warning_time
= jiffies
;
4045 * register_netdevice(x1);
4046 * register_netdevice(x2);
4048 * unregister_netdevice(y1);
4049 * unregister_netdevice(y2);
4055 * We are invoked by rtnl_unlock() after it drops the semaphore.
4056 * This allows us to deal with problems:
4057 * 1) We can delete sysfs objects which invoke hotplug
4058 * without deadlocking with linkwatch via keventd.
4059 * 2) Since we run with the RTNL semaphore not held, we can sleep
4060 * safely in order to wait for the netdev refcnt to drop to zero.
4062 static DEFINE_MUTEX(net_todo_run_mutex
);
4063 void netdev_run_todo(void)
4065 struct list_head list
;
4067 /* Need to guard against multiple cpu's getting out of order. */
4068 mutex_lock(&net_todo_run_mutex
);
4070 /* Not safe to do outside the semaphore. We must not return
4071 * until all unregister events invoked by the local processor
4072 * have been completed (either by this todo run, or one on
4075 if (list_empty(&net_todo_list
))
4078 /* Snapshot list, allow later requests */
4079 spin_lock(&net_todo_list_lock
);
4080 list_replace_init(&net_todo_list
, &list
);
4081 spin_unlock(&net_todo_list_lock
);
4083 while (!list_empty(&list
)) {
4084 struct net_device
*dev
4085 = list_entry(list
.next
, struct net_device
, todo_list
);
4086 list_del(&dev
->todo_list
);
4088 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
4089 printk(KERN_ERR
"network todo '%s' but state %d\n",
4090 dev
->name
, dev
->reg_state
);
4095 dev
->reg_state
= NETREG_UNREGISTERED
;
4097 netdev_wait_allrefs(dev
);
4100 BUG_ON(atomic_read(&dev
->refcnt
));
4101 BUG_TRAP(!dev
->ip_ptr
);
4102 BUG_TRAP(!dev
->ip6_ptr
);
4103 BUG_TRAP(!dev
->dn_ptr
);
4105 if (dev
->destructor
)
4106 dev
->destructor(dev
);
4108 /* Free network device */
4109 kobject_put(&dev
->dev
.kobj
);
4113 mutex_unlock(&net_todo_run_mutex
);
4116 static struct net_device_stats
*internal_stats(struct net_device
*dev
)
4121 static void netdev_init_one_queue(struct net_device
*dev
,
4122 struct netdev_queue
*queue
)
4124 spin_lock_init(&queue
->lock
);
4128 static void netdev_init_queues(struct net_device
*dev
)
4130 netdev_init_one_queue(dev
, &dev
->rx_queue
);
4131 netdev_init_one_queue(dev
, &dev
->tx_queue
);
4135 * alloc_netdev_mq - allocate network device
4136 * @sizeof_priv: size of private data to allocate space for
4137 * @name: device name format string
4138 * @setup: callback to initialize device
4139 * @queue_count: the number of subqueues to allocate
4141 * Allocates a struct net_device with private data area for driver use
4142 * and performs basic initialization. Also allocates subquue structs
4143 * for each queue on the device at the end of the netdevice.
4145 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
4146 void (*setup
)(struct net_device
*), unsigned int queue_count
)
4149 struct net_device
*dev
;
4152 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
4154 alloc_size
= sizeof(struct net_device
) +
4155 sizeof(struct net_device_subqueue
) * (queue_count
- 1);
4157 /* ensure 32-byte alignment of private area */
4158 alloc_size
= (alloc_size
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
4159 alloc_size
+= sizeof_priv
;
4161 /* ensure 32-byte alignment of whole construct */
4162 alloc_size
+= NETDEV_ALIGN_CONST
;
4164 p
= kzalloc(alloc_size
, GFP_KERNEL
);
4166 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
4170 dev
= (struct net_device
*)
4171 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
4172 dev
->padded
= (char *)dev
- (char *)p
;
4173 dev_net_set(dev
, &init_net
);
4176 dev
->priv
= ((char *)dev
+
4177 ((sizeof(struct net_device
) +
4178 (sizeof(struct net_device_subqueue
) *
4179 (queue_count
- 1)) + NETDEV_ALIGN_CONST
)
4180 & ~NETDEV_ALIGN_CONST
));
4183 dev
->egress_subqueue_count
= queue_count
;
4184 dev
->gso_max_size
= GSO_MAX_SIZE
;
4186 netdev_init_queues(dev
);
4188 dev
->get_stats
= internal_stats
;
4189 netpoll_netdev_init(dev
);
4191 strcpy(dev
->name
, name
);
4194 EXPORT_SYMBOL(alloc_netdev_mq
);
4197 * free_netdev - free network device
4200 * This function does the last stage of destroying an allocated device
4201 * interface. The reference to the device object is released.
4202 * If this is the last reference then it will be freed.
4204 void free_netdev(struct net_device
*dev
)
4206 release_net(dev_net(dev
));
4208 /* Compatibility with error handling in drivers */
4209 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4210 kfree((char *)dev
- dev
->padded
);
4214 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
4215 dev
->reg_state
= NETREG_RELEASED
;
4217 /* will free via device release */
4218 put_device(&dev
->dev
);
4221 /* Synchronize with packet receive processing. */
4222 void synchronize_net(void)
4229 * unregister_netdevice - remove device from the kernel
4232 * This function shuts down a device interface and removes it
4233 * from the kernel tables.
4235 * Callers must hold the rtnl semaphore. You may want
4236 * unregister_netdev() instead of this.
4239 void unregister_netdevice(struct net_device
*dev
)
4243 rollback_registered(dev
);
4244 /* Finish processing unregister after unlock */
4249 * unregister_netdev - remove device from the kernel
4252 * This function shuts down a device interface and removes it
4253 * from the kernel tables.
4255 * This is just a wrapper for unregister_netdevice that takes
4256 * the rtnl semaphore. In general you want to use this and not
4257 * unregister_netdevice.
4259 void unregister_netdev(struct net_device
*dev
)
4262 unregister_netdevice(dev
);
4266 EXPORT_SYMBOL(unregister_netdev
);
4269 * dev_change_net_namespace - move device to different nethost namespace
4271 * @net: network namespace
4272 * @pat: If not NULL name pattern to try if the current device name
4273 * is already taken in the destination network namespace.
4275 * This function shuts down a device interface and moves it
4276 * to a new network namespace. On success 0 is returned, on
4277 * a failure a netagive errno code is returned.
4279 * Callers must hold the rtnl semaphore.
4282 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
4285 const char *destname
;
4290 /* Don't allow namespace local devices to be moved. */
4292 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
4295 /* Ensure the device has been registrered */
4297 if (dev
->reg_state
!= NETREG_REGISTERED
)
4300 /* Get out if there is nothing todo */
4302 if (net_eq(dev_net(dev
), net
))
4305 /* Pick the destination device name, and ensure
4306 * we can use it in the destination network namespace.
4309 destname
= dev
->name
;
4310 if (__dev_get_by_name(net
, destname
)) {
4311 /* We get here if we can't use the current device name */
4314 if (!dev_valid_name(pat
))
4316 if (strchr(pat
, '%')) {
4317 if (__dev_alloc_name(net
, pat
, buf
) < 0)
4322 if (__dev_get_by_name(net
, destname
))
4327 * And now a mini version of register_netdevice unregister_netdevice.
4330 /* If device is running close it first. */
4333 /* And unlink it from device chain */
4335 unlist_netdevice(dev
);
4339 /* Shutdown queueing discipline. */
4342 /* Notify protocols, that we are about to destroy
4343 this device. They should clean all the things.
4345 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4348 * Flush the unicast and multicast chains
4350 dev_addr_discard(dev
);
4352 /* Actually switch the network namespace */
4353 dev_net_set(dev
, net
);
4355 /* Assign the new device name */
4356 if (destname
!= dev
->name
)
4357 strcpy(dev
->name
, destname
);
4359 /* If there is an ifindex conflict assign a new one */
4360 if (__dev_get_by_index(net
, dev
->ifindex
)) {
4361 int iflink
= (dev
->iflink
== dev
->ifindex
);
4362 dev
->ifindex
= dev_new_index(net
);
4364 dev
->iflink
= dev
->ifindex
;
4367 /* Fixup kobjects */
4368 netdev_unregister_kobject(dev
);
4369 err
= netdev_register_kobject(dev
);
4372 /* Add the device back in the hashes */
4373 list_netdevice(dev
);
4375 /* Notify protocols, that a new device appeared. */
4376 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
4384 static int dev_cpu_callback(struct notifier_block
*nfb
,
4385 unsigned long action
,
4388 struct sk_buff
**list_skb
;
4389 struct netdev_queue
**list_net
;
4390 struct sk_buff
*skb
;
4391 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
4392 struct softnet_data
*sd
, *oldsd
;
4394 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
4397 local_irq_disable();
4398 cpu
= smp_processor_id();
4399 sd
= &per_cpu(softnet_data
, cpu
);
4400 oldsd
= &per_cpu(softnet_data
, oldcpu
);
4402 /* Find end of our completion_queue. */
4403 list_skb
= &sd
->completion_queue
;
4405 list_skb
= &(*list_skb
)->next
;
4406 /* Append completion queue from offline CPU. */
4407 *list_skb
= oldsd
->completion_queue
;
4408 oldsd
->completion_queue
= NULL
;
4410 /* Find end of our output_queue. */
4411 list_net
= &sd
->output_queue
;
4413 list_net
= &(*list_net
)->next_sched
;
4414 /* Append output queue from offline CPU. */
4415 *list_net
= oldsd
->output_queue
;
4416 oldsd
->output_queue
= NULL
;
4418 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
4421 /* Process offline CPU's input_pkt_queue */
4422 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
4428 #ifdef CONFIG_NET_DMA
4430 * net_dma_rebalance - try to maintain one DMA channel per CPU
4431 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4433 * This is called when the number of channels allocated to the net_dma client
4434 * changes. The net_dma client tries to have one DMA channel per CPU.
4437 static void net_dma_rebalance(struct net_dma
*net_dma
)
4439 unsigned int cpu
, i
, n
, chan_idx
;
4440 struct dma_chan
*chan
;
4442 if (cpus_empty(net_dma
->channel_mask
)) {
4443 for_each_online_cpu(cpu
)
4444 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
4449 cpu
= first_cpu(cpu_online_map
);
4451 for_each_cpu_mask(chan_idx
, net_dma
->channel_mask
) {
4452 chan
= net_dma
->channels
[chan_idx
];
4454 n
= ((num_online_cpus() / cpus_weight(net_dma
->channel_mask
))
4455 + (i
< (num_online_cpus() %
4456 cpus_weight(net_dma
->channel_mask
)) ? 1 : 0));
4459 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
4460 cpu
= next_cpu(cpu
, cpu_online_map
);
4468 * netdev_dma_event - event callback for the net_dma_client
4469 * @client: should always be net_dma_client
4470 * @chan: DMA channel for the event
4471 * @state: DMA state to be handled
4473 static enum dma_state_client
4474 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
4475 enum dma_state state
)
4477 int i
, found
= 0, pos
= -1;
4478 struct net_dma
*net_dma
=
4479 container_of(client
, struct net_dma
, client
);
4480 enum dma_state_client ack
= DMA_DUP
; /* default: take no action */
4482 spin_lock(&net_dma
->lock
);
4484 case DMA_RESOURCE_AVAILABLE
:
4485 for (i
= 0; i
< nr_cpu_ids
; i
++)
4486 if (net_dma
->channels
[i
] == chan
) {
4489 } else if (net_dma
->channels
[i
] == NULL
&& pos
< 0)
4492 if (!found
&& pos
>= 0) {
4494 net_dma
->channels
[pos
] = chan
;
4495 cpu_set(pos
, net_dma
->channel_mask
);
4496 net_dma_rebalance(net_dma
);
4499 case DMA_RESOURCE_REMOVED
:
4500 for (i
= 0; i
< nr_cpu_ids
; i
++)
4501 if (net_dma
->channels
[i
] == chan
) {
4509 cpu_clear(pos
, net_dma
->channel_mask
);
4510 net_dma
->channels
[i
] = NULL
;
4511 net_dma_rebalance(net_dma
);
4517 spin_unlock(&net_dma
->lock
);
4523 * netdev_dma_regiser - register the networking subsystem as a DMA client
4525 static int __init
netdev_dma_register(void)
4527 net_dma
.channels
= kzalloc(nr_cpu_ids
* sizeof(struct net_dma
),
4529 if (unlikely(!net_dma
.channels
)) {
4531 "netdev_dma: no memory for net_dma.channels\n");
4534 spin_lock_init(&net_dma
.lock
);
4535 dma_cap_set(DMA_MEMCPY
, net_dma
.client
.cap_mask
);
4536 dma_async_client_register(&net_dma
.client
);
4537 dma_async_client_chan_request(&net_dma
.client
);
4542 static int __init
netdev_dma_register(void) { return -ENODEV
; }
4543 #endif /* CONFIG_NET_DMA */
4546 * netdev_compute_feature - compute conjunction of two feature sets
4547 * @all: first feature set
4548 * @one: second feature set
4550 * Computes a new feature set after adding a device with feature set
4551 * @one to the master device with current feature set @all. Returns
4552 * the new feature set.
4554 int netdev_compute_features(unsigned long all
, unsigned long one
)
4556 /* if device needs checksumming, downgrade to hw checksumming */
4557 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
4558 all
^= NETIF_F_NO_CSUM
| NETIF_F_HW_CSUM
;
4560 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4561 if (all
& NETIF_F_HW_CSUM
&& !(one
& NETIF_F_HW_CSUM
))
4562 all
^= NETIF_F_HW_CSUM
4563 | NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
;
4565 if (one
& NETIF_F_GSO
)
4566 one
|= NETIF_F_GSO_SOFTWARE
;
4569 /* If even one device supports robust GSO, enable it for all. */
4570 if (one
& NETIF_F_GSO_ROBUST
)
4571 all
|= NETIF_F_GSO_ROBUST
;
4573 all
&= one
| NETIF_F_LLTX
;
4575 if (!(all
& NETIF_F_ALL_CSUM
))
4577 if (!(all
& NETIF_F_SG
))
4578 all
&= ~NETIF_F_GSO_MASK
;
4582 EXPORT_SYMBOL(netdev_compute_features
);
4584 static struct hlist_head
*netdev_create_hash(void)
4587 struct hlist_head
*hash
;
4589 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
4591 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
4592 INIT_HLIST_HEAD(&hash
[i
]);
4597 /* Initialize per network namespace state */
4598 static int __net_init
netdev_init(struct net
*net
)
4600 INIT_LIST_HEAD(&net
->dev_base_head
);
4602 net
->dev_name_head
= netdev_create_hash();
4603 if (net
->dev_name_head
== NULL
)
4606 net
->dev_index_head
= netdev_create_hash();
4607 if (net
->dev_index_head
== NULL
)
4613 kfree(net
->dev_name_head
);
4618 static void __net_exit
netdev_exit(struct net
*net
)
4620 kfree(net
->dev_name_head
);
4621 kfree(net
->dev_index_head
);
4624 static struct pernet_operations __net_initdata netdev_net_ops
= {
4625 .init
= netdev_init
,
4626 .exit
= netdev_exit
,
4629 static void __net_exit
default_device_exit(struct net
*net
)
4631 struct net_device
*dev
, *next
;
4633 * Push all migratable of the network devices back to the
4634 * initial network namespace
4637 for_each_netdev_safe(net
, dev
, next
) {
4639 char fb_name
[IFNAMSIZ
];
4641 /* Ignore unmoveable devices (i.e. loopback) */
4642 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
4645 /* Push remaing network devices to init_net */
4646 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
4647 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
4649 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
4650 __func__
, dev
->name
, err
);
4657 static struct pernet_operations __net_initdata default_device_ops
= {
4658 .exit
= default_device_exit
,
4662 * Initialize the DEV module. At boot time this walks the device list and
4663 * unhooks any devices that fail to initialise (normally hardware not
4664 * present) and leaves us with a valid list of present and active devices.
4669 * This is called single threaded during boot, so no need
4670 * to take the rtnl semaphore.
4672 static int __init
net_dev_init(void)
4674 int i
, rc
= -ENOMEM
;
4676 BUG_ON(!dev_boot_phase
);
4678 if (dev_proc_init())
4681 if (netdev_kobject_init())
4684 INIT_LIST_HEAD(&ptype_all
);
4685 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
4686 INIT_LIST_HEAD(&ptype_base
[i
]);
4688 if (register_pernet_subsys(&netdev_net_ops
))
4691 if (register_pernet_device(&default_device_ops
))
4695 * Initialise the packet receive queues.
4698 for_each_possible_cpu(i
) {
4699 struct softnet_data
*queue
;
4701 queue
= &per_cpu(softnet_data
, i
);
4702 skb_queue_head_init(&queue
->input_pkt_queue
);
4703 queue
->completion_queue
= NULL
;
4704 INIT_LIST_HEAD(&queue
->poll_list
);
4706 queue
->backlog
.poll
= process_backlog
;
4707 queue
->backlog
.weight
= weight_p
;
4710 netdev_dma_register();
4714 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
4715 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
4717 hotcpu_notifier(dev_cpu_callback
, 0);
4725 subsys_initcall(net_dev_init
);
4727 EXPORT_SYMBOL(__dev_get_by_index
);
4728 EXPORT_SYMBOL(__dev_get_by_name
);
4729 EXPORT_SYMBOL(__dev_remove_pack
);
4730 EXPORT_SYMBOL(dev_valid_name
);
4731 EXPORT_SYMBOL(dev_add_pack
);
4732 EXPORT_SYMBOL(dev_alloc_name
);
4733 EXPORT_SYMBOL(dev_close
);
4734 EXPORT_SYMBOL(dev_get_by_flags
);
4735 EXPORT_SYMBOL(dev_get_by_index
);
4736 EXPORT_SYMBOL(dev_get_by_name
);
4737 EXPORT_SYMBOL(dev_open
);
4738 EXPORT_SYMBOL(dev_queue_xmit
);
4739 EXPORT_SYMBOL(dev_remove_pack
);
4740 EXPORT_SYMBOL(dev_set_allmulti
);
4741 EXPORT_SYMBOL(dev_set_promiscuity
);
4742 EXPORT_SYMBOL(dev_change_flags
);
4743 EXPORT_SYMBOL(dev_set_mtu
);
4744 EXPORT_SYMBOL(dev_set_mac_address
);
4745 EXPORT_SYMBOL(free_netdev
);
4746 EXPORT_SYMBOL(netdev_boot_setup_check
);
4747 EXPORT_SYMBOL(netdev_set_master
);
4748 EXPORT_SYMBOL(netdev_state_change
);
4749 EXPORT_SYMBOL(netif_receive_skb
);
4750 EXPORT_SYMBOL(netif_rx
);
4751 EXPORT_SYMBOL(register_gifconf
);
4752 EXPORT_SYMBOL(register_netdevice
);
4753 EXPORT_SYMBOL(register_netdevice_notifier
);
4754 EXPORT_SYMBOL(skb_checksum_help
);
4755 EXPORT_SYMBOL(synchronize_net
);
4756 EXPORT_SYMBOL(unregister_netdevice
);
4757 EXPORT_SYMBOL(unregister_netdevice_notifier
);
4758 EXPORT_SYMBOL(net_enable_timestamp
);
4759 EXPORT_SYMBOL(net_disable_timestamp
);
4760 EXPORT_SYMBOL(dev_get_flags
);
4762 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4763 EXPORT_SYMBOL(br_handle_frame_hook
);
4764 EXPORT_SYMBOL(br_fdb_get_hook
);
4765 EXPORT_SYMBOL(br_fdb_put_hook
);
4769 EXPORT_SYMBOL(dev_load
);
4772 EXPORT_PER_CPU_SYMBOL(softnet_data
);