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 #if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
98 #include <linux/imq.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
103 #include <linux/if_bridge.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 <typedefs.h>
127 * The list of packet types we will receive (as opposed to discard)
128 * and the routines to invoke.
130 * Why 16. Because with 16 the only overlap we get on a hash of the
131 * low nibble of the protocol value is RARP/SNAP/X.25.
133 * NOTE: That is no longer true with the addition of VLAN tags. Not
134 * sure which should go first, but I bet it won't make much
135 * difference if we are running VLANs. The good news is that
136 * this protocol won't be in the list unless compiled in, so
137 * the average user (w/out VLANs) will not be adversely affected.
154 static DEFINE_SPINLOCK(ptype_lock
);
155 static struct list_head ptype_base
[16] __read_mostly
; /* 16 way hashed list */
156 static struct list_head ptype_all __read_mostly
; /* Taps */
158 #ifdef CONFIG_NET_DMA
159 static struct dma_client
*net_dma_client
;
160 static unsigned int net_dma_count
;
161 static spinlock_t net_dma_event_lock
;
165 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
168 * Pure readers hold dev_base_lock for reading.
170 * Writers must hold the rtnl semaphore while they loop through the
171 * dev_base_head list, and hold dev_base_lock for writing when they do the
172 * actual updates. This allows pure readers to access the list even
173 * while a writer is preparing to update it.
175 * To put it another way, dev_base_lock is held for writing only to
176 * protect against pure readers; the rtnl semaphore provides the
177 * protection against other writers.
179 * See, for example usages, register_netdevice() and
180 * unregister_netdevice(), which must be called with the rtnl
183 LIST_HEAD(dev_base_head
);
184 DEFINE_RWLOCK(dev_base_lock
);
186 EXPORT_SYMBOL(dev_base_head
);
187 EXPORT_SYMBOL(dev_base_lock
);
189 #define NETDEV_HASHBITS 8
190 static struct hlist_head dev_name_head
[1<<NETDEV_HASHBITS
];
191 static struct hlist_head dev_index_head
[1<<NETDEV_HASHBITS
];
193 static inline struct hlist_head
*dev_name_hash(const char *name
)
195 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
196 return &dev_name_head
[hash
& ((1<<NETDEV_HASHBITS
)-1)];
199 static inline struct hlist_head
*dev_index_hash(int ifindex
)
201 return &dev_index_head
[ifindex
& ((1<<NETDEV_HASHBITS
)-1)];
208 static RAW_NOTIFIER_HEAD(netdev_chain
);
211 * Device drivers call our routines to queue packets here. We empty the
212 * queue in the local softnet handler.
214 DEFINE_PER_CPU(struct softnet_data
, softnet_data
) = { NULL
};
217 extern int netdev_sysfs_init(void);
218 extern int netdev_register_sysfs(struct net_device
*);
219 extern void netdev_unregister_sysfs(struct net_device
*);
221 #define netdev_sysfs_init() (0)
222 #define netdev_register_sysfs(dev) (0)
223 #define netdev_unregister_sysfs(dev) do { } while(0)
226 #ifdef CONFIG_DEBUG_LOCK_ALLOC
228 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
229 * according to dev->type
231 static const unsigned short netdev_lock_type
[] =
232 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
233 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
234 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
235 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
236 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
237 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
238 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
239 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
240 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
241 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
242 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
243 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
244 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
245 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_VOID
,
248 static const char *netdev_lock_name
[] =
249 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
250 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
251 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
252 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
253 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
254 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
255 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
256 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
257 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
258 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
259 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
260 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
261 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
262 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
265 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
267 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
271 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
272 if (netdev_lock_type
[i
] == dev_type
)
274 /* the last key is used by default */
275 return ARRAY_SIZE(netdev_lock_type
) - 1;
278 static inline void netdev_set_lockdep_class(spinlock_t
*lock
,
279 unsigned short dev_type
)
283 i
= netdev_lock_pos(dev_type
);
284 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
285 netdev_lock_name
[i
]);
288 static inline void netdev_set_lockdep_class(spinlock_t
*lock
,
289 unsigned short dev_type
)
294 /*******************************************************************************
296 Protocol management and registration routines
298 *******************************************************************************/
301 * Add a protocol ID to the list. Now that the input handler is
302 * smarter we can dispense with all the messy stuff that used to be
305 * BEWARE!!! Protocol handlers, mangling input packets,
306 * MUST BE last in hash buckets and checking protocol handlers
307 * MUST start from promiscuous ptype_all chain in net_bh.
308 * It is true now, do not change it.
309 * Explanation follows: if protocol handler, mangling packet, will
310 * be the first on list, it is not able to sense, that packet
311 * is cloned and should be copied-on-write, so that it will
312 * change it and subsequent readers will get broken packet.
317 * dev_add_pack - add packet handler
318 * @pt: packet type declaration
320 * Add a protocol handler to the networking stack. The passed &packet_type
321 * is linked into kernel lists and may not be freed until it has been
322 * removed from the kernel lists.
324 * This call does not sleep therefore it can not
325 * guarantee all CPU's that are in middle of receiving packets
326 * will see the new packet type (until the next received packet).
329 void dev_add_pack(struct packet_type
*pt
)
333 spin_lock_bh(&ptype_lock
);
334 if (pt
->type
== htons(ETH_P_ALL
))
335 list_add_rcu(&pt
->list
, &ptype_all
);
337 hash
= ntohs(pt
->type
) & 15;
338 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
340 spin_unlock_bh(&ptype_lock
);
344 * __dev_remove_pack - remove packet handler
345 * @pt: packet type declaration
347 * Remove a protocol handler that was previously added to the kernel
348 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
349 * from the kernel lists and can be freed or reused once this function
352 * The packet type might still be in use by receivers
353 * and must not be freed until after all the CPU's have gone
354 * through a quiescent state.
356 void __dev_remove_pack(struct packet_type
*pt
)
358 struct list_head
*head
;
359 struct packet_type
*pt1
;
361 spin_lock_bh(&ptype_lock
);
363 if (pt
->type
== htons(ETH_P_ALL
))
366 head
= &ptype_base
[ntohs(pt
->type
) & 15];
368 list_for_each_entry(pt1
, head
, list
) {
370 list_del_rcu(&pt
->list
);
375 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
377 spin_unlock_bh(&ptype_lock
);
380 * dev_remove_pack - remove packet handler
381 * @pt: packet type declaration
383 * Remove a protocol handler that was previously added to the kernel
384 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
385 * from the kernel lists and can be freed or reused once this function
388 * This call sleeps to guarantee that no CPU is looking at the packet
391 void dev_remove_pack(struct packet_type
*pt
)
393 __dev_remove_pack(pt
);
398 /******************************************************************************
400 Device Boot-time Settings Routines
402 *******************************************************************************/
404 /* Boot time configuration table */
405 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
408 * netdev_boot_setup_add - add new setup entry
409 * @name: name of the device
410 * @map: configured settings for the device
412 * Adds new setup entry to the dev_boot_setup list. The function
413 * returns 0 on error and 1 on success. This is a generic routine to
416 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
418 struct netdev_boot_setup
*s
;
422 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
423 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
424 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
425 strcpy(s
[i
].name
, name
);
426 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
431 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
435 * netdev_boot_setup_check - check boot time settings
436 * @dev: the netdevice
438 * Check boot time settings for the device.
439 * The found settings are set for the device to be used
440 * later in the device probing.
441 * Returns 0 if no settings found, 1 if they are.
443 int netdev_boot_setup_check(struct net_device
*dev
)
445 struct netdev_boot_setup
*s
= dev_boot_setup
;
448 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
449 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
450 !strncmp(dev
->name
, s
[i
].name
, strlen(s
[i
].name
))) {
451 dev
->irq
= s
[i
].map
.irq
;
452 dev
->base_addr
= s
[i
].map
.base_addr
;
453 dev
->mem_start
= s
[i
].map
.mem_start
;
454 dev
->mem_end
= s
[i
].map
.mem_end
;
463 * netdev_boot_base - get address from boot time settings
464 * @prefix: prefix for network device
465 * @unit: id for network device
467 * Check boot time settings for the base address of device.
468 * The found settings are set for the device to be used
469 * later in the device probing.
470 * Returns 0 if no settings found.
472 unsigned long netdev_boot_base(const char *prefix
, int unit
)
474 const struct netdev_boot_setup
*s
= dev_boot_setup
;
478 sprintf(name
, "%s%d", prefix
, unit
);
481 * If device already registered then return base of 1
482 * to indicate not to probe for this interface
484 if (__dev_get_by_name(name
))
487 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
488 if (!strcmp(name
, s
[i
].name
))
489 return s
[i
].map
.base_addr
;
494 * Saves at boot time configured settings for any netdevice.
496 int __init
netdev_boot_setup(char *str
)
501 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
506 memset(&map
, 0, sizeof(map
));
510 map
.base_addr
= ints
[2];
512 map
.mem_start
= ints
[3];
514 map
.mem_end
= ints
[4];
516 /* Add new entry to the list */
517 return netdev_boot_setup_add(str
, &map
);
520 __setup("netdev=", netdev_boot_setup
);
522 /*******************************************************************************
524 Device Interface Subroutines
526 *******************************************************************************/
529 * __dev_get_by_name - find a device by its name
530 * @name: name to find
532 * Find an interface by name. Must be called under RTNL semaphore
533 * or @dev_base_lock. If the name is found a pointer to the device
534 * is returned. If the name is not found then %NULL is returned. The
535 * reference counters are not incremented so the caller must be
536 * careful with locks.
539 struct net_device
*__dev_get_by_name(const char *name
)
541 struct hlist_node
*p
;
543 hlist_for_each(p
, dev_name_hash(name
)) {
544 struct net_device
*dev
545 = hlist_entry(p
, struct net_device
, name_hlist
);
546 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
553 * dev_get_by_name - find a device by its name
554 * @name: name to find
556 * Find an interface by name. This can be called from any
557 * context and does its own locking. The returned handle has
558 * the usage count incremented and the caller must use dev_put() to
559 * release it when it is no longer needed. %NULL is returned if no
560 * matching device is found.
563 struct net_device
*dev_get_by_name(const char *name
)
565 struct net_device
*dev
;
567 read_lock(&dev_base_lock
);
568 dev
= __dev_get_by_name(name
);
571 read_unlock(&dev_base_lock
);
576 * __dev_get_by_index - find a device by its ifindex
577 * @ifindex: index of device
579 * Search for an interface by index. Returns %NULL if the device
580 * is not found or a pointer to the device. The device has not
581 * had its reference counter increased so the caller must be careful
582 * about locking. The caller must hold either the RTNL semaphore
586 struct net_device
*__dev_get_by_index(int ifindex
)
588 struct hlist_node
*p
;
590 hlist_for_each(p
, dev_index_hash(ifindex
)) {
591 struct net_device
*dev
592 = hlist_entry(p
, struct net_device
, index_hlist
);
593 if (dev
->ifindex
== ifindex
)
601 * dev_get_by_index - find a device by its ifindex
602 * @ifindex: index of device
604 * Search for an interface by index. Returns NULL if the device
605 * is not found or a pointer to the device. The device returned has
606 * had a reference added and the pointer is safe until the user calls
607 * dev_put to indicate they have finished with it.
610 struct net_device
*dev_get_by_index(int ifindex
)
612 struct net_device
*dev
;
614 read_lock(&dev_base_lock
);
615 dev
= __dev_get_by_index(ifindex
);
618 read_unlock(&dev_base_lock
);
623 * dev_getbyhwaddr - find a device by its hardware address
624 * @type: media type of device
625 * @ha: hardware address
627 * Search for an interface by MAC address. Returns NULL if the device
628 * is not found or a pointer to the device. The caller must hold the
629 * rtnl semaphore. The returned device has not had its ref count increased
630 * and the caller must therefore be careful about locking
633 * If the API was consistent this would be __dev_get_by_hwaddr
636 struct net_device
*dev_getbyhwaddr(unsigned short type
, char *ha
)
638 struct net_device
*dev
;
643 if (dev
->type
== type
&&
644 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
650 EXPORT_SYMBOL(dev_getbyhwaddr
);
652 struct net_device
*__dev_getfirstbyhwtype(unsigned short type
)
654 struct net_device
*dev
;
658 if (dev
->type
== type
)
664 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
666 struct net_device
*dev_getfirstbyhwtype(unsigned short type
)
668 struct net_device
*dev
;
671 dev
= __dev_getfirstbyhwtype(type
);
678 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
681 * dev_get_by_flags - find any device with given flags
682 * @if_flags: IFF_* values
683 * @mask: bitmask of bits in if_flags to check
685 * Search for any interface with the given flags. Returns NULL if a device
686 * is not found or a pointer to the device. The device returned has
687 * had a reference added and the pointer is safe until the user calls
688 * dev_put to indicate they have finished with it.
691 struct net_device
* dev_get_by_flags(unsigned short if_flags
, unsigned short mask
)
693 struct net_device
*dev
, *ret
;
696 read_lock(&dev_base_lock
);
697 for_each_netdev(dev
) {
698 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
704 read_unlock(&dev_base_lock
);
709 * dev_valid_name - check if name is okay for network device
712 * Network device names need to be valid file names to
713 * to allow sysfs to work. We also disallow any kind of
716 int dev_valid_name(const char *name
)
720 if (strlen(name
) >= IFNAMSIZ
)
722 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
726 if (*name
== '/' || isspace(*name
))
734 * dev_alloc_name - allocate a name for a device
736 * @name: name format string
738 * Passed a format string - eg "lt%d" it will try and find a suitable
739 * id. It scans list of devices to build up a free map, then chooses
740 * the first empty slot. The caller must hold the dev_base or rtnl lock
741 * while allocating the name and adding the device in order to avoid
743 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
744 * Returns the number of the unit assigned or a negative errno code.
747 int dev_alloc_name(struct net_device
*dev
, const char *name
)
752 const int max_netdevices
= 8*PAGE_SIZE
;
754 struct net_device
*d
;
756 p
= strnchr(name
, IFNAMSIZ
-1, '%');
759 * Verify the string as this thing may have come from
760 * the user. There must be either one "%d" and no other "%"
763 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
766 /* Use one page as a bit array of possible slots */
767 inuse
= (long *) get_zeroed_page(GFP_ATOMIC
);
772 if (!sscanf(d
->name
, name
, &i
))
774 if (i
< 0 || i
>= max_netdevices
)
777 /* avoid cases where sscanf is not exact inverse of printf */
778 snprintf(buf
, sizeof(buf
), name
, i
);
779 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
783 i
= find_first_zero_bit(inuse
, max_netdevices
);
784 free_page((unsigned long) inuse
);
787 snprintf(buf
, sizeof(buf
), name
, i
);
788 if (!__dev_get_by_name(buf
)) {
789 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
793 /* It is possible to run out of possible slots
794 * when the name is long and there isn't enough space left
795 * for the digits, or if all bits are used.
802 * dev_change_name - change name of a device
804 * @newname: name (or format string) must be at least IFNAMSIZ
806 * Change name of a device, can pass format strings "eth%d".
809 int dev_change_name(struct net_device
*dev
, char *newname
)
815 if (dev
->flags
& IFF_UP
)
818 if (!dev_valid_name(newname
))
821 if (strchr(newname
, '%')) {
822 err
= dev_alloc_name(dev
, newname
);
825 strcpy(newname
, dev
->name
);
827 else if (__dev_get_by_name(newname
))
830 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
832 device_rename(&dev
->dev
, dev
->name
);
833 hlist_del(&dev
->name_hlist
);
834 hlist_add_head(&dev
->name_hlist
, dev_name_hash(dev
->name
));
835 raw_notifier_call_chain(&netdev_chain
, NETDEV_CHANGENAME
, dev
);
841 * netdev_features_change - device changes features
842 * @dev: device to cause notification
844 * Called to indicate a device has changed features.
846 void netdev_features_change(struct net_device
*dev
)
848 raw_notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
850 EXPORT_SYMBOL(netdev_features_change
);
853 * netdev_state_change - device changes state
854 * @dev: device to cause notification
856 * Called to indicate a device has changed state. This function calls
857 * the notifier chains for netdev_chain and sends a NEWLINK message
858 * to the routing socket.
860 void netdev_state_change(struct net_device
*dev
)
862 if (dev
->flags
& IFF_UP
) {
863 raw_notifier_call_chain(&netdev_chain
,
865 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
870 * dev_load - load a network module
871 * @name: name of interface
873 * If a network interface is not present and the process has suitable
874 * privileges this function loads the module. If module loading is not
875 * available in this kernel then it becomes a nop.
878 void dev_load(const char *name
)
880 struct net_device
*dev
;
882 read_lock(&dev_base_lock
);
883 dev
= __dev_get_by_name(name
);
884 read_unlock(&dev_base_lock
);
886 if (!dev
&& capable(CAP_SYS_MODULE
))
887 request_module("%s", name
);
890 static int default_rebuild_header(struct sk_buff
*skb
)
892 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
893 skb
->dev
? skb
->dev
->name
: "NULL!!!");
899 * dev_open - prepare an interface for use.
900 * @dev: device to open
902 * Takes a device from down to up state. The device's private open
903 * function is invoked and then the multicast lists are loaded. Finally
904 * the device is moved into the up state and a %NETDEV_UP message is
905 * sent to the netdev notifier chain.
907 * Calling this function on an active interface is a nop. On a failure
908 * a negative errno code is returned.
910 int dev_open(struct net_device
*dev
)
918 if (dev
->flags
& IFF_UP
)
922 * Is it even present?
924 if (!netif_device_present(dev
))
928 * Call device private open method
930 set_bit(__LINK_STATE_START
, &dev
->state
);
932 ret
= dev
->open(dev
);
934 clear_bit(__LINK_STATE_START
, &dev
->state
);
938 * If it went open OK then:
945 dev
->flags
|= IFF_UP
;
948 * Initialize multicasting status
953 * Wakeup transmit queue engine
958 * ... and announce new interface.
960 raw_notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
966 * dev_close - shutdown an interface.
967 * @dev: device to shutdown
969 * This function moves an active device into down state. A
970 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
971 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
974 int dev_close(struct net_device
*dev
)
976 if (!(dev
->flags
& IFF_UP
))
980 * Tell people we are going down, so that they can
981 * prepare to death, when device is still operating.
983 raw_notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
987 clear_bit(__LINK_STATE_START
, &dev
->state
);
989 /* Synchronize to scheduled poll. We cannot touch poll list,
990 * it can be even on different cpu. So just clear netif_running(),
991 * and wait when poll really will happen. Actually, the best place
992 * for this is inside dev->stop() after device stopped its irq
993 * engine, but this requires more changes in devices. */
995 smp_mb__after_clear_bit(); /* Commit netif_running(). */
996 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
1002 * Call the device specific close. This cannot fail.
1003 * Only if device is UP
1005 * We allow it to be called even after a DETACH hot-plug
1012 * Device is now down.
1015 dev
->flags
&= ~IFF_UP
;
1018 * Tell people we are down
1020 raw_notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
1027 * Device change register/unregister. These are not inline or static
1028 * as we export them to the world.
1032 * register_netdevice_notifier - register a network notifier block
1035 * Register a notifier to be called when network device events occur.
1036 * The notifier passed is linked into the kernel structures and must
1037 * not be reused until it has been unregistered. A negative errno code
1038 * is returned on a failure.
1040 * When registered all registration and up events are replayed
1041 * to the new notifier to allow device to have a race free
1042 * view of the network device list.
1045 int register_netdevice_notifier(struct notifier_block
*nb
)
1047 struct net_device
*dev
;
1051 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1053 for_each_netdev(dev
) {
1054 nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1056 if (dev
->flags
& IFF_UP
)
1057 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1065 * unregister_netdevice_notifier - unregister a network notifier block
1068 * Unregister a notifier previously registered by
1069 * register_netdevice_notifier(). The notifier is unlinked into the
1070 * kernel structures and may then be reused. A negative errno code
1071 * is returned on a failure.
1074 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1079 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1085 * call_netdevice_notifiers - call all network notifier blocks
1086 * @val: value passed unmodified to notifier function
1087 * @v: pointer passed unmodified to notifier function
1089 * Call all network notifier blocks. Parameters and return value
1090 * are as for raw_notifier_call_chain().
1093 int call_netdevice_notifiers(unsigned long val
, void *v
)
1095 return raw_notifier_call_chain(&netdev_chain
, val
, v
);
1098 /* When > 0 there are consumers of rx skb time stamps */
1099 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1101 void net_enable_timestamp(void)
1103 atomic_inc(&netstamp_needed
);
1106 void net_disable_timestamp(void)
1108 atomic_dec(&netstamp_needed
);
1111 static inline void net_timestamp(struct sk_buff
*skb
)
1113 if (atomic_read(&netstamp_needed
))
1114 __net_timestamp(skb
);
1116 skb
->tstamp
.tv64
= 0;
1120 * Support routine. Sends outgoing frames to any network
1121 * taps currently in use.
1124 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1126 struct packet_type
*ptype
;
1130 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1131 /* Never send packets back to the socket
1132 * they originated from - MvS (miquels@drinkel.ow.org)
1134 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1135 (ptype
->af_packet_priv
== NULL
||
1136 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1137 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1141 net_timestamp(skb2
);
1143 /* skb->nh should be correctly
1144 set by sender, so that the second statement is
1145 just protection against buggy protocols.
1147 skb_reset_mac_header(skb2
);
1149 if (skb_network_header(skb2
) < skb2
->data
||
1150 skb2
->network_header
> skb2
->tail
) {
1151 if (net_ratelimit())
1152 printk(KERN_CRIT
"protocol %04x is "
1154 skb2
->protocol
, dev
->name
);
1155 skb_reset_network_header(skb2
);
1158 skb2
->transport_header
= skb2
->network_header
;
1159 skb2
->pkt_type
= PACKET_OUTGOING
;
1160 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1167 void __netif_schedule(struct net_device
*dev
)
1169 if (!test_and_set_bit(__LINK_STATE_SCHED
, &dev
->state
)) {
1170 unsigned long flags
;
1171 struct softnet_data
*sd
;
1173 local_irq_save(flags
);
1174 sd
= &__get_cpu_var(softnet_data
);
1175 dev
->next_sched
= sd
->output_queue
;
1176 sd
->output_queue
= dev
;
1177 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1178 local_irq_restore(flags
);
1181 EXPORT_SYMBOL(__netif_schedule
);
1183 void __netif_rx_schedule(struct net_device
*dev
)
1185 unsigned long flags
;
1187 local_irq_save(flags
);
1189 list_add_tail(&dev
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
1191 dev
->quota
+= dev
->weight
;
1193 dev
->quota
= dev
->weight
;
1194 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1195 local_irq_restore(flags
);
1197 EXPORT_SYMBOL(__netif_rx_schedule
);
1199 void dev_kfree_skb_any(struct sk_buff
*skb
)
1201 if (in_irq() || irqs_disabled())
1202 dev_kfree_skb_irq(skb
);
1206 EXPORT_SYMBOL(dev_kfree_skb_any
);
1210 void netif_device_detach(struct net_device
*dev
)
1212 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1213 netif_running(dev
)) {
1214 netif_stop_queue(dev
);
1217 EXPORT_SYMBOL(netif_device_detach
);
1219 void netif_device_attach(struct net_device
*dev
)
1221 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1222 netif_running(dev
)) {
1223 netif_wake_queue(dev
);
1224 __netdev_watchdog_up(dev
);
1227 EXPORT_SYMBOL(netif_device_attach
);
1231 * Invalidate hardware checksum when packet is to be mangled, and
1232 * complete checksum manually on outgoing path.
1234 int skb_checksum_help(struct sk_buff
*skb
)
1237 int ret
= 0, offset
;
1239 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1240 goto out_set_summed
;
1242 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1243 /* Let GSO fix up the checksum. */
1244 goto out_set_summed
;
1247 offset
= skb
->csum_start
- skb_headroom(skb
);
1248 BUG_ON(offset
>= skb_headlen(skb
));
1249 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1251 offset
+= skb
->csum_offset
;
1252 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1254 if (skb_cloned(skb
) &&
1255 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1256 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1261 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1263 skb
->ip_summed
= CHECKSUM_NONE
;
1269 * skb_gso_segment - Perform segmentation on skb.
1270 * @skb: buffer to segment
1271 * @features: features for the output path (see dev->features)
1273 * This function segments the given skb and returns a list of segments.
1275 * It may return NULL if the skb requires no segmentation. This is
1276 * only possible when GSO is used for verifying header integrity.
1278 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1280 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1281 struct packet_type
*ptype
;
1282 __be16 type
= skb
->protocol
;
1285 BUG_ON(skb_shinfo(skb
)->frag_list
);
1287 skb_reset_mac_header(skb
);
1288 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1289 __skb_pull(skb
, skb
->mac_len
);
1291 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1292 if (skb_header_cloned(skb
) &&
1293 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1294 return ERR_PTR(err
);
1298 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
) & 15], list
) {
1299 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1300 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1301 err
= ptype
->gso_send_check(skb
);
1302 segs
= ERR_PTR(err
);
1303 if (err
|| skb_gso_ok(skb
, features
))
1305 __skb_push(skb
, (skb
->data
-
1306 skb_network_header(skb
)));
1308 segs
= ptype
->gso_segment(skb
, features
);
1314 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1319 EXPORT_SYMBOL(skb_gso_segment
);
1321 /* Take action when hardware reception checksum errors are detected. */
1323 void netdev_rx_csum_fault(struct net_device
*dev
)
1325 if (net_ratelimit()) {
1326 printk(KERN_ERR
"%s: hw csum failure.\n",
1327 dev
? dev
->name
: "<unknown>");
1331 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1334 /* Actually, we should eliminate this check as soon as we know, that:
1335 * 1. IOMMU is present and allows to map all the memory.
1336 * 2. No high memory really exists on this machine.
1339 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1341 #ifdef CONFIG_HIGHMEM
1344 if (dev
->features
& NETIF_F_HIGHDMA
)
1347 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1348 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1356 void (*destructor
)(struct sk_buff
*skb
);
1359 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1361 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1363 struct dev_gso_cb
*cb
;
1366 struct sk_buff
*nskb
= skb
->next
;
1368 skb
->next
= nskb
->next
;
1371 } while (skb
->next
);
1373 cb
= DEV_GSO_CB(skb
);
1375 cb
->destructor(skb
);
1379 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1380 * @skb: buffer to segment
1382 * This function segments the given skb and stores the list of segments
1385 static int dev_gso_segment(struct sk_buff
*skb
)
1387 struct net_device
*dev
= skb
->dev
;
1388 struct sk_buff
*segs
;
1389 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1392 segs
= skb_gso_segment(skb
, features
);
1394 /* Verifying header integrity only. */
1398 if (unlikely(IS_ERR(segs
)))
1399 return PTR_ERR(segs
);
1402 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1403 skb
->destructor
= dev_gso_skb_destructor
;
1408 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1410 if (likely(!skb
->next
)) {
1411 if (!list_empty(&ptype_all
))
1412 #if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1413 if (!(skb
->imq_flags
& IMQ_F_ENQUEUE
))
1415 dev_queue_xmit_nit(skb
, dev
);
1417 if (netif_needs_gso(dev
, skb
)) {
1418 if (unlikely(dev_gso_segment(skb
)))
1424 return dev
->hard_start_xmit(skb
, dev
);
1429 struct sk_buff
*nskb
= skb
->next
;
1432 skb
->next
= nskb
->next
;
1434 rc
= dev
->hard_start_xmit(nskb
, dev
);
1436 nskb
->next
= skb
->next
;
1440 if (unlikely(netif_queue_stopped(dev
) && skb
->next
))
1441 return NETDEV_TX_BUSY
;
1442 } while (skb
->next
);
1444 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1451 #define HARD_TX_LOCK(dev, cpu) { \
1452 if ((dev->features & NETIF_F_LLTX) == 0) { \
1453 netif_tx_lock(dev); \
1457 #define HARD_TX_UNLOCK(dev) { \
1458 if ((dev->features & NETIF_F_LLTX) == 0) { \
1459 netif_tx_unlock(dev); \
1464 * dev_queue_xmit - transmit a buffer
1465 * @skb: buffer to transmit
1467 * Queue a buffer for transmission to a network device. The caller must
1468 * have set the device and priority and built the buffer before calling
1469 * this function. The function can be called from an interrupt.
1471 * A negative errno code is returned on a failure. A success does not
1472 * guarantee the frame will be transmitted as it may be dropped due
1473 * to congestion or traffic shaping.
1475 * -----------------------------------------------------------------------------------
1476 * I notice this method can also return errors from the queue disciplines,
1477 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1480 * Regardless of the return value, the skb is consumed, so it is currently
1481 * difficult to retry a send to this method. (You can bump the ref count
1482 * before sending to hold a reference for retry if you are careful.)
1484 * When calling this method, interrupts MUST be enabled. This is because
1485 * the BH enable code must have IRQs enabled so that it will not deadlock.
1489 int BCMFASTPATH
dev_queue_xmit(struct sk_buff
*skb
)
1491 struct net_device
*dev
= skb
->dev
;
1495 /* GSO will handle the following emulations directly. */
1496 if (netif_needs_gso(dev
, skb
))
1499 if (skb_shinfo(skb
)->frag_list
&&
1500 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1501 __skb_linearize(skb
))
1504 /* Fragmented skb is linearized if device does not support SG,
1505 * or if at least one of fragments is in highmem and device
1506 * does not support DMA from it.
1508 if (skb_shinfo(skb
)->nr_frags
&&
1509 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1510 __skb_linearize(skb
))
1513 /* If packet is not checksummed and device does not support
1514 * checksumming for this protocol, complete checksumming here.
1516 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1517 skb_set_transport_header(skb
, skb
->csum_start
-
1520 if (!(dev
->features
& NETIF_F_GEN_CSUM
) &&
1521 (!(dev
->features
& NETIF_F_IP_CSUM
) ||
1522 skb
->protocol
!= htons(ETH_P_IP
)))
1523 if (skb_checksum_help(skb
))
1528 spin_lock_prefetch(&dev
->queue_lock
);
1530 /* Disable soft irqs for various locks below. Also
1531 * stops preemption for RCU.
1535 /* Updates of qdisc are serialized by queue_lock.
1536 * The struct Qdisc which is pointed to by qdisc is now a
1537 * rcu structure - it may be accessed without acquiring
1538 * a lock (but the structure may be stale.) The freeing of the
1539 * qdisc will be deferred until it's known that there are no
1540 * more references to it.
1542 * If the qdisc has an enqueue function, we still need to
1543 * hold the queue_lock before calling it, since queue_lock
1544 * also serializes access to the device queue.
1547 q
= rcu_dereference(dev
->qdisc
);
1548 #ifdef CONFIG_NET_CLS_ACT
1549 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1552 /* Grab device queue */
1553 spin_lock(&dev
->queue_lock
);
1556 rc
= q
->enqueue(skb
, q
);
1558 spin_unlock(&dev
->queue_lock
);
1560 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1563 spin_unlock(&dev
->queue_lock
);
1566 /* The device has no queue. Common case for software devices:
1567 loopback, all the sorts of tunnels...
1569 Really, it is unlikely that netif_tx_lock protection is necessary
1570 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1572 However, it is possible, that they rely on protection
1575 Check this and shot the lock. It is not prone from deadlocks.
1576 Either shot noqueue qdisc, it is even simpler 8)
1578 if (dev
->flags
& IFF_UP
) {
1579 int cpu
= smp_processor_id(); /* ok because BHs are off */
1581 if (dev
->xmit_lock_owner
!= cpu
) {
1583 HARD_TX_LOCK(dev
, cpu
);
1585 if (!netif_queue_stopped(dev
)) {
1587 if (!dev_hard_start_xmit(skb
, dev
)) {
1588 HARD_TX_UNLOCK(dev
);
1592 HARD_TX_UNLOCK(dev
);
1593 if (net_ratelimit())
1594 printk(KERN_CRIT
"Virtual device %s asks to "
1595 "queue packet!\n", dev
->name
);
1597 /* Recursion is detected! It is possible,
1599 if (net_ratelimit())
1600 printk(KERN_CRIT
"Dead loop on virtual device "
1601 "%s, fix it urgently!\n", dev
->name
);
1606 rcu_read_unlock_bh();
1612 rcu_read_unlock_bh();
1617 /*=======================================================================
1619 =======================================================================*/
1621 int netdev_max_backlog __read_mostly
= 1000;
1622 int netdev_budget __read_mostly
= 300;
1623 int weight_p __read_mostly
= 64; /* old backlog weight */
1625 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1629 * netif_rx - post buffer to the network code
1630 * @skb: buffer to post
1632 * This function receives a packet from a device driver and queues it for
1633 * the upper (protocol) levels to process. It always succeeds. The buffer
1634 * may be dropped during processing for congestion control or by the
1638 * NET_RX_SUCCESS (no congestion)
1639 * NET_RX_CN_LOW (low congestion)
1640 * NET_RX_CN_MOD (moderate congestion)
1641 * NET_RX_CN_HIGH (high congestion)
1642 * NET_RX_DROP (packet was dropped)
1646 int netif_rx(struct sk_buff
*skb
)
1648 struct softnet_data
*queue
;
1649 unsigned long flags
;
1651 /* if netpoll wants it, pretend we never saw it */
1652 if (netpoll_rx(skb
))
1655 if (!skb
->tstamp
.tv64
)
1659 * The code is rearranged so that the path is the most
1660 * short when CPU is congested, but is still operating.
1662 local_irq_save(flags
);
1663 queue
= &__get_cpu_var(softnet_data
);
1665 __get_cpu_var(netdev_rx_stat
).total
++;
1666 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1667 if (queue
->input_pkt_queue
.qlen
) {
1670 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1671 local_irq_restore(flags
);
1672 return NET_RX_SUCCESS
;
1675 netif_rx_schedule(&queue
->backlog_dev
);
1679 __get_cpu_var(netdev_rx_stat
).dropped
++;
1680 local_irq_restore(flags
);
1686 int netif_rx_ni(struct sk_buff
*skb
)
1691 err
= netif_rx(skb
);
1692 if (local_softirq_pending())
1699 EXPORT_SYMBOL(netif_rx_ni
);
1701 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1703 struct net_device
*dev
= skb
->dev
;
1706 if (skb_bond_should_drop(skb
)) {
1710 skb
->dev
= dev
->master
;
1716 static void net_tx_action(struct softirq_action
*h
)
1718 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1720 if (sd
->completion_queue
) {
1721 struct sk_buff
*clist
;
1723 local_irq_disable();
1724 clist
= sd
->completion_queue
;
1725 sd
->completion_queue
= NULL
;
1729 struct sk_buff
*skb
= clist
;
1730 clist
= clist
->next
;
1732 BUG_TRAP(!atomic_read(&skb
->users
));
1737 if (sd
->output_queue
) {
1738 struct net_device
*head
;
1740 local_irq_disable();
1741 head
= sd
->output_queue
;
1742 sd
->output_queue
= NULL
;
1746 struct net_device
*dev
= head
;
1747 head
= head
->next_sched
;
1749 smp_mb__before_clear_bit();
1750 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1752 if (spin_trylock(&dev
->queue_lock
)) {
1754 spin_unlock(&dev
->queue_lock
);
1756 netif_schedule(dev
);
1762 static inline int deliver_skb(struct sk_buff
*skb
,
1763 struct packet_type
*pt_prev
,
1764 struct net_device
*orig_dev
)
1766 atomic_inc(&skb
->users
);
1767 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1770 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1771 /* These hooks defined here for ATM */
1773 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1774 unsigned char *addr
);
1775 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
1778 * If bridge module is loaded call bridging hook.
1779 * returns NULL if packet was consumed.
1781 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
1782 struct sk_buff
*skb
) __read_mostly
;
1783 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
1784 struct packet_type
**pt_prev
, int *ret
,
1785 struct net_device
*orig_dev
)
1787 struct net_bridge_port
*port
;
1789 if (skb
->pkt_type
== PACKET_LOOPBACK
||
1790 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
1794 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1798 return br_handle_frame_hook(port
, skb
);
1801 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1804 #ifdef CONFIG_NET_CLS_ACT
1805 /* TODO: Maybe we should just force sch_ingress to be compiled in
1806 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1807 * a compare and 2 stores extra right now if we dont have it on
1808 * but have CONFIG_NET_CLS_ACT
1809 * NOTE: This doesnt stop any functionality; if you dont have
1810 * the ingress scheduler, you just cant add policies on ingress.
1813 static int ing_filter(struct sk_buff
*skb
)
1816 struct net_device
*dev
= skb
->dev
;
1817 int result
= TC_ACT_OK
;
1818 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
1820 if (MAX_RED_LOOP
< ttl
++) {
1822 "Redir loop detected Dropping packet (%d->%d)\n",
1823 skb
->iif
, dev
->ifindex
);
1827 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
1828 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
1830 spin_lock(&dev
->ingress_lock
);
1831 if ((q
= dev
->qdisc_ingress
) != NULL
)
1832 result
= q
->enqueue(skb
, q
);
1833 spin_unlock(&dev
->ingress_lock
);
1838 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
1839 struct packet_type
**pt_prev
,
1840 int *ret
, struct net_device
*orig_dev
)
1842 if (!skb
->dev
->qdisc_ingress
)
1846 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1849 /* Huh? Why does turning on AF_PACKET affect this? */
1850 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1853 switch (ing_filter(skb
)) {
1866 int netif_receive_skb(struct sk_buff
*skb
)
1868 struct packet_type
*ptype
, *pt_prev
;
1869 struct net_device
*orig_dev
;
1870 int ret
= NET_RX_DROP
;
1873 /* if we've gotten here through NAPI, check netpoll */
1874 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1877 if (!skb
->tstamp
.tv64
)
1881 skb
->iif
= skb
->dev
->ifindex
;
1883 orig_dev
= skb_bond(skb
);
1888 __get_cpu_var(netdev_rx_stat
).total
++;
1890 skb_reset_network_header(skb
);
1891 skb_reset_transport_header(skb
);
1892 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1898 #ifdef CONFIG_NET_CLS_ACT
1899 if (skb
->tc_verd
& TC_NCLS
) {
1900 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1905 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1906 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1908 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1913 #ifdef CONFIG_NET_CLS_ACT
1914 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
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
);
1955 backlog_dev
->weight
= weight_p
;
1957 struct sk_buff
*skb
;
1958 struct net_device
*dev
;
1960 local_irq_disable();
1961 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
1968 netif_receive_skb(skb
);
1979 backlog_dev
->quota
-= work
;
1984 backlog_dev
->quota
-= work
;
1987 list_del(&backlog_dev
->poll_list
);
1988 smp_mb__before_clear_bit();
1989 netif_poll_enable(backlog_dev
);
1995 static void net_rx_action(struct softirq_action
*h
)
1997 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1998 unsigned long start_time
= jiffies
;
1999 int budget
= netdev_budget
;
2002 local_irq_disable();
2004 while (!list_empty(&queue
->poll_list
)) {
2005 struct net_device
*dev
;
2007 if (budget
<= 0 || jiffies
- start_time
> 1)
2012 dev
= list_entry(queue
->poll_list
.next
,
2013 struct net_device
, poll_list
);
2014 have
= netpoll_poll_lock(dev
);
2016 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
2017 netpoll_poll_unlock(have
);
2018 local_irq_disable();
2019 list_move_tail(&dev
->poll_list
, &queue
->poll_list
);
2021 dev
->quota
+= dev
->weight
;
2023 dev
->quota
= dev
->weight
;
2025 netpoll_poll_unlock(have
);
2027 local_irq_disable();
2032 #ifdef CONFIG_NET_DMA
2034 * There may not be any more sk_buffs coming right now, so push
2035 * any pending DMA copies to hardware
2037 if (net_dma_client
) {
2038 struct dma_chan
*chan
;
2040 list_for_each_entry_rcu(chan
, &net_dma_client
->channels
, client_node
)
2041 dma_async_memcpy_issue_pending(chan
);
2048 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2049 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2053 static gifconf_func_t
* gifconf_list
[NPROTO
];
2056 * register_gifconf - register a SIOCGIF handler
2057 * @family: Address family
2058 * @gifconf: Function handler
2060 * Register protocol dependent address dumping routines. The handler
2061 * that is passed must not be freed or reused until it has been replaced
2062 * by another handler.
2064 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2066 if (family
>= NPROTO
)
2068 gifconf_list
[family
] = gifconf
;
2074 * Map an interface index to its name (SIOCGIFNAME)
2078 * We need this ioctl for efficient implementation of the
2079 * if_indextoname() function required by the IPv6 API. Without
2080 * it, we would have to search all the interfaces to find a
2084 static int dev_ifname(struct ifreq __user
*arg
)
2086 struct net_device
*dev
;
2090 * Fetch the caller's info block.
2093 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2096 read_lock(&dev_base_lock
);
2097 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
2099 read_unlock(&dev_base_lock
);
2103 strcpy(ifr
.ifr_name
, dev
->name
);
2104 read_unlock(&dev_base_lock
);
2106 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2112 * Perform a SIOCGIFCONF call. This structure will change
2113 * size eventually, and there is nothing I can do about it.
2114 * Thus we will need a 'compatibility mode'.
2117 static int dev_ifconf(char __user
*arg
)
2120 struct net_device
*dev
;
2127 * Fetch the caller's info block.
2130 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2137 * Loop over the interfaces, and write an info block for each.
2141 for_each_netdev(dev
) {
2142 for (i
= 0; i
< NPROTO
; i
++) {
2143 if (gifconf_list
[i
]) {
2146 done
= gifconf_list
[i
](dev
, NULL
, 0);
2148 done
= gifconf_list
[i
](dev
, pos
+ total
,
2158 * All done. Write the updated control block back to the caller.
2160 ifc
.ifc_len
= total
;
2163 * Both BSD and Solaris return 0 here, so we do too.
2165 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2168 #ifdef CONFIG_PROC_FS
2170 * This is invoked by the /proc filesystem handler to display a device
2173 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2176 struct net_device
*dev
;
2178 read_lock(&dev_base_lock
);
2180 return SEQ_START_TOKEN
;
2183 for_each_netdev(dev
)
2190 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2193 return v
== SEQ_START_TOKEN
?
2194 first_net_device() : next_net_device((struct net_device
*)v
);
2197 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2199 read_unlock(&dev_base_lock
);
2202 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2204 struct net_device_stats
*stats
= dev
->get_stats(dev
);
2206 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2207 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2208 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2210 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2211 stats
->rx_fifo_errors
,
2212 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2213 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2214 stats
->rx_compressed
, stats
->multicast
,
2215 stats
->tx_bytes
, stats
->tx_packets
,
2216 stats
->tx_errors
, stats
->tx_dropped
,
2217 stats
->tx_fifo_errors
, stats
->collisions
,
2218 stats
->tx_carrier_errors
+
2219 stats
->tx_aborted_errors
+
2220 stats
->tx_window_errors
+
2221 stats
->tx_heartbeat_errors
,
2222 stats
->tx_compressed
);
2226 * Called from the PROCfs module. This now uses the new arbitrary sized
2227 * /proc/net interface to create /proc/net/dev
2229 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2231 if (v
== SEQ_START_TOKEN
)
2232 seq_puts(seq
, "Inter-| Receive "
2234 " face |bytes packets errs drop fifo frame "
2235 "compressed multicast|bytes packets errs "
2236 "drop fifo colls carrier compressed\n");
2238 dev_seq_printf_stats(seq
, v
);
2242 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2244 struct netif_rx_stats
*rc
= NULL
;
2246 while (*pos
< NR_CPUS
)
2247 if (cpu_online(*pos
)) {
2248 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2255 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2257 return softnet_get_online(pos
);
2260 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2263 return softnet_get_online(pos
);
2266 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2270 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2272 struct netif_rx_stats
*s
= v
;
2274 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2275 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2276 0, 0, 0, 0, /* was fastroute */
2281 static const struct seq_operations dev_seq_ops
= {
2282 .start
= dev_seq_start
,
2283 .next
= dev_seq_next
,
2284 .stop
= dev_seq_stop
,
2285 .show
= dev_seq_show
,
2288 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2290 return seq_open(file
, &dev_seq_ops
);
2293 static const struct file_operations dev_seq_fops
= {
2294 .owner
= THIS_MODULE
,
2295 .open
= dev_seq_open
,
2297 .llseek
= seq_lseek
,
2298 .release
= seq_release
,
2301 static const struct seq_operations softnet_seq_ops
= {
2302 .start
= softnet_seq_start
,
2303 .next
= softnet_seq_next
,
2304 .stop
= softnet_seq_stop
,
2305 .show
= softnet_seq_show
,
2308 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2310 return seq_open(file
, &softnet_seq_ops
);
2313 static const struct file_operations softnet_seq_fops
= {
2314 .owner
= THIS_MODULE
,
2315 .open
= softnet_seq_open
,
2317 .llseek
= seq_lseek
,
2318 .release
= seq_release
,
2321 static void *ptype_get_idx(loff_t pos
)
2323 struct packet_type
*pt
= NULL
;
2327 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
2333 for (t
= 0; t
< 16; t
++) {
2334 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
2343 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2346 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
2349 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2351 struct packet_type
*pt
;
2352 struct list_head
*nxt
;
2356 if (v
== SEQ_START_TOKEN
)
2357 return ptype_get_idx(0);
2360 nxt
= pt
->list
.next
;
2361 if (pt
->type
== htons(ETH_P_ALL
)) {
2362 if (nxt
!= &ptype_all
)
2365 nxt
= ptype_base
[0].next
;
2367 hash
= ntohs(pt
->type
) & 15;
2369 while (nxt
== &ptype_base
[hash
]) {
2372 nxt
= ptype_base
[hash
].next
;
2375 return list_entry(nxt
, struct packet_type
, list
);
2378 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
2383 static void ptype_seq_decode(struct seq_file
*seq
, void *sym
)
2385 #ifdef CONFIG_KALLSYMS
2386 unsigned long offset
= 0, symsize
;
2387 const char *symname
;
2391 symname
= kallsyms_lookup((unsigned long)sym
, &symsize
, &offset
,
2398 modname
= delim
= "";
2399 seq_printf(seq
, "%s%s%s%s+0x%lx", delim
, modname
, delim
,
2405 seq_printf(seq
, "[%p]", sym
);
2408 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
2410 struct packet_type
*pt
= v
;
2412 if (v
== SEQ_START_TOKEN
)
2413 seq_puts(seq
, "Type Device Function\n");
2415 if (pt
->type
== htons(ETH_P_ALL
))
2416 seq_puts(seq
, "ALL ");
2418 seq_printf(seq
, "%04x", ntohs(pt
->type
));
2420 seq_printf(seq
, " %-8s ",
2421 pt
->dev
? pt
->dev
->name
: "");
2422 ptype_seq_decode(seq
, pt
->func
);
2423 seq_putc(seq
, '\n');
2429 static const struct seq_operations ptype_seq_ops
= {
2430 .start
= ptype_seq_start
,
2431 .next
= ptype_seq_next
,
2432 .stop
= ptype_seq_stop
,
2433 .show
= ptype_seq_show
,
2436 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
2438 return seq_open(file
, &ptype_seq_ops
);
2441 static const struct file_operations ptype_seq_fops
= {
2442 .owner
= THIS_MODULE
,
2443 .open
= ptype_seq_open
,
2445 .llseek
= seq_lseek
,
2446 .release
= seq_release
,
2450 static int __init
dev_proc_init(void)
2454 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2456 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2458 if (!proc_net_fops_create("ptype", S_IRUGO
, &ptype_seq_fops
))
2461 if (wext_proc_init())
2467 proc_net_remove("ptype");
2469 proc_net_remove("softnet_stat");
2471 proc_net_remove("dev");
2475 #define dev_proc_init() 0
2476 #endif /* CONFIG_PROC_FS */
2480 * netdev_set_master - set up master/slave pair
2481 * @slave: slave device
2482 * @master: new master device
2484 * Changes the master device of the slave. Pass %NULL to break the
2485 * bonding. The caller must hold the RTNL semaphore. On a failure
2486 * a negative errno code is returned. On success the reference counts
2487 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2488 * function returns zero.
2490 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2492 struct net_device
*old
= slave
->master
;
2502 slave
->master
= master
;
2509 slave
->flags
|= IFF_SLAVE
;
2511 slave
->flags
&= ~IFF_SLAVE
;
2513 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2518 * dev_set_promiscuity - update promiscuity count on a device
2522 * Add or remove promiscuity from a device. While the count in the device
2523 * remains above zero the interface remains promiscuous. Once it hits zero
2524 * the device reverts back to normal filtering operation. A negative inc
2525 * value is used to drop promiscuity on the device.
2526 * Return 0 if successful or a negative errno code on error.
2528 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
2530 unsigned short old_flags
= dev
->flags
;
2532 dev
->flags
|= IFF_PROMISC
;
2533 dev
->promiscuity
+= inc
;
2534 if (dev
->promiscuity
== 0) {
2537 * If inc causes overflow, untouch promisc and return error.
2540 dev
->flags
&= ~IFF_PROMISC
;
2542 dev
->promiscuity
-= inc
;
2543 printk(KERN_WARNING
"%s: promiscuity touches roof, "
2544 "set promiscuity failed, promiscuity feature "
2545 "of device might be broken.\n", dev
->name
);
2549 if (dev
->flags
!= old_flags
) {
2551 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2552 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2554 audit_log(current
->audit_context
, GFP_ATOMIC
,
2555 AUDIT_ANOM_PROMISCUOUS
,
2556 "dev=%s prom=%d old_prom=%d auid=%u",
2557 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2558 (old_flags
& IFF_PROMISC
),
2559 audit_get_loginuid(current
->audit_context
));
2565 * dev_set_allmulti - update allmulti count on a device
2569 * Add or remove reception of all multicast frames to a device. While the
2570 * count in the device remains above zero the interface remains listening
2571 * to all interfaces. Once it hits zero the device reverts back to normal
2572 * filtering operation. A negative @inc value is used to drop the counter
2573 * when releasing a resource needing all multicasts.
2574 * Return 0 if successful or a negative errno code on error.
2577 int dev_set_allmulti(struct net_device
*dev
, int inc
)
2579 unsigned short old_flags
= dev
->flags
;
2581 dev
->flags
|= IFF_ALLMULTI
;
2582 dev
->allmulti
+= inc
;
2583 if (dev
->allmulti
== 0) {
2586 * If inc causes overflow, untouch allmulti and return error.
2589 dev
->flags
&= ~IFF_ALLMULTI
;
2591 dev
->allmulti
-= inc
;
2592 printk(KERN_WARNING
"%s: allmulti touches roof, "
2593 "set allmulti failed, allmulti feature of "
2594 "device might be broken.\n", dev
->name
);
2598 if (dev
->flags
^ old_flags
)
2603 unsigned dev_get_flags(const struct net_device
*dev
)
2607 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2612 (dev
->gflags
& (IFF_PROMISC
|
2615 if (netif_running(dev
)) {
2616 if (netif_oper_up(dev
))
2617 flags
|= IFF_RUNNING
;
2618 if (netif_carrier_ok(dev
))
2619 flags
|= IFF_LOWER_UP
;
2620 if (netif_dormant(dev
))
2621 flags
|= IFF_DORMANT
;
2627 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2630 int old_flags
= dev
->flags
;
2633 * Set the flags on our device.
2636 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2637 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2639 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2643 * Load in the correct multicast list now the flags have changed.
2649 * Have we downed the interface. We handle IFF_UP ourselves
2650 * according to user attempts to set it, rather than blindly
2655 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2656 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2662 if (dev
->flags
& IFF_UP
&&
2663 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2665 raw_notifier_call_chain(&netdev_chain
,
2666 NETDEV_CHANGE
, dev
);
2668 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2669 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2670 dev
->gflags
^= IFF_PROMISC
;
2671 dev_set_promiscuity(dev
, inc
);
2674 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2675 is important. Some (broken) drivers set IFF_PROMISC, when
2676 IFF_ALLMULTI is requested not asking us and not reporting.
2678 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2679 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2680 dev
->gflags
^= IFF_ALLMULTI
;
2681 dev_set_allmulti(dev
, inc
);
2684 /* Exclude state transition flags, already notified */
2685 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
2687 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
2692 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2696 if (new_mtu
== dev
->mtu
)
2699 /* MTU must be positive. */
2703 if (!netif_device_present(dev
))
2707 if (dev
->change_mtu
)
2708 err
= dev
->change_mtu(dev
, new_mtu
);
2711 if (!err
&& dev
->flags
& IFF_UP
)
2712 raw_notifier_call_chain(&netdev_chain
,
2713 NETDEV_CHANGEMTU
, dev
);
2717 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2721 if (!dev
->set_mac_address
)
2723 if (sa
->sa_family
!= dev
->type
)
2725 if (!netif_device_present(dev
))
2727 err
= dev
->set_mac_address(dev
, sa
);
2729 raw_notifier_call_chain(&netdev_chain
,
2730 NETDEV_CHANGEADDR
, dev
);
2735 * Perform the SIOCxIFxxx calls.
2737 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2740 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2746 case SIOCGIFFLAGS
: /* Get interface flags */
2747 ifr
->ifr_flags
= dev_get_flags(dev
);
2750 case SIOCSIFFLAGS
: /* Set interface flags */
2751 return dev_change_flags(dev
, ifr
->ifr_flags
);
2753 case SIOCGIFMETRIC
: /* Get the metric on the interface
2754 (currently unused) */
2755 ifr
->ifr_metric
= 0;
2758 case SIOCSIFMETRIC
: /* Set the metric on the interface
2759 (currently unused) */
2762 case SIOCGIFMTU
: /* Get the MTU of a device */
2763 ifr
->ifr_mtu
= dev
->mtu
;
2766 case SIOCSIFMTU
: /* Set the MTU of a device */
2767 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
2771 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
2773 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
2774 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2775 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
2779 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
2781 case SIOCSIFHWBROADCAST
:
2782 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
2784 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
2785 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2786 raw_notifier_call_chain(&netdev_chain
,
2787 NETDEV_CHANGEADDR
, dev
);
2791 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
2792 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
2793 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
2794 ifr
->ifr_map
.irq
= dev
->irq
;
2795 ifr
->ifr_map
.dma
= dev
->dma
;
2796 ifr
->ifr_map
.port
= dev
->if_port
;
2800 if (dev
->set_config
) {
2801 if (!netif_device_present(dev
))
2803 return dev
->set_config(dev
, &ifr
->ifr_map
);
2808 if (!dev
->set_multicast_list
||
2809 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2811 if (!netif_device_present(dev
))
2813 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
2817 if (!dev
->set_multicast_list
||
2818 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
2820 if (!netif_device_present(dev
))
2822 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
2826 ifr
->ifr_ifindex
= dev
->ifindex
;
2830 ifr
->ifr_qlen
= dev
->tx_queue_len
;
2834 if (ifr
->ifr_qlen
< 0)
2836 dev
->tx_queue_len
= ifr
->ifr_qlen
;
2840 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
2841 return dev_change_name(dev
, ifr
->ifr_newname
);
2844 * Unknown or private ioctl
2848 if ((cmd
>= SIOCDEVPRIVATE
&&
2849 cmd
<= SIOCDEVPRIVATE
+ 15) ||
2850 cmd
== SIOCBONDENSLAVE
||
2851 cmd
== SIOCBONDRELEASE
||
2852 cmd
== SIOCBONDSETHWADDR
||
2853 cmd
== SIOCBONDSLAVEINFOQUERY
||
2854 cmd
== SIOCBONDINFOQUERY
||
2855 cmd
== SIOCBONDCHANGEACTIVE
||
2856 cmd
== SIOCGMIIPHY
||
2857 cmd
== SIOCGMIIREG
||
2858 cmd
== SIOCSMIIREG
||
2859 cmd
== SIOCBRADDIF
||
2860 cmd
== SIOCBRDELIF
||
2861 cmd
== SIOCWANDEV
) {
2863 if (dev
->do_ioctl
) {
2864 if (netif_device_present(dev
))
2865 err
= dev
->do_ioctl(dev
, ifr
,
2878 * This function handles all "interface"-type I/O control requests. The actual
2879 * 'doing' part of this is dev_ifsioc above.
2883 * dev_ioctl - network device ioctl
2884 * @cmd: command to issue
2885 * @arg: pointer to a struct ifreq in user space
2887 * Issue ioctl functions to devices. This is normally called by the
2888 * user space syscall interfaces but can sometimes be useful for
2889 * other purposes. The return value is the return from the syscall if
2890 * positive or a negative errno code on error.
2893 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
2899 /* One special case: SIOCGIFCONF takes ifconf argument
2900 and requires shared lock, because it sleeps writing
2904 if (cmd
== SIOCGIFCONF
) {
2906 ret
= dev_ifconf((char __user
*) arg
);
2910 if (cmd
== SIOCGIFNAME
)
2911 return dev_ifname((struct ifreq __user
*)arg
);
2913 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2916 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
2918 colon
= strchr(ifr
.ifr_name
, ':');
2923 * See which interface the caller is talking about.
2928 * These ioctl calls:
2929 * - can be done by all.
2930 * - atomic and do not require locking.
2941 dev_load(ifr
.ifr_name
);
2942 read_lock(&dev_base_lock
);
2943 ret
= dev_ifsioc(&ifr
, cmd
);
2944 read_unlock(&dev_base_lock
);
2948 if (copy_to_user(arg
, &ifr
,
2949 sizeof(struct ifreq
)))
2955 dev_load(ifr
.ifr_name
);
2957 ret
= dev_ethtool(&ifr
);
2962 if (copy_to_user(arg
, &ifr
,
2963 sizeof(struct ifreq
)))
2969 * These ioctl calls:
2970 * - require superuser power.
2971 * - require strict serialization.
2977 if (!capable(CAP_NET_ADMIN
))
2979 dev_load(ifr
.ifr_name
);
2981 ret
= dev_ifsioc(&ifr
, cmd
);
2986 if (copy_to_user(arg
, &ifr
,
2987 sizeof(struct ifreq
)))
2993 * These ioctl calls:
2994 * - require superuser power.
2995 * - require strict serialization.
2996 * - do not return a value
3006 case SIOCSIFHWBROADCAST
:
3009 case SIOCBONDENSLAVE
:
3010 case SIOCBONDRELEASE
:
3011 case SIOCBONDSETHWADDR
:
3012 case SIOCBONDCHANGEACTIVE
:
3015 if (!capable(CAP_NET_ADMIN
))
3018 case SIOCBONDSLAVEINFOQUERY
:
3019 case SIOCBONDINFOQUERY
:
3020 dev_load(ifr
.ifr_name
);
3022 ret
= dev_ifsioc(&ifr
, cmd
);
3027 /* Get the per device memory space. We can add this but
3028 * currently do not support it */
3030 /* Set the per device memory buffer space.
3031 * Not applicable in our case */
3036 * Unknown or private ioctl.
3039 if (cmd
== SIOCWANDEV
||
3040 (cmd
>= SIOCDEVPRIVATE
&&
3041 cmd
<= SIOCDEVPRIVATE
+ 15)) {
3042 dev_load(ifr
.ifr_name
);
3044 ret
= dev_ifsioc(&ifr
, cmd
);
3046 if (!ret
&& copy_to_user(arg
, &ifr
,
3047 sizeof(struct ifreq
)))
3051 /* Take care of Wireless Extensions */
3052 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
3053 return wext_handle_ioctl(&ifr
, cmd
, arg
);
3060 * dev_new_index - allocate an ifindex
3062 * Returns a suitable unique value for a new device interface
3063 * number. The caller must hold the rtnl semaphore or the
3064 * dev_base_lock to be sure it remains unique.
3066 static int dev_new_index(void)
3072 if (!__dev_get_by_index(ifindex
))
3077 static int dev_boot_phase
= 1;
3079 /* Delayed registration/unregisteration */
3080 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
3082 static void net_set_todo(struct net_device
*dev
)
3084 list_add_tail(&dev
->todo_list
, &net_todo_list
);
3088 * register_netdevice - register a network device
3089 * @dev: device to register
3091 * Take a completed network device structure and add it to the kernel
3092 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3093 * chain. 0 is returned on success. A negative errno code is returned
3094 * on a failure to set up the device, or if the name is a duplicate.
3096 * Callers must hold the rtnl semaphore. You may want
3097 * register_netdev() instead of this.
3100 * The locking appears insufficient to guarantee two parallel registers
3101 * will not get the same name.
3104 int register_netdevice(struct net_device
*dev
)
3106 struct hlist_head
*head
;
3107 struct hlist_node
*p
;
3110 BUG_ON(dev_boot_phase
);
3115 /* When net_device's are persistent, this will be fatal. */
3116 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
3118 spin_lock_init(&dev
->queue_lock
);
3119 spin_lock_init(&dev
->_xmit_lock
);
3120 netdev_set_lockdep_class(&dev
->_xmit_lock
, dev
->type
);
3121 dev
->xmit_lock_owner
= -1;
3122 spin_lock_init(&dev
->ingress_lock
);
3126 /* Init, if this function is available */
3128 ret
= dev
->init(dev
);
3136 if (!dev_valid_name(dev
->name
)) {
3141 dev
->ifindex
= dev_new_index();
3142 if (dev
->iflink
== -1)
3143 dev
->iflink
= dev
->ifindex
;
3145 /* Check for existence of name */
3146 head
= dev_name_hash(dev
->name
);
3147 hlist_for_each(p
, head
) {
3148 struct net_device
*d
3149 = hlist_entry(p
, struct net_device
, name_hlist
);
3150 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
3156 /* Fix illegal SG+CSUM combinations. */
3157 if ((dev
->features
& NETIF_F_SG
) &&
3158 !(dev
->features
& NETIF_F_ALL_CSUM
)) {
3159 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no checksum feature.\n",
3161 dev
->features
&= ~NETIF_F_SG
;
3164 /* TSO requires that SG is present as well. */
3165 if ((dev
->features
& NETIF_F_TSO
) &&
3166 !(dev
->features
& NETIF_F_SG
)) {
3167 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no SG feature.\n",
3169 dev
->features
&= ~NETIF_F_TSO
;
3171 if (dev
->features
& NETIF_F_UFO
) {
3172 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
3173 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3174 "NETIF_F_HW_CSUM feature.\n",
3176 dev
->features
&= ~NETIF_F_UFO
;
3178 if (!(dev
->features
& NETIF_F_SG
)) {
3179 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3180 "NETIF_F_SG feature.\n",
3182 dev
->features
&= ~NETIF_F_UFO
;
3187 * nil rebuild_header routine,
3188 * that should be never called and used as just bug trap.
3191 if (!dev
->rebuild_header
)
3192 dev
->rebuild_header
= default_rebuild_header
;
3194 ret
= netdev_register_sysfs(dev
);
3197 dev
->reg_state
= NETREG_REGISTERED
;
3200 * Default initial state at registry is that the
3201 * device is present.
3204 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
3206 dev_init_scheduler(dev
);
3207 write_lock_bh(&dev_base_lock
);
3208 list_add_tail(&dev
->dev_list
, &dev_base_head
);
3209 hlist_add_head(&dev
->name_hlist
, head
);
3210 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
3212 write_unlock_bh(&dev_base_lock
);
3214 /* Notify protocols, that a new device appeared. */
3215 raw_notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
3229 * register_netdev - register a network device
3230 * @dev: device to register
3232 * Take a completed network device structure and add it to the kernel
3233 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3234 * chain. 0 is returned on success. A negative errno code is returned
3235 * on a failure to set up the device, or if the name is a duplicate.
3237 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3238 * and expands the device name if you passed a format string to
3241 int register_netdev(struct net_device
*dev
)
3248 * If the name is a format string the caller wants us to do a
3251 if (strchr(dev
->name
, '%')) {
3252 err
= dev_alloc_name(dev
, dev
->name
);
3257 err
= register_netdevice(dev
);
3262 EXPORT_SYMBOL(register_netdev
);
3265 * netdev_wait_allrefs - wait until all references are gone.
3267 * This is called when unregistering network devices.
3269 * Any protocol or device that holds a reference should register
3270 * for netdevice notification, and cleanup and put back the
3271 * reference if they receive an UNREGISTER event.
3272 * We can get stuck here if buggy protocols don't correctly
3275 static void netdev_wait_allrefs(struct net_device
*dev
)
3277 unsigned long rebroadcast_time
, warning_time
;
3279 rebroadcast_time
= warning_time
= jiffies
;
3280 while (atomic_read(&dev
->refcnt
) != 0) {
3281 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
3284 /* Rebroadcast unregister notification */
3285 raw_notifier_call_chain(&netdev_chain
,
3286 NETDEV_UNREGISTER
, dev
);
3288 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
3290 /* We must not have linkwatch events
3291 * pending on unregister. If this
3292 * happens, we simply run the queue
3293 * unscheduled, resulting in a noop
3296 linkwatch_run_queue();
3301 rebroadcast_time
= jiffies
;
3306 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
3307 printk(KERN_EMERG
"unregister_netdevice: "
3308 "waiting for %s to become free. Usage "
3310 dev
->name
, atomic_read(&dev
->refcnt
));
3311 warning_time
= jiffies
;
3320 * register_netdevice(x1);
3321 * register_netdevice(x2);
3323 * unregister_netdevice(y1);
3324 * unregister_netdevice(y2);
3330 * We are invoked by rtnl_unlock().
3331 * This allows us to deal with problems:
3332 * 1) We can delete sysfs objects which invoke hotplug
3333 * without deadlocking with linkwatch via keventd.
3334 * 2) Since we run with the RTNL semaphore not held, we can sleep
3335 * safely in order to wait for the netdev refcnt to drop to zero.
3337 * We must not return until all unregister events added during
3338 * the interval the lock was held have been completed.
3340 void netdev_run_todo(void)
3342 struct list_head list
;
3344 /* Snapshot list, allow later requests */
3345 list_replace_init(&net_todo_list
, &list
);
3349 while (!list_empty(&list
)) {
3350 struct net_device
*dev
3351 = list_entry(list
.next
, struct net_device
, todo_list
);
3352 list_del(&dev
->todo_list
);
3354 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
3355 printk(KERN_ERR
"network todo '%s' but state %d\n",
3356 dev
->name
, dev
->reg_state
);
3361 dev
->reg_state
= NETREG_UNREGISTERED
;
3363 netdev_wait_allrefs(dev
);
3366 BUG_ON(atomic_read(&dev
->refcnt
));
3367 BUG_TRAP(!dev
->ip_ptr
);
3368 BUG_TRAP(!dev
->ip6_ptr
);
3369 BUG_TRAP(!dev
->dn_ptr
);
3371 if (dev
->destructor
)
3372 dev
->destructor(dev
);
3374 /* Free network device */
3375 kobject_put(&dev
->dev
.kobj
);
3379 static struct net_device_stats
*internal_stats(struct net_device
*dev
)
3385 * alloc_netdev - allocate network device
3386 * @sizeof_priv: size of private data to allocate space for
3387 * @name: device name format string
3388 * @setup: callback to initialize device
3390 * Allocates a struct net_device with private data area for driver use
3391 * and performs basic initialization.
3393 struct net_device
*alloc_netdev(int sizeof_priv
, const char *name
,
3394 void (*setup
)(struct net_device
*))
3396 struct net_device
*dev
;
3398 struct net_device
*p
;
3400 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
3402 alloc_size
= sizeof(struct net_device
);
3404 /* ensure 32-byte alignment of private area */
3405 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
3406 alloc_size
+= sizeof_priv
;
3408 /* ensure 32-byte alignment of whole construct */
3409 alloc_size
+= NETDEV_ALIGN
- 1;
3411 p
= kzalloc(alloc_size
, GFP_KERNEL
);
3413 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
3417 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
3418 dev
->padded
= (char *)dev
- (char *)p
;
3421 dev
->priv
= netdev_priv(dev
);
3423 dev
->get_stats
= internal_stats
;
3425 strcpy(dev
->name
, name
);
3428 EXPORT_SYMBOL(alloc_netdev
);
3431 * free_netdev - free network device
3434 * This function does the last stage of destroying an allocated device
3435 * interface. The reference to the device object is released.
3436 * If this is the last reference then it will be freed.
3438 void free_netdev(struct net_device
*dev
)
3441 /* Compatibility with error handling in drivers */
3442 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3443 kfree((char *)dev
- dev
->padded
);
3447 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3448 dev
->reg_state
= NETREG_RELEASED
;
3450 /* will free via device release */
3451 put_device(&dev
->dev
);
3453 kfree((char *)dev
- dev
->padded
);
3457 /* Synchronize with packet receive processing. */
3458 void synchronize_net(void)
3465 * unregister_netdevice - remove device from the kernel
3468 * This function shuts down a device interface and removes it
3469 * from the kernel tables. On success 0 is returned, on a failure
3470 * a negative errno code is returned.
3472 * Callers must hold the rtnl semaphore. You may want
3473 * unregister_netdev() instead of this.
3476 void unregister_netdevice(struct net_device
*dev
)
3478 BUG_ON(dev_boot_phase
);
3481 /* Some devices call without registering for initialization unwind. */
3482 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3483 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3484 "was registered\n", dev
->name
, dev
);
3490 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3492 /* If device is running, close it first. */
3493 if (dev
->flags
& IFF_UP
)
3496 /* And unlink it from device chain. */
3497 write_lock_bh(&dev_base_lock
);
3498 list_del(&dev
->dev_list
);
3499 hlist_del(&dev
->name_hlist
);
3500 hlist_del(&dev
->index_hlist
);
3501 write_unlock_bh(&dev_base_lock
);
3503 dev
->reg_state
= NETREG_UNREGISTERING
;
3507 /* Shutdown queueing discipline. */
3511 /* Notify protocols, that we are about to destroy
3512 this device. They should clean all the things.
3514 raw_notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3517 * Flush the multicast chain
3519 dev_mc_discard(dev
);
3524 /* Notifier chain MUST detach us from master device. */
3525 BUG_TRAP(!dev
->master
);
3527 /* Remove entries from sysfs */
3528 netdev_unregister_sysfs(dev
);
3530 /* Finish processing unregister after unlock */
3539 * unregister_netdev - remove device from the kernel
3542 * This function shuts down a device interface and removes it
3543 * from the kernel tables. On success 0 is returned, on a failure
3544 * a negative errno code is returned.
3546 * This is just a wrapper for unregister_netdevice that takes
3547 * the rtnl semaphore. In general you want to use this and not
3548 * unregister_netdevice.
3550 void unregister_netdev(struct net_device
*dev
)
3553 unregister_netdevice(dev
);
3557 EXPORT_SYMBOL(unregister_netdev
);
3559 static int dev_cpu_callback(struct notifier_block
*nfb
,
3560 unsigned long action
,
3563 struct sk_buff
**list_skb
;
3564 struct net_device
**list_net
;
3565 struct sk_buff
*skb
;
3566 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3567 struct softnet_data
*sd
, *oldsd
;
3569 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
3572 local_irq_disable();
3573 cpu
= smp_processor_id();
3574 sd
= &per_cpu(softnet_data
, cpu
);
3575 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3577 /* Find end of our completion_queue. */
3578 list_skb
= &sd
->completion_queue
;
3580 list_skb
= &(*list_skb
)->next
;
3581 /* Append completion queue from offline CPU. */
3582 *list_skb
= oldsd
->completion_queue
;
3583 oldsd
->completion_queue
= NULL
;
3585 /* Find end of our output_queue. */
3586 list_net
= &sd
->output_queue
;
3588 list_net
= &(*list_net
)->next_sched
;
3589 /* Append output queue from offline CPU. */
3590 *list_net
= oldsd
->output_queue
;
3591 oldsd
->output_queue
= NULL
;
3593 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3596 /* Process offline CPU's input_pkt_queue */
3597 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3603 #ifdef CONFIG_NET_DMA
3605 * net_dma_rebalance -
3606 * This is called when the number of channels allocated to the net_dma_client
3607 * changes. The net_dma_client tries to have one DMA channel per CPU.
3609 static void net_dma_rebalance(void)
3611 unsigned int cpu
, i
, n
;
3612 struct dma_chan
*chan
;
3614 if (net_dma_count
== 0) {
3615 for_each_online_cpu(cpu
)
3616 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
3621 cpu
= first_cpu(cpu_online_map
);
3624 list_for_each_entry(chan
, &net_dma_client
->channels
, client_node
) {
3625 n
= ((num_online_cpus() / net_dma_count
)
3626 + (i
< (num_online_cpus() % net_dma_count
) ? 1 : 0));
3629 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
3630 cpu
= next_cpu(cpu
, cpu_online_map
);
3639 * netdev_dma_event - event callback for the net_dma_client
3640 * @client: should always be net_dma_client
3641 * @chan: DMA channel for the event
3642 * @event: event type
3644 static void netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
3645 enum dma_event event
)
3647 spin_lock(&net_dma_event_lock
);
3649 case DMA_RESOURCE_ADDED
:
3651 net_dma_rebalance();
3653 case DMA_RESOURCE_REMOVED
:
3655 net_dma_rebalance();
3660 spin_unlock(&net_dma_event_lock
);
3664 * netdev_dma_regiser - register the networking subsystem as a DMA client
3666 static int __init
netdev_dma_register(void)
3668 spin_lock_init(&net_dma_event_lock
);
3669 net_dma_client
= dma_async_client_register(netdev_dma_event
);
3670 if (net_dma_client
== NULL
)
3673 dma_async_client_chan_request(net_dma_client
, num_online_cpus());
3678 static int __init
netdev_dma_register(void) { return -ENODEV
; }
3679 #endif /* CONFIG_NET_DMA */
3682 * netdev_compute_feature - compute conjunction of two feature sets
3683 * @all: first feature set
3684 * @one: second feature set
3686 * Computes a new feature set after adding a device with feature set
3687 * @one to the master device with current feature set @all. Returns
3688 * the new feature set.
3690 int netdev_compute_features(unsigned long all
, unsigned long one
)
3692 /* if device needs checksumming, downgrade to hw checksumming */
3693 if (all
& NETIF_F_NO_CSUM
&& !(one
& NETIF_F_NO_CSUM
))
3694 all
^= NETIF_F_NO_CSUM
| NETIF_F_HW_CSUM
;
3696 /* if device can't do all checksum, downgrade to ipv4 */
3697 if (all
& NETIF_F_HW_CSUM
&& !(one
& NETIF_F_HW_CSUM
))
3698 all
^= NETIF_F_HW_CSUM
| NETIF_F_IP_CSUM
;
3700 if (one
& NETIF_F_GSO
)
3701 one
|= NETIF_F_GSO_SOFTWARE
;
3704 /* If even one device supports robust GSO, enable it for all. */
3705 if (one
& NETIF_F_GSO_ROBUST
)
3706 all
|= NETIF_F_GSO_ROBUST
;
3708 all
&= one
| NETIF_F_LLTX
;
3710 if (!(all
& NETIF_F_ALL_CSUM
))
3712 if (!(all
& NETIF_F_SG
))
3713 all
&= ~NETIF_F_GSO_MASK
;
3717 EXPORT_SYMBOL(netdev_compute_features
);
3720 * Initialize the DEV module. At boot time this walks the device list and
3721 * unhooks any devices that fail to initialise (normally hardware not
3722 * present) and leaves us with a valid list of present and active devices.
3727 * This is called single threaded during boot, so no need
3728 * to take the rtnl semaphore.
3730 static int __init
net_dev_init(void)
3732 int i
, rc
= -ENOMEM
;
3734 BUG_ON(!dev_boot_phase
);
3736 if (dev_proc_init())
3739 if (netdev_sysfs_init())
3742 INIT_LIST_HEAD(&ptype_all
);
3743 for (i
= 0; i
< 16; i
++)
3744 INIT_LIST_HEAD(&ptype_base
[i
]);
3746 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
3747 INIT_HLIST_HEAD(&dev_name_head
[i
]);
3749 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
3750 INIT_HLIST_HEAD(&dev_index_head
[i
]);
3753 * Initialise the packet receive queues.
3756 for_each_possible_cpu(i
) {
3757 struct softnet_data
*queue
;
3759 queue
= &per_cpu(softnet_data
, i
);
3760 skb_queue_head_init(&queue
->input_pkt_queue
);
3761 queue
->completion_queue
= NULL
;
3762 INIT_LIST_HEAD(&queue
->poll_list
);
3763 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
3764 queue
->backlog_dev
.weight
= weight_p
;
3765 queue
->backlog_dev
.poll
= process_backlog
;
3766 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
3769 netdev_dma_register();
3773 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
3774 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
3776 hotcpu_notifier(dev_cpu_callback
, 0);
3784 subsys_initcall(net_dev_init
);
3786 EXPORT_SYMBOL(__dev_get_by_index
);
3787 EXPORT_SYMBOL(__dev_get_by_name
);
3788 EXPORT_SYMBOL(__dev_remove_pack
);
3789 EXPORT_SYMBOL(dev_valid_name
);
3790 EXPORT_SYMBOL(dev_add_pack
);
3791 EXPORT_SYMBOL(dev_alloc_name
);
3792 EXPORT_SYMBOL(dev_close
);
3793 EXPORT_SYMBOL(dev_get_by_flags
);
3794 EXPORT_SYMBOL(dev_get_by_index
);
3795 EXPORT_SYMBOL(dev_get_by_name
);
3796 EXPORT_SYMBOL(dev_open
);
3797 EXPORT_SYMBOL(dev_queue_xmit
);
3798 EXPORT_SYMBOL(dev_remove_pack
);
3799 EXPORT_SYMBOL(dev_set_allmulti
);
3800 EXPORT_SYMBOL(dev_set_promiscuity
);
3801 EXPORT_SYMBOL(dev_change_flags
);
3802 EXPORT_SYMBOL(dev_set_mtu
);
3803 EXPORT_SYMBOL(dev_set_mac_address
);
3804 EXPORT_SYMBOL(free_netdev
);
3805 EXPORT_SYMBOL(netdev_boot_setup_check
);
3806 EXPORT_SYMBOL(netdev_set_master
);
3807 EXPORT_SYMBOL(netdev_state_change
);
3808 EXPORT_SYMBOL(netif_receive_skb
);
3809 EXPORT_SYMBOL(netif_rx
);
3810 EXPORT_SYMBOL(register_gifconf
);
3811 EXPORT_SYMBOL(register_netdevice
);
3812 EXPORT_SYMBOL(register_netdevice_notifier
);
3813 EXPORT_SYMBOL(skb_checksum_help
);
3814 EXPORT_SYMBOL(synchronize_net
);
3815 EXPORT_SYMBOL(unregister_netdevice
);
3816 EXPORT_SYMBOL(unregister_netdevice_notifier
);
3817 EXPORT_SYMBOL(net_enable_timestamp
);
3818 EXPORT_SYMBOL(net_disable_timestamp
);
3819 EXPORT_SYMBOL(dev_get_flags
);
3821 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3822 EXPORT_SYMBOL(br_handle_frame_hook
);
3823 EXPORT_SYMBOL(br_fdb_get_hook
);
3824 EXPORT_SYMBOL(br_fdb_put_hook
);
3828 EXPORT_SYMBOL(dev_load
);
3831 EXPORT_PER_CPU_SYMBOL(softnet_data
);