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 txq->_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 netdev_queue *txq)
1325 struct net_device *dev = txq->dev;
1327 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1328 struct softnet_data *sd;
1329 unsigned long flags;
1331 local_irq_save(flags);
1332 sd = &__get_cpu_var(softnet_data);
1333 txq->next_sched = sd->output_queue;
1334 sd->output_queue = txq;
1335 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1336 local_irq_restore(flags);
1339 EXPORT_SYMBOL(__netif_schedule);
1341 void dev_kfree_skb_irq(struct sk_buff *skb)
1343 if (atomic_dec_and_test(&skb->users)) {
1344 struct softnet_data *sd;
1345 unsigned long flags;
1347 local_irq_save(flags);
1348 sd = &__get_cpu_var(softnet_data);
1349 skb->next = sd->completion_queue;
1350 sd->completion_queue = skb;
1351 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1352 local_irq_restore(flags);
1355 EXPORT_SYMBOL(dev_kfree_skb_irq);
1357 void dev_kfree_skb_any(struct sk_buff *skb)
1359 if (in_irq() || irqs_disabled())
1360 dev_kfree_skb_irq(skb);
1364 EXPORT_SYMBOL(dev_kfree_skb_any);
1368 * netif_device_detach - mark device as removed
1369 * @dev: network device
1371 * Mark device as removed from system and therefore no longer available.
1373 void netif_device_detach(struct net_device *dev)
1375 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1376 netif_running(dev)) {
1377 netif_stop_queue(dev);
1380 EXPORT_SYMBOL(netif_device_detach);
1383 * netif_device_attach - mark device as attached
1384 * @dev: network device
1386 * Mark device as attached from system and restart if needed.
1388 void netif_device_attach(struct net_device *dev)
1390 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1391 netif_running(dev)) {
1392 netif_wake_queue(dev);
1393 __netdev_watchdog_up(dev);
1396 EXPORT_SYMBOL(netif_device_attach);
1398 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1400 return ((features & NETIF_F_GEN_CSUM) ||
1401 ((features & NETIF_F_IP_CSUM) &&
1402 protocol == htons(ETH_P_IP)) ||
1403 ((features & NETIF_F_IPV6_CSUM) &&
1404 protocol == htons(ETH_P_IPV6)));
1407 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1409 if (can_checksum_protocol(dev->features, skb->protocol))
1412 if (skb->protocol == htons(ETH_P_8021Q)) {
1413 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1414 if (can_checksum_protocol(dev->features & dev->vlan_features,
1415 veh->h_vlan_encapsulated_proto))
1423 * Invalidate hardware checksum when packet is to be mangled, and
1424 * complete checksum manually on outgoing path.
1426 int skb_checksum_help(struct sk_buff *skb)
1429 int ret = 0, offset;
1431 if (skb->ip_summed == CHECKSUM_COMPLETE)
1432 goto out_set_summed;
1434 if (unlikely(skb_shinfo(skb)->gso_size)) {
1435 /* Let GSO fix up the checksum. */
1436 goto out_set_summed;
1439 offset = skb->csum_start - skb_headroom(skb);
1440 BUG_ON(offset >= skb_headlen(skb));
1441 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1443 offset += skb->csum_offset;
1444 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1446 if (skb_cloned(skb) &&
1447 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1448 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1453 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1455 skb->ip_summed = CHECKSUM_NONE;
1461 * skb_gso_segment - Perform segmentation on skb.
1462 * @skb: buffer to segment
1463 * @features: features for the output path (see dev->features)
1465 * This function segments the given skb and returns a list of segments.
1467 * It may return NULL if the skb requires no segmentation. This is
1468 * only possible when GSO is used for verifying header integrity.
1470 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1472 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1473 struct packet_type *ptype;
1474 __be16 type = skb->protocol;
1477 BUG_ON(skb_shinfo(skb)->frag_list);
1479 skb_reset_mac_header(skb);
1480 skb->mac_len = skb->network_header - skb->mac_header;
1481 __skb_pull(skb, skb->mac_len);
1483 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1484 if (skb_header_cloned(skb) &&
1485 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1486 return ERR_PTR(err);
1490 list_for_each_entry_rcu(ptype,
1491 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1492 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1493 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1494 err = ptype->gso_send_check(skb);
1495 segs = ERR_PTR(err);
1496 if (err || skb_gso_ok(skb, features))
1498 __skb_push(skb, (skb->data -
1499 skb_network_header(skb)));
1501 segs = ptype->gso_segment(skb, features);
1507 __skb_push(skb, skb->data - skb_mac_header(skb));
1512 EXPORT_SYMBOL(skb_gso_segment);
1514 /* Take action when hardware reception checksum errors are detected. */
1516 void netdev_rx_csum_fault(struct net_device *dev)
1518 if (net_ratelimit()) {
1519 printk(KERN_ERR "%s: hw csum failure.\n",
1520 dev ? dev->name : "<unknown>");
1524 EXPORT_SYMBOL(netdev_rx_csum_fault);
1527 /* Actually, we should eliminate this check as soon as we know, that:
1528 * 1. IOMMU is present and allows to map all the memory.
1529 * 2. No high memory really exists on this machine.
1532 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1534 #ifdef CONFIG_HIGHMEM
1537 if (dev->features & NETIF_F_HIGHDMA)
1540 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1541 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1549 void (*destructor)(struct sk_buff *skb);
1552 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1554 static void dev_gso_skb_destructor(struct sk_buff *skb)
1556 struct dev_gso_cb *cb;
1559 struct sk_buff *nskb = skb->next;
1561 skb->next = nskb->next;
1564 } while (skb->next);
1566 cb = DEV_GSO_CB(skb);
1568 cb->destructor(skb);
1572 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1573 * @skb: buffer to segment
1575 * This function segments the given skb and stores the list of segments
1578 static int dev_gso_segment(struct sk_buff *skb)
1580 struct net_device *dev = skb->dev;
1581 struct sk_buff *segs;
1582 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1585 segs = skb_gso_segment(skb, features);
1587 /* Verifying header integrity only. */
1592 return PTR_ERR(segs);
1595 DEV_GSO_CB(skb)->destructor = skb->destructor;
1596 skb->destructor = dev_gso_skb_destructor;
1601 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1602 struct netdev_queue *txq)
1604 if (likely(!skb->next)) {
1605 if (!list_empty(&ptype_all))
1606 dev_queue_xmit_nit(skb, dev);
1608 if (netif_needs_gso(dev, skb)) {
1609 if (unlikely(dev_gso_segment(skb)))
1615 return dev->hard_start_xmit(skb, dev);
1620 struct sk_buff *nskb = skb->next;
1623 skb->next = nskb->next;
1625 rc = dev->hard_start_xmit(nskb, dev);
1627 nskb->next = skb->next;
1631 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
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 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1669 struct sk_buff *skb)
1671 u16 queue_index = 0;
1673 if (dev->select_queue)
1674 queue_index = dev->select_queue(dev, skb);
1676 skb_set_queue_mapping(skb, queue_index);
1677 return netdev_get_tx_queue(dev, queue_index);
1680 int dev_queue_xmit(struct sk_buff *skb)
1682 struct net_device *dev = skb->dev;
1683 struct netdev_queue *txq;
1687 /* GSO will handle the following emulations directly. */
1688 if (netif_needs_gso(dev, skb))
1691 if (skb_shinfo(skb)->frag_list &&
1692 !(dev->features & NETIF_F_FRAGLIST) &&
1693 __skb_linearize(skb))
1696 /* Fragmented skb is linearized if device does not support SG,
1697 * or if at least one of fragments is in highmem and device
1698 * does not support DMA from it.
1700 if (skb_shinfo(skb)->nr_frags &&
1701 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1702 __skb_linearize(skb))
1705 /* If packet is not checksummed and device does not support
1706 * checksumming for this protocol, complete checksumming here.
1708 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1709 skb_set_transport_header(skb, skb->csum_start -
1711 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1716 /* Disable soft irqs for various locks below. Also
1717 * stops preemption for RCU.
1721 txq = dev_pick_tx(dev, skb);
1722 spin_lock_prefetch(&txq->lock);
1724 /* Updates of qdisc are serialized by queue->lock.
1725 * The struct Qdisc which is pointed to by qdisc is now a
1726 * rcu structure - it may be accessed without acquiring
1727 * a lock (but the structure may be stale.) The freeing of the
1728 * qdisc will be deferred until it's known that there are no
1729 * more references to it.
1731 * If the qdisc has an enqueue function, we still need to
1732 * hold the queue->lock before calling it, since queue->lock
1733 * also serializes access to the device queue.
1736 q = rcu_dereference(txq->qdisc);
1737 #ifdef CONFIG_NET_CLS_ACT
1738 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1741 /* Grab device queue */
1742 spin_lock(&txq->lock);
1745 rc = q->enqueue(skb, q);
1747 spin_unlock(&txq->lock);
1749 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1752 spin_unlock(&txq->lock);
1755 /* The device has no queue. Common case for software devices:
1756 loopback, all the sorts of tunnels...
1758 Really, it is unlikely that netif_tx_lock protection is necessary
1759 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1761 However, it is possible, that they rely on protection
1764 Check this and shot the lock. It is not prone from deadlocks.
1765 Either shot noqueue qdisc, it is even simpler 8)
1767 if (dev->flags & IFF_UP) {
1768 int cpu = smp_processor_id(); /* ok because BHs are off */
1770 if (txq->xmit_lock_owner != cpu) {
1772 HARD_TX_LOCK(dev, txq, cpu);
1774 if (!netif_tx_queue_stopped(txq)) {
1776 if (!dev_hard_start_xmit(skb, dev, txq)) {
1777 HARD_TX_UNLOCK(dev, txq);
1781 HARD_TX_UNLOCK(dev, txq);
1782 if (net_ratelimit())
1783 printk(KERN_CRIT "Virtual device %s asks to "
1784 "queue packet!\n", dev->name);
1786 /* Recursion is detected! It is possible,
1788 if (net_ratelimit())
1789 printk(KERN_CRIT "Dead loop on virtual device "
1790 "%s, fix it urgently!\n", dev->name);
1795 rcu_read_unlock_bh();
1801 rcu_read_unlock_bh();
1806 /*=======================================================================
1808 =======================================================================*/
1810 int netdev_max_backlog __read_mostly = 1000;
1811 int netdev_budget __read_mostly = 300;
1812 int weight_p __read_mostly = 64; /* old backlog weight */
1814 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1818 * netif_rx - post buffer to the network code
1819 * @skb: buffer to post
1821 * This function receives a packet from a device driver and queues it for
1822 * the upper (protocol) levels to process. It always succeeds. The buffer
1823 * may be dropped during processing for congestion control or by the
1827 * NET_RX_SUCCESS (no congestion)
1828 * NET_RX_DROP (packet was dropped)
1832 int netif_rx(struct sk_buff *skb)
1834 struct softnet_data *queue;
1835 unsigned long flags;
1837 /* if netpoll wants it, pretend we never saw it */
1838 if (netpoll_rx(skb))
1841 if (!skb->tstamp.tv64)
1845 * The code is rearranged so that the path is the most
1846 * short when CPU is congested, but is still operating.
1848 local_irq_save(flags);
1849 queue = &__get_cpu_var(softnet_data);
1851 __get_cpu_var(netdev_rx_stat).total++;
1852 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1853 if (queue->input_pkt_queue.qlen) {
1856 __skb_queue_tail(&queue->input_pkt_queue, skb);
1857 local_irq_restore(flags);
1858 return NET_RX_SUCCESS;
1861 napi_schedule(&queue->backlog);
1865 __get_cpu_var(netdev_rx_stat).dropped++;
1866 local_irq_restore(flags);
1872 int netif_rx_ni(struct sk_buff *skb)
1877 err = netif_rx(skb);
1878 if (local_softirq_pending())
1885 EXPORT_SYMBOL(netif_rx_ni);
1887 static inline struct net_device *skb_bond(struct sk_buff *skb)
1889 struct net_device *dev = skb->dev;
1892 if (skb_bond_should_drop(skb)) {
1896 skb->dev = dev->master;
1903 static void net_tx_action(struct softirq_action *h)
1905 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1907 if (sd->completion_queue) {
1908 struct sk_buff *clist;
1910 local_irq_disable();
1911 clist = sd->completion_queue;
1912 sd->completion_queue = NULL;
1916 struct sk_buff *skb = clist;
1917 clist = clist->next;
1919 BUG_TRAP(!atomic_read(&skb->users));
1924 if (sd->output_queue) {
1925 struct netdev_queue *head;
1927 local_irq_disable();
1928 head = sd->output_queue;
1929 sd->output_queue = NULL;
1933 struct netdev_queue *txq = head;
1934 struct net_device *dev = txq->dev;
1935 head = head->next_sched;
1937 smp_mb__before_clear_bit();
1938 clear_bit(__LINK_STATE_SCHED, &dev->state);
1940 if (spin_trylock(&txq->lock)) {
1942 spin_unlock(&txq->lock);
1944 netif_schedule_queue(txq);
1950 static inline int deliver_skb(struct sk_buff *skb,
1951 struct packet_type *pt_prev,
1952 struct net_device *orig_dev)
1954 atomic_inc(&skb->users);
1955 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1958 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1959 /* These hooks defined here for ATM */
1961 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1962 unsigned char *addr);
1963 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1966 * If bridge module is loaded call bridging hook.
1967 * returns NULL if packet was consumed.
1969 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1970 struct sk_buff *skb) __read_mostly;
1971 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1972 struct packet_type **pt_prev, int *ret,
1973 struct net_device *orig_dev)
1975 struct net_bridge_port *port;
1977 if (skb->pkt_type == PACKET_LOOPBACK ||
1978 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1982 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1986 return br_handle_frame_hook(port, skb);
1989 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1992 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1993 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
1994 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
1996 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
1997 struct packet_type **pt_prev,
1999 struct net_device *orig_dev)
2001 if (skb->dev->macvlan_port == NULL)
2005 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2008 return macvlan_handle_frame_hook(skb);
2011 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2014 #ifdef CONFIG_NET_CLS_ACT
2015 /* TODO: Maybe we should just force sch_ingress to be compiled in
2016 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2017 * a compare and 2 stores extra right now if we dont have it on
2018 * but have CONFIG_NET_CLS_ACT
2019 * NOTE: This doesnt stop any functionality; if you dont have
2020 * the ingress scheduler, you just cant add policies on ingress.
2023 static int ing_filter(struct sk_buff *skb)
2025 struct net_device *dev = skb->dev;
2026 u32 ttl = G_TC_RTTL(skb->tc_verd);
2027 struct netdev_queue *rxq;
2028 int result = TC_ACT_OK;
2031 if (MAX_RED_LOOP < ttl++) {
2033 "Redir loop detected Dropping packet (%d->%d)\n",
2034 skb->iif, dev->ifindex);
2038 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2039 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2041 rxq = &dev->rx_queue;
2043 spin_lock(&rxq->lock);
2044 if ((q = rxq->qdisc) != NULL)
2045 result = q->enqueue(skb, q);
2046 spin_unlock(&rxq->lock);
2051 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2052 struct packet_type **pt_prev,
2053 int *ret, struct net_device *orig_dev)
2055 if (!skb->dev->rx_queue.qdisc)
2059 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2062 /* Huh? Why does turning on AF_PACKET affect this? */
2063 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2066 switch (ing_filter(skb)) {
2080 * netif_nit_deliver - deliver received packets to network taps
2083 * This function is used to deliver incoming packets to network
2084 * taps. It should be used when the normal netif_receive_skb path
2085 * is bypassed, for example because of VLAN acceleration.
2087 void netif_nit_deliver(struct sk_buff *skb)
2089 struct packet_type *ptype;
2091 if (list_empty(&ptype_all))
2094 skb_reset_network_header(skb);
2095 skb_reset_transport_header(skb);
2096 skb->mac_len = skb->network_header - skb->mac_header;
2099 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2100 if (!ptype->dev || ptype->dev == skb->dev)
2101 deliver_skb(skb, ptype, skb->dev);
2107 * netif_receive_skb - process receive buffer from network
2108 * @skb: buffer to process
2110 * netif_receive_skb() is the main receive data processing function.
2111 * It always succeeds. The buffer may be dropped during processing
2112 * for congestion control or by the protocol layers.
2114 * This function may only be called from softirq context and interrupts
2115 * should be enabled.
2117 * Return values (usually ignored):
2118 * NET_RX_SUCCESS: no congestion
2119 * NET_RX_DROP: packet was dropped
2121 int netif_receive_skb(struct sk_buff *skb)
2123 struct packet_type *ptype, *pt_prev;
2124 struct net_device *orig_dev;
2125 int ret = NET_RX_DROP;
2128 /* if we've gotten here through NAPI, check netpoll */
2129 if (netpoll_receive_skb(skb))
2132 if (!skb->tstamp.tv64)
2136 skb->iif = skb->dev->ifindex;
2138 orig_dev = skb_bond(skb);
2143 __get_cpu_var(netdev_rx_stat).total++;
2145 skb_reset_network_header(skb);
2146 skb_reset_transport_header(skb);
2147 skb->mac_len = skb->network_header - skb->mac_header;
2153 /* Don't receive packets in an exiting network namespace */
2154 if (!net_alive(dev_net(skb->dev)))
2157 #ifdef CONFIG_NET_CLS_ACT
2158 if (skb->tc_verd & TC_NCLS) {
2159 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2164 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2165 if (!ptype->dev || ptype->dev == skb->dev) {
2167 ret = deliver_skb(skb, pt_prev, orig_dev);
2172 #ifdef CONFIG_NET_CLS_ACT
2173 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2179 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2182 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2186 type = skb->protocol;
2187 list_for_each_entry_rcu(ptype,
2188 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2189 if (ptype->type == type &&
2190 (!ptype->dev || ptype->dev == skb->dev)) {
2192 ret = deliver_skb(skb, pt_prev, orig_dev);
2198 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2201 /* Jamal, now you will not able to escape explaining
2202 * me how you were going to use this. :-)
2212 static int process_backlog(struct napi_struct *napi, int quota)
2215 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2216 unsigned long start_time = jiffies;
2218 napi->weight = weight_p;
2220 struct sk_buff *skb;
2221 struct net_device *dev;
2223 local_irq_disable();
2224 skb = __skb_dequeue(&queue->input_pkt_queue);
2226 __napi_complete(napi);
2235 netif_receive_skb(skb);
2238 } while (++work < quota && jiffies == start_time);
2244 * __napi_schedule - schedule for receive
2245 * @n: entry to schedule
2247 * The entry's receive function will be scheduled to run
2249 void __napi_schedule(struct napi_struct *n)
2251 unsigned long flags;
2253 local_irq_save(flags);
2254 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2255 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2256 local_irq_restore(flags);
2258 EXPORT_SYMBOL(__napi_schedule);
2261 static void net_rx_action(struct softirq_action *h)
2263 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2264 unsigned long start_time = jiffies;
2265 int budget = netdev_budget;
2268 local_irq_disable();
2270 while (!list_empty(list)) {
2271 struct napi_struct *n;
2274 /* If softirq window is exhuasted then punt.
2276 * Note that this is a slight policy change from the
2277 * previous NAPI code, which would allow up to 2
2278 * jiffies to pass before breaking out. The test
2279 * used to be "jiffies - start_time > 1".
2281 if (unlikely(budget <= 0 || jiffies != start_time))
2286 /* Even though interrupts have been re-enabled, this
2287 * access is safe because interrupts can only add new
2288 * entries to the tail of this list, and only ->poll()
2289 * calls can remove this head entry from the list.
2291 n = list_entry(list->next, struct napi_struct, poll_list);
2293 have = netpoll_poll_lock(n);
2297 /* This NAPI_STATE_SCHED test is for avoiding a race
2298 * with netpoll's poll_napi(). Only the entity which
2299 * obtains the lock and sees NAPI_STATE_SCHED set will
2300 * actually make the ->poll() call. Therefore we avoid
2301 * accidently calling ->poll() when NAPI is not scheduled.
2304 if (test_bit(NAPI_STATE_SCHED, &n->state))
2305 work = n->poll(n, weight);
2307 WARN_ON_ONCE(work > weight);
2311 local_irq_disable();
2313 /* Drivers must not modify the NAPI state if they
2314 * consume the entire weight. In such cases this code
2315 * still "owns" the NAPI instance and therefore can
2316 * move the instance around on the list at-will.
2318 if (unlikely(work == weight)) {
2319 if (unlikely(napi_disable_pending(n)))
2322 list_move_tail(&n->poll_list, list);
2325 netpoll_poll_unlock(have);
2330 #ifdef CONFIG_NET_DMA
2332 * There may not be any more sk_buffs coming right now, so push
2333 * any pending DMA copies to hardware
2335 if (!cpus_empty(net_dma.channel_mask)) {
2337 for_each_cpu_mask(chan_idx, net_dma.channel_mask) {
2338 struct dma_chan *chan = net_dma.channels[chan_idx];
2340 dma_async_memcpy_issue_pending(chan);
2348 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2349 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2353 static gifconf_func_t * gifconf_list [NPROTO];
2356 * register_gifconf - register a SIOCGIF handler
2357 * @family: Address family
2358 * @gifconf: Function handler
2360 * Register protocol dependent address dumping routines. The handler
2361 * that is passed must not be freed or reused until it has been replaced
2362 * by another handler.
2364 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2366 if (family >= NPROTO)
2368 gifconf_list[family] = gifconf;
2374 * Map an interface index to its name (SIOCGIFNAME)
2378 * We need this ioctl for efficient implementation of the
2379 * if_indextoname() function required by the IPv6 API. Without
2380 * it, we would have to search all the interfaces to find a
2384 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2386 struct net_device *dev;
2390 * Fetch the caller's info block.
2393 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2396 read_lock(&dev_base_lock);
2397 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2399 read_unlock(&dev_base_lock);
2403 strcpy(ifr.ifr_name, dev->name);
2404 read_unlock(&dev_base_lock);
2406 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2412 * Perform a SIOCGIFCONF call. This structure will change
2413 * size eventually, and there is nothing I can do about it.
2414 * Thus we will need a 'compatibility mode'.
2417 static int dev_ifconf(struct net *net, char __user *arg)
2420 struct net_device *dev;
2427 * Fetch the caller's info block.
2430 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2437 * Loop over the interfaces, and write an info block for each.
2441 for_each_netdev(net, dev) {
2442 for (i = 0; i < NPROTO; i++) {
2443 if (gifconf_list[i]) {
2446 done = gifconf_list[i](dev, NULL, 0);
2448 done = gifconf_list[i](dev, pos + total,
2458 * All done. Write the updated control block back to the caller.
2460 ifc.ifc_len = total;
2463 * Both BSD and Solaris return 0 here, so we do too.
2465 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2468 #ifdef CONFIG_PROC_FS
2470 * This is invoked by the /proc filesystem handler to display a device
2473 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2474 __acquires(dev_base_lock)
2476 struct net *net = seq_file_net(seq);
2478 struct net_device *dev;
2480 read_lock(&dev_base_lock);
2482 return SEQ_START_TOKEN;
2485 for_each_netdev(net, dev)
2492 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2494 struct net *net = seq_file_net(seq);
2496 return v == SEQ_START_TOKEN ?
2497 first_net_device(net) : next_net_device((struct net_device *)v);
2500 void dev_seq_stop(struct seq_file *seq, void *v)
2501 __releases(dev_base_lock)
2503 read_unlock(&dev_base_lock);
2506 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2508 struct net_device_stats *stats = dev->get_stats(dev);
2510 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2511 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2512 dev->name, stats->rx_bytes, stats->rx_packets,
2514 stats->rx_dropped + stats->rx_missed_errors,
2515 stats->rx_fifo_errors,
2516 stats->rx_length_errors + stats->rx_over_errors +
2517 stats->rx_crc_errors + stats->rx_frame_errors,
2518 stats->rx_compressed, stats->multicast,
2519 stats->tx_bytes, stats->tx_packets,
2520 stats->tx_errors, stats->tx_dropped,
2521 stats->tx_fifo_errors, stats->collisions,
2522 stats->tx_carrier_errors +
2523 stats->tx_aborted_errors +
2524 stats->tx_window_errors +
2525 stats->tx_heartbeat_errors,
2526 stats->tx_compressed);
2530 * Called from the PROCfs module. This now uses the new arbitrary sized
2531 * /proc/net interface to create /proc/net/dev
2533 static int dev_seq_show(struct seq_file *seq, void *v)
2535 if (v == SEQ_START_TOKEN)
2536 seq_puts(seq, "Inter-| Receive "
2538 " face |bytes packets errs drop fifo frame "
2539 "compressed multicast|bytes packets errs "
2540 "drop fifo colls carrier compressed\n");
2542 dev_seq_printf_stats(seq, v);
2546 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2548 struct netif_rx_stats *rc = NULL;
2550 while (*pos < nr_cpu_ids)
2551 if (cpu_online(*pos)) {
2552 rc = &per_cpu(netdev_rx_stat, *pos);
2559 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2561 return softnet_get_online(pos);
2564 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2567 return softnet_get_online(pos);
2570 static void softnet_seq_stop(struct seq_file *seq, void *v)
2574 static int softnet_seq_show(struct seq_file *seq, void *v)
2576 struct netif_rx_stats *s = v;
2578 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2579 s->total, s->dropped, s->time_squeeze, 0,
2580 0, 0, 0, 0, /* was fastroute */
2585 static const struct seq_operations dev_seq_ops = {
2586 .start = dev_seq_start,
2587 .next = dev_seq_next,
2588 .stop = dev_seq_stop,
2589 .show = dev_seq_show,
2592 static int dev_seq_open(struct inode *inode, struct file *file)
2594 return seq_open_net(inode, file, &dev_seq_ops,
2595 sizeof(struct seq_net_private));
2598 static const struct file_operations dev_seq_fops = {
2599 .owner = THIS_MODULE,
2600 .open = dev_seq_open,
2602 .llseek = seq_lseek,
2603 .release = seq_release_net,
2606 static const struct seq_operations softnet_seq_ops = {
2607 .start = softnet_seq_start,
2608 .next = softnet_seq_next,
2609 .stop = softnet_seq_stop,
2610 .show = softnet_seq_show,
2613 static int softnet_seq_open(struct inode *inode, struct file *file)
2615 return seq_open(file, &softnet_seq_ops);
2618 static const struct file_operations softnet_seq_fops = {
2619 .owner = THIS_MODULE,
2620 .open = softnet_seq_open,
2622 .llseek = seq_lseek,
2623 .release = seq_release,
2626 static void *ptype_get_idx(loff_t pos)
2628 struct packet_type *pt = NULL;
2632 list_for_each_entry_rcu(pt, &ptype_all, list) {
2638 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2639 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2648 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2652 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2655 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2657 struct packet_type *pt;
2658 struct list_head *nxt;
2662 if (v == SEQ_START_TOKEN)
2663 return ptype_get_idx(0);
2666 nxt = pt->list.next;
2667 if (pt->type == htons(ETH_P_ALL)) {
2668 if (nxt != &ptype_all)
2671 nxt = ptype_base[0].next;
2673 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2675 while (nxt == &ptype_base[hash]) {
2676 if (++hash >= PTYPE_HASH_SIZE)
2678 nxt = ptype_base[hash].next;
2681 return list_entry(nxt, struct packet_type, list);
2684 static void ptype_seq_stop(struct seq_file *seq, void *v)
2690 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2692 #ifdef CONFIG_KALLSYMS
2693 unsigned long offset = 0, symsize;
2694 const char *symname;
2698 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2705 modname = delim = "";
2706 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2712 seq_printf(seq, "[%p]", sym);
2715 static int ptype_seq_show(struct seq_file *seq, void *v)
2717 struct packet_type *pt = v;
2719 if (v == SEQ_START_TOKEN)
2720 seq_puts(seq, "Type Device Function\n");
2721 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2722 if (pt->type == htons(ETH_P_ALL))
2723 seq_puts(seq, "ALL ");
2725 seq_printf(seq, "%04x", ntohs(pt->type));
2727 seq_printf(seq, " %-8s ",
2728 pt->dev ? pt->dev->name : "");
2729 ptype_seq_decode(seq, pt->func);
2730 seq_putc(seq, '\n');
2736 static const struct seq_operations ptype_seq_ops = {
2737 .start = ptype_seq_start,
2738 .next = ptype_seq_next,
2739 .stop = ptype_seq_stop,
2740 .show = ptype_seq_show,
2743 static int ptype_seq_open(struct inode *inode, struct file *file)
2745 return seq_open_net(inode, file, &ptype_seq_ops,
2746 sizeof(struct seq_net_private));
2749 static const struct file_operations ptype_seq_fops = {
2750 .owner = THIS_MODULE,
2751 .open = ptype_seq_open,
2753 .llseek = seq_lseek,
2754 .release = seq_release_net,
2758 static int __net_init dev_proc_net_init(struct net *net)
2762 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2764 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2766 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2769 if (wext_proc_init(net))
2775 proc_net_remove(net, "ptype");
2777 proc_net_remove(net, "softnet_stat");
2779 proc_net_remove(net, "dev");
2783 static void __net_exit dev_proc_net_exit(struct net *net)
2785 wext_proc_exit(net);
2787 proc_net_remove(net, "ptype");
2788 proc_net_remove(net, "softnet_stat");
2789 proc_net_remove(net, "dev");
2792 static struct pernet_operations __net_initdata dev_proc_ops = {
2793 .init = dev_proc_net_init,
2794 .exit = dev_proc_net_exit,
2797 static int __init dev_proc_init(void)
2799 return register_pernet_subsys(&dev_proc_ops);
2802 #define dev_proc_init() 0
2803 #endif /* CONFIG_PROC_FS */
2807 * netdev_set_master - set up master/slave pair
2808 * @slave: slave device
2809 * @master: new master device
2811 * Changes the master device of the slave. Pass %NULL to break the
2812 * bonding. The caller must hold the RTNL semaphore. On a failure
2813 * a negative errno code is returned. On success the reference counts
2814 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2815 * function returns zero.
2817 int netdev_set_master(struct net_device *slave, struct net_device *master)
2819 struct net_device *old = slave->master;
2829 slave->master = master;
2837 slave->flags |= IFF_SLAVE;
2839 slave->flags &= ~IFF_SLAVE;
2841 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2845 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2847 unsigned short old_flags = dev->flags;
2851 dev->flags |= IFF_PROMISC;
2852 dev->promiscuity += inc;
2853 if (dev->promiscuity == 0) {
2856 * If inc causes overflow, untouch promisc and return error.
2859 dev->flags &= ~IFF_PROMISC;
2861 dev->promiscuity -= inc;
2862 printk(KERN_WARNING "%s: promiscuity touches roof, "
2863 "set promiscuity failed, promiscuity feature "
2864 "of device might be broken.\n", dev->name);
2868 if (dev->flags != old_flags) {
2869 printk(KERN_INFO "device %s %s promiscuous mode\n",
2870 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2873 audit_log(current->audit_context, GFP_ATOMIC,
2874 AUDIT_ANOM_PROMISCUOUS,
2875 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2876 dev->name, (dev->flags & IFF_PROMISC),
2877 (old_flags & IFF_PROMISC),
2878 audit_get_loginuid(current),
2879 current->uid, current->gid,
2880 audit_get_sessionid(current));
2882 if (dev->change_rx_flags)
2883 dev->change_rx_flags(dev, IFF_PROMISC);
2889 * dev_set_promiscuity - update promiscuity count on a device
2893 * Add or remove promiscuity from a device. While the count in the device
2894 * remains above zero the interface remains promiscuous. Once it hits zero
2895 * the device reverts back to normal filtering operation. A negative inc
2896 * value is used to drop promiscuity on the device.
2897 * Return 0 if successful or a negative errno code on error.
2899 int dev_set_promiscuity(struct net_device *dev, int inc)
2901 unsigned short old_flags = dev->flags;
2904 err = __dev_set_promiscuity(dev, inc);
2907 if (dev->flags != old_flags)
2908 dev_set_rx_mode(dev);
2913 * dev_set_allmulti - update allmulti count on a device
2917 * Add or remove reception of all multicast frames to a device. While the
2918 * count in the device remains above zero the interface remains listening
2919 * to all interfaces. Once it hits zero the device reverts back to normal
2920 * filtering operation. A negative @inc value is used to drop the counter
2921 * when releasing a resource needing all multicasts.
2922 * Return 0 if successful or a negative errno code on error.
2925 int dev_set_allmulti(struct net_device *dev, int inc)
2927 unsigned short old_flags = dev->flags;
2931 dev->flags |= IFF_ALLMULTI;
2932 dev->allmulti += inc;
2933 if (dev->allmulti == 0) {
2936 * If inc causes overflow, untouch allmulti and return error.
2939 dev->flags &= ~IFF_ALLMULTI;
2941 dev->allmulti -= inc;
2942 printk(KERN_WARNING "%s: allmulti touches roof, "
2943 "set allmulti failed, allmulti feature of "
2944 "device might be broken.\n", dev->name);
2948 if (dev->flags ^ old_flags) {
2949 if (dev->change_rx_flags)
2950 dev->change_rx_flags(dev, IFF_ALLMULTI);
2951 dev_set_rx_mode(dev);
2957 * Upload unicast and multicast address lists to device and
2958 * configure RX filtering. When the device doesn't support unicast
2959 * filtering it is put in promiscuous mode while unicast addresses
2962 void __dev_set_rx_mode(struct net_device *dev)
2964 /* dev_open will call this function so the list will stay sane. */
2965 if (!(dev->flags&IFF_UP))
2968 if (!netif_device_present(dev))
2971 if (dev->set_rx_mode)
2972 dev->set_rx_mode(dev);
2974 /* Unicast addresses changes may only happen under the rtnl,
2975 * therefore calling __dev_set_promiscuity here is safe.
2977 if (dev->uc_count > 0 && !dev->uc_promisc) {
2978 __dev_set_promiscuity(dev, 1);
2979 dev->uc_promisc = 1;
2980 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2981 __dev_set_promiscuity(dev, -1);
2982 dev->uc_promisc = 0;
2985 if (dev->set_multicast_list)
2986 dev->set_multicast_list(dev);
2990 void dev_set_rx_mode(struct net_device *dev)
2992 netif_addr_lock_bh(dev);
2993 __dev_set_rx_mode(dev);
2994 netif_addr_unlock_bh(dev);
2997 int __dev_addr_delete(struct dev_addr_list **list, int *count,
2998 void *addr, int alen, int glbl)
3000 struct dev_addr_list *da;
3002 for (; (da = *list) != NULL; list = &da->next) {
3003 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3004 alen == da->da_addrlen) {
3006 int old_glbl = da->da_gusers;
3023 int __dev_addr_add(struct dev_addr_list **list, int *count,
3024 void *addr, int alen, int glbl)
3026 struct dev_addr_list *da;
3028 for (da = *list; da != NULL; da = da->next) {
3029 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3030 da->da_addrlen == alen) {
3032 int old_glbl = da->da_gusers;
3042 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3045 memcpy(da->da_addr, addr, alen);
3046 da->da_addrlen = alen;
3048 da->da_gusers = glbl ? 1 : 0;
3056 * dev_unicast_delete - Release secondary unicast address.
3058 * @addr: address to delete
3059 * @alen: length of @addr
3061 * Release reference to a secondary unicast address and remove it
3062 * from the device if the reference count drops to zero.
3064 * The caller must hold the rtnl_mutex.
3066 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3072 netif_addr_lock_bh(dev);
3073 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3075 __dev_set_rx_mode(dev);
3076 netif_addr_unlock_bh(dev);
3079 EXPORT_SYMBOL(dev_unicast_delete);
3082 * dev_unicast_add - add a secondary unicast address
3084 * @addr: address to add
3085 * @alen: length of @addr
3087 * Add a secondary unicast address to the device or increase
3088 * the reference count if it already exists.
3090 * The caller must hold the rtnl_mutex.
3092 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3098 netif_addr_lock_bh(dev);
3099 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3101 __dev_set_rx_mode(dev);
3102 netif_addr_unlock_bh(dev);
3105 EXPORT_SYMBOL(dev_unicast_add);
3107 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3108 struct dev_addr_list **from, int *from_count)
3110 struct dev_addr_list *da, *next;
3114 while (da != NULL) {
3116 if (!da->da_synced) {
3117 err = __dev_addr_add(to, to_count,
3118 da->da_addr, da->da_addrlen, 0);
3123 } else if (da->da_users == 1) {
3124 __dev_addr_delete(to, to_count,
3125 da->da_addr, da->da_addrlen, 0);
3126 __dev_addr_delete(from, from_count,
3127 da->da_addr, da->da_addrlen, 0);
3134 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3135 struct dev_addr_list **from, int *from_count)
3137 struct dev_addr_list *da, *next;
3140 while (da != NULL) {
3142 if (da->da_synced) {
3143 __dev_addr_delete(to, to_count,
3144 da->da_addr, da->da_addrlen, 0);
3146 __dev_addr_delete(from, from_count,
3147 da->da_addr, da->da_addrlen, 0);
3154 * dev_unicast_sync - Synchronize device's unicast list to another device
3155 * @to: destination device
3156 * @from: source device
3158 * Add newly added addresses to the destination device and release
3159 * addresses that have no users left. The source device must be
3160 * locked by netif_tx_lock_bh.
3162 * This function is intended to be called from the dev->set_rx_mode
3163 * function of layered software devices.
3165 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3169 netif_addr_lock_bh(to);
3170 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3171 &from->uc_list, &from->uc_count);
3173 __dev_set_rx_mode(to);
3174 netif_addr_unlock_bh(to);
3177 EXPORT_SYMBOL(dev_unicast_sync);
3180 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3181 * @to: destination device
3182 * @from: source device
3184 * Remove all addresses that were added to the destination device by
3185 * dev_unicast_sync(). This function is intended to be called from the
3186 * dev->stop function of layered software devices.
3188 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3190 netif_addr_lock_bh(from);
3191 netif_addr_lock(to);
3193 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3194 &from->uc_list, &from->uc_count);
3195 __dev_set_rx_mode(to);
3197 netif_addr_unlock(to);
3198 netif_addr_unlock_bh(from);
3200 EXPORT_SYMBOL(dev_unicast_unsync);
3202 static void __dev_addr_discard(struct dev_addr_list **list)
3204 struct dev_addr_list *tmp;
3206 while (*list != NULL) {
3209 if (tmp->da_users > tmp->da_gusers)
3210 printk("__dev_addr_discard: address leakage! "
3211 "da_users=%d\n", tmp->da_users);
3216 static void dev_addr_discard(struct net_device *dev)
3218 netif_addr_lock_bh(dev);
3220 __dev_addr_discard(&dev->uc_list);
3223 __dev_addr_discard(&dev->mc_list);
3226 netif_addr_unlock_bh(dev);
3229 unsigned dev_get_flags(const struct net_device *dev)
3233 flags = (dev->flags & ~(IFF_PROMISC |
3238 (dev->gflags & (IFF_PROMISC |
3241 if (netif_running(dev)) {
3242 if (netif_oper_up(dev))
3243 flags |= IFF_RUNNING;
3244 if (netif_carrier_ok(dev))
3245 flags |= IFF_LOWER_UP;
3246 if (netif_dormant(dev))
3247 flags |= IFF_DORMANT;
3253 int dev_change_flags(struct net_device *dev, unsigned flags)
3256 int old_flags = dev->flags;
3261 * Set the flags on our device.
3264 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3265 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3267 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3271 * Load in the correct multicast list now the flags have changed.
3274 if (dev->change_rx_flags && (old_flags ^ flags) & IFF_MULTICAST)
3275 dev->change_rx_flags(dev, IFF_MULTICAST);
3277 dev_set_rx_mode(dev);
3280 * Have we downed the interface. We handle IFF_UP ourselves
3281 * according to user attempts to set it, rather than blindly
3286 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3287 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3290 dev_set_rx_mode(dev);
3293 if (dev->flags & IFF_UP &&
3294 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3296 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3298 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3299 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3300 dev->gflags ^= IFF_PROMISC;
3301 dev_set_promiscuity(dev, inc);
3304 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3305 is important. Some (broken) drivers set IFF_PROMISC, when
3306 IFF_ALLMULTI is requested not asking us and not reporting.
3308 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3309 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3310 dev->gflags ^= IFF_ALLMULTI;
3311 dev_set_allmulti(dev, inc);
3314 /* Exclude state transition flags, already notified */
3315 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3317 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3322 int dev_set_mtu(struct net_device *dev, int new_mtu)
3326 if (new_mtu == dev->mtu)
3329 /* MTU must be positive. */
3333 if (!netif_device_present(dev))
3337 if (dev->change_mtu)
3338 err = dev->change_mtu(dev, new_mtu);
3341 if (!err && dev->flags & IFF_UP)
3342 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3346 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3350 if (!dev->set_mac_address)
3352 if (sa->sa_family != dev->type)
3354 if (!netif_device_present(dev))
3356 err = dev->set_mac_address(dev, sa);
3358 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3363 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3365 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3368 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3374 case SIOCGIFFLAGS: /* Get interface flags */
3375 ifr->ifr_flags = dev_get_flags(dev);
3378 case SIOCGIFMETRIC: /* Get the metric on the interface
3379 (currently unused) */
3380 ifr->ifr_metric = 0;
3383 case SIOCGIFMTU: /* Get the MTU of a device */
3384 ifr->ifr_mtu = dev->mtu;
3389 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3391 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3392 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3393 ifr->ifr_hwaddr.sa_family = dev->type;
3401 ifr->ifr_map.mem_start = dev->mem_start;
3402 ifr->ifr_map.mem_end = dev->mem_end;
3403 ifr->ifr_map.base_addr = dev->base_addr;
3404 ifr->ifr_map.irq = dev->irq;
3405 ifr->ifr_map.dma = dev->dma;
3406 ifr->ifr_map.port = dev->if_port;
3410 ifr->ifr_ifindex = dev->ifindex;
3414 ifr->ifr_qlen = dev->tx_queue_len;
3418 /* dev_ioctl() should ensure this case
3430 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3432 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3435 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3441 case SIOCSIFFLAGS: /* Set interface flags */
3442 return dev_change_flags(dev, ifr->ifr_flags);
3444 case SIOCSIFMETRIC: /* Set the metric on the interface
3445 (currently unused) */
3448 case SIOCSIFMTU: /* Set the MTU of a device */
3449 return dev_set_mtu(dev, ifr->ifr_mtu);
3452 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3454 case SIOCSIFHWBROADCAST:
3455 if (ifr->ifr_hwaddr.sa_family != dev->type)
3457 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3458 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3459 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3463 if (dev->set_config) {
3464 if (!netif_device_present(dev))
3466 return dev->set_config(dev, &ifr->ifr_map);
3471 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3472 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3474 if (!netif_device_present(dev))
3476 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3480 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3481 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3483 if (!netif_device_present(dev))
3485 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3489 if (ifr->ifr_qlen < 0)
3491 dev->tx_queue_len = ifr->ifr_qlen;
3495 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3496 return dev_change_name(dev, ifr->ifr_newname);
3499 * Unknown or private ioctl
3503 if ((cmd >= SIOCDEVPRIVATE &&
3504 cmd <= SIOCDEVPRIVATE + 15) ||
3505 cmd == SIOCBONDENSLAVE ||
3506 cmd == SIOCBONDRELEASE ||
3507 cmd == SIOCBONDSETHWADDR ||
3508 cmd == SIOCBONDSLAVEINFOQUERY ||
3509 cmd == SIOCBONDINFOQUERY ||
3510 cmd == SIOCBONDCHANGEACTIVE ||
3511 cmd == SIOCGMIIPHY ||
3512 cmd == SIOCGMIIREG ||
3513 cmd == SIOCSMIIREG ||
3514 cmd == SIOCBRADDIF ||
3515 cmd == SIOCBRDELIF ||
3516 cmd == SIOCWANDEV) {
3518 if (dev->do_ioctl) {
3519 if (netif_device_present(dev))
3520 err = dev->do_ioctl(dev, ifr,
3533 * This function handles all "interface"-type I/O control requests. The actual
3534 * 'doing' part of this is dev_ifsioc above.
3538 * dev_ioctl - network device ioctl
3539 * @net: the applicable net namespace
3540 * @cmd: command to issue
3541 * @arg: pointer to a struct ifreq in user space
3543 * Issue ioctl functions to devices. This is normally called by the
3544 * user space syscall interfaces but can sometimes be useful for
3545 * other purposes. The return value is the return from the syscall if
3546 * positive or a negative errno code on error.
3549 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3555 /* One special case: SIOCGIFCONF takes ifconf argument
3556 and requires shared lock, because it sleeps writing
3560 if (cmd == SIOCGIFCONF) {
3562 ret = dev_ifconf(net, (char __user *) arg);
3566 if (cmd == SIOCGIFNAME)
3567 return dev_ifname(net, (struct ifreq __user *)arg);
3569 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3572 ifr.ifr_name[IFNAMSIZ-1] = 0;
3574 colon = strchr(ifr.ifr_name, ':');
3579 * See which interface the caller is talking about.
3584 * These ioctl calls:
3585 * - can be done by all.
3586 * - atomic and do not require locking.
3597 dev_load(net, ifr.ifr_name);
3598 read_lock(&dev_base_lock);
3599 ret = dev_ifsioc_locked(net, &ifr, cmd);
3600 read_unlock(&dev_base_lock);
3604 if (copy_to_user(arg, &ifr,
3605 sizeof(struct ifreq)))
3611 dev_load(net, ifr.ifr_name);
3613 ret = dev_ethtool(net, &ifr);
3618 if (copy_to_user(arg, &ifr,
3619 sizeof(struct ifreq)))
3625 * These ioctl calls:
3626 * - require superuser power.
3627 * - require strict serialization.
3633 if (!capable(CAP_NET_ADMIN))
3635 dev_load(net, ifr.ifr_name);
3637 ret = dev_ifsioc(net, &ifr, cmd);
3642 if (copy_to_user(arg, &ifr,
3643 sizeof(struct ifreq)))
3649 * These ioctl calls:
3650 * - require superuser power.
3651 * - require strict serialization.
3652 * - do not return a value
3662 case SIOCSIFHWBROADCAST:
3665 case SIOCBONDENSLAVE:
3666 case SIOCBONDRELEASE:
3667 case SIOCBONDSETHWADDR:
3668 case SIOCBONDCHANGEACTIVE:
3671 if (!capable(CAP_NET_ADMIN))
3674 case SIOCBONDSLAVEINFOQUERY:
3675 case SIOCBONDINFOQUERY:
3676 dev_load(net, ifr.ifr_name);
3678 ret = dev_ifsioc(net, &ifr, cmd);
3683 /* Get the per device memory space. We can add this but
3684 * currently do not support it */
3686 /* Set the per device memory buffer space.
3687 * Not applicable in our case */
3692 * Unknown or private ioctl.
3695 if (cmd == SIOCWANDEV ||
3696 (cmd >= SIOCDEVPRIVATE &&
3697 cmd <= SIOCDEVPRIVATE + 15)) {
3698 dev_load(net, ifr.ifr_name);
3700 ret = dev_ifsioc(net, &ifr, cmd);
3702 if (!ret && copy_to_user(arg, &ifr,
3703 sizeof(struct ifreq)))
3707 /* Take care of Wireless Extensions */
3708 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3709 return wext_handle_ioctl(net, &ifr, cmd, arg);
3716 * dev_new_index - allocate an ifindex
3717 * @net: the applicable net namespace
3719 * Returns a suitable unique value for a new device interface
3720 * number. The caller must hold the rtnl semaphore or the
3721 * dev_base_lock to be sure it remains unique.
3723 static int dev_new_index(struct net *net)
3729 if (!__dev_get_by_index(net, ifindex))
3734 /* Delayed registration/unregisteration */
3735 static DEFINE_SPINLOCK(net_todo_list_lock);
3736 static LIST_HEAD(net_todo_list);
3738 static void net_set_todo(struct net_device *dev)
3740 spin_lock(&net_todo_list_lock);
3741 list_add_tail(&dev->todo_list, &net_todo_list);
3742 spin_unlock(&net_todo_list_lock);
3745 static void rollback_registered(struct net_device *dev)
3747 BUG_ON(dev_boot_phase);
3750 /* Some devices call without registering for initialization unwind. */
3751 if (dev->reg_state == NETREG_UNINITIALIZED) {
3752 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3753 "was registered\n", dev->name, dev);
3759 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3761 /* If device is running, close it first. */
3764 /* And unlink it from device chain. */
3765 unlist_netdevice(dev);
3767 dev->reg_state = NETREG_UNREGISTERING;
3771 /* Shutdown queueing discipline. */
3775 /* Notify protocols, that we are about to destroy
3776 this device. They should clean all the things.
3778 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3781 * Flush the unicast and multicast chains
3783 dev_addr_discard(dev);
3788 /* Notifier chain MUST detach us from master device. */
3789 BUG_TRAP(!dev->master);
3791 /* Remove entries from kobject tree */
3792 netdev_unregister_kobject(dev);
3799 static void __netdev_init_queue_locks_one(struct net_device *dev,
3800 struct netdev_queue *dev_queue,
3803 spin_lock_init(&dev_queue->_xmit_lock);
3804 netdev_set_lockdep_class(&dev_queue->_xmit_lock, dev->type);
3805 dev_queue->xmit_lock_owner = -1;
3808 static void netdev_init_queue_locks(struct net_device *dev)
3810 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3811 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
3815 * register_netdevice - register a network device
3816 * @dev: device to register
3818 * Take a completed network device structure and add it to the kernel
3819 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3820 * chain. 0 is returned on success. A negative errno code is returned
3821 * on a failure to set up the device, or if the name is a duplicate.
3823 * Callers must hold the rtnl semaphore. You may want
3824 * register_netdev() instead of this.
3827 * The locking appears insufficient to guarantee two parallel registers
3828 * will not get the same name.
3831 int register_netdevice(struct net_device *dev)
3833 struct hlist_head *head;
3834 struct hlist_node *p;
3838 BUG_ON(dev_boot_phase);
3843 /* When net_device's are persistent, this will be fatal. */
3844 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3845 BUG_ON(!dev_net(dev));
3848 spin_lock_init(&dev->addr_list_lock);
3849 netdev_init_queue_locks(dev);
3853 /* Init, if this function is available */
3855 ret = dev->init(dev);
3863 if (!dev_valid_name(dev->name)) {
3868 dev->ifindex = dev_new_index(net);
3869 if (dev->iflink == -1)
3870 dev->iflink = dev->ifindex;
3872 /* Check for existence of name */
3873 head = dev_name_hash(net, dev->name);
3874 hlist_for_each(p, head) {
3875 struct net_device *d
3876 = hlist_entry(p, struct net_device, name_hlist);
3877 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3883 /* Fix illegal checksum combinations */
3884 if ((dev->features & NETIF_F_HW_CSUM) &&
3885 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3886 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3888 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3891 if ((dev->features & NETIF_F_NO_CSUM) &&
3892 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3893 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3895 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3899 /* Fix illegal SG+CSUM combinations. */
3900 if ((dev->features & NETIF_F_SG) &&
3901 !(dev->features & NETIF_F_ALL_CSUM)) {
3902 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3904 dev->features &= ~NETIF_F_SG;
3907 /* TSO requires that SG is present as well. */
3908 if ((dev->features & NETIF_F_TSO) &&
3909 !(dev->features & NETIF_F_SG)) {
3910 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3912 dev->features &= ~NETIF_F_TSO;
3914 if (dev->features & NETIF_F_UFO) {
3915 if (!(dev->features & NETIF_F_HW_CSUM)) {
3916 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3917 "NETIF_F_HW_CSUM feature.\n",
3919 dev->features &= ~NETIF_F_UFO;
3921 if (!(dev->features & NETIF_F_SG)) {
3922 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3923 "NETIF_F_SG feature.\n",
3925 dev->features &= ~NETIF_F_UFO;
3929 netdev_initialize_kobject(dev);
3930 ret = netdev_register_kobject(dev);
3933 dev->reg_state = NETREG_REGISTERED;
3936 * Default initial state at registry is that the
3937 * device is present.
3940 set_bit(__LINK_STATE_PRESENT, &dev->state);
3942 dev_init_scheduler(dev);
3944 list_netdevice(dev);
3946 /* Notify protocols, that a new device appeared. */
3947 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
3948 ret = notifier_to_errno(ret);
3950 rollback_registered(dev);
3951 dev->reg_state = NETREG_UNREGISTERED;
3964 * register_netdev - register a network device
3965 * @dev: device to register
3967 * Take a completed network device structure and add it to the kernel
3968 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3969 * chain. 0 is returned on success. A negative errno code is returned
3970 * on a failure to set up the device, or if the name is a duplicate.
3972 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3973 * and expands the device name if you passed a format string to
3976 int register_netdev(struct net_device *dev)
3983 * If the name is a format string the caller wants us to do a
3986 if (strchr(dev->name, '%')) {
3987 err = dev_alloc_name(dev, dev->name);
3992 err = register_netdevice(dev);
3997 EXPORT_SYMBOL(register_netdev);
4000 * netdev_wait_allrefs - wait until all references are gone.
4002 * This is called when unregistering network devices.
4004 * Any protocol or device that holds a reference should register
4005 * for netdevice notification, and cleanup and put back the
4006 * reference if they receive an UNREGISTER event.
4007 * We can get stuck here if buggy protocols don't correctly
4010 static void netdev_wait_allrefs(struct net_device *dev)
4012 unsigned long rebroadcast_time, warning_time;
4014 rebroadcast_time = warning_time = jiffies;
4015 while (atomic_read(&dev->refcnt) != 0) {
4016 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4019 /* Rebroadcast unregister notification */
4020 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4022 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4024 /* We must not have linkwatch events
4025 * pending on unregister. If this
4026 * happens, we simply run the queue
4027 * unscheduled, resulting in a noop
4030 linkwatch_run_queue();
4035 rebroadcast_time = jiffies;
4040 if (time_after(jiffies, warning_time + 10 * HZ)) {
4041 printk(KERN_EMERG "unregister_netdevice: "
4042 "waiting for %s to become free. Usage "
4044 dev->name, atomic_read(&dev->refcnt));
4045 warning_time = jiffies;
4054 * register_netdevice(x1);
4055 * register_netdevice(x2);
4057 * unregister_netdevice(y1);
4058 * unregister_netdevice(y2);
4064 * We are invoked by rtnl_unlock() after it drops the semaphore.
4065 * This allows us to deal with problems:
4066 * 1) We can delete sysfs objects which invoke hotplug
4067 * without deadlocking with linkwatch via keventd.
4068 * 2) Since we run with the RTNL semaphore not held, we can sleep
4069 * safely in order to wait for the netdev refcnt to drop to zero.
4071 static DEFINE_MUTEX(net_todo_run_mutex);
4072 void netdev_run_todo(void)
4074 struct list_head list;
4076 /* Need to guard against multiple cpu's getting out of order. */
4077 mutex_lock(&net_todo_run_mutex);
4079 /* Not safe to do outside the semaphore. We must not return
4080 * until all unregister events invoked by the local processor
4081 * have been completed (either by this todo run, or one on
4084 if (list_empty(&net_todo_list))
4087 /* Snapshot list, allow later requests */
4088 spin_lock(&net_todo_list_lock);
4089 list_replace_init(&net_todo_list, &list);
4090 spin_unlock(&net_todo_list_lock);
4092 while (!list_empty(&list)) {
4093 struct net_device *dev
4094 = list_entry(list.next, struct net_device, todo_list);
4095 list_del(&dev->todo_list);
4097 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4098 printk(KERN_ERR "network todo '%s' but state %d\n",
4099 dev->name, dev->reg_state);
4104 dev->reg_state = NETREG_UNREGISTERED;
4106 netdev_wait_allrefs(dev);
4109 BUG_ON(atomic_read(&dev->refcnt));
4110 BUG_TRAP(!dev->ip_ptr);
4111 BUG_TRAP(!dev->ip6_ptr);
4112 BUG_TRAP(!dev->dn_ptr);
4114 if (dev->destructor)
4115 dev->destructor(dev);
4117 /* Free network device */
4118 kobject_put(&dev->dev.kobj);
4122 mutex_unlock(&net_todo_run_mutex);
4125 static struct net_device_stats *internal_stats(struct net_device *dev)
4130 static void netdev_init_one_queue(struct net_device *dev,
4131 struct netdev_queue *queue,
4134 spin_lock_init(&queue->lock);
4138 static void netdev_init_queues(struct net_device *dev)
4140 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4141 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4145 * alloc_netdev_mq - allocate network device
4146 * @sizeof_priv: size of private data to allocate space for
4147 * @name: device name format string
4148 * @setup: callback to initialize device
4149 * @queue_count: the number of subqueues to allocate
4151 * Allocates a struct net_device with private data area for driver use
4152 * and performs basic initialization. Also allocates subquue structs
4153 * for each queue on the device at the end of the netdevice.
4155 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4156 void (*setup)(struct net_device *), unsigned int queue_count)
4158 struct netdev_queue *tx;
4159 struct net_device *dev;
4163 BUG_ON(strlen(name) >= sizeof(dev->name));
4165 alloc_size = sizeof(struct net_device);
4167 /* ensure 32-byte alignment of private area */
4168 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4169 alloc_size += sizeof_priv;
4171 /* ensure 32-byte alignment of whole construct */
4172 alloc_size += NETDEV_ALIGN_CONST;
4174 p = kzalloc(alloc_size, GFP_KERNEL);
4176 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4180 tx = kzalloc(sizeof(struct netdev_queue) * queue_count, GFP_KERNEL);
4182 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4188 dev = (struct net_device *)
4189 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4190 dev->padded = (char *)dev - (char *)p;
4191 dev_net_set(dev, &init_net);
4194 dev->num_tx_queues = queue_count;
4195 dev->real_num_tx_queues = queue_count;
4198 dev->priv = ((char *)dev +
4199 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
4200 & ~NETDEV_ALIGN_CONST));
4203 dev->gso_max_size = GSO_MAX_SIZE;
4205 netdev_init_queues(dev);
4207 dev->get_stats = internal_stats;
4208 netpoll_netdev_init(dev);
4210 strcpy(dev->name, name);
4213 EXPORT_SYMBOL(alloc_netdev_mq);
4216 * free_netdev - free network device
4219 * This function does the last stage of destroying an allocated device
4220 * interface. The reference to the device object is released.
4221 * If this is the last reference then it will be freed.
4223 void free_netdev(struct net_device *dev)
4225 release_net(dev_net(dev));
4229 /* Compatibility with error handling in drivers */
4230 if (dev->reg_state == NETREG_UNINITIALIZED) {
4231 kfree((char *)dev - dev->padded);
4235 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4236 dev->reg_state = NETREG_RELEASED;
4238 /* will free via device release */
4239 put_device(&dev->dev);
4242 /* Synchronize with packet receive processing. */
4243 void synchronize_net(void)
4250 * unregister_netdevice - remove device from the kernel
4253 * This function shuts down a device interface and removes it
4254 * from the kernel tables.
4256 * Callers must hold the rtnl semaphore. You may want
4257 * unregister_netdev() instead of this.
4260 void unregister_netdevice(struct net_device *dev)
4264 rollback_registered(dev);
4265 /* Finish processing unregister after unlock */
4270 * unregister_netdev - remove device from the kernel
4273 * This function shuts down a device interface and removes it
4274 * from the kernel tables.
4276 * This is just a wrapper for unregister_netdevice that takes
4277 * the rtnl semaphore. In general you want to use this and not
4278 * unregister_netdevice.
4280 void unregister_netdev(struct net_device *dev)
4283 unregister_netdevice(dev);
4287 EXPORT_SYMBOL(unregister_netdev);
4290 * dev_change_net_namespace - move device to different nethost namespace
4292 * @net: network namespace
4293 * @pat: If not NULL name pattern to try if the current device name
4294 * is already taken in the destination network namespace.
4296 * This function shuts down a device interface and moves it
4297 * to a new network namespace. On success 0 is returned, on
4298 * a failure a netagive errno code is returned.
4300 * Callers must hold the rtnl semaphore.
4303 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4306 const char *destname;
4311 /* Don't allow namespace local devices to be moved. */
4313 if (dev->features & NETIF_F_NETNS_LOCAL)
4316 /* Ensure the device has been registrered */
4318 if (dev->reg_state != NETREG_REGISTERED)
4321 /* Get out if there is nothing todo */
4323 if (net_eq(dev_net(dev), net))
4326 /* Pick the destination device name, and ensure
4327 * we can use it in the destination network namespace.
4330 destname = dev->name;
4331 if (__dev_get_by_name(net, destname)) {
4332 /* We get here if we can't use the current device name */
4335 if (!dev_valid_name(pat))
4337 if (strchr(pat, '%')) {
4338 if (__dev_alloc_name(net, pat, buf) < 0)
4343 if (__dev_get_by_name(net, destname))
4348 * And now a mini version of register_netdevice unregister_netdevice.
4351 /* If device is running close it first. */
4354 /* And unlink it from device chain */
4356 unlist_netdevice(dev);
4360 /* Shutdown queueing discipline. */
4363 /* Notify protocols, that we are about to destroy
4364 this device. They should clean all the things.
4366 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4369 * Flush the unicast and multicast chains
4371 dev_addr_discard(dev);
4373 /* Actually switch the network namespace */
4374 dev_net_set(dev, net);
4376 /* Assign the new device name */
4377 if (destname != dev->name)
4378 strcpy(dev->name, destname);
4380 /* If there is an ifindex conflict assign a new one */
4381 if (__dev_get_by_index(net, dev->ifindex)) {
4382 int iflink = (dev->iflink == dev->ifindex);
4383 dev->ifindex = dev_new_index(net);
4385 dev->iflink = dev->ifindex;
4388 /* Fixup kobjects */
4389 netdev_unregister_kobject(dev);
4390 err = netdev_register_kobject(dev);
4393 /* Add the device back in the hashes */
4394 list_netdevice(dev);
4396 /* Notify protocols, that a new device appeared. */
4397 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4405 static int dev_cpu_callback(struct notifier_block *nfb,
4406 unsigned long action,
4409 struct sk_buff **list_skb;
4410 struct netdev_queue **list_net;
4411 struct sk_buff *skb;
4412 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4413 struct softnet_data *sd, *oldsd;
4415 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4418 local_irq_disable();
4419 cpu = smp_processor_id();
4420 sd = &per_cpu(softnet_data, cpu);
4421 oldsd = &per_cpu(softnet_data, oldcpu);
4423 /* Find end of our completion_queue. */
4424 list_skb = &sd->completion_queue;
4426 list_skb = &(*list_skb)->next;
4427 /* Append completion queue from offline CPU. */
4428 *list_skb = oldsd->completion_queue;
4429 oldsd->completion_queue = NULL;
4431 /* Find end of our output_queue. */
4432 list_net = &sd->output_queue;
4434 list_net = &(*list_net)->next_sched;
4435 /* Append output queue from offline CPU. */
4436 *list_net = oldsd->output_queue;
4437 oldsd->output_queue = NULL;
4439 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4442 /* Process offline CPU's input_pkt_queue */
4443 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4449 #ifdef CONFIG_NET_DMA
4451 * net_dma_rebalance - try to maintain one DMA channel per CPU
4452 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4454 * This is called when the number of channels allocated to the net_dma client
4455 * changes. The net_dma client tries to have one DMA channel per CPU.
4458 static void net_dma_rebalance(struct net_dma *net_dma)
4460 unsigned int cpu, i, n, chan_idx;
4461 struct dma_chan *chan;
4463 if (cpus_empty(net_dma->channel_mask)) {
4464 for_each_online_cpu(cpu)
4465 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4470 cpu = first_cpu(cpu_online_map);
4472 for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
4473 chan = net_dma->channels[chan_idx];
4475 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4476 + (i < (num_online_cpus() %
4477 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4480 per_cpu(softnet_data, cpu).net_dma = chan;
4481 cpu = next_cpu(cpu, cpu_online_map);
4489 * netdev_dma_event - event callback for the net_dma_client
4490 * @client: should always be net_dma_client
4491 * @chan: DMA channel for the event
4492 * @state: DMA state to be handled
4494 static enum dma_state_client
4495 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4496 enum dma_state state)
4498 int i, found = 0, pos = -1;
4499 struct net_dma *net_dma =
4500 container_of(client, struct net_dma, client);
4501 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4503 spin_lock(&net_dma->lock);
4505 case DMA_RESOURCE_AVAILABLE:
4506 for (i = 0; i < nr_cpu_ids; i++)
4507 if (net_dma->channels[i] == chan) {
4510 } else if (net_dma->channels[i] == NULL && pos < 0)
4513 if (!found && pos >= 0) {
4515 net_dma->channels[pos] = chan;
4516 cpu_set(pos, net_dma->channel_mask);
4517 net_dma_rebalance(net_dma);
4520 case DMA_RESOURCE_REMOVED:
4521 for (i = 0; i < nr_cpu_ids; i++)
4522 if (net_dma->channels[i] == chan) {
4530 cpu_clear(pos, net_dma->channel_mask);
4531 net_dma->channels[i] = NULL;
4532 net_dma_rebalance(net_dma);
4538 spin_unlock(&net_dma->lock);
4544 * netdev_dma_regiser - register the networking subsystem as a DMA client
4546 static int __init netdev_dma_register(void)
4548 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4550 if (unlikely(!net_dma.channels)) {
4552 "netdev_dma: no memory for net_dma.channels\n");
4555 spin_lock_init(&net_dma.lock);
4556 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4557 dma_async_client_register(&net_dma.client);
4558 dma_async_client_chan_request(&net_dma.client);
4563 static int __init netdev_dma_register(void) { return -ENODEV; }
4564 #endif /* CONFIG_NET_DMA */
4567 * netdev_compute_feature - compute conjunction of two feature sets
4568 * @all: first feature set
4569 * @one: second feature set
4571 * Computes a new feature set after adding a device with feature set
4572 * @one to the master device with current feature set @all. Returns
4573 * the new feature set.
4575 int netdev_compute_features(unsigned long all, unsigned long one)
4577 /* if device needs checksumming, downgrade to hw checksumming */
4578 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4579 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4581 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4582 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4583 all ^= NETIF_F_HW_CSUM
4584 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4586 if (one & NETIF_F_GSO)
4587 one |= NETIF_F_GSO_SOFTWARE;
4590 /* If even one device supports robust GSO, enable it for all. */
4591 if (one & NETIF_F_GSO_ROBUST)
4592 all |= NETIF_F_GSO_ROBUST;
4594 all &= one | NETIF_F_LLTX;
4596 if (!(all & NETIF_F_ALL_CSUM))
4598 if (!(all & NETIF_F_SG))
4599 all &= ~NETIF_F_GSO_MASK;
4603 EXPORT_SYMBOL(netdev_compute_features);
4605 static struct hlist_head *netdev_create_hash(void)
4608 struct hlist_head *hash;
4610 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4612 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4613 INIT_HLIST_HEAD(&hash[i]);
4618 /* Initialize per network namespace state */
4619 static int __net_init netdev_init(struct net *net)
4621 INIT_LIST_HEAD(&net->dev_base_head);
4623 net->dev_name_head = netdev_create_hash();
4624 if (net->dev_name_head == NULL)
4627 net->dev_index_head = netdev_create_hash();
4628 if (net->dev_index_head == NULL)
4634 kfree(net->dev_name_head);
4639 static void __net_exit netdev_exit(struct net *net)
4641 kfree(net->dev_name_head);
4642 kfree(net->dev_index_head);
4645 static struct pernet_operations __net_initdata netdev_net_ops = {
4646 .init = netdev_init,
4647 .exit = netdev_exit,
4650 static void __net_exit default_device_exit(struct net *net)
4652 struct net_device *dev, *next;
4654 * Push all migratable of the network devices back to the
4655 * initial network namespace
4658 for_each_netdev_safe(net, dev, next) {
4660 char fb_name[IFNAMSIZ];
4662 /* Ignore unmoveable devices (i.e. loopback) */
4663 if (dev->features & NETIF_F_NETNS_LOCAL)
4666 /* Push remaing network devices to init_net */
4667 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4668 err = dev_change_net_namespace(dev, &init_net, fb_name);
4670 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4671 __func__, dev->name, err);
4678 static struct pernet_operations __net_initdata default_device_ops = {
4679 .exit = default_device_exit,
4683 * Initialize the DEV module. At boot time this walks the device list and
4684 * unhooks any devices that fail to initialise (normally hardware not
4685 * present) and leaves us with a valid list of present and active devices.
4690 * This is called single threaded during boot, so no need
4691 * to take the rtnl semaphore.
4693 static int __init net_dev_init(void)
4695 int i, rc = -ENOMEM;
4697 BUG_ON(!dev_boot_phase);
4699 if (dev_proc_init())
4702 if (netdev_kobject_init())
4705 INIT_LIST_HEAD(&ptype_all);
4706 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4707 INIT_LIST_HEAD(&ptype_base[i]);
4709 if (register_pernet_subsys(&netdev_net_ops))
4712 if (register_pernet_device(&default_device_ops))
4716 * Initialise the packet receive queues.
4719 for_each_possible_cpu(i) {
4720 struct softnet_data *queue;
4722 queue = &per_cpu(softnet_data, i);
4723 skb_queue_head_init(&queue->input_pkt_queue);
4724 queue->completion_queue = NULL;
4725 INIT_LIST_HEAD(&queue->poll_list);
4727 queue->backlog.poll = process_backlog;
4728 queue->backlog.weight = weight_p;
4731 netdev_dma_register();
4735 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
4736 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
4738 hotcpu_notifier(dev_cpu_callback, 0);
4746 subsys_initcall(net_dev_init);
4748 EXPORT_SYMBOL(__dev_get_by_index);
4749 EXPORT_SYMBOL(__dev_get_by_name);
4750 EXPORT_SYMBOL(__dev_remove_pack);
4751 EXPORT_SYMBOL(dev_valid_name);
4752 EXPORT_SYMBOL(dev_add_pack);
4753 EXPORT_SYMBOL(dev_alloc_name);
4754 EXPORT_SYMBOL(dev_close);
4755 EXPORT_SYMBOL(dev_get_by_flags);
4756 EXPORT_SYMBOL(dev_get_by_index);
4757 EXPORT_SYMBOL(dev_get_by_name);
4758 EXPORT_SYMBOL(dev_open);
4759 EXPORT_SYMBOL(dev_queue_xmit);
4760 EXPORT_SYMBOL(dev_remove_pack);
4761 EXPORT_SYMBOL(dev_set_allmulti);
4762 EXPORT_SYMBOL(dev_set_promiscuity);
4763 EXPORT_SYMBOL(dev_change_flags);
4764 EXPORT_SYMBOL(dev_set_mtu);
4765 EXPORT_SYMBOL(dev_set_mac_address);
4766 EXPORT_SYMBOL(free_netdev);
4767 EXPORT_SYMBOL(netdev_boot_setup_check);
4768 EXPORT_SYMBOL(netdev_set_master);
4769 EXPORT_SYMBOL(netdev_state_change);
4770 EXPORT_SYMBOL(netif_receive_skb);
4771 EXPORT_SYMBOL(netif_rx);
4772 EXPORT_SYMBOL(register_gifconf);
4773 EXPORT_SYMBOL(register_netdevice);
4774 EXPORT_SYMBOL(register_netdevice_notifier);
4775 EXPORT_SYMBOL(skb_checksum_help);
4776 EXPORT_SYMBOL(synchronize_net);
4777 EXPORT_SYMBOL(unregister_netdevice);
4778 EXPORT_SYMBOL(unregister_netdevice_notifier);
4779 EXPORT_SYMBOL(net_enable_timestamp);
4780 EXPORT_SYMBOL(net_disable_timestamp);
4781 EXPORT_SYMBOL(dev_get_flags);
4783 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4784 EXPORT_SYMBOL(br_handle_frame_hook);
4785 EXPORT_SYMBOL(br_fdb_get_hook);
4786 EXPORT_SYMBOL(br_fdb_put_hook);
4790 EXPORT_SYMBOL(dev_load);
4793 EXPORT_PER_CPU_SYMBOL(softnet_data);