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
);
843 write_lock_bh(&dev_base_lock
);
844 hlist_del(&dev
->name_hlist
);
845 hlist_add_head(&dev
->name_hlist
, dev_name_hash(dev
->name
));
846 write_unlock_bh(&dev_base_lock
);
848 raw_notifier_call_chain(&netdev_chain
, NETDEV_CHANGENAME
, dev
);
854 * netdev_features_change - device changes features
855 * @dev: device to cause notification
857 * Called to indicate a device has changed features.
859 void netdev_features_change(struct net_device
*dev
)
861 raw_notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
863 EXPORT_SYMBOL(netdev_features_change
);
866 * netdev_state_change - device changes state
867 * @dev: device to cause notification
869 * Called to indicate a device has changed state. This function calls
870 * the notifier chains for netdev_chain and sends a NEWLINK message
871 * to the routing socket.
873 void netdev_state_change(struct net_device
*dev
)
875 if (dev
->flags
& IFF_UP
) {
876 raw_notifier_call_chain(&netdev_chain
,
878 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
883 * dev_load - load a network module
884 * @name: name of interface
886 * If a network interface is not present and the process has suitable
887 * privileges this function loads the module. If module loading is not
888 * available in this kernel then it becomes a nop.
891 void dev_load(const char *name
)
893 struct net_device
*dev
;
895 read_lock(&dev_base_lock
);
896 dev
= __dev_get_by_name(name
);
897 read_unlock(&dev_base_lock
);
899 if (!dev
&& capable(CAP_SYS_MODULE
))
900 request_module("%s", name
);
903 static int default_rebuild_header(struct sk_buff
*skb
)
905 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
906 skb
->dev
? skb
->dev
->name
: "NULL!!!");
912 * dev_open - prepare an interface for use.
913 * @dev: device to open
915 * Takes a device from down to up state. The device's private open
916 * function is invoked and then the multicast lists are loaded. Finally
917 * the device is moved into the up state and a %NETDEV_UP message is
918 * sent to the netdev notifier chain.
920 * Calling this function on an active interface is a nop. On a failure
921 * a negative errno code is returned.
923 int dev_open(struct net_device
*dev
)
931 if (dev
->flags
& IFF_UP
)
935 * Is it even present?
937 if (!netif_device_present(dev
))
941 * Call device private open method
943 set_bit(__LINK_STATE_START
, &dev
->state
);
945 ret
= dev
->open(dev
);
947 clear_bit(__LINK_STATE_START
, &dev
->state
);
951 * If it went open OK then:
958 dev
->flags
|= IFF_UP
;
961 * Initialize multicasting status
963 dev_set_rx_mode(dev
);
966 * Wakeup transmit queue engine
971 * ... and announce new interface.
973 raw_notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
979 * dev_close - shutdown an interface.
980 * @dev: device to shutdown
982 * This function moves an active device into down state. A
983 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
984 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
987 int dev_close(struct net_device
*dev
)
989 if (!(dev
->flags
& IFF_UP
))
993 * Tell people we are going down, so that they can
994 * prepare to death, when device is still operating.
996 raw_notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
1000 clear_bit(__LINK_STATE_START
, &dev
->state
);
1002 /* Synchronize to scheduled poll. We cannot touch poll list,
1003 * it can be even on different cpu. So just clear netif_running(),
1004 * and wait when poll really will happen. Actually, the best place
1005 * for this is inside dev->stop() after device stopped its irq
1006 * engine, but this requires more changes in devices. */
1008 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1009 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
1015 * Call the device specific close. This cannot fail.
1016 * Only if device is UP
1018 * We allow it to be called even after a DETACH hot-plug
1025 * Device is now down.
1028 dev
->flags
&= ~IFF_UP
;
1031 * Tell people we are down
1033 raw_notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
1040 * Device change register/unregister. These are not inline or static
1041 * as we export them to the world.
1045 * register_netdevice_notifier - register a network notifier block
1048 * Register a notifier to be called when network device events occur.
1049 * The notifier passed is linked into the kernel structures and must
1050 * not be reused until it has been unregistered. A negative errno code
1051 * is returned on a failure.
1053 * When registered all registration and up events are replayed
1054 * to the new notifier to allow device to have a race free
1055 * view of the network device list.
1058 int register_netdevice_notifier(struct notifier_block
*nb
)
1060 struct net_device
*dev
;
1064 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1066 for_each_netdev(dev
) {
1067 nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1069 if (dev
->flags
& IFF_UP
)
1070 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1078 * unregister_netdevice_notifier - unregister a network notifier block
1081 * Unregister a notifier previously registered by
1082 * register_netdevice_notifier(). The notifier is unlinked into the
1083 * kernel structures and may then be reused. A negative errno code
1084 * is returned on a failure.
1087 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1092 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1098 * call_netdevice_notifiers - call all network notifier blocks
1099 * @val: value passed unmodified to notifier function
1100 * @v: pointer passed unmodified to notifier function
1102 * Call all network notifier blocks. Parameters and return value
1103 * are as for raw_notifier_call_chain().
1106 int call_netdevice_notifiers(unsigned long val
, void *v
)
1108 return raw_notifier_call_chain(&netdev_chain
, val
, v
);
1111 /* When > 0 there are consumers of rx skb time stamps */
1112 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1114 void net_enable_timestamp(void)
1116 atomic_inc(&netstamp_needed
);
1119 void net_disable_timestamp(void)
1121 atomic_dec(&netstamp_needed
);
1124 static inline void net_timestamp(struct sk_buff
*skb
)
1126 if (atomic_read(&netstamp_needed
))
1127 __net_timestamp(skb
);
1129 skb
->tstamp
.tv64
= 0;
1133 * Support routine. Sends outgoing frames to any network
1134 * taps currently in use.
1137 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1139 struct packet_type
*ptype
;
1144 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1145 /* Never send packets back to the socket
1146 * they originated from - MvS (miquels@drinkel.ow.org)
1148 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1149 (ptype
->af_packet_priv
== NULL
||
1150 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1151 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1155 /* skb->nh should be correctly
1156 set by sender, so that the second statement is
1157 just protection against buggy protocols.
1159 skb_reset_mac_header(skb2
);
1161 if (skb_network_header(skb2
) < skb2
->data
||
1162 skb2
->network_header
> skb2
->tail
) {
1163 if (net_ratelimit())
1164 printk(KERN_CRIT
"protocol %04x is "
1166 skb2
->protocol
, dev
->name
);
1167 skb_reset_network_header(skb2
);
1170 skb2
->transport_header
= skb2
->network_header
;
1171 skb2
->pkt_type
= PACKET_OUTGOING
;
1172 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1179 void __netif_schedule(struct net_device
*dev
)
1181 if (!test_and_set_bit(__LINK_STATE_SCHED
, &dev
->state
)) {
1182 unsigned long flags
;
1183 struct softnet_data
*sd
;
1185 local_irq_save(flags
);
1186 sd
= &__get_cpu_var(softnet_data
);
1187 dev
->next_sched
= sd
->output_queue
;
1188 sd
->output_queue
= dev
;
1189 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1190 local_irq_restore(flags
);
1193 EXPORT_SYMBOL(__netif_schedule
);
1195 void __netif_rx_schedule(struct net_device
*dev
)
1197 unsigned long flags
;
1199 local_irq_save(flags
);
1201 list_add_tail(&dev
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
1203 dev
->quota
+= dev
->weight
;
1205 dev
->quota
= dev
->weight
;
1206 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1207 local_irq_restore(flags
);
1209 EXPORT_SYMBOL(__netif_rx_schedule
);
1211 void dev_kfree_skb_any(struct sk_buff
*skb
)
1213 if (in_irq() || irqs_disabled())
1214 dev_kfree_skb_irq(skb
);
1218 EXPORT_SYMBOL(dev_kfree_skb_any
);
1222 void netif_device_detach(struct net_device
*dev
)
1224 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1225 netif_running(dev
)) {
1226 netif_stop_queue(dev
);
1229 EXPORT_SYMBOL(netif_device_detach
);
1231 void netif_device_attach(struct net_device
*dev
)
1233 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1234 netif_running(dev
)) {
1235 netif_wake_queue(dev
);
1236 __netdev_watchdog_up(dev
);
1239 EXPORT_SYMBOL(netif_device_attach
);
1243 * Invalidate hardware checksum when packet is to be mangled, and
1244 * complete checksum manually on outgoing path.
1246 int skb_checksum_help(struct sk_buff
*skb
)
1249 int ret
= 0, offset
;
1251 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1252 goto out_set_summed
;
1254 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1255 /* Let GSO fix up the checksum. */
1256 goto out_set_summed
;
1259 if (skb_cloned(skb
)) {
1260 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1265 offset
= skb
->csum_start
- skb_headroom(skb
);
1266 BUG_ON(offset
> (int)skb
->len
);
1267 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1269 offset
= skb_headlen(skb
) - offset
;
1270 BUG_ON(offset
<= 0);
1271 BUG_ON(skb
->csum_offset
+ 2 > offset
);
1273 *(__sum16
*)(skb
->head
+ skb
->csum_start
+ skb
->csum_offset
) =
1276 skb
->ip_summed
= CHECKSUM_NONE
;
1282 * skb_gso_segment - Perform segmentation on skb.
1283 * @skb: buffer to segment
1284 * @features: features for the output path (see dev->features)
1286 * This function segments the given skb and returns a list of segments.
1288 * It may return NULL if the skb requires no segmentation. This is
1289 * only possible when GSO is used for verifying header integrity.
1291 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1293 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1294 struct packet_type
*ptype
;
1295 __be16 type
= skb
->protocol
;
1298 BUG_ON(skb_shinfo(skb
)->frag_list
);
1300 skb_reset_mac_header(skb
);
1301 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1302 __skb_pull(skb
, skb
->mac_len
);
1304 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1305 if (skb_header_cloned(skb
) &&
1306 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1307 return ERR_PTR(err
);
1311 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
) & 15], list
) {
1312 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1313 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1314 err
= ptype
->gso_send_check(skb
);
1315 segs
= ERR_PTR(err
);
1316 if (err
|| skb_gso_ok(skb
, features
))
1318 __skb_push(skb
, (skb
->data
-
1319 skb_network_header(skb
)));
1321 segs
= ptype
->gso_segment(skb
, features
);
1327 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1332 EXPORT_SYMBOL(skb_gso_segment
);
1334 /* Take action when hardware reception checksum errors are detected. */
1336 void netdev_rx_csum_fault(struct net_device
*dev
)
1338 if (net_ratelimit()) {
1339 printk(KERN_ERR
"%s: hw csum failure.\n",
1340 dev
? dev
->name
: "<unknown>");
1344 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1347 /* Actually, we should eliminate this check as soon as we know, that:
1348 * 1. IOMMU is present and allows to map all the memory.
1349 * 2. No high memory really exists on this machine.
1352 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1354 #ifdef CONFIG_HIGHMEM
1357 if (dev
->features
& NETIF_F_HIGHDMA
)
1360 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1361 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1369 void (*destructor
)(struct sk_buff
*skb
);
1372 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1374 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1376 struct dev_gso_cb
*cb
;
1379 struct sk_buff
*nskb
= skb
->next
;
1381 skb
->next
= nskb
->next
;
1384 } while (skb
->next
);
1386 cb
= DEV_GSO_CB(skb
);
1388 cb
->destructor(skb
);
1392 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1393 * @skb: buffer to segment
1395 * This function segments the given skb and stores the list of segments
1398 static int dev_gso_segment(struct sk_buff
*skb
)
1400 struct net_device
*dev
= skb
->dev
;
1401 struct sk_buff
*segs
;
1402 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1405 segs
= skb_gso_segment(skb
, features
);
1407 /* Verifying header integrity only. */
1411 if (unlikely(IS_ERR(segs
)))
1412 return PTR_ERR(segs
);
1415 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1416 skb
->destructor
= dev_gso_skb_destructor
;
1421 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1423 if (likely(!skb
->next
)) {
1424 if (!list_empty(&ptype_all
))
1425 dev_queue_xmit_nit(skb
, dev
);
1427 if (netif_needs_gso(dev
, skb
)) {
1428 if (unlikely(dev_gso_segment(skb
)))
1434 return dev
->hard_start_xmit(skb
, dev
);
1439 struct sk_buff
*nskb
= skb
->next
;
1442 skb
->next
= nskb
->next
;
1444 rc
= dev
->hard_start_xmit(nskb
, dev
);
1446 nskb
->next
= skb
->next
;
1450 if (unlikely((netif_queue_stopped(dev
) ||
1451 netif_subqueue_stopped(dev
, skb
->queue_mapping
)) &&
1453 return NETDEV_TX_BUSY
;
1454 } while (skb
->next
);
1456 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1463 #define HARD_TX_LOCK(dev, cpu) { \
1464 if ((dev->features & NETIF_F_LLTX) == 0) { \
1465 netif_tx_lock(dev); \
1469 #define HARD_TX_UNLOCK(dev) { \
1470 if ((dev->features & NETIF_F_LLTX) == 0) { \
1471 netif_tx_unlock(dev); \
1476 * dev_queue_xmit - transmit a buffer
1477 * @skb: buffer to transmit
1479 * Queue a buffer for transmission to a network device. The caller must
1480 * have set the device and priority and built the buffer before calling
1481 * this function. The function can be called from an interrupt.
1483 * A negative errno code is returned on a failure. A success does not
1484 * guarantee the frame will be transmitted as it may be dropped due
1485 * to congestion or traffic shaping.
1487 * -----------------------------------------------------------------------------------
1488 * I notice this method can also return errors from the queue disciplines,
1489 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1492 * Regardless of the return value, the skb is consumed, so it is currently
1493 * difficult to retry a send to this method. (You can bump the ref count
1494 * before sending to hold a reference for retry if you are careful.)
1496 * When calling this method, interrupts MUST be enabled. This is because
1497 * the BH enable code must have IRQs enabled so that it will not deadlock.
1501 int dev_queue_xmit(struct sk_buff
*skb
)
1503 struct net_device
*dev
= skb
->dev
;
1507 /* GSO will handle the following emulations directly. */
1508 if (netif_needs_gso(dev
, skb
))
1511 if (skb_shinfo(skb
)->frag_list
&&
1512 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1513 __skb_linearize(skb
))
1516 /* Fragmented skb is linearized if device does not support SG,
1517 * or if at least one of fragments is in highmem and device
1518 * does not support DMA from it.
1520 if (skb_shinfo(skb
)->nr_frags
&&
1521 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1522 __skb_linearize(skb
))
1525 /* If packet is not checksummed and device does not support
1526 * checksumming for this protocol, complete checksumming here.
1528 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1529 skb_set_transport_header(skb
, skb
->csum_start
-
1532 if (!(dev
->features
& NETIF_F_GEN_CSUM
) &&
1533 !((dev
->features
& NETIF_F_IP_CSUM
) &&
1534 skb
->protocol
== htons(ETH_P_IP
)) &&
1535 !((dev
->features
& NETIF_F_IPV6_CSUM
) &&
1536 skb
->protocol
== htons(ETH_P_IPV6
)))
1537 if (skb_checksum_help(skb
))
1542 spin_lock_prefetch(&dev
->queue_lock
);
1544 /* Disable soft irqs for various locks below. Also
1545 * stops preemption for RCU.
1549 /* Updates of qdisc are serialized by queue_lock.
1550 * The struct Qdisc which is pointed to by qdisc is now a
1551 * rcu structure - it may be accessed without acquiring
1552 * a lock (but the structure may be stale.) The freeing of the
1553 * qdisc will be deferred until it's known that there are no
1554 * more references to it.
1556 * If the qdisc has an enqueue function, we still need to
1557 * hold the queue_lock before calling it, since queue_lock
1558 * also serializes access to the device queue.
1561 q
= rcu_dereference(dev
->qdisc
);
1562 #ifdef CONFIG_NET_CLS_ACT
1563 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1566 /* Grab device queue */
1567 spin_lock(&dev
->queue_lock
);
1570 /* reset queue_mapping to zero */
1571 skb
->queue_mapping
= 0;
1572 rc
= q
->enqueue(skb
, q
);
1574 spin_unlock(&dev
->queue_lock
);
1576 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1579 spin_unlock(&dev
->queue_lock
);
1582 /* The device has no queue. Common case for software devices:
1583 loopback, all the sorts of tunnels...
1585 Really, it is unlikely that netif_tx_lock protection is necessary
1586 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1588 However, it is possible, that they rely on protection
1591 Check this and shot the lock. It is not prone from deadlocks.
1592 Either shot noqueue qdisc, it is even simpler 8)
1594 if (dev
->flags
& IFF_UP
) {
1595 int cpu
= smp_processor_id(); /* ok because BHs are off */
1597 if (dev
->xmit_lock_owner
!= cpu
) {
1599 HARD_TX_LOCK(dev
, cpu
);
1601 if (!netif_queue_stopped(dev
) &&
1602 !netif_subqueue_stopped(dev
, skb
->queue_mapping
)) {
1604 if (!dev_hard_start_xmit(skb
, dev
)) {
1605 HARD_TX_UNLOCK(dev
);
1609 HARD_TX_UNLOCK(dev
);
1610 if (net_ratelimit())
1611 printk(KERN_CRIT
"Virtual device %s asks to "
1612 "queue packet!\n", dev
->name
);
1614 /* Recursion is detected! It is possible,
1616 if (net_ratelimit())
1617 printk(KERN_CRIT
"Dead loop on virtual device "
1618 "%s, fix it urgently!\n", dev
->name
);
1623 rcu_read_unlock_bh();
1629 rcu_read_unlock_bh();
1634 /*=======================================================================
1636 =======================================================================*/
1638 int netdev_max_backlog __read_mostly
= 1000;
1639 int netdev_budget __read_mostly
= 300;
1640 int weight_p __read_mostly
= 64; /* old backlog weight */
1642 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1646 * netif_rx - post buffer to the network code
1647 * @skb: buffer to post
1649 * This function receives a packet from a device driver and queues it for
1650 * the upper (protocol) levels to process. It always succeeds. The buffer
1651 * may be dropped during processing for congestion control or by the
1655 * NET_RX_SUCCESS (no congestion)
1656 * NET_RX_CN_LOW (low congestion)
1657 * NET_RX_CN_MOD (moderate congestion)
1658 * NET_RX_CN_HIGH (high congestion)
1659 * NET_RX_DROP (packet was dropped)
1663 int netif_rx(struct sk_buff
*skb
)
1665 struct softnet_data
*queue
;
1666 unsigned long flags
;
1668 /* if netpoll wants it, pretend we never saw it */
1669 if (netpoll_rx(skb
))
1672 if (!skb
->tstamp
.tv64
)
1676 * The code is rearranged so that the path is the most
1677 * short when CPU is congested, but is still operating.
1679 local_irq_save(flags
);
1680 queue
= &__get_cpu_var(softnet_data
);
1682 __get_cpu_var(netdev_rx_stat
).total
++;
1683 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1684 if (queue
->input_pkt_queue
.qlen
) {
1687 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1688 local_irq_restore(flags
);
1689 return NET_RX_SUCCESS
;
1692 netif_rx_schedule(&queue
->backlog_dev
);
1696 __get_cpu_var(netdev_rx_stat
).dropped
++;
1697 local_irq_restore(flags
);
1703 int netif_rx_ni(struct sk_buff
*skb
)
1708 err
= netif_rx(skb
);
1709 if (local_softirq_pending())
1716 EXPORT_SYMBOL(netif_rx_ni
);
1718 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1720 struct net_device
*dev
= skb
->dev
;
1723 if (skb_bond_should_drop(skb
)) {
1727 skb
->dev
= dev
->master
;
1733 static void net_tx_action(struct softirq_action
*h
)
1735 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1737 if (sd
->completion_queue
) {
1738 struct sk_buff
*clist
;
1740 local_irq_disable();
1741 clist
= sd
->completion_queue
;
1742 sd
->completion_queue
= NULL
;
1746 struct sk_buff
*skb
= clist
;
1747 clist
= clist
->next
;
1749 BUG_TRAP(!atomic_read(&skb
->users
));
1754 if (sd
->output_queue
) {
1755 struct net_device
*head
;
1757 local_irq_disable();
1758 head
= sd
->output_queue
;
1759 sd
->output_queue
= NULL
;
1763 struct net_device
*dev
= head
;
1764 head
= head
->next_sched
;
1766 smp_mb__before_clear_bit();
1767 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1769 if (spin_trylock(&dev
->queue_lock
)) {
1771 spin_unlock(&dev
->queue_lock
);
1773 netif_schedule(dev
);
1779 static inline int deliver_skb(struct sk_buff
*skb
,
1780 struct packet_type
*pt_prev
,
1781 struct net_device
*orig_dev
)
1783 atomic_inc(&skb
->users
);
1784 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1787 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1788 /* These hooks defined here for ATM */
1790 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1791 unsigned char *addr
);
1792 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
1795 * If bridge module is loaded call bridging hook.
1796 * returns NULL if packet was consumed.
1798 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
1799 struct sk_buff
*skb
) __read_mostly
;
1800 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
1801 struct packet_type
**pt_prev
, int *ret
,
1802 struct net_device
*orig_dev
)
1804 struct net_bridge_port
*port
;
1806 if (skb
->pkt_type
== PACKET_LOOPBACK
||
1807 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
1811 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1815 return br_handle_frame_hook(port
, skb
);
1818 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1821 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1822 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
1823 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
1825 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
1826 struct packet_type
**pt_prev
,
1828 struct net_device
*orig_dev
)
1830 if (skb
->dev
->macvlan_port
== NULL
)
1834 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1837 return macvlan_handle_frame_hook(skb
);
1840 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1843 #ifdef CONFIG_NET_CLS_ACT
1844 /* TODO: Maybe we should just force sch_ingress to be compiled in
1845 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1846 * a compare and 2 stores extra right now if we dont have it on
1847 * but have CONFIG_NET_CLS_ACT
1848 * NOTE: This doesnt stop any functionality; if you dont have
1849 * the ingress scheduler, you just cant add policies on ingress.
1852 static int ing_filter(struct sk_buff
*skb
)
1855 struct net_device
*dev
= skb
->dev
;
1856 int result
= TC_ACT_OK
;
1858 if (dev
->qdisc_ingress
) {
1859 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1860 if (MAX_RED_LOOP
< ttl
++) {
1861 printk(KERN_WARNING
"Redir loop detected Dropping packet (%d->%d)\n",
1862 skb
->iif
, skb
->dev
->ifindex
);
1866 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1868 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1870 spin_lock(&dev
->ingress_lock
);
1871 if ((q
= dev
->qdisc_ingress
) != NULL
)
1872 result
= q
->enqueue(skb
, q
);
1873 spin_unlock(&dev
->ingress_lock
);
1881 int netif_receive_skb(struct sk_buff
*skb
)
1883 struct packet_type
*ptype
, *pt_prev
;
1884 struct net_device
*orig_dev
;
1885 int ret
= NET_RX_DROP
;
1888 /* if we've gotten here through NAPI, check netpoll */
1889 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1892 if (!skb
->tstamp
.tv64
)
1896 skb
->iif
= skb
->dev
->ifindex
;
1898 orig_dev
= skb_bond(skb
);
1903 __get_cpu_var(netdev_rx_stat
).total
++;
1905 skb_reset_network_header(skb
);
1906 skb_reset_transport_header(skb
);
1907 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1913 #ifdef CONFIG_NET_CLS_ACT
1914 if (skb
->tc_verd
& TC_NCLS
) {
1915 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1920 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1921 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1923 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1928 #ifdef CONFIG_NET_CLS_ACT
1930 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1931 pt_prev
= NULL
; /* noone else should process this after*/
1933 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1936 ret
= ing_filter(skb
);
1938 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
1947 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
1950 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
1954 type
= skb
->protocol
;
1955 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
1956 if (ptype
->type
== type
&&
1957 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
1959 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1965 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1968 /* Jamal, now you will not able to escape explaining
1969 * me how you were going to use this. :-)
1979 static int process_backlog(struct net_device
*backlog_dev
, int *budget
)
1982 int quota
= min(backlog_dev
->quota
, *budget
);
1983 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
1984 unsigned long start_time
= jiffies
;
1986 backlog_dev
->weight
= weight_p
;
1988 struct sk_buff
*skb
;
1989 struct net_device
*dev
;
1991 local_irq_disable();
1992 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
1999 netif_receive_skb(skb
);
2005 if (work
>= quota
|| jiffies
- start_time
> 1)
2010 backlog_dev
->quota
-= work
;
2015 backlog_dev
->quota
-= work
;
2018 list_del(&backlog_dev
->poll_list
);
2019 smp_mb__before_clear_bit();
2020 netif_poll_enable(backlog_dev
);
2026 static void net_rx_action(struct softirq_action
*h
)
2028 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2029 unsigned long start_time
= jiffies
;
2030 int budget
= netdev_budget
;
2033 local_irq_disable();
2035 while (!list_empty(&queue
->poll_list
)) {
2036 struct net_device
*dev
;
2038 if (budget
<= 0 || jiffies
- start_time
> 1)
2043 dev
= list_entry(queue
->poll_list
.next
,
2044 struct net_device
, poll_list
);
2045 have
= netpoll_poll_lock(dev
);
2047 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
2048 netpoll_poll_unlock(have
);
2049 local_irq_disable();
2050 list_move_tail(&dev
->poll_list
, &queue
->poll_list
);
2052 dev
->quota
+= dev
->weight
;
2054 dev
->quota
= dev
->weight
;
2056 netpoll_poll_unlock(have
);
2058 local_irq_disable();
2063 #ifdef CONFIG_NET_DMA
2065 * There may not be any more sk_buffs coming right now, so push
2066 * any pending DMA copies to hardware
2068 if (!cpus_empty(net_dma
.channel_mask
)) {
2070 for_each_cpu_mask(chan_idx
, net_dma
.channel_mask
) {
2071 struct dma_chan
*chan
= net_dma
.channels
[chan_idx
];
2073 dma_async_memcpy_issue_pending(chan
);
2080 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2081 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2085 static gifconf_func_t
* gifconf_list
[NPROTO
];
2088 * register_gifconf - register a SIOCGIF handler
2089 * @family: Address family
2090 * @gifconf: Function handler
2092 * Register protocol dependent address dumping routines. The handler
2093 * that is passed must not be freed or reused until it has been replaced
2094 * by another handler.
2096 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2098 if (family
>= NPROTO
)
2100 gifconf_list
[family
] = gifconf
;
2106 * Map an interface index to its name (SIOCGIFNAME)
2110 * We need this ioctl for efficient implementation of the
2111 * if_indextoname() function required by the IPv6 API. Without
2112 * it, we would have to search all the interfaces to find a
2116 static int dev_ifname(struct ifreq __user
*arg
)
2118 struct net_device
*dev
;
2122 * Fetch the caller's info block.
2125 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2128 read_lock(&dev_base_lock
);
2129 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
2131 read_unlock(&dev_base_lock
);
2135 strcpy(ifr
.ifr_name
, dev
->name
);
2136 read_unlock(&dev_base_lock
);
2138 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2144 * Perform a SIOCGIFCONF call. This structure will change
2145 * size eventually, and there is nothing I can do about it.
2146 * Thus we will need a 'compatibility mode'.
2149 static int dev_ifconf(char __user
*arg
)
2152 struct net_device
*dev
;
2159 * Fetch the caller's info block.
2162 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2169 * Loop over the interfaces, and write an info block for each.
2173 for_each_netdev(dev
) {
2174 for (i
= 0; i
< NPROTO
; i
++) {
2175 if (gifconf_list
[i
]) {
2178 done
= gifconf_list
[i
](dev
, NULL
, 0);
2180 done
= gifconf_list
[i
](dev
, pos
+ total
,
2190 * All done. Write the updated control block back to the caller.
2192 ifc
.ifc_len
= total
;
2195 * Both BSD and Solaris return 0 here, so we do too.
2197 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2200 #ifdef CONFIG_PROC_FS
2202 * This is invoked by the /proc filesystem handler to display a device
2205 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2208 struct net_device
*dev
;
2210 read_lock(&dev_base_lock
);
2212 return SEQ_START_TOKEN
;
2215 for_each_netdev(dev
)
2222 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2225 return v
== SEQ_START_TOKEN
?
2226 first_net_device() : next_net_device((struct net_device
*)v
);
2229 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2231 read_unlock(&dev_base_lock
);
2234 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2236 struct net_device_stats
*stats
= dev
->get_stats(dev
);
2238 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2239 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2240 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2242 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2243 stats
->rx_fifo_errors
,
2244 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2245 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2246 stats
->rx_compressed
, stats
->multicast
,
2247 stats
->tx_bytes
, stats
->tx_packets
,
2248 stats
->tx_errors
, stats
->tx_dropped
,
2249 stats
->tx_fifo_errors
, stats
->collisions
,
2250 stats
->tx_carrier_errors
+
2251 stats
->tx_aborted_errors
+
2252 stats
->tx_window_errors
+
2253 stats
->tx_heartbeat_errors
,
2254 stats
->tx_compressed
);
2258 * Called from the PROCfs module. This now uses the new arbitrary sized
2259 * /proc/net interface to create /proc/net/dev
2261 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2263 if (v
== SEQ_START_TOKEN
)
2264 seq_puts(seq
, "Inter-| Receive "
2266 " face |bytes packets errs drop fifo frame "
2267 "compressed multicast|bytes packets errs "
2268 "drop fifo colls carrier compressed\n");
2270 dev_seq_printf_stats(seq
, v
);
2274 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2276 struct netif_rx_stats
*rc
= NULL
;
2278 while (*pos
< NR_CPUS
)
2279 if (cpu_online(*pos
)) {
2280 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2287 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2289 return softnet_get_online(pos
);
2292 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2295 return softnet_get_online(pos
);
2298 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2302 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2304 struct netif_rx_stats
*s
= v
;
2306 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2307 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2308 0, 0, 0, 0, /* was fastroute */
2313 static const struct seq_operations dev_seq_ops
= {
2314 .start
= dev_seq_start
,
2315 .next
= dev_seq_next
,
2316 .stop
= dev_seq_stop
,
2317 .show
= dev_seq_show
,
2320 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2322 return seq_open(file
, &dev_seq_ops
);
2325 static const struct file_operations dev_seq_fops
= {
2326 .owner
= THIS_MODULE
,
2327 .open
= dev_seq_open
,
2329 .llseek
= seq_lseek
,
2330 .release
= seq_release
,
2333 static const struct seq_operations softnet_seq_ops
= {
2334 .start
= softnet_seq_start
,
2335 .next
= softnet_seq_next
,
2336 .stop
= softnet_seq_stop
,
2337 .show
= softnet_seq_show
,
2340 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2342 return seq_open(file
, &softnet_seq_ops
);
2345 static const struct file_operations softnet_seq_fops
= {
2346 .owner
= THIS_MODULE
,
2347 .open
= softnet_seq_open
,
2349 .llseek
= seq_lseek
,
2350 .release
= seq_release
,
2353 static void *ptype_get_idx(loff_t pos
)
2355 struct packet_type
*pt
= NULL
;
2359 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
2365 for (t
= 0; t
< 16; t
++) {
2366 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
2375 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2378 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
2381 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2383 struct packet_type
*pt
;
2384 struct list_head
*nxt
;
2388 if (v
== SEQ_START_TOKEN
)
2389 return ptype_get_idx(0);
2392 nxt
= pt
->list
.next
;
2393 if (pt
->type
== htons(ETH_P_ALL
)) {
2394 if (nxt
!= &ptype_all
)
2397 nxt
= ptype_base
[0].next
;
2399 hash
= ntohs(pt
->type
) & 15;
2401 while (nxt
== &ptype_base
[hash
]) {
2404 nxt
= ptype_base
[hash
].next
;
2407 return list_entry(nxt
, struct packet_type
, list
);
2410 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
2415 static void ptype_seq_decode(struct seq_file
*seq
, void *sym
)
2417 #ifdef CONFIG_KALLSYMS
2418 unsigned long offset
= 0, symsize
;
2419 const char *symname
;
2423 symname
= kallsyms_lookup((unsigned long)sym
, &symsize
, &offset
,
2430 modname
= delim
= "";
2431 seq_printf(seq
, "%s%s%s%s+0x%lx", delim
, modname
, delim
,
2437 seq_printf(seq
, "[%p]", sym
);
2440 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
2442 struct packet_type
*pt
= v
;
2444 if (v
== SEQ_START_TOKEN
)
2445 seq_puts(seq
, "Type Device Function\n");
2447 if (pt
->type
== htons(ETH_P_ALL
))
2448 seq_puts(seq
, "ALL ");
2450 seq_printf(seq
, "%04x", ntohs(pt
->type
));
2452 seq_printf(seq
, " %-8s ",
2453 pt
->dev
? pt
->dev
->name
: "");
2454 ptype_seq_decode(seq
, pt
->func
);
2455 seq_putc(seq
, '\n');
2461 static const struct seq_operations ptype_seq_ops
= {
2462 .start
= ptype_seq_start
,
2463 .next
= ptype_seq_next
,
2464 .stop
= ptype_seq_stop
,
2465 .show
= ptype_seq_show
,
2468 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
2470 return seq_open(file
, &ptype_seq_ops
);
2473 static const struct file_operations ptype_seq_fops
= {
2474 .owner
= THIS_MODULE
,
2475 .open
= ptype_seq_open
,
2477 .llseek
= seq_lseek
,
2478 .release
= seq_release
,
2482 static int __init
dev_proc_init(void)
2486 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2488 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2490 if (!proc_net_fops_create("ptype", S_IRUGO
, &ptype_seq_fops
))
2493 if (wext_proc_init())
2499 proc_net_remove("ptype");
2501 proc_net_remove("softnet_stat");
2503 proc_net_remove("dev");
2507 #define dev_proc_init() 0
2508 #endif /* CONFIG_PROC_FS */
2512 * netdev_set_master - set up master/slave pair
2513 * @slave: slave device
2514 * @master: new master device
2516 * Changes the master device of the slave. Pass %NULL to break the
2517 * bonding. The caller must hold the RTNL semaphore. On a failure
2518 * a negative errno code is returned. On success the reference counts
2519 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2520 * function returns zero.
2522 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2524 struct net_device
*old
= slave
->master
;
2534 slave
->master
= master
;
2542 slave
->flags
|= IFF_SLAVE
;
2544 slave
->flags
&= ~IFF_SLAVE
;
2546 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2550 static void __dev_set_promiscuity(struct net_device
*dev
, int inc
)
2552 unsigned short old_flags
= dev
->flags
;
2556 if ((dev
->promiscuity
+= inc
) == 0)
2557 dev
->flags
&= ~IFF_PROMISC
;
2559 dev
->flags
|= IFF_PROMISC
;
2560 if (dev
->flags
!= old_flags
) {
2561 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2562 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2564 audit_log(current
->audit_context
, GFP_ATOMIC
,
2565 AUDIT_ANOM_PROMISCUOUS
,
2566 "dev=%s prom=%d old_prom=%d auid=%u",
2567 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2568 (old_flags
& IFF_PROMISC
),
2569 audit_get_loginuid(current
->audit_context
));
2571 if (dev
->change_rx_flags
)
2572 dev
->change_rx_flags(dev
, IFF_PROMISC
);
2577 * dev_set_promiscuity - update promiscuity count on a device
2581 * Add or remove promiscuity from a device. While the count in the device
2582 * remains above zero the interface remains promiscuous. Once it hits zero
2583 * the device reverts back to normal filtering operation. A negative inc
2584 * value is used to drop promiscuity on the device.
2586 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2588 unsigned short old_flags
= dev
->flags
;
2590 __dev_set_promiscuity(dev
, inc
);
2591 if (dev
->flags
!= old_flags
)
2592 dev_set_rx_mode(dev
);
2596 * dev_set_allmulti - update allmulti count on a device
2600 * Add or remove reception of all multicast frames to a device. While the
2601 * count in the device remains above zero the interface remains listening
2602 * to all interfaces. Once it hits zero the device reverts back to normal
2603 * filtering operation. A negative @inc value is used to drop the counter
2604 * when releasing a resource needing all multicasts.
2607 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2609 unsigned short old_flags
= dev
->flags
;
2613 dev
->flags
|= IFF_ALLMULTI
;
2614 if ((dev
->allmulti
+= inc
) == 0)
2615 dev
->flags
&= ~IFF_ALLMULTI
;
2616 if (dev
->flags
^ old_flags
) {
2617 if (dev
->change_rx_flags
)
2618 dev
->change_rx_flags(dev
, IFF_ALLMULTI
);
2619 dev_set_rx_mode(dev
);
2624 * Upload unicast and multicast address lists to device and
2625 * configure RX filtering. When the device doesn't support unicast
2626 * filtering it is put in promiscous mode while unicast addresses
2629 void __dev_set_rx_mode(struct net_device
*dev
)
2631 /* dev_open will call this function so the list will stay sane. */
2632 if (!(dev
->flags
&IFF_UP
))
2635 if (!netif_device_present(dev
))
2638 if (dev
->set_rx_mode
)
2639 dev
->set_rx_mode(dev
);
2641 /* Unicast addresses changes may only happen under the rtnl,
2642 * therefore calling __dev_set_promiscuity here is safe.
2644 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
2645 __dev_set_promiscuity(dev
, 1);
2646 dev
->uc_promisc
= 1;
2647 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
2648 __dev_set_promiscuity(dev
, -1);
2649 dev
->uc_promisc
= 0;
2652 if (dev
->set_multicast_list
)
2653 dev
->set_multicast_list(dev
);
2657 void dev_set_rx_mode(struct net_device
*dev
)
2659 netif_tx_lock_bh(dev
);
2660 __dev_set_rx_mode(dev
);
2661 netif_tx_unlock_bh(dev
);
2664 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
2665 void *addr
, int alen
, int glbl
)
2667 struct dev_addr_list
*da
;
2669 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
2670 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2671 alen
== da
->da_addrlen
) {
2673 int old_glbl
= da
->da_gusers
;
2690 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
2691 void *addr
, int alen
, int glbl
)
2693 struct dev_addr_list
*da
;
2695 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
2696 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2697 da
->da_addrlen
== alen
) {
2699 int old_glbl
= da
->da_gusers
;
2709 da
= kmalloc(sizeof(*da
), GFP_ATOMIC
);
2712 memcpy(da
->da_addr
, addr
, alen
);
2713 da
->da_addrlen
= alen
;
2715 da
->da_gusers
= glbl
? 1 : 0;
2723 * dev_unicast_delete - Release secondary unicast address.
2725 * @addr: address to delete
2726 * @alen: length of @addr
2728 * Release reference to a secondary unicast address and remove it
2729 * from the device if the reference count drops to zero.
2731 * The caller must hold the rtnl_mutex.
2733 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
2739 netif_tx_lock_bh(dev
);
2740 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
2742 __dev_set_rx_mode(dev
);
2743 netif_tx_unlock_bh(dev
);
2746 EXPORT_SYMBOL(dev_unicast_delete
);
2749 * dev_unicast_add - add a secondary unicast address
2751 * @addr: address to delete
2752 * @alen: length of @addr
2754 * Add a secondary unicast address to the device or increase
2755 * the reference count if it already exists.
2757 * The caller must hold the rtnl_mutex.
2759 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
2765 netif_tx_lock_bh(dev
);
2766 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
2768 __dev_set_rx_mode(dev
);
2769 netif_tx_unlock_bh(dev
);
2772 EXPORT_SYMBOL(dev_unicast_add
);
2774 static void __dev_addr_discard(struct dev_addr_list
**list
)
2776 struct dev_addr_list
*tmp
;
2778 while (*list
!= NULL
) {
2781 if (tmp
->da_users
> tmp
->da_gusers
)
2782 printk("__dev_addr_discard: address leakage! "
2783 "da_users=%d\n", tmp
->da_users
);
2788 static void dev_addr_discard(struct net_device
*dev
)
2790 netif_tx_lock_bh(dev
);
2792 __dev_addr_discard(&dev
->uc_list
);
2795 __dev_addr_discard(&dev
->mc_list
);
2798 netif_tx_unlock_bh(dev
);
2801 unsigned dev_get_flags(const struct net_device
*dev
)
2805 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2810 (dev
->gflags
& (IFF_PROMISC
|
2813 if (netif_running(dev
)) {
2814 if (netif_oper_up(dev
))
2815 flags
|= IFF_RUNNING
;
2816 if (netif_carrier_ok(dev
))
2817 flags
|= IFF_LOWER_UP
;
2818 if (netif_dormant(dev
))
2819 flags
|= IFF_DORMANT
;
2825 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2828 int old_flags
= dev
->flags
;
2833 * Set the flags on our device.
2836 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2837 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2839 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2843 * Load in the correct multicast list now the flags have changed.
2846 if (dev
->change_rx_flags
&& (dev
->flags
^ flags
) & IFF_MULTICAST
)
2847 dev
->change_rx_flags(dev
, IFF_MULTICAST
);
2849 dev_set_rx_mode(dev
);
2852 * Have we downed the interface. We handle IFF_UP ourselves
2853 * according to user attempts to set it, rather than blindly
2858 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2859 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2862 dev_set_rx_mode(dev
);
2865 if (dev
->flags
& IFF_UP
&&
2866 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2868 raw_notifier_call_chain(&netdev_chain
,
2869 NETDEV_CHANGE
, dev
);
2871 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2872 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2873 dev
->gflags
^= IFF_PROMISC
;
2874 dev_set_promiscuity(dev
, inc
);
2877 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2878 is important. Some (broken) drivers set IFF_PROMISC, when
2879 IFF_ALLMULTI is requested not asking us and not reporting.
2881 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2882 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2883 dev
->gflags
^= IFF_ALLMULTI
;
2884 dev_set_allmulti(dev
, inc
);
2887 /* Exclude state transition flags, already notified */
2888 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
2890 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
2895 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2899 if (new_mtu
== dev
->mtu
)
2902 /* MTU must be positive. */
2906 if (!netif_device_present(dev
))
2910 if (dev
->change_mtu
)
2911 err
= dev
->change_mtu(dev
, new_mtu
);
2914 if (!err
&& dev
->flags
& IFF_UP
)
2915 raw_notifier_call_chain(&netdev_chain
,
2916 NETDEV_CHANGEMTU
, dev
);
2920 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2924 if (!dev
->set_mac_address
)
2926 if (sa
->sa_family
!= dev
->type
)
2928 if (!netif_device_present(dev
))
2930 err
= dev
->set_mac_address(dev
, sa
);
2932 raw_notifier_call_chain(&netdev_chain
,
2933 NETDEV_CHANGEADDR
, dev
);
2938 * Perform the SIOCxIFxxx calls.
2940 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2943 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2949 case SIOCGIFFLAGS
: /* Get interface flags */
2950 ifr
->ifr_flags
= dev_get_flags(dev
);
2953 case SIOCSIFFLAGS
: /* Set interface flags */
2954 return dev_change_flags(dev
, ifr
->ifr_flags
);
2956 case SIOCGIFMETRIC
: /* Get the metric on the interface
2957 (currently unused) */
2958 ifr
->ifr_metric
= 0;
2961 case SIOCSIFMETRIC
: /* Set the metric on the interface
2962 (currently unused) */
2965 case SIOCGIFMTU
: /* Get the MTU of a device */
2966 ifr
->ifr_mtu
= dev
->mtu
;
2969 case SIOCSIFMTU
: /* Set the MTU of a device */
2970 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
2974 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
2976 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
2977 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2978 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
2982 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
2984 case SIOCSIFHWBROADCAST
:
2985 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
2987 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
2988 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
2989 raw_notifier_call_chain(&netdev_chain
,
2990 NETDEV_CHANGEADDR
, dev
);
2994 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
2995 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
2996 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
2997 ifr
->ifr_map
.irq
= dev
->irq
;
2998 ifr
->ifr_map
.dma
= dev
->dma
;
2999 ifr
->ifr_map
.port
= dev
->if_port
;
3003 if (dev
->set_config
) {
3004 if (!netif_device_present(dev
))
3006 return dev
->set_config(dev
, &ifr
->ifr_map
);
3011 if (!dev
->set_multicast_list
||
3012 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3014 if (!netif_device_present(dev
))
3016 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
3020 if (!dev
->set_multicast_list
||
3021 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3023 if (!netif_device_present(dev
))
3025 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
3029 ifr
->ifr_ifindex
= dev
->ifindex
;
3033 ifr
->ifr_qlen
= dev
->tx_queue_len
;
3037 if (ifr
->ifr_qlen
< 0)
3039 dev
->tx_queue_len
= ifr
->ifr_qlen
;
3043 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
3044 return dev_change_name(dev
, ifr
->ifr_newname
);
3047 * Unknown or private ioctl
3051 if ((cmd
>= SIOCDEVPRIVATE
&&
3052 cmd
<= SIOCDEVPRIVATE
+ 15) ||
3053 cmd
== SIOCBONDENSLAVE
||
3054 cmd
== SIOCBONDRELEASE
||
3055 cmd
== SIOCBONDSETHWADDR
||
3056 cmd
== SIOCBONDSLAVEINFOQUERY
||
3057 cmd
== SIOCBONDINFOQUERY
||
3058 cmd
== SIOCBONDCHANGEACTIVE
||
3059 cmd
== SIOCGMIIPHY
||
3060 cmd
== SIOCGMIIREG
||
3061 cmd
== SIOCSMIIREG
||
3062 cmd
== SIOCBRADDIF
||
3063 cmd
== SIOCBRDELIF
||
3064 cmd
== SIOCWANDEV
) {
3066 if (dev
->do_ioctl
) {
3067 if (netif_device_present(dev
))
3068 err
= dev
->do_ioctl(dev
, ifr
,
3081 * This function handles all "interface"-type I/O control requests. The actual
3082 * 'doing' part of this is dev_ifsioc above.
3086 * dev_ioctl - network device ioctl
3087 * @cmd: command to issue
3088 * @arg: pointer to a struct ifreq in user space
3090 * Issue ioctl functions to devices. This is normally called by the
3091 * user space syscall interfaces but can sometimes be useful for
3092 * other purposes. The return value is the return from the syscall if
3093 * positive or a negative errno code on error.
3096 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
3102 /* One special case: SIOCGIFCONF takes ifconf argument
3103 and requires shared lock, because it sleeps writing
3107 if (cmd
== SIOCGIFCONF
) {
3109 ret
= dev_ifconf((char __user
*) arg
);
3113 if (cmd
== SIOCGIFNAME
)
3114 return dev_ifname((struct ifreq __user
*)arg
);
3116 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3119 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
3121 colon
= strchr(ifr
.ifr_name
, ':');
3126 * See which interface the caller is talking about.
3131 * These ioctl calls:
3132 * - can be done by all.
3133 * - atomic and do not require locking.
3144 dev_load(ifr
.ifr_name
);
3145 read_lock(&dev_base_lock
);
3146 ret
= dev_ifsioc(&ifr
, cmd
);
3147 read_unlock(&dev_base_lock
);
3151 if (copy_to_user(arg
, &ifr
,
3152 sizeof(struct ifreq
)))
3158 dev_load(ifr
.ifr_name
);
3160 ret
= dev_ethtool(&ifr
);
3165 if (copy_to_user(arg
, &ifr
,
3166 sizeof(struct ifreq
)))
3172 * These ioctl calls:
3173 * - require superuser power.
3174 * - require strict serialization.
3180 if (!capable(CAP_NET_ADMIN
))
3182 dev_load(ifr
.ifr_name
);
3184 ret
= dev_ifsioc(&ifr
, cmd
);
3189 if (copy_to_user(arg
, &ifr
,
3190 sizeof(struct ifreq
)))
3196 * These ioctl calls:
3197 * - require superuser power.
3198 * - require strict serialization.
3199 * - do not return a value
3209 case SIOCSIFHWBROADCAST
:
3212 case SIOCBONDENSLAVE
:
3213 case SIOCBONDRELEASE
:
3214 case SIOCBONDSETHWADDR
:
3215 case SIOCBONDCHANGEACTIVE
:
3218 if (!capable(CAP_NET_ADMIN
))
3221 case SIOCBONDSLAVEINFOQUERY
:
3222 case SIOCBONDINFOQUERY
:
3223 dev_load(ifr
.ifr_name
);
3225 ret
= dev_ifsioc(&ifr
, cmd
);
3230 /* Get the per device memory space. We can add this but
3231 * currently do not support it */
3233 /* Set the per device memory buffer space.
3234 * Not applicable in our case */
3239 * Unknown or private ioctl.
3242 if (cmd
== SIOCWANDEV
||
3243 (cmd
>= SIOCDEVPRIVATE
&&
3244 cmd
<= SIOCDEVPRIVATE
+ 15)) {
3245 dev_load(ifr
.ifr_name
);
3247 ret
= dev_ifsioc(&ifr
, cmd
);
3249 if (!ret
&& copy_to_user(arg
, &ifr
,
3250 sizeof(struct ifreq
)))
3254 /* Take care of Wireless Extensions */
3255 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
3256 return wext_handle_ioctl(&ifr
, cmd
, arg
);
3263 * dev_new_index - allocate an ifindex
3265 * Returns a suitable unique value for a new device interface
3266 * number. The caller must hold the rtnl semaphore or the
3267 * dev_base_lock to be sure it remains unique.
3269 static int dev_new_index(void)
3275 if (!__dev_get_by_index(ifindex
))
3280 static int dev_boot_phase
= 1;
3282 /* Delayed registration/unregisteration */
3283 static DEFINE_SPINLOCK(net_todo_list_lock
);
3284 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
3286 static void net_set_todo(struct net_device
*dev
)
3288 spin_lock(&net_todo_list_lock
);
3289 list_add_tail(&dev
->todo_list
, &net_todo_list
);
3290 spin_unlock(&net_todo_list_lock
);
3294 * register_netdevice - register a network device
3295 * @dev: device to register
3297 * Take a completed network device structure and add it to the kernel
3298 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3299 * chain. 0 is returned on success. A negative errno code is returned
3300 * on a failure to set up the device, or if the name is a duplicate.
3302 * Callers must hold the rtnl semaphore. You may want
3303 * register_netdev() instead of this.
3306 * The locking appears insufficient to guarantee two parallel registers
3307 * will not get the same name.
3310 int register_netdevice(struct net_device
*dev
)
3312 struct hlist_head
*head
;
3313 struct hlist_node
*p
;
3316 BUG_ON(dev_boot_phase
);
3321 /* When net_device's are persistent, this will be fatal. */
3322 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
3324 spin_lock_init(&dev
->queue_lock
);
3325 spin_lock_init(&dev
->_xmit_lock
);
3326 netdev_set_lockdep_class(&dev
->_xmit_lock
, dev
->type
);
3327 dev
->xmit_lock_owner
= -1;
3328 spin_lock_init(&dev
->ingress_lock
);
3332 /* Init, if this function is available */
3334 ret
= dev
->init(dev
);
3342 if (!dev_valid_name(dev
->name
)) {
3347 dev
->ifindex
= dev_new_index();
3348 if (dev
->iflink
== -1)
3349 dev
->iflink
= dev
->ifindex
;
3351 /* Check for existence of name */
3352 head
= dev_name_hash(dev
->name
);
3353 hlist_for_each(p
, head
) {
3354 struct net_device
*d
3355 = hlist_entry(p
, struct net_device
, name_hlist
);
3356 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
3362 /* Fix illegal checksum combinations */
3363 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
3364 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3365 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
3367 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
3370 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
3371 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3372 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
3374 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
3378 /* Fix illegal SG+CSUM combinations. */
3379 if ((dev
->features
& NETIF_F_SG
) &&
3380 !(dev
->features
& NETIF_F_ALL_CSUM
)) {
3381 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no checksum feature.\n",
3383 dev
->features
&= ~NETIF_F_SG
;
3386 /* TSO requires that SG is present as well. */
3387 if ((dev
->features
& NETIF_F_TSO
) &&
3388 !(dev
->features
& NETIF_F_SG
)) {
3389 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no SG feature.\n",
3391 dev
->features
&= ~NETIF_F_TSO
;
3393 if (dev
->features
& NETIF_F_UFO
) {
3394 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
3395 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3396 "NETIF_F_HW_CSUM feature.\n",
3398 dev
->features
&= ~NETIF_F_UFO
;
3400 if (!(dev
->features
& NETIF_F_SG
)) {
3401 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3402 "NETIF_F_SG feature.\n",
3404 dev
->features
&= ~NETIF_F_UFO
;
3409 * nil rebuild_header routine,
3410 * that should be never called and used as just bug trap.
3413 if (!dev
->rebuild_header
)
3414 dev
->rebuild_header
= default_rebuild_header
;
3416 ret
= netdev_register_sysfs(dev
);
3419 dev
->reg_state
= NETREG_REGISTERED
;
3422 * Default initial state at registry is that the
3423 * device is present.
3426 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
3428 dev_init_scheduler(dev
);
3429 write_lock_bh(&dev_base_lock
);
3430 list_add_tail(&dev
->dev_list
, &dev_base_head
);
3431 hlist_add_head(&dev
->name_hlist
, head
);
3432 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
3434 write_unlock_bh(&dev_base_lock
);
3436 /* Notify protocols, that a new device appeared. */
3437 raw_notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
3451 * register_netdev - register a network device
3452 * @dev: device to register
3454 * Take a completed network device structure and add it to the kernel
3455 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3456 * chain. 0 is returned on success. A negative errno code is returned
3457 * on a failure to set up the device, or if the name is a duplicate.
3459 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3460 * and expands the device name if you passed a format string to
3463 int register_netdev(struct net_device
*dev
)
3470 * If the name is a format string the caller wants us to do a
3473 if (strchr(dev
->name
, '%')) {
3474 err
= dev_alloc_name(dev
, dev
->name
);
3479 err
= register_netdevice(dev
);
3484 EXPORT_SYMBOL(register_netdev
);
3487 * netdev_wait_allrefs - wait until all references are gone.
3489 * This is called when unregistering network devices.
3491 * Any protocol or device that holds a reference should register
3492 * for netdevice notification, and cleanup and put back the
3493 * reference if they receive an UNREGISTER event.
3494 * We can get stuck here if buggy protocols don't correctly
3497 static void netdev_wait_allrefs(struct net_device
*dev
)
3499 unsigned long rebroadcast_time
, warning_time
;
3501 rebroadcast_time
= warning_time
= jiffies
;
3502 while (atomic_read(&dev
->refcnt
) != 0) {
3503 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
3506 /* Rebroadcast unregister notification */
3507 raw_notifier_call_chain(&netdev_chain
,
3508 NETDEV_UNREGISTER
, dev
);
3510 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
3512 /* We must not have linkwatch events
3513 * pending on unregister. If this
3514 * happens, we simply run the queue
3515 * unscheduled, resulting in a noop
3518 linkwatch_run_queue();
3523 rebroadcast_time
= jiffies
;
3528 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
3529 printk(KERN_EMERG
"unregister_netdevice: "
3530 "waiting for %s to become free. Usage "
3532 dev
->name
, atomic_read(&dev
->refcnt
));
3533 warning_time
= jiffies
;
3542 * register_netdevice(x1);
3543 * register_netdevice(x2);
3545 * unregister_netdevice(y1);
3546 * unregister_netdevice(y2);
3552 * We are invoked by rtnl_unlock() after it drops the semaphore.
3553 * This allows us to deal with problems:
3554 * 1) We can delete sysfs objects which invoke hotplug
3555 * without deadlocking with linkwatch via keventd.
3556 * 2) Since we run with the RTNL semaphore not held, we can sleep
3557 * safely in order to wait for the netdev refcnt to drop to zero.
3559 static DEFINE_MUTEX(net_todo_run_mutex
);
3560 void netdev_run_todo(void)
3562 struct list_head list
;
3564 /* Need to guard against multiple cpu's getting out of order. */
3565 mutex_lock(&net_todo_run_mutex
);
3567 /* Not safe to do outside the semaphore. We must not return
3568 * until all unregister events invoked by the local processor
3569 * have been completed (either by this todo run, or one on
3572 if (list_empty(&net_todo_list
))
3575 /* Snapshot list, allow later requests */
3576 spin_lock(&net_todo_list_lock
);
3577 list_replace_init(&net_todo_list
, &list
);
3578 spin_unlock(&net_todo_list_lock
);
3580 while (!list_empty(&list
)) {
3581 struct net_device
*dev
3582 = list_entry(list
.next
, struct net_device
, todo_list
);
3583 list_del(&dev
->todo_list
);
3585 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
3586 printk(KERN_ERR
"network todo '%s' but state %d\n",
3587 dev
->name
, dev
->reg_state
);
3592 dev
->reg_state
= NETREG_UNREGISTERED
;
3594 netdev_wait_allrefs(dev
);
3597 BUG_ON(atomic_read(&dev
->refcnt
));
3598 BUG_TRAP(!dev
->ip_ptr
);
3599 BUG_TRAP(!dev
->ip6_ptr
);
3600 BUG_TRAP(!dev
->dn_ptr
);
3602 if (dev
->destructor
)
3603 dev
->destructor(dev
);
3605 /* Free network device */
3606 kobject_put(&dev
->dev
.kobj
);
3610 mutex_unlock(&net_todo_run_mutex
);
3613 static struct net_device_stats
*internal_stats(struct net_device
*dev
)
3619 * alloc_netdev_mq - allocate network device
3620 * @sizeof_priv: size of private data to allocate space for
3621 * @name: device name format string
3622 * @setup: callback to initialize device
3623 * @queue_count: the number of subqueues to allocate
3625 * Allocates a struct net_device with private data area for driver use
3626 * and performs basic initialization. Also allocates subquue structs
3627 * for each queue on the device at the end of the netdevice.
3629 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
3630 void (*setup
)(struct net_device
*), unsigned int queue_count
)
3633 struct net_device
*dev
;
3636 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
3638 /* ensure 32-byte alignment of both the device and private area */
3639 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
+
3640 (sizeof(struct net_device_subqueue
) * (queue_count
- 1))) &
3641 ~NETDEV_ALIGN_CONST
;
3642 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
3644 p
= kzalloc(alloc_size
, GFP_KERNEL
);
3646 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
3650 dev
= (struct net_device
*)
3651 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3652 dev
->padded
= (char *)dev
- (char *)p
;
3655 dev
->priv
= ((char *)dev
+
3656 ((sizeof(struct net_device
) +
3657 (sizeof(struct net_device_subqueue
) *
3658 (queue_count
- 1)) + NETDEV_ALIGN_CONST
)
3659 & ~NETDEV_ALIGN_CONST
));
3662 dev
->egress_subqueue_count
= queue_count
;
3664 dev
->get_stats
= internal_stats
;
3666 strcpy(dev
->name
, name
);
3669 EXPORT_SYMBOL(alloc_netdev_mq
);
3672 * free_netdev - free network device
3675 * This function does the last stage of destroying an allocated device
3676 * interface. The reference to the device object is released.
3677 * If this is the last reference then it will be freed.
3679 void free_netdev(struct net_device
*dev
)
3682 /* Compatibility with error handling in drivers */
3683 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3684 kfree((char *)dev
- dev
->padded
);
3688 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3689 dev
->reg_state
= NETREG_RELEASED
;
3691 /* will free via device release */
3692 put_device(&dev
->dev
);
3694 kfree((char *)dev
- dev
->padded
);
3698 /* Synchronize with packet receive processing. */
3699 void synchronize_net(void)
3706 * unregister_netdevice - remove device from the kernel
3709 * This function shuts down a device interface and removes it
3710 * from the kernel tables. On success 0 is returned, on a failure
3711 * a negative errno code is returned.
3713 * Callers must hold the rtnl semaphore. You may want
3714 * unregister_netdev() instead of this.
3717 void unregister_netdevice(struct net_device
*dev
)
3719 BUG_ON(dev_boot_phase
);
3722 /* Some devices call without registering for initialization unwind. */
3723 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3724 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3725 "was registered\n", dev
->name
, dev
);
3731 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3733 /* If device is running, close it first. */
3734 if (dev
->flags
& IFF_UP
)
3737 /* And unlink it from device chain. */
3738 write_lock_bh(&dev_base_lock
);
3739 list_del(&dev
->dev_list
);
3740 hlist_del(&dev
->name_hlist
);
3741 hlist_del(&dev
->index_hlist
);
3742 write_unlock_bh(&dev_base_lock
);
3744 dev
->reg_state
= NETREG_UNREGISTERING
;
3748 /* Shutdown queueing discipline. */
3752 /* Notify protocols, that we are about to destroy
3753 this device. They should clean all the things.
3755 raw_notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3758 * Flush the unicast and multicast chains
3760 dev_addr_discard(dev
);
3765 /* Notifier chain MUST detach us from master device. */
3766 BUG_TRAP(!dev
->master
);
3768 /* Remove entries from sysfs */
3769 netdev_unregister_sysfs(dev
);
3771 /* Finish processing unregister after unlock */
3780 * unregister_netdev - remove device from the kernel
3783 * This function shuts down a device interface and removes it
3784 * from the kernel tables. On success 0 is returned, on a failure
3785 * a negative errno code is returned.
3787 * This is just a wrapper for unregister_netdevice that takes
3788 * the rtnl semaphore. In general you want to use this and not
3789 * unregister_netdevice.
3791 void unregister_netdev(struct net_device
*dev
)
3794 unregister_netdevice(dev
);
3798 EXPORT_SYMBOL(unregister_netdev
);
3800 static int dev_cpu_callback(struct notifier_block
*nfb
,
3801 unsigned long action
,
3804 struct sk_buff
**list_skb
;
3805 struct net_device
**list_net
;
3806 struct sk_buff
*skb
;
3807 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3808 struct softnet_data
*sd
, *oldsd
;
3810 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
3813 local_irq_disable();
3814 cpu
= smp_processor_id();
3815 sd
= &per_cpu(softnet_data
, cpu
);
3816 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3818 /* Find end of our completion_queue. */
3819 list_skb
= &sd
->completion_queue
;
3821 list_skb
= &(*list_skb
)->next
;
3822 /* Append completion queue from offline CPU. */
3823 *list_skb
= oldsd
->completion_queue
;
3824 oldsd
->completion_queue
= NULL
;
3826 /* Find end of our output_queue. */
3827 list_net
= &sd
->output_queue
;
3829 list_net
= &(*list_net
)->next_sched
;
3830 /* Append output queue from offline CPU. */
3831 *list_net
= oldsd
->output_queue
;
3832 oldsd
->output_queue
= NULL
;
3834 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3837 /* Process offline CPU's input_pkt_queue */
3838 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3844 #ifdef CONFIG_NET_DMA
3846 * net_dma_rebalance - try to maintain one DMA channel per CPU
3847 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
3849 * This is called when the number of channels allocated to the net_dma client
3850 * changes. The net_dma client tries to have one DMA channel per CPU.
3853 static void net_dma_rebalance(struct net_dma
*net_dma
)
3855 unsigned int cpu
, i
, n
, chan_idx
;
3856 struct dma_chan
*chan
;
3858 if (cpus_empty(net_dma
->channel_mask
)) {
3859 for_each_online_cpu(cpu
)
3860 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
3865 cpu
= first_cpu(cpu_online_map
);
3867 for_each_cpu_mask(chan_idx
, net_dma
->channel_mask
) {
3868 chan
= net_dma
->channels
[chan_idx
];
3870 n
= ((num_online_cpus() / cpus_weight(net_dma
->channel_mask
))
3871 + (i
< (num_online_cpus() %
3872 cpus_weight(net_dma
->channel_mask
)) ? 1 : 0));
3875 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
3876 cpu
= next_cpu(cpu
, cpu_online_map
);
3884 * netdev_dma_event - event callback for the net_dma_client
3885 * @client: should always be net_dma_client
3886 * @chan: DMA channel for the event
3887 * @state: DMA state to be handled
3889 static enum dma_state_client
3890 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
3891 enum dma_state state
)
3893 int i
, found
= 0, pos
= -1;
3894 struct net_dma
*net_dma
=
3895 container_of(client
, struct net_dma
, client
);
3896 enum dma_state_client ack
= DMA_DUP
; /* default: take no action */
3898 spin_lock(&net_dma
->lock
);
3900 case DMA_RESOURCE_AVAILABLE
:
3901 for (i
= 0; i
< NR_CPUS
; i
++)
3902 if (net_dma
->channels
[i
] == chan
) {
3905 } else if (net_dma
->channels
[i
] == NULL
&& pos
< 0)
3908 if (!found
&& pos
>= 0) {
3910 net_dma
->channels
[pos
] = chan
;
3911 cpu_set(pos
, net_dma
->channel_mask
);
3912 net_dma_rebalance(net_dma
);
3915 case DMA_RESOURCE_REMOVED
:
3916 for (i
= 0; i
< NR_CPUS
; i
++)
3917 if (net_dma
->channels
[i
] == chan
) {
3925 cpu_clear(pos
, net_dma
->channel_mask
);
3926 net_dma
->channels
[i
] = NULL
;
3927 net_dma_rebalance(net_dma
);
3933 spin_unlock(&net_dma
->lock
);
3939 * netdev_dma_regiser - register the networking subsystem as a DMA client
3941 static int __init
netdev_dma_register(void)
3943 spin_lock_init(&net_dma
.lock
);
3944 dma_cap_set(DMA_MEMCPY
, net_dma
.client
.cap_mask
);
3945 dma_async_client_register(&net_dma
.client
);
3946 dma_async_client_chan_request(&net_dma
.client
);
3951 static int __init
netdev_dma_register(void) { return -ENODEV
; }
3952 #endif /* CONFIG_NET_DMA */
3955 * Initialize the DEV module. At boot time this walks the device list and
3956 * unhooks any devices that fail to initialise (normally hardware not
3957 * present) and leaves us with a valid list of present and active devices.
3962 * This is called single threaded during boot, so no need
3963 * to take the rtnl semaphore.
3965 static int __init
net_dev_init(void)
3967 int i
, rc
= -ENOMEM
;
3969 BUG_ON(!dev_boot_phase
);
3971 if (dev_proc_init())
3974 if (netdev_sysfs_init())
3977 INIT_LIST_HEAD(&ptype_all
);
3978 for (i
= 0; i
< 16; i
++)
3979 INIT_LIST_HEAD(&ptype_base
[i
]);
3981 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
3982 INIT_HLIST_HEAD(&dev_name_head
[i
]);
3984 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
3985 INIT_HLIST_HEAD(&dev_index_head
[i
]);
3988 * Initialise the packet receive queues.
3991 for_each_possible_cpu(i
) {
3992 struct softnet_data
*queue
;
3994 queue
= &per_cpu(softnet_data
, i
);
3995 skb_queue_head_init(&queue
->input_pkt_queue
);
3996 queue
->completion_queue
= NULL
;
3997 INIT_LIST_HEAD(&queue
->poll_list
);
3998 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
3999 queue
->backlog_dev
.weight
= weight_p
;
4000 queue
->backlog_dev
.poll
= process_backlog
;
4001 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
4004 netdev_dma_register();
4008 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
4009 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
4011 hotcpu_notifier(dev_cpu_callback
, 0);
4019 subsys_initcall(net_dev_init
);
4021 EXPORT_SYMBOL(__dev_get_by_index
);
4022 EXPORT_SYMBOL(__dev_get_by_name
);
4023 EXPORT_SYMBOL(__dev_remove_pack
);
4024 EXPORT_SYMBOL(dev_valid_name
);
4025 EXPORT_SYMBOL(dev_add_pack
);
4026 EXPORT_SYMBOL(dev_alloc_name
);
4027 EXPORT_SYMBOL(dev_close
);
4028 EXPORT_SYMBOL(dev_get_by_flags
);
4029 EXPORT_SYMBOL(dev_get_by_index
);
4030 EXPORT_SYMBOL(dev_get_by_name
);
4031 EXPORT_SYMBOL(dev_open
);
4032 EXPORT_SYMBOL(dev_queue_xmit
);
4033 EXPORT_SYMBOL(dev_remove_pack
);
4034 EXPORT_SYMBOL(dev_set_allmulti
);
4035 EXPORT_SYMBOL(dev_set_promiscuity
);
4036 EXPORT_SYMBOL(dev_change_flags
);
4037 EXPORT_SYMBOL(dev_set_mtu
);
4038 EXPORT_SYMBOL(dev_set_mac_address
);
4039 EXPORT_SYMBOL(free_netdev
);
4040 EXPORT_SYMBOL(netdev_boot_setup_check
);
4041 EXPORT_SYMBOL(netdev_set_master
);
4042 EXPORT_SYMBOL(netdev_state_change
);
4043 EXPORT_SYMBOL(netif_receive_skb
);
4044 EXPORT_SYMBOL(netif_rx
);
4045 EXPORT_SYMBOL(register_gifconf
);
4046 EXPORT_SYMBOL(register_netdevice
);
4047 EXPORT_SYMBOL(register_netdevice_notifier
);
4048 EXPORT_SYMBOL(skb_checksum_help
);
4049 EXPORT_SYMBOL(synchronize_net
);
4050 EXPORT_SYMBOL(unregister_netdevice
);
4051 EXPORT_SYMBOL(unregister_netdevice_notifier
);
4052 EXPORT_SYMBOL(net_enable_timestamp
);
4053 EXPORT_SYMBOL(net_disable_timestamp
);
4054 EXPORT_SYMBOL(dev_get_flags
);
4056 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4057 EXPORT_SYMBOL(br_handle_frame_hook
);
4058 EXPORT_SYMBOL(br_fdb_get_hook
);
4059 EXPORT_SYMBOL(br_fdb_put_hook
);
4063 EXPORT_SYMBOL(dev_load
);
4066 EXPORT_PER_CPU_SYMBOL(softnet_data
);