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
)
820 char oldname
[IFNAMSIZ
];
826 if (dev
->flags
& IFF_UP
)
829 if (!dev_valid_name(newname
))
832 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
834 if (strchr(newname
, '%')) {
835 err
= dev_alloc_name(dev
, newname
);
838 strcpy(newname
, dev
->name
);
840 else if (__dev_get_by_name(newname
))
843 strlcpy(dev
->name
, newname
, IFNAMSIZ
);
846 device_rename(&dev
->dev
, dev
->name
);
848 write_lock_bh(&dev_base_lock
);
849 hlist_del(&dev
->name_hlist
);
850 hlist_add_head(&dev
->name_hlist
, dev_name_hash(dev
->name
));
851 write_unlock_bh(&dev_base_lock
);
853 ret
= raw_notifier_call_chain(&netdev_chain
, NETDEV_CHANGENAME
, dev
);
854 ret
= notifier_to_errno(ret
);
859 "%s: name change rollback failed: %d.\n",
863 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
872 * netdev_features_change - device changes features
873 * @dev: device to cause notification
875 * Called to indicate a device has changed features.
877 void netdev_features_change(struct net_device
*dev
)
879 raw_notifier_call_chain(&netdev_chain
, NETDEV_FEAT_CHANGE
, dev
);
881 EXPORT_SYMBOL(netdev_features_change
);
884 * netdev_state_change - device changes state
885 * @dev: device to cause notification
887 * Called to indicate a device has changed state. This function calls
888 * the notifier chains for netdev_chain and sends a NEWLINK message
889 * to the routing socket.
891 void netdev_state_change(struct net_device
*dev
)
893 if (dev
->flags
& IFF_UP
) {
894 raw_notifier_call_chain(&netdev_chain
,
896 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
901 * dev_load - load a network module
902 * @name: name of interface
904 * If a network interface is not present and the process has suitable
905 * privileges this function loads the module. If module loading is not
906 * available in this kernel then it becomes a nop.
909 void dev_load(const char *name
)
911 struct net_device
*dev
;
913 read_lock(&dev_base_lock
);
914 dev
= __dev_get_by_name(name
);
915 read_unlock(&dev_base_lock
);
917 if (!dev
&& capable(CAP_SYS_MODULE
))
918 request_module("%s", name
);
921 static int default_rebuild_header(struct sk_buff
*skb
)
923 printk(KERN_DEBUG
"%s: default_rebuild_header called -- BUG!\n",
924 skb
->dev
? skb
->dev
->name
: "NULL!!!");
930 * dev_open - prepare an interface for use.
931 * @dev: device to open
933 * Takes a device from down to up state. The device's private open
934 * function is invoked and then the multicast lists are loaded. Finally
935 * the device is moved into the up state and a %NETDEV_UP message is
936 * sent to the netdev notifier chain.
938 * Calling this function on an active interface is a nop. On a failure
939 * a negative errno code is returned.
941 int dev_open(struct net_device
*dev
)
949 if (dev
->flags
& IFF_UP
)
953 * Is it even present?
955 if (!netif_device_present(dev
))
959 * Call device private open method
961 set_bit(__LINK_STATE_START
, &dev
->state
);
963 ret
= dev
->open(dev
);
965 clear_bit(__LINK_STATE_START
, &dev
->state
);
969 * If it went open OK then:
976 dev
->flags
|= IFF_UP
;
979 * Initialize multicasting status
981 dev_set_rx_mode(dev
);
984 * Wakeup transmit queue engine
989 * ... and announce new interface.
991 raw_notifier_call_chain(&netdev_chain
, NETDEV_UP
, dev
);
997 * dev_close - shutdown an interface.
998 * @dev: device to shutdown
1000 * This function moves an active device into down state. A
1001 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1002 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1005 int dev_close(struct net_device
*dev
)
1007 if (!(dev
->flags
& IFF_UP
))
1011 * Tell people we are going down, so that they can
1012 * prepare to death, when device is still operating.
1014 raw_notifier_call_chain(&netdev_chain
, NETDEV_GOING_DOWN
, dev
);
1016 dev_deactivate(dev
);
1018 clear_bit(__LINK_STATE_START
, &dev
->state
);
1020 /* Synchronize to scheduled poll. We cannot touch poll list,
1021 * it can be even on different cpu. So just clear netif_running(),
1022 * and wait when poll really will happen. Actually, the best place
1023 * for this is inside dev->stop() after device stopped its irq
1024 * engine, but this requires more changes in devices. */
1026 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1027 while (test_bit(__LINK_STATE_RX_SCHED
, &dev
->state
)) {
1033 * Call the device specific close. This cannot fail.
1034 * Only if device is UP
1036 * We allow it to be called even after a DETACH hot-plug
1043 * Device is now down.
1046 dev
->flags
&= ~IFF_UP
;
1049 * Tell people we are down
1051 raw_notifier_call_chain(&netdev_chain
, NETDEV_DOWN
, dev
);
1058 * Device change register/unregister. These are not inline or static
1059 * as we export them to the world.
1063 * register_netdevice_notifier - register a network notifier block
1066 * Register a notifier to be called when network device events occur.
1067 * The notifier passed is linked into the kernel structures and must
1068 * not be reused until it has been unregistered. A negative errno code
1069 * is returned on a failure.
1071 * When registered all registration and up events are replayed
1072 * to the new notifier to allow device to have a race free
1073 * view of the network device list.
1076 int register_netdevice_notifier(struct notifier_block
*nb
)
1078 struct net_device
*dev
;
1079 struct net_device
*last
;
1083 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1087 for_each_netdev(dev
) {
1088 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1089 err
= notifier_to_errno(err
);
1093 if (!(dev
->flags
& IFF_UP
))
1096 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1105 for_each_netdev(dev
) {
1109 if (dev
->flags
& IFF_UP
) {
1110 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1111 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1113 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1119 * unregister_netdevice_notifier - unregister a network notifier block
1122 * Unregister a notifier previously registered by
1123 * register_netdevice_notifier(). The notifier is unlinked into the
1124 * kernel structures and may then be reused. A negative errno code
1125 * is returned on a failure.
1128 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1133 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1139 * call_netdevice_notifiers - call all network notifier blocks
1140 * @val: value passed unmodified to notifier function
1141 * @v: pointer passed unmodified to notifier function
1143 * Call all network notifier blocks. Parameters and return value
1144 * are as for raw_notifier_call_chain().
1147 int call_netdevice_notifiers(unsigned long val
, void *v
)
1149 return raw_notifier_call_chain(&netdev_chain
, val
, v
);
1152 /* When > 0 there are consumers of rx skb time stamps */
1153 static atomic_t netstamp_needed
= ATOMIC_INIT(0);
1155 void net_enable_timestamp(void)
1157 atomic_inc(&netstamp_needed
);
1160 void net_disable_timestamp(void)
1162 atomic_dec(&netstamp_needed
);
1165 static inline void net_timestamp(struct sk_buff
*skb
)
1167 if (atomic_read(&netstamp_needed
))
1168 __net_timestamp(skb
);
1170 skb
->tstamp
.tv64
= 0;
1174 * Support routine. Sends outgoing frames to any network
1175 * taps currently in use.
1178 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1180 struct packet_type
*ptype
;
1185 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1186 /* Never send packets back to the socket
1187 * they originated from - MvS (miquels@drinkel.ow.org)
1189 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1190 (ptype
->af_packet_priv
== NULL
||
1191 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1192 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
1196 /* skb->nh should be correctly
1197 set by sender, so that the second statement is
1198 just protection against buggy protocols.
1200 skb_reset_mac_header(skb2
);
1202 if (skb_network_header(skb2
) < skb2
->data
||
1203 skb2
->network_header
> skb2
->tail
) {
1204 if (net_ratelimit())
1205 printk(KERN_CRIT
"protocol %04x is "
1207 skb2
->protocol
, dev
->name
);
1208 skb_reset_network_header(skb2
);
1211 skb2
->transport_header
= skb2
->network_header
;
1212 skb2
->pkt_type
= PACKET_OUTGOING
;
1213 ptype
->func(skb2
, skb
->dev
, ptype
, skb
->dev
);
1220 void __netif_schedule(struct net_device
*dev
)
1222 if (!test_and_set_bit(__LINK_STATE_SCHED
, &dev
->state
)) {
1223 unsigned long flags
;
1224 struct softnet_data
*sd
;
1226 local_irq_save(flags
);
1227 sd
= &__get_cpu_var(softnet_data
);
1228 dev
->next_sched
= sd
->output_queue
;
1229 sd
->output_queue
= dev
;
1230 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1231 local_irq_restore(flags
);
1234 EXPORT_SYMBOL(__netif_schedule
);
1236 void __netif_rx_schedule(struct net_device
*dev
)
1238 unsigned long flags
;
1240 local_irq_save(flags
);
1242 list_add_tail(&dev
->poll_list
, &__get_cpu_var(softnet_data
).poll_list
);
1244 dev
->quota
+= dev
->weight
;
1246 dev
->quota
= dev
->weight
;
1247 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
1248 local_irq_restore(flags
);
1250 EXPORT_SYMBOL(__netif_rx_schedule
);
1252 void dev_kfree_skb_any(struct sk_buff
*skb
)
1254 if (in_irq() || irqs_disabled())
1255 dev_kfree_skb_irq(skb
);
1259 EXPORT_SYMBOL(dev_kfree_skb_any
);
1263 void netif_device_detach(struct net_device
*dev
)
1265 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1266 netif_running(dev
)) {
1267 netif_stop_queue(dev
);
1270 EXPORT_SYMBOL(netif_device_detach
);
1272 void netif_device_attach(struct net_device
*dev
)
1274 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1275 netif_running(dev
)) {
1276 netif_wake_queue(dev
);
1277 __netdev_watchdog_up(dev
);
1280 EXPORT_SYMBOL(netif_device_attach
);
1284 * Invalidate hardware checksum when packet is to be mangled, and
1285 * complete checksum manually on outgoing path.
1287 int skb_checksum_help(struct sk_buff
*skb
)
1290 int ret
= 0, offset
;
1292 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1293 goto out_set_summed
;
1295 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1296 /* Let GSO fix up the checksum. */
1297 goto out_set_summed
;
1300 if (skb_cloned(skb
)) {
1301 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1306 offset
= skb
->csum_start
- skb_headroom(skb
);
1307 BUG_ON(offset
> (int)skb
->len
);
1308 csum
= skb_checksum(skb
, offset
, skb
->len
-offset
, 0);
1310 offset
= skb_headlen(skb
) - offset
;
1311 BUG_ON(offset
<= 0);
1312 BUG_ON(skb
->csum_offset
+ 2 > offset
);
1314 *(__sum16
*)(skb
->head
+ skb
->csum_start
+ skb
->csum_offset
) =
1317 skb
->ip_summed
= CHECKSUM_NONE
;
1323 * skb_gso_segment - Perform segmentation on skb.
1324 * @skb: buffer to segment
1325 * @features: features for the output path (see dev->features)
1327 * This function segments the given skb and returns a list of segments.
1329 * It may return NULL if the skb requires no segmentation. This is
1330 * only possible when GSO is used for verifying header integrity.
1332 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
, int features
)
1334 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1335 struct packet_type
*ptype
;
1336 __be16 type
= skb
->protocol
;
1339 BUG_ON(skb_shinfo(skb
)->frag_list
);
1341 skb_reset_mac_header(skb
);
1342 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1343 __skb_pull(skb
, skb
->mac_len
);
1345 if (WARN_ON(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1346 if (skb_header_cloned(skb
) &&
1347 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1348 return ERR_PTR(err
);
1352 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
) & 15], list
) {
1353 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1354 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1355 err
= ptype
->gso_send_check(skb
);
1356 segs
= ERR_PTR(err
);
1357 if (err
|| skb_gso_ok(skb
, features
))
1359 __skb_push(skb
, (skb
->data
-
1360 skb_network_header(skb
)));
1362 segs
= ptype
->gso_segment(skb
, features
);
1368 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1373 EXPORT_SYMBOL(skb_gso_segment
);
1375 /* Take action when hardware reception checksum errors are detected. */
1377 void netdev_rx_csum_fault(struct net_device
*dev
)
1379 if (net_ratelimit()) {
1380 printk(KERN_ERR
"%s: hw csum failure.\n",
1381 dev
? dev
->name
: "<unknown>");
1385 EXPORT_SYMBOL(netdev_rx_csum_fault
);
1388 /* Actually, we should eliminate this check as soon as we know, that:
1389 * 1. IOMMU is present and allows to map all the memory.
1390 * 2. No high memory really exists on this machine.
1393 static inline int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
1395 #ifdef CONFIG_HIGHMEM
1398 if (dev
->features
& NETIF_F_HIGHDMA
)
1401 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1402 if (PageHighMem(skb_shinfo(skb
)->frags
[i
].page
))
1410 void (*destructor
)(struct sk_buff
*skb
);
1413 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1415 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
1417 struct dev_gso_cb
*cb
;
1420 struct sk_buff
*nskb
= skb
->next
;
1422 skb
->next
= nskb
->next
;
1425 } while (skb
->next
);
1427 cb
= DEV_GSO_CB(skb
);
1429 cb
->destructor(skb
);
1433 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1434 * @skb: buffer to segment
1436 * This function segments the given skb and stores the list of segments
1439 static int dev_gso_segment(struct sk_buff
*skb
)
1441 struct net_device
*dev
= skb
->dev
;
1442 struct sk_buff
*segs
;
1443 int features
= dev
->features
& ~(illegal_highdma(dev
, skb
) ?
1446 segs
= skb_gso_segment(skb
, features
);
1448 /* Verifying header integrity only. */
1452 if (unlikely(IS_ERR(segs
)))
1453 return PTR_ERR(segs
);
1456 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
1457 skb
->destructor
= dev_gso_skb_destructor
;
1462 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1464 if (likely(!skb
->next
)) {
1465 if (!list_empty(&ptype_all
))
1466 dev_queue_xmit_nit(skb
, dev
);
1468 if (netif_needs_gso(dev
, skb
)) {
1469 if (unlikely(dev_gso_segment(skb
)))
1475 return dev
->hard_start_xmit(skb
, dev
);
1480 struct sk_buff
*nskb
= skb
->next
;
1483 skb
->next
= nskb
->next
;
1485 rc
= dev
->hard_start_xmit(nskb
, dev
);
1487 nskb
->next
= skb
->next
;
1491 if (unlikely((netif_queue_stopped(dev
) ||
1492 netif_subqueue_stopped(dev
, skb
->queue_mapping
)) &&
1494 return NETDEV_TX_BUSY
;
1495 } while (skb
->next
);
1497 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
1504 #define HARD_TX_LOCK(dev, cpu) { \
1505 if ((dev->features & NETIF_F_LLTX) == 0) { \
1506 netif_tx_lock(dev); \
1510 #define HARD_TX_UNLOCK(dev) { \
1511 if ((dev->features & NETIF_F_LLTX) == 0) { \
1512 netif_tx_unlock(dev); \
1517 * dev_queue_xmit - transmit a buffer
1518 * @skb: buffer to transmit
1520 * Queue a buffer for transmission to a network device. The caller must
1521 * have set the device and priority and built the buffer before calling
1522 * this function. The function can be called from an interrupt.
1524 * A negative errno code is returned on a failure. A success does not
1525 * guarantee the frame will be transmitted as it may be dropped due
1526 * to congestion or traffic shaping.
1528 * -----------------------------------------------------------------------------------
1529 * I notice this method can also return errors from the queue disciplines,
1530 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1533 * Regardless of the return value, the skb is consumed, so it is currently
1534 * difficult to retry a send to this method. (You can bump the ref count
1535 * before sending to hold a reference for retry if you are careful.)
1537 * When calling this method, interrupts MUST be enabled. This is because
1538 * the BH enable code must have IRQs enabled so that it will not deadlock.
1542 int dev_queue_xmit(struct sk_buff
*skb
)
1544 struct net_device
*dev
= skb
->dev
;
1548 /* GSO will handle the following emulations directly. */
1549 if (netif_needs_gso(dev
, skb
))
1552 if (skb_shinfo(skb
)->frag_list
&&
1553 !(dev
->features
& NETIF_F_FRAGLIST
) &&
1554 __skb_linearize(skb
))
1557 /* Fragmented skb is linearized if device does not support SG,
1558 * or if at least one of fragments is in highmem and device
1559 * does not support DMA from it.
1561 if (skb_shinfo(skb
)->nr_frags
&&
1562 (!(dev
->features
& NETIF_F_SG
) || illegal_highdma(dev
, skb
)) &&
1563 __skb_linearize(skb
))
1566 /* If packet is not checksummed and device does not support
1567 * checksumming for this protocol, complete checksumming here.
1569 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1570 skb_set_transport_header(skb
, skb
->csum_start
-
1573 if (!(dev
->features
& NETIF_F_GEN_CSUM
) &&
1574 !((dev
->features
& NETIF_F_IP_CSUM
) &&
1575 skb
->protocol
== htons(ETH_P_IP
)) &&
1576 !((dev
->features
& NETIF_F_IPV6_CSUM
) &&
1577 skb
->protocol
== htons(ETH_P_IPV6
)))
1578 if (skb_checksum_help(skb
))
1583 spin_lock_prefetch(&dev
->queue_lock
);
1585 /* Disable soft irqs for various locks below. Also
1586 * stops preemption for RCU.
1590 /* Updates of qdisc are serialized by queue_lock.
1591 * The struct Qdisc which is pointed to by qdisc is now a
1592 * rcu structure - it may be accessed without acquiring
1593 * a lock (but the structure may be stale.) The freeing of the
1594 * qdisc will be deferred until it's known that there are no
1595 * more references to it.
1597 * If the qdisc has an enqueue function, we still need to
1598 * hold the queue_lock before calling it, since queue_lock
1599 * also serializes access to the device queue.
1602 q
= rcu_dereference(dev
->qdisc
);
1603 #ifdef CONFIG_NET_CLS_ACT
1604 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_EGRESS
);
1607 /* Grab device queue */
1608 spin_lock(&dev
->queue_lock
);
1611 /* reset queue_mapping to zero */
1612 skb
->queue_mapping
= 0;
1613 rc
= q
->enqueue(skb
, q
);
1615 spin_unlock(&dev
->queue_lock
);
1617 rc
= rc
== NET_XMIT_BYPASS
? NET_XMIT_SUCCESS
: rc
;
1620 spin_unlock(&dev
->queue_lock
);
1623 /* The device has no queue. Common case for software devices:
1624 loopback, all the sorts of tunnels...
1626 Really, it is unlikely that netif_tx_lock protection is necessary
1627 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1629 However, it is possible, that they rely on protection
1632 Check this and shot the lock. It is not prone from deadlocks.
1633 Either shot noqueue qdisc, it is even simpler 8)
1635 if (dev
->flags
& IFF_UP
) {
1636 int cpu
= smp_processor_id(); /* ok because BHs are off */
1638 if (dev
->xmit_lock_owner
!= cpu
) {
1640 HARD_TX_LOCK(dev
, cpu
);
1642 if (!netif_queue_stopped(dev
) &&
1643 !netif_subqueue_stopped(dev
, skb
->queue_mapping
)) {
1645 if (!dev_hard_start_xmit(skb
, dev
)) {
1646 HARD_TX_UNLOCK(dev
);
1650 HARD_TX_UNLOCK(dev
);
1651 if (net_ratelimit())
1652 printk(KERN_CRIT
"Virtual device %s asks to "
1653 "queue packet!\n", dev
->name
);
1655 /* Recursion is detected! It is possible,
1657 if (net_ratelimit())
1658 printk(KERN_CRIT
"Dead loop on virtual device "
1659 "%s, fix it urgently!\n", dev
->name
);
1664 rcu_read_unlock_bh();
1670 rcu_read_unlock_bh();
1675 /*=======================================================================
1677 =======================================================================*/
1679 int netdev_max_backlog __read_mostly
= 1000;
1680 int netdev_budget __read_mostly
= 300;
1681 int weight_p __read_mostly
= 64; /* old backlog weight */
1683 DEFINE_PER_CPU(struct netif_rx_stats
, netdev_rx_stat
) = { 0, };
1687 * netif_rx - post buffer to the network code
1688 * @skb: buffer to post
1690 * This function receives a packet from a device driver and queues it for
1691 * the upper (protocol) levels to process. It always succeeds. The buffer
1692 * may be dropped during processing for congestion control or by the
1696 * NET_RX_SUCCESS (no congestion)
1697 * NET_RX_CN_LOW (low congestion)
1698 * NET_RX_CN_MOD (moderate congestion)
1699 * NET_RX_CN_HIGH (high congestion)
1700 * NET_RX_DROP (packet was dropped)
1704 int netif_rx(struct sk_buff
*skb
)
1706 struct softnet_data
*queue
;
1707 unsigned long flags
;
1709 /* if netpoll wants it, pretend we never saw it */
1710 if (netpoll_rx(skb
))
1713 if (!skb
->tstamp
.tv64
)
1717 * The code is rearranged so that the path is the most
1718 * short when CPU is congested, but is still operating.
1720 local_irq_save(flags
);
1721 queue
= &__get_cpu_var(softnet_data
);
1723 __get_cpu_var(netdev_rx_stat
).total
++;
1724 if (queue
->input_pkt_queue
.qlen
<= netdev_max_backlog
) {
1725 if (queue
->input_pkt_queue
.qlen
) {
1728 __skb_queue_tail(&queue
->input_pkt_queue
, skb
);
1729 local_irq_restore(flags
);
1730 return NET_RX_SUCCESS
;
1733 netif_rx_schedule(&queue
->backlog_dev
);
1737 __get_cpu_var(netdev_rx_stat
).dropped
++;
1738 local_irq_restore(flags
);
1744 int netif_rx_ni(struct sk_buff
*skb
)
1749 err
= netif_rx(skb
);
1750 if (local_softirq_pending())
1757 EXPORT_SYMBOL(netif_rx_ni
);
1759 static inline struct net_device
*skb_bond(struct sk_buff
*skb
)
1761 struct net_device
*dev
= skb
->dev
;
1764 if (skb_bond_should_drop(skb
)) {
1768 skb
->dev
= dev
->master
;
1774 static void net_tx_action(struct softirq_action
*h
)
1776 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
1778 if (sd
->completion_queue
) {
1779 struct sk_buff
*clist
;
1781 local_irq_disable();
1782 clist
= sd
->completion_queue
;
1783 sd
->completion_queue
= NULL
;
1787 struct sk_buff
*skb
= clist
;
1788 clist
= clist
->next
;
1790 BUG_TRAP(!atomic_read(&skb
->users
));
1795 if (sd
->output_queue
) {
1796 struct net_device
*head
;
1798 local_irq_disable();
1799 head
= sd
->output_queue
;
1800 sd
->output_queue
= NULL
;
1804 struct net_device
*dev
= head
;
1805 head
= head
->next_sched
;
1807 smp_mb__before_clear_bit();
1808 clear_bit(__LINK_STATE_SCHED
, &dev
->state
);
1810 if (spin_trylock(&dev
->queue_lock
)) {
1812 spin_unlock(&dev
->queue_lock
);
1814 netif_schedule(dev
);
1820 static inline int deliver_skb(struct sk_buff
*skb
,
1821 struct packet_type
*pt_prev
,
1822 struct net_device
*orig_dev
)
1824 atomic_inc(&skb
->users
);
1825 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1828 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1829 /* These hooks defined here for ATM */
1831 struct net_bridge_fdb_entry
*(*br_fdb_get_hook
)(struct net_bridge
*br
,
1832 unsigned char *addr
);
1833 void (*br_fdb_put_hook
)(struct net_bridge_fdb_entry
*ent
) __read_mostly
;
1836 * If bridge module is loaded call bridging hook.
1837 * returns NULL if packet was consumed.
1839 struct sk_buff
*(*br_handle_frame_hook
)(struct net_bridge_port
*p
,
1840 struct sk_buff
*skb
) __read_mostly
;
1841 static inline struct sk_buff
*handle_bridge(struct sk_buff
*skb
,
1842 struct packet_type
**pt_prev
, int *ret
,
1843 struct net_device
*orig_dev
)
1845 struct net_bridge_port
*port
;
1847 if (skb
->pkt_type
== PACKET_LOOPBACK
||
1848 (port
= rcu_dereference(skb
->dev
->br_port
)) == NULL
)
1852 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1856 return br_handle_frame_hook(port
, skb
);
1859 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1862 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1863 struct sk_buff
*(*macvlan_handle_frame_hook
)(struct sk_buff
*skb
) __read_mostly
;
1864 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook
);
1866 static inline struct sk_buff
*handle_macvlan(struct sk_buff
*skb
,
1867 struct packet_type
**pt_prev
,
1869 struct net_device
*orig_dev
)
1871 if (skb
->dev
->macvlan_port
== NULL
)
1875 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
1878 return macvlan_handle_frame_hook(skb
);
1881 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1884 #ifdef CONFIG_NET_CLS_ACT
1885 /* TODO: Maybe we should just force sch_ingress to be compiled in
1886 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1887 * a compare and 2 stores extra right now if we dont have it on
1888 * but have CONFIG_NET_CLS_ACT
1889 * NOTE: This doesnt stop any functionality; if you dont have
1890 * the ingress scheduler, you just cant add policies on ingress.
1893 static int ing_filter(struct sk_buff
*skb
)
1896 struct net_device
*dev
= skb
->dev
;
1897 int result
= TC_ACT_OK
;
1899 if (dev
->qdisc_ingress
) {
1900 __u32 ttl
= (__u32
) G_TC_RTTL(skb
->tc_verd
);
1901 if (MAX_RED_LOOP
< ttl
++) {
1902 printk(KERN_WARNING
"Redir loop detected Dropping packet (%d->%d)\n",
1903 skb
->iif
, skb
->dev
->ifindex
);
1907 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
,ttl
);
1909 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
,AT_INGRESS
);
1911 spin_lock(&dev
->ingress_lock
);
1912 if ((q
= dev
->qdisc_ingress
) != NULL
)
1913 result
= q
->enqueue(skb
, q
);
1914 spin_unlock(&dev
->ingress_lock
);
1922 int netif_receive_skb(struct sk_buff
*skb
)
1924 struct packet_type
*ptype
, *pt_prev
;
1925 struct net_device
*orig_dev
;
1926 int ret
= NET_RX_DROP
;
1929 /* if we've gotten here through NAPI, check netpoll */
1930 if (skb
->dev
->poll
&& netpoll_rx(skb
))
1933 if (!skb
->tstamp
.tv64
)
1937 skb
->iif
= skb
->dev
->ifindex
;
1939 orig_dev
= skb_bond(skb
);
1944 __get_cpu_var(netdev_rx_stat
).total
++;
1946 skb_reset_network_header(skb
);
1947 skb_reset_transport_header(skb
);
1948 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1954 #ifdef CONFIG_NET_CLS_ACT
1955 if (skb
->tc_verd
& TC_NCLS
) {
1956 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
1961 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1962 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
1964 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1969 #ifdef CONFIG_NET_CLS_ACT
1971 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
1972 pt_prev
= NULL
; /* noone else should process this after*/
1974 skb
->tc_verd
= SET_TC_OK2MUNGE(skb
->tc_verd
);
1977 ret
= ing_filter(skb
);
1979 if (ret
== TC_ACT_SHOT
|| (ret
== TC_ACT_STOLEN
)) {
1988 skb
= handle_bridge(skb
, &pt_prev
, &ret
, orig_dev
);
1991 skb
= handle_macvlan(skb
, &pt_prev
, &ret
, orig_dev
);
1995 type
= skb
->protocol
;
1996 list_for_each_entry_rcu(ptype
, &ptype_base
[ntohs(type
)&15], list
) {
1997 if (ptype
->type
== type
&&
1998 (!ptype
->dev
|| ptype
->dev
== skb
->dev
)) {
2000 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
2006 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
2009 /* Jamal, now you will not able to escape explaining
2010 * me how you were going to use this. :-)
2020 static int process_backlog(struct net_device
*backlog_dev
, int *budget
)
2023 int quota
= min(backlog_dev
->quota
, *budget
);
2024 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2025 unsigned long start_time
= jiffies
;
2027 backlog_dev
->weight
= weight_p
;
2029 struct sk_buff
*skb
;
2030 struct net_device
*dev
;
2032 local_irq_disable();
2033 skb
= __skb_dequeue(&queue
->input_pkt_queue
);
2040 netif_receive_skb(skb
);
2046 if (work
>= quota
|| jiffies
- start_time
> 1)
2051 backlog_dev
->quota
-= work
;
2056 backlog_dev
->quota
-= work
;
2059 list_del(&backlog_dev
->poll_list
);
2060 smp_mb__before_clear_bit();
2061 netif_poll_enable(backlog_dev
);
2067 static void net_rx_action(struct softirq_action
*h
)
2069 struct softnet_data
*queue
= &__get_cpu_var(softnet_data
);
2070 unsigned long start_time
= jiffies
;
2071 int budget
= netdev_budget
;
2074 local_irq_disable();
2076 while (!list_empty(&queue
->poll_list
)) {
2077 struct net_device
*dev
;
2079 if (budget
<= 0 || jiffies
- start_time
> 1)
2084 dev
= list_entry(queue
->poll_list
.next
,
2085 struct net_device
, poll_list
);
2086 have
= netpoll_poll_lock(dev
);
2088 if (dev
->quota
<= 0 || dev
->poll(dev
, &budget
)) {
2089 netpoll_poll_unlock(have
);
2090 local_irq_disable();
2091 list_move_tail(&dev
->poll_list
, &queue
->poll_list
);
2093 dev
->quota
+= dev
->weight
;
2095 dev
->quota
= dev
->weight
;
2097 netpoll_poll_unlock(have
);
2099 local_irq_disable();
2104 #ifdef CONFIG_NET_DMA
2106 * There may not be any more sk_buffs coming right now, so push
2107 * any pending DMA copies to hardware
2109 if (!cpus_empty(net_dma
.channel_mask
)) {
2111 for_each_cpu_mask(chan_idx
, net_dma
.channel_mask
) {
2112 struct dma_chan
*chan
= net_dma
.channels
[chan_idx
];
2114 dma_async_memcpy_issue_pending(chan
);
2121 __get_cpu_var(netdev_rx_stat
).time_squeeze
++;
2122 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2126 static gifconf_func_t
* gifconf_list
[NPROTO
];
2129 * register_gifconf - register a SIOCGIF handler
2130 * @family: Address family
2131 * @gifconf: Function handler
2133 * Register protocol dependent address dumping routines. The handler
2134 * that is passed must not be freed or reused until it has been replaced
2135 * by another handler.
2137 int register_gifconf(unsigned int family
, gifconf_func_t
* gifconf
)
2139 if (family
>= NPROTO
)
2141 gifconf_list
[family
] = gifconf
;
2147 * Map an interface index to its name (SIOCGIFNAME)
2151 * We need this ioctl for efficient implementation of the
2152 * if_indextoname() function required by the IPv6 API. Without
2153 * it, we would have to search all the interfaces to find a
2157 static int dev_ifname(struct ifreq __user
*arg
)
2159 struct net_device
*dev
;
2163 * Fetch the caller's info block.
2166 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
2169 read_lock(&dev_base_lock
);
2170 dev
= __dev_get_by_index(ifr
.ifr_ifindex
);
2172 read_unlock(&dev_base_lock
);
2176 strcpy(ifr
.ifr_name
, dev
->name
);
2177 read_unlock(&dev_base_lock
);
2179 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
2185 * Perform a SIOCGIFCONF call. This structure will change
2186 * size eventually, and there is nothing I can do about it.
2187 * Thus we will need a 'compatibility mode'.
2190 static int dev_ifconf(char __user
*arg
)
2193 struct net_device
*dev
;
2200 * Fetch the caller's info block.
2203 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
2210 * Loop over the interfaces, and write an info block for each.
2214 for_each_netdev(dev
) {
2215 for (i
= 0; i
< NPROTO
; i
++) {
2216 if (gifconf_list
[i
]) {
2219 done
= gifconf_list
[i
](dev
, NULL
, 0);
2221 done
= gifconf_list
[i
](dev
, pos
+ total
,
2231 * All done. Write the updated control block back to the caller.
2233 ifc
.ifc_len
= total
;
2236 * Both BSD and Solaris return 0 here, so we do too.
2238 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
2241 #ifdef CONFIG_PROC_FS
2243 * This is invoked by the /proc filesystem handler to display a device
2246 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2249 struct net_device
*dev
;
2251 read_lock(&dev_base_lock
);
2253 return SEQ_START_TOKEN
;
2256 for_each_netdev(dev
)
2263 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2266 return v
== SEQ_START_TOKEN
?
2267 first_net_device() : next_net_device((struct net_device
*)v
);
2270 void dev_seq_stop(struct seq_file
*seq
, void *v
)
2272 read_unlock(&dev_base_lock
);
2275 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
2277 struct net_device_stats
*stats
= dev
->get_stats(dev
);
2279 seq_printf(seq
, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2280 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2281 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
2283 stats
->rx_dropped
+ stats
->rx_missed_errors
,
2284 stats
->rx_fifo_errors
,
2285 stats
->rx_length_errors
+ stats
->rx_over_errors
+
2286 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
2287 stats
->rx_compressed
, stats
->multicast
,
2288 stats
->tx_bytes
, stats
->tx_packets
,
2289 stats
->tx_errors
, stats
->tx_dropped
,
2290 stats
->tx_fifo_errors
, stats
->collisions
,
2291 stats
->tx_carrier_errors
+
2292 stats
->tx_aborted_errors
+
2293 stats
->tx_window_errors
+
2294 stats
->tx_heartbeat_errors
,
2295 stats
->tx_compressed
);
2299 * Called from the PROCfs module. This now uses the new arbitrary sized
2300 * /proc/net interface to create /proc/net/dev
2302 static int dev_seq_show(struct seq_file
*seq
, void *v
)
2304 if (v
== SEQ_START_TOKEN
)
2305 seq_puts(seq
, "Inter-| Receive "
2307 " face |bytes packets errs drop fifo frame "
2308 "compressed multicast|bytes packets errs "
2309 "drop fifo colls carrier compressed\n");
2311 dev_seq_printf_stats(seq
, v
);
2315 static struct netif_rx_stats
*softnet_get_online(loff_t
*pos
)
2317 struct netif_rx_stats
*rc
= NULL
;
2319 while (*pos
< NR_CPUS
)
2320 if (cpu_online(*pos
)) {
2321 rc
= &per_cpu(netdev_rx_stat
, *pos
);
2328 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2330 return softnet_get_online(pos
);
2333 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2336 return softnet_get_online(pos
);
2339 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
2343 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
2345 struct netif_rx_stats
*s
= v
;
2347 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2348 s
->total
, s
->dropped
, s
->time_squeeze
, 0,
2349 0, 0, 0, 0, /* was fastroute */
2354 static const struct seq_operations dev_seq_ops
= {
2355 .start
= dev_seq_start
,
2356 .next
= dev_seq_next
,
2357 .stop
= dev_seq_stop
,
2358 .show
= dev_seq_show
,
2361 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
2363 return seq_open(file
, &dev_seq_ops
);
2366 static const struct file_operations dev_seq_fops
= {
2367 .owner
= THIS_MODULE
,
2368 .open
= dev_seq_open
,
2370 .llseek
= seq_lseek
,
2371 .release
= seq_release
,
2374 static const struct seq_operations softnet_seq_ops
= {
2375 .start
= softnet_seq_start
,
2376 .next
= softnet_seq_next
,
2377 .stop
= softnet_seq_stop
,
2378 .show
= softnet_seq_show
,
2381 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
2383 return seq_open(file
, &softnet_seq_ops
);
2386 static const struct file_operations softnet_seq_fops
= {
2387 .owner
= THIS_MODULE
,
2388 .open
= softnet_seq_open
,
2390 .llseek
= seq_lseek
,
2391 .release
= seq_release
,
2394 static void *ptype_get_idx(loff_t pos
)
2396 struct packet_type
*pt
= NULL
;
2400 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
2406 for (t
= 0; t
< 16; t
++) {
2407 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
2416 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2419 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
2422 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2424 struct packet_type
*pt
;
2425 struct list_head
*nxt
;
2429 if (v
== SEQ_START_TOKEN
)
2430 return ptype_get_idx(0);
2433 nxt
= pt
->list
.next
;
2434 if (pt
->type
== htons(ETH_P_ALL
)) {
2435 if (nxt
!= &ptype_all
)
2438 nxt
= ptype_base
[0].next
;
2440 hash
= ntohs(pt
->type
) & 15;
2442 while (nxt
== &ptype_base
[hash
]) {
2445 nxt
= ptype_base
[hash
].next
;
2448 return list_entry(nxt
, struct packet_type
, list
);
2451 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
2456 static void ptype_seq_decode(struct seq_file
*seq
, void *sym
)
2458 #ifdef CONFIG_KALLSYMS
2459 unsigned long offset
= 0, symsize
;
2460 const char *symname
;
2464 symname
= kallsyms_lookup((unsigned long)sym
, &symsize
, &offset
,
2471 modname
= delim
= "";
2472 seq_printf(seq
, "%s%s%s%s+0x%lx", delim
, modname
, delim
,
2478 seq_printf(seq
, "[%p]", sym
);
2481 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
2483 struct packet_type
*pt
= v
;
2485 if (v
== SEQ_START_TOKEN
)
2486 seq_puts(seq
, "Type Device Function\n");
2488 if (pt
->type
== htons(ETH_P_ALL
))
2489 seq_puts(seq
, "ALL ");
2491 seq_printf(seq
, "%04x", ntohs(pt
->type
));
2493 seq_printf(seq
, " %-8s ",
2494 pt
->dev
? pt
->dev
->name
: "");
2495 ptype_seq_decode(seq
, pt
->func
);
2496 seq_putc(seq
, '\n');
2502 static const struct seq_operations ptype_seq_ops
= {
2503 .start
= ptype_seq_start
,
2504 .next
= ptype_seq_next
,
2505 .stop
= ptype_seq_stop
,
2506 .show
= ptype_seq_show
,
2509 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
2511 return seq_open(file
, &ptype_seq_ops
);
2514 static const struct file_operations ptype_seq_fops
= {
2515 .owner
= THIS_MODULE
,
2516 .open
= ptype_seq_open
,
2518 .llseek
= seq_lseek
,
2519 .release
= seq_release
,
2523 static int __init
dev_proc_init(void)
2527 if (!proc_net_fops_create("dev", S_IRUGO
, &dev_seq_fops
))
2529 if (!proc_net_fops_create("softnet_stat", S_IRUGO
, &softnet_seq_fops
))
2531 if (!proc_net_fops_create("ptype", S_IRUGO
, &ptype_seq_fops
))
2534 if (wext_proc_init())
2540 proc_net_remove("ptype");
2542 proc_net_remove("softnet_stat");
2544 proc_net_remove("dev");
2548 #define dev_proc_init() 0
2549 #endif /* CONFIG_PROC_FS */
2553 * netdev_set_master - set up master/slave pair
2554 * @slave: slave device
2555 * @master: new master device
2557 * Changes the master device of the slave. Pass %NULL to break the
2558 * bonding. The caller must hold the RTNL semaphore. On a failure
2559 * a negative errno code is returned. On success the reference counts
2560 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2561 * function returns zero.
2563 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
2565 struct net_device
*old
= slave
->master
;
2575 slave
->master
= master
;
2583 slave
->flags
|= IFF_SLAVE
;
2585 slave
->flags
&= ~IFF_SLAVE
;
2587 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
2591 static void __dev_set_promiscuity(struct net_device
*dev
, int inc
)
2593 unsigned short old_flags
= dev
->flags
;
2597 if ((dev
->promiscuity
+= inc
) == 0)
2598 dev
->flags
&= ~IFF_PROMISC
;
2600 dev
->flags
|= IFF_PROMISC
;
2601 if (dev
->flags
!= old_flags
) {
2602 printk(KERN_INFO
"device %s %s promiscuous mode\n",
2603 dev
->name
, (dev
->flags
& IFF_PROMISC
) ? "entered" :
2605 audit_log(current
->audit_context
, GFP_ATOMIC
,
2606 AUDIT_ANOM_PROMISCUOUS
,
2607 "dev=%s prom=%d old_prom=%d auid=%u",
2608 dev
->name
, (dev
->flags
& IFF_PROMISC
),
2609 (old_flags
& IFF_PROMISC
),
2610 audit_get_loginuid(current
->audit_context
));
2612 if (dev
->change_rx_flags
)
2613 dev
->change_rx_flags(dev
, IFF_PROMISC
);
2618 * dev_set_promiscuity - update promiscuity count on a device
2622 * Add or remove promiscuity from a device. While the count in the device
2623 * remains above zero the interface remains promiscuous. Once it hits zero
2624 * the device reverts back to normal filtering operation. A negative inc
2625 * value is used to drop promiscuity on the device.
2627 void dev_set_promiscuity(struct net_device
*dev
, int inc
)
2629 unsigned short old_flags
= dev
->flags
;
2631 __dev_set_promiscuity(dev
, inc
);
2632 if (dev
->flags
!= old_flags
)
2633 dev_set_rx_mode(dev
);
2637 * dev_set_allmulti - update allmulti count on a device
2641 * Add or remove reception of all multicast frames to a device. While the
2642 * count in the device remains above zero the interface remains listening
2643 * to all interfaces. Once it hits zero the device reverts back to normal
2644 * filtering operation. A negative @inc value is used to drop the counter
2645 * when releasing a resource needing all multicasts.
2648 void dev_set_allmulti(struct net_device
*dev
, int inc
)
2650 unsigned short old_flags
= dev
->flags
;
2654 dev
->flags
|= IFF_ALLMULTI
;
2655 if ((dev
->allmulti
+= inc
) == 0)
2656 dev
->flags
&= ~IFF_ALLMULTI
;
2657 if (dev
->flags
^ old_flags
) {
2658 if (dev
->change_rx_flags
)
2659 dev
->change_rx_flags(dev
, IFF_ALLMULTI
);
2660 dev_set_rx_mode(dev
);
2665 * Upload unicast and multicast address lists to device and
2666 * configure RX filtering. When the device doesn't support unicast
2667 * filtering it is put in promiscous mode while unicast addresses
2670 void __dev_set_rx_mode(struct net_device
*dev
)
2672 /* dev_open will call this function so the list will stay sane. */
2673 if (!(dev
->flags
&IFF_UP
))
2676 if (!netif_device_present(dev
))
2679 if (dev
->set_rx_mode
)
2680 dev
->set_rx_mode(dev
);
2682 /* Unicast addresses changes may only happen under the rtnl,
2683 * therefore calling __dev_set_promiscuity here is safe.
2685 if (dev
->uc_count
> 0 && !dev
->uc_promisc
) {
2686 __dev_set_promiscuity(dev
, 1);
2687 dev
->uc_promisc
= 1;
2688 } else if (dev
->uc_count
== 0 && dev
->uc_promisc
) {
2689 __dev_set_promiscuity(dev
, -1);
2690 dev
->uc_promisc
= 0;
2693 if (dev
->set_multicast_list
)
2694 dev
->set_multicast_list(dev
);
2698 void dev_set_rx_mode(struct net_device
*dev
)
2700 netif_tx_lock_bh(dev
);
2701 __dev_set_rx_mode(dev
);
2702 netif_tx_unlock_bh(dev
);
2705 int __dev_addr_delete(struct dev_addr_list
**list
, int *count
,
2706 void *addr
, int alen
, int glbl
)
2708 struct dev_addr_list
*da
;
2710 for (; (da
= *list
) != NULL
; list
= &da
->next
) {
2711 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2712 alen
== da
->da_addrlen
) {
2714 int old_glbl
= da
->da_gusers
;
2731 int __dev_addr_add(struct dev_addr_list
**list
, int *count
,
2732 void *addr
, int alen
, int glbl
)
2734 struct dev_addr_list
*da
;
2736 for (da
= *list
; da
!= NULL
; da
= da
->next
) {
2737 if (memcmp(da
->da_addr
, addr
, da
->da_addrlen
) == 0 &&
2738 da
->da_addrlen
== alen
) {
2740 int old_glbl
= da
->da_gusers
;
2750 da
= kmalloc(sizeof(*da
), GFP_ATOMIC
);
2753 memcpy(da
->da_addr
, addr
, alen
);
2754 da
->da_addrlen
= alen
;
2756 da
->da_gusers
= glbl
? 1 : 0;
2764 * dev_unicast_delete - Release secondary unicast address.
2766 * @addr: address to delete
2767 * @alen: length of @addr
2769 * Release reference to a secondary unicast address and remove it
2770 * from the device if the reference count drops to zero.
2772 * The caller must hold the rtnl_mutex.
2774 int dev_unicast_delete(struct net_device
*dev
, void *addr
, int alen
)
2780 netif_tx_lock_bh(dev
);
2781 err
= __dev_addr_delete(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
2783 __dev_set_rx_mode(dev
);
2784 netif_tx_unlock_bh(dev
);
2787 EXPORT_SYMBOL(dev_unicast_delete
);
2790 * dev_unicast_add - add a secondary unicast address
2792 * @addr: address to delete
2793 * @alen: length of @addr
2795 * Add a secondary unicast address to the device or increase
2796 * the reference count if it already exists.
2798 * The caller must hold the rtnl_mutex.
2800 int dev_unicast_add(struct net_device
*dev
, void *addr
, int alen
)
2806 netif_tx_lock_bh(dev
);
2807 err
= __dev_addr_add(&dev
->uc_list
, &dev
->uc_count
, addr
, alen
, 0);
2809 __dev_set_rx_mode(dev
);
2810 netif_tx_unlock_bh(dev
);
2813 EXPORT_SYMBOL(dev_unicast_add
);
2815 static void __dev_addr_discard(struct dev_addr_list
**list
)
2817 struct dev_addr_list
*tmp
;
2819 while (*list
!= NULL
) {
2822 if (tmp
->da_users
> tmp
->da_gusers
)
2823 printk("__dev_addr_discard: address leakage! "
2824 "da_users=%d\n", tmp
->da_users
);
2829 static void dev_addr_discard(struct net_device
*dev
)
2831 netif_tx_lock_bh(dev
);
2833 __dev_addr_discard(&dev
->uc_list
);
2836 __dev_addr_discard(&dev
->mc_list
);
2839 netif_tx_unlock_bh(dev
);
2842 unsigned dev_get_flags(const struct net_device
*dev
)
2846 flags
= (dev
->flags
& ~(IFF_PROMISC
|
2851 (dev
->gflags
& (IFF_PROMISC
|
2854 if (netif_running(dev
)) {
2855 if (netif_oper_up(dev
))
2856 flags
|= IFF_RUNNING
;
2857 if (netif_carrier_ok(dev
))
2858 flags
|= IFF_LOWER_UP
;
2859 if (netif_dormant(dev
))
2860 flags
|= IFF_DORMANT
;
2866 int dev_change_flags(struct net_device
*dev
, unsigned flags
)
2869 int old_flags
= dev
->flags
;
2874 * Set the flags on our device.
2877 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
2878 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
2880 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
2884 * Load in the correct multicast list now the flags have changed.
2887 if (dev
->change_rx_flags
&& (dev
->flags
^ flags
) & IFF_MULTICAST
)
2888 dev
->change_rx_flags(dev
, IFF_MULTICAST
);
2890 dev_set_rx_mode(dev
);
2893 * Have we downed the interface. We handle IFF_UP ourselves
2894 * according to user attempts to set it, rather than blindly
2899 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
2900 ret
= ((old_flags
& IFF_UP
) ? dev_close
: dev_open
)(dev
);
2903 dev_set_rx_mode(dev
);
2906 if (dev
->flags
& IFF_UP
&&
2907 ((old_flags
^ dev
->flags
) &~ (IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
|
2909 raw_notifier_call_chain(&netdev_chain
,
2910 NETDEV_CHANGE
, dev
);
2912 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
2913 int inc
= (flags
& IFF_PROMISC
) ? +1 : -1;
2914 dev
->gflags
^= IFF_PROMISC
;
2915 dev_set_promiscuity(dev
, inc
);
2918 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2919 is important. Some (broken) drivers set IFF_PROMISC, when
2920 IFF_ALLMULTI is requested not asking us and not reporting.
2922 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
2923 int inc
= (flags
& IFF_ALLMULTI
) ? +1 : -1;
2924 dev
->gflags
^= IFF_ALLMULTI
;
2925 dev_set_allmulti(dev
, inc
);
2928 /* Exclude state transition flags, already notified */
2929 changes
= (old_flags
^ dev
->flags
) & ~(IFF_UP
| IFF_RUNNING
);
2931 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
2936 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
2940 if (new_mtu
== dev
->mtu
)
2943 /* MTU must be positive. */
2947 if (!netif_device_present(dev
))
2951 if (dev
->change_mtu
)
2952 err
= dev
->change_mtu(dev
, new_mtu
);
2955 if (!err
&& dev
->flags
& IFF_UP
)
2956 raw_notifier_call_chain(&netdev_chain
,
2957 NETDEV_CHANGEMTU
, dev
);
2961 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
2965 if (!dev
->set_mac_address
)
2967 if (sa
->sa_family
!= dev
->type
)
2969 if (!netif_device_present(dev
))
2971 err
= dev
->set_mac_address(dev
, sa
);
2973 raw_notifier_call_chain(&netdev_chain
,
2974 NETDEV_CHANGEADDR
, dev
);
2979 * Perform the SIOCxIFxxx calls.
2981 static int dev_ifsioc(struct ifreq
*ifr
, unsigned int cmd
)
2984 struct net_device
*dev
= __dev_get_by_name(ifr
->ifr_name
);
2990 case SIOCGIFFLAGS
: /* Get interface flags */
2991 ifr
->ifr_flags
= dev_get_flags(dev
);
2994 case SIOCSIFFLAGS
: /* Set interface flags */
2995 return dev_change_flags(dev
, ifr
->ifr_flags
);
2997 case SIOCGIFMETRIC
: /* Get the metric on the interface
2998 (currently unused) */
2999 ifr
->ifr_metric
= 0;
3002 case SIOCSIFMETRIC
: /* Set the metric on the interface
3003 (currently unused) */
3006 case SIOCGIFMTU
: /* Get the MTU of a device */
3007 ifr
->ifr_mtu
= dev
->mtu
;
3010 case SIOCSIFMTU
: /* Set the MTU of a device */
3011 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
3015 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
3017 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
3018 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3019 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
3023 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
3025 case SIOCSIFHWBROADCAST
:
3026 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
3028 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
3029 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
3030 raw_notifier_call_chain(&netdev_chain
,
3031 NETDEV_CHANGEADDR
, dev
);
3035 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
3036 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
3037 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
3038 ifr
->ifr_map
.irq
= dev
->irq
;
3039 ifr
->ifr_map
.dma
= dev
->dma
;
3040 ifr
->ifr_map
.port
= dev
->if_port
;
3044 if (dev
->set_config
) {
3045 if (!netif_device_present(dev
))
3047 return dev
->set_config(dev
, &ifr
->ifr_map
);
3052 if (!dev
->set_multicast_list
||
3053 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3055 if (!netif_device_present(dev
))
3057 return dev_mc_add(dev
, ifr
->ifr_hwaddr
.sa_data
,
3061 if (!dev
->set_multicast_list
||
3062 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
3064 if (!netif_device_present(dev
))
3066 return dev_mc_delete(dev
, ifr
->ifr_hwaddr
.sa_data
,
3070 ifr
->ifr_ifindex
= dev
->ifindex
;
3074 ifr
->ifr_qlen
= dev
->tx_queue_len
;
3078 if (ifr
->ifr_qlen
< 0)
3080 dev
->tx_queue_len
= ifr
->ifr_qlen
;
3084 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
3085 return dev_change_name(dev
, ifr
->ifr_newname
);
3088 * Unknown or private ioctl
3092 if ((cmd
>= SIOCDEVPRIVATE
&&
3093 cmd
<= SIOCDEVPRIVATE
+ 15) ||
3094 cmd
== SIOCBONDENSLAVE
||
3095 cmd
== SIOCBONDRELEASE
||
3096 cmd
== SIOCBONDSETHWADDR
||
3097 cmd
== SIOCBONDSLAVEINFOQUERY
||
3098 cmd
== SIOCBONDINFOQUERY
||
3099 cmd
== SIOCBONDCHANGEACTIVE
||
3100 cmd
== SIOCGMIIPHY
||
3101 cmd
== SIOCGMIIREG
||
3102 cmd
== SIOCSMIIREG
||
3103 cmd
== SIOCBRADDIF
||
3104 cmd
== SIOCBRDELIF
||
3105 cmd
== SIOCWANDEV
) {
3107 if (dev
->do_ioctl
) {
3108 if (netif_device_present(dev
))
3109 err
= dev
->do_ioctl(dev
, ifr
,
3122 * This function handles all "interface"-type I/O control requests. The actual
3123 * 'doing' part of this is dev_ifsioc above.
3127 * dev_ioctl - network device ioctl
3128 * @cmd: command to issue
3129 * @arg: pointer to a struct ifreq in user space
3131 * Issue ioctl functions to devices. This is normally called by the
3132 * user space syscall interfaces but can sometimes be useful for
3133 * other purposes. The return value is the return from the syscall if
3134 * positive or a negative errno code on error.
3137 int dev_ioctl(unsigned int cmd
, void __user
*arg
)
3143 /* One special case: SIOCGIFCONF takes ifconf argument
3144 and requires shared lock, because it sleeps writing
3148 if (cmd
== SIOCGIFCONF
) {
3150 ret
= dev_ifconf((char __user
*) arg
);
3154 if (cmd
== SIOCGIFNAME
)
3155 return dev_ifname((struct ifreq __user
*)arg
);
3157 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3160 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
3162 colon
= strchr(ifr
.ifr_name
, ':');
3167 * See which interface the caller is talking about.
3172 * These ioctl calls:
3173 * - can be done by all.
3174 * - atomic and do not require locking.
3185 dev_load(ifr
.ifr_name
);
3186 read_lock(&dev_base_lock
);
3187 ret
= dev_ifsioc(&ifr
, cmd
);
3188 read_unlock(&dev_base_lock
);
3192 if (copy_to_user(arg
, &ifr
,
3193 sizeof(struct ifreq
)))
3199 dev_load(ifr
.ifr_name
);
3201 ret
= dev_ethtool(&ifr
);
3206 if (copy_to_user(arg
, &ifr
,
3207 sizeof(struct ifreq
)))
3213 * These ioctl calls:
3214 * - require superuser power.
3215 * - require strict serialization.
3221 if (!capable(CAP_NET_ADMIN
))
3223 dev_load(ifr
.ifr_name
);
3225 ret
= dev_ifsioc(&ifr
, cmd
);
3230 if (copy_to_user(arg
, &ifr
,
3231 sizeof(struct ifreq
)))
3237 * These ioctl calls:
3238 * - require superuser power.
3239 * - require strict serialization.
3240 * - do not return a value
3250 case SIOCSIFHWBROADCAST
:
3253 case SIOCBONDENSLAVE
:
3254 case SIOCBONDRELEASE
:
3255 case SIOCBONDSETHWADDR
:
3256 case SIOCBONDCHANGEACTIVE
:
3259 if (!capable(CAP_NET_ADMIN
))
3262 case SIOCBONDSLAVEINFOQUERY
:
3263 case SIOCBONDINFOQUERY
:
3264 dev_load(ifr
.ifr_name
);
3266 ret
= dev_ifsioc(&ifr
, cmd
);
3271 /* Get the per device memory space. We can add this but
3272 * currently do not support it */
3274 /* Set the per device memory buffer space.
3275 * Not applicable in our case */
3280 * Unknown or private ioctl.
3283 if (cmd
== SIOCWANDEV
||
3284 (cmd
>= SIOCDEVPRIVATE
&&
3285 cmd
<= SIOCDEVPRIVATE
+ 15)) {
3286 dev_load(ifr
.ifr_name
);
3288 ret
= dev_ifsioc(&ifr
, cmd
);
3290 if (!ret
&& copy_to_user(arg
, &ifr
,
3291 sizeof(struct ifreq
)))
3295 /* Take care of Wireless Extensions */
3296 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
3297 return wext_handle_ioctl(&ifr
, cmd
, arg
);
3304 * dev_new_index - allocate an ifindex
3306 * Returns a suitable unique value for a new device interface
3307 * number. The caller must hold the rtnl semaphore or the
3308 * dev_base_lock to be sure it remains unique.
3310 static int dev_new_index(void)
3316 if (!__dev_get_by_index(ifindex
))
3321 static int dev_boot_phase
= 1;
3323 /* Delayed registration/unregisteration */
3324 static DEFINE_SPINLOCK(net_todo_list_lock
);
3325 static struct list_head net_todo_list
= LIST_HEAD_INIT(net_todo_list
);
3327 static void net_set_todo(struct net_device
*dev
)
3329 spin_lock(&net_todo_list_lock
);
3330 list_add_tail(&dev
->todo_list
, &net_todo_list
);
3331 spin_unlock(&net_todo_list_lock
);
3335 * register_netdevice - register a network device
3336 * @dev: device to register
3338 * Take a completed network device structure and add it to the kernel
3339 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3340 * chain. 0 is returned on success. A negative errno code is returned
3341 * on a failure to set up the device, or if the name is a duplicate.
3343 * Callers must hold the rtnl semaphore. You may want
3344 * register_netdev() instead of this.
3347 * The locking appears insufficient to guarantee two parallel registers
3348 * will not get the same name.
3351 int register_netdevice(struct net_device
*dev
)
3353 struct hlist_head
*head
;
3354 struct hlist_node
*p
;
3357 BUG_ON(dev_boot_phase
);
3362 /* When net_device's are persistent, this will be fatal. */
3363 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
3365 spin_lock_init(&dev
->queue_lock
);
3366 spin_lock_init(&dev
->_xmit_lock
);
3367 netdev_set_lockdep_class(&dev
->_xmit_lock
, dev
->type
);
3368 dev
->xmit_lock_owner
= -1;
3369 spin_lock_init(&dev
->ingress_lock
);
3373 /* Init, if this function is available */
3375 ret
= dev
->init(dev
);
3383 if (!dev_valid_name(dev
->name
)) {
3388 dev
->ifindex
= dev_new_index();
3389 if (dev
->iflink
== -1)
3390 dev
->iflink
= dev
->ifindex
;
3392 /* Check for existence of name */
3393 head
= dev_name_hash(dev
->name
);
3394 hlist_for_each(p
, head
) {
3395 struct net_device
*d
3396 = hlist_entry(p
, struct net_device
, name_hlist
);
3397 if (!strncmp(d
->name
, dev
->name
, IFNAMSIZ
)) {
3403 /* Fix illegal checksum combinations */
3404 if ((dev
->features
& NETIF_F_HW_CSUM
) &&
3405 (dev
->features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3406 printk(KERN_NOTICE
"%s: mixed HW and IP checksum settings.\n",
3408 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
3411 if ((dev
->features
& NETIF_F_NO_CSUM
) &&
3412 (dev
->features
& (NETIF_F_HW_CSUM
|NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
3413 printk(KERN_NOTICE
"%s: mixed no checksumming and other settings.\n",
3415 dev
->features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
|NETIF_F_HW_CSUM
);
3419 /* Fix illegal SG+CSUM combinations. */
3420 if ((dev
->features
& NETIF_F_SG
) &&
3421 !(dev
->features
& NETIF_F_ALL_CSUM
)) {
3422 printk(KERN_NOTICE
"%s: Dropping NETIF_F_SG since no checksum feature.\n",
3424 dev
->features
&= ~NETIF_F_SG
;
3427 /* TSO requires that SG is present as well. */
3428 if ((dev
->features
& NETIF_F_TSO
) &&
3429 !(dev
->features
& NETIF_F_SG
)) {
3430 printk(KERN_NOTICE
"%s: Dropping NETIF_F_TSO since no SG feature.\n",
3432 dev
->features
&= ~NETIF_F_TSO
;
3434 if (dev
->features
& NETIF_F_UFO
) {
3435 if (!(dev
->features
& NETIF_F_HW_CSUM
)) {
3436 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3437 "NETIF_F_HW_CSUM feature.\n",
3439 dev
->features
&= ~NETIF_F_UFO
;
3441 if (!(dev
->features
& NETIF_F_SG
)) {
3442 printk(KERN_ERR
"%s: Dropping NETIF_F_UFO since no "
3443 "NETIF_F_SG feature.\n",
3445 dev
->features
&= ~NETIF_F_UFO
;
3450 * nil rebuild_header routine,
3451 * that should be never called and used as just bug trap.
3454 if (!dev
->rebuild_header
)
3455 dev
->rebuild_header
= default_rebuild_header
;
3457 ret
= netdev_register_sysfs(dev
);
3460 dev
->reg_state
= NETREG_REGISTERED
;
3463 * Default initial state at registry is that the
3464 * device is present.
3467 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
3469 dev_init_scheduler(dev
);
3470 write_lock_bh(&dev_base_lock
);
3471 list_add_tail(&dev
->dev_list
, &dev_base_head
);
3472 hlist_add_head(&dev
->name_hlist
, head
);
3473 hlist_add_head(&dev
->index_hlist
, dev_index_hash(dev
->ifindex
));
3475 write_unlock_bh(&dev_base_lock
);
3477 /* Notify protocols, that a new device appeared. */
3478 ret
= raw_notifier_call_chain(&netdev_chain
, NETDEV_REGISTER
, dev
);
3479 ret
= notifier_to_errno(ret
);
3481 unregister_netdevice(dev
);
3493 * register_netdev - register a network device
3494 * @dev: device to register
3496 * Take a completed network device structure and add it to the kernel
3497 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3498 * chain. 0 is returned on success. A negative errno code is returned
3499 * on a failure to set up the device, or if the name is a duplicate.
3501 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3502 * and expands the device name if you passed a format string to
3505 int register_netdev(struct net_device
*dev
)
3512 * If the name is a format string the caller wants us to do a
3515 if (strchr(dev
->name
, '%')) {
3516 err
= dev_alloc_name(dev
, dev
->name
);
3521 err
= register_netdevice(dev
);
3526 EXPORT_SYMBOL(register_netdev
);
3529 * netdev_wait_allrefs - wait until all references are gone.
3531 * This is called when unregistering network devices.
3533 * Any protocol or device that holds a reference should register
3534 * for netdevice notification, and cleanup and put back the
3535 * reference if they receive an UNREGISTER event.
3536 * We can get stuck here if buggy protocols don't correctly
3539 static void netdev_wait_allrefs(struct net_device
*dev
)
3541 unsigned long rebroadcast_time
, warning_time
;
3543 rebroadcast_time
= warning_time
= jiffies
;
3544 while (atomic_read(&dev
->refcnt
) != 0) {
3545 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
3548 /* Rebroadcast unregister notification */
3549 raw_notifier_call_chain(&netdev_chain
,
3550 NETDEV_UNREGISTER
, dev
);
3552 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
3554 /* We must not have linkwatch events
3555 * pending on unregister. If this
3556 * happens, we simply run the queue
3557 * unscheduled, resulting in a noop
3560 linkwatch_run_queue();
3565 rebroadcast_time
= jiffies
;
3570 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
3571 printk(KERN_EMERG
"unregister_netdevice: "
3572 "waiting for %s to become free. Usage "
3574 dev
->name
, atomic_read(&dev
->refcnt
));
3575 warning_time
= jiffies
;
3584 * register_netdevice(x1);
3585 * register_netdevice(x2);
3587 * unregister_netdevice(y1);
3588 * unregister_netdevice(y2);
3594 * We are invoked by rtnl_unlock() after it drops the semaphore.
3595 * This allows us to deal with problems:
3596 * 1) We can delete sysfs objects which invoke hotplug
3597 * without deadlocking with linkwatch via keventd.
3598 * 2) Since we run with the RTNL semaphore not held, we can sleep
3599 * safely in order to wait for the netdev refcnt to drop to zero.
3601 static DEFINE_MUTEX(net_todo_run_mutex
);
3602 void netdev_run_todo(void)
3604 struct list_head list
;
3606 /* Need to guard against multiple cpu's getting out of order. */
3607 mutex_lock(&net_todo_run_mutex
);
3609 /* Not safe to do outside the semaphore. We must not return
3610 * until all unregister events invoked by the local processor
3611 * have been completed (either by this todo run, or one on
3614 if (list_empty(&net_todo_list
))
3617 /* Snapshot list, allow later requests */
3618 spin_lock(&net_todo_list_lock
);
3619 list_replace_init(&net_todo_list
, &list
);
3620 spin_unlock(&net_todo_list_lock
);
3622 while (!list_empty(&list
)) {
3623 struct net_device
*dev
3624 = list_entry(list
.next
, struct net_device
, todo_list
);
3625 list_del(&dev
->todo_list
);
3627 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
3628 printk(KERN_ERR
"network todo '%s' but state %d\n",
3629 dev
->name
, dev
->reg_state
);
3634 dev
->reg_state
= NETREG_UNREGISTERED
;
3636 netdev_wait_allrefs(dev
);
3639 BUG_ON(atomic_read(&dev
->refcnt
));
3640 BUG_TRAP(!dev
->ip_ptr
);
3641 BUG_TRAP(!dev
->ip6_ptr
);
3642 BUG_TRAP(!dev
->dn_ptr
);
3644 if (dev
->destructor
)
3645 dev
->destructor(dev
);
3647 /* Free network device */
3648 kobject_put(&dev
->dev
.kobj
);
3652 mutex_unlock(&net_todo_run_mutex
);
3655 static struct net_device_stats
*internal_stats(struct net_device
*dev
)
3661 * alloc_netdev_mq - allocate network device
3662 * @sizeof_priv: size of private data to allocate space for
3663 * @name: device name format string
3664 * @setup: callback to initialize device
3665 * @queue_count: the number of subqueues to allocate
3667 * Allocates a struct net_device with private data area for driver use
3668 * and performs basic initialization. Also allocates subquue structs
3669 * for each queue on the device at the end of the netdevice.
3671 struct net_device
*alloc_netdev_mq(int sizeof_priv
, const char *name
,
3672 void (*setup
)(struct net_device
*), unsigned int queue_count
)
3675 struct net_device
*dev
;
3678 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
3680 /* ensure 32-byte alignment of both the device and private area */
3681 alloc_size
= (sizeof(*dev
) + NETDEV_ALIGN_CONST
+
3682 (sizeof(struct net_device_subqueue
) * (queue_count
- 1))) &
3683 ~NETDEV_ALIGN_CONST
;
3684 alloc_size
+= sizeof_priv
+ NETDEV_ALIGN_CONST
;
3686 p
= kzalloc(alloc_size
, GFP_KERNEL
);
3688 printk(KERN_ERR
"alloc_netdev: Unable to allocate device.\n");
3692 dev
= (struct net_device
*)
3693 (((long)p
+ NETDEV_ALIGN_CONST
) & ~NETDEV_ALIGN_CONST
);
3694 dev
->padded
= (char *)dev
- (char *)p
;
3697 dev
->priv
= ((char *)dev
+
3698 ((sizeof(struct net_device
) +
3699 (sizeof(struct net_device_subqueue
) *
3700 (queue_count
- 1)) + NETDEV_ALIGN_CONST
)
3701 & ~NETDEV_ALIGN_CONST
));
3704 dev
->egress_subqueue_count
= queue_count
;
3706 dev
->get_stats
= internal_stats
;
3708 strcpy(dev
->name
, name
);
3711 EXPORT_SYMBOL(alloc_netdev_mq
);
3714 * free_netdev - free network device
3717 * This function does the last stage of destroying an allocated device
3718 * interface. The reference to the device object is released.
3719 * If this is the last reference then it will be freed.
3721 void free_netdev(struct net_device
*dev
)
3724 /* Compatibility with error handling in drivers */
3725 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3726 kfree((char *)dev
- dev
->padded
);
3730 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
3731 dev
->reg_state
= NETREG_RELEASED
;
3733 /* will free via device release */
3734 put_device(&dev
->dev
);
3736 kfree((char *)dev
- dev
->padded
);
3740 /* Synchronize with packet receive processing. */
3741 void synchronize_net(void)
3748 * unregister_netdevice - remove device from the kernel
3751 * This function shuts down a device interface and removes it
3752 * from the kernel tables. On success 0 is returned, on a failure
3753 * a negative errno code is returned.
3755 * Callers must hold the rtnl semaphore. You may want
3756 * unregister_netdev() instead of this.
3759 void unregister_netdevice(struct net_device
*dev
)
3761 BUG_ON(dev_boot_phase
);
3764 /* Some devices call without registering for initialization unwind. */
3765 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
3766 printk(KERN_DEBUG
"unregister_netdevice: device %s/%p never "
3767 "was registered\n", dev
->name
, dev
);
3773 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
3775 /* If device is running, close it first. */
3776 if (dev
->flags
& IFF_UP
)
3779 /* And unlink it from device chain. */
3780 write_lock_bh(&dev_base_lock
);
3781 list_del(&dev
->dev_list
);
3782 hlist_del(&dev
->name_hlist
);
3783 hlist_del(&dev
->index_hlist
);
3784 write_unlock_bh(&dev_base_lock
);
3786 dev
->reg_state
= NETREG_UNREGISTERING
;
3790 /* Shutdown queueing discipline. */
3794 /* Notify protocols, that we are about to destroy
3795 this device. They should clean all the things.
3797 raw_notifier_call_chain(&netdev_chain
, NETDEV_UNREGISTER
, dev
);
3800 * Flush the unicast and multicast chains
3802 dev_addr_discard(dev
);
3807 /* Notifier chain MUST detach us from master device. */
3808 BUG_TRAP(!dev
->master
);
3810 /* Remove entries from sysfs */
3811 netdev_unregister_sysfs(dev
);
3813 /* Finish processing unregister after unlock */
3822 * unregister_netdev - remove device from the kernel
3825 * This function shuts down a device interface and removes it
3826 * from the kernel tables. On success 0 is returned, on a failure
3827 * a negative errno code is returned.
3829 * This is just a wrapper for unregister_netdevice that takes
3830 * the rtnl semaphore. In general you want to use this and not
3831 * unregister_netdevice.
3833 void unregister_netdev(struct net_device
*dev
)
3836 unregister_netdevice(dev
);
3840 EXPORT_SYMBOL(unregister_netdev
);
3842 static int dev_cpu_callback(struct notifier_block
*nfb
,
3843 unsigned long action
,
3846 struct sk_buff
**list_skb
;
3847 struct net_device
**list_net
;
3848 struct sk_buff
*skb
;
3849 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
3850 struct softnet_data
*sd
, *oldsd
;
3852 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
3855 local_irq_disable();
3856 cpu
= smp_processor_id();
3857 sd
= &per_cpu(softnet_data
, cpu
);
3858 oldsd
= &per_cpu(softnet_data
, oldcpu
);
3860 /* Find end of our completion_queue. */
3861 list_skb
= &sd
->completion_queue
;
3863 list_skb
= &(*list_skb
)->next
;
3864 /* Append completion queue from offline CPU. */
3865 *list_skb
= oldsd
->completion_queue
;
3866 oldsd
->completion_queue
= NULL
;
3868 /* Find end of our output_queue. */
3869 list_net
= &sd
->output_queue
;
3871 list_net
= &(*list_net
)->next_sched
;
3872 /* Append output queue from offline CPU. */
3873 *list_net
= oldsd
->output_queue
;
3874 oldsd
->output_queue
= NULL
;
3876 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
3879 /* Process offline CPU's input_pkt_queue */
3880 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
)))
3886 #ifdef CONFIG_NET_DMA
3888 * net_dma_rebalance - try to maintain one DMA channel per CPU
3889 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
3891 * This is called when the number of channels allocated to the net_dma client
3892 * changes. The net_dma client tries to have one DMA channel per CPU.
3895 static void net_dma_rebalance(struct net_dma
*net_dma
)
3897 unsigned int cpu
, i
, n
, chan_idx
;
3898 struct dma_chan
*chan
;
3900 if (cpus_empty(net_dma
->channel_mask
)) {
3901 for_each_online_cpu(cpu
)
3902 rcu_assign_pointer(per_cpu(softnet_data
, cpu
).net_dma
, NULL
);
3907 cpu
= first_cpu(cpu_online_map
);
3909 for_each_cpu_mask(chan_idx
, net_dma
->channel_mask
) {
3910 chan
= net_dma
->channels
[chan_idx
];
3912 n
= ((num_online_cpus() / cpus_weight(net_dma
->channel_mask
))
3913 + (i
< (num_online_cpus() %
3914 cpus_weight(net_dma
->channel_mask
)) ? 1 : 0));
3917 per_cpu(softnet_data
, cpu
).net_dma
= chan
;
3918 cpu
= next_cpu(cpu
, cpu_online_map
);
3926 * netdev_dma_event - event callback for the net_dma_client
3927 * @client: should always be net_dma_client
3928 * @chan: DMA channel for the event
3929 * @state: DMA state to be handled
3931 static enum dma_state_client
3932 netdev_dma_event(struct dma_client
*client
, struct dma_chan
*chan
,
3933 enum dma_state state
)
3935 int i
, found
= 0, pos
= -1;
3936 struct net_dma
*net_dma
=
3937 container_of(client
, struct net_dma
, client
);
3938 enum dma_state_client ack
= DMA_DUP
; /* default: take no action */
3940 spin_lock(&net_dma
->lock
);
3942 case DMA_RESOURCE_AVAILABLE
:
3943 for (i
= 0; i
< NR_CPUS
; i
++)
3944 if (net_dma
->channels
[i
] == chan
) {
3947 } else if (net_dma
->channels
[i
] == NULL
&& pos
< 0)
3950 if (!found
&& pos
>= 0) {
3952 net_dma
->channels
[pos
] = chan
;
3953 cpu_set(pos
, net_dma
->channel_mask
);
3954 net_dma_rebalance(net_dma
);
3957 case DMA_RESOURCE_REMOVED
:
3958 for (i
= 0; i
< NR_CPUS
; i
++)
3959 if (net_dma
->channels
[i
] == chan
) {
3967 cpu_clear(pos
, net_dma
->channel_mask
);
3968 net_dma
->channels
[i
] = NULL
;
3969 net_dma_rebalance(net_dma
);
3975 spin_unlock(&net_dma
->lock
);
3981 * netdev_dma_regiser - register the networking subsystem as a DMA client
3983 static int __init
netdev_dma_register(void)
3985 spin_lock_init(&net_dma
.lock
);
3986 dma_cap_set(DMA_MEMCPY
, net_dma
.client
.cap_mask
);
3987 dma_async_client_register(&net_dma
.client
);
3988 dma_async_client_chan_request(&net_dma
.client
);
3993 static int __init
netdev_dma_register(void) { return -ENODEV
; }
3994 #endif /* CONFIG_NET_DMA */
3997 * Initialize the DEV module. At boot time this walks the device list and
3998 * unhooks any devices that fail to initialise (normally hardware not
3999 * present) and leaves us with a valid list of present and active devices.
4004 * This is called single threaded during boot, so no need
4005 * to take the rtnl semaphore.
4007 static int __init
net_dev_init(void)
4009 int i
, rc
= -ENOMEM
;
4011 BUG_ON(!dev_boot_phase
);
4013 if (dev_proc_init())
4016 if (netdev_sysfs_init())
4019 INIT_LIST_HEAD(&ptype_all
);
4020 for (i
= 0; i
< 16; i
++)
4021 INIT_LIST_HEAD(&ptype_base
[i
]);
4023 for (i
= 0; i
< ARRAY_SIZE(dev_name_head
); i
++)
4024 INIT_HLIST_HEAD(&dev_name_head
[i
]);
4026 for (i
= 0; i
< ARRAY_SIZE(dev_index_head
); i
++)
4027 INIT_HLIST_HEAD(&dev_index_head
[i
]);
4030 * Initialise the packet receive queues.
4033 for_each_possible_cpu(i
) {
4034 struct softnet_data
*queue
;
4036 queue
= &per_cpu(softnet_data
, i
);
4037 skb_queue_head_init(&queue
->input_pkt_queue
);
4038 queue
->completion_queue
= NULL
;
4039 INIT_LIST_HEAD(&queue
->poll_list
);
4040 set_bit(__LINK_STATE_START
, &queue
->backlog_dev
.state
);
4041 queue
->backlog_dev
.weight
= weight_p
;
4042 queue
->backlog_dev
.poll
= process_backlog
;
4043 atomic_set(&queue
->backlog_dev
.refcnt
, 1);
4046 netdev_dma_register();
4050 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
, NULL
);
4051 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
, NULL
);
4053 hotcpu_notifier(dev_cpu_callback
, 0);
4061 subsys_initcall(net_dev_init
);
4063 EXPORT_SYMBOL(__dev_get_by_index
);
4064 EXPORT_SYMBOL(__dev_get_by_name
);
4065 EXPORT_SYMBOL(__dev_remove_pack
);
4066 EXPORT_SYMBOL(dev_valid_name
);
4067 EXPORT_SYMBOL(dev_add_pack
);
4068 EXPORT_SYMBOL(dev_alloc_name
);
4069 EXPORT_SYMBOL(dev_close
);
4070 EXPORT_SYMBOL(dev_get_by_flags
);
4071 EXPORT_SYMBOL(dev_get_by_index
);
4072 EXPORT_SYMBOL(dev_get_by_name
);
4073 EXPORT_SYMBOL(dev_open
);
4074 EXPORT_SYMBOL(dev_queue_xmit
);
4075 EXPORT_SYMBOL(dev_remove_pack
);
4076 EXPORT_SYMBOL(dev_set_allmulti
);
4077 EXPORT_SYMBOL(dev_set_promiscuity
);
4078 EXPORT_SYMBOL(dev_change_flags
);
4079 EXPORT_SYMBOL(dev_set_mtu
);
4080 EXPORT_SYMBOL(dev_set_mac_address
);
4081 EXPORT_SYMBOL(free_netdev
);
4082 EXPORT_SYMBOL(netdev_boot_setup_check
);
4083 EXPORT_SYMBOL(netdev_set_master
);
4084 EXPORT_SYMBOL(netdev_state_change
);
4085 EXPORT_SYMBOL(netif_receive_skb
);
4086 EXPORT_SYMBOL(netif_rx
);
4087 EXPORT_SYMBOL(register_gifconf
);
4088 EXPORT_SYMBOL(register_netdevice
);
4089 EXPORT_SYMBOL(register_netdevice_notifier
);
4090 EXPORT_SYMBOL(skb_checksum_help
);
4091 EXPORT_SYMBOL(synchronize_net
);
4092 EXPORT_SYMBOL(unregister_netdevice
);
4093 EXPORT_SYMBOL(unregister_netdevice_notifier
);
4094 EXPORT_SYMBOL(net_enable_timestamp
);
4095 EXPORT_SYMBOL(net_disable_timestamp
);
4096 EXPORT_SYMBOL(dev_get_flags
);
4098 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4099 EXPORT_SYMBOL(br_handle_frame_hook
);
4100 EXPORT_SYMBOL(br_fdb_get_hook
);
4101 EXPORT_SYMBOL(br_fdb_put_hook
);
4105 EXPORT_SYMBOL(dev_load
);
4108 EXPORT_PER_CPU_SYMBOL(softnet_data
);