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>
101 #include <linux/if_macvlan.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/kmod.h>
108 #include <linux/module.h>
109 #include <linux/kallsyms.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #include <net/wext.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.h>
119 #include <linux/ctype.h>
120 #include <linux/if_arp.h>
123 * The list of packet types we will receive (as opposed to discard)
124 * and the routines to invoke.
126 * Why 16. Because with 16 the only overlap we get on a hash of the
127 * low nibble of the protocol value is RARP/SNAP/X.25.
129 * NOTE: That is no longer true with the addition of VLAN tags. Not
130 * sure which should go first, but I bet it won't make much
131 * difference if we are running VLANs. The good news is that
132 * this protocol won't be in the list unless compiled in, so
133 * the average user (w/out VLANs) will not be adversely affected.
150 static DEFINE_SPINLOCK(ptype_lock
);
151 static struct list_head ptype_base
[16] __read_mostly
; /* 16 way hashed list */
152 static struct list_head ptype_all __read_mostly
; /* Taps */
154 #ifdef CONFIG_NET_DMA
156 struct dma_client client
;
158 cpumask_t channel_mask
;
159 struct dma_chan
*channels
[NR_CPUS
];
162 static enum dma_state_client
163 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
164 enum dma_state state
);
166 static struct net_dma net_dma
= {
168 .event_callback
= netdev_dma_event
,
174 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
177 * Pure readers hold dev_base_lock for reading.
179 * Writers must hold the rtnl semaphore while they loop through the
180 * dev_base_head list, and hold dev_base_lock for writing when they do the
181 * actual updates. This allows pure readers to access the list even
182 * while a writer is preparing to update it.
184 * To put it another way, dev_base_lock is held for writing only to
185 * protect against pure readers; the rtnl semaphore provides the
186 * protection against other writers.
188 * See, for example usages, register_netdevice() and
189 * unregister_netdevice(), which must be called with the rtnl
192 LIST_HEAD(dev_base_head
);
193 DEFINE_RWLOCK(dev_base_lock
);
195 EXPORT_SYMBOL(dev_base_head
);
196 EXPORT_SYMBOL(dev_base_lock
);
198 #define NETDEV_HASHBITS 8
199 static struct hlist_head dev_name_head
[1<<NETDEV_HASHBITS
];
200 static struct hlist_head dev_index_head
[1<<NETDEV_HASHBITS
];
202 static inline struct hlist_head
*dev_name_hash(const char *name
)
204 unsigned hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
205 return &dev_name_head
[hash
& ((1<<NETDEV_HASHBITS
)-1)];
208 static inline struct hlist_head
*dev_index_hash(int ifindex
)
210 return &dev_index_head
[ifindex
& ((1<<NETDEV_HASHBITS
)-1)];
217 static RAW_NOTIFIER_HEAD(netdev_chain
);
220 * Device drivers call our routines to queue packets here. We empty the
221 * queue in the local softnet handler.
223 DEFINE_PER_CPU(struct softnet_data
, softnet_data
) = { NULL
};
226 extern int netdev_sysfs_init(void);
227 extern int netdev_register_sysfs(struct net_device
*);
228 extern void netdev_unregister_sysfs(struct net_device
*);
230 #define netdev_sysfs_init() (0)
231 #define netdev_register_sysfs(dev) (0)
232 #define netdev_unregister_sysfs(dev) do { } while(0)
235 #ifdef CONFIG_DEBUG_LOCK_ALLOC
237 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
238 * according to dev->type
240 static const unsigned short netdev_lock_type
[] =
241 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
242 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
243 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
244 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
245 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
246 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
247 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
248 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
249 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
250 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
251 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
252 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
253 ARPHRD_FCFABRIC
, ARPHRD_IEEE802_TR
, ARPHRD_IEEE80211
,
254 ARPHRD_IEEE80211_PRISM
, ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_VOID
,
257 static const char *netdev_lock_name
[] =
258 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
259 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
260 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
261 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
262 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
263 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
264 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
265 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
266 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
267 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
268 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
269 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
270 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
271 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
274 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
276 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
280 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
281 if (netdev_lock_type
[i
] == dev_type
)
283 /* the last key is used by default */
284 return ARRAY_SIZE(netdev_lock_type
) - 1;
287 static inline void netdev_set_lockdep_class(spinlock_t
*lock
,
288 unsigned short dev_type
)
292 i
= netdev_lock_pos(dev_type
);
293 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
294 netdev_lock_name
[i
]);
297 static inline void netdev_set_lockdep_class(spinlock_t
*lock
,
298 unsigned short dev_type
)
303 /*******************************************************************************
305 Protocol management and registration routines
307 *******************************************************************************/
310 * Add a protocol ID to the list. Now that the input handler is
311 * smarter we can dispense with all the messy stuff that used to be
314 * BEWARE!!! Protocol handlers, mangling input packets,
315 * MUST BE last in hash buckets and checking protocol handlers
316 * MUST start from promiscuous ptype_all chain in net_bh.
317 * It is true now, do not change it.
318 * Explanation follows: if protocol handler, mangling packet, will
319 * be the first on list, it is not able to sense, that packet
320 * is cloned and should be copied-on-write, so that it will
321 * change it and subsequent readers will get broken packet.
326 * dev_add_pack - add packet handler
327 * @pt: packet type declaration
329 * Add a protocol handler to the networking stack. The passed &packet_type
330 * is linked into kernel lists and may not be freed until it has been
331 * removed from the kernel lists.
333 * This call does not sleep therefore it can not
334 * guarantee all CPU's that are in middle of receiving packets
335 * will see the new packet type (until the next received packet).
338 void dev_add_pack(struct packet_type
*pt
)
342 spin_lock_bh(&ptype_lock
);
343 if (pt
->type
== htons(ETH_P_ALL
))
344 list_add_rcu(&pt
->list
, &ptype_all
);
346 hash
= ntohs(pt
->type
) & 15;
347 list_add_rcu(&pt
->list
, &ptype_base
[hash
]);
349 spin_unlock_bh(&ptype_lock
);
353 * __dev_remove_pack - remove packet handler
354 * @pt: packet type declaration
356 * Remove a protocol handler that was previously added to the kernel
357 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
358 * from the kernel lists and can be freed or reused once this function
361 * The packet type might still be in use by receivers
362 * and must not be freed until after all the CPU's have gone
363 * through a quiescent state.
365 void __dev_remove_pack(struct packet_type
*pt
)
367 struct list_head
*head
;
368 struct packet_type
*pt1
;
370 spin_lock_bh(&ptype_lock
);
372 if (pt
->type
== htons(ETH_P_ALL
))
375 head
= &ptype_base
[ntohs(pt
->type
) & 15];
377 list_for_each_entry(pt1
, head
, list
) {
379 list_del_rcu(&pt
->list
);
384 printk(KERN_WARNING
"dev_remove_pack: %p not found.\n", pt
);
386 spin_unlock_bh(&ptype_lock
);
389 * dev_remove_pack - remove packet handler
390 * @pt: packet type declaration
392 * Remove a protocol handler that was previously added to the kernel
393 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
394 * from the kernel lists and can be freed or reused once this function
397 * This call sleeps to guarantee that no CPU is looking at the packet
400 void dev_remove_pack(struct packet_type
*pt
)
402 __dev_remove_pack(pt
);
407 /******************************************************************************
409 Device Boot-time Settings Routines
411 *******************************************************************************/
413 /* Boot time configuration table */
414 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
417 * netdev_boot_setup_add - add new setup entry
418 * @name: name of the device
419 * @map: configured settings for the device
421 * Adds new setup entry to the dev_boot_setup list. The function
422 * returns 0 on error and 1 on success. This is a generic routine to
425 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
427 struct netdev_boot_setup
*s
;
431 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
432 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
433 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
434 strcpy(s
[i
].name
, name
);
435 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
440 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
444 * netdev_boot_setup_check - check boot time settings
445 * @dev: the netdevice
447 * Check boot time settings for the device.
448 * The found settings are set for the device to be used
449 * later in the device probing.
450 * Returns 0 if no settings found, 1 if they are.
452 int netdev_boot_setup_check(struct net_device
*dev
)
454 struct netdev_boot_setup
*s
= dev_boot_setup
;
457 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
458 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
459 !strncmp(dev
->name
, s
[i
].name
, strlen(s
[i
].name
))) {
460 dev
->irq
= s
[i
].map
.irq
;
461 dev
->base_addr
= s
[i
].map
.base_addr
;
462 dev
->mem_start
= s
[i
].map
.mem_start
;
463 dev
->mem_end
= s
[i
].map
.mem_end
;
472 * netdev_boot_base - get address from boot time settings
473 * @prefix: prefix for network device
474 * @unit: id for network device
476 * Check boot time settings for the base address of device.
477 * The found settings are set for the device to be used
478 * later in the device probing.
479 * Returns 0 if no settings found.
481 unsigned long netdev_boot_base(const char *prefix
, int unit
)
483 const struct netdev_boot_setup
*s
= dev_boot_setup
;
487 sprintf(name
, "%s%d", prefix
, unit
);
490 * If device already registered then return base of 1
491 * to indicate not to probe for this interface
493 if (__dev_get_by_name(name
))
496 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
497 if (!strcmp(name
, s
[i
].name
))
498 return s
[i
].map
.base_addr
;
503 * Saves at boot time configured settings for any netdevice.
505 int __init
netdev_boot_setup(char *str
)
510 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
515 memset(&map
, 0, sizeof(map
));
519 map
.base_addr
= ints
[2];
521 map
.mem_start
= ints
[3];
523 map
.mem_end
= ints
[4];
525 /* Add new entry to the list */
526 return netdev_boot_setup_add(str
, &map
);
529 __setup("netdev=", netdev_boot_setup
);
531 /*******************************************************************************
533 Device Interface Subroutines
535 *******************************************************************************/
538 * __dev_get_by_name - find a device by its name
539 * @name: name to find
541 * Find an interface by name. Must be called under RTNL semaphore
542 * or @dev_base_lock. If the name is found a pointer to the device
543 * is returned. If the name is not found then %NULL is returned. The
544 * reference counters are not incremented so the caller must be
545 * careful with locks.
548 struct net_device
*__dev_get_by_name(const char *name
)
550 struct hlist_node
*p
;
552 hlist_for_each(p
, dev_name_hash(name
)) {
553 struct net_device
*dev
554 = hlist_entry(p
, struct net_device
, name_hlist
);
555 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
562 * dev_get_by_name - find a device by its name
563 * @name: name to find
565 * Find an interface by name. This can be called from any
566 * context and does its own locking. The returned handle has
567 * the usage count incremented and the caller must use dev_put() to
568 * release it when it is no longer needed. %NULL is returned if no
569 * matching device is found.
572 struct net_device
*dev_get_by_name(const char *name
)
574 struct net_device
*dev
;
576 read_lock(&dev_base_lock
);
577 dev
= __dev_get_by_name(name
);
580 read_unlock(&dev_base_lock
);
585 * __dev_get_by_index - find a device by its ifindex
586 * @ifindex: index of device
588 * Search for an interface by index. Returns %NULL if the device
589 * is not found or a pointer to the device. The device has not
590 * had its reference counter increased so the caller must be careful
591 * about locking. The caller must hold either the RTNL semaphore
595 struct net_device
*__dev_get_by_index(int ifindex
)
597 struct hlist_node
*p
;
599 hlist_for_each(p
, dev_index_hash(ifindex
)) {
600 struct net_device
*dev
601 = hlist_entry(p
, struct net_device
, index_hlist
);
602 if (dev
->ifindex
== ifindex
)
610 * dev_get_by_index - find a device by its ifindex
611 * @ifindex: index of device
613 * Search for an interface by index. Returns NULL if the device
614 * is not found or a pointer to the device. The device returned has
615 * had a reference added and the pointer is safe until the user calls
616 * dev_put to indicate they have finished with it.
619 struct net_device
*dev_get_by_index(int ifindex
)
621 struct net_device
*dev
;
623 read_lock(&dev_base_lock
);
624 dev
= __dev_get_by_index(ifindex
);
627 read_unlock(&dev_base_lock
);
632 * dev_getbyhwaddr - find a device by its hardware address
633 * @type: media type of device
634 * @ha: hardware address
636 * Search for an interface by MAC address. Returns NULL if the device
637 * is not found or a pointer to the device. The caller must hold the
638 * rtnl semaphore. The returned device has not had its ref count increased
639 * and the caller must therefore be careful about locking
642 * If the API was consistent this would be __dev_get_by_hwaddr
645 struct net_device
*dev_getbyhwaddr(unsigned short type
, char *ha
)
647 struct net_device
*dev
;
652 if (dev
->type
== type
&&
653 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
659 EXPORT_SYMBOL(dev_getbyhwaddr
);
661 struct net_device
*__dev_getfirstbyhwtype(unsigned short type
)
663 struct net_device
*dev
;
667 if (dev
->type
== type
)
673 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
675 struct net_device
*dev_getfirstbyhwtype(unsigned short type
)
677 struct net_device
*dev
;
680 dev
= __dev_getfirstbyhwtype(type
);
687 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
690 * dev_get_by_flags - find any device with given flags
691 * @if_flags: IFF_* values
692 * @mask: bitmask of bits in if_flags to check
694 * Search for any interface with the given flags. Returns NULL if a device
695 * is not found or a pointer to the device. The device returned has
696 * had a reference added and the pointer is safe until the user calls
697 * dev_put to indicate they have finished with it.
700 struct net_device
* dev_get_by_flags(unsigned short if_flags
, unsigned short mask
)
702 struct net_device
*dev
, *ret
;
705 read_lock(&dev_base_lock
);
706 for_each_netdev(dev
) {
707 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
713 read_unlock(&dev_base_lock
);
718 * dev_valid_name - check if name is okay for network device
721 * Network device names need to be valid file names to
722 * to allow sysfs to work. We also disallow any kind of
725 int dev_valid_name(const char *name
)
729 if (strlen(name
) >= IFNAMSIZ
)
731 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
735 if (*name
== '/' || isspace(*name
))
743 * dev_alloc_name - allocate a name for a device
745 * @name: name format string
747 * Passed a format string - eg "lt%d" it will try and find a suitable
748 * id. It scans list of devices to build up a free map, then chooses
749 * the first empty slot. The caller must hold the dev_base or rtnl lock
750 * while allocating the name and adding the device in order to avoid
752 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
753 * Returns the number of the unit assigned or a negative errno code.
756 int dev_alloc_name(struct net_device
*dev
, const char *name
)
761 const int max_netdevices
= 8*PAGE_SIZE
;
763 struct net_device
*d
;
765 p
= strnchr(name
, IFNAMSIZ
-1, '%');
768 * Verify the string as this thing may have come from
769 * the user. There must be either one "%d" and no other "%"
772 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
775 /* Use one page as a bit array of possible slots */
776 inuse
= (long *) get_zeroed_page(GFP_ATOMIC
);
781 if (!sscanf(d
->name
, name
, &i
))
783 if (i
< 0 || i
>= max_netdevices
)
786 /* avoid cases where sscanf is not exact inverse of printf */
787 snprintf(buf
, sizeof(buf
), name
, i
);
788 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
792 i
= find_first_zero_bit(inuse
, max_netdevices
);
793 free_page((unsigned long) inuse
);
796 snprintf(buf
, sizeof(buf
), name
, i
);
797 if (!__dev_get_by_name(buf
)) {
798 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
802 /* It is possible to run out of possible slots
803 * when the name is long and there isn't enough space left
804 * for the digits, or if all bits are used.
811 * dev_change_name - change name of a device
813 * @newname: name (or format string) must be at least IFNAMSIZ
815 * Change name of a device, can pass format strings "eth%d".
818 int dev_change_name(struct net_device
*dev
, char *newname
)
824 if (dev
->flags
& IFF_UP
)
827 if (!dev_valid_name(newname
))
830 if (strchr(newname
, '%')) {
831 err
= dev_alloc_name(dev
, newname
);
834 strcpy(newname
, dev
->name
);
836 else if (__dev_get_by_name(newname
))
839 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
841 device_rename(&dev
->dev
, dev
->name
);
842 hlist_del(&dev
->name_hlist
);
843 hlist_add_head(&dev
->name_hlist
, dev_name_hash(dev
->name
));
844 raw_notifier_call_chain(&netdev_chain
, NETDEV_CHANGENAME
, dev
);
850 * netdev_features_change - device changes features
851 * @dev: device to cause notification
853 * Called to indicate a device has changed features.
855 void netdev_features_change(struct net_device
*dev
)
857 raw_notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
859 EXPORT_SYMBOL(netdev_features_change
);
862 * netdev_state_change - device changes state
863 * @dev: device to cause notification
865 * Called to indicate a device has changed state. This function calls
866 * the notifier chains for netdev_chain and sends a NEWLINK message
867 * to the routing socket.
869 void netdev_state_change(struct net_device
*dev
)
871 if (dev
->flags
& IFF_UP
) {
872 raw_notifier_call_chain(&netdev_chain
,
874 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
879 * dev_load - load a network module
880 * @name: name of interface
882 * If a network interface is not present and the process has suitable
883 * privileges this function loads the module. If module loading is not
884 * available in this kernel then it becomes a nop.
887 void dev_load(const char *name
)
889 struct net_device
*dev
;
891 read_lock(&dev_base_lock
);
892 dev
= __dev_get_by_name(name
);
893 read_unlock(&dev_base_lock
);
895 if (!dev
&& capable(CAP_SYS_MODULE
))
896 request_module("%s", name
);
899 static int default_rebuild_header(struct sk_buff
*skb
)
901 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
902 skb
->dev
? skb
->dev
->name
: "NULL!!!");
908 * dev_open - prepare an interface for use.
909 * @dev: device to open
911 * Takes a device from down to up state. The device's private open
912 * function is invoked and then the multicast lists are loaded. Finally
913 * the device is moved into the up state and a %NETDEV_UP message is
914 * sent to the netdev notifier chain.
916 * Calling this function on an active interface is a nop. On a failure
917 * a negative errno code is returned.
919 int dev_open(struct net_device
*dev
)
927 if (dev
->flags
& IFF_UP
)
931 * Is it even present?
933 if (!netif_device_present(dev
))
937 * Call device private open method
939 set_bit(__LINK_STATE_START
, &dev
->state
);
941 ret
= dev
->open(dev
);
943 clear_bit(__LINK_STATE_START
, &dev
->state
);
947 * If it went open OK then:
954 dev
->flags
|= IFF_UP
;
957 * Initialize multicasting status
959 dev_set_rx_mode(dev
);
962 * Wakeup transmit queue engine
967 * ... and announce new interface.
969 raw_notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
975 * dev_close - shutdown an interface.
976 * @dev: device to shutdown
978 * This function moves an active device into down state. A
979 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
980 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
983 int dev_close(struct net_device
*dev
)
985 if (!(dev
->flags
& IFF_UP
))
989 * Tell people we are going down, so that they can
990 * prepare to death, when device is still operating.
992 raw_notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
996 clear_bit(__LINK_STATE_START
, &dev
->state
);
998 /* Synchronize to scheduled poll. We cannot touch poll list,
999 * it can be even on different cpu. So just clear netif_running(),
1000 * and wait when poll really will happen. Actually, the best place
1001 * for this is inside dev->stop() after device stopped its irq
1002 * engine, but this requires more changes in devices. */
1004 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1005 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
1011 * Call the device specific close. This cannot fail.
1012 * Only if device is UP
1014 * We allow it to be called even after a DETACH hot-plug
1021 * Device is now down.
1024 dev
->flags
&= ~IFF_UP
;
1027 * Tell people we are down
1029 raw_notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
1036 * Device change register/unregister. These are not inline or static
1037 * as we export them to the world.
1041 * register_netdevice_notifier - register a network notifier block
1044 * Register a notifier to be called when network device events occur.
1045 * The notifier passed is linked into the kernel structures and must
1046 * not be reused until it has been unregistered. A negative errno code
1047 * is returned on a failure.
1049 * When registered all registration and up events are replayed
1050 * to the new notifier to allow device to have a race free
1051 * view of the network device list.
1054 int register_netdevice_notifier(struct notifier_block
*nb
)
1056 struct net_device
*dev
;
1060 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1062 for_each_netdev(dev
) {
1063 nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1065 if (dev
->flags
& IFF_UP
)
1066 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1074 * unregister_netdevice_notifier - unregister a network notifier block
1077 * Unregister a notifier previously registered by
1078 * register_netdevice_notifier(). The notifier is unlinked into the
1079 * kernel structures and may then be reused. A negative errno code
1080 * is returned on a failure.
1083 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1088 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1094 * call_netdevice_notifiers - call all network notifier blocks
1095 * @val: value passed unmodified to notifier function
1096 * @v: pointer passed unmodified to notifier function
1098 * Call all network notifier blocks. Parameters and return value
1099 * are as for raw_notifier_call_chain().
1102 int call_netdevice_notifiers(unsigned long val
, void *v
)
1104 return raw_notifier_call_chain(&netdev_chain
, val
, v
);
1107 /* When > 0 there are consumers of rx skb time stamps */
1108 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1110 void net_enable_timestamp(void)
1112 atomic_inc(&netstamp_needed
);
1115 void net_disable_timestamp(void)
1117 atomic_dec(&netstamp_needed
);
1120 static inline void net_timestamp(struct sk_buff
*skb
)
1122 if (atomic_read(&netstamp_needed
))
1123 __net_timestamp(skb
);
1125 skb
->tstamp
.tv64
= 0;
1129 * Support routine. Sends outgoing frames to any network
1130 * taps currently in use.
1133 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1135 struct packet_type
*ptype
;
1140 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1141 /* Never send packets back to the socket
1142 * they originated from - MvS (miquels@drinkel.ow.org)
1144 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1145 (ptype
->af_packet_priv
== NULL
||
1146 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1147 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1151 /* skb->nh should be correctly
1152 set by sender, so that the second statement is
1153 just protection against buggy protocols.
1155 skb_reset_mac_header(skb2
);
1157 if (skb_network_header(skb2
) < skb2
->data
||
1158 skb2
->network_header
> skb2
->tail
) {
1159 if (net_ratelimit())
1160 printk(KERN_CRIT
"protocol %04x is "
1162 skb2
->protocol
, dev
->name
);
1163 skb_reset_network_header(skb2
);
1166 skb2
->transport_header
= skb2
->network_header
;
1167 skb2
->pkt_type
= PACKET_OUTGOING
;
1168 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1175 void __netif_schedule(struct net_device
*dev
)
1177 if (!test_and_set_bit(__LINK_STATE_SCHED
, &dev
->state
)) {
1178 unsigned long flags
;
1179 struct softnet_data
*sd
;
1181 local_irq_save(flags
);
1182 sd
= &__get_cpu_var(softnet_data
);
1183 dev
->next_sched
= sd
->output_queue
;
1184 sd
->output_queue
= dev
;
1185 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1186 local_irq_restore(flags
);
1189 EXPORT_SYMBOL(__netif_schedule
);
1191 void __netif_rx_schedule(struct net_device
*dev
)
1193 unsigned long flags
;
1195 local_irq_save(flags
);
1197 list_add_tail(&dev
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
1199 dev
->quota
+= dev
->weight
;
1201 dev
->quota
= dev
->weight
;
1202 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1203 local_irq_restore(flags
);
1205 EXPORT_SYMBOL(__netif_rx_schedule
);
1207 void dev_kfree_skb_any(struct sk_buff
*skb
)
1209 if (in_irq() || irqs_disabled())
1210 dev_kfree_skb_irq(skb
);
1214 EXPORT_SYMBOL(dev_kfree_skb_any
);
1218 void netif_device_detach(struct net_device
*dev
)
1220 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1221 netif_running(dev
)) {
1222 netif_stop_queue(dev
);
1225 EXPORT_SYMBOL(netif_device_detach
);
1227 void netif_device_attach(struct net_device
*dev
)
1229 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1230 netif_running(dev
)) {
1231 netif_wake_queue(dev
);
1232 __netdev_watchdog_up(dev
);
1235 EXPORT_SYMBOL(netif_device_attach
);
1239 * Invalidate hardware checksum when packet is to be mangled, and
1240 * complete checksum manually on outgoing path.
1242 int skb_checksum_help(struct sk_buff
*skb
)
1245 int ret
= 0, offset
;
1247 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1248 goto out_set_summed
;
1250 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1251 /* Let GSO fix up the checksum. */
1252 goto out_set_summed
;
1255 if (skb_cloned(skb
)) {
1256 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1261 offset
= skb
->csum_start
- skb_headroom(skb
);
1262 BUG_ON(offset
> (int)skb
->len
);
1263 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1265 offset
= skb_headlen(skb
) - offset
;
1266 BUG_ON(offset
<= 0);
1267 BUG_ON(skb
->csum_offset
+ 2 > offset
);
1269 *(__sum16
*)(skb
->head
+ skb
->csum_start
+ skb
->csum_offset
) =
1272 skb
->ip_summed
= CHECKSUM_NONE
;
1278 * skb_gso_segment - Perform segmentation on skb.
1279 * @skb: buffer to segment
1280 * @features: features for the output path (see dev->features)
1282 * This function segments the given skb and returns a list of segments.
1284 * It may return NULL if the skb requires no segmentation. This is
1285 * only possible when GSO is used for verifying header integrity.
1287 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1289 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1290 struct packet_type
*ptype
;
1291 __be16 type
= skb
->protocol
;
1294 BUG_ON(skb_shinfo(skb
)->frag_list
);
1296 skb_reset_mac_header(skb
);
1297 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1298 __skb_pull(skb
, skb
->mac_len
);
1300 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1301 if (skb_header_cloned(skb
) &&
1302 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1303 return ERR_PTR(err
);
1307 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
) & 15], list
) {
1308 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1309 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1310 err
= ptype
->gso_send_check(skb
);
1311 segs
= ERR_PTR(err
);
1312 if (err
|| skb_gso_ok(skb
, features
))
1314 __skb_push(skb
, (skb
->data
-
1315 skb_network_header(skb
)));
1317 segs
= ptype
->gso_segment(skb
, features
);
1323 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1328 EXPORT_SYMBOL(skb_gso_segment
);
1330 /* Take action when hardware reception checksum errors are detected. */
1332 void netdev_rx_csum_fault(struct net_device
*dev
)
1334 if (net_ratelimit()) {
1335 printk(KERN_ERR
"%s: hw csum failure.\n",
1336 dev
? dev
->name
: "<unknown>");
1340 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1343 /* Actually, we should eliminate this check as soon as we know, that:
1344 * 1. IOMMU is present and allows to map all the memory.
1345 * 2. No high memory really exists on this machine.
1348 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1350 #ifdef CONFIG_HIGHMEM
1353 if (dev
->features
& NETIF_F_HIGHDMA
)
1356 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1357 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1365 void (*destructor
)(struct sk_buff
*skb
);
1368 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1370 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1372 struct dev_gso_cb
*cb
;
1375 struct sk_buff
*nskb
= skb
->next
;
1377 skb
->next
= nskb
->next
;
1380 } while (skb
->next
);
1382 cb
= DEV_GSO_CB(skb
);
1384 cb
->destructor(skb
);
1388 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1389 * @skb: buffer to segment
1391 * This function segments the given skb and stores the list of segments
1394 static int dev_gso_segment(struct sk_buff
*skb
)
1396 struct net_device
*dev
= skb
->dev
;
1397 struct sk_buff
*segs
;
1398 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1401 segs
= skb_gso_segment(skb
, features
);
1403 /* Verifying header integrity only. */
1407 if (unlikely(IS_ERR(segs
)))
1408 return PTR_ERR(segs
);
1411 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1412 skb
->destructor
= dev_gso_skb_destructor
;
1417 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1419 if (likely(!skb
->next
)) {
1420 if (!list_empty(&ptype_all
))
1421 dev_queue_xmit_nit(skb
, dev
);
1423 if (netif_needs_gso(dev
, skb
)) {
1424 if (unlikely(dev_gso_segment(skb
)))
1430 return dev
->hard_start_xmit(skb
, dev
);
1435 struct sk_buff
*nskb
= skb
->next
;
1438 skb
->next
= nskb
->next
;
1440 rc
= dev
->hard_start_xmit(nskb
, dev
);
1442 nskb
->next
= skb
->next
;
1446 if (unlikely((netif_queue_stopped(dev
) ||
1447 netif_subqueue_stopped(dev
, skb
->queue_mapping
)) &&
1449 return NETDEV_TX_BUSY
;
1450 } while (skb
->next
);
1452 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1459 #define HARD_TX_LOCK(dev, cpu) { \
1460 if ((dev->features & NETIF_F_LLTX) == 0) { \
1461 netif_tx_lock(dev); \
1465 #define HARD_TX_UNLOCK(dev) { \
1466 if ((dev->features & NETIF_F_LLTX) == 0) { \
1467 netif_tx_unlock(dev); \
1472 * dev_queue_xmit - transmit a buffer
1473 * @skb: buffer to transmit
1475 * Queue a buffer for transmission to a network device. The caller must
1476 * have set the device and priority and built the buffer before calling
1477 * this function. The function can be called from an interrupt.
1479 * A negative errno code is returned on a failure. A success does not
1480 * guarantee the frame will be transmitted as it may be dropped due
1481 * to congestion or traffic shaping.
1483 * -----------------------------------------------------------------------------------
1484 * I notice this method can also return errors from the queue disciplines,
1485 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1488 * Regardless of the return value, the skb is consumed, so it is currently
1489 * difficult to retry a send to this method. (You can bump the ref count
1490 * before sending to hold a reference for retry if you are careful.)
1492 * When calling this method, interrupts MUST be enabled. This is because
1493 * the BH enable code must have IRQs enabled so that it will not deadlock.
1497 int dev_queue_xmit(struct sk_buff
*skb
)
1499 struct net_device
*dev
= skb
->dev
;
1503 /* GSO will handle the following emulations directly. */
1504 if (netif_needs_gso(dev
, skb
))
1507 if (skb_shinfo(skb
)->frag_list
&&
1508 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1509 __skb_linearize(skb
))
1512 /* Fragmented skb is linearized if device does not support SG,
1513 * or if at least one of fragments is in highmem and device
1514 * does not support DMA from it.
1516 if (skb_shinfo(skb
)->nr_frags
&&
1517 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1518 __skb_linearize(skb
))
1521 /* If packet is not checksummed and device does not support
1522 * checksumming for this protocol, complete checksumming here.
1524 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1525 skb_set_transport_header(skb
, skb
->csum_start
-
1528 if (!(dev
->features
& NETIF_F_GEN_CSUM
) &&
1529 !((dev
->features
& NETIF_F_IP_CSUM
) &&
1530 skb
->protocol
== htons(ETH_P_IP
)) &&
1531 !((dev
->features
& NETIF_F_IPV6_CSUM
) &&
1532 skb
->protocol
== htons(ETH_P_IPV6
)))
1533 if (skb_checksum_help(skb
))
1538 spin_lock_prefetch(&dev
->queue_lock
);
1540 /* Disable soft irqs for various locks below. Also
1541 * stops preemption for RCU.
1545 /* Updates of qdisc are serialized by queue_lock.
1546 * The struct Qdisc which is pointed to by qdisc is now a
1547 * rcu structure - it may be accessed without acquiring
1548 * a lock (but the structure may be stale.) The freeing of the
1549 * qdisc will be deferred until it's known that there are no
1550 * more references to it.
1552 * If the qdisc has an enqueue function, we still need to
1553 * hold the queue_lock before calling it, since queue_lock
1554 * also serializes access to the device queue.
1557 q
= rcu_dereference(dev
->qdisc
);
1558 #ifdef CONFIG_NET_CLS_ACT
1559 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1562 /* Grab device queue */
1563 spin_lock(&dev
->queue_lock
);
1566 /* reset queue_mapping to zero */
1567 skb
->queue_mapping
= 0;
1568 rc
= q
->enqueue(skb
, q
);
1570 spin_unlock(&dev
->queue_lock
);
1572 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1575 spin_unlock(&dev
->queue_lock
);
1578 /* The device has no queue. Common case for software devices:
1579 loopback, all the sorts of tunnels...
1581 Really, it is unlikely that netif_tx_lock protection is necessary
1582 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1584 However, it is possible, that they rely on protection
1587 Check this and shot the lock. It is not prone from deadlocks.
1588 Either shot noqueue qdisc, it is even simpler 8)
1590 if (dev
->flags
& IFF_UP
) {
1591 int cpu
= smp_processor_id(); /* ok because BHs are off */
1593 if (dev
->xmit_lock_owner
!= cpu
) {
1595 HARD_TX_LOCK(dev
, cpu
);
1597 if (!netif_queue_stopped(dev
) &&
1598 !netif_subqueue_stopped(dev
, skb
->queue_mapping
)) {
1600 if (!dev_hard_start_xmit(skb
, dev
)) {
1601 HARD_TX_UNLOCK(dev
);
1605 HARD_TX_UNLOCK(dev
);
1606 if (net_ratelimit())
1607 printk(KERN_CRIT
"Virtual device %s asks to "
1608 "queue packet!\n", dev
->name
);
1610 /* Recursion is detected! It is possible,
1612 if (net_ratelimit())
1613 printk(KERN_CRIT
"Dead loop on virtual device "
1614 "%s, fix it urgently!\n", dev
->name
);
1619 rcu_read_unlock_bh();
1625 rcu_read_unlock_bh();
1630 /*=======================================================================
1632 =======================================================================*/
1634 int netdev_max_backlog __read_mostly
= 1000;
1635 int netdev_budget __read_mostly
= 300;
1636 int weight_p __read_mostly
= 64; /* old backlog weight */
1638 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1642 * netif_rx - post buffer to the network code
1643 * @skb: buffer to post
1645 * This function receives a packet from a device driver and queues it for
1646 * the upper (protocol) levels to process. It always succeeds. The buffer
1647 * may be dropped during processing for congestion control or by the
1651 * NET_RX_SUCCESS (no congestion)
1652 * NET_RX_CN_LOW (low congestion)
1653 * NET_RX_CN_MOD (moderate congestion)
1654 * NET_RX_CN_HIGH (high congestion)
1655 * NET_RX_DROP (packet was dropped)
1659 int netif_rx(struct sk_buff
*skb
)
1661 struct softnet_data
*queue
;
1662 unsigned long flags
;
1664 /* if netpoll wants it, pretend we never saw it */
1665 if (netpoll_rx(skb
))
1668 if (!skb
->tstamp
.tv64
)
1672 * The code is rearranged so that the path is the most
1673 * short when CPU is congested, but is still operating.
1675 local_irq_save(flags
);
1676 queue
= &__get_cpu_var(softnet_data
);
1678 __get_cpu_var(netdev_rx_stat
).total
++;
1679 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1680 if (queue
->input_pkt_queue
.qlen
) {
1683 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1684 local_irq_restore(flags
);
1685 return NET_RX_SUCCESS
;
1688 netif_rx_schedule(&queue
->backlog_dev
);
1692 __get_cpu_var(netdev_rx_stat
).dropped
++;
1693 local_irq_restore(flags
);
1699 int netif_rx_ni(struct sk_buff
*skb
)
1704 err
= netif_rx(skb
);
1705 if (local_softirq_pending())
1712 EXPORT_SYMBOL(netif_rx_ni
);
1714 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1716 struct net_device
*dev
= skb
->dev
;
1719 if (skb_bond_should_drop(skb
)) {
1723 skb
->dev
= dev
->master
;
1729 static void net_tx_action(struct softirq_action
*h
)
1731 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1733 if (sd
->completion_queue
) {
1734 struct sk_buff
*clist
;
1736 local_irq_disable();
1737 clist
= sd
->completion_queue
;
1738 sd
->completion_queue
= NULL
;
1742 struct sk_buff
*skb
= clist
;
1743 clist
= clist
->next
;
1745 BUG_TRAP(!atomic_read(&skb
->users
));
1750 if (sd
->output_queue
) {
1751 struct net_device
*head
;
1753 local_irq_disable();
1754 head
= sd
->output_queue
;
1755 sd
->output_queue
= NULL
;
1759 struct net_device
*dev
= head
;
1760 head
= head
->next_sched
;
1762 smp_mb__before_clear_bit();
1763 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1765 if (spin_trylock(&dev
->queue_lock
)) {
1767 spin_unlock(&dev
->queue_lock
);
1769 netif_schedule(dev
);
1775 static inline int deliver_skb(struct sk_buff
*skb
,
1776 struct packet_type
*pt_prev
,
1777 struct net_device
*orig_dev
)
1779 atomic_inc(&skb
->users
);
1780 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1783 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1784 /* These hooks defined here for ATM */
1786 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1787 unsigned char *addr
);
1788 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
1791 * If bridge module is loaded call bridging hook.
1792 * returns NULL if packet was consumed.
1794 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
1795 struct sk_buff
*skb
) __read_mostly
;
1796 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
1797 struct packet_type
**pt_prev
, int *ret
,
1798 struct net_device
*orig_dev
)
1800 struct net_bridge_port
*port
;
1802 if (skb
->pkt_type
== PACKET_LOOPBACK
||
1803 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
1807 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1811 return br_handle_frame_hook(port
, skb
);
1814 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1817 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1818 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
1819 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
1821 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
1822 struct packet_type
**pt_prev
,
1824 struct net_device
*orig_dev
)
1826 if (skb
->dev
->macvlan_port
== NULL
)
1830 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1833 return macvlan_handle_frame_hook(skb
);
1836 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1839 #ifdef CONFIG_NET_CLS_ACT
1840 /* TODO: Maybe we should just force sch_ingress to be compiled in
1841 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1842 * a compare and 2 stores extra right now if we dont have it on
1843 * but have CONFIG_NET_CLS_ACT
1844 * NOTE: This doesnt stop any functionality; if you dont have
1845 * the ingress scheduler, you just cant add policies on ingress.
1848 static int ing_filter(struct sk_buff
*skb
)
1851 struct net_device
*dev
= skb
->dev
;
1852 int result
= TC_ACT_OK
;
1854 if (dev
->qdisc_ingress
) {
1855 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1856 if (MAX_RED_LOOP
< ttl
++) {
1857 printk(KERN_WARNING
"Redir loop detected Dropping packet (%d->%d)\n",
1858 skb
->iif
, skb
->dev
->ifindex
);
1862 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1864 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1866 spin_lock(&dev
->ingress_lock
);
1867 if ((q
= dev
->qdisc_ingress
) != NULL
)
1868 result
= q
->enqueue(skb
, q
);
1869 spin_unlock(&dev
->ingress_lock
);
1877 int netif_receive_skb(struct sk_buff
*skb
)
1879 struct packet_type
*ptype
, *pt_prev
;
1880 struct net_device
*orig_dev
;
1881 int ret
= NET_RX_DROP
;
1884 /* if we've gotten here through NAPI, check netpoll */
1885 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1888 if (!skb
->tstamp
.tv64
)
1892 skb
->iif
= skb
->dev
->ifindex
;
1894 orig_dev
= skb_bond(skb
);
1899 __get_cpu_var(netdev_rx_stat
).total
++;
1901 skb_reset_network_header(skb
);
1902 skb_reset_transport_header(skb
);
1903 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1909 #ifdef CONFIG_NET_CLS_ACT
1910 if (skb
->tc_verd
& TC_NCLS
) {
1911 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1916 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1917 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1919 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1924 #ifdef CONFIG_NET_CLS_ACT
1926 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1927 pt_prev
= NULL
; /* noone else should process this after*/
1929 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1932 ret
= ing_filter(skb
);
1934 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
1943 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
1946 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
1950 type
= skb
->protocol
;
1951 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
1952 if (ptype
->type
== type
&&
1953 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
1955 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1961 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1964 /* Jamal, now you will not able to escape explaining
1965 * me how you were going to use this. :-)
1975 static int process_backlog(struct net_device
*backlog_dev
, int *budget
)
1978 int quota
= min(backlog_dev
->quota
, *budget
);
1979 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1980 unsigned long start_time
= jiffies
;
1982 backlog_dev
->weight
= weight_p
;
1984 struct sk_buff
*skb
;
1985 struct net_device
*dev
;
1987 local_irq_disable();
1988 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
1995 netif_receive_skb(skb
);
2001 if (work
>= quota
|| jiffies
- start_time
> 1)
2006 backlog_dev
->quota
-= work
;
2011 backlog_dev
->quota
-= work
;
2014 list_del(&backlog_dev
->poll_list
);
2015 smp_mb__before_clear_bit();
2016 netif_poll_enable(backlog_dev
);
2022 static void net_rx_action(struct softirq_action
*h
)
2024 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2025 unsigned long start_time
= jiffies
;
2026 int budget
= netdev_budget
;
2029 local_irq_disable();
2031 while (!list_empty(&queue
->poll_list
)) {
2032 struct net_device
*dev
;
2034 if (budget
<= 0 || jiffies
- start_time
> 1)
2039 dev
= list_entry(queue
->poll_list
.next
,
2040 struct net_device
, poll_list
);
2041 have
= netpoll_poll_lock(dev
);
2043 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
2044 netpoll_poll_unlock(have
);
2045 local_irq_disable();
2046 list_move_tail(&dev
->poll_list
, &queue
->poll_list
);
2048 dev
->quota
+= dev
->weight
;
2050 dev
->quota
= dev
->weight
;
2052 netpoll_poll_unlock(have
);
2054 local_irq_disable();
2059 #ifdef CONFIG_NET_DMA
2061 * There may not be any more sk_buffs coming right now, so push
2062 * any pending DMA copies to hardware
2064 if (!cpus_empty(net_dma
.channel_mask
)) {
2066 for_each_cpu_mask(chan_idx
, net_dma
.channel_mask
) {
2067 struct dma_chan
*chan
= net_dma
.channels
[chan_idx
];
2069 dma_async_memcpy_issue_pending(chan
);
2076 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2077 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2081 static gifconf_func_t
* gifconf_list
[NPROTO
];
2084 * register_gifconf - register a SIOCGIF handler
2085 * @family: Address family
2086 * @gifconf: Function handler
2088 * Register protocol dependent address dumping routines. The handler
2089 * that is passed must not be freed or reused until it has been replaced
2090 * by another handler.
2092 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2094 if (family
>= NPROTO
)
2096 gifconf_list
[family
] = gifconf
;
2102 * Map an interface index to its name (SIOCGIFNAME)
2106 * We need this ioctl for efficient implementation of the
2107 * if_indextoname() function required by the IPv6 API. Without
2108 * it, we would have to search all the interfaces to find a
2112 static int dev_ifname(struct ifreq __user
*arg
)
2114 struct net_device
*dev
;
2118 * Fetch the caller's info block.
2121 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2124 read_lock(&dev_base_lock
);
2125 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
2127 read_unlock(&dev_base_lock
);
2131 strcpy(ifr
.ifr_name
, dev
->name
);
2132 read_unlock(&dev_base_lock
);
2134 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2140 * Perform a SIOCGIFCONF call. This structure will change
2141 * size eventually, and there is nothing I can do about it.
2142 * Thus we will need a 'compatibility mode'.
2145 static int dev_ifconf(char __user
*arg
)
2148 struct net_device
*dev
;
2155 * Fetch the caller's info block.
2158 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2165 * Loop over the interfaces, and write an info block for each.
2169 for_each_netdev(dev
) {
2170 for (i
= 0; i
< NPROTO
; i
++) {
2171 if (gifconf_list
[i
]) {
2174 done
= gifconf_list
[i
](dev
, NULL
, 0);
2176 done
= gifconf_list
[i
](dev
, pos
+ total
,
2186 * All done. Write the updated control block back to the caller.
2188 ifc
.ifc_len
= total
;
2191 * Both BSD and Solaris return 0 here, so we do too.
2193 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2196 #ifdef CONFIG_PROC_FS
2198 * This is invoked by the /proc filesystem handler to display a device
2201 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2204 struct net_device
*dev
;
2206 read_lock(&dev_base_lock
);
2208 return SEQ_START_TOKEN
;
2211 for_each_netdev(dev
)
2218 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2221 return v
== SEQ_START_TOKEN
?
2222 first_net_device() : next_net_device((struct net_device
*)v
);
2225 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2227 read_unlock(&dev_base_lock
);
2230 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2232 struct net_device_stats
*stats
= dev
->get_stats(dev
);
2234 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2235 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2236 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2238 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2239 stats
->rx_fifo_errors
,
2240 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2241 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2242 stats
->rx_compressed
, stats
->multicast
,
2243 stats
->tx_bytes
, stats
->tx_packets
,
2244 stats
->tx_errors
, stats
->tx_dropped
,
2245 stats
->tx_fifo_errors
, stats
->collisions
,
2246 stats
->tx_carrier_errors
+
2247 stats
->tx_aborted_errors
+
2248 stats
->tx_window_errors
+
2249 stats
->tx_heartbeat_errors
,
2250 stats
->tx_compressed
);
2254 * Called from the PROCfs module. This now uses the new arbitrary sized
2255 * /proc/net interface to create /proc/net/dev
2257 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2259 if (v
== SEQ_START_TOKEN
)
2260 seq_puts(seq
, "Inter-| Receive "
2262 " face |bytes packets errs drop fifo frame "
2263 "compressed multicast|bytes packets errs "
2264 "drop fifo colls carrier compressed\n");
2266 dev_seq_printf_stats(seq
, v
);
2270 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2272 struct netif_rx_stats
*rc
= NULL
;
2274 while (*pos
< NR_CPUS
)
2275 if (cpu_online(*pos
)) {
2276 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2283 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2285 return softnet_get_online(pos
);
2288 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2291 return softnet_get_online(pos
);
2294 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2298 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2300 struct netif_rx_stats
*s
= v
;
2302 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2303 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2304 0, 0, 0, 0, /* was fastroute */
2309 static const struct seq_operations dev_seq_ops
= {
2310 .start
= dev_seq_start
,
2311 .next
= dev_seq_next
,
2312 .stop
= dev_seq_stop
,
2313 .show
= dev_seq_show
,
2316 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2318 return seq_open(file
, &dev_seq_ops
);
2321 static const struct file_operations dev_seq_fops
= {
2322 .owner
= THIS_MODULE
,
2323 .open
= dev_seq_open
,
2325 .llseek
= seq_lseek
,
2326 .release
= seq_release
,
2329 static const struct seq_operations softnet_seq_ops
= {
2330 .start
= softnet_seq_start
,
2331 .next
= softnet_seq_next
,
2332 .stop
= softnet_seq_stop
,
2333 .show
= softnet_seq_show
,
2336 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2338 return seq_open(file
, &softnet_seq_ops
);
2341 static const struct file_operations softnet_seq_fops
= {
2342 .owner
= THIS_MODULE
,
2343 .open
= softnet_seq_open
,
2345 .llseek
= seq_lseek
,
2346 .release
= seq_release
,
2349 static void *ptype_get_idx(loff_t pos
)
2351 struct packet_type
*pt
= NULL
;
2355 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
2361 for (t
= 0; t
< 16; t
++) {
2362 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
2371 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2374 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
2377 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2379 struct packet_type
*pt
;
2380 struct list_head
*nxt
;
2384 if (v
== SEQ_START_TOKEN
)
2385 return ptype_get_idx(0);
2388 nxt
= pt
->list
.next
;
2389 if (pt
->type
== htons(ETH_P_ALL
)) {
2390 if (nxt
!= &ptype_all
)
2393 nxt
= ptype_base
[0].next
;
2395 hash
= ntohs(pt
->type
) & 15;
2397 while (nxt
== &ptype_base
[hash
]) {
2400 nxt
= ptype_base
[hash
].next
;
2403 return list_entry(nxt
, struct packet_type
, list
);
2406 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
2411 static void ptype_seq_decode(struct seq_file
*seq
, void *sym
)
2413 #ifdef CONFIG_KALLSYMS
2414 unsigned long offset
= 0, symsize
;
2415 const char *symname
;
2419 symname
= kallsyms_lookup((unsigned long)sym
, &symsize
, &offset
,
2426 modname
= delim
= "";
2427 seq_printf(seq
, "%s%s%s%s+0x%lx", delim
, modname
, delim
,
2433 seq_printf(seq
, "[%p]", sym
);
2436 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
2438 struct packet_type
*pt
= v
;
2440 if (v
== SEQ_START_TOKEN
)
2441 seq_puts(seq
, "Type Device Function\n");
2443 if (pt
->type
== htons(ETH_P_ALL
))
2444 seq_puts(seq
, "ALL ");
2446 seq_printf(seq
, "%04x", ntohs(pt
->type
));
2448 seq_printf(seq
, " %-8s ",
2449 pt
->dev
? pt
->dev
->name
: "");
2450 ptype_seq_decode(seq
, pt
->func
);
2451 seq_putc(seq
, '\n');
2457 static const struct seq_operations ptype_seq_ops
= {
2458 .start
= ptype_seq_start
,
2459 .next
= ptype_seq_next
,
2460 .stop
= ptype_seq_stop
,
2461 .show
= ptype_seq_show
,
2464 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
2466 return seq_open(file
, &ptype_seq_ops
);
2469 static const struct file_operations ptype_seq_fops
= {
2470 .owner
= THIS_MODULE
,
2471 .open
= ptype_seq_open
,
2473 .llseek
= seq_lseek
,
2474 .release
= seq_release
,
2478 static int __init
dev_proc_init(void)
2482 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2484 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2486 if (!proc_net_fops_create("ptype", S_IRUGO
, &ptype_seq_fops
))
2489 if (wext_proc_init())
2495 proc_net_remove("ptype");
2497 proc_net_remove("softnet_stat");
2499 proc_net_remove("dev");
2503 #define dev_proc_init() 0
2504 #endif /* CONFIG_PROC_FS */
2508 * netdev_set_master - set up master/slave pair
2509 * @slave: slave device
2510 * @master: new master device
2512 * Changes the master device of the slave. Pass %NULL to break the
2513 * bonding. The caller must hold the RTNL semaphore. On a failure
2514 * a negative errno code is returned. On success the reference counts
2515 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2516 * function returns zero.
2518 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2520 struct net_device
*old
= slave
->master
;
2530 slave
->master
= master
;
2538 slave
->flags
|= IFF_SLAVE
;
2540 slave
->flags
&= ~IFF_SLAVE
;
2542 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2546 static void __dev_set_promiscuity(struct net_device
*dev
, int inc
)
2548 unsigned short old_flags
= dev
->flags
;
2552 if ((dev
->promiscuity
+= inc
) == 0)
2553 dev
->flags
&= ~IFF_PROMISC
;
2555 dev
->flags
|= IFF_PROMISC
;
2556 if (dev
->flags
!= old_flags
) {
2557 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2558 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2560 audit_log(current
->audit_context
, GFP_ATOMIC
,
2561 AUDIT_ANOM_PROMISCUOUS
,
2562 "dev=%s prom=%d old_prom=%d auid=%u",
2563 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2564 (old_flags
& IFF_PROMISC
),
2565 audit_get_loginuid(current
->audit_context
));
2567 if (dev
->change_rx_flags
)
2568 dev
->change_rx_flags(dev
, IFF_PROMISC
);
2573 * dev_set_promiscuity - update promiscuity count on a device
2577 * Add or remove promiscuity from a device. While the count in the device
2578 * remains above zero the interface remains promiscuous. Once it hits zero
2579 * the device reverts back to normal filtering operation. A negative inc
2580 * value is used to drop promiscuity on the device.
2582 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2584 unsigned short old_flags
= dev
->flags
;
2586 __dev_set_promiscuity(dev
, inc
);
2587 if (dev
->flags
!= old_flags
)
2588 dev_set_rx_mode(dev
);
2592 * dev_set_allmulti - update allmulti count on a device
2596 * Add or remove reception of all multicast frames to a device. While the
2597 * count in the device remains above zero the interface remains listening
2598 * to all interfaces. Once it hits zero the device reverts back to normal
2599 * filtering operation. A negative @inc value is used to drop the counter
2600 * when releasing a resource needing all multicasts.
2603 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2605 unsigned short old_flags
= dev
->flags
;
2609 dev
->flags
|= IFF_ALLMULTI
;
2610 if ((dev
->allmulti
+= inc
) == 0)
2611 dev
->flags
&= ~IFF_ALLMULTI
;
2612 if (dev
->flags
^ old_flags
) {
2613 if (dev
->change_rx_flags
)
2614 dev
->change_rx_flags(dev
, IFF_ALLMULTI
);
2615 dev_set_rx_mode(dev
);
2620 * Upload unicast and multicast address lists to device and
2621 * configure RX filtering. When the device doesn't support unicast
2622 * filtering it is put in promiscous mode while unicast addresses
2625 void __dev_set_rx_mode(struct net_device
*dev
)
2627 /* dev_open will call this function so the list will stay sane. */
2628 if (!(dev
->flags
&IFF_UP
))
2631 if (!netif_device_present(dev
))
2634 if (dev
->set_rx_mode
)
2635 dev
->set_rx_mode(dev
);
2637 /* Unicast addresses changes may only happen under the rtnl,
2638 * therefore calling __dev_set_promiscuity here is safe.
2640 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
2641 __dev_set_promiscuity(dev
, 1);
2642 dev
->uc_promisc
= 1;
2643 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
2644 __dev_set_promiscuity(dev
, -1);
2645 dev
->uc_promisc
= 0;
2648 if (dev
->set_multicast_list
)
2649 dev
->set_multicast_list(dev
);
2653 void dev_set_rx_mode(struct net_device
*dev
)
2655 netif_tx_lock_bh(dev
);
2656 __dev_set_rx_mode(dev
);
2657 netif_tx_unlock_bh(dev
);
2660 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
2661 void *addr
, int alen
, int glbl
)
2663 struct dev_addr_list
*da
;
2665 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
2666 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2667 alen
== da
->da_addrlen
) {
2669 int old_glbl
= da
->da_gusers
;
2686 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
2687 void *addr
, int alen
, int glbl
)
2689 struct dev_addr_list
*da
;
2691 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
2692 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2693 da
->da_addrlen
== alen
) {
2695 int old_glbl
= da
->da_gusers
;
2705 da
= kmalloc(sizeof(*da
), GFP_ATOMIC
);
2708 memcpy(da
->da_addr
, addr
, alen
);
2709 da
->da_addrlen
= alen
;
2711 da
->da_gusers
= glbl
? 1 : 0;
2719 * dev_unicast_delete - Release secondary unicast address.
2722 * Release reference to a secondary unicast address and remove it
2723 * from the device if the reference count drop to zero.
2725 * The caller must hold the rtnl_mutex.
2727 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
2733 netif_tx_lock_bh(dev
);
2734 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
2736 __dev_set_rx_mode(dev
);
2737 netif_tx_unlock_bh(dev
);
2740 EXPORT_SYMBOL(dev_unicast_delete
);
2743 * dev_unicast_add - add a secondary unicast address
2746 * Add a secondary unicast address to the device or increase
2747 * the reference count if it already exists.
2749 * The caller must hold the rtnl_mutex.
2751 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
2757 netif_tx_lock_bh(dev
);
2758 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
2760 __dev_set_rx_mode(dev
);
2761 netif_tx_unlock_bh(dev
);
2764 EXPORT_SYMBOL(dev_unicast_add
);
2766 static void __dev_addr_discard(struct dev_addr_list
**list
)
2768 struct dev_addr_list
*tmp
;
2770 while (*list
!= NULL
) {
2773 if (tmp
->da_users
> tmp
->da_gusers
)
2774 printk("__dev_addr_discard: address leakage! "
2775 "da_users=%d\n", tmp
->da_users
);
2780 static void dev_addr_discard(struct net_device
*dev
)
2782 netif_tx_lock_bh(dev
);
2784 __dev_addr_discard(&dev
->uc_list
);
2787 __dev_addr_discard(&dev
->mc_list
);
2790 netif_tx_unlock_bh(dev
);
2793 unsigned dev_get_flags(const struct net_device
*dev
)
2797 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2802 (dev
->gflags
& (IFF_PROMISC
|
2805 if (netif_running(dev
)) {
2806 if (netif_oper_up(dev
))
2807 flags
|= IFF_RUNNING
;
2808 if (netif_carrier_ok(dev
))
2809 flags
|= IFF_LOWER_UP
;
2810 if (netif_dormant(dev
))
2811 flags
|= IFF_DORMANT
;
2817 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2820 int old_flags
= dev
->flags
;
2825 * Set the flags on our device.
2828 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2829 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2831 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2835 * Load in the correct multicast list now the flags have changed.
2838 if (dev
->change_rx_flags
&& (dev
->flags
^ flags
) & IFF_MULTICAST
)
2839 dev
->change_rx_flags(dev
, IFF_MULTICAST
);
2841 dev_set_rx_mode(dev
);
2844 * Have we downed the interface. We handle IFF_UP ourselves
2845 * according to user attempts to set it, rather than blindly
2850 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2851 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2854 dev_set_rx_mode(dev
);
2857 if (dev
->flags
& IFF_UP
&&
2858 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2860 raw_notifier_call_chain(&netdev_chain
,
2861 NETDEV_CHANGE
, dev
);
2863 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2864 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2865 dev
->gflags
^= IFF_PROMISC
;
2866 dev_set_promiscuity(dev
, inc
);
2869 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2870 is important. Some (broken) drivers set IFF_PROMISC, when
2871 IFF_ALLMULTI is requested not asking us and not reporting.
2873 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2874 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2875 dev
->gflags
^= IFF_ALLMULTI
;
2876 dev_set_allmulti(dev
, inc
);
2879 /* Exclude state transition flags, already notified */
2880 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
2882 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
2887 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2891 if (new_mtu
== dev
->mtu
)
2894 /* MTU must be positive. */
2898 if (!netif_device_present(dev
))
2902 if (dev
->change_mtu
)
2903 err
= dev
->change_mtu(dev
, new_mtu
);
2906 if (!err
&& dev
->flags
& IFF_UP
)
2907 raw_notifier_call_chain(&netdev_chain
,
2908 NETDEV_CHANGEMTU
, dev
);
2912 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2916 if (!dev
->set_mac_address
)
2918 if (sa
->sa_family
!= dev
->type
)
2920 if (!netif_device_present(dev
))
2922 err
= dev
->set_mac_address(dev
, sa
);
2924 raw_notifier_call_chain(&netdev_chain
,
2925 NETDEV_CHANGEADDR
, dev
);
2930 * Perform the SIOCxIFxxx calls.
2932 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2935 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2941 case SIOCGIFFLAGS
: /* Get interface flags */
2942 ifr
->ifr_flags
= dev_get_flags(dev
);
2945 case SIOCSIFFLAGS
: /* Set interface flags */
2946 return dev_change_flags(dev
, ifr
->ifr_flags
);
2948 case SIOCGIFMETRIC
: /* Get the metric on the interface
2949 (currently unused) */
2950 ifr
->ifr_metric
= 0;
2953 case SIOCSIFMETRIC
: /* Set the metric on the interface
2954 (currently unused) */
2957 case SIOCGIFMTU
: /* Get the MTU of a device */
2958 ifr
->ifr_mtu
= dev
->mtu
;
2961 case SIOCSIFMTU
: /* Set the MTU of a device */
2962 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
2966 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
2968 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
2969 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2970 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
2974 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
2976 case SIOCSIFHWBROADCAST
:
2977 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
2979 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
2980 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2981 raw_notifier_call_chain(&netdev_chain
,
2982 NETDEV_CHANGEADDR
, dev
);
2986 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
2987 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
2988 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
2989 ifr
->ifr_map
.irq
= dev
->irq
;
2990 ifr
->ifr_map
.dma
= dev
->dma
;
2991 ifr
->ifr_map
.port
= dev
->if_port
;
2995 if (dev
->set_config
) {
2996 if (!netif_device_present(dev
))
2998 return dev
->set_config(dev
, &ifr
->ifr_map
);
3003 if (!dev
->set_multicast_list
||
3004 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3006 if (!netif_device_present(dev
))
3008 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
3012 if (!dev
->set_multicast_list
||
3013 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3015 if (!netif_device_present(dev
))
3017 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
3021 ifr
->ifr_ifindex
= dev
->ifindex
;
3025 ifr
->ifr_qlen
= dev
->tx_queue_len
;
3029 if (ifr
->ifr_qlen
< 0)
3031 dev
->tx_queue_len
= ifr
->ifr_qlen
;
3035 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
3036 return dev_change_name(dev
, ifr
->ifr_newname
);
3039 * Unknown or private ioctl
3043 if ((cmd
>= SIOCDEVPRIVATE
&&
3044 cmd
<= SIOCDEVPRIVATE
+ 15) ||
3045 cmd
== SIOCBONDENSLAVE
||
3046 cmd
== SIOCBONDRELEASE
||
3047 cmd
== SIOCBONDSETHWADDR
||
3048 cmd
== SIOCBONDSLAVEINFOQUERY
||
3049 cmd
== SIOCBONDINFOQUERY
||
3050 cmd
== SIOCBONDCHANGEACTIVE
||
3051 cmd
== SIOCGMIIPHY
||
3052 cmd
== SIOCGMIIREG
||
3053 cmd
== SIOCSMIIREG
||
3054 cmd
== SIOCBRADDIF
||
3055 cmd
== SIOCBRDELIF
||
3056 cmd
== SIOCWANDEV
) {
3058 if (dev
->do_ioctl
) {
3059 if (netif_device_present(dev
))
3060 err
= dev
->do_ioctl(dev
, ifr
,
3073 * This function handles all "interface"-type I/O control requests. The actual
3074 * 'doing' part of this is dev_ifsioc above.
3078 * dev_ioctl - network device ioctl
3079 * @cmd: command to issue
3080 * @arg: pointer to a struct ifreq in user space
3082 * Issue ioctl functions to devices. This is normally called by the
3083 * user space syscall interfaces but can sometimes be useful for
3084 * other purposes. The return value is the return from the syscall if
3085 * positive or a negative errno code on error.
3088 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
3094 /* One special case: SIOCGIFCONF takes ifconf argument
3095 and requires shared lock, because it sleeps writing
3099 if (cmd
== SIOCGIFCONF
) {
3101 ret
= dev_ifconf((char __user
*) arg
);
3105 if (cmd
== SIOCGIFNAME
)
3106 return dev_ifname((struct ifreq __user
*)arg
);
3108 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3111 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
3113 colon
= strchr(ifr
.ifr_name
, ':');
3118 * See which interface the caller is talking about.
3123 * These ioctl calls:
3124 * - can be done by all.
3125 * - atomic and do not require locking.
3136 dev_load(ifr
.ifr_name
);
3137 read_lock(&dev_base_lock
);
3138 ret
= dev_ifsioc(&ifr
, cmd
);
3139 read_unlock(&dev_base_lock
);
3143 if (copy_to_user(arg
, &ifr
,
3144 sizeof(struct ifreq
)))
3150 dev_load(ifr
.ifr_name
);
3152 ret
= dev_ethtool(&ifr
);
3157 if (copy_to_user(arg
, &ifr
,
3158 sizeof(struct ifreq
)))
3164 * These ioctl calls:
3165 * - require superuser power.
3166 * - require strict serialization.
3172 if (!capable(CAP_NET_ADMIN
))
3174 dev_load(ifr
.ifr_name
);
3176 ret
= dev_ifsioc(&ifr
, cmd
);
3181 if (copy_to_user(arg
, &ifr
,
3182 sizeof(struct ifreq
)))
3188 * These ioctl calls:
3189 * - require superuser power.
3190 * - require strict serialization.
3191 * - do not return a value
3201 case SIOCSIFHWBROADCAST
:
3204 case SIOCBONDENSLAVE
:
3205 case SIOCBONDRELEASE
:
3206 case SIOCBONDSETHWADDR
:
3207 case SIOCBONDCHANGEACTIVE
:
3210 if (!capable(CAP_NET_ADMIN
))
3213 case SIOCBONDSLAVEINFOQUERY
:
3214 case SIOCBONDINFOQUERY
:
3215 dev_load(ifr
.ifr_name
);
3217 ret
= dev_ifsioc(&ifr
, cmd
);
3222 /* Get the per device memory space. We can add this but
3223 * currently do not support it */
3225 /* Set the per device memory buffer space.
3226 * Not applicable in our case */
3231 * Unknown or private ioctl.
3234 if (cmd
== SIOCWANDEV
||
3235 (cmd
>= SIOCDEVPRIVATE
&&
3236 cmd
<= SIOCDEVPRIVATE
+ 15)) {
3237 dev_load(ifr
.ifr_name
);
3239 ret
= dev_ifsioc(&ifr
, cmd
);
3241 if (!ret
&& copy_to_user(arg
, &ifr
,
3242 sizeof(struct ifreq
)))
3246 /* Take care of Wireless Extensions */
3247 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
3248 return wext_handle_ioctl(&ifr
, cmd
, arg
);
3255 * dev_new_index - allocate an ifindex
3257 * Returns a suitable unique value for a new device interface
3258 * number. The caller must hold the rtnl semaphore or the
3259 * dev_base_lock to be sure it remains unique.
3261 static int dev_new_index(void)
3267 if (!__dev_get_by_index(ifindex
))
3272 static int dev_boot_phase
= 1;
3274 /* Delayed registration/unregisteration */
3275 static DEFINE_SPINLOCK(net_todo_list_lock
);
3276 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
3278 static void net_set_todo(struct net_device
*dev
)
3280 spin_lock(&net_todo_list_lock
);
3281 list_add_tail(&dev
->todo_list
, &net_todo_list
);
3282 spin_unlock(&net_todo_list_lock
);
3286 * register_netdevice - register a network device
3287 * @dev: device to register
3289 * Take a completed network device structure and add it to the kernel
3290 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3291 * chain. 0 is returned on success. A negative errno code is returned
3292 * on a failure to set up the device, or if the name is a duplicate.
3294 * Callers must hold the rtnl semaphore. You may want
3295 * register_netdev() instead of this.
3298 * The locking appears insufficient to guarantee two parallel registers
3299 * will not get the same name.
3302 int register_netdevice(struct net_device
*dev
)
3304 struct hlist_head
*head
;
3305 struct hlist_node
*p
;
3308 BUG_ON(dev_boot_phase
);
3313 /* When net_device's are persistent, this will be fatal. */
3314 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
3316 spin_lock_init(&dev
->queue_lock
);
3317 spin_lock_init(&dev
->_xmit_lock
);
3318 netdev_set_lockdep_class(&dev
->_xmit_lock
, dev
->type
);
3319 dev
->xmit_lock_owner
= -1;
3320 spin_lock_init(&dev
->ingress_lock
);
3324 /* Init, if this function is available */
3326 ret
= dev
->init(dev
);
3334 if (!dev_valid_name(dev
->name
)) {
3339 dev
->ifindex
= dev_new_index();
3340 if (dev
->iflink
== -1)
3341 dev
->iflink
= dev
->ifindex
;
3343 /* Check for existence of name */
3344 head
= dev_name_hash(dev
->name
);
3345 hlist_for_each(p
, head
) {
3346 struct net_device
*d
3347 = hlist_entry(p
, struct net_device
, name_hlist
);
3348 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
3354 /* Fix illegal checksum combinations */
3355 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
3356 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3357 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
3359 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
3362 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
3363 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3364 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
3366 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
3370 /* Fix illegal SG+CSUM combinations. */
3371 if ((dev
->features
& NETIF_F_SG
) &&
3372 !(dev
->features
& NETIF_F_ALL_CSUM
)) {
3373 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no checksum feature.\n",
3375 dev
->features
&= ~NETIF_F_SG
;
3378 /* TSO requires that SG is present as well. */
3379 if ((dev
->features
& NETIF_F_TSO
) &&
3380 !(dev
->features
& NETIF_F_SG
)) {
3381 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no SG feature.\n",
3383 dev
->features
&= ~NETIF_F_TSO
;
3385 if (dev
->features
& NETIF_F_UFO
) {
3386 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
3387 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3388 "NETIF_F_HW_CSUM feature.\n",
3390 dev
->features
&= ~NETIF_F_UFO
;
3392 if (!(dev
->features
& NETIF_F_SG
)) {
3393 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3394 "NETIF_F_SG feature.\n",
3396 dev
->features
&= ~NETIF_F_UFO
;
3401 * nil rebuild_header routine,
3402 * that should be never called and used as just bug trap.
3405 if (!dev
->rebuild_header
)
3406 dev
->rebuild_header
= default_rebuild_header
;
3408 ret
= netdev_register_sysfs(dev
);
3411 dev
->reg_state
= NETREG_REGISTERED
;
3414 * Default initial state at registry is that the
3415 * device is present.
3418 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
3420 dev_init_scheduler(dev
);
3421 write_lock_bh(&dev_base_lock
);
3422 list_add_tail(&dev
->dev_list
, &dev_base_head
);
3423 hlist_add_head(&dev
->name_hlist
, head
);
3424 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
3426 write_unlock_bh(&dev_base_lock
);
3428 /* Notify protocols, that a new device appeared. */
3429 raw_notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
3438 * register_netdev - register a network device
3439 * @dev: device to register
3441 * Take a completed network device structure and add it to the kernel
3442 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3443 * chain. 0 is returned on success. A negative errno code is returned
3444 * on a failure to set up the device, or if the name is a duplicate.
3446 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3447 * and expands the device name if you passed a format string to
3450 int register_netdev(struct net_device
*dev
)
3457 * If the name is a format string the caller wants us to do a
3460 if (strchr(dev
->name
, '%')) {
3461 err
= dev_alloc_name(dev
, dev
->name
);
3466 err
= register_netdevice(dev
);
3471 EXPORT_SYMBOL(register_netdev
);
3474 * netdev_wait_allrefs - wait until all references are gone.
3476 * This is called when unregistering network devices.
3478 * Any protocol or device that holds a reference should register
3479 * for netdevice notification, and cleanup and put back the
3480 * reference if they receive an UNREGISTER event.
3481 * We can get stuck here if buggy protocols don't correctly
3484 static void netdev_wait_allrefs(struct net_device
*dev
)
3486 unsigned long rebroadcast_time
, warning_time
;
3488 rebroadcast_time
= warning_time
= jiffies
;
3489 while (atomic_read(&dev
->refcnt
) != 0) {
3490 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
3493 /* Rebroadcast unregister notification */
3494 raw_notifier_call_chain(&netdev_chain
,
3495 NETDEV_UNREGISTER
, dev
);
3497 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
3499 /* We must not have linkwatch events
3500 * pending on unregister. If this
3501 * happens, we simply run the queue
3502 * unscheduled, resulting in a noop
3505 linkwatch_run_queue();
3510 rebroadcast_time
= jiffies
;
3515 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
3516 printk(KERN_EMERG
"unregister_netdevice: "
3517 "waiting for %s to become free. Usage "
3519 dev
->name
, atomic_read(&dev
->refcnt
));
3520 warning_time
= jiffies
;
3529 * register_netdevice(x1);
3530 * register_netdevice(x2);
3532 * unregister_netdevice(y1);
3533 * unregister_netdevice(y2);
3539 * We are invoked by rtnl_unlock() after it drops the semaphore.
3540 * This allows us to deal with problems:
3541 * 1) We can delete sysfs objects which invoke hotplug
3542 * without deadlocking with linkwatch via keventd.
3543 * 2) Since we run with the RTNL semaphore not held, we can sleep
3544 * safely in order to wait for the netdev refcnt to drop to zero.
3546 static DEFINE_MUTEX(net_todo_run_mutex
);
3547 void netdev_run_todo(void)
3549 struct list_head list
;
3551 /* Need to guard against multiple cpu's getting out of order. */
3552 mutex_lock(&net_todo_run_mutex
);
3554 /* Not safe to do outside the semaphore. We must not return
3555 * until all unregister events invoked by the local processor
3556 * have been completed (either by this todo run, or one on
3559 if (list_empty(&net_todo_list
))
3562 /* Snapshot list, allow later requests */
3563 spin_lock(&net_todo_list_lock
);
3564 list_replace_init(&net_todo_list
, &list
);
3565 spin_unlock(&net_todo_list_lock
);
3567 while (!list_empty(&list
)) {
3568 struct net_device
*dev
3569 = list_entry(list
.next
, struct net_device
, todo_list
);
3570 list_del(&dev
->todo_list
);
3572 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
3573 printk(KERN_ERR
"network todo '%s' but state %d\n",
3574 dev
->name
, dev
->reg_state
);
3579 dev
->reg_state
= NETREG_UNREGISTERED
;
3581 netdev_wait_allrefs(dev
);
3584 BUG_ON(atomic_read(&dev
->refcnt
));
3585 BUG_TRAP(!dev
->ip_ptr
);
3586 BUG_TRAP(!dev
->ip6_ptr
);
3587 BUG_TRAP(!dev
->dn_ptr
);
3589 if (dev
->destructor
)
3590 dev
->destructor(dev
);
3592 /* Free network device */
3593 kobject_put(&dev
->dev
.kobj
);
3597 mutex_unlock(&net_todo_run_mutex
);
3600 static struct net_device_stats
*internal_stats(struct net_device
*dev
)
3606 * alloc_netdev_mq - allocate network device
3607 * @sizeof_priv: size of private data to allocate space for
3608 * @name: device name format string
3609 * @setup: callback to initialize device
3610 * @queue_count: the number of subqueues to allocate
3612 * Allocates a struct net_device with private data area for driver use
3613 * and performs basic initialization. Also allocates subquue structs
3614 * for each queue on the device at the end of the netdevice.
3616 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
3617 void (*setup
)(struct net_device
*), unsigned int queue_count
)
3620 struct net_device
*dev
;
3623 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
3625 /* ensure 32-byte alignment of both the device and private area */
3626 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
+
3627 (sizeof(struct net_device_subqueue
) * (queue_count
- 1))) &
3628 ~NETDEV_ALIGN_CONST
;
3629 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
3631 p
= kzalloc(alloc_size
, GFP_KERNEL
);
3633 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
3637 dev
= (struct net_device
*)
3638 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3639 dev
->padded
= (char *)dev
- (char *)p
;
3642 dev
->priv
= ((char *)dev
+
3643 ((sizeof(struct net_device
) +
3644 (sizeof(struct net_device_subqueue
) *
3645 (queue_count
- 1)) + NETDEV_ALIGN_CONST
)
3646 & ~NETDEV_ALIGN_CONST
));
3649 dev
->egress_subqueue_count
= queue_count
;
3651 dev
->get_stats
= internal_stats
;
3653 strcpy(dev
->name
, name
);
3656 EXPORT_SYMBOL(alloc_netdev_mq
);
3659 * free_netdev - free network device
3662 * This function does the last stage of destroying an allocated device
3663 * interface. The reference to the device object is released.
3664 * If this is the last reference then it will be freed.
3666 void free_netdev(struct net_device
*dev
)
3669 /* Compatibility with error handling in drivers */
3670 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3671 kfree((char *)dev
- dev
->padded
);
3675 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3676 dev
->reg_state
= NETREG_RELEASED
;
3678 /* will free via device release */
3679 put_device(&dev
->dev
);
3681 kfree((char *)dev
- dev
->padded
);
3685 /* Synchronize with packet receive processing. */
3686 void synchronize_net(void)
3693 * unregister_netdevice - remove device from the kernel
3696 * This function shuts down a device interface and removes it
3697 * from the kernel tables. On success 0 is returned, on a failure
3698 * a negative errno code is returned.
3700 * Callers must hold the rtnl semaphore. You may want
3701 * unregister_netdev() instead of this.
3704 void unregister_netdevice(struct net_device
*dev
)
3706 BUG_ON(dev_boot_phase
);
3709 /* Some devices call without registering for initialization unwind. */
3710 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3711 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3712 "was registered\n", dev
->name
, dev
);
3718 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3720 /* If device is running, close it first. */
3721 if (dev
->flags
& IFF_UP
)
3724 /* And unlink it from device chain. */
3725 write_lock_bh(&dev_base_lock
);
3726 list_del(&dev
->dev_list
);
3727 hlist_del(&dev
->name_hlist
);
3728 hlist_del(&dev
->index_hlist
);
3729 write_unlock_bh(&dev_base_lock
);
3731 dev
->reg_state
= NETREG_UNREGISTERING
;
3735 /* Shutdown queueing discipline. */
3739 /* Notify protocols, that we are about to destroy
3740 this device. They should clean all the things.
3742 raw_notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3745 * Flush the unicast and multicast chains
3747 dev_addr_discard(dev
);
3752 /* Notifier chain MUST detach us from master device. */
3753 BUG_TRAP(!dev
->master
);
3755 /* Remove entries from sysfs */
3756 netdev_unregister_sysfs(dev
);
3758 /* Finish processing unregister after unlock */
3767 * unregister_netdev - remove device from the kernel
3770 * This function shuts down a device interface and removes it
3771 * from the kernel tables. On success 0 is returned, on a failure
3772 * a negative errno code is returned.
3774 * This is just a wrapper for unregister_netdevice that takes
3775 * the rtnl semaphore. In general you want to use this and not
3776 * unregister_netdevice.
3778 void unregister_netdev(struct net_device
*dev
)
3781 unregister_netdevice(dev
);
3785 EXPORT_SYMBOL(unregister_netdev
);
3787 static int dev_cpu_callback(struct notifier_block
*nfb
,
3788 unsigned long action
,
3791 struct sk_buff
**list_skb
;
3792 struct net_device
**list_net
;
3793 struct sk_buff
*skb
;
3794 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3795 struct softnet_data
*sd
, *oldsd
;
3797 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
3800 local_irq_disable();
3801 cpu
= smp_processor_id();
3802 sd
= &per_cpu(softnet_data
, cpu
);
3803 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3805 /* Find end of our completion_queue. */
3806 list_skb
= &sd
->completion_queue
;
3808 list_skb
= &(*list_skb
)->next
;
3809 /* Append completion queue from offline CPU. */
3810 *list_skb
= oldsd
->completion_queue
;
3811 oldsd
->completion_queue
= NULL
;
3813 /* Find end of our output_queue. */
3814 list_net
= &sd
->output_queue
;
3816 list_net
= &(*list_net
)->next_sched
;
3817 /* Append output queue from offline CPU. */
3818 *list_net
= oldsd
->output_queue
;
3819 oldsd
->output_queue
= NULL
;
3821 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3824 /* Process offline CPU's input_pkt_queue */
3825 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3831 #ifdef CONFIG_NET_DMA
3833 * net_dma_rebalance -
3834 * This is called when the number of channels allocated to the net_dma_client
3835 * changes. The net_dma_client tries to have one DMA channel per CPU.
3838 static void net_dma_rebalance(struct net_dma
*net_dma
)
3840 unsigned int cpu
, i
, n
, chan_idx
;
3841 struct dma_chan
*chan
;
3843 if (cpus_empty(net_dma
->channel_mask
)) {
3844 for_each_online_cpu(cpu
)
3845 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
3850 cpu
= first_cpu(cpu_online_map
);
3852 for_each_cpu_mask(chan_idx
, net_dma
->channel_mask
) {
3853 chan
= net_dma
->channels
[chan_idx
];
3855 n
= ((num_online_cpus() / cpus_weight(net_dma
->channel_mask
))
3856 + (i
< (num_online_cpus() %
3857 cpus_weight(net_dma
->channel_mask
)) ? 1 : 0));
3860 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
3861 cpu
= next_cpu(cpu
, cpu_online_map
);
3869 * netdev_dma_event - event callback for the net_dma_client
3870 * @client: should always be net_dma_client
3871 * @chan: DMA channel for the event
3872 * @event: event type
3874 static enum dma_state_client
3875 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
3876 enum dma_state state
)
3878 int i
, found
= 0, pos
= -1;
3879 struct net_dma
*net_dma
=
3880 container_of(client
, struct net_dma
, client
);
3881 enum dma_state_client ack
= DMA_DUP
; /* default: take no action */
3883 spin_lock(&net_dma
->lock
);
3885 case DMA_RESOURCE_AVAILABLE
:
3886 for (i
= 0; i
< NR_CPUS
; i
++)
3887 if (net_dma
->channels
[i
] == chan
) {
3890 } else if (net_dma
->channels
[i
] == NULL
&& pos
< 0)
3893 if (!found
&& pos
>= 0) {
3895 net_dma
->channels
[pos
] = chan
;
3896 cpu_set(pos
, net_dma
->channel_mask
);
3897 net_dma_rebalance(net_dma
);
3900 case DMA_RESOURCE_REMOVED
:
3901 for (i
= 0; i
< NR_CPUS
; i
++)
3902 if (net_dma
->channels
[i
] == chan
) {
3910 cpu_clear(pos
, net_dma
->channel_mask
);
3911 net_dma
->channels
[i
] = NULL
;
3912 net_dma_rebalance(net_dma
);
3918 spin_unlock(&net_dma
->lock
);
3924 * netdev_dma_regiser - register the networking subsystem as a DMA client
3926 static int __init
netdev_dma_register(void)
3928 spin_lock_init(&net_dma
.lock
);
3929 dma_cap_set(DMA_MEMCPY
, net_dma
.client
.cap_mask
);
3930 dma_async_client_register(&net_dma
.client
);
3931 dma_async_client_chan_request(&net_dma
.client
);
3936 static int __init
netdev_dma_register(void) { return -ENODEV
; }
3937 #endif /* CONFIG_NET_DMA */
3940 * Initialize the DEV module. At boot time this walks the device list and
3941 * unhooks any devices that fail to initialise (normally hardware not
3942 * present) and leaves us with a valid list of present and active devices.
3947 * This is called single threaded during boot, so no need
3948 * to take the rtnl semaphore.
3950 static int __init
net_dev_init(void)
3952 int i
, rc
= -ENOMEM
;
3954 BUG_ON(!dev_boot_phase
);
3956 if (dev_proc_init())
3959 if (netdev_sysfs_init())
3962 INIT_LIST_HEAD(&ptype_all
);
3963 for (i
= 0; i
< 16; i
++)
3964 INIT_LIST_HEAD(&ptype_base
[i
]);
3966 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
3967 INIT_HLIST_HEAD(&dev_name_head
[i
]);
3969 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
3970 INIT_HLIST_HEAD(&dev_index_head
[i
]);
3973 * Initialise the packet receive queues.
3976 for_each_possible_cpu(i
) {
3977 struct softnet_data
*queue
;
3979 queue
= &per_cpu(softnet_data
, i
);
3980 skb_queue_head_init(&queue
->input_pkt_queue
);
3981 queue
->completion_queue
= NULL
;
3982 INIT_LIST_HEAD(&queue
->poll_list
);
3983 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
3984 queue
->backlog_dev
.weight
= weight_p
;
3985 queue
->backlog_dev
.poll
= process_backlog
;
3986 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
3989 netdev_dma_register();
3993 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
3994 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
3996 hotcpu_notifier(dev_cpu_callback
, 0);
4004 subsys_initcall(net_dev_init
);
4006 EXPORT_SYMBOL(__dev_get_by_index
);
4007 EXPORT_SYMBOL(__dev_get_by_name
);
4008 EXPORT_SYMBOL(__dev_remove_pack
);
4009 EXPORT_SYMBOL(dev_valid_name
);
4010 EXPORT_SYMBOL(dev_add_pack
);
4011 EXPORT_SYMBOL(dev_alloc_name
);
4012 EXPORT_SYMBOL(dev_close
);
4013 EXPORT_SYMBOL(dev_get_by_flags
);
4014 EXPORT_SYMBOL(dev_get_by_index
);
4015 EXPORT_SYMBOL(dev_get_by_name
);
4016 EXPORT_SYMBOL(dev_open
);
4017 EXPORT_SYMBOL(dev_queue_xmit
);
4018 EXPORT_SYMBOL(dev_remove_pack
);
4019 EXPORT_SYMBOL(dev_set_allmulti
);
4020 EXPORT_SYMBOL(dev_set_promiscuity
);
4021 EXPORT_SYMBOL(dev_change_flags
);
4022 EXPORT_SYMBOL(dev_set_mtu
);
4023 EXPORT_SYMBOL(dev_set_mac_address
);
4024 EXPORT_SYMBOL(free_netdev
);
4025 EXPORT_SYMBOL(netdev_boot_setup_check
);
4026 EXPORT_SYMBOL(netdev_set_master
);
4027 EXPORT_SYMBOL(netdev_state_change
);
4028 EXPORT_SYMBOL(netif_receive_skb
);
4029 EXPORT_SYMBOL(netif_rx
);
4030 EXPORT_SYMBOL(register_gifconf
);
4031 EXPORT_SYMBOL(register_netdevice
);
4032 EXPORT_SYMBOL(register_netdevice_notifier
);
4033 EXPORT_SYMBOL(skb_checksum_help
);
4034 EXPORT_SYMBOL(synchronize_net
);
4035 EXPORT_SYMBOL(unregister_netdevice
);
4036 EXPORT_SYMBOL(unregister_netdevice_notifier
);
4037 EXPORT_SYMBOL(net_enable_timestamp
);
4038 EXPORT_SYMBOL(net_disable_timestamp
);
4039 EXPORT_SYMBOL(dev_get_flags
);
4041 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4042 EXPORT_SYMBOL(br_handle_frame_hook
);
4043 EXPORT_SYMBOL(br_fdb_get_hook
);
4044 EXPORT_SYMBOL(br_fdb_put_hook
);
4048 EXPORT_SYMBOL(dev_load
);
4051 EXPORT_PER_CPU_SYMBOL(softnet_data
);