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 dev->_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 net_device
*dev
)
1325 if (!test_and_set_bit(__LINK_STATE_SCHED
, &dev
->state
)) {
1326 unsigned long flags
;
1327 struct softnet_data
*sd
;
1329 local_irq_save(flags
);
1330 sd
= &__get_cpu_var(softnet_data
);
1331 dev
->next_sched
= sd
->output_queue
;
1332 sd
->output_queue
= dev
;
1333 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1334 local_irq_restore(flags
);
1337 EXPORT_SYMBOL(__netif_schedule
);
1339 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1341 if (atomic_dec_and_test(&skb
->users
)) {
1342 struct softnet_data
*sd
;
1343 unsigned long flags
;
1345 local_irq_save(flags
);
1346 sd
= &__get_cpu_var(softnet_data
);
1347 skb
->next
= sd
->completion_queue
;
1348 sd
->completion_queue
= skb
;
1349 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1350 local_irq_restore(flags
);
1353 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1355 void dev_kfree_skb_any(struct sk_buff
*skb
)
1357 if (in_irq() || irqs_disabled())
1358 dev_kfree_skb_irq(skb
);
1362 EXPORT_SYMBOL(dev_kfree_skb_any
);
1366 * netif_device_detach - mark device as removed
1367 * @dev: network device
1369 * Mark device as removed from system and therefore no longer available.
1371 void netif_device_detach(struct net_device
*dev
)
1373 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1374 netif_running(dev
)) {
1375 netif_stop_queue(dev
);
1378 EXPORT_SYMBOL(netif_device_detach
);
1381 * netif_device_attach - mark device as attached
1382 * @dev: network device
1384 * Mark device as attached from system and restart if needed.
1386 void netif_device_attach(struct net_device
*dev
)
1388 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1389 netif_running(dev
)) {
1390 netif_wake_queue(dev
);
1391 __netdev_watchdog_up(dev
);
1394 EXPORT_SYMBOL(netif_device_attach
);
1396 static bool can_checksum_protocol(unsigned long features
, __be16 protocol
)
1398 return ((features
& NETIF_F_GEN_CSUM
) ||
1399 ((features
& NETIF_F_IP_CSUM
) &&
1400 protocol
== htons(ETH_P_IP
)) ||
1401 ((features
& NETIF_F_IPV6_CSUM
) &&
1402 protocol
== htons(ETH_P_IPV6
)));
1405 static bool dev_can_checksum(struct net_device
*dev
, struct sk_buff
*skb
)
1407 if (can_checksum_protocol(dev
->features
, skb
->protocol
))
1410 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1411 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
1412 if (can_checksum_protocol(dev
->features
& dev
->vlan_features
,
1413 veh
->h_vlan_encapsulated_proto
))
1421 * Invalidate hardware checksum when packet is to be mangled, and
1422 * complete checksum manually on outgoing path.
1424 int skb_checksum_help(struct sk_buff
*skb
)
1427 int ret
= 0, offset
;
1429 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1430 goto out_set_summed
;
1432 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1433 /* Let GSO fix up the checksum. */
1434 goto out_set_summed
;
1437 offset
= skb
->csum_start
- skb_headroom(skb
);
1438 BUG_ON(offset
>= skb_headlen(skb
));
1439 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1441 offset
+= skb
->csum_offset
;
1442 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1444 if (skb_cloned(skb
) &&
1445 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1446 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1451 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1453 skb
->ip_summed
= CHECKSUM_NONE
;
1459 * skb_gso_segment - Perform segmentation on skb.
1460 * @skb: buffer to segment
1461 * @features: features for the output path (see dev->features)
1463 * This function segments the given skb and returns a list of segments.
1465 * It may return NULL if the skb requires no segmentation. This is
1466 * only possible when GSO is used for verifying header integrity.
1468 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1470 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1471 struct packet_type
*ptype
;
1472 __be16 type
= skb
->protocol
;
1475 BUG_ON(skb_shinfo(skb
)->frag_list
);
1477 skb_reset_mac_header(skb
);
1478 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1479 __skb_pull(skb
, skb
->mac_len
);
1481 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1482 if (skb_header_cloned(skb
) &&
1483 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1484 return ERR_PTR(err
);
1488 list_for_each_entry_rcu(ptype
,
1489 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1490 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1491 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1492 err
= ptype
->gso_send_check(skb
);
1493 segs
= ERR_PTR(err
);
1494 if (err
|| skb_gso_ok(skb
, features
))
1496 __skb_push(skb
, (skb
->data
-
1497 skb_network_header(skb
)));
1499 segs
= ptype
->gso_segment(skb
, features
);
1505 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1510 EXPORT_SYMBOL(skb_gso_segment
);
1512 /* Take action when hardware reception checksum errors are detected. */
1514 void netdev_rx_csum_fault(struct net_device
*dev
)
1516 if (net_ratelimit()) {
1517 printk(KERN_ERR
"%s: hw csum failure.\n",
1518 dev
? dev
->name
: "<unknown>");
1522 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1525 /* Actually, we should eliminate this check as soon as we know, that:
1526 * 1. IOMMU is present and allows to map all the memory.
1527 * 2. No high memory really exists on this machine.
1530 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1532 #ifdef CONFIG_HIGHMEM
1535 if (dev
->features
& NETIF_F_HIGHDMA
)
1538 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1539 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1547 void (*destructor
)(struct sk_buff
*skb
);
1550 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1552 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1554 struct dev_gso_cb
*cb
;
1557 struct sk_buff
*nskb
= skb
->next
;
1559 skb
->next
= nskb
->next
;
1562 } while (skb
->next
);
1564 cb
= DEV_GSO_CB(skb
);
1566 cb
->destructor(skb
);
1570 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1571 * @skb: buffer to segment
1573 * This function segments the given skb and stores the list of segments
1576 static int dev_gso_segment(struct sk_buff
*skb
)
1578 struct net_device
*dev
= skb
->dev
;
1579 struct sk_buff
*segs
;
1580 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1583 segs
= skb_gso_segment(skb
, features
);
1585 /* Verifying header integrity only. */
1590 return PTR_ERR(segs
);
1593 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1594 skb
->destructor
= dev_gso_skb_destructor
;
1599 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1601 if (likely(!skb
->next
)) {
1602 if (!list_empty(&ptype_all
))
1603 dev_queue_xmit_nit(skb
, dev
);
1605 if (netif_needs_gso(dev
, skb
)) {
1606 if (unlikely(dev_gso_segment(skb
)))
1612 return dev
->hard_start_xmit(skb
, dev
);
1617 struct sk_buff
*nskb
= skb
->next
;
1620 skb
->next
= nskb
->next
;
1622 rc
= dev
->hard_start_xmit(nskb
, dev
);
1624 nskb
->next
= skb
->next
;
1628 if (unlikely((netif_queue_stopped(dev
) ||
1629 netif_subqueue_stopped(dev
, skb
)) &&
1631 return NETDEV_TX_BUSY
;
1632 } while (skb
->next
);
1634 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1642 * dev_queue_xmit - transmit a buffer
1643 * @skb: buffer to transmit
1645 * Queue a buffer for transmission to a network device. The caller must
1646 * have set the device and priority and built the buffer before calling
1647 * this function. The function can be called from an interrupt.
1649 * A negative errno code is returned on a failure. A success does not
1650 * guarantee the frame will be transmitted as it may be dropped due
1651 * to congestion or traffic shaping.
1653 * -----------------------------------------------------------------------------------
1654 * I notice this method can also return errors from the queue disciplines,
1655 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1658 * Regardless of the return value, the skb is consumed, so it is currently
1659 * difficult to retry a send to this method. (You can bump the ref count
1660 * before sending to hold a reference for retry if you are careful.)
1662 * When calling this method, interrupts MUST be enabled. This is because
1663 * the BH enable code must have IRQs enabled so that it will not deadlock.
1667 int dev_queue_xmit(struct sk_buff
*skb
)
1669 struct net_device
*dev
= skb
->dev
;
1670 struct netdev_queue
*txq
;
1674 /* GSO will handle the following emulations directly. */
1675 if (netif_needs_gso(dev
, skb
))
1678 if (skb_shinfo(skb
)->frag_list
&&
1679 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1680 __skb_linearize(skb
))
1683 /* Fragmented skb is linearized if device does not support SG,
1684 * or if at least one of fragments is in highmem and device
1685 * does not support DMA from it.
1687 if (skb_shinfo(skb
)->nr_frags
&&
1688 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1689 __skb_linearize(skb
))
1692 /* If packet is not checksummed and device does not support
1693 * checksumming for this protocol, complete checksumming here.
1695 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1696 skb_set_transport_header(skb
, skb
->csum_start
-
1698 if (!dev_can_checksum(dev
, skb
) && skb_checksum_help(skb
))
1703 txq
= &dev
->tx_queue
;
1704 spin_lock_prefetch(&txq
->lock
);
1706 /* Disable soft irqs for various locks below. Also
1707 * stops preemption for RCU.
1711 /* Updates of qdisc are serialized by queue->lock.
1712 * The struct Qdisc which is pointed to by qdisc is now a
1713 * rcu structure - it may be accessed without acquiring
1714 * a lock (but the structure may be stale.) The freeing of the
1715 * qdisc will be deferred until it's known that there are no
1716 * more references to it.
1718 * If the qdisc has an enqueue function, we still need to
1719 * hold the queue->lock before calling it, since queue->lock
1720 * also serializes access to the device queue.
1723 q
= rcu_dereference(dev
->qdisc
);
1724 #ifdef CONFIG_NET_CLS_ACT
1725 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1728 /* Grab device queue */
1729 spin_lock(&txq
->lock
);
1732 /* reset queue_mapping to zero */
1733 skb_set_queue_mapping(skb
, 0);
1734 rc
= q
->enqueue(skb
, q
);
1736 spin_unlock(&txq
->lock
);
1738 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1741 spin_unlock(&txq
->lock
);
1744 /* The device has no queue. Common case for software devices:
1745 loopback, all the sorts of tunnels...
1747 Really, it is unlikely that netif_tx_lock protection is necessary
1748 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1750 However, it is possible, that they rely on protection
1753 Check this and shot the lock. It is not prone from deadlocks.
1754 Either shot noqueue qdisc, it is even simpler 8)
1756 if (dev
->flags
& IFF_UP
) {
1757 int cpu
= smp_processor_id(); /* ok because BHs are off */
1759 if (dev
->xmit_lock_owner
!= cpu
) {
1761 HARD_TX_LOCK(dev
, cpu
);
1763 if (!netif_queue_stopped(dev
) &&
1764 !netif_subqueue_stopped(dev
, skb
)) {
1766 if (!dev_hard_start_xmit(skb
, dev
)) {
1767 HARD_TX_UNLOCK(dev
);
1771 HARD_TX_UNLOCK(dev
);
1772 if (net_ratelimit())
1773 printk(KERN_CRIT
"Virtual device %s asks to "
1774 "queue packet!\n", dev
->name
);
1776 /* Recursion is detected! It is possible,
1778 if (net_ratelimit())
1779 printk(KERN_CRIT
"Dead loop on virtual device "
1780 "%s, fix it urgently!\n", dev
->name
);
1785 rcu_read_unlock_bh();
1791 rcu_read_unlock_bh();
1796 /*=======================================================================
1798 =======================================================================*/
1800 int netdev_max_backlog __read_mostly
= 1000;
1801 int netdev_budget __read_mostly
= 300;
1802 int weight_p __read_mostly
= 64; /* old backlog weight */
1804 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1808 * netif_rx - post buffer to the network code
1809 * @skb: buffer to post
1811 * This function receives a packet from a device driver and queues it for
1812 * the upper (protocol) levels to process. It always succeeds. The buffer
1813 * may be dropped during processing for congestion control or by the
1817 * NET_RX_SUCCESS (no congestion)
1818 * NET_RX_DROP (packet was dropped)
1822 int netif_rx(struct sk_buff
*skb
)
1824 struct softnet_data
*queue
;
1825 unsigned long flags
;
1827 /* if netpoll wants it, pretend we never saw it */
1828 if (netpoll_rx(skb
))
1831 if (!skb
->tstamp
.tv64
)
1835 * The code is rearranged so that the path is the most
1836 * short when CPU is congested, but is still operating.
1838 local_irq_save(flags
);
1839 queue
= &__get_cpu_var(softnet_data
);
1841 __get_cpu_var(netdev_rx_stat
).total
++;
1842 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1843 if (queue
->input_pkt_queue
.qlen
) {
1846 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1847 local_irq_restore(flags
);
1848 return NET_RX_SUCCESS
;
1851 napi_schedule(&queue
->backlog
);
1855 __get_cpu_var(netdev_rx_stat
).dropped
++;
1856 local_irq_restore(flags
);
1862 int netif_rx_ni(struct sk_buff
*skb
)
1867 err
= netif_rx(skb
);
1868 if (local_softirq_pending())
1875 EXPORT_SYMBOL(netif_rx_ni
);
1877 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1879 struct net_device
*dev
= skb
->dev
;
1882 if (skb_bond_should_drop(skb
)) {
1886 skb
->dev
= dev
->master
;
1893 static void net_tx_action(struct softirq_action
*h
)
1895 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1897 if (sd
->completion_queue
) {
1898 struct sk_buff
*clist
;
1900 local_irq_disable();
1901 clist
= sd
->completion_queue
;
1902 sd
->completion_queue
= NULL
;
1906 struct sk_buff
*skb
= clist
;
1907 clist
= clist
->next
;
1909 BUG_TRAP(!atomic_read(&skb
->users
));
1914 if (sd
->output_queue
) {
1915 struct net_device
*head
;
1917 local_irq_disable();
1918 head
= sd
->output_queue
;
1919 sd
->output_queue
= NULL
;
1923 struct net_device
*dev
= head
;
1924 struct netdev_queue
*txq
;
1925 head
= head
->next_sched
;
1927 txq
= &dev
->tx_queue
;
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(dev
);
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
)
2018 struct net_device
*dev
= skb
->dev
;
2019 int result
= TC_ACT_OK
;
2020 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
2022 if (MAX_RED_LOOP
< ttl
++) {
2024 "Redir loop detected Dropping packet (%d->%d)\n",
2025 skb
->iif
, dev
->ifindex
);
2029 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
2030 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
2032 spin_lock(&dev
->ingress_lock
);
2033 if ((q
= dev
->qdisc_ingress
) != NULL
)
2034 result
= q
->enqueue(skb
, q
);
2035 spin_unlock(&dev
->ingress_lock
);
2040 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
2041 struct packet_type
**pt_prev
,
2042 int *ret
, struct net_device
*orig_dev
)
2044 if (!skb
->dev
->qdisc_ingress
)
2048 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
2051 /* Huh? Why does turning on AF_PACKET affect this? */
2052 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
2055 switch (ing_filter(skb
)) {
2069 * netif_receive_skb - process receive buffer from network
2070 * @skb: buffer to process
2072 * netif_receive_skb() is the main receive data processing function.
2073 * It always succeeds. The buffer may be dropped during processing
2074 * for congestion control or by the protocol layers.
2076 * This function may only be called from softirq context and interrupts
2077 * should be enabled.
2079 * Return values (usually ignored):
2080 * NET_RX_SUCCESS: no congestion
2081 * NET_RX_DROP: packet was dropped
2083 int netif_receive_skb(struct sk_buff
*skb
)
2085 struct packet_type
*ptype
, *pt_prev
;
2086 struct net_device
*orig_dev
;
2087 int ret
= NET_RX_DROP
;
2090 /* if we've gotten here through NAPI, check netpoll */
2091 if (netpoll_receive_skb(skb
))
2094 if (!skb
->tstamp
.tv64
)
2098 skb
->iif
= skb
->dev
->ifindex
;
2100 orig_dev
= skb_bond(skb
);
2105 __get_cpu_var(netdev_rx_stat
).total
++;
2107 skb_reset_network_header(skb
);
2108 skb_reset_transport_header(skb
);
2109 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
2115 /* Don't receive packets in an exiting network namespace */
2116 if (!net_alive(dev_net(skb
->dev
)))
2119 #ifdef CONFIG_NET_CLS_ACT
2120 if (skb
->tc_verd
& TC_NCLS
) {
2121 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
2126 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
2127 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
2129 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2134 #ifdef CONFIG_NET_CLS_ACT
2135 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
2141 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
2144 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
2148 type
= skb
->protocol
;
2149 list_for_each_entry_rcu(ptype
,
2150 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
2151 if (ptype
->type
== type
&&
2152 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
2154 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2160 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2163 /* Jamal, now you will not able to escape explaining
2164 * me how you were going to use this. :-)
2174 static int process_backlog(struct napi_struct
*napi
, int quota
)
2177 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2178 unsigned long start_time
= jiffies
;
2180 napi
->weight
= weight_p
;
2182 struct sk_buff
*skb
;
2183 struct net_device
*dev
;
2185 local_irq_disable();
2186 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
2188 __napi_complete(napi
);
2197 netif_receive_skb(skb
);
2200 } while (++work
< quota
&& jiffies
== start_time
);
2206 * __napi_schedule - schedule for receive
2207 * @n: entry to schedule
2209 * The entry's receive function will be scheduled to run
2211 void __napi_schedule(struct napi_struct
*n
)
2213 unsigned long flags
;
2215 local_irq_save(flags
);
2216 list_add_tail(&n
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
2217 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2218 local_irq_restore(flags
);
2220 EXPORT_SYMBOL(__napi_schedule
);
2223 static void net_rx_action(struct softirq_action
*h
)
2225 struct list_head
*list
= &__get_cpu_var(softnet_data
).poll_list
;
2226 unsigned long start_time
= jiffies
;
2227 int budget
= netdev_budget
;
2230 local_irq_disable();
2232 while (!list_empty(list
)) {
2233 struct napi_struct
*n
;
2236 /* If softirq window is exhuasted then punt.
2238 * Note that this is a slight policy change from the
2239 * previous NAPI code, which would allow up to 2
2240 * jiffies to pass before breaking out. The test
2241 * used to be "jiffies - start_time > 1".
2243 if (unlikely(budget
<= 0 || jiffies
!= start_time
))
2248 /* Even though interrupts have been re-enabled, this
2249 * access is safe because interrupts can only add new
2250 * entries to the tail of this list, and only ->poll()
2251 * calls can remove this head entry from the list.
2253 n
= list_entry(list
->next
, struct napi_struct
, poll_list
);
2255 have
= netpoll_poll_lock(n
);
2259 /* This NAPI_STATE_SCHED test is for avoiding a race
2260 * with netpoll's poll_napi(). Only the entity which
2261 * obtains the lock and sees NAPI_STATE_SCHED set will
2262 * actually make the ->poll() call. Therefore we avoid
2263 * accidently calling ->poll() when NAPI is not scheduled.
2266 if (test_bit(NAPI_STATE_SCHED
, &n
->state
))
2267 work
= n
->poll(n
, weight
);
2269 WARN_ON_ONCE(work
> weight
);
2273 local_irq_disable();
2275 /* Drivers must not modify the NAPI state if they
2276 * consume the entire weight. In such cases this code
2277 * still "owns" the NAPI instance and therefore can
2278 * move the instance around on the list at-will.
2280 if (unlikely(work
== weight
)) {
2281 if (unlikely(napi_disable_pending(n
)))
2284 list_move_tail(&n
->poll_list
, list
);
2287 netpoll_poll_unlock(have
);
2292 #ifdef CONFIG_NET_DMA
2294 * There may not be any more sk_buffs coming right now, so push
2295 * any pending DMA copies to hardware
2297 if (!cpus_empty(net_dma
.channel_mask
)) {
2299 for_each_cpu_mask(chan_idx
, net_dma
.channel_mask
) {
2300 struct dma_chan
*chan
= net_dma
.channels
[chan_idx
];
2302 dma_async_memcpy_issue_pending(chan
);
2310 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2311 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2315 static gifconf_func_t
* gifconf_list
[NPROTO
];
2318 * register_gifconf - register a SIOCGIF handler
2319 * @family: Address family
2320 * @gifconf: Function handler
2322 * Register protocol dependent address dumping routines. The handler
2323 * that is passed must not be freed or reused until it has been replaced
2324 * by another handler.
2326 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2328 if (family
>= NPROTO
)
2330 gifconf_list
[family
] = gifconf
;
2336 * Map an interface index to its name (SIOCGIFNAME)
2340 * We need this ioctl for efficient implementation of the
2341 * if_indextoname() function required by the IPv6 API. Without
2342 * it, we would have to search all the interfaces to find a
2346 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
2348 struct net_device
*dev
;
2352 * Fetch the caller's info block.
2355 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2358 read_lock(&dev_base_lock
);
2359 dev
= __dev_get_by_index(net
, ifr
.ifr_ifindex
);
2361 read_unlock(&dev_base_lock
);
2365 strcpy(ifr
.ifr_name
, dev
->name
);
2366 read_unlock(&dev_base_lock
);
2368 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2374 * Perform a SIOCGIFCONF call. This structure will change
2375 * size eventually, and there is nothing I can do about it.
2376 * Thus we will need a 'compatibility mode'.
2379 static int dev_ifconf(struct net
*net
, char __user
*arg
)
2382 struct net_device
*dev
;
2389 * Fetch the caller's info block.
2392 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2399 * Loop over the interfaces, and write an info block for each.
2403 for_each_netdev(net
, dev
) {
2404 for (i
= 0; i
< NPROTO
; i
++) {
2405 if (gifconf_list
[i
]) {
2408 done
= gifconf_list
[i
](dev
, NULL
, 0);
2410 done
= gifconf_list
[i
](dev
, pos
+ total
,
2420 * All done. Write the updated control block back to the caller.
2422 ifc
.ifc_len
= total
;
2425 * Both BSD and Solaris return 0 here, so we do too.
2427 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2430 #ifdef CONFIG_PROC_FS
2432 * This is invoked by the /proc filesystem handler to display a device
2435 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2436 __acquires(dev_base_lock
)
2438 struct net
*net
= seq_file_net(seq
);
2440 struct net_device
*dev
;
2442 read_lock(&dev_base_lock
);
2444 return SEQ_START_TOKEN
;
2447 for_each_netdev(net
, dev
)
2454 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2456 struct net
*net
= seq_file_net(seq
);
2458 return v
== SEQ_START_TOKEN
?
2459 first_net_device(net
) : next_net_device((struct net_device
*)v
);
2462 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2463 __releases(dev_base_lock
)
2465 read_unlock(&dev_base_lock
);
2468 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2470 struct net_device_stats
*stats
= dev
->get_stats(dev
);
2472 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2473 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2474 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2476 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2477 stats
->rx_fifo_errors
,
2478 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2479 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2480 stats
->rx_compressed
, stats
->multicast
,
2481 stats
->tx_bytes
, stats
->tx_packets
,
2482 stats
->tx_errors
, stats
->tx_dropped
,
2483 stats
->tx_fifo_errors
, stats
->collisions
,
2484 stats
->tx_carrier_errors
+
2485 stats
->tx_aborted_errors
+
2486 stats
->tx_window_errors
+
2487 stats
->tx_heartbeat_errors
,
2488 stats
->tx_compressed
);
2492 * Called from the PROCfs module. This now uses the new arbitrary sized
2493 * /proc/net interface to create /proc/net/dev
2495 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2497 if (v
== SEQ_START_TOKEN
)
2498 seq_puts(seq
, "Inter-| Receive "
2500 " face |bytes packets errs drop fifo frame "
2501 "compressed multicast|bytes packets errs "
2502 "drop fifo colls carrier compressed\n");
2504 dev_seq_printf_stats(seq
, v
);
2508 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2510 struct netif_rx_stats
*rc
= NULL
;
2512 while (*pos
< nr_cpu_ids
)
2513 if (cpu_online(*pos
)) {
2514 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2521 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2523 return softnet_get_online(pos
);
2526 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2529 return softnet_get_online(pos
);
2532 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2536 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2538 struct netif_rx_stats
*s
= v
;
2540 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2541 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2542 0, 0, 0, 0, /* was fastroute */
2547 static const struct seq_operations dev_seq_ops
= {
2548 .start
= dev_seq_start
,
2549 .next
= dev_seq_next
,
2550 .stop
= dev_seq_stop
,
2551 .show
= dev_seq_show
,
2554 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2556 return seq_open_net(inode
, file
, &dev_seq_ops
,
2557 sizeof(struct seq_net_private
));
2560 static const struct file_operations dev_seq_fops
= {
2561 .owner
= THIS_MODULE
,
2562 .open
= dev_seq_open
,
2564 .llseek
= seq_lseek
,
2565 .release
= seq_release_net
,
2568 static const struct seq_operations softnet_seq_ops
= {
2569 .start
= softnet_seq_start
,
2570 .next
= softnet_seq_next
,
2571 .stop
= softnet_seq_stop
,
2572 .show
= softnet_seq_show
,
2575 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2577 return seq_open(file
, &softnet_seq_ops
);
2580 static const struct file_operations softnet_seq_fops
= {
2581 .owner
= THIS_MODULE
,
2582 .open
= softnet_seq_open
,
2584 .llseek
= seq_lseek
,
2585 .release
= seq_release
,
2588 static void *ptype_get_idx(loff_t pos
)
2590 struct packet_type
*pt
= NULL
;
2594 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
2600 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
2601 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
2610 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2614 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
2617 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2619 struct packet_type
*pt
;
2620 struct list_head
*nxt
;
2624 if (v
== SEQ_START_TOKEN
)
2625 return ptype_get_idx(0);
2628 nxt
= pt
->list
.next
;
2629 if (pt
->type
== htons(ETH_P_ALL
)) {
2630 if (nxt
!= &ptype_all
)
2633 nxt
= ptype_base
[0].next
;
2635 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
2637 while (nxt
== &ptype_base
[hash
]) {
2638 if (++hash
>= PTYPE_HASH_SIZE
)
2640 nxt
= ptype_base
[hash
].next
;
2643 return list_entry(nxt
, struct packet_type
, list
);
2646 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
2652 static void ptype_seq_decode(struct seq_file
*seq
, void *sym
)
2654 #ifdef CONFIG_KALLSYMS
2655 unsigned long offset
= 0, symsize
;
2656 const char *symname
;
2660 symname
= kallsyms_lookup((unsigned long)sym
, &symsize
, &offset
,
2667 modname
= delim
= "";
2668 seq_printf(seq
, "%s%s%s%s+0x%lx", delim
, modname
, delim
,
2674 seq_printf(seq
, "[%p]", sym
);
2677 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
2679 struct packet_type
*pt
= v
;
2681 if (v
== SEQ_START_TOKEN
)
2682 seq_puts(seq
, "Type Device Function\n");
2683 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
2684 if (pt
->type
== htons(ETH_P_ALL
))
2685 seq_puts(seq
, "ALL ");
2687 seq_printf(seq
, "%04x", ntohs(pt
->type
));
2689 seq_printf(seq
, " %-8s ",
2690 pt
->dev
? pt
->dev
->name
: "");
2691 ptype_seq_decode(seq
, pt
->func
);
2692 seq_putc(seq
, '\n');
2698 static const struct seq_operations ptype_seq_ops
= {
2699 .start
= ptype_seq_start
,
2700 .next
= ptype_seq_next
,
2701 .stop
= ptype_seq_stop
,
2702 .show
= ptype_seq_show
,
2705 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
2707 return seq_open_net(inode
, file
, &ptype_seq_ops
,
2708 sizeof(struct seq_net_private
));
2711 static const struct file_operations ptype_seq_fops
= {
2712 .owner
= THIS_MODULE
,
2713 .open
= ptype_seq_open
,
2715 .llseek
= seq_lseek
,
2716 .release
= seq_release_net
,
2720 static int __net_init
dev_proc_net_init(struct net
*net
)
2724 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
2726 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2728 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
2731 if (wext_proc_init(net
))
2737 proc_net_remove(net
, "ptype");
2739 proc_net_remove(net
, "softnet_stat");
2741 proc_net_remove(net
, "dev");
2745 static void __net_exit
dev_proc_net_exit(struct net
*net
)
2747 wext_proc_exit(net
);
2749 proc_net_remove(net
, "ptype");
2750 proc_net_remove(net
, "softnet_stat");
2751 proc_net_remove(net
, "dev");
2754 static struct pernet_operations __net_initdata dev_proc_ops
= {
2755 .init
= dev_proc_net_init
,
2756 .exit
= dev_proc_net_exit
,
2759 static int __init
dev_proc_init(void)
2761 return register_pernet_subsys(&dev_proc_ops
);
2764 #define dev_proc_init() 0
2765 #endif /* CONFIG_PROC_FS */
2769 * netdev_set_master - set up master/slave pair
2770 * @slave: slave device
2771 * @master: new master device
2773 * Changes the master device of the slave. Pass %NULL to break the
2774 * bonding. The caller must hold the RTNL semaphore. On a failure
2775 * a negative errno code is returned. On success the reference counts
2776 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2777 * function returns zero.
2779 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2781 struct net_device
*old
= slave
->master
;
2791 slave
->master
= master
;
2799 slave
->flags
|= IFF_SLAVE
;
2801 slave
->flags
&= ~IFF_SLAVE
;
2803 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2807 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
2809 unsigned short old_flags
= dev
->flags
;
2813 dev
->flags
|= IFF_PROMISC
;
2814 dev
->promiscuity
+= inc
;
2815 if (dev
->promiscuity
== 0) {
2818 * If inc causes overflow, untouch promisc and return error.
2821 dev
->flags
&= ~IFF_PROMISC
;
2823 dev
->promiscuity
-= inc
;
2824 printk(KERN_WARNING
"%s: promiscuity touches roof, "
2825 "set promiscuity failed, promiscuity feature "
2826 "of device might be broken.\n", dev
->name
);
2830 if (dev
->flags
!= old_flags
) {
2831 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2832 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2835 audit_log(current
->audit_context
, GFP_ATOMIC
,
2836 AUDIT_ANOM_PROMISCUOUS
,
2837 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2838 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2839 (old_flags
& IFF_PROMISC
),
2840 audit_get_loginuid(current
),
2841 current
->uid
, current
->gid
,
2842 audit_get_sessionid(current
));
2844 if (dev
->change_rx_flags
)
2845 dev
->change_rx_flags(dev
, IFF_PROMISC
);
2851 * dev_set_promiscuity - update promiscuity count on a device
2855 * Add or remove promiscuity from a device. While the count in the device
2856 * remains above zero the interface remains promiscuous. Once it hits zero
2857 * the device reverts back to normal filtering operation. A negative inc
2858 * value is used to drop promiscuity on the device.
2859 * Return 0 if successful or a negative errno code on error.
2861 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
2863 unsigned short old_flags
= dev
->flags
;
2866 err
= __dev_set_promiscuity(dev
, inc
);
2869 if (dev
->flags
!= old_flags
)
2870 dev_set_rx_mode(dev
);
2875 * dev_set_allmulti - update allmulti count on a device
2879 * Add or remove reception of all multicast frames to a device. While the
2880 * count in the device remains above zero the interface remains listening
2881 * to all interfaces. Once it hits zero the device reverts back to normal
2882 * filtering operation. A negative @inc value is used to drop the counter
2883 * when releasing a resource needing all multicasts.
2884 * Return 0 if successful or a negative errno code on error.
2887 int dev_set_allmulti(struct net_device
*dev
, int inc
)
2889 unsigned short old_flags
= dev
->flags
;
2893 dev
->flags
|= IFF_ALLMULTI
;
2894 dev
->allmulti
+= inc
;
2895 if (dev
->allmulti
== 0) {
2898 * If inc causes overflow, untouch allmulti and return error.
2901 dev
->flags
&= ~IFF_ALLMULTI
;
2903 dev
->allmulti
-= inc
;
2904 printk(KERN_WARNING
"%s: allmulti touches roof, "
2905 "set allmulti failed, allmulti feature of "
2906 "device might be broken.\n", dev
->name
);
2910 if (dev
->flags
^ old_flags
) {
2911 if (dev
->change_rx_flags
)
2912 dev
->change_rx_flags(dev
, IFF_ALLMULTI
);
2913 dev_set_rx_mode(dev
);
2919 * Upload unicast and multicast address lists to device and
2920 * configure RX filtering. When the device doesn't support unicast
2921 * filtering it is put in promiscuous mode while unicast addresses
2924 void __dev_set_rx_mode(struct net_device
*dev
)
2926 /* dev_open will call this function so the list will stay sane. */
2927 if (!(dev
->flags
&IFF_UP
))
2930 if (!netif_device_present(dev
))
2933 if (dev
->set_rx_mode
)
2934 dev
->set_rx_mode(dev
);
2936 /* Unicast addresses changes may only happen under the rtnl,
2937 * therefore calling __dev_set_promiscuity here is safe.
2939 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
2940 __dev_set_promiscuity(dev
, 1);
2941 dev
->uc_promisc
= 1;
2942 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
2943 __dev_set_promiscuity(dev
, -1);
2944 dev
->uc_promisc
= 0;
2947 if (dev
->set_multicast_list
)
2948 dev
->set_multicast_list(dev
);
2952 void dev_set_rx_mode(struct net_device
*dev
)
2954 netif_tx_lock_bh(dev
);
2955 __dev_set_rx_mode(dev
);
2956 netif_tx_unlock_bh(dev
);
2959 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
2960 void *addr
, int alen
, int glbl
)
2962 struct dev_addr_list
*da
;
2964 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
2965 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2966 alen
== da
->da_addrlen
) {
2968 int old_glbl
= da
->da_gusers
;
2985 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
2986 void *addr
, int alen
, int glbl
)
2988 struct dev_addr_list
*da
;
2990 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
2991 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2992 da
->da_addrlen
== alen
) {
2994 int old_glbl
= da
->da_gusers
;
3004 da
= kzalloc(sizeof(*da
), GFP_ATOMIC
);
3007 memcpy(da
->da_addr
, addr
, alen
);
3008 da
->da_addrlen
= alen
;
3010 da
->da_gusers
= glbl
? 1 : 0;
3018 * dev_unicast_delete - Release secondary unicast address.
3020 * @addr: address to delete
3021 * @alen: length of @addr
3023 * Release reference to a secondary unicast address and remove it
3024 * from the device if the reference count drops to zero.
3026 * The caller must hold the rtnl_mutex.
3028 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
3034 netif_tx_lock_bh(dev
);
3035 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3037 __dev_set_rx_mode(dev
);
3038 netif_tx_unlock_bh(dev
);
3041 EXPORT_SYMBOL(dev_unicast_delete
);
3044 * dev_unicast_add - add a secondary unicast address
3046 * @addr: address to add
3047 * @alen: length of @addr
3049 * Add a secondary unicast address to the device or increase
3050 * the reference count if it already exists.
3052 * The caller must hold the rtnl_mutex.
3054 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
3060 netif_tx_lock_bh(dev
);
3061 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
3063 __dev_set_rx_mode(dev
);
3064 netif_tx_unlock_bh(dev
);
3067 EXPORT_SYMBOL(dev_unicast_add
);
3069 int __dev_addr_sync(struct dev_addr_list
**to
, int *to_count
,
3070 struct dev_addr_list
**from
, int *from_count
)
3072 struct dev_addr_list
*da
, *next
;
3076 while (da
!= NULL
) {
3078 if (!da
->da_synced
) {
3079 err
= __dev_addr_add(to
, to_count
,
3080 da
->da_addr
, da
->da_addrlen
, 0);
3085 } else if (da
->da_users
== 1) {
3086 __dev_addr_delete(to
, to_count
,
3087 da
->da_addr
, da
->da_addrlen
, 0);
3088 __dev_addr_delete(from
, from_count
,
3089 da
->da_addr
, da
->da_addrlen
, 0);
3096 void __dev_addr_unsync(struct dev_addr_list
**to
, int *to_count
,
3097 struct dev_addr_list
**from
, int *from_count
)
3099 struct dev_addr_list
*da
, *next
;
3102 while (da
!= NULL
) {
3104 if (da
->da_synced
) {
3105 __dev_addr_delete(to
, to_count
,
3106 da
->da_addr
, da
->da_addrlen
, 0);
3108 __dev_addr_delete(from
, from_count
,
3109 da
->da_addr
, da
->da_addrlen
, 0);
3116 * dev_unicast_sync - Synchronize device's unicast list to another device
3117 * @to: destination device
3118 * @from: source device
3120 * Add newly added addresses to the destination device and release
3121 * addresses that have no users left. The source device must be
3122 * locked by netif_tx_lock_bh.
3124 * This function is intended to be called from the dev->set_rx_mode
3125 * function of layered software devices.
3127 int dev_unicast_sync(struct net_device
*to
, struct net_device
*from
)
3131 netif_tx_lock_bh(to
);
3132 err
= __dev_addr_sync(&to
->uc_list
, &to
->uc_count
,
3133 &from
->uc_list
, &from
->uc_count
);
3135 __dev_set_rx_mode(to
);
3136 netif_tx_unlock_bh(to
);
3139 EXPORT_SYMBOL(dev_unicast_sync
);
3142 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3143 * @to: destination device
3144 * @from: source device
3146 * Remove all addresses that were added to the destination device by
3147 * dev_unicast_sync(). This function is intended to be called from the
3148 * dev->stop function of layered software devices.
3150 void dev_unicast_unsync(struct net_device
*to
, struct net_device
*from
)
3152 netif_tx_lock_bh(from
);
3153 netif_tx_lock_bh(to
);
3155 __dev_addr_unsync(&to
->uc_list
, &to
->uc_count
,
3156 &from
->uc_list
, &from
->uc_count
);
3157 __dev_set_rx_mode(to
);
3159 netif_tx_unlock_bh(to
);
3160 netif_tx_unlock_bh(from
);
3162 EXPORT_SYMBOL(dev_unicast_unsync
);
3164 static void __dev_addr_discard(struct dev_addr_list
**list
)
3166 struct dev_addr_list
*tmp
;
3168 while (*list
!= NULL
) {
3171 if (tmp
->da_users
> tmp
->da_gusers
)
3172 printk("__dev_addr_discard: address leakage! "
3173 "da_users=%d\n", tmp
->da_users
);
3178 static void dev_addr_discard(struct net_device
*dev
)
3180 netif_tx_lock_bh(dev
);
3182 __dev_addr_discard(&dev
->uc_list
);
3185 __dev_addr_discard(&dev
->mc_list
);
3188 netif_tx_unlock_bh(dev
);
3191 unsigned dev_get_flags(const struct net_device
*dev
)
3195 flags
= (dev
->flags
& ~(IFF_PROMISC
|
3200 (dev
->gflags
& (IFF_PROMISC
|
3203 if (netif_running(dev
)) {
3204 if (netif_oper_up(dev
))
3205 flags
|= IFF_RUNNING
;
3206 if (netif_carrier_ok(dev
))
3207 flags
|= IFF_LOWER_UP
;
3208 if (netif_dormant(dev
))
3209 flags
|= IFF_DORMANT
;
3215 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
3218 int old_flags
= dev
->flags
;
3223 * Set the flags on our device.
3226 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
3227 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
3229 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
3233 * Load in the correct multicast list now the flags have changed.
3236 if (dev
->change_rx_flags
&& (old_flags
^ flags
) & IFF_MULTICAST
)
3237 dev
->change_rx_flags(dev
, IFF_MULTICAST
);
3239 dev_set_rx_mode(dev
);
3242 * Have we downed the interface. We handle IFF_UP ourselves
3243 * according to user attempts to set it, rather than blindly
3248 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
3249 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
3252 dev_set_rx_mode(dev
);
3255 if (dev
->flags
& IFF_UP
&&
3256 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
3258 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
3260 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
3261 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
3262 dev
->gflags
^= IFF_PROMISC
;
3263 dev_set_promiscuity(dev
, inc
);
3266 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3267 is important. Some (broken) drivers set IFF_PROMISC, when
3268 IFF_ALLMULTI is requested not asking us and not reporting.
3270 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
3271 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
3272 dev
->gflags
^= IFF_ALLMULTI
;
3273 dev_set_allmulti(dev
, inc
);
3276 /* Exclude state transition flags, already notified */
3277 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
3279 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
3284 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
3288 if (new_mtu
== dev
->mtu
)
3291 /* MTU must be positive. */
3295 if (!netif_device_present(dev
))
3299 if (dev
->change_mtu
)
3300 err
= dev
->change_mtu(dev
, new_mtu
);
3303 if (!err
&& dev
->flags
& IFF_UP
)
3304 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
3308 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
3312 if (!dev
->set_mac_address
)
3314 if (sa
->sa_family
!= dev
->type
)
3316 if (!netif_device_present(dev
))
3318 err
= dev
->set_mac_address(dev
, sa
);
3320 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3325 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3327 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3330 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3336 case SIOCGIFFLAGS
: /* Get interface flags */
3337 ifr
->ifr_flags
= dev_get_flags(dev
);
3340 case SIOCGIFMETRIC
: /* Get the metric on the interface
3341 (currently unused) */
3342 ifr
->ifr_metric
= 0;
3345 case SIOCGIFMTU
: /* Get the MTU of a device */
3346 ifr
->ifr_mtu
= dev
->mtu
;
3351 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
3353 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
3354 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3355 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
3363 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
3364 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
3365 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
3366 ifr
->ifr_map
.irq
= dev
->irq
;
3367 ifr
->ifr_map
.dma
= dev
->dma
;
3368 ifr
->ifr_map
.port
= dev
->if_port
;
3372 ifr
->ifr_ifindex
= dev
->ifindex
;
3376 ifr
->ifr_qlen
= dev
->tx_queue_len
;
3380 /* dev_ioctl() should ensure this case
3392 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3394 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
3397 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
3403 case SIOCSIFFLAGS
: /* Set interface flags */
3404 return dev_change_flags(dev
, ifr
->ifr_flags
);
3406 case SIOCSIFMETRIC
: /* Set the metric on the interface
3407 (currently unused) */
3410 case SIOCSIFMTU
: /* Set the MTU of a device */
3411 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
3414 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
3416 case SIOCSIFHWBROADCAST
:
3417 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
3419 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
3420 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3421 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
3425 if (dev
->set_config
) {
3426 if (!netif_device_present(dev
))
3428 return dev
->set_config(dev
, &ifr
->ifr_map
);
3433 if ((!dev
->set_multicast_list
&& !dev
->set_rx_mode
) ||
3434 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3436 if (!netif_device_present(dev
))
3438 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
3442 if ((!dev
->set_multicast_list
&& !dev
->set_rx_mode
) ||
3443 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3445 if (!netif_device_present(dev
))
3447 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
3451 if (ifr
->ifr_qlen
< 0)
3453 dev
->tx_queue_len
= ifr
->ifr_qlen
;
3457 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
3458 return dev_change_name(dev
, ifr
->ifr_newname
);
3461 * Unknown or private ioctl
3465 if ((cmd
>= SIOCDEVPRIVATE
&&
3466 cmd
<= SIOCDEVPRIVATE
+ 15) ||
3467 cmd
== SIOCBONDENSLAVE
||
3468 cmd
== SIOCBONDRELEASE
||
3469 cmd
== SIOCBONDSETHWADDR
||
3470 cmd
== SIOCBONDSLAVEINFOQUERY
||
3471 cmd
== SIOCBONDINFOQUERY
||
3472 cmd
== SIOCBONDCHANGEACTIVE
||
3473 cmd
== SIOCGMIIPHY
||
3474 cmd
== SIOCGMIIREG
||
3475 cmd
== SIOCSMIIREG
||
3476 cmd
== SIOCBRADDIF
||
3477 cmd
== SIOCBRDELIF
||
3478 cmd
== SIOCWANDEV
) {
3480 if (dev
->do_ioctl
) {
3481 if (netif_device_present(dev
))
3482 err
= dev
->do_ioctl(dev
, ifr
,
3495 * This function handles all "interface"-type I/O control requests. The actual
3496 * 'doing' part of this is dev_ifsioc above.
3500 * dev_ioctl - network device ioctl
3501 * @net: the applicable net namespace
3502 * @cmd: command to issue
3503 * @arg: pointer to a struct ifreq in user space
3505 * Issue ioctl functions to devices. This is normally called by the
3506 * user space syscall interfaces but can sometimes be useful for
3507 * other purposes. The return value is the return from the syscall if
3508 * positive or a negative errno code on error.
3511 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
3517 /* One special case: SIOCGIFCONF takes ifconf argument
3518 and requires shared lock, because it sleeps writing
3522 if (cmd
== SIOCGIFCONF
) {
3524 ret
= dev_ifconf(net
, (char __user
*) arg
);
3528 if (cmd
== SIOCGIFNAME
)
3529 return dev_ifname(net
, (struct ifreq __user
*)arg
);
3531 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3534 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
3536 colon
= strchr(ifr
.ifr_name
, ':');
3541 * See which interface the caller is talking about.
3546 * These ioctl calls:
3547 * - can be done by all.
3548 * - atomic and do not require locking.
3559 dev_load(net
, ifr
.ifr_name
);
3560 read_lock(&dev_base_lock
);
3561 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
3562 read_unlock(&dev_base_lock
);
3566 if (copy_to_user(arg
, &ifr
,
3567 sizeof(struct ifreq
)))
3573 dev_load(net
, ifr
.ifr_name
);
3575 ret
= dev_ethtool(net
, &ifr
);
3580 if (copy_to_user(arg
, &ifr
,
3581 sizeof(struct ifreq
)))
3587 * These ioctl calls:
3588 * - require superuser power.
3589 * - require strict serialization.
3595 if (!capable(CAP_NET_ADMIN
))
3597 dev_load(net
, ifr
.ifr_name
);
3599 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3604 if (copy_to_user(arg
, &ifr
,
3605 sizeof(struct ifreq
)))
3611 * These ioctl calls:
3612 * - require superuser power.
3613 * - require strict serialization.
3614 * - do not return a value
3624 case SIOCSIFHWBROADCAST
:
3627 case SIOCBONDENSLAVE
:
3628 case SIOCBONDRELEASE
:
3629 case SIOCBONDSETHWADDR
:
3630 case SIOCBONDCHANGEACTIVE
:
3633 if (!capable(CAP_NET_ADMIN
))
3636 case SIOCBONDSLAVEINFOQUERY
:
3637 case SIOCBONDINFOQUERY
:
3638 dev_load(net
, ifr
.ifr_name
);
3640 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3645 /* Get the per device memory space. We can add this but
3646 * currently do not support it */
3648 /* Set the per device memory buffer space.
3649 * Not applicable in our case */
3654 * Unknown or private ioctl.
3657 if (cmd
== SIOCWANDEV
||
3658 (cmd
>= SIOCDEVPRIVATE
&&
3659 cmd
<= SIOCDEVPRIVATE
+ 15)) {
3660 dev_load(net
, ifr
.ifr_name
);
3662 ret
= dev_ifsioc(net
, &ifr
, cmd
);
3664 if (!ret
&& copy_to_user(arg
, &ifr
,
3665 sizeof(struct ifreq
)))
3669 /* Take care of Wireless Extensions */
3670 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
3671 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
3678 * dev_new_index - allocate an ifindex
3679 * @net: the applicable net namespace
3681 * Returns a suitable unique value for a new device interface
3682 * number. The caller must hold the rtnl semaphore or the
3683 * dev_base_lock to be sure it remains unique.
3685 static int dev_new_index(struct net
*net
)
3691 if (!__dev_get_by_index(net
, ifindex
))
3696 /* Delayed registration/unregisteration */
3697 static DEFINE_SPINLOCK(net_todo_list_lock
);
3698 static LIST_HEAD(net_todo_list
);
3700 static void net_set_todo(struct net_device
*dev
)
3702 spin_lock(&net_todo_list_lock
);
3703 list_add_tail(&dev
->todo_list
, &net_todo_list
);
3704 spin_unlock(&net_todo_list_lock
);
3707 static void rollback_registered(struct net_device
*dev
)
3709 BUG_ON(dev_boot_phase
);
3712 /* Some devices call without registering for initialization unwind. */
3713 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3714 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3715 "was registered\n", dev
->name
, dev
);
3721 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3723 /* If device is running, close it first. */
3726 /* And unlink it from device chain. */
3727 unlist_netdevice(dev
);
3729 dev
->reg_state
= NETREG_UNREGISTERING
;
3733 /* Shutdown queueing discipline. */
3737 /* Notify protocols, that we are about to destroy
3738 this device. They should clean all the things.
3740 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
3743 * Flush the unicast and multicast chains
3745 dev_addr_discard(dev
);
3750 /* Notifier chain MUST detach us from master device. */
3751 BUG_TRAP(!dev
->master
);
3753 /* Remove entries from kobject tree */
3754 netdev_unregister_kobject(dev
);
3762 * register_netdevice - register a network device
3763 * @dev: device to register
3765 * Take a completed network device structure and add it to the kernel
3766 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3767 * chain. 0 is returned on success. A negative errno code is returned
3768 * on a failure to set up the device, or if the name is a duplicate.
3770 * Callers must hold the rtnl semaphore. You may want
3771 * register_netdev() instead of this.
3774 * The locking appears insufficient to guarantee two parallel registers
3775 * will not get the same name.
3778 int register_netdevice(struct net_device
*dev
)
3780 struct hlist_head
*head
;
3781 struct hlist_node
*p
;
3785 BUG_ON(dev_boot_phase
);
3790 /* When net_device's are persistent, this will be fatal. */
3791 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
3792 BUG_ON(!dev_net(dev
));
3795 spin_lock_init(&dev
->_xmit_lock
);
3796 netdev_set_lockdep_class(&dev
->_xmit_lock
, dev
->type
);
3797 dev
->xmit_lock_owner
= -1;
3798 spin_lock_init(&dev
->ingress_lock
);
3802 /* Init, if this function is available */
3804 ret
= dev
->init(dev
);
3812 if (!dev_valid_name(dev
->name
)) {
3817 dev
->ifindex
= dev_new_index(net
);
3818 if (dev
->iflink
== -1)
3819 dev
->iflink
= dev
->ifindex
;
3821 /* Check for existence of name */
3822 head
= dev_name_hash(net
, dev
->name
);
3823 hlist_for_each(p
, head
) {
3824 struct net_device
*d
3825 = hlist_entry(p
, struct net_device
, name_hlist
);
3826 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
3832 /* Fix illegal checksum combinations */
3833 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
3834 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3835 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
3837 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
3840 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
3841 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3842 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
3844 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
3848 /* Fix illegal SG+CSUM combinations. */
3849 if ((dev
->features
& NETIF_F_SG
) &&
3850 !(dev
->features
& NETIF_F_ALL_CSUM
)) {
3851 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no checksum feature.\n",
3853 dev
->features
&= ~NETIF_F_SG
;
3856 /* TSO requires that SG is present as well. */
3857 if ((dev
->features
& NETIF_F_TSO
) &&
3858 !(dev
->features
& NETIF_F_SG
)) {
3859 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no SG feature.\n",
3861 dev
->features
&= ~NETIF_F_TSO
;
3863 if (dev
->features
& NETIF_F_UFO
) {
3864 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
3865 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3866 "NETIF_F_HW_CSUM feature.\n",
3868 dev
->features
&= ~NETIF_F_UFO
;
3870 if (!(dev
->features
& NETIF_F_SG
)) {
3871 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3872 "NETIF_F_SG feature.\n",
3874 dev
->features
&= ~NETIF_F_UFO
;
3878 netdev_initialize_kobject(dev
);
3879 ret
= netdev_register_kobject(dev
);
3882 dev
->reg_state
= NETREG_REGISTERED
;
3885 * Default initial state at registry is that the
3886 * device is present.
3889 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
3891 dev_init_scheduler(dev
);
3893 list_netdevice(dev
);
3895 /* Notify protocols, that a new device appeared. */
3896 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
3897 ret
= notifier_to_errno(ret
);
3899 rollback_registered(dev
);
3900 dev
->reg_state
= NETREG_UNREGISTERED
;
3913 * register_netdev - register a network device
3914 * @dev: device to register
3916 * Take a completed network device structure and add it to the kernel
3917 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3918 * chain. 0 is returned on success. A negative errno code is returned
3919 * on a failure to set up the device, or if the name is a duplicate.
3921 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3922 * and expands the device name if you passed a format string to
3925 int register_netdev(struct net_device
*dev
)
3932 * If the name is a format string the caller wants us to do a
3935 if (strchr(dev
->name
, '%')) {
3936 err
= dev_alloc_name(dev
, dev
->name
);
3941 err
= register_netdevice(dev
);
3946 EXPORT_SYMBOL(register_netdev
);
3949 * netdev_wait_allrefs - wait until all references are gone.
3951 * This is called when unregistering network devices.
3953 * Any protocol or device that holds a reference should register
3954 * for netdevice notification, and cleanup and put back the
3955 * reference if they receive an UNREGISTER event.
3956 * We can get stuck here if buggy protocols don't correctly
3959 static void netdev_wait_allrefs(struct net_device
*dev
)
3961 unsigned long rebroadcast_time
, warning_time
;
3963 rebroadcast_time
= warning_time
= jiffies
;
3964 while (atomic_read(&dev
->refcnt
) != 0) {
3965 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
3968 /* Rebroadcast unregister notification */
3969 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
3971 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
3973 /* We must not have linkwatch events
3974 * pending on unregister. If this
3975 * happens, we simply run the queue
3976 * unscheduled, resulting in a noop
3979 linkwatch_run_queue();
3984 rebroadcast_time
= jiffies
;
3989 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
3990 printk(KERN_EMERG
"unregister_netdevice: "
3991 "waiting for %s to become free. Usage "
3993 dev
->name
, atomic_read(&dev
->refcnt
));
3994 warning_time
= jiffies
;
4003 * register_netdevice(x1);
4004 * register_netdevice(x2);
4006 * unregister_netdevice(y1);
4007 * unregister_netdevice(y2);
4013 * We are invoked by rtnl_unlock() after it drops the semaphore.
4014 * This allows us to deal with problems:
4015 * 1) We can delete sysfs objects which invoke hotplug
4016 * without deadlocking with linkwatch via keventd.
4017 * 2) Since we run with the RTNL semaphore not held, we can sleep
4018 * safely in order to wait for the netdev refcnt to drop to zero.
4020 static DEFINE_MUTEX(net_todo_run_mutex
);
4021 void netdev_run_todo(void)
4023 struct list_head list
;
4025 /* Need to guard against multiple cpu's getting out of order. */
4026 mutex_lock(&net_todo_run_mutex
);
4028 /* Not safe to do outside the semaphore. We must not return
4029 * until all unregister events invoked by the local processor
4030 * have been completed (either by this todo run, or one on
4033 if (list_empty(&net_todo_list
))
4036 /* Snapshot list, allow later requests */
4037 spin_lock(&net_todo_list_lock
);
4038 list_replace_init(&net_todo_list
, &list
);
4039 spin_unlock(&net_todo_list_lock
);
4041 while (!list_empty(&list
)) {
4042 struct net_device
*dev
4043 = list_entry(list
.next
, struct net_device
, todo_list
);
4044 list_del(&dev
->todo_list
);
4046 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
4047 printk(KERN_ERR
"network todo '%s' but state %d\n",
4048 dev
->name
, dev
->reg_state
);
4053 dev
->reg_state
= NETREG_UNREGISTERED
;
4055 netdev_wait_allrefs(dev
);
4058 BUG_ON(atomic_read(&dev
->refcnt
));
4059 BUG_TRAP(!dev
->ip_ptr
);
4060 BUG_TRAP(!dev
->ip6_ptr
);
4061 BUG_TRAP(!dev
->dn_ptr
);
4063 if (dev
->destructor
)
4064 dev
->destructor(dev
);
4066 /* Free network device */
4067 kobject_put(&dev
->dev
.kobj
);
4071 mutex_unlock(&net_todo_run_mutex
);
4074 static struct net_device_stats
*internal_stats(struct net_device
*dev
)
4079 static void netdev_init_one_queue(struct net_device
*dev
,
4080 struct netdev_queue
*queue
)
4082 spin_lock_init(&queue
->lock
);
4086 static void netdev_init_queues(struct net_device
*dev
)
4088 netdev_init_one_queue(dev
, &dev
->rx_queue
);
4089 netdev_init_one_queue(dev
, &dev
->tx_queue
);
4093 * alloc_netdev_mq - allocate network device
4094 * @sizeof_priv: size of private data to allocate space for
4095 * @name: device name format string
4096 * @setup: callback to initialize device
4097 * @queue_count: the number of subqueues to allocate
4099 * Allocates a struct net_device with private data area for driver use
4100 * and performs basic initialization. Also allocates subquue structs
4101 * for each queue on the device at the end of the netdevice.
4103 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
4104 void (*setup
)(struct net_device
*), unsigned int queue_count
)
4107 struct net_device
*dev
;
4110 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
4112 alloc_size
= sizeof(struct net_device
) +
4113 sizeof(struct net_device_subqueue
) * (queue_count
- 1);
4115 /* ensure 32-byte alignment of private area */
4116 alloc_size
= (alloc_size
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
;
4117 alloc_size
+= sizeof_priv
;
4119 /* ensure 32-byte alignment of whole construct */
4120 alloc_size
+= NETDEV_ALIGN_CONST
;
4122 p
= kzalloc(alloc_size
, GFP_KERNEL
);
4124 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
4128 dev
= (struct net_device
*)
4129 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
4130 dev
->padded
= (char *)dev
- (char *)p
;
4131 dev_net_set(dev
, &init_net
);
4134 dev
->priv
= ((char *)dev
+
4135 ((sizeof(struct net_device
) +
4136 (sizeof(struct net_device_subqueue
) *
4137 (queue_count
- 1)) + NETDEV_ALIGN_CONST
)
4138 & ~NETDEV_ALIGN_CONST
));
4141 dev
->egress_subqueue_count
= queue_count
;
4142 dev
->gso_max_size
= GSO_MAX_SIZE
;
4144 netdev_init_queues(dev
);
4146 dev
->get_stats
= internal_stats
;
4147 netpoll_netdev_init(dev
);
4149 strcpy(dev
->name
, name
);
4152 EXPORT_SYMBOL(alloc_netdev_mq
);
4155 * free_netdev - free network device
4158 * This function does the last stage of destroying an allocated device
4159 * interface. The reference to the device object is released.
4160 * If this is the last reference then it will be freed.
4162 void free_netdev(struct net_device
*dev
)
4164 release_net(dev_net(dev
));
4166 /* Compatibility with error handling in drivers */
4167 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
4168 kfree((char *)dev
- dev
->padded
);
4172 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
4173 dev
->reg_state
= NETREG_RELEASED
;
4175 /* will free via device release */
4176 put_device(&dev
->dev
);
4179 /* Synchronize with packet receive processing. */
4180 void synchronize_net(void)
4187 * unregister_netdevice - remove device from the kernel
4190 * This function shuts down a device interface and removes it
4191 * from the kernel tables.
4193 * Callers must hold the rtnl semaphore. You may want
4194 * unregister_netdev() instead of this.
4197 void unregister_netdevice(struct net_device
*dev
)
4201 rollback_registered(dev
);
4202 /* Finish processing unregister after unlock */
4207 * unregister_netdev - remove device from the kernel
4210 * This function shuts down a device interface and removes it
4211 * from the kernel tables.
4213 * This is just a wrapper for unregister_netdevice that takes
4214 * the rtnl semaphore. In general you want to use this and not
4215 * unregister_netdevice.
4217 void unregister_netdev(struct net_device
*dev
)
4220 unregister_netdevice(dev
);
4224 EXPORT_SYMBOL(unregister_netdev
);
4227 * dev_change_net_namespace - move device to different nethost namespace
4229 * @net: network namespace
4230 * @pat: If not NULL name pattern to try if the current device name
4231 * is already taken in the destination network namespace.
4233 * This function shuts down a device interface and moves it
4234 * to a new network namespace. On success 0 is returned, on
4235 * a failure a netagive errno code is returned.
4237 * Callers must hold the rtnl semaphore.
4240 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
4243 const char *destname
;
4248 /* Don't allow namespace local devices to be moved. */
4250 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
4253 /* Ensure the device has been registrered */
4255 if (dev
->reg_state
!= NETREG_REGISTERED
)
4258 /* Get out if there is nothing todo */
4260 if (net_eq(dev_net(dev
), net
))
4263 /* Pick the destination device name, and ensure
4264 * we can use it in the destination network namespace.
4267 destname
= dev
->name
;
4268 if (__dev_get_by_name(net
, destname
)) {
4269 /* We get here if we can't use the current device name */
4272 if (!dev_valid_name(pat
))
4274 if (strchr(pat
, '%')) {
4275 if (__dev_alloc_name(net
, pat
, buf
) < 0)
4280 if (__dev_get_by_name(net
, destname
))
4285 * And now a mini version of register_netdevice unregister_netdevice.
4288 /* If device is running close it first. */
4291 /* And unlink it from device chain */
4293 unlist_netdevice(dev
);
4297 /* Shutdown queueing discipline. */
4300 /* Notify protocols, that we are about to destroy
4301 this device. They should clean all the things.
4303 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
4306 * Flush the unicast and multicast chains
4308 dev_addr_discard(dev
);
4310 /* Actually switch the network namespace */
4311 dev_net_set(dev
, net
);
4313 /* Assign the new device name */
4314 if (destname
!= dev
->name
)
4315 strcpy(dev
->name
, destname
);
4317 /* If there is an ifindex conflict assign a new one */
4318 if (__dev_get_by_index(net
, dev
->ifindex
)) {
4319 int iflink
= (dev
->iflink
== dev
->ifindex
);
4320 dev
->ifindex
= dev_new_index(net
);
4322 dev
->iflink
= dev
->ifindex
;
4325 /* Fixup kobjects */
4326 netdev_unregister_kobject(dev
);
4327 err
= netdev_register_kobject(dev
);
4330 /* Add the device back in the hashes */
4331 list_netdevice(dev
);
4333 /* Notify protocols, that a new device appeared. */
4334 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
4342 static int dev_cpu_callback(struct notifier_block
*nfb
,
4343 unsigned long action
,
4346 struct sk_buff
**list_skb
;
4347 struct net_device
**list_net
;
4348 struct sk_buff
*skb
;
4349 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
4350 struct softnet_data
*sd
, *oldsd
;
4352 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
4355 local_irq_disable();
4356 cpu
= smp_processor_id();
4357 sd
= &per_cpu(softnet_data
, cpu
);
4358 oldsd
= &per_cpu(softnet_data
, oldcpu
);
4360 /* Find end of our completion_queue. */
4361 list_skb
= &sd
->completion_queue
;
4363 list_skb
= &(*list_skb
)->next
;
4364 /* Append completion queue from offline CPU. */
4365 *list_skb
= oldsd
->completion_queue
;
4366 oldsd
->completion_queue
= NULL
;
4368 /* Find end of our output_queue. */
4369 list_net
= &sd
->output_queue
;
4371 list_net
= &(*list_net
)->next_sched
;
4372 /* Append output queue from offline CPU. */
4373 *list_net
= oldsd
->output_queue
;
4374 oldsd
->output_queue
= NULL
;
4376 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
4379 /* Process offline CPU's input_pkt_queue */
4380 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
4386 #ifdef CONFIG_NET_DMA
4388 * net_dma_rebalance - try to maintain one DMA channel per CPU
4389 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4391 * This is called when the number of channels allocated to the net_dma client
4392 * changes. The net_dma client tries to have one DMA channel per CPU.
4395 static void net_dma_rebalance(struct net_dma
*net_dma
)
4397 unsigned int cpu
, i
, n
, chan_idx
;
4398 struct dma_chan
*chan
;
4400 if (cpus_empty(net_dma
->channel_mask
)) {
4401 for_each_online_cpu(cpu
)
4402 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
4407 cpu
= first_cpu(cpu_online_map
);
4409 for_each_cpu_mask(chan_idx
, net_dma
->channel_mask
) {
4410 chan
= net_dma
->channels
[chan_idx
];
4412 n
= ((num_online_cpus() / cpus_weight(net_dma
->channel_mask
))
4413 + (i
< (num_online_cpus() %
4414 cpus_weight(net_dma
->channel_mask
)) ? 1 : 0));
4417 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
4418 cpu
= next_cpu(cpu
, cpu_online_map
);
4426 * netdev_dma_event - event callback for the net_dma_client
4427 * @client: should always be net_dma_client
4428 * @chan: DMA channel for the event
4429 * @state: DMA state to be handled
4431 static enum dma_state_client
4432 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
4433 enum dma_state state
)
4435 int i
, found
= 0, pos
= -1;
4436 struct net_dma
*net_dma
=
4437 container_of(client
, struct net_dma
, client
);
4438 enum dma_state_client ack
= DMA_DUP
; /* default: take no action */
4440 spin_lock(&net_dma
->lock
);
4442 case DMA_RESOURCE_AVAILABLE
:
4443 for (i
= 0; i
< nr_cpu_ids
; i
++)
4444 if (net_dma
->channels
[i
] == chan
) {
4447 } else if (net_dma
->channels
[i
] == NULL
&& pos
< 0)
4450 if (!found
&& pos
>= 0) {
4452 net_dma
->channels
[pos
] = chan
;
4453 cpu_set(pos
, net_dma
->channel_mask
);
4454 net_dma_rebalance(net_dma
);
4457 case DMA_RESOURCE_REMOVED
:
4458 for (i
= 0; i
< nr_cpu_ids
; i
++)
4459 if (net_dma
->channels
[i
] == chan
) {
4467 cpu_clear(pos
, net_dma
->channel_mask
);
4468 net_dma
->channels
[i
] = NULL
;
4469 net_dma_rebalance(net_dma
);
4475 spin_unlock(&net_dma
->lock
);
4481 * netdev_dma_regiser - register the networking subsystem as a DMA client
4483 static int __init
netdev_dma_register(void)
4485 net_dma
.channels
= kzalloc(nr_cpu_ids
* sizeof(struct net_dma
),
4487 if (unlikely(!net_dma
.channels
)) {
4489 "netdev_dma: no memory for net_dma.channels\n");
4492 spin_lock_init(&net_dma
.lock
);
4493 dma_cap_set(DMA_MEMCPY
, net_dma
.client
.cap_mask
);
4494 dma_async_client_register(&net_dma
.client
);
4495 dma_async_client_chan_request(&net_dma
.client
);
4500 static int __init
netdev_dma_register(void) { return -ENODEV
; }
4501 #endif /* CONFIG_NET_DMA */
4504 * netdev_compute_feature - compute conjunction of two feature sets
4505 * @all: first feature set
4506 * @one: second feature set
4508 * Computes a new feature set after adding a device with feature set
4509 * @one to the master device with current feature set @all. Returns
4510 * the new feature set.
4512 int netdev_compute_features(unsigned long all
, unsigned long one
)
4514 /* if device needs checksumming, downgrade to hw checksumming */
4515 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
4516 all
^= NETIF_F_NO_CSUM
| NETIF_F_HW_CSUM
;
4518 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4519 if (all
& NETIF_F_HW_CSUM
&& !(one
& NETIF_F_HW_CSUM
))
4520 all
^= NETIF_F_HW_CSUM
4521 | NETIF_F_IP_CSUM
| NETIF_F_IPV6_CSUM
;
4523 if (one
& NETIF_F_GSO
)
4524 one
|= NETIF_F_GSO_SOFTWARE
;
4527 /* If even one device supports robust GSO, enable it for all. */
4528 if (one
& NETIF_F_GSO_ROBUST
)
4529 all
|= NETIF_F_GSO_ROBUST
;
4531 all
&= one
| NETIF_F_LLTX
;
4533 if (!(all
& NETIF_F_ALL_CSUM
))
4535 if (!(all
& NETIF_F_SG
))
4536 all
&= ~NETIF_F_GSO_MASK
;
4540 EXPORT_SYMBOL(netdev_compute_features
);
4542 static struct hlist_head
*netdev_create_hash(void)
4545 struct hlist_head
*hash
;
4547 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
4549 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
4550 INIT_HLIST_HEAD(&hash
[i
]);
4555 /* Initialize per network namespace state */
4556 static int __net_init
netdev_init(struct net
*net
)
4558 INIT_LIST_HEAD(&net
->dev_base_head
);
4560 net
->dev_name_head
= netdev_create_hash();
4561 if (net
->dev_name_head
== NULL
)
4564 net
->dev_index_head
= netdev_create_hash();
4565 if (net
->dev_index_head
== NULL
)
4571 kfree(net
->dev_name_head
);
4576 static void __net_exit
netdev_exit(struct net
*net
)
4578 kfree(net
->dev_name_head
);
4579 kfree(net
->dev_index_head
);
4582 static struct pernet_operations __net_initdata netdev_net_ops
= {
4583 .init
= netdev_init
,
4584 .exit
= netdev_exit
,
4587 static void __net_exit
default_device_exit(struct net
*net
)
4589 struct net_device
*dev
, *next
;
4591 * Push all migratable of the network devices back to the
4592 * initial network namespace
4595 for_each_netdev_safe(net
, dev
, next
) {
4597 char fb_name
[IFNAMSIZ
];
4599 /* Ignore unmoveable devices (i.e. loopback) */
4600 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
4603 /* Push remaing network devices to init_net */
4604 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
4605 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
4607 printk(KERN_EMERG
"%s: failed to move %s to init_net: %d\n",
4608 __func__
, dev
->name
, err
);
4615 static struct pernet_operations __net_initdata default_device_ops
= {
4616 .exit
= default_device_exit
,
4620 * Initialize the DEV module. At boot time this walks the device list and
4621 * unhooks any devices that fail to initialise (normally hardware not
4622 * present) and leaves us with a valid list of present and active devices.
4627 * This is called single threaded during boot, so no need
4628 * to take the rtnl semaphore.
4630 static int __init
net_dev_init(void)
4632 int i
, rc
= -ENOMEM
;
4634 BUG_ON(!dev_boot_phase
);
4636 if (dev_proc_init())
4639 if (netdev_kobject_init())
4642 INIT_LIST_HEAD(&ptype_all
);
4643 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
4644 INIT_LIST_HEAD(&ptype_base
[i
]);
4646 if (register_pernet_subsys(&netdev_net_ops
))
4649 if (register_pernet_device(&default_device_ops
))
4653 * Initialise the packet receive queues.
4656 for_each_possible_cpu(i
) {
4657 struct softnet_data
*queue
;
4659 queue
= &per_cpu(softnet_data
, i
);
4660 skb_queue_head_init(&queue
->input_pkt_queue
);
4661 queue
->completion_queue
= NULL
;
4662 INIT_LIST_HEAD(&queue
->poll_list
);
4664 queue
->backlog
.poll
= process_backlog
;
4665 queue
->backlog
.weight
= weight_p
;
4668 netdev_dma_register();
4672 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
4673 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
4675 hotcpu_notifier(dev_cpu_callback
, 0);
4683 subsys_initcall(net_dev_init
);
4685 EXPORT_SYMBOL(__dev_get_by_index
);
4686 EXPORT_SYMBOL(__dev_get_by_name
);
4687 EXPORT_SYMBOL(__dev_remove_pack
);
4688 EXPORT_SYMBOL(dev_valid_name
);
4689 EXPORT_SYMBOL(dev_add_pack
);
4690 EXPORT_SYMBOL(dev_alloc_name
);
4691 EXPORT_SYMBOL(dev_close
);
4692 EXPORT_SYMBOL(dev_get_by_flags
);
4693 EXPORT_SYMBOL(dev_get_by_index
);
4694 EXPORT_SYMBOL(dev_get_by_name
);
4695 EXPORT_SYMBOL(dev_open
);
4696 EXPORT_SYMBOL(dev_queue_xmit
);
4697 EXPORT_SYMBOL(dev_remove_pack
);
4698 EXPORT_SYMBOL(dev_set_allmulti
);
4699 EXPORT_SYMBOL(dev_set_promiscuity
);
4700 EXPORT_SYMBOL(dev_change_flags
);
4701 EXPORT_SYMBOL(dev_set_mtu
);
4702 EXPORT_SYMBOL(dev_set_mac_address
);
4703 EXPORT_SYMBOL(free_netdev
);
4704 EXPORT_SYMBOL(netdev_boot_setup_check
);
4705 EXPORT_SYMBOL(netdev_set_master
);
4706 EXPORT_SYMBOL(netdev_state_change
);
4707 EXPORT_SYMBOL(netif_receive_skb
);
4708 EXPORT_SYMBOL(netif_rx
);
4709 EXPORT_SYMBOL(register_gifconf
);
4710 EXPORT_SYMBOL(register_netdevice
);
4711 EXPORT_SYMBOL(register_netdevice_notifier
);
4712 EXPORT_SYMBOL(skb_checksum_help
);
4713 EXPORT_SYMBOL(synchronize_net
);
4714 EXPORT_SYMBOL(unregister_netdevice
);
4715 EXPORT_SYMBOL(unregister_netdevice_notifier
);
4716 EXPORT_SYMBOL(net_enable_timestamp
);
4717 EXPORT_SYMBOL(net_disable_timestamp
);
4718 EXPORT_SYMBOL(dev_get_flags
);
4720 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4721 EXPORT_SYMBOL(br_handle_frame_hook
);
4722 EXPORT_SYMBOL(br_fdb_get_hook
);
4723 EXPORT_SYMBOL(br_fdb_put_hook
);
4727 EXPORT_SYMBOL(dev_load
);
4730 EXPORT_PER_CPU_SYMBOL(softnet_data
);