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/notifier.h>
94 #include <linux/skbuff.h>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <net/wext.h>
113 #include <net/iw_handler.h>
114 #include <asm/current.h>
115 #include <linux/audit.h>
116 #include <linux/dmaengine.h>
117 #include <linux/err.h>
118 #include <linux/ctype.h>
119 #include <linux/if_arp.h>
122 * The list of packet types we will receive (as opposed to discard)
123 * and the routines to invoke.
125 * Why 16. Because with 16 the only overlap we get on a hash of the
126 * low nibble of the protocol value is RARP/SNAP/X.25.
128 * NOTE: That is no longer true with the addition of VLAN tags. Not
129 * sure which should go first, but I bet it won't make much
130 * difference if we are running VLANs. The good news is that
131 * this protocol won't be in the list unless compiled in, so
132 * the average user (w/out VLANs) will not be adversely affected.
149 static DEFINE_SPINLOCK(ptype_lock
);
150 static struct list_head ptype_base
[16] __read_mostly
; /* 16 way hashed list */
151 static struct list_head ptype_all __read_mostly
; /* Taps */
153 #ifdef CONFIG_NET_DMA
155 struct dma_client client
;
157 cpumask_t channel_mask
;
158 struct dma_chan
*channels
[NR_CPUS
];
161 static enum dma_state_client
162 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
163 enum dma_state state
);
165 static struct net_dma net_dma
= {
167 .event_callback
= netdev_dma_event
,
173 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
176 * Pure readers hold dev_base_lock for reading.
178 * Writers must hold the rtnl semaphore while they loop through the
179 * dev_base_head list, and hold dev_base_lock for writing when they do the
180 * actual updates. This allows pure readers to access the list even
181 * while a writer is preparing to update it.
183 * To put it another way, dev_base_lock is held for writing only to
184 * protect against pure readers; the rtnl semaphore provides the
185 * protection against other writers.
187 * See, for example usages, register_netdevice() and
188 * unregister_netdevice(), which must be called with the rtnl
191 LIST_HEAD(dev_base_head
);
192 DEFINE_RWLOCK(dev_base_lock
);
194 EXPORT_SYMBOL(dev_base_head
);
195 EXPORT_SYMBOL(dev_base_lock
);
197 #define NETDEV_HASHBITS 8
198 static struct hlist_head dev_name_head
[1<<NETDEV_HASHBITS
];
199 static struct hlist_head dev_index_head
[1<<NETDEV_HASHBITS
];
201 static inline struct hlist_head
*dev_name_hash(const char *name
)
203 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
204 return &dev_name_head
[hash
& ((1<<NETDEV_HASHBITS
)-1)];
207 static inline struct hlist_head
*dev_index_hash(int ifindex
)
209 return &dev_index_head
[ifindex
& ((1<<NETDEV_HASHBITS
)-1)];
216 static RAW_NOTIFIER_HEAD(netdev_chain
);
219 * Device drivers call our routines to queue packets here. We empty the
220 * queue in the local softnet handler.
222 DEFINE_PER_CPU(struct softnet_data
, softnet_data
) = { NULL
};
225 extern int netdev_sysfs_init(void);
226 extern int netdev_register_sysfs(struct net_device
*);
227 extern void netdev_unregister_sysfs(struct net_device
*);
229 #define netdev_sysfs_init() (0)
230 #define netdev_register_sysfs(dev) (0)
231 #define netdev_unregister_sysfs(dev) do { } while(0)
234 #ifdef CONFIG_DEBUG_LOCK_ALLOC
236 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
237 * according to dev->type
239 static const unsigned short netdev_lock_type
[] =
240 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
241 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
242 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
243 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
244 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
245 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
246 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
247 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
248 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
249 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
250 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
251 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
252 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
253 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_VOID
,
256 static const char *netdev_lock_name
[] =
257 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
258 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
259 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
260 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
261 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
262 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
263 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
264 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
265 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
266 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
267 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
268 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
269 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
270 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
273 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
275 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
279 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
280 if (netdev_lock_type
[i
] == dev_type
)
282 /* the last key is used by default */
283 return ARRAY_SIZE(netdev_lock_type
) - 1;
286 static inline void netdev_set_lockdep_class(spinlock_t
*lock
,
287 unsigned short dev_type
)
291 i
= netdev_lock_pos(dev_type
);
292 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
293 netdev_lock_name
[i
]);
296 static inline void netdev_set_lockdep_class(spinlock_t
*lock
,
297 unsigned short dev_type
)
302 /*******************************************************************************
304 Protocol management and registration routines
306 *******************************************************************************/
309 * Add a protocol ID to the list. Now that the input handler is
310 * smarter we can dispense with all the messy stuff that used to be
313 * BEWARE!!! Protocol handlers, mangling input packets,
314 * MUST BE last in hash buckets and checking protocol handlers
315 * MUST start from promiscuous ptype_all chain in net_bh.
316 * It is true now, do not change it.
317 * Explanation follows: if protocol handler, mangling packet, will
318 * be the first on list, it is not able to sense, that packet
319 * is cloned and should be copied-on-write, so that it will
320 * change it and subsequent readers will get broken packet.
325 * dev_add_pack - add packet handler
326 * @pt: packet type declaration
328 * Add a protocol handler to the networking stack. The passed &packet_type
329 * is linked into kernel lists and may not be freed until it has been
330 * removed from the kernel lists.
332 * This call does not sleep therefore it can not
333 * guarantee all CPU's that are in middle of receiving packets
334 * will see the new packet type (until the next received packet).
337 void dev_add_pack(struct packet_type
*pt
)
341 spin_lock_bh(&ptype_lock
);
342 if (pt
->type
== htons(ETH_P_ALL
))
343 list_add_rcu(&pt
->list
, &ptype_all
);
345 hash
= ntohs(pt
->type
) & 15;
346 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
348 spin_unlock_bh(&ptype_lock
);
352 * __dev_remove_pack - remove packet handler
353 * @pt: packet type declaration
355 * Remove a protocol handler that was previously added to the kernel
356 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
357 * from the kernel lists and can be freed or reused once this function
360 * The packet type might still be in use by receivers
361 * and must not be freed until after all the CPU's have gone
362 * through a quiescent state.
364 void __dev_remove_pack(struct packet_type
*pt
)
366 struct list_head
*head
;
367 struct packet_type
*pt1
;
369 spin_lock_bh(&ptype_lock
);
371 if (pt
->type
== htons(ETH_P_ALL
))
374 head
= &ptype_base
[ntohs(pt
->type
) & 15];
376 list_for_each_entry(pt1
, head
, list
) {
378 list_del_rcu(&pt
->list
);
383 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
385 spin_unlock_bh(&ptype_lock
);
388 * dev_remove_pack - remove packet handler
389 * @pt: packet type declaration
391 * Remove a protocol handler that was previously added to the kernel
392 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
393 * from the kernel lists and can be freed or reused once this function
396 * This call sleeps to guarantee that no CPU is looking at the packet
399 void dev_remove_pack(struct packet_type
*pt
)
401 __dev_remove_pack(pt
);
406 /******************************************************************************
408 Device Boot-time Settings Routines
410 *******************************************************************************/
412 /* Boot time configuration table */
413 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
416 * netdev_boot_setup_add - add new setup entry
417 * @name: name of the device
418 * @map: configured settings for the device
420 * Adds new setup entry to the dev_boot_setup list. The function
421 * returns 0 on error and 1 on success. This is a generic routine to
424 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
426 struct netdev_boot_setup
*s
;
430 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
431 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
432 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
433 strcpy(s
[i
].name
, name
);
434 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
439 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
443 * netdev_boot_setup_check - check boot time settings
444 * @dev: the netdevice
446 * Check boot time settings for the device.
447 * The found settings are set for the device to be used
448 * later in the device probing.
449 * Returns 0 if no settings found, 1 if they are.
451 int netdev_boot_setup_check(struct net_device
*dev
)
453 struct netdev_boot_setup
*s
= dev_boot_setup
;
456 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
457 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
458 !strncmp(dev
->name
, s
[i
].name
, strlen(s
[i
].name
))) {
459 dev
->irq
= s
[i
].map
.irq
;
460 dev
->base_addr
= s
[i
].map
.base_addr
;
461 dev
->mem_start
= s
[i
].map
.mem_start
;
462 dev
->mem_end
= s
[i
].map
.mem_end
;
471 * netdev_boot_base - get address from boot time settings
472 * @prefix: prefix for network device
473 * @unit: id for network device
475 * Check boot time settings for the base address of device.
476 * The found settings are set for the device to be used
477 * later in the device probing.
478 * Returns 0 if no settings found.
480 unsigned long netdev_boot_base(const char *prefix
, int unit
)
482 const struct netdev_boot_setup
*s
= dev_boot_setup
;
486 sprintf(name
, "%s%d", prefix
, unit
);
489 * If device already registered then return base of 1
490 * to indicate not to probe for this interface
492 if (__dev_get_by_name(name
))
495 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
496 if (!strcmp(name
, s
[i
].name
))
497 return s
[i
].map
.base_addr
;
502 * Saves at boot time configured settings for any netdevice.
504 int __init
netdev_boot_setup(char *str
)
509 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
514 memset(&map
, 0, sizeof(map
));
518 map
.base_addr
= ints
[2];
520 map
.mem_start
= ints
[3];
522 map
.mem_end
= ints
[4];
524 /* Add new entry to the list */
525 return netdev_boot_setup_add(str
, &map
);
528 __setup("netdev=", netdev_boot_setup
);
530 /*******************************************************************************
532 Device Interface Subroutines
534 *******************************************************************************/
537 * __dev_get_by_name - find a device by its name
538 * @name: name to find
540 * Find an interface by name. Must be called under RTNL semaphore
541 * or @dev_base_lock. If the name is found a pointer to the device
542 * is returned. If the name is not found then %NULL is returned. The
543 * reference counters are not incremented so the caller must be
544 * careful with locks.
547 struct net_device
*__dev_get_by_name(const char *name
)
549 struct hlist_node
*p
;
551 hlist_for_each(p
, dev_name_hash(name
)) {
552 struct net_device
*dev
553 = hlist_entry(p
, struct net_device
, name_hlist
);
554 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
561 * dev_get_by_name - find a device by its name
562 * @name: name to find
564 * Find an interface by name. This can be called from any
565 * context and does its own locking. The returned handle has
566 * the usage count incremented and the caller must use dev_put() to
567 * release it when it is no longer needed. %NULL is returned if no
568 * matching device is found.
571 struct net_device
*dev_get_by_name(const char *name
)
573 struct net_device
*dev
;
575 read_lock(&dev_base_lock
);
576 dev
= __dev_get_by_name(name
);
579 read_unlock(&dev_base_lock
);
584 * __dev_get_by_index - find a device by its ifindex
585 * @ifindex: index of device
587 * Search for an interface by index. Returns %NULL if the device
588 * is not found or a pointer to the device. The device has not
589 * had its reference counter increased so the caller must be careful
590 * about locking. The caller must hold either the RTNL semaphore
594 struct net_device
*__dev_get_by_index(int ifindex
)
596 struct hlist_node
*p
;
598 hlist_for_each(p
, dev_index_hash(ifindex
)) {
599 struct net_device
*dev
600 = hlist_entry(p
, struct net_device
, index_hlist
);
601 if (dev
->ifindex
== ifindex
)
609 * dev_get_by_index - find a device by its ifindex
610 * @ifindex: index of device
612 * Search for an interface by index. Returns NULL if the device
613 * is not found or a pointer to the device. The device returned has
614 * had a reference added and the pointer is safe until the user calls
615 * dev_put to indicate they have finished with it.
618 struct net_device
*dev_get_by_index(int ifindex
)
620 struct net_device
*dev
;
622 read_lock(&dev_base_lock
);
623 dev
= __dev_get_by_index(ifindex
);
626 read_unlock(&dev_base_lock
);
631 * dev_getbyhwaddr - find a device by its hardware address
632 * @type: media type of device
633 * @ha: hardware address
635 * Search for an interface by MAC address. Returns NULL if the device
636 * is not found or a pointer to the device. The caller must hold the
637 * rtnl semaphore. The returned device has not had its ref count increased
638 * and the caller must therefore be careful about locking
641 * If the API was consistent this would be __dev_get_by_hwaddr
644 struct net_device
*dev_getbyhwaddr(unsigned short type
, char *ha
)
646 struct net_device
*dev
;
651 if (dev
->type
== type
&&
652 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
658 EXPORT_SYMBOL(dev_getbyhwaddr
);
660 struct net_device
*__dev_getfirstbyhwtype(unsigned short type
)
662 struct net_device
*dev
;
666 if (dev
->type
== type
)
672 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
674 struct net_device
*dev_getfirstbyhwtype(unsigned short type
)
676 struct net_device
*dev
;
679 dev
= __dev_getfirstbyhwtype(type
);
686 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
689 * dev_get_by_flags - find any device with given flags
690 * @if_flags: IFF_* values
691 * @mask: bitmask of bits in if_flags to check
693 * Search for any interface with the given flags. Returns NULL if a device
694 * is not found or a pointer to the device. The device returned has
695 * had a reference added and the pointer is safe until the user calls
696 * dev_put to indicate they have finished with it.
699 struct net_device
* dev_get_by_flags(unsigned short if_flags
, unsigned short mask
)
701 struct net_device
*dev
, *ret
;
704 read_lock(&dev_base_lock
);
705 for_each_netdev(dev
) {
706 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
712 read_unlock(&dev_base_lock
);
717 * dev_valid_name - check if name is okay for network device
720 * Network device names need to be valid file names to
721 * to allow sysfs to work. We also disallow any kind of
724 int dev_valid_name(const char *name
)
728 if (strlen(name
) >= IFNAMSIZ
)
730 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
734 if (*name
== '/' || isspace(*name
))
742 * dev_alloc_name - allocate a name for a device
744 * @name: name format string
746 * Passed a format string - eg "lt%d" it will try and find a suitable
747 * id. It scans list of devices to build up a free map, then chooses
748 * the first empty slot. The caller must hold the dev_base or rtnl lock
749 * while allocating the name and adding the device in order to avoid
751 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
752 * Returns the number of the unit assigned or a negative errno code.
755 int dev_alloc_name(struct net_device
*dev
, const char *name
)
760 const int max_netdevices
= 8*PAGE_SIZE
;
762 struct net_device
*d
;
764 p
= strnchr(name
, IFNAMSIZ
-1, '%');
767 * Verify the string as this thing may have come from
768 * the user. There must be either one "%d" and no other "%"
771 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
774 /* Use one page as a bit array of possible slots */
775 inuse
= (long *) get_zeroed_page(GFP_ATOMIC
);
780 if (!sscanf(d
->name
, name
, &i
))
782 if (i
< 0 || i
>= max_netdevices
)
785 /* avoid cases where sscanf is not exact inverse of printf */
786 snprintf(buf
, sizeof(buf
), name
, i
);
787 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
791 i
= find_first_zero_bit(inuse
, max_netdevices
);
792 free_page((unsigned long) inuse
);
795 snprintf(buf
, sizeof(buf
), name
, i
);
796 if (!__dev_get_by_name(buf
)) {
797 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
801 /* It is possible to run out of possible slots
802 * when the name is long and there isn't enough space left
803 * for the digits, or if all bits are used.
810 * dev_change_name - change name of a device
812 * @newname: name (or format string) must be at least IFNAMSIZ
814 * Change name of a device, can pass format strings "eth%d".
817 int dev_change_name(struct net_device
*dev
, char *newname
)
823 if (dev
->flags
& IFF_UP
)
826 if (!dev_valid_name(newname
))
829 if (strchr(newname
, '%')) {
830 err
= dev_alloc_name(dev
, newname
);
833 strcpy(newname
, dev
->name
);
835 else if (__dev_get_by_name(newname
))
838 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
840 device_rename(&dev
->dev
, dev
->name
);
841 hlist_del(&dev
->name_hlist
);
842 hlist_add_head(&dev
->name_hlist
, dev_name_hash(dev
->name
));
843 raw_notifier_call_chain(&netdev_chain
, NETDEV_CHANGENAME
, dev
);
849 * netdev_features_change - device changes features
850 * @dev: device to cause notification
852 * Called to indicate a device has changed features.
854 void netdev_features_change(struct net_device
*dev
)
856 raw_notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
858 EXPORT_SYMBOL(netdev_features_change
);
861 * netdev_state_change - device changes state
862 * @dev: device to cause notification
864 * Called to indicate a device has changed state. This function calls
865 * the notifier chains for netdev_chain and sends a NEWLINK message
866 * to the routing socket.
868 void netdev_state_change(struct net_device
*dev
)
870 if (dev
->flags
& IFF_UP
) {
871 raw_notifier_call_chain(&netdev_chain
,
873 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
878 * dev_load - load a network module
879 * @name: name of interface
881 * If a network interface is not present and the process has suitable
882 * privileges this function loads the module. If module loading is not
883 * available in this kernel then it becomes a nop.
886 void dev_load(const char *name
)
888 struct net_device
*dev
;
890 read_lock(&dev_base_lock
);
891 dev
= __dev_get_by_name(name
);
892 read_unlock(&dev_base_lock
);
894 if (!dev
&& capable(CAP_SYS_MODULE
))
895 request_module("%s", name
);
898 static int default_rebuild_header(struct sk_buff
*skb
)
900 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
901 skb
->dev
? skb
->dev
->name
: "NULL!!!");
907 * dev_open - prepare an interface for use.
908 * @dev: device to open
910 * Takes a device from down to up state. The device's private open
911 * function is invoked and then the multicast lists are loaded. Finally
912 * the device is moved into the up state and a %NETDEV_UP message is
913 * sent to the netdev notifier chain.
915 * Calling this function on an active interface is a nop. On a failure
916 * a negative errno code is returned.
918 int dev_open(struct net_device
*dev
)
926 if (dev
->flags
& IFF_UP
)
930 * Is it even present?
932 if (!netif_device_present(dev
))
936 * Call device private open method
938 set_bit(__LINK_STATE_START
, &dev
->state
);
940 ret
= dev
->open(dev
);
942 clear_bit(__LINK_STATE_START
, &dev
->state
);
946 * If it went open OK then:
953 dev
->flags
|= IFF_UP
;
956 * Initialize multicasting status
958 dev_set_rx_mode(dev
);
961 * Wakeup transmit queue engine
966 * ... and announce new interface.
968 raw_notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
974 * dev_close - shutdown an interface.
975 * @dev: device to shutdown
977 * This function moves an active device into down state. A
978 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
979 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
982 int dev_close(struct net_device
*dev
)
984 if (!(dev
->flags
& IFF_UP
))
988 * Tell people we are going down, so that they can
989 * prepare to death, when device is still operating.
991 raw_notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
995 clear_bit(__LINK_STATE_START
, &dev
->state
);
997 /* Synchronize to scheduled poll. We cannot touch poll list,
998 * it can be even on different cpu. So just clear netif_running(),
999 * and wait when poll really will happen. Actually, the best place
1000 * for this is inside dev->stop() after device stopped its irq
1001 * engine, but this requires more changes in devices. */
1003 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1004 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
1010 * Call the device specific close. This cannot fail.
1011 * Only if device is UP
1013 * We allow it to be called even after a DETACH hot-plug
1020 * Device is now down.
1023 dev
->flags
&= ~IFF_UP
;
1026 * Tell people we are down
1028 raw_notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
1035 * Device change register/unregister. These are not inline or static
1036 * as we export them to the world.
1040 * register_netdevice_notifier - register a network notifier block
1043 * Register a notifier to be called when network device events occur.
1044 * The notifier passed is linked into the kernel structures and must
1045 * not be reused until it has been unregistered. A negative errno code
1046 * is returned on a failure.
1048 * When registered all registration and up events are replayed
1049 * to the new notifier to allow device to have a race free
1050 * view of the network device list.
1053 int register_netdevice_notifier(struct notifier_block
*nb
)
1055 struct net_device
*dev
;
1059 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1061 for_each_netdev(dev
) {
1062 nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1064 if (dev
->flags
& IFF_UP
)
1065 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1073 * unregister_netdevice_notifier - unregister a network notifier block
1076 * Unregister a notifier previously registered by
1077 * register_netdevice_notifier(). The notifier is unlinked into the
1078 * kernel structures and may then be reused. A negative errno code
1079 * is returned on a failure.
1082 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1087 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1093 * call_netdevice_notifiers - call all network notifier blocks
1094 * @val: value passed unmodified to notifier function
1095 * @v: pointer passed unmodified to notifier function
1097 * Call all network notifier blocks. Parameters and return value
1098 * are as for raw_notifier_call_chain().
1101 int call_netdevice_notifiers(unsigned long val
, void *v
)
1103 return raw_notifier_call_chain(&netdev_chain
, val
, v
);
1106 /* When > 0 there are consumers of rx skb time stamps */
1107 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1109 void net_enable_timestamp(void)
1111 atomic_inc(&netstamp_needed
);
1114 void net_disable_timestamp(void)
1116 atomic_dec(&netstamp_needed
);
1119 static inline void net_timestamp(struct sk_buff
*skb
)
1121 if (atomic_read(&netstamp_needed
))
1122 __net_timestamp(skb
);
1124 skb
->tstamp
.tv64
= 0;
1128 * Support routine. Sends outgoing frames to any network
1129 * taps currently in use.
1132 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1134 struct packet_type
*ptype
;
1139 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1140 /* Never send packets back to the socket
1141 * they originated from - MvS (miquels@drinkel.ow.org)
1143 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1144 (ptype
->af_packet_priv
== NULL
||
1145 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1146 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1150 /* skb->nh should be correctly
1151 set by sender, so that the second statement is
1152 just protection against buggy protocols.
1154 skb_reset_mac_header(skb2
);
1156 if (skb_network_header(skb2
) < skb2
->data
||
1157 skb2
->network_header
> skb2
->tail
) {
1158 if (net_ratelimit())
1159 printk(KERN_CRIT
"protocol %04x is "
1161 skb2
->protocol
, dev
->name
);
1162 skb_reset_network_header(skb2
);
1165 skb2
->transport_header
= skb2
->network_header
;
1166 skb2
->pkt_type
= PACKET_OUTGOING
;
1167 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1174 void __netif_schedule(struct net_device
*dev
)
1176 if (!test_and_set_bit(__LINK_STATE_SCHED
, &dev
->state
)) {
1177 unsigned long flags
;
1178 struct softnet_data
*sd
;
1180 local_irq_save(flags
);
1181 sd
= &__get_cpu_var(softnet_data
);
1182 dev
->next_sched
= sd
->output_queue
;
1183 sd
->output_queue
= dev
;
1184 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1185 local_irq_restore(flags
);
1188 EXPORT_SYMBOL(__netif_schedule
);
1190 void __netif_rx_schedule(struct net_device
*dev
)
1192 unsigned long flags
;
1194 local_irq_save(flags
);
1196 list_add_tail(&dev
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
1198 dev
->quota
+= dev
->weight
;
1200 dev
->quota
= dev
->weight
;
1201 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1202 local_irq_restore(flags
);
1204 EXPORT_SYMBOL(__netif_rx_schedule
);
1206 void dev_kfree_skb_any(struct sk_buff
*skb
)
1208 if (in_irq() || irqs_disabled())
1209 dev_kfree_skb_irq(skb
);
1213 EXPORT_SYMBOL(dev_kfree_skb_any
);
1217 void netif_device_detach(struct net_device
*dev
)
1219 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1220 netif_running(dev
)) {
1221 netif_stop_queue(dev
);
1224 EXPORT_SYMBOL(netif_device_detach
);
1226 void netif_device_attach(struct net_device
*dev
)
1228 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1229 netif_running(dev
)) {
1230 netif_wake_queue(dev
);
1231 __netdev_watchdog_up(dev
);
1234 EXPORT_SYMBOL(netif_device_attach
);
1238 * Invalidate hardware checksum when packet is to be mangled, and
1239 * complete checksum manually on outgoing path.
1241 int skb_checksum_help(struct sk_buff
*skb
)
1244 int ret
= 0, offset
;
1246 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1247 goto out_set_summed
;
1249 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1250 /* Let GSO fix up the checksum. */
1251 goto out_set_summed
;
1254 if (skb_cloned(skb
)) {
1255 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1260 offset
= skb
->csum_start
- skb_headroom(skb
);
1261 BUG_ON(offset
> (int)skb
->len
);
1262 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1264 offset
= skb_headlen(skb
) - offset
;
1265 BUG_ON(offset
<= 0);
1266 BUG_ON(skb
->csum_offset
+ 2 > offset
);
1268 *(__sum16
*)(skb
->head
+ skb
->csum_start
+ skb
->csum_offset
) =
1271 skb
->ip_summed
= CHECKSUM_NONE
;
1277 * skb_gso_segment - Perform segmentation on skb.
1278 * @skb: buffer to segment
1279 * @features: features for the output path (see dev->features)
1281 * This function segments the given skb and returns a list of segments.
1283 * It may return NULL if the skb requires no segmentation. This is
1284 * only possible when GSO is used for verifying header integrity.
1286 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1288 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1289 struct packet_type
*ptype
;
1290 __be16 type
= skb
->protocol
;
1293 BUG_ON(skb_shinfo(skb
)->frag_list
);
1295 skb_reset_mac_header(skb
);
1296 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1297 __skb_pull(skb
, skb
->mac_len
);
1299 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1300 if (skb_header_cloned(skb
) &&
1301 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1302 return ERR_PTR(err
);
1306 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
) & 15], list
) {
1307 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1308 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1309 err
= ptype
->gso_send_check(skb
);
1310 segs
= ERR_PTR(err
);
1311 if (err
|| skb_gso_ok(skb
, features
))
1313 __skb_push(skb
, (skb
->data
-
1314 skb_network_header(skb
)));
1316 segs
= ptype
->gso_segment(skb
, features
);
1322 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1327 EXPORT_SYMBOL(skb_gso_segment
);
1329 /* Take action when hardware reception checksum errors are detected. */
1331 void netdev_rx_csum_fault(struct net_device
*dev
)
1333 if (net_ratelimit()) {
1334 printk(KERN_ERR
"%s: hw csum failure.\n",
1335 dev
? dev
->name
: "<unknown>");
1339 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1342 /* Actually, we should eliminate this check as soon as we know, that:
1343 * 1. IOMMU is present and allows to map all the memory.
1344 * 2. No high memory really exists on this machine.
1347 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1349 #ifdef CONFIG_HIGHMEM
1352 if (dev
->features
& NETIF_F_HIGHDMA
)
1355 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1356 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1364 void (*destructor
)(struct sk_buff
*skb
);
1367 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1369 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1371 struct dev_gso_cb
*cb
;
1374 struct sk_buff
*nskb
= skb
->next
;
1376 skb
->next
= nskb
->next
;
1379 } while (skb
->next
);
1381 cb
= DEV_GSO_CB(skb
);
1383 cb
->destructor(skb
);
1387 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1388 * @skb: buffer to segment
1390 * This function segments the given skb and stores the list of segments
1393 static int dev_gso_segment(struct sk_buff
*skb
)
1395 struct net_device
*dev
= skb
->dev
;
1396 struct sk_buff
*segs
;
1397 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1400 segs
= skb_gso_segment(skb
, features
);
1402 /* Verifying header integrity only. */
1406 if (unlikely(IS_ERR(segs
)))
1407 return PTR_ERR(segs
);
1410 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1411 skb
->destructor
= dev_gso_skb_destructor
;
1416 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1418 if (likely(!skb
->next
)) {
1419 if (!list_empty(&ptype_all
))
1420 dev_queue_xmit_nit(skb
, dev
);
1422 if (netif_needs_gso(dev
, skb
)) {
1423 if (unlikely(dev_gso_segment(skb
)))
1429 return dev
->hard_start_xmit(skb
, dev
);
1434 struct sk_buff
*nskb
= skb
->next
;
1437 skb
->next
= nskb
->next
;
1439 rc
= dev
->hard_start_xmit(nskb
, dev
);
1441 nskb
->next
= skb
->next
;
1445 if (unlikely((netif_queue_stopped(dev
) ||
1446 netif_subqueue_stopped(dev
, skb
->queue_mapping
)) &&
1448 return NETDEV_TX_BUSY
;
1449 } while (skb
->next
);
1451 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1458 #define HARD_TX_LOCK(dev, cpu) { \
1459 if ((dev->features & NETIF_F_LLTX) == 0) { \
1460 netif_tx_lock(dev); \
1464 #define HARD_TX_UNLOCK(dev) { \
1465 if ((dev->features & NETIF_F_LLTX) == 0) { \
1466 netif_tx_unlock(dev); \
1471 * dev_queue_xmit - transmit a buffer
1472 * @skb: buffer to transmit
1474 * Queue a buffer for transmission to a network device. The caller must
1475 * have set the device and priority and built the buffer before calling
1476 * this function. The function can be called from an interrupt.
1478 * A negative errno code is returned on a failure. A success does not
1479 * guarantee the frame will be transmitted as it may be dropped due
1480 * to congestion or traffic shaping.
1482 * -----------------------------------------------------------------------------------
1483 * I notice this method can also return errors from the queue disciplines,
1484 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1487 * Regardless of the return value, the skb is consumed, so it is currently
1488 * difficult to retry a send to this method. (You can bump the ref count
1489 * before sending to hold a reference for retry if you are careful.)
1491 * When calling this method, interrupts MUST be enabled. This is because
1492 * the BH enable code must have IRQs enabled so that it will not deadlock.
1496 int dev_queue_xmit(struct sk_buff
*skb
)
1498 struct net_device
*dev
= skb
->dev
;
1502 /* GSO will handle the following emulations directly. */
1503 if (netif_needs_gso(dev
, skb
))
1506 if (skb_shinfo(skb
)->frag_list
&&
1507 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1508 __skb_linearize(skb
))
1511 /* Fragmented skb is linearized if device does not support SG,
1512 * or if at least one of fragments is in highmem and device
1513 * does not support DMA from it.
1515 if (skb_shinfo(skb
)->nr_frags
&&
1516 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1517 __skb_linearize(skb
))
1520 /* If packet is not checksummed and device does not support
1521 * checksumming for this protocol, complete checksumming here.
1523 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1524 skb_set_transport_header(skb
, skb
->csum_start
-
1527 if (!(dev
->features
& NETIF_F_GEN_CSUM
) &&
1528 !((dev
->features
& NETIF_F_IP_CSUM
) &&
1529 skb
->protocol
== htons(ETH_P_IP
)) &&
1530 !((dev
->features
& NETIF_F_IPV6_CSUM
) &&
1531 skb
->protocol
== htons(ETH_P_IPV6
)))
1532 if (skb_checksum_help(skb
))
1537 spin_lock_prefetch(&dev
->queue_lock
);
1539 /* Disable soft irqs for various locks below. Also
1540 * stops preemption for RCU.
1544 /* Updates of qdisc are serialized by queue_lock.
1545 * The struct Qdisc which is pointed to by qdisc is now a
1546 * rcu structure - it may be accessed without acquiring
1547 * a lock (but the structure may be stale.) The freeing of the
1548 * qdisc will be deferred until it's known that there are no
1549 * more references to it.
1551 * If the qdisc has an enqueue function, we still need to
1552 * hold the queue_lock before calling it, since queue_lock
1553 * also serializes access to the device queue.
1556 q
= rcu_dereference(dev
->qdisc
);
1557 #ifdef CONFIG_NET_CLS_ACT
1558 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1561 /* Grab device queue */
1562 spin_lock(&dev
->queue_lock
);
1565 /* reset queue_mapping to zero */
1566 skb
->queue_mapping
= 0;
1567 rc
= q
->enqueue(skb
, q
);
1569 spin_unlock(&dev
->queue_lock
);
1571 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1574 spin_unlock(&dev
->queue_lock
);
1577 /* The device has no queue. Common case for software devices:
1578 loopback, all the sorts of tunnels...
1580 Really, it is unlikely that netif_tx_lock protection is necessary
1581 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1583 However, it is possible, that they rely on protection
1586 Check this and shot the lock. It is not prone from deadlocks.
1587 Either shot noqueue qdisc, it is even simpler 8)
1589 if (dev
->flags
& IFF_UP
) {
1590 int cpu
= smp_processor_id(); /* ok because BHs are off */
1592 if (dev
->xmit_lock_owner
!= cpu
) {
1594 HARD_TX_LOCK(dev
, cpu
);
1596 if (!netif_queue_stopped(dev
) &&
1597 !netif_subqueue_stopped(dev
, skb
->queue_mapping
)) {
1599 if (!dev_hard_start_xmit(skb
, dev
)) {
1600 HARD_TX_UNLOCK(dev
);
1604 HARD_TX_UNLOCK(dev
);
1605 if (net_ratelimit())
1606 printk(KERN_CRIT
"Virtual device %s asks to "
1607 "queue packet!\n", dev
->name
);
1609 /* Recursion is detected! It is possible,
1611 if (net_ratelimit())
1612 printk(KERN_CRIT
"Dead loop on virtual device "
1613 "%s, fix it urgently!\n", dev
->name
);
1618 rcu_read_unlock_bh();
1624 rcu_read_unlock_bh();
1629 /*=======================================================================
1631 =======================================================================*/
1633 int netdev_max_backlog __read_mostly
= 1000;
1634 int netdev_budget __read_mostly
= 300;
1635 int weight_p __read_mostly
= 64; /* old backlog weight */
1637 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1641 * netif_rx - post buffer to the network code
1642 * @skb: buffer to post
1644 * This function receives a packet from a device driver and queues it for
1645 * the upper (protocol) levels to process. It always succeeds. The buffer
1646 * may be dropped during processing for congestion control or by the
1650 * NET_RX_SUCCESS (no congestion)
1651 * NET_RX_CN_LOW (low congestion)
1652 * NET_RX_CN_MOD (moderate congestion)
1653 * NET_RX_CN_HIGH (high congestion)
1654 * NET_RX_DROP (packet was dropped)
1658 int netif_rx(struct sk_buff
*skb
)
1660 struct softnet_data
*queue
;
1661 unsigned long flags
;
1663 /* if netpoll wants it, pretend we never saw it */
1664 if (netpoll_rx(skb
))
1667 if (!skb
->tstamp
.tv64
)
1671 * The code is rearranged so that the path is the most
1672 * short when CPU is congested, but is still operating.
1674 local_irq_save(flags
);
1675 queue
= &__get_cpu_var(softnet_data
);
1677 __get_cpu_var(netdev_rx_stat
).total
++;
1678 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1679 if (queue
->input_pkt_queue
.qlen
) {
1682 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1683 local_irq_restore(flags
);
1684 return NET_RX_SUCCESS
;
1687 netif_rx_schedule(&queue
->backlog_dev
);
1691 __get_cpu_var(netdev_rx_stat
).dropped
++;
1692 local_irq_restore(flags
);
1698 int netif_rx_ni(struct sk_buff
*skb
)
1703 err
= netif_rx(skb
);
1704 if (local_softirq_pending())
1711 EXPORT_SYMBOL(netif_rx_ni
);
1713 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1715 struct net_device
*dev
= skb
->dev
;
1718 if (skb_bond_should_drop(skb
)) {
1722 skb
->dev
= dev
->master
;
1728 static void net_tx_action(struct softirq_action
*h
)
1730 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1732 if (sd
->completion_queue
) {
1733 struct sk_buff
*clist
;
1735 local_irq_disable();
1736 clist
= sd
->completion_queue
;
1737 sd
->completion_queue
= NULL
;
1741 struct sk_buff
*skb
= clist
;
1742 clist
= clist
->next
;
1744 BUG_TRAP(!atomic_read(&skb
->users
));
1749 if (sd
->output_queue
) {
1750 struct net_device
*head
;
1752 local_irq_disable();
1753 head
= sd
->output_queue
;
1754 sd
->output_queue
= NULL
;
1758 struct net_device
*dev
= head
;
1759 head
= head
->next_sched
;
1761 smp_mb__before_clear_bit();
1762 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1764 if (spin_trylock(&dev
->queue_lock
)) {
1766 spin_unlock(&dev
->queue_lock
);
1768 netif_schedule(dev
);
1774 static inline int deliver_skb(struct sk_buff
*skb
,
1775 struct packet_type
*pt_prev
,
1776 struct net_device
*orig_dev
)
1778 atomic_inc(&skb
->users
);
1779 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1782 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1783 /* These hooks defined here for ATM */
1785 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1786 unsigned char *addr
);
1787 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
1790 * If bridge module is loaded call bridging hook.
1791 * returns NULL if packet was consumed.
1793 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
1794 struct sk_buff
*skb
) __read_mostly
;
1795 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
1796 struct packet_type
**pt_prev
, int *ret
,
1797 struct net_device
*orig_dev
)
1799 struct net_bridge_port
*port
;
1801 if (skb
->pkt_type
== PACKET_LOOPBACK
||
1802 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
1806 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1810 return br_handle_frame_hook(port
, skb
);
1813 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1816 #ifdef CONFIG_NET_CLS_ACT
1817 /* TODO: Maybe we should just force sch_ingress to be compiled in
1818 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1819 * a compare and 2 stores extra right now if we dont have it on
1820 * but have CONFIG_NET_CLS_ACT
1821 * NOTE: This doesnt stop any functionality; if you dont have
1822 * the ingress scheduler, you just cant add policies on ingress.
1825 static int ing_filter(struct sk_buff
*skb
)
1828 struct net_device
*dev
= skb
->dev
;
1829 int result
= TC_ACT_OK
;
1831 if (dev
->qdisc_ingress
) {
1832 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1833 if (MAX_RED_LOOP
< ttl
++) {
1834 printk(KERN_WARNING
"Redir loop detected Dropping packet (%d->%d)\n",
1835 skb
->iif
, skb
->dev
->ifindex
);
1839 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1841 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1843 spin_lock(&dev
->ingress_lock
);
1844 if ((q
= dev
->qdisc_ingress
) != NULL
)
1845 result
= q
->enqueue(skb
, q
);
1846 spin_unlock(&dev
->ingress_lock
);
1854 int netif_receive_skb(struct sk_buff
*skb
)
1856 struct packet_type
*ptype
, *pt_prev
;
1857 struct net_device
*orig_dev
;
1858 int ret
= NET_RX_DROP
;
1861 /* if we've gotten here through NAPI, check netpoll */
1862 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1865 if (!skb
->tstamp
.tv64
)
1869 skb
->iif
= skb
->dev
->ifindex
;
1871 orig_dev
= skb_bond(skb
);
1876 __get_cpu_var(netdev_rx_stat
).total
++;
1878 skb_reset_network_header(skb
);
1879 skb_reset_transport_header(skb
);
1880 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1886 #ifdef CONFIG_NET_CLS_ACT
1887 if (skb
->tc_verd
& TC_NCLS
) {
1888 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1893 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1894 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1896 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1901 #ifdef CONFIG_NET_CLS_ACT
1903 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1904 pt_prev
= NULL
; /* noone else should process this after*/
1906 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1909 ret
= ing_filter(skb
);
1911 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
1920 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
1924 type
= skb
->protocol
;
1925 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
1926 if (ptype
->type
== type
&&
1927 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
1929 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1935 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1938 /* Jamal, now you will not able to escape explaining
1939 * me how you were going to use this. :-)
1949 static int process_backlog(struct net_device
*backlog_dev
, int *budget
)
1952 int quota
= min(backlog_dev
->quota
, *budget
);
1953 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1954 unsigned long start_time
= jiffies
;
1956 backlog_dev
->weight
= weight_p
;
1958 struct sk_buff
*skb
;
1959 struct net_device
*dev
;
1961 local_irq_disable();
1962 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
1969 netif_receive_skb(skb
);
1975 if (work
>= quota
|| jiffies
- start_time
> 1)
1980 backlog_dev
->quota
-= work
;
1985 backlog_dev
->quota
-= work
;
1988 list_del(&backlog_dev
->poll_list
);
1989 smp_mb__before_clear_bit();
1990 netif_poll_enable(backlog_dev
);
1996 static void net_rx_action(struct softirq_action
*h
)
1998 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1999 unsigned long start_time
= jiffies
;
2000 int budget
= netdev_budget
;
2003 local_irq_disable();
2005 while (!list_empty(&queue
->poll_list
)) {
2006 struct net_device
*dev
;
2008 if (budget
<= 0 || jiffies
- start_time
> 1)
2013 dev
= list_entry(queue
->poll_list
.next
,
2014 struct net_device
, poll_list
);
2015 have
= netpoll_poll_lock(dev
);
2017 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
2018 netpoll_poll_unlock(have
);
2019 local_irq_disable();
2020 list_move_tail(&dev
->poll_list
, &queue
->poll_list
);
2022 dev
->quota
+= dev
->weight
;
2024 dev
->quota
= dev
->weight
;
2026 netpoll_poll_unlock(have
);
2028 local_irq_disable();
2033 #ifdef CONFIG_NET_DMA
2035 * There may not be any more sk_buffs coming right now, so push
2036 * any pending DMA copies to hardware
2038 if (!cpus_empty(net_dma
.channel_mask
)) {
2040 for_each_cpu_mask(chan_idx
, net_dma
.channel_mask
) {
2041 struct dma_chan
*chan
= net_dma
.channels
[chan_idx
];
2043 dma_async_memcpy_issue_pending(chan
);
2050 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2051 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2055 static gifconf_func_t
* gifconf_list
[NPROTO
];
2058 * register_gifconf - register a SIOCGIF handler
2059 * @family: Address family
2060 * @gifconf: Function handler
2062 * Register protocol dependent address dumping routines. The handler
2063 * that is passed must not be freed or reused until it has been replaced
2064 * by another handler.
2066 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2068 if (family
>= NPROTO
)
2070 gifconf_list
[family
] = gifconf
;
2076 * Map an interface index to its name (SIOCGIFNAME)
2080 * We need this ioctl for efficient implementation of the
2081 * if_indextoname() function required by the IPv6 API. Without
2082 * it, we would have to search all the interfaces to find a
2086 static int dev_ifname(struct ifreq __user
*arg
)
2088 struct net_device
*dev
;
2092 * Fetch the caller's info block.
2095 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2098 read_lock(&dev_base_lock
);
2099 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
2101 read_unlock(&dev_base_lock
);
2105 strcpy(ifr
.ifr_name
, dev
->name
);
2106 read_unlock(&dev_base_lock
);
2108 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2114 * Perform a SIOCGIFCONF call. This structure will change
2115 * size eventually, and there is nothing I can do about it.
2116 * Thus we will need a 'compatibility mode'.
2119 static int dev_ifconf(char __user
*arg
)
2122 struct net_device
*dev
;
2129 * Fetch the caller's info block.
2132 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2139 * Loop over the interfaces, and write an info block for each.
2143 for_each_netdev(dev
) {
2144 for (i
= 0; i
< NPROTO
; i
++) {
2145 if (gifconf_list
[i
]) {
2148 done
= gifconf_list
[i
](dev
, NULL
, 0);
2150 done
= gifconf_list
[i
](dev
, pos
+ total
,
2160 * All done. Write the updated control block back to the caller.
2162 ifc
.ifc_len
= total
;
2165 * Both BSD and Solaris return 0 here, so we do too.
2167 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2170 #ifdef CONFIG_PROC_FS
2172 * This is invoked by the /proc filesystem handler to display a device
2175 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2178 struct net_device
*dev
;
2180 read_lock(&dev_base_lock
);
2182 return SEQ_START_TOKEN
;
2185 for_each_netdev(dev
)
2192 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2195 return v
== SEQ_START_TOKEN
?
2196 first_net_device() : next_net_device((struct net_device
*)v
);
2199 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2201 read_unlock(&dev_base_lock
);
2204 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2206 struct net_device_stats
*stats
= dev
->get_stats(dev
);
2208 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2209 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2210 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2212 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2213 stats
->rx_fifo_errors
,
2214 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2215 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2216 stats
->rx_compressed
, stats
->multicast
,
2217 stats
->tx_bytes
, stats
->tx_packets
,
2218 stats
->tx_errors
, stats
->tx_dropped
,
2219 stats
->tx_fifo_errors
, stats
->collisions
,
2220 stats
->tx_carrier_errors
+
2221 stats
->tx_aborted_errors
+
2222 stats
->tx_window_errors
+
2223 stats
->tx_heartbeat_errors
,
2224 stats
->tx_compressed
);
2228 * Called from the PROCfs module. This now uses the new arbitrary sized
2229 * /proc/net interface to create /proc/net/dev
2231 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2233 if (v
== SEQ_START_TOKEN
)
2234 seq_puts(seq
, "Inter-| Receive "
2236 " face |bytes packets errs drop fifo frame "
2237 "compressed multicast|bytes packets errs "
2238 "drop fifo colls carrier compressed\n");
2240 dev_seq_printf_stats(seq
, v
);
2244 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2246 struct netif_rx_stats
*rc
= NULL
;
2248 while (*pos
< NR_CPUS
)
2249 if (cpu_online(*pos
)) {
2250 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2257 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2259 return softnet_get_online(pos
);
2262 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2265 return softnet_get_online(pos
);
2268 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2272 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2274 struct netif_rx_stats
*s
= v
;
2276 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2277 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2278 0, 0, 0, 0, /* was fastroute */
2283 static const struct seq_operations dev_seq_ops
= {
2284 .start
= dev_seq_start
,
2285 .next
= dev_seq_next
,
2286 .stop
= dev_seq_stop
,
2287 .show
= dev_seq_show
,
2290 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2292 return seq_open(file
, &dev_seq_ops
);
2295 static const struct file_operations dev_seq_fops
= {
2296 .owner
= THIS_MODULE
,
2297 .open
= dev_seq_open
,
2299 .llseek
= seq_lseek
,
2300 .release
= seq_release
,
2303 static const struct seq_operations softnet_seq_ops
= {
2304 .start
= softnet_seq_start
,
2305 .next
= softnet_seq_next
,
2306 .stop
= softnet_seq_stop
,
2307 .show
= softnet_seq_show
,
2310 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2312 return seq_open(file
, &softnet_seq_ops
);
2315 static const struct file_operations softnet_seq_fops
= {
2316 .owner
= THIS_MODULE
,
2317 .open
= softnet_seq_open
,
2319 .llseek
= seq_lseek
,
2320 .release
= seq_release
,
2323 static void *ptype_get_idx(loff_t pos
)
2325 struct packet_type
*pt
= NULL
;
2329 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
2335 for (t
= 0; t
< 16; t
++) {
2336 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
2345 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2348 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
2351 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2353 struct packet_type
*pt
;
2354 struct list_head
*nxt
;
2358 if (v
== SEQ_START_TOKEN
)
2359 return ptype_get_idx(0);
2362 nxt
= pt
->list
.next
;
2363 if (pt
->type
== htons(ETH_P_ALL
)) {
2364 if (nxt
!= &ptype_all
)
2367 nxt
= ptype_base
[0].next
;
2369 hash
= ntohs(pt
->type
) & 15;
2371 while (nxt
== &ptype_base
[hash
]) {
2374 nxt
= ptype_base
[hash
].next
;
2377 return list_entry(nxt
, struct packet_type
, list
);
2380 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
2385 static void ptype_seq_decode(struct seq_file
*seq
, void *sym
)
2387 #ifdef CONFIG_KALLSYMS
2388 unsigned long offset
= 0, symsize
;
2389 const char *symname
;
2393 symname
= kallsyms_lookup((unsigned long)sym
, &symsize
, &offset
,
2400 modname
= delim
= "";
2401 seq_printf(seq
, "%s%s%s%s+0x%lx", delim
, modname
, delim
,
2407 seq_printf(seq
, "[%p]", sym
);
2410 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
2412 struct packet_type
*pt
= v
;
2414 if (v
== SEQ_START_TOKEN
)
2415 seq_puts(seq
, "Type Device Function\n");
2417 if (pt
->type
== htons(ETH_P_ALL
))
2418 seq_puts(seq
, "ALL ");
2420 seq_printf(seq
, "%04x", ntohs(pt
->type
));
2422 seq_printf(seq
, " %-8s ",
2423 pt
->dev
? pt
->dev
->name
: "");
2424 ptype_seq_decode(seq
, pt
->func
);
2425 seq_putc(seq
, '\n');
2431 static const struct seq_operations ptype_seq_ops
= {
2432 .start
= ptype_seq_start
,
2433 .next
= ptype_seq_next
,
2434 .stop
= ptype_seq_stop
,
2435 .show
= ptype_seq_show
,
2438 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
2440 return seq_open(file
, &ptype_seq_ops
);
2443 static const struct file_operations ptype_seq_fops
= {
2444 .owner
= THIS_MODULE
,
2445 .open
= ptype_seq_open
,
2447 .llseek
= seq_lseek
,
2448 .release
= seq_release
,
2452 static int __init
dev_proc_init(void)
2456 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2458 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2460 if (!proc_net_fops_create("ptype", S_IRUGO
, &ptype_seq_fops
))
2463 if (wext_proc_init())
2469 proc_net_remove("ptype");
2471 proc_net_remove("softnet_stat");
2473 proc_net_remove("dev");
2477 #define dev_proc_init() 0
2478 #endif /* CONFIG_PROC_FS */
2482 * netdev_set_master - set up master/slave pair
2483 * @slave: slave device
2484 * @master: new master device
2486 * Changes the master device of the slave. Pass %NULL to break the
2487 * bonding. The caller must hold the RTNL semaphore. On a failure
2488 * a negative errno code is returned. On success the reference counts
2489 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2490 * function returns zero.
2492 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2494 struct net_device
*old
= slave
->master
;
2504 slave
->master
= master
;
2512 slave
->flags
|= IFF_SLAVE
;
2514 slave
->flags
&= ~IFF_SLAVE
;
2516 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2520 static void __dev_set_promiscuity(struct net_device
*dev
, int inc
)
2522 unsigned short old_flags
= dev
->flags
;
2524 if ((dev
->promiscuity
+= inc
) == 0)
2525 dev
->flags
&= ~IFF_PROMISC
;
2527 dev
->flags
|= IFF_PROMISC
;
2528 if (dev
->flags
!= old_flags
) {
2529 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2530 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2532 audit_log(current
->audit_context
, GFP_ATOMIC
,
2533 AUDIT_ANOM_PROMISCUOUS
,
2534 "dev=%s prom=%d old_prom=%d auid=%u",
2535 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2536 (old_flags
& IFF_PROMISC
),
2537 audit_get_loginuid(current
->audit_context
));
2542 * dev_set_promiscuity - update promiscuity count on a device
2546 * Add or remove promiscuity from a device. While the count in the device
2547 * remains above zero the interface remains promiscuous. Once it hits zero
2548 * the device reverts back to normal filtering operation. A negative inc
2549 * value is used to drop promiscuity on the device.
2551 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2553 unsigned short old_flags
= dev
->flags
;
2555 __dev_set_promiscuity(dev
, inc
);
2556 if (dev
->flags
!= old_flags
)
2557 dev_set_rx_mode(dev
);
2561 * dev_set_allmulti - update allmulti count on a device
2565 * Add or remove reception of all multicast frames to a device. While the
2566 * count in the device remains above zero the interface remains listening
2567 * to all interfaces. Once it hits zero the device reverts back to normal
2568 * filtering operation. A negative @inc value is used to drop the counter
2569 * when releasing a resource needing all multicasts.
2572 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2574 unsigned short old_flags
= dev
->flags
;
2576 dev
->flags
|= IFF_ALLMULTI
;
2577 if ((dev
->allmulti
+= inc
) == 0)
2578 dev
->flags
&= ~IFF_ALLMULTI
;
2579 if (dev
->flags
^ old_flags
)
2580 dev_set_rx_mode(dev
);
2584 * Upload unicast and multicast address lists to device and
2585 * configure RX filtering. When the device doesn't support unicast
2586 * filtering it is put in promiscous mode while unicast addresses
2589 void __dev_set_rx_mode(struct net_device
*dev
)
2591 /* dev_open will call this function so the list will stay sane. */
2592 if (!(dev
->flags
&IFF_UP
))
2595 if (!netif_device_present(dev
))
2598 if (dev
->set_rx_mode
)
2599 dev
->set_rx_mode(dev
);
2601 /* Unicast addresses changes may only happen under the rtnl,
2602 * therefore calling __dev_set_promiscuity here is safe.
2604 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
2605 __dev_set_promiscuity(dev
, 1);
2606 dev
->uc_promisc
= 1;
2607 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
2608 __dev_set_promiscuity(dev
, -1);
2609 dev
->uc_promisc
= 0;
2612 if (dev
->set_multicast_list
)
2613 dev
->set_multicast_list(dev
);
2617 void dev_set_rx_mode(struct net_device
*dev
)
2619 netif_tx_lock_bh(dev
);
2620 __dev_set_rx_mode(dev
);
2621 netif_tx_unlock_bh(dev
);
2624 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
2625 void *addr
, int alen
, int glbl
)
2627 struct dev_addr_list
*da
;
2629 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
2630 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2631 alen
== da
->da_addrlen
) {
2633 int old_glbl
= da
->da_gusers
;
2650 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
2651 void *addr
, int alen
, int glbl
)
2653 struct dev_addr_list
*da
;
2655 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
2656 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2657 da
->da_addrlen
== alen
) {
2659 int old_glbl
= da
->da_gusers
;
2669 da
= kmalloc(sizeof(*da
), GFP_ATOMIC
);
2672 memcpy(da
->da_addr
, addr
, alen
);
2673 da
->da_addrlen
= alen
;
2675 da
->da_gusers
= glbl
? 1 : 0;
2682 void __dev_addr_discard(struct dev_addr_list
**list
)
2684 struct dev_addr_list
*tmp
;
2686 while (*list
!= NULL
) {
2689 if (tmp
->da_users
> tmp
->da_gusers
)
2690 printk("__dev_addr_discard: address leakage! "
2691 "da_users=%d\n", tmp
->da_users
);
2697 * dev_unicast_delete - Release secondary unicast address.
2700 * Release reference to a secondary unicast address and remove it
2701 * from the device if the reference count drop to zero.
2703 * The caller must hold the rtnl_mutex.
2705 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
2711 netif_tx_lock_bh(dev
);
2712 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
2714 __dev_set_rx_mode(dev
);
2715 netif_tx_unlock_bh(dev
);
2718 EXPORT_SYMBOL(dev_unicast_delete
);
2721 * dev_unicast_add - add a secondary unicast address
2724 * Add a secondary unicast address to the device or increase
2725 * the reference count if it already exists.
2727 * The caller must hold the rtnl_mutex.
2729 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
2735 netif_tx_lock_bh(dev
);
2736 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
2738 __dev_set_rx_mode(dev
);
2739 netif_tx_unlock_bh(dev
);
2742 EXPORT_SYMBOL(dev_unicast_add
);
2744 static void dev_unicast_discard(struct net_device
*dev
)
2746 netif_tx_lock_bh(dev
);
2747 __dev_addr_discard(&dev
->uc_list
);
2749 netif_tx_unlock_bh(dev
);
2752 unsigned dev_get_flags(const struct net_device
*dev
)
2756 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2761 (dev
->gflags
& (IFF_PROMISC
|
2764 if (netif_running(dev
)) {
2765 if (netif_oper_up(dev
))
2766 flags
|= IFF_RUNNING
;
2767 if (netif_carrier_ok(dev
))
2768 flags
|= IFF_LOWER_UP
;
2769 if (netif_dormant(dev
))
2770 flags
|= IFF_DORMANT
;
2776 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2779 int old_flags
= dev
->flags
;
2782 * Set the flags on our device.
2785 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2786 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2788 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2792 * Load in the correct multicast list now the flags have changed.
2795 dev_set_rx_mode(dev
);
2798 * Have we downed the interface. We handle IFF_UP ourselves
2799 * according to user attempts to set it, rather than blindly
2804 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2805 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2808 dev_set_rx_mode(dev
);
2811 if (dev
->flags
& IFF_UP
&&
2812 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2814 raw_notifier_call_chain(&netdev_chain
,
2815 NETDEV_CHANGE
, dev
);
2817 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2818 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2819 dev
->gflags
^= IFF_PROMISC
;
2820 dev_set_promiscuity(dev
, inc
);
2823 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2824 is important. Some (broken) drivers set IFF_PROMISC, when
2825 IFF_ALLMULTI is requested not asking us and not reporting.
2827 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2828 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2829 dev
->gflags
^= IFF_ALLMULTI
;
2830 dev_set_allmulti(dev
, inc
);
2833 /* Exclude state transition flags, already notified */
2834 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
2836 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
2841 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2845 if (new_mtu
== dev
->mtu
)
2848 /* MTU must be positive. */
2852 if (!netif_device_present(dev
))
2856 if (dev
->change_mtu
)
2857 err
= dev
->change_mtu(dev
, new_mtu
);
2860 if (!err
&& dev
->flags
& IFF_UP
)
2861 raw_notifier_call_chain(&netdev_chain
,
2862 NETDEV_CHANGEMTU
, dev
);
2866 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2870 if (!dev
->set_mac_address
)
2872 if (sa
->sa_family
!= dev
->type
)
2874 if (!netif_device_present(dev
))
2876 err
= dev
->set_mac_address(dev
, sa
);
2878 raw_notifier_call_chain(&netdev_chain
,
2879 NETDEV_CHANGEADDR
, dev
);
2884 * Perform the SIOCxIFxxx calls.
2886 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2889 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2895 case SIOCGIFFLAGS
: /* Get interface flags */
2896 ifr
->ifr_flags
= dev_get_flags(dev
);
2899 case SIOCSIFFLAGS
: /* Set interface flags */
2900 return dev_change_flags(dev
, ifr
->ifr_flags
);
2902 case SIOCGIFMETRIC
: /* Get the metric on the interface
2903 (currently unused) */
2904 ifr
->ifr_metric
= 0;
2907 case SIOCSIFMETRIC
: /* Set the metric on the interface
2908 (currently unused) */
2911 case SIOCGIFMTU
: /* Get the MTU of a device */
2912 ifr
->ifr_mtu
= dev
->mtu
;
2915 case SIOCSIFMTU
: /* Set the MTU of a device */
2916 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
2920 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
2922 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
2923 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2924 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
2928 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
2930 case SIOCSIFHWBROADCAST
:
2931 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
2933 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
2934 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2935 raw_notifier_call_chain(&netdev_chain
,
2936 NETDEV_CHANGEADDR
, dev
);
2940 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
2941 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
2942 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
2943 ifr
->ifr_map
.irq
= dev
->irq
;
2944 ifr
->ifr_map
.dma
= dev
->dma
;
2945 ifr
->ifr_map
.port
= dev
->if_port
;
2949 if (dev
->set_config
) {
2950 if (!netif_device_present(dev
))
2952 return dev
->set_config(dev
, &ifr
->ifr_map
);
2957 if (!dev
->set_multicast_list
||
2958 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2960 if (!netif_device_present(dev
))
2962 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
2966 if (!dev
->set_multicast_list
||
2967 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2969 if (!netif_device_present(dev
))
2971 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
2975 ifr
->ifr_ifindex
= dev
->ifindex
;
2979 ifr
->ifr_qlen
= dev
->tx_queue_len
;
2983 if (ifr
->ifr_qlen
< 0)
2985 dev
->tx_queue_len
= ifr
->ifr_qlen
;
2989 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
2990 return dev_change_name(dev
, ifr
->ifr_newname
);
2993 * Unknown or private ioctl
2997 if ((cmd
>= SIOCDEVPRIVATE
&&
2998 cmd
<= SIOCDEVPRIVATE
+ 15) ||
2999 cmd
== SIOCBONDENSLAVE
||
3000 cmd
== SIOCBONDRELEASE
||
3001 cmd
== SIOCBONDSETHWADDR
||
3002 cmd
== SIOCBONDSLAVEINFOQUERY
||
3003 cmd
== SIOCBONDINFOQUERY
||
3004 cmd
== SIOCBONDCHANGEACTIVE
||
3005 cmd
== SIOCGMIIPHY
||
3006 cmd
== SIOCGMIIREG
||
3007 cmd
== SIOCSMIIREG
||
3008 cmd
== SIOCBRADDIF
||
3009 cmd
== SIOCBRDELIF
||
3010 cmd
== SIOCWANDEV
) {
3012 if (dev
->do_ioctl
) {
3013 if (netif_device_present(dev
))
3014 err
= dev
->do_ioctl(dev
, ifr
,
3027 * This function handles all "interface"-type I/O control requests. The actual
3028 * 'doing' part of this is dev_ifsioc above.
3032 * dev_ioctl - network device ioctl
3033 * @cmd: command to issue
3034 * @arg: pointer to a struct ifreq in user space
3036 * Issue ioctl functions to devices. This is normally called by the
3037 * user space syscall interfaces but can sometimes be useful for
3038 * other purposes. The return value is the return from the syscall if
3039 * positive or a negative errno code on error.
3042 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
3048 /* One special case: SIOCGIFCONF takes ifconf argument
3049 and requires shared lock, because it sleeps writing
3053 if (cmd
== SIOCGIFCONF
) {
3055 ret
= dev_ifconf((char __user
*) arg
);
3059 if (cmd
== SIOCGIFNAME
)
3060 return dev_ifname((struct ifreq __user
*)arg
);
3062 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3065 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
3067 colon
= strchr(ifr
.ifr_name
, ':');
3072 * See which interface the caller is talking about.
3077 * These ioctl calls:
3078 * - can be done by all.
3079 * - atomic and do not require locking.
3090 dev_load(ifr
.ifr_name
);
3091 read_lock(&dev_base_lock
);
3092 ret
= dev_ifsioc(&ifr
, cmd
);
3093 read_unlock(&dev_base_lock
);
3097 if (copy_to_user(arg
, &ifr
,
3098 sizeof(struct ifreq
)))
3104 dev_load(ifr
.ifr_name
);
3106 ret
= dev_ethtool(&ifr
);
3111 if (copy_to_user(arg
, &ifr
,
3112 sizeof(struct ifreq
)))
3118 * These ioctl calls:
3119 * - require superuser power.
3120 * - require strict serialization.
3126 if (!capable(CAP_NET_ADMIN
))
3128 dev_load(ifr
.ifr_name
);
3130 ret
= dev_ifsioc(&ifr
, cmd
);
3135 if (copy_to_user(arg
, &ifr
,
3136 sizeof(struct ifreq
)))
3142 * These ioctl calls:
3143 * - require superuser power.
3144 * - require strict serialization.
3145 * - do not return a value
3155 case SIOCSIFHWBROADCAST
:
3158 case SIOCBONDENSLAVE
:
3159 case SIOCBONDRELEASE
:
3160 case SIOCBONDSETHWADDR
:
3161 case SIOCBONDCHANGEACTIVE
:
3164 if (!capable(CAP_NET_ADMIN
))
3167 case SIOCBONDSLAVEINFOQUERY
:
3168 case SIOCBONDINFOQUERY
:
3169 dev_load(ifr
.ifr_name
);
3171 ret
= dev_ifsioc(&ifr
, cmd
);
3176 /* Get the per device memory space. We can add this but
3177 * currently do not support it */
3179 /* Set the per device memory buffer space.
3180 * Not applicable in our case */
3185 * Unknown or private ioctl.
3188 if (cmd
== SIOCWANDEV
||
3189 (cmd
>= SIOCDEVPRIVATE
&&
3190 cmd
<= SIOCDEVPRIVATE
+ 15)) {
3191 dev_load(ifr
.ifr_name
);
3193 ret
= dev_ifsioc(&ifr
, cmd
);
3195 if (!ret
&& copy_to_user(arg
, &ifr
,
3196 sizeof(struct ifreq
)))
3200 /* Take care of Wireless Extensions */
3201 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
3202 return wext_handle_ioctl(&ifr
, cmd
, arg
);
3209 * dev_new_index - allocate an ifindex
3211 * Returns a suitable unique value for a new device interface
3212 * number. The caller must hold the rtnl semaphore or the
3213 * dev_base_lock to be sure it remains unique.
3215 static int dev_new_index(void)
3221 if (!__dev_get_by_index(ifindex
))
3226 static int dev_boot_phase
= 1;
3228 /* Delayed registration/unregisteration */
3229 static DEFINE_SPINLOCK(net_todo_list_lock
);
3230 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
3232 static void net_set_todo(struct net_device
*dev
)
3234 spin_lock(&net_todo_list_lock
);
3235 list_add_tail(&dev
->todo_list
, &net_todo_list
);
3236 spin_unlock(&net_todo_list_lock
);
3240 * register_netdevice - register a network device
3241 * @dev: device to register
3243 * Take a completed network device structure and add it to the kernel
3244 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3245 * chain. 0 is returned on success. A negative errno code is returned
3246 * on a failure to set up the device, or if the name is a duplicate.
3248 * Callers must hold the rtnl semaphore. You may want
3249 * register_netdev() instead of this.
3252 * The locking appears insufficient to guarantee two parallel registers
3253 * will not get the same name.
3256 int register_netdevice(struct net_device
*dev
)
3258 struct hlist_head
*head
;
3259 struct hlist_node
*p
;
3262 BUG_ON(dev_boot_phase
);
3267 /* When net_device's are persistent, this will be fatal. */
3268 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
3270 spin_lock_init(&dev
->queue_lock
);
3271 spin_lock_init(&dev
->_xmit_lock
);
3272 netdev_set_lockdep_class(&dev
->_xmit_lock
, dev
->type
);
3273 dev
->xmit_lock_owner
= -1;
3274 spin_lock_init(&dev
->ingress_lock
);
3278 /* Init, if this function is available */
3280 ret
= dev
->init(dev
);
3288 if (!dev_valid_name(dev
->name
)) {
3293 dev
->ifindex
= dev_new_index();
3294 if (dev
->iflink
== -1)
3295 dev
->iflink
= dev
->ifindex
;
3297 /* Check for existence of name */
3298 head
= dev_name_hash(dev
->name
);
3299 hlist_for_each(p
, head
) {
3300 struct net_device
*d
3301 = hlist_entry(p
, struct net_device
, name_hlist
);
3302 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
3308 /* Fix illegal checksum combinations */
3309 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
3310 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3311 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
3313 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
3316 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
3317 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3318 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
3320 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
3324 /* Fix illegal SG+CSUM combinations. */
3325 if ((dev
->features
& NETIF_F_SG
) &&
3326 !(dev
->features
& NETIF_F_ALL_CSUM
)) {
3327 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no checksum feature.\n",
3329 dev
->features
&= ~NETIF_F_SG
;
3332 /* TSO requires that SG is present as well. */
3333 if ((dev
->features
& NETIF_F_TSO
) &&
3334 !(dev
->features
& NETIF_F_SG
)) {
3335 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no SG feature.\n",
3337 dev
->features
&= ~NETIF_F_TSO
;
3339 if (dev
->features
& NETIF_F_UFO
) {
3340 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
3341 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3342 "NETIF_F_HW_CSUM feature.\n",
3344 dev
->features
&= ~NETIF_F_UFO
;
3346 if (!(dev
->features
& NETIF_F_SG
)) {
3347 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3348 "NETIF_F_SG feature.\n",
3350 dev
->features
&= ~NETIF_F_UFO
;
3355 * nil rebuild_header routine,
3356 * that should be never called and used as just bug trap.
3359 if (!dev
->rebuild_header
)
3360 dev
->rebuild_header
= default_rebuild_header
;
3362 ret
= netdev_register_sysfs(dev
);
3365 dev
->reg_state
= NETREG_REGISTERED
;
3368 * Default initial state at registry is that the
3369 * device is present.
3372 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
3374 dev_init_scheduler(dev
);
3375 write_lock_bh(&dev_base_lock
);
3376 list_add_tail(&dev
->dev_list
, &dev_base_head
);
3377 hlist_add_head(&dev
->name_hlist
, head
);
3378 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
3380 write_unlock_bh(&dev_base_lock
);
3382 /* Notify protocols, that a new device appeared. */
3383 raw_notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
3392 * register_netdev - register a network device
3393 * @dev: device to register
3395 * Take a completed network device structure and add it to the kernel
3396 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3397 * chain. 0 is returned on success. A negative errno code is returned
3398 * on a failure to set up the device, or if the name is a duplicate.
3400 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3401 * and expands the device name if you passed a format string to
3404 int register_netdev(struct net_device
*dev
)
3411 * If the name is a format string the caller wants us to do a
3414 if (strchr(dev
->name
, '%')) {
3415 err
= dev_alloc_name(dev
, dev
->name
);
3420 err
= register_netdevice(dev
);
3425 EXPORT_SYMBOL(register_netdev
);
3428 * netdev_wait_allrefs - wait until all references are gone.
3430 * This is called when unregistering network devices.
3432 * Any protocol or device that holds a reference should register
3433 * for netdevice notification, and cleanup and put back the
3434 * reference if they receive an UNREGISTER event.
3435 * We can get stuck here if buggy protocols don't correctly
3438 static void netdev_wait_allrefs(struct net_device
*dev
)
3440 unsigned long rebroadcast_time
, warning_time
;
3442 rebroadcast_time
= warning_time
= jiffies
;
3443 while (atomic_read(&dev
->refcnt
) != 0) {
3444 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
3447 /* Rebroadcast unregister notification */
3448 raw_notifier_call_chain(&netdev_chain
,
3449 NETDEV_UNREGISTER
, dev
);
3451 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
3453 /* We must not have linkwatch events
3454 * pending on unregister. If this
3455 * happens, we simply run the queue
3456 * unscheduled, resulting in a noop
3459 linkwatch_run_queue();
3464 rebroadcast_time
= jiffies
;
3469 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
3470 printk(KERN_EMERG
"unregister_netdevice: "
3471 "waiting for %s to become free. Usage "
3473 dev
->name
, atomic_read(&dev
->refcnt
));
3474 warning_time
= jiffies
;
3483 * register_netdevice(x1);
3484 * register_netdevice(x2);
3486 * unregister_netdevice(y1);
3487 * unregister_netdevice(y2);
3493 * We are invoked by rtnl_unlock() after it drops the semaphore.
3494 * This allows us to deal with problems:
3495 * 1) We can delete sysfs objects which invoke hotplug
3496 * without deadlocking with linkwatch via keventd.
3497 * 2) Since we run with the RTNL semaphore not held, we can sleep
3498 * safely in order to wait for the netdev refcnt to drop to zero.
3500 static DEFINE_MUTEX(net_todo_run_mutex
);
3501 void netdev_run_todo(void)
3503 struct list_head list
;
3505 /* Need to guard against multiple cpu's getting out of order. */
3506 mutex_lock(&net_todo_run_mutex
);
3508 /* Not safe to do outside the semaphore. We must not return
3509 * until all unregister events invoked by the local processor
3510 * have been completed (either by this todo run, or one on
3513 if (list_empty(&net_todo_list
))
3516 /* Snapshot list, allow later requests */
3517 spin_lock(&net_todo_list_lock
);
3518 list_replace_init(&net_todo_list
, &list
);
3519 spin_unlock(&net_todo_list_lock
);
3521 while (!list_empty(&list
)) {
3522 struct net_device
*dev
3523 = list_entry(list
.next
, struct net_device
, todo_list
);
3524 list_del(&dev
->todo_list
);
3526 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
3527 printk(KERN_ERR
"network todo '%s' but state %d\n",
3528 dev
->name
, dev
->reg_state
);
3533 dev
->reg_state
= NETREG_UNREGISTERED
;
3535 netdev_wait_allrefs(dev
);
3538 BUG_ON(atomic_read(&dev
->refcnt
));
3539 BUG_TRAP(!dev
->ip_ptr
);
3540 BUG_TRAP(!dev
->ip6_ptr
);
3541 BUG_TRAP(!dev
->dn_ptr
);
3543 if (dev
->destructor
)
3544 dev
->destructor(dev
);
3546 /* Free network device */
3547 kobject_put(&dev
->dev
.kobj
);
3551 mutex_unlock(&net_todo_run_mutex
);
3554 static struct net_device_stats
*internal_stats(struct net_device
*dev
)
3560 * alloc_netdev_mq - allocate network device
3561 * @sizeof_priv: size of private data to allocate space for
3562 * @name: device name format string
3563 * @setup: callback to initialize device
3564 * @queue_count: the number of subqueues to allocate
3566 * Allocates a struct net_device with private data area for driver use
3567 * and performs basic initialization. Also allocates subquue structs
3568 * for each queue on the device at the end of the netdevice.
3570 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
3571 void (*setup
)(struct net_device
*), unsigned int queue_count
)
3574 struct net_device
*dev
;
3577 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
3579 /* ensure 32-byte alignment of both the device and private area */
3580 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
+
3581 (sizeof(struct net_device_subqueue
) * queue_count
)) &
3582 ~NETDEV_ALIGN_CONST
;
3583 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
3585 p
= kzalloc(alloc_size
, GFP_KERNEL
);
3587 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
3591 dev
= (struct net_device
*)
3592 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3593 dev
->padded
= (char *)dev
- (char *)p
;
3596 dev
->priv
= ((char *)dev
+
3597 ((sizeof(struct net_device
) +
3598 (sizeof(struct net_device_subqueue
) *
3599 queue_count
) + NETDEV_ALIGN_CONST
)
3600 & ~NETDEV_ALIGN_CONST
));
3603 dev
->egress_subqueue_count
= queue_count
;
3605 dev
->get_stats
= internal_stats
;
3607 strcpy(dev
->name
, name
);
3610 EXPORT_SYMBOL(alloc_netdev_mq
);
3613 * free_netdev - free network device
3616 * This function does the last stage of destroying an allocated device
3617 * interface. The reference to the device object is released.
3618 * If this is the last reference then it will be freed.
3620 void free_netdev(struct net_device
*dev
)
3623 /* Compatibility with error handling in drivers */
3624 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3625 kfree((char *)dev
- dev
->padded
);
3629 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3630 dev
->reg_state
= NETREG_RELEASED
;
3632 /* will free via device release */
3633 put_device(&dev
->dev
);
3635 kfree((char *)dev
- dev
->padded
);
3639 /* Synchronize with packet receive processing. */
3640 void synchronize_net(void)
3647 * unregister_netdevice - remove device from the kernel
3650 * This function shuts down a device interface and removes it
3651 * from the kernel tables. On success 0 is returned, on a failure
3652 * a negative errno code is returned.
3654 * Callers must hold the rtnl semaphore. You may want
3655 * unregister_netdev() instead of this.
3658 void unregister_netdevice(struct net_device
*dev
)
3660 BUG_ON(dev_boot_phase
);
3663 /* Some devices call without registering for initialization unwind. */
3664 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3665 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3666 "was registered\n", dev
->name
, dev
);
3672 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3674 /* If device is running, close it first. */
3675 if (dev
->flags
& IFF_UP
)
3678 /* And unlink it from device chain. */
3679 write_lock_bh(&dev_base_lock
);
3680 list_del(&dev
->dev_list
);
3681 hlist_del(&dev
->name_hlist
);
3682 hlist_del(&dev
->index_hlist
);
3683 write_unlock_bh(&dev_base_lock
);
3685 dev
->reg_state
= NETREG_UNREGISTERING
;
3689 /* Shutdown queueing discipline. */
3693 /* Notify protocols, that we are about to destroy
3694 this device. They should clean all the things.
3696 raw_notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3699 * Flush the unicast and multicast chains
3701 dev_unicast_discard(dev
);
3702 dev_mc_discard(dev
);
3707 /* Notifier chain MUST detach us from master device. */
3708 BUG_TRAP(!dev
->master
);
3710 /* Remove entries from sysfs */
3711 netdev_unregister_sysfs(dev
);
3713 /* Finish processing unregister after unlock */
3722 * unregister_netdev - remove device from the kernel
3725 * This function shuts down a device interface and removes it
3726 * from the kernel tables. On success 0 is returned, on a failure
3727 * a negative errno code is returned.
3729 * This is just a wrapper for unregister_netdevice that takes
3730 * the rtnl semaphore. In general you want to use this and not
3731 * unregister_netdevice.
3733 void unregister_netdev(struct net_device
*dev
)
3736 unregister_netdevice(dev
);
3740 EXPORT_SYMBOL(unregister_netdev
);
3742 static int dev_cpu_callback(struct notifier_block
*nfb
,
3743 unsigned long action
,
3746 struct sk_buff
**list_skb
;
3747 struct net_device
**list_net
;
3748 struct sk_buff
*skb
;
3749 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3750 struct softnet_data
*sd
, *oldsd
;
3752 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
3755 local_irq_disable();
3756 cpu
= smp_processor_id();
3757 sd
= &per_cpu(softnet_data
, cpu
);
3758 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3760 /* Find end of our completion_queue. */
3761 list_skb
= &sd
->completion_queue
;
3763 list_skb
= &(*list_skb
)->next
;
3764 /* Append completion queue from offline CPU. */
3765 *list_skb
= oldsd
->completion_queue
;
3766 oldsd
->completion_queue
= NULL
;
3768 /* Find end of our output_queue. */
3769 list_net
= &sd
->output_queue
;
3771 list_net
= &(*list_net
)->next_sched
;
3772 /* Append output queue from offline CPU. */
3773 *list_net
= oldsd
->output_queue
;
3774 oldsd
->output_queue
= NULL
;
3776 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3779 /* Process offline CPU's input_pkt_queue */
3780 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3786 #ifdef CONFIG_NET_DMA
3788 * net_dma_rebalance -
3789 * This is called when the number of channels allocated to the net_dma_client
3790 * changes. The net_dma_client tries to have one DMA channel per CPU.
3793 static void net_dma_rebalance(struct net_dma
*net_dma
)
3795 unsigned int cpu
, i
, n
, chan_idx
;
3796 struct dma_chan
*chan
;
3798 if (cpus_empty(net_dma
->channel_mask
)) {
3799 for_each_online_cpu(cpu
)
3800 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
3805 cpu
= first_cpu(cpu_online_map
);
3807 for_each_cpu_mask(chan_idx
, net_dma
->channel_mask
) {
3808 chan
= net_dma
->channels
[chan_idx
];
3810 n
= ((num_online_cpus() / cpus_weight(net_dma
->channel_mask
))
3811 + (i
< (num_online_cpus() %
3812 cpus_weight(net_dma
->channel_mask
)) ? 1 : 0));
3815 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
3816 cpu
= next_cpu(cpu
, cpu_online_map
);
3824 * netdev_dma_event - event callback for the net_dma_client
3825 * @client: should always be net_dma_client
3826 * @chan: DMA channel for the event
3827 * @event: event type
3829 static enum dma_state_client
3830 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
3831 enum dma_state state
)
3833 int i
, found
= 0, pos
= -1;
3834 struct net_dma
*net_dma
=
3835 container_of(client
, struct net_dma
, client
);
3836 enum dma_state_client ack
= DMA_DUP
; /* default: take no action */
3838 spin_lock(&net_dma
->lock
);
3840 case DMA_RESOURCE_AVAILABLE
:
3841 for (i
= 0; i
< NR_CPUS
; i
++)
3842 if (net_dma
->channels
[i
] == chan
) {
3845 } else if (net_dma
->channels
[i
] == NULL
&& pos
< 0)
3848 if (!found
&& pos
>= 0) {
3850 net_dma
->channels
[pos
] = chan
;
3851 cpu_set(pos
, net_dma
->channel_mask
);
3852 net_dma_rebalance(net_dma
);
3855 case DMA_RESOURCE_REMOVED
:
3856 for (i
= 0; i
< NR_CPUS
; i
++)
3857 if (net_dma
->channels
[i
] == chan
) {
3865 cpu_clear(pos
, net_dma
->channel_mask
);
3866 net_dma
->channels
[i
] = NULL
;
3867 net_dma_rebalance(net_dma
);
3873 spin_unlock(&net_dma
->lock
);
3879 * netdev_dma_regiser - register the networking subsystem as a DMA client
3881 static int __init
netdev_dma_register(void)
3883 spin_lock_init(&net_dma
.lock
);
3884 dma_cap_set(DMA_MEMCPY
, net_dma
.client
.cap_mask
);
3885 dma_async_client_register(&net_dma
.client
);
3886 dma_async_client_chan_request(&net_dma
.client
);
3891 static int __init
netdev_dma_register(void) { return -ENODEV
; }
3892 #endif /* CONFIG_NET_DMA */
3895 * Initialize the DEV module. At boot time this walks the device list and
3896 * unhooks any devices that fail to initialise (normally hardware not
3897 * present) and leaves us with a valid list of present and active devices.
3902 * This is called single threaded during boot, so no need
3903 * to take the rtnl semaphore.
3905 static int __init
net_dev_init(void)
3907 int i
, rc
= -ENOMEM
;
3909 BUG_ON(!dev_boot_phase
);
3911 if (dev_proc_init())
3914 if (netdev_sysfs_init())
3917 INIT_LIST_HEAD(&ptype_all
);
3918 for (i
= 0; i
< 16; i
++)
3919 INIT_LIST_HEAD(&ptype_base
[i
]);
3921 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
3922 INIT_HLIST_HEAD(&dev_name_head
[i
]);
3924 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
3925 INIT_HLIST_HEAD(&dev_index_head
[i
]);
3928 * Initialise the packet receive queues.
3931 for_each_possible_cpu(i
) {
3932 struct softnet_data
*queue
;
3934 queue
= &per_cpu(softnet_data
, i
);
3935 skb_queue_head_init(&queue
->input_pkt_queue
);
3936 queue
->completion_queue
= NULL
;
3937 INIT_LIST_HEAD(&queue
->poll_list
);
3938 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
3939 queue
->backlog_dev
.weight
= weight_p
;
3940 queue
->backlog_dev
.poll
= process_backlog
;
3941 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
3944 netdev_dma_register();
3948 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
3949 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
3951 hotcpu_notifier(dev_cpu_callback
, 0);
3959 subsys_initcall(net_dev_init
);
3961 EXPORT_SYMBOL(__dev_get_by_index
);
3962 EXPORT_SYMBOL(__dev_get_by_name
);
3963 EXPORT_SYMBOL(__dev_remove_pack
);
3964 EXPORT_SYMBOL(dev_valid_name
);
3965 EXPORT_SYMBOL(dev_add_pack
);
3966 EXPORT_SYMBOL(dev_alloc_name
);
3967 EXPORT_SYMBOL(dev_close
);
3968 EXPORT_SYMBOL(dev_get_by_flags
);
3969 EXPORT_SYMBOL(dev_get_by_index
);
3970 EXPORT_SYMBOL(dev_get_by_name
);
3971 EXPORT_SYMBOL(dev_open
);
3972 EXPORT_SYMBOL(dev_queue_xmit
);
3973 EXPORT_SYMBOL(dev_remove_pack
);
3974 EXPORT_SYMBOL(dev_set_allmulti
);
3975 EXPORT_SYMBOL(dev_set_promiscuity
);
3976 EXPORT_SYMBOL(dev_change_flags
);
3977 EXPORT_SYMBOL(dev_set_mtu
);
3978 EXPORT_SYMBOL(dev_set_mac_address
);
3979 EXPORT_SYMBOL(free_netdev
);
3980 EXPORT_SYMBOL(netdev_boot_setup_check
);
3981 EXPORT_SYMBOL(netdev_set_master
);
3982 EXPORT_SYMBOL(netdev_state_change
);
3983 EXPORT_SYMBOL(netif_receive_skb
);
3984 EXPORT_SYMBOL(netif_rx
);
3985 EXPORT_SYMBOL(register_gifconf
);
3986 EXPORT_SYMBOL(register_netdevice
);
3987 EXPORT_SYMBOL(register_netdevice_notifier
);
3988 EXPORT_SYMBOL(skb_checksum_help
);
3989 EXPORT_SYMBOL(synchronize_net
);
3990 EXPORT_SYMBOL(unregister_netdevice
);
3991 EXPORT_SYMBOL(unregister_netdevice_notifier
);
3992 EXPORT_SYMBOL(net_enable_timestamp
);
3993 EXPORT_SYMBOL(net_disable_timestamp
);
3994 EXPORT_SYMBOL(dev_get_flags
);
3996 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3997 EXPORT_SYMBOL(br_handle_frame_hook
);
3998 EXPORT_SYMBOL(br_fdb_get_hook
);
3999 EXPORT_SYMBOL(br_fdb_put_hook
);
4003 EXPORT_SYMBOL(dev_load
);
4006 EXPORT_PER_CPU_SYMBOL(softnet_data
);