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/ethtool.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/net_namespace.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/stat.h>
102 #include <linux/if_bridge.h>
103 #include <linux/if_macvlan.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/kallsyms.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
125 #include "net-sysfs.h"
128 * The list of packet types we will receive (as opposed to discard)
129 * and the routines to invoke.
131 * Why 16. Because with 16 the only overlap we get on a hash of the
132 * low nibble of the protocol value is RARP/SNAP/X.25.
134 * NOTE: That is no longer true with the addition of VLAN tags. Not
135 * sure which should go first, but I bet it won't make much
136 * difference if we are running VLANs. The good news is that
137 * this protocol won't be in the list unless compiled in, so
138 * the average user (w/out VLANs) will not be adversely affected.
155 #define PTYPE_HASH_SIZE (16)
156 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
158 static DEFINE_SPINLOCK(ptype_lock);
159 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
160 static struct list_head ptype_all __read_mostly; /* Taps */
162 #ifdef CONFIG_NET_DMA
164 struct dma_client client;
166 cpumask_t channel_mask;
167 struct dma_chan **channels;
170 static enum dma_state_client
171 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
172 enum dma_state state);
174 static struct net_dma net_dma = {
176 .event_callback = netdev_dma_event,
182 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
185 * Pure readers hold dev_base_lock for reading.
187 * Writers must hold the rtnl semaphore while they loop through the
188 * dev_base_head list, and hold dev_base_lock for writing when they do the
189 * actual updates. This allows pure readers to access the list even
190 * while a writer is preparing to update it.
192 * To put it another way, dev_base_lock is held for writing only to
193 * protect against pure readers; the rtnl semaphore provides the
194 * protection against other writers.
196 * See, for example usages, register_netdevice() and
197 * unregister_netdevice(), which must be called with the rtnl
200 DEFINE_RWLOCK(dev_base_lock);
202 EXPORT_SYMBOL(dev_base_lock);
204 #define NETDEV_HASHBITS 8
205 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
207 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
209 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
210 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
213 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
215 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
218 /* Device list insertion */
219 static int list_netdevice(struct net_device *dev)
221 struct net *net = dev_net(dev);
225 write_lock_bh(&dev_base_lock);
226 list_add_tail(&dev->dev_list, &net->dev_base_head);
227 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
228 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
229 write_unlock_bh(&dev_base_lock);
233 /* Device list removal */
234 static void unlist_netdevice(struct net_device *dev)
238 /* Unlink dev from the device chain */
239 write_lock_bh(&dev_base_lock);
240 list_del(&dev->dev_list);
241 hlist_del(&dev->name_hlist);
242 hlist_del(&dev->index_hlist);
243 write_unlock_bh(&dev_base_lock);
250 static RAW_NOTIFIER_HEAD(netdev_chain);
253 * Device drivers call our routines to queue packets here. We empty the
254 * queue in the local softnet handler.
257 DEFINE_PER_CPU(struct softnet_data, softnet_data);
259 #ifdef CONFIG_DEBUG_LOCK_ALLOC
261 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
262 * according to dev->type
264 static const unsigned short netdev_lock_type[] =
265 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
266 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
267 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
268 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
269 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
270 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
271 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
272 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
273 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
274 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
275 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
276 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
277 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
278 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
281 static const char *netdev_lock_name[] =
282 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
283 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
284 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
285 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
286 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
287 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
288 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
289 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
290 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
291 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
292 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
293 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
294 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
295 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
298 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
300 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
304 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
305 if (netdev_lock_type[i] == dev_type)
307 /* the last key is used by default */
308 return ARRAY_SIZE(netdev_lock_type) - 1;
311 static inline void netdev_set_lockdep_class(spinlock_t *lock,
312 unsigned short dev_type)
316 i = netdev_lock_pos(dev_type);
317 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
318 netdev_lock_name[i]);
321 static inline void netdev_set_lockdep_class(spinlock_t *lock,
322 unsigned short dev_type)
327 /*******************************************************************************
329 Protocol management and registration routines
331 *******************************************************************************/
334 * Add a protocol ID to the list. Now that the input handler is
335 * smarter we can dispense with all the messy stuff that used to be
338 * BEWARE!!! Protocol handlers, mangling input packets,
339 * MUST BE last in hash buckets and checking protocol handlers
340 * MUST start from promiscuous ptype_all chain in net_bh.
341 * It is true now, do not change it.
342 * Explanation follows: if protocol handler, mangling packet, will
343 * be the first on list, it is not able to sense, that packet
344 * is cloned and should be copied-on-write, so that it will
345 * change it and subsequent readers will get broken packet.
350 * dev_add_pack - add packet handler
351 * @pt: packet type declaration
353 * Add a protocol handler to the networking stack. The passed &packet_type
354 * is linked into kernel lists and may not be freed until it has been
355 * removed from the kernel lists.
357 * This call does not sleep therefore it can not
358 * guarantee all CPU's that are in middle of receiving packets
359 * will see the new packet type (until the next received packet).
362 void dev_add_pack(struct packet_type *pt)
366 spin_lock_bh(&ptype_lock);
367 if (pt->type == htons(ETH_P_ALL))
368 list_add_rcu(&pt->list, &ptype_all);
370 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
371 list_add_rcu(&pt->list, &ptype_base[hash]);
373 spin_unlock_bh(&ptype_lock);
377 * __dev_remove_pack - remove packet handler
378 * @pt: packet type declaration
380 * Remove a protocol handler that was previously added to the kernel
381 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
382 * from the kernel lists and can be freed or reused once this function
385 * The packet type might still be in use by receivers
386 * and must not be freed until after all the CPU's have gone
387 * through a quiescent state.
389 void __dev_remove_pack(struct packet_type *pt)
391 struct list_head *head;
392 struct packet_type *pt1;
394 spin_lock_bh(&ptype_lock);
396 if (pt->type == htons(ETH_P_ALL))
399 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
401 list_for_each_entry(pt1, head, list) {
403 list_del_rcu(&pt->list);
408 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
410 spin_unlock_bh(&ptype_lock);
413 * dev_remove_pack - remove packet handler
414 * @pt: packet type declaration
416 * Remove a protocol handler that was previously added to the kernel
417 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
418 * from the kernel lists and can be freed or reused once this function
421 * This call sleeps to guarantee that no CPU is looking at the packet
424 void dev_remove_pack(struct packet_type *pt)
426 __dev_remove_pack(pt);
431 /******************************************************************************
433 Device Boot-time Settings Routines
435 *******************************************************************************/
437 /* Boot time configuration table */
438 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
441 * netdev_boot_setup_add - add new setup entry
442 * @name: name of the device
443 * @map: configured settings for the device
445 * Adds new setup entry to the dev_boot_setup list. The function
446 * returns 0 on error and 1 on success. This is a generic routine to
449 static int netdev_boot_setup_add(char *name, struct ifmap *map)
451 struct netdev_boot_setup *s;
455 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
456 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
457 memset(s[i].name, 0, sizeof(s[i].name));
458 strlcpy(s[i].name, name, IFNAMSIZ);
459 memcpy(&s[i].map, map, sizeof(s[i].map));
464 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
468 * netdev_boot_setup_check - check boot time settings
469 * @dev: the netdevice
471 * Check boot time settings for the device.
472 * The found settings are set for the device to be used
473 * later in the device probing.
474 * Returns 0 if no settings found, 1 if they are.
476 int netdev_boot_setup_check(struct net_device *dev)
478 struct netdev_boot_setup *s = dev_boot_setup;
481 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
482 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
483 !strcmp(dev->name, s[i].name)) {
484 dev->irq = s[i].map.irq;
485 dev->base_addr = s[i].map.base_addr;
486 dev->mem_start = s[i].map.mem_start;
487 dev->mem_end = s[i].map.mem_end;
496 * netdev_boot_base - get address from boot time settings
497 * @prefix: prefix for network device
498 * @unit: id for network device
500 * Check boot time settings for the base address of device.
501 * The found settings are set for the device to be used
502 * later in the device probing.
503 * Returns 0 if no settings found.
505 unsigned long netdev_boot_base(const char *prefix, int unit)
507 const struct netdev_boot_setup *s = dev_boot_setup;
511 sprintf(name, "%s%d", prefix, unit);
514 * If device already registered then return base of 1
515 * to indicate not to probe for this interface
517 if (__dev_get_by_name(&init_net, name))
520 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
521 if (!strcmp(name, s[i].name))
522 return s[i].map.base_addr;
527 * Saves at boot time configured settings for any netdevice.
529 int __init netdev_boot_setup(char *str)
534 str = get_options(str, ARRAY_SIZE(ints), ints);
539 memset(&map, 0, sizeof(map));
543 map.base_addr = ints[2];
545 map.mem_start = ints[3];
547 map.mem_end = ints[4];
549 /* Add new entry to the list */
550 return netdev_boot_setup_add(str, &map);
553 __setup("netdev=", netdev_boot_setup);
555 /*******************************************************************************
557 Device Interface Subroutines
559 *******************************************************************************/
562 * __dev_get_by_name - find a device by its name
563 * @net: the applicable net namespace
564 * @name: name to find
566 * Find an interface by name. Must be called under RTNL semaphore
567 * or @dev_base_lock. If the name is found a pointer to the device
568 * is returned. If the name is not found then %NULL is returned. The
569 * reference counters are not incremented so the caller must be
570 * careful with locks.
573 struct net_device *__dev_get_by_name(struct net *net, const char *name)
575 struct hlist_node *p;
577 hlist_for_each(p, dev_name_hash(net, name)) {
578 struct net_device *dev
579 = hlist_entry(p, struct net_device, name_hlist);
580 if (!strncmp(dev->name, name, IFNAMSIZ))
587 * dev_get_by_name - find a device by its name
588 * @net: the applicable net namespace
589 * @name: name to find
591 * Find an interface by name. This can be called from any
592 * context and does its own locking. The returned handle has
593 * the usage count incremented and the caller must use dev_put() to
594 * release it when it is no longer needed. %NULL is returned if no
595 * matching device is found.
598 struct net_device *dev_get_by_name(struct net *net, const char *name)
600 struct net_device *dev;
602 read_lock(&dev_base_lock);
603 dev = __dev_get_by_name(net, name);
606 read_unlock(&dev_base_lock);
611 * __dev_get_by_index - find a device by its ifindex
612 * @net: the applicable net namespace
613 * @ifindex: index of device
615 * Search for an interface by index. Returns %NULL if the device
616 * is not found or a pointer to the device. The device has not
617 * had its reference counter increased so the caller must be careful
618 * about locking. The caller must hold either the RTNL semaphore
622 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
624 struct hlist_node *p;
626 hlist_for_each(p, dev_index_hash(net, ifindex)) {
627 struct net_device *dev
628 = hlist_entry(p, struct net_device, index_hlist);
629 if (dev->ifindex == ifindex)
637 * dev_get_by_index - find a device by its ifindex
638 * @net: the applicable net namespace
639 * @ifindex: index of device
641 * Search for an interface by index. Returns NULL if the device
642 * is not found or a pointer to the device. The device returned has
643 * had a reference added and the pointer is safe until the user calls
644 * dev_put to indicate they have finished with it.
647 struct net_device *dev_get_by_index(struct net *net, int ifindex)
649 struct net_device *dev;
651 read_lock(&dev_base_lock);
652 dev = __dev_get_by_index(net, ifindex);
655 read_unlock(&dev_base_lock);
660 * dev_getbyhwaddr - find a device by its hardware address
661 * @net: the applicable net namespace
662 * @type: media type of device
663 * @ha: hardware address
665 * Search for an interface by MAC address. Returns NULL if the device
666 * is not found or a pointer to the device. The caller must hold the
667 * rtnl semaphore. The returned device has not had its ref count increased
668 * and the caller must therefore be careful about locking
671 * If the API was consistent this would be __dev_get_by_hwaddr
674 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
676 struct net_device *dev;
680 for_each_netdev(net, dev)
681 if (dev->type == type &&
682 !memcmp(dev->dev_addr, ha, dev->addr_len))
688 EXPORT_SYMBOL(dev_getbyhwaddr);
690 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
692 struct net_device *dev;
695 for_each_netdev(net, dev)
696 if (dev->type == type)
702 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
704 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
706 struct net_device *dev;
709 dev = __dev_getfirstbyhwtype(net, type);
716 EXPORT_SYMBOL(dev_getfirstbyhwtype);
719 * dev_get_by_flags - find any device with given flags
720 * @net: the applicable net namespace
721 * @if_flags: IFF_* values
722 * @mask: bitmask of bits in if_flags to check
724 * Search for any interface with the given flags. Returns NULL if a device
725 * is not found or a pointer to the device. The device returned has
726 * had a reference added and the pointer is safe until the user calls
727 * dev_put to indicate they have finished with it.
730 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
732 struct net_device *dev, *ret;
735 read_lock(&dev_base_lock);
736 for_each_netdev(net, dev) {
737 if (((dev->flags ^ if_flags) & mask) == 0) {
743 read_unlock(&dev_base_lock);
748 * dev_valid_name - check if name is okay for network device
751 * Network device names need to be valid file names to
752 * to allow sysfs to work. We also disallow any kind of
755 int dev_valid_name(const char *name)
759 if (strlen(name) >= IFNAMSIZ)
761 if (!strcmp(name, ".") || !strcmp(name, ".."))
765 if (*name == '/' || isspace(*name))
773 * __dev_alloc_name - allocate a name for a device
774 * @net: network namespace to allocate the device name in
775 * @name: name format string
776 * @buf: scratch buffer and result name string
778 * Passed a format string - eg "lt%d" it will try and find a suitable
779 * id. It scans list of devices to build up a free map, then chooses
780 * the first empty slot. The caller must hold the dev_base or rtnl lock
781 * while allocating the name and adding the device in order to avoid
783 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
784 * Returns the number of the unit assigned or a negative errno code.
787 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
791 const int max_netdevices = 8*PAGE_SIZE;
792 unsigned long *inuse;
793 struct net_device *d;
795 p = strnchr(name, IFNAMSIZ-1, '%');
798 * Verify the string as this thing may have come from
799 * the user. There must be either one "%d" and no other "%"
802 if (p[1] != 'd' || strchr(p + 2, '%'))
805 /* Use one page as a bit array of possible slots */
806 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
810 for_each_netdev(net, d) {
811 if (!sscanf(d->name, name, &i))
813 if (i < 0 || i >= max_netdevices)
816 /* avoid cases where sscanf is not exact inverse of printf */
817 snprintf(buf, IFNAMSIZ, name, i);
818 if (!strncmp(buf, d->name, IFNAMSIZ))
822 i = find_first_zero_bit(inuse, max_netdevices);
823 free_page((unsigned long) inuse);
826 snprintf(buf, IFNAMSIZ, name, i);
827 if (!__dev_get_by_name(net, buf))
830 /* It is possible to run out of possible slots
831 * when the name is long and there isn't enough space left
832 * for the digits, or if all bits are used.
838 * dev_alloc_name - allocate a name for a device
840 * @name: name format string
842 * Passed a format string - eg "lt%d" it will try and find a suitable
843 * id. It scans list of devices to build up a free map, then chooses
844 * the first empty slot. The caller must hold the dev_base or rtnl lock
845 * while allocating the name and adding the device in order to avoid
847 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
848 * Returns the number of the unit assigned or a negative errno code.
851 int dev_alloc_name(struct net_device *dev, const char *name)
857 BUG_ON(!dev_net(dev));
859 ret = __dev_alloc_name(net, name, buf);
861 strlcpy(dev->name, buf, IFNAMSIZ);
867 * dev_change_name - change name of a device
869 * @newname: name (or format string) must be at least IFNAMSIZ
871 * Change name of a device, can pass format strings "eth%d".
874 int dev_change_name(struct net_device *dev, char *newname)
876 char oldname[IFNAMSIZ];
882 BUG_ON(!dev_net(dev));
885 if (dev->flags & IFF_UP)
888 if (!dev_valid_name(newname))
891 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
894 memcpy(oldname, dev->name, IFNAMSIZ);
896 if (strchr(newname, '%')) {
897 err = dev_alloc_name(dev, newname);
900 strcpy(newname, dev->name);
902 else if (__dev_get_by_name(net, newname))
905 strlcpy(dev->name, newname, IFNAMSIZ);
908 err = device_rename(&dev->dev, dev->name);
910 memcpy(dev->name, oldname, IFNAMSIZ);
914 write_lock_bh(&dev_base_lock);
915 hlist_del(&dev->name_hlist);
916 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
917 write_unlock_bh(&dev_base_lock);
919 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
920 ret = notifier_to_errno(ret);
925 "%s: name change rollback failed: %d.\n",
929 memcpy(dev->name, oldname, IFNAMSIZ);
938 * netdev_features_change - device changes features
939 * @dev: device to cause notification
941 * Called to indicate a device has changed features.
943 void netdev_features_change(struct net_device *dev)
945 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
947 EXPORT_SYMBOL(netdev_features_change);
950 * netdev_state_change - device changes state
951 * @dev: device to cause notification
953 * Called to indicate a device has changed state. This function calls
954 * the notifier chains for netdev_chain and sends a NEWLINK message
955 * to the routing socket.
957 void netdev_state_change(struct net_device *dev)
959 if (dev->flags & IFF_UP) {
960 call_netdevice_notifiers(NETDEV_CHANGE, dev);
961 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
965 void netdev_bonding_change(struct net_device *dev)
967 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
969 EXPORT_SYMBOL(netdev_bonding_change);
972 * dev_load - load a network module
973 * @net: the applicable net namespace
974 * @name: name of interface
976 * If a network interface is not present and the process has suitable
977 * privileges this function loads the module. If module loading is not
978 * available in this kernel then it becomes a nop.
981 void dev_load(struct net *net, const char *name)
983 struct net_device *dev;
985 read_lock(&dev_base_lock);
986 dev = __dev_get_by_name(net, name);
987 read_unlock(&dev_base_lock);
989 if (!dev && capable(CAP_SYS_MODULE))
990 request_module("%s", name);
994 * dev_open - prepare an interface for use.
995 * @dev: device to open
997 * Takes a device from down to up state. The device's private open
998 * function is invoked and then the multicast lists are loaded. Finally
999 * the device is moved into the up state and a %NETDEV_UP message is
1000 * sent to the netdev notifier chain.
1002 * Calling this function on an active interface is a nop. On a failure
1003 * a negative errno code is returned.
1005 int dev_open(struct net_device *dev)
1015 if (dev->flags & IFF_UP)
1019 * Is it even present?
1021 if (!netif_device_present(dev))
1025 * Call device private open method
1027 set_bit(__LINK_STATE_START, &dev->state);
1029 if (dev->validate_addr)
1030 ret = dev->validate_addr(dev);
1032 if (!ret && dev->open)
1033 ret = dev->open(dev);
1036 * If it went open OK then:
1040 clear_bit(__LINK_STATE_START, &dev->state);
1045 dev->flags |= IFF_UP;
1048 * Initialize multicasting status
1050 dev_set_rx_mode(dev);
1053 * Wakeup transmit queue engine
1058 * ... and announce new interface.
1060 call_netdevice_notifiers(NETDEV_UP, dev);
1067 * dev_close - shutdown an interface.
1068 * @dev: device to shutdown
1070 * This function moves an active device into down state. A
1071 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1072 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1075 int dev_close(struct net_device *dev)
1081 if (!(dev->flags & IFF_UP))
1085 * Tell people we are going down, so that they can
1086 * prepare to death, when device is still operating.
1088 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1090 clear_bit(__LINK_STATE_START, &dev->state);
1092 /* Synchronize to scheduled poll. We cannot touch poll list,
1093 * it can be even on different cpu. So just clear netif_running().
1095 * dev->stop() will invoke napi_disable() on all of it's
1096 * napi_struct instances on this device.
1098 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1100 dev_deactivate(dev);
1103 * Call the device specific close. This cannot fail.
1104 * Only if device is UP
1106 * We allow it to be called even after a DETACH hot-plug
1113 * Device is now down.
1116 dev->flags &= ~IFF_UP;
1119 * Tell people we are down
1121 call_netdevice_notifiers(NETDEV_DOWN, dev);
1128 * dev_disable_lro - disable Large Receive Offload on a device
1131 * Disable Large Receive Offload (LRO) on a net device. Must be
1132 * called under RTNL. This is needed if received packets may be
1133 * forwarded to another interface.
1135 void dev_disable_lro(struct net_device *dev)
1137 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1138 dev->ethtool_ops->set_flags) {
1139 u32 flags = dev->ethtool_ops->get_flags(dev);
1140 if (flags & ETH_FLAG_LRO) {
1141 flags &= ~ETH_FLAG_LRO;
1142 dev->ethtool_ops->set_flags(dev, flags);
1145 WARN_ON(dev->features & NETIF_F_LRO);
1147 EXPORT_SYMBOL(dev_disable_lro);
1150 static int dev_boot_phase = 1;
1153 * Device change register/unregister. These are not inline or static
1154 * as we export them to the world.
1158 * register_netdevice_notifier - register a network notifier block
1161 * Register a notifier to be called when network device events occur.
1162 * The notifier passed is linked into the kernel structures and must
1163 * not be reused until it has been unregistered. A negative errno code
1164 * is returned on a failure.
1166 * When registered all registration and up events are replayed
1167 * to the new notifier to allow device to have a race free
1168 * view of the network device list.
1171 int register_netdevice_notifier(struct notifier_block *nb)
1173 struct net_device *dev;
1174 struct net_device *last;
1179 err = raw_notifier_chain_register(&netdev_chain, nb);
1185 for_each_netdev(net, dev) {
1186 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1187 err = notifier_to_errno(err);
1191 if (!(dev->flags & IFF_UP))
1194 nb->notifier_call(nb, NETDEV_UP, dev);
1205 for_each_netdev(net, dev) {
1209 if (dev->flags & IFF_UP) {
1210 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1211 nb->notifier_call(nb, NETDEV_DOWN, dev);
1213 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1217 raw_notifier_chain_unregister(&netdev_chain, nb);
1222 * unregister_netdevice_notifier - unregister a network notifier block
1225 * Unregister a notifier previously registered by
1226 * register_netdevice_notifier(). The notifier is unlinked into the
1227 * kernel structures and may then be reused. A negative errno code
1228 * is returned on a failure.
1231 int unregister_netdevice_notifier(struct notifier_block *nb)
1236 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1242 * call_netdevice_notifiers - call all network notifier blocks
1243 * @val: value passed unmodified to notifier function
1244 * @dev: net_device pointer passed unmodified to notifier function
1246 * Call all network notifier blocks. Parameters and return value
1247 * are as for raw_notifier_call_chain().
1250 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1252 return raw_notifier_call_chain(&netdev_chain, val, dev);
1255 /* When > 0 there are consumers of rx skb time stamps */
1256 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1258 void net_enable_timestamp(void)
1260 atomic_inc(&netstamp_needed);
1263 void net_disable_timestamp(void)
1265 atomic_dec(&netstamp_needed);
1268 static inline void net_timestamp(struct sk_buff *skb)
1270 if (atomic_read(&netstamp_needed))
1271 __net_timestamp(skb);
1273 skb->tstamp.tv64 = 0;
1277 * Support routine. Sends outgoing frames to any network
1278 * taps currently in use.
1281 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1283 struct packet_type *ptype;
1288 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1289 /* Never send packets back to the socket
1290 * they originated from - MvS (miquels@drinkel.ow.org)
1292 if ((ptype->dev == dev || !ptype->dev) &&
1293 (ptype->af_packet_priv == NULL ||
1294 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1295 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1299 /* skb->nh should be correctly
1300 set by sender, so that the second statement is
1301 just protection against buggy protocols.
1303 skb_reset_mac_header(skb2);
1305 if (skb_network_header(skb2) < skb2->data ||
1306 skb2->network_header > skb2->tail) {
1307 if (net_ratelimit())
1308 printk(KERN_CRIT "protocol %04x is "
1310 skb2->protocol, dev->name);
1311 skb_reset_network_header(skb2);
1314 skb2->transport_header = skb2->network_header;
1315 skb2->pkt_type = PACKET_OUTGOING;
1316 ptype->func(skb2, skb->dev, ptype, skb->dev);
1323 void __netif_schedule(struct net_device *dev)
1325 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1326 struct netdev_queue *txq = &dev->tx_queue;
1327 unsigned long flags;
1328 struct softnet_data *sd;
1330 local_irq_save(flags);
1331 sd = &__get_cpu_var(softnet_data);
1332 txq->next_sched = sd->output_queue;
1333 sd->output_queue = txq;
1334 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1335 local_irq_restore(flags);
1338 EXPORT_SYMBOL(__netif_schedule);
1340 void dev_kfree_skb_irq(struct sk_buff *skb)
1342 if (atomic_dec_and_test(&skb->users)) {
1343 struct softnet_data *sd;
1344 unsigned long flags;
1346 local_irq_save(flags);
1347 sd = &__get_cpu_var(softnet_data);
1348 skb->next = sd->completion_queue;
1349 sd->completion_queue = skb;
1350 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1351 local_irq_restore(flags);
1354 EXPORT_SYMBOL(dev_kfree_skb_irq);
1356 void dev_kfree_skb_any(struct sk_buff *skb)
1358 if (in_irq() || irqs_disabled())
1359 dev_kfree_skb_irq(skb);
1363 EXPORT_SYMBOL(dev_kfree_skb_any);
1367 * netif_device_detach - mark device as removed
1368 * @dev: network device
1370 * Mark device as removed from system and therefore no longer available.
1372 void netif_device_detach(struct net_device *dev)
1374 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1375 netif_running(dev)) {
1376 netif_stop_queue(dev);
1379 EXPORT_SYMBOL(netif_device_detach);
1382 * netif_device_attach - mark device as attached
1383 * @dev: network device
1385 * Mark device as attached from system and restart if needed.
1387 void netif_device_attach(struct net_device *dev)
1389 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1390 netif_running(dev)) {
1391 netif_wake_queue(dev);
1392 __netdev_watchdog_up(dev);
1395 EXPORT_SYMBOL(netif_device_attach);
1397 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1399 return ((features & NETIF_F_GEN_CSUM) ||
1400 ((features & NETIF_F_IP_CSUM) &&
1401 protocol == htons(ETH_P_IP)) ||
1402 ((features & NETIF_F_IPV6_CSUM) &&
1403 protocol == htons(ETH_P_IPV6)));
1406 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1408 if (can_checksum_protocol(dev->features, skb->protocol))
1411 if (skb->protocol == htons(ETH_P_8021Q)) {
1412 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1413 if (can_checksum_protocol(dev->features & dev->vlan_features,
1414 veh->h_vlan_encapsulated_proto))
1422 * Invalidate hardware checksum when packet is to be mangled, and
1423 * complete checksum manually on outgoing path.
1425 int skb_checksum_help(struct sk_buff *skb)
1428 int ret = 0, offset;
1430 if (skb->ip_summed == CHECKSUM_COMPLETE)
1431 goto out_set_summed;
1433 if (unlikely(skb_shinfo(skb)->gso_size)) {
1434 /* Let GSO fix up the checksum. */
1435 goto out_set_summed;
1438 offset = skb->csum_start - skb_headroom(skb);
1439 BUG_ON(offset >= skb_headlen(skb));
1440 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1442 offset += skb->csum_offset;
1443 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1445 if (skb_cloned(skb) &&
1446 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1447 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1452 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1454 skb->ip_summed = CHECKSUM_NONE;
1460 * skb_gso_segment - Perform segmentation on skb.
1461 * @skb: buffer to segment
1462 * @features: features for the output path (see dev->features)
1464 * This function segments the given skb and returns a list of segments.
1466 * It may return NULL if the skb requires no segmentation. This is
1467 * only possible when GSO is used for verifying header integrity.
1469 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1471 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1472 struct packet_type *ptype;
1473 __be16 type = skb->protocol;
1476 BUG_ON(skb_shinfo(skb)->frag_list);
1478 skb_reset_mac_header(skb);
1479 skb->mac_len = skb->network_header - skb->mac_header;
1480 __skb_pull(skb, skb->mac_len);
1482 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1483 if (skb_header_cloned(skb) &&
1484 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1485 return ERR_PTR(err);
1489 list_for_each_entry_rcu(ptype,
1490 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1491 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1492 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1493 err = ptype->gso_send_check(skb);
1494 segs = ERR_PTR(err);
1495 if (err || skb_gso_ok(skb, features))
1497 __skb_push(skb, (skb->data -
1498 skb_network_header(skb)));
1500 segs = ptype->gso_segment(skb, features);
1506 __skb_push(skb, skb->data - skb_mac_header(skb));
1511 EXPORT_SYMBOL(skb_gso_segment);
1513 /* Take action when hardware reception checksum errors are detected. */
1515 void netdev_rx_csum_fault(struct net_device *dev)
1517 if (net_ratelimit()) {
1518 printk(KERN_ERR "%s: hw csum failure.\n",
1519 dev ? dev->name : "<unknown>");
1523 EXPORT_SYMBOL(netdev_rx_csum_fault);
1526 /* Actually, we should eliminate this check as soon as we know, that:
1527 * 1. IOMMU is present and allows to map all the memory.
1528 * 2. No high memory really exists on this machine.
1531 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1533 #ifdef CONFIG_HIGHMEM
1536 if (dev->features & NETIF_F_HIGHDMA)
1539 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1540 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1548 void (*destructor)(struct sk_buff *skb);
1551 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1553 static void dev_gso_skb_destructor(struct sk_buff *skb)
1555 struct dev_gso_cb *cb;
1558 struct sk_buff *nskb = skb->next;
1560 skb->next = nskb->next;
1563 } while (skb->next);
1565 cb = DEV_GSO_CB(skb);
1567 cb->destructor(skb);
1571 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1572 * @skb: buffer to segment
1574 * This function segments the given skb and stores the list of segments
1577 static int dev_gso_segment(struct sk_buff *skb)
1579 struct net_device *dev = skb->dev;
1580 struct sk_buff *segs;
1581 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1584 segs = skb_gso_segment(skb, features);
1586 /* Verifying header integrity only. */
1591 return PTR_ERR(segs);
1594 DEV_GSO_CB(skb)->destructor = skb->destructor;
1595 skb->destructor = dev_gso_skb_destructor;
1600 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1602 if (likely(!skb->next)) {
1603 if (!list_empty(&ptype_all))
1604 dev_queue_xmit_nit(skb, dev);
1606 if (netif_needs_gso(dev, skb)) {
1607 if (unlikely(dev_gso_segment(skb)))
1613 return dev->hard_start_xmit(skb, dev);
1618 struct sk_buff *nskb = skb->next;
1621 skb->next = nskb->next;
1623 rc = dev->hard_start_xmit(nskb, dev);
1625 nskb->next = skb->next;
1629 if (unlikely((netif_queue_stopped(dev) ||
1630 netif_subqueue_stopped(dev, skb)) &&
1632 return NETDEV_TX_BUSY;
1633 } while (skb->next);
1635 skb->destructor = DEV_GSO_CB(skb)->destructor;
1643 * dev_queue_xmit - transmit a buffer
1644 * @skb: buffer to transmit
1646 * Queue a buffer for transmission to a network device. The caller must
1647 * have set the device and priority and built the buffer before calling
1648 * this function. The function can be called from an interrupt.
1650 * A negative errno code is returned on a failure. A success does not
1651 * guarantee the frame will be transmitted as it may be dropped due
1652 * to congestion or traffic shaping.
1654 * -----------------------------------------------------------------------------------
1655 * I notice this method can also return errors from the queue disciplines,
1656 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1659 * Regardless of the return value, the skb is consumed, so it is currently
1660 * difficult to retry a send to this method. (You can bump the ref count
1661 * before sending to hold a reference for retry if you are careful.)
1663 * When calling this method, interrupts MUST be enabled. This is because
1664 * the BH enable code must have IRQs enabled so that it will not deadlock.
1668 int dev_queue_xmit(struct sk_buff *skb)
1670 struct net_device *dev = skb->dev;
1671 struct netdev_queue *txq;
1675 /* GSO will handle the following emulations directly. */
1676 if (netif_needs_gso(dev, skb))
1679 if (skb_shinfo(skb)->frag_list &&
1680 !(dev->features & NETIF_F_FRAGLIST) &&
1681 __skb_linearize(skb))
1684 /* Fragmented skb is linearized if device does not support SG,
1685 * or if at least one of fragments is in highmem and device
1686 * does not support DMA from it.
1688 if (skb_shinfo(skb)->nr_frags &&
1689 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1690 __skb_linearize(skb))
1693 /* If packet is not checksummed and device does not support
1694 * checksumming for this protocol, complete checksumming here.
1696 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1697 skb_set_transport_header(skb, skb->csum_start -
1699 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1704 txq = &dev->tx_queue;
1705 spin_lock_prefetch(&txq->lock);
1707 /* Disable soft irqs for various locks below. Also
1708 * stops preemption for RCU.
1712 /* Updates of qdisc are serialized by queue->lock.
1713 * The struct Qdisc which is pointed to by qdisc is now a
1714 * rcu structure - it may be accessed without acquiring
1715 * a lock (but the structure may be stale.) The freeing of the
1716 * qdisc will be deferred until it's known that there are no
1717 * more references to it.
1719 * If the qdisc has an enqueue function, we still need to
1720 * hold the queue->lock before calling it, since queue->lock
1721 * also serializes access to the device queue.
1724 q = rcu_dereference(txq->qdisc);
1725 #ifdef CONFIG_NET_CLS_ACT
1726 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1729 /* Grab device queue */
1730 spin_lock(&txq->lock);
1733 /* reset queue_mapping to zero */
1734 skb_set_queue_mapping(skb, 0);
1735 rc = q->enqueue(skb, q);
1737 spin_unlock(&txq->lock);
1739 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1742 spin_unlock(&txq->lock);
1745 /* The device has no queue. Common case for software devices:
1746 loopback, all the sorts of tunnels...
1748 Really, it is unlikely that netif_tx_lock protection is necessary
1749 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1751 However, it is possible, that they rely on protection
1754 Check this and shot the lock. It is not prone from deadlocks.
1755 Either shot noqueue qdisc, it is even simpler 8)
1757 if (dev->flags & IFF_UP) {
1758 int cpu = smp_processor_id(); /* ok because BHs are off */
1760 if (dev->xmit_lock_owner != cpu) {
1762 HARD_TX_LOCK(dev, cpu);
1764 if (!netif_queue_stopped(dev) &&
1765 !netif_subqueue_stopped(dev, skb)) {
1767 if (!dev_hard_start_xmit(skb, dev)) {
1768 HARD_TX_UNLOCK(dev);
1772 HARD_TX_UNLOCK(dev);
1773 if (net_ratelimit())
1774 printk(KERN_CRIT "Virtual device %s asks to "
1775 "queue packet!\n", dev->name);
1777 /* Recursion is detected! It is possible,
1779 if (net_ratelimit())
1780 printk(KERN_CRIT "Dead loop on virtual device "
1781 "%s, fix it urgently!\n", dev->name);
1786 rcu_read_unlock_bh();
1792 rcu_read_unlock_bh();
1797 /*=======================================================================
1799 =======================================================================*/
1801 int netdev_max_backlog __read_mostly = 1000;
1802 int netdev_budget __read_mostly = 300;
1803 int weight_p __read_mostly = 64; /* old backlog weight */
1805 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1809 * netif_rx - post buffer to the network code
1810 * @skb: buffer to post
1812 * This function receives a packet from a device driver and queues it for
1813 * the upper (protocol) levels to process. It always succeeds. The buffer
1814 * may be dropped during processing for congestion control or by the
1818 * NET_RX_SUCCESS (no congestion)
1819 * NET_RX_DROP (packet was dropped)
1823 int netif_rx(struct sk_buff *skb)
1825 struct softnet_data *queue;
1826 unsigned long flags;
1828 /* if netpoll wants it, pretend we never saw it */
1829 if (netpoll_rx(skb))
1832 if (!skb->tstamp.tv64)
1836 * The code is rearranged so that the path is the most
1837 * short when CPU is congested, but is still operating.
1839 local_irq_save(flags);
1840 queue = &__get_cpu_var(softnet_data);
1842 __get_cpu_var(netdev_rx_stat).total++;
1843 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1844 if (queue->input_pkt_queue.qlen) {
1847 __skb_queue_tail(&queue->input_pkt_queue, skb);
1848 local_irq_restore(flags);
1849 return NET_RX_SUCCESS;
1852 napi_schedule(&queue->backlog);
1856 __get_cpu_var(netdev_rx_stat).dropped++;
1857 local_irq_restore(flags);
1863 int netif_rx_ni(struct sk_buff *skb)
1868 err = netif_rx(skb);
1869 if (local_softirq_pending())
1876 EXPORT_SYMBOL(netif_rx_ni);
1878 static inline struct net_device *skb_bond(struct sk_buff *skb)
1880 struct net_device *dev = skb->dev;
1883 if (skb_bond_should_drop(skb)) {
1887 skb->dev = dev->master;
1894 static void net_tx_action(struct softirq_action *h)
1896 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1898 if (sd->completion_queue) {
1899 struct sk_buff *clist;
1901 local_irq_disable();
1902 clist = sd->completion_queue;
1903 sd->completion_queue = NULL;
1907 struct sk_buff *skb = clist;
1908 clist = clist->next;
1910 BUG_TRAP(!atomic_read(&skb->users));
1915 if (sd->output_queue) {
1916 struct netdev_queue *head;
1918 local_irq_disable();
1919 head = sd->output_queue;
1920 sd->output_queue = NULL;
1924 struct netdev_queue *txq = head;
1925 struct net_device *dev = txq->dev;
1926 head = head->next_sched;
1928 smp_mb__before_clear_bit();
1929 clear_bit(__LINK_STATE_SCHED, &dev->state);
1931 if (spin_trylock(&txq->lock)) {
1933 spin_unlock(&txq->lock);
1935 netif_schedule(dev);
1941 static inline int deliver_skb(struct sk_buff *skb,
1942 struct packet_type *pt_prev,
1943 struct net_device *orig_dev)
1945 atomic_inc(&skb->users);
1946 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1949 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1950 /* These hooks defined here for ATM */
1952 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1953 unsigned char *addr);
1954 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1957 * If bridge module is loaded call bridging hook.
1958 * returns NULL if packet was consumed.
1960 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1961 struct sk_buff *skb) __read_mostly;
1962 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1963 struct packet_type **pt_prev, int *ret,
1964 struct net_device *orig_dev)
1966 struct net_bridge_port *port;
1968 if (skb->pkt_type == PACKET_LOOPBACK ||
1969 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1973 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1977 return br_handle_frame_hook(port, skb);
1980 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1983 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1984 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
1985 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
1987 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
1988 struct packet_type **pt_prev,
1990 struct net_device *orig_dev)
1992 if (skb->dev->macvlan_port == NULL)
1996 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1999 return macvlan_handle_frame_hook(skb);
2002 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2005 #ifdef CONFIG_NET_CLS_ACT
2006 /* TODO: Maybe we should just force sch_ingress to be compiled in
2007 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2008 * a compare and 2 stores extra right now if we dont have it on
2009 * but have CONFIG_NET_CLS_ACT
2010 * NOTE: This doesnt stop any functionality; if you dont have
2011 * the ingress scheduler, you just cant add policies on ingress.
2014 static int ing_filter(struct sk_buff *skb)
2016 struct net_device *dev = skb->dev;
2017 u32 ttl = G_TC_RTTL(skb->tc_verd);
2018 struct netdev_queue *rxq;
2019 int result = TC_ACT_OK;
2022 if (MAX_RED_LOOP < ttl++) {
2024 "Redir loop detected Dropping packet (%d->%d)\n",
2025 skb->iif, dev->ifindex);
2029 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2030 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2032 rxq = &dev->rx_queue;
2034 spin_lock(&rxq->lock);
2035 if ((q = rxq->qdisc) != NULL)
2036 result = q->enqueue(skb, q);
2037 spin_unlock(&rxq->lock);
2042 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2043 struct packet_type **pt_prev,
2044 int *ret, struct net_device *orig_dev)
2046 if (!skb->dev->rx_queue.qdisc)
2050 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2053 /* Huh? Why does turning on AF_PACKET affect this? */
2054 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2057 switch (ing_filter(skb)) {
2071 * netif_receive_skb - process receive buffer from network
2072 * @skb: buffer to process
2074 * netif_receive_skb() is the main receive data processing function.
2075 * It always succeeds. The buffer may be dropped during processing
2076 * for congestion control or by the protocol layers.
2078 * This function may only be called from softirq context and interrupts
2079 * should be enabled.
2081 * Return values (usually ignored):
2082 * NET_RX_SUCCESS: no congestion
2083 * NET_RX_DROP: packet was dropped
2085 int netif_receive_skb(struct sk_buff *skb)
2087 struct packet_type *ptype, *pt_prev;
2088 struct net_device *orig_dev;
2089 int ret = NET_RX_DROP;
2092 /* if we've gotten here through NAPI, check netpoll */
2093 if (netpoll_receive_skb(skb))
2096 if (!skb->tstamp.tv64)
2100 skb->iif = skb->dev->ifindex;
2102 orig_dev = skb_bond(skb);
2107 __get_cpu_var(netdev_rx_stat).total++;
2109 skb_reset_network_header(skb);
2110 skb_reset_transport_header(skb);
2111 skb->mac_len = skb->network_header - skb->mac_header;
2117 /* Don't receive packets in an exiting network namespace */
2118 if (!net_alive(dev_net(skb->dev)))
2121 #ifdef CONFIG_NET_CLS_ACT
2122 if (skb->tc_verd & TC_NCLS) {
2123 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2128 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2129 if (!ptype->dev || ptype->dev == skb->dev) {
2131 ret = deliver_skb(skb, pt_prev, orig_dev);
2136 #ifdef CONFIG_NET_CLS_ACT
2137 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2143 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2146 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2150 type = skb->protocol;
2151 list_for_each_entry_rcu(ptype,
2152 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2153 if (ptype->type == type &&
2154 (!ptype->dev || ptype->dev == skb->dev)) {
2156 ret = deliver_skb(skb, pt_prev, orig_dev);
2162 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2165 /* Jamal, now you will not able to escape explaining
2166 * me how you were going to use this. :-)
2176 static int process_backlog(struct napi_struct *napi, int quota)
2179 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2180 unsigned long start_time = jiffies;
2182 napi->weight = weight_p;
2184 struct sk_buff *skb;
2185 struct net_device *dev;
2187 local_irq_disable();
2188 skb = __skb_dequeue(&queue->input_pkt_queue);
2190 __napi_complete(napi);
2199 netif_receive_skb(skb);
2202 } while (++work < quota && jiffies == start_time);
2208 * __napi_schedule - schedule for receive
2209 * @n: entry to schedule
2211 * The entry's receive function will be scheduled to run
2213 void __napi_schedule(struct napi_struct *n)
2215 unsigned long flags;
2217 local_irq_save(flags);
2218 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2219 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2220 local_irq_restore(flags);
2222 EXPORT_SYMBOL(__napi_schedule);
2225 static void net_rx_action(struct softirq_action *h)
2227 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2228 unsigned long start_time = jiffies;
2229 int budget = netdev_budget;
2232 local_irq_disable();
2234 while (!list_empty(list)) {
2235 struct napi_struct *n;
2238 /* If softirq window is exhuasted then punt.
2240 * Note that this is a slight policy change from the
2241 * previous NAPI code, which would allow up to 2
2242 * jiffies to pass before breaking out. The test
2243 * used to be "jiffies - start_time > 1".
2245 if (unlikely(budget <= 0 || jiffies != start_time))
2250 /* Even though interrupts have been re-enabled, this
2251 * access is safe because interrupts can only add new
2252 * entries to the tail of this list, and only ->poll()
2253 * calls can remove this head entry from the list.
2255 n = list_entry(list->next, struct napi_struct, poll_list);
2257 have = netpoll_poll_lock(n);
2261 /* This NAPI_STATE_SCHED test is for avoiding a race
2262 * with netpoll's poll_napi(). Only the entity which
2263 * obtains the lock and sees NAPI_STATE_SCHED set will
2264 * actually make the ->poll() call. Therefore we avoid
2265 * accidently calling ->poll() when NAPI is not scheduled.
2268 if (test_bit(NAPI_STATE_SCHED, &n->state))
2269 work = n->poll(n, weight);
2271 WARN_ON_ONCE(work > weight);
2275 local_irq_disable();
2277 /* Drivers must not modify the NAPI state if they
2278 * consume the entire weight. In such cases this code
2279 * still "owns" the NAPI instance and therefore can
2280 * move the instance around on the list at-will.
2282 if (unlikely(work == weight)) {
2283 if (unlikely(napi_disable_pending(n)))
2286 list_move_tail(&n->poll_list, list);
2289 netpoll_poll_unlock(have);
2294 #ifdef CONFIG_NET_DMA
2296 * There may not be any more sk_buffs coming right now, so push
2297 * any pending DMA copies to hardware
2299 if (!cpus_empty(net_dma.channel_mask)) {
2301 for_each_cpu_mask(chan_idx, net_dma.channel_mask) {
2302 struct dma_chan *chan = net_dma.channels[chan_idx];
2304 dma_async_memcpy_issue_pending(chan);
2312 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2313 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2317 static gifconf_func_t * gifconf_list [NPROTO];
2320 * register_gifconf - register a SIOCGIF handler
2321 * @family: Address family
2322 * @gifconf: Function handler
2324 * Register protocol dependent address dumping routines. The handler
2325 * that is passed must not be freed or reused until it has been replaced
2326 * by another handler.
2328 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2330 if (family >= NPROTO)
2332 gifconf_list[family] = gifconf;
2338 * Map an interface index to its name (SIOCGIFNAME)
2342 * We need this ioctl for efficient implementation of the
2343 * if_indextoname() function required by the IPv6 API. Without
2344 * it, we would have to search all the interfaces to find a
2348 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2350 struct net_device *dev;
2354 * Fetch the caller's info block.
2357 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2360 read_lock(&dev_base_lock);
2361 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2363 read_unlock(&dev_base_lock);
2367 strcpy(ifr.ifr_name, dev->name);
2368 read_unlock(&dev_base_lock);
2370 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2376 * Perform a SIOCGIFCONF call. This structure will change
2377 * size eventually, and there is nothing I can do about it.
2378 * Thus we will need a 'compatibility mode'.
2381 static int dev_ifconf(struct net *net, char __user *arg)
2384 struct net_device *dev;
2391 * Fetch the caller's info block.
2394 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2401 * Loop over the interfaces, and write an info block for each.
2405 for_each_netdev(net, dev) {
2406 for (i = 0; i < NPROTO; i++) {
2407 if (gifconf_list[i]) {
2410 done = gifconf_list[i](dev, NULL, 0);
2412 done = gifconf_list[i](dev, pos + total,
2422 * All done. Write the updated control block back to the caller.
2424 ifc.ifc_len = total;
2427 * Both BSD and Solaris return 0 here, so we do too.
2429 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2432 #ifdef CONFIG_PROC_FS
2434 * This is invoked by the /proc filesystem handler to display a device
2437 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2438 __acquires(dev_base_lock)
2440 struct net *net = seq_file_net(seq);
2442 struct net_device *dev;
2444 read_lock(&dev_base_lock);
2446 return SEQ_START_TOKEN;
2449 for_each_netdev(net, dev)
2456 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2458 struct net *net = seq_file_net(seq);
2460 return v == SEQ_START_TOKEN ?
2461 first_net_device(net) : next_net_device((struct net_device *)v);
2464 void dev_seq_stop(struct seq_file *seq, void *v)
2465 __releases(dev_base_lock)
2467 read_unlock(&dev_base_lock);
2470 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2472 struct net_device_stats *stats = dev->get_stats(dev);
2474 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2475 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2476 dev->name, stats->rx_bytes, stats->rx_packets,
2478 stats->rx_dropped + stats->rx_missed_errors,
2479 stats->rx_fifo_errors,
2480 stats->rx_length_errors + stats->rx_over_errors +
2481 stats->rx_crc_errors + stats->rx_frame_errors,
2482 stats->rx_compressed, stats->multicast,
2483 stats->tx_bytes, stats->tx_packets,
2484 stats->tx_errors, stats->tx_dropped,
2485 stats->tx_fifo_errors, stats->collisions,
2486 stats->tx_carrier_errors +
2487 stats->tx_aborted_errors +
2488 stats->tx_window_errors +
2489 stats->tx_heartbeat_errors,
2490 stats->tx_compressed);
2494 * Called from the PROCfs module. This now uses the new arbitrary sized
2495 * /proc/net interface to create /proc/net/dev
2497 static int dev_seq_show(struct seq_file *seq, void *v)
2499 if (v == SEQ_START_TOKEN)
2500 seq_puts(seq, "Inter-| Receive "
2502 " face |bytes packets errs drop fifo frame "
2503 "compressed multicast|bytes packets errs "
2504 "drop fifo colls carrier compressed\n");
2506 dev_seq_printf_stats(seq, v);
2510 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2512 struct netif_rx_stats *rc = NULL;
2514 while (*pos < nr_cpu_ids)
2515 if (cpu_online(*pos)) {
2516 rc = &per_cpu(netdev_rx_stat, *pos);
2523 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2525 return softnet_get_online(pos);
2528 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2531 return softnet_get_online(pos);
2534 static void softnet_seq_stop(struct seq_file *seq, void *v)
2538 static int softnet_seq_show(struct seq_file *seq, void *v)
2540 struct netif_rx_stats *s = v;
2542 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2543 s->total, s->dropped, s->time_squeeze, 0,
2544 0, 0, 0, 0, /* was fastroute */
2549 static const struct seq_operations dev_seq_ops = {
2550 .start = dev_seq_start,
2551 .next = dev_seq_next,
2552 .stop = dev_seq_stop,
2553 .show = dev_seq_show,
2556 static int dev_seq_open(struct inode *inode, struct file *file)
2558 return seq_open_net(inode, file, &dev_seq_ops,
2559 sizeof(struct seq_net_private));
2562 static const struct file_operations dev_seq_fops = {
2563 .owner = THIS_MODULE,
2564 .open = dev_seq_open,
2566 .llseek = seq_lseek,
2567 .release = seq_release_net,
2570 static const struct seq_operations softnet_seq_ops = {
2571 .start = softnet_seq_start,
2572 .next = softnet_seq_next,
2573 .stop = softnet_seq_stop,
2574 .show = softnet_seq_show,
2577 static int softnet_seq_open(struct inode *inode, struct file *file)
2579 return seq_open(file, &softnet_seq_ops);
2582 static const struct file_operations softnet_seq_fops = {
2583 .owner = THIS_MODULE,
2584 .open = softnet_seq_open,
2586 .llseek = seq_lseek,
2587 .release = seq_release,
2590 static void *ptype_get_idx(loff_t pos)
2592 struct packet_type *pt = NULL;
2596 list_for_each_entry_rcu(pt, &ptype_all, list) {
2602 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2603 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2612 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2616 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2619 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2621 struct packet_type *pt;
2622 struct list_head *nxt;
2626 if (v == SEQ_START_TOKEN)
2627 return ptype_get_idx(0);
2630 nxt = pt->list.next;
2631 if (pt->type == htons(ETH_P_ALL)) {
2632 if (nxt != &ptype_all)
2635 nxt = ptype_base[0].next;
2637 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2639 while (nxt == &ptype_base[hash]) {
2640 if (++hash >= PTYPE_HASH_SIZE)
2642 nxt = ptype_base[hash].next;
2645 return list_entry(nxt, struct packet_type, list);
2648 static void ptype_seq_stop(struct seq_file *seq, void *v)
2654 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2656 #ifdef CONFIG_KALLSYMS
2657 unsigned long offset = 0, symsize;
2658 const char *symname;
2662 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2669 modname = delim = "";
2670 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2676 seq_printf(seq, "[%p]", sym);
2679 static int ptype_seq_show(struct seq_file *seq, void *v)
2681 struct packet_type *pt = v;
2683 if (v == SEQ_START_TOKEN)
2684 seq_puts(seq, "Type Device Function\n");
2685 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2686 if (pt->type == htons(ETH_P_ALL))
2687 seq_puts(seq, "ALL ");
2689 seq_printf(seq, "%04x", ntohs(pt->type));
2691 seq_printf(seq, " %-8s ",
2692 pt->dev ? pt->dev->name : "");
2693 ptype_seq_decode(seq, pt->func);
2694 seq_putc(seq, '\n');
2700 static const struct seq_operations ptype_seq_ops = {
2701 .start = ptype_seq_start,
2702 .next = ptype_seq_next,
2703 .stop = ptype_seq_stop,
2704 .show = ptype_seq_show,
2707 static int ptype_seq_open(struct inode *inode, struct file *file)
2709 return seq_open_net(inode, file, &ptype_seq_ops,
2710 sizeof(struct seq_net_private));
2713 static const struct file_operations ptype_seq_fops = {
2714 .owner = THIS_MODULE,
2715 .open = ptype_seq_open,
2717 .llseek = seq_lseek,
2718 .release = seq_release_net,
2722 static int __net_init dev_proc_net_init(struct net *net)
2726 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2728 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2730 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2733 if (wext_proc_init(net))
2739 proc_net_remove(net, "ptype");
2741 proc_net_remove(net, "softnet_stat");
2743 proc_net_remove(net, "dev");
2747 static void __net_exit dev_proc_net_exit(struct net *net)
2749 wext_proc_exit(net);
2751 proc_net_remove(net, "ptype");
2752 proc_net_remove(net, "softnet_stat");
2753 proc_net_remove(net, "dev");
2756 static struct pernet_operations __net_initdata dev_proc_ops = {
2757 .init = dev_proc_net_init,
2758 .exit = dev_proc_net_exit,
2761 static int __init dev_proc_init(void)
2763 return register_pernet_subsys(&dev_proc_ops);
2766 #define dev_proc_init() 0
2767 #endif /* CONFIG_PROC_FS */
2771 * netdev_set_master - set up master/slave pair
2772 * @slave: slave device
2773 * @master: new master device
2775 * Changes the master device of the slave. Pass %NULL to break the
2776 * bonding. The caller must hold the RTNL semaphore. On a failure
2777 * a negative errno code is returned. On success the reference counts
2778 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2779 * function returns zero.
2781 int netdev_set_master(struct net_device *slave, struct net_device *master)
2783 struct net_device *old = slave->master;
2793 slave->master = master;
2801 slave->flags |= IFF_SLAVE;
2803 slave->flags &= ~IFF_SLAVE;
2805 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2809 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2811 unsigned short old_flags = dev->flags;
2815 dev->flags |= IFF_PROMISC;
2816 dev->promiscuity += inc;
2817 if (dev->promiscuity == 0) {
2820 * If inc causes overflow, untouch promisc and return error.
2823 dev->flags &= ~IFF_PROMISC;
2825 dev->promiscuity -= inc;
2826 printk(KERN_WARNING "%s: promiscuity touches roof, "
2827 "set promiscuity failed, promiscuity feature "
2828 "of device might be broken.\n", dev->name);
2832 if (dev->flags != old_flags) {
2833 printk(KERN_INFO "device %s %s promiscuous mode\n",
2834 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2837 audit_log(current->audit_context, GFP_ATOMIC,
2838 AUDIT_ANOM_PROMISCUOUS,
2839 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2840 dev->name, (dev->flags & IFF_PROMISC),
2841 (old_flags & IFF_PROMISC),
2842 audit_get_loginuid(current),
2843 current->uid, current->gid,
2844 audit_get_sessionid(current));
2846 if (dev->change_rx_flags)
2847 dev->change_rx_flags(dev, IFF_PROMISC);
2853 * dev_set_promiscuity - update promiscuity count on a device
2857 * Add or remove promiscuity from a device. While the count in the device
2858 * remains above zero the interface remains promiscuous. Once it hits zero
2859 * the device reverts back to normal filtering operation. A negative inc
2860 * value is used to drop promiscuity on the device.
2861 * Return 0 if successful or a negative errno code on error.
2863 int dev_set_promiscuity(struct net_device *dev, int inc)
2865 unsigned short old_flags = dev->flags;
2868 err = __dev_set_promiscuity(dev, inc);
2871 if (dev->flags != old_flags)
2872 dev_set_rx_mode(dev);
2877 * dev_set_allmulti - update allmulti count on a device
2881 * Add or remove reception of all multicast frames to a device. While the
2882 * count in the device remains above zero the interface remains listening
2883 * to all interfaces. Once it hits zero the device reverts back to normal
2884 * filtering operation. A negative @inc value is used to drop the counter
2885 * when releasing a resource needing all multicasts.
2886 * Return 0 if successful or a negative errno code on error.
2889 int dev_set_allmulti(struct net_device *dev, int inc)
2891 unsigned short old_flags = dev->flags;
2895 dev->flags |= IFF_ALLMULTI;
2896 dev->allmulti += inc;
2897 if (dev->allmulti == 0) {
2900 * If inc causes overflow, untouch allmulti and return error.
2903 dev->flags &= ~IFF_ALLMULTI;
2905 dev->allmulti -= inc;
2906 printk(KERN_WARNING "%s: allmulti touches roof, "
2907 "set allmulti failed, allmulti feature of "
2908 "device might be broken.\n", dev->name);
2912 if (dev->flags ^ old_flags) {
2913 if (dev->change_rx_flags)
2914 dev->change_rx_flags(dev, IFF_ALLMULTI);
2915 dev_set_rx_mode(dev);
2921 * Upload unicast and multicast address lists to device and
2922 * configure RX filtering. When the device doesn't support unicast
2923 * filtering it is put in promiscuous mode while unicast addresses
2926 void __dev_set_rx_mode(struct net_device *dev)
2928 /* dev_open will call this function so the list will stay sane. */
2929 if (!(dev->flags&IFF_UP))
2932 if (!netif_device_present(dev))
2935 if (dev->set_rx_mode)
2936 dev->set_rx_mode(dev);
2938 /* Unicast addresses changes may only happen under the rtnl,
2939 * therefore calling __dev_set_promiscuity here is safe.
2941 if (dev->uc_count > 0 && !dev->uc_promisc) {
2942 __dev_set_promiscuity(dev, 1);
2943 dev->uc_promisc = 1;
2944 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2945 __dev_set_promiscuity(dev, -1);
2946 dev->uc_promisc = 0;
2949 if (dev->set_multicast_list)
2950 dev->set_multicast_list(dev);
2954 void dev_set_rx_mode(struct net_device *dev)
2956 netif_tx_lock_bh(dev);
2957 __dev_set_rx_mode(dev);
2958 netif_tx_unlock_bh(dev);
2961 int __dev_addr_delete(struct dev_addr_list **list, int *count,
2962 void *addr, int alen, int glbl)
2964 struct dev_addr_list *da;
2966 for (; (da = *list) != NULL; list = &da->next) {
2967 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2968 alen == da->da_addrlen) {
2970 int old_glbl = da->da_gusers;
2987 int __dev_addr_add(struct dev_addr_list **list, int *count,
2988 void *addr, int alen, int glbl)
2990 struct dev_addr_list *da;
2992 for (da = *list; da != NULL; da = da->next) {
2993 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2994 da->da_addrlen == alen) {
2996 int old_glbl = da->da_gusers;
3006 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3009 memcpy(da->da_addr, addr, alen);
3010 da->da_addrlen = alen;
3012 da->da_gusers = glbl ? 1 : 0;
3020 * dev_unicast_delete - Release secondary unicast address.
3022 * @addr: address to delete
3023 * @alen: length of @addr
3025 * Release reference to a secondary unicast address and remove it
3026 * from the device if the reference count drops to zero.
3028 * The caller must hold the rtnl_mutex.
3030 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3036 netif_tx_lock_bh(dev);
3037 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3039 __dev_set_rx_mode(dev);
3040 netif_tx_unlock_bh(dev);
3043 EXPORT_SYMBOL(dev_unicast_delete);
3046 * dev_unicast_add - add a secondary unicast address
3048 * @addr: address to add
3049 * @alen: length of @addr
3051 * Add a secondary unicast address to the device or increase
3052 * the reference count if it already exists.
3054 * The caller must hold the rtnl_mutex.
3056 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3062 netif_tx_lock_bh(dev);
3063 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3065 __dev_set_rx_mode(dev);
3066 netif_tx_unlock_bh(dev);
3069 EXPORT_SYMBOL(dev_unicast_add);
3071 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3072 struct dev_addr_list **from, int *from_count)
3074 struct dev_addr_list *da, *next;
3078 while (da != NULL) {
3080 if (!da->da_synced) {
3081 err = __dev_addr_add(to, to_count,
3082 da->da_addr, da->da_addrlen, 0);
3087 } else if (da->da_users == 1) {
3088 __dev_addr_delete(to, to_count,
3089 da->da_addr, da->da_addrlen, 0);
3090 __dev_addr_delete(from, from_count,
3091 da->da_addr, da->da_addrlen, 0);
3098 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3099 struct dev_addr_list **from, int *from_count)
3101 struct dev_addr_list *da, *next;
3104 while (da != NULL) {
3106 if (da->da_synced) {
3107 __dev_addr_delete(to, to_count,
3108 da->da_addr, da->da_addrlen, 0);
3110 __dev_addr_delete(from, from_count,
3111 da->da_addr, da->da_addrlen, 0);
3118 * dev_unicast_sync - Synchronize device's unicast list to another device
3119 * @to: destination device
3120 * @from: source device
3122 * Add newly added addresses to the destination device and release
3123 * addresses that have no users left. The source device must be
3124 * locked by netif_tx_lock_bh.
3126 * This function is intended to be called from the dev->set_rx_mode
3127 * function of layered software devices.
3129 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3133 netif_tx_lock_bh(to);
3134 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3135 &from->uc_list, &from->uc_count);
3137 __dev_set_rx_mode(to);
3138 netif_tx_unlock_bh(to);
3141 EXPORT_SYMBOL(dev_unicast_sync);
3144 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3145 * @to: destination device
3146 * @from: source device
3148 * Remove all addresses that were added to the destination device by
3149 * dev_unicast_sync(). This function is intended to be called from the
3150 * dev->stop function of layered software devices.
3152 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3154 netif_tx_lock_bh(from);
3155 netif_tx_lock_bh(to);
3157 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3158 &from->uc_list, &from->uc_count);
3159 __dev_set_rx_mode(to);
3161 netif_tx_unlock_bh(to);
3162 netif_tx_unlock_bh(from);
3164 EXPORT_SYMBOL(dev_unicast_unsync);
3166 static void __dev_addr_discard(struct dev_addr_list **list)
3168 struct dev_addr_list *tmp;
3170 while (*list != NULL) {
3173 if (tmp->da_users > tmp->da_gusers)
3174 printk("__dev_addr_discard: address leakage! "
3175 "da_users=%d\n", tmp->da_users);
3180 static void dev_addr_discard(struct net_device *dev)
3182 netif_tx_lock_bh(dev);
3184 __dev_addr_discard(&dev->uc_list);
3187 __dev_addr_discard(&dev->mc_list);
3190 netif_tx_unlock_bh(dev);
3193 unsigned dev_get_flags(const struct net_device *dev)
3197 flags = (dev->flags & ~(IFF_PROMISC |
3202 (dev->gflags & (IFF_PROMISC |
3205 if (netif_running(dev)) {
3206 if (netif_oper_up(dev))
3207 flags |= IFF_RUNNING;
3208 if (netif_carrier_ok(dev))
3209 flags |= IFF_LOWER_UP;
3210 if (netif_dormant(dev))
3211 flags |= IFF_DORMANT;
3217 int dev_change_flags(struct net_device *dev, unsigned flags)
3220 int old_flags = dev->flags;
3225 * Set the flags on our device.
3228 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3229 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3231 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3235 * Load in the correct multicast list now the flags have changed.
3238 if (dev->change_rx_flags && (old_flags ^ flags) & IFF_MULTICAST)
3239 dev->change_rx_flags(dev, IFF_MULTICAST);
3241 dev_set_rx_mode(dev);
3244 * Have we downed the interface. We handle IFF_UP ourselves
3245 * according to user attempts to set it, rather than blindly
3250 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3251 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3254 dev_set_rx_mode(dev);
3257 if (dev->flags & IFF_UP &&
3258 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3260 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3262 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3263 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3264 dev->gflags ^= IFF_PROMISC;
3265 dev_set_promiscuity(dev, inc);
3268 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3269 is important. Some (broken) drivers set IFF_PROMISC, when
3270 IFF_ALLMULTI is requested not asking us and not reporting.
3272 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3273 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3274 dev->gflags ^= IFF_ALLMULTI;
3275 dev_set_allmulti(dev, inc);
3278 /* Exclude state transition flags, already notified */
3279 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3281 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3286 int dev_set_mtu(struct net_device *dev, int new_mtu)
3290 if (new_mtu == dev->mtu)
3293 /* MTU must be positive. */
3297 if (!netif_device_present(dev))
3301 if (dev->change_mtu)
3302 err = dev->change_mtu(dev, new_mtu);
3305 if (!err && dev->flags & IFF_UP)
3306 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3310 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3314 if (!dev->set_mac_address)
3316 if (sa->sa_family != dev->type)
3318 if (!netif_device_present(dev))
3320 err = dev->set_mac_address(dev, sa);
3322 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3327 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3329 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3332 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3338 case SIOCGIFFLAGS: /* Get interface flags */
3339 ifr->ifr_flags = dev_get_flags(dev);
3342 case SIOCGIFMETRIC: /* Get the metric on the interface
3343 (currently unused) */
3344 ifr->ifr_metric = 0;
3347 case SIOCGIFMTU: /* Get the MTU of a device */
3348 ifr->ifr_mtu = dev->mtu;
3353 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3355 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3356 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3357 ifr->ifr_hwaddr.sa_family = dev->type;
3365 ifr->ifr_map.mem_start = dev->mem_start;
3366 ifr->ifr_map.mem_end = dev->mem_end;
3367 ifr->ifr_map.base_addr = dev->base_addr;
3368 ifr->ifr_map.irq = dev->irq;
3369 ifr->ifr_map.dma = dev->dma;
3370 ifr->ifr_map.port = dev->if_port;
3374 ifr->ifr_ifindex = dev->ifindex;
3378 ifr->ifr_qlen = dev->tx_queue_len;
3382 /* dev_ioctl() should ensure this case
3394 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3396 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3399 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3405 case SIOCSIFFLAGS: /* Set interface flags */
3406 return dev_change_flags(dev, ifr->ifr_flags);
3408 case SIOCSIFMETRIC: /* Set the metric on the interface
3409 (currently unused) */
3412 case SIOCSIFMTU: /* Set the MTU of a device */
3413 return dev_set_mtu(dev, ifr->ifr_mtu);
3416 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3418 case SIOCSIFHWBROADCAST:
3419 if (ifr->ifr_hwaddr.sa_family != dev->type)
3421 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3422 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3423 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3427 if (dev->set_config) {
3428 if (!netif_device_present(dev))
3430 return dev->set_config(dev, &ifr->ifr_map);
3435 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3436 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3438 if (!netif_device_present(dev))
3440 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3444 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3445 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3447 if (!netif_device_present(dev))
3449 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3453 if (ifr->ifr_qlen < 0)
3455 dev->tx_queue_len = ifr->ifr_qlen;
3459 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3460 return dev_change_name(dev, ifr->ifr_newname);
3463 * Unknown or private ioctl
3467 if ((cmd >= SIOCDEVPRIVATE &&
3468 cmd <= SIOCDEVPRIVATE + 15) ||
3469 cmd == SIOCBONDENSLAVE ||
3470 cmd == SIOCBONDRELEASE ||
3471 cmd == SIOCBONDSETHWADDR ||
3472 cmd == SIOCBONDSLAVEINFOQUERY ||
3473 cmd == SIOCBONDINFOQUERY ||
3474 cmd == SIOCBONDCHANGEACTIVE ||
3475 cmd == SIOCGMIIPHY ||
3476 cmd == SIOCGMIIREG ||
3477 cmd == SIOCSMIIREG ||
3478 cmd == SIOCBRADDIF ||
3479 cmd == SIOCBRDELIF ||
3480 cmd == SIOCWANDEV) {
3482 if (dev->do_ioctl) {
3483 if (netif_device_present(dev))
3484 err = dev->do_ioctl(dev, ifr,
3497 * This function handles all "interface"-type I/O control requests. The actual
3498 * 'doing' part of this is dev_ifsioc above.
3502 * dev_ioctl - network device ioctl
3503 * @net: the applicable net namespace
3504 * @cmd: command to issue
3505 * @arg: pointer to a struct ifreq in user space
3507 * Issue ioctl functions to devices. This is normally called by the
3508 * user space syscall interfaces but can sometimes be useful for
3509 * other purposes. The return value is the return from the syscall if
3510 * positive or a negative errno code on error.
3513 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3519 /* One special case: SIOCGIFCONF takes ifconf argument
3520 and requires shared lock, because it sleeps writing
3524 if (cmd == SIOCGIFCONF) {
3526 ret = dev_ifconf(net, (char __user *) arg);
3530 if (cmd == SIOCGIFNAME)
3531 return dev_ifname(net, (struct ifreq __user *)arg);
3533 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3536 ifr.ifr_name[IFNAMSIZ-1] = 0;
3538 colon = strchr(ifr.ifr_name, ':');
3543 * See which interface the caller is talking about.
3548 * These ioctl calls:
3549 * - can be done by all.
3550 * - atomic and do not require locking.
3561 dev_load(net, ifr.ifr_name);
3562 read_lock(&dev_base_lock);
3563 ret = dev_ifsioc_locked(net, &ifr, cmd);
3564 read_unlock(&dev_base_lock);
3568 if (copy_to_user(arg, &ifr,
3569 sizeof(struct ifreq)))
3575 dev_load(net, ifr.ifr_name);
3577 ret = dev_ethtool(net, &ifr);
3582 if (copy_to_user(arg, &ifr,
3583 sizeof(struct ifreq)))
3589 * These ioctl calls:
3590 * - require superuser power.
3591 * - require strict serialization.
3597 if (!capable(CAP_NET_ADMIN))
3599 dev_load(net, ifr.ifr_name);
3601 ret = dev_ifsioc(net, &ifr, cmd);
3606 if (copy_to_user(arg, &ifr,
3607 sizeof(struct ifreq)))
3613 * These ioctl calls:
3614 * - require superuser power.
3615 * - require strict serialization.
3616 * - do not return a value
3626 case SIOCSIFHWBROADCAST:
3629 case SIOCBONDENSLAVE:
3630 case SIOCBONDRELEASE:
3631 case SIOCBONDSETHWADDR:
3632 case SIOCBONDCHANGEACTIVE:
3635 if (!capable(CAP_NET_ADMIN))
3638 case SIOCBONDSLAVEINFOQUERY:
3639 case SIOCBONDINFOQUERY:
3640 dev_load(net, ifr.ifr_name);
3642 ret = dev_ifsioc(net, &ifr, cmd);
3647 /* Get the per device memory space. We can add this but
3648 * currently do not support it */
3650 /* Set the per device memory buffer space.
3651 * Not applicable in our case */
3656 * Unknown or private ioctl.
3659 if (cmd == SIOCWANDEV ||
3660 (cmd >= SIOCDEVPRIVATE &&
3661 cmd <= SIOCDEVPRIVATE + 15)) {
3662 dev_load(net, ifr.ifr_name);
3664 ret = dev_ifsioc(net, &ifr, cmd);
3666 if (!ret && copy_to_user(arg, &ifr,
3667 sizeof(struct ifreq)))
3671 /* Take care of Wireless Extensions */
3672 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3673 return wext_handle_ioctl(net, &ifr, cmd, arg);
3680 * dev_new_index - allocate an ifindex
3681 * @net: the applicable net namespace
3683 * Returns a suitable unique value for a new device interface
3684 * number. The caller must hold the rtnl semaphore or the
3685 * dev_base_lock to be sure it remains unique.
3687 static int dev_new_index(struct net *net)
3693 if (!__dev_get_by_index(net, ifindex))
3698 /* Delayed registration/unregisteration */
3699 static DEFINE_SPINLOCK(net_todo_list_lock);
3700 static LIST_HEAD(net_todo_list);
3702 static void net_set_todo(struct net_device *dev)
3704 spin_lock(&net_todo_list_lock);
3705 list_add_tail(&dev->todo_list, &net_todo_list);
3706 spin_unlock(&net_todo_list_lock);
3709 static void rollback_registered(struct net_device *dev)
3711 BUG_ON(dev_boot_phase);
3714 /* Some devices call without registering for initialization unwind. */
3715 if (dev->reg_state == NETREG_UNINITIALIZED) {
3716 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3717 "was registered\n", dev->name, dev);
3723 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3725 /* If device is running, close it first. */
3728 /* And unlink it from device chain. */
3729 unlist_netdevice(dev);
3731 dev->reg_state = NETREG_UNREGISTERING;
3735 /* Shutdown queueing discipline. */
3739 /* Notify protocols, that we are about to destroy
3740 this device. They should clean all the things.
3742 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3745 * Flush the unicast and multicast chains
3747 dev_addr_discard(dev);
3752 /* Notifier chain MUST detach us from master device. */
3753 BUG_TRAP(!dev->master);
3755 /* Remove entries from kobject tree */
3756 netdev_unregister_kobject(dev);
3764 * register_netdevice - register a network device
3765 * @dev: device to register
3767 * Take a completed network device structure and add it to the kernel
3768 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3769 * chain. 0 is returned on success. A negative errno code is returned
3770 * on a failure to set up the device, or if the name is a duplicate.
3772 * Callers must hold the rtnl semaphore. You may want
3773 * register_netdev() instead of this.
3776 * The locking appears insufficient to guarantee two parallel registers
3777 * will not get the same name.
3780 int register_netdevice(struct net_device *dev)
3782 struct hlist_head *head;
3783 struct hlist_node *p;
3787 BUG_ON(dev_boot_phase);
3792 /* When net_device's are persistent, this will be fatal. */
3793 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3794 BUG_ON(!dev_net(dev));
3797 spin_lock_init(&dev->_xmit_lock);
3798 netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
3799 dev->xmit_lock_owner = -1;
3803 /* Init, if this function is available */
3805 ret = dev->init(dev);
3813 if (!dev_valid_name(dev->name)) {
3818 dev->ifindex = dev_new_index(net);
3819 if (dev->iflink == -1)
3820 dev->iflink = dev->ifindex;
3822 /* Check for existence of name */
3823 head = dev_name_hash(net, dev->name);
3824 hlist_for_each(p, head) {
3825 struct net_device *d
3826 = hlist_entry(p, struct net_device, name_hlist);
3827 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3833 /* Fix illegal checksum combinations */
3834 if ((dev->features & NETIF_F_HW_CSUM) &&
3835 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3836 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3838 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3841 if ((dev->features & NETIF_F_NO_CSUM) &&
3842 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3843 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3845 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3849 /* Fix illegal SG+CSUM combinations. */
3850 if ((dev->features & NETIF_F_SG) &&
3851 !(dev->features & NETIF_F_ALL_CSUM)) {
3852 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3854 dev->features &= ~NETIF_F_SG;
3857 /* TSO requires that SG is present as well. */
3858 if ((dev->features & NETIF_F_TSO) &&
3859 !(dev->features & NETIF_F_SG)) {
3860 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3862 dev->features &= ~NETIF_F_TSO;
3864 if (dev->features & NETIF_F_UFO) {
3865 if (!(dev->features & NETIF_F_HW_CSUM)) {
3866 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3867 "NETIF_F_HW_CSUM feature.\n",
3869 dev->features &= ~NETIF_F_UFO;
3871 if (!(dev->features & NETIF_F_SG)) {
3872 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3873 "NETIF_F_SG feature.\n",
3875 dev->features &= ~NETIF_F_UFO;
3879 netdev_initialize_kobject(dev);
3880 ret = netdev_register_kobject(dev);
3883 dev->reg_state = NETREG_REGISTERED;
3886 * Default initial state at registry is that the
3887 * device is present.
3890 set_bit(__LINK_STATE_PRESENT, &dev->state);
3892 dev_init_scheduler(dev);
3894 list_netdevice(dev);
3896 /* Notify protocols, that a new device appeared. */
3897 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
3898 ret = notifier_to_errno(ret);
3900 rollback_registered(dev);
3901 dev->reg_state = NETREG_UNREGISTERED;
3914 * register_netdev - register a network device
3915 * @dev: device to register
3917 * Take a completed network device structure and add it to the kernel
3918 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3919 * chain. 0 is returned on success. A negative errno code is returned
3920 * on a failure to set up the device, or if the name is a duplicate.
3922 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3923 * and expands the device name if you passed a format string to
3926 int register_netdev(struct net_device *dev)
3933 * If the name is a format string the caller wants us to do a
3936 if (strchr(dev->name, '%')) {
3937 err = dev_alloc_name(dev, dev->name);
3942 err = register_netdevice(dev);
3947 EXPORT_SYMBOL(register_netdev);
3950 * netdev_wait_allrefs - wait until all references are gone.
3952 * This is called when unregistering network devices.
3954 * Any protocol or device that holds a reference should register
3955 * for netdevice notification, and cleanup and put back the
3956 * reference if they receive an UNREGISTER event.
3957 * We can get stuck here if buggy protocols don't correctly
3960 static void netdev_wait_allrefs(struct net_device *dev)
3962 unsigned long rebroadcast_time, warning_time;
3964 rebroadcast_time = warning_time = jiffies;
3965 while (atomic_read(&dev->refcnt) != 0) {
3966 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3969 /* Rebroadcast unregister notification */
3970 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3972 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3974 /* We must not have linkwatch events
3975 * pending on unregister. If this
3976 * happens, we simply run the queue
3977 * unscheduled, resulting in a noop
3980 linkwatch_run_queue();
3985 rebroadcast_time = jiffies;
3990 if (time_after(jiffies, warning_time + 10 * HZ)) {
3991 printk(KERN_EMERG "unregister_netdevice: "
3992 "waiting for %s to become free. Usage "
3994 dev->name, atomic_read(&dev->refcnt));
3995 warning_time = jiffies;
4004 * register_netdevice(x1);
4005 * register_netdevice(x2);
4007 * unregister_netdevice(y1);
4008 * unregister_netdevice(y2);
4014 * We are invoked by rtnl_unlock() after it drops the semaphore.
4015 * This allows us to deal with problems:
4016 * 1) We can delete sysfs objects which invoke hotplug
4017 * without deadlocking with linkwatch via keventd.
4018 * 2) Since we run with the RTNL semaphore not held, we can sleep
4019 * safely in order to wait for the netdev refcnt to drop to zero.
4021 static DEFINE_MUTEX(net_todo_run_mutex);
4022 void netdev_run_todo(void)
4024 struct list_head list;
4026 /* Need to guard against multiple cpu's getting out of order. */
4027 mutex_lock(&net_todo_run_mutex);
4029 /* Not safe to do outside the semaphore. We must not return
4030 * until all unregister events invoked by the local processor
4031 * have been completed (either by this todo run, or one on
4034 if (list_empty(&net_todo_list))
4037 /* Snapshot list, allow later requests */
4038 spin_lock(&net_todo_list_lock);
4039 list_replace_init(&net_todo_list, &list);
4040 spin_unlock(&net_todo_list_lock);
4042 while (!list_empty(&list)) {
4043 struct net_device *dev
4044 = list_entry(list.next, struct net_device, todo_list);
4045 list_del(&dev->todo_list);
4047 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4048 printk(KERN_ERR "network todo '%s' but state %d\n",
4049 dev->name, dev->reg_state);
4054 dev->reg_state = NETREG_UNREGISTERED;
4056 netdev_wait_allrefs(dev);
4059 BUG_ON(atomic_read(&dev->refcnt));
4060 BUG_TRAP(!dev->ip_ptr);
4061 BUG_TRAP(!dev->ip6_ptr);
4062 BUG_TRAP(!dev->dn_ptr);
4064 if (dev->destructor)
4065 dev->destructor(dev);
4067 /* Free network device */
4068 kobject_put(&dev->dev.kobj);
4072 mutex_unlock(&net_todo_run_mutex);
4075 static struct net_device_stats *internal_stats(struct net_device *dev)
4080 static void netdev_init_one_queue(struct net_device *dev,
4081 struct netdev_queue *queue)
4083 spin_lock_init(&queue->lock);
4087 static void netdev_init_queues(struct net_device *dev)
4089 netdev_init_one_queue(dev, &dev->rx_queue);
4090 netdev_init_one_queue(dev, &dev->tx_queue);
4094 * alloc_netdev_mq - allocate network device
4095 * @sizeof_priv: size of private data to allocate space for
4096 * @name: device name format string
4097 * @setup: callback to initialize device
4098 * @queue_count: the number of subqueues to allocate
4100 * Allocates a struct net_device with private data area for driver use
4101 * and performs basic initialization. Also allocates subquue structs
4102 * for each queue on the device at the end of the netdevice.
4104 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4105 void (*setup)(struct net_device *), unsigned int queue_count)
4108 struct net_device *dev;
4111 BUG_ON(strlen(name) >= sizeof(dev->name));
4113 alloc_size = sizeof(struct net_device) +
4114 sizeof(struct net_device_subqueue) * (queue_count - 1);
4116 /* ensure 32-byte alignment of private area */
4117 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4118 alloc_size += sizeof_priv;
4120 /* ensure 32-byte alignment of whole construct */
4121 alloc_size += NETDEV_ALIGN_CONST;
4123 p = kzalloc(alloc_size, GFP_KERNEL);
4125 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4129 dev = (struct net_device *)
4130 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4131 dev->padded = (char *)dev - (char *)p;
4132 dev_net_set(dev, &init_net);
4135 dev->priv = ((char *)dev +
4136 ((sizeof(struct net_device) +
4137 (sizeof(struct net_device_subqueue) *
4138 (queue_count - 1)) + NETDEV_ALIGN_CONST)
4139 & ~NETDEV_ALIGN_CONST));
4142 dev->egress_subqueue_count = queue_count;
4143 dev->gso_max_size = GSO_MAX_SIZE;
4145 netdev_init_queues(dev);
4147 dev->get_stats = internal_stats;
4148 netpoll_netdev_init(dev);
4150 strcpy(dev->name, name);
4153 EXPORT_SYMBOL(alloc_netdev_mq);
4156 * free_netdev - free network device
4159 * This function does the last stage of destroying an allocated device
4160 * interface. The reference to the device object is released.
4161 * If this is the last reference then it will be freed.
4163 void free_netdev(struct net_device *dev)
4165 release_net(dev_net(dev));
4167 /* Compatibility with error handling in drivers */
4168 if (dev->reg_state == NETREG_UNINITIALIZED) {
4169 kfree((char *)dev - dev->padded);
4173 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4174 dev->reg_state = NETREG_RELEASED;
4176 /* will free via device release */
4177 put_device(&dev->dev);
4180 /* Synchronize with packet receive processing. */
4181 void synchronize_net(void)
4188 * unregister_netdevice - remove device from the kernel
4191 * This function shuts down a device interface and removes it
4192 * from the kernel tables.
4194 * Callers must hold the rtnl semaphore. You may want
4195 * unregister_netdev() instead of this.
4198 void unregister_netdevice(struct net_device *dev)
4202 rollback_registered(dev);
4203 /* Finish processing unregister after unlock */
4208 * unregister_netdev - remove device from the kernel
4211 * This function shuts down a device interface and removes it
4212 * from the kernel tables.
4214 * This is just a wrapper for unregister_netdevice that takes
4215 * the rtnl semaphore. In general you want to use this and not
4216 * unregister_netdevice.
4218 void unregister_netdev(struct net_device *dev)
4221 unregister_netdevice(dev);
4225 EXPORT_SYMBOL(unregister_netdev);
4228 * dev_change_net_namespace - move device to different nethost namespace
4230 * @net: network namespace
4231 * @pat: If not NULL name pattern to try if the current device name
4232 * is already taken in the destination network namespace.
4234 * This function shuts down a device interface and moves it
4235 * to a new network namespace. On success 0 is returned, on
4236 * a failure a netagive errno code is returned.
4238 * Callers must hold the rtnl semaphore.
4241 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4244 const char *destname;
4249 /* Don't allow namespace local devices to be moved. */
4251 if (dev->features & NETIF_F_NETNS_LOCAL)
4254 /* Ensure the device has been registrered */
4256 if (dev->reg_state != NETREG_REGISTERED)
4259 /* Get out if there is nothing todo */
4261 if (net_eq(dev_net(dev), net))
4264 /* Pick the destination device name, and ensure
4265 * we can use it in the destination network namespace.
4268 destname = dev->name;
4269 if (__dev_get_by_name(net, destname)) {
4270 /* We get here if we can't use the current device name */
4273 if (!dev_valid_name(pat))
4275 if (strchr(pat, '%')) {
4276 if (__dev_alloc_name(net, pat, buf) < 0)
4281 if (__dev_get_by_name(net, destname))
4286 * And now a mini version of register_netdevice unregister_netdevice.
4289 /* If device is running close it first. */
4292 /* And unlink it from device chain */
4294 unlist_netdevice(dev);
4298 /* Shutdown queueing discipline. */
4301 /* Notify protocols, that we are about to destroy
4302 this device. They should clean all the things.
4304 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4307 * Flush the unicast and multicast chains
4309 dev_addr_discard(dev);
4311 /* Actually switch the network namespace */
4312 dev_net_set(dev, net);
4314 /* Assign the new device name */
4315 if (destname != dev->name)
4316 strcpy(dev->name, destname);
4318 /* If there is an ifindex conflict assign a new one */
4319 if (__dev_get_by_index(net, dev->ifindex)) {
4320 int iflink = (dev->iflink == dev->ifindex);
4321 dev->ifindex = dev_new_index(net);
4323 dev->iflink = dev->ifindex;
4326 /* Fixup kobjects */
4327 netdev_unregister_kobject(dev);
4328 err = netdev_register_kobject(dev);
4331 /* Add the device back in the hashes */
4332 list_netdevice(dev);
4334 /* Notify protocols, that a new device appeared. */
4335 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4343 static int dev_cpu_callback(struct notifier_block *nfb,
4344 unsigned long action,
4347 struct sk_buff **list_skb;
4348 struct netdev_queue **list_net;
4349 struct sk_buff *skb;
4350 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4351 struct softnet_data *sd, *oldsd;
4353 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4356 local_irq_disable();
4357 cpu = smp_processor_id();
4358 sd = &per_cpu(softnet_data, cpu);
4359 oldsd = &per_cpu(softnet_data, oldcpu);
4361 /* Find end of our completion_queue. */
4362 list_skb = &sd->completion_queue;
4364 list_skb = &(*list_skb)->next;
4365 /* Append completion queue from offline CPU. */
4366 *list_skb = oldsd->completion_queue;
4367 oldsd->completion_queue = NULL;
4369 /* Find end of our output_queue. */
4370 list_net = &sd->output_queue;
4372 list_net = &(*list_net)->next_sched;
4373 /* Append output queue from offline CPU. */
4374 *list_net = oldsd->output_queue;
4375 oldsd->output_queue = NULL;
4377 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4380 /* Process offline CPU's input_pkt_queue */
4381 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4387 #ifdef CONFIG_NET_DMA
4389 * net_dma_rebalance - try to maintain one DMA channel per CPU
4390 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4392 * This is called when the number of channels allocated to the net_dma client
4393 * changes. The net_dma client tries to have one DMA channel per CPU.
4396 static void net_dma_rebalance(struct net_dma *net_dma)
4398 unsigned int cpu, i, n, chan_idx;
4399 struct dma_chan *chan;
4401 if (cpus_empty(net_dma->channel_mask)) {
4402 for_each_online_cpu(cpu)
4403 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4408 cpu = first_cpu(cpu_online_map);
4410 for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
4411 chan = net_dma->channels[chan_idx];
4413 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4414 + (i < (num_online_cpus() %
4415 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4418 per_cpu(softnet_data, cpu).net_dma = chan;
4419 cpu = next_cpu(cpu, cpu_online_map);
4427 * netdev_dma_event - event callback for the net_dma_client
4428 * @client: should always be net_dma_client
4429 * @chan: DMA channel for the event
4430 * @state: DMA state to be handled
4432 static enum dma_state_client
4433 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4434 enum dma_state state)
4436 int i, found = 0, pos = -1;
4437 struct net_dma *net_dma =
4438 container_of(client, struct net_dma, client);
4439 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4441 spin_lock(&net_dma->lock);
4443 case DMA_RESOURCE_AVAILABLE:
4444 for (i = 0; i < nr_cpu_ids; i++)
4445 if (net_dma->channels[i] == chan) {
4448 } else if (net_dma->channels[i] == NULL && pos < 0)
4451 if (!found && pos >= 0) {
4453 net_dma->channels[pos] = chan;
4454 cpu_set(pos, net_dma->channel_mask);
4455 net_dma_rebalance(net_dma);
4458 case DMA_RESOURCE_REMOVED:
4459 for (i = 0; i < nr_cpu_ids; i++)
4460 if (net_dma->channels[i] == chan) {
4468 cpu_clear(pos, net_dma->channel_mask);
4469 net_dma->channels[i] = NULL;
4470 net_dma_rebalance(net_dma);
4476 spin_unlock(&net_dma->lock);
4482 * netdev_dma_regiser - register the networking subsystem as a DMA client
4484 static int __init netdev_dma_register(void)
4486 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4488 if (unlikely(!net_dma.channels)) {
4490 "netdev_dma: no memory for net_dma.channels\n");
4493 spin_lock_init(&net_dma.lock);
4494 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4495 dma_async_client_register(&net_dma.client);
4496 dma_async_client_chan_request(&net_dma.client);
4501 static int __init netdev_dma_register(void) { return -ENODEV; }
4502 #endif /* CONFIG_NET_DMA */
4505 * netdev_compute_feature - compute conjunction of two feature sets
4506 * @all: first feature set
4507 * @one: second feature set
4509 * Computes a new feature set after adding a device with feature set
4510 * @one to the master device with current feature set @all. Returns
4511 * the new feature set.
4513 int netdev_compute_features(unsigned long all, unsigned long one)
4515 /* if device needs checksumming, downgrade to hw checksumming */
4516 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4517 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4519 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4520 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4521 all ^= NETIF_F_HW_CSUM
4522 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4524 if (one & NETIF_F_GSO)
4525 one |= NETIF_F_GSO_SOFTWARE;
4528 /* If even one device supports robust GSO, enable it for all. */
4529 if (one & NETIF_F_GSO_ROBUST)
4530 all |= NETIF_F_GSO_ROBUST;
4532 all &= one | NETIF_F_LLTX;
4534 if (!(all & NETIF_F_ALL_CSUM))
4536 if (!(all & NETIF_F_SG))
4537 all &= ~NETIF_F_GSO_MASK;
4541 EXPORT_SYMBOL(netdev_compute_features);
4543 static struct hlist_head *netdev_create_hash(void)
4546 struct hlist_head *hash;
4548 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4550 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4551 INIT_HLIST_HEAD(&hash[i]);
4556 /* Initialize per network namespace state */
4557 static int __net_init netdev_init(struct net *net)
4559 INIT_LIST_HEAD(&net->dev_base_head);
4561 net->dev_name_head = netdev_create_hash();
4562 if (net->dev_name_head == NULL)
4565 net->dev_index_head = netdev_create_hash();
4566 if (net->dev_index_head == NULL)
4572 kfree(net->dev_name_head);
4577 static void __net_exit netdev_exit(struct net *net)
4579 kfree(net->dev_name_head);
4580 kfree(net->dev_index_head);
4583 static struct pernet_operations __net_initdata netdev_net_ops = {
4584 .init = netdev_init,
4585 .exit = netdev_exit,
4588 static void __net_exit default_device_exit(struct net *net)
4590 struct net_device *dev, *next;
4592 * Push all migratable of the network devices back to the
4593 * initial network namespace
4596 for_each_netdev_safe(net, dev, next) {
4598 char fb_name[IFNAMSIZ];
4600 /* Ignore unmoveable devices (i.e. loopback) */
4601 if (dev->features & NETIF_F_NETNS_LOCAL)
4604 /* Push remaing network devices to init_net */
4605 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4606 err = dev_change_net_namespace(dev, &init_net, fb_name);
4608 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4609 __func__, dev->name, err);
4616 static struct pernet_operations __net_initdata default_device_ops = {
4617 .exit = default_device_exit,
4621 * Initialize the DEV module. At boot time this walks the device list and
4622 * unhooks any devices that fail to initialise (normally hardware not
4623 * present) and leaves us with a valid list of present and active devices.
4628 * This is called single threaded during boot, so no need
4629 * to take the rtnl semaphore.
4631 static int __init net_dev_init(void)
4633 int i, rc = -ENOMEM;
4635 BUG_ON(!dev_boot_phase);
4637 if (dev_proc_init())
4640 if (netdev_kobject_init())
4643 INIT_LIST_HEAD(&ptype_all);
4644 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4645 INIT_LIST_HEAD(&ptype_base[i]);
4647 if (register_pernet_subsys(&netdev_net_ops))
4650 if (register_pernet_device(&default_device_ops))
4654 * Initialise the packet receive queues.
4657 for_each_possible_cpu(i) {
4658 struct softnet_data *queue;
4660 queue = &per_cpu(softnet_data, i);
4661 skb_queue_head_init(&queue->input_pkt_queue);
4662 queue->completion_queue = NULL;
4663 INIT_LIST_HEAD(&queue->poll_list);
4665 queue->backlog.poll = process_backlog;
4666 queue->backlog.weight = weight_p;
4669 netdev_dma_register();
4673 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
4674 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
4676 hotcpu_notifier(dev_cpu_callback, 0);
4684 subsys_initcall(net_dev_init);
4686 EXPORT_SYMBOL(__dev_get_by_index);
4687 EXPORT_SYMBOL(__dev_get_by_name);
4688 EXPORT_SYMBOL(__dev_remove_pack);
4689 EXPORT_SYMBOL(dev_valid_name);
4690 EXPORT_SYMBOL(dev_add_pack);
4691 EXPORT_SYMBOL(dev_alloc_name);
4692 EXPORT_SYMBOL(dev_close);
4693 EXPORT_SYMBOL(dev_get_by_flags);
4694 EXPORT_SYMBOL(dev_get_by_index);
4695 EXPORT_SYMBOL(dev_get_by_name);
4696 EXPORT_SYMBOL(dev_open);
4697 EXPORT_SYMBOL(dev_queue_xmit);
4698 EXPORT_SYMBOL(dev_remove_pack);
4699 EXPORT_SYMBOL(dev_set_allmulti);
4700 EXPORT_SYMBOL(dev_set_promiscuity);
4701 EXPORT_SYMBOL(dev_change_flags);
4702 EXPORT_SYMBOL(dev_set_mtu);
4703 EXPORT_SYMBOL(dev_set_mac_address);
4704 EXPORT_SYMBOL(free_netdev);
4705 EXPORT_SYMBOL(netdev_boot_setup_check);
4706 EXPORT_SYMBOL(netdev_set_master);
4707 EXPORT_SYMBOL(netdev_state_change);
4708 EXPORT_SYMBOL(netif_receive_skb);
4709 EXPORT_SYMBOL(netif_rx);
4710 EXPORT_SYMBOL(register_gifconf);
4711 EXPORT_SYMBOL(register_netdevice);
4712 EXPORT_SYMBOL(register_netdevice_notifier);
4713 EXPORT_SYMBOL(skb_checksum_help);
4714 EXPORT_SYMBOL(synchronize_net);
4715 EXPORT_SYMBOL(unregister_netdevice);
4716 EXPORT_SYMBOL(unregister_netdevice_notifier);
4717 EXPORT_SYMBOL(net_enable_timestamp);
4718 EXPORT_SYMBOL(net_disable_timestamp);
4719 EXPORT_SYMBOL(dev_get_flags);
4721 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4722 EXPORT_SYMBOL(br_handle_frame_hook);
4723 EXPORT_SYMBOL(br_fdb_get_hook);
4724 EXPORT_SYMBOL(br_fdb_put_hook);
4728 EXPORT_SYMBOL(dev_load);
4731 EXPORT_PER_CPU_SYMBOL(softnet_data);